Cosmic X-ray Surveys of Distant Active Galaxies: The Demographics, Physics, and Ecology of Growing Supermassive Black Holes

We review results from cosmic X-ray surveys of active galactic nuclei (AGNs) over the past ~ 15 yr that have dramatically improved our understanding of growing supermassive black holes in the distant universe. First, we discuss the utility of such surveys for AGN investigations and the capabilities of the missions making these surveys, emphasizing Chandra, XMM-Newton, and NuSTAR. Second, we briefly describe the main cosmic X-ray surveys, the essential roles of complementary multiwavelength data, and how AGNs are selected from these surveys. We then review key results from these surveys on the AGN population and its evolution ("demographics"), the physical processes operating in AGNs ("physics"), and the interactions between AGNs and their environments ("ecology"). We conclude by describing some significant unresolved questions and prospects for advancing the field.


General utility of X-ray surveys for studies of active galactic nuclei
Cosmic X-ray surveys have now achieved sufficient sensitivity and sky coverage to allow the study of many distant source populations including active galactic nuclei (AGNs), starburst galaxies, normal galaxies, galaxy clusters, and galaxy groups. Among these, AGNs, representing actively growing supermassive black holes (SMBHs), dominate the source number counts as well as the received integrated X-ray power. This has led to an impressive literature on the demographics, physics, and ecology of distant growing SMBHs found in X-ray surveys.
The intrinsic X-ray emission from AGNs largely originates in the immediate vicinity of the SMBH. The X-ray continuum arises via Compton upscattering in an accretion-disk "corona" over a broad X-ray band, and also perhaps via accretion-disk emission at low X-ray energies (e.g., Mushotzky et al 1993;Reynolds and Nowak 2003;Fabian 2006;Turner and Miller 2009;Done 2010;Gilfanov and Merloni 2014). AGNs hosting powerful jets furthermore often show strong jet-linked X-ray continuum emission (e.g., Worrall 2009;Miller et al 2011). This intrinsic X-ray emission may then interact with matter throughout the nuclear region to produce, via Compton "reflection" and scattering, more distributed X-ray emission. In some cases, when the intrinsic X-rays are obscured, such reflected/scattered emission may dominate the observed luminosity.
Cosmic X-ray surveys of AGNs offer considerable utility for several reasons: 1. X-ray emission appears to be nearly universal from the luminous AGNs that dominate SMBH growth in the Universe. When AGNs have been reliably identified using optical, infrared, and/or radio techniques, they almost always also show X-ray AGN signatures (e.g., see Fig. 1 and Avni and Tananbaum 1986;Brandt et al 2000;Mushotzky 2004;Gibson et al 2008). Thus, the intrinsic X-ray emission from the accretion disk and its corona empirically appears robust, even if its detailed nature is only now becoming clear (e.g., Done 2010;Schnittman and Krolik 2013). This point is discussed further in Section 3.3 and Section 4.2. 2. X-ray emission is penetrating with reduced absorption bias. The highenergy X-ray emission observed from AGNs is capable of directly penetrating through substantial columns with hydrogen column densities of N H = 10 21 -10 24.5 cm −2 (e.g., Wilms et al 2000, and references therein). 1 This is critically important, since the majority of AGNs in the Universe are now known to be absorbed by such column densities (see Section 4.1). X-ray surveys thus aid greatly in identifying the majority AGN populations and, moreover, in allowing their underlying luminosities to be assessed re-1 10 0.01 0.1

90% Conf. Limit on Fraction of Quasars
Factor of X−ray Weakness SDSS SDSS + BQS Fig. 1 Upper limit at 90% confidence on the fraction of Sloan Digital Sky Survey (SDSS; red curve) and SDSS+Bright Quasar Survey (BQS; blue curve) radio-quiet quasars that are X-ray weak by a given factor. The factor of X-ray weakness is computed relative to expectations based on optical/UV luminosity (see Section 4.2), where a value of unity represents the average quasar. Broad Absorption Line quasars, which are known often to have heavy X-ray absorption, have been excluded when making this plot. Note that quasars that are X-ray weak by a factor of 10 represent < ∼ 3% of the population. Adapted from Gibson et al (2008).
liably (in a regime where optical/UV luminosity indicators are generally unreliable). Only in the highly Compton-thick regime (N H 1/σ T , corresponding to N H 1.5 × 10 24 cm −2 ) does direct transmission become impossible, but here one can still investigate the (much fainter) X-rays that are reflected or scattered around the absorber (e.g., Comastri 2004; Georgantopoulos 2013). An additional relevant advantage of X-ray studies is that, as one studies objects at increasing redshift in a fixed observed-frame band, one gains access to increasingly penetrating rest-frame emission (i.e., higher rest-frame energies are probed); note the opposite generally applies in the optical and UV bandpasses where dust-reddening effects increase toward shorter wavelengths (e.g., Cardelli et al 1989). 3. X-rays have low dilution by host-galaxy starlight (i.e., emission at any wavelength associated with stellar processes). AGNs generally have much higher ratios of L X /L Opt and thus f X /f Opt than stars (e.g., Maccacaro et al 1988). Thus, X-rays provide excellent contrast between SMBH accretion Fig. 2 Optical and Chandra 2-8 keV images of a local active galaxy (NGC 3783); each image is 1.5 arcmin on a side. Note the substantial host-galaxy starlight competing with the AGN light in the optical band while, in the X-ray band, only the AGN light is apparent. The large contrast in the X-ray band between AGN light and starlight helps greatly with the identification of distant AGNs.
light and starlight (see Fig. 2), allowing one to construct pure samples of AGNs even down to relatively modest luminosities. This aspect of X-ray surveys is critical, for example, at high redshift where it is often unfeasible, at any wavelength, to resolve spatially the AGN light from host starlight. For weak or highly obscured AGNs, such dilution by host starlight can make AGNs difficult to separate from galaxies in the optical/UV regime (e.g., Moran et al 2002;Hopkins et al 2009). 4. The X-ray spectra of AGNs are rich with diagnostic potential that can be exploited when sufficient source counts are collected. At a basic level, the distinctive X-ray spectral characteristics of AGNs can often aid with their identification, improving still further the purity of AGN samples (see the previous point). Furthermore, measurements of low-energy photoelectric absorption cut-offs, underlying continuum shapes, Compton reflection continua, fluorescent line emission (e.g., from the iron Kα transition), and absorption edges (e.g., the iron K edge) can diagnose system luminosity, obscuration level, nuclear geometry, disk/corona conditions, and Eddington ratio (L Bol /L Edd ).
While these basic points of utility have led to great success for the enterprise of X-ray surveys, such surveys do have their shortcomings; e.g., in the regime of highly Compton-thick absorption or in cases of intrinsically X-ray weak AGNs (see Section 3.3). Thus, when possible, it is critical to complement X-ray surveys with suitably matched multiwavelength surveys in the area of sky under study. These can help considerably in filling the small chinks in the armor of X-ray surveys, thereby allowing nearly complete identification of all significant SMBH growth.

The survey capabilities of relevant distant-universe missions: Chandra, XMM-Newton, and NuSTAR
In this review, we will describe some of the main discoveries on AGNs coming from the intensive activity in X-ray (0.5-100 keV) surveys research over the past 15 yr, mainly focusing on missions that can make sensitive "blank-field" surveys of typical AGNs in the distant (z = 0.1-5) universe. Our emphasis will thus be on results from the Chandra X-ray Observatory (hereafter Chandra; e.g., Weisskopf et al 2000), the X-ray Multi-Mirror Mission (hereafter XMM-Newton; e.g., Jansen et al 2001), and the Nuclear Spectroscopic Telescope Array (hereafter NuSTAR; e.g., Harrison et al 2013). The new results from these missions rest squarely upon a rich heritage of X-ray survey studies with several superb earlier X-ray missions, as briefly described in, e.g., Section 3.1. We will also introduce more local results from, e.g., the Swift Gamma-Ray Burst Explorer (hereafter Swift; e.g., Gehrels et al 2004) and the International Gamma-Ray Astrophysics Laboratory (hereafter INTEGRAL; e.g., Winkler et al 2003) where they connect strongly with results from the distant universe, although these critical local investigations are not our primary focus (and deserve to be the subject of an entirely separate dedicated review).
In terms of basic survey capability, both Chandra (launched in 1999 July) and XMM-Newton (launched in 1999 December) provide X-ray spectroscopic imaging over broad bandpasses (0.3-8 keV and 0.2-10 keV, respectively) and over respectable fields of view (290 arcmin 2 for Chandra ACIS-I, and 720 arcmin 2 for XMM-Newton EPIC-pn). 2 Their imaging point spread functions are excellent (an on-axis half-power diameter of 0.84 arcsec for Chandra) or good (15 arcsec for XMM-Newton), though these degrade significantly with increasing off-axis angle. Their most sensitive surveys reach about 80-400 times deeper than those of previous X-ray missions, and excellent source positions (accurate to 0.5-4 arcsec) allow effective multiwavelength follow-up studies even at the faintest X-ray fluxes. Typical survey projects with Chandra and XMM-Newton generate hundreds-to-thousands of detected AGNs, allowing powerful statistical studies of source populations. Furthermore, systematic public data archiving practices allow effective survey combination, so that source populations spanning wide ranges of luminosity and redshift can be studied together.
NuSTAR is a more recently launched mission (2012 June) that is now transforming surveys of the X-ray universe above 10 keV. It is the first focusing high-energy  keV) X-ray observatory in orbit; for X-ray surveys of the generally faint sources in the distant universe, NuSTAR is most effective up to ≈ 24 keV (at higher energies, rising background levels and dropping photon collecting area limit its sensitivity to faint sources). This coverage of high energies equates to reduced absorption bias even relative to Chandra and XMM-Newton. The reduced bias is particularly key at z < ∼ 1 where the restframe energies covered by Chandra and XMM-Newton are still modest. The NuSTAR field of view for spectroscopic imaging is 140 arcmin 2 , and its imaging point spread function has an on-axis half-power diameter of 58 arcsec but with a sharp core having a full width at half maximum of 18 arcsec (this is about an order of magnitude improvement compared to the imaging capabilities of previous coded-mask instruments in orbit; e.g., Tueller et al 2010). Its sensitivity in probing the hard X-ray sky is about two orders of magnitude better than previous collimated or coded-mask instruments, making it the first genuine surveyor of the distant universe from 10-24 keV. The first publications from the extensive NuSTAR survey programs are presently appearing with more in preparation; these have been delivered by the members of the NuSTAR survey teams, although the underlying data will be made public for investigations by the whole astronomical community. Many additional NuSTAR results are expected in coming years.
1.3 Structure of this review, other relevant reviews, and definitions As noted above, there is a vast literature on the demographics, physics, and ecology of distant growing SMBHs found in X-ray surveys. Indeed, more than 500 papers on this subject have been published over the past 15 yr based on surveys with Chandra, XMM-Newton, and NuSTAR alone. Our primary aim here is to describe briefly some of the main discoveries coming from these efforts to make them more accessible to interested researchers and students. We note in advance that, owing to the vastness of the literature, it will not be possible to cover all relevant work; our apologies in advance if we could not cover your favorite result or paper.
The structure for the rest of this review will be the following: -In Section 2, we will review the main X-ray surveys of the distant universe and their supporting observations. We will also describe how AGNs are selected effectively using the X-ray and multiwavelength data. -Section 3 ("AGN demographics") will cover demographic results for distant X-ray selected AGNs, focusing on AGN evolution over cosmic time. This will also include brief discussions of the population of AGNs missed in cosmic X-ray surveys, the So ltan and related arguments, and the environmental dependence of AGN evolution. -Section 4 ("AGN physics") will describe insights on the physical processes operating in AGNs that have come from X-ray surveys. -Section 5 ("AGN ecology") will describe what X-ray surveys have revealed about interactions between growing SMBHs and their environments (mainly their host galaxies). This section will also discuss the relative radiative output from SMBHs and stars over cosmic time.
-In Section 6, we will outline key outstanding questions. We will also describe prospects for advancing the field both in the near and longer terms using both X-ray and multiwavelength follow-up facilities.
We note that Sections 3,4, and 5 cover strongly inter-related themes, and that there is inevitably a degree of subjectivity in assigning some results to a single one of these sections. Nevertheless, this structure is useful for basic organizational purposes. Over the past 15 yr, a number of other relevant in-depth reviews have been prepared that have some overlap with the topics discussed here. These include Hasinger and Zamorani (2000), Gilli (2004), Brandt and Hasinger (2005), Urry and Treister (2007), Hickox (2009), Brandt and Alexander (2010), , , Kormendy and Ho (2013), Merloni and Heinz (2013), Shankar (2013), and Gilfanov and Merloni (2014). We encourage interested readers to consult these reviews as well, noting that they generally emphasize somewhat different topics than those emphasized here. We also refer interested readers to the chapters in the Astrophysics and Space Science Library volume titled "Supermassive Black Holes in the Distant Universe" (Barger 2004).
Throughout this review we shall adopt J2000 coordinates and a standard cosmology with H 0 = 70 km s −1 Mpc −1 , Ω M = 0.3, and Ω Λ = 0.7. When quoting effective hydrogen column densities estimated from X-ray spectral analyses, we will adopt the cosmic abundances of Anders and Grevesse (1989). When referring to X-ray obscured AGNs, we will be considering systems with N H > ∼ 10 22 cm −2 unless noted otherwise. This threshold value is commonly adopted though admittedly somewhat arbitrary, being close to the typical maximum absorption expected from a galactic disk. It is also broadly consistent with the division in X-ray absorption level between optically obscured and optically unobscured AGNs. When referring to highly X-ray obscured AGNs, we will generally mean systems with column densities at least a factor of ≈ 50 greater; i.e., N H > ∼ 5 × 10 23 cm −2 .
2 The main cosmological X-ray surveys, their supporting observations, and AGN selection 2.1 Description of the current main cosmic X-ray surveys of distant AGNs The capabilities of Chandra, XMM-Newton, and NuSTAR (see Section 1.2) have led to a substantial number of X-ray surveys being conducted of the distant universe. These include targeted surveys, both deep and wide, where a sky area of particular interest is observed; e.g., a field already having excellent multiwavelength data or a field containing a notable object such as a highredshift protocluster. Furthermore, these include serendipitous surveys that investigate the serendipitous sources detected in a number of fields observed for other reasons. Some selected extragalactic surveys conducted with Chandra, XMM-Newton, and NuSTAR are listed in Table 1. A number of aspects of this table deserve  note: 1. These surveys often have wide ranges of sensitivity across their associated solid angles due to, e.g., differing satellite pointing strategies and instrumental effects. The serendipitous surveys (e.g., ChaMP, SEXSI, CSC, HELLAS2XMM, 3XMM) particularly stand out in this regard, often being made up of observations differing by an order of magnitude or more in exposure time. 2. The solid angles quoted generally represent the total sky coverage at bright X-ray flux limits. 3. When listing XMM-Newton exposure times, we have attempted to remove time intervals affected by strong background flaring; such intervals are generally not useful for surveys of faint cosmic sources. 4. The listed exposure times are for a single X-ray telescope and focal plane module (FPM). Note that XMM-Newton and NuSTAR have three and two simultaneously operating telescopes/FPMs, respectively. 5. Some of the surveys have overlap of their solid angles of coverage (e.g., CDF-S vs. E-CDF-S; AEGIS-X Deep vs. AEGIS-X; XDEEP2 Shallow vs. Stripe 82X-Chandra; SXDS vs. XMM-LSS XMDS vs. XMM-LSS vs. XMM-XXL). 6. Some of these surveys are still increasing in solid angle and/or depth. For example, many of the serendipitous surveys continue to grow as more X-ray observations are performed, and the Chandra CDF-S survey is presently being raised to a 7 Ms exposure. 7. Some of the listed surveys have been conducted in multiple epochs spanning up to ≈ 15 yr, allowing assessments of long-timescale X-ray variability for the detected sources (e.g., Paolillo et al 2004;Papadakis et al 2008;Young et al 2012b;Lanzuisi et al 2014).
Interested readers should consult the cited papers in Table 1 for survey-specific details. Plots of solid angle of sky coverage vs. sensitivity in both the 0.5-2 keV and 2-10 keV bands for Chandra and XMM-Newton surveys are given in Figure 3. Together, all these surveys cover a broad part of the practically accessible sensitivity vs. solid-angle "discovery space" via the standard "wedding-cake" design, providing a quite complete understanding of AGN populations in the distant universe (though, as discussed in Section 6.2, there is still room for important improvements). One persistent limitation of X-ray surveys in general, however, has been the lack of sensitive and thoroughly followed-up surveys over hundreds-to-thousands of deg 2 ; the widest sensitive surveys presently are 3XMM and XMM-XXL. This limitation has hindered X-ray constraints upon rare objects, such as the most luminous AGNs in the Universe, although targeted X-ray follow-up studies of such objects selected at other wavelengths have mitigated this issue to some degree (e.g., Vignali et al 2005;Just et al 2007;Stern et al 2014). eROSITA (e.g., Merloni et al 2012) is expected to im-prove this situation significantly in the near future, and dedicated wide-field X-ray telescopes (e.g., Murray et al 2013;Rau et al 2013) could make major additional strides (see Section 6.4 for more detailed discussion).
Surveys with Chandra and XMM-Newton have resolved ≈ 75-80% of the cosmic X-ray background (CXRB) from 0.5-10 keV into point sources (e.g., see Section 1.3 of Brandt and Hasinger 2005 for further discussion). Up-to-date measurements of the resolved fraction of the CXRB below 10 keV in discrete energy bands may be found in, e.g., Hickox and Markevitch (2006), Lehmer et al (2012), Xue et al (2012), and Ranalli et al,in prep;Fig. 4 shows results from one such analysis. These measurements provide useful integral constraints upon remaining undetected X-ray source populations, although they still lie significantly below the peak of the CXRB at 20-40 keV. The deepest surveys with NuSTAR are expected to resolve 30-40% of the 8-24 keV CXRB (e.g., Ballantyne et al 2011;J.A. Aird 2014, private communication), reaching closer to its peak. This is a large improvement over pre-NuSTAR results, where only a few percent of the 10-100 keV CXRB was resolved (e.g., Krivonos et al 2005;Treister et al 2009b).

