Abstract
We study the dynamical properties of massive quiescent galaxies at 1.4 < z < 2.1 using deep Hubble Space Telescope WFC3/F160W imaging and a combination of literature stellar velocity dispersion measurements and new near-infrared spectra obtained using the K-band Multi Object Spectrograph (KMOS) on the ESO Very Large Telescope. We use these data to show that the typical dynamical-to-stellar mass ratio has increased by ∼0.2 dex from z = 2 to the present day, and we investigate this evolution in the context of possible changes in the stellar initial mass function (IMF) and/or fraction of dark matter contained within the galaxy effective radius, fDM[<re]. Comparing our high-redshift sample to their likely descendants at low redshift, we find that fDM[<re] has increased by a factor of more than 4 since z ≈ 1.8, from fDM[<re] = 6.6% ± 1.0% to ∼24%. The observed increase appears robust to changes in the methods used to estimate dynamical masses or match progenitors and descendants. We quantify possible variation of the stellar IMF through the offset parameter α, defined as the ratio of dynamical mass in stars to the stellar mass estimated using a Chabrier IMF. We demonstrate that the correlation between stellar velocity dispersion and α reported among quiescent galaxies at low redshift is already in place at z = 2, and we argue that subsequent evolution through (mostly minor) merging should act to preserve this relation while contributing significantly to galaxies' overall growth in size and stellar mass.
Export citation and abstract BibTeX RIS
1. Introduction
Spectroscopic surveys of the high-redshift universe have shown that well-known scaling relations such as the fundamental and mass planes were already in place by at least z = 2 (e.g., Toft et al. 2012; Bezanson et al. 2013b; van de Sande et al. 2014; Beifiori et al. 2017; Prichard et al. 2017), despite the fact that individual galaxies appear to evolve significantly from the time they join the passive population to the present day. The most conspicuous signature of this evolution is seen in galaxy sizes, where massive quiescent galaxies at z > 1 are significantly smaller than their local counterparts at fixed stellar mass (e.g., Daddi et al. 2005; Trujillo et al. 2006; van Dokkum et al. 2008; Cimatti et al. 2012; van der Wel et al. 2014; Chan et al. 2016, 2018, but see also Carollo et al. 2013), but it is also apparent in measurements of galaxy stellar velocity dispersions and surface brightness profiles (e.g., Kriek et al. 2009; Cenarro & Trujillo 2009; van der Wel et al. 2011; Chang et al. 2013; van de Sande et al. 2013). However, the exact degree to which individual galaxies change as they evolve is still unclear: although some amount of inferred evolution can be explained by a bias in the matching of progenitor and descendant populations (progenitor bias, e.g., van Dokkum & Franx 1996; Saglia et al. 2010; Valentinuzzi et al. 2010; Keating et al. 2015), some evolution is still required to reproduce properties of the full population (e.g., Belli et al. 2015).
Guided by the intrinsically hierarchical assembly of structure in ΛCDM models, the most attractive explanation for the continued structural evolution of quiescent galaxies is by gas-poor merging after the cessation of star formation. Both major (mass ratio μ* > 0.25) and minor (μ* < 0.25) mergers can significantly alter galaxy light profiles, leading to a disproportionate increase in (half-light/half-mass) size relative to stellar mass (e.g., Oser et al. 2010; Hilz et al. 2012, 2013), which seems all but demanded in the most compact, massive high-z galaxies (e.g., Damjanov et al. 2011). Detailed photometric and kinematic analyses of nearby passive galaxies appear to support the idea of a "two-phase" formation scenario characterized by early, rapid formation and subsequent assembly through repeated mergers (e.g., Arnold et al. 2011, 2014; de la Rosa et al. 2016; Foster et al. 2016). But while it appears that mergers with 0.1 < μ* < 1 can account for the evolution of galaxy sizes and velocity dispersions since z ∼ 1, they have more difficulty explaining the dramatic increase in average sizes at earlier epochs (e.g., Newman et al. 2012), suggesting that other mechanisms such as stellar mass loss or feedback from active galactic nuclei (AGNs) may also play some role (e.g., Fan et al. 2008; Damjanov et al. 2009; Fan et al. 2010).
While different evolutionary scenarios predict different physical characteristics for the resulting galaxy population, the persistence of the fundamental plane, mass plane, and other scaling relations over time limits the parameter space available to models describing the evolution of galaxy properties. The existence of a fundamental plane for quiescent galaxies can be understood as a manifestation of the virial relation, where for relaxed systems the dynamical mass , with σ* and re the stellar velocity dispersion and half-light size, respectively. Given measurements of σ* and re, the remaining unknown is the dynamical mass-to-light ratio, Mdyn/L. Following Graves et al. (2009), Mdyn/L can be rewritten in terms of its underlying physical dependencies as
The first and last terms, Mdyn/Mtot and M*,IMF/L, depend on our ability to model certain galaxy properties: the former encapsulates offsets between the derived dynamical mass Mdyn and the true total mass of the system Mtot, while the latter is the stellar mass-to-light ratio for some fiducial stellar initial mass function (IMF), usually obtained by modeling multiband photometric data. The fact that dynamical studies of nearby early-type galaxies can recover the virial relation suggests that, given appropriate assumptions, Mdyn/Mtot ≈ 1 (e.g., Hyde & Bernardi 2009; Cappellari et al. 2013b). Uncertainties in the derivation of M*,IMF/L from multiband photometry, on the other hand, can be significant (of order 0.1–0.2 dex) depending on the treatment of star formation history (SFH), metallicity, and dust (e.g., Leja et al. 2019). The short formation timescales and low attenuation generally inferred for passive galaxies help to reduce these uncertainties considerably (e.g., Pforr et al. 2012), but the extent to which these assumptions remain valid at higher redshift remains to be seen.
The remaining terms of Equation (1), Mtot/M* and M*/M*,IMF, encapsulate the relationship between different physical components of the galaxy and are the most likely to be affected by evolutionary processes. Mtot/M* is the ratio of total to stellar mass and is related to the balance of baryonic and dark matter (DM) within a given aperture—typically the effective radius, re—while M*/M*,IMF accounts for differences between the assumed and true stellar IMF. Variation of the IMF might be expected owing to the evolution of interstellar medium (ISM) properties with redshift and stellar mass, but there is no clear theoretical consensus as to how these changes might manifest in the observed galaxy population (see, e.g., Chabrier et al. 2014; Krumholz 2014, and references therein).
In nearby galaxies, deep photometric and spectroscopic data can be used to study the relationship between galaxies, their stellar populations, and the properties of their dark matter halos in great detail. Van Dokkum & Conroy (2010) used stellar population models to show that massive early-type galaxies host a large population of low-mass stars in their cores (≲re/8), suggesting a very bottom-heavy IMF compared to the Milky Way (MW) and other nearby star-forming galaxies. These results were consistent with a complementary analysis of strong-lensing systems by Treu et al. (2010), who additionally found evidence for systematic variation of the IMF from MW-like at low stellar velocity dispersions to Salpeter (1955) or heavier in the most massive galaxies. Cappellari et al. (2012) obtained similar results based on modeling the spatially resolved stellar kinematics of galaxies in the ATLAS3D survey. Stellar population results from studies like van Dokkum & Conroy (2010) are uniquely sensitive to a galaxy's stellar content, but dynamical IMF constraints cannot necessarily distinguish between IMF variation and changes in the central DM fraction. Cappellari et al. (2013b) showed that the typical dark matter fraction within re, fDM[<re], is relatively low (9%–17%) and, while fDM[<re] tends to increase with increasing galaxy mass, this variation cannot account for the observed trends in total M/L, supporting their conclusion of a systematically varying IMF (Cappellari et al. 2013a); unfortunately, the picture becomes complicated if there is no clear distinction between the baryonic and dark matter distributions (e.g., Thomas et al. 2011). Even though there is no consensus on the exact correlations between IMF normalization (or shape) and observed galaxy properties, variability of the IMF is now supported by a number of different studies using a wide range of stellar population, lensing, and dynamical techniques (e.g., Thomas et al. 2011; Conroy & van Dokkum 2012; Dutton et al. 2012; Cappellari et al. 2013a; Conroy et al. 2013; Ferreras et al. 2013; Spiniello et al. 2014; Martín-Navarro et al. 2015a; Parikh et al. 2018; but see also Smith et al. 2015).
At intermediate redshift, Tortora et al. (2018) used data from the Kilo Degree Survey (KiDS) and Sloan Digital Sky Survey (SDSS) to show that the locally observed correlations between stellar mass, dynamical mass, stellar velocity dispersion, and structural parameters are already in place by z ∼ 0.65, but that high-redshift quiescent galaxies likely have lower fDM[<re] at fixed stellar velocity dispersion than nearby galaxies (see also Beifiori et al. 2014; Tortora et al. 2014). Shetty & Cappellari (2015) found a similar decrease in the central dark matter fraction for massive galaxies at z ≈ 0.8, while at the same time reporting a Salpeter-like IMF consistent with massive galaxies at z = 0 (see also Shetty & Cappellari 2014; Sonnenfeld et al. 2015; Martín-Navarro et al. 2015b). Extending such studies of kinematic scaling relations beyond z > 1 remains challenging. While an abundance of massive, compact red galaxies have been identified using deep Hubble Space Telescope (HST) and ground-based imaging (e.g., Cimatti et al. 2004; Daddi et al. 2005; Whitaker et al. 2011, 2013), kinematic data for individual galaxies have been notoriously difficult to obtain (e.g., Kriek et al. 2009). The development of efficient, highly multiplexed near-infrared spectrographs such as MOSFIRE at Keck (McLean et al. 2012) and the K-band Multi Object Spectrograph (KMOS) at the ESO Very Large Telescope (Sharples et al. 2012, 2013) has led to rapid growth in the number of kinematic measurements at z > 1.4, but individual samples remain relatively small and have been analyzed using a wide variety of methods that make combining the results from different surveys difficult.
In this paper we undertake a homogeneous reanalysis of currently available kinematic data at high redshift in order to study the key parameters governing the behavior of Equation (1) over cosmic time, namely, the central dark matter fraction and normalization of the stellar IMF. Our sample comprises 58 quiescent galaxies at 1.4 ≤ z ≤ 2.1 with stellar velocity dispersion measurements and high-resolution HST/WFC3 imaging available. These data include 17 new stellar velocity dispersion measurements obtained as part of the VLT IR IFU Absorption Line Survey (VIRIAL; Mendel et al., in preparation), in addition to measurements from a variety of samples in the literature. We derive dynamical properties based on both a straightforward application of the virial theorem and more complex dynamical models, allowing us to test the influence of different assumptions about galaxy structure on the study of high-redshift stellar kinematics.
The outline of this paper is as follows: In Section 2 we describe the compilation of high-redshift galaxies, along with a comparison sample at z = 0. In Section 3 we discuss our modeling of galaxy surface brightness profiles and the calculation of dynamical masses. The main results of this work—the relationship between dynamical and stellar masses, central dark matter fraction, and dynamical constraints on the normalization of the stellar IMF—are presented in Section 4. In Section 5 we discuss our results in the context of the high- and low-redshift galaxy populations. We summarize our conclusions in Section 6.
Throughout this paper we use AB magnitudes (Oke & Gunn 1983) and adopt a flat ΛCDM cosmology with ΩΛ = 0.7, ΩM = 0.3, and H0 = 70 km s−1 Mpc−1.
2. Samples and Data
2.1. KMOS Observations at 1.5 < z < 2.0
Our analysis includes new spectroscopic data for 17 galaxies in the redshift range 1.5 < z < 2.0 observed as part of the VIRIAL GTO survey (Mendel et al., in preparation) using KMOS (Sharples et al. 2012, 2013). These galaxies were selected from 3D-HST (Brammer et al. 2012; Skelton et al. 2014; Momcheva et al. 2016) in the COSMOS, GOODS-S, and UDS fields to have mF140W ≤ 22.5 mag and be classified as quiescent according to their rest-frame U − V and V − J colors using the criteria described by Whitaker et al. (2011; see also Williams et al. 2009), shown in the top panel of Figure 1. Their general properties are given in Table 1.
Table 1. General Properties of 17 VIRIAL Targets
Field | ID | R.A. | Decl. | mF160W | (U − V)rf | (V − J)rf | Exposure |
---|---|---|---|---|---|---|---|
(J2000) | (J2000) | (mag) | (min.) | ||||
UDS | 22480 | 34.3353 | −5.2017 | 20.85 | 1.83 | 1.03 | 670 |
UDS | 24891 | 34.4458 | −5.1940 | 21.37 | 1.65 | 1.08 | 735 |
GOODS-S | 39364 | 53.0628 | −27.7265 | 20.99 | 1.70 | 1.06 | 475 |
GOODS-S | 42113 | 53.1279 | −27.7189 | 20.95 | 1.99 | 1.13 | 495 |
GOODS-S | 43548 | 53.1294 | −27.7073 | 21.82 | 1.47 | 0.77 | 505 |
COSMOS | 6977 | 150.0695 | 2.2500 | 21.62 | 1.73 | 0.95 | 650 |
UDS | 22802 | 34.4469 | −5.2007 | 21.05 | 1.70 | 0.96 | 635 |
UDS | 29352 | 34.4696 | −5.1786 | 21.44 | 1.69 | 0.94 | 740 |
UDS | 10237 | 34.3148 | −5.2433 | 20.75 | 1.78 | 1.12 | 440 |
COSMOS | 7411 | 150.1770 | 2.2552 | 21.37 | 1.82 | 1.00 | 630 |
UDS | 35111 | 34.4536 | −5.1589 | 21.63 | 1.67 | 0.87 | 740 |
UDS | 32892 | 34.3896 | −5.1681 | 21.17 | 1.55 | 0.76 | 660 |
UDS | 38073 | 34.3365 | −5.1490 | 21.30 | 1.38 | 0.79 | 635 |
COSMOS | 6396 | 150.1728 | 2.2441 | 21.89 | 1.69 | 0.98 | 615 |
COSMOS | 9227 | 150.0618 | 2.2737 | 21.47 | 1.60 | 0.79 | 620 |
COSMOS | 7391 | 150.0773 | 2.2548 | 22.01 | 1.39 | 0.53 | 650 |
COSMOS | 2816 | 150.1411 | 2.2085 | 21.43 | 1.84 | 1.04 | 650 |
Download table as: ASCIITypeset image
2.1.1. Observations and Data Reduction
Observations of VIRIAL galaxies were carried out between 2014 and 2016 using the KMOS YJ band (1–1.36 μm). Data were taken using a standard object-sky-object pattern with individual exposure times of 300 s. Each science exposure was offset by between 01 and 06 in order to avoid bad pixels in the final extracted spectra. Along with our science targets, we assigned one IFU from each of the three KMOS spectrographs to a reference star that we used to monitor the ambient conditions (seeing, atmospheric transmission, etc.), pointing accuracy, and point-spread function (PSF) shape. Due to the relatively small angular size of the KMOS IFUs (28 × 28), sky exposures were taken nodding completely off-source. Total on-source integration times range from 440 to 740 minutes (see Table 1).
