Main

The Submillimeter Array image of NGC 5253, shown in Fig. 1, reveals a bright CO(3→2) source coincident with the giant cluster and its ‘supernebula’4. ‘Cloud D’ (ref. 3) is one of only two molecular clouds detected within the galaxy; the second cloud is smaller and located ∼5″ (90 pc) to the southwest. A ‘streamer’ of gas extending along the minor axis is also detected in CO(3→2). This streamer, previously detected in lower-J CO lines, seems to be falling into the galaxy near the supernebula3,5. Both the streamer and Cloud D emit 870-μm continuum emission, as shown in Fig. 2. Also shown is an image of 350-μm continuum, in which both Cloud D and the streamer are detected.

Figure 1: CO J = 3→2 emission in NGC 5253.
figure 1

The Submillimeter Array (SMA) CO(3→2) integrated line intensity, in red, is shown atop a λ814 nm Hubble Space Telescope image. The SMA beam is 4″ × 2″ (74 pc × 37 pc). The field covers 40″ × 40″ (740 pc × 740 pc), north up, east left. Image registration is to less than 1″. The CO streamer coincides with the optical dust lane to the east. The massive star cluster is located at the bright, compact CO peak, Cloud D; it is embedded8.9 and is not visible here. Cloud F is to the southwest of Cloud D.

PowerPoint slide

Full size image
Figure 2: Dust and gas in NGC 5253.
figure 2

SMA image of continuum dust emission at 870 μm (greyscale), with 350-μm dust continuum emission from SHARC at the Caltech Submillimeter Observatory (contours) superimposed. The SMA continuum image has been smoothed to 6″ resolution to show emission from the streamer. The SHARC image has been smoothed to 12.7″; contours are 2σ. Coordinates of the SHARC image are uncertain to ∼5″ (see Methods).

PowerPoint slide

Full size image

The molecular gas in Cloud D is hot. This is clear from the increase in brightness from CO(2→1) (ref. 3) to CO(3→2). The intensity ratio of the two lines is I32/I21 = 2.6 ± 0.5(Iline = ∫Tline dv). This ratio is non-thermal, although the thermal limit of 2.25 (∼ν2) is within the uncertainties and is what we adopt. Non-local-thermodynamic-equilibrium (non-LTE) modelling of this ratio using RADEX6 indicates a minimum kinetic temperature of TK > 200 K for the 1σ lower limit, and TK > 350 K for the adopted value of I32/I21 = 2.25 (see Methods). The high gas temperature is consistent with a thermal origin for H2 2.2-μm emission in the region7. Cloud D seems to be a photon-dominated region, heated by ultraviolet radiation from the several thousand cluster O stars in the cluster4,8. The CO(3→2)-emitting gas is dense, with .

By contrast with Cloud D, the streamer consists of more typical cool giant molecular clouds. Its value of I32/I21 = 1.0 ± 0.3 is consistent with optically thick emission, for which RADEX models allow temperatures as low as TK ≈ 15–20 K, and number densities . The mass of the streamer is , which is 1–2% of the stellar mass of the galaxy. The streamer is molecular, dense, and primed for star formation, even before entering the galaxy. This is unlikely to be a primordial collapsing filament9, but is more probably previously enriched gas.

The star-formation efficiency of a cloud or region can be defined as η = Mstars/(Mgas + Mstars), where Mstars is the stellar mass and Mgas the mass of molecular gas. Mgas can be hard to define for star-forming regions within giant molecular clouds, but the association of the isolated Cloud D with the supernebula gives us an opportunity to calculate η directly for the giant molecular cloud giving birth to this massive star cluster—if we can determine the mass of Cloud D.

