A SPECTROSCOPIC CENSUS OF THE M82 STELLAR CLUSTER POPULATION^*,†

The results presented in this paper are based on spectroscopy obtained with the Gemini-North Multi-Object Spectrograph (GMOS-N) on the Gemini-North telescope, as part of observing program GN-2006A-Q38 and the follow-up program GN-2007A-Q21 (PI L. J. Smith; hereafter Q38 and Q21). Overall, two multi-object slit masks were designed and included a total of 54 slits, from which we extracted 62 spectra in the range λλ3700–6500. Figure 1 marks the positions of the slits on HST/Advanced Camera for Surveys (ACS) imaging of M82 (Mutchler et al. 2007) and presents all the cluster spectra used in the subsequent analysis.

Zoom In Zoom Out Reset image size

The source selection was performed within the bounds of the original observing program (Q38). The candidate young massive clusters (YMCs) were selected based on Gemini Multi-Object Spectrograph (GMOS) pre-imaging after cross-identification with HST imaging, with the aim of choosing the brightest isolated clusters, spanning most of the optical extent of the disk. Starting from a masterlist of several hundred candidates, we built two multislit masks (one for each observing program), with a standard 075 slit width, while the slit length was varied to optimize the overall arrangement. In all cases, we attempted to apply a minimum slit length of 65, in order to provide adequate sky for background subtraction. In Q38, we fell below 65 in a few cases, to accommodate more clusters along the slit; this, however, complicated the aperture extraction procedure. For that reason, in Q21 we did not sacrifice the slit length in favor of cluster number, particularly since the mean magnitude was fainter compared to Q38.

The GMOS observational setup was identical for Q38 and Q21. We used the B600 grating (0.9 Å px⁻¹) and the three GMOS CCDs in 2 × 2 binning mode. The observations were executed in pairs of pointings with different central wavelengths (5080 Å and 5120 Å) in order to cover the inter-chip gaps. A brief journal of observations is given in Table 1. The total exposure time for each data set was eight 1800 s exposures for a total on-source integration time of 4 hr. CuAr arc exposures and Quartz–Halogen flat fields were taken in between target exposures, and bias frames were taken as part of the Gemini baseline calibrations (GCAL).

Table 1. Summary of Spectroscopic Observations taken with Gemini-North

Dataset	Dates of observation	Seeinga	No. slits	Exposure Time
GN-2006A-Q38	2006 Apr 5	08	39	8 × 1800 s
GN-2007A-Q21	2007 Jun 4–8	045b	15	8 × 1800 s

Notes. ^aThis is the Gemini "image-quality" (IQ) at 5000 Å. This parameter accounts for wind and telescope performance effects as well as atmospheric seeing. ^bFor all nights except 2007 June 8, when the IQ was 08.

Download table as:  ASCIITypeset image

The data reduction was performed using standard IRAF⁷ and purpose-designed Gemini-IRAF routines. In brief, this consisted of bias-subtraction, flat fielding, and the combination of a three-CCD image into a single frame, while correcting for the variation in detector quantum efficiency. We used the obtained CuAr frames for the wavelength calibration, which resulted in satisfactorily low residuals (typically less than 0.2 Å). As Gemini does not have a functioning atmospheric diffraction corrector, we corrected for differential atmospheric refraction using a purpose-built IDL routine written by B. W. Miller (private communication; this routine is based on Filippenko 1982, where the effects of atmospheric differential refraction are discussed). Finally, all eight available exposures of each target were extracted and then flux calibrated using exposures of standard stars Wolf 1346 (observed 2006 May 25) and HZ 44 (observed 2008 Feb 18) for Q38 and Q21, respectively.

In summary, the two masks contained 54 slits (see Table 1) and the total number of sources observed was 62, of which 28 have a signal-to-noise ratio (S/N) above 20 (conservatively measured in two continuum windows covering λλ4500–4800 and λλ5000–5500). We present spectra for 49 clusters in this paper, and omit the remaining 13 on grounds of low data quality. The spectral resolution of the Q38 and Q21 cluster sample was measured to be 3.5 Å and 3.7 Å, respectively from the CuAr spectra, but we assume a common resolution of 3.7 Å for the sake of simplicity. The spectral range of our data is ≈λλ3700–6500, which includes the Hβ and Hγ lines, our main age indicators, and the Na  i  D doublet, which we use as a tracer of interstellar gas RVs. The exact wavelength range varies with each slit, depending on its positioning with respect to the M82 major axis, which defines the dispersion axis. Our effective wavelength range for all slits is λλ4000–6000; this is set by the poor sensitivity of GMOS below 4000 Å and an uncertain flux calibration above 6000 Å. Figure 2 demonstrates these features on the representative high S/N spectrum of cluster 112.2. Positional and other basic information for all clusters can be found in Table 2 (the coordinates of our sources have been converted to the HST Word Coordinate System, as the mosaic imaging is publicly available). The cataloging convention we have adopted for cluster naming follows the increasing slit number in the masterlist compiled during the initial source selection. The list is not continuous (as more sources were initially selected than could be accommodated), but the increasing number does indicate the position of a cluster along the M82 disk—numbers count from west to east, along the major axis. In addition, several slits include multiple sources; in these cases, we refer to clusters by the slit number with a decimal part, to indicate the increasing number of the source (again, counting from west to east). Our sample spans most of the extent of the optically visible disk, which is approximately 6 by 2 kpc, as seen in HST imaging.

