HELP: modelling the spectral energy distributions of Herschel detected galaxies in the ELAIS N1 field
K. Malek, V. Buat, Y. Roehlly, D. Burgarella, P. D. Hurley, R. Shirley, K. Duncan, A. Efstathiou, A. Papadopoulos, M. Vaccari, D. Farrah, L. Marchetti, S. Oliver
AAstronomy & Astrophysics manuscript no. KMalek_final c (cid:13)
ESO 2018September 5, 2018
HELP: modelling the spectral energy distributions of
Herschel detected galaxies in the ELAIS N1 field
K. Małek , (cid:63) , V. Buat , Y. Roehlly , , D. Burgarella , P. D. Hurley , R. Shirley , K. Duncan , A. Efstathiou ,A. Papadopoulos , M. Vaccari , , D. Farrah , L. Marchetti , , and S. Oliver Aix Marseille Univ. CNRS, CNES, LAM Marseille, France, National Centre for Nuclear Research, ul. Ho˙za 69, 00-681 Warszawa, Poland, Astronomy Centre, Department of Physics and Astronomy, University of Sussex, Falmer, Brighton BN1 9QH, UK Univ Lyon, Univ Lyon1, ENS de Lyon, CNRS, Centre de Recherche Astrophysique de Lyon UMR5574, F-69230, Saint-Genis-Laval, France Leiden Observatory, Leiden University, NL-2300 RA Leiden, Netherlands School of Sciences, European University Cyprus, Diogenes Street, Engomi, 1516, Nicosia, Cyprus Department of Physics and Astronomy, University of the Western Cape, Robert Sobukwe Road, Private Bag X17, 7535 Bellville,Cape Town, SA INAF - Istituto di Radioastronomia, via Gobetti 101, 40129 Bologna, Italy University of Hawaii, 2505 Correa Road, Honolulu, HI 96822, USA Department of Astronomy, University of Cape Town, Private Bag X3, Rondebosch 7701, South AfricaSeptember 5, 2018
ABSTRACT
Aims.
The
Herschel
Extragalactic Legacy Project (HELP) focuses on the data from ESA’s
Herschel mission, which covered over1300 deg and is preparing to publish a multi-wavelength catalogue of millions of objects. Our main goal is to find the best approach tosimultaneously fitting spectral energy distributions (SEDs) of millions of galaxies across a wide redshift range to obtain homogeneousestimates of the main physical parameters of detected infrared (IR) galaxies. Methods.
We perform SED fitting on the ultraviolet(UV) / near-infrared(NIR) to far-infrared(FIR) emission of 42 047 galaxies fromthe pilot HELP field: ELAIS N1. To do this we use the latest release of CIGALE, a galaxy SED fitting code relying on energy balance,to deliver the main physical parameters such as stellar mass, star formation rate, and dust luminosity. We implement additional qualitycriteria to the fits by calculating χ values for the stellar and dust part of the spectra independently. These criteria allow us to identifythe best fits and to identify peculiar galaxies. We perform the SED fitting of ELAIS N1 galaxies by assuming three di ff erent dustattenuation laws separately allowing us to test the impact of the assumed law on estimated physical parameters. Results.
We implemented two additional quality value checks for the SED fitting method based on stellar mass estimation and energybudget. This method allows us to identify possible objects with incorrect matching in the catalogue and peculiar galaxies; we found351 possible candidates of lensed galaxies using two complementary χ s criteria (stellar and infrared χ s) and photometric redshiftscalculated for the IR part of the spectrum only. We find that the attenuation law has an important impact on the stellar mass estimate(on average leading to disparities of a facto33131corrr of two). We derive the relation between stellar mass estimates obtained by threedi ff erent attenuation laws and we find the best recipe for our sample. We also make independent estimates of the total dust luminosityparameter from stellar emission by fitting the galaxies with and without IR data separately. Key words.
Infrared: galaxies – Galaxies: statistics – fundamental parameters
1. Introduction
Multi-wavelength data for extragalactic objects is a necessaryprecondition for a physical analysis of galaxies, as the full com-plexity of the galaxy is only seen when using di ff erent spec-tral ranges simultaneously. The emission from the hot gas com-ponent, active nuclear regions, and the end products of stellarevolution (supernovae and compact remnants) can be observedin the X-rays ( 1 < λ < λ > ∼ µ m to 1 mm spectralrange; IR) spectra of all galaxies arises from stellar light; eitherdirectly or reprocessed by the surrounding interstellar medium(ISM). Old stars can be seen in the near-infrared (NIR) spec-tral range (0.75 < λ < µ m). The dust, composed of a mixture (cid:63) e-mail: [email protected] of carbonaceous and amorphous silicate grains (Draine 2003) isheated by the interstellar radiation field and emits in the mid-( ∼ < λ < µ m) and far-infrared (MIR and FIR, respectively)(10 µ m < λ < ff erent galaxy components see the book written byBoselli (2011).The UV-to-IR spectral energy distribution (SED) contains in-formation about the stars of the galaxy such as the stellar mass orstar formation rate (SFR). For example, information on newbornstars can be inferred from the UV data, making the UV range a Article number, page 1 of 17 a r X i v : . [ a s t r o - ph . GA ] S e p & A proofs: manuscript no. KMalek_final very e ffi cient tracer of SFR. Unfortunately, one of the obstaclesto observing starburst regions in the UV range only is dust. Thenewly born stars are surrounded by gas and dust, which, as wellas obscuring the most interesting regions, absorb a part of theUV light emitted by stars (e.g. Buat et al. 2007). Dust grains ab-sorb or scatter photons emitted by stars and re-emit the energyover the full IR range. Therefwore, only part of the energy fromnewly born stars can be observed in UV wavelengths. Infraredemission, reflecting the dust-obscured star formation activity ofgalaxies (Genzel & Cesarsky 2000), combined with UV and op-tical data, can provide a broad range of information about thestar formation history (SFH) and SFR.As shown in previous studies (e.g. Le Floc’h et al. 2005;Takeuchi et al. 2005) the fraction of hidden SFR estimated byUV emission increases from ∼
50% in the Local Universe to ∼
80% at z =
1. Burgarella et al. (2013) combined the measure-ments of the UV and IR luminosity functions up to redshift 3.6to calculate the redshift evolution of the total SFR UV + IR and dustattenuation. They found that the dust attenuation increases fromz = = ff ective approach to obtaining constraintson, for example, stellar masses, SFRs, and dust properties fromthe large multi-wavelength catalogues of galaxies is to fit phys-ical models to the galaxy’s broad-band SED (i.e. Walcher et al.2011; Leja et al. 2017). For an overview of recent improve-ments in models and methods, Walcher et al. 2011 present a de-tailed review of di ff erent techniques for SED fitting. Many newtools were developed for both UV-optical data or the dust partonly, such as STARLIGHT (Cid Fernandes et al. 2005), ULYSS(Koleva et al. 2009), VESPA (Tojeiro et al. 2007), Hyperz (Bol-zonella et al. 2000), Le Phare (Arnouts et al. 1999; Ilbert et al.2006), PAHFIT (Smith et al. 2007) or to fit specific types ofobjects (e.g. AGNfitter, Calistro Rivera et al. 2016). Also, theVirtual Observatory tools allow for SED fitting. Bayesian analy-sis is included in many di ff erent software packages; for exampleGOSSIP (Franzetti et al. 2008), CIGALE (Noll et al. 2009), andBayeSED (Han & Han 2014). The wealth of public and privateSED fitting tools implies that di ff erent surveys tend to be anal-ysed with di ff erent tools, with no common set of models or pa-rameters. As a consequence, it is di ffi cult to combine, compareor interpret large datasets for statistical analysis, as di ff erent ap-proaches, models, and assumptions result in disparate accuracy,scaling factors, and non-uniform physical parameters across awide redshift range.A lack of homogeneous multi-wavelength catalogues (cover-ing over 10% of the entire sky), together with non-uniform phys-ical parameters obtained based on di ff erent models and software,makes the analysis of the main physical properties of galaxiesstatistically limited and biased.An FP7 project called the Herschel
Extragalactic LegacyProject (HELP, Vaccari 2016, Oliver et al., in preparation)funded by European Union will remove the barriers to multi- λ statistical survey science. The main aim of HELP is to providehomogeneously calibrated multi-wavelength catalogues cover-ing roughly 1300 deg of the Herschel
Space Observatory (Pil- bratt et al. 2010) survey fields. These catalogues are going tomatch individual galaxies across broad wavelengths, allowingfor multi-wavelength SED fitting to be performed and for statis-tical studies of the local-to-intermediate redshift galaxy popula-tion. The selection criteria, depth maps, and master list creationdetails will be published together with the catalogues (Shirleyet al. in prep.). The presence of IR data together with UV–NIRcounterparts makes the HELP multi-wavelength catalogue a per-fect data set for studying galaxy formation and evolution overcosmic time.In this paper, we present the general HELP strategy for SEDfitting for the millions of galaxies observed in multi-wavelengthpass-bands. We discuss the software, models, and parameterswhich are going to be used uniformly for each data set. Thisapproach ensures homogeneity of obtained physical parametersfor the final HELP deliverable of 1300 deg field.The paper is organised as follows. In Sect. 2 we describe theELAIS N1 field as a pilot field of the HELP project. In Sect. 3we present the method applied in this work to fit SEDs and themain physical models and parameters used for the fitting. Sec-tion 4 presents all reliability tests as well as two implementedquality checks. In Sect. 5 we discuss the general properties of thesample, while in Sects. 6 and 7 we explore how all physical pa-rameters obtained by fitting SEDs with di ff erent dust attenuationmodels can be biased and whether or not it is possible to pre-dict total dust luminosity from stellar emission only. Our sum-mary and conclusions are then presented in Sect. 8. In this paperwe use WMAP7 cosmology (Komatsu et al. 2011): Ω m = Ω Λ = = − Mpc − .
