B=0 field output in magnetostatic mode

[jwestern]

I get zero magnetic field in the field output running with the magnetostatic solver on.

I'm running in electromagnetostatic mode via PICMI. I setup the solver like this:

solver = picmi.ElectrostaticSolver(
    grid=grid,
    method='Multigrid',
    required_precision=1e-6,
    warpx_self_fields_verbosity = 0,
    warpx_magnetostatic=True,
)

And I install the field diagnostic like this:

field_diag = picmi.FieldDiagnostic(write_dir = f'./diags/{diagdire}',
                                       warpx_file_prefix = 'field',
                                       grid = grid,
                                       period=100,
                                       warpx_openpmd_backend='h5',
                                       warpx_format='openpmd',
                                       data_list=['B', 'E', 'J', 'rho'])
    sim.add_diagnostic(field_diag)

This happens with both GPU and CPU runs.

Does anyone know what's going on? Let me know if other details are needed, such as the creation of grid or sim.

[roelof-groenewald]

Hi @jwestern. That is strange. Could you please provide a bit more information about the simulation:

• what geometry do you use (1d, 1d, 3d, rz)?
• are you seeing zero values for all B-field components?
• do you get non-zero current values?
• do you get non-zero E-field values?

[clarkse]

@jwestern could you share an input deck or reproducer so we can take a look at it? Can you share what sort of current source you have? Particle injection?

[jwestern]

• what geometry do you use (1d, 1d, 3d, rz)? 3D
• are you seeing zero values for all B-field components? Yes
• do you get non-zero current values? Yes
• do you get non-zero E-field values? Yes

@clarkse I'll work on a reproducer. The issue arises either with particles initialized at t=0 or with injection over time.

The runtime is slower with magnetostatic mode on, suggesting there is extra work being done. But the field diagnostic output gives B=0.

[jwestern]

@clarkse @roelof-groenewald Here's a reproducer PICMI .py file (I had to rename is .txt in order to attach):

zeroB_reproducer.txt

[clarkse]

@clarkse @roelof-groenewald Here's a reproducer PICMI .py file (I had to rename is .txt in order to attach):

zeroB_reproducer.txt

@jwestern Thank you for sending this over. There appears to be an issue with how the B field is calculated from A when not using an embedded boundary. As a work around you can compile with EB support. There is an example in Examples/Tests/magnetostatic_eb that still passes CI. You can probably adapt it to have a box EB for now. I will look at fixing the bug soon so you don't need to use an embedded boundary.

[jwestern]

@clarkse great, thank you! I'll wait for the bug fix, because compiling with EB support reduces the reliability of the Poisson solve to a level where I cannot use it.

[roelof-groenewald]

Thanks for tracking down the issue @jwestern and @clarkse.

@jwestern there's a WIP PR that actually fixes this bug (see #5175) but I won't have time in the next few weeks to finish up that PR. If you wanted to you could take a look at how the issue is fixed in that PR and maybe cherry-pick those parts to create an intermediate PR focused on just fixing this bug (without all the other refactoring done in the WIP PR).

[clarkse]

@jwestern I think this PR should fix your problem.
#5289

You can try to pull it and run your case. There was a bug how the functor for computing B from A was being constructed and the centering coefficients used were uninitialized when not using momentum-conserving field gathering.