Improved detection of chaos with Lagrangian descriptors using differential algebra

Abstract

Lagrangian descriptors (LDs) based on the arc length of orbits previously demonstrated their utility in delineating structures governing the dynamics. Recently, a chaos indicator based on the second derivatives of the LDs, referred to as $\vert\vert \Delta \rm{LD} \vert\vert$, has been introduced to distinguish regular and chaotic trajectories. Thus far, the derivatives are numerically approximated using finite differences on fine meshes of initial conditions. In this paper, we instead use the differential algebra (DA) framework as a form of automatic differentiation to introduce and compute $\vert\vert \Delta \rm{LD} \vert\vert$ up to machine precision. We discuss and exemplify benefits of this framework, such as the determination of reliable thresholds to distinguish ordered from chaotic trajectories. Our extensive parametric study quantitatively assesses the accuracy and sensitivity of both the finite differences and differential arithmetic approaches by focusing on paradigmatic discrete models of Hamiltonian chaos, namely the Chirikov's standard map and coupled $4$-dimensional variants. Our results demonstrate that finite difference techniques for $\vert\vert \Delta \rm{LD} \vert\vert$ might lead to significant misclassification rate, up to $20\%$ when the phase space supports thin resonant webs, due to the difficulty to determine appropriate thresholds. On the contrary, $\vert\vert \Delta \rm{LD} \vert\vert$ computed through DA arithmetic leads to clear bimodal distributions which in turn lead to robust thresholds. As a consequence, the DA framework reveals as sensitive as established first order tangent map based indicators, independently of the underlying dynamical regime. Finally, the benefits of the DA framework are also highlighted for non-uniform depleted meshes of initial conditions.