A Systematic Search for Trace Molecules in the Atmosphere of Exoplanet K2-18 b

Gregory J. Cooke; Lorenzo Pica-Ciamarra; Martin Binet; Nikku Madhusudhan; Savvas Constantinou

arxiv: 2505.10539 · v2 · submitted 2025-05-15 · 🌌 astro-ph.EP

A Systematic Search for Trace Molecules in the Atmosphere of Exoplanet K2-18 b

Lorenzo Pica-Ciamarra , Nikku Madhusudhan , Gregory J. Cooke , Savvas Constantinou , Martin Binet This is my paper

Pith reviewed 2026-05-22 14:23 UTC · model grok-4.3

classification 🌌 astro-ph.EP

keywords K2-18 bexoplanet atmospheredimethyl sulfideJWST transmission spectroscopytrace gasesbiosignaturesBayesian retrieval

0 comments

The pith

Only dimethyl sulfide shows consistent statistical preference when added to models of K2-18 b's JWST transmission spectrum.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper conducts an agnostic search across 661 trace molecules to test whether any improve the fit to the observed JWST spectra of the temperate sub-Neptune K2-18 b beyond a baseline model containing only methane and carbon dioxide. A sympathetic reader cares because the planet lies in the habitable zone and prior work already suggested possible biosignatures, so identifying which additional gases are statistically favored narrows the list of candidates that deserve follow-up. The authors evaluate each molecule using Bayes factors on the mid-infrared dataset, then cross-check promising candidates on the near-infrared dataset while also considering known production pathways. Only dimethyl sulfide meets the ln B ≥ 2.0 threshold consistently across the independent datasets, although the near-infrared result remains sensitive to how detector offsets are handled.

Core claim

The authors tested 661 molecules spanning biotic, abiotic, and anthropogenic gases against JWST mid-infrared and near-infrared transmission spectra of K2-18 b. They report that only dimethyl sulfide yields Bayes factors ln B ≥ 2.0 on both datasets when added to a model that already includes methane and carbon dioxide, with the near-infrared preference depending on detector-offset choices. A handful of other species produce comparable fits to limited subsets of the data or with only certain retrieval codes, yet lack clear plausible sources of production in the atmosphere.

What carries the argument

Bayesian atmospheric retrievals that compute the Bayes factor for each added trace-gas model relative to the CH4-plus-CO2 baseline, applied separately to the mid-infrared and near-infrared JWST datasets.

If this is right

Additional observations are required to confirm whether dimethyl sulfide is truly present or whether the current preference arises from residual data artifacts.
Theoretical modeling of plausible abiotic and biotic production pathways for dimethyl sulfide under K2-18 b conditions becomes necessary.
The same broad-molecule search strategy can be applied to other temperate sub-Neptunes to identify unexpected trace gases without prior assumptions.
Careful treatment of detector offsets and cross-instrument consistency checks will be essential for future JWST analyses of similar planets.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If dimethyl sulfide is ultimately confirmed, targeted laboratory studies of its photochemical stability and surface or atmospheric sources on sub-Neptunes would help distinguish biological from abiotic origins.
The finding that most of the 661 tested molecules produce no improvement underscores the value of data-driven, rather than hypothesis-driven, searches for atmospheric constituents on habitable-zone worlds.
Extending the analysis to include temperature-dependent line lists or cloud parameterizations could test whether the current preference for dimethyl sulfide persists under more flexible atmospheric assumptions.

Load-bearing premise

Any statistical improvement from adding a trace gas reflects the actual presence of that gas rather than unaccounted systematics, incomplete line lists, or choices in aligning data from different detectors.

What would settle it

New JWST or ground-based observations that achieve higher signal-to-noise or use independent reduction pipelines and show no ln B preference for dimethyl sulfide above the baseline model would falsify the central result.

