Neural Network Approach for Characterizing Structural Transformations by X-Ray Absorption Fine Structure Spectroscopy

Abstract

The knowledge of the coordination environment around various atomic species in many functional materials provides a key for explaining their properties and working mechanisms. Many structural motifs and their transformations are difficult to detect and quantify in the process of work (operando conditions), due to their local nature, small changes, low dimensionality of the material, and/or extreme conditions. Here we use an artificial neural network approach to extract the information on the local structure and its in situ changes directly from the x-ray absorption fine structure spectra. We illustrate this capability by extracting the radial distribution function (RDF) of atoms in ferritic and austenitic phases of bulk iron across the temperature-induced transition. Integration of RDFs allows us to quantify the changes in the iron coordination and material density, and to observe the transition from a body-centered to a face-centered cubic arrangement of iron atoms. This method is attractive for a broad range of materials and experimental conditions.

Solving the Structure and Dynamics of Metal Nanoparticles by Combining X-Ray Absorption Fine Structure Spectroscopy and Atomistic Structure Simulations

“…Note also that a number of approaches have been proposed to further refine MD models and to improve agreement with experimental data: (a) isotropic rescaling of model coordinates to account for systematic errors in interatomic distances (22); (b) running MD simulations at a different temperature than the temperature at which the experimental data were collected to account for under-or overestimated atomic vibrations (22,115); (c) refining RDFs via the socalled histogram approach (23,116) or via fine-tuning the cumulants of bond-length distributions (117,118); or (d) fine-tuning the whole 3D structural model by complementing MD simulations with a consequent RMC run (39). An alternative approach has recently been proposed, where instead of comparing MD-EXAFS spectra directly with experiment, they are used to train an artificial neural network, which is later applied to extract RDF from experimental EXAFS data (119,120).…”

Linking the evolution of catalytic properties and structural changes in copper–zinc nanocatalysts using operando EXAFS and neural-networks

“…The trained NN can then be used to directly invert experimental data, and quickly and accurately extract relevant structural information, such as partial radial distribution functions (RDFs). In our previous works, we demonstrated the power of this approach in studies of bulk metals, 45 monometallic nanoparticles, 39,46 as well as bimetallic model systems. 46 However, the previous examples were limited to the applications of the NN-EXAFS method to completely metallic systems.…”

Infrared spectroscopy data- and physics-driven machine learning for characterizing surface microstructure of complex materials

“…Toward characterizing the structure of real nanomaterials, reverse Monte Carlo analysis of X-ray absorption fine structure spectroscopy (EXAFS) data has proven successful in resolving the structure of bimetallic nanoparticles with atomic resolution 8 . Neural networks, trained on X-ray absorption near edge structure (XANES) data, predict average coordination numbers of coordination shells 9 and radial distribution functions 10 given experimental spectra from monometallic nanoparticles. XAS spectroscopy is, though, a bulk technique 11,12 , whereas many phenomena, such as catalysis, depend directly on surface properties.…”