A&A 484, 469-478 (2008)
DOI: 10.1051/0004-6361:20078843
P. Delorme1 - C. J. Willott2 - T. Forveille1 - X. Delfosse1 - C. Reylé3 - E. Bertin4 - L. Albert5 - E. Artigau6 - A. C. Robin3 - F. Allard7 - R. Doyon8 - G. J. Hill9
1 - Laboratoire d'Astrophysique de Grenoble,Université
J. Fourier, CNRS, UMR5571, Grenoble, France
2 - University of Ottawa, Physics Department, 150 Louis Pasteur,
MacDonald Hall, Ottawa, ON K1N 6N5, Canada
3 - Observatoire de Besançon, Institut Utinam, UMR CNRS 6213,
BP 1615, 25010 Besançon Cedex, France
4 - Institut d'Astrophysique de Paris-CNRS, 98bis Boulevard Arago,
75014, Paris, France
5 - Canada-France-Hawaii Telescope Corporation, 65-1238 Mamalahoa
Highway, Kamuela, HI96743, USA
6 - Gemini Observatory Southern Operations Center c/o AURA, Casilla 603
La Serena, Chile
7 - C.R.A.L. (UMR 5574 CNRS), École Normale Supérieure, 69364 Lyon
Cedex 07, France
8 - Département de physique and Observatoire du Mont Mégantic,
Université de Montréal, CP 6128, Succursale Centre-Ville,
Montréal, QC H3C 3J7, Canada
9 - McDonald Observatory, University of Texas at Austin, 1
University Station C1402, Austin, TX 78712-0259, USA
Received 12 October 2007 / Accepted 2 April 2008
Abstract
Aims. We present the first results of a wide field survey for cool brown dwarfs with the MegaCam camera on the CFHT telescope, the Canada-France Brown Dwarf Survey, hereafter CFBDS. Our objectives are to find ultracool brown dwarfs and to constrain the field-brown dwarf mass function thanks to a larger sample of L and T dwarfs.
Methods. We identify candidates in CFHT/MegaCam i' and z' images using optimised psf-fitting within Source Extractor, and follow them up with pointed near-infrared imaging on several telescopes.
Results. We have so far analysed over 350 square degrees and found 770 brown dwarf candidates brighter than
.
We currently have J-band photometry for 220 of these candidates, which confirms 37% as potential L or T dwarfs. Some are among the reddest and farthest brown dwarfs currently known, including an independent identification of the recently published ULAS J003402.77-005206.7 and the discovery of a second brown dwarf later than T8, CFBDS J005910.83-011401.3. Infrared spectra of three T dwarf candidates confirm their nature, and validate the selection process.
Conclusions. The completed survey will discover 100 T dwarfs and
500 L dwarfs or M dwarfs later than M8, approximately doubling the number of currently known brown dwarfs. The resulting sample will have a very well-defined selection function, and will therefore produce a very clean luminosity function.
Key words: stars: low-mass, brown dwarfs - techniques: spectroscopic - techniques: photometric - methods: data analysis - surveys
Much interesting work however remains to be done, and
the advent of wide field cameras on large telescopes
makes an unprecedented volume of the Milky Way accessible
for brown dwarf searches. Here we use two large surveys with
MegaCam
on the CFHT telescope, the Canada-France-Hawaii Telescope Legacy
Survey (CFHTLS
)
and the Red-sequence Cluster
Survey 2 (RCS-2; Yee et al. 2007), and complement them
by additional observations to address three areas of brown dwarf physics:
Field brown dwarfs are extremely cool objects, with a temperature range which
currently extends from 2500 K (early L) to
625 K (late T)
(Delorme et al. 2008; Warren et al. 2007; Golimowski et al. 2004).
Even cooler, yet to found, brown dwarfs should close the temperature
gap between
late type T dwarfs and solar system giant planets (
100 K).
Brown dwarfs spectra very much differ from a black body, and have considerable
structure from deep absorption lines and bands. Their spectral energy
distribution (in
units) peaks in the near
infrared (hereafter NIR), particularly
in the J photometric band, and they are most easily detected
in that wavelength range. Their pure NIR
colours however
do not very effectively distinguish them from other classes at
modest S/N ratio.
