The Curious Case of Lyman Alpha Emitters: Growing Younger from z ~ 3 to z ~ 2?
Viviana Acquaviva, Carlos Vargas, Eric Gawiser, Lucia Guaita
aa r X i v : . [ a s t r o - ph . C O ] M a y Astrophysical Journal Letters, in press
Preprint typeset using L A TEX style emulateapj v. 5/2/11
THE CURIOUS CASE OF LYMAN ALPHA EMITTERS:GROWING YOUNGER FROM z ∼ z ∼ Viviana Acquaviva , Carlos Vargas , Eric Gawiser , Lucia Guaita ABSTRACTLyman Alpha Emitting (LAE) galaxies are thought to be progenitors of present-day L ∗ galaxies.Clustering analyses have suggested that LAEs at z ∼ z ∼
2, but itis unclear whether the physical nature of these galaxies is compatible with this hypothesis. Severalgroups have investigated the properties of LAEs using spectral energy distribution (SED) fitting, butdirect comparison of their results is complicated by inconsistencies in the treatment of the data andin the assumptions made in modeling the stellar populations, which are degenerate with the effects ofgalaxy evolution. By using the same data analysis pipeline and SED fitting software on two stackedsamples of LAEs at z = 3 . z = 2 .
1, and by eliminating several systematic uncertainties thatmight cause a discrepancy, we determine that the physical properties of these two samples of galaxiesare dramatically different. LAEs at z = 3 . ∼ Z < . Z ⊙ ), while LAEs at z = 2 . ∼
50 Myr) and metal-rich (
Z > Z ⊙ ).The difference in the observed stellar ages makes it very unlikely that z =3.1 LAEs evolve directly into z =2.1 LAEs. Larger samples of galaxies, studies of individual objects and spectroscopic measurementsof metallicity at these redshifts are needed to confirm this picture, which is difficult to reconcile withthe effects of 1 Gyr of cosmological evolution. INTRODUCTION
Lyman Alpha Emitting (LAE) galaxies have beenshown to be building blocks of Milky-Way type galaxiestoday (Gawiser et al. 2007, Guaita et al. 2010). Thebrightness of the Lyman Alpha line allows one todetect these galaxies using the narrow-band techniqueeven when the continuum is faint. As a result, largesamples of LAEs have been studied using SED fittingat a variety of redshifts ( e.g.,
Gawiser et al. 2006,Nilsson et al. 2007, Gawiser et al. 2007, Lai et al.2007, Pirzkal et al. 2007, Finkelstein et al. 2007,Lai et al. 2008, Pentericci et al. 2009, Ono et al. 2010a,Ono et al. 2010b, Ouchi et al. 2010, Nilsson et al. 2011,Guaita et al. 2011, Finkelstein et al. 2011b). Thepicture emerging from these collective studies is far fromhomogeneous. While LAEs were once thought to beyoung, dust-free galaxies experiencing their first episodeof star formation, many of these investigations suggesta large spread in their physical properties, and findevidence of an older, more evolved, and dustier stellarpopulation component.An important caveat in the interpretation of re-sults comes from the systematic uncertainties intro-duced by inconsistent modeling of the stellar populationsamong different groups. This issue was studied in de-tail by Acquaviva et al. (2011), where we showed thatthe use of BC03 (Bruzual & Charlot 2003) and CB11(Charlot & Bruzual 2011) stellar population synthesis(SPS) templates, at Solar or variable metallicity, andwith or without including nebular emission, gave rise toa scatter in the estimate of age and mass significantlylarger than the statistical uncertainty for the same data.Further scatter might be created by the use of a different Department of Physics and Astronomy, Rutgers, The StateUniversity of New Jersey, Piscataway, NJ 08854 Institutionen f¨or Astronomi, Stockholms Universitet, SE-10691 Stockholm, Sweden initial mass function (IMF) or of inconsistent statisticalestimators of the physical properties of galaxies, such asthe best-fit parameters as opposed to the mean of theprobability distribution employed by Bayesian statistics. OBJECTIVES, METHODOLOGY, AND DATA
This paper focuses on one particular puzzle, the agesof Lyman Alpha Emitters, and one particular redshiftrange, the Gyr between z ∼ z ∼
2. LAEs havebeen mostly studied through stacking analyses, becausetheir faint continua make it hard to fit the broad-bandphotometry of each galaxy, although Ono et al. (2010a)and Nilsson et al. (2011) have also attempted to fit theSEDs of bright individual objects at these redshifts. At z ∼