2. Data
In our analysis, we focus on the pilot HELP field: EuropeanLarge Area ISO Survey North 1 (hereafter ELAIS N1), 9 deg area centred at 16 h m s + o (cid:48) (cid:48)(cid:48) (Oliver et al. 2000).ELAIS N1 is one of 20 fields making up the European LargeArea ISO Survey (ELAIS, Oliver et al. 2000; Rowan-Robinsonet al. 2004). The Herschel data in ELAIS N1 was obtained aspart of the HerMES project Oliver et al. (2012). It is representa-tive of moderately deep fields for future HELP catalogues.In this paper we briefly summarise the data used forELAIS N1. A detailed description of the data used for the HELPproject (both FIR and ancillary data), the open source pipeline,which was developed for HELP, and the cross-matching proce-dure, astrometry corrections, and full data diagnostics will bepresented in Shirley et al., (in preparation).
Herschel was equipped with two continuum imaging instru-ments, the Photodetector Array Camera and Spectrometer(PACS; Poglitsch et al. 2010) and the Spectral and Photomet-ric Imaging Receiver (SPIRE, Gri ffi n et al. 2010). These instru-ments provided FIR coverage at 100 and 160 µ m from PACS,and at 250, 350, and 500 µ m from SPIRE. To obtain the pho-tometry of Herschel sources a new prior-based source extractiontool was developed, called XID + (Hurley et al. 2017).The XID + is a probabilistic de-blending tool used to ex-tract Herschel
SPIRE source flux densities from
Herschel mapsthat su ff er from source confusion (Nguyen et al. 2010). This isachieved by using a Bayesian inference tool to explore the pos-terior probability distribution. This algorithm is e ffi cient in ob- The software is available at https: // github.com / H-E-L-P / XID_plusArticle number, page 2 of 17. Małek et al.: HELP: modelling the spectral energy distributions of
Herschel detected galaxies in the ELAIS N1 field λ [ µ m] . . . . . . tr a n s m i ss i o n CFHT MegaCam uSubaru HSC gSubaru HSC rSubaru HSC iSubaru HSC zSubaru HSC n921Subaru HSC yUKIRT WFCam JUKIRT WFCam KSpitzer IRAC 3.6 Spitzer IRAC 4.5Spitzer IRAC 5.8Spitzer IRAC 8.0Spitzer MIPS 24Herchel PACS 100Herchel PACS 160Herchel SPIRE 250Herchel SPIRE 350Herchel SPIRE 500
Fig. 1: Transmission curves and demonstrative coverage of primary photometric bands used for SED fitting. Where multiple filterscover the same region, we show only one for clarity. All transmission curves are plotted as exact transmissions.taining Bayesian probability distribution functions (PDFs) forall prior sources, and thus flux uncertainties can be estimated.A detailed description can be found in Hurley et al. (2017). Theoriginal XID + , used for ELAIS N1 field, uses information IRACbands as a prior. We note that XID + can be run with more so-phisticated priors, using both dust luminosity and redshift infor-mation (e.g. Pearson et al. 2018, developed a method of incorpo-rating flux predictions from SED fitting procedure as informedpriors, finding improvements in the detection of faint sources).The XID + was run on the ELAIS N1 Spitzer MIPS, and Her-schel
PACS and SPIRE map. The flux level, at which the aver-age posterior probability distribution of the source flux becomesGaussian ( ? , indicating that the information from data dominatesover the prior, see)or more details]Hurley2017, is 20 mJy forMIPS, 12.5 and 17.5 mJy for PACS and for SPIRE is 4 µ Jy forall three bands.In the deblending work, the priors we use for computingXID + fluxes must satisfy two criteria: they must have a detectionin the Spitzer IRAC 1 band and they must have been detected ineither optical of NIR (this was done to eliminate artefacts). Theentire catalogue of the HELP FIR measurements based on theXID + tool will be published at the end of the program (2018),together with the full multi-wavelength data collected from othersurveys. The HELP master catalogue is built on a positional cross matchof all the public survey data available in the optical to MIRrange. This comprises observations from the Isaac Newton Tele-scope / Wide Field Camera (INT / WFC) survey (González-Solareset al. 2011), the Subaru Telescope / Hyper Suprime-Cam StrategicProgram Catalogues (HSC-SSP) (Aihara et al. 2018), the SpitzerAdaptation of the Red-sequence Cluster Survey (SpARCS) (Tu-dorica et al. 2017), the UK Infrared Telescope Deep Sky Survey- Deep Extragalactic Survey (UKIDSS-DXS) (Swinbank 2013;Lawrence et al. 2007), the Spitzer Extragalactic RepresentativeVolume Survey (SERVS, Mauduit et al. 2012), and the SpitzerWide InfraRed Extragalatic survey (SWIRE, Lonsdale et al.2003; Surace et al. 2005). We use the Spitzer Data Fusion prod-ucts for the final two Spitzer ‘MIR surveys presented in Vaccariet al. (2010) and Vaccari (2015). The cross match is describedin full in Shirley et al. (in prep). The list of filters is shown inTable 1 and the coverage is presented in Fig. 1. Table 1: Data used for SED fitting from ELAIS N1 field.Telescope Instrument FilterCFHT MegaCam u, g, r, y, zSubaru HSC g, r, i, z, N921, yIsaac Newton Wide Field Cam. u, g, r, i, zPanSTARRS1 Gigapixel Cam.1 g, r, i, z, yUKIRT WFCam J, KSpitzer IRAC 3.6, 4.5, 5.8, 8.0 ( µ m)MIPS mips 24 ( µ m) Herschel
PACS 100, 160 ( µ m)SPIRE 250, 350, 500 ( µ m)When multiple measurements are available in similar pass-bands we take the deepest only since rapidly increasing errors onshallower surveys mean those measurements contribute little tooverall photometry. In cases where a detection is available fromtwo similar filters (e.g. MegaCam g and HyperSuprimeCam g)we check the number of measurements in the full catalogue, thedepth associated with each filter, and the distribution of uncer-tainties. Because there is a large di ff erence in the depths of dif-ferent surveys, there is no advantage to using multiple measure-ments where one clearly has an order-of-magnitude lower errorand errors in the shallower catalogues may not include system-atic errors of comparable size to the random errors. Based on ouranalysis we decide to use filters in the same order for each band:HyperSuprimeCam – MegaCam – WFC – GPC1. This meansthat if we have, for example, g band measurements from WFCand MegaCam, for our analysis, we are going to use only mea-surements from MegaCam. If we also have HyperSuprimeCamphotometry then only that one would be used for SED fitting.We also remove objects around bright stars by measuring thesize of the circular region around a star that contains no detec-tions as a function of the star magnitude. This compound selec-tion function can be used to model which objects will propagatethrough to the final SED sample. As part of the HELP database, we provide new photometricredshifts generated using a Bayesian combination approach, de-
Article number, page 3 of 17 & A proofs: manuscript no. KMalek_final redshift
Fig. 2: Redshift distributions of 50 129 galaxies for which iswas possible to fit SEDs (blue histogram) and 42 453 galaxiesremaining after SED quality cleaning according to Sect. 4.3 (redhistogram).scribed in Duncan et al. (2018). They used two independent mul-tiwavelength datasets (NOAO Deep Wide Field Survey Bootesand COSMOS) and performed zphot estimation. Duncan et al.(2018) investigated the performance of three zphot template sets:(1) default EZY reduced galaxy set (Brammer et al. 2008), (2)"XMM COSMOS" templates (Salvato et al. 2009), and (3) atlasof Galaxy SEDs (Brown et al. 2014) as a function of redshift, ra-dio luminosity, and IR / X-ray properties. They found that only acombination of all template libraries is able to provide a consen-sus zphot estimate and they used a hierarchical model Bayesiancombination of the zphot estimates.This method is used for the HELP project and for the nextgeneration of deep radio continuum surveys. The detailed de-scription of the zphot methodology and redshift accuracy is pre-sented in Duncan et al. (2018).
The catalogues produced by XID + contain a flag for thosesources that are either below a flux level in which there is a cleardetection or the related XID + Bayesian P-value maps indicate apoor fit in a region local to the source. The final sample includes50 129 galaxies from the ELAIS N1 survey with flux measure-ments for PACS or SPIRE data (or both). The full sample anddata files can be downloaded at http://hedam.lam.fr/HELP/dataproducts/dmu28/dmu28_ELAIS-N1/data/zphot/ andalso accessed with Virtual Observatory standard protocols at https://herschel-vos.phys.sussex.ac.uk/ .Our procedure gives us a set of 19 bands: u, g, r, i, z, N921, y,J, K, Spitzer IRAC 3.6, 4.5, 5.8 and 8.0 µ m, Spitzer MIPS 24 µ m,and five passbands from Herschel
PACS (100 and 160 µ m) andSPIRE (250, 350 and 500 µ m). The coverage of the wavelengthsis shown in Fig. 1.The mean value of the redshift distribution of the final sam-ple is 0.97, while 50% of galaxies (between quartile 1 and 3) arelocated in the redshift range 0.53 – 1.63. The redshift distribu-tion of the initial sample can be found in Fig. 2 (blue hatchedhistogram).
