Red noise-based false alarm thresholds for astrophysical periodograms via Whittle’s approximation to the likelihood Astropy ( 7 ; 4 ; 6 ) Facilities: ESO:3.6m (European Southern Observatory (ESO) 3.6m Telescope at La Silla Observatory), HET (University of Texas, Austin 9.3m Hobby-Eberly Telescope at
Amna Ejaz, Sarah Dodson-Robinson, Charlotte Haley · 2025
Read Red noise-based false alarm thresholds for astrophysical periodograms via Whittle’s approximation to the likelihood Astropy ( 7 ; 4 ; 6 ) Facilities: ESO:3.6m (European Southern Observatory (ESO) 3.6m Telescope at La Silla Observatory), HET (University of Texas, Austin 9.3m Hobby-Eberly Telescope at by Amna Ejaz, Sarah Dodson-Robinson, Charlotte Haley (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-18205.txt.
Introduction
Amna Ejaz, Sarah Dodson-Robinson, and Charlotte Haley help planet hunters tell true rhythms from red noise. Star light often flickers with slow, rolling patterns called red noise, which can fake a planet's tidy beat. Charts of rhythm strength are judged against a false alarm line, yet that line is often drawn as if the noise were plain white static.
Their method first flattens the roll with a fitted trend, then sets the alarm line from a quick likelihood shortcut, now stretched to uneven sampling. On Kepler star data, the red noise line kept only true spin signals, while the white noise line blessed fakes. Tests on three stars and fake data agree, and the code is public.
Catalog record
| CREATOR | Amna Ejaz, Sarah Dodson-Robinson, Charlotte Haley |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-18205
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Red noise-based false alarm thresholds for astrophysical periodograms via Whittle’s approximation to the likelihood Astropy ( 7 ; 4 ; 6 ) Facilities: ESO:3.6m (European Southern Observatory (ESO) 3.6m Telescope at La Silla Observatory), HET (University of Texas, Austin 9.3m Hobby-Eberly Telescope at McDonald Observatory), Kepler (NASA 0.95m Kepler Satellite Mission) Amna Ejaz, Sarah Dodson-Robinson, Charlotte Haley Abstract Astronomers who search for periodic signals using Lomb-Scargle periodograms rely on false alarm level (FAL) estimates to identify statistically significant peaks. Although FALs are often calculated from white noise models, many astronomical time series suffer from red noise. Prewhitening is a statistical technique in which a continuum model is subtracted from log power spectrum estimate, after which the observer can proceed with a white-noise treatment. Here we present a prewhitening-based method of calculating frequency-dependent FALs. We fit power laws and autoregressive models of order 1 to each Lomb-Scargle periodogram by minimizing the Whittle approximation to the negative log-likelihood (NLL), then calculate FALs based on the best-fit model power spectrum. Our technique is a novel extension of the Whittle NLL to datasets with uneven time sampling. We demonstrate FAL calculations using observations of α \alpha Cen B, GJ 581, HD 192310, synthetic data from the radial velocity (RV) Fitting Challenge, and Kepler observations of a differential rotator. The Kepler data analysis shows that only true rotation signals are detected by red-noise FALs, while white-noise FALs suggest all spurious peaks in the low-frequency range are significant. A high-frequency sinusoid injected into α \alpha Cen B log R H K ′ \log R^{\prime}_{HK} observations exceeds the 1% red-noise FAL despite having only 8.9% of the power of the dominant rotation signal. In a periodogram of HD 192310 RVs, peaks associated with differential rotation and planets are detected against the 5% red-noise FAL without iterative model fitting or subtraction. Software for calculating red noise-based FALs is available on GitHub. \NewPageAfterKeywords I Introduction Extreme-precision spectrographs with radial velocity precision approaching 10 cm s -1 are searching for Earth-mass planets orbiting sunlike stars ( 50 ; 14 ; 101 ) . Since the primary mathematical goal in an RV planet search is to detect and confirm a periodic signal in the time series, the typical first step is to examine a power spectrum estimate. The most commonly used power spectrum estimator for RV time series is the generalized Lomb-Scargle periodogram ( 146 , GLS periodogram;) , though there are other period-search tools such as the Bayesian and stacked Bayesian GLS ( 93 ; 92 ) , the apodized Kepler periodogram ( 60 ) , compressed sensing techniques ( 42 ; 63 ) , and Welch’s method ( 140 ; 40 ) . When