ULTRAVIOLET SPECTROSCOPY OF CIRCUMNUCLEAR STAR CLUSTERS IN M83

The FUV data set consists of a direct image and four two-dimensional (2D) spectrograms. The direct image is projected onto the MAMA detector, which has a field of view of 25'' × 25'' and a plate scale of 0024 pixel⁻¹. The spectrograms, one long slit (25'' × 2'') and three multi-object (slitless), were taken with the G140L grating, which has an average dispersion of 0.584 Å pixel⁻¹. Table 1 gives the observation date (Column 2), exposure time (Column 3), aperture position angle in degrees east of north (Column 4), filter name (Column 5), wavelength coverage for the detector + filter combination (Column 6), and extension of the most processed data file available from the Hubble Data Archive of the FUV observations.

The data were partially processed through the On-the-Fly Reprocessing (OTFR) pipeline of the archive, at retrieval time. This included data linearization, dark subtraction, flat fielding, and summing of the repeated observations. In addition, the pipeline corrected the direct image for geometric distortion. After processing through the pipeline, in order to increase the signal-to-noise ratio (S/N), we added together exposures taken with the same instrument configuration, i.e., the two slitless exposures taken with the F25SFR2 aperture, which we refer to as the F25SFR2 exposure.

Because the spectrograms were taken in either a slitless or a wide slit mode with the G140L grating, the spectral traces have large offsets in the dispersion direction. Therefore, the OTFR pipeline did not wavelength calibrate or convert these spectra to absolute flux. In order to extract wavelength-calibrated one-dimensional (1D) spectra from the 2D spectra, we first located the position of each spectral trace along the cross-dispersion direction, and then used the X1D module (McGrath, Busko, & Hodge, STIS ISR 99-03) available in the Image Reduction and Analysis Facility (IRAF; Tody 1986, 1993) for this task. Several iterations were required in order to find the optimal value of xoffset, which is the parameter used by X1D for shifting a spectrum along the dispersion direction. We used the absorption feature C ii λλ1334, 1335 + λ1336 (hereafter C ii 1335), for finding the optimal value of xoffset. In our spectra, this feature is unresolved and is a blend of Milky Way interstellar and M83 stellar components.

For the width of the extraction box we used the value recommended in the STIS manual for point sources, i.e., 11 pixels (Brown 2002). Background spectra extracted from outer portions of the 2D spectrograms were assumed to represent purely geocoronal emission (airglow) from H i Lyman-α (Lyα) at 1216 Å and/or [O i] 1302 + 1305 + 1306 Å, depending on the aperture, and were subtracted from the cluster spectra. We did not subtract the underlying diffuse light because of its large variations as a function of position.

The direct image and the 2D spectrograms from the STIS are shown in the four panels of Figure 2. Panel (a) features the 25'' × 25'' direct image in logarithmic scale, with the location of M83's optical center marked with a cross, and the spatial scale marked with a vector. Panels (b) and (c) show the long-slit and the multi-object spectrograms, respectively, both taken with the same aperture (25MAMA). The vertical band in panel (b) is due to Lyα airglow filling the 2'' wide slit (it is seen more clearly in the online version). The Lyα airglow contaminates much of the slitless spectrogram shown in panel (c). Finally, panel (d) features the sum of the two slitless spectrograms taken with the F25SFR2 aperture, which rejects the Lyα airglow. No contamination from the [O i] airglow is apparent in panel (d) because the emission brightness geocoronal [O i] is typically less than 10% of that of geocoronal Lyα, depending on the time of observation and the position of the target relative to the Earth (Miskey & Bruhweiler 2003).

Zoom In Zoom Out Reset image size

We used the slitless exposures for extracting spectra of those clusters falling outside the slit as well as for studying the effect of rejecting the Lyα airglow on our results.

