VARIABILITY AND MULTIWAVELENGTH-DETECTED ACTIVE GALACTIC NUCLEI IN THE GOODS FIELDS

Active galactic nuclei (AGNs) are galaxies accreting significant amounts of material onto their central supermassive black holes (SMBHs). These galaxies were once thought to be oddities among the total galaxy population, but are now seen as important mile markers on the broad road of galaxy evolution. Once considered rare, SMBHs are now believed to exist in the centers of all galaxies containing a significant bulge component (Kormendy & Richstone 1995). In addition, the observed relation between the mass of the SMBH and the bulge stellar mass (Ferrarese & Merritt 2000; Gebhardt et al. 2000) suggests a coupled assembly history between the two. To understand the role played by AGNs in the evolution of galaxies, it is thus necessary to identify complete samples of AGNs in galaxy surveys.

AGNs have been identified using many techniques such as UV/optical color selection (Markarian 1967; Schmidt & Green 1983; Richards et al. 2002), spectroscopic signatures of broad emission lines, or narrow emission lines with distinct line flux ratios (Baldwin et al. 1981; Veilleux & Osterbrock 1987), X-ray emission (Fabbiano et al. 1992; Alexander et al. 2003), infrared color selection (Lacy et al. 2004; Stern et al. 2005), radio emission (e.g., Smith & Wright 1980), and variability (e.g., Hook et al. 1994; MacLeod et al. 2010) over a range of timescales. Many of these techniques, however, produce samples biased against galaxies where the AGN light is much less than that of the host galaxy. Additionally, most surveys in the optical/UV are biased against obscured, dusty AGN/host galaxies. In order to identify largely complete samples of AGNs, it is clear that a variety of techniques and selection methods should be employed. While it may be found that AGNs selected using different techniques represent separate populations experiencing different phases in AGN/galaxy evolution (e.g., Hickox et al. 2009), a less biased sample of AGNs will help to make this picture clearer and allow us to better interpret the role AGNs play within their host galaxies.

In this paper, we aim to identify AGNs in the Great Observatories Origins Deep Survey (GOODS) North and South fields beginning with an optical variability-selected sample. AGNs are known to vary on timescales of months to years in the optical with 90%–100% of AGNs identified via other means observed to vary over the course of several years (e.g., Koo et al. 1986; Schmidt et al. 2010). The mechanism behind the variability is still uncertain, though the leading explanation involves disk instabilities (Pereyra et al. 2006) or changes in the amount of infalling material (Hopkins & Beacom 2006). Regardless of the physics behind variability, AGNs are observed to display brightness variations of several percent over many years. There is additional evidence of an increase in variability amplitude for intrinsically fainter AGNs (Bershady et al. 1998; Vanden Berk et al. 2004) making variability a particularly effective means for identifying low-luminosity AGNs. We build on the results of previous successful surveys to identify AGN candidates in the nuclei of galaxies found in deep Hubble Space Telescope (HST) surveys (e.g., Sarajedini et al. 2003, 2006). The high-resolution HST images allow us to obtain accurate photometry within small apertures (r≲ 02), thus allowing us to probe lower AGN/host galaxy luminosity ratios than can be done using ground-based images. We expect ∼60% of AGNs in the GOODS fields to be identified as variable with the time sampling of five epochs separated by 45 day intervals. This estimate is based on the results of Webb & Malkan (2000), who found that 60% of Seyfert/QSOs in their study revealed significant variability on month-to-month timescales.

We also explore X-ray- and mid-IR-selected AGNs identified in the GOODS fields and compare with the variability survey. These wavelengths are powerful probes of AGNs and are especially important for uncovering obscured sources. The multiwavelength data provide important confirmations of the AGN nature of the variables as well as the quantification of the use of optical variability in identifying sources of varying levels of obscuration. We also compare the variables to spectroscopically selected AGNs in the GOODS fields.

We describe the AGN sample drawn from the variability analysis in Section 2. In Section 3, we compare this sample to those identified via X-rays, mid-IR power-law spectral energy distributions (SEDs), and optical spectroscopy. We discuss the redshift distribution, absolute magnitudes, and colors of the variables, X-ray and mid-IR AGNs in Section 4. In Section 5, we discuss how variability-selected AGNs fit with the current picture of AGN/galaxy evolution in part by deriving Eddington ratios for the variables in our survey. Throughout the paper, we assume a flat, cosmological constant-dominated cosmology with parameter values Ω_Λ = 0.7, Ω_M = 0.3, and H_o = 70 km s⁻¹ Mpc⁻¹. The data presented in this paper are in the AB photometric system.

The GOODS fields consist of two regions of the sky, each ∼130 arcmin² in size and located at R.A. = 3^h32^m, decl. = −27^d48^m and R.A. = 12^h37^m, decl. = +62^d15^m. These fields were imaged with HST's Advance Camera for Surveys Wide Field Camera (ACS WFC) in four passbands (F435W, F606W, F775W, and F850LP). All but the B-band images (F435W) were obtained in epochs separated by 45 days. We chose to perform our analysis using the F606W (hereafter V-band) images to maximize sensitivity to varying AGNs within host galaxies, given the fact that several studies find increased variability amplitude with decreasing wavelength (di Clemente et al. 1996; Sarajedini et al. 2003). The south field images were obtained between 2002 August 1 and 2003 February 5 with five epochs separated by ∼45 days for a total time baseline of 6 months. The north field images were obtained between 2002 November 21 and 2003 May 23 with the same time sampling as in the south. Two exposures per pointing per epoch were combined to produce an image with a total exposure time of ∼1000 s.

