Searching for Planets Orbiting Fomalhaut with JWST/NIRCam

The full set of images (summarized in Table 2) was processed using the JWST pipeline (version 2022_4a, calibration version jwst_1019.pmap). The data set can be obtained at:10.17909/kckm-n422. We started from the uncalibrated stage-1 data products as produced by the standard pipeline (Bushouse et al. 2023) with some modifications of the subsequent steps. Specifically, (1) we did not include dark current corrections, which are not well characterized for subarray observations; (2) we performed a modified version of the ramp fitting, as implemented within the SpaceKLIP package (Kammerer et al. 2022) to significantly improve the noise floor in the subarray images; and (3) to reduce saturation effects in the full array images, we allowed fluxes to be measured for ramps that only have a single group before saturation.

We performed additional steps to reject bad pixels and remove horizontal striping resulting from 1/f noise. The standard pipeline removes pixels adjacent to saturation-flagged pixels, which can result in an overly aggressive removal of good data. Since the images do not have any evidence of charge spillage, we utilized less conservative flagging (n_pix_grow_sat set to 0, rather than (1). For identification of truly bad pixels, we used the pipeline DQ flags: any pixels flagged as DO_NOT_USE, e.g., dead pixels, those without a linearity correction, etc., were set to NaN. 5σ outliers—temporally within subexposures or spatially within a 5 × 5 box—were also rejected.

After PSF subtraction, a set of additional bad pixels became apparent in the inner, speckle-dominated, region. The brightness of the speckles helps these bad pixels elude correction in the first steps of correction; we, thus, need to correct for these after PSF subtraction (Section 3.3). For the inner region (∼37 × 37) of the PSF-subtracted non-coadded images, we apply a temporal and spatial bad pixel correction identical to the procedure described in the previous paragraph.

To remove horizontal striping resulting from 1/f noise, we subtracted the median value from each row of data (the direction of fast detector readouts). Figure 1 shows single F356W frames before and after horizontal stripe removal. Figure 2 shows the final coadded images in both F356W and F444W, prior to the next post-processing steps to remove residual starlight in the images.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Following the basic data reduction for individual images, we combined the overall set of two roll angles on the primary target and a set of dithered observations of the reference star, using either reference star differential imaging (RDI) or angular difference imaging (ADI). We used two approaches for this post-processing—classical PSF subtraction and principal component analysis (PCA) taking advantage of the full diversity of the dithered reference PSFs. For the classical RDI, we first created a reference PSF from the nearby star HR8709, shifting and coadding its five dithered observations together to maximize S/N. We scaled and shifted the reference star PSF to align with the target at Roll 1 and Roll 2 independently before performing the PSF subtraction. For the classical ADI, we subtracted the two rolls from one another after applying the corresponding shift and data centering. In both RDI and ADI approaches, the last step after PSF subtraction was to orient both subtracted rolls to the North before coadding them resulting in a negative–positive–negative pattern for sources that are present in both telescope angles.

While the classical PSF subtraction performs well at larger distances from the target star where the noise is limited by the instrument sensitivity, at close separations residual starlight speckles are the dominant limitation to the detection of point sources. PCA analysis is preferred for cleaning the inner speckle field within ∼15. In this speckle-dominated region, we performed a PCA-based algorithm (Amara & Quanz 2012) via Karhunen Loéve Image Projection (KLIP; Soummer et al. 2012) using both the reference frames and the target frames from the opposite roll in the PSF library. We used the open-source Python package pyKLIP (Wang et al. 2015), which provides routines for cleaning the images, calculating detection limits, and quantifying the uncertainty in the flux of any detected sources.

The KLIP reduction was done using all available images (i.e., with six KL modes), with the full array mode data cropped and centered to match the subarray mode data. Only the full array data set was used to produce the PCA full-frame reductions. The reference star data was used for PSF subtraction. Use of the 5-POINT-SMALL-GRID dither pattern mitigated any misalignment between the star and coronagraph focal plane mask.

