Hi there,
From a quick look at your code, I’ve got two comments:
- I recommend using equinoctial coordinates for propagation rather than Cartesian, they’re slow variables so better behaved, especially with adaptive schemes
- about the integration tolerances actually, you pass really tight tolerances. Have you checked the size of the steps effectively taken? I’m afraid they could be very small (<1s) and increase numerical noise
Edit: the difference could also simply come from slightly different atmospheric density predictions. Could you checkwithout drag?
Cheers,
Romain.