AN ABSTRACT OF THE THESIS OF for the SERGEI KALVIN AALTO (Name) Date thesis is presented Title May M.A. (Degree) in Mathematics (Major) 3, 1966 REDUCTION OF FREDHOLM INTEGRAL EQUATIONS WITH GREEN'S FUNCTION KERNELS TO VOLTERRA EQUATIONS Redacted for Privacy Abstract approved (Major professor) G. F. Drukarev has given a method for solving the Fredholm equations which arise in the study of collisions between electrons and atoms. He transforms the Fredholm equations into Volterra equations plus finite algebraic systems. H. Brysk observes that Drukarev's method applies generally to a Fredholm integral equation (I -> G)u In = h with a Green's function kernel. this thesis connections between the Drukarev transforma- tion and boundary value problems for ordinary differential equations are investigated. In particular, it is shown that the induced Volterra operator is independent of the boundary conditions. operator can be expressed in terms regular X . of The resolvent the Volterra operator for The characteristic values of G satisfy a certain transcendental equation. The Neumann expansion provides a means for approximating this resolvent and the characteristic values. To illustrate the theory several classical boundary value problems are solved by this method. Also included is an appendix which relates the resolvent operator mentioned above and the Fredholm resolvent operator. REDUCTION OF FREDHOLM INTEGRAL EQUATIONS WITH GREEN'S FUNCTION KERNELS TO VOLTERRA EQUATIONS by SERGEI KALVIN AALTO A THESIS submitted to OREGON STATE UNIVERSITY partial fulfillment of the requirements for the in degree of MASTER OF ARTS June 1966 APPROVED: Redacted for Privacy Professor of Mathematics In Charge of Major Redacted for Privacy Chairman of Department of Mathematics Redacted for Privacy Dean of Graduate School Date thesis is presented Typed by Carol Baker May 3, 1.966 TABLE OF CONTENTS Page Chapter I. INTRODUCTION 1 II. GREEN'S FUNCTION FOR A SECOND ORDER DIFFERENTIAL EQUATION 7 III. IV. GREEN'S FUNCTION FOR AN NTH ORDER DIFFERENTIAL EQUATION SOLUTION OF THE INTEGRAL EQUATION (I -X K)u V. VI. 13 = h + X fCu) APPROXIMATE SOLUTION OF THE INTEGRAL EQUATION (I -X K)u = h + X fcD (u) 18 31 SOLUTION OF THE INTEGRAL EQUATION (I-X K)u = h + X lrf.(D.(u) 37 BIBLIOGRAPHY 46 APPENDIX 48 REDUCTION OF T'REDHOLM INTEGRAL EQUATIONS WITH GREEN'S FUNCTION KERNELS TO VOLTERRA EQUATIONS CHAPTER I INTRODUCTION We sha1l be concerned with a certain rnethod of solving particular Fredholrn (r. 1) 'W'e inte gral equations of the second kind, u(x) - assurne ^ t, G(x, s)u(s)ds = h(x). that G, h and u are continuous cornplex valued functions on their closed domains of definition. Physicists are interested in obtaining solutions to integral equations of form (1. 1) which often arise in the study of collisions between electrons and atorns IZ,S] . One method physicists use to obtain approxirnate solutions to (1. I) is to calculate one of the Born approximations which are truncations of the Neurnann series solu- tions lZ, p. 1536i L2, p. 1073] . The Born approxirnatisns are use- ful only when the Neurnann series converges and, generally, this occurs only for sufficiently small values of the parameter ). in equation (1.I). It is of physical interest to obtain solutions to (I.1) for larger values of }. Thus there is motivation to study other rnethods of solving Fredholrn equations. 2 Integral equations (1. 1) often come from ordinary differential equations with two point boundary conditions, e. g. a one dimensic,_nal scattering problem. In this case is the Green's function associ- G ated with the given boundary value problem. Thus we are led to con- sider (1. with a Green's function type kernel, 1) G(x, s) (1. 2) In (1. 2) let V(s)f(s)g(x), 0 < s < V(s)f(x)g(s), 0 < x < 1 < s < 1 , = V, f and g x . be continuous complex valued functions defined on the closed unit interval and further suppose that g *0 and kernel of * 0, *0. f G. F. V Drukarev gave a novel method form (1. 2) [ solving (1. of 2, p. 1536; 5, pp. 309 -320] transform the Fredholm equation into a . 1) with the He was able to Volterra equation and a finite algebraic system for certain constants. H. Brysk observes that Drukarev's transformation of the Fredholm equation into a Volterra equation is possible because the kernel is of Green's function type attempts to show that the solution [ 2, p. 1536] of the . Brysk further Volterra equation leads to the solution of the Fredholm equation obtained by using the Fredholm resolvent [ 2, pp. 1537 -1538] . Briefly, the transformation of a Fredholm equation with 3 kernel (1. 2) depends on ('1 J G(x, s)u(s)ds (1.3) x = 0 J V(s)f(s)g(x)u(s)ds 0 1 V(s)f(x)g(s)u(s)ds + x = J V(s)[f(s)g(x)-f(x)g(s)] u(s)ds 0 1 + V(s)g(s)u(s)ds. f(x) 0 Let (1.4) K(x, Then equation (1. 1) s) = V(s)[f(s)g(x)- f(x)g(s)] 1 V(s)g(s)u(s)ds . . 