The Deep SPIRE HerMES Survey: Spectral Energy Distributions and their Astrophysical Indications at High Redshift
D. Brisbin, M. Harwit, B. Altieri, A. Amblard, V. Arumugam, H. Aussel, T. Babbedge, A. Blain, J. Bock, A. Boselli, V. Buat, N. Castro-Rodríguez, A. Cava, P. Chanial, D.L. Clements, A. Conley, L. Conversi, A. Cooray, C.D. Dowell, E. Dwek, S. Eales, D. Elbaz, M. Fox, A. Franceschini, W. Gear, J. Glenn, M. Griffin, M. Halpern, E. Hatziminaoglou, E. Ibar, K. Isaak, R.J. Ivison, G. Lagache, L. Levenson, Carol J. Lonsdale, N. Lu, S. Madden, B. Maffei, G. Mainetti, L. Marchetti, G.E. Morrison, H.T. Nguyen, B. O'Halloran, S.J. Oliver, A. Omont, F.N. Owen, M. Pannella, P. Panuzzo, A. Papageorgiou, C.P. Pearson, I. Pérez-Fournon, M. Pohlen, D. Rizzo, I.G. Roseboom, M. Rowan-Robinson, M. Sánchez Portal, B. Schulz, N. Seymour, D.L. Shupe, A.J. Smith, J.A. Stevens, V. Strazzullo, M. Symeonidis, M. Trichas, K.E. Tugwell, M. Vaccari, I. Valtchanov, L. Vigroux, L. Wang, R. Ward, G. Wright, C.K. Xu, M. Zemcov
TThe Deep SPIRE HerMES Survey: Spectral Energy Distributions and their Astrophysical Indications at High Redshift The Deep SPIRE HerMES Survey: Spectral Energy Distributionsand their Astrophysical Indications at High Redshift (cid:63)
D. Brisbin, † M. Harwit, B. Altieri, A. Amblard, V. Arumugam, H. Aussel, T. Babbedge, A. Blain, J. Bock, , A. Boselli, V. Buat, N. Castro-Rodr´ıguez, , A. Cava, , P. Chanial, D.L. Clements, A. Conley, L. Conversi, A. Cooray, , C.D. Dowell, , E. Dwek, S. Eales, D. Elbaz, M. Fox, A. Franceschini, W. Gear, J. Glenn, M. Gri ffi n, M. Halpern, E. Hatziminaoglou, E. Ibar, K. Isaak, R.J. Ivison, , G. Lagache, L. Levenson, , Carol J. Lonsdale, N. Lu, , S. Madden, B. Ma ff ei, G. Mainetti, L. Marchetti, G.E. Morrison, , H.T. Nguyen, , B. O’Halloran, S.J. Oliver, A. Omont, F.N. Owen, M. Pannella, P. Panuzzo, A. Papageorgiou, C.P. Pearson, , I. P´erez-Fournon, , M. Pohlen, D. Rizzo, I.G. Roseboom, M. Rowan-Robinson, M. S´anchez Portal, B. Schulz, , N. Seymour, D.L. Shupe, , A.J. Smith, J.A. Stevens, V. Strazzullo, M. Symeonidis, M. Trichas, K.E. Tugwell, M. Vaccari, I. Valtchanov, L. Vigroux, L. Wang, R. Ward, G. Wright, C.K. Xu , and M. Zemcov , Space Science Building, Cornell University, Ithaca, NY, 14853-6801, USA: [email protected] Cornell University and 511 H street, SW, Washington, DC 20024-2725, USA Herschel Science Centre, European Space Astronomy Centre, Villanueva de la Ca˜nada, 28691 Madrid, Spain Dept. of Physics & Astronomy, University of California, Irvine, CA 92697, USA Institute for Astronomy, University of Edinburgh, Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK Laboratoire AIM-Paris-Saclay, CEA / DSM / Irfu - CNRS - Universit´e Paris Diderot, CE-Saclay, pt courrier 131, F-91191 Gif-sur-Yvette, France Astrophysics Group, Imperial College London, Blackett Laboratory, Prince Consort Road, London SW7 2AZ, UK California Institute of Technology, 1200 E. California Blvd., Pasadena, CA 91125, USA Jet Propulsion Laboratory, 4800 Oak Grove Drive, Pasadena, CA 91109, USA Laboratoire d’Astrophysique de Marseille, OAMP, Universit´e Aix-marseille, CNRS, 38 rue Fr´ed´eric Joliot-Curie, 13388 Marseille cedex 13, France Instituto de Astrof´ısica de Canarias (IAC), E-38200 La Laguna, Tenerife, Spain Departamento de Astrof´ısica, Universidad de La Laguna (ULL), E-38205 La Laguna, Tenerife, Spain Dept. of Astrophysical and Planetary Sciences, CASA 389-UCB, University of Colorado, Boulder, CO 80309, USA Observational Cosmology Lab, Code 665, NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA Cardi ff School of Physics and Astronomy, Cardi ff University, Queens Buildings, The Parade, Cardi ff CF24 3AA, UK Dipartimento di Astronomia, Universit`a di Padova, vicolo Osservatorio, 3, 35122 Padova, Italy Department of Physics & Astronomy, University of British Columbia, 6224 Agricultural Road, Vancouver, BC V6T 1Z1, Canada ESO, Karl-Schwarzschild-Str. 2, 85748 Garching bei M¨unchen, Germany UK Astronomy Technology Centre, Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK Institut d’Astrophysique Spatiale (IAS), bˆatiment 121, Universit´e Paris-Sud 11 and CNRS (UMR 8617), 91405 Orsay, France National Radio Astronomy Observatory, P.O. Box O, Socorro NM 87801, USA Infrared Processing and Analysis Center, MS 100-22, California Institute of Technology, JPL, Pasadena, CA 91125, USA School of Physics and Astronomy, The University of Manchester, Alan Turing Building, Oxford Road, Manchester M13 9PL, UK Institute for Astronomy, University of Hawaii, Honolulu, HI 96822, USA Canada-France-Hawaii Telescope, Kamuela, HI, 96743, USA Astronomy Centre, Dept. of Physics & Astronomy, University of Sussex, Brighton BN1 9QH, UK Institut d’Astrophysique de Paris, UMR 7095, CNRS, UPMC Univ. Paris 06, 98bis boulevard Arago, F-75014 Paris, France Space Science & Technology Department, Rutherford Appleton Laboratory, Chilton, Didcot, Oxfordshire OX11 0QX, UK Institute for Space Imaging Science, University of Lethbridge, Lethbridge, Alberta, T1K 3M4, Canada Mullard Space Science Laboratory, University College London, Holmbury St. Mary, Dorking, Surrey RH5 6NT, UK Centre for Astrophysics Research, University of Hertfordshire, College Lane, Hatfield, Hertfordshire AL10 9AB, UK Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138, USA
Accepted 2010 June ??. Received 2010 June ??c (cid:13) , 2– ?? a r X i v : . [ a s t r o - ph . C O ] S e p on. Not. R. Astron. Soc. , 2– ?? (2010) Printed 24 October 2018 (MN L A TEX style file v2.2)
ABSTRACT
