[Paper Review] Period Determination of RR Lyrae Stars

R. F. Stellingwerf|arXiv (Cornell University)|Aug 25, 2011

Stellar, planetary, and galactic studies3 citations

TL;DR

This paper presents PDM2, an enhanced version of Phase Dispersion Minimization for period determination in RR Lyrae stars, improving accuracy and efficiency through optimized binning, statistical significance testing via Beta distribution or Monte Carlo methods, and period change detection. It demonstrates superior performance on noisy, gapped data—especially for Blazhko variables—achieving precise period estimates and detecting period changes consistent with O-C analyses.

ABSTRACT

The classic problem of detection of periodic signals in the presence of noise becomes much more challenging if the observation times are themselves periodic, contain large gaps, or consist of data from several different instruments. For RR Lyrae light curves the additional attribute of highly non-sinusoidal variation adds another dimension. This memo discusses and contrasts Discrete Fourier, Periodogram, and Phase Dispersion Minimization (PDM) analysis techniques. A new version of the PDM technique is described with tests and applications.

Motivation & Objective

• To address challenges in period detection for RR Lyrae stars with non-sinusoidal light curves, irregular sampling, and large data gaps.
• To improve upon the original PDM method by enhancing computational efficiency and statistical reliability.
• To enable accurate detection of period changes in Blazhko variables, which are difficult to analyze with fixed-phase methods.
• To provide a robust, user-accessible tool (PDM2) for astronomers analyzing variable star light curves.

Proposed method

• Uses a modified PDM algorithm that computes phase dispersion via binning of folded light curves, minimizing the sum of bin variances relative to total variance.
• Employs non-overlapping or overlapping bins based on data size, with adjustable switching point for optimal resolution.
• Applies statistical significance testing using either the Beta distribution (Schwarzenberg-Czerny, 1997) or Monte Carlo simulations (Nemec & Nemec, 1985) to assess false alarm probability.
• Introduces a bandwidth correction to align the distribution of extreme values with the expected exponential or Beta distribution for better significance estimation.
• Performs segmented analysis on data with large gaps, treating each observation cluster independently to reduce computational load and improve frequency resolution.
• Extends the method to detect period changes by scanning over a range of period derivative values (β), minimizing the residual variance of the folded light curve.

Experimental results

Research questions

• RQ1 Can PDM2 improve period determination accuracy in RR Lyrae stars with irregular sampling and large gaps compared to traditional Fourier and periodogram methods?
• RQ2 How effective is PDM2 in detecting period changes in Blazhko variables, where amplitude and period vary over time?
• RQ3 Does the use of overlapping vs. non-overlapping bins affect the precision of period estimation in low-S/N or sparse data sets?
• RQ4 How do the statistical significance estimates from the Beta distribution and Monte Carlo simulations compare in practice for real variable star data?
• RQ5 Can PDM2 reliably identify period changes in stars where O-C analysis is inconclusive or unavailable?

Key findings

• PDM2 achieves a period estimate of 1.60355 cycles/day for RR Lyrae V04, matching the published value of 1.60354 with high precision.
• For V04, the period change rate β was estimated as 1.5 ± 0.10 d/My, consistent with the O-C analysis result of 0.18 ± 0.05 d/My.
• For Blazhko variable V48, PDM2 detected a period change of β = -0.13 ± 0.05 d/My with a well-defined minimum in the θ(β) scan, reducing scatter on the light curve.
• For V83, the period change was estimated as β = -0.20 ± 0.10 d/My, confirmed by visual inspection of phase-folded data showing decreasing amplitude over time.
• The bandwidth-corrected extreme value distribution in PDM2 provides a clearer and more accurate interpretation of statistical significance than traditional methods.
• Monte Carlo simulations with 250–500 passes are recommended for high-confidence results, especially in borderline or low-S/N cases.