NetNews Usenet Archive 1992 #23/NN_1992_23.iso/spool/sci/math/13246

Internet Message Format | 1992-10-15 | 1.6 KB

Xref: sparky sci.math:13246 sci.math.num-analysis:3016
Path: sparky!uunet!mcsun!uknet!edcastle!hwcs!pjbk
From: [e-mail] (Peter JB King)
Newsgroups: sci.math,sci.math.num-analysis
Subject: Optimal omega for SOR
Keywords: successive over relaxation
Message-ID: <[e-mail]>
Date: 15 Oct 92 13:50:39 GMT
Sender: [e-mail] (News Administrator)
Organization: Dept of Computing & Electrical Engineering, Heriot-Watt University, Scotland
Lines: 23

Does anyone have any good references for dynamic selection of the
optimal omega for successive over relaxation?  I am aware of the
theoretical optimum, but since this depends on the eigenvalues, the
computation of this is likely to be at least as expensive as solving the
equations with a non optimal omega.

The problem which I'm solving involves arrays that are too large to
keep in core, so an iterative method is essential.  Also the problem
gets bigger every week, with new equations being added, and the old
solutions can be used as first estimates for the new solutions.  It may
well be that the optimum omega changes too, so an algorithm that would
adapt from any starting value for omega in range 0-2 would be best.

I have tried using Aitken acceleration on the elements of the solution
vector directly, but it seems to reduce the rate of convergence!

Literature references, code segments or good ideas welcomed.

--
Peter King, Computer Science Department    JANET:    [e-mail]
  Heriot-Watt University        ARPA:    [e-mail]
  Riccarton, Edinburgh EH14 4AS     UUCP:    ..!uknet!cs.hw.ac.uk!pjbk
Phone: [phone]                Fax: [phone]