Exploring the attosecond laser-driven electron dynamics in the hydrogen molecule with different real-time time-dependent configuration interaction approaches

TL;DR

This paper develops algorithms to efficiently model laser-driven electron dynamics in molecules using real-time configuration interaction methods, significantly reducing computational cost while maintaining accuracy in high harmonic spectra predictions.

Contribution

The authors introduce algorithms to select dominant eigenstates in RT-TDCISD, enabling two orders of magnitude reduction in computational complexity for modeling strong field electron dynamics.

Findings

Electron dynamics mainly involve a small subspace of single excitations.

Proper eigenstate selection preserves accuracy in high harmonic spectra.

Reduced models significantly cut computational costs without loss of key observables.

Abstract

Time-dependent quantum chemical methods coupled to Gaussian basis sets are gaining popularity in modeling the electron dynamics of atoms and molecules interacting with intense laser fields. Two approaches most widely used for this purpose, the real-time time-dependent configuration interaction singles and the real-time time-dependent density functional theory, both have their limitations, so the development of more accurate yet computationally efficient time-dependent methods is still in demand. In this work we explore the applicability of the real-time time-dependent configuration interaction singles and doubles (RT-TDCISD) in modeling strong field phenomena. Since the main drawback of RT-TDCISD is its unfavourable scaling, we develop several algorithms for reducing the effective propagation space by selecting these CISD eigenstates that should have dominant contribution to the time-evolution of the wavefunction. We test them by performing calculations of the high harmonic spectra of the H\textsubscript{2} molecule. We find out that the laser-driven electron dynamics is mostly realized in a very small subspace of eigenstates dominated by single excitations, that constitutes about one percent of the whole CISD eigenspectrum. Therefore, by properly selecting this subspace one can reduce the dimension of the propagation equation by two orders of magnitude without affecting the time-resolved observables.