Understanding X-ray Spectroscopy of Carbonaceous Materials by Combining Experiments, Density Functional Theory, and Machine Learning. Part II: Quantitative Fitting of Spectra

Carbon-based nanomaterials are a promising platform for diverse technologies, but their rational design requires a more detailed chemical control over their structure and properties than is currently available. A long-standing challenge for the field has been in the interpretation and use of experimental X-ray spectra, especially for the amorphous and disordered forms of carbon. Here, we outline a unified approach to simultaneously and quantitatively analyze experimental X-ray absorption spectroscopy (XAS) and X-ray photoelectron spectroscopy (XPS) spectra of carbonaceous materials. We employ unsupervised machine learning to identify the most representative chemical environments and deconvolute experimental data according to these spectral contributions. To fit experimental spectra we rely on ab initio references and use all the information available: to fit experimental XAS spectra, the whole XAS fingerprint (reference) spectra of certain sites are taken into account, rather than just peak positions, as is currently the standard procedure. We argue that, even for predominantly pure-carbon materials, carbon K-edge and oxygen K-edge spectra should not be interpreted separately, since the presence of even small amounts of functional groups at the surface manifests itself on the X-ray spectroscopic signatures of both elements in an interlinked manner. Finally, we introduce the idea of carrying out simultaneous fits of XAS and XPS spectra, to reduce the number of degrees of freedom and arbitrariness of the fits. This work opens up a new direction, tightly integrating experiment and simulation, for understanding and ultimately controlling the functionalization of carbon nanomaterials at the atomic level.