Active Learning of Atomic Size Gas/Solid Potential Energy Surfaces via Physics Aware Models.

We propose an active learning (AL) framework to develop classical force fields (FFs) that accurately model the potential energy surfaces (PES) of gas/solid atomic-scale complexes. A central challenge is integrating AL with flexible, computationally efficient physics-aware potentials to achieve quantum-level accuracy for complex interfacial systems. Our approach trains physics-aware potentials, with incorporated flexibility and smoothness, on actively sampled density functional theory (DFT) data to describe interactions between undercoordinated atomic silver (Ag) clusters and gaseous pollutants (CO, CO, SO), relevant for environmental applications like sensing. The AL process follows three stages: (1) FFs are trained using adaptable physics aware potentials of semiempirical descriptors, optimized via a Pareto analysis scheme; (2) new candidate structures are generated through the use of the refined FFs in Metropolis Hastings Monte Carlo (MHMC) or stochastic molecular dynamics (sMD) simulations; (3) a subset of candidates is selected for DFT computations based on an outlier score (OS), which utilizes the existing data descriptor distributions, ensuring diverse PES exploration. This framework produces FFs capable of capturing cohesive, physisorption, and chemisorption interactions with admirable accuracy, close to ab initio methods, while retaining the efficiency of semiempirical potentials. To demonstrate, produced FFs are utilized in molecular dynamics (MD) simulations of single Ag clusters embedded in bulk gas phases, examining condensation characteristics. Our methodology is highly versatile, easily accommodating various choices of descriptors, model basis sets, and sampling techniques.