Exploration of nonlinear optical materials by introducing information science

Abstract

A material search scheme is proposed herein to discover novel deep ultraviolet nonlinear optical (NLO) crystals. The scheme combines first-principles calculations of the optical properties, including birefringence and second-harmonic generation (SHG) susceptibility based on the density functional perturbation theory (DFPT), and prediction of the NLO properties using graph neural networks (GNNs). Systematic comparisons using a common computational code revealed that the DFPT method reproduced the experimental values better than the conventional sum-over-states (SOS) method. The accuracy of the scheme for predicting NLO properties was improved by including angle information in the crystal graph representation. By screening 6660 oxides with bandgaps greater than 4.0 eV in the Materials Project (MP) database, nine promising NLO crystals with dynamic stability were discovered. Moreover, new crystals are expected to be identified by extending the range of applications of the developed scheme from oxides to nitrides and sulfides.