CO is often used to estimate the mass of molecular gas, but this is unreliable in NGC 5253. Here we use the width of the CO line to determine a gas mass for Cloud D based on dynamical considerations. The CO linewidth is σ = 9.2 ± 0.6 km s−1, based on a Gaussian fit. The cloud dimensions, deconvolved from the beam, are 2.8″ × 1.5″ ± 0.1″ (52 pc × 28 pc). The virial mass is for Cloud D, with uncertainties due to the unknown internal mass distribution (see Methods). We assume that the linewidth is gravitational, with no winds or outflow; this is therefore an upper limit to Mvir. The virial mass includes both gas and stars, but we can constrain the stellar mass. The mass in stars exciting the supernebula, Mstars, can be predicted from the Lyman continuum rate of NLyc = (7 ± 2) × 1052 s−1 (ref. 6) and Bracket γ equivalent width of 255 Å (ref. 10). The star cluster has mass (see Methods). We then obtain a gas mass Mgas = Mvir − Mstars = (7 ± 4) × 105 if the cluster is embedded in the cloud. This treatment assumes that the CO kinematics trace all of the cloud and that there is no extensive layer of H2 without CO; however, the dust continuum and CO sizes are nearly identical, consistent with the dust and gas mass being contained within Cloud D.

Other methods to estimate H2 mass are problematic for Cloud D. CO(3→2) is optically thin, but the excitation temperature is not determined, nor is CO/H2 known. The line strength of 41 ± 8 Jy km s−1 gives MCO = (4 ± 1) (T/200 K). For a Galactic abundance ratio of [CO]/[H2] = 8.5 × 10−5, the H2 mass would be (T/200 K). This is one-tenth of the value for Mgas derived above. Intense radiation fields and high temperature will affect the chemistry and the relative abundance of CO.

Dust continuum emission can trace gas mass, but it is also unreliable in Cloud D. The 870-μm continuum flux density for Cloud D is S870μm = 72 ± 10 mJy, consistent with previous measurements11, of which free-free emission8 contributes = 38 ± 4 mJy, leaving dust emission = 34 ± 14 mJy. Our 350-μm image gives = 1.0 ± 0.2 Jy for Cloud D, consistent with the 870-μm flux. Adopting the dust opacity of the Large Magellanic Cloud12 (see Methods), we find an observed dust mass, . To obtain gas mass, we need a gas-to-dust ratio (GTD). Scaled as the oxygen abundance of NGC 5253, 0.2–0.3 solar13,14, GTD ≈ 650, within the range 340–1,200 inferred for the Magellanic clouds15,16,17. The observed dust mass is fivefold the ∼3,000 of dust expected for a 2 × 106 gas cloud at GTD = 650. If instead we compare the observed dust mass with the dynamical estimate of gas mass, we derive GTD ≈ 47 for the embedded cluster, and in the unlikely case that the cluster is not embedded, GTD ≈ 120. Dust is an expected result of mass loss from massive, short-lived stars. Stellar models indicate that a cluster of massive stars of age 4.4 Myr, consistent with recombination line equivalent widths, will expel 20,000–30,000 of elements carbon, oxygen, silicon, magnesium and iron depending on the cluster mass and initial mass function, of which ∼30–50% will be in the form of dust (see Methods). To produce the amounts of dust and ionizing photons observed, given the upper limit imposed by the dynamical mass, suggests that the stellar initial mass function is top-heavy, with a lower mass cutoff of at least 2–3 (see Methods.) The star cluster has probably produced most of the dust. To infer a gas mass on the basis of the observed dust emission for Cloud D from a GTD scaled to the global metallicity of NGC 5253 without accounting for in situ dust production, as has been done for other galaxies18, would give an erroneously high gas mass and underestimate the star-formation efficiency.

Given the peculiarities of Cloud D, the most reliable gas mass is dynamical. We use this mass to calculate the star-formation efficiency, η. A lower limit to η occurs if the gravitational mass is all gas—no stars—so that . Even in this case, η significantly exceeds the η < 1% of Galactic giant molecular clouds and the highest efficiencies of η ≈ 15–20% seen in individual cloud cores19 in the Galaxy. However, it is almost certain that the star cluster is located within the cloud, given the subarcsecond positional coincidence of nebular emission and CO, the precise kinematic coincidence of nebular H53α (ref. 8) and CO line centroids, and the high extinction to the cluster. Thus the stellar mass also contributes to the linewidth. For the more realistic case that the cluster is embedded within Cloud D, . This value exceeds even the canonical ∼50% (see Methods) needed to allow a star cluster to survive in its current bound state with rapid gas dispersal. If dust competes with gas for ultraviolet photons, the Lyman continuum rate and stellar mass have been underestimated, and η is even higher. If there are winds or outflows contributing to the CO linewidth, η is higher. The large dust mass favours larger-mass cluster models for which η is higher. These values for η are uncertain, but they are free of the systematics due to standard assumptions such as gas-to-dust ratio, relative CO abundance or CO conversion factor. The star-formation efficiency of Cloud D is unusually high, implying gas consumption timescales of ∼10 Myr.

