Magnifying the early episodes of star formation: super star clusters at cosmological distances
E. Vanzella, M. Castellano, M. Meneghetti, A. Mercurio, G. B. Caminha, G. Cupani, F. Calura, L. Christensen, E. Merlin, P. Rosati, M. Gronke, M. Dijkstra, M. Mignoli, R. Gilli, S. De Barros, K. Caputi, C. Grillo, I. Balestra, S. Cristiani, M. Nonino, E. Giallongo, A. Grazian, L. Pentericci, A. Fontana, A. Comastri, C. Vignali, G. Zamorani, M. Brusa, P. Bergamini, P. Tozzi
DDraft version September 17, 2018
Preprint typeset using L A TEX style emulateapj v. 11/26/03
MAGNIFYING THE EARLY EPISODES OF STAR FORMATION: SUPER STAR CLUSTERS ATCOSMOLOGICAL DISTANCES
E. Vanzella , M. Castellano , M. Meneghetti , A. Mercurio , G. B. Caminha , G. Cupani , F. Calura , L.Christensen , E. Merlin , P. Rosati , M. Gronke , M. Dijkstra , M. Mignoli , R. Gilli , S. De Barros , K.Caputi , C. Grillo , I. Balestra , S. Cristiani , M. Nonino , E. Giallongo , A. Grazian , L. Pentericci , A.Fontana , A. Comastri , C. Vignali , G. Zamorani , M. Brusa , P. Bergamini , P. Tozzi Draft version September 17, 2018
ABSTRACTWe study the spectrophotometric properties of a highly magnified ( µ (cid:39) −
70) pair of stellar systemsidentified at z=3.2222 behind the Hubble Frontier Field galaxy cluster MACS J0416. Five multipleimages (out of six) have been spectroscopically confirmed by means of VLT/MUSE and VLT/X-Shooter observations. Each image includes two faint ( m UV (cid:39) . (cid:46)
100 Myr), low-mass( < M (cid:12) ), low-metallicity (12+Log(O/H) (cid:39) (cid:39)
300 pc, after correcting for lensing amplification. We measured severalrest-frame ultraviolet and optical narrow ( σ v (cid:46)
25 km s − ) high-ionization lines. These features maybe the signature of very hot ( T > α emission (e.g., C iv / Ly α > α line flux is expectedto be 150 times brighter (inferred from the H β flux). A spatially-offset, strongly-magnified ( µ > α emission with a spatial extent (cid:46) . is instead identified 2 kpc away from the system. Theorigin of such a faint emission can be the result of fluorescent Ly α induced by a transverse leakageof ionizing radiation emerging from the stellar systems and/or can be associated to an underlyingand barely detected object (with m UV >
34 de-lensed). This is the first confirmed metal-line emitterat such low-luminosity and redshift without Ly α emission, suggesting that, at least in some cases, anon-uniform covering factor of the neutral gas might hamper the Ly α detection. Subject headings: cosmology: observations — galaxies: formation INTRODUCTION
The investigation of the ionizing properties of young,low mass star-forming systems caught at z (cid:39)
3, i.e., INAF–Osservatorio Astronomico di Bologna, via Gobetti 93/3,40129 Bologna, Italy INAF–Osservatorio Astronomico di Roma, via Frascati 33,00040 Monteporzio, Italy INAF – Osservatorio Astronomico di Capodimonte, ViaMoiariello 16, I-80131 Napoli, Italy Dipartimento di Fisica e Scienze della Terra, Universit`a di Fer-rara, via Saragat 1, 44122 Ferrara, Italy INAF - Osservatorio Astronomico di Trieste, via G. B. Tiepolo11, I-34131, Trieste, Italy Dark Cosmology Centre, Niels Bohr Institute, University ofCopenhagen, Juliane Maries Vej 30, DK-2100 Copenhagen, Den-mark Institute of Theoretical Astrophysics, University of Oslo, Post-boks 1029 Blindern, NO-0315 Oslo, Norway Observatoire de Gen`eve, Universit´e de Gen`eve, 51 Ch. desMaillettes, 1290, Versoix, Switzerland Kapteyn Astronomical Institute, University of Groningen,Postbus 800, 9700 AV, Groningen, The Netherlands Dipartimento di Fisica, Universit`a degli Studi di Milano, viaCeloria 16, I-20133 Milano, Italy University Observatory Munich, Scheinerstrasse 1, D-819M¨unchen, Germany Dipartimento di Fisica e Astronomia, Universit`a degli Studidi Bologna, Viale Berti Pichat 6/2, 40127 Bologna, Italy Dipartimento di Fisica e Astronomia “G. Galilei”, Universit`adi Padova, Vicolo dell’Osservatorio 3, I-35122, Italy INAF – Osservatorio Astrofisico di Arcetri, Largo E. Fermi,I-50125, Firenze, Italy * [email protected] † Based on observations collected at the European Southern Ob-servatory for Astronomical research in the Southern Hemisphereunder ESO programmes P095.A-0840, P095.A-0653, P186.A-0798. nearly 1 Gyr after reionization ended and the possibleanalogy with similar systems identified during reioniza-tion ( z >
6) represents today a strategic line of research,especially at poorly explored, low-luminosity regimes(e.g., Amor´ın et al. 2017). The detection of nebular highionization emission lines has underscored the consider-able contribution of hot and massive stars to the trans-parency of the medium in low-luminosity (
L << L ∗ ) sys-tems (Vanzella et al. 2016a), as expected in scenarios inwhich stellar winds and supernova explosions blow cavi-ties in the interstellar medium (e.g., Jaskot & Oey 2013;Micheva et al. 2017). In fact, the rest-frame ultravioletand optical line ratios observed in z (cid:39) z = 2 − z > z = 2 − z > < z < a r X i v : . [ a s t r o - ph . GA ] M a y Vanzella et al.2017b). In this regard, the exceptional line flux sensi-tivity of the integral field spectrograph MUSE mountedon the VLT (Bacon et al. 2010) has driven consider-able progress. In fact, MUSE allows the identificationand characterization of extremely faint (and small) lineemitters at 3 < z < .
