Could simplified drag models accumulate large errors over very long game-world timescales?

Your question asks whether simplified drag models—like those used in video games or fast simulations—can accumulate significant errors over long timescales. This is a great question about numerical error growth, but the preprints provided don't directly address drag modeling or its error accumulation.

However, I can point you toward the closest relevant result: [AG-2025.12-078] demonstrates a general principle that applies here. When celestial dynamics simulations use approximations like particle-mesh (PM) methods or modified gravity models, "although the errors and shortcomings of these PM approximations... are small, they cause the instability of the regular dynamics" over large-scale systems. The key finding is that small violations of conservation laws—momentum and angular momentum not being exactly preserved—destabilize trajectories.

A simplified drag model would have a similar problem: if it doesn't conserve energy correctly (drag dissipates energy, but an oversimplified model might do so non-uniformly or incorrectly), those small per-timestep errors compound exponentially over game-world timescales. The longer your simulation runs, the more those errors accumulate.

For a concrete answer specific to drag models—how fast errors grow, what simplified forms fail worst, or techniques to keep them bounded—you'd want papers on numerical integration of dissipative systems or game-engine physics stability, which aren't in this archive.