THE PHYSICAL PROPERTIES OF FOUR ∼600 K T DWARFS

The low-temperature field brown dwarfs are of interest for many reasons. They are important for studies of star formation and the initial mass function (IMF) at low masses, and of the chemistry and physics of cool atmospheres. Furthermore, they populate the substellar temperature range and can act as proxies for giant exoplanets which have similar temperatures and radii, but which have less well-understood interiors and which are harder to observe than isolated brown dwarfs.

The far-red Sloan Digital Sky Survey (SDSS; York et al. 2000), and near-infrared Deep Near-Infrared Survey of the Southern Sky (DENIS; Epchtein 1998) and Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006), led to the discovery of the field population of very low mass and low temperature stars and brown dwarfs—the L and T dwarfs (e.g., Kirkpatrick et al. 1997; Martín et al. 1999; Strauss et al. 1999; Leggett et al. 2000). The L dwarfs have effective temperatures (T_eff) of 1400–2300 K and the T dwarfs have T_eff < 1400 K (Golimowski et al. 2004; Vrba et al. 2004), with a lower limit not yet defined. Burrows et al. (2003) calculate that several changes will occur near T_eff = 400 K which may trigger a new spectral type, including the formation of H₂O clouds, strengthening of NH₃, and the increasing importance of the mid-infrared flux region.

The flux from the T dwarfs is predominantly emitted in the far-red through mid-infrared, and the classification scheme is based on the strength of the near-infrared H₂O and CH₄ absorption bands (Burgasser et al. 2006). At the time of writing, only 14 T dwarfs are known with spectral type later than T7, i.e., with T_eff less than ∼900 K (e.g., http://www.DwarfArchives.org). The near-infrared spectra of the three recently discovered T9 dwarfs (Warren et al. 2007; Burningham et al. 2008; Delorme et al. 2008a), the latest spectral type currently known, hint at an NH₃ feature at 1.5 μm and suggest that we may be at the brink of defining the next spectral class, for which the letter Y has been suggested (Kirkpatrick et al. 1999).

In this paper, we present new mid-infrared spectra for two of these T9 dwarfs, ULAS J003402.77−005206.7 and ULAS J133553.45+113005.2, obtained with the Infrared Spectrograph (IRS; Houck et al. 2004) on the Spitzer Space Telescope (SST; Werner et al. 2004). We combine these data with the previously published near-infrared spectra and mid-infrared Infrared Array Camera (IRAC; Fazio et al. 2004) to generate an almost complete spectral energy distribution (SED) for the dwarfs, and use atmospheric and evolutionary models to constrain their physical properties. We also investigate similar data for 2MASS J09393548−2448279, a T8 dwarf recently found to be very cool (Burgasser et al. 2008), and the near-infrared spectrum for the third T9 dwarf, CFBD J005910.82−011401.3.

Two T9 dwarfs were recently found in the UK Infrared Telescope (UKIRT) Infrared Deep Sky Survey (UKIDSS; Lawrence et al. 2007) and another in the far-red Canada–France–Hawaii Telescope's Brown Dwarf Survey (CFBDS; Delorme et al. 2008b): ULAS J003402.77−005206.7 (hereafter ULAS J0034−00; Warren et al. 2007), ULAS J133553.45+ 113005.2 (ULAS J1335+11; Burningham et al. 2008), and CFBD J005910.82−011401.3 (CFBD J0059−01; Delorme et al. 2008a). The two UKIDSS dwarfs were found in the 1st and 3rd data releases of the Large Area Survey (LAS) component of UKIDSS, which reaches J = 19.6. The 3rd data release covers 900 deg² of a planned 4000 deg² for the LAS. The CFBDS reaches z_AB = 22.5, and 350 deg² have been analyzed. The discovery papers show that the near-infrared spectra, as well as IRAC photometry for the two UKIDSS dwarfs, imply a very low T_eff of ∼600 K for these three dwarfs. Here, we present new data for the two UKIDSS T9 dwarfs; no new data are presented for the CFBDS dwarf but we include it in our analysis as these three dwarfs appear to be similarly cool. We also include in this analysis a dwarf discovered in the 2MASS survey by Tinney et al. (2005): 2MASS J09393548−2448279 (2MASS J0939−24). Although classified as T8 using low-resolution near-infrared spectra and methane imaging, Burgasser et al. (2008) show that mid-infrared data constrain its temperature to be similar to that of the T9 dwarfs.