The GOODS v2.0 source catalog (http://archive.stsci.edu/pub/hlsp/goods/catalog_r2/) provided initial coordinate positions for identifying objects in the individual epoch ACS images. We found 11,519 sources in the combined fields that were visible in all five epochs. To ensure that the centers of all sources were accurately determined and matched across the epochs, objects were recentered using the IRAF task center and sources whose centers had large pixel shifts (i.e., more than 3σ of the average shift for sources in a given ACS image) were visually inspected. Sources that fell too close to the CCD edge or with a cosmic ray close to the object center in one or more epochs were rejected. After this process, 10,161 sources remained.

We used IRAF phot to determine aperture photometry for each source using three aperture sizes, r = 2.5 pixels, 3.5 pixels, and 5 pixels where the pixel scale is 0.05 arcsec pixel⁻¹. The smallest aperture size is ∼twice the FWHM of a typical unresolved source and was initially selected as the primary aperture to identify AGNs in galaxies, which should originate from an unresolved nuclear region. However, we carry out the variability analysis using the 5 pixel aperture data to ensure that changes in the point-spread function (PSF), which would be more significant for smaller apertures, do not result in an overestimate of variables. Magnitude differences were determined between epochs 2 through 5 and the first epoch. Small zero-point offsets of 0.01–0.03 mag were determined among the epochs and applied to the photometry. For every source, we then determined the mean magnitude and standard deviation around the mean. We also calculate the concentration index, CI, as the magnitude difference between the r = 2.5 pixel photometry and the r = 5 pixel photometry. Point-like, unresolved sources should have small CI values, while extended sources will occupy a range of higher values.

Figure 1 shows the mean nuclear V-band magnitude versus standard deviation for r = 2.5 pixel apertures (Figure 1(a)), r = 5 pixel apertures (Figure 1(b)), and the CI (Figure 1(c)). The expected increase in standard deviation values at fainter magnitudes is observed due to increasing photometric noise. The spread in standard deviation values is generally the same using either the small or large aperture photometry and would suggest that variations in the PSF are not producing significantly increased standard deviation scatter. Figure 1(c) shows a clear separation between the resolved and unresolved sources. Based on this figure, we define objects having CI values of ⩽0.2 as unresolved or point-like and those with higher values as resolved or extended.

Zoom In Zoom Out Reset image size

Applying this division, we find that unresolved sources generally have higher standard deviation values than do the extended sources across the range of magnitudes (Figure 2). The scatter in the standard deviation values do not decrease significantly when using the r = 5 pixel photometry. We analyzed the unresolved sources using r = 10 pixel aperture photometry as well and found a similar level of scatter. This higher level of scatter among the standard deviation values of the unresolved population is likely reflecting the difficulty in achieving photometric repeatability for point-like sources and may be enhanced by variations in the PSF which are not as significant for extended sources. Since simply using larger apertures does not reduce the scatter, we account for this difference by continuing the variability analysis separately for the unresolved and extended sources throughout the remainder of this paper. In this way, we determine the variability threshold for unresolved sources independently and can allow for this larger scatter. We use small apertures (r = 2.5 pixel) for the extended sources to isolate the varying nuclei of galaxies and increase sensitivity. For the unresolved sources, we use the r = 5 pixel apertures to avoid any possible sensitivity to PSF changes. In addition, since these are generally point-like in morphology, there is no need to use the smaller aperture to block light from an extended galaxy component.

Zoom In Zoom Out Reset image size

We apply both a bright and faint magnitude cut to the variability survey. A bright limit of V_2.5=20.0 was applied to remove several marginally saturated unresolved objects at the bright end of the distribution. These appear to be mainly foreground stars for which accurate photometry cannot be determined. We identify the faint magnitude limit for our variability survey from the source number counts as a function of magnitude. The number of objects significantly drops beyond V_2.5= 26.5. Therefore, we limit the survey analysis to V_2.5= 26.0 to avoid significant photometric incompleteness.

To identify the variability threshold, the standard deviation values are separated into magnitude bins, allowing the bin limits to vary in size with larger bins at the bright end and smaller bins at the faint end to maintain similar numbers of objects ranging from 15 to 40 sources per bin. The greatest outliers from the mean magnitude of each bin were excluded and the resulting histograms were fit with Gaussians to determine the distribution center (mean standard deviation) and the sigma (i.e., spread) in standard deviation values at a given magnitude. At fainter magnitudes the standard deviation increases as well as the spread in those values. We fit a second-order polynomial to both the mean standard deviation values in each bin and 1σ of the Gaussian distribution in each bin. This is done in order to characterize the photometric error as a smoothly changing function of the source magnitude. Both the mean of the standard deviation and the spread in values around that mean are determined independently. Figures 3(a) and (b) show the variability plot for the GOODS South field with the mean standard deviation (solid thick line) and the mean plus 2σ and plus 3σ (dashed lines) for the point sources and extended sources separately. Figures 3(c) and 3(d) are the same for the north field. Determining the expected error at each magnitude in this empirical way provides the most realistic estimation of the photometric error on the standard deviation since we carry through the effects of object and sky Poisson noise, readout noise, and any limitations imposed by minor errors in the sky zero points used.

