X-ray absorption spectroscopy (XAS) is a technique used to probe unoccupied electronic states by exciting core electrons. The absorption energy of a photon by a transition metal varies depending on the binding energy of the excited electron. The resulting spectral signal corresponds to sample-specific electronic and chemical characteristics, enabling the technique to be used to help identify specific elements, oxidation states (OS), and local coordination (LC). Through computational methods, software packages have been developed that allow for the high-speed generation of XAS spectra based on parameters, including but not limited to OS and LC. However, using spectra to reverse identify the operational parameters of real-world samples is significantly more computationally intensive. Our project aims to improve the speed and computational efficiency of predicting OS and LC from an Fe L<sub>3</sub> XAS spectrum using Machine Learning (ML). A Python script was programmed to build our database using Crispy (an XAS simulation tool) to generate thousands of spectra. These spectra were used to train scikit-learn’s Random Forest algorithm. This algorithm was chosen because of its ability to accurately classify spectra based on OS and LC within the dataset and its simple implementation with python. The train data were artificially convoluted by shifting the data to higher and lower energies to reduce reliance on exact peak locations and simulate real-world variations in experimental data collection and detector calibration. This method achieved just over 70% accuracy when predicting OS and LC within the simulated data. In the future, this method can be adapted to classify the OS and LC of large experimental datasets with greater speed and much less computational power.

Determination of oxidation state and symmetry of Fe from X-ray absorption spectroscopy spectra using machine learning

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.