3. Fit of the spectral energy distribution
Taking advantage of the very dense coverage from the broad-band passbands (from u band to 500 µ m), we use Code Investi-gating GALaxy Emission (CIGALE, Boquien et al., in prepa-ration). CIGALE is designed to estimate the physical parame-ters (i.e. SFR, stellar mass, dust luminosity, dust attenuation,AGN fraction) by comparing modelled galaxy SEDs to observedones. CIGALE conserves the energy balance between the dust-absorbed stellar emission and its re-emission in the IR. Manyauthors (e.g. Buat et al. 2015; Ciesla et al. 2015, 2016; Małeket al. 2017) have presented the methodology and the strategy ofthe code already. A more detailed description of the code will begiven by Boquien et al (in prep). All adopted parameters usedfor all modules are presented in Table. 2. To build the stellar component, we use the stellar population syn-thesis models by Bruzual & Charlot (2003) with the initial massfunction (IMF) given by Chabrier (2003).As the aim of this analysis is to show how to fit the SEDs ofa large sample of galaxies, simplicity and limiting the number ofparameters is crucial. It was shown in Ciesla et al. (2015), than aSFH composed of a delayed form to model the bulk of the stel-lar population with the addition of a flexibility in the recent SFHprovides very good estimates of the SFR–M star relation compar-ing to observations. We refer the reader to Ciesla et al. (2015)for a detailed description of the model. We apply SFH scenarioswhich include delayed SFR with an additional burst. Parameterswhich describe our scenario are: age of the galaxy, decreasingrate, burst fraction, and burst age. The functional form for theSFR is calculated as:SFR(t) = SFR delayd (t) + SFR burst (t) , (1)where SFR delayd (t) ∝ te − t /τ main , and SFR burst (t) ∝ te − (t − t ) /τ burst if t > t , and SFR burst (t) = t < t . The factor τ main representse-folding time of the main stellar population and τ burst representse-folding time of the late starburst population. The adopted pa-rameters used for the stellar component in our fitting procedureare presented in Table 2. We fix values of e-folding times of themain and late stellar population models as those parameters arevery di ffi cult to constrain using the SED fitting procedure (i.e.Noll et al. 2009). The current public version of CIGALE (cigale version 0.12.1)includes four di ff erent modules to calculate dust properties:Casey (2012), Dale et al. (2007), Dale et al. (2014), and an up-dated version of the Draine & Li (2007) model, which we decideto use (this model was also used by e.g. Lo Faro et al. 2017 andPearson et al. 2018 in the framework of the HELP project). In ouranalysis we apply the multi-parameter dust emission model pro-posed by Draine et al. (2014), which is described as a mixture ofcarbonaceous and amorphous silicate grains. Infrared emissionSEDs have been calculated for dust grains heated by starlight forvarious distributions of starlight intensities. The majority of thedust is heated by a radiation field with constant intensity (markedas U min in Table 2) from the di ff use ISM. A much smaller frac-tion of dust ( γ , from Table 2) is illuminated by the starlight with http:\cigale.lam.fr Article number, page 4 of 17. Małek et al.: HELP: modelling the spectral energy distributions of
Herschel detected galaxies in the ELAIS N1 field intensity range from U min to U max . This intensity is characterisedby a power-law distribution.We note that we are not able to obtain a reliable value of the γ parameter due to the degeneracy between γ and radiation in-tensity, and the large photometric uncertainty. We fix the valueof γ using the mean value obtained from a stacking analysis byMagdis et al. (2012) for the average SEDs of main sequencegalaxies at redshifts 1 and 2 ( γ = As shown in the literature (e.g. Ciesla et al. 2015; Leja et al.2018) the SED fitting procedure is a powerful technique, but theaccuracy of estimated physical properties is tightly correlatedwith the accuracy of the models used. For example, the AGNscan substantially contribute to the MIR emission of a galaxy.To improve the derived galaxy properties we add an AGNcomponent to the stellar SED. We derive the fractional contri-bution of the AGN emission from Fritz et al. (2006) templates,which assume two components: (1) point-like isotropic emis-sion of the central source, and (2) radiation from dust with atoroidal geometry in the vicinity of the central engine. The AGNemission is absorbed by the toroidal obscurer and re-emitted at1–1000 µ m wavelengths or scattered by the same obscurer. Weperform SED fitting with a set of parameters from the Fritz et al.(2006) models as described in Table 2. To limit the number ofmodels we fix the value to five variables that parametrizes thedensity distribution of the dust within the torus (ratio of the max-imum to minimum radius of the dust torus, radial and angulardust distribution in the torus, angular opening angle of the torus,and angle between equatorial axis and the line of sight) with typ-ical values found by Fritz et al. (2006), and used, for example,by Hatziminaoglou et al. (2009), Buat et al. (2015), and Cieslaet al. (2015). We perform three SED-fitting runs with three di ff erent dust-attenuation laws to check which law gives the best fits: we usethe Charlot & Fall (2000) model as the main dust attenuationrecipe, and then we redo the whole analysis with the popular at-tenuation law of Calzetti et al. (2000). Subsequently we also testthe attenuation law for z ∼ yr, young stars disrupt their BCs andmigrate into the ambient ISM. The functional form for Charlot &Fall (2000) attenuation law (CIGALE’s dustatt_2powerlaws module) refers to:A( λ ) = (cid:40) A λ (BC) + A λ (ISM), for young stars, age < yr , A λ (ISM), for old stars (age > yr) . (2)where A λ = A V ( λ/λ V ) δ , λ V is 550 µ m, A V is a V-band atten-uation, and for CF00 δ = -0.7. The fraction of the total e ff ec-tive optical depth contributed by the di ff use ISM is defined as µ = f att / (1 + f att ), where f att = A ISMV / A BCV .As part of the HELP project, Lo Faro et al. 2017 com-pared power-law attenuation curves with the results of radiative–transfer calculations for 20 ULIRGs observed by
Herschel . LoFaro et al. (2017) provided an attenuation law similar to but flat-ter than that of CF00, which consists of two power laws but with slopes of the attenuation in the BC and ISM equal to -0.48. Thee ff ect of the change of the slopes from -0.7 to -0.48 results inever more greyer attenuation curves for the UV part of the spec-trum. The functional form of Lo Faro et al. (2017) attenuationlaw is exactly the same as Eq. 2 with δ = -0.48.The functional form for Calzetti et al. (2000) attenuation lawis described as:A( λ ) = E(B − V)k( λ ) , (3)where k( λ ) is the e ff ective attenuation curve defined as:k( λ ) = . − . + . /λ ) , for (0 . µ m (cid:54) λ (cid:54) . µ m )2 . − . + . /λ − . /λ + . /λ ) , for (0 . µ m (cid:54) λ < . µ m ) , (4)and E(B − V) is the colour excess for the stellar continuum.The di ff erence in shape for attenuation curves defined byCalzetti et al. (2000) and CF00 is mostly present for the wave-lengths longer than 5000 Å, when CF00 curve is much flatterthan the one given by Calzetti et al. (2000). A detailed descrip-tion of the di ff erence between the two attenuation laws can befound in Lo Faro et al. (2017).To make the comparison possible, we used only one reduc-tion factor between A ISMV / A BCV (f att ) both for CF00 and Lo Faroet al. (2017). As was shown by Lo Faro et al. (2017) and refer-ences therein, the f att parameter is known to usually be uncon-strained by broad-band SED fitting. We performed a preliminaryanalysis using f att as a free parameter (values correspond to µ parameter in the interval 0.2–0.5), and for a final run we chosethe most probable value obtained based on the full ELAIS N1sample: 0.44 (f att = . Each created SED template consists of five components (SFH,single stellar population, dust attenuation and emission, and theAGN component) which are spread of all possible grid of theinput parameters.All these parameters (with the exception of parameters re-lated to the attenuation curve recipes published by Calzetti et al.2000 and Lo Faro et al. 2017, which were used for comparingthe main physical parameters presented in Sect. 6) are defaultparameters used for all HELP fields. The full list of the inputparameters of the code is presented in Table 2.Based on the number of parameters presented in Table 2 andthe redshift range for our sample of galaxies we built ∼
850 mil-lion SED templates which were fitted with CIGALE to 50 129ELAIS N1 galaxies detected by
Herschel . We used a 20-corepersonal computer (PC) with 252Gb of memory. The total timeto fit all SEDs for ELAIS N1 was ∼ Article number, page 5 of 17 & A proofs: manuscript no. KMalek_final
Table 2: List of the input parameters of the code CIGALE. All free parameters are marked with a star before the name.parameter valuesdelayed star formation history + additional burste-folding time of the main stellar population model [Myr] 3000e-folding time of the late starburst population model [Myr] 10 000 (cid:63) mass fraction of the late burst population 0.001, 0.010, 0.030, 0.100, 0.300 (cid:63) age [Myr] 1000, 2000, 3000, 4000, 5000, 6500, 10000 (cid:63) age of the late burst [Myr] 10.0, 40.0, 70.0single stellar population: Bruzual & Charlot (2003)initial mass function Chabrier (2003)metallicities (solar metallicity) 0.20age of the separation between the young and the old star population [Myr] 10attenuation curvemain recipe: Charlot & Fall (2000) (cid:63) A V in the BCs 0.3, 0.8,1.2,1.7,2.3,2.8,3.3, 3.8power law slopes of the attenuation in the birth clouds and ISM -0.7Lo Faro et al. (2017) (cid:63) V-band attenuation in the birth clouds (Av BC) 0.3, 0.8,1.2,1.7,2.3,2.8,3.3, 3.8power law slopes of the attenuation in the birth clouds and ISM -0.48Calzetti et al. (2000) (cid:63) the colour excess of the stellar continuum light for the young population 0.6, 0.67, 0.74, 0.81, 0.88, 0.95, 1.02, 1.09, 1.16, 1.23,1.29, 1.36, 1.43, 1.50, 1.57, 1.64, 1.71, 1.78, 1.85, 1.92,1.99, 2.06, 2.13, 2.2dust emission: Draine & Li (2007)for our analysis we built templates based on adopted parameters from previous studies:Magdis et al. (2012); Ciesla et al. (2015); Lo Faro et al. (2017); Pearson et al. (2018) (cid:63) mass fraction of PAH 1.12, 2.5, 3.19 (cid:63) minimum radiation field (U min ) 5.0, 10.0, 25.0 (cid:63) power law slope dU / dM ( U α ) 2.0, 2.8fraction illuminated from Umin to Umax ( γ ) 0.02AGN emission: Fritz et al. (2006)For our analysis we used templates built based on average parameters from previous studies:Fritz et al. (2006); Hatziminaoglou et al. (2009); Buat et al. (2015); Ciesla et al. (2015); Małek et al. (2017)ratio of the maximum to minimum radius of the dust torus 60 (cid:63) optical depth at 9.7 microns 1.0, 6.0radial dust distribution in the torus -0.5angular dust distribution in the torus 0.0angular opening angle of the torus [deg] 100.0angle between equatorial axis and the line of sight [deg] 0.001 (cid:63) fractional contribution of AGN 0.0, 0.15, 0.25, 0.8
4. Reliability check for SED fitting procedure andidentification of outliers
After the SED fitting we removed 143 SED fitting failures(galaxies with χ r equal to 99 ) from the initial sample of 50 129ELAIS N1. From now on we use the remaining 49 986 galaxiesas an ELAIS N1 sample. The most common reason for SED fitting failures is overestimatedredshifts which translates into an estimated age of the galaxy larger thanthe age of the universe with the assumed ages provided in the CIGALESED fitting.