6, without the need of specificpre-selection of the targets in a relatively large field ofview (1 (cid:48) × (cid:48) ), aptly matching strongly magnified regionsof the sky. In these investigations, strong lensing turnedout to be essential for two reasons: (1) before the comple-tion of the Extremely Large Telescope (ELT), it is notpossible to spatially resolve sources with effective radiilower than 150 pc ( < z > −
100 pc at 3 < z < . m >
29 and line fluxes of > − × − cgs.MUSE observations on HFFs are leading us to the con-struction of a reference sample at z < z > z (cid:39) − z = 3 . α emission. These fea-tures are very rare among known local and distant star-forming systems (e.g., Stark et al. 2014, 2015; Henry etal. 2015; Mainali et al. 2017; Shibuya et al. 2017; Smitet al. 2017). The intriguing nature of our system andof its ionization structure provides us with an insightfulexample of the geometrical complexity of high-redshiftyoung, ionized systems. The energy and the momentumcontinuously deposited by massive, young stellar associ-ations can lead to puzzling structures such as our object,which also represents a valuable benchmark for interstel-lar photo-ionization models.In this work, we use the Hubble Frontier Fields HST- dataset , in particular three optical and four near-infrared bands: F435W, F606W, F814W (HST/ACS)and F105W, F125W, F140W and F160W (HST/WFC3),with typical limiting AB magnitudes 28.5 - 29.0 calcu-lated at 5 σ depth. In addition to the seven HST bands,the publicly available Hawk-I@VLT K s band and theSpitzer/IRAC 3.6 and 4.6 µ m data have also been in-cluded (the description of the data-set, the extractionof the PSF-matched photometry and the construction ofthe photometric catalog are provided in Castellano et al.2016b and Merlin et al. 2016).This paper is organized as follows. In Sect. 2, a briefdescription of the multiple-image magnification patternof our system is presented. Our main results are pre-sented in Sect. 3, and discussed in Sect. 4. Finally, inSect. 5 some conclusions are drawn.Throughout the paper, we assume a flat cosmologywith Ω M = 0.3, Ω Λ = 0.7 and H = 70 km s − Mpc − . ID14, A DOUBLE LENSED SYSTEM
The source ID14 is multiply imaged by the clusterMACS J0416 into three images separated by up to 50 (cid:48)(cid:48) .The location of these three images is given in the rightpanel of Figure 1 by the three yellow circles labelled Im-age1, Image2, and Image3. Image1 happens to be closeto a pair of elliptical cluster members, indicated as E1and E2 in the left panel of Figure 1. These galaxies act asstrong lenses, further splitting Image1 into the four im-ages ID14a, ID14b, ID14c, and ID14d. For consistency,in the following we will refer to Image2 and Image3 asID14e and ID14f, respectively.This system was first discovered by Zitrin et al. (2013).In each of the images, the source appears to consist of twobright knots. This morphological marker, together withthe image geometry, allowed the first association of theimages to the same source. The two knots are labelled“1” and “2” in Figure 1. As discussed in the next section,the images ID14a,b,c,d are spectroscopically confirmed.The association of Image3 with the sixth multiple imageof ID14 (namely ID14f) is corroborated by the lens modelrecently presented by Caminha et al. (2016c), which pre-dicts the presence of an image at this location within1 σ uncertainty, i.e., < (cid:48)(cid:48) . In addition, the photometricredshift of ID14f is zphot = 3 . ± .
