Variational Methods in Mathematics, Science and Engineering / Prof. RNDr Karel Rektorys, DrSc Second edition D, REIDEL PUBLISHING COMPANY DORDRECHT—HOLLAND / BOSTON—U.S.A. LONDON—ENGLANDLibrary of Congress Cataloging in Publication Data Rektorys, Kare). Variational methods in mathematics, science and engineering. Bibliography: Includes index. 1. Calculus of variations. 2. Hilbert space. 3. Differential equations — Numerical solutions. 4. Boundary value problems — Numerical solutions. L. Title. QA315. R44 515°7 = 74-80530 ISBN 90-277-1060-0 (2nd edn) Translated from the Czech by Michael Basch, 1975 Published by D. Reidel Publishing Company, P. O. Box 17, 3300 AA Dordrecht, Holland in co-edition with SNTL — Publishers of Technical Literature — Prague, Czechoslovakia Distributed in Albania, Bulgaria, Chinese People’s Republic, Cuba, Czechoslovakia, German Democratic Republic, Hungary, Korean People’s Democratic Republic, Mongolia, Poland, Rumania, the U.S.S.R., Vietnam, and Yugoslavia by Artia, Prague Sold and distributed in the USA, Canada and Mexico by D. Reidel Publishing Company, Inc., Lincoln Building, 160 Ol¢ Derby Street, Hingham, Mass. 02043, USA Distributed in all other countries by D. Reidel Publishing Company, Dordrecht All Rights Reserved Copyright © 1977, 1980 by Karel Rektorys No part of the material protected by this copyright notice may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying, recording or by any informational storage and retrieval system, without written permission from the copyright owner Printed in- Czechoslovakia by SNTL, PragueCONTENTS Preface 11 Notation Frequently Used 15 Chapter 1. Introduction 17 Part I, HILBERT SPACE Chapter 2. Inner Product of Functions. Norm, Metric 21 Chapter 3. The Space LZ, 32 Chapter 4. Convergence in the Space £,(G) (Convergence in the Mean). Complete Space. Separable Space 38 a) Convergence in the space L(G) 38 © b) Completeness 41 ¢) Density. Separability 44 Chapter 5. Orthogonal Systems in L,(G) 47 a) Linear dependence and independence in Z,(G) 47 b) Orthogonal and orthonormal systems in L(G) 51 c) Fourier series. Complete systems. The Schmidt orthonormalization 54 d) Decomposition of L(G) into orthogonal subspaces 62 e) Some properties ef'the omer product 64 Chapter 6. Wilbert Space 66 a). BreHilbert space, Hilbert space 66 b) Vimear dependence and independence in a Hilbert space. Orthogonal systems, Fourier series 74 c) Orthogonal subspaces. Some properties of the inner product 78 d) The complex Hilbert space 79 Chapter 7. Some Remarks to the Preceding Chapters. Normed Space, Banach Space 81 Chapter 8. Operators and Functionals, especially in Hilbert Spaces 86 a) Operators in Hilbert spaces 87 b) Symmetric, positive, and positive definite operators. Theorems on density 98 c) Functionals. The Riesz theorem 109 Part H. VARIATIONAL METHODS Chapter 9. Theorem on the Minimum of a Quadratic Functional and its Consequences 113 Chapter 10, The Space Hy 121 Chapter 11. Existence of the Minimum of the Functional F in the Space H,. Generalized Solutions 133VARIATIONAL METHODS IN MATHEMATICS, SCIENCE AND ENGINEERING Chapter 12. The Method of Orthonormal Series, Example 146 Chapter 13. The Ritz Method 153 Chapter 14. The Galerkin Method 161 Chapter 15. The Least Squares Method. The Courant Method 166 Chapter 16. The Method of Steepest Descent. Example 172 Chapter 17. Summary of Chapters 9 to 16 178 Part If. APPLICATION OF VARIATIONAL METHODS TO THE SOLUTION OF BOUNDARY VALUE PROBLEMS IN ORDINARY AND PARTIAL DIFFERENTIAL EQUATIONS Chapter 18. The Friedrichs Inequality. The Poincaré Inequality 188 Chapter 19. Boundary Value Problems in Ordinary Differential Equations 199 a) Second order equations 199 b) Higher order equations 219 Chapter 20. Problem of the Choice of a Base 223 a) General principles 223 b) Choice of the base for ordinary differential equations 234 Chapter 21. Numerical Examples: Ordinary Differential Equations 238 Chapter 22. Boundary Value Problems in Second Order Partial Differential Equations 254 Chapter 23. The Biharmonic Operator. (Equations of Plates and Wall-beams.) 266 Chapter 24, Operators of the Mathematical Theory of Elasticity 276 Chapter 25. The Choice of a Base for Boundary Value Problems in Partial Differential Equations 285 Chapter 26. Numerical Examples: Partial Differential Equations 293 Chapter 27. Summary of Chapters 18 to 26 307 ParttIV. THEORY OF BOUNDARY VALUE PROBLEMS IN DIFFERENTIAL EQUATIONS BASED ON THE CONCEPT OF A WEAK SOLUTION AND ON THE LAX-MILGRAM THEOREM Chapter 28. The Lebesgue Integral. Domains with the Lipschitz Boundary 314 Chapter 29. The Space '#§")(G) 328 Chapter 30. Traces of Functions from the Space W{*)(G). The Space W$*)(G). The Generalized Friedrichs and Poincaré Inequalities 337 Chapter 31. Elliptic Differential Operators of Order 2k. Weak Solutions of Elliptic Equations 344 Chapter 32 The Formulation of Boundary Value Problems 355 a) Stable and unstable boundary conditions 355 b) The weak solution of a boundary value problem. Special cases 359 c) Definition of the weak solution of a boundary value problem. The general