Brown dwarfs are more easily recognised by including
at least one photometric band below 1
m, since their steep
spectral slope at those wavelength produce very distinctively
red i'-z' and z'-J colours. As one recent example,
the T8.5 ULAS 0034 has
(
,
Delorme et al. 2008), and at any S/N ratio where it is safely
detected it cannot be confused with anything, except a slightly earlier
T dwarf or a z=6 quasar. The UKIDSS discovery observation however was
less than 3
away from the K dwarf locus. Since K dwarfs outnumber
T dwarfs by orders of
magnitude in any flux limited sample, that distance would have
been woefully insufficient for a secure identification. The
non-detection of ULAS 0034 at i' and z' in the deep SDSS stripe
82 played
a major role in its identification by Warren et al. 2007), and
other near-IR searches for brown dwarfs similarly
use some <1
m imaging to weed out their
contamination.
We take advantage of the wide field
of view of the MegaCam camera (Boulade et al. 2003) on the CFHT telescope,
and of the trove of observational material obtained
with that instrument, to select brown dwarfs on their i'-z' colour.
The i'-z' colour has excellent dynamics for brown dwarfs, varying
from 1.7 to 4.0 between mid-L and late-T (Fig. 3). It therefore
provides (at least at high S/N ratio) a good spectral type estimator.
At the high galactic latitude of our survey, the i'-z' colour
distinguishes brown dwarfs from almost every astronomical
source type, but it leaves one contaminant, quasars at
.
Those are
of considerable interest in their own right, but need to be distinguished
from the brown dwarfs. As first shown by Fan et al. (2001), the
i'-z' vs. z'-J colour/colour diagram very effectively separates
the two populations (Fig. 1 and Willott et al. 2005).
The very red i'-z' of high redshift quasars is caused by deep
Lyman
absorption on a relatively flat intrinsic spectrum), and they
therefore have a more neutral z'-J.
The spectral distribution of brown dwarfs, in contrast, keeps
rising steeply into the J band. We therefore complement our MegaCam
i' and z' photometry by pointed J-band imaging of the candidates
selected on i'-z'. Besides pinpointing the (few) quasars, the J-band
photometry very effectively rejects any remaining observational
artefact, as well as the (more numerous) moderately red stars
scattered into the brown dwarf/quasar box by large noise excursions.
![]() |
Figure 1:
z'-J and i'-z' colours of brown dwarfs and quasars. These
synthetic colours were computed for the MegaCam i' and z' photometric system and the NTT SOFI
![]() |
Open with DEXTER |
Each square-degree MegaCam image contains a few hundred thousand objects, of
which at most a few are brown dwarfs. We thus need to strike a careful
balance between sample completeness and contamination. To tune this
compromise we need a precise knowledge of the colours of brown dwarfs and quasars
for the exact instruments and filters used in our survey.
As discussed in Willott et al. (2005), these colours are known for
some photometric systems, in particular SDSS and 2MASS, but the
filters and quantum efficiency curves of MegaCam are notably
different (Fig. 2). This is particularly
significant for brown dwarfs and quasars. Due to their highly structured
spectra a modest change to a
response curve can produce significantly different colours
when it includes or excludes a major absorption band or emission
line. We update the synthetic colours of Willott et al. (2005), using
additional brown dwarfs spectra which have become available since 2005, and
adding the many near-IR instruments and filters which we use for
the J-band imaging. We use publically available spectra
(from S. Leggett's website,
Geballe et al. 2001; Leggett et al. 2002; Chiu et al. 2006; Knapp et al. 2004; Martín et al. 1999; Burgasser et al. 2003; Golimowski et al. 2004; Kirkpatrick et al. 2000)
of over 60 brown dwarfs with spectral types L1 to T8 (on the
Burgasser et al. 2006, spectral type scale) and the synthetic quasar
spectra of Willott et al. (2005). We compute their synthetic
MegaCam i' and z' photometry in the AB system (Fukugita et al. 1996)
using detector quantum efficiency and
transmission curves for the atmosphere, telescope, camera
optics, and filters (cf. Fig. 2), obtained from the CFHT
web page. Figure 3 displays the resulting
colours as a function of the spectral type.
