Gabriella Contardo
September 12, 2024
Zoom recording
Abstract:
Stellar infrared excess usually indicates the presence of debris disks or dust around a star. While common in young stars due to their protoplanetary disks, extreme IR excess in older stars is more atypical and could come from either rare or short-lived events. Recent works have investigated such excess at around 5-12μm wavelengths, finding around 20 candidates (“occurrence rate” of 0.01%) with fractional luminosities around 0.02, some tentatively associated with planetary collisions. These extreme “anomalies” are of interest as they potentially sit in an under-populated region of the “disk evolution” path depending on the stellar ages. They could also provide indirect observations of planetary systems. Interestingly, another possible source (while more speculative) of IR excess is mega-structures such as Dyson spheres or swarms. However, distinguishing them from natural processes wouldn’t be trivial. Expanding our set of candidates and our knowledge of such objects is thus crucial. Previous searches have focused on using WISE data notably in the W3 band (12μm) with high SNR requirement, dramatically limiting the amount of data. We propose in our work to rely only on the W1 and W2 bands (3.4-4.6μm) combined with Gaia DR3 and 2MASS, thus unlocking a much larger amount of data, at the cost of detecting only “extreme” excesses showing in those bands. We focus on Sun-like stars (FGK) on the main sequence, leading to a catalog of 4.9M stars. Our search is designed as a data-driven contextual anomaly detection pipeline: to identify outlier candidates, we use a combination of prediction errors from a set of Random Forests trained on our catalog, and statistics using prediction errors of similar neighbouring points. Our approach bypasses the need for actual stellar modelling while providing a high detection sensitivity, crucial in our case. We identify 53 candidates with fractional luminosities between 0.005 to 0.1, consistent with previous EDDs candidates and potential planetary collision events.