We now compensate the possible slight date offset between state and
transit time.

derivatives.
It is indeed much faster than automatic differentiation for this force.
Additionnaly, only the derivatives of density with respect to position
are computed, as density does not depend on velocity or drag
coefficient.

The additional mass was not really useful, it was intended for maneuvers
calibration, but in fact during maneuver calibration we adjust either
flow rate or specific impulse but not directly mass itself. 

The tests were intentional, as they check equals method behaved
correctly when passed wrong type objects.

As some force models depend only on position (gravity, third body,
parametric acceleration when direction is inertial), we can reduce
the number of derivatives computed for these force models.
The other force models that depend on velocity (either directly or due
to a dependency on attitude which itself may depend on velocity,
for example for LOF offset attitudes) are computed with all derivatives
taken into account.

We now compute derivatives with respect to state and derivatives with
respect to estimated parameters simultaneously.

rnveach authored 7 years ago and Luc Maisonobe committed 7 years ago

Github: closes #13

ionospheric/tropospheric modifiers.
The azimuth computation is now consistent with the one in AngularAzEl
class.

Thanks to roberto Alacevich for finding the issue.

The ForceModel API is now much simpler, with only two principal methods
for computing acceleration. One method deals with regular double
elements, the other method deals with field elements.

This is a first step towards removing the last accelerationDerivatives
method.