Munim Technologies

Research/Space

NASA Exoplanet Archive stellar densities stored in the wrong unit

A physical self-consistency check of every parameter set in the NASA Exoplanet Archive found that stellar densities from four references, led by 1,414 rows from Fulton & Petigura 2018, are in solar units, although the column is documented in g/cm³.

Oct 14, 2026 · 5 min read

The NASA Exoplanet Archive is where most researchers get the masses, radii and orbits of known planets and their stars. We tested whether its numbers agree with each other, using relations that must hold within a single published solution, such as Kepler's third law or density = mass ÷ volume. The clearest result is a unit error: stellar densities from four references are stored in solar units, although the Archive documents the column in g/cm³. The largest case, Fulton & Petigura 2018, covers 1,414 rows. A density that is 1.41 times too low can look like an eccentric orbit, so the error can mislead real analyses.

40,336

parameter sets checked in the Planetary Systems table

1,414

Fulton & Petigura 2018 rows with densities in solar units

0.716 → 1.01

their median ratio before and after the ×1.4098 fix

The question

The Archive holds two main tables. The Planetary Systems table (ps) keeps every published solution as its own row, about 40,000 in all. The composite table (pscomppars) picks one value per column for each planet, often from different papers. Errors in either can flow into catalogs, population studies and target lists. Earlier work tested Kepler's third law on the default parameters and found five outliers. We wanted to know how many internal contradictions there are across every row, and which of them are real.

What we did

A standard-library Python script downloads both tables and tests nine relations: Kepler's third law, planet bulk density, a/R★, Rp/R★, transit depth against (Rp/R★)², the transit-derived stellar density, the stored stellar density against 1.4098 g/cm³ · M★/R★³, surface gravity, and insolation. A row is flagged only when the mismatch exceeds a physical tolerance and, where uncertainties are given, three standard deviations. The first full run checked 40,336 ps rows and 6,445 composite rows and raised 17,675 flags.

Planetary Systems rows tested by each check

Run 0.2.0 on the 2026-10-14T14:00Z download. A row is tested only when it has every value a relation needs. Source: results/2026-10-14-v0.2/summary.json

A flag on its own is a lead, not an error. Most flags on a/R★ and transit depth, for example, come from the Kepler pipeline tables, where those quantities are free fit parameters and limb darkening makes depths larger than (Rp/R★)². So we also grouped results by reference. When most rows from one paper share the same offset, the cause is the catalog entry, not scatter, and if the offset equals a known unit factor it points to the unit.

What we found

For four references, the stored stellar density divided by 1.4098 M★/R★³ has a median close to 0.709, which is exactly what solar units (ρ/ρ☉) produce, since the Sun's mean density is 1.4098 g/cm³. Multiplying their densities by 1.4098 brings the median back to 1.

References whose stellar densities are in solar units

ReferenceTableRowsMedian ratioRows within 10% of medianMedian after ×1.4098
Fulton & Petigura 2018ps1,4140.716392.5%1.01
Torres et al. 2017ps220.702590.9%0.9904
Mann et al. 2017ps190.709968.4%1.001
Mann et al. 2017pscomppars70.718471.4%1.013
MacDonald et al. 2016ps110.7056100%0.9948
MacDonald et al. 2016pscomppars50.7056100%0.9948
Ratio = stored st_dens ÷ (1.4098 g/cm³ · M★/R★³) from the same row. Source: results/2026-10-14-v0.2/systematics.csv.

We then read each source paper. Fulton & Petigura 2018 and MacDonald et al. 2016 print solar-unit values under a g/cm³ label in their own tables, so the Archive copied the label faithfully. Torres et al. 2017 and Mann et al. 2017 label their densities ρ☉ correctly, so those look like slips when the data were ingested. The composite table, which many users treat as the parameters, takes 1,049 of its density values from Fulton & Petigura 2018. For Kepler-80 the composite density is 2.33 "g/cm³"; the paper's own mass and radius give 3.30 g/cm³.

This matters because comparing a star's density from a transit fit with an independent density is a standard way to measure orbital eccentricity. A density 1.41 times too low mimics an eccentricity of about 0.11 for an orbit that may be circular, and any population study that pulls these densities inherits a 41% bias.

Following the flags further turned up other kinds of problem, each checked against its source paper, erratum or catalog.

Kinds of error found and checked against sources

KindWhere the error isCases checked
A reference's stellar densities in solar units2 in journal tables, 2 at ingestion4 references, about 1,470 ps rows and 1,069 composite values
One value stored in the wrong unitingestion5 rows
A printed value contradicting the same paper's other numbersthe paper12 papers, 20 planets
A different quantity stored in the columningestionKepler-419 b and c
A published erratum not appliedArchive not updatedHD 48948 b, c and d
A placeholder stored as a measurementthe paper's input tablelog g = 4.50 for KELT-2 A, KELT-7, WASP-33
A linear luminosity stored in the log₁₀ column4 at ingestion, 3 in the paper's label7 papers, 9 stars, 12 rows
Two stars' parameters exchangedingestionHD 97658 and HD 192310
Selected rows from the summary table in FINDINGS.md (eleven kinds in all).

Two of these errors have reached a published catalog of habitable-zone planets. In one, a linear luminosity read as a logarithm made a K dwarf look 1.55 times as bright as the Sun instead of 0.19 times, which placed its planet HIP 57274 d partly in the habitable zone when, with the paper's own luminosity, its orbit lies beyond the outer edge.

What it means

Large catalogs are mostly consistent, and most of our flags turned out to be physics, pipeline conventions or mixed sources rather than mistakes. But a check run once across every row finds errors that no single user would notice, and grouping by reference turns scattered mismatches into one clear fix. The density correction is a single multiplication for a known set of rows.

More research