Searching for Low-Mass Exoplanets Amid Stellar Variability with a Fixed Effects Linear Model of Line-by-Line Shape Changes

TL;DR

This paper introduces a fixed effects linear model that leverages line shape changes in stellar spectra to improve the detection of low-mass exoplanets by reducing stellar activity noise in radial velocity measurements.

Contribution

The study presents a novel fixed effects linear modeling approach that incorporates line shape-change covariates to better distinguish planetary signals from stellar variability.

Findings

Reduced RV residuals by approximately 70% using the new model.

Residuals are significantly lower than traditional methods.

Model performs comparably to state-of-the-art stellar activity mitigation techniques.

Abstract

The radial velocity (RV) method, also known as Doppler spectroscopy, is a powerful technique for exoplanet discovery and characterization. In recent years, progress has been made thanks to the improvements in the quality of spectra from new extreme precision RV spectrometers. However, detecting the RV signals of Earth-like exoplanets remains challenging, as the spectroscopic signatures of low-mass planets can be obscured or confused with intrinsic stellar variability. Changes in the shapes of spectral lines across time can provide valuable information for disentangling stellar activity from true Doppler shifts caused by low-mass exoplanets. In this work, we present a fixed effects linear model to estimate RV signals that controls for changes in line shapes by aggregating information from hundreds of spectral lines. Our methodology incorporates a wild-bootstrap approach for modeling uncertainty and cross-validation to control for overfitting. We evaluate the model's ability to remove stellar activity using solar observations from the NEID spectrograph, as the sun's true center-of-mass motion is precisely known. Including line shape-change covariates reduces the RV root-mean-square errors by approximately 70% (from 1.919 m s$^{-1}$ to 0.575 m s$^{-1}$) relative to using only the line-by-line Doppler shifts. The magnitude of the residuals is significantly less than that from traditional CCF-based RV estimators and comparable to other state-of-the-art methods for mitigating stellar variability.