The spectral resolution depends on the spatial extent of the star clusters. In order to estimate the spectral resolution of our spectra, we determined the full width at half-maximum (FWHM) of the cross-dispersion profiles of the spectral traces. For our point-like representative, GD 153, the FWHM is 009. Due to blending with the profiles of close neighbors, we were only able to determine the FWHM of four clusters by our method. For clusters 1, 4, 10, and 11+12 we obtained FWHM of 017, 030, 019, and 023, respectively, which suggests that the clusters are partially resolved. Since these are neither point sources nor fully extended sources, we expected the spectral resolution of our long-slit spectra to be intermediate between the values given in the STIS instrument handbook, i.e., 0.9 Å (point source) and 48 Å (extended source). Multi-component Gaussian fitting of the C ii feature at λλ1334, 1335 + λ1336, carried out with the Peak Analysis code (available at ftp://ftp.ncnr.nist.gov/pub/staff/dimeo/pan.zip) reveals that the FWHM of this feature ranges from 2.5 to 7 Å. Our field spectra are expected to have the resolution of a fully extended source.

We corrected the cluster spectra for reddening due to the presence of dust in the Galaxy and in M83. For the Galaxy's contribution to the reddening, we adopt E(B − V) = 0.06 (Schlegel et al. 1998), and we assume the Galactic extinction curve of Fitzpatrick (1999). For M83's contribution, we (1) fitted a power law of the form F ∝ λ^β to the continuum in the range 1240–1600 Å, (2) assumed that any deviation of β from −2.6, which is the expected value for a dust-free starburst, is due to extinction by local dust, and (3) de-reddened the observed spectrum until −2.6 was reached, using the starburst obscuration law of Calzetti et al. (2000).

In order to match each cluster with its FUV spectral trace, we started by directly comparing the positions of the clusters in the STIS direct image, with the positions of the spectral traces in the 2D spectrograms. For this, we rotated the image and vertically shifted the spectrograms until the orientations matched, as shown in Figure 3, which shows enlarged portions of the STIS image, side by side with two spectrograms. Note that the slitless spectrograms all have the same orientation, while the orientation of the long-slit spectrogram is different. Therefore, in Figure 3, we show only the slitless spectrogram taken with the F25SFR2 aperture. In panel (a) of Figure 3, we indicate the position of the long slit, the locations where we extracted the sky and field region spectra, and the correspondence between clusters within the slit and their spectral traces. Panel (c), which shows the same image as panel (a), but with a slightly different orientation, shows only one sky region due to the fact that the second falls outside of the field displayed in the panel.

Zoom In Zoom Out Reset image size

Unfortunately, and as can be seen in Figure 3, in spite of the different orientations of the spectrograms, the spectral traces of the two brightest clusters, 11 and 12, always overlap. Furthermore, as seen in panel (d), the faint spectral traces of clusters 4 and 5 also overlap, as do the traces of cluster 14 and a source not in our sample. Fortunately, we can use the long-slit spectra for clusters 4 and 14. However, cluster 5 was rejected from our analysis due to its poor spectral trace.

The spatial correspondence between star clusters and their spectral traces can be checked in Figure 4, where we compare the cluster radial profiles derived from the STIS direct image (bottom curves in left and right panels), with the cross-dispersion profiles derived from the long-slit (top curve, left panel) and the slitless (top curve, right panel) 25MAMA spectrograms. Also shown in Figure 4 is the cross-dispersion profile of a point-like source, represented by white dwarf GD 153 (top dotted curve in left and right panels), observed as part of program 7097, in the same configuration as the long-slit spectrogram. For convenience, we divided each curve in Figure 4 by the amplitude of the tallest peak, which for the clusters is the amplitude of target 11 + 12 (which overlap), and for the white dwarf is its own amplitude. For clarity, the top curves in Figure 4 (cross-dispersion profiles) were all shifted by 0.2. The clusters and regions used for the background subtraction are labeled.

Zoom In Zoom Out Reset image size

For the radial profiles, we first rotated the image to match the orientation of each spectrogram, and we then obtained the count rate at a given vertical position by integrating over the horizontal direction. The radial profiles in the left and right panels look slightly different because of the different angles by which we rotated the image. We note that although the peaks for clusters in our sample are dominated by emission from the clusters themselves, diffuse light and fainter sources also contribute to their shape and amplitude. This, however, does not affect the identification of the spectrum of each cluster.

