The miniJPAS survey: Identification and characterization of galaxy populations with the J-PAS photometric system
R. M. González Delgado, L. A. Díaz-García, A. de Amorim, G. Bruzual, R. Cid Fernandes, E. Pérez, S. Bonoli, A. J. Cenarro, P. R. T. Coelho, A. Cortesi, R. García-Benito, R. López Fernández, G. Martínez-Solaeche, J. E. Rodríguez-Martín, G. Magris, A. Mejía-Narvaez, D. Brito-Silva, L. R. Abramo, J. M. Diego, R. A. Dupke, A. Hernán-Caballero, C. Hernández-Monteagudo, C. López-Sanjuan, A. Marín-Franch, V. Marra, M. Moles, A. Montero-Dorta, C. Queiroz, L. Sodré Jr., J. Varela, H. Vázquez Ramió, J. M. Vílchez, P. O. Baqui, N. Benítez, D. Cristóbal-Hornillos, A. Ederoclite, C. Mendes de Oliveira, T. Civera, D. Muniesa, K. Taylor, E. Tempel, J-PAS collaboration
AAstronomy & Astrophysics manuscript no. GalEvo_AEGIS_25Feb © ESO 2021March 1, 2021
The miniJPAS survey:
Identification and characterization of galaxy populations with the J-PASphotometric system
R. M. González Delgado , L. A. Díaz-García , , A. de Amorim , G. Bruzual , R. Cid Fernandes , E. Pérez ,S. Bonoli , , , A. J. Cenarro , P. R. T. Coelho , A. Cortesi , R. García-Benito , R. López Fernández ,G. Martínez-Solaeche , J. E. Rodríguez-Martín , G. Magris , A. Mejía-Narvaez , D. Brito-Silva , L. R. Abramo ,J. M. Diego , R. A. Dupke , , , A. Hernán-Caballero , C. Hernández-Monteagudo , , , C. López-Sanjuan ,A. Marín-Franch , V. Marra , , , M. Moles , , A. Montero-Dorta , , C. Queiroz , L. Sodré Jr. , J. Varela ,H. Vázquez Ramió , J. M. Vílchez , P. O. Baqui , N. Benítez , D. Cristóbal-Hornillos , A. Ederoclite , C. Mendes deOliveira , T. Civera , D. Muniesa , K. Taylor , E. Tempel , and the J-PAS collaboration (A ffi liations can be found after the references) March 1, 2021
ABSTRACT
The Javalambre-Physics of the Accelerating Universe Astrophysical Survey (J-PAS) will soon start imaging thousands of squaredegrees of the northern sky with its unique set of 56 filters (spectral resolution of R ∼ on the AEGIS field with an interim camera with all the J-PAS filters. Taking advantage of this data, dubbedminiJPAS, we aim at proving the scientific potential of J-PAS to derive the stellar population properties of galaxies via fitting codesfor spectral energy distributions (SEDs), with the ultimate goal of performing galaxy evolution studies across cosmic time. Oneparametric ( BaySeAGal ) and three non-parametric (
MUFFIT , AlStar , TGASPEX ) SED-fitting codes are used to constrain the stellarmass, age, metallicity, extinction, and rest-frame and dust-corrected ( u − r ) colours of a complete flux-limited sample ( r SDSS ≤ . z =
1. We generally find consistent results on the galaxy properties derived from the di ff erentcodes, independently of the galaxy spectral-type or redshift; this is remarkable considering that 25% of the J-spectra have signal-to-noise ratio (SNR) ∼
3. For galaxies with SNR ≥
10, we estimate that the J-PAS photometric system allows to derive stellar populationproperties, such as the rest-frame ( u − r ) colour, stellar mass, extinction, and the mass-weighted age, with a precision of 0 . ± . . ± .
03 dex, 0 . ± .
09 mag, and 0 . ± .
07 dex, respectively. This precision is equivalent to that obtained with spectroscopicsurveys of similar SNR. By using the dust–corrected ( u − r ) colour–mass diagram, a powerful proxy to characterize galaxy populations,we find that: (i) the fraction of red and blue galaxies evolves with cosmic time, with red galaxies being ∼
38% and ∼
18% of the wholepopulation at z = . z = .
5, respectively; (ii) consistent results between codes for the average intrinsic ( u − r ) colour, stellarmass, age and stellar metallicity of blue and red galaxies and their evolution up to z =
1. At all redshifts, the more massive galaxiesbelong to the red sequence and these galaxies are typically older and more metal rich than their counterparts in the blue cloud. Ourresults confirm that with J-PAS data we will be able to analyze large samples of galaxies up to z ∼
1, with galaxy stellar masses aboveof log( M (cid:63) / M (cid:12) ) ∼ .
9, 9 .
5, and 9 . z = .
3, 0 .
5, and 0 .
7, respectively. The SFH of a complete sub-sample of galaxies selected at z ∼ . M (cid:63) / M (cid:12) ) > . z ∼ Key words.
Surveys–Techniques: photometric – galaxies: evolution – galaxies: stellar content – galaxies: fundamental parameters
1. Introduction
Constraining the geometry and the expansion rate of the Uni-verse is of paramount importance in cosmology, since it is inti-mately related to the fundamental components of our Universe.Large cosmological surveys have been designed to perform com-plementary cosmological experiments to unveil the nature ofdark matter and dark energy, the latter by precise measurementsof the expansion rate of the universe (Weinberg et al. 2013).Spectroscopic surveys, such as eBOSS (Ahumada et al. 2020),DESI (DESI Collaboration et al. 2016) and 4MOST (de Jonget al. 2012), aim at exquisitely characterizing the large-scale 3Dclustering of galaxies by targeting large samples of pre-selectedsources. Photometric surveys, instead, image large portions ofthe sky with a few broad-band filters and derive cosmological information from both the galaxy angular distribution and clus-ter and lensing studies. Recent and ongoing e ff orts, to just quotea few, are the Dark Energy Survey (DES, Wester & Dark En-ergy Survey Collaboration 2005), the Panoramic Survey Tele-scope and Rapid Response System 1 (Pan-STARRS1, Chamberset al. 2016), or the Hyper Suprime-Cam Subaru Strategic Pro-gram (HSC-SSP, Aihara et al. 2018). The data delivered bythese cosmological surveys o ff er the opportunity to study theevolution of the galaxy population. Spectroscopic data providemore information on individual objects, with the drawback thatthe selected samples su ff er target-selection biases. Photometricdata, instead, su ff er from the lack of information on individualsources, as each object can be characterized by only a few photo-metric points. Photometric surveys can provide a good descrip- Article number, page 1 of 27 a r X i v : . [ a s t r o - ph . GA ] F e b & A proofs: manuscript no. GalEvo_AEGIS_25Feb tion of the objects detected when a large number of bands isused. Notable examples are the Classifying Objects by Medium-Band Observations (COMBO-17, Wolf et al. 2003), the Cos-mological Evolution Survey (COSMOS, Ilbert et al. 2009), orthe Advanced Large Homogeneous Area Medium Band Redshift(ALHAMBRA, Moles et al. 2008; Molino et al. 2014). AL-HAMBRA used medium-band photometry (with a full width athalf-maximum of
FWHM ∼
300 Å) reaching a better precisionin their estimations of photometric redshifts (hereafter photo- z )and opened the possibility to characterize the physical propertiesof individual objects. However, so far, this approach has beenused only to target a few square degrees of the sky, and there-fore, a small volume of the Universe.The Javalambre-Physics of the Accelerating Universe Astro-physical Survey (J-PAS, Benítez et al. 2009; Benitez et al. 2014)has been conceived to overcome these limitations. J-PAS is aboutto start scanning thousands of square degrees of the northern skywith 56 narrow-band filters and the JPCam instrument (Marín-Franch et al. 2017) onboard the 2 . z , with anaccuracy of up to ∆ z = .
003 (1 + z ), to perform multiple cosmo-logical studies, including BAOs measurements (see also Benitezet al. 2014; Bonoli et al. 2020). Besides the potential of J-PASfor cosmology and theoretical physics, this survey is perfectlysuited for galaxy evolution studies. The main reasons are relatedto its unique photometric system, the characteristics of the imag-ing camera, and the large area of the sky that will be observedover the life time of the survey.The J-PAS photometric system covers the full optical spec-tral range, with a narrow-band filter (FWHM ∼
145 Å) ev-ery ∼
100 Å. This is equivalent to low resolution spectroscopy( R ∼ ∆ v ∼ − , ∆ λ ∼
100 Å) for each pixel overa Sloan Digital Sky Survey (SDSS)-like area. It allows a verygood sampling of the spectral energy distribution (SED) of eachsource of the imaged sky. Up to intermediate redshift, J-PASwill be more competitive than other medium-band imaging sur-veys such as ALHAMBRA ( R ∼
20) due to a higher spectralresolution to obtain the stellar population properties of galaxies(Díaz-García et al. 2015, 2019a,b), as well as the huge numberof galaxies that will be observed. The width of the narrow-bandfilters (
FWHM ∼
145 Å) allows to identify star-forming galax-ies and measure emission lines such as H α up to z ≤ .
4, H β and [OIII] λ z ≤ .
8, and [OII] λ z ≤ . z < . α galaxies and quasi stellar objects (QSOs), canbe detected up to high redshift. In fact, J-PAS will increase thecapability of J-PLUS and S-PLUS to reveal the bright end of theL α luminosity function at z = . (cid:48)(cid:48) pixel − for the broad-band filters and0 . (cid:48)(cid:48) pixel − for the narrow-band filters) opens a pixel-by-pixelinvestigation of the spatially-resolved galaxies. This makes J-PAS a competitive integral field unit-like (IFU) survey of nearbygalaxies ( z ≤ . ff ective radii,thus going further than MaNGA for instance.The large volume and area of the survey ( ∼ )will include millions of galaxies for which their SEDs will beobtained. This will allow us to identify and characterize bluegalaxies (BGs), and luminous red galaxies (LRGs), as well asto study the evolution of galaxy stellar populations up to z = ff erent types of galaxies in a widerange of redshift, halo mass, density field, and inter-galacticmedium environment. Groups and galaxy clusters at intermedi-ate redshifts (0 . ≤ z ≤
1) will be easily detected (Maturi etal. 2020 in prep.), and the SEDs of their galaxy members willbe used to determine in an homogeneous way their stellar pop-ulation and emission line properties, according to their environ-mental density. Further, at low-redshift ( z < . Pathfinder camera instead of the JPCam. The present work,based on miniJPAS galaxies at z <
1, has the goal of paving theway to identify and characterize galaxy populations and theirstellar content across cosmic time. In particular, we show thepower of the J-PAS filter system to dissect the bimodal distri-bution of stellar populations of galaxies and its evolution up to z ∼ z < . ff mann et al. 2003a,b; Baldry et al. 2004; Article number, page 2 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field
Brinchmann et al. 2004; Gallazzi et al. 2005; Mateus et al. 2006,2007). These colour distributions also depend on the galaxy stel-lar mass, important factor in the evolution of galaxies (e.g. Pérez-González et al. 2008; Pérez et al. 2013; González Delgado et al.2014), with the red sequence being populated by the most mas-sive galaxies (Hernán-Caballero et al. 2013; Schawinski et al.2014; Díaz-García et al. 2019a,b). The colour bimodality is alsopresent in the colour–stellar mass diagrams, and is tightly cor-related with the ongoing star formation processes (or star for-mation rates, SFRs) and the stellar mass of the galaxies in thesample (Noeske et al. 2007; Speagle et al. 2014; Renzini & Peng2015; Catalán-Torrecilla et al. 2015; González Delgado et al.2016; ? ; Thorne et al. 2020). Despite selection e ff ects and pho-tometric uncertainties, the colour bimodality has been measuredat intermediate redshifts from large-area surveys like BOSS, us-ing Bayesian statistics (Montero-Dorta et al. 2016). The exis-tence of these two groups beyond the nearby Universe is there-fore accepted, with evidence suggesting that they could be al-ready in place at least at z (cid:39) ? ; García-Benito et al.2017; Goddard et al. 2017a,b; Zibetti et al. 2017; García-Benitoet al. 2019; Sánchez et al. 2019a; Sánchez 2020). The spectro-scopic stellar continuum of a galaxy contains many absorptionlines and stellar features whose intensity allows us to constrainits SFH, extinction, and the age and metallicity of the stellarpopulation. Nevertheless, this technique su ff ers from large un-certainties when applied to spectra of low signal-to-noise ratio(SNR) or broad-band photometry. However, intermediate-bandphotometry at optical wavelengths combined with broad-bandNIR data from the ALHAMBRA survey provided successful re-sults to identify star-forming and quiescent / red galaxies (Díaz-García et al. 2019a,b,c). The capability of the J-PAS filter systemto retrieve the stellar population properties of galaxies was firststudied in Benitez et al. (2014), based on both synthetic mag-nitudes obtained from SDSS spectra of nearby galaxies and onmock galaxies, and later by Mejía-Narváez et al. (2017). Themulti-band data of miniJPAS, or J-spectra, brings us the oppor-tunity to investigate the potential of J-PAS to identify and charac-terise galaxy populations since z ∼ ff ective radius, with a precision similarto that obtained with MaNGA data. Here, several methodologiesare explored to prove the consistency of the results in the identifi-cation and characterization of galaxy populations for a completesample of galaxies, and their evolution up to z ∼ ff er-ent SED-fitting codes used to derive the properties of the galaxystellar populations. In Sect. 4 we present the inferred stellar pop-ulation properties of the galaxies in the sample, while the un-certainties in the galaxy properties are discussed in Sect. 5. Weidentify and characterise blue and red galaxies and their evolu-tion up to z ∼ Λ CDM) cosmology in a flat Universe with H = . − and Ω M = .