Data were reduced using a combination of the Software Package for Astronomical Reductions with KMOS pipeline tools (SPARK; Davies et al. 2013) and custom Python scripts. In the following we briefly outline the steps used to produce calibrated one-dimensional spectra. Details of the VIRIAL reduction will be described in a future paper (Mendel et al., in preparation). Calibration exposures (dark, arc, and flat) were reduced using standard SPARK routines to produce flat-field, wavelength, and spatial calibration frames. When processing science frames, we first corrected each raw image for a readout-channel-dependent bias term estimated from reference pixels around the perimeter of each detector. We then adjusted the wavelength and spatial illumination calibrations for each exposure based on the positions and relative flux of bright sky lines before subtracting the object and sky images. The brightness of atmospheric OH lines can vary significantly between object and sky exposures (∼10% on 5- to 10-minute timescales; Ramsay et al. 1992; Davies 2007), often leading to significant systematic residuals in the initial sky-subtracted frames. In order to limit the impact of these systematics on our final spectra, we performed a second-order correction to the sky for each IFU using residuals measured in other IFUs in the same detector, excluding the IFU of interest.
One-dimensional spectra were extracted directly from the flat-fielded, illumination-corrected, and sky-subtracted detector frames for each exposure separately. Since VIRIAL targets are typically undetected in individual 300 s exposures, we used the available 3D-HST/CANDELS F125W imaging (Grogin et al. 2011; Koekemoer et al. 2011; Skelton et al. 2014) to model the source flux distribution and mask neighboring objects in the optimal extraction. The HST images were convolved to match the KMOS PSF measured from the reference stars in each exposure, which were also used to adjust for changes in transmission between exposures. Individual optimally extracted spectra were then corrected for telluric absorption using synthetic atmospheric models computed with MOLECFIT (Kausch et al. 2014) and combined using inverse variance weights. Uncertainties on the output spectra were estimated using bootstrap combines of the individual 1D spectra for each object. The typical spectral resolution in the extracted 1D spectra (as measured from sky lines) ranges from R = 3000 to 3500 (σinst ≈ 36–42 km s−1) depending on arm and detector (see also Wisnioski et al. 2019).
2.1.2. Stellar Masses and Velocity Dispersions
We estimated stellar velocity dispersions for VIRIAL galaxies using a simultaneous fit to the observed KMOS spectrum and multiband photometry from 3D-HST (Skelton et al. 2014). We generated model spectral energy distributions (SEDs) using FSPS v2.4 (Conroy et al. 2009; Conroy & Gunn 2010) assuming a lognormal SFH with
where t is the age of the universe, t0 is the delay time, and τ controls the width of the distribution (see also Gladders et al. 2013). The additional parameter ttrunc allows for star formation to be abruptly truncated and provides added flexibility when modeling the SFHs of quiescent galaxies. We stress that our adoption of a lognormal SFH is motivated by its flexibility compared to more commonly used τ or delayed-τ models, rather than an assumption that galaxies' SFHs are intrinsically lognormal (e.g., Gladders et al. 2013; Abramson et al. 2016; Diemer et al. 2017). In Appendix A we show that our derived velocity dispersions are not biased by the use of a parametric SFH. We modeled the effects of dust using a two-component extinction law that includes a foreground screen and additional attenuation toward young stellar populations (<107 yr), which are assumed to remain embedded within their birth clouds (see, e.g., Charlot & Fall 2000). We used the reddening curve of Calzetti et al. (2000) and, following Wuyts et al. (2013), adopted a relationship between the total V-band extinction AV and the additional extinction toward young stellar population Aextra such that . For simplicity we assume a fixed solar metallicity; in Appendix A we show that changing the metallicity by ±0.2 dex leads to systematic shifts in the derived velocity dispersion of ≲2%.
Before fitting, templates were smoothed to match the wavelength-dependent KMOS resolution measured from sky lines in extracted 1D spectra. The final matched templates include an additional (constant) offset of σoffset = 65 km s−1 to account for the resolution difference between KMOS (σinst ≈ 40 km s−1) and the adopted MILES spectral library (σMILES ≈ 75 km s−1; Beifiori et al. 2011). This effectively sets a floor for our velocity dispersion measurements of 65 km s−1. We limited our fits to the wavelength range from 3750 to 5300 Å and included a ninth-order additive polynomial—corresponding to ∼1 order per 10,000 km s−1—to minimize the effects of template mismatch on our final velocity dispersion measurements. We verified that our results are not sensitive to the adoption of an additive, as opposed to multiplicative, polynomial. In the end our model has a total of seven free parameters: redshift, z; stellar mass, MSPS;12 three parameters that describe the SFH, τ, t0, and ttrunc; absolute V-band extinction, AV; and stellar velocity dispersion, σ*. Samples from the posterior distribution were generated using emcee (Foreman-Mackey et al. 2013), and our final estimates of velocity dispersion and stellar mass were taken as the medians of their respective marginal posterior distributions, with 1σ uncertainties estimated from the 16th and 84th percentiles. We have confirmed that the derived stellar masses do not change significantly if we refit objects using only the available photometric data (i.e., excluding spectra). The final redshifts, stellar masses, and velocity dispersions are provided in Table 2.13 One-dimensional spectra and the corresponding best-fit models are shown in Figure 2.
Download figure:
Standard image High-resolution imageTable 2. Redshift, Velocity Dispersion, and Stellar Masses of KMOS Galaxies
Field | ID | z | σea | |
---|---|---|---|---|
(km s−1) | ||||
UDS | 22480 | 1.5288 | 11.08 | 323 ± 42 |
UDSb | 24891 | 1.6031 | 10.99 | 146 ± 32 |
GOODS-S | 39364 | 1.6118 | 11.10 | 203 ± 42 |
GOODS-S | 42113 | 1.6140 | 11.20 | 362 ± 65 |
GOODS-S | 43548 | 1.6143 | 10.64 | 169 ± 43 |
COSMOS | 6977 | 1.6424 | 10.86 | 187 ± 32 |
UDSb | 22802 | 1.6660 | 11.13 | 337 ± 28 |
UDSb | 29352 | 1.6886 | 10.91 | 277 ± 46 |
UDS | 10237 | 1.7664 | 11.38 | 233 ± 23 |
COSMOS | 7411 | 1.7808 | 11.09 | 186 ± 28 |
UDS | 35111 | 1.8217 | 10.95 | 228 ± 36 |
UDS | 32892 | 1.8243 | 11.02 | 206 ± 27 |
UDS | 38073 | 1.8245 | 10.94 | 194 ± 49 |
COSMOS | 6396 | 1.8364 | 10.90 | 169 ± 33 |
COSMOS | 9227 | 1.8604 | 10.98 | 273 ± 41 |
COSMOS | 7391 | 1.8681 | 10.54 | 145 ± 38 |
COSMOS | 2816 | 1.9230 | 11.26 | 297 ± 49 |
Notes. The formal statistical uncertainties on stellar masses derived from our SED fitting are of order 0.02 dex. Where relevant we include an additional 0.15 dex uncertainty on in quadrature to account for systematic uncertainties in the determination of stellar masses (see, e.g., Conroy et al. 2009; Mendel et al. 2014).
aVelocity dispersion corrected to re following the procedure described by van de Sande et al. (2013). bThese galaxies are in common with the Belli et al. (2017) sample. See discussion in Section 2.2.Download table as: ASCIITypeset image
In Figure 3 we show a comparison of stellar velocity dispersions obtained with and without the inclusion of photometric data in the fit. The two estimates are generally consistent within their quoted uncertainties, though there is a clear systematic offset in the sense that spectra-only fits return velocity dispersions that are ∼5% lower on average than those that also incorporate photometric data. This stems from the fact that the photometric data generally down-weight the youngest spectral templates—as one might expect from our a priori selection of galaxies based on their U − V and V − J colors—preferring instead solutions with smaller contributions from (rapidly rotating) early stellar types.
Download figure:
Standard image High-resolution image2.2. Literature Data at 1.4 < z < 2.1
In addition to our KMOS data, we have compiled a sample of quiescent galaxies with 1.4 < z < 2.1 from the literature where HST/WFC3 F160W imaging, multiwavelength photometric catalogs, and stellar velocity dispersion measurements were available. A full accounting of the literature data is given in Table 3, along with a few general galaxy properties. In Figure 4 we show the redshift distribution of our full high-redshift galaxy sample.
Download figure:
Standard image High-resolution imageTable 3. Properties of the High-z Literature Sample
Field | 3D-HST ID | z | a | (U − V) | (V − J) | Ref. ID | σe | Reference |
---|---|---|---|---|---|---|---|---|
(km s−1) | ||||||||
COSMOS | 30145 | 1.4010 | 10.90 | 1.84 | 1.09 | 19498 | 250 ± 39 | Belli et al. (2014a) |
AEGIS | 5087 | 1.4060 | 11.00 | 1.84 | 1.15 | 42109 | 369 ± 48 | Belli et al. (2014a) |
E9 | 295 ± 69b | Newman et al. (2010) | ||||||
GOODS-S | 40623 | 1.4149 | 10.89 | 2.07 | 1.25 | 2239 | 116 ± 36b | Cappellari et al. (2009) |
GOODS-S | 42466 | 1.4150 | 11.07 | 1.82 | 1.08 | 5020 | 181 ± 54 | Belli et al. (2014a) |
2470 | 147 ± 27b | Cappellari et al. (2009) | ||||||
GOODS-S | 43042 | 1.4190 | 11.32 | 2.24 | 1.28 | 4906 | 298 ± 26 | Belli et al. (2014a) |
AEGIS | ⋯ | 1.4235 | 11.26 | 1.57 | 0.78 | A17300 | 276 ± 7b | Bezanson et al. (2013a) |
COSMOS | 21628 | 1.4320 | 10.82 | 1.79 | 1.15 | 13880 | 169 ± 70 | Belli et al. (2014a) |
COSMOSd | 31780 | 1.4390 | 10.78 | 1.75 | 1.46 | 20841 | 267 ± 52 | Belli et al. (2014a) |
COSMOS | 31136 | 1.4420 | 10.93 | 1.86 | 1.07 | 20275 | 221 ± 70 | Belli et al. (2014a) |
UDS | 1854 | 1.4560 | 11.49 | 1.74 | 0.98 | 29410 | 355 ± 98 | van de Sande et al. (2013) |
UDSd | ⋯ | 1.4848 | 11.53 | 1.52 | 1.08 | U55531 | 260 ± 24b | Bezanson et al. (2013a) |
COSMOS | ⋯ | 1.5222 | 11.34 | 1.77 | 0.94 | C20866 | 284 ± 24b | Bezanson et al. (2013a) |
COSMOS | ⋯ | 1.5223 | 11.26 | 1.66 | 0.83 | C21434 | 229 ± 17b | Bezanson et al. (2013a) |
COSMOS | 17364 | 1.5260 | 11.02 | 1.84 | 1.12 | 17364 | 168 ± 84 | Belli et al. (2017) |
COSMOS | 17361 | 1.5270 | 10.86 | 1.63 | 0.93 | 17361 | 169 ± 43 | Belli et al. (2017) |
COSMOS | 17641 | 1.5280 | 10.79 | 1.70 | 1.01 | 17641 | 142 ± 54 | Belli et al. (2017) |
COSMOS | 17089 | 1.5280 | 11.37 | 2.02 | 1.23 | 17089 | 348 ± 57 | Belli et al. (2017) |
AEGIS | 17926 | 1.5730 | 11.14 | 1.79 | 1.03 | 17926 | 231 ± 39 | Belli et al. (2017) |
AEGIS | 22719 | 1.5790 | 11.13 | 1.87 | 1.14 | 22719 | 262 ± 51 | Belli et al. (2017) |
COSMOS | 28523 | 1.5825 | 11.38 | 1.82 | 0.91 | 34265 | 377 ± 54 | Belli et al. (2014a) |
18265 | 400 ± 72 | van de Sande et al. (2013) | ||||||
AEGIS | ⋯ | 1.5839 | 11.24 | 1.47 | 0.64 | A21129 | 275 ± 10b | Bezanson et al. (2013a) |
GOODS-N | 17678 | 1.5980 | 11.00 | 1.59 | 0.80 | 2653 | 174 ± 27 | Belli et al. (2014a) |
GN5 | 245 ± 37b | Newman et al. (2010) | ||||||
12632 | 187 ± 36b | Barro et al. (2016) | ||||||
UDSc | 24891 | 1.6035 | 10.99 | 1.65 | 1.08 | 24891 | 391 ± 71 | Belli et al. (2017) |
UDS | 35616 | 1.6090 | 11.19 | 1.64 | 0.79 | 35616 | 198 ± 49 | Belli et al. (2017) |
UDS | 30737 | 1.6200 | 11.37 | 1.77 | 1.04 | 30737 | 307 ± 82 | Belli et al. (2017) |
UDS | ⋯ | 1.6210 | 10.93 | 1.31 | 0.47 | U53937 | 251 ± 21b | Bezanson et al. (2013a) |
UDS | 43367 | 1.6240 | 11.26 | 1.80 | 1.26 | 43367 | 299 ± 74 | Belli et al. (2017) |
UDS | 30475 | 1.6330 | 10.83 | 1.38 | 0.70 | 30475 | 296 ± 109 | Belli et al. (2017) |
UDS | 32707 | 1.6470 | 11.25 | 1.82 | 1.12 | 32707 | 174 ± 30 | Belli et al. (2017) |
COSMOS | 16629 | 1.6570 | 10.67 | 1.64 | 0.79 | 16629 | 358 ± 76 | Belli et al. (2017) |
UDS | 37529 | 1.6650 | 11.13 | 1.78 | 1.23 | 37529 | 232 ± 60 | Belli et al. (2017) |
UDSc | 22802 | 1.6665 | 11.13 | 1.70 | 0.96 | 22802 | 291 ± 31 | Belli et al. (2017) |
GOODS-Nd | 11470 | 1.6740 | 10.77 | 1.25 | 0.51 | 8231 | 221 ± 36b | Barro et al. (2016) |
GOODS-N | 24033 | 1.6740 | 10.80 | 1.64 | 0.84 | 17360 | 155 ± 31b | Barro et al. (2016) |
GOODS-N | 3604 | 1.6750 | 10.69 | 1.67 | 0.88 | 2617 | 317 ± 118b | Barro et al. (2016) |
UDSc | 29352 | 1.6895 | 10.91 | 1.69 | 0.94 | 29352 | 146 ± 31 | Belli et al. (2017) |
COSMOS | 19958 | 1.7220 | 10.75 | 1.50 | 0.73 | 19958 | 169 ± 87 | Belli et al. (2017) |
COSMOS | 17255 | 1.7390 | 10.97 | 1.74 | 1.00 | 17255 | 147 ± 40 | Belli et al. (2017) |
AEGIS | 25526 | 1.7520 | 10.84 | 1.59 | 0.99 | 25526 | 134 ± 36 | Belli et al. (2017) |
COSMOSd | ⋯ | 1.8000 | 11.31 | 1.24 | 0.37 | 7447 | 287 ± 53 | van de Sande et al. (2013) |
UDSd | ⋯ | 2.0360 | 11.20 | 1.25 | 0.52 | 19627 | 304 ± 41 | van de Sande et al. (2013) |
19627 | 335 ± 56b | Toft et al. (2012) | ||||||
COSMOS | 13083 | 2.0880 | 11.10 | 1.76 | 0.90 | 13083 | 197 ± 52 | Belli et al. (2017) |
COSMOS | 11494 | 2.0920 | 11.58 | 1.87 | 1.01 | 7865 | 446 ± 57 | van de Sande et al. (2013) |
31719 | 358 ± 30 | Belli et al. (2014b) | ||||||
11494 | 319 ± 26 | Belli et al. (2017) | ||||||
COSMOSd | 12020 | 2.0960 | 11.34 | 1.69 | 1.44 | 31769 | 312 ± 65 | Belli et al. (2014b) |
Notes.
