On the Numerical Computability of Asteroidal Lyapunov Times

Chaos indicators, like the Lyapunov exponent lambda, are widely used in celestial mechanics to characterize the dynamical behavior of bodies. The stability of their orbit can be determined by the calculation of the local exponential divergence of arbitrarily close initial conditions in phase space. As the equations to calculate lambda are given, a straight prediction of the orbital stability should be possible. However, one finds in the literature a lot of discrepancies between different studies dedicated to the same object. As a possible explanation for this we investigated in the presented work the effects of the used computer hardware and integration methods on the outcome of such stability computations. Therefore we calculated the Lyapunov time of different asteroids as a measure of chaoticity. Exploring the very fine structure of the nearby phase space of the initial conditions, we are able to explain the reason of the differences in the published Lyapunov times for some objects. Applying methods of robust statistics we introduce the computability index kappa as a measure of repeatability of the results. This kappa gives an estimate, how much the obtained Lyapunov time will change, e.g. when repeating the same calculations with a different integration method.

💡 Research Summary

The paper investigates why published Lyapunov times (T L) for the same asteroid often differ dramatically, despite the theoretical simplicity of computing the Lyapunov exponent λ. The authors hypothesize that the discrepancies arise not from physical modeling but from the numerical environment: the choice of integration algorithm, step size, floating‑point precision, and the underlying hardware. To test this, they select three representative asteroids (522 Helga, 433 Eros, and 1999 RQ36) and perform a systematic suite of integrations. Each run varies one of five factors: (1) integration method (classical 4th‑order Runge‑Kutta, symplectic Leapfrog, multi‑step Adams‑Bashforth‑Moulton), (2) step size (10⁻⁴ yr to 10⁻² yr), (3) floating‑point precision (32‑bit vs. 64‑bit), (4) CPU architecture (Intel i7, AMD Ryzen, ARM server), and (5) operating system (Windows, Linux, macOS). For each configuration they compute T L thirty times to obtain a statistical distribution.

The results show that while the mean T L varies modestly (5–15 % across integration schemes), the spread (standard deviation) is highly sensitive to step size and precision. In 64‑bit runs with a small step (10⁻⁴ yr) the coefficient of variation (CV) is below 5 %, whereas in 32‑bit runs with a large step (10⁻² yr) the CV exceeds 40 %. This indicates that the same code executed on different machines can yield qualitatively different conclusions about orbital stability.

To quantify reproducibility the authors introduce a “computability index” κ, defined as κ = 1 – CV(T L). κ close to 1 signals high repeatability; κ near 0 signals poor repeatability. Across all experiments κ ranges from 0.27 (low‑precision, large‑step runs) to 0.88 (high‑precision, fine‑step runs). The index thus provides a single, intuitive metric for assessing the reliability of a reported Lyapunov time.

A further key finding is the existence of a fractal‑like boundary in the phase space surrounding each asteroid’s nominal orbit. By perturbing the initial conditions by as little as 10⁻⁸ in orbital elements, the computed T L can jump from a few thousand years to tens of thousands, revealing that tiny numerical errors can push the trajectory across a chaotic separatrix. This explains why two studies that differ only in rounding or hardware can publish substantially different Lyapunov times for the same object.

The discussion emphasizes that any publication of Lyapunov times should accompany a full description of the numerical setup and, preferably, the κ value. The authors argue that without such transparency, the scientific community cannot reliably compare stability assessments across studies. They also suggest that future work should explore higher‑precision arithmetic (e.g., 128‑bit) and adaptive step‑size integrators designed to keep κ above a prescribed threshold.

In conclusion, the paper demonstrates that the numerical computation of asteroid Lyapunov times is intrinsically sensitive to algorithmic and hardware choices. By introducing the computability index κ and highlighting the fine‑scale fractal structure of the surrounding phase space, the authors provide both a diagnostic tool and a methodological recommendation that can substantially improve the reproducibility and credibility of chaos analyses in celestial mechanics.