Accelerating propagation of many orbits

I am trying to compute ephemerides from orbits with poliastro. Bluntly put, I need Cartesian coordinates in fixed time intervals for longer periods of times. poliastro can generate Orbit objects from ephemerides objects, but - as far as I can see - there is no direct method for going the opposite direction (in some optimized form).

What I came up with simply looks as follows … I am not entirely sure if this the “right” way to do this:

eph = np.zeros((days, 3))
for day in range(0, days):
    time_current = time_zero + day * u.day
    eph[day, :] = orb.propagate(time_current).r.to(u.AU)

The approach is relatively slow … probably due to lots of overhead from (numba) function calls (?).

[astrojuanlu]

We are pleased to inform you that poliastro’s proposal, /Array types for scaling poliastro/ was approved for the amount of $5,000.00. The committee offered this feedback about the proposal:

Well written proposal with clear goals and benefits. Likely that this project would not be accomplished without this funding. Integrating with other NumFOCUS projects is a plus.

🎉 Full text of the proposal: https://github.com/poliastro/documents/blob/master/numfocus-sdg-2021-r3.md

[s-m-e]

Ok, this was a numba issue if I understand this correctly. I uploaded a self contained notebook here. I “fixed” it by separating the contents of cells 6 and 12. If this stuff is on one single cell, it probably re-compiles my farnocchia_coe_fast function probably on every run.

It’s actually faster, from 530 ms down to 390 ms, so roughly 25% less computing time.