We present an analysis of the 5–8 μm Spitzer-IRS spectra of a sample of 68 local ultraluminous infrared galaxies (ULIRGs). Our diagnostic technique allows a clear separation of the active galactic nucleus (AGN) and starburst (SB) components in the observed mid-infrared emission, and a simple analytic model provides a quantitative estimate of the AGN/SB contribution to the bolometric luminosity. We show that AGN are ∼30 times brighter at 6 μm than SBs with the same bolometric luminosity, so that even faint AGN can be detected. Star formation events are confirmed as the dominant power source for extreme IR activity, since ∼85 per cent of ULIRG luminosity arises from the SB component. None the less an AGN is present in the majority (46/68) of our sources.
Ultraluminous infrared galaxies (ULIRGs, LIR > 1012 L⊙) are the local counterparts of the high-redshift objects dominating the cosmic background in the far-infrared (far-IR) and millimetric bands. Unveiling the nature of their energy source is fundamental in order to understand the star formation history and the obscured active galactic nuclei (AGN) activity in the distant Universe.
Since their discovery, several IR indicators have been proposed to determine whether the central engine in ULIRGs is an AGN or a starburst (SB). The presence of high-ionization lines in the mid-IR spectra of ULIRGs points to AGN activity, while intense polycyclic aromatic hydrocarbon (PAH) emission features are typical of SB environments (Genzel et al. 1998; Laurent et al. 2000). Recently, the absorption feature of amorphous silicate grains centred at 9.7 μm has also been used together with the PAH emission to assess the nature of the obscured power source (Spoon et al. 2007). An alternate way to disentangle AGN and SBs in ULIRGs has been proposed by Risaliti et al. (2006, hereafter R06), based on the separation of the two continuum components in 3–4 μm spectra. This method has been successfully applied to a sample of ∼50 nearby ULIRGs (Risaliti, Imanishi & Sani 2007) and provided an estimate of the average AGN/SB contribution to ULIRGs. The key reason for using the continuum emission at λ≃ 3–4 μm as a diagnostic is the difference between the 3 μm to bolometric ratios in AGN and SBs (approximately two orders of magnitude larger in the former). This makes the detection of the AGN component possible even when the AGN is heavily obscured and/or bolometrically weak compared to the SB. However, the original prescription is limited by the low quality of the available L-band spectra of ULIRGs (Imanishi et al. 2006; R06), which makes the results on individual sources highly uncertain, except for the ∼10–15 brightest objects.
At present, we have extended the analysis to the 5–8 μm spectral band, using the observations of the Infrared Spectrograph (IRS; Houck et al. 2004) onboard Spitzer. We disentangled the AGN and SB contributions to the observed 5–8 μm emission of ULIRGs by combining average spectral templates representing the different properties of the two physical processes at work. The high quality of Spitzer-IRS data, in addition to the relatively low dispersion of the intrinsic continuum properties of both AGN and SBs in this spectral range, allows a much more accurate determination of the AGN/SB components than possible at other wavelengths (e.g. X-rays) or with other diagnostic methods based on emission lines. In this paper, we present our decomposition method, and discuss a simple analytical model providing a quantitative estimate of the AGN/SB contribution to the bolometric luminosity of each source.
Observations and Data Reduction
In order to perform a detailed study of a representative sample of ULIRGs in the local Universe, we selected 68 sources with z < 0.15 and a 60-μm flux density f60 > 1 Jy. Most of the objects are taken from the IRAS ULIRG 1-Jy sample (Kim & Sanders 1998), but a few more sources in the Southern hemisphere have also been included. The flux limit at 60 μm ensures an unbiased selection with respect to the relative AGN/SB contributions.
IRS observations were obtained within three different programs: PID 105 (PI J.R.Houck), PID 2306 (PI M.Imanishi), PID 3187 (PI S.Veilleux). The coadded images provided by the Spitzer Science Centre (after the treatment with pipeline version S13.0) have been background-subtracted by differencing the two observations in the nodding cycle. The spectra have been extracted and calibrated following the standard procedure for point-like sources with the package spice. The flux uncertainties have been estimated from source and background counts (in e−/s). Finally, we performed a smooth connection between the Short-Low spectral orders, with no necessity of relative scaling.
Out of the 68 spectra, we already published 48 in Imanishi et al. (2007). Six more spectra are shown in Armus et al. (2006, 2007). The remaining 14 spectra are analysed here for the first time and will be fully presented in a forthcoming paper (Nardini et al., in preparation).
