JWST/NIRCam Imaging of Young Stellar Objects. II. Deep Constraints on Giant Planets and a Planet Candidate Outside of the Spiral Disk Around SAO 206462

Brown et al. (2007) studied the SED of the star and inferred the presence of a disk with a dust cavity at ∼45 au. The first resolved detection of the disk surrounding the star was presented in Grady et al. (2009) and was obtained with HST data. HST only detected the outer regions of the disk, without revealing any substructures. Subsequent polarimetric observations with Subaru/HiCIAO (Muto et al. 2012), VLT/NaCo (Garufi et al. 2013), and VLT/SPHERE (Stolker et al. 2016) revealed a 28 au small dust grain cavity and two spiral arms extending to 06 (∼80 au, Stolker et al. 2016). The spirals show a brighter peak southwest from the star associated with an emission clump (Bae et al. 2016). Furthermore, the scattered light images presented in Stolker et al. (2016) reveal shadow features that the authors associated to a misaligned inner disk. Even though the analysis of Bohn et al. (2022) did not confirm this hypothesis, the authors indicate that the inner disk morphology of SAO 206462 is likely very complex and could not be captured by their model.

The cavity detected in scattered light was also identified by ALMA continuum observations (Pérez et al. 2014; Pinilla et al. 2015; Francis & van der Marel 2020) tracing sub-mm and mm pebbles, but with a larger size of 48 au. The difference in cavity size can be explained by spatial segregation of mm-sized dust grains (Garufi et al. 2013). The gas in the disk has been studied by van der Marel et al. (2016), who found that the cavity in ¹³CO and C¹⁸O is smaller in size, ∼30 au, and the ¹²CO surface density drop at the cavity edge is 4 orders of magnitude. In addition to the inner cavity, ALMA continuum images detected two rings at 52 and 80 au connected by a filament coincident with the Southern spiral and possibly tracing a planetary wake crossing the dust gap (Casassus et al. 2021).

In recent years, it has been proposed that monitoring of the spiral motion can provide constraints on its origin (GI; planet-disk interaction) and, in case of the latter scenario, of the separation of the perturber (see Ren et al. 2020, for an example applied to MWC758). Xie et al. (2021) studied the motion of the spirals in scattered light over a 1 yr period, trying to determine if they are comoving or if they have distinct dynamics. They found that the spiral rotation rate is not consistent with the gravitational instability scenario, as they move at sub-keplerian velocity with respect to keplerian material at their location. Instead, the velocities suggest the spirals move independently from each other (3σ confidence) and the eastern one is consistent with being launched by a companion orbiting with a period of 1130 ± 780 yr (corresponding to a semimajor axis of ${123}_{-18}^{+63}$ au ¹⁰ ). assuming a circular orbit coplanar with the disk. Notably, this method does not provide constraints on the planet mass.

Another key result from analyzing the dynamics of SAO 206462 is that the spirals should not have formed due to gravitational instability: Xie et al. (2021) estimated that in order for GI to be responsible for the spirals, the central object mass would need to be ${0.1}_{-0.05}^{+0.08}\,{M}_{\odot }$, inconsistent with the mass of the star (1.6 ± 0.1, Garufi et al. 2018). Another method to assess the disk potential for GI relies on estimating the total disk mass. Assuming optically thin emission, the observed millimeter continuum flux from the disk corresponds to a dust mass of 136 M_⊕, or a total-disk-to-star-mass ratio of 2.5% assuming a gas-to-dust ratio of 100 (Dong et al. 2018). This is about an order of magnitude lower than needed for GI to produce prominent two-arm spirals (e.g., Dong et al. 2015). Thus the disk is unlikely to be GI unstable, unless the millimeter emission is significantly optically thick, and/or the gas-to-dust ratio is substantially above 100.

