PINN enforces angular momentum conservation with <0.05% error in spin-fluid simulations

bravo_abad · x · 2026-11-16

Conventional numerical methods can conserve energy and linear momentum yet still struggle with angular momentum. Matsuda et al. use a physics-informed neural network (PINN) to solve relativistic hydrodynamics equations while explicitly enforcing angular-momentum conservation during training, simulating the conversion of microscopic spin into macroscopic fluid rotation.

In one simulation, roughly 30% of initial spin angular momentum converts into fluid rotation with less than 0.05% error in total angular momentum; removing the conservation constraint leads to far larger numerical violations.

Original post →

More from Research

Research channel →