1.2 Fixed-Point Iteration Consider a sequence of point which is computed by a formula of the form xn+1 = F ( xn ) ( n ³ 0) . (1) The algorithm defined by such an equation is called functional iteration. Formula (1) can be used to generate sequences that do not converge. For example, the sequence 1, 3,9,27, which arises if x0 = 1 and F ( x ) = 3x . However, our interest lies mainly in those cases where lim xn = s exists. n®¥ Suppose that lim xn = s . n®¥ What is the relation between s and ( F ? If F is continuous, then ) F ( s ) = F lim xn = lim F ( xn ) = lim xn+1 = s n®¥ n®¥ n®¥ Thus, F ( s ) = s , and we will call s a fixed point of the function F. Definition 1.4 (Fixed Point) The real number s is a fixed point of the function F(s) = s F if We could think of a fixed point as a value that the function “locks onto” in the iterative process. Based on formula (1), we have the following algorithm Fixed-Point Iteration (FPI) x0 = initial guess For i = 0,1,2,… xi+1 = F(xi ) end for 1 A mathematical problem often can be reduced to the problem of finding a fixed point of a function. Very interesting applications occur in differential equations, optimization theory, and other areas. Usually the function F whose fixed points are sought will be a mapping from one vector space into another. We intend to analyze the simplest case when F maps closed set C Í R into itself. Definition (Contractive Mapping) A mapping (or function) there exists a number l less than 1 such that () F is said to be contractive if ( ) F x - F y £ l | x - y |. For all points x and y in the domain of (2) F. As illustrates in following figure, the distance between a and b is mapped into a small distance between F ( a ) and F ( b ) by the contractive function F . y F (b) F (a) x a b Theorem (Contractive mapping Theorem) Let C be a closed subset of the real numbers. If F is a contractive mapping of C into C , then F has a unique fixed point. Moreover, this fixed point is the limit of every sequence obtained from equation (1) with a starting point x0 ÎC . Proof We use the contractive property (2) together with equation (1) to write xn - xn-1 = F ( xn-1 ) - F ( xn- 2 ) £ l xn-1 - xn-2 . (3) This argument can be repeated to get xn - xn-1 £ l xn-1 - xn-2 £ l 2 xn-2 - xn-3 £ £ l n-1 x1 - x0 . Since xn can be written as the form xn - x0 = ( x1 - x0 ) + ( x2 - x1 ) + + ( xn - xn-1 ) . 2 ¥ We see that the sequence {xn } converges if and only if the series å( x n =1 n - xn -1 ) converges. To prove that this series converges, it suffices to prove that the series ¥ å( x n =1 n - xn -1 ) converges. This is easy because we can use the comparison test and the previous work: ¥ ¥ n=1 n=1 å xn - xn-1 £ å l n-1 x1 - x0 = 1 x1 - x0 . 1- l Since the sequence converges, let s = lim xn , then F ( s ) = s , as noted previously. (Observe n®¥ that the contractive property implies continuity of point, if x and y are fixed points, then F ). As for the uniqueness of the fixed x - y = F ( x ) - F ( y) £ l x - y . since l < 1, x - y = 0 . Finally, note that the point s that is obtained belongs to C since s is the limit of a sequence lying in C . Example: Show that the sequence {xn } in the contractive Mapping Theorem satisfies the Cauchy criterion for convergence. Solution. Recall the Cauchy criterion for the sequence {xn }: Given any q , there exists an xn - xm < e whenever n, m ³ N . If n ³ m ³ N , then by the integer N such that triangle inequality and the equation following (3), xn - xm £ xn - xn-1 + xn-1 - xn-2 + + xn+1 - xn £ l n-1 x1 - x0 + l n-2 x1 - x0 + +l m x1 - x0 ( + l n-1- m ( )= l £ l m x1 - x0 1 + l + l 2 + £ l N x1 - x0 1 + l + l 2 + For any q > 0 , there exists an increase N ) x1 - x0 (1 - l ) . -1 N such that xn - xm < e whenever n, m ³ N . We just N until l N x1 - x0 (1 - l ) < e . -1 3 Example: Prove that the sequence defined recursively as follows is convergent ì x0 = -15 ï 1 í ïî xn +1 = 3 - 2 xn Solution the function F ( x ) = 3 - ( n ³ 0 ). 1 x is a contraction because 2 F ( x ) - F ( y) = 3 - 1 1 1 1 x - 3+ y = y - x £ y - x 2 2 2 2 by the triangle inequality. By Theorem, the sequence described must converge to the unique fixed point of F , which is readily seen to be. In the theorem, we can take C to be R . Example: Use the Contractive Mapping Theorem to compute a fixed point of the function 1 F ( x ) = 4 + sin 2x . 3 Solution By the mean-value Theorem, we have F ( x ) - F ( y) = 1 2 2 sin 2x - sin 2y = cos 2s x - y £ x - y . 3 3 3 for some s between x and y . This shows that F is contractive, with l = F has a fixed point. Theorem 1, 2 .. By 3 Error analysis (Linear convergence of Fixed-Point Iteration) Let us start with the definition of linear convergence: Definition (Linear Convergence) Let en denote the error at step n of an iterative method. If lim n®¥ en+1 = S <1 en the method is said to obey linear convergence with S . Now let us analyze the errors in the process of functional iteration. We suppose that F has a fixed point s and that a sequence {xn } has been defined by the formula x n+1 = F ( x n ) . Let en = xn - s , 4 If F ¢ exists and is continuous, then by the mean-value theorem, xn+1 - s = F ( xn ) - F ( s) = F ¢ (xn ) ( xn - s) , en+1 = F ¢ (xn ) en Or where xn is a point between xn and s . The condition F ¢ ( x ) < 1 for all x ensures that the error decrease in magnitude. If en is small, then xn is near s and F ' (xn ) » F ' ( s) . One would expect rapid convergence if F ¢ ( s ) is small. An ideal situation would be F ¢ ( s) = 0 . F is Theorem 1.6 (Linear Convergence of Fixed-Point iteration) Assume that continuously differentiable, that F(s) = s , and that S =| F '(s) |< 1 . Then Fixed-Point Iteration converges linearly with rate S to the fixed point close to s . s for initial guesses sufficiently Definition 1.7 (Locally Convergent) An iterative method is called locally convergent to method converges to s for initial guesses sufficiently close to s . Example: Explain why the Fixed-Point Iteration s if the F(x) = cos x converges. Example: Find the fixed points of F(x) = 2.8x - x 2 . Example: Calculate 2 by using FPI. 5