[
https://issues.apache.org/jira/browse/MATH-1137?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14063383#comment-14063383
]
Gilles commented on MATH-1137:
------------------------------
bq. the behaviour between Java and Fortran is different
An example?
bq. \[ branches\] have to be covered in the same way in the same order for
Fortran and Java.
I recall that some statements seemed to rely on the finite precision of
floating-point arithmetic.
If that's the case, I don't think that the goal should be to mimic this
(fragile) behaviour. And if we don't, then the code paths might start to differ.
Finding the same optimum, within some reasonable tolerance, and with about the
same number of evaluations seems (IMHO) sufficient.
Again, you should raise this point on the "dev" ML to gather more opinions.
bq. ever seen the sketch by Morecombe and Wise and Andre Previn?
:)
Thanks for the reference. I knew the relevant excerpt but did not know its
origin.
> BOBYQA incorrect indexing
> -------------------------
>
> Key: MATH-1137
> URL: https://issues.apache.org/jira/browse/MATH-1137
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 3.3
> Reporter: Nigel Goodwin
>
--
This message was sent by Atlassian JIRA
(v6.2#6252)
