Home / Publications / PIBERT
Under review

PIBERT: A Physics-Informed Bi-directional Hybrid Spectral Transformer for Multiscale CFD Surrogate Modeling

PIBERT combines spectral representations, transformer learning, and physics-informed objectives to build efficient surrogate models for multiscale CFD fields.

Code repositoryProject page
PIBERT physics-informed CFD surrogate modeling overview

Plain-language summary

Repeated CFD simulation can be computationally expensive. PIBERT learns a fast approximation of the flow solution while using spectral information and physical constraints to retain important multiscale behavior.

Research contribution

  • Hybrid spectral and transformer representations for multiscale flow fields.
  • Physics-informed training objectives for more reliable surrogate predictions.
  • Evaluation across PDE and CFD benchmark settings.

Publication status

Journal manuscript under review at the Journal of Computational Physics.

Code and dataset

Public code repository. Dataset documentation is available through the project repository and associated manuscript materials.

Recommended citation

Chakraborty, S., Ming, P., and Chen, X. PIBERT: A Physics-Informed Bi-directional Hybrid Spectral Transformer for Multiscale CFD Surrogate Modeling. Manuscript under review.

BibTeX

@article{chakraborty2025pibert, title={PIBERT: A Physics-Informed Bi-directional Hybrid Spectral Transformer for Multiscale CFD Surrogate Modeling}, author={Chakraborty, Somyajit and Ming, Peng and Chen, Xizhong}, journal={Manuscript under review}, year={2025} }

Related work

PIBERT project page · Fourier-Wavelet Transformer · BubbleFieldNet