1 K(x,$)u(s)ds u(x)-X x can be rewritten x (1. 5) 0 < s < < , = h(x)+Xf(x) 0 J 0 The right member of (1. 5) is of the form h(x) where h and f are known and + cf(x) c is a constant depending on 4 u and X . function u. Note that (1.5) is a Volterra equation for the unknown If it is solved for stitution of the solution into u with arbitrary, then sub- c yields an equation for (1. 5) technique we shall develop to solve (1. 5) The c. is somewhat analogous to the "shooting method" discussed by Henrici [ 7, pp. 345 -346] Brysk deals with the special case h (cf. Chapter VI). 1537] = f , [ 2, p. and thus his Volterra equation has the form u(x) - ('x K(x, s)u(s)ds f(x)[ 1+X = (' V(s)g(s)u(s)ds] 1 J 0 1 . 0 This is a brief summary of the results of Drukarev and Brysk dealing with the mathematical aspects of the problem. The author intends to set the problem in a more abstract setting and to extend the results obtained by Drukarev and Brysk. As we will be dealing with integral equations with continuous kernels, it will be convenient to work in the complex Banach space C of continuous complex valued functions defined on the closed unit Ifli interval with the norm = max {If(x)I:xe[0, letters will denote continuous linear mappings For example, define G and (Gu)(x) G(x, s)u(s)ds, = 0 and C by the equations K 1 (1. 6) of 1] }. Capital into itself. 5 (Ku)(x) (1. 7) = ('x J K(x, s)u(s)ds 0 where and G are given by K (1. 2) and (1. 4). Now (1. 1) may be expressed by ( 1. =h (I - XG)u 8) where is the identity operator on I C. Capital Greek letters will denote continuous linear functionals; continuous linear mappings of i. e. The set of all linear functionals on In particular, define (u) (1.9) where SEC* V(s) and g(s) C C into the scalar field forms a Banach space J C*. by the equation 1 = V(s)g(s)u(s)ds 1 J 0 are as above. Note that 1, f 0 . As a final notational convention, the symbols for elements of C will also be used to indicate mappings from the scalars into C. This convention is adopted because of the obvious isomorphism be- tween f: y - (1. 10) and these mappings: for each f EC C C by (fy)(x) = yf(x) define the mapping 6 where C ye into . C With this convention DI. with the one dimensional range operator ft. has rank one where the rank dimension of is a linear operator on {yf: of an NE J } . Thus the operator is the its range. Now (1. 3) and (1. 4) may be (1. 11) G = expressed by K + RD, and (I-X K)u = h (1. 12) Thus, the Fredholm operator of the Volterra operator K G + X fsl (u) . has a decomposition into the sum and the operator f.T. of finite rank. In Chapters II and III of this thesis the above decomposition of a Fredholm operator with a Green's function kernel arising from an ordinary differential equation is investigated. and VI, the solution of (1. ters IV and VI 12) In Chapters IV is developed. Also included in Chap- are examples worked out using the techniques inspired by Drukarev. Approximate solutions of (1. 12) are discussed and error estimates given in Chapter V. Finally the solution of (1. 12) is related to the Fredholm resolvent operator in the Appendix. 7 CHAPTER II GREEN'S FUNCTION FOR A SECOND ORDER DIFFERENTIAL EQUATION In this chapter a brief outline construction of the of a Green's function for a boundary value problem arising from a second order ordinary differential equation is given. Then the integral operator arising from this construction is decomposed as in Chapter I. A close examination of the Volterra operator shows that it is independent of the boundary values. Further discussion is given to show the relation of this decomposition to more classical results of ordinary differential equations. The Green's function will be constructed for the G(x, s) second order differential operator Lu=u" where u is defined on [ + plu' p2u + pi, p2 and 0, 1] C E with boundary conditions (2. 1) a a2u'(0) = 0, (2. 2) ß1u(1)+ 132u'(1) = o, lu(0) + I al Ißl I + + I a2I > 0, IR2I > o . 8 We wish to solve the equation (2.3) Lu (heC) h = subject to boundary conditions (2. 1) and (2. 2). We assume that this boundary value problem has a unique solution. Since it follows from the theory of ordinary dif- pl, p2 e C, ferential equations that there exist two linearly independent functions ul and satisfying the homogeneous equation u2 Lu (2. 4) Thus the Wronskian of = u1 0 [ and u2 xe [ 0, 1] u2(x) t _ ui (x) for The assumption that the solution to the boundary . chosen such that . 0 u2(x) value problem is unique assures us that 378] . is nonzero; that is ul(x) W[ul(x),u2(x)] 3, p. 106] u1 satisfies and u1 (2. 1) and can be u2 satisfies u2 (2. 2) [ 9, p. The method of variation of parameters yields a solution of (2. 1) -(2. 3) in the form çx u u(x) = 0 (s)u (x) W[u (s), u ()] s 2 1 1 h(s)d(s) + x u(x)u (s) W[u (s), u (s)]h(s)ds 2 1 9 [ 9, pp. 378-379] . Let u1(s)u2(x) W[ul(s),u2(s)] G(x, s) x < 1 < s < 1 0 < s < ' , = u1(x)u2(s) w[ul(s), u2(s)] 0 < ' x . Then 1 u(x) (2. 