[petsc-users] SuperLU_dist bug with parallel symbolic factorisation

[Smith, Barry F.]

   Hmm, why would 

> the resolution with *sequential* symbolic factorisation gives ans err around 1e-6 instead of 1e-16 for parallel one (when it works). 

  ? One would think that doing a "sequential" symbolic factorization won't affect the answer to this huge amount? Perhaps this is the problem that needs to be addressed.

   Barry


> On May 22, 2018, at 12:57 PM, Eric Chamberland <Eric.Chamberland at giref.ulaval.ca> wrote:
> 
> On 22/05/18 12:11 PM, Xiaoye S. Li wrote:
> > Default setting is to use sequential symbolic factorization, precisely
> > due to the ParMETIS bugs.
> 
> Ok,
> 
> and I saw you reported the bug "a few years ago" and still have not received a fix...
> 
> I would like to "live with the patch" (ie working in sequential) but our problem is that if we compute err=|Ax-b|, the resolution with *sequential* symbolic factorisation gives ans err around 1e-6 instead of 1e-16 for parallel one (when it works).  MUMPS, also gives 1e-16 error levels.
> 
> In our nightly tests we have to compare computed solutions and they are now acceptable if they are of the order of 1e-15.  I do *not* want to raise this comparison to 1e-6 just because there is a bug in parmetis which forces us to use the "unprecise" sequential options...
> 
> In other words, I am kind of "stuck" with this non-fixed bug now... :/
> 
> Hope this can be fixed...
> 
> Thanks,
> 
> Eric