Forward and inverse design1,2 of equilibrium materials properties have been accelerated with data-driven methods.3Electronic materials exhibiting phase transitions4,5between metastable states, such as metal-insulator transition (MIT) materials6with abrupt electrical resistivity transformations, however, are challenging to decode. Conventional machine learning methods display limited predictive capability for MIT materials, because data scarcity and the absence of features7,8impede model training. Here we demonstrate a discovery strategy based on multi-objective Bayesian optimization to directly circumvent these bottlenecks by utilizing latent variable Gaussian processes combined with high-fidelity electronic structure calculations for validation in the chalcogenide lacunar spinel family. We directly and simultaneously learn phase stability and band gap tunability from chemical composition alone to efficiently discover 12 superior compositions on the Pareto front. Previously unidentified electronic transitions also emerge from our featureless adaptive optimization engine (AOE). Our methodology readily generalizes to optimization of multiple properties, enabling co-design of complex multifunctional materials especially where prior data is sparse.
|Original language||English (US)|
|State||Published - Apr 15 2020|
ASJC Scopus subject areas