On Tue, Apr 18, 2023 at 11:16 AM Karthikeyan Chockalingam - STFC UKRI <
[email protected]> wrote:
> Thank you for your response. I spend some time understanding how
>
> MatSetValuesLocal and ISLocalToGlobalMappingCreate work.
>
You can look at SNES ex28 where we do this with DMCOMPOSITE.
> Q1) Will the matrix K be of type MATMPIAIJ or MATIS?
>
> K = [A P^T
>
> P 0]
>
I assume MPIAIJ since IS is only used for Neumann-Neumann decompositions.
> Q2) Can I use both MatSetValues() to MatSetValuesLocal() to populate K?
> Since I have already used MatSetValues() to construct A.
>
You can, and there would be no changes in serial if K is exactly the upper
left block, but in parallel global indices would change.
> Q3) What are the advantages of using MatSetValuesLocal()? Is it that I
> can construct P directly using local indies and map the entrees to the
> global index in K?
>
You have a monolithic K, so that you can use sparse direct solvers to check
things. THis is impossible with separate storage.
> Q4) I probably don’t have to construct an independent P matrix
>
You wouldn't in this case.
Thanks,
Matt
> Best regards,
>
> Karthik.
>
>
>
>
>
>
>
> *From: *Matthew Knepley <[email protected]>
> *Date: *Tuesday, 18 April 2023 at 11:08
> *To: *Chockalingam, Karthikeyan (STFC,DL,HC) <
> [email protected]>
> *Cc: *[email protected] <[email protected]>
> *Subject: *Re: [petsc-users] Setting up a matrix for Lagrange multiplier
>
> On Tue, Apr 18, 2023 at 5:24 AM Karthikeyan Chockalingam - STFC UKRI via
> petsc-users <[email protected]> wrote:
>
> Hello,
>
>
>
> I'm solving a problem using the Lagrange multiplier, the matrix has the
> form
>
>
>
> K = [A P^T
>
> P 0]
>
>
>
> I am familiar with constructing K using MATMPIAIJ. However, I would like
> to know if had [A], can I augment it with [P], [P^T] and [0] of type
> MATMPIAIJ? Likewise for vectors as well.
>
>
>
> Can you please point me to the right resource, if it is a common operation
> in PETSc?
>
>
>
> You can do this at least 2 ways:
>
>
>
> 1) Assemble you submatrices directly into the larger matrix by
> constructing local-to-global maps for the emplacement. so that you do
>
> not change your assembly code, except to change MatSetValues() to
> MatSetValuesLocal(). This is usually preferable.
>
>
>
> 2) Use MATNEST and VecNEST to put pointers to submatrices and subvectors
> directly in.
>
>
>
> Thanks,
>
>
>
> Matt
>
>
>
> Many thanks.
>
>
>
> Kind regards,
>
> Karthik.
>
>
>
>
>
>
>
>
>
>
>
>
> --
>
> What most experimenters take for granted before they begin their
> experiments is infinitely more interesting than any results to which their
> experiments lead.
> -- Norbert Wiener
>
>
>
> https://www.cse.buffalo.edu/~knepley/
> <http://www.cse.buffalo.edu/~knepley/>
>
--
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
