# Morphologies of mid-IR variability-selected AGN host galaxies

Morphologies of mid-IR variability-selected AGN host galaxies Abstract We use multi-epoch 3.6 and 4.5 μm data from the Spitzer Extended Deep Survey (SEDS) to probe the AGN population among galaxies to redshifts ∼3 via their mid-IR variability. About 1 per cent of all galaxies in our survey contain varying nuclei, 80 per cent of which are likely to be AGN. Twenty-three per cent of mid-IR variables are also X-ray sources. The mid-IR variables have a slightly greater fraction of weakly disturbed morphologies compared to a control sample of normal galaxies. The increased fraction of weakly distorted hosts becomes more significant when we remove the X-ray emitting AGN, while the frequency of strongly disturbed hosts remains similar to the control galaxy sample. These results suggest that mid-IR variability identifies a unique population of obscured, Compton-thick AGN revealing elevated levels of weak distortion among their host galaxies. galaxies: active, galaxies: evolution, infrared: galaxies 1 INTRODUCTION Active galactic nuclei (AGN) are galaxies which accrete significant amounts of material on to their central super massive black holes (SMBHs) and have long been one of the most interesting phenomena in extragalactic astronomy. Questions relating to their formation, structure, and SMBH accretion have been contemplated since theories explaining the powering mechanism of AGN were developed (Lynden-Bell 1969). AGN may represent an evolutionary phase for many galaxies and could have a significant impact on their star formation. To understand the role played by AGN in the evolution of galaxies, it is necessary to identify them in galaxy surveys out to redshifts of ∼3 where significant AGN fuelling and bulge growth occur. Historically, AGN have been identified using several methods such as colour selection (Stern et al. 2005; Kochanek et al. 2012), spectroscopic signatures (Morse, Raymond & Wilson 1996; Veilleux 2002), and variability (Sarajedini, Koo & Klesman 2009; MacLeod et al. 2010). Studies using optical data found that identification is biased against more obscured or faint sources (Richards et al. 2002). To address these issues, other studies have used X-ray selection (Alexander et al. 2003) and mid-IR colour selection (Lacy et al. 2004; Stern et al. 2005; Park et al. 2008), but they each have their own biases. For instance, the X-ray selection is biased against heavily obscured AGN, while colour-selection is biased against AGN that are masked by the intrinsic luminosity of the host galaxy (Kocevski et al. 2015, hereafter K15). AGN have long been identified as variable sources with the variability most likely originating from instabilities in the accretion disc or temperature fluctuations (e.g. Ruan et al. 2014). For sources obscured by dust in the vicinity of the AGN, perhaps in the form of a dusty torus as envisioned in the unified model, the variable UV photons may then be reprocessed and produce variability in the mid-IR. Thus, mid-IR variability may be more sensitive than other selection techniques to obscured, Compton-thick AGN. Recent results from the NOAO Deep Wide Field Survey and Spitzer Deep Wide Field Survey (Kozłowski et al. 2010, 2016) have revealed that as many as 1.1 per cent of galaxies are significant variables in the Spitzer bands at 3.6 and 4.5 μm. Their study covered a 9 deg2 field with short (∼90 s) exposures in five epochs spanning ∼10 yr. The mid-IR variable light curves had lower amplitudes at short time-scales compared to optical AGN which can be interpreted as the accretion disc or dust torus smoothing out the short time-scale variations. An interesting and unsolved question concerning AGN is how accretion is initiated on to the central SMBH. Hopkins, Kocevski & Bundy (2014) suggested that AGN accretion can be triggered by mergers. However, surveys to quantify merger signatures in AGN samples have yielded mixed results. Kocevski et al. (2012) showed that host galaxies of a sample of X-ray detected AGN are no more likely to show morphological disturbances than a mass-matched sample of control galaxies. In a follow-up study, K15 found that the more obscured X-ray detected AGN have higher probability of merger signatures than the unobscured ones suggesting that dust may hide the AGN shortly after the fuelling is triggered by a merger. Here, we address the issue of obscuration and its impact on detecting merger-triggered accretion by identifying AGN in deep extragalactic surveys by their variability in the mid-IR. The data used in our study are part of the Spitzer Extended Deep Survey (SEDS; Ashby et al. 2013) and the Spitzer-Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (S-CANDELS; Ashby et al. 2015), two Exploration programmes completed during Spitzer's warm mission. The multi-epoch, deep images allow us to probe variability on time-scales from months to several years using 3.6 and 4.5 μm images of five extragalactic fields covering ∼0.9 deg2 of the sky. Our survey extends ∼3 mag deeper than the SDWFS field from Kozłowski et al. (2016) and in some regions includes a larger number of epochs for variability analysis. We perform morphological classification for the subset of variables that lie within the near-IR CANDELS survey (Grogin et al. 2011; Koekemoer et al. 2011) to look for evidence of merger-triggered accretion among the AGN host galaxies. In this paper, Section 2 describes the data sample, Section 3 outlines the steps to identify variable galaxies, Section 4 describes the selection of the control sample for morphology classification. The classification process is described in Section 5. Section 6 discusses the results and the following section lists our conclusions. 2 DATA DESCRIPTION The Spitzer IRAC data used for this study are fully described by Ashby et al. (2013, 2015). The survey covered the 30 × 30 arcmin2 Extended GOODS-South (a.k.a. the GEMS field, hereafter ECDFS), 30 × 30 arcmin2 Extended GOODS-North (HDFN), 50 × 50 arcmin2 UKIDSS Ultra-Deep Survey (UDS), a 10 × 60 arcmin2 region within the Extended Groth Strip (EGS), and a 10 × 60 arcmin2 strip within the UltraVista deep survey of the larger COSMOS field. Three epochs of SEDS observations, each consisting of up to 4 h integration time per pointing, were obtained for these fields in 2010 and 2011. However, all of the fields had previous Spitzer observations, some dating back to 2003, providing a total of four to 10 epochs of observations separated by several months to years. Table 1 lists all epochs of data obtained for each SEDS field used in this study. Table 1. Photometric depth of data. Field  PIDa  Date  Source count  Median magnitudeb  Number of  Number of  Number of      yyyy-mm-dd  [3.6]  [4.5]  [3.6] (AB)  [4.5] (AB)  detected sources  analysed sourcesc  variable sourcesd  COSMOS  61043  2010/01/30  20334  19776  23.282  22.225  40635  13673 (34%)  107 (0.8%)    61043  2010/06/18  20857  19633  23.235  23.245          61043  2011/01/31  20049  19341  21.689  21.739          80057  2012/02/11  5812  5249  23.307  23.219          80057  2012/07/03  5674  5804  22.761  23.128        ECDFS  194  2004/02/08  4058  3229  22.874  22.191  76792  18936 (25%)  163 (0.9%)    194  2004/08/13  4399  3125  22.579  22.379          20708  2005/08/23  23417  17413  22.562  22.218          20708  2006/02/09  22203  18161  22.527  22.174          60022  2010/09/24  31680  28366  23.510  23.060          60022  2011/03/28  30676  28557  23.247  23.340          60022  2011/10/15  31648  28831  23.354  23.308          70204  2011/03/21  2838  2778  23.407  23.597          80217  2011/09/27  2085  2012  23.232  23.372        EGS  0008  2003/12/25  30070  25195  23.000  22.928  75217  21471 (29%)  305 (1.4%)    0008  2004/07/01  29476  24459  22.923  23.000          0008  2006/03/28  5845  4537  22.741  22.214          41023  2008/01/25  10798  8266  22.319  21.966          41023  2008/07/23  11181  8821  22.512  21.944          61042  2010/02/06  27856  25803  22.957  22.972          61042  2010/08/05  29047  26028  22.883  23.078          61042  2011/02/10  28085  26155  23.084  22.999          80216  2011/08/19  5181  4626  23.464  23.043          80216  2012/02/23  5711  5213  23.553  22.817          80216  2012/08/31  5004  4658  23.403  22.853        HDFN  00169  2004/05/16  4402  3177  22.640  22.480  57354  12450 (22%)  112 (0.9%)    00169  2004/11/17  4114  3413  22.806  22.393          00169  2005/11/25  1144  1026  22.485  22.597          20218  2005/12/09  4820  2759  22.127  21.805          20218  2006/06/02  3785  3208  22.395  21.932          61040  2010/05/27  23750  22815  22.951  22.893          61040  2011/02/28  24324  20984  22.734  23.142          61040  2011/06/02  24706  22772  22.853  22.914        UDS  40021  2008/01/28  32351  26772  23.132  22.766  67181  21140 (31%)  186 (0.9%)    61041  2009/09/21  32469  29639  22.984  23.373          61041  2010/02/25  31945  30597  23.207  21.539          61041  2010/09/23  32907  29842  23.180  23.294          80218  2012/03/11  4807  4459  22.565  23.422          80218  2012/10/13  4752  4160  21.668  23.162          80218  2013/03/16  3610  3402  23.216  22.790        Field  PIDa  Date  Source count  Median magnitudeb  Number of  Number of  Number of      yyyy-mm-dd  [3.6]  [4.5]  [3.6] (AB)  [4.5] (AB)  detected sources  analysed sourcesc  variable sourcesd  COSMOS  61043  2010/01/30  20334  19776  23.282  22.225  40635  13673 (34%)  107 (0.8%)    61043  2010/06/18  20857  19633  23.235  23.245          61043  2011/01/31  20049  19341  21.689  21.739          80057  2012/02/11  5812  5249  23.307  23.219          80057  2012/07/03  5674  5804  22.761  23.128        ECDFS  194  2004/02/08  4058  3229  22.874  22.191  76792  18936 (25%)  163 (0.9%)    194  2004/08/13  4399  3125  22.579  22.379          20708  2005/08/23  23417  17413  22.562  22.218          20708  2006/02/09  22203  18161  22.527  22.174          60022  2010/09/24  31680  28366  23.510  23.060          60022  2011/03/28  30676  28557  23.247  23.340          60022  2011/10/15  31648  28831  23.354  23.308          70204  2011/03/21  2838  2778  23.407  23.597          80217  2011/09/27  2085  2012  23.232  23.372        EGS  0008  2003/12/25  30070  25195  23.000  22.928  75217  21471 (29%)  305 (1.4%)    0008  2004/07/01  29476  24459  22.923  23.000          0008  2006/03/28  5845  4537  22.741  22.214          41023  2008/01/25  10798  8266  22.319  21.966          41023  2008/07/23  11181  8821  22.512  21.944          61042  2010/02/06  27856  25803  22.957  22.972          61042  2010/08/05  29047  26028  22.883  23.078          61042  2011/02/10  28085  26155  23.084  22.999          80216  2011/08/19  5181  4626  23.464  23.043          80216  2012/02/23  5711  5213  23.553  22.817          80216  2012/08/31  5004  4658  23.403  22.853        HDFN  00169  2004/05/16  4402  3177  22.640  22.480  57354  12450 (22%)  112 (0.9%)    00169  2004/11/17  4114  3413  22.806  22.393          00169  2005/11/25  1144  1026  22.485  22.597          20218  2005/12/09  4820  2759  22.127  21.805          20218  2006/06/02  3785  3208  22.395  21.932          61040  2010/05/27  23750  22815  22.951  22.893          61040  2011/02/28  24324  20984  22.734  23.142          61040  2011/06/02  24706  22772  22.853  22.914        UDS  40021  2008/01/28  32351  26772  23.132  22.766  67181  21140 (31%)  186 (0.9%)    61041  2009/09/21  32469  29639  22.984  23.373          61041  2010/02/25  31945  30597  23.207  21.539          61041  2010/09/23  32907  29842  23.180  23.294          80218  2012/03/11  4807  4459  22.565  23.422          80218  2012/10/13  4752  4160  21.668  23.162          80218  2013/03/16  3610  3402  23.216  22.790        aSpitzer program identification number; bThe number of sources meeting our criteria of detection in both bands with at least 3 epochs and the percentage of total detected sources this represents; cThe number