CALCULUS OF VARIATIONS y icles and exercises) A. Mouvani TNR TOE and amended oy Moser, 1962 Supplementary noes iv Meetie feeste ang Ianan Robin Courant Institute of Mathematiles? Gelerrce: New York UniverstiyTae Courant Institute publishes a number o sets of lecture notes, & list of title: currently availoble will be sent uoon reques t. Courant Institute of Mathematical Sclences 251 Mercer Street, New York, New York 10012 adContents Page Tatroduction 2 I. Formalism of Calculus of Variations 1. The Euler Equation 10 2, Generalizations of the Zuler Equation 18 3. Natural Boundary Conditions 25 4. Degenerate Euler Equation 30 5. Isoperimetric Problems 33 6. Parametric Form of the Theory 42 7. Invariance of the Euler Equation 48 8, The Legondro Condition 52 TI. Hamilton Jacobi Theory ~ Sufficient Conditions 1. The Legendre Transformation Su 2. The Distance Function - Reduction to Canonical Form 56 3+ The Hamilton-Jacobi Partial Differential Equation 60 4. The Two Body Problem 7 5. The Homogensous Case - Geodesics 76 6, Sufficient Conditions 62 7. Construction of a Field ~ The Conjugate Point 87 Supplomentary Notes to Chapter II 1. Tne Hamilton-Jacobi Equation 90 2. Correspondence between Solutions of the Hamilton- Jacobi Equation and Fielas oh 3. Application to Differential Geometry 96 4. Analytical Representation of a Field 97 5. Conjugate Points 100 6. Application to Sturm's Theory 105 III. Direct Methods tn the Calculus of Variations Introduction 109 Compactness in Function Space, Arzelats Theorem and Applications 14 Application to Geodesios: Lipschitz's Condition 18 fiiupplementary Notes aad Exercises Direct Variational Methods in the Theory of Integral Equations 12. Dirichlet's Principle 12k, Dirichlet's Integral 12h Minimizing Sequences 124 Explicit Expression of Dirichlet's Integral for a Cirele. Hadamard's Objection 124 The Gorrect Formulation of Dirichlet's Principle 126 Dirichlet's Principle 126 Lower Semi-Continulty of Dirichlet's Integral for Harmonic Puncttons 126 Proof of Dirichlet's Principle for the Circle 127 "Distance" in Function Space. Triangle Inequalities 129 Construction of a Harmonie Function u by a "Smoothing Process" 132 Convergence of w, 137 Proof that D(u) = a 139 Proof that the Function u Attains the Prescribed Bowndary Values 140 Alternativo Proof of Dirichlet's Principle 143 Numerical Procedures 150 The Ritz Method 150 Method of Finite Differences 153 Boundary Valuo Problem in a Net 156 Existence and Uniqueness of the Solution 156 Practica] Methods 157 Convergence of the Difference Equation to a Differential Equation 158 Method of Gradients 159 Application of the Calculus of Variations to the Eigenvalue Problems 162 Extremun Properties of Zigenvalues 162 The Maximum-Minimum Property of the Eigenvalues 166 1. References 169 2. Notes on the Brachistochrone problem 179 3. The road of quickest ascent to the top of a mountain 171 ivPare 4. The shortest connection between two points in a closed simply-connected region 17h 5. The shortest connection in a plane between two points 178 6. Problems for which no solutions exist 180 7+ Semi-continuity 188 8. Genoralization of the Fundamental Lemma 189 9. Derivation of Euler's equation by special variations 191 10. Characterization of the longer great circle arc 192 11. Integration of Euler's equation in special cases 192 12. The shortest connection between two points on a sphere 196 13. Application of Euler's equation to classical problems 203 ly. Invariance and the Euler expression 213 15. Transformation theory 24? 