I recently stumbled on a problem with propagator converter.
The AbstractPropagatorConverter.adapt method uses a two step fitting, with an initial run to estimate first guess and a second run to refine this guess. I already fixed a problem with the first run (see issue 1976). Then it occurred to me that after the second run returns, there is a call to getRMS(result) where result contains the estimated parameters (orbit and propagation parameters). This call induces a last propagation as getObjectiveFunction().value(parameterSet) is called and then residuals are evaluated.
This seemed a waste of time to me as the optimum field already contains all required data and even has its own getRMS() method. Using optimum.getRMS() however gives exactly the same value when position-only fiiting is used but different results if position and velocity are used. Digging into the problem, it occurred to me that optimum takes into account the weight array that is set up when setSample is call prior to the fitting whereas the method getRMS(result) that is private to AbstractPropagatorConverter just uses unweighted residuals.
I am wondering if we could get rid of the private AbstractPropagatorConverter.getRMS(result) method and the private rms field and just have the public getRMS() method (without any arguments) return optimum.getRMS()? This would mean getRMS() would take weights into account, which seems good to me as the optimizer did use them. This would be more consistent and would avoid one call to propagation, which is a significant gain when optimization itself converges in just a few propagation runs.