THE DISTANCE TO THE MASSIVE GALACTIC CLUSTER WESTERLUND 2 FROM A SPECTROSCOPIC AND HST PHOTOMETRIC STUDY^*,†,‡

2.1.1. HST

2.1.2. Optical Spectroscopy

2.2.1. HST Photometry

We performed photometry on the long and short exposures in each filter using the point-spread function (PSF)-fitting stellar photometry package HSTphot (Dolphin 2000) that was specially designed for the undersampled PSF of the WFPC2. HSTphot provides both instrumental and standard UBVI magnitudes, using the standard reduction recipe and transformations of Holtzman et al. (2006) as updated by Dolphin (2009). This procedure includes masking out the bad pixels, generating sky images, masking cosmic rays, and cleaning hot pixels. HSTphot performs a two-pass search for stars in each WFPC2 chip to locate brightness peaks, performs PSF-fitting photometry in an iterative manner, and produces a "refined" photometric measurement and position for each star with an astrometric precision of 0.03 pixels. The code matches positions of stars detected in more than one filter to provide source-matched multi-color photometry. The initial run of HSTphot yielded 2251 and 577 candidate detections in at least one filter for the "long" and "short" images, respectively.

It is a known characteristic of HST/WFPC2 images that spurious sources arising from, for example, diffraction spikes near bright stars, are included in this list. In an attempt to remove false sources we applied a series of additional criteria to the initial source list. HSTphot reports a series of "global" parameters for each source, based on the combined detections in one or more exposures and one or more filters. These include the source's roundness, sharpness, χ (as a goodness of fit), object type (point source or extended), and S/N. We determined a threshold roundness parameter, used to discriminate between point-like and elongated objects, by fitting a one-sided Lorentzian profile to the distribution of roundness values and retaining only sources having roundness parameter less than 1σ above zero, corresponding to roundness <0.37 and <0.19 for the "long" and "short" exposures, respectively. This retains 64% and 79% of the sources for the "long"/"short" exposures, respectively. The rejected stars are invariably those with the lowest S/N below ∼10. Next, as per the HSTphot manual suggestion, we retained only sources having |sharpness| ⩽ 0.5. This criterion removes only a handful of stars that have not already been rejected by the roundness cut. Finally, we retained only sources with global S/N ⩾ 10 in order to ensure that the final catalog contains only highly reliable sources. Applying these criteria, the final list contains 967 and 360 stars in at least one filter for the "long" and "short" images, respectively: 194/171 in F336W, 227/185 in F439W, 348/286 in F555W, and 918/343 in F814W. We performed a visual inspection of the images to confirm that these selection criteria retained real stars while rejecting obvious spurious sources such as diffraction spikes.

Figure 2 shows the photometric uncertainty as a function of instrumental magnitude for the short exposures (dots; colored blue in the online journal) and long exposures (asterisks; colored red in the online journal) in each filter. Typical uncertainties are <0.01 mag for stars <15 mag in both the long and the short exposures for all filters. Uncertainties rise to 0.2 mag at the faint end. This occurs for stars with F555W ≃ 20 in the short exposures and F555W ≃ 21.5 in the long exposures. Stars with uncertainties exceeding 0.1 mag appear in these plots because, on a per-filter basis, the S/N may be greater than the threshold global S/N of 10:1. The sudden drop in source density in the F814W panel (lower right) above uncertainties of ∼0.1 mag arises because of the global S/N criterion imposed. Since F814W is least affected by reddening, most sources are detected more strongly in this band and, therefore, the F814W photon statistics drive the global S/N values.

Zoom In Zoom Out Reset image size

For the 190 sources detected on both the "short" and "long" exposure images, we compared the photometry in each HST filter to assess any systematic differences between the two orbits observations. Figure 3 plots the differences between the short- and long-exposure instrumental magnitudes for each HST filter as a function of long exposure magnitude. This figure shows that there are no systematic differences between exposures, except at the faintest levels where the asymmetric nature of Poisson noise produces a systematically brighter magnitude measured in the short exposures relative to the long exposures. However, the rms deviation from zero is much larger than the typical photometric uncertainty. The reduced χ² exceeds 8 in all cases, indicating that there are additional sources of photometric uncertainty that are not included in the errors calculated by HSTphot. This is not an effect of chip position or crowding, but rather is a known limitation of existing HST photometry packages when dealing with very short exposures (Dolphin 2000). Therefore, we adopt photometry from the long-exposure images, when available. For stars appearing on only the short-exposure images (138 of the 1136 stars), we adopt a larger photometric uncertainty that is the quadrature sum of the uncertainty estimated by HSTphot and an uncertainty determined empirically from the dispersion in Figure 3 using 1-magnitude wide intervals. This procedure yields $\chi ^{2}_{\nu }$ values of less than two in all of the filters. Stars detected on the "short" or "long" exposure images are identified as such in the last column in Table 2.

Zoom In Zoom Out Reset image size

Table 2. Photometry

ID	MSP91	α (J2000 °)	δ (J2000 °)	F336W	σF336W	F439W	σF439W	F555W	σF555W	F814W	σF814W	U	B	V	I	S/N	Source
1	...	155.93865	−57.75719	99.999	0.000	99.999	0.000	17.621	0.023	16.585	0.009	99.999	99.999	17.597	16.547	147.5	l
2	...	155.94239	−57.75466	17.757	0.016	17.640	0.014	16.687	0.009	15.521	0.004	17.780	17.567	16.661	15.478	320.0	l
3	...	155.94329	−57.75762	99.999	0.000	99.999	0.000	20.203	0.075	17.438	0.010	99.999	99.999	20.270	17.461	109.1	l
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
66	...	155.96852	−57.75353	99.999	0.000	99.999	0.000	99.999	0.000	20.894	0.094	99.999	99.999	99.999	99.999	11.4	l
67	273b	155.96877	−57.76713	18.633	0.018	18.534	0.016	17.529	0.013	16.454	0.006	18.656	18.462	17.503	16.411	231.7	l
68	...	155.96880	−57.75063	99.999	0.000	99.999	0.000	20.986	0.137	18.910	0.022	99.999	99.999	20.996	18.890	49.7	l
69	...	155.96893	−57.74199	99.999	0.000	99.999	0.000	99.999	0.000	21.443	0.111	99.999	99.999	99.999	99.999	10.2	l
70	...	155.96899	−57.75914	99.999	0.000	99.999	0.000	21.375	0.176	19.061	0.023	99.999	99.999	21.401	19.053	47.8	l
71	...	155.96914	−57.75185	99.999	0.000	99.999	0.000	99.999	0.000	20.238	0.055	99.999	99.999	99.999	99.999	19.9	l
72	273a	155.96941	−57.76712	15.801	0.004	15.118	0.004	99.999	0.000	99.999	0.000	15.438	15.059	99.999	99.999	429.6	l
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
1134	...	156.05588	−57.74563	18.930	0.105	18.966	0.127	18.005	0.081	16.861	0.077	18.991	18.895	17.979	16.819	42.2	s
1135	...	156.05922	−57.74585	15.547	0.030	15.330	0.022	14.785	0.030	14.082	0.024	15.548	15.298	14.761	14.045	264.7	s
1136	...	156.05973	−57.74560	19.166	0.223	99.999	0.000	99.999	0.000	18.597	0.202	19.362	99.999	99.999	18.560	10.1	s

