Sample Classification using Machine Learning-Assisted Entangled Two-Photon Absorption

Abstract

Entangled two-photon absorption (eTPA) has been recognized as a potentially powerful tool for the implementation of ultra-sensitive spectroscopy. Unfortunately, there exists a general agreement in the quantum optics community that experimental eTPA signals, particularly those obtained from molecular solutions, are extremely weak. Consequently, obtaining spectroscopic information about an arbitrary sample via conventional methods rapidly becomes an unrealistic endeavor. To address this problem, we introduce an experimental scheme that reduces the amount of data needed to identify and classify unknown samples via their electronic structure. Our proposed method makes use of machine learning (ML) to extract information about the number of intermediate levels that participate in the two-photon excitation of the absorbing medium. This is achieved by training artificial neural networks (ANNs) with various eTPA signals where the delay between the absorbed photons is externally controlled. Inspired by multiple experimental studies of eTPA, we consider model systems comprising one to four intermediate levels, whose energies are randomly chosen from four different intermediate-level band gaps, namely: $\Delta\lambda = 10$, $20$, $30$, and $40$ nm. Within these band gaps, and with the goal of testing the efficiency of our artificial intelligence algorithms, we make use of three different wavelength spacing $1$, $0.5$ and $0.1$ nm. We find that for a proper entanglement time between the absorbed photons, classification average efficiencies exceed 99$\%$ for all configurations. Our results demonstrate the potential of artificial neural networks for facilitating the experimental implementation of eTPA spectroscopy.