A mock catalogue was generated to check the reliability of thecomputed physical parameters. To perform this test, we use anoption included in CIGALE, which allows for the creation of amock object for each galaxy for which the physical parametersare known. To build the artificial catalogue we use the best-fitmodel for each galaxy used for SED fitting (one artificial objectper galaxy). A detailed description of the mock analysis can befound in Lo Faro et al. (2017) and Giovannoli et al. (2011). Inthe next step we perturb the fluxes obtained from the best SEDsaccording to a Gaussian distribution with σ corresponding to theobserved uncertainty for each band. In the final step of the verifi-cation of estimated parameters we run CIGALE on the simulated Article number, page 6 of 17. Małek et al.: HELP: modelling the spectral energy distributions of
Herschel detected galaxies in the ELAIS N1 field log(M star ) [ M (cid:12) ] , r = 0 . y=0.99*x +0.06∆=-0.01 ± log ( L dust ) [ L (cid:12) ], r= 1.00 y=1.00*x -0.02∆=-0.01 ± − log ( SFR ) [ M (cid:12) yr − ], r= 1.00 y=1.01*x +0.00∆=-0.01 ± AGN frac [%], r= 0.85 y=1.16*x -0.53∆=-0.36 ± E x a c t v a l u e s Estimated values − . . . . . . . . log ( M star ) [ M (cid:12) ] − . . . . . . . . log ( L dust ) [ L (cid:12) ] − . − . . . . . . . . . log ( SFR ) [ M (cid:12) yr − ] −
20 0 200 . . . . . . AGN frac [%] N o r m a li ze d c o un t s (estimated − exact) values Fig. 3:
Upper panel:
Comparison between the true value of the output parameter provided by the best-fit model for the mockcatalogue (x-axis) and the value estimated by the code (y-axis), for M star , L dust , SFR and AGN fraction. The Pearson product-moment correlation coe ffi cient is given as an ‘r’ value. The blue line corresponds to the 1:1 relation, while the red dashed line is aregression line with the equation given in the legend. ∆ represents the mean di ff erence between estimated and exact values and thestandard deviation of that di ff erence. Bottom panel: distribution of estimated minus exact parameters from upper panel. Red dashedlines correspond to median values, while black dotted lines represent mean values.sample using the same set of input parameters as for the origi-nal catalogue and compare the output physical parameters of theartificial catalogue with the real ones. A similar reliability checkwas performed for example by da Cunha et al. (2008), Walcheret al. (2011), Yuan et al. (2011), Boquien et al. (2012), Buat et al.(2014), Han & Han (2014) and Ciesla et al. (2015).All the physical parameters presented in Fig. 3 are com-puted from their probability distribution function (PDF) as themean and standard deviation determined from the PDFs (Bo-quien et al., in prep., or Walcher et al. (2008) for a more detailedexplanation of the PDF method.The upper panel of Fig. 3 presents the comparison of theoutput parameters of the mock catalogue with the best valuesestimated by the code for our real galaxy sample. The value,which characterises the reliability of the obtained properties, isthe Pearson product-moment correlation coe ffi cient ( r ). The dis-persion of the main physical parameters is presented in the sameplot as the ∆ value. The lower panel of Fig. 3 shows the distri-bution of estimated minus exact values for M star , L dust , SFR, andAGN frac . We find normal distributions with small ( ∼ star ) and log(L dust ). The third histogram sug-gests a greater overestimation of SFR which is the result of theslight overestimation of L dust which propagates to SFR. The es-timation of the AGN frac is less accurate ( ∆ = − . ± .
50) dueto the grid parameters used for a fraction of the AGNs (we werenot able to put more than four parameters due to the number ofmodels that needed to be created and the time required to analysemillions of galaxies with all models).
The standard global quality of the fitted SEDs is quantified withthe reduced value of χ ( χ divided by the number of data, here-after: χ r ) of the best model; it is not purely reduced χ regarding the statistical definition as the exact number of free parametersfor each galaxy is unknown . The minimum χ r value for thegalaxy still points out the best model from the grid of all pos-sible models created with the input parameters, but due to thevaried number of observed fluxes and unknown number of freeparameters the χ r criteria cannot be use to remove galaxies withunreliable fits from our sample. Instead of χ r , we make use ofthe estimation of physical properties (L dust and M star ) to selectpossible outliers for the SED fitting procedure, and especially,galaxies which do not preserve the energy budget.To calculate the e ffi ciency of using our method to select theouliers, we choose galaxies for which we can estimate the L dust based on the FIR measurements (the coverage of measurements,Fig. 1, and quality of the optical data allow us to calculate stel-lar masses for all of the galaxies, and we do not need to makean additional cut for UV–OPT data). To obtain the most reliablesample of galaxies to test our method based on physical prop-erties of galaxies we select objects with at least two measure-ments for FIR data with signal-to-noise rations (S / N) (cid:62) ∼
50% of the ELAIS N1 catalogue, hereafter:ELAIS N1 3S / N).We run CIGALE two more times: (1) for optical measure-ments only, to estimate the classical stellar mass based on UV–OPT measurements and photometric redshifts only (hereafter:M star , stellar SED ), and (2) for FIR data only, to calculate dust lumi-nosity (hereafter: L dust , IR SED ). In the following steps we compareL dust and M star values obtained from full UV-FIR SED fittingwith L dust , IR SED and M star , stellar SED . the number of degrees of freedom can only be estimated for linearmodels; concerning nonlinear models, the number of degrees of free-dom is absolutely nontrivial and whether or not it can be calculatedproperly is questionable; see e.g. (Andrae et al. 2010) or Chevallard &Charlot (2016) Article number, page 7 of 17 & A proofs: manuscript no. KMalek_final(a)(b)
Fig. 4: Comparison between L dust , UV − FIR SED and L dust , IR SED (panel a) and M star , UV − FIR SED and M star , stellar SED (panel b) . Thesolid black line represents 1:1 relation, while black dashed linescorrespond to 3 σ . Open blue circles show the ELAIS N1 sample,magenta stars (in panel a) show possible energy budget outliers,filled yellow triangles (panel b) show stellar mass outliers, andopen red squares (panels a and b) show objects with inconsisten-cies in both physical parameters (L dust and M star ).We then select galaxies with dust luminosity and / or stellarmasses that are inconsistent with the estimated based on the totalSED fitting. We use two criteria to find those objects: – Criterion 1: L dust inconsistent (within 3 σ level) withL dust , IR SED , Fig. 4a. – Criterion 2: M star inconsistent (within 3 σ level) withM star , stellar SED , Fig. 4b.Based on the combination of these two criteria we selectthree di ff erent groups of objects:(i) Criterion 1 not criterion 2: energy budget issue; we find1 748 (7.08% of the sample) galaxies with inconsistentL dust and L dust , IR SED and consistent estimation of stellar mass (full magenta stars in Fig. 4a) and we refer to themas galaxies with possible energy budget issues. Some ofthe sources are possibly very deeply obscured galaxies (i.e.galaxy in Fig. 5a). After visual inspection we find that ∼
25% of them also show possible photometric redshift prob-lems (incorrect matching between optical and IR counter-parts).(ii) Criterion 2 not criterion 1: problem with matching opti-cal catalogues; we find 353 objects (1.43% of the sample)with inconsistent M star estimation and at the same timeconsistent estimation of dust luminosity (objects markedas yellow triangles in Fig. 4b). Those objects are expectedto have problems with matching between catalogues orthere is a problem with calibration between di ff erent opti-cal data-sets used for the SED fitting (Fig. 5b).(iii) Criteria 1 and 2: 268 objects (1.08% of the sample) withinconsistency between runs in both stellar mass and dustluminosity (criteria 1 and 2, red open squares in Figs. 4aand b). The sub-sample is a mixture of stars, incorrectmatching between stellar and IR catalogues, possible in-correct photometric redshifts and very small S / N for IRmeasurements.In total, based on the physical analysis, we find only 2 369(10% of the ELAIS N1 3S / N sample) peculiar objects (or cat-alogues mismatches) that should not be included in the furtherphysical analysis. χ s criteria To try to remove the objects with inconsistent measurements ofL dust (possible energy budget issue) and / or M star (possible prob-lem with matching optical catalogues) labelled in Sect. 4.2, weuse a combination of di ff erent quantities describing the qualityof the fit. Our selection of galaxies can be correlated with χ scalculated for the stellar (hereafter: χ , stellar ) and the dust (here-after: χ , IR ) parts of the spectra, separately .Figure 6 shows the scatter between both χ s with groups ofoutliers found based on the analysis of physical parameters. Wedefine two rejection criteria used to isolate at least 80% of re-jected objects as: χ , IR (cid:62) . ∧ χ , stellar (cid:54) , black solid lines in Fig. 6 ,χ r , IR (cid:62) ∧ χ r , stellar (cid:62) , black dashed lines in Fig. 6. (5)This simple cut based on the values of χ , IR and χ , stellar allows usto reject: –
81% of objects with a possible problem with energy balance(magenta stars in Figs. 4a and 6, referred to as group (i) inSect. 4.2, –
85% of objects with a possible incorrect match of opticalsurvey or stars (yellow triangles in Figs. 4b and 6, group (ii)in Sect. 4.2), often with stars – we find that 60% of those objects have the nearestGaia star source from GAIA DR1 (Gaia Collaboration et al. 2016) be-tween 0.6 and 2 arcsec distance, or that the stellarity parameter is largerthan 0.5 We defined the wavelength ranges for χ , stellar and χ , IR as ≤ µ m and > µ m in rest frame, respectively, and calculate the values directly basedon the best model.Article number, page 8 of 17. Małek et al.: HELP: modelling the spectral energy distributions of Herschel detected galaxies in the ELAIS N1 field(a) (b) (c)
