Computational screening of materials with extreme gap deformation potentials

Abstract

In this work, we present a large-scale study of gap deformation potentials based on density-functional theory calculations for over 5000 semiconductors. As expected, in most cases the band gap decreases for increasing volume with deformation potentials that can reach values of almost −15 eV. We find, however, also a sizeable number of materials with positive deformation potentials. Notorious members of this group are halide perovskites, known for their applications in photovoltaics. We then focus on understanding the physical reasons for so different values of the deformation potentials by investigating the correlations between this property and a large number of other material and compositional properties. We also train explainable machine learning models as well as graph convolutional networks to predict deformation potentials and establish simple rules to understand predicted values. Finally, we analyze in more detail a series of materials that have record positive and negative deformation potentials.