JWST Excludes Exomoons Down to 0.1 Earth Radii Around a Rocky, Temperate Exoplanet

Abstract: To date, even with JWST, it has not been possible to test for exomoons as small as the Moon. Only two reported searches for exomoons around bound planets have been attempted with JWST, both of which relied on a single JWST transit. We suggest that the below expectation sensitivity to date is, in part, a product of the considerable intrinsic flexibility of the planet+moon transit model when confronted with a single event, and the presence of unanticipated systematic noise. To test this, we present a search around the rocky, temperate-zone exoplanet LP 890-9c using twelve JWST transits. We find no evidence for an exomoon but exclude $0.1R_{\oplus}$ moons to 95% confidence across the entire Hill region, representing by far the most sensitive search to date. Our limits exclude analogs to many real Solar System moons, such as Europa, Rhea and Umbriel. However, we emphasize that the close-in orbit of LP 890-9c (0.04au) would make any moons larger than ${\sim}0.1R_{\oplus}$ unlikely to survive for many Gyr due to tides. Regardless, our study firmly establishes that JWST can indeed probe down to remarkably small moons. Further, we find that even a single transit can deliver impressive limits, with the exception of one epoch that is contaminated by red noise. However, by pairing it with just one cleaner epoch, the sensitivity recovers to a level indistinguishable from any other pairing, which bodes well for a potential second JWST transit of Kepler-167e.

• The paper demonstrates that JWST/NIRSpec observations can detect or constrain satellites around temperate, rocky exoplanets to down to moon radii of approximately 0.1 RADARIUS<|epilogue|>Earth
• N-resistant methods, including multi-transit and photometric performance techniques was used The highly precision analysis ruled out the existence of satellites around LP 890-9 c
• Coherent multi-epoch photodynamic modelling demonstrates enhanced robustness to formally confirm Transit clouds and, Scope to explore nearby stars

Observational objective and target selection

The paper presents a twelve-transit JWST/NIRSpec search for exomoons orbiting LP 890-9 c, a $1.37\,R_{\oplus}$ planet receiving approximately $90\%$ of Earth’s instellation (2609.05301). The central methodological premise is that previous JWST exomoon searches were limited by the combination of single-transit model flexibility and time-correlated systematics. A planet–moon model confronted with one transit has substantial freedom to reinterpret unmodeled structure as a satellite signal, while red noise can broaden the marginalized posterior on the moon radius. A sequence of transits instead requires the putative satellite to maintain a single dynamically coherent orbit across epochs.

LP 890-9 c is therefore selected as an observationally favorable target. Its transits are comparatively short, the host is quiescent, and twelve visits were obtained in a single JWST program. The observations used NIRSpec PRISM in Bright Object Time Series mode, with NRSRAPID readout, the CLEAR/PRISM configuration, a SUB512 subarray, five groups per integration, and an effective integration time of $0.903$ seconds. The visits span August 2025 to February 2026 and sample twelve planetary transits.

The target is not necessarily dynamically favorable for retaining a large moon. At an orbital separation of approximately $0.04$ au from its star, tidal evolution is expected to remove or destroy substantial satellites over gigayear timescales. This distinction is important: the analysis is primarily an instrumental and methodological demonstration of JWST sensitivity, rather than a high-prior-probability search for a long-lived satellite around LP 890-9 c.

Reduction strategy and photometric performance

The authors reduce the data independently with Eureka! and ExoTiC-JEDI. Both reductions omit reference-pixel correction, which is unavailable for the SUB512 subarray, and jump detection, whose false-positive rate is problematic for five-group integrations. Both also perform group-level column-by-column removal of NIRSpec $1/f$ noise before ramp fitting. Relative spectrophotometry is obtained without applying flat-field, photometric, or wavelength-zero-point corrections, thereby avoiding calibration operations that are unnecessary for white-light transit depths.

