7.3 The Jacobi and Gauss-Seidel Iterative Methods (cont’d) The Gauss-Seidel Method. For each ( ) generate the components ( ) [ ∑( ) ( ∑( ) ( ) ) of ( ) from ( ) by ] Namely, ( ) ( ( ) ) ( ) ( ( ( ) ) ) ( ( ) ) ( ) Matrix form of Gauss-Seidel method. ( ( ) ( Define ) ) ( ( ) ( ) ( and ( ) ) ( ( ) ) ( , Gauss-Seidel Step 4 If || ( || ) ∑ ( ) , then OUTPUT ( method can be written as ) Numerical Algorithm of Gauss-Seidel Method ( ) Input: , , tolerance TOL, maximum number of iterations Step 1 Set Step 2 while ( ) do Steps 3-6 Step 3 For [ ∑ ) . ] ); 1 STOP. Step 5 Set Step 6 For Set Step 7 OUTPUT ( STOP. ); Convergence theorems of the iteration methods Let the iteration ( ) ( Lemma 7.18 If the spectral radius satisfies ( ) ( Theorem 7.19 For any ( ) , then ( ( Since ( ) ) ( written as exists, and ∑ ( ) , the sequence defined by ) if and only if ( ) converges to the unique solution of ( ) ) be ) ( ) Proof (only show ( ) method ) is sufficient condition) ( ( ) ( ) ) ( ) ( ) ( ) (∑ ) ( ) 2 Corollary 7.20 If || || ( ) ( ) for any natural matrix norm and converges, for any (i) || ( ) || (ii) || ( ) || Theorem 7.21 If sequences ( ) || || || || || || || || ( ) ( ) || ( ) ( ) is a given vector, then the sequence , to a vector with ( ) defined by , and the following error bound hold: || is strictly diagonally dominant, then for any choice of that converges to the unique solution of . ( ) , both the Jacobi and Gauss-Seidel methods give Rate of Convergence Corollary 7.20 (i) implies || ( ) || Theorem 7.22 (Stein-Rosenberg) If following statements holds: (i) (ii) (iii) (iv) ( ) || ( ) for each || and , for each , then one and only one of ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) 3 7.4 Relaxation Techniques for Solving Linear Systems Definition Suppose ̃ is an approximation to the solution of the linear system defined by ̃ with respect to this system is ̃. . The residual vector for Objective of accelerating convergence: Let residual vector converge to 0 rapidly. In Gauss-Seidel method, we first ( ) associate ( ) ( ( with ) ( ) each ( ) calculation ( ) ) of an approximate component to the solution a residual vector ( ) The ith component of ( ) ( ) ( ) ( ( ) ( ) ) is ( ) ∑( ) ∑ ( ) ( ) ( ) ( ) ) ( ) so ( Also, ( ) ( ) ∑( ) ∑ ( ( ) ) is computed by ( ) [ ∑( ( ) ) ( ∑( ) ) ( ) ] Therefore ( ) ( ) ( ) 4 Gauss-Seidel method is characterized by ( ) Now consider ( ( ) ) the ( ) residual ( ) ( ) The ith component of ( ), ( ( ) ( ) ( ) ( associated ) ( ) with the vector ) is ( ) By ( ) vector ∑( ( ) ) ( ∑ ( ) ) ( ) ( ) Idea of Successive Over-Relaxation (SOR) (technique to accelerate convergence) ( ) to Modify ( ) so that norm of residual vector ( ( ) ) ( ) ( ) converges to 0 rapidly. Here Under-relaxation method when Over-relaxation method when Use ( ) and ( ) ( ( ), ) ( ) [ ∑( ( ) ) ∑ ( ( ) )] ( ) 5 Matrix form of SOR Rewrite Eq. (5) as ( ) ∑( ( ) ( ) ( ) ( ) ( ( ) ( Define ) SOR can be written as Example Use SOR with with ( ) ( ( ) ( ) ( , ( ) ( ( ) ( ) ∑ ( ( ) ( ) ) ( ) ) ) ( ) ) ) to solve ) 6