{{ message }}
Repository navigation
LombScargle(..., method="fast") disagrees with the exact method by more than expected on tightly clustered, non-uniform sampling #20449
Description
Description
No response
Expected behavior
For the standard normalization (power bounded to [0, 1]), fast and slow should agree to within a small absolute budget (on the order of 1e-6) across the range of sampling patterns the method is documented to support.
How to Reproduce
import numpy as np
from astropy.timeseries import LombScargle
rng = np.random.default_rng(0)
span = 1094.6
count = 155
t = np.sort(np.r_[rng.normal(0, span * 1e-8, count - 1), span]) # 154 points
# clustered near t=0,
# plus one at t=span
y = rng.normal(size=count)
frequency = 0.8 + 0.0037 * np.arange(791)
ls = LombScargle(t, y)
fast_power = ls.power(frequency, method="fast",
assume_regular_frequency=True, normalization="standard")
slow_power = ls.power(frequency, method="slow", normalization="standard")
print(np.max(np.abs(fast_power - slow_power))) # 7.53e-06Versions
Latest
Activity