Venturini Et Al. 2020
Venturini Et Al. 2020
Venturini Et Al. 2020
3 School of Physics, University of Bristol, HH Wills Physics Laboratory, Tyndall Avenue, Bristol BS8 1TL, UK
4 Department of Astronomy and Astrophysics, University of California, Santa Cruz, CA 95064, USA
5 Center for Astrophysics | Harvard & Smithsonian, 60 Garden St, Cambridge, MA 02138, USA
6 Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218
7 Eureka Scientific, Inc. 2452 Delmer Street Suite 100 Oakland, CA 94602-3017
8 Department of Physics, Imperial College London, Prince Consort Road, London SW7 2AZ, UK
9 Department of Planetary Sciences and Lunar and Planetary Laboratory, University of Arizona, Tuscon, AZ, USA
ABSTRACT
Planets between the sizes of Earth and Neptune are the most common in the Galaxy, bridging the
gap between the terrestrial and giant planets in our Solar System. Now that we are firmly in the era of
JWST, we can begin to measure, in more detail, the atmospheres of these ubiquitous planets to better
understand their evolutionary trajectories. The two planets in the TOI-836 system are ideal candidates
for such a study, as they fall on either side of the radius valley, allowing for direct comparisons of the
present-day atmospheres of planets that formed in the same environment but had different ultimate
end states. We present results from the JWST NIRSpec G395H transit observation of the larger and
outer of the planets in this system, TOI-836c (2.587 R⊕ , 9.6 M⊕ , Teq ∼665 K). While we measure
average 30-pixel binned precisions of ∼24ppm for NRS1 and ∼43ppm for NRS2 per spectral bin, we
do find residual correlated noise in the data, which we attempt to correct using the JWST Engineering
Database. We find a featureless transmission spectrum for this sub-Neptune planet, and are able to
rule out atmospheric metallicities <175× Solar in the absence of aerosols at ≲1 millibar. We leverage
microphysical models to determine that aerosols at such low pressures are physically plausible. The
results presented herein represent the first observation from the COMPASS (Compositions of Mini-
Planet Atmospheres for Statistical Study) JWST program, which also includes TOI-836b and will
ultimately compare the presence and compositions of atmospheres for 12 super-Earths/sub-Neptunes.
loss (Ginzburg et al. 2018; Gupta & Schlichting 2019), or tories and the Hubble Space Telescope (HST) yielded
a combination (e.g. Rogers et al. 2023a). Alternatively, only a handful of targets with robust evidence of molec-
the division of the two populations may derive from their ular features, including HD 97658b (Guo et al. 2020),
initial formation in a gas-rich versus gas-poor environ- HD 3167c (Guilluy et al. 2020; Mikal-Evans et al. 2021),
ment (e.g. Lee et al. 2022) or formation interior and K2-18b (Benneke et al. 2019), and 55 Cnc e (Tsiaras
exterior to the ice line (Luque & Pallé 2022; Burn et al. et al. 2016). In comparison, observations of most other
2021). Most recently, Burn et al. (2024) used a coupled objects in this size class were consistent with a lack of
formation and evolution model, including new equations features caused by clouds and hazes and/or a lack of an
of state and interior structure models in the treatment of atmosphere altogether, e.g., GJ 1214b (Kreidberg et al.
water mixed with H/He. With this model, they showed 2014), LHS 1140b (Edwards et al. 2021), L 98–59c and
that the radius valley can be interpreted as separating d (Barclay et al. 2023; Zhou et al. 2023), Kepler-138d
rocky, in-situ-formed super-Earths from water-rich sub- (Piaulet et al. 2022), LHS 3844b (Diamond-Lowe et al.
Neptunes formed farther out in the disk. Each of these 2020), LTT 1445Ab (Diamond-Lowe et al. 2023), GJ
mechanisms can explain the overall radius distribution 1132b (Libby-Roberts et al. 2022), and TRAPPIST-1 b,
of the population, but specific mechanisms result in dif- c, d, e, f, g, and h (Wakeford et al. 2019; de Wit et al.
fering predictions about the atmospheric compositions 2016, 2018; Garcia et al. 2022).
of planets across this distribution (and with other vari- More recently, JWST observations of sub-Neptunes
ables like stellar mass and system age). Therefore, un- (∼2-4 R⊕ ) have started to reveal new insights into the
derstanding the atmospheres of these small planets is compositions of these planets. Phase-curve observations
crucial to understanding how they formed. of GJ 1214b (8.17 M⊕ , 2.628 R⊕ ; Cloutier et al. 2021;
For example, it may be that some sub-Neptunes did Gao et al. 2023) indicated that while the planet had
not form at their current locations, but outside of the a featureless transmission spectrum with HST observa-
snow line and then migrated inward, versus more rocky tions, at MIRI LRS wavelengths (∼5-12µm), there is
super-Earths that formed interior to the snow line (e.g., evidence of a molecular feature at >3σ, which has been
Raymond et al. 2018; Mousis et al. 2020; Burn et al. attributed as most likely due to the presence of water
2021). Evidence for this scenario might be bulk plan- vapor, on both the dayside and nightside of the planet
etary compositions that have a significant fraction of (Kempton et al. 2023; Gao et al. 2023).
water/ice (“water worlds”). Recently, building on the Transmission observations with NIRISS SOSS and
findings of Zeng et al. (2019) but using just the bulk den- NIRSpec G395H of another sub-Neptune planet K2-18b
sities of these planets, Luque & Pallé (2022) indeed iden- (8.63 M⊕ , 2.61 R⊕ ; Cloutier et al. 2019) detected CH4
tified three distinct populations of small planets around and CO2 in an H2 -rich atmosphere (Madhusudhan et al.
M dwarfs – rocky, water-rich, and gas-rich. The au- 2023). Madhusudhan et al. (2023) argued the carbon
thors thus suggest that densities (and therefore the bulk species and lack of detected NH3 are consistent with a
compositions), not radii, delineate the observed small liquid water ocean under a H2 -rich atmosphere, suggest-
planet sub-populations. However, Rogers et al. (2023b) ing that at least part of the radius valley may indeed be
were able to match the observed small planet densi- sculpted by the presence of these water worlds. How-
ties with self-consistent, physically-motivated evolution ever, more recently, Wogan et al. 2024 used photochem-
models (both for photoevaporation and core-powered ical and climate models to show that the K2-18b data
mass loss) that yield rocky core planets with a range does not necessarily imply an ocean and can instead
of H2 /He mass fractions that scale with planet mass. be explained by a ∼ 100× Solar metallicity atmosphere
That is, they find that the mass-radius space is actually sub-Neptune. In any case, it appears likely our knowl-
compositionally degenerate, and not able to conclusively edge surrounding sub-Neptune atmospheres will grow
provide evidence for a population of water worlds (see quickly with more JWST observations.
also Valencia et al. 2007; Rogers & Seager 2010). How- Up to present, JWST observations of super-Earths
ever, determining the atmospheric compositions of sub- (1.0 R⊕ ≲ Rp ≲ 1.7 R⊕ ; Fulton & Petigura 2018) have
Neptunes would help break the degeneracies and pro- been less informative than similar observations of sub-
vide more definite evidence for this population, further Neptunes. Using recent NIRSpec G395H transmission
underlining the importance of understanding the atmo- observations of the super-Earth GJ 486b (1.3 R⊕ , 3.0
spheres of these objects in gaining a full picture of their M⊕ ), Moran & Stevenson et al. 2023 found a slope in
formation and evolution. the spectrum consistent with either stellar activity or a
Previous efforts to observe the atmospheres of super- water-rich atmosphere. In two different observations of
Earths and sub-Neptunes using ground-based observa- GJ 1132b, May & MacDonald et al. 2023 found one of
3
their observations is consistent with a water dominated Table 1. System Properties for TOI-
atmosphere with ∼1% methane, and the other observa- 836c
tion is featureless. These ambiguous JWST observations
of super-Earths demonstrate that further efforts are nec- Property Value
essary to understand the smallest planets and their at- K (mag) 6.804 ± 0.018
mospheres or lack thereof, and the spectra of feature-less T∗ (K) 4552±154
to feature-full 1–3.5 R⊕ planets. R∗ (R⊙ ) 0.665±0.010
Motivated by the prevalence of 1-3 R⊕ planets in gen- log(g) 4.743±0.105
eral and the new possibilities for understanding their [Fe/H]∗ -0.284±-0.067
atmospheres unlocked by JWST, we initiated the large Period (days) 8.59545 ± 0.00001
JWST Cycle 1 program COMPASS (Compositions of Mass (M⊕ ) 9.6+2.7
−2.5
Mini-Planet Atmospheres for Statistical Study) (GO- Radius (R⊕ ) 2.587 ± 0.088
2512, PIs N. E. Batalha & J. Teske), a transmission spec- Teq (K) 665±27
troscopy survey of 11 super-Earth/sub-Neptune plan- e 0.078±0.056
ets with NIRSpec G395H (our full sample also in- ω (°) -28 ± 113
cludes GTO target TOI-175.02 for a total of 12 planets;
inclination (°) 88.7 ±1.5
Batalha et al. 2021). Broadly, the program aims to bet-
a (AU) 0.0750±0.0016
ter understand whether small planets have atmospheres,
and if so, the compositional diversity of this population. All values from Hawthorn et al.
