Abstract
A protocol for generating potential energy surfaces and performing photoinduced nonadiabatic multidimensional wave packet propagation is presented. The workflow starts with the parameterization of a linear vibronic coupling (LVC) Hamiltonian using the BSE@GW approach. In a second step, the LVC model is used as input for multi-layer multi-configurational time-dependent Hartree (ML-MCTDH) wave packet propagation. To facilitate automated ML tree generation, a spectral clustering algorithm is applied based on a correlation matrix obtained from nuclear coordinate expectation values of a full-dimensional Time-dependent Hartree (TDH) simulation. The performance of the protocol is tested on the photoinduced spin-vibronic dynamics of a transition metal complex, [Fe(cpmp)]$^{+2}$. For this example, it is shown that BSE@GW provides a more robust description of the character of the transitions contributing to the absorption spectrum compared to TD-DFT. Furthermore, the LVC parameterization is tested against explicit calculations of potential energy curves to find the validity of the linear approximation over a wide range of normal mode elongation. Finally, the flexibility of spectral clustering is used to generate different ML trees, resulting in very different numerical efficiencies for ML-MCTDH propagation. In terms of electronic structure and dimensionality, [Fe(cpmp)]$^{+2}$ is a challenging example, suggesting that the new protocol should be applicable to a wide range of systems.
Supplementary materials
Title
additional results
Description
The ESI contains further information on the choice of the basis set, the validity of LVC model, the structure of the ML-trees, the adjacency matrix of model (IV), the population dynamics for a larger convergence threshold, a 42-dimensional model, and some discussion of accuracy of state couplings.
Actions