The Spectral and Photometric Imaging Receiver (SPIRE) on Herschel has been carrying outdeep extragalactic surveys, one of whose aims is to establish spectral energy distributions(SED)s of individual galaxies spanning the infrared / submillimeter (IR / SMM) wavelength re-gion. We report observations of the (IR / SMM) emission from the Lockman North field (LN)and Great Observatories Origins Deep Survey field North (GOODS-N). Because galaxy im-ages in the wavelength range covered by Herschel generally represent a blend with contribu-tions from neighboring galaxies, we present sets of galaxies in each field especially free ofblending at 250, 350, and 500 µ m. We identify the cumulative emission of these galaxies andthe fraction of the far infrared cosmic background radiation they contribute. Our surveys re-veal a number of highly luminous galaxies at redshift z ∼ < Key words:
IR galaxies: spectral energy distributions: galaxy luminosities
The Herschel Space Observatory (Pilbratt et al. 2010) hasopened wide astronomical access to the far-infrared / submillimeter(FIR / SMM) spectral range. With the Spectral and PhotometricImaging Receiver (SPIRE) (Gri ffi n et al. 2010), deep cosmologi-cal surveys are studying galaxies out to redshifts of order z ∼ ) project (Oliveret al. 2010) have been surveys of galaxies in GOODS-N and North-ern portions of the Lockman Hole (LN) field [see Oliver et al.(2010a) for a description of these early observations.] Source con-fusion, as defined and discussed in detail by Takeuchi, T. T. &Ishii, T. T. (2004), results in blending of far-infrared sources andcomplicates the analysis of survey data. In light of the large de-gree of source blending expected at SPIRE wavelengths, novel op-tions for source extraction have been pursued [(Roseboom et al.2010), (Smith et al. 2010), (B´ethermin et al. 2010)]. Rather thanlooking for sources based on SPIRE intensity maps alone or rely-ing on traditional source detection and extraction techniques for theSPIRE data, which are heavily a ff ected by confusion, Roseboom etal. (2010) measure the SPIRE flux at the position of known 24 µ msources using a linear inversion technique to account for sourceblending. The rationale for this is provided by the results of theBalloon-borne Large Aperture Submillimetre Telescope (BLAST)extragalactic survey (Marsden et al. 2009), which showed that the24 µ m and the FIR flux densities are at least statistically correlated. The aims of this paper are twofold; our primary aim is to de-rive spectral energy distributions for distant galaxies observed bySPIRE. Before this can be achieved, however, a robust way of iden-tifying sources least a ff ected by confusion and blending must bedevised.The GOODS-N catalogue of Roseboom et al. (2010) provides † hermes.sussex.ac.uk † E-mail: [email protected] a cross-identification (XID) of FIR / SMM flux density at 250, 350and 500 µ m with 1951 possible 24 µ m counterparts having mini-mum flux densities of 20 µ Jy. Many of the identified 24 µ m galaxiesare further cross-identified with ultraviolet, optical, near-infrared(NIR) and radio counterparts. The survey covered a 12 . × . ∼
230 arcmin The SPIRE beam diameters at full-width-half maximum(FWHM) respectively measure 18.1, 25.2 and 36.9 arcsec at 250,350 and 500 µ m. For present purposes, we take the beams to beclose to circular; their ellipticity varies from pixel to pixel, but isapproximately 1 . ± .
05, the longer direction lying in the space-craft horizontal direction, parallel to the ecliptic plane (BernhardSchulz private communication). The beam at 500 µ m thus has anarea ∼ . With 1951 possible 24 µ m sources, we can ex-pect a typical 500 µ m beam to contain 2.5 possible sources. At 250 µ m the crowding is a factor of 4 less severe, but still appreciable.An example of the crowded source distribution is seen in Figure1 where the SPIRE beam outlines are overlaid on a patch of theGOODS-N field centered on a 24 µ m source.In the Northern Lockman region, a 40.1’ × µ m counterparts with mini-mum flux densities of 50 µ Jy in an area subtending ∼ — again corresponding to more than 1 potential 500 µ m sourceper beam. Identification of sources least a ff ected by confusionand blending is therefore important. These sources have also beencross-identified at multiple wavelengths and assigned photometricredshifts as detailed in Strazzullo et al. (2010). Using the existing high spatial resolution Spitzer 24 µ m data andthe known far-infrared instrumental point response function (PRF)as inputs, Roseboom et al. (2010) determined best-fit 250, 350, and500 µ m fluxes by a procedure they detail in their paper. In eachSPIRE wavelength band, their tabulated cross-identifications pro-vide both their best estimate of the flux density F ν and the fluxdensity PRF ν in a PRF-convolved map centered on the position ofan associated 24 µ m source. They make no assumptions about a Herschel is an ESA space observatory with science instruments providedby European-led Principal Investigator consortia and with important partic-ipation from NASA. hermes.sussex.ac.uk c (cid:13) he Deep SPIRE HerMES Survey: Spectral Energy Distributions and their Astrophysical Indications at High Redshift proportionality between 24 µ m and FIR flux densities, but assumethat SPIRE sources will only be detected at positions of 24 µ msources. For each source, ancillary data at other wavelengths areincluded, as well as several flags that can be used to identify de-generate cases. Of particular importance is the redshift of the asso-ciated 24 µ m source, which enables derivation of rest-frame SEDsand thus source luminosities. Wherever we refer to F ν and PRF ν inthe remainder of this paper these quantities are to be thought of asthose defined by Roseboom et al. (2010).The flux density in a PRF-convolved region centered on theposition of the 24 µ m galaxy associated with each far-infraredsource represents the system response not only to the flux densityattributed to this source (referred to as F ν (cid:12)(cid:12)(cid:12) λ ) but also to contribu-tions from nearby sources. It thus provides a measure of blendingcharacterizing each source. We define a purity index Π λ for eachsource as the ratio Π λ = F ν / PRF ν (cid:12)(cid:12)(cid:12) λ , (1)where λ specifies the wavelength band, 250, 350, or 500 µ m, and PRF ν (cid:12)(cid:12)(cid:12) λ , as supplied in the XID catalogs by Roseboom et al. 2010is the PRF-smoothed flux density at the position of the source. Ahigh value of Π λ indicates low confusion and blending in wave-length band λ ; a low value indicates high blending. The fractionalcontributions by ambient sources to the PRF-convolved flux den-sity within a PRF is simply (1 − Π λ ). In principle, the purity indexmust assume values 0 ∼ < Π λ ∼ <
