197 96 6MB
English Pages 339 [340] Year 1987
Nonlinear and
Convex Analysis
PURE AND APPLIED MATHEMATICS A Program of Monographs, Textbooks, and Lecture Notes
EXECUTIVE EDITORS Earl J. Taft
Zuhair Nashed
Rutgers University New Brunswick. New Jersey
Unfrersity of Del.aware Newark. Delaware
CHAIRMEN OF THE EDITORIAL BOARD
S. Kobayashi
Edwin Hewitt
University of California, Berkeley Berkeley, California
University of Washington Seattle, Washington
EDITORIAL BOARD M. S. Baouendi Purdue University
Donald Passman University of Wisconsin-Madison
Jack K. Hale Brown University
Fred S. Roberts Rutgers University
Marvin l'v!arcus University of California, Santa Barbara W. S. Massey Yale University Leopoldo Nachbin Centro Brasileiro de Pesquisas Fisicas and University of Rochester Anil Nerode Cornell University
Gian-Carlo Rota Massachusetts Institute of Technology David Russell University of Wisconsin-Madison Jane Cronin Scanlon Rutgers University Walter Schempp Universitii t Siegen
Mark Teply University of Wisconsin-Milwaukee
LECTURE NOTES IN PURE AND APPLIED MATHEMATICS 1. N. .lacohson, Lxccptional Lie AlgL·bras 2. L. -A'. U1ulahl and F Poulsen, Thin Sets in llannonic Analysis
3. I. Sa take. Classification Theory of Semi-Simple Algebraic Croups 4. F Hir:ehruch, W D. Newmann, and S.S. Koh, Dillerentiable Manifolds and Quadratic Forms (out of print) 5. I. Cizal'c/. Riemannian Symmetric Spaces of Rank 01\L' (out of print) 6. R. B. Burckcl, Characterization of C(X) Among Its Subalgcbras 7. B. R. McDonald..-1. R. Magid, and K. C Smith, Ring Theorv: Proceedings of the Okbhoma ConfrrenL-c 8. Y.-T. Siu, Techniques of Fxtension on Analytic Objects 9. S. R. Caradus, W. /:. Pfajji•11hcrger. and B. Yood, Calkin Algebras and Algebras ol Operators on Banach SpaL"L'S 10. /:". 0. Roxin. P.-T. Uu. and R. L. Stcmherg. Ditlen:ntial Can1csand Contrnl lheor, 11 . .l'vl. Orzech and C. S111all. The Brauer Group of Commutative Rings 12. S. Thomeier, Topology ,ind Its Applications 13 . ./. ,"1. Lopez and K. A. Ross. Sidon Sets 14. W. W. ComjcJrl and S. ,Vcgrepontis, Continuous P,eudometncs 15. K. McKennon and .I. iv!. Roher/son, Locally Convex Spaces 16. M. Canncli and S. Malin. Representations of the Rotation and Lorentz Groups: An Introduction 17. G. B. Seligman, Rational Methods in Lie Algebras 18. D. G. de Figueiredo. Functional Analysis: Proccedint:s of 1hc Brazilian l\lathematical Society Symposium 19. L. Cesari, R. Kannan, and .I. D. Sclzuur, Nonlinear I unctional Analysis and Differential Equations: Proceedings of the Michigan State University Conference 211 . ./. ./. Schaffer, Geometry of Spheres in Normed Spaces 21. K. Yano and M. Kon, Anti-Invariant Subnianifolds 22. W. V. Vasco11cf'ios, The Rings of Dimension Two 23. R. J-:. Chandler, Hausdorff Compactifications 24. S. P. Franklin and B. VS. Thomas, Topology: Proceedings of the Memphis State University Conference 25. S. K. Jain, Ring Theory: Proceedings the Ohio University Conference 26. B. R. McDonald and R. A. Morris. Ring Theory II: Proceedings of the Second OkLihom,1 Conference 27. R. B. Mura and A. Rhe11Hulla. Orderablc (;roups 28 . ./. R. Graef: Stability of Dynamical Systems: Thl'Ory and Application, 29. lf..C. Wang, Homo~eneous Branch Algebras 30. F. 0. Roxin, P.-T. /,i11, and R. /,. Sternhcrg. Differential Games and Control The,m II 31. R. D. Porter. Introduction to Fibre Bundles 32. M. Altman, Contrac·tors and Contractor Direction, Theory and Applications 33 . .I. S. Colan. Decomposition and Dimension in .\1odule Categories 34. c;_ Fairweather, 1-inite flemL'llt Galcrkin Methods for Di!lerL'nti,il 1-.quatlllns 35 . .I D. Sall_\', Numbers of Cennators of Ideals in Lllcal Rings 36. S S. Miller, Comple~ Analysis: Proceedings of the S.U.:\i.Y. Brnckport ConlcrenCL' 37. R. Gordon, Representation Theory of Algebra,: ProcL·edings ,,f the Phil:tdclphia ConferenLT 38. J'vl. Goto and F !)_ Grosshans. Scmisnnpk Lie Al~d,ra, 39. A. I. Arruda. X. CA. da Costa, and R. Ch1ia1111i. .\Li1hematical 1.o~ic: Procccdin).!s of the 1-'irst Brazilian Conference
or
40. F. Van 0ystaeyen, Ring Theory: Proceedings of the 1977 Antwerp Conference 41. F. Van 0ystaeyen and A. Verschoren. Reflectors and Localization: Application to Sheaf Theory 42. M. Satyanarayana, Positively Ordered Semigroups 43. D. L. Russell, Mathematics of finite-Dimensional Control Systems 44. P.-T. Liu and E. Roxin, Differential Games and Control Theory III: Proceedings of the Third Kingston Conference, Part A 45. A. Geramita and J. Seberry. Orthogonal Designs: Quadratic Forms and Hadamard Matrices 46. J. Cigler, V. Losert, and P. Michor, Banach Modules and Functors on Categories of Banach Spaces 4 7. P.-T. Liu and J. G. Sutinen. Control Theory in Mathematical Economics: Proceedings of the Third Kingston Conference, Part B 48. C. Byrnes, Partial Differential Equations and Geometry 49. G. Klambauer, Problems and Propositions in Analysis 50. J. Knopfmacher, Analytic Arithmetic of Algebraic Function Fields 51. F. Van 0ystaeyen, Ring Theory: Proceedings of the 1978 Antwerp Conference 52. B. Kedem, Binary Time Series 53. J. Barros-Neto and R. A. Artino. Hypoelliptic Boundary-Value Problems 54. R. L. Sternberg, A. J. Kalinowski. and J. S. Papadakis. Nonlinear Partial Differential Equations in Engineering and Applied Science 55. B. R. McDonald, Ring Theory and Algebra III: Proceedings of the Third Oklahoma Conference 56. J. S. Golan, Structure Sheaves over a Noncommutative Ring 57. T. V. Narayana, J. G. Williams, and R. M. Mathsen. Combinatorics, Representation Theory and Statistical Methods in Groups: YOUNG DAY Proceedings 58. T. A. Burton, Modeling and Differential Equations in Biology 59. K. H. Kim and F. W. Roush, Introduction to Mathematical Consensus Theory 60. J. Banas and K. Goebel, Measures of Noncompactness in Banach Spaces 61. 0. A. Nielson, Direct Integral Theory 62. J. E Smith, G. 0. Kenny. and R. N. Ball, Ordered Groups: Proceedings of the Boise State Conference 63. J. Cronin, Mathematics of Cell Electrophysiology 64. J. W. Brewer, Power Series Over Commutative Rings 65. P. K. Kamthan and M. Gupta, Sequence Spaces and Series 66. T. G. McLaughlin, Regressive Sets and the Theory of !sols 67. T. L. Herdman, S. M. Rankin, III, and H. W. Stech. Integral and Functional Differential Equations 68. R. Draper, Commutative Algebra: Analytic Methods 69. W. G. McKay and J. Patera. Tables of Dimensions, Indices, and Branching Rules for Representations of Simple Lie Algebras 70. R. L. Devaney and Z. H. Nitecki, Classical Mechanics and Dynamical Systems 71. J. Van Geel, Places and Valuations in Noncommutative Ring Theory 72. C. Faith, Injective Modules and Injective Quotient Rings 73. A. Fiacco, Mathematical Programming with Data Perturbations I 74. P. Schultz, C. Praeger, and R. Su/Iii-an, Algebraic Structures and Applications Proceedings of the First Western Australian Conference on Algebra 75. L. Bican, T. Kepka, and P. Nemec, Rings, Modules, and Preradicals 76. D. C. Kay and M. Breen, Convexity and Related Combinatorial Geometry: Proceedings of the Second University of Oklahoma Conference 77. P. Fletcher and W. F. Lindgren, Quasi-Uniform Spaces 78. C.-C Yang, Factorization Theory of Meromorphic l'unctions 79. 0. Taussky, Ternary Quadratic Forms and Norms 80. S. P. Singh and J. H. Burry. Nonlinear Analysis and Applications 81. K. B. Hannsgen, T. L. Herdman, H. W. Stech. and R. L. Wheeler, Volterra and Functional Differential Equations
82. N. L. Johnson, M. J. Kallaher, and C. T. Long, Finite Geometries: Proceedings of a Conference in Honor of T. G. Ostrom 83. G. I. Zapata, Functional Analysis, Holomorphy, and Approximation Theory 84. S. Greco and G. Valla, Commutative Algebra: Proceedings of the Trento Conference 85. A. V. Fiacco, Mathematical Programming with Data Perturbations II 86. J.-B. Hiriart-Urruty, W. Oettli, and J. Stoer, Optimization: Theory and Algorithms 87. A. Figa Talamanca and M. A. Picardello, Harmonic Analysis on Free Groups 88. M Harada, Factor Categories with Applications to Direct Decomposition of Modules 89. V. I. Istriifescu, Strict Convexity and Complex Strict Convexity: Theory and Applications 90. V. Lakshmikantham, Trends in Theory and Practice of Nonlinear Differential Equations 91. H. L. Manocha and J. B. Srivastava, Algebra and Its Applications 92. D. V. Chudnovsky and G. V. Chudnovsky, Classical and Quantum Models and Arithmetic Problems 93. J. W. Longley, Least Squares Computations Using Orthogonalization Methods 94. L. P. de Alcantara, Mathematical Logic and Formal Systems 95. C. E. Aull, Rings of Continuous Functions 96. R. Chuaqui, Analysis, Geometry, and Probability 97. L. Fuchs and L. Salce, Modules Over Valuation Domains 98. P. Fischer and W. R. Smith, Chaos, Fractals, and Dynamics 99. W. B. Powell and C. Tsinakis, Ordered Algebraic Structures 100. G. M Rassias and T. M Rassias, Differential Geometry, Calculus of Variations, and Their Applications 101. R.-E. Hoffmann and K. H. Hofmann, Continuous Lattices and Their Applications 102. J. H. Lightbourne, III, and S. M. Rankin, III, Physical Mathematics and Nonlinear Partial Differential Equations 103. C. A. Baker and L. M. Batten, Finite Geometries 104. J. W. Brewer, J. W. Bunce, and F. S. Van Vleck, Linear Systems Over Commutative Rings 105. C. McCrory and T. Shifrin, Geometry and Topology: Manifolds, Varieties, and Knots 106. D. W. Kueker, E. G. K. Lopez-Escobar, and C. H. Smith, Mathematical Logic and Theoretical Computer Science 107. B.-L. Lin and S. Simons, Nonlinear and Convex Analysis: Proceedings in Honor of Ky Fan Other Volumes in Preparation
Nonlinear and Convex Analysis PROCEEDINGS IN HONOR OF KY FAN Edited by
BOR,LUH LIN 'The University of Iowa Iowa City, Iowa
STEPHEN SIMONS University of California Santa Barbara, California
~ CRC Press v
Taylor & Francis Group Boca Raton London New York
CRC Press is an imprint of the Taylor & Francis Group, an informa business
Press Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487-2742 © 1987 by Taylor & Francis Group, LLC CRC Press is an imprint of Taylor & Francis Group, an lnforma business No claim to original U.S. Government works ISBN-13: 978-0-824-77777-7 (pbk) This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot assume responsibility for the validity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint. Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www. copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Publisher's Note The publisher has gone to great lengths to ensure the quality of this reprint but points out that some imperfections in the original copies may be apparent.
Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com and the CRC Press Web site at http://www.crcpress.com DOI: 10.1201/9781003420040
Preface
Professor Ky Fan retired at the end of the 1984-85 academic year, and the Department of Mathematics of the University of California at Santa Barbara held a conference in his honor June 23-26, 1985. This volume contains expanded versions of the talks given at the conference, as well as papers contributed by others, many of whom would have liked to attend the conference but were unable to come. Before writing about the conference, I will give a brief history of Professor Fan's career, although it is not really possible to do it justice in a few lines. Professor Fan obtained a B.S. from the National Peking University in 1936 and a D.Sc. Math from the University of Paris in 1941.
He was a mem-
ber of the Institute for Advanced Study at Princeton from 1945 to 1947, and held professorial positions at the University of Notre Dame, Wayne State University, and Northwestern University before coming to the University of California at Santa Barbara in 1965. He has been a visiting professor at the University of Texas at Austin, Hamburg University in Germany, and the University of Paris IX in France. Elected a member of the Academics Sinica in 1964, he served as director of the Institute of Mathematics at the Academics Sinica for two terms from 1978 to 1984.
He has been on many editorial boards, including that
of the Journal of Mathematical Analysis and Applications since its founding in 1960. A student and collaborator of M. Frechet, Professor Fan was also influenced by J. von Neumann and H. Weyl.
He made fundamental contributions
to operator and matrix theory, convex analysis and inequalities, linear and nonlinear programming, topology and fixed-point theory, and topological iii
iv
Preface
groups.
Many of his results have become part of the basic literature in
these fields.
Furthermore, his work in fixed-point theory has found wide
application in mathematical economics and game theory, potential theory, calculus of variations, and differential equations, and has had many consequences in nonlinear functional analysis. Professor Fan has trained twenty-two Ph.D. students, many of whom hold positions at prestigious universities both in the United States and abroad.
Seven of them were at the conference:
John Cantwell, Michael
Geraghty, Donald Hartig, Charles llimmelberg, Ronald Knill, Bor-Luh Lin, Michael Powell, and Raimond Struble. In all, sixty-three mathematicians from three continents came to the University of California at Santa Barbara for the conference.
The one hour
talks were given by Professors Jean-Pierre Aubin of the University of Paris, Felix Browder of the University of Chicago, Zhang Gong-Qing (K. C. Chang) of Peking University, Andrzej Granas of the University of Montreal, Paul Ilalmos of the University of Santa Clara, and Shizuo Kakutani of Yale University.
Professor Samuel Karlin of Stanford University was also scheduled
to give a one hour talk, but was prevented from coming by i 11 ness.
Twenty-
s even shorter talks were also given, many in areas pioneered by Professor Fan. On the evening of June 25, those attending the conference joined members of the Department of Mathematics for a banquet 1n honor of Professor Fan.
The banqucteers were rather startled when one of the spcakers---a
student of Professor Fan who prefers to remain nameless-started to undress.
To the relief (and amusement) of everyone it was merely to demon-
strate that he was wearing a Ky Fan T-shirt. I spoke to many of the participants in the conference, trying to find out more about Professor Fan's career before he came to Santa Barbara. They confirmed unanimously my own feelings; they spoke of his high standards in mathematics, not only for himself, hut also for his colleagues and students.
But they also spoke of his fine qualities as a human being and
his generosity. Although Professor Fan has now retired from his teaching duties, he remains mathematically active, and will undoubtedly continue to do so for many years to come. Thanks are due to Marvin Marcus, Associate Vice Chancellor of Research and Academic Development, and David Sprecher, Provost of the College of Letters and Science at the University of California at Santa Barbara, whose
Preface
V
offices provided financial support for the conference; to James Robertson, Chairman of the Department of Mathematics at the University of California at Santa Barbara and all his staff, who worked so hard to make the conference a success; to all those who participated - in many cases coming a great distance and braving an unpleasant airline strike to get to Santa Barbara; to all those who could not participate in the conference but sent their good wishes; and to Professor Earl Taft and Professor Zuhair Nashed and Marcel Dekker, Inc., for including the Proceedings in their Lecture Notes in Pure and Applied Mathematics Series.
Stephen Simons
Contents
Preface
iii ix
Contributors
xiii
Conference Participants
xv
List of Publications of Ky Fan Smooth and Heavy Viable Solutions to Control Problems
Jean-Pierre Aubin
On Simplified Proofs of Theorems of von Neumann, Heinz, and Ky Fan, and Their Extended Versions
Gong-ning Chen
Vandermonde Determinant and Lagrange Interpolation in Rs Charles K. Chui and M. J. Lai
Applications of Nonstandard Theory of Locally Convex Spaces Carl L. Devito
Local Invertibility of Set-Valued Maps
Halina Frankowska
Some Minimax Theorems Without Convexity
Andrzej Granas and Fon-Che Liu
Weak Compactness and the Minimax Equality
Chung-Wei Ha
Nonlinear Volterra Equations with Positive Kernels
Norimichi Hirano
Some Results on Multivalued Mappings and Inequalities Without Convexity Charles D. Horvath
Strong Equilibria
1
15
23 37
47 61 77
83
99 107
Tatsuro Ichiishi
vii
viii
Contents
A Variational Principle ,\ppl i cat ion to the \onl inear Complementarity Problem George Isac and M.
Thera
On a Best Approximation Theorem Hidetoshi Kamiya
A Vector-Minimization Problem in a Stochastic ContinuousTime n-Person Game Hang-Chin Lai and Kensukc Tanaka
Calculation of the A-Function for Several Classes of \ormed Linear Spaces
Robert H. Lohman and Thaddeus J. Shura
Existence of Positive Eigenvectors and Fixed Points for A-Proper Type Naps 1n Cones
W. V. Petryshyn
On the Nethod of Successive Approximations for Nonexpansive Mappings Simeon Reich and I. Shafrir
147
151
167
1 75
19 3
Quasilinear Ellipticity on the \-Torus Victor L. Shapiro
A Local ~linimax 'li1eorern \\'i thout Compactness
Shi Shu-zhong and Chang Kung-ching (Zhang Gong-qing)
203
211
Covering Theorems of Convex Sets Related to Fixed-Point Theorems
Mau-Hsiang Shih and Kok-Keong Tan
Shapley Selections anJ Covering Theorems of Simplexes
Mau-Hsiang Shih and Kok-Keong Tan
235
245
Cer:erali zatim;s of Convex Suprcmi zation Duality Ivan Singer
On the Asymptotic Behavior of i\lmost-Drbits of Commutative Semigroups in Banach Spaces fvataru Takahashi and Jong Yeoul Park
On a Factorization of Operators Through a Subspace of c 0 Yau-Chuen Wong
Trace Formula for Almost Lie Algebra of Operators and Cyclic One-Cocycles
253
271
295
Daoxing Xia
299
Index
309
Contributors
JEAN-PIERRE AUBIN Mathematical Research Center, University of ParisDauphine, Paris, France CHANG KUNG-CHING (ZHANG GONG-QING) Department of Mathematics, Peking University, Beijing, People's Republic of China GONG-NING CHEN Department of Mathematics, Beijing Normal University, Beijing, People's Republic of China CHARLES K. CHUI Department of Mathematics, Texas A&M University, College Station, Texas CARLL. DeVITO* Tucson, Arizona
Department of Mathematics, University of Arizona,
HALINA FRANKOWSKA Mathematical Research Center, University of ParisDauphine, Paris, France ANDRZEJ GRANAS Department of Mathematics, University of Montreal, Montreal, Quebec, Canada CHUNG-WEI HA Department of Mathematics, ]l;ational Tsing Hua University, llsinchu, Taiwan, Republic of China NO RIMI CHI HIRANO Yokohama, Japan
Department of Ma thematics, Yokohama National University,
CHARLES D. HORVATH Canada
Champlain Regional College, St. Lambert, Quebec,
TATSURO ICHIISHI** Iowa City, Iowa
Department of Economics, The University of Iowa,
GEORGE ISAC Department of ~!athematics, Royal Military College, St.-Jean, Quebec, Canada *Current affiliation: **Current affiliation:
Naval Postgraduate School, Monterey, California Ohio State University, Columbus, Ohio
ix
x HIDETOSHI KOMIYA
Contributors College of Commerce, Nihon University, Tokyo, Japan
HANG-CHIN LAI Institute of Mathematics, National Tsing Hua University, Hsinchu, Taiwan, Republic of China M. J. LAI Department of Mathematics, Texas A&M University, College Station, Texas FON-CHE LIU Institute of Mathematics, Academia Sinica, Taipei, Taiwan, Republic of China ROBERT H. LOHMAN Department of Mathematics and Sciences, Kent State University, Kent, Ohio JONG YEOUL PARK Department of Mathematics, Busan National University, Busan, Republic of Korea W. V. PETRYSHYN Department of Mathematics, Rutgers University, New Brunswick, New Jersey SIMEON REICH* Department of Mathematics, The Technion-Israel Institute of Technology, Haifa, Israel I. SHAFRIR Department of Mathematics, The Technion-Israel Institute of Technology, Haifa, Israel VICTOR L. SHAPIRO, Department of Mathematics and Computer Science, University of California, Riverside, Riverside, California SHI SHU-ZHONG** Paris, France
Mathematical Research Center, University of Paris-Dauphine,
MAU-HSIANG SHIH Department of Mathematics, Chung Yuan University, ChungLi, Taiwan, Republic of China THADDEUS J. SHURA Department of Mathematics and Sciences, Kent State University at Salem, Salem, Ohio IVAN SINGER Department of Mathematics, National Institute for Scientific and Technical Creation, Bucharest, Romania WATARU TAKAHASHI Department of Information Science, Tokyo Institute of Technology, Tokyo, Japan KOK-KEONG TAN Department of Mathematics, Statistics and Computing Science, Dalhousie University, Halifax, Nova Scotia, Canada KENSlJKE TANAKA Japan
Department of Mathematics, Niigata University, Niigata,
*Current affiliation: California
The University of Southern California, Los Angeles,
**Current affiliation: China
Nankai University, Tianjin, People's Republic of
xi
Contributors
M. TI-IERA France
Department of Mathematics, University of Limoges, Limoges,
YAU-CHUEN WONG Department of Mathematics, The Chinese University of Hong.Kong, Hong Kong DAOXING XIA Tennessee
Department of Mathematics, Vanderbilt University, Nashville,
Conference Participants*
Archbold, R. (Aberdeen, Scotland) Aubin, J.-P. (Paris, France) Browder, F.E. (Chicago, IL) Beer, G.A. (Los Angeles, CA) Bruck, R.E. (Los Angeles, CA)
Kottman, C .A. (Springfield, VA) Laetsch, T. (Tucson, AZ) Lai, H.-C. (Taiwan, China) Lai, M. (College Station, TX) Lange, L.H. (San Jose, CA)
Border, K.C. (Pasadena, CA) Cantwell, J.C. (St. Louis, MO) Chang, K.C. (Beijing, China) Chu, II. (College Park, MD) Chuan, J.-C. (Taiwan, China)
Lapidus, M. (Iowa City, IA) Lassonde, M. (Clermont, France) Lau, A.T. (Edmonton, Alberta) Lee, T.Y. (University, AL) Lin, B.-L. (Iowa City, IA)
Chui, C. K. (College Station, TX) Crandall, M.G. (Madison, WI) Deprima, C.R. (Pasadena, CA) Donoghue, W. F. (Irvine, CA) Fournier, G. (Sherbrooke, Quebec)
Liu, F.-C. (Taiwan, China) Liu, S.-C. (Taiwan, China) McLinden, L. (Urbana, IL) Mikusinksi, P. (Orlando, FL) Namioka, I. (Seattle, WA)
Frankowska, H. (Paris, France) Geraghty, M.A. (Iowa City, IA) Granas, A. (Montreal, Quebec) Greim, P. (Charleston, SC) Gretsky, N. (Riverside, CA)
Ng, K.F. (Hong Kong) Pak, H. (Detroit, MI) Petryshyn, W. V. (New Brunswick, NJ) Phelps, R. (Seattle, WA) Powell, M.H. (Santa Cruz, CA)
Halmos, P.R. (Santa Clara, CA) Hartig, D.G. (San Luis Obispo, CA) Hilario, R. (Chino, CA) Himmelberg, C.J. (Lawrence, KS) Hirano, N. (Yokahama, Japan)
Reich, S. (Los Angeles, CA) Schechter, M. (Irvine, CA) Shapiro, V.L. (Riverside, CA) Shekhtman, B. (Riverside, CA) Shih, H.-H. (Taiwan, China)
Hoffman, H. (Los Angeles, CA) Horvath, C. (Montreal, Quebec) Hu, T. (Taiwan, China) Ichiishi, T. (Iowa City, IA) Kakutani, S. (New Haven, CT)
Struble, R. (Rayleigh, NC) Swaminathan, S. (Halifax, Nova Scotia) Takahashi, W. (Tokyo, Japan) Tan, K.-K. (Halifax, Nova Scotia) Thera, M. (Limoges, France)
Kalisch, G.K. (Irvine, CA) Knill, R.J. (New Orleans, LA) Kamiya, H. (TokyQ, Japan) Komuro, N. (Hokkaido, Japan)
Wang, J.L.-M. (University, AL) Wong, Y. C. (!long Kong) Wu, S.Y. (Iowa City, IA) Xia, D. (Nashville, TN)
*Other than those from The University of California at Santa Barbara xiii
List of Publications of Ky Fan 1.
Sur une representation des fonctions abstraites continues, C.R. Acad. Sci. Paris 210(1940), 429-431.
2.
Surles types homogenes de dimensions, C.R. Acad. Sci. Paris 211 (1940), 175-177.
3.
Espaces quasi-reguliers, quasi-normaux et quasi-distancies, C.R. Acad. Sci. Paris 211(1940), 348-351. Caracterisation topologique des arcs simples dans les espaces accessibles de M. Frechet, C.R. Acad. Sci. Paris 212(1949), 1024-1026.
4. 5.
Surles ensembles possedant la propriete des quatre points, C.R. Acad. Sci. Paris 213(1941), 518-520.
6.
Surles ensembles monotones-connexes, les ensembles filiformes et les ensembles possedant la propriete des quatre points, Bull. Soc. Royale Sci. Liege 10(1941), 625-642.
7.
Sur le theoreme d'existence des equations differentielles dans !'analyse generale, Bull. Sci. Math. 65(1941), 253-264.
8.
Sur quelques notions fondamentales de !'analyse generale, J. Math. Pures et Appl. 21(1942), 289-368.
9.
Expose sur le calcul symbolique de Heaviside, Revue Scientifique 80 (1942), 147-163.
10.
Sur le comportement asymptotique des solutions d'equations lineaires aux differences finies du second ordre, Bull. Soc. Math. France 70(1942), 76-96.
11.
Les fonctions asymptotiquement presque-periodiques d'une variable entiere et leur application a l'etude de !'iteration des transformations continues, Math. Zeitschr. 48(1943), 685-711.
12.
Une propriete asymptotique des solutions de certaines equations Lineaires aux differences finies, C.R. Acad. Sci. Paris 216(1943), 169-171.
13.
Quelques proprietes caracteristiques des ensembles possedant la propriete des quatre points et des ensembles filiformes, C.R. Acad. Sci. Paris 216(1943), 553-555.
14.
Nouvelles definitions des ensembles possedant la propriete des quatre points et des ensembles filiformes, Bull. Sci. Math. 67 (1943), 187-202.
15.
Entfernung zweier zufalligen Grossen und die Konvergenz nach Wahrscheinlichkeit, Math. Zeitschr. 49(1943/44), 681-683.
16.
Sur !'extension de la formule generale d'interpolation de M. Borel aux fonctions aleatoires, C.R. Acad. Sci. Paris 218(1944), 260-262.
17.
Apropos de la definition de connexion de Cantor, Bull. Sci. Math. 68(1944), 111-116.
18.
Un theoreme general sur les probabilites associees a un systeme d'evenements dependants, C.R. Acad. Sci. Paris 218(1944), 380-382.
19.
Une definition descriptive de l'integrale stochastique, C.R. Acad. Sci. Paris 218(1944), 953-955.
20.
Sur !'approximation et !'integration des fonctions aleatoires, Bull. Soc. Math. France 72(1944), 97-117.
21.
Le prolongement des fonctionnelles continues sur un espace semixv
xvi
Publications List of Ky Fan ordonne, Revue Scientifique 82(1944), 131-139.
22.
Conditions d'existence de suites illimitees d'evenements correspondant a certaines probabilites donnees, Revue Scientifique 82 (1944), 235-240.
23.
Generalisations du theoreme de M. Khintchine sur la validite de la loi des grands nombres pour les suites stationnaires de variables aleatoires, C.R. Acad. Sci. Paris 220(1945), 102-104.
24.
Remarques sur un theoreme de M. Khintchine, Bull. Sci. Math. 69 (1945), 81-92.
25.
Two mean theorems in Hilbert space, Proc. Nat. Acad. Sci. U.S.A. 31 (1945), 417-421.
26.
(with M. Frechet) Introduction a la topologie combinatoire, I. Initiation, Vuibert, Paris (1946).
Spanish translation: Introduccion a la topologfa combinatoria (traducida por D.A.H. Nogues), Editorial Universitaria de Buenos Aires (la. edicion, 1959; 2a. edicion, 1961; 3a edicion, 1967). English translation: Initiation to combinatorial topology (translated by H.W. Eves), Prindle, Weber & Schmidt, Boston, LondonSydney (1967). 27.
On positive definite sequences, Ann. of Math. 47(1946), 593-607.
28.
(with S. Bochner) Distributive order-preserving operations in partially ordered vector sets, Ann. of Math. 48(1947), 168-179.
29.
On a theorem of Weyl concerning eigenvalues of linear transformations, I, Proc. Nat. Acad. Sci. U.S.A. 35(1949), 652-655.
30.
On a theorem of Weyl concerning eigenvalues of linear transformations, II, Proc. Nat. Acad. Sci. U.S.A. 36(1950), 31-35.
31.
Partially ordered additive groups of continuous functions, Ann. of Math. 51(1950), 409-427. --
32.
Les fonctions definies-positives et les fonctions completement monotones (Memorial des Sci. Math., Fasc. 114), Gauthier-Villars, Paris (1950).
33.
(with A. Appert) Espaces topologiques intermediaires (Actualites Sci. et Industr., Fasc. 1121), Hermann, Paris (1951).
34.
Maximum properties and inequalities for the eigenvalues of completely continuous operators, Proc. Nat. Acad. Sci. U.S.A. 37 (1951), 760-766.
35.
Fixed-point and minimax theorems in locally convex topological linear spaces, Proc. Nat. Acad. Sci. U.S.A. 38(1952), 121-126.
36.
Note on a theorem of Banach, Math. Zeitschr. 55(1952), 308-309.
37.
A generalization of Tucker's combinatorial lemma with topological applications, Ann. of Math. 56(1952), 431-437.
38.
(with N. Gottesman) On compactifications of Freudenthal and Wallman, Proc. Kon. Nederl. Akad. Wetensch. Amsterdam, Ser. A, 55 (1952), 504-510. Also in: Indag. Math. 14 (1952), 504-510.
39.
Minimax theorems, Proc. Nat. Acad. Sci. U.S.A. 39(1953), 42-47.
40.
Some remarks on commutators of matrices, Archiv. der Math. 5 (1954), 102-107. (with R.A. Struble) Continuity in terms of connectedness, Proc. Kon.Nederl. Akad ..Wetensch. Amsterdam, Ser. A, 57(1954), 161-164. Also in: Indag. Math. 16(1954), 161-164.
41.
Publications List of Ky Fan
xvii
42.
(with A.J. Hoffman) Lower bounds for the rank and location of the eigenvalues of a matrix, Contributions to the Solution of Systems of Linear Equations and the Determination of Eigenvalues, 117-130. (National Bureau of Standards Applied Math. Series, vol. 39, Washington 1954.)
43.
Inequalities for eigenvalues of !Iermitian matrices, Contributions to the Solution of Systems of Linear Equations and the Determination of Eigenvalues, 131-139. (National Bureau of Standards Applied Math. Series, vol. 39, Washington 1954.)
44.
(with G.G. Lorentz) An integral inequality, Amer. Math. Monthly 61(1954), 626-631.
45.
(with J. Todd) A determinantal inequality, J. London Math. Soc. 30 (1955), 58-64.
46.
(with A.J. Hoffman) Some metric inequalities in the space of matrices, Proc. Amer. Math. Soc. 6(1955), 111-116.
47.
(with O. Taussky and J. Todd) Discrete analogs of inequalities of Wortinger, Monatsh. Math. 59(1955), 73-90.
48.
Some inequalities concerning positive-definite Hermitian matrices, Proc. Cambridge Philos. Soc. 51(1955), 414-421.
49.
(with O. Taussky and J. Todd) An algebraic proof of the isoperimetric inequality for polygons, J. Washington Acad. Sci. 45(1955), 339-342.
50.
(with I. Glicksberg) Fully convex normed linear spaces, Proc. Nat. Acad. Sci. U.S.A. 41(1955), 947-953.
51.
A comparison theorem for eigenvalues of normal matrices, Pacific J . Ma th . 5 ( 19 5 5 ) , 911- 913 .
52.
On systems of linear inequalities, Linear Inequalities and Related Systems, 99-156 (Annals of Math. Studies, No. 38, Princeton Univ. Press, 1956). Also in Russian translation, Moscow, 1959.
53.
(with G. Pall) Imbedding conditions for Hermitian and normal matrices, Canad. J. Math. 9(1957), 298-304.
