Predicting novel superconducting hydrides using machine learning approaches

Abstract

Searching for superconducting hydrides has so far largely focused on finding materials exhibiting the highest possible critical temperatures (Tc). This has led to a bias towards materials stabilised at very high pressures, which introduces a number of technical difficulties in experiment. Here we apply machine learning methods in an effort to identify superconducting hydrides which can operate closer to ambient conditions. The output of these models informs structure searches, from which we identify and screen stable candidates before performing electron-phonon calculations to obtain Tc. Hydrides of alkali and alkaline earth metals are identified as particularly promising; a Tc of up to 115 K is calculated for RbH12 at 50 GPa and a Tc of up to 90 K is calculated for CsH7 at 100 GPa.

Strong correlation between electronic bonding network and critical temperature in hydrogen-based superconductors

“…Attempts have been made to increase T c further through ternary compounds, for instance with H 3 S 1− x P x 18 and Li 2 MgH 16 19 . Aiming at extracting useful information from this dataset, two main routes are being explored: on the one hand, machine learning methods 20 – 22 are starting to be employed to further increase the list of predicted systems, although the obtained new compounds so far do not beat the already known; on the other hand, additional efforts are being invested into classifying these superconductors using simple footprints based on structural, chemical, and electronic properties 14 , 21 , 23 . These studies suggest hydrogen rich systems with highly symmetrical structures and high density of states (DOS) at the Fermi level are the best candidates for high-temperature superconductivity.…”

High-temperature conventional superconductivity in the boron-carbon system: Material trends

“…Combined with methods for high-throughput database (DB) screening and machine learning, CSP methods are an unprecedentedly powerful tool driving a sudden acceleration in material discoveries recently [4,27,28]. However, compared to other problems of material research, their application to superconductivity is still at a very early stage [29][30][31][32] due to two intrinsic problems: (i) for a large class of unconventional superconductors, including the high-T c cuprates, a quantitative theory of superconductivity is currently missing and (ii) for conventional superconductors where, on the other hand, T c can be predicted with quantitative accuracy, the cost of a single T c calculation is too high to directly perform highthroughput screening of large DBs of hypothetical materials [14].…”

Chemical templates that assemble the metal superhydrides

“…6). A rough estimation of Tc using the ratio of the density of hydrogen related states versus the total density of states 42 at the Fermi level show that the mixing metals in superhydrides can potentially improve Tc (Supplementary Fig. 7).…”