aStellar masses are rederived in this work following the method described in Section 2.1.2. bDispersions corrected to Re following the Appendix B of van de Sande et al. (2013). cThese galaxies are in common with the KMOS sample. See discussion in Section 2.2. dThese galaxies fall outside of the UVJ quiescent selection defined by Whitaker et al. (2011); however, their spectra show strong absorption features characteristic of post-starburst galaxies, as well as weak or absent [O II] emission, so we include them in our analysis.Download table as: ASCIITypeset image
This literature sample includes 15 galaxies from Newman et al. (2010), Bezanson et al. (2013a), and Belli et al. (2014a) observed with Keck LRIS at 1.4 < z < 1.6, as well as two galaxies from the GMASS spectroscopic sample (Cimatti et al. 2008) with velocity dispersions published by Cappellari et al. (2009). Belli et al. (2014a) incorporate the LRIS data from Newman et al. (2010) in their analysis, and there is one galaxy in common between Belli et al. (2014a) and Cappellari et al. (2009). Velocity dispersion measurements derived using near-IR spectroscopy are available for 29 additional galaxies with 1.4 < z < 2.1 from Toft et al. (2012) and van de Sande et al. (2013), obtained using VLT XShooter, and from Belli et al. (2014b), Barro et al. (2016), and Belli et al. (2017) using Keck MOSFIRE. The sample of Barro et al. (2016) includes one galaxy in common with the LRIS sample of Newman et al. (2010) and Belli et al. (2014a), and there are several galaxies in common between Toft et al. (2012), van de Sande et al. (2013), Belli et al. (2014b), and Belli et al. (2017). See Table 3 for details.
There are three galaxies in common between Belli et al. (2017) and the KMOS sample described in Section 2.1—UDS 24891, UDS 29352, and UDS 22802—which are highlighted in Tables 2 and 3. For UDS 22802, the two independent velocity dispersion measurements are in relatively good agreement (337 ± 28 vs. 291 ± 31 km s−1); however, for the other two galaxies the discrepancy is larger: 277 ± 46 km s−1 versus 146 ± 31 km s−1 for UDS 29352 (2.4σ offset) and 146 ± 32 km s−1 versus 391 ± 71 km s−1 for UDS 24891 (3.1σ offset). Although we are not in a position to assess which of these measurements are "correct," we note that adopting σe as measured by Belli et al. (2017) for these galaxies results in large offsets between their dynamical and stellar masses (see Section 4.1), such that UDS 29352 (UDS 24891) would have the highest (lowest) dynamical-to-stellar mass ratio in the sample. Nevertheless, in the absence of additional data we adopt a final σe for these objects based on an error-weighted average of the quoted measurements, with an increased uncertainty to reflect the large discrepancy between quoted values; in Section 3.1 we describe in more detail how we combine data for galaxies with multiple velocity dispersion measurements.
In order to ensure that our high-redshift sample is as homogeneous as possible, we remeasured stellar masses for all galaxies using the SED fitting procedure described in Section 2.1.2. In most cases, multiwavelength photometric catalogs were available from either the Newfirm Medium Band Survey (NMBS; Whitaker et al. 2011) or 3D-HST (Skelton et al. 2014). Several galaxies in the UDS field—UDS 55531 and UDS 53937 from Bezanson et al. (2013a), as well as UDS 19627 from Toft et al. (2012) and van de Sande et al. (2013)—fall outside of the 3D-HST footprint, and for these objects we adopted the combined Subaru/XMM-Newton Deep Survey (SXDS; Furusawa et al. 2008) and UKIRT Infrared Deep Sky Survey (Lawrence et al. 2007) catalogs described by Simpson et al. (2012). We supplemented these data with deep Spitzer/IRAC 3.6 and 4.5 μm flux measurements from Ashby et al. (2013), which were corrected to match the 3'' apertures used by Simpson et al. (2012) using the UKIDSS K-band mosaics. Stellar masses derived for the literature sample are provided in Table 3.
2.3. Comparison Sample at z ≈ 0
We identified a comparison sample of quiescent galaxies at low redshift from the SDSS Legacy Survey (Sloan Digital Sky Survey; Abazajian et al. 2009) using the same color-based selection criteria as at high redshift. We select galaxies with 0.02 ≤ z ≤ 0.2 and that also have stellar velocity dispersions measured by the Portsmouth group (see Thomas et al. 2013) using pPXF (Cappellari & Emsellem 2004). In order to avoid potential biases in the SED fitting between our high- and low-redshift data, we limit our selection to galaxies in the GAMA DR2 survey area (Galaxy and Mass Assembly; Driver et al. 2011; Liske et al. 2015), where Wright et al. (2016) provide aperture-matched photometric catalogs covering from the ultraviolet to infrared.
We computed rest-frame colors for these galaxies using EAZY (Brammer et al. 2008), and the resulting distribution is shown in the bottom panel of Figure 1, where we again adopt the color criteria of Whitaker et al. (2011) to select quiescent galaxies. Stellar masses for the 4546 galaxies satisfying this selection were estimated from fits to their far-UV to K-band photometry using the procedures described in Sections 2.1.2 and 2.2. Based on these data, we derive a redshift-dependent stellar mass limit following the approach of Sohn et al. (2017) and Zahid et al. (2019), such that our final sample of 3108 galaxies is mass complete at the 97.5% level.
The passive galaxy population appears to grow significantly from z = 2 to the present day, suggesting that our low-redshift data contain galaxies that are too young to be descendants of the galaxies in our high-z sample. While there is no consensus on the magnitude of such progenitor bias effects (e.g., Carollo et al. 2013; Belli et al. 2015; Fagioli et al. 2016), it is nevertheless important to account for them in our analysis. We use here the mass-weighted stellar ages derived by Comparat et al. (2017) for SDSS galaxies using FIREFLY (Wilkinson et al. 2017), and we select those galaxies with ages older than 9 Gyr as the most likely descendants of our high-z sample. This identifies a subsample of 792 galaxies, or ∼27% of the full quiescent sample. In the following we will discuss results for both the full and age-selected samples.
3. Dynamical Modeling
The main focus of this work is a discussion of the dynamical constraints afforded by current high-redshift quiescent galaxy samples and a comparison with low-redshift data. In this section we describe the key quantities required for this analysis—stellar velocity dispersions, structural parameters (sizes, Sérsic indices, etc.)—as well as our estimates of dynamical masses and their related quantities.
3.1. Stellar Velocity Dispersions
Galaxies in our high-redshift sample have stellar velocity dispersions derived within a range of apertures and are based on data obtained with a variety of instruments and extraction methods. These measurements therefore require some degree of homogenization in order to be meaningfully combined. In many cases authors quote velocity dispersions corrected such that they represent the luminosity-weighted mean with one effective radius, σe, and we adopt these values when available. Where velocity dispersions are quoted within a different aperture—as is the case for Cappellari et al. (2009), Newman et al. (2010), Toft et al. (2012), Bezanson et al. (2013a), and Barro et al. (2016)—we correct the quoted velocity dispersion to one effective radius following the procedure outlined in van de Sande et al. (2013).
In cases where multiple velocity dispersion measurements were available we used an inverse-variance-weighted average of the published dispersions, after correcting them to a common re aperture. In addition to their propagated uncertainties, we included a term (in quadrature) to account for large offsets between quoted dispersions, taken as half of the range of dispersion measurements. The one exception to this procedure is GOODS-N 17678, where the velocity dispersion measured by Newman et al. (2010) differs significantly from the measurements of Belli et al. (2014b) and Barro et al. (2016); for this object we used an average of only the Belli et al. (2014b) and Barro et al. (2016) dispersions.
In our low-redshift sample all dispersions were measured from spectra within a common 3'' aperture, corresponding to the SDSS fiber diameter. Where we quote individual stellar velocity dispersions, these aperture values have been corrected to one effective radius, again following the procedure described by van de Sande et al. (2013) and using structural parameters described below. However, in our dynamical modeling (see Section 3.3) we fit directly to model dispersions computed within the 3'' (fiber) aperture, accounting for seeing effects. Although the physical scale subtended by the 3'' SDSS fibers increases dramatically over the redshift range of our low-z sample, the physical quantities derived from our dynamical models are independent of redshift at fixed stellar mass, suggesting that the use of aperture measurements does not bias our results.
3.2. Structural Parameters
We adopted two different approaches to measuring structural properties for our galaxy samples: first using galfit (Peng et al. 2002) to model their two-dimensional surface brightness distributions using a single Sérsic profile (Sérsic 1963), and second using the Multi-Gaussian Expansion (MGE) approach described by Emsellem et al. (1994; see also Cappellari 2002).
3.2.1. Sérsic Profile Fits
In our high-redshift sample 50/58 galaxies fall within the HST WFC3/F160W imaging footprint of the CANDELS survey (Grogin et al. 2011; Koekemoer et al. 2011), and for these objects we used the mosaics and composite PSFs described by Skelton et al. (2014).14 The remaining eight galaxies were observed separately using HST WFC3/F160W as part of HST-GO-12167 (PI: Franx; AEGIS 17300, AEGIS 21129, COSMOS 21434, COSMOS 20866, COSMOS 07447, UDS 53937, and UDS 55531) and HST-GO-13002 (PI: Williams; UDS 19627) for a single orbit each with total exposure times of 2611 or 2411 s, respectively. Level 2 data products were retrieved from the Hubble Legacy Archive (HLA),15 and we constructed empirical PSFs for these objects by stacking the images of bright unsaturated stars in each combined frame. We generated segmentation maps for the HLA images using SExtractor (Bertin & Arnouts 1996) with parameters similar to those given by Skelton et al. (2014) for 3D-HST.
We used galfit to model the surface brightness distribution of the primary galaxy, while also including in the fit neighboring galaxies with mF160W < 25 and projected separations rp < 5 (rprimary + rneighbor). Initial estimates of the galaxy sizes, i.e., rprimary and rneighbor, were taken from the SExtractor output. The local sky background for each object was estimated using the full image by first masking all pixels within 3 Kron radii of nearby sources using the ellipse parameters produced by SExtractor. We then identified the nearest 10,000 unmasked pixels as sky. An initial estimate of the background was taken as the mode of these sky pixels, which was then iteratively refined to obtain our final estimates of the local sky background. Postage stamps for individual objects were then extracted and the local sky background removed; the background level was subsequently held fixed during fitting. The structural parameters derived in this way are consistent with those available in the literature; a direct comparison with literature values is given in Appendix B.1. An example of our photometric modeling for COSMOS 30145 is shown in Figure 5, with figures for the remaining galaxies included in Appendix C.
Download figure:
Standard image High-resolution imageAlthough there are numerous existing catalogs of structural parameter measurements for the SDSS and GAMA (e.g., Simard et al. 2002; Kelvin et al. 2012; Meert et al. 2015), for consistency with our high-z data we chose to rederive these quantities using the methodology described above. We retrieved "corrected" r-band images from the SDSS Data Archive Server (DAS), along with their associated mask and PSF files. We then used SExtractor to generate segmentation images following the procedures described by Simard et al. (2011), and the local sky background for each source was estimated using the method described above. Individual postage stamps and PSFs16 for each galaxy were then extracted and the background removed. galfit was used to simultaneously fit the primary galaxy and any neighboring sources with mr < 22 and rp < 5 (rprimary + rneighbor). All other sources were masked during the fit. A comparison of our measurements with several different literature catalogs can be found in Appendix B.2.
