[DrZoidberg]

Dear CFD experts,

I observe odd convergence behaviour in transient VOF:
At some point, when the flow situation gets more complex, residuals begin to stall for some timesteps (oscillating around 1e-4 to 1e-6). Looking at the cell-specific residuals it is because of momentum and/or energy residuals in some cells (0.01% of mesh) where residuals rise so high, that it "destroys" the global residual. Often it is just 2 cells at a wall or an interface that are affected. At least the solver does not diverge and keeps running (although at very small timestep, due to the error the flux-controlled adaptive timestep decreases to CFL~0.2). After some time residuals behave normal again, i.e. they drop below the target value of 1e-7 within 30 iterations.

The problem is very persistent, I tried the following:
* change timestep to CFL=2, 1, 0.5 (no effect)
* decrease mesh size 1/2 and 1/4 (finer mesh make this worse, slower progress due to lower timesteps)
* adding 10 boundary layers at walls for y+<100 (make things worse)
* 1stOrder, 2ndOrder with HOTR instead of QUICK (no effect)
* differentiable spatial limiter (no effect comp to default)
* acivate energy noise filter (no effect)
* GMRES instread of BCGSTAB (no effect)
* AMG off (coarsening levels =0, had a positive effect, otherwise divergence or velo/temperature limiting)


Basic case infos: 2ddp, pbcs, kwsst, phase1=air ideal-gas, phase2=incompressible fluid, homogenous VOF explicit with GeoReconstruct, transient first order, pressure level 1e7Pa, Velocity Range 10...400m/s, Temperature 800K
Mesh stats (for non-BL mesh): 12k cells, Mostly Quad Elements, Min Orthogonal Quality=0.3, Max Aspect Ratio=6.7, no negative volume or areas


I added some images (vof and T residual) close to the end of the filling simulation at 95% fillfraction. Not the best example, but it happens for very good cells (perfect squares), too.


Any ideas what could be the issue?