Metal Deficiency in Two Massive Dead Galaxies at z ∼ 2

Stellar chemical compositions are a powerful means to link galaxies across cosmic time and therefore understand their evolution. Massive early-type galaxies in the local universe exhibit solar or super-solar stellar metallicities (Z_*; e.g., Thomas et al. 2005). This holds beyond z ∼ 0—previous observations report similar values in massive dead galaxies up to z ∼ 1.6 (e.g., Gallazzi et al. 2014; Lonoce et al. 2015; Onodera et al. 2015), indicating that either they formed from metal-enriched gas or retained a lot of their metals after forming from pristine gas.

These observations spanning ∼10 Gyr seem to fit in a recently popular evolution scenario for massive galaxies, where dry minor merging might grow galaxy sizes without adding much mass (e.g., Naab et al. 2009), explaining the observed size evolution by a factor of ≳3 from z ∼ 2 to 0 (e.g., Szomoru et al. 2012; van der Wel et al. 2014). As these galaxies are already dead and near the mass of local ellipticals (e.g., Stiavelli et al. 1999; Daddi et al. 2005; Glazebrook et al. 2017), they are believed to evolve without further substantial episodes of in situ star formation over the rest of their lives. If so, the main z ∼ 2 progenitor must already have at least its z = 0 descendant's metallicity: the low-mass satellites that it will accrete are probably metal poor (Gallazzi et al. 2005; Kirby et al. 2013), and thus can only dilute the final system's total metallicity (see also Choi et al. 2014).

The above scenario, however, may not be the complete picture. Other works (e.g., Newman et al. 2012; Nipoti et al. 2012; Sonnenfeld et al. 2014) show that minor mergers alone cannot accomplish the necessary size evolution. Furthermore, recent stellar metallicity estimates at z ≳ 2 yield values ∼2.5σ below the local M_*–Z_* locus (Kriek et al. 2016; Toft et al. 2017), adding to arguments for more-than-minor-merging growth. To clarify the picture, we must take steps toward a more complete galaxy census.

Here we report two additional massive ($\mathrm{log}{M}_{* }/{M}_{\odot }\sim 11$) dead galaxies at z ∼ 2.2 that appear substantially under-enriched. Both galaxies have typical A-type spectra, but $\mathrm{log}{Z}_{* }/{Z}_{\odot }\,=-0.40\pm 0.02$ and −0.49 ± 0.03, respectively. With the other spectrophotometric results at z ∼ 2, these galaxies suggest something other than dry minor merging is required to explain the subsequent evolution of some fraction of z ≳ 2 passive galaxies. While results based on a larger sample will be presented in a future study, this Letter reports the discovery of these two galaxies and briefly discusses its implications. Throughout, magnitudes are quoted in the AB system assuming Ω_m = 0.3, Ω_Λ = 0.7, H₀ =70 km s⁻¹ Mpc⁻¹.

2.1. Hubble Space Telescope (HST) Data

2.2. K_s and Infrared Array Camera (IRAC) Photometry

2.3. Sample Galaxies

With extensive multi-band photometry and deep NIR spectra, we deconstruct the stellar populations of the two high-z passive galaxies. A forthcoming paper presents full details of the SED fitting, but our scheme was established and tested by Kelson et al. (2014) using a similar data set at z ≲ 1. The core of the method is to find the best combination of amplitudes, {a_{i=[0,1,2,3,4]}}, for a set of stellar templates of different ages, {t_{i=[0,1,2,3,4]}}, that matches the data. This flexibility avoids systematic biases in, e.g., stellar mass that can arise when using functional forms for star formation histories.

Our baseline results adopt the Flexible Stellar Population Synthesis code (FSPS; Conroy et al. 2009; Conroy & Gunn 2010) to generate four simple stellar population (SSP) templates with ages of [0.01, 0.1, 0.5, 1.0, 3.0] Gyr (roughly O, B, early/late-A, and F stars), based on Padova isochrones and the MILES stellar library. We assume a Salpeter IMF and Calzetti dust law. Our main results are robust to those assumptions except for the choice of isochrones, which systematically affects one system (see below). The number and choice of templates was set to fully exploit the data based on result reproducibility tests using simulations. Assuming non-zero durations of star formation for each template—e.g., exponentially declining star formation histories—pushes systems to older ages, and thus lower metallicities to compensate for the subsequently redder SEDs. Thus, SSPs produce conservative results for our purposes.

The interpolation scheme in the FSPS python module generates each template at metallicities spanning $\mathrm{log}{Z}_{* }/{Z}_{\odot }\,\in [-1.0,0.5]$ in increments of 0.05 dex. Metallicity and extinction are fixed across all five SSP templates; i.e., the fits have 5 + 1 + 1 free parameters.