More specifically, COMPASS aims to: (1) map out at- (2023)
mospheric detectability as a function of radius across the
small planet regime—i.e. is there a radius below which
we stop seeing detectable atmospheric features? (2) ex-
plore what the diversity in atmospheric composition im- planets on either side of the radius valley that formed
plies for the origin(s) of these planets, (3) compare the within the same stellar environment. The transmission
compositions of “sibling” planets orbiting the same star spectrum of the interior planet, TOI-836b, as well as a
but which have different radii/periods, and (4) investi- comparison of the NIRSpec G395H transmission spectra
gate what population-level inferences we can draw from of the two confirmed planets in this system are presented
the entire sample. Importantly, the sample was selected in Alderson & COMPASS et al. (2024).
with a quantitative, reproducible metric to enable ro- In Sections 2 and 3 we detail our observations of TOI-
bust inferences about the properties of somewhat cool 836c and present three different data reductions. In Sec-
(∼400-1000 K), 1-3 R⊕ planets. More details about the tion 4 we present our fits to the white light curves and
motivation and benefits of our selection method, and an our transmission spectrum. In Section 5 we interpret
example of a simple population-level trend which could our transmission spectrum through the use of different
be investigated, can be found in Batalha et al. (2023). atmospheric models. In Section 6 we contextualize our
Here we present observations of TOI-836c (TOI- results and present new bulk composition models for
836.01, HIP 73427c), a 2.587±0.088 R⊕ , 9.6+2.7 −2.5 M⊕ TOI-836c. We present our conclusions in Section 7.
sub-Neptune (Hawthorn et al. 2023) discovered by the
Transiting Exoplanet Survey Satellite (TESS; Ricker
2. OBSERVATIONS
et al. 2015) and confirmed by ground-based transit and
radial velocity follow-up. The bulk density of the planet One transit of TOI-836c was obtained by the Near-
(∼0.6 ρ⊕ ) suggests a relatively extended gaseous en- Infrared Spectrograph (NIRSpec) G395H/F290LP grat-
velope. TOI-836c is one of two planets known to ex- ing on 16 February 2023 in the NIRSpec Bright Object
ist in this system and exhibits transit timing variations Timeseries mode with the SUB2048 subarray and the
(TTVs) on the order of 20 minutes. However, its sibling, NRSRAPID readout pattern. We utilized 3 groups per
TOI-836b, a super-Earth in an interior orbit, does not integration, with 6755 integrations, and a total expo-
exhibit any TTVs, pointing to the possible existence sure time of 6.8 hours. The spectral traces using NIR-
of an additional exterior planet. While these planets Spec G395H (2.87 - 5.18 microns) are taken across two
were chosen agnostically using our aforementioned se- separate detectors, NRS1 and NRS2, with a small gap
lection criteria (Batalha et al. 2023), the presence of in wavelength coverage (between 3.72 and 3.82 microns)
the two confirmed planets in this multi-planet system in caused by the gap between the two detectors.
our COMPASS sample allows us to directly compare two 3. DATA REDUCTION
4
We reduced our observations using three independent and 3.819918–5.082485 µm for NRS2) using combina-
pipelines to determine the robustness of our derived tions of: extraction aperture half-widths of 4-8 pixels,
best-fit parameters and transmission spectrum. We de- background aperture half-widths of 8-11 pixels, sigma
tail our Eureka! (Bell et al. 2022) reduction in Sec- thresholds for outlier rejection for the optimal extrac-
tion 3.1, our ExoTiC-JEDI (Alderson et al. 2022, 2023) tion of 10 and 60 (the latter of which approximates
reduction in Section 3.2, and our Aesop (Alam et al. in standard box extraction), and two different methods
prep) reduction in Section 3.3. of background subtraction (an additional column-by-
column background subtraction and a full frame median
3.1. Eureka! Reduction
background subtraction). In order to determine the op-
We first reduced the data using Eureka!1 (Bell et al. timal combination of these extraction parameters, we
2022), an end-to-end pipeline for analyzing both HST generated a white light curve using Stage 4 of Eureka!
and JWST data. The JWST Eureka! pipeline relies for each combination, and chose the reduction that pro-
on the STScI jwst pipeline (Bushouse et al. 2022) for duced the lowest median absolute deviation in the white
its initial two stages (we use version 1.8.2 of the jwst light curves. This process was done separately for NRS1
pipeline and version 0.9 of Eureka!). We analyzed data and NRS2, meaning that the reduction parameters that
from the two detectors separately in order to ascertain resulted in the optimal light curves were not necessar-
any flux calibration offsets between the detectors. Initial ily the same for each detector. For NRS1, our optimal
calibration was done in Stage 1 of Eureka!, following the aperture half-width for the extraction of the trace was 4
suggested jwst pipeline steps of converting the raw data pixels and our optimal inner bound for the background
ramps to slopes. We utilize the default Eureka! pa- subtraction was 9 pixels (i.e. our background was from
rameters, with the exception of using a jump detection 9 pixels away from the trace location outwards to the
threshold of 15. Additionally, we utilized group-level lower and upper edges of the frame). For NRS2, our op-
background subtraction to account for the 1/f noise, timal aperture half-width for the extraction of the trace
which has been found to often be the dominant source was 6 pixels and our optimal inner bound for the back-
of systematic noise in NIRSpec data (Alderson et al. ground subtraction was 8 pixels. For both detectors,
2023; Rustamkulov et al. 2023). We did not utilize the we favored an additional column-by-column background
in-built Eureka! group level background subtraction subtraction over a median frame background subtrac-
method, instead opting to utilize the 1/f noise correction tion and a sigma threshold for the optimal extraction of
methodology employed in ExoTiC-JEDI2 (see Alderson 60. We then utilized a custom light curve fitting code
et al. (2023) for more details about the 1/f noise cor- instead of the default Eureka! fitting code for increased
rection), as it provided more robust results with fewer flexibility, but here continue to refer to this reduction as
temporal outliers. We used the default parameters for the Eureka! reduction.
Eureka! Stage 2, which again utilized the default re- From our white light curves, we then produced spec-
duction steps in the jwst pipeline in order to further troscopically binned light curves of 30-pixels (R∼200).
calibrate the Stage 1 outputs. We iteratively trim 3σ outliers three times from a 50
We then utilized Stage 3 of Eureka!, which does the point rolling median in both the white light curves and
data reduction and allows for different choices of trace the binned light curves. We then fit the optimal light
and background apertures, different sigma thresholds curves for NRS1 and NRS2 separately using a combina-
for rejection of background pixels, and different polyno- tion of a Levenberg–Marquardt least-squares minimiza-
mial orders for an additional column-by-column back- tion and the affine-invariant Markov chain Monte Carlo
ground subtraction. In addition to masking the pixels (MCMC) ensemble sampler emcee (Foreman-Mackey
flagged as do not use by the jwst pipeline (which are et al. 2013) where we fit for a combined astrophysi-
masked by default within Stage 3 of Eureka!), we also cal and systematic noise model and optimize the log-
mask pixels flagged as saturated, dead, hot, low quan- likelihood. Our astrophysical model is calculated using
tum efficiency, or no gain value, as well as any indi- the batman package (Mandel & Agol 2002; Kreidberg
vidual columns that resulted in anomalous time series 2015), assuming quadratic limb-darkening coefficients
compared to the columns around it (there were nine calculated using ExoTiC-LD (Grant & Wakeford 2022)
such columns in NRS2). We then generated white light using Set One of the MPS-ATLAS models (Kostogryz
curves (summing over 2.862704–3.714356 µm for NRS1 et al. 2022, 2023). We fit for the system inclination,
a/R∗ (the orbital semi-major axis with respect to the
1 https://github.com/kevin218/Eureka host star), T0 (time of transit), and Rp /R∗ (the radius
2 https://github.com/Exo-TiC/ExoTiC-JEDI of the planet with respect to the radius of the host star),
5
fixing the period, eccentricity, and stellar parameters to ExoTiC-JEDI begins with a modified version of Stage
those shown in Table 1. Our instrumental noise model, 1 of the jwst pipeline (v.1.8.6, context map 1078), per-
S, was of the form forming linearity, dark current and saturation correc-
tions, and using a jump detection threshold of 15. Next,
S = p1 + p2 ∗ T + p3 ∗ X + p4 ∗ Y, (1) a custom destriping routine is used to remove 1/f noise,
masking the spectral trace at 15 times the standard de-
where pN is a free parameter, T is the vector of times,
viation of the PSF from the dispersion axis in each in-
and X and Y are vectors of the positions of the trace
tegration, subtracting the median pixel value from each
from Stage 3 of Eureka!. We also fit for an additional
detector column at the group level. We also perform
error term that we add in quadrature with the per-point
a custom bias subtraction, computing the median of
measured errors. We trim the initial 1,000 points to
each detector pixel in the first group across all the in-
remove any initial ramp. We chose to trim a conser-
tegrations in the time series, and then subtracting this
vatively long initial duration to account for any initial
new median “pseudo-bias” image from all groups. This
ramp in our data that might bias the linear slope in
method was found to improve the standard deviation of
our systematic noise model given our abundance of pre-
the out-of-transit points for both detector white light
transit baseline and the apparent small initial ramp in
curves, and reduced an offset seen between the resulting
the NRS1 white light curve (see Figure 1). We initialized
NRS1 and NRS2 transmission spectra. ExoTiC-JEDI
three times the number of free parameters for the num-
uses the standard jwst Stage 1 ramp fitting step and
ber of walkers (for a total of 27 walkers for these fits) at
Stage 2 steps to produce the 2D wavelength map.
the best-fit values from the initial Levenberg–Marquardt
We next extract our 1D stellar spectra, with further
least-squares minimization. For the MCMC fit, we used
pixel cleaning steps and 1/f correction. With the stan-
uninformed priors for all of our parameters and used a
dard data quality flags produced by the jwst pipeline,
burn-in of 50,000 steps which was discarded then an ad-
we replace any pixels flagged as do not use, saturated,
ditional 50,000 steps to ensure adequate sampling of the
dead, hot, low quantum efficiency, or no gain value, with
posterior. We adopt the median of each chain as our
the median value of the next 4 pixels in each row. We
best-fitting value for each parameter and take the stan-
additionally identify any remaining pixels that are out-
dard deviation of each chain as the uncertainty for each
liers from their nearest neighbors on the detector, or
parameter. We show the white light curves for NRS1
throughout the time series (such as cosmic rays), with
and NRS2, the best-fit models, and the residuals in the
a 20σ threshold in time and a 6σ threshold spatially.
top panels in Figure 1 and the best-fit astrophysical pa-
The flagged pixel is replaced with the median value of
rameters in Table 2.
the surrounding 10 integrations or 20 pixels in the row.
We used the resulting MCMC chains of the astrophys-
Remaining 1/f noise and background is removed by sub-
ical parameters from the white light curve fits as Gaus-
tracting the median unilluminated pixel value from each
sian priors on the spectroscopic light curve fits. We com-
column in each integration. To extract the 1D stellar
bined the MCMC chains from the NRS1 and NRS2 fits
spectra, we fit a Gaussian to each column to obtain the
for the inclination, a/R∗ , and T0 and used the result-
spectral trace and width, fitting a fourth-order polyno-
ing combined chains as Gaussian priors for each of those
mial to each and smoothing with a median filter to de-
parameters for the fits to the spectroscopically binned
termine the aperture region. We used an aperture five
light curves centered at the median of each of the com-
times the FWHM of the trace, approximately 8 pixels
bined chains and with a conservative width of 3× the
wide, using an intrapixel extraction. The resulting 1D
standard deviation of the combined chains. We utilized
stellar spectra are cross-correlated to obtain the x- and
a wide flat prior for the Rp /R∗ of each spectroscopic bin.
y-positional shifts throughout the observation for use in
In order to further validate the robustness of our re-
systematic light curve detrending.
sults, we then reduced our data with additional inde-
We fit white light curves for both NRS1 and NRS2, as
pendent reductions, using ExoTiC-JEDI and Aesop.
well as spectroscopic light curves across the full NIR-
3.2. ExoTiC-JEDI Reduction Spec/G395H wavelength range. For the white light
curves (spanning 2.814–3.717 µm for NRS1 and 3.824–
The Exoplanet Timeseries Characterisation - JWST
5.111 µm for NRS2), we fit for i, a/R∗ , T0 , and Rp /R∗
Extraction and Diagnostic Investigator (ExoTiC-JEDI,
(Table 2), holding the period and eccentricity fixed to
Alderson et al. 2022) package produces an end-to-end
values presented in Hawthorn et al. (2023) (see Ta-
reduction of JWST time-series data, beginning with the
ble 1). We use stellar limb darkening coefficients cal-
uncal files. Throughout the analysis, data from NRS1
culated with ExoTiC-LD based on the stellar T∗ , log(g),
and NRS2 are treated separately.