10 (see Figure 2),which is fully consistent with the spectroscopic redshiftof ID14. For the analysis discussed in this paper, it is crucialto estimate the magnification of the brightest images ofID14, namely images a, b and c . For this goal, the lensmodel by Caminha et al. (2016c) could in principle beused. There are, however, two complications. The firstis that these images are located in a region of high mag-nification. In a collaborative work which involved severalgroups employing different lens modeling algorithms, andamong them the one used by Caminha et al. (2016c),Meneghetti et al. (2017) recently showed that the modelmagnifications in this regime are affected by large uncer-tainties. For example, at µ >
10 the error on the localmagnification estimates is (cid:38) The photometric redshift is also reported in Castellanoet al. (2016b). In particular, an example of the best-fitSED for ID14f is shown at following address http://astrodeep.u-strasbg.fr/ff/?ffid=FF M0416CL & id=1141 . esolving parsec-scale star-forming regions at z=3.22 3 Fig. 1.—
Panel A: the HST color image of the galaxy cluster MACS J0416 showing the position of the three multiple images expected atthe redshift of ID14 generated by the galaxy cluster (“Image1”, “Image2” and “Image3”). “Image1” is further magnified by the ellipticalgalaxy cluster members E1+E2 into four additional multiple images (ID14a,b,c,d). The left and right top insets show the F814W imagewith and without the subtraction of the galaxies E1+E2. Each image of ID14 contains two further components, “1” and “2”. Magnitudescorrespond to the average values obtained from all the HST bands (except F435W). The other two multiple images “Image2” (ID14e)and “Image3” (ID14f) are also shown in the insets (red dotted circles), where the black crosses mark the position predicted by the lensmodel. All insets have a size of 3 . (cid:48)(cid:48) . Panel B: magnification map ( µ , coded according to the greyscale bar on the right) derived from thelens model by Caminha et al. (2016c): large magnifications are expected close to the critical lines, between “Image1” and “Image2” andbetween ID14a,b and c. Fig. 2.—
The best-fit SED solution (blue template) compared tothe photometry (black points) for image ID14f. The photometricredshift with its 1 σ uncertainty is reported (see also Castellano etal. 2016b). The inset shows the zoomed ID14f image in the F814Wband. The photometric black points includes both components“1,2”. strong magnification gradient near the critical lines. Thesecond complication is that images ID14a,b,c are the re- sult of a galaxy-galaxy strong lensing event. Meneghettiet al. (2017) also showed that the model uncertaintiesincrease near cluster members.For these reasons, instead of reading the magnifi-cations off the map derived from the Caminha et al.(2016c) model, we opted to derive the magnifications ofID14a,b,c using the same method employed in Vanzellaet al. (2016a, 2017b). Meneghetti et al. (2017) showedthat magnification estimates are significantly more ro-bust if µ <
5. Using the lens model by Caminha etal. (2016c), we estimate that the magnification of ID14fis µ = 2 . ± .
1. This estimate is consistent with oth-ers based on different models of MACS J0416, which weobtained from the Hubble Frontier Field Magnificationcalculator. If we assume that the model magnification ofthis image is secure, we can then derive the magnifica-tions of the other images by means of the measured fluxratios.ID14f is identified with the Astrodeep object . ± . Fig. 3.—
Left panel: the multiple images ID14a,b,c,d are shownin the F814W band. The morphological details of images ID14aand ID14c are indicated by the black arrows and the zoomed inset(for ID14c). The corresponding
Galfit models of each componentare shown on the right side (the numbers indicate the magnifica-tion, as derived in Sect. 2). computed multi-band, PSF-matched photometry of im-ages ID14a,b,c as described in Merlin et al. (2016), butusing the F814W band as detection image instead ofF160W. HST photometry has been measured from 30maspixel-scale images after subtracting the elliptical galax-ies E1+E2 with
Galfit (Peng et al. 2010). T-PHOTv2.0 (Merlin et al. 2016) has been used to measure Kand IRAC photometry. An example of subtracted ellip-ticals in the F814W-band is shown in Figure 1 (top-leftinset), where also the average magnitudes (over the fiveHST bands) are quoted for each component. For exam-ple, image ID14a is composed of two components “1,2”with magnitudes 26.52 and 26.61, respectively. The to-tal magnitude, combining the two components, is 25.81.The total magnification for ID14a is therefore µ (ID14a)= µ (ID14f) × (flux[ID14a]/flux[ID14f]) = 41 ±
7. Similarly, µ (ID14b)= 36 ± ±
10. The errors estimatedwith this method are < ∼ (cid:39) . σ sta-tistical errors from Caminha et al. for images ID14a(1)and ID14a(2) are µ = 33 +38 − and µ = 23 +15 − , respec-tively ( µ = 29 +64 − and µ = 22 +34 − for ID14b1 and 2).They change significantly if different and independentlens models are applied. In particular, the magnifi-cations of images ID14a(1) and ID14a(2) inferred fromeleven lens models span the interval 8 −
110 with me-dians µ (cid:39) −
40. Also the reported 1 σ uncertaintiesin each model are larger than 50%. The same happensalso for the other strongly magnified images, ID14b andID14c. The scatter among the various model predictionsis of the order of the predicted magnifications, suggest- We used the Hubble Frontier Field Magnification calculator, https://archive.stsci.edu/prepds/frontier/lensmodels/
Table 1. Observed spectral lines. line/ λ vacuum [ F obs ]( SN )( σ v ) [ F corr ](EW) RedshiftLy α (cid:39) .
1) [M] (3.226)[C iv ] λ .
20 [2.18](20.0)( (cid:46)
64) [0.87](16.9) [M] 3.222[C iv ] λ .
78 [1.20](11.0)( (cid:46)
64) [0.48](9.3) [M] 3.223He ii λ .
42 [0.33](4.8)( <
38) [0.13](2.8) [M] 3.223O iii ] λ .
81 [0.48](5.4)( <
42) [0.19](4.3) [M] 3.222O iii ] λ .
15 [1.02](11.7)( <
42) [0.41](9.2) [M] 3.222[C iii ] λ .
68 [0.55](6.2)( <
42) [0.22](6.5) [M] 3.223C iii ] λ .
73 [0.45](5.1)( <
42) [0.18](5.3) [M] 3.222[O ii ] λ .
09 [ < . ii ] λ .
88 [ < . ii ] λ .
44 [ < . Hβλ .
69 [1.06](3.7)(–) [0.71](136) [X] (3.2222)[O iii ] λ .
30 [2.76](9.6)( <
23) [1.84](367) [X] 3.2222[O iii ] λ .