Zoom In Zoom Out Reset image size

We also make use of imaging data of M82 to derive cluster colors and deduce information about their surroundings. We use the publicly available Hubble Heritage ACS mosaic imaging of M82 (for a full description, we refer the reader to Mutchler et al. 2007), which covers the galaxy disk, nucleus, and outskirts in six pointings and in three filters, F435W, F555W, F814W. These are roughly equivalent to the standard Johnson BVI, so we use this notation throughout this paper. Note, however, that we do not perform any conversion between the two systems. All of our observed clusters lie within the field of view (FOV) of the ACS imaging. The spatial scale of the images is 0.05 arcsec px⁻¹, and one pixel corresponds to ∼0.9 pc at the distance of M82.

We used the HST images to perform aperture photometry on all sources in our spectroscopic sample. For most disk clusters, we applied an aperture radius of 10 pixels to avoid contamination by nearby objects, with a 2 pixel wide sky annulus placed at 12 pixels. As most of our sources are of considerable size, we calculated aperture corrections for each filter as −0.34 mag for the B and V bands and −0.45 mag for the I band, based on the flux difference between 10 and 30 pixel apertures for isolated clusters. After close inspection of each source individually, it was deemed necessary to limit the aperture size in some cases, due to the presence of contaminating neighboring sources. Thus, we applied 5 pixel apertures with the same annulus for eight disk clusters as well as all sources in the nuclear region (in the nucleus, the sky annulus was placed between 7 and 9 px), and calculated the corrections to 30 pixels as −0.94, −0.90, and 1.04 mag for the B, V, and I bands, respectively. While these aperture corrections are significant and dependent on cluster size, the cluster colors should be largely unaffected. The photometric results are presented in Table 2; this includes cluster reddening, which is discussed in Section 3.2.

In this section, we present our methodology for deriving cluster ages and illustrate our techniques through example fits. We employ the Cumulative χ² Minimization method (CCM) presented in Paper I to obtain ages for the full sample of 49 star clusters in the disk and nuclear region of M82. This method is based on the goodness of fit between observed cluster spectra and evolutionary population synthesis (EPS) models. We chose to compare our spectra to the González-Delgado et al. (2005, hereafter GD05) model library, a range of theoretical models specifically compiled for young star clusters. The major advantages these models present over more established empirical cluster SEDs (e.g., Bruzual & Charlot 2003) are the high spectral and temporal resolution (0.3 Å and a few Myr for the first 100 Myr). We used the full range of models (4 Myr to 17 Gyr) compiled using Padova isochrones and a Salpeter IMF. We have assumed a solar metallicity following the findings of McLeod et al. (1993) for the M82 ISM.

Before the fit can take place, we rectify both observed spectrum and template (using two continuum windows adjacent to the absorption line) and smooth the models by a factor of three and then re-bin them to the wavelength scale of the observed spectra. In order to obtain a first estimate of the cluster age, we apply an approximate RV correction on the observed spectra (matching the Balmer line profiles by eye; this calculation is later refined). We use the lower Balmer series line profiles (Hβ and Hγ) to perform the fits, and take care not to include any emission from ionized gas in the line of sight (which is plentiful in this gas rich, edge-on galaxy). This makes our age determination technique largely independent of the effects of foreground gas and dust.

In brief, the CCM calculates the χ² for each point in the fit, and picks the best fitting age as the model with the lowest "reduced" χ², expressed through χ²_ν = Σχ²/d (note that we will omit the subscript ν hereafter); d is the number of degrees of freedom, taken as the number of points included in the fit. This approach follows the reasoning outlined in Lampton et al. (1976). Our fitting technique is therefore sensitive to the overall line profile (core and wings) rather than just the line depth. This approach overcomes both the line strength degeneracy (discussed in Section 1) that exists between models of age ∼200 ± 100 Myr and ∼600 ± 100 Myr, and the lack of information on the line core where the superimposed emission is strong or complex. We therefore obtain two measurements of the cluster age, based on the fits on the Hγ and Hβ features. As the two measurements do not always agree perfectly (usually due to the lower S/N in the Hγ spectral region), we examine each case individually. In a few cases, one of the two measurements is excluded from the fit; we apply this principle by enforcing a minimum acceptable S/N ratio of 10 (this applies to 12 clusters in the sample).

The uncertainty on a given fit is defined as the range of ages that lie within 1σ of the minimum χ²—quantified by the criterion χ²_i ⩽ χ²_min + N, where N is the number of free parameters in the fit (Lampton et al. 1976). In this case, the only varying parameter is the age of the model, therefore N = 1. This process is slightly complicated by the fact that a large number of our adopted ages consist of the statistical mean of the results of two χ² fits. In these cases, we combine the probability distributions derived for each fit (simply by adding them) and use a value of N = 2 to define the 1σ confidence region. We stress that, while this error determination is not statistically flawless, it provides an adequate description of the range of models that can plausibly describe the spectra. This range therefore matches what one might pick "by eye." This statistical analysis was also successfully used in Bastian et al. (2008).

Our fitting method is demonstrated in Figure 3 for three clusters of varying luminosity and S/N. We present a BVI color-composite image (the spatial scale is a square of side 200 px or 175 pc) and the spectrum of each cluster, with the best-fit model overplotted. We also show the probability distributions of all available models as a function of age below each spectrum. We note that, although not demonstrated, the Hγ and Hβ fits do provide overlapping χ² distributions (i.e., equivalent fits). In the top panel, we show a low S/N spectrum (cluster 19, S/N < 20), where the Hβ region provides the only reliable fit. The middle panel shows the Hγ profile in a medium S/N (∼25) spectrum. For this cluster, we are able to obtain two age measurements in good agreement. In the bottom panel, we show the high S/N (∼40) spectrum of cluster 112.2, where the Hβ and Hγ measurements give the same age. We note that these plots do not present a subset of high-quality fits, but are meant to provide an overview of all problems faced by our fitting method, be they due to low S/N, embedded/blended emission, or related to the template fitting process.

