machine-learning-interatomic-potential

Open-source AI infrastructure for materials science

EquiformerV3: Scaling Efficient, Expressive, and General SE(3)-Equivariant Graph Attention Transformers

Generating Deep Potential with Python

Interface for simulations and standalone Python script generation with universal machine-learning interatomic potentials (MACE, CHGNet, SevenNet, NequIP, ORB, Allegro, MatterSim, UPET, GRACE, UMA, ALIGNN-FF)

End-to-end Reaction-Path Modeling from PDB Structures Using Machine-Learning Interatomic Potentials

Machine‑Learning / Molecular‑Mechanics (ML/MM) hybrid calculator and CLI toolset for Mechanistic Investigation of Enzymatic Reactions.

PySlice is a Python package for simulating and analyzing multslice simulations from molecular dynamics trajectories. In addition to standard multislice simulations such as diffraction and HAADF image generation, it implements the TACAW method to convert time-domain electron scattering data into frequency-domain spectra.

MLIP (Machine Learning Interatomic Potential) plugins for ORCA ExtTool (ProgExt) interface.

Model zoo and experimental features of machine learning interatomic potentials.

🦀 CPU-based neighbor list construction in Rust for atomistic simulations — naive O(N²) and cell list O(N) algorithms

MLIP (Machine Learning Interatomic Potential) plugins for Gaussian 16 External interface.

Implement SE(3)-equivariant graph attention transformers for efficient and expressive molecular modeling in PyTorch.

LCAONet - MPNN including electronic structure and orbital information, physically motivatied by the LCAO method.

MLIP (Machine Learning Interatomic Potential) plugins for ML/MM MD simulations with AmberTools25.

Foundation MLIP benchmark for β-Sn (MACE-MPA-0, ORB v3, SevenNet-Omni vs DFT/PBE)

MLIPs as ASE calculators, benchmarked on BCC TiZrNb properties: a distilled compact potential vs pretrained foundation potentials, accuracy vs speed

From-scratch SchNet-style message-passing neural network potential for BCC TiZrNb, benchmarked head to head against descriptor baselines at matched data and budget. Surrogate teacher labels (CHGNet), real-DFT anchors, leakage-safe splits, autograd forces, NVE stability.

PFP/PBE + OpenMX/PBE data for surface energies and works of adhesion at α-CoSn₃ / β-Sn and Si / α-CoSn₃ interfaces — companion to Wang, Tatsumi et al. on β-Sn orientation control via α-CoSn₃ seed layers.

Query-by-committee active learning for TiZrNb machine-learned interatomic potentials: a from-scratch deep ensemble, structural label-budget accounting, and an honest benchmark where random selection wins on a mixed pool and the committee earns a 2.2x label saving on a scarce one

From-scratch Behler-Parrinello neural network potential for BCC TiZrNb: original ACSF descriptors, autograd forces, and a leakage-disciplined benchmark against tuned linear and pair-potential baselines (surrogate teacher labels, not DFT)