Lawrence Livermore runs one-million‑orbit simulation to chart collision risks in cislunar space

Researchers at Lawrence Livermore National Laboratory have produced a large‑scale simulation that models one million orbital trajectories spanning six years to better understand how objects move in the region between Earth and the Moon. The work leverages publicly available orbital data combined with substantial supercomputing time to generate a dataset intended for machine‑learning analysis and anomaly detection. Results show that roughly half of the simulated paths remain predictable for at least a year, but fewer than one in ten retain stability across the entire six‑year window, underscoring the dynamic and chaotic nature of many cislunar paths. Achieving this required about 1.6 million CPU hours of computation; on LLNL’s dedicated systems the workload completed in days rather than the centuries it would take on a single processor. The scale of the run lets analysts explore subtle behaviors and identify likely “busy” regions where trajectories cluster and collision probability rises. By producing many near‑realistic scenarios, the team can train algorithms to flag trajectories that diverge from expected behavior and to estimate orbital lifetimes under varying perturbations. The dataset does not remove uncertainty from short‑term tracking—predicting precise positions still demands stepwise propagation and frequent updates—but it provides a statistical backbone for prioritizing tracking resources and designing safer mission profiles. The approach is adaptable: agencies and operators could use similar large‑ensemble modeling to assess the risk of planned launches, or to decide where maneuvering capability is most valuable. As more nations and commercial groups place assets into low‑Earth and cislunar orbits without centralized traffic coordination, datasets of this size will become practical tools for multinational risk assessment and coordination. The simulation’s public methods also increase transparency and reproducibility, enabling independent teams to test alternative models and mitigation strategies. While not a substitute for active surveillance networks, this work offers a way to move from isolated position fixes to probabilistic maps of congestion and instability across months and years. Taken together, the results point to a future in which large‑ensemble orbital modeling complements radar and optical tracking to reduce collision risk across crowded orbital regimes.