JWST/NIRSpec Observations of the Coldest Known Brown Dwarf^*

We compare the NIRSpec data to previous observations of WISE 0855 between 1 and 5 μm. We first consider the photometric measurements with the smallest errors, consisting of data in the following filters: the F105W, F110W, F127M, and F160W bands on HST's Wide Field Camera 3 (WFC3; Kimble et al. 2008) measured by Luhman & Esplin (2016) and Schneider et al. (2016), the [3.6] and [4.5] bands of Spitzer measured by Luhman & Esplin (2016), and W2 from the NEOWISE-R Single Exposure Source Table (Cutri et al. 2023), which contains 18 epochs for WISE 0855 between 2014 and 2022. Each of the NEOWISE epochs consisted of ∼10–20 measurements that span a few days. For each epoch, we calculated the median and standard error of W2. We exclude the two epochs from the original WISE survey in 2010 since the photometry of WISE 0855 was contaminated by emission from background stars (Luhman 2014b). For bands that have multiple epochs of photometry with the exception of [3.6], we adopt the midpoint of the measurements and an uncertainty that is represented by the range spanned by those measurements and their errors. The mean, median, and midpoint are nearly identical for [4.5], and the same is true for W2 as well. The adopted values of [4.5] and W2 are 13.87 ± 0.08 and 13.90 ± 0.16, respectively. Since [3.6] was measured in only four of the nine [4.5] epochs from Luhman & Esplin (2016), we have calculated the weighted mean of the [3.6] − [4.5] colors among those four epochs (3.49) and applied it to the adopted measurement of [4.5] to arrive at our adopted value for [3.6]. The uncertainty for [4.5] is also assigned to [3.6].

We have calculated synthetic Vega magnitudes in the aforementioned bands from the NIRSpec PRISM spectrum of WISE 0855. In addition, photometry has been derived from the G395M spectrum in the two bands that it fully covers, [3.6] and [4.5]. These calculations were performed by convolving the spectra of WISE 0855 and a spectrum of Sirius (Rieke et al. 2023) with the photon-counting spectral responses of the filters (Hora et al. 2008; Jarrett et al. 2011; Krick et al. 2021), ¹⁶ integrating the resulting fluxes in photons, and assuming that Sirius has a magnitude of −1.395 in each band.

In Figure 1, we have plotted the differences between the imaging and synthetic photometry. For most bands, the synthetic photometry falls within the range of previous measurements. The primary exception is [3.6], which is much fainter (∼0.4 mag) in both NIRSpec spectra than in the Spitzer images. The spectra and imaging agree more closely in [4.5], which means that [3.6] − [4.5] is redder in the spectra than in the images. To investigate this discrepancy further, we have derived synthetic [3.6] and [4.5] for the one other Y dwarf with an available NIRSpec PRISM spectrum, WISE J035934.06−540154.6 (hereafter WISE 0359; Beiler et al. 2023). We find that its color from NIRSpec is 0.28 mag redder than the value measured with Spitzer (Kirkpatrick et al. 2012), resembling the result for WISE 0855. We also have compared the synthetic and imaging colors in W2 − [4.5] for these Y dwarfs. The PRISM data indicate W2 − [4.5] = −0.02 for both WISE 0855 and WISE 0359 whereas Spitzer and WISE/NEOWISE have measured 0.03 and 0.06, respectively. It would be useful to perform similar comparisons for a larger sample of T and Y dwarfs observed by NIRSpec.

Zoom In Zoom Out Reset image size

The differences between synthetic and imaging values of [3.6] − [4.5] and W2 − [4.5] for WISE 0855 and WISE 0359 could be caused by errors in the filter response functions or variability, although neither Y dwarf has shown large enough variability to explain the discrepancies in [3.6], as discussed earlier in this section and elsewhere (Esplin et al. 2016; Brooks et al. 2023). Synthetic [3.6] photometry for T and Y dwarfs is particularly sensitive to errors in the response function since their fluxes vary significantly across the bandpass, which overlaps with the fundamental band of CH₄. The response functions for the IRAC filters do shift slightly in wavelength with position on the detector arrays (Hora et al. 2008). For the [3.6] band, the maximum redward shift relative to the average response function that we have adopted is ∼80 Å, but we find that a shift of ∼500 Å would be needed to account for the discrepancy in the photometry. An error in the response function could partially explain why synthetic values of [3.6] − [4.5] from model spectra for Y dwarfs are redder than observed colors (Leggett et al. 2021).

