Abstract
Scenarios that invoke multiple episodes of star formation within young globular clusters (GCs) to explain the observed chemical and photometric anomalies in GCs require that clusters can retain the stellar ejecta of the stars within them and accrete large amounts of gas from their surroundings. Hence, it should be possible to find young massive clusters in the local Universe that contain significant amounts (>10 per cent) of the cluster mass of gas and/or dust within them. Recent theoretical studies have suggested that clusters in the Large Magellanic Cloud (LMC) with masses in excess of 104 M⊙ and ages between 30 and ∼300 Myr should contain such gas reservoirs. We have searched for H i gas within 12 LMC (and 1 Small Magellanic Cloud) clusters and also for dust using Spitzer 70 and 160 μm images. No clusters were found to contain gas and/or dust. While two of the clusters have H i at the same (projected) position and velocity, the gas does not appear to be centred on the clusters, but rather part of nearby clouds or filaments, suggesting that the gas and cluster are not directly related. This lack of gas (<1 per cent of the stellar mass) is in strong tension with model predictions, and may be due to higher stellar feedback than has been previously assumed or due to the assumptions used in the previous calculations.
1 INTRODUCTION
Stellar clusters have traditionally been thought of as simple stellar populations (SSPs), with all of the stars within a given cluster having the same abundance and age, within some small tolerance. However, all old globular clusters (GCs) in the Galaxy that have been studied in the necessary detail show abundance anomalies not observed in field stars of the same metallicity. The most common of these is a Na–O anticorrelation (e.g. Carretta et al. 2009). Many GCs with precise photometry from the Hubble Space Telescope (HST) also show features in the colour–magnitude diagram that are inconsistent with being SSPs, such as multiple sequences, in regions from the main-sequence turnoff to the giant branch. The origin of these anomalies is still under debate.
One potential solution is to have multiple epochs of star formation within massive clusters, i.e. when the GCs were young (<200–300 Myr). In this scenario, material processed by stars of an initial population in the form of stellar ejecta is mixed with a large amount of primordial (unenriched) gas, which then forms a second (or multiple) generation(s). The stars that contribute to the processed material have been suggested to be rapidly rotating massive stars (e.g. Decressin et al. 2007), interacting high-mass binaries (e.g. de Mink et al. 2009) and asymptotic giant branch (AGB) stars (e.g. D'Ercole et al. 2008), among other candidates. If AGB stars are the contributor, which do not process material for the first ∼30 Myr of a cluster's life, the cluster must be able to retain large amounts of primordial gas or accrete new gas from its surroundings (e.g. Conroy & Spergel 2011 – hereafter CS11).
CS11 have theoretically investigated the conditions necessary for a cluster to accrete gas from its surroundings. There are two main mechanisms to bring gas into an existing cluster (1) through Bondi–Hoyle type accretion into the cluster and (2) if the cluster already has some fraction of its mass in gas, that internal gas can sweep up material as the cluster passes through the interstellar medium (ISM). In the second case, the cluster gas can also be stripped from the cluster due to ram pressure.
CS11 have applied their calculations to clusters in the Large Magellanic Cloud (LMC), taking the observed current ISM density and cluster velocities within the galaxy. The authors suggest that there is a minimum mass, ∼104 M⊙, above which clusters should be able to retain (and further accrete) gas from their surroundings. These clusters may then undergo a second star formation epoch.
Such secondary (or continuous or multiple burst) have been suggested to have happened in a number of intermediate age (1–2 Gyr), massive (>104 M⊙) clusters in the LMC and Small Magellanic Cloud (SMC; e.g. Mackey & Broby Nielsen 2007; Milone et al. 2009; Goudfrooij et al. 2011a,b). These works have suggested (semi)continuous star-forming episodes lasting for 200–500 Myr based on extended main-sequence turn-offs (eMSTOs) within the clusters. However, observations of massive (104–105 M⊙) clusters in the LMC with younger ages, where such age spreads should be readily apparent, show that these clusters are all consistent with a single burst of star formation (Δ(age) < 10–40 Myr – Bastian & Silva-Villa 2013; Niederhofer et al. 2014). Additionally, observations of 130 Galactic and extragalactic clusters with masses between 104 and 108 M⊙ and ages between 10 and 1000 Myr found no evidence of ongoing star formation within any of the clusters (Bastian et al. 2013b). If star formation continued for 100–200 Myr after an initial burst, at least 50 per cent of their sample should have shown evidence for ongoing star formation, unless the initial mass function of stars was severely deficient in high-mass stars (>15 M⊙). These observations call into question whether clusters are able to retain/accrete the necessary gas and dust in order to form stars after the initial burst.