1. In practice, however, the value F ν has been calculated on top of a locally determined background,whereas the PRF ν values do not take a local background variationinto account. This can result in Π λ > We have found it useful to identify sources whose purity indicesrespectively exceed Π λ measures of 0.7, 0.5, and 0.3 at 250, 350,and 500 µ m. We say that a source is secure at each wavelength if itmeets this criterion. If the source is found to be secure in all threewavelength bands, we call it triply secure . High-purity sources areof special interest because their isolated nature makes them lesssusceptible to blending by neighbors. It is these sources which willbe especially useful for follow up studies with other instrumentsand also may provide confirmation of the deblending approachused. It should be noted, however, that a flux density estimate froma highly pure source might still be inaccurate if there is signifi-cant contribution from an infrared source that is not observed at 24 µ m. Furthermore, sources with low purity do have well-defined de-blending solutions and hence well-characterized flux densities anduncertainties from Roseboom et al. (2010). In crowded fields, thetrue flux density is very likely described by this characterization,although the margins of uncertainty tend to be large. Our choiceof purity criteria is somewhat subjective but o ff ers a reasonablecompromise for extracting relatively reliable SEDs despite sourceblending. Understandably, these criteria may be expected to varydepending on the type of information an astronomer expects to ex-tract from the survey data. Our
PRF ν | λ corresponds to the quantity d in equation (2) of Roseboomet al. 2010, convolved with the point response function centered on the pri-mary source whose flux is F ν | λ . The entries in the XID tables list our F ν | λ as F ( λ ), and our PRF ν | λ flux density as PRF ( λ ). When a reliable redshift is available, the single most importantquantity that can be determined from an SED is source luminos-ity. With this, one can begin discussing the luminosity distributionat specific redshifts, as well as luminosity evolution as a functionof redshift, particularly among ultraluminous galaxies that emit thedominant fraction of their energy in the infrared. However, to obtaina reliable SED and thus a reliable luminosity, we require sourceswhose flux densities are well determined at all three SPIRE wave-lengths in order to optimally constrain the flux density defining thebroad wavelength region around peak emission.To explain the consequences of our choice of purity criteria inthis context, we may consider a toy model which, as pointed out inSection 1, will respectively exhibit an average number of sources n λ ∼ .
5, 1.225, and 0.625 per GOODS-N beam, at 500, 350, and250 µ m. Let us inject an additional source into such a beam and callit the primary source. If all the sources involved are equally bright,on average, i.e., make equal contributions to the PRF-smoothed fluxdensity, the purity of the primary source will be Π λ = (1 + n λ ) − ,i.e. 0.29, 0.45, and 0.62, respectively at 500, 350 and 250 µ m. Halfthe sources in each waveband will have purities higher than thesepurity cuts, and half lower.Turning now to our preferred adoption of purity cuts of 0.3, 0.5and 0.7 at 500, 350, and 250 µ m, we see that they assure two prop-erties: (i) that they yield sources whose purities are above average atall three wavelengths, and (ii) that the fraction of sources with pu-rity above the cut is roughly comparable at all three wavelengths —a balance, which is important to assure a well-defined SED. Table1, described below, confirms these traits for the GOODS-N sample.It shows that a fraction f λ = .
23 of the sources has purity exceed-ing 0.7 at 250 µ m, a fraction 0.32 exceeding purity 0.5 at 350 µ m,and a fraction 0.36 exceeding purity 0.3 at 500 µ m. These fractionscluster around a value of 0.3, thus lending roughly equal weight tothe flux density in each waveband in the determination of the SED.In Table 1, we list the fraction of sources in GOODS-N andLN whose purity indices lie above certain cuts. We permit theseindices to slightly exceed a value of one, with a cut-o ff of Π = . Π = . Π = .
1, thenumber of triply secure sources rises to 59. In LN there are 633sources with detections at all five wavelengths with known red-shift, 165 of which are triply secure; this number increases to 287if we remove the upper limits on purity. Although the numbers ofthese galaxies are quite modest, they nevertheless yield informa-tive statistics on the luminosities and luminosity distributions ofgalaxies observed out to redshifts z ∼
3. These will be discussed inSection 6.The larger aperture of the Herschel telescope and the higherspatial resolution this enables have permitted the SPIRE surveys toreach depths beyond those attained by BLAST. Nevertheless, Mars-den et al. (2009) succeeded in acquiring reliable measurements ofstacked source flux densities at comparable wavelengths. Their re-sults indicate surface brightnesses of 8.60 ± ± ± − sr − at 250, 350, and 500 µ m respectively.These stacked source flux densities represent the major compo-nent of the cosmic infrared background (CIB) measured by theCosmic Background Explorer’s Far-Infrared Absolute Spectrom-eter (FIRAS) to be 10.4 ± ± ± − sr − at 250, 350, and 500 µ m respectively (Fixsen et al. 1998). To c (cid:13) , 2– ?? D. Brisbin et al.
Table 1.
