New Python library enables solving complex fluid equations using discrete neural networks and collocation-based physics-informed training

Researchers develop DVF-CRVPINN, a programming environment for solving PDEs through discrete weak formulations instead of traditional continuous approaches

The framework uses discrete neural networks trained over point sets with Kronecker delta test functions and discrete finite difference derivatives

Demonstrates the method on Stokes equations in 2D, a computationally challenging fluid dynamics problem

Training employs the Adamax algorithm with discrete automatic differentiation for optimization