We similarly synthesize J-band photometry for each of the
instruments and J filters used in the J-band follow up. These
instruments have significantly different response curves,
which must be taken into account to obtain homogeneous
selection criteria. We found, in particular, that brown dwarfs
colours which include J photometry obtained at the NTT
with SOFI and its (default) wide J filter are not as red as
we initially expected: contrary to most J filters, its
wide bandpass includes water vapor bands which
are strongly absorbed in L and (particularly) T dwarfs. As a
result, the i'-J and z'-J colours which use this filter
are bluer by 0.15 mag for early L and
0.5 mag
for late T. After we realised this we switched our SOFI observations
to the alternate
filter, which better separates
T dwarfs from quasars. We use the synthetic colours to
shift our selection boxes according to the actual filter.
![]() |
Figure 2: Compared spectral response functions of the CFHT (thick lines) and SDSS (thin dashed lines) instruments for their i' (dark blue) and z' (red) filters. These factor in the average atmospheric transmission of the two observatory sites, the telescope reflectivities, the transmissions of the camera optics and filters, and the quantum efficiencies of the CCDs. Contrary to the SDSS bandpasses, the CFHT i' and z' filters overlap significantly, leading to less contrasted colours. |
Open with DEXTER |
![]() |
Figure 3: i'-z' synthetic colour versus spectral type for the MegaCam photometric system. The colour for a spectral type is the average over the brown dwarf spectra from Chiu et al. (2006), Golimowski et al. (2004) and Knapp et al. (2004) with that spectral type, and the error bars represent the dispersion (set to 0 when only one template per spectral bin is available). |
Open with DEXTER |
![]() |
Figure 4:
Sky chart of sky area covered by CFBDS so far. Black curve
marks the galactic plane while dotted curves mark
![]() ![]() |
Open with DEXTER |
Table 1: Characteristics of the optical surveys used by CFBDS.
The CFHTLS survey has three components, named Deep, Wide, and Very Wide, (described in detail on the CFHTLS web page, http://www.cfht.hawaii.edu/Science/CFHLS/), and we use all three.
CFHTLS Deep.
This deepest component of the CFHTLS covers four high galactic
latitude and low extinction 1 square degree
fields in the
u*g'r'i'z' filters, and it is primarily motivated by the
SNLS type Ia supernovae search (Pain & SNLS Collaboration 2003).
The expected total exposures times
per field of the completed survey will be 66h at z' and 132h at i'
with 10
depths of
and
.
This sensitivity is sufficient to identify mid-L dwarfs out to
1300 parsecs. At such distances, and at high galactic latitude,
the thick disk becomes dominant and we therefore have good hopes
to probe its brown dwarf population. The present analysis uses the
T0003 release (T0004 was very recently released, but we have yet
to analyse and follow-up the corresponding detections) which has
10
depths of
and
.
Thanks to their large number of contributing exposures and their extended
time base, the T0003 images are extremely clean, with essentially no
contamination by cosmic rays or bad pixels, or by variable or moving
objects such as supernovae and asteroids.
CFHTLS Wide. This component of the CFHTLS will cover 186 deg2
divided between four high galactic latitude low extinction
fields, in
u*g'r'i'z', and it is primarily motivated by
cosmological weak
lensing. We have analysed the 20 deg2 that have both i' and
z' coverage in the T0003 CFHTLS release (i' coverage is
considerably more extensive, due to priorities set by the main
drivers of the Wide survey). The average 10
depths are
(for total exposure times of 7200 s) and
(for total exposure times of 4300 s), with small field to
field variations due to seeing and sky background fluctuations. Of
the three components the Wide probes the largest volume. The Wide
images have enough coadded subexposures (9 for z' and 7 for i')
to reject all cosmic rays and bad pixels, and the overall exposure
times are sufficiently long to eliminate all but the slowest moving
objects. The i' and z' images on the other hand are usually
not contemporaneous, and variable sources (in practice mostly
supernovae) which are serendipitously bright in the z' can
erroneously pass our i'-z' colour filter. Those need to be
eliminated at a later stage.