315 (Planck Collaboration et al. 2018). All the stel-lar masses in this work are quoted in solar mass units (M (cid:12) ) andare scaled according to a universal (Chabrier 2003) initial stellarmass function. All the magnitudes are in the AB-system (Oke &Gunn 1983).
2. Data and sample
This section presents a brief summary of the main instrumentalcharacteristics and observations contained in miniJPAS. A de-tailed and broader description of miniJPAS can be found in theminiJPAS presentation paper (Bonoli et al. 2020).
The miniJPAS survey was carried out at the Observatorio As-trofísico de Javalambre (OAJ, Cenarro et al. 2014) located at thePico del Buitre in the Sierra de Javalambre in Teruel (Spain). Theacquisition of the data was performed with the 2.5m Javalam-bre Survey Telescope (JST / T250) that is optimized to provide agood image quality in the optical spectral range (3300–11000 Å)across the 7 deg of the focal plane.The miniJPAS data was obtained with the JPAS- Pathfinder camera, which is the first scientific instrument installed at theJST / T250 before the arrival of the Javalambre Panoramic Cam-era (JPCam, Taylor et al. 2014; Marín-Franch et al. 2017). TheJPCam has an e ff ective field of view (FoV) of 4 . and it iscomposed of 14 CCDs. In contrast, the JPAS- Pathfinder instru-ment is a single CCD direct imager located at the center of theJST / T250 FoV with a pixel scale of 0 . (cid:48)(cid:48) pixel − , which is vi-gnetted on its periphery, providing an e ff ective FoV of 0 .
27 deg .The filter system of J-PAS contains 54 narrow band (NB) fil-ters ranging from 3780 Å to 9100 Å plus two broader filters atthe blue and red ends of the optical range, centered at 3479 Å(named u JAVA ) and 9316 Å (named J ∼
100 Å, whereas the FWHM ofthe u JAVA band is 495 Å and J z up to z ∼ ff erent epochs. (ii) Delivering low-resolution( R ∼
60) photo-spectra (or J-spectra) that allow us to identify andcharacterize the stellar populations in galaxies up to z ∼
1. (iii)The measurement of emission lines in galaxies and broad emis-sion line features of QSOs and supernovae. A detailed technicaldescription of the J-PAS filter design and characterization can
Article number, page 3 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb be found in Marín-Franch et al. (2012). In addition, miniJPASincludes four SDSS-like broad-band (BB) filters u JPAS , g SDSS , r SDSS and i SDSS . The miniJPAS observations consist on four pointings in the Ex-tended Groth Strip along a stripe aligned 45 ◦ with respect to theNorth at ( α , δ ) = (215 . ◦ , + . ◦ ) amounting to a total areaof ∼ . The depth achieved is fainter than 22 mag (AB) forfilters with λ < ∼
22 mag (AB) for longer wave-lengths.The images were processed by the Data Processing andArchiving Unit group at CEFCA (Cristóbal-Hornillos et al.2014). All the images and catalogues are available throughthe CEFCA web portal , which o ff ers advanced tools for datasearch, visualization, and data query (Civera et al. 2020, inprep.). In this work, we only analyze the photometric data ob-tained with SExtractor (Bertin & Arnouts 1996) in the so-called “dual-mode”. The r SDSS filter was used as the detectionband. For the rest of the filters the r SDSS band is used as a refer-ence to define the aperture and build the source catalogue. Fur-ther details of the observations and the data reduction can befound in Bonoli et al. (2020).
The sample of galaxies analyzed here is extracted from the dual-mode miniJPAS catalogue, selecting according to r SDSS magni-tude, stellarity index, and photo- z . The resulting catalogue con-tains 64293 objects, of which ∼ zz ≤ CLASS_STAR ≤ . SExtractor , as provided by the
CLASS_STAR parameter, tobuild our galaxy sample. Other stellarity indeces, such as the‘stellar-galaxy locus classification’ total_prob_star param-eter (López-Sanjuan et al. 2019; Baqui et al. 2020) are alsolisted in the miniJPAS catalogue and can be used to selectextended sources. We checked that for total_prob_star ≤ . CLASS_STAR ≤ . z values for miniJPAS were estimated using the JPHOTOZ package developed by the photo- z team at CEFCA.This package is a modified version of the L e P hare code (Arnouts& Ilbert 2011) adding a new set of stellar population synthesisgalaxy templates (Hernan-Caballero et al. 2020, in prep.). Asa result, the miniJPAS catalogue contains several estimates ofphoto- z . To select our galaxy sample we use the photo- z valuecorresponding to the mode or the maximum of the probabilitydensity function (PDF).The third condition to select the sample galaxies is related tothe quality of the detection and galaxy brightness. Only objectswith SExtractor flag-mask equal to zero in the r SDSS band wereallowed. This parameter guarantees that our sources are detectedin a well-defined aperture and that our sample does not containsources close to problematic regions of the image such as brightstars or out of the window frame. The miniJPAS dual mode cata-logue includes measurements for di ff erent types of aperture. Weuse the MAG_AUTO aperture as a proxy for the total galaxy mag-nitude. To define a flux-limited sample we impose the condition http://archive.cefca.es/catalogues/miniJPAS-pdr201912/ that the galaxies have to be brighter than 22 . r SDSS ac-cording to the
MAG_AUTO photometry.Bonoli et al. (2020) showed that the catalogue is com-plete up to r SDSS = . r SDSS = . z ac-curacy reaches a precision of 0.8% for most of the sources upto r SDSS = . ≥ z ≤
1. Our final sample contains ∼ z ≤ r SDSS ≤ . CLASS_STAR ≤ . z ∼ . z ∼ . − − z = .
07 in Fig. 1),and even radio-jet structures at higher redshifts (e.g., 2406 − z = .
18 in Fig. 1). Galaxies in groups are also well identified(e.g., 2243 − − z = .
28 and0 .
24, respectively), as well as in less dense environments. The J-spectra of these galaxies (lower panels of Fig. 1) show the powerof J-PAS to identify blue and star-forming galaxies, to detectthe recombination nebular lines H α up to z ∼ . β up to z ∼ λ z ∼ λ z ≤ Fig. 2 compares the distributions of redshift, brightness, and er-ror in the r SDSS band of the final sample with those of the orig-inal sample of miniJPAS galaxies up to z =
1. A large fractionof galaxies in the sample have photo- z between 0 . .
6. Fewgalaxies are selected with photo- z between 0 . . r SDSS . Most of the extended ob-jects at z ≥ . . r SDSS band for the selectedsample is 0 . . ∼
95% of the ∼ ff er from codeto code. Fig. 2 shows that the selected and the whole samplehave similar distributions in redshift, r SDSS band brightness anderror, and, in terms of detection, both data sets have similar SNRdistributions.
AUTO and
PSFCOR photometry
In addition to
MAG_AUTO , the dual catalogue also includes the
MAG_PSFCOR photometry. This photometric method was devel-oped for the analysis of the ALHAMBRA data (Molino et al.2014) and adapted for miniJPAS. For each object it performs cor-rections that take into account the di ff erences in the point spread Article number, page 4 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field m a g r = 16.76z = 0.072470-10291 H 5000 7500 1000018192021 r = 18.82z = 0.122406-3900
H 5000 7500 100001819202122 r = 19.94z = 0.302243-4843
H 5000 7500 100001920212223 r = 20.35z = 0.332241-5981
H5000 7500 10000wavelength [Å]151617181920 m a g r = 16.74z = 0.072470-9821 r = 18.83z = 0.182406-3162 r = 19.39z = 0.282243-12066 r = 20.55z = 0.242241-16643 Fig. 1.
Top panels : A mosaic image of red and blue galaxies.
Bottom panels : Corresponding J-spectra (coloured points) in
MAG_PSFCOR for thegalaxies above. Note the H α emission in the blue galaxies. The star-shaped markers are the broad-band magnitudes ( u JAVA , g SDSS , r SDSS , and i SDSS ). function (PSF) among the di ff erent bands. This process guaran-tees a good determination of the galaxy colour and the shape ofthe continuum, crucial to retrieve the properties of the galaxystellar populations via the full-spectral fitting techniques appliedhere. Instead of the total magnitude that estimates the flux withinan elliptical aperture determined by the KRON_RADIUS (such as
MAG_AUTO ), MAG_PSFCOR is measured for a smaller aperture, i.e.,it does not provide the total flux.In this work we perform the analysis of the sample usingboth the
MAG_PSFCOR and the
MAG_AUTO photometry, which maydi ff er (Fig. 3). The magnitude distributions in the r SDSS
BB fil-ter and in the NB filter with central wavelength at λ NB MAG_PSFCOR by0 .
55 mag because the aperture is typically smaller than for
MAG_AUTO . Flux uncertainties in r SDSS for
MAG_PSFCOR areslightly smaller, by 0 .
03 mag, than for
MAG_AUTO . The sameis true for the NB filters. In particular, the error in flux for the NB .
05 mag for
MAG_PSFCOR , and,in general, by ∼ .
04 mag for the rest of the NB filters, andthe SNR distribution for
MAG_PSFCOR is slightly better than for
MAG_AUTO .As expected, there is a clear correlation between the error inthe magnitude and the magnitude of the object (Fig. 4). Errorsin the r SDSS and NB filters increase exponentially at faint magni-tudes. The slope is steeper for the NB filters than for the BB r SDSS filter. For
MAG_AUTO magnitudes, an r SDSS =
20 (AB) galaxy hasan error < .
05 mag, while the median error for the NB filtersis < . r SDSS =
22 (AB) galaxy, the respective er-rors are < .
15 and < . r SDSS band, the median average SNRis low, SNR ∼ r SDSS = . MAG_PSFCOR are smaller than for
MAG_AUTO but both show simi-lar behaviours. For this reason we perform the spectral fits to theJ-spectra using the
MAG_PSFCOR magnitudes, and we use them
Article number, page 5 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb N SDSS N galaxiesselectedfitted 0.0 0.2 0.4error r SDSS
SDSS r e d s h i f t c o un t s Fig. 2.
From top to bottom left : distribution of r SDSS magnitude and er-rors, and redshift of the galaxies with photo- z ≤ r SDSS ≤ . z ≤ MAG_AUTO photometry.
Bottom rightpanel : density map of the number count of galaxies with z ≤
1, and thedensity contours of the sample selected (dashed lines) and the galaxiesthat were fitted (solid lines).
15 20 25r
SDSS N SDSS N AUTOPSFCOR
15 20 25mag 620601002003004005006000.00 0.25 0.50 0.75 1.00median error0100200300400
Fig. 3.
Distribution of galaxy magnitudes for the r SDSS broad-bandand the narrow-band filter at λ top panels ). The distributionof the magnitude errors in the r SDSS broad-band and the median errorof the narrow-band filters for each galaxy are in the bottom panels .The coloured and black lines correspond to the values obtained by the
MAG_AUTO and
MAG_PSFCOR photometry, respectively. as reference values, despite the fact that the sample was definedaccording to
MAG_AUTO .The magnitude in the r SDSS band shows a clear dependenceon redshift (Fig. 5). Galaxies with r SDSS ≤
20 (
MAG_AUTO ) aretypically found at z < .
5, while fainter galaxies are at anydistance. The SNR is also a clear function of the brightness ofthe galaxy, as indicated by the average median errors of the r SDSS and the NB filters (Fig. 4). The increase of the SNR for
SDSS e rr o r r S D SS AUTO 0.00.20.40.60.81.015.0 17.5 20.0 22.5r
SDSS e rr o r r S D SS PSFCOR 0.00.20.40.60.81.0 15.0 17.5 20.0 22.5r
SDSS m e d i a n e rr o r AUTO 0.00.20.40.60.81.0 r e d s h i f t SDSS m e d i a n e rr o r PSFCOR 0.00.20.40.60.81.0 r e d s h i f t Fig. 4.
Errors of the r SDSS band (left) and narrow-band (right) filtersas a function of the r SDSS magnitude (detection band).
Top and bottompanels are for the
MAG_AUTO and
MAG_PSFCOR apertures, respectively.The colour bar illustrates the redshift of each galaxy.
MAG_PSFCOR with respect to
MAG_AUTO is also clear in Fig. 5, al-though this increase is small. For instance, galaxies with r SDSS ∼
22 (AB) have a SNR of ∼ MAG_PSFCOR apertures, whilefor
MAG_AUTO this value is ∼ r SDSS <
21) at z ∼ .
82 in the miniJPAS catalogue, which proba-bly are red dwarf stars, with not well estimated photo- z . Of thesesources, 14 with CLASS_STAR < . z ∼ .
82 in Fig. 5;8 of these sources have total_prob_star > .
9, as expectedfor red dwarf stars. The fraction of these sources in our sampleis quite small, less than 0 .
2% of the fitted J-spectra, and their in-clusion in the sample will not have any statistical impact on ourresults.
