[
https://issues.apache.org/jira/browse/MATH-380?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12882304#action_12882304
]
Luc Maisonobe commented on MATH-380:
------------------------------------
You are perfectly right.
The FirstOrderIntegratorWithJacobians class is a brand new one and it clearly
has some design flaws.
It will most probably be deprecated in its current form and replaced by a new
mechanism, better integrated (sorry for the joke) with the standard ODE solvers.
The ability for user to set an initial value for dydy0 will be present in the
new design, but will probably not be back-ported to the current one.
In the meantime, you can save the final value of the jacobian matrix dydy0
after first part of integration, which we could call dy1dy0 as it represents
dy(t1)/dy(t0). Start the second part from t1 to t2 that will reset the initial
matrix to identity and hence compute compute dy(t2)/dy(t1) and do the
multiplication by yourself of the two matrices to really get what you need:
dy(t2)/dy(t1) = dy(t2)/dy(t1) * dy(t1)/dy(t0).
Thanks for reporting the issue
> Need to (re)initialize dYdY0 for multiple integrate with
> FirstOrderIntegratorWithJacobians
> ------------------------------------------------------------------------------------------
>
> Key: MATH-380
> URL: https://issues.apache.org/jira/browse/MATH-380
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 2.1
> Reporter: Pascal Parraud
> Priority: Minor
>
> There is a lack in the method integrate of FirstOrderIntegratorWithJacobians.
> The jacobian DYDY0 can't be initialized by the user, unlike DFDP with DF0DP.
> So, for several successive integrations, the matrix is reinitialized to
> identity and that is not what we might want.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