CFHTLS Very Wide (VW). This shallowest component of the CFHTLS is
motivated by transneptunian objects and was initially set to cover
1000 deg2 in the ecliptic plane with g'r'i' images. It was
later downsized to 250 deg2 when it was realised that the three
components could not all be completed within the allocated time. We
use the
150 deg2 from the Very Wide with
absolute value of the galactic latitude above 30 degrees to ensure
low absorption (see Table 1). The
average 10
depth
of the 540 s i' VW exposures is
,
and we complement
them by 420 s z' exposures, with typical 10
depths of
22.8. Both sets of images are coadditions of 3 subexposures
separated by at least one night. We therefore have enough
information to reject the vast majority of bad pixels, cosmic ray
hits, and moving solar system objects. The time span of the 3 z' exposures on the other hand is usually too short to reliably
recognize supernovae, which vary on time scales of a few weeks. We
therefore again need to reject these contaminants at a later stage.
Red-sequence Cluster Survey 2 (RCS-2).
The RCS-2, designed to identify
distant galaxy clusters through their galaxies on the red
sequence, (Yee et al. 2007)
is an ongoing g'r'z' survey of 800 deg2 at high galactic
latitude to lower the absorption (see Table 1. We have to
date used 600 deg2 kindly made
available to us by the RCS-2 consortium, and we complement their 360 s
z' band images by 500 s or 680 s i' exposures, depending on the
seeing. The resulting 10
depths of (
and
are
similar to those of the CFHTLS-VW. Both the RCS-2 images and our
complementary i' data are single exposures, which maximize
the depth achieved for a given observing time.
We use the RCS-2 g' and r' images, which
are contemporaneous with the z' ones, to identify and reject
both supernovae and moving solar system objects.
All images are pre-processed by the CFHT staff using the ELIXIR package (Magnier & Cuillandre 2004). The CFHTLS Deep and Wide images are aligned and coadded by the Terapix data center (Bertin et al. 2002). For the CFHTLS Very Wide and RCS-2 datasets, we carry out our own processing to check and refine the astrometry and (for fields which overlap the SDSS) photometry. For the CFHTLS Very Wide, each pointing has 3 subexposures per filter which are combined whilst rejecting bad pixels and cosmic ray impacts. The CFHTLS Very Wide and RCS-2 images in different filters are aligned (with distortion correction) and trimmed to their common area.
To date most of our volume coverage originates in the CFHTLS Very Wide and RCS-2, due in part to the late start of the z' part of CFHTLS-Wide. We have currently analysed 350 deg2 of the 800 deg2 expected for these two surveys. Their relative shallowness has the advantage of producing targets for which spectroscopy can be obtained relatively easily on 8m-class telescopes.
Table 1 summarizes the properties of the four
surveys, listing the limiting magnitude, the maximum distances
at which mid-L and late-T dwarfs can be detected to these magnitudes,
and the current and final areas covered by the survey. Our
full survey probes
several times the SDSS volume for T dwarfs and we expect to
detect 100 new T dwarfs (compared with the
150 currently
known)
As explained above, we use J-band photometry to distinguish between
brown dwarfs and z>5.8 quasars. For brown dwarfs (and very low mass
stars) the z'-J colour also provide a good spectral type diagnostic, for which
we obtain better S/N ratio than i'-z'. That is in particular very
helpful in eliminating mid-M dwarfs scattered into our i'-z'
selection box by several
noise excursions. Given the relative
numbers of mid-M stars and brown dwarfs in a magnitude-limited sample,
these noise excursions are sufficiently frequent to very
significantly contaminate our i'-z' selection, but the better S/N ratio
of the z'-J colour (and the low likelihood of large noise excursions
at both i'-z' and z'-J) makes them obvious once we obtain J-band images.
