NGC5253 is host to hundreds of large star clusters6, including dozens of extremely bright, young super star clusters7 (SSCs). Only a few million years old8,9, these clusters are found in the central “starburst”9 region of the nearby (3.8 Mpc) galaxy. Subarcsecond radio and infrared imaging reveal3,4,5 a bright “supernebula” within the starburst, optically invisible, probably powered by a young SSC1,2. The luminosity required to ionize the supernebula is (0.8–1.2) × 109L, where the subscript refers to the Sun, within a 1–2 pc (ref. 1) region—perhaps the most concentrated star-forming luminosity known.

To verify the nature of the supernebula and to study its dynamics, we observed the Brackett α and γ recombination lines of hydrogen at 4.05 µm and 2.17 µm using the NIRSPEC10 on the Keck Telescope on 11 March 2000. Spectra were taken through a 0.579″ × 24″ slit at a spectral resolution of R ≈ 25,000, or about 12 km s-1. SCAM, a 256 × 256 array camera within NIRSPEC, simultaneously imaged the slit location on the galaxy at the K band (2.2 µm), allowing us to pinpoint where the spectra were taken. The seeing was 0.55″–0.8″.

The 2.2 µm broadband image of NGC5253 reveals hundreds of bright infrared star clusters, shown in Figs 1 and 2. From their optical colours, these SSCs are estimated to be only 2–50 Myr in age8,9. We find that the brightest infrared source does not coincide with any of the optical clusters. The brightest 2.2-µm source is offset by 0.3″ ± 0.1″ to the northwest of the youngest8 optical source. This visually obscured K-band source is the location of the strong Brackett line emission that we observe from the supernebula.

Figure 1: Infrared–optical colour view of the young SSCs of NGC5253.
figure 1

The λ = 2.2 µm image (red channel) was constructed from SCAM images made during the night. The infrared image is combined with an optical image from the Hubble Space Telescope (HST) (blue and green channels). Seeing was 0.7–0.8″ for the infrared image; the HST image was convolved to match. The brightest K-band source appears as an extended red source to the north of the dust lane. A gaussian fit to this source gives a size (FWHM) of 1.10″ × 0.9″, position angle 36°. From smaller clusters we estimate the point spread function for the SCAM image to be 0.75″ ± 0.2″. If the K-band source is gaussian, this would imply a source size of 0.8″ × 0.5″, along position angle 40°. This size is uncertain because of variable seeing. The entire SCAM image is 46″ square. The orientation of the image is east to the left, north up.

Figure 2: Echellogram and slit position for Brackett spectra of the supernebula.
figure 2

The 46″ × 46″ SCAM image shows the position of the 0.579″ × 24″ slit on the brightest K-band source in NGC5253. In the inset is the Brackett α echellogram, with frequency/velocity running horizontally and the spatial dimension vertically. The full slit length is 24″, but only 6″ is shown in the inset. The orientation of the image is such that north is at an angle of -43° (clockwise) from vertical. The Brackett line emission is less than 1.3″ in spatial extent on the echellogram, the same as the standard star. Spectra taken at eight other positions showed weak or no emission.

Figure 2 shows the image of the slit position with the strongest Brackett line emission and the corresponding Brackett α echellogram. Line plots of both Brackett lines are shown in Fig. 3. Continuum-subtracted line fluxes are SBrαobs = (3.4 ± 1) × 10-16 W m-2 and SBrγobs = (3.9 ± 1) × 10-17 W m-2 for a region within 3″ of the continuum peak. For comparison, ref. 11 measures SBrαobs = (7.0 ± 1) × 10-16 W m-2 and SBrγobs = (1.5 ± 1) × 10-16 W m-2 in a 10″ × 20″ aperture; as much as half of their Br α flux may be contributed by an unsubtracted continuum. The large aperture flux at the K band appears to be dominated by stellar emission, while much of the L′ flux is from nebular dust. Our photometry and mapping confirm that most, if not all, of the Br α and at least 30% of the Br γ emission in NGC5253 emerges from a source coincident with the brightest K-band continuum source.

Figure 3: Spectra of Br α and Br γ in NGC5253.
figure 3

The spectra were integrated over the central 3″ centred on the brightest infrared continuum source. Each grating setting was followed by a calibration A star. On/off slit SCAM images of the standard indicate that 50% of the light entered the slit. Because the Brackett line emission has a spatial extent comparable to the standard star, 6–9 pixels (0.8″–1.3″), calibration using the standard should automatically correct for this effect. We estimate an uncertainty of 30% in the line fluxes due to variability in seeing. Near-infrared continuum emission from the brightest K-band source is so strong that it can be seen even in these highly dispersed spectra. We measure continuum fluxes of 186 ± 60 mJy and 12 ± 4 mJy at L′ and K, respectively, as compared to the 144 mJy and 20 mJy fluxes in these bands observed in ref. 29 with 8–9″ apertures.

