[Help-gsl] rk4

[Brian Powell]

Hello

This email is in reference to what appears to be a notable change in
performance of the rk4 integrator between older and newer versions of gsl.
 In particular, rk4 from version 1.5 and version 1.8.  The particular test
that I am performing is I am integrating a system of ODE's from some
initial point to some final point, then reversing the integration. 
Clearly, one expects such evolutions to overlap.  Surprisingly, they
overlap quite well with the older version of rk4, but they do not match up
well at all with the newer version.  I'm trying to track down the cause of
this, and, so far, can't seem to find a system specific reason for this.

Thank you in advance,
Brian