54.
(with P. Davis) Complete sequences and approximations in normed linear spaces, Duke Math. J. 24(1957), 183-192.
55.
(with I. Glicksberg and A.J. Iloffman) Systems of inequalities involving convex functions, Proc. Amer. Math. Soc. 8(1957), 617-622.
56.
Existence theorems and extreme solutions for inequalities concerning convex functions or linear transformations, Math. Zeitschr. 68 (1957), 205-216.
57.
Topological proofs for certain theorems on matrices with non-negative elements, Monatsh. Math. 62(1958), 219-237.
58.
Note on circular disks containing the eigenvalues of a matrix, Duke Math. J. 25(1958), 441-445.
59.
Linear inequalities and closure properties in normed linear spaces, Seminar on Analytic Functions, Vol. 2, 202-212 (Institute for Advanced Study, Princeton 1958).
60.
(with I. Glicksburg) Some geometric properties of the spheres in a normed linear space, Duke Math. J. 25(1958), 553-568.
61.
On the equilibrium value of a system of convex and concave functions, Math. Zeitschr. 70(1958), 271-280.
62.
Convex sets and their applications, Argonne National Laboratory, Argonne (19 59) .
xviii
Publications List of Ky Fan
63.
(with A,S. Householder) A note concerning positive matrices and M-matrices, Monatsh. Math. 63(1959), 265-270.
64.
Note on M-matrices, Quarterly J. Math. (1960), 43-49.
65.
Combinatorial properties of certain simplicial and cubical vertex maps, Archiv der Math. 11(1960), 368-377.
66.
A generalization of Tychonoff's fixed point theorem, Math. Ann. 142(1961), 305-310.
67.
On the Krein-Milman theorem, Proceedings of Symposia in Pure Mathematics, Vol. 7, Convexity, 211-219 · (Amer. Math. Soc., Providence 1963).
68.
(with R, Bellman) On systems of linear inequalities in Hermitian matrix variables, Proceedings of Symposia in Pure Mathematics, Vol. 7, Convexity, 1-11 (Amer. Math. Soc., Providence 1963).
69.
Invariant subspaces of certain linear operators, Bull. Amer, Math. Soc. 69(1963), 773-777.
70.
Invariant cross-sections and invariant linear subspaces, Israel J. Math. 2(1964), 19-26.
(Oxford Second Ser.) 11
Russian Translation: Matematika: Period, Sb. Perevodov Inostran. Statei. (Mathematics: Periodical Collection of Translations of Foreign Articles) 13 (6), Moscow, 1969. MR 40 #7057, 71.
Inequalities for M-matrices, Proc. Kon. Nederl. Akad. Wetensch. Amsterdam, Ser. A, 67(1964), 602-610. Also in:
Indag. Math. 26(1964), 602-610.
72.
Sur un theoreme minimax, C.R. Acad. Sci. Paris 259(1964), 39253928.
73.
A generalization of the Alaoglu-Bourbaki theorem and its applications, Math. Zeitschr. 88(1965), 48-60.
74.
Invariant subspaces for a semigroup of linear operators, Proc. Kon. Nederl. Akad. Wetensch. Amsterdam, Ser A. 68(1965), 447-451, Also in:
Indag. Math. 27(1965), 447-451.
Russian Translation: Matematika: Period. Sb. Perevodov Inostran. Sta tei. (Mathematics: Periodical Collection of Translations of Foreign Articles) 13 (6), Moscow, 1969. MR 40 #7057. 75.
Applications Of a theorem concerning sets with convex sections, Math. Ann. 163(1966), 189-203. Also in: Contributions to Functional Analysis Heidelberg-New York, 1966).
(Springer, Berlin-
76.
Some matrix inequalities, Abhandl. Math. Seminar Univ. Hamburg 29 (1966), 185-196.
77.
Sets with convex sections, Proceedings of the Colloquium on Convexity, Copenhagen, 1965, 72-77 (K¢benhavns Univ. Mat. Inst., 1967.
78.
Subadditive functions on a distributive lattice and an extension of Szasz's inequality, J. Math. Anal. Appl. 18(1967), 262-268.
79.
Inequalities for the sum of two M-matrices, Inequalities, Proceedings of a Symposium, 105-117 (Academic Press 1967).
80.
Simplicial maps from an orientable n-pseudomanifold onto sm with the octahedral triangulation, J. Combinatorial Theory 2(1967), 588-602.
Publications List of Ky Fan
xix
81.
An inequality for subadditive functions on a distributive lattice, with application to determinantal inequalities, Linear Algebra and ~ - 1 (1968), 33-38.
82.
On infinite systems of linear inequalities, J. Math. Anal. Appl. 21(1968), 475-478.
83.
A covering property of simplexes, Math. Scandinavica 22(1968), 17-20.
84.
Asymptotic cones and duality of linear relations, J. Approximation Theory 2(1969), 152-159. Also in: Inequalities II, Proceedings of the Second Symposium on Inequalities, 179-186 (Academic Press, 1970).
85.
Extensions of two fixed point theorems of F. E. Browder, Math. Zeitschr. 112(1969), 234-240. Abstract in:
Set-valued mappings, selections and topological
properties of 2x. Proceedings of the Conference held at SUNY at Buffalo, May 1969, edited by W. M. Fleischman (Lecture Notes in Mathematics, Vol. 171, Springer, 1970), 12-16. 86.
On a theorem of Pontryagin, Studies and Essays presented to Yu-Why Chen, Mathematics Research Center, National Taiwan Univ. (1970), 197-200.
87.
A combinatorial property of pseudomanifolds and convering properties of simplexes, J. Math. Anal. Appl. 31(1970), 68-80.
88.
On local connectedness of locally compact Abelian groups, Math. Ann . 18 7 ( 19 7 o) , 114 -116 .
89.
Simplicial maps of pseudomanifolds, Annals of the New York Academy of Sciences 175 (Art. 1), (1970), International Conference on Combinatorial Mathematics, 125-130.
90.
Combinatorial properties of simplicial maps and convex sets, Proceedings of the Twelfth Biennial Seminar of the Canadian Mathematical Congress: Time Series and Stochastic Processes, Convexity and Combinatorics (Canadian Math. Congress, 1970) 231-241.
91.
On the singular values of compact operators, J. London Math. Soc. (2) 3 (1971), 187-189.
92.
A minimax inequality and applications, Inequalities III, Proceedings of the Third Symposium on Inequalities, 103-113 (Academic Press, 1972).
93.
Covering properties of convex sets and fixed point theorems in topological vector spaces, Symposium on Infinite Dimensional Topology (Annals of Math. Studies, No. 69, Princeton Univ. Press, 1972), 79-92.
94.
Generalized Cayley transforms and strictly dissipative matrices, Linear Algebre a n d ~ - 5(1972), 155-172.
95.
Fixed-point theorems in functional analysis (Supplementary manual for recording of a lecture), Audio Recordings of Mathematical Lectures, No. 46, Amer. Math. Soc. (1972), 7 pages.
96.
On Dilworth's coding theorem, Math. Zeitschr. 127(1972), 92-94.
97.
On real matrices with positive definite symmetric component, Linear and Multilinear Algebra 1(1973), 1-4.
98.
Sums of eigenvalues of strictly J-positive compact operators, J. Math. Anal. Appl. 42(1973), 431-437.
xx 99.
Publications List of Ky Fan On similarity of operators, Advances in Math. 10(1973), 395-400.
100.
On strictly dissipative matrices, Linear Algebra and Appl. 9 (1974), 223-241.
101.
Two applications of a consistency theorem for systems of linear inequalities, Linear Algebra and Appl. 11(1975), 171-180.
102.
Orbits of semi-groups of contractions and groups of isometries, Abh. Math. Sem. Univ. Hamburg 45(1976), 245-250.
103.
Extension of invariant linear functionals, Proc. Amer. Math. Soc. 66 (1977), 23-29.
104.
Analytic functions of a proper contraction, Math. Zeitschr. 160 (1978), 275-290.
105.
Distortion of univalent functions, J. Math. Anal. Appl. 66(1978), 626-631.
106.
Julia's lemma for operators, Math. Annalen 239(1979), 241-245.
107.
(with T. Ando) Pick-Julia theorems for operators, Math. Zeitschr. 168 (1979), 23-34.
108.
Fixed-point and related Game Theory and Related Game Theory and Related 1978; Managing Editors: Holland, 1979, 151-156.
109.
Schwarz's lemma for operators on Hilbert space (lecture presented at the Romanian-American Seminar on Operator Theory and Applications, 20-24 March 1978), Analele stiintifice ale Universitatii Al. I. Cuza Iasi, Supliment la tomul 25, s. I a 1979, 103-106.
110.
Harnack's inequalities for operators, General Inequalities, Oberwolfach; edited by E.F. Beckenbach), Birkhauser Verlag, BaselBoston-Stuttgart, 1980; 333-339.
111.
A further generalization of Shapley's generalization of the Knaster-Kuratowski-Mazurkiewicz theorem, Game Theory and Mathematical Economics (edited by 0. Moeschlin and D. Pallaschke), North-Holland, 1981, 275-279.
112.
Evenly distributed subsets of Sn and a combinatorial application, Pacific J. Math. 98 (1982), 323-325.
113.
Iteration of analytic functions of operators, Math. Zeitschr. 179 (1982), 293-298.
114.
Iteration of analytic functions of operators. II, Linear and Multilinear Algebra 12(1983), 295-304.
115.
Normalizable operators, Linear Algebra and Appl. 52/53(1983), 253-263.
116.
Some properties of convex sets related to fixed point theorems, Ma th . Ann. 2 6 6 ( 19 8 4 ) , 519- 5 3 7 .
117.
An identity for symmetric bilinear forms, Linear Algebra and Appl. 65 (1985), 273-279.
118.
The angular derivative of an operator-valued analytic function, Pacific J. Math. 121 (1986), 67-72.
theorems for non-compact convex sets, Topics (Proceedings of the Seminar on Topics, Bonn/Hagen, 26-29 September, 0. Moeschlin and D. Pallaschke), North-
Nonlinear and Convex Analysis
and Heavy Viable Solutions to Control Problems JEAN-PIERRE AUBIN Mathematical Research Center, University of ParisDauphine, Paris, France
ABSTRACT We introduce the concept of viability domain of a set-valued map, which we study and use for providing the existence of smooth solutions to differential inclusions. We then define and study the concept of heavy viable trajectories of a controlled system with feedbacks.
Viable trajectories
are trajectories satisfying at each instant given constraints on the state.
The controls regulating viable trajectories evolve
according a set-valued feedback map.
Heavy viable trajectories
are the ones which are associated to the controls in the feedback map whose velocity has at each instant the minimal norm.
We con-
struct the differential equation governing the evolution of the controls associated to heavy viable trajectories and we state their existence. DEDICATION I would have liked to find an original way to dedicate this lecture to Professor Ky Fan, but I did not see any better solution than to simply confess that it is both an honor and a pleasure to have been invited to this conference held in his honor. I have been deeply influenced by the theorems discovered and proved by Professor Ky Fan, and, in particular, by his 1968 famous inequality.
Let me just repeat what I tell my students 1
DOI: 10.1201/9781003420040-1
Aubin
2
when I begin to teach the Ky Fan inequality.
I tell them a lot
of stories, how the young Ky Fan came to Paris in 1939 for one year with only a metro map, how he had to survive during the darkest years of the history of my country, how he met Frechet and worked with him, etc.
But most important, I choose the
Ky Fan inequality as the best illustration of the concept of "labor value" of a theorem. Indeed, most of the theorems of nonlinear functional analysis are equivalent to the Brouwer fixed point theorem.
But when
we prove that statement (A) is equivalent to statement (B), there is always one implication, say "A implies B", that is more difficult to prove than the other one. ment (B)
We then can say that state-
"incorporates" more labor value than statement (A).
An empirical law shows that the more labor value a theorem incorporates, the more useful it is.
And my point is that among
all the theorems equivalent to the Brouwer fixed point theorem I know, the Ky Fan inequality is one which is the most valuable.
1.
VIABLE SOLUTIONS TO A CONTROL PROBLEM
IRn, U: X + X be a set-valued map with closed graph Let X and f : Graph U-->- X be a continuous map. We consider the control problem with feedbacks
( 1. 1)
i)
x' (t) = f(x(t),u(t)}
ii)
for almost all t _:: _ 0,
iii)
u (t) E u(x (t))
x(0) = x 0 given in Dom U
Instead of selecting a solution x( •) to (1) which minimizes a given functional, as in optimal control theory( 1 ),we are only selecting solutions which are viable in the sense that, given a closed subset KC X ( 1 • 2)
\It > 0
,
X ( t)
E K
A first issue is to provide necessary and sufficient con-
Smooth and Heavy Viable Solutions to Control Problems
3
ditions linking the dynamics of the system (described by f and U) and the constraints bearing on the system (described by the closed subset K) such that the viability property
vx 0 EK , there exists a solution to ( 1)
( 1 • 3)
viable in K
holds true.
This allows us to describe the evolution of the
viable controls u(•),
(the controls which govern viable solutions)
A second issue is to provide conditions for having smooth viable solutions to a control problem, in the sense that the viable control function is absolutely continuous instead of being simply measurable. A third issue is to give a mathematical description of the "heavy viable solutions" of the control system which we observe in the evolution of large systems arising in biology and economic and social sciences.
Such large systems keep the same control
whenever they can and change them only when the viability is at stakes, and do that as slowly as possible.
In other words, heavy
viable solutions are governed by those controls who minimize at each instant the norm of the velocity of the viable controls. In the case when f(x,u)
= u, system (1) reduces to the differen-
tial inclusion x' (t) E U(x (t)), x (0)
=
x0 :
heavy (viable) solu-
tions to this system minimize at each instant the norm of the acceleration of viable solutions; in other words, they evolve with maximal inertia. Hence the name heavy viable solutions (or inert viable solutions). 0,
and
such that
sup Re £ 1 (co h (Ue)) s c - t
.e 1 (h (x)),
< c s Re
r = 1.
a contradiction, so that (3) holds when
k
the function into
A.
In the general case of
k(y) = h(ry)
defined by
is holomorphic
(3) now follows directly from the result just proved
and from the convexity of balls. co u
Obviously,
of
(3) whether
A
is hermitian and continuous, we have
bra pseudonorm on (3).
if not, by a separation
c
is bounded in
e
A,
A
(4)
is a simple consequence
or not.
If the involution for
([3], [8]) that
and is continuous on
A,
whence
p
is an alge-
(5)
is valid by
For the last part of the theorem, as before, it will be enough to r = 1.
verify the result in the case
But this follows directly from
the mean value property for vector-valued holomorphic functions [5, p. 99]:
(2n) -1J2n
h(x) = h(f(O)) = for each (6).
x
in
A
( 7)
0
!xl 0 < 1
with
and
p(x) < 1,
where
f
is as in
That completes the proof of the theorem. We note that there exists an alternative proof of Theorem 1
REMARK.
by deriving it from the formula COROLLARY 1. Let
Suppose that
A
(7). has hermitian and continuous involution.
be a domain containing
G
rt
for some
0 < r s 1,
r,
and define
A1 as in Theorem 1 and holomorphic function on
A0 = [x EA: o(x) c OJ. If k is a complex 0, then (3)-(5) of Theorem 1 are valid with
D
h(x)
replaced by k(x)
where
r
Proof. function
A0
and
defined by
k(x)
= (2ni) -1 ~k(A) (Ae-x) -1 dA, n
( 8)
is any contour that surrounds It is well known [10] that k
defined by so that
(8) k : Ao
Ar, into A, virtue of the properties of whenever
by the function
O
~
rt.
A0
o(x)
in
is open in
0. A
and that the
is a continuously differentiable mapping of -➔
A
is holomorphic by definition.
In
p given above, we have A0 ~ r co Ue Corollary 1 now follows from Theorem 1.
3. SIMPLIFIED DISCUSSION OF THEOREMS I, II, III, AND THEIR EXTENDED VERSIONS In this section we shall introduce the use of the results of the pre-
Proofs of Theorems of von Neumann, Heinz, and Ky Fan
19
ceding section in the simplified proofs of Theorems I, II, III, and their extended versions of Tao [11].
A result (see [6, p. 109] and
[10, p. 309]) of fundamental importance for us is the following LEMMA I.
Let
E
normal operator
be the spectral resolution of the identity for a V
in
B(H),
and suppose that
O
morphic function on a domain 0
< r s 1.
Then
g(rV)
Jcc l V) 0
containing
g
is a complex holoin
o(rV)
C
for some
r,
g(rt)dE(t),
( 9)
where the integral exists with respect to the norm topology in
B(11),
and fig(rV)li = sup[ Jg(r,\.)
i: ,\.
( 10)
E CT\V)j.
We note that the involution on
B(Il),
the Hilbert space adjoint, is
clearly hermitian and continuous, and that [T le B(Il): l:T
B (II) l
Proof of Theorem I. since
o(V)
s
[Tl
CT
s p(T) =
iiTL
for each
V
so that
l}.
Observe that
O ~
lies on the unit circle.
E
~
o(V)
By Corollary l
with
in
U
r = l
and
v(V)
for
by (10),
i: u (T) ,i
s sup[ u (V)
V E
uJ
l} s 1,
s sup[ lu(,\.) I
as desired. Proof of Theorem II. each
V
D,
in
0
s;•i
(Re v(V)w,w) for all
w
"
. 2
ii WI! C,
is in
•
in E
w,w Since
where E (U) = (E(u)w,w). w,w is a positive measure on o(V) Rev(,\.) ~
t.
It is well known [10]
0
~
0
whose total variation
is continuous on the compact set
tor each
w
in
I!,
i.e.,
Re v(V)
~
o(V) 0
by
The result now follows directly from (3) of Corollary 1.
Proof of Theorem I I I . exist real
v(v)*
Re v(ciO)dEw,w(O)
II,
(Re v(V)w,w)
definition.
and since
i Re v ( e , ) dE ( 0 ) ,
J,
and therefore
that each
l
r
we have
r2,1
Re v(V)
By (9), with
r,
T
If
0 < r < 1,
is a proper contraction of and a contraction
;\
T
of
B (H),
B(H)
there
such that
s sup[ I h (r,\.) I : i ,\. I = 1] < l. Thus, as before, By /! h (T) i applying a similar argument to that used in the preceding theorem, we
T
=
rT.
can obtain
Re k (T) > 0
for each
11
T ii < 1.
It is natural to expect that the proofs of the extended versions of Theorems I, II, and III, given by Tao [11], should be also greatly
20
Chen
simplified by the same approach mentioned above. case.
For notations, let
the set of all A
A
A
0.
Recall [11] that
A
and ~(T)
A
g(A)
is normal in
denotes a
B(H) 0 =[TE B(H): cr(T) c OJ,
function
~
C
and denote by
B(H)-valued holomorphic functions
g(cr)g(A) = g(A)g(cr) main
be a domain in
These are indeed the
on
O
B(H)
A
g
on O for all
NH{O)
such that A,cr in
B{H)-valued function, with dothat arise from a
B(H)-valued
by the formula
A
g(T) where
r
(11)
is as in (1). A A
A
THEOREM IV. Suppose that u,v are in NH(O) and h is in NH (t.), each of which commutes with T in B(H). Then Theorems I, II, and III A v, A hold J.'f u, v, and h therein are replaced by u, and h/,, respectively. Proof.
Consider only the extended result with respect to Theorem I.
In a similar way one can verify the others.
By a slight extension of
Theorem 10.38 in [10], we can assert that the function by (11) is a holomorphic mapping of Lemma I
B(H) 0
into
B(H).
~(T)
defined
Instead of
we use an analogous lemma, that is Lemma 6.2 of [11].
Thus,
as before, we are able to conclude that the extended result is valid. We conclude with the remark that each of Theorems 3.3, 3.4 (2)-(4), and 3.5 of [11] may be shown to be valid by means of the same methods, and is merely an immediate consequence of Theorem 1 and Corollary 1 together with either Lemma I or Lemma 6.2 of [11]. REFERENCES 1. N. Dunford and J.T. Schwartz, "Linear Operators," Interscience, New York, pt. I, 1958; pt. II, 1963. 2. Ky Fan, Analysis functions of a proper contraction, Math. Z. 160 (1978), 275-290. 3. L. Harris, Banach algebras with involution and Mobius transformations, J. Functional Analysis 11(1972), 1-16. 4. E. Heinz, Ein v. Neumannscher Satz uber beschrankte Operatoren im Hilbertschen Raum, Nachr. Akad. Wiss. Gottingen Math.-Phys. KI.II (19 5 2) , 5- 6. 5. E. Hille and R.S. Phillips, "Functional Analysis and Semigroups," Rv. Ed., Amer. Math. Soc., Providence, R.I. 1957. 6. B. Sz.-Nagy and C. Foias, "Harmonic Analysis of Operators on a Hilbert Space," North-Holland, Amsterdam, 1970. 7. J. Von Neumann, Eine Spektraltheorie fur allgemeine Operatoren eines unitaren Raumes, Math. Nachr. 4(1950/1951), 258-281. 8. V. Ptak, On the spectral radius in Banach algebras with involution, Bull. London Math. Soc. 2(1970), 327-334.
Proofs of Theorems of von Neumann, Heinz, and Ky Fan 9. F. Riesz and B. Sz.-Nagy, "Functional Analysis," Fredrick Ungar, New York, 1955. 10.
w.
Rudin, "Functional Analysis," McGraw-Hill, New York, 1973.
11. Tao Zhiguang, Analytic operator functions, J. Math. Anal. Appl. 103(1984), 293-320.
21
Determinant and Lagrange Interpolation in Rs CHARLES K. CIIUI Department of Mathematics, Texas A&M University, College Station, Texas HANG-CHIN LAI Institute of Mathematics, National Tsing Hua University, Hsinchu, Taiwan, Republic of China
The problem of multivariate polynomial interpolation is very old. Among the papers published during the last decade, we only include [1-16] in the References. Let N~ lem can be stated as follows: study the location of nodes (or sample points) in
R•
such that for every data {f; : i
= 1, ... , Nn, p(w) =
= (n;•). The prob{ x; : i = 1, ... , N n
there is a unique polynomial
L aiwi
lilSn
with total degree n which interpolates the given data at the nodes, namely: p(x;)
=
1, ... , N~. Here and throughout, we use the usual multivariate notation wi
w~•,
j1
+ ... + is
= w{
1 •••
f;, i
=
Iii =
(j1 , ... , is E Z+), etc. Of course, the problem is equivalent to the study of the
nonsingularity of the square matrix
= [x{JT, Iii :S: n, is the i th column of the matrix. While the determinant of this matrix in the cases = l is the wellknown Vandermonde determinant, which is always nonzero for arbitrary
where 0 F,Eint R(T,F,)
( 5)
The purpose of this section is to provide a sufficient condition for (5) when F, is an equilibrium of F, i.e. OEF(F,). We shall apply results of Section 1. The set of solutions ST(F,) is closed in w111 (O,T) whenever Graph (F) is closed in JR.n
Consider the continuous linear operator A from the Banach space w111 (O,T) into the finite dimensional space JR.n x JR
n_
defined by A(x)
1 1
x(T) for all xEW '
Theorem 1.1. states then that if x
0
(0,T)
denotes the constant tra-
jectory x 0 (•) = F, and {w(T) :wECST(F,)(x 0 lation (5) holds true.
)}
= ]Rn then the re-
Let B denote the closed unit ball in JR.n.
We say that a
set-valued map Fis Lipschitzian (in the Hausdorff metric) on an open neighborhood V of F, if for a constant L> 0 x,yE V F ( x) C F ( y) + L llx-y II B
and all
Frankowska
52
Thanks to this property we can compute a subset of c 8 T(s) (x 0 ) : Theorem 2.1. Assume that F has closed graph and is Lipschitzian around the equilibriums- Then every solution of the differential inclusion
(6)
I
w' ( t) E CF ( s, 0) w (t)
a.e. in [ 0, T]
w (0) = 0
belongs to c 8 T(s) (x 0
).
•
The proof of the last result is based on a Filippov Theorem (1967]. We say that the inclusion (6) is aontrottabte if its reachable set at some time T > 0 is equal to the whole space. Theorems 1.1. and 2.1. together imply Theorem 2.2. Assume that F has closed graph and is Lipschitzian around the equilibriums- The inclusion (3) is locally controllable around s if the inclusion (6) is controllable.• Remark. Actually the idea of the proof of Theorem 1.1. allows to prove a stronger result: We denote by co F(~) the closed convex hull of the set F(~). Theorem 2.3. Assume that F has closed graph and is Lipschitzian around the equilibrium S• The inclusion (3) is locally controllable around s if the inclusion
(7)
~ w'
~
E
w(O)
cl [CF ( s , 0) w + CcoF ( ~ ) ( 0) ] = 0
is controllable. • The proof requires a very careful calculation of variations of solutions (see Frankowska (1984]). A necessary condition for the controllability of the inclusions (6), (7) is Dom CF ( s , 0) : = { w E lRn : CF ( s , 0) w 'f' j} = lRn
Local Invertibility of Set-Valued Maps
53
Whenever it holds true the right-hand sides of (6),
(7)
are set-valued maps whose graphs are closed convex cones. Such maps, called "closed convex processes", are set-valued analogues of linear operators.
The controllability of such
differential inclusions is the issues of the next section. Before, we provide the following Using Theorem 2.3 one can obtain a classical result
Example.
on local controllability of control system (4) without assuming too much regularity.
Let Ube a compact in lRmand let
f : lRn x u-+ lRn be a continuous function. (E;,u)ElRnxU, f(E;,u)
=
Assume that for some
0 and for some S>O, L>O and all uEU;
x,y Es+ SB
~ llf(x,u) - f(y,u) 11.:_Lllx-yll
?~; (• ,u) Theorem 2.4.
is continuous on
s + SB
If the sublinearized differential inclusion
) w' E
~!
? w(O) =
(E;,u)w+Cco f(E;,U) (0) 0
is controllable, then the system (4) is locally controllable around t;. 3.
Controllability of Convex Processes A convex process A from lRn to itself is a set-valued map
satisfying Vx, yEDom A,\,µ> 0 ,
\A(x) +µA(y) C A(\x+µy)
or, equivalently, a set-valued map whose graph is a convex cone.
Convex processes are the set-valued analogues of linear
operators.
We shall say that a convex process is closed if its
graph is closed and that it is strict if its domain is the whole space.
Convex processes were introduced and studied in
Frankowska
54
Rockafellar [1967], [1984]).
[1970],
[1974)
(see also Aubin-Ekeland
We associate with a strict closed convex process A
the Cauchy problem for the differential inclusion
(8)
{
x' (t) EA(x(t))
a.e.
x(O) = 0
We say that the differential inclusion (8) is controllable if the reachable set R:={x(t):xEw 111 (0,t) is a solution of (8), t>O} is equal to the whole space JRn. A particular case of (8) is a linear control system
( 9)
f x'
- Fx+GU
lx(O~
=
uEU
0
where U is an m-dimensional space and FE L (JRn, JRn), GEL (JRm, JRn) are linear operators. We observe that the reachable set R(T,O) of (8) at time T is convex.
Since OE A(O) the family {R(T,0) }T>O is increasing.
Moreover, R = U R(T,0). Hence (8) is controllable if and only 'I'>O if it is controllable at some time T>O, i.e.3 T>O such that R(T,O) = JRn a)
The rank condition Let A be a strict closed convex process.
Set A1 (0)
A(O)
and for all integer i > 2 set Ai (0) = A(Ai- 1 (0)) Theorem 3.1. if and only if
The differential inclusion (8) is controllable
Local Invertibility of Set-Valued Maps
55
•
In the case of system (9) for all x E m.n Ax
Fx + Im G.
Thus (-A)
m
(0) = Im G + F (Im G) + ... +F
m-1
(Im G)
The Cayley-Hamilton theorem implies then the Kalman rank condition for the controllability of the linear system (9):
r k [ G, FG, . . F
n-1
G] = n
Theorem 3.1. is a consequence of the following "Eigenvalue" criterion for controllability b) We say that a subspace P of m.n is invariant by a strict closed convex process A if A (P) c P. A real number A is called an eigenvalue of A if Im(A-\I) f f
m.n, where I denotes the identity operator.
Theorem 3.2.
The differential inclusion (8)
is controllable if
and only if A has neither proper invariant subspace nor eigen-
•
values.
It is more convenient to write the above criterion in a "a:Aal forn1":
c)
"Eigenvector" criterion for controllability The convex processes can be transposed as linear operators.
Let A be a convex process; we define its transpose A* by pEA*(q)
¢>
V(x,y) EGraph A,
<
It can easily be shown that\ is an eigenvalue of A if and only if for some q E Im (A-,\I)
1
, q f
0
AgEA*q We call such a vector q f is equivalent then to
O an eigenvector of A*.
Theorem 3.2
56
Frankowska
The differential inclusion (8) is controllable if
Theorem 3.3.
and only if A* has neither proper invariant subspace nor eigenvectors.
A
The proof of Theorem 3.3 is based on a separation theorem and the KY-FAN coincidence theorem [1972].
(See Aubin-
Frankowska-Olech [1985] ). Examples:
a)
Let F be a linear operator from :m.n to its elf, L
be a closed convex cone of controls and A be the strict closed convex process defined by A(x)
:=Fx+L
Then its transpose is equal to A* (q)
I
= F:q ,,
When L = { 0}, i.e. , when A= F, we deduce that A* = F*, so that transposition of co11'.;ex processes is a legitimate extension of transposition of linear operators. Consider the control system
{X '
( 1 0)
= Ax+u,
uEL
lx(O)=O
Corollary 3.4. The following conditions a"e equivalent. a)
the system (10) is controllable
b)
For
SOffi,;
m +1 L + F(L)+ ••• +Fm(L)=
L-F(L)+ ..• +(-1)mFm(L) = :m.n (see Korobov (1980]).
c)
F has neither proper invariant subspace containing L nor eigenvalue A satisfying Im(F-AI) + L-1 :m.n.
d) e)
F* has neither proper invariant suspace contained in . . L. + L+ nor eigenvector in n-1 the subspace spanned by L, F(L) , ..• ,F (L) is equal to JRn and F* has no eigenvector in L+ (see Brammer (1972])
A
Local Invertibility of Set-Valued Maps
b)
57
Consider the control system with feedback in m.2 :
( 11 )
+y +U +
X
1
=
y
I
= -x +w
XV
XU
U, W E U
= [ 0 , 1]
j +1 vEV(x)=}- 1
X?O
x
either
0
(a)
there is x0
(b)
there is y0 E Y such that g(x,y 0 ) 2).. - c for all
E
X such that
f(x 0 ,y)
s)..
for all
y
£
Y ,
or Proof. Sup f(x,y)
YEY
>)..
(i ) ,
=>
hence
(ii )
If
(ii) (a)
Inf Sup f(x,y)
2)..
x EX
does not hold, then for each x EX ,
XEX YEY
which implies by (i) that
This research was supported in part by a grant from the National Research Council of Canada. 61
DOI: 10.1201/9781003420040-6
62
Granas and Liu
Sup Inf g(x,y) cc>...
Thus there is
y EY X EX
g ( x ,y O ) cc >.. -
for al l
£
(ii)=> (i)
Let
x EX . y
=
y 0 c Y such that So ( i i ) ( b)
Inf g(x,y 0 ) cc>.. -
X ,X
£,
i.e.,
ho l d s .
Inf Sup f(x ,y) EX y EY
We may assume that
y
> - oo.
If
X
y < +
00
we apply ( i i ) with
,
(ii)(b)
holds, i.e.,there is
x EX ,
or
>..
=y
-
y 0 EY
such that
Sup Inf g ( x ,Y) cc y - 2 £
Sup Inf g(x ,y) cc y .
Thus {i) holds.
does not hold, hence there is
If
y = +
y 0 e Y such that
does not hold for this
g(x,y 0 ) cc.\.-£ =y-2£
But since
y eY x eX
y eY X eX
Since (ii)(a)
£
for all
0 is arbitrary, we have
E >
oo,
>..,
then for any
g(x,y 0 ) cc>.. -
>.. E R , (i i )(a)
for all
£
x EX
Thus Sup Inf g(x,y) cc A X EX
y eY
If we let
>..
+
+
oo,
then
complete.
{l .2)
Lemma.
Let
Sup Inf g{x,y) yeY xEX
f ,g : X
x
+
00
•
E
•
Hence (i) holds.
The proof is
Y + R be two numerical functions and suppose
that one of the following conditions holds :
(*)
X
is a compact topological space and
X for each
(**)
+
f{x,y)
is lower semi-continuous on
y
+
q(x,y)
is upper semi-continuous on
ye Y.
Y is a compact topological space and Y for each
x
x c X .
Then the following two statements are equivalent (i)
Inf Sup f(x,y) s Sup Inf g(x,y) xeX ycY yeY xeX
(ii)
For each
>.. e R , either
(a)
there is
such that
f(x 0 ,y) s >.. -
for all
y
(b)
there is
y 0 e Y such that
g(x,y 0 ) cc>..
for all
x EX .
E
y ,
or
Some Minimax Theorems Without Convexity
Proof.
Suppose that(*) holds.
straightforward manner from (l. l).
63
The implication (II)
To show that (i)
=>
=>
(i)
(ii) , let
follows in a \ E R be given
and assume that (ii)(b)
does not hold. By Lemma (l.l), for each n = 1,2, ... , we have Inf Sup f(x,y) ,; A +nl- . But, since x + Sup f(x ,y) is lower semi continuous yEY XEX YEY l on X , there is xn E X such that Sup f(xn,y) = Inf Sup f(x,y) ,; A + n . From XEX yEY YEY this letting n + oo , and using the compactness of X and the lower-semicontinuity on
X of
Thus
we get a point X + Sup f(x,y) ' YEY (ii)(a) holds and we have shown that
XO E X such that Sup f(x 0 ,y) ,; \ . YEY ( i ) => ( i i ) The proof of our asser-
tion in the case (**) being similar is omitted. Now we are able to formulate the main result of this section representing a generalization of a known theorem of Ky Fan Ll J. (l.3)
Theorem.