We use a Monte Carlo Markov Chain to find the best combination of a_i, A_V, Z_*, and their covariances in the parameter ranges (${a}_{i}\in [0,200]$ and ${A}_{V}\in [0,4]\,\mathrm{mag}$) with flat priors. We set the number of walkers to 200 and use N_mc = 10000 realizations. Redshifts determined beforehand via χ²-minimization (and visual inspection of absorption lines) are fixed during parameter estimation. Final results are drawn from the last N_mc/2 steps to avoid biases from the initial exploration. Emission lines detected at >1.5σ across 3 pixels (∼135 Å) are masked from the best-fit template of each iteration and later fit with gaussians.

One caveat is that the metallicity determined above is not a direct measurement, as metallicity sensitive indicators, such as Fe and Mg, fall outside the G141 spectral range. The broadband-defined "Z_*" here is thus different from the total metallicity derived from absorption line measurements (see Section 5).

Figure 1 shows the data and fits and Table 1 summarizes the best-fit parameters. Both galaxies prefer sub-solar metallicities under the standard assumption of Padova isochrones: $\mathrm{log}{Z}_{* }/{Z}_{\odot }=-{0.40}_{-0.02}^{+0.02}$ and $-{0.49}_{-0.03}^{+0.04}$ for 00141 and 00227, respectively, with estimates and uncertainties reflecting the 16th/50th/84th percentile values.⁸ Also, both best-fit templates are dominated by ∼0.5-Gyr-old stellar populations: ∼99% of 00141's light is in the a₂-template, while 00227's stellar population is younger, with ≳44% in a₁ and a₂. Stellar masses are M_* = 1.8 × 10¹¹ M_⊙ and 5.8 × 10¹⁰ M_⊙, respectively. 00141 exhibits no significant emission, while 00227 shows very weak Balmer infilling (equivalent widths of Hδ ∼ 0.8 ± 0.2 Å and Hγ ∼ 1.8 ± 0.2 Å) and [O ii] emission (∼3.4 ± 0.7 Å), suggesting the latter was indeed quenched more recently. Both galaxies have compact/centrally concentrated morphologies with effective radii r_e = 1.8–2 kpc and Sérsic indices of n = 8. These values are typical for passive galaxies at similar redshifts (e.g., Szomoru et al. 2012), and confirm that significant dry minor merging is needed if they are to grow to match the size of z = 0 early-types.

Zoom In Zoom Out Reset image size

Table 1.  Summary of the SED Fitting Parameters

ID	R.A.	Decl.	zspec.	μ	$\mathrm{log}{Z}_{* }$	a0	a1	a2	a3	a4	AV	M*
	(degree)	(degree)			(Z⊙)	(10 Myr)	(100 Myr)	(500 Myr)	(1 Gyr)	(3 Gyr)	(mag)	(1010M⊙)
Padova Isochrone
00141	177.4034	22.4185	${1.967}_{-0.001}^{+0.002}$	${1.85}_{-0.11}^{+0.83}$	$-{0.40}_{-0.02}^{+0.02}$	${0.18}_{-0.13}^{+0.26}$	${0.22}_{-0.16}^{+0.40}$	${74.06}_{-4.85}^{+2.86}$	${1.47}_{-1.09}^{+2.07}$	${0.52}_{-0.38}^{+0.73}$	${1.79}_{-0.03}^{+0.02}$	${17.86}_{-0.54}^{+0.43}$
00227	177.4068	22.41624	${2.412}_{-0.004}^{+0.004}$	${1.68}_{-0.09}^{+0.37}$	$-{0.49}_{-0.03}^{+0.04}$	${0.71}_{-0.62}^{+1.77}$	${14.72}_{-1.89}^{+1.81}$	${14.10}_{-1.52}^{+1.59}$	${0.88}_{-0.67}^{+1.31}$	${1.37}_{-0.94}^{+1.22}$	${0.66}_{-0.07}^{+0.06}$	${5.78}_{-0.56}^{+0.68}$
MESA Isochrones and Stellar Tracks (MIST) Isochrone
00141	177.403	22.419	${1.967}_{-0.002}^{+0.001}$	${1.85}_{-0.11}^{+0.83}$	${0.10}_{-0.07}^{+0.07}$	${0.23}_{-0.17}^{+0.34}$	${0.31}_{-0.23}^{+0.52}$	${42.78}_{-5.59}^{+5.36}$	${2.79}_{-2.00}^{+3.28}$	${0.94}_{-0.70}^{+1.19}$	${1.37}_{-0.08}^{+0.08}$	${14.72}_{-1.14}^{+1.22}$
00227	177.407	22.416	${2.411}_{-0.002}^{+0.004}$	${1.68}_{-0.09}^{+0.37}$	$-{0.50}_{-0.06}^{+0.07}$	${0.67}_{-0.51}^{+1.10}$	${11.98}_{-2.13}^{+1.80}$	${17.75}_{-1.31}^{+1.09}$	${0.01}_{-0.01}^{+0.05}$	${0.40}_{-0.31}^{+0.62}$	${0.62}_{-0.08}^{+0.06}$	${5.39}_{-0.40}^{+0.46}$

