Title:Symbolic regression for the work function of metal/organic interfaces: when are descriptors physically meaningful?

Abstract:We apply symbolic regression in the form of the Sure Independence Screening and Sparsifying Operator (SISSO) to hybrid inorganic/organic interfaces, with the intent to extract physically meaningful correlations between the adsorption-induced work-function modifications and the properties of the constituents. We separately investigate two cases: Hypothetical, free standing self-assembled monolayers with a large intrinsic dipole moment, and metal-organic interfaces with a large charge-transfer induced dipole. For the former, we find - without notable prior assumptions - the Topping model, as expected from literature. For the latter, where presently no sensible correlations are known, we find expressions with high predictive accuracy. These are, however, partly unphysical. We show that this is because even first-principles data are noisy, and due to intrinsic limitations of the SISSO approach.