[
https://issues.apache.org/jira/browse/MATH-1137?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14064877#comment-14064877
]
Nigel Goodwin commented on MATH-1137:
-------------------------------------
This is now drawing to a close...
The good news is that I managed to reproduce the Fortran behaviour until close
to convergence, when rounding errors start to become more important. I compare
the values of all the function evaluations until close to convergence, and the
paths.
I altered some comparison statements between doubles, as sometimes two numbers
almost identical were being compared and the rounding errors caused changes in
logic (as you know).
I did a few changes of initialisation of indices, Fortran set them to zero, I
set them to minus one, this may have an effect.
So, in short:
- I am now pretty confident my Java version works as it should
- I still couldn't get the Fortran to exercise the lines which started all this
(although the Java does), so I'm not much further on whether my fix is
confirmed correct or not.
Some time soon I will try to put this to bed. It is very time consuming. But
also, I have provided a further visual check of quite a lot of the code, so
that can't be a bad thing.
> 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)
