SpeX SPECTROSCOPY OF UNRESOLVED VERY LOW MASS BINARIES. I. IDENTIFICATION OF 17 CANDIDATE BINARIES STRADDLING THE L DWARF/T DWARF TRANSITION

Ninety-nine L and T dwarfs were observed by the authors over several runs spanning 2004 September through 2008 September, as summarized in Table 1. These data were acquired in a variety of weather (clear to partly cloudy) and seeing conditions (03–12), but were uniformly observed using the 05 or 08 slit aligned at parallactic angle and generally at low air mass (sec z ≲ 2). Each source was observed in multiple exposures, dithering in an ABBA pattern along the slit, with total integration times of 240–1200 s depending on source brightness and weather conditions. Nearby A0 V stars were observed for flux calibration and telluric absorption correction, and internal flat field and argon arc lamps were observed with each flux standard for pixel response and wavelength calibration. All data were reduced using the SpeXtool reduction package (Vacca et al. 2003; Cushing et al. 2004) using standard settings, as described in detail in Burgasser (2007b).

In addition to our new observations, we compiled from the literature SpeX observations of 134 L and T dwarfs with published optical and/or near-infrared spectral types spanning L3 to T8 and with data having median signal-to-noise ratios of 20 or greater over the 0.9–2.4 μm window. Those data¹³ were acquired and reduced following similar procedures as described above. Combining the new and previous observations results in a sample of 253 spectra for 233 L3–T8 dwarfs.

In order to identify and characterize unresolved binaries, we required a "clean" sample of spectral templates. An initial sample was constructed by excluding known binaries (Burgasser et al. 2007b, and references therein) young cluster objects (e.g., Muench et al. 2007), and sources specifically noted to have peculiar spectra associated with low surface gravities, subsolar metallicities or unusual cloud properties (e.g., Lucas & Roche 2000; Burgasser et al. 2003d; Knapp et al. 2004; Chiu et al. 2006; Cruz et al. 2007, 2009; Looper et al. 2008b). This left us with a sample of 189 spectra of 178 sources, whose properties are summarized in Table 2.

Figure 1 shows the distribution of published spectral types for these spectra, based on optical data for L3–L8 dwarfs or near-infrared data for L9–T8 dwarfs and those L dwarfs without optical classifications. The distribution is relatively flat, albeit with fewer T0–T3 dwarfs and T8 dwarfs reflecting the rarity of these sources in current search samples. The literature classifications are based primarily on the Kirkpatrick et al. (1999), Geballe et al. (2002), or Burgasser et al. (2006b) schemes. However, many L dwarfs have never been typed in the near-infrared, and those with near-infrared types have been classified following different schemes (e.g., Reid et al. 2001a; Testi et al. 2001; McLean et al. 2003; Nakajima et al. 2004; Burgasser 2007a). To place the sample on a self-consistent, homogeneous basis, we calculated near-infrared spectral types for each source directly from the SpeX data, using the spectral indices H₂O-J, CH₄-J, H₂O-H, CH₄-H, H₂O-K, and CH₄-K defined in Burgasser et al. (2006b) and index/spectral types relations given in Burgasser (2007a). As in the latter study, the set of indices used to classify each source is selected based on an initial (in this case, published) spectral type estimate. The H₂O-J, H₂O-H, and H₂O-K indices are combined for types L8 and earlier and the H₂O-J, CH₄-J, H₂O-H, CH₄-H, and CH₄-K indices are used for later types. As there can be significant differences between published and near-infrared spectral types, particularly for the L dwarfs (e.g., Geballe et al. 2002), we repeated the classification process three times for each source, using each iteration's classification to determine the indices used for the next iteration. The spectral type adopted for each source is the mean of its individual index types.

Zoom In Zoom Out Reset image size

Table 2 lists the index-based types, with uncertainties (standard deviation of index types) given for those sources with considerable scatter (default classification uncertainty is ±0.5 subtypes). Figure 1 shows the distribution of these classifications, which is not significantly different from the literature distribution. Figure 2 compares the SpeX and literature classifications against each other. Over the entire template sample, the standard deviation between classifications is 0.9 subtypes, although average deviations are generally larger among the L dwarfs (σ = 1.1 subtypes) than the T dwarfs (σ = 0.5 subtypes). In most cases, the discrepancies are between optical and near-infrared types for L dwarfs. Only 11 sources had SpeX classifications that differed by more than two subtypes from the published classifications, one of which is a candidate binary (2MASS J1711+2232; see Section 3.2). For completeness, we used both published and index-based spectral types in our analysis.

