# Benchmarking Compositional Generalisation for Machine Learning Interatomic Potentials

arXiv: https://arxiv.org/abs/2605.08988 · HTML: https://mlsanity.com/paper/2605.08988/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 44 references that are in the map.
- First posted on arXiv 9 May 2026.

## Builds on (most cited first, 22 of 22)

- [Neural Message Passing for Quantum Chemistry](https://mlsanity.com/paper/1704.01212/) (2017)
- [E(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentials](https://mlsanity.com/paper/2101.03164/) (2021)
- [MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields](https://mlsanity.com/paper/2206.07697/) (2022)
- [Equivariant message passing for the prediction of tensorial properties and molecular spectra](https://mlsanity.com/paper/2102.03150/) (2021)
- [GemNet: Universal Directional Graph Neural Networks for Molecules](https://mlsanity.com/paper/2106.08903/) (2021)
- [Compositionality decomposed: how do neural networks generalise?](https://mlsanity.com/paper/1908.08351/) (2019)
- [Fast and Uncertainty-Aware Directional Message Passing for Non-Equilibrium Molecules](https://mlsanity.com/paper/2011.14115/) (2020)
- [EquiformerV2: Improved Equivariant Transformer for Scaling to Higher-Degree Representations](https://mlsanity.com/paper/2306.12059/) (2023)
- [What Algorithms can Transformers Learn? A Study in Length Generalization](https://mlsanity.com/paper/2310.16028/) (2023)
- [Reducing SO(3) Convolutions to SO(2) for Efficient Equivariant GNNs](https://mlsanity.com/paper/2302.03655/) (2023)
- [UMA: A Family of Universal Models for Atoms](https://mlsanity.com/paper/2506.23971/) (2025)
- [OOD-GNN: Out-of-Distribution Generalized Graph Neural Network](https://mlsanity.com/paper/2112.03806/) (2021)
- [Orb: A Fast, Scalable Neural Network Potential](https://mlsanity.com/paper/2410.22570/) (2024)
- [Isambard-AI: a leadership class supercomputer optimised specifically for Artificial Intelligence](https://mlsanity.com/paper/2410.11199/) (2024)
- [GemNet-OC: Developing Graph Neural Networks for Large and Diverse Molecular Simulation Datasets](https://mlsanity.com/paper/2204.02782/) (2022)
- [Transition1x -- a Dataset for Building Generalizable Reactive Machine Learning Potentials](https://mlsanity.com/paper/2207.12858/) (2022)
- [PET-MAD, a lightweight universal interatomic potential for advanced materials modeling](https://mlsanity.com/paper/2503.14118/) (2025)
- [Structure-based out-of-distribution (OOD) materials property prediction: a benchmark study](https://mlsanity.com/paper/2401.08032/) (2024)
- [May the Force be with You: Unified Force-Centric Pre-Training for 3D Molecular Conformations](https://mlsanity.com/paper/2308.14759/) (2023)
- [FlashMD: long-stride, universal prediction of molecular dynamics](https://mlsanity.com/paper/2505.19350/) (2025)
- [Beyond MD17: the reactive xxMD dataset](https://mlsanity.com/paper/2308.11155/) (2023)
- [Systematic Relational Reasoning With Epistemic Graph Neural Networks](https://mlsanity.com/paper/2407.17396/) (2024)

## Abstract (arXiv)

Machine Learning Interatomic Potentials play a fundamental role in computational chemistry and materials science, enabling applications from molecular dynamics simulations to drug design and materials discovery. While recent approaches can estimate inter-atomic forces with high precision, it remains unclear to what extent they can generalise to previously unseen molecules. Do they learn the compositional structure of chemistry, capturing how molecular fragments and their combinations determine properties, or do they primarily learn to interpolate patterns that are specific to the training examples? To address this question, we propose a benchmark consisting of four tasks that require some form of compositional generalisation. In each task, models are tested on molecules that were unseen during training, but the training data is chosen such that generalisation to the test examples should be feasible for models that learn the underlying physical principles. Our empirical analysis shows that the considered tasks are highly challenging for state-of-the-art models, with errors on out-of-distribution examples often an order of magnitude higher than on in-distribution examples, even when using foundation models that have been pre-trained on millions of molecules.