Hydrodynamical simulations of the SAO 206462 disk allow constraints on the mass and location of the planet generating the detected spirals. Using their scaling relation between star–planet mass ratio q and azimuthal separation of the primary and secondary spirals ϕ_sep, Fung & Dong (2015) found the planet mass to be 5.4 ± 1.6 M_J, where we used the updated stellar mass from Table 1. Bae et al. (2016) performed 2D two-fluids hydrodynamical simulations, confirming that a gas giant planet could be responsible for the observations of both the spiral arms in scattered light and the asymmetric thermal emission observed in ALMA continuum. They inferred the mass of the perturber to be of the order of 10–15 M_J to reproduce the observed features. Finally, Dong & Fung (2017) used the contrast ratio of the spirals to constrain the planet mass, finding that M_p ∼ 5–10 M_J (assuming a planet at ∼100 au). In addition, the presence of only two spirals suggest that M_p ≳ 5.1 M_J (Bae & Zhu 2018a, 2018b). All these works agree that the necessary planetary mass to launch both spirals must be at least ∼5 M_J at 100–120 au. Assuming the spiral-driving companion is on a circular and coplanar orbit, and the disk has reached a steady state response to the companion (i.e., the companion did not form in the recent past), more massive companions could be located at larger separations from the spirals, because a massive companion opens a bigger gap around its orbit (Dong et al. 2016a).

SAO 206462 has been observed with the high-contrast imagers GPI, SPHERE and NaCo to search for the thermal emission from planets that are predicted to be responsible for the spirals (Vicente et al. 2011; Wahhaj et al. 2015; Maire et al. 2017; Asensio-Torres et al. 2021). So far no gas giant has been detected, and Asensio-Torres et al. (2021) reported SPHERE/IRDIS contrast limits (ΔK = 14 mag) that exclude planets with M_p > 4 M_J beyond 100 au when using hot-start models (AMES-DUSTY, Chabrier et al. 2000) and M_p > 11–12 M_J when using warm-start models (Linder et al. 2019; Marleau et al. 2019, BEX-WARM) and assuming no extinction affects the planet emission. These were the deepest detection limits available until now. Another mechanism to search for forming planets is to detect localized Hα emission associated with accretion processes onto the planet. SAO 206462 was part of the Hα surveys presented in Zurlo et al. (2020), Follette et al. (2023), and Cugno et al. (2019). The latter obtained an Hα luminosity upper limit of 8.5 × 10⁻⁷ L_⊙ with SPHERE/ZIMPOL.

We observed SAO 206462 as part of the NIRCam GTO program (PID 1179, PI: J. Leisenring). The program targets five YSOs showing strong signposts of ongoing planet formation. The SAO 206462 observations were executed on UT2023-02-16 in subarray imaging mode (SUB160) using four different filters of the NIRCam instrument (Rieke et al. 2023): F187N, F200W, F405N, and F410M (see Table 2 for details). The narrow filters are centered on the Pa-α and Br-α emission lines expected to contribute significantly to the protoplanet emitted flux in case it is accreting material, while the medium and wide filters are much broader and focus on detecting thermal continuum emission from a substellar companion. Observations were obtained simultaneously for the F187N/F405N and the F200W/F410M filter combinations, increasing overall observing efficiency. For each filter pair, we observed the target at two roll angles, separated by ∼10°, enabling angular differential imaging (Marois et al. 2006). For each position angle, we placed the star at four different dither positions to mitigate the effect of bad pixels. To minimize wave front variations, observations with the two rolls were executed one after the other. Indeed a variation of the wave front could lead to slightly different point spread functions (PSFs), increasing the stellar residuals in the final images and reducing the achieved contrast. No reference star was observed with the intent of performing reference differential imaging (RDI). Table 2 reports technical details for the observations.

Table 2. Summary of Observations

Target	Prog. ID	Filter	λpivot	Weff a	Readout	Subarray b	Ngr	Nint	Ndither	Nroll c	ttot	FWHM d
			(μm)	(μm)							(s)	('')
SAO 206462	1179	F187N	1.874	0.024	RAPID	SUB160	10	120	4	2	2674	0064
SAO 206462	1179	F200W	1.990	0.461	RAPID	SUB160	10	120	4	2	2674	0066
SAO 206462	1179	F405N	4.055	0.046	RAPID	SUB160	10	120	4	2	2674	0136
SAO 206462	1179	F410M	4.092	0.436	RAPID	SUB160	10	120	4	2	2674	0137
P330-E	1538	F187N	1.874	0.024	RAPID	SUB160	7	2	4	1	15.6	0064
P330-E	1538	F200W	1.990	0.461	RAPID	SUB160	3	2	4	1	6.7	0066
P330-E	1538	F405N	4.055	0.046	RAPID	SUB160	10	2	4	1	22.3	0136
P330-E	1538	F410M	4.092	0.436	RAPID	SUB160	3	2	4	1	6.7	0137

Notes.

