commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From S├ębastien Brisard <sebastien.bris...@m4x.org>
Subject [math] Testing for convergence in iterative algorithms
Date Mon, 08 Aug 2011 18:46:47 GMT
Hi,
discussions regarding the development of iterative linear solvers
(MATH-581) triggered the following issue. Usually, a default
(efficiently implemented) convergence checker is provided for each
standard linear iterative solver. However, it would be very useful to
have the ability to use a custom convergence checker. This is very
similar to what is done in o.a.c.m.optimization. Only,
ConvergenceChecker is too specific in my view, since the decision on
convergence is based on the last two iterations. While a perfectly
valid approach, this is not the only one, and I was wondering whether
we could be a little bit more general ?
For example, in iterative linear solvers, the convergence test is
(often) based on the residual r = b - Ax, not on the difference
between newX and oldX.
Thanks in advance for your thoughts on this topic.

Sebastien

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org


Mime
View raw message