Efficient use of simultaneous multi-band observations for variable star analysis

TL;DR

This paper demonstrates how principal component analysis can be effectively used to analyze multi-band variable star data, accounting for noise and outliers, to identify new variable star candidates.

Contribution

It introduces a robust PCA-based methodology tailored for multi-band variable star data, addressing noise, outliers, and error estimation challenges.

Findings

PCA effectively identifies variable star candidates in multi-band data.

Robust methods improve variability detection accuracy.

Application to SDSS Stripe 82 data yields new RR Lyrae and SX Phoenicis candidates.

Abstract

The luminosity changes of most types of variable stars are correlated in the different wavelengths, and these correlations may be exploited for several purposes: for variability detection, for distinction of microvariability from noise, for period search or for classification. Principal component analysis is a simple and well-developed statistical tool to analyze correlated data. We will discuss its use on variable objects of Stripe 82 of the Sloan Digital Sky Survey, with the aim of identifying new RR Lyrae and SX Phoenicis-type candidates. The application is not straightforward because of different noise levels in the different bands, the presence of outliers that can be confused with real extreme observations, under- or overestimated errors and the dependence of errors on the magnitudes. These particularities require robust methods to be applied together with the principal component analysis. The results show that PCA is a valuable aid in variability analysis with multi-band data.