Multivariate statistical analysis of in situ and operando X-ray Absorption Spectroscopy data

X-ray Absorption Spectroscopy (XAS) is an element specific analysis which focuses on the local order and electronic structure of the absorbing atom. The high penetrating power of X-rays allows this method to be carried out in situ and operando. Provided that measurements are performed with adequate time-resolution, the underlying reaction mechanisms and the nature of the chemical intermediates involved can be identified. The large amount of data collected during a typical operando X-ray absorption fine structure experiment and the interest to thoroughly investigate and comprehend the processes occurring, need an advanced data analysis approach. Established methods, such as the linear combination of known standard materials, do often not accurately represent the measured data at relevant temperatures, pressure and reactants present. This mismatch is explained by a different chemical nature and often stable pure states of a compound do not exist to be measured ex situ. In this context, multivariate statistical analysis has gained attention, which allows the identification of the number and their abundance of the chemical species involved, with limited a priori information on the studied system. [1] After a brief historical introduction and the basic insights on the technique of multivariate statistical analysis, I will provide a selection of XAFS examples and case studies, to discuss and demonstrate approaches to determine the number of components and their abundance in the dataset by matrix factorization. Given this information, initial guesses by blind-source separation through Evolving Factor Analysis [2], independent component analysis [3] or purest variables [4] will be discussed. Rank deficiency and strategies to try to resolve beyond will be presented. Furthermore, a closer look at the regression and the applied constraints will be elaborated, for example for samples where the absorption step is not constant. A comparison to compare information obtained by commonly applied methods such as white-line intensity, edge-position, pre-edge features and linear combination will be showcased. References [1] - Multivariate Curve Resolution: https://mcrals.wordpress.com/theory/mcr-als/ [2] - Keller HR, Massart DL. Chemom Intell Lab Syst. 1991,12:209-224. [3] - Common P. Signal Processing. 1994,36:287-314. [4] - Windig W. Comprehensive Chemometrics. 2009:275-307.