A NEAR-INFRARED SURVEY OF THE INNER GALACTIC PLANE FOR WOLF–RAYET STARS. I. METHODS AND FIRST RESULTS: 41 NEW WR STARS

It is extraordinary but true that the galaxy of the Local Group whose global stellar populations are least well observed is our own Milky Way. Deeply immersed within the optically opaque, dusty lanes of our Galaxy's spiral arms, astronomers have been frustrated for centuries in their attempts to map the stellar populations of the Milky Way. A complete census of our Galaxy for every member of even one class of star would have seemed like an impossible goal even a decade ago. Major advances in instrumentation are transforming this daunting task from near-impossibility to increasingly likely. Wide-field, high-resolution, sensitive surveys, particularly in the near-infrared and X-ray parts of the electromagnetic spectrum (where the Galaxy is relatively transparent), are key to locating and characterizing all members of one or more classes of stellar object. The goal of the project described in this paper is to detect and spectrographically characterize at least 90% of the Wolf–Rayet (WR) stars in the Milky Way within 10 years.

A set of well defined tests of stellar evolution theory will follow from detections of complete samples of WR and other related stars. For example, the radial abundance gradient across our Galaxy and the increase of the WR/O number-ratio with increasing Z suggests that many more WR stars will be found in the inner parts of the Milky Way than in the outer regions. In addition, our previous Hubble Space Telescope (HST) survey of the H ii regions in the ScIII galaxy NGC 2403 (Drissen et al. 1999) suggested that the distribution of WR and red supergiant stars (RSG) is a sensitive diagnostic of the recent star-forming history of these large complexes: young cores of O and WR stars are surrounded by older halos containing RSG. Theory predicts that the number-ratio WR/O increases with increasing metallicity; thus, relatively fewer WR stars form at lower Z. We will also be able to determine if superclusters, dominated by WR stars, are common in the Milky Way. Finally, we note that WR stars are predicted to end their lives as supernovae, and in rare cases as gamma-ray bursts. The WR stars in the Milky Way may be abundant enough for one to erupt as a Type Ib or Ic supernova within a few generations. This comes from the assumption that the MW contains ∼6000 WR stars, each lasting ∼5 × 10⁵ yr. The clear identification of a WR star as the progenitor of one of these eruptions would be a dramatic confirmation of a key prediction of stellar evolution theory.

WR stars are massive (with initial masses greater than ∼20 M_☉ at Z_☉) stars with strong winds ($\dot{M} \sim 10^{-5}\;M_\odot$ yr⁻¹) displaying the heavier elements created by what are normally internal nuclear processes. Distinctive spectra with strong, broad emission lines of helium, and either nitrogen (WN) or carbon (WC) are the defining observational characteristics of WR stars. As they have relatively short lifetimes (about 5 × 10⁵ yr), WR stars are excellent tracers of star formation, and they are also believed to be type Ib or Ic supernova progenitors, because they have removed their outer H-rich layers (WN) or even He-rich layers (WC/O).

About 300 WR stars have been previously identified in the Milky Way (van der Hucht 2006), with distribution models predicting ∼1000–6500 total expected (Shara et al. 1999; van der Hucht 2001). Optical narrowband surveys have been severely limited by interstellar extinction (Shara et al. 1999), and so the natural solution is to turn to the near infrared. Emission line magnitudes of 40 known WR stars are presented in Appendix A. A model of the Milky Way, predicting the numbers and distributions of WR stars visible in the K band is presented in Appendix B.

In an initial attempt at a narrowband near-infrared survey Homeier et al. (2003) had limited success, while Hadfield et al. (2007) were somewhat more successful with their color-based selection of objects from the Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006) and GLIMPSE (Benjamin et al. 2003) surveys.

Utilizing a new narrowband survey, described in Section 2 along with the candidate selection methods, we have found 41 new WR stars: 15 WN and 26 WC. Spectrographic follow-up and data reduction are described in Section 3. Our resulting K-band spectra, line measurements, and subtype classification are presented in Section 4. Distances to the new WR stars are calculated and their distribution within the Galaxy considered in Section 5. Our conclusions are summarized in Section 6.

The more than 77,000 science and dome flat images of the survey require a customized, streamlined pipeline to reduce this large amount of data. The pipeline was constructed (by JG and DZ) in IDL and uses the 2MASS catalog extensively as a reference for both astrometry and photometry. Each of the ∼1200 fields has been imaged in each of the eight filters at seven dither positions with separations of ∼15''. Dome flats were used to flatten each of the images for sensitivity and chip defects. These seven dither positions were median combined together without shifting to remove most of the stellar sources and to create a sky image. The sky image was subtracted from each image.

