Generating synthetic detection rates for RV observations

One of the most effective techniques for detecting exoplanets is the radial velocity (RV) method, which tracks tiny shifts in a star’s RV caused by the gravitational pull of orbiting planets. In this work, which constitutes the first step in an effort to quantify observational biases and develop an AI-based tool (Davoult et al., in prep), we developed a simulation framework that injects synthetic planetary signals, generated using the Bern model, into pre-simulated stellar RV curves to compute detection rates across various orbital and planetary parameters. These detection rates are subsequently compared to those derived in Mayor et al. 2011 revealing differences in sensitivity, particularly for smaller planets and shorter orbital periods. Notably, our method shows higher detection rates for planets with larger RV amplitudes (K > 1 m·s⁻¹), and occasionally detects low-mass planets (< 1 Earth mass) under favorable conditions. In contrast, Mayor et al.’s approach tends to outperform our method in the low-amplitude regime (K < 1 m·s⁻¹), likely due to the presence of instrumental noise in our model. In a future work (Davoult et al. in prep.), this physically motivated framework will serve as a foundation for training a machine learning model capable of reproducing detection rates in a much faster way by learning from the outcomes of our simulations.