The Atmospheric Composition of Sub-Neptune K2-18 b and Implications for its Formation

TL;DR

This study critically reexamines the JWST transmission spectrum of the temperate sub-Neptune K2-18 b to test the robustness of claimed molecular detections and quantify how data-reduction and retrieval choices bias atmospheric inferences. By generating 12 spectrum variants and performing TauREx 3.1 retrievals across multiple configurations, the authors find robust CH$_4$ detection (≈4σ) but only tentative, model-dependent CO$_2$ and inconclusive DMS; spot-correction offsets and retrieval complexity significantly influence inferred mean molecular weight and composition. The results consistently point to a hydrogen-rich, low-MMW atmosphere with elevated C/O, and the inferred carbon and oxygen abundances align with Inside-Out Planet Formation (IOPF) expectations for formation interior to the soot line. The work underscores the importance of using multiple reductions and model configurations to assess the robustness of atmospheric inferences from JWST data and provides guidelines for interpreting sub-Neptune atmospheres in the context of planet formation, with implications for future missions like Ariel.

Abstract

Unlocking the atmospheres of sub-Neptunes is among JWST's major achievements, yet such observations demand complex analyses that strongly affect interpretations. We present an independent reanalysis of the original JWST transmission spectrum of K2-18 b, to assess the robustness of previously claimed detections, explore the parameter space, and implications for its formation. The observations were reduced using a combination of public and customized pipelines producing a total of 12 different versions of the transmission spectrum by varying: spectral binning, limb-darkening, and a novel correction for the occulted stellar spot. We then performed atmospheric retrievals using TauREx 3, comparing models of varying complexity, robustly detecting CH$_4$ (3-4$σ$) across all configurations. The evidence for CO$_2$ is weaker and highly model-dependent. The tentative detection of dimethyl sulphide (DMS) vanishes in our most comprehensive retrieval models. We find that correcting the stellar spot in the NIRISS transit is a critical step, introducing a uniform offset that primarily drives the inference of a lower mean molecular weight atmosphere. Furthermore, the assumed complexity of the retrieval model itself introduces significant biases; including more molecules systematically increases the retrieved CH$_4$ abundance and atmospheric mean molecular weight, even for species without spectral features. The data are consistent with a hydrogen-rich atmosphere with an elevated O and an even more elevated C abundance, leading to a super-solar C/O. We show that the physical properties of the system planets K2-18 c, and K2-18 b are consistent with those expected by the in situ formation theory of Inside-Out Planet Formation (IOPF), interior to the carbon "soot" line, where an elevated C/O ratio of a primordial atmosphere is expected to be inherited from the protoplanetary disk.

Figures (10)

• Figure 1: White light curves for NIRISS NIRSpec NRS1 and NIRSpec NRS2 with best fit models. The blue line in the left-most panel residuals (NIRISS) indicates the resulting profile from the subtraction of the spot and spotless models. The red dashed line is the spotless model. The coloured points are 10 minute binned light curves.
• Figure 2: Comparison between different reduction modified AICs in the top panel and BICs in the lower panel (all reduction names and their characteristics are described in Section \ref{['sec:spec_lc_fits']}) taking the Blc-EQ-S (binned before fitting light curves - empirical limb darkening - with spot correction) reduction as baseline. The median of the modified AIC or BIC is shown in the legend. For those cases where the NIRSpec data is shared between reductions, only the median for the NIRISS data is shown. Although there are peaks below, the majority of the points lie above the 0 $\Delta \mathrm{\,AIC_{mod}}$ and $\Delta \mathrm{\,BIC}$ lines.
• Figure 3: Fiducial reduction and retrieval which includes: spot correction, 2 coefficient empirical limb darkening law, light curves binned to a resolution of 100 before the fit (Blc-EQ-S). The data (lighter shaded points) are binned to a resolution of $R\sim15$ (darker shaded points) to highlight the spectral features. The bottom panel shows the residuals, at $R\sim100$, which is the resolution at which the retrieval was done.
• Figure 4: Left: Main molecular abundances obtained from different spectral extractions, using simplified (blue) and comprehensive (cyan) retrieval setups. Note that CO$_2$ was not included in the Blc-PhQ-S and Blc-PhQ-S$'$ simplified retrievals, as its significance was below 2$\sigma$. Similarly, DMS was not included in the majority of simplified retrievals. Right: Retrieved mean molecular weights, temperature at a pressure of 10 mbar, and reduced chi-squared values.
• Figure 5: Posterior distribution correlations between mean molecular weight (MMW) and temperature at 10 mbar. The left-most panel shows all retrievals which include the full suite of molecules. The middle panel shows the same but only including CO$_2$, CH$_4$ and DMS with blue and red indicating no spot and spot correction reductions respectively. The right-most panel shows all retrievals for the fiducial spectrum (Blc-EQ-S, binned before fitting light curves - empirical limb darkening - with spot correction) with blue indicating those with all molecules and red those with only CO$_2$ and CH$_4$.