We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cond-mat.mtrl-sci

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Condensed Matter > Materials Science

Title: Ultra-fast interpretable machine-learning potentials

Abstract: All-atom dynamics simulations are an indispensable quantitative tool in physics, chemistry, and materials science, but large systems and long simulation times remain challenging due to the trade-off between computational efficiency and predictive accuracy. To address this challenge, we combine effective two- and three-body potentials in a cubic B-spline basis with regularized linear regression to obtain machine-learning potentials that are physically interpretable, sufficiently accurate for applications, as fast as the fastest traditional empirical potentials, and two to four orders of magnitude faster than state-of-the-art machine-learning potentials. For data from empirical potentials, we demonstrate exact retrieval of the potential. For data from density functional theory, the predicted energies, forces, and derived properties, including phonon spectra, elastic constants, and melting points, closely match those of the reference method. The introduced potentials might contribute towards accurate all-atom dynamics simulations of large atomistic systems over long time scales.
Subjects: Materials Science (cond-mat.mtrl-sci); Computational Physics (physics.comp-ph)
Cite as: arXiv:2110.00624 [cond-mat.mtrl-sci]
  (or arXiv:2110.00624v1 [cond-mat.mtrl-sci] for this version)

Submission history

From: Richard G. Hennig [view email]
[v1] Fri, 1 Oct 2021 19:23:06 GMT (410kb,D)

Link back to: arXiv, form interface, contact.