Zoom In Zoom Out Reset image size

Sources at or above the mean standard deviation plus 3σ were flagged for visual inspection. In the south field, 126 extended sources and 27 point sources were flagged and 94 extended and 10 point sources in the north field. Visual inspection of these outliers revealed that ∼50 in each field contained evidence of cosmic rays or fell in a location too close to the edge of a CCD and thus suffered from high background noise in one or more epochs. These sources were removed from the survey. About 15 objects per field were determined to have centering problems and these were manually recentered. With all outliers checked for accuracy, we recalculated the variability threshold for each field.

After comparing the variables against the known supernova (SN) catalogs of Strolger et al. (2004), we identified one SN in the north and one in the south field from our extended source list. We also compare to several published redshift surveys (see Section 4) and identify a number of spectroscopically confirmed stars (11 in each of the fields). Since the aim of our paper is to study the AGN population in these fields, we remove the known SNe and stars from the survey.

The final sample of galaxies surveyed for AGN variability is shown in Figure 4. We normalize the y-axis by dividing the standard deviation of each galaxy (with the mean standard deviation at that magnitude subtracted) by the 1σ Gaussian fit determined for the galaxies at each magnitude, the majority of which are assumed to be non-varying. We call this the variability significance parameter. The variability significance will have a value of 3 for objects that are varying by exactly three times the 1σ Gaussian fit (i.e., spread) in standard deviation values at a given magnitude. Since we subtract off the mean standard deviation, some sources will have negative variability significance values.

Zoom In Zoom Out Reset image size

To determine the threshold of variability and estimate the number of false positives, we employ an empirical approach. We examine the cumulative distribution of variability significance values for resolved and unresolved galaxies in GOODS (Figure 5). For a non-varying population, a smooth decline in the distribution should be observed. A population of significant outliers should appear as a break in that distribution with a shallower decline at higher variability significance values. A fit to the distribution of the non-varying sources (long dashed line) helps to identify the break at a significance value of ∼3.1 for the extended sources and at ∼3 for the unresolved sources (short dashed line). We use this value for the threshold of variability and indicate it with a dashed line in Figure 4. To estimate the number of outliers, we extend the fit to the smooth distribution of non-varying galaxies at low significance values to the y-intercept. Integrating below this line and beyond the variability threshold (shaded region in Figure 5) yields an estimate of the expected number of outliers from the non-varying population. With a total of 3775 extended sources and 399 unresolved sources, the shaded regions represent ∼4 false positives among the extended and ∼3 false positives among the unresolved sources for a total of 7 expected false positive variables.

Zoom In Zoom Out Reset image size

Based on our determined variability thresholds, we identify 42 variable galaxies out of a possible 2055 galaxies in the GOODS South field and 43 variables out of 2119 galaxies in the GOODS North field. This represents 2% of all galaxies within the flux limits of the survey that display significantly varying nuclei over the 6 month time interval. We estimate that ∼7 of the 85 total variables, or ∼8%, are false positives. Table 1 lists the variables identified in the GOODS survey.

2.1. Comparison with Other Variability Surveys in GOODS

The GOODS North and South fields have been the target of deep X-ray observations with Chandra and mid-IR imaging with Spitzer IRAC and MIPS. We compare AGN candidates selected via X-ray detection and mid-IR power-law SED fitting to our variability-selected AGN candidates to investigate their multiwavelength properties and various biases among the different selection techniques. We also examine evidence for AGN based on emission line properties from published redshift surveys in these fields.

3.1. X-ray-Detected AGN

All optical sources surveyed for variability were matched against the list of X-ray point source detections from the 2Ms Chandra Deep Field surveys in the north field (Alexander et al. 2003) and the south field (Luo et al. 2008). We refer to the "variability survey" as those optical sources from the GOODS v2.0 catalog that were visible in all five epochs with good-quality images (i.e., containing no cosmic rays or CCD edge effects) and having nuclear magnitudes brighter than V_2.5 = 26.

In the south field, Luo et al. (2008) found 218 optical matches to the 462 X-ray sources in the main catalog with a matching radius of 05. We find that 115 of these X-ray/optically matched sources fall within our variability survey. A supplementary catalog, produced with greater sensitivity in the outer regions of the CDF-S due to the inclusion of ∼250 ks from the E-CDF-S, consists of 86 sources, of which 3 match to sources in the variability survey. A third catalog of bright optical sources with lower X-ray significance detections consists of 30 sources, of which 15 have counterparts in the variability survey. Although most objects in the optically bright catalog are not likely to be AGNs based on their low X-ray-to-optical flux ratios, we identify these matches since variability would cast light on the nature of the X-ray emission.