6
and [Fe/H]∗ listed in Table 1, using Set One of the MPS- series stellar spectra, we fit a Gaussian profile to each
ATLAS stellar models (Kostogryz et al. 2022, 2023) and column of a given integration to find the center of the
the non-linear limb darkening law (Claret 2000), which spectral trace and smoothed the trace centers with a
are held fixed in a batman transit model. We used a median filter, followed by fitting a fourth-order polyno-
least-squares optimizer to fit the transit model simulta- mial to the smoothed trace centers. We then extracted
neously with our systematic model S(λ), which took the the 1D stellar spectra by summing up the flux within a
form 10-pixel wide aperture, calculating the uncertainties in
the stellar spectra assuming photon noise.
S(λ) = s0 + (s1 × xs |ys |) + (s2 × t), (2) We generated the white light curves for NRS1
and NRS2 by summing the flux between 2.862704–
where xs is the x-positional shift of the spectral trace,
3.714356 µm and 3.819918–5.082485 µm, respectively,
|ys | is the absolute magnitude of the y-positional shift
and inflated the flux uncertainties following Pont et al.
of the spectral trace, t is the time and s0 , s1 , s2 are co-
(2006). We then fit the broadband and spectroscopic
efficient terms. We use the same binning scheme as was
light curves using a least-squares minimizer. We fit each
utilized for the Eureka! reduction. For these spectro-
light curve with a two-component function consisting of
scopic light curves, we fit for Rp /R∗ , holding T0 , i, and
a transit model (generated using batman) multiplied by
a/R∗ fixed to the respective white light curve fit value.
a systematics model including a linear polynomial in the
For all fitted parameters, we take the optimized values
x- and y-pixel positions on the detector. We first fit the
and the standard deviation errors as the value and un-
broadband light curve by fixing the period and eccen-
certainty for each parameter. For both the white and
tricity to the Hawthorn et al. (2023) values, and fitting
spectroscopic light curves, we removed any data points
for T0 , a/R∗ , i, Rp /R∗ , stellar baseline flux, and sys-
that were greater than 4σ outliers in the residuals, and
tematic trends using wide uniform bounds. We adopt
refit the light curves until no such points remained. We
the optimized parameters and their standard deviations
also rescaled the flux time series errors using the beta
as the best-fitting value and associated uncertainty for
value (Pont et al. 2006) measured from the white and
our fitted parameters. We again use the same binning
red noise values calculated for the Allan variance plots
scheme as was utilized for the other two reductions. For
to account for any remaining red noise in the data (see
these spectroscopic light curves, we fixed T0 , a/R∗ , and
middle panel of Figure 3).
i to the best-fit values from the broadband light curve
3.3. Aesop Reduction (Table 2) and fit for Rp /R∗ . We held the non-linear
limb-darkening coefficients fixed to theoretical values,
Aesop is a JWST data reduction pipeline developed
which we computed using ExoTiC-LD and the 3D stellar
for the reduction and analysis of NIRSpec G395H obser-
model grid from Magic et al. (2015).
vations that has been benchmarked with several other
pipelines in the literature (see e.g., Alderson et al. 2023). 4. RESULTS
The pipeline’s methodology is described in detail in
Alam et al. (in prep), which we briefly summarize 4.1. Correlated noise and timing offsets
here. We began with the uncal uncalibrated JWST data The fits to the white light curves for NRS1 and NRS2
products, treating the data sets for the NRS1 and NRS2 for all three independent reductions were consistent in
detectors separately. We first ran the standard Stage 1 their best-fit parameters (see Table 2). For all three
steps of the jwst pipeline for time-series observations, reductions, the best-fit times of mid-transit differed by
including corrections for saturation, bias, linearity, and greater than 4σ between the two detectors (see Figure 2
dark current. We set the detection threshold for the and Table 2). This is likely not a physical effect, as the
jump step to 15, apply a custom group-level background NRS1 and NRS2 data comprise a single observation in
subtraction to remove 1/f noise using a 15σ threshold which the light was dispersed across the two detectors
and a second-order polynomial, and perform standard simultaneously. While the detectors are utilized simul-
ramp fitting. We used the standard steps of the jwst taneously, the time stamps between the two detectors
pipeline for Stage 2 to extract the integration exposure vary slightly. It is important to note that this time array
times and the 2D wavelength array. difference is more than an order of magnitude smaller
We then performed additional cleaning steps to re- than the timing offset between the transit times from
place poor data quality pixels (flagged as bad, saturated, the fits to the white light curves of the two detectors, so
hot, or dead) with the median of the neighboring pixels is not likely to be the reason for the transit timing offset.
and remove residual 1/f noise by calculating the median Due to the fact that all three reductions resulted in the
pixel value of each column. To extract the 1D time- same timing offset despite different reduction parame-
7
Eureka! White Light Curves 0.0005
Eureka! Residuals
Relative Flux
1.000
0.999 0.0000
0.998
NRS1 0.0005
1.001
0.0005
Relative Flux
1.000
0.0000
0.999
0.998 NRS2 0.0005
4 3 2 1 0 1 2 4 3 2 1 0 1 2
Time From Mid-Transit (hours) Time From Mid-Transit (hours)
ExoTiC-JEDI White Light Curves ExoTiC-JEDI Residuals
1.002 0.0005
Relative Flux
1.000 0.0000
0.0005
0.998 NRS1
1.001 0.0010
Relative Flux
1.000 0.0005
0.999 0.0000
0.998 NRS2 0.0005
4 3 2 1 0 1 2 4 3 2 1 0 1 2
Time From Mid-Transit (hours) Time From Mid-Transit (hours)
Aesop White Light Curves Aesop Residuals
0.0005
Relative Flux
1.000
0.0000
0.998
NRS1 0.0005
1.001
0.0005
Relative Flux
1.000
0.0000
0.999
NRS2 0.0005
0.998
4 3 2 1 0 1 2 4 3 2 1 0 1 2
Time From Mid-Transit (hours) Time From Mid-Transit (hours)
Figure 1. White light curves for each reduction with the best-fit models and associated residuals shown as both a time series
and histogram. NRS1 (blue) and NRS2 (red) are shown separately for each reduction. The Eureka! reduction is shown at the
top, ExoTiC-Jedi in the middle, and Aesop at the bottom.
ter choices and fitting methods, this offset is likely also lations of the observations from PandExo (Batalha et al.
not reduction or fitting specific, but instead due to the 2017, see Figure 4).
residual correlated noise present in our fits to the white
light curves (Figure 3). Using the reduction and fitting
method described in Section 3.1, we also investigated It is important to determine whether the correlated
whether utilizing a higher order and more complicated noise that is likely causing the offset in the transit times
systematic noise model could reduce the amount of cor- is affecting the resulting transit depths and therefore the
related noise and the timing offset, but did not find that overall transmission spectrum. We are also interested in
the added complexity reduced the amount of residual pinning down the correct time of transit to allow for an
correlated noise or remedied the timing offset. We note improved understanding of the dynamics of this system.
that the precisions we achieve, despite the correlated Moreover, given the evidence of TTVs on the order of
noise, are still comparable to those expected from simu- 20 minutes for this planet (Hawthorn et al. 2023), we
cannot adopt a fixed center of transit time and apply it
8
Table 2. Best-fit system parameters from the fits to the white light curves for all three reductions.
tors.
a/R*
.50
.75
.00
.25
.50
3.5
3.7
4.0
4.2
4.5
24
25
27
88
88
89
89
89
T0 (minutes) a/R* Inclination (°) model. We first individually add each engineering pa-
rameter to the systematic model in Equation 1, meaning
Figure 2. We show the time in minutes from the predicted that our new systematic model is of the form
mid-transit time of 59991.7252 BMJD (T0 ), the a/R∗ , and
the inclination from the fits to our white light curves. The
contours show the posterior distributions from the MCMC S = p1 + p2 ∗ T + p3 ∗ X + p4 ∗ Y + p5 ∗ E, (3)
fits to the Eureka! reduction and the best-fit values from
the fits to the other reductions are shown as points above where E represents the time series of a single engineer-
the distribution for each parameter (as we did not run an
ing parameter that has been resampled to the time se-
MCMC for the other reductions). All of our reductions agree
on the presence of an offset in the times of transit between
ries for the white light curve from the detector that is
NRS1 and NRS2. being fit. We loop over each of the more than 1,000
engineering parameters separately, fitting using a Lev-
to both the NRS1 and NRS2 white light curves since we enberg–Marquardt least-squares minimization. We use
cannot be certain which of the two derived T0 values is the same process and leave the same parameters as free
closest to the truth. as described in the fits in Section 3.1, but elect to not
In order to investigate the cause of the residual run an MCMC for every combination for computational
correlated noise, we utilized the JWST Engineering efficiency.
Database. There are thousands of telemetry points,
designated by mnemonics, that are taken onboard the 3 https://mast.stsci.edu/portal/Mashup/Clients/jwstedb/
spacecraft at different time intervals, which are all avail- jwstedb.html
9
Eureka! ExoTiC-JEDI Aesop
10 4
RMS
10 5
50 10 1
40
30
20 10 2
100 101 102 100 101 102
10 Bin Size (Number of Points) Bin Size (Number of Points)
3.0 3.5 4.0 4.5 5.0
Wavelength (microns) Figure 5. The Allan variance plots for both detectors for the
Eureka! reduction for each of the spectroscopic bins fit with
Figure 4. The expected errors on our transmission spec- a systematic model of the form in Equation 1. Each colored
trum simulated from PandExo (Batalha et al. 2017) with line represents a different bin and the black line represents
our measured errors for each of our three reductions. We the normalized standard error. We have normalized the root
achieve comparable precisions between our three reductions, mean squared (RMS) values √ for the data in order to show
and generally are not able to achieve errors as expected from that the bins follow a 1/ N scaling and that there is little
PandExo, likely due to the noise structure in our observation excess red noise in the residuals of the fits to the binned light
that only utilizes three groups. curves.
