Cross-state prediction of dynamic ferroelectric catalysis: Efficient screening for bypassing the Sabatier limit

Abstract

Ferroelectric (FE) switching offers a pathway to bypass the Sabatier limit by toggling binding states, yet the computational cost for screening such dynamic materials remains prohibitive. Here, we present a machine learning-accelerated approach to discover two-dimensional FE metal–organic frameworks for switchable hydrogen evolution. We establish a geometric-electronic descriptor that enables quantitative cross-state prediction of adsorption energies, effectively halving the computational cost by forecasting the switched state solely from the initial-state features. Using this model, we identify candidate materials where polarization reversal drives the hydrogen adsorption free energy (ΔGH) from negative to positive, facilitating a direct transition from strong adsorption to spontaneous desorption. Microscopically, polarization-driven coordination field reconfiguration serves as a geometric switch to tune the energy level and occupation of the metal dz2 orbitals, while the magnitude of spin-polarization splitting ultimately governs the thermodynamic stability. This work provides a systematic framework for discovering intrinsic switchable catalysts via efficient cross-state prediction.