The 5–8 μm AGN/SB Separation
Despite the diversity of the global IRS spectra of pure AGN and pure SBs, and the complexity of the physics involved, little dispersion is seen at wavelengths shortward of the 9.7-μm silicate feature. This makes possible the use of universal AGN/SB templates to reproduce the spectral properties of ULIRGs in the 5–8 μm interval. In the following, we describe the templates adopted in our model.
Starburst. The mid-IR spectral features of local SB galaxies show very little variations from one object to another in the 5–8 μm wavelength range, while larger differences are present in the ∼9–30 μm band (Brandl et al. 2006, hereafter B06). In order to check if this is the case in the ULIRG luminosity range as well, we analysed the sources in our sample estabilished to be SB-dominated by multiband studies. We did not find significant variations among the spectra, concluding that a fixed template can be used to represent the 5–8 μm SB component in local ULIRGs. We built such template using the five brighest objects among the pure SBs in our sample (IRAS 10190+1322, IRAS 12112+0305, IRAS 17028+0014, IRAS 20414−1651 and IRAS 22491−1808), whose underlying continuum has been reproduced by a power law and normalized at 6 μm before adding. Our SB template is shown in Fig. 1, together with its dispersion in the whole IRS spectral band and the B06 template. The little spectral dispersion below 8 μm among SBs of different luminosity (to be compared with their large differences at longer wavelengths) is in itself an interesting result, which should be fully investigated through detailed emission and radiative transfer models. Concerning this we only note that such a remarkable similarity can result from the spatial integration over a large number of individual star-forming regions.
AGN. Our recent L-band analysis of bright ULIRGs shows that the intrinsic AGN emission is due to hot dust grains and the flux density is well described by a featureless power law with a fixed spectral index: fν∝λ1.5. Here, we adopt the same spectral shape up to 8 μm, in agreement with new Spitzer observations of a large sample of local type 1 quasars (Netzer et al. 2007).
An active nucleus is much more compact than a circumnuclear SB region. As a consequence, the near-IR radiation due to thin dust reprocessing can be itself strongly reddened because of a compact absorber along the line of sight. We therefore introduce an exponential attenuation factor e−τ(λ), where the optical depth follows the conventional law τ(λ) ∝λ−1.75 (Draine 1989). A similar correction is not needed in the SB template. We stress that this does not imply that the SB spectrum is not affected by inner obscuration; the possible effects of this obscuration (which are clear at longer wavelengths, for example, in the silicate absorption features at 9.7 and ∼18 μm) are, however, already accounted for in the adopted observational template.Table 1.
Additional high-ionization emission lines and molecular absorption features (due to ices and aliphatic hydrocarbons), whenever present, were fitted by means of Gaussian profiles except for the water ice absorption at ∼6 μm, reproduced with the laboratory profile from the Leiden data base corresponding to pure H2O ice at 30 K.
Fig. 2 shows the spectral decomposition of three representative ULIRGs. In spite of the great diversity of the observed spectra, our simple model provides a good fit of each spectrum in the sample: the residuals from the best fits are in all sources smaller than 10 per cent at all wavelengths (though the fits are not formally acceptable in a statistical sense, with a reduced χ2≳ 2, due to the small error bars and the remaining unfitted minor features). In particular, both the PAH emission and the continuum are always well reproduced: this implies that the large variations in the 5–8 μm spectral shape of ULIRGs are entirely due to the AGN contribution and its obscuration. A detailed analysis of the results for each source and a physical interpretation of peculiar cases are the subject of a forthcoming paper (Nardini et al., in preparation).
AGN/SB Bolometric Contributions
The large difference between the 5–8 μm to bolometric ratios in AGN and SBs implies that this ratio is itself an indicator of AGN activity, and can be used (i) to test the consistency of our decomposition method and (ii) to estimate the relative AGN and SB contributions to the bolometric luminosity of our sample. We define the 6 μm to bolometric ratio asSanders & Mirabel (1996). Since the integrated IR luminosity of ULIRGs coincides almost exactly with their bolometric luminosity, R is a fair approximation to the fraction of the total energy output that is intrinsically emitted in the 5–8 μm range. Reminding that the intrinsic AGN/SB contributions are α6fint6 and (1 −α6)fint6, respectively, and decomposing FIR as FagnIR+FsbIR, a simple connection between R and α6 is brought out: equation (2). We fitted the theoretical R(α6) relation (equation 3) to our data considering Ragn and Rsb as free parameters, and found: Elvis et al. (1994). This suggests that the local quasars (mostly PG quasars) used to build the mentioned SED can be contaminated to some extent by an SB contribution, in agreement with recent studies (Netzer et al. 2007).
