THE RISE AND FALL OF PASSIVE DISK GALAXIES: MORPHOLOGICAL EVOLUTION ALONG THE RED SEQUENCE REVEALED BY COSMOS

The COSMOS field (10^h00^m286, + 02°12'210) is the largest contiguous area (1.64 deg²) imaged by the HST and is almost 20 times the size of the two GOODS fields combined. As such, it marks an unprecedented step forward in studies of the morphologies of distant (z ∼ 1) galaxies at high resolution. Specifically for this work, COSMOS provides such large samples that we can statistically track evolving sub-samples of field galaxies defined by both rest-frame color as well as morphology as a function of M_* and redshift from z ∼ 1 to z ∼ 0.1.

A large team of scientists has assembled to collaborate on a wide range of topics using observations from the COSMOS field (for a review, see Scoville et al. 2007b). The primary data sets of interest for this work are the i_814W observations from the HST Advanced Camera for Surveys (ACS; Scoville et al. 2007a; Koekemoer et al. 2007), which provide morphological information, and the ground-based optical to near-IR photometry presented in detail in Capak et al. (2007b) and updated in Ilbert et al. (2009). As we discuss further below, the ground-based catalog provides the basis for high-precision photometric redshifts and a set of rest-frame colors that separate truly passive stellar populations from those that are actively star-forming but dusty.

We briefly review the relevant components of the ground-based catalog, referring to Capak et al. (2007b) and Ilbert et al. (2009) for details. The catalog includes u*- and i*-band imaging from Mega-Prime (Aune et al. 2003; Boulade et al. 2003) and was cross-referenced with K_s-band observations (McCracken et al. 2010) from the Wide-field InfraRed Camera (WIRCAM; Puget et al. 2004), both of which are instruments on the Canada–France–Hawaii Telescope (CFHT). The majority of photometric bands in the catalog were observed with the Subaru Telescope equipped with Suprime-Cam (Komiyama et al. 2003; filters B_J, V_J, g⁺, r⁺, i⁺, z⁺). The catalog was selected based on detections in the Subaru i⁺ band and the seeing of the final mosaics is better than 15 in all cases. Capak et al. (2007b) point-spread function (PSF) match all data, optimizing for photometry performed in 3'' diameter apertures. All of the optical bands have a 5σ point-source (3'' diameter aperture) depth of at least 25th magnitude (AB) with the K_s band limited to K_s < 24. Given these depths, the final sample completeness is determined mostly by the ACS data because morphological classification begins to degrade at magnitudes fainter than i_814W ≈ 24. The HST ACS imaging is described in Scoville et al. (2007a) and Koekemoer et al. (2007), and is used only for morphological classification in this work. We use the Zurich Estimator of Structural Types (ZEST) morphology catalog (Scarlata et al. 2007a)—discussed further in Section 3.4—which is based on the ACS source catalog presented in Leauthaud et al. (2007) who estimate the ACS depth at i_814W = 26.1 for photometry of a 1'' galaxy.

In the analysis that follows, our primary sample is based on the i-selected catalog from Capak et al. (2007b). We impose magnitude limits of K_s < 24 and i < 26.5 and divide our sample into six redshift bins: [0.2–0.4), [0.4–0.6), [0.6–0.8), [0.8–1.0), [1.0–1.1), [1.1–1.2). Given the magnitude cuts, we estimate the mass completeness of each redshift bin by considering the observed magnitude of a maximal M_*/L stellar population model with solar metallicity, no dust, and a τ = 0.5 Gyr burst of star formation that occurred at z_form = 5. This exercise provides limits that roughly match the 80% completeness limits determined when deeper samples are available (Bundy et al. 2006), and in this case gives values of 8.8, 9.3, 9.7, 10.0, 10.1, and 10.2 in units of log M_*/M_☉.

In practice, morphological classification is not possible for every source down to the K_s- and i-band limits. This means that the effective M_* above which morphological samples are complete is higher than the full sample. These depend on the nature of the source, however, but can be easily defined with respect to the deeper K_s < 24 and i < 26.5 sample. We consider the morphological sample incomplete when the fraction of galaxies with a successful ZEST morphological classification (see Section 3.4) falls below 50% of the primary sample. We also limit the sample to ensure the robustness of ZEST classifications using ellipticity distributions tests (see Section 4.2). For each of the redshift bins defined above, we adopt mass limits of 9.7, 9.9, 10, 10.4, 10.5, and 10.8 in units of log M_*/M_☉. After removing stars and bright AGN, the number of galaxies in the primary sample in each redshift bin is listed in Table 1.

2.2.1. Cosmic Variance

While a spectroscopic survey, zCOSMOS, is ongoing in the COSMOS field (Lilly et al. 2007), the large samples required for this work necessitate the use of photometric redshifts. We use the catalog of photo-z's published by Ilbert et al. (2009) which are based on 30 bands of imaging in the COSMOS field. In addition to the observations mentioned in Section 2, these include UV bands from the Galaxy Evolution Explorer (GALEX), four mid-IR bands (3.6–8 μm) from the Infrared Array Camera (IRAC) on the Spitzer Space Telescope, J-band imaging from the Wide Field IR Camera (WFCAM) on the UK IR Telescope (UKIRT), and an additional 12 medium-band and two narrowband filters observed with Suprime-Cam on Subaru. This extended data set is described in detail by P. Capak et al. (2010, in preparation).