We then generate a ranked list of all of the engineering ton noise limit (i.e., the RMS of the difference of the ex-
parameters for each detector (as the engineering param- pected and measured lines in Figure 3), and then rank all
eter exploration is done independently). We calculate of the engineering parameters by this calculated metric.
the RMS of the difference between the measured stan- Ranking the parameters in this way ensures that their
dard deviation of the residuals and the predicted pho-
10
inclusion will decrease the amount of residual correlated engineering parameters did not agree between the two
noise in the white light curves. detectors.
In order to not be limited to a single additional engi- For NRS2, we found that, in all systematic noise mod-
neering parameter (and indeed some of the mnemonics els including up to 15 different engineering terms, all of
are associated with other mnemonics), we then com- the times of transit were in agreement with each other
bine the top ranked mnemonics. Treating each detec- (and the time derived from the non-engineering model)
tor separately, we begin with only including the single within 0.65σ. Thus, adding the engineering parame-
mnemonic that improves the residual correlated noise ters did not make a significant difference in the mid-
the most (e.g. the top mnemonic in our ranked list de- transit time for NRS2, although the inclusion of addi-
scribed above) and fit the white light curve using the tional terms did help to reduce the amount of corre-
MCMC procedure detailed in Section 3.1. We then add lated noise. The amount of residual correlated noise in
the second best mnemonic to our systematic noise model NRS2 is lower than for NRS1 regardless of the system-
in the form atic model used and all of the inclinations and a/R∗
values in the fits with and without any number of engi-
neering parameters were all within 0.5σ of each other,
S = p1 + p2 ∗ T + p3 ∗ X + p4 ∗ Y + p5 ∗ E1 + p6 ∗ E2 , (4) meaning that while the additional engineering parame-
ters helped to reduce the amount of residual correlated
noise, any residual correlated noise from the fits using
where E1 is the top ranked mnemonic and E2 is the
the nominal systematic model do not seem to be alter-
second highest ranked mnemonic. We then repeat the
ing the NRS2 best-fit parameters. The NRS1 detector,
full MCMC fit on this two mnemonic engineering pa-
on the other hand, shows much more residual correlated
rameter model. We then add another mnemonic to the
noise and the additional engineering terms more appre-
systematic model and run an MCMC fit. We continue
ciably affect the red noise. Moreover, the addition of
this process to include up to 15 mnemonics, meaning
engineering terms results in times of transit that differ
that we have 15 different systematic noise models that
by up to 1.8σ from the fit without the engineering pa-
include engineering parameters that we try for each de-
rameters.
tector. We then determine the optimal number of engi-
For consistency, we opt to use the optimal engineer-
neering parameters as the number that corresponds to
ing parameter systematic noise model (including 13 pa-
the engineering model that most reduced the residual
rameters) from NRS1 to fit both detectors. We find
correlated noise without adding additional scatter.
that the best-fit white light curve astrophysical param-
The purpose of this exercise was to determine if a re-
eters for NRS2 are still within 1σ of those derived from
duction in residual correlated noise would result in times
the model without the inclusion of engineering parame-
of transit that were more in agreement between the two
ters, further showing that any residual correlated noise
detectors, and therefore if the transit timing offset was
in NRS2 is not altering the white light curve best-fit pa-
likely caused by the residual correlated noise. However,
rameters. As different engineering models do not make
we are not claiming to determine an underlying physi-
a difference to our derived best-fit values for NRS2, we
cal process that is being tracked by our best engineering
opt for this model (where we only use the NRS1 derived
parameter model. We first optimized the two detec-
parameters) as our engineering model for simplicity.
tors separately, so it was not a requirement that the
We find that utilizing the best-fit engineering
mnemonics utilized in the best engineering model be
mnemonics from NRS1 for the systematic models for
identical between the two detectors, or have the same
both NRS1 and NRS2 reduces the amount of residual
number of additional engineering parameters in the sys-
correlated noise for both detectors as evident in Fig-
tematic model. Interestingly, the two top mnemonics
ure 6 (although the inclusion of the engineering terms
between the two detectors, IGDP MIR IC V DETHTR
reduces the amount of residual correlated noise more for
and IFGS TFG BADCNTCNT respectively, were in
NRS1 than for NRS2), and causes the best-fitting times
agreement. The former mnemonic is a focal plane elec-
of transit to be within 2.3σ of each other (Figure 7).
tronic SCE1-IC Detector Heater Voltage and the latter
This means that while none of our 15 different system-
is the Fine Guidance Sensor Track/Fine Guide Image
atic models that included engineering parameters could
Data Processing Telemetry Bad Centroid Data Count.
entirely resolve the discrepancy in transit times between
While utilizing only these two mnemonics did improve
the two detectors, the values are in closer agreement us-
the residual correlated noise, this alone did not remedy
ing additional engineering parameters than for the sys-
the transit timing offset, so we do not further discuss the
tematic model outlined in Section 3.1, which does not
physical implications of these parameters. The other top
11
Eureka! NRS1 Eureka! NRS2
With Engineering With Engineering
Without Engineering Without Engineering
10 4
RMS
Inclination (°)
100 101 102 100 101 102
Bin Size (Number of Points) Bin Size (Number of Points)
Figure 6. The Allan variance plots for both detectors for
the Eureka! reductions where we include 13 engineering pa-
rameters in the systematic model are shown in colors while
0
5
0
5
22 0
.5
.0
.5
.0
.50
.75
.00
.25
.50
3.5
3.7
4.0
4.2
4.5
24
25
27
88
88
89
89
89
the corresponding curves without the engineering parame- T0 (minutes) a/R* Inclination (°)
ters are shown in grey for each detector (the grey curves are
identical to those shown in panel one of Figure 3). In the ab- Figure 7. The posteriors for the time (T0 in minutes from
sence of red noise, the residuals would follow the black lines. the predicted mid-transit time of 59991.7252 BMJD), the
While there is still residual correlated noise, the amount of a/R∗ , and the inclination from the fits to our white light
red noise has been reduced. curves of our Eureka! reduction including 13 engineering
parameters in our systematic model are shown in colors. The
consider engineering parameters. This, coupled with a corresponding posteriors from the fits without the engineer-
reduction in the residual correlated noise when includ- ing parameters are shown in grey and outlined in their re-
ing engineering terms, suggests that further exploration spective detector colors. The posteriors for the fits without
the engineering parameters are identical to those shown in
is warranted of such engineering terms in future efforts
Figure 2. The T0 values are more in agreement between
to reduce correlated noise in similar observations using NRS1 and NRS2 when we include engineering parameters.
NIRSpec G395H.
Using these different reductions, we then assess the
in Section 4.1), which reduced the correlated noise in
impact of the correlated noise (and timing offsets) on
both NRS1 and NRS2. This exercise is important in de-
our spectrum. We next calculate and compare the trans-
termining the robustness of our spectrum and any con-
mission spectrum using our engineering systematic noise
straints that we can place on the planet’s atmosphere.
model and our nominal position and time noise models.
We find that all of the bins in the transmission spec-
trum generated with and without the engineering pa-
4.2. Transmission Spectrum
rameters agree at the 2.3σ level or better, indicating that
Despite the residual correlated noise present in the the noise that is likely affecting the transit times does
fits to the white light curves, there did not seem to be not alter our spectrum significantly (Figure 9). This,
a substantial amount of residual correlated noise in the along with the lack of residual correlated noise in the
spectroscopic bins (see Figure 5), so we first determine spectroscopically binned light curves (Figure 5), indi-
the transmission spectrum using each of our three re- cates that our transmission spectrum is robust despite
ductions with the nominal (not including engineering the timing offset between the white light curves from
mnemonics) noise models described in Section 3. We the two detectors. Additionally, we extracted the spec-
find that all three of our reductions agree per-point at trum for our Eureka! reduction using only the MCMC
the ∼2σ level or better (Figure 8). This indicates that chains from each detector separately as priors (i.e. at
individual reduction and fitting specific choices do not the two different transit times) and find that the per-
seem to greatly impact our derived transmission spec- point transmission spectra using the best-fit NRS1 time
trum. All three of our reductions are consistent with a and the best-fit NRS2 time each agree with the nominal
relatively flat spectrum with no easily identifiable fea- spectrum at the 0.08σ level or better. Since our nom-
tures. inal Eureka! fits utilize wide Gaussian priors derived
We also generate a transmission spectrum using our from each chain (of 3× the standard deviation in the
13 parameter best-fitting engineering model (described
12
Eureka!
1400 ExoTiC-JEDI
Aesop
1300
(Rp/R*)2 (ppm)
1200
1100
1000
0
2
3.0 3.5 4.0 4.5 5.0
Wavelength (microns)
Figure 8. We show the transmission spectrum for each reduction in the top panel and the differences between the reductions
(in units of sigma) in the bottom panel. All three of our reductions agree per-point, at the ∼2σ level or better, with a median
difference of 6ppm.
combined chains), this further shows that neither our mospheric mean molecular weight and opaque pressure
choice of wide Gaussian prior nor the bimodal times al- level allowed by our data. To give context to the “non-
ter the resulting transmission spectrum. Since all of our physical models” and the “fixed-climate spectral mod-
transmission spectra agree regardless of the inclusion of els”, we also compute a small grid of radiative-convective
engineering parameters or prior on the time of transit, (RC) models using PICASO 3.1 (Mukherjee et al. 2023)
for the Eureka! reduction, we focus our subsequent that allows us to gain a deeper understanding of the
analysis on the spectrum derived without engineering potential climate structure of TOI-836c. Finally, using
parameters, using the wide combined chain priors for these RC temperature-pressure profiles we leverage a mi-
each parameter when modeling the atmosphere. crophysical model (CARMA; Gao et al. 2023) to compute
an altitude-dependent haze profile. Ultimately, this al-
5. MODELS
lows us to put into context the physical plausibility of
an optically thick aerosol at low pressures that could be
In order to interpret our transmission spectrum, we impeding the detection of strong molecular features.
employ three different techniques on all three data re-
ductions. First, we fit for a series of “non-physical mod-
els” to understand the overall shape of the spectra and
determine the significance of any structure that appears 5.1. Non-Physical Models
to be a feature. Second, we computed a grid of “fixed- Using the “MLFriends” Nested Sampling Algorithm
climate spectral models” using PICASO (Batalha et al. (Buchner 2016, 2019) implemented in the open source
2019) with a fixed parameterized temperature-pressure UltraNest code (Buchner 2021) we test seven different
profile for a range of metallicities and grey-cloud slab non-physical models, which contain increasing levels of
pressure levels. This gives us an overview of the at- complexity. The model suite is as follows:
13
Eureka!
