Research/Space
Most of Gaia's close brown-dwarf orbits cannot be real
Gaia DR3's single-lined orbits with brown-dwarf minimum masses at 1-13 days should have produced about 408 TESS transits; 3 were seen. Even after counting heavier stars seen face-on, at most about a third of these orbits can be right, and under a typical population about 3%.
Nov 19, 2026 · 7 min read

Gaia's third data release contains thousands of single-lined spectroscopic orbits whose companions, at their minimum mass, would be brown dwarfs on orbits of one to thirteen days. If those orbits were real, many of the companions would pass in front of their stars, and NASA's TESS would see them. We checked 2,180 such stars with TESS data good enough to see a transit. About 408 transits were expected if every orbit were real with its minimum mass. TESS shows 3, all of them already known TESS Objects of Interest. A pre-registered test then asked whether heavier stars seen nearly face-on could explain the gap. They cannot: under every population we fixed in advance, most of these orbits must be wrong.
This matters because these orbits are a ready-made list of close brown-dwarf candidates, the rarest kind of close companion, and they feed population studies of the brown-dwarf desert. Our result says the list, taken as published, is mostly not real at these periods.
2,180
Gaia DR3 brown-dwarf-class orbits with sensitive TESS data
408
transits expected if every orbit were real at its minimum mass
3
substellar-size transits seen (0.5 expected by chance)
≤ 32%
of orbits can be right, even under the most generous population (95%)
The question
A spectroscopic orbit gives only a minimum companion mass. A missing transit therefore has two explanations. Either the orbit is wrong, with a period or amplitude that describes nothing real, or it is right and the companion is a heavier star whose orbit we see nearly face-on, which can never transit. Brown dwarfs are rare in close orbits, so the second explanation is serious: follow-up of Gaia brown-dwarf candidates has often found stellar binaries instead (Stevenson et al. 2023; 13 of 20 in arXiv:2609.08590). Before the shortfall could be read as wrong orbits, the face-on explanation had to be ruled out.
What we did
For each star we searched its TESS light curve at Gaia's period with an outlier-proof fold search, measured its sensitivity by injecting transits into its own data, and estimated false alarms from control folds at wrong periods. Known transiting companions were recovered in all eight data chunks. For each star we then used Gaia's mass function and primary mass to compute how likely the near-edge-on configuration is when companion masses follow a population of brown dwarfs and low-mass stars. We set a grid of 36 populations in advance: brown dwarfs at 0.05%, 0.2% or 0.5% per star, stars at 2%, 4% or 8%, and two mass laws each. The fraction of orbits that are right is φ = (hits − false alarms) / expected hits. The rules, the labels and their exact wording were committed before the job ran.
Brown-dwarf class: transits expected under each assumption vs seen
transits
What we found
The reproduction check passed: the job found the same 3 hits, 0.5 expected false alarms and 407.7 expected transits as the combined grade. Under the central population (4% low-mass stars, 0.2% brown dwarfs), face-on stars cut the expected count to 241, and φ = 0.010, at most 0.034 at 95%. Under the population most generous to real orbits (8% stars weighted toward low masses, only 0.05% brown dwarfs), the expected count falls to 25.7, but φ is still 0.097, at most 0.32. All 36 scenarios keep the upper bound below one half. The pre-registered label is MOSTLY_SPURIOUS, and its fixed wording reads: even if brown dwarfs are as rare as any published rate allows and every face-on star is counted, most of Gaia DR3's brown-dwarf-mass SB1 orbits at 1-13 days must be wrong.
Upper 95% bound on the share of orbits that can be right
Another way to say it: for every orbit to be real, close brown dwarfs would have to be far rarer, relative to close low-mass stars, than any population in the grid.
How dry the brown-dwarf desert would have to be for every orbit to be real
| Mass laws | Largest allowed brown-dwarf/star ratio | Drier than central ratio (0.05) | Drier than floor ratio (0.00625) |
|---|---|---|---|
| Flat, stars γ = 0 | 0.00058 | 87× | 11× |
| Log-flat, stars γ = −1 | 0.0018 | 28× | 3.5× |
The failures are concentrated at the shortest periods. In the combined grade, orbits at 5-13 days produce about 13% of their expected transits, against 1.5% below 5 days.
All classes: share of expected transits seen, by period
What it means
Individual Gaia DR3 orbits of this kind can be right: Unger et al. 2023 verified three with TESS, and one of them, TOI-5427, is among our three hits. At the opposite end of the catalogue, El-Badry et al. 2026 found that about half of the SB1-based compact-object candidates have spurious solutions. As far as we found, ours is the first population-level estimate for the low-mass end, and it is harsher: at 1-13 days, brown-dwarf-mass SB1 orbits should be treated as unconfirmed until other data back them. Gaia DR4, due on 2 December 2026, will show which DR3 orbits survive a longer baseline. Our test of that was pre-registered before the release, with its code already rehearsed.
Our predictions did not all hold. We had made the weaker label (wrong orbits unless the desert is very dry) the most likely outcome, at 55%, and gave the stronger label 35%. The stronger one came out. Both of our predicted ranges for φ held.