Redshift estimates were obtained with the Le Phare¹⁴ code using templates drawn from Polletta et al. (2007). Ilbert et al. (2009) develop a novel technique that accounts for the contribution of emission line flux (especially important in the narrow and medium filter bands) to the spectral energy distributions (SEDs) of star-forming galaxies. This correction improves the photo-z precision by a factor of ∼2. When comparing to zCOSMOS spectroscopic redshifts with 17.5 ⩽ i⁺ ⩽ 22.5, Ilbert et al. (2009) find a precision of σ_Δz/(1 + z) ≈ 0.01 after removing an outlier fraction that accounts for 0.6% of the comparison sample. They also compare to spec-z's obtained with Keck DEIMOS (Faber et al. 2003) for a sample of very faint and 24 μm selected sources. At i⁺ ≈ 25.5 the uncertainties are estimated at σ_Δz/(1 + z) ≈ 0.06.

We use the Bayesian code described in Bundy et al. (2006) to estimate the current mass in stars of the galaxies in our sample. The observed SED of each galaxy is compared to a grid of 13,440 models from the Bruzual & Charlot (2003, hereafter BC03) population synthesis code that spans a range of metallicities, star formation histories (parameterized as exponentials), ages, and dust content. No bursts are included in our models and only those models with ages (roughly) less than the cosmic age at each redshift are considered. We use a Chabrier initial mass function (IMF; Chabrier 2003) and assume a Hubble constant of 70 km s⁻¹ Mpc⁻¹.

At each grid point, the K_s-band M_*/L_K ratios, inferred M_*, and probability that the model matches the observed SED is stored. This probability is marginalized over the grid, giving the stellar mass probability distribution, the median of which is taken as the final estimate of M_*. The width of the distribution provides the uncertainty which is typically 0.1–0.2 dex. This is added in quadrature to the K_s-band magnitude uncertainty to determine the final error on M_*. Stellar mass estimates for galaxies with photo-z's also suffer from the uncertainty in luminosity distance introduced by the photo-z error and the possibility of catastrophically wrong redshift information (Bundy et al. 2005; Kitzbichler & White 2007).

More broadly, any stellar mass estimate suffers potential systematic errors¹⁵ from uncertainties inherent in stellar population modeling and various required assumptions, such as the form of the IMF. Several papers have stressed the importance of treating thermally pulsating asymptotic giant branch stars (TP-AGBs; e.g., Maraston et al. 2006), an element that is missing in the Bruzual & Charlot (2003) models. The recent and thorough investigation of population synthesis modeling by Conroy et al. (2009), however, argues that M/L ratios estimated from BC03 are largely resistant to the uncertain contribution from TP-AGBs as well as other limitations of current models. A further test is provided in Drory et al. (2009) who compare mass functions computed using BC03 and BC07 stellar population codes and conclude that ∼20% differences in abundances can be introduced, especially for star-forming galaxies. It is also important to recognize that M_* estimates may be affected by unrecognized systematic uncertainties at the 0.1–0.2 dex level.

Because our goal in this paper is to study disk galaxies that have stopped forming stars, in this section we discuss the rest-frame galaxy colors we use to identify passive stellar populations. Previous work has often relied on a single rest-frame UV-optical color to select passive galaxies, but this technique leads to possible contamination from dusty star-formers (e.g., Yan et al. 2006). In this work, we make use of color–color diagrams that include rest-frame UV, optical, and near-IR colors, inspired by recent work showing how this approach separates dusty star-formers from truly passive populations (e.g., Pozzetti & Mannucci 2000; Labbé et al. 2005; Wuyts et al. 2007; Williams et al. 2009).

The utility of the rest-frame UV–optical–near-IR "bicolor" diagram for the COSMOS data set is illustrated in Figure 1 where we show interpolated rest-frame colors in various redshift bins using the following filter bands: the GALEX near-UV band (NUV), the Suprime-Cam r⁺ band, which we label as R, and a generic J-band filter.¹⁶ Rest-frame NUV- and R-band absolute magnitudes were estimated by Ilbert et al. (2009) during the SED fitting necessary for measuring photo-z's. To derive the rest-frame J-band absolute magnitudes needed for the NUV–R–J diagram, we used the Kcorrect software package (Blanton & Roweis 2007) and the full suite of photometry described in Section 2. The broad wavelength coverage of COSMOS ensures that the flux corresponding to a given rest-frame filter is well sampled. A similar diagram was discussed in Ilbert et al. (2010).

Zoom In Zoom Out Reset image size

The upper left portion of each redshift panel in Figure 1 exhibits a relatively tight clump of galaxies that we associate with passive stellar populations. A sequence of star-forming galaxies is evident below and to the right of the clump in each panel. As a test, we have colored in red those galaxies with SEDs best fit by early "spectral" type templates, as determined by the photo-z code (this category includes types 1–8 in the COSMOS photo-z fits; see Ilbert et al. 2009). Highly star-forming templates are shown in blue. At most redshifts, the early-type SED designation overlaps well with the associated passive clump. Note that the location and orientation of the passive clump and star-forming sequence evolves somewhat given our choice of filters. In what follows, we will isolate passive clump galaxies (also referred to hereafter as the "red sequence") using a joint color cut that evolves accordingly and is indicated by the two lines in Figure 1. Recognizing that such color cuts are somewhat arbitrary, our goal is to evenly split the now meandering "green valley" that separates the two populations. We specify our thresholds in terms of the given filters because they correspond well to our observations and do a good job of making the passive clump distinct, although we acknowledge that a more general treatment would be valuable in future work. Still, we emphasize that adjusting the details of the cuts defined below has little effect on the analysis that follows.