For the spectral profiles, the count rate at a given position in the cross-dispersion direction was obtained by integrating over the dispersion axis. Note the higher background of the top right curve with respect to the top left curve, which is due to contamination with Lyα airglow over a larger portion of the detector in the slitless spectrogram (right panel of Figure 4).

Circular aperture photometry was performed on the clusters using the IRAF task PHOT, using an aperture radius of 3 pixels and a background annulus with radii of 10 and 13 pixels. Aperture corrections, based on the estimated size of each cluster, were applied to convert the fixed aperture magnitudes to total magnitudes. The instrumental magnitudes in the F336W, F438W, F555W, F658N, and F814W filters were converted to the VEGAMAG photometric system by applying zero points available at the following URL: http://www.stsci.edu/hst/wfc3/phot_zp_lbn. We refer to the broadband filters as the "U," "B," "V," and "I " magnitudes for simplicity, although they were not transformed to the Johnson–Cousins system. For more details, refer to Chandar et al. (2010).

The semi-empirical models were built from spectral libraries of Galactic O and B stars observed in the 1000–1200 Å range at 0.12 Å resolution with the Far Ultraviolet Spectroscopic Explorer (Pellerin et al. 2002) and in the 1200–1800 Å range at 0.75 Å resolution with the International Ultraviolet Explorer (Robert et al. 1993; de Mello et al. 2000). Comparisons of the semi-empirical models with observations of local and distant galaxies generally yield quite good agreement (Leitherer 2010).

The theoretical models were built from a spectral library of hot massive stars (T_eff ≳ 20, 000 K, M ≳ 8 M_☉) that was computed as described in Leitherer et al. (2010), using the stellar atmosphere code WM-Basic (Pauldrach et al. 2001), following the approach pioneered by Rix et al. (2004). The models span the spectral range 900–3000 Å at high (0.5 Å) resolution. This resolution is sufficient to resolve the main UV stellar wind lines, which have widths in excess of 1000 km s⁻¹.

The strengths and weaknesses of the theoretical stellar library are the following. (1) The two main publicly available codes optimized for modeling the UV spectra of OB stars with winds are CMFGEN (Hillier & Miller 1998) and WM-Basic. WM-Basic, which we utilized, solves the radiative transfer and the hydrodynamics self-consistently to derive the density structure of the stellar wind, while CMFGEN assumes an analytical approximation for the wind structure a priori. (2) Wolf–Rayet (W-R) stars, whose stellar evolution is quite uncertain, are modeled using WM-Basic atmospheres. This ensures that their continuum fluxes are accounted for, but W-R-specific lines are not reproduced correctly. Fortunately, W-R stars in the local universe do not contribute significantly to the UV spectrum of OB stellar populations, except for He ii λ1640, which should be viewed with care. In individual W-R stars, the equivalent width (EW) of the He ii 1640 emission component ranges from 3 to 100 Å (Conti & Morris 1990), with a preference for smaller values (3–30 Å or so) for late-type WN stars, which tend to dominate observed statistics due to their high absolute visual magnitude. In starburst clusters, the typical EW of the 1640 bump is about 2–3 Å (Chandar et al. 2004). If EW = 15 Å is the typical value for late-type WN stars, this suggests a continuum diluting about 80% of the light at 1640 Å, and W-R stars cannot contribute more than 20% to the UV continuum. The dilution is caused by the strong continuum of O main sequence and O supergiant stars. (3) We omit the geometrical effect of clumps in the stellar wind, but the associated X-ray emission from shocks propagating in the wind is included.³ Clumping affects the lines from trace ions, e.g., O vi λ1035 and O v λ1371 (Bouret et al. 2005). In addition, the strength of the wind lines depends on the assumed ionization balance. While photoionization is the primary mechanism determining the ionization state of the wind, X-rays arising from shocked material can be important as well (e.g., Pauldrach et al. 1994). (4) We omit Stark broadening of the hydrogen lowest Lyman lines, which is computationally expensive. A pure Doppler profile of hydrogen is very similar to one with Stark broadening included, as long as the upper principal quantum of the transition is small (Repolust et al. 2005). This is the result of the Stark width's dependence on the fourth power of the upper principal quantum number. In practice, this means Stark broadening becomes negligible in hydrogen lines at higher energies and originating at lower densities. Lyα, owing to its large optical depth, is formed far out in the wind even at relatively low mass-loss rates. Therefore, Stark broadening is not important at ages younger than 10 Myr. (5) The evolution of massive stars in binary systems and of rotating massive stars requires further study before it can be included in the models (Eldridge & Stanway 2009; Vázquez et al. 2007). (6) The models can be used for interpreting the rest-frame UV spectra of objects with metallicities other than near-solar or those of the Large (Z = 0.007) and Small (Z = 0.002) Magellanic Clouds (LMC, SMC; Maeder et al. 1999).