Sources for each field/filter were matched to 2MASS to determine the world coordinate system (WCS). Once a WCS was fit, another iterative process was performed to minimize the residuals to 2MASS and determine the best geometric distortion solution. The IDL procedure WARP_TRI applied the geometric distortion solution. The IDL procedure HASTOM aligned all the images taken in a dither series. These images were combined to create a final deep exposure.

Sources were identified as WR candidates through emission in the narrowband filters. This made it necessary to have not only the magnitude of a source in the narrowband filter, but also the magnitude of the continuum at that wavelength. As a result, each narrowband filter had to be examined concurrently with the CONT1 and CONT2 filters.

Sources on the final deep exposures were detected using the IDL procedure FIND. Aperture photometry was then carried out using the IDL procedure APER with a 2 pixel aperture and a sky annulus from 10 to 20 pixels. Objects matched to the 2MASS catalog (at least 100 in each field) were used to determine a zero point for each filter in each deep image. An object with a flat spectrum through our filters has the same magnitude in all six filters.

Sources were considered matched across filters if their positions were consistent within 0.5 pixel. These matched sources with 2MASS-calibrated magnitudes were then used to construct emission–magnitude diagrams (EMDs). An EMD was constructed for each narrowband filter with the wavelength-interpolated continuum magnitude versus continuum-subtracted narrowband magnitude. The stars scatter around a continuum-subtracted narrowband magnitude of zero. In each magnitude bin we calculate the standard deviation sigma of the continuum-subtracted narrowband magnitude. Objects that are 5σ or more in the negative direction from the locus of stars are considered candidates. We determined the offset to convert instrumental magnitudes to apparent magnitudes using the 2MASS Ks band catalog. The images were divided into an 8 × 8 grid, with each of the 64 areas having an individually determined offset, to compensate for an observed color dependence (probably due to variable reddening) across the field. The IDL procedure APER determined the magnitude for the source at the coordinates given by 2MASS.

Once the offsets to convert to apparent magnitude were calculated, FIND, an IDL procedure, identified sources in an image that were a given deviation above the background. APER found the instrumental magnitude for the sources, which were then converted to apparent magnitudes. The CONT1 filter was taken as a reference image and HASTROM aligned all filters from a field. The sources from CONT1 were then matched to the sources from the CONT2 image and sources within 0.5 pixels were kept as matches. The matched sources were then compared to the sources of a narrowband filter. The result was a list of sources found within 0.5 pixels of each other in both of the continuum filters and the narrowband filter. A linear fit was found between the CONT1 and CONT2 filter magnitudes of each source. Then, using the central wavelength for the narrowband filter, an interpolated continuum value at the narrowband wavelength was determined. The magnitude of the source in the narrowband filter was then subtracted from the interpolated continuum value, giving the negative emission magnitude for the source in that narrowband filter. The emission magnitude for a source was also estimated by subtracting the CONT1 and the CONT2 magnitude from the narrowband magnitude, resulting in a total of three estimates for the emission magnitude of a source. An EMD of continuum magnitude versus emission magnitude was created for each narrowband filter and the standard deviation was determined for the sources in bins of 1 mag. Sources that had emission magnitudes of 5σ or greater from the center of the EMD were marked as WR candidates. 1 arcmin × 1 arcmin finder charts were produced for each WR candidate, showing the candidate in all filters of the survey and also in XDSS red and XDSS infrared images. These are presented in Appendix C. Candidates with emission magnitudes that were similar to those of the known WR stars covered in the survey were selected for spectrographic follow-up. This refined list of candidates was blinked by eye to remove any candidates that did not resemble stars.

In this initial, exploratory phase of the survey we used two techniques when selecting targets for spectroscopic follow-up. We began by selecting targets with such powerful emission lines that the star appeared brighter in the narrowband images when compared to the continuum images even when examined by eye. This corresponds to a minimum of 0.5 to 1 mag difference in brightness between the narrow- and broadband images. We initially selected candidates displaying a brightening of at least 0.5 mag in at least one narrowband image relative to the continuum. This resulted in the detection of 34 new planetary nebulae (which will be reported elsewhere) whose very strong, sharp emission lines, He i 2.058 μm and Brγ 2.166 μm, fell within our He i and Brγ narrowband imaged fields. A few of the planetary nebulae were slightly resolved on the Brγ narrowband images. No new WR stars have yet been found this way.

