|
|
|
THE FORTHCOMING KAPUY LECTURE
|
|
|
is scheduled for the 25th of Szeptember, 2025
|
|
|
starting at 3:00 PM in auditory 063 (Bruckner)
|
|
|
at the Institute of Chemistry,
ELTE Eötvös Loránd University,
|
|
1117 Budapest, Pázmány
Péter sétány 1/A
|
|
|
The lecture will be delivered by
|
|
Prof. Martin Head-Gordon
|
|
( UC Berkeley, USA)
|
|
titled
|
|
|
Perturbation theory for electron correlation made better and faster?
|
Rayleigh-Schrödinger perturbation theory, and its ab initio application to many-electron systems via Moller-Plesset (MP) perturbation theory, is not regular order by order, and therefore can exhibit erratic behavior for systems with small HOMO-LUMO gaps, and even for intermolecular interaction energies of large molecules. With this in mind, in the first part of this talk, I will discuss a new approach to regularize MP2 theory against divergence in the small gap limit. The idea is to modify Brillouin-Wigner perturbation theory to be size-consistent as well as regular at second order to define BW-s2. This can also be viewed as a form of Rayleigh-Schrodinger perturbation theory with a non-Moller-Plesset partitioning, which appears to offer considerable promise.
My second topic is revisiting the design of local correlation methods at the doubles level, for MP2 and BW-s2, as well as higher level methods. The main challenge for local correlation is to achieve full control over errors such that a user only needs to select a single numerical drop tolerance. To achieve this goal, we have designed a new “single threshold” approach to local correlation, that also avoids use of projected AOs and PNOs to span the virtual space, by instead employing a localized orthogonal virtual basis. Accuracy and performance will be assessed via a range of example calculations. Kapuy’s double perturbation theory plays a significant role in this approach, as it does in many local correlation models. We will show that our latest implementation of this single-threshold numerical sparsity approach provides significantly higher numerical accuracy and/or significantly lower compute cost than the domain-localized pair natural orbital scheme, as implemented in ORCA.
|
|
|
Click here to view the official invitation card.
|
Members