(197bb) Physics-Informed Bayesian Optimization Framework for Material Discovery

The lack of efficient discovery tools for advanced functional materials is a major bottleneck to enabling future-generation energy, health, and sustainability technologies. One main factor contributing to this inefficiency is the large combinatorial space of materials which is very sparsely observed. Searches of this large combinatorial space are often biased by expert knowledge and clustered close to material configurations that are known to perform well. Moreover, experimental characterization or first principles quantum mechanical calculations of all possible materials are extremely expensive leading to small data sets that are not suitable for a number of approaches, such as Deep Learning. As a result, there is a need for the development of computational algorithms that can efficiently search this large space for a given material application. Here, we introduce a method that combines a physics-informed belief model with Bayesian optimization. The material is characterized by physical and chemical properties of components of the material in a complex manner but a priori knowledge of the identity of the important properties is often lacking. The key contributing factor of our proposed framework is in the creation of a hypothesis space with all possible Gaussian process representations of the domain using these different elemental/molecular properties and the ability to select the hypothesis (belief model) that best represents our material design domain. The best hypothesis is then used to perform a search of the material space. Our method is unique since it picks out the physical descriptors that are most representative of the material domain making the search unbiased toward expert knowledge, which in many cases is unknown. The model also provides valuable chemical insight into the domain that can be used to develop new materials that were outside the domain that was initially searched.