Figure 1 shows the near-infrared spectra of these four dwarfs as well as the spectrum of the T8 spectral standard 2MASS J04151954−0935066 (2MASS J0415−09; Burgasser et al. 2002, 2006) which has T_eff= 725–775 K (Saumon et al. 2007). At these low temperatures very little flux remains to be absorbed in the troughs of the strong H₂O and CH₄ bands and the spectral typing based on these fluxes becomes difficult. Burningham et al. (2008), Delorme et al. (2008a), and Warren et al. (2007) show that spectral indices which measure the blue wings of the J and H flux peaks, "W_J" at ≈1.21 μm and "NH₃-H" at ≈1.55 μm, appear to be useful type indicators, and that the previously defined "H₂O-H" at ≈1.50 μm also retains some use at these extreme types (see, e.g., Figures 7–9 of Burningham et al.). It is likely that there is an NH₃ absorption feature in the blue wing of the H band for the coolest dwarfs, and Delorme et al. (2008a) suggest that these objects may be the prototype of the Y spectral class. However, the feature is difficult to distinguish from the strong H₂O and CH₄ bands and so until even later-type dwarfs are available to define the Y class we adopt a type of T9 here (as also suggested by Burningham et al.).

Zoom In Zoom Out Reset image size

The three dwarfs classified as T9 have narrower J and H (1.2 and 1.6 μm) flux peaks than 2MASS J0415−09, suggesting that they are cooler than 750 K. The near-infrared indices, including the "W_J" and "NH₃-H" indices, confirm an earlier type of T8 for 2MASS J0939−24. However, as shown by Burgasser et al. (2008) and as we demonstrate below, the mid-infrared data for 2MASS J0939−24 constrain its temperature to be similar to the T9 dwarfs. The near-infrared spectrum of 2MASS J0939−24 is very low resolution with R ≈ 150 or Δλ ≈ 0.01 μm, so that the detection of the variation in the width of the J and H bands is somewhat compromised; however, the object is further complicated by possible binarity (Burgasser et al. and Section 5.4) and we show later that multiplicity can address the discrepancy between near-infrared spectral type and temperature.

Table 1 lists astrometric and photometric data for the three T9 dwarfs and 2MASS J0939−24, taken from the sources referenced in the table. The z photometry for the UKIDSS dwarfs has been converted to the SDSS AB system using transformations given by Warren et al. (2007). The YJHK photometry is on the Mauna Kea Observatories Vega-based system (Tokunaga et al. 2002; the JHK for 2MASS J0939−24 is synthesized from its spectrum which was flux calibrated using 2MASS photometry). The IRAC photometry is reported in the standard IRAC system, which is calibrated against Vega. The IRAC values provide a flux density at a nominal wavelength assuming a flux distribution where νf_ν is constant, which is not the case for T dwarfs. Hence the IRAC magnitudes do not give the flux at the nominal wavelengths listed in Table 1, but do give a scaling factor for the spectrum over the IRAC filter bandpasses (e.g., Cushing et al. 2006).

We used the Short-Low (SL) module of the IRS to obtain first-order 7.6–15.2 μm spectra of ULAS J0034−00 and ULAS J1335+11 in the standard staring mode, which produces low-resolution spectra with λ/Δλ≈ 60–120. Wavelengths longer than ∼14.5 μm are compromised by an overlapping second order. The blue peakup band was used to center the targets in the 37 slit by offsetting from a nearby bright star, and cycles were repeated after nodding the target along the slit by roughly one- and two-thirds of its length.

ULAS J0034−00 was observed as part of program 40419 using instrument-team guaranteed time in Cycle 4. The dwarf was observed on 2008 January 15 and January 16, each time using a 240 s ramp duration and 32 cycles. The total time on source was 8.5 hr, and the total Astronomical Observation Request (AOR) duration was 10.8 hr.

