Solving Navier-Stokes Equations With Mixed Equation Physics Informed Neural Networks

This paper presents a study on the implementation and testing of mixed-precision and mixed-equation approaches for optimizing the performance of physics-informed neural networks. Mixed-equation approach involves utilizing equations in a multi-step manner, which leads to a significant reduction in computational costs during the network's training while capturing complex physical phenomena. Specifically, we demonstrate the effectiveness of the proposed methodology in approximating the Navier-Stokes equations for incompressible flow around a 2D cylinder.