[petsc-dev] Why does TSDefaultComputeJacobianColor exist?

Barry Smith bsmith at mcs.anl.gov
Thu Apr 15 12:57:26 CDT 2010

    Sadly, I do not remember why I needed to have a separate version  
for TS, but I do remember not being happy with needing one but not  
finding a good alternative.

    The function in the TS case has an extra t argument, that is  
definitely part of the reason there is separate code but I do not  
remember if there are other reasons.

    It would be a good thing if they could be all merged together,


On Apr 15, 2010, at 9:43 AM, Jed Brown wrote:

> And MatFDColoringApplyTS?  It seems to me that TS defines the  
> nonlinear
> function and SNES already has the ability to do these things.  I ask
> because the DAE variants can't use these, and I'd rather not duplicate
> code for IJacobian variants.  The trouble seems to be that the  
> interface
> of just passing in a different function to TSSet{RHS,I}Jacobian  
> doesn't
> work any more.  SNES has all the necessary information, using an
> interface like this seems unnecessary since it seems like all we  
> really
> need is to get the MatFDColoring to the SNES.
> Note that if the shift is scaled very much differently from the right
> hand side, it is conceivable that the SNES operator may have poor
> conditioning compared to the RHS, but I think this case will be
> overwhelmed by the cases where the SNES operator is better conditioned
> (because transient Jacobians usually have better conditioning than
> steady-state Jacobians).
> Am I missing something?
> Jed

More information about the petsc-dev mailing list