Zoom In Zoom Out Reset image size

Prior identification of unresolved L dwarf/T dwarf binaries from SpeX data has been based largely on the presence of peculiar spectral features, such as the sharp dip at 1.6 μm observed in the spectra of 2MASS J0320−0446, SDSS J0805+4812 (Burgasser 2007b), and Kelu-1A (Stumpf et al. 2008), attributed to overlapping FeH and CH₄ absorption features from unresolved L dwarf and T dwarf components. However, such specific features need not arise in all L/T binary combinations; see, for example, Figure 4 in Burgasser (2007a).

In an effort to make the selection process more robust, we examined the near-infrared spectra of six resolved L/T binaries (Table 3): 2MASS J0518−2828, SDSS J0423−0414 (Burgasser et al. 2005b), SDSS J1021−0304 (Burgasser et al. 2006c), 2MASS J1404+3159 (Looper et al. 2008a), SDSS J1534+1615 (Liu et al. 2006), and 2MASS J2252−1730 (Reid et al. 2006). Figure 3 compares the combined light spectrum of each of these systems to its closest match in our template library (see Section 4.2). Spectral deviations between the binaries and templates range from pronounced (in the case of 2MASS J0518−2828) to subtle (in the case of SDSS J1021-0304), but several common trends stand out.

• 1.  
  The 1.6 μm CH₄ absorption band is typically stronger relative to 2.2 μm band in the binaries, the former occasionally appearing in the absence of the latter (e.g., 2MASS J0518−2828).
• 2.  
  The 1.27 μm flux peak is more pronounced and H₂O and CH₄ absorption deeper at 1.1 μm in the binaries (e.g., SDSS J1534+1615).
• 3.  
  The 2.1 μm flux peak is slightly shifted toward the blue in the binaries (e.g., SDSS J1021−0304).
• 4.  
  The overall near-infrared spectral energy distribution is generally bluer in the binaries (e.g., 2MASS J2252−1730).

Zoom In Zoom Out Reset image size

These traits suggest qualitative means of differentiating unresolved binaries from single sources, and systems such as 2MASS J0518−2828 readily stand out in direct spectral comparisons. However, comparative methods implicitly assume that the templates themselves are "normal" single spectra, an assumption we cannot make given that we are searching for unresolved binaries within the same sample. To overcome this ambiguity, we used a suite of spectral indices to identify sources whose overall spectral properties are similar to known binaries and therefore indicative of unresolved multiplicity.

Eight spectral indices, defined in Table 4, were measured for each spectrum in our full spectral sample (templates and known binaries/peculiar sources). These include the six classification indices used in Section 2.2, the K/J flux peak ratio defined in Burgasser et al. (2006b), and an additional index sampling the 1.6 μm feature noted above. We compared all eight indices against each other for every source in our sample, and examined trends in individual indices and index ratios as a function of spectral type using both literature and index-based classifications.

From these combinations, we identified six pairings that best segregated known L dwarf/T dwarf binaries from the bulk of the spectral sample, as shown in Figure 4. These pairings reflect the peculiar traits outlined above, in particular the strength of the H-band CH₄ feature (CH₄-K versus CH₄-H and H-dip versus H₂O-H), the unusual brightness of the J-band peak (H₂O-K versus H₂O-J, H₂O-J/H₂O-H and H₂O-J/CH₄-K versus spectral type), and blue near-infrared colors (K/J versus CH₄-H). In these comparison spaces, we identified regions (Table 5) which clearly segregated known binaries, but were conservative enough so as not to over-select unusual spectra that presumably arise from non-multiplicity effects (e.g., variations in surface gravity, metallicity, or cloud properties). The criteria also allowed us to retain enough non-binary-like late-L/early-T dwarf spectra for our spectral template analysis (Section 4). These criteria clearly provide an incomplete selection of unresolved binaries, as they do not select several known binary systems with either weak signatures of CH₄ absorption (e.g., 2MASS J0850+1057 and 2MASS J1728+3948; Reid et al. 2001b; Gizis et al. 2003) or systems with identical components (e.g., SDSS J0926+584; Burgasser et al. 2006c). However, as a first attempt at identifying unresolved pairs, we chose to maximize the reliability of our candidate selection over sample completeness.