In the north field, Alexander et al. (2003) produced a main source catalog containing 503 sources in the CDF-N, and a bright source catalog containing 75 sources with lower significance X-ray detections matched to optically bright galaxies. Barger et al. (2003) identified optical matches and optical source positions for the main source catalog objects based on ground-based Subaru Suprime-Cam observations (Capak et al. 2004). We match the GOODS-N v2.0 positions to these optical positions rather than the X-ray positions for greater matching accuracy. There are 139 objects in the variability survey with matches to X-ray sources in the main catalog within 05 of the optical counterpart position given in Barger et al. (2003). Two objects with offsets between 05 and 10 were included as matches since their X-ray positions matched the GOODS-N optical positions within ∼06. Among the bright source catalog, 49 objects match within 05 of an X-ray source.

We find 21 of the 42 variables in the south field and 21 of the 43 variables in the north field are X-ray detected, yielding a total of 42/85 or 49% of the variables associated with X-ray sources (blue triangles in Figure 4). This fraction is similar to that found in the HDF-N (Sarajedini et al. 2003) and higher than recent ground-based optical variability surveys (Trevese et al. 2008; Morokuma et al. 2008), where 60%–70% of variables were not detected in X-rays. In Figure 6, we plot the soft band flux (0.5–2 keV) against the R-band AB magnitude for all X-ray-detected sources in the variability survey.⁶ The solid lines indicate the location of most AGNs with log(F_x/F_opt) between −1 and 1 and the dashed line indicates log(F_x/F_opt) = −2. Most non-AGN X-ray sources fall below the dashed line. It can be seen that the majority of bright source catalog objects (open circles) are not likely to be AGNs based on their low F_x/F_opt values. The green points indicate the variable sources. We find that the majority of variables are likely to be AGNs based on their location in this diagram but a significant number (19%) have low F_x/F_opt values. None of the objects in the X-ray bright source catalog for the GOODS-S field were found to be variable and only one from the GOODS-N field (ID 2426) is variable. This object lies close to the limit of likely AGN F_x/F_opt values.

Zoom In Zoom Out Reset image size

Figure 6 also shows the R-band magnitude for variables without an X-ray counterpart (green asterisks). These non-X-ray-detected variables cover a range of optical magnitudes but tend to be fainter than the X-ray-detected variables sources (average R_AB = 22.1 for X-ray-detected sources and 23.3 for non-X-ray-detected sources). Thus, in general, they are expected to be more difficult to detect at the flux limits of the Chandra survey. The range of optical magnitudes for the non-X-ray-detected sources would indicate that if the optical variability is due to the presence of an AGN, they are likely to have low F_x/F_opt values, similar to what is found for ∼20% of the X-ray-detected sources. Additionally, since the optical magnitudes represent the total galaxy+AGN light, all of the points on the diagram will have brighter R-band magnitudes than for the AGN component alone, the extent of which depends on the ratio of AGN to host galaxy luminosity. Therefore, some of the X-ray non-detections may be extremely faint optical AGNs that lie below the flux limits of the X-ray survey.

We explore the variable nature of the X-ray-detected sources and find that 21 of the 118 X-ray sources in GOODS-S (combining the main and extended CDF-S catalogs) are significant variables. In GOODS-N, 21 of the 140 X-ray sources are significant variables. Thus, 16.3% of the X-ray sources with optical counterparts in the variability survey are identified as significant variables. We do not include the non-variable bright source catalog objects in this statistic since the aim of our paper is to investigate the varying and non-varying AGN population. With the same reasoning, we may also consider X-ray sources below the dashed line in Figure 6 as less likely AGN candidates. Only two variables lie below this line in the north field (one very close to the limit) and none in the south field. If we exclude X-ray sources below log(F_x/F_opt) = −2, we find 21 of 75 X-ray sources in the south and 19 of 92 in the north are variable, increasing the percentage of X-ray sources with optically varying counterparts to 24%. This percentage is similar to that found in other variability studies. Klesman & Sarajedini (2007) found that 26% of the X-ray source population in the GOODS-S field were optical variables. Since this earlier analysis examined X-ray sources detected in the previously published 1Ms Chandra survey, many of the fainter X-ray sources and those likely to have lower F_x/F_opt values were already omitted from their sample (see discussion in Klesman & Sarajedini 2007). Of the 21 objects we identify as optically variable X-ray sources, 18 were also identified in Klesman & Sarajedini (2007). Of the three that were not identified in that paper, one was not included in their analysis since it was only detected in the 2Ms survey. The other two fell just below their variability significance threshold.

We expect that some X-ray-detected sources do not reveal significant variability in their optical counterparts due to obscuration. We quantify obscuration using the hardness ratio, HR, defined as the hard band source counts (HB; 2–7 keV) divided by the soft band counts (SB; 0.5–2 keV). Higher values of HR are found for more obscured sources. Figure 7 shows the hardness ratio versus the variability significance for all of the X-ray-detected sources in our survey with enough counts in either the soft or hard band. Open points are upper limits without enough counts in the hard band to compute a ratio. The solid line shows the threshold above which we consider the source to be significantly varying. As in Klesman & Sarajedini (2007), we find that the majority of variables are "soft" sources, having HR values less than 0.5. These are also the most significant variables in our sample. We find that 36% of the softest sources are variable, a dramatic increase from the 16% identified as variable overall. The number of optical variables falls significantly among harder, more obscured, X-ray sources. For sources with HR values between 0.5 and 2, we find 8% are optically variable. For the hardest sources with HR values greater than 2, only 1 variable is found accounting for 3% of the X-ray-detected sources in our survey. We observe that the variability significance declines with increasing hardness ratio consistent with findings in Klesman & Sarajedini (2007).