Supporting multiwavelength observations and spectroscopic follow-up
Characterization of the detected X-ray sources using both multiwavelength photometric data and spectroscopic observations is crucial for investigating their nature. At the most basic level, a detected X-ray source must be matched reliably to a multiwavelength photometric counterpart so that, e.g., the feasibility of spectroscopic observations can be determined; such matching is done most effectively using a likelihood-ratio technique (e.g., Sutherland and Saunders 1992;Rutledge et al 2000;Ciliegi et al 2003;Naylor et al 2013). Counterpart matching is straightforward for the majority of sources in Chandra surveys owing to the excellent angular resolution of Chandra (generally providing 0.5-1.5 positions; e.g., Evans et al 2010), although there are genuine matching challenges for the faintest optical, near-infrared (NIR; about 1-5 µm), and/or mid-infrared (MIR; about 5-30 µm) counterparts. For XMM-Newton surveys the counterpart matching is more challenging (generally 2-4 positions; e.g., Watson et al 2009), although the majority of sources in XMM-Newton surveys can also be matched to optical/NIR/MIR counterparts. In current NuSTAR surveys (generally 10-20 positions; e.g., Harrison et al 2013), the detected sources are typically first matched to Chandra/XMM-Newton (or other X-ray) sources which then are matched to optical/NIR/MIR counterparts. The X-ray sources found in the surveys listed in Table 1 span an extremely broad range of optical/NIR/MIR flux; e.g., the I-band magnitudes for AGNs range from brighter than 15th to fainter than 28th magnitude (e.g., Barger et al 2003b;Szokoly et al 2004;Laird et al 2009;Brusa et al 2010;Luo et al 2010;Pineau et al 2011;Rovilos et al 2011;Xue et al 2011;Trichas et al 2012).  Table 1 from Chandra (blue) and XMM-Newton (green). For comparison purposes, a few surveys from previous X-ray missions are shown in red. The circles around some of the points indicate serendipitous surveys as also denoted in Table 1. Some of the surveys are labeled by name (sometimes abbreviated) in regions where symbol crowding is not too strong. The vertical dotted line shows the solid angle for the whole sky. Each of the surveys has a range of sensitivity across its solid angle, and different authors use somewhat different methodologies for computing and quoting sensitivity; this leads to small uncertainties in the precise relative locations of the data points. The resolved fraction of the CXRB in the 4 Ms CDF-S observation as a function of energy from 0.5-8 keV. Shown are the resolved fraction from the 4 Ms CDF-S sources (blue), the "bright-end correction" that accounts for bright sources too rare to be found within the studied CDF-S field of view (red), and results from stacking X-ray photons coincident with z-band identified galaxies that are not individually X-ray detected (green, titled "Sample A"). The summation of these three components is also shown (magenta). The total CXRB intensity is taken from Hickox and Markevitch (2006) with (non-negligible) uncertainty indicated by the gray area. The unresolved CXRB below ≈ 5 keV, even after inclusion of the stacked emission coincident with galaxies, is likely associated with groups/clusters of galaxies. Adapted from Xue et al (2012).
The X-ray data from current surveys generally do not allow direct redshift determination, although there are occasional cases where redshifts can be measured based on the strong iron Kα line and/or iron K edge in X-ray spectra (e.g., Iwasawa et al 2012;Del Moro et al 2014). Thus, spectroscopic and/or photometric determination of redshifts, usually in the optical/infrared, is essential for, e.g., calculation of source luminosity, the most meaningful X-ray spectral modeling, and studies of source cosmic evolution. In line with the broad range of optical/NIR/MIR flux for the counterparts, a wide variety of facilities have been used productively for spectroscopic redshift determination, including the largest telescopes on Earth (e.g., Keck, the Very Large Telescope, and Subaru) for the faintest sources in deep surveys. Enormous progress has been made with spectroscopic redshift determination for X-ray survey sources, particularly when multi-object spectrographs can be utilized to target efficiently large numbers of X-ray sources simultaneously; the field Fig. 5 (a) Distribution of the photometric redshift (photo-z, or z photo ) accuracy, ∆z/(1 + zspec), derived from an unbiased blind-test sample of X-ray AGNs in the CDF-S (zspec values are spectroscopic redshifts). The typical photo-z accuracy for the sample is evaluated with a robust estimator, the normalized median absolute deviation (σ NMAD ): σ NMAD = 1.48 × median[(|∆z − median(∆z)|)/(1 + zspec)]. (b) Photo-z accuracy, ∆z/(1 + zspec), vs. R-band magnitude for the blind-test sample. The solid line indicates z photo = zspec, and the dotted lines represent relations of z photo = zspec ± 0.15(1 + zspec). The photo-z accuracy declines toward faint R-band magnitudes, particularly when considering the frequency of catastrophically incorrect photo-z values. Adapted from Luo et al (2010).
sizes of these spectrographs often match the sizes of the X-ray survey fields well. Large samples of X-ray sources with reasonably complete spectroscopic identification down to I = 22-24 are now available for statistical studies (e.g., Barger et al 2003b;Fiore et al 2003;Szokoly et al 2004;Eckart et al 2006;Della Ceca et al 2008;Trouille et al 2008;Trump et al 2009;Silverman et al 2010;Kochanek et al 2012;Trichas et al 2012). This being said, much work is still needed to improve the spectroscopic completeness for some key fields, in some cases by systematically publishing spectra already acquired. At fainter fluxes of I = 24-28 where many X-ray sources are found, particularly in the deepest X-ray surveys, the spectroscopic completeness drops rapidly (e.g., even in the intensively studied CDF-S, only ≈ 65% of the X-ray sources overall have spectroscopic redshifts). This bottleneck remains one persistent driver for the construction of future Extremely Large Telescopes in the optical/NIR (see Section 6.3).
It is often possible to derive photometric redshifts of reasonable quality for X-ray AGNs when spectroscopic redshifts are unavailable, although these are generally of lower quality than those for comparable non-AGN galaxies (e.g., see Fig. 5). Photometric redshifts also provide a useful cross-check even when spectroscopic redshifts are available; e.g., if only one emission line is clearly detected then there can be significant uncertainty in the correct spectroscopic redshift determination. In the best cases, photometric redshifts are derived from > ∼ 20 bands of MIR-to-UV photometric data, utilize dedicated templates suitable for X-ray sources such as AGN/galaxy hybrids, utilize dedicated sets of medium-band filters, and/or allow for AGN optical/NIR/MIR variability effects between different filters observed non-simultaneously (e.g., Salvato et al 2009;Cardamone et al 2010;Luo et al 2010;Xue et al 2012;Hsu et al 2014). Photometric redshift derivations for X-ray sources can reach much fainter optical magnitudes than can be reached spectroscopically (e.g., to I ≈ 28). When compared with relatively bright optical sources having spectroscopic redshifts, they have a (magnitude-dependent) typical accuracy in ∆z/(1 + z) of 1-10% (using σ NMAD , see Fig. 5) with an outlier fraction of catastrophically incorrect redshifts of 3-20%. Further improvements in statistical redshift estimation for optically faint X-ray sources should be possible in the near future using clustering-based techniques (e.g., Matthews and Newman 2010;Ménard et al 2014).
In addition to counterpart identification and redshift determination, multiwavelength observations play many further key roles in the effective investigation of sources from cosmic X-ray surveys. These include basic characterization of the nature of the detected sources (see Section 2.3), measurements of broad-band AGN spectral energy distributions (SEDs) to determine more reliable bolometric luminosities and investigate accretion processes, and measurements of AGN host-galaxy properties [e.g., stellar mass, star-formation rate (SFR), morphology, interaction status, and large-scale environment; see Section 5]. Furthermore, these multiwavelength data have been used to identify AGNs and AGN candidates missed by the X-ray selection technique (e.g., Compton-thick and/or intrinsically X-ray weak AGNs; see Section 3.3).

AGN selection from the general X-ray source population
The main classes of extragalactic X-ray sources detected in cosmic surveys include AGNs, starburst galaxies, normal galaxies, galaxy clusters, and galaxy groups. We will not review the non-AGN classes here and instead refer readers to the other reviews cited in Section 1.3 for relevant details (e.g., see Section 2.2 of Brandt and Hasinger 2005). Instead, consistent with our focus in this review, we will describe how AGNs are selected from the general X-ray source population.
Multiple methods can be used to derive a highly reliable sample of AGNs, including obscured and low-luminosity systems, from a sample of X-ray survey point sources. Those relying upon direct use of the X-ray data include the following: 1. X-ray luminosity. Sources with 0.5-10 keV luminosities above 3×10 42 erg s −1 are predominantly AGNs. Only rare extreme starburst galaxies in the distant universe, such as luminous submillimeter galaxies, can exceed this threshold without an AGN being present; caution is needed in applying the luminosity threshold when such sources are under study. 2. X-ray luminosity vs. SFR. Many researchers have established relations between X-ray luminosity and SFR for starburst/normal galaxies that lack AGNs (e.g., Bauer et al 2002;Ranalli et al 2003;Persic and Rephaeli 2007;Lehmer et al 2010;Mineo et al 2014). X-ray sources lying well above (typically, > ∼ 5 times is used) such relations are found to be AGNs. When quality data capable of constraining host-galaxy SFR are available (e.g., in the radio, infrared, and/or UV), this method is both more reliable and more complete than the straight X-ray luminosity cut of method 1 above. If host-galaxy stellar mass is also available, then the expected X-ray luminosity can be estimated as a function of both SFR and stellar mass. 3. X-ray-to-optical/NIR flux ratio. Consistent with the low dilution of X-ray emission by host-galaxy starlight noted in Section 1.1, AGNs tend to have higher X-ray-to-optical/NIR flux ratios than starburst/normal galaxies. Typically thresholds of log(f 0.5−10 keV /f R ) > −1 using the observed-frame R band or log(f 0.5−10 keV /f 3.6µm ) > −1 using the observed-frame 3.6 µm band from Spitzer (e.g., Werner et al 2004) serve to select samples that are 90-95% AGNs (other bands can also be used in a similar fashion, although the requisite threshold will vary). Ideally these ratios would be derived using rest-frame rather than observed-frame bands, since the X-ray and optical/NIR fluxes have significantly different k-corrections. However, this is often not practical or possible, and when using observed-frame bands it is generally best to use the reddest optical/NIR band available (e.g., to at least minimize the effects of dust extinction in high-redshift sources). 4. X-ray spectral shape. X-ray sources with flat effective power-law photon indices in the 0.5-10 keV band of Γ eff < 1.0 are generally obscured AGNs (obscured AGNs can also have larger Γ eff values, but use of a larger Γ eff threshold can lead to uncertain AGN identifications). The effective photon index from a simple power-law fit is a useful first-order indicator of spectral shape even when the observed spectrum does not precisely have a powerlaw form. It can be estimated based upon direct X-ray spectral fitting or, in cases of limited counts, based upon a hardness/band ratio. In Γ eff < 1.0 cases, the effective photon index is generally flat owing to X-ray absorption and/or Compton reflection. The X-ray binary populations that dominate the emission from starburst/normal galaxies are empirically found to produce steeper X-ray spectra with Γ eff > ∼ 1.5. 5. X-ray variability. Rapid X-ray variability by large amplitudes is commonly seen among those AGNs where the direct continuum produced close to the SMBH is observable. This variability is generally stronger than that seen from collections of X-ray binaries in starburst/normal galaxies (e.g., Young et al 2012b). Furthermore, as noted above, some of the X-ray surveys in Table 1 allow variability studies over much longer timescales. Significantly variable sources that also have X-ray luminosities larger than ≈ 10 41 erg s −1 are likely to be AGNs, although there are rare notable exceptions (e.g., Kaaret et al 2001;Webb et al 2010). 6. X-ray position. The X-ray positions of AGNs are generally coincident with the apparent nuclei of their host galaxies (as opposed to, e.g., X-ray binaries, which can be identified across the extent of the host galaxy often as off-nuclear sources). This positional coincidence can be checked for relatively low-redshift objects (z < ∼ 0.5) when high-resolution X-ray (e.g., Chan- Note that AGNs remain the numerically dominant source population down to faint fluxes, although at still-fainter 0.5-2 keV fluxes galaxies will become numerically dominant. The AGN number counts reach ≈ 14, 900 deg −2 at the faintest 0.5-2 keV fluxes, and this is the highest sky density of reliably identified AGNs found at any wavelength. Taken from Lehmer et al (2012). dra) and optical/NIR (e.g., HST ) imaging are available (e.g., Lehmer et al 2006).
Some of these methods have a long history (e.g., Maccacaro et al 1988 (2012), . Note that some of these methods rely upon having fairly precise redshift information available while others depend much less upon redshift; AGN samples can often be selected reasonably well using methods 3-6 together prior to redshift determination. AGNs are generally found to make up 75-95% of the sources by number in current X-ray surveys, with their percentage contribution dropping with survey depth as many starburst/normal galaxies are detected at faint fluxes (primarily at low X-ray energies). The precise fractional contribution from AGNs as a function of survey depth has been quantified in number counts apportioned by source type (see Fig. 6; e.g., Bauer et al 2004;Lehmer et al 2012). In addition to the approaches above relying upon the direct use of X-ray data, approaches relying upon independent multiwavelength data can also be used for AGN selection/confirmation from a sample of X-ray sources. These include the detection of broad and/or high-ionization emission lines in optical/NIR spectra, high surface brightness radio core emission or extended radio jets/lobes, strong infrared emission from hot dust heated by an AGN continuum, and distinctive optical variability. In well-studied X-ray and multiwavelength survey fields, multiple independent methods have been applied to cross-validate AGN candidates, leading to the most reliable and pure samples of distant AGNs available.

The status of AGN evolution studies before Chandra and XMM-Newton
The number-density evolution of the AGN population over cosmic time has been a topic of intense interest since the 1960's (e.g., Schmidt 1968). Early work focused on the evolution of luminous quasars and radio galaxies detected in wide-field optical/radio surveys (see, e.g., Hartwick and Schade 1990;Hewett and Foltz 1994;Boyle 2001;Osmer 2004; Heinz 2013 for reviews), and such wide-field studies have continued to advance until the present (e.g., Wall et al 2005;Richards et al 2006b;Massardi et al 2010;Ross et al 2013). Wide-field optical surveys remain largely constrained to the relatively rare systems where the AGN significantly outshines the host galaxy, owing to the selection techniques employed (typically based upon source colors measured in photometric data).
Prior to the launches of Chandra and XMM-Newton, the population of luminous quasars had been well established to evolve strongly over cosmic time, peaking in number density at z ≈ 2-3. Most analyses up to the year ≈ 2000 found that the form of the evolution at z < ∼ 2.5 could be fit acceptably with pure luminosity evolution models, although there were reports of more complex evolution at bright magnitudes (e.g., Hartwick and Schade 1990;Boyle et al 2000). Many researchers reasonably expected that the basic evolutionary behavior derived for luminous quasars would also apply at lower luminosities.
The German-USA-UK ROSAT mission was the most effective surveyor of the distant X-ray (0.1-2.4 keV) universe prior to Chandra/XMM-Newton, contributing to a number of fundamental results on AGN demographics. The ROSAT Deep Survey in the Lockman Hole, resolving 70-80% of the 0.5-2 keV CXRB, directly showed that AGNs produce most of the background in this band (e.g., Hasinger et al 1998;Schmidt et al 1998). The responsible AGNs showed a range of optical spectral types, including broad-line and narrowline spectra, and X-ray-to-optical flux ratios [−1 < ∼ log(f 0.5−2 keV /f R ) < ∼ 1]. Results on the X-ray luminosity function (XLF) derived from ROSAT surveys spanning a range of depths indicated that luminosity-dependent density evolution (LDDE) described the data better than pure luminosity evolution or pure density evolution (e.g., Miyaji et al 2000Miyaji et al , 2001; but see also Page et al 1997), importantly indicating a luminosity dependence of AGN evolution. At higher energies, the Japanese-USA ASCA and Dutch-Italian BeppoSAX missions were able to resolve ≈ 30% of the 2-10 keV CXRB, and follow-up studies indicated that the majority of the sources in this band were also AGNs (e.g., Ueda et al 1998;Fiore et al 1999). Many of these sources had hard X-ray spectra with effective power-law photon indices of Γ Eff = 1.3-1.7, consistent with longstanding expectations that obscured AGNs make much of the CXRB which has Γ Eff = 1.4 from 2-10 keV (e.g., Setti and Woltjer 1989;Comastri et al 1995).
The demographics of high-redshift AGNs at z = 3-6 were highly uncertain when Chandra and XMM-Newton began operation. Optical and radio surveys of luminous quasars both indicated a consistent strong decline in number density above z ≈ 3 (e.g., Schmidt et al 1995;Shaver et al 1999). However, X-ray surveys of somewhat less luminous quasars suggested a lack of any strong decline (e.g., Miyaji et al 2000) and were consistent with a constant number density at z > 2. Theoretical considerations offered few further constraints, allowing extremely high quasar space densities in principle (e.g., Haiman and Loeb 1999). Given the available constraints, it was feasible that the Universe was reionized at z ≈ 5-7 by AGNs.

