Lomb-Scargle Periodogram
The Lomb-Scargle periodogram is the standard tool for finding periods in unevenly sampled astronomical time series data. It's particularly useful for detecting periodic signals in light curves from space missions like Kepler, K2, and TESS.
Overview
The Lomb-Scargle periodogram extends the classical periodogram to handle unevenly sampled data, which is common in astronomy due to observing constraints, data gaps, and variable cadences.
Basic Usage with Lightkurve
import lightkurve as lk
import numpy as np
# Create a light curve object
lc = lk.LightCurve(time=time, flux=flux, flux_err=error)
# Create periodogram (specify maximum period to search)
pg = lc.to_periodogram(maximum_period=15) # Search up to 15 days
# Find strongest period
strongest_period = pg.period_at_max_power
max_power = pg.max_power
print(f"Strongest period: {strongest_period:.5f} days")
print(f"Power: {max_power:.5f}")
to_periodogram defaults to normalization='amplitude': power is the amplitude of the best-fit sinusoid in flux units. Pass normalization='psd' for power spectral density, e.g. lc.to_periodogram(normalization='psd', maximum_period=15). Neither is a significance measure; see False-Alarm Probability with astropy below.
Plotting Periodograms
import matplotlib.pyplot as plt
pg.plot(view='period') # View vs period (not frequency)
plt.xlabel('Period [days]')
plt.ylabel('Power')
plt.show()
Important: Use view='period' to see periods directly, not frequencies. The default view='frequency' shows frequency (1/period).
Period Range Selection
Choose appropriate period ranges based on your science case:
- Stellar rotation: 0.1 - 100 days
- Exoplanet transits: use BLS or TLS instead (see the end of this guide); short box-shaped dips spread Lomb-Scargle power over many harmonics
- Eclipsing binaries: 0.1 - 100 days
- Stellar pulsations: 0.001 - 1 day
# Search specific period range
pg = lc.to_periodogram(minimum_period=2.0, maximum_period=7.0)
Interpreting Results
Power Significance
Higher power indicates stronger periodic signal, but be cautious:
- High power: not significance by itself; compute a false-alarm probability (below)
- Multiple peaks: Could indicate harmonics (period/2, period*2)
- Aliasing: Very short periods may be aliases of longer periods
Common Patterns
- Single strong peak: Likely the true period
- Harmonics: Peaks at period/2, period*2 suggest the fundamental period
- Aliases: Check if period*2 or period/2 also show signals
False-Alarm Probability with astropy
lightkurve wraps astropy.timeseries.LombScargle; use it directly for normalization control and false-alarm probabilities (FAP).
import numpy as np
from astropy.timeseries import LombScargle
fmin, fmax = 1 / 15, 1 / 0.1 # 1/day: periods 0.1-15 days
ls = LombScargle(t, y, dy, normalization='standard')
frequency, power = ls.autopower(minimum_frequency=fmin, maximum_frequency=fmax,
samples_per_peak=10)
best = np.argmax(power)
best_period = 1 / frequency[best]
grid = dict(minimum_frequency=fmin, maximum_frequency=fmax, samples_per_peak=10)
fap = ls.false_alarm_probability(power[best], method='baluev', **grid)
power_1pct = ls.false_alarm_level(0.01, method='baluev', **grid) # power needed for FAP = 1%
normalization:'standard'(default; 0-1, fraction of chi-square reduced by the sinusoid),'model','log', or'psd'(unnormalized). Compare powers only under the same normalization.method:'baluev'(default, fast analytic upper bound, accurate for small FAP),'bootstrap'(resamples the data, slow but makes no analytic approximation), plus'davies'and'naive'. FAP is implemented fornterms=1only.- Frequency grid:
autopowerspaces frequencies by1 / (samples_per_peak * baseline); use 5-10 samples per peak and setmaximum_frequencyexplicitly, since the default upper limit comes fromnyquist_factortimes the mean sampling rate. - FAP assumes white noise; red noise from stellar variability makes it optimistic.
Model Fitting
Once you find a period, you can fit a model:
# Get the frequency at maximum power
frequency = pg.frequency_at_max_power
# Create a model light curve
model = pg.model(time=lc.time, frequency=frequency)
# Plot data and model
import matplotlib.pyplot as plt
lc.plot(label='Data')
model.plot(label='Model')
plt.legend()
plt.show()
Dependencies
pip install lightkurve numpy matplotlib
References
When to Use This vs. Other Methods
- Lomb-Scargle: General periodic signals (rotation, pulsation, eclipsing binaries)
- Transit Least Squares (TLS): Specifically for exoplanet transits (more sensitive)
- Box Least Squares (BLS): Alternative transit detection method
For exoplanet transits, search with BLS (lc.to_periodogram(method='bls') or astropy.timeseries.BoxLeastSquares) or TLS; use Lomb-Scargle to find and remove stellar variability first.