Math 346 Lecture #7 7.1 Contraction Mapping Principle Definition 7.1.1. For a nonempty set X and a function f : X ! X, a pointx ¯ 2 X is called a fixed point of f if f(¯x) =x ¯. Recall the One-Dimensional Brouwer Fixed Point Theorem: if f : [0; 1] ! [0; 1] is con- tinuous, then there existsx ¯ 2 [0; 1] such that f(¯x) =x ¯, i.e., f has a fixed point, although it may have more than one fixed point. There are other kinds of continuous functions that have fixed points. Definition 7.1.3. Let (X; k · k) be a normed linear space, and D a nonempty subset of X. A function f : D ! D is called a contraction mapping if there exists 0 ≤ k < 1 such that for all x; y 2 D there holds kf(x) − f(y)k ≤ kkx − yk: Remark 7.1.4. A contraction mapping f : D ! D is Lipschitz continuous on D with constant k, and hence continuous on D. Example 7.1.6. For µ 2 [0; 1] define T : [0; 1] ! [0; 1] by T (x) = 4µx(1 − x): That the codomain of T is correct for each µ 2 [0; 1] follows from the unique maximum of T occurring at x = 1=2 when µ > 0, so that the maximum value of T is 4µ(1=4) = µ. The function T , known as the logistic map, is an iterative model in population dynamics, i.e., starting with a population percentage of x0, the next population percentage is x1 = 2 T (x0), the next is x2 = T (x1) = T (x0), etc. To show that T is contraction mapping for some µ, we start with inequality j1 − x − yj ≤ 1 that holds for all x; y 2 [0; 1]. Then jT (x) − T (y)j = j4µx(1 − x) − 4µy(1 − y)j = 4µjx − x2 − y + y2j = 4µjx − y − (x2 − y2)j = 4µjx − y − (x − y)(x + y)j = 4µj(x − y)(1 − (x + y))j = 4µj(x − y)(1 − x − y)j = 4µjx − yj j1 − x − yj ≤ 4µjx − yj: This implies that T is contraction mapping when 4µ < 1, i.e., when µ 2 [0; 1=4). Theorem 7.1.8 (Contraction Mapping Principle). For a nonempty closed subset D of a Banach space (X; k · k), if f : D ! D is a contraction mapping, then there exists a unique fixed point ¯x 2 D for f. 1 Proof. For a fixed x0 2 D we form a sequence (xn)n=0 iteratively by xn+1 = f(xn); n = 0; 1; 2; 3;:::: 1 We show that the sequence (xn)n=0 is Cauchy in X. Let k 2 [0; 1) be the constant for the contraction mapping f : D ! D, i.e., kf(x) − f(y)k ≤ kkx − yk for all x; y 2 D. Then for n ≥ 2 we have kxn − xn−1k = kf(xn−1) − f(xn−2)k ≤ kkxn−1 − xn−2k = kkf(xn−2) − f(xn−3)k 2 ≤ k kxn−2 − xn−3k . n−2 = k kx2 − x1k n−2 = k kf(x1) − f(x0)k n−1 ≤ k kx1 − x0k: Hence for n > m ≥ 0 we have kxn − xmk = kxn − xn−1 + xn−1 − xn−2 + ··· + xm+2 − xm+1 + xm+1 − xmk ≤ kxn − xn−1k + kxn−1 − xn−2k + ··· + kxm+2 − xm+1k + kxm+1 − xmk n−1 n−2 m+1 m ≤ k kx1 − x0k + k kx1 − x0k + ··· + k kx1 − x0k + k kx1 − x0k m 2 n−m−1 = k (1 + k + k + ··· + k )kx1 − x0k: We get an upper bound on 1+k+k2 +···+kn−m−1 using the geometric series in k 2 [0; 1), i.e., 1 X 1 kl = 1 − k l=0 and that k ≥ 0, so that 1 X 1 1 + k + k2 + ··· + kn−m−1 ≤ kl = : 1 − k l=0 Thus for n > m ≥ 0 we have km kx − x k ≤ kx − x k: n m 1 − k 1 0 Since k 2 [0; 1), we have km ! 0 as m ! 1. 1 To see that (xn)n=0 is Cauchy in X, for > 0 we choose N 2 N so that kN kx − x k < , 1 − k 1 0 for then kN kx − x k ≤ kx − x k < . n m 1 − k 1 0 1 The completeness of X implies that the Cauchy sequence (xn)n=0 converges to an element ¯x 2 X. 1 Since D is closed and (xn)n=0 ⊂ D, the limit point ¯xbelongs to D. To see thatx ¯ is a fixed point of f we have for > 0 the existence of N 2 N such that kxn − ¯xk < /2 for all n ≥ N; hence for all n ≥ N we have kf(¯x) − x¯k = kf(¯x) − f(xn) + f(xn) − ¯xk ≤ kf(¯x) − f(xn)k + kf(xn) − ¯xk ≤ kk¯x − xnk + kxn+1 − ¯xk k ≤ + 2 2 < + 2 2 = . Since > 0 is arbitrary, we obtain kf(¯x) − ¯xk = 0, i.e., f(¯x)= ¯x. To show the uniqueness of the fixed point ¯xsuppose