Zoom In Zoom Out Reset image size

We continue to examine the non-X-ray-detected variables by performing an X-ray stacking analysis for these sources. The stacking was done for the full, soft and hard bands (Table 2). The last column lists the signal-to-noise ratio (S/N) for the stacked samples. Calculating the source significance instead, the significance for the full, soft, and hard bands for the GOODS-N is 1.6σ, 3.4σ, and 0σ, indicating a slightly significant result only for the soft band. For the GOODS-S field, source significance is 2.5σ, 2.3σ, and 1.5σ. The overall results of the analysis are consistent with the trends observed in Figures 6 and 7, that optical variability is more likely to be detected among X-ray soft sources and that many non-X-ray-detected variables may emit very weak X-rays and have F_x/F_opt values lower than typical AGNs.

Table 2. X-Ray Stacking Results

Energy Band (keV)	No. of Stacked Sources	Total Countsa	Total Background Countsb	S/Nc
GOODS-N
0.5–0.7	20	228 ± 15	205.1 ± 0.5	1.5
0.5–2.0	20	96 ± 10	67.7 ± 0.3	2.9
2.0–7.0	19	127 ± 11	130.3 ± 0.4	−0.3
GOODS-S
0.5–0.7	19	273 ± 17	234.1 ± 0.5	2.4
0.5–2.0	20	99 ± 10	78.4 ± 0.3	2.1
2.0–7.0	18	170 ± 13	152.0 ± 0.4	1.4

Notes. ^aTotal counts extracted within a radius r = 175 of source location. ^bBased on 1000 samples taken within an annulus r = 10'' to r = 30'' around source position. ^cS/N = S/(sqrt(S+B)), where S is net source count flux per source and B is background count flux per source.

Download table as:  ASCIITypeset image

The lack of X-ray emission from about half of the sources is not well understood, but is also observed in several previous variability surveys (Trevese et al. 2008; Morokuma et al. 2008). Brandt & Hasinger (2005) discuss some possible explanations (e.g., the lack of an accretion-disk X-ray corona) which may play a role in addition to the possibility that weak X-ray emission is present but not detectable at the current X-ray survey depths.

3.2. IRAC Power-law-detected AGNs

3.3. Spectroscopically Identified AGNs

Both the north and south GOODS fields have been the target of extensive spectroscopic follow-up. For the south field, we have the spectroscopic catalogs of Szokoly et al. (2004), Cimatti et al. (2002), La Fevre et al. (2004), Vanzella et al. (2008), Mignoli et al. (2005), and Popesso et al. (2009). In the north, we have the compilation of spectroscopic data in Barger et al. (2008) which includes the catalogs of Cowie et al. (2004), Reddy et al. (2006), Cohen et al. (2000), and Cohen (2001). For sources without spectroscopic information, photometric redshifts were obtained from the Rainbow Database (Perez-Gonzalez et al. 2008; Barro et al. 2010) for the south field and from Fernanez-Soto et al. (1999), Capak et al. (2004), and Bundy et al. (2009) for the north.

Among the 4174 galaxies surveyed for AGN variability in the combined fields, 2371 have spectroscopic redshifts (57%) and an additional 1497 have photometric redshifts yielding redshift information for 93% of the galaxies. Among the 85 variable galaxies, 62 have spectroscopic redshifts (73%) and the remaining 23 have photometric redshifts.

Figure 8(a) shows the redshift distribution for all galaxies in our survey compared to the distribution of optical variables multiplied by 12 for comparison. The variables have a higher median redshift than the total population with 35 lying at redshifts greater than 1. The percentage of galaxies hosting optically varying nuclei increases with increasing redshift, from ≲2% at low redshifts to almost 10% at redshifts greater than 3 (Figure 8(b)). This is consistent with the fact that the AGN luminosity function shows their numbers peak at redshifts between 0.7 ≲ z ≲ 2, with more luminous AGN peaking at higher redshifts than fainter AGN (e.g., Bongiorno et al. 2007).

Zoom In Zoom Out Reset image size

To further explore the nature of the AGN candidates, Figure 9 shows redshift versus rest frame absolute V magnitude for all galaxies in our survey with the variable AGN candidates (green circles), X-ray-detected sources (blue triangles), and mid-IR sources (red squares) indicated. We determine rest-frame magnitudes using the v2.0 GOODS catalog b, v, i, and z (F435W, F606W, F775W, and F850LP) galaxy photometry (MAG-AUTO) and U-band photometry from Wolf et al. (2004, 2008; GOODS-S) and Capak et al. (2004; GOODS-N). These data, together with the redshift information, allow us to compute k-corrected, rest-frame magnitudes and colors for each galaxy in our survey using $\emph {kcorrect}$ (Blanton et al. 2003).