^a Filter bandwidth, defined as the integral of the normalized transmission curve (https://jwst-docs.stsci.edu/jwst-near-infrared-camera/nircam-instrumentation/nircam-filters). ^b SUB160 has a group time of t_gr of 0.27864 s. Each integration length was t_gr × N_gr. ^c Number of spacecraft rolls in which the observations is repeated. ^d Empirical PSF full width at half maximum (FWHM) provided in JWST documentation (https://jwst-docs.stsci.edu/jwst-near-infrared-camera/nircam-performance/nircam-point-spread-functions).

Download table as:  ASCIITypeset image

As described later, the PSF core of the SAO 206462 data was heavily saturated in every filter. Hence, we used the standard G star P330-E to perform photometric calibration of our data. P330-E was observed as part of the NIRCam calibration program (Prog. ID 1538) on 2022 August 29 for all available NIRCam filters and detector combinations with the SUB160 subarray. The star was placed at the four corners of the field of view (FoV), but the PSF core necessary for the photometric calibration was always well within the detector. Table 2 reports details of these observations as well.

Our data reduction starts with the _uncal.fits frames downloaded from the Mikulski Archive for Space Telescope. Images underwent the standard Stage 1 and Stage 2 of the jwst pipeline (version 1.9.4 with crds version 11.16.20) publicly available, which performed all the necessary calibration steps. ¹¹ Following the directives in Carter et al. (2023), we switched off dark current correction and we changed the detection threshold of jumps in the ramp fitting to 5. After noticing the strong saturation in the central core of the stellar PSFs, we set the parameter ramp_fit.suppress_one_group to False, in order to minimize the area where no data is available. Despite this intervention, the central 02–03 of the images are already saturated within the first read, and therefore those pixels are flagged as NaNs.

In addition, the jwst pipeline flags pixels presenting jumps as NaNs in the final images. To remove image NaNs not connected with the PSF core, for each dither position of the two rolls, we substituted those pixels with the median of the PSFs imaged at different locations after being shifted at the same location as the PSF at hand. To center the images, we used cross-correlation on a synthetic and perfectly centered PSF generated with webbpsf (Perrin et al. 2014) using the closest in time optical path difference map. As an input for the stellar spectrum, we fit a stellar+blackbody model to the Near- to Mid-Infrared photometries available in VizieR Photometry Viewer, where the blackbody accounts for inner disk excess emission from ∼3 μm onwards.

At this point the data are treated differently when they are used as science and when they are used as reference to remove the stellar PSF from images taken with the other roll angle. Science data are binned every 5 integrations as we found that this does not affect contrast performance, but speeds up every subsequent process. This is not done for references, as more images carry more information and help with the PSF modeling and subtraction. We include a random shift in every reference image, centered around 0 and with standard deviation equal 0.05 pixels. This step is similar to the small grid dither pattern employed during coronagraphic observations. ¹² It helps minimizing the impact of small pointing error/shifts and the imperfections of the centering algorithm in order to more accurately subtract the central PSF. Finally, to deal with the sometimes irregular saturation, we applied a round central mask at the location of the star with radius r = 03 (08 for F410M). At this point images have sizes of 38 × 38 and 74 × 74 at short and long wavelengths, respectively.

Then, we applied principal component analysis (PCA; Amara & Quanz 2012; Soummer et al. 2012) as performed in PynPoint (Stolker et al. 2019) in order to first model the stellar emission in one roll and then subtract it from the other one. Images are then derotated to a common orientation and residuals from both rolls are finally combined. We found that at least seven principal components (PCs) need to be removed in order to best reveal the disk structures and residuals from potential planets. Removing larger numbers of PCs does not improve the residuals or the contrast, but, contrary to ground-based observations, it does not increase self-subtraction either thanks to the two-rolls observing strategy.