The brighter dwarf ULAS J1335+11 was observed as part of program 40449 using general observer time in Cycle 4. The dwarf was observed on 2008 August 9, using a 60 s ramp duration and 64 cycles. The total time on source was 2.1 hr, and the total AOR duration was 2.8 hr.

Low-level processing of the spectra was done automatically by versions 17.2.0 and 18.1.0 of the IRS pipeline for ULAS J0034−00 and ULAS J1335+11, respectively. The IRS pipeline removed all the instrumental artifacts to produce Basic Calibrated Data (BCD) for each exposure. The data were further reduced as described in Cushing et al. (2006). Briefly, the spectra were extracted using a modified version of the Spextool package with a fixed aperture of 6''. Observations of α Lac, obtained as part of the IRS calibration observations, were used to remove the instrument response and flux calibrate the spectra. The flux calibration was checked using the IRAC band 4 (6.4–9.3 μm) photometry, extrapolating the observed spectrum blueward from 7.6 to 6.4 μm using the observed spectrum of the T8 dwarf 2MASS J0415−09 as a template. Very little flux is emitted at these wavelengths due to strong absorption by H₂O at ∼5–7 μm and CH₄ at ∼7–9 μm. This photometric check showed that the IRS calibration is accurate to ⩽10%.

Figure 2 shows the IRS spectra of ULAS J0034−00 and ULAS J1335+11.

Zoom In Zoom Out Reset image size

Atmospheric and evolutionary models were generated by members of our team (Marley et al. 2002; Saumon & Marley 2008). The atmospheric models include all the significant sources of gas opacity (Freedman et al. 2008). However, there are known deficiencies in the molecular opacity line lists at 0.9 < λ μm <1.4, where the line list for CH₄ is incomplete and the line list for NH₃ does not exist. At the temperatures considered here these opacities are significant and we return to this issue later in Sections 5 and 6. The models assume solar system abundances and calculate the equilibrium abundances of the important C-, N-, and O-bearing gases as detailed by Freedman et al. (2008). These equilibrium values are then used to determine the abundances of the various gas species in each atmospheric layer during the construction of the temperature–pressure profile. We iteratively compute models in radiative–convective equilibrium, solving for a temperature–pressure profile while employing the source-function technique of Toon et al. (1989) to solve the equation of transfer. During this process, condensates of certain refractory elements are included in atmospheric layers where physical conditions favor such condensation. At the temperatures considered here the condensate clouds are located deep below the photosphere and the opacity of the clouds is ignored although grain condensation is still accounted for in the chemistry.

The models also include vertical transport in the atmosphere, which significantly affects the chemical abundances of C-, N-, and O-bearing species. The very stable molecules CO and N₂ are dredged up from deep layers to the photosphere, causing enhanced abundances of these species and decreased abundances of CH₄, H₂O, and NH₃ (e.g., Fegley & Lodders 1996; Hubeny & Burrows 2007; Saumon et al. 2007). Vertical transport is parameterized in our models by an eddy diffusion coefficient K_zz (cm² s⁻¹) that is related to the mixing timescale. Generally, larger values of K_zz imply greater enhancement of CO and N₂ over CH₄ and NH₃, respectively. Values of log K_zz= 2–6, corresponding to mixing timescales of ∼10 yr to ∼1 hr, respectively, reproduce the observations of T dwarfs (e.g., Leggett et al. 2007b; Saumon et al. 2007).

The evolutionary sequences of Saumon & Marley (2008), computed with cloudless models as the surface-boundary condition, provide radius, mass, and age for a variety of atmospheric parameters. For this investigation, synthetic spectra were generated with T_eff of 500, 550, 600, 650, and 700 K; gravity g of 100, 300, 1000, and 2000 m s⁻² (corresponding to log g in cgs units of 4.00, 4.48, 5.00, and 5.30); metallicities of solar and twice solar ([m/H] =+0.3); and vertical mixing diffusion coefficient K_zz of zero (corresponding to equilibrium) and 10⁴ cm² s⁻¹. A limited set was also calculated with K_zz = 10⁶ cm² s⁻¹ and [m/H]=−0.3. The evolutionary models show that a brown dwarf with T_eff ⩽ 700 K must have g ⩽ 2000 m s⁻² if its age is ⩽10 Gyr (Saumon & Marley 2008).

