Towards Extended Active Learning for Modelling Ferroelectric Switching: the Need for 'Gold Standards'New

This paper addresses the need for reliable reference data ('gold standards') in active learning workflows for ferroelectric switching simulations. The authors evaluate four computationally efficient DFT functionals (PBE, PBEsol, r2SCAN, and r2SCAN-rVV10) against high-level ab initio methods including the random-phase approximation (RPA) and second-order Møller-Plesset perturbation theory (MP2). The study focuses on wurtzite-structured ferroelectric materials, providing guidance for selecting appropriate DFT methods when building surrogate models for active learning campaigns in materials discovery.