SERIES ON CONCRETE AND APPLICABLE MATHEMATICS ISSN: 1793-1142 Series Editor: Professor George A. Anastassiou Department...
114 downloads
2167 Views
2MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
SERIES ON CONCRETE AND APPLICABLE MATHEMATICS ISSN: 1793-1142 Series Editor: Professor George A. Anastassiou Department of Mathematical Sciences The University of Memphis Memphis, TN 38152, USA
Published Vol. 1
Long Time Behaviour of Classical and Quantum Systems edited by S. Graffi & A. Martinez
Vol. 2
Problems in Probability by T. M. Mills
Vol. 3
Introduction to Matrix Theory: With Applications to Business and Economics by F. Szidarovszky & S. Molnár
Vol. 4
Stochastic Models with Applications to Genetics, Cancers, Aids and Other Biomedical Systems by Tan Wai-Yuan
Vol. 5
Defects of Properties in Mathematics: Quantitative Characterizations by Adrian I. Ban & Sorin G. Gal
Vol. 6
Topics on Stability and Periodicity in Abstract Differential Equations by James H. Liu, Gaston M. N’Guérékata & Nguyen Van Minh
Vol. 7
Probabilistic Inequalities by George A. Anastassiou
Vol. 8
Approximation by Complex Bernstein and Convolution Type Operators by Sorin G. Gal
Vol. 9
Distribution Theory and Applications by Abdellah El Kinani & Mohamed Oudadess
Vol. 11 Advanced Inequalities by George A. Anastassiou
Forthcoming Vol. 10 Theory and Examples of Ordinary Differential Equations by Chin-Yuan Lin
ZhangJi - Advanced Inequalities.pmd
2
8/31/2010, 2:20 PM
Series on Concrete and Applicable Mathematics – Vol.11
George A Anastassiou University of Memphis, USA
World Scientific NEW JERSEY
•
LONDON
•
SINGAPORE
•
BEIJING
•
SHANGHAI
•
HONG KONG
•
TA I P E I
•
CHENNAI
Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.
ADVANCED INEQUALITIES Series on Concrete and Applicable Mathematics — Vol. 11 Copyright © 2011 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN-13 978-981-4317-62-7 ISBN-10 981-4317-62-4
Printed in Singapore.
ZhangJi - Advanced Inequalities.pmd
1
8/31/2010, 2:20 PM
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Dedicated to my daughter Peggy
v
Book˙Adv˙Ineq
This page intentionally left blank
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Preface
In this monograph we present univariate and multivariate classical analysis advanced inequalities. This treatise relies on author’s last thirteen years related research work, more precisely see [15]-[61], and it is a natural outgrowth of it. Chapters are self-contained and several advanced courses can be taught out of this book. Extensive background and motivations are given per chapter. A very extensive list of references is given at the end. The topics covered are very diverse. We present recent advances on Ostrowski type inequalities, see Chapters 2-9; on Opial type inequalities, see Chapters 10-16; Poincare and Sobolev type inequalities, see Chapters 17-20. Also we present HardyOpial type inequalities, as well as Ostrowski like ones, see Chapters 21-22. Then we present some work on ordinary and distributional Taylor formulae with estimates for their remainders and applications, see Chapters 23-24. Finally in Chapters 25-29 we study Chebyshev-Gruss, Gruss, Comparison of Means inequalities. Our book’s results appeared for the first time in our published articles mentioned above, and they are mostly optimal, that is usually our inequalities are sharp and attained. They are expected to find applications in many areas of pure and applied mathematics, such as mathematical analysis, probability, ordinary and partial differential equations, numerical analysis, information theory, etc. As such this monograph is suitable for researchers, graduate students, and seminars of the above subjects, also to be in all science libraries. The preparation of book took place during 2008-2009 in Memphis, Tennessee, USA. I would like to thank my family for their dedication and love to me, which was the strongest support during the writing of this monograph. Also many thanks go to my typist Rodica Gal for an excellent and on time technical job. January 1, 2010 George A. Anastassiou Department of Mathematical Sciences The University of Memphis, TN U.S.A. vii
Book˙Adv˙Ineq
This page intentionally left blank
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Contents
Preface
vii
1. Introduction
1
2. Advanced Univariate Ostrowski Type Inequalities
5
2.1 2.2 2.3
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Auxilliary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3. Higher Order Ostrowski Inequalities 3.1 3.2
21
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4. Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities 4.1 4.2 4.3 4.4 4.5
Introduction . Background . Main Results Applications . Sharpness . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Auxilliary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6. Ostrowski Inequalities on Euclidean Domains 6.1 6.2
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
21 22
27
5. More on Multidimensional Ostrowski Type Inequalities 5.1 5.2 5.3
5 6 14
27 30 31 53 76 81 81 82 87 93 93 93
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
x
7. High Order Ostrowski Inequalities on Euclidean Domains 7.1 7.2 7.3
99
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Functions on General Domains . . . . . . . . . . . . . . . . . . . . 106
8. Ostrowski Inequalities on Spherical Shells 8.1 8.2 8.3
109
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Addendum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
9. Ostrowski Inequalities on Balls and Shells Via Taylor–Widder Formula 9.1 9.2 9.3 9.4 9.5
Introduction . . . . . . Background . . . . . . Results on the Shell . Results on the Sphere Addendum . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
125 . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
10. Multivariate Opial Type Inequalities for Functions Vanishing at an Interior Point 10.1 10.2
139
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
11. General Multivariate Weighted Opial Inequalities 11.1 11.2
149
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
12. Opial Inequalities for Widder Derivatives 12.1 12.2 12.3
161
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
13. Opial Inequalities for Linear Differential Operators 13.1 13.2
171
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
14. Opial Inequalities for Vector Valued Functions 14.1 14.2 14.3 14.4
125 126 128 132 138
Introduction Background . Results . . . Applications .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
179 . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
179 179 180 186
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Contents
xi
15. Opial Inequalities for Semigroups 15.1 15.2 15.3
187
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
16. Opial Inequalities for Cosine and Sine Operator Functions 16.1 16.2 16.3 16.4
Introduction . Background . Results . . . Applications .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
197 . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
17. Poincar´e Like Inequalities for Linear Differential Operators 17.1 17.2
209
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
18. Poincar´e and Sobolev Like Inequalities for Widder Derivatives 18.1 18.2
215
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
19. Poincar´e and Sobolev Like Inequalities for Vector Valued Functions 19.1 19.2 19.3 19.4
Introduction . Background . Results . . . Applications .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
229 . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
20. Poincar´e Type Inequalities for Semigroups, Cosine and Sine Operator Functions 20.1 20.2 20.3 20.4 20.5
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . Semigroups Background . . . . . . . . . . . . . . . . . . Poincar´e Type Inequalities for Semigroups . . . . . . . . Cosine and Sine Operator Functions Background . . . . Poincar´e Type Inequalities for Cosine and Sine Operator Functions . . . . . . . . . . . . . . . . . . . . . . . . . .
21. Hardy–Opial Type Inequalities 21.1
Results
229 230 231 240
243 . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
243 243 245 249
. . . . . . 250 261
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
22. A Basic Sharp Integral Inequality 22.1 22.2
197 197 199 206
271
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
xii
23. Estimates of the Remainder in Taylor’s Formula 23.1 23.2 23.3 23.4 23.5
Introduction . . . . . . . . . . . . . . . . Some New Bounds for the Remainder . Some Further Bounds of the Remainder Some Inequalities for Special Cases . . . Taylor-Multivariate Case Estimates . . .
. . . . .
275 . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
24. The Distributional Taylor Formula 24.1 24.2 24.3
319
331
341
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
29. Multivariate Fink Type Identity Applied to Multivariate Inequalities 29.1 29.2 29.3
305
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
28. Multivariate Chebyshev–Gr¨ uss and Comparison of Integral Means Inequalities 28.1 28.2
293
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319 Auxiliary Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
27. Chebyshev–Gr¨ uss Type Inequalities on Spherical Shells and Balls 27.1 27.2
275 276 280 282 284
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
26. Gr¨ uss Type Multivariate Integral Inequalities 26.1 26.2 26.3
. . . . .
Introduction and Background . . . . . . . . . . . . . . . . . . . . . 293 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
25. Chebyshev–Gr¨ uss Type Inequalities Using Euler Type and Fink Identities 25.1 25.2
. . . . .
365
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
Bibliography
395
List of Symbols
407
Index
409
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Chapter 1
Introduction
Here we describe the material contained in this monograph. Our results are mostly optimal, i.e. sharp, attained inequalities. We give also applications. The exposed results are brought for the first time in a book form. CHAPTER 2: Very general univariate Ostrowski type inequalities are given, involving the k · k∞ and k · kp , p ≥ 1 norms of the engaged nth order derivative, n ≥ 1. In establishing them, several important univariate identities of Montgomery type are developed. CHAPTER 3: We generalize Ostrowski inequality for higher order derivatives, by using a generalized Euler-type identity. Some of the produced inequalities are sharp, namely attained by basic functions. The rest of the estimates are tight. We give applications to trapezoidal and midpoint rules. Estimates are given with respect to L∞ norm. CHAPTER 4: We present a general multivariate Euler type identity. Using it we derive general tight multivariate high order Ostrowski type inequalities for the estimate on the error of a multivariate function f evaluated at a point from its average. The estimates are involving only the single partial derivatives of f and are with respect to k · kp , 1 ≤ p ≤ ∞. We give specific applications of our main results to the multivariate trapezoid and midpoint rules for functions f differentiable up to order 6. We prove sharpness of our inequalities for differentiation orders m = 1, 2, 4 and with respect to k · k∞ . CHAPTER 5: Very general multidimensional Ostrowski type inequalities are presented, some of them are proved to be sharp. They involve the k · k∞ and k · kp norms of the engaged mixed partial of nth order n ≥ 1. In establishing them, other important multivariate results of Montgomery type identity are developed. CHAPTER 6: The classical Ostrowski inequality for functions on intervals is extended to functions on general domains in Euclidean space. For radial functions on balls the inequality is sharp. CHAPTER 7: The classical Ostrowski inequality for functions on intervals estimates the value of the function minus its average in terms of the maximum of its first derivative. This result is extended to functions on general domains using the 1
Book˙Adv˙Ineq
17th August 2010
2
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
L∞ norm of its nth partial derivatives. For radial functions on balls the inequality is sharp. CHAPTER 8: Here are presented Ostrowski type inequalities over spherical shells. These regard sharp or close to sharp estimates to the difference of the average of a multivariate function from its value at a point. CHAPTER 9: The classical Ostrowski inequality for functions on intervals estimates the value of the function minus its average in terms of the maximum of its first derivative. This result is extended to higher order over shells and balls of RN , N ≥ 1, with respect to an extended complete Chebyshev system and the generalized radial derivatives of Widder type. We treat radial and non-radial functions. CHAPTER 10: Here we generalize Opial inequalities in the multidimensional case over balls. The inequalities carry weights and are proved to be sharp. The functions under consideration vanish at the center of the ball. CHAPTER 11: Here we give Opial type weighted multidimensional inequalities over balls and arbitrary smooth bounded domains. The inequalities are mostly sharp. The functions under consideration vanish on the boundary. CHAPTER 12: Various Lp form Opial type inequalities are given for Widder derivatives. CHAPTER 13: Various Lp form Opial type inequalities are presented for a linear differential operator L, involving its related initial value problem solution y, Ly, the associated Green’s function H and initial conditions point x0 ∈ R. CHAPTER 14: Various Lp form Opial type inequalities are shown for functions valued in a Banach vector space. We give applications in C. CHAPTER 15: Various Lp form Opial type inequalities are given for semigroups with applications. CHAPTER 16: Various Lp form Opial type inequalities are shown for cosine and sine operator functions with applications. CHAPTER 17: Various Lp form Poincar´e like inequalities, forward and reverse, are given for a linear differential operator L, involving its related initial value problem solution y, Ly, the associated Green’s function H and initial conditions at point x0 ∈ R. CHAPTER 18: Various Lp form Poincar´e and Sobolev like inequalities, forward and reverse, are presented involving Widder derivative. CHAPTER 19: Various Lp form Poincar´e and Sobolev like inequalities are given for functions valued in a Banach vector space. We give applications on C. CHAPTER 20: Here we present Poincar´e type general Lp inequalities regarding semigroups, cosine and sine operator functions. We apply inequalities to specific cases of them. CHAPTER 21: Various Lp form Hardy–Opial type sharp integral inequalities are derived involving two functions. CHAPTER 22: A sharp multidimensional integral type inequality is presented involving nth order (n ∈ N) mixed partial derivatives. This is subject to some basic
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Introduction
Book˙Adv˙Ineq
3
boundary condition satisfied by the involved multivariate function. CHAPTER 23: Important estimates of the remainder in Taylor’s formula are given. We treat both univariate and multivariate cases. CHAPTER 24:We derive a distributional Taylor formula with precise integral remainder. We give applications of it and estimates for the remainder. CHAPTER 25: Here we present Chebyshev–Gr¨ uss type univariate inequalities by using the generalized Euler type and Fink identities. The results involve the functions f, g, f (n) , g (n) , n ∈ N, and are with respect to k · kp , 1 ≤ p ≤ ∞. CHAPTER 26: Here are presented Gr¨ uss type multivariate integral inequalities. CHAPTER 27: Here we give Chebyshev–Gr¨ uss type inequalities on R N over spherical shells and balls by extending some basic univariate results of Pachpatte [206]. CHAPTER 28: We present tight multivariate Chebyshev–Gr¨ uss and comparison of integral means inequalities by using a general multivariate Euler type identity. Our results involve the functions f, g and their high order single partials and are with respect to k · kp , 1 ≤ p ≤ ∞. CHAPTER 29: We describe a general multivariate Fink type identity which is a representation formula for a multivariate function. Using it we derive general tight multivariate high order Ostrowski type, comparison of means and Gr¨ uss type inequalities. The estimates involve Lp norms, any 1 ≤ p ≤ ∞. Our presented results are expected to find applications in many areas of pure and applied mathematics, such as mathematical analysis, probability, ordinary and partial differential equations, numerical analysis, information theory, etc.
This page intentionally left blank
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Chapter 2
Advanced Univariate Ostrowski Type Inequalities
Very general univariate Ostrowski type inequalities are presented, involving the k · k∞ and k · kp , p ≥ 1 norms of the engaged nth order derivative, n ≥ 1. In proving them, several important univariate identities of Montgomery type are given. This chapter follows [24]. 2.1
Introduction
In 1938, A. Ostrowski [196] proved the following important inequality: Theorem 2.1. Let f : [a, b] → R be continuous on [a, b] and differentiable on (a, b) whose derivative f 0 : (a, b) → R is bounded on (a, b), i.e., kf 0 k∞ := supt∈(a,b) |f 0 (t)| < +∞. Then " # 1 Z b 2 ) 1 (x − a+b 2 f (t)dt − f (x) ≤ · (b − a)kf 0 k∞ , (2.1) + b − a a 4 (b − a)2
for any x ∈ [a, b]. The constant
1 4
is the best possible.
Since then there has been a lot of activity around these inequalities with important applications to Numerical Analysis and Probability. This chapter is greatly motivated and inspired also by the following results. Theorem 2.2 (see [16]). Let f ∈ C n+1 ([a, b]), n ∈ N and x ∈ [a, b] be fixed, such that f (k) (x) = 0, k = 1, . . . , n. Then it holds 1 Z b kf (n+1) k (x − a)n+2 + (b − x)n+2 ∞ f (y)dy − f (x) ≤ . (2.2) · b − a a (n + 2)! b−a
Inequality (2.2) is sharp. In particular, when n is odd is attained by f ∗ (y) := (y − x)n+1 · (b − a), while when n is even the optimal function is f˜(y) := |y − x|n+α · (b − a),
α > 1.
Clearly inequality (2.2) generalizes inequality (2.1) for higher order derivatives of f. 5
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
6
One of the goals of this chapter is to give a generalization Theorem 4.17, p. 191 of [85], see here Theorem 2.13 later. We would like to mention that result. Theorem 2.3. Let f : [a, b] → R be continuous on [a, b] and twice differentiable function on (a, b), whose second derivative f 00 : (a, b) → R is bounded on (a, b). Then Z b a + b 1 f (b) − f (a) x− f (t)dt − f (x) − b−a a b−a 2 " #2 2 1 (x − a+b 1 1 2 ) ≤ + + 2 (b − a)2 4 12 ·(b − a)2 · kf 00 k∞ ≤
kf 00 k∞ (b − a)2 , 6
for all x ∈ [a, b].
(2.3)
In this chapter we present very general Ostrowski type inequalities in the one dimensional case. Involved integrals here are also of Riemann–Stieltjes form. Along the way to establishing these results, we show several important side univariate related results, generalizing Montgomery’s identity ([188], Ch. XVII, p. 565), see Section 2.2. 2.2
Auxilliary Results
We present the following results we need, which by themselves are of independent interest. Theorem 2.4. Let f : [a, b] → R be n-times differentiable on [a, b], n ∈ N. The nth derivative f (n) : [a, b] → R is integrable on [a, b]. Let x ∈ [a, b]. Define the kernel s−a , a ≤ s ≤ r, b−a P (r, s) := s − b , r < s ≤ b, b−a where r, s ∈ [a, b]. Then
Z b 1 f (s1 )ds1 θ1,n := f (x) − (b − a) a Z b Z b n−2 X f (k) (b) − f (k) (a) · ··· − b−a k=0 | a {z a}
(k+1)th−integral
· · · dsk+1 = =: θ2,n .
Z
b
a
···
Z
b
P (x, s1 )
a
n−1 Y i=1
P (x, s1 )
k Y
i=1
P (si , si+1 ) · ds1 ds2
P (si , si+1 )f (n) (sn )ds1 ds2 · · · dsn (2.4)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Advanced Univariate Ostrowski Type Inequalities −1 Q
Here and later we make the conventions that
k=0
• = 0,
0 Q
i=1
7
• = 1.
Proof. Applying integration by parts twice we obtain Montgomery identity ([188], Ch. XVII, p. 565) Z b Z b 1 f (s1 )ds1 + P (x, s1 )f 0 (s1 )ds1 . f (x) = b−a a a Doing the same for the derivative of f we have Z b Z b 1 0 0 f (s1 ) = f (s2 )ds2 + P (s1 , s2 )f 00 (s2 )ds2 . b−a a a That is
f (b) − f (a) f (s1 ) = + b−a 0
Similarly for f 00 we get
f 00 (s2 ) = And in general we obtain f (n−1) (sn−1 ) =
Z
f 0 (b) − f 0 (a) + b−a
b
P (s1 , s2 )f 00 (s2 )ds2 .
a
Z
b
P (s2 , s3 )f 000 (s3 )ds3 .
a
f (n−2) (b) − f (n−2) (a) + b−a
Z
b
P (sn−1 , sn )f (n) (sn )dsn . a
We observe that Z b Z b f (b) − f (a) P (x, s1 )ds1 P (x, s1 )f 0 (s1 )ds1 = b−a a a Z bZ b + P (x, s1 )P (s1 , s2 )f 00 (s2 )ds2 ds1 . a
That is f (x) =
Z b Z b f (b) − f (a) 1 f (s1 )ds1 + P (x, s1 )ds1 b−a a b−a a Z bZ b + P (x, s1 )P (s1 , s2 )f 00 (s2 )ds2 ds1 . a
Next we see that Z bZ b a
a
a
P (x, s1 )P (s1 , s2 )f 00 (s2 )ds2 ds1
a
Z bZ b f 0 (b) − f 0 (a) P (x, s1 )P (s1 , s2 )ds2 ds1 = b−a a a Z bZ bZ b + P (x, s1 )P (s1 , s2 )P (s2 , s3 )f 000 (s3 )ds3 ds2 ds1 .
a
a
a
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
8
Hence we get f (x) =
Z b Z b f (b) − f (a) 1 P (x, s1 )ds1 f (s1 )ds1 + b−a a b−a a 0 Z bZ b f (b) − f 0 (a) + P (x, s1 )P (s1 , s2 )ds2 ds1 b−a a a Z bZ bZ b + P (x, s1 )P (s1 , s2 )P (s2 , s3 )f 000 (s3 )ds3 ds2 ds1 . a
a
a
Therefore in general we derive f (x)
Z b Z b 1 f (b) − f (a) f (s1 )ds1 + P (x, s1 )ds1 b−a a b−a a Z bZ b 0 f (b) − f 0 (a) + P (x, s1 )P (s1 , s2 )ds2 ds1 b−a a a 00 Z Z bZ b b f (b) − f 00 (a) P (x, s1 )P (s1 , s2 )P (s2 , s3 )ds3 ds2 ds1 + b−a a a a 000 Z b Z b Z b Z b f (b) − f 000 (a) P (x, s1 )P (s1 , s2 )P (s2 , s3 )P (s3 , s4 )ds4 · · · ds1 + b−a a a a a (n−2) Z b Z b n−2 Y f (b) − f (n−2) (a) + ···+ P (x, s1 ) P (si , si+1 )dsn−1 · · · ds1 ··· b−a i=1 | a {z a} Z b Z b n−1 Y + ··· P (x, s1 ) P (si , si+1 )f (n) (sn )dsn · · · ds1 . a a i=1 | {z }
=
nth−integral
The last equality is written briefly as follows Z b n−2 X f (k) (b) − f (k) (a) 1 f (x) = f (s1 )ds1 + b−a a b−a k=0 Z b Z b ··· · a | {z a}
P (x, s1 )
(k+1)th−integral
+
Z
b
a
···
Z
b
P (x, s1 )
a
So we have proved (2.4).
n−1 Y i=1
P (si , si+1 )ds1 ds2 · · · dsk+1 i=1 k Y
P (si , si+1 )f (n) (sn )ds1 ds2 · · · dsn .
(2.5)
A special very common case follows. Corollary 2.5. Under the assumptions and notations of Theorem 2.4, additionally assume that f (k) (a) = f (k) (b),
k = 0, 1, . . . , n − 2; when n ≥ 2.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Advanced Univariate Ostrowski Type Inequalities
9
Then we get θ1,n
Z b 1 = f (x) − f (s1 )ds1 b−a a Z b Z b n−1 Y = ··· P (x, s1 ) · P (si , si+1 ) · f (n) (sn )ds1 · · · dsn = θ2,n , (2.6) a
a
i=1
for n ∈ N, x ∈ [a, b]. Proof.
Directly from (2.4).
Next we generalize Montgomery’s identity for Riemann–Stieltjes integrals (see also in [102] a related but different result). Proposition 2.6. Let f : [a, b] → R be differentiable on [a, b]. The derivative f 0 : [a, b] → R is integrable on [a, b]. Let also g : [a, b] → R be of bounded variation, such that g(a) 6= g(b). Let x ∈ [a, b]. We define g(t) − g(a) , a ≤ t ≤ x, g(b) − g(a) P (g(x), g(t)) = g(t) − g(b) , x < t ≤ b. g(b) − g(a) Then
f (x) =
1 (g(b) − g(a))
Z
b
f (t)dg(t) + a
Z
b
P (g(x), g(t))f 0 (t)dt.
(2.7)
a
Proof. Here f is continuous on [a, b], and g of bounded variation on [a, b]. Then f is Riemann–Stieltjes integrable with respect to g, and it holds Z b Z b f (x)dg(x) + g(x)df (x) = f (b)g(b) − f (a)g(a). a
a
Also it holds that g is Riemann–Stieltjes integrable with respect to f . Here f is differentiable on [a, b]. By the above we obtain Z x Z x (g(t) − g(a))f 0 (t)dt = (g(t) − g(a))df (t) a a x Z x = f (t)(g(t) − g(a)) − f (t)dg(t) a a Z x = f (x)(g(x) − g(a)) − f (t)dg(t). a
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
10
Also we have Z
b x
(g(t) − g(b))f 0 (t)dt =
Z
b x
(g(t) − g(b))df (t)
b Z b f (t)dg(t) = (g(t) − g(b))f (t) − x x Z b = (g(b) − g(x))f (x) − f (t)dg(t). x
Adding the above equalities we get Z b Z x (g(t) − g(b))f 0 (t)dt (g(t) − g(a))f 0 (t)dt + x
a
= f (x)(g(b) − g(a)) −
Z
b
f (t)dg(t). a
Hence we derive that f (x) =
Z b 1 1 f (t)dg(t) + (g(b) − g(a)) a (g(b) − g(a)) "Z # Z b x 0 0 · (g(t) − g(b))f (t)dt . (g(t) − g(a))f (t)dt + x
a
So we have established (2.7).
For twice differentiable functions we obtain the following general Montgomery type identity. Proposition 2.7. Let f : [a, b] → R be twice differentiable on [a, b]. The second derivative f 00 : [a, b] → R is integrable on [a, b]. Let also g : [a, b] → R be of bounded variation, such that g(a) 6= g(b). Let x ∈ [a, b]. The kernel P is defined as in Proposition 2.6. Then Z b 1 1 f (t)dg(t) + f (x) = (g(b) − g(a)) a (g(b) − g(a)) ! ! Z Z b
·
+
Proof.
a
Z
b a
b
P (g(x), g(t)) · dt
Z
b a
·
f 0 (t1 )dg(t1 )
a
P (g(x), g(t)) · P (g(t), g(t1 )) · f 00 (t1 )dt1 dt.
Apply (2.7) for f 0 , we get Z b Z b 1 f 0 (t1 )dg(t1 ) + P (g(t), g(t1 ))f 00 (t1 )dt1 . f 0 (t) = (g(b) − g(a)) a a
(2.8)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Advanced Univariate Ostrowski Type Inequalities
11
Therefore it holds Z b Z b P (g(x), g(t)) P (g(x), g(t)) · f 0 (t)dt = a a " # Z b Z b 1 · f 0 (t1 )dg(t1 ) + P (g(t), g(t1 ))f 00 (t1 )dt1 dt (g(b) − g(a)) a a ! Z b Z b 1 0 · P (g(x), g(t)) · = f (t1 )dg(t1 ) dt (g(b) − g(a)) a a ! Z b Z b P (g(x), g(t)) · + P (g(t), g(t1 ))f 00 (t1 )dt1 dt a
a
! ! Z b 1 0 = · P (g(x), g(t)) · dt · f (t1 )dg(t1 ) (g(b) − g(a)) a a Z bZ b + P (g(x), g(t)) · P (g(t), g(t1 ))f 00 (t1 )dt1 dt. Z
a
b
a
Using the last equality along with (2.7), we have proved (2.8).
At the greatest generality it holds Theorem 2.8. Let f : [a, b] → R be n-times differentiable on [a, b], n ∈ N. The nth derivative f (n) : [a, b] → R is integrable on [a, b]. Let also g : [a, b] → R be of bounded variation, such that g(a) 6= g(b). Let x ∈ [a, b]. The kernel P is defined as in Proposition 2.6. Then Z b 1 1 · f (s1 )dg(s1 ) + f (x) = (g(b) − g(a)) a (g(b) − g(a)) ! Z n−2 b X · f (k+1) (s1 )dg(s1 ) a
k=0
Z b Z b ··· · a | {z a}
P (g(x), g(s1 ))
(k+1)th−integral
+
Proof.
Z
b
a
···
Z
b
P (g(x), g(s1 ))
a
n−1 Y i=1
P (g(si ), g(si+1 )) · ds1 ds2 · · · dsk+1 i=1 k Y
P (g(si ), g(si+1 ))f (n) (sn )ds1 · · · dsn . (2.9)
Similar to Proposition 2.7.
Another important special case follows. Corollary 2.9. All as in Theorem 2.8. Additionally suppose that Z b f (k+1) (s1 )dg(s1 ) = 0, k = 0, 1, . . . , n − 2; when n ≥ 2. a
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
12
Then M1,n
Z b 1 := f (x) − · f (s1 )dg(s1 ) (g(b) − g(a)) a Z b Z b n−1 Y = ··· P (g(x), g(s1 )) P (g(si ), g(si+1 ))f (n) (sn ) · ds1 · · · dsn a
a
i=1
=: M2,n ,
(2.10)
for n ∈ N, x ∈ [a, b]. Proof.
From Theorem 2.8.
A similar multivariate result comes next. Theorem 2.10. Let Q be a compact convex subset of Rk , k ≥ 2; x := (x1 , . . . , xk ) and O := (0, . . . , 0) ∈ Q. Let f ∈ C 1 (Q) and suppose that all partial derivatives of f of order one are coordinatewise absolutely continuous functions. Then Z 1 Z 1Z 1 k X ∂f (t1 t2 x)dt1 dt2 f (x) = f (t1 x)dt1 + xj t1 ∂xj 0 0 0 j=1 +
k X k X
xi xj
i=1 j=1
Z
Z
1
0
1
t21 t2
0
∂ 2 f (t1 t2 x) dt1 dt2 . ∂xi ∂xj
(2.11)
Note. Parameterized integrals as above are met in Radon Transform Theory and in Computerized Tomography. Proof of Theorem 2.10. Set gx (t) := f (tx), x := (x1 , . . . , xk ), x ∈ Rk , k ≥ 2, 0 ≤ t ≤ 1. Then gx (0) = f (0), gx (1) = f (x). Integrating by parts we obtain that Z 1 Z 1 tgx0 (t)dt. gx (t)dt + gx (1) = 0
0
Here we have
gx0 (t) =
k X
xi
i=1
Therefore f (x) = So that f (x) =
Z
1
f (tx)dt + 0
1
f (t1 x)dt1 + 0
Next apply (2.12) for ∂f (x) = ∂xj
Z
Z
1
t 0
k X i=1
∂f ∂xj
Z
1 0
to obtain
∂f (tx). ∂xi
xi
k X
∂f (tx) xi ∂xi i=1
Z
1
t1 0
!
∂f (t1 x) ∂xi
dt.
dt1 .
Z 1 k X ∂f ∂ 2 f (t2 x) (t2 x)dt2 + xi dt2 . t2 ∂xj ∂xi ∂xj 0 i=1
(2.12)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Advanced Univariate Ostrowski Type Inequalities
Book˙Adv˙Ineq
13
Hence t1
∂f (t1 x) = t1 ∂xj
Z
Z 1 k X ∂f ∂ 2 f (t2 t1 x) t21 xi (t2 t1 x)dt2 + dt2 . t2 ∂xj ∂xi ∂xj 0 i=1
1 0
Furthermore it holds Z 1 Z 1 Z 1 ∂f (t1 x) ∂f dt1 = (t2 t1 x)dt2 dt1 t1 t1 ∂xj 0 0 0 ∂xj Z 1 Z 1X k ∂ 2 f (t1 t2 x) t21 xi t2 dt2 dt1 . + ∂xi ∂xj 0 0 i=1
(2.13)
Consequently by (2.13) we obtain Z 1 Z 1Z 1 k k X X ∂f (t1 x) ∂f xj t1 xj dt1 = t1 (t1 t2 x)dt2 dt1 ∂x ∂x j j 0 0 0 j=1 j=1 +
k X
xj
j=1
=
k X
xj
j=1
+
Z
1 0
Z
k X k X
0
k X
t21 xi
i=1
1Z
1
t1 0
xj xi
j=1 i=1
Z
1 0
Z
1
t2 0
∂ 2 f (t1 t2 x) dt2 dt1 ∂xi ∂xj
∂f (t1 t2 x)dt2 dt1 ∂xj Z
1 0
t21 t2
∂ 2 f (t1 t2 x) dt2 dt1 . ∂xi ∂xj
At last combining (2.12) and (2.14) we have established (2.11).
(2.14)
Corollary 2.11. All as in Theorem 2.10. Additionally suppose that Z 1Z 1 ∂f (t1 t2 x)dt1 dt2 = 0, all j = 1, . . . , k. t1 ∂x j 0 0 Then f (x) −
Z
1
f (t1 x)dt1 = 0
k X k X i=1 j=1
xi xj
Z
1 0
Z
1 0
t21 t2
∂ 2 f (t1 t2 x) dt1 dt2 , ∂xi ∂xj
(2.15)
for any x ∈ Q; 0 ∈ Q. We will use the next Lemma 2.12. Let f : [a, b] → R be 3-times differentiable on [a, b]. Assume that f 000 is integrable on [a, b]. Let x ∈ [a, b]. Define s−a , a ≤ s ≤ r, b−a P (r, s) := s−b , r < s ≤ b; r, s ∈ [a, b]. b−a
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
14
Then θ1,3 := f (x) −
1 (b − a)
Z
b a
f (s1 )ds1 −
f (b) − f (a) b−a
Z
b
P (x, s1 )ds1 a
Z bZ b f 0 (b) − f 0 (a) P (x, s1 )P (s1 , s2 )ds1 ds2 − b−a a a Z bZ bZ b = P (x, s1 )P (s1 , s2 )P (s2 , s3 )f 000 (s3 )ds1 ds2 ds3 =: θ2,3 . (2.16)
a
Proof.
2.3
a
a
By Theorem 2.4.
Main Results
We present an Ostrowski type inequality. Theorem 2.13. Let f : [a, b] → R be 3-times differentiable on [a, b]. Assume that f 000 is bounded on [a, b]. Let x ∈ [a, b]. Then Z b a+b 1 f (b) − f (a) x− f (s1 )ds1 − f (x) − (b − a) a b−a 2 0 f (b) − f 0 (a) a2 + b2 + 4ab − · x2 − (a + b)x + 2(b − a) 6 000 kf k∞ · A, (2.17) ≤ (b − a)3 where 1 1 1 A := abx4 − a2 b3 x + a3 bx2 − ab2 x3 − a3 b2 x 3 3 3 1 1 1 3 2 + ab x + a2 b2 x2 − a2 bx3 − ax5 − bx5 3 2 2 1 6 3 2 4 3 2 4 1 2 4 2 3 3 + x + a x + b x + b a − a x 6 4 4 3 3 2 3 3 1 3 3 5 4 2 5 4 2 − b x − b a + a x + b x 3 3 12 12 1 4 2 2 5 2 5 1 5 1 a6 b6 . + b a − ba − ab − a x − b5 x + + 3 15 15 6 6 20 20
(2.18)
Inequality (2.17) is attained by f (x) = (x − a)3 + (b − x)3 , in that case both sides of the inequality equal zero.
(2.19)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Advanced Univariate Ostrowski Type Inequalities
15
Since kf 000 k∞ < +∞, by (2.16) we obtain Z bZ bZ b |θ1,3 | = |θ2,3 | ≤ kf 000 k∞ · |P (x, s1 )| · |P (s1 , s2 )| · |P (s2 , s3 )|ds1 ds2 ds3 a a a ! ! Z b Z b Z b kf 000 k∞ |p(s2 , s3 )|ds3 ds2 ds1 , |p(s1 , s2 )| · |p(x, s1 )| = (b − a)3 a a a
Proof.
(2.20) where p(r, s) := We see that
Z
s − a, s − b,
a ≤ s ≤ r, r < s ≤ b; r, s ∈ [a, b].
b a
|p(s2 , s3 )|ds3 =
(s2 − a)2 + (b − s2 )2 . 2
Furthermore we find with some calculations that Z b (s2 − a)2 + (b − s2 )2 |p(s1 , s2 )| ds2 2 a # " 4 2 1 a+b a+b 7 2 4 + 3(b − a) s1 − + (b − a) . = 6 s1 − 12 2 2 8 To finish calculating the integral in (2.20) we find " 4 Z b 1 a+b |p(x, s1 )| 6 s1 − 12 a 2 # 2 a+b 7 2 4 +3(b − a) s1 − + (b − a) ds1 2 8 " Z x 4 1 a+b (s1 − a) 6 s1 − = 12 a 2 # 2 7 a+b 4 2 + (b − a) ds1 +3(b − a) s1 − 2 8 " 4 Z b 1 a+b + (b − s1 ) 6 s1 − 12 x 2 # 2 7 a+b 2 4 + (b − a) ds1 +3(b − a) s1 − 2 8 = A. Consequently we have |θ1,3 | ≤
kf 000 k∞ · A. (b − a)3
(2.21)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
16
Next we find that
Z
b
P (x, s1 )ds1 = a
x−
a+b 2
.
(2.22)
Furthermore we obtain Z bZ b P (x, s1 )P (s1 , s2 )ds1 ds2 a a (5a2 + 5b2 + 8ab) 1 2 2 (x + (a + b) ) − (a + b)x − . = 2 6 Therefore we derive Z b 1 θ1,3 = f (x) − f (s1 )ds1 (b − a) a 0 a+b f (b) − f 0 (a) f (b) − f (a) x− − − b−a 2 2(b − a) 2 2 (5a + 5b + 8ab) · (x2 + (a + b)2 ) − (a + b)x − . 6 So we have established inequality (2.17).
(2.23)
(2.24)
We have an interesting special case. Corollary 2.14. All as in Theorem 2.13. Additionally suppose that f (a) = f (b), and f 0 (a) = f 0 (b). Then Z b 1 kf 000 k∞ · A. (2.25) f (s1 )ds1 − f (x) ≤ (b − a)3 (b − a) a Other Ostrowski type results follow:
Theorem 2.15. All as in Theorem 2.4. Additionally assume that kf (n) k∞ < +∞. Then Z b Z b n−1 Y (n) |θ1,n | ≤ kf k∞ · ··· |P (x, s1 )| · |P (si , si+1 )|ds1 · · · dsn . (2.26) a
a
i=1
Theorem 2.16. All as in Proposition 2.6. Additionally suppose that kf 0 k∞ < +∞. Then Z b Z b 1 0 |P (g(x), g(t))|dt. (2.27) f (t)dg(t) − f (x) ≤ kf k∞ · (g(b) − g(a)) a a
Theorem 2.17. All as in Proposition 2.7. Additionally assume that kf 00 k∞ < +∞. Then Z b 1 1 f (x) − f (t)dg(t) − (g(b) − g(a)) a (g(b) − g(a)) ! ! Z b Z b 0 · P (g(x), g(t))dt · f (t1 )dg(t1 ) a
00
≤ kf k∞ ·
a
Z
a
bZ
b
a
!
|P (g(x), g(t))| · |P (g(t), g(t1 ))| · dt1 dt .
(2.28)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Advanced Univariate Ostrowski Type Inequalities
17
Theorem 2.18. All as in Theorem 2.8. Additionally suppose that kf (n) k∞ < +∞. Then ! Z b n−2 X Z b 1 1 (k+1) f (s1 )dg(s1 ) f (s1 )dg(s1 ) − · f (x) − (g(b) − g(a)) a (g(b) − g(a)) a k=0 Z b Z b k Y · ··· P (g(x), g(s1 )) · P (g(si ), g(si+1 ))ds1 · · · dsk+1 a a i=1 ! Z b Z b n−1 Y ≤ kf (n) k∞ · |P (g(si ), g(si+1 ))|ds1 · · · dsn . ··· |P (g(x), g(s1 ))| · a
a
i=1
(2.29)
Theorem 2.19. All as 2.10. Additionally assume that
in2Theorem
∂ f
for all i, j = 1, . . . , k. γij :=
∂xi ∂xj < +∞, ∞ Then Z 1 Z 1Z 1 k X ∂f f (x) − (t t x)dt dt f (t x)dt − x t 1 2 1 2 1 1 j 1 ∂xj 0 0 0 j=1 k k 1 XX |xi | |xj | · γij . ≤ 6 i=1 j=1
(2.30)
Next we present Lp , p > 1, Ostrowski type results.
Theorem 2.20. All as in Theorem 2.4. Additionally assume that kf (n) kp < +∞, p > 1. Here p, q : p1 + 1q = 1. Then ! Z b Z b n−2 Y (n) |θ1,n | ≤ kf kp · ··· |P (x, s1 )| · |P (si , si+1 )| a
a
i=1
· kP (sn−1 , •)kq ds1 ds2 · · · dsn−1 .
(2.31)
0
Theorem 2.21. All as in Proposition 2.6. Additionally assume that kf kp < +∞, p > 1. Here p, q : p1 + q1 = 1. Then Z b 1 f (t)dg(t) ≤ kf 0 kp · kP (g(x), g(t))kq,t . (2.32) f (x) − (g(b) − g(a)) a Here k · kq,t means integration with respect to t. Theorem 2.22. All as in Proposition 2.8. Additionally assume that kf 00 kp < +∞, p > 1. Here p, q : 1p + q1 = 1. Then Z b 1 1 f (x) − f (t)dg(t) − (g(b) − g(a)) a (g(b) − g(a)) ! Z ! Z b b 0 · P (g(x), g(t))dt f (t1 )dg(t1 ) a
≤ kf 00 kp ·
a
Z
b
a
|P (g(x), g(t))| · kP (g(t), g(t1 )kq,t1 · dt.
(2.33)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
18
Theorem 2.23. All as in Theorem 2.8. Additionally suppose that kf (n) kp < +∞, p > 1. Here p, q : p1 + q1 = 1. Then Z b 1 1 f (x) − f (s1 )dg(s1 ) − (g(b) − g(a)) a (g(b) − g(a)) ! Z b Z b n−2 X Z b (k+1) · f (s1 )dg(s1 ) · ··· P (g(x), g(s1 )) k=0
a
a
! k Y · P (g(si ), g(si+1 )) · ds1 · · · dsk+1 i=1
≤ kf (n) kp ·
Z
b
a
···
Z
b
a
|P (g(x), g(s1 ))| ·
a
n−2 Y i=1
|P (g(si ), g(si+1 ))|
· kP (g(sn−1 ), g(sn ))kq,sn · ds1 ds2 · · · dsn−1 .
! (2.34)
Finally we present L1 Ostrowski type results. Theorem 2.24. All as in Theorem 2.4. Additionally suppose that kf (n) k1 < +∞. Then ! Z b Z b n−2 Y (n) |θ1,n | ≤ kf k1 · |P (si , si+1 )| |P (x, s1 )| · ··· a
a
i=1
· kP (sn−1 , •)k∞ · ds1 ds2 · · · dsn−1 .
(2.35)
The interested reader can consult with [10], which is a further study on this basic chapter. Theorem 2.25. All as in Proposition 2.6. Additionally assume that kf 0 k1 < +∞. Then Z b 1 f (t)dg(t) ≤ kf 0 k1 · kP (g(x), g(t))k∞,t . (2.36) f (x) − (g(b) − g(a)) a
Here k · k∞,t is taken with respect to t.
Theorem 2.26. All as in Proposition 2.8. Additionally assume that kf 00 k1 < +∞. Then Z b 1 1 f (x) − f (t)dg(t) − (g(b) − g(a)) a (g(b) − g(a)) ! ! Z b Z b 0 · P (g(x), g(t))dt · f (t1 )dg(t1 ) a
00
≤ kf k1 ·
a
Z
b
a
!
|P (g(x), g(t))| · kP (g(t), g(t1 ))k∞,t1 · dt .
(2.37)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Advanced Univariate Ostrowski Type Inequalities
19
Finally we give Theorem 2.27. All as in Theorem 2.8. Additionally suppose that kf (n) k1 < +∞. Then Z b 1 1 f (x) − f (s1 )dg(s1 ) − · (g(b) − g(a)) a (g(b) − g(a)) ! Z b Z b n−2 X Z b (k+1) P (g(x), g(s1 )) ··· f (s1 )dg(s1 ) · · k=0
·
k Y
i=1
a
a
! P (g(si ), g(si+1 ))ds1 ds2 · · · dsk+1
≤ kf (n) k1 ·
Z
b
a
···
Z
b
a
|P (g(x), g(s1 ))| ·
a
n−2 Y i=1
·kP (g(sn−1 ), g(sn ))k∞,sn · ds1 · · · dsn−1 ) .
|P (g(si ), g(si+1 ))|
! (2.38)
This page intentionally left blank
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Chapter 3
Higher Order Ostrowski Inequalities
We present a generalization of Ostrowski inequality for higher order derivatives, by using a generalized Euler-type identity. Some of the produced inequalities are sharp, namely attained by basic functions. The rest of the estimates are tight. We give applications to trapezoidal and midpoint rules. Estimates are given with respect to L∞ norm. This treatement is based on [35].
3.1
Introduction
We mention as motivation to our work the great Ostrowski inequality [196]: " 2 # Z b x − a+b 1 1 2 (b − a)kf 0 k∞ , (3.1) + f (t)dt ≤ f (x) − b−a a 4 (b − a)2
where f ∈ C 0 ([a, b]), x ∈ [a, b], which is a sharp inequality, see [16]. Other motivations come from [10], [24], [16], [98] and [121]. We use here the sequence {Bk (t), k ≥ 0} of Bernoulli polynomials which is uniquely determined by the following identities: Bk0 (t) = kBk−1 (t),
k ≥ 1,
and
Bk (t + 1) − Bk (t) = ktk−1 ,
B0 (t) = 1 k ≥ 0.
The values Bk = Bk (0), k ≥ 0 are the known Bernoulli numbers. We need to mention 1 1 B0 (t) = 1, B1 (t) = t − , B2 (t) = t2 − t + , 2 6 3 1 1 B3 (t) = t3 − t2 + t, B4 (t) = t4 − 2t3 + t2 − , 2 2 30 5 1 5 t2 1 5 . B5 (t) = t5 − t4 + t3 − t, and B6 (t) = t6 − 3t5 + t4 − + 2 3 6 2 2 42 ∗ Let {Bk (t), k ≥ 0} be a sequence of periodic functions of period 1, related to Bernoulli polynomials as Bk∗ (t) = Bk (t),
Bk∗ (t + 1) = Bk∗ (t),
0 ≤ t < 1, 21
t ∈ R.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
22
We have that B0∗ (t) = 1, B1∗ is a discontinuous function with a jump of −1 at each integer, while Bk∗ , k ≥ 2 are continuous functions. Notice that Bk (0) = Bk (1) = Bk , k ≥ 2. We use the following result. Theorem 3.1. Let f : [a, b] → R be such that f (n−1) , n ≥ 1, is a continuous function and f (n) (x) exists and is finite for all but a countable set of x in (a, b) and that f (n) ∈ L1 ([a, b]). Then for every x ∈ [a, b] we have Z b n−1 X (b − a)k−1 x − a 1 Bk f (t)dt + [f (k−1) (b) − f (k−1) (a)] f (x) = b−a a k! b−a k=1 " # n−1 Z x−a (b − a) x − t + Bn − Bn∗ f (n) (t)dt. (3.2) n! b−a b−a [a,b]
The sum in (3.2) when n = 1 is zero.
Proof. By using Theorem 2 of [98], Exercise 18.41(d), p. 299 in [158] and Problem 14(c), p. 264 in [224]. And that f (n−1) as implied absolutely continuous it is also of bounded variation. If f (n−1) is just absolutely continuous then (3.2) is valid again. Formula (3.2) is a generalized Euler type identity, see also [171]. We set Z b 1 f (t)dt ∆n (x) := f (x) − b−a a n−1 X (b − a)k−1 x − a − [f (k−1) (b) − f (k−1) (a)], x ∈ [a, b]. (3.3) Bk k! b−a k=1
We have by (3.2) that Z (b − a)n−1 x−a x−t ∆n (x) = Bn − Bn∗ f (n) (t)dt. (3.4) n! b − a b − a [a,b] In this chapter we give sharp, namely attained, upper bounds for |∆4 (x)| and tight upper bounds for |∆n (x)|, n ≥ 5, x ∈ [a, b], with respect to L∞ norm. That is generalizing (3.1) for higher order derivatives. High computational difficulties in this direction prevent us for shoming sharpness for n ≥ 5 cases. 3.2
Main Results
We give Theorem 3.2. Let f : [a, b] → R be such that f (n−1) , n ≥ 1, is a continuous function and f (n) (x) exists and is finite for all but a countable set of x in (a, b) and that f (n) ∈ L∞ ([a, b]). Then for every x ∈ [a, b] we have ! Z b (b − a)n−1 x−a ∗ x−t |∆n (x)| ≤ − Bn Bn dt kf (n) k∞ . (3.5) n! b−a b−a a
Proof.
By Theorem 3.1.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Higher Order Ostrowski Inequalities
23
Performing the change of variable method in the integral of (3.5) we derive Corollary 3.3. All assumptions as in Theorem 3.2. Then for every x ∈ [a, b] we have Z 1 (b − a)n x − a (n) (3.6) |∆n (x)| ≤ Bn (t) − Bn b − a dt kf k∞ , n ≥ 1. n! 0
Note. Inequality (3.6) appeared first as Theorem 7, p. 350, in [98], wrongly under the sole assumption of f (n) ∈ L∞ ([a, b]). Also in the rest of [98] the complete assumptions of our Theorem 3.2 are missing, whenever it applies. Using Cauchy–Schwartz inequality we find that 2 !1/2 Z 1 Z 1 x−a Bn (t) − Bn x − a dt ≤ Bn (t) − Bn dt b−a b−a 0 0 s x−a (n!)2 2 |B2n | + Bn , n ≥ 1, (3.7) = (2n)! b−a the last comes by [98], p. 352. We give Corollary 3.4. All assumptions as in Theorem 3.2. Then for every x ∈ [a, b] we have s ! (n!)2 x − a (b − a)n kf (n) k∞ , n ≥ 1. (3.8) |B2n | + Bn2 |∆n (x)| ≤ n! (2n)! b−a Here we elaborate on (3.6) and (3.8). We introduce the parameter λ := We see that
x−a , b−a
a ≤ x ≤ b.
(3.9)
λ = 0 iff x = a, λ = 1 iff x = b, and λ=
1 2
iff x =
a+b . 2
Consider p4 (t) := B4 (t) − B4 (λ) = t4 − 2t3 + t2 − λ4 + 2λ3 − λ2 .
(3.10)
We need to compute I4 (λ) :=
Z
1 0
|p4 (t)|dt,
0 ≤ λ ≤ 1.
(3.11)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
24
Lemma 3.5. We find 14 3 1 16λ5 4 2 5 − 7λ + 3 λ − λ + 30 , I4 (λ) = 5 3 − 16λ + 9λ4 − 26λ + 3λ2 − 1 , 5 3 10
0 ≤ λ ≤ 1/2,
(3.12)
1/2 ≤ λ ≤ 1,
which is continuous in λ ∈ [0, 1]. We obtain
1 , I4 (0) = I4 (1) = 30 7 1 = I4 . 2 240
(3.13)
Proof. Here all calculations are done by Mathematica 4. The equation p4 (t) = 0 has four real roots, p 1 r1 = 1 − λ, r2 = λ, r3 = 1 − 1 + 4λ − 4λ2 2 and
r4 = We find the following orders:
p 1 1 + 1 + 4λ − 4λ2 . 2
i) if 0 ≤ λ ≤ 1/2, then r3 ≤ 0 ≤ r 2 ≤ r 1 ≤ 1 ≤ r 4 , and ii) if 1/2 ≤ λ ≤ 1, then r3 ≤ 0 ≤ r 1 ≤ r 2 ≤ 1 ≤ r 4 . So we have p4 (t) = (t − r1 )(t − r2 )(t − r3 )(t − r4 ), t ∈ [0, 1]. We easily derive that when 0 ≤ λ ≤ 1/2, p4 (t) is greater equal zero over [λ, 1 − λ] and smaller equal zero over [0, λ] and [1 − λ, 1]. Similarly when 1/2 ≤ λ ≤ 1 we get that p4 (t) ≥ 0 over [1 − λ, λ] and p4 (t) ≤ 0 over [0, 1 − λ] and [λ, 1]. Therefore when 0 ≤ λ ≤ 1/2 we have Z 0 Z 1−λ Z 1−λ I4 (λ) = p4 (t)dt + p4 (t)dt + p4 (t)dt, λ
λ
while when 1/2 ≤ λ ≤ 1 we have Z Z 0 p4 (t)dt + I4 (λ) = 1−λ
1
λ
p4 (t)dt + 1−λ
proving (3.12) after the computations are done.
Z
λ
p4 (t)dt, 1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Higher Order Ostrowski Inequalities
Book˙Adv˙Ineq
25
Using basic calculus we obtain Lemma 3.6.
1 3 5 min I4 (λ) = I4 = I4 = = 0.01953125, λ∈[0,1] 4 4 256
(3.14)
and max I4 (λ) = I4 (0) = I4 (1) =
λ∈[0,1]
1 = 0.033333 . 30
(3.15)
Consequently by Lemmas 3.5 and 3.6 we get Theorem 3.7. Assumptions as in Theorem 3.2, case of n = 4. For every x ∈ [a, b] it holds (b − a)4 I4 (λ)kf (4) k∞ , (3.16) |∆4 (x)| ≤ 24 where I4 (λ) is given by (3.12) with λ = x−a b−a . Furthermore we have that |∆4 (x)| ≤
(b − a)4 (4) kf k∞ , 720
∀x ∈ [a, b].
(3.17)
Optimality is achieved in Theorem 3.8. Assumptions as in Theorem 3.2, case of n = 4. Inequality (3.17) is sharp, namely it is attained when x = a, b by the functions (t − a)4 and (t − b)4 . Proof.
We have
∆4 (a) = ∆4 (b) =
Z b (b − a) 0 1 f (a) + f (b) − f (t)dt. (3.18) (f (b)−f 0 (a))− 2 12 b−a a
So by (3.17) we have
|∆4 (a)| = |∆4 (b)| ≤
(b − a)4 (4) kf k∞ . 720
(3.19)
Let f (t) = (t − a)4 or f (t) = (t − b)4 , then
(b − a)4 = R.H.S.(3.19), 30 proving that (3.19) is attained. That is proving (3.17) sharp. |∆4 (a)| = |∆4 (b)| =
The trapezoid and midpoint inequalities follow. Corollary 3.9. Assumptions as in Theorem 3.2, case of n = 4. It holds Z b f (a) + f (b) (b − a) 1 0 0 − (f (b) − f (a)) − f (t)dt 2 12 b−a a
(b − a)4 (4) kf k∞ , 720 the last inequality is attained by (t − a)4 and (t − b)4 , that is sharp. ≤
(3.20)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
26
Furthermore we obtain Z b a + b (b − a) 1 f (t)dt (f 0 (b) − f 0 (a)) − + f 2 24 b−a a ≤
7 (b − a)4 kf (4) k∞ . 5760
(3.21)
Remark 3.10. We do obtain the trapezoidal formula (f (a) + f (b)) (b − a) 0 − (f (b) − f 0 (a)) 2 12 Z b 1 (b − a)3 000 f (t)dt. (3.22) (f (b) − f 000 (a)) − + 720 b−a a
∆5 (a) = ∆5 (b) = ∆6 (a) = ∆6 (b) =
We also find the midpoint formula a+b (b − a) 0 a+b a+b ∆5 = ∆6 =f + (f (b) − f 0 (a)) 2 2 2 24 Z b 7(b − a)3 000 1 − (f (b) − f 000 (a)) − f (t)dt. 5760 b−a a
(3.23)
Using (3.8) and Mathematica 7 we get
Theorem 3.11. Assumptions as in Theorem 3.2, cases of n = 5, 6. It holds 5 (a)|, |∆ (b − a)5 ≤ √ kf (5) k∞ , (3.24) a + b 144 2310 ∆5 2
and
1 |∆6 (a)| ≤ 1440 with
r
101 (b − a)6 kf (6) k∞ , 30030
r 1 7081 a + b 6 (6) ≤ ∆6 46080 2145 (b − a) kf k∞ . 2
(3.25)
(3.26)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Chapter 4
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities We develop and establish a general multivariate Euler type identity. Using it we derive general tight multivariate high order Ostrowski type inequalities for the estimate on the error of a multivariate function f evaluated at a point from its average. The estimates are involving only the single partial derivatives of f and are with respect to k · kp , 1 ≤ p ≤ ∞. We give specific applications of the main results to the multidimensional trapezoid and midpoint rules for functions f differentiable up to order 6. We show sharpness of the inequalities for differentiation orders m = 1, 2, 4 and with respect to k · k∞ . This treatment relies on [38]. 4.1
Introduction
We mention as motivation to this chapter the great Ostrowski inequality, see [196], [16], [21], [24]. Theorem 4.1. It holds # " Z b a+b 2 x − 1 1 2 f (x) − f (t)dt ≤ (b − a)kf 0 k∞ , + b−a a 4 (b − a)2
where f ∈ C 1 ([a, b]), x ∈ [a, b], which is a sharp inequality. The author proved also in [17], [21] the following result:
Theorem 4.2. Let f ∈ C n+1 ([a, b]), n ∈ N and x ∈ [a, b] be fixed, such that f (k) (x) = 0, k = 1, . . . , n. Then Z b 1 kf (n+1) k∞ (x − a)n+2 + (b − x)n+2 . · f (y)dy − f (x) ≤ b − a (n + 2)! b−a a
The last inequality is sharp. In particular, when n is odd it is attained by f ∗ (y) := (y − x)n+1 · (b − a), while when n is even the optimal function is f˜(y) := |y − x|n+α · (b − a),
Recently the author in [35], proved 27
α > 1.
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
28
Theorem 4.3. Let f : [a, b] → R be such that f (n−1) , n ≥ 1, is a continuous function and f (n) (x) exists and is finite for all but a countable set of x in (a, b) and that f (n) ∈ L∞ ([a, b]). Then for every x ∈ [a, b] we have ! Z b (b − a)n−1 x − t x − a ∗ B n dt kf (n) k∞ , − Bn |∆n (x)| ≤ n! b − a b − a a
where
∆n (x) := f (x) − −
n−1 X k=1
1 b−a
Z
b
f (t)dt a
x−a (b − a)k−1 Bk [f (k−1) (b) − f (k−1) (a)], x ∈ [a, b], k! b−a
with Bk (t) the Bernoulli polynomial of degree k ≥ 1. Here {Bk∗ (t), k ≥ 0} is the sequence of periodic functions of period 1, related to Bernoulli polynomials as Bk∗ (t) = Bk (t), 0 ≤ t < 1, Bk∗ (t + 1) = Bk∗ (t), t ∈ R. We give also Corollary 4.4. All assumptions as in Theorem 4.3. Then for every x ∈ [a, b] we have Z 1 x − a (b − a)n (n) |∆n (x)| ≤ Bn (t) − Bn b − a dt kf k∞ , n ≥ 1. n! 0
Note. The last inequality appeared first as Theorem 7, p. 350, in [98], wrongly under the sole assumption of f (n) ∈ L∞ ([a, b]). Theorem 4.3 and Corollary 4.4 along with [35], [98] are the greatest motivations for the Euler identity method we use in this chapter. Further motivation is given by (see also [17], [21]) n Q 1 Theorem 4.5. Let f ∈ C [ai , bi ] , where ai < bi ; ai , bi ∈ R, i = 1, . . . , k, i=1
k Q − and let → x 0 := (x01 , . . . , x0k ) ∈ [ai , bi ] be fixed. Then i=1
Z b1 Z bi Z bk 1 → − ··· ··· f (z1 , . . . , zk )dz1 · · · dzk − f ( x 0 ) k Q ai ak (bi − ai ) a1 i=1
k X
(x0i − ai )2 + (bi − x0i )2
∂f . ≤
2(bi − ai ) ∂zi ∞ i=1
The last inequality is sharp, here the optimal function is f ∗ (z1 , . . . , zk ) :=
k X i=1
|zi − x0i |αi , αi > 1.
