A TREASURY STUDY OF STAR-FORMING REGIONS IN THE LOCAL GROUP. I. HST PHOTOMETRY OF YOUNG POPULATIONS IN SIX DWARF GALAXIES

HSTphot offers two different options for background subtraction: local background estimation (option 2) and use of a sky background image ("re-fitted sky," option 512). We compared the PSF photometry obtained with both options. They compare well if the default aperture corrections are used. We computed individual aperture corrections with HSTphot: they are comparable to the default values when we use the "re-fitted sky" option, but are very different when we use the local background option. Magnitudes computed with local background estimate and individual aperture corrections are up to 0.25 mag (depending on the detector and image) fainter than magnitudes computed using local background estimate and default aperture corrections, or re-fitted sky background subtraction with either default or individual aperture corrections. The effect is due to source crowding in most of our fields. Using the re-fitted sky option for background subtraction, we compared the aperture corrections computed individually for every chip and every filter by HSTphot, with the default (average) values. We find that the highest deviations, up to 0.05 mag, occur in the best exposed filters (F555W and F439W) while for shorter-wavelength filters, where the sources are much more sparse, the agreement is excellent (⩽0.01 mag in most cases). Moreover, the highest differences are seen in the WF3 chip, which contains the largest part of our target SF regions (as a consequence of how we centered the targets), therefore is much more populated of sources, and the smallest differences (almost zero) are found in the PC chip, which is the least populated. We conclude that in some fields, and especially for WF3 F555W and F439W images, crowding affects the choice of isolated stars by the HSTphot package to derive individual aperture correction. We have experienced this problem previously when analyzing HST imaging in crowded stellar fields from earlier programs. Because the problem does not subsist for shorter-wavelength filters where stars are more sparse, the error in aperture correction, however small (⩽0.05 mag), would introduce a small color bias for crowded field chips if we used individual aperture correction values. Therefore, for homogeneous robustness across the whole catalog, we adopted the default aperture correction values. We discuss here at length these additional systematic uncertainties, because in the final catalog we report only the formal photometry error for each source, as is customary. Finally, a potential additional source of systematic uncertainty is the CTE correction.

In order to evaluate which are the best HSTphot options to use for our photometry of crowded stellar associations and to assess uncertainties in the CTE corrections, we compared photometry of sources in overlapping portions of some fields. Such overlaps are few, since our goal was to maximize the imaging coverage of SF regions and to avoid repetitions. M31 and M33 photometry will be presented in a forthcoming paper, but all of the tests on photometry accuracy are reported here. There is some overlap between fields 1, 2 and, 3 in WLM, between fields 49 and D8 in M31, and between fields 18 and 20 in M33. In addition, visit D8 in M31 is a repetition of visit 48, which failed during reacquisition and therefore the second orbit worth of observations was lost, but exposures in F336W, F439W, and F555W are useable. Since their orientation is similar to the make-up visit D8, the comparison between the two provides a check for repeatability with the same detector, while the other overlaps provide a test for consistency among different detectors. Most comparisons are limited by statistical significance, and the agreement is much better when a larger number of overlapping stars is available. Stars with repeated measurements in WLM are few: from 21(F336W) to 43/64 (F555W/F814W) stars overlap between field1-WF3 and field3-WF4, and 17 to 57 stars (F336W to F814W) are in common between field2-WF4 and field3-WF2 (plus a few WF3 sources), when we limit the sample to sources with photometric error better than 0.1 mag and "sharpness" and "type" restrictions as described below. The comparison shows that magnitudes from both chips of field 3 are brighter than in field 1 or field 2, the mean difference being 0.1/0.08 mag in F555W/F814W (σ = 0.15/0.12 mag) for WF4, but almost zero for WF3 in these filters, and being larger for bluer filters than for longer-wavelength filters, probably just because of the much poorer statistics at shorter wavelengths. All differences are less than the standard deviations. More matched stars are available if we do not apply restrictions of magnitude errors, but the larger errors then blur the comparisons. Photometry from visit 18-WF3 and visit 20-WF2 in M33 (38 to 98 stars, F336W to F814W) differs on average by 0.045 mag (F336W and F439W, σ = 0.16 mag) or 0.006 mag (F555W, σ = 0.1 mag), visit 18-WF3 being fainter, but again not significantly. Comparison between visit 18-WF4 and visit 20-WF2 shows again the first to be fainter by 0.187/0.06/0.04/0.03 mag (F336W/F439W/F555W/F814W, with 11/16/24/50 overlapping stars, respectively), all differences being much smaller than the standard deviations and therefore not significant. Comparison of M31 visit D8-WF2 or WF3 with visit 49-WF4 indicates D8 measurements to be fainter by 0.021–0.027 mag in all filters (with standard deviations between 0.12 and 0.3 mag, and a few hundred stars overlap). Finally, for M31 repeated observations visit D8 and visit 48 where we compare stars reobserved in the same chip, the agreement is within 0.015 mag for all repeated filters in WF3 and WF2 detectors (a few hundred stars overlap in each), but slightly worse, about 0.04 mag, for the WF4 detector, but again all differences are well within 1σ. The difference is again in the sense of fainter magnitudes from visit D8. If we took these differences at face value, photometric stability seems worse in WF4 than in WF3 and WF2 (by about 0.02 mag), but other possible systematic small offsets may occur from visit to visit, of the order of ≳0.02 mag, that we suspect may arise from CTE corrections since they are detector- and filter-independent. Comparisons with PC measurements are worse, as is often found to be the case, but stars are also very few in this higher resolution detector. The comparison figures given above are for HSTphot option "520," i.e., re-fitted sky background estimate and default aperture correction values, which we adopted for our catalog. Using other options, such as the aperture corrections for individual chips, the consistency is only slightly worse. In summary, our data do not provide any significant evidence for systematic differences and do not add information to the thorough tests by the WFPC2 team cited above, since we particularly avoided repetitions; these tests only show that repeatability is within the significance of the measurements, and worse in WF4 than in other chips.