3, Gawiser et al. (2007) and Lai et al. (2008) (here-after L08) examined the SEDs of average-stacked LAEs,segregated according to their detection in the IRAC3.6 µ m band, and found that both samples were essen-tially dust-free, and that LAEs Undetected in IRAC werefairly young (age ∼ ∼ < z ∼
2, Guaita et al. (2011) revealed the LAEsto be young (age < µ m band lu-minosity. Nilsson et al. (2011) reported finding a young(age < ∼ z = 2 . z ∼ z ∼ F l u x d e n s i t y ( µ J y ) data, z = 3.1 (x 2.05)data, z = 2.1 F l u x d e n s i t y ( µ J y ) data, z = 3.1Best Fit, z = 3.1 λ rest frame (˚ A ) F l u x d e n s i t y ( µ J y ) data, z = 2.1Best Fit, z = 2.1 Fig. 1.—
Data points and best fit spectra for the full LAE sam-ples. The black diamonds show the flux predicted by the best-fitmodel in each band. The Ly − α line has been subtracted from thephotometry and is not included in the templates used to fit theSED. In the top panel, data at z = 3 . ∼ µ m is typical of low-metallicity, old stellarpopulations. pipeline to compose the stacked samples and employingthe same, state-of-the-art algorithm to analyze the SEDs.This procedure will discriminate between physical evo-lution of LAEs and artificial differences introduced byinconsistencies in building the stacks and modeling thestellar populations.The LAE stacks at z = 2 . z = 3.1) that comprises70 LAEs from Gronwall et al. (2007). At both redshifts,LAEs are selected based on rest-frame Ly α equivalentwidth >
20 and on the flux in the Ly- α line. The fluxlimits used in the two surveys lead to very similar rest-frame Ly- α luminosity limits, making the samples at z = 3 . z = 2 . z = 3 .
1, we have UBVRIzJKdata from MUSYC (Gawiser et al. 2006) and IRAC pho-tometry, while at z = 2 . ∼ gawiser/MUSYC/data.html. RESULTS
The SEDs of the median-stacked full samples at red-shifts z = 3 . z = 2 . Z , as discussed in Acquaviva et al.2011). We use the latest CB11 models, use the Calzettilaw (Calzetti et al. 2000) to compute the attenuation,assume constant star formation history, favored by theprevious analyses of L08 and Gu11, and adopt a SalpeterIMF. We include the contribution of nebular emission asdescribed in Acquaviva et al. (2011), following the pro-cedure outlined by Schaerer & de Barros (2009). Resultsfrom SED fitting are shown in Figs. 2 and 3 and sum-marized in Table 1.Unsurprisingly, given the different shapes of theirSEDs, we find strong differences in the properties of z = 3 . z = 2 . ∼ . × M ⊙ ,and are significantly older, with mean ages of 1 Gyr, thantheir lower-redshift counterparts, which present a verymoderate amount of dust, appear to have Solar metallic-ity, have mean masses of ∼ × M ⊙ , and have meanages of 45 Myr.Results for the full, IRAC-bright, and IRAC-dimstacked samples of LAEs at z = 2.1 can be compareddirectly to the ones for the same stacks of Gu11; themain difference is the lower amount of dust (E(B-V) ∼ Sample z Notes Z/Z ⊙ Age (Gyr) E(B-V) M ∗ (10 M/M ⊙ ) best fit χ /d.o f.full 3.1 0.025 [0.005-0.04] 0.98 [0.84-2.0] 0.019 [0.-0.022] 16 [13-19] 11.9/8full 3.1 fixed Z Z Z Z Z Z TABLE 1Mean expectation values and 68% credible intervals from SED fitting for the six different samples considered in thiswork. For the two-population fit, we report the mean value of the total mass and the ratio between the mass in the oldstellar population and the total mass and its uncertainty in parentheses.
L og (A ge ) P o s t e r i o r p r o b a b ili t y E (B - V ) P o s t e r i o r p r o b a b ili t y −2 −1 0 L og (Z /Z ⊙ ) P o s t e r i o r p r o b a b ili t y z = 2 .1z = 3 .1 L og (M ∗ /M ⊙ ) P o s t e r i o r p r o b a b ili t y Fig. 2.—
Marginalized probability distributions for the SED fitting parameters for z = 3 . z = 2 . µ m bands by a factor of 1.8, and in part from allowingmetallicity to vary and finding Z/Z ⊙ < .
2. The quotedL08 results assumed Solar metallicity, although the au-thors reported, consistent with our findings, that theirage estimate for the Undetected population increased if Z = 0 . Z ⊙ was used. THE MYSTERY OF THE AGES OF LAES
The evolutionary picture of LAEs between z ∼ z ∼ z ∼ Data quality tests
We begin by examining in detail the stacked SEDs ofFig. 1. At z = 3 .