If the anomalies seen in the intermediate-age LMC clusters (i.e. the eMSTO) are due to (continuous) extended star formation histories, then at any given time during the star formation epoch, a significant amount of gas needs to be present. The exact amount depends on when the cluster is observed during its extended formation. Bastian et al. (2013b) have estimated the amount of ongoing star formation, as a function of the current mass of the clusters, expected in clusters with different star formation histories. If we adopt the continuous star formation histories estimated by Goudfrooij et al. (2011b), which are reasonably well approximated by Gaussians with dispersions of 200–500 Myr, then the clusters would be expected to be presently forming >7–10 per cent of their current mass (averaged over 7 Myr). For 100 per cent star formation efficiency, we then would expect >700–1000 M⊙ of gas within the clusters in our survey (Mcluster, stars > 104 M⊙).
However, the amount of gas predicted to be in clusters by CS11 is not as straightforward. While the Bondi–Hoyle accretion of the gas is a direct function of the cluster mass and ISM density, most of the gas in their model is swept up from the surrounding ISM by the initial gas in the cluster (i.e. the cluster gas acts as a net as the cluster passes through the ISM). CS11 adopt an initial gas mass of 10 per cent of the mass of the cluster, which is effective in sweeping up the material, hence a lower limit predicted for the amount of gas present in our sample would be 103 M⊙, which would increase as the clusters accrete gas from their surroundings.
We note an important difference between the CS11 models and the observations and interpretation presented by Mackey & Broby Nielsen (2007) and Goudfrooij et al. (2011b). The CS11 models predict that gas stays within the cluster for the first 100–200 Myr, and then cools rapidly, allowing a second, discrete, burst of star formation. However, the observations suggest a smooth spread in the eMSTOs of the intermediate-age clusters, which suggests (if interpreted as being caused by differences in age) a continuous star formation episode lasting ∼500 Myr. Additionally, D'Ercole, D'Antona & Vesperini (2011) have argued against the CS11 scenario on the basis of the observed chemistry within GCs. If AGB stars are the source of the enriched material, in order to produce the ‘extreme’ stars (in terms of their abundances), these stars need to be formed entirely from material processed through AGB stars, i.e. without any diluting pristine material. The CS11 scenario results in a smooth mass accretion, whereas D'Ercole et al. (2011) argue that a time variable accretion rate of the pristine material from the surroundings is required to fit the observations.
In this work, we test the above predictions, that young massive clusters in the LMC (and SMC) have significant amounts of gas within them. We study a sample of clusters with masses >104 M⊙ and ages between 25 and 315 Myr (see Section 2.1), i.e. those predicted to host a gas reservoir. In our search, we are guided by the predictions of CS11, that the gas should be cool (T ∼ 100 K), so should be bright in H i and possibly in dust emission. We use the publicly available Australia Telescope Compact Array (ATCA)+Parkes H i surveys of the LMC (Kim et al. 2003; Staveley-Smith et al. 2003) and SMC (Stanimirović et al. 1999) to search for H i at the position and velocity of the clusters in our sample. For the dust, we use the Spitzer SAGE (Surveying the Agents of Galaxy Evolution) survey (Meixner et al. 2006) and the SAGE-SMC survey (Gordon et al. 2011) to search for emission at the location of the clusters, and we also search for optical colour gradients within the clusters, as the gas/dust is expected to be centrally concentrated in the cluster, so the inner portions are predicted to be redder than the outer parts.
We note that alternative theories have been put forward to explain the observed abundance anomalies in GCs and eMSTOs in the intermediate-age LMC/SMC clusters that do not invoke multiple (or continuous) epochs of star formation. For GCs this is the early disc accretion scenario (Bastian et al. 2013a) and for the eMSTOs, both stellar rotation (Bastian & de Mink 2009) and interacting binary stars (Yang et al. 2011) have been suggested.
The paper is organized as follows. In Section 2, we describe the cluster sample as well as the observations used in this work. In Section 3, we present our results on the search for gas/dust within the clusters and in Section 4, we present our conclusions.