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
Book˙Adv˙Ineq
29
Theorem 4.5 generalizes Theorem 4.1 to multidimension. We also mention (see [17], [21]) Theorem 4.6. Let Q be a compact and convex subset of Rk , k ≥ 1. Let f ∈ α − C n+1 (Q), n ∈ N and → x 0 ∈ Q be fixed such that all partial derivatives fα := ∂∂zαf , k P where α = (α1 , . . . , αk ), αi ∈ Z+ , i = 1, . . . , k, |α| = αi = j, j = 1, . . . , n fulfill i=1 − fα (→ x 0 ) = 0. ZThen Z 1 Dn+1 (f ) → − → − → − − − − f ( z )d z − f ( x 0 ) ≤ (k→ z −→ x 0 k`1 )n+1 d→ z, Vol(Q) (n + 1)!Vol(Q) Q Q where Dn+1 (f ) := max kfα k∞ α : |α|=n+1
and − − k→ z −→ x 0 k`1 := As a related result we give
k X i=1
|zi − x0i |.
Corollary 4.7. Under the assumptions of Theorem 4.6 we find that Z 1 → − → − → − f ( z )d z − f ( x ) 0 Vol(Q) Q "
n+1 # Z X k
∂ 1 −
|zi − x0i | · f d→ z. ≤
∂zi (n + 1)!Vol(Q) Q ∞ i=1 Furthermore, the last inequality is sharp : when n is odd it is attained by k X f ∗ (z1 , . . . , zk ) := (zi − x0i )n+1 , i=1
while when n is even the optimal function is k X f˜(z1 , . . . , zk ) := |zi − x0i |n+αi , αi > 1. i=1
Theorems 4.5, 4.6 and Corollary 4.7 motivate this chapter in the multivariate setting. Here we develop a very general Multivariate Euler type identity, see Theorems 4.14, 4.15, 4.18, 4.21. Based on this identity we prove some general tight multivariate Ostrowski type inequalities, see Theorems 4.32, 4.33, 4.34, 4.35, 4.60. Then we give lots of specific and important applications of Theorems 4.32, 4.33, 4.34, see Corollaries 4.36–4.59. There are produced many multivariate Ostrowski type inequalities for differentiation orders m = 1, . . . , 6, mostly related to multivariate trapezoid and midpoint rules. When we impose some basic and natural boundary conditions, then inequalities become very simple and elegant, see Corollaries 4.56 – 4.59. The surprising fact there is, that only a very small number of sets of boundary conditions is needed comparely to the higher order of differentiation of the involved functions. At the end we establish sharpness of the inequalities with respect to k · k∞ and for differentiation orders m = 1, 2, 4, see Corollaries 4.61–4.69.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
30
4.2
Book˙Adv˙Ineq
Background
We will use Theorem 4.8 ([171, p. 17]). Let f ∈ C n ([a, b]), n ∈ N, x ∈ [a, b]. Then Z b 1 f (x) = f (t)dt + φn−1 (x) + Rn (x), b−a a where for m ∈ N we call m X (b − a)k−1 x−a [f (k−1) (b) − f (k−1) (a)] φm (x) := Bk k! b−a
(4.1)
k=1
with the convention φ0 (x) = 0, and Z (b − a)n−1 b ∗ x − t x−a Bn Rn (x) := − − Bn f (n) (t)dt. (4.2) n! b − a b − a a Here Bk (x), k ≥ 0, are the Bernoulli polynomials, Bk = Bk (0), k ≥ 0, the Bernoulli numbers, and Bk∗ (x), k ≥ 0, are periodic functions of period one, related to the Bernoulli polynomials as and
Bk∗ (x) = Bk (x), 0 ≤ x < 1,
(4.3)
Bk∗ (x + 1) = Bk∗ (x), x ∈ R.
(4.4)
Some basic properties of Bernoulli polynomials follow (see [1, 23.1]). We have 1 1 B0 (x) = 1, B1 (x) = x − , B2 (x) = x2 − x + , 2 6 and Bk0 (x) = kBk−1 (x), k ∈ N, B0∗
B1∗
Bk (x + 1) − Bk (x) = kx
k−1
(4.5) , k ≥ 0.
(4.6)
Clearly = 1, is a discontinuous function with a jump of −1 at each integer, and Bk∗ , k ≥ 2, is a continuous function. Notice that Bk (0) = Bk (1) = Bk , k ≥ 2. We need also the more general Theorem 4.9 (see [35]). Let f : [a, b] → R be such that f (n−1) , n ≥ 1, is a continuous function and f (n) (x) exists and is finite for all but a countable set of x in (a, b) and that f (n) ∈ L1 ([a, b]). Then for every x ∈ [a, b] we have Z b n−1 X (b − a)k−1 x − a 1 f (x) = [f (k−1) (b) − f (k−1) (a)] Bk f (t)dt + b−a a k! b−a k=1 Z x−a (b − a)n−1 ∗ x−t + Bn − Bn f (n) (t)dt. (4.7) n! b−a b−a [a,b]
The sum in (4.7) when n = 1 is zero. If f (n−1) is just absolutely continuous then (4.7) is valid again. Formula (4.7) is a generalized Euler type identity, see also [98].
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
4.3
31
Main Results
We make `
∂ f Assumption 4.10. Let f and the existing ∂x ` , all ` = 1, . . . , m; j = 1, . . . , n, be j n Q continuous real valued functions on [ai , bi ]; m, n ∈ N, ai , bi ∈ R. i=1
We give
Proposition 4.11. All as in Assumption 4.10 for m = n = 2, xi ∈ [ai , bi ], i = 1, 2. Then ! Z b1 Z b 2 1 f (s1 , s2 )ds1 ds2 + T2 (x2 ) + T1 (x1 , x2 ), f (x1 , x2 ) = (b1 − a1 )(b2 − a2 ) a2 a1 (4.8) where ! Z b1 1 x2 − a 2 T2 := T2 (x2 ) := B1 (f (s1 , b2 ) − f (s1 , a2 ))ds1 (b1 − a1 ) b2 − a 2 a1 # ! 2 Z b 1 Z b2 " (b2 − a2 ) x2 − a 2 ∂ f ∗ x2 − s 2 + B2 − B2 (s1 , s2 ) ds1 ds2 , 2(b1 − a1 ) a1 a2 b2 − a 2 b2 − a 2 ∂x22 (4.9) and T1 := T1 (x1 , x2 ) := B1 (b1 − a1 ) + 2
Z
b1 a1
x1 − a 1 (f (b1 , x2 ) − f (a1 , x2 )) b1 − a 1 ! 2 x1 − a 1 x − s ∂ f 1 1 − B2∗ B2 (s1 , x2 )ds1 . b1 − a 1 b1 − a 1 ∂x21 (4.10)
By Theorem 4.8 we have Z b1 x1 − a 1 1 f (s1 , x2 )ds1 + B1 (f (b1 , x2 ) − f (a1 , x2 )) f (x1 , x2 ) = b 1 − a 1 a1 b1 − a 1 ! 2 Z x1 − a 1 (b1 − a1 ) b1 ∂ f ∗ x1 − s 1 + B2 − B2 (s1 , x2 )ds1 2 b1 − a 1 b1 − a 1 ∂x21 a1 Z b1 1 f (s1 , x2 )ds1 + T1 (x1 , x2 ). (4.11) = b 1 − a 1 a1 And also we obtain Z b2 1 x2 − a 2 f (s1 , x2 ) = f (s1 , s2 )ds2 + B1 (f (s1 , b2 ) − f (s1 , a2 )) b 2 − a 2 a2 b2 − a 2 ! 2 Z (b2 − a2 ) b2 x2 − a 2 x − s ∂ f 2 2 + − B2∗ B2 (s1 , s2 )ds2 . 2 b − a b − a ∂x22 2 2 2 2 a2
Proof.
(4.12)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
32
Thus we derive
Z b1 ( Z b2 1 1 f (s1 , s2 )ds2 f (x1 , x2 ) = b 1 − a 1 a1 b 2 − a 2 a2 x2 − a 2 + B1 (f (s1 , b2 ) − f (s1 , a2 )) b2 − a 2 Z x2 − a 2 (b2 − a2 ) b2 B2 + 2 b2 − a 2 a2 ) ! 2 ∂ f ∗ x2 − s 2 − B2 (s1 , s2 )ds2 ds1 + T1 (x1 , x2 ), b2 − a 2 ∂x22
proving the claim.
(4.13)
We continue with Proposition 4.12. All as in Assumption 4.10, case n = 3 and m ∈ N, xi ∈ [ai , bi ], i = 1, 2, 3. Then Z b1 Z b2 Z b3 1 f s1 , s2 , s3 ds1 ds2 ds3 f (x1 , x2 , x3 ) = 3 Q (bi − ai ) a1 a2 a3 i=1
+ T3 (x3 ) + T2 (x2 , x3 ) + T1 (x1 , x2 , x3 ),
(4.14)
where m−1 X (b3 − a3 )k−1 x3 − a3 1 T3 := T3 (x3 ) := Bk (b1 − a1 )(b2 − a2 ) k! b3 − a 3 k=1 ! Z b1 Z b 2 ∂ k−1 f ∂ k−1 f × (s1 , s2 , b3 ) − s1 , s2 , a 3 ds1 ds2 ∂x3k−1 ∂x3k−1 a2 a1 Z b1 Z b2 Z b3 x3 − a 3 (b3 − a3 )m−1 Bm + m!(b1 − a1 )(b2 − a2 ) a1 a2 a3 b3 − a 3 ! x3 − s 3 ∂mf ∗ −Bm (s1 , s2 , s3 )ds1 ds2 ds3 , (4.15) b3 − a 3 ∂xm 3
T2 :=
m−1 X (b2 − a2 )k−1 x2 − a2 Z b1 ∂ k−1 f 1 Bk (s1 , b2 , x3 ) (b1 − a1 ) k! b2 − a 2 ∂x2k−1 a1 k=1 ! Z Z b2 (b2 − a2 )m−1 b1 ∂ k−1 f x2 − a 2 (s1 , a2 , x3 ) ds1 + − Bm m!(b1 − a1 ) a1 b2 − a 2 ∂x2k−1 a2 ! ! m x2 − s 2 ∂ f ∗ − Bm (4.16) (s1 , s2 , x3 )ds2 ds1 , b2 − a 2 ∂xm 2
T2 (x2 , x3 ) :=
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
33
and x1 − a 1 (b1 − a1 )k−1 T1 := T1 (x1 , x2 , x3 ) := Bk k! b1 − a 1 k=1 ! ∂ k−1 f ∂ k−1 f (b1 , x2 , x3 ) − (a1 , x2 , x3 ) ∂x1k−1 ∂x1k−1 ! m Z x1 − s 1 ∂ f x1 − a 1 (b1 − a1 )m−1 b1 ∗ Bm − Bm (s1 , x2 , x3 )ds1 . + m! b1 − a 1 b1 − a 1 ∂xm a1 1 m−1 X
(4.17) When m = 1 then the sums in (4.15)–(4.17) are zero. Proof.
Applying Theorem 4.8 we have 1 f (x1 , x2 , x3 ) = b1 − a 1
Z
b1
f (s1 , x2 , x3 )ds1 + T1 (x1 , x2 , x3 ).
(4.18)
a1
Furthermore we find Z b2 1 f (s1 , s2 , x3 )ds2 b 2 − a 2 a2 ! m−1 X (b2 − a2 )k−1 x2 − a2 ∂ k−1 f ∂ k−1 f Bk (s1 , b2 , x3 ) − (s1 , a2 , x3 ) + k! b2 − a 2 ∂x2k−1 ∂x2k−1 k=1 ! m Z x − s ∂ f x2 − a 2 (b2 − a2 )m−1 b2 2 2 ∗ Bm − Bm (s1 , s2 , x3 )ds2 + m! b − a b − a ∂xm 2 2 2 2 a2 2 f (s1 , x2 , x3 ) =
(4.19) and Z b3 1 f (s1 , s2 , s3 )ds3 b 3 − a 3 a3 ! m−1 X (b3 − a3 )k−1 x3 − a3 ∂ k−1 f ∂ k−1 f + (s1 , s2 , b3 ) − (s1 , s2 , a3 ) Bk k! b3 − a 3 ∂x3k−1 ∂x3k−1 k=1 ! m Z (b3 − a3 )m−1 b3 x3 − a 3 x − s ∂ f 3 3 ∗ + − Bm Bm (s1 , s2 , s3 )ds3 . m! b − a b − a ∂xm 3 3 3 3 a3 3 f (s1 , s2 , x3 ) =
(4.20)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
34
Putting the above together we derive Z b1 ( Z b2 1 1 f (s1 , s2 , x3 )ds2 f (x1 , x2 , x3 ) = b 1 − a 1 a1 b 2 − a 2 a2 m−1 X (b2 − a2 )k−1 x2 − a2 ∂ k−1 f ∂ k−1 f + (s1 , b2 , x3 ) − (s1 , a2 , x3 ) Bk k! b2 − a 2 ∂x2k−1 ∂x2k−1 k=1 Z x2 − a 2 (b2 − a2 )m−1 b2 Bm + m! b2 − a 2 a2 ) ! m x − s ∂ f 2 2 ∗ − Bm (4.21) (s1 , s2 , x3 )ds2 ds1 + T1 b2 − a 2 ∂xm 2 Z b 1 Z b2 1 = f (s1 , s2 , x3 )ds2 ds1 (b1 − a1 )(b2 − a2 ) a1 a2 m−1 X (b2 − a2 )k−1 x2 − a2 Z b1 ∂ k−1 f 1 Bk (s , b , x3 )ds1 + k−1 1 2 (b1 − a1 ) k! b2 − a 2 a1 ∂x2 k=1 ! Z Z b2 Z b1 k−1 (b2 − a2 )m−1 b1 x2 − a 2 ∂ f B + (s , a , x )ds − m 3 1 k−1 1 2 m!(b1 − a1 ) a1 b2 − a 2 a2 a1 ∂x2 ! ! m x2 − s 2 ∂ f ∗ − Bm (4.22) (s1 , s2 , x3 )ds2 ds1 + T1 . b2 − a 2 ∂xm 2 So far we have f (x1 , x2 , x3 ) =
1 (b1 − a1 )(b2 − a2 )
Z
b1 a1
Z
b2
f (s1 , s2 , x3 )ds2 ds1 + T2 (x2 , x3 ) + T1 (x1 , x2 , x3 ). a2
(4.23)
Finally we conclude that f (x1 , x2 , x3 )
Z
"
Z b3 1 f (s1 , s2 , s3 )ds3 b 3 − a 3 a3 a1 a2 m−1 X (b3 − a3 )k−1 x3 − a3 ∂ k−1 f ∂ k−1 f Bk (s , s , b ) − (s , s , a ) + 1 2 3 1 2 3 k! b3 − a 3 ∂x3k−1 ∂x3k−1 k=1 Z (b3 − a3 )m−1 b3 x3 − a 3 + Bm m! b3 − a 3 a3 # ! m x3 − s 3 ∂ f ∗ −Bm (4.24) (s1 , s2 , s3 )ds3 ds2 ds1 + T2 + T1 b3 − a 3 ∂xm 3
1 = (b1 − a1 )(b2 − a2 )
b1
Z
b2
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
=
1 3 Q
i=1
(bi − ai )
Z
Z
b1 a1
b2 a2
Z
35
b3
f (s1 , s2 , s3 )ds1 ds2 ds3 + T3 + T2 + T1 ,
(4.25)
a3
proving the claim.
We further present Proposition 4.13. All as in Assumption 4.10, case of n = 4 and m ∈ N, xi ∈ [ai , bi ], i = 1, . . . , 4. Then 1
f (x1 , x2 , x3 , x4 ) =
+
4 Q
(bi − ai )
i=1 4 X
Z
4 Q
f (s1 , s2 , s3 , s4 )ds1 ds2 ds3 ds4 [ai ,bi ]
i=1
Tj .
(4.26)
j=1
Here 1
T4 := T4 (x4 ) := Z +
i=1
3 Q
[ai ,bi ]
i=1
(bi − ai )
∗ Bm
k=1
x4 − a 4 (b4 − a4 )k−1 Bk k! b3 − a 4
! ∂ k−1 f ∂ k−1 f (s1 , s2 , s3 , b4 ) − (s1 , s2 , s3 , a4 ) ds1 ds2 ds3 ∂x4k−1 ∂x4k−1
(b4 − a4 )m−1 3 Q (bi − ai )m!
i=1
−
3 Q
m−1 X
x4 − s 4 b4 − a 4
Z
4 Q
i=1
!
Bm [ai ,bi ]
x4 − a 4 b4 − a 4
! ∂ mf (s1 , s2 , s3 , s4 ) ds1 ds2 ds3 ds4 , ∂xm 4
(4.27)
m−1 X (b3 − a3 )k−1 x3 − a3 1 Bk T3 := T3 (x3 , x4 ) := (b1 − a1 )(b2 − a2 ) k! b3 − a 3 k=1 ! Z b1 Z b2 k−1 ∂ f ∂ k−1 f × (s1 , s2 , b3 , x4 ) − (s1 , s2 , a3 , x4 ) ds1 ds2 ∂x3k−1 ∂x3k−1 a1 a2 Z b1 Z b2 Z b3 (b3 − a3 )m−1 x3 − a 3 + Bm m!(b1 − a1 )(b2 − a2 ) a1 a2 a3 b3 − a 3 ! ! ! m x3 − s 3 ∂ f ∗ − Bm (4.28) (s1 , s2 , s3 , x4 ) ds1 ds2 ds3 , b3 − a 3 ∂xm 3
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
36
T2 := T2 (x2 , x3 , x4 )
and
m−1 X (b2 − a2 )k−1 x2 − a2 1 := Bk (b1 − a1 ) k! b2 − a 2 k=1 !! Z b1 k−1 ∂ f ∂ k−1 (s1 , b2 , x3 , x4 ) − f (s1 , a2 , x3 , x4 ) ds1 × ∂x2k−1 ∂x2k−1 a1 Z b1 Z b 2 (b2 − a2 )m−1 x2 − a 2 + Bm m!(b1 − a1 ) b2 − a 2 a1 a2 ! ! ! m ∂ f x2 − s 2 ∗ (4.29) (s1 , s2 , x3 , x4 ) ds1 ds2 , − Bm b2 − a 2 ∂xm 2
T1 := T1 (x1 , x2 , x3 , x4 ) :=
k−1 (b1 − a1 )k−1 ∂ f x1 − a 1 (b , x2 , x3 , x4 ) Bk k−1 1 k! b1 − a 1 ∂x 1 k=1 Z ∂ k−1 f x1 − a 1 (b1 − a1 )m−1 b1 − k−1 (a1 , x2 , x3 , x4 ) + Bm m! b1 − a 1 ∂x1 a1 ! m ∂ f x1 − s 1 ∗ (s1 , x2 , x3 , x4 )ds1 . (4.30) − Bm b1 − a 1 ∂xm 1
m−1 X
When m = 1 then the sums in (4.27)–(4.30) are zero. Proof.
By Theorem 4.8 we have
Z b1 1 f (s1 , x2 , x3 , x4 )ds1 b 1 − a 1 a1 m−1 X (b1 − a1 )k−1 x1 − a1 ∂ k−1 f + Bk (b1 , x2 , x3 , x4 ) k! b1 − a 1 ∂x1k−1 k=1 Z (b1 − a1 )m−1 b1 x1 − a 1 ∂ k−1 f Bm (a1 , x2 , x3 , x4 ) + − m! b1 − a 1 ∂x1k−1 a1 ! x1 − s 1 ∂mf ∗ − Bm (s1 , x2 , x3 , x4 )ds1 b1 − a 1 ∂xm 1 Z b1 1 f (s1 , x2 , x3 , x4 )ds1 + T1 (x1 , x2 , x3 , x4 ). = (b1 − a1 ) a1 Similarly we obtain Z b2 1 f (s1 , s2 , x3 , x4 )ds2 f (s1 , x2 , x3 , x4 ) = b 2 − a 2 a2 m−1 X (b2 − a2 )k−1 x2 − a2 ∂ k−1 f + Bk (s1 , b2 , x3 , x4 ) k! b2 − a 2 ∂x2k−1 k=1 f (x1 , x2 , x3 , x4 ) =
(4.31) (4.32)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
Z x2 − a 2 (b2 − a2 )m−1 b2 ∂ k−1 f (s , a , x , x ) + B 1 2 3 4 m m! b2 − a 2 ∂x2k−1 a2 ! m x2 − s 2 ∂ f ∗ − Bm (s1 , s2 , x3 , x4 )ds2 , b2 − a 2 ∂xm 2 Z b3 1 f (s1 , s2 , s3 , x4 )ds3 f (s1 , s2 , x3 , x4 ) = b 3 − a 3 a3 m−1 X (b3 − a3 )k−1 x3 − a3 ∂ k−1 f + Bk (s1 , s2 , b3 , x4 ) k! b3 − a 3 ∂x3k−1 k=1 Z x3 − a 3 (b3 − a3 )m−1 b3 ∂ k−1 f B (s , s , a , x ) + − m 1 2 3 4 m! b3 − a 3 ∂x3k−1 a3 ! m x3 − s 3 ∂ f ∗ − Bm (s1 , s2 , s3 , x4 )ds3 , b3 − a 3 ∂xm 3
37
−
(4.33)
(4.34)
and finally Z b4 1 f (s1 , s2 , s3 , s4 )ds4 f (s1 , s2 , s3 , x4 ) = b 4 − a 4 a4 m−1 X (b4 − a4 )k−1 x4 − a4 ∂ k−1 f (s1 , s2 , s3 , b4 ) + Bk k! b4 − a 4 ∂x4k−1 k=1 Z ∂ k−1 f (b4 − a4 )m−1 b4 x4 − a 4 − (s , s , s , a ) + B 1 2 3 4 m m! b4 − a 4 ∂x4k−1 a4 ! ∂mf x4 − s 4 ∗ (s1 , s2 , s3 , s4 )ds4 . − Bm b4 − a 4 ∂xm 4
(4.35)
Consequently, by using the above we derive f (x1 , x2 , x3 , x4 ) Z b1 " Z b2 1 1 = f (s1 , s2 , x3 , x4 )ds2 (b1 − a1 ) a1 (b2 − a2 ) a2 m−1 X (b2 − a2 )k−1 x2 − a2 ∂ k−1 f + Bk (s1 , b2 , x3 , x4 ) k! b2 − a 2 ∂x2k−1 k=1 Z (b2 − a2 )m−1 b2 ∂ k−1 f x2 − a 2 − k−1 (s1 , a2 , x3 , x4 ) + Bm m! b2 − a 2 ∂x2 a2 # ! m x2 − s 2 ∂ f ∗ −Bm (s1 , s2 , x3 , x4 )ds2 ds1 + T1 b2 − a 2 ∂xm 2
(4.36)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
38
1 = (b1 − a1 )(b2 − a2 )
Z
b1 a1
Z
b2
f (s1 , s2 , x3 , x4 )ds1 ds2 a2
m−1 X (b2 − a2 )k−1 x2 − a2 Z b1 ∂ k−1 f 1 + Bk (s1 , b2 , x3 , x4 ) (b1 − a1 ) k! b2 − a 2 ∂x2k−1 a1 k=1 Z Z x2 − a 2 (b2 − a2 )m−1 b1 b2 ∂ k−1 f (s1 , a2 , x3 , x4 )ds1 + Bm − m!(b1 − a1 ) a1 a2 b2 − a 2 ∂x2k−1 ! x2 − s 2 ∂mf ∗ − Bm (s1 , s2 , x3 , x4 )ds1 ds2 + T1 (4.37) b2 − a 2 ∂xm 2 Z b 1 Z b2 ( Z b3 1 1 = f (s1 , s2 , s3 , x4 )ds3 (b1 − a1 )(b2 − a2 ) a1 a2 b3 − a3 a3 m−1 X (b3 − a3 )k−1 x3 − a3 ∂ k−1 f (s1 , s2 , b3 , x4 ) Bk + k! b3 − a 3 ∂x3k−1 k=1 Z b3 x3 − a 3 ∂ k−1 f (b3 − a3 )m−1 − Bm (s1 , s2 , a3 , x3 ) + m! b3 − a 3 ∂x3k−1 a3 ! ! ! ∂mf x3 − s 3 ∗ (s1 , s2 , s3 , x4 )ds3 ds1 ds2 + T2 + T1 . − Bm (4.38) b3 − a 3 ∂xm 3 That is, we found that Z 3 X 1 f (s , s , s , x )ds ds ds + Tj . (4.39) f (x1 , x2 , x3 , x4 ) = 3 1 2 3 4 1 2 3 3 Q Q j=1 (bi − ai ) i=1[ai ,bi ] i=1
At last we observe that
f (x1 , x2 , x3 , x4 ) Z b4 Z 1 1 f (s1 , s2 , s3 , s4 )ds4 = 3 3 Q Q [ai ,bi ] b4 − a4 a4 (bi − ai ) i=1 i=1 m−1 X
k−1 (b4 − a4 )k−1 x4 − a 4 f ∂ ∂ k−1 f + Bk (s1 , s2 , s3 , b4 ) − (s1 , s2 , s3 , a4 ) k! b4 − a 4 ∂x4k−1 ∂x4k−1 k=1 Z x4 − a 4 (b4 − a4 )m−1 b4 Bm + m! b4 − a 4 a4 ! ! m 3 X ∂ f x4 − s 4 ∗ −Bm (s , s , s , s )ds ds ds ds + Ti (4.40) 1 2 3 4 4 1 2 3 b4 − a 4 ∂xm 4 i=1 Z 4 X 1 f (s , s , s , s )ds ds ds ds + Tj , (4.41) = 4 1 2 3 4 1 2 3 4 4 Q Q j=1 (bi − ai ) i=1[ai ,bi ] i=1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
39
proving the claim.
We finally give the following general Multivariate Euler-type identity in Theorem 4.14. All as in Assumption 4.10 for m, n ∈ N, xi ∈ [ai , bi ], i = 1, 2, . . . , n. Then f (x1 , x2 , . . . , xn ) = Q n
i=1
1 (bi − ai )
Z
n Q
[ai ,bi ]
f (s1 , s2 , . . . , sn )ds1 ds2 · · · dsn +
i=1
n X
Tj ,
j=1
(4.42)
where for j = 1, . . . , n we have
Tj := Tj (xj , xj+1 , . . . , xn ) :=
×
Z
j−1 Q i=1
[ai ,bi ]
1 j−1 Q i=1
(bi − ai )
(m−1 X (bj − aj )k−1 xj − aj Bk k! bj − a j k=1
∂ k−1 f (s1 , s2 , . . . , sj−1 , bj , xj+1 , . . . , xn ) ∂xjk−1
!) ∂ k−1 f − k−1 (s1 , s2 , . . . , sj−1 , aj , xj+1 , . . . , xn ) ds1 · · · dsj−1 ∂xj (Z xj − a j (bj − aj )m−1 B + m j Q j−1 bj − a j Q [ai ,bi ] m! (bi − ai ) i=1 i=1 ! ) ! m ∂ f xj − s j ∗ −Bm (s1 , s2 , . . . , sj , xj+1 , . . . , xn ) ds1 ds2 · · · dsj . bj − a j ∂xm j (4.43) When m = 1 then the sum in (4.43) is zero. Proof. Theorem 4.14 is true for n = 3, 4, see Propositions 4.12, 4.13. Assume that Theorem 4.14, is true for n, see (4.42), (4.43). Then we prove it for n+1, n ∈ N. We have Z 1 f (s1 , . . . , sn , xn+1 )ds1 , . . . dsn f (x1 , x2 , . . . , xn , xn+1 ) = Q n n Q [a ,b ] i i (bi − ai ) i=1 +
i=1 n X j=1
where
Tj , for j = 1, . . . , n,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
40
Tj := Tj xj , xj+1 , . . . , xn , xn+1 =
Bk
xj − a j bj − a j
! Z
j−1 Q i=1
[ai ,bi ]
1 (j−1) Q i=1
(bi − ai )
(m−1 X (bj − aj )k−1 k! k=1
∂ k−1 f (s1 , . . . , sj−1 , bj , xj+1 , . . . , xn , xn+1 ) ∂xjk−1 !) !!
∂ k−1 f ds1 . . . dsj−1 s1 , . . . , sj−1 , aj , xj+1 , . . . , xn , xn+1 ∂xjk−1 (Z xj − a j (bj − aj )m−1 + Bm j Q j−1 bj − a j Q [ai ,bi ] m! (bi − ai ) i=1 i=1 ) !! ! m ∂ f xj − s j ∗ ds1 . . . dsj . s1 , . . . , sj , xj+1 , . . . , xn , xn+1 − Bm bj − a j ∂xm j −
But it holds 1 f (s1 . . . , sn , xn+1 ) = (bn+1 − an+1 )
Z
bn+1
f (s1 , . . . , sn , sn+1 )dsn+1 an+1
! xn+1 − an+1 (bn+1 − an+1 )k−1 + Bk k! bn+1 − an+1 k=1 # " ∂ k−1 f (s1 , . . . sn , bn+1 ) ∂ k−1 f − s , . . . , s , a 1 n n+1 k−1 k−1 ∂xn+1 ∂xn+1 ! m−1 Z b " n+1 bn+1 − an+1 xn+1 − an+1 Bm + m! bn+1 − an+1 an+1 !# ∂ m f (s1 , . . . , sn , sn+1 ) xn+1 − sn+1 ∗ dsn+1 . − Bm bn+1 − an+1 ∂xm n+1 m−1 X
Thus we get f (x1 , x2 , . . . , xn , xn+1 ) ( Z bn+1 Z 1 1 = Q f (s1 , . . . , sn , sn+1 )dsn+1 n n Q (bn+1 − an+1 ) an+1 (bi − ai ) i=1[ai ,bi ] i=1
+ "
m−1 X k=1
(bn+1 − an+1 )k−1 xn+1 − an+1 Bk k! bn+1 − an+1
∂ k−1 f (s1 , . . . , sn , bn+1 ) ∂ k−1 f (s1 , . . . , sn , an+1 ) − k−1 k−1 ∂xn+1 ∂xn+1
#
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
41
" Z xn+1 − an+1 (bn+1 − an+1 )m−1 bn+1 Bm + m! bn+1 − an+1 an+1 ) # m n X xn+1 − sn+1 ∂ f (s1 , . . . , sn , sn+1 ) ∗ −Bm ds . . . , ds + Tj . ds 1 n n+1 bn+1 − an+1 ∂xm n+1 j=1 Hence f (x1 , . . . , xn+1 ) =
1 n+1 Q i=1
(bi − ai )
Z
n+1 Q
f (s1 , . . . , sn+1 ) ds1 . . . dsn+1 + [ai ,bi ]
n+1 X
Tj ,
j=1
i=1
where Tn+1 (xn+1 ) := Tn+1 := Q n
(Z
n Q
i=1
[ai ,bi ]
"
i=1
∂
1 (bi − ai )
(m−1 X (bn+1 − an+1 )k−1 k!
k=1
Bk
xn+1 − an+1 bn+1 − an+1
)) # f (s1 , . . . , sn , bn+1 ) ∂ k−1 f − k−1 s1 , . . . , sn , an+1 ds1 . . . dsn k−1 ∂xn+1 ∂xn+1 (Z "" (bn+1 − an+1 )m−1 xn+1 − an+1 n+1 n + Bm Q Q bn+1 − an+1 [ai ,bi ] m! bi − a i i=1 i=1 ) # m # xn+1 − sn+1 ∂ f ∗ −Bm s1 , . . . , sn , sn+1 ds1 . . . dsn+1 . bn+1 − an+1 ∂xm n+1
k−1
Thus is proving the claim.
Next we rewrite the last theorem. Theorem 4.15. All as in Assumption 4.10 for m, n ∈ N, xi ∈ [ai , bi ], i = 1, 2, . . . , n. Then
f Em (x1 , x2 , . . . , xn ) := f (x1 , x2 , . . . , xn ) Z 1 − Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ]
−
i=1 n X j=1
Aj =
n X j=1
Bj ,
(4.44)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
42
where for j = 1, . . . , n we have 1
Aj := Aj (xj , xj+1 , . . . , xn ) =
×
Z
j−1 Q i=1
[ai ,bi ]
j−1 Q i=1
(bi − ai )
(m−1 X (bj − aj )k−1 xj − aj Bk k! bj − a j k=1
∂ k−1 f (s1 , s2 , . . . , sj−1 , bj , xj+1 , . . . , xn ) ∂xjk−1
!) ∂ k−1 f , − (s1 , s2 , . . . , sj−1 , aj , xj+1 , . . . , xn ) ds1 · · · dsj−1 ∂xjk−1
(4.45)
and
(Z (bj − aj )m−1 xj − a j Bj := Bj (xj , xj+1 , . . . , xn ) := B m j Q j−1 bj − a j Q [ai ,bi ] m! (bi − ai ) i=1 i=1 ) ! ! m xj − s j ∂ f ∗ − Bm (s1 , s2 , . . . , sj , xj+1 , . . . , xn ) ds1 ds2 · · · dsj . bj − a j ∂xm j (4.46)
When m = 1 then Aj = 0, and Tj = Bj , j = 1, . . . , n. Remark 4.16. Notice above that Tj = Aj + Bj , j = 1, . . . , n. Also we have that f |Em (x1 , x2 , . . . , xn )| ≤
Also by denoting ∆ := f (x1 , . . . , xn ) − Q n
1
i=1
(bi − ai )
Z
n Q
[ai ,bi ]
n X j=1
|Bj |.
f (s1 , . . . , sn )ds1 · · · dsn
(4.47)
(4.48)
i=1
we get
|∆| ≤
n X j=1
(|Aj | + |Bj |).
(4.49)
Later we will estimate Aj , Bj . A general set of suppositions follow
Assumption 4.17. Here m ∈ N, j = 1, . . . , n.We suppose 1) f :
n Q
i=1
2)
∂`f ∂x`j
[ai , bi ] → R is continuous.
are existing real valued functions for all j = 1, . . . , n; ` = 1, . . . , m − 2.
3) For each j = 1, . . . , n we assume that continuous real valued function.
∂ m−1 f (x1 , . . . , xj−1 , ·, xj+1 , . . . , xn ) ∂xjm−1
is a
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
43
m
4) For each j = 1, . . . , n we assume that gj (·) := ∂∂xmf (x1 , . . . , xj−1 , ·, xj+1 , . . . , xn ) j exists and is real valued with the possibility of being infinite only over an at most countable subset of (aj , bj ). 5) Parts #3, #4 are true for all (x1 , . . . , xj−1 , xj+1 , . . . , xn ) ∈
n Y
[ai , bi ].
i=1 i6=j
6) The functions for j = 2, . . . , n; ` = 1, . . . , m − 2, j−1 j−1 ` z }| { z }| { ∂ f ·, ·, ·, · · · , ·, xj , xj+1 , . . . , xn qj ·, ·, · · · , · := ∂x`j are continuous on
j−1 Q i=1
[ai , bi ], for each (xj , xj+1 , . . . , xn ) ∈
n Q
[ai , bi ].
i=j
7) The functions for each j = 1, . . . , n, j j m }| { z }| { z ∂ f ·, ·, ·, · · · , ·, xj+1 , . . . , xn ∈ L1 ϕj ·, ·, ·, · · · , ·, ·, · := ∂xm j for any (xj+1 , . . . , xn ) ∈ We present
n Q
j Y
i=1
!
[ai , bi ] ,
[ai , bi ].
i=j+1
Theorem 4.18. All as in Assumption 4.17. Then (4.42) and (4.44) are true again. Proof. Use of Assumption 4.17, Theorem 4.9 and it is similar to the proof of the so far results of this chapter. Remark 4.19. The results of Propositions 4.11, 4.12, 4.13 are still valid under Assumption 4.17 for their cases. Some weaker general suppositions follow. Assumption 4.20. Here m ∈ N, j = 1, . . . , n and only the Parts #1, #2, #6, #7 of Assumption 4.17 remain the same. We further suppose that for each j = 1, . . . , n and over [aj , bj ], the function ∂ m−1 f (x1 , . . . , xj−1 , ·, xj+1 , . . . , xn ) ∂xjm−1 is absolutely continuous, and this is true for all (x1 , . . . , xj−1 , xj+1 , . . . , xn ) ∈ We give
n Y
i=1 i6=j
[ai , bi ].
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
44
Theorem 4.21. All as in Assumption 4.20. Then (4.42) and (4.44) are still true. Remark 4.22. The results of Propositions 4.11, 4.12, 4.13 are again valid under Assumption 4.20 for their cases. We proceed with the estimates of our remainders in (4.42) and (4.44). We need to make Remark 4.23. We are operating under the Assumptions 4.10 or 4.17 or 4.20. We observe for j = 1, . . . , n that |Bj | ≤ Γj ,
(4.50)
where Γj :=
Z xj − sj (bj − aj )m−1 xj − a j ∗ − B B m j m Q j−1 bj − a j bj − a j Q [ai ,bi ] m! (bi − ai ) i=1 i=1 ! ∂mf × m (s1 , s2 , . . . , sj , xj+1 , . . . , xn ) ds1 · · · dsj . (4.51) ∂xj
Here we assume
j ∂ m f z}|{ · · · , xj+1 , . . . , xn ∈ L∞ ∂xm j
for any (xj+1 , . . . , xn ) ∈ Thus we obtain
n Q
i=j+1
j Y
i=1
[ai , bi ]
!
[ai , bi ], any xj ∈ [aj , bj ].
! Z (bj − aj )m−1 xj − sj xj − a j ∗ Γj ≤ j Bn bj − aj − Bm bj − aj ds1 · · · dsj Q j−1 Q [ai ,bi ] m! (bi − ai ) i=1
i=1
m
j
∂ f z }| {
·, ·, ·, · · · , ·, xj+1 , . . . , xn × (4.52)
∂xm
j
j
Q ∞,
[ai ,bi ]
i=1
! Z bj x − a x − s (bj − aj )m−1 j j j j ∗ Bm dsj − Bm = m! bj − a j bj − a j aj
m
j
∂ f z }| {
× m · · · , · · ·, xj+1 , . . . , xn
j
∂xj
Q ∞,
(4.53)
[ai ,bi ]
i=1
(by letting
λj := λj (xj ) :=
xj − a j bj − a j
(4.54)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
we have) (bj − aj )m = m!
m Z 1 j
∂ f z}|{
|Bm (λj ) − Bm (tj )|dtj m · · · , xj+1 , . . . , xn ·
0
∂xj
45
(4.55) ∞,
j Q
[ai ,bi ]
i=1
(bj − aj )m sm!
m
j Z 1
∂ f z}|{
· · · , xj+1 , . . . , xn (Bm (λj ) − Bm (tj ))2 dtj ·
∂xm
0
j
≤
(4.56) ∞,
(using [98], p. 352 we get) (bj − aj )m = m! s !
m
j 2
z}|{ xj − a j (m!) 2
∂ f · · · , xj+1 , . . . , xn |B2m | + Bm ·
m
(2m)! bj − a j
∂xj
j Q
[ai ,bi ]
i=1
. ∞,
j Q
[ai ,bi ]
i=1
(4.57)
We have established Lemma 4.24. Suppose Assumptions 4.10 or 4.17 or 4.20. Let Bj as in (4.46). In particular suppose that ! j j Y ∂ m f z}|{ · · · , xj+1 , . . . , xn ∈ L∞ [ai , bi ] , ∂xm j i=1
for any (xj+1 , . . . , xn ) (xj , xj+1 , . . . , xn ) ∈
n Q
∈
n Q
[ai , bi ], all j
= 1, . . . , n.
Then for any
i=j+1
[ai , bi ] we have
i=j
(bj − aj )m |Bj | = |Bj (xj , xj+1 , . . . , xn )| ≤ m!
m
j
∂ f z}|{
× m · · · , xj+1 , . . . , xn
j
∂xj
Q ∞,
s
(m!)2 xj − a j 2 |B2m | + Bm (2m)! bj − a j ,
!
(4.58)
[ai ,bi ]
i=1
for all j = 1, . . . , n. We continue with
Remark 4.25. Continuing from Remark 4.23. Let pj , qj > 1: j = 1, . . . , n, with the assumption that ! j Y ∂mf (· · · , xj+1 , . . . , xn ) ∈ Lqj [ai , bi ] , ∂xm j i=1
1 pj
+
1 qj
= 1;
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
46
for any (xj+1 , . . . , xn ) ∈
n Q
i=j+1
[ai , bi ], xj ∈ [aj , bj ]. From (4.50) and (4.51) we get Z
(bj − aj )m−1 |Bj | ≤ j−1 Q (bi − ai ) m! i=1
−
∗ Bm
Z ×
xj − a j j Bm bj − aj Q [ai ,bi ]
i=1
!1/pj p xj − sj j ds1 · · · dsj bj − a j
1/qj m qj ∂ f j m (s1 , . . . , sj , xj+1 , . . . , xn ) ds1 · · · dsj Q [ai ,bi ] ∂xj
(4.59)
i=1
1/pj Z bj j−1 Y (bj − aj )m−1 Bm (λj ) = (bi − ai ) j−1 Q a j m! (bi − ai ) i=1 i=1
∗ − Bm
=
!1/pj
p
∂mf xj − sj j
(4.60) dsj j
∂xm (· · · , xj+1 , . . . , xn ) Q bj − a j j qj , [ai ,bi ] i=1
(bj − aj ) m!
m− q1
j
pj
− Bm (tj )| dtj We have derived
j−1 Y
(bi − ai )
i=1 1/pj
!−1/qj Z
1
0
|Bm (λj )
m
∂ f (· · · , xj+1 , . . . , xn )
∂xm j
qj ,
j Q
.
(4.61)
[ai ,bi ]
i=1
Lemma 4.26. Suppose Assumptions 4.10 or 4.17 or 4.20. Let Bj as in (4.46) and pj , qj > 1 : p1j + q1j = 1; j = 1, . . . , n. In particular we suppose that ! j Y ∂mf (· · · , xj+1 , . . . , xn ) ∈ Lqj [ai , bi ] , ∂xm j i=1 for any (xj+1 , . . . , xn ) ∈ (xj , xj+1 , . . . , xn ) ∈
n Q
n Q
[ai , bi ], for all j = 1, . . . , n.
Then for any
i=j+1
[ai , bi ] we have
i=j
|Bj | = |Bj (xj , xj+1 , . . . , xn )| (bj − aj ) ≤ m!
m− q1
j
pj − Bm (tj ) dtj
j−1 Y
(bi − ai )
i=1 1/pj
!−1/qj
Z 1 B m xj − a j bj − a j 0
m
∂ f (· · · , xj+1 , . . . , xn )
∂xm
j
qj ,
j Q
i=1
[ai ,bi ]
.
(4.62)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
47
When pj = qj = 2, then we obtain !−1/2 1 j−1 (bj − aj )m− 2 Y |Bj | ≤ (bi − ai ) m! i=1
∂mf
× m (· · · , xj+1 , . . . , xn ) j
∂xj
Q
s
! (m!)2 x − a j j 2 |B2m | + Bm (2m)! bj − a j
.
(4.63)
[ai ,bi ]
2,
i=1
The inequalities (4.62), (4.63) are true for all j = 1, . . . , n. At last we make Remark 4.27. Continuing from Remark 4.25. Assume for j = 1, . . . , n that ! j Y ∂mf [ai , bi ] , (· · · , xj+1 , . . . , xn ) ∈ L1 ∂xm j i=1 Q for any (xj+1 , . . . , xn ) ∈ ni=j+1 [ai , bi ], any xj ∈ [aj , bj ]. Then from (4.50) and (4.51) we obtain |Bj | ≤
(bj − aj )m−1 j−1 Q m! (bi − ai ) i=1
Z
j Q
[ai ,bi ]
i=1
! m ∂ f ∂xm (s1 , . . . , sj , xj+1 , . . . , xn ) ds1 · · · dsj j
xj − a j xj − s j ∗
· B m − Bm bj − a j bj − aj ∞,[aj ,bj ]
(bj − aj )m−1 xj − a j xj − s j
∗ = − Bm
B m Qj−1 bj − a j bj − a j m! i=1 (bi − ai ) ∞,[aj ,bj ]
∂ mf
× m (· · · , xj+1 , . . . , xn ) j
Q
∂xj 1,
(4.64)
(4.65)
[ai ,bi ]
i=1
(bj − aj )m−1
Bm (t) − Bm xj − aj
j−1 bj − aj ∞,[0,1] Q m! (bi − ai ) i=1
∂ mf
. (4.66) × m (· · · , xj+1 , . . . , xn ) j
∂xj
Q
(by [98], p. 347) =
[ai ,bi ]
1,
i=1
From [98], pp. 347–348 we have: i) case m = 2r, r ∈ N, then
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
48
xj − a j
= B2r (t) − B2r bj − aj ∞,[0,1] ∞,[0,1] xj − aj ; (4.67) = (1 − 2−2r )|B2r | + 2−2r B2r − B2r bj − a j
Bm (t) − Bm xj − aj
bj − a j
ii) case m = 2r + 1, r ∈ N, then
Bm (t) − Bm xj − aj
bj − a j
xj − a j
= B2r+1 (t) − B2r+1 bj − aj ∞,[0,1] ∞,[0,1] 2(2r + 1)! B2r+1 xj − aj ; (4.68) + ≤ 2r+1 −2r (2π) (1 − 2 ) bj − a j
iii) special case of m = 1, then
Bm (t) − Bm xj − aj
B1 (t) − B1 xj − aj =
bj − a j bj − aj ∞,[0,1] ∞,[0,1] aj + bj 1 = + xj − . 2 2
(4.69)
We have proved
Lemma 4.28. Suppose Assumptions 4.10 or 4.17 or 4.20. Let Bj as in (4.46), j = 1, . . . , n. In particular we suppose for j = 1, . . . , n that ! j Y ∂mf [ai , bi ] , (. . . , xj+1 , . . . , xn ) ∈ L1 ∂xm j i=1 for any (xj+1 , . . . , xn ) ∈ we have
n Q
i=j+1
[ai , bi ]. Then for any (xj , xj+1 , . . . , xn ) ∈
n Q
[ai , bi ]
i=j
m
(bj − aj )m−1
∂ f (. . . , xj+1 , . . . , xn ) j
m
Q j−1 ∂xj Q 1, [ai ,bi ] m! (bi − ai ) i=1 i=1
xj − a j
. (4.70) × B (t) − B m m
bj − aj ∞,[0,1]
|Bj | = |Bj (xj , xj+1 , . . . , xn )| ≤
The special cases follow: 1) When m = 2r, r ∈ N we have |Bj | ≤
2r
(bj − aj )2r−1
∂ f (. . . , xj+1 , . . . , xn ) j 2r
Q
j−1 ∂xj Q 1, [ai ,bi ] (2r)! (bi − ai ) i=1 i=1 xj − aj . × (1 − 2−2r )|B2r | + 2−2r B2r − B2r bj − a j
(4.71)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
49
2) When m = 2r + 1, r ∈ N we obtain
∂ 2r+1 f
(bj − aj )2r
|Bj | ≤ (. . . , x , . . . , x )
j+1 n 2r+1 j−1 j
Q Q ∂xj 1, [ai ,bi ] (2r + 1)! (bi − ai ) i=1 i=1 2(2r + 1)! xj − aj + B2r+1 . (4.72) × (2π)2r+1 (1 − 2−2r ) bj − a j
3) When m = 1 we get |Bj | ≤
1 j−1 Q i=1
(bi − ai )
∂f
∂xj (. . . , xj+1 , . . . , xn )
1,
j Q
[ai ,bi ]
"
i=1
# aj + bj 1 + xj − . 2 2
(4.73)
We need θ Q − − [ai , bi ], θ ∈ N, where k→ x k := Definition 4.29. Let → x = (x1 , . . . , xθ ) ∈
p
i=1
x21 + · · · + x2θ . Let F :
lus of continuity of F by
θ Q
i=1
[ai , bi ] → R be continuous. We define the (first) modu-
ω1 (F, δ) :=
sup − − all → x ,→ y∈
θ Q
[ai ,bi ],
− − |F (→ x ) − F (→ y )|,
(4.74)
i=1
− − with k→ x −→ y k≤δ
for all δ > 0. Notice 4.30. Under Assumption 4.10 we have valid that ∂ k−1 f ∂ k−1 f (s , . . . , s , a , x , . . . , x ) k−1 (s1 , . . . , sj−1 , bj , xj+1 , . . . , xn ) − 1 j−1 j j+1 n ∂xj ∂xjk−1 j ∂ k−1 f z}|{ · · · , xj+1 , . . . , xn , bj − aj , all j = 1, . . . , n; k = 1, . . . , m − 1. ≤ ω1 ∂xjk−1
(4.75)
We give Lemma 4.31. Suppose Assumption 4.10. Let Aj as in (4.45), j = 1, . . . , n. Then for any (xj , xj+1 , . . . , xn ) ∈
n Y i=j
[ai , bi ]
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
50
we have |Aj | = |Aj (xj , xj+1 , . . . , xn )| ≤ · ω1
m−1 X k=1
xj − aj (bj − aj )k−1 B k b j − a j k!
∂ f z}|{ · · · , xj+1 , . . . , xn ), bj − aj , for all j = 1, . . . , n. (4.76) ∂xjk−1 j
k−1
Putting together all these above auxilliary results, we derive the following multivariate Ostrowski type inequalities.
f Theorem 4.32. Suppose Assumptions 4.10 or 4.17 or 4.20. Let Em (x1 , x2 , . . . , xn ) as in (4.44) and Aj for j = 1, . . . , n as in (4.45), m ∈ N. In particular we suppose that ! j j Y ∂ m f z}|{ [ai , bi ] , · · · , xj+1 , . . . , xn ∈ L∞ ∂xm j i=1
for any (xj+1 , . . . , xn ) ∈
n Q
[ai , bi ], all j = 1, . . . , n. Then
i=j+1
f |Em (x1 , . . . , xn )| = f (x1 , . . . , xn ) − Q n
1
Z
n X
Aj
f (s1 , . . . , sn )ds1 · · · dsn − n Q j=1 (bi − ai ) i=1[ai ,bi ] i=1 s " ! n 1 X (m!)2 xj − a j m 2 ≤ (bj − aj ) |B2m | + Bm m! j=1 (2m)! bj − a j #
m j
∂ f z}|{
· · · , x , . . . , x ) . (4.77) × j+1 n j
∂xm Q j
∞,
[ai ,bi ]
i=1
Proof.
By Theorems 4.15, 4.18, 4.21 and Lemma 4.24.
Next comes f Theorem 4.33. Suppose Assumptions 4.10 or 4.17 or 4.20. Let Em (x1 , . . . , xn ) 1 1 as in (4.44), m ∈ N. Let pj , qj > 1 : pj + qj = 1; j = 1, . . . , n. In particular we assume that ! j Y ∂mf (. . . , xj+1 , . . . , xn ) ∈ Lqj [ai , bi ] , ∂xm j i=1
for any (xj+1 , . . . , xn ) ∈
n Q
i=j+1
[ai , bi ], for all j = 1, . . . , n. Then
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
51
" j−1 − q1 Z 1 n X Y j xj − a j 1 m− q1 f j (bj − aj ) |Em (x1 , . . . , xn )| ≤ (bi − ai ) B m b j − a j m! j=1 0 i=1 #
pj 1/pj m
∂ f
. −Bm (tj ) dtj j
∂xm (. . . , xj+1 , . . . , xn ) Q j qj , [ai ,bi ] i=1
(4.78)
When pj = qj = 2, all j = 1, . . . , n, then
f |Em (x1 , . . . , xn )|
" −1/2 j−1 n Y 1 1 X (bi − ai ) (bj − aj )m− 2 ≤ m! j=1 i=1 s ! x − a (m!)2 j j 2 × |B2m | + Bm (2m)! bj − a j #
m
∂ f
. × m (. . . , xj+1 , . . . , xn ) Q j ∂xj 2, [ai ,bi ]
(4.79)
i=1
Proof.
By Theorems 4.15, 4.18, 4.21 and Lemma 4.26.
Finally we have f Theorem 4.34. Suppose Assumptions 4.10 or 4.17 or 4.20. Let Em (x1 , . . . , xn ) as in (4.44), m ∈ N. In particular we assume for j = 1, . . . , n that
∂mf (. . . , xj+1 , . . . , xn ) ∈ L1 ∂xm j for any (xj+1 , . . . , xn ) ∈
f |Em (x1 , . . . , xn )|
Qn
i=j+1 [ai , bi ].
j Y
i=1
!
[ai , bi ] ,
Then
" n 1 X (bj − aj )m−1 ≤ m! j=1 j−1 Q (bi − ai ) i=1
#
m
∂ f
xj − a j
. j
∂xm (. . . , xj+1 , . . . , xn ) Q
Bm (t) − Bm bj − aj j 1, [ai ,bi ] ∞,[0,1] i=1
(4.80)
The special cases are calculated and estimated further as follows: 1) When m = 2r, r ∈ N, then
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
52
f |E2r (x1 , . . . , xn )| ( n 1 X (bj − aj )2r−1 ≤ (2r)! j=1 j−1 Q (bi − ai ) i=1
!
2r
∂ f
j
∂x2r (. . . , xj+1 , . . . , xn ) Q j 1, [ai ,bi ] i=1
) −2r xj − aj −2r × (1 − 2 )|B2r | + 2 B2r − B2r . bj − a j
(4.81)
2) When m = 2r + 1, r ∈ N, then
f |E2r+1 (x1 , . . . , xn )| ( n X (bj − aj )2r 1 ≤ j−1 (2r + 1)! j=1 Q (bi − ai ) i=1
!
2r+1
∂
f
j
∂x2r+1 (. . . , xj+1 , . . . , xn ) Q 1, [ai ,bi ] j i=1
) x − a 2(2r + 1)! j j . + B2r+1 × (2π)2r+1 (1 − 2−2r ) bj − a j
(4.82)
And at last 3) When m = 1, then |E1f (x1 , . . . , xn )|
≤
n X j=1
(
j−1 Q i=1
(bi − ai )
"
∂f
(. . . , x , . . . , x ) j+1 n
∂xj
#) 1 aj + bj × + xj − . 2 2
Proof.
1
By Theorems 4.15, 4.18, 4.21 and Lemma 4.28.
1,
j Q
i=1
[ai ,bi ]
#
(4.83)
The final general Ostrowski type estimate follows: Theorem 4.35. All as in Assumption 4.10, m, n ∈ N, xi ∈ [ai , bi ], i = 1, 2, . . . , n. Here Bj is as in (4.46), for j = 1, . . . , n. Then Z 1 f (x1 , . . . , xn ) − f (s , s , . . . , s )ds ds · · · ds 1 2 n 1 2 n n n Q Q [ai ,bi ] (b − a ) i i i=1 i=1 " n n m−1 X X X (bj − aj )k−1 B k ≤ |Bj | + k! j=1 j=1 k=1 k−1 # ∂ f xj − aj ω1 (. . . , xj+1 , . . . , xn ), bj − aj . (4.84) bj − a j ∂xjk−1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
53
Estimates for Bj are given by Lemmas 4.24, 4.26, 4.28. Proof. (4.49). 4.4
By Theorems 4.14, 4.15 and Lemma 4.31, see also Remark 4.16, (4.48),
Applications
Here we apply Theorems 4.32, 4.33, 4.34. We give Corollary 4.36. Suppose Assumptions 4.10 jor 4.17 or4.20, case m = 1. Q ∂f ([ai , bi ]) , for any (xj+1 , . . . , xn ) ∈ (. . . , xj+1 , . . . , xn ) ∈ L∞ i) Assume ∂x j n Q
i=1
[ai , bi ], all j = 1, . . . , n. Then
i=j+1
Z 1 f (x1 , x2 , . . . , xn ) − f (s1 , . . . , sn )ds1 · · · dsn n n Q Q (bi − ai ) i=1[ai ,bi ] i=1 s ! " 2 n X
1 1 xj − a j
(bj − aj ) + − ≤
12 bj − a j 2 j=1 #
∂f
× (. . . , xj+1 , . . . , xn ) Q . j ∂xj ∞, [ai ,bi ]
(4.85)
i=1
ii) Let pj , qj > 1 :
1 pj
+
1 qj
= 1; j = 1, . . . , n. In particular we suppose that ! j Y ∂f [ai , bi ] , (. . . , xj+1 , . . . , xn ) ∈ Lqj ∂xj i=1
for any (xj+1 , . . . , xn ) ∈
n Q
[ai , bi ], for all j = 1, . . . , n. Then
i=j+1
Z 1 f (x1 , x2 , . . . , xn ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [a ,b ] i i (bi − ai ) i=1 i=1 " pj 1/pj j−1 − q1 Z 1 n X Y xj − a j j 1− q1 j ≤ (bj − aj ) (bi − ai ) bj − aj − tj dtj 0 j=1 i=1 #
∂f
. (4.86) × j
∂xj (. . . , xj+1 , . . . , xn ) Q qj , [ai ,bi ] i=1
When pj = qj = 2, all j = 1, . . . , n, then
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
54
Z 1 f (x1 , . . . , xn ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [a ,b ] i i (bi − ai ) i=1 i=1 v v ! u " 2 ! u n q uj−1 Y X u1 1 x − a j j t + (bj − aj ) t (bi − ai ) − ≤ 12 bj − a j 2 i=1 j=1 #
∂f
(. . . , xj+1 , . . . , xn ) . (4.87) × j
Q ∂xj 2, [ai ,bi ] i=1
iii) Here assume for j = 1, . . . , n that
∂f (. . . , xj+1 , . . . , xn ) ∈ L1 ∂xj for any (xj+1 , . . . , xn ) ∈
n Q
j Y
!