The J-band follow up has been carried out at several observatories:
La Silla (NTT, 3.6 m), McDonald (2.7 m), Kitt Peak (2.1 m), La Palma
(NOT, 2.5 m), see Table 2. We adjust our integration times
to achieve either
a clear detection or a limiting magnitude that excludes any dwarf
and demonstrates that the candidate is a high redshift quasar (usually around
). For the few candidates which are not detected at i' and where
we cannot exclude that the object was a supernovae in the z' image,
we integrate deeper to detect the quasar at J. Any supernova has long
faded, and cannot be detected at any reasonable depth.
At the NTT, which accounts for most of our J-band follow up, we
usually need exposures times of 5 to 10 min, obtained as
40 s individual exposures which we jitter to
measure and substract the sky background. The typical integration time on
2 meter-class telescope was 30 min.
Table 2: Technical characteristics of the telescopes used for the J-band follow up.
Fortunately, high-redshift quasars (at the resolution of the MegaCam images) and brown dwarfs are point-like. We therefore only need to distinguish point-like sources from both artefacts and extended objects, and don't have to tackle the much more difficult task of separating general astronomical sources from artefacts. Point Spread Function (hereafter PSF) fitting provides an excellent stellarity diagnostic, as well as optimal photometry and astrometry for point sources. It therefore forms the basis of our selection procedure.
We use the well known SExtractor (Bertin & Arnouts 1996) photometry package, to which two of us (Bertin & Delorme, in prep.) recently added a PSF-fitting module. In keeping with the general SExtractor philosophy, this module implements a dual-image mode, where source positions in a ``detection image'' precisely determine where photometry will be extracted in a ``photometry image''. This dual image mode is particularly well matched to the extreme colours of our targets: given the relative depths of the i' and z' images, any object of interest is very robustly detected at z' but faintly, if at all, at i'. We therefore use the z' image as the detection image for both i' and z', naturally producing matched catalogues of i' and z' photometry for every object that is well detected at z', independently of its i' significance. This eliminates the delicate task of handling unmatched sources in independent catalogues: those might be weakly detected in the i' image, though with too low a significance for inclusion in any modestly reliable single image i' catalogue, and they therefore cannot validly be handled as pure upper limits.
SExtractor implements simultaneous fitting of multiple PSFs to blended
objects, providing accurate parameters for close binaries and usable
measurements for point-like sources blended with galaxies.
The latter is particulary important for the quasar search, since
it can recover some lensed quasars which would otherwise be lost
to confusion with their lensing galaxy. In addition to more accurate
parameters for the affected objects, this better handling of blends
produces more complete catalogues in crowded fields. Introducing multiple
PSFs recovers
additional sources in the relatively
shallow CFHTLS-VW and
RCS-2 images, and
in the deeper CFHTLS-Deep images.
PSF-fitting also improves the photometric precision by
10% over optimum
aperture photometry (Bertin & Delorme, in prep.),
and it therefore allows us to use slightly deeper catalogues,
for another 15% gain in sample size.
Since low significance i' detections provide colours with
complex error distributions, we replace them by the
5
detection limit on their image and compute
a lower limit on i'-z'. We note that the resampling involved
in the coaddition of images built from multiple exposures,
and in the filter to filter alignment, generates noise
correlations on scales of 1-2 pixels. Thanks to the
generous sampling of our MegaCam images (0.186
/pixel and seeing
mostly above 0.6
)
and the use of a Lanczos3 interpolation
function, source profiles are negligibly affected
(Bertin et al. 2002), but resampling has a measurable low-pass filtering
effect on photon noise. As we decided for practical reasons to ignore
noise covariances in our fitting, the net effect on photometry is that
errors estimates must be multiplied by a factor
1.4. That factor is well determined for the CFHTLS-Deep and
CFHTLS-Wide images, which are built from a large number of individual
exposures, but for the CFHTLS-VW images it significantly varies from
field to field according to the sub-pixel relative positions of the
3 coadded exposures.
