Can a machine learn chemistry?

Speaker: Jesús Pérez Ríos, Department of Physics and Astronomy, Stony Brook University (USA)

We live in a new scientific paradigm: the big data era. Instead of using computers to simulate physical models in this paradigm, the idea is to use machine learning techniques through computers to simulate physical models from the data itself. This paradigm has proven helpful in simulating realistic physical, biological, and chemical models, yielding impressive results. On the other hand, the insight gained via machine learning techniques could be used to answer fundamental questions. In that regard, we want to answer the question: Can a machine learn chemistry? In our group, we plan to find an answer by analyzing first the simplest molecules: diatomics. Specifically, we will show that it is possible to predict molecular properties of diatomic molecules (spectroscopic constants and dipole moments) from atomic ones. Indeed, the level of accuracy is nearly as good as the gold standard in quantum chemistry. Next, we will discuss our results on predicting atom-diatom reactions, showing that it is possible to predict the outcome of a reaction across the chemical space given by the isotopologues when using artificial neural networks. Finally, we will discuss how to generalize our approach to other chemical reactions. 

Coffee and cake will be served.

Michael Drewsen