Zoom In Zoom Out Reset image size

In all, 46 sources from our full SpeX sample of 253 spectra satisfied at least one selection criterion. These include nine known binaries and all six sources listed in Table 3. We refined our selection by requiring binary candidates to satisfy at least two criteria. We also segregated "strong" candidates (satisfying at least three criteria) from "weak" candidates (satisfying only two criteria), a distinction intended to test the robustness of the spectral index criteria. These 20 sources are listed in Table 6, and constitute our initial candidate pool. All have near-infrared classifications from SpeX data spanning L8.5–T3; several have been previously noted in the literature as having peculiar or highly uncertain spectral types. We discuss each candidate in detail in Section 5.

The final test of the binary nature of our candidates and characterization of their components is a comparison of their spectra against synthetic composites generated from the template spectra. The templates, purged of binary candidates (reducing the sample to 170 spectra of 161 sources), were interpolated onto a common wavelength scale and set on an absolute flux scale (in F_λ flux units) using the MKO¹⁴ M_K/spectral type relations of Liu et al. (2006). We considered the two M_K relations defined in that study, one constructed by rejecting known binaries (hereafter, the "bright" relation) and one constructed from rejecting known and candidate binaries (hereafter, the "faint" relation; see Figure 5 in Liu et al. 2006). These two relations encompass our current best constraints on absolute magnitude trends across the L dwarf/T dwarf transition, despite diverging by nearly ∼1 mag over spectral types L8–T5. Absolute magnitudes were assigned according to either literature or index-based spectral types, resulting in four independent flux scalings for the templates. As our baseline calibration set, we adopted the faint calibration applied to literature classifications (see Section 6.1).

For each candidate, we first determined best matches to individual template spectra following a procedure similar to that outlined in Burgasser et al. (2008a). All spectra were initially normalized to the maximum flux in the 1.2–1.3 μm region. We then computed a weighted χ² statistic between each candidate (C[λ]) and template (T[λ]) spectrum:

$$\begin{equation} \chi ^2 \equiv \sum _{\lbrace \lambda \rbrace }w[{\lambda }]\left[ \frac{C[{\lambda }]-{\alpha }T[{\lambda }]}{\sigma _c[{\lambda }]} \right]^2 \end{equation} \tag{ 1 }$$

(see Cushing et al. 2008). Here, w[λ] is a vector of weights satisfying ∑_{λ}w[λ] = 1, α is a scaling factor that minimizes χ² (see Equation (2) in Cushing et al. 2008), σ_c[λ] is the noise spectrum for the candidate, and the sum is performed over the wavelength ranges {λ} = 0.95–1.35 μm, 1.45–1.8 μm, and 2.0–2.35 μm in order to avoid regions of strong telluric absorption. We adopted the same weighting scheme used in Cushing et al. (2008), with each pixel weighted by its spectral width (i.e., w_i ∝ Δλ_i).

The candidate spectra were then compared to a larger set of synthetic composites, constructed by combining all possible pairs of flux-calibrated templates for which one source was of equal or later spectral type (using either literature or index-based classifications). The resulting 13,581 composites were also normalized and compared to the candidate spectra using the same χ² statistic as Equation (1). We note that because we are comparing the candidate spectra against a finite sample of spectra of distinct sources, each of which are modulated by some degree of noise, our expectation is not to achieve χ² ≈ 1 for our best-fit cases; indeed, this expectation is not even realized in spectral model fits (see Figure 4 in Cushing et al. 2008). What matters is whether a binary template provides a significantly improved fit, as discussed below.