Only a portion of this table is shown here to demonstrate its form and content. Machine-readable and Virtual Observatory (VO) versions of the full table are available.

Download table as:  Machine-readable (MRT)Virtual Observatory (VOT)Typeset image

Coordinates for each star were obtained using the task METRIC in the Space Telescope Science Data Analysis System (STSDAS) package. For a subset of 120 bright isolated stars we identified counterparts in the 2MASS Point Source Catalog (PSC) and used the mean differences between positions to correct the headers of the HST images. The position header parameters obtained from the guide stars were corrected by (Δα, Δδ) = (027182, −057479) for the "short" exposure images, while the "long" exposure images required a correction of (Δα, Δδ) = (− 001137, −052219). After astrometric correction, the rms deviation between positions measured on the HST images and the 2MASS coordinates is 01 in R.A. and Dec. This is consistent with the limitations imposed by the WFPC2 pixel size coupled with the precision of the PSC astrometry.

Table 2 lists the final band- and exposure-merged photometry in each filter, totaling 1136 unique source IDs ordered by right ascension. A few lines appear in the printed journal as a sample of the table format. The entire content is available as a machine-readable table in the online journal. Column 1 gives the ID number from this work. Column 2 provides a cross-identification with the nomenclature of MSP91. In many instances the angular resolution of HST reveals several sources within a ∼2'' radius of the nominal MSP91 star. These are denoted by the MSP91 (MSP) designation with an alphanumeric suffix "a" for the brightest component, "b" for the next brightest component, as measured in the F555W band, etc. Columns 3 and 4 give the (J2000) right ascension and declination in degrees. Columns 5 through 12 list the instrumental magnitudes and uncertainties in the HST filter system. Columns 13 through 16 list the transformed magnitudes in the standard UBVI. Column 17 lists the global S/N value. Column 18 lists whether the photometry is from the "long" exposure image (l) or the "short" exposure image (s).

We caution that the transformations from the HST to Johnson system used by HSTphot are based on stars having a limited range of color, and that many of the Wd2 stars fall at the extreme red end of those calibrations. These reddest stars are late-type, low-gravity giants rather than heavily reddened early-type stars with high surface gravity that typify the Wd2 membership. Holtzman et al. (1995) further caution that the transformation are known to depend on metallicity and surface gravity. We conducted our own synthetic photometry on reddened O-star spectra to assess the extent to which the HSTphot transformations are appropriate to our targets ((B − V) ≃ 1.4; (U − B) ≃ 0.3.) We found that, for the B and V bands, the transformations are appropriate (within 0.02 mag) for reddened O stars. In the U and I bands, systematic uncertainties exceeding 0.05 mag may be present, as noted by Holtzman et al. (1995). Hence, where high levels of photometric accuracy are required, the HST instrumental magnitudes are preferred over the UBVI magnitudes. We use the HST instrumental magnitudes hereafter in our analysis, but we provide the URVI magnitudes as a convenience and compare them to the MSP91 values in the next subsection.

Transformations from F336W to the standard U-band are particularly problematic. One complication is the red leak in the F336W filter, which becomes more significant for redder stars. However, for the great majority of Wd2 stars in the range, −0.5 ≲ (F336W–F439W) ≲ 0.5, or (U − B) ≃ 0.3 (MSP91), this only amounts to ∼2–3%, as shown by Figure 3 of Holtzman et al. (1995) and confirmed by our own synthetic photometry of reddened early type stars. Accordingly, we correct all of the reported U and F336W magnitudes by +0.02 mag, acknowledging that this is only a zeroth-order correction appropriate to the mean Wd2 stars. More problematic is that the standard Johnson U band straddles the Balmer discontinuity while the F336W filter lies short ward of it. Holtzman et al. (1995) provides a discussion of difficulties in transforming the F336W magnitudes into standard U-band magnitudes, notes that the F336W zero points and U-band transformations are based on stars having (U − B) < −0.1 and (V − I) < 1.0, a regime that excludes the reddened O stars of Wd2. Accordingly, we urge caution in any interpretation of the transformed U-band values. Nevertheless, our U-band photometry agrees well with that of MSP91.

Figure 4 displays the positions of the 1136 detected stars in equatorial coordinates. The concentration of cluster stars located on the PC1 chip of the "long" exposure images and in the WF3 chip of the "short" exposure images, is the dominant feature of the cluster, a secondary concentration of stars about 45'' to the north is also apparent. Equally striking is the near-total absence of stars over a ∼25'' diameter area to the northeast of the Wd2 core. This corresponds to a minimum in the mid-IR [5.8] and [8.0] emission in Figure 1, suggesting the presence of an infrared dark cloud (IRDC) having extremely high extinction. Our examination of the CO molecular maps of Furukawa et al. (2009) reveals no obvious molecular clouds at this location, although the angular resolution of those data is 15, several times the size of the region of interest. The JHK_s infrared images of Ascenso et al. (2007) show no obvious dearth of stars at this location, lending credence to the hypothesis of a localized region of high extinction that affects the optical wavelength data most significantly.