Because there are no empirical UV spectral stellar libraries at Z>0.020, we have to rely on fully theoretical star cluster spectra, at Z = 0.040. Here, we discuss differences in the theoretical spectra of SSPs that were computed with stellar evolution tracks corresponding to metallicities of Z = 0.020 and Z = 0.040.

Line-driven wind theory predicts lower mass-loss rates and lower wind terminal velocities for O stars of lower metallicity (Castor et al. 1975; Puls et al. 1996), and therefore, weaker wind features in the spectra of these stars. Observations of MS O stars in the Magellanic Clouds and the Galaxy support the theory. For instance, see Mokiem et al. (2007), where empirical stellar wind results for the Milky Way, the LMC, and the SMC are compared with the OB star theoretical wind models of Vink et al. (2001), which are an extension of line-driven wind theory. WM-Basic models do a good job of reproducing lower wind terminal velocities at lower metallicity (Leitherer et al. 2010), as can be seen in Figure 5, which shows a comparison between the Z = 0.020 and the Z = 0.040 models, for ages from 1 to 20 Myr (as labeled on the right), in the wavelength range 1200–1700 Å, at 0.75 Å resolution. For instance, compare the absorption components of C iv 1550 at Z = 0.020 and Z = 0.040. In addition, the photospheric lines are deeper at the higher metallicity, as expected.

Zoom In Zoom Out Reset image size

Figure 6 shows semi-empirical and theoretical models for SSPs with different high-mass limits of the stellar IMF. The figures show ages 2, 3, 5, and 10 Myr and focus on C iv 1550 and Si iv 1400, whose profiles are sensitive to the distributions of O main sequence and O supergiant stars, respectively. All models in Figure 6 have a lower mass limit of 0.1 M_☉. The thin black curves use a Kroupa IMF from 0.1 to 100 M_☉, while the thick gray curves (red in the online version) use IMF high-mass limits (M_up) of 10, 30, and 50 M_☉.

Zoom In Zoom Out Reset image size

Lowering M_up, i.e., gradually removing the O and the most luminous B stars, has the following effects. (1) As M_up decreases, the P-Cygni profiles of Si iv 1400 and of C iv 1550 become weaker. In particular, the emission component of these profiles is non-existent at M_up = 30 M_☉ for the semi-empirical models and drops in strength more gradually in the theoretical models. Also the strong absorption component of these profiles is purely photospheric at M_u = 10 M_☉, for both models. (2) For the theoretical models at M_up = 10 M_☉, the absorption lines are dominated by photospheric lines from B stars.

