Enhancing Precision Radial Velocity Analysis: Validating the Template Method and Developing Generative Models

Dhvani Doshi ( Université McGill )

Precision radial velocity (pRV) is a highly effective method for determining a planet's mass, which in turn informs its density and whether it is rocky or gaseous. This study focuses on pRV analysis using ground-based telescopes in the near-infrared for Mdwarfs. In this regime, there are two sources of modeling difficulties: the stellar spectrum and the telluric spectrum. To construct a ground-truth stellar spectrum, researchers typically use the template method, where observations are averaged to reduce noise. Despite its widespread use, it remains unclear whether this method effectively eliminates telluric contamination from the stellar spectrum. This study aims to validate the template method by assessing its efficacy under different conditions. We employ synthetic observations based on PHOENIX stellar models and HITRAN tellurics to characterize the residual noise left in templates and its impact on radial velocity estimation. Furthermore, we develop score-based generative diffusion models to learn the ground truth stellar and telluric spectra. These models are then implemented as priors in our pRV analysis to generate posterior samples of the underlying spectra in our observations. This approach aims to bypass the need for the template method and improve the accuracy of RV retrieval by better characterizing observational noise.