Zoom In Zoom Out Reset image size

Figures 5–7 show three-color WFPC2 images of Wd2 with F439W in blue, F555W in green, and F814W in red. The star ID numbers on the figures denote stars of interest to this study according to the numeration of Table 2. The color scale reveals that the core cluster stars are red compared to other bright foreground field stars that appear white, such as MSP 158 (#436), confirmed as a foreground object by R07. The excellent angular resolution reveals that many of the bright core stars identified by MSP91 have close neighbors that are blended or unresolved at ground-based resolutions.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

2.2.2. Photometry Comparison

We compared the photometry obtained with HSTphot to that presented by MSP91 and R07 for the 75 and 11 stars in common with those works, respectively.⁶ The catalog provided by MSP91 does not contain positions, so it was necessary to do a visual match between our images and their finding chart. Since the MSP91 observations were performed with ground-based telescopes that do not possess the resolution of HST, some of the MSP stars are actually blends of several stars that are now resolved in our data. For such cases we choose the brightest star in the HST image to be the one that was originally identified by MSP91. In 63 of the 75 cases, either the MSP star is isolated and unambiguous, or we can identify the MSP source with the brightest of several close sources in our photometry. In 8 cases, we found two or more stars of roughly equal brightness within a radius of 2'', so the cross identifications with MSP91 are ambiguous.

Figures 8 and 9 show differences between our photometry and the works of MSP91 and R07 (Δm = m_HST − m_them) versus HST magnitudes transformed to the Johnson system. Numbered points indicate the nomenclature of MSP91 in both figures. Where significant photometric differences exist, they are seen in both the V and B bands. The median differences compared with MSP91 are 0.06, 0.19 and 0.15 mag, for the U, B, and V bands, respectively. Compared with R07 the median differences are 0.22 and 0.12 mag for the B and V bands, respectively. Both figures show that our measurements are systematically fainter than theirs in both the BV bands, particularly for the brightest stars which lie in the most crowded regions. Part of this effect can be explained as a result of ground-based photometry that inevitably includes close neighbors in the measurement of the central star. These neighbors are well-resolved and individually measured in the HST data.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To further investigate this systematic magnitude offset we performed a series of tests on the HST images. First, we verified that the images are processed using the latest calibration pipeline routine, which in this case is the OPUS version 2009_2k with calwp2 version 2.5.5 (2009 April 17).⁷ This represents the final and best calibration of WFPC2. Next, we performed aperture photometry using the Image Reduction and Analysis Facility (IRAF)⁸ package and compared it with the PSF results of HSTphot. To this photometry we applied the corresponding filter's zeropoint and Charge Transfer Efficiency (CTE) corrections according to Dolphin (2009) and subsequent updates from the WFPC2 Calibration and CTE Corrections Web site.⁹ We found excellent agreement between the results of the aperture and PSF photometry. Furthermore, the HST zeropoints are well calibrated and their uncertainties range between 0.02 and 0.04 mag, depending on the filter used (Heyer et al. 2004).

Another possible source of discrepancy with ground-based observations could be the color transformations applied to the HST flight system filters. The color transformation of Holtzman et al. (1995) was made using unreddened standard stars and stars from the cluster ω Cen rather than reddened O stars like those that dominate the population of Wd2. We used the IRAF/STSDAS synphot package to create synthetic photometry of reddened O stars and compare it with the color transformations of Holtzman et al. (1995). We found the transformations for reddened O stars to be compatible with the color transformations based on late-type stars, to within a few percent. In most cases the color transformations are indeed small and the HST's flight system filters can be used as a proxy for standard filters, given the Holtzman et al. (1995) transformations. For example, the differences between the F555W and the standard V-band are on the order of ∼0.01 mag for a wide range of stellar colors (see Figure 4 in Holtzman et al. 1995).

We also investigated the known long-short exposure difference (Casertano & Mutchler 1998) and the low background level (Golimowski & Biretta 2010) problems that affect the WFPC2 camera. Both effects are well-calibrated by HSTphot, and any residuals in the current CTE corrections are small (∼0.15 mmag row⁻¹ for images with low-to-moderate background signal of <50e⁻) and cannot account for the systematic difference between the ground-based and WFPC2 observations.

Others have also noticed discrepancies between HST and ground-based observations. For example, Dotter et al. (2008) reported offsets in the V and I bands of 0.068 and 0.007 mag, respectively. Turner (2012) also reports offsets with UBV photometry. From his Figure 2 we estimated an offset of ∼0.2 mag. At the moment we can only conclude that there is an offset of unknown origin between the values reported here and those of MSP91 and R07.

2.2.3. Spectral Reductions

Figure 10 shows the CMD for Wd2 in the HST's standard filters. The CMD is the F555W magnitude versus the (F439W–F555W) color. Detections with MSP91 counterparts, 81 stars, are identified by an asterisk (colored red in the online journal). Stars of relevance to this work are labeled using the numeration system of this work. The CMD clearly shows two distinct populations. On the CMD the blue "arm" represents the field stars, while the redder "arm" clearly shows the cluster's members. Two stars, MSP 91 and 158 (#832 and 436, respectively), fall on the field "arm" and are likely foreground objects, consistent with the conclusions of R07.

Zoom In Zoom Out Reset image size

We attempted to clean the CMD of field stars by using the 3D field star decontamination algorithm developed by Bonatto & Bica (2007) and then improved in Bica et al. (2008) and Bonatto & Bica (2009). Unfortunately, the limited WFPC2 field of view means that we lack off-source observations of field stars needed to make a proper field star decontamination. However, given the richness of the cluster core in the center of the small WFPC2 field, these data are dominated by cluster stars and, at the brightest magnitudes, suffer little field star contamination in the red arm.

The vast majority of the new spectra for Wd2 exhibit lines of neutral and/or ionized He, indicative of O or early-B-type stars. Four sources exhibit Hα in emission. Figure 11 shows the normalized spectra for early-type stars labeled according to the IDs of this work and organized by decreasing value of temperature/spectral type from top to bottom. Figure 12 shows the spectra for stars having spectral type consistent with late-O/early-B, undetermined, or field stars. The main stellar absorption lines, interstellar lines (mainly Na i D lines λλ5889, 5895), and Diffuse Interstellar Bands (DIB) are labeled. We classified stars having new spectra using the ratio of the equivalent width (EW) of He ii (λ5411) to the EW of He i (λ5876) using the diagram and analytic fit from Kobulnicky et al. (2012):

$$\begin{eqnarray} \frac{{\rm EW}(\lambda 5411)}{{\rm EW}(\lambda 5876)} &=& 1.16208 \times 10^{-12} T^{3} - 1.19205 \times 10^{-7} T^{2} \nonumber\\ && + 4.22137 \times 10^{-3} T - 50.5093. \end{eqnarray} \tag{ 1 }$$