Figures 3, 4, 5, and 6 show the results of the classical PSF and PCA subtractions. The presence of the expected negative–positive–negative image pattern is a good indication that a candidate object is real, even if its overall S/N is too low for reliable extraction of its position and brightness. Sources S1-S10 discussed in Section 3.5 all show this pattern.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The contrast limits reported in this work were obtained by normalizing the flux to a synthetic peak flux. Predicted fluxes in the JWST wave bands were calculated based on BOSZ stellar models (Bohlin et al. 2017) fit to optical and near-IR photometry (Table 1). Convolving the stellar model with JWST bandpasses gives an estimated flux of 115.6 Jy (0.93 mag) in the F356W filter and 80.9 Jy (0.89 mag) in F444W. To estimate the peak flux of the instrument's off-axis coronagraphic PSF we simulated this PSF using WebbPSF (Perrin et al. 2014). Measured fluxes in the NIRCam images were divided by these stellar fluxes to obtain contrast ratios. While the brightness of Fomalhaut often saturates CCD detectors at near-IR wavelengths (e.g., in the Two Micron All Sky Survey and Wide-field Infrared Survey Explorer all-sky surveys), Carter (1990) was able to use the 75-cm South African Astronomical Observatory to obtain JHK photometry with 2% accuracy (see Table 1), resulting in ∼2% overall uncertainty in our stellar flux estimates at JWST wavelengths; note that this uncertainty only contributes to the error budget for contrast ratios, not for the measured photometry that is used to calculate the contrast ratios.

The 3σ contrast curves (Figure 7(a)) were obtained using pyKLIP. While a 3σ threshold allows for the possibility of spurious detections, we find below that the overwhelming majority of sources above this threshold (all but one) are also detected by other telescopes, validating this threshold as appropriate for identifying potential sources for follow-up analysis. The noise was computed in an azimuthal annulus at each separation of the reduced image (before any smoothing), and we used a Gaussian cross correlation (kernel size = 2 pixels) to remove high-frequency noise. The contrast was calculated using the normalization peak value for the target star. We corrected for algorithmic throughput losses by injecting and retrieving fake sources at different separations. The contrast was also corrected for small sample statistics (Mawet et al. 2014) at the closest angular separations, although saturation is dominating around 1'' and below. At separations closer than 2'' the contrast is limited against the residuals from the PSF subtraction methods (∼4 × 10⁻⁷ at 1'', ∼1 × 10⁻⁷ at 2''), and further than 2'' the performance is limited by the background level (∼19 mag in F444W), consistent with the expectations of the instrument given the exposure time. Figure 7(b) converts the sensitivity curves into detection limits in Jupiter masses appropriate to Fomalhaut's age and distance (<1 M_Jup beyond 2'').

Zoom In Zoom Out Reset image size

Losses due to the coronagraphic mask were taken into account in both the contrast curves and in the reported fluxes for any detected sources (the following section), although this is a negligible effect outside of ∼1''. Transmission losses from the Lyot stop and coronagraph mask substrate were also included in both the contrast calibration and the point source photometry (Section 3.5 below). The mask substrate throughput is 0.95 and 0.93 for the F356W and F444W filters, while the transmission of the Lyot stop wedge is 0.98 in both.

We adopted two methods to identify sources in the PSF-subtracted images, whether we considered the speckle-dominated or the outer region.

First, we focused on the inner, speckle-dominated region, looking for planet candidates within (∼1.5''; ∼10–35 au) of Fomalhaut. Taking advantage of the telescope stability to generate a well-defined PSF, we applied a forward model match filter (FMMF; Ruffio et al. 2017) method. FMMF corrects for the KLIP's over-subtraction effects with a forward model (Pueyo 2016). This forward model was then used as a match filter to enhance the S/N of sources for which the spatial structure is well matched to the expected point source PSF. The PSF for the forward model was computed for each filter with WebbPSF using the OPD wave-front error map closest in time prior to the observations. No sources were reliably detected within this speckle-dominated region. The FMMF method is most useful for point source detection in the speckle-dominated region so we used a different method to identify sources in the outer region of the image.

Second, we visually examined the entire region interior and just the exterior of the debris ring, which has a radius of 140 au. Sources were initially detected using a Gaussian-smoothed image (kernel size = 2 pixels) to search for >3σ candidates. These candidates were examined visually to identify stellar diffraction and other image artifacts.

The photometry and astrometry of the detected sources were recovered via a Markov Chain Monte Carlo (MCMC; emcee, Foreman-Mackey et al. 2013) fit to the PSF-subtracted data using pyKLIP. Sources for which the MCMC fit was unsatisfactory were analyzed with aperture photometry. For sources that manifest themselves differently from one method to another, the differences in the methodologies serve as a diagnostic for the source characteristics. Point sources, for example, will be detected more strongly with a method that matches the source to the predicted PSF like MCMC. Extended structures, on the other hand, will have poor fits to point source models.

