Probabilistic deep learning approach for targeted hybrid organic-inorganic perovskites

TL;DR

This paper introduces a probabilistic deep learning model to efficiently screen and predict the electronic band gaps of diverse hybrid organic-inorganic perovskites, accounting for data uncertainty and identifying promising new materials.

Contribution

It presents a novel probabilistic machine learning approach for material property prediction that explicitly models uncertainty due to atomic structure diversity.

Findings

Identified dozens of new HOIP formulas with band gaps between 1.25 and 1.50 eV.

Validated predictions against first-principles computations.

Demonstrated robustness and versatility of the probabilistic approach.

Abstract

We develop a probabilistic machine learning model and use it to screen for new hybrid organic-inorganic perovskites (HOIPs) with targeted electronic band gap. The data set used for this work is highly diverse, containing multiple atomic structures for each of 192 chemically distinct HOIP formulas. Therefore, any property prediction on a given formula must be associated with an irreducible "uncertainty" that comes from its unknown atomic details. As a result, dozens of new HOIP formulas with band gap falling between 1.25 and 1.50 eV were identified and validated against suitable first-principles computations. Through this demonstration we show that the probabilistic deep learning approach is robust, versatile, and can be used to properly quantify this uncertainty. In conclusion, the probabilistic standpoint and approach described herein could be widely useful for the very common and inevitable data uncertainty which is rooted at the incompleteness of information during experiments and/or computations.