This calibration is based on model spectra from the Lanz & Hubeny (2003) Tlusty models and hot stars from the Jacoby et al. (1984) spectral atlas. Because we have not observed spectral standards using our spectral setup, this may introduce additional uncertainty into our procedure for placing our targets on the MK system. We used the robust curve fitting package MPFIT (Markwardt 2008) as implemented in IDL to measure the EW of the He ii λ5411 and He i λ5876 lines and their uncertainties. To improve the robustness of the fit for weak lines, we constrained the Gaussian width to that measured for the stronger of the two lines. Table 3 list the stars having spectral types determined using the above procedure. Columns 1 and 2 are the stars' identification numbers according to this work and MPS91, where available. Columns 3 and 4 are the photometry, V mag and B − V color, for such stars. Columns 5 and 6 are the measured EW for He ii and He i lines having detections above 2σ; negative values correspond to emission lines, and upper limits are indicated. Column 7 is the ratio of the He ii/He i EW and Column 8 is the stellar effective temperature corresponding to the ratio as determined by the relation in Equation (1). Column 9 is the spectral type determined from the temperature approximation to one subtype (Martins et al. 2005). Because our SOAR/IMACS data do not cover the blue portion of the optical spectrum typically used for classification, we are not able to constrain the luminosity class, thus we adopt a default of dwarf (V). This is likely to be correct in most cases, since our targets are not among the brightest cluster members, assigned V or III by R07. A constraint on the luminosity is obtained when analyzing the MIKE spectra for stars #714, and #738. The lack of N iii λλ4634, 4640, 4662 in emission in spectra of Figures 15 and 16 confirms that these stars are not evolved. Remarkably, there are no known supergiants in the cluster, consistent with it being very young.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

#137 (MSP 165). We tentatively classify this star as O4 on the basis of the relatively large He ii EW (0.97 ± 0.09 Å) and relatively small He i EW (0.30  ±  0.07 Å) in the SOAR spectrum (S/N = 50). This ratio indicates a very hot star with ∼43,000 K, very close to the maximum range calibrated in the Kobulnicky et al. (2012) diagnostic diagram. However, the detection of He i precludes classification as a true O3 given the Walborn (1971) criteria requiring its absence. This star is among the brightest in the sample at V = 15.5 and also one of the reddest with B − V = 1.66. It does not have detectable Hα emission. Its location on the cluster outskirts near a region of diffuse IR emission (see Figures 1 and 6) is consistent with higher dust extinction at a location that has not yet been fully evacuated by the powerful winds of the central cluster.

#260 (MSP 201). Located on the cluster outskirts not far from #137 (see Figure 6), we classify as a late O or early B because of the lack of a He ii line from our SOAR spectrum (S/N = 21). Hα appears in emission, along with [O iii] λλ4959, 5007, despite our best attempts at background subtraction, indicating the presence of small-scale variations in nebular emission near the star.

#463 (MSP 125). This star, together with #483 at 15 to the E, constitute MSP 125. Our SOAR spectrum of this star (S/N = 11) has insufficient S/N for spectral classification. With (B − V) = 0.87, this star is unusually blue for a cluster member.

#505 (MSP 196). This particular spectrum from SOAR (S/N = 27) is dominated by light from #505 but may contain contributions from #495 and #502 located within about 1''. Hence, MSP 196 is a blend of at least three stars. The dominant component, #505, has (B − V) = 1.3 and spectral type O8.5, consistent with cluster membership.

#528 (MSP 229). This source and its companion #523, separated by ∼04, constitute MSP 229. Our two spectra of this object, one from Magellan (S/N = 42) and one from SOAR (S/N = 53), yield O8.5 and O8, respectively, so we adopt O8 as a weighted mean of the brighter component which dominates the spectrum.

#547 (MSP 28). This V = 17.4 source has [O iii] emission and Hα emission. He ii is a non-detection, and He i is marginally detected in our S/N = 22 SOAR spectrum, making this a late-O or early-B star. The position of this star is interesting since it is located near the secondary cluster, centered on MSP 18, and it lies upon a ring of diffuse mid-IR emission which surrounds the secondary cluster. Figure 13 is a Spitzer three-color image of the region surrounding MSP 18 showing this ∼36'' diameter ring-like structure, which is present in all IRAC bands but best seen at [8.0], consistent with PAH emission from a photo-dissociation region. We believe that this is the first recognition of this feature, which possibly demarcates the boundary of the H ii region surrounding #889 (MSP 18), although #889 does not lie at its geometric center. The ring is similar to ones found by Gvaramadze et al. (2010) and Wachter et al. (2010) and indicative that star #889 might be a luminous blue variable (LBV).

Zoom In Zoom Out Reset image size

#548 (MSP 151). Although this star is located ∼13'' NW of the cluster core in an area that may be affected by small-scale dust extinction. Observed with SOAR (S/N = 80), a spectral type of O4 was determined. This classification is at odds with MSP91 and R07 classification of O6–7III. The He ii is very strong with EW =0.94  ±  0.04 Å and He i is rather weak with EW =0.31  ±  0.06 Å. Nebular lines are well subtracted, so we do not have an explanation for the earlier spectral type compared to R07.

#549 (MSP 44). Star #549, along with the faint companion #538 constitute the source MSP 44. Indeed, Rauw et al. (2011, hereafter R11) identify this system as a P = 5.176 day eclipsing binary (ΔV = 0.6 mag) composed of an B1V + pre-main-sequence star, or possibly even a triple system. We obtained two spectra for this source, one as target (S/N = 59), and the second being serendipitous (S/N = 28). Both observations, obtained with the SOAR telescope, yield consistent spectra where the He lines are very weak or non-detections. For this star we also detected emission at [O iii] and Hα. The presence of emission lines may indicate a localized knot of nebular emission. The residual He i emission may affect the stellar EW measurement, rendering stellar classification unreliable, but consistent with a very late-O or early-B star, in agreement with R11. R11 reports nebular emission associated with MSP 18, MSP 223, and several other stars. Evidently, the nebular structure in the Wd2 region is complex both spatially and, as indicated by R11, in velocity as well. Like #547, which also exhibits nebular emission, #549 lies near the ring of mid-IR emission encircling MSP 18. Mid-IR PAH emission usually traces the UV-heated surfaces of molecular clouds. Such a location could indicate that the MSP 44 system is very young, consistent with the interpretation of R11 that the secondary star is a pre-main-sequence object.