5) G(x, s)h(s)ds = . 0 The function is the Green's function associated with the G(x, s) differential operator Now using the L with boundary conditions (2. and (2. 2). 1) decomposition developed in Chapter I the right side of (2. 5) may be rewritten (s)u2(x) - ul(x)u2(s)] 1 G(x, s)h(s)ds 0 = 0 1 u2(s) 1 + u1(x) S-'0 Let h (s)ds W[ul(s), u2(s)] W[u 1 (s), u 2 (s)] h(s)ds . 10 K(x,$) - u1(s)u2(x) ul(x)u2(s) - , W[ul(s), uz 0 < s < x< 1 1 Then 1 u2(s)h(s)ds ('x r J K(x, s)h(s)ds + u 1 (x) J W[u (s) ,u (s)] 0 0 2 1 (2. 6) G(x, s)h(s)ds SI = 0 1 The kernel of the Volterra operator has the property that it is in- variant under linear combinations of the functions u1 and u2, that is, if vi(x) = y1u1(x) + Y2u2(x) v2(x) = 61u1(x) + 62u2(x) (2. 7) and Ni62 - N261 f 0, then vi(s)v2(x) K(x,$) - - v1(x)v2(s) W[v1(s),v2(s)] . This follows since vi(s)v2(x)-vi(x)v2(s) = (Y162-Y261)(ul(s)u2(x)-u1(x)u2(s) ) and w[vl(s),v2(s)] _ (v152-Y251)w[ul(s), u2(s)] 1 1 If boundary conditions (2. and (2. 2) are changed to 1) (2. 1)' alu(0) + a2u'(0) = 0, I (2. 2)' plum + R2u'(1) = 0, I then the linearly independent functions the homogeneous equation (2. and + I u2. (2. 7) for some > 0, and v2 I v1 satisfying and the boundary conditions (2. 1)' In other words they can be Y2, Y1, 51 and 52. Ylul(x)+Y2u2(x) ('x K(x, s)h(s)ds = a2I > 0, ßl I+ ß2I expressed as in equations The solution to the new boundary value problem written in terms of u(x) I respectively can be expressed as linear combinations of and (2. 2)' u1 4) al + S 0 1 Therefore the kernel 2 - K(x, s) Y 2 S 1 , u1 and u2 is 1[S1u1(s)+S2u2(s)] h(s)ds W[u (s), u(s)) 0 1 2 is independent of the boundary con- ditions and the second term on the right varies with the boundary conditions. It is easy to u (x) p satisfies equation In fact u (x) verify that = xK(x, s)h(s)ds 1 0 (2. 3) and the initial conditions u(0) = u'(0) = O. is the "particular" solution to (2. 3) which can be 12 found by the method of variation of parameters if any two linearly independent solutions to equation (2. 4) are given. other hand On the it is well known from the theory of ordinary differential equations that any solution to equation (2. 3) can be written u In our = u + + 13u2 [ 9, p. 356] . case we have 1 a and aul ß = O. - S' 0 u2(s) h(s)ds W[u1(s),u2(s)] Thus the solution obtained via the Green's function and the decomposition gives the parameters a and ß as functionals operating on the function h. As a final remark it should be noted that boundary value problems with inhomogeneous boundary conditions can be transformed into boundary value problems with homogeneous boundary conditions. The term on the right side of (2. 3) is modified by this transformation, but the discussion is simpler for homogeneous boundary value problems. Thus the techniques used here apply with greater generality than indicated above. 13 CHAPTER III GREEN'S FUNCTION FOR AN NTH ORDER DIFFERENTIAL EQUATION results In this chapter the of the previous chapter are generalized to a boundary value problem arising from an nth order ordinary differential equation. The more general results do not appear in as simple a form as the second order case considered in the previous chapter. In the second order case the Fredholm opera- tor G admitted the decomposition G In the nth K + f(1). = order case we obtain a decomposition n (3. 1) G = K+ f.(D. i i . i=1 However the Volterra operator is still independent of the boundary conditions. The Green's function will be constructed for the differential operator n (3. 2) Lu dn-iu = pj j=0 dxnj- nth order 14 where u p0(x) 0 > is defined on for X [ 0, [ 0, 1] pi EC, , j = ,n and 0, 1, with boundary conditions 1] n -1 (3. 3) Ui(u) ßlJu(j)(1)] alJu(j)(0) = 0 = j =0 i = 1, 2, , We wish to solve the n. boundary value problem given by (3.4) Lu = (hEC) h As before we assume that this and boundary conditions (3.3). boundary value problem has a unique solution. The usual definition of the Green's function for the operator L given in (3. 2) with boundary conditions (3. and its derivatives up to and including the G(x, s) (a) derivative are continuous for (n -2) (b) ElimO+ (c) - axn -1 0+ for each fixed Ui(G) = In the region tion G(x, s) 0, i x, s 0 < < 1, an -1 an -1 E is 3) = G(s +E, s) -axn -1 G(s s E [ 1, 2, 0 < s < and all 0, 1] , x < 1, has the representation n we [ -E x , # s)} - p0(s) L(G(x, s)) s 8, p. 254] , = 0, . assume the Green's func- 15 n G (x, s) ai(s)ui() x = i=1 and in the region 0 < x < s < 1 n G (x, s) b. (s )u. (x) i i = i=1 {u.(x): i where i = ,n} 1, 2, is a linearly independent set of solutions to the equation Lu = 0. Using conditions (a) and (b) unique solutions for the quantities c.(s) i i = 1, 2, . . . ,n are obtained K(x, s) a.(s) i = [ - b.(s) i 8, pp. 254 -255] ci(S)ui(x) , _ Let . 