Does Hessian Data Improve the Performance of Machine Learning Potentials?

Austin Rodriguez, Justin S. Smith, Jose L. Mendoza-Cortes · 2025

Read Does Hessian Data Improve the Performance of Machine Learning Potentials? by Austin Rodriguez, Justin S. Smith, Jose L. Mendoza-Cortes (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-07839.txt.

Description

Austin Rodriguez, Justin S Smith, and Jose L Mendoza-Cortes ask whether adding Hessian data improves machine learning models of how atoms interact, published in 2025. Such models predict energies and forces at near quantum accuracy, and are used in chemistry, materials, and drug design. Adding force data to training already helps a lot.

Hessian data add the curvature of the energy surface, a second-order detail. The team trains models on stable molecule shapes only, then tests them on off-balance shapes. Models trained with Hessian data predicted energies, forces, and Hessians better for those new shapes. They also modeled reaction paths and transition states more exactly, and gave better vibration spectra, while needing less data overall in the comparisons.

Catalog record

CREATORAustin Rodriguez, Justin S. Smith, Jose L. Mendoza-Cortes
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE72 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-07839

Browse the full library · About the Tower of Babel