3. Method of analysis
There is a long list of SED-fitting codes developed to fit dataranging from the far ultraviolet to the far infrared (FIR), e.g.,PROSPECT (Robotham et al. 2020), CIGALE (Boquien et al.2019), PROSPECTOR (Leja et al. 2019), BAGPIPES (Carnallet al. 2018), BEAGLE (Chevallard & Charlot 2016) or MAG-PHYS (da Cunha et al. 2008). Their ingredients, input and out-put are di ff erent. They include parametric and / or non-parametricSFHs, diverse stellar initial mass functions (IMFs), a variety ofdust attenuation laws and dust emission models to fit the FIR, aswell as state-of-the-art stellar population templates. In this workwe use our own SED-fitting codes, specifically developed oradapted to fit the galaxy SEDs as traced by the 56 J-PAS bands.The objective is to estimate the SFH of each galaxy in order toderive various properties of the galaxy stellar population. Oneof these codes ( BaySeAGal ) is a parametric code, i.e., the SFHis described by an analytical model. In contrast, the other codes(
MUFFIT , AlStar , and
TGASPEX ) are non-parametric, i.e., theSFH is expressed as an arbitrary superposition of di ff erent sim-ple stellar populations. This section describes the common fea-tures among these codes, as well as their specific aspects, whichmay cause di ff erences in the results. Article number, page 6 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field
16 18 20 22 240.000.250.500.751.00 r e d s h i f t AUTO 05101520 S / N
16 18 20 22 24r
SDSS r e d s h i f t PSFCOR 05101520 S / N Fig. 5.
Magnitudes observed in the r SDSS band as a function of redshiftfor
MAG_AUTO and
MAG_PSFCOR apertures. The colour bar shows themedian signal-to-noise ratio in the narrow-band filters.
Stellar Population models:
In order to minimize discrepanciesin the results due to the methodologies embedded in the codes,the four fitting codes used the latest version of the Bruzual &Charlot (2003) stellar population synthesis models (Plat et al.2019, hereafter C&B). The C&B models follow the PARSECevolutionary tracks (Marigo et al. 2013; Chen et al. 2015) anduse the Miles (Sánchez-Blázquez et al. 2006; Falcón-Barrosoet al. 2011; Prugniel et al. 2011) and IndoUS (Valdes et al. 2004;Sharma et al. 2016) stellar libraries in the spectral range cov-ered by the J-PAS data. These models are available for di ff erentmetallicities ranging from log ( Z (cid:63) / Z (cid:12) ) = − .
23 to 0 .
55 dex andrun in age from ∼ Z (cid:63) / Z (cid:12) ) = − . − . − . − .
33, 0, 0 .
25, and 0 .
55, and all the ages to build the SFHof the parametric models in
BaySeAGal , the ’two burst’ compos-ite stellar population (CSP) models in
MUFFIT , and the ’squareburst’ CSP models in
AlStar and
TGASPEX . Dust attenuation law:
Attenuation by dust is a key ingredi-ent for a proper interpretation of galaxy colours. We assume acommon attenuation law for the four codes. Model spectra areattenuated by a factor formally expressed as e − q λ τ V , where τ V isthe dust attenuation parameter in the V band and q λ ≡ τ λ /τ V denotes the reddening law. For the present work we chose the at-tenuation law proposed by Calzetti et al. (2000), which we addedas a unique foreground screen with a fixed ratio of R V = . Emission lines:
The presence of emission lines from ei-ther young star forming regions or an active galactic nucleus(AGN) component may strongly increase the flux in certain J-PAS bands. Since emission lines are not included in SSP models,the narrow-bands a ff ected by strong emission lines (mainly H α ;H β ; [NII] λ λ λ Maximum age of the stellar population and redshift:
In thenon-parametric codes, the age-span of the simple stellar pop-ulation (SSP) base used to fit a given galaxy SED is trun-cated at t max , the age of the Universe at the galaxy redshift. In BaySeAGal , the look-backtime for the onset of star formation islimited to 0 . × t max . The redshift of each galaxy is fixed to themode of the photo- z value provided in the miniJPAS catalogues(Hernán-Caballero et al. 2020, in preparation). In AlStar , TGASPEX , and
MUFFIT the best-fitting solution is ob-tained by computing the non-negative values of the coe ffi cients x tZ that minimize the merit function χ = (cid:88) λ [ F obs λ − (cid:80) t , Z x tZ f λ, tZ ( τ V )] σ λ , (1)used to measure the goodness-of-fit. In Eq. (1), F obs λ is the ob-served flux in each of the miniJPAS bands, f λ, tZ are the modelphoto-spectra, and σ l is the corresponding uncertainty in eachband. The sum is done over all the filters (index λ ) and mod-els (indices t and Z ). Both AlStar and
TGASPEX use the non-negative least squares (NNLS) algorithm (Lawson & Hanson1974) to find the vector x tZ that minimizes χ , using an outerloop to minimize by the dust attenuation τ V . MUFFIT , on theother hand, uses an algebraic solution to compute the valuesof the coe ffi cients, with the dust attenuation embedded in thepre-computed models. In contrast, BaySeAGal follows a MarkovChain Monte Carlo (MCMC) approach, using the same figure-of-merit function to compute the probability of each step, and τ V is one of the model parameters.Given the nature of the photometric uncertainties and theknown correlations and degeneracies amongst colours and stel-lar population properties, it is essential to perform a statisticalanalysis based on the observed photon-noise and its impact onthe results. MUFFIT , AlStar , and
TGASPEX follow the frequen-tist approach of Monte Carlo-ing the input (by adding Gaussiannoise with observationally defined amplitudes) and repeating thefit many times, assuming that the errors in the di ff erent bandsare uncorrelated. A PDF for each stellar population property isbuilt by weighting the results from each iteration by the likeli-hood (cid:76) ∝ exp( − χ / BaySeAGal the problem is treated in aBayesian way, the PDF for each parameter results naturally fromthe MCMC algorithm.For the four codes the inferred value for each galaxy prop-erty is obtained directly from the corresponding marginalizedPDF. Each property is thus characterized by the mean, the me-dian, and the percentiles defining the confidence interval in thedistribution.
The four codes use the same definitions of the stellar populationproperties to characterize the AEGIS galaxies. A brief descrip-tion follows: – Galaxy stellar mass (M (cid:63) ) : Stellar mass of a galaxy at present.It is calculated from the mass converted into stars accord-ing to the SFH of the galaxy, taking into account the massloss of the SSP owing to stellar evolution. Usually we plotlog M (cid:63) [ M (cid:12) ]. Article number, page 7 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb – Age of the stellar population : We define the mass-weightedlogarithmic age (hereafter mass-weighted age) following CidFernandes et al. (2013, eq. 9) as (cid:104) log age (cid:105) M = (cid:88) t , Z µ tZ × log t , (2)where µ tZ is the fraction of mass of the base element withage t and metallicity Z . Similarly, the light-weighted age isdefined as (cid:104) log age (cid:105) L = (cid:88) t , Z x tZ × log t , (3)where x tZ is the fraction of light at the normalization wave-length (5635 Å) corresponding to the base element with age t and metallicity Z . – Intrinsic and rest-frame ( u − r ) colour : The intrinsic ( u − r )colour is calculated by convolving the resulting fitted syn-thetic spectrum at rest-frame with the u JAVA and r SDSS filtertransmission functions. The rest-frame ( u − r ) is calculatedin a similar manner but including also the reddenning e ff ectsinto the synthetic SED (i.e. e − q λ τ V ). BaySeAGal : specific aspects
BaySeAGal is based on the method developed by ? to fit theGALEX and CALIFA data for a sample of nearby galaxies. Thismethodology is now modified to fit J-spectra (de Amorim et al.2020, in prep.) instead of the GALEX colours and the stellar fea-tures measured in the CALIFA spectra. The code assumes a SFH = SFH( t ; Θ ), where t is the lookback time and Θ is a parametervector, including the stellar metallicity ( Z (cid:63) ), a dust attenuationparameter ( τ V ), and parameters ( k , t , τ ) controlling the temporalbehavior of the SFR, ψ ( t ).The synthetic spectrum for a given ψ and Θ is computed from L λ ( Θ ) = e − q λ τ V (cid:90) SSP λ ( t , Z ) ψ ( t ; Θ ) dt , (4)where SSP λ ( t , Z ) is the spectrum at age t of an SSP of metal-licity Z and initial mass 1 M (cid:12) , and Θ = ( k , t , τ, Z (cid:63) , τ V ). In thisstudy we use the exponential- τ and the delayed- τ SFR laws, fre-quently used in the literarture. For the exponential- τ SFR, ψ ( t ) = ψ e − ( t − t ) /τ , and for the delayed- τ SFR, ψ ( t ) = k ( t − t ) /τ e − ( t − t ) /τ .In both cases, t is the lookback-time, τ is the SFR e-folding timeand t the look-back time for the onset of star formation. ψ and k are normalization constants related to the mass formed in stars.The code explores parameter space and constrains the pa-rameters Θ that fit our data using an MCMC method. The advan-tage of this Bayesian analysis is that we can marginalize over theparameters. The main characteristics of the code are: – Given a set of 56 narrow-band magnitudes, the MCMC algo-rithm draws a set of values of Θ in parameter space accordingto some probabilistic rules. These values are a representationof the PDF over the entire parameter space. – The code determines the corresponding total stellar massformed by maximizing the likelihood of the scale dependentobservables (the photometric magnitudes, in our case). – Using the SFH library the code derives the PDF for eachof the stellar population properties (mass, stellar age, dustattenuation, stellar metallicity, and colours). The PDFs rep-resent the complete solution to the inference problem. Foreach galaxy property we use the median and sigma of thecorresponding PDF as its inferred value and error.
MUFFIT : specific aspects
The code
MUFFIT (MUlti-Filter FITting for stellar populationdiagnostics, extensively detailed in Díaz-García et al. 2015) isa generic SED-fitting tool specifically designed for the detailedanalysis of the SED of stars and galaxies. Namely,
MUFFIT isoptimised to deal with multi-band photometric data and allowsto determine the stellar population properties of the mini-JPASgalaxies: age, metallicity, stellar mass, rest-frame luminosities,extinction, photo- z , etc. To constrain these fundamental param-eters, the code performs an error-weighted χ -test assumingtwo-burst composite models of stellar populations. MUFFIT hasproven to be a reliable and powerful tool with capabilities fortracing the evolution of the stellar content of galaxies in agree-ment with spectroscopic studies at intermediate redshifts (seee.g., San Roman et al. 2018; Díaz-García et al. 2019c), as wellas with mock galaxy samples and SDSS data based on spectro-scopic indices (Díaz-García et al. 2015).
AlStar : specific aspects
Instead of postulating an a priori ψ ( t ; Θ ) function, one can de-scribe galaxy evolution in terms of an arbitrary positive com-bination of N stellar populations of di ff erent age and metallic-ity. This non-parametric approach is the one implemented in the starlight code (Cid Fernandes et al. 2005), employed in numer-ous spectroscopic studies based on SDSS, CALIFA, MaNGA,and other data sets. The advantage with respect to parametriccodes is that the more flexible description of the SFH in generalleads to better spectral fits, particularly in objects which undergomultiple star formation episodes (parametric codes have a hardtime dealing with multi-modal SFHs). The disadvantage is thatthe huge parameter space ( N ∼ starlight , henceforth dubbed AlStar . The code performs the same spectral decomposition interms of a base of N model-spectra, but solving the NNLS prob-lem instead of the MCMC-like sampling scheme implementedin starlight . The non-linear e ff ects of dust attenuation are dealtwith separately, solving the problem for di ff erent values of τ V and finding the best solution. The specific base used in this pa-per comprises populations of 16 ages and 7 metallicities as inWerle et al. (2019), except that here each age corresponds to a‘square burst’ (a period of constant SFR between adjacent agebins). TGASPEX : specific aspects
TGASPEX is a non-parametric spectral fitting code which can beused to estimate the SFH of a stellar population. The possibil-ity of determining the physical properties of galaxies from theJ-PAS narrow band-filters was studied by Mejía-Narváez et al.(2017) for a variety of SFHs using the
TGASPEX and
DynBaS spectral fitting codes described in detail by Magris et al. (2015).These authors compared results obtained from SDSS galaxyspectra and J-PAS synthetic photometry, concluding that the J-PAS filter system yields the same trend in the age–metallicity re-lation as SDSS spectroscopy for typical SNR values. They alsoquantified the biases, correlations and degeneracies a ff ecting theretrieved parameters using mock galaxy samples. As AlStar , TGASPEX derives the SFH by using the NNLS algorithm. Thebest-fit values of the physical properties of the galaxy are deter-mined directly from the best-fit solution, computed for several
Article number, page 8 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field independent values of τ V . For the implementation of TGASPEX used in this paper we use the 9 metallicities available in the C&Bmodels described above. A maximum of 100 time steps rangingin age from 0 . t max are considered for each metallic-ity. The best solution rarely comprises more than 10 spectra ofdi ff erent ages and metallicities.
4. The stellar population properties of the sample
This section presents the general picture of the distributions ofthe stellar population properties obtained with our four SED-fitting codes for the miniJPAS photometry. We explore the di ff er-ences and the impact on the results introduced by the methodolo-gies embedded in BaySeAGal , MUFFIT , AlStar , and
TGASPEX . As mentioned above, SED-fitting solutions reproduce quite wellthe 56 miniJPAS narrow-band magnitudes of galaxies of di ff er-ent types (e.g., ELGs or LRGs) within the uncertainties and in-dependently of the redshift and brightness range (see Fig. 6).Indeed, spectral features of ELGs such as the H α , [OIII] λ λ ff erent bands, suchas the noticeable 4000 Å-break in LRGs, are also well repro-duced (see bottom panels in Fig. 6). The residuals obtained fromthe fits, defined as the di ff erence between the observed and thebest-fitting magnitudes, are higher for fainter galaxies (see rightpanels in Fig. 6) and typically increase towards the bluer bands(see middle panels in Fig. 6). This behavior is related to thehigher uncertainties in the data, which increase at fainter mag-nitudes. It is also worth noting that the uncertainty in the redfilters ( λ ≥ ff ects, a ff ecting more fainter galaxies (e.g., 2243–10498).The J-spectra are well reproduced by the four codes. Thequality of the fits for the whole sample can be assessed throughdi ff erent estimators or parameters such as the reduced χ value( χ ) and the normalized residual. See Appendix A for a dis-cussion of these parameters as derived by BaySeAGal . The qual-ity assessment of the fits for the four codes is very similar toFig. A.1, yielding equivalent conclusions about the fitting errors.