The horizontal NUV − R color cut is tuned to the usual luminosity-dependent bimodality (e.g., Willmer et al. 2006) in this color and is defined as follows:

$$\begin{equation} \mbox{NUV} - R > 4.2(1+z)^{-0.43} - 0.2(M_K + 20), \end{equation} \tag{ 1 }$$

where M_K is the rest-frame absolute K_s-band magnitude. Passive galaxies must also satisfy the following diagonal cut in Figure 1:

$$\begin{equation} \mbox{NUV} - R > C_1(z) + C_2(z)(R-J), \end{equation} \tag{ 2 }$$

where the constants, C₁ and C₂, have been chosen by inspection in redshift bins of width 0.2 with central values of z = [0.30, 0.50, 0.70, 0.85, 0.95, 1.10]. They are given by C₁(z) = [4.4, 4.2, 4.0, 3.9, 3.8, 4.2] and C₂(z) = [2.41, 2.41, 2.5, 2.6, 3.0, 3.7]. At z ≲ 0.6 the combination of the color cuts above select 10%–20% more passive galaxies than those with early-type SED templates, many of them scattered between star-forming and passive sequences. At higher redshifts, however, the SED method appears to miss genuine galaxies in the passive clump, as many as 20% at z ≈ 1. We note that our conclusions regarding passive disk galaxies are unchanged if we use the SED fits to define the passive population.

Finally, we have checked for obscured star formation among the passive population as indicated by MIPS 24 μm emission. We note that a significant fraction of 24 μm emission, especially among the passive population, might arise from AGNs (e.g., Fiore et al. 2008). In addition, among star-forming galaxies, the implied SFR from 24 μm emission must be added to that from the UV flux which will dominate the output luminosity from star formation in our sample (e.g., Noeske et al. 2007). As a result, our constraints on hidden star formation in passive galaxies compared to star-formers are upper limits. Using the morphological classifications defined below (grouping "early" and "late" disks together), our first test compares 3σ MIPS detection rates of passive versus star-forming disk galaxies. For z>1, no passive disks are detected with MIPS while the detection rate among star-forming disks is roughly 10%. In our lowest redshift bin, 0.2 < z < 0.4, 1% of passive disks are detected compared to 10% of star-formers. At redshifts in between, the highest detection rates of passive disks occur at 0.4 < z < 0.6, where 5% are detected compared to 8% for star-forming disks. Considering the full redshift range, these numbers confirm that the color–color methodology described above does not lead to significant contamination from hidden star formation. However, given the small numbers of MIPS detections, it is not possible to compare 24 μm flux as a function of M_* or put constraints on the relative amount of SFR hidden among passive population but below the MIPS detection limits.

For this, we perform a stacking analysis of the MIPS data in bins of stellar mass (of width 0.5 dex) and in the redshift intervals used throughout (the minimum number of galaxies per bin is roughly 20). Using Rieke et al. (2009), we estimate mean sSFR given the stacked 24 μm emission. Across our ∼30 stacked bins, the average passive sSFR is roughly an order of magnitude less than the star-forming sSFR. In two bins this difference approaches a factor of 2, and in several other bins, there is no stacked MIPS detection for passive galaxies. We note that the IR/SFR calibrations become uncertain at low 24 μm luminosities below log L₂₄/L_☉ <8.5 and several of our bins reach this limit. With this caveat and again ignoring contributions from AGN, our stacking analysis provides a rough upper limit to the possible amount of hidden star formation (SF) in the average passive galaxy that corresponds to growth rates of 3%–5% in M_* per Gyr.

Understanding the morphological makeup of the galaxy population is crucial to our goals in this paper. Given the large samples (typically more than 10⁵ galaxies) made possible by the COSMOS data, automated morphological classifiers figure prominently. It is well known, however, that techniques such as CAS (e.g., Conselice 2003) or Gini/M20 (see Abraham et al. 2007) often disagree when compared to visual morphologies and can differ from each other. Indeed, Ilbert et al. (2010) use two different classifiers—the Gini/M20 method described in Abraham et al. (2007) and a CAS-like technique used in Cassata et al. (2007)—to gauge the systematic uncertainty of broad categories such as "spheroidal" or "disk-like." Their analysis of the evolving distribution of morphologies in COSMOS from z ∼ 1 serves as the basis for this work. We should also note that "wiki-morphologies" of the kind pioneered by GZ (Lintott et al. 2008) are an alternative way to obtain visual morphologies for large samples.

In this work, we are specifically interested in disk components living in passive galaxies. As such, the exact definitions of morphological type are less important than confidence in the ability of the chosen classifier to find disk-like structure even if embedded in bulge-dominated galaxies and, ideally, to parameterize the prominence of disks compared to an accompanying bulge component. The ZEST automated classifier satisfies these requirements and has also been trained to match visual classifications of local surveys. Our strategy is to adopt the ZEST classification in the COSMOS sample and test it both through visual inspection of COSMOS galaxies and an independent check of our results using visual morphologies determined in GOODS-N and published in Bundy et al. (2005).

