Mitigating stellar activity jitter with different line lists for least-squares deconvolution: analysis of a parametric and a randomised line selection

TL;DR

This study compares line list selection methods for least-squares deconvolution to reduce stellar activity jitter in radial velocity data, finding that a randomized approach significantly improves jitter mitigation and aids in exoplanet detection around active M dwarfs.

Contribution

It introduces a randomized line selection algorithm that outperforms parametric methods in reducing stellar activity jitter in RV measurements of M dwarfs.

Findings

Randomized line selection reduces RV jitter by over 50%.

Parametric line selection reduces jitter by less than 10%.

The method enables detection of synthetic planets despite stellar activity.

Abstract

Stellar activity limits the radial velocity search and characterisation of exoplanets, as it introduces spurious jitter in the data sets and prevents the correct retrieval of a planetary signal. This is key for M dwarfs, considering that they manifest high activity levels and are primary targets for present and future searches of habitable Earth-like planets. Effective filtering of activity is crucial to achieve the sensitivity required for small planet detections. Here, we analyse the impact of selecting different line lists for least-squares deconvolution on the dispersion in our RV data sets, to identify the line list that most effectively reduces the jitter. We employ optical observations of the active M dwarf EV Lac collected with ESPaDOnS and NARVAL, and study two line selection approaches: a parametric one based on line properties and a randomised algorithm that samples the line combination space. We test the latter further to find the line list that singles out the activity signal from other sources of noise, and on AD Leo and DS Leo to examine its consistency at mitigating jitter for different activity levels. The analysis is complemented with planetary injection tests. The parametric selection yields a RV RMS reduction of less than 10%, while the randomised selection a systematic >50% improvement, regardless of the activity level of the star examined. Furthermore, if activity is the dominant source of noise, this approach allows the construction of lists containing mainly activity-sensitive lines, which could be used to enhance the rotational modulation of the resulting data sets and determine the stellar rotation period more robustly. Finally, the output line lists allow the recovery of a synthetic planet (0.3-0.6 MJup on a 10 d orbit) in the presence of both moderate (20 m/s semi-amplitude) and high (200 m/s) activity levels, without affecting the planet signal substantially.