Viscousous fluids with ST(and dealing with volume loss) along with a deep discussion about FLIP sdfs:
ASwaab: you can also just crank surface sampling up on the emitter without reseeding everything. may not work for you, but it is an option I used a lot
DStewart: also scattering on the source surface and merging into the source volume points is super handy, but yeah no reseeding for those
JLait: You shouldn't need it, and we used to not need one before narrow band(about initializing SDFs). But its essential for narrow band
DStewart: If you're leaning on heavy reseeding for smooth surfacing and sourcing in having a smooth source SDF can make a huge difference. I've got some examples of that somewhere, need to dig 'em up. But probably most important for small-scale viscous sims, and I'm not doing the reseeding thing on most of those these days
It can also make a difference if you're sourcing and using high ST values, since a rough SDF tends to break that pretty quickly
Dave: ST can be stable with fewer substeps
But it does slow down the solve
A. Echeverry: SDF is extremelly important in the solver, and it has to be regenerated every timestep. The advection at the end is just to give you options if you want to export the surface data, so you can just advect it or regenerated, and advection is enough I think. But this last step has nothig to do with the sdf initialization for the sim, is just to export the final data.
And regards why we need sdf from the begining is because the solver needs a defined zone for the computations, in this case the sdf is like an alpha channel to apply everthing that is inside. The FLIP solver slowly became like a sparse solver but working on non sparse fields containers. JeffL explained that in a forum some time ago.
Also you can do black magig using that freesurface zone, to manipulate your fluid nicely! The thing is that I leaved almost all the computations to the particle stage, mainly because you can loose so much details working just with the volume stage. Thats why I also apply an uniform distribution to the particles mainly to have a much better fluid representation without artifacts, so at the end you dont need so many particles to have a good looking mesh
(about initlializing SDFs in Sims) - This is very important if you dont want to source particles where there is already a portion of the fluid present. So this sdf that came with your source is masking your particles to not emmit in an already filled zone. Is very useful when filling a tank with a source is at the bottom.
(About ST):Yes the ST from the flip solver is amazing but very expensive and slow. So one thing that is very usefull to make it work nicely is to always use a low timescale across the whole solver.This will make the ST works way better and without huge amount of substeps to converge correctly.
(about Re-init SDF and Particle to SDF setups in the fluild solver(look at 02_volume_setup)):The reinitialize sdf helps to make the sdf ranges more clean, and to remain as an sdf. When you use points to construct a sdf and then you want to expand the witdh or ranges, you can use a reinitialize sdf to sort of expand the ranges and to still make it valid.
TSlanik: resizing doesn't update values nor it would know how, so you need to reinitialize
it just updates the dimensions of the grid.
A. Echeverry: You can even use this to for example propagate your sdf rabges across the whole domain, this is so isefull if you will use this for attractors for example.
(About sPBD fluid):Pop fluid is sPBD if I'm not wrong! I think is a very nice solver, but yes It needs more options and capabilities.
TSlanik: POP Fluid is super nice especially if you need exploding fluid
DStewart:Have you been able to get natural-looking fluid behavior out of Pop Fluid? Other than ST droplet formation etc?
So flowing/stacking/splashing and high-viscosity behaviors?
And animated collision behavior is something I always have issues with
A. Echeverry: Unfortunately I failed to make it work nicely, yeah its perfect for droplets and even to fill objects (way better than FLIP). And yes the collisions are a nightmare, one think I do sometimes is to just change my colliders to use just sdf without proving a geometric representation, so the solver uses only the sdf and this fixes many collision issues on pops.