Note. 50th percentile is quoted as the best-fit parameter with 16th/84th percentiles as statistical uncertainties. Systematic uncertainties derived from simulations (Section 4) are ∼0.25 dex, 0.25 dex, and 0.17 dex for A_V, $\mathrm{log}{Z}_{* }/{Z}_{\odot }$, and $\mathrm{log}{M}_{* }/{M}_{\odot }$, respectively. a_i: Amplitudes of the templates, with i = [0, 1, 2, 3, 4] representing t_i = [0.01, 0.1, 0.5, 1, 3]/Gyr. μ: Median of magnification by the foreground cluster. Magnification errors are not included in the parameter errors as all are μ-independent except for M_*.

Download table as:  ASCIITypeset image

Figure 2 shows the fit parameter covariances. While some covary strongly (e.g., a₂ and A_V), none significantly impact the metallicity assessment. Simulations of our SEDs (discussed momentarily) confirm these results.

Zoom In Zoom Out Reset image size

As we lack detailed spectra, the metallicities reported here might depend on the isochrones adopted during SED fitting. We tested all isochrones available in FSPS, and found that, while 00227's result is robust, Z_* changes significantly for 00141, reaching a low of −0.6 (PAdova & TRieste Stellar Evolution Code (PARSEC); Bressan et al. 2012) and a high of $\mathrm{log}{Z}_{* }/{Z}_{\odot }\sim 0.10\pm 0.07$ (MIST; Choi et al. 2016). This is mostly caused by the different treatments of the asymptotic giant branch, which contributes significantly to the galaxy's luminosity at ages near 00141's of ∼500 Myr. Because we cannot determine the best isochrone with the current data set, we base our discussion on the result with the standard Padova isochrone to facilitate contextualizing our results with those in the literature. However, we present the MIST results in Table 1 and related figures to illustrate the effect of this choice.

In addition to isochrone systematics, there is a chance that, due to dust or statistical fluctuations, an observed sub-solar SED could be measured from a galaxy that is truly enriched to solar levels. To estimate this contamination rate, we tested our fitting method on 1000 simulated SEDs covering the same spectrophotometric bandpasses at comparable S/N with parameters randomly chosen over the fitting range. While full results will be discussed in our forthcoming study, the simulation shows that the input and extracted estimates are unbiased up to A_V = 1.3 mag. Above that, our inferences underestimate A_V by up to 0.2 mag. If so, to return a sufficiently reddened SED, the fitter would tend to increase Z_*, returning an overestimate of the true metallicity, making our results conservative. Finally, we find ∼20% and ∼5% chances that 00141's $\mathrm{log}{Z}_{* }/{Z}_{\odot }=-0.4$ and 00227's $\mathrm{log}{Z}_{* }/{Z}_{\odot }=-0.49$, respectively, could have come from a galaxy with truly solar metallicity.

The sub-solar abundances found in this study, suggest that these galaxies are in the early stages of chemical evolution, if already quite mature in terms of mass. While perhaps not surprising given the age of the universe (e.g., De Lucia & Blaizot 2007), these results do differ from the solar to super-solar values reported previously for massive quiescent galaxies at lower redshifts (Gallazzi et al. 2005, 2014; Choi et al. 2014; Onodera et al. 2015) as summarized in Figure 3. (Gallazzi et al. 2014 and Leethochawalit et al. 2018, however, show Z_* evolution at z < 0.7.)

Zoom In Zoom Out Reset image size

Now, Z_* here derives from broad SED features. Given the complexities of stellar evolution, it is therefore perhaps not easily interpreted, e.g., as reflecting iron or total (iron + α) abundances. This fact complicates comparisons to other measurements based on highly detailed spectroscopy. As such, in those cases (Choi et al. 2014; Lonoce et al. 2015; Onodera et al. 2015; Kriek et al. 2016) we show in Figure 3 both [Fe/H] and the total abundance, [Z/H], where [Z/H] = [Fe/H] +0.94 [α/Fe] (Thomas et al. 2003). Other results reported only as "$\mathrm{log}{Z}_{* }/{Z}_{\odot }$" are reproduced as published (Gallazzi et al. 2005, 2014; Toft et al. 2017).