Aperture photometry was performed with a 025 radius aperture (about 4 pixels) at the same location as the MCMC fit, with the background calculated within a 16 to 24 pixel annulus. While this aperture size is large relative to the nominal telescope resolution, diffraction by the coronagraphic mask and Lyot stop results in a much broader PSF, with a large fraction of the flux dispersed to wider angles; only 20%/25% of the flux is enclosed within the aperture for the F356W/F444W filters.

We simulated the coronagraphic PSF with WebbPSF_ext (J. Leisenring 2023 in preparation), deriving aperture corrections of 3.87 and 4.81 (multiplicative factors on the flux) for F356W and F444W, respectively, for our chosen aperture. We also measured the flux of each target within smaller and larger apertures (ranging from 01 to 10), finding that most of our sources yield consistent fluxes independent of aperture size (after applying the aperture correction), as expected for point sources. Some sources, however, have fluxes that rise significantly with aperture (e.g., S2 and S5), suggesting source sizes that are a fraction of an arcsecond. We determined the photometric uncertainty by placing the aperture at locations throughout the background annulus and measuring the dispersion in fluxes (again multiplied by the aperture correction factors). The standard aperture size (025) was chosen to yield optimal S/N for point sources but also served adequately for the sources that are moderately extended. The measured fluxes and uncertainties are given in Table 3, with those determined via aperture photometry explicitly flagged. The calibration uncertainty for NIRCam (not included in the Table 3 uncertainties) is currently estimated as ∼5%, ¹² much smaller than the photometric uncertainties for our low S/N detections.

Table 3. Sources Detected by JWST/NIRCam

Source	Other	Δα	Δδ	F356W			F444W
	Images a	('')	('')	Peak S/N b	(μJy)	(mag)	Peak S/N	(μJy)	(mag)
S1	MIRI c , K09	−6.10 ± 0.01	−2.28 ± 0.01	3.7	${6.84}_{-0.55}^{+0.54}$	19.0 ± 0.12	3.7	${5.17}_{-0.47}^{+0.51}$	18.9 ± 0.15
S2	K14	+7.22 ± 0.03	−5.94 ± 0.04	3.3	${3.09}_{-1.42}^{+1.26}$	19.8 ± 0.6	2.3	${1.24}_{-1.14}^{+1.11}$	20.4 ± 0.7
S3 d	MIRI,HST/STIS,K17	+14.89 ± 0.04	−8.49 ± 0.06	3.4	5.08 ± 1.06	19.3 ± 0.2	3.8	3.39 ± 0.90	19.3 ± 0.3
S4 d	MIRI,HST/WFC3	+21.29 ± 0.04	−17.23 ± 0.04	4.2	2.94 ± 1.19	19.9 ± 0.4	4.1	3.02 ± 1.18	19.5 ± 0.4
S5 d	HST/WFC3/Spitzer	+4.69 ± 0.01	−25.31 ± 0.01	3.6	9.11 ± 0.94	18.7 ± 0.1	5.2	12.81 ± 0.69	17.9 ± 0.1
S6 d	K12	+2.80 ± 0.03	−4.59 ± 0.03	2.0	5.36 ± 1.00	19.3 ± 0.2	⋯	2.02 ± 1.00	19.9 ± 0.4
S7 d	N/A	+0.45 ± 0.03	+3.91 ± 0.03	⋯	<4.26	>19.4	3.0	5.96 ± 1.13	18.7 ± 0.2
S8	HST/STIS,K13	+5.77 ± 0.003	−13.29 ± 0.002	25.8	${56.22}_{-2.02}^{+2.01}$	16.7 ± 0.07	26.7	${42.91}_{-1.14}^{+1.10}$	16.6 ± 0.05
S9	HST/STIS,K07	−20.09 ± 0.013	−8.29 ± 0.011	6.0	${6.94}_{-0.80}^{+0.76}$	19.0 ± 0.2	8.1	${5.86}_{-0.61}^{+0.63}$	18.8 ± 0.2
S10	HST/STIS,K04	−28.21 ± 0.001	+9.74 ± 0.001	31.7	${76.46}_{-1.16}^{+1.18}$	16.4 ± 0.03	26.8	${64.01}_{-0.84}^{+0.85}$	16.2 ± 0.02
Fomalhaut b e	HST	−9.377	11.144	⋯	<2.83	>19.4	⋯	<3.30	>19.4

Notes. The position offsets are relative to the current location of Fomalhaut (Epoch = 2022.808), located at α,δ = 22^h57^m39615, −29°37' 2387, including the effects of proper motion and parallax. As denoted in Figure 8, the offsets in earlier HST epochs are consistent with the star's proper motion. Astrometric precision is based on the uncertainty in peak fitting; distortion contributes an additional systematic uncertainty of ∼10–20 mas.

