Online material

Appendix A: The inclusion of a central point source in GALFIT fits

Nuclear activity can systematically alter the appearance of AGN host galaxies by adding excess emission in the center, which concentrates the light profile, increases the resultant best-fit Sérsic index and makes the system appear more circular. Even at the highest attainable resolutions, the nucleus is completely unresolved in distant galaxies, appearing as a point source. It is reasonable, therefore, to model AGN host galaxies as a combination of galaxy structural components and a central PSF. However, bulges, which also make light profiles more central concentrated, cannot be easily distinguished from low levels of nuclear point source emission. Galaxy bulges at z ~ 2 have characteristic half-light diameters of ~2 kpc or 0.24′′ (e.g., Bruce et al. 2012), only slightly larger than the FWHM of the WFC3/F160W PSF (0.15′′). Therefore, accurately distinguishing between bulges and point sources requires careful modelling of the instrumental PSF, and even then may still be inaccurate, since galaxies do not usually have regular light profiles.

To test the performance of GALFIT fits with a PSF component for real z ~ 2 galaxies, we compared single Sérsic and two component (Sérsic + central PSF) fits to the F160W images of galaxies in the CANDELS GOODS-S field. This field has the deepest NIR imaging among the CANDELS fields. For this exercise, we used a subset of galaxies from the full sample introduced in Sect. 2.4, which have been carefully fit with different light profile models for a study of the bulge properties of distant galaxies by (Lang et al. 2014). Details of the methodology and fitting setup are published in that work. Only galaxies at 1.5 <z< 2.5 were considered in this exercise, since it is at these redshifts where the difference in light profiles between AGNs and inactive galaxies is most pronounced (Sect. 3.3). The galaxies were all selected to have with M_∗> 10¹⁰ M_⊙.

Each galaxy was first fit with a single elliptical Sérsic profile, iterated numerous times over a grid of initial values to prevent the best fit from falling into a local minimum. The best fitting single Sérsic profile was then used to initialise a second fit with the addition of a PSF component to the galaxy model. The center of the PSF was restricted to within 2 pixels (0.12′′) of the center of the Sérsic profile, and its flux was initialised to 1% of the integrated magnitude of the galaxy. All galaxies, whether identified as inactive or active, were fit identically in this fashion.

Fig. A.1 Distributions of the Sérsic index (n) from GALFIT fits to AGNs and inactive control galaxies in the WFC3/F160W band. 1000 draws of a mass-matched control sample are analyzed to determine the 1σ/2σ scatter in the n distributions for inactive galaxies, shown as dark/light grey zones in the histograms. Red open histograms show the distributions for the AGNs. Left: results from fits of a single Sérsic elliptical model. Right: results from fits with two components – a Sérsic elliptical model and a central point source. The inclusion of a point source results in lower Sérsic indices for both AGNs and inactive galaxies.

Open with DEXTER

In Fig. A.1, we compare the resultant Sérsic index distributions of the X-ray AGNs to those of mass-matched inactive galaxies, where the distributions for the inactive galaxies have been determined using the bootstrapping procedure discussed in Sect. 2.4.1. In the left panel, we show the results from single Sérsic fits. These are qualitatively similar to those shown in Figs. 5 and 7, in that AGNs typically show significantly higher Sérsic indices than inactive galaxies. In detail, we find a slightly higher fraction of n = 4 AGNs than in the full sample. However, given the small number of AGNs in this subsample, the difference could arise from Poissonian variation.

In the right panel, we plot the distributions of the best-fit Sérsic index of the galaxy component in the two component fits. As expected, including a PSF component has lowered the resultant n of the AGN hosts, greatly increasing the fraction of disk-dominated systems. On the other hand, an comparison of the distributions of the inactive galaxies between both panels in the figure also demonstrates a reduction in the typical n for these galaxies as well. Since the inactive population is not expected to show widespread nuclear point source emission, we conclude that simple two component fits as used in this exercise also tends to remove light from potential bulge components, systematically leading to best-fit Sérsic index distributions that are too disky.

Fig. A.2 Distributions of the flux ratio of the Sérsic component to the PSF component (in magnitudes) from two component GALFIT fits to AGNs and inactive galaxies in the WFC3/F160W band. 1000 draws of a mass-matched control sample are analyzed to determine the 1σ/2σ scatter in the n distributions for inactive galaxies, shown as dark/light grey zones in the histograms. Red open histograms show the distributions for the AGNs. The dashed black line is the distribution for the full inactive galaxy sample for which two component fits were performed, including many lower mass galaxies not widely found among AGN hosts. This histogram has been scaled down to overlap with the peak of the AGN histogram, to allow a simple visual comparison of the distributions.

