Speaker
Prof.
Francesco Paesani
(University of California, San Diego)
Description
Two of the most challenging problems at the frontier of
contemporary electronic structure theory are the
accurate
representation of intermolecular interactions and the
development of reduced-scaling algorithms applicable to
large systems. To some extent, these two problems are
antithetical, since the accurate calculation of non-
covalent
interactions typically requires correlated, post-Hartree–
Fock
methods whose computational scaling with respect to
system
size precludes the application to large systems. In this
talk, I
describe our theoretical/computational framework (MB-
pol)
for the development of chemically accurate and
transferable
intermolecular potentials derived entirely from “first
principles”. MB-pol potentials are built upon the many-
body
expansion of molecular interactions with explicit 2-body
and
3-body terms represented by high-degree invariant
polynomials obtained from application of machine
learning
techniques to CCSD(T)/CBS reference data. These terms
smoothly transition at intermediate range into a sum of
electrostatic and dispersion interactions that reproduce
the
correct asymptotic behavior. The induction contributions
to
non-pairwise additive interactions are taken into account
using polarizable point dipoles. The accuracy of the MB-
pol
approach is assessed here through the analysis of the
properties of water from the gas to the condensed
phase with
a particular emphasis on nuclear quantum effects and
vibrational spectroscopy.