of variable sources in the field and the percentage of analysed sources this represents; dMedian magnitude of sources detected in this field. View Large The spectral energy distributions of AGN have a minimum at a rest wavelength of 1 μm (Elvis et al. 2009; Assef et al. 2010) with bluer light originating from the accretion disc and redder light coming from hot dust in the vicinity of the AGN (Urry & Padovani 1995; Asmus et al. 2014; Vazquez et al. 2015). The Spitzer images from our survey probe from just near the AGN minimum at z ∼ 3 into the dust-dominated regime at lower redshifts. Because our goal is to identify variability in the mid-IR, we make use of images produced from the individual epochs for each field. Identical reduction procedures were applied to all data to ensure a uniform data quality throughout the survey. Each individual epoch consists of the combined images, or mosaics, which were re-sampled to 0.6 arcsec. per pixel resolution and locked to the same world coordinate system. In this way, any particular (x, y) pixel coordinate corresponds to the same RA and Dec. in every epoch. We used SExtractor (Bertin & Arnouts 1996) to identify extended and point-like sources in each epoch of every field in our survey. Photometry was measured in both the 3.6 and 4.5 μm images using apertures ranging from 2.4 to 12 arcsec in diameter. A very low detection threshold (1σ) was adopted to improve source completeness in the individual epochs. Sources found in each epoch were matched by RA and DEC to within 0.5 pixels of a source in the SEDS catalogues (Ashby et al. 2013). This eliminated spurious detections found in the individual epochs due to bad pixels around bright sources or near the noisy edges of the mosaic. The photometry from each epoch was zero-point corrected to match the photometry in the SEDS catalogues. To determine the offsets, we used sources in the bright-to-mid-range between 18th and 21st magnitude which consisted of ∼1800 to ∼14000 sources depending on the image size of the epoch. Offsets were found up to ±0.03 mag but were generally closer to ±0.01 mag. These offsets were applied to each epoch independently, ensuring photometric consistency in each epoch. Based on the PSF for unresolved sources in the Spitzer images, we chose the 3.6 arcsec diameter aperture for our variability analysis. This aperture is large enough to include nuclear light from a varying active nucleus while excluding light from the non-varying, outer portions of the host galaxy and provides maximum sensitivity to flux variations originating from the AGN. Table 1 shows the number of sources found in each epoch of each observing field and the median depth of the observation. The differences in the numbers of sources are a reflection of the areas observed. Our analysis includes only those sources detected in both bands in three or more epochs in each field, which amounts to about 30 per cent of all sources detected. Thus, the total number of sources used in our variability analysis is 87670. 3 VARIABILITY SELECTION For each source in the catalogue detected in both bands with at least three epochs of data, we calculated the standard deviation v[X] as a function of the apparent magnitude as done by Kozłowski et al. (2010),   \begin{eqnarray} \ v[X] = \bigg [ \frac{1}{N-1} \sum _{j=1}^{N}(m[X]_j - \langle m[X] \rangle )^2\bigg ] ^\frac{1}{2}, \end{eqnarray} (1)where N is the number of epochs for an observed source, m[X]j is the magnitude of the source in the jth epoch, X is the 3.6 or 4.5 μm band, and 〈m[X]〉 is the average magnitude of the source in band X. Fig. 1 shows the v[3.6] values for all galaxies in the COSMOS field as a function of magnitude. Because the vast majority of galaxies in our survey are expected to be non-varying, the spread in the v[X] values represents the photometric noise which increases as a function of magnitude. We quantify this noise term by first calculating the median v[X] value, vm[X], in various magnitude bins along the x-axis. The width of the bins used for this calculation ranged from 1 mag at the bright end to 0.5 mag at the faint end to ensure an adequate number of sources per bin. The vm[X] values were then fitted with a line to produce a smooth distribution (red line in Fig. 1). Next, we calculate the dispersion, σ[X], as the standard deviation of v[X] values from vm[X] in each magnitude bin. These values were also fitted to produce a smooth distribution of σ[X] values as a function of magnitude (black line in Fig. 1). Making use of both the 3.6 and 4.5 μm data available for each source, we calculate the joint significance parameter, which quantifies the degree of variability of the source in both bands as   \begin{eqnarray} \sigma _{12} = \bigg [\bigg (\frac{v[3.6] - v_{\rm m}[3.6]}{\sigma [3.6]}\bigg )^2 + \bigg (\frac{v[4.5] - v_{\rm m}[4.5]}{\sigma [4.5]}\bigg )^2 \bigg ]^\frac{1}{2} \end{eqnarray} (2)where σ12 is essentially the significance level of the variability in each band added in quadrature. Fig. 2 shows the joint significance values for all sources with the thresholds of 3 and 4σ indicated. Figure 1. View largeDownload slide Standard deviation (equation 1) versus magnitude in the 3.6 μm band of the COSMOS field. The blue dots represent all the sources, the red line represents a fit to the vm [3.6] median values, and the black line represents the 1σ dispersion above the median line. This plot is representative of the other fields in our survey. Figure 1. View largeDownload slide Standard deviation (equation 1) versus magnitude in the 3.6 μm band of the COSMOS field. The blue dots represent all the sources, the red line represents a fit to the vm [3.6] median values, and the black line represents the 1σ dispersion above the median line. This plot is representative of the other fields in our survey. Figure 2. View largeDownload slide Joint significance (equation 2) versus magnitude in the 3.6 μm band for all sources in all five fields. The red and black dashed lines represent the 3 and 4σ thresholds. Figure 2. View largeDownload slide Joint significance (equation 2) versus magnitude in the 3.6 μm band for all sources in all five fields. The red and black dashed lines represent the 3 and 4σ thresholds. To quantify the coupling of variances in both the bands, we calculate a variability covariance C and Pearson's correlation factor r as   \begin{eqnarray} C = \frac{1}{N-1} \Sigma (m[3.6]_j - \langle m[3.6] \rangle ) (m[4.5]_j - \langle m[4.5] \rangle )) \end{eqnarray} (3)  \begin{eqnarray} r = \frac{C}{v[3.6]v[4.5]}. \end{eqnarray} (4)Fig. 3 is a histogram of the correlation factors for all 3 and 4σ variables. The correlation factor r is constrained in the closed interval [−1, 1], where −1 shows absolute anti-correlation between the variability in the two bands and +1 shows complete correlation. Figure 3. View largeDownload slide Histogram of the correlation factor r (equation 4) for all sources with cut-offs as σ12 ≥ 3 (blue) and σ12 ≥ 4 (green). The red and black dashed lines represent the average r number of objects per bin at r < 0.8 for the 3 and 4σ sources, respectively. Figure 3. View largeDownload slide Histogram of the correlation factor r (equation 4) for all sources with cut-offs as σ12 ≥ 3 (blue) and σ12 ≥ 4 (green). The red and black dashed lines represent the average r number of objects per bin at r < 0.8 for the 3 and 4σ sources, respectively. To ensure a robust selection of true variables above the photometric noise and with highly correlated light curves, we set the criteria for variable galaxies as σ12 ≥ 3 and r ≥ 0.8. The selected galaxies were examined by eye, of which 16 per cent were classified as spurious detections due to artefacts like diffraction spikes from nearby foreground stars or close proximity of a source to the edge. Since variability could not be accurately determined for these sources, they were removed from the survey. Fig. 4 shows the distribution of [3.6] mag of all detected sources and all the variability-selected sources in all five fields. The variables span the entire magnitude range of the galaxy survey and extend about 3 mag deeper than the mid-IR variability survey in the SDWFS (Kozłowski et al. 2010, 2016). Figure 4. View largeDownload slide Histogram of magnitudes of all sources (blue line) and the variability-selected AGN (green) in all five fields. Variable numbers have been multiplied by 100 to be compared with all sources. Figure 4. View largeDownload slide Histogram of magnitudes of all sources (blue line) and the variability-selected AGN (green) in all five fields. Variable numbers have been multiplied by 100 to be compared with all sources. The results of this analysis are summarized in the last column of Table 1. The number of variables in each field are as follows: 107 (0.8 per cent) in COSMOS, 163 (0.9 per cent) in ECDFS, 305 (1.4 per cent) in EGS, 112 (0.9 per cent) in HDFN, and 186 (0.9 per cent) in UDS. This gives us a total of 873 variables or ∼ 1.0 per cent of analysed sources in our survey which is in agreement with the result from Kozłowski et al. (2010) where the AGN fraction was found to be ∼ 1.1 per cent. Although our aim is to identify AGN, other sources such as supernovae can be detected as variable objects. An optical variability survey by Falocco et al. (2015) identified 0.06 per cent of sources as SNe among ∼33 300 sources with their SNe selection criteria being r ≥ 23 and Nepoch ≥ 6. Given this statistic, the number of SNe in our galaxy survey of ∼87 000 sources should be about 52, which is ∼ 6 per cent of the variability-selected sources. However, the spectral energy distribution models of SNe from Smitka (2016) show that the luminosity of the SNe dips in the mid-IR, suggesting that the fraction of SNe among our mid-IR variability-selected galaxies has an upper limit of 6 per cent and is likely to be even lower than the percentage based on the Falocco et al. (2015) statistics. We estimate the number of spurious detections in our survey using the average number of sources with σ12 ≥ 3 but having correlation factor values r < 0.8. These are sources which have uncorrelated light curves and are unlikely to be true variables. The average number of sources found per bin having r < 0.8 is 84 (red dashed line in Fig. 3). Extrapolating this line to sources with r > 0.8 implies that about 15 per cent of the variables meeting our variability criteria are spurious. This is much smaller than the number of spurious sources found by Kozłowski et al. (2016). They found 34 to 47 per cent false positives among variables identified with the same criteria we use (σ12 > 3 and r > 0.8) among variables having four to five epochs of data. Our estimated percentage of false positives combined with the percentage of sources which may be SNe results in ∼ 20 per cent of our variables being spurious or non-AGN in nature. Therefore, we estimate ∼ 80 per cent of the mid-IR variability-selected sources are likely due to the presence of an AGN. It can be seen that the false positive rate cited by Kozłowski et al. (2016) is much higher than our estimated rate. But, the rate drops significantly in Kozłowski et al. (2016) when they consider presumably brighter sources with magnitudes measurable in all four Spitzer bands. These have only a 6 to 7 per cent false positive rate. Therefore, we could say that since our false positive rate falls between that of their two photometry groups, this may indicate that the parameters used for photometry of sources in our survey result in a sample somewhere between these two photometry groups. 