Nominals for initial derivatives
From what I can tell, the initial derivatives are not scaled before they are put in the state vector. Is there a reason for this? I was thinking of having a nominal like
nominal_init_derivative = 0.01 * nominal_state / dt)
This is assuming that the time step is typically chosen such that changes in states are significant relative to that state (but not too much of course). Above, I chose the 0.01 accordingly, assuming that most changes from time step to time step are in the order of 1E-3 to 0.1 times nominal_state
. As for dt, we might just take t[1] - t[0] as a good enough guess, and a proper t[0] - t[start-1] if history data is available.