Zoom In Zoom Out Reset image size

Figure 9 reveals that AGN candidates generally have brighter optical magnitudes than the non-AGN galaxies. Their luminosity may be partly enhanced by the AGN component, though AGN host galaxies are also typically found to be brighter and more massive than galaxies not hosting AGNs (e.g., Brusa et al. 2009). The X-ray-detected AGN candidates appear to lie in the brightest galaxies. Mid-IR power-law galaxies cover a range of magnitudes with most lying at the bright end and a few occupying faint galaxies. The mid-IR sources are also generally at higher redshift than the X-ray sources, consistent with other studies of X-ray and mid-IR AGNs (Alonso-Herrero et al. 2006). Most optically bright variable AGN candidates are also X-ray sources. The non-X-ray-detected optical variables tend to have fainter absolute magnitudes at all redshifts. Thus, optical variability appears to identify some intrinsically faint AGN/host galaxies and may do so more efficiently than X-ray or mid-IR AGN identification.

Figure 10 is the color–magnitude diagram for the galaxies in our survey with the same symbols used to identify optical variables, X-ray, and mid-IR sources. We plot the rest-frame U − V color versus the V-band absolute magnitude for galaxies divided into two redshift bins, z = 0.2 < z < 0.76 and 0.76 < z < 1.6. The second bin is more than twice the co-moving volume of the first bin. We see the bimodal color distribution observed in recent surveys (e.g., Hogg et al. 2002) among our survey galaxies. The dashed line indicates the limit of red sequence galaxies as determined in Bell et al. (2004) at the center of each redshift bin.

Zoom In Zoom Out Reset image size

Several recent findings suggest that the integrated galaxy colors for the AGN candidates in our survey should largely represent the colors of their host galaxies and in most cases are only marginally effected by the light from the nuclear AGN. Recent results by Cardamone et al. (2010) for X-ray-selected AGNs show little color contamination (<0.1 mag) from the nucleus in rest-frame U − V. Likewise, Hickox et al. (2009) find a maximum of 0.4–0.5 mag color corrections are needed for the bluest AGN sources in (u–r). Exceptions are the most luminous X-ray/optically variable AGN candidates, where we expect the blue colors are indeed significantly affected by the AGN light. These would include several bright (M_V < −23) blue optical variables/X-ray sources in the lower left quadrant of Figure 10(b).

The AGN galaxy colors cover a range in U–V and fill the region known as the green valley, as observed previously for X-ray-detected AGNs (e.g., Nandra et al. 2007). Figure 11 shows the color histograms for optical variables, X-ray- and mid-IR-selected AGN candidates compared to the total galaxy color population. Here, we include only galaxies at 0.2 < z < 1.6 and correct for the slope of the dashed line in Figure 10. We have also removed the nine bright, blue sources in the high-redshift color–magnitude diagram (Figure 10(b) that are likely to be affected by the AGN light and may not represent the true host galaxy colors. We find that the X-ray sources and optical variables occupy galaxies at a range of colors, including many within the green valley. While the X-ray sources appear to peak in the green valley, the variables have a flatter distribution that extends blueward of the X-ray host distribution and into the red sequence galaxies. There are few mid-IR AGN candidates at z < 1.6, but those in this redshift range show a relatively flat distribution. This is consistent with the results of Hickox et al. (2009) who find similar distributions among a sample of 585 AGNs selected via X-ray and mid-IR colors. Recently, Cardamone et al. (2010) found that 75% of X-ray hosts in the green valley have colors consistent with young star-forming galaxies reddened by dust. The blue extension of the optically variable AGN hosts may imply that variability-selected AGNs are generally less dust reddened. In any case, we find a number of AGN hosts, several identified only through their variable nature, with blue colors indicative of current star formation.

Zoom In Zoom Out Reset image size

We have shown that optical variability can be used to identify a significant population of AGN candidates in deep HST fields. The colors and magnitude ranges of the variable hosts are similar to those of other AGN samples selected via X-rays or mid-IR emission but extending to bluer colors and fainter magnitudes. With about 50% of the variables not identified as AGNs through other means, we investigate how the inclusion of this population fits with recent theories of AGN/galaxy evolution.

In the current picture of AGN and galaxy evolution, dark matter halos grow in the early universe to masses of 10¹²–10¹³ M_☉ when a dramatic event (e.g., major merger) triggers luminous quasar activity and rapid growth of the central BH producing a dynamically hot stellar bulge component in the host galaxy (Hopkins et al. 2008, 2009). Other events such as disk instabilities may also trigger accretion onto the BH and grow the stellar bulge. After this period of growth, star formation in the galaxy must be quenched in order to produce the observed population of passively evolving galaxies on the red sequence. There has been speculation that the quenching results from the event that fuels the quasar and/or is related to radiative feedback during the powerful AGN phase (Alexander et al. 2005; Brand et al. 2006). The high virial temperature of the dark matter halo should stop accretion of cold gas and thus inhibit further star formation (e.g., Faber et al. 2007). The transition of a blue, star-forming galaxy to a galaxy on the red sequence occurs relatively quickly (1–2 Gyr; Barger et al. 1996) which is supported by the dichotomy of observed galaxy colors seen here and in other surveys. There is some evidence that further accretion onto the SMBH occurs during this transition and would explain the observed "green" colors of X-ray-selected AGNs (Schawinski et al. 2007), though recent results suggest that many of those may actually be dust reddened star-forming galaxies (Cardamone et al. 2010). Galaxies on the red sequence may continue to experience quenched star formation due to mechanical feedback from the AGN which prevents the accretion of cool gas onto the SMBH (Churazov et al. 2002).

