Snapshot 1: the Runge–Kutta method gives an orbit that spirals out to infinity
Snapshot 2: The averaged leapfrog method gives an orbit that spirals towards the origin. It deviates from the exact orbit (shown in green) considerably slower than in the Runge–Kutta case.
Snapshot 3: the non-averaged leapfrog method gives orbits that always stay close to the exact one
Snapshot 4 and 5: The tiny radial error varies in a characteristically different way with time for the ALF and DALF methods. It scales with
so that halving the time step divides the error by 16.
Snapshots 5 and 6: ADALF and Runge–Kutta show a radial error that scales with
. For Runge–Kutta, the error is about a factor of four larger than for ADALF. Notice that the sign of the error is different in the two cases.
Snapshot 7: The phase error scales in all cases with
and always grows linearly with time. So even for very short time steps, the computed position and the exact position eventually run out of phase.
The integrators under consideration are formulated for an ODE of the general form
, where the number of vector components of
only appears in the initial values for
; the linear algebra used to formulate the integration step is dimension generic in the Wolfram Language. In the simple case actually treated,
is independent of
and is of the form
. The asynchronous leapfrog integrators under consideration are described in .