Here, we discuss differences between the semi-empirical and the theoretical models at Z = 0.020. Figure 7 shows the theoretical and semi-empirical models, using the same format of Figure 5. There are several noteworthy features. (1) In the region below 1240 Å the semi-empirical spectra suffer from Lyα absorption in the Galactic plane. (2) The P-Cygni profiles of the resonant doublets of N v at 1238.8+1242.8 Å (hereafter N v 1240) and of C iv at 1548.2+1550.8 Å (hereafter C iv 1550), which originate in the stellar wind, decrease in strength as the O stars evolve and disappear. (3) The resonant doublet of Si iv at 1393.8+1402.8 Å (hereafter Si iv 1400) is absent in main-sequence (MS) O stars and develops P-Cygni profiles in giant and supergiant O stars. This is why the P-Cygni profile of Si iv 1400 is strong at ages 3–5 Myr, when O supergiants are present. Note that the lack of strong N v 1240, C iv 1550, or Si iv 1400 P-Cygni profiles, in the spectrum of a star cluster, indicates either that the cluster did not form very massive O stars or that the cluster is too old to show the signatures of such stars. (4) At 3 Myr, the semi-empirical model has somewhat stronger N v, Si iv, and C iv lines than the theoretical model. At this particular time step, stars with stronger winds make a stronger contribution to the semi-empirical than to the theoretical spectrum. This difference serves as a reminder of a crucial assumption made when linking model spectra to stellar evolution models. In the case of the theoretical spectra, one simply adopts the effective temperature of the atmospheres for matching the positions of the evolutionary tracks. In contrast, for the empirical spectra, we obtain the effective temperatures from spectral types via a temperature scale. Starburst99 assumes the relation between spectral type and effective temperature of O stars derived by Martins et al. (2005). We note that switching between the relations of Martins et al. and the earlier work of Vacca et al. (1996) or Schmidt-Kaler (1982) modifies the peak of the Si iv emission by almost 10%. Careful comparisons with observations should provide guidance for choosing the preferred approach in the synthesis models. In addition, the WM-Basic models use the revised solar composition of Asplund et al. (2009), which has a lesser content in metals than the solar composition of the stellar evolution tracks. While the revised solar abundances have a small effect on the evolutionary tracks of massive stars, they do affect the computed UV spectra via changed opacities and wind properties. (5) Many of the narrow absorption lines in the semi-empirical models are of interstellar origin. For the first four Myr, the absorption line Si ii λ1264 is stronger in the semi-empirical models because of the interstellar contribution. (6) O v λ1371, which is strong at ages ≲2 Myr, i.e., in the hottest stars, is stronger in the theoretical models due to the neglect of wind clumping (Bouret et al. 2005).

Zoom In Zoom Out Reset image size

Table 2 allows an assessment of the uncertainties in the reddened flux at 1500 Å, F₁₅₀₀, and in the power-law index of the reddened FUV continuum, β_obs, for the 13 clusters in our final sample. For each spectral exposure and each cluster, Columns 2–12 in the table give (1) the cluster ID, (2) and (3) the right ascension and declination measured from the STIS direct image, (4)–(6) the median S/N of the reddened spectrum, (7)–(9) F₁₅₀₀, and (10)–(12) β_obs.

The faint clusters 4 and 14 have the lowest S/N values. The rest of the clusters have reasonably good S/N (⩾10). The F₁₅₀₀ fluxes from the slitless exposures taken with the 25MAMA and F25SFR2 apertures agree within 10%. Therefore, the Lyα airglow has little effect on F₁₅₀₀. The slit blocks the light from nearby sources and lowers F₁₅₀₀ by 20%–50%. The value of β_obs (1) increases by 20% when the wavelength range for fitting the continuum is extended (compare the two values in Column 10); (2) varies by at most 10%, due to geocoronal Lyα contamination (compare Columns 11 and 12); and (3) increases by at most 10% when the slit is removed (compare Columns 10 and 11 or 12). The properties of clusters 1–3 derived from the slitless spectra are less reliable than those derived from the long-slit spectra and are only shown for comparison.

Since the two slitless exposures yield similar values in Table 2, we decided to discard the exposure spanning the shortest wavelength range, i.e., the one taken with the F25SFR2 aperture. For the nine clusters within the slit, i.e., 1, 2, 3, 4, 10, 11, 12, 13, and 14, the "good" data span the wavelength range 1240–1690 Å, while for the rest of the clusters, i.e., 6, 7, 8, and 9, they span the range 1240–1670 Å.

We adopted the algorithm developed by Tremonti et al. (2001) for deriving the age and mass of each cluster from the best-fit model to the data. The algorithm takes as input the de-reddened cluster spectrum and a grid of Starburst99 models spanning ages from 1 to 20 Myr in step sizes of 0.1 Myr. The goodness of the fit is characterized by χ², where χ² = (o_i − m_i)²w_i/σ²_i, and where o_i represents the observed data for the ith pixel, m_i is the model data, σ_i is the error in the observed spectrum, and w_i is the assigned weight. The weighting scheme eliminates from consideration the interstellar lines and gives the stellar wind lines a weight 1.5 times greater than that of the continuum. Prior to fitting, the resolution of the models was degraded to match the approximate resolution of the observed spectra. The spectroscopic masses were derived by (1) fitting power laws to the continua of the observed and the model spectra of each cluster, where the model corresponds to the best-fit age and a stellar mass of 10⁶ M_☉; (2) computing the ratio of the observed to the model power law at each wavelength point; (3) taking the average of the sum of the ratios obtained in (2); and (4) multiplying the result by 10⁶ M_☉.

