Active delta-learning for surface hopping: efficient fitting of potential energy surfaces

We combine active learning with Δ-learning into an end-to-end protocol for active delta-learning of potential energy surfaces for trajectory surface hopping dynamics.