Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

Remarkably the discoveries of the first exoplanet orbiting a normal star (51 Peg b) and the first uncontested field brown dwarf (Gliese 229B) were announced on the same day, October 6, 1995, in Florence, Italy at the 9th Cambridge Workshop on Cool Stars, Stellar Systems, and the Sun. Brown dwarfs are low-mass objects whose central temperatures are too cool for them to reach an equilibrium configuration where the surface radiative losses are balanced by the energy generated by hydrogen fusion in their cores (Kumar 1963; Hayashi and Nakano 1963). Of course without an internal energy source, brown dwarfs cool over time “like dying embers plucked from a fire” (Burrows et al. 2001). Consequently they obey a mass-luminosity-age relation rather than a mass-luminosity relation like the hotter hydrogen-fusing stars. The study and characterization of the more than two thousand brown dwarfs and exoplanets discovered since 1995 has become a major focus of both the stellar and planetary astrophysics communities. Our understanding of their properties has evolved in parallel because “[w]hatever the mass (M) or origin of an extrasolar giant planet or brown dwarf, the same physics, chemistry, and compositions obtain for both” (Burrows et al. 2001). That is, when we learn something about brown dwarfs, we learn something about gas giant planets, and vice versa.

The emergent spectra of brown dwarfs are so distinct from those of the cool M stars, that the creation of three new spectral classes, L, T, and most recently Y, was required in order to classify them. In this chapter, I present a historical review of the discovery of the L, T, and Y dwarfs. In particular, I will review the underlying concepts of spectral classification as embodied in the MK System (Sect. 3) and then discuss how the L, T, and Y classification schemes were devised (Sects. 4–6). Finally, I will briefly touch on some of the underlying physics that give rise to the different spectral classes (Sect. 7) and note some outstanding issues that require future attention (Sect. 8). For more detail on these subjects, as well as a discussion of many other aspects of brown dwarf astrophysics, the interested reader is referred to the excellent reviews by J. D. Kirkpatrick and A. J. Burgasser in Stellar Spectral Classification (Gray and Corbally 2009), Basri (2000), and Kirkpatrick (2005). For a more theoretical perspective on brown dwarfs please see Stevenson (1991), Burrows and Liebert (1993), Allard et al. (1997), Chabrier and Baraffe (2000), Burrows et al. (2001), and the chapter by I. Baraffe in this book.

2 On the Precipice of the Main Sequence

Over the course of the twentieth century, astronomers slowly extended the M-dwarf spectral sequence to later M subtypes and cooler effective temperatures (e.g., Adams et al. 1926; Morgan 1938; Kuiper 1942; Joy 1947; Boeshaar 1976; Boeshaar and Tyson 1985).Footnote 1 The early classification schemes were devised in the visible region (3,900–7,000 Å) and used the strengths of the titanium oxide (TiO) bands as subtype discriminators. However, M-dwarf spectral energy distributions peak beyond the visible wavelength range so once detectors were invented that were capable of collecting light at λ  > 7,000 Å, the classification schemes naturally moved to the red optical (6,000–10,000 Å) to take advantage of the increased number of photons (e.g., Bessell 1991; Kirkpatrick et al. 1991). The most widely used M-dwarf classification system is that of Kirkpatrick et al. (1991) which classifies M dwarfs from M0 to M9 based on the strengths of the TiO and vanadium oxide (VO) bands over the 6,300–9,000 Å wavelength range.

One of the motivations in the search for cooler and cooler dwarf stars was to eventually breach the barrier between the bottom of the hydrogen-burning main sequence and the realm of the brown dwarfs. However, it was not until more sensitive red and near-infrared (1–2.5 μm) detectors were developed and deployed in the late 1980s and early 1990s that brown dwarfs were finally discovered (see chapters by R. Rebolo, G. Basri, and B. Oppenheimer in this book). Modern evolutionary theory places the edge of the main sequence at roughly M  ≈ 0.075 M ⊙ (78.5 M Jup), L bol  ≈ 6 × 10−5  L ⊙, and T eff  ≈ 1,600–1,750 K for solar metallicity (Burrows et al. 2001) which means that L dwarfs encompass both very low-mass stars and brown dwarfs while T and Y dwarfs are exclusively brown dwarfs. Collectively, any star or brown dwarf with a spectral type later than M6 is known as an “ultracool” dwarf. Particularly germane in the context of the new spectral classes was the discovery of two ultracool dwarfs, GD 165B and Gliese 229B, because they became the archetypes of the L and T spectral classes.

Fig. 1
figure 1

Discovery images of the L-, T-, and Y-dwarf archetypes. Left: The roughly 7′ ′× 7′ ′ image of GD 165 A and B was taken in the K band (Becklin and Zuckerman 1988). GD 165B is the brighter of the two components. Middle: The \(2{5}^{{\prime\prime}}\times \ 2{5}^{{\prime\prime}}\) image of Gliese 229 A and B was taken in the z band (Nakajima et al. 1995). Gliese 229B can be seen as a faint companion at the 7 o’clock position. Right: A \({2}^{{\prime}}\times \ {2}^{{\prime}}\) three-color image of WISE J182831.08+265037.7 obtained with Wide-field Infrared Survey Explorer (WISE) (Cushing et al. 2011). Blue corresponds to the W1 band (3.4 μm), green to W2 (4.6 μm), and red to W3 (12 μm). All images are oriented so that North is up and East is to the left. The images of GD 165B and Gliese 229B are reprinted by permission from Mcmillan Publishers Ltd: (Nature), copyright 1988, 1995

