A survey of 63,254 stars puts the brown dwarf desert on a relative scale

A desert that moves with the star

The “brown dwarf desert” is a shortage of companions between the populations of giant planets and stellar companions. Previous surveys established this scarcity around Sun-like stars, but testing whether its location changes with stellar mass has been difficult: different observing methods cover different stars and orbital distances.

William Thompson and colleagues analyzed 63,254 stars using Hipparcos and Gaia measurements. Their central finding is that the shortage follows the companion-to-host mass ratio, called q. Companions become particularly scarce near q of 2%—about one-fiftieth of the host’s mass.

That distinction matters. If the boundary scales with the host, an object on the planet-like side around a massive star can outweigh an object on the stellar-like side around a smaller star. One universal companion-mass threshold would not separate those populations.

Reading the gravitational wobble

An orbiting companion changes a star system’s apparent motion across the sky. The researchers combined Hipparcos observations with Gaia’s second and third data releases, using differences in measured motion and departures from a simple astrometric fit. Additional constraints came from Gaia radial-velocity variability, resolved neighboring sources and stellar brightness measurements.

The main analysis concerns orbital periods of approximately 3–94 years, corresponding roughly to distances of 2–20 astronomical units, depending on host mass. One astronomical unit is the Earth–Sun distance.

Crucially, this is not a catalog of 63,254 confirmed companion detections. Many individual systems have ambiguous companion masses and periods. Instead, a hierarchical Bayesian model fits the population and its constituent systems together, allowing each system to contain zero to three companions.

Strong measurements help interpret weaker ones, while uncertainty about companion numbers and properties carries through the calculation. The resulting occurrence rates describe expected companions per star, not simply the fraction of stars with companions. Accounting for missed companions depends on correctly modeling each star’s sensitivity, rather than applying a separate detection-completeness correction.

How deep is the gap?

Across the full sample, the inferred occurrence at the desert floor, between mass ratios of 1.8% and 3.2%, was about 33 times lower than the planet-side peak and 19 times lower than the stellar-side peak. Those contrasts have substantial uncertainties.

To test how the minimum shifts with stellar mass, the team separately analyzed host-mass groups, retaining sufficiently sensitive stars. Across the K-through-B groups, a constant-ratio fit placed the minimum at 1.76%, with reported uncertainties of minus 0.19 and plus 0.26 percentage points.

The smallest-host group contained only 119 sensitivity-selected M dwarfs. It showed evidence consistent with the trend but was excluded from that quantitative fit. Thus, the strongest measured scaling does not have equally strong support at every stellar mass.

The authors illustrate the overlap with stellar-like companions around K stars near 13 Jupiter masses and planet-like companions around B stars reaching roughly 27 Jupiter masses. Their interpretation challenges using the deuterium-burning threshold alone as a formation-based dividing line; it does not establish how every individual object formed.

Chemistry offers a second clue

For 18,672 FGK hosts, the researchers examined metallicity: the abundance of elements heavier than helium. Companions with mass ratios between 0.1% and 1%, below the desert, became more common with increasing metallicity. Companions between 1.8% and 10% showed the opposite trend.

The evidence against a flat metallicity relation was stronger below the desert, at 4.4 standard deviations, than above it, at 2.6 standard deviations.

The authors interpret this reversal as support for two formation channels: bottom-up growth through core accretion and top-down fragmentation. However, population correlations are not direct observations of formation. Two channels remain a plausible explanation, not a demonstrated history for each companion.

Where the inference stops

The sample is limited by brightness, distance and catalog requirements, which mildly select against some binaries. The population model also assumes that systems grouped together share companion distributions. Its ability to extract structure from ambiguous measurements makes these assumptions important.

Beyond approximately 100-year periods, companion mass and orbital separation become difficult to disentangle. The study cannot establish whether the desert persists there, nor whether planet occurrence remains flat out to 100 years. Individual planetary periods and orbital eccentricities remain poorly constrained.

Comparisons with overlapping radial-velocity and imaging surveys broadly support the results where sensitivities overlap. However, inferred occurrence rates are substantially lower than those of one other combined radial-velocity and astrometry study; the authors suggest completeness corrections as a possible explanation, not a settled resolution.

A preliminary population result

The source is arXiv version 1, posted November 9, 2026, and is a preprint. Its principal contribution is population-level evidence that the desert tracks relative mass. Better Gaia measurements and complementary observations could sharpen that conclusion and test its formation-based interpretation.