A measurement of star-formation efficiency is a snapshot in time; η could be high because the gas has been incorporated into stars or because the young stars have already dispersed the gas. How and when a young star cluster disperses its gas is crucial to its survival2,20. For Cloud D, gas dispersal models are strongly constrained by the youth of the embedded star cluster10, its positional coincidence with the cloud, lack of evidence for supernovae21, and small CO(3→2) linewidth. Apparently not much has yet escaped this cloud.

Cloud D is a strange molecular cloud: hot, dusty, and small in mass relative to its young star cluster. It is found in a dark-matter-dominated galaxy. Its unusual properties may indicate a mode of star formation different from that observed in disk galaxies, including luminous infrared galaxies. Models of stochastic star formation for turbulently supported giant molecular clouds in our Galaxy suggest that star-formation efficiencies are 1% in a free-fall time22, which implies that the ultimate efficiency can be limited if star formation is quenched by massive stellar feedback. An extended period of star formation might be facilitated if Cloud D is compressed by an external influence, as for example by a streamer of gas force-fed into the star-forming region by the galactic potential. Our data in NGC 5253 could support such a model. The streamer contains ∼2 × 106 of gas extending ∼200–300 pc along the minor axis, entering the galaxy at a rate of ∼20 pc Myr−1. The streamer can fuel star formation at the present rate of ∼0.1–0.2 yr−1 for the next 10 Myr. This dwarf spheroidal galaxy is not rotationally supported23; multiple accreting streams from its extensive H i halo24,25 could be responsible for its global dynamics and morphology26 as well as its spheroidal system of massive star clusters spanning billions of years in age27,28,29. NGC 5253 may illustrate a new mode of highly efficient star cluster formation triggered by cold-stream accretion30.

Methods

Submillimeter Array observations

NGC 5253 was observed with the Submillimeter Array (SMA)31 on 2011 April 17. The observing frequency was νLO = 340.323 GHz with 48 adjacent spectral windows covering 4 GHz bandwidth for each of two sidebands. The CO J = 3→2 rotational transition at ν0 = 345.79599 GHz was in the upper sideband. The array was in the subcompact configuration covering the visibility baselines between 9 and 80 kλ corresponding to the angular scales between 29″ and 2″. The phase centre was αJ2000 = 13 h 39 min 56.249 s, δJ2000 = −31° 38′ 29.00″. Calibration and reduction were performed with MIRIAD32. The instrumental bandpass was corrected using the quasar 3C 279; complex gains were calibrated using the nearby quasar J1316−336; the flux density scale was determined from the planet model of Neptune. Continuum and line emission were separated using the task UVLIN by fitting a linear model to line-free channels. The result is a cube of 25 channels of 10 km s−1 and a continuum map with an effective bandwidth of 8 GHz, convolved to a beam 4″ × 2″, position angle = 0°, shown in Extended Data Fig. 1. Final noise levels were 3 mJy per beam in the continuum map, and 50 mJy per beam in the individual 10 km s−1 channels.

CSO SHARC observations

The 350-μm continuum observations were made with the SHARC camera33 at the Caltech Submillimeter Observatory on 1999 January 11–12, with 225-GHz opacities around 0.035 and 0.075 for the respective dates. The data consist of 2.2 h of on-the-fly mapping with a ∼60″ chopping secondary at 4.132 Hz, and were reduced with CRUSH34, using an enhanced implementation of the Emerson II deconvolution algorithm35, which uses sky rotation to fill in the poorly sampled spatial frequencies of the dual-beam chop. CRUSH removes direct-current detector offset and correlated sky-noise residuals; flatfields detectors based on sky response; and performs noise weighting, whitening and despiking. The main beam is 9″ full width at half-maximum at 350 μm, but the image presented here was smoothed to 12.7″ resolution. From observations of Mars taken immediately before the SHARC observations, at a similar elevation, it is estimated that the pointing is good to ∼5″ root mean squared (r.m.s.). The systematic aperture flux calibration of the 350-μm image is estimated to be good to within 7% r.m.s.