We estimated the completeness of our photometry by using the fake_image/HSTphot routine to create artificial images on which we performed detection and photometry running HSTphot with the same parameters as for our photometry. The incompleteness reaches 20% at 21.0, 22.8, 22.9, and 22.0 mag for F336W, F439W, F555W, and F814W, respectively. The detections in the two ultraviolet filters F170W and F255W are incomplete even at bright magnitudes.

In the final photometry catalogs we want to retain as many true sources as possible, while at the same time eliminate artifacts. After testing a number of options and considering the completeness limits given above, we first chose to retain sources detected in at least one of the two filters where we have the deepest exposures: F439W or F555W, plus one other band (any filter), in order to eliminate possible spurious detections that would appear in a single filter. By imposing this requirement we typically include in the catalogs between 30% and 55% of the total original detections by HSTphot in any filter. The rejected sources are detections in only one filter, and they are artifacts, typically small residual cosmic rays or hot pixels. The percentage of retained sources varies, as we may expect, being higher for fields with higher number of detections. In fact, more total detections generally mean more real stars, if the number of artifacts (the single-filter detections), mostly related to cosmic rays, is constant, since the exposure times are the same in all fields. For one field however, visit 09 in NGC 6822, imposing detection in two filters reduces the total number to about 15%. Visual inspection reveals this field to be particularly plagued with artifacts. We further restricted the catalogs, for the purpose of our analysis, to sources with HSTphot parameters sharpness between −0.3 and 0.3, and type = 1 (defined as "good star"), although sources of with any sharpness value and HSTphot type are included in the online catalogs. Such restrictions conservatively eliminate most remaining artifacts (cosmic rays, background objects), which appear mostly as sources with unrealistic extremely small errors and faintest magnitudes in the error–magnitude plots (Figure 2). We note that applying only a cut in sharpness would be sufficient to effectively eliminate most of these artifacts (between 5% and 7% of the sources across our galaxy sample) and the additional restriction of type = 1 removes about 10 times less sources, about 0.4%–0.7% of the sources (except for Sextans A where it is 2%). This criterion may eliminate also some actual stars, but very few. There are only a few type = 2 objects (defined as unresolved binary) per galaxy, and these fall on the acceptable locus in the error–magnitude diagram (Figure 2); however, the analysis of their composite color would be meaningless. They are therefore retained in the online catalogs for record, but not used in the following analysis. Figure 2 shows error–magnitude plots for a typical field in each filter.

Zoom In Zoom Out Reset image size

We then examined the color–magnitude diagrams of the sources (discussed in the next section) after applying the above restrictions, and we found that a number of sources detected in the two filters F255W and F439W have unrealistic colors of m_F255W − m_F439W ⩽ −4. All these implausible sources have large errors, typically larger than 0.2 mag; however, our maximum error (from the imposed 3.5σ detection limit) is 0.27 mag, and the corresponding ∼0.4 mag error (3σ) on color could not cause even the hottest stars to have such negative m_F255W− m_F439W color. We therefore suspected these sources to be spurious, and inspection of the images indeed revealed most of them to be one-pixel small artifacts such as sharp cosmic rays or hot pixels, that happen to have a match in the other image, because both F255W and F439W have fairly long exposures, and therefore, abundances of small cosmic rays. Much fewer (almost none) of these spurious detections are seen in the shorter F555W and F814W exposures, as can be expected. We further investigated the HSTphot output parameters: sharpness, roundness, χ², and "major axis," in search of criteria to eliminate these spurious sources. Unfortunately, all of these parameters for the artifacts have values in the range of the good stellar sources, therefore they are not helpful. The only useful coincidence allowing us to sieve these remaining artifacts is that they group at color m_F255W − m_F439W ⩽ −3.5 mag, well separated from the stellar sequence (Figure 4), and this coincidence just reflects the difference in the zero points: ZP = 17.037 mag for m_F255W and ZP = 20.886/20.870 (WF2–WF3/WF4) for m_F439W. Most of these spurious sources do not have a counterpart in F555W, thanks to the shorter exposure (which reduces the probability of a cosmic ray coinciding in the same position as in another filter image) which is another criterion to identify most though not all possible artifacts. For example, in NGC 6822, the galaxy where we have most fields (seven), and 0.83 million detected sources overall, we find 101 sources with detection in both m_F255W and m_F439W and m_F255W − m_F439W ⩽ −3.5 mag; 94 sources detected in both m_F255W and m_F439W but not in F555W, all but five of these having m_F255W − m_F439W ⩽ −3.5 mag, and five having normal m_F255W − m_F439W and m_F336W − m_F439W colors. Therefore, in the final analysis catalog we imposed the further requirement of detection in the F555W band (equally deep as F439W but with a five-time shorter exposure), which eliminates an additional ∼0.04% of the sources. This criterion may also eliminate a few real stars, but errors of these objects are mostly close to our faintest-source limit, and the analysis that follows, concerning the most massive stars, is not at all affected by these restrictions. We try nonetheless to apply the best possible criteria in order to make the catalogs generally useful. The final number of sources in each galaxy with these criteria are given in Column 3 (see footnotes "a" and "b") of Table 3. Imposing detection in F814W in our case would unnecessarily eliminate hot sources, since the completeness limit is brighter in this filter, consistent with our program choices, and therefore this criterion was avoided.