According to equation (4), AGN are ∼30 times more luminous at 6 μm than SBs with the same bolometric luminosity. We are now able to quantify the AGN contribution (αbol=FagnIR/FIR) to the total IR luminosity of each source:Genzel et al. (1998) and Laurent et al. (2000) mid-IR diagnostic diagrams. Considering the whole sample, our results can be compared with the optical classification and with L-band and hard X-ray studies, when available. A substantial agreement is obtained in all cases. It is worth noting that the optical classification alone gives incomplete information: all the sources classified as Seyferts show clear traces of AGN activity, but seven out of eight among the ULIRGs with αbol > 0.25 and τ > 1 are indeed classified as LINERs or H ii regions. LINERs are again confirmed to be rather heterogeneous with respect to the nature of their energy source. Such ambiguities can be solved by applying our diagnostic.
By inverting equation (5) the relation between R and αbol takes the neat form R=αbolRagn+ (1 −αbol)Rsb, and is plotted in Fig. 3(a). As a final check, we have computed for each source the following quantities:
The results are shown in Fig. 3(b) and prove that our decomposition method is reliable in estimating the AGN/SB contributions both to the 6-μm and to the bolometric luminosity of local ULIRGs. In fact, the estimated 6 μm to bolometric ratios of the SB component in the composite sources (which are located at the bottom right-hand side of the plot and are heavily dependent on our modelling) are fully consistent, within the errors, with the ratios of the pure SBs (located at the bottom left-hand side and directly computed from the measured 6-μm and IRAS fluxes). This success is promising in anticipation of a forthcoming study about the role of black hole accretion and star formation in the intense IR activity characterizing the distant galaxies.
However, it is important to keep in mind the following main limitations of our approach.
The narrow wavelength range used in this work simplifies the decomposition analysis, but prevents us from a complete study of the dust composition, density and geometrical distribution. These elements strongly affect the overall mid-IR emission longward of the silicate absorption feature, and can be investigated only through an analysis of the whole IRS spectrum.
While the 5–8 μm templates seem to have little dispersion (as discussed in detail in Section 3), the spread in the 6 μm to bolometric ratios Ragn and Rsb can be much higher, making our estimates of the AGN/SB bolometric fractions more uncertain than those in the 5–8 μm band. The uncertainties in αbol reported in Table 1 are obtained assuming the mean ratios, with the errors on the mean given in equation (4). However, the dispersion around the best fit in Fig. 3(a) is significantly larger. We therefore consider this dispersion (0.3 dex, constant at all values of αbol) as the actual uncertainty in the bolometric ratios of the individual sources. The numerical results on the individuale sources are anyway precise enough to estabilish which is the dominant source of the observed luminosity.
Overall, if we consider as a confidence limit the value αbol= 0.01 (i.e. the dispersion around our SB template), an AGN is present in 46 of the 68 ULIRGs in our sample (including several of those optically classified as H ii regions), but it is significant (αbol≳ 0.25) only in ∼30 per cent of the cases. The SB process is responsible for almost 90 per cent of the observed IR luminosity of ULIRGs, with no significant (i.e. >5 per cent) bias due to the sample selection. A similar fraction holds for the subsample of 34 sources optically classified as LINERs. Our analysis is also consistent with the findings about the nature of high-redshift IR-bright galaxies detected in 24-μm Spitzer–MIPS surveys. IRS spectroscopy shows that they are mostly z∼ 1–3 galaxies, with an apparent bias toward AGN-dominated sources (Houck et al. 2005). This is in agreement with the ∼30 times higher AGN relative emission in the 5–8 μm rest-frame wavelength range we have pointed out.
The use of average templates for AGN and SB emission has allowed us to disentangle the two components in the 5–8 μm spectra of 68 local ULIRGs, observed with the Spitzer Space Telescope. We have been able to detect an AGN in more than 60 per cent of our sources, and estimate its contribution to the bolometric luminosity. In a statistical sense, we confirm that local ULIRGs are powered for ∼85 per cent by intense star formation and for the remaining ∼15 per cent by AGN activity. Our method proves to be successful in unveiling an intrinsically faint or obscured AGN inside a ULIRG. In this context, we also put on a sound basis our initial assumption that the wavelength interval 5–8 μm is an appropriate spectral range in order to search for AGN: an AGN turns out to be approximately 30 times more luminous at 6 μm than an SB with the same bolometric luminosity.
We are grateful to the anonymous referee for his/her helpful and constructive comments. We acknowledge financial support from the PRIN-MIUR 2006025203 grant and the ASI-INAF grant I/023/05/0.