4 COMPARISON TO MULTIWAVELENGTH DATA Smaller regions within the SEDS fields have extremely large amounts of multiwavelength data from X-ray to radio wavelengths. For the purposes of our study, we focus on the sub-regions of the fields with deep, HST imaging and well-sampled spectral energy distributions for determining masses, redshifts, and morphological classification of the host galaxies. The Rainbow data base (Pérez-González et al. 2008)1 is a thorough compilation of photometry and spectroscopy for several extragalactic fields which includes portions of the SEDS fields. Multiwavelength photometry is used to determine photometric redshifts as well as stellar mass. We searched for sources in the Rainbow data base within 1 arcsec of sources included in our survey. We cross-referenced both the mid-IR variability-selected sources and the non-varying sources and identified 7146 sources in COSMOS (28 of which were mid-IR variables), 1992 in ECDFS (13 variables), 5574 in EGS (18 variables), 2054 in HDFN (20 variables), and 5173 in the UDS (25 variables). We further required that the Rainbow sources are within the CANDELS F160W field of view. This is necessary for the morphology classification process (see Section 5). This reduced the number of sources in the ECDFS field to 6, EGS to 17, and HDFN to 19, with the number of sources in the other fields remaining the same. Thus, a total of 95 mid-IR variables were found in the Rainbow data base within the CANDELS field of view, representing 11 per cent of all mid-IR variable sources. A visual inspection of the spectral energy distribution fits to model templates indicate that the variable sources are well fitted by the templates and should yield accurate redshifts and masses for the galaxies. To produce a control sample for our morphological study of AGN host galaxies, we require a randomly selected sample of galaxies that are non-varying and therefore not likely to be AGN but whose physical parameters are similar to the AGN host galaxy parameters. In particular we use galaxy mass, luminosity, and redshift to constrain the control galaxy population. For each field, we used the upper and lower limits of the mass, redshift, and magnitude ranges to search for sources to constitute our control sample. From the pool of candidates, we chose three random, non-varying galaxies per variable source whose masses fall within the range Mvar/2 < Mcontrol < 2 * Mvar, where Mvar is the mass of the AGN host galaxy displaying variability. Of the 285 control galaxies initially selected from the Rainbow data base, 55 were discarded since they were not within the CANDELS image field of view. Fig. 5 shows the mass versus redshift for all mid-IR sources in our survey with available data in the Rainbow data base (grey points), mid-IR-selected variables (red points), and control galaxies (blue triangles). The distributions of the control and variable samples are well matched in the redshift range between z = 0.25 and 3 and in the mass range between log(mass [M⊙]) = 8.5 and 11.5. This is further displayed in the histograms of these galaxy samples shown in Figs 6 and 7. The sharp decline in AGN at redshifts beyond z = 3 may be due to the combination of several effects. At these high redshifts, we probe the part of the AGN SED that becomes more strongly dominated by the host galaxy, making us less sensitive to the AGN variability. Additionally, the decrease in surface brightness at high redshift leads to greater incompleteness in the Rainbow data base. We also find few AGN hosts among the low-mass tail of the galaxy distribution. Since AGN are generally more common among massive galaxies, it is not too surprising that we find very few among the lowest mass galaxies. Figure 5. View largeDownload slide Redshifts versus masses of the galaxies (for which data were available from the RAINBOW data base) in our survey (grey dots), the mid-IR-selected AGN host galaxies (red circles), and the control sample galaxies (blue triangles). Note: The three outlier points lie outside the CANDELS fields and thus have no control galaxies nearby. Figure 5. View largeDownload slide Redshifts versus masses of the galaxies (for which data were available from the RAINBOW data base) in our survey (grey dots), the mid-IR-selected AGN host galaxies (red circles), and the control sample galaxies (blue triangles). Note: The three outlier points lie outside the CANDELS fields and thus have no control galaxies nearby. Figure 6. View largeDownload slide Normalized histogram showing the masses of all galaxies (for which data were available from the RAINBOW data base) in our survey (blue line), the mid-IR-selected AGN host galaxies (green shaded), and the control sample galaxies (red lines). Figure 6. View largeDownload slide Normalized histogram showing the masses of all galaxies (for which data were available from the RAINBOW data base) in our survey (blue line), the mid-IR-selected AGN host galaxies (green shaded), and the control sample galaxies (red lines). Figure 7. View largeDownload slide Normalized histogram showing the redshift distribution of all galaxies (for which data were available from the RAINBOW data base) in our survey (blue line), the mid-IR-selected AGN host galaxies (green shaded), and the control sample galaxies (red lines). Figure 7. View largeDownload slide Normalized histogram showing the redshift distribution of all galaxies (for which data were available from the RAINBOW data base) in our survey (blue line), the mid-IR-selected AGN host galaxies (green shaded), and the control sample galaxies (red lines). 5 MORPHOLOGY CLASSIFICATION The aim of our morphology study is to look for evidence of mergers or interactions by examining the physical features of the AGN host galaxies. If AGN accretion is indeed triggered by mergers or interactions, we should observe a higher fraction of morphologically disturbed galaxies among our variability-selected sample compared to the control sample. We visually classify 30 × 30 pixel (0.9 × 0.9 arcsec2) thumbnails of the AGN hosts extracted from the F160W (1600nm) images, available in the CANDELS (Grogin et al. 2011) Public Access data base. The classification was done using the criteria described in more detail in Kocevski et al. (2012). First, the general morphology was classified as disc, spheroid, irregular, or point-like. Secondly, the degree of disturbance was classified as Merger/Interacting (highly disturbed with multiple nuclei or two distinct galaxies with interacting features such as tidal arms), Distorted/Asymmetric (single asymmetric or distorted galaxies with no visible interacting companion), Close Pair (near-neighbour pair with both the galaxies in a single frame), Double Nuclei (multiple nuclei in a single system) or Undisturbed (none of the above). Fig. 8 shows representative galaxies from our survey in the various morphological classes. Figure 8. View largeDownload slide Sample galaxies belonging to each of the classification groups employed in our analysis. The size of each thumbnail is 30 × 30 pixels or 0.9 × 0.9 arcsec2. Figure 8. View largeDownload slide Sample galaxies belonging to each of the classification groups employed in our analysis. The size of each thumbnail is 30 × 30 pixels or 0.9 × 0.9 arcsec2. We performed a blind inspection, which means that the images of the variable and control galaxies within each field were mixed to avoid any biases during the classification process of the galaxies. The classification was performed by two of the authors of this paper individually (M. Polimera and V. Sarajedini), and the results were merged. For ∼ 30 per cent of the galaxies, each reviewer had classified them differently and these galaxies were reinspected until a consensus could be reached on the proper classification. We chose to employ this type of visual inspection to resolve the subtle, low surface brightness features which are easy to miss using modern automated classification techniques. However, we hope to use these classifications as training sets for a machine-learning-based classifier in the future. 6 RESULTS Figs 9 and 10 show the visual morphology classification results tabulated in Table 2. The 1σ error bars are calculated as confidence intervals for a binomial population as   \begin{eqnarray} {\rm Error} = z_{1 - \alpha /2} \sqrt{\frac{p (1-p)}{n}}, \end{eqnarray} (5)where p is the fraction of variable galaxies in the each disturbance category and n is the total number of galaxies (e.g. Cameron 2011). We estimate the 1σ error where the confidence level is c = 0.683, so the variate value from the standard normal distribution z1−α/2 = 1 (α = 1 − c). Figure 9. View largeDownload slide Fraction of variable galaxies (blue diamonds) and control galaxies (red squares) in each of the major morphology classes. Figure 9. View largeDownload slide Fraction of variable galaxies (blue diamonds) and control galaxies (red squares) in each of the major morphology classes. Figure 10. View largeDownload slide Fraction of variable galaxies (blue diamonds) and control galaxies (red squares) in different disturbance classes. Figure 10. View largeDownload slide Fraction of variable galaxies (blue diamonds) and control galaxies (red squares) in different disturbance classes. Table 2. Disturbance classification. Galaxy type  Disturbance class  COSMOS  UDS  EGS  HDFN  ECDFS  Total  Variable  Merg/Int  6 (21 ± 8 per cent)  5 (21 ± 8 per cent)  2 (11 ± 7 per cent)  2 (11 ± 7 per cent)  3 (50 ± 20 per cent)  18 (19 ± 4 per cent)    Dist/Asy  10 (35 ± 9 per cent)  12 (50±10 per cent)  12 (70 ± 11 per cent)  5 (26 ± 10 per cent)  1 (17 ± 15 per cent)  40 (42 ± 5 per cent)    Undisturbed  12 (42 ± 9 per cent)  8 (33 ± 9 per cent)  3 (17 ± 9 per cent)  12 (63 ± 11 per cent)  2 (33 ± 19 per cent)  37 (39 ± 5 per cent)    Total  28  25  17  19  6  95  Control  Merg/Int  10 (22 ± 6 per cent)  17 (26 ± 5 per cent)  12 (25 ± 6 per cent)  11 (20 ± 5 per cent)  7 (35 ± 10 per cent)  57 (25 ± 3 per cent)    Dist/Asy  17 (37 ± 7 per cent)  22 (35 ± 6 per cent)  13 (27 ± 6 per cent)  19 (34 ± 6 per cent)  3 (15 ± 7 per cent)  74 (32 ± 3 per cent)    Undisturbed  17 (37 ± 7 per cent)  25 (39 ± 6 per cent)  22 (46 ± 7 per cent)  25 (45 ± 7 per cent)  10 (50 ± 11 per cent)  99 (43 ± 3 per cent)    Total  45  63  47  55  20  230  Galaxy type  Disturbance class  COSMOS  UDS  EGS  HDFN  ECDFS  Total  Variable  Merg/Int  6 (21 ± 8 per cent)  5 (21 ± 8 per cent)  2 (11 ± 7 per cent)  2 (11 ± 7 per cent)  3 (50 ± 20 per cent)  18 (19 ± 4 per cent)    Dist/Asy  10 (35 ± 9 per cent)  12 (50±10 per cent)  12 (70 ± 11 per cent)  5 (26 ± 10 per cent)  1 (17 ± 15 per cent)  40 (42 ± 5 per cent)    Undisturbed  12 (42 ± 9 per cent)  8 (33 ± 9 per cent)  3 (17 ± 9 per cent)  12 (63 ± 11 per cent)  2 (33 ± 19 per cent)  37 (39 ± 5 per cent)    Total  28  25  17  19  6  95  Control  Merg/Int  10 (22 ± 6 per cent)  17 (26 ± 5 per cent)  12 (25 ± 6 per cent)  11 (20 ± 5 per cent)  7 (35 ± 10 per cent)  57 (25 ± 3 per cent)    Dist/Asy  17 (37 ± 7 per cent)  22 (35 ± 6 per cent)  13 (27 ± 6 per cent)  19 (34 ± 6 per cent)  3 (15 ± 7 per cent)  74 (32 ± 3 per cent)    Undisturbed  17 (37 ± 7 per cent)  25 (39 ± 6 per cent)  22 (46 ± 7 per cent)  25 (45 ± 7 per cent)  10 (50 ± 11 per cent)  99 (43 ± 3 per cent)    Total  45  63  47  55  20  230  * The galaxies used to calculate these percentages are only those that are present in the RAINBOW data base for which the visual morphology classification described in Section 5 has been performed. View Large Fig. 9 shows no significant differences between the AGN host galaxy classifications and those for the control sample in terms of galaxy classified as disc, spheroidal, irregular, or point-like. We also find no difference in the percentage of galaxies in a close pair. We find a similar fraction of AGN hosts in galaxies with a discernible disc (40 ± 10 per cent) as did Kocevski et al. (2012) for an X-ray-selected sample (51.4 ± 5.8 per cent). About 30 ± 10 per cent of the mid-IR variables are found in pure elliptical (spheroidal) hosts. This is also quite similar to that found by Kocevski et al. (2012) for X-ray-selected AGN hosts (27.8 ± 5 per cent). In Fig. 10, we compare the percentage of galaxies displaying signs of mergers/interactions and distortion/asymmetry using the same symbols as Fig. 9. We find that AGN hosts and control galaxies have the same fraction of merger/interaction