Figure 3 plots a selection of 500 K and 600 K synthetic spectra to demonstrate the effects of varying the atmospheric parameters. The top panel shows that metallicity and gravity variations can affect the SEDs similarly—an increase in metallicity or a decrease in gravity leads to significantly increased flux in the K band at 2.2 μm, and decreased flux at 4.5 μm (impacting IRAC band 2 photometry for example). The opposite changes occur if metallicity is decreased or gravity is increased. The lower panel shows that varying T_eff impacts the overall slope of the SED—decreasing T_eff increases the ratio of the mid- to near-infrared fluxes (although surface flux at all wavelengths decreases with decreasing temperature). The lower panel also shows that the mixing parameter K_zz strongly affects the 4.5 μm and 9–15 μm fluxes: increasing K_zz reduces the 4.5 μm flux as CO is dredged up, and increases the 9–15 μm flux as N₂ is favored over NH₃. For the objects considered here, the CH₄ and H₂O bands are not significantly affected by CO mixing, as the amount of carbon and oxygen tied up in CO is a small fraction of that in these other molecules at these temperatures.

Zoom In Zoom Out Reset image size

5.1. The 0.9–2.5 μm and 7.6–14.5 μm Spectra of ULAS J0034−00 and ULAS J1335+11

The synthetic spectra were compared to the observed 0.9–2.5 μm and 7.6–14.5 μm spectra of ULAS J0034−00 and ULAS J1335+11 using the fitting procedure described by Cushing et al. (2008), except that here we do not mask out the 1.58 to 1.75 μm spectral region. No regions were ignored in the fitting, despite the known problems at 0.9 < λ μm <1.4 (Section 4). The procedure uses weighted least-squares fitting to identify the synthetic spectrum that best matches the observed spectrum. The scaling of the synthetic spectra to the observations is determined as part of the fitting procedure, by minimizing the weighted deviation between the two data sets.

Fits using the entire wavelength range as well as only the near- or mid-infrared were carried out. For both of these T9 dwarfs the fits to either the near- or mid-infrared regions alone imply a value of T_eff of 600 ⩽ T_eff ⩽ 700. However, the fit to the mid-infrared spectrum does not constrain the temperature strongly, and the fit to the near-infrared is very poor in the H band. The fits to the full spectrum give a lower temperature of 500 ⩽ T_eff ⩽ 600 for ULAS J0034−00 and 500 ⩽ T_eff ⩽ 550 for ULAS J1335+11; in these full-range fits the match to the H band is improved at the expense of a poorer fit to the K band. Increasing T_eff makes the overall spectrum bluer (e.g., Figure 3), and temperatures as high as suggested by the near-infrared fits cannot reproduce the observed full SEDs.

The best fitting model to the spectrum of ULAS J0034−00 has effective temperature (K)/gravity (m s⁻²)/metallicity ([m/H] dex) of 550/100/0.0; this fit is shown in the top panel of Figure 4. However, temperatures in the range 550–600 K, gravities of 100–300 ms⁻² and metallicities of 0.0–0.3 dex do not give significantly worse fits, as shown in the top two panels of Figures 4 and 5. The best fitting model to the spectrum of ULAS J1335+11 has effective temperature (K)/gravity (m s⁻²)/metallicity ([m/H] dex) of 500/100/0.0; this fit is shown in the top panel of Figure 6. Temperatures in the range 500–550 K, gravities of 100–300 ms⁻², and metallicities of 0.0–0.3 dex give similar quality fits as shown in Figure 6.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The full spectral range supports a low gravity for both dwarfs of 100 ⩽ g m s⁻² ⩽300. Although an increase in gravity can be partly compensated for by increasing metallicity, a gravity as high as g = 1000 m s⁻² gives a poorer least-squares fit to both dwarfs, and this can be seen for ULAS J0034−00 in the bottom panel of Figure 4.