The number of composites used to fit the candidate spectra vastly outnumbers the number of single templates, so we are almost assured of finding a better fit (lower χ²) with the former. It is therefore necessary to assess the statistical significance of the fit improvement in order to rule out the null hypothesis; i.e., that the candidate is not a binary. We used the one-sided F-test for this purpose, using as our distribution statistic the ratio

$$\begin{equation} \eta _{\rm SB} \equiv \frac{{\rm min}\big(\big\lbrace \chi ^2_{\rm single}\big\rbrace\big)\big/\nu _{\rm single}}{{\rm min}\big(\big\lbrace \chi ^2_{\rm composite}\big\rbrace\big)\big/\nu _{\rm binary}}, \end{equation} \tag{ 2 }$$

where ν is the degrees of freedom in each fit. For both single and composite fits, one might initially assume that ν_single = ν_composite = ν is the number of data points used in the fit (N = 296), minus one to account for the relative scaling between data and template spectra. However, individual weighting of the spectral points implies that they do not contribute equally to the total χ² value. We therefore define the effective number of data points,

$$\begin{equation} N_{\rm eff} \equiv \frac{1}{{\rm max}(\lbrace w\rbrace)}\sum _{i=1}^Nw_i \end{equation} \tag{ 3 }$$

which reduces to N for w_i = constant. For our weighting scheme, N_eff = 254 and hence ν = 253. To rule out our null hypothesis at the 99% confidence level (CL), we required η_SB>1.34 as our final selection criteria.

Typically, multiple single or composite templates yield similar χ² values for a given candidate. We therefore computed mean values and uncertainties for the component parameters (i.e., spectral types and relative brightnesses) using a weighting scheme based on the F-distribution. In effect, each fit's parameter set was weighted by the likelihood that that fit is equivalent to the best fit:

$$\begin{equation} W_i \propto 1-F(\eta _{i0} \mid \nu,\nu). \end{equation} \tag{ 4 }$$

Here, η_i0≡ χ²_i/min([χ²]) is the ratio of χ² residuals between the best-fit template and the ith template, and F(η_i0 ∣ ν, ν) is the F-distribution probability distribution function. Parameter means ($\bar{p}$) and uncertainties (σ_p) were then computed as

$$\begin{equation} \bar{p} \equiv \frac{\sum _iW_ip_i}{\sum _iW_i} \end{equation} \tag{ 5 }$$

and

$$\begin{equation} \sigma _p^2 = \frac{\sum _iW_i(p_i-\bar{p})^2}{\sum _iW_i}, \end{equation} \tag{ 6 }$$

where the sums are over all fits.

Before proceeding to examine our binary candidates, we first assessed how well our index selection and spectral fitting procedures identified and reproduced the properties of the known binary systems listed in Table 3. The best-fit templates for these sources are shown in Figure 5, while Table 5 summarizes the component parameters based on the faint calibration scale. For each system, best-fit composites provided a statistically significant better match (CL > 99%) to the combined light spectrum than the best-fit single template. Visual inspection of these fits also indicates clear improvement in the cases of 2MASS J2252−1730, SDSS J0423−0414, 2MASS J0518−2828, and 2MASS J1404+3159. The fits for SDSS J1021−0304 and SDSS J1534+1615 are more subtle improvements, but notably reproduce the blueshifted peak at the K band in the spectrum of the former and the strong 1.1 μm H₂O/CH₄ band in the spectrum of the latter.

Zoom In Zoom Out Reset image size

Resolved photometric measurements have been made for these systems, so we examined how well our technique reproduces reported relative magnitudes. For the composites, synthetic MKO JHK and HST/NICMOS F110W and F170M photometry were computed directly from the flux-calibrated SpeX data by integrating these and a Kurucz model spectrum of Vega with the respective filter profiles (see Cushing et al. 2005). Mean relative magnitudes and their uncertainties were calculated as in Equations (5) and (6), and are also listed in Table 7. In all six cases, synthetic photometry is consistent with measurements to within 2σ, although uncertainties in the former are as high as 0.8 mag. For the two sources with the most reliable photometry, SDSS J0423−0414 and 2MASS J1404+3159, the agreement is within 1.5σ.