The metallicity of M83's starburst is Z ≃ 0.030 with a large error bar (see Bresolin & Kennicutt 2002). We considered theoretical models with metallicities of Z = 0.020 and Z = 0.040.

Chandar et al. (2010) derived the age and extinction for each cluster by finding the minimum χ² when comparing observed UBVIHα magnitudes with those predicted by the S. Charlot & G. Bruzual (2009, private communication) stellar population synthesis models, assuming a Milky-Way-type extinction curve (Fitzpatrick 1999), metallicities of Z = 0.017 and Z = 0.034, and a Chabrier IMF from 0.1–100 M_☉ (Chabrier 2003). Here, the narrowband filter contains both stellar continuum and nebular line emission, i.e., no continuum subtraction was performed for the narrowband image, and the predictions were modeled as line plus continuum.

The mass of each cluster was estimated from its extinction-corrected V-band luminosity and the age-dependent mass-to-light ratio from the stellar population models, assuming a Chabrier stellar IMF and a distance of 4.5 Mpc. The ages and masses derived using this technique are compared to the spectroscopic values below.

On average, the theoretical spectra fit the data better for Z = 0.040 than for Z = 0.020 (on average, the reduced χ² is 1.3 times larger at Z = 0.020 than at Z = 0.040, with a standard deviation of 0.3). Note that some of the differences between the data and the best-fit models are not due to the metallicity choice, but to the quality of the data, which suffers from crowding, as well as from worse spectral resolution than when a narrow slit is used. In any case, on average, the spectroscopic ages and masses at the two metallicities are identical. Hereafter, we only consider spectroscopic results at Z = 0.020, which is the metallicity of the semi-empirical models.

Cluster 14 is the youngest (∼ 3 Myr, average of spectroscopic and photometric ages) and the least massive (photometric mass of about 4 × 10³ M_☉) in our sample. According to Weidner et al. 2010 cluster 14 should have a maximum stellar mass of about 80 M_☉ (i.e., below 100 M_☉). Unfortunately, the spectrum of this cluster is too noisy to reliably pin down its maximum stellar mass. Cluster 1 is the next best cluster in our sample to test the upper end of the stellar IMF and has a mean age of ∼4 Myr and a photometric mass of ∼ 3×10⁴ M_☉. As shown in Figure 8, left panel, the strong Si iv 1400 and C iv 1550 P-Cygni profiles of cluster 1 cannot be reproduced with models having maximum stellar masses of 30 M_☉ or less. Models with maximum stellar masses of 50 and 100 M_☉ fit cluster 1 equally well. Note that Bresolin & Kennicutt (2002) found optical signatures of the presence of WN stars in the region of our clusters 1–3. The P-Cygni profile of He ii at 1640 Å, which is characteristic of WN stars, is barely detected in the FUV spectra of clusters 1–3, but given the point that we make in Section 3.2, this is not surprising. The rest of the clusters with strong P-Cygni profiles of N v, Si iv and C iv are consistent with having formed stars more massive than 30 M_☉. The right panel of Figure 8 shows that the upper mass limit of the IMF is harder to constrain for cluster 10, which is without P-Cygni profiles. Clusters with strong absorption profiles of Si iv and C iv are consistent with having formed at least some B stars, if we assume that the absorptions are not of interstellar origin.

Zoom In Zoom Out Reset image size

Figures 9–11 show the background and reddening-corrected rest-frame spectra of the clusters in our sample, along with best-fit semi-empirical and theoretical models corresponding to a metallicity of Z = 0.020 and a Kroupa IMF from 0.1to100 M_☉. We used the latter models to determine the spectroscopic ages and masses of all clusters, including those without strong P-Cygni profiles of N v, Si iv, and C iv in their spectra. Therefore, we are assuming that clusters without strong P-Cygni profiles formed massive O stars but are older than ∼12 Myr. Another possibility for the latter clusters is that they are younger than ∼12 Myr and did not form very massive O stars.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Here, we compare our ages derived using spectroscopic and photometric techniques. We also compare our ages with estimates from the literature.

