Exploring the robust extrapolation of high-dimensional machine learning potentials

TL;DR

This paper investigates the extrapolation behavior of high-dimensional machine learning potentials, revealing they often predict outside the training data's convex hull, and offers a probabilistic framework to understand their domain of robustness.

Contribution

It challenges assumptions about the domain of ML potentials and introduces a probabilistic perspective to rationalize their extrapolation capabilities.

Findings

Predictions often occur outside the convex hull of training data.

A probabilistic framework explains the domain of robust extrapolation.

Insights could improve the design of more reliable ML potentials.

Abstract

We show that, contrary to popular assumptions, predictions from machine learning potentials built upon high-dimensional atom-density representations almost exclusively occur in regions of the representation space which lie outside the convex hull defined by the training set points. We then propose a perspective to rationalize the domain of robust extrapolation and accurate prediction of atomistic machine learning potentials in terms of the probability density induced by training points in the representation space