Fraction of detected SPIRE sources in GOODS-N and LN with a Π λ value within the range indicated in the top row. The columns marked“detections” denote the total number of SPIRE sources detected at a givenwavelength in the current HerMES survey.GOODS-N1.1 > Π > µ m 0.106 0.231 0.348 0.493 1032350 µ m 0.069 0.199 0.316 0.451 697500 µ m 0.061 0.141 0.227 0.362 475LN1.2 > Π > µ m 0.275 0.435 0.579 0.703 4646350 µ m 0.184 0.343 0.500 0.670 2968500 µ m 0.144 0.281 0.419 0.570 2127 investigate the extent to which this background is resolved withHerschel, we summed the estimated flux densities for our secureSPIRE sources in the deepest field (GOODS-N) and then attributedthis flux density to the entire survey region of 230 arcmin . Usingthe flux densities for sources contained in the XID catalog of Rose-boom et al. (2010), our cumulative surface brightness for GOODS-N at 250, 350, and 500 µ m is 1.49, 0.70, and 0.41 nW m − sr − or 14%, 13%, and 17% of the estimated CIB. At 250 and 350 µ mthese values are within 1 σ of those corrected for blending and in-completeness by Oliver et al. (2010a). Figures 2 and 3 exhibit SEDs for triply secure sources in LN andGOODS-N. For LN sources, we have set an additional criterionfor inclusion, namely that they have observed flux densities also atPACS wavelengths of 100 and 170 µ m. Along with examining theobserved SED, we show a fit using starburst models developed bySiebenmorgen & Kr¨ugel (2007) (S&K). These models are basedon a nuclear concentration of massive young stars embedded in amatrix of gas and dust referred to as “hot spots”. S&K use a fiveparameter SED fit which incorporates a variable nuclear bulge sizewith old and new stellar components as well as the e ff ects of dust.They provide their models in the form of a library of 7000 SEDsavailable as text files online. By using their models, we are able tonot only find realistic intrinsic luminosities, but also star formationrates (SFR) for highly luminous sources at high redshift for whichthe Kennicut infrared - SFR relations apply (Kennicutt 1998).The S&K model fits observations quite well, although someof our SEDs exhibit considerable deviations from the data at vis-ible and near infrared wavelengths. This is largely due to variableshielding of starlight by dust, see section 6 below.In Figures 2 and 3, we have focused on high redshift (z > The S&K SED library of models is available athttp: // / ∼ rsiebenm / sb models / . Figure 1.
A GOODS-N SPIRE and 24 µ m source with equatorial coordi-nates (J2000.0) 189.1506 / µ msources. Known ambient 24 µ m sources are shown as squares with nestedsymbols. Plus signs indicate sources with known redshifts and triangles in-dicate sources with unknown redshift. The spatial distribution of 24 µ msources is based on the XID information in Roseboom et al. (2010). Solidcircles represent the full-width-half-maxima (FWHM) of the Airy profilePRF at the three SPIRE wavelengths and dashed circles represent the sec-ond Airy minimum. Coordinates are in J2000. most current models, however, vastly di ff erent dust masses and nu-clear sizes are able to yield similar SEDs, as witnessed by the largedi ff erences in dust masses assigned to some galaxies listed in Ta-bles 2 and 3 that have nearly identical redshifts and luminosities.This is not surprising because the models are only required to pro-vide su ffi cient dust to convert most of the visible and ultravioletradiation produced in the starburst into infrared radiation at the ob-served temperature. If this criterion is satisfied, the models produceroughly correct SEDs.While these figures and tables emphasize the relatively fewtriply-secure SEDs among high redshift objects, it is important tonote that for many statistical trends, triple-security is not neces-sary. In Figure 4, we show the luminosities we have determined asa function of redshift, based on the S&K models for all sources inGOODS-N and LN that have detections of any kind (secure or not)at all three SPIRE bands. While some of the flux densities, espe-cially at 500 µ m, are quite uncertain (see Figures 2 and 3), this willnot greatly a ff ect the overall luminosity distribution. We estimatethe source luminosity uncertainty to be ∼ Λ = .
7, H =
70 km s − Mpc − cosmology).Throughout this paper we refer to multiple luminosities,namely total bolometric luminosity and infrared luminosity inte-grated over 8 - 1000 µ m [as used in star formation estimates byKennicutt (1998)]. For our purposes, the di ff erence between thetwo is small as the majority of a luminous star-forming galaxy’senergy is emitted in the infrared. In galaxies with L IR ∼ > . L (cid:12) ,Buat et al. (2010) find that ∼
95% of the total star formation rate isaccounted for by L IR . Nonetheless, we explicitly di ff erentiate be-tween the two when the distinction is significant. As Figure 4 attests, our surveys reveal a number of highly luminoussources, mainly at redshifts between z = . c (cid:13) , 2– ?? he Deep SPIRE HerMES Survey: Spectral Energy Distributions and their Astrophysical Indications at High Redshift all of these in detail is beyond the scope of the present paper, if forno other reason than that information on these sources is still quitemodest. Nevertheless, we here discuss four of these sources to showthat their luminosities appear to be intrinsic. Neither gravitationallensing nor blending from neighboring sources appear to contributesignificantly to the observed luminosities. This may not be true ofall of the luminous SPIRE sources, but it appears to be so for atleast those about which we have the most information right now. Anumber of other uncertainties also warrant comment.(i) For most of the sources observed in LN only photometricredshifts are available. For the more luminous sources shown, es-timated redshift errors range from ∼ + z ) , a 17% error at z = . ∼ − ff use background,this distinction may not be of primary importance; but for chartingthe luminosities of distant sources their intrinsic luminosities needto be determined.(iii) Because our deep surveys are highly sensitive, they re-veal a large number of faint sources. This leads to potential mis-identification of sources. It can also lead to blending. But seven ofthe eleven ultraluminous sources in Figure 5 and Table 4 are triplysecure; the remaining four are doubly secure within observationaluncertainties. Severe blending is thus unlikely.(iv) For some of the sources, the available data points strad-dle but do not directly constrain the region where the SED reachesa maximum (see Figure 5), so that our SED-derived luminositiescould be over-estimated.At the present stage of data reduction we are not yet in a posi-tion to account for all these uncertainties. However, for the citedsources in GOODS-N and for a few of the sources in LN reli-able spectroscopic redshifts are available. Among four of these wewere able, below, to search for potential lensing, assess a degree ofblending, and justify our confidence in their derived luminosities insome detail. These sources are referred to by letters correspondingto their designations in Table 4. Note that in Table 4 we give bolo-metric luminosities whereas below we quote infrared luminosities.The LN source (j) at equatorial position (J2000.0) 161.554052 / z = .
