Hi @Paul,
I wrote you up a little code comparing state transition matrix computation with Orekit partial derivatives class on one side and derivative structures on the other side.
TestSTMWithDS.java (10.9 KB)
I’ll let you try it and see if you can use it in your application.
Hope it will help.
@evan.ward I’ve been thinking for a while about this bug in the PDE providing only Cartesian derivatives. And I don’t think that this is the source of the problem.
If you look into the results of the code I sent to Paul you’ll see that the PDE and JacobianMapper classes do give the state transition matrix with the proper orbit type but that the computation is quite bad. There is a lot of numerical “noise”, especially for the derivatives of the semi major axis.
My guess is that it comes from numerical errors in getJacobianWrtParameters which uses a matrix inversion on an un-normalized matrix (see method Orbit.createInverseJacobian)
But I’m not 100% sure and it would require more testing.
What do you think ?
Regards,
Maxime