case 366 Chapter 33. Existence of the Weak Solution of a Boundary Value Problem. V-ellipticity. The Lax- Milgram Theorem 383 Chapter 34. Application of Direct Variational Methods to the Construction of an Approximation of the Weak Solution 399 a) Homogeneous boundary conditions 399 b) Nonhomogeneous boundary conditions 408 c) The method of least squares 414CONTENTS Chapter 35. The Neumann Problem for Equations of Order 2k (the Case when the For1a ((v, u)) is not V-elliptic) 417 Chapter 36. Summary and Some Comments to Chapters 28 to 35 434 Part V. THE EIGENVALUE PROBLEM Chapter 37. Introduction 441 Chapter 38. Completely Continuous Operaiors 445 Chapter 39. The Eigenvalue Problem for Differential Operators 462 Chapter 40. The Ritz Method in the Eigenvalue Problem 477 a) The Ritz method 477 b) The problem of the error estimate 485 Chapter 4!. Numerical Examples 495 Part VI. SOME SPECIAL METHODS. REGULARITY OF THE WEAK SOLUTION Chapter 42. The Finite Element Method 503 Chapter 43. The Méthod of Least Squares on the Boundary for the Biharmonic Equation (for the Prob- lem of Wall-beams). The Trefftz Method of the Solution of the Dirichlet Problem for the Laplace Equation 511 a) First boundary value problem for the biharmonic equation (the problem of wail-beams) 512 b) The idea of the method of least squares on the boundary 514 c) Convergence of the method 517 d) The Trefftz method 521 Chapter 44. The Method of Orthogonal Projections 524 Chapter 45. Application of the Ritz Method to the Solution of Parabolic Boundary Value’ Problems 533 Chapter 46. Regularity of the Weak Solution, Fulfilment of the Given Equaiion and of the Boundary Con- ditions in the Classical Sense. Existence of the Function we W!(G) satisfying the Given Boundary Conditions 545 a) Smoothness of the weak solution 545 b) Existence of the function w ¢ W{"\(G) satisfying the given boundary conditions 549 Chapter 47. Concluding Remarks, Perspectives of the Presented Theory 551 Takle for the Construction of Most Current Functionals and of Systems of Ritz Equations 554 References 565 Index 567PREFACE There were three main impulses which led me to the writing of the present book: My many years of lecturing in special courses for selected students and graduate students-engineers at the College of Civil Engineering of the Technical University in Prague, frequent consultations with technicians and physicists who have asked for advice in overcoming difficulties encountered in solving their problems, and finally many discussions with mathematicians themselves, “pure” as well as “applied”. This is also why the book is proposed for a relatively wide range of readers. The first half of the book (Parts I-III) is devoted to the theory based on the theorem on the minimum of the functional of energy and contains current variational methods with examples of their applications. These problems are relatively familiar to engineers and physicists. The second half (Part IV—VI) is grounded on the rather more abstract Lax-Milgram theorem. Roughly speaking, it is written “more” for mathematicians while the first half is written “more” for consumers of mathematics. Nevertheless, I tried very much that the whole book be for use and interest to both categories of readers. The realization of such a conception is not at all simple. Indeed, the mentioned categories of readers have often quite opposing ideas as concerns such a book and, consequently, different requirements which cannot be met simultaneously. For example, mathematicians cannot be met in their desire that the book be written in concise mathematical form, with sufficiently rapid gradation. In fact, for a reader who is not a professional mathematician and who has to under- stand well gil the individual relations and consequences (or who even intends to engage himself in the creative development of the theory), it is necessary to become acquainted with modern mathematical tools, at least with the foundations of functional analysis. This mathematical discipline — familiar to a mathematician — brings a lot of abstract concepts to a noa-mathe- matician which cannot be absorbed with haste. This is why I have advanced very cautiously. To begin with, I have defined the inner product, the norm and the metrics on the set of sufficiently smooth functions where these concepts — discussed later in more general functional spaces — are rather intuitive. Further concepts were prepared in the space L,, one of the most simple Hil- bert functional spaces. Only then did I proceed to the definition of the abstract Hilbert space. This inductive rather than deductive approach has been applied in many other places of the book, even in its (more abstract) second half. As concerns mathematicians, I have applied my best efforts to prepare topics of interest also for them. In the book they find sufficiently general existence theorems on the one hand and a lot of my own results, some of them published here for the first time (eg., the new methods discussed 11VARIATIONAL METHODS IN MATHEMATICS, SCIENCE AND ENGINEERING in Chaps. 