Development of a universal machine learning descriptor from density functional theory data of ion adsorption and diffusion properties on Sulfer-functionalized MXenes for battery applicatations

Two-dimensional materials composed of transition metal carbides and nitrides (MXenes) are hopeful candidates for energy storage devices, such as ion batteries and supercapacitors, on account of their high capacity and high-power capabilities.  Additionally, adding Sulfur terminations decreases the diffusion energy barrier for some ions, increasing the devices’ efficiencies.  In this work, we developed a database of adsorption energy and diffusion barriers for various adatoms (Li, Ca, Mg, Na, Al, and Zn) on nine M₂CS₂ monolayers (M: Cr, Hf, Mo, Nb, Ta, Ti, B, W, Zr).  We found that Ca binds the most strongly to all of the MXenes, and that Zn binds the weakest in dilute concentrations.  Our database also consists of coverage dependent formation energies and open-circuit voltages (OCVs) derived through cluster expansion.  Finally, we go one step further and develop a machine learning descriptor from our data set to predict binding energy, adsorption geometry, charge transfer. This universal descriptor is trained to work successfully on multilayer M2CS2s and their heterostructures as well.