4 Survey description

 

Table 2: Prominent lines in Wolf-Rayet K-band spectra.

Central $\lambda$ ($\mu$m)	Transition	Type
2.0587	He I 2s-2p	WN, WC
2.0705/2.0796/2.0842	C IV 3p-3d	WC
2.1126/2.1137	He I 4s-3p	WN, WC
2.1038/2.1152/2.1155/2.1156	C III/N III 8-7	WC
2.1632	He I 7-4	WN, WC
2.1652	He II 14-8	WN, WC
2.166	Br $\gamma$, H I 7-4	WN
2.189	He II 10-7	WN, WC
2.2779	C IV 15-12	WC
2.3178	C IV 17-13	WC
2.3470	He II 13-8	WN

Data is reproduced from Table 2 of Figer et al. (1997), see references therein. Wavelengths listed are vacuum wavelengths.

Figure 1: A representative sample of WR stars with filter transmission curves overplotted, WN stars left panel and WC stars right panel. Dotted curves represent the line filters at 2.06, 2.08, 2.166, and 2.19 $\mu$m. Dot-dashed lines represent the continuum filters at 2.03, 2.14, and 2.248 $\mu$m. The He II filter transmission curve is difficult to see in the left panel due to the strong He II lines in the WN stars. WR spectra in both panels are kindly provided by P. Eenens.

In Table 1 we present central wavelengths and FWHMs for our chosen set of K-band filters (Blum & Damineli 1999a; Homeier et al. 2002). Four filters are centered on the characteristic stellar wind emission lines of He I 2.06 $\mu$m, C IV 2.08 $\mu$m, H I Br$\gamma$ 2.166 $\mu$m, and He II 2.189 $\mu$m, and the additional three continuum filters are at 2.03 $\mu$m, 2.14 $\mu$m, and 2.248 $\mu$m. Thus each line filter measurement has a continuum point to the red and blue. It is extremely important to have continuum points to the red and the blue of each line filter because of the variation in continuum slope caused by dust extinction.

In Table 2 we list the most prominent emission lines in WR spectra, and the filter response curves are overplotted in Figs. 1a and 1b for unpublished K-band spectra of WN and WC stars kindly provided by P. Eenens. In both, the three continuum filters are overplotted as dot-dash lines, and the four line filters as dotted lines. This illustrates the sensitivity of the 2.17 and 2.19 filters to WN stars, whereas the 2.06 and 2.08 filters are most sensitive to WC stars.

4.1 Observations and reductions

Figure 2: Map of the observations from 1996, 1997, and 1998. The x and y axes are in Galactic coordinates. Lines of constant right ascension are overplotted in intervals of 2 min as dotted lines, and lines of constant declination are overplotted every half degree as solid lines.

Figure 3: The same data as in Fig. 2 overplotted on the 90 cm radio image of the Galactic Center region presented in LaRosa et al. 2000. Major features are labeled, and the small rectangle indicates the approximate position of Fig. 5.

Figure 4: Same as Fig. 2, but for the 2000 data.

Figure 5: This is a sample of our data from 1996. The image shown here was taken with the 2.248 $\mu$m filter. The center coordinates are 17:46:35.7, -28:42:35 J2000. The y dimension spans $5 \hbox {$^\prime $ }$ and the x dimension spans $\sim$19 $\hbox {$^\prime $ }$. South is up and east is to the left. Large scale variations in stellar density are apparent and caused by intervening dark clouds.

In 1996 the survey was begun at the 1.5m at the Cerro Tololo Interamerican Observatory, and continued in 1997, 1998, and 2000. For 1996, 97, and 98 we used the same telescope and instrument configuration: the Cerro Tololo Infrared Imager (CIRIM) with the f/8 mode, giving us a $5 \hbox{$^\prime$ }\times 5 \hbox{$^\prime$ }$ field of view and $1.16 \hbox{$^{\prime\prime}$ }$ per pixel. In 2000, we used the Ohio State Infrared Imaging Spectrometer in the f/8 mode for a $10 \hbox{$^\prime$ }\times 10 \hbox{$^\prime$ }$ field of view and $1.16 \hbox{$^{\prime\prime}$ }$ per pixel. Our images are taken in "strips'' composed of 35-45 images. We offset 1/3 of a chip in RA (1996-1998 data) or Dec (2000 data), keeping the other coordinate constant. Thus each spot on the image strip is a composite of three exposures. Images have been obtained over $\sim$3 degrees of longitude between $\pm$0.5 near the Galactic Center. This is shown in Fig. 2, and overplotted on a 90 cm image of the Galactic Center region in Fig. 3. In 2000, we moved outward along the plane to l = 316, at the edge of Centaurus looking towards the Scutum-Crux spiral arm. In this region we obtained $\sim$1 degree of Galactic longitude between $\pm$1.0 Galactic Latitude. This is shown in Fig. 4. A sample of our data is shown in Fig. 5, with a small rectangle indicating the approximate position on the 90 cm radio image in Fig. 3.