Let
X,Y
be two compact topological spaces and
f,g
Xx Y • R two real-valued functions such that
(i)
x
+
f(x,y)
is lower-sernicontinuous on
X for each
y
E
Y
(ii)
y
+
g(x ,y)
is upper-semi continuous on
Y for each x
E
X.
Then the following conditions are equivalent:
A. exists
For any two finite sets
XO E X and
For each
for all
,;
,; n
X,
{yl , ... ,ylll}
and
l ,; k ,; m
x0 EX
1c
for all
y
E
Y
~ 1c
for all
X E
X
y0 E Y such that g(x,y0 )
C.
l
\ER, either there is f ( x0 ,Y) ,;
or there is
C
Yo E y such that
f(xo,yk) ,; g(xi ,yo) B.
{xo,· .. ,xn}
The following minimax inequality holds
such that
C
y
'
there
64
Granas and Liu
Min Sup f(x,y) s Max Inf q(x,y) . XEX YEY YEY XEX Proof.
We al ready know from Lemma (1.2) that B and
The implication C
=>
C
are equivalent.
A being obvious, it remains only to prove that
A implies
Assume that A is verified ; we are now going to prove that B holds. A
ER be given.
B Let
Define
L(y) = {x E X
f(x,y)
,s A}
for each y E Y
R(x) = {y E Y
q(x,y)
~ A}
for each x EX
n L(y)"' ¢ YEY are closed subsets of
To show that B holds
is equivalent to show that either
X and Y or n R(x) "' ¢ • Since the sets L(y) and R(x) XEX respectively. it is sufficient to show that either {L(y)}yEY or {R(x)}xEX has the finite intersection property.
y does not have the finite inYE m tersection property: there are y1 , .•. ,Ym E Y such that n L(yk') = ¢ , i.e., k=l
(*)
Max f(x,y.) 1 l,sk~
Let now {x 1 , .•. ,xn}
>
A
Suppose that {L(y)}
for each x EX
be any finite subset of X . By (i) there exist x0 EX and
y0 E Y such that
or :5
:5
n •
From(*) we have then
Thus y0
n
€
n R(xi) . Hence {R(x)} XE X has the finite intersection property and i =l
the proof is complete. 2.
A Minimax Inequality of the von Neumann type. In this section, we prove a minimax theorem which represents a further gene-
ralization of a minimax theorem of Ky Fan ll J under "convexity conditions" milder than in our previous note [4].
Some Minimax Theorems Without Convexity
65
In what follows given positive integer t , we let [tJ = {i E N I s i s t} t-1 t t and denote by ti = {(x 1 , ... ,xt) ER I x1 , ... ,xt.: 0 , i~l xi = l} the standard simplex in Rt (2.1)
We introduce first some terminology. Definition.
function on X
Let X be a set and F
=
{f} a family of real-valued
We shall say that X is finitely-convex with respect to
F
provided for any f 1 ,f2 , ... ,fn E F and any x1 ,x 2 EX there is a point xd such that
for all
i E [nJ ; X is called finitely-concave with respect to
above definition the symbol (2.2)
Definition.
"s"
is replaced by
11
2:"
for
in the
•
Let X,Y be sets and f : Xx Y + R a real-valued
function ; consider the family {fy}yEY of functions f(x,y)
F , if
fy : X + R where fy{x) =
x EX . We shall say that X is finitely f-convex (resp. f-concave)
if X is finitely convex (resp. concave) with respect to the family {fy}yEY Similar definitions can also be given for
Y , by considering the family
We can now state and prove the main result of this section : (2.3) Theorem.
Suppose that X and
Y are two non-empty compact spaces
and that f,s,t,g : X x Y + R are four real-valued functions such that (i )
f(x,y) s s(x,y) s t(x,y) s g(x,y)
(ii)
x
(iii)
X is finitely s-convex ; i.e., for any x1 ,x 2 • X and y1 , ... ,Yn E Y there
+
f{x,y)
for all
( X ,y)
€
X
X
y ;
is lower semicontinuous on X for each y E Y ;
is x EX such that s(x,yi) s
½[s{x 1 ,yi)
~ s(x 2 ,yi)J for all
= l ,2, ... ,n ;
(iv)
Y is finitely t-concave ; i.e., for any y1 ,y 2 E Y and x1 , ... ,xm EX there exist y E Y such that t(xj,y).:} [t(xj,yl) ~ t(xj,y 2)J for all j = l ,2, ... ,m ;
(v)
y
+
g(x,y)
is upper semicontinuous on
Y for each x EX .
66
Granas and Liu
Then Min Sup f(x,y) s Max Inf g(x,y) . X Y Y X Proof.
Let
E >
be given.
0
To establish our assertion, it is clearly
sufficient to show that Min Sup f(x,y) s Max Inf{g(x,y)} X
Y
Y
X
+ E
= Max{Inf g(x,y) Y
+ E} •
X
Hence, in view of Theorem (l.3), we need to show only that for any two finite sets x1 , ••• ,xn "- X and y1 , ••• ,Ym "- Y there exist x0
"-
X and
y0
"-
Y such that
(l )
"- [nJ,
for all
k "- [mJ
.
by
For this purpose define a function n
m
L L
G(a,B)
i =l k= l
G
L:in-l
x
L:im-l -->R
aiBk s(xi ,yk)
for a = (a1 , ... ,an) "- lln- l ,
By the von Neumann Minimax
Theorem in Rn , there exist - ) "- lln- l a- = (-a 1 , ... ,an
and
such that (2)
G(a,B) s G(a.S)
for all
(a,B) "- L:in-l
From the definition of G and (2)
x
L:im-l .
we get
(3)
Next we choose
a= (;1 , ...
,;n) "- lln-l
and
B=
(s1 , ... ,Bi
11 )
rationals as their components such that (4)
(5)
n
I
i =l m
I
k=l
ai s (xi ,yk) s Sk t(xi ,Yi)
s
n ai s(xi ,yk) + f.2 i =l
I
m
I
k=l
E
Bk t(xi ,yk) + 2
for all
k d[mJ
for all
"- [n J
with dyadic
Some Minimax Theorems Without Convexity
Now, because exist
x0
and
EX
X is finitely s-convex and y0
s(xo,yk)
(7)
t(xi,y 0 )"" i
E
I
$
i =l Ill
I
k= l
In I and
Y is finitely t-concave, there
Y such that
E
n
(6)
Let
67
~ ai s(xi,yk)
for a 11
Bi t(xi,yk)
for a11
k
E
Un J be given.
k
E 1111 )
c Ln J .
Then using the assumption (i) of the
Theorem and (4), (5,), (6), and (7) we get (8)
(9)
Finally, from (3), (8), and (9) we conclude that
Since
i c [nJ and
kc [mJ were fixed arbitrarily, this implies (1), and the proof
of the theorem is complete. As an immediate consequence we obtain the following result established in our Note [4] :
(2.4)
Theorem.
Let
X,Y
be two compact spaces and let
f,s,t,g : X xy
be four real-valued functions satisfying conditions (i), (ii) and (v) of Theorem (2.3).
Assume furthermore that
(iii)*
For any
xc
x 1 ,x 2 , ... ,xn c X and
X such that for a 11 s(x ,Yl
n
'° I
i =l
(a 1 ,a. 2 , ... ,an) c lln-l
y c Y
ai s(xi ,_vl
there is a point
➔R
68
Granas and Liu
For any y1 ,y 2 , ... ,Ym y
€
Y such that all t(x,y) ~
€
Y and
x
€
m
I
j =l
there is a point
X
t(x,yi)
i3J•
Then Min Sup f(x,y) X
Proof.
Y
~
Max Inf g(x,y) . Y
X
It is enough to observe that (iii)* (resp. (iv)*)
implies the condi-
tion (iii) (resp. (iv)) of Theorem (2.3) . By taking in (2.3)
f = s = t = g we obtain among special ca~es of Theorem
(2.3) the following result (2.5) Theorem.
Let X,Y be two non-empty compact spaces and f
Xx Y ... R
be a real-valued function such that : (i)
x ... f(x,y)
is lower semi-continuous on
X for each y
€
Y
(ii)
y ... f(x,y)
is upper semi-continuous on
Y for each x
€
X
(iii)
For any x1 ,x 2 f(x,yi) ~
(iv)
€
X and y1 ,y 2 , ... ,Yn
½[f(x 1 ,yi)
+ f(x 2 ,yi)J
Y there is x such that
€
for all
For any y1 ,y 2 EX and any x1 ,x 2 , ... ,xm EX
i
E
LnJ ;
there is y
€
Y such that
The the following minimax equality holds Min Max f(x,y) = Max Min f(x,y) . X
Y
Y
X
As a special case we have the following (2.6) Corollary (Nikaido- von Neumann).: Let X and
Y be two compact
convex subsets in linear topological spaces and f : Xx Y... R a real-valued function such that
Some Minimax Theorems Without Convexity
x ... f{x,y)
(a)
(b) y ... f(x,y)
69
is lower semi-continuous and convex on X for each y is· upper semi-continuous and concave on
E
Y;
for each x EX.
Y
Thenthe following minimax equality holds Min Max f{x,y) = Max Min f(x,y)
X Y
Y X
A theorem concerning systems of inequalities.
3.
In this section we establish a further generalization of a theorem of Ky Fan [2] concerning systems of inequalities ; our result extends three previous generali-
zations given in C3J,[4J and [5J. X be a non-empty set and
Let
G =
F = {f},
{g}
two families of real-
valued functions defined on X ; we write F s G provided for each f g
€
G such that f(x) s g(x)
for all
€
F there is
x EX
We need the following Definition. A family
(3.1)
any f 1 ,f 2
for all
€
F there is an
f
E
F =
{f}
is weakly F-concave on X provided for
F such that
x e X . *) We may now formulate our second main result Theorem.
(3.2)
Let X be a compact space and
families of real-valued functions defined on
*)
We recall that a family
F = {f}
~
-~ aifi(x)
l
=l
Ky Fan [lJ.
for all
x
E
be three non-empty
X satisfying :
is called F-concave on X provided for any
f 1 ,f2 ,... ,fn e F and a = (a l ' a 2 '· · · •0 n ) f{x)
F,G,H
E
~n-1 0
there is f
E
F
such that
X . This notion was first formulated by
70
Granas and Liu
2)
Each function in
F is lower-semicontinuous on
3)
X is finitely convex with respect to
4)
The family
H is weakly
X
G;
F-concave.
Then (I)
(II)
A ER , the following alternative holds :
For each (a)
There is
h EH
such that
h(x)
(b)
There is
x0 E X such that
> A
for all
x EX ;
f(x 0 ) s: A for all
f
E
F.
Min Sup f(x) ,; Sup Inf h(x) XEX fEF hEH XEX Proof.
Since (II) follows e:isily from (I), we only need to prove (I). A ER
To show (I), let
be given ;
we assume that the condition (I)(a)
is not verified and proceed to show that (I)(b) holds. n and each
For each positive integer
f E F , we let Sn (f) = {x E X I f(x) s: 1' + ~}
Clearly each
Sn(f)
is closed in
first that for each fixed intersection property. gl , ... ,g 111
E
G
and
X and therefore compact.
n , the family of compact sets
Let
f 1 , ... ,f111
h1 , ... ,h111
E I{
F(x ,E;)
G(x ,E;)
H(x,E;) = for
m
I
F , G, H
E;ifi (x)
'
.I E;igi (x) l =1
'
i =l
F be given.
for all
=
Since
x EX
and
Xx 6m-l ➔ R defined
has the finite
F s: G ,; H , there are
i E CmJ. by
Ill
m
.I i=,;.h.(x) l =1 l l
,
x E X and We are going to apply Theorem (2.3) with
g
{Sn(f)}fEf
such that
fi(x) s g(x) s hi(x) Consider now the functions
E
We are going to show
H.
We observe first that
Y
6m-l ,
f
F, s
= t =G
and
71
Some Minimax Theorems Without Convexity
i' ,
F(,,()
(*)
E,
G(x ,E,)
➔
E, ➔ H(s ,E,)
is lower semi-continuous on X for each t, m- l is concave on 6 for each X E X ; m- l is upper semi-continuous on 6 for each
Furthermore since (by the assumption) given
x1 ,x 2 c X there is
x
E
X
X
6m- l
E
X E
X •
is finitely convex with respect to
G ,
such that E [m
J ,
and this implies
for all
E,
E
L\m-l .
Thus
X is finitely G-convex.
the hypotheses of Theorem (2. 3) and
g
=
H.
are verified with
Y
= lim- l , f
=
F ,
s
=t =G
Consequently by Theorem (2.3) we get
(**)
Min Max X 6m-l Let
Because of this fact and(*),
n
E
F(x ,E,) s Max 6m- l
N be a given integer.
Inf H(x ,E,) X
From (**), by applying Theorem ( l. 3), we
have the following alternative, either there is
(a)
such that
., ,\ +
ln
for all
t, EL\nl-1'
or there is
(b)
H(x ,r;) :, ,\ + -l n
such that
for all
X E
X •
By adapting arguments of S. Simons [5J and those of our earlier Note [3J we are going to show that (b) is not true. holds.
By restri ctinq
we may assume that
E,
Suppose to the contrary that the condition (b)
the following set-up to some face of L\m-l is in the interior of
L\
m-1
•
Define
if necessary,
h : L\ m-1 -, R u
{ - ro}
by k(E,) = Inf H(x,E,) XEX
Since E, ,
H(x,E,) ;,, F(x,E,) k
and
x
➔
,
F(x,E,)
for
E,
E
L\m-l .
is lower semi-continuous on
is real-valued and is bounded from gel0\'1 on
L\m-l .
X for each
Furthermore, since
k
Granas and Liu
72
m-1
is concave on the convex
t:,
,
k is continuous at
and k is continuous at 1; , there is a point n
=
Because
1; •
(n 1 , .•. ,11,i) with dyadic raand
tional coordinates n1 ,n2 , ... ,11,i such that n E Int t:,m-l since the family H h(x)
m 2
I
n1hi(x) = H(x,n)
i =l
2
k(n)
Finally,
> >..
x EX . This contradicts our assumption that (I)(a)
for all
> >.. •
is weakly F-concave, there is h E H such that
{h}
a:
k(n)
is not verified.
Thus
the condition (b) does not hold and hence in view of (a) there is x0 EX such that F(xo,1;)
s A
nl
T
for all
I;
m-1
t:,
€
•
This means that
XO
€
family {Sn{f)}fEF has the finite intersection property.
m n s {f.) ' i.e.' the i =l n ,
A = n S {f) "' f EF 11 n is compact and An~ An+l) that n An"' ..
f E F.
for all
n =l
Thus (I}{b)
Next it is clear Obviously, if holds and the proof
As an immediate consequence of (3.2) we obtain the following result Let X be a compact space, Y an arbitrary set and
(3.3) Theorem.
f,g,h : Xx Y ➔ R three real-valued functions verifying l.
f (x ,y) s g (x ,Y) s h( x ,y)
2.
x
➔
f{x,y)
for a11
(x ,y)
€
Xx Y;
is lower semi-continuous on X for each y E Y .
Assume furthermore that one of the following conditions (A), (B), C)
is
satisfied : Al..
For any x1 ,x 2 E X and y1 ,y2 , ... ,Yn E Y there is x E X such that l [g{x, ,Yi) + g(x2,Yi) J for all s 2 g ( x,yi)
(A)
A4.
For any y1 ,y 2 E Y there is h( x,y)
2
y€
i1 [h(x,y 1 ) + h(x,y 2 )J
i
E (n
€
X.
Y such that for a11
x
J •
Some Minimax Theorems Without Convexity
83.
there is
For any x1 ,x 2 , ... ,xm E X and
x E )(
such that
g(x,y) 84.
(B)
73
m
.l
s
l
=1
aig(xi ,y)
For any y1 ,y 2 , ... ,Yn
E
for all
y
€
Y
there is
Y and
y E Y such that h(x ,y)
(C)
{"
n 2:
l
j =1
S}(x ,Y)
for all
x EX •
X is convex
C4.
X-+
C5.
The family {hy}yEY
g(x,y)
is convex on
for each y E y •
X
is F-concave.
Then the following two assertions hold : (I)
For any
A
ER either (a)
x0 EX
there is
f(x 0 ,y)
such that
s A
for all
y EY
or
(s) there is y0 E Y such that h(x,y 0 ) (II)
A
>
x EX .
for all
Inf Sup f(x,y) s Sup Inf h(x,y) X
y
y
X
By taking in Theorem (3.3),
f
=
g
=
h we obtain as a special case the fol-
lowing result, which should be compared with Theorem (2.5) : (3.4) Theorem. f : Xx
Y-+
for each y 1.
Let
X be a compact space,
R a function such that x-+ f(x,y) E
Y.
Y an arbitrary set and
is lower semi-continuous on
X
Assume that the following two additional conditions are verified there is
x- EX
such that
74
Granas and Liu
2.
there is
y-
E
Y such that
f(x,y);:, [f{x,y 1 ) +f(x,y 2 )J
for all
x EX.
Under the above hypotheses we have : (I )
For any
"
€
R either
there is
(a)
XO
E
f ( X ,y) 0
X such that
"
$
or there is
( 13)
(II)
Inf Sup f(x ,y) X
€
y
f(x,y 0 )
>
Yo
for a11
y
€
y
such that for a11
"
X E
X•
Sup Inf f ( x ,y) . y X
y
As a special case we get the following
(3.5) Corollary. (Ky Fan-tnkaido-K!,eser).
Let
X be a compact convex space,
Y an arbitrary set and f : Xx Y ➔ R a real-valued function such that x ➔ f(x,y) is lower-semicontinuous on X for each y E Y . Assume that one of the following additional hypotheses is verified : 1.
( Ky Fan).
The family
2.
(Nikaido)
Y is a convex subset of a vector space and
Y for each 3.
(Kneser)
{f} y y, y
is F-concave. y
➔
f(x,y)
is concave on
x EX
Y is a vector space and
y
+
f(x,y)
Then Inf Sup f(x ,y) X
y
Sup Inf f(x,y) . y
X
is affine on
Y for each
x EX •
Some Minimax Theorems Without Convexity
75
References 1.
2.
Ky Fan, Minimax Theorems, Proc. Nat. Acad. Sci:, U.S.A., 39 (1953), 42-47. , Existence theorems and extreme solutions for inequalities concerning convex functions or linear transformations, Math. Z. 68 (1957), 205-217.
3.
A. Granas & F.C. Liu, Remark on a theorem of Ky Fan concerning systems of inequalities, Bull. Inst. Math. Acad. Sinica 11 (1983), 639-643.
4.
_________ , Theoremes du minimax sans convexite, C.R. Acad. Sci. Paris, t. 300, no ll (1985), 347-350.
5.
S. Simons, Remark on a remark of Granas and Liu concerning systems of inequalities, Bull. Inst. Math. Acad. Sinica (to appear).
Compactness and the Minimax Equality CHUNG-WEI HA Department of Mathematics, National Tsing Hua University, Hsinchu, Taiwan, Republic of China
Let E be a real locally convex Hausdorff topological vector space, E' be the topological dual of E and let X be a bounded subset of E. (We suppose throughout that X and Y are nonempty.) We denote by cr(E,E') and t(E,E'), respectively, the weak topology and the Mackey topology on E induced by E'. It is easy to show (see, e.g., [SJ, Remark 6) that if Xis cr(E,E')-compact, then, for any subset Y of E', I]
inf sup inf S sup inf sup t E ,jT (X) y E Y XE t E $(Y) XE X yE I]
(1)
where 3-" (X) and ,y; (Y) denote the families of nonempty finite subsets of X and Y, respectively. Moreover, if Xis convex and cr(E,E')-compact, then for any convex subset Y of E', inf sup ( y ,x> < y EY XE X
sup inf XE X y E y
(2)
(2) follows immediately from (1) and the fact that for any convex sets
X C E and y CE I ' we always have I]
and
inf E $ (Y) sup
t E $(X)
sup XEX
inf YE 11
inf yEY
sup XE t
< Y,x>
= inf yE y
sup < Y,x> XEX
(3)
( y ,x) = sup inf ( y ,x) XE X yE y
(4)
(see, e.g., [SJ, Lemma 11). The relation (2) is the minimax equality referred to in the title; obviously (2) with the inequality sign reversed is always true. In this paper, we shall show that each of the properties (1) and (2) in a way characterizes the weak compactness of 77
DOI: 10.1201/9781003420040-7
78
Ha
X. Our main results are Theorems 1 and 2 below. As applications, we obtain directly from Theorem 2 DeBranges' s characterization [ 1] of the t(E' ,£)-open convex sets in E' and a theorem of Fan [2] generalizing the Alaoglu-Bourbaki theorem. If E is assumed to be t(E,E')-complete, then Theorem 2 becomes a result of Simons [5]. In the following we shall assume the notations given above. The reader is referred to [4] for the terminology not herein defined. Theorem 1. Let X be a bounded set in E. If (1) holds for any convex set Y in E' for which the left-hand side of (1) is positive, then Xis relatively a(E,E')-compact in E. Proof: Suppose that X is not relatively a(E,E' )-compact in E, then the a(E'*,E')-closure of X in the algebraic dual E'* of E' contains an element$ which is not in E. For any elements x 1 , ... ,xm in X, there exists y E E' satisfying $(y) > 1 but I< y,xi>I < 1/3 for all 1 ~ i ~ m. Let y = {y EE 1 :
Then Y is convex in E'
t
sup E S(X)
'
$(y)
~
l}
(5)
and
inf yE y
sup xEt
I< y ,x >I
~
1/3.
On the other hand, since$ is in the a(E'*,E')-closure of X, for any elements y 1 , ... ,Yn in Y, there exists x E X satisfying I$ (y j )- < y j ,x > < 1/3 and so < y j ,x > > 2/3 for all 1 ~ j ~ n. Thus,
I
inf r)ES(Y)
sup xEX
inf y Er)
< y,x>
~
2/3.
(6)
Hence (1) does not hold for the convex set Y defined in (5), which by (6) makes the left-hand side of (1) positive. This completes the proof. Theorem 2. Let X be i bounded convex set in E. lf (2) holds for any convex set Y in E' for which the left-hand side of (2) is positive, then Xis relatively a(E,E')-compact in E. Clearly Theorem 2 follows from Theorem 1, ( 3) and (4). Now we shall use Theorem 2 to prove the following result of DeBranges ([1], Theorem 4):
Weak Compactness and the Minimax Equality
79
set- in Theorem 3. Let Ube a convex E'. U is t(E',E)-open if and only if A n u = fJ for any convex set H in E' such that H n u = fJ' - -where A denotes the a(E' ,£)-closure of H in E'. Since the a(E'E)-closure and the t(E' ,£)-closure of a Proof: convex set Hin E' are the same, the condition is clearly necessary. Conversely, by a translation we may assume that the origin O of E' belongs to U. It suffices to prove that O is in the t(E' ,£)-interior of U. Let X be the polar U0 of U in E, that is, X = {x E E : (u,x>
~
1
for
u E U}
then X is bounded, closed, convex and contains O. (X is bounded because U is radial round all of its points.) We shall show that Xis a(E,E')-compact. To this end, let Y be a convex set in E' and a> 0 be a real number such that a < inf sup yEY XEX
(7)
By dividing both sides of (7) by a, we may assume that a= 1. Then (7) implies that Y n U = {J. By Zorn's lemma, there exists a maximal convex set H in E' containing Y but disjoint from U. Clearly H is a(E' ,E)-closed. It follows from the separation theorem that there exists an element x E E and a real number ~ such that ( y,x>
~
~
(8)
>
for all y E Y and u E U. Since OE U, Thus, (8) implies that x E X and so 1
~
sup
XE
X
~
> 0 and so we can assume
~
= 1.
(9)
inf < y ,x > yE Y
By Theorem 2, X is a(E,E' )-compact. Now if y ff U, then by applying the preceding argument to {y} in place of Y, we obtain an element x EX such that< y,x> ~ 1. Hence {u
which shows completed.
that
E
E'
O is
u ,X ) in
< 1
for
X E
X}
C
the t(E' ,£)-interior of U.
U The proof is
80
Ha
Next we shall show that the following result of Fan ([2], Theorem 1) is a consequence of Theorem 2. Theorem 4.
Let X be a bounded set in E.
U = {u
E
u ,x >
E'
~
1
for
x E X}
of X in E' has ~ nonempty t(E' ,E)- interior, cr(E,E')-compact in E.
(10)
then X is relatively
Proof: Since X and the closed convex hull of X u {0} have the same polar in E', we may assume that Xis closed, convex and contains 0. To apply Theorem 2 as in the proof of Theorem 3, let Y be a convex set in E' such that 1 < inf sup ( y,x) yEY xEX
(11)
Then (11) implies that Y n U = {J. Since U has a nonempty t(E' ,£)interior, by the Hahn-Banach theorem there exist a nonzero element x EE and a real number ~ such that
~
~
~
(u,x>
(12)
for all y E Y and u E U. Since O E U, ~ ~ 0. Suppose first that = 0. By the bipolar theorem ( [4], p.126) x E X. But then X also contains all the positive multiples of x. Since x 1 0, this contradicts the boundedness of X. Hence ~ > 0 and so we may assume that ~ = 1. Then again x EX and so (12) implies (9). This completes the proof. ~
As pointed out in [2], the t(E,E')-interior of the set U in (10) does not necessarily contain the origin of E', and so Theorem 4 strictly generalizes the Alaoglu-Bourbaki theorem. Our proofs show a dual relationship between Theorems 3 and 4. Theorem 5. Let X be a closed bounded set in E andlet -the - closed convex hull C of X in E be t (E ,E' )- complete. Then the following statements are equivalent: (a) (b)
Xis cr(E,E')-compact; If y is~ real number and Z is the convex hull of an equi-
Weak Compactness and the Minimax Equality
81
continuous sequence {y.} in E' such that J
inf ZEZ
sup ( z-u,x XEX
(13)
for any a(E' ,E)- cluster point u of the sequence {yj}, then y ~ O; (c) li a, ~ are real numbers, {yj} is an equicontinuous sequence in E' and {xi} is~ sequence in X such that (yj,xi>
~
a
for
j
( y. ,x. )
~
~
for
j > i
J
Then
1
~
i
l
(14)
; (d) The inequality (1) holds for any equicontinuous (convex) set Yin E' (for which the left-hand side of (1) is positive). a~~
Proof: Suppose that Xis a(E,E')-compact. (b) is a property of the vector space C(X) of real-valued continuous functions on X. For given a a(E' ,£)-cluster point u of the sequence {yj} in E', there exists a subsequence of {yj} which converges pointwise on X to u (see [4], p.185). Without disturbing (13) we may assume that {yj} itself converges pointwise on X to u. Since {yj} is uniformly bounded in C(X), from Lebesgue's bounded convergence theorem, {yj} converges to u in the weak topology of C(X). Since the closures of Zin C(X) in the weak and the uniform topology coincide, u can be arbitrarily approximated uniformly on X by elements in Z. This shows that y ~ 0. To prove that (b) implies (c), let Z be the convex hull of the sequence {yjl given in (c). Then (14) implies that (13) holds for any a(E' ,£)cluster point u of {yjl, where y = a - ~- Hence a~~- Now we assume (c). Let Y be an equicontinuous set in E', for which we suppose that the left-hand side of (1) is not -oo; the right-hand side of (1) is not +oo. Let a' ~ be real numbers such that a < ~
>
rt E !7(Y) XE X
inf y E rt
( y ,x)
(15)
sup inf E !7(X) yEY
sup xEt
( y ,x)
(16)
inf
f;
We choose arbitrarily an x 1 E X such that < Yi ,x 1 > exists y 2 E Y such that Y1,··•,Yn-l E Y have been
sup
element y 1 EY, then by (15) there exists > a. For the element x 1 EX by (16) there < ~ for all 1 ~ i ~ n-1, and again by (15) there exists XnE X such that< yj,xn> > a for all 1 ~ j ~ n. Continuing in this way we obtain an equicontinuous sequence {y.} in E' J and a sequence {xi} in X satisfying (14). Hence a~ ~- This proves (d). In proving (d) implies (a), we shall make use of the assumption that the closed convex hull C of Xis t(E,E')-complete. Suppose that X is not a(E,E')-compact, then the closure of X in the algebraic dual E'* of E' contains an element$ which is not in the t(E,E')-completion of E. By Grothendieck' s completion theorem ( see [ 4] , p .148), there exists a balanced convex equicontinuous set U in E' such that the restriction of $ to U is not a(E' ,£)-continuous at OE U. After suitably scaling U, the proof of Theorem 1 goes through for the set Y={yEU:
$(y)~l}
The proof is completed. It is clear that if a bounded set X in E satisfies Theorem 5(b), so does also its closed convex hull C in E. Thus, we obtain a new proof of Krein' s theorem ( see [ 4] , p .189), which says that if the closed convex hull C of a a(E,E' )-compact set X in E is l(E,E' )complete, then C is also a(E,E')-compact. It is noteworthy that Theorem 5(b) was used by Pryce (3) as a starting point in his proof of James's theorem on weakly compact sets (see also [5]). Finally we remark that, if X is assumed to be convex, then Theorem 5(d) can be replaced by the condition that (d') The relation (2) holds for any equicontinuous convex set Y in E' (for which the left-hand side of (2) is positive). Thus, we obtain Simons's result Theorem 15 in [5).
References 1.
L. DeBranges, Vectorial topology, J. Math. Anal. Appl., 69(1979), 443-454.
2.
K. Fan, A generalization of the Alaoglu-Bourbaki theorem and its applications, Math. Z., 88(1965), 48-60.
3.
J. D. Pryce, Weak compactness in locally convex spaces, Proc. Amer. Math. Soc., 11(1966), 148-155.
4.
H. H. Schaefer, Topological vector spaces, Springer-Verlag, New York-Heidelberg-Berlin, 1970.
5.
S. Simons, Maximinimax, minimax, and antiminimax theorems and a result of R. C. James, Pacific J. Math., 40(1972), 709-718.
Volterra Equations with Positive Kernels NORIMICHI HIRANO Yokohama, Japan
1•
Department of Mathematics, Yokohama National University,
Our purpose in this paper is to consider
Introduction.
the existence of solutions to the nonlinear Volterra equation u(t)
( 1. 1 )
+
f
t O a(t-s)Au(s) ds
f ( t),
0 < t _:_ T,
where O < T _:_ oo, a(t) is a real valued function and A is a nonlinear operator from one space to another.
In case where A is
a maximal monotone operator on a real Hilbert space H,
the
existence of solutions of (1 .1) has been studied by several authors(cf.
(1 ],[6]). The setting in which A is a maximal monotone
operator from a real Banach space V c H to its dual V'
has also
been studied by Barbu[1 ], Crandall et al.[5], and Kiffe and Stecher[9].
In [7],
the author considered the case where A is a
pseudo-monotone operator from H into itself and gave existence theorems for the equation (1 .1 ).
From the point of view of
applications to the case where A is a differential operator, this assumption on A is restrictive.
In this paper, we extend the
result in [7] to the case where A is a pseudo-monotone operator from a real Hilbert space V CH into its dual V'. Assuming that a(t) is a kernel of positive type, we show the existence of solutions of (1.1)
2
Statement of
for f
EL
2
(0,T;V).
the main result.
Throughout this paper, V
will denote a real Hilbert space, densely and continuously imbedded in a real Hilbert space H. Identifying H with its own 83
DOI: 10.1201/9781003420040-8
Hirano
84
dual H',
v CH cv'.
we have
Let (x,y) be the pairing between
an element x EV' and y EV. If x, y EH, then (x,y) is the ordinary inner product in H.
II • II , I • I
By
II • II*,
and
we denote '\,
'\,
'.\,
the norms of V, Hand V', respectively. We denote by V, H, v' • spaces L 2 (O,T,V), L 2 (O,T;H) and
L 2 (O,T;V'), respectively. We
denote by the pairing between~• and u, v EH,
~- Then for each
is the inner product in~-
and~• are again
the
l•I
denoted by 11•11,
'\,
The norms of~.~
and 11•11*' respectivel-y.
-v
-If"
the cl-osed convex
For a subset D of V, we denote by co D and co D
hull of D with respect to the topology of~ and~, respectively. A nonlinear mapping A from V into V' is said to be monotone if for all yi E
Axi, i = 1,2.
A is said
to be maximal monotone if it has no proper monotone extension, A nonlinear (single-valued) mapping from V into V' is said to be pseudo-monotone (*)
if
A satisfies the following conition:
If a sequence {un}C V satisfies that un
in V
and
(Au, u - v)
lim sup (Aun' un - u)
~
n+co
~
lim inf (Aun' un - v) n+co
u(weak convergence)
+
O, it follows that for all v EV,
and Aun
converges weakly to Au in V'. In the following,
we will assume that A is a pseudo-monotone
operator from V into V' with domain D(A) = V.
In addition,
we
will assume that A satisfies the conditions
II ull>
( 2. 1 )
II Aull*
( 2. 2)
c 2 11ull2 ~ c3 + (Au, u)
~
c, ( 1 +
for u EV, for
U
E
v,
where c, , c2 and c 3 are positive constants. 1 For each a(t) EL (O,T),
L
a
denotes the linear continuous
Nonlinear Volterra Equations with Positive Kernels
85
operator defined by
f
(Laf)(t) = for each f £ L 2 (O,T). =
I
T
t
for O
:a(t-s)f(s) ds
t
~
~
T,
Then the adjoint operator La* is given by for O
a(s-t)f(s) ds
t
~
~
T.