Optical variability is a ubiquitous observational characteristic of accretion onto the SMBH and thus we might expect to identify AGNs in the full range of evolutionary stages described above. However, since the variability likely originates from the UV/optically bright accretion disk, variability selection should be most sensitive to unobscured, less dust-reddened AGNs which is largely consistent with our findings. The variables that are also X-ray and mid-IR sources have luminosities, redshifts, and host galaxy colors consistent with the findings of Hickox et al. (2009). Using the observed colors, clustering properties, and SMBH accretion rates, they speculate that the X-ray-selected AGNs observed at 0.25 < z < 0.8 live in large dark matter halo galaxies that have recently experienced the buildup of their bulges and quenching of star formation. These galaxies also exhibit a range of Eddington ratios (10⁻³ to 1) which would indicate declining accretion rates onto the SMBH. The mid-IR AGNs in their study are associated with galaxies having less massive dark matter halos and higher Eddington ratios than the X-ray-selected sample. Thus, these galaxies have not reached the critical halo mass for the quenching of star formation and therefore are more commonly found in blue host galaxies. Though we do not examine the clustering properties of the different samples here, the colors and magnitudes of the X-ray-selected and mid-IR-selected variables in our survey would appear to be consistent with this interpretation.

To further investigate the variable AGN candidates found in our survey and attempt to understand their placement in the above evolutionary scenario, we estimate the BH masses and accretion rates for the variability-selected AGN candidates at redshifts less than z < 1.6. Higher redshift data are excluded in this analysis to avoid the high-luminosity AGN which may contaminate galaxy colors and bulge estimates for the host. To determine Eddington ratios, however, we first need to estimate the luminosity of the AGN. We expect the AGN component to be ∼10% or less of the total galaxy flux in most cases. This is based on the observed variability in the nucleus compared with that expected from AGN structure function flux amplitudes at time intervals similar to our data (e.g., Butler & Bloom 2011). Nonetheless, we can make a crude estimate of the AGN flux by measuring the nuclear magnitude within r = 0125 apertures, twice the FWHM for the ACS images. This aperture contains the highest concentration of the AGN light and represents an upper limit on the AGN luminosity since this measurement will contain essentially all of the AGN plus some underlying flux from the host galaxy. The nuclear V-band magnitudes are converted to rest-frame B-band absolute magnitudes. We use the bolometric corrections of Hopkins et al. (2007) to convert to bolometric luminosity. Figure 12(a) shows the upper limit bolometric luminosity versus redshift for the variables at z < 1.6, where X-ray-detected variables are shown as black points and non-X-ray-detected variables as green. We find that the upper limit bolometric luminosities for non-X-ray-detected variables are generally fainter than that for the X-ray-selected sample at all redshifts but with a large dispersion.

Zoom In Zoom Out Reset image size

To estimate BH masses, we compute the stellar bulge masses for our variable galaxies. We used GALFIT (Peng et al. 2002) to fit a two-component model consisting of an exponential disk and n = 4 sersic profile bulge in both the V-band (F606W) and i-band (F775W) images. Fits with an n = 2 sersic bulge were also computed but showed no significant differences with the n = 4 fits. With the magnitudes for each component determined, we then calculated the total galaxy and bulge masses using the color and galaxy redshift. Reliable bulge masses were calculated for 48 of the 62 variable galaxies at z < 1.6 using this technique. In some cases a significant bulge component was not identified in one of the bands, making a bulge mass measurement impossible. We use the relation from Marconi & Hunt (2003) to convert bulge mass to BH mass and plot this as a function of redshift in Figure 12(b). The variables have a range of BH masses with an average of 7.2 log(M_BH[M_☉]). There is no apparent difference between the non-X-ray- and X-ray-detected variables at z < 1. At higher redshifts, the non-X-ray-detected variables appear to have lower mass BHs in general, though the number of sources at these redshifts with reliable bulge measurements is fewer. We note that any significant contribution by the AGN component could produce an artificially high bulge mass and consequently overestimate the BH mass. While we expect the contribution of the AGN component to be negligible in the majority of our sources, the higher redshift and X-ray-detected sources are more likely to be susceptible to this effect based on their overall colors and luminosities (as discussed in Section 4). This may account for the slightly higher BH masses for z>1 variables that are also X-ray detected in Figure 12(b).

Finally, we compute the Eddington ratios using the equation L_Edd = 1.3 × 10³⁸ (M_BH/M_☉) erg s⁻¹. Figure 12(c) shows the Eddington ratio as a function of redshift. Several sources at z>1 have luminosities brighter than Eddington which is likely caused by the fact that our bolometric luminosity is an upper limit. Thus, these values are less reliable estimates of the true Eddington ratios of these sources. Nonetheless, we find no discernible difference among the X-ray and non-X-ray-detected variable AGNs across the range of observed redshifts.