Our spectroscopic and photometric ages are compiled in Columns 1–9 of Table 3, which give (1) the cluster ID, (2) and (4) the spectroscopic ages from the semi-empirical and theoretical models, (3) and (5) the reduced χ² value of the best fits to the observed spectra, (6) and (7) the photometric ages from the Z = 0.017 and Z = 0.034 models, and (8)–(13) ratios of the age estimates derived from the different techniques. The last four rows of Table 3 give the minimum, maximum, mean, and standard deviation values for relevant columns. Figure 12 shows the photometric age determinations for cluster 1 (mean photometric age ∼4 Myr) and cluster 10 (mean photometric age ∼19 Myr). They were derived using the technique described in Chandar et al. (2010). Figure 12 also shows the measured V − B and V − I colors for clusters 1 and 10. Considering the mean of the spectroscopic and photometric ages, clusters 1 and 10 are representative of young age (<10 Myr) and older age (>10 Myr) clusters, respectively.

Zoom In Zoom Out Reset image size

The ages estimated from the four different techniques are in excellent agreement. Ages derived by comparing UV spectroscopy with semi-empirical and theoretical predictions are within a factor of 1.5. This is similar to the 1σ uncertainties of ∼1 Myr found previously for young clusters from UV spectroscopy (Tremonti et al. 2001; Chandar et al. 2003). Ages determined by comparing the HST/WFC3 optical colors of the clusters with predictions from stellar population synthesis models at the two metallicities agree within a factor of 2. Finally, the spectroscopic and photometric ages are in excellent agreement (within a factor of 1.2 on average). Table 3 shows that most clusters are quite young, with ages <5 Myr. Clusters 6, 7, and 10, which have the oldest spectroscopic ages, also have the oldest photometric ages. The worst agreement between the spectroscopic and photometric ages is for cluster 8 (difference of a factor of 4). The FUV spectrum of this cluster lacks the strong P-Cygni profiles found in young star clusters, yet the cluster has very blue colors, resulting in a young (∼4 Myr) photometric age.

Comparing with previous work, our ages agree with the photometric ages derived by Harris et al. (2001) based on HST/WFPC2 photometry, except for clusters 6, 7, and 10, which are older than 6 Myr in our case. We found 3.6, 3.4, and 4.1 Myr, for the ages of clusters 1, 2, and 3, respectively (means of spectroscopic and photometric ages), which is in excellent agreement with the ages found by Bresolin & Kennicutt (2002) for region A (all ∼4 Myr), using the same spectroscopic data as here. On the other hand, Bresolin & Kennicutt (2002) determined an age of 7 Myr for their region B, from the EW of Hβ. Region B corresponds to the location of our clusters 7–9. We found mean ages of 15.5 Myr for cluster 7 and 3.4 Myr for cluster 9. As explained in the previous paragraph, we were unable to pin down the age of cluster 8, which is also in region B. Finally, region D of Bresolin & Kennicutt (2002) corresponds to a cluster that we did not include in our sample, due to its truncated spectral trace.

We conclude that our spectroscopic and photometric age estimates are within the expected uncertainties for 12 out of 13 of the clusters. The clusters are ∼3–20 Myr old and were not all formed at the same time. In addition, we do not find any compelling evidence for an age gradient along M83's arc-shaped starburst, but rather see clumping of clusters of similar age with very young clusters dominating the northernmost and southernmost portion of the starburst that we studied, and somewhat older clusters found between. This seems contrary to the finding of Puxley et al. (1997), Díaz et al. (2006b), and Knapen et al. (2010) that there is an age gradient along the starburst's arc, and consider regions that are large compared to our HST resolution.