We state the assumptions for the kernel function a(t). (i)
a(t) £ L 2 (0,T) and is of positive type on [O,T], i.e., for
each f £ L2 (O,T),
f:
(ii)
La*
f(t)
f:
a(t-s)f(s) ds dt
* is injective, i.e., Laf
Remark.
~
0,
=O
for O
means f
~
t
~
T.
= O.
L and L* are bounded linear on a a are positive on these spaces if a(t)
the operator L*
and ~•. Also L and a a satisfies the condition (i). Sufficient conditions for a(t) to satisfy (i) are investigated in (11] and (12]. (ii)
if a ( 0)
~
0 and a' ( t)
£
1
a(t) satisfies
L ( O, T) •
We now state our main result: Theorem.
Let A:V
V' be a pseudo~monotone operator satisfying (2.1) and (2,2), Let a(t) £ t 2 (0,T) be a function satisfying (i) and (ii). Then for each f £ L 2 (o,T;V), (1 .1) has a solution in L 2 (0,T;V). ~
Hirano
86
Remark.
Since a monotone hemicontinuous operator is
pseudo-monotone, our result is an extension of the results for monotone operators (cf.
[ 1, 9 J).
It is well known that the sum of
monotone hemicontinuous operator and completely continuous operator is pseudo-monotone. Then Corollary. Let A:V
➔
we have
Suppose that Vis compactly imbedded in H.
V' be a monotone single-valued hemicontinuous operator
satisfying (2.1) and B:H
+
H be a continuous operator satisfying
(2.1 ). Suppose that A+ B satisfies (2.2) with A replaced by A+ B. Let a(t) E L 2 (0,T) be a function satisfying (i) and (ii). Then for each f EL 2 (O,T;V), the equation
f t0a(t-s)(Au(s)
u(t) +
( 2. 3)
+ Bu(s)) ds
f ( t),
0 < t
< T,
has a solution in L 2 (0,T;V).
3.
Proof of Theorem.
We first state a well known result
which is crucial for our argument. Proposition A(cf. [3), [4)). of~-
Let T :~
+
~•
Let K be a closed convex subset
be a monotone operator and T 0 :K ➔
pseudo-monotone operator.
~•
such that (g + Tou, v - u) > 0
for all g E Tv,
v EK.
The following lemma is also known and easy to verify. Lemma A. sup Then
u E 11'_
Let u be an element of~•. (U
1
V)
be
Then there exists an element u of K
VE
'\;
V,
!vi
~
Suppose that
0 A-Ix EB,
Hence, every closed convex cone with a base is necessarily pointed. Proposition2.l
(V. Klee[31])
Let K be a closed convex pointed cone in equivalent:
E. The following are
The Nonlinear Complementarity Problem
129
(1)
K
is locally (weakly) compact;
(2)
K
admits a (weakly) compact base.
When
K
is not pointed, an useful property initially remarked by
V•. Klee [31] ( see al so B. Anger and J. Lembcke [ I J) and pointed out to the authors by J.M. Borwein [8]) will be a key tool for what follows. Proposition 2.2.
Let
K be a convex closed cone in
(1)
K
is locally (weakly) compact;
(2)
K
is the direct sum of a locally (weakly) compact convex S with a finite dimensional subspace L ,.
pointed subcone (3)
there exists a continuous sublinear functional
B := {x EK : g(x)=J}
exists
c
Proof
E. The following are equivalent
>
0
is (weakly) compact, generates
such that
(1)=(2)
g (x) ~ c
11
x
for each
As easily seen, the set
which is finite dimensional since if we denote by
11
K
Since for each we deduce that S+L = K
K. Furthermore, there
is a subspace
is locally (weakly) compact. Hence, L
L
and by
P (resp. Q)
(resp. onto M), the set
is a locally (weakly) compact convex subcone of S n ( -S) = (Kn ( -K)) n M
such that
x E K.
L :=Kn (-K)
M a topological co~plement of
the continuous linear projection onto
g
K
S :=KnM
which is pointed since
LnM={O}. x EK,
Q(x) = x-P(x) E K-L cK+K cK
Q(K) c Kn M = S, and therefore
and
Q(x) EM,
Kc P(K) +Sc L+S c K. Hence,
as desired. (2)=> (3) Since
S+L = K
and
S
is a locally (weakly) compact
convex cone, the standard Klee's result, provides a compact convex base B for s. If e * EE * separates B from the origin, then X* :=e * o Q EK * ·= {x * EE * :x*(x) ~o for all x E K} and x * (x) > 0 for each XEK\{O}. For any equivalent norm in g(x) :=
L, the functional
II P(x) II
g
given by
+x* (x)
is sublinear and continuous. As easily seen, B := {x EK: g(x)=l} K
and is (weakly) compact. Hence, B
such that
g(x) ~cllxll
for all
xEK.
is bounded and there exists
generates c
>
0
Isac and Thera
130
(3)= (I)
Let
a< I
and set
U := {x EE: g(x) l' a}. Then
convex closed neighbourhood of the origin such that (weakly) compact. Hence
Un K is (weakly) compact and therefore
locally (weakly) compact, as desired,
that generates the topology on iEI
which is K is
D
In the locally convex setting, He say that uniformly positive if for a given g(x) >,cpi(x), for all
U is a
Un Kc [0,a] B
g : E ->JR
is K-
saturated family of semi-norms
E, there exists and each
c >0
{p.}. I l.
such that
l.E
xEK, xfO,
We observe that if we only require the existence of a sublinear continuous functional, which is K-uniformly positive, we only get a convex open neighbourhood
U of the origin such that
U n K is bounded. Then we
K is locally bounded [22]. In fact we have :
say that
Proposition 2.3
Let
K
be a cone in
z,Jhich is supposed to be locally convex.
E
The following are equivalent (1)
There exists a sublinear functional
g
which is K-uniformly
positive and continuous; (2)
K is locally bounded.
Proof : We only have to prove (2)=>(1). Choose a neighbourhood origin, which is closed, convex and circled. given by
pU(x) := inf{t > 0 : t
Furthermore, if with
therefore i.e.,
pU(x) =0 -I
x = ti (ti x)
x EU}
for some
and
ti> 0, lim ti= 0
-I
I
Then, the gauge
U of the
Pu
of
U
is sublinear and continuous, x EK\ {0}
t: x EU. Since
we can select a net -I
(x.). I l.
l.E
K is a cone, ti x EU n K and
belongs to the asymptotic cone
T (UnK) 00
of
U n K,
xe:{de:E: '3(ti)id->0, ti>0, -3(di)iEI' die:E, d=lim tidi}. Since
u n K is bounded, T ,,(UnK) 0
contradiction. Hence, PU
reduces to
{0}, so that
x =0
and we get a
is a convenient sublinear mapping.
REMARK: We observe that in Proposition 2,2, whenever convex locally (weakly) compact cone, L
reduces to
Klee's result and its well-known consequence: E*
D
K is a pointed {0}
and we recapture
admits a K-uniformly
continuous linear functional which is necessarily strictly positive.
131
The Nonlinear Complementarity Problem 3
A VARIATIONAL PRINCIPLE.
Theorem 3. I
Let
K
E
c
that satisfy: is positively homogeneous of order
nals defined on each x
E
f
K, x
0
be functio-
and
p
for
T 1 (x) > 0
T 2 (x)
---:,:0. II XII p
lim sup llx 11 ➔ + 00 XEK
(?,)
T1, T2
K
TI
(1)
be a cone in a named space and let
is closed, locally (weakly) compact and convex, and T 1 is (weakly) lower semi-continuous, then {x EE: T 1 (x)-T 2 (x) :s >..} is (weakly) If
K
>.. ER.
relatively compact for all Since
Proof :
K
is a closed locally (weakly) compact convex cone, by and a sublinear continuous mapping
c >0
Proposition 2.2(3) we get
g
such that
~
g(x)
cl/pll xii
x EK.
for all
Note LA:= {xEK :T 1 (x)-T 2 (x) :,:>..}, the >..-level set of T 1-T 2 • Since g is weakly continuous, it suffices to show that sup g(x) is finite. X XELA n If not, pick xn ELA such that g(xn) ~ n. Then y :=-(--) belongs to g xn n -1 g ({!}) nK which is (weakly) compact. Thus, there exists a subsequence with
(xnk), which tends (weakly) t o y
Ilk
)
----::: - - - - : s - - - [g(x
Ilk
)JP
[g(x
so that, T 1 (y)
Thus, T 1 (y):,:O,
yEK
A+T 2 (x~)
A+T 2 (x~)
T 1 (x~)
Tl(y
g(y)=l. Hence,
and
:sf
Ilk
cllx
)JP
~
llp
T2 (x~)
lim inf k -->-+oo
----:so II x
Ilk
11P
;;(y)=I, which is impossible.
□
Corollary 3. 2
Let
T
be a p-positively homogeneous functional defined on a cone E, which satisfies
K given in a named space
T(x)>O
forxEK,x/0.
Isac and Thera
132
If K is closed,locally (weakly) compact and convex and Tis (weakly) lower semi-continuous,then each level set of Tis (weakly) compact. REMARK:
A similar result was proved by Janos [28 :Thm. 1.2] for a positive-
ly continuous functional. The next result, extends the Weierstrass variational principle. Theorem 3.3
Under the assumptions of Proposition 3,1, if T1-T 2 is (weakly) lower semi-continuous then T1-T 2 achieves its minimum on K and the set of minimal points of T1-T2 is (weakly) compact. Proof :
Every level set of
T 1-T 2
is (weakly) relatively compact, hence
(weakly) compact, since (weakly) closed by (weak) lower semi-continuity of T 1-T 2 . Therefore,
m :=inf {T 1 (x)-T 2 (x)} X€K
tion property applied to the family
is finite and the finite intersec-
(FE)E>O
given by
X€
provides an element
FE :={xEK :T 1 (x)-T 2 (x) :;:m+d satisfies :
min {T 1 (x)-T 2 (x)} XEK
n
that
F
E>O E
□
Corollary 3.4
Under the assumptions of Corollary 3.2, every p-positively homogeneous functional T that satisfies T(x)
achieves its minimum on (weakly) compact, Proof
Set
T2 =0
>
0
for x
€
K, x -IO ,
K and moreover the set of minimal points of
in the preceding Theorem.
T is
□
See also J.P. Penot [43] for a more general criterion using the concept of asymptotically compact set, REMARK :
We notice that assumption (2) of Theorem 2.1 applies with
for a quasi-bounded mapping i.e, a mapping such that
p
>
1,
The Nonlinear Complementarity Problem
133
l..U20J
q(f) := lim sup
llxll->-+oo llxll
< +oo •
XEK
As observed in [ 21], this occurs whenever D00 f EE *
that is, whenever we may find
I f(x)-D f(x) I 00
lim
II
is asymptotically linear,
f
such that
ll->-+
X
XEK
II XII
00
= 0
The reader interested by these concepts should consult for instance
t 21],
[ 33].
4
APPLICATION TO THE NONLINEAR COMPLEMENTARITY PROBLEM. Let
T map a closed convex cone
Kc E
nonlinear complementarity problem relative to
into
E*
T and
and consider the
K
N,C.P. (T,K) : Find XE K such that T(x) EK* and = Since
K is a cone, as easily observed, N.C.P, (T,K)
to the variational inequality V.I(T,K) V.I(T,K): Find XEK, such that
o.
is equivalent
defined by ~O for each yEK.
which can also be rewritten as, V,I(T,K) : Find x EK where
31/J
K
-T(x)E 31/JK(x),
stands for the convex subdifferential of the indicator
(x)
function of
such that
K defined by, 1/J
K
(x)
=0
if
X E
K
and
1/J
K
(x) = +
00 ,
else.
The complementarity problem was studied in infinite dimensional spacesin[2]
[4]
[7]
[24]
[25]
[26]
[29]
andvariationaline-
qualities for potential operators in [44]. Let
K be a closed convex set of a Banach space
that a continuous operator P(E) =K
and
P(x) =x
P
for each
defined on
E
E. We recall
is a projection onto
K if
x EK.
We obtain the existence of continuous projections using different approaches. When
K is a closed convex cone, the following holds true:
Isac and Thera
134
Theorem 4 .1
Let a > O,
[34]
K be a closed convex cone in a Banach space
there exists a projection
P
0:
onto
K
such that for all
llx-Po:(x)II:: (J+o:)d(x,K)
where as usual
xEE
d(x,K) := inf llx-yjj yEK
We recall that a norm
11
satisfies the property t>,O
E. For each
or else
defined on
11
is strictly convex if it
E
implies
llx+yll = llxll + llyll
x =t y
for some
y=O.
The corresponding normed space is then called strictly convex. The class of strictly convex Banach spaces introduced by Clarkson [13] includes all Hilbert spaces, as well all
Lp((t,1,µ)
for
1
, 0}
has a bounded convex base then
E+
has
a Schauder basis. Proof : Let E+
B be a bounded base for
E+. Since
E
is a Banach lattice,
is closed and as observed before, B can always be chosen convex closed
with the uniqueness decomposition property. Since
O ,/. B, pick
a> 0, b
>
Let
x
0
such that for each
a::= 11 x 11 : = b
( 1)
be an extreme point of
OO, "2 >O, bl, b2 ~
y
x
is a convex combination of
x E ext(B), x =b 1 =b 2
x,yEext(B), with
b1
and the claim is proved, In par-
x11yfO, the relations
O a Galer>kin appr>oximation for>
T : K->E *
(a) if conver>gent to
y,
be a closed convex cone.
Kc E
(K) n nE ]N
of K. Suppose that
the following assumption ar>e satisfied (xn)
then
is weakly conver>gent to
x
and if
(yn)
is stY'Ongly
138
Isac and Thera
lim sup ~ n-++oo
(b) Then,
1,,J
(xn)
solves
'.t>
X
is sequentially weakly lower semi-continuous,
xl->
solves
N,C.P, (T,Kn), each weak sequential Zimit point of N.C.P,(T,K). Moreover, if K is locally compact or E is
n
reflexive, it suffices that Proof :
x=
be bounded.
(xn)
Let
x
be a weak sequential limit point of
(w) - lim
x
• Since
k➔+oo
Ilk
solves
Letting
X
Ilk convexity of
:= p
Ilk
N,C.P. (T ,Kn),
(x)
and
lim inf = 0
I
x +(I --)x and using the ~Ilk Ilk~ we have z EK • Furthermore, since X solves Ilk Ilk ~ also solves V,I(T,K ) so that, z
Ilk
:= -
n
Hence, >, 0 k
and therefore, since
k
assumption (a) yields Let
~
N.C.P.(T,K
(xn), that is,
y EK
>,O. Hence
(;;-n) k
norm-converges to
x,
=O.
be arbitrary and set
solves
N.C.P.(T,K obtain,
) , we have >,O, and using again assumption (a), we Ilk Ilk Ilk >, 0 for every y E K. Hence T(x) E K* and x so 1ves
N,C.P(T,K), as
desired. □
REMARK
Assumption (a) is fulfilled if
(I)
continuous. In this case, if norm-converges to
(xn)
and
(T(x ))
REMARK
n
(2) :
n
x
and
(yn)
- n n n
is strongly bounded, we have If
is sequentially weakly
y, since
- n
T
weakly converges to
T
lim
n-++oo
n
n
=
is completely continuous, i.e, T maps weakly
convergent sequences to
x
into strongly convergent sequences to
T(x),
as do most of the operators used in the deformation equations of Elasticity theory, then
xt->
tion 4.4 is trivial.
is sequentially weakly continuous and Proposi-
The Nonlinear Complementarity Problem REMARK
i.e. for each sequence zero i_n
(resp. in
E
has the Dunford-Pettis property (D.P.P in short),
E
Suppose
(3)
139
{x } CE (resp. {x*} c E*) n n * E ) then lim X * (X ) = Q • n oo n n
that tends weakly to
It is well-known that for each compact set D.P.P, as well each If E
E
T, C(T)
is Dunford-Pettis then for each operator E*
equipped with the weak topology into
xl-->
admits the
for each probability measure.
1 1 (µ)
T
continuous from
equipped with the weak topology
is weakly continuous and Proposition 4.4 is again trivial.
Lemma 4.5. -
Let
T be the Gateaux derivative of a convex functional
following implication holds true:
'
lim sup 11 X 11 ➔ +oo Proof
11
::=
X 11 p
0
=
Because of the convexity of
¢(x)-¢(0)
. llm sup 11 X 11 ➔ +oo
¢ we have
¢'(x)(x)
::=
1. The
from which we derive the desired result.
O
Theorem 4.6. -
Let Let
K be a closed locally compact convex cone in a Banach space.
T :=T 1-T 2
where
Ti
(i=l,2)
maps K into
and suppose the
E*
following assumptions hold true (1)
If
y,
converges to (2)
then
n
n
oo
n
is sequentially weakly upper semicontinuous; is the Gateaux derivative of
T.
1.
¢ 1 (x) >O
for each
is convex and Then, N.C.P(T,K) Let
1.
{Kn}
is lower
p,lower semicontinuous
XEK\{O};
lim sup
1
::=
X I I Hoo 11 X 11 p+ XE:K 11
Proof :
and
¢.
is positively homogeneous of order
¢l
and satisfies
norm-
is sequentially weakly lower semicontinuous while
semicontinuous; ( 4)
and if
x
lim sup :;;
x t->
x l-> ( 3)
is weakly convergent to
{xn}
O.
is solvable. be a Galerkin approximation of
conditions (3), (4) and (5), Theorem 3.3 applied to
K. Using Lemma 4.5, provides
X
n
EK
n
Isac and Thera
140
such that
on X is both upper
is a continuous sublinear DOI: 10.1201/9781003420040-12
Komiya
148
functional on E and A is a continuous multi-valued mapping of X into E. Theorem 1 Let N be a normed vector space, X a compact convex subset of N1 and pa continuous sublinear functional on N. Let F be a continuous multi-valued mapping of X into N with nonempty values Fx such that for any u, v £ Fx 1 there exists w £ Fx with p(w - (u + v)/2) s O. Then there exist x £ X and f £ N' such that max f(X) f s ¢, f(x) inf p(Fx).
inf f(Fx)
Proof The dual space N' of N will have the weak* topology. If. Y = (f £ N': f s ¢}, then Y is compact, Define a multi-valued mapping A of X into Y by Ax
{f £ Y: inf f(Fx) = inf (Fx)}.
Ax is nonempty by [4, Basic Theorem) and it is easy to see Ax is convex. Let {(xa,fa)} be a net in the graph Gr(A) of A converging to (x,f) £ X x Y. Take a point x I of Fx. Since the functional y ~ inf ¢ (Fy - x 1 ) is upper semicontinuous, for any£> O, inf (Fxa - x') < £ eventually. Hence there exist x~ £ Fxa with (x~ - x 1 ) < £ eventually. Then we have f
a
(X 1 a
)
1 .: f a (x') a - ¢(x a
-
x')
Hence we have f(x') ~ inf (Fx), since the functional y i-+ inf (Fy) is lower semicontinuous, and hence inf f(Fx) ~ inf (Fx). The reverse inequality is easy, hence we have (x,f) £ Gr(A). We have proved that Gr(A) is closed. Since Y is compact, A is upper semicontinuous and has closed values. Define a multi-valued mapping B of Y into X by Bf
{x £ X: f(x)
max f(X)}.
It is easily seen that Bf is nonempty and convex. If {(f ,x )} is a net a a
On a Best Approximation Theorem
in Gr(B) converging to (f,x) for any x 1 £ X,
£
149
Y x X, then
f(x)
f(x 1 ) = lim a f a (x') :a lim a f a (x) = f(x). a Therefore we have (f,x) £ Gr(B). We have proved that Gr(B) is closed. Since Xis compact, Bis upper semicontinuous and has closed values. By the coincidence theorem, there exist x £ X and f £ Y such that f £ Ax and x £ Bf. These x and fare the desired ones. Remark (1) The conditions for the multi-valued mapping F of Theorem 1 are satisfied if Fis a continuous multi-valued mapping of X into N with nonempty convex values. (2) We obtain [6,Theorem 5,1] by taking Fx = Px - Qx. We see from Theorem 1 that the assumption that P and Q have closed values in [6, Theorem 5.1] is unnecessary. (3) Theorem 1 holds even if the normed vector space N is replaced by a Macky space, but it is not known whether Theorem 1 holds or not when N is replaced by a general locally convex topological vector space.
3, A normed vector space N with a vector order.: is called an ordered normed vector space. The functional¢ on N defined by cj>(x) = inf {JJyJJ: y.: x} is called the canonical half-norm for the given vector order (cf. [1]). The functional¢ is continuous and sublinear. A point f of N1 is said to be poisitve if f(x) .: 0 for all x.: O, and we write f.: 0 if f is positive. When the positive cone is closed, x ~ 0 if and only if cj>(x) = O, and f s cjl if and only if f.: 0 and jfj s 1. Now we have the following thorem. Theorem 2 Let N be an ordered normed vector space with closed positive cone and X be a compact convex subset of N. Let F be a continuous multi-valued mapping of X into N with nonempty values Fx such that for any u, v £ Fx, there exists w £ Fx with w s (u + v)/2. Then there exist x £ X and f £ N1 such that f.: O,
JJfJJ :a 1,
and inf f(Fx)
f(x) = max f(X) inf cj>(Fx).
It is known that if N is a normed vector lattice (cf. [5]), then
150
Komiya
the canonical half-norm is the norm of the positive part, that 1s, l(x) = ~x+~. Hence we have the following corollary of Theorem 2. Corollary 3 Let L be a normed vector lattice and convex subset of L. Let F be a continuous multi-valued L with nonempty values Fx such that for any u, v € Fx, w E Fx with w ~ (u + v)/2. Then there exist x EX and f 2: 0, II fll s 1 , f ( x) = max f ( X) , inf f(Fx)
inf {lly+II: y
€
X be a compact mapping of X into there exists f EN' such that
Fx}.
REFERENCES [1] Arendt W., Chernoff P.R. and Kato T., 11 A Generalization of Dissipa:tivity and Positive semigroup 11 , J. Optim. Theory, vol. 8, pp. 167180, 1983. [2] Fan K., 11 Fixed-point and Minimax Theorems in Locally Convex Topological Linear Spaces", Proc. Nat. Acad. Sci. U.S.A., vol. 38, pp. 121.,.125, 1952. [3) Kakutani S., 11 A Generalization of Brouwer's Fixed Point Theorem", Duke Math. J., vol. 8, pp. 457-459, 1941. [4] K~nig H., 11 0n Certain Applications of the Hahn-Banach and Minimax Theorems", Arch. Math., vol. 21, pp. 583-591, 1970. [5] Schaefer H.H., "Banach Lattices and Positive Operators'', Berlin/ Heidelberg/New York: Springer, 1974. [6] Simons S., 11 An Existence Theorem for Quasiconcave Functions with Applications", an expanded version of the talk given at AMS meeting in Eugene, Oregon in August 1984.
Vector-Minimization Problem in a Stochastic Continuous-Time n-Person Game HANG-CHIN LAI Institute of Mathematics, National Tsing Hua University, Hsinchu, Taiwan, Republic of China KENSUKE TANAKA Japan
1.
Department of Mathematics, Niigata University, Niigata,
Introduction The authors investigated a cooperative n-person game with
a discount factor in the case of discrete time countable state space.
( See Lai
and Tanaka
space [10]
).
and We
study the game system with a side-payment in case of the tinuous time space.
a now con-
In this case we would apply the Kolmogorov
forward differential equation to settle the transitionprobabilities which plays an important role in our game system.
From
the transition probability, we can define a one parameter contraction semigroup of operators on a space of all bounded real valued functions on the countable state space so that it has a fixed point.
Then it will reduce to an optimal
multistrategy
on which the collective total expected multiloss will 151
make
a
DOI: 10.1201/9781003420040-13
152
Lai and Tanaka
collective stability in the dynamic game system for a side-payment.
This vector valued optimization problem in game theory
is pretty different from usual multiobjectivemathematical problems which one could refer to Yu [16], Tanino and Sav,aragi [15], Lai and Ho [11] and their references.
The reasons
come
from
that the total expected multi loss is depending on each component which is steered by each player upon his chosen in the strategy space.
The cooperative discounted Markov
game
considered
in this paper has many connection with the noncooperative cases which are developed by the authors in [7],
[8],
[9]
We organize this paper into five sections.
and
[1~.
Section 2
is
the formulation of our game system which will be discussed
in
this paper.
In section 3, we give some necessary
assumptions
and the preliminaries on this game system in. which we introduce the Kolomogorov forward differential equation which wU 1 derive a one paprmeter semigroup of operators on the space of bounded real valued functions of the state space
S.
Section
4
will
give a cone dominated structure in which we would describe optimal multi-strategy existed in the game so that
an
the
total
expected multiloss makes a collective stability under a
side-
payment, and finally we summarized the main results of our present game system that under some conditions, there is a payment
side-
forced a vultistratevy to be optimum, and the related
properties, for example, a convex
multiplier, lower
support
function and super-differential are discussed in the context.
A Minimization Problem in a Stochastic n-Person Game
2,
153
Formulation of an E_-person discounted dynamic game Consider a game system as the following Cl,
2n+3
objects:
) r1, r 2 ,···,rn ,ex.
(2.1)
Here (1)
S = {1, 2,
•··,}
is a countable set, namely,
the
state
space; (2)
Ai
is the action space of player
Ai
is a compact metric space, i
We assume that each
i. E
:t-T
= {1,
2,
• · •, n}
is
the player set; (3)
q = q(• Is, a)
is a bounded function, namely a transition
rate function, for any
-a
where (4)
i r ( s, a)
= (a
1
'
a
(s, a) 2
'
... '
a
an)
X
E
A, A
n -
IT
i=1
s
is a real valued function on
loss rate function of player (5)
Es
X
Ai. A,
namely
a
i EN.
is a positive number, as a discount factor.
In order to know
easily our game system,
we
give some
interpretation as follows. For any time
t
E
[O,oo),
all players observe the state of
the game process and classify it to one possible state which is determined by a stochastic process Then all players choose their actions
{X(t)l
at FA
st ES
at time t.
under some pro-
such that each player t After this the game process moves to a new state st, according bability
JJs
to the transition function Markov process.
q(st' Jst, at)
associated
with
a
154
Lai and Tanaka
A strategy [O, oo)
1T
i
of the player
i
N
£
chosen at time
is described as a function which maps the information
game process from past history up to present state space carried by
i th player.
dent of the time
t
Especially if
n
i =
of
t
£
the
jnto
the
ni
is indepen-
( t)
action
and depends only on the present state, the strategy
is said to be stationary. In this case, for any t £ (0' 00)' there is mapping µ i = µ i : s ... P(Ai), that is, µ i £ [P(Ai)]s, for simplicity,we t write P(Ai) instead of [P(Ai)]S as the stationary strategy space of the player
i
N.
£
Here
P(Ai)
on the Borel measurable space field of the metric space
Ai
is the set of all probability measures (Ai, B(Ai)) for each
i
B(Ai)
where £
N.
Since
is the Borel Ai
is assumed
C(A i ), the space of all real-valued continuous func-
to be compact so
tions on A\ is a separable Banach space in the supremum norm topology. It is know that space of M(Ai).
P(Ai)
C(Ai).
is a weak* closed unit sphere in
M(Ai), the dual
Hence P(Ai) is a metrizable compact convex subset of
(Cf. Lai and Tanaka [8]).
Throughout this
paper,
we
assume
that each player uses only the stationary strategy. In this game systems, as the multistrategy players, the i th tial state
i
1, 2,
s
£
ii
is
chosen by all
player's total expected discounted loss under an iniS
n.
is defined by
We denote a vector (2.3)
and ask whether there exists an optimal mul tistrategy 1/J(ii)(s)
µ*
such
that
makes a collective stability rather than the individual opti-
mal strategies of players under some convex dominated structure. 1 2 n n this paper, we will use a side-payment d = (d, d, ·· · ,d) £ 1R
In with
A Minimization Problem in a Stochastic n-Person Game
li~lldi\ = 1
155
to make a collective loss function
w(µ)(s)> =
+ Q(µ*, o )u(µ*)(s)}
min {-.-Function for Several Classes of Normed Linear Spaces ROBERT H. LOHMAN Department of Mathematics and Sciences, Kent State University, Kent, Ohio THADDEUS J. SHURA Department of Mathematics and Sciences, Kent State University at Salem, Salem, Ohio
Let X be a normed space and let x E BX.
A triple (e, y, :\.) is said to be
amenable to x in case e E ext(¾:), y E BX , 0 < :\.
~
1 and x = :\.e + (1-:\.)y,
In this
case, :>i.(x) is defined by :>i.(x) = sup(:\.: (e, y, :>i.) is amenable to x}. Xis said to have the A-property if each x E ¾: admits an amenable triple, ally, if X has the :\.-property, then the :\.-function:\.: ¾:
~
Inform-
[O, 17 gives a measure
of the "largest" positive weight assigned to an extreme point e in the different representations x = :\.e + (1 - :>i.)y as x ranges over the closed unit ball BX of X, If X has the :\.-property, define :>i.(X)
inf {:\.(x): x E ¾:l,
Xis said to have the
uniform :\.-property in case :\.(X) > 0, These ideas were introduced and studied in [2],
In particular, spaces with
the :\.-property or uniform :\.-property were shown to have geometric features which may be of potential use in applications,
For example, if X has the :\.-property,
then every real-valued convex function on¾: which attains its maximum value must attain this value at a member of ext(¾:),
Consequently, if Xis a Banach
space with the :\.-property, then BX is the closed convex hull of its set of extreme points (Theorem 3.3 of [2]).
A much stronger conclusion holds if X has the uniform
:\.-property (Theorem 3.1 of [2]).
Namely, if O < :\. < :\.(X), then for each x E BX
there is a sequence (ek) in ext(BX) such that for every n, we have n llx - 6 :\.(1 - :>i.)k-le II ~ (1 - :>i.)n. k=l k Evidence is given in [2] that neither the :\.-property nor uniform :\.-property is It is shown that many spaces of the type t 1(X), t 00 (X), c(X), CX(T) possess
rare,
one of these properties and formulas for the :\.-functions of these spaces are obtained,
Also, every finite-dimensional normed space X possesses the uniform :\.-property and, for each x E Bx, we have :>i.(x) ~ (1 + n)- 1 , where n is the dimension of X considered as a real vector space, Since normed spaces with the :\.-property or uniform :\.-property satisfy strong geometric conditions, it is of interest to expand the catalog of spaces with one 167
DOI: 10.1201/9781003420040-14
Lohman and Shura
168
of these properties,
That is the purpose of this note, We first consider t -direct co sums, noting that parts of the following result were obtained independently by
David Trautman, Let (X ) be a sequence of normed spaces and let X = ($ !) X ) , • n n~ co Assume that each space Xn has the >..-property and let >..n denote its >..-function, (a) X has the >..-property if and only if there is a finite subset P0 of the set P of positive integers such that inf [>..n(Xn): n E I"P 0 } > 0, In this case, Theorem 1,
x = (xn ) E BX' then >..(x) = inf>.. n n(xn ) • (b) X has the uniform >..-property i f and only if inf >.. (X ) > 0, n n n 11 11 Proof. ( a) ~ Write B = R_ • Let x = (x ) E BX and note that we have n -xn n a= inf>.. (x) > 0, If O .. (x ) - s < >.. ~ 1, such that x = >.. e + (1 - >.. )y • n n n nn n n nn nn Since O < a - s ~ >.. (x) - s, Proposition 1,2 of [2] guarantees that we can write if
n
n
x = (a - s)e + (1 - ~ - s)z, where z EB. Thus, x = (a - e)e + (1 - a - e)z, n n n n n where e = (en) E ext(Bx), z = (zn) E Bx• Since a - £ > O, X has the >--property and >..(x) i!: a - s. Letting e tend to O yields >..(x) i!: a. On the other hand, assume x = >..e + (1 - >..)y, where e =(en) E ext(Bx), y = (yn) E BX and O < >.. ~ 1, For each n, we have e E ext(B ), y EB and x = >..e + (1 - >..)y. Therefore,>..~>.. (x) n n n n n n n nn for all n, This implies >.. ~ a and, taking the supremum over all such >.., shows that >..(x) ~ a, Assume that inf [>..n(Xn): n E P\P O } = 0 for each finite subset PO of P. Then there is an increasing sequence (n.) in P and vectors z EB such that -1 K Ilk nk >.. (z ) < k for all k, If we define (x) E BX by x = z , k E P, and x = 0 Ilk Ilk n Ilk Ilk n otherwise, then there is no triple amenable to x, Consequently, X fails to have the >..-property, 11
=>11
(b)
away from
11
« 11
Follows from (a), noting that a i!: inf>.. (X ) and hence is bounded n n n
o, independent of x.