Finally, while none of the known binaries listed in Table 3 have reported resolved spectroscopy, the inferred component types are generally consistent with estimated types from the literature (±1 subtype) with the exceptions of 2MASS J0518−2828, for which we infer later-type components (L7.5+T5 versus L6:+T4:), and DENIS J2252−1730, for which we infer an earlier primary and later secondary (L4.5+T4.5 versus L6+T2). However, reported types for the former are highly uncertain (Cruz et al. 2004; Burgasser et al. 2006c) while our spectral fit to the latter is superior to that presented in Reid et al. (2006).

In summary, we find that our method is capable of robustly identifying and characterizing known L dwarf/T dwarf transition pairs, providing a measure of confidence in our candidate binary selection and component characterization.

5.1.1. SDSS J024749.90−163112.6

5.1.2. SDSS J035104.37+481046.8

5.1.3. SDSS J103931.35+325625.5

5.1.4. 2MASS J11061197+2754225

5.1.5. 2MASS J13243559+6358284

5.1.6. SDSS J141530.05+572428.7

5.1.7. SDSS J143553.25+112948.6

5.1.8. SDSS J143945.86+304220.6

5.1.9. SDSS J151114.66+060742.9

5.1.10. SDSS J151603.03+025928.9

5.1.11. 2MASS J1711457+223204

5.1.12. 2MASS J21392676+0220226

5.2.1. SDSS J011912.22+240331.6

5.2.2. SDSS J075840.33+324723.4

5.2.3. SDSS J090900.73+652527.2

5.2.4. 2MASS J09490860−1545485

5.2.5. SDSS J120602.51+281328.7

5.2.6. SDSS J120747.17+024424.8

5.2.7. SDSS J151643.01+305344.4

5.2.8. SDSS J205235.31−160929.8

5.2.9. Summary

As discussed in Section 1, there remains considerable uncertainty in absolute magnitude trends across the L dwarf/T dwarf transition due to the presence of unresolved multiples, gravity and metallicity effects, and ultimately the behavior of condensate clouds as they disperse out of the photosphere. With a large number of candidate binaries in our sample whose components straddle this transition, we considered whether it was possible to use these systems to better constrain the trends.

We first considered the quality of fits provided by the composites for the two flux calibration relations of Liu et al. (2006), which differ by nearly a full magnitude between types L7 and T5. For the 17 sources identified as promising binary candidates, we find that for 65% (11/17) of these the bright relation provides a lower χ² value for the best-fit composite (the same fraction is derived whether literature or index-based classifications are used). However, for only one case—SDSS J1207+0244—does the bright relation provide a statistically significant better fit; i.e., CL > 99% in the ratio of minimum χ² values between the best-fit composites for the bright and faint relations. On the other hand, for SDSS J1039+3256 and 2MASS J1324+6358 the bright relation provides a statistically significant worse fit (CL < 1%). Figure 8 compares the best-fit composites based on both faint and bright flux calibrations for these three sources, and indeed the lower χ² fits are clear improvements. All three of these systems also contain a T2–T4 secondary, within the spectral type range where the faint and bright relations of Liu et al. (2006) differ the most. Given the inconsistency in which relation provides better fits for our candidate binaries, we cannot definitely select one over the other on this analysis alone.

Zoom In Zoom Out Reset image size

A more direct assessment can be made by focusing on individual brown dwarfs (singles and binary components) with parallax distance measurements. Figure 9 displays absolute MKO J magnitudes¹⁶ for 58 L and T dwarfs with measured parallaxes and absolute magnitude errors ⩽0.3 mag. Eighteen of these sources are known binaries; we include component magnitudes of the only candidate binary in our sample with a parallax measurement, 2MASS J1711+2232, along with measured component magnitudes for SDSS J0423-0414, SDSS J1021−0304, and Indi BC (McCaughrean et al. 2004). Nearly all of the unresolved sources and binary components follow the faint M_J relation up to spectral type ∼T2; only the L9.5 2MASS J0328+2302 stands out as clearly overluminous in this range, with M_J magnitudes comparable to combined light photometry for Gliese 337CD. Between types T2 and T5, there are only two unresolved sources with parallax measurements, SDSS J1750+1759 (T3.5) and 2MASS J0559−1404 (T4.5), both of which lie well above the faint relation; indeed, the latter lies ∼0.6 mag above the bright M_J relation. This could indicate a dramatic brightening over this spectral type range, as suggested by the component brightnesses of the resolved binary 2MASS J1404+3159 (Looper et al. 2008a) and our unresolved candidate SDSS J0119+2403 (Section 5.2.1). On the other hand, SDSS J1750+1759 and 2MASS J0559-1404 have both been suggested as unresolved multiples (Golimowski et al. 2004; Liu et al. 2006; Stephens et al. 2009), although neither satisfy any of our selection criteria. We therefore cannot rule out the possibility that they are young or otherwise peculiar sources.