037 determined by the strong Ly- α line. The visi-ble continuum flux in this spectrum is ∼ × − erg cm − s − Å − ,equivalent to ∼ × − Jy, stretching down to ∼ ,
000 Å, for atotal flux of roughly 1 . × − W m − , which is comparable to theinfrared flux observed from the source. The IPAC extragalactic database lists the optical source as brighter, by more than a magnitude,than any other source within a radius of an arc minute. Trouille etal. (2008) list a 0.5-2 keV X-ray flux of 38 . × − W m − , and a2-8keV flux of 22 . × − W m − for the source, jointly about 20times lower than the visible flux. Owen & Morrison (2008) detecteda 20-cm continuum radio flux density of 64 µ Jy from this source,within an apparent size < . ∼
34 arcsec.Because of their relatively large displacement, these and other am-bient galaxies are unlikely to contribute appreciably to the SPIREflux densities assigned to LN (j) by Roseboom et al. (2010). All thisgives confidence that both the visible and the infrared flux comefrom the same source, that there is no lens magnification, and thatwe are indeed dealing with an un-lensed ultraluminous source withinfrared luminosity ∼ . × L (cid:12) . The simplest explanation forthese data is that we are viewing a QSO with a surrounding torusalong a sightline coinciding with the torus axis. The visible lightreaches us directly along this axis; the infrared emission comesfrom dust heated in part by the QSO and possibly also by massivestar formation.The GOODS-N source (c), with equatorial coordinates(J2000.0) 188.990097 / z = . µ m flux den-sity listed in the Barger catalog is 109 µ Jy. The rest-frame 2-8keV luminosity assigned to this source by Trouille et al. (2008)is 3 . × W ∼ . × L (cid:12) , but their paper provides no rest-frame 0.5-2keV luminosity. The nearest SPIRE source listed in theXID catalogs lies at a distance of 35 arcsec, where its contributionto our primary galaxy’s flux listed in the XID catalog can at mostbe minor. Our primary source also displays weak X-ray fluxes, butMorrison et al. (2010) list no 20-cm source within 3.5 arcmin. In-tegrating the flux densities indicated by the fitted SED in Figure 5,leads to an infrared luminosity of 6 . × L (cid:12) .One of our ultraluminous sources [LN (f)] has previously beendiscussed by Polletta et al. (2006). They observed the LockmanSWIRE source at (J2000.0) 161.041521 / . ± . × − erg cm − s − in the 0.3- 8 keV range. The Spitzer 24 µ m flux is 4.0 mJy, strong for asource at spectroscopic redshift 2.54, and much brighter than any-thing within an arcminute of its location. Because of its initial de-tection by Spitzer, the authors characterize the source as an infraredselected Compton-thick AGN on the basis of the rest-frame hydro-gen column density, which they estimate to be N H ∼ × cm − with an uncertainty envelope extending a factor of three times lowerand arbitrarily higher. The infrared luminosity derived on the basisof our SPIRE and PACS observations (see Figure 5 and Table 4) is2 . × L (cid:12) . It appears to be fairly well isolated in the infrared, thenearest comparably bright 250 micron source being located half anarcminute away.Some of these ultraluminous galaxies could be lensed but afirst look has not yet revealed these in our sample. The GOODS-N source (d), with coordinates (J2000.0) 189.309509 / z = . ff ect the SPIRE fluxattributed to our source of primary interest. Another source only ∼
10 arcsec away is also noted in the NASA / IPAC Extragalac-tic Database (NED). This appears not to have a measured 24 µ mflux and is not listed by Barger et al. (2008). However, Law et al.(2007) have included this source in their discussion of distant ir-regular galaxies. The object designated as BX 150 appears elon-gated roughly along a north / south direction, is ∼ . z = .
28. Atoptical wavelengths, it is 1.3 magnitudes fainter than the ultralumi-nous infrared source and, at its rather high displacement of ∼ ∼ . × L (cid:12) .Figure 5 provides the SEDs of these ultraluminous sources. c (cid:13) , 2– ?? D. Brisbin et al.
Our combined surveys of GOODS-N and LN cover ∼ .
47 squaredegrees, or one part in 85,000 of the sky. Given that we observe sev-eral high-luminosity sources in the small area covered, it suggeststhat approximately 10 sources in the infrared luminosity range ∼ L (cid:12) should be observable in the Universe, at redshifts z = . ff ects inherent in our obser-vations cast this conclusion into serious doubt:(i) The first is that Figure 4, on which the conclusion is based,only includes sources detected at all three SPIRE wavelengths. Forhighly redshifted sources, the 100 µ m infrared emission peak isredshifted into the 500 µ m range, favoring the detection of galaxiesat all three wavelengths, including 500 µ m.(ii) However, compensating for this e ff ect, lower redshiftsources are more readily detected by a factor inversely proportionalto luminosity distance squared. Although these two e ff ects partiallycancel, high-luminosity sources should be more readily detected atlow than at high redshifts.(iii) The XID catalogs search for SPIRE sources solely at loca-tions where Spitzer 24 µ m sources exhibit flux densities (cid:62) µ Jy inthe GOODS-N field and (cid:62) µ Jy in LN. We may thus be missingsources at redshifts at which poorly emitting spectral regions areredshifted to 24 µ m. At redshifts z ∼ .
4, for example, the 9.7 µ msilicate absorption dip shown by Spoon et al. (2007) to be prevalentin many ULIRGS is shifted to 24 µ m. This may account for thestriking absence of low-luminosity sources, at z ∼ .
4, i.e. the lackof sources hugging the luminosity distance curve at this redshift inFigure 4.