We start by requiring a >
detection in the z' filter and a
i'-z'>1.7. These criteria without any additional filtering
typically yield over 10 000 candidates per RCS-2 1 square degree
field. The single exposures per filter used for the RCS-2 survey
(and for our follow-up of its fields) are most affected by cosmic
ray hits and bad pixels, and the stacked images used in the
other components contain fewer such artefacts. Flagging of
known bad-pixel positions and simple morphological rejection
of cosmic ray hits lowers this number to under 1000,
but not to a point where visual examination would be practical.
We then assess the stellarity of each candidate from its SExtractor
output parameters to further decrease the number of false
detections. After experimenting with several parameter
combination, we have
converged to the quality of the fit between the image and the
PSF model, as summarised by the
of the residuals, as our main
diagnostic. We found that Sextractor's default stellarity index,
based on a specifically trained neural network, works well at high
signal to noise ratios, but that it becomes ineffective for the
faint objects which dominate our catalogues. We similarly found that
tests based on comparisons of fluxes through different apertures,
or on peak surface brightness versus flux, are always less distinctive
than
filtering. Since we currently prefer to visually inspect
all final candidates, we very conservatively set our filtering threshold
to a level where
10% of the candidates are visually acceptable.
These filtering criteria typically yield under 50 candidates per
square degree.
A lower threshold would not very significantly decrease
the inspection workload, and might conceivably eliminate a few valid
candidates. We will probably revisit this tuning as we gain
experience with, and confidence in, our selection process.
With our current settings,
filtering reduces the number
of artefacts by a factor of 10-15 (Fig. 5).
Visual inspection of the more than 2000 sources with i'-z'>1.7in a 4 square
degrees test region showed that
filtering rejected none
of the 14 valid point-like sources, and our resolution of the few
initial discrepancies was always in favour of the
filtering
diagnostic. Further tests also verified that our current threshold is
comfortably above the highest
measured for valid
candidates, and therefore very conservative.
After this pruning of the initial catalogue to just bona fide
point sources, we select candidates with an i'-z' criterion. Our
synthetic photometry (Fig. 3) demonstrates that the
L dwarf domain begins at
i'-z'>1.45. M dwarfs however, with i'-z'just below 1.45, massively outnumber brown dwarfs in a magnitude-limited
sample. Poissonian photometric errors consequently scatter a
significant number of M dwarfs into this L dwarf box. We therefore
set our colour filtering to i'-z'>1.7, or nominally to later than L4. With M 8 and M 9 stars having
,
and assuming
Gaussian noise at our z' S/N limit, this colour threshold eliminates
of these very late M dwarfs. Mid-M dwarfs have
and we eliminate
of them. Since brown dwarfs are
intrinsically much rarer, our candidate list nonetheless has some
significant contamination by late-M dwarfs, but at a level which no
longer overwhelms our follow-up capacity.
![]() |
Figure 5:
Histogram of the ![]() ![]() |
Open with DEXTER |
Our final filtering step is to eliminate as many as we can of
the candidates which actually are supernovae that were bright
at the time of the z' image. All RCS-2 fields have contemporaneous
g', r' and z' images, and for supernovae the g' and r' images
are very significantly deeper than the z' one. We therefore
very reliably reject their supernovae by inspecting these g' and r' images. The CFHTLS-Deep images are stacks of exposures
obtained over several years, and any supernova is eliminated by
the sigma-clipping applied during stacking. The exposures which
contribute to our CFHTLS-Wide and CFHTLS-VW z' images, on the
other hand, were usually obtained over shorter time spans than
the 6 weeks (Pain & SNLS Collaboration 2003) timescale of supernovae
photometric evolution. We therefore mostly cannot recognize their
supernovae based on their photometric variation between the
individual exposures, and the exposures in other filters
are usually not sufficiently contemporaneous to reject them
based on a blue instaneous colour. We must therefore handle
some supernovae contamination at a later stage.