43 and 45) on the other hand. Some of my results have been published only here. To mention some of them: refined estimates in inequalities of the Friedrichs type (Chap. 18), new results in eigenvalue problems and eigenvalue estimates for elliptic equations of the form Au — — 4Bu = 0 (Chaps. 39, 40), etc. See also the untraditional approach to the problems of Chaps. 19, 34, 35, 44 etc. Considerable attention has been devoted to practical aspects of the methods (see the complete numerical solution of examples in Chaps. 21, 26 and 41) as well as to the questions of numerical stability and of the proper choice of a base. In this connection, J have pointed out the practical advantages and disadvantages of the individual methods, including the finite element methods. The question of the error estimate of an approximate solution represents a rather difficult problem. One of the possibilities of establishing this estimate is given by the method of orthogonal projections or the Trefftz method (Chap. 44). However, a number of drawbacks of practical nature are encountered here. Therefore, I have tried to develop a rather simple estimate (11.21) and to increase its efficiency namely by finding more effective estimates for the constant of positive definiteness C (Chap. 18). As mentioned above, the second half of the book and especially Part I'V is to a certain degree more abstract. Especially, I have sought to analyse in detail problems (nonsymmetric in general) in differential equations with nonhomogenous boundary conditions, including the Neumann problem for higher order equations. This is a complex of problems which happens to be rather cumbersome in the theory based on the theorem on the minimum of functional of energy. At the same time, I have tried to investigate these topics without introducing the so-called factor spaces, which are conceptually difficult for a technically oriented reader. By mentioning Part IV as relatively more difficult I do not mean to say that it is dedicated to mathematicians only. On the contrary, my aim was to present the problems there in sufficient detail so that it be well comprehensible even to technicians and physicists to whom it may bring — in my opinion — precisely that what they find not in the “classical” theory of variational methods. Part V is devoted to eigenvalue problems for elliptic equations of the form Au — Au = 0, or, more generally, of the form Au — ABu = 0. In Part VI several special methods are presented. It was my intention to describe these methods in such a way that the reader interested in their application be able to understand them without necessarily reading the more complicated Part IV. In the final Chapter 47, I have attempted to provide the reader with information concerning several other topics related to the discussed complex of problems (nonlinear problems, problems on infinite domains, etc.). For the convenience of the reader, a table of functionals is added at the end of the book for the most common types of problems from the theory of differential equations with boundary conditions, including the corresponding Ritz system of equations. Finally, a brief note concerning terminology. There is a significant disunity in using concepts such as linear set, linear space, etc., in the literature. Some of the authors understand by “linear * space’’ a set of elements possessing certain properties of linearity, while others denote by the same terms the set in which there is already introduced a certain (linear) metric or topology. In this 12PREFACE publication we understand by a linear space a linear metric space. By a subspace of a complete linear metric space a complete linear subspace is always understood. In concluding the preface, I sincerely wish to thank all those who have by their work or advice contributed to the improvement of this book. Prague, 1979 Karel Rektorys 13VARIATIONAL METHODS IN MATHEMATICS, SCIENCE AND ENGINEERING In Part VI several special methods are presented, some of them entirely new. It was my intention to describe these methods in such a way that the reader interested in their application would be able to understand them without the necessity of reading the more complicated Part IV in advance. By mentioning Part IV as relatively more difficult I do not mean to say that it is dedicated to mathematicians only. On the contrary, my aim was to present the problems there in sufficient detail so that it be well comprehensible even to technicians and physicists to whom it may bring — in my opinion — precisely that what they find not in the “‘classical” theory of variational methods. In the concluding Chap. 47, I have attempted to provide the reader with information concerning several other topics related to the discussed complex of problems (nonlinear problems, problems on infinite domains, etc.). For the convenience of the reader, a table of functionals is added at the end of the book for the most ‘common types of problems from the theory of differential equations with boundary conditions, including the corresponding Ritz system of equations. Let me add a short comment concerning terminology: There