hosts within the uncertainties (19 ± 5 to 23 ± 3 per cent). However, AGN hosts have a slightly higher tendency to reside in galaxies that are either distorted or asymmetric (42 ± 5 per cent) compared to the control galaxy population (32 ± 3 per cent). This result is significant at the 1.7σ level. Dividing our sample into high- and low-mass hosts and control galaxies also produced similar results. As previously mentioned, Kocevski et al. (2012) studied 72 X-ray-selected AGN hosts in the GOODS-S field using the same CANDELS NIR images analysed in this study. However, they found no statistical difference between AGN hosts and the control galaxies having strongly disturbed (merger/interaction) or weakly disturbed (distorted/asymmetric) morphologies. One possibility they suggest is that X-ray-selected samples of AGN still miss the most obscured sources which could be hiding hosts displaying merger signatures. A follow-up study by K15 explored this issue by determining the level of obscuration in the X-ray-selected AGN using reflection-dominated X-ray spectroscopic analysis. They found that the more obscured X-ray detected AGN (i.e. those with the highest hydrogen column densities) had a higher fraction of morphologically disturbed host galaxies than less obscured X-ray detected AGN. Our results indicate that mid-IR variability-selected AGN samples have marginally significant levels of weakly disturbed host galaxy morphologies. This could be interpreted as an indication that our sample includes a significant number of obscured, Compton-thick AGN. In order to test this hypothesis, we cross-referenced our sample of mid-IR variables with X-ray identified AGN in the five survey fields [COSMOS: Elvis et al. (2009); ECDFS: Alexander et al. (2003) and Xue et al. (2011); EGS: Nandra et al. (2015); HDFN: Alexander et al. (2003); UDS: Ueda et al. (2008)]. We initially searched for X-ray sources within 1 arcsec of the position of each mid-IR variable and found that matched sources generally lie within a 0.5 arcsec radius. Setting this as our matching threshold, we found 22 of our mid-IR variables are also X-ray detected AGN (23 per cent). This means that 77 per cent of our sources, or 73 mid-IR variables, are not identified in deep X-ray surveys and are likely to be either spurious sources (expected among ∼ 20 per cent of our sample) or may be highly obscured AGN whose X-ray emission has been absorbed. If these represent the most obscured sources, we may expect to find their morphologies to be the most disturbed. In Fig. 11, we compare the fraction of host galaxies of different morphological classifications for the mid-IR variables that are also identified in X-ray surveys (green diamonds) to the mid-IR variable sources without an X-ray counterpart (black diamonds). We see that while the two samples have the same, low fraction of highly disturbed host galaxies, a much higher fraction of mid-IR-only sources have weakly disturbed hosts (48 ± 5 per cent) than those with X-ray counterparts (23 ± 8 per cent). This difference is significant at the 3σ level. When we compare these percentages with those of the most highly obscured, Compton-thick X-ray sources from K15, we find that they agree favourably. We find that 22 ± 5 per cent of mid-IR-only-selected AGN have highly disturbed host morphologies compared to $$21.5^{+4.2}_{-3.3} \, {\rm per \, cent}$$ of the most obscured X-ray sources in K15. Furthermore, 40 ± 5 per cent of mid-IR-only-selected AGN have weakly disturbed hosts compared to $$43.0^{+4.6}_{-4.4} \, {\rm per \, cent}$$ in K15. This further supports the claim that AGN selected via mid-IR variability are likely to contain a larger fraction of obscured AGN, possibly more obscured than the most absorbed X-ray-selected sources, revealing a higher fraction of asymmetric hosts and highlighting the unique nature of this selection technique. Figure 11. View largeDownload slide Fraction of mid-IR variable galaxies with an X-ray counterpart (green diamonds) and those with no X-ray counterpart detected within 0.5 arcsec (black diamonds) in different disturbance classes. Figure 11. View largeDownload slide Fraction of mid-IR variable galaxies with an X-ray counterpart (green diamonds) and those with no X-ray counterpart detected within 0.5 arcsec (black diamonds) in different disturbance classes. 7 DISCUSSION It has been speculated that mergers tidally induce non-axis-symmetric modes in the merging galaxy and these modes sweep the interstellar gaseous medium into a disc with a radius of a few parsec around the central SMBH (Shlosman, Frank & Begelman 1989; Shlosman, Begelman & Frank 1990). The disc then is accreted on to the SMBH. Simulations from Barnes & Hernquist (1992) have shown that these discs are more likely to form in a major head-on merger scenario. Given such a theory, we may expect to see the AGN hosts displaying a higher fraction of major mergers as compared to the control group. Our results, however, do not show a significant fraction of strong interaction signatures among the mid-IR variability-selected AGN hosts. We do find a marginally significant increase in the presence of weak interaction signatures in the host galaxies when compared to a control sample and the significance increases when we remove X-ray-selected AGN and isolate mid-IR variability-selected AGN. One explanation is that during the entire process of a galaxy merger, the galaxies spend more time in stages where they appear to be weakly disturbed versus having a highly disturbed morphology. Based on recent simulations by Capelo et al. (2015), the time a galaxy spends in the weakly disturbed phases is about 7.9 Gyr, but the system shows a highly disturbed morphology only for about 4.6 Gyr. This makes it less likely to ‘catch’ the galaxy in the highly disturbed phase because their signatures dissipate too fast and we see less dramatic distortions in the hosts. Another possibility is that the dominant AGN triggering mechanism is not major mergers but that minor mergers are more important for AGN to begin accretion as the perturbation is sufficient for the orbits of material in the central part of the galaxy to get randomized and disturbed just enough to infall into the central SMBH (Hernquist & Mihos 1995; Menci et al. 2014). Hernquist & Mihos (1995) also point out that the occurrences of major mergers are more unusual as compared to minor mergers/interactions. Yet another possibility is that the low fraction of major merger signatures seen in our AGN hosts is due to heavy obscuration which would absorb the variable UV photons from the accretion disc and weaken the mid-IR variable flux necessary for our selection method. As shown in K15, there is a significant increase in the fraction of AGN host galaxy morphology disturbance as a function of obscuration, with the Compton-thick AGN hosts showing the highest fraction of disturbed morphologies compared to unobscured hosts. Our study found that the host galaxies of mid-IR variables have the same fraction of disturbed morphologies as the most obscured X-ray-selected sources. It is possible that higher levels of disturbance are hidden by extreme obscuration such that neither X-ray nor mid-IR variability is able to detect the AGN nature of the source. 8 CONCLUSION The results of our study are summarized as follows. 1 per cent of all galaxies with [3.6] < 24.5 are significantly variable in the mid-IR Spitzer/IRAC mosaics with 80 per cent likelihood of being AGN. The AGN host galaxies show a marginally significant trend (1.7σ significance) towards higher fractions of Disturbed/Asymmetric morphologies (42 ± 5 per cent) as compared to the control sample (32 ± 3 per cent), whereas Merger/Interaction fractions are not statistically different for the AGN and control galaxies. 23 per cent of mid-IR variables are also identified in X-ray surveys. The mid-IR variables without an X-ray counterpart show a higher fraction of weakly disturbed hosts than those also identified with X-ray emission. This suggests that mid-IR variability selects a unique population of obscured AGN that would likely be missed using other selection techniques. Acknowledgements This work is based on observations made with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA. This work has also made use of the Rainbow Cosmological Surveys data base, which is operated by the Universidad Complutense de Madrid (UCM), partnered with the University of California Observatories at Santa Cruz (UCO/Lick,UCSC) and used observations taken by the CANDELS Multi-Cycle Treasury Program with the NASA/ESA HST, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555. Footnotes 1 Website: http://rainbow.fis.ucm.es REFERENCES Alexander D. M. et al.  , 2003, AJ , 126, 539 CrossRef Search ADS   Ashby M. L. N. et al.  , 2013, ApJ , 769, 80 CrossRef Search ADS   Ashby M. L. N. et al.  , 2015, ApJS , 218, 33 CrossRef Search ADS   Asmus D., Hönig S. F., Gandhi P., Smette A., Duschl W. J., 2014, MNRAS , 439, 1648 CrossRef Search ADS   Assef R. J. et al.  , 2010, ApJ , 713, 970 CrossRef Search ADS   Barnes J. E., Hernquist L., 1992, ARA&A , 30, 705 CrossRef Search ADS   Bertin E., Arnouts S., 1996, A&AS , 117, 393 CrossRef Search ADS   Cameron E., 2011, PASA , 28, 128 CrossRef Search ADS   Capelo P. R., Volonteri M., Dotti M., Bellovary J. M., Mayer L., Governato F., 2015, MNRAS , 447, 2123 CrossRef Search ADS   Elvis M. et al.  , 2009, ApJS , 184, 158 CrossRef Search ADS   Falocco S. et al.  , 2015, A&A , 579, A115 CrossRef Search ADS   Grogin N. A. et al.  , 2011, ApJS , 197, 35 CrossRef Search ADS   Hernquist L., Mihos J. C., 1995, ApJ , 448, 41 CrossRef Search ADS   Hopkins P. F., Kocevski D. D., Bundy K., 2014, MNRAS , 445, 823 CrossRef Search ADS   Kocevski D. D. et al.  , 2012, ApJ , 744, 148 CrossRef Search ADS   Kocevski D. D. et al.  , 2015, ApJ , 814, 104 (K15) CrossRef Search ADS   Kochanek C. S. et al.  , 2012, ApJS , 200, 8 CrossRef Search ADS   Koekemoer A. M. et al.  , 2011, ApJS , 197, 36 CrossRef Search ADS   Kozłowski S. et al.  , 2010, ApJ , 716, 530 CrossRef Search ADS   Kozłowski S., Kochanek C. S., Ashby M. L. N., Assef R. J., Brodwin M., Eisenhardt P. R., Jannuzi B. T., Stern D., 2016, ApJ , 817, 119 CrossRef Search ADS   Lacy M. et al.  , 2004, ApJS , 154, 166 CrossRef Search ADS   Lynden-Bell D., 1969, Nature , 223, 690 CrossRef Search ADS   MacLeod C. L. et al.  , 2010, ApJ , 721, 1014 CrossRef Search ADS   Menci N., Gatti M., Fiore F., Lamastra A., 2014, A&A , 569, A37 CrossRef Search ADS   Morse J. A., Raymond J. C., Wilson A. S., 1996, PASP , 108, 426 CrossRef Search ADS   Nandra K. et al.  , 2015, ApJS , 220, 10 CrossRef Search ADS   Park S. Q. et al.  , 2008, ApJ , 678, 744 CrossRef Search ADS   Pérez-González P. G. et al.  , 2008, ApJ , 675, 234 CrossRef Search ADS   Richards G. T. et al.  , 2002, AJ , 123, 2945 CrossRef Search ADS   Ruan J. J., Anderson S. F., Dexter J., Agol E., 2014, ApJ , 783, 105 CrossRef Search ADS   Sarajedini V., Koo D., Klesman A., 2009, Am. Astron. Soc. Meeting Abst. #213 . 238 Shlosman I., Frank J., Begelman M. C., 1989, Nature , 338, 45 CrossRef Search ADS   Shlosman I., Begelman M. C., Frank J., 1990, Nature , 345, 679 CrossRef Search ADS   Smitka M. T., 2016, PhD thesis , Texas A&M University Stern D. et al.  , 2005, ApJ , 631, 163 CrossRef Search ADS   Ueda Y. et al.  , 2008, ApJS , 179, 124 CrossRef Search ADS   Urry C. M., Padovani P., 1995, PASP , 107, 803 CrossRef Search ADS   Vazquez B. et al.  , 2015, ApJ , 801, 127 CrossRef Search ADS   Veilleux S., 2002, in Green R. F., Khachikian E. Y., Sanders D. B., eds, ASP Conf. Ser. Vol. 284, IAU Colloq. 184: AGN Surveys . Astron. Soc. Pac., San Francisco, p. 111 (astro-ph/0201118) Xue Y. Q. et al.  , 2011, ApJS , 195, 10 CrossRef Search ADS   © 2018 The Author(s) Published by Oxford University Press on behalf of the Royal Astronomical Society http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Monthly Notices of the Royal Astronomical Society Oxford University Press

