Numerical Accuracy Matters: Applications of Machine Learned Potential Energy Surfaces

2023/11/29 by Silvan Käser, Käser, Silvan, Markus Meuwly +1 · 2 citations
Materials Science · #Chemical Physics (physics.chem-ph) #FOS: Physical sciences #Machine Learning in Materials Science

paper · pdf · doi:10.48550/arxiv.2311.17398

openalex publication_date 2023/11/29 · openalex created_date 2023/12/01 · openalex updated_date 2026/07/28

Abstract

The role of numerical accuracy in training and evaluating neural network-based potential energy surfaces is examined for different experimental observables. For observables that require third- and fourth-order derivatives of the total energy with respect to Cartesian coordinates single-precision arithmetics as is typically used in ML-based approaches is insufficient and leads to roughness of the underlying PES as is explicitly demonstrated. Increasing the numerical accuracy to double-precision yields a smooth PES with higher-order derivatives that are numerically stable and yield meaningful anharmonic frequencies and tunneling splitting as is demonstrated for H₂CO and malonaldehyde. For molecular dynamics simulations, which only require first-order derivatives, single-precision arithmetics appears to be sufficient, though.

Cited by

• Perturbatively corrected ring-polymer instanton rate theory rigorously captures anharmonicity and deep tunneling
• Dynamics of Protonated Oxalate from Machine-Learned Simulations and Experiment: Infrared Signatures, Proton Transfer Dynamics and Tunneling Splittings

Related

• Sampling-free Inference for Ab-Initio Potential Energy Surface Networks
• On the Effect of Aleatoric and Epistemic Errors on the Learnability and Quality of NN-based Potential Energy Surfaces
• The Bigger the Better? Accurate Molecular Potential Energy Surfaces from Minimalist Neural Networks
• Machine Learning Potential Energy Surfaces
• Neural Scaling Laws Surpass Chemical Accuracy for the Many-Electron Schrödinger Equation