Full size image
Fig. 2
figure 2

Left: Red optical spectrum of BRI 0021–0214 (M9.5 V, Reid et al. 1999), GD 165B (Kirkpatrick et al. 1999a), and Gliese 229B (Oppenheimer et al. 1998). Spectra are normalized to unity and offset with constants (dotted lines). Right: Near-infrared spectra of BRI 0021–0214 (M9.5 V, Cushing et al. 2005), GD 165B (Kirkpatrick et al. 1999a), and Gliese 229B (Geballe et al. 1996). Wavelengths of high telluric absorption, and thus low signal-to-noise or missing data, are denoted as grey bars. The spectra are normalized to unity at 1.27 μm and offset with constants (dotted lines). Prominent atomic and molecular absorption features are indicated

Full size image

GD 165B was discovered as part of a near-infrared survey for low-mass star and brown dwarf companions to white dwarfs (see Fig. 1 for the discovery image; Becklin and Zuckerman 1988). Later Kirkpatrick et al. (1993) obtained a spectrum of GD 165B over the 6,300–9,000 Å wavelength range and noted that the spectrum lacked the prominent TiO and VO absorption bands typically found in the spectra of late-type M dwarfs. Thus they were “forced to delay classification until other such objects are discovered.”Footnote 2 The left panel of Fig. 2 shows the red spectrum of GD 165B along with the spectrum of BRI 0021–0214, an M9.5 dwarf. The spectrum of GD 165B not only lacks the prominent TiO (7,053, 7,589, 8,206, 8,432, 8,859 Å) and VO ( ∼ 7,334, ∼ 7,851 Å) bands present in the spectrum of BRI 0021–0214 but also exhibits the 8,611 Å band head of cromium hydride (CrH), the 8,692 Å band head of iron hydride (FeH), and the additional alkali lines of Rb (7,800, 7,948 Å) and Cs (8,521 Å). The K I resonance doublet (7,665, 7,699 Å) also appears heavily broadened such that its profile appears as a single, broad absorption feature. These spectral features would become the hallmark features of the new L spectral class.

Gliese 229B was discovered as a companion to the M1 dwarf Gliese 229A in a search for brown dwarf companions to nearby stars (Nakajima et al. 1995)Footnote 3 and one of its discovery images is shown in the middle panel of Fig. 1. Its near-infrared spectrum, along with that of BRI 0021–0214 and GD 165B is shown in the right panel of Fig. 2. The infrared spectra of late-type M dwarfs are characterized by broad absorption bands of water (H2O; centered at 0.95, 1.1, 1.4, 1.8, and 2.3 μm), deep K I (1.169, 1.177, 1.244, 1.253 μm) and Na I (1.138, 1.140 μm) lines in the J band, and prominent Δ ν  = + 2 CO overtone band heads beginning at 2.29 μm. Similar spectral features are seen in the infrared spectra of L dwarfs, i.e. GD 165B. In stark contrast, the spectrum of Gliese 229B exhibits deep absorption bands of methane (CH4; centered at 1.15, 1.7, and 2.3 μm) and collision-induced absorption (CIA) of molecular hydrogen (H2; 1.8–2.8 μm)Footnote 4 and is qualitatively similar to the spectrum of Jupiter (e.g., Oppenheimer et al. 1995).Footnote 5 These spectral features – H2O, CH4, and CIA H2 – would become the hallmark features of the new T spectral class.

Although the brown dwarf status of GD 165B remains uncertain to this day, there was very little doubt that Gliese 229B was a bona fide brown dwarf. Chemical equilibrium calculations indicate that the two dominant carbon-bearing species are CO and CH4 with CO dominating at higher temperatures and CH4 dominating at lower temperatures. The two have equal abundances at a temperature of roughly 1,100 K at a pressure of 1 bar (e.g. Lodders 1999). Given the prominent CH4 absorption bands present in its spectrum, Gliese 229B’s effective temperature must therefore be below ∼ 1,100 K which, in combination with its low bolometric luminosity of ≤ 10−5  L ⊙, was a solid indication of its substellar nature.

Although it was clear that at least one new spectral class was going to be required in order to classify GD 165B- and Gliese 229B-like objects, the choice of letter(s) was not. Martín et al. (1997) were the first to propose ‘L’ as the new spectral class because it would be suggestive of “Low-temperature”. They further suggested that objects confirmed as brown dwarfs via the so-called lithium test (see chapters by R. Rebolo and G. Basri) could be designated ‘L Li ’ and that objects with spectra similar to that of Gliese 229B could be designated as ‘L\(_{CH_{4}}\)’. In contrast, Kirkpatrick (1998) suggested that two new letters were required, one for GD 165B-like objects, and one for Gliese 229B-like objects. Kirkpatrick et al. (1999b) describe in detail how they settled on ‘L’ for the GD 165B-like spectra and ‘T’ for the Gliese 229B-like spectra. Table 1 is a reproduction of table 5 from Kirkpatrick et al. (1999b) that lists the 26 Latin letters, their usage in astronomical parlance, and a note on the suitability of the letter as a new spectral class. Based on this analysis, Kirkpatrick et al. suggested the letters ‘H’, ‘L’, ‘T’, and ‘Y’ were possible choices for the new spectral classes. There was already consensus on ‘L’ for at least one class, but Martín et al. (1999) suggested that ‘H’ be used for the Gliese 229B-like objects because ‘T’ could be confused with T Tauri stars and T-associations. However, ‘H’ was never adopted and the Gliese 229B-like brown dwarfs became known as T dwarfs.