[ai , bi ] ,
i=1
[ai , bi ]. Then
i=j+1
Z 1 f (x1 , . . . , xn ) − f (s1 , . . . , sn )ds1 · · · dsn n n Q Q (bi − ai ) i=1[ai ,bi ] i=1 # " (
n X
∂f
1
(. . . , xj+1 , . . . , xn ) ≤ j
Q j−1 ∂xj Q 1, [ai ,bi ] j=1 (bi − ai ) i=1 i=1 ) 1 aj + bj × + xj − . 2 2
(4.88)
Notice 4.37. We have for j = 1, . . . , n:
xj − a j =0 bj − a j = 1 1 = 2
λj :=
iff
xj = aj ,
λj
iff
x j = bj , aj + b j xj = . 2
λj
iff
(4.89)
We continue with Corollaries to Corollary 4.36. Corollary 4.38. Suppose Assumptions 4.10 or 4.17 or 4.20, Case m = 1, all xj = aj , j = 1, . . . , n. i) Assume ! j Y ∂f [ai , bi ] , all j = 1, . . . , n. (. . . , aj+1 , . . . , an ) ∈ L∞ ∂xj i=1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
55
Then Z 1 f (a1 , a2 , . . . , an ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [ai ,bi ] (b − a ) i i i=1 i=1 ) √ ( n
∂f
3 X
≤ . (bj − aj ) (. . . , a , . . . , a ) j+1 n j
∂xj
Q 3 j=1 [ai ,bi ] ∞,
(4.90)
i=1
ii) Let pj , qj > 1 :
1 pj
+
1 qj
= 1; j = 1, . . . , n. In particular we suppose that ! j Y ∂f [ai , bi ] , (. . . , aj+1 , . . . , an ) ∈ Lqj ∂xj i=1
for all j = 1, . . . , n. Then Z 1 f (a1 , . . . , an ) − f (s1 , . . . , sn )ds1 · · · dsn n n Q Q (bi − ai ) i=1[ai ,bi ] i=1
≤
j−1 1 " (bj − aj )1− qj Q (bi − ai )−1/qj n X i=1
(pj + 1)1/pj #
∂f
× . j
∂xj (. . . , aj+1 , . . . , an ) Q qj , [ai ,bi ] j=1
(4.91)
i=1
When pj = qj = 2, all j = 1, . . . , n, then Z 1 f (a1 , . . . , an ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [ai ,bi ] (b − a ) i i i=1 i=1 v " √ u n q uj−1 Y 3 X ≤ (bj − aj ) t (bi − ai ) 3 j=1 i=1 #!
∂f
. × j
∂xj (. . . , aj+1 , . . . , an ) Q [ai ,bi ] 2, i=1
iii) Here assume for j = 1, . . . , n, that
∂f (. . . , aj+1 , . . . , an ) ∈ L1 ∂xj Then
j Y
i=1
!
[ai , bi ] .
(4.92)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
56
Z 1 f (a1 , . . . , an ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [a ,b ] i i (bi − ai ) i=1 i=1 ) (
n
1 X (bj − aj + 1)
∂f (. . . , aj+1 , . . . , an ) . ≤ j
Q ∂xj 2 j=1 j−1 Q [ai ,bi ] 1, (bi − ai ) i=1
(4.93)
i=1
Corollary 4.39. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 1, all xj = bj , j = 1, . . . , n. i) Assume j Y
∂f (. . . , bj+1 , . . . , bn ) ∈ L∞ ∂xj
i=1
!
[ai , bi ] ,
for all j = 1, . . . , n. Then Z 1 f (b1 , . . . , bn ) − f (s1 , . . . , sn )ds1 · · · dsn n n Q Q (bi − ai ) i=1[ai ,bi ] i=1 # √ n "
∂f
3X
≤ (bj − aj ) . j
∂xj (. . . , bj+1 , . . . , bn ) Q 3 j=1 ∞, [ai ,bi ]
(4.94)
i=1
ii) Let pj , qj > 1 :
1 pj
+
1 qj
= 1; j = 1, . . . , n. In particular we assume that j Y
∂f (. . . , bj+1 , . . . , bn ) ∈ Lqj ∂xj
i=1
!
[ai , bi ] ,
for all j = 1, . . . , n. Then Z 1 f (b1 , . . . , bn ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [ai ,bi ] (b − a ) i i i=1 i=1 " j−1 − q1 n Y X j 1− 1 (bi − ai ) ≤ (pj + 1)−1/pj (bj − aj ) qj i=1
j=1
∂f
(. . . , b , . . . , b ) j+1 n
∂xj
qj ,
j Q
i=1
[ai ,bi ]
#
.
(4.95)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
57
When pj = qj = 2, all j = 1, . . . , n, then Z 1 f (b1 , . . . , bn ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [ai ,bi ] (b − a ) i i i=1 i=1 v # √ n "
u uj−1 Y
∂f
3X p t
(bi − ai ) bj − a j ≤ . j
∂xj (. . . , bj+1 , . . . , bn ) Q 3 j=1 2, [ai ,bi ] i=1 i=1
(4.96)
iii) Here suppose for j = 1, . . . , n that ∂f (. . . , bj+1 , . . . , bn ) ∈ L1 ∂xj Then
j Y
!
[ai , bi ] .
i=1
Z 1 f (b1 , . . . , bn ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [a ,b ] i i (bi − ai ) i=1 i=1 ( )
n
1 X (1 + bj − aj )
∂f (. . . , bj+1 , . . . , bn ) . ≤ j
Q ∂xj 2 j=1 j−1 Q [ai ,bi ] 1, (bi − ai ) i=1
(4.97)
i=1
Next come the multivariate midpoint rule inequalities.
Corollary 4.40. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 1, all a +b xj = j 2 j , j = 1, . . . , n. i) Assume ! j Y aj+1 + bj+1 ∂f an + b n ..., [ai , bi ] , all j = 1, . . . , n. ∈ L∞ ,..., ∂xj 2 2 i=1
Then Z a1 + b 1 a2 + b 2 1 an + b n f f (s , . . . , s )ds · · · ds − , , . . . , 1 n 1 n n n Q Q 2 2 2 [ai ,bi ] (b − a ) i i i=1 i=1
n X
1 (bj − aj ) ∂f . . . , aj+1 + bj+1 , . . . , an + bn (. 4.98) ≤ √ j
∂xj
Q 2 2 2 3 j=1 ∞, [ai ,bi ] i=1
ii) Let pj , qj > 1 :
1 pj
+
1 qj
= 1; j = 1, . . . , n. In particular we suppose that ! j Y aj+1 + bj+1 an + b n ∂f [ai , bi ] , ..., ,..., ∈ L qj ∂xj 2 2 i=1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
58
for all j = 1, . . . , n. Then Z a1 + b 1 1 an + b n f − f (s , . . . , s )ds · · · ds , . . . , 1 n 1 n n n Q Q 2 2 [ai ,bi ] (b − a ) i i i=1 i=1
!− q1j j−1 Y 1 − p1 1− q1 ≤ (bi − ai ) (pj+1 ) j (bj − aj ) j 2 j=1 i=1 #
∂f a + b a + b n n j+1 j+1
. × ,..., j
Q
∂xj . . . , 2 2 [ai ,bi ] qj , n X
"
(4.99)
i=1
When pj = qj = 2, all j = 1, . . . , n, then Z a1 + b 1 1 an + b n f − , . . . , f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q 2 2 [a ,b ] i i (bi − ai ) i=1 i=1 v " u n uj−1 Y 1 X p t √ (bi − ai ) bj − a j ≤ 2 3 j=1 i=1 #
∂f
a + b a + b j+1 j+1 n n
,..., × . (4.100) j
∂xj . . . ,
Q 2 2 2, [ai ,bi ] i=1
iii) Here assume for j = 1, . . . , n that ∂f aj+1 + bj+1 an + b n ∈ L1 ..., ,..., ∂xj 2 2
j Y
i=1
!
[ai , bi ] .
Then Z a1 + b 1 an + b n 1 f , . . . , f (s , . . . , s )ds · · · ds − 1 n 1 n n n Q Q 2 2 [ai ,bi ] (b − a ) i i i=1 i=1 ) (
n
∂f aj+1 + bj+1 1X an + b n 1
. . . . , ≤ , . . . , j
Q ∂xj 2 j=1 j−1 2 2 Q [ai ,bi ] 1, (bi − ai ) i=1 i=1
(4.101)
Next we treat the case of m = 2 and only for the norms k · k∞ , k · k2 , and specifically for λj = 0, 1, 12 , j = 1, . . . , n. The multivariate trapezoid rule estimates follow immediately. Corollary 4.41. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 2, all xj = aj , j = 1, . . . , n.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
59
i) Assume ∂2f (. . . , aj+1 , . . . , an ) ∈ L∞ ∂x2j
j Y
i=1
!
[ai , bi ] ,
all j = 1, . . . , n. Then f (a , a , . . . , a ) + f (b , a , . . . , a ) 1 2 n 1 2 n K2 := 2 Z 1 f (s1 , . . . , sn )ds1 · · · dsn − Q n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( n Z 1 1 X f (s1 , s2 , . . . , sj−1 , bj , aj+1 , . . . , an ) + Q j−1 2 j=2 j−1 Q [ai ,bi ] (bi − ai ) i=1 i=1 ) − f (s1 , . . . , sj−1 , aj , aj+1 , . . . , an ) ds1 · · · dsj−1 ( n
2
) X
1 ∂ f
≤ √ (bi − aj )2 . (4.102) j
∂x2 (. . . , aj+1 , . . . , an ) Q 2 30 j=1 j [ai ,bi ] ∞, i=1
ii) Suppose
∂ 2f (. . . , aj+1 , . . . , an ) ∈ L2 ∂x2j
j Y
!
[ai , bi ] ,
i=1
all j = 1, . . . , n. Then K2 ( n
) j−1 −1/2 2 Y X
∂ f
1 3/2
. (bi − ai ) (bj − aj ) ≤ √ j
∂x2 (. . . , aj+1 , . . . , an ) Q 2 30 j=1 j 2, [ai ,bi ] i=1 i=1
(4.103)
We continue with trapezoid rule estimates.
Corollary 4.42. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 2, all xj = bj , j = 1, . . . , n. i) Assume ∂2f (. . . , bj+1 , . . . , bn ) ∈ L∞ ∂x2j
j Y
i=1
!
[ai , bi ] ,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
60
all j = 1, . . . , n. Then Λ 2 f (b , . . . , b ) + f (a , b , . . . , b ) 1 n 1 2 n := − Q n 2 −
( n " 1 X
2
j=2
1
j−1 Q i=1
(bi − ai )
(Z
i=1
j−1 Q
1 (bi − ai )
Z
n Q
[ai ,bi ]
f (s1 , . . . , sn )ds1 · · · dsn
i=1
f (s1 , . . . , sj−1 , bj , bj+1 , . . . , bn ) [ai ,bi ]
i=1
#) − f (s1 , . . . , sj−1 , aj , bj+1 , . . . , bn ) ds1 · · · dsj−1
2
n X
∂ f 1 . (bj − aj )2 (. . . , bj+1 , . . . , bn ) ≤ √ j 2
Q ∂xj 2 30 j=1 ∞, [ai ,bi ]
(4.104)
i=1
ii) Assume
! j Y ∂2f [ai , bi ] , (. . . , bj+1 , . . . , bn ) ∈ L2 ∂x2j i=1 all j = 1, . . . , n. Then ( n " j−1 −1/2 X Y 1 3/2 Λ2 ≤ √ (bj − aj ) (bi − ai ) 2 30 j=1 i=1
#)
∂2f
. × 2 (. . . , bj+1 , . . . , bn j
Q
∂xj 2,
(4.105)
[ai ,bi ]
i=1
The multivariate midpoint rule estimates follow.
Corollary 4.43. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 2, all a +b xj = j 2 j , j = 1, . . . , n. i) Assume ! j Y ∂2f aj+1 + bj+1 an + b n [ai , bi ] , ∈ L∞ ..., ,..., ∂x2j 2 2 i=1 all j = 1, .. . , n. Then a +b an + b n 1 1 ,..., M2 := f 2 2 Z 1 f (s , . . . , s )ds · · · ds − Q 1 n 1 n n n Q (bi − ai ) i=1[ai ,bi ] i=1
#) ( n "
∂2f
X 1 a + b a + b
j+1 j+1 n n ≤ √ (bj − aj )2 2 . . . , . ,...,
j
∂xj
Q 2 2 8 5 j=1 ∞, [a ,b ] i
i
i=1
(4.106)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
ii) Suppose aj+1 + bj+1 an + b n ∂ 2f ∈ L2 . . . , , . . . , ∂x2j 2 2 all j = 1, . . . , n. Then
j Y
61
!
[ai , bi ] ,
i=1
( n " j−1 −1/2 2 Y X
∂ f 1
(bi − ai ) M2 ≤ √ (bj − aj )3/2
∂x2 8 5 j=1 j i=1 #) an + b n aj+1 + bj+1
,..., . ..., j
Q 2 2 2, [ai ,bi ]
(4.107)
i=1
We continue with trapezoid and midpoint rules inequalities for m = 3. Corollary 4.44. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 3, all xj = aj , j = 1, . . . , n. i) Assume ! j Y ∂3f [ai , bi ] , (. . . , aj+1 , . . . , an ) ∈ L∞ ∂x3j i=1 all j = 1, . . . , n. Then f (a , . . . , a ) + f (b , a , . . . , a ) 1 n 1 2 n K3 := 2 Z 1 − Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( X 2 n X (bj − aj )k−1 1 Bk (0) − j−1 k! Q j=1 k=1 (bi − ai ) with j=k6=1 ×
Z
j−1 Q i=1
[ai ,bi ]
i=1
∂ k−1 f (s1 , . . . , sj−1 , bj , aj+1 , . . . , an ) ∂xjk−1
!) ∂ k−1 f − (s1 , . . . , sj−1 , aj , aj+1 , . . . , an ) ds1 · · · dsj−1 ∂xjk−1
3
n X
∂ f 1
√ (bj − aj )3 . ≤ j
∂x3 (. . . , an+1 , . . . , an ) Q 12 210 j=1 j [ai ,bi ] ∞, i=1
(4.108)
ii) Assume ∂ 3f (. . . , aj+1 , . . . , an ) ∈ L2 ∂x3j
j Y
i=1
!
[ai , bi ] ,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
62
all j = 1, . . . , n. Then
" −1/2 3 j−1 n Y X
∂ f 1 5/2
(bi − ai ) K3 ≤ √ (bj − aj )
∂x3 12 210 j=1 j i=1
. (. . . , aj+1 , . . . , an ) j
Q 2,
(4.109)
[ai ,bi ]
i=1
Corollary 4.45. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 3, all xj = bj , j = 1, . . . , n. i) Assume ! j Y ∂3f [ai , bi ] , (. . . , bj+1 , . . . , bn ) ∈ L∞ ∂x3j i=1 all j = 1, . . . , n. Then f (b , . . . , b ) + f (a , b , . . . , b ) 1 n 1 2 n Λ3 := 2 Z 1 − Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( 2 ( n X (bj − aj )k−1 X 1 Bk (1) − j−1 k! Q j=1 k=1 (bi − ai ) with j=k6=1 ×
Z
j−1 Q i=1
[ai ,bi ]
i=1
∂ k−1 f (s1 , s2 , . . . , sj−1 , bj , bj+1 , . . . , bn ) ∂xjk−1
!)) ∂ k−1 f − (s , s , . . . , s , a , b , . . . , b ) ds · · · ds 1 2 j−1 j j+1 n 1 j−1 ∂xjk−1
3
n X
1 . (4.110) (bj − aj )3 ∂ f (. . . , bj+1 , . . . , bn ) √ ≤ j 3
Q ∂xj 12 210 j=1 ∞, [ai ,bi ] i=1
ii) Suppose
∂3f (. . . , bj+1 , . . . , bn ) ∈ L2 ∂x3j all j = 1, . . . , n. Then
j Y
!
[ai , bi ] ,
i=1
" j−1 −1/2 3 n Y X
∂ f 1 5/2
(bi − ai ) Λ3 ≤ √ (bj − aj )
∂x3 12 210 j=1 j i=1
. (. . . , bj+1 , . . . , bn ) j
Q 2,
[ai ,bi ]
i=1
(4.111)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
63
Corollary 4.46. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 3, all a +b xj = j 2 j , j = 1, . . . , n. i) Assume ! j Y ∂3f aj+1 + bj+1 an + b n ..., ,..., ∈ L∞ [ai , bi ] , ∂x3j 2 2 i=1
all j = 1, . . . , n. Then a +b an + b n 1 1 ,..., M3 := f 2 2 Z 1 − Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( Z n (bj − aj ) 1 X ∂f + Q j−1 24 j=1 j−1 ∂x Q j [ai ,bi ] (bi − ai ) i=1 i=1 aj+1 + bj+1 an + b n (s1 , s2 , . . . , sj−1 , bj , ,..., 2 2 ) ! aj+1 + bj+1 an + b n ∂f s1 , . . . , sj−1 , aj , ,..., ds1 · · · dsj−1 − ∂xj 2 2 # "
3 n X
1 aj+1 + bj+1 an + b n 3 ∂ f
√ (bj − aj ) 3 . . . , . ≤ ,..., j
Q ∂xj 2 2 12 210 j=1 [ai ,bi ] ∞, i=1
(4.112)
ii) Suppose ∂3f ∂x3j
aj+1 + bj+1 an + b n ..., ,..., 2 2
all j = 1, . . . , n. Then
∈ L2
j Y
!
[ai , bi ] ,
i=1
" j−1 −1/2 n X Y 1 5/2 M3 ≤ √ (bj − aj ) (bi − ai ) 12 210 j=1 i=1
∂3f aj+1 + bj+1 an + b n
,..., × 3 ...,
j
Q
∂xj 2 2 2,
[ai ,bi ]
#
.
(4.113)
i=1
Next we present trapezoid and midpoint rules inequalities for m = 4. Corollary 4.47. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 4, all xj = aj , j = 1, . . . , n. i) Assume ! j Y ∂4f [ai , bi ] , (. . . , aj+1 , . . . , an ) ∈ L∞ ∂x4j i=1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
64
all j = 1, . . . , n. Then f (a , . . . , a ) + f (b , a , . . . , a ) 1 n 1 2 n K3 = 2 Z 1 −Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( ( 2 n X X (bj − aj )k−1 1 − j−1 k! Q j=1 (bi − ai ) k=1 (with j=k6=1) i=1
∂ k−1 f × Bk (0) j−1 (s1 , . . . , sj−1 , bj , aj+1 , . . . , an k−1 Q [ai ,bi ] ∂xj i=1 !)) ! ∂ k−1 f (s , s , . . . , s , a , a , . . . , a ) ds · · · ds − 1 2 j−1 j j+1 n 1 j−1 ∂xjk−1
" # n
∂4f
X 1
≤ √ (bj − aj )4 4 (. . . , aj+1 , . . . , an ) . (4.114) j
∂xj
Q 24 630 Z
j=1
∞,
[ai ,bi ]
i=1
ii) Assume j Y
∂ 4f (. . . , aj+1 , . . . , an ) ∈ L2 ∂x4j
!
[ai , bi ] ,
i=1
all j = 1, . . . , n. Then " −1/2 4 j−1 n X Y
∂ f 1 7/2
K3 ≤ √ (bj − aj ) (bi − ai )
∂x4 24 630 j=1 j i=1 #
(. . . , aj+1 , . . . , an ) . j
Q 2,
(4.115)
[ai ,bi ]
i=1
Corollary 4.48. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 4, all xj = bj , j = 1, . . . , n. i) Assume ∂4f (. . . , bj+1 , . . . , bn ) ∈ L∞ ∂x4j
j Y
i=1
!
[ai , bi ] ,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
65
all j = 1, . . . , n. Then f (b , . . . , b ) + f (a , b , . . . , b ) 1 n 1 2 n Λ3 = 2 Z 1 − Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( ( 2 n X (bj − aj )k−1 X 1 − j−1 k! Q j=1 (bi − ai ) k=1 (with j=k6=1) i=1
Z
∂ k−1 f (s1 , s2 , . . . , sj−1 , bj , bj+1 , . . . , bn ) Qj−1 ∂xjk−1 i=1 [ai ,bi ] !)) ∂ k−1 f − (s , s , . . . , s , a , b , . . . , b ) ds · · · ds 1 2 j−1 j j+1 n 1 j−1 ∂xjk−1 ( n " #)
4
X
∂ f
1
(bj − aj )4 . ≤ √ j
4 (. . . , bj+1 , . . . , bn ) Q ∂x 24 630 j=1 j ∞, [ai ,bi ] × Bk (1)
(4.116)
i=1
ii) Suppose j Y
∂4f (. . . , bj+1 , . . . , bn ) ∈ L2 ∂x4j
!
[ai , bi ] ,
i=1
all j = 1, . . . , n. Then " −1/2 4 j−1 n Y X
∂ f 1
Λ3 ≤ √ (bj − aj )7/2 (bi − ai )
∂x4 24 630 j=1 j i=1 #
(. . . , bj+1 , . . . , bn ) . j
Q 2,
(4.117)
[ai ,bi ]
i=1
Corollary 4.49. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 4, all a +b xj = j 2 j , j = 1, . . . , n. i) Assume ∂4f ∂x4j
aj+1 + bj+1 an + b n ..., ,..., 2 2
∈ L∞
j Y
i=1
!
[ai , bi ] ,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
66
all j = 1, . . . , n. Then a +b an + b n 1 1 M3 = f ,..., 2 2 Z 1 −Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( n ( Z (bj − aj ) 1 X ∂f s1 , s2 , . . . , sj−1 , + Q j−1 24 j=1 j−1 Q [ai ,bi ] ∂xj (bi − ai ) i=1 i=1 aj+1 + bj+1 an + b n bj , ,..., 2 2 !)) ! aj+1 + bj+1 an + b n ∂f s1 , s2 , . . . , sj−1 , aj , ds1 · · · dsj−1 ,..., − ∂xj 2 2 ( n " r
∂4f 1 107 X ≤ (bj − aj )4
∂x4 152 35 j=1 j #) aj+1 + bj+1 an + b n
. (4.118) ..., ,..., j
Q 2 2 ∞, [ai ,bi ] i=1
ii) Assume
∂4f ∂x4j
aj+1 + bj+1 an + b n ..., ,..., 2 2
∈ L2
j Y
!
[ai , bi ] ,
i=1
all j = 1, . . . , n. Then !−1/2 ( n " r j−1 Y 1 107 X 7/2 (bi − ai ) (bj − aj ) M3 ≤ 1152 35 j=1 i=1
#)
∂4f aj+1 + bj+1 an + b n
× 4 ..., ,..., .
j
∂xj
Q 2 2 2,
(4.119)
[ai ,bi ]
i=1
Also we present trapezoid and midpoint rules inequalities for m = 5. Corollary 4.50. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 5, all xj = aj , j = 1, . . . , n. i) Assume ! j Y ∂5f [ai , bi ] , (. . . , aj+1 , . . . , an ) ∈ L∞ ∂x5j i=1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
67
all j = 1, . . . , n. Then f (a , . . . , a ) + f (b , a , . . . , a ) 1 n 1 2 n K5 := 2 Z 1 − Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( ( n X X (bj − aj )k−1 1 − j−1 k! Q j=1 (bi − ai ) k∈{1,2,4} (with j=k6=1) Z
i=1
∂ k−1 f (s , . . . , sj−1 , bj , aj+1 , . . . , an ) j−1 k−1 1 Q [ai ,bi ] ∂xj i=1 ! !)) k−1 ∂ f − (s , . . . , sj−1 , aj , aj+1 , . . . , an ) ds1 · · · dsj−1 k−1 1 ∂xj ( n " #) r
5
5 X 1 5 ∂ f
. (bj − aj ) 5 (. . . , aj+1 , . . . , an ) Q ≤ j 720 462 j=1 ∂xj [ai ,bi ] ∞, × Bk (0)
i=1
(4.120)
ii) Suppose ∂ 5f (. . . , aj+1 , . . . , an ) ∈ L2 ∂x5j
j Y
!
[ai , bi ] ,
i=1
all j = 1, . . . , n. Then ( n " r −1/2 j−1 Y 5 X 1 (bi − ai ) (bj − aj )9/2 K5 ≤ 720 462 j=1 i=1
#)
∂5f
× 5 (. . . , aj+1 , . . . , an ) . j
∂xj
Q 2,
(4.121)
[ai ,bi ]
i=1
Corollary 4.51. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 5, all xj = bj , j = 1, . . . , n. i) Assume ∂5f (. . . , bj+1 , . . . , bn ) ∈ L∞ ∂x5j
j Y
i=1
!
[ai , bi ] ,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
68
all j = 1, . . . , n. Then f (b , . . . , b ) + f (a , b , . . . , b ) 1 n 1 2 n Λ5 := 2 Z 1 − Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( ( n X X (bj − aj )k−1 1 − j−1 k! Q j=1 (bi − ai ) k∈{1,2,4} (with j=k6=1) Z
i=1
∂ k−1 f (s , . . . , sj−1 , bj , bj+1 , . . . , bn ) j−1 k−1 1 Q [ai ,bi ] ∂xj i=1 ! !)) k−1 ∂ f − (s , . . . , sj−1 , aj , bj+1 , . . . , bn ) ds1 · · · dsj−1 k−1 1 ∂xj ( n " #) r
5
5 X 1 5 ∂ f
. (bj − aj ) 5 (. . . , bj+1 , . . . , bn ) Q ≤ j 720 462 j=1 ∂xj [ai ,bi ] ∞, × Bk (1)
i=1
(4.122)
ii) Suppose j Y
∂5f (. . . , bj+1 , . . . , bn ) ∈ L2 ∂x5j
!
[ai , bi ] ,
i=1
all j = 1, . . . , n. Then ( n " r −1/2 j−1 Y 5 X 1 (bi − ai ) (bj − aj )9/2 Λ5 ≤ 720 462 j=1 i=1
#)
∂5f
× 5 (. . . , bj+1 , . . . , bn ) . j
∂xj
Q 2,
(4.123)
[ai ,bi ]
i=1
Corollary 4.52. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 5, all a +b xj = j 2 j , j = 1, . . . , n. i) Assume ∂5f ∂x5j
aj+1 + bj+1 an + b n ..., ,..., 2 2
∈ L∞
j Y
i=1
!
[ai , bi ] ,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
69
all j = 1, . . . , n. Then a +b an + b n 1 1 ,..., M5 := f 2 2 Z 1 − Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( ( n X (bj − aj )k−1 X 1 − j−1 k! Q j=1 (bi − ai ) k∈{2,4} i=1
Z an + b n aj+1 + bj+1 1 ∂ k−1 f ,..., s1 , . . . , sj−1 , bj , × Bk j−1 k−1 Q 2 2 2 [ai ,bi ] ∂xj i=1 !)) ! ∂ k−1 f an + b n aj+1 + bj+1 ds · · · ds − , . . . , s , . . . , s , a , 1 j−1 1 j−1 j 2 2 ∂xjk−1 ( n " r
∂5f 1 5 X ≤ (bj − aj )5
∂x5 . . . , 720 462 j=1 j #) aj+1 + bj+1 an + b n
. (4.124) ,..., j
Q 2 2 ∞, [ai ,bi ] i=1
ii) Assume
∂5f ∂x5j
aj+1 + bj+1 an + b n ..., ,..., 2 2
∈ L2
j Y
!
[ai , bi ] ,
i=1
all j = 1, . . . , n. Then ( n " r −1/2 j−1 Y 1 5 X 9/2 (bi − ai ) M5 ≤ (bj − aj ) 720 462 j=1 i=1
∂5f aj+1 + bj+1 an + b n
× 5 ..., ,...,
j
∂xj
Q 2 2 2,
[ai ,bi ]
#)
.
(4.125)
i=1
Finally we present trapezoid and midpoint rules inequalities for m = 6.
Corollary 4.53. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 6, all xj = aj , j = 1, . . . , n. i) Assume ! j Y ∂6f [ai , bi ] , (. . . , aj+1 , . . . , an ) ∈ L∞ ∂x6j i=1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
70
all j = 1, . . . , n. Then f (a , . . . , a ) + f (b , a , . . . , a ) 1 n 1 2 n K5 = 2 Z 1 f (s1 , . . . , sn )ds1 · · · dsn − Q n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( ( n X X (bj − aj )k−1 1 − j−1 k! Q j=1 k∈{1,2,4} (b − a ) i i (with j=k6=1) Z
i=1
∂ k−1 f (s , . . . , sj−1 , bj , aj+1 , . . . , an ) k−1 1 [ai ,bi ] ∂xj i=1 !)) ! k−1 ∂ f − (s , . . . , s , a , a , . . . , a ) ds · · · ds 1 j−1 j j+1 n 1 j−1 k−1 ∂xj #) ( n " r
6
1 101 X 6 ∂ f . (bj − aj ) 6 (. . . , aj+1 , . . . , an ) ≤ j
Q 1440 30030 j=1 ∂xj [ai ,bi ] ∞, × Bk (0)
j−1 Q
i=1
(4.126)
ii) Suppose ∂ 6f (. . . , aj+1 , . . . , an ) ∈ L2 ∂x6j
j Y
!
[ai , bi ] ,
i=1
all j = 1, . . . , n. Then ( n " r −1/2 j−1 Y 1 101 X (bi − ai ) K5 ≤ (bj − aj )11/2 1440 30030 j=1 i=1
#)
∂6f
× 6 (. . . , aj+1 , . . . , an ) . j
∂xj
Q 2,
(4.127)
[ai ,bi ]
i=1
Corollary 4.54. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 6, all xj = bj , j = 1, . . . , n. i) Assume ∂6f (. . . , bj+1 , . . . , bn ) ∈ L∞ ∂x6j all j = 1, . . . , n. Then
j Y
i=1
!
[ai , bi ] ,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
71
f (b , . . . , b ) + f (a , b , . . . , b ) 1 n 1 2 n Λ5 = 2 Z 1 − Q f (s1 , . . . , sn )ds1 · · · dsn n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( ( X X (bj − aj )k−1 1 − j−1 k! Q j=1 k∈{1,2,4} (b − a ) i i (with j=k6=1) Z
i=1
∂ k−1 f (s , . . . , sj−1 , bj , bj+1 , . . . , bn ) k−1 1 [ai ,bi ] ∂xj i=1 !)) ! k−1 ∂ f (s , . . . , s , a , b , . . . , b ) ds · · · ds − 1 j−1 j j+1 n 1 j−1 k−1 ∂xj ( n " #) r
6
101 X 1 6 ∂ f
(bj − aj ) 6 (. . . , bj+1 , . . . , bn ) Q . ≤ j 1440 30030 j=1 ∂xj ∞, [ai ,bi ] × Bk (1)
j−1 Q
i=1
(4.128)
ii) Assume j Y
∂6f (. . . , bj+1 , . . . , bn ) ∈ L2 ∂x6j
!
[ai , bi ] ,
i=1
all j = 1, . . . , n. Then ( n " r j−1 −1/2 Y 1 101 X Λ5 ≤ (bi − ai ) (bj − aj )11/2 1440 30030 j=1 i=1
#)
∂6f
. × 6 (. . . , bj+1 , . . . , bn ) j
∂xj
Q 2,
(4.129)
[ai ,bi ]
i=1
Corollary 4.55. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 6, all a +b xj = j 2 j , j = 1, . . . , n. i) Assume ∂6f ∂x6j
aj+1 + bj+1 an + b n ..., ,..., 2 2
∈ L∞
j Y
i=1
!
[ai , bi ] ,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
72
all j = 1, . . . , n. Then a +b an + b n 1 1 M 5 = f ,..., 2 2 Z 1 f (s1 , . . . , sn )ds1 · · · dsn − Q n n Q (bi − ai ) i=1[ai ,bi ] i=1 ( ( n X X (bj − aj )k−1 1 − j−1 k! Q j=1 (bi − ai ) k∈{2,4} i=1
Z an + b n aj+1 + bj+1 1 ∂ k−1 f , . . . , s , . . . , s , b , × Bk 1 j−1 j j−1 k−1 Q 2 2 2 [ai ,bi ] ∂xj i=1 !)) ! aj+1 + bj+1 an + b n ∂ k−1 f ds · · · ds s , . . . , s , a , , . . . , − 1 j−1 1 j−1 j k−1 2 2 ∂xj ( n " r 1 7081 X ≤ (bj − aj )6 46080 2145 j=1 #)
6
∂ f
a + b a + b j+1 j+1 n n
. (4.130) ,..., × j
∂x6 . . . ,
Q 2 2 j ∞, [ai ,bi ] i=1
ii) Suppose
aj+1 + bj+1 an + b n ∂ 6f ..., ,..., ∈ L2 ∂x6j 2 2
j Y
!
[ai , bi ] ,
i=1
all j = 1, . . . , n. Then ( n " r j−1 −1/2 Y 1 7081 X 11/2 M5 ≤ (bj − aj ) (bi − ai ) 46080 2145 j=1 i=1
#)
∂6f aj+1 + bj+1 an + b n
. × 6 ..., ,...,
j
Q
∂xj 2 2 2,
(4.131)
[ai ,bi ]
i=1
At the end we give a simplified special case of Theorems 4.32 and 4.33. Corollary 4.56. Suppose Assumptions 4.10 or 4.17 or 4.20. We further assume that ∂ `f ∂`f (. . . , bj , . . .) = (. . . , aj , . . .), ` ∂xj ∂x`j
(4.132)
for all j = 1, . . . , n and all ` = 0, 1, . . . , m − 2. Here m, n ∈ N, xi ∈ [ai , bi ], i = 1, 2, . . . , n.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
i) In particular we assume that ∂ mf (. . . , xj+1 , . . . , xn ) ∈ L∞ ∂xm j
j Y
i=1
!
[ai , bi ] ,
n Q for any (xj+1 , . . . , xn ) ∈ [ai , bi ], all j = 1, . . . , n. Then i=j+1 Z 1 f (x1 , . . . , xn ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [ai ,bi ] (b − a ) i i i=1 i=1 s ! ( n " xj − a j 1 X (m!)2 2 (bj − aj )m ≤ |B2m | + Bm m! j=1 (2m)! bj − a j
#)
∂mf
, × m (. . . , xj+1 , . . . , xn ) j
Q
∂xj ∞,
73
(4.133)
[ai ,bi ]
i=1
true ∀(x1 , . . . , xn ) ∈ ii) Suppose
n Q
[ai , bi ].
i=1
∂mf (. . . , xj+1 , . . . , xn ) ∈ L2 ∂xm j
j Y
i=1
!
[ai , bi ] ,
n Q for any (xj+1 , . . . , xn ) ∈ [ai , bi ], for all j = 1, . . . , n. Then i=j+1 Z 1 f (x1 , . . . , xn ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [a ,b ] i i (bi − ai ) i=1 i=1 s ( n " −1/2 j−1 ! Y 1 X (m!)2 xj − a j m− 12 2 ≤ (bi − ai ) (bj − aj ) |B2m | + Bm m! j=1 (2m)! bj − a j i=1
#)
∂ mf
× m (. . . , xj+1 , . . . , xn ) , (4.134) j
∂xj
Q 2,
[ai ,bi ]
i=1
true ∀(x1 , . . . , xn ) ∈ Proof.
n Q
[ai , bi ].
i=1
Clearly here Aj = 0, j = 1, . . . , n. Then proof is obvious.
Similarly as in Corollary 4.56 we obtain Corollary 4.57. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 6, all xj = aj , j = 1, . . . , n. Also assume ∂`f ∂`f (. . . , b , a , . . . , a ) = (. . . , aj , aj+1 , . . . , an ), (4.135) j j+1 n ∂x`j ∂x`j
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
74
for all j = 1, . . . , n, and all ` = 0, 1, 3. i) Assume ∂6f (. . . , aj+1 , . . . , an ) ∈ L∞ ∂x6j
j Y
!
[ai , bi ] , all j = 1, . . . , n.
i=1
Then Z 1 f (a1 , . . . , an ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [ai ,bi ] (b − a ) i i i=1 i=1 " # r
6 n
101 X ∂ f 1
(. . . , a , . . . , a ) (bj − aj )6 . ≤ j+1 n j
Q
∂x6 1440 30030 j=1 j ∞, [ai ,bi ] i=1
(4.136)
ii) Suppose ∂6f (. . . , aj+1 , . . . , an ) ∈ L2 ∂x6j Then
j Y
!
[ai , bi ] , all j = 1, . . . , n.
i=1
Z 1 f (a1 , . . . , an ) − f (s1 , . . . , sn )ds1 · · · dsn n n Q Q (bi − ai ) i=1[ai ,bi ] i=1 ( n r j−1 −1/2 Y 1 101 X ≤ (bi − ai ) (bj − aj )11/2 1440 30030 j=1 i=1
#)
∂6f
. × 6 (. . . , aj+1 , . . . , an ) j
∂xj
Q 2,
(4.137)
[ai ,bi ]
i=1
Proof.
By Corollary 4.53.
Corollary 4.58. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 6, all xj = bj , j = 1, . . . , n. Also assume ∂ `f ∂`f (. . . , b , b , . . . , b ) = (. . . , aj , bj+1 , . . . , bn ), j j+1 n ∂x`j ∂x`j for all j = 1, . . . , n and all ` = 0, 1, 3. i) Assume ∂6f (. . . , bj+1 , . . . , bn ) ∈ L∞ ∂x6j
j Y
i=1
!
[ai , bi ] , all j = 1, . . . , n.
(4.138)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
75
Then Z 1 f (b1 , . . . , bn ) − f (s1 , . . . , sn )ds1 · · · dsn n n Q Q (bi − ai ) i=1[ai ,bi ] i=1 # " r
6 n X
101 ∂ f 1 (. . . , bj+1 , . . . , bn ) . (bj − aj )6 ≤ j 6
Q 1440 30030 j=1 ∂xj ∞, [ai ,bi ] i=1
(4.139)
ii) Suppose ∂ 6f (. . . , bj+1 , . . . , bn ) ∈ L2 ∂x6j
j Y
!
[ai , bi ] , all j = 1, . . . , n.
i=1
Then Z 1 f (b1 , . . . , bn ) − f (s , . . . , s )ds · · · ds 1 n 1 n n n Q Q [ai ,bi ] (b − a ) i i i=1 i=1 ( n " r j−1 −1/2 Y 101 X 1 (bj − aj )11/2 (bi − ai ) ≤ 1440 30030 j=1 i=1
#)
∂6f
. × 6 (. . . , bj+1 , . . . , bn ) j
Q
∂xj 2,
(4.140)
[ai ,bi ]
i=1
Proof.
By Corollary 4.54.
Corollary 4.59. Suppose Assumptions 4.10 or 4.17 or 4.20, case m = 6, all a +b xj = j 2 j , j = 1, . . . , n. Also assume ∂`f an + b n an + b n aj+1 + bj+1 aj+1 + bj+1 ∂`f , . . . , , . . . , . . . , b , = . . . , a , , j j 2 2 2 2 ∂x`j ∂x`j (4.141) for all j = 1, . . . , n and ` = 1, 3. i) Assume aj+1 + bj+1 an + b n ∂6f . . . , , . . . , ∈ L∞ ∂x6j 2 2
j Y
i=1
!
[ai , bi ] ,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
76
all j = 1, . . . , n. Then a +b an + b n 1 1 − Q ,..., f n 2 2 ≤
(
r
i=1
"
1 (bi − ai )
Z
f (s , . . . , s )ds · · · ds 1 n 1 n n Q [ai ,bi ]
i=1
n 1 7081 X (bj − aj )6 46080 2145 j=1
∂6f aj+1 + bj+1 an + b n
,..., × 6 ...,
∂xj 2 2
∞,
j Q
[ai ,bi ]
#)
.
(4.142)
i=1
ii) Suppose
∂6f ∂x6j
an + b n aj+1 + bj+1 ,..., ..., 2 2
all j = 1, . . . , n. Then a +b an + b n 1 1 − Q ,..., f n 2 2 ≤
r
(
"
i=1
1 (bi − ai )
Z
∈ L2
j Y
!
[ai , bi ] ,
i=1
f (s , . . . , s )ds · · · ds 1 n 1 n n Q [ai ,bi ]
i=1
j−1 −1/2 n Y 7081 X 1 (bj − aj )11/2 (bi − ai ) 46080 2145 j=1 i=1
#)
∂6f aj+1 + bj+1 aj + b n
,..., . × 6 ...,
j
Q
∂xj 2 2 2,
(4.143)
[ai ,bi ]
i=1
Proof.
By Corollary 4.55.
One can apply similar conditions to (4.132) for the cases of m = 2, 3, 4, 5 and simplify a lot the results of Corollaries 4.41, 4.42 and of Corollaries 4.44 – 4.52, exactly as we did in Corollaries 4.56–4.59 for general m ∈ N and m = 6, etc. 4.5
Sharpness
We need to include f Theorem 4.60. Suppose Assumptions 4.10 or 4.17 or 4.20. Let Em (x1 , . . . , xn ) as in (4.44), m ∈ N. In particular we assume that ! j j Y ∂ m f z}|{ [ai , bi ] , · · · , xj+1 , . . . , xn ∈ L∞ ∂xm j i=1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
for any (xj+1 , . . . , xn ) ∈
n Q
77
[ai , bi ], all j = 1, . . . , n. Then
i=j+1
" Z 1 n X xj − a j 1 f m |Em (x1 , . . . , xn )| ≤ (bj − aj ) Bm bj − aj − Bm (tj ) dtj m! j=1 0
#
∂mf
. (4.144) × m (. . . , xj+1 , . . . , xn ) j
Q
∂xj ∞,
[ai ,bi ]
i=1
Proof.
By Remark 4.23, see (4.55).
We give the important f Corollary 4.61. Suppose Assumptions 4.10 or 4.17 or 4.20. Let Em (x1 , . . . , xn ) as in (4.44), m ∈ N. In particular we assume that ! j j Y ∂ m f z}|{ [ai , bi ] , · · · , xj+1 , . . . , xn ∈ L∞ ∂xm j i=1
for any (xj+1 , . . . , xn ) ∈ L∞
n Q
i=1
n Q
[ai , bi ], all j = 1, . . . , n. And also suppose
i=j+1
∂mf ∂xm j
∈
[ai , bi ] , j = 1, . . . , n − 1. Call
(
)
∂mf
. Dm (f ) := max
1≤j≤n ∂xm j ∞
Then f |Em (x1 , . . . , xn )|
(4.145)
" !# Z 1 n x − a Dm (f ) X j j m Bm . − Bm (tj ) dtj (bj − aj ) ≤ m! j=1 bj − a j 0
(4.146)
Comment 4.62. We observe that (see also [98]) Z 1 a +b 2 xj − j 2 j 1 |B1 (λj ) − B1 (tj )|dtj = + I1 (λj ) := , 4 (bj − aj )2 0
where λj =
xj −aj bj −aj ,
(4.147)
j = 1, . . . , n. Notice that
max I1 (λj ) = I1 (0) = I1 (1) =
λj ∈[0,1]
1 , 2
i.e. when xj = aj or bj . Thus we have Corollary 4.63. All here assumed as in Corollary 4.61 when m = 1. Then Z 1 f (x1 , . . . , xn ) − f (s1 , . . . , sn )ds1 · · · dsn n n Q Q (bi − ai ) i=1[ai ,bi ] i=1
(4.148)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
78
≤
n D1 (f ) X (bj − aj ) . 2 j=1
(4.149)
Inequality (4.149) is sharp, that is attained by f1 (s1 , . . . , sn ) := xj = aj , j = 1, . . . , n, and by f2 (s1 , . . . , sn ) :=
n P
j=1
1, . . . , n.
Proof. i) Case of xj = aj , j = 1, . . . , n. Then
∂f1
∂xj ∞ = 1 and D1 (f1 ) = 1. Clearly then we have L.H.S.(4.149) = R.H.S.(4.149) =
proving sharpness. ii) Case of xj = bj , j = 1, . . . , n. Then and D1 (f2 ) = 1. Clearly we have
∂f2 ∂xj
n P
j=1
(sj − aj ) when
(bj − sj ) when xj = bj , j = ∂f1 ∂xj
= 1, j = 1, . . . , n, i.e.
n
1X (bj − aj ), 2 j=1
∂f2
=1 = −1, j = 1, . . . , n, i.e. ∂x ∞ j n
L.H.S.(4.149) = R.H.S.(4.149) = proving again sharpness.
1X (bj − aj ), 2 j=1
Comment 4.64. We observe that ([98]) Z 1 8 1 I2 (λj ) := |B2 (λj ) − B2 (tj )|dtj = δj3 (x) − δj2 (x) + , j = 1, . . . , n, (4.150) 3 12 0 where
xj − aj +bj 2 δj (xj ) := , bj − a j
xj ∈ [aj , bj ].
Also from [98] we have that
max I2 (λj ) = I2 (0) = I2 (1) =
0≤λj ≤1
1 , 6
(4.151)
i.e. when xj = aj or bj . We continue with Corollary 4.65. All here assumed as in Corollary 4.61 when m = 2. Then n
|E2f (x1 , . . . , xn )| ≤
D2 (f ) X (bj − aj )2 . 12 j=1
(4.152)
n P Inequality (4.152) is sharp, that is attained by f1 (s1 , . . . , sn ) := (sj − aj )2 when j=1 Pn 2 xj = aj , j = 1, . . . , n and by f2 (s1 , . . . , sn ) := j=1 (sj − bj ) when xj = bj , j = 1, . . . , n.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multidimensional Euler Identity and Optimal Multidimensional Ostrowski Inequalities
Proof.
∂2f
21 ∂xj
∞
i) Case of xj = aj , j = 1, . . . , n. Then
∂f1 ∂xj
∂ 2 f1 ∂x2j
= 2(sj − aj ),
79
= 2, and
= 2, with D2 (f ) = 2. Clearly then we have n
L.H.S.(4.152) = R.H.S.(4.152) =
1X (bi − aj )2 , 6 j=1
proving sharpness. ∂f2 ii) Case of xj = bj , j = 1, . . . , n. Then ∂x = 2(sj − bj ), j
∂2f
22 = 2, with D2 (f ) = 2. Clearly again we have ∂x ∞
∂ 2 f2 ∂x2j
= 2, and
j
n
L.H.S.(4.152) = R.H.S.(4.152) =
1X (bi − aj )2 , 6 j=1
proving again sharpness.
Comment 4.66. By [10] we have that √ ! 3− 3 = I3 max I3 (λj ) = I3 0≤λj ≤1 6 where I3 (λj ) :=
Z
√ ! √ 3+ 3 3 = , 6 36
(4.153)
j = 1, . . . , n.
(4.154)
1 0
|B3 (λj ) − B3 (tj )|dtj ,
Consequently we have Corollary 4.67. All here assumed as in Corollary 4.61 when m = 3. Then √ n 3D3 (f ) X f |E3 (x1 , . . . , xn )| ≤ (bj − aj )3 . (4.155) 216 j=1 Comment 4.68. We call Im (λj ) := where λj :=
xj −aj bj −aj ,
Z
1 0
|Bm (λj ) − Bm (tj )|dtj ,
(4.156)
j = 1, . . . , n, m ∈ N. In [35] we found that max I4 (λj ) = I4 (0) = I4 (1) =
λj ∈[0,1]
1 . 30
(4.157)
So we give Corollary 4.69. All here supposed as in Corollary 4.61 when m = 4. Then n D4 (f ) X f (bj − aj )4 . |E4 (x1 , . . . , xn )| ≤ (4.158) 720 j=1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
80
Inequality (4.158) is sharp, that is attained by f1 (s1 , . . . , sn ) := xj = aj , j = 1, . . . , n and by f2 (s1 , . . . , sn ) :=
n P
j=1
1, . . . , n. Proof.
n P
j=1 4
(sj − aj )4 when
(sj − bj ) when xj = bj , j =
Case of xj = aj , j = 1, . . . , n. Then
∂f1 ∂ 2 f1 ∂ 3 f1 ∂ 4 f1 = 4(sj − aj )3 , = 12(sj − aj )2 , = 24(sj − aj ), = 24, 2 3 ∂xj ∂xj ∂xj ∂x4j
4 with ∂∂xf41 ∞ = 24 and D4 (f ) = 24. Clearly then we have j n 1 X L.H.S.(4.158) = R.H.S.(4.158) = (bj − aj )4 , 30 j=1
proving sharpness. ii) Case xj = bj , j = 1, . . . , n. Then
∂ 2 f2 ∂ 3 f2 ∂ 4 f2 ∂f2 2 = 4(sj − bj )3 , = 12(s − b ) , = 24(s − b ), = 24, j j j j ∂xj ∂x2j ∂x3j ∂x4j with D4 (f2 ) = 24. Clearly again we have
n 1 X L.H.S.(4.158) = R.H.S.(4.158) = (bj − aj )4 , 30 j=1
proving again sharpness.
Comment 4.70. Inequality (4.144) is sharper than (4.77), however the integral Im (λj ) (see (4.156)) in its right hand side, is difficult to compute and find its maximum value for m ≥ 5. That is why (4.77) is more practical, also less restrictive, and we used it extensively here in the applications.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Chapter 5
More on Multidimensional Ostrowski Type Inequalities
Very general multidimensional Ostrowski type inequalities are established, some of them prove to be sharp. They involve the k · k∞ and k · kp norms of the engaged mixed partial of nth order n ≥ 1. In establishing them, other important multivariate results of Montgomery type identity are developed and presented. This chapter relies on [23].
5.1
Introduction
In 1938, A. Ostrowski [196] proved the following inequality: Theorem 5.1. Let f : [a, b] → R be continuous on [a, b] and differentiable on (a, b) whose derivative f 0 : (a, b) → R is bounded on (a, b), i.e., kf 0 k∞ = sup |f 0 (t)| < t∈(a,b)
+∞. Then " 2 # 1 Z b x − a+b) 1 2 + f (t)dt − f (x) ≤ · (b − a)kf 0 k∞ , b − a a 4 (b − a)2
(5.1)
for all x ∈ [a, b]. The constant 14 is the best possible. Since then there has been a lot of activity around these inequalities with important applications to Numerical Analysis. This chapter is greatly motivated by the following results: k Q [ai , bi ] , where ai < bi ; ai , bi ∈ R, Theorem 5.2 (see [17]). Let f ∈ C 1 i=1
k Q − i = 1, . . . , k, and let → x 0 := (x01 , . . . , x0k ) ∈ [ai , bi ] be fixed. Then i=1 Z bk Z bi Z b1 1 − f (z1 , . . . , zk )dz1 · · · dzk − f (→ x 0 ) ··· ··· k Q ak ai (bi − ai ) a1 i=1
k X
(x0i − ai )2 + (bi − x0i )2
∂f . ≤
2(bi − ai ) ∂zi ∞ i=1
81
(5.2)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
82
Inequality (5.2) is sharp, here the optimal function is f ∗ (z1 , . . . , zk ) :=
k X i=1
|zi − x0i |αi ,
αi > 1.
Clearly inequality (5.2) generalizes inequality (5.1) to multidimension. We also would like to mention Theorem 5.3 (see [122]). Let f : [a, b] × [c, d] → R be a continuous mapping on ∂2f 00 exists on (a, b) × (c, d) and is in Lp ((a, b) × (c, d)), i.e., [a, b] × [c, d], fx,y := ∂x∂y !1/p Z bZ d 2 ∂ f (x, y) p 00 < +∞, p > 1. kfs,t kp := ∂x∂y dxdy a c
Then
"
− (b − a)
Z
d c
Z Z b d f (s, t)dsdt a c
f (x, t)dt + (d − c)
(x − a)q+1 + (b − x)q+1 ≤ q+1
Z
b
a
# f (s, y)ds − (d − c)(b − a)f (x, y)
1/q 1/q (y − c)q+1 + (d − y)q+1 00 · · kfs,t kp , q+1
(5.3)
for all (x, y) ∈ [a, b] × [c, d], where p1 + 1q = 1. In this chapter we develop more Ostrowski type inequalities in the multidimension, some of them are sharp. Along the way to establishing them we produce important side related results. Among others, one of the purposes of the chapter is to generalize Theorem 5.3, as far as possible, over ×ni=n [ai , bi ], n ∈ N, where we have n f (x1 ,...,xn ) . The main results are Ostrowski type inequalities involving involved ∂ ∂x 1 ···∂xn the norms k · k∞ and k · kp . Reference [62] provided us with important analytical tools. 5.2
Auxilliary Results
The following results are also by themselves of interest. We give Theorem 5.4. Let f : [a, A] × [b, B] × [c, C] → R be a continuous mapping on ∂3f 000 [a, A] × [b, B] × [c, C], and fx,y,z := ∂x∂y∂z exists on [a, A] × [b, B] × [c, C] and is integrable. Let also (x, y, z) ∈ [a, A] × [b, B] × [c, C] be fixed. We define the kernels p : [a, A]2 → R, q : [b, B]2 → R, and θ : [c, C]2 → R: s − a, s ∈ [a, x] p(x, s) := s − A, s ∈ (x, A],
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
More on Multidimensional Ostrowski Type Inequalities
q(y, t) :=
θ(z, r) :=
and
Then
Z
A
Z
B
Z
83
t − b, t ∈ [b, y], t − B, t ∈ (y, B] r − c, r ∈ [c, z], r − C, r ∈ (z, C].
C
000 p(x, s)q(y, t)θ(z, r)fs,t,r (s, t, r)dsdtdr Z = {(A − a)(B − b)(C − c) · f (x, y, z)} − (B − b)(C − c)
θ1,3 :=
a
b
c
+ (A − a)(C − c)
+ (C − c) + (A − a) −
Z
A a
Z
B b
Z
Z
A
a B
b
Z
C
Z
Z
Z
B
f (x, t, z)dt + (A − a)(B − b)
b B
f (s, t, z)dsdt + (B − b)
b C
f (x, t, r)dtdr c
Z
Z
A a
Z
A
f (s, y, z)ds a
C
f (x, y, r)dr c
C
f (s, y, r)dsdr c
f (s, t, r)dsdtdr =: θ2,3 .
(5.4)
c
Proof. Z Integrating by parts repeatedly we obtain the following eight equalities: xZ yZ z 000 (s − a)(t − b)(r − c)fs,t,r (s, t, r)drdtds a b c Z x f (s, y, z)ds = (x − a)(y − b)(z − c)f (x, y, z) − (y − b)(z − c) Z y Za z − (x − a)(z − c) f (x, t, z)dt − (x − a)(y − b) f (x, y, r)dr Z xZ z c Z xZ y b f (s, y, r)drds f (s, t, z)dtds + (y − b) + (z − c) Za y Zb z Z x Z y Za z c + (x − a) f (x, t, r)drdt − f (s, t, r)drdtds. (5.5) Similarly, we have Z AZ BZ C x
y
z
b
c
a
b
c
000 (s − A)(t − B)(r − C)fs,t,r (s, t, r)drdtds
Z
= (A − x)(B − y)(C − z)f (x, y, z) − (B − y)(C − z) − (A − x)(C − z) + (C − z) + (A − x)
Z
A
x
Z
B y
Z
Z
y
f (x, t, z)dt − (A − x)(B − y)
f (s, t, z)dtds + (B − y)
y
Z
C z
f (x, t, r)drdt −
Z
A x
Z
Z
B y
A x
Z
Z
C
f (s, y, z)ds
x
B
B
A
Z
C
f (x, y, r)dr z
f (s, y, r)drds z
C
f (s, t, r)drdtds. z
(5.6)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
84
Furthermore, Z
x a
Z
y b
Z
C z
000 (s − a)(t − b)(r − C)fs,t,r (s, t, r)drdtds
Z
= (x − a)(y − b)(C − z)f (x, y, z) − (y − b)(C − z) − (x − a)(C − z) + (C − z) + (x − a)
Z
x
a Z y
Z
Z
b
y
Z
f (x, t, z)dt − (x − a)(y − b)
f (s, t, z)dtds + (y − b)
b C
f (x, t, r)drdt −
z
Z
x a
Z
Z
y b
a
Z
xZ C
f (s, y, z)ds a
Z
y b
x
C
f (x, y, r)dr z
f (s, y, r)drds
z
C
f (s, t, r)drdtds.
(5.7)
z
Also, Z
x A
Z
y B
Z
z c
000 (s − A)(t − B)(r − c)fs,t,r (s, t, r)drdtds
= (A − x)(B − y)(z − c)f (x, y, z) − (B − y)(z − c) − (A − x)(z − c) + (z − c)
Z
+ (A − x)
A x
Z
Z
B y
Z
B
f (x, t, z)dt − (A − x)(B − y)
y
B
f (s, t, z)dtds + (B − y)
y
Z
z
f (x, t, r)drdt −
c
Z
Z
A x
Z
Z
B y
A x
Z
Z
z
A
f (s, y, z)ds x
Z
z
f (x, y, r)dr c
f (s, y, r)drds c
z
f (s, t, r)drdtds.
(5.8)
c
Next we find, Z
a
xZ B y
Z
C z
000 (s − a)(t − B)(r − C)fs,t,r (s, t, r)drdtds
= (x − a)(B − y)(C − z)f (x, y, z) − (B − y)(C − z) − (x − a)(C − z) + (C − z) + (x − a)
Z
xZ B
a Z B y
Z
Z
y
y
B
f (x, t, z)dt − (x − a)(B − y)
f (s, t, z)dtds + (B − y)
C
z
f (x, t, r)drdt −
Z
x a
Z
B y
Z
x a
Z
C
Z
C
Z
Z
x
f (s, y, z)ds
a C
f (x, y, r)dr z
f (s, y, r)drds z
f (s, t, r)drdtds. z
(5.9)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
More on Multidimensional Ostrowski Type Inequalities
85
Also, Z
A x
Z
y b
Z
z c
000 (s − A)(t − b)(r − c)fs,t,r (s, t, r)drdtds
Z A = (A − x)(y − b)(z − c)f (x, y, z) − (y − b)(z − c) f (s, y, z)ds x Z z Z y f (x, y, r)dr f (x, t, z)dt − (A − x)(y − b) − (A − x)(z − c) + (y − b)
Z
+ (A − x)
A
x
Z
y b
Z
b
Z
z
f (s, y, r)drds + (z − c)
c
Z
z
f (x, t, r)drdt −
c
Z
A x
Z
y b
A
x
Z
z
Z
c
y
f (s, t, z)dtds b
f (s, t, r)drdtds.
