PETSC SparseDirectMUMPS and factorization reuse

[Yann Jobic]

Hello all,

I want to reuse the factorization of a PETSC/MUMPS solve for different
right-hand side vectors. I found the following discussion:

https://groups.google.com/g/dealii/c/Sw0wHKEFuiA

and the issue: https://github.com/dealii/dealii/issues/2869

However, it's quite old. Is there a way to do it properly ? For example,
an option as keep the factorized matrices outside of the solve method ?

Thanks,

Best,

Yann

[Wolfgang Bangerth]

On 2/12/25 09:21, Yann Jobic wrote:
>
> I want to reuse the factorization of a PETSC/MUMPS solve for different
> right-hand side vectors. I found the following discussion:
>

> https://nam10.safelinks.protection.outlook.com/?
> url=https%3A%2F%2Fgroups.google.com%2Fg%2Fdealii%2Fc%2FSw0wHKEFuiA&data=05%7C02%7CWolfgang.Bangerth%40colostate.edu%7Ccc9f132e5dd34347116708dd4b815658%7Cafb58802ff7a4bb1ab21367ff2ecfc8b%7C0%7C0%7C638749741055859417%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=tYsPCfUrZbq7pQEyTSZ6ld7S8ykwLbGWsiZ%2Fre43UzE%3D&reserved=0
>
> and the issue: https://nam10.safelinks.protection.outlook.com/?
> url=https%3A%2F%2Fgithub.com%2Fdealii%2Fdealii%2Fissues%2F2869&data=05%7C02%7CWolfgang.Bangerth%40colostate.edu%7Ccc9f132e5dd34347116708dd4b815658%7Cafb58802ff7a4bb1ab21367ff2ecfc8b%7C0%7C0%7C638749741055880166%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=LArEuAs1qSPsEZSrU2jGkLyfH%2BgyQVQOBAQ8qy8Osdo%3D&reserved=0

>
> However, it's quite old. Is there a way to do it properly ? For example,
> an option as keep the factorized matrices outside of the solve method ?

Yann -- I don't think there is any specific news I can share (though
good job finding old discussions on the topic!). The interface of
SparseDirectMUMPS does not currently provide a way to separate the
factorization step from the solve step.

I suspect that it would not be very difficult to actually change the
interface in such a way that it looks more like the one of
SparseDirectUMFPACK that does exactly that. It's just that nobody so far
has found the time or motivation to do that. But if you wanted to, that
would of course be fantastic and we'd be excited to see a patch (and
help you with it)!

Best
Wolfgang