3.4.1. The ZEST Automated Classifier

We compute mass functions in this work using the V_max technique (Schmidt 1968). We weight galaxies by the maximum volume in which they would be detected within the K_s-band and i⁺-band limits in a given redshift interval. In practice, our magnitude limits (K < 24 and i⁺ < 26.5) impose restrictions only in the highest redshift bins (z ≳ 1) and at masses below the limit where morphological classification becomes difficult. For each host galaxy i in the redshift interval j, the value of Vⁱ_max is given by the minimum redshift at which the galaxy would drop out of the sample

$$\begin{equation} V^i_{\rm max} = \int _{z_{\rm low}}^{z_{\rm high}} d \Omega \frac{dV}{dz} dz, \end{equation} \tag{ 3 }$$

where dΩ is the solid angle subtended by the survey area, and dV/dz is the comoving volume element. The redshift limits are given as

$$\begin{equation} z_{\rm high} = {\rm min}\left(z^j_{\rm max}, z^j_{K_{\rm lim}}, z^j_{i_{\rm lim}}\right) \end{equation} \tag{ 4 }$$

$$\begin{equation} z_{\rm low} = z^j_{\rm min}, \end{equation} \tag{ 5 }$$

where the redshift interval, j, is defined by [z^j_min, z^j_max], and $z^j_{K_{\rm lim}}$ and $z^j_{i_{\rm lim}}$ refer to the redshift at which the galaxy would still be detected below the K_s-band and i⁺-band limits. We use the best-fit SED template as determined by the stellar mass estimator to calculate these values, thereby accounting for the K-corrections necessary to compute V_max values (no evolutionary correction is applied).

Our main results are shown in Figures 2 and 3 which present the absolute mass function and fractional contribution of passive galaxies (which we refer to as the "red sequence") divided by ZEST morphological type into spheroidals, early disks, and late disks. The mass distributions of all red-sequence galaxies (red dashed line), all star-forming galaxies (blue dash-dotted line), and the full sample (black solid line) are also indicated. The statistical uncertainties are encoded by the width of the shaded regions. The sample (or cosmic) variance in COSMOS (see Section 2.2.1) is estimated at 0.2 dex and is likely responsible for the high apparent abundance values (especially for passive objects) in the 0.8 < z < 1.0 bin, a feature seen in previous work (Ilbert et al. 2010; Drory et al. 2009; Pannella et al. 2009; Pozzetti et al. 2009). Extreme disks with passive colors are so rare that we do not plot them and for clarity we omit the morphological mass functions of star-forming galaxies. Though we do not show it, it is important to note that at low masses star-forming disks of all three types far outnumber passive disks, while at the highest masses they are roughly equal in absolute numbers. The 50% completeness of the ZEST morphological sample is indicated with vertical dotted lines in Figure 2 while data points below this limit are simply omitted in Figure 3. These limits are nearly identical to the 50% completeness limits that would be obtained imposing a brightness limit of i_814W < 24.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figures 2 and 3 show that galaxies on the red sequence, even after removing those contaminated by dust, are remarkably diverse in their morphologies. This may have been missed in previous data sets that were limited to the highest masses where spheroidals do indeed dominate the red sequence. Thanks to the greater depth of the COSMOS photometry, the ZEST morphologies reveal a significant population of red-sequence galaxies at lower masses that appear to harbor disk components. These become prominent below a stellar mass scale that evolves downward with time from (1–2) ×10¹¹M_☉ at z ≈ 1.2 to ≈ 10¹⁰ M_☉ at z ≈ 0.3. Considering the full red sequence (defined by the mass range over which the fraction of passive galaxies is f_red>10%) we find that at every redshift, passive disks (early + late) are identified in as much as half of red-sequence galaxies and become dominant at the low-mass end. This increase in passive disks is robust to the maximum combined biases that could result from morphological classifications as well as cosmic variance. Dividing disks further into "early" and "late" depending on whether the average Sérsic 〈n〉 values are greater than or less than 1.25, we see that disk galaxies with prominent disk components (lower 〈n〉) are more common at lower masses (≲10¹⁰ M_☉) becoming comparable to the more bulge-dominated early disks with higher 〈n〉 values. On the other hand, these early disks are more frequent at higher masses and there is a redshift-dependent mass range (near ∼10¹¹M_☉) where they are roughly as abundant as passive spheroidals. This general behavior seems to mirror the broader pattern seen when the total population, regardless of rest-frame color, is split by morphological type (e.g., Bundy et al. 2005; Ilbert et al. 2009).

The evolution of morphologies on the red sequence is most easily seen in Figure 3 where the fraction of passive galaxies with a given morphology is plotted as a function of redshift in four M_* bins. While spheroidals dominate the highest masses of the red sequence at all times, at lower masses they start off as a small fraction of passive galaxies at early epochs and rise in importance with time. Meanwhile, the fraction of early and late passive disks tends to decline with time. As we discuss below, it seems plausible from Figure 3 that the passive disks' fraction declines because they transform into passive spheroidals, but Figure 2 shows that their absolute numbers evolve less strongly. It is therefore likely that some star-forming systems (including star-forming spheroidals) transform directly into passive spheroidals. This is supported by Figure 3, which shows that at every redshift, there is an ample supply of star-forming galaxies which may later transform into passive spheroidals.