Open with DEXTER

Another valuable test of these fits is a comparison of the fraction of light in the PSF and galaxy components for AGNs and mass-matched inactive galaxies. The nuclear luminosities of X-ray AGNs will be higher than any weak or heavily obscured nuclear emission that may remain undetected in the inactive population. If the central excesses are indeed due to widespread nuclear point source contamination, we expect to find higher PSF fractions among AGNs. In Fig. A.2, we plot histograms of the difference between the H-band magnitudes of the best-fit Sérsic component and the best-fit PSF components from our fits. In addition to the AGNs (red) and mass-matched inactive galaxies (grey regions), we also show the full distribution for galaxies with two-component fits (black dashed line), which includes many more low mass galaxies than generally found among AGN hosts. The latter histogram has been scaled down in number to allow a visual comparison to the other distributions in the figure.

Despite their higher nuclear luminosities, the two component fits yield essentially indistinguishable PSF fractions in the AGNs and equally massive inactive galaxies. This is not simply due to limitations of the fits or local minima, since the overall distribution of PSF fractions includes a long tail to very low values not found among the more massive AGN hosts (or other massive inactive galaxies). These two component fits were initialised with 5 mag between the PSF and the Sérsic component, but the best-fit difference is about 2.5 mag (10%). Considering that the excess central light is found to a similar degree both in AGNs and inactive galaxies, one may conclude that the central excess is likely to arise in a bulge rather than in a nuclear point source. This is consistent with arguments based on central colors and energetics from Sect. 3.3.1. We refrain from commenting on bulge fractions here – this requires detailed bulge+disk decomposition fits to both AGNs and inactive galaxies.

Appendix B: Comparison with the CANDELS study of Kocevski et al. (2012)

Kocevski et al. (2012) studied structural differences between AGN and mass-matched inactive control galaxies at 1.5 <z< 2.5 in the CDF-S using essentially the same visual classification scheme as in this work. However, there are important differences related to sample selections that should be borne in mind when comparing our results to theirs.

The X-ray source catalogs used in the two CANDELS studies are based on distinct reductions and source detection algorithms, leading to different sized parent samples. We employ the CDF-S 4Msec catalog of Xue et al. (2011) which consists of 740 sources, while Kocevski et al. (2012) use a more conservative catalog of 569 sources. The differences between these catalogs are discussed in Rangel et al. (2014). In addition, we adopt AGN-specific photometric redshifts from Luo et al. (2010), while Kocevski et al. (2012) take redshifts from Wuyts et al. (2008) which are not optimized for AGNs. Despite these differences, the number of X-ray AGNs at 1.5 <z< 2.5, after the application of a lower L_X limit, is nearly the same in both studies (≈70). Unlike us, however, Kocevski et al. (2012) do not restrict their sample to the GOODS-MUSIC footprint or apply cuts in m_H and M_∗. Our sample of AGNs for the visual classification analysis at 1.5 <z< 2.5 is 55, in total after cuts, a reduction of 25%. This smaller sample shares the same uniform analysis, quality checks and photometric selections of the full GOODS-MUSIC dataset, enabling a consistent analysis of stellar mass and other galaxy properties between AGNs and inactive galaxies in this work.

We can compare the fractions of disks and spheroids in our sample of AGNs with those reported by Kocevski et al. (2012). From their Table 1, ≈80% of AGNs are classified to have a visible disk or spheroid. In contrast, 100% of our AGNs have one of these components. This is due to the m_H< 24.5 cut applied to the galaxies in our visual classification catalog; we have less AGNs in our sample, but all have accurate structural assessments. As a consequence, our fractions of disks and spheroids will necessarily be higher than those in Kocevski et al. (2012), simply due to our different sample sizes. Therefore, we scale our fractions down by a factor of 1.25 to ease the comparison.

In our AGN sample, we find disk galaxy fractions of % of which % are pure disks, with no reported spheroid component. These fractions may be compared to and % respectively from Table 1 of Kocevski et al. (2012). We also find pure spheroid fractions of % compared to from Kocevski et al. (2012). As for morphologically disturbed systems (Sect. 3.4), the fractions of visual disks and spheroids are completely consistent in both CANDELS works. Our ability to reproduce the results of Kocevski et al. (2012), despite the differences of approach and numbers of classifiers, highlights the stability of the CANDELS visual classification scheme.

This being said, we prefer in this work to use an analytical measure of the galaxy light profile rather than visual measures of diskiness. As stated in Sect. 5.2, a visual classifier has difficulty discriminating between subtle variations in the light profile gradient. For e.g., inactive galaxies at z ~ 2 that are classified visually as having dominant disk components exhibit a range in Sérsic index of 0.7−3.2 (80th percentile), while the range is 1.7−6.6 for spheroid-dominated galaxies.

© ESO, 2014