Machine learning yields new candidates for layered superconductors
Machine learning yields new candidates for layered superconductors lead image
Most of today’s superconductors have been discovered through a combination of trial-and-error alongside the experience of researchers, which tend to incur high costs and potentially take long periods of time to explore. Mapping chemical components of materials to superconducting critical temperatures is extremely complicated. One new approach looks to use materials informatics and machine learning to point a path forward for superconductor discovery.
Liu et al. present a machine learning model for efficiently predicting critical temperatures of layered high-temperature superconductors based on material descriptors with the help of materials informatics. The multi-step learning and multi-algorithm cross-verification approach draws on molecular attributes such as a compound’s average atomic mass, the average number of electrons in unfilled shell and the average ground state atomic magnetic moments.
“In this study, we let computers learn the hidden laws of known layered superconductors in a database and then screen for new layered superconductors not known before,” said author Zhong-Li Liu. “Our findings will guide experiments that synthesize new, high-temperature superconductors.”
The group screened for approximately 2,500 layered materials from the 200,000 materials in the Inorganic Crystal Structure Database. They then used the model to screen for 25 potential new layered superconductors, including 12 cuprates, 7 iron-based crystals and 6 others.
The group’s approach predicts excellent critical temperature values with over 92 percent confidence, ranging from about 32 kelvins to about 138 kelvins.
Liu said that of all the descriptors used with their machine learning model, the maximum difference of electronegativity appeared to be most important.
Liu hopes that such work stimulates more experimental work in layered high-critical-temperature materials and new findings in the mechanism of non-cuprate superconductors.
Source: “Material informatics for layered high-TC superconductors,” by Zhong-Li Liu, Peng Kang, Yu Zhu, Lei Liu, and Hong Guo, APL Materials (2020). The article can be accessed at https://doi.org/10.1063/5.0004641 .