2 OBSERVATIONS AND TECHNIQUES
2.1 Cluster catalogue
We selected our sample of clusters from the McLaughlin & van der Marel (2005) catalogue, searching for clusters in the LMC/SMC with ages between 10 and 500 Myr and masses in excess of 104 M⊙. The masses were obtained from profile fitting as well as by comparison of the cluster luminosities with the mass-to-light ratio of SSP models of the appropriate age. The clusters and their ages and masses are given in Table 1, mostly from McLaughlin & van der Marel (2005); however, NGC 1850, NGC 1856 and NGC 1866 have updated values taken from Niederhofer et al. (2014) and Bastian & Silva-Villa (2013). None of the clusters in the current sample have been suggested to have significant age spreads within their stellar populations. However, the sample clusters were predicted by CS11 to host significant amounts of gas within them and have the potential to form further stellar generations.
The clusters and their properties used in this work, and the results of the H i survey. The ages and masses have been taken from the compilation of McLaughlin & van der Marel (2005) with the exceptions of NGC 1856 and NGC 1866 which are from Bastian & Silva-Villa (2013) and NGC 1850 which is from Niederhofer et al. (2014).
| Cluster | Age | Mass | H i | Relative velocity | Reference |
|---|---|---|---|---|---|
| Myr | [103 M⊙] | detection | (H i and cluster) | (cluster velocity) | |
| NGC 1711 | 50 | 17 | no | – | – |
| NGC 1818 | 25 | 26 | yes | 5 ± 7 km s−1 | Freeman et al. (1983) |
| NGC 1831 | 315 | 39 | yes | >10 km s−1 | Schommer et al. (1992) |
| NGC 1847 | 26 | 25 | yes | 10 km s−1 | this work |
| NGC 1850 | 100 | 140 | yes | >10 km s−1 | Fischer, Welch & Mateo (1992) |
| NGC 1856 | 280 | 76 | yes | >10 km s−1 | Freeman et al. (1983) |
| NGC 1866 | 180 | 81 | no | – | – |
| NGC 2031 | 160 | 30 | yes | 0 ± 2 km s−1 | Storm et al. (2005) |
| NGC 2136 | 100 | 20 | yes | >10 km s−1 | Mucciarelli et al. (2012) |
| NGC 2157 | 40 | 20 | yes | >10 km s−1 | Freeman et al. (1983) |
| NGC 2164 | 50 | 15 | no | – | – |
| NGC 2214 | 40 | 11 | no | – | – |
| NGC 330 | 25 | 50 | yes | >10 km s−1 | Hill (1999) |
| Cluster | Age | Mass | H i | Relative velocity | Reference |
|---|---|---|---|---|---|
| Myr | [103 M⊙] | detection | (H i and cluster) | (cluster velocity) | |
| NGC 1711 | 50 | 17 | no | – | – |
| NGC 1818 | 25 | 26 | yes | 5 ± 7 km s−1 | Freeman et al. (1983) |
| NGC 1831 | 315 | 39 | yes | >10 km s−1 | Schommer et al. (1992) |
| NGC 1847 | 26 | 25 | yes | 10 km s−1 | this work |
| NGC 1850 | 100 | 140 | yes | >10 km s−1 | Fischer, Welch & Mateo (1992) |
| NGC 1856 | 280 | 76 | yes | >10 km s−1 | Freeman et al. (1983) |
| NGC 1866 | 180 | 81 | no | – | – |
| NGC 2031 | 160 | 30 | yes | 0 ± 2 km s−1 | Storm et al. (2005) |
| NGC 2136 | 100 | 20 | yes | >10 km s−1 | Mucciarelli et al. (2012) |
| NGC 2157 | 40 | 20 | yes | >10 km s−1 | Freeman et al. (1983) |
| NGC 2164 | 50 | 15 | no | – | – |
| NGC 2214 | 40 | 11 | no | – | – |
| NGC 330 | 25 | 50 | yes | >10 km s−1 | Hill (1999) |
The clusters and their properties used in this work, and the results of the H i survey. The ages and masses have been taken from the compilation of McLaughlin & van der Marel (2005) with the exceptions of NGC 1856 and NGC 1866 which are from Bastian & Silva-Villa (2013) and NGC 1850 which is from Niederhofer et al. (2014).
