Re: [SOSlib-devel] Bug in sensitivity analysis?

[Rainer M. <ra...@tb...>]

Hi,

> CvodeSettings_setTime(settings, 7.5*200, 200);

The success of CVODES integration really depends on the number of 
printsteps because the error tolerances are used separately for each 
printstep. Errors can accumulate and cause numerical instabilities that 
CVODES cannot handle.

The behaviour you described in the previous emails, that CVODES 
convergence fails with less parameters and succeeds with more parameters 
is really strange.

I cannot explain this, but currently couldn't think of any bug that causes 
this. To find out if it is a SOSlib bug, we could try to change the input 
file, i.e. localize/globalize the parameters by hand and see what happens.

It might be some problem of the algorithm. Maybe 
you can play with the error tolerances and printsteps. If that succeeds 
that we could be sure that it's not a SOSlib bug.

Rainer


On Thu, 11 May 2006, Witold Wolski wrote:

> Hi,
>
> Sorry, I have forgotten a sentence in the last e-mail.
>
> However, after changing:
>
> CvodeSettings_setTime(settings, 300, 10);
> to:
> CvodeSettings_setTime(settings, 7.5*200, 200);
>
> The sensitivities are computed ....
>
> I am just not getting my head around this.
>
> Eryk
>