Abstract: Can Gaia combined with AI help-us plant seeds in the Brown-Dwarf desert?

Gathering statistics on brown-dwarfs and better understanding their mass distribution  is critical to uncovering the underlying mechanisms for their formation. The current census of brown-dwarfs has revealed a scarcity of these sub-stellar objects on short period orbits (P<~100 days),  around solar-type stars, with minimum around 30–35 Mjup (Grether et al. 2004,Persson C. et al. 2019, Ma B. et al 2013, Stevenson J. et al. 2023).  This raises some interesting questions and several theories regarding the possible formation paths or post-formation migration for brown-dwarfs.  In this study, we take advantage of the rich dataset available in Gaia DR3 to try to shead some light into  this interesting question, by providing a mechanism for the systematic detection of these (and possibly other)  sub-stellar objects within Gaia DR3 data. To do so, we generate a deep learning model that takes advantage of the correlation existing  between quality of the astrometic fit performed by the Gaia data reduction system and the presence of an unseen companion.   To generate training data for our deep neural network , we use a probabilistic generative model, that simulates a stellar population composed of both single and binary systems  with primary masses ranging from 0.1 to 1.5 solar masses and secondary companion masses spanning 10 to 80 Jupiter masses  (a range intentionally selected to corver the "brown-dwarf desert"). We then generate astrometric epoch data for each system and also simulate the observations that Gaia would perform on each of our systems.  From these, we perform an astrometric fit similar to that one performed by Gaia data reduction system and obtain a set of quality-of-fit statistics  that, together with a known (single/binary) flag are fed into a deep neural network (DNN) to map the underlying correlation  between the presence of unseen companions and astrometric quality-of-fit.  We then apply our model to a sample of F,G,K,M stars from the Gaia DR3 and obtain the probablity of each of those stars to host  a sub-stellar companion. Using this model we identify ~8000 new candidate stars to host sub-stellar companions. Using suplementary data from existing ground RV surveys we can constrain the masses for the potential companions,  but only for a small subset of 20 of these candidates. The  estimated masses of these potential 20 sub-stellar companions are in the range  30-50M_Jup, and although statistics are scarce, this would locate them in the dry-est part of the brown-dwarf desert.