Zoom In Zoom Out Reset image size

One immediately noticeable feature in these fits is the existence of two best-fit troughs. While the two fits are statistically equivalent, there is a significant volume of evidence that favors the younger ages. This is based on the comparison to ages independently derived through UBVI photometry, as well as a number of considerations of the star formation history of M82 that will be discussed in Section 4.3. The age measurements are shown in Table 3, where we have listed the acceptable range as the extent of the preferred best fitting trough.

A final consideration concerns the effect of our choice of model library. We used the GD05 templates to achieve the highest quality fits, given their high spectral and temporal resolution. As a consistency check, we fitted some of the high S/N clusters in our sample with BC03 (Bruzual & Charlot 2003) models of similar characteristics (same isochrones, metallicity, and IMF) and obtained very similar results. For instance, cluster 91 was found to have an identical age under GD05 and BC03. Further in support of our choice of templates is the fact that, under BC03, the best-fit troughs are represented by very few consecutive models (perhaps three or four), whereas with GD05 the χ² probability distribution appears smooth and therefore more reliable.

Having measured the age of each cluster, we can now determine their RVs. We used the IRAF FXCOR routine to cross-correlate the best-fitting model with the cluster spectrum. We also manually measured the velocities of the Na i D lines (λλ5890, 5896) which originate from cold interstellar gas along the line of sight (LOS). We transformed the measured cluster velocities to the heliocentric frame of reference and corrected for the systemic velocity of M82 (200  km s⁻¹; McKeith et al. 1993). The final RV values vary between 0 and 220  km s⁻¹, with associated Na  iD velocities in the range 5–200  km s⁻¹ (in absolute value). These measurements are also presented in Table 3. We also used the RVs to refine our age measurements: we repeated the age fitting process, this time applying a more accurate Doppler shift according to the derived RVs. This allowed for a better alignment between spectrum and model, which had been estimated in the first iteration. This second iteration of the age fits did not alter any of the best-fitting ages.

We divide our sample into disk clusters and nuclear region knots, where the nuclear region is defined as a projected radius of ∼200 pc about the dynamical center of the galaxy (see McKeith et al. 1993, for exact location). Note here that we use the terms "cluster," "knot," and "cluster complex" interchangeably for these nuclear region objects; they are probably not individual clusters but cluster complexes (physically associated) or asterisms (chance projections). We find all disk clusters to have ages between ∼30 and 270 Myr, with a mean (and median) value of ∼140 Myr. For the knots in the nuclear region (slit numbers 67 and 78), these values range from 7 to 30 Myr with a mean of 10 Myr. Unfortunately, it is not possible to obtain accurate age dating in the nuclear region, as the fits are dominated by strong emission lines. This restricts the fitting to a spectral range that hardly covers the Balmer absorption lines. The CCM picks very young ages due to the flatness of these spectra; in the case of knots in slit 67, the fit favors the very youngest models in GD05 library, as all models above a few 10 Myr feature noticeably stronger metal lines.⁸ Further evidence of the young age of these clusters is offered by the strong, embedded emission lines. Such strong features centred at the same wavelength as the absorption can only be associated with the clusters. The clusters in slit 78 display a similar appearance, however, with some more structure (i.e., they are not as "flat" as the sources in slit 67), thus they are assigned slightly older ages (in the third and fourth decade of Myr). In addition, we find offset emission in cluster 78.1 perhaps indicating that it is not situated deep in the galaxy like the rest of the "nuclear" knots. Finally, our sample also contains two halo objects, which we have dated at >8 Gyr; we discuss these candidate old populous clusters (OPC) in Section 3.3.

We use the photometric measurements to derive cluster colors (B−V and V−I) and present these results in Figure 4. Here, we compare the observed cluster colors with GALEV evolutionary models for SSPs (Anders & Fritze-v. Alvensleben 2003). As for our chosen SEDs, the photometric models use the Padova isochrones, a Salpeter IMF, and solar metallicity.

Zoom In Zoom Out Reset image size

This plot can be used as a rough diagnostic of age but by no means an accurate one, given the lack of information in the Balmer jump region and the consequent degeneracy in the BVI baseline. From this it becomes obvious that there is no clear distinction when age-dating is performed photometrically between a cluster of, e.g., age 4 Myr and 2.5 Gyr. Given some a priori knowledge of extinction, only some rough deductions can be made.

More importantly, given a strong prior knowledge of age (which we possess from spectroscopy), colors can be used to estimate the reddening for each cluster. To do this, we used a simple code to transfer each cluster color along the extinction vector until it is close to the intrinsic (model) color at the derived spectroscopic age. Assuming a standard Galactic extinction law with R_V = 3.1 (Rieke & Lebofsky 1985), we measured extinctions in the M82 galaxy disk in the A_V range 0.4–3.2, with a median value of 1.3 mag.⁹ The corresponding range of values for the nuclear region is 1.4–3.5 with a median of 2.2 mag. To derive approximate errors for the A_V values, we have adopted an age range of 50 Myr (a representative value for the error in age measurement) and calculated the corresponding extinctions.

All values are listed in Table 2. Note that the foreground extinction from the Milky Way is ∼0.5 mag in the V band, so the lowest values we derive correspond to roughly no extinction from material associated with M82. Interestingly, the measured values do not reflect the large amounts of extinction one might expect to find in this galaxy—reaching 15 mag, as we mention in the introduction. This is due to our sample, which sought out the brightest possible sources to secure the acquisition of high S/N spectra. As a result, no highly extinguished sources were included.