| Cluster | Age | Mass | H i | Relative velocity | Reference |
|---|---|---|---|---|---|
| Myr | [103 M⊙] | detection | (H i and cluster) | (cluster velocity) | |
| NGC 1711 | 50 | 17 | no | – | – |
| NGC 1818 | 25 | 26 | yes | 5 ± 7 km s−1 | Freeman et al. (1983) |
| NGC 1831 | 315 | 39 | yes | >10 km s−1 | Schommer et al. (1992) |
| NGC 1847 | 26 | 25 | yes | 10 km s−1 | this work |
| NGC 1850 | 100 | 140 | yes | >10 km s−1 | Fischer, Welch & Mateo (1992) |
| NGC 1856 | 280 | 76 | yes | >10 km s−1 | Freeman et al. (1983) |
| NGC 1866 | 180 | 81 | no | – | – |
| NGC 2031 | 160 | 30 | yes | 0 ± 2 km s−1 | Storm et al. (2005) |
| NGC 2136 | 100 | 20 | yes | >10 km s−1 | Mucciarelli et al. (2012) |
| NGC 2157 | 40 | 20 | yes | >10 km s−1 | Freeman et al. (1983) |
| NGC 2164 | 50 | 15 | no | – | – |
| NGC 2214 | 40 | 11 | no | – | – |
| NGC 330 | 25 | 50 | yes | >10 km s−1 | Hill (1999) |
| Cluster | Age | Mass | H i | Relative velocity | Reference |
|---|---|---|---|---|---|
| Myr | [103 M⊙] | detection | (H i and cluster) | (cluster velocity) | |
| NGC 1711 | 50 | 17 | no | – | – |
| NGC 1818 | 25 | 26 | yes | 5 ± 7 km s−1 | Freeman et al. (1983) |
| NGC 1831 | 315 | 39 | yes | >10 km s−1 | Schommer et al. (1992) |
| NGC 1847 | 26 | 25 | yes | 10 km s−1 | this work |
| NGC 1850 | 100 | 140 | yes | >10 km s−1 | Fischer, Welch & Mateo (1992) |
| NGC 1856 | 280 | 76 | yes | >10 km s−1 | Freeman et al. (1983) |
| NGC 1866 | 180 | 81 | no | – | – |
| NGC 2031 | 160 | 30 | yes | 0 ± 2 km s−1 | Storm et al. (2005) |
| NGC 2136 | 100 | 20 | yes | >10 km s−1 | Mucciarelli et al. (2012) |
| NGC 2157 | 40 | 20 | yes | >10 km s−1 | Freeman et al. (1983) |
| NGC 2164 | 50 | 15 | no | – | – |
| NGC 2214 | 40 | 11 | no | – | – |
| NGC 330 | 25 | 50 | yes | >10 km s−1 | Hill (1999) |
2.2 H i data
Kim et al. (2003), Staveley-Smith et al. (2003) and Stanimirović et al. (1999) published H i maps of the LMC and SMC that combined single dish (Parkes) and multi dish (ATCA) for sensitivity to gas at a range of scales. The 1–1.6 arcmin resolution of these surveys is comparable to the typical sizes of massive clusters in the Magellanic Clouds, and so is a reasonable match (modulo the large uncertainties in model predictions) to the scales on which we would expect to find gas if it were present.
For each star cluster in the survey, we extracted H i emission spectra at the location of the cluster.1 We then determined whether there was any H i detected in the spectrum, and, if so, the radial velocity of the associated gas. These data are interpreted in Section 3.
2.3 Spitzer imaging
In order to search for cool/warm dust within the clusters, we used the publicly available SAGE Spitzer Space Telescope survey of the Magellanic Clouds (Meixner et al. 2006). We downloaded the processed images2 in a cutout around the position of the cluster centres. Specifically, we used the 3.6, 70 and 160 μm images. NGC 2214 was outside the SAGE field of view, so no limits on the dust mass present in that cluster can be given.
The 3.6 μm image was used to locate the position of the cluster, as stars with cool photospheres (giants) are readily detected. The 70 and 160 μm images were then used to search for dust within the clusters. Images of the clusters in the three wavelengths are shown in Figs 1 and 2.