For both dwarfs, models which include gas transport, i.e., nonequilibrium chemistry with K_zz > 0, are strongly preferred by the fits to the mid-infrared only and the fits to the entire spectral range. The equilibrium chemistry models do not reproduce the shape of the 9–15 μm spectrum, where NH₃ absorption is strong, but not as strong as it would be in the K_zz = 0 case; this is shown in the bottom panel of Figure 5. The flux at 4.5 μm is also dependent on the value of K_zz, which we discuss further when considering the IRAC photometry below. Vertical mixing in ULAS J1335+11 appears stronger than in ULAS J0034−00 based on the 4.5 μm IRAC photometry; we determine K_zz ≈ 10⁶ cm²s⁻¹ compared to 10⁴ cm²s⁻¹ for ULAS J0034−00 (Section 5.2).

5.2. The 3–9 μm Photometry of ULAS J0034−00 and ULAS J1335+11

5.3. Trends in Color

Fitting the synthetic spectra to the observations showed that the ratio of the mid- to near-infrared flux is very sensitive to temperature. Combining ground-based near-infrared photometry with IRAC photometry therefore should be a useful indicator of T_eff, in particular the [4.5] band as there is very little flux at [3.6], [5.8], and [8.0]. Warren et al. (2007) investigate various color combinations and find that H−[4.5] is a good indicator of T_eff for T dwarfs. While the K band is sensitive to metallicity and gravity, the H band is insensitive (e.g., Figure 3). Stephens et al. (2009) show that T dwarf colors involving J and IRAC magnitudes show a lot of scatter with T_eff, at least some of which is due to condensates clearing out of the photospheric region probed by 1.2 μm radiation in the earlier T dwarfs.

Figure 7 shows H−[4.5] as a function of H − K. Model sequences are shown for a range of gravity and metallicity, and for nonequilibrium chemistry models with K_zz = 10⁴ cm² s⁻¹. Increasing K_zz by a factor of 100 leads to no significant change in H − K, but a decrease in H−[4.5] of 0.1–0.3 mag. From this and other studies it appears that the range of K_zz in T dwarfs is 10²–10⁶ cm² s⁻¹ and so the uncertainty in the modeled H−[4.5] due to unknown K_zz should be ≈0.2 mag. Plausible changes in gravity and metallicity for a local disk sample also affect H−[4.5] by ∼0.2 mag and affect H − K by ∼0.3–0.6 mag at the temperatures considered here (note that the effect on H − K increases with decreasing temperature). As described before, the trends with metallicity and gravity are degenerate. Although these ∼0.2 mag uncertainties are not ideal in a temperature indicator, Figure 7 shows that for T_eff < 700 K H−[4.5] increases by 0.6–0.8 mag per 100 K and so temperatures can still be accurately determined. The Spitzer warm mission retains IRAC [4.5] capability and hence will continue to provide important data for studies of brown dwarfs.

Zoom In Zoom Out Reset image size

Colors for ULAS J0034−00, ULAS J1335+11, and 2MASS J0939−24 are shown in Figure 7, as well as those of other dwarfs taken from the literature (Patten et al. 2006; Leggett et al. 2007b; Luhman et al. 2007). The T7.5 dwarfs Gl 570D and HD 3651B are benchmark dwarfs with well constrained ages and metallicities from their primary stars, and well constrained T_eff based on their measured luminosity and the known distance to the primary star (Burgasser et al. 2000; Burgasser 2007; Liu et al. 2007; Mugrauer et al. 2006; Saumon et al. 2006). Gl 570D has T_eff= 750–825 K, g= 630–2000 m s⁻², and [m/H] = 0.01–0.10; HD 3651B has T_eff= 780–840 K, g= 1260–3160 m s⁻², and [m/H] = 0.09–0.16. 2MASS J0415−09 has a well-measured distance and SED which indicates T_eff= 725–775 K, g= 1000–2500 m s⁻², and [m/H] = 0.0–0.3 dex (Saumon et al. 2007). If the sequences shown in Figure 7 are shifted redward in H − K by ∼0.2 mag and redward in H−[4.5] by ∼0.1 mag there is good agreement with the observations of these slightly metal-rich dwarfs, and with the overall population in the diagram. The models appear to be too faint at K, and this can also be seen in Figures 4–6. Most likely this is due to some inadequacy in the CH₄ line list or the H₂ collision-induced absorption, as discussed in Saumon & Marley (2008; their Section 4.1).

