Machine learning using structural representations for discovery of high temperature superconductors

TL;DR

This paper develops a machine learning approach using structural representations to efficiently predict high temperature superconductors, significantly aiding the exploration of vast compositional spaces in condensed matter physics.

Contribution

It introduces a novel structural representation for machine learning that accurately predicts superconducting transition temperatures, facilitating accelerated discovery of new high-Tc materials.

Findings

Achieved an $r^2$ above 0.94 in predicting $T_c$.

Enabled fast screening of structural polymorphisms under pressure.

Demonstrated the effectiveness of ML in exploring complex phase spaces.

Abstract

The expansiveness of compositional phase space is too vast to fully search using current theoretical tools for many emergent problems in condensed matter physics. The reliance on a deep chemical understanding is one method to identify local minima of relevance to investigate further, minimizing sample space. Utilizing machine learning methods can permit a deeper appreciation of correlations in higher order parameter space and be trained to behave as a predictive tool in the exploration of new materials. We have applied this approach in our search for new high temperature superconductors by incorporating models which can differentiate structural polymorphisms, in a pressure landscape, a critical component for understanding high temperature superconductivity. Our development of a representation for machine learning superconductivity with structural properties allows fast predictions of superconducting transition temperatures ($T_c$) providing a $r^2$ above 0.94.