To summarize, in the catalogs available online we retained sources with detection in either F555W or F439W plus one other filter in favor of completeness; however, we restrict the analysis sample to sources with detection in F555W (plus another filter), HSTphot type = 1 and |sharpness| < 0.3. The inclusive photometry catalogs are given in the online version of this paper and are also available, with other supporting data, from the author's Web site (http://dolomiti.pha.jhu.edu/LocalGroup/). A small sample is shown in Table 4. In the catalogs, the magnitudes of objects detected in overlapping images are average of the individual measurements weighted by their errors. Their coordinates refer to one of the images, which is provided in Column 24 of the table; in Column 25 we also record on which detector (chip) the source has been measured so that it is possible to retrace sources measured on WF4, which has slightly larger systematic errors. The columns are: (1) running number, (2) parent galaxy, (3) object ID, following IAU designation, (4 and 5) R.A. and decl., (6–17) magnitude and error, in each filter from F170W to F814W, (18–21) UBVI magnitudes, transformed from the m_F336W, m_F439W, m_F555W, m_F814W magnitudes to the Johnson system with HSTphot, (22) sharpness parameter, (23) "type," (24 and 25) visit/field (see Table 2) and detector (chip). Small offsets were added to the coordinates of visit 02 (Δ − R.A. = −0.00018 deg, Δ − decl. = −0.00018 deg) and visit 10 (Δ  −  R.A. = −0.00003 deg, Δ − decl. = −0.00020 deg) to match their sources in overlapping fields (Figure 1).

The astrometry was derived using the metric/wfpc2 IRAF routine to translate the PC/WF coordinates into right ascension and declination. We kept the astrometry in the frame of the parent WFPC2 image (except as noted above) which is useful to produce finding charts for selected objects from this program. In Table 2, we also give offsets to reduce the astrometry to a standard reference frame, obtained by comparing reference stars with the astrometry solution of Massey et al. (2007) photometry. The offsets were derived by comparing stars measured in the ground-based surveys with only one HST counterpart; after examining the distribution of distances between matches in the two catalogs and the multiplicity of matches with decreasing match radius, we concluded that all systematic differences are much less than 1 arcsec and vary from visit to visit. More details are given below.

We compared our HST photometry in m_F555W and m_F439W, transformed to B and V, to the broadband photometry from the ground-based LGGS (Massey et al. 2007). Comparison of astrometry shows the coordinates to agree to much better than 1 arcsec (Section 3.4), with no corrections applied, and the majority of distances between matched sources to be a few tens of an arcsec (for the closest match). These distances decrease to <0.2 arcsec for most sources after the corrections given in Table 2 are applied. If we use a match radius of 2.5 arcsec, we find between 67% and 95% of the Massey et al. (2007) sources to be resolved into multiple HST sources (up to a maximum of 13 HST sources within 25 of a ground-based source). This is due to the HST spatial resolution, primarily, and to the HST photometry being deeper for most galaxies (Figure 3). For four of our sample galaxies, Massey et al. (2007) photometry has a limit of V ∼ 22 mag, and for WLM and NGC 6822 it includes fainter sources: we show in Figure 3 NGC 6822 as the deepest example in the LGGS ground-based catalog and Sextans A as a typical-depth example, both with numerous matches.

Zoom In Zoom Out Reset image size

The number of multiple matches is reduced if we restrict the match radius to 1.5 arcsec, for example, which is comparable to a typical ground-based seeing, and it is also reduced if we consider HST sources brighter than our faintest limit. Obviously, source crowding increases with fainter flux limits, although only unresolved sources of comparable brightness to the closest match would affect unresolved photometric measurements. A ground-based source measurement including a number of unresolved stars would be brighter than the HST resolved sources, and this effect is seen in Figure 3, increasing as fainter sources are considered. Red dots are HST sources for which only one ground-based source is found within the match radius; green dots are cases where both the HST source has a single ground-based match and the ground-based source matches only one HST star within the match radius; these cases are much fewer, and not surprisingly they are found among the brightest sources. Specifically, for a match radius of 2.5 arcsec, we find the following fraction of HST sources to have ground-based matches (and fraction among these matches having a unique ground-based counterpart): 73% (68%), 29% (78%), 44% (68%), 36% (78%), 88% (45%), and 83% (50%) for Phoenix, Pegasus, Sextans A, Sextans B, WLM, and NGC 6822, respectively. Also, 67%, 83%, 90%, 94%, 75%, and 78%, respectively, of the ground-based counterparts match multiple HST sources (i.e., they are resolved into multiple components in HST imaging), and only 24%, 14%, 8%, 5%, 11%, and 12% of the total matched HST sources have a unique ground-based match that is matching only that HST source. As we will see in the next sections, the hottest stars tend to be in crowded regions, making the HST study all the more important for our purpose.