Spitzer images of each cluster at 3.6, 70 and 160 μm. The circle in each image has a radius of 61 arcsec, which corresponds to ∼15 and ∼18 pc for the LMC and SMC clusters, respectively. For all the images, dark colours show emission.
With the possible exception of NGC 2136 and 2164 (see Section 3.2), we did not detect emission at 70 or 160 μm for any of the clusters. Like for the H i, some clusters (e.g. NGC 2031) did have emission within the aperture; however, this emission was not centred on the clusters. Instead, it appears to be related to nearby filaments. NGC 2157 may have a slight detection in the 160 μm image, although no corresponding emission was found in the 70 μm image.
2.4 SOAR observations
One of the LMC clusters in our sample (NGC 1847) had no published radial velocity to compare to the detection of H i at its position. To estimate its velocity, we obtained long-slit spectra of two bright stars near the cluster centre using the Goodman High Throughput Spectrograph (Clemens, Crain & Anderson 2004) on the Southern Astrophysical Research (SOAR) 4.1-m telescope. We used a 2100 l/mm grating and a 1.03 arcsec slit for the single 300 s observation, giving a resolution of 0.9 Å over the wavelength range ∼4960–5600 Å. The spectra were reduced in the standard manner, and heliocentric radial velocities derived through cross-correlation with standard stars observed with the same setup. The two stars had radial velocities consistent to within 3 km s−1, suggesting they are both cluster members. The mean velocity is 292 ± 3 km s−1, which we adopt as the systemic radial velocity for NGC 1847.
3 RESULTS
3.1 H i Sensitivity and detections
The results from the H i survey are presented in Table 1. 11 of the 13 clusters in our sample did not have H i detected at the position and velocity of the cluster. In order to determine the upper limit to the amount of H i gas present, we generously take the range ±10 km s−1 from the cluster (20 km s−1 total). For the typical noise of the LMC and SMC surveys (the SMC survey was somewhat deeper), this gives 3σ column density limits of 7.9 × 1019 and 4.3 × 1019 atoms cm−2, respectively. To convert these values into H i masses, we need to assume a size of the cluster, and find |$M_{{\rm H}\small {\rm I},\text{LMC}} = 1.99 R_{\rm cl}^2 \,\mathrm{M}_{{\odot }}$| (for the single SMC cluster in our sample, the prefactor is 1.08 rather than 1.99). For a radius of 10 pc, the 3σ upper limit for the LMC clusters is 199 M⊙. If we were less conservative and adopted ±5 km s−1, this value would be lower by a factor of 2. Similarly, if the gas was more concentrated, i.e. within a radius of 5 pc, the upper limit would shift to ∼50 M⊙.
Two of the clusters (NGC 1818 and NGC 2031) have H i detected at the spatial position (with the 15 pc aperture) and velocity of the cluster. The inferred H i mass summed over a range ±10 km s−1 around the peak velocity along the line of sight for NGC 1818 is |$M_{{\rm H}\small {\rm I}} = 26 \times R_{\rm cl}^2 \,\mathrm{M}_{{\odot }}$|, i.e. ∼2600 M⊙ for a radius of 10 pc. However, spatially, the gas is not centred on the cluster, but rather appears to be projected on the edge of a large H i cloud to its north-east. This is shown in Fig. 3. Hence, it is reasonable to argue that the H i is not likely to be associated with the cluster. However, we cannot prove or disprove this with the currently available H i data alone; better resolution imaging is required. However, looking at the Spitzer images (see Section 3.2), which have higher resolution (Fig. 1), we see that the emission is not coming from the cluster itself, but rather from nearby filamentary gas/dust clouds.
H i emission maps for the two clusters (NGC 1818 and NGC 2013) that have detected H i within an aperture of 15 pc of the cluster with the same (within the errors) radial velocity. The circles show a radius of 15 pc (the same as used in Fig. 1). Note that in both clusters the emission does not appear to be from the clusters, but rather due to nearby larger scale filaments.
For NGC 2031, the inferred H i mass is |$M_{\text{H}\,\small {\rm I}} = 17 \times R_{\rm cl}^2 \,\mathrm{M}_{{\odot }}$|, corresponding to ∼1700 M⊙ for a radius of 10 pc. However, NGC 2031 is located (in projection) on the edge of a filament (also seen in cold dust in Fig. 2), suggesting that the detected H i is not associated with the cluster directly. As was the case for NGC 1818, higher resolution data would be required to assess whether the H i was likely associated with the cluster. Again, the Spitzer imaging shows (Fig. 2) that the emission is not due to the cluster, but rather from nearby filamentary clouds. Hence, from the current data, it appears that the gas is not physically associated with the clusters.