X-ray luminosity functions and the luminosity dependence of AGN evolution
Owing to the advantages of X-ray surveys detailed in Section 1.1, Chandra and XMM-Newton have allowed the effective selection of distant AGNs, including obscured systems, that are up to ≈ 100 times less bolometrically luminous than those found in wide-field quasar surveys such as the SDSS (e.g., Ross et al 2012; the vast majority of SDSS broad-line quasars are straightforwardly detected in moderate-depth X-ray surveys). Objects similar to local moderate-luminosity Seyfert galaxies can be identified out to z ≈ 5. The sky density of X-ray selected AGNs in the deepest Chandra surveys has now reached ≈ 14, 900 deg −2 (see Fig. 6; Lehmer et al 2012), making them ≈ 500 times more numerous on the sky than SDSS quasars. As another comparison, the deepest optical photometric AGN surveys reaching B ≈ 24.5 have delivered AGN sky densities of ≈ 400 deg −2 (e.g., Wolf et al 2004;Beck-Winchatz and Anderson 2007; also see Palanque-Delabrouille et al 2013). Furthermore, the ≈ 14, 900 deg −2 sky density is ≈ 15 times larger than that from the ROSAT Deep Survey (970 deg −2 ;Hasinger et al 1998), the deepest X-ray survey conducted prior to Chandra and XMM-Newton. NuSTAR is now further broadening the parameter space of discovery, allowing improved identification of highly obscured AGNs at relatively bright flux levels (e.g., Del . AGN samples from Chandra and XMM-Newton are now thought to be sufficiently complete over a broad part of the luminosity-redshift plane to allow many fundamental issues regarding AGN evolution to be addressed (though further improvements are still critical; e.g., see Section 3.3). A key finding, first reported by Cowie et al (2003) and now a subject of many studies, is a notable "anti-hierarchical" luminosity dependence of AGN evolution, such that the number density of lower-luminosity AGNs peaks later in cosmic time than that of powerful quasars (see Fig. 7; e.g., Barger et al 2005;Hasinger et al 2005;La Franca et al 2005;Silverman et al 2008a;Ebrero et al 2009;Yencho et al 2009;Aird et al 2010;Ueda et al 2014). This qualitative behavior is also broadly seen in star-forming galaxies (e.g., Cowie et al 1996) and is often referred to as "cosmic downsizing", although this term has developed a number of usages with respect to galaxies (e.g., Bundy et al 2006;Cimatti et al 2006;Faber et al 2007;Fontanot et al 2009). AGN downsizing was not widely anticipated prior to its observational discovery; AGN population synthesis models for the CXRB at the time generally adopted the basic evolutionary behavior of luminous quasars for AGNs of all luminosities. The most X-ray luminous AGNs with L X = 10 45 -10 47 erg s −1 peak in number density at z ≈ 2-3, consistent with the behavior of optically selected quasars, while more common AGNs with L X = 10 43 -10 44 erg s −1 peak at z ≈ 0.8-1.5. Fig. 7 shows that the luminosity dependence of AGN evolution appears sufficiently important to shift the overall peak of cosmic SMBH power production to lower redshifts (z ≈ 1.5-2) than would be expected solely from the study of luminous quasars (z ≈ 2-3). Roughly, at z < 1, z = 1-2, and z > 2 the integrated fractions of SMBH growth (i.e., the total accreted mass onto SMBHs) are broadly comparable at 25-35%, 37-47%, and 23-33%, respectively. AGNs with L X = 10 44 -10 45 erg s −1 dominate SMBH power production at z = 1.5-4, while those with L X = 10 43 -10 44 erg s −1 dominate at lower redshifts (at very low redshifts of z < ∼ 0.5, AGNs with L X = 10 41 -10 43 erg s −1 also make large fractional contributions). Ongoing NuSTAR surveys at higher X-ray energies up to ≈ 24 keV, while providing valuable insights, do not suggest any qualitative revisions to this basic picture (e.g., Alexander et al 2013;Del Moro et al 2014;Civano et al, in prep;Mullaney et al, in prep); e.g., the vast majority of the NuSTAR survey sources were previously detected by Chandra and/or XMM-Newton. The downsizing behavior of AGNs has also now been found in optically selected (e.g., Bongiorno  Measurement of the quantitative details of the downsizing behavior for X-ray AGNs depends upon many challenging issues, including initial detection completeness (and corrections for missed AGNs), multiwavelength counterpart identification, completeness in redshift determination, X-ray spectral modeling (e.g., X-ray absorption corrections to luminosity estimates), and statistical methodology in XLF calculation. Thus, while the basic downsizing phenomenon appears securely established, there is still some remaining debate over the precise form of the XLF and its evolution. The most recent in-depth studies have proposed either LDDE (e.g., Ueda et al 2014) or luminosity and density evolution (LADE; e.g., Aird et al 2010); in the latter the shape of the XLF is constant with redshift, but it undergoes strong luminos- Fig. 7 (a) Comoving number density vs. redshift for AGNs, selected from multiple X-ray surveys, in four rest-frame 2-10 keV luminosity classes [as labeled; in units of log(erg s −1 )]. Note that the number density of moderate-luminosity AGNs peaks later in cosmic time than that of powerful quasars (i.e., AGN cosmic downsizing). (b) Comoving bolometric luminosity density vs. redshift for the same AGN sample in six bolometric luminosity classes [as labeled; in units of log(erg s −1 )]. Note the peak luminosity density at z ≈ 1.8 for AGNs over the broad range of L Bol = 10 43 -10 48 erg s −1 . Taken from Ueda et al (2014).
ity evolution at z < ∼ 1, and overall negative density evolution toward increasing redshift. Especially at high redshifts of z > ∼ 3, the LDDE and LADE models predict quite different numbers of AGNs. The latest high-redshift constraints appear to favor the LDDE model, but further testing is needed.
The observed AGN downsizing behavior seen via the measured XLF could arise due to changes in the mass of the typical active SMBH and/or changes in the typical accretion rate. A variety of modeling efforts have been made to understand the physical nature of AGN downsizing, including analytic models, semi-analytic models, and large-scale numerical simulations (e.g., Hopkins et al 2008;Degraf et al 2010;Fanidakis et al 2012;Hirschmann et al 2012Hirschmann et al , 2014Menci et al 2013); these efforts sometimes also attempt to model simultaneously the growth of AGN host galaxies and their downsizing behavior. The numerical simulations continue to advance rapidly and include many of the mechanisms relevant to SMBH fueling and growth, including galaxy interactions, disk instabilities, and gas cooling. They have had genuine success in plausibly reproducing the apparent basic anti-hierarchical behavior of SMBH growth within the context of the hierarchical paradigm for cosmic structure formation. This being said, even the most intensive simulations to date lack the spatial/mass resolution to model in detail the still uncertain but essential accretion and feedback processes operating on small scales, and approximate "sub-grid" approaches are often adopted for these processes.
Given the immense modeling challenges, it is understandable that the model predictions for XLF evolution and the nature of downsizing end up differing in detail. Broadly, and as would initially be expected from the XLF results, the models generally indicate that more massive SMBHs grew earlier in cosmic time. Some furthermore quantitatively predict a decline in average Eddington ratio with redshift (e.g., Fanidakis et al 2012;Hirschmann et al 2012Hirschmann et al , 2014. The strong early drop in number density of the luminous AGN population is predicted to result from the exhaustion of cold gas in massive halos due to strong early star formation and feedback, as well as a decline over time in merging activity. Less-luminous AGNs evolve more mildly and can remain numerous later in cosmic time, however, as the gas content of their generally lower mass halos evolves more mildly. A significant fraction of the less-luminous AGNs are also remnants of formerly luminous AGNs that have faded; i.e., objects with low Eddington ratios, perhaps intermittently triggered, whose massive SMBHs are no longer rapidly growing (in a fractional sense) but can still appear as AGNs. Additional observational evidence consistent with this basic picture includes estimates of SMBH masses and Eddington ratios for distant AGNs in X-ray surveys (see Section 5.4) and observations of the mass-dependent growth timescales of local SMBHs (e.g., Heckman et al 2004).
Chandra and XMM-Newton have also greatly clarified the demographics of AGNs at z = 3-6, although there is still scope for significant improvements. In contrast to the earlier suggestions from ROSAT surveys (see Section 3.1), the X-ray data now clearly support an exponential decline in the number density of luminous AGNs above z ≈ 3 (e.g., Barger  , ruling out some of the more exotic early predictions (e.g., Haiman and Loeb 1999) by ≈ 2 orders of magnitude. Furthermore, quantitative comparisons of space densities for optically selected quasars (e.g., McGreer et al 2013) and X-ray selected quasars indicate statistical agreement to within factors of 2-3. At lower AGN luminosities, the situation is significantly less clear, largely owing to the small solid angles with sufficiently sensitive X-ray data as well as substantial challenges with spectroscopic/photometric follow-up studies. The current data generally suggest a decline in space density for moderate-luminosity AGNs at z > 3 (e.g., Fiore et al 2012a; Vito et al 2013Vito et al , 2014aKalfountzou et al 2014;Ueda et al 2014). Some recent work has found this decline may be less pronounced at the lowest luminosities, and if this trend holds then "cosmic upsizing" would apply in this regime, perhaps consistent with expectations for hierarchical structure formation in the early universe (i.e., given the young age of the Universe, most SMBHs would not yet have sufficient masses to generate high luminosities). Improved measurements are required for clarification, especially for the most highly obscured AGNs at high redshift (e.g., Gilli et al 2011). The available space-density estimates indicate that, in the absence of dramatic changes in the XLF at very low luminosities, SMBHs probably did not produce sufficient power to reionize the Universe at z ≈ 5-7; they likely have secondary effects upon reionization (e.g., Grissom et al 2014; but see Giallongo et al 2012). Stacking analyses of high-redshift galaxies have also set upper limits on the accreted mass density in black holes out to z ≈ 8, providing useful inputs into models of early SMBH formation (e.g., Cowie

AGNs missed in cosmic X-ray surveys and their importance
As noted in Section 1.1, X-ray surveys do have (small) shortcomings, and thus it is essential to utilize well-matched multiwavelength surveys to find AGNs missed by the X-ray technique (as well as to characterize more reliably the underlying bolometric luminosities of X-ray detected AGNs; see Section 2.2). Particularly important are missed AGNs that still have sufficiently large bolometric luminosities to contribute materially to cosmic SMBH growth. The primary way that luminous AGNs can be missed in X-ray surveys involves heavy obscuration, and that will be the focus of this section below. Additionally, however, there is growing evidence that a small fraction of the luminous AGN population is intrinsically X-ray weak (e.g., Leighly et al 2007;Wu et al 2011;Luo et al 2014;Teng et al 2014), thereby mildly challenging the rule of universal luminous X-ray emission from AGNs (see point 1 of Section 1.1). Even in the absence of obscuration, such AGNs will often be difficult to detect, but thankfully current assessments indicate that intrinsically X-ray weak AGNs are sufficiently rare that they should not substantially impact demographic studies (e.g., Gibson et al 2008;Wu et al 2011;Luo et al 2014).
Heavy obscuration with N H = (5-50) × 10 23 cm −2 or more, as is commonly seen in the local universe, can diminish the measurable X-ray emission from an AGN to below the limits of detectability, even for intrinsically luminous AGNs in deep X-ray surveys. The factor of diminution depends upon the sampled rest-frame energy band, the absorption column density, and the absorption geometry (the latter setting the level of flux that is Compton reflected/scattered around the absorber); the factor can be 10-100 or more in the 0.5-10 keV band for large column densities (e.g., Comastri 2004; Burlon et al 2011). Furthermore, owing to the largely energy independent nature of Compton scattering below a couple hundred keV, observations at tens of keV offer only limited improvements for sources with highly Compton-thick absorption column densities.
Multiple methods have been applied to identify AGN candidates that are undetected in the X-ray band, and tens of papers have now been written on this subject. Arguably the most successful methods have utilized infrared observations, largely from Spitzer , that home in on AGN "waste heat" (i.e., AGN-heated dust emission) and have reduced extinction bias. Such infrared selection techniques include the following:  Optical, radio, and other techniques have also been utilized to identify AGN candidates, including the following: to enable cross-checking of candidates. X-ray stacking analyses using samples of AGN candidates derived with the methods above often show hard average X-ray spectral shapes, indicating that at least some highly obscured AGNs are indeed present. However, large uncertainties remain about the fraction of candidates that are bona-fide AGNs and the corresponding AGN luminosities, particularly among the infrared-selected samples, and much further candidate characterization is required; e.g., with deeper or harder X-ray observations and high-quality optical/NIR/MIR spectroscopy. Furthermore, many of these methods, on their own, are not effective at distinguishing between highly obscured and moderately obscured AGNs. The X-ray undetected obscured AGNs presumably produce the ≈ 25% of the 6-8 keV CXRB that remains unresolved (see Section 2.1), and this provides one integral limit upon their overall importance to cosmic SMBH growth (e.g., Xue et al 2012;Ueda et al 2014); also see Section 3.4. If they have low intrinsic luminosities, as some analyses suggest, they might increase the current AGN number counts (≈ 14, 900 deg −2 ; Lehmer et al 2012) by 50% or more.

The So ltan argument for X-ray selected AGNs
The established relations between the bulge properties of local galaxies and the masses of the SMBHs at their centers (e.g., Kormendy and Ho 2013; Shankar 2013; and references therein) allow estimation of the total local mass density of SMBHs (ρ •,loc , in M Mpc −3 ). Following So ltan (1982), this quantity then serves as an integral constraint upon the allowed amount of total cosmic SMBH growth. Multiple authors have performed this local SMBH mass-density estimation; e.g., Marconi et al (2004) (2013) have recently argued that some past SMBH mass estimates need to be revised upward by a factor of about 2-4, owing to improvements in data, improvements in modeling, and the identification of downward biases in emission-line based SMBH masses (see their Section 6.6). This leads to an increase in ρ •,loc by a factor of ≈ 2 (L.C. Ho 2014, private communication). Given that this change would be well in excess of the error bars of past ρ •,loc estimates, it is clear that one must presently tread cautiously when using the So ltan argument! The size of the required upward revision of ρ •,loc is still a matter of debate (A. Marconi 2014, private communication).
X-ray AGN demographers have integrated AGN bolometric luminosity functions, derived from XLFs via bolometric corrections, to estimate the total amount of cosmic SMBH growth, ρ •,XLF , for comparison to ρ •,loc (e.g., Marconi et al 2004;Shankar 2013;Ueda et al 2014; also see Delvecchio et al 2014 for a recent infrared-based perspective). Acceptable agreement can be achieved, broadly supporting the idea that radiatively efficient gas accre-tion (i.e., AGN phases) has driven much of cosmic SMBH growth (significant SMBH merging is allowed, provided the merging progenitors grew via radiatively efficient accretion). For example, Ueda et al (2014) find ρ •,XLF = (3.9 ± 0.6) × 10 5 η −1 0.1 M Mpc −3 , where η 0.1 is the average mass-to-energy conversion efficiency of accretion divided by 0.1. η 0.1 ≈ 0.8 would give consistency with ρ •,loc estimates from 2000-2012, while η 0.1 ≈ 0.4 would be required if the recent upward revision of Kormendy and Ho (2013) is adopted. For comparison, accretion onto a Schwarzschild SMBH would give η 0.1 ≈ 0.57 or higher (with the exact η 0.1 depending upon the role of magnetic stress at the innermost stable circular orbit around the SMBH), while (prograde) accretion onto a Kerr SMBH would give η 0.1 as high as ≈ 3.6 (e.g., Agol and Krolik 2000;Noble et al 2009Noble et al , 2011J.H. Krolik 2014, private communication). The ρ •,XLF vs. ρ •,loc agreement is thus fair, though the implied accretion efficiency would appear low for upward revisions of ρ •,loc as suggested by Kormendy and Ho (2013). This could be remedied if XLFs still suffer from incompleteness due to, e.g., highly obscured and/or intrinsically X-ray weak AGNs (see Section 3.3). 3 Such objects are very difficult to detect in X-ray surveys, making this hypothesis challenging to assess. Detailed considerations also indicate that η may depend significantly upon SMBH mass and redshift (e.g., Li et al 2012;Shankar 2013;Ueda et al 2014).
The So ltan argument can be utilized in a differential, rather than an integral, manner to investigate the evolution of the mass function of SMBHs (e.g., Marconi et al 2004;Shankar 2013;Ueda et al 2014). The most robust conclusion arising from such work is that more massive SMBHs generally grew earlier in cosmic time (also see Section 3.2). Unfortunately, present uncertainties in bolometric corrections, η, ρ •,loc , and XLFs limit the precision of So ltanargument constraints for, e.g., constraining the amount of SMBH growth missed by X-ray surveys (see Section 3.3). So ltan-argument constraints are currently a good first-order "sanity check" but should not be over-interpreted.

The environmental dependence of AGN evolution
Theoretical models of structure formation predict that galaxy growth is environmentally dependent (i.e., it is accelerated in high-density regions; e.g., Kauffmann 1996;De Lucia et al 2006). Observational support for this hypothesis comes from the finding that the most evolved and massive spheroids reside in galaxy clusters (high-density regions) by the present day (e.g., Baldry et al 2004;Smith et al 2009). How is the growth of SMBHs influenced by the largescale environment? A powerful way to address this question is to compare the fraction of galaxies hosting AGNs in galaxy clusters (and their high-redshift progenitors, protoclusters) to that in the field, as measured from blank-field X-ray surveys.
To first order, the fraction of galaxies in clusters hosting X-ray luminous AGNs (L X > ∼ 10 43 erg s −1 ) is found to evolve strongly with redshift out to z ≈ 1, qualitatively similar to what is seen for the field-galaxy population. However, the cluster AGN fraction is lower than that in the field by about an order of magnitude at low redshift, and it appears to rise more rapidly with redshift (e.g., Eastman et al 2007;Martini et al 2009Martini et al , 2013. This evolution of the AGN fraction with redshift also broadly tracks that seen for the star-forming galaxy population in galaxy clusters. There is evidence that the relative suppression of AGNs in galaxy clusters is dependent on the richness of the cluster, the clustercentric radius explored, and the adopted luminosity threshold for AGN activity (e.g., Kocevski et al 2009;Klesman andSarajedini 2012, 2014;Ehlert et al 2014b,a;Koulouridis et al 2014), complicating comparisons between different studies. In less extreme large-scale environments, such as galaxy groups, there is no clear evidence for a decrease in the AGN fraction when compared to the field out to at least z ≈ 1 (e.g., Georgakakis et al 2008;Silverman et al 2009a;Pentericci et al 2013;Oh et al 2014).
The source statistics are more limited at z > ∼ 1, but, on the basis of current results, the AGN fraction in galaxy clusters at z = 1.0-1.5 appears broadly similar to that in the field (e.g., Martini et al 2013). At yet higher redshifts there is a deficit of galaxy cluster systems, although studies of AGNs in protoclusters (large-scale overdense regions at z > ∼ 2) have found an order of magnitude increase in the AGN fraction when compared to the field (e.g., Lehmer et al 2009bLehmer et al , 2013Digby-North et al 2010), an apparent reversal of what is seen at z < 1. There is also tentative evidence that the AGN fraction in protoclusters increases in regions of higher galaxy density (e.g., Lehmer et al 2009bLehmer et al , 2013, potentially the opposite to what is seen in massive lower-redshift galaxy clusters (e.g., Ehlert et al 2014b). Overall, the current results suggest that the large-scale environment has an effect on the growth of SMBHs, which is presumably driven by the availability of a cold-gas supply in the vicinity of the SMBH and may be sparse in massive, evolved (i.e., virialized) galaxy clusters but is prevalent in less-evolved galaxy clusters and protoclusters (e.g., van Breukelen et al 2009). However, there are potentially conflicting results between the suite of published studies, which may be driven by a number of effects, including the analyses employed to measure AGN activity, the approach adopted when comparing to non AGNs (e.g., luminosity and mass thresholds when calculating the AGN fraction), and the selection of the galaxy clusters and protoclusters.
The large-scale environment can also be quantified with the two-point correlation function, which measures the clustering strength of selected populations and provides an estimate of the dark-matter halo mass. Clusteringstrength measurements for X-ray selected AGNs over the broad redshift range of z ≈ 0-3 imply a characteristic halo mass of ≈ 10 12 -10 13 M (e.g.  Koutoulidis et al 2013). This range in halo mass is in agreement with that found for luminous star-forming galaxies, optically selected quasars, and massive galaxies, but it is lower than that found for radioselected AGNs (e.g., Hickox et al 2009Hickox et al , 2011Hickox et al , 2012Georgakakis et al 2014a;see Fig. 5 of Alexander and Hickox 2012). On the basis of darkmatter halo models, ≈ 10 13 M broadly corresponds to the maximum mass where a halo can support a large cold-gas supply-the gas in halos significantly more massive than this is mostly in a hot form, which is less easily accreted onto the SMBH (e.g., Cattaneo et al 2006;Croton 2009;Kereš et al 2009). The dark-matter halo may, therefore, have a strong controlling influence on the fuelling of AGNs (e.g., Booth and Schaye 2010;Volonteri et al 2011).
Further useful constraints can be placed by measuring how X-ray AGNs are distributed in dark-matter halos using the halo occupation distribution (HoD; e.g., Berlind and Weinberg 2002;Zheng et al 2005) formalism. The main parameters of the HoD are the fraction of central and satellite galaxies hosting AGN activity as a function of the halo mass. Current constraints suggest a preference for X-ray AGNs to reside in central galaxies, with < 5% identified with satellites (e.g., Starikova et al 2011;Richardson et al 2013); however, models with X-ray AGNs solely hosted in satellite galaxies can also fit the observed clustering in some cases (e.g., Miyaji et al 2011).

