Electronic structure theory describes the motions of electrons in atoms or molecules, and provides a versatile framework for the calculation of molecular geometries, chemical bonding, electronic and spectroscopic properties, reaction barriers, intermolecular interactions, and more. Wave function theory-based methods such as coupled-cluster methods, provide accurate results in a systematic manner, but they typically carry a significant computational cost.
The Vogiatzis group uses machine learning, the field of study allowing computers to learn without explicit programming, to provide novel approaches on the learning of the underlying electronic structure while subverting a significant portion of the computational expense. In a recent article published in the Journal of Chemical Theory and Computation, Jacob Townsend, under the supervision of Kostas Vogiatzis, presents a new efficient methodology that explores the local nature of the correlated motion of electrons which offers scalability and transferability between different chemical systems.
The novel approach provides coupled-cluster quality electronic energies at the cost of second-order perturbation theory (MP2), a computationally more affordable method. As a result, the authors were able to predict energies of a large molecular database, known as the GDB-9 dataset, at high accuracy at 
a fraction of the cost. Additionally, it was shown that the introduced method could be used to accurately predict energies of large chemical systems based on models trained on smaller ones.
Townsend recently received his PhD in the chemistry doctoral program at UT.