BaySeAGal
In this section, we present the results of the stellar populationproperties obtained by
BaySeAGal using the
MAG_PSFCOR and
MAG_AUTO photometry.
The grey histograms in Fig. 7 show the distributions of mass,age, extinction, rest-frame colour, and metallicity (log M (cid:63) , (cid:104) log age (cid:105) M and (cid:104) log age (cid:105) L , A V , ( u − r ) res , and (cid:104) log Z (cid:105) M , respec-tively) obtained with BaySeAGal for the galaxies in the AEGISfield. The distribution of stellar mass ranges from 8 to ∼
12 dex,with a plateau from 9 . ∼ .
6. Extinction is distributed be-tween 0 and 2 mag, peaking at ∼ . (cid:104) log Z (cid:105) M coversmost of the metallicity range of the SSPs, but the distributionpeaks at around solar and super-solar metallicity. The colour( u − r ) res shows the well-known bimodal distribution of galax-ies, where the maximum density for the blue cloud galaxies is at( u − r ) res ∼ .
5, and at ( u − r ) res ∼ . (cid:104) log age (cid:105) L with peaks at (cid:104) log age (cid:105) L [yr] ∼ . ∼ .
5, while the wholedistribution ranges from (cid:104) log age (cid:105) L [yr] 8 to 10. (cid:104) log age (cid:105) M isshifted towrds older ages with respect to (cid:104) log age (cid:105) L . This is ex-pected because small amounts of mass in recent star-formationepisodes translate into a high contribution to the luminosity ofthe galaxy, decreasing the light-weighted age. The distributionin (cid:104) log age (cid:105) M [yr] peaks at ∼ . (cid:104) log age (cid:105) M [yr] ∼ . u − r ) res distribution clearly showsthat the fraction of blue galaxies detected with high SNR is lowerthan the fraction of red galaxies. Similarly, the fraction of mas-sive galaxies with high SNR is higher than for the less massivegalaxies (that populate mainly the blue cloud), and the fractionof old galaxies with high SNR is also higher than for the youngergalaxies. In summary, the red, old, and massive galaxies are de-tected at higher SNR than the blue, young, and less massive ones. MAG_AUTO vs.
MAG_PSFCOR
We fitted the J-spectra using both
MAG_PSFCOR and
MAG_AUTO magnitudes. In this way we can assess the impact of the pho-tometric choice on the stellar population properties of the mini-JPAS galaxies. As mentioned in Sect.2.5, the flux is integratedthrough a larger aperture in
MAG_AUTO than in
MAG_PSFCOR , theformer being closer to the total flux. Then, the fitting to the J-spectra performed using
MAG_AUTO should provide a more repre-sentative estimate of the integrated stellar population propertiesthan the ones obtained using
MAG_PSFCOR .From Fig. 8 we see that the main di ff erences appear in thelog M (cid:63) and ( u − r ) res distributions. MAG_AUTO results are shiftedto larger masses by 0 . σ = . − .
09 ( σ = . MAG_AUTO with respect to
MAG_PSFCOR . Thesmall shift to younger ages by (cid:104) log age (cid:105) L ∼ − .
07 ( σ = . (cid:104) log age (cid:105) M is not a ff ected by thelarger aperture because this property is biased towards the olderstellar populations, which are concentrated in the inner regionsof the galaxies. A V and (cid:104) log Z (cid:63) (cid:105) M are not a ff ected by the aper-ture, although the small decrease of (cid:104) log Z (cid:63) (cid:105) M is possibly dueto the negative gradient of log Z (cid:63) in galaxies (González Delgadoet al. 2015). MUFFIT , AlStar , and
TGASPEX
Next, the J-spectra (in
MAG_PSFCOR ) were fitted using the non-parametric codes
MUFFIT , AlStar , and
TGASPEX describedin Sects. 3.2.2–3.2.4. The main di ff erences with respect to BaySeAGal are: (i) these three codes use combinations of SSPsinstead of a parametric ψ ( t ). (ii) BaySeAGal builds the full PDFof the properties via a MCMC process, whereas the other codesuse a Monte Carlo approach. These codes provide a characteriza-tion of the stellar populations in the miniJPAS galaxies in agree-ment with the
BaySeAGal results. The figures of merit for thethree codes are comparable to Fig. A.1. The χ distributions in-dicate fits similar in quality to the BaySeAGal ones.In Fig. 9 we compare the distributions of galaxy propertiesobtained with the four codes. The distributions of ( u − r ) res areremarkably similar, indicating that the four codes reach similarquality fits. The distributions of log M (cid:63) are quite compatible, al-though AlStar and
TGASPEX show a mild excess of galaxieswith stellar mass 10 ≤ log M (cid:63) <
11 [M (cid:12) ] with respect to
MUFFIT
Article number, page 9 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb m a g r = 19.14z = 0.202243-10648 H[OIII][OII] r = 20.35z = 0.332241-5981 H[OIII][OII] r = 22.37z = 0.612243-10498 [OIII][OII] m a g r = 18.83z = 0.182406-3162 r = 19.49z = 0.292470-1010 r = 21.70z = 0.772243-10643 Fig. 6.
MAG_PSFCOR
J-spectra of di ff erent galaxies with redshift z = . .
77 and brightness 19 . < r SDSS < . MAG_AUTO in r SDSS ). Maskedfilters (white coloured circles) and filters overlapping with H α , [NII], [OIII], H β , or [OII] lines (grey coloured circles) are not used in the fit. Thebest model fitted to the continuum with BaySeAGal is plotted as black points, and the grey band shows the magnitudes of the mean model ± one σ uncertainty level. The di ff erence between the observed and the best model fitted magnitudes are plotted as a small coloured points around theblack bottom line, that represents a null di ff erence between the observed and the best fitted model. The grey band around these dots shows thedi ff erence ± one sigma variation. and BaySeAGal . On average, the
AlStar and
TGASPEX stel-lar masses are 0 .
06 and 0 .
11 dex larger, respectively, than the
BaySeAGal values.
MUFFIT stellar masses are 0 .
05 dex lowerthan
BaySeAGal ’s. The distributions of (cid:104) log Z (cid:63) (cid:105) M are very simi-lar for the four codes, although MUFFIT predicts a larger fractionof galaxies with super-solar metallicity. It is worth mentioningthat the
BaySeAGal
PDF is more extended at the low-metallicityend than for the other codes. On average,
MUFFIT , AlStar , and
TGASPEX galaxies are 0 .
24, 0 .
1, and 0 . BaySeAGal ’s. However, these di ff erences are small consider-ing the coarse distribution of metallicity in the input models.The distributions of A V from BaySeAGal , MUFFIT , and
AlStar are similar, although the
BaySeAGal distribution shows a flat-ter slope and a cuto ff at A V ≤ .
3. The distribution of A V from TGASPEX is significantly di ff erent from the other three codes.On average, A V resultes from MUFFIT , AlStar , and
TGASPEX ,are larger by 0.35, 0.2, and 0.1, respectively, than
BaySeAGal ’s.We find that there are more discrepancies in the distributions ofage than for the other properties. On the one hand, the distribu-tions of (cid:104) log age (cid:105) L and (cid:104) log age (cid:105) M obtained by BaySeAGal and
MUFFIT are almost identical, showing a double-peaked distribu-tion. On the other hand, the
AlStar and
TGASPEX (cid:104) log age (cid:105) L dis-tributions show one maximum, shifted towards older ages withrespect to the BaySeAGal and
MUFFIT results. The (cid:104) log age (cid:105) M results from AlStar and
TGASPEX are typically 0 .
15 dex and0 . ff er- ences in the mean value and dispersion of the galaxy proper-ties measured with the non-parametric codes and BaySeAGal arelisted in Table 1.
The ( u − r ) res colour of miniJPAS galaxies shows a bimodaldistribution indicating the presence of two galaxy populations,which we refer to as red and blue galaxies. One of the diagramsused more frequently to study the di ff erences between these twopopulations is the galaxy stellar mass–colour diagram (see e.g.,Moresco et al. 2013; Schawinski et al. 2014; Díaz-García et al.2019a, and references therein). If we focus on the distribution ofgalaxies in the log M (cid:63) –( u − r ) res diagram (Fig. 10), the blue cloudand red sequence of galaxies are clearly identified. Galaxies thatbelong to the red sequence are typically old and metal rich. Re-garding age, we find that typically (cid:104) log age (cid:105) M [yr] is above 9 . (cid:104) log age (cid:105) M [yr] down to 9. Most of the galaxies in the red se-quence have solar and super-solar (cid:104) log Z (cid:63) (cid:105) M metallicity, whilefor the blue cloud it is below solar values. Overall, the extinc-tion is similar in the blue cloud and in the red sequence, A V < A V values are located at di ff erent positions in thediagram depending of the code. For example, BaySeAGal galax-ies with A V > . Article number, page 10 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field N res N M [yr]0250500750100012501500 8 9 10
Fig. 7.
Stellar population properties from the full spectral fitting of the J-spectra with
MAG_PSFCOR obtained with
BaySeAGal . From left to right,and from top to bottom : stellar mass, mean mass-weighted age, intrinsic extinction, rest frame ( u − r ) colour, mean light-weighted age (at 5630 Å),and stellar metallicity. The di ff erent coloured histograms correspond to di ff erent values of the median signal-to-noise ratio in the narrow-bandfilters. Table 1.
Mean and standard deviation of the di ff erence between the value of each galaxy property derived with MUFFIT , AlStar , and
TGASPEX with respect to
BaySeAGal and the delayed- τ model. SP MUFFIT AlStar TGASPEX ( u − r ) res . ± .
11 0 . ± .
16 0 . ± . M (cid:63) − . ± .
14 0 . ± .
27 0 . ± . A V − . ± . − . ± . − . ± . (cid:104) log Z (cid:63) (cid:105) M . ± .
52 0 . ± .
50 0 . ± . (cid:104) log age (cid:105) M . ± .
21 0 . ± .
30 0 . ± . (cid:104) log age (cid:105) L − . ± .
26 0 . ± . − . ± . TGASPEX are located inthe right-ridge of the histogram, where the high redshift galax-ies of the sample are located, which have larger uncertainties inthe observed magnitudes. These di ff erences in the properties ofthe galaxies among the four fitting codes illustrate the degener-acy between the outputs and the dependence of the results on theuncertainties in the observed magnitudes.
5. Uncertainties in the stellar population properties
One of our aims is to explore how the uncertainties in the ob-served magnitudes and redshift a ff ect the locus of galaxies in the( u − r ) res diagram, as well as their spectral classification as red or blue galaxies (see Fig. 11). Faint galaxies are mainly locatedin the right ridge of the log M (cid:63) – ( u − r ) res diagram; this reflectsthe position of the galaxies with higher photo- z , which are alsofainter. The precision in the estimation of the photo- z also a ff ectsthe fits and the values of log M (cid:63) and ( u − r ) res . If we measurethe precision in photo- z by the relative error, σ z / (1 + z ) × z = photo- z ), then galaxies in the red sequence exhibitlower photo- z uncertainty than galaxies in the blue cloud. The4000 Å break, more prominent in LRGs than in blue galaxies,controls this behaviour. This stellar continuum feature helps toobtain more precise values of photo- z , and also of the propertiesof the galaxy, since it is correlated with the age of the stellar pop-ulation (González Delgado et al. 2005). Nonetheless, the identi-fication of red and blue galaxies in the log M (cid:63) – ( u − r ) res diagramdoes not show a strong limitation according to the quality of theJ-spectra. Galaxies with similar SNR, as traced by the median er-ror in the narrow-band magnitudes, are equally distributed acrossthe red sequence and blue cloud, but most of the galaxies with Article number, page 11 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb N = -0.20 = 0.17 res N = 0.09 =0.17 PSFCORAuto M (yr)0250500750100012501500 = 0.03 =0.14 L (yr)0500100015002000 = 0.07 =0.19 V = 0.02 =0.32 M [Z ]0200400600 = -0.01 =0.57 Fig. 8.
Comparison of the distributions of the stellar population properties derived with
BaySeAGal by spectral fitting using
MAG_PSFCOR (black)and
MAG_AUTO (orange) magnitudes for stellar mass (present and initial), rest-frame intrinsic ( u − r ) colour, mass and light weighted ages, stellarextinction, and metallicity. In each panel, µ and σ indicate the mean and standard deviation of the di ff erence between the value of each galaxyproperty derived with MAG_PSFCOR and
MAG_AUTO . low SNR fall near the border of the area filled by the bulk of thegalaxies in the log M (cid:63) – ( u − r ) res diagram in Fig. 11. With
BaySeAGal we can evaluate easily the uncertainties onthe galaxy SFH and the inferred stellar population proper-ties thanks to the MCMC process and the use of a paramet-ric SFR. For a given ψ ( t ) we obtain the PDF and from it themean and the standard deviation ( σ ) for each galaxy prop-erty ( t , τ , A V , log Z , log M (cid:63) ). Overall, ( u − r ) res and log M (cid:63) arethe properties with lower uncertainties (see upper left panel inFig. 12) with (cid:104) σ (( u − r ) res ) (cid:105) = . ± .
