machine-learning-potentials

🌟 [NeurIPS '25 Spotlight] Fair and transparent benchmark of machine learning interatomic potentials (MLIPs), beyond basic error metrics https://openreview.net/forum?id=SAT0KPA5UO

A simple and fast python library to handle the data generated from molecular dynamics simulations

Deep Potential Evolution Accelerator

Take the pain out of installing and running machine-learning interatomic potentials: every major MLIP in one registry, set up, verified and run by your LLM.

Reference-audited AI coding-agent skill library for computational chemistry, materials modeling, atomistic simulation, scientific ML, and related workflows.

This repository contains a suite of scripts designed to automate thermodynamic properties of materials, using MatterSim, a universal machine learning interatomic potential, for geometry optimization and force constant calculations, integrated with Phonopy-QHA for Quasi-Harmonic Approximation.

Awesome benchmark list for MLIPS or MLFFs or MLPs or LAMs

Autonomous active-learning pipeline (MACE + jobflow) that trains an MLIP to DFT-level accuracy, automates CI-NEB calculations and extracts Brønsted–Evans–Polanyi scaling relations for H₂ dissociation on N-doped graphene-supported Pd single-atom catalysts.

Scripts for "Hierarchical universal open structure dataset for atomistic modeling" paper

PET machine-learning interatomic potentials in C++ with Kokkos: CUDA, HIP and CPU, with a LAMMPS package

How close do universal and DFT-fine-tuned MACE potentials come to DFT for Li transport in Li6PS5Cl? 42 MD trajectories benchmarked against DFT forces and AIMD diffusion.

Code to reproduce "Quantum machine learning interatomic potential" (arXiv:2607.27841): inserting quantum circuits into ANI-style neural network interatomic potentials via transfer learning.

A graphite heat shield heated to 10,000 K and simulated atom by atom with the Orb-v3 machine-learned potential.

Automatic-differentiation-based Gaussian processes for molecular and materials potential energy surfaces.

Code for computationally predicted NaSbBr4 and LiSbBr4 crystal structure models (prototype transplant + MACE + QE PBE-D3). Data: NOMAD doi:10.17172/nomad.j0ce-ctz1

Runnable workflows on the ORCHID data (PhaseForge+, ESPEI, one-shot MLIP phase diagrams)

MLIP screening of Cu bimetallic alloys for acetylene electroreduction — built on EleCatML by J. Chen.

Automated phonon and thermal conductivity screening harness, featuring a 103-compound benchmark exposing systematic anharmonic bias in foundation MLIPs.

Computational investigation of defect chemistry and structural disorder in IGZO, a transparent conducting oxide and amorphous oxide semiconductor, using DFT, molecular dynamics and machine-learning potentials.

Loss-aware converter for computational-chemistry file formats — reports what every conversion keeps, drops, or fabricates.