# Morphologies of mid-IR variability-selected AGN host galaxies

, Volume 476 (1) – May 1, 2018
9 pages

/lp/ou_press/morphologies-of-mid-ir-variability-selected-agn-host-galaxies-5MIQQag6zM
Publisher
The Royal Astronomical Society
ISSN
0035-8711
eISSN
1365-2966
D.O.I.
10.1093/mnras/sty164
Publisher site
See Article on Publisher Site

### Abstract

Abstract We use multi-epoch 3.6 and 4.5 μm data from the Spitzer Extended Deep Survey (SEDS) to probe the AGN population among galaxies to redshifts ∼3 via their mid-IR variability. About 1 per cent of all galaxies in our survey contain varying nuclei, 80 per cent of which are likely to be AGN. Twenty-three per cent of mid-IR variables are also X-ray sources. The mid-IR variables have a slightly greater fraction of weakly disturbed morphologies compared to a control sample of normal galaxies. The increased fraction of weakly distorted hosts becomes more significant when we remove the X-ray emitting AGN, while the frequency of strongly disturbed hosts remains similar to the control galaxy sample. These results suggest that mid-IR variability identifies a unique population of obscured, Compton-thick AGN revealing elevated levels of weak distortion among their host galaxies. galaxies: active, galaxies: evolution, infrared: galaxies 1 INTRODUCTION Active galactic nuclei (AGN) are galaxies which accrete significant amounts of material on to their central super massive black holes (SMBHs) and have long been one of the most interesting phenomena in extragalactic astronomy. Questions relating to their formation, structure, and SMBH accretion have been contemplated since theories explaining the powering mechanism of AGN were developed (Lynden-Bell 1969). AGN may represent an evolutionary phase for many galaxies and could have a significant impact on their star formation. To understand the role played by AGN in the evolution of galaxies, it is necessary to identify them in galaxy surveys out to redshifts of ∼3 where significant AGN fuelling and bulge growth occur. Historically, AGN have been identified using several methods such as colour selection (Stern et al. 2005; Kochanek et al. 2012), spectroscopic signatures (Morse, Raymond & Wilson 1996; Veilleux 2002), and variability (Sarajedini, Koo & Klesman 2009; MacLeod et al. 2010). Studies using optical data found that identification is biased against more obscured or faint sources (Richards et al. 2002). To address these issues, other studies have used X-ray selection (Alexander et al. 2003) and mid-IR colour selection (Lacy et al. 2004; Stern et al. 2005; Park et al. 2008), but they each have their own biases. For instance, the X-ray selection is biased against heavily obscured AGN, while colour-selection is biased against AGN that are masked by the intrinsic luminosity of the host galaxy (Kocevski et al. 2015, hereafter K15). AGN have long been identified as variable sources with the variability most likely originating from instabilities in the accretion disc or temperature fluctuations (e.g. Ruan et al. 2014). For sources obscured by dust in the vicinity of the AGN, perhaps in the form of a dusty torus as envisioned in the unified model, the variable UV photons may then be reprocessed and produce variability in the mid-IR. Thus, mid-IR variability may be more sensitive than other selection techniques to obscured, Compton-thick AGN. Recent results from the NOAO Deep Wide Field Survey and Spitzer Deep Wide Field Survey (Kozłowski et al. 2010, 2016) have revealed that as many as 1.1 per cent of galaxies are significant variables in the Spitzer bands at 3.6 and 4.5 μm. Their study covered a 9 deg2 field with short (∼90 s) exposures in five epochs spanning ∼10 yr. The mid-IR variable light curves had lower amplitudes at short time-scales compared to optical AGN which can be interpreted as the accretion disc or dust torus smoothing out the short time-scale variations. An interesting and unsolved question concerning AGN is how accretion is initiated on to the central SMBH. Hopkins, Kocevski & Bundy (2014) suggested that AGN accretion can be triggered by mergers. However, surveys to quantify merger signatures in AGN samples have yielded mixed results. Kocevski et al. (2012) showed that host galaxies of a sample of X-ray detected AGN are no more likely to show morphological disturbances than a mass-matched sample of control galaxies. In a follow-up study, K15 found that the more obscured X-ray detected AGN have higher probability of merger signatures than the unobscured ones suggesting that dust may hide the AGN shortly after the fuelling is triggered by a merger. Here, we address the issue of obscuration and its impact on detecting merger-triggered accretion by identifying AGN in deep extragalactic surveys by their variability in the mid-IR. The data used in our study are part of the Spitzer Extended Deep Survey (SEDS; Ashby et al. 2013) and the Spitzer-Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (S-CANDELS; Ashby et al. 2015), two Exploration programmes completed during Spitzer's warm mission. The multi-epoch, deep images allow us to probe variability on time-scales from months to several years using 3.6 and 4.5 μm images of five extragalactic fields covering ∼0.9 deg2 of the sky. Our survey extends ∼3 mag deeper than the SDWFS field from Kozłowski et al. (2016) and in some regions includes a larger number of epochs for variability analysis. We perform morphological classification for the subset of variables that lie within the near-IR CANDELS survey (Grogin et al. 2011; Koekemoer et al. 2011) to look for evidence of merger-triggered accretion among the AGN host galaxies. In this paper, Section 2 describes the data sample, Section 3 outlines the steps to identify variable galaxies, Section 4 describes the selection of the control sample for morphology classification. The classification process is described in Section 5. Section 6 discusses the results and the following section lists our conclusions. 2 DATA DESCRIPTION The Spitzer IRAC data used for this study are fully described by Ashby et al. (2013, 2015). The survey covered the 30 × 30 arcmin2 Extended GOODS-South (a.k.a. the GEMS field, hereafter ECDFS), 30 × 30 arcmin2 Extended GOODS-North (HDFN), 50 × 50 arcmin2 UKIDSS Ultra-Deep Survey (UDS), a 10 × 60 arcmin2 region within the Extended Groth Strip (EGS), and a 10 × 60 arcmin2 strip within the UltraVista deep survey of the larger COSMOS field. Three epochs of SEDS observations, each consisting of up to 4 h integration time per pointing, were obtained for these fields in 2010 and 2011. However, all of the fields had previous Spitzer observations, some dating back to 2003, providing a total of four to 10 epochs of observations separated by several months to years. Table 1 lists all epochs of data obtained for each SEDS field used in this study. Table 1. Photometric depth of data. Field  PIDa  Date  Source count  Median magnitudeb  Number of  Number of  Number of      yyyy-mm-dd  [3.6]  [4.5]  [3.6] (AB)  [4.5] (AB)  detected sources  analysed sourcesc  variable sourcesd  COSMOS  61043  2010/01/30  20334  19776  23.282  22.225  40635  13673 (34%)  107 (0.8%)    61043  2010/06/18  20857  19633  23.235  23.245          61043  2011/01/31  20049  19341  21.689  21.739          80057  2012/02/11  5812  5249  23.307  23.219          80057  2012/07/03  5674  5804  22.761  23.128        ECDFS  194  2004/02/08  4058  3229  22.874  22.191  76792  18936 (25%)  163 (0.9%)    194  2004/08/13  4399  3125  22.579  22.379          20708  2005/08/23  23417  17413  22.562  22.218          20708  2006/02/09  22203  18161  22.527  22.174          60022  2010/09/24  31680  28366  23.510  23.060          60022  2011/03/28  30676  28557  23.247  23.340          60022  2011/10/15  31648  28831  23.354  23.308          70204  2011/03/21  2838  2778  23.407  23.597          80217  2011/09/27  2085  2012  23.232  23.372        EGS  0008  2003/12/25  30070  25195  23.000  22.928  75217  21471 (29%)  305 (1.4%)    0008  2004/07/01  29476  24459  22.923  23.000          0008  2006/03/28  5845  4537  22.741  22.214          41023  2008/01/25  10798  8266  22.319  21.966          41023  2008/07/23  11181  8821  22.512  21.944          61042  2010/02/06  27856  25803  22.957  22.972          61042  2010/08/05  29047  26028  22.883  23.078          61042  2011/02/10  28085  26155  23.084  22.999          80216  2011/08/19  5181  4626  23.464  23.043          80216  2012/02/23  5711  5213  23.553  22.817          80216  2012/08/31  5004  4658  23.403  22.853        HDFN  00169  2004/05/16  4402  3177  22.640  22.480  57354  12450 (22%)  112 (0.9%)    00169  2004/11/17  4114  3413  22.806  22.393          00169  2005/11/25  1144  1026  22.485  22.597          20218  2005/12/09  4820  2759  22.127  21.805          20218  2006/06/02  3785  3208  22.395  21.932          61040  2010/05/27  23750  22815  22.951  22.893          61040  2011/02/28  24324  20984  22.734  23.142          61040  2011/06/02  24706  22772  22.853  22.914        UDS  40021  2008/01/28  32351  26772  23.132  22.766  67181  21140 (31%)  186 (0.9%)    61041  2009/09/21  32469  29639  22.984  23.373          61041  2010/02/25  31945  30597  23.207  21.539          61041  2010/09/23  32907  29842  23.180  23.294          80218  2012/03/11  4807  4459  22.565  23.422          80218  2012/10/13  4752  4160  21.668  23.162          80218  2013/03/16  3610  3402  23.216  22.790        Field  PIDa  Date  Source count  Median magnitudeb  Number of  Number of  Number of      yyyy-mm-dd  [3.6]  [4.5]  [3.6] (AB)  [4.5] (AB)  detected sources  analysed sourcesc  variable sourcesd  COSMOS  61043  2010/01/30  20334  19776  23.282  22.225  40635  13673 (34%)  107 (0.8%)    61043  2010/06/18  20857  19633  23.235  23.245          61043  2011/01/31  20049  19341  21.689  21.739          80057  2012/02/11  5812  5249  23.307  23.219          80057  2012/07/03  5674  5804  22.761  23.128        ECDFS  194  2004/02/08  4058  3229  22.874  22.191  76792  18936 (25%)  163 (0.9%)    194  2004/08/13  4399  3125  22.579  22.379          20708  2005/08/23  23417  17413  22.562  22.218          20708  2006/02/09  22203  18161  22.527  22.174          60022  2010/09/24  31680  28366  23.510  23.060          60022  2011/03/28  30676  28557  23.247  23.340          60022  2011/10/15  31648  28831  23.354  23.308          70204  2011/03/21  2838  2778  23.407  23.597          80217  2011/09/27  2085  2012  23.232  23.372        EGS  0008  2003/12/25  30070  25195  23.000  22.928  75217  21471 (29%)  305 (1.4%)    0008  2004/07/01  29476  24459  22.923  23.000          0008  2006/03/28  5845  4537  22.741  22.214          41023  2008/01/25  10798  8266  22.319  21.966          41023  2008/07/23  11181  8821  22.512  21.944          61042  2010/02/06  27856  25803  22.957  22.972          61042  2010/08/05  29047  26028  22.883  23.078          61042  2011/02/10  28085  26155  23.084  22.999          80216  2011/08/19  5181  4626  23.464  23.043          80216  2012/02/23  5711  5213  23.553  22.817          80216  2012/08/31  5004  4658  23.403  22.853        HDFN  00169  2004/05/16  4402  3177  22.640  22.480  57354  12450 (22%)  112 (0.9%)    00169  2004/11/17  4114  3413  22.806  22.393          00169  2005/11/25  1144  1026  22.485  22.597          20218  2005/12/09  4820  2759  22.127  21.805          20218  2006/06/02  3785  3208  22.395  21.932          61040  2010/05/27  23750  22815  22.951  22.893          61040  2011/02/28  24324  20984  22.734  23.142          61040  2011/06/02  24706  22772  22.853  22.914        UDS  40021  2008/01/28  32351  26772  23.132  22.766  67181  21140 (31%)  186 (0.9%)    61041  2009/09/21  32469  29639  22.984  23.373          61041  2010/02/25  31945  30597  23.207  21.539          61041  2010/09/23  32907  29842  23.180  23.294          80218  2012/03/11  4807  4459  22.565  23.422          80218  2012/10/13  4752  4160  21.668  23.162          80218  2013/03/16  3610  3402  23.216  22.790        aSpitzer program identification number; bThe number of sources meeting our criteria of detection in both bands with at least 3 epochs and the percentage of total detected sources this represents; cThe number of variable sources