[
https://issues.apache.org/jira/browse/MATH-896?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13495301#comment-13495301
]
Luc Maisonobe commented on MATH-896:
------------------------------------
Yes, this value is used only for an initial check in this algorithm.
The implementation of the loop corresponds to the original Brent algorithm, not
to something specific created for the Apache Commons Math library. Our
implementation of a classical algorithm should be as close as possible to the
reference. Any departure from it would prevent comparison with other
implementations and would only confuse users.
The value is inherited from an higher level base class and other
implementations with different algorithms exist, which do use this parameter in
the convergence loop. Examples are the secant based solvers and also
BracketingNthOrderBrentSolver.
For most purposes, BracketingNthOrderBrentSolver is the recommended solver, it
is often the most efficient one (typically with a setting for order 5) and has
interesting features (bracketing).
Also note that the JIRA issue tracker is not the place for questions, it is a
place for bug reports and features requests. Usage questions should be asked on
the user list (see [http://commons.apache.org/mail-lists.html]). So if you
agree, I will close this report as NOTAPROBLEM.
> BrentSolver : function value accuracy convergence criterion
> -----------------------------------------------------------
>
> Key: MATH-896
> URL: https://issues.apache.org/jira/browse/MATH-896
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 3.0
> Reporter: Vincent Ruch
> Priority: Minor
>
> Hi everybody,
> In the current implementation of the Brent solver, it seems that attribute
> "functionValueAccuracy" is only used in the "doSolve" method, before entering
> the iteration loop in "brent" method.
> Shouldn't it be a convergence test on the current function value to ensure
> that function value accuracy is reached?
> This is the first time I'm posting here, so maybe I have missed something.
> Sorry if it's the case.
> Thanks for your time,
> Vincent
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
