Christopoulou, Georgia; (2021) Developing and applying efficient DD-vMCG method for nonadiabatic simulations. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text G_Christopoulou_Thesis.pdf - Accepted Version Download (13MB) | Preview

Abstract

Gaussian wavepacket methods have been widely employed for the investigation of nonadiabatic molecular dynamics. The Direct Dynamics variational Multi-Configurational Gaussian (DD-vMCG) method provides a fully quantum mechanical solution to the time-dependent Schrödinger equation for the time evolution of nuclei with potential surfaces calculated on-the-fly using a quantum chemistry program. The first strand of this research study was to develop new, more efficient algorithms and improve the existing code for DD-vMCG aiming to increase both the accuracy and efficiency of this method. Thus, a new, efficient parallel algorithm to control the DD-vMCG database of quantum chemistry points is presented along with improvements to the interpolation scheme. Benchmark calculations on butatriene, allene and formamide showed that the new scheme is a very accurate, efficient and general method to employ for full-dimensional dynamics calculations. The aforementioned algorithm was then used to describe the photodissociation dynamics of phenol including all degrees of freedom, as the second strand of this research work was to explore more complex chemical systems. Full-dimensional quantum dynamics calculations including for the first time six electronic states, along with a detailed comparison with existing 3-state and 4-state models are presented. Including the fifth singlet excited state has been shown to be vital in unravelling the photodissociation of phenol. State population and flux analysis provided new insights into the decay mechanism of phenol confirming the idea of rapid relaxation to the ground state through the ¹ππ/1¹πσ* conical intersection. Finally, an effort to further improve the accuracy of DD-vMCG was made by employing a state-of-the-art approach where a Gaussian process regression scheme is introduced and machined-learned potential energy surfaces are obtained. All the findings suggest that this method could be promising to calculate potential energy surface matrix elements. However, further development is essential to take advantage of its benefits and to deal with the computational cost.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as JSONCSVXML

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item