24 [8.92](31.5)( <
18) [5.95](1209) [X] 3.2222Ly α <
40) – [M] 3.223
Note . — Column − ergs − cm − ), S/N and σ v (instrument-corrected velocity dispersion in kms − ) for ID14 extracted from the apertures that include the multipleimages. Flux limits are reported at 1 σ . Column F corr ) (see text for details). The rest-frame EW (˚A),derived adopting the SED-fitted continuum of “1”+“2” ( m (cid:39) .
80) isalso reported. The de-lensed fluxes can be obtained by dividing thesevalues by µ = 40. In column σ v of the lines with S/N > α flux calculated within the magenta ellipse (whosearea is 6.7 sq. arcsec) shown in Figure 4 is reported in the last row. ing that the systematic errors dominate the uncertain-ties, especially in this complex double-lensed object (seeMeneghetti et al. 2017 for a detailed discussion). There-fore, the method used in this work significantly decreasesthe uncertainties of the amplification factors and conse-quently limits the error budget on the intrinsic physicalproperties discussed in the next section. CHARACTERIZING SYSTEM ID14
The main goal of our investigation is to estimate thesize and the age of our system, as well as to assess thenature of the radiation field originated by its two close,compact sources. The gas phase metallicity is also esti-mated, as well as the dynamical mass, useful to constrainthe baryon content of ID14.
Sizes of the components “1” and “2”
Galfit fitting (Peng et al. 2010) has been performedon both components “1,2” of images ID14a,b,c (followingthe methodology described in Vanzella et al. 2017b) afterthe elliptical galaxies, E1 and E2, have been subtracted(Figure 1, top-left). All the multiple images are verynucleated and tangentially elongated along the arc shapeshowing an average effective radius < R e > = 1 . ± . q = b/a ) < q > = 0 . ± . Galfit modeling is shown in Figure 3). Thecircularized radius R c (= R e × q . ) is 0.8 pix (1 pix =0 . (cid:48)(cid:48) ) corresponding to a de-lensed size of (cid:39)
30 pc atz=3.222, adopting µ = 40, (or sizes lower than 50 pc ifwe increase R e to 2 pix and decrease µ to 30). The twocomponents “1,2” are separated by 300 pc on the sourceplane.esolving parsec-scale star-forming regions at z=3.22 5 Fig. 4.—
Panoramic view of the MUSE spectra. Top panels: the zoomed region centered on ID14a,b,c,d,e is shown (6 . (cid:48)(cid:48) wide). Fromleft to right we show the MUSE images at the Ly α (panel A), at 1548 ˚A (panel B), in the HST/F814W band (panel C), and the gray-codedmagnification map (panel D), in which the yellow contours enclose the region with µ >
25. The arc-shaped black line marks the apertureused to extract the MUSE spectrum as the sum of images ID14a,b,c. The black dashed circle identifies the image ID14e and the magentaellipse the diffuse Ly α emission at the same redshift as ID14 on top of the critical line of the galaxy cluster. Panel (B) shows the rectangularred aperture with the same width and position angle of the X-Shooter slit (also shown in panel C, top-right, with black dotted thin lines).The orange dotted circle in panel C marks the position of a possible counterpart of the diffuse Ly α emission (magenta ellipse). The bluecircle (1 . (cid:48)(cid:48) diameter) marks the aperture used to derive the spectrum for image ID14a only. Panel (E), shows the MUSE spectrum ofID14a,b,c (black line) extracted using the arc-shaped aperture. The continuum of the elliptical galaxies E1+E2 has been subtracted (whitethick line) and the high ionization lines are shown in the corresponding insets. The colors of the spectra in the insets corresponds to thecolors of the apertures shown in panel B. The flux scale of all images is F ( λ ) [10 − erg s − cm − ˚A − ]. Additional details on the observed multiple images arehighlighted in Figure 3 (see ID14c), but we will not dis-cuss them further in this work. We stress that these ad-ditional features are at least two magnitudes fainter thanthe main components (or constitute sub-structures of thecomponents themselves) and, given the large magnifica-tion, they would have a de-lensed ultraviolet magnitudefainter than 31 and de-lensed sizes possibly consistentwith forming star-clusters on parsec-scale. These addi-tional features might also be the result of the extremelylarge magnification gradient characterizing this field.