The *_cal.fits P330-E data were downloaded from the archive and were centered using cross-correlation with a synthetic PSF generated, once again, with webbpsf. The images were then median combined, and cropped to 10 in order to focus only on the central region of the PSF. The P330-E spectrum has been taken from Rieke et al., submitted to AJ.

At large separations the residuals in the F410M images revealed a companion candidate at ∼22 from the star that we designate CC1. The candidate is best visible when using a larger mask to cover the central region of the images, allowing the principal components to focus on removing the PSF pattern at large separations. In Figure 2 we report the residuals when using a central mask of 19 (left panel) and a zoom-in on the area of interest (central panel). Despite several speckles having similar strengths as CC1, the negative-positive-negative pattern expected for a point source in observations obtained with the two-rolls strategy is clearly visible in the residuals, strongly supporting a point source detection. We additionally inspected the residuals from individual rolls. In both of them there is a positive signal at the location of the source, and a negative feature on one of the two sides (as expected), providing additional evidence that the detection is real. Using the metric proposed by Mawet et al. (2014), we estimated the signal-to-noise ratio of CC1 to be 4.4 and its false-positive fraction to be 1.6 × 10⁻⁵, corresponding to 4.1σ assuming Gaussian noise. This estimate only considers the central positive signal, and being able to consider the correlated symmetric pattern would certainly provide a higher confidence. Note that even though other regions in the image have similar brightness (in particular above the northeast edge of the central mask), they are part of an extended diagonal noise feature resulting from PSF spiders residuals and are not physical sources (see Figure 7, where we masked out the extended feature). In Appendix B we provide examples of injected sources at the same separation as CC1 and with similar intensity, but at different position angles. In most cases the injected sources indeed look very similar to CC1 in Figure 2.

Zoom In Zoom Out Reset image size

To characterize the companion candidate, we adapted the methods used in Stolker et al. (2020b) to the NIRCam data structure to determine three parameters: separation, position angle, and contrast. Briefly, we used an Markov chain Monte Carlo (MCMC) algorithm to explore the parameter space with 200 walkers performing a chain of 400 steps each. At each step, artificial negative copies of the PSF (obtained from the P330-E data) were inserted in the data at the location of companion candidates. Then, PSF is removed with the same approach described in Section 4.2 and the residuals are minimized in an aperture of radius 03 (Wertz et al. 2017; Stolker et al. 2019, 2020b).

To avoid measurement biases and explore potential systematics due to remaining PSF and detector artifacts, we inserted artificial planets at the same separation measured from the  MCMC algorithm with the same contrast at 360 different position angles. We then used the same algorithm to retrieve these values and calculate the difference with the input for the 360 experiments. The mean of the differences is added to the MCMC results as a bias offset, while the standard deviation is used to account for speckle noise and systematics in the measurement and is added to the MCMC uncertainties.

We found that the separation from the central star is ${2.225}_{-0.073}^{+0.071}$ arcsec and its position angle is ${303.5}_{-2.1}^{+2.0}$°. Considering the disk geometry (Table 1) and assuming the candidate is a real companion coplanar with the circumstellar disk, we find its physical separation to be ${300.8}_{-9.5}^{+9.9}$ au.

The measured companion photometry in F410M is ${1.7}_{-0.5}^{+0.6}\,\mu {Jy}$. Assuming unobscured photospheric emission, we used the BEX evolutionary models (Linder et al. 2019) coupled with the HELIOS (Malik et al. 2017) atmospheric models assuming solar metallicity, to estimate the mass of CC1. We find M_CC1 = 0.8 ± 0.3 M_J. This value is consistent with the non-detection in F200W (see below). We note that inserting one negative point source in the data at the CC1 location removes both the negative wings and the central positive signal (see right panel of Figure 2), confirming once again the correlation between the elements of the signal pattern and excluding independent speckles residuals could generate the CC1 signal.

The point source is not detected in the F405N filter, as the LW narrowband data are less sensitive in that region than the F410M data (see Section 5.2). Furthermore, due to the SW data's smaller pixel scale, the location of CC1 falls outside of the FoV for some of the dither positions. Hence, we used only two dither positions for each roll where the star is located such that the location of CC1 is included in the frames. CC1 is not detected in the final residuals, and using the approach described in the next Section we estimated a flux upper limit of 0.8 μ Jy in the F200W filter at the separation of CC1. This implies that CC1 is a very red object.

