Molecularity: a fast and efficient criterion for probing superconductivity

TL;DR

This paper introduces a new criterion based on electron localization to efficiently identify and predict high-temperature superconductors, especially hydrogen-based ones, using a combination of DFT and machine learning.

Contribution

It extends 3D electron localization descriptors to superconducting states and develops an index for automatic screening and prediction of superconducting critical temperatures.

Findings

Electron localization changes little during superconducting transition.

Molecular character can be characterized within the framework.

The new index improves machine learning predictions of Tc.

Abstract

We present an efficient criterion for probing the critical temperature of hydrogen based superconductors. We start by expanding the applicability of 3D descriptors of electron localization to superconducting states within the framework of superconducting DFT. We first apply this descriptor to a model system, the hydrogen chain, which allows to prove two main concepts: i) that the electron localization changes very little when the transition from the normal to the superconducting state takes place, i.e. that it can be described at the DFT level from the normal state; and ii) that the formation of molecules can be characterized within this theoretical framework, enabling to filter out systems with marked molecular character and hence with low potential to be good superconductors. These two ideas, are then exploited in real binary and ternary systems, showing i) that the bonding type can be characterized automatically; and ii) that this provides a new index which enables to feed machine learning algorithms for a better prediction of critical temperatures. Overall, this sets a grounded theoretical scenario for an automatic and efficient high-throughput of potential hydrogen based superconductors.