Analysis of the ODE Integration Methods

June 9, 1996

52

Our prediction is that, as extrapolation methods like

Bulirsch-Stoer continue to gain sophistication, they will

eventually beat out PC methods in all applications. We are

willing, however, to be corrected.(14:592)

Let's look at the properties of our particular problem. First, the evaluation of the

right-hand side, i.e.

*f()*

, is

*very*

expensive, even if a fast force evaluation method is used.

The star system movement is normally very smooth, but it can also have singularities or

poles in the complex plane when stars pass close to each other. The ODE integration rou-

tines need to be accurate, but since the results are just used to draw pictures, very high

accuracy is not really required. All that is required is that it has to be accurate enough to

avoid the worst of the movement errors.

Another atypical property of our star movement problem is that the smaller the

step size is, the less a `near collision' looks like a singularity and the more it looks like a

smooth path. This is because the stars tend to move around each other so the point of the

singularity (i.e., the other star) tends to move out of the way. Thus, by using a simpler

method such as the Adam-Bashford method which requires a very smooth function, you

can make the step size smaller and get a smooth function. If a more expensive method,

such as the Gragg-Bulirsch-Stoer's is used, a much large step size must be used and thus

you get the singularities that it can't handle.

This document is best viewed as n-body.pdf because the translation to html was buggy.