Figures 2 and 3 reveal a surprisingly abundant population of passive galaxies that appear to contain disk components. As we argue in Section 5, these galaxies offer valuable clues to the mechanisms driving galaxy evolution. We have visually inspected early and late passive disks identified by ZEST in COSMOS and present random examples drawn from specific redshift and M_* ranges in Figures 4 and 5. Early disks tend to be bulge-dominated as expected, with many examples that might be classified as Sa/S0 galaxies. These galaxies often show rounded central regions surrounded by lower surface brightness material that we attribute to the disk component responsible for the relatively low Sérsic values (1.25 < 〈n〉 < 2.5). Note that our choice of ZEST classifications groups objects exhibiting even higher n values (〈n〉>2.5) in the spheroidals category. The late disks (Figure 5) tend to be more elongated with smaller bulges and comparably brighter disk components. In either category and across the full redshift range, there is little evidence for the presence of significant amounts of dust and only a small degree of spiral structure. Compared to the clumpier appearance of star-forming disks with spiral arms and associated star-forming regions, the morphology of passive disk galaxies is consistent with the notion that star formation has been largely shut down in these systems.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

While direct inspection supports the automated ZEST classifications, systematic biases could remain. In Section 3.4, we discuss how at faint magnitudes ZEST may misclassify up to 30% of spheroidals as disks. This potential bias is illustrated by the systematic error bar in Figure 2 which shows that passive disks would still dominate the low-mass end, even if the maximum 30% correction were applied. We have explored this bias further by comparing the ellipticity distributions of passive and star-forming disks. At the mass completeness limits of our analysis, the ellipticities of passive and star-forming early disks are similar, but late passive disks are weighted toward lower ellipticities. This occurs at a level that is consistent with the expected 30% contamination of the late disk sample by misclassified early disks and spheroidals. Thus, at the mass completeness limits and below, the passive late disk abundance should be viewed with caution given that as much as 30% could be misclassified.

We emphasize, however, that the basic trends in our results are robust because they are still present when we limit our sample to bright magnitudes (i_814W < 22.5) where the classifications are highly secure. In absolute terms, we see by comparing to Ilbert et al. (2010) that other automated methods classify some fraction of ZEST disks as spheroidals. This is expected given that broad divisions are somewhat arbitrary and that our goal is to identify evidence for disk components even in bulge-dominated galaxies that might otherwise be grouped with spheroidals. The galaxies with disk morphologies in our sample are associated with average Sérsic indices of 〈n〉 < 2.5 for the early disks and 〈n〉 < 1.25 for late disks. A value of n = 2.5 has often been used to divide disks from spheroidals (e.g., Barden et al. 2005). However, we should note that surface brightness dimming at high redshifts may make some disk galaxies appear more bulge-dominated, although this works in the opposite sense as the estimated bias at faint magnitudes from Scarlata et al. (2007b). So, until verification through bulge-to-disk decompositions is performed, a conservative interpretation would be to view the abundance of ZEST disks as an upper limit.

A more specific issue is a possible bias from inclination. We expect any type of disk component to be more easily identified in edge-on systems where flatter isophotes are easier to distinguish. Of particular concern is the possibility that dust contamination in edge-on systems could hide residual star formation given that the integrated colors could be dominated by a quiescent bulge. To check this, we have recomputed the mass functions of passive galaxies with the requirement that all galaxies have a ZEST ellipticity class of 0, corresponding to completely face-on systems. This reduces the sample size by ∼80%. While the mass functions shift a bit under this assumption, the essential behavior captured in Figure 3 is unchanged, with the same relative abundance and mass dependence of the various morphological types preserved. We can even compute mass functions requiring significant elongation (ellipticity class 3). While this disfavors spheroidals because they are intrinsically rounder and boosts late disks which are intrinsically flatter, the basic mass and redshift trends are still unchanged. We can therefore rule out an inclination bias in our ability to identify passive disks.

Another concern is the morphological K-correction which could mimic the rising abundance of passive disks at early times if, as expected, galaxies appear more disk-dominated at bluer rest-frame wavelengths. Over the redshift range of the sample (0.2 < z < 1.2) the ACS i_814W filter corresponds to a rest-frame wavelength range of 4000–6700 Å. It has been shown that morphological K-corrections within this range are small for most galaxies (e.g., Lotz et al. 2004; Cassata et al. 2005), but with only one band of ACS data, it is hard to test this effect directly. Capak et al. (2007a) use a subsample of COSMOS galaxies to show that automated estimators are generally robust to K-corrections, but to address this problem further and check our ZEST-based results more generally, we use the visually classified morphologies in the northern GOODS-N (Giavalisco et al. 2004) as presented in Bundy et al. (2005). These observations have a comparable depth as the COSMOS i_814W imaging, but include four HST/ACS bands (BViz), enabling morphological classifications robust to potential K-corrections.

We identify quiescent stellar populations using similar color–color cuts for the GOODS data as for COSMOS and do indeed find a population of passive disks with visual morphological types of Sa–Scd. Figure 6 shows several examples. While the small size of the GOODS comparison sample prevents a detailed comparison, it nonetheless shows similar evidence for an increasing fraction of passive disks as a function of redshift. For 0.4 < z < 0.7 and above the completeness limit of log M_*/M_☉ ≈ 10.2 we find ≲20% of the quiescent population is identified as disk-like. At 1.0 < z < 1.1 and log M_*/M_☉ ≳ 10.7, the fraction rises to roughly 35%. While the statistical uncertainty is large, the sense of this trend reinforces the results of the larger COSMOS sample. It should be noted that biases among visual morphologies favor early-type classifications at faint magnitudes and high redshifts due to surface brightness dimming (e.g., Brinchmann et al. 1998). We also note that the GOODS objects show no apparent bias with inclination, in agreement with the inclination tests above. The use of color information in refining the GOODS morphologies indicates that the evolutionary signal is not likely to result from K-corrections and the fact that passive disks identified visually exhibit similar abundances and evolution lends support to the ZEST classifications.