read the original abstract

The first transmission spectrum of the habitable-zone sub-Neptune K2-18 b with JWST has opened a new avenue for atmospheric characterisation of temperate low-mass exoplanets. The observations led to inferences of methane and carbon dioxide, as well as of dimethyl sulfide (DMS) and/or dimethyl disulfide (DMDS), both potential biosignatures. In the present work we conduct a broad and agnostic search for other chemical species in the atmosphere of K2-18 b. Our exploration includes 661 molecules, spanning a wide range of trace gases, including biotic, abiotic, and anthropogenic gases on Earth. We investigate possible preference for any of these gases, compared to a model only including the previously-detected CH$_4$ and CO$_2$, using three metrics: (a) preference in the JWST mid-infrared (MIR) spectrum, (b) preference in the JWST near-infrared spectrum, for species preferred in MIR and (c) plausible sources of production. We find that only DMS consistently results in Bayes factors $\ln B \geq 2.0$ across the datasets considered independently, though in the near-infrared this depends on detector offsets, as previously reported. The threshold of $\ln B \geq 2.0$ is motivated by the conventional threshold of $\ln B \geq 2.5$ for moderate preference, allowing for an empirical uncertainty of 0.5. A few other gases also provide comparable fits to a subset of the data or only with some of the retrieval codes used, but with limited known plausible sources. Our study highlights the need for further observations to distinguish between possible trace gases in the atmosphere of K2-18 b and theoretical work to establish their plausible sources.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

DMS is the only molecule that meets the Bayes factor threshold across both datasets, but the NIR result depends on detector offset choices.

read the letter

The punchline is that only DMS shows a Bayes factor above their threshold across the MIR and NIR JWST data for K2-18 b, but the NIR result depends on the detector offset values used in the reduction. The paper does a good job of running a systematic search across 661 molecules using standard atmospheric retrieval methods on the existing JWST transmission spectra. This is new in its scale for this target and helps rule out many other trace gases that might have been plausible. They compare models with each added molecule against a baseline with just CH4 and CO2, and they check for plausible sources of the gases that perform well. That approach is straightforward and covers a lot of ground without overclaiming. The soft spots come from the data processing sensitivities. The NIR preference for DMS is not fully independent because it relies on specific choices for aligning the detectors, and different but reasonable offset values could drop the Bayes factor below 2.0. They adjusted the conventional ln B threshold of 2.5 down to 2.0 to allow for some uncertainty, which is reasonable but makes the evidence a bit softer. The analysis also takes the statistical preference at face value as indicating the molecule's presence, without fully exploring how unaccounted systematics or line list incompleteness might produce similar signals. This paper is for astronomers and planetary scientists who are tracking the JWST observations of K2-18 b and the ongoing discussion about possible atmospheric composition and biosignatures. A reader looking for a broad check on alternative molecules would get value from the results. It deserves a serious referee because the systematic nature adds to the literature, even with the noted dependencies on processing choices. I recommend sending it for peer review, with reviewers asked to focus on the robustness to offset variations and the justification for the Bayes factor cutoff.

Referee Report

2 major / 2 minor

Summary. The manuscript reports a systematic, agnostic search for trace molecules in the transmission spectrum of the habitable-zone sub-Neptune K2-18 b using JWST MIR and NIR data. After fixing CH4 and CO2, the authors test 661 additional species via Bayesian atmospheric retrievals and evaluate preference with Bayes factors. They conclude that only DMS yields ln B ≥ 2.0 in both datasets considered independently (with the NIR result depending on detector offsets), while a few other gases fit subsets of the data or only with certain retrieval codes but lack clear plausible sources. The work calls for additional observations and theoretical studies on production pathways.

Significance. If the reported preference for DMS survives quantitative tests of data-reduction choices, the result would be significant for biosignature searches on temperate sub-Neptunes, as DMS is a potential biological indicator. The broad exploration of 661 molecules constitutes a genuine strength, providing a more comprehensive search than targeted studies. The quantitative use of Bayes factors supplies a clear metric, though the adjusted threshold and offset sensitivity limit the strength of the cross-dataset claim.

major comments (2)

[Abstract] Abstract: the central claim that 'only DMS consistently results in Bayes factors ln B ≥ 2.0 across the datasets considered independently' is qualified by the statement that the NIR result 'depends on detector offsets, as previously reported.' Because the manuscript acknowledges this dependence, the cross-dataset consistency is not fully independent. Alternative but plausible offset values within the uncertainty of the data reduction could reduce the NIR ln B below 2.0 while the MIR value remains above threshold, undermining the attribution of the preference to the molecule rather than to alignment systematics. A quantitative sensitivity analysis showing the range of offsets for which ln B stays ≥ 2.0 in NIR is required to support the claim.
[§4 (Results)] §4 (Results) and associated tables: the choice of ln B ≥ 2.0 (motivated by allowing an empirical uncertainty of 0.5 around the conventional 2.5 threshold) combined with the documented sensitivity of the NIR DMS result to detector offsets introduces moderate uncertainty in the strength of the claimed preference. The manuscript should report the actual ln B values obtained for DMS under the nominal and alternative offset choices so readers can assess how close the NIR result sits to the threshold.