With a more restricted match radius of 1.5 arcsec, after registering the coordinates as shown in Table 2, the above fractions become: 59% (94%), 19% (91%), 29% (90%), 21% (90%), 69% (82%), and 64% (86%) total matches (and fraction of them where the HST source has only one ground-based match); 51%, 63%, 75%, 80%, 59%, and 64% of the matched ground-based sources are matching multiple HST sources (this is the fraction of unresolved stars in the ground-based survey); and finally, for 48%, 35%, 24%, 20%, 34%, and 32% of the matches, the HST source has a unique counterpart and this counterpart is not closer than 1.5 arcsec to any other HST source. We reported results with two often used match radii, although the second is more representative of a ground-based seeing, to aid the interpretation. Within a smaller radius, fewer matches are found, but those that are eliminated are mostly spurious matches; in fact, the number of total matches decreases but the fraction of "single" matches becomes much higher. This is also confirmed by the distribution of distances between HST sources and their ground-based counterparts, which is peaked at small values (≪1 arcsec) for the closest match, while it is flat or grows with distance for secondary matches. Such arguments were also discussed by Bianchi et al. (2011a) regarding other data sets, and are relevant for assessing the significance of comparisons.

Because unresolved ground-based sources include flux from more than one HST source, we compared magnitudes among the two data sets only for stars with a unique match (the green points in Figure 3). We further restricted the comparison to sources brighter than 21.5 mag to avoid large photometric errors. The green dots in Figure 3 indicate a very small offset between HST and ground-based measurements. The average difference between magnitudes (in the restricted sample as defined above) is V (HST)−V (Massey et al. 2007) = 0.086 mag (1σ = 0.187 mag), and B (HST)−B (Massey et al. 2007) = 0.068 mag (1σ = 0.136 mag) for NGC 6822 where the maximum number of unique matches is found, brighter than 21.5 mag (594 stars in V and 384 stars in B). In other galaxies, much fewer matches are found, with these restrictions, between 19 (Pegasus) and 103 (WLM); the average magnitude difference in all galaxies is always in the sense of HST magnitudes being fainter. Restricting the comparison to stars brighter than 21.0 [20.5, 20.0] mag decreases the sample but at the same time reduces the scatter because brighter stars have smaller photometric errors: for NGC 6822 (393 [263, 193] stars in V, 249 [160, 95] stars in B), V (HST)−V (LGGS) = 0.091[0.083, 0.081] mag (1σ = 0.167[0.082, 0.076] mag), and B (HST)−B (LGGS) = 0.062[0.060, 0.061] mag (1σ = 0.099[0.081, 0.067] mag), respectively. In sum, the same offset is found at any magnitude range, and the average difference becomes significant when we consider stars with small photometric errors.

Figure 4 shows observed color–magnitude diagrams (CMD) for the six galaxies in different colors constructed from our set of WFPC2 filters. The data are not corrected for reddening. In order to indicate the expected location of the young stellar populations, we show with vertical lines stellar model colors, computed in the WFPC2 passbands, for T_eff = 50,000, 30,000, 20,000, and 15,000 K. Both intrinsic (long lines) and reddened (short lines) colors with E(B − V) = 0.25 (MW-type dust, R_V = 3.1) are shown. Model colors are shown for two gravity values, bracketing most expected values. The colors from different filters illustrate quantitatively the advantage of using violet and ultraviolet measurements to photometrically characterize the hottest stars, separating them from intermediate temperature stars. This can be further appreciated from Figure 5. In all galaxies a young or very young stellar population (blue stars) is present, although it is very inconspicuous in Pegasus and Phoenix compared with the other galaxies that are richer in massive stars. Comparison of the data with intrinsic and reddened model colors also indicates at a glance that foreground reddening is overall negligible in most galaxies, consistent with the foreground extinction estimates given in Table 3, and with values derived in the next section, except for NGC 6822, which is at low Galactic latitude. In NGC 6822's CMD, the "blue plume," the feature consisting of the upper main-sequence stars with m_F439W − m_F555W ∼ color between 0 and −0.2 mag or m_F336W − m_F439W ∼−1.6 mag, appears displaced by a reddening amount similar to the estimated foreground reddening, E(B − V) ∼ 0.25–0.22 (Massey et al. 2007; Bianchi & Efremova 2006; Bianchi et al. 2001b; Efremova et al. 2011; Table 3). The low Galactic latitude of NGC 6822 also entails a significant contamination by foreground MW-halo stars, visible as a large number of redder objects, mostly between the "blue plume" and the red supergiants.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 6 shows color–color diagrams with two combinations of WFPC2 filters and models (metallicity z = 0.002) of different gravity for two sample galaxies. Sources with best photometry are shown with larger dots in all figures. The plot emphasizes again the sensitivity of our program to the hottest stars. Color–color diagrams for other galaxies are similar to Sextans A; only for NGC 6822 most stars are more reddened than the model curves with E(B − V) = 0.25 mag.

Zoom In Zoom Out Reset image size

We analyze the photometry of individual stars to infer their T_eff and radius, as well as the extinction by interstellar dust. The derived parameters will be used in Section 5 to characterize the young populations.