Our spectroscopic data set includes two halo objects located at a projected distance of ∼400 pc to the south of the eastern galaxy disk, selected from the initial masterlist as candidate OPCs. We fit those objects (nos. 152 and 139) with GD05 models of $\frac{1}{5}\hbox{Z$_\odot $}$ (a value characteristic of old globular clusters) and find best-fit ages of ≳8 Gyr. In the case of 152, we also find a peculiar/halo RV, as it lags significantly behind the disk population. Unfortunately, the low S/N of cluster 139 does not allow for a certain age fit or a high-confidence RV cross-correlation, therefore only 152 was investigated further (see Section 4.6). These clusters have not been observed in the past (they are not included in the list by Saito et al. 2005, who studied a sample of GC candidates around M82 and reviewed the case for GCs in the galaxy) and make very interesting targets for further study.

The designed multi-object mask features two long slits placed over the M82 nuclear region, covering a total of seven knots. The precise location of these sources is not easy to pinpoint, given the apparent complexity of the nuclear region. We therefore reserve the possibility that they lie near the surface of the galaxy and are projected onto the nuclear region (this matter is revisited in Section 4.5). In addition, the extracted apertures are likely to contain blended light from more than one star cluster, therefore we do not treat these as observations of individual sources. The activity in the nuclear region is key to understanding the starburst history of M82. A recent study by Barker et al. (2008), found no evidence of cluster formation over the last 4–6 Myr in the nuclear region. However, we find Wolf–Rayet (WR) features in the spectra of three clusters in the region (maybe cluster complexes; sources 1, 2, and perhaps also 4 within slit 67), which are indicative of such young ages. More specifically, we find evidence of He ii λ4686, C iii λ4651, and possibly C iv λ4659, N iii λ4634 − 4640 and N v λ4603 − 4619; all these features form part of the "blue WR bump" at 4600 − 4700 Å (Sidoli et al. 2006; Bibby et al. 2008). In Figure 5, we provide a finding chart for slit No. 67 and we present the spectra of apertures 1 and 2, indicating the locations of the proposed WR features. This detection needs to be confirmed with higher S/N in this spectral region, and if proven to be true, it will classify M82 as a WR galaxy.

Zoom In Zoom Out Reset image size

We stress at this point that the exact source of these emission lines is uncertain. They could originate from (or be enhanced by) scattered light from the nuclear region, rather than the observed knots. In order to avoid contamination by nebular features in our analysis, we have extracted these spectra in narrow apertures. Despite having taken that precaution, we reserve the possibility that the feature at λ4658 is in fact nebular [Fe iii]. In essence, it is not the source of this emission that we report here, but simply its existence. Even though our data are not suitable for an in depth analysis of the detected lines, they provide evidence that the nuclear starburst is still active. WR features only appear in the youngest stars (2–6 Myr), and are a reliable age-dating tool for young star clusters (Sidoli et al. 2006; Bastian et al. 2006). More evidence for an ongoing nuclear starburst was offered by Smith et al. (2006), who found an age of 6.4 Myr for cluster A1 in the nuclear region. The high mass of this cluster (∼10⁶ M_☉) implies the existence of a fully sampled cluster mass function at that era, and therefore the contemporary formation of a large number of clusters—such a massive cluster represents but the "tip of the iceberg" in terms of the underlying distribution. Hence we confirm that star/cluster formation is currently continuing in the nuclear region of M82.

Additionally, it was proposed by Barker et al. (2008) that while stars are forming in the nuclear region of M82, star and cluster formation have become dissociated in this region. Our detection of young clusters in the same region studied by Barker et al. indicates that star and cluster formation is still ongoing. Furthermore, Bastian (2008) showed that cluster A1 fits the relation between the brightest cluster in a population and the SFR of the host galaxy (a relation which extends from quiescent to extreme starburst galaxies). This suggests that star and cluster formation is proceeding normally in the nuclear region—subject to our interpretation that these clusters reside in the nucleus (see Section 4.5 for further discussion).

Cluster colors are presented in Figure 4; this plot works well in confirming the timescale for the starburst, with many data points distributed closely about the 250 Myr extinction vector. In addition, the nuclear region clusters are intercepted by younger age vectors, and the OPC candidates by vectors in accord with their spectroscopic ages. There are two isolated pairs of clusters (nos. 113 and 135, 156 and 157) that do not agree with the spectroscopic ages; of those, cluster 113 appears to have an underestimated flux in blue light, which we interpret as the result of differential extinction across the face of the cluster (this will shift the data point redward in blue bands preferentially; see Bastian et al. 2007, and Paper I for more discussion of such cases). Cluster 135 appears diffuse in the ACS imaging, and could therefore have disproportionately underestimated fluxes. Cluster 156 has an underestimated B-band flux, as it resides behind a thin, filamentary dust lane; cluster 157 lies near the edge of the optically visible disk, at the boundary between the bright, blue disk and wide, dark band to the north of region B, and is therefore differentially extinguished.

We present cluster reddenings in the bottom panel of Figure 8, where in this case the values have been corrected for foreground galactic extinction (A_V = 0.53; Schlegel et al. 1998). The values presented elsewhere in the paper do not incorporate such a correction.