Fix m E P and let x E B • Define x = (x ) E BX by x = x if n = m and m n n x = 0 otherwise, Given O < s < >..(x), we can write x = >..e + (1 - >..)y, where n = (e) n E ext(R_), -x y = (y) n E BX and O < >..(x)-s ;!;>..~ 1, Since em E ext(B m), ym EB m and x = >..e + (1 - >..)y, we have>.. (x) i!: Ai!: >..(x) - 8 i!: >..(X) - s, Since g is m m m arbitrary, we obtain>.. (x) l!: >..(X), Therefore, inf>.. (X) l!: >..(x) > 0, m n n n Remarks 2, If each space X is strictly convex, then>.. (x ) = (1 + llx II )/2 n n n n and the formula for >..(x) given in part (a) of Theorem 1 coincides with that given 11
=>11
e
in [2] (see Remark 1,14), It is also clear that Theorem 1 holds for tco-direct sums of arbitrary families of normed spaces, Next, we consider t 1 -direct sums of normed spaces, The situation is more complicated than for tco-direct sums and we obtain upper and lower estimates for >..(x). A precise formula for >..(x) is not yet known,
Calculation of the A-Function for Normed Linear Spaces
169
Let (Xn) be a sequence of normed spaces and let X = (13:1 6 X ) • n t1
Theorem 3,
Assume that each space Xn has the :>..-property and let An denote its :>..-function. Then X has the :>..-property and if x = (xn) E
¾,
f
x
O, we have ~(x)
:§
:>..(x)
\(x),
:§
where -:>..(x) = (1 - !!xii + 2max (A n (x n /!Ix n!I )!Ix n II : x n
and
f
0) )/2
f(x) = min ( sup:>.. (x ), (1- !!xii+ 2max!!x !J)/2), n n n n Proof.
Since !Ix II ... 0, the sequence (A (x /llx II )!Ix 11: x ,/ 0) is either finite n n n n n n
or is a null sequence.
In either case, there is a positive integer N such that
:>..i~/ll~ll lll~ll = max p..n(xn/11xn11 lllxn11 : xn Let O < e < :>..N(~/ll~lllll~II.
f
0).
There is a triple (eN, ~' :>..N) amenable to ~/llxNIJ
such that :>..N(~/JJ~JJ) - e < AN may assume that 1 - :>..Nll~II
f
0.
:§
By making AN slightly smaller, if necessary, we
1.
Since ~/11~11 = :>..NeN + (1 - :>..N)~, we obtain
~ = :>..N!l~\JeN + (1 - :>..Nll~!!)vN' where vN = [(1 - :>..N)l!~l!/1 - :>..NJJ~!J]~.
Define
e = (en) E ext(BX) and y = (yn) EX by
e
n
=
l
o,
n '/- N
eN'
n = N
[2/(1 + llxll - 2:>..N!l~li)]~, y
n
=
n
2(1 - :>..N!l~[J)vN + (IJx/1 - 1 )~ 1 + JJx\\ - 2:>..Nl\~Jl
and let A = (1 - !Jx!J + 2:>..N\l~JJ)/2.
N
n = N
'
I t is clear that x = :>..e + (1 - :>..)y.
f
Moreover,
a routine computation shows IIYil
:§
2 (!1xll - 11~11) + 2(1 - :>..Nll~II )JlvNII + 1 - llxll 1 + \\xi[ - 2ANJ\xJ\
:§
2(llx!I - llxJI) + 2(1 - :>..N)l!~ll + 1 - 11:,ql 1 + llx\\ - 2:>..Nll~II
= 1.
Since 1 ~A= [l - !Ix/I+ 2:>..N(~/l!xJll!l~IIJ/2 + [AN - Ai~/llxJl)Jll~\I
~ ~(x) -
e!!xJI,
we obtain :>..(x) ~ ~(x) - e!lxJI.
Letting e tend to zero shows :>..(x)
.
-
:§
:>..(x), which
establishes the first inequality and the fact that X has the :>..-property, That :>..(x)
:§
(1 - llx!J + 2max\1xn11)/2 is established in the same manner as in the
proof of Theorem 1.1 of [2].
In addition, if e
=
(en) E ext(¾), y
and O < :>. ~ 1 are such that x = :>..e + (1 - :>..)y, we have e Since !!Y JI :§ 1 and x = :>..e + (1 - :>..)y , we have A m mm m follows that :>. ( x) :§ I"( x) •
:§
m
=
(yn) E BX
E ext(R_ ) for some m.
:>. (x ) mm
-xm
:§
sup>.. (x ) • nnn
It
170
Lohman and Shura
Theorem 3 easily extends to t 1 -direct sums of arbitrary families of normed spaces. Also, if each Xn is strictly convex and xn f o, then we have Remarks 4.
:>..n(xn/11xn11) = 1 and :>..n(xn) = (1 + llxnl\)/2.
~ ( x)
=
i (x)
Hence, in this case, - 11 xi I
= :>,. ( x) = ( 1
2maxl Ixnl I)/ 2 ,
>
which agrees with the formula given in [2] (see Remark 1.12). space X of Theorem 3 never has the uniform :>..-property. positive integer and choose~ E Xk such that
11~\I
= 1, 1
Observe that the
To see this, let m be a c'.e
k c'.e m.
Then the unit
vector x = ( xJ /m, ••• , x /m, O, O, ... ) of X satisfies :>..(x) c'.e i(x) c'.e 1/m and so .
:>..(X) = O.
m
Finally, if one of the spaces Xm fails to have the :>..-property, there is
a unit vector x EX
m
which fails to admit an amenable triple.
in the proof of Theorem 1, it is easy to check that triple in the space X of Theorem 3.
x fails
If~ is defined as
to admit an amenable
Therefore, X fails to have the :>..-property.
We next consider Lorentz sequence spaces of the type d(a, 1).
Before proceed-
ing to our results, we review some notation and known facts about this class of spaces.
For our purposes, let a
sequence with a 1 in c 0 such that
(an) E c 0\t 1 be a positive, strictly decreasing
= 1. The space d(a, 1) consists of all real sequences x = (xn)
IxTI ( n ) Ia n
sup 0
< "' ,
where the supremum is taken over all permutations
TI
of the set of positive integers.
If llxll is taken to be this supremum, then d(a, 1) is a Banach space.
There is con-
siderable literature on the Lorentz sequence spaces d(a, 1.) and, more generally, on the Lorentz sequence spaces d(a, p).
The interested reader should refer to refer-
ences [1], [3]-[6]. If x = (xn) E d(a, 1) and x M1 (x)
f
O, write
= llxt,
F1 (x)
= (n : lxnl = M1 (x)}.
If cF (x) denotes the characteristic function of F1 (x), define 1
M2 (x) = l\x - xcF 1 (x)I\"',
F2 (x)
= (n: \xn\ = M2 (x)}
and write M3(x) = llx - xcF 1 (x) - xcF 2 (x)t' etc. Then M1 (x)>O, ~(x)IO and if
~(x)>O, then ~(x) > ~+ 1 (x). Also, if M/xl > o, ~(x) > 0 and j f k, then Fj(x) and Fk(x) are disjoint. We let N(x) = (k ~(x) - ~+ 1 (x) > O} and, if k E N(x), we define
~(x) = card(
k
U
i=l
F.(x)), l
0, then we can write llxll =
0
kEN(x)
II xii
as
~(x)[sk(x) - sk_ 1 (x)].
Calculation of the \-Function for Normed Linear Spaces
171
The extreme points of Bd( a, l) are precisely the vectors of the form k
(I;an) n=l
e=
where jl < ... < ik, .:n (see
[3]
and
-J
k
(I;.:nei ), n=l n
:!: land (ei) is the standard unit vector basis of d(a, 1)
[7]).
Also, observe that if b = ( b ) is a sequence of ± l's, then the mapping n Tb:d(a, 1) - d(a, 1), defined by Tb((x )) =(bx), is a linear isometry of d(a, 1) n n n onto d(a, 1). In addition, any rearrangement of coordinates is a linear isometry
1)
of d(a,
1).
onto d(a,
Y is
Finally, note that if T:X -
a linear isometry of the
normed space X onto a normed space Y and x E BX' then x admits an .amenable triple if and only if T(x) admits an amenable triple. Theorem~.
Let x
=
f
(xn) E Bd(a, l)' x A( x) ~
In this case, \(x) = \(T(x)). Then
0,
sup [ Ic~ ( x) - I•~+ 1 ( x)] sk ( x) • kEJlf(x)
Consequently, the space d(a, 1) has the A-property, Proof. For each n, choose b = + l such that bx = Ix I. n n n n ing Tb(x), where b = (bn), we may assume xn ~ 0 for all n, If k
E N(x),
let e
E
ext(Bd(a, l)) be defined by
e
= (
sk ( x) )
and let A = [ M ( x) - M 1 ( x) ] s ( x) . K k+ _ k First, if A< l, define y
=
-l
k
6 e) i=l nEF. (x) n
( I;
l
There are two cases to consider,
(yn) E d(a, 1) by
- A)' y
=
n
[H ( x) k
-
~
~
k
r/. U
n - M,
. ( x) K+J
1 - \
l f l
Then, by consider-
J
F. (x)
i=l
l
n E Yi(x) for some i E (1, ... , k)
'
k - l, then (x) - [t~(x) - l\+ 1 (x)J ~ //1 0 and xn = 0 for n ~ N + 1. If x E ext(Bd(a, l)), then A(x) = 1 and, since the nonzero values of x are the constant M1 (x) and M2 (x)
= O,
n1 (x)
we obtain
1 =
llxll = M1 (x)(
6 an)
n=l
Thus, we may assume x ~ ext(BX). Suppose that x = Ae + (1 - A)y, where O NJ
L
(N+l, N+2, ••• )'-.(il' ... ,
\l-
In what follows, we will denote the nth coordinate of a sequence z by either zn or z(n).
Then, noting that the nth coordinate of the sequence x = Ae +
zero if n ~ N + 1 and e(n)
=
0 if n ~ (i 1 , ... , ik), we have 1 = llxll =
N
6 x a
n=l n n
(1 - A)y is
Calculation of the A-Function for Normed Linear Spaces 6
nEI +
[>.. c
E
ck(
+ (1 - :>..)y. ]a. + 6 (1 - :>..)y a n n l n l n nEJ
[:>..e(i ) + (1 - :>..)y. ]a. + E [:>..e(n) + (1 - :>..)y ]a n in in nEL n n
nEK
= :>. ck(
Since
E:
k n
.,
a. ) + :>. ck( I: E: a. nEI n in nEK n in
E E:
k
+ ( 1 - :>..)
E
y a n=l n n
.,
CD
a. n=l n in EE:
1 and I; y a ~ E IYnlan ~ n=l n n n=l
~
ck(
( 1) and
k
E
JlyjJ ~ 1, it follows that
1
E: a. n=l n in
6 y a = 1. n=l n n
(2) From ( 1), we obtain for 1
173
~
n
~
k.
k
E E: n ai
n=l
k
E
n=l
n
an' which forces
E:
CD
n
= 1 for 1
~
n
~
k and i
n
= n
a,
CIO
E y a < E I y I a , forcing E y a < 1. n=l n n n=l n n n=l n n
If yn < 0 for some n, then
Therefore, (2) implies yn ~ 0 for all n. This means that e = ck(
(3)
x
(4)
x
k
E
n=l
n n
en) and
=:>..c
~
N.
n
= ( l - :>. )y ,
(3)
l~n!k n ~ k + 1.
n
If N < k, then N + 1 ~ k and Therefore, we have k
+(1-:>..)y,
k
implies ~+l ~:>..ck>
o,
which is impossible.
By (3) and (4), y is of the form
y = ( xl - :>..ck, ••• , xk - :>..ck, 1 - :>. 1 - X
xk+1, ••• ,
r---:---x-
~ ,
r---:---x-
0, 0, • • • ) •
Then 1 = IIYII N
[n~/nan - ¾¾ -xk+lak+l + ¾+lak + xk¾+1
k 1 - :>..c (Ea ) + :>..c a - Ac a ](1 - A)k n=l n k k k k+l [l - A + (Ac
k
+
:x_
K+l
-
:x_
)(a. - a
K
K
K+l
)](1 - A)-l
Since¾ - ¾+l > O, it must be the case that Ack+ ¾+l - ¾ ;e
•
o.
Therefore, we
Lohman and Shura
174
have A;,;;
("k -
xk+l)ck-l"
But A> 0 implies xk > xk+l"
Hence, there is an integer
j such that M.(x) = xk, M. J(x) = xk 1 and n.(x) = k (i.e., s.(x) = c -l). J J+ + J J k yields A( x) ;,;;
rhis
max [ M. ( x) - M. 1 ( x) ] s . ( x) . jEN(x) J J+ J
An appeal to Theorem 5 proves the asserted equality for A( x). To prove that d(a, 1) fails to have the uniform A-property, let m = 2k be an even positive integer.
X
Define the unit vector x by
=
m m [ 6 (m - n + l)a ]-1[ L (-m_-_n_+_l)e ]. n=l m n n=l m n
From the first part of Theorem 6, we have m
m
\(x) = ![ 0 (-m_-_n_+_l)a ]- 1 ( 6 a ) m n=l m n n=l n m m m ( 6a )/[0a + 6(m- n)a] n= 1 n n= 1 n n= 1 n m
k
m
;,;; ( 0 a ) /[ 0 a + k( 6 a ) ] n=l n n=l n n=l n m m m ;,;; (0a )/[Ea+ (~)6an] n=l n n=l n n=l = 4/(4 +
m).
As a consequence, \(d(a, 1)) = O, completing the proof. REFERENCES 1.
Altshuler,
z.,
Casazza, P. G. and Lin, B. L., On symmetric basic sequences in
Lorentz sequence spaces, Israel 2.
J.
Math., 1'._5(1973), 140-155.
Aron, R. M. and Lohman, R. H., A geometric function determined by extreme points of the unit ball of a normed space, (preprint).
3.
Calder, J. R. and Hill, J.B., A collection of sequence spaces, Trans. Amer.
4.
Casazza, P. G. and Lin, B. L., On symmetric basic sequences in Lorentz sequence
5.
Casazza, P. G. and Lin, B. L., On Lorentz sequence spaces, Bull. Inst. Math.
6.
Casazza, P. G. and Lin, B. L., Some geometric properties of Lorentz sequence
7.
Davis, W. J., Positive bases in Banach spaces, Rev. Roum. Math. Pures Et Appl.,
Math. Soc., 152(1970), 107-118. spaces. II, Israel J. Math., 17(1974), 191-218. Acad. Sinica, 2(1974), 233-240. spaces, Rocky Mount. J. Math., 7(1977), 683-698. 16(1971), 487-492.
of Positive Eigenvectors and Fixed Points for A-Proper Type Maps in Cones W. V. PETRYSHYN Department of Mathematics, Rutgers University, New Brunswick, New Jersey
0.
Introduction
Let
X
be a real Banach space, K
projectionally complete scheme for and
D
X
0
a bounded neighborhood of
boundary and
DK
the closure of
a cone in with
ln
DK
r
X,
= {Xn,Pn} a for each n EN+
Pn(K)CK
with
X
relative to
The purpose of this paper is two-fold.
the
DK=
K.
First, using the approach
of [14, 23] and the index theory for A-proper vector fields developed by Fitzpatrick and Petryshyn in [8], in Section 1 we give a complete proof of Theorem 1, which is an improved version of Theorem 3 stated without proof in [23] and which asserts that if such that
T11
and some fixed and Ci):
I-T-µF y
A e (O,d/6]
>
0
T, F:DK + K
is A-proper w.r.t. with
such that
r
6 = inf{il Fxll: x e clDK} > dy!CT, where
T = 0, then clearly
be A-proper for each
assumption that
qI-F: DK+ X
µ
e
a = a(K) e [1/2 ,l] K
is the
and
IK(O,DK)
=
(0,1/y]
{l} 'le {O}, d
=
=
b
is equivalent to the
is A-proper for each
Py-compact in the sense of [17, 18].
x e cl DK
6 > by/a, and the requirement
sup{[lxll: x E 3DK} > O, (i) reduces to I-µF
E [0,1/y]
provided
d = sup{ II x-Txll : x e 3DK}.
that
µ
IK(T ,DK) 'le { D}, then there exist x-Tx = \Fx
"quasinormality" constant associated with Note that if
are bounded maps
for each
q
~
y, i.e., F
is
In this case, Theorem 1 yields a
new result, i.e., Corollary 1, which establishes the existence of an eigenvalue such that·
n > 6/b
Fx = nx.
and the corresponding eigenvector Since every k-ball-contractive map
Py-compact for any fixed exist
n
>
o/b
and
x E 3DK
of
F:DK + K
F is
y > k, it follows from Corollary 1 that there
x E 3DK
such that
6 = inf{//F,x/1: x e 3DK} > bk/a.
:x = nx
provided
The latter result was proved by Massabo
and Stuart in [14] for the case when
F:DK + K
is k-set-contractive and
Supported by NSF Grant MCSS0-3002 175
DOI: 10.1201/9781003420040-15
176
K
Petryshyn
is a normal cone with
being the constant of normality.
0
For the
extension of some eigenvalue results in [12, 9, 14] to multivalue k-setcontractive maps see [23]. Let us add that when
o
> 0
(i.e., when
= I-T-µF
T
)J
is A-proper for each
µ > 0
and
y = 0), then Theorem 1 and its Corollary 1 (when
T = 0) were first proved by Fitzpatrick and Petryshyn in [8] by a method which cannot be extended to the more general class of maps studied in In particular, Corollary 1 contains Theorem 5.5 of Krasno-
this paper.
selskii [12] when
F
is completely continuous.
Using Theorem 1 and the fixed point index properties in [8], it is shown in Theorem 2 in Section 2 that if D1 and D2 are bounded neigh-2 borhoods of O in X, T, F, C: DK+ K are bounded mappings such that I-T-µF
is A-proper for each
µ e [0,1/y], condition Ci) holds on
aD~,
then under very general boundary conditions (see (Cl) and (C2)) the map -2 -1 T has a fixed point x 0 e DK\DK. In case Tµ = I-T-µF is A-proper for each µ > 0 and o > O, Theorem 2 was proved by the author in [24] by a method which cannot be extended to maps treated in this paper. special cases of Theorem 2, when k-ball-contractive, or
T
and/or
F
Various
are completely continuous,
T
is P 1 -compact, will be considered. cular, it will be shown that, for suitable choices of C and
In partiF, Theorem
2 contains, on the one hand, fixed point theorems which are of ''coneexpansion and cone-compression" type in the sense of Krasnoselskii [12] for various classes of maps
T
and, on the other hand, it contains fixed
point theorems for T when and/or T00 at oo along K
has a Frechet derivative T T at 0 with T 0 and/or T00 being of type
in the sense of Amann [l]. or L~(X) authors will be given in Section 2.
1.
0
L~(X) The exact references to various
Basic eigenvalue results
We first introduce some definitions and state those results which will be needed in the sequel.
Let
X
be a real Banach space, { Xn} C X
a sequence of increasing finite dimensional subspaces, and for each in
let
N+
Pn(x)
+
x
in
P n
convergence in
be a linear projection of for each
X
X
and set
x
in
X.
X
We use when
onto
"+"
Xn
n
such that
to denote the strong
a = supjj Pn II n
•
The following class of A-proper (and, in particular, Py-compact) mappings introduced and studied by the author in [18, 19] (see [20] for the survey) proved to be quite useful since, on the one hand, this class includes completely continuous, Py -compact and ball-condensing vector
Positive Eigenvectors and Fixed Points for Maps in Cones
177
fields, strongly accretive maps and their perturbations by k-ballcontractive operators and others and, on the other hand, the notion of the A-proper mapping proved to be also useful in the constructive solvability of operator equations; furthermore, the theory of A-proper maps can be applied directly to the solvability of differential equations when the latter are formulated as operator equations with operators acting between two different spaces. (1)
Definition.
Let
D
be any subset of
X.
A-proper w.r.t. continuous for each
n
{x
a subsequence
if
=
and
T:D C:: X
X
+
is said to be
nk
}
q > 0
X
and i f
{x n. [ x n.
n.
) +
g
for some
X
D
such that
J
and
E
y > 0, the map
-
=
y
if
q
Dn. }
E
J
J
is A-proper for each
qI-F
y > 0
J
For a fixed
g.
N+
E
T n. (x
sequence such that
Tx
A map
iff
F:D
J
in
g X
nk
X
+
dominating
+
is
n
is any bounded
X, then there exist k
as
X
+
and
00
is said to be p y-compact y
(i.e., q > y
if
0) .
The notion of an A-proper mapping evolved from the· concept of a P-compact map introduced by the author in [17] to obtain constructive fixed point theorems and surjectivity theorems for monotone type maps. In terms of Definition 1 is A-proper for each
q
, we say that >
D, (i.e., F
use, we recall that, for a bounded set B.
J
is a ball in
with
X
diam(Q.) < d} J
-
with radius
r}
F:D
+
is P-compact if
X
is P0 -compact).
n
U B. ,
Q, S ( Q) = inf { r > 0: Q C and
qI-F
For subsequent j =1 J
a(Q) = inf{d > 0: QC
n
U
Q. j=l J
are ball-measure and set-measure if noncompactness
A continuous map F: DC X + X is said to be k-ball-contractive Ck-set-contractive) if S(F(Q)) 2 kS(Q) (a(F(Q)) 2 ka(Q)) for each bounded set QC D and some k > O. Clear-
of
Q, respectively.
ly, if
F:D
+
X
is completely continuous, then
(and 0-set-contractive).
If
D
F
semicontractive in the sense of Browder [3], then contractive for some
k
E
(0,1)
said to be ball-condensing if that
S(Q) # 0.
is
(see [21,26]). S(F(Q)) < S(Q)
For various properties of
□ -ball-contractive
F:D
is a open set and F
A map
for each
S(Q)
and
+
is strictly
X
is k-ballF:D
X
+
QCD a(Q)
is such
and examples
of the above classes of mappings see [6, 16, 25]. It is known that A-properness is invariant under compact perturbaIn some cases we can say more to indicate the generality of
tions.
178
Petryshyn
A-proper maps. and
T:X
w.r.t. when
F:D
➔
X
is k-ball-contractive
is c-accretive and continuous, then Tµ _ T+µF is A-propel for each µ e (-ck- 1 ,ck- 1 ). If c =a= 1, the same holds
r1
F
Thus, for example, if
X
➔
is ball-condensing and
is k-ball-contractive, then
µ e [-1,1].
F
In particular, if
F:D
➔
y > k.
is P -compact for any fixed y
X
For
other examples of A-proper maps see [4, 6, 15, 20, 27] and the more recent papers by the author, Browder, Fitzpatrick, Milojevi~, Webb, Fitzpatrick-Petryshyn, Petryshyn-Yu, Massabo-Nistri, Xianling, Dupuis, Toland, and others. Since the fixed point index for A-proper vector fields introduced in [8] will play an essential role in what follows, we introduce this notion here and state those of its properties which we shall use.
KCX
Let
be a closed and convex set and suppose that Let
for each DK= D ()Ki 0 DK
and let
relative to
w.r.t.
ra
D
K.
and
DK
x-Tx
n > No' where
and
oDK
and
Suppose
0
~
for
x e aDK.
DK
is such that
P
(K)
CK
n such that
I-T
is A-proper
Then the A-properness of
such that
x-Tn(x) i O
relative to
n
X
the closure and the boundary of
D = D () xn, DK K = K () X n n' n
is the boundary of
n
aDK
T:DK ➔ K
N0 e N+
implies the existence of and
be a bounded and open set in
n
= Dnn Kn' T n
K n
I-T
x e aDK
for
n
p TJn DK
Following in the
spirit of the multivalued topological degree defined by Browder and Petryshyn in [ 4], the following notion was introduced in [ 8 J .
(2) Definition. Let X e ;)DK. denoted by
X, r
K, DK and T:DK K be as above and X i Tx for a' fixed 12.oint index of T on D with respect to K, Then the --➔
Z' = NU {-N} U {±
IK(T,DK)' is a subset of
00 }
determined
as follows:
(i) {n.}CN+
J
An integer such that
me IK(T,DK) iK
nj
(ii) such that Here
(Tn. ,DK J nj
if there is an infinite sequence . + ) = m f or each J e N .
if there is an infinite sequence
lim iK (Tn. ,DK j nj J nj "iK" n
= ± 00 •
{n.}
J
C N+
is the finite dimensional fixed point index defined
via the Brouwer degree.
Note that the observation immediately preceding
Definition 2 implies that
IK(T,DK) ¥ 0.
In what follows we need the following properties: (Pl)
If
IK(T,DK)
t {O}, then
T
has a fixed point in
DK.
n
Positive Eigenvectors and Fixed Points for Maps in Cones ( :, 2 )
I£
T(x)
(F3)
If
F:[
x-F(x,t) e K t
x e DK
for
I"'
(,L)
i~ either tion:
anc
t
(I-F(T,•),DK)
U C0
J :::>
U 3D 2 K' then
3DlK
(T,
or
e [O,l], r(t,x)
~
O
is constant in
t
e [O,l].
Cl U J 2 , D1 () D 2
D
,
IK(:,DK) C lK)
and
is an A-proper homotopy such that
,l]xl\
then
[O,l],
E
x e DK
for
J
~
179
IK(T,
0
x e 3DK
£or
and if
We sa"; that
:KC:,D 2 K)
F:[O,l]xQ
Q C X, then
anc if
r)
➔
{t
X n.
is an A-proper homotopy
} C
[ O , l] , { x
"
r, ·,
:" (
g
➔
) = g. We rec al J trEt t
ax+b:;
If
E
K
a,L e
for
K
+
[c],oo)
is a cone, we wri~e
we have
I xii
In case
T
,'.:_
ell YI!
i
e~Ji:alent to the
some
ye ~
C)
(C,J]
ex
I ·II Ii ·II
_
c:
X
llx+-,,11
r-1)_
re
e:i.ger.',
K
E
I7 is eas
to
I x+ i
i.ndeper1dent of
Y/lxl]
>
x
a=
y
y
(Q)
[l,oo)
for
[23] _
µ
in the Banach e
(0,1), or
a!'e EOt norma: b ·,+ Ll ' C (Q)
and
( Q)
~s a bo~~ded domain.
C
2:t v.'a2 sf'.:.._.-:i,h'r. b\·
and that
K.
€
~onriegative furictioris
:he same cones in
t
l
CT
0 for each ye~
normality constant
o = o(K)
by
If one sets
o(y) =
and defines the quasio
o(K) = sup{o(y): ye K}, then it was
shown in [5] that for any cone in
X
one has the estimate
1/2 < o < 1
( 1. 3)
and in general the constant
1/2
in (1.3) is the best possible.
In view of the above results, our first theorem in this paper is the following improvement of Theorem 3 stated in [23] without proof. Theorem 1. scheme in
X
Let
K
Pn(K) CK
with
bounded neighborhood of such that
I-T:DK
be a cone in
+
X
X,
{Xn,Pn}
a projection
+
n e N , and let
for each
O e X.
ra =
Suppose
T,F:DK
ra
is A-proper w.r.t.
+
K
DC X
be a
are bounded maps
and the following condi-
tions hold: (Hl)
IK(T,DK) # {O}.
CH2)
There is a constant
is A-proper w.r.t. µ > 0
if (H3)
y
=
o.
r
y
for each
~
0
such that
6 = inf{II Fxll: x e aDK} > dd, where
quasinormality constant associated with Then there are
x e aDK
and
K
Tµ = I-T-µF:DK
if
µ e (0,1/y]
and
A e (O,d/6]
y > 0
+
X
and for each
a e [1/2,1]
is the
d = sup{llx-Txll :x e aDK}. such that
x-Tx = AFx,
i.e., (1. 2) holds. Proof.
We shall first consider the case when
y > 0.
Now, choose
Positive Eigenvectors and Fixed Points for Maps in Cones E
( 0 , o)
E
0 > yd/(o-E).
such that
II x+yll and
X
0
in
K,
11
= II tO/t)x+y)II
x+tyll
and
y = YE E K such that o(y)+E > o(K) and for all X in X. Hence, for all t > 0
> ( O-E l II xii
> o(y)II xii
o(y)
Then, by definition of
0
a = o ( K) , there exists
181
> t ( o - El
= tll n yod = ~
a> 1, in contradiction to
It is obvious that
H(t,x) ¥ 0
We claim that
0, i.e., x -Tx o o
(H3), d ..::_ II x 0 -Tx 0 II
Tn
be arbitrary and consider the map H(t,x) = Tn(x)-tmy.
If not, then there exist
H(t 0 ,x 0
T (x ) n o
Tn = I-T-nF, then
Let
given by
is an A-proper homotopy. that
and
d
and
Thus ( 0, l]
and
Then the preceding discussion, the choice of
x 0 -Tx 0 = nFx 0 +t 0 my.
E > 0, and condition (H3) imply that d > llx 0 -Tx 0 II
= llnFx 0 +t 0 my!! > (o-Elnllrx 0 II > Co-Elna
> (o-E)n•yd/(o-E)
which is impossible.
=d
since
Thus, H(t,x)
= 1/y,
n ~
0
x e aDK
for
t e [0,l].
and
Therefore, by the homotopy property (P3), IK(T+nF,DK) = IK(T+nF+my,DK).
me N+
Now, the above equality holds for any implies that
IK(T+nF,DK) = {0}.
then for each
m
in
(I-T-nF)(xm) = my. T
and
F
N+
we could choose
Since
and this, we claim,
Indeed, if this were not the case,
y ¥ 0
and
xm e DK
such that
(I-T-nF)(DK)
is bounded because
are bounded maps, we have a contradiction when
m
is large.
Consequently, IK(T+nF,DK) = {0}. Consider now the mapping x-Tx-tnFx
for
t
boundedness of
E [0,l]
F
that
H0 :[0,l] DK ➔ X
and H0
x E DK.
given by
is an A-proper homotopy.
IK(T,DK) ~ {0}, while we have just proven that Hence there exist some
t e (0,1)
i.e., x = Tx+AFx
x
In case
o
y
> dE/(o-E),
o(y) > o-E
and
=
with
E dDK
0, we let
n
=
A= tn
1/E
and choose
As before, there exists llx+yll ::_ (o-Elllxll
such that
and
Ee (0,o) 0
x
H0 (t,x) = 0,
(O,d/8].
y = yE € K
for all
argument is the same as in the case when
€
By (Hl) we have
IK(T+nF,DK) = {0}
x e aDK
and
H0 (t,x) =
It follows from (H2) and the
in
such that
such that K.
The rest of the
y > 0. Q .E .D.
We note that Theorem 1 includes a number of special cases.
In what
182
Petryshyn
follows we only state here two new special cases which extend some known results. Corollary 1. be fixed and let
=
6
Let F:DK
inf{[[ Fxl[ :x
E
➔
clDK} > by/ CT , b
Then there exist Proof. when
T
=
K, CT ' r and D be as in Theorem 1. Let y -> 0 K be a bounded and P -compact map such that y
x E clDK
=
and
sup{[lxl[
E
X
A E (0,b/6]
such that
Corollary l follows from ~heorem l when
0, then
=
IK(O,DK)
{l}-/ {O}
( 1. 4)
cl DK}.
x = AFX.
T = 0.
by (P2), i.e.,
Indeed,
(Hl) holds.
The hypothesis (H2) reduces in this case to the condition that is Py-compact, while (H3) reduces to (1.4).
F:DK ➔ K
Consequently, Corollary l
follows from Theorem 1. Remark 1.
When
y = 0, ~heorem land Corollary 1 were proved in
[8] by a method which cannot be extended to the more general
maps when 0
y > 0.
Since a completely continuous map
F:DK
class of K
➔
is
0 -compact, Theorem Vl.l in [12] is a special case of Corollary 1 when
the Banach space
r
has a projectionally complete scheme
X
~oreover, since every k-ball-contractive map
F
= {Xn,Pn}.
can be easily shown to
y > k when r = r 1 by using the argument of Webb, it follows that Corollary l in [23] and, in particular, Theorem
be ~-compact for any fixed l
in [14]
(for
w.r.t.
x = AF
Corollary 2.
and
r1 .
provided Let
Suppose that
1
::.s } -compact and
\ e ( 0,b / 6]
8 > b/CT.
K, CT T: DK
Let
IK(T,D) f. {O}.
k > 0
fork-ball-contractive
r 1 , Corollary l shows that there exist
such that
=
normal) follow from Corollary l
Furthermore, since a ball-condensing map
maps.
r
K
and ➔
F:
D
be as in Theorem 1 and let k 0 e [0,1)
is k 0 -ball-contractive with
K ➔
K
beak-ball-contractive map with
and such that
dk 0 = inf{IIFx[I: x e clDK} > CT(l-k), 0
where
d = sup{[lx-Tx[I: x
A e (O,d/6]
Proof.
such that
E
clDK}.
Then there exist
x e clDK
and
x-Tx = \Fx.
Note that when we take
y
such t:iat
y
>
k
1-k
and 0
8 > dy/CT, then to deduce Corollary 2 from Theorem 1, it suffices to
show that
Tµ = I-T-µF
But, for any such
is A-proper w.r.--'.::".
µ, the map
~+µ?
is
r1
for each
µ
E
(0,1/y].
(k 0 +k/y)-ball-contractive with
Positive Eigenvectors and Fixed Points for Maps in Cones
183
k 0 +k/y < k 0 +k(l-k 0 )/k = 1.
is
Hence, by the results of Webb [27], Tµ
A-proper.
Q.E.D.
Let us note that for some applications, and from the practical point of view, the following version of Theorem 1 may prove to be useful since we don't have to compute
in case
d = inf{llx-Txl/: x E clDK}
K
is a normal cone. Theorem l'.
If
K
is a normal cone in
Theorem 1 remains valid with
A E (D,Tb/o]
X, then the conclusion of if condition (H3) is replaced
by (H3') and
o = inf{/IFxll: x E clDK} > by/o, where
o = llT
with
T > 1
such that
b = sup{l/xll: x e clDK} if
II ul/ .'.:_ T/1 vii
u,v EK
and
D < u < v.
Proof.
The proof of Theorem l' is similar to that of Theorem 1.
