Autonomous atomic Hamiltonian construction and active sampling of X-ray absorption spectroscopy by adversarial Bayesian optimization

X-ray absorption spectroscopy (XAS) is a well-established method for in-depth characterization of electronic structure. In practice hundreds of energy-points should be sampled during the measurements, and most of them are redundant. Additionally, it is also tedious to estimate reasonable parameters in the atomic Hamiltonians for mechanistic understanding. We implement an Adversarial Bayesian optimization (ABO) algorithm comprising two coupled BOs to automatically fit the many-body model Hamiltonians and to sample effectively based on active learning (AL). Taking NiO as an example, we find that less than 30 sampling points are sufficient to recover the complete XAS with the corresponding crystal field and charge transfer models, which can be selected based on intuitive hypothesis learning. Further applications on the experimental XAS spectra reveal that less than 80 sampling points give reasonable XAS and reliable atomic model parameters. Our ABO algorithm has a great potential for future applications on automated physics-driven XAS analysis and AL sampling.

Automated summary

X-ray absorption spectroscopy (XAS) is a well-established method for characterizing electronic structure. We propose an Adversarial Bayesian optimization (ABO) algorithm that combines active learning to efficiently fit many-body model Hamiltonians and sample points. Using NiO as an example, we demonstrate that less than 30 sampling points can recover the complete XAS and corresponding models. Experimental XAS spectra analysis also shows that less than 80 sampling points provide reasonable XAS and reliable atomic model parameters. Our ABO algorithm has great potential for automated physics-driven XAS analysis and active learning sampling in the future.