[petsc-users] superlu_dist has some problem for SAME_NONZERO_PATTERN

[Barry Smith]

On Jan 3, 2011, at 2:26 AM, Gong Ding wrote:

> Dear Barry,
> Trouble you again.
> I found that my nonlinear solver failed to convergence with superlu_dist (even broken some times), but works well with mumps and pastix.
> 
> When I change SAME_NONZERO_PATTERN to DIFFERENT_NONZERO_PATTERN, everything goes well.

  Are you sure that non-zero pattern isn't changing? After the first solve call MatSetOption(mat, MAT_NEW_NONZERO_LOCATION_ERR,PETSC_TRUE and it will stop if the code tries to add a new nonzero location. 

  Also please switch to petsc-dev http://www.mcs.anl.gov/petsc/petsc-as/developers/index.html Sherry Li has fixed a bunch of things in SuperLU_dist that you can only use from petsc-dev.

   Barry


> 
> Will you please try to fix this?
> 
> Gong Ding