First-principles based physics-informed models for sensitivity and thermo-mechanical of energetic crystals

Development of first-principles based property databases for energetics materials and comparing them with available experimental records are currently done by multiple sources. Benchmarking of first-principles methods for a set of known energetic crystals and creating a repository is an important need for the community. In this effort, we have made progress in benchmarking different dispersion corrected DFT methods and compared their accuracy against experimental datasets. We will discuss the results from first-principles training data for bulk modulus and pressure derivative, temperature-dependent equation of state, thermal expansion coefficient, isobaric heat capacity, Grüneisen parameter, Hugoniot EoS and ranking for shock sensitivity using an overtone-based model. This data set is used for training a machine learning model that can provide initial estimates for properties of unknown HE crystals and co-crystals. The physics-informed machine learning models and the order of features of importance for sensitivity are also pointing to the complexity of sensitivity prediction models long recognized in the community. Integration of models for property predictions grounded in fundamental electronic structure and crystal packing perform better than past empirical models.