Accurate computational prediction of core-electron binding energies in carbon-based materials: A machine-learning model combining density-functional theory and $\boldsymbol{GW}$
We present a quantitatively accurate machine-learning (ML) model for the computational prediction of core-electron binding energies, from which x-ray photoelectron spectroscopy (XPS) spectra can be readily obtained. Our model combines density functional theory (DFT) with $GW$ and uses kernel ridge regression for the ML predictions. We apply the new approach to materials and molecules containing carbon, hydrogen and oxygen, and obtain qualitative and quantitative agreement with experiment, resolving spectral features within 0.1 eV of reference experimental spectra. The method only requires the user to provide a structural model for the material under study to obtain an XPS prediction within seconds. Our new tool is freely available online through the XPS Prediction Server.