minor comments (2)

[Methods] The methods section should explicitly state the prior ranges adopted for the molecular volume mixing ratios of the 661 species and confirm that the same priors were used across all retrieval codes.
Figure captions should clearly distinguish MIR versus NIR datasets and label which retrieval code produced each posterior distribution shown.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments, which help clarify the robustness of our claims. We address each major comment below and will revise the manuscript to incorporate additional quantitative details on the NIR offset sensitivity and reported Bayes factors.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that 'only DMS consistently results in Bayes factors ln B ≥ 2.0 across the datasets considered independently' is qualified by the statement that the NIR result 'depends on detector offsets, as previously reported.' Because the manuscript acknowledges this dependence, the cross-dataset consistency is not fully independent. Alternative but plausible offset values within the uncertainty of the data reduction could reduce the NIR ln B below 2.0 while the MIR value remains above threshold, undermining the attribution of the preference to the molecule rather than to alignment systematics. A quantitative sensitivity analysis showing the range of offsets for which ln B stays ≥ 2.0 in NIR is required to support the claim.

Authors: We agree that the NIR result's dependence on detector offsets qualifies the degree of independence between the two datasets. To strengthen the manuscript, we will add a quantitative sensitivity analysis in the revised Section 4. This will vary the NIR detector offsets across their reported uncertainty range and identify the specific interval of offset values for which the DMS ln B remains ≥ 2.0. The results will be presented in the text and, if space permits, a supplementary figure, allowing readers to assess the robustness directly. revision: yes
Referee: [§4 (Results)] §4 (Results) and associated tables: the choice of ln B ≥ 2.0 (motivated by allowing an empirical uncertainty of 0.5 around the conventional 2.5 threshold) combined with the documented sensitivity of the NIR DMS result to detector offsets introduces moderate uncertainty in the strength of the claimed preference. The manuscript should report the actual ln B values obtained for DMS under the nominal and alternative offset choices so readers can assess how close the NIR result sits to the threshold.

Authors: We accept that explicit numerical values would improve transparency. In the revised manuscript we will report the precise ln B values obtained for DMS under the nominal offset choice as well as under the alternative plausible offset values tested in the NIR retrievals. These numbers will be stated in the main text of Section 4 and added to the relevant summary table of retrieval metrics, enabling readers to judge the margin relative to the adopted threshold. revision: yes

Circularity Check

0 steps flagged

Standard Bayesian retrieval on external data; minor self-citation on offsets not load-bearing

full rationale

The paper applies conventional Bayesian atmospheric retrieval and model comparison (Bayes factors) to external JWST MIR and NIR transmission spectra of K2-18 b. The reported preference for DMS (ln B ≥ 2.0 across datasets) is computed directly from fits to the observational data rather than being defined in terms of the result itself. The caveat that the NIR result 'depends on detector offsets, as previously reported' is a minor self-citation that does not carry the central claim; the manuscript explicitly qualifies the consistency and grounds the analysis in independent data. No self-definitional steps, fitted inputs renamed as predictions, ansatz smuggling, or uniqueness theorems imported from the authors' prior work appear in the derivation chain. The analysis remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The claim rests on the validity of Bayesian model comparison between a baseline retrieval and single-molecule extensions, which depends on standard assumptions in exoplanet atmospheric modeling and the quality of the input spectra.

free parameters (1)

Molecular volume mixing ratios
Abundances of each tested trace gas are free parameters fitted during retrieval.

axioms (1)

domain assumption Standard 1D plane-parallel radiative transfer and equilibrium or parameterized chemistry assumptions hold for the atmosphere.
Retrieval codes rely on these background modeling choices to compute spectra.

pith-pipeline@v0.9.0 · 5871 in / 1304 out tokens · 38845 ms · 2026-05-22T14:23:07.575346+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Transmission Spectrum of the Benchmark Temperate Exo-Neptune TOI-1231 b
astro-ph.EP 2026-07 unverdicted novelty 7.0

The JWST transmission spectrum of TOI-1231 b shows strong CH4 detection and moderate CO2 evidence, consistent with a deep H2-rich atmosphere and no distinct surface.