AGN physics
The large and relatively complete samples of AGNs detected in X-ray surveys can provide unique insights, usually of a statistical character, into the physical processes that shape their emission (i.e., "AGN physics"). These physical processes span scales from the immediate vicinity of the SMBH (light minutes to light hours) to that of the obscuring material (light days to light years). Such survey-based physical investigations critically complement the in-depth targeted X-ray studies of individual and small samples of AGNs that are another major thrust of X-ray astronomy (e.g., Mushotzky et al 1993;Reynolds and Nowak 2003;Fabian 2006;Turner and Miller 2009;Done 2010;Gilfanov and Merloni 2014). In this section, we will briefly review results on AGN physics coming from three key areas of survey-based investigation: the basic properties, luminosity dependence, and redshift dependence of AGN X-ray obscuration (Section 4.1), X-ray-to-optical/UV SEDs (Section 4.2), and the X-ray continuum shape and its connection to Eddington ratio (Section 4.3).

AGN X-ray obscuration: Basic properties, luminosity dependence, and redshift dependence
Understanding of the nature of the obscuring material in AGNs, often referred to generally as the obscuring "torus" of orientation-based unification models (e.g., Antonucci 1993), continues to improve rapidly. In recent years, for example, it has become possible to obtain direct estimates of the (luminosity dependent) extent of the torus (e.g., Burtscher et al 2013;Koshida et al 2014) and to ascertain its apparently clumpy nature (e.g., Elitzur and Shlosman 2006;Mor et al 2009). Studies based on X-ray surveys have advanced understanding of AGN obscuration in several regards. First, X-ray surveys provide arguably the clearest evidence that the majority of the AGNs in the Universe are obscured. Basic X-ray spectral analyses of the detected AGNs in deep X-ray surveys find that the majority show evidence for obscuration (e.g., Dwelly and Page 2006;Tozzi et al 2006;Merloni et al 2014), even before accounting for biases against detecting heavily obscured objects (see Section 3.3). The inferred column-density distribution of the underlying obscured population, before separation into luminosity and redshift bins, appears to peak roughly around N H ≈ 10 23 cm −2 with an approximately log-normal shape having a logarithmic σ ≈ 1. This result implies that a substantial fraction of AGNs will be highly obscured and even Compton-thick. Additionally, there is a significant minority of low-obscuration systems with N H values consistent with zero, in agreement with expectations from unification models.
Another key advance, where recent X-ray surveys covering a broad part of the luminosity-redshift plane have contributed substantially, has been in clarifying the long-suspected (e.g., Lawrence and Elvis 1982;Lawrence 1991) luminosity dependence of the fraction of obscured AGNs (e.g., Treister and Urry 2006;Hasinger 2008;Burlon et al 2011;Brightman et al 2014;Merloni et al 2014;Ueda et al 2014). The fraction of AGNs showing X-ray obscuration drops strongly with rising X-ray luminosity, from ≈ 70% at L 2−10 keV = 10 43 erg s −1 to ≈ 20% at L 2−10 keV = 10 45 erg s −1 (for z < 1). Results from recent infrared (e.g., Ballantyne et al 2006;Treister et al 2008;Assef et al 2013;Lusso et al 2013;Toba et al 2014) and optical (e.g., Simpson 2005) surveys broadly confirm this basic revision of orientation-based unification models (but see Lawrence and Elvis 2010). AGNs of higher luminosities may be able to evacuate or destroy, via radiative feedback, circumnuclear gas and dust more effectively, leading to a so-called "receding torus" with larger opening angle (e.g., Menci et al 2008;Nenkova et al 2008). Alternatively, luminosity-dependent changes in the underlying X-ray-to-optical/UV SED (see Section 4.2) may drive changes in the absorption properties; e.g., if the obscuring material is often in the form of a radiatively driven wind. The precise numerical values for obscured fractions differ between different authors and have remaining systematic uncertainties owing to, e.g., selection biases against highly obscured AGNs and limited photon statistics in absorption spectral modeling; at faint X-ray fluxes currently only crude hardness-ratio based absorption estimates can be effectively used, and these cannot appropriately characterize complex X-ray absorption (e.g., Mayo and Lawrence 2013;Buchner et al 2014). Moreover, wide-field infrared surveys suggest, somewhat surprisingly, that the fraction of highly obscured AGNs may rise upward substantially again at the highest luminosities (L Bol ∼ 10 47 erg s −1 ) reaching ≈ 50% (e.g., Assef et al 2014;Stern et al 2014). At the highest AGN luminosities, the nature of the material typically providing the obscuration may change from the standard small-scale torus to something else, perhaps more distributed gas and dust that has been perturbed by galaxy major mergers leading to high Eddington-ratio AGNs (e.g., Draper and Ballantyne 2010).
The fraction of AGNs showing X-ray obscuration, after allowing for luminosity effects, also appears to rise with redshift (e.g., Treister and Urry 2006;Hasinger 2008;Hiroi et al 2012;Iwasawa et al 2012;Vito et al 2013Vito et al , 2014aBrightman et al 2014;Merloni et al 2014;Ueda et al 2014). Recent studies find this rise can be parameterized as proportional to (1 + z) 0.4−0.6 at least up to z ≈ 2, beyond which the uncertainties become substantial. Such evolution of the obscured fraction might arise due to the generally greater availability of nuclear gas and dust in galaxies at earlier cosmic times. There remains debate regarding whether this redshift evolution applies for all AGNs or primarily for the most-luminous ones (e.g., Iwasawa et al 2012;Vito et al 2013;Merloni et al 2014;Ueda et al 2014); a luminosity dependence of the evolution might arise if low-to-moderate luminosity AGNs and high-luminosity AGNs have different fueling mechanisms (see Section 5.3). This evolution of AGNs on the physical scale of the absorbing medium is notably not accompanied by apparent small-scale evolution of the accretion disk and its corona (see Section 4.2).
Finally, we note briefly that this section has primarily focused on X-ray obscuration. The relations between X-ray obscuration and that at other wavelengths (e.g., optical/UV reddening, blocking of line emission from the Broad Line Region, UV absorption lines) remain an extremely complex issue requiring much further work, and understanding such relations thoroughly will elucidate the structure and kinematics of the obscuring material (e.g.

Basic X-ray-to-optical/UV spectral energy distribution properties
One of the most effective ways to investigate the accretion physics of the SMBHs in AGNs is to study their overall broad-band SEDs. The part of the AGN SED spanning from the X-ray to the optical/UV (hereafter, the "X-rayto-optical/UV SED") is the spectral region where direct accretion emission dominates for relatively unobscured systems (after subtraction/removal of the light from the AGN host galaxy; see Section 5.1). Studies of X-ray-tooptical/UV SEDs have a long history (e.g., Avni and Tananbaum 1986), and they continue to deliver fundamental insights as improved data, analysis techniques, and theoretical modeling become available. In addition to providing constraints upon accretion physics, these studies also are critically important, e.g., when 1. Assessing the universality of AGN X-ray emission and selecting remarkable AGNs that deviate from the typical X-ray-to-optical/UV SED (see Section 1.1 and 3.3). 2. Making bolometric corrections in the So ltan argument (see Section 3.4).
3. Modeling the winds responsible for many of the absorption-line properties of AGNs and likely feedback; the UV and extreme-UV (EUV) continuum largely drives these winds, which must also be protected from overionization by the underlying X-ray emission (e.g., Murray et al 1995;Proga et al 2000;Luo et al 2014).
While the X-ray-to-optical/UV portion of the AGN SED is complex (e.g., Vasudevan and Fabian 2009;Trump et al 2011a;Elvis et al 2012;Jin et al 2012a;Scott and Stewart 2014), its first-order properties can be described in large samples with the use of simple parameters. The most common such parameter is α ox , defined to be the slope of a nominal power law connecting the rest-frame 2500Å and 2 keV monochromatic luminosities; i.e., α ox = 0.3838 log(L 2 keV /L 2500Å ). For systems where the direct accretion emission is dominant, α ox compares the relative amounts of power coming from the optically thick accretion disk (at rest-frame 2500Å) and the accretion-disk corona (at rest-frame 2 keV).
X-ray surveys have now provided sensitive and often relatively uniform coverage for substantial numbers of unobscured or mildly obscured AGNs that span a broad part of the AGN luminosity-redshift plane. These have been used, often in conjunction with other AGN samples, to extend studies of X-ray-to-optical/UV SEDs using α ox (e.g., Steffen et al 2006;Just et al 2007;Kelly et al 2007;Green et al 2009;Lusso et al 2010;Young et al 2010); importantly, sensitive X-ray surveys have allowed the majority population of moderate-luminosity AGNs at high redshift to be included in such analyses. α ox shows a clear correlation with optical/UV luminosity (L 2500Å ) at all investigated redshifts (z = 0-6), such that AGNs with higher L 2500Å produce less X-ray emission per unit optical/UV emission (see Fig. 9). This finding is qualitatively in agreement with earlier studies (e.g., Avni and Tananbaum 1986;Anderson and Margon 1987;Wilkes et al 1994), and the correlation with L 2500Å appears stronger and tighter than that with estimates of L Bol /L Edd . The correlation is now well established over at least five orders of magnitude in L 2500Å , and the observed range of α ox corresponds to a substantial range of ≈ 20 in X-ray vs. optical/UV luminosity. There is considerable object-toobject scatter, corresponding to a factor of ≈ 3 in X-ray vs. optical/UV luminosity in either direction around the α ox -log(L 2500Å ) relation. Some of this scatter simply arises from (generally) non-simultaneous observations of AGNs that vary both in the X-ray and optical/UV, but the majority appears to be genuine intrinsic scatter (e.g., Gibson and Brandt 2012;Vagnetti et al 2013). Additional physical parameters beyond L 2500Å , such as L Bol /L Edd , likely can explain much of this scatter (e.g., Kelly et al 2008;Shemmer et al 2008;Lusso et al 2010;Young et al 2010;Jin et al 2012a). The α ox -log(L 2500Å ) relation is also likely nonlinear (e.g., Steffen et al 2006;Maoz 2007;Green et al 2009;Vagnetti et al 2013), appearing flatter at low luminosities and steeper at high luminosities, but detailed constraints on the form of the nonlinearity remain limited. Fig. 9 The αox parameter vs. 2500Å monochromatic luminosity for (a) optically selected AGNs from Steffen et al (2006) and Just et al (2007), and (b) X-ray selected AGNs from Lusso et al (2010). Larger negative values of αox correspond to weaker X-ray emission relative to optical/UV emission. The bottom portions of each panel show the residuals relative to the best-fit relation. In panel (a), the Kendall's τ coefficient and its significance are presented, along with the functional form of the best-fit relation. In panel (b), the abbreviated references for the plotted fits refer to Steffen et al (2006), Just et al (2007), and Lusso et al (2010). The symbol colors/types in each panel denote the AGN samples utilized; see the cited papers for details. AGNs become relatively X-ray weaker with increasing optical/UV luminosity, over a very wide range of this luminosity. Taken from Just et al (2007) and Lusso et al (2010).
The functional form and parameters of the α ox -log(L 2500Å ) relation provide fundamental constraints that any successful model of the SMBH diskcorona system must be able to reproduce. For example, as large-scale numerical magnetohydrodynamic simulations of black-hole accretion flows continue their rapid advances (e.g., Schnittman and Krolik 2013), it is expected that researchers will soon be able to determine which physical parameters (L Bol /L Edd , SMBH mass, SMBH spin, magnetic-field structure) set the diskto-corona power balance and the observed X-ray-to-optical/UV SED. Given this importance for accretion models, it is of concern that the recent improved studies of the α ox -log(L 2500Å ) relation, while agreeing on its existence, sometimes disagree about its slope/intercept and functional form; e.g., fitted parameters disagree by more than is allowed by their statistical uncertainties. Until such systematic uncertainties in the observational results are resolved, it will be difficult to use them to inform physical disk-corona models with high fidelity. Furthermore, α ox was defined at rest-frame 2500Å and 2 keV largely for observational convenience, rather than for fundamental reasons, and it is not obvious that these are the optimal wavelength/energy choices to consider for characterization of X-ray-to-optical/UV SEDs. In this vein, Young et al (2010) have considered the effects of varying these choices. The slope of the α ox -log(L Opt ) relation does depend significantly upon chosen X-ray energy, generally becoming steeper/flatter as the definition is moved to lower/higher energies. On the other hand, the slope does not appear to depend strongly upon the chosen optical/UV wavelength.
Most recent studies of X-ray-to-optical/UV SEDs using α ox find no significant evolution with redshift (e.g., Steffen et al 2006;Just et al 2007;Green et al 2009;Lusso et al 2010); the tightest limits require the X-ray-to-optical/UV luminosity ratio to change by < ∼ 30% out to z = 5-6. However, there are counterclaims finding evidence for redshift evolution (e.g., Kelly et al 2007), and the issue requires further scrutiny with even further improved samples. The current consensus is that, in spite of the dramatic evolution of AGN number densities over cosmic time (see Section 3.2), the inner accretion properties of the individual AGN unit are notably stable.

X-ray continuum shape as an estimator of Eddington ratio
As the primary indicator of SMBH growth rate, the Eddington ratio is of critical importance in studies of SMBH demographics (see Section 3.2), physics (see Section 4.2), and ecology (see Section 5.4). L Bol /L Edd is typically derived by estimating the mass of a SMBH (and thus its Eddington limit) and also its bolometric luminosity. Unfortunately, mass estimates and bolometric corrections for SMBHs in the distant universe generally have substantial uncertainties (e.g., see Shen 2013 and Peterson 2014 for discussion of virial SMBH mass estimators and Hao et al 2014 for discussion of bolometric corrections; also see Section 5.4). Thus, it is important to have as many methods as possible for L Bol /L Edd estimation, so that different approaches can be cross checked.
It has long been suspected that the intrinsic hard X-ray (rest-frame 2-10 keV) power-law photon index (Γ ) of a radio-quiet AGN can be used as an estimator of L Bol /L Edd (e.g., Pounds et al 1995;Brandt et al 1997). Higher L Bol /L Edd is expected to lead to increased Compton cooling of the accretion-disk corona, and thus steeper 2-10 keV power-law spectra (i.e., larger values of Γ ; higher L Bol /L Edd also often leads to the production of a strong "soft X-ray excess" affecting the spectrum below rest-frame 2 keV). The early suspicions have now been confirmed via both targeted (e.g., Shemmer et al 2006Shemmer et al , 2008) and X-ray survey-based (e.g., Risaliti et al 2009;Jin et al 2012a;Brightman et al 2013;Fanali et al 2013) studies, which find clear Γ − L Bol /L Edd correlations across a broad range of luminosity and redshift (and no apparent redshift dependence); see Fig. 10. The Γ − L Bol /L Edd correlation does have significant object-to-object scatter (a factor of ≈ 3, when high-quality Γ measurements are available; e.g., Shemmer et al 2008) and thus, as with other methods of L Bol /L Edd estimation, it is best used in a statistical sense to characterize the average Eddington ratio of a sample of objects. The Γ − L Bol /L Edd technique, of course, requires a reliable measurement of the intrinsic power-law photon index; i.e., corrected for X-ray absorption and Compton-reflection effects. The penetrating nature of 2-10 keV X-rays naturally mitigates absorption effects, and this technique should be effective for moderate column densities up to  (2008) and Risaliti et al (2009), respectively. The apparent systematic offsets of the three correlations may be due to differences in SMBH mass estimation and spectral fitting methodology. Taken from Brightman et al (2013). N H ≈ 10 22.5 cm −2 where some other techniques fail (e.g., due to reddening of optical line emission from the Broad Line Region). It may be possible, with broad-band X-ray coverage and in-depth modeling, to recover the intrinsic photon index for even larger values of N H (e.g., Arévalo et al 2014;Puccetti et al 2014). Finally, the Γ − L Bol /L Edd technique, once calibrated, is a direct L Bol /L Edd estimator that does not require intermediate estimation of SMBH mass. In fact, it can be utilized to serve as another SMBH mass-estimation technique via the dependence of L Edd on SMBH mass (e.g., Shemmer et al 2008).