It suffices to show that if we let with
E > 0
such that
H:[D,l]xDK ➔ X and
x e clDK.
by
=
o > bE/o
=
n bET
1/y if
if y
H(t,x) = Tn(x)-tmy, then
=
y > D 0)
H(t,x) 1 0
H(t,x) 7 D, then
If we can show that
and this together with (Hl) yields the conclusion. H(t,x) 1 0 when
for
y > D
there are
x 0 e clDK
x 0 -Tx 0 -nFx 0 -t 0 my 11•11
t E [0,1]
and
(the case when
=
and 0.
semimonotone on
t
0
X
D
for
1/E
t E [0,1]
IK(T+nf,DK) = {O} To show that
is handled similarly).
E [D,l] x0
=
n
x E clDK, we first consider the case
y
Then
(and
and define
=
such that
H(t 0 ,x 0
Tx 0 +nfx 0 +t 0 my > nfx 0
and (H3') holds, we have
)
Suppose that
=
and thus, since
b > Jlx 0 /I
> onl1Fx 0 I
.:_
ono > on•by/o = b, a contradiction. Q .E.D.
2.
Existence of positive fixed points
Using Theorem 1 and the properties of the fixed point index stated in Section 1, we are now in position to extend Theorem 2 for cones proved in [24] so as to obtain the following new theorem concerning the existence of positive fixed points. Theorem 2.
Let
K
be a cone in
scheme as in Theorem 1, D1 with
D1 C D2 , and let
is A-proper w.r.t. ( Cl)
r.
-2 T:DK
D2
and ➔
K
X,
r
= {Xn,Pn}
a projectional
bounded neighborhoods of
be a bounded map such that
D
in
-2 I-T:DK
X ➔
Assume also that the following conditions hold:
There is a bounded map
-1
C:DK
➔
K and
such that
X
184
Petryshyn
the restriction for
X €
for
X €
C
n
Pn C
of
to
µ > 1 and n > N furthermore, X t- tT (x)+Cl-t)C ( X) aD 1 o' K ' n n n aD 1 t € [ 0, 1] and n > N 0 K ' n
(C2)
-2 F:DK
There is a bounded map
(H2) and (H3) of Theorem 1 and A E
is continuous and
K
➔
x-Tx t- Afx
which satisfies conditions 2 for all X E aDK and
(O,d/8]. Then there exists
x0
-2 -1 DK\DK
such that
x 0 = T(x 0
The same assertion is true if we assume that condition (Cl) holds on D2 while E
).
K
1
(C2) holds on
DK.
Proof. First, we may assume without loss of generality that T 1 2 has no fixed points on aDK and on aDK. Second, we will only provide a detailed proof for the case where y > 0 in (H2) since the case when
y = O
is handled in a similar way.
Now, for each tC ( x)
for
n
n > N0
t E [0,1]
define a map
,
and
of condition (Cl) that
x E
Dl
K
n
x-Mn(t,x) t- 0
-1
Mn:[O,l]xDK ➔ Kn by Mn(t,x) = n It follows from the first part
for
t
E
[0,1], x
1
oDK
E
and
n
the properties of the finite dimensional fixed point for all
= { l}
n > N . 0
part in condition (Cl), the map by
n > N0 t
E
all
= tTn(x)+(l-t)Cn(x) and
[O,l]
n > N0 -1
I-T:DK
➔
X
X
€
1
aDK. n
➔
for
x t- Gn(t,x)
for
by Definition 2 since x-Tx t- 0
for
in virtue of condition (C2), Theorem 1 (with Indeed, if
given for each
n
) = iK (Cn,D~) = {l} n n
and, therefore,
IK(T,D~) = {O}.
K
is such that
Hence
is A-proper and
Now, in view of the second
IK(T,D~)
1 x E aDK.
On the other hand,
DK= D~) implies that
were not equal to
{O}, then
since the first part of condition (C2) implies (H2) and (H3) of Theorem
1, it would follow from that theorem that there exist A E
(O,d/o]
tion in (C2). Now, let
x E aD~
and
such that Thus Q
=
x-Tx = Afx, in contradiction to the second assump2 IK(T,DK) = {O}.
D2 \n 1
has no fixed points on
and observe that oQK.
aQK
=
aD~ U aD~
and that
Thus, by the additivity property (P4),
T
Positive Eigenvectors and Fixed Points for Maps in Cones
since
is the singleton.
that
{-1}.
-2,-1
such that
DK DK
It follows from the last equality
Hence, by property (Pl), there exists
= T(x 0
x0
185 E QK =
x0
).
To pr'ove the second part of Theorem 2, note that when condition 2 DK then, by the same arguments as above, we see that Similarly, when (C2) holds on D1 then, in view of IK(T,D~) = {l}. 1 K 1 Theorem 1 with DK= DK, we see that IK(T,DK) = {O}. The property (P4) (Cl) holds on
implies in this case that
{l}
ceding case, there exists
x0
and so again, as in the pre-
such that
QK
E
x
T(
0
x ). 0
Q.E.D.
For the purpose of applications we state Theorem 2 in the following practically useful form which, as we will indicate below, unifies and extends a number of results obtained earlier by this and other authors for special classes of maps by using various methods. Theorem 2'.
Let
K
and
be as in Theorem 2, let
T
r 1 ,r 2 e (0, 00 )
J\(
with
0 ,r) ➔ K r = max{r 1 ,r 2 }, let BK(O,r) = B(O,r)() K and let T: be a bounded map such that I-T:BK(O,r) ➔ X is A-proper. Suppose also 2 that (Cl) and (C2) of Theorem 2 hold with D1 B( □ ,r 1 ) and D = B ( 0 , r 2 ) . Then there exist X e K such that min{r 1 ,r 2 } ~ ~x 0 j ~ max{r 1 ,r 2 }
and
x
0
= T(x ).
0
0
Special cases. the maps
T, F
We shall now show that, for suitable choices of
and
point theorems:
C, Theorems' 2 and 2' contain two classes of fixed
(A)
Variations of the cone-expansion and the cone-
compression theorems of the type established in [12] when
T
is compact
and Nussbaum, Potter, Reich, Hahn, Fitzpatrick-Petryshyn, Milojevic and others when T is k-ball-contractive (see [2, 22]) for exact references. ( B) The fixed point theorems when T:K ➔ K has a Frechet derivative
T0
at
0
are of type (A)
and/or
L~(X)
T00
or
at
Li(X)
along
such that
K
T
in the sense of [l].
and/or
0
Cone-expansion and cone-compression type theorems.
X E aD 1 K
and
X
II h[[
n
fc tTx
> dy/o
defining
\ft
E
for and
-2 F:DK
[O,l] t
➔
K
by
n 2:_ N0 ", holds when and
[ 0 'l]
E
x-Tx
\f
fc
11h
for
Fx = h
while (H2) also holds since
X
1
e oDK.
X
c oD 2
for
X
I-T-µF
K
and
-2 T: DK
Choosing /\ E
➔
is P 1 -compact
K h
0
€
K
such that
( 0' d/[[ h[[], and then
-2 DK we see that (H3) holds, is A-proper for each ).1 > 0 E
C=0
"x fc tP n Tx
in (Cl), then condition (Cl), which in this case reduces to for
When
T
186
Petryshyn
because
F
is compact.
Thus, in this case we deduce from Theorem 2'
the earlier results in Goncharov [10] when and of Hamilton [11] when
T
D2 = B(0,r 2 ), and of Milojevic [15] when 2 and
T:BK(0,r) + K is P-compact D1 = B(0,r 1 ) and
is P 1 -compact with D1
D2
and
are as in Theorem
is singlevalued. Theorem 2 in [28] is also a special case of Theorem 2.
T
An immediate consequence of Theorem 2' is the following extension of Corollary 1 in [24] which is a new result. Corollary 3. with
(Dl) x
E
for
3BK(0,r 1 )
(D2) with
+
K
There is a bounded ball-condensing map
Cx t µx
that for
Let r = r 1 and T:BK(0,r) Suppose further that:
k 0 e (0,1).
x e 3BK(0,r 1 ) and and t e [0,l].
µ > 1, and
be k 0 -ball-contractive C:BK(0,r 1 )
There is a bounded k-ball-contractive map
k > 0
such that
x-Tx t AFX
for
x
0
F:BK(0,r 2 )
dk o = inf{II Fxll: x e 3BK(0,r 2 )} > 0(1-k)
x e 3BK(0,r2)
Then there is
+
such
+
K
and
0
A e (0,d/o].
and
K
x t tCx+Cl-t)Tx
e K
Proof.
To deduce Corollary 3 from Theorem 2' it suffices to show ' (Dl) (Cl) , and that implies that the map Tµ = I-T-µF lS A-proper w.r.t. k and rl for each µ e (0,1/y], where y lS chosen such that y > 1-k 0
o > dy/0.
The latter fact has been shown to be the case in our proof
of Corollary 2, while the fact that (Dl) implies (Cl) follows in the standard way from the fact that
C
is ball-condensing and
Cx t µx
for x e 3BK(0,r 1 ) and µ > 1 and that F(t,•) = tC+(l-t)T is also ball-condensing (in the sense that S(F([0,l]xA)) < S(A) for each A
E
BK(0,r)
with
SCA) t 0)
and
x t Cx + (1-t)Tx
for
t e [0,1]
x e 3KB(0,r 2 ).
When all
A> 0
F
and
Q.E.D. is compact (i.e., k = 0
in (D2)) and
x-Tx t AFx
for
and
x e 3BK(0,r 2 ), Corollary 3 was first proved by the author in [22] for a general Banach space X where it was also shown that in this case it extends and unifies the corresponding results of Krasnoselskii, Gustafson-Schmitt, Turner, Gatica-Smith, Edmunds-PotterStuart, Nussbaum, Potter, Fitzpatrick-Petryshyn, Amann, Milojevic, Petryshyn and others.
mentioned authors.
See [22] for the exact references of the above
187
Positive Eigenvectors and Fixed Points for Maps in Cones ~onzero fixed points of differentiable maps.
( B)
In order to
deduce from Theorem 2 the existence of nonzero fixed points for compact maps, which are along
0
rechet differentiable either at
K, we recall firs~ some definit~ons.
1
0
er at
0
We emit some details except
for one lemma which is necessary bGt net proved in [24].
A map
T:K
X
➔
is said to be (Freche~) differentiable at
if there exists a map
K
( C , h)
T(O)+c'
is said to be
The ma? e ~(X)
It
h
1,;i th
~ 00
are u~iq~ely determined. T~,(K) CK
and
K
I
ti I )
00
( 2 .1)
➔ 0•
if there is
K
II hll
as
(2•2)
➔
is a tctal cone, then the maps
C
ancl
It ~s easy to
'l (K) CK
II hll
as
along
K 0 (
at
the derivati__\/es o~
,
in
(h ) =
not hard to show that i"
co
o(llhll)
w(O,h) =
sy~pto!icaJly linear along
such that for each
T(h)
anC
1,Jith
0
h e K
such that for each
e L(X)
along
K
ee that if
respectively,
T:K
➔
K, then
,,;henever they exist.
Before we appJ.y Theorc~ 2'
t.o ob~a~~ the exister1ce of nonzero
0 1 -ccnpact maps T:K ➔ K satis~ying either (2.1) or .2), we will need ~he fol:owing s le fact (see [12]): If u,v e K,
fixed points for u 1-
C
} C
a
ex::sts
II
()
(J,oo)
S'Jcr_ that
a
lS
3·J.cf'. ~:-1at
e
¢ : 1
}
for all
e K
\1-A
and this implies that
then
n
;\
i~
co::_r,c
BC::,,·)
n
in
X.
Then
K
such that ;\h
A~ f
for all
BK(G,~) = 8(0,r)
K, the closure
~K(C,r)
K, ,:,nc:; the bo:nc'a:n·
of
3BK(O,r)
n
K
BK(C,r) of
3(=,llnK.
( 0, r)
the ~ollcw1ng lemma which is related 1:1
r,
~errirr.a 1.
If
[
= ].
(',r)
_,_
co~tinuc~s 3nd A-proper w.r.t. to any closed subset
rcst_:-ict
( :,1 ) }C
M
of
clos e
S\;c'i
tl:at
G(
)
➔
g
as
188
Petryshyn
k .-,.
for some
oo
all
g
in
n, Pn(x) .-,. x
that for each
K.
Since
for each and
k
x
in
ck= 1/k
= 1, Pn
Pn(K) CK, IIPnll K, and
G
C
Pn+l
for
is continuous, it follows n(k) EN+
there exists
n(k) > k
with
such that ( 2. 3)
for sufficiently large
and
k
JI Txk-Twn(k)IJ ~ Ek .-,. 0
r1
and our properties of the scheme Pn(k)Gwn(k) .-,. g
k .-,.
as
= {Xn,Pn}
wn(j) .-,. x
x . .-,. x J
as
j .-,.
as
j .-,. and
00
k .-,.
This
00
imply that
whence on account of the A-properness of
00 ,
it follows that there exist a subsequence that
as
x EM
wn(j)
Gx = g.
and
00
since
M
and
x E BK(0,r)
G such
This and (2.3) imply that is closed.
This proves Lemma 1.
We now are in position to deduce from Theorem 2 the following corol-
laries for differentiable maps. Assuming (Cl) of Theorem 2, our first corollary in this section shows how (C2) of Theorem 2 is verified when
T0
+ L (X).
or
T00
is of type
1
Corollary 4.
Let
is A-proper w.r.t.
T:K .-,. K
be a bounded map such that
I-T:K .-,. X
and assume also that:
fl
> 0' N e N+ and a bounded map 0 such that (Cl) of Theorem 2 holds for Dl = BK(0,r). K (al)
There exist
r
Suppose further that either (bl) or (dl) holds, where T(0) = 0, T
(bl) that
(I-T 0 )jK T
T0
has the derivative
is A-proper w.r.t.
r1
and at
has the derivative
oo
T0
at E
along +
is A-proper w.r.t.
TOOELl(X).
Then, in either case, T
has a fixed point in
+
0
along
K
such
L1 (X) K
such that
0
K.
Since (al) holds, to deduce Corollary 4 from Theorem 2, it
Proof.
suffices to show that condition (C2) of Theorem 2 is implies by either (bl) or (dl) for a suitable ball B = D2 . Since the arguments are almost identical in both cases, we prove these implications simultaneously for T
satisfying (bl) and (dl).
First note that (bl) and (dl) can be stated
as: (dB) where
B
iK
=
T(h) = TB(h) + QB(h) 0
or
B
= oo
and
is A-proper and
Now, since
TB
E
1
TB
TB E
with E
+
L(X)
IIOB(h)JI
= o(ilhll) .-,. B
is such that
(h E K)'
TB(K) ~ K,
L 1 (X).
L+(X), there are
AB> 1
and
hB
€
0
K
such that
Positive Eigenvectors and Fixed Points for Maps in Cones
TB(hB) = ABhS. r 00 > r
We claim that we can choose F:BK(o,r 6 )
and define a map
K
+
rS > 0
by
189
with
F 6 (x) ~ hB
r
0
< r
and
for
x e BK(o,r 6 ) such that (C2) of Theorem 2 holds with y = 0 in (H2) and (H3). Indeed since FS:BK(O,rS) + K is compact, it follows that I-T-µF:BK(o,r 6 )
+
X
y = O.
holds for
is A-proper w.r.t.
for each
µ
>O, i.e.,
inf{JIFxll: x e aBK(o,r 6 )} = lih 6 11
6
Since
r1
( H 2)
we
> 0
r 6 > 0 chosen 0. Indeed, if above is such x-Tx ~ AFx for x e aBK(O,rS) and A> this were not the case, then there would exist sequences {rs} n ' see that (H3) holds when
OB} C n
or
CO 00 ) '
and
S = oo, AS > 0 n
{ K6 } C
n
and
Finally, we show that
y = 0.
11
K
x~II
such that
= rS n
rS n
+
for each
S
as
n
with
N
E
+
S = 0
and
It follows from last equality and (d 6 ) that
Since
llzSII n
as
1
is bounded.
llxSII n
+
S, it follows that
Hence we may assume that
as
{zB} and n. J with ns > 0 (I-TS)(z 6 ) = nSh s
rl, it follows from Lemma 1 that there is a subsequence ZS n. J because
ZS e aBK(O,l)
such that
since
ZS
➔
and
Since
lows from the last equality, by induction, that zB-nS(l+AS+A~+ ... +A~)(hS) e K and
for each
t'+
' .
in
n
Since
ns > 0
AS> 1, the last relation is impossible by the remark preceding
Corollary 4.
Thus,
(C2) holds and so Corollary 4 follows from Theorem 2
provided that (al) and either (bl) or (dl) hold.
Q.E.D. Remark 2.
Under the stronger assumption that
compact and the restrictions of
T
0
and
Corollary 4 was first obtained in [24].
+ K is P 1 are also P 1 -compact, For the proof given in [24] to
T00
to
T:K
K
be correct it seems that one should use Lemma 1 instead of Proposition l.lC from [20]. In our next corollary, assuming (C2) of Theorem 2, we show how (Cl) of that theorem is verified when T00 1K
T:K
is also~ -compact and either
+
T0
K or
is P 1 -compact, T 0 \K T00
lies in
L~(X).
or
Petryshyn
190
Corollary 5.
Suppose
T:K
K
+
is a bounded P 1 -compact map such
that: (a2)
There exist
r > 0
and a bounded map
F:BK(O,r)
K
+
such
that condition (C2) of Theorem 2 holds. Suppose further that either (b2) or (d2) holds, where: (b2)
T(O) = 0
such that
T 0 1K
( d 2)
and
T
has the derivative
is P 1 -compact and
has the derivative
T
P1-compact and
T e L~(X). 00
Proof.
T
T 0 e L~(X). T00
at
along
00
Then, in either case,
at
0
0
along
K such that
K
T j K is 00
0
T has a fixed point in K.
To deduce Corollary 5 from Theorem 2, it suffices to show
that (Cl) of Theorem 2 is implied by either (b2) or (d2) for a suitable B = D1 .
ball
As before, the proof that either (b2) or (d2) implies (Cl) will be carried simultaneously. derivative at TS(x) i µx x-TSx i O
S
for for
NS e N+
and
n ~ NS.
Choose
mS/2jxj
for
all
along µ ~ 1
=
0
K.
Sine;
and
x e K.
or
S
=
00
and let
TS
be the
TS e L~(X), it follows that
with
3BK(O,r 8 ).
Now, since
TS\K
II x-P nT S (x)II ~ msll xii
such that
rs> 0 E
t e [O,l], all
ro < r
and
roo > r
for
x e Kn
such that
and
IITx-Tsxll
NS
we have
is P 1 -compact and TB e L~(X), it follows from PnTS(x) i µx for all x € aBK(O,rs), all n ~ NB and
Moreover, since Lemma 1 that -
S
is P 1 -compact and x e aKB(O,l), it follows from Lemma 1 that there exist
ms > 0 x
Let
TslK
+
NB c N . Thus, (Cl) holds with C = TB and so Corollary S follows from Theorem 2 provided that (a2) and either (b2) or (d2) hold. Q.E.D.
some
Remark 3.
When
y = 0
in condition (a2), Corollary 5 was proved
by the author in [24]. An immediate consequence of Corollaries 4 and 5 and Remarks 2 and 3 is the following special case established in [24]. K
is a bounded and ? 1 -compact map
and the derivatives
Too at 0 and at 00 are P 1 -compact. ~ 00 ,K K provided one of the following
Corollary 6. with along
T(O) = 0 K Then
Suppose
are such that T
T:K
➔
To\K and has a fixed point in
T
0
and
Positive Eigenvectors and Fixed Points for Maps in Cones
191
conditions hold: (a)
T0
E
Li(X)
and
Too
E
L~(X)
(b)
T
E
L~(X)
and
T
E
Li (X).
0
0
Corollary 6 contains the corresponding results of [12] when T completely continuous and of [l], [7] and [22] when T is k-ballcontractive with
k
0
(4)
is also satisfied. Our first two results (Theorems 5 and 8) are valid in every Banach space.
On the other hand, our next two results
are valid only in special Banach spaces.
(Theorems 11 and 14)
As a matter of fact, in order
to formulate Theorem 14 we introduce a new geometric property of (infinite-dimensional) Banach spaces.
We conclude with a consequence of
Theorem 8 which involves the fixed point property for nonexpansive mapings (Theorem 15). Our first result will be preceded by several lemmas. Lemma 1.
(y
l
is defined by (2) and xt E D(Jt), then n-1 (1+1/t)nlJtxt-ynj s _II (1+(1-c.)/t) !Jtxt-Yol i=0 l. n-1 • n-1 + (1/t) -~ ci(l+l/t)l.lxt-Tyil -~ (1+(1-ck)/t) i-0 k=i+l for all n .e 1.
Proof.
If
Since
n
(l+t}Jtxt
(1+1/t) n+l IJtxt - Yn+l I
s (1+1/t) (1-cn) (1+1/tlnlJtxt-ynl + cn(l+l/t)n(l/t) (l+t) IJtxt-TYnl s (1+1/t) (1-cn) (1+1/t)n!Jtxt-ynl + cn(l+l/t)n(l/t) (lxt-TYnl+tlJtxt-ynll (1+(1-c n )/t) (1+1/t)nlJtxt-y n I+ c n (1+1/t)n(l/t) lxt-Ty n
I-
The result now follows by induction. Although the next lemma is well known, we include a proof for completeness. (znJ are defined by (2), then n-1 ly 0 -ynl s 2ly 0 -z 0 l + (i~ol lz 0 -Tz 0 1
Lemma 2.
for all
If
(ynJ
n .e 1.
and
Successive Approximations for Nonexpansive Mappings Proof.
We first note that
195
lzn+ 1 -Tzn+ll :s: lzn+l-Tznl + !Tzn-Tzn+ll
,, I (1-c n ) z n +c n Tz n -Tz n I + I z n -z n +l I [lz -Tz IJ is decreasing. Hence n n
I z n -Tz n I,
so that the sequence
jz 0 -zn+ll:s:iz 0 -znl +cnlzn-Tznl,, lz 0 -znl +cnlz 0 -Tz 0 j. Since jyo-Yn+ll $ IYo-zol + lzo-zn+ll + lzn+l-yn+ll +jz 0 -zn+ll' the result now follows by induction.
$
2jyo-zol
The range condition (3) is satisfied if and only if there exists a sequence
[(ti,xt_JJ
such that
J.
ti ➔ °', xt. E D(Jt.l J.
(5)
lxt_l/ti ➔ 0.
and
J.
J.
Our next lemma is also known [8, Proposition 3], but we present a different and simpler proof. main
D(A)
and range
,;; lx 1 -x 2 +r(y 1 -y 2 ) clear that
I
A= I-T
Recall that a subset
R(A)
for all
A
of
is said to be accretive if [xi,yi] EA,
i = 1,2,
is accretive whenever
T
and
X
x
X
with do-
lx 1 -x 2 1
r > 0.
is nonexpansive.
It is
The
analog of condition (3) is therefore lim inf d(0,R(I+tA))/t = 0.
(6)
t ➔"'
Lemma 3.
Let
X
be a Banach space,
that satisfies (6), and
( (ti,xt_JJ
A c X xX
an accreti ve opera tor
a sequence satisfying (5).
Then
J.
Proof.
Let
d= d(0,R(A)).
Since
(xt-Jtxt)/t
belongs to
is clear that
0 < s < t
On the other hand, for
and
[x,y] EA
jJs(x+s~-Js( (s/t)xt+(l-(s/t) )Jtxt) ,;; lx+sy-{(s/t)xt+(l-(s/t))Jtxt)
we have
I
I
,. (1-(s/t)) jx-Jtxtl + (s/t) lx-xtl + sly!. Hence and lim sup !Jt x i ➔~
Therefore
i ti
l/t. ,, IYI· J.
li~ sup jJt xt l/t. ,;; d i ➔ ro
i
i
J.
and the result follows.
Finally, we note the following simple fact.
R(A),
it
Reich and Shafrir
196
Lemma 4.
For each
n
~
1,
lim t(l-
t➔m
n-1
n-1
I1 (1-c./(l+t)))
i=0
L
i=0
1
C .•
1
We are now in a position to establish our first result. Theorem 5. T: D
X
➔
(ynJ
Let
D
be a closed subset of a Banach space
a nonexpansive mapping which satisfies (3).
X
and
If the sequence
is defined by (2), then n-1 lim IY I/( L c.) = d(0,R(I-T)). n➔ m n i=0 1 n-1
L'. c. by a n and d ( 0 , R ( I -T) ) by d. On the one i=0 1 hand, Lemma 2 shows that lim sup iY I/a :!. d. On the other hand, n n
Proof.
Denote
Lemma 1 shows that
IY 0 -ynl
n➔ °'
IYo-Jtxtl - !Jtxt-ynl ~ n-1 n-1 n-1 (1- I1 (l-c./(l+t)l)!Jtxt-Yol- (1/(l+t)) L c.lxt-Ty.l I1 (1-ck/(l+t)) 1 1 k=i+l i=0 i=0 1
for all that
n
!y 0 -y
~
l
I~ n
~
and
xt E D(Jt).
ad n
for all
Therefore Lemmas 3 and 4 now imply
n ~ 1,
and the result follows.
In order to establish our second result we need two more lemmas. Lemma 6. If {en} is bounded away from defined by (2), then for all k ~ 1,
Proof.
Fix
{!y n -Ty n IJ
k
~
1
0
and
1,
and
{yn}
is
and denote the limit of the decreasing sequence
by
L. Since n+k-1 L'. (yJ.+1-yJ.l Yn+k - Yn = j=n
it is clear that
lim sup IYn+k-yn I/ ( n➔ m
n+k-1
L
j=n
C.) J
n+k-1
L
j=n
:!.
c. (Ty. -y.) , J J J
L.
Now we note that by [4, Proposition l] IYn+k-yn! ~ ITYn+k-Tyn!
~ I Ty n+ k -yn I - ITyn -yn I ~
n+k-1
n+k-1 1 (1-cJ.)- (ITYn+k-Yn+kl-!Tyn-yn!)+( _L CJ.) jTyn-yn'· J=n J=n
.rr
Assuming, as we do, that the sequence and l we conclude that n+k-1 lim inf IYn+k-yn[/( _L cj) ~ L. n➔ °' J=n
{c
n
J
is bounded away from
0
Successive Approximations for Nonexpansive Mappings
197
Hence the result. Lemma 7.
[c n J
If
is bounded away from
O
and
1,
and
defined by (2), then the limit of the decreasing sequence is independent of the initial point Proof.
Let
limits of tively.
{z J
y0 •
be another sequence defined by (2), and denote the
n
{!y -Ty IJ and (lz -Tz IJ n n n n We first note that
I !Yn+k-yn!
Since the sequence
s,
(en}
n.
L(z 0 )
and
respec-
n+k-1
( L
j=n
C . ) -
J
1.
0,
is bounded away from
find an integer
I I y + k -y I I n n for all
L(y 0 )
- lzn+k-znl I "" 1Yn+k- 2 n+kl + [yn-znl "" 2 lyn-znl "- 2!y 0 -z 0
positive
by
Iz
k
we can, given a
such that
n+k-1 + k -z I/ ( C . ) I < s n n j=n J
z:
By Lemma 6 we can now find, for this
k,
an integer
N
such that
and
n+k-1 I lz +k-z I/( Z: c.) -L(z 0 ) I< n n j=n J
for all .:
n ~ N.
!L(y 0 )-L(z 0 ) I< 3s,
Thus
is arbitrary.
Theorem 8. T: D
➔
X
Let
i.
L(y 0 )
and
be a closed subset of a Banach space
D
a nonexpansive mapping which satisfies
bounded away from
0
(2) , then for all
k
lim [y +l-y l/c n n n
n ➔ 0>
L(z 0 ) because
and ~
1,
and the sequence
1,
~~:
( 3) •
(yn}
X
and
If {c J is n is defined by
n+k-1 !Yn+k-yn[/( j~n cj)
n-1 lim IY [/( Z: c.) = d(O,R(I-T)). n➔.,, n j=O J Proof.
Since
lim IYn+l-ynj/cn
is independent of the initial point
n ➔ ro
y0
by Lemma 7, it must equal
d(O,R(I-T)).
The result now follows by
combining Theorem 5 with Lemma 6.
We do not know if this result remains valid
when condition (3) is not
assumed.
{c
It is indeed true if the sequence
Theorem 2.1]. (3)
n
J
is constant [1,
In this case the limit of Theorem 5 also exists even if
is not satisfied [8, Lemma 1], but it is not always equal to
d ( 0, R ( I-T) ) .
198
Reich and Shafrir
We now turn our attention to special Banach spaces.
We begin with
another lemma. Lemma 9.
Let
T: D
a nonexpansive mapping which satisfies (3).
by
➔
an
X
and
D
be a closed subset of a Banach space
d(0,R(I-T))
by
d.
If the sequence
Ex*
(2), then there is a functional z 2 ((y 0 -yn)/an,z) ;;,: d for all n.
Proof.
with
let zt belong to { (ti,xt_l} satisfy (5). Let
For
of a subnet of (yO-yn,zt)
J.
lzl = d
such that
J((y 0 -Jtxt)/t),
xt E D(Jt)
the sequence
X and n-l Denote L c. i=0 1 {yn} is defined by
z
and let
be a weak-star limit
Since
{zt_J. J.
(yO-Jtxt,zt) + (Jtxt-yn,zt) ;;,: lztl(lyo-Jtxtl- !Jtxt-yn!l,
we can use Lemmas 1, 3 and 4 to conclude that for all
n.
This inequality and Theorem 5 now show that clear that
lzl ~ d,
lzl ~ d.
Since it is
the proof is complete.
Recall that the norm of a Banach space ferentiable if for each
x
lim ( I x+ty I - Ix I ) / t t""0 is attained uniformly for
X
is said to be Frechet dif-
in its unit sphere
U = fx EX: Ix! = 1}, (7)
y
in
U.
We shall then write that
X
is
(F) •
Our next lemma is known (cf. Lemma 10.
x*
[2]).
is (F) i f and only i f every sequence
which there exists
w E
x*
with
lwl = 1
such that
{x n J
C
X
for
lim lxnl n"""'
lim (xn,w) converges. n"""' Theorem 11. Let D be a closed subset of a Banach space X and T: D _, X a nonexpansive mapping which satisfies ( 3) • Let the sen-1 quence {ynJ be defined by (2) and denote 1: c. by a n If x* i=0 J. has a Frechet differentiable norm, then v = strong limy /a exists, n➔"' n n and -v is the unique point of least norm in cl (R (I-T)). Proof.
Let the sequence
f (ti,xt_lJ J.
the functional obtained in Lemma 9. u = strong
lim Jt.xt_/ti i """' J. J.
satisfy (5), and let
z
Ex*
It is known [8, Theorem 3] that
exists, and that
-u
is the unique point
be
Successive Approximations for Nonexpansive Mappings
199
of least norm in cl(R(I-T)). Moreover, the proof of [8, Theorem 3] shows that z E J(-u). Combining Theorem 5, Lemma 9 and Lemma 10 we see that Since u
=
v
X
v.
strong
limy /a also exists, and that z E J(-v). n n is certainly strictly convex, [8, Lemma 5] now shows that n ➔"'
This completes the proof.
exists when X is lim y /a n n assumed to be only reflexive and strictly convex. These results improve upon Theorem 1 and Corollary 2 of [3] because we show that the weak range condition (3) is sufficient for the conclusions to hold.
A similar argument shows that the weak
n➔°'
(This condition, introduced in [8], is, in fact, also necessary.) Since X is uniformly convex if and only if the norm of x* is uniformly Frechct differentiable, they also improve upon previous results of the first author [11, 12]. We continue with a convergence result in the setting of Theorem 8.
To
this end, we introduce a new geometric property of (infinite-dimensional) Banach spaces. Recall that to each functional w in the unit sphere of x* there corresponds a face F of the unit sphere U of X,
namely all those
x EU
the norm of a Banach space
tiable (LUF) if for each face formly for all y in U and
x*
whenever
(x,w) = 1.
for which X
We shall say that
is locally uniformly F~echct differenF x
of in
is uniformly convex or
U the limit (7) is attained uniF. It is clear that X is (LUF) X
is (F) with compact faces.
In
analogy with Lemma 10, we also have the following characterization. Lemma 12.
x*
is (LUF) if and only if every sequence [x J c X for n lxl = 1 and a sequence [w} c J(x) n converges.
which there exists x EX with such that limlx I = lim(x ,w) n ➔~
n
n ➔ oo
n
n
We also note in passing that if convex, then x* is (LUF). Lemma 13.
If
x*
X
is reflexive and locally uniformly
is (LUF), then every accretive operator
that satisfies (6) has the following property: in
R(A)
Proof.
for which Denote
least norm in
limlb
n➔ °'
d(0,R(A)) cl(R(A)).
n
I =
d(0,R(A))
by
d
and let
converges. -u
A c Xx X
every sequence
[b n J
be the unique point of
The proof of [8, Theorem 4] and Lemma 3 show
that to each b there corresponds a point z in J(-u) such that n n lim (b , z ) = d2, (bn,zn) :.,; d2. Since lim i b I = d, we have, in fact, 11 .. °' n n n➔ °' n and the conclusion follows by Lemma 12. This lemma improves upon Lemma 3.2 in [11] where
X
was assumed to be
200
Reich and Shafrir
uniformly convex and
A
is assumed to satisfy the stronger range con-
dition corresponding to (4).
It can be shown that the converse of
Lemma 13 is also valid. We can now establish our convergence result.
Such a result has been
known so far only in uniformly convex spaces [11, Theorem 3.7(c)].
