[telotortium]

I was a grad student in an ab initio quantum chemistry group about a decade and a half ago. I was working on using DFT with correction from various post-Hartree-Fock methods for long-range correlation - it worked okay, but it was clear that it would never scale up to large non-crystalline molecules. DFT did somewhat better on solid-state systems. The scaling issue really killed my motivation to work on the field, and led me to taking a master's degree and leaving early. So it's been fascinating to hear about deep learning approaches to computational chemistry recently - almost like the revenge of the molecular mechanics models, which our group disdained a little but was also by far the most-used feature of the software package for which we wrote our codes.