Main

The analysis is based on mid-infrared (5–35-μm) spectra of 48 ultraluminous infrared galaxies7 (ULIRGs) measured using the Infrared Spectrograph8 (data are available at http://ssc.spitzer.caltech.edu) on board NASA’s Spitzer Space Telescope and publicly available through the Spitzer Heritage Archive. The objects were selected from the Infrared Astronomical Satellite 1-Jy sample to include only those ULIRGs that did not have signatures of powerful supermassive-black-hole activity in their optical spectra7. For these objects, I measure the strength of the 10-μm silicate absorption feature (S[9.7 μm]), the strength of 6-μm water-ice absorption (S[6.0 μm]), the fluxes of H2 emission lines, the fluxes of polycyclic aromatic hydrocarbon (PAH) emission features, which are used as star formation indicators9, and the bolometric fluxes, F(IR), of the galaxies (as described in Supplementary Information). In addition, I use other samples of nearby ULIRGs for which some of these measurements are available in the literature3,10,11. As a comparison data set, I use the publicly available spectroscopic data from the Spitzer Infrared Nearby Galaxies Survey (SINGS12; data available at http://sings.stsci.edu), which I analyse using the same methods as ULIRG spectra. Nuclei of SINGS galaxies that show activity associated with accretion onto a supermassive black hole have been excluded from the comparison sample. Therefore, star formation is the likely dominant energy source for both the ULIRGs and the comparison galaxies selected for this Letter.

Even though the starbursts supply most of the energy output of ULIRGs, which is in excess of 1012 solar luminosities, only a small fraction of each ULIRG’s bolometric luminosity is observed as UV and optical continua of young stars1, as most of this emission is absorbed by dust and re-emitted thermally at infrared wavelengths. If the observed PAH and H2 emission is powered by the embedded starbursts, then these features should also be extinguished by the intervening matter. Because the amount of extinction depends on wavelength, different features will be affected by different amounts (Fig. 1); those that fall within the strong opacity peak at 9.7 μm due to astronomical silicates should be affected most. By taking ratios of emission line fluxes within and outside the absorption feature, it is possible to test where the emitting region is located relative to the source of opacity. Because ULIRGs are optically thick even at mid-infrared wavelengths, having a median silicate absorption strength of 1.6, the effect of opacity is expected to be strong.

Figure 1: Wavelengths of emission features present in ULIRG spectra and representative opacity curves.
figure 1

If emission regions are embedded in dust or ice, those emission features that are near the peaks of opacity should be the ones most strongly affected. If H2 emission originates inside silicate dust obscuration (example opacities25,26,27 in black), then S(3) should be strongly extinguished, S(1) somewhat less so, and other lines still less. If PAH emission originates inside obscuration, the features centred at 8.5 μm and 11.3 μm should be more affected by silicates than are the other features, and PAH[6.2 μm] is the only feature that may be affected by water ice (grey; smoothed data for the whole complex28 and laboratory data for water ice29, shown at the maximum strength according to the S[6.0 μm] = 0.6S[9.7 μm] relation; Supplementary Information).

PowerPoint slide

Full size image

Indeed, the observed PAH ratios depend on the strength of absorption (Fig. 2), confirming that the PAH emitting regions are located behind silicates and water ices. This observation provides a quantitative measure of the absorption correction that needs to be applied to the PAHs to derive star formation rates in ULIRGs. The slopes of the observed correlations between PAH emission ratios and strengths and S[9.7 μm] are somewhat flatter than those expected if there is a screen of dust between the emission regions and the observer; that is, PAH emission is absorbed less than expected from the simplest model. This difference suggests that the PAH emitting regions are spread out within the absorbing medium rather than being concentrated at one central point. Because of the extinction, we are more likely to observe light from the PAH emitting regions that are closest to us and are therefore the least obscured.