Relation of NGC 5253 to M83 and distance

NGC 5253 is a dwarf spheroidal galaxy of the Cen A/M83 galaxy complex36, with a stellar mass of ∼1.5 × 108 (ref. 37) and an estimated38 total mass, including dark matter, about tenfold higher. It is close to the large spiral galaxy M83 in projection. However, the distance to M83, 4.8 Mpc (ref. 39), is significantly larger than the distance to NGC 5253, at 3.8 Mpc (ref. 40). The H i streamer system24,25 in the halo of NGC 5253, from which the CO streamer seems to emanate, strongly suggests that this dwarf galaxy has had some encounter in its past, but M83 does not seem to be responsible.

Cloud D CO emission

The J = 3 level of CO corresponds to an energy Eu/k of 33 K, a temperature that begins to distinguish actively star-forming clumps from giant molecular clouds. Cloud D is bright in CO(3→2), but only weakly detected in CO(2→1) (ref. 3), and not at all in CO(1→0) (ref. 5). The total flux of CO(3→2) emission in the galaxy and streamer is 110 ± 20 Jy km s−1, about 30% less than the single-dish flux41. This is a typical value for local galaxies, because the array configuration is insensitive to structures <30″ in extent; the value is consistent with the extended streamer emission and with the JCMT-SCUBA continuum image42. The CO(2→1) image is shown in Extended Data Fig. 2, overlaid on the SMA CO(3→2) image. CO(3→2) was not detected in previous SMA observations11 because of insufficient signal-to-noise. Located at αJ2000 = 13 h 39 min 55.943 s ± 0.003 s, δJ2000 = −31° 38′ 25.097″ ± 0.05″, the Cloud D CO(3→2) source is coincident to within ±0.5″ with the core of the supernebula as defined by high-brightness 7-mm free–free emission43. The CO(3→2) line centre is at a heliocentric velocity of 397.5 ± 0.6 km s−1 and the CO flux of Cloud D is 41 Jy km s−1. The size of the CO source in the integrated intensity map deconvolved from the beam is 2.8″ × 1.5″ ± 7%, position angle 12° ± 1°. The slight northward extension is consistent with features seen4,8,21,43,44 in free–free emission, but it is also in the same direction as the elongation of the beams for northern synthesis arrays for this source.

Cloud D virial mass

The width of the CO(3→2) line is σ = 9.2 ± 0.6 km s−1, based on a least-squares fit to a Gaussian line profile. We adopt a value for the radius of half the full width at half-maximum of the geometric mean of the deconvolved source size, using , where v = 2.35σ (in km s−1) and r is in parsecs, with coefficients of α = 190, for ρ ≈ r−1, adopted here, and α = 126 for ρ ≈ r−2 and 210 for ρ ≈ r0 giving the uncertainty limits in Mvir (ref. 45). We assume that the cloud is turbulently supported against gravity and dispersion-dominated, as for Galactic giant molecular clouds46,47; inclination effects should therefore not be important. If Cloud D is not bound, or has flows that are super-gravitational, our estimate for the virial mass is an overestimate.

Supernebula stellar mass

The stellar mass is based on STARBURST99 (refs 48, 49) modelling with the following constraints. First, the 7-mm flux density of the supernebula is 47 ± 4 mJy for the central 2″ (refs 8, 43). From this we obtain a Lyman continuum rate of NLyc = 7 × 1052 s−1 for a nebula at 12,000 K (refs 50, 51). Second, the cluster age must be consistent with Bracket γ equivalent width of 255 Å (refs 10, 52) and mid-infrared ionic line ratios53. Third, the cluster must be old enough to have Wolf–Rayet (WR) stars, which would explain the WR spectral signatures13,14,50,51,54,55. Fourth, the cluster must have mass less than the virial mass of 1.8 × 106, which is also the maximum mass allowed by the [SIV] and Bracket α linewidths56,57. We considered both Geneva high and Geneva v = 40% breakup stellar models; the high-velocity models permit older cluster ages. The main parameter to vary in fitting the models is the lower mass cutoff to initial mass function (IMF). We consider cluster models with standard Kroupa IMFs with upper mass cutoffs of 150, and top-heavy IMFs. Kroupa IMFs with stellar masses down to 0.1 cannot give cluster ages sufficiently old to have both WR stars and the given NLyc, in view of the upper limit on the cluster mass. The IMF must be top-heavy: a cluster of requires a lower mass cutoff of >3. This cluster mass is consistent with previous estimates58,59. The Lyman continuum rate inferred from free–free emission may be less than the true value because of leakage from the H ii region. In addition, dust can absorb as much as 50% of the ionizing photons in dense Galactic H ii regions60; this could also increase the stellar mass. Studies of the extended ionized gas in NGC 5253 (ref. 61) indicate a total galactic star formation rate of twice what we calculate for the supernebula, but this could be ultraviolet photons from nearby slightly older clusters27,28,29,58.