05 mag and (cid:104) σ (log M (cid:63) ) (cid:105) = . ± .
06 dex. The property with the largest uncertainty is (cid:104) log Z (cid:63) (cid:105) M , with (cid:104) σ ( (cid:104) log Z (cid:63) (cid:105) M ) (cid:105) = . ± .
28 dex. This is ex-pected from the coarse sampling of metallicity in the SSP mod-els. The stellar extinction A V is significantly better estimatedthan (cid:104) log Z (cid:63) (cid:105) M , with (cid:104) σ ( A V ) (cid:105) = . ± . (cid:104) log age (cid:105) M is estimated better than A V , andglobally worse than ( u − r ) res . σ ( (cid:104) log age (cid:105) M ) presents a bimodaldistribution with peaks at ∼ . ∼ .
15. These peaks are re-lated to the bimodal distribution of the galaxy population. Theage for the blue galaxies is less constrained than for the redgalaxies, with (cid:104) σ ( (cid:104) log age (cid:105) M ) (cid:105) = . ± .
06 dex and 0 . ± . u − r ) res smaller or larger than 1 . ∼
8. In particular, if we consider a sub-sampleof blue galaxies with SNR ≥
10, which is the mean SNR ofspectroscopic samples in future surveys like WEAVE / STePs (Costantin et al. 2019), the precision of the inferred galaxyproperties improves considerably. For this sub-sample of mini-JPAS blue cloud galaxies, the stellar metallicity is estimatedwith (cid:104) σ ( (cid:104) log Z (cid:63) (cid:105) M ) (cid:105) = . ± .
25 dex, and (cid:104) σ ( (cid:104) log age (cid:105) M ) (cid:105) = . ± .
07 dex. Higher precision is obtained for SNR ≥
20 – 30,approaching IFU surveys of nearby galaxies, e.g., CALIFA (CidFernandes et al. 2014; González Delgado et al. 2015). See Table2. Exploring the distribution of the uncertainty across thelog M (cid:63) – ( u − r ) res diagram (see also Fig. 12) we find that theseproperties are not estimated with equal precision for the di ff er-ent galaxy populations, so the precision is not evenly distributedacross the log M (cid:63) – ( u − r ) res diagram. The bimodality of the dis-tribution is clearly seen in σ ( (cid:104) log age (cid:105) M ) (bottom left panel inFig. 12). For red sequence galaxies (cid:104) log age (cid:105) M is estimated bet-ter ( σ ( (cid:104) log age (cid:105) M ) ≤ .
15 dex) than for galaxies in the bluecloud ( σ ( (cid:104) log age (cid:105) M ) ∼ . M (cid:63) (topcentral panel in Fig. 12) are also smaller in the red sequence thanin the blue cloud, although the larger uncertainties in log M (cid:63) cor-respond to galaxies in the right ridge of the log M (cid:63) – ( u − r ) res diagram, as discussed above for Fig. 10. In addition, ( u − r ) res isconstrained slightly better in the red sequence than in the bluecloud (top right panel in Fig. 10). The uncertainties in A V (bot-tom central panel in Fig. 10) are similar throughout the bluecloud, somewhat lower for the less massive galaxies, while thehigher σ ( A V ) correspond to the more massive galaxies in thered sequence. The precision in (cid:104) log Z (cid:63) (cid:105) M (bottom right panel inFig. 10) shows a clear distribution in the log M (cid:63) – ( u − r ) res dia-gram; it is worse constrained for galaxies in the blue cloud, with Article number, page 12 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field N res N M [yr]0500100015002000 7 8 9 10
Fig. 9.
Distributions of mean values of stellar population properties obtained by our SED-fitting codes
BaySeAGal , MUFFIT , AlStar , and
TGASPEX for the miniJPAS galaxies (black, red, blue, and green lines, respectively). The grey curves also illustrate the posterior probability distributionfunctions (PDFs) obtained by
BaySeAGal assuming a delayed- τ star formation history (SFH). All the results were obtained using the PSFCOR photometry. ( u - r ) r e s BaySeAGal < l o g a g e > M A V < l o g Z > M [ Z ] ( u - r ) r e s MUFFIT < l o g a g e > M A V < l o g Z > M [ Z ] ( u - r ) r e s AlStar < l o g a g e > M A V < l o g Z > M [ Z ] ( u - r ) r e s TGASPEX < l o g a g e > M A V < l o g Z > M [ Z ] Fig. 10.
Mass–colour relation with the coloured bar showing the stellar population properties of the galaxies (age, extinction, and metallicity).Properties were derived by using the
PSFCOR photometry.
From top to bottom : results from
BaySeAGal , MUFFIT , AlStar , and
TGASPEX .Article number, page 13 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb ( u - r ) r e s e rr _ r z e rr / ( + z ) * m e d i a n e rr o r Fig. 11.
Mass–colour relation derived by
BaySeAGal and the
PSFCOR photometry with the coloured bar showing the error in the magnitude in the r SDSS band, error in redshift, and the median SNR in the narrow band filters. ( u - r ) r e s s t d < l o g a g e > M s t d A V s t d < l o g Z > M ( u - r ) r e s s t d l o g M a ss s t d ( u - r ) r e s N log Mass(u-r) res
Distribution of standard deviations of galaxy properties ( top-left panel ) and colour–mass plane coloured by the standard deviation of thestellar populations properties: log M (cid:63) , ( u − r ) res , (cid:104) log age (cid:105) M , A V , and (cid:104) log Z (cid:63) (cid:105) M , respectively. σ ( (cid:104) log Z (cid:63) (cid:105) M ) up to 1 dex. However, for galaxies in the red se-quence, σ ( (cid:104) log Z (cid:63) (cid:105) M ) can be as low as 0 . Here we explore the influence of ψ ( t ) on the inferred stel-lar population properties using BaySeAGal and the two SFRlaws defined in Sect. 3.2.1. The results in Fig. 13 indicate thatthe PDF for log M (cid:63) , ( u − r ) res and (cid:104) log Z (cid:63) (cid:105) M are very simi-lar for the two cases. The exponential- τ SFR produces valuesof A V lower on average by 0 . ± .
65 magnitudes than thedelayed- τ SFR. The peaks of the (cid:104) log age (cid:105) M and (cid:104) log age (cid:105) L PDFs for the exponential- τ SFR are shifted to older ages withrespect to the delayed- τ SFR by (cid:104) log age (cid:105) M = . ± .
09 dexand (cid:104) log age (cid:105) L = . ± .
12 dex. This aging of the galaxies isrelated to the decrease of A V in the delayed- τ SFR model with re-spect to the exponential- τ SFR model. This degeneracy betweenage and extinction is well known in stellar population synthesis(e.g. Cid Fernandes et al. 2005).Table 3 lists the average di ff erences between the propertiesfrom the four codes and the exponential- τ SFR model. Figure 13compares the distributions of the properties from
BaySeAGal with the results from the exponential and delayed- τ models plusthe results for the non-parametric codes. Notice that the distri-butions of A V and (cid:104) log age (cid:105) M from the exponential- τ SFR arecloser to the results from
AlStar and
TGASPEX than the re-
Article number, page 14 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field
Table 2.
Precision (mean standard deviation) of the stellar population properties of miniJPAS sub-sample with SNR ≥
30, 20, 10, 5, and 3. Theunits of (cid:104) σ (( u − r ) res ) (cid:105) , and (cid:104) σ ( A V ) (cid:105) are in magnitudes, and (cid:104) σ (log M (cid:63) ) (cid:105) , (cid:104) σ ( (cid:104) log age (cid:105) M ) (cid:105) ( (cid:104) σ ( (cid:104) log age (cid:105) L )) , (cid:105) and (cid:104) σ ( (cid:104) log Z (cid:63) (cid:105) M ) (cid:105) are in dex of[M (cid:12) ], [yr], and [Z (cid:12) ]. The number of galaxies included in each sub-sample are indicated. SP SNR ≥
30 SNR ≥
20 SNR ≥
10 SNR ≥ ≥ (cid:104) σ (( u − r ) res ) (cid:105) . ± .
02 0 . ± .
02 0 . ± .
02 0 . ± .
03 0 . ± . (cid:104) σ (log M (cid:63) ) (cid:105) . ± .
03 0 . ± .
03 0 . ± .
03 0 . ± .
03 0 . ± . (cid:104) σ ( A V ) (cid:105) . ± .
05 0 . ± .
06 0 . ± .
09 0 . ± .
12 0 . ± . (cid:104) σ ( (cid:104) log age (cid:105) M ) (cid:105) . ± .
07 0 . ± .
07 0 . ± .
07 0 . ± .
08 0 . ± . (cid:104) σ ( (cid:104) log age (cid:105) L ) (cid:105) . ± .
07 0 . ± .
07 0 . ± .
08 0 . ± .
08 0 . ± . (cid:104) σ ( (cid:104) log Z (cid:63) (cid:105) M ) (cid:105) . ± .
20 0 . ± .
22 0 . ± .
25 0 . ± .
28 0 . ± . N res N M (yr)0500100015002000 8 9 10
Fig. 13.
Distributions of stellar population properties obtained by
BaySeAGal for the miniJPAS galaxies using the
PSFCOR photometry andassuming di ff erent star formation laws: delayed (black) and exponential (violet) star formation laws. Results obtained with MUFFIT and
AlStar are also shown (see inset). sults from the delayed- τ SFR model. On average, the di ff er-ence between galaxy properties inferred with AlStar and theexponential- τ SFR model are − . ± . . ± . A V and (cid:104) log age (cid:105) M , respectively. Thus, we conclude thatthe results from the non-parametric codes, in particular AlStar and
MUFFIT , are represented better by an exponential- τ than adelayed- τ SFR. However, it is known that exponential SFR mod-els produce an early and rapid growth of the galaxy stellar massthat is not compatible with the peak in the cosmic density of starformation at z ∼ ? ).
6. Discussion
The photo- z precision makes J-PAS an excellent survey to studythe properties of galaxies across cosmic time. The miniJPAS sur-vey comprises only 1 deg of the sky, and the number of galaxiesper redshift bin is somehow limited. Nevertheless, it allows us to explore how galaxy stellar properties evolve with cosmic time.Further, it allows us to identify and characterize blue and redgalaxies, and their cosmic evolution.In this work we do not carry out a comprehensive and com-plete study of the evolution and formation of galaxies with red-shift. Rather, we roughly explore the global properties of thegalaxies available in the miniJPAS survey; and we also comparethe results from di ff erent SED-fitting codes. For a proper inter-pretation of the results, we should take into account that flux-limited samples are a ff ected by the so-called Malmquist bias,meaning that galaxies below a certain luminosity limit are notobserved at a given redshift. In this regard, less massive galax-ies are typically a ff ected by this bias and it is convenient to de-fine a stellar mass completeness limit for a more detailed andcomprehensive study. The definition of these limits, as well asother aspects and properties such as comoving number densitiesas a function of mass and redshift or stellar mass and luminos- Article number, page 15 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb
Table 3.
Mean and standard deviation of the di ff erence between the value of each galaxy property derived with MUFFIT , AlStar , and
TGASPEX with respect to
BaySeAGal exponential model. Last column shows the di ff erences between the results with BaySeAGal and a delayed and theexponential models. SP MUFFIT AlStar TGASPEX BaySeAGal ( u − r ) res − . ± .
10 0 . ± .
16 0 . ± . − . ± . M (cid:63) − . ± .
13 0 . ± .
26 0 . ± . − . ± . A V − . ± . − . ± .
50 0 . ± .
65 0 . ± . (cid:104) log Z (cid:63) (cid:105) M . ± .
56 0 . ± .
59 0 . ± .
79 0 . ± . (cid:104) log age (cid:105) M − . ± .
22 0 . ± . − . ± . − . ± . (cid:104) log age (cid:105) L − . ± . − . ± . − . ± . − . ± . ff ected by this e ff ect ow-ing to a higher mass–luminosity relation. For instance, quiescentgalaxies of log M (cid:63) ∼ . z < .
35, whereas star-forming galaxies at this mass range canbe easily imaged at z > . The bimodal distribution of galaxies is also reflected on the stel-lar mass–age diagram (log M (cid:63) – (cid:104) log age (cid:105) M ). Figure 14 (onlyshowing the results from BaySeAGal with the delayed- τ model)shows the bimodal distribution of the galaxy populations inAEGIS. Old galaxies are typically redder and more metal richthan young galaxies. Galaxies in the green valley have larger A V than both red and blue galaxies. The log M (cid:63) – (cid:104) log age (cid:105) M diagram also shows a bimodal correlation between the mass andthe stellar population age. Late-type galaxies, which mainly pop-ulate the blue cloud at the nearby universe, show a linear cor-relation with the stellar mass, which is steeper than for early-type galaxies or red sequence galaxies (Kau ff mann et al. 2003a;González Delgado et al. 2014).These results are revealed better when only galaxies of sim-ilar cosmic epoch are plotted (lower panels in Fig. 14). For thisreason, we split our sample in three di ff erent redshift bins, wherethe width of each subset was defined to include a su ffi cient num-ber of galaxies: z ≤ .
35, 0 . < z ≤ .
