Open ikalash opened 5 days ago
Curiously, the oscillations in acceleration near the Schwarz boundary do not lessen when I change the convergence criterion to include acceleration or reduce the Schwarz tolerance to as low as 1e-20. This was surprising.
In setting up / running the clamped problem with Schwarz, I have discovered discontinuities in the acceleration field (see screenshots below). It appears this issue actually shows up only in the initial time-step, but, curiously, there are noticeable errors in the acceleration field in subsequent time-steps (see screenshots below), suggesting Schwarz may not be converged. The tolerance is pretty tight already: https://github.com/lxmota/Norma.jl/blob/main/examples/ahead/overlap/clamped/dynamic/clamped.yaml#L11-L12 . I can try tightening it to see if it helps. I wonder if the acceleration should go into the convergence criterion? I can try adding that too to see if it helps.
It seems the issue with the discontinuity may have to do with the interpolation of the acceleration data not happening at the right time relative to the acceleration solve (https://github.com/lxmota/Norma.jl/blob/main/src/time_integrator.jl#L193).