The Eureka! reduction uses optimal extraction, with the extraction aperture, background exclusion, detector window, outlier threshold, and white-light wavelength range optimized on the first visit. The adopted aperture has a three-pixel half-width, and the white-light curve integrates $0.6$–$5.3\,\mu{\rm m}$. The optimal aperture produces a median absolute deviation of $393$ ppm per $0.9$-second integration on the optimization visit. The ExoTiC-JEDI analysis uses a distinct implementation while retaining the same high-level detector choices. In contrast to Eureka!, box extraction outperforms optimal extraction for this faint M6V host, reflecting instability in the estimated spatial profile.

The two reductions are compared only after identical binning and detrending. At 30-second cadence, ExoTiC-JEDI achieves a median out-of-transit RMS of $117$ ppm and a median formal uncertainty of $90\%$0 ppm, compared with $90\%$1 ppm and $90\%$2 ppm for Eureka!. The approximately $90\%$3 precision advantage of ExoTiC-JEDI persists across nearly all visits.

Figure 1: Twelve LP 890-9 c transits reduced with Eureka!, showing unbinned and 30-second-binned white-light photometry with per-epoch polynomial baselines.

The data are not purely photon-limited. Residual RMS scales approximately as $90\%$4 with binning factor $90\%$5 under the white-jitter likelihood, rather than the $90\%$6 behavior expected for independent Gaussian noise. Introducing a GP substantially flattens the RMS–bin-size relation. This establishes that correlated noise is present, but its characteristic scale is generally short compared with the transit duration.

Figure 2: RMS as a function of bin size for the four reduction–likelihood combinations, illustrating sub-photon-limited binning slopes under GP treatment.

Photodynamical and noise models

The astrophysical signal is modeled with LUNA, a full planet–moon photodynamical model incorporating planetary and satellite transits, barycentric motion, dynamical transit timing effects, transit-duration effects, and instrumental trends. The base model includes planetary radius, stellar and planetary density, impact parameter, limb-darkening parameters, planetary ephemeris, satellite semimajor axis, phase, inclination, node, mass ratio, and radius ratio.

A key modification is the inclusion of an independent transit-time offset $90\%$7 for each of the twelve epochs. These offsets are necessary because the data exhibit strong TTVs that are not plausibly produced by the candidate moon. The offsets move the planet–moon barycenter along the planetary orbit while preserving one globally coherent satellite orbit. This construction prevents the model from resetting the moon’s phase independently at every transit, which would make the satellite model excessively flexible. However, it introduces a substantial degeneracy between the epoch offsets and the satellite-to-planet mass ratio. Consequently, the inferred satellite mass and density are treated as nuisance quantities rather than physically informative measurements.

The baseline for each visit is a multiplicative polynomial in centered and scaled time. Polynomial order is selected independently for each epoch using ten-fold contiguous cross-validation on the out-of-transit data. The selected orders range from one to four, with most visits favoring quadratic or lower-order trends. This procedure explicitly avoids choosing a uniform baseline order or selecting it by visual inspection.

Two noise models are considered. The first adds an epoch-independent white jitter term in quadrature to the empirical photometric uncertainties. The second uses a shared GP amplitude and correlation scale for all visits, with a stochastically driven simple harmonic oscillator kernel. Polynomial coefficients are profiled analytically by generalized least squares. The SHO covariance belongs to the celerite family, allowing exact linear-time factorization and solves rather than the usual cubic-time GP computation (Foreman-Mackey et al., 2017). This computational structure is essential for the MultiNest analyses, which use 4000 live points and require a large number of likelihood evaluations.

The moon inclination is parameterized through an unbounded latent variable that produces a uniform prior in $90\%$8, preserving isotropic orientations while avoiding hard sampling boundaries. The authors also impose a Rayleigh-statistic penalty against artificial concentration of the moon’s orbital phase across the twelve epochs. This is intended to suppress fine-tuned phase configurations that can mimic repeated instrumental distortions, although it constitutes an informative modeling choice and therefore contributes to the prior structure of the upper limits.

