Differentiable moving-mesh Voronoi fluid solver in JAX runs autodiff through mesh motion
CatAstro_Piyush · x · 2026-11-04
PMocz released differentiable-voronoi-fluids, a 2D compressible Euler solver on an Arepo-style moving Voronoi mesh written in JAX, so reverse-mode autodiff runs through the entire simulation including mesh motion.
Key trick: mesh connectivity is piecewise constant and non-differentiable, so it's computed on the host via SciPy Delaunay (Qhull) through jax.purecallback with gradient-stopped positions, while continuous face geometry stays inside JAX. It's a prototype: periodic box, no walls, no refinement, hydro only.
More from Research
- Allen Institute teams with Google on NeuralGCM+ for years-to-decades climate projections — allen_ai · 2026-11-05
- NeuralGCM+ shifts to decadal projections, coupling atmosphere and ocean models — allen_ai · 2026-11-05
- Weight edits lift Qwen3.5-9B SecEval from 32.5% to 51.5% without real security capability gains — evilsocket · 2026-11-05
- PianoBench: open audio models top out at 52.8% on musical reasoning — kastnerkyle · 2026-11-05
- Same model, new harness: success rate jumps from 68% to 88% — Arindam_1729 · 2026-11-05
- AI is tackling longer research tasks, but scientific taste remains the hard part — nathanbenaich · 2026-11-05