Cloud D continuum and dust mass

The strong 870-μm continuum source towards Cloud D consists of equal parts free–free emission from the H ii region and dust emission. The continuum source is located at αJ2000 = 13 h 39 min 55.948 s ± 0.005 s, δJ2000 = − 31° 38′ 24.88″ ± 0.11″ (J2000). The 870-μm peak is 0.5″ north and 0.32″ west of the 7-mm continuum supernebula core43, which is within the uncertainties of our 4″ × 2″ beam. The continuum source agrees in position and size with the CO source. The total 870-μm flux density of Cloud D is 72 ± 10 mJy. This flux is consistent with previous observations11, and constitutes ∼40% of the total 870-μm flux of 192 mJy for the galaxy as determined from a JCMT/SCUBA map42. We extrapolate a free–free flux density from the 7-mm value of 47 ± 4 mJy (refs 8, 43), using the Sν ∝ ν−0.1 spectrum of optically thin emission, giving = 38 ± 4 mJy for Cloud D. The dust emission is then = S870μm −  = 34 ± 14 mJy. This value is just consistent with the upper limit set at 1.3 mm (ref. 3). The SHARC flux is = 3.7 ± 0.5 Jy, of which ∼1.0 ± 0.2 Jy originates in Cloud D.

The dust mass is dependent on the submillimetre dust opacity and dust temperature. We adopt the opacity of the Large Magellanic Cloud12, extrapolating from κ(160 μm) = 16 cm2 g−1 and β = 1.7 to obtain κ(870) = 0.9 cm2 g−1. The 350-μm and 870-μm fluxes for Cloud D are consistent with β = 1.7. For dust temperature we adopt Tdust = 45 K based on IRAS Point Source Catalog fluxes62; Tdust < 57 K based on IRAS 60-μm flux. These values of temperature and opacity give Mdust = (1.5 ± 0.1) × 104 (T/45 K)−1 for the dust mass of Cloud D, with the uncertainty based on the flux. Previous determinations of the dust mass in NGC 5253 were for the entire galaxy, including the streamer, based on the large-aperture JCMT/SCUBA42 flux density; if scaled to our flux of 34 mJy for Cloud D only, these models59,63 would give a dust mass of Mdust ≈ (2–3) × 104, consistent with a cooler dust temperature.

Dust yield from the cluster and GTD

The GTD estimated from our dust mass of 1.5 × 104 and our gas mass derived from virial and stellar masses is ∼47 (embedded cluster) or 120 (cluster outside cloud). Either is significantly lower than the value of 650 predicted64 from scaling the Galactic value of 160 (ref. 65) to the metallicity of NGC 5253. We argue that the high dust mass is from in situ enrichment by the cluster. The Brackett γ equivalent width of 255 Å (refs 10, 52) indicates a cluster age consistent with the presence of WR activity, so mass loss is expected. If the original progenitor cloud had a mass of ∼2 × 106, there would initially have been ∼3,000 of dust in Cloud D, one-fifth of the amount that we observe. From STARBURST99 models with z = 0.004, Geneva high-mass-loss stellar models, for a 1.1 × 106 cluster of age 4.4 Myr with an initial mass function of range 3–150 and log NLyc = 52.84, one would expect a yield in elements C, N, O, Mg, Si and Fe of ∼24,000, of which an estimated 30–50% would be in the form of dust65,66. If the cluster is more massive or more top-heavy, the mass loss can be higher. The Geneva rotating models give yields that are lower but still sufficient to explain the dust mass. The optical spectrum of NGC 5253 does not yet reflect this localized enrichment: the metallicity of ∼0.25 solar is based on nebular lines13,14 from the extended nuclear H ii region, whereas the embedded star cluster is behind at least 16 magnitudes of extinction. However, the small CO linewidth suggests that what has been produced in the cloud has so far largely stayed in the cloud.

