AI flags 10,000+ TESS exoplanet candidates

Exoplanet hunting just got a lot more crowded. A Princeton-led team went back through NASA TESS data and surfaced 11,554 planet candidates in one sweep, with 10,091 of them newly flagged rather than already sitting in the catalog. (arxiv.org) That matters because TESS has been good at finding planets, but a lot of its search machinery has stayed biased toward brighter, easier stars. This new pass pushes much deeper into the faint-star pile — and that is where a huge amount of missed signal can hide. ### What did they actually do? They used the T16 light-curve set — a uniformly detrended, systematics-corrected reprocessing of TESS Cycle 1 full-frame images — and ran a semi-automated transit search across essentially the whole sample. The scale is the eye-popper: 83,717,159 light curves, stars down to TESS magnitude 16, and candidate orbital periods between 0.5 and 27 days. (arxiv.org) ### Why is TESS data so hard to search? TESS looks for tiny dips in starlight when a planet crosses in front of its star. But lots of other things can fake that dip — stellar noise, instrumental systematics, eclipsing binaries, and one-off glitches. (arxiv.org) When you are scanning tens of millions of stars, the real problem is not just “can you find a dip?” It is “can you rank which dips are worth a telescope night?” ### Where does the AI part come in? The paper describes the search as machine-learning-assisted, not magic autopilot. (arxiv.org) Basically, the models help clean up the flood of possible signals and push the search into regions that older pipelines handled less well, especially faint stars in full-frame images. NASA has been moving in the same direction with ExoMiner++, another AI system built to sift TESS and Kepler signals and sort likely planets from lookalikes. ### Are these all real planets? No — and that is the catch. “Candidate” means a signal looks planet-like but still needs vetting and, often, follow-up observations. (science.nasa.gov) In the new haul, 411 are single-transit events, which means the team did not even try to pin down full orbital parameters for those cases. ### So how do we know the pipeline works? They did not just dump a giant list and walk away. The team followed up one host star, TIC 183374187, with radial-velocity measurements from Magellan/PFS and confirmed a newly identified hot Jupiter. (arxiv.org) One confirmation does not validate 10,000 planets, obviously, but it does show the pipeline can pull out at least some genuinely new worlds from the noise. ### Why do faint stars matter so much? Because older TESS searches have leaned toward bright stars, where signals are easier and follow-up is cheaper. (arxiv.org) But planet occurrence rates imply there should be many more planets around fainter stars too. This survey is basically a reminder that the archive still has a huge long tail of undiscovered systems if you are willing to accept harder vetting work later. ### Does this change the exoplanet field right now? It changes the to-do list more than the confirmed planet count. NASA said in January that ExoMiner++ alone pulled out 7,000 TESS candidate targets on an initial run, which already showed how much unmined material remains in the archive. (arxiv.org) Add this T16 haul, and the bottleneck shifts even harder from “finding signals” to “verifying the best ones first.” ### Bottom line? The headline is not “10,000 new planets.” It is “10,000 new places to look.” But that is still a big deal. (arxiv.org) The new catalog more than doubles known TESS exoplanet candidates, and it turns the next phase of planet hunting into a triage problem — deciding which faint, promising signals deserve precious telescope time first. (science.nasa.gov)