Influence of Additional Gaussian Noises on Mixed Quantum-Classical Nonadiabatic Dynamics Simulations of Photoisomerization of <i>cis</i>-Azobenzene

An efficient potential energy surface from cutting-edge technologies such as quantum computing and deep learning has been incorporated into mixed quantum-classical dynamics. However, the intrinsic noise embedded in those methodologies continues to be the sword of Damocles, as the simulation results of nonadiabatic dynamics are heavily dependent on the numerical stability of potential energy surfaces as well as nonadiabatic couplings. To address this concern, we perform surface hopping and Ehrenfest mean field dynamics simulations on the photoisomerization of <i>cis</i>-azobenzene and investigate the influence of additional noises on the collective results by introducing Gaussian random numbers into on-the-fly electronic structure calculations at each dynamic step. Noises added to nuclear gradients reduce the stability of the dynamic simulations regardless of the method employed. The excited-state lifetime simulated with the original fewest switches surface hopping method is decreased in the presence of Gaussian noise applied to nonadiabatic coupling terms, while the simulation with the branching correction mean field method exhibits the opposite tendency. Finally, we observe that branching correction surface hopping, which combines the robustness of nuclear motion during surface hopping dynamics and the elimination of accumulated perturbations by resetting the electronic density matrix for decoherence correction, is less sensitive to additional noises.