Far-infrared surveys with Herschel need to take source confusionand source blending into account, particularly at the longest wave-lengths, 350 and 500 µ m. GOODS-N and LN are the two deepestsurveys undertaken as part of the HerMES project to date. In thesedeep fields, crowding of sources presents especially serious prob-lems. In establishing a set of criteria that assess source blending,we have taken a preliminary step toward estimating the utility ofthe survey data for di ff erent purposes. This has proven itself use-ful in our analysis of the ultraluminous galaxies, some of whichwe described in Section 6 and whose characteristics are exhibitedin Figure 5 and Table 4. In view of the high infrared luminositieswe find, it is particularly satisfying that seven of the eleven sourcescited turn out to be triply secure, i.e. with high purities in all threeSPIRE wave-bands, and that five of the sources also are observedby PACS where blending is not severe, particularly in the 100 µ mwaveband. In this context, we have placed no upper limit on ac-ceptable values of Π , which are especially high for GOODS(c) andLN(f), suggesting especially low ambient source contributions attheir locations.In compiling the SEDs for GOODS-N and LN, we haveelected to work with the S&K models because they are based ona limited set of well-defined physical parameters. The models thusmake predictions that our SEDs may be able to verify, refute, or ex-tend. S&K do not specifically address the e ff ects of adding an AGNcomponent to a starburst model. However, they do provide a star-burst fit for NGC 6240 and propose that addition of a small AGNcomponent could provide an improved fit. Most starbursts generallyalso exhibit some AGN activity. Perhaps because of this, the S&Kmodels appear to provide reasonable fits. The major weakness of Table 5.
Ratios ( R ≡ F FIR / F optical ) of SPIRE flux densities (consistentlymeasured at 250 µ m to minimize source blending) to optical flux densitiesat a rest wavelength ∼ N / R Bright ) to that of the N / R Dim )where N is the total number of L > L (cid:12) sources.z N R Bright R Dim R Bright / R Dim . < z < .
05 18 64500 72000 0.901 . < z < . . < z < . the S&K models, as well as that of all others, tends to be the di ffi -culty in accounting for the seemingly random relationship betweenthe infrared and optical portions of the SEDs that is so apparent inFigures 2, 3, and 5.We investigated the relationship between flux-density ratios atoptical and far-infrared wavelengths in high- and low-luminositygalaxies. Current theory suggests that starbursts involve stellarmass distributions obeying the Salpeter initial mass function (Zin-necker & Yorke 2007). The drop in luminosities from the most mas-sive O type stars with mass ∼ M (cid:12) to the early B type stars at 10 M (cid:12) , can then be shown to be roughly in a ratio of 500:1, i.e. witha contrast considerably higher than that of the mass ratio, roughly12:1. The highest mass ranges will thus be depleted most rapidly,ending their lives in supernova explosions in which at least someof the dust will be destroyed or expelled from the galaxy. The mostluminous galaxies found using our SEDs and their associated red-shifts would thus be expected to be the very youngest as well asthose most densely shrouded by dust, i.e. having the lowest frac-tional optical luminosities. To test this hypothesis we restricted our-selves to galaxies with total bolometric luminosities, L > L (cid:12) ,as these have long been considered likely starburst mergers, albeitwith potential contributions from AGNs (Sanders & Mirabel 1996).In Table 5, we compare the flux density ratios for the most and leastluminous sources in three redshift bins. It is evident that larger ra-tios correspond to more luminous sources at all redshifts but thatthese di ff erences greatly diminish at lower redshifts. This finding isboth new and significant. It indicates evolutionary trends that mayneed to be incorporated into more advanced models of starburstsdesigned to yield SEDs which not only mirror observed ratios ofoptical to infrared emission, but also define a galaxy’s place in itsevolutionary history.Figure 4 provides a capsule history of galaxy evolution overcosmological epochs for the sample included in our two deep sur-veys. The shapes of these distributions are nearly identical in LNand GOODS-N, which motivated us to plot both in the same figure.A glance confirms that sources having the highest luminosities arefound at highest redshifts, i.e. earliest epochs. Luminosities higherthan 10 L (cid:12) are generally observed at redshifts z ∼ − .
2, thehighest redshifts reached in our surveys.
Confusion, which can be troubling at 250 µ m, becomes increas-ingly severe at 350 and 500 µ m. Yet the data at these longer wave-lengths are particularly important given how little is known aboutthis spectral domain. We believe that the triply-secure sources listedin Tables 2, 3, and 4 will be in demand for follow-on studies that X- c (cid:13) , 2– ?? he Deep SPIRE HerMES Survey: Spectral Energy Distributions and their Astrophysical Indications at High Redshift ray astronomers, spectroscopists, and others may wish to undertakeon sources known to be especially free of confusion.With SPIRE photometry data in hand along with cross identi-fications at several shorter wavelengths, we have constructed SEDsfor a handful of trustworthy sources in the GOODS-N and LN re-gions. Many of these can be fit by starburst SED models, such asthose created by Siebenmorgen & Kr¨ugel (2007), which yield in-formation on luminosity, dust mass, and size. Figures 4 and 5 showa number of ultraluminous galaxies with L IR ∼ L (cid:12) . Althoughthese are extreme systems, they do not appear to deviate from thegeneral distribution at high redshift. A major strength of the deepHerMES surveys is their ability to obtain reliable source luminosi-ties and star-formation rates based on flux densities in the infraredand at auxiliary wavelengths as well as redshifts compiled in theXID catalogues. ACKNOWLEDGMENTS
This work is based in part on observations made with Herschel, aEuropean Space Agency Cornerstone Mission with significant par-ticipation by NASA. Support for this work was provided by NASAthrough an award issued by JPL / Caltech.SPIRE has been developed by a consortium of institutes led byCardi ff University (UK) and including Univ. Lethbridge (Canada);NAOC (China); CEA, OAMP (France); IFSI, Univ. Padua (Italy);IAC (Spain); Stockholm Observatory (Sweden); Imperial Col-lege London, RAL, UCL-MSSL, UKATC, Univ. Sussex (UK);and Caltech / JPL, IPAC, Univ. Colorado (USA). This developmenthas been supported by national funding agencies: CSA (Canada);NAOC (China); CEA, CNES, CNRS (France); ASI (Italy); MCINN(Spain); SNSB (Sweden); STFC (UK); and NASA (USA).We thank the referee of this paper, Stephen J. Messenger, forhis incisive comments and helpful recommendations.This paper has been typeset from a TEX / L A TEX file prepared by theauthor.