To estimate the probability of CC1 being a background object within the Milky Way galaxy, we used the Trilegal galactic model (Girardi et al. 2012), which provides, given the coordinates, a model for the stellar population observable in an area of 1 deg². To each element of the catalog consistent with the F200W limits, we assign a value corresponding to the probability function of the CC1 F410M photometry. We then sum up these values and normalized to the considered FoV (we used 55 arcsec² for the NIRCam data, equivalent to the FoV of the F410M data). This translates in a probability of 0.008 that CC1 is a background galactic object.

Additionally, we used the mock extragalactic catalog from Williams et al. (2018) to estimate the probability of CC1 being a background extragalactic object. The catalog includes two populations of galaxies (quiescent and star-forming) and both need to be considered. Given the position of SAO 206462 in the Milky Way (latitude 332392; longitude +17325), galactic extinction (A_V = 0.15 mag ¹³ ) toward a background object can be considered negligible, especially in the infrared. We found that for the star-forming and quiescent galaxy populations CC1 has a probability of 0.354 and 0.001 of being a background contaminant, respectively.

The location of CC1 in a color–magnitude diagram is shown in Figure 3. In gray, we show the population of galaxies with the contours indicating the areas including 50%, 80%, 90%, and 95% of the elements. The very red colors of CC1 are inconsistent with the vast majority of the extragalactic objects.

Zoom In Zoom Out Reset image size

Combining results from the galactic and extragalactic models, there is a probability of ∼36.3% that CC1 is a background object based on the NIRCam observations. This value should be treated as an upper limit, as many of the background contaminant galaxies from the mock catalog are spatially resolved, while CC1 is unresolved. To include the requirement that the contaminant should be unresolved, we can take the effective radius r_e of each source and convolve it with the FWHM of the F410M PSF (see Table 2). In practical terms, we verify that the term $\sqrt{{r}_{e}^{2}+{\mathrm{FWHM}}_{{\rm{F}}410{\rm{M}}}^{2}}$ is smaller than a threshold that we set to 1.5 FWHM_F410M. This reduces the probability of the contaminant scenario to ∼4.2% (galactic and extragalactic). Follow-up observations in future cycles will help determining the true nature of CC1.

To estimate the contrast limits we used applefy (Bonse et al. 2023), a python package that estimates contrast curves by injecting artificial sources in the data in steps of 1 FWHM (see Table 2), where at each separation we inspected six different azimuthal positions. We set the detection threshold to a false-positive fraction of 2.87 × 10⁻⁷, corresponding to 5σin the case of Gaussian noise. The noise was estimated as individual pixels separated by 1 FWHM (Bonse et al. 2023). To take into account the speckles field in the residuals, at each separation the noise configuration was measured 360 different times, each with slightly different angular orientation, using the obtained median value for the contrast and the standard deviation as an uncertainty on the contrast measurement.

Due to the central saturation in our images, we used PSFs from P330-E (see Section 3) to inject fake companions. This choice prevents us from obtaining contrast estimates compared to the central star, but guarantees an accurate photometric calibration of our limits. Indeed, the contrast estimated with respect to the P330-E PSF can be directly combined with the flux of the standard star (20.17, 21.23, 5.81, and 5.84 mJy for F187N, F200W, F405N, and F410M, respectively) in order to have a calibrated flux limit.

Following this procedure we estimated the 5σ contrast limits for our SAO 206462 NIRCam data sets compared to P330-E. From there, we used the calibrated fluxes to convert the contrast measurements into flux limits, which are reported in Figure 4 and Table 3. The gray dashed vertical line represents the size of the spirals detected in scattered light (consistent with Stolker et al. 2016), and therefore marks the region where the limits can not be trusted. Indeed, at shorter separations the physical signal from the disk and the strong self-subtraction prevent the noise from being independent and identically distributed, and the statistics should be treated with caution (Cugno et al. 2023a). For the SW filters, the narrow filter reaches a lower sensitivity than the broad filter. This is likely because only the inner 08 of the F187N data are contrast dominated, while at larger separation the data are background limited. Conversely, F200W images are contrast limited at every separation, and the sensitivity limit of the filter is expected to be almost two orders of magnitude lower than what we achieve at large separations. In the LW channel, F405N is more sensitive <09, while at larger separation F410M could detect fainter sources. Indeed, at short separations the effect of saturation and its consequences like the brighter-fatter effect and charge migration play an important role in limiting our sensitivity in F410M. Both LW filters are far away from being background limited, even at large separations.