At both high and low redshift we derive sizes in fixed photometric bands (HST WFC3/F160W and SDSS r band, respectively), which probe different rest-frame wavelengths at different redshifts. In the presence of strong color gradients this shift in rest-frame wavelength can systematically bias our size measurements and must be taken into account. Following van der Wel et al. (2014), we define the corrected, in this case r-band, semimajor axis size as
where is the measured half-light size in either the WFC3/F160W or SDSS r-band filter and zp is the "pivot redshift." zp = 0 by definition for the GAMA/SDSS sample, as we are correcting to the rest-frame r-band size, while for F160W imaging zp = 1.49. Kelvin et al. (2012) used GAMA data to show that for early-type galaxies on average. Chan et al. (2016) and van der Wel et al. (2014) derive similar values based on their analyses of quiescent galaxies' high redshift, and we therefore adopt for all galaxies in our sample. The typical correction derived in this way is of order 2%–3%, and we adopt these corrected r-band sizes for the remainder of this work.
3.2.2. Multi-Gauss Expansion Fits
While the single-component Sérsic fits described in Section 3.2.1 provide a straightforward summary of the overall surface brightness profile, Sérsic models have several drawbacks that complicate their use in constructing dynamical models. As well as providing a poor description of multicomponent profiles (e.g., bulge + disk), the coupling between inner and outer profile shapes makes the Sérsic models extremely sensitive to sky background: over-/undersubtraction of the sky level can significantly affect the inferred inner profile shape. In addition, with the exception of a few special cases, Sérsic profiles cannot be deprojected analytically, making their use for constructing dynamical models computationally expensive compared to simpler functional forms. In this context, modeling galaxies as a sum of individual Gaussian components—so-called multi-Gaussian expansion (MGE; Emsellem et al. 1994; Cappellari 2002)—provides a flexible description of surface brightness profiles that does not require any extrapolation of the profile to large radii, can accommodate multiple photometric components, and can be easily deprojected to obtain an estimate of the three-dimensional luminosity density (see Section 3.3.2).
The starting points for our MGE models were the background-subtracted postage stamps produced as described in Section 3.2.1. We used the results of our Sérsic model fits to subtract neighboring sources before identifying the primary object and producing binned two-dimensional surface brightness measurements using the find_galaxy and sectors_photometry routines described by Cappellari (2002).17 A model of the surface photometry in terms of nested Gaussians was then derived using the mge_fit_sectors method of Cappellari (2002). For high-redshift galaxies we constructed MGE-based PSF models per field using either composite PSFs provided by Skelton et al. (2014) for galaxies within the CANDELS/3D-HST footprint or else the stacked images of bright stars within the same field for stand-alone observations. In the right panel of Figure 5 we show a comparison of the observed and MGE-derived surface brightness contours for one object, COSMOS 30145.
MGE PSF models for the low-redshift SDSS/GAMA data were constructed on a galaxy-by-galaxy basis using the PSF extracted from the SDSS drField files. In all cases—that is, both high and low redshift—we tied the ellipticity of the fitted Gaussian components together to avoid large variations in the derived axis ratios for low surface brightness components; however, we confirmed that our results are not qualitatively sensitive to this assumption.
3.3. Dynamical Masses
The final piece of information we require is an estimate of total galaxy mass, including both stellar and dark matter components. For this work we investigate two broad approaches to estimating dynamical masses in order to test their sensitivity to underlying assumptions: the first is based on a simple application of the virial theorem and scaling relations derived for nearby galaxies, while the second relies on more detailed dynamical modeling of the stellar density profile and velocity dispersion.
3.3.1. Virial Mass Estimates
As outlined in Section 1, the tight relationship between size, stellar velocity dispersion, and mass for nearby early-type galaxies can be understood as a consequence of virial equilibrium, where for a pressure-supported system the total mass is given by
Here κ(n) is the so-called virial coefficient, and in this case it is taken as an analytic function of Sérsic index that encapsulates the effects of structural and orbital nonhomology (e.g., Bertin et al. 2002; Cappellari et al. 2006). We adopt the relation derived by Cappellari et al. (2006) based on spherical, isotropic models,
which has been shown to provide a reliable estimate of the total mass for nearby early-type galaxies in the SAURON and ATLAS3D samples (e.g., Cappellari et al. 2006, 2013b). Note that in Equation (4) we used the semimajor axis size, , following the discussion of Cappellari et al. (2013a, their Figure 14). The semimajor axis size is expected to be more robust to systematic changes in galaxy shapes than the harmonic mean size (e.g., , where a and b are the semimajor- and semiminor-axis sizes, respectively), especially for (thin) disk galaxies where the observed b/a is an indicator of inclination rather than intrinsic shape.
3.3.2. Jeans Models
The assumptions of spherical symmetry and isotropy discussed above appear to be reasonable at low redshift; however, high-z quiescent galaxies are known to be flatter on average—that is, have intrinsically lower b/a—than their low-redshift counterparts (e.g., van der Wel et al. 2011; Chang et al. 2013), leading to a possible bias in their derived masses when using Equation (5). We therefore consider an alternative approach to computing dynamical masses based on the Jeans Anisotropic MGE (JAM) method discussed by Cappellari (2008), which allows us to relax these assumptions. The modeling requires as input the MGE-derived surface brightness profile described in Section 3.2.2 and a measurement of the stellar velocity dispersion (see Section 2.1.2 and 2.2).
Following Cappellari (2002), the deprojected luminosity density can be computed from the best-fit MGE decomposition given assumptions about the inclination, which is related to the intrinsic axis ratio of an oblate ellipsoid qint by
where i is the inclination and qobs is the observed axis ratio. Since in this work we are concerned with the sensitivity of our dynamical mass estimates to possible changes of the intrinsic axis ratio, we computed JAM models over a grid of qint from 0.05 ≤ qint ≤ min(0.95, qobs) in steps of 0.05; unless otherwise stated, our results are based on marginalizing over qint. In our default modeling we assumed that the velocity ellipsoid is marginally anisotropic with an anisotropy parameter (where z and R define directions parallel and perpendicular to the symmetry axis for an axisymmetric system) based on local early-type galaxies (e.g., Cappellari et al. 2007; Thomas et al. 2009). We explored possible systematic effects over a range of anisotropies from 0 ≤ β ≤ 0.8 and found that they resulted in variations of the derived dynamical masses of at most a few percent, consistent with previous results (e.g., Wolf et al. 2010; Dutton et al. 2013); all results are therefore quoted adopting our fiducial value of β = 0.2.
We adopt two different implementations of the JAM modeling procedure distinguished by their treatment of baryonic and dark matter components. In the first instance we assume that the total mass is proportional to the light at all radii, i.e., mass-follows-light (MFL). This provides a self-consistent estimate of the dynamical mass-to-light ratio (M/L)MFL. MFL models have been shown to reliably recover the total mass within relatively small apertures (≲re), even in the presence of multiple mass components (e.g., Cappellari et al. 2006; Williams et al. 2010), and provide a baseline comparison for dynamical masses computed following Equation (5). A similar approach was used by Shetty & Cappellari (2014) to study quiescent galaxies at z ∼ 0.8 in the DEEP2 survey. In the MFL case the best-fitting value of (M/L)MFL for a given combination of qint and β is simply given by (M/L)MFL = (σe,obs/σe,model)2, where σe,obs is the observed aperture velocity dispersion and σe,model is the model prediction assuming (M/L)MFL = 1. Instead, our second implementation includes an explicit dark matter component described by a spherical NFW halo profile (Navarro et al. 1996). With sufficient sampling of the velocity field it is possible to independently constrain the stellar mass-to-light ratio, (M/L)*,NFW, and properties of the dark matter halo (e.g., Cappellari et al. 2013b; Übler et al. 2018). However, the aperture velocity dispersions used here cannot be used to break the degeneracy between stellar and dark matter components, leading us to impose additional constraints on the properties of the dark matter halo. Starting from our photometric estimates of galaxy stellar mass, we assigned dark matter halo masses based on the evolving stellar-to-halo mass relation derived by Moster et al. (2013). We then used the calculations of Diemer & Kravtsov (2015) to assign a halo concentration. This halo profile was then fed back into the JAM modeling procedure along with the MGE-based stellar density profile, and a grid search was used to determine the mass-to-light ratio of the stellar component, (M/L)*,NFW, as a function of qint and β.
In the explicit DM halo case we obtain an estimate of the dark matter fraction within re, fDM[<re], defined as
We compute fDM[<re] within a volume defined by , where for consistency with the literature re is the circularized half-light radius (), and the relevant masses are computed using the derived M/L values and deprojected MGE luminosity densities. For all galaxies we use the rest-frame r-band sizes computed following Equation (3).
4. Results
In this section we present the main results of this work, which are focused in two areas: the relationship between dynamical and stellar masses, and the interplay between dark matter content and the IMF at high redshift. Derived quantities for our high-redshift sample are provided in Table 4, and they are described in more detail in Sections 2 and 3.
Table 4. Derived Properties of High-redshift Galaxies
Field | IDa | c | n | qobs | σ* | Reference | |||||
---|---|---|---|---|---|---|---|---|---|---|---|
(kpc) | (L⊙) | (km s−1) | (M⊙/L⊙) | (M⊙/L⊙) | (M⊙/L⊙) | ||||||
COSMOS | 30145 | 1.25 ± 0.26 | 4.31 | 0.42 | 11.28 ± 0.01 | 250 ± 39 | −0.30 ± 0.17 | −0.32 ± 0.14 | 2 | ||
AEGIS | 05087 | 1.18 ± 0.26 | 2.53 | 0.41 | 11.36 ± 0.01 | 345 ± 54 | −0.02 ± 0.17 | −0.12 ± 0.14 | 2, 3 | ||
GOODS-S | 40623 | 2.29 ± 0.27 | 2.98 | 0.88 | 11.18 ± 0.01 | 116 ± 36 | −0.52 ± 0.27 | −0.50 ± 0.27 | 4 | ||
GOODS-S | 42466 | 2.53 ± 0.30 | 6.42 | 0.91 | 11.45 ± 0.01 | 154 ± 30 | −0.75 ± 0.18 | −0.51 ± 0.17 | 2, 4 | ||
GOODS-S | 43042 | 3.48 ± 0.30 | 5.62 | 0.63 | 11.68 ± 0.01 | 298 ± 26 | −0.20 ± 0.09 | −0.11 ± 0.08 | 2 | ||
AEGIS | A17300b | 2.93 ± 0.29 | 5.48 | 0.68 | 11.89 ± 0.01 | 276 ± 7 | −0.54 ± 0.06 | −0.47 ± 0.02 | 5 | ||
COSMOS | 21628 | 1.09 ± 0.26 | 3.07 | 0.70 | 11.18 ± 0.01 | 169 ± 70 | −0.52 ± 0.37 | −0.48 ± 0.36 | 2 | ||