Data reduction is performed with the CIRRED package of routines in IRAF, written specifically for CIRIM and OSIRIS reductions by R. D. Blum. The reduction steps are as follows. First, we trim the flat-field images if needed. For each filter there is a set of images taken with the lamps on and off. We take the median of each set of "on'' and "off'' frames to make a single "on'' and "off'' image. We subtract the "off'' frame from the "on'' frame to make a flat image, which is then normalized by the mean. A bad pixel mask is made by comparing the dome flats with the lights on and the lights off and using the histogram of pixel intensities to distinguish good and bad pixels. This bad pixel mask is then used to correct the flat field images.

Next, the survey images are trimmed if needed, linearity corrected with IRLINCOR, divided by exposure time, fixed with the bad pixel mask, and divided by the flat field image. Then the entire stack of images within a strip is used to make a sky image for each filter. The images are median combined using "minmax'' rejection with approximately half the images thrown out to reject contributions from stars. The resulting sky image is subtracted from each of the individual frames, and a constant is added back to maintain an appropriate sky level in the reduced image. These images are then assembled into a strip approximately one degree long by cross-correlation of the overlapping regions on each frame. Finally, we derive astrometric solutions for each of our strips by comparing the 248 filter images to the 2MASS catalog images from IPAC and the IRAF task CCMAP.

4.2 Photometry

Each frame is analyzed with DoPhot (Schechter et al. 1993). DoPhot identifies, classifies, and performs photometry on objects in an image. It makes successive passes over the image, subtracting the objects and searching in the next pass for fainter objects. The model parameters are found iteratively in the image itself, with reasonable first guesses supplied by the user in parameter files.

4.3 Apparent magnitude comparison

Our program has no strict requirement for calibrated photometry since emission-line stars are found through continuum independent line indices. Furthermore, much of the data was taken in non-photometric conditions. However, it is of interest to have an order-of-magnitude calibration in order to help assess how deep the images typically go. We have done a comparison between a typical image section of one of our image strips and the 2MASS survey images. We compared our instrumental magnitude for the 2.06 $\mu$m filter, which is the least sensitive of all our filters due to the combination of transmission efficiency and filter width. 2MASS magnitudes were obtained from the Infrared Science Archive (IRSA) through the GATOR query page. Our comparison shows that an instrumental magnitude of -3 corresponds roughly to an apparent K magnitude of 12 for this strip.

4.4 Completeness

Figure 6: The fraction of stars recovered in artificial star experiments for the four line images and its dependence on the continuum magnitude of the star. See text.

Figure 7: Standard deviation of the difference between the input color index and the measured line index for the 2.06 line artificial star experiment. The continuum magnitude is shown on the top of each small panel. The error in line index does not depend on line strength, but does depend on continuum magnitude.

To test our sensitivity to emission line stars, we created grids of fake stars with continuum magnitudes ranging from -2.0 to -7.0 in instrumental continuum magnitude, and emission line magnitudes ranging from 0.1 to 1.0 magnitudes brighter than the continuum. That is, we created grids where 100 stars have continuum (2.03, 2.14, and 2.248 $\mu$m) magnitudes of -2.0, and line (2.062, 2.077, 2.166, 2.191 $\mu$m) magnitudes of -2.1, 100 with the same continuum magnitudes, and line magnitudes of -2.2, and so on. At each step in continuum magnitude, we added 1000 stars with varying line magnitude.

The psf for each filter image was constructed using several bright and relatively well-isolated stars in the image and the IRAF tasks "pstselect'' and "psf''. Images were then created using "mkobjects''. These images were analyzed in exactly the same way as a real image. In Fig. 6 we plot the fraction recovered at increasing brightness of the continuum magnitude. There is a dependence on overall brightness, but no significant dependence on emission line strength. This is similar to what one would expect for a broad-band color.

The survey is limited by crowding (especially in the GC region) and the relatively coarse angular resolution element provided by the 1.5 m telescope. Given the order-of-magnitude comparison made above with the comparison to 2MASS, th  2.06 $\mu$m filter the data are $\ge$75% complete at K = 12 mag. However, the photometric accuracy decreases at faint magnitudes. This is shown in Fig. 7 for the same stars that are plotted in Fig. 6. The line shown here is the $2.062~ \mu$m line, but all lines (2.077, 2.166, and 2.191 $\mu$m) show very similar behavior.

