Our two papers on understanding experimental X-ray spectra of materials utilizing simulated references just got accepted (and appeared as “just accepted” manuscripts) in Chemistry of Materials as Part I (more qualitative) & Part II (more quantitative).
These papers are a collaboration of our group (Anja Aarva, who did most of the work, Tomi Laurila and myself) with Volker Deringer (now at Oxford) and Sami Sainio (Stanford), supported by the Academy of Finland (funding), the CSC supercomputing center (CPU time) and the HPCEuropa3 program (traveling support). I will try to explain here what we accomplished in this work.
The problem: X-ray spectroscopy (e.g., XPS and XAS) is a popular experimental tool to probe the microscopic structure of materials. The link is typically made by fitting X-ray spectra to a weighted combination of reference molecular spectra. Unfortunately, these molecular references are not able to account for all the features of the full X-ray spectra of the materials, and they sometimes overlap among themselves, leading to inconclusive or even contradictory results.
Our solution: With simulation (DFT and machine learning), we can make a one to one correspondence between the structure of a local atomic motif (or environment) and its corresponding X-ray signature. We use these “fingerprint” spectra as building blocks to fit experimental spectra and gain insight into the atomic structure of materials. These papers are a first attempt at full quantitative fitting of experimental spectra. In the future, we hope to improve the accuracy of the method and extend its range of applicability to more materials.