Zoom In Zoom Out Reset image size

How significant are passive disks in the evolutionary history of galaxies? Ideally, we could answer this question by tracking the mass-dependent flow of galaxies transitioning through different evolutionary phases, but unfortunately, as discussed in Section 2.2.1, the strong cosmic variance uncertainties in the single COSMOS field prevent a detailed treatment. Broadly speaking, Figures 2 and 3 in conjunction with previous work suggest that passive disks may represent a significant and relatively common phase of evolution. While the red-sequence abundance grows with time (e.g., Bundy et al. 2006; Borch et al. 2006), passive disks always dominate at lower masses which is also where the newest arrivals to the red sequence can be found (e.g., Treu et al. 2006). This suggests that a large fraction of these quenching galaxies experience an initial passive disk phase. This is supported by Figure 2 which shows that, as with red-sequence spheroidals, the absolute number density of passive disks below ∼10¹¹M_☉ increases with time, even though, compared to all passive galaxies in this mass range, their fraction decreases with time. Choosing a mass between 10¹⁰ and 10¹¹M_☉ and comparing the highest and lowest redshifts sampled, we see that ∼20%–40% of new red-sequence galaxies formed over this interval can be accounted for by the increase in the number of passive disks. At masses above ∼10¹¹M_☉, evolution in the absolute abundance of passive disks is more difficult to discern but, as with their fraction, likely declines by 20%–50% (compare the two lower redshift bins in Figure 2 and note the lack of high-mass examples at late times). This reinforces the visual impression of Figure 3, namely, that passive disks evolve into spheroidals. Indeed, as we discuss below, one expects morphological evolution of passive disks to be enhanced compared to their star-forming counterparts because lower gas fractions (f_gas) enable more minor and, hence, more frequent mergers to drive transformations (Hopkins et al. 2009a). If we make the assumption that all passive disks transform into spheroidals within 1–2 Gyr (e.g., van Dokkum et al. 2005), the fraction of new arrivals to the red-sequence transitioning through a disk phase must increase to compensate, and could be as high as 60%.¹⁷ Passive disk galaxies may therefore represent a fairly common phenomenon and highlight the importance of further morphological evolution along the red sequence.

At all times, there are as many if not more star-forming disks than passive disks, so it is easy to imagine that passive disks may be drawn from the star-forming population with one important caveat: there are almost no passive galaxies with "extreme disk" (Sérsic 〈n〉 < 0.75) morphologies. Such extreme disks are common in large numbers among star-forming galaxies at all masses and even dominate at the lowest masses where passive disks are most abundant. They would be easily detected if they existed in the passive population. Therefore, the transformation to the red sequence either leads to the formation of a bulge component (e.g., Domínguez-Palmero & Balcells 2009) or requires the existence of one. In the latter case, star-forming and passive disks with similar bulge components should differ only in color but not in their intrinsic structure. We will postpone a detailed comparison of bulge/disk decompositions, scale lengths, and M/L evolution to future work and focus here on a simple comparison of the concentration (C = R_80%/R_20%) of passive and star-forming early disks (Figure 7). We consider the nearest redshift bin 0.2 < z < 0.4 so that the ACS i_814W band probes the reddest possible wavelengths (least affected by star formation), and consider the mass range of 9.7 < log M_*/M_☉ < 10.2. Passive disks appear to be moderately more concentrated with a peak at C ≈ 3.9 compared to blue disks with a peak of C ≈ 3.6 (a Kolmogorov–Smirnov probability test shows that the probability of 0.02 for the distributions being drawn from the same sample). Even after dividing the star-forming disks into low and high SFR subsamples (based on their NUV luminosities) and comparing just the low SFR star-formers to passive disks, this difference in concentration remains. It also holds to z ≈ 0.9 despite rest-frame morphologies that sample bluer wavelengths. Still, this signature could largely be driven by differences in M/L gradients between star-forming and passive systems. Given this tentative result and the stronger requirement that passive disks be more bulge-dominated than star-forming disks, it seems unlikely that passive disks form simply via the quenching and fading of star formation disk galaxies.

Zoom In Zoom Out Reset image size

Could passive disks be a product of dense environments? Their morphologies are sometimes reminiscent of S0 galaxies (see Figure 6) that are thought to become increasingly common in clusters (e.g., Dressler et al. 1997; Smith et al. 2005) although recent work disputes this (Holden et al. 2009). While we have specifically attempted to exclude S0s from our disk sample (they are classified as spheroidals), distinguishing them at distant epochs remains difficult (e.g., Moran et al. 2007b). A more promising link exists with "passive spirals" (defined via far UV and spectra line diagnostics) which likely transform into S0s in cluster environments (Moran et al. 2005, 2006, 2007a). This transformation process is thought to begin slowly (1–2 Gyr timescales) in infalling groups, perhaps as a result of "gentle" interactions with other galaxies and the intergroup medium. If the intracluster medium (ICM) is sufficiently dense, ram pressure stripping strongly accelerates the quenching of star formation, producing S0s and ellipticals closer to the cluster core and driving evolution in the morphology–density relation (e.g., Capak et al. 2007a).