in the field and the percentage of analysed sources this represents; dMedian magnitude of sources detected in this field. View Large The spectral energy distributions of AGN have a minimum at a rest wavelength of 1 μm (Elvis et al. 2009; Assef et al. 2010) with bluer light originating from the accretion disc and redder light coming from hot dust in the vicinity of the AGN (Urry & Padovani 1995; Asmus et al. 2014; Vazquez et al. 2015). The Spitzer images from our survey probe from just near the AGN minimum at z ∼ 3 into the dust-dominated regime at lower redshifts. Because our goal is to identify variability in the mid-IR, we make use of images produced from the individual epochs for each field. Identical reduction procedures were applied to all data to ensure a uniform data quality throughout the survey. Each individual epoch consists of the combined images, or mosaics, which were re-sampled to 0.6 arcsec. per pixel resolution and locked to the same world coordinate system. In this way, any particular (x, y) pixel coordinate corresponds to the same RA and Dec. in every epoch. We used SExtractor (Bertin & Arnouts 1996) to identify extended and point-like sources in each epoch of every field in our survey. Photometry was measured in both the 3.6 and 4.5 μm images using apertures ranging from 2.4 to 12 arcsec in diameter. A very low detection threshold (1σ) was adopted to improve source completeness in the individual epochs. Sources found in each epoch were matched by RA and DEC to within 0.5 pixels of a source in the SEDS catalogues (Ashby et al. 2013). This eliminated spurious detections found in the individual epochs due to bad pixels around bright sources or near the noisy edges of the mosaic. The photometry from each epoch was zero-point corrected to match the photometry in the SEDS catalogues. To determine the offsets, we used sources in the bright-to-mid-range between 18th and 21st magnitude which consisted of ∼1800 to ∼14000 sources depending on the image size of the epoch. Offsets were found up to ±0.03 mag but were generally closer to ±0.01 mag. These offsets were applied to each epoch independently, ensuring photometric consistency in each epoch. Based on the PSF for unresolved sources in the Spitzer images, we chose the 3.6 arcsec diameter aperture for our variability analysis. This aperture is large enough to include nuclear light from a varying active nucleus while excluding light from the non-varying, outer portions of the host galaxy and provides maximum sensitivity to flux variations originating from the AGN. Table 1 shows the number of sources found in each epoch of each observing field and the median depth of the observation. The differences in the numbers of sources are a reflection of the areas observed. Our analysis includes only those sources detected in both bands in three or more epochs in each field, which amounts to about 30 per cent of all sources detected. Thus, the total number of sources used in our variability analysis is 87670. 3 VARIABILITY SELECTION For each source in the catalogue detected in both bands with at least three epochs of data, we calculated the standard deviation v[X] as a function of the apparent magnitude as done by Kozłowski et al. (2010),   \begin{eqnarray} \ v[X] = \bigg [ \frac{1}{N-1} \sum _{j=1}^{N}(m[X]_j - \langle m[X] \rangle )^2\bigg ] ^\frac{1}{2}, \end{eqnarray} (1)where N is the number of epochs for an observed source, m[X]j is the magnitude of the source in the jth epoch, X is the 3.6 or 4.5 μm band, and 〈m[X]〉 is the average magnitude of the source in band X. Fig. 1 shows the v[3.6] values for all galaxies in the COSMOS field as a function of magnitude. Because the vast majority of galaxies in our survey are expected to be non-varying, the spread in the v[X] values represents the photometric noise which increases as a function of magnitude. We quantify this noise term by first calculating the median v[X] value, vm[X], in various magnitude bins along the x-axis. The width of the bins used for this calculation ranged from 1 mag at the bright end to 0.5 mag at the faint end to ensure an adequate number of sources per bin. The vm[X] values were then fitted with a line to produce a smooth distribution (red line in Fig. 1). Next, we calculate the dispersion, σ[X], as the standard deviation of v[X] values from vm[X] in each magnitude bin. These values were also fitted to produce a smooth distribution of σ[X] values as a function of magnitude (black line in Fig. 1). Making use of both the 3.6 and 4.5 μm data available for each source, we calculate the joint significance parameter, which quantifies the degree of variability of the source in both bands as   \begin{eqnarray} \sigma _{12} = \bigg [\bigg (\frac{v[3.6] - v_{\rm m}[3.6]}{\sigma [3.6]}\bigg )^2 + \bigg (\frac{v[4.5] - v_{\rm m}[4.5]}{\sigma [4.5]}\bigg )^2 \bigg ]^\frac{1}{2} \end{eqnarray} (2)where σ12 is essentially the significance level of the variability in each band added in quadrature. Fig. 2 shows the joint significance values for all sources with the thresholds of 3 and 4σ indicated. Figure 1. View largeDownload slide Standard deviation (equation 1) versus magnitude in the 3.6 μm band of the COSMOS field. The blue dots represent all the sources, the red line represents a fit to the vm [3.6] median values, and the black line represents the 1σ dispersion above the median line. This plot is representative of the other fields in our survey. Figure 1. View largeDownload slide Standard deviation (equation 1) versus magnitude in the 3.6 μm band of the COSMOS field. The blue dots represent all the sources, the red line represents a fit to the vm [3.6] median values, and the black line represents the 1σ dispersion above the median line. This plot is representative of the other fields in our survey. Figure 2. View largeDownload slide Joint significance (equation 2) versus magnitude in the 3.6 μm band for all sources in all five fields. The red and black dashed lines represent the 3 and 4σ thresholds. Figure 2. View largeDownload slide Joint significance (equation 2) versus magnitude in the 3.6 μm band for all sources in all five fields. The red and black dashed lines represent the 3 and 4σ thresholds. To quantify the coupling of variances in both the bands, we calculate a variability covariance C and Pearson's correlation factor r as   \begin{eqnarray} C = \frac{1}{N-1} \Sigma (m[3.6]_j - \langle m[3.6] \rangle ) (m[4.5]_j - \langle m[4.5] \rangle )) \end{eqnarray} (3)  \begin{eqnarray} r = \frac{C}{v[3.6]v[4.5]}. \end{eqnarray} (4)Fig. 3 is a histogram of the correlation factors for all 3 and 4σ variables. The correlation factor r is constrained in the closed interval [−1, 1], where −1 shows absolute anti-correlation between the variability in the two bands and +1 shows complete correlation. Figure 3. View largeDownload slide Histogram of the correlation factor r (equation 4) for all sources with cut-offs as σ12 ≥ 3 (blue) and σ12 ≥ 4 (green). The red and black dashed lines represent the average r number of objects per bin at r < 0.8 for the 3 and 4σ sources, respectively. Figure 3. View largeDownload slide Histogram of the correlation factor r (equation 4) for all sources with cut-offs as σ12 ≥ 3 (blue) and σ12 ≥ 4 (green). The red and black dashed lines represent the average r number of objects per bin at r < 0.8 for the 3 and 4σ sources, respectively. To ensure a robust selection of true variables above the photometric noise and with highly correlated light curves, we set the criteria for variable galaxies as σ12 ≥ 3 and r ≥ 0.8. The selected galaxies were examined by eye, of which 16 per cent were classified as spurious detections due to artefacts like diffraction spikes from nearby foreground stars or close proximity of a source to the edge. Since variability could not be accurately determined for these sources, they were removed from the survey. Fig. 4 shows the distribution of [3.6] mag of all detected sources and all the variability-selected sources in all five fields. The variables span the entire magnitude range of the galaxy survey and extend about 3 mag deeper than the mid-IR variability survey in the SDWFS (Kozłowski et al. 2010, 2016). Figure 4. View largeDownload slide Histogram of magnitudes of all sources (blue line) and the variability-selected AGN (green) in all five fields. Variable numbers have been multiplied by 100 to be compared with all sources. Figure 4. View largeDownload slide Histogram of magnitudes of all sources (blue line) and the variability-selected AGN (green) in all five fields. Variable numbers have been multiplied by 100 to be compared with all sources. The results of this analysis are summarized in the last column of Table 1. The number of variables in each field are as follows: 107 (0.8 per cent) in COSMOS, 163 (0.9 per cent) in ECDFS, 305 (1.4 per cent) in EGS, 112 (0.9 per cent) in HDFN, and 186 (0.9 per cent) in UDS. This gives us a total of 873 variables or ∼ 1.0 per cent of analysed sources in our survey which is in agreement with the result from Kozłowski et al. (2010) where the AGN fraction was found to be ∼ 1.1 per cent. Although our aim is to identify AGN, other sources such as supernovae can be detected as variable objects. An optical variability survey by Falocco et al. (2015) identified 0.06 per cent of sources as SNe among ∼33 300 sources with their SNe selection criteria being r ≥ 23 and Nepoch ≥ 6. Given this statistic, the number of SNe in our galaxy survey of ∼87 000 sources should be about 52, which is ∼ 6 per cent of the variability-selected sources. However, the spectral energy distribution models of SNe from Smitka (2016) show that the luminosity of the SNe dips in the mid-IR, suggesting that the fraction of SNe among our mid-IR variability-selected galaxies has an upper limit of 6 per cent and is likely to be even lower than the percentage based on the Falocco et al. (2015) statistics. We estimate the number of spurious detections in our survey using the average number of sources with σ12 ≥ 3 but having correlation factor values r < 0.8. These are sources which have uncorrelated light curves and are unlikely to be true variables. The average number of sources found per bin having r < 0.8 is 84 (red dashed line in Fig. 3). Extrapolating this line to sources with r > 0.8 implies that about 15 per cent of the variables meeting our variability criteria are spurious. This is much smaller than the number of spurious sources found by Kozłowski et al. (2016). They found 34 to 47 per cent false positives among variables identified with the same criteria we use (σ12 > 3 and r > 0.8) among variables having four to five epochs of data. Our estimated percentage of false positives combined with the percentage of sources which may be SNe results in ∼ 20 per cent of our variables being spurious or non-AGN in nature. Therefore, we estimate ∼ 80 per cent of the mid-IR variability-selected sources are likely due to the presence of an AGN. It can be seen that the false positive rate cited by Kozłowski et al. (2016) is much higher than our estimated rate. But, the rate drops significantly in Kozłowski et al. (2016) when they consider presumably brighter sources with magnitudes measurable in all four Spitzer bands. These have only a 6 to 7 per cent false positive rate. Therefore, we could say that since our false positive rate falls between that of their two photometry groups, this may indicate that the parameters used for photometry of sources in our survey result in a sample somewhere between these two photometry groups. 