Fine-tuning ab initio XANES spectra calculations using the Bayesian optimization algorithm

Theoretical modeling of X-ray absorption near-edge structure (XANES) spectra
is often performed to interpret experimental spectra and to relate spectral
features to the atomic and electronic structure of materials. The sensitivity of the
XANES spectra to coordination environment, oxidation state and structural
disorder is explored using FEFF and FDMNES simulation codes. With a suitable
configuration of these instruments, one can also explore the quantitative char-
acteristics of the studied materials, such as lattice constant, nanoparticle
geometry, coordination numbers and longer-range ordering. However, fine-
tuning the input parameters is often resource-consuming and non-intuitive. We
test the application of the Bayesian optimization (BO) algorithm in finding the
most optimal simulation parameters for the theoretical XANES spectra and
compare the results using different spectrum similarity metrics. The BO method
outperforms the random search technique by a factor of three in speed and
shows that the correlation-based metrics provide better shape-level agreement
than those defined as normalized distance.