Data-Driven Materials Design

Automated workflows for data generation and -exploitation (supervised and unsupervised) have long been employed in fields like bio-chemistry and drug discovery, but are only now making their way into materials science. I will begin this talk by introducing the Atomic Simulation Recipes (ASR) – an open source Python framework for constructing and executing materials simulation workflows [1, 2]. Next, I will give some examples of its use in materials design studies including the search for indirect band gap semiconductors for thin-film photovoltaics [3] and the Computational 2D Materials Database (C2DB), which contains calculated properties of several thousand 2D materials in monolayer and bilayer form [4]. Finally, I will discuss how deep generative models can help to suggest new types of materials by learning the patterns in known data sets of stable materials [5]. If time allows, I will discuss machine learning of GW band structures from descriptors encoding the electronic structure of standard DFT calculations [6].