4.5 Candidate selection

In this section we detail our candidate selection procedure. As a first step, the object coordinates must be transformed to a single system. We have arbitrarily chosen to use the longest wavelength filter image as our template. Thus, we scale and offset the coordinates for objects in the other lists to the 2.248 $\mu$m object list. This is accomplished with the "transform'' package which accompanies DoPhot. Transform uses a triangular search routine to match stars in different images by comparing the relative scales and orientations of stars in groups of three. Hereafter, we refer to the task as "offset''.

We look at each emission-line filter separately, meaning we consider sets of three filters each time, the line and its two continuum filters (2.03 and 2.14 for 2.06 and 2.08; 2.14 and 2.248 for 2.17 and 2.19). The raw output of DoPhot contains 10 000 to 30 000 objects per image, and this output must first be gleaned for stars with good statistics. The first cut is on object type. In the DoPhot output each object is assigned a number indicating the whether it is a star or a galaxy, and how well it thinks it can determine the magnitude. There are three object types we are interested in: 1 s are single stars with good statistics, 7 s are single stars with decent statistics, and 3 s are fit as members of a blend of 2 single stars. During the transformation to 2.248 $\mu$m coordinates, "offset'' keeps track of the object types in both lists, and records it in the output file. We keep only combinations of single stars (such as 11 s, 17 s, 71 s, or 77 s) and doubles (33 s). We throw out stars which have been classified as a member of a double in one image and a single star in another, as the photometry in this case is very unreliable.

After cleaning the lists for object type, we perform a cut on photometric error. The error is calculated simply as the errors of the individual magnitudes added in quadrature. The error cut is a free parameter in our selection routine, but we usually select it to be between 0.1 and 0.3 mag.

We now have three lists of objects, cleaned on the basis of object type and error. The coordinates in these three lists are now matched, and a "line index'' is computed as the magnitude difference between the measured value and the expected value derived from a linear interpolation using the two continuum filters magnitudes.

Essentially all stars in the images should have a constant line index (which is not zero due to filter transmission differences and sky transmission variations), that we can subtract to set the mean line index to zero. When the nights were not photometric, systematic line index variations as a function of position (time) along the strip occur. These are relatively straightforward to account for since they produce large scale changes in the line indices with position on the image. In our analysis we subtract these variations by calculating a "mean neighbor index'' for each star, using a bin centered on the star, with a size between 50-100 pixels. Smaller bins produce output with less scatter, but the bin must be large enough to contain stars to compute a significant "neighbor index''. This mean value is then subtracted from the individual objects.

Figure 8: The scatter as a function of $2.03 ~\mu$m magnitude is fit with a linear function to aid in candidate selection.

We also consider the scatter in line index as a function of the short wavelength continuum magnitude. An example is shown in Fig. 8, where the standard deviation in line index is fit as a linear function of short wavelength continuum magnitude. Our final selection is done by considering the line index, the error on this index, and the scatter in line index at this magnitude. We employ a criterion similar to the "S'' parameter employed by Damineli et al. (1997).

An example of our photometry and candidate selection is shown in Figs. 9 and 10, and candidate emission line objects are overplotted as asterisks. One is a confirmed WR star of late WC subtype (line index $\sim -0.6$, inst. mag. $\sim -3.7$) (Homeier et al. 2002). In right hand panel of each figure is the the same data plotted, but without correcting for systematic positional variations in line index as described above. One can see that the apparent scatter is larger for the objects in the right-hand panels, and in the case of the 2.08 filter, it affects which objects are chosen as emission-line candidates. In this particular case, a bona-fide WR star would be missed.

Figure 9: 2.06 Filter: an example of our photometric data and candidate selection. The stars marked with asterisks are emission line candidates, and one is a confirmed WR star (cindex $\sim -0.6$, mag $\sim -3.7$; Homeier et al. 2002). In the right panel is the same but without correcting for positional line index variations; see text.

Figure 10: Same as Fig. 9, but for the $2.08~ \mu$m filter. The confirmed WR star is located at index $\sim -0.6$, mag $\sim -3.7$.

4.6 Image subtraction

We also use image subtraction as a way to select high priority candidates. This step is only performed for those objects that are selected as photometric candidates. All procedures discussed here are accomplished with standard IRAF routines. An image section is selected within +/-125 pixels of a photometric candidate's x position, and a y pixel range from 1 to 250 pixels. First the images are cross-correlated and shifted to a common position. The two continuum filters are averaged and this result is normalized by the mean. The line filter is also normalized by the mean and then divided by the averaged, normalized continuum image. The result is displayed on the screen and examined visually. We have found that this provides a useful and necessary check on the photometric selection. Effects such as bad pixels and residual images are easily seen with this method. We will discuss candidate selection in further detail in the upcoming paper on the spectroscopic follow-up (Homeier et al. 2003, in prep).