Improved Initial Guesses for Numerical Solutions of Kepler's Equation

Abstract

Numerical solutions of Kepler's Equation are critical components of celestial mechanics software, and are often computation hot spots. This work uses symbolic regression and a genetic learning algorithm to find new initial guesses for iterative Kepler solvers for both elliptical and hyperbolic orbits. The new initial guesses are simple to implement, and result in modest speed improvements for elliptical orbits, and major speed improvements for hyperbolic orbits.