Having obtained accurate photometry (magnitudes and extinctions) and ages for each star cluster in our spectroscopic sample, we derived their photometric masses (see Table 3 for results). We find average and median values of 2.5 × 10⁵ and 7.2 × 10⁴ M_☉. According to our photometric mass estimates, the most massive individual clusters in our sample are M82-F and H (nos. 51 and 91 on our list) and no. 152 (the candidate OPC), with values of 4.5 × 10⁶, 2.1 × 10⁶, and 2.1 × 10⁶ M_☉. The total baryonic mass contained in the observed clusters sums to 1.1 × 10⁷ M_☉.

Based on these mass measurements, we can derive an estimate for the total mass in clusters for the entire galaxy. We generated an array of simple stochastic populations of star clusters, governed by a power-law mass function of index −2, with a minimum mass of 50 M_☉ and no upper mass limit. As we know of five clusters in M82 with mass above 10⁶ M_☉, we then varied the number of clusters in the population, N, until the average (over 100 runs) number of clusters with mass greater than 10⁶ M_☉ is 5. From this, we find a total cluster mass for M82 of ∼10⁸ M_☉, the mass of a stochastic cluster population with N = 10⁵. This is an approximate lower limit, as there may exist more star clusters in M82 with mass greater than 10⁶ M_☉ that have not been observed due to extinction.

We repeated this exercise analytically, by calculating the missing mass according to a power-law mass function of index −2. Given five clusters with mass of the order of 10⁶, we integrate the mass function and extrapolate a mass of ∼6 × 10⁷ M_☉ for the observable population. Taking this consideration further, and assuming that: (1) all the clusters we have studied formed during the most recent starburst; and (2) that during the starburst, 10% of stars formed in clusters that are still bound in the currently observable era, then we estimate the mass of the young field star population at ∼5 × 10⁸ M_☉. This value is an underestimate, as it was calculated based on the observable cluster population. In addition, recent studies by Larsen (2009) and Gieles (2009) find that the underlying cluster mass function is likely to be a Schechter function, with a limiting mass of M_* ∼ 10⁶ M_☉. This is another factor that may lead to a false estimate of this total mass, as we have integrated over a power law with no imposed maximum mass. While this value represents a lower limit, we note that the survival of a significant number of massive clusters to ∼150 Myr argues against a prolonged era of mass-independent cluster disruption.

Using the spectroscopically derived ages, we have plotted the age distribution of all clusters in the sample in Figure 7, where the solid and dashed lines represent the disk and nuclear region cluster populations, respectively. We chose to bin the data in 50 Myr bins to reflect the errors in some of the measured ages; also, the first bin has half the size of the rest, to mirror the lack of disk clusters younger than ∼40 Myr. We note that the overall shape of the histogram is maintained when a smaller bin size is applied, however the low number of clusters introduces uncertainties. The plotted error bars account for Poissonian sampling and go as $\sqrt{N}$ for each bin. As the spectroscopic data sample the entire galaxy disk and nucleus, this can be treated as a rough indicator of the M82 star formation history (SFH), where the number of clusters formed reflects the mean SFR at that epoch. Naturally, the numbers cannot be treated as correctly "to scale" as our data set is not complete.

The histogram confirms some of the "milestones" we have discussed in our recent work on the M82 cluster population (Smith et al. 2007; Bastian et al. 2007, and Paper I). First, by placing the burst between 200 and 250 Myr, it reflects the timescale for the last encounter with M81 at 220 Myr (Yun 1999), the event that most likely triggered the disk starburst and stripped away part of the disk. Note that this marks the onset of the disk starburst, not the period of maximum formation that follows closely. This sets another observational constraint on the age of the disk starburst (this was discussed in Paper I and Smith et al. 2007, for region B). Finally, the first bin reflects the ongoing nuclear starburst.

In addition, the age distribution peak for the full disk sample (that we place roughly at 150 Myr) is consistent with the 156 Myr distribution peak for region B, as derived in Smith et al. (2007) based on a sample of 35 UBVI age dated clusters. This supports our interpretation of region B as a regular part of the galaxy disk (Smith et al. 2007, Paper I).

In Section 3.1, we discussed the possibility of a degeneracy in our cluster age dating technique. This is expressed through the existence of two best-fit troughs (occurring in the majority of cases), roughly placed at 200 ± 100 Myr and 600 ± 100 Myr. This degeneracy in line profile arises because of the evolution of the Balmer line profile both in strength and width. These line properties fluctuate in the first few hundred Myr, before commencing a more or less monotonic evolution. More specifically, stellar Balmer lines gradually broaden until the age of ∼300 Myr and become narrower once again at ∼600 Myr. This makes it difficult to distinguish between spectra either side of this 300 Myr mark using a statistical method. One deduction can therefore be made with certainty: that the clusters we observed are all younger than ∼600 Myr. This adds to the findings of Paper I, where we concluded that the population of region M82-B is not as old as previously believed (in excess of 1 Gyr, as presented in de Grijs et al. 2003, and consequent works). This still does not exclude the possibility of a broader age distribution for M82, possibly stretching to ∼600 Myr.

In order to discern the better fit let us consider the existing evidence. First, the aforementioned study presented in Smith et al. (2007) found the (independently derived) age distribution to be extremely similar to the one presented in Figure 7: virtual inactivity prior to the burst at ∼250 Myr and a gradual quenching in more recent times. In addition, this distribution is centred at ∼150 Myr and finds no clusters older than ∼300 Myr. Second, adopting the older age range in these cases results in an extended distribution with a gap between ∼300–500 Myr; this cannot be explained without employing complex scenario such as multiple bursts.¹⁰ In addition, recent work by McQuinn et al. (2009) suggests that a starburst should last no longer than ∼400 Myr, therefore advocating against an extended distribution for the starburst cluster population. Finally, a note on the method itself: the two-minimum structure does not always exist in both the Hβ and Hγ fits: for instance, cluster 112.2 (presented in Figure 3) only shows a double minimum in the Hβ fit. In the Hγ, fit there is only one trough, describing the same age range as the Hβ fit.