1400 Eureka! with Engineering
1300
(Rp/R*)2 (ppm)
1200
1100
1000
900
3
2
Difference ( )
1
0
1
2
3 3.0 3.5 4.0 4.5 5.0
Wavelength (microns)
Figure 9. We show the transmission spectrum for the nominal Eureka! case (detailed in Section 3.1) and the spectrum where
we utilize a 13 engineering parameter systematic noise model. We show the differences between the nominal spectrum and
spectrum that included engineering parameters (in units of sigma) in the bottom panel. The two spectra agree per-point at the
2.3σ level or better with a median difference of 17ppm.
Table 3. The results from our seven non-physical models for each of our three reductions. We show the log likelihood, the
χ2 /N, and the offset (=NRS1−NRS2) between the detectors (where applicable) for each model.
Eureka! Exo-TiC-JEDI Aesop
2
Model Type log Z χ /N Offset [ppm] log Z χ2 /N Offset [ppm] log Z 2
χ /N Offset[ppm]
Slope Zero -106 1.92 N/A -91 1.64 N/A -96 1.74 N/A
Step Function -72 1.24 51 -70 1.19 41 -83 1.43 34
Line -79 1.38 N/A -77 1.33 N/A -85 1.48 N/A
Gaussian -78 1.25 N/A -77 1.23 N/A -87 1.45 N/A
NRS1 Gaussian -72 1.21 48 -71 1.18 43 -82 1.32 31
NRS2 Gaussian -72 1.23 53 -70 1.19 43 -83 1.42 35
Gaussian + Line -78 1.31 N/A -77 1.30 N/A -85 1.44 N/A
• Model 1 [Zero-sloped line, 1 free parameter]: Fits • Model 4 [Gaussian, 4 free parameters]: Tests for
for the y-intercept assuming zero slope. the case of a single Gaussian-shaped spectral fea-
ture. Multiple features, if they were significant,
• Model 2 [Step function, 2 free parameters]: Fits are captured via degenerate solution. This fits for
for two y-intercepts independently for NRS1 and 2 2
y = c + Ae(λ−λ0 ) /σ , where c is the baseline, A is
NRS2. Assumes zero slope for both. the amplitude of the feature, λ0 is the wavelength
• Model 3 [Line, 2 free parameters]: Fits for the midpoint of the feature, and σ is the width of the
slope and the y-intercept. feature.
14
Figure 10. We show the transmission spectrum for the nominal Eureka! reduction (detailed in Section 3.1), the ExoTiC-JEDI,
and Aesop reductions, respectively compared to the fitting results of the three highest-likelihood non-physical models. The
three highest-likelihood non-physical models are consistent between reductions. In all panels, the solid colored line indicates
the median fit from the samples, the dark shaded region is the 68-percentile region (1σ) and the light shaded region is the
99-percentile region. The Bayes factor, as defined by the difference in evidence, for each of the non-physical models prefers the
step function model in all three reductions.
• Model 5 [NRS1 Gaussian + offset, 5 free param- or NRS2 Gaussian + offset models are all strongly pre-
eters]: A superposition of Model 2 and 4 and ferred (| lnB12 | > 13) when compared to the slope-zero
therefore accounts for an offset between NRS1 and line. However, when comparing the simple step function
NRS2 while allowing for a Gaussian feature in the model against any of the Gaussian models, none of the
NRS1 wavelength region. The prior on λ0 is con- latter are preferred for any of the reductions because the
strained to that of NRS1. increase in number of free parameters does not improve
the evidence despite minor improvements to the χ2 /N.
• Model 6 [NRS2 Gaussian + offset, 5 free param- This suggests that the transmission spectrum, regard-
eters]: Accounts for an offset between NRS1 and less of data reduction method, is well-explained by two
NRS2 while allowing for a Gaussian feature in the zero-sloped lines (step function) with an offset of 34, 41,
NRS2 wavelength region. and 51 ppm for the Aesop, ExoTiC-JEDI, and Eureka!
reductions, respectively. The tentative bump at 3.25µm
• Model 7 [Gaussian + line, 5 free parameters]: A appears within 1σ of the highest likelihood solution of
superposition of Model 3 and 4 and therefore a NRS1 Gaussian + offset model and the Gaussian + line
Gaussian feature with a sloped line. This would model for both reductions. However, when compared
account for any changes to a continuum, along with a featureless offset model, it is not preferred (ra-
with a feature. tio of the model evidences is < 1). Other even more
tentative features in NRS1 and NRS2 shown within the
Given the resultant evidence of seven model runs, we 99-percentile region of Figure 10 are likely due to sys-
compute the Bayes factor (lnB12 = lnZ1 [Model 1] − tematic scatter. Similarly, when compared with the step
lnZ2 [Model 2]; Trotta (2008)) in order to determine if function model, these features are not preferred because
one model is preferred over another. The Bayes fac- the increase in likelihood is not enough to outweigh the
tor only directly translates to a σ-significance (e.g., increase in free parameters. Even though the scatter
Eqn. 27 in Trotta (2008)) if the model’s free param- seems to be consistent between reductions, additional
eters are nested (e.g. f (c1 , c2 , c3 = A, ..., cN = Z) vs. visits of TOI-836c would be needed to determine the
f (c1 , c2 , c3 , ..., cN ), where A and Z are constants less origin of this scatter.
than N). Therefore in order to compare all models, we
focus on the Bayes factor. Figure 10 shows the results 5.2. Fixed-Climate Spectral Models
of the three highest likelihood models (step function, The non-physical model results showcase the agree-
and NRS1 or NRS2 Gaussian + offset) for Eureka!, ment between all three data reduction methods. There-
ExoTiC-JEDI, and Aesop reductions, respectively, from fore, in this section, we show spectral models compared
left to right. The full numerical results are shown in Ta- against the nominal Eureka! reduction, without the
ble 3. Overall, the fitting results are consistent between addition of the engineering parameters. The ultimate
reductions within 1σ for each of the fit parameters. For goal here is to determine what range of mean molecu-
all reductions, the step function and either the NRS1 lar weights and “opaque pressure levels” can be ruled
15
(a) (b)
Figure 11. We show the transmission spectrum for the nominal Eureka! case (as detailed in Section 3.1) along with four
example models depicting 10×, 150×, and 1000× Solar metallicity, as well as a 100% H2 O model. In the legend, we also show
the χ2 /N and the corresponding σ at which the model can be confidently ruled out. On the left (b) we show the full parameter
space of opaque pressure level and metallicity. Here, the color scale of the heatmap similarly indicates at what σ each model
can be confidently-ruled out. Ultimately, we are unable to rule out metallicities >175× Solar (log[M/H]>2.2).
out. Here, we use metallicity as a proxy for tuning the H2 S, PH3 , C2 H2 , C2 H6 , Na, K, TiO, VO, F, H, H2, He)
mean molecular weight of the atmosphere. The “opaque at 26 metallicities (log [M/H]=-2–3), and 25 C/O ra-
pressure level” is a term adapted from the term “appar- tios (C/O = 0.01–2). We choose a fixed solar C/O ratio
ent surface pressure” that was used in Lustig-Yaeger & (0.55) in accordance with Asplund et al. (2009) because
Fu et al. 2023 and term “cloud top pressure” that is we are interested in leveraging metallicity as a proxy for
frequently used in previous studies of exoplanets (e.g. changes in mean molecular weight. While changes in
Kreidberg et al. 2014; Knutson et al. 2014). All these C/O can change the expected dominant molecular spec-
terms generally define a grey opacity below which our tral features, we have demonstrated in §5.1 that there
measurements are not able to probe. The simplest way are no statistically significant features in any of the data
to implement this is to imagine a grey optically thick reductions and thus an exploration of C/O variations is
cloud deck. not warranted. Here we are simply interested in deter-
For the fixed-temperature-pressure profile, we use a mining what range of mean molecular weights can be
1D, 5-parameter double-grey analytic formula (Guillot statistically ruled out. Lastly, we implement the cloud
2010), which for weakly irradiated systems approximates deck as an optically thick slab below a certain pressure
to T 4 ∼ 0.75 ∗ Teqt4
(p[bar] + 2/3)). This is similarly (varied between 1–10−4 bars). Given these three com-
adopted in several other works that aim to approxi- ponents (temperature-pressure profile, chemistry, cloud)
mate the climate structure when in-depth modeling is we use the PICASO radiative transfer code (Batalha et al.
not warranted by the precision of the data (e.g., Lustig- 2019) to compute transmission spectra between 3–5 µm
Yaeger & Fu et al. 2023; Moran & Stevenson et al. to compare to our NIRSpec G395H data. We rebin our
2023; May & MacDonald et al. 2023). Given the climate high resolution spectral models (R=60,000) to the wave-
structure, we compute atmospheric abundance profiles length grid shown in Figure 8.
assuming chemical equilibrium using the chemical equi- Figure 11 shows the result of this analysis, which
librium grid computed by Line et al. (2013) with NASA’s shows we can generally rule out atmospheric com-
CEA code (Gordon & McBride 1994). The chemistry positions with [M/H]< 175×Solar (mean molecular
grid is publicly available on GitHub as part of CHIMERA’s weight∼6). In Figure 11a we show three chemical equi-
open source code4 . The grid computes the abundances librium models (10×, 150×, and 1000×), all of which
of 19 molecules (H2 O, CH4 , CO, CO2 , NH3 , N2 , HCN, have a grey opacity at 0.1 bar. We additionally show
a 100% H2 O model, which cannot be ruled out, for ref-
4
erence. In this analysis we fit for detector offsets in-
https://github.com/mrline/CHIMERA
dividually to each model, before computing the χ2 /N .
16
However, in the figure we add these offsets directly to the which then settle downwards in the atmosphere under
data in order to better visualize the model. For a visual the effect of gravity and vertical mixing. We parameter-
representation of the offsets, see Figure 10. Shown in ize vertical mixing using eddy diffusion, with the eddy
Figure 11b is the full parameter space explored. Above diffusion coefficient derived for a 100× Solar metallicity
a metallicity of 175× Solar (mean molecular wight∼6) atmosphere from Charnay et al. (2015). As the parti-
we lose the ability to rule out physical cases at 3σ. This cles sediment, they are allowed to grow through colli-
result is relatively insensitive to the optical depth of the sions (coagulation), with the assumption that they re-
cloud slab until pressures <10−4 bar, where we are un- main spherical.
able to rule out even Solar metallicity cases. In the We compute the optical depth per layer of the result-
following section, we conduct more in-depth modeling ing model haze distributions assuming Mie scattering
to understand whether it is physically plausible to form and both the purely scattering and Titan tholin refrac-
aerosols in the atmosphere of TOI-836c at such low pres- tive indices (Khare et al. 1984) as in Gao et al. (2023).
sures. We then sum up the layer optical depths to arrive at the
nadir cumulative optical depth, shown in Figure 12 for
5.3. Haze Modeling with CARMA the different haze production rates and haze refractive
indices. To compute the slant optical depth, we use Eq.