3.2 Spitzer sensitivity
In order to determine the sensitivity of the observations we use the approach of Gordon et al. (2010 – their equation 2) to estimate the amount of dust mass present in both the 70 and 160 μm images. Here, the dust mass is a function of the grain size, density, temperature, distance, flux and the emissivity of the dust at that wavelength. We adopt the same parameters as Gordon et al. (2010), and an emissivity of the dust at 70 μm that is five times higher than at 160 μm. We note that if we used the more recent emissivity value at 70 μm that is 3.4 times higher than at 160 μm (Planck Collaboration 2014), the estimated masses at 70 μm would be increased by a factor of ∼1.5, which would not significantly impact our results.
Since most of the clusters had a clear detection at the location of the clusters, we adopt the flux limits given by the SAGE survey, namely 0.1 and 1 Jy at 70 and 160 μm, respectively. This just leaves the dust temperature as a free parameter. Adopting these limits, the upper limit to the amount of dust present (in each of the bands) as a function of temperature is shown in Fig. 4. Note the strong temperature dependence on the upper limit of the dust mass.
The estimated upper limit on the amount of dust present within the clusters, based on the detection limit for each of the Spitzer bands used.
If we adopt a dust temperature of 30 K, we obtain an upper limit of ∼0.15 and ∼2 M⊙ for the dust mass for the 70 and 160 μm fluxes, respectively. Assuming a gas-to-dust ratio of 500-to-1 (e.g. van Loon et al. 2005a; Roman-Duval et al. 2010), this leads to an upper limit of 75 and 1000 M⊙ of gas for the two bands. However, if the dust was this cool, we may expect high extinction in the optical, or for the gas to catastrophically cool (CS11) and begin forming stars. Hence, if gas does exist within the clusters, we may expect it to be at a higher temperature. Adopting a dust temperature of 100 K leads to significantly lower upper limits on the dust mass, 0.001 and 0.15 M⊙, for the 70 and 160 μm fluxes, respectively. Again, assuming a gas-to-dust ratio of 500-to-1, leads to upper limits on the total gas mass within any of the clusters of 0.5 and 75 M⊙ for the two bands, respectively.
Due to the steep dependence of the upper limit of the dust mass on the dust temperature, temperatures above 100 K lead to dust and gas masses well below 1 per cent of the stellar mass of the cluster.
NGC 2136 has a significant amount of flux in the 160 μm image (see Fig. 2), which appears to be an extension of a nearby gas/dust cloud. No detection is found in the corresponding 70 μm image.
3.3 The curious case of NGC 2164
One cluster, NGC 2164, shows a marginal detection at 70 μm. While most of the emission is not centred on the cluster, the cluster resides at one end of the extended emission, suggestive that the emission may originate from the cluster. Less than 10 per cent of the 70 μm flux of the extended feature is seen within 15 pc of the cluster centre. No trace of the extended emission is detected in the 24 or 160 μm images. Additionally, there is no H i emission detected at any velocity at the position of the cluster (or along the extended filament).
In order to see if there is indeed dust within the cluster, we used archival HSTF555W (V) and F814W (I) band imaging in order to search for a colour gradient within the cluster. If there is gas/dust within the cluster, the expectation is that it would collect in the centre of the cluster and be more centrally concentrated than the stars (e.g. CS11). As such, we would expect the integrated (radially) cluster colour to become redder closer to the centre (e.g. Bastian & Silva-Villa 2013). We do not find any colour gradient within the cluster from |$1\text{ to }5$| pc from the cluster centre. Within 1 pc, however, the colour becomes bluer at smaller radii. The overall, <5 pc, colour of the cluster (F555W–F814W =0.6) is consistent with an age of ∼40 Myr (McLaughlin & van der Marel 2005) and a small amount of extinction, AV ≲ 0.3 mag (e.g. Kotulla et al. 2009). Hence, we also do not find evidence for dust within the cluster based on optical imaging.
The origin of the extended 70 μm emission near NGC 2164, with no counterpart detected at any other wavelength, remains unknown.
