2026 preprint

A Physics-Informed Fourier-Wavelet Transformer for Multiscale Computational Fluid Dynamics Surrogate Modeling

A physics-informed surrogate-modeling framework that combines Fourier and wavelet representations to learn multiscale computational-fluid-dynamics fields while retaining physically meaningful structure.

Multiscale flow fieldsFourier featuresWavelet featuresPhysics-informed prediction

Plain-language summary

High-fidelity CFD simulations can be expensive to run repeatedly. This work investigates a transformer-based surrogate that uses global Fourier information together with localized wavelet information, allowing the model to represent flow behavior occurring at different spatial scales.

Research contribution

  • Combines Fourier and wavelet representations for global and local flow structure.
  • Uses physics-informed learning to guide the surrogate toward physically meaningful predictions.
  • Targets multiscale CFD surrogate modeling for faster simulation-assisted analysis.

Publication status

arXiv preprint arXiv:2606.24696. Submitted to Engineering Applications of Artificial Intelligence.

Code and dataset

Code and dataset links will be added when a public release is available.

Recommended citation

Chakraborty, S., Pan, M., and Chen, X. (2026). A Physics-Informed Fourier-Wavelet Transformer for Multiscale Computational Fluid Dynamics Surrogate Modeling. arXiv:2606.24696.

BibTeX

@article{chakraborty2026physics, title={A Physics-Informed Fourier-Wavelet Transformer for Multiscale Computational Fluid Dynamics Surrogate Modeling}, author={Chakraborty, Somyajit and Pan, Ming and Chen, Xizhong}, journal={arXiv preprint arXiv:2606.24696}, year={2026} }

Related work

PIBERT publication page · PIBERT project · BubbleFieldNet