Unsupervised machine learning for supercooled liquids

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Qiu, Yunrui, Jang, Inhyuk, Huang, Xuhui, Yethiraj, Arun
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910422653730816
author Qiu, Yunrui
Jang, Inhyuk
Huang, Xuhui
Yethiraj, Arun
author_facet Qiu, Yunrui
Jang, Inhyuk
Huang, Xuhui
Yethiraj, Arun
contents Unraveling the relation between structural information and the dynamic properties of supercooled liquids is one of the grand challenges of physics. Dynamic heterogeneity, characterized by the propensity of particles, is often used as a proxy for the dynamic slowing down. In this work, we introduce an unsupervised machine learning approach based on a time-lagged autoencoder (TAE) to elucidate the effect of structural features on the long-time dynamic heterogeneity of supercooled liquids. The TAE uses an autoencoder to reconstruct features at time $t + Δt$ from input features at time $t$ for individual particles, and the resulting latent space variables are considered as order parameters. In the Kob-Andersen system, with a $Δt$ about a thousand times smaller than the relaxation time, the TAE order parameter exhibits a remarkable correlation with the long-time propensity. We find that radial features on all length-scales are required to capture the long-time dynamics, consistent with recent simulations. This shows that fluctuations of structural features contain sufficient information about the long-time dynamic heterogeneity.
format Preprint
id arxiv_https___arxiv_org_abs_2404_04473
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Unsupervised machine learning for supercooled liquids
Qiu, Yunrui
Jang, Inhyuk
Huang, Xuhui
Yethiraj, Arun
Disordered Systems and Neural Networks
Unraveling the relation between structural information and the dynamic properties of supercooled liquids is one of the grand challenges of physics. Dynamic heterogeneity, characterized by the propensity of particles, is often used as a proxy for the dynamic slowing down. In this work, we introduce an unsupervised machine learning approach based on a time-lagged autoencoder (TAE) to elucidate the effect of structural features on the long-time dynamic heterogeneity of supercooled liquids. The TAE uses an autoencoder to reconstruct features at time $t + Δt$ from input features at time $t$ for individual particles, and the resulting latent space variables are considered as order parameters. In the Kob-Andersen system, with a $Δt$ about a thousand times smaller than the relaxation time, the TAE order parameter exhibits a remarkable correlation with the long-time propensity. We find that radial features on all length-scales are required to capture the long-time dynamics, consistent with recent simulations. This shows that fluctuations of structural features contain sufficient information about the long-time dynamic heterogeneity.
title Unsupervised machine learning for supercooled liquids
topic Disordered Systems and Neural Networks
url https://arxiv.org/abs/2404.04473