AGN ecology
Ecology refers to the relationship between "organisms" and their physical surroundings. From the point of view of this review, the ecology of AGNs refers to the relationship between the AGN and the host-galaxy environment. 4 Until comparatively recently AGNs were considered an exotic phenomenon with no close connection to their host galaxies. However, over the past 20 years, the identification of tight relationships between the mass of the SMBH and various host-galaxy properties for nearby galaxies (e.g., the bulge luminosity, mass, and velocity dispersion; Kormendy and Richstone 1995; Magorrian et al 1998;Gebhardt et al 2000;Kormendy and Ho 2013) has comprehensively dismissed this view by implying that (1) most (if not all) massive galaxies have hosted AGN activity at some time over the past ≈ 13 Gyr of cosmic evolution and (2) the growth of SMBHs and galaxies is connected, potentially via a link between AGN activity and star formation (the two primary processes whereby SMBHs and galaxies are grown). Since the majority of the growth of SMBHs has occurred at z > ∼ 0.2 (see Section 3.2 and Fig. 7), X-ray surveys of distant AGNs provide key insight into when, where, and how these SMBHs grew, and can shed light on any connections between AGN activity and star formation (when combined with other multiwavelength observations).
In this section we review research on the host galaxies of distant X-ray AGNs and explore if anything "special" is happening in the AGN-hosting galaxies by making comparisons to galaxies that lack AGN activity. We start with a brief overview of the broad-band emission from galaxies and AGNs to illustrate how the properties of an AGN host galaxy can be measured without significant contamination from the AGN itself. To provide the broadest redshift baseline to search for trends in the host-galaxy properties of distant AGNs we often include comparisons to the host-galaxy properties of AGNs in the nearby universe, principally those detected by the Swift-BAT survey (e.g., Tueller et al 2010;Baumgartner et al 2013).

The broad-band emission from galaxies and AGNs
The bulk of the emission from galaxies is produced at UV-submillimeter wavelengths (≈ 0.1-1000 µm; e.g., Kennicutt and Evans 2012) and is primarily due to the radiation produced by populations of stars as well as AGN activity, when present. The intrinsic emission from these stellar populations peaks at UV-NIR wavelengths and corresponds to the (approximate) black-body radiation from stars over a range of masses and ages (Kurucz 1979). This intrinsic emission is typically modified by the presence of dust, particularly in regions of young and forming stars, which are generally optically thick to short-wavelength UV-NIR radiation. Consequently, the emission from young and forming stars is Fig. 11 The median rest-frame UV-MIR SEDs of X-ray AGNs at z ≈ 0-4. The range in X-ray luminosity [in units of log(erg s −1 )] for subsets of the X-ray AGNs are shown in addition to the number of X-ray AGNs used to produce the median SEDs. These SEDs are compared to the mean SED of the low-luminosity quasars from the SDSS, with the host-galaxy contribution removed to indicate the expected SED from a "pure AGN" (gray curve; Richards et al 2006a). The emission at rest-frame ≈ 0.2-4 µm from the X-ray AGNs is predominantly due to starlight from the host galaxy since the majority of the AGNs are obscured or intrinsically weak at optical wavelengths. Taken from Luo et al (2010). often most efficiently measured using infrared observations since the starlight will heat the dust and thermally reradiate the emission at FIR wavelengths (λ ≈ 30-300 µm; typical temperatures of T ≈ 10-100 K). The majority of the emission from galaxies undergoing intense star formation is, therefore, produced at FIR wavelengths, while the majority of the emission from quiescent galaxies (i.e., those with little on-going star formation) is produced at UV-NIR wavelengths.
A significant fraction of the AGNs detected in X-ray surveys are obscured or intrinsically weak at UV-NIR wavelengths; see Section 1.1 and 4.1. While this makes it challenging to determine the properties of the AGN over that band pass, it allows for convenient measurements of the host-galaxy properties without significant contamination from the AGN (e.g., Simmons  can measure at UV-NIR wavelengths are luminosity, color, and morphology. The host-galaxy color provides a basic characterization of the galaxy integrated stellar populations and, when combined with the luminosity, a reliable estimate of the mass of the host galaxy (e.g., Bell and de Jong 2001; Zibetti et al 2009;Conroy 2013). The morphology can provide clues to the formation and evolution of galaxies and, when spatially resolved color information is available, basic constraints on the stellar populations across the galaxy. We explore the host-galaxy masses, colors, and morphologies of X-ray AGNs in Section 5.2 and 5.3.
AGNs are often luminous at infrared wavelengths due to the thermal emission from dust in the vicinity of the accretion disk (e.g., the putative obscuring torus; Antonucci 1993), potentially contaminating infrared-based SFR estimates. However, since the accretion disk is hotter than young stars, the dust is typically heated to higher temperatures (≈ 100-1000 K) and, therefore, the majority of the infrared emission from the AGN is shifted to shorter NIR-MIR wavelengths than that from star formation (≈ 3-30 µm; e.g., Netzer et al 2007;Mullaney et al 2011). Consequently, for many AGN studies, the FIR emission is taken as a measurement of the SFR; however, to provide the most accurate SFR constraints it is necessary to decompose the infrared emission into the AGN and star-formation components, which is essential for reliable SFR measurements from intrinsically luminous AGNs (e.g. . We explore the SFRs of X-ray AGNs in Section 5.5.

Host-galaxy masses and colors
The large numbers of optically obscured and intrinsically weak AGNs detected in Chandra and XMM-Newton surveys provided the first detailed measurements of the host-galaxy properties of distant AGNs without significant contamination from AGN activity at UV-NIR wavelengths. Initial studies noted that the AGN host galaxies were typically luminous ( > ∼ L * ; i.e., the knee of the galaxy luminosity function; Schechter 1976) and have red optical colors, suggesting massive early type systems (Hubble types Sa-E; e.g., Alexander et al 2001Alexander et al , 2002Barger et al 2001bBarger et al , 2002Cowie et al 2001;Severgnini et al 2003;Gandhi et al 2004). More recent studies estimated the host-galaxy stellar masses by fitting the UV-NIR emission with stellar-population models and applied appropriate mass-to-light conversions to the host-galaxy luminosities. These analyses were significantly helped by the launch of Spitzer in 2004, which provided the first extensive rest-frame NIR data for distant sources using the IRAC instrument (3.6-8 µm; Fazio et al 2004). As expected, given the results from the initial studies, the majority of the distant X-ray AGNs were found to be hosted by massive galaxies (> 3 × 10 10 M ; e.g., AGNs are also identified in lower-mass host galaxies (down to ≈ 10 7 -10 9 M ) and are most prevalent in the deepest Chandra surveys (e.g., Shi et al 2008;Brusa et al 2009b;Xue et al 2010Xue et al , 2012 but they always comprise a minority of the AGN population in X-ray surveys. As discussed in Section 5.4 this is more due to challenges in detecting these sources rather than their intrinsic rarity. No strong trend for an increase or decrease in the average stellar mass of X-ray AGNs with redshift is found down to z ≈ 0.2. However, by contrast, the average stellar mass of Swift-BAT selected AGNs at z < 0.05 is found to be substantially lower than the distant X-ray AGNs (≈ 2 × 10 10 M ; e.g., Koss et al 2011). This lower average stellar mass is not obviously due to selection effects since the majority of the sources have X-ray luminosities comparable to the distant AGNs (14-195 keV luminosities of ≈ 10 43 -10 44 erg s −1 ; Tueller et al 2010; Koss et al 2011). Therefore, the differences in the average stellar masses of distant and nearby X-ray AGNs appear to be due to either a genuine decrease in the host-galaxy masses over the narrow redshift range of z ≈ 0.05-0.2 or different approaches in the estimation of the host-galaxy masses.
A common diagnostic to characterize the properties of galaxies is the colormagnitude diagram (CMD), which plots rest-frame optical colors vs. absolute magnitude and provides insight into the integrated stellar populations (e.g., Strateva et al 2001;Baldry et al 2004). The CMD for galaxies is found to be bimodal out to at least z ≈ 1-2 (e.g., Bell et al 2004;Brammer et al 2009;Xue et al 2010), with the majority of galaxies either falling on the "red sequence" or the "blue cloud", which are believed to correspond broadly to quiescent and star-forming galaxies, respectively. Intense star-forming galaxies can also lie on the red sequence due to the presence of dust-obscured star formation rather than quiescent stellar populations (e.g., Cardamone et al 2010;Bongiorno et al 2012;Rosario et al 2013b;Wang et al 2013). Consequently, the CMD is not, in isolation, a reliable indicator of the degree of on-going star formation and other analyses are often required to determine which red-sequence galaxies are quiescent and which are intensely forming stars (e.g., Rosario et al 2013b); see Section 5.1. A modest fraction of the galaxy population lies in a narrow "green valley" between these two dominant regions, which is likely to comprise galaxies with a mix of red and blue stellar populations (e.g., a galaxy with both significant ongoing star-formation and a significant old stellar population) in addition to galaxies transiting from the blue cloud to the red sequence due to the (potentially rapid) shut down of star formation in the host galaxy (e.g., The first CMD analyses of distant X-ray AGNs showed that many lie in the green valley (e.g., Nandra et al 2007;Silverman et al 2008b;Hickox et al 2009), where only a minority of the optically selected galaxy population is found. The distinct difference between the locations of the AGNs and the optically selected galaxy populations in the CMD implies that distant AGNs are found in a subset of the galaxy population and, tentatively, suggests that they are the catalysts for the transition of galaxies from the blue cloud to the red sequence (e.g., through the suppression of star formation via energetic winds, outflows, and jets; see Veilleux et al 2005;Fabian 2012 for reviews). However, later studies showed that clear distinctions between the host-galaxy colors of the AGN and the galaxy populations mostly disappear when the galaxy sample is matched in mass to the AGN sample (e.g., Silverman et al 2009b;Xue et al 2010;Pierce et al 2010b;Rosario et al 2013b), with broadly similar fractions of coeval galaxies and AGNs found in the red sequence, green valley, and blue cloud out to at least z ≈ 3. 5 The lack of significant differences in the host-galaxy colors appears to suggest that, in general, distant X-ray AGNs are not found in "special" host-galaxy environments (at least in terms of the color-mass plane). This general conclusion is in contrast to that found for X-ray AGNs in the nearby universe, where the hosts of Swift-BAT AGNs at z < ∼ 0.05 are found to have bluer colors than the coeval galaxy population (when matched in mass to the AGN sample; e.g., Koss et al 2011), suggesting a connection between the presence of young stars and AGN activity. These apparent disagreements are not necessarily in contradiction since there is evidence that the host galaxies of X-ray AGNs out to z ≈ 1 have experienced more recent star formation than the coeval galaxy population (e.g., from the resolved host-galaxy colors and the depth of the 4000Å break; e.g.

Host-galaxy morphologies
Many of the cosmic X-ray survey fields have extensive coverage at UV-NIR wavelengths from Hubble Space Telescope (HST) observations. The excellent spatial resolution of HST (≈ 0.1 ) allows moderately detailed characterization of the rest-frame optical host-galaxy properties of distant AGNs through kpc-scale measurements of the host-galaxy morphology and the identification of potential triggers of AGN activity (e.g., galaxy merger and galaxy interaction signatures). A number of different approaches for measuring host-galaxy morphologies have been developed (e.g., automated classifications based on the distribution of light in the galaxy, two-dimensional fits to the surface brightness profiles to provide disk/bulge decompositions, and human "eyeball" classification), which appear to give broadly similar results (e.g., Huertas-Company et al 2014). The contribution of the AGN to the optical emission can significantly bias morphological measurements (e.g., Simmons and Urry 2008;Pierce et al 2010b) and, therefore, the majority of the studies focus on optically obscured or intrinsically weak AGNs.  Fig. 12 for a comparison of the morphologies of AGNs and mass-matched galaxies at 1.5 < z < 2.5. The first studies found that AGNs typically reside in more bulge-dominated systems than the coeval galaxy population (e.g., Grogin et al 2005;Pierce et al 2007). However, as with the CMD analyses, clear differences mostly disappeared when the galaxy sample was matched in mass to the AGN sample (e.g., Kocevski et al 2012;Böhm et al 2013;Fan et al 2014;Villforth et al 2014); we note that recent evidence from a systematic morphological analysis of AGNs and mass-matched galaxies over the broad redshift range of z = 0.5-2.5 has found evidence that AGNs at z ≈ 1 are preferentially hosted in disk-dominated galaxies when compared to the galaxy population, although the differences are comparatively subtle . More significant morphological differences are found by the present day, with Swift-BAT AGNs ≈ 2 times more likely to reside in spiral (disk dominated) galaxies than comparably massive inactive galaxies (e.g., Koss et al 2011), suggesting that not all host-galaxy environments are capable of hosting significant AGN activity by z ≈ 0.
How is the AGN activity triggered? Prior to the launches of Chandra and XMM-Newton, it was widely predicted that distant AGNs are triggered by gas-rich major mergers, violent events where two similar-mass galaxies interact and merge, torquing the gas and driving it toward the central SMBH, where it can be accreted (e.g., Sanders et al 1988;Barnes and Hernquist 1992;Mihos and Hernquist 1996;Hopkins et al 2008). Contrary to these expectations, only a minority ( < ∼ 20%) of the AGN population over the broad redshift range of z ≈ 0.2-2.5 were found to have the clear signatures expected for major mergers (disturbed host-galaxy morphologies and tidal tails; Silverman et al 2011;Cisternas et al 2011;Kocevski et al 2012;Böhm et al 2013;Villforth et al 2014). The fraction of Swift-BAT AGNs at z < 0.05 with major-merger signatures is also ≈ 20%, consistent with that found for the distant AGNs Cotini et al 2013). These results, therefore, suggest that other processes such as galaxy interactions, minor mergers, and secular processes (e.g., galaxy bars, disk instabilities, and clumpy cloud accretion) may be responsible for triggering the majority of AGN activity (e.g., Silverman et al 2011;Cisternas et al 2011Cisternas et al , 2014Bournaud et al 2012;Kocevski et al 2012;Böhm et al 2013;Cheung et al 2014;Trump et al 2014;Villforth et al 2014). However, we note that there is evidence that the most luminous AGNs are preferentially triggered by major mergers, as also found for the most powerful star-forming galaxies (e.g., Kartaltepe et al 2012), which could indicate that major mergers are required to drive sufficient quantities of gas into the central regions of galaxies to power the most luminous systems.
In general, the fraction of the coeval galaxy population with major merger signatures is comparable to that found for the (majority of) distant AGNs, when the galaxy sample is matched in mass to the AGN sample; however, see Rosario et al (2014) for tentative evidence of a factor ≈ 2 decrease in the fraction of z = 0.5-1.0 galaxies with major-merger signatures when compared to the coeval AGN population. More significant differences are found by the present day, with a ≈ 5 times lower fraction of the galaxy population hosted in major mergers when compared to Swift-BAT selected AGNs (≈ 4% vs. ≈ 20%; Cotini et al 2013). The emerging picture, therefore, suggests that, while the absolute fraction of moderate-luminosity AGNs with major merger signatures is comparatively constant at ≈ 20% over the broad redshift range of z ≈ 0-3, there are significant differences between the major-merger fraction of the AGN and coeval galaxy populations by z ≈ 0. As we discuss in Section 5.5, these differences may be related to the greater availability of a cold-gas supply in the distant universe when compared to the local universe (i.e., the ubiquity of gas-rich galaxies may be a key factor).