Our second, much more successful, technique relied on using known WR stars to calibrate our selection of targets. Forty known WR stars were selected within the survey area, and patterns were found in their narrowband Δm values which distinguished broad subtypes of WR stars (see Appendix A). WC stars generally showed strong (−0.8 mag or less) emission excess in the C iv filter, and slightly weaker emission (between −0.4 and −0.8 mag) in the He i filter, due to the blue side of the C iv line extending into the range of the He i narrowband filter. Early WN stars generally showed moderate emission in the Brγ filter and the He ii filter, but slight absorption (0.1 mag) in both He i and C iv.

Using these criteria, 173 candidate targets were selected which appeared at least 5σ brighter in their narrowband filter than did other stars in the field and which also fit the criteria suggested by the known WRs. Though we are reliably detecting stars to magnitude 14–14.5 in all of the filters (by judicious choice of exposure times), during this exploratory stage candidates were selected to have emission-band magnitudes brighter than K = 11.5. This is because the initial spectrographic follow-up is being done with a 1.5 m telescope (see below). (Thousands of uncrowded 5σ candidates as faint as K = 14.5 will be the subjects of future papers.) There are strong selection effects for those WR subtypes which were used to determine the selection criteria, and as a result, no WC9 or late WN type (>WN6) stars have yet been found. Improvements now underway in survey image reduction will permit discovery of the less strongly distinguished subtypes.

The iraf task splot was used to fit the continuum and then optimize and deblend Gaussian functions to fit the observed emission lines in all spectra. This gave measurements of line centers, equivalent widths (EWs), and FWHMs.

A number of errors contribute to reduce the accuracy of these measurements. Blending of many lines creates the largest emission features, and introduces errors into line center measurements up to 10 Å. It also skews the shape of the feature to be a poor fit to a Gaussian. In general, the rms error of the residuals for the fits was about 10% of the continuum value. Additionally, the continuum was quite difficult to determine precisely due to the abundance of emission lines throughout the spectrum, especially in WC stars, resulting in errors in the equivalent width measurements. A number of our WN detections were rather faint, with peak fluxes only twice the continuum level, resulting in lower signal-to-noise ratios for these spectra. Finally, ground-based observatories must peer through the murky atmosphere, so that the strong C iv 2.08 μm line in WC stars falls on the edge of the equally strong atmospheric absorption feature at 2.06 μm. Division by a standard star removes most of this feature; however, changes in atmospheric conditions between object and standard-star observations result in residual features on the blue side of the emission line.

The strong emission lines visible in the spectra, while easily distinguished as belonging to either type WC or WN, are the result of overlapping blends of emission lines of various elements. In the optical, WR stars are categorized by looking at ratios of equivalent widths of various nitrogen species for WN, carbon and oxygen species for WC, and He ii and He i for both. However, the heavy blending of lines present in the K band makes it more difficult to find either isolated spectral lines or distinguishing ratios of blended lines (Figer et al. 1997).

Spectral subtypes were assigned following Crowther et al. (2006). These are presented in Tables 3 and 4, along with the ratio of equivalent widths used for categorization. The ratio W_2.189/W_2.165 was used to categorize the WN stars, and the ratio W_2.076/W_2.110 was used to categorize the WC stars. Also following Crowther et al. (2006), WN stars with FWHM(He ii 2.189) ⩾ 130 Å were classified as broad/strong, and the letter "b" appended to the subtype designation. Subtypes thus assigned are expected to be accurate to within one subtype. The 40 known WR stars were similarly assigned subtypes, which agreed within one subtype of their published spectral classifications.

Those WC stars with especially broad, heavily blended C iv and C iii lines, as presented in Figure 1(e) and which match our observed spectra of WR19 (WC4) and spectra observed in Figer et al. (1997) of WR146(WC4), WR143(WC5), and WR150(WC5), are classified more generally as WCE.

Blended lines complicate the calculations of accurate radial velocities (RVs), as there is no longer any fixed line center with which to compare theoretical and actual wavelengths. The purest line in our K-band spectra, He ii at 21891 Å, allows calculation of RV with respect to the motion of the Sun in principal; however, this line is extremely weak or not discernable in most of the WC stars. Line-center measurements are further complicated by the difficulty of accurately fitting Gaussian curves to WR emission lines, as described in Section 4.1, resulting in errors of 5 Å to 10 Å in determination of peak wavelengths. This corresponds to errors on the order of 50–150 km s⁻¹. In general, the direction of motion with respect to the Sun follows the clockwise rotation of the Galaxy; however, as noted, the error on these measurements is large. Measured RVs (based on the measured line centers, which may be shifted, depending on details of the line formation mechanism) for WNs and, when possible, WCs, are included in Tables 3 and 4.