For the stellar sources we have measurements in up to six filters, only the hottest, most luminous massive stars having good photometry in the far-UV filter as well. We derived the stellar physical parameters by fitting the photometry with grids of model colors, computed in the WFPC2 passbands from stellar atmosphere models with a range of T_eff and gravity values (and metallicity), progressively reddened by increasing E(B − V) values in small steps. Extinction by interstellar dust is determined concurrently with the stellar parameters. SED-fitting results for sample objects are shown in Figure 7. In practice, our model color grids (computed from Kurucz model spectra from various sources and our own) have the greatest T_eff coverage for log g = 5, the hottest temperature models being harder to calculate for lower gravities. In order to gauge accuracy and limitations of the models of the SED-fitting results and the effect of various assumptions (see later), we first compared model colors computed in the U, B, V bands from our grids with canonical U, B, V color calibrations for MW early-type stars; we found the model colors to be in agreement within ≲0.08 mag (U − B) and ≲0.02 mag (B − V) for the hottest stars, and much better for T_eff values lower than 20,000 K, Kurucz-model grids showing better agreement than TLUSTY model grids. Part of the discrepancy may be due to the traditional passbands used for the UBV system and the consistency may be better for HST colors (see Maíz-Apelániz 2006). Although colors at a given T_eff depend also on metallicity and gravity, for hot stars the difference between model colors and empirical calibrations (e.g., Schmidt-Kaler 1982) are larger than this effect. This was also found by Martins & Plez (2006, see their Figure 6); the color sensitivity to gravity is shown in Figure 5 and discussed later. Our previous comparisons between colors from Kurucz models and models including winds (computed with WM-basic or CMFGEN e.g., by Bianchi & Garcia 2002; Bianchi et al. 2009; Garcia & Bianchi 2004) showed the wind lines to have negligible effect on broadband colors, compared with other effects at play, and typical magnitude uncertainties. In any case, resulting T_eff values are necessarily more uncertain at the hottest end of the model range. However, very hot stars are identified as such, more unambiguously from our filter set (extending to the far-UV) than from other combinations commonly used.

Zoom In Zoom Out Reset image size

The color locus of hot stars has little foreground contamination. Colors for hot stars do not vary much with gravity as they do for A- and later-type stars (see Figure 5; also Bianchi & Efremova 2006; Figure 7 in Bianchi et al. 2007; Figures 3–5 in Bianchi 2009). Their log g can hardly be determined by photometric colors; however, once T_eff and extinction are derived, because we know the distance (of their parent galaxy) we can use the extinction-corrected photometry also to approximately separate stars by luminosity class. For cool stars instead, the apparent magnitude of a cool supergiant, e.g., in NGC 6822, does not differ from that of a dwarf of similar temperature in the MW halo (e.g., Bianchi et al. 2001a; Catanzaro et al. 2003). Therefore, their brightness provides a more uncertain separation of galaxy members from MW stars. But cool star colors do differ with gravity, as seen in Figures 5 and 6, and in the above cited papers, and as also shown by Massey et al. (2006) from photometry of objects with known spectral type. Therefore, multi-band photometry also provides some sensitivity to gravity for cool stars.

Once T_eff and extinction are determined from SED fitting (i.e., basically from the SED colors), the scaling between observed flux of a source and best-fit model flux provides the stellar radius, hence also L_bol, since the distance is known (to its parent galaxy) and the reddening correction has been derived (Figure 7). In practice, for every galaxy we perform SED fitting for the stars that have good photometry in enough filters; we use both high gravity and lower gravity models, we compare the resulting T_eff's to spot spurious or unstable results, then we use the resulting stellar radius to assign a luminosity class, and we adopt the [T_eff, E(B − V)] solution from the appropriate set of models. Again, the known distance helps us to choose the most probable solution, by consistency of [T_eff, R_*] with the photometry. Usually, the difference in derived [T_eff, E(B − V)] values from using different log g grids is very subtle for hot stars, and the gravity of the adopted model grid finally does not influence their resulting T_eff (as can be expected from Figure 5). We will return to this point later. The χ² procedure weighs each magnitude by its photometric uncertainty; formal errors in T_eff and E(B − V) are determined by χ² contour levels equal to min(χ²) + 1; for T_eff, they are within 20% for the majority of the hot stars, and much less for cooler stars, while most uncertainties in E(B − V) are less than 30%.

In addition to gravity, another secondary parameter that influences the derived [T_eff, E(B − V)] values from SED modeling is the adopted metallicity, again the effect being less relevant for hot stars. Figure 5 shows that metallicity does not affect colors dramatically; however, it slightly affects the magnitudes (owing to the different line blanketing) and hence the derived radius. Comparison of model spectra with T_eff = 50,000 K, for example, shows a difference of less than 10% in flux between solar and 1/100th solar metallicity (the difference being higher around 2000 Å than in the optical, and about 5% between Lyα and 2000 Å). Below we present more comparisons to illustrate how results depend on the assumed metallicity.

A final issue, quite relevant for hot stars, concerns the extinction by interstellar dust. As mentioned above, our model color grids were obtained by progressively reddening model spectra for E(B − V) values increasing in very small steps (0.01 mag) and computing synthetic magnitudes for each reddened model (this is more accurate than applying an average absorption at the λ_eff of each filter; see Bianchi 2011). We know from UV spectroscopy of hot stars in the MW, Magellanic Clouds, M31, and M33 (e.g., Misselt et al. 1999; Bianchi et al. 1996; Bianchi et al. 2004 and references therein; see also Bianchi 2011 for comparison of known curves) that the selective extinction A_λ/E(B − V) may vary significantly, especially at UV wavelengths and higher UV A_λ/E(B − V) values being often associated with regions of intense star formation (see also, e.g., Calzetti et al. 2005). Such differences in the properties of interstellar dust have negligible effects on reddening of optical colors, but affect the F255W magnitudes (the passband includes the 2175 Å broad feature) and the F170W magnitudes (the A_λ/E(B − V) steepness in this wavelength range varies according to the properties of small grains, e.g., Mathis 1994).

