CONVERGENCE ANALYSIS OF LEAPFROG FOR GEODESICS

Geodesics are of fundamental interest in mathematics, physics, computer science, and many other subjects. The so-called leapfrog algorithm was proposed in [L. Noakes, J. Aust. Math. Soc., 65 (1998), pp. 37-50] (but not named there as such) to find geodesics joining two given points x0 and x1 on a path-connected complete Riemannian manifold. The basic idea is to choose some junctions between x0 and x1 that can be joined by geodesics locally and then adjust these junctions. It was proved that the sequence of piecewise geodesics {γ^k}_k≥₁ generated by this algorithm converges to a geodesic joining x0 and x1. The present paper investigates leapfrog's convergence rate τi,n of ith junction depending on the manifold M. A relationship is found with the maximal root λn of a polynomial of degree n-3, where n (n > 3) is the number of geodesic segments. That is, the minimal τi,n is upper bounded by λn(1 + c+), where c+ is a sufficiently small positive constant depending on the curvature of the manifold M. Moreover, we show that λn increases as n increases. These results are illustrated by implementing leapfrog on two Riemannian manifolds: the unit 2-sphere and the manifold of all 2 × 2 symmetric positive definite matrices.