It
sharpens the conclusion of Theorem 11. Theorem 14. T: D
➔
X
Let
D
be a closed subset of a Banach space
a nonexpansive mapping which satisfies (3).
X
and
Assume that
(en} is bounded away from O and 1, and let the sequence [yn} be defined by (2). If x* is (LUF), then the strong lim(y -Ty) exists n➔ m n n and coincides with the unique point of least norm in cl(R(I-T)). Proof.
[y n -Ty n JC R(I-T)
Since
limly n -Ty n I
and
limly +l-y l/c n n n by Theorem 8, the result follows by Lemma 13.
= d(O,R(I-T))
x*
It can be shown that if 14 is no longer valid.
n➔m
n➔m
is not (F), then the conclusion of Theorem
We do not know, however, if this differentia-
bility condition is sufficient for Theorem 14 to hold. We shall say that the norm of a Banach space
X
with a unit sphere
is locally uniformly Gateaux differentiable (LUG) if for each and each face x
in
F.
F
of
U
y Eu
the limit (7) is attained uniformly for all
x*
It can be shown that if, in the setting of Theorem 14,
is (LUG), then the weak
U
lim(y -Ty) exists. We expect to present a n n more complete discussion of the differentiability properties (LUF) and n➔ m
(LUG), as well as their consequences, elsewhere. We t~rn now to another consequence of Theorem 8. Recall that a closed convex subset C of a Banach space has the fixed point property for nonexpansive mappings (FPP for short) if every nonexpansive
T: C
➔
C
has a fixed point [6, 7, 9]. Theorem 15. and
T: C
away from
Let C
➔
O
C
be a closed convex subset of a Banach space
a nonexpansive mapping. and
1,
Assume that
If each bounded closed convex subset of fixed point free if and only if Proof.
[yn]
and let the sequence limly
n➔ m
n
I
X
X
[c] is bounded n be defined by (2).
has the FPP, then
T
is
m
(yn} of (yn} is bounded, we k let R = lim sup!y 0 -y I and B = [x EC: limk➔ ~uplx-ynkl $ R}. It is k➔ m nk ,_ clear that B is a nonempty, bounded closed convex subset of C. Since
Assuming that a subsequence
lim(y
n➔ m
nk
-Ty
nk
)
=
0
by Theorem 8, it is also invariant under
T.
Successive Approximations for Nonexpansive Mappings
201
Hence the result. It can also be shown that if, in the setting of Theorem 15, uniformly convex and (F), and lim yn
n-->co
T
X
is
has a fixed point, then the weak
exists and is a fixed point of
T.
In this case, however, a
better result is already known [10, Theorem 2]. Finally, we note that Theorems 5, 8 and 15 can be shown to carry over to these self-mappings of the Hilbert ball which are nonexpansive with resepct to the hyperbolic metric [5].
These results extend several
theorems in [13] and provide an affirmative answer to the question raised at the end of that paper.
It is expected that a complete dis-
cussion, as well as related results on implicit iterations, will be presented elsewhere. REFERENCES 1.
J. B. Baillon, R. E. Bruck and S. Reich, On the asymptotic behavior of nonexpansive mappings and semigroups in Banach spaces, Houston J. Math. 4(1978), 1-9.
2.
K. Fan and I. Glicksberg, Some geometric properties of the spheres in a normed linear space, Duke Math. J. 25(1958), 553-568.
3.
T. Fujihira, Asymptotic behavior of nonexpansive mappings and some geometric properties in Banach spaces, Tokyo J. Math. 7(1984), 119-128.
4.
K. Goebel and W. A. Kirk, Iteration processes for nonexpansive mappings, Contemporary Math. 21(1983), 115-123.
5.
K. Goebel and S. Reich, "Uniform Convexity, Hyperbolic Geometry, and Nonexpansive Mappings," Marcel Dekker, New York and Basel, 1984.
6.
W. A. Kirk, Fixed point theory for nonexpansive mappings, Lecture Notes in Math., Vol. 886, Springer, Berlin and New York, 1981, pp. 484-505.
7.
W. A. Kirk, Fixed point theory for nonexpansive mappings II, Contemporary Math. 18(1983), 121-140.
8.
A. T. Plant and S. Reich, The asymptotics of nonexpansive iterations, J. Functional Anal. 54(1983), 308-319. s. Reich, The fixed point property for nonexpansive mappings, I, II, Arner. Math. Monthly 83(1976), 266-268; 87(1980), 292-294.
9. 10.
S. Reich, Weak convergence theorems for nonexpansive mappings in Banach spaces, J. Math. Anal. Appl. 67(1979), 274-276.
11.
S. Reich, On the asymptotic behavior of nonlinear semigroups and the range of accretive operators, J. Math. Anal. Appl. 79(1981), 113-126.
12.
S. Reich, On the asymptotic behavior of nonlinear semigroups and the range of accretive operators II, Mathematics Research Center Report #2198, 1981; J. Math. Anal. Appl. 87(1982), 134-146.
13.
S. Reich, Averaged mappings in the Hilbert ball, J. Math. Anal. Appl. 109(1985), 199-206.
Quasilinear Ellipticity on the N-Torus VICTOR L. SHAPIRO, Department of Mathematics and Computer Science, University of California, Riverside, Riverside, California
1.
Introduction.
In this paper, we intend to establish one new result and
extend two previous results [3, Thins. 1 & 2], for second order quasilinear elliptic partial differential equations defined on the N-torus.
The
main point of the extension is that now the leading order coefficients
aii(x,u) are allowed to have also a linear growth condition with respect to u, i.e., laii(x, u)I ~ a(x) + 11lul where a(x) E L 2 (0) and f7 is a positive constant. Here, O={x:-1r 0 such that lii(x,z)l::;a(x)+11lzl for every zER and a.e. xEO, i,i=l, ... ,N. (Q-3) Q is symmetric; that is, aii(x,z) = aii(x,z) for every z ER and a.e. x E 0, i,J· = 1, ... , N. 203
204
Shapiro
(Q-4) Q is uniformly elliptic almost everywhere in O; that is, there is a constant r,0 > 0 such that
e
et>.
for every z ER, a.e. x E O, and every E RN Clel 2 =er+···+ (Q-6) There is a nonnegat,i.ve :function b(x) E L 2 (0) and a positive constants tJ1 and tJ2 such that
:for every pERN, zER, and a.e. zEO, j=l, ... ,N. (Q-6) For every u E W 1 •2 (0), the vector b(z,u,Du) = [b1(z,u,Du),
... ,bN{z,u,Du)] is weakly solenoidal, i.e.,
f0
bi(x,u,Du)D;v(x)dx = 0 :for
every u and v E W 1•2 (0) where the summation convention is used.
(Q-7) If {un}~=l is a sequence of functions in L2 (0) which tend strongly
to u E L2 {0) and {wn}~=l is a sequence of vector-valued :functions which
tend weakly tow E [L2 {0)]N, then {b(z,un,wn)}~=l tends weakly to b(z,u,w)
E [L 2 {0)]N, i.e., with bi(z,u,w) b{z,u,w), then lim j=l, ... ,N.
n--oo
f0
= bi(x,u,w1, .. ,,wn) f0
bi(x,un,wn)vdx =
the j-th component of
bi(x,u,w)vdx V v E L 2 {0) and
By strong convergence in (Q-7), we mean convergence in norm. In this paper, we intend to establish the following new result.
Assume (Q-1) - (Q-7) and let sufficient condition that the equation THEOREM 1.
Qu=
f
E L 2 (0). Then a neceuary and
f
{1.2)
has a distribution solution u E W 1 •2 ( 0) is that fO f dx = 0. If Qu = -~u where ~ is the Laplacian, the above theorem is well known. It is also well known if bi = 0, j = 1, ... , N and aii does not depend on u and is smooth as a :function of z.
(See [2, p. 674] .)
However
in the case considered here where the aii(x,u) are unbounded in:,; and possibly grow linearly in u, the result obtained in Theorem 1 is new. To be quite explicit, what we mean by u E W 1 •2 (0) being a distribution solution of Qu = f is the following:
(1.3)
205
Quasilinear Ellipticity on the N-Torus
for every ¢> E 0 00 (0), where the summation convention is used for i,j
l, ... ,N. 2.
We first dea.l with the sufficiency condition of
Proof of Theorem 1.
the theorem a.nd establish this pa.rt via. a. Ga.lerkin argument. to accomplish this, we observe tha.t there is a. sequence va.lued functions in 0 (a.) (b)
00
In order
{1Pk}k=l
of rea.1-
(0) with the following properties:
= (2,r)-N/2; J0 1Pk1Ptdx = Dkt
1/11
Also, given
,p E 0
00
where Dkt is the Kronecker-5 k, l
= l, 2, ....
(2.1)
(0) a.nd t > 0, :3 constants c1, ... , Cn such tha.t
lt/J(x) - Cq,Pq(x)I < t uniformly for x E O a.nd j in (2. 2) with q
=
= l, ... , n.
ID;,f;(x) - cqDjt/,q(x)I < t
a.nd
= 1, ... , N
(2.2)
where the summation convention is used
Here, a.nd in the sequel throughout the rest of the pa.per, we sha.11 use the notation
=
(v, w)o for v a.nd w in L 2 (0).
lo
v(x)w(x)dx
In ca.se v and w a.re a.lso in W 112 (0), we sha.11 set
(v, w)i
= (v, w)o + (D;v, D;w)
where the summation convention is used for j set
(2.3)
llvllo = (v, v)g'
and
=
1, ... , N.
(2.4) Also we sha.11
llvlli = (v, v)f.
(2.5)
H~ sha.11 designate the Hilbert space spanned by {1Pk}r'= 2 using the Thus w E H~ means :l{wn}~=l such tha.t llw - wnll1 -+ 0 a.s
W 112 (0)-norm.
n-+ oo where ea.ch Wn is a. finite linear combination of elements of the
sequence
,p2,,f;3, ... ,,pk,··· .
lows tha.t v E
W 112 (0)
As a. consequence of (2.1) and (2.2), it fol-
ca.n be written uniquely in the form V
= C11/}1 + W.
where w EH~ a.nd c1 is a. constant.
(2.6)
Also, it is clear tha.t
(2.7)
206
Shapiro
Furthermore, it is easy to see from (2.1)(a), (2.4), (2.6), and (2.7) and elementary Fourier analysis that
llwll~ $
(Diw, Diw}o
(2.8)
V wEH~
where the summation convention is used for j = 1, ... , N. The first lemma we state is the following. LEMMA 1.
Let f( x) be the function in L 2 ( 0) given in the statement of Theorem
1, and assume that Q satisfies (Q-1) - (Q-6). ~
2, there is a function un
= 72'¢;2 + · · · + "f~"Pn
Then if n is a given positive integer such that
(ii(·, un)Diun, Di'I/Jk}o + (bi(·, Un, Dun)Diun, "Pk}o + (un, "Pk}on- 1 = (1/Jk,J)o
(2.9)
fork= 2, .. . ,n where the summation convention is used for i,j = 1, ... ,N and ,; are constants q = 2, ... , n. The proof of this lemma is essentially the same as the proof given in [3, Lemma 1], and the reader should have no difficulty in filling in the details. Next for n fixed and~ 2, we multiply both sides of (2.9) by 'k for k = 2, ... , n and sum on k. Observing that Di(un) 2 = 2un D;un and using (Q-6), we see that (aii(.,un)D;un,Diun)o
+ (un,un)on-1 = (un,/}o.
We conclude from (Q-4) and this last fact that
(2.10) where the summation convention is used for j = l, ... ,N. Since un is in 1. H-, it follows from (2.8) that llunllo$(D;un,D;un)g. We therefore obtain from (2.4), (2.6), and (2.10) that for n = 2,3, ....
{2.11)
From this last fact, it follows that there is a subsequence of {un}:= 2 • which for ease of notation we take to be the. full sequence, with the following properties:
3u E H- such that lim un(x) = u(x)
n-+oo
lim
n- 00
[ wD;undx
lo
= [
lo
lim llun - ullo = O;
(2.12)
for a. e. x E O;
(2.13)
n-+oo
wD;udx V w E L 2 (0),j
= 1, ... , N.
(2.14)
Quasilinear Ellipticity on the N-Torus
207
Using (2.11) - (2.14), in conjunction with (Q-6) - (Q-7), it follows exactly as in the proof [3, Thm. 1] that
fork= 2,3, .... Next, we fix i and j and set (2.16) and observe from (2.13) and (Q-1) that for a.e. zEO.
(2.17)
With a(z) and~ as in (Q-2), we set On(z) = [2a(z} + ~lunl + ~(u}] 2 and observe from (2.13) that lim On(z) = 4[a(z) +~u(z)]2 a.e. inn. n-oo we have that for a.e. zEO and n=2,3, ...
(2.18} Also
(2.19)
Since a(z) E L 2 (0}, it follows from (2 .12) that lim
n - 00
f
}0
On(z)dz = 4
2 Jof [a(z) + ~juj dz.
(2.20)
We invoke [2, Thm. 16, p. 89] with hn replacing fn and conclude from (2.16) - (2.20) that lim J0 hn(z)dz = 0. Now i and j were fixed but arbitrary; n-oo so we record this last fact, using (2.16), as follows: (2.21) for i,j= 1, ... ,N. But then using (2.21) and proceeding exactly as in the proof of [3, Thm. 1], we see that (2.22) fork= 2,3, ...
208
Shapiro
Next, we see from (2.11) that lim {un,ifak)on- 1 = 0 fork= 2,3, .... n-+oo We consequently obtain from (2.9) on passing to the limit as n--+ oo and using (2.16) and (2.22) that
(ii(·, u)D;u, D;ifak)o + {b'°(-, u, Du)D;u, 1Pk)o = {ifak,J)o
(2.23)
for k = 2, 3,... . Now using (Q-6) and (2 .1) (a), we see that the lefthand side of (2.23) is zero when k = 1. Likewise from the hypothesis of the theorem, we see the right-hand side of (2.23) is zero when k = 1. We conclude that (2.23) also holds fork= 1. But then it follows from (2.23) and the uniform convergence aspect of (2.2) that
for all JR 1
Suppose that
of Banach spaces.
Ex F
is a
c 1 -function de-
of two closed convex sets
We say that
f
E
and
satisfies the twist Palais Smale
condition (T.P.S), if { (xn,yn)} c Ex F,
any sequence (T.P.S.)
= Sup f(x ,y)
f(x ,y) n n
yEF
n
is bounded and
n
inf ( f' (x , y ) , -11 h II) hEE-x x n n h n
along which
-;>
a. .: O,
possesses a
convergent subsequence. Remark 1.1. F
F
is a compact set, then any function
satisfies (G.P.S.).
tion
f
defined on
Remark 1.2. g
If
If
F
If
Ex
E
F,
and
F
satisfies (G.P.S.).
If
defined on
are compact sets, then any func-
satisfies (T.P.S.).
is a Banach space, and if
satisfies P.S. over
g
E
Ex F,
and then
F f
g
satisfies P.S., then
are Banach spaces, and if
f
satisfies (T.P.S.).
The main result of this paper is the following Theorem A. let
f E c1 (1)
Let
E, F
be two closed convex sets of Banach spaces, and
(EX F,lR 1 )
V XE E,
be a function satisfying
y I--> -f(x,y)
is bounded below and quasi-convex,
and satisfies (G.P.S.), (2) (3)
3 y0 E F
Then there exists (1)
(2)
such that
xi--> f(x,y 0 )
(T.P.S.) holds for (x,y)
is bounded below,
f.
E E
X
F
such that
min max f (x,y), = max t(x,y) xEE yEF yEF V h E E-x. (f~ (x,y) ,h) ~ 0
f (x,y)
Moreover, if
X 1------;l>
f (x,y)
there is a neighborhood
is locally pseudo-convex at
U(x)
of
x
such that
x,
i.e.,
A Local Minimax Theorem Without Compactness Y x E U (x) ~ f (x,y) ;:: f (x,y)
(f' (x,y) ,x-x) ;:: 0 X
then
(x,y)
213
is a saddle point of
f,
V x E U (x)
i.e.,
f (x,y) :,; f (x,y) :,; f (x,y) V x E U(x),
Vy E F.
Remark 1.3.
It is well known that the crucial point in proving von
Neumann-Sion Minimax Theorem, is to prove the equality: min max f(x,y) = max min f(x,y) xEE yEF yEF xEE which depends heavily on the convexity hypothesis of the functions
Vy E F.
x I--> f(x,y) point
of
view
of
However, our approach, which starts from the
critical point
theory,
concerns merely
the
local
behavior near the "saddle point," so that the convexity hypothesis is not necessary. The proof of Theorem A is based upon the following two fundamental principles: (Ekeland's Variational Princiele [ 7]) • Let E be a complete metric Suppose G:E -a> JRl u [ +a, J be 1. s. c. with G(x) F +"' + that XO E E and € > 0 are given such that
space, and let
G(x 0 ) :,; inf G(x) + €. xEE Then
V k > O,
there exists
x€k EE
such that
G(x 8 k) :;;G(x 0 ), 1 d(x0,x€k):,; k,
G(x) > G(x€k) - ked(x,x~k) where
V x /
xek'
m. 1
is the metric of E. + This principle is very useful in many problems. d : E X E -,:,
It has been appli-
ed to give a new proof of the Mountain Pass Lemma (cf. [2] and [12]), and will be applied time to time in this paper. (Loe-sided Minimax Theorem).
be two convex sets of Hausdorff linear topological spaces, and let f: Ex F -,:, :JR 1 . Suppose that (1)
Let
E, F
V x E E,y I--> f(x,y)
is u.s.c. and quasi-concave,
(2)
Y y E F,x I--> f(x,y)
is 1.s.c. and quasi-convex,
(3)
3 xC E E
f(x 0 ,y) :::
A]
and
is compact.
Then there exists
y E F
A. < inf sup f (x,y)
xEE yEF
such that
such that the set
(y E F
I
214
Shi and Chang
inf sup f(x,y) = sup inf f(x,y). xEE yEF yEF xEE
inf f(x,y) xEE
The reader is referred to [2], [3] and [4]. The paper is organized as follows: given in §2.
The proof of Theorem A will be
Some extensions, which need weaker smooth conditions on
the function will be discussed in §3, in which the locally Lipschitzian f are used to replace the c 1 -condition.
condition for the function
§4 deals with applications, in which a variational inequality problem, an elliptic system and an existence theorem of nonlinear programming are studied.
2.
Proof of Theorem A. The proof is divided into several steps.
In order to make it easy
to understand, we start with a slightly abstract version. Theorem 2.1. and let
Let E, F be two closed convex sets of Banach spaces, f E c1 (Ex F, JR 1 ). Suppose that
(1)
V x E E,y I--> f(x,y)
(2)
3 y 0 E F,
(3) (4)
is bounded above and quasi-concave,
such that
(T.P.S.) holds for
V x EE,
x
f(x,y 0 )
is bounded below,
the set
M(x) = {y E F
J
is compact and nonempty, (5) V x EE, V h E E-x, there exist y n E M(x+tnh) , mulate point y E M(x). Then there exists a pair (1)
f (x,y)
(2)
(f' (x,y) ,h)
=
1-->
f, f(x,y)
sup f (x,y)] yEF
V positive sequence
n = 1, 2, . . .
such that
(x,y) E E X F such that = min max f(x,y), xEE yEF v h E E-x. 0
max f(x,y) yEF
X
Proof of Theorem 2.1.
;::
Let G(x)
sup f(x,y). yEF
It is easily seen from the assumptions (1) and (2) that is l.s.c. and bounded below. Principle,
v
€
with tn I 0, has an accu-
> 0,
According to the Ekeland's Variational
3 x 8 EE G(xe)
G: E -~ m. 1
such that ~
inf G(x) + e, xEE
(2 .1)
215
A Local Minimax Theorem Without Compactness For each
h E E-xe,
we have G(x +th)-G(x) e t e .: -eilhll-
G+I (x ;h) ~ lim e l-->+O Since for any positive sequence
[tn]
with
tn" 0,
( 2. 2)
we have
~ lim tl [G(x e +t n h)-G(x e )], n-->"" n
G+' (x ;h) e
~ 1 im tl [ f ( x e + t n h, y n ) - f ( x e , y n ) ] n-->"" n lim (f' (x +t e h,y ) ,h) x e n n n for some
en E (0,1),
provided by the assumption (4), with
yn E
M(xe +tnh). According to the assumption (5), we may so choose {y] 1 . 1: that [y ] has an accumulate point yn E M(xe). The C -continuity of n f implies that ( 2. 3)
Combining (2.2) with (2.3), it follows 'th E E-x .
(2. 4)
€
P.: : (E-xe) x M (xe) -;;:,. lR 1
Let us define a function
by
(f' (x ,y) ,h) + ellhll x e
Pe (h,y) We shall verify that (1) (2) ( 3)
V y E M(xe), h V h E E-x e, y
is continuous and convex,
1--->
Pe (h,y)
is continuous and quasi-concave,
M(xe)
y 1--7 f(x,y).
function
P .: (h,y)
is compact and convex.
M(xe)
The convexity of tion
1--->
It remains to prove the quasi-concavity of the
y 1--> Pe (h,y);
In fact,
follows from the quasi-concavity of the funcbecause all other items are easily seen.
V y 1 ,y 2 E M(xe),
and
Vt.. E [O,l],
(f~(xe, (1-t..)y 1 +t..y 2 ) ,h) 1 lim t[f(xs:+th, (1-t..)y 1 +t..y 2 )-f(xe' (l-t..)y 1 +t..y 2 )]
uo
.: lim min[½[f(xe+th,y 1 )-G(x€)], t[f(xe+th,y 2 )-G(xs:)]]
uo
min[ (f~ (xs: ,y 1 ) ,h), (f~ (x,y 2 ) ,h)], i.e., the function Now
Pe
y
t-->
PE:(h,y)
is quasi-concave.
satisfies all conditions of the Lop-sided Minimax Theorem,
we conclude that
j
ys: E M(xs:)
such that
216
Shi and Chang inf hEE-x
P E:
E:
(h, y ) = inf E: hEE-x
sup ( (f' (x ,y) ,h) + E: ilhl!} ;;,: 0 yEM(x) x E: E:
E:
provided by (2,4), that is, V h E E-x . E:
( 2. 5)
By definition,
(2.1) reads as inf sup f(x,y) xEE yEF
infG(x) :.G(xE:) xEE
=
f(xE:,yE:) s;
Let
{E:n]
Ex F,
inf sup f(x,y) xEE yEF
(2. 6)
E: j O. Then, the sequence n possesses a convergent subsequence (xE:.'YE:.l -» (x,y)
n
provided by (T.P.S.).
(f' (x,y) ,h) ;;,: 0 X
(x,y)
(2. 5).
J
E
min max f(x,y) xEE yEF
yEF
by (2.6); and
J
Thus we have
= max f(x,y)
f(x,y)
by
+ E:.
be a positive sequence with
l (XE: , y E: ) } n
sup f(xE:,y) yEF
V h E E-x
is the solution of our theorem.
Now we turn to the proof of Theorem A.
It remains to prove that
(G.P.S.) of the functions y I----> -f(x,y)
(4) and (5) of theorem 2.1.
implies the assumptions Lemma 2.2. g E c 1 (E,JR 1 )
Let
E
be a closed convex set of a Banach space, and let
be a function bounded below and satisfying (G.P.S.).
Then the set M
=
rx
E E
I g {x)
is nonempty and compact. If
OM
V x EE
is a neighborhood of
M,
inf g (x)] xEE with boundary
oOM,
g(x) > inf g(x). inf xEE xEEfloOM Proof.
(2. 7)
The first assertion follows directly from the Ekeland's Varia-
tional Principle.
In fact,
VE:> 0 3 x
g(x) 8
and
then
s;
E:
EE
inf g(x) + E: xEE
such that
A Local Minimax Theorem Without Compactness
i.e.
217
I
Inf (g' (x ) ,____b__) hEE-x e I hll
:?.
-e.
e
[x
We obtain a convergent subsequence Therefore
e.
} -a> x,
provided by (G.P.S.).
l
x EM.
Again, by (G.P.S.), the set
M
is compact.
Denote (2. 8)
which is positive because
M
is compact.
We shall prove the second assertion by contradiction. were
[xn} c
En
oOM
If there
such that g(xn) -a> inf g(x), xEE
then we would have
z
n
EE
such that
g(zn) s g(x ) < inf g(x) + e, n xEE
for all
e > g (xn) - inf g (x) xEE
:?.
0,
and which imply
Again, by (G.P.S.), -,, x.
Therefore
[ z
n
}
possesses a convergent subsequence g (x)
i.e.,
xEM,
but
. nJ
inf g (x) , xEM
dist(x,oOM) s llx-z s
z
½+
-11
nJ
+ llz
.-x -II nJ nJ
llx-znjll _,,
½,
which contradicts with (2.8). Remark 2.1.
In lemma 2.2, the same conclusion holds true, if the con-
dition (G.P.S.) is weakened as follows: any sequence (G.P .S.) *
[x] n
CE,
along which
inf (g' (x ) , II hhll) hEE-x n n a convergent subsequence.
and
g(xn) -a> inf g(x) xEE possesses
It is interesting to note that (G.P.S.)* is also a necessary condition for the conclusion.
In fact, if
[x] n
c E
is a sequence such that
218
Shi and Chang g(xn} -;;,, inf g(x) xEE
h
and
inf (g' (xn), -llhll) -;;,, i:l hEE-x n
Then we have
;;e
0.
dist(xn,M) -;;,, 0 provided by (2. 7). we have
[x~]
x~ E M
II xn -x~II
0
such that (2. 9)
y 0 E M(x 0 ) arbitrarily, and connecting y 0 by a segment, the segment must intersect with the set oO, Choosing
3 An E (0,1)
with i.e.,
such that
=
zn
(1-An)y 0 + AnYn E oO,
n = 1,2, • · ·.
On one hand, by lemma 2.2, we have sup f(x 0 ,zn) n
sup f(x 0 ,y)
~
oO
< sup f(x 0 ,y) yEF
(2
.10)
On the other hand, f(x ,z) n n
2:
min[f(x ,y ),f(x ,y 0 )J = f(xn,y 0 ), n n n
provided by the quasi-convexity of the function For large
y
I---?>
-f(x,y).
n,
We obtain, lim f(x 0 ,zn)
n-tQ;I
2:
lim f(xn,Yol
n-tro
which contradicts with (2.10). Proof of Theorem A. We mentioned before that we only need to verify the assumptions (4) and (5) of theorem 2.1. Actually, assumption (4) was verified in lemma 2.4. According to lemma 2.4, we know that
V h E E-x,
V positive sequence
choose arbitrarily point
y
E M(x),
Theorem A
yn E M(xn).
x t--> M(x)
is u.s.c.
Now,
{tnJ
with
Then
{ynJ possesses an accumulation
provided by lemma 2.3.
tn I 0,
xn
x+tnh,
we
Assumption (5) is verified.
follows directly from theorem 2.1.
220 3.
Shi and Chang Extensions and Remarks. It is our purpose of this section to weaken the smoothness condi-
tion of the function Theorem A.
f
and the convexity condition of the set
E
of
In addition, we shall discuss how important the convexity
of the function
x I--> f(x,y)
plays a role in von Neumann-Sion-Ky Fan
Theorem. (I) Locally Lip. function. Carefully analyzing the proof of Theorem A, the the function
f
c 1 -condition of
was used in the following points:
(1)
A formulation of the P.S. condition.
(2)
The inequality (2.3), where a mean value property and the
. . C1 -continuity are use d .
(3)
A locally Lipschitzian property in (2.9).
These inspire us to generalize our results to loc. Lip. functions. Let us recall the Clarke directional derivative for Loe. lip. functions [6]. (x 0 ,y 0 ) EE
For each
x F,
(h,l)
~
E (E-x 0 )
x (F-y 0 ),
we define
= lim f[f(x'+th,y' )-f(x' ,y' )]
f~(x 0 ,y 0 ;h)
tlO
(x' ,y') ~ (xo,Yo) fO(x 0 ,y 0 ;1} = lim t[f(x' ,y'+tl)-f(x 1 ,y' )] y tlO (x'
,y') _,,, (x 0 ,y 0 )
to be the Clarke partial directional derivatives. f
is said to be regular in
at
x0,
if
= lim f[f(x 0 +th,y 0 )-f(x 0 ,y 0 )]
f~x(x 0 ,y 0 ;h) exists for each
x
uo
h E E-x 0 ,
and
f~x(x 0 ,y 0 ;h) Employing these notions, we use
0 -f y (x , y·l) '
(or
0
fx(x,y;h))
(-f' (x,y) ,1) (or (f' (x,y) ,h) resp.) in the definition of y X (f (x,y)). ((T.P.S.) resp.) of the function y 1--> -f(x,y) In the deduction of (2.3), we carry out as follows lim tl [f(x +t h,y )-f(x ,y )] G+' (x , h) ~ e:n n e:n e:
n...,"'
~
n
fox(xe:,yh;h). -
to replace (G.P.S.)
221
A Local Minimax Theorem Without Compactness In order to obtain the quasi-concavity of the function
0
fx(x 8 ,y;h),
we shall assume that the function
f(x,y)
y
i----,,.
is regular in
x. In summary, Theorem A holds true for loc. Lip. functions is regular in E
(II)
f, which
x.
without convexity.
The convexity of
E
was used only in defining the directional de-
rivative of the function
f.
Let us recall the notions of the contin-
gent cone and the Clarke tangential cone of a nonempty subset Banach space
X
[2].
V x O EE,
E
of a
let such that
x 0 +t h
EE]
[h EX IV xn-,.. x 0 ,xn EE, V tn •+O, 3 hn ---,,>h,
such
n n
and
that
x
n
+ t h E EJ• n n
They are called the contingent cone and the Clarke tangential cone of E
at
respectively. By definition,
CE(x 0 )
is convex, and both
say that
E
TE(x 0 )
and
CE(x 0 ) c TE(x 0 ). If CE(x 0 ) = TE(x 0 ), is regular at x 0 in the Clarke sense. It is easily seen
are closed cones with that
U .!. (E-x 0 )
V x 0 E E,
A.>0 II.
and
CE (x 0 ) = TE (x 0 ) = the tangent space of Let g
at
at
E
1 g : E ------:;, .IR , and let x 0 E E. along a direction h E TE(x 0 )
c;i:_'+E (xO; h)
=
if
E
x 0 , if
is convex,
is
E
at
The contingent derivative of is defined to be
½[g (x 0 +th ')-g (x 0 )].
lim
h' -+h
x 0 +th' EE t!O Similarly, the Clarke directional derivative for a loc. Lip. function And
g
g
x O,
lim
h1 ➔h
x O+th' EE t ♦ O
exists with
0
(x O;h)
is defined along directions
is said to be regular at g'+E(x O ;h) =
g
if
E
xO
h E CE(x O).
is regular at
t[g(x O+th')-g(x O)]
at x O,
and
Shi and Chang
222
Employing these notions, we use f~x(xn,yn;ll~IT) -~a~ 0
CE(x)
to replace
E-x
and
inf (f~ (x ,y ) , hEE-x X n n
to replace
n
h
llhlf ~a~ 0
in the definition of (T.P.S.), Theorem A holds true
also. In summary, Theorem A holds true for any closed set convex sets x EE
F,
on which
in direction
f(x,y)
E
and closed
is loc. Lip. and regular for each
x.
(III) Further discussion. We shall emphasize the main difficult point of Theorem A being the lack of convexity
of the function
x
f-->
f(x,y).
Provided by the Lop-Sided Minimax Theorem via weak topology we have the following Theorem 3.1.
Suppose that
reflexive Banach spaces. (1)
'
E
and
F
are two closed convex sets of
We assume that f(x,y)
is l.s.c. and quasi-convex,
3 Yo E F such that the function below and coercive, (2)
(3)
'
-f(x,y)
below and coercive. Then there exists a saddle point f(x,y)
~
f(x,y)
~
X 1-->
f(x,yo)
is bounded
is 1. s .c. and quasi-convex, y
I--,>
(x,y) EE x F
f(x,y)
'
!Ix II
as
=
[x EE
I g(x) =
For
b E :JR1 ,
is nonempty and compact. gb and c
=
C
< +°',
=
{x E E
sup{b E
We shall prove that
°',
x E E.
then
c
=
]R
1
inf g(x)} xEE define the level set
I g (x) gb
+°',
s: b},
is bounded}.
so that
g
is coercive.
is a bounded set, but
g
g
-n c-2 g -n c BR, an open ball with radius c-2 n
This implies
g(x) > c-2-n
V x E E\BR
In fact,
-n
c+2 Rn;;,: n,
is not. i.e.,
n
-n with llx n -n is unbounded, there is xn E g c+2 c+2 Applying Ekeland's Variational Principle to the function g
Since Rn+2.
-.-
According to lemma 2.2, the set M
if not,
+°',
g
the metric space
E\BR, n
we obtain
x~ E E\BR, n
II> on
such that
c-2-n < g(x') s:g(x) s: c+2-n, n n
( 3. 2) ( 3. 3)
and V
X
,f
( 3. 4)
x E E\BR. n
The inequalities (3.2) and (3.4) imply , ( , 9'..+E xn,
Nh )
;;,:
2
1-n
,
Therefore there is a convergent subsequence of (G.P.S. )'.
However, from (3.3),
It is impossible. Corollary 3.1. x ~ f(x,y 0 )
(x' J n
provided by
The contradiction proves the theorem.
Theorem 3.1 holds, if the coerciveness of the functions and
y I--> -f(x 0 ,y)
is replaced by (G.P.S.)'.
We compare Theorem A with Corollary 3.1.
The conditions (1) and
(2) of Corollary 3.1 are weaker than the conditions (1) and (2) of Theorem A.