#556. Located ∼66'' to the north of the cluster core, #556 (V = 17.1) seems to be surrounded by diffuse IR emission, very likely coming from the ring surrounding MSP 18. He i is detected in our S/N = 23 SOAR spectrum with an EW of 0.53  ±  0.11 Å, but He ii is not detected, yielding only an upper limit on the spectral type of late O or early B.

#584 (MSP 157b). Located at ∼10'' from the cluster core where blending becomes problematic, MSP 157 contains contributions from as many as seven stars, the brightest three of which are stars #597, #584, and #568 located within ∼12 radius. Our Magellan spectrum (S/N = 58) and classification implies an O8. This is "slightly" later type than the O5.5–6.5V by R07, but their spectrum is likely dominated by the brighter component, MSP 157a (#597).

#620 (MSP 96). Located in a relatively isolated region at ∼26'' to the north of the main cluster, #620 along with #663, constitute MSP 96, and its photometry is consistent with the measurement from MSP91. Both He lines are non-detections in our S/N = 43 SOAR spectrum. This limits our classification to a very late O or early B star.

#640 (MSP 233). This V = 16.2 source is relatively isolated and lies close to diffuse IR emission at ∼14'' to the SW of the main cluster. The clear detections of the He lines in our Magellan spectrum (S/N = 36) provide a classification of about O9.5.

#664 (MSP 188). This bright star has two faint (ΔV = 4.3 mag) neighbors at ∼15. The very strong He ii and the weak He i in our S/N = 91 Magellan spectrum yield an EW ratio of the He lines of 8.99  ±  4.20, placing this star at the upper limit of our classification scheme; therefore, we classify it as an O4 or earlier. This is in agreement with R07 who revised the MSP91 spectral type to O4. As portrayed in the velocity plot of the He ii and He i lines in Figure 14 the spectrum shows that both He lines are double-peaked, with He i showing two approximately equal-depth absorption lines, and He ii displaying different strength absorption lines. We measure the velocity difference between the peaks for He i to be Δv≊360.5  ±  103. km s⁻¹ and for He ii to be Δv≊489.6  ±  36.8 km s⁻¹. The He lines were deblended to calculate individual EW and EW ratios, so as to determine the spectral types of what we infer to be a binary system. The two components have EW ratios of 5.70  ±  3.32 and 2.04  ±  1.18, implying spectral types of at least O4 and O5.5, respectively. The binary nature of this system partially explains why #664 is so luminous compared to other very early members of Wd2.

Zoom In Zoom Out Reset image size

#704 (MSP 175). Star #704, along with the faint and close neighbor #713 (lying ∼044 away, midway between #704 and #714), forms MSP 175. It shows very strong He ii in our S/N = 47 Magellan spectrum, while He i is very weak (EW =0.15  ±  0.05), implying a spectral type near O4. This is slightly earlier than the O5V–O6V type assigned by R07. R07 notes that the He lines in MSP 175 are broad making classification uncertain; He ii lines appear asymmetric with a blue wing in Figure 3 of R07. Our Magellan spectrum shows asymmetry with a red wing in both the He ii and He i lines, suggesting a possible binary system.

#714. Star #714, located at ∼3'' to the W of the main cluster core, is in a region of heavy crowding. Stars #664 (MSP 188), and #704 (MSP 175) lie in a crowded region near the cluster core within 15 of several other stars, including #714 (no MSP identification) and #713. Figure 15 shows the MIKE spectrum of #714 (S/N = 91), smoothed to a resolution of about 1 Å with key spectral features of hot stars labeled. This spectrum is consistent with the lower resolution IMACS spectrum in that He i λ4471 is extremely weak (not detected here), while the He ii lines are very strong, indicating an extremely hot star. By comparison with the O-star atlases of Walborn & Fitzpatrick (1990) and Walborn et al. (2002) we assign this star an O3 spectral type and dwarf (V) luminosity class. Signatures of O2 spectral types, such as N v λλ4604, 4620 absorption or strong emission lines of N iv λ4058 and N iii λλ4634, 4640, 4642 are not seen, although these latter N features are weakly present. The lack of He emission lines and lack of strong N emission lines further supports a dwarf luminosity classification. Nebular [O iii] λλ4959, 5007 and Hβ lines also appear in this spectrum, but we consider these to originate from the diffuse ionized gas that pervades the RCW49 H ii region; these emission lines remain after background subtraction, presumably because of small-scale variations in the nebular emission along the slit. In our single spectrum we find no evidence of binarity. However, all of the principle spectral stellar features in #714 are redshifted by about 1 Å (67 km s⁻¹) compared to expected rest wavelengths, and compared to molecular gas associated with the cluster which spans the range −11 to 11 km s⁻¹ LSR (Ohama et al. 2010). This redshift may constitute evidence of being a single-lined spectroscopic binary or may indicate that #714 has a peculiar velocity sufficient to unbind it from the cluster.

Zoom In Zoom Out Reset image size

#738 (MSP 168). Figure 16 shows our single MIKE spectrum of #738 (V = 14.9) boxcar smoothed to about 1 Å resolution (S/N = 48). He i λ4471, with an equivalent width of 0.29 Å, appears in this spectrum, making #738 later than O3. The EW ratio EW(λ4686)/EW(λ4471) is 1.86, similar to O5–O6 stars from the Walborn & Fitzpatrick (1990) atlas. The radial velocity, as judged from the mean of key spectral features, is consistent with zero km s⁻¹ LSR, making it consistent with the molecular gas surrounding the cluster (Ohama et al. 2010). We adopt a classification of O5.5.

Zoom In Zoom Out Reset image size

#769 (MSP 219). This star is located at ∼9'' to the S of the main cluster core in a relatively isolated area. He ii is a marginal detection in the Magellan spectrum (S/N = 31) but He i is strong with EW =0.85  ±  0.05. This suggests that #769 is O9.5 or later.

#771 (MSP 167). MSP 167 is a blend of the stars #804, #771, #765, #777, and #772. A broad He ii line and a well-defined He i line produced an EW ratio of 1.00  ±  0.13, implying that this star is O8.