COSMOS | 31780 | 2.45 ± 0.27 | 1.28 | 0.32 | 11.11 ± 0.01 | 267 ± 52 | 0.39 ± 0.18 | 0.30 ± 0.17 | 2 | ||
COSMOS | 31136 | 2.01 ± 0.28 | 4.24 | 0.45 | 11.33 ± 0.01 | 221 ± 70 | −0.24 ± 0.28 | −0.29 ± 0.28 | 2 | ||
UDS | 01854 | 2.40 ± 0.27 | 2.69 | 0.49 | 11.99 ± 0.01 | 355 ± 98 | −0.32 ± 0.24 | −0.39 ± 0.24 | 6 | ||
UDS | U55531b | 8.01 ± 0.40 | 3.90 | 0.75 | 11.94 ± 0.01 | 260 ± 24 | −0.09 ± 0.08 | −0.05 ± 0.08 | 5 | ||
COSMOS | C20866b | 2.71 ± 0.27 | 3.45 | 0.66 | 11.84 ± 0.01 | 284 ± 24 | −0.35 ± 0.09 | −0.38 ± 0.07 | 5 | ||
COSMOS | C21434b | 1.99 ± 0.26 | 3.43 | 0.72 | 11.87 ± 0.01 | 229 ± 17 | −0.70 ± 0.09 | −0.69 ± 0.06 | 5 | ||
COSMOS | 17364 | 2.86 ± 0.28 | 2.77 | 0.48 | 11.37 ± 0.01 | 168 ± 84 | −0.28 ± 0.44 | −0.35 ± 0.43 | 7 | ||
COSMOS | 17361 | 1.91 ± 0.26 | 2.13 | 0.70 | 11.45 ± 0.01 | 169 ± 43 | −0.49 ± 0.23 | −0.58 ± 0.22 | 7 | ||
COSMOS | 17089 | 5.65 ± 0.40 | 4.58 | 0.86 | 11.85 ± 0.01 | 348 ± 57 | 0.06 ± 0.15 | 0.12 ± 0.14 | 7 | ||
COSMOS | 17641 | 1.19 ± 0.27 | 5.32 | 0.92 | 11.26 ± 0.02 | 142 ± 54 | −0.86 ± 0.35 | −0.68 ± 0.33 | 7 | ||
UDS | 22480 | 1.84 ± 0.26 | 4.54 | 0.55 | 11.62 ± 0.01 | 323 ± 42 | −0.26 ± 0.13 | −0.24 ± 0.11 | 1 | ||
AEGIS | 17926 | 5.45 ± 0.43 | 4.19 | 0.72 | 11.61 ± 0.02 | 231 ± 39 | −0.04 ± 0.15 | −0.05 ± 0.15 | 7 | ||
AEGIS | 22719 | 2.25 ± 0.30 | 6.08 | 0.94 | 11.54 ± 0.01 | 262 ± 51 | −0.39 ± 0.18 | −0.26 ± 0.17 | 7 | ||
COSMOS | 28523 | 1.88 ± 0.26 | 2.93 | 0.25 | 11.95 ± 0.01 | 385 ± 45 | −0.32 ± 0.12 | −0.43 ± 0.10 | 2, 6 | ||
AEGIS | A21129b | 1.96 ± 0.26 | 7.15 | 0.49 | 11.97 ± 0.01 | 275 ± 10 | −0.94 ± 0.08 | −0.64 ± 0.03 | 5 | ||
GOODS-N | 17678 | 1.29 ± 0.26 | 8.00 | 0.70 | 11.61 ± 0.01 | 179 ± 23 | −1.25 ± 0.16 | −0.91 ± 0.11 | 2, 3, 8 | ||
UDS | 24891 | 1.88 ± 0.26 | 2.45 | 0.88 | 11.48 ± 0.01 | 187 ± 126 | −0.46 ± 0.59 | −0.43 ± 0.58 | 1, 7 | ||
UDS | 35616 | 4.47 ± 0.48 | 6.07 | 0.65 | 11.82 ± 0.02 | 198 ± 49 | −0.61 ± 0.23 | −0.50 ± 0.21 | 7 | ||
GOODS-S | 39364 | 1.58 ± 0.26 | 2.97 | 0.95 | 11.64 ± 0.01 | 203 ± 42 | −0.64 ± 0.19 | −0.56 ± 0.18 | 1 | ||
GOODS-S | 42113 | 1.96 ± 0.27 | 6.53 | 0.78 | 11.66 ± 0.01 | 362 ± 65 | −0.33 ± 0.17 | −0.16 ± 0.16 | 1 | ||
GOODS-S | 43548 | 0.94 ± 0.25 | 3.91 | 0.62 | 11.30 ± 0.01 | 169 ± 43 | −0.74 ± 0.25 | −0.66 ± 0.22 | 1 | ||
UDS | 30737 | 3.32 ± 0.27 | 2.71 | 0.51 | 11.82 ± 0.01 | 307 ± 82 | −0.13 ± 0.24 | −0.21 ± 0.23 | 7 | ||
UDS | U53937b | 0.63 ± 0.25 | 3.78 | 0.78 | 11.74 ± 0.01 | 251 ± 21 | −1.01 ± 0.19 | −0.86 ± 0.07 | 5 | ||
UDS | 43367 | 2.69 ± 0.28 | 5.18 | 0.53 | 11.60 ± 0.01 | 299 ± 74 | −0.18 ± 0.22 | −0.16 ± 0.21 | 7 | ||
UDS | 30475 | 0.98 ± 0.25 | 3.04 | 0.75 | 11.55 ± 0.01 | 296 ± 109 | −0.44 ± 0.34 | −0.39 ± 0.32 | 7 | ||
COSMOS | 06977 | 1.49 ± 0.25 | 1.43 | 0.79 | 11.41 ± 0.01 | 187 ± 32 | −0.43 ± 0.17 | −0.48 ± 0.15 | 1 | ||
UDS | 32707 | 1.75 ± 0.26 | 3.62 | 0.25 | 11.70 ± 0.01 | 174 ± 30 | −0.83 ± 0.16 | −0.92 ± 0.15 | 7 | ||
COSMOS | 16629 | 0.74 ± 0.25 | 2.40 | 0.72 | 11.32 ± 0.01 | 358 ± 76 | −0.13 ± 0.24 | −0.07 ± 0.18 | 7 | ||
UDS | 37529 | 2.32 ± 0.28 | 3.83 | 0.64 | 11.52 ± 0.01 | 232 ± 60 | −0.29 ± 0.23 | −0.28 ± 0.23 | 7 | ||
UDS | 22802 | 1.50 ± 0.25 | 2.33 | 0.36 | 11.66 ± 0.01 | 316 ± 31 | −0.27 ± 0.11 | −0.23 ± 0.09 | 1, 7 | ||
GOODS-N | 11470 | 2.85 ± 0.29 | 4.00 | 0.71 | 11.60 ± 0.01 | 221 ± 36 | −0.33 ± 0.15 | −0.31 ± 0.14 | 8 | ||
GOODS-N | 24033 | 1.09 ± 0.25 | 3.20 | 0.72 | 11.43 ± 0.01 | 155 ± 31 | −0.84 ± 0.20 | −0.85 ± 0.17 | 8 | ||
GOODS-N | 03604 | 0.82 ± 0.25 | 2.55 | 0.25 | 11.31 ± 0.01 | 317 ± 118 | −0.19 ± 0.35 | −0.25 ± 0.32 | 8 | ||
UDS | 29352 | 1.08 ± 0.25 | 4.79 | 0.77 | 11.52 ± 0.01 | 187 ± 70 | −0.87 ± 0.34 | −0.73 ± 0.33 | 1, 7 | ||
COSMOS | 19958 | 2.62 ± 0.29 | 2.94 | 0.84 | 11.45 ± 0.01 | 169 ± 87 | −0.39 ± 0.45 | −0.39 ± 0.45 | 7 | ||
COSMOS | 17255 | 1.54 ± 0.26 | 2.93 | 0.60 | 11.46 ± 0.01 | 147 ± 40 | −0.75 ± 0.25 | −0.76 ± 0.24 | 7 | ||
AEGIS | 25526 | 0.83 ± 0.25 | 2.48 | 0.51 | 11.37 ± 0.01 | 134 ± 36 | −0.98 ± 0.27 | −0.99 ± 0.23 | 7 | ||
UDS | 10237 | 3.07 ± 0.29 | 4.07 | 0.67 | 11.88 ± 0.01 | 233 ± 23 | −0.54 ± 0.10 | −0.58 ± 0.09 | 1 | ||
COSMOS | 07411 | 2.04 ± 0.27 | 3.95 | 0.85 | 11.64 ± 0.01 | 186 ± 28 | −0.66 ± 0.14 | −0.69 ± 0.13 | 1 | ||
COSMOS | C07447b | 1.67 ± 0.25 | 5.58 | 0.68 | 12.19 ± 0.01 | 287 ± 53 | −1.03 ± 0.18 | −0.87 ± 0.16 | 6 | ||
UDS | 35111 | 0.78 ± 0.25 | 2.56 | 0.31 | 11.56 ± 0.01 | 228 ± 36 | −0.74 ± 0.19 | −0.75 ± 0.14 | 1 | ||
UDS | 32892 | 1.63 ± 0.26 | 4.05 | 0.87 | 11.73 ± 0.01 | 206 ± 27 | −0.76 ± 0.14 | −0.68 ± 0.11 | 1 | ||
UDS | 38073 | 2.78 ± 0.29 | 7.06 | 0.84 | 11.68 ± 0.02 | 194 ± 49 | −0.78 ± 0.23 | −0.54 ± 0.22 | 1 | ||
COSMOS | 06396 | 1.44 ± 0.25 | 1.38 | 0.86 | 11.47 ± 0.01 | 169 ± 33 | −0.58 ± 0.19 | −0.61 ± 0.17 | 1 | ||
COSMOS | 09227 | 1.16 ± 0.25 | 2.75 | 0.68 | 11.65 ± 0.01 | 273 ± 41 | −0.51 ± 0.16 | −0.55 ± 0.13 | 1 | ||
COSMOS | 07391 | 0.69 ± 0.25 | 8.00 | 0.80 | 11.40 ± 0.01 | 145 ± 38 | −1.48 ± 0.29 | −0.94 ± 0.23 | 1 | ||
COSMOS | 02816 | 1.93 ± 0.25 | 1.82 | 0.24 | 11.74 ± 0.01 | 297 ± 49 | −0.25 ± 0.15 | −0.30 ± 0.14 | 1 | ||
UDS | U19627b | 2.08 ± 0.25 | 3.48 | 0.48 | 12.06 ± 0.01 | 315 ± 37 | −0.57 ± 0.12 | −0.65 ± 0.10 | 6, 9 | ||
COSMOS | 13083 | 1.38 ± 0.24 | 3.43 | 0.88 | 11.75 ± 0.01 | 197 ± 52 | −0.84 ± 0.24 | −0.73 ± 0.23 | 7 | ||
COSMOS | 11494 | 2.70 ± 0.28 | 4.68 | 0.80 | 12.19 ± 0.01 | 348 ± 66 | −0.58 ± 0.17 | −0.53 ± 0.17 | 6, 7, 10 | ||
COSMOS | 12020 | 1.98 ± 0.26 | 3.87 | 0.59 | 11.70 ± 0.01 | 312 ± 65 | −0.26 ± 0.19 | −0.38 ± 0.18 | 10 |
References. (1) this work; (2) Belli et al. 2014a; (3) Newman et al. 2010; (4) Cappellari et al. 2009; (5) Bezanson et al. 2013a; (6) van de Sande et al. 2013; (7) Belli et al. 2017; (8) Barro et al. 2016; (9) Toft et al. 2012; (10) Belli et al. 2014b.
aUnless otherwise noted, IDs correspond to those provided by Skelton et al. (2014) for galaxies in the 3D-HST fields. bThese galaxies fall outside the 3D-HST footprint; IDs therefore correspond to those given in their originating publications. cSizes have been corrected to the rest-frame r band according to Equation (3).4.1. The Relationship between Dynamical and Stellar Mass
In Figure 6 we show a comparison of dynamical and stellar masses for the different dynamical mass estimates described in Section 3.3. Two features are apparent. First, fixed stellar mass high-redshift galaxies appear to have dynamical masses that are ∼0.20 dex lower on average than their low-redshift counterparts. This offset appears regardless of the dynamical mass estimate used (i.e., MMFL vs. Mvir). Second, the correlation between dynamical and stellar mass is superlinear regardless of redshift, in the sense that the ratio of dynamical to stellar mass increases with increasing stellar mass. Such a "tilt" in the relationship between dynamical and stellar mass has been studied extensively at low redshift and has commonly been interpreted as variation of the central dark matter fraction and/or stellar IMF (e.g., Renzini & Ciotti 1993; Cappellari et al. 2013a; Dutton et al. 2013); in the following sections we will consider evidence for changes in the dark matter fraction and stellar IMF among high-redshift galaxies in more detail.
Download figure:
Standard image High-resolution imageFinally, there are a number of galaxies in Figure 6 with stellar mass estimates formally larger than their derived dynamical masses. While this cannot physically be the case, there a number of factors that influence the apparent trend, particularly at low stellar masses. Observational uncertainties at increase dramatically, driven primarily by the increased uncertainty on galaxy size as one pushes down the size−mass relation. While these increased uncertainties cannot in and of themselves explain the apparent shift toward low dynamical masses, when combined with the tilt of the relation described above they can nevertheless increase the fraction of galaxies with low dynamical-to-stellar mass ratios. In addition, we will show in Section 4.1.2 that our dynamical modeling likely underestimates the dynamical mass for galaxies that are intrinsically flat. Enforcing a flat structure for face-on galaxies can increase dynamical mass estimates by as much as ∼0.2 dex. Indeed, nearly 59% of galaxies with qobs ≥ 0.7 have dynamical masses smaller than their derived stellar mass, compared to only 24% for galaxies with qobs < 0.7.
4.1.1. Central Dark Matter Fractions
The tendency for high-redshift quiescent galaxies to have lower dynamical-to-stellar mass ratios compared to low redshift has been reported in a number of previous studies (e.g., Toft et al. 2012; van de Sande et al. 2013; Belli et al. 2017) and is generally interpreted as reflecting a systematic decrease in the central dark matter fraction, fDM[<re]. This decline in dynamical-to-stellar mass ratio appears to occur relatively smoothly with increasing redshift, as shown by a number of studies based on large spectroscopic surveys at z < 1 (e.g., Beifiori et al. 2014; Tortora et al. 2014, 2018). Figure 7 shows the cumulative distribution of dynamical-to-stellar mass ratio in the samples considered here. We find an offset in the mean dynamical-to-stellar mass ratio of −0.20 dex when moving to high redshift— at z = 0 compared to 0.09 ± 0.05 at 1.4 ≤ z ≤ 2.1—which is consistent with the results of previous studies. The magnitude of this offset is independent of the dynamical mass estimator used (e.g., MMFL vs. Mvir) and does not change when considering only the oldest galaxies at z = 0 (shown as contours in Figure 6 and light-gray lines in Figure 7). The right panel of Figure 7 shows the same distribution of dynamical-to-stellar mass ratios as the left panel, but with individual measurements normalized by their uncertainties. These can be compared to the dotted–dashed (black) line, which shows the prediction for a standard normal distribution. Although nearly 40% of galaxies in the high-redshift sample have photometrically derived stellar masses that exceed their dynamical masses, given measurement uncertainties the overall distribution is consistent with a positive (albeit small) dynamical-to-stellar mass ratio on average.
Download figure:
Standard image High-resolution imageWe can examine the evolution of fDM[<re] more directly using our dynamical models that include an explicit dark matter component, where dark matter fractions are computed following Equation (7). In Figure 8 we show fDM[<re] as a function of M*,NFW, the dynamical mass of the stellar component. While there is significant uncertainty in the individual measurements of fDM[<re] at high redshift, the overall trends support our interpretation of Figures 6 and 7 in terms of an evolution in the central dark matter fraction: galaxies at 1.4 < z < 2.1 have a mean fDM[<re] = 6.6 ± 1.0%, a factor of >2 lower than galaxies of a similar mass in our SDSS/GAMA sample at z = 0 (fDM[<re] ≈ 16.3 ± 0.3%; cf. 17% from Cappellari et al. 2013b). Furthermore, our low-redshift fDM[<re] measurements are consistent with the values derived by Thomas et al. (2011) and Cappellari et al. (2013b) based on more detailed dynamical modeling of low-redshift galaxies, suggesting that the observed offset in fDM[<re] between different redshifts is unlikely to be due to differences in the modeling approach. The above comparison between high- and low-redshift galaxies at fixed mass must nevertheless be made with some caution, as individual galaxies are expected to evolve from z = 2 to 0; we will revisit the evolution of fDM[<re] using more carefully matched progenitor and descendant samples in Section 5.
