The newly identified carbon stars, presented here, were found by estimating
the magnitudes of stars present on IIIaJ and IIIaF UK Schmidt sky
survey plates. The plates were measured using the APM facility (Kibblewhite
et al. 1984) and the image lists for each pair were combined to
produce a catalogue of magnitudes and 
colours for each
field. The instrumental magnitudes were internally mapped to a "linear''
magnitude scale using the technique described by Bunclark & Irwin
(1983). These instrumental magnitudes were then transformed into R
and 
by making use of the known properties of the foreground
Galactic stellar population and the presence (in most fields) of a well
defined red horizontal branch/clump due to the Cloud population.
We have successfully used this so-called "internal calibration'' technique
on many occasions where no standards are directly available for calibration,
see for example Irwin et al. (1990). This method of
calibration is reliable at the 0.1-0.2 magnitude level and is more than
adequate for selecting candidate carbon stars from the broad swathe of AGB
stars easily visible in most fields (see for example Fig. 4 of Irwin et al.
1990). For two of the fields surveyed in DIK we had of the order of
100 calibration stars available per field which provided an independent
check of this methodology and colour transformation equations given below.
To facilitate comparison with other carbon star surveys, it is possible to
define a transformation from the R, 
system to the B, V
system using simple equations such as the ones given by Demers & Irwin
(1991):
Candidate Cloud carbon stars were selected from stars with 
redder than 
, and bounded by 14 < R < 17 with the exact
colour boundary varying by 
from field to field in order to
minimise the likely foreground contamination. A 
colour of
2.4 corresponds to B-V = 1.85, i.e. equivalent to selecting spectral types
later than about M 5.
The UKST fields surveyed up to the end of 1994/95 season are listed in Table 1 (click here). The F number refers to the ESO/SERC atlas field descriptors. The last three fields do not coincide with ESO/SERC field centres, and are the fields from the DIK pilot study. Figure 1 (click here) shows the location of the carbon stars identified in the fields of Table 1 (click here), together with the plate boundaries. Particularly toward the inner parts of the LMC/SMC not all of the candidates were followed up, since we were mainly interested in obtaining good spatial coverage around the Clouds rather than an exhaustive list of carbon stars. However, that caveat notwithstanding, all candidates in the periphery regions were followed up since we wanted as a byproduct of the survey to find the extent of the intermediate age population of the Clouds. The WORC and LMC stars, given in Tables 16 and 17, are not plotted. These stars are much closer to the LMC centre.
Figure 1: Distribution of the newly identified carbon stars.
Candidates were selected to have colours 
and 14 < R <
17 and were spectroscopically confirmed as Cloud carbon stars. Toward the
inner parts of both the SMC and LMC not all candidates were followed up
Table 1: UK Schmidt fields surveyed
On all but three out of 23 nights between August 18th,
1991 and January 5th,
1995 slit spectra were obtained with the MODular spectrograph (MODspec) on the
duPont 2.5 m telescope. A Boller and Chivens spectrograph was briefly used
early in January 1992 but abandoned after showing a
lesser throughput. With a 831/7500 grating MODspec covered the range
at a
resolution of 2.8 Å FWHM with a CRAF 
thick
CCD sporting 
pixels, and 0.66 arcsec per pixel spacing along the
slit. Trials at higher
dispersions with a lower efficiency 1200/7500 grating
were abandoned after two
nights, to maintain the higher discovery rate. Slit
widths of one arc-second
were used throughout the survey. Exposure times varied between 120 s
and 1200 s, with 60 percent taken at 180 s. The longest exposures
were reserved for a class of carbon stars showing only weak CN bands,
identified by the symbol "wk C'' in the tables. It is quite likely that
many of these wk C stars could be late-type M giants rather than carbon
stars, although spectra of much longer integration times would be
required to unambiguously assign types. We have left these objects in the
tables because of the uncertainty of their nature and because their velocities
indicate that the majority have a high probability of being Cloud members and
hence provide additional useful kinematic constraints.
Although HD 16115 was the only template star observed,
a set of ten program
stars was observed on most of the nights as a means
of controlling the
internal velocity system. Six of these are in common with the work of
Hardy
et al. (1989) and with Blanco et al. (1980), and another
four (DIK 01, DIK 02, DIK 03 and DIK 08)
from the Inter Cloud Region (ICR). Fe-Ar lamp spectra
were obtained with every stellar spectrum. In the course of the survey a
number of additional stars were reobserved for a variety of motives. Some
stars showed unexpected velocities, occasionally more than 
different
from the expected values. Such stars were observed
three times, to permit
discarding velocity variables or bad measurements.
