Localized Coulomb Descriptors for the Gaussian Approximation Potential

TL;DR

This paper introduces LC-GAP, a machine learning approach using localized Coulomb matrix-based descriptors to accurately predict atomic properties and energies in molecules, outperforming previous Coulomb matrix methods in efficiency and accuracy.

Contribution

The paper presents a new localized Coulomb matrix descriptor integrated with Gaussian approximation potentials, improving prediction accuracy and computational efficiency for molecular property prediction.

Findings

Successfully predicts atomization energies for larger molecules.

Achieves chemical accuracy in property predictions.

Offers a linear dimensionality representation in local environments.

Abstract

We introduce a novel class of localized atomic environment representations, based upon the Coulomb matrix. By combining these functions with the Gaussian approximation potential approach, we present LC-GAP, a new system for generating atomic potentials through machine learning (ML). Tests on the QM7, QM7b and GDB9 biomolecular datasets demonstrate that potentials created with LC-GAP can successfully predict atomization energies for molecules larger than those used for training to chemical accuracy, and can (in the case of QM7b) also be used to predict a range of other atomic properties with accuracy in line with the recent literature. As the best-performing representation has only linear dimensionality in the number of atoms in a local atomic environment, this represents an improvement both in prediction accuracy and computational cost when considered against similar Coulomb matrix-based methods.