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


Ancillary-file links:

Ancillary files (details):

Current browse context:


Change to browse by:

References & Citations


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

Condensed Matter > Disordered Systems and Neural Networks

Title: Unraveling intricate processes of glassy dynamics from static structure by machine learning relative motion

Abstract: The recent developments in machine learning methodologies have enabled accurate predictions of glassy dynamics. The machine learning models developed thus far ensure that the hidden characteristic features of the local structures around each particle can be autonomously related to the hotspots where particles are more prone to move. Herein, we present a distinct approach for developing a novel graph neural network model that learns the relative motion between neighboring particle pairs as its target quantity, in addition to the self-displacement of each particle. This enables precise predictions of the distance of separation of each neighbor pair from its static structure. Furthermore, if we enable the model to simultaneously learn the self-displacement and relative motion, it can predict the particle propensity for self-motion within a short period, with an accuracy drastically higher than those of existing models. This is achieved by utilizing the physical information underlying these quantities that complement the intricate dynamics of both the short-time collective vibrations and structural changes. This model represents a new state-of-the-art prediction method for glassy dynamics, thus introducing a new approach for elucidating the crossover from short-time vibrational motion to long-time structural relaxation in glasses.
Comments: 10 pages, 6 figures; SI is included as an ancillary file; Minor revision from the previous version
Subjects: Disordered Systems and Neural Networks (cond-mat.dis-nn); Soft Condensed Matter (cond-mat.soft); Statistical Mechanics (cond-mat.stat-mech)
Cite as: arXiv:2206.14024 [cond-mat.dis-nn]
  (or arXiv:2206.14024v2 [cond-mat.dis-nn] for this version)

Submission history

From: Hayato Shiba [view email]
[v1] Tue, 28 Jun 2022 14:14:01 GMT (4251kb,AD)
[v2] Sat, 16 Jul 2022 22:04:00 GMT (4252kb,AD)

Link back to: arXiv, form interface, contact.