Atomic Scattering For Chemical Analysis Of Surfaces

TL;DR

This paper develops machine learning techniques to analyze Helium spin-echo spectroscopy data, enabling ultra-sensitive surface chemical analysis by modeling co-adsorbed systems and estimating partial surface concentrations with high accuracy.

Contribution

It introduces deep learning methods for chemical sensitivity analysis in Helium spin-echo spectroscopy, achieving quantification of surface concentrations in complex systems.

Findings

Partial surface concentrations are obtainable up to 20% occupancy.

Deep learning models can handle noise levels of 4%.

Modeling bi-species co-adsorption improves surface analysis accuracy.

Abstract

The study explores machine learning methods for revealing chemical sensitivity in Helium spin-echo spectroscopy, in order to obtain ultra-sensitive surface analytic technique. We model bi-species co-adsorbed systems and demonstrate that by using deep-learning neural-networks partial surface concentrations are obtainable. An example system of particles with mass 50 and 100 a.m.u was tested with characteristic inter-adsorbate and adsorbate-substrate interactions, with partial surface concentrations being resolvable up to 20% occupancy of adsorption sites, and with modestly high noise level of 4%.