0 < s < x < 1 . i=1 Using condition (c) unique solutions for the terms of the a.(s) c.(s) i = i ci( s) + b.(s), i and the boundary terms i = 1, 2, ,n and thus b.(s) i [ are obtained in 8, p. 255] a.(s) i and . But bi(s) can be found such that the assumed representation of G(x, s) in 16 the appropriate regions are satisfied. Therefore n n c.(s)u.(x) i=1 G(x, s) b.(s)u.(x), i i + 0 < s < x < = _ _ 1 , i=1 = n Ib.(s)u.(x), 0 < x < s < 1 i=1 n K(x, s) bi(s)ui(x), + i 0 < s < x < 1 , i i=1 n 0 < bi(s)ui(x), x< s < 1 . i=1 Thus the solution to (3. with boundary values (3. 3) can be repre- 4) sented by x 1 (3. 5) u(x) G(x, s)h(s)ds _ bi(s)ui(x))h(s)ds K(x,$)h(s)ds+ = 0 0 0 n (' + J bi(s)ui(x))h(s)ds 1( x i=1 n K(x, s)h(s)ds = 0 1 u.(x) + i=1 b.(s)h(s)ds. 1 0 i i i 17 1 Let (Gh)(x) _ G(x, s)h(s)ds 1 0 (Kh)(x) J = K(x, s)h(s)ds 0 and 1 i(h) Then (3. 5) = J bi(s)h(s)ds i = ,n 1, 2, . 0 can be rewritten as u=Gh=Kh+ ui1)i(h) i =1 and we see that the Fredholm operator (3. 1) where K (b) admits the decomposition is a Volterra operator and the functionals. Further, we note that c.(s) G K(x, s) ,i is determined by the which were given by conditions (a) and (b). are independent of the are linear But (a) and boundary conditions (3. 3). Hence K(x, s) is independent of the boundary conditions. The same remarks made about the inhomogeneous boundary conditions in Chapter II can be repeated here, so that there is no need to consider inhomogeneous boundary conditions separately. 18 CHAPTER IV SOLUTION OF THE INTEGRAL EQUATION (I -XK)u In this (4. = h +Xf.1)(u) chapter the equation (I-A G)u 1) = h (h E C) is solved assuming that the Fredholm operator G has the de- composition (4. 2) where G K = K + f is a Volterra operator, f E C, (DE C *, f tO and 1, # O. This is the decomposition considered in Chapter II. The solution obtained for (4. and an entire function in comprise all of the X . 1) is a quotient of an operator The zeros of this entire function characteristic values of G (characteristic values are inverses of eigenvalues). The resolvent operator ob- tained by solving (4. 1) exists for all noncharacteristic values Thus this resolvent and the Fredholm resolvent are equal [ X. 11, p. 15] . Also to be discussed are solutions of characteristic value problems for integral equations. Finally several examples of characteristic value problems are solved using techniques developed in this chapter. At this point it is convenient to note that the operator G is 19 an operator such that the Fredholm alternative holds for (4. follows since asserts that G(x, s) (4. 1) 1). This The Fredholm alternative is continuous. has a unique solution for arbitrary h E C iff the homogeneous equation (I-X G)u (4. 3) = 0 has only the zero solution [ 11, p. 46] are called eigenfunctions From (I-AK)u is equivalent to (4. exists for all X (4. 5) u for 1). (I-X K) - lh h K and the corresponding G. that Aft(u) + is a Volterra operator, + X (I-X K) - lft(u). it and transposing yields the following equation 4(u); (u) [1-X t(I-A K) (4. 6) Equation (4. 6) (I -X K)- 1 such that (4. 5) holds, then operating on both u 5) by Since = G 3) Thus equation (4. 4) is equivalent to . = there exists sides of (4. of (4. 1) and (4. 2) it follows (4. 4) If operator of the are called characteristic values X Nonzero solutions of (4. . lf] = has a unique solution for (I-X K) - 4(u) lh. iff 20 d(X)= Assuming 1-X (I- K) - we can solve for d(X) f 0, if f 0. and we obtain Vu) equation (4. 7) u = That is to say, if (I-K)-1 h+d() (I-XK)-lf(I-XK)-lh. d(X) which implies that contrapositive of value then d(X) value of 0 then (4. is not a X 7) = O. if Now suppose 1) that 6) since (I -X K) E -1 we assumed that If is not a characteristic X d(X) is a one to one mapping of g f 0, hence d(X) f The G. of 1) is a characteristic X holds for arbitrary h holds for arbitrary h C. that (4. is the unique solution to (4. characteristic value this statement is, Then (4. G. t O. = E C 0, which implies then c = 0 onto itself. C But To summarize, the following two theorems are recorded. Theorem 4. 1 Theorem 4. 2 X is a characteristic value of If X 1) d(X) = 0 7) gives 1). has a unique solution iff the one dimensional system (4. 6) has a unique solution. operator iff is not a characteristic value, then (4. the unique solution to (4. Note that (4. G (I -XG) Thus the Fredholm alternative for the reduces to the Fredholm alternative for the one 21 dimensional system (4. 6). From Theorems 4. and 4. 2, 1 (I -X G) -1 exists iff d(),.) t in which case (I-AG)-1= (I-XK)-1+d() (I-XK)-lf(I-aK)-1 (4.8) As K . is a Volterra operator 00 (I-XK)-1 = X nKn n=0 where K = I and Kn operator norm for all +1 X . KKn = Letting fx (4. 8) may be The series converges in the . (I-X K) - if = rewritten oo 1 n 0 n[ d(X)Kn+l X. (4. 9) (I-X G)-1 =I+ X d(%) The Fredholm resolvent operator (I -AG) -1 of F'X (I-xG)-1 whenever + fX (DKn] exists. Thus by = I+ (4. 