4 COMPARISON TO MULTIWAVELENGTH DATA Smaller regions within the SEDS fields have extremely large amounts of multiwavelength data from X-ray to radio wavelengths. For the purposes of our study, we focus on the sub-regions of the fields with deep, HST imaging and well-sampled spectral energy distributions for determining masses, redshifts, and morphological classification of the host galaxies. The Rainbow data base (Pérez-González et al. 2008)1 is a thorough compilation of photometry and spectroscopy for several extragalactic fields which includes portions of the SEDS fields. Multiwavelength photometry is used to determine photometric redshifts as well as stellar mass. We searched for sources in the Rainbow data base within 1 arcsec of sources included in our survey. We cross-referenced both the mid-IR variability-selected sources and the non-varying sources and identified 7146 sources in COSMOS (28 of which were mid-IR variables), 1992 in ECDFS (13 variables), 5574 in EGS (18 variables), 2054 in HDFN (20 variables), and 5173 in the UDS (25 variables). We further required that the Rainbow sources are within the CANDELS F160W field of view. This is necessary for the morphology classification process (see Section 5). This reduced the number of sources in the ECDFS field to 6, EGS to 17, and HDFN to 19, with the number of sources in the other fields remaining the same. Thus, a total of 95 mid-IR variables were found in the Rainbow data base within the CANDELS field of view, representing 11 per cent of all mid-IR variable sources. A visual inspection of the spectral energy distribution fits to model templates indicate that the variable sources are well fitted by the templates and should yield accurate redshifts and masses for the galaxies. To produce a control sample for our morphological study of AGN host galaxies, we require a randomly selected sample of galaxies that are non-varying and therefore not likely to be AGN but whose physical parameters are similar to the AGN host galaxy parameters. In particular we use galaxy mass, luminosity, and redshift to constrain the control galaxy population. For each field, we used the upper and lower limits of the mass, redshift, and magnitude ranges to search for sources to constitute our control sample. From the pool of candidates, we chose three random, non-varying galaxies per variable source whose masses fall within the range Mvar/2 < Mcontrol < 2 * Mvar, where Mvar is the mass of the AGN host galaxy displaying variability. Of the 285 control galaxies initially selected from the Rainbow data base, 55 were discarded since they were not within the CANDELS image field of view. Fig. 5 shows the mass versus redshift for all mid-IR sources in our survey with available data in the Rainbow data base (grey points), mid-IR-selected variables (red points), and control galaxies (blue triangles). The distributions of the control and variable samples are well matched in the redshift range between z = 0.25 and 3 and in the mass range between log(mass [M⊙]) = 8.5 and 11.5. This is further displayed in the histograms of these galaxy samples shown in Figs 6 and 7. The sharp decline in AGN at redshifts beyond z = 3 may be due to the combination of several effects. At these high redshifts, we probe the part of the AGN SED that becomes more strongly dominated by the host galaxy, making us less sensitive to the AGN variability. Additionally, the decrease in surface brightness at high redshift leads to greater incompleteness in the Rainbow data base. We also find few AGN hosts among the low-mass tail of the galaxy distribution. Since AGN are generally more common among massive galaxies, it is not too surprising that we find very few among the lowest mass galaxies. Figure 5. View largeDownload slide Redshifts versus masses of the galaxies (for which data were available from the RAINBOW data base) in our survey (grey dots), the mid-IR-selected AGN host galaxies (red circles), and the control sample galaxies (blue triangles). Note: The three outlier points lie outside the CANDELS fields and thus have no control galaxies nearby. Figure 5. View largeDownload slide Redshifts versus masses of the galaxies (for which data were available from the RAINBOW data base) in our survey (grey dots), the mid-IR-selected AGN host galaxies (red circles), and the control sample galaxies (blue triangles). Note: The three outlier points lie outside the CANDELS fields and thus have no control galaxies nearby. Figure 6. View largeDownload slide Normalized histogram showing the masses of all galaxies (for which data were available from the RAINBOW data base) in our survey (blue line), the mid-IR-selected AGN host galaxies (green shaded), and the control sample galaxies (red lines). Figure 6. View largeDownload slide Normalized histogram showing the masses of all galaxies (for which data were available from the RAINBOW data base) in our survey (blue line), the mid-IR-selected AGN host galaxies (green shaded), and the control sample galaxies (red lines). Figure 7. View largeDownload slide Normalized histogram showing the redshift distribution of all galaxies (for which data were available from the RAINBOW data base) in our survey (blue line), the mid-IR-selected AGN host galaxies (green shaded), and the control sample galaxies (red lines). Figure 7. View largeDownload slide Normalized histogram showing the redshift distribution of all galaxies (for which data were available from the RAINBOW data base) in our survey (blue line), the mid-IR-selected AGN host galaxies (green shaded), and the control sample galaxies (red lines). 5 MORPHOLOGY CLASSIFICATION The aim of our morphology study is to look for evidence of mergers or interactions by examining the physical features of the AGN host galaxies. If AGN accretion is indeed triggered by mergers or interactions, we should observe a higher fraction of morphologically disturbed galaxies among our variability-selected sample compared to the control sample. We visually classify 30 × 30 pixel (0.9 × 0.9 arcsec2) thumbnails of the AGN hosts extracted from the F160W (1600nm) images, available in the CANDELS (Grogin et al. 2011) Public Access data base. The classification was done using the criteria described in more detail in Kocevski et al. (2012). First, the general morphology was classified as disc, spheroid, irregular, or point-like. Secondly, the degree of disturbance was classified as Merger/Interacting (highly disturbed with multiple nuclei or two distinct galaxies with interacting features such as tidal arms), Distorted/Asymmetric (single asymmetric or distorted galaxies with no visible interacting companion), Close Pair (near-neighbour pair with both the galaxies in a single frame), Double Nuclei (multiple nuclei in a single system) or Undisturbed (none of the above). Fig. 8 shows representative galaxies from our survey in the various morphological classes. Figure 8. View largeDownload slide Sample galaxies belonging to each of the classification groups employed in our analysis. The size of each thumbnail is 30 × 30 pixels or 0.9 × 0.9 arcsec2. Figure 8. View largeDownload slide Sample galaxies belonging to each of the classification groups employed in our analysis. The size of each thumbnail is 30 × 30 pixels or 0.9 × 0.9 arcsec2. We performed a blind inspection, which means that the images of the variable and control galaxies within each field were mixed to avoid any biases during the classification process of the galaxies. The classification was performed by two of the authors of this paper individually (M. Polimera and V. Sarajedini), and the results were merged. For ∼ 30 per cent of the galaxies, each reviewer had classified them differently and these galaxies were reinspected until a consensus could be reached on the proper classification. We chose to employ this type of visual inspection to resolve the subtle, low surface brightness features which are easy to miss using modern automated classification techniques. However, we hope to use these classifications as training sets for a machine-learning-based classifier in the future. 6 RESULTS Figs 9 and 10 show the visual morphology classification results tabulated in Table 2. The 1σ error bars are calculated as confidence intervals for a binomial population as   \begin{eqnarray} {\rm Error} = z_{1 - \alpha /2} \sqrt{\frac{p (1-p)}{n}}, \end{eqnarray} (5)where p is the fraction of variable galaxies in the each disturbance category and n is the total number of galaxies (e.g. Cameron 2011). We estimate the 1σ error where the confidence level is c = 0.683, so the variate value from the standard normal distribution z1−α/2 = 1 (α = 1 − c). Figure 9. View largeDownload slide Fraction of variable galaxies (blue diamonds) and control galaxies (red squares) in each of the major morphology classes. Figure 9. View largeDownload slide Fraction of variable galaxies (blue diamonds) and control galaxies (red squares) in each of the major morphology classes. Figure 10. View largeDownload slide Fraction of variable galaxies (blue diamonds) and control galaxies (red squares) in different disturbance classes. Figure 10. View largeDownload slide Fraction of variable galaxies (blue diamonds) and control galaxies (red squares) in different disturbance classes. Table 2. Disturbance classification. Galaxy type  Disturbance class  COSMOS  UDS  EGS  HDFN  ECDFS  Total  Variable  Merg/Int  6 (21 ± 8 per cent)  5 (21 ± 8 per cent)  2 (11 ± 7 per cent)  2 (11 ± 7 per cent)  3 (50 ± 20 per cent)  18 (19 ± 4 per cent)    Dist/Asy  10 (35 ± 9 per cent)  12 (50±10 per cent)  12 (70 ± 11 per cent)  5 (26 ± 10 per cent)  1 (17 ± 15 per cent)  40 (42 ± 5 per cent)    Undisturbed  12 (42 ± 9 per cent)  8 (33 ± 9 per cent)  3 (17 ± 9 per cent)  12 (63 ± 11 per cent)  2 (33 ± 19 per cent)  37 (39 ± 5 per cent)    Total  28  25  17  19  6  95  Control  Merg/Int  10 (22 ± 6 per cent)  17 (26 ± 5 per cent)  12 (25 ± 6 per cent)  11 (20 ± 5 per cent)  7 (35 ± 10 per cent)  57 (25 ± 3 per cent)    Dist/Asy  17 (37 ± 7 per cent)  22 (35 ± 6 per cent)  13 (27 ± 6 per cent)  19 (34 ± 6 per cent)  3 (15 ± 7 per cent)  74 (32 ± 3 per cent)    Undisturbed  17 (37 ± 7 per cent)  25 (39 ± 6 per cent)  22 (46 ± 7 per cent)  25 (45 ± 7 per cent)  10 (50 ± 11 per cent)  99 (43 ± 3 per cent)    Total  45  63  47  55  20  230  Galaxy type  Disturbance class  COSMOS  UDS  EGS  HDFN  ECDFS  Total  Variable  Merg/Int  6 (21 ± 8 per cent)  5 (21 ± 8 per cent)  2 (11 ± 7 per cent)  2 (11 ± 7 per cent)  3 (50 ± 20 per cent)  18 (19 ± 4 per cent)    Dist/Asy  10 (35 ± 9 per cent)  12 (50±10 per cent)  12 (70 ± 11 per cent)  5 (26 ± 10 per cent)  1 (17 ± 15 per cent)  40 (42 ± 5 per cent)    Undisturbed  12 (42 ± 9 per cent)  8 (33 ± 9 per cent)  3 (17 ± 9 per cent)  12 (63 ± 11 per cent)  2 (33 ± 19 per cent)  37 (39 ± 5 per cent)    Total  28  25  17  19  6  95  Control  Merg/Int  10 (22 ± 6 per cent)  17 (26 ± 5 per cent)  12 (25 ± 6 per cent)  11 (20 ± 5 per cent)  7 (35 ± 10 per cent)  57 (25 ± 3 per cent)    Dist/Asy  17 (37 ± 7 per cent)  22 (35 ± 6 per cent)  13 (27 ± 6 per cent)  19 (34 ± 6 per cent)  3 (15 ± 7 per cent)  74 (32 ± 3 per cent)    Undisturbed  17 (37 ± 7 per cent)  25 (39 ± 6 per cent)  22 (46 ± 7 per cent)  25 (45 ± 7 per cent)  10 (50 ± 11 per cent)  99 (43 ± 3 per cent)    Total  45  63  47  55  20  230  * The galaxies used to calculate these percentages are only those that are present in the RAINBOW data base for which the visual morphology classification described in Section 5 has been performed. View Large Fig. 9 shows no significant differences between the AGN host galaxy classifications and those for the control sample in terms of galaxy classified as disc, spheroidal, irregular, or point-like. We also find no difference in the percentage of galaxies in a close pair. We find a similar fraction of AGN hosts in galaxies with a discernible disc (40 ± 10 per cent) as did Kocevski et al. (2012) for an X-ray-selected sample (51.4 ± 5.8 per cent). About 30 ± 10 per cent of the mid-IR variables are found in pure elliptical (spheroidal) hosts. This is also quite similar to that found by Kocevski et al. (2012) for X-ray-selected AGN hosts (27.8 ± 5 per cent). In Fig. 10, we compare the percentage of galaxies displaying signs of mergers/interactions and distortion/asymmetry using the same symbols as Fig. 9. We find that AGN hosts and control galaxies have the same fraction of merger/interaction hosts within the uncertainties (19 ± 5 to 23 ± 3 per cent). However, AGN hosts have a slightly higher tendency to reside in galaxies that are either distorted or asymmetric (42 ± 5 per cent) compared to the control galaxy population (32 ± 3 per cent). This result is significant at the 1.7σ level. Dividing our sample into high- and low-mass hosts and control galaxies also produced similar results. As previously mentioned, Kocevski et al. (2012) studied 72 X-ray-selected AGN hosts in the GOODS-S field