Coming from an SPH background, I implemented and tested an MPM fluid solver. The solver uses a density-based incompressiblity contraint similar to Kugelstadt et al. (2021) instead of solely relying on an estimated velocity divergence. This way, the fluid does not compress over time. Through a particle sampling, complex geometries can be naturally incorporated into the simulation.
SPH fluids, especially implicit pressure solvers with small kernel support, suffer from high artificial viscosity. This is quite unfortunate, since they are otherwise robust, performant and intuitive to use for simulating free-surface flows. I'm currently exploring some tweaks related to the SPH interpolation that result in a much less viscous fluid with a comparable number of neighbors!
A lid driven cavity with the experimental SPH interpolation. To emphasize the turbulent fluid flow, only tracer particles are rendered here.