Geometry optimization of large molecules with quantum Monte Carlo
Jonas FELDT
Mercredi 18 Septembre 2019, 11h00
bibliothèque LCT, tour 12 - 13, 4ème étage
Computing excited states is highly demanding for electronic structure methods, which often struggle to ensure high accuracy. The alternative framework of quantum Monte Carlo methods which use stochastic algorithms to solve the Schrödinger equation, scale well with system size, and can offer a balanced description of the ground and electronic excited states. In this work I am using the real-space methods variational (VMC) and diffusion Monte Carlo (DMC). The following developments allow for a consistent computational approach for both VMC and DMC without relying on structures determined at a lower level of theory.
First, the necessary requirements with respect to the wave function are explored in the context of a selective configuration interaction method in order to obtain accurate excited state geometries. The efficiency of the optimizations is increased by using an internal coordinate system. Second, the analytical expressions for an approximate scheme to compute the forces in DMC have been implemented and its limitations and possibilities are being explored. Finally, further improvements aiming at the development of improved estimators which reduce the statistical fluctuations of the interatomic forces and address the infinite variance of these are under development.