Deep Learning Interatomic Potentials

Machine learning interatomic potentials (MLIPs) based on deep equivariant neural networks achieve quantum-accurate results, but realizing their full potential requires careful attention to software design and computational efficiency. I led the redesign of the NequIP software framework for deep equivariant graph neural network potentials, focusing on robustness, extensibility, and performance. The rewrite achieved 5× speedups for training and 5–18× speedups for molecular dynamics simulations.

  1. C.W. Tan , M.L. Descoteaux , M. Kotak , G. Miranda Nascimento , S.R. Kavanagh , L. Zichi , M. Wang , A. Saluja , Y.R. Hu , T. Smidt , others . High-performance training and inference for deep equivariant interatomic potentials. Digital Discovery 5, 1558–1567 (2026).

Building on the NequIP software infrastructure, I support and maintain a variety of extension packages that extend the capabilities of our MLIP architectures to new modeling capabilities.

  1. S. Falletta , A. Cepellotti , A. Johansson , C.W. Tan , M.L. Descoteaux , A. Musaelian , C.J. Owen , B. Kozinsky . Unified differentiable learning of electric response. Nature Communications 16, 4031 (2025).
  1. L.A. Gomes , S. Larmore , M. Wang , C.W. Tan , B. Kozinsky , S.A. Lopez . Machine-learned nonadiabatic couplings enable reactive photochemical reaction dynamics simulations. ChemRxiv (2026).
  1. G.d.M. Nascimento , M.L. Descoteaux , L. Zichi , C.W. Tan , W.C. Witt , N. Molinari , S. Mantha , D. Kitchaev , M. Kornbluth , K. Gadelrab , others . Mixture of Experts Framework in Machine Learning Interatomic Potentials for Atomistic Simulations. arXiv preprint arXiv:2604.26143 (2026).

Computational Materials Screening

Computational materials screening cover

Computational materials discovery often prioritizes device performance while overlooking materials-level sustainability. To address this, I developed a sustainability-guided materials screening protocol and applied it to ultrawide bandgap layered materials. This work identified 25 low-risk, sustainable ultrawide bandgap layered candidate semiconductors through in silico materials screening for nanoelectronic device applications, such as dielectric, power-electronics, and ultraviolet-photonics.

  1. C.W. Tan , L. Xu , C.C. Er , S.P. Chai , B. Kozinsky , H.Y. Yang , S.A. Yang , J. Lu , Y.S. Ang . Toward sustainable ultrawide bandgap van der Waals materials: An ab initio screening effort. Advanced Functional Materials, 2308679 (2023).

Orbital-Free Density Functional Theory

Orbital-free density functional theory (OFDFT) is a promising method for faster, more large-scale atomistic simulations. My prior work involves method development for OFDFT, including a differentiable OFDFT code, PROFESS-AD.

  1. C.W. Tan , C.J. Pickard , W.C. Witt . Automatic differentiation for orbital-free density functional theory. The Journal of Chemical Physics 158, 124801 (2023).
  1. W.C. Witt , B.W. Shires , C.W. Tan , W.J. Jankowski , C.J. Pickard . Random structure searching with orbital-free density functional theory. The Journal of Physical Chemistry A 125, 1650–1660 (2021).