seems to be a significant disunity when using concepts such as linear set, linear space, etc., in the Czech as well as in the foreign literature. Some of the authors understand by “‘linear space” a set of elements possessing certain properties of linearity, while others denote by the same term the set in which there is already introduced a certain (linear) metric or topology. In this publication we understand by a linear space a linear metric space. By a subspace of a complete linear metric space a complete linear subspace is always understood. In concluding the preface, I sincerely wish to thank all those who have by their work or advice contributed to the improvement of this book. Prague, January 27", £972. Karel Rektorys t4NOTATION FREQUENTLY USED ueM (oru¢M) Ew u(x) Jog u(x) dx co“) c%(G) CG) [or 6(G)] CHG) [or 2(G)] H (x, ») ulo H=H,@H, tt belongs (or does not belong) to the set M. N-dimensional Euclidean space with the usual definition of the distance o(A, B) of two points A(a,, ..., ay), B(By,.... by), (4, B) = Vo - a,y +o t (by - ay)"). domain in Ey, i.e., an open connected set in Ey. In this book, we consider bounded domains only with the so-called Lipschitz boundary (Chap. 2, p. 21; Chap. 28, p.324).In the case of N = 1, the domain G is an open interval (a, b). boundary of the domain G. closure of the set G in the space Ey (thus, @ = G + I). Instead of the closure of the domain G in the space Ey we speak, in brief, of the closed domain G. brief notation for u(x), ...5 xy). brief notation for J... fg u(x1,.... xy) dx, ...dxy. If Nt, then Jo u(x) dx = §2 u(x) dx. the set of functions u(x) whose (partial) derivatives up to the k-th order inclusive are continuous in G. By the, senvative of order zero we understand the function u(x). Insteag of C°(G) we write C(G). Thus, ué C(G) means that the function u(x) is continuous in G; ue CG) means that che functions u(x), dujdx,,..., @uf@xy are continuous in G. the set.of functions u(x) whose(partial) derivatives up to the k-th order inclusive are continuous in G. Instead of C(G) we write C(G). the set of functions u(x) whose (partial) derivatives of all orders are continuous in G. the set of functions with compact supports in G. See Chap. 8, p. 99. Hilbert space, Chap. 6, p. 73. inner (scalar) product of the elements u, » in a Hilbert (or pre-Hilbert) space. the elements u, » of a Hilbert space are orthogonal; 4 1 v <> (u, v) = 0. decomposition of the Hilbert space H into an orthogonal sum of sub- spaces H,, H2, p. 79. 15VARIATIONAL METHODS IN MATHEMATICS, SCIENCE AND ENGINEERING H,=HOH, I o(u, v) L,(G) Ww2(6) WS(6) HAG) = WG) LP) WPF) Vv D4 (u, v), Hy, A(v, u) orthogonal complement to the subspace H, in the Hilbert space H. norm of the element u, pp. 24, 34, 70, 83. distance of the elements u, v, pp. 27, 34, 70, 81. Hilbert space of functions square integrable in the domain G, Chap. 3. Hilbert space whose elements are those functions from L,(G) which have generalized derivatives in G up to the k-th order inclusive, Chap. 29. closure of the set C§(G) in the metric of the space W$°(G), Chap. 30. p. 335. p. 327. p. 340, Def. 32.1, 32.2. See also p. 359. domain of the operator A. inner product defined in Chap. 10, p. 121. Hilbert space with the inner product (u, v),, Chap. 10. bilinear form gorresponding to the operator A, Chap. 32, p. 369. ((v, u)) = A(v, u) + afv, u) {v; P} 16 Chap. 32, p. 375. set whose elements have the property P. For instance, we read M = = {v; ve C26), » = 0 on F} as follows: M is the set of all functions which belong to CG) and are equai to zero on the boundary I. If the elements v belong to some metric space, it is usual to understand by the above symbol the set M with the metric of that space (thus, a metric space as well). E.g. {e veL,(6), { ” dx = 0} is the subspace of the space L,(G) the elements of which are those func- tions from L,(G) for which f, v(x) dx = 0 holds.Chapter 1 INTRODUCTION The theory of differential equations is one of the most attractive mathematical disciplines for engineers as well as for scientists, although it is not one of the youngest. The reason lies in the fact that through differential equations the majority of phenomena studied in physics may be described, and — consequently — also of phenomena arising in fields related to physics as well as in fields which have broken away from physics as independent scientific disciplines (as, for instance, the theory of elasticity, etc.). At the same time, problems leading to the solution of boundary value problems in differential equations — especially in partial differential equations — are the source of considerable theoretical as well as practical difficulties to specialists in applied fields (and even to mathematicians). In some cases, when both the given equation and the given domain are very simple, it is possible to obtain the solution in the so-called closed form — most often in the form of an infinite series. For instance, the so-called Fourier method (cf. p. 442) and the method of the Laplace transformation lead to this type of results. The effort of mathematicians, throughout the