[petsc-users] Problem on LU factorization

[Matthew Knepley]

On Mon, Jan 24, 2011 at 3:32 PM, Rongliang Chen <rongliang.chan at gmail.com>wrote:

> Hi,
>
> I face a problem on the LU factorization. When I use the PETSC's default LU
> factorization, my code does not converge for KSP.


This is no enough information. Note that if you use this in parallel, it is
just BlokcJacobi LU and thus it is not
surprising that it does not converge.


> When I use superlu with command line "-sub_pc_factor_mat_solver_package
> superlu", it said

"[43]PETSC ERROR:
> ------------------------------------------------------------------------
> [43]PETSC ERROR: Caught signal number 11 SEGV: Segmentation Violation,
> probably memory access out of range
> [43]PETSC ERROR: Try option -start_in_debugger or -on_error_attach_debugger
> [43]PETSC ERROR: or see
> http://www.mcs.anl.gov/petsc/petsc-as/documentation/troubleshooting.html#Signal[43]PETSCERROR: or try
> http://valgrind.org on GNU/linux and Apple Mac OS X to find memory
> corruption errors
> [43]PETSC ERROR: likely location of problem given in stack below
> [43]PETSC ERROR: ---------------------  Stack Frames
> ------------------------------------
> [43]PETSC ERROR: Note: The EXACT line numbers in the stack are not
> available,
> [43]PETSC ERROR:       INSTEAD the line number of the start of the function
> [43]PETSC ERROR:       is given.
> [43]PETSC ERROR: [43] MatLUFactorNumeric_SuperLU line 121
> src/mat/impls/aij/seq/superlu/superlu.c
> [43]PETSC ERROR: [43] MatLUFactorNumeric line 2575
> src/mat/interface/matrix.c
> ............................
>  "
>

Please confirm that you have the latest patch level. If so, send the matrix
in PETSc binary format to petsc-maint at mcs.anl.gov
along with the precise solver options and output of -ksp_view.


> When I use superlu_dist, my code converges well, but I found that the
> compute time is very high. What's maybe the problem?  Thanks.
>

This is not surprising either. Sparse LU factorization can be expensive.

   Matt


>
> Regards,
>
> Rongliang
>



-- 
What most experimenters take for granted before they begin their experiments
is infinitely more interesting than any results to which their experiments
lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20110124/5483bedb/attachment-0001.htm>