Lookup NU author(s): Dr Thomas Penfold
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).
Methods using a swarm of Gaussian basis functions to represent the nuclear wavefunction are a very appealing way to solve the time-dependent Schrödinger equation (TDSE) as they avoid the conventional scaling bottleneck of grid-based methods and provide a grid-free trajectory rep- resentation of the dynamics understudy. When coupled with direct (on-the-fly) dynamics, these methods offer the ability to simulate quantum dynamics of larger systems in full nuclear config- uration space and avoid the requirement of a priori fitting of a potential energy surface. During such simulations, it is often assumed that the limiting factor is the computational cost of the quan- tum chemistry calculations. To combat this, in the present paper the direct dynamics variational multi-configurational Gaussian (DD-vMCG) method is combined with electronic structure calcula- tions accelerated by Graphical Processing Units (GPUs). For the systems studied, a protonated ammonia dimer and the imidazole dimer, it is shown that the cost of the term responsible for the quantum behaviour of the nuclear dynamics means that the computational time associated with the quantum chemistry quickly becomes a small part of the overall computational time. Us- ing these simulations, an estimated scaling of the vMCG method, with respect to the number of Gaussian basis functions is reported. This can be used to identify when quantum chemistry is the limiting factor and when GPU acceleration will have a significant effect for both ground and excited state simulations.
Author(s): Penfold TJ
Publication type: Article
Publication status: Published
Journal: Physical Chemistry Chemical Physics
Print publication date: 14/08/2017
Online publication date: 31/03/2017
Acceptance date: 30/03/2017
Date deposited: 31/03/2017
ISSN (print): 1463-9076
ISSN (electronic): 1463-9084
Publisher: Royal Society of Chemistry
Altmetrics provided by Altmetric