Download figure:
Standard image High-resolution imageUsing data from the SINS survey, Förster Schreiber et al. (2009) found that star-forming galaxies at z ∼ 2 are strongly baryon dominated, even for a Chabrier (2003) IMF, suggesting little room for either a bottom-heavy Salpeter IMF or significant dark matter. These results have recently been supported by kinematic data for hundreds of early star-forming disks in the KMOS3D (Wisnioski et al. 2015, 2019) and MOSDEF (Kriek et al. 2015) surveys (e.g., Price et al. 2016; Wuyts et al. 2016; Lang et al. 2017; Übler et al. 2017), as well as the detailed analysis of outer rotation curves for individual high-redshift disks (Genzel et al. 2017; Genzel et al. 2020; but see also Tiley et al. 2019). For comparison, in Figure 8 we show the dark matter fractions derived by Genzel et al. (2017, shown as squares and upper limits for a subsample of high-redshift, star-forming galaxies), which are in good agreement with the fDM[<re] measurements derived here.
4.1.2. The Effects of Unresolved Rotation
One of our goals in comparing multiple dynamical mass estimators is to assess the impact and importance of different modeling assumptions on the inferred properties of high-redshift galaxies. In that regard, the main distinctions between Mvir and MMFL are the assumptions of spherical symmetry and isotropy inherited through the application of Equations (4) and (5).
In practice, the dynamical masses derived here are only weakly dependent on changes of the anisotropy, β, at fixed qint for a range of values consistent with local early-type galaxies (0 ≤ β ≤ 0.6; Cappellari et al. 2007). It is therefore unlikely that the assumption of isotropy has a significant impact on the results presented in Figures 6–8, particularly given the typical uncertainties on measurements of σe (20%–30%). On the other hand, changes in assumed galaxy structure—for example, from spherically symmetric to oblate and axisymmetric—can systematically bias dynamical mass estimates depending on the degree of intrinsic flattening and relative importance of rotation versus pressure support.
Crucially, there is growing observational evidence that quiescent galaxies at high redshift may indeed be rotationally flattened, violating the assumption of spherical symmetry inherent in Equation (5). Bezanson et al. (2018) showed that passive galaxies at z ∼ 1 have on average a higher proportion of rotational support (higher V/σ) than galaxies of the same mass at low redshift (see also van der Wel & van der Marel 2008). These results are consistent with the observed evolution of photometrically derived axis ratios over the same redshift range, which favor a significant portion of the quiescent galaxy population having 0.2 ≤ qint ≤ 0.3 (van der Wel et al. 2011; Chang et al. 2013; Hill et al. 2019). Belli et al. (2017) argued that the dynamical masses of quiescent galaxies at z > 1.5 are statistically consistent with a factor of ∼2 increase in V/σ compared to z = 0 based on their correlation with observed axis ratios. More directly, a handful of strongly lensed passive galaxies at z > 2 have resolved kinematic profiles that are consistent with being rotationally flattened disks (e.g., Newman et al. 2015; Toft et al. 2017; Newman et al. 2018).
In the case of integrated (as opposed to resolved) absorption-line kinematics, rotation is expected to manifest as a dependence of the measured velocity dispersion—and, by extension, dynamical mass—on galaxy inclination. For an oblate model observed at inclination i with no azimuthal variation of the velocity ellipsoid (i.e., σϕ/σr = 1, with σϕ and σr describing velocity dispersion in the azimuthal and radial directions, respectively), the second moment of the velocity distribution σobs can be written as
where V and σ are the flux-weighted mean circular velocity and velocity dispersion within the effective radius for the edge-on case (i = 90°) with , and β is the anisotropy parameter as defined in Section 3.3.2. Following Belli et al. (2017), substituting Equation (8) into Equation (4) and normalizing by the dynamical mass predicted for the face-on case (i.e., i = 0°) gives
with the relationship between qint, qobs, and i given by Equation (6). In the isotropic case where β = 0, Equation (9) reduces to Equation (5) of Belli et al. (2017) modulo a factor γ = 1.51.18
In Figure 9 we show the dynamical-to-stellar mass ratio as a function of qobs for both Mvir and MMFL estimates. In the virial theorem case we find evidence for a weak negative correlation between Mvir and qobs, with a Spearman rank coefficient ρ = −0.23 ± 0.10 (p = 0.007), while for the MFL models the correlation is not significant (ρ = −0.14 ± 0.10; p = 0.068). Individual galaxies are color-coded according to their Sérsic indices as derived from the profile fits described in Section 3.2.1. In contrast to Belli et al. (2017), we find no significant dependence of the dynamical-to-stellar mass ratio on Sérsic index in either case with ρ = −0.20 ± 0.10 (p = 0.04) and −0.08 ± 0.10 (p = 0.44), which appears to preclude a simple exclusion of disk-dominated systems based on their structure and motivates a more detailed examination of the correlation between qobs and dynamical-to-stellar mass ratio.
Download figure:
Standard image High-resolution imageLines in the top panel of Figure 9 show predicted behavior of the dynamical-to-stellar mass ratio for a galaxy with qint = 0.25 observed at different inclinations as given by Equation (9). We set β = 0.7(1−qint) based on the results of Cappellari et al. (2007) and Emsellem et al. (2011) for nearby fast-rotating early-type galaxies. In the case of an oblate system, V/σ and anisotropy are related by (Binney & Tremaine 1987; Binney 2005)
where α is a dimensionless number that quantifies the contribution of streaming motions to the line-of-sight velocity dispersion and (Wzz/Wxx) is a shape parameter related to the intrinsic axis ratio q as
We adopt a value of α = 0.15, which provides a good description for nearby galaxies (Cappellari et al. 2007). Models are offset in Mvir/MSPS to reflect a range of dynamical-to-stellar mass ratios. The predicted trends qualitatively reproduce the observed correlation between mass ratio and qobs, supporting previous statistical evidence of rotational support among a fraction of high-redshift quiescent galaxies (e.g., Belli et al. 2017).
In the bottom panel of Figure 9 the correlation between qobs and dynamical-to-stellar mass ratio for MFL models is notably weaker than in the virial theorem case, both visually and as measured by Spearman ρ, though galaxies with higher qobs still tend toward lower dynamical-to-stellar mass ratios. Unlike the virial theorem case, we can explicitly test the impact of intrinsic structure on our MFL mass estimates through application of a prior on qint in our modeling.19 The vertical lines in the bottom panel of Figure 9 show the effect of assuming that galaxies are intrinsically flat, with qint = 0.25, as opposed to the default case where we adopt a uniform prior on qint. For galaxies with low qobs, the effect of assuming a different intrinsic structure is minimal, but for galaxies with qobs > 0.6 (i < 55°) the estimated dynamical mass can increase by as much as ∼65% (0.22 dex), with a median increase of ∼15% (0.06 dex). The resulting correlation between dynamical-to-stellar mass ratio and qobs is also flatter, with ρ = 0.03 ± 0.06 (p = 0.95). Furthermore, assuming an intrinsically flat structure for these objects reduces the number of galaxies with dynamical masses significantly lower than their photometrically derived stellar mass MSPS.
In summary, the data considered here support the conclusions of previous studies suggesting that rotational support is prevalent among quiescent galaxies at high redshift (e.g., Chang et al. 2013; Newman et al. 2015; Belli et al. 2017; Toft et al. 2017; Newman et al. 2018; Hill et al. 2019). While we expect rotational flattening to have a minimal impact on dynamical mass estimates for galaxies with qobs < 0.6, galaxies with high qobs can have their dynamical masses underestimated by 0.2 dex or more depending on their intrinsic structure (i.e., if they are intrinsically spherical vs. flattened systems viewed face-on). As mentioned in Section 4.1, such a discrepancy between intrinsic and assumed structure can at least partially explain those galaxies in our sample that have dynamical masses formally less than their photometrically derived stellar mass, though it may not be the only factor affecting this comparison. Finally, we note that enforcing an intrinsically flat structure for all galaxies in our sample (e.g., qint = 0.25) shifts the results presented in Section 4.1.1 toward lower central dark matter fractions and cannot explain the apparent evolution of fDM[<re] without also appealing to significant changes in the stellar initial mass function (see Section 4.2).
4.2. The Normalization of the Stellar IMF at 1.4 ≤ z ≤ 2.1
In the case that we include an explicit dark matter component in our dynamical models, then we obtain an independent estimate of the stellar dynamical mass, M*,NFW, that can be used to diagnose changes in the normalization of the stellar IMF. A similar approach has been used to highlight possible IMF variation in low- and intermediate-redshift galaxies through the IMF offset parameter α ≡ M*,dyn/M*,IMF, where M*,IMF is the stellar mass computed for some default IMF (e.g., Treu et al. 2010; Thomas et al. 2011; Cappellari et al. 2013b; Conroy et al. 2013; Dutton et al. 2013; Spiniello et al. 2014; Smith et al. 2015). In our case α is measured with respect to the Chabrier (2003) IMF used in our SPS models (i.e., M*,IMF ≡ MSPS). While in principle α does not rely on any assumptions about how the IMF varies, significant deviations from a Salpeter-like IMF above 1–2 M⊙ are difficult to reconcile with observations of color and luminosity evolution for elliptical galaxies (e.g., Tinsley 1978; van Dokkum 2008). We therefore assume that any variation in the IMF occurs at stellar masses that contribute very little to the overall luminosity of the population, i.e., well below the MS turnoff, which is <2M⊙ for stellar populations older than ∼1 Gyr (the typical age for galaxies in our high-redshift sample; see, e.g., Mendel et al. 2015).
In Figure 10 we consider two limiting cases for the derivation of α: one where total mass follows the light profile and fDM[<re] = 0 (top panel), and a second where we include a static NFW dark matter halo following the procedure outlined in Section 3.3 (bottom panel). In each case we show the combined constraint obtained from stacking individual posterior probability distribution functions (PDFs) for galaxies in our high-redshift sample.
Download figure:
Standard image High-resolution imageWe find that the high-redshift data prefer an overall normalization of the IMF that is lighter than reported for nearby early-type galaxies of a similar mass ( ∼ 11), which tend to favor Salpeter (1955) or heavier IMFs (Conroy & van Dokkum 2012; Conroy et al. 2013; Cappellari et al. 2013a; Li et al. 2017; but see also Smith et al. 2015). There is an offset between the MFL and NFW models such that models including an explicit dark matter halo predict , consistent with a Chabrier IMF, while MFL models prefer a slightly heavier IMF normalization with . There is little evidence for the bottom-heavy IMFs that have been reported in the central regions (≲1/8re) of massive nearby early-type galaxies (e.g., van Dokkum et al. 2017; Parikh et al. 2018), which one might expect if all quiescent galaxies seen at z > 1 are the seeds of local massive ellipticals. We will discuss this further in Section 5.2.
As highlighted by Section 4.1.2, systematic differences in galaxy structure can influence the derivation of dynamical masses, and by extension our inferences about the IMF. In order to estimate the magnitude of this effect, we recomputed the stacked α PDFs, imposing a Gaussian prior on qint = N(0.25, 0.05) following the result of Chang et al. (2013); the results are shown as dotted–dashed lines in Figure 10. Assuming an intrinsically flattened structure for all galaxies results in a slightly heavier overall normalization of IMF, such that for the MFL (NFW) case (0.05 ± 0.03). It therefore seems unlikely that structural evolution alone can account for the apparent IMF differences between galaxies in our high-redshift sample and the cores of local early-type galaxies.
4.2.1. The Degeneracy between Central Dark Matter Fraction and IMF Normalization
One of the key assumptions in computing M*,NFW is that the dark matter halo component is well represented by an NFW profile, with no accounting for the possible influence of baryons on the dark matter profile shape. However, if the timescale for galaxy formation is long compared to the halo dynamical time, then the halo is expected to contract adiabatically as a result of baryonic collapse (e.g., Blumenthal et al. 1986; Gnedin et al. 2004). Dutton et al. (2016) argue that the dark matter halo can contract or expand depending on the relative balance of inflows, outflows, and feedback (see also Lovell et al. 2018), suggesting that our assumption of a static halo may bias the derived values of M*,NFW and, by extension, α. In this section we therefore explore a broader set of dynamical models that explicitly probe the effect of a variable dark matter halo response on our results.
In the case of spherical symmetry and circular dark matter particle orbits, the adiabatic invariant is given by rMtot(r)—where Mtot(r) is the total (baryonic plus dark matter) mass within radius r—so that rf/ri = Mtot,i(ri)/Mtot,f(rf). Therefore, given an initial mass distribution Mtot,i(r) and a final baryonic mass profile Mbar,f(r), we can derive the final dark matter profile MDM,f(r). Here we assume that the initial dark matter distribution is described by an NFW profile with mass and concentration parameter set by the scaling relations adopted in Section 3.3, and that the baryonic mass is distributed in the same way, i.e., Mbar,i(r) = fbarMtot,i(r), with fbar set by the stellar-to-halo mass relation. The final baryonic profile Mbar,f(r) is given by the deprojected MGE luminosity density scaled to match the galaxy stellar mass. We note that this assumes that star formation is distributed throughout the halo and is therefore likely an upper limit on the expected contraction.
If we assume no shell crossing of the dark matter, then MDM,i(ri) = MDM,f(rf), and the final mapping between rf and ri is given by
with following the generalized contraction formula suggested by Dutton et al. (2007). In this framework ν = 1 reproduces the standard adiabatic contraction derived by Blumenthal et al. (1986), while ν = 0.8 reproduces the modified contraction scenario described by Gnedin et al. (2004). ν = 0 is equivalent to an unmodified NFW profile. We also include a milder model for the halo response derived from the NIHAO simulations discussed by Dutton et al. (2015), such that
For each contraction model we solve for the mapping between rf and ri, and we use this modified dark matter profile as input to the JAM modeling procedure. While we do not explicitly include any models for halo expansion (i.e., ν < 0), our default MFL models can be interpreted as the extreme case where dark matter is completely evacuated within re, setting an upper limit for the dynamical effects of an expanded halo.