Having both Brackett lines allows us to correct the line fluxes for extinction and obtain a good estimate of the true, extinction-free recombination line flux, and thus the total ionizing flux. For an intrinsic SBrαobs/SBrγobs flux ratio of 2.8 (ref. 12), a temperature of 12,000 K (ref. 13), and a Rieke and Lebofsky14 extinction law, we obtain extinctions of Aα = 0.8 mag at 4.05 µm and Aγ = 2.0 mag at 2.17 µm. The corresponding optical extinction is Av = 18 mag. Optical Balmer recombination lines give Av = 3 (ref. 13). The apparent contradiction between the low optical extinction and higher infrared extinction is resolved if the extinction is high and internal to the nebula. The extinction-corrected Brackett line fluxes are SBrαobs = 7.1 × 10-16 W m-2 and SBrγobs = 2.5 × 10-16 W m-2 for the inner 1.5″. These fluxes predict a 15-GHz free–free flux of 24 ± 8 mJy: the observed 15 GHz flux in this region, corrected for opacity2, is 24 ± 3 mJy (ref. 4). The Lyman continuum rate required to maintain the ionization of the supernebula is NLyc = (4 ± 1) × 1052 s-1, equivalent to 4,000 O7 stars. This agrees with centimetre-wave and millimetre-wave free–free fluxes1,2,4 and radio recombination lines15 (Owens Valley millimetre-wave fluxes2 give a value of 6,000 O7 stars from a larger aperture). Both the excellent agreement of the Brackett line fluxes and radio fluxes and the observed compactness of the emission argue that the Brackett line emission arises from the radio ‘supernebula’.

The velocity information presented here is new; these spectra probe the dynamics of the nebula at high spatial and spectral resolution. The Brackett line profiles were fitted with gaussians, with line centroids at vLSR = 377 ± 2 km s-1 and full-widths at half-maxima (FWHMs) of 76 ± 1 km s-1, consistent with Hα (ref. 16) and radio recombination lines15. For gas at 12,000 K, this linewidth is supersonic. Supersonic gas motions are expected in nebulae: in addition to nebular expansion there is also the interaction of the nebular gas with winds from the massive stars in the cluster. For comparison, compact H ii regions in our Galaxy around small groups of O stars have linewidths of up to 64 km s-1 (ref. 17); nebulae around individual O stars in the Wolf–Rayet phase can have linewidths of 50–200 km s-1 (ref. 18).

The first implication of the linewidth of 75 km s-1 for the ‘supernebula’ is that if the nebula were actually expanding at the implied speed of 38 km s-1, then the mean radius of 0.7 pc (ref. 1) of the nebula implies an implausibly short dynamical age of 7,000 years. Dynamical ages of compact H ii regions in our Galaxy are generally too short to explain the numbers of observed nebulae21. It is thought that confinement mechanisms such as the pressure of dense molecular clouds are responsible for the retardation of their expansion. This could lengthen the lifetime of the supernebula phase, if there were molecular gas nearby, which is as yet undetected2.

The more unusual implication of the linewidth is that although they are supersonic, these nebular lines are remarkably narrow given the size of the nebula and the high luminosity of its embedded star cluster. Brackett line, radio and infrared fluxes all require LOB ≈ (0.8–1.2) × 109L for excitation of the nebula. For a cluster with a Salpeter initial mass function (IMF) and a lower mass cut-off of 1 M, the mass in stars corresponding to this luminosity is (5–7) × 105M; if the IMF extends down to stars less than 1 M, the cluster mass may reach 106 M. The size of the radio nebula is well-determined, and should be the same as the Brackett size because both scale with emission measure; VLA images show that the nebula is 0.9 pc × 1.8 pc in size (0.05″ × 0.1″, ± 0.02″) (ref. 1). We assume that the star cluster exciting the nebula lies within the supernebula, otherwise the implied excitation luminosity would be higher than the total observed IRAS luminosity of the entire galaxy, 1.8 × 109L (ref. 5). For a radius of 0.5–0.9 pc, the escape velocity is 25–30 km s-1 for a cluster of O stars alone; vesc ≈ 50–70 km s-1 for an IMF cut-off at 1 M; and vesc ≈ 85–110 km s-1 for a cluster IMF extending below 1 M. Gravity must play a significant role in the dynamics of this nebula, slowing its expansion, if not halting it. Virial linewidths are only slightly smaller than vesc, so the nebula could be in gravitational equilibrium.

If the ‘supernebula’ stage of SSC formation should prove to be common and long-lived, it may have implications for the detection of Lyman α and other ultraviolet lines in star-forming galaxies. There is evidence that Ly α is detectable primarily in galaxies in which the potential absorbing gas is Doppler-shifted out of the path of the Ly α photons20. This would be most likely in galaxies with superwinds and superbubbles, phenomena which can be caused by large concentrations of hot and windy O stars21. If a significant fraction of the lifetime of the ionizing phase of an SSC is spent in a confined, dust-enshrouded state such as the supernebula, it is likely that these winds will develop only late in the formation process of the cluster, perhaps at the occurrence of the first supernovae. By that time the ionizing flux of the cluster is already declining. SSCs may be an important mode of star formation in the early Universe. Ly α searches for primeval galaxies may therefore fall far short of detecting the true star formation rate, as suggested by observations20,22,23,24.

To put the supernebula in context, we can compare it to a more traditional nebula, 30 Doradus in the Large Magellanic Cloud. 30 Dor is a large, luminous nebula ionized by an optically visible star cluster with an estimated age of 10 Myr (ref. 25), 150–200 pc in extent and with a gas density of 20–100 cm-3 (ref. 26). 30 Dor is 200 times larger and 100–1,000 times less dense than the supernebula in NGC5253, and its exciting star cluster R136 is 10–100 times less massive than the cluster in NGC5253 (ref. 27). The escape velocity for 30 Dor is less than the thermal sound speed of 10 km s-1 so its gas motions are determined by winds and turbulence rather than by gravity28. How the supernebula in NGC5253 will free itself from ‘gravitational bondage’ and evolve into an extended, diffuse nebula like 30 Dor will probably depend on the evolution of its underlying star cluster.