[aspect-devel] Using MUMPS through Trillions instead of PETSc

[Wolfgang Bangerth]

On 02/25/2017 11:57 AM, cedric thieulot wrote:
> out of curiosity, for our type of applications/matrices, which solver
> (SuperLU or MUMPS) do you think would be better/faster/less memory
> consuming/etc … ?

I don't think I have intuition. Given the fact that both have been developed 
for (among other applications) and tested with finite element problems, I 
would imagine that there is no fundamental, asymptotic difference between the 
two of them. I wouldn't be surprised if there is a difference of a factor of 
two or three in run times, memory consumption, etc, but I would be surprised 
if there was a factor of ten. It would be interesting to see a comparison, 
just so we gain some intuition.

Both are direct solvers and so will likely yield the same accuracy. In that 
sense, neither is going to be "better" than the other.

Best
  W.

-- 
------------------------------------------------------------------------
Wolfgang Bangerth          email:                 bangerth at colostate.edu
                            www: http://www.math.colostate.edu/~bangerth/