Learning Conductance: Gaussian Process Regression for Molecular Electronics.

Area of Science:

• Molecular electronics
• Quantum transport phenomena
• Computational condensed matter physics

Background:

• Experimental charge transport studies use break junction setups, yielding statistical conductance distributions.
• Modeling these distributions and experimental structural changes is computationally demanding for theoretical methods.
• Accurate quantum transport calculations are limited to few conformations, neglecting experimental conformational diversity.

Purpose of the Study:

• To investigate machine learning, particularly Gaussian process regression, for modeling molecular conductance histograms.
• To reduce the computational expense associated with simulating conductance histograms.
• To enable routine evaluation of conductance histograms using computationally intensive methods.

Main Methods:

• Utilizing Gaussian process regression with specific structural parameters as features.
• Predicting zero-bias conductance from molecular structures.
• Reducing the number of required charge transport calculations.

Main Results:

• Gaussian process regression efficiently predicts zero-bias conductance.
• Computational cost for simulating conductance histograms is reduced by an order of magnitude.
• Enables efficient calculation of conductance histograms even with first-principles approaches.

Conclusions:

• Machine learning offers a viable solution for the computational challenges in modeling conductance histograms.
• This approach facilitates routine evaluation of conductance histograms, bridging theory and experiment.
• Paves the way for more comprehensive understanding of charge transport in molecular junctions.