A precise derivation of the UV extinction curve toward individual stars requires UV spectroscopy, but such studies are necessarily limited to a few sightlines. In order to obtain from the photometry an overall indication of what kind of extinction curve is more appropriate for our targets, we have performed the stellar SED analysis assuming different known curves of selective extinction, from the MW "average" with R_V = 3.1 to LMC and SMC UV-steep curves, more representative of SF regions and low metallicity, and compared the results in various ways. In Figures 8–12 we show some of the comparisons. We indicate with "R_V = 3.1" results obtained from model grids reddened assuming the MW typical curve with R_V = 3.1, taken from Cardelli et al. (1989), "LMC2" indicates that we used the curve from Misselt et al. (1999), "avgLMC" labels results assuming the LMC extinction curve in sightlines excluding 30 Dor from Gordon & Clayton (1998), and "SMC" indicates the very steep curve derived along SMC sightlines by the same authors. First, because variations in the A_λ/E(B − V) curve mostly concern the UV wavelengths, we examined which A_λ/E(B − V) curve produces the best agreement between E(B − V) results derived from optical colors only and from UV-to-optical colors. We expect the difference to be appreciable only where the intrinsic extinction (the component that we expect to be peculiar) is significant compared with the foreground MW component. In Figure 8, we examine the effects of dust type on our results by comparing E(B − V) values obtained for the stars with good photometry in all bands, with results obtained using optical colors only, for the same stars (i.e., neglecting our UV measurements). The latter, however, gives much more uncertain results because optical colors are not a sensitive probe of high T_eff's and vary little (typically less than the photometric errors) for stars hotter than ∼18,000 K (see Figure 5; also, e.g., Bianchi 2007 and references therein), which limits the significance of the comparison. In fact, Figure 8 is more illustrative of the effects in which the SED-fitting procedure may incur (for hot stars) when no UV information is available. Results are compared using z = 0.002 models, taken as a reference metallicity value, but in fact both extinction type and metallicity influence the derived parameters and it is not always possible to disentangle their effects. In WLM and Sextans A, neglecting the UV information results in lower T_eff's for the hottest stars; it also causes the SED fitting to find, for a number of objects, a solution for T_eff equal to the highest value of the model grid (50,000 K) with a gap at T_eff ≳ 35,000 K, the extent and the limit of the gap depending on the assumed dust type. The smoothest and most consistent distribution is obtained for "LMC-avg" or "LMC2"-type A_λ/E(B − V). A similar effect seems present in Sextans B and in the smaller Pegasus and Phoenix, but the paucity of stars with good photometry all the way to far-UV and larger scatter makes the comparison inconclusive in these cases. In NGC 6822, the most consistent results (and the least deviations to extreme values) are seen when "LMC2"-type A_λ/E(B − V) is assumed for hot stars, and an effect opposite from the other galaxies is seen when an extremely steep, "SMC"-like A_λ/E(B − V) curve is used. In NGC 6822, however, the total E(B − V) for each source includes a substantial foreground component, and the combination of two components with possibly differing dust properties complicates matters further. In all galaxies, assuming MW-type dust (R_V = 3.1) causes a clustering of points around T_eff ∼ 40,000 K, in the solution accounting for UV photometry, which is not realistic, and indicates that the procedure may not find a "good" minimum and tends to match the SED primarily with E(B − V) adjustments rather than by variations of both [T_eff, E(B − V)]. This is a known pitfall of the χ² procedure, since both free parameters (in our case T_eff and E(B − V)), over the range considered induce changes in photometric colors, but at particular regimes one of the two may drive the changes more than the other; the circumstances when this happens vary according to the parameters range, the model grid, and especially the measured colors available, each color being particularly sensitive to some parameter at some range of values. More colors and finer model grids help averting or mitigating such insidious pitfalls, but finally it is the sensitivity of the available colors to the free parameters that drives the limitations of the results. Of course, for stars cooler than ∼25,000 K the agreement is always excellent in Figure 8; it does not make a difference if UV measurements are neglected, because their optical colors are not saturated, on one hand, and dust properties vary only at the wavelengths of the UV filters. Ultimately, this comparison gives more of a sense of the hidden errors the procedure may incur in when no UV information is available for hot stars.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We then compared statistically χ² values from the SED fitting performed under various assumptions (for objects with small photometric errors to avoid blurring the results with "false good" χ²). Figures 9 and 10 show χ² values to be somewhat reduced for hot stars (in most cases) when solar metallicity is used; it also shows subsolar metallicity to reduce the χ² for a subset of cool stars in NGC 6822. Note that sources with smaller (combined) photometric errors have larger χ² (cyan points) because χ² measures the best-fit-model departure from the data relative to the observational errors, which is a misleading indication of the physical meaningfulness if taken alone, and not considered in conjunction with the photometric errors. Finally, visual inspection of SED fits (such as those shown in the top of Figure 7) for the best cases, and of χ² contours, indicates, as the other comparisons above, that a UV-steep A_λ/E(B − V) curve often produces better results in the hot star samples, and, in particular, improves the match with the F255W observed magnitudes, in many cases.

