By Martin Head-Gordon
Department of Chemistry, University of California
& Chemical Sciences Division, Lawrence Berkeley National Laboratory, Berkeley CA 94720, USA.

I shall discuss progress in the design of a next generation of density functional theories [1,2].  In contrast to most approaches to functional design, we have adopted a combinatorial approach, in which we have trained a huge number of functionals (over 100,000 in the case of the generalized gradient approximation models [1], and over 1010 in the case of meta GGA functionals [2]).  The functional from each class that performs best on independent test data (survival of the most transferable) is self-consistently trained to yield a new generation functional that seems very promising for application purposes.  The resulting functionals involve significantly fewer parameters than many of the best functionals of recent years.

References

1. “ωB97X-V: A 10 parameter range separated hybrid density functional including non-local correlation, designed by a survival of the fittest strategy”, N. Mardirossian and M. Head-Gordon, Phys. Chem. Chem. Phys. 16, 9904-9924 (2014).
2. “Mapping the genome of meta-generalized gradient approximation density functionals: The search for B97M-V”, N. Mardirossian and M. Head-Gordon, J. Chem. Phys. 142, 074111 (2015).