there was another fixed point ¯yof f in D. Then k¯x − ¯yk = kf(¯x) − f(¯y)k ≤ kk¯x − ¯yk: This implies, since 0 ≤ k < 1 that k¯x − ¯yk = 0, i.e., ¯x= ¯y. Remark 7.1.9. The contraction mapping principle also applies to contraction mappings f : D ! D on a closed subset D of a complete metric space (X; d) by replacing every occurrence of kx − yk with d(x; y), etc., in the definition of contraction mapping and the proof of the Contraction Mapping Principle. Remark 7.1.10. The proof of the Contraction Mapping Principle uses a technique called the method of successive approximations for finding the fixed point. 1 No matter the initial guess x0 for the fixed point, the sequence (xn)n=1 determined iter- atively by xn+1 = f(xn) always converges to the unique fixed point ¯xof f on D. th An bound on the error of the m approximation xm for ¯xis obtained from the Cauchy estimate, km k¯x − xmk = lim kxn − xmk ≤ kx1 − x0k; n!1 1 − k where we have pulled the limit outside the norm by its the continuity. Vista 7.1.11. The contraction mapping principle is used to prove existence of solutions to many equations, such as initial value problems for systems of first-order differential equations (coming your way in Volume 4, if not sooner in disguise). Example 7.1.12. An algorithm for finding the positive square roots of real numbers greater than one uses the method of successive approximations. For c2 = b ≥ 1 the number c ≥ 1 (the positive square root of c2) is a fixed point of the function f : (0; 1) ! (0; 1) defined by 1 b f(x) = x + 2 x because 1 b 1 c2 f(c) = c + = c + = c: 2 c 2 c The function f is not a contraction mapping on (0; 1): for x = 2=n and y = 1=n we have 1 2 bn 1 1 bn 2 bn2 2 − bn2 f(x) − f(y) = + − − bn = − = − = 2 n 2 n 2n 4 4n 4n 4n so that for all sufficiently large n we have bn2 − 2 jf(2=n) − f(1=n)j = 4n while 2 1 1 − = ; n n n so that for all sufficiently large n we have 2 bn − 4 1 2 1 jf(2=n) − f(1=n)j = > = − : 4n n n n But f is a contraction mapping on the closed interval [pb=2; 1). To see this we start with 1 b b 1 b(y − x) 1 b jf(x) − f(y)j = x − y + − = x − y + = (x − y) 1 − : 2 x y 2 xy 2 xy When x; y ≥ pb=2, then xy ≥ b=2 and −2 ≤ −b=xy, so that b −1 = 1 − 2 ≤ 1 − ≤ 1; xy whence 1 b 1 jf(x) − f(y)j = jx − yj − 1 ≤ jx − yj: 2 xy 2 Thus f is a contraction mapping on [pb=2; 1) with constant k = 1=2. p 1 Taking any initial guess x0 ≥ b=2 gives a sequence (xn)n=0 determined by xn+1 = f(xn) that converges (quite rapidly because k = 1=2) to the square root c. Example (in lieu of 7.1.13). A subset of the Banach space (C([0; 1]; R); k · k1) is the set D = fx(t) 2 C([0; 1]; R): x(0) = 0g: This D is a subspace because for x(t); y(t) 2 D we have ax(0)+by(0) = 0 for all a; b 2 R. 1 This D is also closed because for any convergent sequence (xn(t))n=1 ⊂ D, the limit function x(t) 2 C([0; 1]; R) satisfies x(0) = lim xn(0) = 0. Consider the operator F : D ! D defined by Z t F (x)(t) = f(x(τ)) dτ; t 2 [0; 1]; 0 where f : R ! R is Lipschitz with constant 0 ≤ L < 1, i.e., for all x; y 2 R there holds jf(x) − f(y)j ≤ Ljx − yj: That F (x)(t) 2 D when x(t) 2 D follows because the continuity of f(x(τ)) on [0; 1] R t implies the continuity of F (x)(t) = 0 f(x(τ)) dτ on [0; 1] (see Theorem 5.10.12(v)), and because F (x)(0) = 0. We show that F is a contraction mapping on D. For x(t); y(t) 2 D we have Z 1 jF (x)(t) − F (y)(t)j = f(x(τ)) − f(y(τ)) dt 0 Z t ≤ jf(x(τ)) − f(y(τ))j dt 0 Z t ≤ Ljx(τ) − y(τ)j dt 0 Z 1 ≤ Ljx(τ) − y(τ)j dt 0 Z 1 ≤ Lkx − yk1 0 = Lkx − yk1: This implies that kF (x) − F (y)k1 = sup jF (x)(t) − F (y)(t)j ≤ Lkx − yk1: t2[0;1] Since 0 ≤ L < 1, the operator F is a contraction mapping on D.