(5.10)
c
And, Z
x a
Z
B y
Z
z c
000 (s − a)(t − B)(r − c)fs,t,r (s, t, r)drdtds
= (x − a)(B − y)(z − c)f (x, y, z) − (B − y)(z − c) − (x − a)(z − c) + (z − c) + (x − a)
Z
a
Z
Z
xZ B y
B
y
Z
B
f (x, t, z)dt − (x − a)(B − y)
y
f (s, t, z)dtds + (B − y)
z
f (x, t, r)drdt −
c
Z
Z
x a
Z
B y
Z
x a
Z
z
Z
z
x
f (s, y, z)ds a
Z
z
f (x, y, r)dr c
f (s, y, r)drds c
f (s, t, r)drdtds.
(5.11)
c
Finally we have, Z
A x
Z
y b
Z
C z
000 (s − A)(t − b)(r − C)fs,t,r (s, t, r)drdtds
= (A − x)(y − b)(C − z)f (x, y, z) − (y − b)(C − z) − (A − x)(C − z) + (C − z) + (A − x)
Z
A x
Z
b
Z
y
b yZ C z
Z
y b
Z
f (x, t, z)dt − (A − x)(y − b)
f (s, t, z)dtds + (y − b) f (x, t, r)drdt −
Z
A x
Z
Z
y b
AZ C
x
Z
A
f (s, y, z)ds x
Z
C
f (x, y, r)dr z
f (s, y, r)drds
z
C
f (s, t, r)drdtds. z
(5.12)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
86
Adding all the right-hand sides of (5.5) – (5.12), we derive:
Big R.H.S. = (A − a)(B − b)(C − c)f (x, y, z) − (B − b)(C − c) − (A − a)(C − c) + (C − c) + (A − a)
Z
A a
Z
B b
Z
Z
B
f (x, t, z)dt − (A − a)(B − b)
b
B
f (s, t, z)dtds + (B − b)
b
Z
C c
Z
f (x, t, r)dtdr −
Z
A a
Z
Z
B b
A
a
Z
Z
C
A
f (s, y, z)ds a
Z
C
f (x, y, r)dr c
f (s, y, r)dsdr c
C
f (s, t, r)drdtds. c
(5.13)
Adding all the left-hand sides of (5.5) – (5.12), we find
Big L.H.S. =
Z
A a
Z
B b
Z
C c
000 p(x, s)q(y, t)θ(z, r)fs,t,r (s, t, r)drdtds.
(5.14)
Clearly we have Big L.H.S. = Big R.H.S. We have established (5.4).
(5.15)
In general we state Theorem 5.5. Let f : ×ni=1 [ai , bi ] → R be a continuous mapping on ×ni=1 [ai , bi ], n f (x1 ,...,xn ) and ∂ ∂x exists on ×ni=1 [ai , bi ] and is integrable. Let also (x1 , . . . , xn ) ∈ 1 ···∂xn ×ni=1 [ai , bi ] be fixed. We define the kernels pi : [ai , bi ]2 → R: pi (xi , si ) :=
si − ai , si ∈ [ai , xi ] si − bi , si ∈ (xi , bi ],
for all i = 1, . . . , n. Then θ1,n := =
Z
n Y
×n i=1 [ai ,bi ] i=1
(
n Y
i=1
pi (xi , si )
(bi − ai )
!
∂ n f (s1 , . . . , sn ) ds1 · · · dsn ∂s1 · · · ∂sn )
· f (x1 , . . . , xn )
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
More on Multidimensional Ostrowski Type Inequalities
n (X Z 1) n Y − (bj − aj )
i=1
j=1 j6=i
bi ai
k=1 k6=i,j
f (x1 , . . . , si , . . . , xn )dsi
(n2 ) Z n X Y + (bk − ak ) `=1
87
bi
ai
Z
bj
f (x1 , . . . , si , . . . , sj , . . . , xn )dsi dsj aj
!
(`)
− + · · · − + · · · + (−1)n−1 n (X Z n−1) cj · · · dsn (bj − aj ) f (s1 , . . . , xj , . . . , sn )ds1 · · · ds · + (−1)
Z
×n [a ,b ] i=1 i i i6=j
j=1
n
×n i=1 [ai ,bi ]
f (s1 , . . . , sn )ds1 · · · dsn =: θ2,n .
(5.16)
cj means dsj The above ` counts all the (i, j)’s, i < j and i, j = 1, . . . , n. Also ds is missing. Proof.
5.3
Similar to Theorem 5.4.
Main Results
Here we present the consequences of Section 5.2. We have Theorem 5.6. Under the notations and assumptions of Theorem 5.4, additionally 000 we suppose that kfs,t,r k∞ < +∞. Then 000 kfs,t,r k∞ · ((x − a)2 + (A − x)2 ) · ((y − b)2 + (B − y)2 ) 8 ·((z − c)2 + (C − z)2 ) , for all (x, y, z) ∈ [a, A] × [b, B] × [c, C].
|θ2,3 | ≤ Proof.
Notice that
000 |θ2,3 | = |θ1,3 | ≤ kfs,t,r k∞
Also observe that
etc.
Z
Z
A a
|p(x, s)|ds
A a
|p(x, s)|ds =
! Z
B b
|q(y, t)|dt
! Z
(5.17)
C c
|θ(z, r)|dr
!
.
1 {(x − a)2 + (A − x)2 }, 2
The counterpart of the last theorem is Theorem 5.7. Under the notations and assumptions of Theorem 5.5, additionally we suppose that
n
∂ f (s1 , . . . , sn )
∂s1 · · · ∂sn < +∞. ∞
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
88
Then
∂ n f (s1 ,...,sn )
∂s1 ···∂sn 2n
|θ2,n | ≤ Proof.
∞
·
(
n Y
i=1
2
2
)
[(xi − ai ) + (bi − xi ) ] ,
for all (x1 , . . . , xn ) ∈ ×ni=1 [ai , bi ].
(5.18)
As in Theorem 5.6.
The Lp -analogues follow. Theorem 5.8. Under the notations and assumptions of Theorem 5.4, additionally, 000 we assume that fs,t,r ∈ Lp ([a, A] × [b, B] × [c, C]), i.e., !1/p Z AZ BZ C 3 ∂ f (x, y, z) p 000 kfs,t,r kp := < +∞, ∂x∂y∂z dxdydz a b c
where p > 1. Then 000 kfs,t,r kp · [((x − a)q+1 + (A − x)q+1 ) · ((y − b)q+1 + (B − y)q+1 ) |θ2,3 | ≤ (q + 1)3/q · ((z − c)q+1 + (C − z)q+1 )]1/q , for all (x, y, z) ∈ [a, A] × [b, B] × [c, C],
(5.19)
1 p
where q is such that Proof.
1 q
+
= 1.
Notice that
Z Z Z A B C 000 |θ2,3 | = |θ1,3 | = p(x, s)q(y, t)θ(z, r)fs,t,r (s, t, r)dsdtdr a b c Z AZ BZ C 000 ≤ |p(x, s)| |q(y, t)| |θ(z, r)| |fs,t,r (s, t, r)|dsdtdr a
a
c
(by H¨ older’s inequality)
Z
≤ ·
A a
Z
Z
A a
B b
Z
B b
000 = kfs,t,r kp ·
·
Z
Z
C c
Z
000 |fs,t,r (s, t, r)|p dsdtdr
C
q
(|p(x, s)| |q(y, t)| |θ(z, r)|) dsdtdr
c
Z
A a
|p(x, s)|q ds
C q
c
!1/p
|θ(z, r)| dr
!
!
·
Z
B b
!1/q
|q(y, t)|q dt
!
1/q
(x − a)q+1 + (A − x)q+1 = · q+1 1/q q+1 (y − b) + (B − y)q+1 (z − c)q+1 + (C − z)q+1 . · · q+1 q+1 000 kfs,t,r kp
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
More on Multidimensional Ostrowski Type Inequalities
89
The corresponding general Lp -case follows. Theorem 5.9. Under the notations and assumptions of Theorem 5.5, additionally ∂nf n we assume that ∂x1 ···∂xn ∈ Lp ×i=1 [ai , bi ] , i.e.
n
∂ f (x1 , . . . , xn )
where p > 1.
∂x1 · · · ∂xn < +∞, p Then
|θ2,n | ≤
∂ n f (x1 ,...,xn )
∂x1 ···∂xn (q + 1)n/q
p
·
(
n Y
i=1
[(xi − ai )
for any (x1 , . . . , xn ) ∈ ×ni=1 [ai , bi ], and q : Proof.
1 p
+
q+1
1 q
+ (bi − xi )
q+1
]
)1/q
,
(5.20)
= 1.
Same as in Theorem 5.8.
Remark 5.10. Equalities (5.4) and (5.16) can simplify dramatically, if for instance we suppose in Theorem 5.4 that there exists an (x0 , y0 , z0 ) ∈ [a, A] × [b, B] × [c, C] such that f (x0 , ·, ·) = f (·, y0 , ·) = f (·, ·, z0 ) = 0.
Also in Theorem 5.5 we may assume that there exists an (x01 , x02 , . . . , x0n ) ∈ ×ni=1 [ai , bi ] such that f (x01 , x2 , . . . , xn ) = f (x1 , x02 , x3 , . . . , xn ) = · · · = f (x1 , . . . , xn−1 , x0n ) = 0,
for any (x1 , . . . , xn ) ∈ ×ni=1 [ai , bi ]. So in these particular cases we derive that Z AZ BZ C θ2,3 (x0 , y0 , z0 ) = − f (s, t, r)dsdtdr, (5.21) a
and
θ2,n (x01 , x02 , . . . , x0n )
= (−1)
n
Z
b
×n i=1 [ai ,bi ]
c
f (s1 , . . . , sn )ds1 · · · dsn .
(5.22)
Hence in these cases we have
and
Z Z Z A B C |θ2,3 (x0 , y0 , z0 )| = f (s, t, r)dsdtdr , a b c
(5.23)
Z f (s1 , . . . , sn )ds1 · · · dsn . |θ2,n (x01 , x02 , . . . , x0n )| = ×ni=1 [ai ,bi ]
So according to Theorems 5.6 – 5.9 we get respectively: Z Z Z kf 000 k A B C s,t,r ∞ f (s, t, r)dsdtdr ≤ a b 8 c · ((x0 − a)2 + (A − x0 )2 ) · ((y0 − b)2 + (B − y0 )2 ) · ((z0 − c)2 + (C − z0 )2 ) ;
(5.24)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
90
and
Z ∂ n f (s ,...,s ) 1 n ∂s1 ···∂sn ∞ f (s1 , . . . , sn )ds1 · · · dsn ≤ ×ni=1 [ai ,bi ] 2n (n ) Y 0 2 0 2 · [(xi − ai ) + (bi − xi ) ] .
(5.25)
i=1
Also it holds Z Z Z A B C f (s, t, r)dsdtdr a b c ≤
and
000 kfs,t,r kp · [((x0 − a)q+1 + (A − x0 )q+1 ) · ((y0 − b)q+1 + (B − y0 )q+1 ) (q + 1)3/q · ((z0 − c)q+1 + (C − z0 )q+1 )]1/q ; (5.26)
∂ n f (x1 ,...,xn ) Z
∂x1 ···∂xn p f (s1 , . . . , sn )ds1 · · · dsn ≤ ×ni=1 [ai ,bi ] (q + 1)n/q )1/q (n Y 0 q+1 0 q+1 [(xi − ai ) + (bi − xi ) ] · .
(5.27)
i=1
Finally we present the optimality of inequalities in Theorems 5.6 and 5.7. Theorem 5.11. Inequalities (5.17) and (5.18) are sharp. Proof. be
It is enough to prove that (5.18) is sharp. Here the optimal function will
f ∗ (s1 , . . . , sn ) :=
n Y
i=1
|si − x0i |α (bi − αi ),
α > 1,
(5.28)
where (x01 , x02 , . . . , x0n ) is fixed in ×ni=1 [ai , bi ]. Notice here that f ∗ (s1 , . . . , x0j , . . . , sn ) = 0,
for all j = 1, . . . , n, and any (s1 , . . . , sn ) ∈ ×ni=1 [ai , bi ].
Therefore by Remark 5.10, inequality (5.18) collapses to inequality (5.25). We see that ! ! n n Y Y ∂ n f ∗ (s1 , . . . , sn ) = αn · (bi − ai ) · |si − x0i |α−1 sign(si − x0i ) , (5.29) ∂s1 · · · ∂sn i=1 i=1 and n ∗ ∂ f (s1 , . . . , sn ) n ∂s1 · · · ∂sn = α ·
n Y
i=1
(bi − ai )
!
·
n Y
i=1
|si −
x0i |α−1
!
.
(5.30)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
More on Multidimensional Ostrowski Type Inequalities
Consequently we find that
n ∗
∂ f (s1 , . . . , sn )
∂s1 · · · ∂sn
=α
n Y
n
∞
i=1 n Y
·
i=1
(bi − ai )
(max(bi −
91
! x0i , x0i
− ai ))
α−1
!
.
(5.31)
First we calculate the left-hand side of corresponding inequality (5.25). We have Z f ∗ (s1 , . . . , sn )ds1 · · · dsn ×ni=1 [ai ,bi ] ! ! Z n n Y Y 0 α ds1 · · · dsn (bi − ai ) · |si − xi | = ×n i=1 [ai ,bi ]
= = =
n Y
i=1 n Y
i=1 n Y i=1
That is,
L.H.S.(5.25) =
i=1
(bi − ai ) (bi − ai ) (bi − ai ) n Q
i=1
!Z !
!
R.H.S.(5.25) =
·
(
n Q
i=1
n Y
Z n Y i=1
·
n Y
lim R.H.S.(5.25) =
α→1
n Q
i=1
|si −
i=1
x0i |α dsi
ds1 · · · dsn
!!
− ai )α+1 + (bi − x0i )α+1 α+1
((x0i
− ai )
α+1
+ (bi −
!
.
!
.
x0i )α+1 )
(5.32)
i=1
2n
2
− ai ) + (bi −
Now let α → 1. We find that n Q (bi − ai ) lim L.H.S.(5.25) = i=1 n · α→1 2 and
bi
|si − x0i |α
!
n Q (bi − ai ) · (max(bi − x0i , x0i − ai ))α−1
((x0i
i=1
i=1
ai i=1 n Y (x0i
·
(α + 1)n αn ·
×n i=1 [ai ,bi ]
·
(bi − ai )
And next we observe that
i=1
n Y
(bi − ai ) 2n
That is,
·
n Y
x0i )2 )
((x0i
i=1
n Y
i=1
)
.
(5.33)
2
− ai ) + (bi −
hence proving the sharpness of (5.18).
α→1
!
,
!
((x0i − ai )2 + (bi − x0i )2 ) .
lim L.H.S.(5.25) = lim R.H.S.(5.25),
α→1
x0i )2 )
(5.34)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
92
Remark 5.12. Another interesting case for (5.4) and (5.16) is to suppose that for specific (x, y, z) ((x1 , . . . , xn ), respectively) all the marginal integrals of f are equal to zero. Then we find Z AZ BZ C f (s, t, r)dsdtdr, (5.35) θ2,3 = (A − a)(B − b)(C − c) · f (x, y, z) − a
and
θ2,n =
n Y
i=1
(bi − ai )
!
· f (x1 , . . . , xn ) + (−1)
n
Z
a
c
×n i=1 [ai ,bi ]
f (s1 , . . . , sn )ds1 · · · dsn .
(5.36) Hence inequalities (5.17), (5.18), (5.19), and (5.20) become again much simpler.
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Chapter 6
Ostrowski Inequalities on Euclidean Domains
The classical Ostrowski inequality for functions on intervals is extended to functions on general domains in Euclidean space. For radial functions on balls the inequality is sharp. This treatment relies on [56].
6.1
Introduction
The classical Ostrowski inequality (of 1938) [196] is 2 ! Z b 1 x − a+b 1 2 (b − a)kf 0 k∞ , + f (y)dy − f (x) ≤ b − a 4 (b − a)2 a
for f ∈ C 1 ([a, b]), x ∈ [a, b], and it is a sharp inequality. This was extended from intervals to rectangles in RN , N ≥ 1, see [21], p. 507. For other recent results related to Ostrowski’s inequality, see [110], [126] and [199], [214], [231]. The extension to general domains in RN is presented here. We deduce Ostrowski type inequalities on general bounded domains in RN , and the inequalities are shown to be sharp on balls.
6.2
Main Results
Let N > 1, B(0, R) := {x ∈ RN : |x| < R} be the ball in RN centered at the origin and of radius R > 0. Let S N −1 := {x ∈ RN : |x| = 1} be the unit sphereR in RN . Let dω be the element of surface measure on S N −1 and let ωN = S N −1 dω = N/2 2π . For x ∈ RN −{0} we can write x = rω, where r = |x| > 0 and ω = xr ∈ S N −1 . Γ
N 2
R
N
dy = ωNNR is the Lebesgue measure of the ball. For f ∈ C B(0, R) let Z Z 1 − f (y)dy := f (y)dy, Vol(B(0, R)) B(0,R)
Note that
B(0,R)
B(0,R)
93
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
94
and Z Z 1 − f (rω)dω = f (rω)dω ωN S N −1
S N −1
be the averages of f over the ball and the sphere, respectively. Here f can be real or complex valued. Let Z ˜ := − f (rω)dω f(r) S N −1
be the average of f (x) as x ranges over {y ∈ RN : |y| = r}. Then N (f ) :=
sup x∈B(0,R)
˜∞ |f (x) − f˜(r)| = kf − fk
(6.1)
measures howfar f is from being a radial function. More precisely, N is a seminorm on C B(0, R) , and N (f ) = 0 if and only if f is a radial function, i.e. f (x) = g(r) for some function g ∈ C([0, R]). We view how close f is to being radial by measuring N (f ); the closer f is to being radial, the smaller N (f ) is and conversely. Let Ω be a domain in RN and let Lip(Ω) = f ∈ C(Ω) : |f (x) − f (y)| ≤ K|x − y| for some K > 0 and all x, y ∈ Ω .
(6.2)
The Lipschitz constant of f ∈ Lip(Ω) is
kf kLip = inf{K : K as in (6.2)}. Then X := Lip(Ω) is a Banach space with norm f → kf k∞ + kf kLip =: kf kX . Equivalently, X is the Sobolev space W 1,∞ (Ω) (cf. [139]). We present Theorem 6.1. Let f ∈ Lip(B(0, R)) = W 1,∞ (B(0, R)). Then for x = rω as above Z N +1 |x| R N f (x) − − f (y)dy ≤ N (f ) + N k∇f k∞ 2|x| + R − . RN N (N + 1) N +1 N B(0,R)
(6.3) The constants in (6.3) are best possible, equality can be attained for nontrivial radial functions at any r ∈ [0, R].
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Euclidean Domains
95
Proof. Let f ∈ Lip(B(0, R)). Then Z f (x) − − f (y)dy B(0,R) Z Z Z R N 0 N −1 0 ˜ ≤ |f (x) − f(r)| + − f (rω 0 )dω 0 − f (sω )s dsdω N ωN R S N −1 0 N −1 S # Z "Z R N |f (rω 0 ) − f (sω 0 )|S N −1 ds dω 0 ≤ N (f ) + N − R 0 S N −1
N ≤ N (f ) + N R
Z Z −
S N −1
0
∂f (ω 0 )
∂r
R
L∞ ((0,R))
|s − r|sN −1 dsdω 0
!
Z R
∂f N N −1
≤ N (f ) + |s − r|s ds
∂r ∞ N 0 L (B(0,R)) R
∂f 2rN +1 R r N N
. + R − = N (f ) +
∂r ∞ N N (N + 1) N +1 N L (B(0,R)) R
Then (6.3) follows since
(6.4)
∂f
≤ k∇f k∞ .
∂r ∞
In particular, a stronger form of (6.3) actually holds in all cases, replacing k∇f k ∞
. Let r ∈ [0, R] and g ∗ (z) = |z − r|. We can view g ∗ as a radial function by ∂f ∂r ∞ on B(0, R). Then sign(z − r), z 6= r, 0 < r < R g ∗0 (z) = 1, r=0 −1 r = R.
Thus kg ∗0 k∞ = 1. Therefore Z R Z R N N ∗ ∗ N −1 N −1 L.H.S.(6.3) = g (z) − N g (s)s ds = |z − r| − N |s − r|s ds R R 0 0 N r 2rN +1 R . = |z − r| − N + RN − R N (N + 1) N +1 N Also N r 2rN +1 R R.H.S.(6.3) = N + RN − . R N (N + 1) N +1 N Hence equality holds in (6.3) at z = r. Note that the function g ∗ (z) = |z − r|, is in C 1 ([0, R]) only for r = 0 and r = R; for 0 < r < R, g ∗ ∈ Lip([0, R]) − C 1 ([0, R]). Of course for 0 < r < R, g ∗ can be 1 approximated by C 1 functions, namely gn (z) = |z − r|1+ n .
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
96
Remark 6.2. A key step in the proof is the fact that we can evaluate exactly Z R Q(r) = |s − r|p(s)ds 0
RR for 0 ≤ r ≤ R, where p is a nonnegative continuous function satisfying 0 p(s)ds = 1 1. In the Ostrowski case (N = 1), p(s) = R ; in our N -dimensional case, p(s) = N −1 Ns RN . This works for many other cases including linear combinations of p(s) = m P aj sqj , where aj > 0, qj ≥ 0 (not necessarily an integer) and j=1
m X
aj
j=1
p(s) =
m P1
j=1
Rqj +1 = 1, qj + 1
aj eλj s , where aj > 0, λj ∈ R − {0}, m1 X j=1
aj
e λj R − 1 λj
= 1,
p(s) = sums of the form aj sin(bj s + cj ) + dj cos(ej s + fj ), RR where the coefficients are such that p(s) ≥ 0 and 0 p(s)ds = 1. The space Lip(Ω) ∩ C0 (Ω) consists of all Lipschitz continuous functions on Ω vanishing on the boundary ∂Ω of Ω. Note that Lip(Ω) ∩ C0 (Ω) = f ∈ W 1,∞ (Ω) ∩ C(Ω) : f = 0 on ∂Ω (cf. [139]). Next comes a more general result when we consider functions over general domains.
Theorem 6.3. Let f ∈ Lip(Ω) ∩ C0 (Ω), where Ω is a bounded domain in RN . Extend f by zero to F on B(0, R), the smallest ball centered at the origin and containing Ω. Then for all x ∈ Ω, Z Z Vol(Ω) − f (y)dy f (x) − − f (y)dy ≤ N (F ) + 1 − Vol(B(0, R)) Ω Ω N 2|x|N +1 |x| R + N k∇f kL∞(Ω) . (6.5) + RN − R N (N + 1) N +1 N Proof.
Let R := inf{R0 > 0 : Ω ⊂ B(0, R0 )}. Then f (x), x ∈ Ω, F (x) := 0, x ∈ B(0, R) − Ω,
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Euclidean Domains
97
satisfies F ∈ Lip(B(0, R)) ∩ C0 (B(0, R)). Then for x ∈ Ω Z Z Z Z f (x) − − f (y)dy ≤ F (x) − − F (y)dy + − F (y)dy − − f (y)dy Ω
B(0,R)
B(0,R)
Ω
= J1 + J2 ,
where
Z J1 := F (x) − − F (y)dy B(0,R)
and
By Theorem 6.1,
Z 1 1 J2 := − f (y)dy . Vol(B(0, R)) Vol(Ω) Ω
J1 ≤ N (F ) +
2|x|N +1 R |x| N N . k∇f k + R − ∞ RN N (N + 1) N +1 N
and
Vol(Ω) J2 = 1 − Vol(B(0, R)) This completes the proof of Theorem 6.3.
Z − f (y)dy . Ω
Remark 6.4. Note that N (F ) appears in (6.5). In this context, N (f ) does not make sense. Also, N (F ) need not be small (of course, it is small if f is approximately spherically symmetric). Here is a simple example to illustrate that N (F ) can be large. Let x0 ∈ Ω and choose ε > 0, small enough so that B(x0 , ε) ⊂ Ω. Let f ∈ C ∞ (Ω) with f have support in B(x0 , ε) and satisfy f (y) = g(ρ) where ρ = |y − x0 |, for 0 ≤ ρ ≤ ε. Assume further that g is nonincreasing and g(0) = f (x0 ) = M > 0. Fix M , then Z 0 < f (x0 ) − − f (y)dy → M = kf k∞ as ε → 0 + . Ω
This page intentionally left blank
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Chapter 7
High Order Ostrowski Inequalities on Euclidean Domains
The original Ostrowski the function minus its This result is extended nth partial derivatives. chapter relies on [57].
7.1
inequality for functions on intervals estimates the value of average in terms of the maximum of its first derivative. to functions on general domains using the L∞ norm of its For radial functions on balls the inequality is sharp. This
Introduction
The classical Ostrowski inequality (of 1938) [196] is 1 Z b f (y)dy − f (x) ≤ b − a a
2 ! x − a+b 1 2 (b − a)kf 0 k∞ , + 4 (b − a)2
(7.1)
for f ∈ C 1 ([a, b]), x ∈ [a, b], and it is a sharp inequality. Further related work was done in [16]. Inequality (7.1) was extended from intervals to rectangles in R N , N ≥ 1, see [21], p. 507. The extension to general domains in RN was done [56]. The resulting inequality was sharp, and equality was shown to hold for certain radial functions on balls. The right hand side included the factor k∇f k∞ . The extension of (7.1) to high order derivative bounds on an interval was obtained in [21], p. 502. The purpose here is to extend the results of [56] and [21], p. 502, to the higher order case when k∇f k∞ is replaced by max kDα f k∞ for W n,∞ functions on balls |α|=n
and more general domains. The obtained inequalities are sharp on balls.
7.2
Main Results
Let N > 1, B(0, R) := {x ∈ RN : |x| < R} be the ball in RN centered at the origin and of radius R > 0. Let S N −1 := {x ∈ RN : |x| = 1} be the unit sphere in RN . 99
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
100
Let dω be the element of surface measure on S N −1 and let Z 2π N/2 ωN = . dω = Γ(N/2) S N −1
For x ∈ RN − {0} we can write x = rω, where r = |x| > 0 and ω = R N Note that B(0,R) dy = ωNNR is the Lebesgue measure of the ball. For f ∈ C B(0, R) let Z Z 1 f (y)dy, − f (y)dy := Vol(B(0, R)) B(0,R)
x r
∈ S N −1 .
B(0,R)
and
Z Z 1 − f (rω)dω = f (rω)dω ωN S N −1
S N −1
be the averages of f over the ball and the sphere, respectively. Here f can be real or complex valued. Let Z ˜ := − f (rω)dω f(r) S N −1
be the average of f (x) as x ranges over {y ∈ RN : |y| = r}. Therefore N (f ) :=
sup x∈B(0,R)
˜∞ |f (x) − f˜(r)| = kf − fk
(7.2)
measures howfar f is from being a radial function. More precisely, N is a seminorm on C B(0, R) , and N (f ) = 0 if and only if f is a radial function, i.e. f (x) = g(r) for some function g ∈ C([0, R]). We view how close f is to being radial by measuring N (f ); the closer f is to being radial, the smaller N (f ) is and conversely. Let Ω be a domain in RN and let Lip(Ω) = f ∈ C(Ω) : |f (x) − f (y)| ≤ K|x − y| for some K > 0 and all x, y ∈ Ω . (7.3)
The Lipschitz constant of f ∈ Lip(Ω) is
kf kLip = inf{K : K as in (7.3)}.
(7.4)
n,∞
Let n ∈ Z+ = {0, 1, 2, . . .}. Define the Sobolev space W (Ω) in the usual way: W n,∞ (Ω) := u : Ω → C : the distributional derivative D α u exists and is in L∞ (Ω) for 0 ≤ |α| ≤ n = u ∈ C n−1 (Ω) : Dα u ∈ Lip(Ω) for |α| = n . (7.5)
This is a Banach space with norm
kuk := max kDα ukL∞ (Ω) , 0≤|α|≤n
(7.6)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
High Order Ostrowski Inequalities on Euclidean Domains
Book˙Adv˙Ineq
101
(see [139]). Clearly W n,∞ (Ω) ⊃ C n (Ω). The first main result is the following. Theorem 7.1. Let f ∈ W n,∞ (B(0, R)). Then for x = rω as above Z f (x) − − f (y)dy B(0,R) ( n−1 Z N X 1 ∂ k f (rω 0 ) 0 ≤ N (f ) + N dω − R k! ∂rk k=1
S N −1
N +k−m ! k X k R × (−1)m rm m N + k − m m=0
∂nf n
n ∞ X RN +n−m n ∂r L (B(0,R)) + (−1)m rm n! N +n−m m m=0 !) n X rN +n n − βn , (−1)m N +n−m m m=0
(7.7)
where n ∈ N, βn = 0 or 2, according as n is even or odd. The constants in (7.7) are best possible, equality can be attained for nontrivial radial functions at any r ∈ [0, R]. We remark that
n
∂ f n
max kDα f k∞ ,
∂rn ≤ kD f k∞ := |α|=n ∞
n n and the larger bound kD f k∞ replacing ∂∂rnf ∞ gives perhaps a simpler version of the inequality (7.7).
Proof. Let f ∈ W n,∞ (B(0, R)). Hence Z Z f (x) − − f (y)dy ≤ |f (x) − f˜(r)| + − f (rω 0 )dω 0 B(0,R) S N −1 Z Z R N 0 N −1 0 − f (sω )s dsdω (7.8) ωN RN S N −1 0 Z Z R N 0 0 N −1 ≤ N (f ) + N − [f (rω ) − f (sω )]s ds dω 0 = N (f ) + (∗). R 0 S N −1
(7.9)
We see by Taylor’s formula that f (sω 0 ) − f (ρω 0 ) =
n−1 X k=1
∂ k f (ρω 0 ) (s − ρ)k + Rn−1 (ρ, s), k!∂rk
(7.10)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
102
where
Z s
∂ n−1 f (tω 0 ) ∂ n−1 f (ρω 0 ) (s − t)n−2 Rn−1 (ρ, s) := − dt, (7.11) ∂ρn−1 ∂ρn−1 (n − 2)! ρ for fixed ω 0 ∈ S N −1 and all s, ρ ∈ (0, R]. As in p. 500 of [21] we derive n
∂ nf |Rn−1 (ρ, s)| ≤ ∂r ∞ |s − ρ|n , ∀s, ρ ∈ [0, R]. (7.12) n! Consequently it holds " # Z R n−1 Z X ∂ k f (rω 0 ) N k N −1 0 (s − r) + R (r, s) s ds dω (∗) = N − (7.13) n−1 R k!∂rk 0 k=1 N −1 S ( n−1 Z Z R k N X ∂ f (rω 0 ) k N −1 ≤ N − (s − r) s ds dω 0 k R k!∂r 0 k=1 N −1 S ) Z Z R
+
−
S N −1
0
|Rn−1 (r, s)|sN −1 dsdω 0
( n−1 Z Z R ∂ k f (rω 0 ) 0 N X 1 dω − (s − r)k sN −1 ds ≤ N k R k! ∂r 0 k=1 S N −1
∂ nf ) Z R
n ∞ ∂r L (B(0,R)) n N −1 |s − r| s ds =: A. + n! 0
(7.14)
(7.15)
To evaluate the integrals in (7.15) we use an elementary calculation as follows. Lemma 7.2. (i) Z R k X k RN +k−m , k ∈ N. sN −1 (s − r)k ds = (−1)m rm N +k−m m 0 m=0 (ii) Z R n X n RN +n−m N −1 n s |s − r| ds = (−1)m rm N +n−m m 0 m=0 n X n rN +n (−1)m − βn , m N +n−m m=0 where n ∈ N, βn = 0 or 2, according as n is even or odd. Proof.
(i) We see that Z R Z sN −1 (s − r)k ds = 0
! k X k m k−m m (−1) s r ds s m 0 m=0 Z R k X k m m = (−1) r sN +k−m−1 ds m 0 m=0 k X k RN +k−m . (−1)m rm = N +k−m m m=0 R
(7.16)
(7.17)
N −1
(7.18)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
High Order Ostrowski Inequalities on Euclidean Domains
Book˙Adv˙Ineq
103
(ii) Using (i) twice, Z
R
s 0
N −1
n
|s − r| ds =
Z
r
s
N −1
0
n
(r − s) ds +
Z
R r
sN −1 (s − r)n ds
n X rN +n n (−1)m = (−1) m N +n−m m=0 n X n RN +n−m − rN +n−m + (−1)m rm N +n−m m m=0 n X n RN +n−m = (−1)m rm N +n−m m m=0 n X rN +n n + . (−1)m [(−1)n − 1] N +n−m m m=0 n
(7.19)
Part (ii) now follows. Continuing the calculation in (7.15) N A= N R
( n−1 Z X 1 ∂ k f (rω 0 ) 0 dω − k! ∂rk k=1
S N −1
N +k−m ! k R m m × (−1) r N + k − m m m=0
∂ nf n
n ∞ X n RN +n−m ∂r L (B(0,R)) (−1)m rm + m n! N +n−m m=0 !) n X rN +n n (−1)m − βn , m N +n−m m=0 k X
(7.20)
by Lemma 7.2. Hence proving (7.7). Let n be even. Let f be a radial function so that f (x) = g(r). Then equality in (7.7) is attained by g(r) = (r − z)n for any fixed z satisfying 0 ≤ z ≤ R. We observe that g (j) (z) = 0, j = 0, 1, . . . , n − 1 and g (n) (z) = n!, kg (n) k∞ = n!. We look at inequality (7.7) evaluating the function at z: L.H.S.(7.7) =
N RN
Z
R 0
(s − z)n sN −1 ds
n N +n−m N X n m m R = (−1) z RN m=0 m N +n−m ! n X n Rn−m m m . = N (−1) z N +n−m m m=0
(7.16)
(7.21) (7.22)
17th August 2010
15:56
104
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
Because N (f ) = 0 and βn = 0 the right hand side of (7.7) collapses to ! n N +n−m X N n m m R (−1) z R.H.S.(7.7) = N R m N +n−m m=0 ! n X Rn−m n m m (−1) z . (7.23) =N m N +n−m m=0 That is equality holds in (7.7). Note that g ∈ C n B(0, R) if z > 0, but g ∈ W n,∞ (B(0, R)) − C n B(0, R) if z = 0. The optimal function in the case of n odd is g(r) := |z − r|n , 0 ≤ r ≤ R, z is fixed in [0, R].
(7.24)
Direct calculations show, for n odd, g (k) (r) = n(n − 1) · · · (n − k + 1)|r − z|n−k αk (r),
(7.25)
where αk (r) = 1 αk (r) = sign(r − z)
if k is even, if k is odd and r 6= z.
(7.26)
It follows that g ∈ C ∞ ([0, R] − {z}) ∩ C n−1 ([0, R]) and g (n) ∈ L∞ ((0, R)) with a jump discontinuity at r = z, whence g ∈ W n,∞ ((0, R)). It follows that f ∈ W n,∞ (B(0, R)). Moreover, g (k) (z) = 0, for k = 0, 1, . . . , n − 1 and |g (k) (z + ε)| → n! as ε → 0 (with ε 6= 0). Thus kg (n) k∞ = n!. The left-hand side of (7.7) is Z R N |z − s|n sN −1 ds L.H.S.(7.7) = RN 0 n N +n−m X n (7.17) N m m R = (−1) z RN m=0 m N +n−m ! n X n z N +n m −2 (−1) . (7.27) m N +n−m m=0
Since N (f ) = 0 and n is odd, using the calculation on g given above leads to the conclusion that ( n RN +n−m N n! X n (−1)m z m R.H.S.(7.7) = N R n! m=0 m N +n−m !) n N +n X n z −2 (−1)m . (7.28) m N +n−m m=0
Hence equality holds in (7.7). This completes the proof of the sharpness of inequality (7.7). The proof of the theorem is now complete.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
High Order Ostrowski Inequalities on Euclidean Domains
Book˙Adv˙Ineq
105
Note that the extremalfunctions which give equality in (7.7) in the case of n odd are not in C n B(0, R) . To find a sequence {fν }ν∈N of approximate extremal functions in C n B(0, R) , simply take fν (x) = gν (r), where 1
gν (r) = |z − r|n+ ν , ν ∈ N.
(7.29)
Next we treat the case of r = R. k
Theorem 7.3. Let f ∈ W n,∞ (B(0, R)) such that ∂∂rfk , k = 1, . . . , n − 1 vanish on ∂B(0, R). Then for all w ∈ S N −1 ,
n Z Rn ∂∂rnf L∞ (B(0,R)) . (7.30) f (Rω) − − f (y)dy ≤ N (f ) + (N + 1) · · · (N + n) B(0,R)
The constants in (7.30) are best possible, equality can be attained for nontrivial radial functions. Proof.
Here from (7.7) we obtain Z f (Rω) − − f (y)dy ≤ N (f ) B(0,R)
n n m n X N Rn n m (−1)
∂ f (−1) + = (∗),
n n! ∂r L∞ (B(0,R)) N +n−m m=0
since 1 − βn = (−1)n . Next, n X n (−1)m tN +n−m−1 = tN −1 (t − 1)n , m m=0
by the Binomial Theorem. Integrate (7.32) over t ∈ [0, 1], the result is Z 1 n X n (−1)m (1 − t)n tN −1 dt = (−1)n m N + n − m 0 m=0 = (−1)n B(n + 1, N ) n!(N − 1)! = (−1)n . (N + n)!
By (7.34) we derive
n
n Rn ∂∂rnf L∞ (B(0,R))
(N − 1)! ∂ f
(∗) = N Rn = ,
∂rn ∞ (N + 1) · · · (N + n) L (B(0,R)) (N + n)!
(7.31)
(7.32)
(7.33)
(7.34)
(7.35)
that is proving (7.30). The function f (x) = g(r) = (R − r)n gives equality in (7.30), whether n is even or odd.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
106
7.3
Book˙Adv˙Ineq
Functions on General Domains
The next result gives high order Ostrowski type inequalities functions on general bounded domains in RN . Theorem 7.4. Let f ∈ W n,∞ (Ω) ∩ C0n−1 (Ω), where Ω is a bounded domain in RN and C0n−1 (Ω) = {u ∈ C n−1 (Ω) : Dα u = 0 on ∂Ω for |α| ≤ n − 1}. Extend f by zero to F on B(0, R), the smallest ball centered at the origin and containing Ω. Then F ∈ W n,∞ (B(0, R)), and for all x = rω ∈ Ω, Z Z Vol(Ω) − f (y)dy f (x) − − f (y)dy ≤ N (F ) + 1 − Vol(B(0, R)) Ω Ω ( n−1 Z N +k−m ! X k R k ∂ k F (rω 0 ) 0 N X 1 m m (−1) r dω − + N k m R k! ∂r N +k−m k=1
S n−1 n X
m=0
kDn f kL∞(Ω) n RN +n−m (−1)m rm n! N +n−m m m=0 !) n X rN +n n (−1)m − βn , m N +n−m m=0 +
(7.36)
where n ∈ N, βn = 0 or 2, according as n is even or odd. Here kDn f kL∞(Ω) = max kDα f kL∞(Ω) . |α|=n
Proof.
satisfies
Let R := inf{R0 > 0 : Ω ⊂ B(0, R0 )}. Then f (x), x ∈ Ω F (x) := 0, x ∈ B(0, R) − Ω,
(7.37)
(7.38)
F ∈ W n,∞ (B(0, R)) ∩ C0n−1 (B(0, R)). Then for x ∈ Ω Z f (x) − − f (y)dy ≤ Ω
where
Z F (x) − − F (y)dy B(0,R) Z Z + − F (y)dy − − f (y)dy = J1 + J2 , B(0,R) Ω
Z J1 := F (x) − − F (y)dy B(0,R)
(7.39)
(7.40)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
High Order Ostrowski Inequalities on Euclidean Domains
and
Book˙Adv˙Ineq
107
Z 1 1 − J2 := f (y)dy Vol(B(0, R)) Vol(Ω) Ω Z Vol(Ω) (7.41) − f (y)dy . = 1− Vol(B(0, R)) Ω
n Theorem 7.1 applies and we may replace ∂∂rFn L∞ (B(0,R)) by its majorant kDn f kL∞(Ω) . The theorem then follows.
This page intentionally left blank
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Chapter 8
Ostrowski Inequalities on Spherical Shells
Here are presented Ostrowski type inequalities over spherical shells. These regard sharp or close to sharp estimates to the difference of the average of a multivariate function from its value at a point. This chapter relies on [47]. 8.1
Introduction
The famous Ostrowski’s inequality (1938), see [196], is ! 1 Z b 2 1 (x − a+b 2 ) (b − a)kf 0 k∞ , f (y)dy − f (x) ≤ + b − a a 4 (b − a)2
for f ∈ C 1 ([a, b]), x ∈ [a, b], and it is a sharp inequality. This was generalized from intervals to boxes in RN , N ≥ 1 , see [17], [21], p. 507. Here we establish Ostrowski type inequalities over spherical shells. We present first the sharp results for the radial functions, then we move to the non-radial case. We use the polar method. The estimates in both cases involve radial derivatives of arbitrary order of the engaged function. At the end we give the connection of radial derivative to the ordinary partial derivative of the function. 8.2
Main Results
We make Remark 8.1. Let A be a spherical shell ⊆ RN , N ≥ 1, i.e. A := B(0, R2 ) − B(0, R1 ), 0 < R1 < R2 . Here the ball B(0, R) := {x ∈ RN : |x| < R}, R > 0 , where | · | is the Euclidean norm , also S N −1 := {x ∈ RN : |x| = 1} is the unit sphere in RN with surface 2π N/2 area ωN := Γ(N/2) . For x ∈ RN − {0} one can write uniquely x = rω, where r > 0, ω ∈ S N −1 . 109
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
110
¯ We suppose first that f is radial i.e f (x) = g(r), where r = Let f ∈ C 1 (A). |x|, R1 ≤ r ≤ R2 . Clearly here g ∈ C 1 ([R1 , R2 ]). In general it holds k ∂f ∂r k∞ ≤ k 5 f k∞ , with equality in the radial case. ¯ we have For F ∈ C(A) ! Z R2 Z Z N −1 F (rω)r dr dω. F (x)dx = S N −1
A
We notice that
Z
N N R2 − R1N
and
R1
R2
sN −1 ds = 1,
ωN (R2N − R1N ) . N Let x ∈ A. Then by using the polar method we derive R R R2 R N −1 N f (sω)s ds dω N −1 S R1 f (x) − A f (y)dy = f (x) − V ol(A) ωN (R2N − R1N ) V ol(A) =
R R R N S N −1 R12 g(s)sN −1 ds dω = g(r) − ωN (R2N − R1N ) Z R2 N N −1 = g(r) − g(s)s ds N N R2 − R 1 R1
! Z R2 Z R2 N N −1 N −1 g(s)s ds = g(r)s ds − R2N − R1N R1 R1 =
N N R2 − R1N
≤
N R2N − R1N
R2
R1
N R2N − R1 =
R1
N N R2 − R1N
(8.2)
(8.3)
(8.4)
(8.5)
(8.6)
|g(r) − g(s)| sN −1 ds ≤
Z R2 N 0 kg k∞ |r − s| sN −1 ds R2N − R1N R1 # "Z Z R2 r N −1 0 N −1 (s − r)s ds kg k∞ (r − s)s ds + N
=
Z R2 (g(r) − g(s)) sN −1 ds R1
Z
(8.1)
R1
r
N 2r − (R1N + R2N ) kg 0 k∞ r N
(8.7)
(8.8)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Ostrowski Inequalities on Spherical Shells
R1N +1 + R2N +1 − 2rN +1 N +1
+
!#
Book˙Adv˙Ineq
111
.
(8.9)
So we have established the first main result. ¯ be radial, i.e. f (x) = g(r), R1 ≤ r ≤ R2 , x ∈ A. ¯ Theorem 8.2. Let f ∈ C 1 (A) Then R Z R2 f (y)dy N N −1 A f (x) − = g(r) − g(s)s ds ≤ V ol(A) RN − RN 2
N N R2 − R1N
N 2r − (R1N + R2N ) kg k∞ r N 0
R1N +1 + R2N +1 − 2rN +1 N +1
+
=
R1
1
N R2N − R1N
(8.10)
2|x|N − (R1N + R2N ) k5f k∞ |x| N
R1N +1 + R2N +1 − 2|x|N +1 N +1
+
!#
!#
.
(8.11)
Optimality comes next Theorem 8.3. Inequality (8.10) is sharp. More precisely (i) it is asymptotically attained by g ∗ (z) := |z−r|α , 1 < α ≤ k, when 0 < r < R. (ii) It is attained by g ∗ (z) = (z − R1 ), when r = R1 . (iii) It is attained by g ∗ (z) = (z − R2 ), when r = R2 . Proof.
(i) We observe that 0
g ∗ (z) = α|z − r|α−1 sign(z − r) and 0
kg ∗ k∞ = α(max{R2 − r, r − R1 })α−1 , along with g ∗ (r) = 0. We see that L.H.S(8.10) =
α→1
−→
N R2N − R1N
N N R2 − R1N
Z
Z R2
R1
R2 R1
α
|s − r| sN −1 ds
|s − r| sN −1 ds
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
112
=
Book˙Adv˙Ineq
N R2N − R1N
" N 2r − (R1N + R2N ) r + N
We also have R.H.S(8.10) =
N N R2 − R1N
" 2rN − (R1N + R2N ) + r N
N R2N − R1N
R1N +1 + R2N +1 − 2rN +1 N +1
!#
. (8.12)
α(max{R2 − r, r − R1 })α−1
R1N +1 + R2N +1 − 2rN +1 N +1
" N 2r − (R1N + R2N ) + r N
!#
α→1
−→
R1N +1 + R2N +1 − 2rN +1 N +1
!#
.
(8.13)
That is lim L.H.S(8.10) = lim R.H.S(8.10),
α→1
α→1
proving sharpness for the case. 0 (ii) We have g ∗ (R1 ) = 0 and kg ∗ k∞ = 1. Thus Z R2 N L.H.S(8.10) = (s − R1 )sN −1 ds R2N − R1N R1 ! # " N R2N − R1N R2N +1 − R1N +1 = − R1 = R.H.S (8.10), (8.14) N +1 N R2N − R1N proving the attainability for the case. (iii) We have g ∗ (R2 ) = 0, and kg ∗0 k∞ = 1. Hence Z R2 N L.H.S(8.10) = (R2 − s)sN −1 ds = R2N − R1N R1 !# " N R2 − R1N R2N +1 − R1N +1 N R2 − = R.H.S (8.10), N N +1 R2N − R1N proving attainability of last case.
(8.15)
We would line to rewrite Theorem 8.3 for the equivalent inequality (8.10) in terms of f . We have ¯ Inequality (8.10) is sharp as follows: Theorem 8.4. Let x ∈ A. (i) Let 0 < |x| < R, then it is asymptotically attained by f ∗ (w) :=| |w| − |x| |α , 1 < α ≤ k.
(ii) Let |x| = R1 , then it is asymptotically attained by f ∗ (w) = |w| − R1 .
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Ostrowski Inequalities on Spherical Shells
Book˙Adv˙Ineq
113
(iii) Let |x| = R2 , then it is asymptotically attained by f ∗ (w) = |w| − R2 . We continue from Remark 8.1 into ¯ n ∈ N , again radial such that f (x) = g(r), where Remark 8.5. Now f ∈ C n (A), ¯ r = |x|, x ∈ A, R1 ≤ r ≤ R2 . Hence g ∈ C n ([R1 , R2 ]). Using the polar method we get again R Z R2 f (y)dy N N −1 A = E := f (x) − (g(r) − g(s))s ds (8.16) . V ol(A) R2N − R1N R1 Let s, r ∈ [R1 , R2 ], then by Taylor’s formula we obtain g(s) − g(r) = where Rn−1 (r, s) :=
Z
n−1 X k=1
s r
As in [21], p. 500, we find
g (k) (r) (s − r)k + Rn−1 (r, s), k!
g (n−1) (t) − g (n−1) (r)
|Rn−1 (r, s)| ≤ ∀ s, r ∈ [R1 , R2 ]. Therefore Z R2 N E= R2N − R1N R1 ≤
N R2N − R1N
"n−1 X k=1
≤
n−1 X k=1
|g (k) (r)| k!
N N R2 − R1N
But one finds that
kg (n) k∞,[R1 ,R2 ] |s − r|n , n!
! g (k) (r) k N −1 (s − r) + Rn−1 (r, s) s ds k!
(8.18)
(8.19)
(8.20)
Z Z # R2 R2 N −1 N −1 k s |Rn−1 (r, s)|ds s (s − r) ds + R1 R1
"n−1 X
R2N +k−m − R1N +k−m N +k−m
(s − t)n−2 dt. (n − 2)!
(8.17)
k=1
|g k (r)| k!
k X k (−1)m rm m m=0
! # kg (n) k Z R2 ∞ N −1 n s |s − r| ds . + n! R1 Z
R2
R1
sn−1 |s − r|n ds
(8.21)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
114
" n X n m rn−m (−1) = m m=0
rm+N − R1m+N m+N
!
+r
m
R2N +n−m − rN +n−m N +n−m
!#
.
(8.22)
Putting all the above together we have derived ¯ n ∈ N, be radial, i.e. f (x) = g(r), R1 ≤ r ≤ Theorem 8.6. Let f ∈ C n (A), ¯ R2 , x ∈ A. Then R Z R2 f (y)dy N N −1 A f (x) − = g(r) − g(s)s ds V ol(A) R2N − R1N R1 ≤
N R2N − R1N
"n−1 X k=1
+ g (n)
∞
"
k X (−1)m rm |g (k) (r)| m!(k − m)! m=0
" (−1)m rn−m m!(n − m)! m=0 n X
+rm
=
N N R2 − R1N
R2N +k−m − R1N +k−m N +k−m
rm+N − R1m+N m+N
R2N +n−m − rN +n−m N +n−m
!###
k (n−1 X ∂ k f (x) X (−1)m |x|m ∂rk m!(k − m)! m=0 k=1
" " n
n X (−1)m
∂ f
|x|n−m · + n ∂r ∞,A¯ m=0 m!(n − m)! + |x|
m
!
!
R2N +n−m − |x|N +n−m N +n−m
(8.23)
R2N +k−m − R1N +k−m N +k−m |x|m+N − R1m+N m+N !##)
.
!
!
(8.24)
We give ¯ n ∈ N, be radial i.e. f (x) = f (r), R1 ≤ r ≤ Corollary 8.7. Let f ∈ C n (A), i ∂ f (x ) 0 ¯ Suppose R2 , x ∈ A. = g (i) (r0 ) = 0, i = 1, . . . , n − 1, for r0 ∈ [R1 , R2 ], x0 = ∂ i ri N −1 ¯ r0 ω ∈ A, ω ∈ S . Then R Z R2 f (y)dy N N −1 A = g(r0 ) − f (x0 ) − g(s)s ds N N V ol(A) R2 − R 1 R1 N kg (n) k∞ ≤ N R2 − R1N
"
n X
(−1)m m!(n − m)! m=0
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Spherical Shells
"
!
r0m+N − R1m+N m+N
r0n−m
+
n
= "
|x0 |n−m
N k ∂∂rnf k∞,A R2N
|x0 |m+N − R1m+N m+N
−
!
R2N +n−m − r0N +n−m N +n−m
r0m
R1N
"
115
!##
n X
(−1)m m!(n − m)! m=0
+ |x0 |m
R2N +n−m − |x0 |N +n−m N +n−m
(8.25)
!##
.
(8.26)
We also have the extreme cases. ¯ n ∈ N, be radial ; f (x) = g(r), r ∈ [R1 , R2 ], x ∈ Corollary 8.8. Let f ∈ C n (A), i ∂ f (x) ¯ x = rω. Suppose that A, ∂r i , i = 1, . . . , n − 1, are zero on ∂B(0, R1 ), i.e. g (i) (R1 ) = 0, i = 1, . . . , n − 1. Then for x ∈ ∂B(0, R1 ) we have R Z R2 f (y)dy N N −1 A = g(R1 ) − f (x) − g(s)s ds V ol(A) R2N − R1N R1 ≤
=
N N R2 − R1N
N R2N − R1N
kg
(n)
k∞
"
n X
(−1)m R1m m!(n − m)! m=0
R2N +n−m − R1N +n−m N +n−m
" n n X (−1)m
∂ f
R1m
∂rn ¯ m!(n − m)! ∞,A m=0
!#
(8.27)
R2N +n−m − R1N +n−m N +n−m
!#
.
(8.28)
Another extreme case follows.
¯ n ∈ N, be radial; f (x) = g(r), r ∈ [R1 , R2 ], x ∈ Corollary 8.9. Let f ∈ C n (A), i f (x) barA; x = rω. Assume that ∂ ∂r i , i = 1, . . . , n − 1, are zero on ∂B(0, R2 ), i.e., (i) g (R2 ) = 0, i = 1, . . . , n − 1. Then for x ∈ ∂B(0, R2 ) we have R Z R2 f (y)dy N N −1 f (x) − A = g(R2 ) − g(s)s ds V ol(A) R2N − R1N R1
=
≤
N N R2 − R1N
N R2N − R1N
kg (n) k∞
"
n X
(−1)m R2n−m m!(n − m)! m=0
" n n X (−1)m
∂ f
R2n−m
∂rn ¯ ∞,A m=0 m!(n − m)!
R2m+N − R1m+N m+N
!#
R2m+N − R1m+N m+N
Optimality follows.
Proposition 8.10. Inequality (8.25) is sharp, namely it is attained by g ∗ (y) := (y − r0 )n ,
y,
r0 ∈ [R1 , R2 ], when n is even.
!#
(8.29)
. (8.30)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
116
Proof.
Notice (j)
g ∗ (r0 ) = 0,
j = 0, 1, . . . ,
We observe L.H.S(8.25) =
=
N R2N − R1N
Next we see that
"X n n (−1)m r0m m m=0
R.H.S (8.25) = " =
N R2N − R1N
=
!
+
R2 R1
(n)
k∞ = n!.
(s − r0 )n sN −1 ds
R2N +n−m − R1N +n−m N +n−m
!#
.
"X n n (−1)m m m=0
r0m R2N +n−m − r0N +n N +n−m
!##
n X n 1 1 m − (−1) N +m N +n−m m m=0
N R2N − R1N
=
N R2N − R1N
=
N N R2 − R1N
(8.33)
(8.34)
(X n rm RN +n−m n (−1)m 0 2 N +n−m m m=0
) n N +n−m X n m m R1 − (−1) r0 m N +n−m m=0
(8.32)
(X n n rm RN +n−m (−1)m 0 2 N +n−m m m=0
) n n−m m+N X n R1 m r0 − (−1) m+N m m=0
(8.31)
" # (X n m N +n−m n r0n−m R1m+N m r0 R2 (−1) − m N +n−m m+N m=0
+r0n+N
Z
N R2N − R1N
r0n+N − r0n−m R1m+N m+N
N R2N − R1N
n − 1 and kg ∗
(8.35)
) (X n N +n−m n − R1N +n−m ) m m (R2 (−1) r0 N +n−m m m=0 = L.H.S
That is proving the claim.
(8.25).
(8.36)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Spherical Shells
117
The other optimal case follows. Proposition 8.11. Inequality (8.25) is sharp, namely it is asymptotically attained by g ∗ (y) := |y − r0 |n−1+α , y, r0 ∈ [R1 , R2 ], 1 < α ≤ T , in the case of n is odd. g∗
Proof. It holds Also we have g∗ That is
(n)
(
k)
(r0 ) = 0,
k = 0, 1, . . . , n − 1.
(y) = (n − 1 + α)(n − 2 + α) · · · (α + 1)α|y − r0 |α−1 sign(y − r0 ). n (n) ∗ Y (y) = (n − j + α) |y − r0 |α−1 , g j=1
and
(n)
∗
g
∞
Consequently,
=
n Y
j=1
(n − j + α) (max{R2 − r0 , r0 − R1 })α−1 .
Z R2 N α→1 |s − r0 |n−1+α sN −1 ds −→ R2N − R1N R1 Z R2 N |s − r0 |n sN −1 ds R2N − R1N R1 ! " (X n N n r0m+N − R1m+N n−m m = (−1) r0 m+N m R2N − R1N m=0 !#) R2N +n−m − r0N +n−m +r0m =: µ. N +n−m L.H.S(8.25) =
Next we derive
R.H.S(8.25) = (
n X n (−1)m m m=0
(
N N R2 − R1N ( r0n−m
(8.38) (8.39)
(8.40)
Qn (n − j + α) (max{R − r , r − R })α−1 2 0 0 1 j=1 n!
r0m+N − R1m+N m+N
!
+ r0m
R2N +n−m − r0N +n−m N +n−m
(X n n N α→1 (−1)m −→ N N m R2 − R 1 m=0 ! !)) r0m+N − R1m+N R2N +n−m − r0N +n−m m + r0 = µ. m+N N +n−m
r0n−m
(8.37)
!)) (8.41)
(8.42)
That is, L.H.S (8.25), R.H.S(8.25) → µ, proving asymptotic attainability and sharpness of (8.25).