In Figure 11 we show the distributions of α derived for these different models of halo response, along with the standard MFL and NFW cases described in the previous section. The expected trend of a decreasing stellar contribution—that is, a lighter overall IMF normalization—with increasingly strong halo contraction is clearly visible, with pure adiabatic contraction (e.g., Blumenthal et al. 1986) predicting stellar masses that are a factor of >3 lighter than those obtained assuming a Chabrier (2003) IMF. Even the mildest model for halo response, Dutton et al. (2015), predicts values of α that are ∼60% lighter than Chabrier, and all of the contraction models considered here predict IMF normalizations that are lighter than observed or inferred for nearby stellar systems (e.g., Chabrier 2003; Bastian et al. 2010). Taken together, the results in Figure 11 suggest that any contraction of the dark matter halo due to gas inflow should be counterbalanced by equally violent outflows during the formation process.
Download figure:
Standard image High-resolution image5. Evolutionary Trends
The data presented in Figures 6–8 suggest an evolution of quiescent galaxy properties from high to low redshift, with high-redshift galaxies having on average lower central dark matter fractions and/or a lighter overall IMF compared to their low-redshift counterparts. These trends appear to persist even when considering only the oldest galaxies at low redshift. However, some care must be taken when making such comparisons, as individual galaxies are expected to evolve over the ∼8 Gyr separating our two samples. The challenge of connecting progenitor and descendant populations is therefore that it relies on having at least some a priori knowledge of how this evolution proceeds. Here we consider two different methods for connecting galaxies across redshift based on either their number density or central stellar velocity dispersion.
The stochastic nature of dark matter halo assembly in a ΛCDM cosmology leads to a broad range of plausible descendants for any single galaxy at high redshift. This diversity of assembly histories can bias inferred evolutionary trends when considering samples selected at either a single fixed or evolving (median) cumulative number density (e.g., Mundy et al. 2015; Wellons & Torrey 2017). As an alternative, one can identify descendants based on the full number density PDF as opposed to a single value, which we do here using NDPredict20 as described by Wellons & Torrey (2017). For each galaxy in our high-redshift sample, NDPredict provides an estimate of the likely descendant stellar mass distribution based on both expected evolution of the median number density and its scatter. This distribution is then used as a weight to select probable descendants from our low-redshift sample. We use the stellar mass functions published by Muzzin et al. (2013) to translate between stellar mass and number density at any given redshift, but we have verified that our results are insensitive to the particular choice of mass function.
As an alternative to the number density matching described above, we also construct a population of descendants matched at fixed central stellar velocity dispersion. Numerical studies have shown that central stellar velocity dispersion is relatively insensitive to assembly via dissipationless mergers (Hopkins et al. 2009b; Nipoti et al. 2012; Oser et al. 2012), which is expected to be the main growth channel for quiescent galaxies from high redshift. Hopkins et al. (2009b) argue that repeated dry mergers tend to decrease stellar velocity dispersions by at most 30%, as any increase in size leads to a corresponding increase in the dark matter content. To first order this assumption is consistent with the global trends shown in Figure 8. Based on these arguments, we identify likely descendants as those with σe within 0.05 dex of galaxies in our high-redshift sample, allowing for replacement—that is, the same low-redshift galaxies can be matched to multiple galaxies in our high-redshift sample.
5.1. Galaxy Evolution in Size, Stellar Mass, and σe
Assuming that high-redshift galaxies evolve to reproduce the typical properties of their matched descendants, we can use the relative difference in , MSPS, and σe to study the evolutionary processes acting on galaxies over their lifetimes. In the context of ΛCDM, individual quiescent galaxies are expected to evolve from z ∼ 2 to the present day through continued mergers after quenching. Bezanson et al. (2009) used scaling relations based on the virial theorem to argue that minor, gas-poor mergers can efficiently increase half-light sizes and decrease stellar velocity dispersions, a result that has been reiterated in a number of theoretical studies (e.g., Naab et al. 2009; Hopkins et al. 2009b; Oser et al. 2010, 2012; Hilz et al. 2013; but see also Nipoti et al. 2012). These predictions are supported by results that show that minor merging can explain the observed evolution of galaxy properties since z ≈ 2 (e.g., Damjanov et al. 2011; van de Sande et al. 2013; Belli et al. 2014a, 2017).
For each high-redshift galaxy we compute the average difference between its properties and those of its matched descendants. In Table 5 we quote the median evolution of MSPS, σe, and derived in this way for the full sample, with uncertainties estimated by jackknife resampling. There is little dependence of the inferred parameters on the method used to define descendant populations, with both the number density and velocity dispersion matched samples pointing toward a dramatic increase in size relative to stellar mass (a factor of ∼4.5 in compared to only ∼1.5 in MSPS). For samples matched on number density we can additionally infer the evolution of σe, which appears to decrease by only 12% on average from z > 1.4 to the present day, in good agreement with the predictions of numerical simulations (e.g., Hopkins et al. 2009b; Hilz et al. 2012).
Table 5. Inferred Evolution of Stellar Mass, Velocity Dispersion, and Size since 1.4 < z < 2.1
Parameter | Evol. Number | Fixed σe |
---|---|---|
Density | ||
0.24 ± 0.01 | 0.22 ± 0.03 | |
−0.05 ± 0.02 | ⋯ | |
0.68 ± 0.02 | 0.63 ± 0.03 |
Download table as: ASCIITypeset image
Figure 12 shows the pairwise distributions of MSPS, σe, and , where the observed evolution can be compared to simple energetic arguments for major and minor merging as in Bezanson et al. (2009, shown as inset coordinate arrows in each panel). The factor of 4–5 evolution in discussed above is immediately apparent, as is the comparably milder evolution in MSPS and σe. These results are quantitatively similar regardless of how we choose to identify low-redshift descendants (i.e., evolving number density or fixed σe), and they are consistent with the simple predictions for evolution by predominantly minor merging, where with α ≈ 2 (cf. α ≈ 1 for major mergers). In contrast, the distributions of both versus σe and σe versus MSPS suggest a more complicated interpretation, whereby those galaxies with or evolve relatively more in MSPS and/or σe than those with lower masses or velocity dispersions. The inclusion of velocity dispersion also complicates our otherwise straightforward interpretation of MSPS and evolution in terms of minor merging, although we note that more detailed numerical simulations find a range of behaviors depending on the properties of galaxies' host dark matter halos (e.g., Oser et al. 2012; Nipoti et al. 2012; Hilz et al. 2012, 2013). We will discuss these trends further in Section 5.2.
Download figure:
Standard image High-resolution imageIn order to investigate these differences further, in Figure 13 we show the same galaxy samples as in Figure 12, however this time in terms of and dynamical mass, often referred to as the mass plane. Galaxies have additionally been color-coded according to their dynamical mass-to-light ratio, M/L. Given the low dark matter fractions inferred for a majority of galaxies in our high-redshift sample, we expect that M/LMFL is primarily a tracer of stellar population variations. The mass plane provides a useful parameter space within which to understand systematic variations of galaxy properties and their evolution and has been used extensively to study the interrelationship between dynamics, structural properties, and stellar populations (e.g., Cappellari et al. 2013a; McDermid et al. 2015; Cappellari 2016; Belli et al. 2017; Scott et al. 2017).
Download figure:
Standard image High-resolution imageWe find that those galaxies with the highest M/L follow lines of roughly constant σe (dashed lines in Figure 13), while at lower M/L galaxies more closely follow lines of constant stellar mass surface density, Σe (dotted lines in Figure 13). Under the assumption that variations in (M/L)MFL are primarily driven by changes in the mean stellar age, these differences are in stark contrast to observational results at low redshift, where σe is by far the best predictor of galaxy stellar populations (e.g., Graves et al. 2009; McDermid et al. 2015; Scott et al. 2017). Nevertheless, numerous studies have shown that stellar surface density is a strong predictor of galaxy quiescence at low redshift (e.g., Cheung et al. 2012; Fang et al. 2013; Woo et al. 2015; but see also Wake et al. 2012), and high densities appear to be a necessary (if not sufficient) condition for quenching at high redshift (e.g., Barro et al. 2013; van Dokkum et al. 2015; Barro et al. 2017). It is therefore unsurprising to see such a correlation borne out in our M/L measurements: at any given epoch galaxies are likely added to the quiescent population as a function of their stellar surface density. Subsequent merger-driven assembly will then tend to move galaxies along lines of constant velocity dispersion (or steeper; see, e.g., Nipoti et al. 2012; Hilz et al. 2013).
5.1.1. Size Evolution as the Primary Driver of fDM
We showed in Section 4.1.1 and Figure 8 that, at fixed stellar mass, fDM[<re] increases by a factor of ∼2 from high redshift to the present day. However, based on the matched progenitor and descendant samples described in Section 5, we find that the growth of fDM[<re] for individual galaxies is likely even larger, reaching 0.64 ± 0.05 dex (0.52 ± 0.07 dex) on average for progenitors and descendants matched on evolving stellar mass (fixed σe).
Just as major and minor mergers are expected to have different effects on a given galaxy's evolution in size, stellar mass, and stellar velocity dispersion (e.g., Figure 13), they also have a distinct influence on the evolution of fDM. Hilz et al. (2012, 2013) show that minor mergers can dramatically increase fDM[<re], so that a factor of 2 growth in stellar mass can nearly double the central dark matter fraction. In contrast, a single equal-mass (major) merger may only increase fDM[<re] by 50%. This difference is driven by the relatively efficient growth of sizes in minor compared to major mergers; for an NFW-like halo the dark matter mass within a given (small) radius r scales roughly with the virial mass of the halo as (Boylan-Kolchin et al. 2005). In an equal-mass merger where both re and the stellar mass within re double, the dark matter mass within re can increase by up to a factor of 8 (assuming that the stellar mass and halo mass increase by a similar amount). By comparison, a similar doubling of stellar mass through multiple minor mergers can increase re by a factor of 4 and the enclosed dark matter mass by a factor of ≳30.
These differences are particularly apparent when viewed in terms of stellar mass surface density, Σe (), and fDM[<re], which we show in Figure 14. We find that Σe and fDM[<re] are anticorrelated, in the sense that those galaxies with the highest stellar mass surface densities have the lowest fDM[<re]; a similar anticorrelation has been demonstrated previously for both late- and early-type galaxies (e.g., McGaugh 2005; Sonnenfeld et al. 2015). Figure 14 additionally shows that Σe decreases by ∼1 dex on average from z > 1.5 to z ≈ 0, which is a direct consequence of the size and stellar mass evolution discussed in Section 5.1 (see also Table 5 and Figure 12). This can be compared to the results of Remus et al. (2017), who show that simulated galaxies follow well-defined tracks in Σe–fDM regardless of redshift (shown as dashed and dotted–dashed lines in Figure 14). Nevertheless, the simulation results roughly reproduce the observational trends at any given epoch, with the primary difference between the two simulations discussed by Remus et al. (2017) being their treatment of black hole feedback.
Download figure:
Standard image High-resolution imageWe can use simple scaling relations to better understand galaxies' expected evolution in Figure 14 given various assumptions. Following Figure 12, we adopt a simple model for size growth during mergers such that , with α = 1 or 2 for major and minor mergers, respectively. We additionally assume that the enclosed dark matter mass scales with the total (virial) mass of the halo and radius as MDM(r) ∝ Mvirr2, and that the stellar and halo mass grow at the same rate (). Arrows in the right panel of Figure 14 show the predicted evolution for a galaxy doubling its stellar mass through either a single major merger or successive 10:1 (minor) mergers. As expected, the efficient size growth associated with minor mergers in our toy model drives rapid evolution in both Σe and fDM[<re]. However, while minor mergers can explain the observed decrease in Σe from high to low redshift, our toy model overpredicts the increase of fDM[<re]. Tortora et al. (2018) report a similar result for galaxies at z ≈ 0.7 and suggest that allowing for variation in the stellar-to-halo mass ratio of accreted galaxies (i.e., ) can help to lessen the tension between predicted and observed evolution in fDM. This is particularly likely for the massive quiescent galaxies in our high-redshift sample, which are expected to trace the most massive halos at their respective redshifts (e.g., Lin et al. 2019).
5.2. Evolution of the IMF at Fixed σe
In Section 4.2 we showed that the kinematics of massive quiescent galaxies at z > 1.5 are consistent with an MW-like IMF on average. In contrast, a number of studies based on stellar absorption features, lensing, and dynamical modeling have shown that a Salpeter or even heavier IMF may be more typical in the inner 0.1–0.2 Re of present-day early-type galaxies (e.g., Thomas et al. 2011; Cappellari et al. 2012, 2013a; Conroy & van Dokkum 2012; Spiniello et al. 2014; Conroy et al. 2017; van Dokkum et al. 2017; but see also Smith et al. 2015), suggesting some tension between our high-redshift results and those in the nearby universe.
Figure 15 shows a comparison of the IMF offset parameter, α, as a function of stellar velocity dispersion, σe, at different redshifts. We include in this figure constraints derived from low-redshift galaxies in the ATLAS3D and SLACS samples by Posacki et al. (2015), as well as binned results for the matched GAMA/SDSS samples described in Section 5. This comparison shows that the apparent offset in mean α between high and low redshift depends on σe, such that galaxies with the highest velocity dispersions generally have α values consistent with observations at low redshift, while lower velocity dispersion galaxies are offset toward lower α—that is, toward a "lighter" IMF—at fixed stellar velocity dispersion. Assuming an intrinsically flat axis ratio for high-redshift galaxies (e.g., qint = 0.25; bottom panel of Figure 15) reduces the apparent offset of α at low σe, though the qualitative trend remains unchanged. We note that this offset is unlikely to be the result of (correlated) uncertainties in σe and α, shown by the error ellipse in the lower right corner of Figure 15, which tend to scatter galaxies along the low-redshift α–σe correlation rather than away from it. It may be that the differential evolution seen between high- and low-σe galaxies is tied to the overall buildup of the velocity dispersion function (VDF) over time. Bezanson et al. (2012) show that galaxies with form early and their number density has changed little since at least z ≈ 1.5, while the population at lower σe evolves significantly owing to the addition of newly quenched galaxies. A more detailed dissection of the α–σe relation in the context of galaxy SFHs will be the subject of future work.
