Importance Sampling for Equilibrium Properties in Non-Adiabatic Systems: Application to Anomalous Interfacial Charge Transfer

Interfacial charge transfer plays a significant role
in the performance
of optoelectronic devices. Compared with real-time non-adiabatic dynamics
simulations, statistical mechanics could obtain the equilibrium properties
directly. We here present an importance sampling approach to calculate
the electron distribution at general interfaces. The configuration
distribution based on a series of reference potential energy surfaces
is utilized to reweight the properties of interest. The high accuracy,
precision, and efficiency of the new method are verified by comparing
the results of different sampling approaches and non-adiabatic dynamics
simulations. We find that numbers of states, electronic couplings,
reorganization energies, and energetic disorders all have strong impacts
on the anomalous interfacial charge transfer across a high energy
barrier. In addition, based on the data computed by importance sampling
in relatively small but diverse systems, we propose an effective machine
learning approach to train artificial neural networks, which could
predict the interfacial charge transfer in extremely large systems.
Due to its high performance, our importance sampling method is promising
for the statistical study of equilibrium properties in general non-adiabatic
systems.