Spectral features
The 2h VLT/MUSE (ID 094.A-0115B, PI: J. Richard)and 2h VLT/X-Shooter (ID 098.A-0665B, PI: E.Vanzella) spectra of images ID14 are shown in Figures 4and 5, obtained with good seeing conditions, 0 . (cid:48)(cid:48) and0 . (cid:48)(cid:48) , respectively. The apertures over which the spectrahave been extracted are also shown in the same figure.Data reduction for X-Shooter and MUSE have been per-formed as described in Vanzella et al. (2016a) and Cam-inha et al. (2016c), respectively.The most prominent line in the MUSE spectrum is the C iv λ iv doublet. We usethis line to measure the redshift of the multiple imagescontained in the data-cube (only ID14f is not includedin the MUSE field of view). A continuum-subtracteddata-cube has been computed by averaging the signalin a central window of 3 spectral elements (one elementcorresponds to 1.25˚A) and subtracting the average signalfrom two adjacent windows of 10 spectral elements each(see also Vanzella et al. 2017a). Figure 6 (left panel)shows the result. The C iv λ > iv emission (especially for ID14a) showsan elongated shape that follows the spatial orientation ofcomponents “1,2”. This suggests that both componentscontribute to the line emission and in general to the ul-traviolet metal lines. Possibly, MUSE equipped with theadaptive optics module may allow these spectral features Vanzella et al.to be spatially resolved.We exploited the MUSE data-cube by choosing a free-form aperture encompassing the three images ID14a,b,c(increasing the S/N); on the other hand, the X-Shooterslit captures the image ID14b and one component (“1”)of the image ID14c (being the component “2” less mag-nified, see Figure 5, top left). Line fluxes, signal-to-noise, velocity dispersions, rest-frame equivalent widths(EW) and redshifts are reported in Table 1. High ion-ization lines C iv λ , ii λ iii ] λ , iii ] λλ , β , [O iii ] λλ , −
31. High ioniza-tion lines have also been confirmed with X-Shooter (with R = 7400), though only C iv λ iii ] λ (cid:38) α emissionhas been detected (S/N <
3) in the same spatial regionwhere both the metal lines and stellar continuum arise.We discuss this in Sect. 4.3. The observed line fluxesderived from the aforementioned apertures are reportedin the column F corr , obtained by dividingthe measured fluxes reported in column F corr and the inferred continuum magni-tude from the SED-fitting of both components “1,2” (seeSect. 3.3). Typical errors mainly reflect the uncertaintyin the line flux, as the underlying images are well detectedin the deep HFF images. Conservatively, even includingthe possible systematic errors (due to the subtraction ofE1+E2) the final error on the EWs is not larger than50%.Beside the presence of ultraviolet metal lines, alsothe prominent [O iii ] λλ , iii ] λλ , / [O ii ] λ , >
15 and[O iii ] λλ , β >
10 suggest a highly ionizedmedium and a possibly large Ly α escape fraction (Erbet al. 2016; Henry et al. 2015). Despite that, the Ly α emission in ID14 is deficient (see Sect. 4), independentlyfrom the adopted aperture shape. Physical Properties
SED fitting has been performed using BC03 templates(Bruzual & Charlot 2003) on each component “1,2”,for images ID14a,b and the component “1” of ID14c(see Castellano et al. 2016b) after subtracting the el-liptical galaxies E1+E2 in all the HST images and us-ing the T-PHOT v2.0 algorithm, Merlin et al. (2016)(see Sect. 2). The subtraction of E1+E2, however, mayintroduce not negligible color trends, especially in thenear-infrared bands. Such residuals may introduce sys-tematics in the inferred physical quantities. In order toadd constraints to our photometric analysis, we also esti-mate the dynamical mass and the gas phase metallicity.The inferred dynamical mass from the σ v ([O iii ] λ (cid:46)
20 km s − velocity dispersion is M dyn < × M (cid:12) , Table 2. Properties of Images ID14a and ID14f.
ID14a(1) ID14a(2) ID14f(1+2)M stellar [10 M (cid:12) ] 2 . − . . − . − SFR [M (cid:12) yr − ] 0 . − .
15 0 . − . − Age [Myr] 16 −
100 13 − − E(B-V) (cid:39) . (cid:39) . − R c (UV) [pc] 30 ±
11 30 ± − m UV (obs) 26 . ± .
03 26 . ± .
03 29 . ± . m UV (int) 30 . ± .
19 30 . ± .
19 29 . ± . β UV − . ± . − . ± . − . ± . µ ± ± . ± . Note . — De-lensed physical properties are derived from SED-fittingfor the single components “1” and “2” of image ID14a and the combi-nation “1+2” of ID14f by adopting the measured metallicity (see text).The 68% central intervals are reported for the stellar mass, star forma-tion rate and age, and the 1 σ errors for the remaining quantities. Theobserved (“obs”) and intrinsic (“int”) magnitudes are also reported.The intrinsic stellar masses and SFRs are obtained dividing by µ thebest fit parameters resulting from the SED fit. It is worth noting thatthe values for ID14f include both components “1+2”. The magnifica-tion µ is also reported and derived as described in the text. where we assume the [O iii ] λ T e = 19000 ± iii ] λ , iii ] λ T e is not sensitive on the assumed electrondensity in the range n e = 10 − − . We obtain12+log(O/H) (cid:39) . ± . M dyn estimate). In general, theinferred quantities agree among the multiple images ofthe same component.In particular we focus on images ID14a.1 and ID14a.2that are the least contaminated by the elliptical galax-ies E1+E2. The best-fit yields stellar masses (cid:46) M (cid:12) ,ages (cid:46)