#857 (MSP 444). MSP 444 is one of those hard-to-identify stars owing to its position in the cluster core and its close proximity to #843 (MSP 203; ∼06). With HST MSP 444 was resolved into three components, #857  and #895, and #824. The Magellan spectrum (S/N = 78) of #857 possesses a strong and well-defined He ii line while the He i is broader and weaker than He ii, implying a hot star of type O4.5, consistent with O4–O5 by R11. With (B − V) = 1.4, its color and magnitude are about right to be an early-type cluster member.

#878. Located at ∼8'' to the SW at the outskirts of the main cluster, the two Magellan spectra obtained for this faint star (V = 16.6) have insufficient S/N (S/N = 15) to perform a reliable classification. He ii is not detected, and He i is weak, hinting at a late O or early B.

#879 (MSP 235). This star is in relative isolation at ∼16'' to the SE of the main cluster. The Magellan observations, with a S/N = 36, show strong He i but minimal He ii, implying a spectral type of O9.5.

#896 and #903 (MSP 183 subcomponents). These stars are located on the outskirts of the main cluster core where heavy blending in ground-based observations is inevitable. Within a radius of ∼15 there are seven stars that could contribute to the brightness of MSP 183, namely, #869, #896, #903, #826, #880, #887, and #856. These two stars, separated by ∼15, became partially blended on the longslit observations with the Magellan Telescope (S/N = 43 for #896 and S/N = 39 for #903). The strong He lines in both stars resulted in an EW ratio of 0.83  ±  0.13 for #896 and 0.90  ±  0.14 for #903, implying an O8.5 type for both.

#1004 (MSP 32), #1012 (MSP 24), and #1026 (MSP 20). This stellar trio, observed spectroscopically with SOAR, is located on the diffuse mid-IR ring that surrounds the secondary cluster, at about ∼12'' to the W of MSP 18. Star #1004's spectrum (S/N = 53) possesses strong He lines, although He ii is noisy and broad, implying a classification of O9.5. The spectra of both #1012 (S/N = 36) and #1026 (S/N = 14) display [O iii] and Hα in emission and have insufficient S/N for reliable classification.

#1028. With V = 14.3, and (B − V) = 0.37, this star is clearly a field star in the vicinity of the cluster core. The serendipitous SOAR spectrum (S/N = 62) shows only strong Balmer lines with EW(Hα) = 7.67 Å, making it a probable A star.

#1031. Greatly isolated from the cluster core at ∼73'' to the SE it lies projected against a bright mid-IR rim illuminated by the central cluster. The S/N = 32 SOAR spectrum is heavily contaminated by emission lines, and no spectral type determination is possible. Interestingly, as shown in Figure 12, #1031 shows Mg ii λ5173 Å absorption lines.

We measured the extinction, A_V, and the ratio of selective to total extinction R_V = A_V/E(B − V), for each star by individually fitting a spectral energy distribution (SED) based on the known spectral type/temperature of the star and the theoretical absolute magnitudes intrinsic colors of the Padova stellar evolution tracks and isochrones (CMD v.2.2;¹⁰ Girardi et al. 2000; Marigo et al. 2008; Girardi et al. 2002). Each SED was reddened by each of three reddening laws (Cardelli et al. 1989; Fitzpatrick 2004; Fitzpatrick & Massa 2007, hereafter CCM89, F04, and FM07, respectively) and compared, after appropriate normalization, with the HST¹¹ and Ascenso et al. (2007) JHK_s photometry to find a global chi-squared minimum over all plausible values for R_V, and A_V. We note here that, while CCM89 provide an analytical approximation of the mean interstellar extinction with R_V and A_V as free parameters, F04 tabulates 38 reddening parameterizations as a function of R_V, and FM07 tabulates parameterizations of 328 sightlines, creating 328 discrete reddening laws. As an additional method to estimate R_V we also used the method of Fitzpatrick & Massa (hereafter FM09 2009) which is based on an average relation between BVK_s photometry in their Figure 7. The fitted methods are sightly dependent on the anchor point used to normalize the Padova magnitudes to the data. To eliminate this dependency the fitting was done using each of the seven photometric values, when available, as anchors points and the resultant values of R_V, A_V and A_λ are weighted averages of all seven fits. Figure 17 shows examples of the best fits for stars #137, 528, 704, and 1039. Each subplot displays the HST photometry as squares and the Ascenso et al. (2007) IR photometry as circles. The lines are the best fit to the photometry after reddening the predicted magnitudes for the appropriate star's spectral type. For comparison purposes the best fit for each method was included: the solid lines is the CCM89, the dash line is the F04, and the dotted line is the FM07 reddening laws (each one is colored in blue, cyan, and magenta in the online journal).

Zoom In Zoom Out Reset image size

Table 4 compiles the fit results and shows that best-fitting values range between 5.71 < A_V < 7.54 and 3.44 < R_V < 4.31. Column 1 lists the star's ID according to this work, Column 2 lists the adopted spectral type, Columns 3–6 are the best-fitting R_V using the methods of CCM89, F04, FM07, and FM09, respectively, and Columns 7–10 are the best-fitting A_V values. The results indicate that R_V is significantly higher than the mean interstellar value of R_V = 3.1, meaning that the obscuring dust produces more extinction for a given amount of reddening, often interpreted as a result of a population of larger "gray" dust grains. On the other hand, FM07 and FM09 argue that there is no universal reddening law or extinction curves and that extinction cannot be parameterized in terms of a single variable such as R_V. The so-called "anomalous" dust would, in reality, be "normal" for the Wd2 region and may be different from any other dust in the Galaxy.

Stars #664, 869, 889 are not included in Table 4 since these only have photometry in three bands (F336W, F439W, F555W) because of saturation or blending. We regard that the adopted fit procedure provides unreliable results and we adopt, for these stars, the average R_V and A_V from the 29 other stars. Excluding these three stars, the average values for the cluster are 〈R_V〉 = 3.77  ±  0.09 and 〈A_V〉 = 6.51  ±  0.38 mag where the uncertainties are the dispersions of the sample. We conclude that the mean R_V is substantially larger than mean interstellar values, and that all stars are consistent, within the uncertainties, of this mean. On the other hand, there appears to be a significant scatter in A_V, consistent with appreciable internal extinction variations.

