A lava lamp-like thing to help pass them time.
(Note: this doesn't work on Firefox. I'm using OffScreenCanvas and WebWorkers to avoid tanking the main thread, but Firefox doesn't support 2d contexts on Offscreen Canvases, and Offscreen Canvases are behind a config flag anyways).
Each physics step consists of an advection step, and a divergence-removal step. The advection step is semi-Lagrangian; basically, we look backwards along the velocity vector for each cell to see where the fluid particle at the center of the cell would have come from, and then use that origin location's velocity for the new cell velocity. (I can kind of understand the inuition there, but if I think about it too hard, it makes my brain hurt). The divergence-removal is run many times per timestep, and consists of computing the gradient of the divergence across each 3x3 patch, and adding it back in to cancel the divergence. Repeating many times per timestep allows the 3x3 convolution to spread its effects through the environment.
The tracer particles have a bit of jitter, to help keep them from clumping up.
I worked from Christopher Batty's slides and Karl Sim's page (although I had to derive the divergence-removal step separately; his pseudo-code caused some wild behavior, but I probably just implemented it wrong).
Nearly all of the simulation time is spent in the divergence-removal step. I'd like to try converting that to wasm, or perhaps trying to do it from a GPU shader.
Additionally, I'd like to add some basic surface-level tracking to simulate water; right now the simulation is for a gas or liquid that totally fills the box.