Among the bands that we have discussed, [4.5] is the best one for refining the flux calibrations of the PRISM and G395M spectra of WISE 0855 since it contains a large fraction of the flux in those data, is fully encompassed by both spectra, and has the most accurate photometry available from previous imaging. Based on their synthetic photometry, the PRISM spectrum is 0.053 mag fainter and the G395M spectrum is 0.022 mag brighter than our adopted photometry of [4.5] = 13.87. We have elected to multiply the spectra by the appropriate factors to bring them into agreement with the latter. Doing so results in a modest discrepancy of ∼0.1 mag between synthetic and imaging photometry in F160W, which could be explained by variability.

Skemer et al. (2016) and Morley et al. (2018) obtained M- and L-band spectra, respectively, of WISE 0855 with the Gemini Near-Infrared Spectrograph (GNIRS; Elias et al. 2006). An updated reduction of the data from Skemer et al. (2016) was presented by Miles et al. (2020). The data were binned to resolutions of ∼250 (L) and ∼300 (M), after which they had median S/Ns of ∼5 (L) and 13 (M). In Figure 2, we compare the binned spectra from Morley et al. (2018) and Miles et al. (2020) to the NIRSpec PRISM spectrum, which has a similar resolution. The errors in the NIRSpec fluxes are not plotted since most are too small to be distinguishable from the spectrum. The GNIRS and NIRSpec spectra have similar spectral slopes and show many of the same absorption features, although the strengths of some of the features differ by substantially more than 1σ.

Zoom In Zoom Out Reset image size

The NIRSpec spectra of WISE 0855 are presented in Figures 3 and 4 (PRISM and G395M). The spectra are available in electronic files associated with those figures. Since the fluxes span a very large range, the spectra are shown on both linear and logarithmic flux scales. Fluxes that are consistent with zero at 1σ are not plotted. Given that the fluxes vary significantly with wavelength, the S/Ns do so as well. In the fainter half of the PRISM spectrum (<2.5 μm), the brightest 50% of pixels have a median S/N of ∼20 while no significant flux is detected in the strongest absorption bands. At the wavelengths of that half of the PRISM spectrum, only photometry (usually with low S/N) has been available from previous observations (Section 4.1). The brightest fluxes in the PRISM data are at 4–5 μm, where the median S/N is ∼90. Given their S/Ns, wavelength coverage, and spectral resolution (for G395M), the NIRSpec data have provided a dramatic improvement in the characterization of the SED of WISE 0855.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The SED of WISE 0855 is shaped by absorption from several molecular species. In the bottom panels of Figures 3 and 4, we have plotted a representation of the opacities for the dominant absorbers that are expected in Y dwarfs, consisting of H₂O, CH₄, NH₃, CO, CO₂, and PH₃ as well as collision-induced absorption (CIA) of H₂ and He. In those opacity diagrams, the vertical thickness of each band is proportional to the logarithm of the abundance-weighted absorption cross section for a given molecule at P = 1 bar and T_eff = 250 K based on the model of WISE 0855 from Leggett et al. (2021). We have summed the abundance-weighted cross sections for the CIA absorption of H₂ and He. For the model in question, the opacities of CO₂, CO, and PH₃ are much smaller than those of the other species, so for the purposes of Figures 3 and 4, we have scaled the opacities of CO₂, CO, and PH₃ by factors of 1000, 100, and 1000, respectively.

For comparison to WISE 0855, we have included in Figure 3 the NIRSpec PRISM spectrum of WISE 0359 (Beiler et al. 2023), which has a spectral type of Y0 (Kirkpatrick et al. 2012). The data for WISE 0359 have been scaled to roughly match the fluxes of WISE 0855 at 4–5 μm. WISE 0359 is bluer from NIR to MIR wavelengths than WISE 0855, which reflects its earlier spectral type and higher temperature (∼450 K; Beiler et al. 2023). WISE 0359 has strong absorption from CO₂ and CO at 4–5 μm, whereas WISE 0855 lacks CO₂ and has much weaker CO (Miles et al. 2020; Section 4.5). Beiler et al. (2023) noted the detection of a relatively narrow feature at 3 μm in WISE 0359, which they attributed to the Q branch of the ν₁ band of NH₃. That feature is detected in WISE 0855 as well.