100 Myr, SFRs (cid:39) (cid:12) yr − and E(B-V) < (cid:39) iii ] λλ , iii ] λ (cid:39) z ∼ .
2) on the basis of high equivalent width opticalemission lines and found strong Ly α emission in all ofthem. Remarkably, despite ID14 shows many propertiesin common with the GPs sources, the Ly α is deficient(see next section).esolving parsec-scale star-forming regions at z=3.22 7 Fig. 5.—
The X-Shooter spectra in the VIS and NIR arms are shown for the relevant emission lines. In panel (A) the two-dimensionalzoomed spectra of the UV metal high-ionization lines are shown (the faint continuum comes from the foreground elliptical galaxy). In panel(B) the color image of ID14a,b,c,d is shown with the MUSE aperture used to extract the spectra (blue curve) and the orientation and width(0 . (cid:48)(cid:48) ) of the X-Shooter slit (white lines). The size of the image is 3 (cid:48)(cid:48) on a side. The one and two-dimensional H β and [O iii ] λλ , (cid:39) µm are shown in panel (C). The left inset shows the two-dimensional spectrum centered at the position of the expected[O ii ] λ , iii ] λ Fig. 6.—
The continuum-subtracted data-cube at the C iv λ . (cid:48)(cid:48) wide). Theblue circles mark the multiple images of ID14 detected on the HSTcolor image, reported in the right panel. In particular, faint lineemissions arise at the position of multiple images ID14d and ID14e. DISCUSSION
The nature of the ionizing source
ID14 shows similarities with other star-forming sys-tems identified at z = 3 − ii λ iv λ , iii ] λ , σ upper limits on the X-ray flux of 1 . × − and 8 . × − erg s − cm − in the soft (0.5-2 keV) and hard (2-7 keV) band,respectively, taking into account the strong backgroundemission due to the intra-cluster gas. At the redshift ofID14 these limits correspond to luminosities of 1 . × and 6 . × erg s − in the soft and hard band, re-spectively. They cannot exclude the possibility that anabsorbed Seyfert-like AGN could be present.In addition to the above limits and the line ratios, thefact that components “1,2” are spatially resolved andshow very narrow nebular emission lines further supportsthe evidence that the source of ionizing radiation is dom-inated by stellar emission. This is also reminiscent of re-cent studies in the local universe (e.g., Smith et al. 2016;Annibali et al. 2015; Kehrig et al. 2015), where youngsuper star clusters show narrow high ionization nebu-lar lines. Although a low-luminosity AGN can not com-pletely excluded, the presence of sub-structures in ID14and the aforementioned characteristics support the con-clusion that the star-formation is likely to be the mainsource of ionizing radiation. A L=0.004L (cid:63) super star clusters at z=3.2222
ID14 is made of two main star-forming regions (“1,2”)of (cid:39)
30 pc effective radius each, with de-lensed mag-nitudes (cid:39) . (cid:63)z =3 ), possibly showing addi-tional fainter sub-components. The low stellar mass( < M (cid:12) ), low metallicity, young ages ( (cid:46)
100 Myr)and hot stars are also distinctive features of this system.The double knot morphology of ID14 suggests two su-per star clusters of several 10 M (cid:12) are present and possi-bly composed by a mixture of stellar populations. In fact,the high-ionization (e.g., He ii λ iv λ , Fig. 7.—
Best-fit SED solutions for the components 1 and 2 of the ID14a image. The inferred physical quantities are reported in Table 2.The greed dashed line is the ultraviolet slope β . The photometric discontinuity at 22000 ˚ A corresponds to the K-band and is fully consistentwith the measured flux of the optical emission lines observed with X-Shooter (see text). lines require a young ( (cid:46)
10 Myr) stellar component,characterized by blackbody effective temperatures higherthan 50000 K (Steidel et al. 2014; Stark et al. 2015), andthe ages inferred from the SED fitting also support thepresence of an underlying older ( (cid:46)
100 Myr) population.The inferred specific star formation rate (sSFR) is large( >
20 Gyr − ), a value comparable to those inferred byStark et al. (2014) and Karman et al. (2017) and suggest-ing the system is still in a starburst phase and rapidlygrowing. Conservatively, adopting the lowest stellar mass(2 . × M (cid:12) ) and the largest effective radius ( R C = 50pc) of the 68% intervals derived above, the observed stel-lar mass density of the star-clusters is relatively large, > (cid:12) pc − (see also, Vanzella et al. 2017b). Deficient Ly α emission in ID14 From the H β line flux, an estimate of the theo-retical Ly α emission can be computed as f(Ly α ) =8 . × f (H α ) × . × E ( B − V ) , where we assume a caseB recombination theory, f(H α )/f(H β )=2.7 (Brocklehurst1971). Adopting E(B-V)=0.0(0.2) (for the nebular emis-sion), the expected intrinsic Ly α flux is 1 . . × − erg s − , i.e., more than 150 times brighter thanwhat was observed, and more than 12 times brighterthan C iv λ , iv λ , α ) is (cid:38)
15. Therefore, Ly α emis-sion is strongly deficient. A conversion of the intrinsicLy α emission into absorption can in principle be achievedby dust suppression of Ly α photons, by geometrical ef-fects leading to the scattering of Ly α photons out ofthe line of sight, or by a combination of both. Simi-lar physical conditions observed in local low-metallicitystar-forming dwarf galaxies might help to better under-stand the nature of ID14 (though at different scales). Forexample, the low-metallicity star-forming dwarf galaxyIZw 18 shows Ly α absorption, despite its low metallicityand young age. Also, galaxies of the LARS sample (or:local, H α selected galaxies) with low velocity dispersionshow a reduced Ly α escape (Herenz et al. 2016). A largenumber of scatterings in static, high column density gascan lead to an efficient suppression of Ly α photons byeven small amounts of dust. The Ly α absorption couldalso be caused by diffusion of the photons out of the ob- server direction (not by dust destruction). As a result,Ly α absorption can be observed also in dust-free galaxies(see, Atek et al. 2009). Verhamme et al. (2012) showdirectional variation in the Ly α escape from a simulateddisk galaxy (see also Zheng & Wallace 2014). This direc-tional variation is reduced significantly in the presenceof outflows (e.g., Duval et al. 2016) and/or for non-diskgeometries including modified shell models (Behrens etal. 2014) and/or clumpy ISM models (Gronke & Dijkstra2014), where the Ly α escape fraction varies by only a fac-tor of order unity and thus do not provide the variationsrequired