past centrury and in the first years of this one, aimed at suitable modifications of methods of this type, or at the derivation of new methods which would lead to solutions in closed form for even more general cases. Essen- tially, it can be said that these efforts did not lead to success, and various methods of finding a suitable approximate solution began to be developed. Among these the finite-difference method and variational methods proved especially true. Of the variational methods, the Ritz method, the Galerkin method and, recently, the finite element method gained considerable popularity. The simplicity of principle, universality and relatively easy treatment on a computer are pref- erable properties of the finite-difference method. However, the application of this method is not of advantage whenever it is necessary to determine, approximately, the derivatives of the desired function on the basis of the obtained approximate solution. For instance, in problems of plane elasticity (wall-beams, etc.) we look for the so-called Airy function U(x, y) as the solution of a boundary value problem for the biharmonic equation. The components of the stress tensor eu _ eu #U y,=-— Tt. by?’ 7 at? ax ay are then computed from this function, When solving the considered problem by ihe finite-dif- ference method, the approximate values of the functions v,, v,, t,, — obtained as the difference quotients of the solution — are rather unreliable, especially in the neighbourhood of the boundary of the considered domain’). 4) It was one of the reasons why we have presented the so-called least squares method on the boundary (see 17VARIATIONAL METHODS IN MATHEMATICS, SCIENCE AND ENGINEERING As mentioned above, variational methods have gained considerable popularity in the solution of theoretical problems in engineering as well as in science. The origin of variational methods may be dated to the time of formulation of the so-called Dirichlet principle: Consider the Dirich- let problem for the Laplace equaiion in the plane, Ou eu —+-—>5=0 in G, 1.1 Ox? ay? (1.1) u = g(s) on TI, (1.2) where G is a bounded plane domain and I its boundary (we are not going to enter into more details concerning their properties at this stage). Let M be the set of functions which are continuous in G = G + I, have continuous first partial derivatives in G, and satisfy the condition (1.2). The Dirichlet principle asserts that the function which minimizes the so-called Dirichlet integral = ff) : among all the functions from M, is harmonic in G and thus constitutes the solution of the problem 1.1), (1.2). ( > Riemann applied this principle to the solution of a number of problems (especially to prob- lems from the theory of functions of a complex variable). However, Riemann assumed that it is evident that for every function g(s) continuous in IF there exists a function, from the considered set M, for which the integral (1.3) actually assumes the minimal value on the set M. The fact that the existence of such a minimizing function is not clear was shown by the German mathematician K. Weierstrass who constructed an example involving an integral, similar to (1.3) !), which does not assume the minimal value for any function from a given set of functions (of very “reasonable” properties). Some time later, an example of a function was presented by J. Hadamard which is a solution (in the classical sense) of the problem (1.1), (1.2) and for which (1.3) is divergent so that the Dirichlet principle cannot be applied. Because of these examples the Dirichlet principle remained in the background for a long time, to be revived only by the works of one of the greatest mathematicians, D. Hilbert. Hilbert showed that in the complex of problems concerning the justification of the application of the Dirichlet principle there occurred not a “small error” in formulation, but that there was in question a far deeper circumstance related to what is at present called the completeness of a metric space (discussed in detail in Chaps. 10 and 11 and in Part IV of this book). A further outstanding factor, drawing the attention of theoreticians as well as of practitioners to the application of the Dirichlet principle and of some of its generalizations, was the method submitted in 1913 by the German engineer W. Ritz. To illustrate the basic idea of the Ritz method let us mention one of the first problems to the solution of which it was applied by Ritz himself. The Dirichlet problem for the Poisson equation, 2. 2, Ou , 2M _ s(x, y) inthe domain G, (1.4) ax? ay? Chap. 43 of this book) for the approximate solution of this problem, since, according to our opinion, it is essentially more suitable for this case. 1) To be more precise, Weierstrass considered the case of functions of one variable. 