Cloud D excitation modelling

The CO(3→2) to CO(2→1) line ratio can be used to constrain gas density and temperature. Extended Data Fig. 2 demonstrates clear differences between the emission in Cloud D and the streamer. RADEX6 was used to perform non-LTE modelling of the line ratios. We assumed a black-body radiation field and a column density of 1016 cm−2, consistent with the observed brightness. Allowed values of density and kinetic temperature for the adopted value of I32/I21 = 2.25 and the 1σ lower limit of 2.0 are shown in Extended Data Fig. 3. For the value of 2.25, the lower limit on the gas kinetic temperature is TK ≈ 350 K; however, the ratio of 2.1 is within the uncertainties and would give TK > 200 K. The indication of thermal ratios in the near-infrared H2 line ratios7 would also support a high temperature for this cloud.

Mass of Cloud D based on XCO

XCO masses are based on CO(1→0) emission, which has not been detected in Cloud D. Given that the CO emission is optically thin, it seems that the XCO value3 would underpredict the H2 mass by a factor of ∼8.

Streamer kinematics

The streamer has been detected previously in CO(1→0) and CO(2→1) (refs 3, 5). The emission is found at heliocentric velocities of 410–430 km s−1, which is red-shifted by about 20 km s−1 with respect to the galaxy and the supernebula Cloud D. High-resolution VLA images24 also detect H i emission coincident with this streamer. The streamer coincides with filamentary emission in nebular lines of oxygen67 and sulphur68; it has been suggested that this is an ‘ionization cone’, possibly even due to an active galactic nucleus68. We suggest that this emission is due to leakage of photons from the starburst, which are ionizing the surface of the infalling streamer.

Streamer CO and dust properties and GTD

The CO(3→2) emission originates largely from a single cloud, Cloud ‘C’ (ref. 3). For its line strength of I32 = 3.9 ± 0.6 K km s−1, the ratio I32/I21 is 1.0 ± 0.3. RADEX models of the ratio are consistent with optically thick and cooler gas, T ≈ 20 K, with n ≈ 103.5–104 cm−3 (Extended Data Fig. 3). The molecular mass of the streamer based on CO(2→1) (ref. 3) is for a Galactic conversion factor, XCO = 2 × 1020 cm−2 (K km s−1)−1. The virial mass is 3 × 106 (ref. 3). The 870-μm dust continuum emission follows the CO (Fig. 2). The 870-μm flux density of the streamer is 26 ± 8 mJy, which is all dust (Fig. 2). Adopting the dust opacity κ(870 μm) = 0.9 cm2 g−1 and dust temperature Td = 20 K, we obtain a dust mass of Mdust = 2.6 × 104. The streamer thus has GTD ≈ 120 using the virial mass for the gas mass. If the H i gas24 is added, the total H + H2 mass becomes 4.3 × 106, which gives GTD = 170. That the streamer is molecular gas, and—even more surprisingly—dense molecular gas, is difficult to understand. Molecular gas favours high-pressure environments69 such as the midplanes of the central regions of spiral disks. Even though the filament seems to be in a low-pressure environment, it is not only molecular, but also dense. Models of the streamer as an example of a primordial cooling filament, in which the gas collapses towards the centre of the dark-matter potential, are able to produce the observed inflow rate of gas of but are unable to reproduce the formation of the observed giant molecular clouds9. The streamer may be previously enriched gas.

Cloud F

Cloud F, located about 5″ to the southwest of Cloud D, was not detected in previous CO observations3,5. Clouds D and F are the only two detected giant molecular clouds within NGC 5253 proper. Using the Galactic CO conversion factor and assuming optically thick emission, for the observed flux of 17 ± 6 Jy km s−1 we obtain a mass of for Cloud F.

Star-formation efficiency and cluster survival

The canonical value of η = 50% is from virial considerations for the survival of a bound cluster with mass loss on timescales less than the crossing time70. It is possible for a cluster to survive with lower efficiency, to ∼30%, if the gas is lost slowly and the cluster expands2,71.