Zoom In Zoom Out Reset image size

We used the same evolutionary models from Section 5.1 to interpret the sensitivity limits. At each separation, we converted fluxes into planet masses for the four filters and then considered only the most constraining mass limit achieved. The final mass limits can be found on the left panel of Figure 5 and Table 3. At 120 au, the NIRCam data exclude planets more massive than ${2.2}_{-0.9}^{+0.5}\,{M}_{{\rm{J}}}$ for a system age of ${11.9}_{-5.8}^{+3.7}$ Myr, while at larger separations (≳20) we would have detected objects with $M\gt {1.5}_{-0.5}^{+0.3}\,{M}_{{\rm{J}}}$, assuming negligible extinction. These mass detection limits are much deeper than those previously available through VLT/SPHERE (Asensio-Torres et al. 2021).

Zoom In Zoom Out Reset image size

In addition to the model-based mass estimates, we want to pursue a more empirical approach. Following Cugno et al. (2023a), we assumed that forming planets emit like blackbodies. This assumption is consistent with observations of PDS70 b and c (Stolker et al. 2020a; Wang et al. 2020; Cugno et al. 2021) where no significant molecular features have been detected so far. ¹⁴ We fixed the planet radii to 1, 2 and 3 R_J representing a range of radii consistent with planet formation and evolutionary models (e.g., Allard et al. 2001; Baraffe et al. 2003; Ginzburg & Chiang 2019) or observations (Stolker et al. 2020a). We then explore T_eff to find the maximum value that produces a blackbody with emission within the NIRCam filters below our limits. The T_eff limits as a function of separation are showed in the right panel of Figure 5 and are reported in Table 3. At the predicted separation of the planet (Xie et al. 2021) our data are sensitive to objects with T_eff ≳ 650–830 K (depending on R_p), while beyond 20 we hit the 500–750 K limit.

The two narrow filters used in this work allow to investigate the presence of ongoing planetary accretion processes, as they are centered on two hydrogen recombination lines (Pa-α and Br-α). In this section, we will consider the limits estimated at 120 au (see Figure 4), which provide line luminosity upper limits of 2.1 × 10⁻¹⁰ L_⊙ at Pa-α (λ = 1.87 μm) and 8.8 × 10⁻⁹ L_⊙ at Br-α (λ = 4.05 μm). We assumed three mass values of 1, 5, and 10 M_J, where 5 M_J is the predicted mass of the perturber generating both spirals (see Section 2.2), 10 M_J represents the upper limit of the possible mass range obtained by hydrodynamical simulations and 1 M_J represents a low-mass scenario. For each mass, we estimate the planet radius using the BEX cooling curves.

Using the accretion shock models from Aoyama et al. (2018, 2020, 2021) determined the relation between L_line and L_acc for forming protoplanets, which can be used to determine L_acc and subsequently ${\dot{M}}_{\mathrm{acc}}$ via

$$\begin{eqnarray}&&{\dot{M}}_{\mathrm{acc}}=\displaystyle \frac{{L}_{\mathrm{acc}}\,{R}_{{\rm{p}}}}{G\,{M}_{{\rm{p}}}}\end{eqnarray} \tag{ 1 }$$

where G is the gravitational constant.

Figure 6 reports the mass accretion rate upper limit of the potential planet as a function of the extinction affecting the line emission in case of accretion processes. Following theoretical work on embedded protoplanets, we consider extinction values up to A_V = 150 mag (Szulágyi et al. 2019; Sanchis et al. 2020). The orange and the violet lines report mass accretion rate upper limits based on the new NIRCam data on Pa-α and Br-α, respectively, presented in this work. In addition, we plot the Hα flux upper limit obtained by Cugno et al. (2019) with SPHERE/ZIMPOL (L_Hα = 8.5 × 10⁻⁷ L_⊙ at separations ≳025).