Since now the convexity of the function
y
~
not assumed, we need a more complicated condition (T.P.S.).
-f(x 0 ,y)
is
224
Shi and Chang
(IV) An improvement of Theorem A. Employing the contingent derivative of a function, we have generalized (G.P.S.) to (G.P.S.)'. A can be weakened by
Now, we mention that (T.P.S.) in Theorem
(G.P.S.)' of the function
G(x)
=
One can verify it directly, the proof is even shorter.
sup f(x,y). yEF
We point out here that (G.P.S.)' plus assumption (1) of Theorem A imply (T.P.S.).
In fact, if
( (x ,y l} CE X F, n n
along which
is bounded and inf G'+E (xn, h) /[[ h [[ ~ a. "" 0, hEE-xn then we conclude that there is a convergent subsequence via (G.P.S.)'.
According to lemma 2.4 and lemma 2.3,
accumulate point, provided by
ynj E M(xnj).
x . ~ x* {yn;J
has an
We arrive at (T.P.S.).
4. Applications In this section, we present three applications of Theorem A: (1)
A variational inequality,
(2)
An elliptic system BVP,
(3)
Infinite dimensional nonlinear programming.
We shall give some new existence theorems, in which the solutions are saddle points rather than loc. minima of functionals. (I) A variational inequality. in
Let M, N be two bounded closed convex sets containing the origin JRm and JRn respectively. Let O CJRP be a bounded open set.
Assume that 1
2
m
1
m
u = (u (x),u (x), •.. ,u (x)) E H0 (0,JR ), 1 2 m 1 n (v (x),v (x), ... ,v (x)) E H0 (0,JR );
v and denote
Suppose that
E
{u E
F
[v E
H; H~
g E C1 (M
(0,JRm)
u(x) E M a.e.},
(O,JRn)
v(x) E N
X
N,JR 1 ) ,
Vu EM, Theorem 4.1.
and that
v ~> g(u,v)
There exists a pair of solutions
ing variational inequalities:
a.e.}.
is concave. (u 0 ,v 0 )
( 4 .1)
of the follow-
A Local Minimax Theorem Without Compactness
V qi E Ht(O,JRm)
Proof.
225
~ [vu 0 vqi + g~ (u 0 ,v 0 )qi]dx ;;e 0,
(4.2)
J0
(4 . 3 J
with
[ v v Ov l/J - g ~ (u O, v O) l/J Jdx qi E E-u 0 ,
o,
;;e
V ljJ E Ht(O,JRn)
and
with
ljJ E F-v 0 .
Define a functional I (u,v) =
J [½l'vu 12 -
½/'vv / 2 + g (u,v) ]dx
0
on the closed convex set
Ex F.
We shall verify that
fies all assumptions of Theorem A. function
v t---> -I(u,v)
V u 0 EE;
[vk} c F
satis-
Actually, only the (G.P.S.) of the
and (T.P.S.) of
we consider the function
Assume that
I(u,v)
I
are needed to prove
v I--> I(u 0 ,v).
is a sequence, along which (4. 4)
is bounded, and ~ ['vvk'vlji- g~(u 0 ,vk)lji]dx
[vk]
From (4.4),
O (Ill/JIil,
1 n H0 (0,JR),
is bounded in
V ljJ E F-vk.
(4. 5)
so that
1 n (HO (0,JR ) ) ,
v*
~
;;e
a. e.,
for a subsequence.
Thus
v* E F.
t0
Now for each
E F-v*, we shall prove (4. 6)
In fact, let
l/Jk.
V
J
f
0
* + lj!O-vk.,
'vv*'vl)! 0 dx
~ 'vvkj vl)! 0 dx + 0 (1)
J 'vvk 0
;;e
and
then
J
j
J
'vljik dx+ 'vvk 'v' (vk -v*)dx+ 0 (1) j O j j
O(1) + Jvvk _vl)ik. dx, 0
J
(4. 7)
J
~ g~(u 0 ,vk/l)! 0 dx+ 0(1)
= provided by Lebesgue Theorem. obtain (4.6).
f
(,
g' (u 0 ,vk )ljik dx+ 0(1) V
j
j
Combining (4. 5) ,
(4. 8)
(4. 7) with ( 4. 8) , we
226
Shi and Chang Putting
in (4.5), and
~ = v*-vk
in (4.6), and adding
vk-v*
~O
these two i~equalities, we obtain
Thus
1 n (HO (0,JR ) ) .
-» v* This verifies
(G.P.S.) of
[ukJ c E,
Next, we assume
J[l
-I(u 0 ,v). {vk] c F
such that
E½lvukl 2 -½l'vvkl 2 +g(uk,vk)]dx
is bounded,
( 4. 9)
(4.10) and (4.11)
From (4.10), take
v 0 -vk
~
for any fixed
v 0 E F,
we obtain
1 n is bounded in H0 (0,JR). SubM1 , so that {vk] 1 m stituting this into (4.9), we see that [ uk} is bounded in H0 (0,JR ) •
for some constant
Then we have subsequences Uk.
J
uk.
-~
1 m HO (0,JR ) ,
u*
vk.
__,.,
V*
--,>
u*
J
u*
a.e.
vk.
J
1 n HO (0,JR ) , a.e.
J
Similarly, we prove
f[l and
['vu*vcp + g' (u* ,v*)cp]dx u
O
2:
V cp E E-u*
(4 .12)
u* EE. Combining (4.10) with (4.12), we have lluk.-u*ll 2 ,;; O(lluk.-u*II). J
Similarly, we have O(llvk. -v*II J
verified.
>.
Thus
( 4. 6) with * uk. -» u I J
* uo = u
I(u 0 ,v 0 ) and
--,>
vk.
J
Applying Theorem A, we obtain
=
J I
and then u*
.
sup I(u 0 ,v)
J
The (T.P.S.) condition is
(u 0 ,v 0 ) EE x F
vEF
llvk. -v*II 2 ,;;
such that (4.13)
(4 .14)
A Local Minimax Theorem Without Compactness
227
Now (4.14) is just (4.2), and (4.13) implies (4.3). Remark 4.1.
The same conclusion holds true, if
g
depends on
x E 0
with some dominant conditions, say, lg (x,u,v) U
g 0 E L1 (0),
Remark 4.2. and
Ig' I
g 0 (x)
I, lg'V (x,u,v) I :.
lg' (x,u,v) where
I :.
V (u,v) E M x N
g 1 (x)
V (u,v) E M x N
g 1 E L 2 (O).
and
The same conclusion holds if is bounded on
M, N
are closed convex sets
M x N.
II. An elliptic system.
Employing the same notations as before we are looking for a weak solution of the following elliptic system: {
Liu = g~ (u,v)
(4 .15)
Liv = -g~ (u,v),
i.e.,
J0 [vuvcp + g'u (u,v)cp]dx
J0 ['ii'v\i'ljJ- g'V (u,v)ljJ]dx Here we assume
g E c 1 (JR.m x JR.n,JR.l)
(a)
V u E JR.m,
(bl
3 constants
and
c 1 > O,
lg(u,O) 3 constant
I :.
I ~
0
1 n V ljJ E HO (0,JR. ) .
3 constants
By Young's inequality,
cl (l+lul 13 J
m
)
1
0
~
I
~
c 2 +g(u,v).
(c) implies a constant
~
C 3 ( l+
Iv I l +y
+
c3
such that
g,_(l+y)
Iu Iy C.
The system (4.16) has a weak solution
x Ho (0, R
n
.17)
13 < 1,
such that
Hereafter, we denote various constants by 1
(4
c 1 (l+lula+lvlal,
~ c 1 (l+lula+lvlYl, 1 c 2 > 0 and e E (0, 2 ) such that
I g ( u, v) I
Ho (G., R
.16)
such that
p-
' 1 gv ' ( u,v)v Sgu(u,v) · u+ 2
Theorem 4.2.
(4
satisfying
is concave, and a E (0 ,p+22)
y E [0,1)
jg'u (u,v) (d)
V cp E H~ (0,JR.m)
v I--> g (u,v)
lg~ (u,v)
(cl
0
) •
(u 0 ,v 0 J E
228
Shi and Chang
Proof.
We introduce the functional I (u,v)
Obviously,
v 1----7" I(u,v)
is quasi-concave, via (a), and is
bounded above via (4.17). It remains to verify the P.S. condition and (T.P.S.).
[vk]
assume that
c H~(O,JRn)
Firstly,
along which ( 4 .18)
and (4.19) is bounded, for a fixed
1 m u 0 E H0 (0,JR ) .
According to (4.17) and (4.19), fvk] is bounded in H~(O,JRn), and then by (4.18) and (c), there is a subsequence v ~ v* in 1 n kj H0 (0,JR ). Therefore v 1----.,> I(u 0 ,v) satisfies P.S. condition V u 0 E
1 m HO (0,JR ) .
Next, we turn to verify (T.P.S.). 1 n [vk ] c H0 (0,JR), along which
f
n
and
Assume that
c½l'vukl 2 -½ l
gi
i
) t- 0
(x)
=
0
and
j
to which
for all for these
and then
for infi-
J
= 0.
h. (x)
f (x N) +
)
provided by the compactness of Nk V j, i f h. (x ) = 0 Similarly,
So we shall restrict ourselves to those gi (x
Nk
O,
$
i
the compactness of nitely many
- x,
) - ~ g~ (x),
g~ (x
many
[xN]
f(xN)
Thus
k, i
and
Sim-
j.
Ng. (x N) + N
L..
i
gi (x )>O This implies g. (x
Nk
i
)
+
L
is bounded.
h. (x
N J h. (x k)=O J
Therefore g. (xl
lim gi (x k ..."'
h. (x) J
lim h. (x
i
and
Thus we proved
k....,.
Nk
)
0
)
0.
x EK.
Now we t urn t o prove t ha t quence.
J
Nk
(,N,µN) ~
possesses convergen t su b se-
Combining (4.25) with (4.26), we have N
f' (x k) +
L
iEI (x)
N
N
N
N
,.._ k.g'. (x k) + µ k.h' (x k) i
i
e.
According to the K.T. condition (assumption (1)), (g' (x
(4. 29)
Nk
) ,h. (x J
Nk
)]
232 i
Shi and Chang
E I (x),
j=l, ... ,q
are linearly independent for
Nk Ai
determinants do not vanish. (4. 28).
for
i
Therefore we can solve Nk is bounded, all Ai and
Since
E I (x).
ity, denote by Nk have x - ~ x.
large; their Gram Nk and µ by Nk µ are bounded
We obtain a convergent subsequence, no loss of generalN N (A k,µ k) --.,,. (f,µ). By the way, again via (4.28), we
Finally, we shall verify that fact,
k
(x,f,µ)
solves our problem.
In
from (4.24), we have f (xl + f • g (xl +
v·h (xl
f(x)+
sup
[A·g(x)+µ·h(x)]
(A,µ)EJR~xlRq f (xl
:. f(x) +
sup
[A·g(x) +µ•h(x)]
(A,µ) EJR~ XlRq ¥ x EX.
This implies f (x) :. f (x) ,
¥
E K,
X
and f•g(x) = O,
and then
figi(x) = 0,
i=l,2, ••• ,p.
The conclusion (2) follows from a limiting process of
(4.25).
The
proof is complete. As an example, we consider the following problem. Suppose that ···•~p'
OE lRn
n 1 , ... ,nq
is a bounded open domain, and that
are Caratheodory functions defined on
q:,,~ 1 ,
Ox JR 1 ,
sat-
isfying the following growth condition: Jg(x,t) where n
I :.
a. E (1 n+2) 'n-2 '
is a constant, and
C
= 1,2. Assume that
J
1
jS
q:,(x,t)dt,
f·
Q
J
t
n .: 3,
and
JSO
l
(x,t)dt,
n- (x,t)dt, J
i = 1, .•. , p; j = 1, •.. , q;
and
i=l, ... ,p; j=l, ..• ,q, 1 ElR. Let
f (u) =
0
r ~A. (X,$) =
if
(~.(x,t),n.(x,t)},
ear independent for each fixed
i (x, s)
C(l+JtJa.)
J ½Ivu I 2 + 0
J0 'i' l. ( x , u) dx ,
gi (u) h. (u) · J
!f ( x , u) dx ,
=
1 u E HO (0).
J
0
A . ( x , u) dx , J
a. .: 1
if
is lin-
233
A Local Minimax Theorem Without Compactness If
where with
x JRq
-a< A1 ,
c 1 is a constant, and 0-Dirichlet data in 0,
the first eigenvalue of
then there exists
-t:.
(u,X°,µ) E H~(O) X JR~
such that
(1)
f(u) = min[f(u)
Iu
t H~(O), gi(u)
j=l,2, ... ,q}, ( 2)
-t:.u(x)+qJ(x,u(x))+
( 3)
X°.f l [l
hj (x,u(x) )dx
(5)
J
Yi(x,u(x))dx s. 0.
Remark 4.3.
If
o,
i= 1,2, ... ,p;
r
0,
l
J0
ever, in case
0, hj(u)
tX-.\jJ.(x,u(x))+ µJ.T]J.(x,u(x)) i=l l l i=l Y. (x,u(x))dx = O,
(4)
0
s.
dim X < +w, dim X
+w,
O,
then Theorem 4.3 is obviously true.
How-
it seems worth proving. REFERENCES
1.
Aubin, J.P., Mathematical Methods of Game and Economic Theory, North-Holland, Amsterdam-New York-Oxford, Rev. ed. 1982.
2.
Aubin, J.P. and Ekeland, I., Applied Nonlinear Analysis, WileyInterscience, New York, 1984.
3.
Barbu, V. and Precupanu, Th., Convexity and Optimization in Banach Spaces, Sijthoff & Noordhoff, Bucharest, 1978.
4.
Brezis, H., Nirenberg, L. and Stampacchia, G., A remark on Ky Fan's minimax principle, Boll. Un. Math. Ital. 5(1973), 293-300.
5.
Chang, Kung-ching and Eells, J., Unstable minimal surface coboundaries, preprint, Univ. of Warwick, 1985. (To appear in Acta Math. Sinica.)
6.
Clarke, F.H., Optimization and Nonsmooth Analysis, Wiley-Interscience, New York, 1983.
7.
Ekeland, I., Nonconvex minimization problems, Bull. Amer. Math. Soc. (n.s.) 1(1979), 443-474.
8.
Fan, Ky, Minimax theorems, Proc. Nat. Acad. Sci. U.S.A. 39(1953), 42-47.
9.
Fan, Ky, A minimax inequality and applications, in "Inequalities III" (Shisha, 0., ed.), Academic Press, New York, 1972, 151-156.
10.
Li, Shujie, A multiple critical point theorem and its application in nonlinear partial differential equations (in Chinese), Acta Mathematica Scientia, 4(1984), 135-340.
11.
Nirenberg, L., Variational and topological methods in nonlinear problems, Bull. Amer. Math. Soc. (n.s.), 4 (1981), 267-302.
12.
Shi, Shuzhong, Ekeland's variational principle and the mountain pass lemma, Cahiers de CEREMADE, Univ. de Paris-Dauphine, n°8425. (To appear in Acta Math. Sinica.) Sion, M., On general minimax theorems, Pacific J. Math. 8(1958), 171-176.
13.
Theorems of Convex Sets Related to Fixed-Point Theorems MAU-HSIANG SHIH Department of Mathematics, Chung Yuan University, ChungLi, Taiwan, Republic of China KOK-KEONG TAN Department of Mathematics, Statistics and Computing Science, Dalhousie University, Halifax, Nova Scotia, Canada
l.
A Conman Generalization of Theorems of Knaster-Kuratowski-Mazurkiewicz, Shapley and Ky Fan.
The a in of this rarer is to prove the foll owing (Jenera l coverin9 theorem of convex sets in locally convex topological vector spaces. THEOREM l. Let X be a paracompact convex subset of a real locally convex Hausdorff topological vector space E , xO a non-empty compact convex convex ~ubs_et of X and K a non-empty compact subset of X . Let {Ai: c I} be a locally finite family of closed subsets of X such that X = u A. , and let id
1
{C.: , I} .ti_~ a family (indexed al so by I) of non-empty subsets of E Let 1 s: X ➔ P(E) be an upper hemi-continuous set-valued map with each s(x) a weakly compact_convex set . If for each x e (Kn ax) u (X\K) , the closed convex hull of ,,{Ci+ s(x): X \} meets the tangent cones e
X
+
IJ
,\>O
X
+
lJ
:\>O
:\(X-x)
if
X
' Kn
:\(X O-x)
if
X
E
ax
'
X\I(
then there exist a non-~~pty finite subset I O of and a roint x c X such that X is in the closed convex hull of u{Ci + s(x): i E Io} as well as in A
In Theoren l, P(E) stands for the collection of all non-emnty subsets of E denote the boundary operation and the closure the synbol s a and operation, respectively. Theorem l remains valid if the tan(Jent cones are replaced by X + u :\(X-x) if X (Kn ax , and X + u :\(Xo-xl if XE X\K, 11
11
11 - 11
:\ O
x E f(x)
•
Proof. We first observe that if x is an interior point of X , then x + u A(X-xJ = E ; thus condition (a) is equivalent to the following condition: A>O (a') For each XE K, f(x) meets X + ·-;;--~(X-x) A>O Suppose that the assertion of the theorem is false. Let x E X ; then x 4: f (x) , so by the Hahn-Banach theorem, there exists 1;x E E' and tx E ~ such that sup{ : v E f(x}} < tx < . As f is upper he~i-continuous and hood Ux of x satisfying
1;x
is continuous, there is an open neighbor-
(2) sup < tx < for all u E Ux yEf(U) As {Ux: x EX} is an open covering of the paracompact space X , there is a continuous partition of unity {¢x: x EX} subordinate to this open coverin~. Thus (i) for each x EX , ¢x is a non-negative real-valued continuous function on X such that its support supp ¢x c Ux , (ii) the fa~ily {supp ¢x: x EX} is a locally finite covering of X and (iii) z ¢x(y) = l for all y EX Define XEX p: X ➔ E' by setting
p(y) := -
Z ¢2 (y)1; 2 for all y E X . ZEX Let y, X ; note that whenever ¢2 (y) t O , we have y E supp ¢2 we must have
thus
c
U2
so by (2),
Covering Theorems Related to Fixed-Point Theorems
< t 2 < As
¢2 (y)
r
O for at least one
z
241
for all
v
f(y) .
c
(3)
X , it follows from (3) that
c
= - l:
zcX
¢z(y)
< - z ¢ (y) ZEX
for all
z
X
v
c
f(y) .
Thus we have shown that for each ye X , < Now we define w: Xx X ➔ JR w(x,y)
for all
v
(4)
f(y)
c
by
for all
(x,y)
Xx X .
E
We observe that (I) For each fixed x e X , w(x,y) is a lower semi-continuous function of y be a net in X for which on X • Indeed, let t e JR and let y ➔ y E X and w(x,y ) ~ t for all Because {supp ¢2 : z EX} is a a Ny of y such that tly n SUJ1Jl ¢ 2 f 0 locally finite, there is an open neiqhborhood • for at most finitely many z c X say, {z , X: N n supp ¢2 r 0} {zl ,z2•· .. ,zn}. y Choose an a 0 c I such that ya e NY for all a? a 0 , ~,e have t
~
w(x,ya)
¢z(ya) ¢z_(ya) < r,z_, x-ya> 1
1
(if
a? ao)
n
➔ -
z ¢z. (y)
i=l
1
1
- z ¢z(y) zeX w(x,y) Hence w(x,y) is a lower semi-continuous function of y on X . (II) For each fixed y EX, w(x,y) is an affine function and hence a quasiconcave function of x on X (III) Clearly w(x,x) 0 for all x e X • (IV) The set C := {y e X: w(x,y) ;i O for all x e x0} is compact. Indeed, by (I), it is sufficient to show that Cc K Suppose ye X\K ; then by hypothesis (b), there exists u E f(y) n [y + u t.(X 0-y)] . As u e f(y) , it ).>0
follows from (4) that < . By continuity of there is an open neighborhood Nu of u such that < O u A>O there exist x E x0 and r > 0 for which u' = y + r(x-y) . Therefore
As
u Ey +
~(x,y)
- z ¢ (y) ZEX
Z
l z
r ZEX
lr
Z
¢ (y) > O
Hence C c K . This shows that y E X\C Thus all hypotheses in Theorem 3 are satisfied, it follows that there exists y EX such that ~(x,y); 0 for all x EX; that is, ; for all Note that
x EX.
y E K ; thus by (a'), there exists w E f(y)
(6)
u >-(X-y)] He ;\>O shall verify that ~ Indeed, let E > 0 be given; because p(y) is continuous at w , there is an open neighborhood U of w such that
J - I < As w E
yt
u
J->O
>-(X-y) , we can find
< +
E
and
u0
n
for all
E
u0 E Wn [y +
u
J->0
U
E
W
>-(X-y)] , it follows that
for some
y + r(u - y)
[y +
u E X and
r >0
By
(6), we have ; , and hence = r ~ o .
Therefore
< + f
+
E
E.
E > 0 is arbitrary, we must have ; which contradicts (4) w E f(y) . This completes the proof. □ We remark that our proof of Theorem 4 follows the method given in Ky Fan [8]. THEOREM 5. Let X be a _2_aracoll1J}act convex subset of a real locally convex lj_~l!_sdorff topolog_f_~~Lv_~cto_r:__~JJ..~~~ E , x0 ~_n_o_r~::.empty compact convex subset of X , and K a non-elllJ}t,r conp ct subset of X • Let f: X ➔ P(E) be up~er hemicontinuous with each f(x) a closed convex subset of E ~l!.~~-t~~t (a) For each x EK n ax , f(x) ----·--meets x + Au-(X-x) - - - ---
As
as
(b)
l:_o__r:__~~ch
X E X\K ' f(x)
Then there exists
x EX
meets
such that
X
+ ~--:\(Xo-xf J-0
x
c
Kn
ax ,
if
f(x)
Similarly, for each
243
meets
x
x, X\K , if
u 11(X 0-x) , then g(x) meets x + u A(X 0-x) 11O a fixed point and therefore f has a fixed point. x
u
;.0 11>0 Applying Theorem 4, there exists a point x c X such that x, f(i) . If we take I 0 := I(x), then the proof of Theorem l is complete. REFERENCES l.
G. Allen, Variational inequalities, complementarity problems, and duality theorems, J. Math. Anal. Appl. 58(1977), l-10.
2.
J.-P. Aubin, Mathematical Methods of Game and Economic Theory, North-Holland, Amsterdam, Revised Edition, 1982.
3.
J.-P. Aubin and I. Ekeland, Applied Nonlinear Analysis, John Wiley and Sons, New York, 1984.
4.
H. Brezis, L. Nirenberg and G. Stampacchia, A remark on Ky Fan's minimax principle, Boll. Un. Mat. Ital. 6(1972), 293-300.
5.
K. Fan, Fixed-point and minimax theorems in locally convex topological linear spaces, Proc. fJat. Acad. Sci. USA 38(1952), 121-126.
6.
K. Fan, A miniriax inequality and applications, in "Inequalities III", Proceedings Third Symposium on Inequalities (O. Shisha, Ed.), pp. 103-113, Academic Press, New York, 1972.
244
Shih and Tan
7.
K. Fan, A further generalization of Shapley's generalization of the KnasterKuratowski-Mazurkiewicz theorem, in "Game Theory and Related Topics", (0. Moeschlin and D. Pallaschke, Eds.), pp. 275-279, North-Holland, Amsterdam, 1981.
8.
K. Fan, Some properties of convex sets related to fixed point theorems, Math. Ann. 266(1984), 519-537.
9.
I.L. Glicksberg, A further generalization of the Kakutani fixed point theore~, with application to Nash equilibrium points, Proc. Amer. Math. Soc. 3(1952), 170-174.
10.
B. Knaster, C. Kuratowski and S. Mazurkiewicz, Ein Beweis des Fixpunktsatzes fur n-dimensionale Simplexe. Fund. Math. 14(1929), 132-137.
'n. L.S. Shapley, On balanced games without side payments, in "Mathematical
Programming", (T.C. Hu and S.M. Robinson, Eds.), pp. 261-290, Academic Press, New York, 1973.
Selections and Covering Theorems of Simplexes MAU-HSIANG SHIH Department of Mathematics, Chung Yuan University, ChungLi, Taiwan, Republic of China KOK-KEONG TAN Department of Mathematics, Statistics and Computing Science, Dalhousie University, Halifax, Nova Scotia, Canada
§l.
KKMS Theorem
Throughout this paper, 0 will denote a simplex in a euclidean snace. The family of all faces of o (of a 11 dimensions) is denoted by F For each T c F , let c(T) denote the barycenter of 1 . A set V of faces of 0 is said to be balanced if the convex hull of the set {c(1): 1 c V} contains the barycenter c(o) of o The collection of all subsets of o is denoted by 2° . To formulate the results, we need the following DEFINITION. 1
E
A set-valued map
B : F-----+ 2° is called a Shanley-map if for each
f ,
The fol lowing remarkable generalization, which plays an important role in game theory, of the classical Knaster-Kuratowski-Mazurkiewicz Theorem [5] is due to Shapley [6]. KKMS Theorem. If A : F-----+ 2° is a Shaoley-map with each A(p) a closed subset of o , then there exists a balanced set V of faces of o such that n TEV
A(T) 'f 0 .
In case A(p) 'f 0 only for 0-dimensional faces of o , KKMS theorem becomes the classical Knaster-Kuratowski-Mazurkiewicz theorem [SJ. Shapley's Proof of the KKMS theorem is based on a generalization [6] of the Snerner combinatorial lemma [7]. A oroof usin9 a Ky Fan's coincidence theorem [1] has been recently given by Ichiishi [3]. In the present paper, we first orove a selection theorem for Shapley-maps, and 245
DOI: 10.1201/9781003420040-19
Shih and Tan
246
then apply it together with the KKMS theorem to give two new covering theorems of simplexes with facial structures. As a consequence, we obtain a general and more direct method for dealing with a recent basic covering theorem of Ky Fan [2, Theorem 2]. §2.
A Selection Theorem.
Let A,B : F->- 2° be set-valued mans. If A(p) c B(p) for each p E F , then A is called a selection for B . If A is both a Shapley-man and a selection for B , then A is called a Shapley-selection for B . The closure of a set U in a is denoted by TI. We now establish THEOREM l. If B : F-.. 2° is a Shapley-map with each B(p) an open subset of a , then B admits a Shapley-selection A : F->- 2° such that each A(p) is a a closed subset of o Since B is a Shapley-map , o = u B(p) .
Proof.
pd
Hy := n{B(pj : y
E
For each
y
E
o, let
B(p)}
Then Hy is an open set in a containing y , and therefore there exists an ooen neighbourhood Uy of y in a such that For each ,
E
F , we have
u{B{p) : pc,}= u{UY: y As B is a Shapley-map, the compactness of , set BT of u{B{p) : pc,} such that We now let K is a finite set .
then
K := u{B, : ,
E
E
B{p)
for some
pc,} .
ensures that there exists a finite
F} ;
Define A: F-.. 2° A(p) := ulUY :y
E
by K and
Uy c B(p)}
for each p E F . Then for each p E F , A(p) is a closed subset of B(p) A(p) lt remains to verity that A is a Shapley-map. Let arbitrarily fixed. For each z E, , there exist p E F with pc, such that z E Uy; it follows that y E K and Uy c TIy c Hy c B(p) z E A(p) for some pc, . Thus C
A(p). u ropEF lhis concludes the proof of our theorem. L
o such that
, E F be and y E B(p) , and hence
C
D
Shapley Selections and Covering Theorems of Simplexes
§3.
247
Some Covering Theorems of Simplexes KKMS and Theorem 1 imply the following
THEOREM 2. lf B : F-+ 2CT is a Shapley-map with each B(p) an ooen subset of CT , then there exists a balanced set V of faces of CT such that 8(1:) f
n
i:cV
0
In what follows we shall show that Theorem 2 also imolies the KKMS theorem. Indeed, ·1et A : F-+ 2CT be a Shapley-map with each A(p) a closed subset of CT For each k = 1,2, ... , we define B(k) : F-,. 2CT by B(k) (p) : = {x
CT : dist(x,A(p))
E
1/k}
(e,,,.,)=
t Mjk(-J--s(1,,.,1!e))·
. k-1
J, -
T/k
T/J
3. Hitherto, suppose Aj,1. = 1, 2 are two subalgebras of the algebra A satisfying (11) [a1, a2] = 0 for ai E Aj, 1· = 1, 2,
where [a 1,a2] is the commutator
aja2 - a2a 1 .
For¢ E Zl(A), define
for 1· = 1, 2, where/ = 2 for 1· = 1 and/= 1 for 1· = 2.
Lemma 2. The functionals ¢j, 1· = 1, 2 are tri-linear,
1>i(l, TJ; ~) = -¢i(TJ, c; ~), (b¢i)(fo, 6, 6; ~) = ¢(~; fo66),
¢1(fo, c1; [TJo, T/1]) = ¢2(TJ0, T/1; [co, c1]),
(13) (14) (15)
1>(c1T/1, c2TJ2) = 1>1(c1, 6; ,,.,1,,.,2) + 1>2(TJ1, T/2, 66)) = ¢1(c1, 6; 8(TJ1TJ2)) + ¢2(r11, T/2; 8(6c2)).
(16) (17)
and
Proof. It is obvious that ¢i are tri-linear. To prove (13), notice that
1>i (e, ,,.,; ~)
+ 1>i (,,.,, e; ~)
1
=
2(¢(~€,,,.,) + ¢(,,.,~, c) + ¢( e, ~,,.,) + ¢(,,.,, e~))
=
2(¢(~, e,,.,J -
1
1>(~, TJc)) =
1
21>(~, [e, ,,.,]),
(18)
by (1) and (4). As functions of€ and T/, ¢i(C.TJ;~) + (~, 1e, ,,.,])
= o,
(19)
Xia
302
for ~ E AJ,, c, r, E AJ. From ¢(~6, 6fo) ¢(~fo, 66)
and
+ c/>(fo, ~66) + c1>(c2, fo~6) = o, + ¢(6, 6~fo) + ¢(6, ~foci)= o
(14) follows. To prove (15), observe that and Thus, if [Ci, r:,3]
= 0 for
i,j = 1, 2, then
(20) and similarly
(21)
From (20) and (21) it follows (15) immediately. By a simple calculation, it is easy to see (16) and (17). Theorem 2. Let the algebra A over C be generated by .4 1 and .4 2 satisfying (11). Then¢ E Zl(A), if and only if¢(-, ·) is a bilinear functional on A X A satisfying (17), where ¢ 3 ,j = 1,2 are tri-linear functionals satisfying (13), (14) and
(15).
Proof. By Lemma 2, we only have to prove the "if" part of the theorem. Suppose ¢(·, ·) is a bilinear functional satisfying (13), (14), (15) and (17). It is obvious that ¢(·, •) is skew-symmetric, by (13) and (17). From (15), there is a bilinear functional 1/J(·, ·) on .4 1 x .4 2 such that
To prove (4), we have to prove ¢1(fo, c1c2; S(r,0[111112])) + ¢1(6, c2fo; S(171[r,2r:,o])) + ¢1((2, foci; S(r:,2[110111])) + ¢2(110, r11112; S(fo[ci6])) + ¢2(111, 112110; S(c1 [c2fo])) + ¢2(112, r/or/1; S(c2[foci])) = 0.
(23)
Trace Formula for Almost Lie Algebra of Operators
303
By means of (14) and (22), we ha.ve
and
¢( fo, Cl c2; 110111112) + ¢i( Cl' C2 fo; 111112110) + ¢( 6' fo C1; 112110111) = ¢(110111112, foE1c2) + '1j;([E1, E2fol, [111112, 110]) + '1j;([c2, foE1l, [112, 110111]), ¢1(fo, E16; 111112110) + ..-
174 viability
3,5,8
Range condition 194-195,200 Riesz-Dunford integral 15 Selection 102-103 Shapley- 245-246
balanced 37,245,247-249 contractible 99-105 D-convex 158,162,164 finite-concave 65,67 finite-convex 65,70-71 reachable 51-52,54 weakly compact 77-78,80-82
Side-payment
23,31-32 25,32
C
Set
Space ab- 42 bornological 42 Frechet 38,40 Lorentz sequence 170 Montel 38,40 Spectral theorem for unitary operator 16 Strategy correlated 108,115,117,
120-122
mixed 108-115 pure 109,111-114,119-122 randomized 108,116-119,
121,123
repeated-game- 119-121 Strongly summable sequence Sum ll. 00 -direct 168 11. 1 -direct
168,170
Supremization problem
259,265
Surrogate duality
264
298
253-254,
255-256,259,
Theorem cone-compression 176,185 cone-expansion 176-185 Debranges 78-79 Fan 6-7,15,56,61,63-64,
69,74,78,80,99,108, 112-113 fixed-point 103 Heinz 15
Knaster-Kuratowski-Mazurkiewicz
99-101,236,245,248 235,245,247,250 Krein 82 lop sided minimax 213,215, 222 minimax 61,64-68,211,213 separation 17,56 Simons 82 von Neumann 15,61,64,66,68 Total expected multiloss 151-152 Trace formula 299 Transition probability 151,159, 164 KKMS
Ulitmately bounded filter
42
37,40,
311
Index
Ultrafilter 38,42 Upper demi-continuity Upper hemi-continuity Utility 107,109,121 expected 115,117 Variational inequality 224
237-239 235-239
131,214,
Viability domains 5-9 largest 6 Viable solutions 1-5,8-11 heavy 3,8, 11 Volterra equation 83 Weak convergence 84,87-93, 271,273,275,286