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

Download:

Current browse context:

physics.comp-ph

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

Physics > Computational Physics

Title: Solving Newton's Equations of Motion with Large Timesteps using Recurrent Neural Networks based Operators

Abstract: Classical molecular dynamics simulations are based on solving Newton's equations of motion. Using a small timestep, numerical integrators such as Verlet generate trajectories of particles as solutions to Newton's equations. We introduce operators derived using recurrent neural networks that accurately solve Newton's equations utilizing sequences of past trajectory data, and produce energy-conserving dynamics of particles using timesteps up to 4000 times larger compared to the Verlet timestep. We demonstrate significant speedup in many example problems including 3D systems of up to 16 particles.
Comments: 15 pages, 12 figures; updated content
Subjects: Computational Physics (physics.comp-ph); Soft Condensed Matter (cond-mat.soft); Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:2004.06493 [physics.comp-ph]
  (or arXiv:2004.06493v3 [physics.comp-ph] for this version)

Submission history

From: Vikram Jadhao [view email]
[v1] Sun, 12 Apr 2020 16:15:21 GMT (1812kb,D)
[v2] Sun, 17 May 2020 23:09:34 GMT (2076kb,D)
[v3] Tue, 14 Dec 2021 00:08:23 GMT (2108kb,D)

Link back to: arXiv, form interface, contact.