Figure 28.

(a) Data flow for using OpenMolcas and LVC models with SHARC: A frequency calculation provides the normal modes and reference harmonic oscillator. The SHARC–OpenMolcas interface then computes energies (E), gradients (Grad), NAC vectors, (optionally) SOCs, and (transition) dipole moments (DM). These data are transformed into normal-mode coordinates to obtain a LVC template. Using suitable initial conditions, LVC parameters are used to simulate nonadiabatic trajectories via the pysharc driver.^172,500 (b) Electronic population of SHARC/LVC dynamics of the VCl₃(ddpd) complex on a picosecond time scale.⁵⁰¹ (c) Data flow for using SchNarc ML models with SHARC and OpenMolcas: Desired properties (E, Grad, NAC, SOC, DM) are calculated with OpenMolcas and stored via the SHARC–OpenMolcas interface and the SchNarc package in a database. SchNarc uses this database to train machine learning (ML) models that predicts all required properties needed by pysharc. Together with suitable initial conditions, nonadiabatic dynamics trajectories are generated.⁴⁹⁸ (d) A trajectory of a roaming hydrogen atom in excited tyrosine, based on simulations on a picosecond time-scale.⁵⁰²