We also compared [T_eff, E(B − V)] values derived by SED fitting using models with different gravities (Figure 11). In the grids used here, lower gravity models do not extend to T_eff hotter than 31,000 K (above which we found UV-steep extinction to be more adequate from the previous comparisons), and this explains in part why log  g = 3.5 models tend to give lower T_eff toward the upper temperature limit. Over most of the range, the method does not diverge to extreme T_eff or E(B − V) values. For cool stars, with T_eff < 10,000 K, photometric colors largely differ with gravity (Figures 5 and 6), and the fitting solution may have two actual minima, a lower T_eff with lower E(B − V), or a hotter temperature and higher reddening, as one can see, for example, by dereddening the data points along reddening vectors in Figure 6 (the points would intersect the model intrinsic-color curve twice, at two different E(B − V) values), or from classical reddening-free indices, e.g., Bianchi & Efremova (2006) and Romaniello et al. (2002). One or the other of the two minima prevails in the numerical procedure (χ²) when slightly different model grids are used. In such cases, the known distance helps us to assess, by the consistency of T_eff and radius, which of the two solutions is more physically meaningful.

It is important, finally, to remember that we cannot expect the dust properties and metallicity to be uniform within the entire sample of each galaxy; in fact, we expect them to largely vary, the UV extinction curves being extreme in regions of high star formation, and the young (hot) stars having possibly higher metallicity than older stars, as indicated also by the existing measurements of metal abundances compiled in Section 1. For this reason, we provide results from our SED analysis obtained under a range of assumptions, the most appropriate combination likely being different case by case. However, from our present standpoint of presenting and examining comprehensive photometric catalogs, we wanted to illustrate the properties of the majority of objects and to look for conspicuous trends and differences among galaxies. Therefore, in the comparisons above and in the following, we only divided the samples in hot and cool stars, which can be taken as a proxy for young/old (most cool stars being low-mass main-sequence objects). When the comparisons show inconclusive trends, they indicate either the limitations of the method and/or actual variations across the galaxy.

We examined the resulting E(B − V) values from the SED fitting of individual stars for each galaxy to derive average values and variations across the sample. Our HST fields contain the most conspicuous OB associations and SF regions in each galaxy, except for NGC 6822 (Figure 1), and some small portions of field population. This is reflected in the histograms of the derived E(B − V) values, which are shown in Figure 12. Such histograms show in many cases both a peak around a minimum value, appropriate for the field population (shared by the majority of cool stars), and not surprisingly similar to the foreground value, and another peak at higher E(B − V) values, mostly due to the hotter stars, representative of regions of intense star-formation. The reddening difference between hot and cool stars is more evident when there is a large number of sources; it also depends on the environments sampled by our HST imaging, which include more or less "field" areas in each case. The median values for the total sample, and for the hot and cool stars' subsamples, are shown in Figure 12 and given in Table 3. The average and median values are from the hot and cool stars for which SED fitting was performed (i.e., when sufficient measurements are available, see Table 3). As discussed in the previous section, the exact result depends on the type of extinction curve assumed for hot stars where UV measurements are relevant and therefore various cases are shown in Figure 12 and reported in Table 3. In Table 3, we give values obtained using both the assumption of MW-type dust (R_V = 3.1) and "LMC2-type" (Misselt et al. 1999) dust, representing two typical extremes among known extinction curves and for two metallicities. The difference between results from these assumptions again gives an indication of the uncertainty that the lack of detailed information on dust characteristics may cause on the results. Because extinction properties vary along individual sightlines, as is well known in the MW and the MC, uncertainties due to the assumed dust properties somewhat average out in the global results, but can be more severe in individual sources. For all galaxies except NGC 6822, the extinction is very small, and therefore its uncertainties do not heavily weigh on the results and on the estimate of the massive star content. Finally, we recall again that, in more detail, dust characteristics may significantly vary across the galaxy with strong deviations (UV-steep curves) in the cores of SF regions, where the UV radiation from the hottest stars modify grain properties of the "average" ISM. While such level of detail requires spectroscopy, the present photometric characterization is necessarily confined to general terms. In the following discussion and figures, for the stars with good photometry enabling SED fitting we use the individually derived E(B − V) values, and for the rest we adopt the median values from Column 7 or 9 of Table 3 according to their m_F439W − m_F555W color (bluer or redder than 0.4 mag) or the average between the cool-star and hot-star sample median values. From the SED fitting, we also culled MW foreground stellar candidates and binary candidates (example in Figure 7); these were excluded from the above-derived average values and from the following figures.

Having derived T_eff and extinction for the stars, we can construct HR diagrams for comparison with evolutionary tracks and isochrones, shown in Figure 13. Dereddened absolute magnitudes are scaled using the distances compiled in Column 2 of Table 1. In order to also plot the fainter/redder stars detected in fewer filters where T_eff cannot be determined by SED fitting (or by analytical solutions such as reddening-free indices) and therefore would be much less accurate, we plot absolute magnitudes versus dereddened color rather than T_eff, L_bol in Figure 13. We overplot isochrones from Padua evolutionary models from Marigo et al. (2008) with the corrections from Case A in Girardi et al. (2010) and Girardi et al. (2008) (which include the isochrones by Bertelli et al. 1994 for initial masses >7 M_☉). From the analysis described in the previous sections, we derive the number of hot massive stars within our HST fields in each galaxy. We counted the stars brighter than the M_F555W value predicted by the isochrones for (initial) masses of 20 M_☉ and 9 M_☉ at age = 4 Myr (M_F555W = −4.36 and −2.041 mag, respectively) and (M_F439W − M_F555W)₀ <−0.05 mag. We report the hot-star counts in Table 3, with the caveats that (1) they refer not to the entire galaxy, but to the portions included in the HST imaging, which is a fraction of the total area largely varying from galaxy to galaxy; and (2) with this selection, we only count the massive stars of the youngest populations (few Myr), not the evolved ones (given our color cut). In reality, our chosen color cut of (M_F439W − M_F555W)₀ <−0.05 mag may also include some stars on the 10 Myr isochrone, as can be seen from Figure 13; on the other hand, due to the spread from photometric errors, reddening correction uncertainties and metallicity, a more stringent color cut would not be adequate. The uncertainties in the number of massive stars are derived by applying the 1σ error to the photometry and counting how many stars would be included/excluded in our color–magnitude limits. When there is a substantial population along slightly older isochrones, the uncertainty estimated in this way becomes very large. While the HST footprints enclose almost all of the young populations for the small galaxies (except for NGC 6822), as shown in Figure 1, one cannot use these numbers, for example, to infer the density of hot stars and compare it among fields without accounting for the spatial distribution of the star-forming complexes. We will perform further analysis with complementary data and divide the sample in individual star-forming regions, as shown in Bianchi et al. (2011c). In NGC 6822, we observed seven fields (a few others with conspicuous H ii regions were studied in earlier programs by Bianchi et al. 2001b; Bianchi & Efremova 2006), and the stellar counts cannot be directly scaled by area to the entire galaxy, given the great variety of properties and the patchy distribution that the star-forming sites display. For the other galaxies, our HST fields include basically all the regions rich in hot stars, but not the entire galaxy area (Figure 1).

