The colors and sizes of recently quenched galaxies: a result of compact starburst before quenching

We are interested in galaxies whose SFR is low but that was much higher in the recent past. Accordingly, our sample must satisfy two criteria: they must have had a rapid drop in star-formation, as evidenced by enhanced Balmer absorption, and they must currently quiescent.

We use the H$\delta$ absorption as the probe for the recent star-formation activity. We require a fiducial cut of $\mbox{EW}(\mbox{H}\delta)\geq 4$Å after subtracting the contribution from the emission line derived from a joint fitting of the stellar continuum and emission lines using the Penalized Pixel-Fitting (pPXF) method (Cappellari & Emsellem, 2004; Cappellari, 2017). We select galaxies with S/N higher than 10 per pixel at rest-frame 4000Å ($\texttt{SN\_RF\_4000}\geq 10$) and quality flags $\texttt{f\_use}=1$ in the LEGA-C DR2 catalog (Straatman et al., 2018) to minimize mis-classification. The typical uncertainties in EW(H$\delta$) is $\sim 0.4$Å.

There are different methods to select quiescent galaxies. The lack of emission lines associated with star-formation activity are commonly used as a criterion, for example, Balmer lines. However, H$\alpha$ redshifts to the near IR at $z\gtrsim 0.5$ and only less than half of the LEGA-C spectra cover both H$\delta$ and H$\beta$ lines. An alternative is the [O II]$\lambda 3727,3729$ emission. Nevertheless, recently quenched galaxies commonly have detectable [O II]$\lambda 3727,3729$ emission from ionizing sources other than star-forming regions (Yan et al., 2006). Requiring the lack of [O II] emission leads to an incomplete sample and potential biases against AGN hosts (Lemaux et al., 2010; Wu et al., 2014; Lemaux et al., 2017).

Instead, we use the rest-frame $U-V$ and $V-J$ broadband colors to define quiescence. We derive the color by fitting the observed multiwavelength spectral energy distributions (SEDs) from the UltraVISTA catalog (Muzzin et al., 2013a) using the FAST code (Kriek et al., 2009). We adopt the demarcation between star-forming and quiescent galaxies proposed by Muzzin et al. (2013b). Recently, selections based on only broadband photometry have been extensively applied (e.g. Kriek et al., 2010; Whitaker et al., 2012). We will discuss the different methods in Section 5.

We derive $R_{e}$ of galaxies, the semi-major radius enclosing 50% of light, using the HST ACS F814W images from the COSMOS program (GO-9822, GO-10092, PI: N. Scoville Scoville et al., 2007), following the procedure of van der Wel et al. (2012) using the galfit program (Peng et al., 2010). Although galaxies do not necessarily have simple Sérsic profiles (e.g., Margalef-Bentabol et al., 2016), tests on simulated galaxy images suggest that as long as the S/N ratios of the images are reasonably high ($S/N\gtrsim 100$), $R_{e}$ can generally be recovered with a $\sim$20% accuracy even for galaxies with double Sérsic profiles (Davari et al., 2014, 2016). While recently quenched galaxies may intrinsically consist of multiple components (Section 4), we choose to adopt the $R_{e}$ from single Sérsic fits for all galaxies for a consistency in methodology. The HST images of all our recently quenched galaxies have $S/N\gtrsim 100$, therefore, we expect the $R_{e}$ is reliable.

We exclude 2 galaxies with catastrophic failures in the galfit fitting or highly disturbed morphologies. One of them exhibits three bright clumps and the other has a diffuse asymmetric low surface brightness feature extending towards a nearby faint object. However, their disturbed morphologies carry important clues for the formation of recently quenched galaxies as will be discussed in Section 5.2. We also exclude 3 galaxies with close companions that potentially affect broadband photometry thus the estimate of stellar masses and rest-frame colors ($\texttt{f\_int}=1$). The final sample consists of 32 recently quenched galaxies (Table 1). Fig. 1 shows their redshifts, stellar masses, $U-V$ and $V-J$ colors, and EW(H$\delta$). Fig. 2 shows the HST images and spectra of 3 recently quenched galaxies with different colors as examples. Recently quenched galaxies in our sample reside mainly in the average density environment with local overdensity $\log(1+\delta)=0.1\pm 0.4$ (zero is the average) according to the estimate of Darvish et al. (2015). Our sample thus does not probe extreme environments like galaxy clusters.