6, and 0 . < z ≤
1. Atlow redshift, the mass–age relation shows a remarkable changein its slope for galaxies more massive than ∼ . ff -mann et al. 2003a; González Delgado et al. 2014), for whichthe relation log M (cid:63) – (cid:104) log age (cid:105) M is almost flat. This mass limitmarks the green valley, the transition from the blue cloud to thered sequence. However, red galaxies in the AEGIS log M (cid:63) – (cid:104) log age (cid:105) M diagram show a decrease of (cid:104) log age (cid:105) M at decreasinglog M (cid:63) , pointing out that less massive galaxies were assembledin more recent cosmic epochs. This behaviour is also reproducedat any redshift z ∼ Here we explore how the bimodal distribution of galaxy popula-tions changes with cosmic time in the mass–colour diagram. Wediscuss the di ff erences in the distribution of galaxy populationswhen the dust-corrected colour (refered to as intrinsic colour)( u − r ) int is considered instead of the ( u − r ) res . Certainly, dust ex- tinction reddens galaxies and to correct colours for extinction itis relevant to get the demographics of galaxies in the blue cloudand red sequence (e.g. Hernán-Caballero et al. 2013; Schawinskiet al. 2014; Díaz-García et al. 2019a).The mass–( u − r ) int diagram shows that the distribution ofgalaxies across the mass-colour diagram is remarkably di ff er-ent after correcting colours for extinction e ff ects. This fact isdue to the presence of a non-negligible fraction of dusty star-forming galaxies populating the so-called green valley. Many ofthe galaxies move to bluer colours after correcting ( u − r ) res forextinction, and the fraction of galaxies that de-populate the redsequence is larger, increasing towards higher redshifts. A signif-icant fraction of the galaxies in the green valley (30–65% Díaz-García et al. 2019a) are actually obscured star-forming galax-ies, whose fraction depends on redshift and stellar mass. In thenearby Universe ( z ≤ . BaySeAGal and thedelayed- τ models. A similar diagram is derived with MUFFIT .However, some di ff erences are found with respect to the resultsfrom AlStar and
TGASPEX , which impact the estimation of redand blue galaxies of the sample. In the Appendix, we discussthe similarities and discrepancies of the distributions of log M (cid:63) ,and ( u − r ) int amongst the results from the four codes. In partic-ular, we focus on the distributions of log M (cid:63) , and ( u − r ) int , atthe redshift bins of z ≤ .
35, 0 . < z ≤ .
6, and 0 . < z ≤ Díaz-García et al. (2019a) developed a method to discern be-tween red and blue galaxies using a sample of galaxies fromthe ALHAMBRA survey at redshift 0 . < z < .
1. The au-thors also used
MUFFIT to perform an SED-fitting analysis ofthe 20 intermediate band filters covering the optical range andthe J , H , and K s NIR bands. They obtained the stellar mass, rest-frame colour, and extinction of each galaxy from ALHAMBRAand subsequently performed a classification to identify quies-cent and star-forming galaxies. They determined the fraction ofdusty star-forming galaxies in the green valley through the in-trinsic ( m F − m F ) and ( m F − J ) colours, as well as thecontamination in quiescent galaxy samples defined via classi-cal colour–colour diagrams owing to this obscured star-forminggalaxies. As a result, Díaz-García et al. (2019a) concluded thatthe log M (cid:63) − ( u − r ) int diagram can reduce the contamination ofthe red sample by a fraction of 20% with respect to previouscolour–colour diagrams, also without any bias at the low stellarmass regime. Article number, page 16 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field < l o g a g e > M ( u - r ) r e s A V < l o g Z > M [ Z ] < l o g a g e > M z z z ( u - r ) r e s Fig. 14.
Distributions of ( u − r ) res , A V , and (cid:104) log Z (cid:63) (cid:105) M across the mass–age relation. The values of all the parameters are coded according to thecolour bars. Bottom panels illustrate the points and contour distribution of ( u − r ) res in the mass–age relation at z ≤ .
35, 0 . < z ≤ .
6, 0 . < z ≤ z ≤ ( u - r ) r e s BaySeAGal < l o g a g e > M [ y r ] ( u - r ) i n t
0 < z z z < l o g a g e > M [ y r ] Fig. 15.
Rest-frame ( top panels ) and intrinsic ( bottom panels ) colour ( u − r ) vs. stellar mass for the redshift bins z ≤ .
35, 0 . < z ≤ .
6, and0 . < z ≤ from left to right ) for the results determined by BaySeAGal and the delayed- τ model. The dashed line shows the ( u − r ) limint for themean redshift in each bin (details in the text). Article number, page 17 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb (u-r) res N BaySeAGalDelayed 0 1 2 3 (u-r) res (u-r) res (u-r) res (u-r) res A V N A V A V A V A V log Mass [M ] N log Mass [M ] log Mass [M ] log Mass [M ] log Mass [M ] Fig. 16.
Distributions of colours, ( u − r ) res , (upper panels), extinction (middle panels), and log M (cid:63) , of red (red lines) and blue (blue lines) galaxiesof the AEGIS sample identified by BaySeAGal with delayed- τ or exponential models, MUFFIT , AlStar , and
TGASPEX (from left to right columns).Filled-blue histograms are the distributions of blue galaxies with ( u − r ) res > . < ( u − r ) res < Here, we follow a similar method to that in Díaz-García et al.(2019a) and classify galaxies as red or blue (quiescent or star-forming) according to their intrinsic colour, stellar mass, andredshift. The limiting intrinsic colour originally computed byDíaz-García et al. (2019a, see eq. 3) was adapted to match theminiJPAS photometric system. As a result, miniJPAS galaxiesare labelled as quiescent when they exhibit intrinsic colours red-der than the limiting value of ( u − r ) limint , which is formally ex-pressed as,( u − r ) limint = . × (log M (cid:63) − . ) − . × ( z − . + . , (5)(eq. 3 of Díaz-García et al. 2019a), where z is the photo- z ofthe galaxy and log M (cid:63) its stellar mass. Otherwise, the galaxyis labelled as star-forming. After a visual inspection, the colourlimit set by Eq. (5) clearly separates the blue cloud and the redsequence at any redshift (see the dashed line in Fig. 15).With this classification, we can estimate the fraction of redand blue galaxies in the AEGIS field that are brighter than r SDSS = .
5, and z ≤
1. We find a 85% (
BaySeAGal ), 81% (
MUFFIT ),69% (
AlStar ), and 76% (
TGASPEX ) fraction of galaxies in thesample that are in the blue cloud. In general, red galaxies haveintrinsic red colours and lower extinction values than blue galax-ies (see Fig. 16). There is, however, a fraction of galaxies withintrinsic blue colours but reddened by dust. For example, thereis a 13% (
BaySeAGal ), 11% (
MUFFIT ), 8% (
AlStar ), and 9%(
TGASPEX ) fraction of galaxies with ( u − r ) res > u − r ) int below to the ( u − r ) limint . The four codes identify these galaxiesas dusty galaxies and as the most massive ones of the wholesample (Fig. 17). Notice that the di ff erences between codes inthe fraction of galaxies in the blue cloud cannot be only due to di ff erences in the estimation of blue galaxies with red colours( u − r ) res >
2. While the maximum di ff erence in the fraction ofblue galaxies between the results from the codes is ∼
16% for
BaySeAGal and
AlStar , the di ff erence in the fraction of bluegalaxies with ( u − r ) res > ∼
4% between these twocodes. However, di ff erences between codes in the estimations of A V of galaxies with intermediate colour 1 . < ( u − r ) res < ∼
1% (
BaySeAGal ), 3% (
MUFFIT ), 9% (
AlStar ) and 4%(
TGASPEX ) of galaxies with ( u − r ) int close to the ( u − r ) limint thatare identified as red galaxies. They have lower extinctions thanmost of the red but intrinsically blue galaxies and they are lessmassive than the reddest blue galaxies (filled red and blue his-tograms in the middle panels of Fig. 16). The mean extinctionof these 9% of AlStar red galaxies with 1 . < ( u − r ) res < A V = . ± .
18, while for these galaxies the mean ex-tinction is 0 . ± .
29 (
TGASPEX ), 0 . ± .
29 (
MUFFIT ), and1 . ± .
35 (
BaySeAGal ). The four codes obtain a similar log M (cid:63) ( ∼ . ± .
5) and stellar metallicity ( (cid:104) log Z (cid:63) (cid:105) M ∼ − . ±
4) forthese galaxies. The 9% fraction of
AlStar red galaxies have z = . ± .
20 and a median SNR of only 3 . MAG_AUTO pho-tometry), although the results from the four codes show signif-icant di ff erences in the highest redshift bins. For example, at z ∼ . z = . ∼ ∼ ff erent Article number, page 18 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field f r a c t i o n o f g a l a x i e s intrinsic redintrinsic blue Fig. 17.
Fraction of red and blue galaxies at di ff erent redshifts ob-tained by BaySeAGal for a delayed- τ SFH (circles),
MUFFIT (squares),
AlStar (stars), and
TGASPEX (crosses). codes and it ranges from values of ∼ BaySeAGal and
MUFFIT to ∼
30% with
AlStar . A fraction of red to blue galax-ies of 20% has been reported in the analysis of ALHAMBRAdata (Díaz-García et al. 2019a,b).A further study based on the detection of emission lines inthese galaxies will be necessary to really discriminate betweenred and blue galaxies with 1 . < ( u − r ) res < .
5. However, an SNR above 5 will also help to classify themwith better agreement between the di ff erent codes. The colour segregation of galaxy populations may be interpretedin terms of the di ff erences of either their stellar content or theirevolutionary pathways. Here we discuss the stellar populationproperties of red and blue galaxies as a function of redshift.In particular, we explore the evolution of ( u − r ) int , log M (cid:63) , (cid:104) log age (cid:105) M , and (cid:104) log Z (cid:63) (cid:105) M for blue and red galaxies (see alsoFig. 18). We average the value of each galaxy property at anyredshift bin. We notice that red and blue galaxies are properlydistinguished according to their stellar content, where the prop-erties of red galaxies are better constrained than for blue galax-ies. The colour ( u − r ) int of blue and red galaxies is bluer at higherredshift. For red galaxies, the two bins at lower redshift showslightly bluer colours than galaxies at intermediate redshift. Asa reference, a colour ( u − r ) int = u − r ) int = ∼ . ff erence being smaller at z = z = .
1. The mass of both blue and red galaxies is typicallylarger at higher redshift. This is a consequence of the incom-pleteness of the sample, because faint and / or less massive galax-ies are not imaged at high redshift. In addition, the larger num-ber of massive galaxies at higher redshifts is just a consequenceof the larger volume observed. Blue and red galaxies are verysegregated in terms of the their values of (cid:104) log age (cid:105) M . At any red-shift, red galaxies are older by ∼ . ff erent star formation histories and / or forma-tion epochs in the blue cloud and in the red sequence, at least at z =
1. However, the (cid:104) log age (cid:105) M of both blue and red galaxies de-creases at increasing redshift, indicating an ongoing star forma-tion and / or reflecting a biased sample for the low mass galaxiesat higher redshift.Blue and red galaxies also di ff er in (cid:104) log Z (cid:63) (cid:105) M . While redgalaxies mostly show solar metallicity and above, blue galax-ies are always below solar values. However, the metallicity val-ues obtained from the di ff erent SED-fitting codes is more uncer-tain, which makes more di ffi cult the distinction of a separated (cid:104) log Z (cid:63) (cid:105) M relation between the two galaxy populations. Giventhe large uncertainty of the results, there is no evidence of ametallicity evolution with redshift. However, there is a drop of ∼ . z ≤ . Before we can estimate the star formation rate density of the Uni-verse from our data, we need to characterize the volume incom-pleteness of our sample and the stellar mass limits as a functionof redshift. For this particular case we used the stellar massesobtained using the
MAG_AUTO photometry, as they provide a bet-ter estimate of the total galaxy stellar mass than the ones derivedusing
MAG_PSFCOR . Firstly, we needed to set the minimum andmaximum redshifts ( z min and z max , respectively) at which everygalaxy can be detected in our sample owing to the miniJPASdetection limits. We estimated the z min and z max values (see theblack lines in Fig. 19) by using the SFH and stellar populationproperties of each galaxy and calculating at which redshift theobserved magnitude in the r band would be equal to 14 . .
5, respectively. Then, the average z min and z max and their 1 σ dispersion were determined after binning the sample in stellarmass bins of ∆ log M (cid:63) = . z min and z max are based on the SEDs obtained from theSED-fitting analysis of the J-spectra fits rather than other tradi-tional methods in which the k -correction is based on a predefinedset of galaxy templates.As a result, galaxies with a stellar mass of log( M (cid:63) / M (cid:12) ) ∼
10 can be detected up to z = .
8, and low mass galaxies oflog( M (cid:63) / M (cid:12) ) ∼ . z = .
15 (see Fig. 19). Our resultspoint out that J-PAS will be able to study samples of galaxieswith a stellar mass above of log( M (cid:63) / M (cid:12) ) ∼ .
9, 9 .
5, and 9 . z = .
3, 0 .
5, and 0 .
7, respectively. These limits are ∼ . . / StePS.
One of the most significant observational results obtained fromgalaxy redshift surveys is the cosmic evolution of the star forma-tion rate of the Universe. It is well-known that the SFR density( ρ (cid:63) ) peaks at ∼ . z ∼
2, and since then ρ (cid:63) decreases until the present days (Lilly et al. 1996; Madauet al. 1998; Hopkins & Beacom 2006; Fardal et al. 2007; Gu-nawardhana et al. 2013; Madau & Dickinson 2014; Driver et al.2018). This result has been also obtained by low- z galaxy sur-veys using fossil record methods of nearby galaxies (Heavenset al. 2004b; Panter et al. 2007; López Fernández et al. 2018;Sánchez et al. 2019b; Bellstedt et al. 2020). Here, we test the Article number, page 19 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb ( u - r ) i n t ( u - r ) r e s l o g M a ss [ M ] ( u - r ) r e s < l o g a g e > M [ y r ] ( u - r ) r e s < l o g Z > M [ Z ] ( u - r ) r e s Fig. 18.