Download figure:
Standard image High-resolution imageAn interesting implication of Figure 15 is that the α–σe relation found at low redshift is established early on, and any scenario invoked to explain quiescent galaxies' subsequent evolution in , σe, and MSPS (e.g., Figure 12) should largely preserve the underlying correlation; this is especially true at the highest stellar velocity dispersions. Using high signal-to-noise ratio, long-slit spectra of six nearby early-type galaxies, van Dokkum et al. (2017) found evidence for strong radial gradients in α that increased from MW-like at R > 0.4Re to Salpeter or heavier at 0.1Re (see also Martín-Navarro et al. 2015b; Lyubenova et al. 2016). Under the assumption that gas-poor mergers primarily contribute to the buildup of an extended stellar envelope (e.g., Hopkins et al. 2009a; van Dokkum et al. 2010; Karademir et al. 2019), then we would expect that the remnants of our high-redshift sample should survive in the cores of massive nearby ellipticals. Indeed, the typical half-light sizes of galaxies in our sample (1–1.5 kpc) are comparable to the physical extent over which the IMF is found to vary significantly in van Dokkum et al. (2017).
Finally, although we have tried to be comprehensive in our measurement of stellar masses, we cannot rule out that at least some part of the evolutionary trends implied by Figure 15 could be the result of redshift-dependent systematic uncertainties in our derivation of MSPS. These could be, for example, due to our adopted SFH or stellar population synthesis models (e.g., Pforr et al. 2012; Leja et al. 2019). Based on the analysis of mock galaxy spectra generated from semianalytic models, Pforr et al. (2012) showed that stellar masses for passive galaxies can be recovered to better than ∼0.05 dex for a wide range of SFHs, suggesting that our results are unlikely to be due entirely to the details of our SFH modeling. Furthermore, while Pforr et al. (2012) show that mismatches in metallicity can lead to systematic offsets of up to 0.2–0.3 dex, direct observations at z > 1.5 support our adoption of a solar-metallicity template library (e.g., Onodera et al. 2015; Kriek et al. 2016).
6. Conclusions
We present an analysis of 58 massive quiescent galaxies at 1.4 < z < 2.1 with measured stellar velocity dispersions and deep near-infrared HST imaging. We use these data to study the evolution of dynamical masses and the dynamical-to-stellar mass ratio, which are sensitive to the central dark matter fraction and normalization of the stellar IMF. We find the following:
- (i)The median dynamical-to-stellar mass ratio of quiescent galaxies is lower by ∼0.2 dex at 1.4 < z < 2.1 compared to low redshift. In Figures 6 and 7 we showed that this offset appears to be independent of the method used to derive dynamical masses (e.g., Jeans models vs. virial mass estimates). The observed evolution is consistent with a decrease in the fraction (by mass) of dark matter within the galaxy effective radius, fDM[<re], which is lower by a factor of ~2 in our high-redshift sample compared to nearby galaxies in the SDSS/GAMA surveys (fDM[<re] = 6.6% ± 1.0% at z ≈ 1.8 vs. 16.3% ± 0.3% at z = 0) at fixed stellar mass. These measurements appear consistent with recent results based on the rotation curves of high-redshift star-forming galaxies. Based on the matching of progenitor and descendant populations, we argue in Section 5.1.1 that the evolution of individual galaxies is likely even larger, with fDM[<re] increasing by a factor of 4–5 from high redshift to the present day.
- (ii)Under the assumption that central dark matter fractions are intrinsically low in high-redshift galaxies, the dynamical-to-stellar mass ratio can be used as a probe of the stellar IMF. For MFL models, we find that high-redshift data are consistent with a Kroupa-like IMF on average, while models including an explicit NFW dark matter halo are consistent with a Chabrier IMF (see Figure 10). We find a correlation between stellar velocity dispersion and the IMF offset parameter, α, at high redshift that is consistent with low-redshift data, suggesting that any subsequent evolution should act to preserve this underlying correlation (see Figure 15). We argue that minor mergers are the most likely drivers of galaxies' growth in and MSPS, as they primarily add material at large radii while preserving the stellar populations in the inner regions.
- (iii)Simple models for the contraction of dark matter halos in response to baryonic collapse predict high central dark matter fractions. In the most extreme case of pure adiabatic contraction, such models require IMF normalizations a factor of ∼3 lighter than Chabrier to explain the observed kinematics (see Figure 11). Significant contraction of the dark matter halo is difficult to accommodate given current observational constraints unless other baryonic process (e.g., outflows or AGN feedback) act to reduce the central dark matter content after collapse.
- (iv)A comparison of kinematics and structural properties between high-redshift galaxies and their likely descendants at low redshift supports minor merging as the dominant evolutionary pathway after quenching; however, those galaxies with the highest stellar masses and/or stellar velocity dispersions appear to evolve relatively more than lower-mass/dispersion objects. This separation is apparent for descendants matched both on evolving number density and at fixed σe (see Figure 12).
- (v)In the two-dimensional parameter space of size and dynamical mass—the so-called "mass plane" (Figure 13)—high-redshift galaxies both are smaller and have lower dynamical masses than their low-redshift descendants. Separating the galaxy population in terms of total mass-to-light ratio (M/L), which we interpret here as a proxy for stellar population age, galaxies with the highest M/L follow lines of roughly constant σe, while those with lower M/L follow more closely lines of constant stellar surface density. We interpret these differences as being driven by two separate phases of passive galaxy formation, whereby galaxies first quench as a strong function of their stellar mass surface density and their subsequent evolution on the mass plane is driven by minor merging.
Taken together, our results point toward an evolutionary scenario for massive quiescent galaxies that sees their formation occurring rapidly at z > 2. Subsequent quenching of star formation appears to preserve the disk-like structural and kinematic signatures associated with massive star-forming galaxies at those redshifts (e.g., Wuyts et al. 2011; Wisnioski et al. 2015, 2019), as well as their dark matter properties (e.g., Wuyts et al. 2016; Genzel et al. 2017, 2020; Lang et al. 2017). Subsequent evolution through continued merging is then required to transform both their structural and kinematics properties to reproduce the massive, predominantly slowly rotating galaxies that constitute their likely descendants at z = 0 (e.g., Veale et al. 2017). Based on galaxies' evolution in size, stellar mass, stellar velocity dispersion, and central dark matter fraction, it appears that the most likely channel for this evolution is through the accretion of relatively lower mass galaxies (i.e., minor mergers). While these results are based on the best currently available kinematic and photometric data, future spectroscopic observations with the James Webb Space Telescope and 30+ m ground-based observatories will enable systematic surveys of high-redshift stellar kinematics.
We thank the referee for constructive feedback that helped to improve the overall quality of this manuscript, as well as Sandesh Kulkarni and Kaushi Bandara for their help with KMOS observations for the VIRIAL survey. J.T.M. acknowledges the support of the Australian Research Council Centre of Excellence for All Sky Astrophysics in 3 Dimensions (ASTRO 3D), through project No. CE170100013. D.J.W. and M.F. acknowledge the support of the Deutsche Forschungsgemeinschaft via Project IDs 3871/1-1 and 3871/1-2. M.F. has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (grant agreement No. 757535).
This work is based on observations taken by the CANDELS Multi-Cycle Treasury Program (GO 12060 and 12099) and 3D-HST Treasury Program (GO 12177 and 12328) with the NASA/ESA HST, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555. It also includes data obtained from the Hubble Legacy Archive, which is a collaboration between the Space Telescope Science Institute (STScI/NASA), the Space Telescope European Coordinating Facility (ST-ECF/ESA), and the Canadian Astronomy Data Centre (CADC/NRC/CSA).
GAMA is a joint European-Australasian project based around a spectroscopic campaign using the Anglo-Australian Telescope. The GAMA input catalog is based on data taken from the Sloan Digital Sky Survey and the UKIRT Infrared Deep Sky Survey. Complementary imaging of the GAMA regions is being obtained by a number of independent survey programs, including GALEX MIS, VST KiDS, VISTA VIKING, WISE, Herschel-ATLAS, GMRT, and ASKAP, providing UV to radio coverage. GAMA is funded by the STFC (UK), the ARC (Australia), the AAO, and the participating institutions. The VISTA VIKING data used here were taken using ESO Telescopes at the La Silla Paranal Observatory under program ID 179.A-2004. The GAMA website is http://www.gama-survey.org/.
Funding for the SDSS and SDSS-II has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, the U.S. Department of Energy, the National Aeronautics and Space Administration, the Japanese Monbukagakusho, the Max Planck Society, and the Higher Education Funding Council for England. The SDSS website is http://www.sdss.org/.
The SDSS is managed by the Astrophysical Research Consortium for the Participating Institutions. The Participating Institutions are the American Museum of Natural History, Astrophysical Institute Potsdam, University of Basel, University of Cambridge, Case Western Reserve University, University of Chicago, Drexel University, Fermilab, the Institute for Advanced Study, the Japan Participation Group, Johns Hopkins University, the Joint Institute for Nuclear Astrophysics, the Kavli Institute for Particle Astrophysics and Cosmology, the Korean Scientist Group, the Chinese Academy of Sciences (LAMOST), Los Alamos National Laboratory, the Max-Planck-Institute for Astronomy (MPIA), the Max-Planck-Institute for Astrophysics (MPA), New Mexico State University, Ohio State University, University of Pittsburgh, University of Portsmouth, Princeton University, the United States Naval Observatory, and the University of Washington.
Appendix A: Additional Tests of Velocity Dispersion Measurements
In this appendix we explore possible systematics in our derived velocity dispersions stemming from assumptions made in our spectrophotometric modeling approach. There are two effects in particular that we are concerned with: the impact of assuming a fixed solar metallicity for our SED template library, and the effect of our adopted (parametric) SFH.
A.1. Metallicity Effects
While the majority of low-redshift massive early-type galaxies are consistent with solar metallicity or higher (e.g., Gallazzi et al. 2006; Thomas et al. 2010), the picture at high redshift is still uncertain, with different results suggesting variations in total metallicity of up to 0.8 dex (e.g., Lonoce et al. 2015; Onodera et al. 2015; Kriek et al. 2016; Morishita et al. 2018). Given our adoption of a fixed solar metallicity for our template grid, it is therefore worthwhile to investigate the possible impact of this assumption on our derived velocity dispersions. In Figure A1 we show a comparison of stellar velocity dispersions obtained with templates a factor of ∼2 higher or lower in metallicity. While there is a clear systematic shift in the resulting values of σ*, the offsets are of order a few percent, significantly smaller than the typical 20% uncertainties on our measurements of σ*, and are therefore unlikely to bias our results.
Download figure:
Standard image High-resolution imageA.2. Star Formation History Effects
While the adoption of a parametric SFH to describe the observed photometry is common practice, the use of such models in performing kinematic measurements is less common and bears further investigation. In brief, we modified the Markov Chain Monte Carlo (MCMC) fitting code described in Section 2.1.1 to construct the best-fit template from a linear combination of simple stellar population templates, as opposed to the parametric SFHs adopted previously. This approach mimics the internal fitting procedure adopted in well-known fitting codes such as pPXF (Cappellari 2002), while at the same time providing improved handling of low signal-to-noise data thanks to the MCMC sampling of the line-of-sight velocity distribution. We show the results of this refitting in Figure A2.
Download figure:
Standard image High-resolution imageThere is significant scatter between the two velocity dispersion estimates, but in nearly all cases the derived σ* values are consistent within uncertainties. The values derived in our default modeling are larger by ∼0.06 dex on average; however, this does not significantly affect our results.
Appendix B: Comparisons with Previous Structural Parameter Measurements
In this appendix we present a comparison of the galaxy structural parameters derived here with those available in the literature (where available).
B.1. High Redshift
The comparison of apparent magnitude, Sérsic index, and size for galaxies in our high-redshift sample is shown in Figure B1. For most galaxies these quantities were taken from van der Wel et al. (2014), which are based on the same CANDELS/3D-HST WFC3/F160W imaging. As discussed in Section 2.2, several galaxies from Bezanson et al. (2013a) and van de Sande et al. (2013) fall outside the CANDELS/3D-HST footprint, and there our measurements are based on pipeline-processed data retrieved from the HLA. In most cases the agreement between different measurements is excellent.
Download figure:
Standard image High-resolution imageB.2. Low Redshift
Figures B2 and B3 show a comparison of apparent r-band magnitude, Sérsic index, and size for galaxies in our low-redshift SDSS/GAMA sample with measurements from Meert et al. (2015) and Simard et al. (2011), respectively. While there is generally good agreement between measurements in the different structural catalogs, there is significant scatter driven by differences in approach used to mask/model neighboring objects, the size of fitted images, and the method for measuring the sky background. In most cases any systematic biases are relatively small, ≲20%, and do not affect the conclusions of this work.
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageAppendix C: Photometric Fits for High-redshift Galaxies
In Figure C1 we show the Sérsic and MGE fits derived for each galaxy in the high-redshift sample.
Download figure:
Standard image High-resolution imageFootnotes
- *
Based on observations obtained at the Very Large Telescope (VLT) of the European Southern Observatory (ESO), Paranal, Chile (ESO program IDs 092.A-0091, 093.A-0079, 093.A-0187, and 094.A-0287).
- 12
- 13
The dispersions quoted in Table 2 have been corrected for the effects of seeing and scaled to the luminosity-weighted mean within the half-light radius following the procedure outlined by van de Sande et al. (2013). The derived corrections range between 1.02 and 1.1 and are consistent with similar corrections derived directly from the dynamical modeling discussed in Section 3.3.2.
- 14
- 15
- 16
Source-specific PSFs were extracted from the SDSS drField files using the read_PSF routine described at http://classic.sdss.org/dr7/products/images/read_psf.html.
- 17
Available at http://purl.org/cappellari/software.
- 18
Belli et al. (2017) adopt the value of γ determined by Cappellari et al. (2013b), which relates the measured second moment σe to the circular velocity at . Here we instead define V/σ in terms of the flux-weighted mean within re, so that all of σobs, V, σ, and (V/σ)e are defined over the same aperture.
- 19
Functionally speaking, we enforce different intrinsic axis ratios by deprojecting our MGE models at inclination i given qobs and qint by inverting Equation (6).
- 20
Available at https://github.com/sawellons/NDpredict.