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
118
Optimality of external inequality (8.25) follows. Proposition 8.12. External inequality (8.25) is asymptotically attained, that is sharp as follows: ¯ (i) when n is even, then optimal function is f ∗ (w) := (|w| − |x0 |)n , w ∈ A. ∗ n−1+α (ii) when n is odd, then optional function is f (w) :=| |w| − |x0 | | ,1< ¯ α ≤ T, w ∈ A. Similarly we obtain
Proposition 8.13. Inequalities (8.23), (8.27) and (8.29) are asymptotically attained, therefore sharp, as inequality (8.25). A simple but general result follows. Theorem 8.14. Let ∅ 6= R be a convex bounded region of RN , N ≥ 1. Let f ∈ C 1 (R). Then R Z f (y)dy k∇f k∞ f (x) − R ≤ |x − y|dy, x ∈ R. (8.43) V ol(R) V ol(R) R
Proof.
We observe that R Z 1 f (x) − R f (y)dy = f (x)V ol(R) − f (y)dy V ol(R) V ol(R) R Z 1 (f (x) − f (y))dy = V ol(R) R Z 1 ≤ |f (x) − f (y)|dy V ol(R) R = (z belongs to the line segment from x to y) Z 1 |∇f (z) · (x − y)|dy V ol(R) R ≤
1 V ol(R)
Z
R
k∇f k∞ V ol(R)
|∇f (z)| · |x − y|dy ≤
Specializing on the shell and sphere we have
Z
R
|x − y|dy.
N Γ( N2 ) k∇f k∞ 2π N/2 (R2N − R1N )
Z
A
|x − y| dy,
(8.45)
(8.46)
¯ or f ∈ C 1 (B(0, R)), R > 0. Then Proposition 8.15. Let f ∈ C 1 (A), (i) Z N Γ( N2 ) f (y)dy f (x) − N/2 N 2π (R2 − R1N ) A ≤
(8.44)
¯ x ∈ A;
(8.47)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Spherical Shells
119
also it holds (ii) Z Z N Γ( N2 ) N Γ( N2 ) f (y)dy ≤ |x − y| dy, k∇f k∞ f (x) − N/2 N 2π N/2 RN 2π R B(0,R) B(0,R)
x ∈ B(0, R). More precise Ostrowski type inequalities for general, not necessarily radial functions, follow. ¯ x ∈ A¯ , x = rω, r > 0. Then Theorem 8.16. Let f ∈ C 1 (A), R R f (y)dy A ≤ f (x) − S N −1 f (rω)dω f (x) − V ol(A) ωN +
N R2N − R1N +
" Proof.
|x|
∂f
∂r
∞
|x|
2|x|N − (R1N + R2N ) N
R1N +1 + R2N +1 − 2|x|N +1 N +1
!#
(8.48)
R Γ( N2 ) S N −1 f (rω)dω N k∇f k∞ + ≤ f (x) − 2π N/2 R2N − R1N
2|x|N − (R1N + R2N ) N
+
R1N +1 + R2N +1 − 2|x|N +1 N +1
!#
.
(8.49)
Applying internal (8.10) to f (rω) we get Z R2 N N −1 f (sω)s ds f (rω) − R2N − R1N R1
≤
N R2N − R1N
N
∂f (rω) 2r − (R1N + R2N )
r
∂r N ∞,r∈[R1 ,R2 ] +
≤
N N R2 − R1N +
R1N +1 + R2N +1 − 2rN +1 N +1
!#
(8.50)
∂f 2|x|N − (R1N + R2N )
∂r ¯ |x| N ∞,A R1N +1 + R2N +1 − 2|x|N +1 N +1
!#
.
(8.51)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
120
Hence it holds R Z N S N −1 f (rω)dω − ωN ωN (R2N − R1N ) S n−1 ≤
N R2n − R1N
Z
R2
f (sω)s R1
ds dω !
∂f 2|x|N − (R1N + R2N )
|x|
∂r N ∞
R1N +1 + R2N +1 − 2|x|N +1 N +1
+
N −1
!#
,
(8.52)
proving the claim.
We continue with ¯ ¯ x = rω, r > 0. Then Theorem 8.17. Let f ∈ C n (A), n ∈ N, x ∈ A, R R N f (y)dy Γ( f (rω)dω ) N −1 S 2 f (x) − A ≤ f (x) − V ol(A) 2π N/2 +
N R2N − R1N
(
Γ( N2 ) 2π N/2
k X (−1)m |x|m m!(k − m)! m=0
(n−1 Z X k=1
S
k ∂ f (rω) ∂rk dω N −1
R2N +k−m − R1N +k−m N +k−m
"
n "X n
∂ f (−1)m
|x|n−m + n ∂r ∞ m=0 m!(n − m)! +|x|
Proof.
m
! )
|x|m+N − R1m+N m+N
R2N +n−m − |x|N +n−m N +n−m
!##)
!
.
Applying internal (8.23) to f (rω) we obtain Z R2 N N −1 f (sω)s ds f (rω) − R2N − R1N R1 ≤
N R2N − R1N
(n−1 X ∂ k f (rω) ∂rk k=1
R2N +k−m − R1N +k−m N +k−m
k X (−1)m |x|m m! (k − m)! m=0
! " n ∂nf X (−1)m
+ n
∂r ∞,A¯ m=0 m!(n − m)!
(8.53)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Spherical Shells
"
|x|
n−m
|x|m+N − R1m+N m+N
Therefore it holds
≤
!
+ |x|
m
121
R2N +n−m − |x|N +n−m N +n−m
!##)
.
(8.54)
Z Γ( N ) Z 1 2 f (y)dy f (rω)dω − N/2 2π V ol(A) A S N −1
N R2 − R1N
(
Γ(N/2) 2π N/2
k X (−1)m |x|m m!(k − m)! m=0
(n−1 Z X k=1
S
k ∂ f (rω) ∂rk dω N −1
R2N +k−m − R1N +k−m N +k−m
" n "
n X (−1)m
∂ f
+ n |x|n−m ∂r ∞,A¯ m=0 m!(n − m)! +|x|
m
! )
|x|m+N − R1m+N m+N
R2N +n−m − |x|N +n−m N +n−m
!##)
!
,
(8.55)
proving the claim.
We also give i
¯ n ∈ N, such that ∂ fi , i = 1, . . . , n − 1, are zero Proposition 8.18. Let f ∈ C n (A), ∂r on ∂B(0, r0 ), r0 ∈ (R1 , R2 ). Then for x0 ∈ ∂B(0, r0 ) we have R Z f (x0 ) − A f (y)dy ≤ f (x0 ) − Γ(N/2) f (r ω)dω 0 N/2 V ol(A) 2π S N −1 +
N R2N − R1N
" n "X n m
∂ f (−1)
|x0 |n−m
∂rn m!(n − m)! ∞ m=0 +|x0 |
Proof.
m
R2N +n−m − |x0 |N +n−m N +n−m
|x0 |m+N − R1m+N m+N
!##
.
Applying internal (8.25) to f (rω) we find Z R2 N N −1 f (sω)s ds f (r0 ω) − R2N − R1N R1 ≤
N R2N − R1N
" n n X (−1)m
∂ f
∂rn ¯ ∞,A m=0 m!(n − m)!
!
(8.56)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
122
"
|x0 |
n−m
|x0 |m+N − R1m+N m+N
Therefore
≤
N N R2 − R1N
!
+ |x0 |
m
R2N +n−m − |x0 |N +n−m N +n−m
!##
.
Z Γ( N ) Z 1 2 f (r0 ω)dω − f (y)dy N/2 2π V ol(A) A S N −1
" n "X n
∂ f (−1)m
|x0 |n−m
∂rn ∞ m=0 m!(n − m)! +|x0 |
m
R2N +n−m − |x0 |N +n−m N +n−m
|x0 |m+N − R1m+N m+N
!##
.
(8.57)
! (8.58)
Claim is clear.
We present the extreme cases. i ¯ Proposition 8.19. Let f ∈ C n (A), n ∈ N such that ∂∂rfi , i = 1, . . . , n − 1, are zero on ∂B(0, R1 ). Then for x0 ∈ ∂B(0, R1 ) it holds R N Z f (x0 ) − A f (y)dy ≤ f (x0 ) − Γ( 2 ) f (R ω)dω 1 V ol(A) 2π N/2 S N −1
+
N N R2 − R1N
Proof.
n "X n
∂ f (−1)m
R1m
∂rn ∞ m=0 m!(n − m)!
R2N +n−m − R1N +n−m N +n−m
!#
By internal (8.27).
(8.59)
We finish the main results with i ¯ Proposition 8.20. Let f ∈ C n (A), n ∈ N, such that ∂∂rfi , i = 1, . . . , n − 1, are zero on ∂B(0, R2 ). Then for x0 ∈ ∂B(0, R2 ) we find R (N ) Z f (x0 ) − A f (y)dy ≤ f (x0 ) − Γ 2 f (R ω)dω 2 V ol(A) 2π N/2 S N −1
+
N N R2 − R1N
Proof.
n "X n
∂ f (−1)m
R2n−m
∂rn ∞ m=0 m!(n − m)!
By internal (8.29).
R2m+N − R1m+N m+N
!#
.
(8.60)
The radial derivatives appearing in the right hand sides of the inequalities can be expressed and estimated by regular partial derivatives in terms of x1 , . . . , xN . See Addendum next.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Ostrowski Inequalities on Spherical Shells
8.3
Book˙Adv˙Ineq
123
Addendum
Let u ∈ C n (B(0, R)), the open ball B(0, R) ⊆ RN , n, N ∈ N. (x1 , · · · , xN ) ∈ B(0, R) and the radial derivative of u is given also by x ∂u(x) = ∇u(x) · , ∂r |x|
That is,
∂u(x) 1 = ∂r |x|
N X ∂u(x)
∂xi
i=1
Here x =
x 6= 0.
xi
!
,
x 6= 0.
(8.61)
In general for 1 ≤ l ≤ n it holds (by induction) that X 1 l! ∂ l u(x) ∂ l u(x) k = · N k ΠN x j , x 6= 0. N l l ∂r |x| Πj=1 kj ! Πj=1 ∂ j xj j=1 j PN k1 ;...;kN ,
j=1
kj =l,kj ∈Z+
(8.62)
For example, when n = N = 2 we obtain 1 ∂ 2 u(x) 2 2∂ 2 u(x) ∂ 2 u(x) 2 ∂ 2 u(x) = x + 2 x x + x , x 6= 0. 1 2 ∂r2 |x|2 ∂x21 1 ∂x1 ∂x2 ∂x22 2 Thus we find l ∂ u(x) ∂rl ≤
or better in brief,
k1 ;...;kN ,
X
PN
j=1 kj =l,kj ∈Z+
l ∂ u(x) ∂rl ≤
∂ l u(x) · N k , x 6= 0, N j Πj=1 kj ! Πj=1 ∂ xj l!
N X ∂ ∂xi i=1
!l
(u(x)) ,
(8.63)
(8.64)
(8.65)
all x ∈ B(0, R) − {0}, all 1 ≤ l ≤ n. So if all the u partial derivatives vanish then the corresponding radial derivative is zero. Consequently, from (8.65) it holds for the essential supreme k · k∞ that !l
l N X
∂
∂ u
≤ (u) < +∞. (8.66)
∂xi
∂rl ∞,B(0,R) ∞,B(0,R) i=1
II. Continuing and specializing in the study of higher order radial derivatives of radial functions. Let now u be radial i.e. u(x) = g(|x|) = g(r); x = rω, r ∈ R+ , ω ∈ S N −1 . Here we will suppose x 6= 0 and N = 2. Let further x1 , x2 6= 0, then by chain rule one has q ∂u r ∂u r g 0 (r) = = , where r = x21 + x22 . (8.67) ∂x1 x1 ∂x2 x2
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
124
That is,
∂u |x| 1 ∂u |x| + . g (r) = 2 ∂x1 x1 ∂x2 x2 0
So one has
Book˙Adv˙Ineq
∂u(x) , |∇u(x)| = |g (r)| = ∂r 0
(8.68)
(8.69)
for any x ∈ B(0, R). ∂u But if u radial, not necessarily ∂x is radial. Here we put x1 = r cos θ, x2 = j r sin θ. Again for x1 , x2 6= 0 and via chain rule we derive ∂ 2 u(x) ∂ 2 u(x) ∂ 2 u(x) ∂ 2 u(x) g 00 (r) = + tan θ = + cot θ. (8.70) ∂x21 ∂x2 ∂x1 ∂x22 ∂x1 ∂x2 That is ∂ 2 u(x) 1 ∂ 2 u(x) ∂ 2 u(x) + + (tan θ + cot θ) . (8.71) g 00 (r) = 2 ∂x21 ∂x22 ∂x1 ∂x2 Or better, by using the Laplacian ∆ we find ∂ 2 u(x) 1 g 00 (r) = ∆u(x) + csc(2 θ), (8.72) 2 ∂x1 ∂x2 or 1 ∂ 2 u(x) |x|2 00 g (r) = ∆u(x) + , (8.73) 2 ∂x1 ∂x2 x1 x2 x1 , x2 6= 0. Similarly, one has that ∂ 3 u(x) ∂ 3 u(x) sin2 θ ∂ 3 u(x) cos θ + 2 sin θ + , x1 , x2 6= 0. (8.74) g 000 (r) = ∂x31 ∂x2 ∂x21 ∂x22 ∂x1 cos θ Also we get ∂ 3 u(x) ∂ 3 u(x) ∂ 3 u(x) cos2 θ g 000 (r) = 2 , x1 , x2 6= 0. (8.75) cos θ + sin θ + ∂x1 ∂x22 ∂x32 ∂x21 ∂x2 sin θ That is, by (8.74) and (8.75) we have ∂ 3 u(x) 1 ∂ 3 u(x) 000 cos θ + sin θ+ g (r) = 2 ∂x31 ∂x32 3 ∂ 3 u(x) cos2 θ ∂ 3 u(x) ∂ 3 u(x) sin2 θ ∂ u(x) + sin θ + cos θ + , x1 , x2 6= 0. 2 ∂x2 ∂x21 ∂x1 ∂x22 ∂x22 ∂x1 cos θ ∂x21 ∂x2 sin θ (8.76) Clearly, it holds 1 ∂ 3 u(x) x1 ∂ 3 u(x) x2 g 000 (r) = + + 2 ∂x31 |x| ∂x32 |x| 2 2 x2 x1 x1 ∂ 3 u(x) x2 ∂ 3 u(x) + 2 + 2 + , x1 , x2 6= 0. (8.77) ∂x21 ∂x2 x2 |x| |x| ∂x22 ∂x1 x1 |x| |x| Not general formula, as in (8.62), can be derived for g (l) (r), l ∈ N, in the radial case. Of course (8.62) is valid for both non-radial and radial cases. Notice in (8.67), (8.70), (8.74) and (8.75) are used less number of terms than in the corresponding non-radial cases.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Chapter 9
Ostrowski Inequalities on Balls and Shells Via Taylor Widder Formula
The classical Ostrowski inequality for functions on intervals estimates the value of the function minus its average in terms of the maximum of its first derivative. This result is extended to higher order over shells and balls of RN , N ≥ 1, with respect to an extended complete Tschebyshev system and the generalized radial derivatives of Widder type. We treat radial and non-radial functions. This chapter relies on [42].
9.1
Introduction
The classical Ostrowski inequality (of 1938, see [196]) is 2 a+b x− 1 1 Z b 2 f (y)dy − f (x) ≤ + (b − a)kf 0 k∞ , 4 b − a a (b − a)2 for f ∈ C 1 ([a, b]), x ∈ [a, b], and it is a sharp inequality. This was extended to RN , N ≥ 1, over balls and shells in [56], [57], [47]. Also this extension was done over boxes and rectangles, see [21], pp. 507-520, and [17], and [16]. The produced Ostrowski inequalities, in the above mentioned references, were mostly sharp and they involved the first and higher order derivatives of the engaged function f . Here we derive a set of very general higher order Ostrowski type inequalities over shells and balls with respect to an extended complete Tschebyshev system (see [163])and generalized derivatives if Widder type (see [243]). The proofs are based on the polar method and the general Taylor-Widder formula (see [243], 1928). These results generalize the higher order Ostrowski type inequalities established in the above mentioned references. 125
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
126
9.2
Book˙Adv˙Ineq
Background
The following are taken from [243]. Let f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]),n ≥ 0, and the Wronskians u0 (x) u1 (x) . . . ui (x) u0 (x) u0 (x) . . . u0 (x) 0 1 i , i = 0, 1, . . . , n. Wi (x) := W [u0 (x), u1 (x), . . . , ui (x)] := .. . (i) (i) u0 (x) u(i) 1 (x) . . . ui (x) Assume Wi (x) > 0 over [a, b]. Clearly then φ0 (x) := W0 (x) = u0 (x), φ1 (x) :=
Wi (x)Wi−2 (x) W1 (x) , . . . , φi (x) := , (W0 (x))2 (Wi−1 (x))2
i = 2, 3, . . . , n, are positive on [a, b]. For i ≥ 0, the linear differentiable operator of order i: Li f (x) :=
W [u0 (x), u1 (x), . . . , ui−1 (x), f (x)] , Wi−1 (x)
i = 1, . . . , n + 1;
L0 f (x) := f (x), ∀ x ∈ [a, b]. Then for i = 1, . . . , n + 1 we have Li f (x) = φ0 (x)φ1 (x) . . . φi−1 (x) Consider also
d d 1 d f (x) 1 d 1 d ... . dx φi−1 (x) dx φi−2 (x) dx dx φ1 (x) dx φ0 (x)
u0 (t) u1 (t) u0 (t) u01 (t) 1 0 ... gi (x, t) := ... Wi (t) (i−1) (i−1) u0 (t) u (t) 1 u0 (x) u1 (x)
. . . ui (t) . . . u0i (t) ... . . . , (i−1) . . . ui (t) . . . ui (x)
u0 (x) , ∀ x, t ∈ [a, b]. u0 (t) Note that gi (x, t) as a function of x is a linear combination of u0 (x), u1 (x), . . . , ui (x) and it holds
i = 1, 2, . . . , n; g0 (x, t) :=
gi (x, t)
Z xi−1 Z xi−2 Z x Z x1 φ0 (x) φi (xi )dxi dxi−1 . . . dx1 φi−1 (xi−1 ) ... φ1 (x) = φ0 (t) . . . φi (t) t t t t Z x 1 = φ0 (s) . . . φi (s)gi−1 (x, s)ds, i = 1, 2, . . . , n. φ0 (t) . . . φi (t) t
Example 9.1. ([243]). The sets {1, x, x2 , . . . , xn }, {1, sin x, − cos x, − sin 2x, cos 2x, . . . , (−1)n−1 sin nx, (−1)n cos nx}
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Ostrowski Inequalities on Balls and Shells Via Taylor–WidderFormula
Book˙Adv˙Ineq
127
fulfill the above theory. We mention Theorem 9.2. (Karlin and Studded (1966), see p. 376, [163]). Let u0 , u1 , . . . , un ∈ C n ([a, b]), n ≥ 0. Then {ui }ni=0 is an extended complete Tschebyshev system on [a, b] iff Wi (x) > 0 on [a, b], i = 0, 1, . . . , n. We also mention Theorem 9.3. (D. Widder, p. 138, [243]). Let the functions f (x), u0 (x), u1 (x), . . . , un (x) ∈ C n+1 ([a, b]), and the Wronskians W0 (x), W1 (x), . . . , Wn (x) > 0 on [a, b], x ∈ [a, b]. Then for t ∈ [a, b] we have f (x) = f (t)
u0 (x) + L1 f (t)g1 (x, t) + · · · + Ln f (t)gn (x, t) + Rn (x), u0 (t)
where Rn (x) :=
Z
x
gn (x, s)Ln+1 f (s)ds. t
For example one could take u0 (x) = c > 0. If ui (x) = xi , i = 0, 1, . . . , n, defined on [a, b], then (x − t)i , t ∈ [a, b]. i! So under the assumptions of Theorem 9.3 we get Z x n u0 (x) X f (x) = f (y) + Li f (y)gi (x, y) + gn (x, t)Ln+1 f (t)dt, ∀ x, y ∈ [a, b]. u0 (y) i=1 y (9.1) If u0 (x) = c > 0, then Z x n X f (x) = f (y) + Li f (y)gi (x, y) + gn (x, t)Ln+1 f (t)dt, ∀ x, y ∈ [a, b]. (9.2) Li f (t) = f (i) (t)
i=1
and gi (x, t) =
y
We call Li the generalized Widder-type derivative. We need
Notation 9.4. Let A be a spherical shell ⊆ RN , N ≥ 1, i.e. A := B(0, R2 ) − B(0, R1 ), 0 < R1 < R2 . Here the ball B(0, R) := {x ∈ RN : |x| < R}, R > 0, where | · | is the Euclidean norm, also S N −1 := {x ∈ RN : |x| = 1} is the unit sphere in RN with surface area 2π N/2 . For x ∈ RN − {0} one can write uniquely x = rω, where r > 0, ωN := Γ(N/2) ω ∈ S N −1 . Let f ∈ C n+1 (A), n ≥ 0. If f is radial i.e. f (x) = g(r), where r = |x|, R1 ≤ r ≤ R2 , then g ∈ C n+1 ([R1 , R2 ]).
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
128
For radial f define θi f (x) := Li g(r),
all i = 1, . . . , n + 1, ∀ x ∈ A.
(9.3)
Here g (i) (r) =
∂ i f (x) , ∂ri
For F ∈ C(A) we have Z Z F (x)dx = A
S N −1
We notice that N N R2 − R1N
and
V ol(A) =
9.3
Z Z
i = 1, . . . , n + 1.
R2
F (rω)r
N −1
R1
R2
dr
!
dω.
sN −1 ds = 1,
(9.4)
(9.5)
R1
ωN (R2N − R1N ) . N
Results on the Shell
We make Remark 9.5. Let here u0 , u1 , . . . , un ∈ C n+1 ([R1 , R2 ]), and W0 , W1 , . . . , Wn > 0 on [R1 , R2 ], 0 < R1 < R2 , n ≥ 0 integer, with u0 (r) = c > 0. Let also f ∈ C n+1 (A). We suppose first that f is radial, i.e. there exists g such that f (x) = g(r), r = |x|, R1 ≤ r ≤ R2 . Clearly g ∈ C n+1 ([R1 , R2 ]). Let x ∈ A. Then by using the polar method (9.4) we obtain ! Z R2 Z Z N −1 f (sω)s ds dω N f (y)dy N −1 R S 1 A (9.6) E(x) := f (x) − = f (x) − N N V ol(A) ωN (R2 − R1 ) ! Z R2 Z g(s)sN −1 ds dω N R1 S N −1 = g(r) − N − RN ) ω (R N 2 1 Z R2 N N −1 = g(r) − g(s)s ds N N R2 − R 1 R1
! Z R2 Z R2 N N −1 N −1 g(s)s ds = g(r)s ds − N N R2 − R 1 R1 R1
(9.7)
(9.8)
(9.9)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Balls and Shells Via Taylor–WidderFormula
=
N R2N − R1N
Z R2 N −1 (g(r) − g(s))s ds =: (∗). R1
129
(9.10)
Let s, r ∈ [R1 , R2 ], then by generalized Taylor’s formula (9.2) we have n X
g(s) − g(r) = where
Rn (r, s) := But it holds
By calling
we find
Z |Rn (r, s)| ≤ Z Nn (r, s) :=
s r
s
r
Li g(r)gi (s, r) + Rn (r, s),
Z
s
gn (s, t)Ln+1 g(t)dt.
(9.12)
r
|gn (s, t)|dt kLn+1 gk∞,[R1 ,R2 ] .
|gn (s, t)|dt , ∀ s, r ∈ [R1 , R2 ],
|Rn (r, s)| ≤ Nn (r, s)kLn+1 gk∞,[R1 ,R2 ] , ∀ s, r ∈ [R1 , R2 ]. Therefore by (9.10), (9.11), we have # Z R2 "X n N N −1 L g(r)g (s, r) + R (r, s) s ds (∗) = i i n R2N − R1N R1 i=1 ≤
N R2N − R1N
(9.11)
i=1
(9.13)
(9.14)
(9.15)
(9.16)
Z # "X n Z R2 R2 N −1 N −1 |Rn (r, s)|s ds Li g(r)gi (s, r)s ds + R1 R1 i=1
by (9.15) ≤
N N R2 − R1N
"X n i=1
+(kLn+1 gk∞,[R1 ,R2 ] )
Z R2 N −1 |Li g(r)| gi (s, r)s ds R1
Z
R2
Nn (r, s)s
R1
N −1
#
ds .
(9.17)
We have established the following result. Theorem 9.6. Let u0 , u1 , . . . , un ∈ C n+1 ([R1 , R2 ]), W0 , W1 , . . . , Wn > 0 on [R1 , R2 ], 0 < R1 < R2 , n ≥ 0 integer, with u0 (r) = c > 0. Let f ∈ C n+1 (A) be radial, i.e. there exists g such that f (x) = g(r), r = |x|, R1 ≤ r ≤ R2 , x ∈ A. Then Z Z R2 f (y)dy N N −1 A = g(r) − E(x) := f (x) − g(s)s ds N − RN V ol(A) R R 2 1 1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
130
≤
N R2N − R1N
"X n i=1
Z R2 N −1 gi (s, r)s ds |Li g(r)| R1
+(kLn+1 gk∞,[R1 ,R2 ] )
=
N R2N − R1N
"X n i=1
Z
R2
Nn (r, s)s
N −1
ds
R1
#
(9.18)
Z R2 N −1 gi (s, |x|)s ds |θi f (x)| R1
+(kθn+1 f k∞,A )
Z
R2
#
Nn (|x|, s)sN −1 ds .
R1
We give Corollary 9.7. Same terms and assumptions as in Theorem 9.6. Suppose further that Li g(r0 ) = 0, i = 1, . . . , n, for a fixed r0 ∈ [R1 , R2 ]; considered all x0 = r0 ω ∈ A, for any ω ∈ S N −1 . Then Z Z R2 f (y)dy N N −1 A = g(r0 ) − g(s)s ds E(x0 ) = f (x0 ) − N − RN V ol(A) R R 2 1 1 ≤
=
N R2N − R1N
N R2N − R1N
Z
(kLn+1 gk∞,[R1 ,R2 ] )
(kθn+1 f k∞,A )
Z
R2
Nn (r0 , s)s
ds
R1
R2 R1
N −1
Nn (|x0 |, s)s
N −1
!
!
ds .
(9.19)
Interesting cases also arise when r0 = R1 or R2 . We continue Remark 9.5 with Remark 9.8. Let now f ∈ C n+1 (A), n ≥ 0, x ∈ A, x = rω, r > 0. Clearly for fixed ω ∈ S N −1 , the function f (rω), r ∈ [R1 , R2 ] is radial, it also belongs to C n+1 ([R1 , R2 ]). By applying internal inequality (9.18) we obtain Z R2 N N −1 f (sω)s ds f (rω) − R2N − R1N R1 ≤
N R2N − R1N
"X n i=1
Z R2 N −1 gi (s, r)s ds |(Li f (·ω))(r)| R1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Balls and Shells Via Taylor–WidderFormula
+(kLn+1 f (·ω)k∞,[R1 ,R2 ] )
Z
R2
Nn (r, s)s
N −1
R1
#
ds .
131
(9.20)
For non-radial f we define again all i = 1, . . . , n + 1, ∀ x ∈ A.
θi f (x) = θi f (rω) := (Li f (·ω))(r),
(9.21)
Here the involved ∂ i f (x) ∂ i f (rω) = , i = 1, . . . , n + 1, i ∂r ∂ri are the radial derivatives. In a sense θi is a generalized radial derivative of Widdertype. Hence "X n N R.H.S.(9.20) ≤ |θi f (rω)| R2N − R1N i=1 Z # Z R2 R2 N −1 N −1 Nn (|x|, s)s ds . gi (s, r)s ds + kθn+1 f k∞,A R1 R1
(9.22)
Therefore, by (9.20) and (9.22) we have N Γ Z Z 1 2 f (y)dy f (rω)dω − 2π N/2 V ol(A) A S N −1
≤
N R2N − R1N
"
Γ(N/2) 2π N/2
(
n Z X i=1
+kθn+1 f k∞,A
Z
) Z R2 gi (s, r)sN −1 ds |θi f (rω)|dω N −1 R1 S R2
R1
#
Nn (|x|, s)sN −1 ds .
(9.23)
We have established Theorem 9.9. Let u0 , u1 , . . . , un ∈ C n+1 ([R1 , R2 ]); W0 , W1 , . . . , Wn > 0 on [R1 , R2 ], 0 < R1 < R2 , n ≥ 0 integer, with u0 (r) = c > 0. Let f ∈ C n+1 (A), x ∈ A; x = rω, r ∈ [R1 , R2 ], ω ∈ S N −1 . Then Z N R Γ f (rω)dω f (y)dy N −1 S 2 ≤ f (x) − E(x) = f (x) − A N/2 V ol(A) 2π N ( ) Γ Z R2 n Z X N 2 N −1 + |θ f (rω)|dω g (s, r)s ds i i R1 R2N − R1N 2π N/2 N −1 S i=1
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
132
Z
+kθn+1 f k∞,A
R2
Nn (|x|, s)s
N −1
R1
#
ds , ∀ x ∈ A.
(9.24)
We give Corollary 9.10. Same terms and assumptions as in Theorem 9.9. Suppose that θi f (r0 ω) = 0, i = 1, . . . , n, for a fixed r0 ∈ [R1 , R2 ], ∀ ω ∈ S N −1 ; also consider all x0 = r0 ω ∈ A for any ω ∈ S N −1 . Then Z Z N f (r ω)dω ) Γ f (y)dy 0 2 N −1 S A E(x0 ) = f (x0 ) − ≤ f (x0 ) − N/2 V ol(A) 2π +
N N R2 − R1N
kθn+1 f k∞,A
Z
R2
R1
Nn (|x0 |, s)s
N −1
!
ds .
(9.25)
When r0 = R1 or R2 is of special interest. 9.4
Results on the Sphere
Notation 9.11. Let N ≥ 1, B(0, R) := {x ∈ RN : |x| < R} be the ball in RN ωN R N . centered at the origin and of radius R > 0. Note that V ol(B(0, R)) = N Let f from B(0, R) into R and consider f be radial, i.e. f (x) = g(r), where r = |x|, 0 ≤ r ≤ R, we assume g ∈ C n+1 ([0, R]), n ≥ 0. Clearly then f ∈ C(B(0, R)). For F ∈ C(B(0, R)) we have ! Z Z Z R
F (x)dx =
S N −1
B(0,R)
F (rω)rN −1 dr
dω.
(9.26)
0
We notice that N RN
Z
R
sN −1 ds = 1.
(9.27)
0
The operator θi in the radial case, is defined as in (9.3), now ∀ x ∈ B(0, R). In the non-radial case, for f ∈ C n+1 (B(0, R)), θi is defined as in (9.21), ∀ x ∈ B(0, R)−{0}. We make Remark 9.12. Let here u0 , u1 , . . . , un ∈ C n+1 ([0, R]) and W0 , W1 , . . . , Wn > 0 on [0, R], R > 0, n ≥ 0 integer, with u0 (r) = c > 0.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Balls and Shells Via Taylor–WidderFormula
133
We again first suppose that f is radial on B(0, R), i.e. there exists g such that f (x) = g(r), r = |x|, 0 ≤ r ≤ R. Assume further that g ∈ C n+1 ([0, R]). Let x ∈ B(0, R). Then by using the polar method (9.26) we obtain ! Z R Z Z n−1 N g(s)s ds dω f (y)dy 0 S N −1 B(0,R) = g(r) − E(x) := f (x) − N V ol(B(0, R)) ωN R (9.28) Z R (by 9.27) N Z R N N −1 (g(r) − g(s))s ds = g(r) − N g(s)sn−1 ds = =: (∗). R RN 0 0 (9.29) Let s, r ∈ [0, R], then by generalized Taylor’s formula (9.2) we have n X
g(s) − g(r) = where
Rn (r, s) := But it holds
Z |Rn (r, s)| ≤
By calling
Z Nn (r, s) :=
we find
Li g(r)gi (s, r) + Rn (r, s),
(9.30)
i=1
s r
r
s
Z
s
gn (s, t)Ln+1 g(t)dt.
(9.31)
r
|gn (s, t)|dt kLn+1 gk∞,[0,R] .
|gn (s, t)|dt , ∀ s, r ∈ [0, R],
|Rn (r, s)| ≤ Nn (r, s)kLn+1 gk∞,[0,R] , ∀ s, r ∈ [0, R]. Therefore by (9.29) and (9.30), we have Z " n # N R X (∗) = N Li g(r)gi (s, r) + Rn (r, s) sN −1 ds R 0
(9.32)
(9.33)
(9.34)
i=1
≤
N RN
"
Z # n Z R R X N −1 N −1 Li g(r)gi (s, r)s ds + |Rn (r, s)|s ds 0 0
(9.35)
i=1
by
(9.34)
≤
Z " n R N X N −1 |L g(r)| g (s, r)s ds i i 0 RN i=1
+(kLn+1 gk∞,[0,R] )
Z
R
Nn (r, s)s
0
N −1
#
ds .
(9.36)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
134
We have established the next result. Theorem 9.13. Let u0 , u1 , . . . , un ∈ C n+1 ([0, R]), W0 , W1 , . . . , Wn > 0 on [0, R], R > 0, n ≥ 0 integer, with u0 (r) = c > 0. Let f from B(0, R) into R be radial, i.e. there exists g such that f (x) = g(r), r = |x|, 0 ≤ r ≤ R, ∀ x ∈ B(0, R); further suppose that g ∈ C n+1 ([0, R]). Then Z f (y)dy Z R N B(0,R) n−1 = g(r) − g(s)s ds E(x) := f (x) − N V ol(B(0, R)) R 0 Z # " n Z R R N X N −1 N −1 |Li g(r)| gi (s, r)s ds + (kLn+1 gk∞,[0,R] ) Nn (r, s)s ds ≤ N 0 R 0 i=1 Z " n R N X N −1 |θi f (x)| = N gi (s, |x|)s ds 0 R i=1
+(kθn+1 f k∞,B(0,R) )
Z
R
Nn (|x|, s)s
N −1
#
ds .
0
(9.37)
We give Corollary 9.14. Same terms and assumptions as in Theorem 9.13. Assume further that Li g(r0 ) = 0, i = 1, . . . , n, for a fixed r0 ∈ [0, R]; consider all x0 = r0 ω ∈ B(0, R), any ω ∈ S N −1 . Then Z f (y)dy Z R N B(0,R) n−1 E(x0 ) := f (x0 ) − = g(r ) − g(s)s ds 0 V ol(B(0, R)) RN 0 N ≤ N (kLn+1 gk∞,[0,R] ) R
Z
N = N (kθn+1 f k∞,B(0,R) ) R
Z
R
Nn (r0 , s)s
ds
0 R
0
Nn (|x0 |, s)s
Interesting cases especially arise when r0 = 0 or R. We continue Remark 9.12 with Remark 9.15. Let f be non-radial.
N −1
N −1
!
!
ds .
(9.38)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Balls and Shells Via Taylor–WidderFormula
135
Here assume that f ∈ C n+1 (B(0, R)). Consider x ∈ B(0, R) − {0}, i.e. uniquely x = rω, r ∈ (0, R], ω ∈ S N −1 . Then by using again the polar method (9.26) we get Z Z f (y)dy S N −1 f (rω)dω B(0,R) (9.39) − ω V ol(B(0, R)) N Z Z N N −1 f (rω)dω S N −1 = S − ωN by
(9.27)
=
N ωN R N
N − ωN R N N = ωN R N
Z S N −1
Z
Z
S N −1
Z
S N −1
Z
R 0
Z
Z
R
f (sω)s
0
ωN R N
f (rω)s
N −1
f (sω)s
ds dω
ds dω
0
N −1
!
!
R
R 0
N −1
(9.40)
!
!
! ds dω
(f (rω) − f (sω))sN −1 ds dω =: (∗). !
(9.41)
Clearly here f (·ω) ∈ C n+1 ((0, R]). Let ρ ∈ (0, R], then by (9.2) we obtain f (ρω) − f (rω) = where
n X
Rn (r, ρ) := That is
Z
(9.42)
ρ
gn (ρ, t)(Ln+1 (f (·ω)))(t)dt.
(9.43)
r
f (ρω) − f (rω) = with Rn (r, ρ) =
n X (θi f (rω))gi (ρ, r) + Rn (r, ρ),
(9.44)
i=1
Z
ρ
gn (ρ, t)θn+1 f (tω)dt.
(9.45)
θ := kθn+1 f k∞,B(0,R)−{0} < +∞.
(9.46)
We further assume that
Therefore we derive
((Li (f (·ω)))(r))gi (ρ, r) + Rn (r, ρ),
i=1
r
Z |Rn (r, ρ)| ≤
ρ r
|gn (ρ, t)|dt θ.
(9.47)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
136
By calling
Z Nn (r, ρ) :=
we get
|gn (ρ, t)|dt , ∀ ρ, r ∈ [0, R],
ρ r
(9.48)
|Rn (r, ρ)| ≤ Nn (r, ρ)θ.
(9.49)
Hence by (9.44) we obtain |f (ρω) − f (rω)| ≤ ≤
n X i=1
n X
|θi f (rω)||gi (ρ, r)| + |Rn (r, ρ)|
i=1
|θi f (rω)||gi (ρ, r)| + Nn (r, ρ)θ.
(9.50)
By continuity of f and gi , i = 1, . . . , n,and by taking the limit as ρ −→ 0 in the external inequality (9.50), we find |f (0) − f (rω)| ≤
n X
|θi f (rω)||gi (0, r)| + Nn (r, 0)θ.
i=1
(9.51)
Notice here gn (ρ, t) is jointly continuous in (ρ, t) ∈ [0, R]2 , hence Nn (r, ρ) is continuous in ρ ∈ [0, R]. That is, ∀ s ∈ [0, R] we get |f (sω) − f (rω)| ≤
n X i=1
|θi f (rω)||gi (s, r)| + Nn (r, s)θ.
Consequently by (9.41) and (9.52) we find ! Z R Z N N −1 |f (sω) − f (rω)|s ds dω (∗) ≤ ωN RN S N −1 0 " n Z X N ≤ N ωN R S N −1 i=1 +θ
=
Z n X i=1
Z
Z
S N −1
|θi f (rω)||gi (s, r)|s
0
Z
S N −1
Z
R
R
!
Nn (r, s)sN −1 ds dω
0
R
|θi f (rω)||gi (s, r)|s
0
N −1
N −1
!
ds dω
!
(9.52)
(9.53)
(9.54)
# !
ds dω
!
V ol(B(0, R)) θN +
Z
R
Nn (r, s)sN −1 ds
0
RN
!
(9.55)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Ostrowski Inequalities on Balls and Shells Via Taylor–WidderFormula
Z
N X = i=1
S N −1
|θi f (rω)|dω
R
|gi (s, r)|s
0
N −1
V ol(B(0, R))
θN + Consequently,
Z
Z
R
Nn (r, s)s
N −1
ds
0
!
RN
n X ≤ i=1
S N −1
|θi f (rω)|dω
Z
R
|gi (s, r)|s
0
V ol(V (0, R))
θN +
Z
R
Nn (r, s)sN −1 ds 0
RN
!
!
ds
.
(9.56)
Z Z f (y)dy S N −1 f (rω)dω B(0,R) ∆ := − ω V ol(B(0, R)) N
Z
137
N −1
!
ds
.
(9.57)
We have proved Theorem 9.16. Let u0 , u1 , . . . , un ∈ C n+1 ([0, R]), W0 , W1 , . . . , Wn > 0 on [0, R], R > 0, n ≥ 0 integer, with u0 (r) = c > 0. Let f ∈ C n+1 (B(0, R)). Suppose that θ := kθn+1 f k∞,B(0,R)−{0} < +∞.
(9.58)
Let x ∈ B(0, R) − {0}, i.e. uniquely x = rω, r ∈ (0, R], ω ∈ S N −1 . Then Z f (y)dy Z B(0,R) 0 0 ≤ f (x) − Γ(N/2) E(x) := f (x) − f (rω )dω N/2 V ol(B(0, R)) 2π S N −1 Z
n X + i=1
S N −1
Z
R
!
|gi (s, r)|sN −1 ds V ol(B(0, R))
|θi f (rω 0 )dω 0
0
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
138
θN +
Z
R
Nn (r, s)s 0
N −1
ds
!
RN
.
(9.59)
We finally give Corollary 9.17. Same terms and assumptions as in Theorem 9.16. Assume further that θi f (r0 ω 0 ) = 0, ∀ ω 0 ∈ S N −1 , for some r0 ∈ (0, R], for all i = 1, . . . , n. Consider all x0 = r0 ω ∈ B(0, R) − {0}, any ω ∈ S N −1 . Then Z f (y)dy Z Γ(N/2) B(0,r) 0 0 E(x0 ) := f (x0 ) − f (|x0 |ω )dω ≤ f (x0 ) − 2π N/2 V ol(B(0, R)) N −1 S ! Z R N −1 θN Nn (|x0 |, s)s ds +
0
RN
.
(9.60)
An interesting case is when r0 = R.
9.5
Addendum
We give Proposition 9.18. Let f ∈ C 1 (B(0, R)) such that f is radial, i.e. f (x) = g(r), r = |x|, 0 ≤ r ≤ R, ∀ x ∈ B(0, R). Then (i) ∃ g 0 ∈ C((0, R]), (9.61) 0 (ii) ∃ g (0) = 0. (9.62) Proof. (i) is obvious. (ii) Let u be a unit vector, h > 0, then f (hu) − f (0) g(h) − g(0) = h h
∇f (0) · hu + o(h) h o(h) ∇f (0) · hu o(h) + = ∇f (0) · u + = h h h o(h) h→0 −→ ∇f (0) · u, by → 0. h The last is true for any unit vector u. Thus ∇f (0) = 0, proving the claim. (by first order multivariate Taylor’s formula) =
Comment 9.19. By Proposition 9.18 we see that, it may well be that g 0 is discontinuous at zero, if only f ∈ C 1 (B(0, R)). Therefore the assumption that g ∈ C n+1 ([0, R]) in Theorem 9.13 seems to be the best.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Chapter 10
Multivariate Opial Type Inequalities for Functions Vanishing at an Interior Point
In this chapter we generalize Opial inequalities in the multivariate case on the balls. The inequalities carry weights and are proved to be sharp. The functions we study vanish at the center of the ball. This treatment relies on [58]. 10.1
Introduction
Z. Opial [195] and C. Olech [194] in 1960 proved the following famous inequality. Theorem 10.1. Let c > 0, and y(x) be real, continuously differentiable on [0, c], with y(0) = y(c) = 0. Then Z Z c c c 0 (y (x))2 dx. |y(x)y 0 (x)| dx ≤ 4 0 0 c Equality holds for the function y(x) = x on 0, 2 , and y(x) = c − x on 2c , c . In 1962 P. Beesack [75] gave the following improvement.
Theorem 10.2. Let b > 0. If y(x) is real, continuously differentiable on [0, b], and y(0) = 0, then Z b Z b b 0 |y(x)y 0 (x)| dx ≤ (y (x))2 dx. 2 0 0 Equality holds only for y(x) = mx, where m is a constant. Since then many people have worked on this type of inequalities in many directions; for an account see the important monograph of 1995 by R. Agarwal and P. Pang [6]. One motivation for this chapter is the interesting article of W. Troy [237] of 2001. His relevant result follows. Theorem 10.3. Let p > −1. Let a, b ∈ R with 0 ≤ a < b. If y(x) is continuously differentiable on [a, b], and y(a) = 0, then Z b Z b 1 tp |y(t)y 0 (t)| dt ≤ √ (bp+1 − atp )(y 0 (t))2 dt. 2 p + 1 a a 139
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
140
Let |·| denote the Euclidean norm in RN , N ≥ 1. Let B(0, R) := {x ∈ RN : |x| < R} be the open ball of radius R with center 0 in RN . Let S N −1 := {x ∈ RN : |x| = 1} N/2 be the unit sphere in RN , centered at zero. Let ωN = NNπ N (see p. 220, [142]) 2
Γ
2
be its surface area. In this chapter we estimate the integral Z |x|p |u(x)| |∇u(x)| dx, I= B(0,R)
for p ∈ R and u ∈ C 1 (B(0, R)). Using polar coordinates we obtain ! Z Z R p N −1 I= r |u(rω)| |∇u(rω)|r dr dω, S N −1
0
where 0 6= x = rω with r := |x| and ω := xr . For radial (spherically symmetric) functions u, this reduces to Z R ∂u p+N −1 r |u(r)| (r) dr, I = ωN ∂r 0 ∂u ∂u(x) x where here |∇u| = ∂r , with ∂r = ∇u(x) · |x| the radial derivative of u. Spherically symmetric function means that u(x) = u(rω) = u(r).
In general one has ∂u(x) ∂r ≤ |∇u(x)|, We will prove that Z
B(0,R)
for any u ∈ C 1 (B(0, R)).
|x|p |u(x)| |∇u(x)| dx ≤ C
Z
B(0,R)
|x|p |∇u(x)|2 dx
for some constant C and functions vanishing at the origin. The idea is to use ! Z R Z Z N −1 F (rω)r dr dω, F (x) dx = B(0,R)
S N −1
0
and to do first a one-dimensional analysis on the inner integral with ω fixed. The interior constraint (u(0) = 0) becomes a boundary condition (F (0) = 0).
10.2
Main Results
We present a basic result. Theorem 10.4. Let R > 0, N ≥ 1, B(0, R) the ball centered at 0 of radius R in RN . Let p ∈ R such that p + N ∈ (0, 2). Consider u ∈ C 1 (B(0, R)) such that u(0) = 0. Then ! Z Z R p 2 p √ |x| |∇u(x)| dx , |x| |u(x)| |∇u(x)| dx ≤ 2 2−p−N B(0,R) B(0,R) (10.1)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multivariate Opial Type Inequalities for Functions Vanishing at an Interior Point
141
where ∇ is the gradient operator. Proposition 10.5. Inequality (10.1) is sharp, more precisely it is attained by u(x) = |x|, for all x ∈ B(0, R), when p + N = 1. Proof.
Call r := |x|, 0 ≤ r ≤ R. Clearly u(0) = 0. Notice that ∂u = 1 and |∇u(x)| = 1. ∂r
We observe that L.H.S.(10.1) = ωN
Rp+N +1 , p+N +1
and R.H.S(10.1) = ωN Thus
Rp+N +1 p . 2(p + N ) 2 − (p + N )
L.H.S.(10.1) = R.H.S.(10.1) iff (p + N ) + 1 = 2(p + N ) (calling y := p + N , y ∈ (0, 2)) y + 1 = 2y
p 2 − (p + N ) iff
p 2−y
iff
g(y) := 4y 3 − 7y 2 + 2y + 1 = 0,
y ∈ (0, 2).
See that g(1) = 0, g(0) = 1, g(2) = 9, and 1 g 0 (y) = 12(y − 1) y − . 6
Thus g has critical numbers 1, 61 with local maximum g 16 = 1.1574078, and minimum g(1) = 0. So y = p + N = 1 is the only optimal value making inequality (10.1) attained. Proof of Theorem 10.4. The integral in the R.H.S(10.1) is finite since 2 > p + N > 0. We can rewrite inequality (10.1) by the use of polar coordinates as follows, cf. p. 217, [142]: ! Z Z R p N −1 r |u(rω)| |∇u(rω)|r dr dω S N −1
0
R ≤ √ 2 2−p−N
Z
S N −1
Z
R 0
p
2 N −1
r |∇u(rω)| r
dr
!
!
dω .
(10.2)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
142
Here 0 = 6 x ∈ B(0, R) is written as x := rω with r := |x|, 0 < r ≤ R, and x ω = |x| ∈ S N −1 . So it is enough to prove that ! Z R Z R R p+N −1 2 p+N −1 √ r |∇u(rω)| dr . r |u(rω)| |∇u(rω)| dr ≤ 2 2−p−N 0 0 (10.3) We set Z r sp+N −1 |∇u(sω)|2 ds, 0 ≤ r ≤ R. z(r) := 0
Here z(r) ≥ 0 and z(0) = 0. Therefore z 0 (r) = rp+N −1 |∇u(rω)|2 ≥ 0,
0 < r ≤ R.
Whence r
p+N −1 2
|∇u(rω)| = (z 0 (r))1/2 ,
0 < r ≤ R.
(10.4)
By the fundamental theorem of calculus we have Z r ∂u(sω) ds. u(rω) = ∂s 0
Consequently it holds Z r Z r p+N −1 ∂u −1 ∂u − p+N 2 2 s |u(rω)| ≤ s ∂s (sω) ds = ∂s (sω) ds 0 0 (by the Cauchy–Schwarz inequality) 2 !1/2 Z r 1/2 Z r −(p+N −1) (p+N −1) ∂u ≤ s ds s ∂s (sω) ds 0 0 2−p−N 1/2 Z r 1/2 r ≤ s(p+N −1) |∇u(sω)|2 ds 2−p−N 0 2−p−N 1/2 r = (z(r))1/2 . 2−p−N So we have proved that |u(rω)| ≤ Furthermore, r
p+N −1 2
r2−p−N 2−p−N
|u(rω)| ≤ √
1/2 √
(z(r))1/2 ,
r (z(r))1/2 , 2−p−N
all 0 ≤ r ≤ R.
all 0 ≤ r ≤ R.
(10.5)
Consequently by (10.4) and (10.5) we derive √ r r(p+N −1) |u(rω)| |∇u(rω)| ≤ √ (z(r))1/2 (z 0 (r))1/2 , all 0 < r ≤ R. (10.6) 2−p−N
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multivariate Opial Type Inequalities for Functions Vanishing at an Interior Point
143
Next we integrate (10.6) and use the Cauchy–Schwarz inequality to get Z R Z R √ 1 r(z(r))1/2 (z 0 (r))1/2 dr rp+N −1 |u(rω)| |∇u(rω)| dr ≤ √ 2 − p − N 0 0 !1/2 Z R R z(r) dz(r) ≤ √ √ 2 2−p−N 0 ! Z R R R p+N −1 2 = √ z(R) = √ r |∇u(rω)| dr , 2 2−p−N 2 2−p−N 0 establishing (10.3).
A similar result follows. Theorem 10.6. Let R > 0, N ≥ 1, B(0, R) the ball centered at 0 of radius R in RN . Let p ∈ R such that p + N > 0. Consider u ∈ C 1 (B(0, R)) such that u(0) = 0. Then Z Z Rp+N p √ |x|1−N |∇u(x)|2 dx. (10.7) |x| |u(x)| |∇u(x)| dx ≤ 2 p + N B(0,R) B(0,R) Proposition 10.7. Inequality (10.7) is sharp, namely it is attained by u(x) = |x|, for all x ∈ B(0, R), when p + N = 1. Proof. Call r := |x|, 0 ≤ r ≤ R. Clearly u(0) = 0. Note that |∇u(x)| = 1. We have that L.H.S.(10.7) = ωN
∂u ∂r
= 1 and
Rp+N +1 , p+N +1
and Rp+N +1 R.H.S.(10.7) = ωN √ . 2 p+N Thus L.H.S.(10.7) = R.H.S.(10.7) iff p (p + N ) + 1 = 2 p + N iff √ y + 1 = 2 y, where y := p + N > 0, y = 1.
iff
Proof of Theorem 10.6. Clearly the R.H.S.(10.7) is finite. As in the proof of Theorem 10.4 it is enough to prove ! Z R Z R Rp+N p N −1 2 r |u(rω)| |∇u(rω)|r dr ≤ √ |∇u(rω)| dr . (10.8) 2 p+N 0 0
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
144
Therefore we begin with Z
R 0
p
Z
N −1
R
r |u(rω)| |∇u(rω)|r dr = rp+N −1 |u(rω)| |∇u(rω)| dr 0 Z R p+N −2 p+N 2 2 = |∇u(rω)| r |u(rω)| dr r 0
Z
≤
0
(by the Cauchy–Schwarz inequality) !1/2 Z !1/2 R R p+N 2 (p+N −1) −1 2 r |∇u(rω)| dr =: (∗). r r (u(rω)) dr 0
We have again u(rω) =
Z
r 0
∂u (sω) ds, ∂s
0 ≤ r ≤ R.
Hence ∂u 1 · (sω) ds ∂s 0 (by the Cauchy–Schwarz inequality) 2 !1/2 1/2 Z r Z r ∂u √ √ 2 (sω) ds ≤ r |∇u(sω)| ds r . ≤ ∂s 0 0
|u(rω)| ≤
That is
Z
r
2
(u(rω)) ≤ r
Z
r 0
|∇u(sω)|2 ds.
(10.9)
Thus by (10.9) we derive Z
(∗) ≤
Z
= −
R
r
p+N
0 R
r
|∇u(rω)| dr
1 p+N
!1/2
·
Z
R
r
(p+N −1)
0
Z
r 0
2
|∇u(sω)| ds dr
!1/2 ! Z R Rp+N 2 · |∇u(rω)| dr |∇u(rω)| dr p+N 0 1/2 Z R p+N 2 =: (∗∗). r |∇u(rω)| dr
p+N
0
2
!1/2
2
0
If A ≥ 0, B ≥ 0 and ε > 0 we have (AB)1/2 ≤
ε 1 A + B. 2 2ε
(10.10)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multivariate Opial Type Inequalities for Functions Vanishing at an Interior Point
Therefore
145
! Z R 1 Rp+N 2 r |∇u(rω)| dr + |∇u(rω)| dr 2ε p + N 0 0 ! Z R 1 p+N 2 − r |∇u(rω)| dr p+N 0 ! Z R 1 1 p+N 2 r |∇u(rω)| dr = ε− 2 ε(p + N ) 0 ! Z R Rp+N 2 + |∇u(rω)| dr 2ε(p + N ) 0 ! Z R 1 Rp+N choosing ε = √ = √ |∇u(rω)|2 dr . p+N 2 p+N 0
ε (∗∗) ≤ 2
Z
R
p+N
!
2
We have established (10.8).
Finally we present a generalization and extension of Theorem 10.4. Theorem 10.8. Let R > 0, N ≥ 1, ρ > 1, α, β > 0, ρ ≥ α + β and p ∈ R such that 0 < p + N < ρ. Consider u ∈ C 1 (B(0, R)) such that u(0) = 0. Then Z α+β |x| (p+N −1) ρ +(1−N ) |u(x)|β |∇u(x)|α dx B(0,R)
ρ−α−β ρ
≤ LωN where L :=
ρ−1 ρ−p−N
β
ρ−1 ρ
Z
B(0,R)
ρ−a β(ρ − 1) + (ρ − α)
|x|p |∇u(x)|ρ dx
ρ−α ρ
α α+β
!
αρ
α+β ρ
R
,
(10.11)
β(ρ−1)+(ρ−α) ρ
.
(10.12)
Proposition 10.9. Inequality (10.11) is sharp, namely it is attained by u(x) = |x|, α for all x ∈ B(0, R), when p + N = 1, ρ = α + β and α α+β = α+β 1+β . Proof. Call r := |x|, 0 ≤ r ≤ R. Clearly u(0) = 0. Note that |∇u(x)| = 1. We have L.H.S.(10.11) = ωN
∂u ∂r
= 1 and
Rβ+1 , β+1
and R.H.S.(10.11) = ωN
αα/ρ β+1 R . ρ
It is obvious now that (10.11) holds as equality.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
146
Proof of Theorem 10.8. The integral in the R.H.S.(10.11) is finite by p + N > 0. x Here for 0 6= x ∈ B(0, R) we set ω := |x| = xr , where r := |x|, 0 < r ≤ R. That is x = rω. By the fundamental theorem of calculus we have Z r ∂u (sω) ds, 0 ≤ r ≤ R. u(rω) = 0 ∂s We see that
Z r p+N −1 ∂u −1 − p+N (sω) ds = ∂u (sω) ds ρ ρ s s ∂s ∂s 0 0 ρ (using H¨ older’s inequality with indices ρ and ρ−1 ) ρ−1 Z r ρ Z r 1/ρ ρ ∂u (p+N −1) ≤ s− ρ−1 ds s(p+N −1) (sω) ds ∂s 0 0 ρ−1 Z 1/ρ r ρ (ρ−p−N ) ρ−1 s(p+N −1) |∇u(sω)|ρ ds ≤ r ρ . ρ−p−N 0
|u(rω)| ≤
That is,
Z
r
ρ−1 ρ−p−N
+∞ > z(r) :=
Z
|u(rω)| ≤
Here we call
ρ−1 ρ
r 0
ρ−p−N ρ
r
z(r)1/ρ ,
0 ≤ r ≤ R.
s(p+N −1) |∇u(sω)|ρ ds ≥ 0,
with z(0) = 0. We have z 0 (r) = rp+N −1 |∇u(rω)|ρ ≥ 0, and p+N −1 ρ
r We observe that
|u(rω)|β ≤ and r
p+N −1 ρ
β
β
|u(rω)| ≤
α
0 < r ≤ R.
|∇u(rω)|α = (z 0 (r))α/ρ ,
ρ−1 ρ−p−N
ρ−1 ρ−p−N
β
β ρ−1 ρ
ρ−1 ρ
·r
β
rβ
0 < r ≤ R.
(ρ−p−N ) ρ
ρ−1 ρ
(10.13)
(z(r))β/ρ ,
z(r)β/ρ ,
By multiplying (10.13) and (10.14) we get α+β r(p+N −1) ρ |u(rω)|β |∇u(rω)|α β ρ−1 ρ ρ−1 ρ−1 ≤ rβ ρ (z(r))β/ρ (z 0 (r))α/ρ , ρ−p−N
0 ≤ r ≤ R. (10.14)
0 < r ≤ R. (10.15)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Multivariate Opial Type Inequalities for Functions Vanishing at an Interior Point
147
ρ Integrate (10.15) and use H¨ older’s inequality with indices αρ , ρ−α to find Z R α+β r(p+N −1) ρ |u(rω)|β |∇u(rω)|α dr 0 β ρ−1 Z R ρ ρ−1 β ρ−1 ρ ≤ (z(r))β/ρ (z 0 (r))α/ρ dr r ρ−p−N 0 ! ρ−α !α/ρ β ρ−1 Z R Z R ρ ρ ρ−1 ρ−1 β ρ−α β/α 0 ≤ dr · r (z(r)) z (r) dr ρ−p−N 0 0 ! α+β Z R ρ α+β (10.12) . rp+N −1 |∇u(rω)|ρ dr = L · z(R) ρ = L 0
We have established Z R r (p+N −1)
α+β ρ
0
Z
≤ L
+(1−N )
α N −1
|u(rω)| |∇u(rω)| r
R
rp |∇u(rω)|ρ rN −1 dr
0
β
!