^a Identification numbers for detected objects as seen in imaging by Keck (Kennedy et al. 2023), HST (Currie et al. 2013), and/or JWST/MIRI (Gáspár et al. 2023). ^b The peak S/N is the S/N found by the initial peak finding routine (sensitive to single-pixel outliers), while the measured fluxes are based on either an MCMC-driven fit to the post-processed instrument PSF or, in the case of extended sources, from aperture photometry. ^c The GDC source in the MIRI image is located at Δα, Δδ = −605 ± 03,−242 ± 03. The offset between MIRI's GDC and NIRCam's S1 is −005 ± 03, +014 ± 03. ^d Flux densities and uncertainties for this source are derived from aperture photometry, rather than PSF fitting. ^e Upper limit at the predicted position of Fomalhaut b, if it is on a bound orbit. A similar limit applies at the position of Fomalhaut b on an unbound orbit.

Download table as:  ASCIITypeset image

Source positions were measured relative to Fomalhaut in the detector frame and converted to celestial coordinates using the star's position at the time of observation (including proper motion and parallax). As described in Greenbaum et al. (2023), the position of the star in the detector was computed by performing cross-correlations of the data with synthetic PSFs, computed using WebbPSF. We use the chi2_shift functions in the image-registration Python package. ¹³ This method yields uncertainties of ∼7 mas, consistent with Carter et al. (2023).

We applied up to ∼30 mas distortion corrections to the WCS coordinate frame, based on a distortion map derived from on-sky observation of a dense stellar field (jwst_nircam_distortion_0173). The correction was performed with the jwst Calibration Pipeline (Bushouse et al. 2023). We estimated the astrometric accuracy to range from ∼10 mas for sources close to the central star up to ∼30 mas for sources outside of the Fomalhaut ring.

Images at F356W and F444W were treated separately, and the results are presented in Table 3. By design the integration time in the shorter wavelength filter was smaller than at the longer wavelength, making the sensitivity at F356W worse than at F444W. The primary purpose of the F356W observation was the rejection of objects with typical stellar or galaxy-like SEDs.

We report in Table 3 the "peak S/N" associated with the brightest pixel of a given source. This is reported to quantify how much the source visually stands out over the noise. However, when extracting the photometry using the joint photometry and astrometry model fit, as explained above, the error bars may differ from the peak S/N for lower signal detections. This discrepancy is due to the detections being excessively noisy due to bad pixels or poorly subtracted speckles. The figures in the Appendix show the model fit results of the data and model-selected point-like sources, e.g., S1, as well as the MCMC corner plots for the astrometric and photometric fits. The photometry and astrometry error bars provide the most complete accounting of the properties of the sources, whereas the peak S/N describes how well a source can be seen over the smoothed noise in the image.

Figure 4(a) shows the final images for the inner speckle field (<15) around Fomalhaut. No sources are apparent, just a residual noise floor from imperfect speckle subtraction. Figure 4(b) shows the corresponding S/N map from the FMMF analysis of the inner region; applying an S/N = 5 threshold, the data yield no detections.

Figure 8 shows the sources extracted from the final NIRCam reduced image, along with a superposition of the MIRI data at F2300C. Some objects are clearly detected in Keck (circled in green), HST (circled in orange), or both HST and Keck (circled in yellow) images from earlier epochs. ¹⁴ One source, identified with a white arrow in Figure 8, is visible in the MIRI data but has no counterpart in the NIRCam data. The location of this source coincides with the red point source identified in Spizter data from 2013 (see Figure 6 of Janson et al. 2015), indicating a background object. The source may be variable and not detectable at the NIRCam sensitivity levels, particularly at the very edge of the field of view. Alternatively, if the Spitzer object is not associated with the MIRI source and is indeed part of the Fomalhaut system, then the proper motion of Fomalhaut from 2013 to the present time would have resulted in the object falling very close to the edge of the field of view of the current NIRCam observations.