The color trends shown in Figure 7 support the temperatures, gravities, and metallicities found above for ULAS J0034−00 and ULAS J1335+11: that they are relatively low-gravity dwarf with solar or enhanced metallicities, and that ULAS J1335+11 is cooler than ULAS J0034−00. The colors also support the conclusion of Burgasser et al. (2008) in their analysis of the T8 dwarf 2MASS J0939−24: that it has a similar temperature to these T9 dwarfs, but a higher gravity and/or lower metallicity. We explore fits to this object's spectral distribution in the following section.

5.4. Fitting the SED of 2MASS J0939−24

Burgasser et al. (2008) find the following parameters for 2MASS J0939−24 using near- through mid-infrared observational data: T_eff = 600 ± 50 K, g= 600 m s⁻², and [m/H] =−0.4 (the uncertainties in g and [m/H] are quite large). The same model set and a similar least-squares fitting routine were used as we use here for ULAS J0034−00 and ULAS J1335+11. In this case, however, the distance to the dwarf is known and so the scaling factor required to match the models to the observations gives the radius of the dwarf, or alternatively if the radius is known from evolutionary arguments then the scaling factor can be derived. Adopting the first approach Burgasser et al. (2008) show that the radius is larger than expected from the models, and they suggest that the dwarf is a binary.

Figure 8 shows this dwarf's observed spectroscopy, IRAC photometry, and a selection of the modeled data. We have selected models with T_eff = 600 K and values of gravity and metallicity within the range determined by the Burgasser et al. (2008) least-squares fitting. Scaling of the modeled spectra and photometry is achieved using the known distance to the dwarf and the radius from evolutionary models. We assume that either it is single or an equal luminosity unresolved pair, as best matches the data. The lower gravity fit, which would be marginally consistent with this object being single, is a significantly poorer fit to the spectra and photometry than the higher gravity fits, which require it to be a binary system. Also a warmer model with T_eff= 650 K can reproduce the near-infrared data well if the dwarf is single, but is underluminous in the mid-infrared by ∼30%. A reasonable fit can be found if indeed, as suggested by Burgasser et al., this is a pair of similar luminosity 600 K dwarfs with g ≈ 1000 m s⁻² and [m/H] =−0.3–0.0 dex.

Zoom In Zoom Out Reset image size

Given that a temperature as low as 600 K is unexpected for an object with the same near-infrared spectral type as the 750 K dwarf 2MASS J0415−09, and that the near-infrared data led Leggett et al. (2007a) to assign T_eff = 750 K to 2MASS J0939−24, we explored the possibility that this object consists of a more dissimilar pair of dwarfs. We find that a satisfactory fit is produced if 2MASS J0939−24 is composed of a 700 K and a 500 K dwarf. The gravities must be high so that the radii are small, otherwise the total flux of the system is too large. The relative gravities are also constrained by the evolutionary models, as the dwarfs presumably are coeval. Fits with g= 1000 m s⁻² for the 500 K dwarf constrains the gravity of the 700 K dwarf to 1000–2000 m s⁻². The system still appears to be somewhat metal-poor with [m/H] =−0.3–0.0 dex. Figure 9 shows composite fits for this range of parameters. To improve the agreement with the previously determined temperature of this object, and to make the temperature:spectral type correlation stronger, this dissimilar-pair solution is preferred to the equal-luminosity pair solution. The near-infrared spectrum is then dominated by the warmer dwarf (see Figure 9) and becomes consistent with the T8 spectral type. The modeled dissimilar pair has slightly wider J and H flux peaks than the equal luminosity pair, consistent with the wider peaks observed for 2MASS J0939−24 and 2MASS J0415−09 compared to the three T9 dwarfs (Figure 1).