AGN fraction and Eddington-ratio distribution
On the basis of the suite of studies explored in the previous sections, there are few significant differences between the host-galaxy properties of distant X-ray AGNs and the coeval galaxy population, when the samples are matched in mass. More significant differences are evident by z ≈ 0 and some differences may already be in place by z ≈ 1. However, to first order, these results suggest that any galaxy with similar properties to the AGN host galaxies is also capable of hosting an AGN and, therefore, the fraction of galaxies with AGN activity provides a basic measurement of the AGN duty cycle (i.e., how often mass accretion onto the SMBH switches on and off). Fig. 13 The fraction of galaxies hosting an AGN as a function of stellar mass for different redshift and X-ray luminosity ranges [as labeled; luminosity ranges in log(erg s −1 )]. A strong stellar-mass dependence on the AGN fraction is found for all redshifts. Adapted from Xue et al (2010).
As revealed from a large number of studies to date, the fraction of galaxies hosting AGN activity above a fixed X-ray luminosity threshold rises steeply with host-galaxy mass out to at least z ≈ 2-3 (e.g., Bundy et al 2008;Brusa et al 2009b;Xue et al 2010;Georgakakis et al 2011;Bluck et al 2011;Aird et al 2012;Bongiorno et al 2012;Mullaney et al 2012b). See Fig. 13 for the AGN fraction as a function of stellar mass across different redshift and X-ray luminosity ranges. For example, the fraction of z ≈ 1 galaxies hosting AGN activity with L X > 10 43 erg s −1 increases from ≈ 0.3% at a stellar mass of ≈ 10 10 M to ≈ 3% at a stellar mass of ≈ 10 11 M ; the AGN fraction globally increases by a further factor of ≈ 5 for a lower luminosity threshold of L X > 10 42 erg s −1 . The AGN fraction does not appear to have a strong dependence on host-galaxy color (e.g., Xue et al 2010;Georgakakis et al 2011), although it is a strong function of SFR; see Section 5.5. There are no comparably detailed analyses for X-ray AGNs in the nearby universe; however, on the basis of optically selected AGNs from the SDSS (e.g., Kauffmann et al 2003;Best et al 2005), the AGN fraction is found to rise with increasing stellar mass but then flattens out at masses of > ∼ 10 11 M . No statistically significant evidence for a flattening in the X-ray AGN fraction at high stellar masses is found for distant AGNs, leaving some uncertainty over whether this result is specific to optically selected AGNs or whether a significant decrease in the duty cycle of X-ray AGN activity has occurred in the most massive galaxies at z < 0.2 that is not seen at z > 0.2. It is also possible that the observational signature for a decrease in the duty cycle of X-ray AGN activity at high stellar masses is too subtle to identify in the current studies.
Overall these results suggest that AGNs are more common in massive galaxies than less massive galaxies, suggesting that the SMBHs in massive galaxies are growing more rapidly than the SMBHs in less massive galaxies. However, we must be careful in our physical interpretation of this result as there are strong biases against detecting AGNs at a fixed luminosity threshold in lower-mass galaxies than higher-mass galaxies since a lower-mass SMBH would need to be accreting at a higher Eddington ratio than a higher-mass SMBH to be detected. 6 To directly explore whether there is a mass dependence to the growth rates of SMBHs requires measuring the Eddington ratios of SMBHs across a broad range in mass.
The calculation of an Eddington ratio (referred to here as λ Edd ) relies on knowing a number of uncertain quantities, including the SMBH mass and the bolometric AGN luminosity, which typically has to be indirectly estimated from a single-wavelength measurement of the AGN luminosity (e.g., from the X-ray band, which represents only a few percent of the bolometric luminosity; Elvis et al 1994;Marconi et al 2004;Hopkins et al 2007;Hao et al 2014). 7 Consequently, it is challenging to derive accurate Eddington ratios, particularly for large numbers of distant AGNs where the majority of the sources lack direct SMBH mass measurements and indirect methods are required to provide mass constraints (e.g., a proxy for the SMBH mass such as the hostgalaxy mass, luminosity, velocity dispersion, or bulge luminosity). To remove the uncertainty on the SMBH mass, an approach adopted by some researchers is to calculate the "specific accretion rate", where the SMBH mass is replaced with the stellar mass, which is a more directly measured quantity (e.g., Brusa et al 2009b;Aird et al 2012;Bongiorno et al 2012). In this review we will refer to Eddington ratios but we caution that this is not a directly measured quantity and any Eddington-ratio measurements are subject to considerable (factor of a few) systematic uncertainties.
The Eddington ratios estimated for distant X-ray AGNs cover a broad range: λ Edd ≈ 10 −5 -1 for implied SMBH masses of ≈ 10 6 -10 9 M , with the majority at λ Edd ≈ 10 −4 -10 −1 (e.g., Babić et al 2007;Ballo et al 2007;Rovilos and Georgantopoulos 2007;Alonso-Herrero et al 2008;Hickox et al 2009;Raimundo et al 2010;Simmons et al 2011;Trump et al 2011a;Lusso et al 2012;Nobuta et al 2012;Matsuoka et al 2013). The distribution of Eddington ratios implied from these studies is, at least partially, dictated by limitations in the data (i.e., incompleteness effects such as the X-ray sensitivity limits and the X-ray luminosity threshold adopted to identify AGN activity in these studies) and, consequently, does not provide a reliable measurement of the intrinsic Eddington-ratio distribution. The first studies to correct for these limitations and construct an intrinsic Eddington-ratio distribution revealed striking results: the Eddington-ratio distribution can be characterized by a power law with a slope that is independent of both host-galaxy mass and redshift out to z ≈ 2-3 (Aird et al 2012(Aird et al , 2013aBongiorno et al 2012); however, we note that the current data would also be consistent with a broad log-normal distribu-6 Here it is assumed that higher-mass galaxies have more massive SMBHs than lower-mass galaxies, which is reasonable since (1) there is a broad relationship between host-galaxy mass and SMBH mass and (2) any evolution in the stellar-SMBH mass relationship with redshift appears to be modest (e.g., Jahnke et al 2009;Bennert et al 2011;. 7 Ideally the bolometric luminosity would be directly measured from the primary AGN continuum over the optical-X-ray waveband. However, it is expected to peak at far-UV wavelengths, which is unobservable due to absorption from the Galaxy. See, for example, Vasudevan and Fabian (2009) and Jin et al (2012b) for some observational approaches to estimating the primary AGN continuum of nearby AGNs. tion, if the peak of the distribution lies below current sensitivity limits. The slope of the power law is steep and corresponds to a ≈ 5-10 increase in the probability to detect an SMBH that is accreting at a ≈ 10 times lower Eddington ratio. See Fig. 14 for the derived Eddington-ratio distribution for a range of stellar masses at 0.2 < z < 1.0. A power-law Eddington-ratio distribution is qualitatively consistent with the latest constraints for distant optically selected quasars (e.g., Schulze and Wisotzki 2010;Shen and Kelly 2012;Kelly and Shen 2013); however, more quantitative comparisons are limited by the different approaches adopted in estimating SMBH masses between the X-ray studies (scaled from the stellar mass) and the optical quasar studies (virial SMBH mass) and the different AGN selection approaches. No comparably detailed measurements of the Eddington-ratio distribution have been produced for X-ray AGNs in the nearby universe. Nevertheless, on the basis of optically selected AGNs from the SDSS, Kauffmann and Heckman (2009) have argued that there are two regimes of growth in nearby AGNs: for systems with significant star formation the Eddington-ratio distribution is characterized by a broad log-normal distribution while for quiescent systems with little or no star formation the Eddington-ratio distribution is characterized by a power-law distribution. Further work is required to understand the differences between these results and those found for the distant X-ray AGNs.
The fraction of galaxies hosting AGN activity at a given Eddington ratio is found to increase with redshift out to at least z ≈ 2-3 (e.g., Aird et al 2012;Bongiorno et al 2012); i.e., the normalization of the Eddington-ratio distribution increases with redshift. The current constraints suggest that this redshift dependence is strong [≈ (1 + z) 3.5−4.0 ; Aird et al 2012; Bongiorno et al 2012], indicating that AGN activity was about an order of magnitude more common in galaxies at z ≈ 1 for a fixed Eddington ratio than at z ≈ 0; however, it is not clear whether this evolution is independent of mass across all redshifts. This strong redshift evolution is consistent with the evolution in SFR found for star-forming galaxies (see Section 5.5).
What insight do these results provide on "AGN downsizing" (see Section 3.2)? The strong increase in the AGN fraction with redshift can be explained by either (1) an increase in the duty cycle of AGN activity at a given Eddington ratio or (2) an increase in the characteristic Eddington ratio with redshift. The majority of the theoretical models predict the second scenario as being a major driver of AGN downsizing (see Section 3.2); however, since the observed Eddington-ratio distribution is a power law it is not possible to distinguish between these different scenarios with the current data (i.e., there are no distinctive features to measure a shift in the characteristic Eddington ratio). Many of the theoretical models also predict a redshift dependence in the characteristic mass of accreting SMBHs, which is not clearly observed but may be due to limitations in the current data (see Section 6.1.1).
Fig. 14 Eddington-ratio distribution (i.e., the probability that a galaxy will host an AGN at a given Eddington ratio) for different ranges in stellar mass for galaxies at 0.2 < z < 1.0. The data have been corrected for X-ray sensitivity incompleteness effects to provide an accurate measurement of the intrinsic Eddington-ratio distribution. The final panel compares the observed Eddington-ratio distribution (gray histogram) to the intrinsic Eddington-ratio distribution measured across the full range in mass and demonstrates the need to account for the effects of incompleteness in the X-ray data. The dashed line is a best-fit power-law model to the Eddington-ratio distribution, evaluated at z = 0.6. The Eddington-ratio distribution is consistent with being independent of stellar mass over 9.5 < log(M * /M ) < 12.0 at 0.2 < z < 1.0. Taken from Aird et al (2012).

Star formation and specific star formation rates
The suite of studies explored in this review have revealed when, where, and how SMBHs have grown in the distant universe. However, we have not yet investigated how the growth of the SMBH relates to the growth of the host galaxy; i.e., the connection between AGN activity and star formation. We would expect at least a broad connection between AGN activity and star formation since (1) the volume averaged star formation and SMBH mass accretion rates track each other (with a 3-4 orders of magnitude offset) out to at least z ≈ 2 (e.g., Heckman et al 2004;Merloni et al 2004;Silverman et al 2008a;Aird et al 2010;Mullaney et al 2012a) and (2) there is a tight relationship between the SMBH mass and spheroid mass for galaxies in the nearby universe (e.g., Kormendy and Richstone 1995;Magorrian et al 1998;Gebhardt et al 2000;Kormendy and Ho 2013), which provides "archaeological" evidence for past joint SMBH-galaxy growth. However, these results only afford broad integrated constraints on the overall SMBH-galaxy growth and do not provide clear clues on how the SMBH and galaxy have grown in individual systems, which requires more direct SFR measurements.
The first studies to constrain the SFRs of distant X-ray AGNs used a broad variety of star-formation indicators, including optical spectroscopy, MIR data, and submillimeter (submm) observations (e.g., Alexander et al 2005b;Polletta et al 2007;Silverman et al 2009b;Trichas et al 2009;Mullaney et al 2010;Lutz et al 2010;Xue et al 2010;Rafferty et al 2011). These initial studies showed that AGNs of a fixed X-ray luminosity can have a broad range of SFRs (up to ≈ 5 orders of magnitude variation between individual sources; see Fig. 14 in Rafferty et al 2011) and provided evidence that the average SFRs of AGNs increase with redshift. However, significant uncertainties remained in the SFR measurements, such as (1) potential contamination from AGN activity to the SFR estimates, (2) potential underestimation of the SFR due to obscuration by dust, and (3) uncertain extrapolation from the observed wavelength to the total SFR. These issues are best addressed by FIR observations, which trace the peak of the star-formation emission with less contamination from AGN activity; see Section 5.1. Consequently, the launch of the Herschel observatory in 2009, the first observatory with high sensitivity at FIR wavelengths (six photometric bands over ≈ 70-500 µm; Pilbratt et al 2010), offered the potential to make the first accurate SFR measurements of distant AGNs (see Lutz 2014 for a recent review of Herschel survey results).
On the basis of a large number of studies using Herschel data, it is now abundantly clear that the average SFRs of X-ray AGNs increase strongly with redshift out to at least z ≈ 3 (e.g., Shao et al 2010;Harrison et al 2012;Mullaney et al 2012b;Rosario et al 2012Rosario et al , 2013bRovilos et al 2012;Santini et al 2012), confirming the trend found from the first SFR studies. See Fig. 15a for an example of the average SFR as a function of redshift and AGN luminosity. The majority of the studies used stacking analyses to provide average FIR constraints since only a modest fraction of the X-ray AGNs are detected by Herschel ; however, these results are also broadly reproduced using more detailed SED fitting analyses taking account of photometric upper limits and calculating average constraints using survival analysis techniques (F. Stanley et al. in preparation). 8 In general, most studies showed no clear evidence for any luminosity dependence on the average SFR for moderate-luminosity AGNs (L X = 10 42 -10 44 erg s −1 ) and the average SFR was found to be broadly constant over this luminosity range, at any given redshift; however, we note that some X-ray luminosity dependence on the average SFR is often seen at z < ∼ 1 and is most prominent at z ≈ 0 (see Fig. 15).
By contrast, a broad range of results have been found for high-luminosity AGNs (L X > ∼ 10 44 erg s −1 ) at z > ∼ 1, with researchers arguing that either the average SFR increases with both redshift and X-ray luminosity, increases only with redshift (following the trend seen for the moderate-luminosity AGNs), or decreases with X-ray luminosity (e.g., Harrison et al 2012;Page et al 2012;Fig. 15 (a) Median 60 µm (FIR) luminosity vs. AGN luminosity for X-ray AGNs over z ≈ 0-2.5 (as labeled), showing the observed relationship between AGN activity and star formation. The solid curve shows the two-component functional fit to the local AGN data from Swift-BAT, the dotted line is the expected FIR luminosity for typical star-forming galaxies at z = 2 (see also panel b), the dashed line is the AGN-star formation luminosity relationship for AGN-dominated systems from Netzer (2009), and the shaded region corresponds to the estimated 1 σ range for an AGN SED (see Section 3.1 of Rosario et al 2012). L AGN corresponds to the bolometric AGN luminosity: L 2−10keV = 10 42 erg s −1 and L 2−10keV = 10 44 erg s −1 correspond to L AGN = 5.7 × 10 42 erg s −1 and L AGN = 3.4 × 10 45 erg s −1 , respectively; (b) Average sSFR vs. redshift for X-ray AGNs with L X = 10 42 -10 44 erg s −1 (as labeled) over z = 0.5-3. The AGNs are compared to FIR-detected star-forming galaxies not hosting AGN activity (non AGNs) and the tracks trace the evolution in sSFR found for star-forming galaxies with redshift, as defined by Pannella et al (2009) and Elbaz et al (2011). Overall, the X-ray AGNs broadly trace the evolution in SFR and sSFR found for the star-forming galaxy population. However, the observed relationship between the AGN and star-formation luminosity is complex and is probably, at least partially, driven by the different timescales of stability between star formation and AGN activity; see Section 5.5. Adapted from Mullaney et al (2012b) and Rosario et al (2012). Rosario et al 2012Rosario et al , 2013bRovilos et al 2012;Santini et al 2012). Some of the variation between the different results for high-luminosity AGNs is due to two practical factors: (1) the SFRs of luminous AGNs are more difficult to measure reliably since the AGN can contribute significantly to the FIR emission and (2) luminous AGNs are less common than moderate-luminosity AGNs, limiting the statistical power of studies restricted to small-area fields. Indeed, the studies performed in large-area fields with good source statistics all find that the average SFR of luminous AGNs is either constant with X-ray luminosity (extending the trend seen for the moderate-luminosity AGNs) or rises with X-ray luminosity, with the change from a rising trend to a flat trend found to be a function of redshift (e.g., Harrison et al 2012;Rosario et al 2012; see Fig. 15.
The strong increase in the average SFR for the X-ray AGNs with redshift tracks the increase seen in the overall star-forming galaxy population [≈ (1 + z) 4 ; e.g., Daddi et al 2007b;Noeske et al 2007;Rodighiero et al 2010;Elbaz et al 2011]. The increase in SFR with redshift for star-forming galaxies is found to be independent of galaxy mass, such that the specific star formation rate (sSFR; the ratio of stellar mass to SFR) evolves strongly with redshift across all stellar masses, and is thought to be driven by the availability of a cold-gas supply (i.e., the distant galaxies are more gas rich than the nearby galaxies; e.g., Daddi et al 2010;Genzel et al 2010;Tacconi et al 2013). The tightness of the sSFR at any given redshift and the lack of a strong mass dependence is often referred to as the "main sequence" of star formation. The average SFRs and sSFRs of the X-ray AGNs are in good quantitative agreement with those found for the star-forming galaxy population (e.g., Xue et al 2010;Mainieri et al 2011;Mullaney et al 2012b;Rosario et al 2012Rosario et al , 2013b, suggesting that the same factors that drive star formation also drive AGN activity. See Fig. 15b for a comparison of sSFR between X-ray AGNs and star-forming galaxies over z = 0.5-3. This connection between AGN activity and star formation is further strengthened by the following additional results: 1. The average SFRs of distant X-ray AGNs are systematically higher than the overall galaxy population, which includes quiescent galaxies in addition to star-forming galaxies (e.g., Santini et al 2012;Symeonidis et al 2013;Vito et al 2014b), demonstrating that AGN activity is more closely connected to star-forming galaxies than the general galaxy population. We note that AGNs are also detected in quiescent galaxies but they appear to comprise a minority of the population (e.g., These results imply a general connection between AGN activity and star formation. However, the observed relationship between AGN and star-formation luminosity does not clearly indicate a correlation, particularly at moderate AGN luminosities where the average SFR is flat across at least 2 orders of magnitude in AGN luminosity; see Fig. 15a. As shown by several models (e.g., Gabor and Bournaud 2013;Hickox et al 2014;Neistein and Netzer 2014;Thacker et al 2014), these apparently uncorrelated data can be explained by differences in the timescales of stability between star formation and AGN activity: star formation is assumed to be relatively stable over long periods (of order ≈ 100 Myr) while AGNs are assumed to vary significantly on short timescales ( < ∼ Myrs). As an example, the Hickox et al (2014) model assumes that the long-term average of AGN activity and star formation is constant (i.e., when averaged over many duty cycles of mass accretion) but allows the observed X-ray luminosity to vary by orders of magnitude on short timescales on the basis of an assumed Eddington-ratio distribution. This model, and the other models referenced above, reproduce the broad trends seen between X-ray luminosity and average SFR and demonstrate how short-term variations can significantly disguise an underlying correlation over longer timescales. It, therefore, seems likely that the general hypothesis that AGN activity varies significantly on short timescales while the star formation is comparatively stable is broadly correct. However, it is not yet clear which model provides the best physical description of the observed trends between X-ray luminosity and average SFR, and further observational diagnostics beyond a simple average SFR (e.g., the distribution of SFRs as a function of X-ray luminosity, the fraction of quiescent galaxies hosting AGNs as a function of X-ray luminosity, the stellar-mass dependence) are required to provide greater diagnostic power.
We finally conclude our discussion of the connection between AGN activity and star formation by considering whether AGNs have significant impact on the star formation in the host galaxy (e.g., by driving energetic winds, outflows, and jets, commonly referred to as "AGN feedback"; see Veilleux et al 2005;Fabian 2012 for reviews). On the basis of many AGN-feedback models we would expect a decrease (i.e., a suppression) in the SFR of X-ray AGNs when compared to the overall galaxy population, particularly at the highest X-ray luminosities. However, depending on how quickly the star formation is suppressed, the signatures of suppressed star formation could be comparatively subtle, particularly when averaged over AGN populations (e.g., as a function of X-ray luminosity; see Section 4 of Harrison et al 2012 for a further discussion of some potential limitations). More sensitive SFR constraints for individual systems, in addition to the measurement of sensitive SFR distributions as a function of, e.g., X-ray luminosity and redshift, are required to provide more sensitive tests of the impact of AGN activity on star formation. 5.6 The cosmic balance of power: SMBH mass accretion vs. stellar radiation How much does SMBH mass accretion contribute to the balance of power in the cosmos? As mentioned in Section 2.1 and 3.3, AGNs dominate the CXRB. However, the CXRB only comprises a minority of the overall cosmic background radiation, which has broadly equal contributions from the cosmic UV-optical background and the cosmic infrared background (CIRB; Hauser and Dwek 2001), when the strongly dominant cosmic microwave background is excluded (i.e., the relic emission from the Big Bang; e.g., Penzias and Wilson 1965).
The cosmic UV-optical background is dominated by stellar emission (e.g., Madau 1992;Totani et al 2001;Finke et al 2010) but, prior to the launch of Chandra and XMM-Newton, it was predicted that AGNs may contribute several tens of percent of the CIRB (e.g., Almaini et al 1999;Fabian and Iwasawa 1999;Gunn and Shanks 1999). These models typically assumed that powerful obscured quasars would contribute significantly to the CIRB through the re-radiation of the dust-obscured AGN emission at infrared wavelengths. However, direct observational measurements from the AGNs detected in the Chandra and XMM-Newton surveys showed that the contribution to the CIRB from AGNs is much more modest at ≈ 5-10% (even before subtracting contributions from star formation to the infrared emission from the X-ray AGN host galaxies), with the majority of the CIRB produced by star formation (e.g., Elbaz et al 2002;Fadda et al 2002;Silva et al 2004;Treister and Urry 2006;Ballantyne and Papovich 2007). Therefore, the cosmic background emission since the formation of galaxies is dominated by stellar radiation processes rather than mass accretion onto SMBHs. Part of the origin for the discrepancy between the original predictions and the observations is the assumed AGN evolution (i.e., the original models did not anticipate the "AGN downsizing" results; see Section 3.2) and the assumed contribution from Compton-thick AGNs to the CIRB, a component that is still relatively poorly constrained but is unlikely to fundamentally change these broad conclusions.