Zoom In Zoom Out Reset image size

Table 3 summarizes the properties of the objects detected within 30'' of the star, corresponding to a projected separation of 230 au. Ten point sources are detected at either or both F356W and F444W with greater than 3σ significance. We have compared these sources against those detected in observations by Keck (Kennedy et al. 2023), HST (Currie et al. 2013), and JWST/MIRI (Gáspár et al. 2023); all but one source—S7—have been previously imaged. We find that none of these previously identified sources are co-moving with Fomalhaut and we conclude that they are background objects.

A brief description of each source follows:

• 1.  
  S1: The source sits within the combined NIRCam-MIRI positional uncertainties at the clump of emission seen in the MIRI coronagraphic image at 23.0 and 25.5 μm and denoted as the Great Dust Cloud (or GDC) by Gáspár et al. (2023). Detected at both F356W and F444W, it is separated from the GDC by (α, δ) = (−50 ± 300, + 140 ± 300) mas where the dominant source of uncertainty comes from the lower resolution MIRI image. The F356W/F444W color, [F356W]–[F444W] =0.0 ± 0.2 Vega mag), is relatively blue (corresponding to a color temperature of ∼1700 ± 400 K). There is a Keck object, K9, as well as an ALMA source seen at this position confirming this as a background object (Kennedy et al. 2023). Astrometric coincidence within ∼100 mas with Keck H-band objects from earlier epochs (2005–2011) is used to identify S1 and all but one of the other NIRCam detections discussed below as background objects (Section 5). The F356W data for S1 show a hint of being extended, consistent with the galaxy interpretation (Figures 5, 12). As discussed further below (Section 5), comparing the combined NIRCam/MIRI spectral distribution with Spitzer SWIRE templates (Polletta et al. 2007) suggests that the object is an ultra-luminous IR galaxy like Arp 220 or M82 at a redshift z = 0.8–1.0
• 2.  
  S2: This source is clearly detected at F356W with a peak S/N of 3.3 and marginally in F444W (Figure 13). Although S2 is positioned tantalizingly close to the outer edge of the innermost disk, close to the gap between the disk and the intermediate belt seen in the MIRI images (Gáspár et al. 2023), the presence of a Keck source at this position, K14, makes association with Fomalhaut improbable.
• 3.  
  S3 and S4: These objects have high peak S/Ns in both F356W and F444W while the point source MCMC analysis provides a poor fit to the data. Aperture photometry of a slightly extended source (∼025–05) provides robust photometric detections in both bands for both objects. S3 is detected in the Keck deep imaging, the 2014 HST/Space Telescope Imaging Spectrograph(STIS) image, and the MIRI data at F2550W (Gáspár et al. 2023 Figure 9). Although S4 is not seen in Keck imaging, it is detected by HST/WFC3 and its relatively blue color ([F356W]–[F444W] ∼ 0 (Vega mag) is consistent with its being a distant galaxy. S4 is not seen in the F2550W imaging and falls outside the field of view of the F2300W data.
• 4.  
  S5: This object sits well outside the Fomalhaut outer ring. It is robustly detected at both F356W and F444W. It is not seen in the Keck imaging and falls outside of the field of view of available HST data. Although the source is well fit with the PSF MCMC analysis (Figure 17), the aperture photometry finds increasing flux for larger apertures, suggesting an extended object. Although the [F356W]–[F444W] = 0.8 ± 0.2 mag (Vega) color is more typical of a distant brown dwarf than a field galaxy (Section 5), the hint of extended emission suggests the latter interpretation.
• 5.  
  S6: This object has marginal detections at F356W and F444W but is spatially coincident with an object seen in deep Keck imaging (K12). With [F356W]–[F444W] = 0.6 ± 0.4 (Vega mag) it is likely to be a background galaxy.
• 6.  
  S7: This object is a ∼5σ detection at F444W and not detected at F356W, giving it a [F356W]–[F444W] color ≥0.7 (Vega mag). It is without a Keck counterpart. If it were an exoplanet, it would have a mass of ∼1 M_Jup as suggested by Figure 10 and various models (Baraffe et al. 2003; Linder et al. 2019). What is most intriguing about this object, the only NIRCam object that cannot be immediately associated with a background source, is its proximity to the inner dust disk newly identified in the MIRI imaging (Gáspár et al. 2023, Figure 8). This disk extends from >12 outward to 10''–12'', compared with the location of S7 at 4'' (∼30 au) separation from Fomalhaut. If associated with Fomalhaut and of the indicated mass, it should have substantial dynamical interactions with the inner debris disk, which are not evident in the 25.5 μm image. It will be important to address its possible effects on the structure of the inner disk if its planetary nature is confirmed.
• 7.  
  S8, S9, and S10: These three objects are easily detected in both earlier HST/Keck imaging and the NIRCam data, making them obvious background objects. In particular, S10 looks extended in HST images.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 10 compares the magnitudes and colors of our detected point sources. Most sources have 0 < [F356W]–[F444W] < 1 Vega mag (or 0.7 ± 0.5 AB mag), consistent with typical galaxy colors at this sensitivity level (Figures 30 and 31 in Ashby et al. 2013; Bisigello et al. 2023). The newly identified source S7, however, only has an upper limit on its [F356W]–[F444W] color. For comparison with S7, we show planet evolutionary curves from AMES-Cond (Baraffe et al. 2003) and BEX (Linder et al. 2019). For the BEX cooling curves, we consider two different radiative transfer models for the planetary atmosphere, HELIOS and the version of petitCODE with clouds (Linder et al. 2019). With only an upper limit to the brightness of this object at F356W, we cannot make a definitive statement about the nature of S7, but based on its brightness at F444W alone, its mass is at or below 1 M_Jup. Note that there is no sign of the planet predicted by Janson et al. (2020) as our detection limit (19 mag or ∼330 ${M}_{{M}_{\oplus }}$ in F444W) is higher than the one they had predicted for those observations (24 mag or ∼66 ${M}_{{M}_{\oplus }}$ in F444W).