Evolution of intrinsic colour, stellar mass, age, and metallicity of galaxies obtained by
BaySeAGal for a delayed- τ SFH. Dots represent theaverage values of each property in each redshift bin derived by
BaySeAGal (circles),
MUFFIT (squares),
AlStar (stars), and
TGASPEX (crosses).The dispersion with respect to the average values is shown as error-bars. Blue and red dots correspond to blue cloud and red-sequence galaxies,respectively. capability of J-PAS data, along with our methods, to derive theSFH of galaxies and their properties, as well as to compare thecosmic evolution of the ρ (cid:63) of the Universe from a subset ofnearby galaxies in our sample.Our results (see Fig. 19) suggest that galaxies in a redshiftbin centred at z ∼ . ρ (cid:63) using the SFH of nearby galaxies (0 . ≤ z ≤ . z = .
15 miniJPAS includes galaxies with stellarmass above 2 × M (cid:12) . While galaxies with stellar mass belowthis limit present a minor contribution to the total stellar massdensity, they may significantly contribute to the star formationrate of the Universe in the last 4 Gyr. This is because low-massgalaxies experienced their main star formation processes at re-cent epochs (Asari et al. 2007; Bellstedt et al. 2020). However,we have detected a small fraction of these galaxies in our sampleand it is di ffi cult to account for the mass incompleteness below2 × M (cid:12) .To get ρ (cid:63) and take the volume incompleteness e ff ect intoaccount, we divide the SFR of each galaxy at 0 . ≤ z ≤ . V max , over which thegalaxy could be observed. However, we note that only a smallfraction ( ∼ z = .
15 (i.e. z max < . V max is equalto the comoving volume ( V c ) at this redshift range, meaning that V max = ∆ V = V c ( z = . − V c ( z = . BaySeAGal , the SFRs were obtained from the paramet-ric SFHs for a 1 Gyr resampling, while for the non-parametriccodes we used a larger interval in lookback time of ∼ . ≤ z ≤ .
15 includes a smallnumber of galaxies (360 galaxies) and the ages of the compo-nents in the non-parametric codes is discretized according to theages of the SSP model set. This is not an issue for parametriccodes because the SFH is described by a smoother and continu-ous composition of stellar population models. The error of log ρ (cid:63) in each epoch is obtained by propagating the dispersion in thestellar mass formed in each bin, although these errors are typ-ically lower than 0 .
05 dex for most of the epochs. The resultsobtained from the four codes (see Fig. 20) point out that ρ (cid:63) in-creases with redshift, up to z ∼ . BaySeAGal and to higherredshift for the parametric codes. It is worth mentioning that fos-sil record methods hardly distinguish between stellar populationsof ages 11 to 13 Gyr. Thereby, it is hard to distinguish betweenpopulations formed above z > ρ (cid:63) obtained in thiswork with previous results in the literature (Panter et al. 2007; Article number, page 20 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field l o g M a ss [ M ] r max = 22.5r min = 14.6 r Fig. 19.
Redshifts and stellar masses obtained by
BaySeAGal usingthe
MAG_AUTO photometry for each of the galaxies in our sample. Theblack lines show the z max and z min values corresponding to limitingmagnitudes of the miniJPAS galaxy sample selection analyzed here(14 . ≤ r ≤ . . ≤ z ≤ .
15 used to explore the evolution of the star formation ratedensity (SFRD, ρ (cid:63) ). All the points are colour-coded according to thegalaxy magnitudes in the r SDSS band and
MAG_AUTO apertures.
Madau & Dickinson 2014; Driver et al. 2018; López Fernándezet al. 2018; Sánchez et al. 2019b; Leja et al. 2019; Bellstedt et al.2020, see Fig. 20). These results are derived from di ff erent datasets and using di ff erent approaches to the analysis. For instance,the works by Madau & Dickinson (2014) and Driver et al. (2018)were performed with data from galaxy redshift surveys. The restof studies are based on nearby galaxies and fossil record methodsof stellar populations, where the cosmic SFR was constrainedby using non-parametric (Panter et al. 2007; Leja et al. 2019;Sánchez et al. 2019b) and parametric (López Fernández et al.2018; Bellstedt et al. 2020) SFH models. There are many simi-larities and discrepancies between the results that are mainly re-lated to the properties of the samples, the quality of the data, themethodologies, as well as the ability to correct for cosmic vari-ance, AGN contribution, and dust e ff ects. Overall, we concludethat the analysis of miniJPAS data provides successful resultsthat are in good agreement with cosmological surveys (Madau &Dickinson 2014; Driver et al. 2018), and the fossil record anal-ysis of SDSS (Panter et al. 2007), IFS CALIFA (López Fernán-dez et al. 2018), and GAMA data (Bellstedt et al. 2020). In thisregard, we have proven that our data and analysis techniques ex-hibit a remarkable potential to predict the evolution of both theSFR and stellar mass density of the Universe with cosmic time.
7. Summary and conclusions
In this paper we presented an analysis of miniJPAS data us-ing the full J-PAS filter system to evaluate the potential of J-PAS for galaxy evolution studies. Our primary aim is to iden-tify and characterize the stellar population properties of galax-ies and their evolution up to z =
1. Using the fossil recordmethod for stellar populations, we analyzed the observed opti-cal SEDs, the J-spectra, of ∼ r SDSS ≤ . MAG_AUTO mag- nitudes), z ≤
1, and
CLASS_STAR ≤ .
1. The J-spectra of thesegalaxies were fitted with di ff erent codes to constrain the stellarmass, rest-frame and intrinsic ( u − r ) colours, extinction, age, andmetallicity of their stellar populations. The bimodal distributionof galaxies was identified by using the stellar mass–colour dia-gram corrected for extinction, log M (cid:63) – ( u − r ) int , and its evolutionwas explored across the cosmic time up to z =
1. The impact inthe results of the photon-noise errors in the J-spectra photometryand the photo- z uncertainties were explored, together with theuncertainties in the estimation of the di ff erent stellar populationproperties across the log M (cid:63) –( u − r ) int plane.One parametric ( BaySeAGal ) and three non-parametricSED-fitting codes (
MUFFIT , AlStar , and
TGASPEX ) were used toobtain the distribution of stellar population properties of the sam-ple, and to check the consistency of the results amongst codes.We used a common set of SSP models from an updated versionof the Bruzual & Charlot (2003) synthesis models, with the at-tenuation law by Calzetti et al. (2000), and assuming the IMFof Chabrier (2003). The main di ff erences between the codes are:(i) BaySeAGal is a Bayesian approach that derives the PDF foreach of the stellar population parameters by assuming a delayed- τ or exponential SFR. (ii) MUFFIT combines two-burst SSP mod-els. (iii)
AlStar and
TGASPEX use an arbitrary combination ofSSP models to solve the non-negative least-squares problem.
MUFFIT , AlStar , and
TGASPEX do not perform a fully Bayesianevaluation of the PDF; instead, they follow a Monte-Carlo ap-proach by adding Gaussian noise to the observed fluxes to itera-tively repeat the SED-fitting process. The median and mean val-ues of the properties obtained for each galaxy by the four SED-fitting codes were compared. The full PDF from
BaySeAGal were also compared to the median / mean distributions.From the results obtained by BaySeAGal , we can drawsome conclusions that can be extended to the other codes.The galaxy stellar mass, extinction, metallicity, mass- andluminosity-weighted ages, and rest-frame and dust-correctedcolours can be estimated from the fits of J-spectra of galaxiesbrighter than 22 . r SDSS band that have a mean ofthe median signal-to-noise in the narrow band filters of ∼ . ± . . ± .
06 dex, and 0 . ± . u − r ) res colour, stellar mass,and extinction, respectively. The precision of the mass-weightedage of the stellar populations is 0 . ± .
05 and 0 . ± . BaySeAGal . The precision in the resultsis remarkable considering that ∼
25% of the sample have SNR ≤
3. Better precision is obtained when only galaxies with SNR ≥
10 are selected. In this case, the precision for ( u − r ) res colour,stellar mass, extinction, mass-weighted age, and stellar metallic-ity are: 0 . ± .
02, 0 . ± .
03 dex, 0 . ± .
09, 0 . ± . . ± .
25 dex, respectively. This precision is only slightlybelow to what will obtained in spectroscopic surveys of simi-lar SNR, such as WEAVE / StePs, which will get precision in themass-weighted age of ∼ . . < z < . I AB < . – The distributions of galaxy properties such stellar mass, rest-frame colours, extinction, and metallicity, are very similar.However, the distributions of the stellar ages show di ff er-ences. The distributions of age from BaySeAGal (assuming
Article number, page 21 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb l o g ( M y r M p c ) miniJPAS (z = 0.05-0.15)BaySeAGalMUFFITAlStarTGASPEX Madau & Dickinson Driver + 18
Panter + 07
Lopez Fernandez + 18
Sanchez + 19
Leja + 19 ( log M prior ) Leja + 19 ( continuity prior ) Bellstedt + 20
Fig. 20.
Cosmic evolution of the star formation rate density (SFRD, ρ (cid:63) ) obtained from the SED-fitting results of BaySeAGal (black dots),
MUFFIT (coral dots),
AlStar (cyan dots), and
TGASPEX (olive dots), with the nearby galaxies (0 . ≤ z ≤ . BaySeAGal ) and cyan (
AlStar )shaded regions represent the uncertainties associated to the results. The di ff erent lines represent the SFRD obtained in other works (see inset) andrecently compiled by Bellstedt et al. (2020). a delayed- τ SFR model) and
MUFFIT are similar, and shiftedtoward younger ages with respect to the distributions ob-tained with
AlStar and
TGASPEX . This is probably a conse-quence of the earlier formation epoch and rapid mass growthin the SFH of blue galaxies found by
AlStar and
TGASPEX with respect to
MUFFIT and
BaySeAGal . But, the consistencyin the results (e.g. between
BaySeAGal + delayed- τ and AlStar that are the two models with larger discrepancies)is within 0 .
06 in log M (cid:63) , 0 . A V , 0 . (cid:104) log Z (cid:63) (cid:105) M ,0 .
15 in (cid:104) log age (cid:105) M , and 0 .
04 in (cid:104) log age (cid:105) L . – Red and blue galaxies are well identified in the stellar massand dust-corrected ( u − r ) diagram. However, the fractionof red and blue galaxies varies according to the SED-fittingcode, especially in the highest redshift bins considered here.We have estimated that the fraction of galaxies in the bluecloud varies between 85% ( BaySeAGal ), 81% (
MUFFIT ),69% (
AlStar ), and 76% (
TGASPEX ); but the fraction of‘dusty-star forming galaxies’ (blue galaxies with ( u − r ) res > A V > BaySeAGal ), 11% (
MUFFIT ), 8%(
AlStar ), and 9% (
TGASPEX ); consistent within a few per-cent. – Red and blue galaxy populations can be characterized by cor-relations between the stellar population properties and theposition of each galaxy in the stellar mass and dust-corrected( u − r ) diagram. All the SED-fitting codes provided consistentresults for the stellar population properties and its evolutionwith redshift. Red and blue galaxies are well separated bytheir ( u − r ) int colour with mean values equal to ∼ ∼ – In terms of redshift evolution, blue and red galaxies are olderat the present-day than at z ∼
1. The mean stellar mass ofgalaxies increases with redshift, as expected from selectione ff ects of flux-limited samples, where at high redshift onlythe brightest galaxies are detected. An increase of the metal-licity since z ∼ z ∼
1; but this is prob-ably a consequence of the observed bias in stellar mass.The miniJPAS survey and its novel J-PAS filter system haveproven their capability to identify and characterise galaxy popu-lations and their evolution across cosmic time up to z =
1. TheJ-PAS survey will cover several thousand times the sky area ob-
Article number, page 22 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field served by miniJPAS, and it will provide complete and statisti-cally significant samples of galaxies with J-spectra with SNRabove 3, to retrieve the stellar population properties as a func-tion of redshift and environment. Similar precision to the anal-ysis of the spectroscopic data sets will be derived by analyzingthe sub-sample of galaxies with SNR ≥
10. In addition, J-PASwill be able to study samples of galaxies with stellar mass abovelog( M (cid:63) / M (cid:12) ) ∼ .
9, 9 .
5, and 9 . z = .
3, 0 .
5, and 0 .
7, re-spectively. These limits are > / StePS. The analysis of the miniJPAS galaxieswith log( M (cid:63) / M (cid:12) ) > . z ∼ . ρ (cid:63) up to z ∼ Acknowledgements.
