Work

A deep neural network approach for predicting the melting point of low-molecular phase change materials

Year: 2026

Type: article

Abstract: The melting point of phase change material (PCM) is a critical property that determines practical applications, but its prediction remains a challenge, particularly for low-molecular PCMs. In this work, we develop a deep-learning neural network to predict the melting points of low-molecular PCMs with high efficiency and accuracy. To train the proposed models, we build a dataset containing over 650,000 configurations of various molecular systems (sugars, polyols, fatty acids, amides, etc.) generated from large-scale molecular dynamics simulations. We demonstrate that the developed model can successfully predict the melting points of diverse low-molecular PCMs, exhibiting impressive accuracy with root mean square errors of 4.2 and 3.0 K for crystals and liquids, respectively. The proposed model is general and could be applicable to predict the critical points of other phase transitions, providing an appealing framework for the rational design of PCMs and expediting material discovery with desirable functionalities.

Source: npj Soft Matter

DOI: 10.1038/s44431-025-00017-2

Authors: Jingsi Liao, Wenjin Gao, Ke Xu, Wenbo Han, Yibin Li, Miao Zhou

Institutions: Harbin Institute of Technology, Tianmushan Laboratory, Beihang University

Language: English

Topic: Machine Learning in Materials Science

Subfield: Materials Chemistry

Field: Materials Science

Domain: Physical Sciences

Open Access status: hybrid

Estimated APC paid: $2,590