to explain the observed flux ratio. However, thetheoretical studies have considered systems with ∼ kpcextents, and ignored observational aperture effects. Theanisotropic escape of Ly α is strongest if the source is“shielded” by a gas cloud of similar or bigger size than theemitting region (in this case few tens parsec), and weak-ened by photons scattering back into the line-of-sight.Therefore, having a very compact source as in this workis likely to enhance the directional variations, but furtherstudies are needed to explore this possibility. Observa-tional aperture effects are not an issue in the present case,given the wide MUSE FoV. The strong lensing effect canin principle modulate the spatial appearance of the Ly α emission, however in the present case no Ly α emissionis observed on top of the stellar ultraviolet continuum ofimages ID14a,b and ID14c. Plausibly, ID14 would ap-pear as a Ly α -emitter if observed from a different view-angle and would require a variation of the Ly α intensityof two orders of magnitudes. A possibility is that even asmall amount of dust extinction in a region enshroudedby neutral gas (or a partially neutral IGM) may explainthe strong depression of the Ly α line. It is also worthnoting that the C iv λ , > − , see also Vanzella et al.2016a), further supporting an efficient Ly α attenuation. A spatially offset ultra-faint Ly α emission As discussed in the previous section, ID14 is a systemwith prominent high-ionization lines with energetic pho-tons reaching at least 4 Ry. The deficiency of the Ly α emission at the spatial position where the metal lines andesolving parsec-scale star-forming regions at z=3.22 9 Fig. 8.—
A diagnostic diagram separating AGN and star-formingpowered sources with superimposed object ID14 and a similar ob-ject described in Vanzella et al. (2016a). Predictions are from pho-toionization models of Feltre et al. (2016) (AGNs, gray filled circles)and Gutkin et al. (2016). Star-forming galaxies are color-coded ac-cording to their gas phase metallicity, Z gas , as defined in Gutkinet al. (2016). the stellar continuum arise suggests an efficient attenua-tion by dust and/or gas. However, such process could benot so efficient in other directions. In the case of a non-uniform covering factor of the neutral gas we may expectto recover the diffused Ly α photons in other directionsand/or observe spatially-offset fluorescent Ly α emission(Furlanetto et al. 2005; Weidinger et al. 2005). Indeed,we detected an offset and spatially-extended Ly α emis-sion (S/N=7.4) at the same redshift as ID14 (separatedby (cid:39) . α emission lies across the critical line pro-duced by the galaxy cluster at the given redshift. Theposition of the critical line (the highest magnification)is strongly supported by the confirmed images of ID14,“image1” and “image2” (Figure 1, panel B). This alsomeans the Ly α is strongly magnified and consequentlyhighly spatially distorted (see Figure 4, magenta ellipse).The observed emitting region (6.7 arcsec ) correspondsto de-lensed 7.6 kpc and a de-lensed integrated flux of1 . × − erg s − cm − if we adopt a median magnifi-cation µ = 50 (calculated within the same region). How-ever the real physical size and shape of the emission is af-fected by large uncertainties, especially where the magni-fications formally diverge. This spatially offset Ly α sug-gests the possible presence of a neutral gas cloud at a dis-tance of 2.1 kpc from ID14, reached by the ionizing pho-tons escaped from the system, where they can produceLy α fluorescence (Mas-Ribas & Dijkstra 2016; Behrens etal. 2014). The emerging spectral line profile apparentlysymmetric and narrow ( < − ) would also supportthe fluorescence scenario. It is worth noting that ID14shows optical Oxygen line equivalent widths and ratiossimilar to the ones of the Lyman continuum emitter dis-covered at z=3.2, named Ion2 (with confirmed > and some local extreme GreenPeas sources (also Lyman continuum leakers, Izotov etal. 2016; Schaerer et al. 2016). A possible link betweenthe large O32 index and the optically thin interstellarmedium to ionizing radiation was explored with photo-ionization models by Jaskot & Oey (2013). The largevalue observed here (O32 >
10) may indicate a density-bounded condition (due to a deficient [O ii ] λ , α nebula might reside into an elongated cavity directly con-necting it with ID14. We would also expect a large es-cape fraction of Ly α photons along the same transversedirection (Behrens et al. 2014).Alternatively, such Ly α emission might be produced byadditional faint/undetected sources. We identified a faintobject behind the diffuse Ly α emission (see the orange-dashed circle in Figure 4, panel C), detected in the HSTbands with an average observed magnitude of (cid:39)
30. Thefaintness of the object prevents us from deriving any re-liable photometric redshift or physical properties. Weonly note from Figure 9 an apparent drop in the F435Wand F606W bands when compared to the other redderbands. If this object (or another undetected source) isresponsible for the entire Ly α diffuse emission, then itsde-lensed magnitude would be (cid:38)
34 ( µ (cid:38)
50) with anEW(Ly α ) > (cid:48)(cid:48) × (cid:48)(cid:48) centered onthe object, orange circle in Figure 9). This would corre-spond to an absolute magnitude of M UV = −
11 and thevery large Ly α equivalent width would imply the pres-ence of unusual stellar populations (e.g., Schaerer 2003;Dijkstra & Wyithe 2007). Since the redshift of this ob-ject is unconstrained and/or we don’t know if other un-detected sources were present, more specific analysis willneed deeper imaging with future instrumentation such asJWST or ELT. CONCLUSIONS
We presented a strongly lensed pair of super star-clusters at z = 3 . ∼ m (cid:39) . iv λ , ii λ iii ] λ , iii ] λλ , β ,[O iii ] λλ , −
31, allof them showing narrow velocity dispersions, σ v < − (derived from MUSE) or σ v (cid:46) − (derivedfrom X-Shooter). The presence of high-ionization lines(like He ii λ Though more massive ( < M (cid:12) ) and larger ( (cid:39) Ion2 shows an equivalent width of the [O iii ] λ > > iii ] λλ , ii ] λ , Fig. 9.—
HST/ACS and WFC3 cutouts (1 . (cid:48)(cid:48) wide) centered on the diffuse Ly α emission marked with the magenta ellipse. The dashed-orange circle marks the position of the faint m ∼
30 (or m >
34 de-lensed) underlying object if placed at the same redshift of the Ly α emission. the observed colors (SED-fitting) suggest young stars arepresent ( <