REFERENCES
Barger, A.J., et al., 2008, ApJ, 689, 687B´ethermin, M., et al., 2010, A&A,516, 43Buat V., et al., 2010, MNRAS, this issueFixsen, D. J., et al., 1998, ApJ, 508, 123Gri ffi n, et al., 2010, A&A 518, L3Kennicutt, R., 1998, Ann. Rev. A&A, 36, 189Law, D. R., et al., 2007, ApJ 656, 1Marsden, G., et al., 2009, ApJ, 707, 1729Morrison, G. E., et al., 2010, ApJS 188,178Oliver, et al., 2010, A&A 518, L2Oliver, et al., 2010a, in preparationOwen, F. N. & Morrison, G. E., 2008, AJ 136,1889Pilbratt, et al., 2010, A&A 518, L1Poglitsch, et al., 2010, A&A 518, L2Polletta, M., et al., 2006, ESASP 604, 807Roseboom I., et al., 2010, MNRAS, this issueSanders, D. B. & Mirabel, I. F., 1996, ARA&A 1996, 34, 749Schulz B., 2010, private communicationSiebenmorgen R., Kr¨ugel E., 2007, A&A, 461, 445Smith A. J., et al., 2010, MNRAS, this issueSpoon, H. W. W., et al., 2007, ApJ 654, L49 Figure 2.
Lockman N galaxy SEDs plotted with arbitrary flux density o ff -sets. These galaxies all met or exceeded our selection criteria of being se-curely identified at all three wavelengths, having a high purity index, (30%,50%, and 70% pure at 500, 350, and 250 µ m respectively), having knownPACS detections at 100 and 170 µ m, and having known redshifts, z > . µ m measurement is compatible with zero, so we have plotted onlythe upper error bar. The solid line is an S&K model fit to the 24 µ m through500 µ m error-weighted observations. We weighted the 24 µ m observationsonly a quarter as heavily as the longer wavelengths which play a dominantrole in determining starburst luminosities. Observational data shortward of24 µ m is plotted for reference but not used in fitting. Source Π s are shown inTable 2. As with many of the examples quoted by S&K, the visual compo-nent of the SED often needs to be fitted by hand because it bears little rela-tion to the starburst characteristics responsible for the mid- and far-infraredflux densities. The extent to which visible stars may or may not contributeto the SED is determined in part by the degree to which an older populationof stars is obscured by dust without significantly contributing to its heatingor by massive young stars whose shrouding by dust has gradually declined.Shrouding by dust may explain the significant drop in optical luminositiesexhibited by some of the sources at the shortest wavelengths. Strazzullo, V., et al., 2010, ApJ 714, 1305Takeuchi, T. T. &Ishii, T. T., 2004, ApJ 604, 40Trouille, L., et al., 2008, ApJS 179, 1Zinnecker, H., & Yorke, H., 2007 Ann. Rev. A&A, 45, 481 c (cid:13) , 2–, 2–
Lockman N galaxy SEDs plotted with arbitrary flux density o ff -sets. These galaxies all met or exceeded our selection criteria of being se-curely identified at all three wavelengths, having a high purity index, (30%,50%, and 70% pure at 500, 350, and 250 µ m respectively), having knownPACS detections at 100 and 170 µ m, and having known redshifts, z > . µ m measurement is compatible with zero, so we have plotted onlythe upper error bar. The solid line is an S&K model fit to the 24 µ m through500 µ m error-weighted observations. We weighted the 24 µ m observationsonly a quarter as heavily as the longer wavelengths which play a dominantrole in determining starburst luminosities. Observational data shortward of24 µ m is plotted for reference but not used in fitting. Source Π s are shown inTable 2. As with many of the examples quoted by S&K, the visual compo-nent of the SED often needs to be fitted by hand because it bears little rela-tion to the starburst characteristics responsible for the mid- and far-infraredflux densities. The extent to which visible stars may or may not contributeto the SED is determined in part by the degree to which an older populationof stars is obscured by dust without significantly contributing to its heatingor by massive young stars whose shrouding by dust has gradually declined.Shrouding by dust may explain the significant drop in optical luminositiesexhibited by some of the sources at the shortest wavelengths. Strazzullo, V., et al., 2010, ApJ 714, 1305Takeuchi, T. T. &Ishii, T. T., 2004, ApJ 604, 40Trouille, L., et al., 2008, ApJS 179, 1Zinnecker, H., & Yorke, H., 2007 Ann. Rev. A&A, 45, 481 c (cid:13) , 2–, 2– ?? D. Brisbin et al.
Figure 3.
GOODS-N galaxy SEDs plotted as in Figure 2. The 500 µ mmeasurements for GOODS sources A, C, E, F, H, K, and L are compatiblewith zero, so here we plot only the upper error bar, barely visible within thediamond symbol. As in Figure 2, we weighted the 24 µ m data a quarter asheavily as the longer wavelengths. Observational data shortward of 24 µ mis plotted for reference but not used in fitting. Figure 4.
Infrared source luminosities (integrated over 8-1000 µ m) inGOODS-N and LN plotted as a function of redshift for all sources detectedat all three SPIRE wavelengths. Diamonds indicate GOODS-N luminositiesobtained from SEDs fitted by S&K models, and triangles indicate similarluminosities for LN sources. The solid line shows the growth of luminositydistance squared with z. It serves as a rough lower bound to the luminosi-ties in our selection of observed sources; the scatter of data points about thecurve provides a visual impression of the uncertainties in those luminosities. Figure 5.
Spectral energy distributions of the most luminous sources inGOODS North (a) to (d) Lockman North (e) to (k). As in Figures 2 and 3,we solely plot the upper error bars at 500 µ m for sources (a) to (e), whoselower error bars are compatible with zero. We similarly plot an upper errorbar at 170 µ m for LN (k). Luminosities and positions for these sources arepresented in Table 4. Seven of these eleven sources are triply secure; two ofthese are also included in Figures 2 and 3. A brief description of sources (c),(d), (f) and (i) is provided in Section 6. Although some of the ultraluminoussources exhibit significant AGN activity, we have applied S&K model fits,as discussed in Section 7. As in Figures 2 and 3 we have weighted the 24 µ mdata only one quarter as heavily as the longer wavelengths. Observationaldata shortward of 24 µ m is plotted for reference but not used in fitting.c (cid:13) , 2– ?? he Deep SPIRE HerMES Survey: Spectral Energy Distributions and their Astrophysical Indications at High Redshift Table 2.