Because there are only 1–2 clusters in the 1.6 deg² COSMOS field (Meneux et al. 2009), cluster mechanisms cannot fully account for the passive disk population studied here. Still, group scale environments may play an important role (see Wilman et al. 2009). To investigate further, we define a "field" sample by removing sources along the line of sight to dense structures. These were identified in the X-ray-selected COSMOS group catalog assembled by Finoguenov et al. (2007) and Leauthaud et al. (2010) which contains 206 groups with weak-lensing calibrated virial (total) masses complete to M₂₀₀ ≈ 3 × 10¹³M_☉ to z = 0.5 and M₂₀₀ ≈ 3 × 10¹⁴ M_☉ to z = 1.0. Any galaxy within the projected R₂₀₀ virial radius of any group, regardless of its redshift, is removed from the field sample, reducing the full sample by ≈ 30%. Remarkably, the mass dependence and evolution of passive galaxy morphologies (Figure 3) in this field sample is virtually identical to the full sample. This is not to say that there are no environmental trends in the passive disk fraction which would be expected from previous studies of morphology–density relation. Our "field" test simply shows that a large fraction of passive disks live and evolve in halos less massive than 10¹³M_☉ (given our two lowest z-bins), ruling out environmental processes in more massive halos as the sole explanation.

Even at M₂₀₀ ≈ 10¹³M_☉ environmental effects can be important as very weak ram pressure (2–3 orders of magnitude weaker than in cluster cores) can cause "strangulation" or "starvation" via stripping of the galaxy's hot gas halo, thereby shutting off a potential fuel source for subsequent star formation (see McCarthy et al. 2008; Bekki 2009). The starvation scenario would help explain the prevalence of passive disks at low masses, since simulations show that lower mass galaxies are more easily affected. But the starved remnants should have structural properties similar to star-forming disks, a result that seems to be at odds with Figure 7. In addition, McCarthy et al. (2008) show that less concentrated disks are more susceptible to starvation. Yet, the very low concentration "extreme disk" morphologies found in the star-forming population are almost completely absent among passive galaxies. Furthermore, because as M_* increases starvation becomes less effective, we would expect the least concentrated passive disks (the "late types") to occur at the highest masses. Our observations show the opposite trend. We also might expect the group abundance and typical gas density, n_ICM, to grow with time from z = 1.5 (e.g., McGee et al. 2009). Yet, the most massive examples of passive disks, for which n_ICM must be large to induce starvation, occur in our highest redshift bins (z ∼ 1).

We therefore conclude that while it is possible that passive disks are a product of low-mass group environments, proposed environmental processes at these scales appear to have some difficulty explaining the observed mass dependence, evolution, and concentration of these galaxies.

Beyond environmental processes, star formation quenching has often been associated with feedback mechanisms triggered by major mergers (e.g., Hopkins et al. 2008). The existence of passive disks presents a problem for this hypothesis: how can a major merger shut down star formation in these galaxies without destroying the underlying stellar disk? One possible solution is that before quenching completes, an entirely new disk is formed (Hammer et al. 2005). Recent high-resolution merger simulations have illustrated exactly this effect (Springel & Hernquist 2005; Robertson et al. 2006; Governato et al. 2009). Roughly speaking, new gaseous disks can form in gas-rich mergers with f_gas>0.5 because a lack of central instabilities prevents gas transport to the center and suppresses bulge growth. Given observations of f_gas as a function of M_*, mergers with M_* ≈ 10¹⁰M_☉ are much more likely to result in a remnant with a disk-like morphology than similar mergers at M_* ≈ 10¹¹M_☉ (see Hopkins et al. 2009a).

One of the key aspects of the merger/regrowth model is that the strength of the remnant disk component depends on f_gas (Stewart et al. 2009b; Hopkins et al. 2009b). To investigate how predictions from this model compare to our observations, we used the relationships between f_gas, merger mass ratio (μ), M_*, and remnant bulge-to-total ratio (B/T) that Hopkins et al. (2009a) derive based on a large suite of merger simulations. We assume a scatter of 0.25 dex in the predicted B/T and consider mergers with 0.3 < μ < 1.0. We use f_gas(M_*) evaluated at z = 2 to correspond to passive disk progenitors at z ∼ 1, and assume that the z = 0 relation applies adequately to progenitors at z ∼ 0.3. Remnant disks are those with B/T < 0.6. The results of this exercise are plotted in Figure 8 as two shaded regions for the two redshifts evaluated. These regions represent the predicted disk fractions spanned by mergers of different mass ratios. The region corresponding to the z = 0 f_gas(M_*) is offset to lower M_*, reflecting the decline of f_gas at higher masses. Overplotted are the observed fractions of passive disks (early + late disks) at the two redshifts considered. While in detail, the scatter in predicted B/T likely depends on mass ratio and f_gas, this toy remnant model does a good job matching the observed passive disk fractions as well as their mass and redshift dependence if most formed in mergers with μ ∼ 1.0. A more detailed comparison between predicted and observed morphologies is needed to test this further.

Zoom In Zoom Out Reset image size

This merger/regrowth scenario is consistent with the lack of a strong environmental dependence (Section 5.3) and would predict that some passive disks will be isolated. This could be tested through clustering and pair count studies. Furthermore, the somewhat higher concentration in passive disks compared to star-formers is naturally explained by the extra bulge mass added during the merger. The big drawback remains understanding how mergers lead to the quenching of star formation. If quenching is short-lived and violent, as in models employing a quasar phase (e.g., Menci et al. 2006; Hopkins et al. 2008), it must somehow be delayed until enough time passes for a stellar disk to regrow (at least 1 Gyr). At the same time, recent simulations have cast doubt on the ability of quasar-mode AGN feedback to directly quench star formation (e.g., DeBuhr et al. 2010).