Other repeat observations
occurred fortuitously, as candidates appearing on more than one search list.
Table 19 (click here) summarizes data
statistics of stars
observed four or more times, including the ten internal velocity controls.
The table shows that two of the "control'' stars in
common with Hardy et al.
(1989) and Blanco et al. (1980) are velocity variables:
Wing-4 (Blanco 13) and Wing-5 (Blanco 26).
75 stars were observed at least three times,
and almost half were observed twice.
In all, 1340 pair comparisons allowed setting velocity offsets for any given
night to within 
of the adopted mean for the
entire data set. The
heliocentric velocity (v) system of our data differs from that of
Hardy et
al. (1989) in the sense: 
. Eliminating
the two velocity variables from the list reduces the
difference to 
.
Radial velocities were determined using a cross-correlation technique adapted to the heterogeneous nature of the data (containing a mix of various carbon types) and to the presence of telluric features contaminating portions of the spectral range. Early trials relied exclusively on the CN bands at 7910, 8100, and at 8320 Å. Later trials led to partly masking the 8100 and 8320 Å CN bands, and adding a spectral region centered on the Ca triplet (from 8480 to 8700 Å), thereby avoiding telluric contamination and adding metallic lines useful with spectra with weak CN features. These windows were given a trapezoidal shape to minimise the effects of accidental contributions from window edges ("ringing'' in FFT jargon). The windows finally adopted for the work here reported cover from 7780 to 8128 Å and Fe, Ti, and Ca lines from 8138 to 8690 Å. Prior to correlation the data were high-pass filtered with an Ormsby zero phase shift filter that begins attenuating features longer than 30 Å, completely suppressing those longer than 60 Å. The mix of carbon types encountered produced correlation functions containing odd terms so that the method of Tonry & Davis (1979) no longer yields reliable error estimates. The presence of the odd terms is seen in an asymmetry of the correlation function about the zero-lag peak; its origin in the data (as opposed to noise) is indicated by its systematic repetition in well-exposed spectra of certain stars, and has led to an independent method for estimating observational errors, described below.
It is well known that exposure levels yielding good cross-correlation functions are often inadequately exposed to permit assigning spectral types from visual inspection. That has proved to be true in the data set here described. Observations of stars from the list of Westerlund et al. (1978) and re-observed by them later (Westerlund et al. 1991) proved mostly to be of spectral types C3 through C5. A class of stars was encountered in which the CN bands, though present, were quite weak, requiring stronger exposures to produce satisfactory correlation plots. After the early preliminary reductions it is simple to assign classification to the carbon type by looking for common features in the cross-correlation function.
When ordered according to the strength of exposure, the growth of
the zero-lag peak in the correlation function (normalised to unity for
a correlation of the template with itself) was found to provide a
reliable estimator for sequentially classifying spectra according the
precision with which velocities can be measured, as shown from repeated
observations. Figure 2 (click here) shows a sequence of correlation
functions (cf. Tonry & Davis 1979) in order of increasing
strength (roughly) of the correlation peak with respect to sidebands
lying within 
on either side. The weakest spectra, assigned Q = 0, do not
permit identifying a zero-lag peak. Q = 1 was assigned when a zero-lag
peak could be identified, but sidebands were of nearly comparable height,
or the width of the zero-lag peak was significantly broadened. Q = 2
was assigned when zero-lag peaks were as narrow as high quality spectra,
and the two strongest sidebands lying within 
summed to more than the zero-lag peak. Q = 3 was assigned when the sum
of these two sidebands was less than the zero-lag peak. The assignment of
higher Q-values was somewhat subjective, and relied on the orderly
appearance of sidebands, none with maxima greater than half the zero-lag
peak. Correlation functions with Q's of 5, 6 or 7 had high zero-lag
peak values > 0.3, and side-band patterns whose structure tended to
repeat in other spectra; the afore-mentioned asymmetries. From repeat
observations spectra with Q-values of 4 or greater give full precision;
spectra with Q-values of 3 give useful velocities when combined with at
least one additional observation of comparable quality. Q's of 2 were
used to plan follow-up observations later in the program.
Figure 2: Sequence of correlation functions used to assign
the quality of the radial velocities. Ordinates are normalised with respect
to the zero-lag peak of the template spectrum cross-correlated with itself
(top panel). The small tick marks on the abscissa denote intervals of
100 km/s
The colour boundary used is targetted at finding cool carbon stars near the
tip of the AGB and is specifically aimed at selecting a uniform sample of
an intermediate age population. A colour of 
corresponds to B - V = 1.8 to 1.9 and guarantees a very clean sample of
Cloud stars free from significant Galactic foreground contamination.
