Software & Data Downloads — ab-pinns

Adaptive-Basis Physics-Informed Neural Networks for Residual-Driven Domain Decomposition for physics-informed neural networks (PINNs).

Adaptive-Basis Physics-Informed Neural Networks (AB-PINNs) for Residual-Driven Domain Decomposition in PINNs.

This repository contains the PyTorch implementation of Adaptive-Basis Physics-Informed Neural Networks (AB-PINNs). AB-PINNs are an adaptive approach to domain decomposition in PINNs. During training, local subdomains can move and deform to fit the geometry of solution features, while AB-PINN+ can add new subdomains in regions with high PDE residual.

    •  Botvinick-Greenhouse, J., Ali, W.H., Benosman, M., Mowlavi, S., "AB-PINNs: Adaptive-Basis Physics-Informed Neural Networks for Residual-Driven Domain Decomposition", Machine Learning: Science and Technology, DOI: 10.1088/​2632-2153/​ae8638, Vol. 7, No. 045025, July 2026.
      BibTeX TR2026-116 PDF Software
      • @article{Botvinick-Greenhouse2026jul,
      • author = {Botvinick-Greenhouse, Jonah and Ali, Wael H. and Benosman, Mouhacine and Mowlavi, Saviz},
      • title = {{AB-PINNs: Adaptive-Basis Physics-Informed Neural Networks for Residual-Driven Domain Decomposition}},
      • journal = {Machine Learning: Science and Technology},
      • year = 2026,
      • volume = 7,
      • number = 045025,
      • month = jul,
      • doi = {10.1088/2632-2153/ae8638},
      • url = {https://www.merl.com/publications/TR2026-116}
      • }

    Access software at https://github.com/merlresearch/ab-pinns.