9) G xF'x is defined by 0 22 oo X rk )Kn+ d(k 1 + fX Kn] n=0 - It might be n[ d(X ) remarked at this point that Brysk attempts to prove similar result by showing that the numerator and denominator a of his solution are the same as the numerator and denominator of the solution obtained via the Fredholm resolvent [ 2, pp. 1537- 1538] . His proof is faulty, but a proof can be established using techniques developed by Manning In [ 10] , (cf. appendix). general d(X) is an entire function in since X 00 d(X) = 1 - X (I-X K) - if = 1 n+1 - n f. n= Thus there is some difficulty in attempting to use the equation d(X) = 0 to calculate the characteristic values of it is easier to calculate determinant [11, p. 56] d(X) . G. However than to calculate the Fredholm More specifically, in making approximate calculations of characteristic values it may be easier to use a truncation of d(X) than to use a truncation -of the Fredholm determinant. Consider the characteristic value problem, (4.10) (I-X G)u = 0 . 23 By (4. 2) this may be rewritten (I-X K)u = X f0u) or u = 1. (u)X (I-X K) -1 f Thus the general form of the eigenfunctions of . G will be co (4. 11) uX = aX(I-XK)-1f = XnKnf, ax n=0 and uX will satisfy (4. 10) only if d(X) = 0. To conclude this chapter two examples of classical differen- tial eigenvalue problems are solved using the integral equation generated by the Green's function for the given eigenvalue problem. The first is the eigenvalue problem for the vibrating string problem. The second example is the heat equation in cylindrical coordinates: Bessel's equation with two boundary conditions. This boundary value problem does not have an ordinary Green's function since the coefficient of the highest derivative vanishes. However in this special case an integral equation for the eigenfunctions can be derived. Furthermore this integral equation has a kernel sidered in Chapter I. of the type con- Thus we can solve this problem by methods developed in this chapter. 24 Example 4. where Consider the vibrating string problem; 1 a2v a2 ax2 a is defined for v(x, t) t2 x< _ 0 < _ 1 and t > 0 with boundary conditions v(0, t) = v(1, t) = 0 and the initial condition v(x, 0) = g(x). Separating variables, the following boundary value problem is obtained; u" (x) (4. 12) _ -X u(0)'= u(1) (4. 13) 0<x< u(x), = 0 1, . The Green's function associated with the differential operator L = d2 with boundary conditions (4. 13) is dx G(x, s) s(x-1), 0 < s < x(s-1), 0 < x < 1 , < s < 1 . = x 25 Thus the following characteristic value problem arises; u(x) (4. 14) = ('1 J G(x, s)u(s)ds -X . 0 Using the decomposition outlined in Chapter I we obtain x u(x) = X J 1 (s-x)u(s)ds + (1-s)u(s)ds Xx 0 0 or in symbolic form u where (Ku)(x) = = XKu + Xf(1)(u) ('x j (s- x)u(s)ds, f(x) = and x (u) =5. 0 In order that there exist nontrivial d(X) u satisfying (4. = 1-X K) - if = O. Now 00 (I-X K) l f] (x) n(Knf)(x) = 1 = siAx ' n=u iD(I-XK)-lf (1- s)u(s)ds. o sufficient that [ 1 sinNFX 3/2 X 12) it is 26 and thus d (A Hence values d(A) of iff = 0 (4.12) are TA X n = = Also we have a sin(nTrx) = n a 2v is defined for av s as 1 at - as2 v(s, t) Thus the eigen- n2Tr2. = Consider the heat equation in cylindrical coordinates; av where X Thus known results are obtained. as eigenfunctions. 2 s inNFA _ or ±nTr n2Tr2. u (x) n Example 4. ) + 0 < s < 1 and t > 0 with the boundary conditions v( l, t) = v(0, t) 0, < co and the initial condition v(s, 0) = g(s). Separating variables the following boundary value problem is obtained; (4. 15) [ su'(s)] ' = -X su(s), 0 < s < 1, 27 u(1) (4. 16) = 0, u(0) < oo . Although no ordinary Green's function exists for the operator (Lu)(s) s = 0), = [su'(s)] ' a function (since the coefficient of right member of (4. 15) µ vanishes at can be found which is integrable with G(s, r) respect to the measure u" (dr) = rdr. Since s it may be reasonable to appears in the try working in this measure space. Proceeding formally, we notice that u1(s) = satisfies the boundary condition at u2(s) satisfies the boundary condition at u2 1 s = = and 0 log s Furthermore s = 1. satisfy the homogeneous equation associated with u1 (4. 15). and Car- rying out the calculations in the same spirit as suggested in Chapter II we find a function log s, H(s,r) 0 < r < s < 1, = log r, 0 < s < r < 1 . Thus formally we expect that a solution to the equation 28 (4. 17) [ su'(s)] satisfying boundary conditions ' = sh(s) (4. 16) would be 1 u(s) (4. 18) Let G(s, r) = rH(s, r). J H(s, r)h(r)rdr. 0 Then (4. 18) can be rewritten u(s) (4. 19) = ('1 = \ G(s, r)h(r)dr . 0 The kernel G(s, r) shows that u(s) Further is continuous. as given by (4. 19) satisfies a simple calculation (4. 17) as well as the Thus we expect that solutions to the boundary conditions (4. 16). characteristic value problem 1 u(s) (4. 20) G(s, r)u(r)dr _ -X 5" J 0 will give eigenfunctions for (4. 