Our ability to rule out metallicity cases less than
6 from Fortney et al. (2005) and assume a mean molecu-
175× is dependent on the location of an optically thick
lar weight of 4.25 g mol−1 for the 100× Solar metallicity
aerosol layer. Therefore, here we leverage a micro-
atmosphere. We find that column haze production rates
physical model (CARMA) to determine if an optically
>10−10 g cm−2 s−1 are needed to produce haze slant op-
thick aerosol layer is physically plausible below 10−4
tical depth unity pressure levels <10−4 bar. These haze
bar. First, we generated self-consistent climate models
production rates are at the upper limits of those pre-
in order to more closely investigate the physics of the
dicted by photochemical models that source haze from
planet’s atmosphere using PICASO 3.1 (Batalha et al.
the photolysis of atmospheric hydrocarbon and nitrogen
2019; Mukherjee et al. 2023). By “self-consistent,” we
species (Kawashima & Ikoma 2019; Lavvas et al. 2019),
mean that the model is initiated with a parametric
but are not implausible, and are on par with those pre-
pressure-temperature profile and chemical equilibrium
dicted for GJ 1214b based on its near-to-mid-infrared
abundances, then iterated until it achieves radiative-
transmission spectrum (Gao et al. 2023). As such, TOI-
convective equilibrium. In our models, we assume full
836c may have an atmosphere with similar chemistry to
day-to-night heat redistribution. For this use case,
GJ 1214b that is conducive to haze formation and an
PICASO supports climate calculations up to 100× So-
extremely high metallicity of ≥1000× Solar (Kempton
lar metallicity. We select the 100× Solar case as a rep-
et al. 2023).
resentative case to probe the physical plausibility for
high-altitude aerosols. 6. DISCUSSION
We use CARMA (Community Aerosol and Radiation
Model for Atmospheres; Turco et al. 1979; Toon et al. 6.1. Contextualizing TOI-836c
1988; Jacobson & Turco 1994; Ackerman et al. 1995) The only two sub-Neptune exoplanets with published
to simulate the optical depth of photochemical hazes JWST transmission spectra to date are GJ 1214b (8.17
in the atmosphere of TOI-836c, with the 100× Solar M⊕ , 2.628 R⊕ , 553K; Gao et al. 2023) and K2-18b (8.63
metallicity model atmosphere described in the previous M⊕ , 2.61 R⊕ , ∼250-300K; Madhusudhan et al. 2023).
paragraph as a nominal background. CARMA is a 1D TOI-836c (9.6 M⊕ , 2.587 R⊕ , 665K; Hawthorn et al.
aerosol microphysics code that accounts for the nucle- 2023) is more massive and hotter than both of these
ation, condensation, coagulation, evaporation, sedimen- planets, but has a smaller radius, meaning that it has the
tation, diffusion, and advection of cloud and haze par- highest gravity. Our observations of TOI-836c rule out
ticles in planetary atmospheres (Gao et al. 2018). Here atmospheric metallicities <175× Solar in the absence of
we apply it in a similar way as in Gao et al. (2023), clouds or hazes, which is similar to the high metallicities
where we compute the vertical and size distribution of inferred for both GJ 1214b and K2-18b. Transmission
photochemical hazes given a range of column haze pro- spectra of GJ 1214b from near- and mid-infrared ob-
duction rates (10−14 –10−9 g cm−2 s−1 as informed by servations (including the aforementioned JWST MIRI
sub-Neptune photochemical models, e.g. Kawashima & LRS observation), are consistent with a metallicity of
Ikoma 2019; Lavvas et al. 2019). Briefly, we assume the >300× Solar and prefer a metallicity of ≥ 1000× So-
production of 10 nm spherical “seed” particles with in- lar with a haze production rate of ≥ 10−10 g cm−2 s−1
ternal density of 1 g cm−3 at pressure levels <1 µbar, (Gao et al. 2023). Such a haze production rate is also
17
10 6 Purely Scattering and Titan Tholin Hazes ( =4 m) structure models is 75% to 85% water by mass (Nettel-
Titan Tholin
10 5 Purely Scattering mann et al. 2013; Podolak et al. 2019). These values
10 9 g/cm 2/s
2
correspond to average metallicities of the envelope of
10 4 10 10 g/cm2/s
10 11 g/cm2/s 200–380× Solar. The existence of a two-layer envelope
10 12 g/cm2/s
Pressure (bar)
both K2-18b and GJ 1214b show direct evidence for towards a potential population of water worlds orbiting
molecular features. Both the dayside and nightside spec- M dwarfs (see Luque & Pallé (2022), although Rogers
tra of GJ 1214b are inconsistent with blackbody expec- et al. (2023b) provide an alternative explanation). In re-
tations at >3σ (Kempton et al. 2023), while K2-18b ality, all intermediate compositions are also possible. If
shows direct evidence of CH4 and CO2 . However, TOI- the atmosphere and the envelope have a similar metallic-
836c is both hotter and has a higher gravity than these ity, then an atmospheric metallicity of 175× Solar could
other two sub-Neptunes, possibly explaining the lack of corresponds to an interior that is 1.6 wt% H2 -He, 4.1
detected features at current S/N in the transmission wt% H2 O, and 94.3 wt% of an Earth-like core. Atmo-
spectrum of this planet. To further contextualize this sphere molecular detections could narrow the range of
planet, we next model the bulk composition of TOI- possible bulk composition even further with information
836c to determine its likelihood as a water world as has on elemental ratios such as C/O or C/N.
been previously argued for K2-18b.
As our NIRSpec observations are unable to place non- We present the first JWST NIRSpec G395H tran-
degenerate constraints on the atmosphere of TOI-836c, sit observations of the sub-Neptune TOI-836c, the first
we instead place model-dependent limits on its struc- planet observed as part of the COMPASS program.
ture. To infer the bulk composition of TOI-836c, we Using three independent reductions, we find a consis-
use the open source code smint5 developed by Piaulet tently featureless transmission spectrum. Our spectrum
et al. (2021) to perform a MCMC retrieval of planetary is able to rule out a clear low metallicity atmosphere (<
bulk compositions from pre-computed grids of theoreti- 175× Solar), as well as low metallicity atmospheres with
cal interior structure models. We consider two possible clouds and hazes at pressures greater than ∼0.1-1 mbar.
compositions for the interior: 1) an Earth-like core with However, the planet likely has a sizable H2 /He envelope
a H2 -He envelope of Solar metallicity (Lopez & Fortney as inferred from its mass and radius, and we show that
2014), and 2) a refractory core with a variable core mass the presence of clouds and hazes at low pressures cannot
fraction and a pure H2 O envelope and atmosphere on be ruled out, as in the case of GJ 1214b, which has a
top (Aguichine et al. 2021). These compositions repre- similar equilibrium temperature and mass.
sent end-member cases between an envelope that would While our median measured transit depth precisions
form with a Solar-like composition, and a H2 -He free en- are only ∼1.13-1.18× those expected from PandExo for
velope. While our observations do rule out a clear Solar our three reductions, we do find a significant amount
metallicity atmosphere, a Solar metallicity end-member of residual correlated noise in the white light curves.
case is still possible given sufficient aerosol opacity. Our We also find that the mid-transit times derived from
goal is to determine the range of possible bulk volatile the two different NIRSpec G395H detectors are incon-
contents in TOI-836c in the absence of such observa- sistent at >4σ for this single simultaneous transit ob-
tional constraints. servation. We include mnemonics from the JWST En-
We find that TOI-836c can have an envelope mass gineering Database in our systematic noise model and
fraction of 1.74+0.55
−0.48 % in the case of Solar metallicity
find that their inclusion can reduce both the residual
gas, or a water mass fraction of 52+15 −14 % in the pure
correlated noise and the timing discrepancy to within
H2 O case (see Figure 13). Detailed results and meth- 2.3σ, but does not fully remedy the offset or entirely re-
ods to produce Figure 13 are provided in Appendix A. move the residual correlated noise. Nonetheless, further
A 50% water and 50% rocky composition is interesting investigation of the use of mnemonics from the JWST
as it represents the bulk composition of the solid build- Engineering Database may help to reduce residual cor-
ing blocks of planets (dust, pebbles, and planetesimals) related noise in NIRSpec G395H observations of small
when all the icy material has condensed, i.e. in the cold- planets around bright stars.
est region of the protoplanetary disk (Zeng et al. 2019). While the results presented herein are only for the
Despite the fact that most interior structure models use first planet observed as part of the COMPASS program,
water as a proxy for all volatiles, the 50% water and 50% these results will be included in the entire COMPASS
rock models match the masses and radii of icy moons of sample of 12 super-Earths and sub-Neptunes, which will
the Solar System (e.g. Sotin et al. 2007) and also point be presented in forthcoming publications. Once the sur-
vey is complete, we will have a better understanding of
the observability and compositions of the atmospheres
5 https://github.com/cpiaulet/smint of the most common type of planet in the Galaxy and
19
80 4
Radius (R )
TOI-836c TOI-836c
K2-18b 60 K2-18b 3
50% Steam Water
Like 1% H2-He Like
2 th- 40 2 th- 2
Ear Ear
ater
id W
Liqu
50% 20 1
1 0 1 0
1 2 5 10 15 20 1 2 5 10 15 20
Mass (M ) Mass (M )
Figure 13. Mass-radius diagram of small planets colored by bulk water mass fraction (left) and bulk H2 -He mass fraction
(right). We show the sub-Neptunes with JWST observations (TOI-836c, GJ 1214b, and K2-18b) as stars and planets from the
NASA Exoplanet Archive with mass errors of <20% and radius errors of <5% as circles. Zeng et al. (2016) curves are used for
the Earth-like and 50% liquid water compositions. Aguichine et al. (2021) curves are used for the 50% and 100% steam water
composition, and are shown for Teq = 600 K and an Earth-like core. Lopez & Fortney (2014) curves are used for the 1%, 2%
and 5% H2 -He composition, and are shown for Teq = 600 K and an age of 5 Gyr. For each planet, the bulk volatile content
(H2 O or H2 -He) has been determined with the method described in Appendix A.
provide crucial context for understanding the composi- ber 19-ICAR19 2-0041. We acknowledge use of the lux
tion of planets in our own Solar System. supercomputer at UC Santa Cruz, funded by NSF MRI
grant AST 1828315. This research has made use of the
This work is based on observations made with the NASA Exoplanet Archive, which is operated by the Cal-
NASA/ESA/CSA James Webb Space Telescope. The ifornia Institute of Technology, under contract with the
data were obtained from the Mikulski Archive for Space National Aeronautics and Space Administration under
Telescopes at the Space Telescope Science Institute, the Exoplanet Exploration Program. This paper makes
which is operated by the Association of Universities use of data from the first public release of the WASP
for Research in Astronomy, Inc., under NASA contract data (Butters et al. 2010) as provided by the WASP con-
NAS 5-03127 for JWST. These observations are associ- sortium and services at the NASA Exoplanet Archive,
ated with program #2512. Support for program #2512 which is operated by the California Institute of Technol-
was provided by NASA through a grant from the Space ogy, under contract with the National Aeronautics and
Telescope Science Institute, which is operated by the Space Administration under the Exoplanet Exploration
Association of Universities for Research in Astronomy, Program.
