Fig. 1: Materials design workflow used in this work.
From: Identification of high-dielectric constant compounds from statistical design
The three parts of the design workflow shown here together complete a single design cycle. The newly computed DFPT results from a design cycle are fed back into the training-data for the next design cycle. Since the dielectric tensor of a crystal is of shape 3 × 3, optimizing all nine components of the dielectric tensor leads to a nine-objective optimization problem. Thus, we specifically optimize the largest value of dielectric constant (ϵ) among all crystallographic directions, also referred to as the target property hereinafter. This scalar value is quantified as the largest eigenvalue of the total dielectric tensor. The training-data in this work consist of nearly 2000 compounds from the MP for which ϵ and Eg were known and the training-data came from the OQMD, for which only Eg were known at the beginning of this work.