Transit timing variations are not moon-like

Before fitting planet–moon models, the authors fit a planet-only model with independent epoch timing offsets. The result is a highly significant TTV signal with an amplitude of $90\%$9 seconds. A Lomb–Scargle periodogram yields $0.903$0 for all four reduction–likelihood combinations, with a dominant feature near twelve cycles. The signal is more complex than a sinusoid, but its timescale is outside the expected exomoon TTV corridor.

Figure 3: The twelve measured TTVs and their Lomb–Scargle periodogram, showing a coherent approximately 14-second timing signal inconsistent with a typical exomoon frequency pattern.

The amplitude also disfavors a satellite interpretation. Given the expected planetary mass of approximately $0.903$1 and Hill radius of roughly $0.903$2, a moon at the Hill radius would require a mass ratio of approximately $0.903$3 to generate a 14-second TTV. Any moon at a smaller semimajor axis would need to be more massive. Thus, the observed timing signal would require an unusually massive satellite, close to the maximum dynamically available lever arm.

The authors test this possibility with transit origami, which conditions the expected moon transit phases on a proposed satellite mass ratio and searches for the corresponding photometric dips. No candidate yields a significant signal. After accounting for red-noise inflation, the maximum improvement is $0.903$4 in either reduction. The implied moon radii are also generally below the physically expected mass–radius relation, often requiring implausibly high densities.

Figure 4: Transit-origami spectra searching for moon transits at the phases required to reproduce the observed TTVs; no significant dip is detected.

The analysis therefore treats the TTVs as evidence for an external perturber, possibly LP 890-9 d or another planet, rather than as the primary signature of a moon. The individual $0.903$5 parameters are retained in all subsequent moon fits to prevent the unrelated timing signal from biasing satellite inference.

Null detection and radius constraints

The main comparison is between a planet–moon model $0.903$6 and a zero-radius satellite model $0.903$7. The four combinations of reduction and likelihood all disfavor the moon model in marginal likelihood. The reported log Bayes factors $0.903$8 are:

Reduction and likelihood	Log Bayes factor
Eureka! + jitter	$0.903$9
Eureka! + GP	$0.04$0
ExoTiC-JEDI + jitter	$0.04$1
ExoTiC-JEDI + GP	$0.04$2

The Savage–Dickey calculations provide consistent, though not identical, evidence against a nonzero moon radius. The posterior radius estimates remain close to zero:

Reduction and likelihood	Posterior $0.04$3	$0.04$4 upper limit
Eureka! + jitter	$0.04$5	$0.04$6
Eureka! + GP	$0.04$7	$0.04$8
ExoTiC-JEDI + jitter	$0.04$9	$1/f$0
ExoTiC-JEDI + GP	$1/f$1	$1/f$2

At $1/f$3, the limits range from $1/f$4 to $1/f$5. Thus, the principal numerical claim is that moons near $1/f$6 are excluded at approximately $1/f$7 confidence across the allowed orbital region. The consistency among independent reductions and fundamentally different noise treatments is important because it reduces the likelihood that the constraint is an artifact of one extraction pipeline or one covariance assumption.

Figure 5: Marginalized moon-radius posteriors for the four analyses, all consistent with a null satellite radius.

The semimajor-axis grid shows that the limit remains near $1/f$8 from approximately $1/f$9 to $0.6$0. Sensitivity degrades modestly at wide separations because the finite observing baselines do not always capture the full range of possible moon transit phases. The expected Hill radius is approximately $0.6$1, so the search covers the dynamically relevant region with some extension beyond it.

The limits exclude analogs of eight Solar System moons: Europa, Ganymede, Io, Rhea, Titan, Titania, Triton, and Umbriel. This is a strong observational statement, but it should be interpreted as a constraint on transit-compatible radius and orbital configurations, not as evidence that such bodies could survive around LP 890-9 c. In this system, tidal evolution independently makes many of these satellites unlikely.

