Source-linked AI summary

Fast and Accurate Modeling of Molecular Atomization Energies with Machine Learning

Matthias Rupp, Alexandre Tkatchenko, Klaus-Robert Müller, O. Anatole von Lilienfeld

arXiv:1109.2618v1physics.chem-phcond-mat.dis-nncond-mat.mtrl-scistat.ML

TL;DR

Accurate molecular atomization energies are difficult to obtain because exact Schrödinger-equation solutions are intractable beyond very small systems. The paper replaces explicit solution with machine-learning regression from molecular nuclear charges and positions, achieving 9.9 kcal/mol MAE across 7165 molecules and predicting atomization-energy curves.

  • Problem

    Exact Schrödinger-equation solutions for assemblies of atoms are intractable beyond the smallest systems, motivating computationally efficient approximations.

  • Method

    The model represents molecules with Coulomb matrices and learns a nonlinear map from nuclear charges and atomic positions to atomization energies computed at the PBE0 DFT level.

  • Results

    9.9 kcal/mol MAE was obtained by cross-validation on 7165 small organic molecules, with transferability to 6000 held-out molecules at roughly 15 kcal/mol accuracy.

  • Takeaways & Limitations

    The approach predicts atomization energies for new molecules and beyond equilibrium geometries, supporting faster exploration of molecular chemical compound space.

  • Takeaways & Limitations

    Reference atomization energies used for training were computed at the PBE0 DFT level of theory.

Abstract

from arXiv · show

We introduce a machine learning model to predict atomization energies of a diverse set of organic molecules, based on nuclear charges and atomic positions only. The problem of solving the molecular Schrödinger equation is mapped onto a non-linear statistical regression problem of reduced complexity. Regression models are trained on and compared to atomization energies computed with hybrid density-functional theory. Cross-validation over more than seven thousand small organic molecules yields a mean absolute error of ~10 kcal/mol. Applicability is demonstrated for the prediction of molecular atomization potential energy curves.

Loading 1109.2618v1…