During my time at the University of Florida I got involved in a theory-experiment collaboration focused on the optimization of new superconductors discovery.
Almost surprisingly, nowadays the search for new superconductors is carried out in nearly the same manner as decades ago: each study focuses on a class of materials, and discoveries are made more or less out of serendipity. Increases in computing power and novel machine learning algorithms offer new opportunities to revolutionize superconductor discovery enabling the rapid prediction of structures and properties of novel materials in an automated, high-throughput fashion.
A first step in this direction can be found in our recent work, published in npj Computational Materials, where we show that modern machine-learning techniques can substantially improve the estimate of the critical temperature (the temperature below which superconductivity occurs). We overcome the limit of the tiny dataset, limited to a few low-T c superconductors, including a database of artificially generated material-specific Eliashberg functions. The main result is the derivation of a formula for the critical temperature that performs as well as the traditional one for low-Tc superconductors and substantially better for higher-Tc ones.
This is just a start! We are currently working to identify the best descriptors to be included in the machine learning database as well we to develope new approaches to extend our analysis beyond electron-phonon mediated superconductivity and explore spin/charge mediated superconductivity. You can find an overview of our research activity in our contribution to the 2021 room-temperature superconductivity roadmap.
Laura Fanfarillo 