CrossCorrelation Gives Strange Results

[wendeborn8]

I am using Stingray's CrossCorrelation function to determine the time-lag between two light-curves, but am having some trouble.

As a simple case, if I calculate the time-lag between two identical light-curves, the result of stingray.crosscorrelation.time_shift depends on the number of data points. An odd number of points yields a time-shift of 0, as one would expect. However, an even number of points yields some fraction of the time sampling. This is also reflected when plotting the CrossCorrelation for various time lags.

Code to reproduce:

from stingray import Lightcurve as LC
from stingray.crosscorrelation import CrossCorrelation as XC

n = 15
dt = 1
times = np.arange(0, n * dt, dt)
data = np.sin(times)
lc = LC(times, data)
xc = XC(lc, lc) # CrossCorrelation of identical light-curves

print(xc.time_shift)

For any odd n, the result is 0. But for any even n, the result is >~0.5, getting closer to 0.5 for larger n. For n=2, the result is 1.0. The final result approximates half the value of dt.

Am I using/interpreting these functions incorrectly, or is Stingray doing something wrong here?

[matteobachetti]

Thanks @wendeborn8 !
I think this is something unsurprising but nonetheless troubling :D
Unsurprising, because it is common for these methods to have corner cases based on the evenness/oddity of the input light curves.
Troubling, because we should account for them making the result as intuitive as possible for the users, and we thought we had done it right.
We'll investigate it. In the mean time, please consider a half step in those conditions as a zero lag.

[mihirtripathi97]

Working on this!

[Ayushag1]

Working on this!

[mihirtripathi97]

Working on this!

Hey, I have already almost done the work on this as commented earlier. Please find another issue.