Clearly, it will not generally be sensitive to the bluer CH-type Carbon
stars, which are themselves readily detectable in conventional IIIaJ prism
surveys due to the strong 
bands blueward of the emulsion cutoff
(e.g. Morgan & Hatzidimitriou 1995). Our sample is only
incomplete within a degree or so of the inward limits of the survey where
the number of candidates is much larger than could be observed, (e.g.
fields F033, F055, F084). In such cases two search modes were adopted,
with time divided roughly evenly between them. One mode was to work
inwards, beginning on the periphery of the LMC and working toward the
rotation center. The other approach was to start with the reddest objects in
the list, and work blueward. Both methods still allow the formulating of
conclusions based on the sampling statistics, which will be the subject of a
later paper. However, apart from the preceding caveat we note that since
all the carbon stars in our sample are cool AGB stars, because of the
uniform colour limits (i.e. 
) imposed, they form part
of a well defined survey for intermediate age cool stars.
There is considerable overlap between the processed regions from the different survey fields and we used these overlap regions to estimate the completeness of the survey for those parts of the Clouds where we have followed up all the candidates. The probability that a candidate found on one plate is also found on the overlapping area of a second plate varies from 67 percent to 90 percent, the mean being 85 percent. This efficiency shows a slight dependence on crowding.
There are three main reasons why the candidate selection will not be 100
percent complete. First, the degree of image crowding, particularly toward
the inner parts of both Clouds, can be severe on long exposure UKST survey
plates where the typical FWHM of stellar images is between 2 and 3 arcsec.
This causes a varying fraction of the images to overlap neighbouring images
making it difficult to both detect them as discrete entities and to estimate
reliable magnitudes. In order to keep the number of candidates to a minimum
we required images to be classified by the APM as stellar on the R plate
and to be approximately stellar on the 
plates (i.e. not
obviously overlapped with neighbours). This produces a much cleaner CMD
with significantly fewer outliers at the expense of losing between 5 and 20
percent (depending on the image crowding) of the real outliers. Second, the
image classifier is statistical in nature and at the magnitudes of interest
can misclassify up to 5 percent of the images due to faint overlapping images,
spurious grain noise etc.. Finally, photographic plates suffer both
from systematic, or field, errors at the 
percent level, and random
photometric errors (
) level. Of necessity this causes the colour
selection, and to a lesser extent the magnitude, boundary to be somewhat
"fuzzy'' with respect to true colours (and magnitudes). Taken together these
artefacts of selection from photographic survey material imply that we
will miss between 10 and 25 percent of the true outliers in a CMD and are
consistent with the number of carbon stars found in common in the overlapping
regions between different fields.
In practice the estimate from the overlaping zones tends to be pessimistic
because these outer regions are precisely the zones of the plate where
the worst systematic effects occur and because with the benefit of the
overlap between fields, at least 25 percent of the area is surveyed more
than once. Consequently, our best estimate of the likely completeness of
the survey is 
percent.
Statistics on the "success rate'' of the whole selection procedure are not
straightforward to estimate because of several effects. Firstly,
the fields where the AGB tip was most clearly populated, and hence defined,
were precisely those fields where not all candidates were followed up.
Secondly, the accuracy of the photometry, particularly the percentage of
spurious outliers, depends strongly on the degree of crowding in the field.
Thirdly, and related to the first two points, the ratio of the surface
density of Galactic stars to Cloud stars varies strongly from field to
field. Finally, the spectra were never intended to be of good enough quality
to differentiate reliably between M dwarf Galactic foreground stars and M
giants in either Galaxy or Clouds. However, in those fields for which we have
completely sampled the candidates we can define three categories of object:
Cloud carbon stars; M giants with the Cloud velocity and a miscellaneous
category of leftovers. The latter category includes: otherwise bluer objects
thrown into the sample by photometric errors; indeterminate spectra
and M stars with velocities seemingly incompatible with Cloud membership.
(Distant cool Galactic carbon stars are so rare that they are not a
problem). For these completely surveyed fields, the breakdown for detecting
carbon stars was 
; Cloud M stars / wk C stars 
;
and the remainder 
. The high success rate at finding
bona-fide Cloud carbon stars rather than Cloud M giants is skewed by the
higher proportion of LMC population surveyed, where the C star to M star
ratio is much higher than for the SMC.