Zoom In Zoom Out Reset image size

The new JWST limits are much more constraining than the previous ground -based optical observations for every value of extinction. For A_V = 0 mag, the difference between the existing Hα limits and the new Pa-α limits in terms of mass accretion rate is ∼2 orders of magnitude. The difference increases with increasing extinctions, mainly because of the strong effect of extinction at optical wavelengths. At A_V ≈ 40 mag, JWST is as sensitive at Pa-α as the unobscured (A_V = 0 mag) Hα SPHERE data. Because the effect of extinction continues to weaken with increasing wavelengths, at A_V ≈ 60 mag the strongest constraints are provided by Br-α data, as Pa-α becomes too obscured. Figure 6 highlights which filter is more sensitive depending on the obscuration level, and demonstrates the potential of JWST in identifying accreting embedded planets.

At 120 au, where the spiral dynamics suggest a planet is located (Xie et al. 2021), the NIRCam data are able to exclude objects with fluxes of 4.8 ± 0.5 and 103.7 ± 16.6 μ Jy in the F200W and F410M filters, respectively (see Figure 4). When translated into BEX masses, these values exclude planets with M_p ≳ 2.2 M_J, at odds with theoretical predictions based on the spirals morphology and brightness, which suggested M_p ≳ 5 M_J. When translated into effective temperatures, these flux limits exclude planets hotter than 817 ± 11, 712 ± 9 and 662 ± 8 K assuming a radius of 1, 2, and 3 R_J. Hence, both our M_p and T_eff limits are not consistent with unobscured hot-start models. For example, AMES-DUSTY and BT-Settl evolutionary models (Chabrier et al. 2000; Allard et al. 2001, 2003) predict that a 5–10 M_J planet at the age of SAO 206462 would have an effective temperature of 1200–1700 K with a radius of ∼1.5 R_J. Such a companion would be clearly visible in multiple JWST/NIRCam filters. These results point toward two possible scenarios similar to what suggested in Wagner et al. (2023): either the planet forms cold (e.g., Marley et al. 2007; Spiegel & Burrows 2012), or it is highly extincted.

A much cooler evolutionary track would explain many non-detections of predicted planets embedded in disks in the thermal continuum (Asensio-Torres et al. 2021; Cugno et al. 2023a) and the NIRCam data provide tight constraints on the possible temperature of the planet around SAO 206462. Cold-start models predict smaller planet radii compared to hot-start models (R_p ∼ 1 R_J), and our constraints suggest the planet T_eff should be below 800 K. This is consistent, for example, with the cold-start models presented in Marley et al. (2007) and Spiegel & Burrows (2012). We note however, that this scenario is at odds with preliminary results on dynamical masses of giant planets, where it appears that giant planets form warm to hot (e.g., Brandt et al. 2021a, 2021b; Franson et al. 2023). However, two factors must be considered: (i) most planets with dynamical mass measurements are much closer to their star than the planet launching the spirals in SAO 206462, possibly indicating they underwent a different formation pathway or were influenced by different environments at birth; and (ii) this might be an observational bias, as only hot-start planets might be bright enough to be detected by currently available instrumentation.

In the other scenario, the planet is deeply embedded in the circumplanetary disk/envelope, and the high extinction is hindering its detection at NIR wavelengths. This possibility is supported by observations of the PDS70 system (Isella et al. 2019; Benisty et al. 2021; Cugno et al. 2021) and was already explored by Wagner et al. (2023) for MWC758 c. We estimated that a visual extinction of A_V ≳ 20 mag would be necessary to make sure that a 5 M_J perturber following the BEX models would not appear in any of our NIRCam filters. If obscuration from the circumplanetary material is the main cause for the poor detection rate of protoplanets, longer wavelength observations might be able to trace colder temperatures and reveal the light reprocessed by the planet surroundings.

When considering the spiral rotation rate from Xie et al. (2021), CC1's semimajor axis is located at 2.25σ from the dynamical prediction for the perturber in the eastern spiral. The rotation velocity from Xie et al. (2021) is based on a 1 yr baseline, and it is likely that a longer period between observations will provide a tighter and more precise constraint on the separation of the perturber. As described in Section 5.1, assuming its emission is due to photospheric emission from a gravitationally bound object, we estimated the mass of CC1 to be 0.8 ± 0.3 M_J. It is unlikely that such a small mass at such large separation is responsible for launching both spirals, but if two perturbers are launching the spirals independently, the mass requirement is relaxed and it is possible that a ∼M_J planet drives one of them.