Fig. 5: An example of (a) possible energy budget issue, (b) incorrectly matched optical data, (c) lensed candidate found using adisagreement between stellar and infrared χ of the SED fitting. The optical part of the spectrum is fitted very well when χ , IR value is on the tail of the χ , IR distribution. The photometric redshift calculated for the IR part only is equal to 3.28 ± / observed flux, areplotted at the bottom of each spectrum. − − χ , IR − − χ r , s t e ll a r ELAIS N1 samplepossible energy balance problem Ldust inconsistent: 1748 (7.08%)possible stars/catalogue problem Mstar inconsistent: 353 (1.43%)Ldust and Mstar inconsistent: 268 (1.08%)
Fig. 6: Scatter between both χ s. Grey-filled hexagons corre-spond to the full ELAIS N1 sample, magenta stars represent thesample of galaxies with possible energy budget problems, yel-low triangles correspond to the galaxies with a possible prob-lem with optical survey matching or stars, and red open squarescorrespond to objects showing inconsistency for both physicalparameters used in our analysis. – and 92% of galaxies with both M star and L dust parametersinconsistent with single runs for stellar and IR parts, respec-tively (open red squares in Figs. 4a, b and 6, group (iii) inSect. 4.2).We stress that the rejection criteria in Eq. 5 are very con-servative in order to avoid excluding too many sources; the fullsample with all calculated χ s will be released to allow the userto make their own quality cuts. Values used for Eq. 5 are validfor ELAIS N1 field, and the thresholds for other HELP fieldsshould be revised.Our two χ s criteria (Eq. 5) remove an additional 2 634galaxies (5.25% of the sample) lying below magenta stars in Fig. 6. We check what kind of objects we reject. We suspectthat some of the sources have problems with matching betweenoptical and IR data and that the rejected sample includes pos-sible incorrect photometric redshift, as in, for example, lensedgalaxies, which are a good example of objects with energy bud-get issues.To check whether or not the mismatch between optical andFIR data is at the basis of the rejection, we calculate z phot , IR -photometric redshifts based on the PACS and SPIRE data only.We make a standard check for photometric redshift outliers (Il-bert et al. 2006): δ (redshift − z phot , IR ) / (1 + redshift) > N, whereredshift corresponds to the photometric redshift calculated basedon UV–OPT data. In the literature, comparisons of photometricand spectroscopic data use a value of N equal to 0.15, but thiscould be too small for comparison between two photometric val-ues. We find that using double this value (N = / N (cid:62) phot , IR estimates re-duces the ELAIS N1 3S / N sub-sample to 1 059 objects only. Wealso find that for this secure sample, only 10% of objects not re-jected by the χ s, and 99% of galaxies selected by Eqs. 5, havea possible problem with photometric redshifts. This suggest thatthe majority of rejected galaxies are incorrect matches betweenoptical and FIR data or peculiar galaxies, as for example lensedgalaxies. To summarise, the contamination of rejected objects isas low as ∼ ff erence between photometric red-shifts estimated based on the UV–OPT and IR measurements forall of them is higher than 0.63 (twice the mean uncertainties ob-tained for the z phot , IR ). An example of a possible lensed candidateis shown in Fig. 5c.Using two χ s to select peculiar objects is not a new tech-nique (but used for the first time in CIGALE tool). For example,Rowan-Robinson et al. (2014) used the quality of the fit of the Article number, page 9 of 17 & A proofs: manuscript no. KMalek_final
IR part of the galaxy spectra to select possible lensing candidatesout of 967 SPIRE sources in the HerMES Lockman survey andfound 109 candidates. We checked if our method to select lensedcandidates (object rejected by Eqs. 5 ∧ | redshift − z phot , IR | > .
63, the nearest Gaia star is further than 0.6 arcsec (we usedflag gaia _ f lag == < χ , stellar rejection criteria (galaxies defined by 5.5 < χ , sterllar <
26 and χ , IR >
11) have special properties. We find that the majorityof them are characterised by very small error bars for UV–NIRmeasurements and with much higher flux uncertainties for
Her-schel fluxes. This implies imprecise SED fitting, but at the sametime, both M star and L dust , calculated separately for UV–IR andIR data, have similar values (below 3 σ threshold, with high un-certainties coming from PDF analysis). Nonetheless, we have nosolid basis to remove those objects from our analysis.To provide general properties of the ELAIS N1 sample weremoved 7 533 galaxies in total based on the two χ s cuts(Eqs. 5). We define the remaining 42 453 galaxies (84% of theoriginal ELAIS N1 sample) as a clean sample for the physicalanalysis. The redshift distribution of the final sample used forSED fitting is presented in Fig. 2 (the red hatched histogram).We stress that χ , χ , stellar , and χ , IR are a part of deliv-erables and each user is free to choose di ff erent criteria (orto not use them at all). For each field we provide the mainphysical parameters without any subjective flags describingthe quality of the fit. Both, the data and the SED fitting re-sults can be downloaded at http://hedam.lam.fr/HELP/dataproducts/dmu28/dmu28_ELAIS-N1/data/zphot/ andalso accessed with Virtual Observatory standard protocols at https://herschel-vos.phys.sussex.ac.uk/ .
5. General properties of the ELAIS N1 sample
The majority of our sample consists of (LIRGs) and ultra lu-minous infrared galaxies (ULIRGs), characterised by 11 ≤ log(L dust / L (cid:12) ) <
12, and 12 ≤ log(L dust / L (cid:12) ) <
13, respectively.We have found 23 214 LIRGs (54.68% of the sample), 11 947ULIRGs (28.14%), and 318 (0.75%) hyper LIRGs (HLIRGs)(log(L dust / L (cid:12) ) ≥ dust / L (cid:12) ) <
11 from the ELAIS N1 field, can be found inTable 3. All delivered parameters (dust luminosity, stellar mass,SFR, χ r , χ , stellar , χ , IR , and calculated thresholds) will be pub-lished while the HELP database is being completed (end of2018). As it was shown in, for example, Ciesla et al. (2015), Salim et al.(2016) and Leja et al. (2018), that not taking into account anAGN component when performing broad band SED fitting of AGN host galaxies results in substantial biases in their derivedparameters.The AGN fraction in CIGALE is calculated as the AGN con-tribution to dust luminosity obtained by SED fitting based onthe bayesian analysis. We find 1 762 galaxies (268 galaxies withlog(L dust ) <
11, 724 LIRGs, 705 ULIRGs and 65 HLIRG) witha significant ( (cid:61) ff er between tworuns. We find that log(M star ) is generally well recovered, even forgalaxies with a large AGN contribution. Galaxies with a signifi-cant AGN contribution have higher SFR when calculating with-out the AGN component, and the di ff erence can be as high as0.5 dex. The result of the test is described in Appendix A. Weconclude that the use of an AGN component is necessary to runCIGALE’s SED fitting for HELP.We also check whether our method is in agreement with vari-ous criteria based on the IRAC colour selections. We select 9 196ELAIS N1 galaxies (21.66% of the final sample) which havemeasurements at all four IRAC bands. For 580 of them, we findan AGN component according to the CIGALE SED fitting. Wefocus on four di ff erent criteria for AGN selection based on theNIR data: Stern et al. (2005); Donley et al. (2012); Lacy et al.(2007, 2013). The majority of the 580 SED AGNs meet the listedcriteria (see Table 4). Figure 7 shows criteria listed above (areasmarked with black dashed lines) and galaxies with a significantAGN contribution according to the SED fitting (marked as redpoints).We conclude that the usage of the AGN module (Fritz et al.2006) for all ELAIS N1 galaxies gives us a reasonable (accord-ing to NIR critera) sample of AGNs. For example, only ∼ star , SFR, L dust ). The AGN identi-fication is a by-product of our analysis. Tests presented aboveshow that the inclusion of emission from dusty AGNs does notgenerate significant biases for estimated properties.
6. Dust attenuation recipes
Based on our sample we explore how physical parameters ob-tained by fitting SEDs with di ff erent dust attenuation modelscan be biased. We apply three di ff erent attenuation laws to ex-plore this subject. We perform two additional SED fitting runs:(1) the Calzetti et al. (2000) recipe that is widely used in theliterature, and (2) the Lo Faro et al. (2017) law obtained for IRbright galaxies. The main change between those attenuation lawsis the slope for the part of the spectrum with λ > ff erences between all three recipes for dust attenua-tion considered in this paper.We find, based on the quality of the fits, that 45% of galaxiesare fitted better with CF00 than the other two attenuation laws.The Lo Faro et al. (2017) procedure works very well for 24%of cases and Calzetti et al. (2000) for 29%. It was expected that Article number, page 10 of 17. Małek et al.: HELP: modelling the spectral energy distributions of
Herschel detected galaxies in the ELAIS N1 field
Table 3: The main physical properties of the sample of 42 453 galaxies presented in total IR luminosity bins. The first columncorresponds to the galaxy type according to their total log(L dust ) value, the second shows the number of galaxies (first row), and thetotal percentage of the sample (second row), column (3) presents the number of galaxies with an AGN contribution higher than 20%(first row – total number, second row – percentage value of galaxiwhiches with AGN contribution within the given IR luminositybin). Median values for redshift, stellar mass, SFR and specific SFR are given in columns (4) - (7). The errors are calculated asmedian absolute deviation. N gal N AGN redshift log(M star ) log(SFR) log(sSFR)(% ) (%) [M (cid:12) ] [M (cid:12) yr −
1] [yr − ](1) (2) (3) (4) (5) (6) (7) log(L dust ) < ± ± ± ± ≤ log(L dust ) <
23 214 724 0.96 ± ± ± ± ≤ log(L dust ) <
11 947 705 1.89 ± ± ± ± log(L dust ) ≥
318 65 4.92 ± ± ± ± − − − [5 . − [8 .
