Looks like a tricky problem!
I will play around with it as well, but there are some smart guys in this thread already, so not sure if I will be able to solve that specific issue.

Sometimes a little bit of shader displacement (along the gradient, tracking with the rest position) might be enough to help hide the problem too - that's more of a fix than a real solution though.

Apart from that it does look like banding of some sort, I've had that happen when the container gets resampled and the resolutions between containers are not the same. It might be good to analyze the velocity in that region.
Also the confinement exagerates the eddies/rotational vortex movement. But if that is too strong it can cause density to collapse in on itself. (Kinda like a black hole). The pressure field should then work as a counterforce to prevent this from happening as that region would cause a zone of high pressure.

I would try to visualize the various fields in that area, specifically your velocity and pressure fields.
– it might be that the confinement is messing with the areas of low velocity.
– or it is a sampling issue. I know for some things only half the res of your fields is used. For example the density field for a resize-able container gets sampled at half res. There might be other areas within the solver where this is happening, but I don't know by heart.
– I would be curious to see if you could make it go away by resampling the field in a vdb and filtering it a bit.

If all else fails, send a pm to Jeff Lait (jlait) or support .

Thanks Peter. I'll try this.

I've seen this in different solvers, not just houdini.

I can often reproduce the problem with turbulence alone, and it is often evident with high magnitude in the turbulence - I believe the culprit to be the non divergent grid / pressure solve dealing with high velocities/when turbulence is high.

Consider this- if turbulence is extremely high and constant everywhere, how does the pressure solve resolve this? I would expect it to create the artifacts we see in an imperfect algorithm. I suspect that the magnitude in the equation is favouring the xyz vectors, potentially it doesn't correct the pressure properly based on length but instead favours the magnitude isolated in each axis seperately. creating xyz noise within a box domain (instead of spherical) can do similar things.

Thats my best guess. I come across this in different situations over the years and it would be a good one to crack. I often get it to go away by cracking open turbulence in pyro and trying different noise for turb velocities.

another guess that is related is - I also wonder what happens in a pressure solve when a velocity vector is perpendicular to the grid and aligned with an opposing vector… seems like keeping only the non divergent component in this situation could be unrealistic and nasty.

this would explain the artifacts too - it appears like velocity can be killed completely in these artifacts by the non divergent grid solve which would explain also why we see these perpendicular artifacts.

You can go to your Solver- Advanced- max substeps increase it to 2or3. that should solve your problem.