High-Pass Filtering and Gaussian Process Regularization: Stellar Activity Characterization Techniques Applied to the 55 Cancri Planetary System

Abstract: Doppler planet searches are complicated by stellar activity, through which cyclical changes in the host star's photosphere and chromosphere can mask or mimic planetary signals. A popular technique for modeling stellar activity is to apply a quasiperiodic Gaussian process (GP) kernel, which provides a flexible model with rigorous error propagation. However, observers must guard against overfitting, as a GP may be flexible enough to subsume other signals besides the one it is intended to model. To counteract overfitting, we introduce a curvature-penalizing objective function for fitting GP models to long-term magnetic activity cycles. We also demonstrate that a Gaussian filter can be an effective method of detrending radial velocities (RVs) so that shorter-period signals can be extracted even in the absence of a mathematical model of the long-term trend. We apply our methods to the heavily studied 55 Cancri system, fitting Keplerian orbits plus the GP activity-cycle model. We show that a 4-Keplerian model that includes planets b, c, e, and f combined with a GP for the activity cycle performs at least as well as the widely agreed-upon 5-planet system with its own GP activity model. Our results suggest that the existence of planet d cannot be established from the RVs alone; additional data are required for confirmation.