Zoom In Zoom Out Reset image size

However, the detailed spectral shape is sensitive to gravity and metallicity as well as temperature, and the temperature:spectral type relationship will have some scatter. Hence unless the object is resolved it cannot be determined whether the equal or nonequal pair solution is correct. The physical properties of both solutions are listed in Table 2. The 500 K + 700 K pair are constrained to be high gravity and therefore quite old at 4–10 Gyr; perhaps appropriate for the slightly metal-poor nature of the system.

5.5. The 0.9–2.5 μm Spectrum of CFBD J0059−01

We have used new and published near- through mid-infrared spectroscopy and photometry to constrain the atmospheric parameters of the very low-temperature dwarfs ULAS J0034−00, ULAS J1335+11, and 2MASS J0939−24. We find strong evidence that photospheric mixing continues to be an important process, even at these cool temperatures. Any 5 μm survey for Y dwarfs (e.g., WISE or Spitzer warm missions) must take this mixing into account, as the flux will be less than predicted by equilibrium chemistry. Considering the complexity of the atmospheres, the models reproduce the data well. However, there are known deficiencies in the molecular opacity line lists, and discrepancies can be seen in Figures 4–6 and 8. The known problems are large at 0.9 < λ μm <1.4, where the line list for CH₄ is incomplete, the line list for NH₃ does not exist, and where the treatment of the red wing of the very broad 0.77 μm K i line is difficult. These problems most likely account for the excess model flux seen in our fits at 1.0 < λ μm <1.3.

Given these deficiencies, the systematic errors in our derived parameters are difficult to determine. However, the good agreement with evolutionary models for the ∼800 K brown dwarf companions to stars of known age and metallicity means that it is unlikely that the errors are larger than 100 K in T_eff, or a factor of 2 in gravity and metallicity (see the trends shown in Figures 3 and 7). Nevertheless, it is of paramount importance that the line lists for CH₄ and NH₃ are improved, now that objects as cool as 500 K are being discovered.

It is also very helpful at these low temperatures to have mid-infrared observational data. The absorption features in the near-infrared are effectively saturated, and the changes with temperature have become subtle. The mid-infrared flux level is very sensitive to temperature and can remove this degeneracy; this sensitivity increases with decreasing temperature as shown in Figure 7. Mid-infrared data offers the possibility of identifying low-temperature systems such as 2MASS J0939−24, which may be a 500 K + 700 K brown dwarf pair. We find that two of the three currently known T9 dwarfs have 550 ⩽ T_eff K ⩽600 and the third appears to be even cooler with 500 ⩽ T_eff K ⩽550. Vertical transport of gas remains significant even at these very low temperatures, so that enhanced CO opacity reduces the flux at 4.5 μm and decreased NH₃ opacity increases the flux at 9–15 μm.

The evolutionary models show that the parameters we derive for ULAS J0034−00 and ULAS J1335+11 imply that they are both very low mass and relatively young—their masses are between 5 and 20 M_Jupiter corresponding to an age range of 0.1–2.0 Gyr. The ranges for CFBD J0059−01 are larger—10–50 M_Jupiter and 0.5–10 Gyr. If 2MASS J0939−24 consists of a 500 K + 700 K binary then it is relatively old and massive: a 4–10 Gyr system composed of a 25 M_Jupiter and a 40 M_Jupiter brown dwarf pair.

This work is based on observations made with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA. Support for this work was provided by NASA through an award issued by JPL/Caltech. S.K.L.'s research is supported by the Gemini Observatory, which is operated by the Association of Universities for Research in Astronomy, Inc., on behalf of the international Gemini partnership of Argentina, Australia, Brazil, Canada, Chile, the United Kingdom, and the United States of America. E.L.M.'s research is supported by The Spanish Ministry of Science via project AYA2007-67458.