3.4 Dust production within the clusters
While no H i gas or dust was observed within the clusters studied here, many of the clusters studied here have luminous evolved stars that are expected to produce dust. Would we expect to observe such dust? The average cluster in our sample is expected to have 10 or fewer luminous AGB stars (Mucciarelli et al. 2006), whereas the younger clusters (e.g. NGC 1818, NGC 1847 and NGC 330) may have up to ∼20–30 red supergiants (RSGs) within them (e.g. Davies et al. 2008). The typical total (gas + dust) mass-loss rates for AGB and RSGs in the LMC are ∼1 × 10−6 M⊙ yr−1 (Bonanos et al. 2010).
Assuming a gas-to-dust ratio of 500-to-1, the amount of dust produced within the clusters per Myr is expected to be |${\sim }0.04\text{--}0.06$| M⊙ or less. If the temperature of the dust is larger than ∼30 K, it may be detectable at 70 μm (see Fig. 4). The fact that we do not detect the clusters at 70 μm (with the possible exception of NGC 2164) means that either the assumed number of evolved massive stars (and their mass-loss rates) are overestimated (as they are meant to be upper limits), or the clusters are efficient at removing any gas/dust released in the cluster. We note that some of the younger clusters (e.g. NGC 330) may have a significant number of Be stars within them that may also contribute gas and dust to the host cluster (e.g. Martayan et al. 2007), further emphasizing the efficiency of the gas/dust removal process from young (≲300 Myr) clusters. Also, we note that the mass-loss rates of RSGs and AGBs are expected to be highly time variable, with the high mass-loss rate phase being short-lived (e.g. van Loon et al. 2005b). Hence, in order to set strict limits on the amount of gas/dust expected, detailed study of the stellar populations within each cluster is required, meaning that the above calculations are meant to be indicative only.
The winds of AGB stars in the LMC have velocities between 8 and 25 km s−1 (e.g. Marshall et al. 2004), which are at, or near, the escape velocities of the clusters in our sample (Niederhofer et al. 2014); hence, it is possible that the RSG/AGB ejecta are able to flow freely from the cluster. Additionally, radiation pressure and the ionizing flux from stars still on the main sequence (or near the main-sequence turnoff) will also act to remove gas/dust from the clusters. Finally, since the amount of gas/dust from RSG/AGB stars within the cluster, at any given time, is expected to be a small fraction of the cluster mass <1 per cent, it is possible that the gas/dust is efficiently stripped by the intracluster medium as the cluster moves through the galaxy (CS11). In a future work, we will place constraints on the amount of dust within much more massive (although more distant) clusters than those presented here.
4 DISCUSSION AND CONCLUSIONS
We have searched for evidence of H i gas and/or dust within a sample of massive (>104 M⊙) clusters within the Large and Small Magellanic Clouds. The clusters have ages between ∼25 and ∼300 Myr, hence would be expected to contain gas if they were able to form a second (or continuous) generation of stars within them, as predicted by recent theory (e.g. CS11). In the majority of our sample, no gas or dust was found at the location (and velocity) of the clusters. In two cases, H i gas was detected at the location and velocity of the cluster (NGC 1818 and NGC 2031); however, upon further examination, the gas was not found to be centred on the clusters, but rather appears to belong to nearby filaments.
We have placed upper limits of ∼200 M⊙ of gas within the clusters based on |${{\rm H}\small{\rm I}}$| measurements and |${\sim }1\text{--}75$| M⊙ of total gas based on the dust measurements. In Fig. 5, we show the upper limit on the fraction of gas + dust within the cluster compared to the stellar mass as a function of stellar mass. In all cases, we can place strict upper limits of <10 per cent of the (stellar) mass of clusters present in gas + dust, and in most cases the upper limit is <1 per cent. Hence, we conclude that these clusters are not able to retain or accrete significant amounts of gas.
The estimated upper limit on the mass fraction of gas (and dust) to stars within the clusters. The limits from H i and 160 μm are shown, and the limits from the 70 μm are well below those shown.