15). Clearly G(s, r) is the same type of kernel as was encountered in Chapter I (cf. equation (1. Thus we find that (4. 20) can be written s u(s) = X 1 (r log r -r log s)u(r)dr 1 0 or symbolically, + X Ç J ( 0 -r log r)u(r)dr 2)).. 29 u XKu = + X flu) s where (Ku)(s) = J (r log r - r log s)u(r)dr, f(s) = and 1 0 1 (10(u) = J ( -r log r)u(r)dr. 0 In order that there exist f uX 0 satisfying (4. 20) it is sufficient that d(X) = 1 -A.(I-XK)-1f = 0 . As before co [(I-XK)-lf] (s) Xn(Knf)(s) + = 1 n=1 By induction it can be (Knf)(s) - verified that (-1) n 2n s ... 22- 42. n = (2n)2 1,2, and therefore oo [(I-XK)-lf](s)= 1 n=1 Furthermore -ln(s)2n + 2 2 4 2 (2n)2 = JO( s) . 30 00 AflI-XK)- lf = (- 1)n(NFX )2n 2. 42. (2n)2 - ... n= and thus d(X) = JO(NFX ). As expected the squares of the eigenvalues of (4. 15) are the zeros of the Bessel function uk (s) = J0 and the eigenfunctions are ak JO(Nrik s), k = 1, 2, 31 CHAPTER V APPROXIMATE SOLUTION OF THE INTEGRAL EQUATION (I -X K)u In this chapter of the = h + X f,D(u) is approximated by truncating all (I -X G) -1 series which appear in the expression on the right side of equation (4. 9). Error estimates are calculated giving an error bound for the approximate solution. From this calculation an error bound for the approximate calculation of the characteristic values arises. Equation (4. 9) is / co (I-x G)-1 = I + X Xn[d(X)Kn+l+fX .T.Kn] n=0 d(X ) where / co (5. 1) d(X) = 1 - X X nKnf n=0 and 00 (5. 2) XnKnf fX = . n=0 Truncatingthe series which appear on the right side obtain of (4. 9) we 32 Xn[ dm(X (I-X G)-1 m = I + )Kn+ 1+fX mCKn] 0 Xn d m (A) where dm(A ) (5. 3) = 1 - X XnKnf (1. n=0 and m fXm (5. 4) IAnKnf. = n=0 Let 00 o(X)= ) X X n[ )Kn+ l d(X. fx dKn] + n=IO and m(X) the analogous truncated expression. X r' II (I-XG)1-(I-XG)mll= - o(X ) d() Then and II d- II dm a° I d-ami I = - Ild I of Idi I I I +[d1 m I m11 ' I a ml l I 33 To find an and dl I I K(x, s) error bound I I A pm K(x, n I It is and a lower bound for I Mn(x-s)n- 1 I<- s) Id!. Id-dm I, HAIL Let Then (0 < s < x < 1). < M, I I we need upper bounds for [11, p. (n-1)! 16] . easily verified that MnI,IfII IIKnfII < (5. 5) for any Thus f EC. n (5. 6) From IIKnII CX = I (5.7) X I ' I I [d(X) (I d(X I" I I ) f I< 1+ I I e m(X) m = 1 n m+ that I e -d m (X)I co where Mn, (5. 1) and (5. 5) we have I where < (IX I!M) n. 1 < = I CX M 1x1I' and from (5. 1), (5.3) and (5. IMI' IIfII"ETTl(X) n From (5. 2) and (5 5) it follows 5) 34 IIfx Again using (5. H < IIfII.eIxIM easy to show that 5) it is IIfX -fXmll <_ m() IIfIIE . Now 00 I I A I I 00 i = II X n[ dKn+ l+fx Kn] I n+1IIKn+1ll < Idl n=-1 I n=0 00 Ix I. IIfx II' + nllKnll IMI' < [ 1+2Cx] eIx IM n=0 A short calculation shows that , 00 IIA-AmII<_ IdI n=m+ 1 CO In+1IIKn+IIl + IX I' IIfx II' IMI IX InIIKnII n=m+l m + I d-d IX In+lIlKn+lll n=0 m + 'XI' IIf f nll IMI' i, n=0 Then using (5. 6) and the estimates I InIIKnII' . 35 oo Ix 2, In+lliKn+lll <Em(), n=m+ 1 m (Ix IM)n+l ek IM < (n+1)! n=0 \ m (1XIM)n LL <eIXIM n! n=0 we obtain II- mIl < (1 + 4Cx.)Em(X) . Then 2 I I G) 1 -(I-X G) 1+ 6CX +6CX 1 I I< I To complete the for Idl From dm error analysis I I d E m(A ). i a lower bound (5. 7) Idmi - Idi < I' IMI' IIfIIEm(x) lx I' IIiI. IIfIIEm(X) IX or Idi > IdmI - must be found 36 For fixed X assuming , E and further I d(X) I m (X > O. for sufficiently large m is not a characteristic value, we have X ) Since that Thus for sufficiently large m, dm(X) dm I I --- - I X + 6C II < X + _ la ml(Idml-IX I' I' HO' O' on the error. Thus I I f II0. scalar field, then we have that the I --- I E 00 we have m(X) > 0. error is 6C 2 If (5. 3) is used to compute approximate on any compact subset of the as m d(X) the final form of the 1 II(I-XG)1-(I-XG)m m -00 as 0 IIfII'Em(X)) Em(X). characteristic values (5. 7) gives a bound characteristic values can be uniformly approximated on compact sets. Due to results obtained in the appendix an error analysis developed by Glahn[6, pp. 7 two methods of analyzing the -16] also applies to this problem. error are The not directly comparable since different parameters appear in the two methods. Thus more work could be done here. 37 CHAPTER VI SOLUTION OF THE INTEGRAL EQUATIONS (I-X K)u In this h / + X f.(D.(u) chapter we shall solve the equation (I-X G)u (6. 1) where = = h (h E C) has the decomposition G n (6.2) G =K+ i=1 As before . i E C*, i K = is a Volterra operator, 1, 2, ,n and the f. i f. and E C, '. i i = 1, 2, ,n, are linearly inde- pendent. Results similar to those in Chapter IV are obtained. However, instead of the Fredholm alternative reducing to the alternative for a one dimensional system, it reduces to the alternative for an n- dimensional algebraic system. Again the solution is expressible as a quotient of an operator and an entire function of X Also similar to the case dealt with in Chapter IV, the zeros of this entire function comprise all operator G. of the characteristic values of the Finally an example is worked using the techniques . 