Hybrid Fourier Neural Operator-Lattice Boltzmann Method

TL;DR

This paper introduces a hybrid Fourier Neural Operator-Lattice Boltzmann Method that accelerates fluid dynamics simulations, improves accuracy, and stabilizes long-term predictions for both steady and unsteady flows.

Contribution

The work presents a novel hybrid framework combining FNO and LBM that significantly speeds up simulations and enhances accuracy compared to traditional methods.

Findings

Achieves up to 70% speed-up in steady-state convergence.

Hybrid coupling reduces error by 96-99.8% with a lightweight model.

Enhances stability and accuracy of long-horizon flow predictions.

Abstract

We propose an accelerated computational fluid dynamics framework based on a hybrid Fourier Neural Operator-Lattice Boltzmann Method (FNO-LBM) for steady and unsteady weakly compressible flows. FNO-based initialization significantly accelerates LBM in reaching steady-states of porous media flows across all macroscopic fields, achieving up to 70% speed-up in convergence of density and more than 40% of pressure-drop while preserving the final steady-state accuracy. Simulations of unsteady flows can be accelerated by hybrid coupling strategies that employ FNO rollouts embedded into LBM time advancement in a way of super-time-stepping. Global and time-resolved error metrics across 100 trajectories for generic 2D flows demonstrate that hybridization consistently improves accuracy and stabilizes long-horizon rollouts. Best efficiency is achieved for a lightweight 2.6M-parameter FNO, which diverges under pure autoregressive rollout but achieves 96-99.8% error reduction under hybrid coupling, matching the predictive capability of a much more expensive 11.2M-parameter model. The hybrid framework enhances predictive fidelity, suppresses error accumulation, and enables small and cheap surrogate models to operate effectively within the same error regime as larger surrogates. These results demonstrate that hybrid neural-operator coupling achieves robust and computationally efficient accelerated LBM while maintaining physically consistent flow evolution.