16. An approximation theorem for minimum problems with aide conditions 270INTRODUCTION The Calculus of Variations has assumed an increasingly important role in modern developments in analysis, geometry, and physics. Originating as a study of certain maximum and mininsn problems not treatable by the methods of elementary calculus, variational calculus in its present form provides powerful methods for the treatment of differential equations, the theory of invariants, existence theorems in geometric function theory, variational principles in mechanics. Also important are the applications to boundary value problems in partial differential equations and in the numerical calcu- lation of many types of problems which can be stated in vari- ational form, No literature representing those diverging view- points is to be found among standard texts on calculus of vari- ations, and in this course an attempt will be made to do justice to this variety of problems. The subject matter with which calculus of variations is concerned is a class of extremum (i.e. maximum or minimum) problems which can be considered an extension of the familiar class of extremum problems dealt with by elementary differen- tial calculus, In the elementary problems one seeks extremal values of a function of one or more (but in any case a finite number) real variables. In the more general problems consi- dered by calculus of variations, the functions to be extremized, sometimes called functionals, have functions as independent variablos. The area A(f) below a curve y = f (x), for oxample, is a functional since its value depends upon a whole function f. (It is possible to treat a functional as a function of an enumerable set of Fourier coefficients, but this attack usually leads to almost insuperable difficulties.) One of the earliest problems of this type was the iso- perimetric problem considered by the ancient Greeks, This is to find, among all closed curves of a given length, the one which encloses the maximum area, It is intuitively evident that the solution is a circle, but this fact has beensatisfactorily proved only in recent times, and the corres- ponding theorem concerning the sphere is even mare difficult. The modern development of calculus of variations, however, began in 1696 with the formulation of the brachis- tochrone problem by John Bernoulli. This problem is to find, among all curves connecting two given points, that one which has the property that a particle sliding along it under the action of gravity alone falls from one point to the other in the least tine. This problem excited great interest among the mathematioians of that day, and gave rise to a train of research which is still continuing. Subsequent developments in classioal calculus of vari- ations were the derivation of necessary conditions for an extremum {corresponding to the conditions grad f(x. yoXoe 09%, 50 for a function f of n variables ) by Euler and more rigorous- ly by Lagrange; and the development of sufficient conditions (corresponding to the consideration of the quadratic form in seoond derivatives of £(x),%5,+++X,) at a stationary point) vy Hamilton, Jacobi, and others; culpinating in the completion of this theory by Weierstrass. The broader aspects of physical variational principles were first set forth by Maupertius, and were given a firmer foundation by the work of Euler, Hamilton, Jacobi and Gauss. We will now consider the mathematioal formulation of several problems; A —-¥ X a) The Brachistochrone A partiole P slides \ under the influence of gra- PL vity along a curve connecting 1 “~B( a) \ tae A and B. The velocity v at ut any point is given by vee ym, so that the time of fall T is (B) ate ®B) as (a) (ay¥ 2guSuppose the curve is given by u = f(x), where £(0) = 0, f(x) = uj, and f\x) is assumed to be piecewise difforontiebie. Then co = /1 + ul"dx. Hence the solution of the peoblem cen oe obtained sy finding the function u = f(x) which minimizes the integral T (e functional) x vi 2 pede Y/R ik. VE % sernoulli obtained the solution to this provlem using en en- tively different line of reesoning. He approximated the path uu? f\x) by a series of line segments dividing the distance frllon into equel parts, t!.e particle velocity bei sued constant through- A out each segment. It 18 an elementary exercise in caloeulus to devive Snell's law of refraction u B = constant ain i . gins M4 Yy ag the condition for the path of rinimm time across a cis- continuity. Taking; the livit os tie segments are made smeller, Bernoulli argued thet the curve would’ be given by sin @ eee which is indeed the correct answer, charecteriaing tie cyelord. Gf course, Bernoulli's solution is only an indication rather than a vroof, since A x he neither justified 8 the limating process, nor stowed that his solution was unique. = constant%) biniwun Area of a Surface of Revolution Consider the surface generated by revolving tie curve AB svout the x-axis. If the equataon of this curve is y = f(x), where fx) = Yor and f is piecewise differentiable, then the area of the surface is given by the functional x 2 x(t) = an § cAvite® ax. ‘1 The problem, then, is co determine f so that I(f) is a minimum. ‘he prooler. can be "solved" physicolly by stretching a soap film between the two circles (made of wire) st A and Be Surface tension in the film will minimize the erce. ce) Geodesics The curve of shortest length connecting two voints in @ plane is a straight line. This need not be taken as an axiom, but can be proved. Similarly, on the sucfece of a sphere, the curve of least length is the arc of a great circle. In general, on any surface, the curves of least length conxec- ting vairs of points are celled geodesics and their determi- nation leads to problems in calculus of variations. In case the surfoce is developable (i.e. one which can be deformed into a plens without eltering length -- e.g. a cone) the geodesics are given by the corresponding atraight lines in the plane. a) The isoperimetric Provliem Consider a plane closed curve given in pera- weteie form by x = ft}, g(t) where f and g have piecewise continuous deri- vatives, are of period an in t, 4and the curve has a given length L, en l= f [PTF at 0 The problem 1s to find, among all f and g satisfying these condi- tions, the pair that maximizes the area A 2 a-h (xy - yx) at = 3 0 This problem is different from the foregoing three problems in that we seek to extremize a functional A of two variables (the functions f and g) subject to a prescribed condition, L= constant. All such problems in caleulus of variations are called tso- perimetric problems; the analogy with the carresponding elementary problem of extremizing a funetion F(x.y) of two real variables subject to an auxiliary condition, say G(x.y) = 0, is evident. We will now prove that the circle, i.e. c(t) asin t + aycos t alt) = a,sin t - aycos t maximizes A, subject to L = constant. Consider the expression 2 L beqe-a For the cirele, I = 0. We then w: h to other curves. Let t = aut, where s ls ar wo that T > 0 for ell henjgth. Then and HenceSince (x + y)? # (7 - x)? 2 0, we will consider aa fw + x? )at + fw = y®) at. Under the conditions imposed, we may expand x and y in Fourier series 2 x~B (a,cos nt + bein nt} oO ~Z_ (aicoa nt + bt ain nt) , y a (ax ) By taking the center of gravity of the curve as the orivin (i.e. translating the axes x! = x + Xe yosye Yq) so thet 2n xi dts jr are =e! 20, ay tay 20 Then, droppin; the primes, we have ‘aor x f (8? » yet = Le (n®(a2 + v2) ~ (a2 + v2)) 0 waich 4s positive unless a, = b, = 0 for n> 1, i.e. unless x = a)008 t+ byein te J - 37) ato 0 unless y = aj con t + bisin t. Similarly But in the case that x end y are both of thie foxm, we have req (rary = x)?) at =E (ia, + apedy - by)? + (a, = af + dy + OY)7) 6whieh ta zero only if a, = bj, af =v. { je Hence I > 0 unless X= a,008 t + by sin t y= bycos t - a) sin t which are the parametric equations of a circle. In all the problems that we have considered so far we have tacitly assumed that they make sense, i.e, that a solution exists. Howaver, this is by no means always the case. For example, consider the D integral q i 1b) = ( —* 1 2 o 1+ [Or (xh 3 $(0) = 0, 602) #2 A c x where $ is subject to the condition that it pass through the two points A and B, and let us try to find a continuous and piecewise differentiable function > which wither maximizes or minimizes I($). By inspection we see that o< (bp <2 since the integrand is positive and always less than one. However from the figure it is easily seen that by picking point © very close to x = 1 we can make I take values as close to unity as we please for the curve AC B, and by taking the ordinate of D large enough we can make I as small as we please for the curve ADB, Since there is no