Ab lnitio Simulations and Materials Chemistry in the Age of Big Data

Abstract

In this perspective, we discuss computational advances in the last decades, both in algorithms as well as in technologies, that enabled the development, widespread use, and maturity of simulation methods for molecular and materials systems. Such advances led to the generation of large amounts of data, which required the creation of several computational databases. Within this scenario, with the democratization of data access, the field now encounters several opportunities for data-driven approaches toward chemical and materials problems. Specifically, machine learning methods for predictions of novel materials or properties are being increasingly used with great success. However, black box usage fails in many instances; several technical details require expert knowledge in order for the predictions to be useful, such as with descriptors and algorithm selection. These approaches represent a direction for further developments, notably allowing advances for both developed and emerging countries with modest computational infrastructures.