18INTRODUCTION = 0 onthe boundary I, (1.5) leads to the problem of finding the minimum of the functional ry [Leff ao 0 on the set N *) whose elements are sufficiently smooth?) functions satisfying the condition (1.5). Choose, in N, n linearly independent functions P1(%, Y)s +9 Pal % ¥) > , (4.7) and look for an approximate solution of the considered problem as for such a function which minimizes the functional (1.6) not on the set N, but on its “n-dimensional” subset P the elements of which are all linear combinations of the functions (1.7), thus a!l functions of the form by 94(x, y) tot b, P(x, Y) » (1.8) with arbitrary coefficients b,, ..., b,. Substituiing (1.8) into (1.6), in place of u(x, y), the functional (1.6) becomes a certain function of these coefficients — let us denote it by h(b,, ..., b,). Now, if the functional (1.6) is to assume its minimum on the set P for some function _ u(x, y) = cy pi(x, y) + «0. + cn Pr(X, VY), (1.9) the equations oh ” 3b, — (cy, ..5¢,) = 0 (1.10) os Pe. weep Cn) = 0, have to be satisfied. However, it is obvious from the form (1.6) of the considered functional that the function h(b,, ..., b,) is a quadratic function in the variables b,,..., b,, so that the system (1.10) will be linear in the unknowns c¢),..., ¢,. If we find the solution of this system, then the function (1.9) will be, in some sense, an approximation of the desired solution of the problem (1.4), (1.5). The Ritz method — very simple in principle, as seen from the example — has become the basis of the so-called direct variational methods. In essence, even the finite element method is its special case, for a particular choice of the so-called base (see Chap. 42). In the linear case, the Galerkin method is also closely related. Variational methods were exploited by engineers and scientists as a véry effective tool for the solution of their problems. At the same time, a series of problems emerged of both theoretical and practical character. It came to be seen that even a correct, or proper, formulation of the problem and an appropriate concept of the solution are far from clear. Attention was drawn to questions of convergence, especially to questions as to under which conditions, in what sense and how fast does the sequence of approximate solutions converge to the “exact” one. New methods were developed and questions concerning their application to the solution of a certain class of ly By (1.6), to every function u(x, y) € N there is assigned a certain (real) number F(u). Such a relation is called a functional (given on the set N). (1.3) is also a functional, given on the set Af. More details conceming the concept of a functional will be found in Chap. 8, p. 109. 2) In this chapter, we do not have the possibility to base our assertions on concepts introduced only later on in the text. Because of that, our statements will not te formulated with sufficient accuracy, at some points. 19VARIATIONAL METHODS IN MATHEMATICS, SCIENCE AND ENGINEERING problems were studied. The impulse to the rise of new methods (for instance, the finite element method) and to modifications of “classical” methods came also from the rapid development of computers whose exploitation enabled a relatively very fast solution of systems of equations in a large number of unknowns. However, coon it came to be seen that inaccuracies arising from round-off errors in the calculations of coefficients of these systems as well as in the solutions of the systems may substantially affect the accuracy of the entire calculation. Surprising and obviously absurd results, frequently obtained in the course of numerical processes, provoked an intense study of the stability of these processes, which is connected with a suitable choice of the so-called base, etc. (See Chap. 20.) It proved that to find an answer to a number of questions — which are theoretically interesting and practically urgent — is not simple. It was functional analysis which played an essential role in the solution of these problems. Its tools proved to be sufficiently effectiye devices to clarify the situation. On the other hand, the complex of problems associated with variational methods contributed, to a considerable extent, to the development of some of the branches of functional analysis. Variational methods have experienced tumultuous progress from the time of their origin. As far as publications of fundamental significance are concerned, one has to mention at least the classical papers by Hilbert and Courant, further the works (which are, to a certain degree, a con- tinuation of the former) by Sobolev and Mikhlin in the theory based on the theorem on the minimum of a quadratic functional (“functional of energy”), and the results of Browder, Lions and Nefas in the theory based on the concept of a weak solution and on the Lax-Milgram theorem. Their results are of both practical (new methods, convergence theorems, etc.) and theoretical (existence theorems) significance. A considerable number of contemporary publications, dealing with linear problems, are devoted to the finite element method. As mentioned already, it is difficult to imagine the study of variational methods without know- ledge of at least some of the basic concepts and results from functional analysis, especially without being familiar with the theory of the Hilbert space. Therefore, we proceed to the investigation of this theory.PART I HILBERT SPACE Shapter 2 INNER