Table 5 compiles the corresponding A_λ for the HST filters corresponding to the best model fit in Table 4. Column 1 lists the star's ID according to this work, Column 2 lists the adopted spectral type, Columns 3, 4, and 5 are the A_F336W values for CCM89, F04 and FM07, and Column 6 contains the weighted average of Columns 3, 4, and 5. The same pattern is followed by Columns 5 through 18 for filters F439W, F555W, and F814W. We use the values of Table 5 to deredden the listed stars in the HST filter system and correct for the heavy, patchy extinction toward Wd2 which produces considerable dispersion in the raw CMD (Figure 10).

The (F336W–F439W) versus (F439W–F555W) color–color diagram (CCD) in Figure 18 shows a portion of a reddened zero-age main sequence (ZAMS) Padova isochrone¹² using the global A_λ averages from Tables 4 and 5 (R_V = 3.77 and A_V = 6.51 with corresponding color excess of E(F336W–F439W) = 1.47 and E(F439W–F555W) = 1.81). The dashed line (magenta in the online journal) is part of the reddening vector connecting the unreddened ZAMS O stars and the reddened Wd2 stars. Asterisks (red in the online journal) represent Wd2 stars having spectral types that we used to obtain the reddening parameters. The reddened isochrone provides a good match to the cluster stars, given that these stars exhibit a dispersion that is parallel to the vector, indicating substantial differential reddening and consistency with the mean reddening parameters found for this sightline. Additionally, the dotted curve (cyan in the online journal), is a reddening track that takes into account small systematic color trends due to the broadband nature of the HST filters and the large reddening toward Wd2. The curve was constructed by performing synthetic photometry with the software package synphot using a CCM89 reddening law. As pointed out by Schmidt-Kaler (1981) and Turner (1989), the effective wavelength of a filter changes as a function of spectral type and reddening, leading to curved reddening tracks. Along this curve we label points that correspond to particular extinction values A_V. According to Turner (1989) the difference between a linear and a curved reddening track is ∼2% per E(B − V). Although this is a small factor we included it in Figure 18 for illustrative purposes. When compared to the random and systematic errors of the intrinsic and observed colors that are discussed previously and in the next section, this difference is small so we adopt the straight vector in our reddening analysis. From the figure it is evident that there exists a systemic offset between the linear vector and the curved track, but this is small when compared to the dispersion in the photometry. This offset can be explained by the method we used in the creation of the linear vector. This vector was created using the average A_λ from the reddening laws of CCM89, F04, and FM07. Such values do not necessary have to reproduce the curve created by synphot that only uses the CCM89 reddening law.

Zoom In Zoom Out Reset image size

Support for the "anomalous" value of R_V toward Wd2 can be found in the studies of other open clusters and their Cepheid stars in the vicinity of the Carina arm. The open cluster Ruprecht 91, Collinder 236, and Shorlin 1; with projected distances of ∼32, ∼56, and ∼64, respectively, away from Wd2, also have different values from the nominal R_V = 3.1; Ruprecht 91 with R_V = 3.82  ±  0.13 (Turner et al. 2005), Collinder 236 with R_V = 3.82  ±  0.13 (the R_V was an adopted value because of similar reddening law with Ruprecht 91; Turner et al. 2009), and Shorlin 1 with R_V = 4.0  ±  0.1 (Turner 2012). Furthermore Turner (2012) made a compilation showing that various clusters in the Carina complex have values greater than 3.1 in general (see his Table 1).

We adopt the Padova isochrones to determine the cluster's distance modulus (DM). Isochrones with ages between 1 and 4 Myr and solar metallicity (Z = 0.019) were chosen to match the expected properties of Wd2. The Padova database provides absolute magnitudes (and therefore intrinsic colors) for stars in the HST filter system which we use to compute a spectroscopic and photometric distance for each star, following the prescription used by Hanson (2003) in finding the distance to Cygnus OB2. We adopt either our own spectral classifications or those of R07 and R11 (based on the Walborn & Fitzpatrick (1990) calibration) in order to calculate the appropriate M_F555W and (F439W–F555W)₀ values and their associated uncertainties by interpolating the Padova data tables. For stars classified using our procedure described above, we estimate absolute magnitude and color uncertainties by propagating errors from our determination of the effective temperature. For stars classified by R07, we adopt temperature uncertainties given in that work. We deredden each star according to the values in Table 5 and then solve for the distance modulus of each.

Table 6 displays the results of our distance analysis. The columns list (1) the star ID numbers according to this work, (2) the MSP91 identification numbers, (3) the star's spectral type, (4) the observed apparent F555W magnitude, (5) the observed (F439W–F555W) color, (6) the absolute F555W magnitude, (7) the intrinsic (F439W–F555W) color, (8) the extinction A_F555W, (9) the distance modulus, and (10) the distance, in kiloparsecs, obtained for each star. The weighted mean are 〈DM〉 = 12.98  ±  0.03, and 〈d〉 = 3.85  ±  0.06 kpc (the simple mean yields 〈d〉 = 4.20  ±  0.06 kpc). This weighted mean is taking into account uncertainties in the spectral/luminosity classification. We exclude from this calculation stars #664, 869, 889, which, as stated in the previous section, lack complete photometry. This distance is substantially smaller than the estimate that would result if we were to assume the canonical R_V = 3.1. If we were to assume such, along with 〈A_V〉 = 5.16 (calculated from the A_V values in R07), then the resulting distance modulus becomes 〈DM〉 = 14.246  ±  0.038 and 〈d〉 = 7.07  ±  0.12 kpc, about twice the distance than when using 〈R_V〉 = 3.77.

Figure 19 shows the extinction- and distance-corrected Wd2 CMD in the HST filter system, illustrating the application of the results from Table 6, along with ZAMS, 1–4, and 10 Myr isochrones. Each star was individually dereddened, extinction corrected, and then shifted by the weighted mean DM. For easier comparison, key spectral types are labeled at the right edge of the figure. Each star is labeled with its ID and spectral type. The dispersion around the ZAMS is minimal, and very few stars have an observed spectral type inconsistent with the spectral type expected along the ZAMS isochrone. Star #889 (MSP 18) is not shown in the figure because of its odd position in the CMD after adopting the average A_λ (overcorrection in color). Since this star lies to the north of the main cluster, it is very likely that the average reddening determined for the main cluster is not applicable to this star.