α+β ρ
dr
!
.
(10.16)
Integrating (10.16) over S N −1 we find ! Z R Z (p+N −1) α+β +(1−N ) β α N −1 ρ r |u(rω)| |∇u(rω)| r dr dω S N −1
≤ L·
0
Z
S N −1
Z
R 0
rp |∇u(rω)|ρ rN −1 dr
!
α+β ρ
dω =: (∗ ∗ ∗).
If ρ > α + β then apply again H¨ older’s inequality with indices to obtain (∗ ∗ ∗) ≤L
Z
ρ
1 ρ−α−β dω S N −1
ρ−α−β
= LωN
ρ
·
Z
S N −1
ρ−α−β ρ
Z
R 0
·
Z
S N −1
Z
R 0
rp |∇u(rω)|ρ rN −1 dr
From (10.17) and (10.18) we conclude (10.11).
ρ ρ−α−β
rp |∇u(rω)|ρ rN −1 dr !
dω
!
α+β ρ
(10.17)
.
!
ρ α+β
abd
dω
!
α+β ρ
(10.18)
The work of Neˇceav [193] in 1973 is related to this chapter, see also [6, p. 275]. Theorem 10.10 [193]. Let u ∈ C 1 (B(0, R)) be such that u(0) = 0, and N + p < 2, N ≥ 1. Then Z Z R2−N −p 1−N |x| |u(x)| |∇u(x)| dx ≤ |x|p |∇u(x)|2 dx, 2(2 − N − p) B(0,R) B(0,R)
17th August 2010
15:56
148
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
with equality holding when u(x) = c|x|2−N −p for a real constant c. This result can be compared to Theorem 10.6 if and only if p = 1 − N , in which case the conclusions are the same, with the common constant being R/2.
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Chapter 11
General Multivariate Weighted Opial Inequalities
In this chapter we expose Opial type weighted multivariate inequalities on balls and arbitrary smooth bounded domains. The inequalities are mostly sharp. The functions we consider vanish on the boundary. This chapter relies on [59]. 11.1
Introduction
Z. Opial [195] and C. Olech [194] in 1960 proved the following well-known inequality. Theorem 11.1. Let c > 0, and y(x) be real, continuously differentiable on [0, c], with y(0) = y(c) = 0. Then Z Z c c c 0 (y (x))2 dx. |y(x)y 0 (x)| dx ≤ 4 0 0 c Equality holds for the function y(x) = x on 0, 2 , and y(x) = c − x on 2c , c . In 1962 P. Beesack [75] gave the following variant.
Theorem 11.2. Let b > 0. If y(x) is real, continuously differentiable on [0, b], and y(0) = 0, then Z b Z b b 0 |y(x)y 0 (x)| dx ≤ (y (x))2 dx. 2 0 0 Equality holds only for y = mx, where m is a constant. Since then many researchers have been extending this type of inequalities in many directions, for a thorough account see the important monograph of 1995 by R. Agarwal and P. Pang [6]. An inspiration for this chapter is the interesting article by W. Troy [237] of 2001. His relevant result is as follows. Theorem 11.3. Let p > −1. Let b and c be real with 0 ≤ b < c. If y(x) is continuously differentiable on [b, c], and y(c) = 0, then Z c Z c 1 tp |y(t)y 0 (t)| dt ≤ √ (ctp − bp+1) (y 0 (t))2 dt. 2 p + 1 b b 149
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
150
R. Agarwal in 1981, see [4] and p. 208 of [6], proved the following two dimensional result. Theorem 11.4. If u(t, s) ∈ C (1,1) ([a, T ] × [c, S]), u(a, s) = u(t, c) = 0, then Z TZ S Z Z (T − a)(S − c) T S √ |uts (t, s)|2 dtds. |u(t, s)uts (t, s)| dtds ≤ 2 2 a a c c In 1982, G.S. Yang [247] proved the following Opial-type inequality in two variables: Theorem 11.5. If f (s, t), f1 (s, t), and f12 (s, t) are continuous functions on [a, b]× [c, d], and if f (a, t) = f (b, t) = f1 (s, c) = f1 (s, d) = 0, for a ≤ s ≤ b, c ≤ t ≤ d, then Z Z Z bZ d (b − a)(d − c) b d (f12 (s, t))2 dtds. |f (s, t)| |f12 (s, t)| dtds ≤ 8 a c a c In 1983, C.T. Lin and G.S. Yang [179] generalized Theorem 11.5 in the following form: Theorem 11.6. If f (s, t), f1 (s, t), and f12 (s, t) are continuous functions on [a, b]× [c, d], and if f (a, t) = f (b, t) = f1 (s, c) = f1 (s, d) = 0, for a ≤ s ≤ b, c ≤ t ≤ d, then Z bZ d |f (s, t)|m |f12 (s, t)|n dtds a
≤
n m+n
c
(b − a)(d − c) 4
m Z
b a
Z
d c
|f12 (s, t)|m+n dtds.
The above Theorems 11.4 – 11.6 motivate this chapter. Here we prove weighted Opial type inequalities over a ball (or other domain) in RN , N ≥ 1, where the functions under consideration vanish on the boundary. Let | · | denote the Euclidean distance for vectors in RN . Let B(0, R) := {x ∈ N R : |x| < R} be the open ball of radius R with center 0 in RN . Let S N −1 := {x ∈ RN : |x| = 1} be the unit sphere in RN centered at zero, with N/2 ωN = NNπ N to be its surface area (cf. [142], p. 220). 2
Γ
2
The basic results here are extended to Sobolev spaces of order 1. We mention
Definition 11.7 (cf. [139]). The Sobolev space of order 1 denoted by H01 (Ω) is the completion of Cc1 (Ω). Here Cc1 (Ω) is the space of one time continuously differentiable functions with compact support on the bounded domain Ω ⊂ RN , N ≥ 1. The norm in H01 (Ω) is given by Z 1/2 Z kuk∗ := |u(x)|2 dx + |∇u(x)|2 dx , for u ∈ H01 (Ω). Ω
Ω
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
General Multivariate Weighted Opial Inequalities
Book˙Adv˙Ineq
151
In this chapter we estimate the integral Z I= |x|p |u(x)| |∇u(x)| dx, B(0,R)
for p ∈ R and u ∈
Cc1 (B(0, R)).
I=
Z
S N −1
Using polar coordinates we have ! Z R p N −1 r |u(rω)| |∇u(rω)|r dr dω, 0
where x = rω with r := |x| and ω := xr . For radial (spherically symmetric) functions u, this reduces to Z R ∂u p+N −1 r |u(r)| (r) dr, I = ωN ∂r 0 ∂u x where here |∇u| = ∂r , with ∂u(x) ∂r = ∇u(x) · |x| the radial derivative of u. Spherically symmetric function means that u(x) = u(rω) = u(r).
In general one has ∂u(x) ∂r ≤ |∇u(x)|, We will prove that Z
B(0,R)
for any u ∈ Cc1 (B(0, R)).
|x|p |u(x)| |∇u(x)| dx ≤ C
Z
B(0,R)
|x|p |∇u(x)|2 dx
for some constant C and functions u as above. The idea is to use ! Z R Z Z F (rω)rN −1 dr dω, F (x) dx = B(0,R)
S N −1
0
and to do first a one-dimensional analysis on the inner integral with ω fixed. The boundary condition u = 0 on ∂B(0, R) becomes the boundary condition F (R) = 0. 11.2
Main Results
We present a basic result. Theorem 11.8. Let R > 0, N ≥ 1, B(0, R) the ball centered at 0 of radius R in R N . Let p ∈ R such that p + N ≥ 1. Let u ∈ C 1 (B(0, R)) be such that u(∂B(0, R)) = 0. Then Z Z R |x|p |u(x)| |∇u(x)| dx ≤ |x|p |∇u(x)|2 dx, (11.1) 2 B(0,R) B(0,R) where ∇ is the gradient operator.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
152
Note that Theorem 11.8 holds for more general choices of u. By a simple approximation argument, the condition u ∈ C 1 (B(0, R)) can be weakened to 1,2 B(0, R) − {0} ∩ C B(0, R) u ∈ Wloc
such that u = 0 on ∂B(0, R) and the right hand side of (11.1) is finite. The extremal in the following proposition satisfies this weaker hypothesis. Similar remarks apply to the Theorems 11.13 and 11.17 below. Proposition 11.9. Inequality (11.1) is sharp, namely it is attained by u(x) = R − |x|, for all x ∈ B(0, R), when p + N = 1. Proof. Call r := |x|, 0 ≤ r ≤ R. Notice u(∂B(0, R)) = 0. Since u(x) = R − r we find ∂u ∂r = −1 and |∇u(x)| = 1. Clearly we get L.H.S.(11.1) = ωN
Rp+N +1 , (p + N )(p + N + 1)
and R.H.S.(11.1) = ωN
Rp+N +1 , 2(p + N )
where ωN is the surface area of the unit sphere S N −1 in RN . Thus L.H.S.(11.1) = R.H.S.(11.1) iff p + N = 1. Proof of Theorem 11.8. The integral in the R.H.S.(11.1) is finite since p+N > 0. We can rewrite inequality (11.1) by the use of polar coordinates as follows ! Z Z R rp |u(rω)| |∇u(rω)|rN −1 dr dω S N −1
0
R ≤ 2
Z
S N −1
Z
R p
0
2 N −1
r |∇u(rω)| r
dr
!
dω.
(11.2)
Here x ∈ B(0, R) is given in polar coordinates by x = rω with r := |x|, 0 ≤ r ≤ R. x Clearly ω = |x| if r > 0. So it is enough to prove that, for each ω, Z R Z R R p p N −1 r |∇u(rω)|2 rN −1 dr. (11.3) r |u(rω)| |∇u(rω)|r dr ≤ 2 0 0
We set
z(r) :=
Z
R r
sp+N −1 |∇u(sω)|2 ds,
0 ≤ r ≤ R.
Here z(r) ≥ 0 and z(R) = 0. Then
−z 0 (r) = rp+N −1 |∇u(rω)|2 ≥ 0,
0 ≤ r ≤ R.
Thus r
p+N −1 2
|∇u(rω)| = (−z 0 (r))1/2 .
(11.4)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
General Multivariate Weighted Opial Inequalities
Since u = 0 on ∂B(0, R) we find u(rω) = − We observe that r
p+N −1 2
Z
R r
153
∂u (sω) ds. ∂s
∂u (sω) ds ∂s r Z R ∂u p+N −1 2 ≤ 1·s ∂s (sω) ds
|u(rω)| ≤
Z
R
p+N −1 2
r
r
(by the Cauchy–Schwarz inequality) 2 !1/2 Z R √ p+N −1 ∂u ≤ R−r s ∂s (sω) ds r !1/2 Z R √ √ ≤ R−r sp+N −1 |∇u(sω)|2 ds = R − r(z(r))1/2 . r
That is, we have
r
p+N −1 2
|u(rω)| ≤
√
R − r(z(r))1/2 ,
(11.5)
all 0 ≤ r ≤ R. Multiplying (11.4) by (11.5) we obtain √ rp+N −1 |u(rω)| |∇u(rω)| ≤ R − r(z(r))1/2 (−z 0 (r))1/2 ,
(11.6)
for all 0 ≤ r ≤ R. Therefore by integrating (11.6) and using the Cauchy–Schwarz inequality we get Z R rp+N −1 |u(rω)| |∇u(rω)| dr 0
≤ ≤
Z
R
√
0
Z
R − r(z(r))1/2 (−z 0 (r))1/2 dr
R 0
(R − r) dr
!1/2
−
Z
R
0
z(r)z (r) dr 0
Z R R R p+N −1 r |∇u(rω)|2 dr. = z(0) = 2 2 0 Clearly we have established (11.3), and (11.1) follows. We have
!1/2
Corollary 11.10. Let p ∈ R be such that p + N ≥ 1 and let u ∈ H01 (B(0, R)). Then Z Z R |x|p |u(x)| |∇u(x)| dx ≤ |x|p |∇u(x)|2 dx. (11.7) 2 B(0,R) B(0,R) Proof. By (11.1), inequality (11.7) is valid for u ∈ C 1 (B(0, R)) with u = 0 on ∂B(0, R). If u ∈ H01 (B(0, R)) then there exists a sequence {um } from C 1 (B(0, R)) k·k∗
such that um = 0 on ∂B(0, R) and um −→ u. Then the corollary follows by using (11.1) for um and letting m → ∞.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
154
Next we are ready to give Theorem 11.11. Let Ω be a smooth bounded domain in RN , N ≥ 1. Let p ∈ R such that p + N ≥ 1. Then there exists a constant K = K(Ω) such that Z Z p |x|p |∇u(x)|2 dx, (11.8) |x| |u(x)| |∇u(x)| dx ≤ K(Ω) Ω
Ω
for all u ∈
H01 (Ω).
Here
R , 2 where R is the radius of the smallest ball centered at the origin and containing Ω. K(Ω) :=
Proof. Since ∂Ω is smooth any function u in H01 (Ω) can be viewed as a member of H01 (B(0, R)) by defining it to be zero on B(0, R) − Ω. For the last step, see p. 245 of [139]. Now Corollary 11.10 applies. Remark 11.12. (i) When u is spherically symmetric then (11.7) reduces to Z R Z R R p+N −1 0 rp+N −1 |u(r)| |u0 (r)| dr ≤ r (u (r))2 dr. (11.9) 2 0 0 (ii) When p = 0 then (11.8) collapses into Z Z |u(x)| |∇u(x)| dx ≤ K(Ω) |∇u(x)|2 dx, Ω
(11.10)
Ω
for all u ∈ H01 (Ω). The counterpart of Theorem 11.8 follows
Theorem 11.13. Let R > 0, N ≥ 1, B(0, R) the ball centered at 0 of radius R in RN . Let p ∈ R be such that p + N > 1. Let u ∈ C 1 (B(0, R)) such that u(∂B(0, R)) = 0. Then √ Z Z R p+N p |x| |u(x)| |∇u(x)| dx ≤ |x|p |∇u(x)|2 dx. (11.11) 2(p + N − 1) B(0,R) B(0,R) Proof. The integral in the R.H.S.(11.11) is finite by p + N > 0. We estimate Z R Z R p+N −2 p+N rp+N −1 |u(rω)| |∇u(rω)|dr = r 2 |∇u(rω)| r 2 |u(rω)| dr 0
0
≤
Z
R
(by the Cauchy–Schwarz inequality) !1/2 Z R
r 0
p+N
2
·
|∇u(rω)| dr
r
p+N −2
0
Since u = 0 on ∂B(0, R) we find u(rω) = −
Z
R r
∂u (sω) ds. ∂s
2
|u(rω)| dr
!1/2
=: (∗∗).
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
General Multivariate Weighted Opial Inequalities
155
Therefore |u(rω)| ≤ ≤
Z
R r
√
∂u 1 · (sω) ds ≤ ∂s
R−r
Z
R
r
Z
R
2
1 ds r
|∇u(sω)|2 ds
!1/2
!1/2
Z
.
That is Z
2
|u(rω)| ≤ (R − r)
R
2
R r
2 !1/2 ∂u (sω) ds ∂s
!
|∇u(sω)| ds .
r
Consequently we have Z
(∗∗) ≤ ·
Z
Z
=
=
Z
r
p+N
0
2
|∇u(rω)| ds
R
r 0 R
0 R 0
R
p+N −2
(R − r)
rp+N |∇u(rω)|2 dr rp+N |∇u(rω)|2 dr
Z
!1/2
!
R 2
r
!1/2
!1/2
|∇u(sω)| ds dr · ·
Z
Z
R 0 R 0
Z
s 0
!1/2
rp+N −2 (R − r) dr |∇u(sω)|2 ds
sp+N Rsp+N −1 − p+N −1 p+N
|∇u(sω)|2 ds
!1/2
!1/2
ε 1 using the inequality (AB)1/2 ≤ A + B, for any A ≥ 0, B ≥ 0 and ε > 0 2 2ε ! " ! Z R Z R 1 ε R p+N 2 p+N −1 2 r |∇u(rω)| dr + r |∇u(rω)| dr ≤ 2 2ε p+N −1 0 0 !# Z R 1 − rp+N |∇u(rω)|2 dr p+N 0 ! Z R 1 p+N 2 r |∇u(rω)| dr ε− ε(p + N ) 0 ! Z R R p+N −1 2 r |∇u(rω)| dr + 2ε(p + N − 1) 0 1 by choosing ε = √ p+N ! √ Z R R p+N p+N −1 2 = r |∇u(rω)| dr . 2(p + N − 1) 0
1 = 2
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
156
We have proved that Z R rp+N −1 |u(rω)| |∇u(rω)| dr ≤ 0
Finally we obtain Z Z S N −1
R
r
p+N −1
0
√ R p+N 2(p + N − 1)
Z
!
dw
|u(rω)| |∇u(rω)| dr
√ Z R p+N ≤ 2(p + N − 1) S N −1
Z
R
r
p+N −1
0
2
|∇u(rω)| dr .
R
r
p+N −1
0
!
2
|∇u(rω)| dr
proving (11.11).
!
dω,
Corollary 11.14. Let p ∈ R be such that p + N > 1 and let u ∈ H01 (B(0, R)). Then √ Z Z R p+N |x|p |∇u(x)|2 dx. (11.12) |x|p |u(x)| |∇u(x)| dx ≤ 2(p + N − 1) B(0,R) B(0,R) Proof.
Similar to Corollary 11.10.
Next we are ready to give Theorem 11.15. Let Ω be a smooth bounded domain in RN , N ≥ 1. Let p ∈ R be such that p + N > 1. Then there exists a constant K = K(Ω, p, N ) such that Z Z |x|p |u(x)| |∇u(x)| dx ≤ K |x|p |∇u(x)|2 dx, (11.13) Ω
Ω
for all u ∈ H01 (Ω). Here
√ R p+N K := , 2(p + N − 1)
where R is the radius of the smallest ball centered at the origin and containing Ω. Proof.
Similar to Theorem 11.11 by applying Corollary 11.14.
Remark 11.16. (i) When u is spherically symmetric then (11.12) reduces to √ Z R Z R R p+N p+N −1 0 r |u(r)| |u (r)| dr ≤ rp+N −1 (u0 (r))2 dr. (11.14) 2(p + N − 1) 0 0 (ii) When p = 0 then (11.13) collapses into Z Z ˜ |∇u(x)|2 dx, |u(x)| |∇u(x)| dx ≤ K Ω
Ω
for all u ∈ H01 (Ω), where
˜ := K
√ R N , 2(N − 1)
Theorem 11.8 is generalized as follows.
N ≥ 2.
(11.15)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
General Multivariate Weighted Opial Inequalities
157
Theorem 11.17 Let R > 0, N ≥ 1, ρ > 1, α, β > 0, ρ ≥ α + β, p ≥ 1 − N , and u ∈ C 1 (B(0, R)) with u(∂B(0, R)) = 0. Then Z α+β |x| (p+N −1) ρ +(1−N ) |u(x)|β |∇u(x)|α dx B(0,R)
ρ−α−β ρ
≤ M ωN where M :=
ρ−α (ρ − 1)β + (ρ − α)
ρ−α ρ
Z
·
B(0,R)
|x|p |∇u(x)|ρ dx
α α+β
α/ρ
·R
!
α+β ρ
(ρ−1)β+(ρ−α) ρ
,
(11.16)
.
(11.17)
Proposition 11.18. Inequality (11.16) is sharp, namely it is attained by u(x) := R − |x|, for x ∈ B(0, R), when p + N = 1, ρ = α + β and α = 1. Proof.
Notice that L.H.S.(11.16) = ωN
Rβ+1 , β+1
and R.H.S.(11.16) = ωN
αα/ρ β+1 R . ρ
Equality holds in (11.16) if α
α α+β =
α+β . 1+β
But this implies α = 1. α Here is a proof that if α > 0 and β > 0, then α α+β =
α+β 1+β
implies that α = 1. α
Put = t and = s. Then 0 < t < 1 and s > 0. The condition α α+β = α+β 1+β yields the equation s − st = 1 − t. Put g(x) = xt − xt, for x > 0. Note that g 0 (x) = txt−1 − t = t(xt−1 − 1). It follows that g has a unique global maximum at x = 1. Since g(s) = g(1), we infer that s = 1, and hence α = 1s = 1. α α+β
1 α t
Proof of Theorem 11.17. The integral in the R.H.S.(11.16) is finite by p+N > 0. We have Z R ∂u (sω) ds, all 0 ≤ r ≤ R. u(rω) = − ∂s r ρ Next we apply H¨ older’s inequality with indices ρ and ρ−1 to derive Z R Z R p+N −1 p+N −1 ∂u(sω) p+N −1 ∂u ρ ρ ρ r |u(rω)| ≤ r 1·s ∂s ds ≤ ∂s (sω) ds r r ρ !1/ρ Z R Z R ! ρ−1 ρ p+N −1 ∂u s ≤ ds ∂s (sω) ds r r !1/ρ Z ≤ (R − r)
R
ρ−1 ρ
r
sp+N −1 |∇u(sω)|ρ ds
.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
158
So by calling
Z
z(r) := we got p+N −1 ρ
r and
r(p+N −1)
β ρ
R
sp+N −1 |∇u(sω)|ρ ds
r
|u(rω)| ≤ (R − r)
ρ−1 ρ
(11.18) 1
(z(r)) ρ ,
|u(rω)|β ≤ (R − r)(ρ−1)
β ρ
(11.19) β
(z(r)) ρ ,
(11.20)
all 0 ≤ r ≤ R. Here notice that z(r) ≥ 0 and z(R) = 0. Also we have Z r sp+N −1 |∇u(sω)|ρ ds −z(r) = R
and
−z 0 (r) = rp+N −1 |∇u(rω)|ρ ≥ 0.
Consequently
p+N −1 ρ
r and
1
|∇u(rω)| = (−z 0 (r)) ρ
α α (11.21) r(p+N −1) ρ |∇u(rω)|α = (−z 0 (r)) ρ , for all 0 ≤ r ≤ R. Multiplying (11.20) by (11.21) we get α+β β β α r(p+N −1) ρ |u(rω)|β |∇u(rω)|α ≤ (R − r)(ρ−1) ρ (z(r)) ρ (−z 0 (r)) ρ , (11.22) for all 0 ≤ r ≤ R. ρ Next we integrate (11.22) and apply H¨ older’s inequality with indices αρ and (ρ−α) to find Z R (p+N −1) α+β ρ r |u(rω)|β |∇u(rω)|α dr 0
≤
Z
R
0
≤
Z
=
Z
= =
(R − r)(ρ−1) R
(R − r)
0 R
(R − r)
0
β ρ
(ρ−1)
(ρ−1)β (ρ−α)
dr
ρ−α (ρ − 1)β + (ρ − α) ρ−α (ρ − 1)β + (ρ − α) Z
R
r 0
p+N −1
β
α
(z(r)) ρ (−z 0 (r)) ρ dr ! ρ−α ρ ρ β ρ−α ρ dr · ! ρ−α ρ
ρ−α ρ
ρ−α ρ
ρ
·
|∇u(rω)| dr
R
!
−
Z
R
Z
R
β ρ
(z(r)) (−z (r))
0
(z(r)) z (r) dr 0
α α+β α+β
αρ
ρ
.
R
α ρ
0
β α
(ρ−1)β+(ρ−α) ρ
0
α α+β
! αρ
αρ
(ρ−1)β+(ρ−α) ρ
(z(0))
α+β ρ
αρ
dr
! αρ
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
General Multivariate Weighted Opial Inequalities
We have proved that Z R
r(p+N −1)
α+β ρ
0
Z
≤ M Finally we derive Z Z S N −1
R
r 0
(p+N −1)
≤ M
Z
|u(rω)|β |∇u(rω)|α dr R
rp+N −1 |∇u(rω)|ρ dr
0
α+β ρ
Z
S N −1
R
+(1−N )
!
α+β ρ
.
(11.23) !
|u(rω)|β |∇u(rω)|α rN −1 dr dω
p
ρ N −1
r |∇u(rω)| r
0
159
dr
! (α+β) ρ
dω ρ ρ−α−β
(if ρ > α + β then we apply again H¨ older’s inequality with indices ρ α+β to find) Z ρ−α−β ρ ρ ≤M 1 ρ−α−β dω S N −1 ! ! α+β Z Z R ρ · rp |∇u(rω)|ρ rN −1 dr dω S N −1
0
ρ−α−β ρ
= M ωN
Z
Z
S N −1
R
rp |∇u(rω)|ρ rN −1 dr
0
!
dω
!
α+β ρ
We have established inequality (11.16).
and
.
Corollary 11.19. Let all the conditions of Theorem 11.17 hold, except now consider u ∈ H01 (B(0, R)). Then Z α+β |x| (p+N −1) ρ +(1−N ) |u(x)|β |∇u(x)|α dx B(0,R)
ρ−α−β ρ
≤ M ωN Proof.
Similar to Corollary 11.10.
Z
B(0,R)
|x|p |∇u(x)|ρ dx
!
α+β ρ
.
(11.24)
Next we give the final main result of this chapter. Theorem 11.20. Let Ω be a smooth bounded domain in RN , N ≥ 1. Let ρ > 1, α, β > 0, ρ ≥ α + β, p ≥ 1 − N . Then there exists a constant K = K(Ω, α, β, ρ, N ) such that Z Ω
|x|
(p+N −1)
α+β ρ
+(1−N )
|u(x)|β |∇u(x)|α dx Z α+β ρ ≤ K |x|p |∇u(x)|ρ dx , Ω
(11.25)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
160
for all u ∈ H01 (Ω). Here ρ−α−β ρ
K = M ωN
(11.26)
is the constant of inequality (11.16). The M in (11.26) is given by (11.17), where R is the radius of the smallest ball centered at the origin and containing Ω. Proof.
Similar to Theorem 11.11 by applying Corollary 11.19.
Remark 11.21. (i) When u is spherically symmetric then (11.24) reduces to ! α+β Z R Z R ρ α+β , r(p+N −1) ρ |u(r)|β |u0 (r)|α dr ≤ M rp+N −1 |u0 (r)|ρ dr 0
0
(11.27)
where M is given by (11.17). (ii) When p = 0 then (11.25) collapses into Z
Ω
|x|(N −1)
α+β−ρ ρ
for all u ∈ H01 (Ω).
|u(x)|β |∇u(x)|α dx ≤ K
Z
Ω
|∇u(x)|ρ dx
α+β ρ
,
(11.28)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Chapter 12
Opial Inequalities for Widder Derivatives
Various Lp form Opial type inequalities are given for Widder derivatives. This treatment relies on [36]. 12.1
Introduction
This chapter is greatly motivated by the article of Z. Opial [195]. Theorem 12.1. (Opial [195]) Let x(t) ∈ C 1 ([0, h]) be such that x(0) = x(h) = 0, and x(t) > 0 in (0, h). Then, Z h Z h h 0 0 (x (t))2 dt. (12.1) |x(t)x (t)|dt ≤ 4 0 0 In the last inequality the constant h/4 is the best possible. Opial type inequalities have applications in establishing uniqueness of solution to initial value problems in differential equations, see [244]. 12.2
Background
The following are taken from [243]. Let f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]), n ≥ 0, and the Wronskians u0 (x) u1 (x) . . . ui (x) u0 (x) u0 (x) . . . u0 (x) 0 1 i , Wi (x) := W [u0 (x), u1 (x), . . . , ui (x)] := .. . (i) (i) u0 (x) u(i) 1 (x) . . . ui (x)
(12.2)
i = 0, 1, . . . , n. Here W0 (x) = u0 (x). Assume Wi (x) > 0 over [a, b], i = 0, 1, . . . , n. For i ≥ 0, the differential operator of order i (Widder derivative): Li f (x) :=
W [u0 (x), u1 (x), . . . , ui−1 (x), f (x)] , Wi−1 (x) 161
(12.3)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
162
i = 1, . . . , n + 1; L0 f (x) := f (x), ∀ x ∈ [a, b]. Consider also u0 (t) u1 (t) u0 (t) u01 (t) 0 1 .. gi (x, t) := . Wi (t) (i−1) (i−1) u0 (t) u1 (t) u0 (x) u1 (x)
i = 1, 2, . . . , n; g0 (x, t) :=
u0 (x) , ∀ x, t ∈ [a, b]. u0 (t)
, (i−1) . . . ui (t) . . . ui (x)
... ...
ui (t) u0i (t)
(12.4)
Example ([243]). Sets of the form {u0 , u1 , . . . , un } are {1, x, x2 , . . . , xn }, {1, sin x, − cos x, − sin 2x, cos 2x, . . . , (−1)n−1 sin nx, (−1)n cos nx}, etc. We also mention the generalized Widder–Taylor’s formula, see [243]. Theorem 12.2. Let the functions f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]), and the Wronskians W0 (x), W1 (x), . . . , Wn (x) > 0 on [a, b], x ∈ [a, b]. Then for t ∈ [a, b] we have u0 (x) f (x) = f (t) + L1 f (t)g1 (x, t) + · · · + Ln f (t)gn (x, t) + Rn (x), (12.5) u0 (t) where Rn (x) :=
Z
x
gn (x, s)Ln+1 f (s)ds.
(12.6)
t
For example ([243]) one could take u0 (x) = c > 0. If ui (x) = xi , i = 0, 1, . . . , n, defined on [a, b], then Li f (t) = f (i) (t)
and gi (x, t) =
(x − t)i , i!
t ∈ [a, b].
We need Corollary 12.3 (on Theorem 12.2). By additionally assuming for fixed x0 ∈ [a, b] that Li f (x0 ) = 0, i = 0, 1, . . . , n, we get that Z x f (x) = gn (x, t)Ln+1 f (t)dt, ∀ x ∈ [a, b]. (12.7) x0
12.3
Results
From now on we are working under the terms and assumptions of Theorem 12.2 and Corollary 12.3. We first present the following
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Widder Derivatives
163
1 1 + = 1. Then p q 1/p Z x Z w Z x (gn (w, t))p dt dw |f (w)||Ln+1 f (w)|dw ≤ 2−1/q
Theorem 12.4. Let x ≥ x0 , x0 , x ∈ [a, b] and p, q > 1:
x0
x0
x0
×
Z
x x0
|Ln+1 f (w)|q dw
By H¨ older’s inequality we have Z x 1/p Z p |f (x)| ≤ |gn (x, t)| dt
Proof.
x0
Call z(w) := Thus
Z
2/q
.
x x0
(12.8)
q
|Ln+1 f (t)| dt
1/q
.
(12.9)
w
|Ln+1 f (t)|q dt,
x0
x0 ≤ w ≤ x
(z(x0 ) = 0).
z 0 (w) = |Ln+1 f (w)|q and |Ln+1 f (w)| = (z 0 (w))1/q . From (12.9) we obtain |f (w)||Ln+1 f (w)| ≤
Z
w x0
p
|gn (w, t)| dt
1/p
(z(w)z 0 (w))1/q .
Integrating the last inequality over [x0 , x] we obtain 1/p Z x Z x Z w |f (w)||Ln+1 f (w)|dw ≤ (z(w)z 0 (w))1/q dw |gn (w, t)|p dt x0
x0
≤
Z
x x0
Z
=
w x0
Z
1/p Z |gn (w, t)|p dt dw
x x0
x0
Z
w x0
p
|gn (w, t)| dt dw
x
z(w)z 0 (w)dw
x0
1/p
1/q
(z(x))2/q , 21/q
proving the claim of Theorem 12.4. See also Remark 22.11 next. The counter part of the previous result follows. 1 1 + = 1. Then is p q 1/p Z x Z w Z x0 |gn (w, t)|p dt dw |f (w)||Ln+1 f (w)|dw ≤ 2−1/q
Theorem 12.5. Let x ≤ x0 , x0 ∈ [a, b] and p, q > 1:
x
x0
x0
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
164
Z
×
Proof.
x0
|Ln+1 f (w)|q dw
x
2/q
Here take x ≤ x0 and p, q > 1 such that
inequality we have
Z
≤
Z |f (x)| =
x0 x
z(x) :=
Z
x0 x
x
Z
|gn (x, t)||Ln+1 f (t)|dt ≤
Call
x0
x0
p
1/p Z
|Ln+1 f (t)|q dt ≥ 0,
That is −z(x) =
Z
(12.10)
1 1 + = 1. From H¨ older’s p q
gn (x, t)Ln+1 f (t)dt |gn (x, t)| dt
x
.
x0 x
q
|Ln+1 f (t)| dt
1/q
.
(12.11)
z(x0 ) = 0.
x x0
|Ln+1 f (t)|q dt ≤ 0,
and −z 0 (x) = |Ln+1 f (x)|q ≥ 0, and |Ln+1 f (x)| = (−z 0 (x))1/q ,
x ∈ [a, x0 ].
Hence by (12.11) (x ≤ w ≤ x0 ) Z x0 1/p |f (w)||Ln+1 f (w)| ≤ |gn (w, t)|p dt (z(w)(−z 0 (w))1/q . w
Integrating the last inequality over [x, x0 ] we obtain 1/p Z x0 Z x0 Z x0 p (z(w)(−z 0 (w)))1/q dw |gn (w, t)| dt |f (w)||Ln+1 f (w)|dw ≤ w
x
x
≤
Z
x0 x
Z
x0 w
= 2−1/q proving the claim.
1/p Z |gn (w, t)|p dt dw Z
x x0
Z
w x0
x0
z(w)(−z 0 (w))dw
x
1/q
1/p (z(x))2/q , |gn (w, t)|p dt dw
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Widder Derivatives
165
Next we study the extreme cases. Theorem 12.6. Here p = 1, q = ∞ and x ≥ x0 . Then Z x Z w Z x |f (w)||Ln+1 f (w)|dw ≤ gn (w, t)dt dw kLn+1 f k2∞ . x0
x0
By
Proof.
Z
f (w) = we get |f (w)| ≤ and
Z
(12.12)
x0
w
gn (w, t)Ln+1 f (t)dt x0
w x0
|f (w)||Ln+1 f (w)| ≤
|gn (w, t)|dt kLn+1 f k∞ , Z
w x0
|gn (w, t)|dt kLn+1 f k2∞ .
Integrating the last inequality we find (12.12). See also Remark 12.11 next. Theorem 12.7. Again p = 1, q = ∞, but x ≤ x0 . Then Z x Z w Z x0 |f (w)||Ln+1 f (w)|dw ≤ gn (w, t)|dt dw (kLn+1 f k2∞ ). x
x0
Here x ≤ w ≤ x0 , and
Proof.
Z =
Thus
and
Z
Z |f (w)| =
gn (w, t)Ln+1 f (t)dt x0 Z x0 x0 gn (w, t)Ln+1 f (t)dt ≤ |gn (w, t)|dt kLn+1 f k∞ . w
w
w
|f (w)||Ln+1 f (w)| ≤
x0 x
(12.13)
x0
|f (w)||Ln+1 f (w)|dw ≤
proving the claim.
Z
Z
x0
w
x0 x
|gn (w, t)|dt kLn+1 f k2∞ ,
Z
x0 w
|gn (w, t)|dt dw kLn+1 f k2∞ ,
1 1 + = 1. Then p q Z x Z x −1/q |f (w)||Ln+1 f (w)|dw ≤ 2
Corollary 12.8. Let p, q > 1: Z
Proof.
x0
x0
Z ×
x
x0
By Theorems 12.4, 12.5.
w x0
p
|gn (w, t)| dt dw
2/q . |Ln+1 f (w)|q dw
1/p (12.14)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
166
In particular when p = q = 2 we have Corollary 12.9. It holds Z x Z −1/2 |f (w)||Ln+1 f (w)|dw ≤ 2 x0
x Z w
x0
2
(gn (w, t)) dt dw
x0
Z x × (Ln+1 f (w))2 dw .
1/2 (12.15)
x0
Furthermore we get
Corollary 12.10. Let p = 1, q = ∞. Then Z x Z x Z w ≤ |f (w)||L f (w)|dw |g (w, t)|dt dw kLn+1 f k2∞ . n+1 n x0
x0
By Theorems 12.6, 12.7.
Proof.
(12.16)
x0
We need to make Remark 12.11. We define (see [243]) φ0 (x) := W0 (x), φ1 (x) :=
W1 (x) ,..., (W0 (x))2
in general φk (x) :=
Wk (x)Wk−2 (x) , (Wk−1 (x))2
k = 2, 3, . . . , n.
The functions φi (x) are positive on [a, b]. According to [243] we get, for x, x0 not fixed, that Z x Z xn−2 Z x1 φ0 (x) gn (x, x0 ) = φn−1 (xn−1 ) ... φ1 (x1 ) φ0 (x0 ) . . . φn (x0 ) x0 x0 x0 Z xn−1 φn (xn )dx1 dx2 . . . dxn × x0
1 = φ0 (x0 ) . . . φn (x0 )
Z
x
φ0 (s) . . . φn (s)gn−1 (x, s)ds.
(12.17)
x0
We get that gn (x, x) = 0, all x ∈ [a, b], and gn (x, x0 ) > 0, x > x0 , x, x0 ∈ [a, b], ∀ n ≥ 1. Also g0 (x, x0 ) > 0 for any x, x0 ∈ [a, b]. To complete the chapter we present 1 1 + = 1, and x > x0 , p q x, x0 ∈ [a, b]. Assume that Ln+1 f is of fixed sign and nowhere zero. Then 1/p Z x Z w Z x (gn (w, t))p dt dw |f (w)||Ln+1 f (w)|dw ≥ 2−1/q
Theorem 12.12. Let 0 < p < 1 and q < 0 be such that
x0
x0
x0
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Widder Derivatives
Z
×
Proof.
x x0
|Ln+1 f (w)|q dw
2/q
167
.
(12.18)
Here for x0 ≤ w ≤ x we have Z w |f (w)| = gn (w, t)|Ln+1 f (t)|dt,
(12.19)
x0
by Remark 22.11. From (12.19) by H¨ older’s inequality we have Z x 1/p Z w 1/q p q |f (w)| ≥ (gn (w, t)) dt , |Ln+1 f (t)| dt x0
(12.20)
x0
for w > x0 . Consider z(w) :=
Z
w x0
|Ln+1 f (t)|q dt > 0,
z(x0 ) = 0.
So that z 0 (w) = |Ln+1 f (w)|q > 0 and|Ln+1 f (w)| = (z 0 (w))1/q , all x0 ≤ w ≤ x. Thus by (12.20) we get Z w 1/p |f (w)||Ln+1 f (w)| ≥ (z(w)z 0 (w))1/q , (gn (w, t))p dt x0
all x0 < w ≤ x. Let x0 < θ ≤ w ≤ x and θ ↓ x0 , then by integration of the last inequality we obtain Z x Z x |f (w)||Ln+1 f (w)|dw |f (w)||Ln+1 f (w)|dw = lim θ↓x0
x0
Z
≥ lim
θ↓x0
≥ lim
θ↓x0
Z
=2
x θ
−1/q
Z Z
=2
x θ
Z
w
p
(gn (w, t)) dt x0
w
p
(gn (w, t)) dt dw x0
x x0
−1/q
Z
w
p
θ
1/p
1/p
Z
x Z w
x0
That is establishing (12.18).
x0
p
(z(w)z (w))
lim
θ↓x0
(gn (w, t)) dt dw x0
1/p
0
Z
x
1/q
dw
!
0
z(w)z (w)dw θ
1/q !
lim (z 2 (x) − z 2 (θ))1/q
θ↓x0
(gn (w, t)) dt dw
1/p
(z(x))2/q .
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
168
We make Remark 12.13. We notice by (12.17) that gn (x, t) < 0,
x < t,
n odd,
gn (x, t) > 0,
x < t,
n even,
where x, t ∈ [a, b]. We give 1 1 + = 1, and x < x0 , p q x, x0 ∈ [a, b]. Let n be odd. Also assume that Ln+1 f is of fixed sign and nowhere zero. Then 1/p Z x Z w Z x0 p −1/q (−gn (w, t)) dt dw |f (w)||Ln+1 f (w)|dw ≥ 2 Theorem 12.14. Let 0 < p < 1 and q < 0 such that
× Proof.
x0
x0
x
Z
x0 x
|(Ln+1 f )(w)|q dw
2/q
.
(12.21)
Here for x ≤ w ≤ x0 we have Z w |f (w)| = gn (w, t)Ln+1 f (t)dt x0
Z =
x0
w
Z gn (w, t)Ln+1 f (t)dt = =
Z
x0
w
x0
(−gn (w, t))Ln+1 f (t)dt
(−gn (w, t))|Ln+1 f (t)|dt.
From (12.22) by H¨ older’s inequality we get Z x0 1/p Z |f (w)| ≥ (−gn (w, t))p dt w
for w < x0 . That is Z |f (w)| ≥
x0 w
|gn (w, t)|p dt
for w < x0 . Consider
z(w) := So that
Z
(12.22)
w
x0 w
1/p Z
x0 w
x0 w
|Ln+1 f (t)|q dt
|Ln+1 f (t)|q dt > 0,
−z(w) =
Z
w x0
|Ln+1 f (t)|q dt
1/q
z(x0 ) = 0.
|Ln+1 f (t)|q dt
1/q
,
(12.23)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Widder Derivatives
169
and −z 0 (w) = |Ln+1 f (w)|q > 0, i.e. z 0 (w) < 0. Hence |Ln+1 f (w)| = (−z 0 (w))1/q > 0, all x ≤ w ≤ x0 . Therefore by (12.23) we derive Z |f (w)||Ln+1 f (w)| ≥
x0
p
|gn (w, t)| dt
w
1/p
(z(w)(−z 0 (w))1/q ,
all x ≤ w < x0 . Let x ≤ w ≤ θ < x0 and θ ↑ x0 , then by integration of the last inequality we obtain Z x0 Z θ |f (w)||Ln+1 f (w)|dw = lim |f (w)||Ln+1 f (w)|dw θ↑x0
x
≥ lim
θ↑x0
≥ lim
θ↑x0
=
Z
Z
=2
Z
θ x
x0 x
−1/q
Z
x
Z
x0 w
|gn (w, t)|p dt
x0
p
x0 w
x0 x
= 2−1/q
p
!1/p
1/p
|gn (w, t)| dt dw Z Z
x0
p
x0
Z
x0 w
(z(w)(−z 0 (w)))1/q dw
Z
lim
θ↑x0
θ
1/p
!
0
z(w)(−z (w))dw x
−
lim
θ↑x0
|gn (w, t)| dt dw
w
x
1/p
|gn (w, t)| dt dw
w
Z
Z
θ
x
Z
θ
z(w)dz(w) x
!1/q
!1/q
lim (z 2 (x) − z 2 (θ))1/q
θ↑x0
1/p |gn (w, t)|p dt dw (z(x))2/q .
Hence Z
x0 x
|f (w)||Ln+1 f (w)|dw ≥ 2−1/q
× that is proving (12.21).
Z
x0 x
Z
x x0
Z
w x0
|Ln+1 f (w)|q dw
1/p |gn (w, t)|p dt dw
2/q
,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
170
Putting things together we have 1 1 + = 1, and x0 , x ∈ [a, b] p q such that x = 6 x0 . Let n be odd. Assume that Ln+1 f is of fixed sign and nowhere zero. Then Z x Z x Z w 1/p −1/q p |f (w)||Ln+1 f (w)|dw ≥ 2 |gn (w, t)| dt dw Corollary 12.15. Let 0 < p < 1 and q < 0 be such that
x0
Proof.
x0
Z ×
x
x0
x0
2/q . |Ln+1 f (w)| dw q
By Theorems 12.12, 12.14.
(12.24)
We give without proof the similar Theorem 12.14, see Remark 12.13. 1 1 + = 1, and x < x0 , p q x, x0 ∈ [a, b]. Let n be even. Also assume that Ln+1 f is of fixed sign and nowhere zero. Then 1/p Z x Z w Z x0 (gn (w, t))p dt dw |f (w)||Ln+1 f (w)|dw ≥ 2−1/q Theorem 12.16. Let 0 < p < 1 and q < 0 such that
x
x0
×
Z
x0 x
x0
|Ln+1 f (w)|q dw
2/q
.
(12.25)
We finish here with 1 1 + = 1, and x0 , x ∈ [a, b] p q such that x = 6 x0 . Let n be even. Assume that Ln+1 f is of fixed sign and nowhere zero. Then Z x 1/p Z x Z w p −1/q (gn (w, t)) dt dw |f (w)||Ln+1 f (w)|dw ≥ 2 Corollary 12.17. Let 0 < p < 1 and q < 0 be such that
x0
x0
Proof.
Z ×
x
x0
x0
2/q . |Ln+1 f (w)| dw
By Theorems 12.12, 12.16.
q
(12.26)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Chapter 13
Opial Inequalities for Linear Differential Operators
Various Lp form Opial inequalities [195] are given for a Linear Differential Operator L, involving its related initial value problem solution y, Ly, the associated Green’s function H and initial conditions point x0 ∈ R. This chapter follows [18]. 13.1
Background
Here we follow [169], pp. 145–154. Let I be a closed interval of R. Let ai (x), i = 0, 1, . . . , n − 1 (n ∈ N), h(x) be continuous functions on I and let L = D n + an−1 (x)Dn−1 + · · · + a0 (x) be a fixed linear differential operator on C n (I). Let y1 (x), . . . , yn (x) be a set of linear independent solutions to Ly = 0. Here the associated Green’s function for L is y1 (t) · · · yn (t) y1 (t) · · · yn (t) 0 0 0 y1 (t) . . . yn0 (t) , y1 (t) · · · yn (t) .. .. H(x, t) := . . , (n−2) (n−2) y (n−2) (t) · · · y (n−2) (t) y (t) · · · yn (t) n 1 1 (n−1) (n−1) y1 (x) · · · yn (x) y (t) · · · yn (t) 1 which is a continuous function on I 2 . Consider fixed x0 ∈ I, then Z x y(x) = H(x, t)h(t)dt, all x ∈ I x0
is the unique solution to the initial value problem Ly = h; y (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. 13.2
Results
We first give 171
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
172
Proposition 13.1. Let x ≥ x0 ; x0 , x ∈ I and p, q > 1: p1 + 1q = 1. Then 1/p Z x Z w Z x |H(w, t)|p dt dw |y(w)||(Ly)(w)|dw ≤ 2−1/q · x0
·
Z
x
x0
x0
q
2/q
|(Ly)(w)| dw
x0
.
(13.1)
older’s Proof. Here take x ≥ x0 and p, q > 1 such that p1 + 1q = 1. From H¨ inequality we have Z x 1/p Z x 1/q . (13.2) |y(x)| ≤ |H(x, t)|p dt |h(t)|q dt x0
Set
z(w) := Thus
Z
w x0
x0
|h(t)|q dt,
x0 ≤ w ≤ x (z(x0 ) = 0).
z 0 (w) = |h(w)|q and |h(w)| = (z 0 (w))1/q . From (13.2) we obtain |y(w)| |h(w)| ≤
Z
w x0
|H(w, t)|p dt
1/p
· (z(w) · z 0 (w))1/q .
Integrating the last inequality over [x0 , x] we derive 1/p Z x Z x Z w p |y(w)| |h(w)|dw ≤ · (z(w) · z 0 (w))1/q dw |H(w, t)| dt x0
≤ =
Z
Z
x
x0 x x0
Z
Z
x0
w
x0 w x0
x0
1/p Z · |H(w, t)| dt dw
p
x
x0
0
z(w) · z (w)dw
1/p (z(x))2/q |H(w, t)|p dt dw · , 21/q
1/q
proving the claim of Proposition 13.1.
The counterpart of the previous result follows. Proposition 13.2. Let x ≤ x0 ; x0 , x ∈ I and p, q > 1: p1 + 1q = 1. Then 1/p Z x Z w Z x0 p −1/q |H(w, t)| dt dw |y(w)| |(Ly)(w)|dw ≤ 2 · x
x0
·
Z
x0 x
|(Ly)(w)|q dw
x0
2/q
.
(13.3)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Linear Differential Operators
173
older’s Proof. Here take x ≤ x0 and p, q > 1 such that p1 + 1q = 1. From H¨ inequality we have Z x 0 Z x0 |H(x, t)| |h(t)|dt |y(x)| = H(x, t)h(t)dt ≤ x
x
≤
Set
Z
x0
p
|H(x, t)| dt
x
Z
z(x) := That is
x0 x
x0
Z
x x0
q
|h(t)| dt
x
|h(t)|q dt ≥ 0,
−z(x) = and
1/p Z
1/q
.
(13.4)
z(x0 ) = 0.
|h(t)|q dt ≤ 0,
−z 0 (x) = |h(x)|q ≥ 0, and |h(x)| = (−z 0 (x))1/q , Hence by (13.4) (x ≤ w ≤ x0 ) Z |y(w)| · |h(w)| ≤
x0 w
|H(w, t)|p dt
x ∈ I.
1/p
· (z(w) · (−z 0 (w)))1/q .
Integrating the last inequality over [x, x0 ] we find 1/p Z x0 Z x0 Z x0 p |H(w, t)| dt |y(w)| |h(w)|dw ≤ w
x
x
·(z(w) · (−z 0 (w)))1/q · dw Z x0 Z x0 1/p Z ≤ |H(w, t)|p dt · dw x
= 2−1/q ·
w
Z
x
x0
Z
w
x0
|H(w, t)|p dt · dw
x0 x
1/p
(−z 0 (w) · z(w)) · dw
1/q
· (z(x))2/q ,
proving the claim of Proposition 13.2.
Extreme cases come next. Proposition 13.3. Here p = 1, q = ∞ and x ≥ x0 . Then Z x |y(w)||(Ly)(w)|dw x0
≤
Z
x x0
Z
w x0
|H(w, t)|dt dw · kLyk2∞ .
(13.5)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
174
From
Proof.
Z
y(w) = we obtain |y(w)| ≤
Z
and
w
H(w, t)h(t)dt x0
|H(w, t)|dt khk∞ ,
x0
Z
|y(w)| |(Ly)(w)| ≤
w
w x0
|H(w, t)|dt · (kLyk∞ )2 .
Integrating the last inequality we obtain (13.5).
Proposition 13.4. Again p = 1, q = ∞, but x ≤ x0 . Then Z x0 |y(w)| |(Ly)(w)|dw x
≤
Z
x x0
Z
w x0
|H(w, t)|dt dw · (k(Ly)k2∞ ).
(13.6)
Here x ≤ w ≤ x0 , and Z x 0 Z w H(w, t)h(t)dt H(w, t)h(t)dt = |y(w)| = w x Z 0 x0 ≤ |H(w, t)|dt khk∞ .
Proof.
w
Therefore |y(w)| |(Ly)(w)| ≤ and
Z
x0
|y(w)| |(Ly)(w)|dw ≤
x
Corollary 13.5. Let p, q > 1: Z
Proof.
x x0
1 p
Z
Z +
x0 w
x0 x
1 q
|H(w, t)|dt (kLyk∞ )2 ,
Z
w
x
x0
|H(w, t)|dt dw
x Z w
x0
x0
· (kLyk∞ )2 .
p
|H(w, t)| dt dw
2/q . |(Ly)(w)|q dw
By Propositions 13.1, 13.2.
= 1. Then
Z −1/q |y(w)| |(Ly)(w)|dw ≤ 2 · Z ·
x0
1/p (13.7)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Linear Differential Operators
175
In particular when p = q = 2 we obtain Corollary 13.6. It holds Z x Z ≤ 2−1/2 · |y(w)| |(Ly)(w)|dw x0
x x0
Z
w x0
1/2 (H(w, t))2 dt dw
Z x ((Ly)(w))2 dw . ·
(13.8)
x0
Furthermore we have
Corollary 13.7 Let p = 1, q = ∞. Then Z x |y(w)| |(Ly)(w)|dw x0
≤
Proof.
Z
Z
x
x0
w
x0
|H(w, t)|dt dw · kLyk2∞.
(13.9)
By Propositions 13.3 and 13.4.
To complete the chapter we present Proposition 13.8. Let 0 < p < 1 and q be such that x0 , x ∈ I. Suppose that
1 p
+
1 q
= 1, and x > x0 ,
H(w, t) ≥ 0 for x0 ≤ t ≤ w, w ∈ I. Also assume that Ly = h is of fixed sign and nowhere zero. Then Z x Z w 1/p Z x −1/q p |y(w)| |(Ly)(w)|dw ≥ 2 (H(w, t)) dt dw x0
x0
· Proof.
Z
x x0
x0
|(Ly)(w)|q dw
2/q
.
(13.10)
Here for x0 ≤ w ≤ x we have Z w H(w, t)|h(t)|dt. |y(w)| =
(13.11)
x0
From (13.11) by H¨ older’s inequality we derive 1/p Z Z w p (H(w, t)) dt · |y(w)| ≥ x0
for w > x0 . Consider z(w) :=
Z
w x0
|h(t)|q dt,
w x0
q
|h(t)| dt
z(x0 ) = 0.
1/q
,
(13.12)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
176
So that z 0 (w) = |h(w)|q and |h(w)| = (z 0 (w))1/q , all x0 ≤ w ≤ x. Hence by (13.12) we get 1/p Z w · (z(w) · z 0 (w))1/q , (H(w, t))p dt |y(w)| |h(w)| ≥ x0
all x0 < w ≤ x. Let x0 < θ ≤ w ≤ x and θ ↓ x0 , then by integration of the last inequality we obtain Z x Z x |y(w)| |h(w)|dw = lim |y(w)| |h(w)|dw θ↓x0 θ x0 ! Z Z x
≥ lim
θ↓x0
≥ lim
θ↓x0
= 2
w
θ
Z
−1/q
= 2−1/q
·
1/p
· (z(w) · z 0 (w))1/q · dw
(H(w, t))p dt
x0
x θ
Z
Z
Z
x x0
w p
(H(w, t)) dt dw x0
Z
x Z
x0
w
p
1/p
(H(w, t)) dt dw x0
w
(H(w, t))p dt dw x0
Z
· lim
θ↓x0
1/p
1/p
x 0
z(w)z (w)dw θ
1/q !
· lim (z 2 (x) − z 2 (θ))1/q θ↓x0
· (z(x))2/q .
That is establishing (13.10).
Proposition 13.9. Let 0 < p < 1 and q be such that x0 , x ∈ I. Suppose that
1 p
+
1 q
= 1, and x < x0 ,
H(w, t) ≤ 0 for w ≤ t ≤ x0 , w ∈ I. Also assume that Ly = h is of fixed sign and nowhere zero. Then Z x Z w 1/p Z x0 |y(w)| |(Ly)(w)|dw ≥ 2−1/q · |H(w, t)|p dt dw x
· Proof.
Z
x0 x
x0
x0
q
2/q
|(Ly)(w)| dw
.
(13.13)
Here for x ≤ w ≤ x0 we have Z x0 Z w H(w, t)h(t)dt H(w, t)h(t)dt = |y(w)| = Zw x0 Z xx0 0 (−H(w, t))|h(t)|dt. (−H(w, t))h(t)dt = =
(13.14)
w
w
From (13.14) by H¨ older’s inequality we find Z x0 1/p Z |y(w)| ≥ (−H(w, t))p dt w
x0 w
|h(t)|q dt
1/q
,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Linear Differential Operators
for w < x0 . That is |y(w)| ≥ for w < x0 . Consider
Z
x0
p
|H(w, t)| dt
w
z(w) := So that
Z
x0
1/p Z
|h(t)|q dt,
w
−z(w) = and
Z
w x0
x0
q
|h(t)| dt
w
177
1/q
,
(13.15)
z(x0 ) = 0.
|h(t)|q dt
−z 0 (w) = |h(w)|q ,
with
|h(w)| = (−z 0 (w))1/q ,
Therefore by (13.15) we find Z |y(w)| |h(w)| ≥
x0 w
all x ≤ w ≤ x0 .
|H(w, t)|p dt
1/p
· (z(w) · (−z 0 (w)))1/q ,
all x ≤ w < x0 . Let x ≤ w ≤ θ < x0 and θ ↑ x0 , then by integration of the last inequality we obtain Z θ Z x0 |y(w)| |h(w)|dw = lim |y(w)| |h(w)|dw θ↑x0 x x ! Z Z θ
≥ lim
θ↑x0
x
Z
≥ lim
θ↑x0
=
Z
= 2
x0
x0 x
−1/q
= 2−1/q · Therefore Z x0 x
θ x
Z ·
w
Z
1/p
|H(w, t)|p dt
x0 w
p
p
|H(w, t)| dt dw
w
Z
|H(w, t)| dt dw
x0
Z
· (z(w) · (−z 0 (w)))1/q · dw
x0 x
Z
x0 Z
x
x0 w x0 w
1/p
p
|H(w, t)|p dt dw
that is proving (13.13).
x0 x
Z
|h(w)|q dw
θ↑x0
−
θ↑x0
|H(w, t)| dt dw
Z
· lim
· lim
|y(w)| |h(w)|dw ≥ 2−1/q · ·
!1/p
1/p
1/p
0
x
(−z(w))(z (w)) · dw
θ
z(w)dz(w) x
!1/q
!1/q
θ↑x0
· (z(x))2/q .
x0
x0
θ
· lim (z 2 (x) − z 2 (θ))1/q
x Z w
2/q
Z
Z
1/p |H(w, t)|p dt dw
,
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
178
Putting things together we have Corollary 13.10. Let 0 < p < 1 and q be such that x 6= x0 . Suppose that
1 p
+
1 q
= 1, and x0 , x ∈ I that
(w − x0 ) · H(w, t) ≥ 0, for all t between x0 , w ∈ I. Also assume that Ly = h is of fixed sign and nowhere zero. Then Z x Z x Z w 1/p p ≥ 2−1/q · |y(w)| |(Ly)(w)|dw |H(w, t)| dt dw x0
Proof.
x0
Z ·
x
2/q . |(Ly)(w)| dw q
x0
x0
By Propositions 13.8 and 13.9.