R.G.D., L.A.D.G., R.G.B., G.M.S., J.R.M., and E.P. ac-knowledge financial support from the State Agency for Research of the Span-ish MCIU through the "Center of Excellence Severo Ochoa" award to the Insti-tuto de Astrofísica de Andalucía (SEV-2017-0709), and to the AYA2016-77846-P and PID2019-109067-GB100. L.A.D.G. also acknowledges financial supportby the Ministry of Science and Technology of Taiwan (grant MOST 106-2628-M-001-003-MY3) and by the Academia Sinica (grant AS-IA-107-M01).G.B.acknowledges financial support from the National Autonomous University ofMéxico (UNAM) through grant DGAPA / PAPIIT IG100319 and from CONA-CyT through grant CB2015-252364. SB acknowledges PGC2018-097585- B-C22, MINECO / FEDER, UE of the Spanish Ministerio de Economia, Indus-tria y Competitividad. L.S.J. acknowledges support from Brazilian agenciesFAPESP (2019 / / / / / T250 tele-scope and PathFinder camera for the miniJPAS project at the Observatorio As-trofísico de Javalambre (OAJ), in Teruel, owned, managed, and operated by theCentro de Estudios de Física del Cosmos de Aragón (CEFCA). We acknowl-edge the OAJ Data Processing and Archiving Unit (UPAD) for reducing andcalibrating the OAJ data used in this work. Funding for OAJ, UPAD, and CE-FCA has been provided by the Governments of Spain and Aragón through theFondo de Inversiones de Teruel; the Aragón Government through the ResearchGroups E96, E103, and E16_17R; the Spanish Ministry of Science, Innovationand Universities (MCIU / AEI / FEDER, UE) with grant PGC2018-097585-B-C21;the Spanish Ministry of Economy and Competitiveness (MINECO / FEDER, UE)under AYA2015-66211-C2-1-P, AYA2015-66211-C2-2, AYA2012-30789, andICTS-2009-14; and European FEDER funding (FCDD10-4E-867, FCDD13-4E-2685).
References
Ahumada, R., Allende Prieto, C., Almeida, A., et al. 2020, ApJS, 249, 3,[1912.02905].Aihara, H., Arimoto, N., Armstrong, R., et al. 2018, PASJ, 70, S4, [1704.05858].Arnouts, S. & Ilbert, O. 2011, LePHARE: Photometric Analysis for RedshiftEstimateAsari, N. V., Cid Fernandes, R., Stasi´nska, G., et al. 2007, MNRAS, 381, 263,[0707.3578].Baldry, I. K., Glazebrook, K., Brinkmann, J., et al. 2004, ApJ, 600, 681, [astro-ph / / / / / / / / / / Article number, page 23 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb
Goddard, D., Thomas, D., Maraston, C., et al. 2017b, MNRAS, 466, 4731,[1612.01546].González Delgado, R. M., Cerviño, M., Martins, L. P., Leitherer, C., &Hauschildt, P. H. 2005, MNRAS, 357, 945, [astro-ph / / / / ff mann, G., Heckman, T. M., White, S. D. M., et al. 2003b, MNRAS, 341,33, [astro-ph / ff mann, G., Heckman, T. M., White, S. D. M., et al. 2003a, MNRAS, 341,54, [astro-ph / ff s, New Jersey)Le Fèvre, O., Cassata, P., Cucciati, O., et al. 2013, A&A, 559, A14, [1307.0545].Leja, J., Carnall, A. C., Johnson, B. D., Conroy, C., & Speagle, J. S. 2019, ApJ,876, 3, [1811.03637].Lilly, S. J., Le Fevre, O., Hammer, F., & Crampton, D. 1996, ApJ, 460, L1,[astro-ph / / / / / / / / / / / ff & T. R. Lauer, 152Whitaker, K. E., van Dokkum, P. G., Brammer, G., et al. 2010, ApJ, 719, 1715,[1006.5453].Williams, R. J., Quadri, R. F., Franx, M., van Dokkum, P., & Labbé, I. 2009,ApJ, 691, 1879, [0806.0625].Wolf, C., Meisenheimer, K., Rix, H. W., et al. 2003, A&A, 401, 73, [astro-ph / / Article number, page 24 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field Instituto de Astrofísica de Andalucía (CSIC), P.O. Box 3004, 18080Granada, Spaine-mail: [email protected] Academia Sinica Institute of Astronomy & Astrophysics (ASIAA),No. 1, Section 4, Roosevelt Road, Taipei 10617, Taiwan Departamento de Física, Universidade Federal de Santa Catarina,P.O. Box 476, 88040-900, Florianópolis, SC, Brazil Instituto de Radioastronomía y Astrofísica (IRyA), Universidad Na-cional Autónoma de México, Campus Morelia, Michoacán 58089,México Centro de Estudios de Física del Cosmos de Aragón (CEFCA),Unidad Asociada al CSIC, Plaza San Juan 1, E-44001, Teruel, Spain Donostia International Physics Center (DIPC), Manuel LardizabalIbilbidea 4, San Sebastián, Spain Ikerbasque, Basque Foundation for Science, E-48013 Bilbao, Spain Universidade de São Paulo, Instituto de Astronomia, Geofísica eCiências Atmosféricas, R. do Matão 1226, 05508-090, São Paulo,Brazil Observatório do Valongo, Universidade Federal do Rio de Janeiro,20080-090, Rio de Janeiro, RJ, Brazil Centro de Investigaciones de Astronomía (CIDA), AP 264, Mérida5101-A, Venezuela Instituto de Astronomía, Universidad Nacional Autónoma de Méx-ico, A.P. 70-264, 04510 México, CDMX, México Núcleo de Astronomía, Universidad Diego Portales, Av. EjércitoLibertador 441, Santiago, Región Metropolitana, Chile Instituto de Física, Universidade de São Paulo, Rua do Matão 1371,CEP 05508-090, São Paulo, Brazil Instituto de Física de Cantabria (CSIC-UC). Avda. Los Castros s / n.39005, Santander, Spain Observatório Nacional, Ministério da Ciencia, Tecnologia, Inovaçãoe Comunicações, Rua General José Cristino, 77, São Cristóvão,20921-400, Rio de Janeiro, Brazil Department of Astronomy, University of Michigan, 311 West Hall,1085 South University Ave., Ann Arbor, USA Department of Physics and Astronomy, University of Alabama, Box870324, Tuscaloosa, AL, USA Instituto de Astrofísica de Canarias, Calle Vía Láctea SN, ES38205La Laguna, Spain Departamento de Astrofísica, Universidad de La Laguna, ES38205,La Laguna, Spain Núcleo de Astrofísica e Cosmologia, PPGCosmo & Dep. de Física,Universidade Federal do Espírito Santo, 29075-910, ES, Brazil INAF – Osservatorio Astronomico di Trieste, via Tiepolo 11, 34131Trieste, Italy IFPU – Institute for Fundamental Physics of the Universe, via Beirut2, 34151, Trieste, Italy Departamento de Astronomia, Instituto de Física, Universidade Fed-eral do Rio Grande do Sul (UFRGS), Av. Bento Gonçalves 9500,Porto Alegre, R.S, Brazil Centro de Estudios de Física del Cosmos de Aragón (CEFCA), PlazaSan Juan 1, E-44001, Teruel, Spain Tartu Observatory, University of Tartu, Observatooriumi 1, 61602Tõravere, Estonia Física, Universidad Técnica Federico Santa María, Casilla 110-V,Avda. España 1680, Valparaíso, Chile
Appendix A: Fits and quality assessment with
BaySeAGal
The quality of J-spectra fits for the whole sample was confrontedin multiple ways. Firstly, by its reduced χ value ( χ , seetop-right panel in Fig. A.1), which is defined as the χ value di-vided by the number of available bands used during the SED-fitting analysis. Secondly, by the residuals, ∆ mag, calculatedas the di ff erence between the observed and the model magni-tudes for each narrow-band filter (see top-left panel in Fig. A.1). Model magnitudes correspond to the SFH with the minimum χ value. Finally, by the normalized residual, meaning the resid-uals divided by the photon-noise uncertainties of each band(see bottom-right panel, and middle- and bottom-left panels ofFig. A.1).The distribution of χ values has a mean equal to 1 . .
01, where the standard deviation of the distri-bution is σ = .
83. It is also worth mentioning that the distribu-tion of the normalized residuals is properly centred with mean − .
06, median − .
15, and σ = .
4. These two distributions in-dicate that the J-spectra are properly fitted within the errors. Theresiduals and the normalized residuals change across the spec-trum, although always within the errors. ∆ mag and the disper-sion are smaller for the red than for the blue bands ( λ < MAG_PSFCOR
J-spectra, similar results were obtained using the
MAG_AUTO photometry. For this case and for all the filters, thenormalized residual (i.e. ∆ mag / error) is close to unity, pointingout that the errors of MAG_AUTO are not underestimated.
Appendix B: Similarities and discrepanciesbetween the codes for the evolution of galaxypopulations in the mass–colour diagram
We have discussed the di ff erences in the distribution of galaxypopulations when the dust-corrected colour ( u − r ) int is consid-ered instead of the ( u − r ) res . Here, we present first the mass-( u − r ) int diagram obtained from MUFFIT , AlStar , and
TGASPEX (Fig. B.1); and then we discuss the similarities and di ff erences inthe distributions of log M (cid:63) , ( u − r ) res , and ( u − r ) int for three red-shift bins: z ≤ .
35, 0 . < z ≤ .
6, and 0 . < z ≤
1. Notice thatthe distribution of galaxy populations in the mass–( u − r ) int planefrom MUFFIT is very similar to the results with
BaySeAGal ; andshow clear di ff erences with respect to the results from AlStar and
TGASPEX . In particular, the line traced by ( u − r ) limint shows asharper and clearer division between red and blue galaxies in theplane derived by BaySeAGal and
TGASPEX and by
AlStar and
TGASPEX ; thus, as we have already discussed, it gives a largernumber of red galaxies from
AlStar with respect to
BaySeAGal and
MUFFIT .In general, the distributions of stellar mass present little dif-ferences amongst the SED-fitting codes at the di ff erent cosmicepochs explored in this work. As expected for flux-limited sam-ples, the four codes retrieved a higher number of massive galax-ies at increasing redshift owing to both the survey detection limit,or depth, and the higher volume observed at higher redshifts.However, there are still some discrepancies between the resultsof the four codes. At z > .
35 (see top-middle and top-rightpanels in Fig. B.2), there is a larger fraction of massive galaxiesderived by
TGASPEX and
AlStar than with
MUFFIT , while thesame distribution for
BaySeAGal is a bit broader than the other,with values in between the distributions of
MUFFIT and
AlStar .However, these di ff erences are not as significant that would indi-cate that there are some problems to determine the stellar massdistribution of the miniJPAS galaxies, but, rather, there are inher-ent discrepancies in the four methodologies that yield this kindof di ff erences in the distributions.The distributions of ( u − r ) int from the four codes show largerdi ff erences. At low redshift, the distributions are very similar. Atintermediate redshift, BaySeAGal and
MUFFIT distributions are
Article number, page 25 of 27 & A proofs: manuscript no. GalEvo_AEGIS_25Feb m a g s s t d m e a n Fig. A.1.
Residuals and figures of merit of the
MAG_PSFCOR
J-spectra fits.
Top-right panel : distribution of χ . The mean, the median, and standarddeviation of the χ distribution are indicated. Top-left panel : mean value of the di ff erence between the observed and the best-fitting model; errorbar is the 1 sigma uncertainty level for each filter, while dashed lines are the global averaged for 1 and 2 sigma uncertainty levels. Bottom-rightpanel : distribution of the ratio between ∆ mag and error for all the filters and galaxies. The mean, median, and standard deviation of this distributionare indicated. Middle- and bottom-left panels : variation of the mean and standard deviation for each filter.
Upper panels are for
MAG_PSFCOR and bottom panels for
MAG_AUTO apertures. equal, and shifted to bluer colours than the distributions from
AlStar and
TGASPEX . At the highest redshift bin, 0 . ≤ z ≤ BaySeAGal and the re-sults from the non-parametric codes. The ( u − r ) int distributionsfrom AlStar and
TGASPEX are similar and more widespread inthe colour range of the SSPs than the
MUFFIT and
BaySeAGal results. These di ff erences in the ( u − r ) int distributions are moreremarkable in the blue cloud galaxies and are mainly related todi ff erences in the values of extinction, which in turn concern theSFH assumptions adopted in each of the codes. A similar conclu-sion can be extended to the di ff erences between the distributionsof (cid:104) log age (cid:105) M and (cid:104) log age (cid:105) L from BaySeAGal and
MUFFIT withrespect to
AlStar and
TGASPEX , where the last two codes pro-duce older ages for the miniJPAS galaxies (i.e. redder colours).
Article number, page 26 of 27. M. González Delgado et al.: Galaxy populations in the AEGIS field ( u - r ) i n t MUFFIT0 < z z z < l o g a g e > M [ y r ] ( u - r ) i n t AlStar 9.09.29.49.6 8 9 10 11 120123 9.09.29.49.6 8 9 10 11 120123 9.09.29.49.6 < l o g a g e > M [ y r ] ( u - r ) i n t TGASPEX 9.09.29.49.6 8 9 10 11 12log Mass [M ]0123 9.09.29.49.6 8 9 10 11 12log Mass [M ]0123 9.09.29.49.6 < l o g a g e > M [ y r ] Fig. B.1.
Intrinsic colour ( u − r ) vs. stellar mass for the redshift bins z ≤ .
35, 0 . < z ≤ .
6, and 0 . < z ≤ from left to right ) for the resultsdetermined by MUFFIT , AlStar , and
TGASPEX . The dashed line shows the ( u − r ) limint for the mean redshift in each bin (details in the text). N
0 < z int N TGASPEXAlStarMUFFITBaySeAGal z int z
10 1 2 3(u-r) int
Fig. B.2.
Distributions of stellar mass, and ( u − r ) colour corrected for extinction (top, and bottom panels, respectively) obtained by our SED-fittingcodes BaySeAGal , MUFFIT , AlStar , and
TGASPEX for the miniJPAS galaxies at di ff erent redshift bins (see panels). The grey curves also illustratethe posterior probability distribution functions (PDFs) obtained by BaySeAGal assuming a delayed- ττ