In light of this evidence, we adopt the younger ages for the all clusters where this degeneracy is present.

Incompleteness is an issue that needs to be addressed in any study of star clusters in extragalactic environments; an analysis that is applied to a fraction of a population can lead to misinterpretations (as treated in Smith et al. 2007; Gieles et al. 2007). In this case, the lack of clusters with age greater than ∼300 Myr could be the effect of such an observational bias. The completeness of our sample was estimated by inspecting the age–mass diagram of the clusters and by adopting a conservative detection limit of m_V = 21 (see Figure 6). We find that if a significant population of clusters with ages between 300 Myr and 1 Gyr and masses above ∼10⁴ M_☉ existed, it would have been included in our sample (assuming that they formed with the same intensity as their younger counterparts). We therefore confirm that the decrease in the number of clusters with ages greater than ∼300 Myr (see Figure 6) is real and not a result of incompleteness. In addition, the lack of older clusters in the photometric (UBVI) sample presented in Smith et al. (2007, where the limiting magnitude was fainter) also argues for a statistically viable sample.

Zoom In Zoom Out Reset image size

Thus we conclude that the full disk of M82 (mostly likely including the nucleus) entered an era of increased star/cluster formation rate no more than ∼300 Myr ago, which continued in the disk until at least τ ∼ 50 Myr, and is ongoing in the nucleus. Furthermore, no area in the galaxy, other than possibly the nuclear region, appears to have a cluster formation history distinct to the rest of the disk. Unfortunately, we cannot derive the absolute cluster formation history for M82 from this data set due to sample selection and detection limits (e.g., Bastian et al. 2005) and uncertain knowledge of the cluster disruption timescale for this galaxy (e.g., Gieles et al. 2005; Lamers et al. 2005). The bottom panel of Figure 7 combines the age and spatial distribution of clusters and clearly shows the scatter of all disk cluster ages about 150 Myr, and the very young ages of all nuclear clusters. It also shows the lack of young clusters at large galactocentric distance: beyond ∼2 kpc, no clusters lie significantly below the distribution mean. This is unlikely to be a selection effect, as the brightest clusters were chosen for this programme, and these are most likely to be the youngest ones.

Zoom In Zoom Out Reset image size

We will now use the presented age measurements to investigate the SFH of M82. Although the sample is incomplete, star cluster ages can still be used as rough indicators of the cluster formation rate (CFR) of the galaxy.

The observational evidence for SF activity in the disk is as follows: there appears to be little or no recent (in the last Gyr or so) SF, prior to the onset of the starburst. The starburst itself is observable in the increased cluster formation at τ ∼ 200 Myr. The overall CFR in the disk appears to have dropped steadily since the peak starburst epoch at ∼150 Myr, reaching an apparent hiatus ∼50 Myr ago. However, given the errors inherent in this analysis (mainly due to incompleteness), we cannot firmly establish whether the CFR did indeed decrease or if it remained stable. Furthermore, our sample contains no evidence for the existence of young clusters in the outer regions of the galaxy disk, with younger disk clusters being situated closer to the nucleus. This deduction can be assumed to be true with some degree of confidence, as a young cluster forming region would shine bright in the outer regions, where the extinction is below average. Also, our observations were planned in such a way as to include the brightest (and therefore youngest) available sources; the ages of these clusters place their formation in the first ∼100 Myr after the onset of the burst. These observations pose questions about the mechanisms that maintained the M82 starburst over the past ∼250 Myr.

The nuclear burst appears at first glance to offer a simpler picture, with zero activity for most of the burst, and intense cluster formation in the current era. However, incompleteness effects are very grave in the nucleus, the result of both technical/observational and physical factors: first, the nuclear region spreads over a mere ∼200 pc, where only two slits could be accommodated. Second, older clusters will be outshone by the very youngest clusters and cluster-forming clumps. In effect, the nucleus may have been producing clusters at a steady rate over the course of the disk starburst, as material was funnelled into the regions around the bar. Furthermore, the precise location of the observed knots is not simple to pinpoint in this optical data set. We discuss this further in the following section.

Having considered this body of evidence, we propose the following scenario for the SFH of M82: following a prolonged era of low activity in the disk, the SFR increased rapidly, probably triggered by the encounter with M81, some ∼220 Myr ago. As the starburst spread across the disk, the SFR increased and reached a maximum soon after (τ ∼ 200 Myr). During this time (starting at the time of the burst), material started to slowly gravitate toward the nucleus (as proposed by Förster Schreiber et al. 2003). This withdrawal affected the outermost regions first and continued inward, causing the SFR to drop gradually, as fewer and fewer regions were forming stars intensely. This process continued until τ ∼ 50 Myr, when disk activity dropped to a similar level as before the starburst. The nucleus may have entered an era of high SF along with the disk and maintained this high SFR throughout, to the present day, fed by the continuous supply of material, channelled in by the bar.

We present the measured cluster RVs (derived by cross-correlation with the best-fitting spectroscopic models) and line-of-sight neutral interstellar gas RVs (derived from the Na  iD doublet line at 5890/5895 Å) in Figure 8. For comparison, we have also plotted the major axis RV measurements derived from the near-IR Ca ii stellar absorption and Pa(10) and [S iii] nebular emission lines taken from the literature (McKeith et al. 1993; Greve 2004).

