Dear Group,
Just a reminder that Mark Watson will be presenting his research talk at 5pm
TODAY at 60 Oxford Street, Room 311. See abstract below.
Thanks,
Anna
************************
Dr. Mark Watson
University of Tokyo
Monday 15 December, 5:00pm
60 Oxford Street, Room 311
'Towards an accurate and efficient density-functional theory for large
molecules'
Density-functional theory (DFT) has emerged over the last 2 decades to
become the preeminent method for ab initio computational chemistry studies
in a wide variety of systems. This is surely due in large part to the
extremely favourable balance between accuracy and computational cost that
DFT can provide. In my talk, I discuss a contribution to the on-going
development of DFT methodology with respect to both of these points. I
first discuss a newly proposed hybrid exchange-correlation functional based
on the long-range correction (LC) scheme, which can dramatically improve
some of the failings of conventional GGA and hybrid
functionals, such as the poor description of Rydberg or charge-transfer
excitation energies. The new functional improves previous LC functionals by
providing a simultaneously good description of a wide range of excitation
energies, reaction barrier heights, and thermochemistry. Next, I discuss
our recent contributions to
linear-scaling DFT, with the introduction of two new methods for the rapid
calculation of the expensive electronic Coulomb potential in large
molecules, which is a traditional bottleneck for DFT calculations. Both
approaches employ real-space techniques and the fast multipole method to
achieve linear-scaling of CPU time. The first solves the Poisson equation
in a mixed basis of Gaussian and low-order finite-element functions. The
second uses a basis of high-order, adaptive, spectral
elements and avoids the Poisson equation by an efficient direct integration
of the potential using a numerical low-rank representation of the Coulomb
operator. Both approaches are shown to be highly effective for large-scale
calculations.
Show replies by thread