The expected position of Fomalhaut b at the time of the JWST observations depends on whether the object is in a bound or unbound orbit (Gáspár & Rieke 2020). The expected offsets with respect to Fomalhaut are: (Δα, Δδ) = (−9377, 11144) and (−9809, 11665), for the bound and unbound orbits, respectively. The failure to detect any F356W or F444W emission at the predicted position(s) of Fomalhaut b rules out the presence of any object more massive than ∼1 M_Jup and is consistent with the hypothesis that the object is a dispersing remnant of a collision between two planetesimals. The scattered light seen at shorter wavelengths by HST would be much fainter at NIRCam wavelengths due to lower stellar flux (∝λ⁻²) and the expected lower scattering cross sections. Janson et al. (2020) suggest that the disruption of a planetesimal in the tidal field of a ∼0.2 M_Jup planet located at a semimajor axis of 117 au might be the cause of the observed collisional remnant. Such an object is below the sensitivity of the current observation.

We do not detect scattered light from the outer debris ring. Starting with the RDI-processed images (not ADI images, where the extended disk would self-subtract during the roll subtraction), we summed the flux within an ellipsoid annulus corresponding to the known ring location but did not find any emission above the background. Based on the azimuthal rms deviation between 30° bins, we find an upper limit on the disk emission of ∼17 Δmag/arcsec² in both the F356W and F444W filters. This contrast is more than an order of magnitude less than the optical contrast detected by HST/ACS in combined F606W and F814W data (∼20 Δmag/arcsec²; Kalas et al. 2005). While JWST/NIRCam has lower effective throughput than HST/ACS (partially negating the advantage in primary mirror size), the poorer contrast limit relative to HST is primarily a result of fainter stellar flux and lower scattering cross sections at the longer wavelengths. Typical interstellar grains have a factor of ∼10 smaller albedo at NIRCam wavelengths than at HST wavelengths Draine (2003) although large (a = 3–5 μm), ice-dominated grains can have comparable visible and near-IR albedos around 3–4 μm (McCabe et al. 2011; Tazaki et al. 2021).

Comparing the JWST upper limit against the integrated thermal emission for the dust ring (L_disk/L_⋆ = 5.4 × 10⁻⁵; Acke et al. 2012), we place an upper limit on the dust albedo of <0.6 at JWST wavelengths (3.56, 4.44 μm). While this rules out grains of pure ice, such dust has already been excluded by the much lower HST albedo measurement of ∼0.05–0.10 at optical wavelengths.