Searching for new heavy fermions with deep learning

TL;DR

This paper employs deep learning to predict properties of potential new heavy fermion compounds, using a compiled database and chemical composition as the primary predictor, aiming to accelerate discovery in condensed matter physics.

Contribution

It introduces a deep learning framework for predicting heavy fermion properties and classifying their physical states based solely on chemical composition.

Findings

Successful regression predictions of coherence temperature, Sommerfeld coefficient, and effective mass.

Classification models effectively identified superconducting and antiferromagnetic heavy fermions.

Proposed future database expansion and AI methods for improved predictions.

Abstract

Deep learning models were developed and implemented to aid the search for new heavy fermion compounds. For the purpose of these calculations a database of more than 200 heavy fermions was compiled from the literature. The deep learning networks trained on the database were then used for regression calculations, and predictions were made about the coherence temperature, Sommerfeld coefficient and carrier effective mass of potential new heavy fermions. Classification calculations were also performed in order to check whether predicted heavy fermions are superconducting and/or antiferromagnetic. Chemical composition was the only physical predictor used during the learning process. Suggestions were made for future improvements in terms of expanding the database, as well as for other artificial intelligence calculations.