Our observations are in conflict with the predictions of CS11, who predicted that all clusters in the LMC with masses above 104 M⊙ (and have ages ≳30 Myr) should be able to retain ejected gas from the stars within them and accrete new gas from their surroundings. They use this prediction to explain the potential existence of large (200–500 Myr) age spreads in intermediate-age clusters in the LMC (e.g. Goudfrooij et al. 2011b). Additionally, they use their predictions to support the ‘AGB scenario for the formation of multiple populations within clusters, where GCs must retain the AGB ejecta of stars within them and accrete large amounts of ‘pristine’ material in order to form a second generation of stars within the young cluster.
The lack of observed gas, in contradiction to the above prediction, is likely to be due to one of two reasons (or a combination of both):
stellar feedback within cluster is not negligible, causing gas to be efficiently removed from the cluster;
the accretion of gas from the surroundings is less efficient than assumed. CS11 assume that 10 per cent of the mass of the cluster is already initially in gas within the cluster. This gas acts as a net that can sweep up additional material from the ISM. If the simulations start with no gas in the cluster, Bondi–Hoyle accretion is not enough to accrete gas to the necessary amounts to sweep up further material. The initial gas has to come from somewhere in order to initiate the gas accretion (i.e. the sweep up). If no gas is initially available then not enough further gas will be accreted by the cluster.
For comparison, NGC 1850 at an age of ∼100 Myr, has a stellar mass of ∼1.4 × 105 M⊙, hence, the assumption is that it should have ∼14 000 M⊙ of gas within it. The upper limits imposed by the observations presented here are ∼1–200 M⊙ (depending on the gas/dust temperature and whether one adopts the limits from H i or dust). The same can be said about the older massive clusters, NGC 1866 and NGC 1856 (180 and 280 Myr, respectively), both would be expected to have more than 7000 M⊙ worth of gas within them, or more if they were able to accrete new gas. The lack of gas within young clusters likely explains why these massive clusters do not show evidence for age spreads within them (Bastian & Silva-Villa 2013; Niederhofer et al. 2014).
In the CS11 model, the collected material (i.e. that shed from the stars and accreted from the surroundings) may not be neutral, but rather reside in dense clumps that are ionized by the large number of B-stars in the clusters. This has not been tested with the present observations, however such ionized clumps would be expected to produce strong optical emission lines. In a survey of 130 Galactic and extragalactic massive clusters (104 < M/M⊙ < 108) with ages greater than 10 Myr, none were found with emission lines in their integrated spectrum, implying that there is <100 M⊙ of ionized gas within the clusters. Additionally, optical spectroscopy of LMC clusters has not detected emission lines associated with the clusters (e.g. Beasley, Hoyle & Sharples 2002; Leonardi & Rose 2003; Palma et al. 2008).
Whether more massive clusters are able to retain and accrete significant amounts of gas remains unclear. Massive old GCs display a well-known deficit of gas within them, relative to the observed outflows of their red giant stars within them, showing that even massive clusters (with low stellar feedback relative to that expected in young clusters) are unable to hold on to gas within them (e.g. van Loon et al. 2009 and references therein). Potentially, observations with Atacama Large Millimeter Array (ALMA) will be able to find or rule out gas in more distant young clusters with masses in excess of 105 or 106 M⊙. However, we note that no clusters with (main population) ages above 10 Myr, including those with masses up to 108 M⊙, have been found with evidence of ongoing star formation within them (Bastian et al. 2013b). Additionally, integrated spectroscopy of a young massive cluster (∼107 M⊙) has shown that it is consistent with single burst of star formation (Cabrera-Ziri et al. 2014).
Many popular models for the production of multiple generations of stars in massive clusters predict that significant neutral gas must be present in the clusters for the second generation(s) to form. Observations presented in this paper, especially when combined with the above results, appear to be in strong tension with these predictions.
We would like to thank Mikako Matsuura for helpful discussions on dust masses and the SAGE survey and Charlie Conroy, Laura Chomiuk and Steve Longmore for helpful comments and discussions. We would also like to thank Alex McComb for help with data reduction. The referee, Jacco van Loon, is gratefully acknowledged for his helpful and constructive report. NB is partially funded by a Royal Society University Research Fellowship. This paper was partially based on observations obtained at the SOAR telescope, which is a joint project of the Ministrio da Cincia, Tecnologia, e Inovao (MCTI) da Repblica Federativa do Brasil, the US National Optical Astronomy Observatory (NOAO), the University of North Carolina at Chapel Hill (UNC), and Michigan State University (MSU).