Zoom In Zoom Out Reset image size

The presence of close, unresolved binaries in a stellar population produces brighter apparent magnitudes than if no binaries were present. Consequently, the derived distances are systematically low compared to the single-star case. Kiminki & Kobulnicky (2012) modeled the light contribution from secondary components among massive binaries in Cygnus OB2 and found that they contribute 16% of the luminosity, on average. This translates into a systematic distance error of 8%, in the sense that the derived distances are too small unless binaries are taken into account. We apply this statistical correction to find a distance of 4.16  ±  0.07 kpc.

A possible source of systematic error in our distance measurement is the intrinsic colors adopted for massive stars. There is a difference in the intrinsic colors of stars between the Padova isochrones and Martins & Plez (2006) colors. The Martins & Plez (2006) (B − V)₀ colors are, on average, 0.04 mag redder than those that result from the Kurucz models or the photometric measurements of Johnson (1966) and Fitzgerald (1970). This difference implies a smaller color excess and, therefore, a smaller extinction and larger distance modulus. Martins & Plez (2006) warn that a redder intrinsic color introduces a difference of ΔA_V ∼ 0.124 mag assuming R_V = 3.1. For our mean value of R_V = 3.77, the change in extinction required is ΔA_V = 0.15, equivalent to a systematic distance error of +0.26 kpc. We conclude that our measurement of the Wd2 distance is 4.16  ±  0.06 (random) +0.26 (systematic) kpc.

The distance that we determined favors a small value, like some in the literature, but not as small as the 2.8 kpc suggested by Ascenso et al. (2007). Ascenso et al. (2007) measured a distance by fitting deep JHK_s CMDs to pre-main-sequence isochrones after correcting for reddening, assuming a single value for A_V derived from the four brightest cluster stars. Despite adopting an A_V smaller than our preferred value, they obtain a smaller distance, which we attribute to uncertainties in the pre-main-sequence models and the inherent ambiguities of fitting such models to the data. Furthermore, the distance calculation is based on a predetermined age for Wd2, which has not yet been properly constrained. Our measurement, like that of Uzpen et al. (2005), falls in the middle of the d = 2–5 kpc constraints determined in the X-ray study of pre-main-sequence stars in Wd2 (Tsujimoto et al. 2007). Our result is also very close to the 4.2 kpc kinematic distance adopted by Churchwell et al. (2004) on the basis of 21 cm absorption arguments (Caswell & Haynes 1987). Other kinematic estimates from the CO radial velocity of the associated molecular cloud (Dame 2007; Furukawa et al. 2009) are slightly larger, 5–6 kpc—still broadly consistent with our result but with larger uncertainties owing to the range of molecular cloud velocities along this sightline. However, our measurement is inconsistent with distances larger than 5 kpc. The main reason why our preferred distance is smaller than that found by Piatti et al. (1998), Carraro & Munari (2004), Rauw et al. (2005), R07, and R11 is that all of these works adopted a standard R_V = 3.1 which leads to a substantially larger distance. Our adopted distance of 4.16 kpc places Wd2 comfortably within the Carina spiral arm and near the solar circle, about 40% further away than where it is pictured in Figure 3 of the review of massive stellar clusters by Portegies Zwart et al. (2010).

Due to the discrepancy between our transformed UBVI photometry and that reported via ground-based observations, as attested in Section 2.2.2, we consider here what would happen if we brighten our flight system photometry by 0.2 mag to achieve better agreement with those works. When such is the case, we find that Wd2 has a slightly smaller reddening, 〈R_V〉 = 3.60  ±  0.10 and 〈A_V〉 = 6.23  ±  0.39. This smaller A_V more than offsets the 0.2 mag photometric correction and yields 〈DM〉 = 13.26  ±  0.03, and 〈d〉 = 4.38  ±  0.07 kpc. After corrections for binary stars this becomes 4.73  ±  0.08 (random) +0.26 (systematic) kpc. In summary, the photometric offset compared to other works has only a small impact on the final distance determination. This alternative distance is still much smaller than the "far" distances in the literature, substantially larger than some of the "near" distances, and still broadly consistent with the kinematic distance estimates.

In Figure 19 we used the values of Table 5 to deredden all the available photometry from Figure 10, and we compared the data with 1–4 Myr isochrones in order to constrain Wd2's age. From the inspection it becomes evident that isochrones with ages 1–3 Myr are degenerate, and it can be "safely" assumed the majority of the stars are dwarf class. The 4 Myr isochrone shows that there should be a population of evolved giant stars that is absent in the data. At the most we can conclude that Wd2 has an age between 1 and 3 Myr. Other authors have determined the same age constraint on the basis of integrated spectra, CMD fitting, and the lack of evolved stars within the cluster (Piatti et al. 1998; Ascenso et al. 2007).

We attempt here a different kind of age constraint using the size of the cavity carved in the ISM by the winds of massive stars (Churchwell et al. 2004). Weaver et al. (1977) derives (Equation (51)) the radius of an interstellar bubble as a function of time t₆ in Myr, ambient density n₀, and mechanical energy injection L₃₆, in units of 10³⁶ erg s⁻¹. We invert this expression to find

$$\begin{equation} t_{6} = \left[ \left(\frac{R_{2}({\rm pc})}{27} \right)^{5} \frac{n_{0}}{ L_{36}} \right]^{1/3} \ {\rm Myr}. \end{equation} \tag{ 2 }$$

From the angular size of the cavity in Figure 1 (see also Churchwell et al. 2004) we adopt a radius of 105'' or 2.0 pc at the adopted distance. We assume n₀ = 10⁵ cm⁻³, typical of molecular clouds. Based on the wind speeds and (highly uncertain) mass loss rates for O stars from Mokiem et al. (2007) we estimate a time-averaged energy injection rate of L₃₆ = 1.0 from the ensemble of massive stars. The mechanical wind luminosity is dominated by the few most massive stars, so the exact number and type of stars later than about O5 is inconsequential. The resulting expansion timescale of the (still ill-defined) interstellar cavity is ∼0.6 Myr, indicating extreme youth, consistent with the abundance of molecular material just outside the cluster core. We caution that the Weaver et al. (1977) analysis was developed for interstellar bubbles in uniform media and densities more typical of the diffuse ISM. The presence of a radio "blister" structure on the west side of the Wd2 H ii region (Whiteoak & Gardner 1977) suggests that the uniform density assumption is not valid and that the bubble may be expanding asymmetrically into a region of lower density on that side. The derived expansion timescale should be considered a lower limit.