there is a peak in the periodogram, the peak frequency serves as the estimator for the frequency of the periodic process ( 91 , e.g.) . Possible sources of periodic signals are p-modes, planets, star rotation, and magnetic activity cycles. If the time series had infinite length, each periodogram peak would correspond to a true physical periodicity, but this is not the case in practice. Even if the spectrograph has extreme precision, the Lomb-Scargle periodogram ( 112 , as normalized by) gives spurious positive detections associated with statistical fluctuations above statistical significance levels given by \chi^{2}_{2} quantiles ( 122 ) . (See Appendix A for a discussion of the relationship between false alarm levels and \chi^{2}_{2} quantiles). To address the false positive problem and utilize error estimates in the observations, detection significance is quantified by the false alarm probability (FAP) that a time series with no true periodic signal could produce a periodogram peak with the observed power. The FAP calculation tools that are implemented in the timeseries.LombScargle class of the astropy software package ( 5 ) are the analytical method of 8 , which approximates FAP as a function of periodogram power based on the theory of extreme values of stochastic processes ( 82 ) , and the bootstrap method ( 34 ; 28 ) , which involves creating an ensemble of noise series by randomly sampling the observations with replacement. Both methods assume that the time series contains an uncorrelated white noise background. However, RV and magnetic activity-indicator time series often have a red noise background, which is indicated by a downward trend in a GLS periodogram plot with frequency on the x-axis and a logarithmic scale in the y-axis ( 40 , see Appendix B of) . For example, red noise is obvious in Figure 1 , which shows the GLS periodogram of the \alpha Cen B \log R_{HK}^{\prime} time series from 46 . The pioneering work by 9 illustrates how red noise in the RVs of GJ 876 led to inaccurate estimates of orbital eccentricities and parameter uncertainties under the traditional white noise-based analysis method. Methods for computing red noise-based FAPs have been proposed by 134 and 135 for standard periodograms of time series with uniform timesteps and by 10 ; 11 ; 84 ; 83 ; 35 ; 64 for GLS periodograms. These methods are based on models of the covariance matrix. Figure 1: GLS periodogram \hat{S}^{LS}(f_{k}) (green) of activity indicator \log R^{\prime}_{HK} of \alpha Cen B ( 46 ) with a red noise background. The vertical navy dotted line is the rotation signal, the horizontal black dash-dot line is a white noise fit to the spectrum, and the purple solid line and red dotted line are the best-fit versions of two red noise models, AR(1) and power law (see section § II for a detailed description of the red noise models used in this work). Just as noise properties can be inferred from a covariance matrix, they can also be estimated using parametric models in the frequency domain ( 75 ; 39 ) . For uniformly sampled data with constant timesteps, statisticians often prewhiten by fitting a continuum model to \log[\hat{S}(f)] (where \hat{S}(f) is the estimated power spectrum at frequency f ), then subtracting the model ( 58 ; 12 , e.g.) . The periodic signals are then identified as peaks in the residual spectrum at which the power exceeds a statistically significant \chi^{2} threshold. In this work, we implement the first step of a prewhitening process by fitting a continuum model in the frequency domain. However, instead of subtracting the model from \log[\hat{S}(f)] and searching for peaks in the residuals, we compute frequency-dependent false alarm levels that account for noise artifacts arising from uneven sampling, in which the timesteps vary. Here we extend Whittle’s negative log-likelihood approximation ( 142 ; 141 ; 71 , see § III.1 for a mathematical description) for power spectrum model fitting to time series with uneven observing cadence. The best-fit models are used to calculate red noise-based false alarm levels (FALs), which are curves with constant FAP. Although the Whittle negative log-likelihood (NLL) has been applied to diverse topics such as hydrology ( 90 ) , econometrics ( 22 ) , and Earth’s polar motion ( 119 ) , this is the first use on unevenly sampled astronomical time series. Here we focus on planet-search data, but the method can be applied to any type of time series. This paper is accompanied by a software package that implements Whittle NLL fitting