Inc., under NASA contract NAS 5-03127. This work is Co-Author contributions are as follows: NLW led the
funded in part by the Alfred P. Sloan Foundation un- data analysis of this study. NEB led the atmospheric
der grant G202114194. Support for this work was pro- modeling efforts with contributions from NS and PG.
vided by NASA through grant 80NSSC19K0290 to JT LA, MKA, and JIAR provided additional reductions and
and NW. This work benefited from the 2022 and 2023 analyses of the data. AA provided interior models. All
Exoplanet Summer Program in the Other Worlds Lab- authors provided detailed comments and conversations
oratory (OWL) at the University of California, Santa that greatly improved the quality of the manuscript.
Cruz, a program funded by the Heising-Simons Foun-
Software: CARMA (Gao et al. 2023), emcee
dation. This material is based upon work supported
(Foreman-Mackey et al. 2013), Eureka! (Bell et al. 2022),
by NASA’S Interdisciplinary Consortia for Astrobiology
ExoTiC-Jedi (Alderson et al. 2022), ExoTiC-LD (Grant
Research (NNH19ZDA001N-ICAR) under award num-
20
& Wakeford 2022), Matplotlib (Hunter 2007), NumPy (Piaulet et al. 2021), Tiberius (Kirk et al. 2017, 2021),
(Harris et al. 2020), PandExo (Batalha et al. 2017), (Vir- ultranest(Buchner 2021)
tanen et al. 2020), PICASO (Batalha et al. 2019), SMINT
REFERENCES
Ackerman, A. S., Toon, O. B., & Hobbs, P. V. 1995, Burn, R., Schlecker, M., Mordasini, C., et al. 2021, A&A,
J. Geophys. Res., 100, 7121, doi: 10.1029/95JD00026 656, A72, doi: 10.1051/0004-6361/202140390
Aguichine, A., Mousis, O., Deleuil, M., & Marcq, E. 2021, Bushouse, H., Eisenhamer, J., Dencheva, N., et al. 2022,
ApJ, 914, 84, doi: 10.3847/1538-4357/abfa99 spacetelescope/jwst: JWST 1.6.2, 1.6.2, Zenodo,
Alderson, L., & COMPASS et al. 2024, AAS Journals, doi: 10.5281/zenodo.6984366
accepted Butters, O. W., West, R. G., Anderson, D. R., et al. 2010,
Alderson, L., Grant, D., & Wakeford, H. 2022, A&A, 520, L10, doi: 10.1051/0004-6361/201015655
Exo-TiC/ExoTiC-JEDI: v0.1-beta-release, v0.1, Zenodo, Charnay, B., Meadows, V., & Leconte, J. 2015, ApJ, 813, 15
Zenodo, doi: 10.5281/zenodo.7185855 Claret, A. 2000, A&A, 363, 1081
Alderson, L., Wakeford, H. R., Alam, M. K., et al. 2023, Cloutier, R., Charbonneau, D., Deming, D., Bonfils, X., &
Nature, 614, 664, doi: 10.1038/s41586-022-05591-3 Astudillo-Defru, N. 2021, AJ, 162, 174,
Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009, doi: 10.3847/1538-3881/ac1584
ARA&A, 47, 481, Cloutier, R., Astudillo-Defru, N., Doyon, R., et al. 2019,
doi: 10.1146/annurev.astro.46.060407.145222 A&A, 621, A49, doi: 10.1051/0004-6361/201833995
Atreya, S. K., Hofstadter, M. H., In, J. H., et al. 2020, de Wit, J., Wakeford, H. R., Gillon, M., et al. 2016, Nature,
SSRv, 216, 18, doi: 10.1007/s11214-020-0640-8 537, 69, doi: 10.1038/nature18641
de Wit, J., Wakeford, H. R., Lewis, N. K., et al. 2018,
Barclay, T., Sheppard, K. B., Latouf, N., et al. 2023, arXiv
Nature Astronomy, 2, 214,
e-prints, arXiv:2301.10866,
doi: 10.1038/s41550-017-0374-z
doi: 10.48550/arXiv.2301.10866
Diamond-Lowe, H., Charbonneau, D., Malik, M., Kempton,
Batalha, N., Teske, J., Alam, M., et al. 2021, Seeing the
E. M. R., & Beletsky, Y. 2020, AJ, 160, 188,
Forest and the Trees: Unveiling Small Planet
doi: 10.3847/1538-3881/abaf4f
Atmospheres with a Population-Level Framework, JWST
Diamond-Lowe, H., Mendonça, J. M., Charbonneau, D., &
Proposal. Cycle 1, ID. #2512
Buchhave, L. A. 2023, AJ, 165, 169,
Batalha, N. E., Marley, M. S., Lewis, N. K., & Fortney,
doi: 10.3847/1538-3881/acbf39
J. J. 2019, ApJ, 878, 70, doi: 10.3847/1538-4357/ab1b51
Edwards, B., Changeat, Q., Mori, M., et al. 2021, AJ, 161,
Batalha, N. E., Wolfgang, A., Teske, J., et al. 2023, AJ,
44, doi: 10.3847/1538-3881/abc6a5
165, 14, doi: 10.3847/1538-3881/ac9f45
Elliot, J. L., & Nicholson, P. D. 1984, in IAU Colloq. 75:
Batalha, N. E., Mandell, A., Pontoppidan, K., et al. 2017,
Planetary Rings, ed. R. Greenberg & A. Brahic, 25–72
PASP, 129, 064501, doi: 10.1088/1538-3873/aa65b0
Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman,
Batalha, N. M., Rowe, J. F., Bryson, S. T., et al. 2013, J. 2013, Publications of the Astronomical Society of the
ApJS, 204, 24, doi: 10.1088/0067-0049/204/2/24 Pacific, 125, 306, doi: 10.1086/670067
Bell, T. J., Ahrer, E.-M., Brande, J., et al. 2022, Journal of Fortney, J. J., Marley, M. S., Lodders, K., Saumon, D., &
Open Source Software, 7, 4503, doi: 10.21105/joss.04503 Freedman, R. 2005, ApJ, 627, L69, doi: 10.1086/431952
Benneke, B., Wong, I., Piaulet, C., et al. 2019, ApJL, 887, Fortney, J. J., Mordasini, C., Nettelmann, N., et al. 2013,
L14, doi: 10.3847/2041-8213/ab59dc ApJ, 775, 80, doi: 10.1088/0004-637X/775/1/80
Buchner, J. 2016, Statistics and Computing, 26, 383, Fulton, B. J., & Petigura, E. A. 2018, AJ, 156, 264,
doi: 10.1007/s11222-014-9512-y doi: 10.3847/1538-3881/aae828
—. 2019, PASP, 131, 108005, Gao, P., Marley, M. S., & Ackerman, A. S. 2018, ApJ, 855,
doi: 10.1088/1538-3873/aae7fc 86, doi: 10.3847/1538-4357/aab0a1
—. 2021, The Journal of Open Source Software, 6, 3001, Gao, P., Piette, A. A. A., Steinrueck, M. E., et al. 2023,
doi: 10.21105/joss.03001 ApJ, 951, 96, doi: 10.3847/1538-4357/acd16f
Burn, R., Mordasini, C., Mishra, L., et al. 2024, Nature Garcia, L. J., Moran, S. E., Rackham, B. V., et al. 2022,
Astronomy, doi: 10.1038/s41550-023-02183-7 A&A, 665, A19, doi: 10.1051/0004-6361/202142603
21
Ginzburg, S., Schlichting, H. E., & Sari, R. 2018, MNRAS, Kreidberg, L. 2015, Publications of the Astronomical
476, 759, doi: 10.1093/mnras/sty290 Society of the Pacific, 127, 1161, doi: 10.1086/683602
Gordon, S., & McBride, B. J. 1994, Computer program for Kreidberg, L., Bean, J. L., Désert, J.-M., et al. 2014,
calculation of complex chemical equilibrium compositions Nature, 505, 69, doi: 10.1038/nature12888
and applications. Part 1: Analysis, Tech. rep. Lambrechts, M., & Johansen, A. 2012, A&A, 544, A32,
Grant, D., & Wakeford, H. R. 2022, Exo-TiC/ExoTiC-LD: doi: 10.1051/0004-6361/201219127
ExoTiC-LD v3.0.0, v3.0.0, Zenodo, Zenodo, Lambrechts, M., Johansen, A., & Morbidelli, A. 2014,
doi: 10.5281/zenodo.7437681 A&A, 572, A35, doi: 10.1051/0004-6361/201423814
Guillot, T. 2010, A&A, 520, A27, Lavvas, P., Koskinen, T., Steinrueck, M. E., Garcı́a Muñoz,
doi: 10.1051/0004-6361/200913396 A., & Showman, A. P. 2019, ApJ, 878, 118,
Guilluy, G., Gressier, A., Wright, S., et al. 2020, The doi: 10.3847/1538-4357/ab204e
Astronomical Journal, 161, 19, Lee, E. J., Karalis, A., & Thorngren, D. P. 2022, ApJ, 941,
doi: 10.3847/1538-3881/abc3c8 186, doi: 10.3847/1538-4357/ac9c66
Guo, X., Crossfield, I. J. M., Dragomir, D., et al. 2020, AJ, Libby-Roberts, J. E., Berta-Thompson, Z. K.,
159, 239, doi: 10.3847/1538-3881/ab8815 Diamond-Lowe, H., et al. 2022, AJ, 164, 59,
Gupta, A., & Schlichting, H. E. 2019, MNRAS, 487, 24, doi: 10.3847/1538-3881/ac75de
doi: 10.1093/mnras/stz1230 Line, M. R., Wolf, A. S., Zhang, X., et al. 2013, ApJ, 775,
Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 137, doi: 10.1088/0004-637X/775/2/137
2020, Nature, 585, 357, doi: 10.1038/s41586-020-2649-2
Lopez, E. D., & Fortney, J. J. 2013, ApJ, 776, 2,
Hawthorn, F., Bayliss, D., Wilson, T. G., et al. 2023,
doi: 10.1088/0004-637X/776/1/2
MNRAS, 520, 3649, doi: 10.1093/mnras/stad306
—. 2014, ApJ, 792, 1, doi: 10.1088/0004-637X/792/1/1
Hu, R., Damiano, M., Scheucher, M., et al. 2021, ApJL,
Luque, R., & Pallé, E. 2022, Science, 377, 1211,
921, L8, doi: 10.3847/2041-8213/ac1f92
doi: 10.1126/science.abl7164
Hunter, J. D. 2007, Computing in Science & Engineering, 9,
Lustig-Yaeger, J., Fu, G., May, E. M., et al. 2023, Nature
90, doi: 10.1109/MCSE.2007.55
Astronomy, doi: 10.1038/s41550-023-02064-z
Jacobson, M. Z., & Turco, R. P. 1994, Atmospheric
Madhusudhan, N., Sarkar, S., Constantinou, S., et al. 2023,
Environment, 28, 1327
arXiv e-prints, arXiv:2309.05566,
Jacobson, R. A. 2009, in AAS/Division of Dynamical
doi: 10.48550/arXiv.2309.05566
Astronomy Meeting, Vol. 40, AAS/Division of Dynamical
Magic, Z., Chiavassa, A., Collet, R., & Asplund, M. 2015,
Astronomy Meeting #40, 8.01
A&A, 573, A90, doi: 10.1051/0004-6361/201423804
Jacobson, R. A., Antreasian, P. G., Bordi, J. J., et al. 2006,
Mandel, K., & Agol, E. 2002, The Astrophysical Journal,
AJ, 132, 2520, doi: 10.1086/508812
580, L171, doi: 10.1086/345520
Kawashima, Y., & Ikoma, M. 2019, ApJ, 877, 109,
doi: 10.3847/1538-4357/ab1b1d May, E. M., MacDonald, R. J., Bennett, K. A., et al. 2023,
Kempton, E. M. R., Zhang, M., Bean, J. L., et al. 2023, arXiv e-prints, arXiv:2310.10711,
Nature, 620, 67, doi: 10.1038/s41586-023-06159-5 doi: 10.48550/arXiv.2310.10711
Khare, B. N., Sagan, C., Arakawa, E. T., et al. 1984, Mikal-Evans, T., Crossfield, I. J. M., Benneke, B., et al.
