Combining Multiscale MD Simulations and Machine Learning Methods to Study Electronic Transport in Molecular Junctions at Finite Temperatures

Abstract

We propose an efficient method to analyze the influence of thermal fluctuations of a molecular junction on its electronic transport properties, and consequently, reliably predict the time-averaged value of the conductance at a finite temperature in the zero-bias regime. Our multiscale approach combines three complementary techniques, namely, large-scale quantum mechanics/molecular mechanics (QM/MM) molecular dynamics (MD) simulations, active machine learning methods, and nonequilibrium Green’s function transport calculations. Results for the exemplary Au(111)-S-C6H4-C6H4-S-Au(111) and Au(111)-N-C6H4-C6H4-N-Au(111) junctions indicate the substantial impact of the thermal evolution of the junction on its transport properties, which cannot be forecasted based just on the ground-state geometry. In the experimentally relevant temperature range around the room temperature, the predicted conductance values are 30-40% larger than those calculated for the minimum-energy configuration of the respective junction at 0 K.