Zoom In Zoom Out Reset image size

Our observations extend the major axis coverage of RV measurements significantly further than previous studies, although we caution that our data are not strictly "major axis" measurements since our sample includes clusters located both above and below the disk midplane, at heights up to ∼500 pc. The rotation curve, as traced by the Ca ii, Pa(10) and [S iii] data, is a composite of a steep inner section tracing the nuclear bar (Wills et al. 2000; Greve 2004; Westmoquette et al. 2007), two flat (or gently rising) sections tracing the main disk (hereafter the flat part of the curve), and an intermediate transition region.

In general, the cluster RVs are consistent with the Ca ii measurements throughout the disk. This is in itself quite interesting considering the large range in heights above and below the disk sampled by our data set. There are, of course, a number of notable exceptions: the off-disk cluster 152 (at ∼−2000 pc) has a significantly lower RV than the stellar content and is discussed below as a candidate OPC. Clusters 108, 112.2, 125, and 126, located between −500 and −900 pc, have RVs 60–100  km s⁻¹ higher than the corresponding Ca ii measurements, and neighboring clusters. Besides their discrepant RVs, this group of clusters have no other distinguishing characteristics (their ages and colors are consistent with those of their neighbors), however, they are associated with region B. Their high RVs therefore support the hypothesis that region B represents a window of lower-than-average extinction, since if the clusters are located deeper within the galaxy, we would expect the radial component of their orbital velocity to be larger than that of material nearer the surface. We note that the discrepancy between the Ca ii and cluster RVs in region B implies that the Ca ii absorption originates principally from the field stars (or their scattered light), not from the star clusters.

Our observations include three slits in the nuclear region, from which we have extracted six high S/N spectra. These probably represent bright "cluster complexes" containing blended light from a number of sources.¹¹ This is because our spatial resolution is not high enough to identify individual clusters within the crowded nuclear region. The RVs of five of these six cluster complexes place them on the steep part of the rotation curve, implying that they are associated with the nuclear bar. The exact location of these clusters with respect to the galaxy nucleus was briefly discussed in Section 4.5; the evidence presented by the RVs suggest that they may not lie at the very deepest part of the galaxy nucleus—as also suggested by their relatively low extinction—therefore their characterisation as "nuclear" may be inaccurate. Having said that, they provide the closest tracer to the behavior and state of the nuclear region in our data set. In order to study the true nuclear cluster population of this galaxy, IR observations are necessary.

Interestingly, we find that, according to the Na  iD velocity measurements, the neutral gas does not appear to follow the same rotation pattern as the stars, clusters, or ionized gas. Throughout the disk, the Na  iD measurements are well described by a single, flat rotation curve with a gradient approximately equal to that of the stars/clusters in the outer disk. Na  iD absorption must therefore originate from neutral gas located further out in the disk (given that they absorb cluster light), where the radial component of its orbital velocity is less. This also explains why none of the Na  iD measurements are associated with the fast bar orbits in the nuclear region.

We now return to the two discussed OPC candidates. As discussed above, cluster 152 was found to have a lower-than-expected RV for its position, with a measurement more consistent with the neutral gas RVs, implying that it may lie at a considerable distance away from the disk. No certain RV measurement could be performed for the other OPC candidate, No. 139, due to the low S/N of its spectrum (however a low-confidence fit places the cluster within the M82 system).

Both clusters are located at projected heights below the disk of >200 pc, implying that they may be part of the galaxy halo. A brief inspection of the area surrounding these clusters reveals the presence of numerous red stars, possibly red giants (RGs), or other, highly reddened stars. We investigated the possibility of photometric contamination by varying the aperture/sky annulus size and found no evidence of such an effect.

Given the red color and faintness of cluster 139 compared to the majority of clusters in the sample, there is a possibility that this object may in fact be a background galaxy—although we have no RV data to confirm or refute this. We can reject the possibility of cluster 152 being a background galaxy (regardless of its photometry), based simply on its RV, which places it within the M82 system. In Section 3.1, we found evidence that this cluster is older than ∼8 Gyr (using a $\frac{1}{5}\hbox{Z$_\odot $}$ fit). Given the uncertainty of the Padova isochrones for intermediate/old ages, we assume that this cluster is an old globular. However, the possibility still exists that cluster 152 is of intermediate age, constituting the first potential detection of the sort in M82. In any case, this object deems further study.

Our results find a distribution of RVs typical of a disk galaxy, with solid body rotation in the nucleus (slits 67 and 78) and flat (or gently outward increasing) components for disk velocities. There are, however, a handful of clusters in region M82-B (200 to 1000 pc east of the nucleus) whose velocities do not follow the observed trend: clusters 108, 112.2, 125, and 126 all have significantly higher velocities than their neighbors and the underlying stellar component. A simple consideration of M82 as it would be seen face-on reveals that clusters situated at similar projected galactocentric distance may reside at a range of depths along the line of sight. A representation of this effect may be found in Figure 9, left panel, where we have assumed that the galaxy disk is roughly circular. In this schematic, the thick line represents the nuclear bar, and crosses map out a grid limited to a 3 kpc radius (the projected visible radius of M82); vectors represent the expected RV of an object in differential rotation at that location. We note here that this is a simplified representation of the galaxy, where the intricate dynamics of the regions near the bar are not modelled, and only regular, circular disk rotation is considered. This is simply because there is very little information to be revealed by nuclear clusters concerning galaxy kinematics.

