Id: arxiv/2611.22624 Url: https://arxiv.org/abs/2611.22624 Tags: physics.flu-dyn, math.AP, math.PR Score: 0 Title: Stochastic Representation-Inspired Physics-Informed Neural Networks for Incompressible Fluid Flows Domain: physics.flu-dyn Source: arxiv Authors: Zihao Guo, Zhongmin Qian Reviews: 0 Subject: physics.flu-dyn Category: research Comments: 0 Disputed: false Division: physics Mentions: 0 Discussed: 0 Published: 2026-11-17 Page: https://delphi.science/r/arxiv/2611.22624 The effectiveness of neural networks in solving partial differential equations depends not only on their approximation capability but also on the suitability of their representations for the underlying solutions. Motivated by stochastic integral representations of incompressible fluid flows and the associated Biot--Savart kernel, we propose a Biot--Savart-informed (i.e. physics-informed) neural network (BS-PINN) for the incompressible Navier--Stokes equations. By extracting a rational profile from a regularized Biot--Savart kernel, we construct a smooth activation function with fixed and…