I wonder if UE5 or Blender have similar and perhaps competitive methods for the Loki water state machine in generating realistic water and mist. The article uses a lot of jargon:
Navier-Stokes simulators, Lagrangian wave packets attached to spline curves, incompressible two-phase Navier-Stokes solvers on a Eulerian grid, and air phase representation by FLIP (Fluid Implicit Particle) and APIC (Affine Particle-in-Cell) solvers.
I remember the Quake I water simulation, which I found amazing at the time. I knew water is hard to simulate, but understanding fluid dynamics and the above jargon seems hard to grok too.
The techniques that control the way electrons flow in silicon is hard to grok. But that gives way to logic gates, that let you build CPUs with branch predictors. The way that branch predictors actually work in CPUs is hard to grok but how to write code that keeps their implications in mind is more approachable. Learning how to use libraries that leverage the above is even easier to grok.
Technology is built upon a stack of useful abstractions, and the ability to understand in deep technical detail the entire stack is a super power. There's no reason why computational physics shouldn't work the same way.
Computational Fluid Dynamics is a tricky area. The science (Fluid Dynamics) was worked out over 100 years ago, with the Navier-Stokes equations (3D partial differential equations). The engineering (the Computational part) is a whole bunch of complex and clever tricks, to overcome surprising problems like instability and dissipation, and often for some particular engineering trade-off (rather than being universally true).
Avatar 2 seems to use an underlying FLIP simulation for the water (which uses a (eularian) grid for pressure, and particles for movement), with cinematic tricks on top: foam, bubbles, and smaller (lagrangian) waves superimposed.
The Fluid Flux plug-in uses a simplified version of the Navier-Stokes equations, which represents the upward direction as an average - so it's 2D internally but looks 3D. It also just calculates once, instead of iterating. Both make it much faster. It uses this paper, which has several other tricks: https://matthias-research.github.io/pages/publications/hfFlu...
I remember the Quake I water simulation, which I found amazing at the time. I knew water is hard to simulate, but understanding fluid dynamics and the above jargon seems hard to grok too.