Force-resolved and recurrence-based identification of dynamical heterogeneity in liquid water

Understanding how microscopic fluctuations in liquid water govern macroscopic dynamics remains a central challenge. Here, we introduce a force-resolved, recurrence-based framework using a charge-aware neuroevolution potential (q-NEP) to identify dynamically active molecular environments. By combining force-derived instability metrics with recurrence statistics, we distinguish transient extreme fluctuations from persistently active molecules that drive hydrogen-bond rearrangement. We show that only a small fraction (∼5-10%) of molecules repeatedly participate in high-instability events, forming spatially localized hotspots of activity that are not uniquely captured by conventional structural descriptors. This reveals a clear separation between extreme and persistent dynamics, demonstrating that recurrence-weighted force fluctuations provide a more sensitive and physically grounded measure of molecular activity. This framework enables direct identification of molecular environments governing transport and reactivity in aqueous systems, offering a generalizable pathway to link microscopic dynamics with macroscopic behavior.