We examine the H$\beta$ emission line strengths of recently quenched galaxies to confirm their quiescence. The spectra of 12 galaxies in the sample cover the H$\beta$ line. Eight of them have $\mbox{EW}(\mbox{H}\beta)_{em}>-1$Å, consistent with being quiescent (Wu et al., 2018a). For the remaining 4 galaxies, two galaxies have $\mbox{EW}(\mbox{H}\beta)_{em}$ consistent with $-1$Å (ID: 211263, $-1.20\pm 0.23$Å and ID: 230053, $-1.04\pm 0.36$Å), one galaxy (ID: 232099, $-3.92\pm 0.45$Å) has strong [Ne III] and [O II] emission, indicating an ionizing source other than star formation. One galaxy (ID: 184250, $-2.24\pm 0.58$Å) would be considered as weakly star-forming. Its H$\beta$ emission strength is $\sim 0.3$ dex lower than the median value of star-forming galaxies in the LEGA-C DR2 sample ($-4.70$Å). We conclude that our sample has low SFRs.

We cross-check the SFH derived from full-spectral fitting technique used by Chauke et al. (2018) and find that our selected galaxies all have an elevated period of star-formation activity in the past $\lesssim 1$ Gyr. The strong H$\delta$ absorption feature ensures that these galaxies are truly ‘recently quenched’.

Galaxies have a wide range of mass growth and chemical evolutionary histories. Recovering the true formation histories from observables remains a challenging task. In this paper, we use simple star-formation histories combined with stellar population synthesis models to illustrate how recent star-formation events affect observables. We adopt the following analytical form for our fiducial analysis:

This star-formation history consists of two components: a constant star-formation phase followed by a subsequent burst with an exponential decay timescale $\tau$ with different strength of bursts (Fig. 3a).

In this paper, we define the burst strength $b$ as the ratio between the masses formed in the burst to the mass formed in the constant star-formation period, $b\equiv M_{burst}/M_{const}$. We generate 4 sets of models of $b$ = 3, 1, 0.3, and 0.1, with 2 different $\tau$ = 0.1 Gyr and 0.5 Gyr, respectively. The combination of $b$ and $\tau$ determines the amount of stars formed right before galaxies become quiescent, which is the key feature for recently quenched galaxies. A shorter $\tau$ and larger $b$ represent a more drastic decline of star-formation rate. A sudden truncation of star-formation corresponds to $\tau=0$. We will see later in Section 3 that 0.5 Gyr is roughly the longest $\tau$ that can produce the observation signatures of recently quenched galaxies with our assumed star-formation history.

The $b=0.1$ model only increases the SFR by $\sim 2$ times, averaging over the first 100 Myr after the burst. The $b\geq 1$ models would be considered as strong starburst events. More extreme bursts with $b>3$ behave qualitatively similarly to the $b=3$ model for the parameters that we are interested in. We make one more model with $b=0$, mimicking a sudden truncation of star-formation without forming new stars. Fig. 3a shows the star-formation histories of the 9 models used in this paper. Real recently quenched galaxies do not have to follow Equation 1. The parameterization, $b$ and $\tau$, provides insight into how the strength and the timescale of the last period of star-formation affect the color evolution.

We then use the Python implementation of the Flexible Stellar Population Synthesis package (FSPS; Conroy et al., 2009; Conroy & Gunn, 2010; Foreman-Mackey et al., 2014) to generate the spectra and broadband colors of each model for each 100 Myr time step, adopting the MILES spectral library (Sánchez-Blázquez et al., 2006), Padova isochrones (Girardi et al., 2000; Marigo & Girardi, 2007; Marigo et al., 2008), and a Chabrier IMF (Chabrier, 2003).

We choose solar metallicity as the fiducial value (Gallazzi et al., 2014). For the dust attenuation, we use the Cardelli et al. (1989) dust reddening curve and a 2-component dust model. Young stellar populations ($<10^{7}$ yr) have $\mbox{A}_{\mbox{v},young}=1.5$ and the rest have $\mbox{A}_{\mbox{v},old}=0.5$. Therefore, when the SFR is low, like the recently quenched galaxies, the total $\mbox{A}_{\mbox{v}}$ of the models approach $\mbox{A}_{\mbox{v}}\simeq 0.5$, which is motivated by the constraints from broadband SED fitting, as well as spectral-photometric fitting of galaxies at similar and slightly higher redshifts (Carnall et al., 2019; Belli et al., 2019). Different $\mbox{A}_{\mbox{v},young}$ makes negligible difference. We use these fiducial values in Section 3 and Section 4 and discuss the assumptions in Section 5.

Fig. 3 shows the evolution of EW(H$\delta$), $V-J$, and $U-V$ colors of the 9 different star-formation histories with our fiducial choice of metallicity and dust attenuation. The thick segments indicate the period that model galaxies would be classified as recently quenched galaxies based on our selection criteria.

Galaxies with shorter $\tau$ can be identified as recently quenched galaxies for longer periods of time. We will compare these observables to our recently quenched galaxy sample in Section 3 and discuss scenarios that deviate from the fiducial parameters in Section 4 and Section 5.