![]() |
Figure 6: Histograms of the number of i'-z' candidates as a function of i'-z' colour (light green), and of the number of these candidates for which J-band photometry is currently available (dark red). |
Open with DEXTER |
![]() |
Figure 7:
i'-z' vs. z'-J colour-colour diagram of our cool dwarf
candidates (black). The error bars are 1![]() |
Open with DEXTER |
This first set of i'z'J photometry allows a good assessment of the actual nature of these 215 candidates. As summarised in Table 3, they include 23 likely T dwarfs, 57 L dwarfs and very late M dwarfs candidates (M8 and M9 dwarfs have very similar z'-J colours to L dwarfs), and at least 4 high redshift quasars (published in Willott et al. 2007). 22 targets remain undetected in deep J-band images and are most likely artefacts which our filtering did not catch. 109 objects have z'-J<1.6 and most of those are likely M dwarf contaminants. Some however have J-band upper limits which are insufficiently deep to ascertain whether they are artefacts, quasars or Mid-M dwarfs. Those will need additional follow-up to clarify their status.
Since we prioritised the analysis of the reddest candidates,
this first i'z'J sample is strongly biased towards T dwarfs.
Appproximately correcting for this bias, we estimate that
40%
of our i'-z' candidates are actual cool dwarfs, of which
15% are T dwarfs.
The T dwarfs include several with extreme colours, which ongoing
spectroscopic observations will characterize further. One,
CFBDS J003402-005206,
actually is an independent discovery of ULAS J003402.77-005206.7
which Warren et al. (2007) recently identified with the UKIDSS
survey (Burningham et al. 2007) as a brown dwarf later than T8 (
). Even more recently, we obtained spectroscopic observations
of an even cooler brown dwarf (
625 K, Delorme et al. 2008).
Figure 7 shows several other candidates that
are at least as promising, and those are currently queued for
near-IR spectroscopy.
Table 3: Preliminary classification of the candidates with J-band photometry.
Table 4: Observational properties of the three T dwarfs whose spectra are presented here.
Table 5: Spectral indices and the resulting spectral types.
![]() |
Figure 8: GNIRS near-IR spectra of the three T Dwarfs observed on Gemini-South. The two fainter targets originate from the CFHTLS-Deep, while the brighter ones comes from the CFHTLS-VW. Templates of spectral types T2 and T4.5 from McLean et al. (2003) are shown for comparison. All spectra are normalised at 1.26 microns and displayed with integer offsets for clarity. |
Open with DEXTER |
We present here near-IR spectroscopic observations of our three first T dwarf
candidates, as an illustration of the content of our candidates
catalogue. Two, which we originally published in
Willott et al. (2005), originate in the CFHTLS-Deep survey and are thus
fainter than most of our candidates. The last comes from the Very Wide
component of the CFHTLS and is more representative.
Cross dispersed spectra were obtained during semester 2006A
with GNIRS (Elias et al. 2006) at Gemini South in Service mode.
The slit width of 0.68 arcsec coupled with the short camera and the
31.7 l/mm grating yielded a resolving power of 900, and the spectra
have full wavelength coverage between 0.9 and 2.4 microns. A-B (not ABBA) sequences were used, with individual
5-minutes exposures for the brighter target and
10-minutes for the fainter CFHTLS-Deep targets. The total on-source
integration time is 30 min for
CFBDS193430-214221 (
),
180 minutes for
CFBDS100113+022622 (
)
and 200 minutes for
CFBDS095914+023655 (
). The OH sky lines were used for
wavelength calibration, and an A-type star was observed immediately
before each sequence for relative flux calibration and telluric
absorption correction. The spectra were extracted
and calibrated using our own IDL procedures.