0] VEGA − . . . . . [ . ] − [ . ] V E G A Stern et al. 2005
S05criterion: no. ofgal.: 1672SEDAGNsinsideS05: 393(67.76%) − . − . . . log(S . / S . ) − . − . . . . l og ( S . / S . ) Donley et al. 2012
D12criterion: no. ofgal.: 779SEDAGNsinsideD12: 339(58.45%) − . − . . . log(S . / S . ) − . − . . . . l og ( S . / S . ) Lacy et al. 2007
L07criterion: no. ofgal.: 2439SEDAGNsinsideL07: 435(75.0%) − log(S . / S . ) − . − . . . . l og ( S . / S . ) Lacy et al. 2013
L13criterion: no. ofgal.: 5082SEDAGNsinsideL13: 485(83.62%) . . . . . A G N f r a c t i o n Fig. 7: Stern et al. (2005), Donley et al. (2012), Lacy et al. (2007) and Lacy et al. (2013) MIR selection of AGNs. Black dashedlines correspond to the criteria. 546 galaxies from our sample for which the
AGN f rac > = . ∼ ∼
2, and morethan half of the objects have log(L dust ) ≥ (cid:12) ]).As was already shown in Lo Faro et al. (2017) and Mitchellet al. (2013), the influence of the attenuation curve on the derived stellar masses can be strong (e.g. Mitchell et al. 2013, showedthat for the most extreme cases the di ff erence can reach a factorof 10, with the median value around 1.4) compared to greyer at-tenuation curves with the standard Calzetti et al. (2000) recipe.A flatter attenuation curve at longer wavelengths results in largerstellar masses. We now compare these findings to our ELAIS-N1sample. For both additional runs with Calzetti et al. (2000) andLo Faro et al. (2017) recipes, we used exactly the same param-eters for the other modules. All parameters used for SED fittingare listed in Table 2.For comparison we use only galaxies without possible out-liers (we used Eqs. 5 presented is Sect. 4.3 for runs with Calzettiet al. (2000) and Lo Faro et al. (2017) attenuation laws). In to-tal we perform a comparison between CF00 and Calzetti et al.(2000) for 37 682 galaxies, and Charlot & Fall 2000 and LoFaro et al. (2017) for 38 301 objects. We find an agreement be-tween all three runs for L dust and SFR (1:1 with slight scatter: seeFig. 10 for details). We conclude that using either of the attenua-tion laws of CF00, Calzetti et al. (2000), or Lo Faro et al. (2017)has no important impact on dust luminosity or SFR estimation.Nevertheless, the choice of attenuation law has a substantial im-pact on calculation of stellar mass. Article number, page 11 of 17 & A proofs: manuscript no. KMalek_final . . . . . . . log(M star ) [M (cid:12) ] . . . . . . . . . . N o r m a li ze d c o un t s CF00Calzetti2000LoFaro2017
Fig. 8: Comparison of stellar masses obtained with three di ff er-ent dust attenuation laws: Charlot & Fall (2000) – black stripedhistogram, Calzetti et al. (2000) - blue right hatched histogram,and Lo Faro et al. (2017) - red left hatched histogram.Fig. 9: Total dust attenuation in NIR filter estimated underthe assumption of the Charlot & Fall (2000) dust attenuationcurve with the Calzzetti dust attenuation law (full blue dots) andLo Faro model (full red dots). The black solid line correspondsto the 1:1 relation while blue and red dashed lines correspondto the relation between Charlot & Fall (2000) and Calzetti et al.(2000), and Charlot & Fall (2000) and Lo Faro et al. (2017) at-tenuation laws, respectively.The right panels of Fig. 10 show the di ff erence betweenlog(M star ) obtained with CF00 and Calzetti et al. (2000) (upperpanel) and CF00 and Lo Faro et al. (2017) (bottom panel) laws.Here we can see a significant shift between both values, whichbecomes stronger for more massive galaxies for Calzetti versusCF00 attenuation laws.Figure 8 shows log(M star ) distributions obtained for runs withthree di ff erent attenuation laws. It is clearly visible that stellarmasses obtained with the Calzetti et al. (2000) law are on aver-age lower than those from CF00 and Lo Faro et al. (2017). De-rived mean values of log(M star ) [M (cid:12) ] are equal to 10.50 ± ± ± ff erent attenuation laws: For Calzetti et al. (2000) and CF00:log(M star ) Calzetti = . × log(M star ) CF00 + . , (6) and for Lo Faro et al. (2017) and CF00:log(M star ) LoFaro = . × log(M star ) CF00 + . . (7)We note that we find the same relations excluding IR datafrom the SED fitting. Our analysis of the influence of the dustattenuation law for estimated dust luminosity, SFR, and stel-lar mass based on the stellar emission only is presented in Ap-pendix B.As was also checked by Lo Faro et al. (2017) for a sampleof IR selected [U]LIRGs, the total amount of attenuation for ayoung population (in UV ranges) is usually well constrained bydi ff erent attenuation laws, but it this is not the case in NIR wave-length range. We find that attenuation in the NIR range is notpreserved between di ff erent attenuation laws, and is on averagelarger for Lo Faro et al. (2017) and Charlot & Fall (2000) thanfor Calzetti et al. (2000) (Fig. 9). This relation is a direct out-come of the shape of adopted attenuation curves. We find thatthe ratio between mean M star obtained from Calzetti et al. (2000)and CF00 is equal to 0.98 ± star obtained from Lo Faro et al. (2017) and CF00 is 1.49 ± ± ± / CF00 and Lo-Faro / CF00, respectively. We summarise that the shape of the at-tenuation laws in the NIR band is responsible for inconsistencyin stellar masses.The change in stellar mass has an impact on specific SFR andmay also introduce a flatter shape for high-mass galaxies at theso-called main sequence galaxies (stellar mass vs. SFR relation,see Fig. 11), as for the high-mass end we have more massivegalaxies for Charlot & Fall (2000) and Lo Faro et al. (2017) thanfor Calzetti et al. (2000) with the same SFR values.In summary the attenuation law used for SED fitting has animpact on the measure of stellar mass, but has little influenceon other parameters, and it should be carefully chosen and takeninto account when the properties of stellar masses are discussed.
7. Dust luminosity prediction from stellar emission
The FIR emission is a key component to accurately determinethe SFR of galaxies. The strong correlation between L dust andSFR implies that an accurate estimate of total dust luminositycorresponds to a better SFR estimation. However, IR surveys areextremely expensive, and also usually su ff er from low spatialresolution, with respect to optical, and therefore result in sourceconfusion. Moreover, the high-redshift galaxies are mostly ex-plored in the visible and NIR wavelengths only. We investigatedthe relation between L dust estimation based on di ff erent wave-length ranges. Based on the sample of ELAIS N1 galaxies weverified whether or not it was possible to predict the total dust lu-minosity based on the optical and NIR data only. Two additionalruns were performed. We ran CIGALE with the same parametersas for the initial sample of ELAIS N1, but without IR data (rest-frame λ ≤ µ m); we refer to that run as stellar . We also madea second run with IR data only ( Herschel
PACS and SPIRE mea-surements), with the same models and parameters (run
IRonly );we refer to our initial run, with all 19 bands, as
UV(cid:21)IR .For our analysis we use galaxies with at least one PACS mea-surement with S / N ≥ / N ≥
2. Selected galaxies have at leastsix UV–NIR measurements and three
Herschel measurements(PACS and SPIRE) to cover all strategic parts of the spectra forSED fitting. Our selection, together with χ s criteria (Eqs. 5), Article number, page 12 of 17. Małek et al.: HELP: modelling the spectral energy distributions of
Herschel detected galaxies in the ELAIS N1 field log(L dust ) CF [L (cid:12) ] l og ( L du s t ) C a l ze tt i [ L (cid:12) ] N = 37682 r= 0.98y=1.01*x -0.151:1 (a) log(L dust ) log(SFR) CF200 [M (cid:12) yr − ] l og ( S F R ) C a l ze tt i [ M (cid:12) y r − ] N = 37682 r= 0.98y=1.00*x +0.021:1 (b) log(SFR) log(M star ) CF [M (cid:12) ] l og ( M s t a r ) C a l ze tt i [ M (cid:12) ] N = 37682 r= 0.96y=0.91*x +0.791:1 (c) log(M star ) log(L dust ) CF [L (cid:12) ] l og ( L du s t ) L o F a r o [ L (cid:12) ] N = 38301 r= 0.99y=1.02*x -0.151:1 (d) log(L dust ) log(SFR) CF200 [M (cid:12) yr − ] l og ( S F R ) L o F a r o [ M (cid:12) y r − ] N = 38301 r= 0.99y=1.00*x +0.021:1 (e) log(SFR) log(M star ) CF [M (cid:12) ] l og ( M (cid:12) ) L o F a r o [ M (cid:12) ] N = 38301 r= 0.97y=0.99*x +0.261:1 (f) log(M star ) Fig. 10: Comparison of main physical parameters (log(L dust ), log(SFR), and log(M star )) for a sample of 37 682 galaxies fitted withthe Calzetti et al. 2000 attenuation law (y-axis), or the Charlot & Fall 2000 attenuation law (x-axis) ( upper panel ), and for a sampleof 38 301 galaxies fitted with the Lo Faro et al. 2017 attenuation law (y-axis), or the Charlot & Fall 2000 attenuation law (x-axis) ( bottom panel ). 1:1 relation is marked with blue solid line, fitted relation - as red dashed line. Each panel includes Pearsonproduct-moment correlation coe ffi cient ( r ).Fig. 11: SFR – M star relation for Calzetti et al. (2000), CF00 and Lo Faro et al. (2017) attenuation laws with colour-coded redshiftvalues. Black dashed-dotted line represents log(M star ) = . (cid:12) ] and dashed line log(M star ) = . (cid:12) ]. Both lines can help todistinguish the di ff erence between the number of massive galaxies according to the scenario used for SED fitting. Numbers given ineach panel correspond to the number of galaxies inside the range 11.5-12.3 log(M star ).results in a sample of 586 galaxies with good coverage of thespectrum and the best photometric measurements.We check the relation between total dust luminosity for allthree runs. First of all we find very good agreement betweenL dust obtained from stellar(cid:21)IR and stellar runs (Fig. 12a).We find the relation to be:log(L dust ) stellar = (1 . ± . × log(L dust ) UV(cid:21)IR + . . (8) We combined a full run on ( UV(cid:21)IR ) with one on
