search

matthewholman / assist

ASSIST is a software package for ephemeris-quality integrations of test particles.
https://assist.readthedocs.io/
GNU General Public License v3.0
24 stars 10 forks source link

Roundtrip unit test #43

Closed hannorein closed 1 year ago

hannorein commented 1 year ago

While coding up the roundtrip test from the paper as a unit test, I've noticed something that doesn't seem quite right. The error seems to depend on the range in a very non-smooth way (have a look at the output below). The plot in the notebook and paper doesn't have that many data points, so I'm not sure if this is a new problem. In any case, I don't think this is correct as of right now. I suspect this is related to how the reb_integrate function handles the last timestep, which depending on where the timesteps fall, might have to be much smaller than a "normal" timestep. (but it could also be something completely different)

trange = 41161.355954   error: 0.139017m
trange = 43219.423752   error: 2897.963781m
trange = 45380.394939   error: 0.061683m
trange = 47649.414686   error: 0.146603m
trange = 50031.885420   error: 0.314383m
trange = 52533.479691   error: 0.077800m
trange = 55160.153676   error: 0.495380m
trange = 57918.161360   error: 0.321240m
trange = 60814.069428   error: 4068.455833m
trange = 63854.772899   error: 0.308811m
trange = 67047.511544   error: 7.797359m
trange = 70399.887121   error: 0.589816m
hannorein commented 1 year ago

@aryaakmal I agree that figure 6 would be the right one to look at, but there are no jumps...?!

aryaakmal commented 1 year ago

@hannorein - certainly no discontinuities. But I imagine @dhernan is referring to the large change in magnitude in the right panel?

On Tue, Feb 21, 2023 at 12:51 PM Hanno Rein @.***> wrote:

@aryaakmal https://github.com/aryaakmal I agree that figure 6 would be the right one to look at, but there are no jumps...?!

— Reply to this email directly, view it on GitHub https://github.com/matthewholman/assist/pull/43#issuecomment-1438880154, or unsubscribe https://github.com/notifications/unsubscribe-auth/AOMJBQ7UKZ3MIXC5LG6J433WYT6AJANCNFSM6AAAAAAUMOYS6Q . You are receiving this because you were mentioned.Message ID: @.***>

aryaakmal commented 1 year ago

I am just trying to understand the concern. I should let @dmhernan speak for himself.

dmhernan commented 1 year ago

Hi all, I was referring to the scatter plot in Fig. 2, the penaltimate point in time, which has the largest error of all points. It may be due to a time step jump, as we've seen happens in several plots in this thread. That wouldn't be surprising, and it would clarify that figure significantly. Whether it is or not is not important--- the point is that these error jumps have caused a headache to me and possibly all of us, and it seems they are potential headaches for other researchers.

hannorein commented 1 year ago

I agree that Figure 2 caused us some headaches. But it's very contrived and not an interesting test from a physical point of view. The scale is arbitrary. There is no a priori reason to think that we need a precision of 100m over 1e5 days. Nothing will break down if we only have 1km or 10km precision.

dmhernan commented 1 year ago

OK, that logic seems good. Unless a user demonstrates other impacts of time step jumps, I am fine assuming there is none.