Add P argument to solve function

[juliusgh]

The class LinearSolver already supports passing a custom preconditioner matrix P to the PETSc solver.
However, this is so far not available through the solve function.

This PR adds an argument P to the solve function that is passed to the LinearSolver. At least in my case, it would be convenient to pass a custom preconditioner matrix directly in this way.

[connorjward]

This seems sensible. Thanks!

It would be nice to have a test for this but I think that might be tricky to come up with. We don't even test the P argument to LinearSolver as things stand.

[juliusgh]

Thanks for the quick reply!

I think a short unit test would indeed be difficult to find. I am using the P matrix passing to solve homogenization problems from computational mechanics with a custom preconditioner. FFT-based preconditioners are very effective for these problems, but in addition to A, a suitable operator P must also be passed. I could provide a documented demo that uses this feature, but maybe that's too specific for firedrake...

[dham]

Thank you!

[connorjward]

Thanks for the quick reply!

I think a short unit test would indeed be difficult to find. I am using the P matrix passing to solve homogenization problems from computational mechanics with a custom preconditioner. FFT-based preconditioners are very effective for these problems, but in addition to A, a suitable operator P must also be passed. I could provide a documented demo that uses this feature, but maybe that's too specific for firedrake...

To quote @dham re the suggested demo: "yes yes yes please". Demos are always great to have.

[connorjward]

@juliusgh please could you rebase this code on top of the latest master/merge in the latest master and push it again? The tests are failing and that should fix them. Then I will merge this. Thanks.

[juliusgh]

@juliusgh please could you rebase this code on top of the latest master/merge in the latest master and push it again? The tests are failing and that should fix them. Then I will merge this. Thanks.

Ok, I have rebased on top of master. I hope that the CI works now.

To quote @dham re the suggested demo: "yes yes yes please". Demos are always great to have.

Thanks for the feedback! I need some time to finish a well documented demo and would submit a new PR later.

[connorjward]

The rebase appears to have gone a bit wrong (the "Files changed" is now quite large). I have created a fresh PR with just your commits: #3571. I will close this one.

[juliusgh]

The rebase appears to have gone a bit wrong (the "Files changed" is now quite large). I have created a fresh PR with just your commits: #3571. I will close this one.

Oh sorry. Thanks!