Classification of superconducting transition temperature over 150K of metal polyhydrides using support vector machine

Abstract

The search for room-temperature superconductors have shed light on metal polyhydrides as their superconducting transition temperature (TC) vary from a few kelvins to near-room temperatures. One major drawback in the quest for hydrogen-based superconductors is that the calculation of TC through the electron-phonon spectral function consumes large amount of time and computational resources. Here we propose a method of using the support vector machine (SVM) to classify the ranges of TC of the metal polyhydrides without actual calculation of the electron-phonon spectral function. The input variables are chosen based on the effect of hydrogen atoms and the electron localization function (ELF). These set of features can be obtained quickly from the electronics simulations, compare with the actual spectral function calculations. We found that the SVM can classify the superconductors with accuracy over 80 percent with respect to all the metal polyhydrides in the dataset. Our goal of this work is to help screen for the high value of TC of the hydrogen-based compounds, and reduce the time required for the direct calculations of TC.