Data on LN Galaxies whose SEDs appear in Figure 2. We list the ID as established by Roseboom et al. (2010), right ascension and declination(J2000), photometric redshift z > .
5, our purity indices Π λ , total bolometric luminosity as estimated by an S&K model, star formation rate based on theinfrared luminosity relation (Kennicutt 1998), and dust mass estimated by the S&K model. Along with these derived parameters, we list the model parameters:nuclear radius, visual extinction to center (A v ), and gas density within hotspots (n hs ). ID RA Dec z L bolometric ( L (cid:12) ) SFR ( M (cid:12) / yr) M dust ( M (cid:12) ) Π Π Π
500 Radius (kpc) A v Log10(n hs × cm3)LN (A) 161.8365 59.1211 0.56 4.0 × ×
108 0.89 0.67 0.74 3.0 35.9 3LN (B) 161.5001 58.8732 0.60 7.9 × ×
108 0.74 0.62 0.88 3.0 72.0 4LN (C) 161.1277 59.1956 0.72 1.3 × ×
108 1.10 1.00 1.07 3.0 72.0 2LN (D) 161.3232 59.2086 0.82 1.3 × ×
108 0.88 0.80 0.98 3.0 72.0 4LN (E) 161.0530 59.0762 0.94 2.0 × ×
108 0.92 0.65 0.69 3.0 35.9 3LN (F) 161.3680 59.2242 0.96 2.5 × ×
108 0.96 0.97 0.56 9.0 9.0 4LN (G) 161.3429 59.2269 1.02 4.0 × ×
108 0.89 0.87 0.64 3.0 35.9 2LN (H) 161.4871 58.8886 1.06 2.5 × ×
108 0.76 0.82 0.93 3.0 35.9 2LN (I) 161.8669 58.8708 2.76 2.5 × ×
109 0.91 0.96 0.77 9.0 120.0 4
Table 3.
Data on GOODS-N galaxies whose SEDs appear in Figure 3, listed as in Table 2. The redshifts here are spectroscopic.
ID RA Dec z L bolometric ( L (cid:12) ) SFR ( M (cid:12) / yr) M dust ( M (cid:12) ) Π Π Π
500 Radius (kpc) A v Log10(n hs × cm3)GOODS (A) 189.0274 62.1643 0.6380 5.0 × ×
108 0.83 0.84 0.44 3.0 72.0 4GOODS (B) 189.3938 62.2898 0.6402 2.0 × ×
107 0.98 0.64 0.90 1.0 119.0 2GOODS (C) 189.2979 62.1820 0.8549 1.3 × ×
108 1.00 0.81 0.74 3.0 72.0 4GOODS (D) 189.1403 62.1683 1.0160 1.6 × ×
108 0.99 0.75 0.54 3.0 35.9 3GOODS (E) 189.0633 62.1691 1.0270 1.3 × ×
108 0.92 0.52 0.45 3.0 35.9 2GOODS (F) 189.3171 62.3541 1.1440 1.0 × ×
107 0.73 0.68 0.68 3.0 17.9 2GOODS (G) 189.1438 62.2114 1.2242 2.5 × ×
107 0.88 0.89 0.98 1.0 119.0 4GOODS (H) 189.2137 62.1810 1.2258 6.3 × ×
108 0.71 0.83 0.31 3.0 35.9 4GOODS (I) 189.2614 62.2338 1.2480 1.3 × ×
108 0.98 0.93 0.97 9.0 9.0 4GOODS (J) 189.2566 62.1962 1.7600 6.3 × ×
108 1.02 1.04 1.03 9.0 18.0 4GOODS (K) 189.3036 62.1955 1.8150 2.0 × ×
107 1.08 0.86 0.50 3.0 17.9 3.4GOODS (L) 189.0764 62.2640 2.0000 2.0 × ×
107 0.82 0.71 0.33 1.0 119.0 2
Table 4.
Data on luminous galaxies whose SEDs appear in Figure 5, listed as in previous tables. The GOODS sources have spectroscopic redshifts, as do LN(f) and (j). The rest of the LN sources have photometric redshifts. The sources GOODS (a) and LN (i) correspond to the sources GOODS (L) and LN (I) shownin the previous figures and tables.
ID RA Dec z L bolometric ( L (cid:12) ) SFR ( M (cid:12) / yr) M dust ( M (cid:12) ) Π Π Π
500 Radius (kpc) A v Log10(n hs × cm3)GOODS (a) 189.0764 62.2640 2.0000 2.0 × ×
107 0.82 0.71 0.33 1.0 119.0 2.0GOODS (b) 189.1483 62.2400 2.0050 2.5 × ×
107 0.69 0.24 0.38 1.0 71.0 2.0GOODS (c) 188.9901 62.1734 3.0750 6.3 × ×
109 1.61 1.22 0.84 15.0 18.0 4.0GOODS (d) 189.3096 62.2024 3.1569 1.3 × ×
107 0.57 0.55 0.84 3.0 9.0 2.0LN (e) 161.7059 59.3247 1.28 1.3 × ×
108 0.99 0.85 0.20 3.0 72.0 2.0LN (f) 161.0415 58.8735 2.28 2.5 × ×
105 0.98 1.10 1.71 0.3 6.7 2.0LN (g) 161.5408 58.7950 2.58 1.3 × ×
109 0.82 0.95 0.58 15.0 18.0 4.0LN (h) 160.9635 58.9555 2.68 2.5 × ×
106 0.18 0.95 0.96 0.3 144.0 2.0LN (i) 161.8667 58.8704 2.76 2.5 × ×
109 0.91 0.96 0.77 9.0 120.0 4.0LN (j) 161.5541 58.7886 2.96 2.0 × ×
107 0.76 0.61 0.60 3.0 9.0 3.0LN (k) 161.8259 59.2771 3.12 1.3 × ×
107 0.98 0.70 0.54 3.0 17.8 2.0 c (cid:13) , 2–, 2–