Detecting Exoplanets with Sparse Recovery Techniques
Intervenant : Nathan Hara
Observatoire de Genève
When a star has planetary companions, it describes an epicyclic motion around the center of mass of the system. The component of the star velocity in the observer's direction, or radial velocity, can be estimated from spectroscopic measurements thanks to Doppler effect. In the radial velocity time series, one searches for periodic variations due to planets. Various phenomena complicate the task: the signals from several planets can be hard to disentangle, stellar noise might mimic planetary signals, correlated noise buries other signals etc. Most recent approaches privilege fitting a complete model instead of fitting planets one by one, since global searches reduce both rates of false positives and false negatives. Unfortunately, the performance improvement comes at the cost of a greater computational complexity.
We will present a method based on sparse recovery algorithms that allows to search for several planets at once but without random searches, which makes the algorithm faster than random searches algorithms. The tool can be used as a Lomb-Scargle periodogram or a fast Fourier transform, but has the advantage of avoiding most cases where the tallest peak of the periodogram is spurious. We will show several examples of application of the method, a brief historic of similar ideas in astronomy and how it can be adapted to other inference problems.