Future prospects
In this review we have highlighted how cosmic X-ray surveys of distant AGNs have provided key insight into the demographics, physics, and ecology of growing SMBHs. Thanks to the revolutionary X-ray facilities of Chandra, XMM-Newton, and NuSTAR, we now have a dramatically improved picture of how SMBHs grew through cosmic time, their accretion and obscuration physics, and the connection between SMBHs, their host galaxies and the larger scale environment. However, despite these great advances, many fundamental questions remain unanswered. In this final section we focus on some of the key remaining big questions and discuss how current and future facilities can be used to address them over the coming decades.

What drives AGN downsizing?
Excellent progress has been made over the past decade in measuring the space density and luminosity density evolution of X-ray AGNs, and it is now clear that lower-luminosity AGNs peaked at lower redshifts than higher-luminosity AGNs, often referred to as "AGN downsizing"; see Section 3.2 and Fig. 7. However, what is not yet clear from the observational data are the factors that drive this luminosity dependence. Theoretical models broadly predict that this behavior is driven by a decrease in the characteristic Eddington ratio and/or a decrease in the characteristic active SMBH mass with decreasing redshift. The current observational constraints (e.g., Aird et al 2012;Bongiorno et al 2012) suggest a decrease in the characteristic Eddington ratio with decreasing redshift but they do not find any clear SMBH mass dependences (modulo that the stellar mass is often used as a proxy for the SMBH mass in these studies); see Section 5.4. However, on the basis of optical studies, at z < ∼ 0.2 there is clear evidence for a strong mass dependence on the volume-average growth rates of SMBHs, where the growth times of the most massive SMBHs are orders of magnitude longer than those of lower-mass SMBHs, indicating that the most massive SMBHs must have been growing more rapidly in the past (e.g., Heckman et al 2004). Testing this result with X-rays is important to verify that the mass dependence of SMBH growth is not specific to optical studies and will require sensitive X-ray data and good source statistics at both z < ∼ 0.2 and z > ∼ 0.2. Connecting our picture of SMBH growth at z < ∼ 0.2 to the emerging picture of SMBH growth at z > ∼ 0.2 will be key to a greater understanding of AGN downsizing and the cosmological growth of SMBHs.

How did the first SMBHs form and grow?
There are several theoretical models for the formation of the first SMBHs (remnants of population III stars; direct collapse of primordial gas clouds; BH merging in dense stellar clusters; e.g., see Volonteri 2010 for a recent review), which predict different initial SMBH masses (typically referred to as "seed" SMBHs) of ≈ 100-100,000 M . Since the mass of a SMBH dictates the maximum luminosity that can be produced through accretion (i.e., the Eddington luminosity), high-redshift XLFs can help distinguish between these different formation scenarios and, from the evolution of the XLFs, constrain the growth of the first SMBHs. The small bias against absorption makes X-rays a powerful probe of accretion in the early growth of SMBHs (e.g., rest-frame 14-70 keV at z > 6 for observed-frame 2-10 keV), particularly since the early SMBH growth phases were likely to be heavily absorbed [as may be expected given the gas-rich environment and small physical sizes of the first SMBHs (M. Volonteri et al. in preparation) and implied by the redshift-dependence of X-ray absorption; see Section 4.1]. One clear observational signature of early SMBH growth is "cosmic upsizing" (e.g., Ueda et al 2014), which would be revealed by a change in the relative ratio between the number density of highluminosity and lower-luminosity AGNs (i.e., a change in the shape of the XLF) at z > ∼ 3-4; see Section 3.2. To model accurately the early growth of SMBHs it will be necessary to construct XLFs in several redshift bins down to moderate luminosities over z ≈ 4-8, requiring excellent high-redshift source statistics. The current XLF constraints are weak at z > ∼ 4, which is due to the intrinsic rarity of such AGNs as well as source identification challenges (see Section 6.2).
Constraining the evolution of the highest-redshift AGNs is a primary goal of several future X-ray observatories (see Section 6.4), although deep Chandra surveys over large areas, in combination with ambitious targeted follow-up programs (see Section 6.3), can better constrain the faint end of the z > ∼ 4 XLF.
6.1.3 How many obscured AGNs are missed in the current X-ray surveys?
X-ray surveys arguably provide the most efficient selection of obscured AGNs, particularly at high energies and at high redshifts where the rest-frame energies allow for penetration of large absorbing column densities (up to N H ≈ 10 24 cm −2 ). However, many of the most highly obscured AGNs will be missed in the current X-ray surveys, restricting our census of the overall AGN population; see Section 3.3. The identification of the most highly obscured AGNs could be more than just a "book keeping" exercise since they may reside in qualitatively different environments than less obscured AGNs (e.g., in galaxy major mergers and the most intense starbursts, where there is potentially more gas to obscure the AGN) and, therefore, may evolve differently, possibly modifying results on the fraction of obscured AGNs with redshift and luminosity; see Section 4.1. The majority of the current studies are based around Chandra and XMM-Newton surveys, but significant progress will be made using NuSTAR, where the higher-energy sensitivity provides a "cleaner" selection of AGNs with less absorption bias, particularly at z < ∼ 1 where the rest-frame energies probed by Chandra and XMM-Newton are modest. On longer timescales Athena will also provide greatly improved identification and characterization of heavily obscured AGNs from X-ray spectral fitting (i.e., accurate N H and reflection measurements; see Section 6.4). Multiwavelength observations will also allow for the identification of X-ray undetected AGNs that produce luminous emission at other wavelengths (e.g., at infrared, radio, and optical wavelengths, when the contaminating emission from the host galaxy is reliably accounted for using SED decomposition and/or spectroscopy) and will become very powerful when the James Webb Space Telescope (JWST), the successor to HST , is launched (see Section 6.3).
6.1.4 What causes the dependence of α ox on optical/UV luminosity, and why are there intrinsically X-ray weak outliers?
Although the basic dependence of α ox upon optical/UV luminosity has been known for about three decades, recent studies have substantially improved measurements of the form of this relation (see Section 4.2). This being said, further improvement is still needed since the quantitative results of some recent studies disagree by considerably more than their statistical uncertainties. Future work must aim to reduce and realistically assess the inevitable systematic errors that enter such analyses (e.g., AGN variability effects, detection-fraction effects, absorption effects, host-galaxy light contamination, and AGN misclassification). The effects of additional physical parameters, particularly Edding-ton ratio, also need better investigation. Furthermore, the few outstanding claims for a significant dependence of α ox upon redshift need checking given the general consensus against a measurable redshift dependence (e.g., via reanalysis of the original data used to claim redshift dependence). At the same time as the observational situation is advanced, improved numerical simulations of the SMBH disk-corona system are required so that expectations for the behavior of the X-ray-to-optical/UV SED, including the basic cause of the α ox -log(L 2500Å ) relation, can be derived from first-principles physics.
Additionally, outliers from the α ox -log(L 2500Å ) relation that appear to be intrinsically X-ray weak need further investigation (see Section 3.3). While these objects seem sufficiently rare that they should not affect AGN demographic studies materially, they may nevertheless provide novel insights (cf. Eddington 1922) into the SMBH disk-corona system and emission-line regions. For example, some X-ray weak outliers may be systems with extremely high Eddington ratios where radiation-trapping effects largely prevent X-ray emission from escaping the accretion flow. Alternatively, perhaps these outliers are not truly intrinsically X-ray weak, and we have simply been tricked by a complex absorption scenario that is not yet properly understood or appreciated.

What host-galaxy environments are conducive to AGN activity?
A somewhat surprising result is how inconspicuous the host galaxies of distant X-ray AGNs are when compared to galaxies not hosting AGN activity. There are no clear differences in the host-galaxy colors and morphologies (broad-scale galactic structure and galaxy merger/interaction signatures) of X-ray AGNs and galaxies when matched in stellar mass (see Sections 5. 2 and 5.3), at least at z > ∼ 1 (there is evidence for some differences at z < ∼ 1). The star-formation properties of distant X-ray AGNs also appear to be broadly similar to those of distant star-forming galaxies and elevated when compared to the overall galaxy population (see Section 5.5), suggesting a general connection between AGN activity and star formation. Clearer differences between the host galaxies of X-ray AGNs and galaxies are more evident by the present day, when it appears that X-ray AGNs reside in a subset of the overall galaxy population. However, it is not yet clear when and how these differences arose, primarily because no study has yet uniformly measured the host-galaxy properties of both nearby and distant systems, as required to remove any potential differences between the methods used in the broad suite of studies published to date. From the point of view of the star-formation properties, deeper infrared-mm data are required to provide more powerful tests than a comparison of average SFRs (see Section 6.3.2); for example, a comparison of the SFR distributions between AGNs and galaxies and improved measurements on the fraction of AGNs in quiescent galaxies.

How does large-scale environment affect AGN activity?
Large-scale environment appears to have a significant affect on AGN activity, as demonstrated, for example, from the increased fraction (decreased fraction) of galaxies hosting AGN activity in protoclusters (rich galaxy clusters) versus the field; see Section 3.5. The availability of a cold-gas supply in the < ∼ pc vicinity of the SMBH is presumably the essential requirement for mass accretion but these results suggest that the gas availability may be controlled by the large-scale environment. For example, there is no lack of gas in the most massive galaxy clusters, but it is mostly in a hot form rather than in the cold form that can be easily accreted by SMBHs. Protoclusters and rich galaxy clusters represent the most extreme high-density environments but to provide a fully coherent picture of the role of large-scale environment on the growth of SMBHs requires measurements of AGN activity across all environments as a function of redshift (e.g., voids; field; groups; poor clusters; rich clusters; protoclusters). To achieve this aim requires X-ray observations covering a large enough cosmic volume to detect significant numbers of X-ray sources across the full range of large-scale environments down to X-ray sensitivity limits sufficient to detect the majority of the SMBH growth. With sufficient data it would then be possible to construct XLFs and explore the host-galaxy properties as a function of environment and redshift. Improved measurements of the clustering of AGNs as a function of physical parameters (e.g., AGN type, AGN luminosity, redshift) will further reveal how the dark-matter halo affects the growth of SMBHs, particularly for HoD analyses, which require excellent source statistics to more accurately constrain the fractions of X-ray AGNs in the satellite and central galaxies within halos (see Section 3.5). Good progress can be made with current facilities (see Section 6.2) while the large FOVs of future planned and proposed X-ray observatories will make them ideally suited to addressing this key question (see Section 6.4).

Additional targeted X-ray surveys with operating missions
The Chandra, XMM-Newton, and NuSTAR observatories are performing well and, subject to funding considerations, have the capability to undertake productive observations of the cosmos for at least the next ≈ 5 yr. Having operated successfully over the last ≈ 15 yr, Chandra and XMM-Newton have already covered much of the accessible flux-solid angle plane, from deep pencil-beam surveys to shallow wide-area surveys; see Fig. 3. Considering all of the Chandra and XMM-Newton surveys listed in Table 1, there are an estimated ≈ 500,000 unique X-ray sources detected over ≈ 1,000 deg 2 of the sky. However, the source statistics and areal coverage are strongly dominated by the serendipitous surveys, which are non contiguous and typically have limited spectroscopic coverage; see Section 6.3 for future large-area/all sky multiwavelength survey plans. The blank-field surveys comprise of order ≈ 45,000 unique X-ray sources detected over ≈ 80 deg 2 , of which about half are from the on-going, and currently unpublished, XMM-Newton XXL survey.
The current suite of Chandra and XMM-Newton surveys with good spectroscopic completeness have covered a broad swathe of the L X -z plane (e.g., see Fig. 3 of Ueda et al 2014). The majority of the detected sources lie at z ≈ 0.3-4 and have L X ≈ 10 43 -10 45 erg s −1 . The excellent source statistics over this redshift range allow for the accurate construction of XLFs in discrete redshift ranges and reliable inferences about the overall source properties in discrete bins across the L X -z plane. However, only a modest number of AGNs are detected at z < 0.3 and z > 4 in the current Chandra and XMM-Newton surveys ( < ∼ 50; e.g., Kalfountzou et al 2014;Vito et al 2014a), which is at least partially due to the small cosmological volumes probed at these redshifts; for example, the predicted yield of z < 0.3 (z > 4) AGNs with 2-10 keV luminosities of > ∼ 10 43 erg s −1 ( > ∼ 10 44 erg s −1 ) is ≈ 2-3 deg −2 (≈ 10-65 deg −2 , depending on whether there is a high-redshift space-density decline) down to 2-10 keV fluxes of ≈ 10 −14 erg cm −2 s −1 (≈ 10 −15 erg cm −2 s −1 ). 9 The modest number of AGNs at z < 0.3 from the Chandra and XMM-Newton surveys is partially mitigated by the ASCA medium sensitivity survey (e.g., Akiyama et al 2003) and the high-energy all-sky X-ray surveys, such as Swift-BAT (e.g., Tueller et al 2010;Baumgartner et al 2013) and will be greatly bolstered by the XMM-Newton XXL survey, in addition to the Chandra and XMM-Newton serendipitous surveys and eROSITA (see Section 6.4). At z > 4, a major challenge in addition to the relative rarity of high-redshift AGNs, is obtaining a reliable redshift constraint, particularly for moderate-luminosity AGNs where the majority are likely to be optically faint (i.e., R > ∼ 25 mag). However, improving the source statistics of z > 4 and z < 0.3 AGNs is a worthwhile endeavor since z > 4 corresponds to < 1.5 Gyr after the Big Bang and the era of the first growth and formation of SMBHs (see Section 6.1.2) while the z = 0.0-0.3 redshift range corresponds to 1/4 of cosmic history and an epoch that connects the evolution of AGNs from higher redshifts to the present day.
Despite the effective exploration of parameter space from the current Chandra and XMM-Newton surveys, an area of inquiry that is still relatively unexplored is the role of environment in the growth of SMBHs; see Sections 3.5 and 6.1.6. Current results suggest that the large-scale environment has a significant effect on the growth of SMBHs. However, the current Chandra and XMM-Newton surveys have not yet had the combination of both area and sensitivity sufficient to detect the majority of the growth of SMBHs (i.e., a factor of ≈ 10 below the knee of the XLF, L * ) across the full range of large-scale structure environments; based on cosmological simulations, regions of > ∼ 4 deg 2 are required to map out the largest structures (e.g., Springel et al 2005). The XMM-Newton XXL survey covers sufficient survey volume (≈ 50 deg 2 in two fields) but is only sensitive to the most-luminous AGNs, close to the knee of the XLF, while the Chandra and XMM-Newton observations of COSMOS have sufficient sensitivity but only cover ≈ 2 deg 2 and, therefore, have a comparatively poor sampling of all large-scale structure environments. The richest galaxy clusters will typically be too rare to be included in a blank-field survey of ≈ 4 deg 2 but analyses of the pointed observations of galaxy clusters (e.g., Martini et al 2009Martini et al , 2013 could be included to trace the most extreme largescale structure environments. To map out the large-scale structures requires large contiguous fields and, therefore, this is not a scientific project that could be undertaken by the serendipitous surveys. NuSTAR has only been operating for ≈ 2 yr and, with a comparatively small FOV, has covered a more limited swathe of the flux-solid angle plane than Chandra and XMM-Newton. The current source statistics from the combination of all of the NuSTAR surveys (see Table 1) are good at intermediate X-ray fluxes [ > ∼ 100 sources with 8-24 keV fluxes of (5−50)×10 −14 erg cm −2 s −1 ], allowing for statistically significant inferences about the source populations in this flux range. However, < 10 sources are detected at both faint and bright X-ray fluxes (8-24 keV fluxes of < 5 × 10 −14 erg cm −2 s −1 and > 5 × 10 −13 erg cm −2 s −1 ), restricting the reliability of any inferences about the properties of these source populations. Further NuSTAR surveys with comparable exposure times to the deepest current surveys in addition to a shallow NuSTAR survey will improve the source statistics at both the faint and bright X-ray flux ends, and also bridge the (approximate) order of magnitude difference in flux between the brightest NuSTAR survey sources and the faintest Swift-BAT survey sources.

Multiwavelength follow-up observations of X-ray surveys
The heart of an X-ray survey is of course the X-ray data. But the backbones are the supporting multiwavelength observations, which provide the necessary data to measure the key properties and environments of the detected sources and to construct XLFs. The majority of the extragalactic X-ray surveys listed in Table 1 are in well-established survey fields with extensive multiwavelength data. The available multiwavelength data are more limited for the serendipitous surveys, which cover non-contiguous areas across the sky and often have to rely on shallow all-sky survey data. Here we review current and future plans for multiwavelength follow-up observations of X-ray surveys, focusing on both all-sky and large-area survey plans in addition to deeper targeted multiwavelength follow-up facilities. As a demonstration of the sensitivity of several selected future observatories, see Fig. 16.