Assuming that our two galaxies do not represent an outlier population, the differences illustrated in Figure 3 raise interesting questions about the diversity of evolutionary paths of these objects to z ∼ 0. What can be securely said is that the popular scenario for high-z passive galaxy evolution does not hold for these objects: while it may grow their sizes sufficiently, dry minor merging cannot correspondingly raise their metallicities to those of their z = 0 counterparts. Given their low mass, any satellites are almost certain to have lower Z_* than these centrals (Pasquali et al. 2010; Kirby et al. 2013), so such merging would only further dilute these systems' metal content. Furthermore, most metallicity gradients in local early-types are negative—or at least flat—with higher metallicity in their centers (e.g., Martin-Navarro et al. 2018). If the objects studied here were to somehow merge with more metal-rich satellites, their z = 0 gradients would go in the opposite direction (Nipoti et al. 2012).

Given this boundary condition, there appear to be three plausible routes for our galaxies to take.

• (1)  
  Major mergers: Merging with another ∼equal-mass galaxy of Z_* ≳ Z_⊙ would bring them toward/onto the local mass–metallicity relation. While major mergers are rare, globally, and probably too rare to happen to all massive galaxies (e.g., Bundy et al. 2009), given the amount of time from the observed redshift to today and the increasing merger rate increases toward higher-z (e.g., Lotz et al. 2011), this route cannot be excluded here. If this is taken as the preferred solution, measurements like this may provide independent tests of the merger rate once sufficient samples exist.
• (2)  
  A second burst of star formation. Newly accreted gas delivered either by wet satellite galaxies or cosmic accretion can raise metallicities at later times via in situ star formation. Dissipation efficiently supplies the gas to the central part of the system and induces starbursts. Despite uncertainties in the metallicity of the inflowing gas and the size of these bursts—perhaps ∼10%–15% of the systems' extant stellar mass (e.g., MacArthur et al. 2008)—the large numbers of minor mergers in massive systems since z ∼ 2 (e.g., Rodriguez-Gomez et al. 2015) make this scenario attractive, as do observations of blue core early-types (e.g., Treu et al. 2005), and other metrics of minor merger-triggered star formation at intermediate redshift (e.g., Kaviraj et al. 2011). The centrally concentrated nature of star formation in this scenario would also produce negative metallicity gradients, qualitatively consistent with those in the local universe. However, the increase of the central density would result in smaller galaxy size, contradicting the observed size evolution.
• (3)  
  The galaxies remain as they are. If this is the case, their descendant systems must be outliers today, suggesting most of the size and metallicity evolution of the passive population is due purely to progenitor bias. The situation is unclear at present: Poggianti et al. (2013) found local compact galaxies (size outliers) to have younger light-weighted ages, which could also reflect low metallicities. Other studies, however, find about solar values for compact local relics (e.g., Martin-Navarro et al. 2018). More spectroscopic studies of local compact galaxies will help resolve this picture further.

We identify two galaxies with SEDs that imply substantially sub-solar stellar metallicities under standard assumptions, adding to recent, similar results. Taken together, these findings suggest that some—if not many—z ≳ 2 passive galaxies are unlikely to evolve exclusively through dry minor merging (a currently popular scenario). As Kriek et al. (2016) also argued, this might imply that z ≳ 2 galaxies are a distinct population from those even at z ∼ 1.5, which already exhibit solar metallicities (see also Newman et al. 2012). Rather, the observed metallicity evolution of massive galaxies could be driven by newly quenched galaxies (a.k.a. progenitor bias; van Dokkum & Franx 1996; Carollo et al. 2013), as may also partially explain their size growth.

Ground-based K_s and IRAC photometry covers the rest-NIR in this study, a regime key to robust metallicity measurements (e.g., Lee et al. 2007). Despite the depth of such data, our lack of high-resolution spectroscopy does leave systematic uncertainties that must wait to be resolved by the James Webb Space Telescope (JWST). Most notable among these are isochrone effects, which can have a substantial impact at ages ∼500 Myr. Once in hand, large samples of direct JWST iron and α-element abundances at high-z would bear perhaps on our understanding of the cosmic merger rate, and the duty cycle of late-time star formation in once-dead galaxies in addition to their early chemical evolution.

We thank the anonymous referee for insightful suggestions, and Camilla Pacifici and Mia Bovill for fruitful discussion. Support for GLASS (HST-GO-13459, 14041) was provided by NASA through a grant from the Space Telescope Science Institute, operated by the Association of Universities for Research in Astronomy, Inc., under contract NAS 5-26555. This work uses gravitational lensing models by PIs Bradač Natarajan & Kneib (CATS v4.1), Merten & Zitrin, Sharon, Williams, Keeton, Bernstein & Diego, and the GLAFIC group. B.V. acknowledges support from an Australian Research Council Discovery Early Career Researcher Award (PD0028506).

Software: Astropy (Muna et al. 2016), emcee (Foreman-Mackey et al. 2013), FSPS (Conroy et al. 2009; Conroy & Gunn 2010; Foreman-Mackey et al. 2014), Grizli (Brammer 2018), lmfit (Newville et al. 2017).