[petsc-users] KSPComputeRitz, GMRES, matrix with complex number

[feng wang]

Hi Mark,

I use "KSPComputeEigenvalues" and it works with complex numbers. For the moment, this should be enough for me.

Besides, the actual number of eigenvalues computed changes from iteration to iteration. Just for curiosity, how does the code decide how many to output?

Thanks,
Feng


________________________________
From: Mark Adams <mfadams at lbl.gov>
Sent: 14 July 2022 15:49
To: feng wang <snailsoar at hotmail.com>
Cc: petsc-users at mcs.anl.gov <petsc-users at mcs.anl.gov>
Subject: Re: [petsc-users] KSPComputeRitz, GMRES, matrix with complex number



On Thu, Jul 14, 2022 at 11:29 AM feng wang <snailsoar at hotmail.com<mailto:snailsoar at hotmail.com>> wrote:
Thanks for the clarification. any recommendations to get around this?

I would look at the existing code and see if you can see what is limiting it to real (comments hopefully), if it looks like something you could do, start with a git repo of PETSc, configure with complex scalar values, modify an existing test or make your own to test it, and make it work.
At that point you can contribute that to PETSc (there is documentation on how to do that including our coding practices that you should be able to infer to a large extent from the code you are modifying).
At that point PETSc will maintain it.

Mark


I am solving a set of linear systems of A_i x_i = B_i with matrix-based GMRES. The entries of A_i and B_i are complex numbers. It is pre-conditioned with ASM and the preconditioning matrix is A_i itself. I would like to show the Ritz value to understand the convergence behaviour for each A_i x_i = B_i.
________________________________
From: Mark Adams <mfadams at lbl.gov<mailto:mfadams at lbl.gov>>
Sent: 14 July 2022 14:16
To: feng wang <snailsoar at hotmail.com<mailto:snailsoar at hotmail.com>>
Cc: petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov> <petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov>>
Subject: Re: [petsc-users] KSPComputeRitz, GMRES, matrix with complex number

No. I missed that note.
Complex has not been implemented.


On Thu, Jul 14, 2022 at 9:54 AM feng wang <snailsoar at hotmail.com<mailto:snailsoar at hotmail.com>> wrote:
Hi Mark,

Thanks for your reply!

Maybe I have a misunderstanding of the documentation of KSPComputeRitz (https://petsc.org/main/docs/manualpages/KSP/KSPComputeRitz/).  It says "-this is currently not implemented when PETSc is built with complex numbers".

But if it works, I will give it a try.

Thanks,
Feng


<https://petsc.org/main/docs/manualpages/KSP/KSPComputeRitz/>

________________________________
From: Mark Adams <mfadams at lbl.gov<mailto:mfadams at lbl.gov>>
Sent: 14 July 2022 12:23
To: feng wang <snailsoar at hotmail.com<mailto:snailsoar at hotmail.com>>
Cc: petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov> <petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov>>
Subject: Re: [petsc-users] KSPComputeRitz, GMRES, matrix with complex number

Why do you say it only works with Real? You need to configure PETSc to use complex numbers and it should work.
This is an advanced method and it does not have any tests, which basically says it is very experimental.
Don't expect it to be very robust.

Mark

On Thu, Jul 14, 2022 at 7:19 AM feng wang <snailsoar at hotmail.com<mailto:snailsoar at hotmail.com>> wrote:
Dear All,

I am using GMRES with complex numbers and would like to compute the Ritz values. I am looking at KSPComputeRitz<https://petsc.org/main/docs/manualpages/KSP/KSPComputeRitz/>  and it seems to me that it only works with Petsc compiled with real numbers. Is there an easy alternative way to compute Ritz values for matrices with complex numbers?

Thanks,
Feng
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20220717/787bbfc0/attachment-0001.html>