# 49854 074927 1 0 000146 000073 0 01 1 1 1 0 0 0 0

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: on (right). Example 9.2.5. We'll try to quantify the di erences in the convergence rates of Jacobi and Gauss-Seidel iteration for Poisson's equation on a rectangle. Jacobi's method satis es (9.2.12) and we let p be an eigenvector of MJ with corresponding eigenvalue thus, MJ p = p: (9.2.17a) Using (9.2.12), we see that the component form of this relation is pjk = x (pj;1 k + pj+1 k) + y (pj k;1 + pj k+1) j = 1 2 ::: J ; 1 k = 1 2 ::: K ;1 (9.2.17b) where pjk is a component of p. (The double subscript notation for a vector component is non-standard, but convenient in this case since it corresponds to a position in the nite di erence mesh.) One may easily verify that pjk = sin mj sin nk J K (9.2.17c) 16 Solution Techniques for Elliptic Problems and = 1 ; 4 x sin2 m ; 4 y sin2 n 2J 2K m = 1 2 ::: J ; 1 n = 1 2 : : : K ; 1: (9.2.17d) Remark 1. The (J ; 1)(K ; 1) eigenvectors and eigenvalues of MJ are indexed by m and n. Remark 2. The eigenvector p is the eigenfunction of the Laplacian sampled at the mesh points j = 1 2 : : : J ; 1, k = 1 2 : : : K ; 1. The eigenvalue is, however, not an eigenvalue of the Laplacian. The largest eigenvalue and, hence, the spectral radius of MJ may be obtained by setting m = n = 1 in (9.2.17d) to obtain (MJ ) = 1 ; 4 x sin2 2J ; 4 y sin2 2K : In the special case of a square, x = y (9.2.17e) = 1=4 and J = K thus, (MJ ) = 1 ; 2 sin2 2J : For large values of J , we may approximate this as (MJ ) 1 ; 2 2J 2 = 1 ; C x2 since J = a= x for an a a square region. Thus, the spectral radius approaches unity and the convergence rate slows as J increases (or as x decreases). In a similar manner, Let q and be an eigenvector-eigenvector pair of the GaussSeidel iteration matrix MGS for Poisson's equation on a rectangle (9.2.16). Thus, using (9.2.15b) MGS q = (D ; L);1 Uq = q (9.2.18a) or, using (9.2.15b), in component form qjk ; ( xqj;1 k + y qj k;1) = ( xqj+1 k + y qj k+1) j = 1 2 ::: J ;1 k = 1 2 : : : K ; 1: (9.2.18b) 9.2. Basic Iterative Solution Methods 17 This problem appea...
View Full Document

• Spring '14
• JosephE.Flaherty
• Articles with example pseudocode, Gauss–Seidel method, Jacobi method, Iterative method, elliptic problems

{[ snackBarMessage ]}

Ask a homework question - tutors are online