Zoom In Zoom Out Reset image size

Because T_eff and E(B − V) are derived concurrently in the SED-fitting process (both quantities significantly affecting the spectral slope), adopting a different extinction curve results in a different solution for both quantities, as discussed above. For this reason, we report in Table 3 the count of hot stars obtained using different assumptions. Again, variations average out to some extent over large samples, making the differences less relevant than for individual objects.

The HRDs in Figure 13 show that all galaxies do have stars along the youngest population tracks, but these are minimal in Phoenix (the only Spheroidal, or transition Sph-Irr galaxy, of the sample) and Pegasus. Note also that the HRD of Phoenix extends to fainter stars, because this galaxy is the closest in our sample (NGC 6822 has similar distance but suffers from higher foreground extinction). Older stars are clearly seen along the 100 Myr isochrone. In this galaxy in particular, the star formation history in older epochs has been studied with HST V − I photometry (which is sensitive to old populations) in the central region by Holtzmann et al. (2000) and Young et al. (2007), who found it to be declining over the past 1.0–0.5 Gyr, with a last SF episode occurring around 100 Myr or 50 Myr (according to these two works, respectively), and a gap before that episode. Such ∼10⁸ yr population was first suspected by Ortolani & Gratton (1988) from ground-based imaging and is clearly confirmed by our HRD, which is more sensitive than V − I even to intermediate age stars. The minimal presence of stars with possible masses above 9 M_☉ (Figure 13) would be unnoticeable in these other surveys.

Sextans A and WLM show the most vigorous star-formation in the sample, and two major populations are clearly seen in their HRD (Figure 13), one of a few Myrs of age and the other of a few tens of Myrs. These appear in Figure 14 as concentrations of very hot stars (blue–purple dots) and a more diffuse spread of ∼B-type stars (green dots). In Sextans B we see similar populations, but less conspicuous in numbers than in Sextans A, especially along the youngest isochrone compared with the plume of stars about the log(age) = 7.5 yr isochrone. These differences are reflected in the number of massive stars in Table 3. Remember, however, that in this galaxy only one field was observed, and two and three in Sextans A and WLM, respectively (according to their respective areas of star-formation revealed by GALEX FUV imaging). However, the three galaxies have very similar M_V.

Zoom In Zoom Out Reset image size

View extended figure (10 panels)

In the nearer, larger galaxy NGC 6822 where we imaged seven fields, the HRD shows again a very young population stretching along the 4 Myr isochrone up to high masses and older populations as well. We must recall that the higher extinction toward this galaxy also causes larger uncertainty in the extinction correction. MW foreground stars are not negligible at intermediate colors, but not expected at very blue colors.

The HST-WFPC2 resolution, from 0.2 pc in Phoenix to 0.7 pc in Sextans B, allows us to examine the spatial distribution of the hot massive stars over entire young associations yet with great detail even in the crowded SF sites. In Figure 14, we overlay the position of the brightest sources with SED-fitting results on the HST images, color-coding the stars by T_eff and indicating their derived radius (which means also luminosity) by the symbol size. Only the stars with good photometry in most filters were included in the subsample for SED fitting and are shown, which favors the brightest stars and in particular the hot stars, given our purpose and filter choice; therefore the maps give a snapshot of the youngest populations. Not surprisingly, the youngest and most massive associations also appear more compact (examples in Sextans A, Sextans B, WLM), while lower temperature stars (green and yellow dots), presumably tracing slightly older populations, are more spread out.

Pegasus and Phoenix, by far the poorest in massive stars among our sample, show sparse associations. When looking at Figure 14, one must also take into account that, e.g., Phoenix is three times closer than Sextans A, therefore we are viewing its population comparatively with a zoom-in factor of three, and so for NGC 6822. Moreover, in this galaxy the most massive and compact SF regions were studied by Bianchi et al. (2001b) and Bianchi & Efremova (2006), and the present fields include smaller complexes. Yet, within each galaxy, we invariably find the most massive and youngest associations to be the most compact. Only a subsample of the maps is shown in Figure 14. Similar maps for all fields are posted in the online version and on the author's Web site http://dolomiti.pha.jhu.edu/LocalGroup for this program.