The reduction proceeds as follows. The sequence of spectral images are flat-fielded using an internal flat taken immediately after the science frames. The five useful cross-dispersed orders are then extracted in five individual images that are corrected for distortion in the spectral dimension. For most objects, the trace is too faint over many wavelengths intervals to determine trace position, so its curvature is derived from the reference star spectrum. These individual order frames are then pair-subtracted, effectively removing most of the sky, dark current and hot pixels contributions. Each frame is then collapsed along the spectral dimension to determine the positive and negative traces positions. We then extract the spectra using positive and negative extraction boxes that have identical but opposite integrals; this minimizes the contribution from residual sky line that would remain from the pair subtraction. The same operation is performed on the A0 telluric calibration star. Spectra derived from individual image pairs are then median-combined into final target and calibration star spectra. A telluric absorption spectrum is then derived using the calibration star spectra. A black body spectrum with a temperature of 10 000 K is assumed for the A0 stars and hydrogen-lines are interpolated over. The target spectrum is then divided by the derived telluric transmission spectrum. A first order wavelength calibration is obtained from an argon-lamp spectrum, and fine-tuned by registering the bright OH-lines obtained from a sum of the pair of images of interest. Table 4 summmarizes the properties of the 3 objects.
The spectra (Fig. 8) confirm that all three candidates
with spectroscopic observations are T-dwarfs. We determined
their spectral types using the Burgasser et al. (2006) spectral
indices. Table 5 lists thoses indices and the corresponding
spectral type. We retain as our prefered determination
the ``weighted'' spectral types, rounded to the closest
half-integer. These ``weighted'' spectral types take into account the
better sensitivity
of those indices that vary most for a given subtype range.
The reddest target, CFBDS100113+022622 (
i'-z'=3.75), turn out to
be a T5
dwarf, and its 45 to 110 parsecs photometric
distance makes it one of the farthest mid/late-T dwarf currently
known. The large uncertainty on its distance is dominated by the
spectral type uncertainty
and the >1 dimming between T4 and T6 dwarfs, (Vrba et al. 2004), with
photometric uncertainties contributing less than 5%. However, the
absolute magnitudes of mid-T dwarfs are
themselves uncertain by as much as 1 mag (see for instance Liu et al. 2006). The faintest of
the three dwarfs, CFBDS095914+023655 (i'-z'>3.4) turns out to
have an earlier spectral type, T3, but lies even farther, between 120
and 130 parsecs. The indices of the brighest one, CFBDS193430-214221
(i'-z'=3.6), indicate a T3.5
spectral type. The spectral
type uncertainties are derived from the scatter
between the estimates from the various spectral indices.
Our survey has to date found 23 T dwarf candidates, 57 L or very-late
M dwarf candidates, and 4 high redshift quasars, out of 215 candidates
with i', z' and J magnitudes. These were drawn from
a larger sample of 770 candidates with i' and z' magnitudes, found
in 357 deg2 of i' and z' MegaCam images. Taking into account our
prioritising of the reddest candidates for J-band observations, we expect
that complete follow-up of these 770 candidates will yield
45 T dwarfs and 200 L dwarfs. Scaling to our final
800 deg2 of shallow surveys, RCS-2+Very Wide, then predicts
100T dwarfs and over 450 L or very late-M dwarfs, approximately doubling the
number of known brown dwarfs.
Our analysis of the CFHTLS-Deep and CFHTLS-Wide
surveys has, and will, yield additional candidates at large
distances, which will constrain the galactic scale height of
brown dwarfs. We plan to obtain spectra for the most exciting
of these many brown dwarfs, and expect that the large discovery
volume will produce even cooler objects than our recent T9/Y0 discovery, described in Delorme et al. (2008).
Acknowledgements
Thanks to the queue observers at CFHT and Gemini who obtained data for this paper (Gemini program GS-2006A-Q-16). Thanks to J.J. Kavelaars for advice on planning our MegaCam observations in the CFHTLS Very Wide and to Howard Yee and the RCS-2 team for making their proprietary data available. This research has made use of the VizieR catalogue access tool, of SIMBAD database and of Aladin, operated at CDS, Strasbourg.
![]() |
Figure A.1:
Finding charts from z' band images of the 3 T dwarfs
whose spectra are presented in this article. The fields are
![]() |
Open with DEXTER |