Variable classification in the LSST era: Exploring a model for quasi-periodic light curves

TL;DR

This paper evaluates stochastic process models, specifically DRW and QPO, for classifying variable stars and quasars in large LSST datasets, highlighting the effectiveness and limitations of these models in automated classification.

Contribution

It introduces a comparative analysis of DRW and QPO models for variable star classification, emphasizing the potential of QPO parameters to distinguish variable types.

Findings

QPO models better describe periodic and quasi-periodic variables.

DRW models are more suitable for quasars.

QPO parameters effectively differentiate variable classes.

Abstract

LSST is expected to yield ~10^7 light curves over the course of its mission, which will require a concerted effort in automated classification. Stochastic processes provide one means of quantitatively describing variability with the potential advantage over simple light curve statistics that the parameters may be physically meaningful. Here, we survey a large sample of periodic, quasi-periodic, and stochastic OGLE-III variables using the damped random walk (DRW, CARMA(1,0)) and quasi-periodic oscillation (QPO, CARMA(2,1)) stochastic process models. The QPO model is described by an amplitude, a period, and a coherence time-scale, while the DRW has only an amplitude and a time-scale. We find that the periodic and quasi-periodic stellar variables are generally better described by a QPO than a DRW, while quasars are better described by the DRW model. There are ambiguities in interpreting the QPO coherence time due to non-sinusoidal light curve shapes, signal-to-noise, error mischaracterizations, and cadence. Higher-order implementations of the QPO model that better capture light curve shapes are necessary for the coherence time to have its implied physical meaning. Independent of physical meaning, the extra parameter of the QPO model successfully distinguishes most of the classes of periodic and quasi-periodic variables we consider.