AI and the World's Hard Problems · Physics · arXiv · 2026-10-19
PINN-LPCDD: Physics-Informed Neural Network with Learnable and Physically Consistent Domain Decomposition for Compressible Multi-Medium Flows
Physics-informed neural networks struggle with compressible multi-medium flows, including locating interfaces, approximating discontinuities, and handling high pressure ratios between media with very different stiffness. The proposed PINN-LPCDD integrates the sharp interface method, uses sub-networks with learnable parameters for each medium to learn interface positions automatically, solves the Euler equations with artificial viscosity for shock waves, and exploits scaling invariance for high pressure ratios. Numerical experiments on air-air, air-helium, air-water and air-R22 problems show acceptable accuracy in simulating multi-medium flows.