Table 1 The Latin letters and their usage in Astronomy
Full size table

The discovery of GD 165B and Gliese 229B in targeted searches of nearby stars hinted that astronomers were on the precipice of discovering a large population of solivagant field L and T dwarfs if only large areas of the sky could be surveyed at red optical and/or near-infrared wavelengths. Fortunately, three such surveys came online in the last decade of the twentieth century: the Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006) which surveyed the entire sky at J (1.25 μm), H (1.65 μm), and K s (2.16 μm), the Deep Near Infrared Southern Sky Survey (DENIS; Epchtein et al. 1997) which surveyed the southern sky at I (0.85 μm), J (1.25 μm), and K s (2.15 μm), and the Sloan Digital Sky Survey (SDSS; York et al. 2000) which surveyed roughly 10,000 deg2 of the northern hemisphere at u′ (3,596 Å), g′ (4,639 Å), r′ (6,122 Å), i′ (7,439 Å), and z′ (8,896 Å). With deep red and infrared observations of large swathes of sky in hand, the stage was set for the first substantive change/addition to the MK System in nearly 50 years.

3 Spectral Classification Precepts

Before describing the discovery of the L, T, and Y dwarfs, it is important to review the roots of the modern stellar classification scheme because it is the foundation on which these new spectral classes were appended. The seemingly odd collection of letters that comprise the modern stellar spectral classes first appeared in the The Draper Catalogue of stellar spectra photographed with the 8-inch Bache telescope as a part of the Henry Draper memorial (Pickering 1890). These letters, OBAFGKM, were actually just seven of the thirteen letter types used by Williamina Fleming to classify the initial 10,351 stellar spectra obtained as part of the Harvard Observatory all-sky spectroscopic survey initiated by Edward C. Pickering in 1885. It was Annie Jump Cannon who first placed them into their now familiar order after removing six of the thirteen Fleming types and reordering the remaining letters in her effort to classify the brightest southern-hemisphere stars (Cannon and Pickering 1901). Ms. Cannon also initiated the practice of subdividing each spectral class using Arabic numerals, e.g. B1. The final component of a modern spectral type, the luminosity class, was added by Morgan et al. (1943) in their seminal work entitled An Atlas of Stellar Spectra. It is this classification system that forms the basis of what is now known as the MK System of spectral classification.

Although most astronomers have a vague notion of how the classification of stars proceeds, it is nevertheless prudent to review how this classification scheme was set up and how the spectrum of an unknown source is classified. The process by which the MK System was created is now known as the ‘MK Process’. The salient details of this process were described in detail by one of the originators of the MK system, W.W. Morgan (1984):

  1. 1.

    “By ‘MK Process,’ we label a specific methodology that makes possible the construction and use of systems of classification based on the particular observed characteristics of stellar spectra that have been selected to define the frames of reference. These systems must be autonomous; that is, they are to be defined completely by the appearance of the spectral features in arrays of standard stellar spectra, in a specified interval of wavelength.”

  2. 2.

    “Each of these autonomous systems must also be self-consistent; that is the array of individual standard stars must constitute - and define - an orderly assemblage, from the point of view of the behavior of the spectral lines, bands and patterns, within the spectral intervals of the standard array.”

  3. 3.

    “The autonomy of each array is achieved through its liberation from dependence on the results of stellar-atmospheric computations - or on any other theoretical models.”

  4. 4.

    “Each new system is to be defined by a network of boxes (as in the MK System); and each of the boxes is to be defined by a specific stellar spectrum in a specific wavelength interval.”

Some clarifying comments are in order. First, although it is not stated explicitly, precept #1 should be amended to include “at a specified spectral resolution” since the visibility of weak, yet potentially important spectral features is a function of the spectral resolution. For example, the T spectral class (see Sect. 5) is defined by the appearance of overtone and combination bands of CH4 in the H and K bands at low spectral resolving power (R \(\equiv \) λ∕Δ λ ≈ 150). However, weak CH4 absorption features are clearly detectable at spectral types earlier than T0 in more moderate resolution R  ≈ 2,000 spectra (e.g., McLean et al. 2001; Cushing et al. 2005). Yet these brown dwarfs are not classified as T dwarfs because the T-dwarf classification system is defined at low spectral resolution. Precept #2 can be amplified by imagining the spectral sequence in an old-fashioned flip book, where each subtype appears only slightly different than the subtype before or after it. When viewed in rapid succession, the spectral sequence shows a smooth variation in spectral morphology. Precept #3 is particularly germane in the creation of classification schemes for ultracool dwarfs. Although great strides in our understanding of ultracool atmospheres have been made in the last two decades, they remain particularly difficult to model because the opacity due to millions of overlapping molecular lines and a prescription for condensate (i.e. grain) particle growth and sedimentation are required (see Sect. 7). For example, Cushing et al. (2008) used the model atmospheres of Marley and Saumon to fit the 0.6–14.5 μm spectra of a sample of L and T dwarfs and found large variations in the derived effective temperatures depending on what wavelength range was used. Clearly a classification system based on model-derived effective temperatures is undesirable since it would be constantly changing as the model atmospheres improved. What makes a spectral type so useful is that it is forever fixed and independent of models. The calibration of spectral type as a function of physical parameters such as mass, radius, effective temperature, surface gravity, etc. is done after the purely observational classification system has been devised.

Once a grid of spectral standards has been defined, how does one go about classifying the spectrum of an unknown source? W.W. Morgan again provides the answer (Morgan 1984):

  1. 5.

    ‘The classification of an unknown stellar spectrum makes use of all features (lines, bands, blends, patterns) within the specified wavelength interval. The classification act itself consists of comparisons with the series of standard spectra that define the boxes, with the question: “Is the unknown spectrum (x) ‘like’ or ‘not like’ this particular standard spectrum?”’