Optical-MIR wavelengths
At optical wavelengths, often the band pass of choice for initial counterpart identification and spectroscopic observations of X-ray sources, the entire sky has been observed in three bands down to an R-band magnitude of R ≈ 21-22 (the SuperCOSMOS Sky Survey; e.g., Hambly et al 2001). Recently completed Fig. 16 Multiwavelength SED of a z = 7 obscured AGN with L X = 3 × 10 43 erg s −1 and N H = 10 23 cm −2 , based on the template SED constructed by Lusso et al (2011). The relative sensitivities of selected future observatories are shown to illustrate their potential for characterizing a high-redshift AGN; ALMA has started operations but has not yet reached its full potential. The 3 σ sensitivity limits for a ≈ 40 ks exposure are plotted for all of the observatories except for Athena, where a 300 ks exposure was assumed. Adapted from Aird et al (2013b). and on-going optical large-area surveys will extend the coverage down to optical magnitudes of 23-24 in five optical bands over practically the entire sky: for example, the SDSS (e.g., York et al 2000;Aihara et al 2011), VST ATLAS (e.g., Shanks et al 2013), PanSTARRS (e.g., Kaiser et al 2010;Magnier et al 2013), and DES (e.g., Flaugher 2005). From 2022-2032, LSST will increase the optical coverage over ≈ 18,000-30,000 deg 2 in six bands down to r ≈ 27.5 with a 10 yr survey on a dedicated 8.4 m telescope (e.g., Ivezic et al 2008), sufficient to identify optical counterparts for almost all X-ray detected sources, including the z > ∼ 4 AGNs required to constrain the early phases of SMBH growth in the universe.
Broader wavelength coverage over optical-MIR wavelengths is required to extend simple counterpart identification to the measurement of accurate photometric redshifts and host-galaxy masses of the X-ray sources. In the NIR-MIR band pass, 2MASS (Skrutskie et al 2006) and WISE (Wright et al 2010) have provided shallow to moderate-depth data across the entire sky. While providing a good general resource, the 1.1-2.3 µm 2MASS survey is too shallow (K ≈ 14 mag) to detect many of the sources found in X-ray surveys; however, the WISE 3.4-22 µm survey is an excellent complement to wide-field or shallow X-ray surveys, particularly for the NuSTAR surveys where the majority of the sources are bright at MIR wavelengths (e.g., Alexander et al 2013;Lansbury et al 2014). The on-going VISTA surveys are observing ≈ 20,000 deg 2 of the Southern hemisphere at NIR wavelengths to > ∼ 4 mags deeper than 2MASS (principally the VHS survey; e.g., McMahon et al 2013;Sutherland et al 2014) and the UKIDSS surveys have observed ≈ 4,000 deg 2 of, predominantly, the Northern hemisphere to > ∼ 4 mags deeper than 2MASS (principally the LAS survey; Lawrence et al 2007), with substantially deeper data in smaller regions (e.g., in the VST ATLAS survey; e.g., Arnaboldi et al 2007). These surveys are particularly useful for characterizing the properties of the X-ray sources detected in large-area and serendipitious X-ray surveys (e.g., providing photometric redshifts and host-galaxy masses), where the existing NIR coverage is comparatively shallow. As described below, deeper optical-MIR coverage over smaller areas of the sky will be possible over the coming decade with, for example, the Wide-Field Infrared Survey Telescope (WFIRST), JWST , and the operation of the first Extremely Large Telescopes (ELTs; telescopes with > ∼ 30 m diameter mirrors).
The deployment of highly multiplexing optical-NIR spectrographs on large telescopes over the next ≈ 5 yr (e.g., Blanco-DESpec; Euclid; Mayall-DESI; SDSS IV-eBOSS; Subaru-PFS; VISTA-4MOST; VLT-MOONS; e.g., Laureijs et al 2011;Abdalla et al 2012;de Jong et al 2012;Levi et al 2013;Sugai et al 2014;Takada et al 2014;Cirasuolo et al 2014) will provide the spectroscopic complement to large-area and serendipitious X-ray surveys by yielding redshifts for hundreds to thousands of X-ray sources in an individual observation over large FOVs (typically ≈ 1-5 deg 2 ) down to optical magnitudes of ≈ 24. In addition to providing accurate source redshifts, the spectroscopic observations allow for characterization of the emission-line properties of the X-ray sources (e.g., identification of broad emission lines; measuring the emission-line gas conditions from emission-line ratios; e.g., Veilleux and Osterbrock 1987;Kewley et al 2013) and the identification of the large-scale structure environment in which they reside (e.g., identifying "filaments" of galaxies and AGNs in narrow redshift ranges). These analyses are particularly powerful when using spectrographs with sensitivity out to NIR wavelengths since they allow for the identification of rest-frame optical emission lines of distant AGNs at z > ∼ 1 (e.g., Euclid; Subaru-PSF; VLT-MOONS). Many ambitious observing programs are already planned using these instruments to obtain spectroscopic redshifts for millions of galaxies and AGNs (e.g., SDSS IV-eBOSS and 4MOST follow up of eROSITA sources). On slightly longer timescales (mid 2020's), WFIRST is expected to launch and observe > 1,000 deg 2 of the sky down to faint NIR depths at HST resolution and obtain NIR spectroscopic-grism redshifts for millions of distant galaxies and AGNs out to z > ∼ 8 (e.g., Spergel et al 2013;Gehrels and Spergel 2014).
A non-negligible fraction ( < ∼ 3%) of the sources detected in the deepest X-ray surveys remain undetected at optical-MIR wavelengths even with the most sensitive current observatories (e.g., HST; 8-10 m class telescopes; Spitzer ), and a larger fraction ( < ∼ 40%) lack spectroscopic redshifts (e.g., Luo et al 2010;Xue et al 2011;Hsu et al 2014). Revolutionary advances in ultrafaint imaging and spectroscopy will be made over the next 5-10 yr with the launch of JWST and the first ELTs. JWST (Gardner et al 2006) is a NASA-ESA-Canadian Space Agency satellite hosting a 6.5 m telescope with high sensitivity at 0.6-28 µm and diffraction-limited spatial resolution (≈ 0.1-1 arcsec).
JWST is planned to be launched in the next ≈ 5 yr and will provide NIR-MIR imaging and spectroscopy a factor > ∼ 10 times below current facilities, providing the potential to achieve (near) complete spectroscopic redshifts for the X-ray sources. 10 Several ELTs are planned over the next decade, including the European ELT (E-ELT), the Giant Magellan Telescope, and the Thirty Meter Telescope. The ELTs will complement JWST in an analogous manner to how the current 8-10 m telescopes complement HST, with innovative instruments and ultra-deep imaging and spectroscopy over large FOVs. 11

Infrared-radio wavelengths
Long-wavelength data at MIR-submm wavelengths are required to measure the energetics for the dust-obscured star formation and AGN components of the X-ray sources. Spitzer and Herschel have provided an excellent resource at MIR-FIR wavelengths over the past decade for the majority of the X-ray survey fields. However, they have only covered a comparatively small fraction of the overall sky (≈ 1,000 deg 2 ; see Lutz 2014), which often restricts MIR-FIR studies of the X-ray sources detected in the serendipitious surveys to the all-sky MIR-FIR coverage. WISE has provided moderate-depth MIR imaging across the whole sky (see above), although the available all-sky survey data at FIR wavelengths (≈ 30-200 µm) is limited to the shallow IRAS and AKARI surveys, which typically only provide useful FIR measurements for z < ∼ 0.1 sources, severely restricting SFR constraints for serendipitous X-ray sources that lack Herschel coverage. The only currently operating FIR telescope is SOFIA (e.g., Young et al 2012a; the effective operational lifetimes of Spitzer and Herschel at MIR-FIR wavelengths were limited by their helium cryogen supply), an airborne observatory with moderate sensitivity over the broad ≈ 0.3-1600 µm band pass. While providing good FIR constraints for nearby AGNs, only the brightest distant AGNs will be detected by SOFIA and it, therefore, does not provide significant improvements over existing SFR constraints for the X-ray sources detected in the serendipitious surveys.
On longer timescales (≈ 2025-2030), the Space Infrared Telescope for Cosmology and Astrophysics (SPICA) is currently the leading concept for a nextgeneration MIR-FIR observatory, although it is not yet fully funded. The primary aim of SPICA (e.g., Nakagawa et al 2014;Roelfsema et al 2014) is faint medium-resolution spectroscopy over the broad 20-210 µm band pass, which will allow for the first extensive studies of the MIR-FIR emission lines of distant systems to provide accurate measurements of the star-formation properties and the interstellar medium of distant X-ray AGNs; however, SPICA will also provide deep broadband MIR-FIR photometry to sensitivity levels slightly below those of Spitzer and Herschel , allowing for deep MIR-FIR observations beyond the X-ray survey regions with existing Spitzer-Herschel coverage. Future MIR-FIR space-borne observatories such as SPICA are key to drive forward our understanding of the role that distant AGNs play in the formation and evolution of galaxies.
The terrestrial atmosphere is mostly opaque at MIR-FIR wavelengths and the majority of the MIR-FIR telescopes described above are mounted inside space-borne observatories (SOFIA is the exception but it is still operated from an aircraft). However, there are discrete band passes at ≈ 0.3-3 mm, in the submm-mm wave band, where the atmosphere is sufficiently transparent to allow for sensitive ground-based observations (see Fig. 4 of Casey et al 2014). The negative K-correction for typical AGN and galaxy SEDs at these wavelengths means that a distant AGN or galaxy with a given SFR has a comparable submm-mm flux across the broad redshift range of z ≈ 0.5-6 (e.g., Blain et al 2002;Casey et al 2014). Over the past two decades there have been several ground-based observatories sensitive at submm-mm wavelengths (e.g., APEX; ASTE; CARMA; IRAM; JCMT; SMA; see Casey et al 2014 for a recent review), which have provided the first clear views of the submm emission from distant AGNs and galaxies. The most recent submm-mm observatories, which have either started operations or are planned to start in the next ≈ 5 yr are the Atacama Large Millimeter/Submillimeter Array (ALMA; Wootten and Thompson 2009), the Large Millimeter Telescope (LMT; e.g., Hughes et al 2010), and CCAT (e.g., Woody et al 2012). ALMA is an interferometer and provides high-resolution imaging and spectroscopy down to sensitivities over an order of magnitude deeper than previous submm-mm facilities with subarcsec resolution. ALMA has the potential to provide unprecedented insight into the star-formation properties of distant X-ray AGNs (e.g., SFR constraints down to those of quiescent galaxies and potentially resolving the extent of star formation in some sources); see Fig. 16. Both the LMT and CCAT aim to reach a broadly similar submm-mm sensitivity limit as ALMA but at a lower spatial resolution over larger FOVs (e.g., up to 1 deg 2 for CCAT) and will provide deep submm-mm measurements of the star-formation properties for large samples of distant X-ray AGN, particularly when combined with sensitive MIR-FIR data.
Radio observations provide independent constraints on the amount of star formation and (radio bright) AGN activity in distant X-ray sources, allowing for a more complete census of AGN activity and the exploration of the connection between AGN activity and star formation. Over the past two decades, large area X-ray surveys and serendipitous X-ray surveys have largely relied on the NRAO VLA Sky Survey (NVSS; Condon et al 1998) and the VLA FIRST survey (Becker et al 1995), which cover ≈ 10,000-33,000 deg 2 down to mJy levels at 1.4 GHz. These data are sufficient to detect radio-bright X-ray sources, although the majority of the X-ray source population is ≈ 1-2 orders of magnitude fainter than the sensitivity of NVSS and FIRST. However, substantially deeper radio surveys are now being undertaken over > ∼ 1,000 deg 2 and many aim to cover the majority of the visible sky (e.g., at low < ∼ 200 MHz frequencies: GMRT-TGSS; LOFAR; MWA; at mid ≈ 1 GHz frequencies: Apertif-WODAN; ASKAP-EMU; MeerKAT-MIGHTEE; VLA-VLASS; see Norris et al 2011Norris et al , 2013Lazio et al 2014 for a recent summary). Many of these surveys are pre-cursors and pathfinders to the Square Kilometer Array (SKA; e.g., Dewdney et al 2009;Carilli 2014), an international project to build the world's largest radio telescope, planned to start operations over the next ≈ 10-20 yr. The SKA is designed to operate over a very broad range of frequencies (≈ 50 MHz to 20 GHz) down to sub-µJy levels at sub-arcsec resolution, sufficient to detect essentially all of the X-ray sources at radio frequencies and provide an independent method of AGN selection (i.e., radio bright AGNs undetected at X-ray energies) in addition to sensitive SFR constraints; see Fig. 16. Due to the high sensitivity and large FOV, SKA will be able to undertake a near allsky survey (3 π steradians) to sensitivities equivalent to the deepest current radio surveys (≈ 1-2 µJy rms at 1.4 GHz; R. P. Norris, 2014, private communication), sufficient to detect the majority of the X-ray sources in the current serendipitious and all-sky X-ray surveys.

New X-ray survey missions
Several new X-ray observatories are planned in the near (< 5 yr) and long (> 10 yr) term that will extend the great progress that Chandra, XMM-Newton, and NuSTAR have made toward our understanding of the demographics, physics, and ecology of distant SMBHs.
In the near term, both eROSITA and ASTRO-H are expected to become operational, with launch dates of 2015-2016. eROSITA  is a joint Russian-German mission and will provide imaging and spectroscopy over ≈ 0.5-10 keV. The primary objective of eROSITA is to perform a moderatedepth survey of the entire sky within the first 4 yr of launch at relatively low spatial resolution (effective half-energy width of ≈ 28-40 arcsec, depending on energy), detecting ≈ 3 million AGNs out to z ≈ 6. The expected sensitivities of the all-sky survey are ≈ 1 × 10 −14 erg cm −2 s −1 (0.5-2 keV) and ≈ 2 × 10 −13 erg cm −2 s −1 (2-10 keV) and will be ≈ 4 times more sensitive in the deepest regions at the ecliptic poles; the all-sky sensitivity limits are ≈ 20 times deeper than ROSAT at 0.5-2 keV (Voges et al 1999) and ≈ 200 times deeper than HEAO 1 A-2 at 2-10 keV (Piccinotti et al 1982). ASTRO-H (Takahashi et al 2012) is a joint JAXA-NASA mission and will provide imaging and spectroscopy of cosmic X-ray sources over the broad energy range of ≈ 0.3-80 keV. ASTRO-H should have a comparable angular resolution and sensitivity limit as NuSTAR at > ∼ 3 keV but over a broader energy range. ASTRO-H also has a wide-field imaging spectrometer and a high-resolution micro-calorimeter that will provide spectroscopy of cosmic X-ray sources over 0.3-12 keV at a higher spectral resolution than Chandra, XMM-Newton, and NuSTAR (∆E < 7 eV). The relatively small FOV of ASTRO-H for imaging at hard X-ray energies (≈ 9 × 9 arcmin; see Table 2 of Takahashi et al 2012) means that it is potentially better suited to observing the well-established survey fields than undertaking new wide-area surveys at hard X-ray energies.
In the longer term, many exciting X-ray observatory concepts have been proposed that will provide revolutionary advances in our understanding of the properties and evolution of the sources detected in cosmic X-ray surveys. 12 We do not have the space in this review to discuss all of these concepts and, to date, only one has been selected for long-term financial support (Athena; Nandra et al 2013). However, to provide some flavor of the X-ray facilities that may be available in the next ≈ 10-30 yr, we briefly describe four different proposed concepts: Athena, HEX-P , SMART-X , and WFXT .
Athena is an ESA-led mission that is scheduled for launch in 2028. With a large collecting area (≈ 2.0-2.5 m 2 at 1 keV), large FOV (≈ 40 × 40 arcmin), and good spatial resolution (≈ 3-5 arcsec half-energy width), Athena will be an excellent general-purpose X-ray observatory and a devastatingly effective survey machine, achieving a given flux-solid angle limit ≈ 2 orders of magnitude more quickly than Chandra and XMM-Newton. The large effective area and good angular resolution combination is achieved from innovative Siliconpore optic technology. The ultimate sensitivity limit of Athena is dictated by the confusion limit and will be comparable to that of a ≈ 2 Ms Chandra observation [0.5-2 keV fluxes of ≈ (2-3) ×10 −17 erg cm −2 s −1 ; e.g., Alexander et al 2003;Luo et al 2008]. However, Athena will be able to achieve that sensitivity limit in ≈ 200-400 ks and, therefore, a 2 Ms Athena survey would cover ≈ 1 deg 2 to this depth, as compared to Chandra which only reaches this sensitivity limit over the central region. The large collecting area of Athena will also provide high signal-to-noise X-ray spectroscopy of distant X-ray AGNs at ≈ 100 eV resolution, allowing for direct redshift measurements from the identification of iron Kα emission lines, and accurate measurements of their spectral properties (i.e., N H , Γ , reflection, intrinsic X-ray luminosity); a high spectral resolution microcalorimeter (called the X-IFU) will provide ≈ 2.5 eV resolution but over a smaller ≈ 5 × 5 arcmin FOV. With these capabilities, Athena will (among other things) efficiently identify moderate-luminosity AGNs at z > ∼ 6 (see Fig. 16), potentially constraining the seeds of SMBHs (see Section 6.1.2), and perform a near-complete census of AGNs out to at least z ≈ 3, even identifying many Compton-thick systems from the detection of strong iron Kα emission (e.g., Aird et al 2013b).
HEX-P (PI: F. Harrison) is a natural successor to NuSTAR and combines an optimized optics design at high energies (half-power diameter resolution of ≈ 10-15 arcsec) with a broader energy bandpass of ≈ 0.1-200 keV and a larger effective area than NuSTAR and XMM-Newton. HEX-P aims to be ≈ 40 times more sensitive than NuSTAR, sufficient to resolve ≈ 90% of the CXRB at its ≈ 20-40 keV peak and to detect AGNs almost independent of the presence of absorption out to z ≈ 6. SMART-X (PI: A. Vikhlinin) is a natural successor to Chandra and aims to use adaptive optics to achieve excellent ≈ 0.5 arcsec resolution (half-power diameter) at 0.2-10 keV with ≈ 30 times the effective area of Chandra. SMART-X is predicted to reach the depth of the 4 Ms CDF-S survey over 5 deg 2 in 4 Ms of exposure or ≈ 3 × 10 −19 erg cm −2 s −1 at 0.5-2 keV in a single pointing of 4 Ms, sufficient to detect the first growing SMBHs at z ≈ 10-20 [L X = (0.4-2) ×10 42 erg s −1 ; e.g., Vikhlinin and SMARTX Collaboration 2013]. WFXT (PI: S. Murray) is a natural successor to eROSITA and combines a large ≈ 1 deg 2 FOV with good angular resolution of ≈ 5 arcsec (half-energy width) to provide mapping of large areas of the sky to faint flux limits at ≈ 0.5-7 keV. WFXT is predicted to achieve a sensitivity limit comparable to that of a 2 Ms Chandra observation and with a dedicated 3 yr program would be able to survey ≈ 100 deg 2 and ≈ 3,000 deg 2 to 0.5-2 keV flux limits of ≈ 4 × 10 −17 erg cm −2 s −1 and ≈ 5 × 10 −16 erg cm −2 s −1 , respectively, sufficient to detect ≈ 5 million AGNs overall and tens of AGNs even at z = 8-10 (e.g., Murray et al 2013). With high X-ray sensitivity, large FOV (with good spatial resolution across the full FOV), and good positional accuracy, WFXT provides a great complement to the next-generation optical imaging surveys undertaken by, for example, LSST (see Section 6.3.1).
Ever since the first rocket flights of the 1960's, cosmic X-ray surveys have been an essential tool for elucidating the processes of mass accretion onto SMBHs. With orders of magnitude improvements in sensitivity over previous generation X-ray missions, the Chandra, XMM-Newton, and, most recently, NuSTAR observatories have (arguably) provided greater leaps forward in our understanding of the demographics, physics, and ecology of distant growing SMBHs than any other facility over the past two decades. This is a subject area strongly driven by technological advances in telescope and instrument design, and the revolutionary developments in X-ray and multiwavelength facilities over the coming decades promise yet greater advances in our understanding of when, where, and how SMBHs have grown in the distant universe.