AtomWorld-Mirror introduces a time‑aware macro‑step world model that captures the critical evolution backbone of atomic systems. Conventional atomistic simulators advance at microscopic resolution, spending most of the computation on low‑impact local updates before reaching structurally significant states, which creates a bottleneck for long‑horizon simulations. The proposed model distills short micro‑event segments into physically reachable transitions between key states, jointly predicting sparse structural edits and accumulated physical time via latent macro‑step dynamics. Each transition is constrained by local reachability, inventory conservation, and continuous‑time consistency. By amortizing local atomic physics into a reusable latent macro model and replacing explicit micro‑event replay with macro‑step inference, the approach offers a path to substantially faster prediction of long‑term materials evolution while preserving structural validity and temporal semantics. Benchmarks on five atomic systems—including Cu‑rich RPV steel irradiation aging, Cu‑Zr metallic glass, and Li$_3$N‑based anti‑perovskite solid electrolyte—show macro‑step inference achieving speed‑ups of $10^3$ to $10^4$ times over event‑by‑event simulation.
Review