Competing chemical signatures in the atmosphere of TOI-270 d: Inference of sulfur and carbon chemistry

TL;DR

This paper reanalyzes TOI-270 d’s JWST transit spectra using independent data reduction and the BeAR retrieval framework, emphasizing the critical role of data resolution and the instrument line spread function in atmospheric inference. The results robustly detect CH4 and CO2 in a hydrogen-dominated, high-metallicity atmosphere, with CS2 emerging as a strong sulfur-bearing absorber and water appearing depleted. However, there is a significant degeneracy between sulfur-chemistry (CS2/CS/H2CS) and methyl-chemistry scenarios (CH3Cl, CH3F), as well as potential degeneracy with H2CS, driven by overlapping opacities below 4 μm; Bayesian evidence is highly sensitive to the data resolution and LSF treatment. The study cautions against relying on native-resolution spectra without LSF accounting and suggests that additional photochemical modeling and extended wavelength observations (e.g., with MIRI) are needed to resolve the chemical ambiguity and to place TOI-270 d in the context of sub-Neptune atmospheric diversity.

Abstract

Recent JWST measurements allow access to the near-infrared spectrum of the sub-Neptune TOI-270 d, for which two different interpretations, a high-metallicity miscible envelope and a lower metallicity hycean world, are currently in conflict. Here, we reanalyze the published data and reproduce previously retrieved molecular abundances based on an independent data reduction and a different retrieval framework. The aim of this study is to refine the understanding of TOI-270 d and highlight considerations for JWST data analysis. Additionally, we test the impact of data resolution on atmospheric retrieval calculations. We reduce one JWST NIRSpec G395H and one NIRISS SOSS GR700XD transit dataset using the Eureka! pipeline and a custom MCMC-based light curve fitting algorithm at the instruments' native resolutions. The atmospheric composition is estimated with the updated BeAR retrieval code across a grid of retrieval setups and spectral resolutions. Our transit spectrum is consistent with previous studies, except at the red end of the NIRISS data. Our retrievals support a higher mean molecular weight atmosphere for TOI-270 d. We provide refined abundance constraints and find statistically favored model extensions indicating either sulfur-rich chemistry with species such as CS2, CS, and H2CS, or the possible presence of CH3Cl or CH3F. However, Bayesian inference cannot distinguish between these scenarios due to similar opacities below 4 microns. Our analysis reinforces TOI-270 d as a highly interesting warm sub-Neptune for atmospheric studies, with a complex chemistry in a cloud-free upper atmosphere. However, its exact nature remains uncertain and warrants further detailed photochemical modeling and observations.

Figures (20)

• Figure 1: Broadband light curve fits for both JWST observations. Left: Combined NIRSpec NRS1 and NRS2 broadband light curve fit of the overlapping transits of TOI-270 b and TOI-270 d. JWST ETC calculations predict noise levels of $\approx85$ ppm, which is roughly 12% below the standard deviation of our residuals. Right: NIRISS first order broadband light curve fit of the transit of TOI-270 d. From S/N calculations with the ETC one would expect a noise level of $\approx69$ ppm, much lower than we find to be the case for the broadband light curve. This could partially be caused by stellar activity, see \ref{['app:niriss_order2']}.
• Figure 2: Transit spectrum of TOI-270 d binned to R $\approx50$ for comparison to benneke_jwst_2024 and holmberg_possible_2024. We note that the spectrum from this work and holmberg_possible_2024 were originally reduced and fit at the native instrument resolution. We see excellent agreement over most of the probed wavelength range, with only the $\geq2.5\,µm$ portion of our NIRISS spectrum deviating significantly from the other reduction and diverging from the shortest wavelength NRS1 data. We test for the impact of these data in \ref{['app:niriss_beyond2.5']}. In general, our uncertainties are slightly larger than in holmberg_possible_2024 but significantly smaller for the NIRSpec data when compared to benneke_jwst_2024. No offsets have been applied to the spectra.
• Figure 3: Posterior distributions and transit spectra of our fiducial model at the full spectral resolution of NIRISS SOSS GR700XD and NIRSpec G395H. The prior distributions for the retrieval parameters are shown in red. The top right shows the resulting transit spectrum models and the initial native resolution data. For visual aid, we also overplot a binned set of points for each dataset, the 32-pixel binning.
• Figure 4: Breakdown of our best-fit model for the transit spectrum of TOI-270 d. Absorption bands of CH$_4$, CO$_2$ are clearly visible and one isolated absorption band of CS$_2$ contributes at $4.6\,µm$. The impact of water, CO and SO$_2$ on the spectrum is negligible. The model has been smoothed for visual clarity.
• Figure 5: Histograms of the posterior distributions for all chemical species and the temperature of our retrieval calculations listed in Table \ref{['table:bayes_factors']}. The corresponding retrieved median parameters and $1\sigma$ intervals from benneke_jwst_2024 and the canonical retrieval from holmberg_possible_2024 are shown for comparison.