Note that the entire spectrum over the specified wavelength range is used in the process of classifying an unknown source.

The use of spectral indices, ratios of integrated or average fluxes in two different wavelength intervals, to quantify the variations in the strengths of various absorption lines and bands has become common practice in the brown dwarf community. Indeed spectral indices are central to some of the classification schemes described in the following sections. However useful spectral indices are for capturing how a particular spectral feature changes quantitatively, their use as a primary spectral type indicator violates precept #5 of the MK Process.

4 The L Dwarfs

When GD 165B was discovered in 1988, it was a unique object and therefore defied classification. However, once the 2MASS and DENIS surveys began in earnest and discovered tens of objects with similar red optical spectra, it was clear that a new spectral class was indeed required. Two different classification schemes using the 6,000–10,000 Å wavelength range emerged based on discoveries from 2MASS and DENIS (Kirkpatrick et al. 1999b; Martín et al. 1999).

Table 2 MK-based LTY spectral standards
Full size table

Kirkpatrick et al. (1999b) presented an L-dwarf scheme based primarily on 2MASS discoveries that is rooted in the MK Process and which consists of nine spectral standards ranging from L0 to L8. Table 2 lists the red optical L-dwarf spectral standards and the left panel of Fig. 3 presents the spectral sequence. The standards were chosen to constitute an “orderly assemblage” based on the behavior of the various spectral features. In particular, the L-dwarf sequence is marked by weakening oxide (TiO, VO) bands, enhanced hydride (CrH, FeH) bands, and enhanced alkalai lines of Rb I and Cs I. Kirkpatrick et al. also devised spectral indices that measure the strengths of the various absorption features and the slope (redness) of the spectrum. A final spectral type is calculated by the median of types given by four of the ratios.

Martín et al. (1999) presented a second classification scheme for the L dwarfs based primarily on discoveries from DENIS. They defined seven subtypes ranging from L0 to L6 but in contrast to the MK Process, the subtypes were assigned a difference in effective temperature of 100 K based on the effective temperature estimates of Basri et al. (2000). The spectral type of an unknown source is determined by computing a single spectral index, PC3, that measures the spectral slope from 7,540 to 8,270 Å. Although spectral types from both systems appear in the literature, the Kirkpatrick et al. (1999b) system is the most widely used red optical L-dwarf classification system.

The peak of the spectral energy distributions of the L dwarfs is in the near-infrared so a classification scheme at these wavelengths would be optimal. The near-infrared spectra of L dwarfs are qualitatively similar to those of the late-type M dwarfs in that they exhibit deep bands of H2O, strong alkali lines of K I and Na I, band heads of FeH, and overtone bands of CO (see Fig. 2). Reid et al. (2001) were the first to show that the near-infrared spectra of L dwarfs formed a smooth spectral sequence when ordered by their Kirkpatrick et al. (1999b) red optical spectral types. Over the years, various spectral indices were devised to measure the depths of the various molecular bands and absorption lines (e.g., Tokunaga and Kobayashi 1999; Delfosse et al. 1999; Burgasser et al. 2002a; Gorlova et al. 2003; McLean et al. 2003; Allers et al. 2007) but these indices are all tied to red optical subtypes.

The first truly near-infrared L-dwarf classification scheme was proposed by Geballe et al. (2002) and has subtypes ranging from L0 to L9. This system does not consist of a grid of spectral standards but rather is defined by the range of values of certain spectral indices and thus does not confirm to the MK Process. For example, an unknown source would be classified as L0 if the value of its H2O 1.5 μm index fell between the values of 1.20 and 1.27. A final subtype is assigned by averaging the subtypes derived from the H2O 1. 5 μm index and other spectral indices. This system is by far the most common near-infrared L-dwarf classification system in the literature.

Fig. 3
figure 3

Left: Red optical spectral sequence (spectral resolution of Δ λ  = 9 Å) of the L dwarf spectral standards given in Table 2. The spectra are from Kirkpatrick et al. (1999b), normalized to unity at 8,250 Å, and offset with constants (dotted lines). Right: Near infrared spectral sequence (R ≈ 150) of the L dwarf near-infrared standards given in Table 2. The spectra are from Kirkpatrick et al. (2010), normalized to unity at 1.27 μm, and are offset by constants (dotted lines). Wavelengths of high telluric absorption, and thus low signal-to-noise data, are denoted as grey bars. The identification of prominent atomic and molecular absorption features can be found in Fig. 2

Full size image

Interestingly, an MK-based near-infrared scheme was devised only a few years ago and as a result, it is currently used infrequently. This system consists of ten near-infrared spectral standards ranging from L0 to L9 (Kirkpatrick et al. 2010). Table 2 lists the L-dwarf near-infrared spectral standards and the right panel of Fig. 3 shows the spectral sequence. Classification of an unknown spectrum is accomplished by comparing the entire 0.9–1.4 μm spectrum to the spectral standards and identifying the best match. In this way, this system is fully consistent with the precepts of the MK Process.

5 The T Dwarfs

As droves of L dwarfs were being discovered with 2MASS, DENIS, and eventually SDSS, brown dwarfs with near-infrared spectra qualitatively similar to that of Gliese 229B were also being discovered with 2MASS and SDSS. These objects eventually came to populate the T spectral class which is characterized by CH4 absorption at near-infrared wavelengths. Since the bulk of the T dwarfs identified in the field were discovered using data from 2MASS or SDSS, it is perhaps not surprising that two different spectral classification systems again emerged (Burgasser et al. 2002a; Geballe et al. 2002).

