Graduate student John Snyder (Burke group) has used machine learning, a branch of artificial intelligence, to find the most accurate density functional ever, as reported in a recent Physical Review Letter. The table is a small subset of the 500,000 numbers that characterize the new approximation. This functional is only a demo for a model problem, and he is now trying to apply these methods to general electronic structure problems, in collaboration with computer scientists from TU Berlin and quantum chemists from Yonsei University, South Korea.
Friday, June 29, 2012
Phys Rev Lett. 108, 253002