Stochastic Representation-Inspired Physics-Informed Neural Networks for Incompressible Fluid Flows

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 learnable kernel scales. Both variants retain the conventional velocity--pressure formulation and physics-informed training framework, allowing the influence of the activation function to be examined independently. We evaluate the method on five benchmarks, including Kovasznay flow, cylinder wake, three-dimensional Beltrami flow, Taylor--Green vortex, and double shear layer. Across 21 benchmark--architecture comparisons, both variants consistently reduce the combined velocity error relative to the conventional $\tanh$-based PINN, achieving mean relative error reductions of 31.7\% and 33.2\%, respectively. Improvements are also observed in reported pressure and vorticity errors. Notably, the double-shear-layer experiments demonstrate improved resolution of localized shear structures and, for BS-PINN-F, qualitatively clearer vortex-roll-up patterns in the long-time simulation. These findings highlight the potential of incorporating analytical flow structures into neural-network representations for incompressible flow simulations.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-17
DOI
https://doi.org/10.48550/arxiv.2611.22624
Primary Topic
Model Reduction and Neural Networks
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Stochastic Representation-Inspired Physics-Informed Neural Networks for Incompressible Fluid Flows

Zhongmin Qian, Zihao Guo
arXiv (Cornell University)
Model Reduction and Neural Networks
preprint

Stochastic Representation-Inspired Physics-Informed Neural Networks for Incompressible Fluid Flows

Zhongmin Qian, Zihao Guo
preprint en

Abstract

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 learnable kernel scales. Both variants retain the conventional velocity--pressure formulation and physics-informed training framework, allowing the influence of the activation function to be examined independently. We evaluate the method on five benchmarks, including Kovasznay flow, cylinder wake, three-dimensional Beltrami flow, Taylor--Green vortex, and double shear layer. Across 21 benchmark--architecture comparisons, both variants consistently reduce the combined velocity error relative to the conventional $\tanh$-based PINN, achieving mean relative error reductions of 31.7\% and 33.2\%, respectively. Improvements are also observed in reported pressure and vorticity errors. Notably, the double-shear-layer experiments demonstrate improved resolution of localized shear structures and, for BS-PINN-F, qualitatively clearer vortex-roll-up patterns in the long-time simulation. These findings highlight the potential of incorporating analytical flow structures into neural-network representations for incompressible flow simulations.

arXiv (Cornell University)
Model Reduction and Neural Networks
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon