Wrap a routine for nonsymmetric LHSs

Open lkapelevich opened this issue 6 years ago • 4 comments

Our use case is a square, nonsymmetric KKT system, with multiple right hand sides.

Sep 29 '19 23:09 lkapelevich

Did you try https://github.com/JuliaSmoothOptimizers/MUMPS.jl ?

Sep 30 '19 01:09 dpo

I plan to try it. I was hoping to try routines from both MUMPs and HSL. I don't know if I should expect one to be superior for our problems.

Sep 30 '19 01:09 lkapelevich

The focus in MUMPS is on distributed computations. The factorization is efficient but my experience is that HSL subroutines are often hard to beat. That said, I've never compared the LU factorizations. I presume you've already tried UMFPACK?!

Also, contributions are most welcome. I'm quite tied up at the moment but I'd be happy to review PRs.

Sep 30 '19 14:09 dpo

Thanks a lot for the input! Yes we are playing with UMFPack simultaneously.

Also, contributions are most welcome. I'm quite tied up at the moment but I'd be happy to review PRs.

I won't get to this right away, but sounds good!

Sep 30 '19 15:09 lkapelevich