Regensburg 2025 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

O: Fachverband Oberflächenphysik

O 85: New Methods: Theory

O 85.1: Talk

Thursday, March 20, 2025, 10:30–10:45, H25

Quantifying the conductance of molecular structural variables using machine learning — •Hector Vazquez — Inst. of Physics, Czech Academy of Sciences

In single molecule electronics, where individual molecules are placed between two nanoscale electrodes, conductance depends critically on the geometry at the junction. Atomistic simulations using DFT-NEGF are ideally suited to address this, but their computational cost restricts their use to only few junction geometries. In experiments, however, molecular geometry is thought to change significantly since measurements are often carried out at room temperature.

Here we use an approximate method to calculate molecular conductance within DFT for thousands of geometries [1]. The method uses small Au-molecule-Au clusters and is thus computationally very efficient, yet reproduces DFT-NEGF conductance well. Combined with MD simulations of the junction, we compute for thousands of geometries the variation in conductance arising from thermally-induced conformational changes in the molecule.

We use machine-learning methods to identify which of the molecular structural parameters, all of which are changing continuously and simultaneously during the MD simulations, have a greater impact in conductance. This elucidates how molecular conformational changes contribute to the width of the conductance signal in single molecule junctions.

[1] H. Vazquez, J. Phys. Chem. Lett. 13 9326 (2022)

Keywords: Single molecule conductance; Machine learning; Molecular electronics; Structure-conductance relationships; Molecular junctions