Non-Gaussian Noise Modeling for Radial Velocity Exoplanet Detection

RV methods (Hara & Ford 2022) are limited by correlated noise from stellar activity and instruments, which are typically modeled as Gaussian processes. This research would challenge that norm by using variational autoencoders to learn non-Gaussian noise patterns directly from RV time-series data. Inspired by Garvin et al.'s (2024) rejection of Gaussian assumptions in spectroscopy, the approach would incorporate stellar magnetic cycle models and instrumental drift simulations into the training process. Unlike parametric models (e.g., Muthukrishna et al. 2021), this would detect subtle signals of Earth-like planets buried in complex noise. The method could be validated on synthetic data with injected non-Gaussian noise, then applied to archival HARPS/ESPRESSO data to identify missed low-mass exoplanets. This directly addresses the "complex, temporally correlated signals" barrier highlighted by Hara & Ford (2022).

References:

1. Real-time Detection of Anomalies in Multivariate Time Series of Astronomical Data. D. Muthukrishna, K. Mandel, M. Lochner, S. Webb, G. Narayan (2021). arXiv.org.
2. Machine Learning for Exoplanet Detection in High-Contrast Spectroscopy: Revealing Exoplanets by Leveraging Hidden Molecular Signatures in Cross-Correlated Spectra with Convolutional Neural Networks. E. O. Garvin, M. Bonse, Jean Hayoz, G. Cugno, Jonas Spiller, P. Patapis, Dominique Petit Dit de la Roche, Rakesh Nath-Ranga, O. Absil, Nicolai F. Meinshausen, S. Quanz (2024). Astronomy & Astrophysics.
3. Statistical Methods for Exoplanet Detection with Radial Velocities. N. Hara, E. Ford (2022). Annual Review of Statistics and Its Application.