(13.16)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Chapter 14
Opial Inequalities for Vector Valued Functions
Various Lp form Opial type inequalities are given for functions valued in a Banach vector space with applications in C. This chapter relies on [52].
14.1
Introduction
This chapter is greatly motivated by the article of Z. Opial [195]. Theorem 14.1. (Opial [195]) Let x(t) ∈ C 1 ([0, h]) be such that x(0) = x(h) = 0, and x(t) > 0 in (0, h). Then Z Z h h h 0 0 (x (t))2 dt. (14.1) |x(t)x (t)|dt ≤ 4 0 0 In the last inequality the constant h/4 is the best possible. Equality holds for the function x(t) = t on [0, h/2] and x(t) = h − t on [h/2, h]. Opial type inequalities have applications in establishing uniqueness of solution to initial value problems in differential equations, see [244]. We are also motivated by [16], [21]. We need
14.2
Background
(see [230], pp. 83–94) Let f (t) be a function defined on [a, b] ⊆ R taking values in a real or complex normed linear space (X, k · k). Then f (t) is said to be differentiable at a point t0 ∈ [a, b] if the limit f (t0 + h) − f (t0 ) (14.2) f 0 (t0 ) := lim h→0 h 179
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
180
exists in X, the convergence is in k · k. This is called the derivative of f (t) at t = t 0 . We call f (t) differentiable on [a,b], iff there exists f 0 (t) ∈ X for all t ∈ [a, b]. Similarly and inductively are defined higher order derivatives of f , denoted 00 f , f (3) , ..., f (k) , k ∈ N, just as for numerical functions. For all the properties of derivatives see [230], pp. 83–86. Let now (X, k · k) be a Banach space, and f : [a, R b b] → X. We define the vector valued Riemann integral a f (t)dt ∈ X as the limit of the vector valued Riemann sums in X, convergence is in k · k. The definition is as for the numerical valued functions. Rb If a f (t)dt ∈ X we call f integrable on [a,b]. For the properties of vector valued Riemann integrals see [230], pp. 86–91. We define the space C n ([a, b], X), n ∈ N, of n-times continuously differentiable functions from [a,b] into X; here continuity is with respect to k · k and defined in the usual way as for numerical functions. Let (X, k · k) be a Banach space and f ∈ C n ([a, b], X), then we have the vector valued Taylor’s formula, see [230], pp. 93–94, and also [227], (IV, 9; 47): It holds En (x, y) := f (y) − f (x) − f 0 (x)(y − x) 1 1 f (n−1) (x)(y − x)n−1 − f 00 (x)(y − x)2 − · · · − 2 (n − 1)! =
1 (n − 1)!
Z
y x
(y − t)n−1 f (n) (t)dt,
∀ x, y ∈ [a, b].
(14.3)
In particular (14.3) is true when X = Rm , Cm , m ∈ N, etc. In case of some x0 ∈ [a, b] such that f (k) (x0 ) = 0, k = 0, 1, ..., n − 1, then Z y 1 f (y) = (y − t)n−1 f (n) (t)dt, ∀ y ∈ [a, b]. (14.4) (n − 1)! x0 In that case En (x0 , y) = f (y). 14.3
Results
Here we consider always X to be a Banach space, n ∈ N, and f ∈ C n ([a, b], X), [a, b] ⊆ R. We fix x0 ∈ [a, b]. We present the first result of the chapter. Theorem 14.2. Let p, q > 1 : Then I1 :=
Z
1 p
y x0
+
1 q
= 1, y ≥ x0 ; y, x0 ∈ [a, b].
kEn (x0 , w)k kf (n) (w)kdw
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Vector Valued Functions 2
≤
(y − x0 )n−1+ p 21/q (n − 1)![(p(n − 1) + 1)(p(n − 1) + 2)]1/p
Z
y x0
181
kf (n) (w)kq dw
2/q
.
(14.5)
We give Corollary 14.3. For y ∈ [x0 , b] we have I1 ≤
Z
(y − x0 )n p 2(n − 1)! n(2n − 1)
y x0
kf (n) (w)k2 dw .
Proof of Theorem 14.2. We have by (14.3) that Z y 1 (y − t)n−1 f (n) (t)dt, ∀ y ≥ x0 ; x0 , y ∈ [a, b]. En (x0 , y) = (n − 1)! x0 By H¨ older’s inequality we derive 1 kEn (x0 , y)k ≤ (n − 1)! 1 ≤ (n − 1)!
Z
Z
x0
(y − t)
(14.7)
y x0
y p(n−1)
(14.6)
dt
(y − t)n−1 kf (n) (t)kdt
1/p Z
y x0
kf
(n)
q
(t)k dt
1/q
1
= where z(y) :=
(y − x0 )n−1+ p 1 (z(y))1/q , (n − 1)! (p(n − 1) + 1)1/p
(14.8)
Z
(14.9)
y x0
kf (n) (t)kq dt, x0 ≤ y ≤ b (z(x0 ) = 0).
Thus z 0 (y) = kf (n) (y)kq and kf (n) (y)k = (z 0 (y))1/q .
(14.10)
Therefore we obtain 1
kEn (x0 , y)k kf
(n)
1 (y − x0 )n−1+ p (y)k ≤ (z(y)z 0 (y))1/q . (n − 1)! (p(n − 1) + 1)1/p
Integrating the last inequality we have Z y kEn (x0 , w)k kf (n) (w)kdw x0
≤
1 (n − 1)!(p(n − 1) + 1)1/p
Z
y x0
1
(w − x0 )n−1+ p (z(w)z 0 (w))1/q dw
(14.11)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
182
Z
1 (n − 1)!(p(n − 1) + 1)1/p
≤
y x0
(w − x0 )p(n−1)+1 dw
1/p Z
y
z(w)z 0 (w)dw
x0
1/q
n−1+ p2
=
21/q (n
(y − x0 ) (z(y))2/q , − 1)![(p(n − 1) + 1)(p(n − 1) + 2)]1/p
(14.12)
proving the claim of the theorem. We continue with
Theorem 14.4. Let p, q > 1 : p1 + q1 = 1, y ∈ [a, x0 ]. Then Z x0 kEn (x0 , w)k kf (n) (w)kdw I2 := y
2
(x0 − y)n−1+ p ≤ 1/q 2 (n − 1)![(p(n − 1) + 1)(p(n − 1) + 2)]1/p
Z
x0
kf
y
(n)
q
(w)k dw
2/q
. (14.13)
We give Corollary 14.5. For y ∈ [a, x0 ] we have
(x0 − y)n p I2 ≤ 2(n − 1)! n(2n − 1)
Z
x0 y
kf
(n)
2
(w)k dw .
(14.14)
Proof of Theorem 14.4. We have by (14.3) that
Z x 0
1 n−1 (n)
kEn (x0 , y)k = (y − t) f (t)dt
(n − 1)! y
1 (n − 1)!
≤
≤
1 (n − 1)!
Z
x0 y
Z
x0
y
(t − y)n−1 kf (n) (t)kdt
(t − y)p(n−1) dt
1/p Z
x0 y
kf (n) (t)kq dt
1/q
1
= Here
(x0 − y)n−1+ p (z(y))1/q . (n − 1)!(p(n − 1) + 1)1/p
z(y) :=
Z
x0 y
kf (n) (t)kq dt,
and −z(y) =
Z
y x0
(z(x0 ) = 0),
kf (n) (t)kq dt ≤ 0,
−z 0 (y) = kf (n) (y)kq ≥ 0,
(14.15)
(14.16)
(14.17) (14.18)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Vector Valued Functions
183
and kf (n) (y)k = (−z 0 (y))1/q ,
a ≤ y ≤ x0 .
(14.19)
Therefore kEn (x0 , y)k kf (n) (y)k 1
(x0 − y)n−1+ p (z(y)(−z 0 (y)))1/q , ≤ (n − 1)!(p(n − 1) + 1)1/p
Hence we find
≤
Z
x0
a ≤ y ≤ x0 .
(14.20)
kEn (x0 , w)k kf (n) (w)kdw
y
1 (n − 1)!(p(n − 1) + 1)1/p
Z
x0 y
1 ≤ (n − 1)!(p(n − 1) + 1)1/p Z
x0
1
(x0 − w)n−1+ p (z(w)(−z 0 (w)))1/q dw Z
x0 y
(x0 − w)
z(w)(−z 0 (w))dw
y
p(n−1)+1
dw
(14.21)
1/p
1/q
(14.22)
2
(x0 − y)n−1+ p (z(y))2/q , ∀ y ∈ [a, x0 ], = 1/q 2 (n − 1)![(p(n − 1) + 1)(p(n − 1) + 2)]1/p
proving the claim of the theorem. Combining Theorems 14.2 and 14.4 we obtain
(14.23)
Theorem 14.6. Let p, q > 1 : p1 + q1 = 1 and y, x0 ∈ [a, b]. Then Z y kEn (x0 , w)k kf (n) (w)kdw I := x0
Z y 2/q 2 |y − x0 |n−1+ p (n) q . ≤ 1/q kf (w)k dw 2 (n − 1)![(p(n − 1) + 1)(p(n − 1) + 2)]1/p x0
(14.24)
Combining Corollaries 14.3 and 14.5 we derive
Corollary 14.7. For y, x0 ∈ [a, b] it holds I≤
Z y |y − x0 |n (n) 2 . p kf (w)k dw 2(n − 1)! n(2n − 1) x0
A special but important case follows
(14.25)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
184
Theorem 14.8. Let p, q > 1 : p1 + 1q = 1 and y, x0 ∈ [a, b]. Suppose further that f (k) (x0 ) = 0, k = 0, 1, ..., n − 1. Then Z y (n) kf (w)k kf (w)kdw x0
≤ min
Proof.
Z y 2/q 2 |y − x0 |n−1+ p (n) q , kf (w)k dw 1/q 1/p 2 (n − 1)![(p(n − 1) + 1)(p(n − 1) + 2)] x0 Z y ! |y − x0 |n p kf (n) (w)k2 dw . 2(n − 1)! n(2n − 1) x0
By Theorem 14.6 and Corollary 14.7.
(14.26)
We continue with Theorem 14.9. Let p = 1, q = ∞ and y ∈ [x0 , b]. Then Z y (y − x0 )n+1 (n) 2 kf k∞,[x0 ,b] . kEn (x0 , w)k kf (n) (w)kdw ≤ (n + 1)! x0
(14.27)
We have by (14.3) that Z y (y − x0 )n 1 (y − t)n−1 kf (n) (t)kdt ≤ kf (n) k∞,[x0 ,b] kEn (x0 , y)k ≤ . (n − 1)! x0 n! (14.28) Consequently it holds Proof.
kEn (x0 , y)k kf (n)(y)k ≤ kf (n) k2∞,[x0 ,b]
(y − x0 )n , n!
(14.29)
all x0 ≤ y ≤ b. Hence we derive Z y kf (n) k2∞,[x0 ,b] Z y (n) kEn (x0 , w)k kf (w)kdw ≤ (w − x0 )n dw n! x0 x0 =
(y − x0 )n+1 (n) 2 kf k∞,[x0 ,b] . (n + 1)!
The counterpart of Theorem 14.9 follows. Theorem 14.10. Let p = 1, q = ∞ and y ∈ [a, x0 ]. Then
(14.30)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Vector Valued Functions
Z Proof.
x0 y
kEn (x0 , w)k kf (n) (w)kdw ≤
185
(x0 − y)n+1 (n) 2 kf k∞,[a,x0 ] . (n + 1)!
(14.31)
We have by (14.3) that
Z y
1 n−1 (n)
kEn (x0 , y)k = (y − t) f (t)dt
(n − 1)! x0
=
Z x 0
Z x0
1 1 n−1 (n)
≤ (y − t) f (t)dt (t−y)n−1 kf (n) (t)kdt (14.32)
(n − 1)! (n − 1)! y y kf (n) k∞,[a,x0 ] ≤ (n − 1)!
Z
x0
y
(t − y)n−1 dt = kf (n) k∞,[a,x0 ]
(x0 − y)n . n!
(14.33)
Hence it holds kEn (x0 , y)k kf (n) (y)k ≤ kf (n) k2∞,[a,x0 ]
(x0 − y)n , n!
(14.34)
all a ≤ y ≤ x0 . Consequently Z
x0 y
kEn (x0 , w)kf
(n)
(w)kdw ≤
kf (n) k2∞,[a,x0 ] Z
= kf (n) k2∞,[a,x0 ]
n!
x0 y
(x0 − w)n dw
(x0 − y)n+1 . (n + 1)!
(14.35)
Combining Theorems 14.9 and 14.10 we obtain Proposition 14.11. Let p = 1, q = ∞, y, x0 ∈ [a, b]. Then Z y |y − x0 |n+1 (n) 2 (n) ≤ kf k∞ . kE (x , w)k kf (w)kdw n 0 (n + 1)! x0
(14.36)
In particular we get
Proposition 14.12. Let p = 1, q = ∞, y, x0 ∈ [a, b]. Further suppose that f (k) (x0 ) = 0, k = 0, 1, ..., n − 1. Then Z y |y − x0 |n+1 (n) 2 (n) ≤ kf k∞ . (14.37) kf (w)k kf (w)kdw (n + 1)! x0
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
186
14.4
Applications
Here X = C, f ∈ C n ([a, b], C), n ∈ N, x0 , y ∈ [a, b]. Furthermore suppose that f (k) (x0 ) = 0, k = 0, 1, ..., n − 1. Applying Theorem 14.8 we obtain Theorem 14.13. Let p, q > 1 : Then Z ≤ min
1 p
+
1 q
= 1.
y x0
|f (w)| |f
(n)
(w)|dw
Z y 2/q 2 |y − x0 |n−1+ p (n) q |f (w)| dw , 1/q 1/p 2 (n − 1)![(p(n − 1) + 1)(p(n − 1) + 2)] x0 Z y ! |y − x0 |n p |f (n) (w)|2 dw . 2(n − 1)! n(2n − 1) x0
(14.38)
By Proposition 14.12 we derive
Proposition 14.14. Let p = 1, q = ∞. Then Z y |y − x0 |n+1 (n) 2 (n) ≤ kf k∞ . |f (w)| |f (w)|dw (n + 1)! x0
(14.39)
Let now f ∈ C 1 ([a, b], C), x0 , y ∈ [a, b] with f (x0 ) = 0. Applying Theorem 14.13 we get
Theorem 14.15. Let p, q > 1 : p1 + 1q = 1. Then Z y 0 |f (w)| |f (w)|dw x0
Z y 2/q Z y ! 1 |f 0 (w)|q dw , |y − x0 | |f 0 (w)|2 dw . ≤ min |y − x0 |2/p 2 x0 x0
(14.40)
Finally by applying Proposition 14.14 we find
Proposition 14.16. Let p = 1, q = ∞. Then Z y (y − x0 )2 0 2 0 ≤ |f (w)| |f (w)|dw kf k∞ . 2 x0
(14.41)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Chapter 15
Opial Inequalities for Semigroups
Various Lp form Opial type inequalities are given for semigroups with applications. This chapter relies on [31].
15.1
Introduction
The chapter is also greatly motivated by the article of Z. Opial ([195]): Theorem 15.1 (Opial [195]) Let x(t) ∈ C 1 ([0, h]) be such that x(0) = x(h) = 0, and x(t) > 0 in (0, h). The following inequality holds Z
h 0
|x(t)x0 (t)|dt ≤ (h/4)
Z
h
(x0 (t))2 dt.
(15.1)
0
In the last inequality the constant h/4 is the best possible. Opial type inequalities have applications in establishing uniqueness of solution to initial value problems in differential equations, see [244]. We are also motivated by [16], [21]. The generalized Opial type inequalities we prove here are in Theorems 15.8, 15.10 and 15.15.
15.2
Background
All this background comes from [79] (in general see also [147]). Let X a real or complex Banach space with elements f, g, · · · having norm kf k, kgk, · · · and let E(X) be the Banach algebra of endomorphisms of X. If T ∈ E(X), kT k denotes the norm of T . Definition 15.2. If T (t) is an operator function on the non- negative real axis 0 ≤ t < ∞ to the Banach algebra E(X) satisfying the following conditions: 187
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
188
(i) T (t1 + t2 ) = T (t1 )T (t2 ), (t1 , t2 ≥ 0), (ii) T (0) = I (I = identity operator),
(15.2)
then {T (t); 0 ≤ t < ∞} is called a one-parameter semi-group of operators in E(X). The semi-group {T (t); 0 ≤ t < ∞} is said to be of class C0 if it satisfies the further property (iii)s − lim T (t)f = f, (f ∈ X) t→0+
(15.3)
referred to as the strong continuity of T (t) at the origin. In this chapter we shall assume that the family of bounded linear operators {T (t); 0 ≤ t < ∞} mapping X to itself is a semi-group of class C0 , thus that all three conditions of the above definition are satisfied. Proposition 15.3. (a) kT (t)k is bounded on every finite subinterval of [0, ∞). (b) For each f ∈ X, the vector-valued function T (t)f on [0, ∞) is strongly continuous. Definition 15.4. The infinitesimal generator A of the semi-group {T (t); 0 ≤ t < ∞} is defined by Af = s − lim Aτ f, Aτ f = τ →0+
1 [T (τ ) − I] τ
(15.4)
whenever the limit exists; the domain of A, in symbols D(A), is the set of elements f for which the limit exists. Proposition 15.5. (a) D(A) is a linear manifold in X and A is a linear operator. (b) If f ∈ D(A), then T (t)f ∈ D(A) for each t ≥ 0 and d T (t)f = AT (t)f = T (t)Af dt
(t ≥ 0);
(15.5)
furthermore,
T (t)f − f =
Z
t
T (u)Af du
(t > 0).
(15.6)
0
(c) D(A) is dense in X, i.e. D(A) = X, and A is a closed operator. Definition 15.6. For r = 0, 1, 2, . . . the operator Ar is defined inductively by the relations A0 = I, A1 = A, and D(Ar ) = {f ; f ∈ D(Ar−1 ) and Ar−1 f ∈ D(A)} Ar f = A(Ar−1 f ) = s − lim Aτ (Ar−1 f ) (f ∈ D(Ar )). τ →0+
(15.7)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Semigroups
189
For the operator Ar and its domain D(Ar ) we have the following Proposition 15.7. (a) D(Ar ) is a linear subspace in X and Ar is a linear operator. (b) If f ∈ D(Ar ), so does T (t)f for each t ≥ 0 and dr T (t)f = Ar T (t)f = T (t)Ar f. dtr
(15.8)
Furthermore T (t)f −
r−1 k X t
k=0
k!
Ak f =
1 (r − 1)!
Z
t 0
(t − u)r−1 T (u)Ar f du,
(15.9)
the Taylor’s formula for semigroups. Additionally it holds Z tZ t Z t [T (t) − I]r = ··· T (u1 + u2 + · + ur )Ar f du1 du2 . . . dur . 0
0
(15.10)
0
r (c) D(Ar ) is dense in X for r = 1, 2 . . .; moreover, ∩∞ r=1 D(A ) is dense in X. A is a closed operator. Integrals in (15.9) and (15.10) are vector valued Riemann integrals, see [79], [164]. r
15.3
Results
Here always we consider T (t), A as in the Background of this chapter and f ∈ D(Ar ), r ∈ N. We present the first main result. Theorem 15.8. Let p, q > 1 : p1 + q1 = 1. One has Z t k∆(w)f k kT (w)Ar f kdw ≤ 0
2
tr−1+ p 21/q (r − 1)! [(p(r − 1) + 1)(p(r − 1) + 2)]1/p
for all t ∈ R+ = [0, ∞), where
∆(t)f := T (t)f −
Z
t 0
r−1 k X t
k=0
k!
kT (w)Ar f kq dw
Ak f.
We give Corollary 15.9. Let p = q = 2. Then Z
t 0
k∆(w)f k kT (w)Ar f kdw ≤
2/q
,
(15.11)
(15.12)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
190
2(r − 1)!
tr p
r(2r − 1)
Z
t r
0
2
kT (w)A f k dw , for all t ∈ R+ .
Proof of Theorem 15.8. By Taylor’s formula (15.9) we have Z t 1 (t − u)r−1 T (u)Ar f du. ∆(t)f = (r − 1)! 0 Therefore
k∆(t)f k = ≤ 1 (r − 1)!
≤
Z
1 k (r − 1)!
1 (r − 1)! t 0
Z
t 0
t
(15.14)
(t − u)r−1 T (u)Ar f duk
0
(t − u)r−1 kT (u)Ar f kdu
(t − u)p(r−1) du 1
=
Z
(15.13)
1/p Z
tr−1+ p (r − 1)! (p(r − 1) + 1)1/p
So far we have found that
Z
t 0
t 0
kT (u)Ar f kq du
kT (u)Ar f kq du
1/q
1/q
.
1
tr−1+ p 1/q (z(t)) , k∆(t)f k ≤ (r − 1)! (p(r − 1) + 1)1/p
where
z(t) := Clearly
Z
(15.15)
t 0
kT (u)Ar f kq du, for all t ∈ R+ .
z 0 (w) := kT (w)Ar f kq , 0 ≤ w ≤ t, z(0) = 0.
(15.16)
(15.17)
In particular kT (w)Ar f k = (z 0 (w))
1/q
.
Hence we have that 1
wr−1+ p 1/q k∆(w)f k kT (w)A f k ≤ (z(w)z 0 (w)) . (r − 1)! (p(r − 1) + 1)1/p r
Consequently it follows Z
Z
t 0 t
0
k∆(w)f k kT (w)Ar f kdw ≤ 1
wr−1+ p (z(w)z 0 (w))
1/q
1 (r − 1)! (p(r − 1) + 1)1/p
dw ≤
1 (r − 1)! (p(r − 1) + 1)1/p
(15.18)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Semigroups
191
Z t 1/q p 1/p Z t r−1+ p1 0 dw · z(w)z (w)dw w 0
0
2 r−1+ p
t 1 · = (r − 1)! (p(r − 1) + 1)1/p (p(r − 1) + 2)1/p 2
1/q z 2 (w) t 2 0
1 tr−1+ p 2/q = · (z(t)) (r − 1)! (p(r − 1) + 1)1/p 21/q (p(r − 1) + 2)1/p Z t 2/q 2 tr−1+ p 1 r q · = kT (u)A f k du , (r − 1)! (p(r − 1) + 1)1/p 21/q (p(r − 1) + 2)1/p 0 (15.19) proving the claim. We give Theorem 15.10. Here p = 1, q = ∞ and suppose w w w w w kT (u)Ar f k w := sup kT (u)Ar f k < ∞. ∞
ThenZ
t
k∆(w)f k kT (w)Ar f kdw ≤
0
where
w2 tr+1 w w w w ||T (u)Ar f k w , for all t ∈ R+ , (r + 1)! ∞
∆(t)f := T (t)f − We have
Proof.
(15.20)
u∈R+
r−1 k X t
k=0
k!
Ak f.
(15.21)
Z t 1 (t − u)r−1 kT (u)Ar f kdu (r − 1)! 0 Z t w w w w (t − u)r−1 du w kT (u)Ar f k w
k∆(t)f k ≤ ≤
That is
1 (r − 1)! 0 w tr w w w = w kT (u)Ar f k w , for all t ∈ R+ . r! ∞
k∆(t)f k ≤ Therefore Hence
w tr w w w w kT (u)Ar f k w , for all t ∈ R+ . r! ∞
k∆(w)f k kT (w)Ar f k ≤ Z
t 0
w2 wr w w w w kT (u)Ar f k w , all 0 ≤ w ≤ t. r! ∞
k∆(w)f k kT (w)Ar f kdw ≤
proving the claim.
∞
Z
t
0
wr dw
(15.22)
w w2 w w w kT (u)Ar f k w
∞
r!
w2 tr+1 w w w = w kT (u)Ar f k w , for all t ∈ R+ , (r + 1)! ∞
(15.23)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
192
We give Application 15.11 (see also [80]). It is known that the classical diffusion equation ∂2W ∂W = , −∞ < x < ∞, t > 0 ∂t ∂x2
(15.24)
lim W (x, t) = f (x),
(15.25)
with initial condition t→0+
has under general conditions its solution given by Z ∞ 2 1 f (x + u)e−u /4t du, W (x, t, f ) = [T (t)f ](x) = √ 2 πt −∞
(15.26)
the so called Gauss–Weierstrass singular integral. The infinitesimal generator of the semigroup {T (t); 0 ≤ t < ∞} is A = ∂ 2 /∂x2 ([159], p. 578). Here we suppose that f, f (2k) , k = 1, . . . , r, all belong to the Banach space U CB(R), the space of bounded and uniformly continuous functions from R into itself, with norm kf kC := sup |f (x)|.
(15.27)
x∈R
Here we define ∆(t)f (x) := W (x, t, f ) −
r−1 k X t
k=0
k!
f (2k) (x), for all x ∈ R.
(15.28)
Therefore by Theorem 15.8 we derive
Proposition 15.12. Let p, q > 1 : p1 + 1q = 1. Then Z t k∆(t)f kC kW (·, w, f (2r) )kC dw ≤ It := 0
2
tr−1+ p 21/q (r − 1)! [(p(r − 1) + 1)(p(r − 1) + 2)]1/p ·
Z
t 0
kW (·, w, f (2r) )kqC dw
2/q
, for all t ∈ R+ .
We make Remark 15.13. We notice here that 1 |W (x, t, f (2r) )| = √ 2 πt
Z
∞ −∞
f (2r) (x + u)e−u
2
/4t
du
(15.29)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Semigroups
1 ≤ √ 2 πt
Z
∞ −∞
|f (2r) (x + u)|e−u
1 ≤ kf (2r)kC √ 2 πt
Z
∞
e−u
2
2
193
/4t
/4t
du
du
−∞
= kf (2r) kC · 1 = kf (2r) kC < ∞, for all t ∈ R.
(15.30)
kW (·, t, f (2r) )kC ≤ kf (2r) kC .
(15.31)
That is
And thus we obtain
w w w w Φ := wkW (·, t, f (2r) )kC w . ∞
:= sup kW (·, t, f (2r) )kC ≤ kf (2r) kC < ∞.
(15.32)
t∈R+
We give Proposition 15.14. Here p = 1, q = ∞. Then tr+1 It ≤ Φ2 , for all t ∈ R+ . (r + 1)!
Proof.
(15.33)
By Theorem 15.10.
We finally get Theorem 15.15. Let p, q > 1 : Choose x∗t ∈ [0, rt] such that
1 p
+
1 q
= 1, t ≥ 0.
ρt := kT (x∗t )Ar f k = sup kT (x)Ar f k.
(15.34)
x∈[0,rt]
Then
Z
t 0
k(T (w) − I)r f k kT (rw)Ar f kdw
ρt tr+(1/p) ≤ (rp + 1)1/p Proof.
Z
t 0
r
q
kT (rw)A f k dw
Call Γ(t) := T (t) − I, t ≥ 0.
Then by (15.10) we have
1/q
.
(15.35)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
194
r
Γ (t)f =
Z tZ 0
t
···
0
Z
t
T (u1 + u2 + · + ur )Ar f du1 du2 . . . dur , t ≥ 0.
0
(15.36)
Therefore it follows that Z tZ t Z t kΓr (t)f k ≤ ··· kT (u1 + u2 + · + ur )Ar f kdu1 du2 . . . dur 0
0
Z t Z
≤ ·
Z t Z
=t
t
0
0
r/p
Z
···
0
0
Z
t
···
0 t
···
0
Z
t
1p du1 du2 . . . dur 0
t
r
q
kT (u1 + u2 + · + ur )A f k du1 du2 . . . dur
0
Z
1/p
t
r
q
1/q
kT (u1 + u2 + · + ur )A f k du1 du2 . . . dur
0
1/q
.
That is we derive
kΓr (t)f k ≤ tr/p t ≥ 0.
Z t Z 0
Set G(t1 , t2 , . . . , tr ) :=
Z
t
···
0
t1 0
Z
Z
t 0
1/q
,
(15.37)
t2
···
0
kT (u1 + u2 + · + ur )Ar f kq du1 du2 . . . dur
Z
tr 0
where
kT (u1 + u2 + · + ur )Ar f kq du1 du2 . . . dur ,
t1 , t2 , . . . , tr ≥ 0.
(15.38)
Thus kΓr (t)f k ≤ tr/p (G(t, t, . . . , t))
1/q
, t ≥ 0.
(15.39)
Clearly we find ∂ r G(t1 , t2 , . . . , tr ) = kT (t1 + t2 + . . . + tr )Ar f kq , all ti ≥ 0, i = 1, r, ∂t1 ∂t2 . . . ∂tr
(15.40)
and in particular we get ∂ r G(t, t, . . . , t) = kT (rt)Ar f kq , ∂t1 ∂t2 . . . ∂tr
(15.41)
and r
kT (rt)A f k = Consequently we have
∂ r G(t, t, . . . , t) ∂t1 ∂t2 . . . ∂tr
1/q
, for all t ≥ 0.
(15.42)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Semigroups
195
kΓr (t)f k kT (rt)Ar f k ≤ 1/q ∂ r G(t, t, . . . , t) r/p G(t, t, . . . , t) t , for all t ≥ 0. ∂t1 ∂t2 . . . ∂tr
Thus
Z
t 0
Z
t
kΓr (w)f k kT (rw)Ar f kdw ≤
0
wr/p G(w, w, . . . , w)
∂ r G(w, w, . . . , w) ∂t1 ∂t2 . . . ∂tr
Notice here that
G(w, w, . . . , w) =
Z
w 0
≤w
r
Z
(15.43)
w
···
0
Z
w 0
1/q
dw =: J
(15.44)
kT (u1 + u2 + · + ur )Ar f kq du1 du2 . . . dur
sup kT (x)Ar f kq = wr kT (x∗t )Ar f kq ,
(15.45)
x∈[0,rt]
for some x∗t ∈ [0, rt], 0 ≤ w ≤ t. That is G(w, w, . . . , w) ≤ wr kT (x∗t )Ar f kq .
Hence we derive
(15.46)
1/q ∂ r G(w, w, . . . , w) dw ∂t1 ∂t2 . . . ∂tr 0 r 1/q Z t ∂ G(w, w, . . . , w) = kT (x∗t )Ar f k wr dw ∂t1 ∂t2 . . . ∂tr 0
J ≤
Z
≤ kT (x∗t )Ar f k
t
wr kT (x∗t )Ar f k
Z
t
wrp dw
0
1/p Z t 0
= kT (x∗t )Ar f k
tr+(1/p) (rp + 1)1/p
We have established that Z
t 0
Z
∂ r G(w, w, . . . , w) ∂t1 ∂t2 . . . ∂tr
t
kT (rw)Ar f kq dw
0
dw
1/q
1/q
.
(15.47)
1/q
,
(15.48)
kΓr (w)f k kT (rw)Ar f kdw
tr+(1/p) ≤ kT (x∗t )Ar f k (rp + 1)1/p that is proving the claim.
Z
t 0
r
q
kT (rw)A f k dw
This page intentionally left blank
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Chapter 16
Opial Inequalities for Cosine and Sine Operator Functions
Various Lp form Opial type inequalities are given for Cosine and Sine Operator functions with applications. This chapter relies on [45]. 16.1
Introduction
This chapter is also greatly motivated by the article of Z. Opial [195]. Theorem 16.1. (Opial [195]) Let x(t) ∈ C 1 ([0, h]) be such that x(0) = x(h) = 0, and x(t) > 0 in (0, h). Then, Z Z h h h 0 0 (x (t))2 dt. (16.1) |x(t)x (t)|dt ≤ 4 0 0 In the last inequality the constant h/4 is the best possible. Opial type inequalities have applications in establishing uniqueness of solution to initial value problems in differential equations, see [244]. We are also motivated by [16], [21]. We need 16.2
Background
(see [93], [147], [191], [192]) Let (X, k·k) be a real or complex Banach space. By definition, a Cosine operator function is a family {C(t); t ∈ R} of bounded linear operators from X into itself, satisfying (i) C(0) = I, I the identity operator; (ii) C(t + s) + C(t − s) = 2C(t)C(s), ∀ t, s ∈ R; (the last product is composition ) (iii) C(·)f is continuous an R, ∀ f ∈ X. 197
(16.2)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
198
Notice that C(t) = C(−t) ∀ t ∈ R. The associated Sine operator function S(·) is defined by S(t)f :=
Z
t
∀ t ∈ R,
C(s)f ds, 0
∀ f ∈ X.
(16.3)
The Cosine operator function C(·) is such that kC(t)k ≤ M eωt , for some M ≥ 1, ω ≥ 0, ∀ t ∈ R+ , here k · k is the norm of the operator. The infinitesimal generator A of C(·) is the operator from X into itself defined as Af := lim
t→0+
2 (C(t) − I) f t2
(16.4)
with domain D(A). Operator A is closed and D(A) is dense in X, i.e. D(A) = X, and satisfies: Z t Z t S(s)f ds = C(t)f − f, ∀f ∈ X. (16.5) S(s)f ds ∈ D(A) and A 0
0
Also it holds A = C 00 (0), and D(A) is the set of f ∈ X : C(t)f is twice differentiable at t = 0 ; equivalently, D(A) = {f ∈ X : C(·)f ∈ C 2 (R, X)}.
(16.6)
If f ∈ D(A), then C(t)f ∈ D(A), and C 00 (t)f = C(t)Af = AC(t)f, ∀t ∈ R; C 0 (0)f = 0, see [140], [232]. We define A0 = I, A2 = A ◦ A, ..., An = A ◦ An−1 , n ∈ N. Let f ∈ D(An ), then C(t)f ∈ C 2n (R, X), and C (2n) (t)f = C(t)An f = An C(t)f, ∀t ∈ R, and C (2k−1) (0)f = 0, 1 ≤ k ≤ n, see [191]. For f ∈ D(An ), t ∈ R, we have the Cosine operator function’s Taylor formula ([191], [192]) saying that Tn (t)f := C(t)f −
n−1 X k=0
t2k k A f= (2k)!
Z
t 0
(t − s)2n−1 C(s)An f ds. (2n − 1)!
(16.7)
By integrating (16.7) we obtain the Sine operator function’s Taylor formula Mn (t)f := S(t)f − f t − Z
t 0
t2n−1 t3 Af − · · · − An−1 f = 3! (2n − 1)!
(t − s)2n C(s)An f ds, ∀t ∈ R, (2n)!
(16.8)
all f ∈ D(An ). Integrals in (16.7) and (16.8) are vector valued Riemann integrals, see [79], [164].
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Cosine and Sine Operator Functions
16.3
199
Results
Here, always, we consider C(t), S(t), A as in the Background and f ∈ D(An ), n ∈ N. Theorem 16.2. Let p, q > 1 : p1 + q1 = 1, t ∈ R+ . Then i) Z t I := kTn (w)f k kC(w)An f kdw ≤ 0
2
t2n−1+ p 1/q 2 (2n − 1)![(p(2n − 1) + 1)(p(2n − 1) + 2)]1/p
Z
t 0
kC(w)An f kq dw
2/q
,
(16.9)
and ii) J :=
Z
t 0
kMn (w)f k kC(w)An f kdw ≤ Z
1
t2(n+ p ) 2(2n)![(2pn + 1)(pn + 1)]1/p
t n
0
q
kC(w)A f k dw
2/q
.
(16.10)
We give Corollary 16.3. It holds i) I≤ and
Z
t2n p 2(2n − 1)! 2n(4n − 1)
t 0
kC(w)An f k2 dw, ∀t ∈ R+ ,
(16.11)
ii) t2n+1 p J≤ 2(2n)! (2n + 1)(4n + 1)
Z
t 0
kC(w)An f k2 dw, ∀t ∈ R+ .
(16.12)
Proof of Theorem 16.2. For convenience we call m := 2n − 1. By (16.7) we have Z t (t − s)m Tn (t)f = C(s)An f ds, t ∈ R+ . (16.13) m! 0 Thus
kTn (t)f k ≤ 1 m!
Z
t 0
1 m!
Z
(t − s)pm ds
t 0
(t − s)m kC(s)An f kds ≤
1/p Z
t 0
kC(s)An f kq ds
1/q
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
200
1
tm+ p = m!(pm + 1)1/p
Z
t 0
n
q
kC(s)A f k ds
1/q
.
(16.14)
So far we have found that 1
tm+ p (z(t))1/q , kTn (t)f k ≤ m!(pm + 1)1/p where z(t) :=
Z
(16.15)
t 0
kC(s)An f kq ds, ∀t ∈ R+ .
(16.16)
Clearly z 0 (w) = kC(w)An f kq , 0 ≤ w ≤ t, z(0) = 0.
(16.17)
kC(w)An f k = (z 0 (w))1/q .
(16.18)
Also
Therefore we have that 1
wm+ p (z(w)z 0 (w))1/q . m!(pm + 1)1/p
kTn (w)f k kC(w)An f k ≤ Consequently it holds Z t kTn (w)f k kC(w)An f kdw ≤ 0
≤
1 m!(pm + 1)1/p
Z
t
1 m!(pm + 1)1/p
wpm+1 dw
0
1/p Z 2
=
tm+ p 1 m!(pm + 1)1/p (pm + 2)1/p 2
tm+ p 1 = m!(pm + 1)1/p 21/q (pm + 2)1/p
Z
Z
t 0
t
(16.19)
1
wm+ p (z(w)z 0 (w))1/q dw
0
t
(16.20) z(w)z 0 (w)dw
0
z 2 (w) t 0 2 n
1/q
1/q
q
kC(w)A f k dw
2/q
(16.21)
,
proving (16.9). Setting now m := 2n and working similarly with (16.8) we get (16.10). The proof of the theorem is now complete. We present
(16.22)
Theorem 16.4. Here p = 1, q = ∞, and k kC(s)An f k k∞,+ := sup kC(s)An f k < ∞. s∈R+
Then
(16.23)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Opial Inequalities for Cosine and Sine Operator Functions
Book˙Adv˙Ineq
201
i) t2n+1 2 k kC(s)An f k k∞,+ , ∀t ∈ R+ , (2n + 1)!
(16.24)
t2(n+1) 2 k kC(s)An f k k∞,+ , ∀t ∈ R+ . (2(n + 1))!
(16.25)
I≤ and ii) J≤
We have from (16.7) that Z t t2n (t − s)2n−1 kC(s)An f kds ≤ k kC(s)An f k k∞,+ , ∀t ∈ R+ . kTn (t)f k ≤ (2n − 1)! (2n)! 0 (16.26) Thus Proof.
kTn (w)f k kC(w)An f k ≤
w2n k kC(s)An f k k2∞,+ (2n)!
(16.27)
all 0 ≤ w ≤ t. So that I= Z
t
w2n dw 0
Z
t 0
kTn (w)f k kC(w)An f kdw ≤
k kC(s)An f k k2∞ , + t2n+1 = k kC(s)An f k k2∞,+, (2n)! (2n + 1)!
(16.28)
proving (16.24). Similarly from (16.8) we obtain kMn (t)f k ≤
t2n+1 k kC(s)An f k k∞,+, ∀t ∈ R+ . (2n + 1)!
(16.29)
Furthermore it holds kMn (w)f k kC(w)An f k ≤ all 0 ≤ w ≤ t. Therefore Z t J= kMn (w)f k kC(w)An f kdw ≤
w2n+1 k kC(s)An f k k2∞,+, (2n + 1)!
(16.30)
t2(n+1) k kC(s)An f k k2∞,+, ∀t ∈ R+ . (2(n + 1))! 0 (16.31) That is establishing (16.25). The theorem is proved.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
202
We continue with Theorem 16.5. Let p, q > 1 : p1 + q1 = 1, t ∈ R− . Then i) Z 0 ∗ I := kTn (w)f k kC(w)An f kdw t
2
(−t)2n−1+ p ≤ 1/q 2 (2n − 1)![(p(2n − 1) + 1)(p(2n − 1) + 2)]1/p
Z
0 n
t
q
kC(w)A f k dw
2/q
,
(16.32)
and ii) J ∗ :=
Z
0 t
kMn (w)f k kC(w)An f kdw
1
≤
(−t)2(n+ p ) 2(2n)![(pn + 1)(2pn + 1)]1/p
Z
0
kC(w)An f kq dw
2/q
.
(16.33)
kC(w)An f k2 dw, ∀t ∈ R− ,
(16.34)
t
We give Corollary 16.6. It holds i) t2n p I ≤ 2(2n − 1)! 2n(4n − 1) ∗
and
Z
0 t
ii) J∗ ≤
(−t)2n+1 p 2(2n)! (2n + 1)(4n + 1)
Z
0 t
kC(w)An f k2 dw, ∀t ∈ R− .
Proof of Theorem 16.5. For convenience call m := 2n − 1. By (16.7) we have Z t (t − s)m C(s)An f ds, t ∈ R− . Tn (t)f = m! 0 Thus
Z t
1 m n
(t − s) C(s)A f ds kTn (t)f k =
m!
(16.35)
(16.36)
0
Z 0
Z 0
1 1 m n
= (t − s) C(s)A f ds ≤ (s − t)m kC(s)An f kds m! t m! t ≤
1 m!
Z
0
t
(s − t)pm ds
1/p Z
0
t
kC(s)An f kq ds
1/q
(16.37)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Opial Inequalities for Cosine and Sine Operator Functions
Z
1
1 (−t)m+ p = m! (pm + 1)1/p So far we have found that
0 t
n
q
kC(s)A f k ds
1/q
203
.
(16.38)
1
(−t)m+ p (z(t))1/q , kTn (t)f k ≤ m!(pm + 1)1/p where z(t) := That is
Z
0 t
−z(t) = and
(16.39)
kC(s)An f kq ds, z(0) = 0.
(16.40)
Z
(16.41)
t 0
kC(s)An f kq ds ≥ 0,
−z 0 (t) = kC(t)An f kq ≥ 0,
(16.42)
kC(t)An f k = (−z 0 (t))1/q , t ∈ R−
(16.43)
and
Hence we have 1
kTn (w)f k kC(w)An f k ≤
(−w)m+ p (z(w)(−z 0 (w)))1/q . m!(pm + 1)1/p
(16.44)
Integrating the last inequality we derive Z 0 kTn (w)f k kC(w)An f kdw t
≤ ≤
1 m!(pm + 1)1/p
1 m!(pm + 1)1/p
Z
0
Z
0
1
(−w)m+ p (z(w)(−z 0 (w))1/q dw
(16.45)
t
(−w)pm+1 dw
t
1/p Z
0
z(w)(−z 0 (w))dw
t
1/q
(16.46)
2
1 (−t)m+ p (z(t))2/q = m!(pm + 1)1/p (pm + 2)1/p 21/q 2
=
(−t)m+ p 21/q m![(pm + 1)(pm + 2)]1/p
Z
0 t
kC(w)An f kq dw
(16.47) 2/q
,
proving (16.32). Setting now m := 2n and working similarly with (16.8) we get (16.33). The proof of the theorem is now complete. Combining Theorems 16.2 and 16.5 we derive Proposition 16.7. Let p, q > 1 :
1 p
+
1 q
= 1, t ∈ R. Then
(16.48)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
204
i) Z t kTn (w)f k kC(w)An f kdw I¯ := 0
and
Z t 2/q 2 |t|2n−1+ p n q , ≤ 1/q kC(w)A f k dw 2 (2n − 1)![(p(2n − 1) + 1)(p(2n − 1) + 2)]1/p 0 (16.49) ii) Z t n ¯ J := kMn (w)f k kC(w)A f kdw 0
≤
Z t 2/q 1 |t|2(n+ p ) n q . kC(w)A f k dw 1/p 2(2n)![(pn + 1)(2pn + 1)] 0
(16.50)
Combining Corollaries 16.3 and 16.6 we obtain Corollary 16.8. It holds i) I¯ ≤ and ii) J¯ ≤
Z t t2n n 2 p kC(w)A f k dw , ∀t ∈ R, 2(2n − 1)! 2n(4n − 1) 0 Z t |t|2n+1 n 2 p kC(w)A f k dw , ∀t ∈ R. 2(2n)! (2n + 1)(4n + 1) 0
(16.51)
(16.52)
Next we present
Theorem 16.9. Here p = 1, q = ∞, and k kC(s)An f k k∞,− := sup kC(s)An f k < ∞. s∈R−
Then i) I∗ ≤
(−t)2n+1 k kC(s)An f k k2∞,− , ∀t ∈ R− , (2n + 1)!
(16.53)
t2(n+1) k kC(s)An f k k2∞,− , ∀t ∈ R− . (2(n + 1))!
(16.54)
and ii) J∗ ≤
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Opial Inequalities for Cosine and Sine Operator Functions
Proof.
Thus
For convenience call m := 2n − 1. By (16.7) we have Z t (t − s)m C(s)An f ds, t ∈ R− . Tn (t)f = m! 0
Z 0
1
(t − s)m C(s)An f ds
m! t Z 0 1 ≤ (s − t)m kC(s)An f kds m! t Z 0 (−t)m+1 m . (s − t) ds = k kC(s)An f k k∞,− (m + 1)! t
Book˙Adv˙Ineq
205
(16.55)
kTn (t)f k =
k kC(s)An f k k∞,− m! Hence ≤
kTn (t)f k ≤ k kC(s)An f k k∞,−
(−t)m+1 , ∀t ∈ R− . (m + 1)!
(16.56) (16.57)
(16.58)
So that it holds kTn (w)f k kC(w)An f k ≤ k kC(s)An f k k2∞,−
(−t)m+1 , t≤w≤0 (m + 1)!
(16.59)
Consequently we get Z Z 0 k kC(s)An f k k2∞,− 0 n ∗ (−w)m+1 dw (16.60) kTn (w)f k kC(w)A f kdw ≤ I = (m + 1)! t t =
(−t)m+2 k kC(s)An f k k2∞,− (m + 2)!
proving (16.53). Similarly, from (16.8), for m = 2n we obtain (16.54).
(16.61)
Finally we give Proposition 16.10. Here p = 1, q = ∞, and
k kC(s)An f k k∞ := sup kC(s)An f k < ∞.
(16.62)
s∈R
Then i) Z t 2n+1 n n 2 ≤ |t| kT (w)f k kC(w)A f kdw n (2n + 1)! k kC(s)A f k k∞ , ∀t ∈ R, 0
(16.63)
and
ii) Z t t2(n+1) n k kC(s)An f k k2∞ , ∀t ∈ R. kMn (w)f k kC(w)A f kdw ≤ (2(n + 1))! 0
Proof.
By Theorems 16.4, 16.9.
(16.64)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
206
16.4
Book˙Adv˙Ineq
Applications
(see [147], p. 121) Let X be the Banach space of odd, 2π-periodic real functions in the space of d2 bounded uniformly continuous functions from R into itself : BU C(R). Let A := dx 2 with D(An ) = {f ∈ X : f (2k) ∈ X , k = 1, ..., n}, n ∈ N. A generates a Cosine functions C ∗ given by 1 [f (x + t) + f (x − t)] , ∀x, t ∈ R. 2 The corresponding Sine function S ∗ is given by Z t Z t 1 S ∗ (t)f (x) = f (x + s)ds + f (x − s)ds , ∀x, t ∈ R. 2 0 0 C ∗ (t)f (x) =
(16.65)
(16.66)
Here we consider f ∈ D(An ), n ∈ N, as above. By (16.7) we obtain n−1
Tn∗ (t)f :=
X t2k 1 [f (· + t) + f (· − t)] − f (2k) 2 (2k)! k=0
=
Z
t 0
i (t − s)2n−1 h (2n) f (· + s) + f (2n) (· − s) ds, ∀t ∈ R. 2(2n − 1)!
(16.67)
By (16.8) we get Mn∗ (t)f
1 := 2
=
Z
Z t
0
t
f (· + s)ds + 0
Z
t 0
f (· − s)ds −
n X
k=1
t2k−1 f (2(k−1)) (2k − 1)!
i (t − s)2n h (2n) f (· + s) + f (2n) (· − s) ds, ∀t ∈ R. 2(2n)!
(16.68)
Let g ∈ BU C(R), we define kgk = kgk∞ := sup |g(x)| < ∞. x∈R
Notice also that k kC ∗ (s)An f k∞ k∞ = k kC ∗ (s)f (2n) k∞ k∞ = ≤ where
1 k k(f (2n) (· + s) + f (2n) (· − s))k∞ k∞ 2
i 1h k kf (2n)(· + s)k∞ k∞ + k kf (2n) (· − s)k∞ k∞ ≤ Θ < ∞, 2 Θ := kf (2n) k∞ .
We have the following applications. From Proposition 16.7 we derive Proposition 16.11. Let p, q > 1 :
1 p
+
1 q
= 1, t ∈ R. Then
(16.69)
(16.70)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Opial Inequalities for Cosine and Sine Operator Functions
Book˙Adv˙Ineq
207
i) Z t
∗ (2n) ∗ ε1 := kTn (w)f k∞ C (w)f
dw ∞ 0
≤
2/q Z t 2
q |t|2n−1+ p
∗ (2n) , dw (w)f
C
1/q 1/p ∞ 2 (2n − 1)![(p(2n − 1) + 1)(p(2n − 1) + 2)] 0 ∀t ∈ R,
(16.71)
and ii) Z t
∗ ∗ (2n) ε2 := kMn (w)f k∞ C (w)f
dw ∞ 0
Z t 2/q 1
q |t|2(n+ p )
∗ (2n) ≤
C (w)f
dw , ∀t ∈ R. 1/p 2(2n)![(pn + 1)(2pn + 1)] ∞ 0
(16.72)
From Corollary 16.8 we get Corollary 16.12. It holds i)
and ii)
Z t
2 t2n
∗ (2n) p ε1 ≤
C (w)f
dw , ∀t ∈ R, ∞ 2(2n − 1)! 2n(4n − 1) 0
Z t
2 |t|2n+1
∗ (2n) p ε2 ≤
C (w)f
dw , ∀t ∈ R. ∞ 2(2n)! (2n + 1)(4n + 1) 0
(16.73)
(16.74)
Finally from Proposition 16.10 we find
Proposition 16.13. Here p = 1, q = ∞. Then i)
|t|2n+1
(2n) 2 ε1 ≤
f
, ∀t ∈ R, (2n + 1)! ∞
(16.75)
and
ii)
ε2 ≤
t2(n+1)
(2n) 2
f
, ∀t ∈ R. (2(n + 1))! ∞
(16.76)
This page intentionally left blank
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Chapter 17
Poincar´ e Like Inequalities for Linear Differential Operators
Various Lp form Poincar´e like inequalities [2], forward and reverse, are given for a linear differential operator L, involving its related initial value problem solution y, Ly, the associated Green’s function H and initial conditions at point x0 ∈ R. This chapter follows [48].
17.1
Background
Here we use [169], pp. 145–154. Let [a, b] ⊂ R, ai (x) , i = 0, 1, . . . , n − 1 (n ∈ N) , h (x) be continuous functions on [a, b] and let L = D n + an−1 (x) Dn−1 + . . . + a0 (x) be a fixed linear differential operator on C n ([a, b]) . Let y1 (x) , . . . , yn (x) be a set of linear independent solutions to Ly = 0. Here the associated Green’s functions for L is y1 (t) . . . yn (t) y 0 (t) . . . y 0 (t) 1 n ... ... ... (n−2) (n−2) y1 (t) . . . yn (t) y1 (x) . . . yn (x) , (17.1) H (x, t) := y1 (t) . . . yn (t) y 0 (t) . . . y 0 (t) 1 n ... ... ... (n−2) (n−2) y1 (t) . . . yn (t) (n−1) (n−1) y (t) . . . yn (t) 1
2
which is a continuous function on [a, b] . Consider a fixed x0 ∈ [a, b] , then Z x y (x) = H (x, t) h (t) dt, ∀x ∈ [a, b]
(17.2)
x0
is the unique solution to the initial value problem
Ly = h; y (i) (x0 ) = 0, i = 0, 1, . . . , n − 1. 209
(17.3)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
210
Next we suppose all of the above.
17.2
Results
We present the following Poincar´e like inequalities. Theorem 17.1. Let x0 < b and x ∈ [x0 , b] , and p, q > 1 : Then !1/ν Z Z b
1) kykLν (x0 ,b) ≤
x
x0
1 q
+
= 1; ν > 0.
ν/p
p
|H (x, t)| dt
x0
1 p
dx
kLykLq (x0 ,b) .
(17.4)
kLykLq (x0 ,b) .
(17.5)
When ν = q we have 2) kykLq (x0 ,b) ≤
Z
Z
b x0
x
p
|H (x, t)| dt
x0
q/p
dx
!1/q
When ν = p = q = 2 we obtain Z
3) kykL2 (x0 ,b) ≤ Proof.
b x0
Z
x
2
H (x, t) dt dx x0
!1/2
kLykL2 (x0 ,b) .
(17.6)
From (17.2) we have |y (x)| ≤
≤
Z
≤
Z
x x0
x x0
|H (x, t)| |h (t)| dt
|H (x, t)|p dt
x x0
Z
p
|H (x, t)| dt
1/p Z
1/p
Z
x x0
|h (t)|q dt
b x0
q
|h (t)| dt
1/q
!1/q
.
(17.7)
kLykLq (x0 ,b) ,
(17.8)
That is ν
|y (x)| ≤ Hence Z b x0
ν
|y (x)| dx ≤
proving the claim.
Z
Z b
x0
x x0
Z
p
|H (x, t)| dt
x x0
p
ν/p
|H (x, t)| dt
ν/p
ν
!
ν
dx kLykLq (x0 ,b) ,
(17.9)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e Like Inequalities for Linear Differential Operators
211
We continue with Theorem 17.2. Let x0 > a and x ∈ [a, x0 ] , and p, q > 1 : p1 + 1q = 1; ν > 0. Then ν/p !1/ν Z x0 Z x0 p kLykLq (a,x0 ) . (17.10) dx |H (x, t)| dt 1) kykLν (a,x0 ) ≤ x
a
When ν = q we obtain Z x0 Z 2) kykLq (a,x0 ) ≤ a
x0
p
|H (x, t)| dt
x
q/p
dx
!1/q
kLykLq (a,x0 ) .
When ν = p = q = 2 we obtain 1/2 Z x0 Z x0 2 H (x, t) dt dx 3) kykL2 (a,x0 ) ≤ kLykL2 (a,x0 ) . From (17.2) we have
|y (x)| ≤ ≤ ≤ That is
Z
Z
x0
Z
x0
p
ν
|y (x)| ≤
a
Z
|H (x, t)| |h (t)| dt
x
|H (x, t)| dt
x
Therefore Z x0 |y (x)|ν dx ≤
x0
|H (x, t)|p dt
x
(17.12)
x
a
Proof.
(17.11)
Z
x0 a
x0
1/p Z p
|H (x, t)| dt
x
Z
1/p Z
x0
p
x0 x x0
a
q
|h (t)| dt
ν/p
|H (x, t)| dt
x
|h (t)|q dt
1/q
1/q
.
(17.13)
kLykLq (a,x0 ) .
(17.14)
ν
ν/p
!
dx kLykνLq (a,x0 ) ,
proving the claim.
(17.15)
Extreme cases follow Proposition 17.3. Here x0 < b, x ∈ [x0 , b] , and p = 1, q = ∞. Then ν !1/ν Z b Z x 1) kykLν (x0 ,b) ≤ kLykL∞ (x0 ,b) . |H (x, t)| dt dx x0
(17.16)
x0
When ν = 1 we have
2) kykL1 (x0 ,b) ≤
Z
b x0
Z
x x0
!
|H (x, t)| dt dx kLykL∞ (x0 ,b) .
(17.17)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
212
From (17.2) we have
Proof.
|y (x)| ≤ Z
≤ That is
x x0
Z
ν
|y (x)| ≤
Z
x x0
|H (x, t)| |h (t)| dt
|H (x, t)| dt khkL∞ (x0 ,b) . x
|H (x, t)| dt
x0
ν
ν
kLykL∞ (x0 ,b) ,
and Z
b x0
Z
ν
|y (x)| dx ≤
Z
b x0
x x0
|H (x, t)| dt
ν
(17.18)
!
ν
dx kLykL∞ (x0 ,b) ,
proving the claim.
(17.19)
(17.20)
We continue with Proposition 17.4. Here x0 > a, x ∈ [a, x0 ] , and p = 1, q = ∞. Then ν 1/ν Z x0 Z x0 |H (x, t)| dt dx 1) kykLν (a,x0 ) ≤ kLykL∞ (a,x0 ) .
(17.21)
x
a
When ν = 1 we derive
2) kykL1 (a,x0 ) ≤
Z
Z
x0 a
x0 x
|y (x)| ≤ ≤
Z
That is ν
|y (x)| ≤ Z
(17.22)
From (17.2) we have
Proof.
and
|H (x, t)| dt dx kLykL∞ (a,x0 ) .
x0 a
ν
|y (x)| dx ≤
proving the claim.
Z
x
Z
x0 x
|H (x, t)| |h (t)| dt
x0
x0 a
Z
|H (x, t)| dt khkL∞ (a,x0 ) . x0
|H (x, t)| dt
x
Z
ν
x0 x
|H (x, t)| dt
ν
kLykL∞ (a,x0 ) ,
ν
ν dx kLykL∞ (a,x0 ) ,
(17.23)
(17.24)
(17.25)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e Like Inequalities for Linear Differential Operators
213
Next we give reverse Poincar´e like inequalities. Theorem 17.5. Let x0 < b, x ∈ [x0 , b] , and 0 < p < 1, q < 0 : p1 + q1 = 1, ν > 0. Suppose H (x, t) ≥ 0 for x0 ≤ t ≤ x, and Ly = h is of fixed sign and nowhere zero. Then !1/ν Z Z x
b
1) kykLν (x0 ,b) ≥
ν/p
p
kLykLq (x0 ,b) .
dx
(H (x, t)) dt
x0
x0
(17.26)
When ν = p we obtain Z
2) kykLp (x0 ,b) ≥
b x0
Z
(H (x, t)) dt dx
!1/p
1/p
!
x
p
x0
kLykLq (x0 ,b) .
(17.27)
dx kLykLq (x0 ,b) .
(17.28)
When ν = 1 we get Z
3) kykL1 (x0 ,b) ≥ Proof.
b x0
Z
x
p
(H (x, t)) dt x0
By (17.2) we have Z x |y (x)| = H (x, t) |h (t)| dt, for all x0 ≤ x ≤ b.
