Machine Learning\nModels for High Explosive Crystal\nDensity and Performance

The rate of discovery of new explosives with superior\nenergy density\nand performance has largely stalled. Rapid property prediction through\nmachine learning has the potential to accelerate the discovery of\nnew molecules by screening of large numbers of molecules before they\nare ever synthesized. To support this goal, we assembled a 21,000-molecule\ndatabase of experimentally synthesized molecules containing energetic\nfunctional groups. Using a combination of experimental density measurements\nand high throughput electronic structure and atomistic calculations,\nwe calculated detonation velocities and pressures for all 21,000 compounds.\nUsing these values, we trained machine learning models for the prediction\nof density, detonation velocity and detonation pressure. Notably,\nour model for crystal density surpassed the accuracy of all current\nmodels and decreased the root-mean square error (RMSE) of the previous\nbest model by 20%. This improvement in model performance relative\nto past works is attributed to our handling of chiral-specified Simplified\nMolecular-Input Line-Entry System (SMILES) strings and introduction\nof a new molecular descriptor, MolDensity. To elucidate descriptor\nimportance, we evaluated interpretable descriptors in terms of importance\nand compared the accuracy of a statistics-driven machine learning\nmodel against a model comprised of descriptors typically assumed to\ncontrol material density. The inexpensive, yet highly accurate predictions\nfrom our models should enable creation of future artificial intelligence\n(AI) models that are able to screen large numbers (>10<sup>6</sup>) of compounds to find the highest performing compounds in terms\nof crystal density, detonation velocity and detonation pressure.