Zoom In Zoom Out Reset image size

The paucity of absolute magnitude measurements in the T2–T5 range makes it impossible to determine at this point whether absolute J magnitudes remain relatively flat across the entire L/T transition (a "J-band plateau;" Liu et al. 2006), favoring of modest flux reversal and more gradual cloud depletion; or undergo a sharp increase over types T2–T5 (a "J-band spike;" Looper et al. 2008a), requiring very rapid cloud depletion. Indeed, neither of these scenarios may encompass the behavior of all brown dwarfs. Increasing the number of T2–T5 dwarfs with parallax measurements, and obtaining resolved photometry of the candidates presented here, are both necessary if we are to properly constrain trends and distributions of absolute magnitudes across the L/T transition.

Higher rates of multiplicity among early-type T dwarfs have been observed in resolved imaging studies (Burgasser et al. 2006c), although statistics in these samples remain poor due to small numbers (Goldman et al. 2008). As our candidate sample includes sources that might be undetectable in high resolution imaging, it is relevant to assess how their inclusion in multiplicity statistics modifies brown dwarf binary fraction trends across the L/T transition, and what constraint can be made on the intrinsic binary fraction.

As discussed in Section 2, our spectral sample is neither complete nor volume-limited, and selection biases are uncertain given that sources were identified from different search programs using different color and magnitude selection criteria. However, as a first approach we made the simplifying assumption that the sample is roughly magnitude-limited, and estimated the fraction of L8–T6 dwarfs with SpeX data that are either resolved or candidate binaries. These numbers are summarized in Table 11, and indicate the same high multiplicity rate among early-type T dwarfs as found in resolved imaging studies. For T0–T4 dwarfs, we deduce a binary fraction of 53% ± 7% (23/43 sources), peaking above 60% for T0–T2 dwarfs alone. Over half of the early-type T dwarfs in this spectral sample appear to be multiples, compared to ∼30% in resolved imaging samples (Burgasser 2007a). Moreover, the fraction inferred here is likely to be a lower limit, given that we are unable to identify unresolved binaries with identical components or combinations that fall outside our spectral index selection criteria (see Section 3.2).

Such a high apparent rate of multiplicity does not necessary translate into a high intrinsic (i.e., volume-limited) rate, however. As discussed in Burgasser (2007a), the enhanced binary fraction of early-type T dwarfs arises predominantly from the small change in luminosity across the L dwarf/T dwarf transition, which induces both a sharp minimum in the luminosity function of brown dwarfs at type T0 (roughly three times rarer in number than early- and mid-type L dwarfs) and makes late-type L and early-type T dwarfs comparably bright in the near-infrared. Combined with the preferential selection of binaries in magnitude-limited samples, these effects collude to amplify the apparent multiplicity rate of early-type T dwarfs. To infer the intrinsic rate, we reproduced the Monte Carlo population simulations of Burgasser (2007a), using the baseline parameters defined in that study and assuming a magnitude-limited sample. We find that an intrinsic rate of only 15% can produce an apparent binary fraction of 56% for T0–T4 dwarfs in a magnitude-limited sample, comparable to the fraction estimated above. This intrinsic rate is in fact on the low end of current estimates that attempt to correct for unresolved multiplicity (e.g., Basri & Reiners 2006; Lodieu et al. 2007; Joergens 2008), and remains consistent with a brown dwarf multiplicity fraction that is significantly lower than those of more massive stars (e.g., Duquennoy & Mayor 1991; Fischer & Marcy 1992). We reiterate, however, that our result constitutes a lower limit, and a robust measure requires detailed modeling of selection effects (e.g., component spectral type sensitivity range, mass ratio sensitivity as a function of age) that are beyond the scope of this paper.