Burgasser et al. (2002a) created a near-infrared MK-based system that was composed of seven spectral standards ranging from T1 to T8. However, like the Kirkpatrick et al. (1999b) L-dwarf system, the Burgasser et al. system broke from the MK Process because the subtype of an unknown T dwarf is determined by averaging the subtypes derived using H2O/CH4 spectral indices and various flux ratios. Geballe et al. (2002) also developed an independent T-dwarf classification scheme that was an extension to their L-dwarf scheme described in Sect. 4 and as such, was not MK-based because it assigned subtypes based on the values of spectral ratios (see Sect. 4 for more details).

Subtypes derived using the two different systems agreed reasonably well, but the existence of two different T-dwarf classification schemes in the literature can lead to confusion. To remedy this situation, the 2MASS and SDSS teams published a unified near-infrared classification scheme for T dwarfs (Burgasser et al. 2006b). This system is MK-based and therefore consists of nine spectral standards spanning from T0 to T8. Table 2 lists the near-infrared T-dwarf spectral standards and the right panel of Fig. 4 shows the T-dwarf spectral sequence.Footnote 6 The T spectral sequence is marked by ever increasing H2O and CH4 absorption until the spectrum of the T8/T9 dwarfs consists of narrow emission-like peaks centered at 1.05, 1.27, 1.6, and 2.2 μm. The primary means of assigning a subtype to the spectrum of an unknown T dwarf is via direct comparison to the spectral standards and thus their system is in full accord with the MK Process. However, Burgasser et al. (2006b) also defined several spectral indices that could be used as secondary classifiers. This system is now the only scheme used in the literature for T dwarfs with spectral types between T0 and T8.

Fig. 4
figure 4

Left: Red optical spectral sequence (Δ λ  = 7 Å) of the optical T dwarf spectral standards given in Table 2. The spectra are from Burgasser et al. (2003b) and Kirkpatrick et al. (2010, 2011), normalized to unity at 9,000 Å, and offset by constants (dotted lines). Right: Near-infrared spectral sequence (R ≈ 150) of the T dwarf spectral standards given in Table 2. The spectra are from Burgasser et al. (2006b) and Cushing et al. (2011), normalized to unity at 1.27 μm, and offset with constants (dotted lines). Wavelengths of high telluric absorption, and thus low signal-to-noise data, are denoted as grey bars. Prominent atomic and molecular absorption features can be identified using Fig. 2

Full size image

Although the near-infrared is the primary wavelength range over which T dwarfs are now classified, a red optical scheme was also created by Burgasser et al. (2003b). This system is MK-based, and consists of only four spectral standards at T o 2, T o 5, T o 6, and T o 8 (where the subscript o is used to denote an optical spectral type). Of course the subtypes of the spectral standards could have easily been defined as T0, T1, T2, and T3, but were instead chosen to correspond to the near-infrared types for ease of use. This system serves to underscore the fact that it is the spectral standards that are the pillars of a classification system and that the labels are arbitrary. Kirkpatrick et al. (2012) recently proposed adding SDSSp J083717.22–000018.3 as the T o 0 standard and WISE J174124.25+255319.6 as the T o 9 standard. Table 2 lists the red optical T-dwarf spectral standards and the left panel of Fig. 4 shows the spectral sequence. This system is not often used given that large telescopes are required to obtain red optical data of T dwarfs.

6 The T/Y Boundary and the Y Dwarfs

In just a handful of years the 2MASS, DENIS, and SDSS surveys uncovered hundreds of low-mass stars and brown dwarfs that eventually came to populate the L and T spectral classes. The coolest T dwarfs discovered by these surveys have estimated effective temperatures of order 750 K (e.g., Burgasser et al. 2006a; Saumon et al. 2007) which left a gap of roughly 600 K between these T8 dwarfs and Jupiter at 124 K (Hanel et al. 1981) which is a commonly used benchmark for the low-mass end of the brown dwarf regime. Two of the foremost questions were (1) what would brown dwarfs with T eff  < 700 K look like spectroscopically and (2) would a new spectral class be required to classify them?

Leggett et al. (2007) and Kirkpatrick (2008) discuss the various spectral features that might trigger the need for a new spectral class from both an observational and theoretical perspective. Atmospheric models (e.g., Burrows et al. 2003) as well as the spectra of both Jupiter and Saturn indicate that ammonia (NH3) absorption bands emerge across the near-infrared as T eff falls below 600 K.Footnote 7 However, using the K-band NH3 feature from 1.95–2.05 μm is impractical observationally speaking as too little flux emerges from the atmosphere at these wavelengths due to the strong H2O, CH4, and CIA H2 absorption. Thus, attention focused on the NH3 features centered at 1.03, 1.21, 1.31, and 1.52 μm. Other predictions include the weakening and eventual loss of the red optical Na I and K I resonance lines as sodium and potassium condense out of the gas to form Na2S and KCl condensates and an eventual turn towards the red in the J − K colors.

The search for these cool field brown dwarfs continued unabated with the commencement of the UKIRT Infrared Deep Sky Survey (UKIDSS; Lawrence et al. 2007) and the Canada-France Brown Dwarf Survey (CFBDS; Delorme et al. 2008b). UKIDSS is a multi-component infrared survey of the northern sky in the Z (0.88 μm), Y (1.03 μm), J, H, and K bands. The most important component from a brown dwarf perspective is the Large Area Survey which covers 4,028 deg2 to a K-band depth significantly deeper than 2MASS (18.2 mag versus 14.3 mag).Footnote 8 The Canada-France Brown Dwarf Survey will eventually survey 780 deg2 of the sky in the red optical at i ′ and z ′. Delorme et al. (2010) also recently began the Canada-France Brown Dwarfs Survey-InfraRed (CFBDSIR) that will image 335 deg2 of the CFBDS survey footprint in the J band.