Most known Y dwarfs have been classified as Y0 or Y1 (Kirkpatrick et al. 2021). One Y dwarf, WISEPA J182831.08+265037.8, has a spectral type of ≥Y2 (Kirkpatrick et al. 2012). WISE 0855 is the one remaining Y dwarf, which has been assigned a tentative classification of Y4 based on its colors and absolute magnitudes (Kirkpatrick et al. 2019, 2021). NIRSpec has provided the first spectrum of WISE 0855 that is suitable for spectral classification. Those data are consistent with an effective temperature that is significantly cooler than those of other known Y dwarfs (Figure 3, Section 4.5), so Y4 remains a reasonable choice for its spectral type. The definition of classes later than Y1 will require JWST spectra of a larger sample of the coolest known Y dwarfs (e.g., JWST program 2302, PI: M. Cushing) and would be facilitated by the discovery of objects that more closely approach the temperature of WISE 0855.

Previous studies have compared photometry and spectroscopy of WISE 0855 to the predictions of atmospheric models in an attempt to characterize its atmosphere and test the models. Such models are categorized in part based on whether they assume chemical equilibrium or nonequilibrium chemistry due to vertical mixing, and whether they assume clear or cloudy atmospheres. Warmer Y dwarfs may have sulfide clouds (Morley et al. 2012) and show evidence of nonequilibrium chemistry (Cushing et al. 2011; Miles et al. 2020; Leggett et al. 2021; Beiler et al. 2023). A brown dwarf as cold as WISE 0855 is expected to have clouds of H₂O ice (<350 K; Burrows et al. 2003; Morley et al. 2014b) and could experience nonequilibrium chemistry given its presence in the atmosphere of Jupiter (Prinn & Lewis 1975; Visscher et al. 2006).

Early studies of WISE 0855 found that its photometric SED was poorly matched by all of the models from available grids and did not favor one model prescription over another (e.g., cloudless versus cloudy; Luhman & Esplin 2014, 2016; Schneider et al. 2016; Leggett et al. 2017). Photometric monitoring with Spitzer has measured the MIR variability of WISE 0855, which is similar to that of warmer Y dwarfs and is inconclusive regarding the presence of H₂O ice clouds (Esplin et al. 2016). Skemer et al. (2016), Morley et al. (2018), and Miles et al. (2020) found that cloudy models provided a better match to their M-band spectra than cloudless models, which was cited as evidence of H₂O clouds. However, given the untested nature of models in the temperature regime of WISE 0855, if one model prescription appears to be favored by data in a limited range of wavelengths, the same may not be true at other wavelengths or with other suites of models. Indeed, the cloudy models from those studies were discrepant in multiple bands of the photometric SED while Leggett et al. (2021) were able to use cloudless models (Phillips et al. 2020) to simultaneously reproduce the slopes of the L/M-band spectra and provide a good fit to the SED. Meanwhile, detections of PH₃ and CO would be two of the most obvious signatures of nonequilibrium chemistry. The former was not detected in the L/M-band spectra (Skemer et al. 2016; Morley et al. 2018; Miles et al. 2020), but Miles et al. (2020) reported the presence of weak CO absorption in the M-band spectrum. In their modeling of the SED of WISE 0855, Lacy & Burrows (2023) found that models with clouds and equilibrium chemistry provided the best fit.

Given that the SED of WISE 0855 has been sampled sparsely and typically with low S/Ns, and given the degeneracies and uncertainties in the model spectra, it has been unclear from modeling of previous data whether the atmosphere of WISE 0855 experiences H₂O ice clouds or nonequilibrium chemistry.