PRODUCT OF FUNCTIONS. NORM, METRIC To start with, a few comments concerning the notation employed: By the symbol G we denote — throughout the book — an N-dimensional domain, i.e., an open connected set in the Euclidean space Ey. Only bounded domains with the so-called Lipschitz boundary will be considered in this book. The definition of this concept is rather complicated and would cause the reader unnecessary difficulties at this opening stage. Therefore, it is left over to Chap. 28. Here we note only that the definition is sufficiently general to include the domains most frequently encountered in engineering applications as far as the solution of boundary value problems in differential equations of the elliptic type is in question. For instance, to this kind of domains there belong plane domains with a smooth or piecewise smooth boundary having no cuspidal points; in the three-dimensional space domains whose boundary is smooth or piece- wise smooth and has no singularities corresponding, in a certain sense, to cuspidal points of plane curves (cuspidal edges, etc.). Examples of plane and space domains (i.e. for N = 2 and N = 3) with the Lipschitz boundary are circles, annuli, squares, triangles, spheres, cubes, etc. For N = 1, the domain reduces to an interval (a, b). Whenever more than one domain is considered sub- scripts are used to distinguish them: G,, G, etc. The boundary of the domain G is denoted by I, or by I, 2, ..., if we have the domains G,, Gz, ... The closure of the domain G in Ey, i.e., the set G + I, is denoted by G. Instead of the “closure of a domain in Ey” we speak briefly of a closed domain. The coordinates of a point of Ey are most usually denoted by x,, ..., xy. Instead of the point (x1, -.., Xy) we speak often of the point x, in brief. Instead of f-- [ue weep Xy) OX 2. Ay we write — in brief — {a dx. Of course, if N = 1, we write fue dx. The notation x, y, or x, y, Z, is frequently used instead of x,, x2, or of x,, X2, X3, a is usual in the plane or in the three-dimensional space, respectively. 21HILBERT SPACE Let a domain G be given and let us proceed to the investigation of functions given on this domain, especially to the definition of the inner (scalar) product of two functions. Throughout this book real functions are considered (as far as the contrary is not explicitly stated). Also, the constants encountered in the text, especially in definitions, are assumed real. Definition 2.1. The set M whose elements are functions') given on some set S *) is called a linear set (linear system, linear space) if with every two functions u,(x), u2(x), from M, also the function a, u,(x) + a2 u(x) (2.1) belongs to M, where aj, a, are arbitrary (real) constants. Here the sum of two functions, and the multiplication of a function by a constant, are defined in the usual sense as familiar from classical calculus. In particular, with a function u, € M also its arbitrary multiple belongs to M (as may be seen from (2.1) if a, = 0), and with two functions u, € M, u2 €M also their sum belongs to M (as may be likewise seen from (2.1) if a, = 1, a, = 1). Example 2.1. Denote by L the set of all functions continuous in a closed domain G, with the usual definitions of the sum of two functions and of the multiplication of a function by a constant. Then Lis a linear set, since — as well known — if u,(x) and u,(x) are two functions continuous in G, belonging thus to L, then also the function a; u,(x) + a2 u2(x) is continuous in G and, consequently, belongs to L. From Def. 2.1 it follows (the reader may readily verify the validity of this statement on the just presented Ex, 2.1): If M is a linear set, then together with any n functions u,(x), u2(x), -.., ua(x) also their arbitrary linear combination a, u(x) + a2 u(x) +... + a,u,(x) belongs to M. Example 2.2. Select from L as defined in Ex. 2.1 the set of all functions u(x) such that |u(x)| < 7 is valid for every x e G. Denote this new set by £. The set Lis not a linear set since, for example, it is not true that for every u(x) e £ and for every constant a we have a u(x) € L. For instance, the function defined by u(x) = 4 in G (thus, a constant function in G) belongs to £ since it is con- tinuous in & and |u(x)| < 7 holds. However, the function 2u(x) does not belong to £, since 2u(x) = 8 and thus |2u(x)| > 7. Consider now the linear set L from Ex. 2.1. For arbitrary two functions u(x)eL, o(x)eL we define: Definition 2.2. By the inner product (or scalar product) of the functions u(x), o(x) from L 4) At this stage, we limit ourselves to this special case to make the exposition as intuitive as possible, In the definition of a linear set it is in no way substantial whether the elements of the given set are functions or not. Cf. Definition 6.1 on p. 66. 2) As arule a domain is considered. 