The physical feature that ties both absolute magnitude trends and enhanced binary fractions across the L/T transition is the "sudden" depletion of condensate clouds from the photosphere. This is particularly evident in the increased surface fluxes at the 1.05 μm and 1.27 μm flux peaks where cloud particles are a dominant source of opacity (Ackerman & Marley 2001). Among the binary candidates identified here, 59% (10/17) of the entire sample and 82% (9/11) of candidates with L7–T5 components show similar flux reversals in the J-band region. While these reversals are dictated in part by our adopted flux calibration (which are, however, tied to the K-band flux; see Section 4.1), the quality of the fits coupled with observed absolute magnitude/spectral type trends (Section 6.1) suggest that these flux reversals are real.

With a relatively large sample of possible L dwarf/T dwarf pairs, we can also assess how these flux reversals vary between systems with similar spectral compositions. Consider the candidates SDSS J0119+2403 and SDSS J1106+2754, which have component types of T0+T4 and T0+T4.5 and relative brightnesses of ΔJ = −0.42± 0.19 and −0.37 ± 0.06, respectively. Examination of the best-fit composites for these sources (Figures 6 and 7) reveals that the former has a far more pronounced flux reversal than the latter, with the secondary of SDSS J0119+2403 being potentially brighter even at the 1.6 μm flux peak. The best-fit primary of this system was inferred to be an unusually cloudy source, which translates in greater opacity in the J-band region. In contrast, the primary of 2MASS J1106+2754 was inferred to be slightly blue for its spectral type, which could be attributed to thinner condensate clouds (e.g., Knapp et al. 2004; Burgasser et al. 2008b). If clouds are the main driver for the unusual colors of the primary components, and their secondaries are roughly equivalent, the different flux reversals can be tied directly to variations in cloud properties. Specifically, an initially cloudy brown dwarf should lose more opacity in the J-band region, and experience a greater increase in brightness, as it evolves across the L/T transition as compared to a brown dwarf that initially has thinner clouds. This correlation between initial cloud content and degree of flux reversal suggests that absolute magnitude trends may have more spread across the L/T transition than can be discerned in the currently sparse data set; spread has also been suggested to arise from age/surface gravity effects (e.g., Metchev & Hillenbrand 2006; Mohanty et al. 2007) and metallicity variations (e.g., Burrows et al. 2006). Again, resolved studies of confirmed binaries in our sample would provide insight into how atmospheric properties modulate the removal of condensate clouds.

Despite the large number of potential L dwarf/T dwarf pairs uncovered in this investigation, there are clearly limitations as pertaining to the completeness of the sample and robustness of the component characterizations. We reiterate that the 17 candidates listed in Table 10 may not represent the full complement of potential binaries in our spectral sample. Our selection criteria were specifically chosen to be conservative so as to not eliminate too many spectral templates, particularly over the L8–T5 spectral type range. As such, we failed to select known binary systems with early-type T dwarf classifications (Section 3.2) and a number of suspected binary pairs such as 2MASS J0328+2302. We are also unable to select systems with similar spectral component types, and are likely biased against higher-order multiples such as Kelu 1 (Stumpf et al. 2008) and DENIS-P J020529.0–115925 (Bouy et al. 2005). The retention of unresolved binaries in our template sample may also skew the statistical significance of our candidates. For instance, our rejected candidate SDSS J1206+2813, whose spectrum is a near clone to that of SDSS J1750+1759, may prove to be a binary after all should the latter be determined as such. On the other hand, incomplete sampling of single L dwarf/T dwarf transition objects in our spectral template library may cause us to over-select candidates, a possibility for the poorly fit candidate 2MASS J2139+0220. It is in this vein that we encourage follow-up high resolution imaging (or second epoch imaging in the cases of 2MASS J1106+2754 and 2MASS J1711+2232) and spectroscopy of the candidates listed in Table 10 to both verify their binary nature and better characterize their components.