Actually this calls for another clarification. With drag, Orekit has two ways to compute the derivatives of the atmospheric density w.r.t. the Cartesian vector (the default one being the first):
- Finite differences
- AD
This is because the Field version of some models e.g. NRLMSISE00 can be computationnally intensive. In fact it can easily become the bottleneck of propagation.
Last year I proposed a potential improvement but it received mixed opinions so I’ve not implemented it.
Cheers,
Romain.