Modelling Ultrafast Dynamics at a Conical Intersection with Regularized Diabatic States: An Approach Based on Multiplicative Neural Networks

Neural-network (NN) potentials are employed in conjunction with the Multi-Configuration Time-Dependent Hartree (MCTDH) method in order to simulate an ultrafast photoinduced cis-trans type isomerization process induced by a conical intersection. To this end, NN potentials are fitted to a diabatic potential of regularized diabatic states type [Köppel et al., J. Chem. Phys. 115, 2377 (2001)], which entirely relies on adiabatic potential information. Multiplicative NNs are employed which match the sum-of-products form of the multiconfigurational MCTDH wavefunction. Good agreement with the reference dynamics is obtained for simulations of the highly correlated dynamics with up to 13 degrees of freedom. This study contributes to developing NN methodologies suitable for photochemical dynamics at complex excited-state topologies.