2INNER PRODUCT OF FUNCTIONS. NORM, METRIC we understand the integral I “u(x) o(x) dx . G The inner product of the functions u(x), v(x) is denoted by (u, »), thus (u, 2) = | u(x) v(x) dx. (2.2) G Consequently, the inner product of two functions from L is a real number determined by the value of the integral (2.2). Example 2.3. Let G be the square 0 < x < 1,0 S$ y S 1, and let u(x, y) = x? + y’, (x, y) = = 2. Then (u, v) = [fe + y*)-2dxdy = $. The properties of the inner product, as stated in the following Theorem 2.1, are immediate consequences of some familiar properties of the integral: Theorem 2.1. For the inner product, as defined in Def. 2.2, we have (u(x), v(x), u(x), ua(x) are arbitrary functions from L, a,, a2 are arbitrary real numbers): (u,v) = (2, x), (2.3) (ayu, + aguz, v) = a,(us, 0) + ao(ua, 0), (2.4) (u,u) 20, (2.5) (u, u) = O<> u(x) = 0 in G. (2.6) The relation (2.3) expresses the symmetry of the inner product, while (2.4) expresses its ad- ditivity and homogeneity. Relations (2.5) and (2.6) signify that the inner product of u(x) with itself is always nonnegative, and equals zero if and only if u(x) = 0. Proof of Theorem 2.1 is straightforward. Property (2.3) follows from the equality u(x) o(x) dx = J 2(x) u(x) dx, G G (2.4) from the equality { [ay uy(x) + a2 u2(x)] o(x) dx = af u(x) o(x) dx + a u(x) v(x) dx . Gc G G Further, we have (u,u) = { u(x) dx 20. G 23HILBERT SPACE Obviously, (u, u) = 0 if u(x) = 0. On the other hand, if f u(x) dx =0, G then u(x) = 0, as known from integral calculus since the function u(x) is continuous in G ac- cording to the assumptions’). Note that from (2.4) and (2.3) it immediately follows that (u,a,v, + a0) = a,(u,v,) + a,(u, 2), (2.7) where u(x), »,(x), v2(x) are functions from L and a,, a, are real constants, since (u, a1, + 202) = (a9, + 4202, 4) = a, (v1, u) + aa(v2, u) = = a,(u, 0)) + a2(u, v2). From (2.4) and (2.7) it then easily follows that (ayuy + @g¥z, 343 + Agus) = 41a3(Uy, us) + 2243(42, us) + + aya4(u,, Us) + a244(u2, U4). (2.8) (That is, the procedure of inner multiplication of functions is formally identical to the procedure of multiplication of polynomials.) In particular, for arbitrary functions u(x), v(x) from L and for arbitrary real constants a, b we have (au, bv) = ab(u, v). (2.9) The inncr product and its properties make it possible to define a further very useful concept, the norm of a function. Definition 2.3. By the norm of a function u(x) from the linear set L we understand the non- negative number Il = vey = [(f eee). (2.10) For example, for the norm of u = sin xx sin zy, on the squareO S$ x $1,0S y S 1, we get lpi 1 1 ull = Af | sin? nx sin? ny dx 4) = ma sin? nx dx ‘| sin? ny «) = oJo ° ° =VG-)=4. Theorem 2.2. The norm from Def.2.3 possesses the following properties (u(x), Xx) are arbitrary functions from L, a is an arbitrary real constant), 1) The reader has certainly noted that the assumption concerning the continuity of u(x) and v(x) in G was not exploited throughout the proof with the exception of this last step. It was sufficient that the respective integrals were defined. Consequently, there should be no difficulties in transferring the basic properties of the inner product to the next chapter, where this concept wil? be extended to a more general class of functions. 24INNER PRODUCT OF FUNCTIONS. NORM, METRIC lu 20, (2.11) lu] =O u(x) = ofx), (2.21) o(u, v) = o(v, x), (2.22) o(u, z) S o(u, v) + ef», z). (2.23) Proof: Properties (2.20) and (2.21) follow directly from the definition of the distance (2.19) and from (2.11) and (2.12). Further, from the definition of the distance and from the relation (2.13), for a = —1, it follows that’ (u,v) = fu — ol], ofc, u) = fo — ul] = [-(@ — of = [=a] Ju - off = fv - of, whence we immediately obtain (2.22). Further, by (2.15) we have o(u, z) = |u — 2] = (@ — ») + (@— 2) s fu - of + fo — a (2.24) which implies (2.23) since the sum of the last terms in (2.24) is equal — in accordance with the definition of the distance — to the number g(u, v) + af», z). Remark 2.2. Properties (2.20) to (2.23) of the distance also have a simple geometrical analogue, similarly as the properties of the inner product and of the norm (see Fig. 1): Denoting by @(u, v) the distance of the endpoints of the vectors u, v with initial points at the origin, we have, first, o(u, v) = ofv, u) [cf. (2.22)] since g(u, v) is determined by the length of the vector u — y, or of the vector v — u, which is the same. Further, it is obvious that o{u, v) 2 0, with o(u, v) = 0 if and only if the vectors u and v coincide [cf. (2.20).and (2.21)]. The inequality e(u, z) < e(a, v) + + ov, z), whose geometrical significarite 1s obvious from Fig. 2, is an analogue of (2.23). Thus, second coordinate first coordinate Fig. 2. we see that the relationship to elementary geometrical concepts, and to their properties, is rather intuitive in the case of this concept as well. Remark 2.3, Def. 2.4 makes it possible to “measure”, by (2.19), the distance of two functions belonging to L. We say that a metric is given on the linear set L by the formula (2.19), and the linear set L (or any other set in which a metric is given) is then called a metric space. In general, we define: Definition 2.5. The set M is called a metric space if for every pair u, v of its elements there is defined a number g(u, v), called the distance of the elements u, v, with properties (2.20) to (2.23). 28