In general, the recently quenched galaxies have bluer $V-J$ and $U-V$ colors than the rest of quiescent galaxies, reflecting their younger stellar population (Fig. 1a). On the other hand, they spread over a wide range in the color-color space between $0.7\lesssim V-J\lesssim 1.3$ and $1.4\lesssim U-V\lesssim 1.9$.

This sample of recently quenched galaxies cannot be separated based on rest-frame $U-V$ and $V-J$ colors alone. The distributions of colors are much wider than prevalent photometric selections, which usually select UVJ quiescent galaxies with $V-J\lesssim 0.9$ (e.g. Whitaker et al., 2012; Belli et al., 2019); only $\sim$30% of the sample fulfill the color selection. We will discuss the different color distributions in Section 5.

Fig. 4a shows the recently quenched galaxies and the evolutionary tracks of different star-formation histories on the UVJ diagram. The model galaxies evolve from the bottom-left of the diagram, become redder and move to the top-right. The thick segments indicate the period when the galaxy is classified as a recently quenched galaxy.

On the UVJ diagram, the models tracks cover the observed distribution of recently quenched galaxies reasonably well. The wide range of color can be partially due to different decay time scales $\tau$. For the same burst fraction $b$, recently quenched galaxies with $\tau=0.1$ Gyr (solid lines) are bluer than galaxies with $\tau=0.5$ Gyr (dashed lines) in the $V-J$ color.

In addition, the $V-J$ color also depends on the burst fraction $b$. Recently quenched galaxies with weaker bursts are on average redder (see also Fig. 3). Even with a short decaying timescale of $\tau=0.1$ Gyr, a recently quenched galaxy with $b=0.3$ is hardly ever bluer than $V-J\simeq 0.9$. Recently quenched galaxies with even bluer colors are predominantly those experienced strong bursts, which have at least doubled their stellar masses in the starburst events.

Fig. 4b shows the EW(H$\delta$) and $V-J$ colors of recently quenched galaxies. In our sample, EW(H$\delta$) ranges from 4Å to $\simeq$8Å. However, the $\tau=0.5$ Gyr models only produce recently quenched galaxies with $\mbox{EW(H}\delta\mbox{)}\lesssim 5$Å. About half of the sample must have shorter quenching timescales in order to produce the strong H$\delta$ absorption.

The model tracks evolve from the top-left to the bottom-right in Fig. 4b and are in broad agreement with the majority of the sample, suggesting that our fiducial parameter set is reasonable. However, at the reddest $V-J$ colors, several recently quenched galaxies have high EW(H$\delta$) that are not well explained by the fiducial models. These galaxies possibly have much larger $A_{v}$ than the fiducial value. We will discuss our assumptions in Section 5.

Wu et al. (2018a) have observed that the $R_{e}$ of recently quenched galaxies at $z\sim 0.8$ are on average smaller than the quiescent population and attributed it to central, compact starbursts happened in the centers of their progenitor galaxies before quenched. This hypothesis leads to a prediction that recently quenched galaxies with stronger bursts should have on average smaller $R_{e}$.

As we see in Fig. 3 and Fig. 4, the burst strengths correlate with $V-J$ colors. Recently quenched galaxies with blue $V-J$ colors can only be the products of strong starburst events that roughly doubled the stellar masses of galaxies. On the other hand, recently quenched galaxies with red $V-J$ colors can be a combination of galaxies with weak bursts, longer decaying timescale, and are observed later after the starburst. Therefore, we should expect that recently quenched galaxies with blue $V-J$ colors have on average smaller $R_{e}$.

Fig. 5a shows the $R_{e}$ and the stellar masses of recently quenched galaxies, color-coded by their $V-J$ colors. The $R_{e}$ range from 0.7 kpc to 4.9 kpc, with 16th, 50th, 84th percentiles of 1.2, 2.2, and 4.0 kpc, respectively. We see in general that recently quenched galaxies with blue $V-J$ colors tend to have smaller $R_{e}$. Next, in order to take into account the intrinsic correlation between the sizes and the stellar masses, we compare the $R_{e}$ of recently quenched galaxies with only galaxies of similar masses.

We first fit a linear relation between the median sizes in each 0.1 dex mass bin and the mass in log-log space as:

where (a,b) are ($0.48^{+0.05}_{-0.06}$, $0.38^{+0.01}_{-0.01}$) and ($0.29^{+0.06}_{-0.07}$, $0.71^{+0.01}_{-0.01}$) for quiescent and star-forming galaxies, respectively (dashed lines in Fig. 5a). The 1-$\sigma$ dispersion around the median is $\sim 0.2$ dex for both quiescent and star-forming populations in the stellar mass range of interest (also see Wu et al., 2018a). We then calculate $\Delta\log R_{e}$, the logarithmic size difference between recently quenched galaxies and the best-fit mass-size relations of quiescent and star-forming galaxies, respectively. We find the $V-J$ colors correlate strongly with $\Delta\log R_{e}$ when comparing either to quiescent or star-forming galaxies (Fig. 5b,c). Blue recently quenched galaxies are relatively smaller than red recently quenched galaxies. The null hypothesis that the correlation is due to random uncertainty can be rejected with high confidence. The clear correlation between colors and $R_{e}$ indicates that the star-formation histories and the structures of recently quenched galaxies are coupled.