These results suggest that the hosts of variability-detected AGNs comprise a mix of galaxy-AGN types. Some of the variables appear to be similar to other X-ray-selected sources, having hosts with significant stellar bulges and recently quenched star formation, followed by a decline in the accretion rate onto the BH (Hickox et al. 2009). In other cases, the SMBHs may be continuously or stochastically fueled, possibly at lower levels by stellar winds or bars in disk galaxies. Such systems could have slightly higher Eddington ratios and be in intermediate mass, later-type hosts (Hopkins et al. 2009) much like the mid-IR/optical AGNs discussed in Hickox et al. (2009). This scenario would be consistent with the observed lower-luminosity AGNs within bluer host galaxies.

We have conducted a variability survey for AGN candidates in the GOODS fields and compared this selection technique with results from other multiwavelength surveys of these fields. Using small aperture photometry, we identify 85 galaxies with significantly varying nuclei in the HST ACS V-band images. This represents 2% of all galaxies surveyed to a limiting magnitude of V_2.5 = 26 with an estimated 7 false positives. Forty-two of the 85 variables are associated with X-ray sources detected in the Chandra 2Ms surveys of these fields and 14 of the variables have power-law SEDs through the Spitzer IRAC mid-IR bands. All but one of the mid-IR sources is also an X-ray source, resulting in a total of 43 of 85 variables (51%) with additional multiwavelength evidence for the presence of an AGN. Optical variability is more sensitive to soft X-ray sources with fewer optical variables found to be associated with the harder X-ray-detected sources in these fields. This is consistent with the results of X-ray stacking for the non-X-ray-detected variables which reveals a marginally significant detection in the soft X-ray band (0.5–2 keV) only. We also find that the slope of the power-law fit through the mid-IR bands for optical variables is consistent with BLAGN-like SEDs for two-thirds of the variable AGN candidates with the other third having steeper slopes like those for NLAGNs and ULIRGs. Likewise, among spectroscopically identified AGNs, 74% of BLAGNs and 15% of NLAGNs show optical variability.

Our findings indicate that optical variability most efficiently identifies less obscured, Type I like AGNs, though a surprisingly significant fraction (15%–30%) appear to have some degree of obscuration revealed through the lack of an observed broad-line region, slightly harder X-ray emission, or the slope of the mid-IR SED. The overall number density of variable AGN candidates is about half that of likely AGN X-ray sources with optical counterparts in the GOODS fields and almost three times that of mid-IR power-law AGNs with optical counterparts. Thus, variability uncovers a significant population of AGNs which complements detection methods at other wavelengths.

The variable AGN candidates span a broad range in redshift with the percentage of galaxies hosting variable nuclei increasing from ≲2% at low redshift to ∼10% at z ∼ 3. The total galaxy+AGN absolute magnitudes reveal that AGN candidates (variable, X-ray-, and mid-IR-detected candidates) are generally found in brighter galaxies at a range of colors spanning that of the total galaxy population. The color distribution of X-ray-detected AGN hosts reveals a peak in number density between the blue, star-forming and red sequence galaxies. The variables follow a similar trend but with an extension toward bluer colors and fainter absolute magnitudes.

Based on the derived bolometric luminosities, BH masses and Eddington ratios for the variable galaxies in our survey, we find that non-X-ray-detected variables at z < 1.6 generally have lower luminosities than X-ray-detected variables, but similar BH masses and Eddington ratios. Thus, some of the variables have characteristics like other X-ray-selected samples, revealing significant bulges and the onset of quenched star formation and decreased BH accretion rates, while others may be intermediate mass systems still forming stars and accreting at a continuous, lower-level, possibly through stellar winds or bars.

The results of this survey suggest that variability is a promising technique to identify samples of AGNs extending to low luminosities out to z ∼ 3. These results are especially encouraging in light of future planned multi-epoch observing programs like that for the LSST. Observations with longer and better sampled light curves will produce more robust and complete variability-selected AGN samples in the low-luminosity, z>1 regime comparable to the well-quantified light curves and structure functions for QSOs (e.g., MacLeod et al. 2010). Together with multiwavelength surveys using a variety of detection techniques, the AGN population can be more powerfully probed and understood in the context of galaxy evolution.

We thank Seth Cohen and Namish Hathi for providing a photometric redshift compilation for the GOODS North field, Amy Barger for supplying the list of spectroscopic AGNs identified in the GOODS South field, Conor Mancone for assistance with k-correction of galaxy magnitudes, and Philip Orange for work on the X-ray stacking analysis presented here. We also thank the referee for a careful reading of this paper and helpful comments and suggestions.

This paper is based on observations with the NASA/ESA Hubble Space Telescope, obtained at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555 and observations with the Keck telescope, made possible by the W. M. Keck Foundation and NASA. Funding was provided by STScI grant AR-09948.01 and NSF CAREER grant 0346691.

P.G.P-G. acknowledges support from grants AYA2009-10368 and CSD2006-00070, and the Ramón y Cajal Program, all financed by the Spanish Government and/or the European Union.