MLMG convergence issue

[yopology]

Hi, I am trying the magnetic reconnection test problem that uses the Hybrid solver (https://warpx.readthedocs.io/en/latest/usage/examples/ohm_solver_magnetic_reconnection/README.html) I noticed that when I increase the resolution to 1024x1024, the MLMG Projection B-Field divergence cleaner failed to converge after 5000 iterations. What can I do to fix this problem? Thanks!

[roelof-groenewald]

Hi @yopology. That problem should not require the B-field divergence cleaning as the analytic field should not have any B-field divergence. You can turn off the divergence cleaning by setting warpx_do_initial_div_cleaning=False in the B_ext initialization, i.e.,

        B_ext = picmi.AnalyticInitialField(
            Bx_expression=self.Bx, By_expression=self.By, Bz_expression=self.Bz,
            warpx_do_initial_div_cleaning=False
        )

Let us know if that does not work for you.

[yopology]

It works. Thank you!

[EZoni]

@roelof-groenewald

I saw this and thought of a question. Shouldn't one nonetheless expect that divergence cleaning does simply nothing (as opposed to failing to converge) when the analytical field does not have any B-field divergence, i.e., when there is no error in $div(B) = 0$ (I guess) to take care of? Without knowing the details of how this works, intuitively I would expect convergence in one iteration, as opposed to no convergence at all, in this situation.

[roelof-groenewald]

@EZoni yes, I agree in principle that should be the case, but the divergence will still have non-zero values up to machine precision which is what I expect causes the problem here. We could take more care with setting the absolute tolerance for the divergence cleaner so that it basically considers the solution converged on the first iteration.