Lecture 5

Machine-Learned Interatomic Potentials (MLIP)

Accelerating molecular dynamics simulations with GNNs - achieving quantum accuracy at classical cost. In this lecture, we will explore how to represent molecules as graphs and how to build and train graph neural networks (GNNs) for predicting energies and forces of atomic configurations.

Madeline Murphy
90 min
Back to schedule
Slides (PDF) Recording (YouTube) Companion Tutorial



Ryan et al. — "A Practical Guide to Machine Learning Interatomic Potentials" Paper
2025 · Overview of best practices for building and training ML interatomic potentials
Reiser et al. — "Graph Neural Networks for Materials Science and Chemistry" Paper
Communications Materials, 2022 · Review of GNN architectures applied to materials and chemical systems
"A Critical Review of ML Interatomic Potentials and Hamiltonian" Paper
Journal of Material Informatics, 2025 · Critical assessment of ML interatomic potential and Hamiltonian learning methods
"Reactive Machine Learning Interatomic Potentials" Paper
Chemical Reviews, 2026 · Review of MLIPs capable of modeling bond breaking and forming
Bronstein et al. — "Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges" Paper
2021 · Unifying framework for deep learning on non-Euclidean data
Gilmer et al. — "Neural Message Passing for Quantum Chemistry" Paper
ICML, 2017 · Introduces the MPNN framework unifying prior GNN architectures
Batzner et al. — "NequIP" Paper
Nature Communications, 2022 · E(3)-equivariant GNN interatomic potential
Musaelian et al. — "Allegro" Paper
Nature Communications, 2023 · Local equivariant GNN potential designed for scalability
Batatia et al. — "MACE" Paper
NeurIPS, 2022 · Higher-order equivariant message passing interatomic potential
Dusson et al. — "Atomic Cluster Expansion (ACE)" Paper
2019 · Systematic body-ordered expansion for atomic descriptors
Bartók et al. — "Smooth Overlap of Atomic Positions (SOAP)" Paper
2010 · Rotationally invariant descriptor of local atomic environments
Behler & Parrinello — "Atom-Centered Symmetry Functions (ACSF)" Paper
2007 · Symmetry function descriptors for neural network potentials