Improving single-molecule conductance measurements with change point detection from the econometrics toolbox

• Main Authors: Hamill, Joseph M, Bro-Jørgensen, William, Balogh, Zoltán, Li, Haixing, Leitherer, Susanne, Solomon, David, Halbritter, András, Solomon, Gemma
• Format: Journal Article
• Language: English
• Published: 23-01-2024
• Subjects: Physics - Mesoscale and Nanoscale Physics; Physics - Soft Condensed Matter
• DOI: 10.48550/arxiv.2401.12769

Structural breaks occur in timeseries data across a broad range of fields,
from economics to nanosciences. For measurements of single-molecule break
junctions, structural breaks in conductance versus displacement data occur when
the molecular junction ruptures. This moment is significant because the
molecule is likely in its most extended geometry, and therefore resembles most
closely the geometry used in theoretical predictions. Conventional
single-molecule break junction data analysis, on the other hand, typically uses
the entire molecular plateau to estimate the single-molecule conductance, which
skews the estimate when the plateau is sloped. Borrowing from econometrics,
where the study of structural breaks is well established, we present change
point detection (CPD) as a tool to search for junction rupture in
single-molecule break junction data, and improve estimates in single-molecule
conductance. We demonstrate that using CPD instead of the conventional 1D
conductance histogram to determine the mean molecular conductance yields a
standard deviation in the estimate of typically half that of the conventional
approach, greatly improving accuracy. We apply CPD to three separate data sets,
two on 4,4'-bipyridine and one on a silane, two at room temperature and one at
4 K, two in one lab, one in another, to show the wide applicability of even the
simplest of CPD algorithms: the Chow test. This versatility and better accuracy
will propagate into more accurate theoretical simulations. These improved
metrics, in turn, will further improve any downstream analyses, including all
emerging machine learning approaches.