The term post-starburst galaxy implies that these quiescent galaxies with strong Balmer absorption had experienced periods of elevated star-formation rates before becoming quiescent. However, such a star-formation history is not required. A sudden quenching without a starburst can also produce quiescent galaxies with strong Balmer absorption (Le Borgne et al., 2006).

The sizes and colors together suggest that a pure truncation scenario may contribute to the recently quenched population but cannot account for the full sample. Most recently quenched galaxies are significantly smaller than star-forming galaxies (Fig. 5). If the quenching mechanisms only shut down star-formation without producing new stars in the center, the $R_{e}$ would be similar to the progenitor star-forming galaxies. In addition, without a newly-formed stellar component, the colors of recently quenched galaxies would be too red for at least half of the sample (the gray area in Fig 7b). We thus conclude that most recently quenched galaxies had at least a low level starburst event before turning into quiescence.

Spatially-resolved information will further verify this scenario. The central compact starburst makes the stellar ages younger in the center. Recently quenched galaxies may appear to have blue centers in multi-wavelength HST images if the dust attenuation is not severer in the center. While spectroscopy can provide better estimate of stellar ages, large ground-based telescopes with AO or JWST will be needed for good S/N and spatial resolution.

Traditionally, the Balmer absorption lines are used to select recently quenched galaxies, or post-starburst galaxies as often called in the literature, because they are one of the key features of A-type stars and thus are sensitive to recent star-formation activities of galaxies. However, beyond the nearby universe, high S/N spectra of galaxies are observationally expansive thus other techniques have been explored.

In recent years, deep multi-wavelength photometry has become a popular alternative. Regardless of details, these photometric selections either explicitly or implicitly select the bluest quiescent galaxies mainly with rest-frame $V-J\lesssim 0.9$ (Kriek et al., 2011; Whitaker et al., 2012; Wild et al., 2014; Yano et al., 2016; Belli et al., 2019). However, we have demonstrated in the paper that spectroscopically-identified recently quenched galaxies span a wide range of color: $0.7\lesssim V-J\lesssim 1.3$ and only $\sim 30\%$ (10/32) of the sample fulfill the prevalent color cut.

While the majority of recently quenched galaxies in this paper have $V-J>0.9$, they present only a small fraction of quiescent galaxies at given $U-V$ and $V-J$ colors (Fig. 1a). The photometric-only selections may not be able to effectively identify these redder recently quenched galaxies without introducing severe contamination to the sample. Based on our LEGA-C DR2 mother sample (Fig. 1a), a more generous color cut of $V-J<1.0$ selects 35 quiescent galaxies and 18 of them have $\mbox{EW}(\mbox{H}\delta)<4$Å. If we adopt an even more generous color cut of $V-J<1.3$ and $U-V<1.9$ to include all but one recently quenched galaxies, the contamination reaches $\sim 70\%$: 112 out of 143 quiescent galaxies that fulfill the color cut have $\mbox{EW}(\mbox{H}\delta)<4$Å.

On the other hand, by selecting extremely blue quiescent galaxies, the photometric selections can obtain clean samples. The blue colors ensure that most of the stars must have formed recently. Therefore, the star-formation rates have to be high in the recent past. These blue quiescent galaxies also have strong Balmer absorption lines. Among 13 UVJ quiescent galaxies with $V-J\leq 0.9$ in Fig. 1a, all of them have $\mbox{EW}(\mbox{H}\delta)\geq 3$Å and 10 of them have $\mbox{EW}(\mbox{H}\delta)\geq 4$Å (also see Maltby et al., 2018). Nevertheless, this selection is very incomplete.

Under the framework of Section 4, this incompleteness also introduces a selection bias. The starbursts need to produce nearly comparable stellar masses to the pre-existing masses to make the whole galaxies blue enough to be distinguishable from other quiescent galaxies in the color space. Recently quenched galaxies with $V-J<0.9$ are predominantly galaxies with strong starbursts ($b\geq 1$). Current photometric selections serve as effective alternatives when good spectra are not available, but they may only pick up those had at least doubled their stellar masses in the starburst events. Recently quenched galaxies with weaker bursts may be missing from the sample, even though they can be produced by the same mechanism. This potential bias should be kept in mind when interpreting the results.