using the same CANDELS NIR images analysed in this study. However, they found no statistical difference between AGN hosts and the control galaxies having strongly disturbed (merger/interaction) or weakly disturbed (distorted/asymmetric) morphologies. One possibility they suggest is that X-ray-selected samples of AGN still miss the most obscured sources which could be hiding hosts displaying merger signatures. A follow-up study by K15 explored this issue by determining the level of obscuration in the X-ray-selected AGN using reflection-dominated X-ray spectroscopic analysis. They found that the more obscured X-ray detected AGN (i.e. those with the highest hydrogen column densities) had a higher fraction of morphologically disturbed host galaxies than less obscured X-ray detected AGN. Our results indicate that mid-IR variability-selected AGN samples have marginally significant levels of weakly disturbed host galaxy morphologies. This could be interpreted as an indication that our sample includes a significant number of obscured, Compton-thick AGN. In order to test this hypothesis, we cross-referenced our sample of mid-IR variables with X-ray identified AGN in the five survey fields [COSMOS: Elvis et al. (2009); ECDFS: Alexander et al. (2003) and Xue et al. (2011); EGS: Nandra et al. (2015); HDFN: Alexander et al. (2003); UDS: Ueda et al. (2008)]. We initially searched for X-ray sources within 1 arcsec of the position of each mid-IR variable and found that matched sources generally lie within a 0.5 arcsec radius. Setting this as our matching threshold, we found 22 of our mid-IR variables are also X-ray detected AGN (23 per cent). This means that 77 per cent of our sources, or 73 mid-IR variables, are not identified in deep X-ray surveys and are likely to be either spurious sources (expected among ∼ 20 per cent of our sample) or may be highly obscured AGN whose X-ray emission has been absorbed. If these represent the most obscured sources, we may expect to find their morphologies to be the most disturbed. In Fig. 11, we compare the fraction of host galaxies of different morphological classifications for the mid-IR variables that are also identified in X-ray surveys (green diamonds) to the mid-IR variable sources without an X-ray counterpart (black diamonds). We see that while the two samples have the same, low fraction of highly disturbed host galaxies, a much higher fraction of mid-IR-only sources have weakly disturbed hosts (48 ± 5 per cent) than those with X-ray counterparts (23 ± 8 per cent). This difference is significant at the 3σ level. When we compare these percentages with those of the most highly obscured, Compton-thick X-ray sources from K15, we find that they agree favourably. We find that 22 ± 5 per cent of mid-IR-only-selected AGN have highly disturbed host morphologies compared to $$21.5^{+4.2}_{-3.3} \, {\rm per \, cent}$$ of the most obscured X-ray sources in K15. Furthermore, 40 ± 5 per cent of mid-IR-only-selected AGN have weakly disturbed hosts compared to $$43.0^{+4.6}_{-4.4} \, {\rm per \, cent}$$ in K15. This further supports the claim that AGN selected via mid-IR variability are likely to contain a larger fraction of obscured AGN, possibly more obscured than the most absorbed X-ray-selected sources, revealing a higher fraction of asymmetric hosts and highlighting the unique nature of this selection technique. Figure 11. View largeDownload slide Fraction of mid-IR variable galaxies with an X-ray counterpart (green diamonds) and those with no X-ray counterpart detected within 0.5 arcsec (black diamonds) in different disturbance classes. Figure 11. View largeDownload slide Fraction of mid-IR variable galaxies with an X-ray counterpart (green diamonds) and those with no X-ray counterpart detected within 0.5 arcsec (black diamonds) in different disturbance classes. 7 DISCUSSION It has been speculated that mergers tidally induce non-axis-symmetric modes in the merging galaxy and these modes sweep the interstellar gaseous medium into a disc with a radius of a few parsec around the central SMBH (Shlosman, Frank & Begelman 1989; Shlosman, Begelman & Frank 1990). The disc then is accreted on to the SMBH. Simulations from Barnes & Hernquist (1992) have shown that these discs are more likely to form in a major head-on merger scenario. Given such a theory, we may expect to see the AGN hosts displaying a higher fraction of major mergers as compared to the control group. Our results, however, do not show a significant fraction of strong interaction signatures among the mid-IR variability-selected AGN hosts. We do find a marginally significant increase in the presence of weak interaction signatures in the host galaxies when compared to a control sample and the significance increases when we remove X-ray-selected AGN and isolate mid-IR variability-selected AGN. One explanation is that during the entire process of a galaxy merger, the galaxies spend more time in stages where they appear to be weakly disturbed versus having a highly disturbed morphology. Based on recent simulations by Capelo et al. (2015), the time a galaxy spends in the weakly disturbed phases is about 7.9 Gyr, but the system shows a highly disturbed morphology only for about 4.6 Gyr. This makes it less likely to ‘catch’ the galaxy in the highly disturbed phase because their signatures dissipate too fast and we see less dramatic distortions in the hosts. Another possibility is that the dominant AGN triggering mechanism is not major mergers but that minor mergers are more important for AGN to begin accretion as the perturbation is sufficient for the orbits of material in the central part of the galaxy to get randomized and disturbed just enough to infall into the central SMBH (Hernquist & Mihos 1995; Menci et al. 2014). Hernquist & Mihos (1995) also point out that the occurrences of major mergers are more unusual as compared to minor mergers/interactions. Yet another possibility is that the low fraction of major merger signatures seen in our AGN hosts is due to heavy obscuration which would absorb the variable UV photons from the accretion disc and weaken the mid-IR variable flux necessary for our selection method. As shown in K15, there is a significant increase in the fraction of AGN host galaxy morphology disturbance as a function of obscuration, with the Compton-thick AGN hosts showing the highest fraction of disturbed morphologies compared to unobscured hosts. Our study found that the host galaxies of mid-IR variables have the same fraction of disturbed morphologies as the most obscured X-ray-selected sources. It is possible that higher levels of disturbance are hidden by extreme obscuration such that neither X-ray nor mid-IR variability is able to detect the AGN nature of the source. 8 CONCLUSION The results of our study are summarized as follows. 1 per cent of all galaxies with [3.6] < 24.5 are significantly variable in the mid-IR Spitzer/IRAC mosaics with 80 per cent likelihood of being AGN. The AGN host galaxies show a marginally significant trend (1.7σ significance) towards higher fractions of Disturbed/Asymmetric morphologies (42 ± 5 per cent) as compared to the control sample (32 ± 3 per cent), whereas Merger/Interaction fractions are not statistically different for the AGN and control galaxies. 23 per cent of mid-IR variables are also identified in X-ray surveys. The mid-IR variables without an X-ray counterpart show a higher fraction of weakly disturbed hosts than those also identified with X-ray emission. This suggests that mid-IR variability selects a unique population of obscured AGN that would likely be missed using other selection techniques. Acknowledgements This work is based on observations made with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA. This work has also made use of the Rainbow Cosmological Surveys data base, which is operated by the Universidad Complutense de Madrid (UCM), partnered with the University of California Observatories at Santa Cruz (UCO/Lick,UCSC) and used observations taken by the CANDELS Multi-Cycle Treasury Program with the NASA/ESA HST, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555. Footnotes 1 Website: http://rainbow.fis.ucm.es REFERENCES Alexander D. M. et al.  , 2003, AJ , 126, 539 CrossRef Search ADS   Ashby M. L. N. et al.  , 2013, ApJ , 769, 80 CrossRef Search ADS   Ashby M. L. N. et al.  , 2015, ApJS , 218, 33 CrossRef Search ADS   Asmus D., Hönig S. F., Gandhi P., Smette A., Duschl W. J., 2014, MNRAS , 439, 1648 CrossRef Search ADS   Assef R. J. et al.  , 2010, ApJ , 713, 970 CrossRef Search ADS   Barnes J. E., Hernquist L., 1992, ARA&A , 30, 705 CrossRef Search ADS   Bertin E., Arnouts S., 1996, A&AS , 117, 393 CrossRef Search ADS   Cameron E., 2011, PASA , 28, 128 CrossRef Search ADS   Capelo P. R., Volonteri M., Dotti M., Bellovary J. M., Mayer L., Governato F., 2015, MNRAS , 447, 2123 CrossRef Search ADS   Elvis M. et al.  , 2009, ApJS , 184, 158 CrossRef Search ADS   Falocco S. et al.  , 2015, A&A , 579, A115 CrossRef Search ADS   Grogin N. A. et al.  , 2011, ApJS , 197, 35 CrossRef Search ADS   Hernquist L., Mihos J. C., 1995, ApJ , 448, 41 CrossRef Search ADS   Hopkins P. F., Kocevski D. D., Bundy K., 2014, MNRAS , 445, 823 CrossRef Search ADS   Kocevski D. D. et al.  , 2012, ApJ , 744, 148 CrossRef Search ADS   Kocevski D. D. et al.  , 2015, ApJ , 814, 104 (K15) CrossRef Search ADS   Kochanek C. S. et al.  , 2012, ApJS , 200, 8 CrossRef Search ADS   Koekemoer A. M. et al.  , 2011, ApJS , 197, 36 CrossRef Search ADS   Kozłowski S. et al.  , 2010, ApJ , 716, 530 CrossRef Search ADS   Kozłowski S., Kochanek C. S., Ashby M. L. N., Assef R. J., Brodwin M., Eisenhardt P. R., Jannuzi B. T., Stern D., 2016, ApJ , 817, 119 CrossRef Search ADS   Lacy M. et al.  , 2004, ApJS , 154, 166 CrossRef Search ADS   Lynden-Bell D., 1969, Nature , 223, 690 CrossRef Search ADS   MacLeod C. L. et al.  , 2010, ApJ , 721, 1014 CrossRef Search ADS   Menci N., Gatti M., Fiore F., Lamastra A., 2014, A&A , 569, A37 CrossRef Search ADS   Morse J. A., Raymond J. C., Wilson A. S., 1996, PASP , 108, 426 CrossRef Search ADS   Nandra K. et al.  , 2015, ApJS , 220, 10 CrossRef Search ADS   Park S. Q. et al.  , 2008, ApJ , 678, 744 CrossRef Search ADS   Pérez-González P. G. et al.  , 2008, ApJ , 675, 234 CrossRef Search ADS   Richards G. T. et al.  , 2002, AJ , 123, 2945 CrossRef Search ADS   Ruan J. J., Anderson S. F., Dexter J., Agol E., 2014, ApJ , 783, 105 CrossRef Search ADS   Sarajedini V., Koo D., Klesman A., 2009, Am. Astron. Soc. Meeting Abst. #213 . 238 Shlosman I., Frank J., Begelman M. C., 1989, Nature , 338, 45 CrossRef Search ADS   Shlosman I., Begelman M. C., Frank J., 1990, Nature , 345, 679 CrossRef Search ADS   Smitka M. T., 2016, PhD thesis , Texas A&M University Stern D. et al.  , 2005, ApJ , 631, 163 CrossRef Search ADS   Ueda Y. et al.  , 2008, ApJS , 179, 124 CrossRef Search ADS   Urry C. M., Padovani P., 1995, PASP , 107, 803 CrossRef Search ADS   Vazquez B. et al.  , 2015, ApJ , 801, 127 CrossRef Search ADS   Veilleux S., 2002, in Green R. F., Khachikian E. Y., Sanders D. B., eds, ASP Conf. Ser. Vol. 284, IAU Colloq. 184: AGN Surveys . Astron. Soc. Pac., San Francisco, p. 111 (astro-ph/0201118) Xue Y. Q. et al.  , 2011, ApJS , 195, 10 CrossRef Search ADS   © 2018 The Author(s) Published by Oxford University Press on behalf of the Royal Astronomical Society

### Journal

Monthly Notices of the Royal Astronomical SocietyOxford University Press

Published: May 1, 2018

## You’re reading a free preview. Subscribe to read the entire article.

### DeepDyve is your personal research library

It’s your single place to instantly
that matters to you.

over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month ### Explore the DeepDyve Library ### Search Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly ### Organize Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place. ### Access Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals. ### Your journals are on DeepDyve Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more. All the latest content is available, no embargo periods. DeepDyve ### Freelancer DeepDyve ### Pro Price FREE$49/month
\$360/year

Save searches from
PubMed

Create lists to

Export lists, citations