pyaneti – II. A multidimensional Gaussian process approach to analysing spectroscopic time-series

Abstract

The two most successful methods for exoplanet detection rely on the detection of planetary signals in photometric and radial velocity time-series. This depends on numerical techniques that exploit the synergy between data and theory to estimate planetary, orbital, and/or stellar parameters. In this work we present a new version of the exoplanet modelling code pyaneti. This new release has a special emphasis on the modelling of stellar signals in radial velocity time-series. The code has a built-in multi-dimensional Gaussian process approach to modelling radial velocity and activity indicator time-series with different underlying covariance functions. This new version of the code also allows multi-band and single transit modelling; it runs on Python 3, and features overall improvements in performance. We describe the new implementation and provide tests to validate the new routines that have direct application to exoplanet detection and characterisation. We have made the code public and freely available at https://github.com/oscaribv/pyaneti. We also present the codes citlalicue and citlalatonac that allow one to create synthetic photometric and spectroscopic time-series, respectively, with planetary and stellar-like signals.

One year of AU Mic with HARPS – II. Stellar activity and star–planet interaction

“…From our estimates of 𝐾 b = 7.1 ± 3.8 m s −1 and 𝐾 c = 13.3 ± 4.1 m s −1 , we obtain planet masses of 𝑀 b = 14.3 ± 7.7 M ⊕ and 𝑀 c = 34.9 ± 10.8 M ⊕ . These estimates are slightly larger but still consistent within 1 𝜎 with the planet masses obtained in paper I by jointly modelling the stellar activity signals and the planet RV signatures using a multi-dimensional Gaussian Process framework (Rajpaul et al 2015;Barragán et al 2022). Over the 121 days spanned by our 2021 data set, AU Mic's brightness topology has significantly evolved (see Fig.…”

“…Unsurprinsingly, the periodograms of the CCF-based activity indicators exhibit prominent peaks at P rot and its first harmonic. By modelling each of these time series independently using a Gaussian Process (GP; Rajpaul et al 2015) with the quasi-periodic covariance kernel introduced in Haywood et al (2014) as implemented in the pyaneti python package (Barragán et al 2022), we find consistent rotation periods of 4.863 ± 0.004 d, 4.86 ± 0.01 d and 4.867 ± 0.004 d for the RV, FWHM and BIS time-series, respectively 2 . As illustrated in Fig.…”

The EXPRES Stellar Signals Project II. State of the Field in Disentangling Photospheric Velocities

“…; Dodson-Robinson submitted; Camacho in Prep. ; Barragán et al 2022). Indicator-dependent methods will only be sensitive to signals that are reflected in the provided indicators; for example, if the used indicators do not track the effects of oscillation or granulation, then these methods will not return models sensitive to these effects.…”

“…This makes use of the flexibility of a GP model while also constraining the model with indicator time series to capture only stellar signal related variations. GLOM can be thought of as a generalization of the multi-dimensional GP method implemented in pyaneti (Rajpaul et al 2015;Barragán et al 2019Barragán et al , 2022. This method requires dense sampling throughout the characteristic timescale of the signal being modeled (e.g., the stellar rotation period for spots).…”

Modelling stellar activity with Gaussian process regression networks

“…More recently, Barragán et al (2021) extended this approach to a more general set of activity indicators and presented an efficient implementation within the pyaneti package (Barragán et al 2019a). The framework has also been extended by Jones et al (2017), who included terms proportional to the second time derivative of the latent GP and by Gilbertson et al (2020), who developed the GLOM model allowing for the use of any covariance kernel.…”