This preview shows page 1. Sign up to view the full content.
Unformatted text preview: venience as
2
32
32
3
C1 D1
x1
b1
6D C D
7 6 x2 7 6 b2 7
76
Ax = 6 2 2 . . 2
6
76 . 7 = 6 . 7
4
54 . 7 6 . 7
.
.54.5 DK ;1 CK ;1 xK ;1 bK ;1 The matrices Ck, etc. were de ned by (8.1.4c8.1.4d). The SOR procedure is
applied to an entire row, i.e., we compute
)
^
Ck x(k +1) = ;Dk x(k;+11) ; Dk x(k+1 + bk ^
x(k +1) = !x(k +1) + (1 ; !)x(k ) k = 1 2 ::: K ; 1 = 0 1 ::: : Thus, we have to solve a tridiagonal system at each step of the process. This
p
procedure converges faster than point SOR iteration by a factor of 2.
4. Alternating direction implicit (ADI) methods. By considering an elliptic problem as
the steady state limit of a transient parabolic problem, we can use some methods for
timedependent problems to solve them. In particular, the ADI method (cf. Section
5.2) has been adapted to the solution of elliptic problems. The goal, when using this
approach, is to select the \time step" so that the ADI scheme converges to steady
state as fast as possible. The ADI method with a single acceleration parameter
(arti cial time step) has the same convergence rate as SOR. Further acceleration
is possible by choosing a sequence of arti cial time steps, changing them after each
predictorcorrector sweep, and applying them cyclically. Wachspress 7] has shown
how to select nearly optimal acceleration parameters.
The results shown in Table 9.2.5 summarize the convergence rates of the methods that
we have studied in this section. They are all obtained by solving a Dirichlet problem on a 9.2. Basic Iterative Solution Methods 29 square mesh with uniform spacing h = x = y. The convergence rates of all methods
decline as the mesh spacing decreases. Degradation in performance is least with the
ADI method however, computing optimal parameters can be problematical in realistic
situations.
Method
Conv. Rate
Jacobi
h2
GaussSeidel
2h2
SOR (with !OPT )
2h
8
ADI ( with m parameters) m ( h )1=m
2
Table 9.2.5: Convergence rates for various iterative methods as a function of mesh spacing
h for a Dirichlet problem on...
View Full
Document
 Spring '14
 JosephE.Flaherty

Click to edit the document details