admissible curve } which will make I() take on the values 0 or 1, there is no solution to either the minimum or maximum problem, Let us now consider a problem in which the existence of a solution depends on the class of admissible curves, We look for a closed curve of minimum erea, within which a line of given lengthcan turn through a complete revolution. If we limit ourselves to convex curves, the solution is given by the equilateral triangle having the given line as 7altitude. However, if we remove the restriction of convexity, it can be shown that the area can be made as small as we please, and since no solution exists having zero area, there is no solution to the problem. We therefore see that in order for a variational ->ro- blem to be solvable, some care :.ust be teken in the choice of admissible functions. Problema 1) Verify that the property S108 = constant rz characterizes the cycloid. 2) Characterize the larger great circle ore between two points on a sphere as the solution of ¢ riinimex problem. Generalize to a great ciroie aro which winds around n times. 3) Find the shortest path between two points A and B B lying botween two convex curves as boundaries. This is an example of a general type of problem in which auxiliary conditions in the form of anequalities are imposed | in oontradistinction to isoperimetric problems where auxiliary equations must be satisfied). 4) Pind the path which requiroa the least time in ascending a rotationally symuetric mountain using, a car with its velo- oity a civen funotion f of the angle of inclination a, such that 8M0) ey, £5) = 0, f(a) and f(a) monotonic. £\ 5) Show thet any admissible functicn $ \x) cen be approxi- mated by admissible functions $,(x) and $5.) such that 16) oan be made as amall as we ploase and 1(6)5) as close to unity as we please, where 1d) = , gx } d+ (dixieI. FORMALISH OF CALCULUS OF VARIATIONS 1. The Yuler Equation. ‘he simplest type of problem in the calculus of variations is to extrenize a functional I(>) of only one independent variable, the function (x). In practice, the functional is usually an integral of a given form, and we will henceforth restrict our discussion to functionsls which are integrals, In Genersl terms, then, the siinplest tyve of problem is to extrenize p fy 1p) = s PC, 1x) P(x) Jax where F is a given function which we will assume }.aa conti- nuous first partirl derivatives and piecewise continuous second pastial derivatives, ‘tno function $\x) will be rea- tricted to the class of admissible functions satisfying the conditions (2) O(a) = a, (db) = B $(x) continuous P'(x) piecewLse continuous The brachistochrone is an example of this type of problem. Assuming that an adnissible function u(x) exists for whieh Iju) is an extremum, we first wish to find a necessary condition wiich this function must satisfy. Consider a function $(x,t) such that 43) O(x,t) 1s admissible for ell t Q(x,t) end 9, .x,t) are continuous 9,4 (48) is plecewise continuous o(x,0) = wx) Por example, we may choose $(x,t) = u(x) + tZ(x); however, any function satisfying (3) will suffice. If wo define (yd) Ot) = 109(x,t)) lothen G(t) has a stationary point at t = 0. Accordingly (5) a -& t Fiapsbtdax | = 0 Differentiating under the integral sign, we have b 16) i) (Pye + Py oax 20, where+ . (n) (x) = O40 4,0) « According to (2), (3), and (7) we see that (9) % Ls continuous &' piecewise continuous ta) =2(b) 20, the last equation being true since ((a,t) 5 A and $(b,t) = B. If we modify (6) by integrating by parts, the anelogy with the corresvonding necessary condition for an extremum of a function of n voriables is revealed. 9) 2 f (Pye +R Sax = Fe + (ik Fy eax le a b diy “f iP, - wy itax 0. a In the case of a function f(x,,--+,x,) of n variables, we may derive the necessary conditions for an extremum by considering g(t) > f(xy (t), Kyit), +0 ex, (t)) where the equations, 44" x(t) define a ourve in n-dimensional space. If 1%, (0), (0), 00x, (0) is to be am extremum, then (io) q =O. # |-o In the literature & and are usually oalled the Variation of u ond u! respectively, and written 6u, du! . il