Disorder signatures in coherent electronic waveguides

Abstract

Microscopic disorder in a coherent conductor is encoded in the magnitude of the transmitted current and in the energy- and channel-resolved structure of the scattering response. We develop a conductance-balanced learning framework for identifying microscopic disorder mechanisms from coherent quantum-transport spectra. Using armchair graphene nanoribbons as controlled multichannel tight-binding waveguides, we generate ensembles in which distinct disorder classes occupy the same intervals of integrated transmission, removing the dominant shortcut associated with average conductance. Within this constrained setting, we compare scalar transmission spectra with transmission-eigenvalue spectra, and use supervised classification, scale-normalizing controls, plateau-resolved tests, and principal-component analysis to identify the spectral structures that remain informative. The learned distinctions persist after normalization of the conductance envelope and are strongest in multichannel energy windows, where mode mixing and channel redistribution shape the scattering response. Principal components provide interpretable transport coordinates whose loadings identify the energy and eigenchannel sectors responsible for the dominant spectral deformations. The results establish a controlled AI-assisted protocol for learning disorder signatures from nonlinear quantum-scattering data.