Icarus, 60, 127, doi: 10.1016/0019-1035(84)90142-8 2021, AJ, 161, 18, doi: 10.3847/1538-3881/abc874
Kirk, J., Wheatley, P. J., Louden, T., et al. 2017, MNRAS, Moran, S. E., Stevenson, K. B., Sing, D. K., et al. 2023,
468, 3907, doi: 10.1093/mnras/stx752 ApJL, 948, L11, doi: 10.3847/2041-8213/accb9c
Kirk, J., Rackham, B. V., MacDonald, R. J., et al. 2021, Mousis, O., Deleuil, M., Aguichine, A., et al. 2020, ApJL,
AJ, 162, 34, doi: 10.3847/1538-3881/abfcd2 896, L22, doi: 10.3847/2041-8213/ab9530
Knutson, H. A., Dragomir, D., Kreidberg, L., et al. 2014, Mousis, O., Atkinson, D. H., Cavalié, T., et al. 2018,
ApJ, 794, 155, doi: 10.1088/0004-637X/794/2/155 Planet. Space Sci., 155, 12, doi: 10.1016/j.pss.2017.10.005
Kopparapu, R. K., Ramirez, R., Kasting, J. F., et al. 2013, Mukherjee, S., Batalha, N. E., Fortney, J. J., & Marley,
ApJ, 765, 131, doi: 10.1088/0004-637X/765/2/131 M. S. 2023, The Astrophysical Journal, 942, 71,
Kostogryz, N., Shapiro, A., Witzke, V., et al. 2023, doi: 10.3847/1538-4357/ac9f48
Research Notes of the AAS, 7, 39 Nettelmann, N., Helled, R., Fortney, J. J., & Redmer, R.
Kostogryz, N., Witzke, V., Shapiro, A., et al. 2022, arXiv 2013, Planet. Space Sci., 77, 143,
preprint arXiv:2206.06641 doi: 10.1016/j.pss.2012.06.019
22
Owen, J. E., & Wu, Y. 2013, ApJ, 775, 105, Toon, O. B., Turco, R. P., Westphal, D., Malone, R., & Liu,
doi: 10.1088/0004-637X/775/2/105 M. S. 1988, Journal of the Atmospheric Sciences, 45, 2123
Piaulet, C., Benneke, B., Rubenzahl, R. A., et al. 2021, AJ, Trotta, R. 2008, Contemporary Physics, 49, 71,
161, 70, doi: 10.3847/1538-3881/abcd3c doi: 10.1080/00107510802066753
Tsiaras, A., Rocchetto, M., Waldmann, I. P., et al. 2016,
Piaulet, C., Benneke, B., Almenara, J. M., et al. 2022,
ApJ, 820, 99, doi: 10.3847/0004-637X/820/2/99
Nature Astronomy, doi: 10.1038/s41550-022-01835-4
Turbet, M., Ehrenreich, D., Lovis, C., Bolmont, E., &
Pierrehumbert, R. T. 2023, ApJ, 944, 20, Fauchez, T. 2019, A&A, 628, A12,
doi: 10.3847/1538-4357/acafdf doi: 10.1051/0004-6361/201935585
Podolak, M., Helled, R., & Schubert, G. 2019, MNRAS, Turco, R. P., Hamill, P., Toon, O. B., Whitten, R. C., &
487, 2653, doi: 10.1093/mnras/stz1467 Kiang, C. S. 1979, Journal of the Atmospheric Sciences,
Pont, F., Zucker, S., & Queloz, D. 2006, MNRAS, 373, 231, 36, 699
doi: 10.1111/j.1365-2966.2006.11012.x Valencia, D., Sasselov, D. D., & O’Connell, R. J. 2007,
ApJ, 665, 1413, doi: 10.1086/519554
Raymond, S. N., Boulet, T., Izidoro, A., Esteves, L., &
Van Eylen, V., Agentoft, C., Lundkvist, M. S., et al. 2018,
Bitsch, B. 2018, MNRAS, 479, L81,
MNRAS, 479, 4786, doi: 10.1093/mnras/sty1783
doi: 10.1093/mnrasl/sly100
Venturini, J., Guilera, O. M., Haldemann, J., Ronco, M. P.,
Ricker, G. R., Winn, J. N., Vanderspek, R., et al. 2015,
& Mordasini, C. 2020, A&A, 643, L1,
Journal of Astronomical Telescopes, Instruments, and
doi: 10.1051/0004-6361/202039141
Systems, 1, 014003, doi: 10.1117/1.JATIS.1.1.014003 Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020,
Rogers, J. G., Owen, J. E., & Schlichting, H. E. 2023a, Nature Methods, 17, 261, doi: 10.1038/s41592-019-0686-2
arXiv e-prints, arXiv:2311.12295, Wakeford, H. R., Lewis, N. K., Fowler, J., et al. 2019, AJ,
doi: 10.48550/arXiv.2311.12295 157, 11, doi: 10.3847/1538-3881/aaf04d
Rogers, J. G., Schlichting, H. E., & Owen, J. E. 2023b, Wogan, N. F., Batalha, N. E., Zahnle, K., et al. 2024, arXiv
ApJL, 947, L19, doi: 10.3847/2041-8213/acc86f e-prints, arXiv:2401.11082,
Rogers, L. A., & Seager, S. 2010, ApJ, 712, 974, doi: 10.48550/arXiv.2401.11082
Zeng, L., Sasselov, D. D., & Jacobsen, S. B. 2016, ApJ, 819,
doi: 10.1088/0004-637X/712/2/974
127, doi: 10.3847/0004-637X/819/2/127
Rustamkulov, Z., Sing, D. K., Mukherjee, S., et al. 2023,
Zeng, L., Jacobsen, S. B., Hyung, E., et al. 2017, in 48th
Nature, 614, 659, doi: 10.1038/s41586-022-05677-y
Annual Lunar and Planetary Science Conference, Lunar
Sotin, C., Grasset, O., & Mocquet, A. 2007, Icarus, 191, and Planetary Science Conference, 1576
337, doi: 10.1016/j.icarus.2007.04.006 Zeng, L., Jacobsen, S. B., Sasselov, D. D., et al. 2019,
Sromovsky, L. A., Karkoschka, E., Fry, P. M., de Pater, I., Proceedings of the National Academy of Science, 116,
& Hammel, H. B. 2019, Icarus, 317, 266, 9723, doi: 10.1073/pnas.1812905116
doi: 10.1016/j.icarus.2018.06.026 Zhou, L., Ma, B., Wang, Y.-H., & Zhu, Y.-N. 2023,
Tollefson, J., de Pater, I., Luszcz-Cook, S., & DeBoer, D. Research in Astronomy and Astrophysics, 23, 025011,
doi: 10.1088/1674-4527/acaceb
2019, AJ, 157, 251, doi: 10.3847/1538-3881/ab1fdf
23
APPENDIX
8 12 16
Mp (M )
12.93
3.21
8 12 16 50 00 50 00 50
5 6 6 7 7
4 6 8
Age (Gyr)
Tirr (K)
Mp (M ) = 9.96+1.03
0.97
Mp (M ) = 9.52+2.64
2.61 Sinc (S ) = 26.707+0.003
2
0.003
6 8 10 12 14
Mp (M )
.69 .70 .71 .72
Mp (M )
26 26 Sinc26 (S 26)
4
0
25
50
75
6
8
10
12
14
0.2
0.4
0.6
0.8
0.2
0.4
0.6
0.8
0
0
0
0
0
4
8
12
16
0.8
0.0
0.8
1.6
4
8
12
16
2
4
6
8
26 9
26 0
26 1
.72
55
60
65
70
75
.6
.7
.7
fH2O (%) Mp (M )
26
′
Figure A1. Results from interior structure modeling with smint. Left panel: fcore , fH2 O , Teq and Mp for TOI-836c using the
model from Aguichine et al. (2021). Middle panel: fenv , Mp , age and Sinc for TOI-836c using the model from Lopez & Fortney
(2014). Right panel: fH2 O and Mp for K2-18b using the model from Zeng et al. (2016).