(17.29)
x0
From (17.29) by reverse H¨ older’s inequality we find |y (x)| ≥
≥
Z
Z
x
p
(H (x, t)) dt x0
x
p
(H (x, t)) dt x0
1/p Z
1/p
Z
x x0
b x0
q
|h (t)| dt
q
|h (t)| dt
1/q
!1/q
,
(17.30)
khkνLq (x0 ,b) ,
(17.31)
for all x0 < x ≤ b. That is it holds |y (x)|ν ≥ for all x0 ≤ x ≤ b, and Z b ν |y (x)| dx ≥ x0
proving the claim.
Z
b x0
Z
x
(H (x, t))p dt
x0
Z
x
p
ν/p
(H (x, t)) dt x0
ν/p
!
ν
dx khkLq (x0 ,b) ,
(17.32)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
214
We continue with Theorem 17.6. Let x0 > a, x ∈ [a, x0 ] , and 0 < p < 1, q < 0 : p1 + q1 = 1, ν > 0. Suppose H (x, t) ≤ 0 for x ≤ t ≤ x0 , and Ly = h is of fixed sign and nowhere zero. Then ν/p !1/ν Z x0 Z x0 p kLykLq (a,x0 ) . (17.33) dx (−H (x, t)) dt 1) kykLν (a,x0 ) ≥ x
a
When ν = p we obtain Z x0 Z 2) kykLp (a,x0 ) ≥ a
When ν = 1 we have Z 3) kykL1 (a,x0 ) ≥ Proof.
x0 a
Z
x0 x
1/p p kLykLq (a,x0 ) . (−H (x, t)) dt dx
x0
p
(−H (x, t)) dt x
From (17.2) we have
Z |y (x)| =
x x0
1/p
!
dx kLykLq (a,x0 ) .
(17.34)
(17.35)
H (x, t) h (t) dt
Z x0 H (x, t) h (t) dt = x Z x 0 = (−H (x, t)) h (t) dt x
=
Z
x0
(−H (x, t)) |h (t)| dt.
x
(17.36)
From (17.36) by reverse H¨ older’s inequality we find Z x0 1/p Z x0 1/q p q |y (x)| ≥ (−H (x, t)) dt |h (t)| dt x
≥ for all a ≤ x < x0 . That is it holds
Z
x0
(−H (x, t))p dt
x
ν
|y (x)| ≥ for all a ≤ x ≤ x0 , and Z x0 ν |y (x)| dx ≥ a
proving the claim.
x
Z
Z
x0 a
x0
1/p Z
p
(−H (x, t)) dt x
Z
x0
p
x0 a
ν/p
(−H (x, t)) dt x
|h (t)|q dt
1/q
ν
kLykLq (a,x0 ) ,
ν/p
!
ν
dx kLykLq (a,x0 ) ,
(17.37)
(17.38)
(17.39)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Chapter 18
Poincar´ e and Sobolev Like Inequalities for Widder Derivatives
Various Lp form Poincar´e and Sobolev like inequalities, forward and reverse, are given involving Widder derivative ([243]). This chapter follows [54]. 18.1
Background
The following come from [243]. Let f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]) , n ≥ 0, and the Wronskians Wi (x) := W [u0 (x) , u1 (x) , . . . , ui (x)] := u0 (x) u1 (x) . . . ui (x) u0 (x) u0 (x) . . . u0 (x) 0 1 i , .. . (i) (i) u0 (x) u(i) (x) . . . u (x) 1 i
(18.1)
i = 0, 1, . . . , n. Here W0 (x) = u0 (x) . Assume Wi (x) > 0 over [a, b] , i = 0, 1, . . . , n. For i ≥ 0, the differential operator of order i (Widder derivative): Li f (x) :=
W [u0 (x) , u1 (x) , . . . , ui−1 (x) , f (x)] , Wi−1 (x)
i = 1, . . . , n + 1; L0 f (x) := f (x) , ∀x ∈ [a, b] . Consider also u0 (t) u1 (t) u0 (t) u01 (t) 0 1 .. gi (x, t) := . Wi (t) (i−1) (i−1) u0 (t) u1 (t) u0 (x) u1 (x) i = 1, 2, . . . , n; g0 (x, t) :=
u0 (x) u0 (t) ,
∀x, t ∈ [a, b] .
, (i−1) . . . ui (t) . . . ui (x)
... ...
Example ([243]). Sets of the form {u0 , u1 , . . . , un } are 215
(18.2)
ui (t) u0i (t)
(18.3)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
216
n
Book˙Adv˙Ineq
1, x, x2 , . . . , xn ,
1, sin x, − cos x, − sin 2x, cos 2x, . . . , (−1)
n−1
o n sin nx, (−1) cos nx , etc.
We also mention the generalized Widder–Taylor’s formula, see [243].
Theorem 18.1. Let the functions f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]) , and the Wronskians W0 (x) , W1 (x) , . . . , Wn (x) > 0 on [a, b] , x ∈ [a, b] . Then for t ∈ [a, b] we have f (x) = f (t)
u0 (x) + L1 f (t) g1 (x, t) + . . . + Ln f (t) gn (x, t) + Rn (x) , u0 (t)
where Rn (x) :=
Z
(18.4)
x
gn (x, s) Ln+1 f (s) ds.
(18.5)
t
For example ([243]) one could take u0 (x) = c > 0. If ui (x) = xi , i = 0, 1, . . . , n, defined on [a, b] , then Li f (t) = f (i) (t) and gi (x, t) =
(x − t)i , t ∈ [a, b] . i!
(18.6)
We need Corollary 18.2. (to Theorem 18.1) Additionally suppose that for fixed x 0 ∈ [a, b] we have Li f (x0 ) = 0, i = 0, 1, . . . , n. Then Z x f (x) = gn (x, t) Ln+1 f (t) dt, ∀x ∈ [a, b] . (18.7) x0
Corollary 18.3. (to Theorem 18.1) Let f, u0 ∈ C 1 ([a, b]) , u0 (x) > 0, for all x ∈ [a, b] . Then Z x u0 (x) L1 f (s) + u0 (x) ds, ∀x, t ∈ [a, b] , (18.8) f (x) = f (t) u0 (t) u0 (s) t where L1 f (s) =
W [u0 (s) , f (s)] u0 (s) f 0 (s) − u00 (s) f (s) = . u0 (s) u0 (s)
(18.9)
We need to make Remark 18.4. We define (see [243]) φ0 (x) := W0 (x) , φ1 (x) :=
W1 (x) (W0 (x))
2,...,
in general φk (x) :=
Wk (x) Wk−2 (x) (Wk−1 (x))2
, k = 2, 3, . . . , n.
(18.10)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Poincar´ e and Sobolev Like Inequalities for Widder Derivatives
Book˙Adv˙Ineq
217
The functions φi (x) are positive on [a, b] . According to [243] we get, for any x, x0 ∈ [a, b] that Z x Z x1 φ0 (x) φ1 (x1 ) gn (x, x0 ) = ... φ0 (x0 ) . . . φn (x0 ) x0 x0 Z xn−2 Z xn−1 φn−1 (xn−1 ) φn (xn ) dx1 dx2 . . . dxn x0
=
x0
1 φ0 (x0 ) . . . φn (x0 )
Z
x
φ0 (s) . . . φn (s) gn−1 (x, s) ds.
(18.11)
x0
We get that gn (x, x) = 0, all x ∈ [a, b] , and gn (x, x0 ) > 0, x > x0 , x, x0 ∈ [a, b] , any n ∈ N. Also g0 (x, x0 ) > 0 for any x, x0 ∈ [a, b] . By (18.11) we notice that gn (x, t) < 0, x < t, n odd, gn (x, t) > 0, x < t, n even,
(18.12)
where x, t ∈ [a, b] . In the next we work under the terms and assumptions of Theorem 18.1 and Corollary 18.2, and the rest of the above conclusions. 18.2
Results
We give the following weighted Neumann–Poincar´e like inequality. Theorem 18.5. Let p, q > 1 : p1 + u0 > 0 on [a, b] . Then
Z b
f 1 f (t)
1) − dt
u0 b − a a u0 (t)
1 q
Lν (a,b)
When ν = q we obtain
Z b
f f (t) 1
2) − dt
u0 b − a a u0 (t)
When ν = p = q = 2 we have
Z b
f 1 f (t)
3) − dt
u0 b − a a u0 (t)
= 1, ν > 0. Consider f, u0 ∈ C 1 ([a, b]) ,
≤ (b − a)( p
Lq (a,b)
L2 (a,b)
Equivalently,
Z
b a
1
f (x) 1 − u0 (x) b − a
+ ν1 )
L1 f
.
u0 Lq (a,b)
L1 f
≤ (b − a) .
u0 Lq (a,b)
L1 f
. ≤ (b − a)
u0 L2 (a,b)
Z
b a
f (t) dt u0 (t)
!2
dx
(18.13)
(18.14)
(18.15)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
218
≤ (b − a) Proof.
1 b−a
Let A = f (c) u0 (c)
Rb
f (t) dt. a u0 (t)
Z
2
b a
L1 f (x) u0 (x)
f (c) f (x) − = u0 (x) u0 (c) i.e.
Z
f (x) −A = u0 (x) Z f (x) ≤ − A u0 (x) ≤ (b − a) Therefore
Z
1/p
b
a
dx.
(18.15*)
By the intermediate value theorem there exists
c ∈ [a, b] with = A. Thus by (18.8) we get
Hence
2
x c
x c
Z
x
L1 f (s) ds, u0 (s)
c
(18.16)
L1 f (s) ds, ∀x ∈ [a, b] . u0 (s)
(18.17)
Z b |L1 f (s)| |L1 f (s)| ds ≤ ds u0 (s) u0 (s) a
|L1 f (s)| u0 (s)
ν f (x) ν p u0 (x) − A ≤ (b − a)
q
ds
!1/q
, ∀x ∈ [a, b] .
L1 f ν
, ∀x ∈ [a, b] .
u0 Lq (a,b)
Consequently we obtain
ν
ν Z b
f (x) ν dx ≤ (b − a)( p +1) L1 f , − A
u0 u0 (x) a Lq (a,b)
proving the claim.
Corollary 18.6. (to Theorem 18.5) Let
f
1)
u0
Lν (a,b)
f
2)
u0
equivalently, Z
b a
f (x) u0 (x)
L2 (a,b)
2
It follows the related result.
f (t) dt a u0 (t)
L1 f
≤ (b − a) u0
dx ≤ (b − a)
Z
b
a
(18.20)
,
(18.21)
Lq (a,b)
L1 f
≤ (b − a) ,
u0 Lq (a,b)
2
(18.19)
= 0. Then
1 1 L1 f ( p+ν )
≤ (b − a)
u0
Lq (a,b)
f
3)
u0
Rb
(18.18)
(18.22)
,
(18.23)
L2 (a,b)
L1 f (x) u0 (x)
2
dx.
(18.23*)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Poincar´ e and Sobolev Like Inequalities for Widder Derivatives
Book˙Adv˙Ineq
219
Proposition 18.7. Let ν > 0 and consider f, u0 ∈ C 1 ([a, b]) , u0 > 0 on [a, b] . Then
Z b
f f (t) 1 L1 f
1/ν
1) − dt ≤ (b − a) . (18.24)
u0 b − a a u0 (t) u0 L1 (a,b) Lν (a,b)
When ν = 1 we have
Z b
f 1 f (t)
dt 2) −
u0 b − a a u0 (t)
Proof.
L1 (a,b)
L1 f
≤ (b − a) . u0 L1 (a,b)
By (18.18) we have ν Z b f (x) 1 f (t) − dt u0 (x) b − a a u0 (t)
L 1 f ν
≤ , ∀x ∈ [a, b] . u0 L1 (a,b)
(18.25)
(18.26)
So that ν
Z b Z b
L 1 f ν 1 f (t) f (x)
− dt dx ≤ (b − a) , b − a a u0 (t) u0 L1 (a,b) a u0 (x)
proving the claim.
(18.27)
We continue with generalized Dirichlet–Poincar´e like inequalities. Theorem 18.8. Let f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]) , n ∈ Z+ ; W0 , W1 , . . . , Wn > 0 on [a, b] , and for fixed x0 ∈ [a, b] suppose that Li f (x0 ) = 0, i = 0, 1, . . . , n. Let p, q > 1 : p1 + 1q = 1, ν > 0. Then ν/p !1/ν Z b Z x p 1) kf kLν (x0 ,b) ≤ kLn+1 f kLq (x0 ,b) . (18.28) dx (gn (x, t)) dt x0
2) kf kLq (x0 ,b) ≤
Z
b x0
x0
Z
x
p
(gn (x, t)) dt x0
q/p
dx
!1/q
kLn+1 f kLq (x0 ,b) .
(18.29)
Also, 3) kf kL2 (x0 ,b) ≤ Proof.
Z
b x0
Z
x
2
(gn (x, t)) dt dx x0
!1/2
kLn+1 f kL2 (x0 ,b) .
By (18.7) we have Z |f (x)| =
x x0
gn (x, t) Ln+1 f (t) dt
(18.30)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
220
Z
≤ Z
≤ ≤
Z
x
x
p
(gn (x, t)) dt x0
|f (x)| ≤ and
x0
gn (x, t) |Ln+1 f (t)| dt
x0
ν
b
x0
(gn (x, t))p dt
That is
Z
x
Z
ν
|f (x)| dx ≤
b x0
Z
1/p
1/p Z
x0
|Ln+1 f (t)|q dt
1/q
kLn+1 f kLq (x0 ,b) , ∀x ∈ [x0 , b] .
x
p
(gn (x, t)) dt x0
Z
x
x
ν/p
p
(gn (x, t)) dt x0
ν/p
ν
kLn+1 f kLq (x0 ,b) , !
ν
dx kLn+1 f kLq (x0 ,b) ,
proving the claim.
(18.31)
(18.32)
(18.33)
We give Theorem 18.9. Same assumptions as in Theorem 18.8. Then ν/p !1/ν Z x0 Z x0 p 1) kf kLν (a,x0 ) ≤ |gn (x, t)| dt dx kLn+1 f kLq (a,x0 ) . (18.34) a
2) kf kLq (a,x0 ) ≤
Z
x
x0 a
Z
x0
p
|gn (x, t)| dt
x
q/p
dx
!1/q
kLn+1 f kLq (a,x0 ) . (18.35)
Also, 3) kf kL2 (a,x0 ) ≤ Proof.
Z
x0 a
Z
1/2 2 (gn (x, t)) dt dx kLn+1 f kL2 (a,x0 ) .
x0 x
By (18.7) we have
Z |f (x)| = Z =
≤ ≤
Z
x0 x
Z
x0
x
x0
x
x x0
gn (x, t) Ln+1 f (t) dt
gn (x, t) Ln+1 f (t) dt
|gn (x, t)| |Ln+1 f (t)| dt
|gn (x, t)|p dt
1/p Z
x0 x
|Ln+1 f (t)|q dt
1/q
(18.36)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e and Sobolev Like Inequalities for Widder Derivatives
≤
Z
x0
p
|gn (x, t)| dt
x
1/p
kLn+1 f kLq (a,x0 ) .
221
(18.37)
That is Z
ν
|f (x)| ≤ and Z
x0
ν
|f (x)| dx ≤
a
Z
x0 a
x0
p
|gn (x, t)| dt
x
Z
x0
ν/p
p
|gn (x, t)| dt
x
ν
kLn+1 f kLq (a,x0 ) ,
ν/p
!
ν
dx kLn+1 f kLq (a,x0 ) ,
proving the claim.
(18.38)
(18.39)
We continue with Proposition 18.10. Let f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]) , n ∈ Z+ ; W0 , W1 , . . . , Wn > 0 on [a, b] , and for fixed x0 ∈ [a, b] suppose that Li f (x0 ) = 0, i = 0, 1, . . . , n; ν > 0. Then !1/ν Z Z b
1) kf kLν (x0 ,b) ≤
ν
x
gn (x, t) dt
x0
Z
2) kf kL1 (x0 ,b) ≤
kLn+1 f kL∞ (x0 ,b) .
dx
x0
b x0
Z
x x0
!
gn (x, t) dt dx kLn+1 f kL∞ (x0 ,b) .
(18.40)
(18.41)
By (18.7) we have
Proof.
|f (x)| ≤
≤
Z
Z
x0
gn (x, t) |Ln+1 f (t)| dt
x x0
x
gn (x, t) dt kLn+1 f k∞,[x0 ,b] .
(18.42)
That is ν
|f (x)| ≤
Z
x
gn (x, t) dt x0
ν
ν
kLn+1 f k∞,[x0 ,b] ,
(18.43)
and Z
b x0
ν
|f (x)| dx ≤
proving the claim.
Z
b x0
Z
x
gn (x, t) dt x0
ν
!
ν
dx kLn+1 f k∞,[x0 ,b] ,
(18.44)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
222
We give Proposition 18.11. All as in Proposition 18.10. Then ν 1/ν Z x0 Z x0 kLn+1 f kL∞ (a,x0 ) . |gn (x, t)| dt dx 1) kf kLν (a,x0 ) ≤ 2) kf kL1 (a,x0 ) ≤ Proof.
Z
Z
x0 a
x
By (18.7) we have
Z
≤ That is
ν
|f (x)| ≤ Z
x0 a
ν
|f (x)| dx ≤
proving the claim.
Z
Z
|gn (x, t)| |Ln+1 f (t)| dt
x
|gn (x, t)| dt kLn+1 f k∞,[a,x0 ] .
x
Z
x0
|gn (x, t)| dt
x
Z
ν
x0 x
(18.46)
x0
x0
x0 a
|gn (x, t)| dt dx kLn+1 f kL∞ (a,x0 ) .
x0
|f (x)| ≤
and
(18.45)
x
a
|gn (x, t)| dt
(18.47)
ν
kLn+1 f k∞,[a,x0 ] ,
ν
ν
dx kLn+1 f k∞,[a,x0 ] ,
(18.48)
(18.49)
We continue with reverse Dirichlet–Poincar´e like inequalities. Theorem 18.12. Let f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]) , n ∈ Z+ ; W0 , W1 , . . . , Wn > 0 on [a, b] , and for a fixed a ≤ x0 < b assume that Li f (x0 ) = 0, i = 0, 1, . . . , n. Let 0 < p < 1, q < 0 : 1p + 1q = 1, ν > 0. Further suppose that Ln+1 f is of fixed sign and nowhere zero on [x0 , b] . Then ν/p !1/ν Z b Z x p (gn (x, t)) dt dx kLn+1 f kLq (x0 ,b) . (18.50) 1) kf kLν (x0 ,b) ≥ x0
x0
2) kf kLp (x0 ,b) ≥
Z
3) kf kL1 (x0 ,b) ≥
Z
Proof.
b x0 b
x0
Z Z
(gn (x, t)) dt dx
!1/p
1/p
!
x
p
x0 x
(gn (x, t))p dt
x0
kLn+1 f kLq (x0 ,b) .
(18.51)
dx kLn+1 f kLq (x0 ,b) .
(18.52)
Here we have by (18.7) and assumption that Z x gn (x, t) |Ln+1 f (t)| dt, |f (x)| = x0
(18.53)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e and Sobolev Like Inequalities for Widder Derivatives
223
∀x ∈ [x0 , b] . Hence by reverse H¨ older’s inequality we derive Z x 1/p Z x 1/q p q |f (x)| ≥ (gn (x, t)) dt |Ln+1 f (t)| dt x0
≥
Z
x0
x
p
(gn (x, t)) dt x0
1/p
Z
b x0
q
|Ln+1 f (t)| dt
!1/q
,
(18.54)
kLn+1 f kνLq (x0 ,b) ,
(18.55)
true for all x ∈ (x0 , b] . That is Z
ν
|f (x)| ≥ true for all x ∈ [x0 , b] , and Z b Z ν |f (x)| dx ≥ x0
b x0
x
p
(gn (x, t)) dt x0
Z
x
p
ν/p
(gn (x, t)) dt x0
ν/p
!
ν
dx kLn+1 f kLq (x0 ,b) ,
proving the claim.
(18.56)
We give Theorem 18.13. Let f, u0 , u1 , . . . , un ∈ C n+1 ([a, b]) , n is odd; W0 , W1 , . . . , Wn > 0 on [a, b] , and for a fixed a < x0 ≤ b assume that Li f (x0 ) = 0, i = 0, 1, . . . , n. Let 0 < p < 1, q < 0 : 1p + 1q = 1, ν > 0. Further suppose that Ln+1 f is of fixed sign and nowhere zero on [a, x0 ] . Then ν/p !1/ν Z x0 Z x0 p kLn+1 f kLq (a,x0 ) . dx (−gn (x, t)) dt 1) kf kLν (a,x0 ) ≥ x
a
(18.57)
2) kf kLp (a,x0 ) ≥
Z
3) kf kL1 (a,x0 ) ≥
Z
Proof.
a
x0 Z
x0 a
Z
(−gn (x, t)) dt dx
1/p
1/p
!
x0
p
x x0
p
(−gn (x, t)) dt x
kLn+1 f kLq (a,x0 ) . (18.58)
dx kLn+1 f kLq (a,x0 ) . (18.59)
Here by (18.7) and assumption we have Z x |f (x)| = gn (x, t) Ln+1 f (t) dt x0
Z =
x0
x
gn (x, t) Ln+1 f (t) dt
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
224
Z
(18.12)
=
=
Z
x0 x
x0
x
(−gn (x, t)) Ln+1 f (t) dt
(−gn (x, t)) |Ln+1 f (t)| dt.
So by reverse H¨ older’s inequality we find 1/p Z Z x0 p (−gn (x, t)) dt |f (x)| ≥ x
≥
Z
x0 x
true for all x ∈ [a, x0 ) , and ν
|f (x)| ≥ true for all x ∈ [a, x0 ] . Thus Z Z x0 ν |f (x)| dx ≥ a
(−gn (x, t))p dt
Z
x0 a
x0
1/p
p
(−gn (x, t)) dt x
Z
x0
x0
1/q
kLn+1 f kLq (a,x0 ) ,
(−gn (x, t)) dt x
q
|Ln+1 f (t)| dt
x
ν/p
p
(18.60)
(18.61)
ν
kLn+1 f kLq (a,x0 ) ,
ν/p
!
(18.62)
ν
dx kLn+1 f kLq (a,x0 ) , (18.63)
proving the claim.
We add Theorem 18.14. Now n is even, the rest as in Theorem 18.13. Then ν/p !1/ν Z x0 Z x0 p kLn+1 f kLq (a,x0 ) . (18.64) dx (gn (x, t)) dt 1) kf kLν (a,x0 ) ≥ x
a
2) kf kLp (a,x0 ) ≥
Z
3) kf kL1 (a,x0 ) ≥
Z
Proof.
x0 a x0
a
Z Z
x0
p
1/p
kLn+1 f kLq (a,x0 ) .
(18.65)
1/p
dx kLn+1 f kLq (a,x0 ) .
(18.66)
(gn (x, t)) dt dx x x0
p
(gn (x, t)) dt x
!
Similar to Theorem 18.13.
We continue with Sobolev like inequalities. Theorem 18.15. Same assumptions as in Theorem 18.8. Call ν/p !1/ν Z b Z x p (gj (x, t)) dt dx , Mν,1 := max 0≤j≤n x0 x0
(18.67)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e and Sobolev Like Inequalities for Widder Derivatives
ν > 0. Then 1) kf kLν (x0 ,b) ≤
Mν,1 n+1
2) kf kLq (x0 ,b) ≤
Mq,1 n+1
n X j=0
225
kLj+1 f kLq (x0 ,b) ,
(18.68)
,
(18.69)
n X kLj+1 f k
Lq (x0 ,b)
j=0
and when ν = p = q = 2 we obtain n M2,1 X 3) kf kL2 (x0 ,b) ≤ kLj+1 f kL2 (x0 ,b) , n+1 j=0
where
M2,1 := max
Z
0≤j≤n
b x0
Z
x
2
(gj (x, t)) dt dx x0
!1/2
(18.70)
.
(18.71)
Proof. The assumptions of Theorem 18.8 are fulfilled for all j = 0, 1, . . . , n. Thus by (18.28) we derive !1/ν Z Z b
kf kLν (x0 ,b) ≤
x
ν/p
p
(gj (x, t)) dt
x0
dx
x0
kLj+1 f kLq (x0 ,b)
≤ Mν,1 kLj+1 f kLq (x0 ,b) , for all j = 0, 1, . . . , n. From (18.72) by addition we get
(n + 1) kf kLν (x0 ,b) ≤ Mν,1
n X j=0
proving the claim.
(18.72)
kLj+1 f kLq (x0 ,b) ,
(18.73)
We continue with Theorem 18.16. Same assumptions as in Theorem 18.8. Call ν/p !1/ν Z x0 Z x0 p , dx |gj (x, t)| dt Mν,2 := max 0≤j≤n x a
ν > 0. Then
1) kf kLν (a,x0 ) ≤
Mν,2 n+1
n X j=0
kLj+1 f kLq (a,x0 ) ,
(18.74)
(18.75)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
226
2) kf kLq (a,x0 ) ≤
Mq,2 n+1
n X kLj+1 f k
Lq (a,x0 )
j=0
,
(18.76)
and when ν = p = q = 2 we obtain n M2,2 X kLj+1 f kL2 (a,x0 ) , 3) kf kL2 (a,x0 ) ≤ n+1 j=0 where
M2,2 := max
0≤j≤n
Proof.
(Z
x0 a
Z
x0
2
(gj (x, t)) dt dx x
1/2 )
(18.77)
.
Similar to Theorem 18.15, based on Theorem 18.9.
(18.78)
We continue with L∞ results. Proposition 18.17. All as in Proposition 18.10. Call ν !1/ν Z b Z x , gj (x, t) dt dx Kν,1 := max 0≤j≤n x0 x0 ν > 0. Then
1) kf kLν (x0 ,b) ≤
Kν,1 n+1
2) kf kL1 (x0 ,b) ≤
K1,1 n+1
and
Proof.
n X kLj+1 f k
L∞ (x0 ,b)
j=0
n X kLj+1 f k
L∞ (x0 ,b)
j=0
,
(18.79)
.
(18.80)
Similar to Theorem 18.15, based on Proposition 18.10.
Proposition 18.18. All as in Proposition 18.10. Call (Z ν 1/ν ) x0 Z x0 Kν,2 := max |gn (x, t)| dt dx , 0≤j≤n
a
1) kf kLν (a,x0 ) ≤
Kν,2 n+1
2) kf kL1 (a,x0 ) ≤
K1,2 n+1
and
Based on Proposition 18.11.
(18.81)
x
ν > 0. Then
Proof.
n X kLj+1 f k
L∞ (a,x0 )
j=0
n X kLj+1 f k
L∞ (a,x0 )
j=0
,
(18.82)
.
(18.83)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e and Sobolev Like Inequalities for Widder Derivatives
227
We continue with reverse Sobolev like inequalities. Theorem 18.19. Assume here that Lj+1 f is of fixed sign and nowhere zero on [x0 , b] , for j = 0, 1, . . . , n. The rest are supposed as in Theorem 18.12. Call ν/p !1/ν Z b Z x p Sν,1 := min , (18.84) dx (gj (x, t)) dt 0≤j≤n x0 x0
ν > 0. Then
1) kf kLν (x0 ,b) ≥
Sν,1 n+1
2) kf kLp (x0 ,b) ≥
Sp,1 n+1
3) kf kL1 (x0 ,b) ≥
S1,1 n+1
and
n X j=0
kLj+1 f kLq (x0 ,b) ,
(18.85)
,
(18.86)
n X kLj+1 f k
Lq (x0 ,b)
j=0
n X j=0
kLj+1 f kLq (x0 ,b) .
(18.87)
Proof. The assumptions of Theorem 18.12 are fulfilled for all j = 0, 1, . . . , n. Thus by (18.50) we obtain !1/ν Z Z b
kf kLν (x0 ,b) ≥
x
ν/p
p
(gj (x, t)) dt
x0
kLj+1 f kLq (x0 ,b)
dx
x0
≥ Sν,1 kLj+1 f kLq (x0 ,b) ,
(18.88)
for all j = 0, 1, . . . , n. From (18.88) by addition we obtain (n + 1) kf kLν (x0 ,b) proving the claim.
n X kLj+1 f k ≥ Sν,1
Lq (x0 ,b)
j=0
,
(18.89)
We continue with Theorem 18.20. Assume all as in Theorem 18.13. Here n = 2k + 1, k ∈ Z + ; ν > 0. Further suppose that Lj+1 f is of fixed sign and nowhere zero on [a, x0 ] for all odds j ∈ [1, n] . Call ν/p !1/ν Z x0 Z x0 p Sν,2 := min (−gj (x, t)) dt dx . (18.90) odd j∈[1,n] a x
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
228
Then 1) kf kLν (a,x0 ) ≥
2) kf kLp (a,x0 ) ≥
3) kf kL1 (a,x0 ) ≥
Sν,2 k+1
Sp,2 k+1
S1,2 k+1
and
(18.91)
(18.92)
(18.93)
X kLj+1 f kLq (a,x0 ) , j∈[1,n] odd
X kLj+1 f kLq (a,x0 ) , j∈[1,n] odd
X kLj+1 f kLq (a,x0 ) . j∈[1,n] odd
Proof. As in Theorem 18.19, based on Theorem 18.13. Since n = 2k + 1, k ∈ Z + , there are (k + 1) odd numbers in [1, n] , so we apply (18.57) (k + 1) times. We finish with Theorem 18.21. Assume all as in Theorem 18.14. Here n = 2k, k ∈ Z+ ; ν > 0. Further suppose that Lj+1 f is of fixed sign and nowhere zero on [a, x0 ] for all evens j ∈ [0, n] . Call ν/p !1/ν Z x0 Z x0 . (18.94) dx (gj (x, t))p dt Sν,3 := min even j∈[0,n] x a Then
1) kf kLν (a,x0 ) ≥
2) kf kLp (a,x0 ) ≥
3) kf kL1 (a,x0 ) ≥
Sν,3 k+1
Sp,3 k+1
S1,3 k+1
and
(18.95)
(18.96)
(18.97)
X kLj+1 f kLq (a,x0 ) , j∈[0,n] even
X kLj+1 f kLq (a,x0 ) , j∈[0,n] even
X kLj+1 f kLq (a,x0 ) . j∈[0,n] even
Proof. As in Theorem 18.19, based on Theorem 18.14. Since n = 2k, k ∈ Z+ , there are (k + 1) even numbers in [0, n] , so we apply (18.64) (k + 1) times.
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Chapter 19
Poincar´ e and Sobolev Like Inequalities for Vector Valued Functions
Various Lp form Poincar´e and Sobolev like inequalities are given for functions valued in a Banach vector space with applications on C. This chapter relies on [44]. 19.1
Introduction
This chapter is motivated by the famous Poincar´e inequality, see [2]: Given a bounded domain Ω ⊂ Rn , n ∈ N, it holds kukLp (Ω) ≤ C k∇ukLp (Ω) for real valued functions u with vanishing mean value over Ω, 1 ≤ p ≤ ∞, under very general assumptions on Ω, where k∇ukLp (Ω) is defined as the Lp −norm of the euclidian norm of ∇u. Especially in [2] is proved for a convex domain Ω ⊂ Rn with diameter d that
d k∇ukL1 (Ω) 2 for any u with zero mean value on Ω, also the constant 1/2 is optimal. We are also motivated by [92], where the authors prove the following: Let 1 < p < ∞, −∞ < a < b < ∞. The best constant C (independent of a, b) for which the 1−dimensional Poincar´e inequality
Rb
f (t) dt
2− 1 a ≤ C (b − a) p kf 0 kLp ([a,b])
f −
b−a 1 kukL1 (Ω) ≤
L ([a,b])
0
holds for all real valued Lipschitz continuous functions f, is C = 12 (1 + p0 )−1/p , where p0 > 1 : p1 + p10 = 1. This chapter is also motivated by the famous Sobolev inequality of the following form, see [139], p. 263: (the Gagliardo–Nirenberg–Sobolev inequality). Assume 1 ≤ p ≤ n. Then exists a constant C, depending only on p and n, such that kukLp∗ (Rn ) ≤ C kDukLp (Rn ) , 229
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
230
for all real valued u ∈ Cc1 (Rn ) . np Here p∗ := n−p , p∗ > p, and Du is the gradient of u. Also we are motivated by the following result, p. 265, [139]: estimates for Sobolev space W 1,p , 1 ≤ p < n. Let U be a bounded, open subset of Rn , and suppose the boundary ∂U is C 1 . Assume 1 ≤ p < n, and real valued u ∈ W 1,p (U ) . ∗ Then u ∈ Lp (U ) , with the estimate kukLp∗ (U ) ≤ C kukW 1,p (U ) , the constant C depending only on p, n and U . So here we derive Poincar´e and Sobolev like inequalities of Lp form for Banach vector space valued functions. Most of the Poincar´e and Sobolev like inequalities here are of Dirichlet type. We assume in this case initial conditions prescribed equal to zero. But we give also a Neumann–Poincar´e like inequality involving the average of the engaged function, see Theorem 19.3. At the end we apply the results to C-valued functions. We need
19.2
Background
(see [230], pp. 83–94) Let f (t) be a function defined on [a, b] ⊆ R taking values in a real or complex normed linear space (X, k·k) . Then f (t) is said to be differentiable at a point t0 ∈ [a, b] if the limit
f (t0 + h) − f (t0 ) (19.1) h→0 h exists in X, the convergence is in k·k . This is called the derivative of f (t) at t = t 0 . We call f (t) differentiable on [a, b] , iff there exists f 0 (t) ∈ X for all t ∈ [a, b] . Similarly and inductively are defined higher order derivatives of f, denoted f 00 , (3) f , . . . , f (k) , k ∈ N, just as for numerical functions. For all the properties of derivatives see [230], pp. 83–86. Let now (X, k·k) be a Banach space, and f : [a, R bb] → X. We define the vector valued Riemann integral a f (t) dt ∈ X as the limit of the vector valued Riemann sums in X, convergence is in k·k . The definition is as for the numerical valued functions. Rb If a f (t) dt ∈ X we call f integrable on [a, b] . If f ∈ C ([a, b] , X), then f is integrable, [230], p. 87. For all the properties of vector valued Riemann integrals see [230], pp. 86–91. We define the space C n ([a, b] , X) , n ∈ N, of n-times continuously differentiable functions from [a, b] into X; here continuity is with respect to k·k and defined in the usual way as for numerical functions. f 0 (t0 ) := lim
Book˙Adv˙Ineq
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e and Sobolev Like Inequalities for Vector ValuedFunctions
231
Let (X, k·k) be a Banach space and f ∈ C n ([a, b] , X) , then we have the vector valued Taylor’s formula, see [230], pp. 93–94, and also [227], (IV, 9; 47). It holds En (x, y) := f (y) − f (x) − f 0 (x) (y − x) − 1 1 00 2 n−1 f (x) (y − x) − . . . − f (n−1) (x) (y − x) 2 (n − 1)! Z y 1 n−1 (n) = (y − t) f (t) dt, ∀x, y ∈ [a, b] . (n − 1)! x
(19.2)
In particular (19.2) is true when X = Rm , Cm , m ∈ N, etc. In case of some x0 ∈ [a, b] such that f (k) (x0 ) = 0, k = 0, 1, . . . , n − 1, then Z y 1 n−1 (n) (y − t) f (t) dt, ∀y ∈ [a, b] , (19.3) f (y) = (n − 1)! x0 see also [52]. In that case En (x0 , y) = f (y) .
19.3
Results
Here we consider always X to be a Banach space, n ∈ N, and f ∈ C n ([a, b] , X) , [a, b] ⊆ R. We fix x0 ∈ [a, b] ; p, q > 1 : p1 + q1 = 1, ν > 0. We present the first results which are of Dirichlet–Poincar´e like inequalities. Theorem 19.1. Let y ∈ [x0 , b] , x0 < b. Assume f (k) (x0 ) = 0, k = 0, 1, . . . , n − 1. Then
1
1 (b − x0 )(n−1+ p + ν )
f (n)
L (x ,b) q 0 1) kkf kkLν (x0 ,b) ≤ 1/ν . (19.4) 1/p 1 ((n − 1)!) (p (n − 1) + 1) ν n−1+ p +1 When ν = q we obtain
2) kkf kkLq (x0 ,b) ≤
((n − 1)!) (p (n − 1) + 1)1/p (qn)1/q
And for ν = q = p = 2 we have 3) kkf kkL2 (x0 ,b) Proof.
By (19.3) we have
(b − x0 )n
f (n)
Lq (x0 ,b)
n
(b − x0 )
f (n)
L2 (x0 ,b) √ ≤ . √ ((n − 1)!) 2n − 1 2n
Z y
1 n−1 (n)
(y − t) f (t) dt
(n − 1)! x0 Z y
1
n−1 (n) ≤ (y − t)
f (t) dt (n − 1)! x0
kf (y)k =
.
(19.5)
(19.6)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
232
≤
1 (n − 1)!
Z
y x0
(y − t)
p(n−1)
dt
1/p Z
y x0
)
(y − x0 )( 1
(n)
.
f
(n − 1)! (p (n − 1) + 1)1/p Lq (x0 ,b) 1 n−1+ p
≤ Thus
1/q
(n) q
f (t) dt (19.7)
1 ν n−1+ p
)
1 (y − x0 ) (
(n)
ν kf (y)k ≤ ,
f
ν ((n − 1)!) (p (n − 1) + 1)ν/p Lq (x0 ,b)
ν
and Z
(19.8)
1 ν (n−1+ p )+1
f (n)
ν (b − x ) 0 1 Lq (x0 ,b) ν , kf (y)k dy ≤ ν ((n − 1)!) ν n − 1 + 1 + 1 (p (n − 1) + 1)ν/p x0 p b
(19.9)
proving the claim.
For the next result we need Theorem 19.2. (see [230], p. 90) Let f (t) be an integrable function on an interval [a, b] , with values in a Banach space X. Then the mean value of f (t) on [a, b] belongs to the closed convex hull of the set of values of f (t) on [a, b] . That is we have that Z b n−1 X 1 1 lim f (t) dt = f (τk ) ∆tk , (19.10) b−a a b − a d(Π)→0 k=0
where
Π := {a = t0 ≤ τ0 ≤ t1 ≤ τ1 ≤ t2 ≤ . . . ≤ tn−1 ≤ τn−1 ≤ tn = b} ,
(19.11)
with d (Π) := max {∆t0 , ∆t1 , . . . , ∆tn−1 } ,
(19.12)
∆tk := tk+1 − tk . Next we give Neumann–Poincar´e like inequalities. Theorem 19.3. Here f ∈ C 1 ([a, b] , X) , X a Banach space, p, q > 1 : ν > 0. One has
Z b
1 1 1
≤ (b − a)( p + ν ) kkf 0 kkLq (a,b) . 1)
f − f (t) dt
b−a a
1 p
+
1 q
= 1,
(19.13)
Lν (a,b)
When ν = q we obtain
Z b
1
f (t) dt
2)
f −
b−a a
Lq (a,b)
≤ (b − a) kkf 0 kkLq (a,b) .
(19.14)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e and Sobolev Like Inequalities for Vector ValuedFunctions
When ν = p = q = 2 we have
Z b
1
3)
f − f (t) dt
b−a a
L2 (a,b)
≤ (b − a) kkf 0 kkL2 (a,b) ,
equivalently,
2 Z b Z b
1
2 f (t) dt ds ≤ (b − a)
f (s) −
b − a a a
Proof.
Z
b a
0
2
!
kf (s)k ds .
Since f ∈ C 1 ([a, b] , X) we have by (19.2) that Z y f (y) = f (x) + f 0 (t) dt, ∀x, y ∈ [a, b] .
233
(19.15)
(19.15*)
(19.16)
x
So by (19.16) we have f (s) = f (τk ) +
Z
s
f 0 (r) dr,
(19.17)
τk
∀s ∈ [a, b] and k = 0, 1, . . . , n − 1. Hence Z s ∆tk ∆tk ∆tk = f (τk ) + f 0 (r) dr, f (s) b−a b − a b − a τk
(19.18)
and f (s)
n−1 1 1 X ∆tk = b−a b−a k=0
·
n−1 X
f (τk ) ∆tk +
k=0
− and
1 b−a
(19.19)
k=0
Therefore we observe that f (s) −
Z s n−1 1 X f 0 (r) dr. ∆tk b−a τk
1 b−a Z
Z
b
f (t) dt = a
b
f (t) dt + a
n−1 1 X f (τk ) ∆tk b−a k=0
Z s n−1 1 X ∆tk f 0 (r) dr, b−a τk k=0
Z b
1
f (t) dt
f (s) −
b−a a
Z b
1 n−1
X 1
≤ f (τk ) ∆tk − f (t) dt
b − a
b−a a k=0
(19.20)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
234
+
Z s
n−1
1 X 0
∆tk f (r) dr
b−a τk k=0
Z b
1 n−1 X 1
f (τk ) ∆tk − f (t) dt ≤
b − a b−a a k=0
Z s n−1 1 X 0 + ∆tk kf (r)k dr b−a τk k=0
Z
Z b
1 n−1 b X 1
f (τk ) ∆tk − f (t) dt + kf 0 (r)k dr. ≤
b − a b−a a a
(19.21)
k=0
That is we proved
Z b
1
f (t) dt
f (s) −
b−a a
Z Z b
1 n−1
b X 1
≤ kf 0 (r)k dr, ∀s ∈ [a, b] , (19.22) f (t) dt + f (τk ) ∆tk −
b − a
b−a a a k=0
for any partition Π. Taking the limit in both sides of (19.22) as d (Π) → 0 we obtain by (19.10) that
Z Z b
b 1
f (t) dt ≤ kf 0 (r)k dr
f (s) −
b−a a a ≤ (b − a)
1/p
Z
b
a
0
q
kf (r)k dr
Consequently we find
ν Z b
1
ν/p f (t) dt ≤ (b − a)
f (s) −
b−a a
!1/q
Z
b a
q
0
kf (r)k dr
Finally we have
ν Z b Z b
ν 1
+1 f (t) dt ds ≤ (b − a) p
f (s) −
b − a a a
proving the claim.
, ∀s ∈ [a, b] .
!ν/q
(19.23)
, ∀s ∈ [a, b] . (19.24)
Z
b a
0
q
kf (r)k dr
!ν/q
,
(19.25)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e and Sobolev Like Inequalities for Vector ValuedFunctions
Corollary 19.4. (to Theorem 19.3). Additionally assume that One has 1
1
Rb a
235
f (t) dt = 0.
1) kkf kkLν (a,b) ≤ (b − a) p + ν kkf 0 kkLq (a,b) .
(19.26)
When ν = q we obtain 2) kkf kkLq (a,b) ≤ (b − a) kkf 0 kkLq (a,b) .
(19.27)
When ν = p = q = 2 we have 3) kkf kkL2 (a,b) ≤ (b − a) kkf 0 kkL2 (a,b) , equivalently Z
b a
2
kf (s)k ds ≤ (b − a)
2
Z
b a
2
0
(19.28) !
kf (s)k ds .
(19.28*)
We continue with Theorem 19.5. Let y ∈ [a, x0 ] , x0 > a. Assume f (k) (x0 ) = 0, k = 0, 1, . . . , n − 1. Then
n−1+ p1 + ν1 )
f (n)
(x0 − a)( Lq (a,x0 ) 1) kkf kkLν (a,x0 ) ≤ 1/ν . (19.29) 1/p ((n − 1)!) (p (n − 1) + 1) ν n − 1 + p1 + 1 When ν = q we derive
2) kkf kkLq (a,x0 ) ≤
n
(x0 − a)
f (n)
Lq (a,x0 )
((n − 1)!) (p (n − 1) + 1)
And for ν = q = p = 2 we have 3) kkf kkL2 (a,x0 ) Proof.
By (19.3) we have kf (y)k =
1 (n − 1)!
Z
Z
1/q
.
n
(x0 − a)
f (n)
L2 (a,x0 ) √ ≤ . √ ((n − 1)!) 2n − 1 2n
x0
y
x0 y
(qn)
(19.30)
(19.31)
Z x 0
1 n−1 (n)
(t − y) f (t) dt
(n − 1)! y
1 ≤ (n − 1)! ≤
1/p
(t − y)
(t − y)
p(n−1)
dt
n−1
(n)
f (t) dt
1/p Z
x0
y
1/q
(n) q
f (t) dt
1 n−1+ p
)
(x0 − y)( 1
(n)
. ≤
f
(n − 1)! (p (n − 1) + 1)1/p Lq (a,x0 )
(19.32)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
236
Hence ν n−1+ p1 )
1 (x0 − y) (
(n)
ν kf (y)k ≤ ,
f
ν ((n − 1)!) (p (n − 1) + 1)ν/p Lq (a,x0 ) ν
and
Z
x0
ν
kf (y)k dy ≤
a
1 ν ((n − 1)!)
1 ν n−1+ p )+1)
f (n)
ν (x0 − a)( ( Lq (a,x0 ) , · ν/p 1 ν n−1+ p +1 (p (n − 1) + 1)
proving the claim.
(19.33)
(19.34)
We continue with L1 results. Proposition 19.6. Let y ∈ [x0 , b] , x0 < b. Suppose f (k) (x0 ) = 0, k = 0, 1, . . . , n − 1. Then
n−1+ ν1 )
f (n)
(b − x0 )( L1 (x0 ,b) 1) kkf kkLν (x0 ,b) ≤ . (19.35) 1/ν (n − 1)! (ν (n − 1) + 1) When ν = 1 we get 2) kkf kkL1 (x0 ,b) ≤ Proof.
By (19.3) , (19.7) we have kf (y)k ≤
≤
(y − x0 )n−1 (n − 1)!
That is
Z
1 (n − 1)!
ν
and b x0
proving the claim.
n!
Z
y x0
.
(19.36)
(y − t)n−1 f (n) (t) dt
(y − x0 )n−1
(n)
(n)
.
f (t) dt =
f
(n − 1)! L1 (x0 ,b) x0 b
kf (y)k ≤ Z
n
(b − x0 )
f (n)
L1 (x0 ,b)
ν
(y − x0 )ν(n−1)
(n)
ν ,
f
ν ((n − 1)!) L1 (x0 ,b)
kf (y)k dy ≤
(b − x0 )
ν(n−1)+1
(n)
ν
f
L1 (x0 ,b)
ν
((n − 1)!) (ν (n − 1) + 1)
(19.37)
(19.38)
,
(19.39)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e and Sobolev Like Inequalities for Vector ValuedFunctions
237
The countercase of the last result follows. Proposition 19.7. Let y ∈ [a, x0 ] , x0 > a. Suppose f (k) (x0 ) = 0, k = 0, 1, . . . , n − 1. Then
n−1+ ν1 )
f (n)
(x0 − a)( L1 (a,x0 ) 1) kkf kkLν (a,x0 ) ≤ . (19.40) ((n − 1)!) (ν (n − 1) + 1)1/ν When ν = 1 we get
n
(x0 − a)
f (n)
L1 (a,x0 ) . (19.41) 2) kkf kkL1 (a,x0 ) ≤ n! Proof. By (19.3) and (19.32) we have Z x0
1
n−1 (n) kf (y)k ≤ (t − y)
f (t) dt (n − 1)! y n−1
≤
(x0 − y) (n − 1)!
n−1
= That is ν
(x0 − y) (n − 1)!
kf (y)k ≤ and Z
x0
ν
x0
a
(n)
f (t) dt
(n)
f
ν(n−1)
(x0 − y) ν ((n − 1)!)
kf (y)k dy ≤
a
Z
(x0 − a)
proving the claim.
L1 (a,x0 )
.
(19.42)
(n)
ν
f
(ν(n−1)+1)
L1 (a,x0 )
,
(19.43)
(n)
ν
f
L1 (a,x0 )
((n − 1)!)ν (ν (n − 1) + 1)
,
(19.44)
We continue with Sobolev like inequalities Theorem 19.8. Let y ∈ [x0 , b] , x0 < b. Assume f (k) (x0 ) = 0, k = 0, 1, . . . , n − 1. Then ! n
Mν X
(i)
, (19.45) 1) kkf kkLν (x0 ,b) ≤
f
n Lq (x0 ,b) i=1 where
1 i−1+ p + ν1 ) ( (b − x0 ) Mν := max 1/ν . 1≤i≤n ((i − 1)!) (p (i − 1) + 1)1/p ν i − 1 + 1 + 1 p
When ν = q we derive
2) kkf kkLq (x0 ,b)
Mq ≤ n
n
X
(i)
f
i=1
Lq (x0 ,b)
!
,
(19.46)
(19.47)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
238
where Mq := max
1≤i≤n
(
For ν = q = p = 2 we have
(b − x0 )
i
((i − 1)!) (p (i − 1) + 1)
3) kkf kkL2 (x0 ,b) where M2 := max
1≤i≤n
M2 ≤ n (
1/p
n
X
(i)
f
i=1
(qi)
1/q
L2 (x0 ,b)
(b − x0 )i √ √ ((i − 1)!) 2i − 1 2i
)
)
!
.
,
.
(19.48)
(19.49)
(19.50)
Proof. By assumption of Theorem 19.1 we have that f ∈ C i ([a, b] , X) , for all i = 1, . . . , n, and that f (k) (x0 ) = 0, k = 0, 1, . . . , i − 1; for all i = 1, . . . , n, n ∈ N. Thus by (19.4) we find
1 + ν1 ) i−1+ p
f (i)
(b − x0 )( Lq (x0 ,b) kkf kkLν (x0 ,b) ≤ 1/ν 1/p ν i − 1 + p1 + 1 ((i − 1)!) (p (i − 1) + 1)
for all i = 1, . . . , n. So that
≤ Mν
f (i)
n kkf kkLν (x0 ,b) ≤ Mν proving the claim.
(19.51)
Lq (x0 ,b)
n
X
(i)
f
i=1
Lq (x0 ,b)
!
,
(19.52)
We continue with Theorem 19.9. Let y ∈ [a, x0 ] , x0 > a. Assume f (k) (x0 ) = 0, k = 0, 1, . . . , n − 1. Then ! n
∆ν X
(i)
1) kkf kkLν (a,x0 ) ≤ , (19.53)
f
n Lq (a,x0 ) i=1 where
i−1+ p1 + ν1 ) ( (x0 − a) ∆ν := max . 1/ν 1≤i≤n ((i − 1)!) (p (i − 1) + 1)1/p ν i − 1 + 1 + 1 p
When ν = q we obtain
2) kkf kkLq (a,x0 )
∆q ≤ n
n
X
(i)
f
i=1
Lq (a,x0 )
!
,
(19.54)
(19.55)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e and Sobolev Like Inequalities for Vector ValuedFunctions
where ∆q := max
1≤i≤n
(
For ν = q = p = 2 we have
(x0 − a)
((i − 1)!) (p (i − 1) + 1)1/p (qi)1/q
3) kkf kkL2 (a,x0 ) where ∆2 := max
1≤i≤n
Proof.
i
n
X
(i)
f
∆2 ≤ n (
i=1
L2 (a,x0 )
i
(x0 − a) √ √ ((i − 1)!) 2i − 1 2i
)
)
!
.
,
.
239
(19.56)
(19.57)
(19.58)
Based on Theorem 19.5, similar to the proof of Theorem 19.8.
We continue with L1 results. Proposition 19.10. Let y ∈ [x0 , b] , x0 < b. Suppose f (k) (x0 ) = 0, k = 0, 1, . . . , n− 1. Then ! n
θν X
(i)
, (19.59) 1) kkf kkLν (x0 ,b) ≤
f
n i=1 L1 (x0 ,b)
where
θν := max
1≤i≤n
When ν = 1 we get
(
(b − x0 )(
i−1+ ν1 )
((i − 1)!) (ν (i − 1) + 1)
2) kkf kkL1 (x0 ,b)
θ1 ≤ n
where θ1 := max
1≤i≤n
n
X
(i)
f
i=1
(
(b − x0 ) i!
i
)
1/ν
)
L1 (x0 ,b)
.
!
(19.60)
,
.
(19.61)
(19.62)
Based on Proposition 19.6. We have from (19.35) that
1 (b − x0 )(i−1+ ν )
f (i)
L (x ,b) 1 0 kkf kkLν (x0 ,b) ≤ 1/ν ((i − 1)!) (ν (i − 1) + 1) for all i = 1, . . . , n. Consequently we see that
, kkf kkLν (x0 ,b) ≤ θν
f (i)
Proof.
L1 (x0 ,b)
all i = 1, . . . , n. Thus
n kkf kkLν (x0 ,b) ≤ θν proving the claim.
n
X
(i)
f
i=1
L1 (x0 ,b)
!
,
(19.63)
(19.64)
(19.65)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
ADVANCED INEQUALITIES
240
Similarly we give Proposition 19.11. Let y ∈ [a, x0 ] , x0 > a. Assume f (k) (x0 ) = 0, k = 0, 1, . . . , n− 1. Then ! n
Tν X
(i)
, (19.66) 1) kkf kkLν (a,x0 ) ≤
f
n i=1 L1 (a,x0 )
where
Tν := max
1≤i≤n
(
(x0 − a)(
i−1+ ν1 )
((i − 1)!) (ν (i − 1) + 1)1/ν
)
.
(19.67)
When ν = 1 we get 2) kkf kkL1 (a,x0 ) where T1 := max
1≤i≤n
Proof.
19.4
n
X
(i)
f
T1 ≤ n
i=1
(
(x0 − a)i i!
)
L1 (a,x0 )
!
,
.
(19.68)
(19.69)
Based on Proposition 19.7 as in Proposition 19.10.
Applications
Here X = C, f ∈ C n ([a, b] , C) , n ∈ N, x0 , y ∈ [a, b] . Furthermore suppose that f (k) (x0 ) = 0, k = 0, 1, . . . , n − 1; p, q > 1 : p1 + q1 = 1, ν > 0. First we give the following Poincar´e like inequalities results. Corollary 19.12. Let x0 < b. One has kf kLν (x0 ,b) Proof.
(n)
f Lq (x0 ,b) ≤ 1/ν . 1/p ν n − 1 + 1p + 1 ((n − 1)!) (p (n − 1) + 1) (b − x0 )(
n−1+ p1 + ν1 )
Use of (19.4) .
(19.70)
We continue with Corollary 19.13. One has
Z b
1
f (t) dt
f −
b−a a
Proof.
Use of (19.13) .
≤ (b − a)( p 1
Lν (a,b)
+ ν1 )
kf 0 kLq (a,b) .
(19.71)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Poincar´ e and Sobolev Like Inequalities for Vector ValuedFunctions
241
Corollary 19.14. Let x0 > a. One has kf kLν (a,x0 ) Proof.
1 1 (x0 − a)(n−1+ p + ν ) f (n) L (a,x ) q 0 ≤ 1/ν . 1/p 1 ((n − 1)!) (p (n − 1) + 1) ν n−1+ p +1
Use of (19.29).
(19.72)
We continue with L1 Poincar´e like results on C. Corollary 19.15. Let x0 < b. Then (b − x0 )(
kf kLν (x0 ,b) ≤ Proof.
n−1+ ν1 )
(n)
f
L1 (x0 ,b) . 1/ν
((n − 1)!) (ν (n − 1) + 1)
Use of (19.35) .
(19.73)
Corollary 19.16. Let x0 > a. Then kf kLν (a,x0 ) ≤ Proof.
(x0 − a)(
Use of (19.40) .
n−1+ ν1 )
(n)
f
L1 (a,x0 ) . 1/ν
((n − 1)!) (ν (n − 1) + 1)
(19.74)
We finish with Sobolev like inequalities results on C. Corollary 19.17. Let x0 < b. Then kf kLν (x0 ,b)
Mν ≤ n
where Mν as in (19.46). Proof.
n
X
(i)
f i=1
Lq (x0 ,b)
,
Use of (19.45).
kf kLν (a,x0 )
(19.75)
Corollary 19.18. Let x0 > a. Then ∆ν ≤ n
where ∆ν as in (19.54) . Proof.
!
n
X
(i)
f i=1
Lq (a,x0 )
!
Use of (19.53).
,
(19.76)
At last we give L1 Sobolev like results on C. Corollary 19.19. Let x0 < b. Then kf kLν (x0 ,b) where θν as in (19.60) . Proof.
Use of (19.59).
θν ≤ n
n
X
(i)
f i=1
L1 (x0 ,b)
!
,
(19.77)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
ADVANCED INEQUALITIES
242
Corollary 19.20. Let x0 > a. Then kf kLν (a,x0 ) where Tν as in (19.67) . Proof.
Book˙Adv˙Ineq
Use of (19.66).
Tν ≤ n
n
X
(i)
f i=1
L1 (a,x0 )
!
,
(19.78)
17th August 2010
15:56
World Scientific Book - 9.75in x 6.5in
Book˙Adv˙Ineq
Chapter 20
Poincar´ e Type Inequalities for Semigroups, Cosine and Sine Operator Functions Here we present Poincar´e type general Lp inequalities regarding Semigroups, Cosine and Sine Operator functions. We give applications. This chapter relies on [50].
20.1
Introduction
This chapter is motivated by the famous Poincar´e inequality, see [2]: Given a bounded domain Ω ⊂ Rn , n ∈ N, it holds kukLp (Ω) ≤ C k∇ukLp (Ω)
for functions u with vanishing mean value over Ω, 1 ≤ p ≤ ∞, under very general assumptions on Ω, where k∇ukLp (Ω) is defined as the Lp −norm of the euclidean norm of ∇u. Especially in [2] is proved for a convex domain Ω ⊂ Rn with diameter d that d kukL1 (Ω) ≤ k∇ukL1 (Ω) 2 for any u with zero mean value on Ω, also the constant 1/2 is optimal. We are also motivated by [92], where the authors prove the following: Let 1 < p < ∞, −∞ < a < b < ∞. The best constant C (independent of a, b) for which the 1−dimensional Poincar´e inequality
Rb
f (t) dt
2− 1 a ≤ C (b − a) p kf 0 kLp ([a,b])
f −
b−a 1 L ([a,b])
−1/p0