Here, we compare the masses estimated for our clusters using spectroscopic and photometric techniques. These are compiled in Columns 1–7 of Table 4, which give (1) the cluster ID, (2) and (3) the masses derived by fitting the semi-empirical and the theoretical models to the FUV spectroscopy, (4) and (5) the masses derived from the optical photometry assuming metallicities of Z = 0.013 and Z = 0.017, and (6)–(11) ratios of masses obtained from the different techniques.

Column 6 of Table 4 shows that the spectroscopic masses from the semi-empirical models are in very good agreement with those from the theoretical models, and that the semi-empirical masses are systematically larger by a factor of 1.5 on average than the theoretical masses. The latter is because the semi-empirical models have systematically larger values of the mean ratio described in (2) of the last paragraph of Section 4.1. Column 7 of Table 4 shows that the photometric masses are not significantly affected by metallicity. Finally, Columns 8–11 show that the spectroscopic masses are systematically larger than the photometric masses by a factor of 4–6 on average. We suggest that this is the result of the high background contamination of our wide slit and slitless crowded spectra.

Optical photometry provides more leverage for determining the cluster mass. This is because photometry captures light from low- and intermediate-mass stars, which contribute more by mass to the IMF than massive stars. In addition, the total light from the cluster is estimated better from the photometry. Therefore, we consider our photometric masses (M_p) as more reliable than our spectroscopic masses.

Our most massive cluster has M_p between 2 and 3 × 10⁵ M_☉, comparable to the virial mass of the ionizing cluster of 30 Doradus, NGC 2070 (4.5 × 10⁵ M_☉; Bosch et al. 2009), in the LMC. According to Larsen (2010), young star clusters with masses larger than 10⁵ M_☉ can last an age comparable to or exceeding the age of the universe. Therefore, this massive cluster in M83 could be a globular cluster progenitor, while two other clusters, which have M_p ≈ 10⁵ M_☉, may also survive. Cluster 14 has M_p ≈ 4 × 10³ M_☉ and appears to be the most compact in size in the FUV image. Finally, the rest of the clusters have M_p of a few×10⁴ M_☉.

Note that our older clusters tend to be more massive than the younger ones. This is probably an observational effect. Older clusters at lower masses are harder to detect in the FUV.

Columns 1–5 of Table 5 give (1) the cluster's ID; (2) β_i, the power-law index of the continuum over the wavelength range 1240–1670 Å, corrected for a foreground Milky Way reddening of E(B − V) = 0.06; (3) the intrinsic reddening of the cluster derived from the FUV spectra; (4) the intrinsic reddening derived from the optical photometry; and (5) L₁₅₀₀, the intrinsic luminosity of the cluster at 1500 Å, derived by adopting a distance of 4.5 Mpc to M83.

Overall, the agreement between the reddening values derived from the spectroscopic and the photometric analysis is fair. For clusters 3 and 14, which have dusty local environments (see Figure 1), the agreement is good. Indeed, both clusters have large intrinsic spectroscopic and photometric reddenings. The agreement is also good for clusters 11 and 12, which appear to be the least reddened by both measures of the intrinsic extinction. However, for cluster 1, E(B − V) = 0.18 from the spectroscopy, while E(B − V) < 0.06 from the photometry. Cluster 3 has the largest intrinsic value of L₁₅₀₀, while clusters 4, 13, and 14, which appear faint in the optical images, have the lowest values of L₁₅₀₀.

In addition to compact star clusters, there is UV emission from the diffuse field star population. We obtained the spectrum representing the diffuse stellar field by summing spectra extracted from the long-slit exposure, at the locations indicated by the dashed lines in Figure 3. We performed no background or reddening correction for the field spectrum. In addition, we did not shift the spectral traces in the dispersion direction, due to the lack of narrow interstellar lines that could be used for reference. Figure 13 shows the result. The spectrum lacks the P-Cygni profiles of N v, Si iv, and C iv, which are signatures of O stars, but it shows strong absorption components in these features, which indicates the presence of B stars. This is consistent with a picture where (massive) O stars form mostly in clusters but not in the field, and where the field is dominated by B stars from clusters which disrupt rapidly, on timescales of ∼10 Myr (Tremonti et al. 2001; Chandar et al. 2006; Pellerin et al. 2007).

Zoom In Zoom Out Reset image size