The quantum mechanical mapping from structure to energy does not have rules like games do. Hence, it would seem that there is no machine learning algorithm that could predict energies without having to solve the complex QM equations. In my talk, I will show that, contrary to expectations, one can train a neural network to predict energies, forces, and charges, with results statistically indistinguishable from the true QM results. <br/> <br/>The applications are limitless, but perhaps more interesting is the understanding of how and what has the program learned about chemistry. I will show examples of how one can relate simple chemical concepts to individual atomic energies predicted by out method<br/>

Can the rules of quantum chemistry be learned? A perfect force field without a functional form

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.