danieljprice
commented
2 months ago a simple fix here is to switch off the velocity iterations. If you set the tolv > 1e2 then it just won't try to do these at all.
The code does try to adjust the timestep when tolv causes iterations, because taking a velocity iteration is as expensive as taking a smaller timestep, so one might as well take a smaller timestep to avoid doing iterations. But this timestep adjustment could definitely send dt to zero if things were going wrong
Often this kind of thing is a symptom of something else being wrong, like div B being out of control?
teogeo1996
Whilst using global time steps to try and examine an issue arising from individual tilmestep simulations that caused the tilmestep to reduce and caused Fatal errors, I discovered something. In mid simulations, as the density increases in colliding flows, the velocity does not converge which throws two errors. Error 1: fatal error velocities not converged. Error 2 dtc:0 Since tilmestep goes to zero, since err-> +00. I can see that in both cases, the iterations of velocity cause the error to grow until errmax=infinity. I have tried reducing the Courant and force parameters to limit maximum tilmestep allowed (as well as dtmax) and increase tree accuracy (from 0.5 -> 0.1). This sometimes works but still a lot of my simulations throw this error. Is there a work around this? I think that this is what causes individual tilmestep simulations to crush as well and this is an issue located primarily at MHD simulations.