There are a few other possibilities that could connect the distant CC1 and the two prominent spirals: (i) an eccentric orbit, (ii) a very large mass for CC1 and (iii) a flyby encounter. These scenarios are further discussed in the next paragraphs.

If CC1 is on a highly eccentric orbit (e ≳ 0.6 assuming CC1 is currently at its apocenter and its pericenter is about the disk size), it can be capable of launching spirals if it is massive enough. Zhu & Zhang (2022) found that if the companion is on an eccentric orbit, a higher mass is required to form the same spiral pattern. This would imply that the mass of CC1 should be ≫5 M_J. Again, the requirement would be for the object to either start very cold or be highly obscured (or both). We note that an eccentric orbit could also explain the different spiral velocities measured by Xie et al. (2021).

If CC1 is highly obscured, it is possible that its mass is much larger than what we found in Section 5.1. If that is the case we are looking at a system similar to HD100453 (Collins et al. 2009; Benisty et al. 2017; Wagner et al. 2018), in which a massive (stellar) companion is driving a pair of near-symmetric spirals, which has been shown in simulations as well (Dong et al. 2016a, 2016b). In addition, such a massive companion is expected to open a wide gap (Artymowicz & Lubow 1994), which can be tested in gas observations. We note that given the very deep NIRCam measurements at F410M, it is unlikely that CC1 could be as massive as HD100453 B (0.2 ± 0.04 M_⊙, Collins et al. 2009). Indeed a highly embedded stellar companion with properties similar to a Class 0 object would require a dusty envelope that most likely would have been detected in the ALMA data. Hence, the massive companion scenario seems rather unlikely.

CC1 could be an older (and hence more massive) flyby object able to produce the spiral structure (e.g., Cuello et al. 2019). Shuai et al. (2022) excluded that a flyby object detected by Gaia could have induced the spirals, but CC1 does not appear in the Gaia catalog. Assuming an age of 1 Gyr, we estimate a mass of 30–50 M_J for an object consistent with the F410M measurement. Smallwood et al. (2023) tested perturber masses in the 10 M_J–1 M_⊙ range with 3D hydrodynamical simulations, showing that small mass objects produce faint spirals, while more massive bodies are necessary to see prominent features. In this scenario, it is not clear what mass would be required to reproduce the bright spirals of SAO 206462 due to insufficient information on its orbit, and determining this value is beyond the scope of this paper.

Finally, CC1 could be a faint companion to SAO 206462, but not being the object responsible for lauching the disk spirals. Such an object is still below our sensitivity and deeper observations at long wavelength are necessary to directly detect it.

The very tight limits indicate that even for very extincted objects (A_V = 100 mag), if accretion onto the planet is ongoing at a moderate rate (10⁻⁶ M_J yr⁻¹) we should have detected emission in the line filters. For less extincted scenarios (A_V = 40–60 mag) we exclude even low-mass accretion rates of the order 10⁻⁸–10⁻⁷ M_J yr⁻¹ (see Figure 6). These are the most stringent constraints of ongoing accretion processes in any protoplanetary disk, and we highlight that even the extreme AO systems at Hα are not able to achieve such sensitivity (Cugno et al. 2023b).

The lack of detection of accretion processes despite the extremely deep limits is hard to reconcile with the scenario of a planet strongly interacting with and likely accreting disk material. However, recent radiation hydrodynamics simulations by Marleau et al. (2023) showed that only a fraction of about 1% of the accreting material falls onto the planet fast enough to produce a shock and emit hydrogen lines. Alternatively, due to the thick envelope, the incoming material does not shock at the planet surface but at higher elevations, where it does not have enough energy to excite hydrogen atoms and thus produce emission lines. The low-emission efficiency and the higher elevation of the shock surface could explain the lack of accretion line detections in the very deep data presented in this work. Further work is warranted both on the theoretical and observational side in order to unveil how young gas giants grow their mass.