Instead AGN may be responsible for subtle feedback cycles that prevent gas cooling on longer timescales (1–2 Gyr) after the cold gas is depleted (e.g., Croton et al. 2006). If this feedback is tied to the central black hole mass, this could explain why passive disks have large bulges (Section 5.2), but it does not make clear why other star-forming disk galaxies with similar bulges have not been quenched (Figure 7). Finally, if quenching is initiated by low luminosity AGN feedback, are mergers even required (e.g., Ciotti & Ostriker 2007)? While the specific quenching process in these scenarios remains unknown, the agreement in Figure 8 nevertheless supports a formation mechanism in which the frequency and appearance of passive disks is strongly linked to their progenitor gas fractions.

If star formation in passive disks is shut down because the gas supply is depleted, simulations show that such gas-poor disks can be more easily disrupted by mergers with lower mass ratios—that is, more minor interactions—than disks with high f_gas, owing to the ability of gas to regrow disks and suppress bulge formation (Hopkins et al. 2009a). This suggests a two-step process where first, late-type star-forming disks are quenched and transformed into bulge-dominated, passive disks. This makes them vulnerable to minor interactions that, second, disrupt the disk component and convert them into spheroidals.¹⁸

Figure 2 suggests that this two-step evolution may have a broad impact on interpretations of how morphological populations evolve. The figure shows that passive disks account for as much as one-third of all disk galaxies at M_* ∼ 10¹¹M_☉ while the number declines toward lower masses (and redshifts). Assuming they are gas-poor and susceptible to more minor interactions, then because minor mergers are far more common than major mergers (e.g., Stewart et al. 2009a), this passive one-third of the disk population is likely to account for many of the newly formed spheroidals since z ∼ 1 (e.g., Bundy et al. 2005; Ilbert et al. 2010). But recent efforts to compare the formation rate of spheroidals to the merger rate have focused only on rarer, major mergers, assuming perhaps incorrectly that single mergers of gas-rich disks are the only channel for spheroidal formation (e.g., Bundy et al. 2007; Genel et al. 2008; Bundy et al. 2009). These studies have suggested that the major merger rate, both as observed and predicted, may be 2–3 times too low to explain the formation of new spheroidals. While a more precise comparison requires consistency in defining mass ratios, morphologies, and understanding gas effects (Hopkins et al. 2010), including more prevalent minor mergers involving passive disks could help bring current measures of the two phenomena into better agreement.

The environmental effects and merger/regrowth model we have considered depend on phenomena that are external to the galaxy. It is also possible that passive disks form as a result of internal processes. An example of this is "morphological quenching" (MQ; Martig et al. 2009). In this model, after the cold streams that fuel high-z star formation shut down or become clumpy (Dekel et al. 2009), the further fragmentation of gaseous disks—and therefore its ability to form stars—can be internally suppressed by two processes. The first is the declining self-gravity of the gas as it is converted to stars and the second is the deepening central potential caused by stellar bulge growth resulting from instabilities in the stellar disk or even minor mergers. The predicted remnants of morphological quenching are consistent with many of our observations. They would appear as passive disks with large bulges, would have a weak environmental dependence, and exhibit slow quenching, in this case initiated by gradual changes in internal structure.

While more detailed comparisons between observations and simulations of morphological quenching (Martig et al. 2009) are clearly needed, we note that many of the examples in Figures 4–6 show large, thick disks. In future work, this may help distinguish between an early formation with thickening from disk instabilities that could accelerate MQ (Bournaud et al. 2009) and later disk regrowth (Hopkins et al. 2009a). It is also important to determine if MQ can explain why bulge growth leads to quenching in some galaxies but not in others as well as why passive disks are more disk-dominated at lower M_* (Figure 2). The higher gas fractions at these masses would naively require larger, not smaller bulges. Finally, we note that morphological quenching requires testable deviations from the Kennicutt–Schmidt relation (e.g., Donovan et al. 2009) and implies that gas disks in passive galaxies remain intact. Subsequent mergers of passive disks in this scenario could therefore produce new star-forming galaxies instead of passive spheroidals.

Finally, while this work has focused on mass-limited samples at high redshift, a detailed study by Masters et al. (2010) of a subset of passive disks identified using GZ in SDSS provides additional insight. A direct comparison is not possible because the GZ passive disks were selected using a single color (g − r) and were required to be mainly face-on and to have spiral arms as identified by GZ users. Still, the two samples paint a consistent picture. Red GZ disks are more common at higher masses (see also van der Wel et al. 2009) and tend to host larger bulges than their star-forming counterparts. Their emission lines indicate little ongoing SF but are inconsistent with post-starburst spectra, favoring instead a gradual decline of SF over 1–2 Gyr. They tend to live in intermediate environments consistent with the ≲10¹³M_☉ halos our observations suggest. More than 75% host nuclear bars and 50%–80% show evidence for LINER emission. Assuming the GZ sample is representative of passive disks in general, the Masters et al. (2010) analysis supports a formation mechanism that is strongly tied to a bulge and a central bar, somewhat weakly tied to environment, and capable of quenching star formation slowly.