38 of this chapter. In order to obtain the solution in a form comparable to the solution in Chapter IV, it is necessary to introduce certain notation. Let (6. 3) G where fn: n- = K fnen is defined by C r- al a 2 n fn LLLL , a.f. i i i=1 an n and ,n: C -n is defined by 1(u) 2(u) tn(u) 1)n (u) where the C. f. and For any norm in (D. are as above. Thus cn g e. g. fn1.n maps C into 39 a 1 a 2 max = { la i : + i = 1, 2, , n} an n n fn J bounded linear mapping of From (6. 3) we see C = h 1) into n into that (6. (I-X K)u (6. 4) n C 11 is a bounded linear mapping of C Hence and e fne is bounded. is equivalent to + X fne(u) and hence to (6. 5) Let u fn : g- n C = (I-k K) ih + k (I-A K) - l fne(u). be defined by MN. a 1 a2 n f n I-XK)- lf i = k i=1 an Then (6. 5) may be rewritten . is a 40 (6. 6) u (I-X K) ih = Assuming there exists a en(u) The operator = nf fn such that (6. u e on both sides of (6. 6) with (6. 7) + X en(I-X K) (u) (1)11 6) . holds, we may operat to obtain lh + X enfxn en(u) . is a linear mapping of may therefore be characterized by a matrix Ax n 7 the identity mapping of onto itself. n into itself and . Let Then (6. 7) be In may be re- written (In-Axn )en(u) (6. 8) The matrix equation (6. (In-An )- B of = l exists iff adj(In -An cofactors of ) 8) (I- K) lh . has a unique solution for dn(X) where = = det(In -Axn) f adj(In -Axn) (In -Axn ). If O. Let is the transpose of the matrix dn(X) f 0, the solution for is given by .1.n(u) B - dn(X If dn(X) f 0, then n(I- K) - lh X ) n(u) iff . .1.n(u) 41 (6. 9) u (I-ñ K) = - lh + dn(X is equivalent to (6. - lh fn BnX tn(I-X K) X ) By exactly the same reasoning as in Chapter 1). IV we have the following two theorems. Theorem 6. iff Theorem 6. 2 tion to (6. 1). dn(X) 1 If = 0 dn(X) 0, # is a characteristic value of X then equation (6. As in Chapter IV we have if (I-XG)-1 (6. 10) = (I-XK)-1 dn(X) + X dn(X It is of fact interest to note that B (6. 10) is Finally as in Chapter value of G iff dn(X) = IV we 0. gives the solu- then 0, fnnn(I-XK)-1 similar in form to see that dn(X) Now if . ) the adjoint matrix, is equal to (1) if , 9) = 1. 0, then a2 =(I-XK)1fn an are the eigenfunctions of G where the a. In is a characteristic X = n (4. 8). al ux G. are appropriate 42 constants. As an example of a we shall solve the of G the form considered in this chapter characteristic value problem which arises from the transverse oscillations of a homogeneous bar clamped at one end and free at the other. Example 6. The oscillations are determined by the equation 1 a4z a22 + where - at 2 ax 0 {(x,t): is defined on the strip z(x,t) 0 < x < 1, t > 0} and the boundary conditions are z(0,t) [ 13, pp. 26 -29 ] . = z x (0,t) = z xx (1,t) = z Assuming z(x, t) = xxx (1,t) = 0 u(x)ei Wt, we are led to the ordinary differential equation (6. 11) d 4 ux) A 4u(x), 0< x< 1 , dx with the boundary conditions (6. 12) u(0) = u'(0) = u"(1) = u"'(1) The Green's function for the operator L = = 0. d 4 dx 4 with boundary 43 conditions (6. 12) is 2x 3s x - s3 G(x, s) - - x <- 1 , x < s 1 . 0 < s < 3! = 3 sx2 - x3 0 < 3! We have the following integral equation for < u; 1 u(x) = X 4 G(x, s)u(s)ds 11 . 0 Decomposing the integral operator, we obtain X (6. 13) u(x) - X 4 0 (x-s)3 u(s)ds 3! = x4 1 { su(s)ds J 0 He re it may be of some treatment of J 2 2! 3 1 + [ - x u(s)ds] } . o interest to briefly mention Tricomi's this problem so that similarities in both methods can be compared . Initially, Tricorni develops a method by which the solution to an initial value problem can be written as a Volterra equation [ 13, pp. 18 -19] that u" (0) = c . In the and case u "'(0) of equation (6. 11), = c3. Tricorni assumes Then by the above mentioned method for solving initial value problems, he arrives at the equation 44 u(x)-X (6. 14) [ 13, p. 2 8] ('x 4 j0 3S) 3 u(s)ds = X 4( c2x2 2! + c3x3/ 3! ) These two methods are connected by the "shooting" . method mentioned by Henrici [7, pp. 345 -346] . Briefly the "shooting" method assumes that every boundary value problem is equivalent to an initial value problem. The solution to the boundary value problem is represented as a parameterized solution to the initial value problem (e. are c and Now equation (6. 14) where the parameters Then the parameters are determined. c 2). method used to obtain (6. functionals of g. 13) The gives these parameters as linear u. let us proceed to the problem of finding the character- rewritten as istic values. Let (6. 13) be (6. 15) (I -X4K)u = A4f2cb 2(u) where a f2 a2l = al x2,2! + a2 x3/3! and 1 su(s)ds (u) 1, 1 2(u) = ('0 2(u) -J u(s)ds 0 45 There exists a solution - 1 2 d (X) = A if u d2(X) = 41(I-X 4 K) -1 x2/2! where 0, _ - A 41(I-X 4K) -1 x3/ 3! det -X 41,2(I -x K) -1 x2/2! 