A Neural Network Approach for Structural Characterization of Catalysts
The quest for unlocking the elusive nature of catalytic active sites in nanometer-scale catalysts often dictates the make-up of a research team. It includes chemists, able to make well-defined nano-structures that can range from shape-controlled nanoparticles to size-selected clusters to single site catalysts. Characterization experts develop new methods, required for their atomic-level characterization in operando conditions. The interpretation of such data should result in the desired structural details, but that last step is often the most difficult to make, because extracting the real-space structure from experimental spectra is often an ill-posed problem. The “data science revolution” offers new opportunities in this field. Here we report on the use of X-ray absorption spectroscopy (XAS) and supervised machine learning (SML) for determining the three-dimensional geometry of metal catalysts. In our method, artificial neural network is used to rapidly unravel the hidden relationship between the XAS features and catalyst geometry. As a result, computers can be trained to invert the experimental spectrum and obtain the underlying structural descriptors on the fly, during the chemical reaction. I will demonstrate our approach by taking the SML to the task of reconstructing the average size, shape, compositional motifs and morphology of mono- and hetero-atomic nanoparticles that range in size from hundreds of atoms to just a few, from their experimental spectra.
 J. Timoshenko, C. J. Wrasman, M. Luneau, T. Shirman, M. Cargnello, S. R. Bare, J. Aizenberg, C. M. Friend, A. I. Frenkel Nano Letters DOI: 10.1021/acs.nanolett.8b04461
 J. Timoshenko, A. Halder, B. Yang, S. Seifert, M. Pellin, S. Vajda, A. I. Frenkel J. Phys. Chem. C 122, 21686-21693 (2018)
 J. Timoshenko, A. Anspoks, A. Cintins, A. Kuzmin, J. Purans, A. I. Frenkel Phys. Rev. Lett. 120, 225502 (2018)
 J. Timoshenko, D. Lu, Y. Lin, A. I. Frenkel J. Phys. Chem. Lett., 8, 5091-5098 (2017)