Abstract
Simulating trajectories of multi-particle systems on complex energylandscapes is a central task in molecular dynamics (MD) and drug discovery, butremains challenging at scale due to computationally expensive and longsimulations. Previous approaches leverage techniques such as flow orSchr\"odinger bridge matching to implicitly learn joint trajectories throughdata snapshots. However, many systems, including biomolecular systems andheterogeneous cell populations, undergo dynamic interactions that evolve overtheir trajectory and cannot be captured through static snapshots. To close thisgap, we introduce Entangled Schr\"odinger Bridge Matching (EntangledSBM), aframework that learns the first- and second-order stochastic dynamics ofinteracting, multi-particle systems where the direction and magnitude of eachparticle's path depend dynamically on the paths of the other particles. Wedefine the Entangled Schr\"odinger Bridge (EntangledSB) problem as solving acoupled system of bias forces that entangle particle velocities. We show thatour framework accurately simulates heterogeneous cell populations underperturbations and rare transitions in high-dimensional biomolecular systems.