Ten brown dwarfs with spectral types later than T8 have been identified in UKIDSS and CFBDS (Warren et al. 2007; Burningham et al. 2008, 2009, 2010, 2011; Delorme et al. 2008a; Lucas et al. 2010; Liu et al. 2011). Burningham et al. (2008) designated ULAS J133553.45+113005.2 as the tentative T9 spectral standard based on the correlation of the W J index, which measures the width of the J-band peak at 1.27 μm (Warren et al. 2007), with subtype. Lucas et al. (2010) proposed that UGPS J072227.51–054031.2 (hereafter UGPS 0722–05) be designated the T10 spectral standard. The first hint of NH3 absorption at near-infrared wavelengths came from Delorme et al. (2008a) who found evidence for the 1.52 μm NH3 band based on their NH3-H spectral index that measures absorption strength on the blue wing of the H-band peak. However, no evidence of this absorption was found by Burningham et al. (2010) in the spectra of late-type T dwarfs with similar spectral types.

These > T8 dwarfs have effective temperature estimates of 500–600 K (e.g., Leggett et al. 2009), which is not significantly cooler than some late-type T dwarfs identified with 2MASS, e.g. 2MASS 09393548–2448279 (Burgasser et al. 2008). However, the discovery of two brown dwarfs, WD 0806–661B and CFBDSIR J145829+101343B, with estimated effective temperatures of only 300–400 K left little doubt that cooler brown dwarfs existed. WD 0806–661B was discovered by Luhman et al. (2011) with the Spitzer Space Telescope as a wide (2,500 AU) proper motion companion to the white dwarf WD 0806–661 (d  = 19.2 pc). It remains undetected in the near-infrared to a limit of 23.9 mag in the J band (Luhman et al. 2012), but its brightness in the mid-infrared at 3.6 and 4.5 μm suggests an effective temperature of 300–345 K. CFBDSIR J145829+101343B was discovered as a tight (2.6 AU) proper motion companion to a ∼ T9.5 brown dwarf CFBDSIR J145829+101343A (Liu et al. 2011). An effective temperature estimate of 370 ± 40 K is slightly higher than that of WD 0806–661B but still qualifies it as one of the coolest brown dwarfs known. Unfortunately, attempts at obtaining spectra of both brown dwarfs have been hampered by their extreme faintness, and in the case of CFBDSIR J145829+101343B, proximity to its primary.

The discovery of WD 0806–661B and CFBDSIR J145829+101343B confirmed that cooler field brown dwarfs existed but it took the launch of the Wide-field Infrared Survey Explorer (WISE, Wright et al. 2010) before a comparatively cool solivagant field object bright enough for follow-up spectroscopy was identified. WISE is an Earth-orbiting NASA mission that surveyed the entire sky at wavelengths of 3.4, 4.6, 12, and 22 μm (hereafter W1, W2, W3, and W4). The W1 and W2 bands were designed specifically to sample the deep CH4 absorption band centered at 3.3 μm and the region relatively free of opacity centered at 4.6 μm in the spectra of cold brown dwarfs. Since the peak of the Planck function at these low temperatures is in the mid-infrared, a large amount of flux emerges at 4.6 μm, making the W1 − W2 color extremely red.

Fig. 5
figure 5

J- and H-band spectra of UGPS 0722–05 and WISE 1828+2650 (Cushing et al. 2011). Spectra are normalized to unity at 1.59 μm. The ratio of the peak fluxes at 1.27 and 1.58 μm in the spectrum of WISE 1828+2650 is near unity which is distinct from the population of known T dwarfs

Full size image

Just as GD 165B was the prototype L dwarf and Gliese 229B was the prototype T dwarf, WISE J182831.08+265037.7 (hereafter WISE 1828+2650) became the prototype Y dwarf. Its discovery image is shown in Fig. 1. Figure 5 shows the J- and H-band spectrum of WISE 1828+2650 obtained with the Wide Field Camera 3 (WFC3; Kimble et al. 2008) on-board the Hubble Space Telescope (HST) and UGPS 0722–05. The peak flux in the J band is roughly the same height as the peak flux in the H band in units of f λ making the spectrum of WISE 1828+2650 unique.Footnote 9 For this reason, Cushing et al. (2011) identified it as the archetype Y dwarf.

With a prototype Y dwarf in hand, Cushing et al. (2011) attempted to identify the T/Y boundary. The T spectral sequence had already been extended to T10 based on the extrapolation of certain spectral ratios beyond T8 (Burningham et al. 2008; Lucas et al. 2010). In addition to WISE 1828+2650, Cushing et al. (2011) also identified six additional brown dwarfs whose spectra appeared later than UGPS 0722–05 based on the width of the J-band peak. Using the MK Process (in particular precept #2, cf. Sect. 3), UGPS 0722–05 was selected as the T9 spectral standard based on its near-infrared spectrum. WISE J173835.53+273259.0 (hereafter WISE 1738+2732) was then chosen as the tentative Y0 spectral standard because it exhibited absorption (both visually and by the NH3-H index of Delorme et al. 2008a) on the blue wing of the H-band peak that was tentatively ascribed to NH3. Bochanski et al. (2011) and Saumon et al. (2012) have identified weak NH3 features in higher resolution (R  ≈ 6,000) spectra of UGPS 0722–05 over this wavelength range suggesting the correct carrier of the absorption has indeed been identified. Kirkpatrick et al. (2012) extended the Y sequence by identifying WISE J035000.32–565830.2 as the tentative Y1 spectral standard. They also noted that the peak flux in the H band relative to the peak flux in the J band is higher than in WISE 1738+2732 presaging the equal flux heights in WISE 1828+2650 which is currently classified as \(\geq \) Y2.