1, the best-model has a very reasonablereduced χ of 1.4. This is a desired property because thestacking procedure relies on the strong assumption that“the typical LAE” exists, and that the artificial SEDbuilt by stacking many real SEDs is a reasonable rep-resentation of the spectrum of this typical LAE galaxy.A reasonable reduced χ value suggests that the boot-strap uncertainties correctly account for the spread inthe SEDs of LAEs and does not illuminate a templateincompleteness problem. Nonetheless, there is a relevantamount of scatter in the stacked SED, with three pointsparticularly standing out, the low measurements in the J and IRAC 5.8 µ m band, and the high measurement inthe K band. We re-did the SED fitting excluding each ofthem, and each pair, in turn, and we found that our re-sults are extremely robust to the exclusion of these datapoints, with changes in the parameter estimates by lessthan 0.2 σ . We also note that the results at z = 3 . α systems by Prochaska et al. (2003).At z = 2 .
1, the SED visually appears to be smoother;however, its reduced χ has a slightly higher value of2.2. This feature suggests that the stacking at this red-shift might be an overly aggressive compression of theinformation contained in the individual SEDs. Breakingdown the sample in IRAC-bright and IRAC-Dim LAEsin the attempt to increase the homogeneity of the stacksdoes not afford a better fit, although it shows that theproperties of these two stacks follow closely the ones ofthe full sample. Metallicity
The SED fitting parameters are correlated with eachother, and in particular, there is a well-know age-metallicity relation that can be seen easily as the axisof the 2-D credible region in the age-metallicity plane ofFig. 3. Our estimates of metallicity depend on the SPStemplates we use and (although weakly) on the flat prioron log Z that we assume. We think that the latter is Acquaviva et al. L og (A ge ) L og ( Z / Z ⊙ ) z = 3. 1 L og (A ge ) L og ( M ∗ / M ⊙ ) z = 3. 1 L og (A ge ) E ( B - V ) z = 3. 1 L og (Z /Z ⊙ ) L og ( M ∗ / M ⊙ ) z = 3. 1 −2 −1.5 −1 −0.599.19.29.39.4 L og (A ge ) L og ( Z / Z ⊙ ) z = 2. 1 L og (A ge ) L og ( M ∗ / M ⊙ ) z = 2. 1 L og (A ge ) E ( B - V ) z = 2. 1 L og (Z /Z ⊙ ) L og ( M ∗ / M ⊙ ) z = 2. 1 −0.5 0 0.58.28.48.68.89 Fig. 3.— physically motivated, and we have no reason not to trustthe models; however, it is possible that the observed dif-ference in metallicity between LAEs at the two redshiftis overestimated. If the true metallicity at z = 3 . z = 2 . Z = 0 . Z ⊙ ; this value issuggested by the spectroscopic analysis of three LAEs at z ∼ . Star Formation History
The assumption of a particular functional form for thestar formation history of LAEs can also influence thedetermination of ages. We investigate this issue by re-peating the SED fitting procedure on the full samplesusing exponentially increasing and decreasing star for-mation histories (SFHs), ψ ( t ) ∝ e ± t/τ . By sampling in1 /τ , both increasing and decreasing SFHs are exploredas part of a contiguous parameter space, as explainedin Acquaviva et al. (2011). This parametrization is alsoable to capture constant or starburst SFHs for appro-priate values of τ /age. The results of this SED fittingrun (where the metallicities were held fixed at the best-fit value for both samples) are shown in Fig. 4; in bothcases the estimates of ages do not change significantly.The quality of the fit does not improve significantly ineither case by allowing this larger range of star forma- tion histories, as can be seen in Table 1. The data favora nearly constant SFH at both redshifts, with a best-fitvalue of τ around 4 Gyr for both samples. Gu11 hadalso performed the SED fitting using these SFHs for the z = 2 . A hidden old stellar population
A possible explanation of the difference in the phys-ical nature of LAEs at z = 3 . z = 2 . e.g., Finkelstein et al. 2011; Lee 2010 and references therein).For z = 2 . z = 2 .
1, this test is meant to reveal an olderstellar population; at z = 3 .
