EXAFS analysis using Reverse Monte Carlo and machine learning: XAS Journal Club, Janis Timoshenko

Title: Beyond the first peak: EXAFS analysis using Reverse Monte Carlo and machine learning
Speaker: Dr. Janis Timoshenko (Fritz-Haber Institute, Berlin, Germany)
Abstract: Extended X-ray absorption fine structure (EXAFS) spectroscopy is an invaluable tool for the characterization of local environment in a broad range of materials, including well-defined crystalline materials, disordered nanomaterials and functional materials experiencing structural and compositional changes under working conditions. Information on the arrangement of atoms at distances up to 10 Å from the absorbing atom is often recorded in experimental spectra. In well-ordered materials, the conventional approaches for EXAFS data analysis can provide an accurate description of the radial distribution of atoms within the first coordination shell around the absorber. However, the conventional EXAFS fitting often cannot be easily extended to the analysis of contributions of distant coordination shells, many-atom distribution functions and to the studies of strongly disordered materials, hindering the application of EXAFS method for the identification of 3D structures of complex materials, for studies of small nanoparticles, materials at high temperature and functional materials experiencing phase transitions. Here we discuss two complementary techniques that can address this problem. Reverse Monte Carlo (RMC) simulations combine the structure optimization methods with ab-initio EXAFS calculations, to find in an iterative process a 3D structure model of the material, consistent with the available experimental information.[1,2] On the other hand, machine learning (ML) based approach for inversion of EXAFS spectra, after a training on a large set of theoretical spectra, provides the ability to quickly extract relevant structure descriptors from the experimental data, and is particularly useful for the studies of the materials that cannot be easily described with a single structure model (e.g., mixtures, or materials experiencing in-situ transformations).[3,4] Both methods provide access to distant coordination shells and can be applied to studies of strongly disordered materials, allowing studies of complex materials under harsh environment.[5]

Suggested Reading:
[1] J. Timoshenko et al, J. Phys.: Condens. Matter 26 055401 (2014), https://doi.org/10.1088/0953-8984/26/...
[2] J. Timoshenko et al, Annu. Rev. Anal. Chem. 12, 501 (2019), https://doi.org/10.1146/annurev-anche...
[3] J. Timoshenko, et al, Phys. Rev. Lett., 120, 225502 (2018), https://doi.org/10.1103/PhysRevLett.1...
[4] J. Timoshenko, A. I. Frenkel, ACS Catal., 9, 10192 (2019), https://doi.org/10.1021/acscatal.9b03599
[5] J. Timoshenko, B. Roldan Cuenya. Chem. Rev. 121, 882 (2021), https://doi.org/10.1021/acs.chemrev.0...