Fig. 6
figure 6

J- and H-band spectral sequence (R ≈ 130) of the tentative Y0 and Y1 spectral standards (Cushing et al. 2011; Kirkpatrick et al. 2012), and the \(\geq \) Y2 dwarf WISE J1828+2650 (Cushing et al. 2011). Spectra are normalized to unity and offset with constants (dotted lines). Note that the ratio of the peak fluxes at 1.27 and 1.58 μm approaches unity with later subtype

Full size image

Table 2 lists the two tentative Y-dwarf spectral standards and Fig. 6 shows the current Y spectral sequence. Y dwarfs are currently classified over the 1.1–1.7 μm wavelength range due to the difficulty of obtaining spectra at other wavelengths (see also Sect. 8). The sequence is marked both by a narrowing of the J-band peak, but also a slow drift in the J∕H-peak flux ratio towards unity. There are currently only 14 published Y dwarfs (Cushing et al. 2011; Kirkpatrick et al. 2012; Tinney et al. 2012) but continued mining of the WISE survey will no doubt uncover more.

7 The Underlying Physics

Effective temperature is the primary parameter that controls almost the entire spectral sequence. In the hotter OBAFGK dwarf stars, it is the variations in atomic line strengths with decreasing temperature (via the Saha Boltzmann equation) that drive variations in the spectral morphology and thus spectral type. However, in the cooler MLTY dwarfs, it is variations in both molecular and condensate (i.e. grain) chemistry with decreasing effective temperature that drive the spectral morphological changes. Indeed the M/L and L/T (and possibly the T/Y) transitions are controlled in part by the formation and subsequent evolution of condensates (see chapter by I. Baraffe).

At atmospheric temperatures of less than 2,400 K, condensates form from the refractory elements (Ti, V, Ca, Al, Fe, Si) which in turn gravitationally settle in the atmosphere to form clouds. With decreasing effective temperature, more and more species condense out until the atmosphere consists of layer upon layer of clouds. The formation of these clouds has two major consequences for the atmosphere. First, the chemistry in the atmospheric layers above the cloud decks is forever altered because the atoms and molecules that make up the condensates are no longer available to participate in chemical reactions. Second, the condensates add opacity to the atmosphere which can, in some cases, dramatically alter the emergent spectrum of the object. Indeed as first suggested by Jones and Tsuji (1997), the weakening of the TiO and VO bands at the M/L transition is a result of the formation of condensates like perovskite (CaTiO3) and other Ti-bearing species (e.g., Ca4Ti3O10, Ca3Ti2O7, Ti2O3) as well as solid solution VO and VO2 (Lodders 1999; Burrows and Sharp 1999; Allard et al. 2001).Footnote 10 Silicate and liquid iron condensates form at slightly cooler temperatures and make the near-infrared colors of cloudy L dwarfs much redder than equivalent-mass objects with clear atmospheres.

Fig. 7
figure 7

Effective temperature as a function of spectral type. L-dwarf spectral types are based on red optical spectra and T- and Y-dwarf spectral types are based on near-infrared spectra. A gap is left between L8 and T0 because there is no L9 subtype in the optical sequence. Temperatures for the L dwarfs and T dwarfs with spectral types earlier than T9 are from Vrba et al. (2004). However, only dwarfs with trigonometric parallax errors less than 30 % of the parallax value are plotted. Temperatures for the T dwarfs with spectral types later than T8 are from Leggett et al. (2009, 2012) while the temperatures for the Y dwarfs are from Leggett et al. (2013). WISE J1828+2650 is currently typed as a \(\geq \) Y2 and so is plotted as a solitary object beyond a spectral type of Y2

Full size image

Figure 7 shows the effective temperature as a function of spectral type for a sample of L, T, and Y dwarfs. The L-dwarf spectral types are based on the red optical system of Kirkpatrick et al. (1999b) and the T- and Y-dwarf spectral types are based on the near-infrared systems of Burgasser et al. (2006b), Cushing et al. (2011), and Kirkpatrick et al. (2012). The effective temperatures of stars are often estimated by fitting model atmospheres to observed spectra and the same is certainly true for brown dwarfs (e.g. Stephens et al. 2009; Testi 2009; Witte et al. 2011). However, a reasonably model-independent estimate of brown dwarf effective temperatures can be obtained by exploiting the fact that the radii of brown dwarfs are nearly independent of both mass and age. Due to their partially electron-degenerate cores (e.g., Kumar 1963), the mass-radius relation is effectively flat at ∼ 1 R Jup across two orders of magnitude in mass from 0.1 to 0.001 M ⊙ (see figure 1 of Burrows and Liebert 1993). The effective temperatures of the L and T dwarfs shown in Fig. 7 are from the Vrba et al. (2004) and were estimated by measuring their bolometric luminosities and then using the Stefan Boltzmann law (\(L_{\mathrm{bol}} = 4\pi {R}^{2}\sigma T_{\mathrm{eff}}^{4}\)) to compute their effective temperatures assuming a fixed radius of 0.90 ± 0.15 R Jup. With decreasing effective temperature, it becomes more and more difficult to measure a model independent bolometric luminosity because wavelengths that are easily accessible from the ground contain a smaller and smaller fraction of the total emergent flux. For example, at T eff  = 300 K, only 10 % of the emergent luminosity emerges at λ  < 2.5 μm (M. Marley, private communication). Therefore, the effective temperature estimates of the brown dwarfs with spectral types later than T8 in Fig. 7 are based on atmospheric model fits to their near-infrared spectra (Leggett et al. 2013).

