Open efaulhaber opened 9 months ago
I think we should look again after the surface tension model is merged since part of this model can be used to calculate a repulsive force at the boundary.
I created this issue to discuss options and see if we can find anything about this in the literature (I couldn't).
I just found this in section 4.8 in https://doi.org/10.1016/j.compfluid.2018.11.023 The authors also use the BC from Adami.
"Change to the implementation" indicates that Adami himself didn't do that? Or did these authors (like us) just not find this anywhere in the paper and assume it has been done by Adami without mentioning? It would probably be the easiest to just ask someone in Adami's group.
Left: main. Right: main, but no attractive forces between fluid and boundary.
There are some particles catapulted away in the right version, but at least they're not sticking.
With
ContinuityDensity
, we allow negative particle pressure and corresponding attractive forces between particles. This is a good thing, as clipping the pressure results in unphysical rarefaction of the fluid in some situations. A realistic incompressible or weakly compressible fluid can't just be pulled apart to a significantly lower density.However, we usually (always?) don't want attractive forces between fluid and boundary particles on (free-slip?) walls. But with dummy particle boundaries, we do get attractive forces if the pressure becomes negative. To demonstrate the severity of the problem, here is a dam break with dummy boundary particles using
SummationDensity
(i.e. their density is calculated through kernel summation and their pressure from the state equation). This is the result when the problem is not treated at all: WithSummationDensity
, all boundary particles away from the fluid end up with a large negative pressure. Therefore, the fluid is sucked up the walls of the tank (I should've used a tank without ceiling, that would look even funnier).Currently, we just limit the pressure of dummy boundary particles to be non-negative. This eliminates the sucking, but still produces some sticking artifacts, as there are still attractive forces when a fluid particle close to the boundary has a negative pressure. The example looks like this, then. As can be seen very clearly, there are still sticking artifacts in the form of fluid particles gliding along the boundary.
Another strategy that came to my mind is to limit the coefficient in front of the kernel gradient in the pressure acceleration to be non-negative for fluid-boundary interaction, eliminating all attractive forces between fluid and boundary particles. This is the result: Obviously, without attractive forces, the fluid will bounce off the boundary after the impact.
I created this issue to discuss options and see if we can find anything about this in the literature (I couldn't).