In an initial attempt at modeling the NIRSpec data for WISE 0855, we have searched for model spectra that reproduce the PRISM data from among available grids that are defined by the following features: cloudless atmospheres with either chemical equilibrium and nonequilibrium chemistry (ATMO 2020; Tremblin et al. 2015; Phillips et al. 2020; Leggett et al. 2021; Meisner et al. 2023), cloudless atmospheres and chemical equilibrium (Sonora Bobcat; Marley et al. 2021), cloudless atmospheres and nonequilibrium chemistry (Saumon et al. 2012), partly cloudy atmospheres and chemical equilibrium (Morley et al. 2012, 2014b), and cloudy and cloudless atmospheres with chemical equilibrium and nonequilibrium chemistry (Lacy & Burrows 2023). None of the grids provided a sufficiently close match to the PRISM spectrum to be useful.

For all of the model suites that we considered, it is possible that the parameters defining the grids are too sparsely sampled and that a good fit could be achieved with custom models in which the parameters are tuned for WISE 0855. We have performed such an exercise with the ATMO 2020 models for both chemical equilibrium and nonequilibrium chemistry. As an initial model, we adopted the one for WISE 0855 from Leggett et al. (2021), which had T_eff = 260 K, a surface gravity of log g = 4 [cm s⁻²], an eddy diffusion coefficient of K_zz = 10^8.7, an effective adiabatic index of γ = 1.33, and solar values for metallicity and C/O. We then increased the effective temperature to improve the fit at 2–4 μm, scaled all fluxes downward to maintain agreement near 4.5–5 μm (i.e., decreased the radius), and decreased γ to further improve the fit at <3 μm. The resulting model, which we refer to as the "base model," has T_eff = 285 K and γ = 1.3 while the remaining parameters are unchanged from the model in Leggett et al. (2021).

In Figure 5, we show a portion of the G395M spectrum that includes bands from PH₃, NH₃, and CO. The top panel compares the data to the spectrum from our base model. The prescription for nonequilibrium chemistry in that model produces mixing ratios of 7.7 × 10⁻⁵ for NH₃ and 2 × 10⁻⁷ for CO. The model does not account for mixing of PH₃, and instead uses the mixing ratio of 5 × 10⁻⁷ derived for chemical equilibrium (Phillips et al. 2020; Leggett et al. 2021). The strongest features from NH₃ and PH₃ are evident in the model spectrum at 4.2 and 4.3 μm, respectively. Additional absorption at other wavelengths from PH₃ and CO is present in the model, but it does not appear as well-defined features against the backdrop of CH₄ and H₂O bands.

Zoom In Zoom Out Reset image size

To illustrate the influence of PH₃ and CO on the model spectra, we show in Figure 5 modified versions of the base model after removing the opacities of PH₃, removing both PH₃ and CO, and reducing the mixing ratio of PH₃ to 10⁻⁸. For each of these models, the pressure–temperature profile was reconverged to maintain hydrostatic and energy equilibria. The model with no PH₃ agrees well with the data near the wavelength of the strongest feature of PH₃ (4.3 μm), indicating that PH₃ is not detected. The model that lacks both PH₃ and CO is significantly brighter than the data across the wavelength range of CO (4.5–4.9 μm), demonstrating that CO is clearly detected. Meanwhile, the spectrum in that range is well matched by the base model. The mixing ratio of CO in the base model is similar to the value derived by Miles et al. (2020) from modeling of the ground-based M-band spectrum obtained by Skemer et al. (2016). In the model with a reduced mixing ratio of 10⁻⁸ for PH₃, the 4.3 μm feature is weak but detectable, which provides a rough upper limit on the abundance of that species. In Figure 5, the models in which PH₃ has been removed or reduced are brighter than the data at 4.3–4.5 μm, which is discussed later in the context of the PRISM data.

As discussed in Miles et al. (2020), the detection of CO in WISE 0855 indicates the presence of nonequilibrium chemistry due to vertical mixing. The degree of vertical mixing needed to account for the observed CO absorption in WISE 0855 is expected to produce strong PH₃ as well, but it is not detected. Like WISE 0855, the Y dwarf WISE 0359 exhibits absorption in CO but not PH₃ in its NIRSpec data (Beiler et al. 2023).

In Figure 6, we compare the PRISM data to the spectra produced by the base model and the modified version with PH₃ removed, which provided the best match to the G395M spectrum. When normalizing the model to the data at 4.5–5 μm as we have done, the model agrees with the data at 0.8–1.6 μm, is too faint at 1.6–4 and 5–5.6 μm, and is too bright at 4.3–4.5 μm (see also Figure 5). It is possible that the fit could be improved by adjusting the abundances of some of the dominant absorbers. For instance, the abundance of NH₃ in our model may need to be reduced based on the strength of the 4.2 μm absorption feature (Figure 5). The source of the discrepancy at 4.3–4.5 μm is unclear. We note that this wavelength range corresponds to the minimum in the opacities of the dominant absorbers (Figure 6).