Figure 6: Two-sigma upper limits across satellite semimajor axis, compared with Solar System moons, the Hill radius, the Roche limit, and the radius required for total solar eclipses.

The analysis also tests whether LP 890-9 c could host a moon producing total stellar eclipses. The eclipse condition requires the satellite’s angular radius as seen from the planet to match the star’s angular radius. The upper limits intersect this condition near $0.6$2; at larger separations, eclipse-producing satellites become increasingly inconsistent with the data. The paper therefore concludes that total eclipses are unlikely for LP 890-9 c, subject to the adopted stellar, planetary, and satellite-radius assumptions.

Sensitivity to the number of transits

The epoch-subset analysis isolates the role of repeated observations. With increasing transit count, the moon-radius upper limit improves approximately as $0.6$3, as expected for partially independent measurements. Individual epochs generally provide strong constraints: excluding the contaminated first visit, the median single-transit $0.6$4 limit is approximately $0.6$5. Observation #51 is a clear outlier, producing a limit of $0.6$6 when analyzed alone because of its excess red noise.

Figure 7: Moon-radius upper limits for progressively larger epoch sets, individual transits, and representative two-transit combinations.

The result most relevant to observing strategy is that a contaminated epoch does not necessarily compromise a multi-epoch search. Pairing observation #51 with another transit produces limits of $0.6$7 and $0.6$8, comparable to the $0.6$9 distribution from the other representative two-epoch combinations. The implication is specific: a second clean transit can substantially reduce the degeneracy created by a red-noise-dominated first transit because a moon model must explain both events with one Keplerian orbit. The data do not establish whether two heavily contaminated epochs would behave similarly.

Limitations and open questions

The upper limits depend on the adopted photodynamical model, prior structure, and treatment of correlated noise. The circular-orbit assumption excludes eccentric satellite configurations, although close-in satellites are generally expected to circularize efficiently. The Rayleigh phase penalty suppresses phase clustering that the authors regard as fine-tuned, but it is an informative prior rather than a likelihood-derived constraint. The planetary density prior is based on a probabilistic mass–radius relation rather than a dynamical mass measurement, and the assumed Hill radius therefore remains uncertain.

The treatment of epoch timing offsets is necessary but weakens constraints on satellite mass. Because the twelve independent $5.3\,\mu{\rm m}$0 terms can absorb barycentric TTVs, the posterior on $5.3\,\mu{\rm m}$1 and the inferred satellite density cannot be interpreted as a measurement. The Bayes-factor comparison is also described as optimistic because the zero-radius model does not remove all other satellite parameters. Finally, the correlated-noise GP is shared across epochs through two hyperparameters, while the polynomial baselines remain visit-specific. Whether this is the uniquely appropriate hierarchical covariance structure is not demonstrated by the data.

The principal unresolved astrophysical question is the origin of the approximately 14-second TTV signal. If it is produced by an additional planet, a joint dynamical model could refine the ephemeris and perhaps alter the treatment of the epoch offsets. The present analysis deliberately avoids using that information because the exomoon question only requires the timing variations to be modeled as nuisance structure.

Conclusion

The paper establishes that a multi-transit JWST/NIRSpec campaign can constrain exomoons around a rocky, temperate exoplanet to radii of order $5.3\,\mu{\rm m}$2. Twelve LP 890-9 c transits, independently reduced with Eureka! and ExoTiC-JEDI and analyzed with both white-jitter and GP likelihoods, yield no evidence for a moon and exclude $5.3\,\mu{\rm m}$3 satellites at approximately $5.3\,\mu{\rm m}$4 confidence across the Hill region (2609.05301). The analysis also demonstrates that coherent multi-epoch photodynamical modeling is more robust to isolated red-noise contamination than single-transit inference. These constraints are observationally stringent even though tidal evolution makes large, long-lived moons around LP 890-9 c intrinsically improbable.