1, it may allow us to sin-gle out two components of different ages, although we donot expect to find a large component of very young stars,since they would dominate the rest-frame UV part of thetotal spectrum. In the case of z = 3 . χ r = 1.4 to χ r = 1.0). Young agesfor the primary stellar population are allowed, but onlywhen the fraction of mass in this stellar population issmall, as seen in the third panel of Fig. 4. For example,ages of the order of 100 Myr are allowed at 68% confi-he curious case of Lyman Alpha Emitters 5 L og ( Age ) P o s t e r i o r p r o b a b ili z = 3. 1 z = 2. 1L og ( Age ) P o s t e r i o r p r o b a b ili L og ( Age ) M O S P / M Y S P z = 2 . 1 L og ( Age ) M O S P / M Y S P z = 3 . 1 Fig. 4.—
Left two panels : Variation in the posterior probability distribution of age of the full samples, for different assumptions describedin the text. The black (solid, thick) line is the reference case ran with CSF and varying metallicity. The red (dotted-dashed) line is for Z fixed at 0 . Z ⊙ , the green (dashed) line corresponds to exponential star formation, and the magenta (solid, thin) line is the age of theprimary population when a second, 1 Gyr old SSP is added. Right panels:
Probability distribution of the fraction of mass in the old stellarpopulation versus the age of the primary population. For the z = 3.1 LAEs, a young stellar component is allowed only when most of themass is in the old stellar population. dence when the mass in young stars is less than 10% ofthe total stellar mass. As expected since the two stellarpopulations are so similar, the mass ratio in the two pop-ulations is unconstrained. For z = 2 . ∼ <
50% at 2 σ confidence for allcases considered. We conclude that the two-componentscenario cannot be confirmed by analyzing the stackedSEDs, although the improvement in the χ at z = 3.1might indicate a heterogeneity of the stellar populations,masked by the stacking process. CONCLUSIONS
The clustering analysis of LAEs (Guaita et al. 2010)has shown that LAEs at z ∼ z ∼ z = 3 . z = 2 .
1, which appear to be several timesyounger and more metal-rich than their higher-redshiftcounterparts. If these samples are a fair representation of LAEs at each redshift, this result directly rules out thehypothesis that z=3.1 LAEs evolve into z=2.1 LAEs, im-plying that the LAE phase lasts significantly less than 1Gyr. Moreover, such stark differences between emission-line selected galaxy samples become difficult to explainas the effect of 1 Gyr of cosmological evolution.We find hints that stacking analysis, meant to provideinsights on the nature of the typical
LAE galaxy, mightbe overly aggressive and perhaps unable to capture themulti-component nature of LAEs at each redshift. Tobe properly investigated, these issues require constraintsfrom SED fitting of large samples of individual objects,as well as spectroscopic constraints on the metallicity ofLAEs. The former have been especially challenging toacquire so far, because of the lack of deep data in theobserved NIR region of the spectrum. This gap is beingfilled by deep YJH HST observations by the Cosmic As-sembly Near-infrared Deep Extragalactic Survey (CAN-DELS, Grogin et al. 2011; Koekemoer et al. 2011), andwe plan to address this issue by using CANDELS datain the MUSYC fields in a subsequent paper (Vargas etal 2012, in preparation).It is a pleasure to thank Kamson Lai for advice onthe photometry at z = 3.1, the Astrophysics group atRutgers, in particular Saurabh Jha and Curtis McCully,and the anonymous referee, for useful conversations.This research was supported in part by the AmericanAstronomical Society’s Small Research Grant Program. REFERENCESAcquaviva, V., Gawiser, E., & Guaita, L. 2011, ApJ, 737, 47Bruzual, G. & Charlot, S. 2003, MNRAS, 344, 1000Calzetti, D. et al. 2000, ApJ, 533, 682Charlot, S. & Bruzual, G. 2011, private communication.Ciardullo, R., Gronwall, C., Wolf, C., et al. 2012, ApJ, 744, 110Finkelstein, K. D., Papovich, C., Finkelstein, S. L., et al. 2011,ApJ, 742, 108Finkelstein, S. L., Cohen, S. H., Moustakas, J., et al. 2011b, ApJ,733, 117Finkelstein, S. L., Hill, G. J., Gebhardt, K., et al. 2011c, ApJ,729, 140Finkelstein, S. L., Rhoads, J. E., Malhotra, S., Pirzkal, N., &Wang, J. 2007, ApJ, 660, 1023Gawiser, E., Francke, H., Lai, K., et al. 2007, ApJ, 671, 278 Gawiser, E., van Dokkum, P. G., Herrera, D., et al. 2006a, ApJS,162, 1Grogin, N. A., Kocevski, D. D., Faber, S. M., et al. 2011, ApJS,197, 35Gronwall, C. et al. 2007, Astrophys. J., 667, 79Guaita, L., Acquaviva, V., Padilla, N., et al. 2011 ApJ, 733, 114Guaita, L., Gawiser, E., Padilla, N., et al. 2010, ApJ, 714, 255Koekemoer, A. M., Faber, S. M., Ferguson, H. C., et al. 2011,ApJS, 197, 36Lai, K., Huang, J., Fazio, G., Cowie, L. L., Hu, E. M., & Kakazu,Y. 2007, ApJ, 655, 704Lai, K. et al. 2008, Astrophys. J., 674, 70Lee, S.-K. J. 2010, PhD thesis, The Johns Hopkins UniversityLewis, A. & Bridle, S. 2002, Phys. Rev., D66, 103511