Figure 2: PAH features are affected by dust extinction.
figure 2

Evidence for the effect of dust extinction on PAH features comes from correlations of the observed ratios of PAH fluxes (a, b) and the PAH/F(IR) ratios (c, d) with the apparent strength of the silicate opacity feature. Spearman’s rank probabilities of the null hypothesis that the plotted values are uncorrelated are P[NH] = 10-4 (a), P[NH] = 10-5 (c) and P[NH] = 10-5 (d). In b, the observed PAH[11.3 μm]/PAH[6.2 μm] ratio (grey points) is uncorrelated with the absorption strength (P[NH] = 0.22); but when PAH[6.2 μm] is corrected for water-ice absorption (black points), the correlation becomes apparent (P[NH] = 2 × 10-3), suggesting that the PAH emitting region is located behind both silicate absorption and water-ice absorption. In a and b, PAH ratios calculated for the comparison sample of nearby star-forming galaxies12,30 are shown by an ellipse whose semi-axes are determined by the standard deviation of the corresponding measure. Although PAH ratios are known to vary as functions of physical conditions30, those seen in ULIRGs are consistent with those found in the comparison sample when extrapolated to low absorption. The dashed lines in a–d illustrate how unobscured ratios would change in the presence of an increasing amount of cold dust between the emitter and the observer for representative opacity curves26,27. In c and d, the model calculation assumes that all absorbed flux is re-emitted at longer wavelengths, but that the total flux does not change. PAHs constitute a greater fraction of the total luminosity output in low-luminosity galaxies (estimates shown with dotted ellipses) than they do in ULIRGs13. The grey shaded areas show the 1σ range in the vertical offset around the best linear fit for ULIRG data. Points denote detections, arrows denote 3σ upper limits, and 1σ standard errors are shown in each panel except b, where individual error bars are omitted for clarity and the median error is shown in the bottom left corner.

PowerPoint slide

Full size image

By contrast, as was previously concluded3, H2 is not affected by absorption. H2 fluxes and ratios fail to show any correlation with the strength of silicate opacity (Fig. 3a–c), even though the S(3) line coincides in wavelength with the peak of the silicate dust opacity curve (Fig. 1) and should be very strongly affected. In the absence of an observable dependence, the question arises of just how strong a correlation would be expected if H2 emission were behind the absorbing material. Molecular hydrogen is known to be present at a range of excitation temperatures3, Texc, and ortho/para ratios4, so even a simple model should take into account a wide range of physical conditions. To calculate the expected H2 line ratios, I assume that dM ∝ T-pdT, where dM is the mass of H2 gas with excitation temperatures in the range [T, T + dT]. For each value of p, I calculate the emitted H2 spectrum for ortho/para ratios between 1 and 3. The values 2.5 < p < 5.0 are sufficient to reproduce the observed range of galactic H2 excitation diagrams (as shown in Supplementary Information). I then calculate the H2 line ratios that would be expected if the H2 emitting region were behind a screen of cold dust and select those of the line ratios that are expected to be most sensitive to extinction.

Figure 3: H 2 emission in ULIRGs is not affected by extinction and shows excess over the H 2 /PAH ratio observed in normal galaxies.
figure 3

The ratios of H2 fluxes (a, b) and the H2/F(IR) ratios (c) for ULIRGs from two samples (points7; crosses3) are uncorrelated with the apparent strength of the silicate absorption. In b, R[J1/J2, J3] is the ratio of the observed flux of the line S(J1) to that expected on the basis of S(J2) and S(J3) assuming all three lines come from a region with a single excitation temperature. Grey areas in a and b and black dashed lines in c show the expected trends of line ratios with apparent silicate strength if H2 emission is behind a screen of dust. Dark grey areas correspond to models with an ortho/para ratio of 3 and a realistic range of excitation temperatures, whereas light grey areas are an extension of the model to include ortho/para ratios between 1 and 3. Although correlations are expected if H2 is affected by silicate absorption, none are detected (P[NH] = 5–75%). In d, if H2 and PAH emission had the same spatial distribution, the H2/PAH ratios would be expected to decrease (dashed lines as in Fig. 2) because dust opacity at the wavelength of the H2 line is greater than that at the wavelength of the PAH feature; but in fact an increase is observed (P[NH] = 0.007 (top), 0.016 (bottom)). The H2/PAH ratios are described better by a model that combines an obscured component of H2 associated with star formation and an unobscured H2 component with luminosity equal to or somewhat greater than that of the obscured component (solid lines: L[H2, outer]/L[H2, inner] = 1 (top), 1.5 (bottom)). Arrows denote 3σ upper and lower limits, and 1σ standard errors are shown for all measurements. Ellipses show flux ratios for comparison galaxies (semi-axes are determined by the standard deviation of the corresponding measure).

PowerPoint slide

Full size image

Even given a generous range of model conditions and a wide intrinsic range of H2 line fluxes and ratios, their dependence on extinction would have been unambiguous if it were present in Fig. 3a–c. It is also noteworthy that models at zero extinction succeed in reproducing the observed range of line ratios, which confirms that the models adequately represent the range of physical conditions in real objects. Using different line combinations, 12 line ratios were computed; none shows dependence on S[9.7 μm]. Neither the line ratios associated with the cooler component (median Texc = 330 K as measured from S(1), S(2) and S(3)) nor those from the warmer component (median Texc = 1,200 K as measured from S(3), S(4) and S(7)) show any evidence for extinction.