10 Myr), as well as a relatively evolved stellarcomponent ( <
100 Myr). A moderate dust attenuation(E(B-V) (cid:39) .
06) is also consistent with the observed col-ors. The inferred stellar masses are of a few 10 M (cid:12) witha metallicity 1/10 solar.The main results can be summarized as follow:1. Despite the fact that the above properties are oftenobserved in Ly α emitting galaxies such as GreenPeas, (e.g., Henry et al. 2015), ID14 does not showany Ly α emission aligned with the detected stellarcontinuum and ultraviolet metal lines, in all theobserved multiple images ID14a,b,c. In particularthe ratio C iv λ , α (cid:38)
15 is uncommon,and it is the first case confirmed at high redshfitand very low-luminosity regimes as discussed here.A relatively low velocity outflow, the presence of ascreen of neutral gas and the presence of an (evenmoderate) amount of dust are all ingredients thatmay explain the Ly α attenuation.2. A spatially offset and strongly magnified Ly α emis-sion is detected at ∼ α photons along the same route,see, e.g., Behrens et al. 2014) and/or (B) an in-situstar formation activity of an object barely (or not)detected in the deep HST images, with de-lensed M UV (cid:39) −
11 ( m >
34) and EW(Ly α ) > µ = 40 ±
7, offers the opportunity to anticipatethe expected future ELT capabilities in terms of the spa-tial resolution and magnitudes in the not lensed fields(see also, Christensen et al. 2012; Vanzella et al. 2016a). In particular, besides the detected pair of super star clus-ters (components “1,2”), additional sub-structures havebeen identified possibly representing single H ii regionsand/or candidate proto-globular clusters (Vanzella et al.2017b). More investigation is needed to fully characterizethese systems. The spatial resolution (3-7 mas) achiev-able with the ELT instrumentation in conjunction withstrong lensing (as in this case) will offer the opportunityto spatially resolve details of 1-2 pc at z >
3. It is alsoworth reporting what integration time would be neededwith MUSE to achieve the same S/N ratios of UV metallines for a ID14-like not lensed object. The most promi-nent among them is the C iv λ . × − erg s − cm − that would require (cid:39) α / C iv ratio often reported in litera-ture (e.g., Stark et al. 2014, 2015; Mainali et al. 2017) orany line ratio that includes the Ly α flux (e.g., Figure 7in Shibuya et al. 2017) can be affected by large line-of-sight variation of the Ly α visibility also in low-luminosityregimes, therefore weakening its use as a diagnostic fea-ture (e.g., Smit et al. 2017).We thank the referee for useful comments that im-proved the manuscript. We thank K. Schmidt, L. Mas-Ribas and R. Amorin for useful discussion. C.G. ac-knowledges support by VILLUM FONDEN Young Inves-tigator Programme through grant no. 10123. K.C. ac-knowledges funding from the European Research Councilthrough the award of the Consolidator Grant ID 681627-BUILDUP. LC is supported by Grant DFF 4090-00079.Based on observations collected at the European South-ern Observatory for Astronomical research in the South-ern Hemisphere under ESO programmes P095.A-0653,P094.A-0115 (B) and ID 094.A-0525(A). MM, AM andPR acknowledge the financial support from PRIN-INAF2014 1.05.01.94.02. REFERENCESAlavi, A., Siana, B., Richard, J., et al. 2016, arXiv:1606.00469Amor´ın, R., P´erez-Montero, E., Contini, T., et al. 2015, A&A, 578,A105Amor´ın, R., Fontana, A., P´erez-Montero, E., et al. 2017, NatureAstronomy, 1, 0052Annibali, F., Tosi, M., Pasquali, A., et al. 2015, AJ, 150, 143 Atek, H., Schaerer, D., & Kunth, D. 2009, A&A, 502, 791Bacon, R., Accardo, M., Adjali, L., et al. 2010, Proc. SPIE, 7735,773508Behrens, C., Dijkstra, M., & Niemeyer, J. C. 2014, A&A, 563, A77Brocklehurst M., 1971, MNRAS, 153, 471 esolving parsec-scale star-forming regions at z=3.22 11esolving parsec-scale star-forming regions at z=3.22 11