Zoom In Zoom Out Reset image size

This basic trigonometric consideration reveals that a cluster on a regular orbit near the visible surface of the galaxy will appear to move significantly slower than another regularly orbiting cluster situated at great depth, r. The weight of this effect ranges with projected galactocentric distance (indicated by d in Figure 9): at the extremes of the galaxy disk this difference will be within the measurement errors; in the mid-regions of the galaxy, assuming the M82 disk is roughly circular, the velocities will differ by a factor of two. Clusters projected on top of the nucleus will appear to have near-zero RVs, as their velocity is mostly in the tangential direction and is therefore not measurable; just to the side of the nucleus the discrepancy reaches a factor of 10.

Any existing spiral structure in M82 will affect its appearance, even in our highly inclined viewpoint. Given the orientation of the bar, which extends along our line of sight, one would expect spiral arms to originate from the visible central regions of the galaxy on the near side, and a corresponding site in the obscured side. This could result in a highly irregular dust distribution, and therefore significant discrepancies in the visible components of the disk.

In order to find evidence of the two-armed, barred spiral structure (as hypothesized in the literature, see Yun et al. 1994, and references therein) one must also consider the inclination of the disk, which provides a view of M82 from below. Assuming no significant spiral substructure other than the two dominant arms, we should expect to face one strong extinction front in the one side of the galaxy (the near spiral arm), and an "opening" in the dust distribution on the other side, along lines of sight underneath the bulk of the galaxy, and into the disk. More specifically, the near spiral arm should stem from the tip of the bar (the approximate location of SSC A1; Smith et al. 2006) and extend westward, and the far-side arm should originate at a symmetric location and "wrap" around region B. Such a conclusion was reached by Mayya et al. (2005), who derived the approximate location of the spiral arms by subtracting a hypothetical axisymmetric exponential disk from NIR imaging of the galaxy.

This structure is observed in the optical and our hypothesis on the approximate placing of the two spiral arms is supported by long wavelength observations (IR and radio). Examining the optical data presented in this paper, we assign the proposed structure as follows: the western side of the galaxy hosts the near spiral arm, hence the highly obscuring extinction front in the southwest disk. A faint trail of reddening dust can be seen superimposed on the central region, stemming from the general vicinity of tip of the nuclear bar, and spreading toward this extinction front. Therefore, the eastern side is expected to host the far spiral arm; as it hosts region B, where kinematics reveal lines of sight deep into the body of the galaxy, it is consistent with this concept.

Longer wavelength studies lend additional support to this model. First, we consider Hubble NICMOS data (Alonso-Herrero et al. 2001) in the J and H near-infrared bands, which trace star light, with much less obscuration from interstellar dust.¹² These data show the eastern disk as a crowded stellar field, with a large number of star clusters that are invisible in submicron wavebands. The part of region B closest to the nucleus (200 to 700 pc) is more crowded, as it represents a cross section of more than ∼4 kpc in depth, and projections of stars and clusters are the norm. Unfortunately, no equivalent high-resolution IR data exist for the western disk.

As a tracer of spiral arms, we have sought out radio data with continuous coverage of the galaxy. We use the CO molecular data in Walter et al. (2002, see their Figure 2 where radio and optical data are overlayed) as an appropriate diagnostic of spiral structure—CO is a tracer of H₂. We find a good agreement between the structures we propose as representing the spiral arms and the CO streamers discovered by Walter et al.. The RVs found for these streamers are consistent with material delineating the outer parts of the disk, as a spiral arm would.

This model finds support in our analysis of disk kinematics; in the right panel of Figure 9, we have matched observed cluster velocities onto the modelled RV field, according to their measured projected galactocentric distance (d). Applying this method, we find clusters in region B to be consistent with being located at large radii (r), given their high RVs.

Finally, we consider the placement of line of sight (LOS) gas, traced by the Na  iD doublet line (plotted onto Figure 9 as filled red boxes). We expect Na  iD line velocities to have maxima at extreme d (i.e., the western and eastern edges of the disk), and a linear trend between those extrema crossing (d, r) = (0, 0); this is confirmed by our observations. There is one deviation from this, as we find the interstellar gas to be situated at greater depth across region B (but at the surface between r = 600–800), thus lending some more support to our hypothesis.

Na  i D measurements can also provide more insight on the exact location of their corresponding clusters. As it is not possible to distinguish kinematically between symmetric near/far side locations, we resort to indirect means to choose between the two solutions to this problem: in some cases, the Na  iD clump is placed behind the respective cluster (with respect to our LOS), which is impossible as the line occurs in absorption of cluster light. In these cases, we deduce a far side location for the cluster.

We therefore present new observational evidence for the morphological structure of M82, and confirm that it consists of two dominant spiral arms and very weak, if any, other spiral structure, as deduced by Ichikawa et al. (1995) using NIR imaging of the galaxy, and followed up by Mayya et al. (2005).

Under the proposed model, part of the eastern M82 disk is left relatively free of extinction. The spiral arm is situated at the far reaches of the galaxy, therefore allowing a clear view into the body of the galaxy. In Paper I and Smith et al. (2007), we argued that region B in the eastern disk is seen through low extinction windows, similar to "Baade's windows" in the Milky Way. Here, we offer the interpretation that these windows form as a result of the inclination of M82 and provide a line of sight under the spiral arm extinction front. Thus, we confirm that region B is not intrinsically brighter, but simply seen under a clearer, relatively unextinguished line of sight. This interpretation is supported by the high RVs of the mentioned group of region B clusters, as they lie within these low extinction windows. This means that they can be seen clearly, even though they are situated deep within the body of the galaxy, and they do not form a distinct population, in terms of SFR, intrinsic luminosity, or actual kinematics.