Zoom In Zoom Out Reset image size

The temperature for our best model for WISE 0855 (285 K) is somewhat higher than the value of 253–276 K derived by combining our luminosity estimate with the luminosities predicted by evolutionary models (Section 4.6). As an alternative illustration of the difference, the model temperature combined with our luminosity estimate correspond to a radius of 0.092 R_⊙, which is smaller than the radii of 0.10–0.11 R_⊙ that are predicted by evolutionary models for brown dwarfs at the luminosity of WISE 0855 for ages of 1–10 Gyr.

As mentioned in Section 4.1, Skemer et al. (2016), Morley et al. (2018), and Miles et al. (2020) found that their ground-based M-band spectrum of WISE 0855 was better matched by cloudy models than cloudless models, which was cited as evidence of H₂O ice clouds. However, the spectrum produced by our cloudless model agrees well with the NIRSpec data in that wavelength range (Figure 5).

Future studies can improve upon our modeling of WISE 0855 by incorporating 5–28 μm data that were collected with the Mid-infrared Instrument (MIRI; Rieke et al. 2015) on JWST during the same visit as the NIRSpec observations, optimizing the abundances of the dominant absorbers in cloudless models like those from ATMO 2020, tuning the latest cloudy models to fit WISE 0855 (Lacy & Burrows 2023), and performing retrieval analysis (Burningham et al. 2017; Line et al. 2017; Zalesky et al. 2019; Rowland et al. 2023).

According to the model spectra for WISE 0855, the NIRSpec PRISM data (0.77–5.53 μm) should encompass roughly one-third of its total flux. We have estimated the bolometric luminosity of WISE 0855 by integrating the flux in the PRISM spectrum and our 285 K model at shorter and longer wavelengths and applying the parallactic distance measured by Kirkpatrick et al. (2021), arriving at log L/L_⊙ = −7.305 ± 0.020. This calculation is similar to simply quoting the luminosity of our best-fit model. As mentioned in Section 4.1, the NIRSpec spectra have been flux calibrated using the mean of multiple epochs of photometry at 4.5 μm with Spitzer. We have adopted the standard deviation of those measurements as the uncertainty in the flux calibration. The error in our luminosity estimate does not include the uncertainty in the model flux beyond the wavelength range of the PRISM data. It will be possible to eventually eliminate most of that uncertainty by incorporating the 5–28 μm data from MIRI.

We can estimate the effective temperature of WISE 0855 from its luminosity and the radii predicted by evolutionary models. In Figure 7, we have plotted the luminosity estimate for WISE 0855 with the luminosities as a function of temperature predicted by two sets of evolutionary models for ages of 1 and 10 Gyr, which encompass most members of the solar neighborhood. We have included the luminosity for WISE 0359 from Beiler et al. (2023) for comparison. We have shown models of cloudless atmospheres with chemical equilibrium and solar metallicity (Marley et al. 2021; Chabrier et al. 2023). Very similar luminosities are produced by models with cloudy atmospheres and nonequilibrium chemistry (Saumon et al. 2012; Morley et al. 2014b; Chabrier et al. 2023). Based on the evolutionary models, the luminosity of WISE 0855 corresponds to a temperature of 253–276 K for an age of 1–10 Gyr, which is somewhat cooler than the value implied by the model spectra (Section 4.5). As mentioned earlier, the true uncertainty in our luminosity estimate is larger than the value we have quoted and plotted in Figure 7, so the same is true for the temperature estimate.

Zoom In Zoom Out Reset image size

To illustrate the expected surface gravity of WISE 0855, we have included in Figure 7 a diagram of the predicted luminosities as a function of surface gravity for 1 and 10 Gyr. The luminosity of WISE 0855 is indicative of log g ∼ 4 [cm s⁻²]. Meanwhile, the luminosity corresponds to a mass range of 3–10 M_Jup for ages of 1–10 Gyr according to the evolutionary models.