Effective temperature correlates reasonably well with spectral type through the entire LTY sequence, except at the L/T transition where the effective temperature is roughly constant from a spectral type of L8 to T5 (see also Burgasser et al. 2002a; Golimowski et al. 2004; Testi 2009). This indicates that changes in a parameter (or parameters) other than effective temperature is driving the evolution in the spectral morphology over these spectral types. It is generally accepted that this change is due to rapid loss of the silicate and iron condensate opacity due to some unknown mechanism. Possible candidates include the break up of the condensates clouds (Ackerman and Marley 2001; Burgasser et al. 2002b) or a sudden downpour of the condensates (Knapp et al. 2004).

The changes in spectral morphology at the L/T transition also provide a cautionary tale against violating precept #3 of the MK process, i.e. using models to guide the selection of subtypes. Any classification system that attempted to force the subtypes at the L/T transition to have a uniform gradient in effective temperature would create both a dramatic and jarring evolution of the spectral morphology over just a few subtypes.

At spectral types later than roughly T5, effective temperature decreases down to the T/Y boundary. WISE 1828+2650 is plotted as a solitary object beyond a spectral type of Y2 because its exact subtype is currently unknown. Although there is clearly a paucity of objects between the Y0s and WISE 1828+2650 which makes drawing any firm conclusions dangerous, it is nevertheless tempting to suggest that a second temperature plateau may exist at the T/Y boundary. Morley et al. (2012) have shown that the inclusion of additional condensates (Na2S, KCl, ZnS, MnS, and Cr) significantly improves the agreement between the model spectra and the observations of late-type T dwarfs and Y dwarfs. This suggests that condensates, in particular Na2S, play an important role in shaping the emergent spectra of the late-type T and Y dwarfs. Perhaps the loss of these condensates in a fashion similar to the L/T transition is responsible for the tentative plateau seen in T eff at the T/Y transition.

8 Outstanding Issues

In just under 20 years, a smooth spectral sequence extending from the edge of the hydrogen-burning main sequence down to brown dwarfs with effective temperatures of roughly 300 K has emerged. However, much work remains before the LTY classification systems become as refined as the MK System for the OBAFGKM stars. Below I discuss just two outstanding issues related to the spectral classification of the L, T, and Y dwarfs.

Issue #1:

As noted in Sect. 7, effective temperature is the primary parameter that controls the spectral morphology of the L, T, and Y dwarfs. Completely absent from that discussion is the impact that variations in surface gravity g and metallicity [Fe/H] can have on the emergent spectra and thus spectral type. For the hotter OBAFGKM stars, variations in surface gravity manifest themselves in a variety of ways (e.g. a change in the width of atomic lines) and are encapsulated in the luminosity class (see Sect. 3). Brown dwarfs with spectral “peculiarities” indicative of unusual surface gravities and/or metallicities have been known for some time, e.g. 2MASS J01415823–4633574 a low surface-gravity L dwarf (Kirkpatrick et al. 2006) and 2MASS J05325346+8246465 a metal-poor L dwarf (Burgasser et al. 2003a) but all of the current classification systems are unable to account for these variations because the systems are still one dimensional (e.g., Kirkpatrick 2005).

Cruz et al. (2009) presented the first L-dwarf spectral classification system that uses both temperature and surface gravity sensitive features as criteria for assigning subtypes. The system expands on the Kirkpatrick et al. (1999b) scheme to include three gravity classes labeled α, β, and γ. The γ class has a lower surface gravity than the β class which in turn has a lower surface gravity than the α class. However, this system is currently not MK-based because rather than assign spectral standards for each T eff/g bin, it uses surface gravity sensitive spectral indices to assign a gravity class. Nevertheless, this scheme is an important proof-of-concept that shows that it is possible to expand the current LTY classification systems to include the effects of gravity and hopefully in the future, metallicity.

Issue #2:

The tentative Y-dwarf classification scheme proposed by Cushing et al. (2011) and Kirkpatrick et al. (2012) is based in the near-infrared, and in particular focuses on the width of the J-band peak at 1.27 μm and the blue wing of the H-band peak at 1.6 μm. There are, of course, other wavelengths that can be used for classification. As noted by Leggett et al. (2007), there is an additional NH3 absorption feature centered at 1.02 μm that would manifest itself as a “divot” in the Y -band peak. This feature would be much easier to identify than the H-band absorption feature currently used but to date few high S/N Y -band spectra of the Y dwarfs have been obtained.

The near-infrared is currently the easiest wavelength range over which to obtain a spectrum of a Y dwarf, but as noted above, it contains only a small fraction of the total bolometric luminosity emitted. Given that Y dwarfs emit most of their radiation in the mid infrared (3–20 μm), it seems reasonable to devise a scheme at these wavelengths. Indeed historically as cooler and cooler objects were discovered, the wavelength range used for spectral classification has moved redward from the visible for the hotter stars, to the red optical for the M and L dwarfs, and then to the near-infrared for the T dwarfs. In addition, Cushing et al. (2006) have already shown that T-dwarf spectra over the 5.5–14.5 μm wavelength range exhibit smooth variations in the strengths of H2O, CH4, and NH3 absorption bands suggesting that a Y dwarf scheme based at these wavelengths is entirely plausible. However, as of the writing of this chapter, there is no ground- or spaced-based facility capable of obtaining mid-infrared spectroscopy of the late-type T and Y dwarfs. The NIRSpec and MIRI instruments on board the James Web Space Telescope will be capable of obtaining low- to moderate-resolution spectra over the entire 0.6–29 μm wavelength range and will therefore revolutionize the study of the Y dwarfs.