1 -X 412(I -X 4K) -1 1 1-X 2/2 = 1 l 1 s(coshX s-cosX.$)ds (-X)/2 Ç J0 s(sinhXs-sinT.$)ds det 1 X 1 2/2 J (coshXs-cosXs)ds 1+X/2 0 = x3/3! (sinhXs-sinXs)ds 0 2(1 +coshX cos X). Hence we obtain the anticipated result that the characteristic values are the zeros of the 1 + transcendental equation coshX cos X = 0 [ 13, p. 29] . 46 BIBLIOGRAPHY 1. Allis, W. P. and Philip M. Morse. The effect of exchange on the scattering of slow neutrons from atoms. Physical Review 44: 269-27 6. 1933. 2. Brysk, Henry. Determinantal solution of the Fredholm equation with Green's function kernel. Journal of Mathematical Physics 4: 153 6 -1538. 1963. 3. Coddington, Earl A. An introduction to ordinary differential equations. Englewood Cliffs, N. J. , Prentice Hall, 1961. 292 p. 4. Davis, H.T. The theory of linear operators. Bloomington, Indiana, Principia Press, 1936. 628 p. 5. Drukarev, 6. Glahn, Thomas Leroy. An error bound for an iterative method of solving Fredholm integral equations. Master's thesis. Corvallis, Oregon State University, 1954. 22 numb. leaves. 7. Henrici, Peter. Discrete variable methods in ordinary differential equations. New York, Wiley, 1962. 407 p. 8. Ince, E. L. Ordinary differential equations. New York, Longmans, Green and Company, 1927. 558 p. 9. Indritz, Jack. Methods in analysis. New York, Macmillan, G. F. The theory of collisions of electrons and atoms. Soviet Physics. JETP 4: 309 -320. 1957. 1963. 481 p. 10. Manning, Irwin. A theorem on the determinantal solution of the Fredholm equation. Journal of Mathematical Physics 5: 1223 -1225. 1964. 11. Miklin, S. G. 12. Morse, Philip M. and Herman Feshbach. Methods of theoretical physics. 2 vol. New York, McGraw -Hill, 1953. 1978 p. Integral equations and their applications to certain problems in mechanics, mathematical physics and technology. New York, Pergamon Press, 1957. 338 p. 47 13. Tricorni, Francesco. Integral equations. New York, Interscience, 1957. 238 p. APPENDICES 48 APPENDIX COMPARISON OF THE FREDHOLM RESOLVENT OPERATOR AND EQUATION (4. 9) As remarked in Chapter IV, Brysk's assertion can be proved using tools developed by Manning. This result can be used to show that the numerator and denominator of co (1) X n[ d(X )Kn+ l+fX IKn] / d(X ) nLL=LO n=0 are equal to the numerator and denominator respectively Fredholm resolvent operator. Theorems immediately from the equality 4. of (1) and the 1 and 4. 2 Fredholm resolvent The Fredholm resolvent operator is given by r (2) _ nDn) d0 = D0 1, = / ( X ndn ) n=0 n=0 where / oo X ( G, 1 dn and the D n = the then follow operator. 00 of (-1)n-1 J Dn-1(s, s)ds, n 0 = 1, 2, are given by the recursion relation , Dn (3) [ dnG 10, pp. 1223 - 1224; 11, p. 54] + GDn- n l' = 49 , 1,2, Brysk's assertion was that . co (4) X nGKnf /d(X ) n=0 and r were identical f [ 2, pp. 1537 -1538] obtained from (1) by operating on f , (note that (4) is and simplifying). developed by Manning are stated below in Lemmas 1 proof. Lemma {Gh} 1 where {h} = for Ch) lim h(x) /f(x) x Lemma = hEC . 0+ dn= - {Dn- if }, 2 n = 1,2, These tools yield Theorem 1 do +l=- cIKnf, n = n = 0,1,, and GKnf = D f, n 0,1, . .. The tools and 2 without 50 Proof: (by induction) For GKof = Gf = n D0f dn+2 {Dn+lf} = {dn = {- 13(Knf)Gf = GDnf} + +1f - {Gf} 13(Knf)Cf) + = - c1,K0f, and 1 f ) (by the inductive hypothesis) {G(GKnf)} + .13.(GKnf) - (D(Knf)(1) (f) + Kn+ (by (3) G(GKnf)} + -''(Knf) {Gf} = - = = 2 = = Also by - {D0f} = . By Lemma - dl 0, = [ (by Lemma 1) (by equation (4. 2) (K +f (D)Knf] ) . (3) Dn+ 1f + GDnf = dn+ 1Gf = -(1)(Knf)Gf = G(G- f')Knf = GK Knf = GKn+ 1 + G(GKnf) (by the inductive hypothesis) (by equation (4. 2) ) 51 Hence the conclusion follows by induction. An interesting conclusion which may be drawn from Theorem is 1 00 d(A dnX n ) = . n n=0 Using this we can write co 00 X n[d(X)Kn+ l+fX Kn] p P[ = Then defining D' D' mKp-m+1 + Kmf CKP-m)] m=0 p=0 n=0 d by d = P m Kp-m+l elf (KP-m), m=0 it is easy to verify that D0 = G and that D' P satisfies the recur - sion relation (3). Hence (1) is identical to (2) and thus Theorems 4. 1 and 4. 2 follow immediately since these properties hold for the Fredholm resolvent operator. This approach while being more tedious (the proof of Lemma 2 as proved by Manning is largely bookkeeping) yields the more satisfying result that do o +1 (1.Knf and D = P D' P . This conclusion cannot be drawn from the remarks in Chapter IV leading up to the 52 equality of the expression given in (1) and rx . The method given in this thesis of solving equation (4. does 1) give greater insight into the nature of the decomposition and how it is used to solve the problem. Also it generalizes to operators with the type of decomposition considered in Chapter VI with greater ease (once the notation was developed the proof of Theorems and 6. 2 were identical to the proofs of Theorems 4. 1 6. and 4. 2). 1