The fractional contribution of H2, and that of PAHs (extrapolated to low extinction), to the total luminosity is smaller in ULIRGs than in comparison galaxies. For PAHs, this finding is in agreement with the known nonlinear relationship between bolometric and PAH luminosities13. For H2, there has to be a component directly associated with star formation, just like that in normal galaxies. However, as the extinction is increased this component is quickly extinguished, so we can only see H2 emission that originates outside obscuration. The luminosity of this component may be estimated by observing that the H2/PAH ratios in ULIRGs exceed those seen in normal galaxies (Fig. 3d). The data are well described by a model in which about as much (and up to 50% more) H2 emission is emitted outside the bulk of the obscuration as is directly associated with the starbursts, if the intrinsic PAH/H2 ratios are the same in ULIRGs and in comparison galaxies.

If dust were mixed in with emission regions or had a patchy distribution, the correlations with silicate strength would be expected to be less steep than in the ‘screen-of-dust’ case. If the dust were not completely cold, its thermal emission would partly fill in the absorption feature, making the observed S[9.7 μm] an underestimate of the true optical depth, and the correlations would be expected to be steeper than in the screen-of-dust case. However, such effects cannot change the sign of the expected correlations. Because H2/PAH ratios increase with opacity whereas a decrease is expected from the models (Fig. 3d), any model with the same spatial distribution of H2 and PAHs is ruled out by the data.

The first step in understanding the origin of the rotational H2 emission in ULIRGs is the determination of its spatial extent and, in particular, whether it is confined to the outer parts of starburst regions or is extended on galaxy-wide scales. One way to proceed is to study the spatial distribution of the rovibrational lines in the near infrared, where observational capabilities for spatially resolved spectroscopy or narrow-band imaging are not as limited as in the mid infrared. This approach yielded two successful observations of ULIRGs, one of NGC 624014 and one of Arp 22015. In both cases, the object is composed of two well-separated merging components, and the near-infrared H2 emission peaks between them, near the collision front.

Because most ULIRGs are either actively merging or dynamically disturbed16, these observations suggest the possibility that H2 lines trace shocks excited by intergalactic interactions17. Recent observations with the Spitzer Space Telescope uncovered a class of unusually H2-luminous extragalactic objects18,19,20, in which little or no star formation is seen. One of these objects, a 40-kpc intergalactic shock in Stephan’s Quintet18,21, emits tens of per cent of its bolometric luminosity in rotational H2 lines. A similar process—albeit one producing twenty times more mid-infrared H2 luminosity in a median ULIRG than in Stephan’s Quintet—may be operating in the outer parts of ULIRGs. Further support for this hypothesis is provided by observations of optical line emission from six ULIRGs22, including four from this Letter, in which a contribution from intergalactic shocks is directly seen. The median excitation temperature measured between the S(1) and S(3) transitions for ULIRGs with S[9.7 μm] > 1 in which the unobscured component dominates the observed H2 flux is 306 K, similar to that measured in Stephan’s Quintet between the same transitions (350 K; ref. 21).

Processes other than intergalactic shocks could in principle lead to excitation of H2 emission. However, as discussed in Supplementary Information, excitation by X-rays, by cosmic rays or by emission due to accretion onto the supermassive black hole is unlikely to be dominant producers of the observed H2 emission on energy grounds. Alternatively, H2 could also be excited in shocks other than those produced by galactic collisions. For example, such excitation could be due to supergalactic winds or supernova remnants that, having emptied the environment around them, may be not as obscured as the photon-dominated regions that produce most of the PAH emission. The available data on ULIRGs neither supports nor rules out these processes as the dominant source of H2 excitation. If they turn out to be important, it will be in contrast to lower-luminosity starburst galaxies in which these processes do not dominate H2 excitation23. Future observations of the spatial distribution and kinematics of H2 emission will help distinguish between intergalactic shocks and other processes.

The observations of ULIRGs presented here raise the possibility that excitation of H2 in shocks, and their subsequent cooling through the mid-infrared rotational lines of H2, is a much more common phenomenon than previously thought. Instead of being an astronomical curiosity illustrated by a handful of H2-luminous objects and a few interacting galaxies, such cooling may be occurring under a wide range of conditions, as the large sample of ULIRGs demonstrates. If similar conditions were commonly encountered in the early Universe24, when H2 was one of the primary coolants, cooling by means of H2 emission in shocks could have had a dramatic effect on the formation of the first objects.