IRonly , inwhich dust luminosity was calculated based on the
Herschel dataonly. Again, we found very good agreement (see Fig. 12b).log(L dust ) IRonly = (0 . ± . × log(L dust ) UV(cid:21)IR + . . (9)Finally we performed the comparison of total dust luminos-ity calculated based on the stellar ( stellar ) data with L dust based on the IR data only ( IRonly ) (Fig. 12c). We have found a
Article number, page 13 of 17 & A proofs: manuscript no. KMalek_final
10 11 12 13 14 log(Ldust) stellar+IR [L (cid:12) ] l og ( L du s t) s t e ll a r [ L (cid:12) ] N = 586, r= 0.96 y=1.00*x +0.05∆=-0.08 ± (a) stellar(cid:21)IR vs stellar
10 11 12 13 log(Ldust) stellar+IR [L (cid:12) ] . . . . . . . . l og ( L du s t) I R o n l y [ L (cid:12) ] N = 586, r= 0.98 y=0.96*x +0.46∆=-0.06 ± (b) stellar(cid:21)IR vs IRonly
10 11 12 13 14 log(Ldust) stellar [L (cid:12) ] l og ( L du s t) I R [ L (cid:12) ] N = 586, r= 0.95 y=0.89*x +1.24∆=0.01 ± (c) stellar vs IRonly
Fig. 12: Comparison of total dust luminosity estimated using di ff erent wavelength range for 686 ELAIS N1 galaxies withPACS green and PACS red measurements with S / N ≥ / N ≥ dust estimated based on the UV–NIRdata only (y-axis); (b) results of SED fitting based on the full sets of data (x-axis) compared to estimates from PACS and SPIREdata only (y-axis); and (c) results of SED fitting based on the UV–NIR data (x-axis) vs. estimates from PACS and SPIRE data only(y-axis). Blue solid lines represent 1:1 relations, while red dashed lines correspond to linear fits to the data. Pearson product-momentcorrelation coe ffi cient is given as an r value. ∆ represents the mean di ff erence between x-axis and y-axis values and the standarddeviation of that di ff erence.clear relation for L dust values:log(L dust ) IRonly = (0 . ± . × log(L dust ) stellar + . . (10)The relation has a much larger slope and scatter but again, weconclude that with CIGALE and the sets of parameters presentedin Table 2 we are able to predict the dust luminosity of star form-ing galaxies based on the UV–NIR data only. The L dust estimatedbased on the UV–NIR data is slightly overestimated for faint ob-jects and underestimated for LIRGS and HLIRGS but Eq. 10,which can be used to calibrate L dust obtained from the stellaremission only, takes this into account. Estimated and calibratedvalues of L dust can be used for a proper estimation of SFR.
8. Summary and conclusions
We present a strategy for SED fitting that is applied to the
Herschel
Extragalactic Legacy Project (HELP) which coversroughly 1 300 deg of the Herschel
Space Observatory. We havefocused on the ELAIS N1 as a pilot field for the HELP SED fit-ting pipeline. We show how the quality of the fits and all mainphysical parameters were obtained. We present automated relia-bility checks to ensure that the results from the ELAIS-N1 fieldare of consistently high quality, with measures of the quality forevery object. This ensures that the final HELP deliverable can beused for further statistical analysis.We introduced the two χ s procedure to remove incorrectSED fits. We calculate two new χ s, apart from a standard χ calculated in CIAGLE for the global fit: one for the stellar part( χ , stellar ) and one for the IR part ( χ , IR ) of the spectrum. Wedefined a threshold between χ , stellar and χ , IR as 8 µ m in the rest-frame. We demonstrate that the combination of two di ff erent χ scan e ffi ciently select outliers and peculiar objects based on theenergy balance assumption.We find that only 4.15% of galaxies of the ELAIS N1field show an AGN contribution. We check, using four di ff erentwidely known criteria for AGN selection based on the nNIR data (Stern et al. 2005; Donley et al. 2012; Lacy et al. 2007, 2013),that the majority of AGNs found based on the SED fitting proce-dure (67.7%, 58.4%, 75.0%, and 83.6% respectively) fulfill theNIR criteria.We test the influence that di ff erent attenuation laws have onthe main physical parameters of galaxies. We compare resultsusing Charlot & Fall (2000), Calzetti et al. (2000), and Lo Faroet al. (2017) recipes. We conclude that using di ff erent attenua-tion laws has almost no influence on the calculation of total dustluminosity or SFR, but we find a discrepancy between obtainedstellar masses, which we find to be a direct result of the shapeof the adopted attenuation curves in NIR wavelengths. We pro-vide relations between stellar masses obtained under those threeassumptions of attenuation curves. We find that on average thevalues of stellar masses for the ELAIS N1 sample can vary upto a factor of approximately two when calculated with di ff erentattenuation laws. We demonstrate that the recipe given by Char-lot & Fall (2000) more often (for 45% of ELAIS N1 galaxies)outperforms that proposed by Calzetti et al. (2000) and Lo Faroet al. (2017) (according to χ , IR and χ , stellar values).We check the accuracy of estimating dust luminosity fromstellar emission only and conclude that with CIGALE and thesets of parameters presented in Table 2 we are able to pre-dict L dust for standard IR galaxies, which preserve energy bud-get, based on the UV–NIR data only. Predicted L dust is in verygood agreement with the dust luminosity estimated based on fullspectra and stellar emission only. We are not able to estimatemonochromatic fluxes but only the total value of L dust . Our testsshow that the SFR, tightly correlated with total dust luminosity,is also properly estimated. Our predictions can be used to designnew surveys and as priors for the IR extraction pipeline (e.g. forXID + , Pearson et al. 2018). References
Aihara, H., Arimoto, N., Armstrong, R., et al. 2018, PASJ, 70, S4Andrae, R., Schulze-Hartung, T., & Melchior, P. 2010, ArXiv e-printsArnouts, S., Cristiani, S., Moscardini, L., et al. 1999, MNRAS, 310, 540
Article number, page 14 of 17. Małek et al.: HELP: modelling the spectral energy distributions of
Herschel detected galaxies in the ELAIS N1 field
Bolzonella, M., Miralles, J.-M., & Pelló, R. 2000, A&A, 363, 476Boquien, M., Buat, V., Boselli, A., et al. 2012, A&A, 539, A145Boselli, A. 2011, A Panchromatic View of GalaxiesBrammer, G. B., van Dokkum, P. G., & Coppi, P. 2008, ApJ, 686, 1503Brown, M. J. I., Moustakas, J., Smith, J.-D. T., et al. 2014, ApJS, 212, 18Bruzual, G. & Charlot, S. 2003, MNRAS, 344, 1000Buat, V. 2005, ApJ, 619, L51Buat, V., Heinis, S., Boquien, M., et al. 2014, A&A, 561, A39Buat, V., Oi, N., Heinis, S., et al. 2015, A&A, 577, A141Buat, V., Takeuchi, T. T., Iglesias-Páramo, J., et al. 2007, ApJ Supplement Series,173, 404Burgarella, D., Buat, V., Gruppioni, C., et al. 2013, A&A, 554, A70Calistro Rivera, G., Lusso, E., Hennawi, J. F., & Hogg, D. W. 2016, ApJ, 833, 98Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682Calzetti, D., Liu, G., & Koda, J. 2012, ApJ, 752, 98Casey, C. M. 2012, MNRAS, 425, 3094Chabrier, G. 2003, PASP, 115, 763Charlot, S. & Fall, S. M. 2000, ApJ, 539, 718Chevallard, J. & Charlot, S. 2016, MNRAS, 462, 1415Cid Fernandes, R., Mateus, A., Sodré, L., Stasi´nska, G., & Gomes, J. M. 2005,MNRAS, 358, 363Ciesla, L., Boselli, A., Elbaz, D., et al. 2016, A&A, 585, A43Ciesla, L., Charmandaris, V., Georgakakis, A., et al. 2015, A&A, 576, A10da Cunha, E., Charlot, S., & Elbaz, D. 2008, MNRAS, 388, 1595da Cunha, E., Charmandaris, V., Díaz-Santos, T., et al. 2010, A&A, 523, A78Dale, D. A., Gil de Paz, A., Gordon, K. D., et al. 2007, APJ, 655, 863Dale, D. A., Helou, G., Magdis, G. E., et al. 2014, A Two-parameter Model forthe Infrared / Submillimeter / Radio Spectral Energy Distributions of Galaxiesand Active Galactic NucleiDonley, J. L., Koekemoer, A. M., Brusa, M., et al. 2012, ApJ, 748, 142Draine, B. T. 2003, ARA&A, 41, 241Draine, B. T., Aniano, G., Krause, O., et al. 2014, ApJ, 780, 172Draine, B. T. & Li, A. 2007, ApJ, 657, 810Duncan, K. J., Brown, M. J. I., Williams, W. L., et al. 2018, MNRAS, 473, 2655Fabbiano, G. 2006, ARA&A, 44, 323Franzetti, P., Scodeggio, M., Garilli, B., Fumana, M., & Paioro, L. 2008, in As-tronomical Society of the Pacific Conference Series, Vol. 394, AstronomicalData Analysis Software and Systems XVII, ed. R. W. Argyle, P. S. Bunclark,& J. R. Lewis, 642Fritz, J., Franceschini, A., & Hatziminaoglou, E. 2006, MNRAS, 366, 767Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. 