Accelerating Discovery of Extreme Lattice Thermal Conductivity by Crystal Attention Graph Neural Network (CATGNN) Using Chemical Bonding Intuitive Descriptors

TL;DR

This paper introduces a machine learning approach using crystal bonding descriptors and a graph neural network to efficiently predict and discover inorganic materials with extreme lattice thermal conductivities, aiding thermal management applications.

Contribution

The study develops novel bonding descriptors and a GNN model that outperform empirical rules, enabling rapid screening and discovery of materials with ultralow or ultrahigh LTC.

Findings

Identified 106 stable compounds with LTC <5 W/mK

Discovered 13 stable compounds with LTC >100 W/mK

Descriptors outperform empirical rules in predicting LTC

Abstract

Designing materials with targeted lattice thermal conductivity (LTC) demands electronic-level insight into chemical bonding. We introduce two bonding descriptors, namely normalized negative integrated crystal orbital Hamilton populations (-ICOHP) and normalized integrated crystal orbital bond index (ICOBI), that strongly correlate with LTC and rattling (mean-squared displacement), surpassing empirical rules and the unnormalized -ICOHP across >4,500 inorganic crystals by first-principles. We train a Crystal Attention Graph Neural Network (CATGNN) to predict these descriptors and screen ~200,000 database structures for extreme LTCs. From 367 (533) candidates with low (high) normalized -ICOHP and normalized ICOBI, first-principles validation identifies 106 dynamically stable compounds with LTC <5 W/mK (68% <2 W/mK) and 13 stable compounds with LTC >100 W/mK. The descriptors' low cost and clear physical meaning provide a rapid, reliable route to high-throughput discovery and inverse design of crystalline materials with ultralow or ultrahigh LTC for applications in thermal insulation, thermoelectrics, and electronics cooling.