2016, A&A, 595, A2Genzel, R. & Cesarsky, C. J. 2000, Annual A&AR, 38, 761Giovannoli, E., Buat, V., Noll, S., Burgarella, D., & Magnelli, B. 2011, A&A,525, A150González-Solares, E. A., Irwin, M., McMahon, R. G., et al. 2011, MNRAS, 416,927Gri ffi n, M. J., Abergel, A., Abreu, A., et al. 2010, A&A, 518, L3Han, Y. & Han, Z. 2014, ApJS, 215, 2Hao, C.-N., Kennicutt, R. C., Johnson, B. D., et al. 2011, ApJ, 741, 124Hatziminaoglou, E., Fritz, J., & Jarrett, T. H. 2009, MNRAS, 399, 1206Hurley, P. D., Oliver, S., Betancourt, M., et al. 2017, MNRAS, 464, 885Ilbert, O., Arnouts, S., McCracken, H. J., et al. 2006, A&A, 457, 841Kennicutt, Jr., R. C. 1998, ARA&A, 36, 189Koleva, M., Prugniel, P., Bouchard, A., & Wu, Y. 2009, A&A, 501, 1269Komatsu, E., Smith, K. M., Dunkley, J., et al. 2011, ApJS, 192, 18Lacy, M., Petric, A. O., Sajina, A., et al. 2007, AJ, 133, 186Lacy, M., Ridgway, S. E., Gates, E. L., et al. 2013, ApJS, 208, 24Lawrence, A., Warren, S. J., Almaini, O., et al. 2007, MNRAS, 379, 1599Le Floc’h, E., Papovich, C., Dole, H., et al. 2005, ApJ, 632, 169Leja, J., Johnson, B. D., Conroy, C., & van Dokkum, P. 2018, ApJ, 854, 62Leja, J., Johnson, B. D., Conroy, C., van Dokkum, P. G., & Byler, N. 2017, ApJ,837, 170Lo Faro, B., Buat, V., Roehlly, Y., et al. 2017, MNRAS, 472, 1372Lonsdale, C. J., Smith, H. E., Rowan-Robinson, M., et al. 2003, PASP, 115, 897Magdis, G. E., Daddi, E., Béthermin, M., et al. 2012, ApJ, 760, 6Małek, K., Bankowicz, M., Pollo, A., et al. 2017, A&A, 598, A1Mauduit, J.-C., Lacy, M., Farrah, D., et al. 2012, PASP, 124, 714Mitchell, P. D., Lacey, C. G., Baugh, C. M., & Cole, S. 2013, MNRAS, 435, 87Nguyen, H. T., Schulz, B., Levenson, L., et al. 2010, A&A, 518, L5Noll, S., Burgarella, D., Giovannoli, E., et al. 2009, A&A, 507, 1793Oliver, S., Rowan-Robinson, M., Alexander, D. M., et al. 2000, MNRAS, 316,749Oliver, S. J., Bock, J., Altieri, B., et al. 2012, MNRAS, 424, 1614Pearson, W. J., Wang, L., Hurley, P. D., et al. 2018, ArXiv e-printsPilbratt, G. L., Riedinger, J. R., Passvogel, T., et al. 2010, A&A, 518, L1Poglitsch, A., Waelkens, C., Geis, N., et al. 2010, A&A, 518, L2Rowan-Robinson, M., Lari, C., Perez-Fournon, I., et al. 2004, MNRAS, 351,1290Rowan-Robinson, M., Wang, L., Wardlow, J., et al. 2014, MNRAS, 445, 3848Salim, S., Lee, J. C., Janowiecki, S., et al. 2016, ApJS, 227, 2 Salvato, M., Hasinger, G., Ilbert, O., et al. 2009, ApJ, 690, 1250Smith, J. D. T., Draine, B. T., Dale, D. A., et al. 2007, ApJ, 656, 770Stern, D., Eisenhardt, P., Gorjian, V., et al. 2005, ApJ, 631, 163Surace, J. A., Shupe, D. L., Fang, F., et al. 2005, in Bulletin of the AmericanAstronomical Society, Vol. 37, American Astronomical Society Meeting Ab-stracts, 1246Swinbank, A. M. 2013, in Astrophysics and Space Science Proceedings, Vol. 37,Thirty Years of Astronomical Discovery with UKIRT, ed. A. Adamson,J. Davies, & I. Robson, 299Takeuchi, T. T., Buat, V., & Burgarella, D. 2005, A&A, 440, L17Tojeiro, R., Heavens, A. F., Jimenez, R., & Panter, B. 2007, MNRAS, 381, 1252Tudorica, A., Hildebrandt, H., Tewes, M., et al. 2017, A&A, 608, A141Vaccari, M. 2015, SALT Science Conference 2015 (SSC2015), 17Vaccari, M. 2016, The Universe of Digital Sky Surveys, 42, 71Vaccari, M., Marchetti, L., Franceschini, A., et al. 2010, A&A, 518, L20Walcher, C. J., Lamareille, F., Vergani, D., et al. 2008, A&A, 491, 713Walcher, J., Groves, B., Budavári, T., & Dale, D. 2011, Ap&SS, 331, 1Yuan, F., Takeuchi, T. T., Buat, V., et al. 2011, PASJ, 63, 1207 Acknowledgements.
The project has received funding from the European UnionSeventh Framework Programme FP7 / / under grant agreement number60725. KM has been supported by the National Science Centre (grant UMO-2013 / / D / ST9 / Article number, page 15 of 17 & A proofs: manuscript no. KMalek_final
Appendix A: log(M star ) and SFR obtained with andwithout AGN module l og ( M s t a r ) n o A G N [ M (cid:12) ] − . . . l og ( S F R ) n o A G N [ M (cid:12) y r − ] A G N f r a c log(M star ) AGN [M (cid:12) ] − . . . A G N − n o A G N σ = 0 . − log(SFR) AGN [M (cid:12) yr − ] − . − . − . . . A G N − n o A G N σ = 0 . Fig. A.1: Comparison of the log(M star ) and SFR obtained bySED fitting without (y-axis) and with (x-axis) AGN module withcolour-coded AGN fraction. Red dashed dotted lines in upperpanels represent 1:1 relations. Bottom panels show di ff erencebetween both runs. Black solid lines correspond to no di ff erence,while the dashed lines represent 3 σ dispersion.We perform a run of the SED fitting with the same pa-rameters for physical models but without AGN component tocheck how the stellar mass and the SFR di ff er between two runs(Fig. A.1). We find that log(M star ) is generally well recovered,even for galaxies with high AGN contribution, with the standarddeviation calculated for the di ff erence of log(M star ) estimatedwith and without AGN module equal to 0.026. Galaxies withsignificant AGN contribution have higher SFR when calculatingwithout the AGN component. For galaxies characterised by thehigh fraction of AGN emission the di ff erence can be as high as0.5 dex. Our result is in agreement with Ciesla et al. (2015) whiofind that the presence of an AGN can bias SFR estimates startingat AGN contributions higher than 10%, reaching 100% overesti-mation for AGN fraction ∼
70% (Ciesla et al. 2015, Appendix A,results for Type 2 AGNs used for our analysis).
Appendix B: Dust attenuation recipes for stellaremission only
We check if the IR data usage can change our result presentedin Sect. 6 obtained with data covering wavelength range fromoptical to IR. We perform the analysis of influence of Charlot &Fall (2000) and Calzetti et al. (2000) attenuation laws for SEDmodelling of stellar emission only. From our test we excludedthe Lo Faro et al. (2017) formula as an extreme case: this law ismuch flatter than those by CF00 and Calzetti et al. (2000) and isdi ff erent suitable for very large attenuation.We run CIGALE with the same set of modules and param-eters as for the main analysis (see Table 2) but without IR data.First we run it with CF00 dust attenuation module and, in thesecond step, with Calzetti et al. (2000).We find that SED fitting with Charlot & Fall (2000) law givesbetter quality fits: 71% of galaxies were fitted better (accordingto the χ r values) with CF00 than with Calzetti et al. (2000). Thisresults is in agreement with our previous finding presented inSect. 6.Similarly to in Sect. 6, we check the influence of the dustattenuation law on estimations of stellar mass. For comparison we use only galaxies with at least ten optical measurements. Intotal we perform the comparison for 22 262 galaxies.The L dust (Fig. B.1 a), and consistently the SFR (Fig. B.1 b),obtained from two runs are scattered in the region of ULIRGsand HLIRGs (we obtained a similar scatter in Fig. 12 c compar-ing the the total dust luminosity estimated using di ff erent wave-length ranges. As L dust and SFR are tightly related, the samee ff ect is visible in Fig. B.1 b. Figure B.1 c shows the relationbetween log(M star ) obtained with the attenuation laws of Charlot& Fall (2000) and Calzetti et al. (2000) by fitting models for thestellar part only. Obtained relation (Eq. B.1) is in a perfect agree-ment with Eq. 6 obtained for the full spectral fitting (includingIR data). log ( M star ) Calzetti = . × log ( M star ) CF + . . (B.1)We conclude that the relation between stellar mass estimatedwith attenution laws by CF00 and Calzetti et al. (2000) is validwith and without IR data included for SED fitting.We check whether or not using CF00 or Calzetti et al. (2000)attenuation law to calculate L dust from the stellar emission onlycan give some bias in function of redshift or L dust estimatedbased on the full (optical + IR emission) SED fitting. We findthat in both cases the scatter is very similar but in case of theCalzetti et al. (2000) recipe we find a clear dependence with L dust and redshift: for bright IR galaxies, the L dust calculated based onthe stellar emission only is underestimated. The same relationapplies for more distant objects (Fig. B.2).
Article number, page 16 of 17. Małek et al.: HELP: modelling the spectral energy distributions of