Observations and Models of High Redshift Radio Galaxies and Quasars from the 3Rd Cambridge Catalog”

”Observations and models of high redshift Radio Galaxies and Quasars from the 3rd Cambridge catalog”

Dissertation

zur Erlangung des Grades ”Doktor der Naturwissenschaften” an der Fakultat¨ fur¨ Physik und Astronomie der Ruhr-Universitat¨ Bochum

von Frank Heymann

aus Leipzig

Bochum 2010

b.w. 2

1. Gutachter Prof. Dr. Rolf Chini 2. Gutachter Priv.-Doz. Dr. Dominik Bomans Datum der Disputation 05.07.2010 3

Abstract

This thesis provides new observations of the most powerful high redshift Active Galactic Nuclei (AGN), namely the complete z > 1 3CR sample, and new dust radiative transfer modeling of their measured spectral energy distribution in the infrared. This work is separated into three main parts, two observational sections and one section containing the modeling.

The first part shows observational results in the near (NIR) and mid (MIR) infrared obtained with the Spitzer Space Telescope, to extend the knowledge on high redshift sources. The main aspect of these observations is to study orientation dependence of the NIR and MIR emission and to confirm the unification scheme for the most powerful high redshift AGNs.

The second part reports on a pilot study to detect galaxy clustering around high redshift radio sources using the Spitzer data. Because the radio AGN reside in massive host galaxies, they are expected to serve as signposts for cosmic mass peaks. These galaxy clusters are among the most distant known structures and therefore of particular cosmological interest.

The third part explains a newly developed method to solve the radiative transfer equation in three dimensional conﬁgurations. This method makes use of the parallelization capabilities of modern vector computing units, like the graphics cards. The speed improvement is about a factor of 100. This enables us to model the close environment of AGN in so far unprecedented detail within reasonable computing time. 4 Contents

1 Introduction 7 1.1 Aimofthisthesis...... 7 1.2 ActiveGalacticNuclei ...... 7 1.2.1 Seyfert Galaxies ...... 9 1.2.2 Quasars & Radio Galaxies ...... 10 1.2.3 Uniﬁcation...... 10 1.3 GalaxyClustering...... 11 1.4 Dustradiation...... 12 1.4.1 Theemissivityofdust ...... 13 1.4.2 Thetemperatureofbiggrains ...... 13 1.4.3 Verysmallgrains ...... 14 1.5 BasicRadiativetransport ...... 15 1.5.1 Deﬁnitions...... 15 1.5.2 The general transfer equation ...... 16 1.5.3 Analytical solutions ...... 17

2 Near- and mid infrared photometry 19 2.1 Motivation...... 20 2.2 ObservationsandData ...... 21 2.3 ResultsandDiscussion ...... 24 2.3.1 Radio galaxies as obscured quasars ...... 24 2.3.2 Evolutionary eﬀects and non-thermal contributions . . . . 27 2.4 Conclusions ...... 28

3 The cluster search 35 3.1 Motivation...... 35

5 6 CONTENTS

3.2 Clusteringaround3C270.1 ...... 37 3.2.1 Observationsanddata ...... 37 3.2.2 Results...... 38 3.3 No evidence for clustering around 3C 437 ...... 51 3.3.1 Observationsanddata ...... 51 3.3.2 Results...... 51 3.3.3 Preliminary Conclusion ...... 53

4 Parallel 3D radiative transport 55 4.1 Theory...... 55 4.1.1 MonteCarlomethod ...... 55 4.1.2 Parallelization ...... 58 4.1.3 Pseudo random number generator in parallel ...... 60 TM 4.1.4 CUDA parallelization on graphic cards ...... 60 4.2 Imaging ...... 61 4.2.1 Solarsystem...... 65 4.3 Benchmarktest ...... 65 4.3.1 Sphericalsymmetry(1D)...... 65 4.3.2 Diskgeometry(2D)...... 67 4.3.3 Dustproperties ...... 69 4.3.4 Spiral expansion of disk structure (3D) ...... 69 4.3.5 Clumpy Torus geometry (3D) ...... 70 4.4 Modelingaveragespectra&SEDs...... 75

5 Summary and Outlook 81

Bibliography 87

List of Figures 93

List of Acronyms 101 Chapter 1 Introduction

1.1 Aim of this thesis

At the begin of my thesis, new unprecedented infrared observations of the complete high-redshift 3CR sample have been obtained with the Spitzer Space Tele- scope. Therefore, the aim of this thesis is

• to explore the near- and mid-infrared spectral energy distributions of this sample, comprising the most powerful radio-loud AGN,

• to test, how far it is possible with these data to detect galaxy clustering around these mass peaks of the early universe,

• and to develop a new proper 3D Monte Carlo radiative transfer code to model the spectral energy distributions.

These three tasks represent new challenges and thus lead to new results.

1.2 Active Galactic Nuclei

Active Galactic Nuclei (AGN) belong to the the most luminous objects in the universe. The luminosity of an AGN is provided by accretion of matter onto the central supermassive black hole:

˙ ˙ 2 12 ǫ M LAGN = ǫ M c ≈ 1.2 10 L⊙ , (1.1) 0.1 M⊙/yr where ǫ is the eﬃciency of the mass to radiative energy transfer, L⊙ solar luminosity, M⊙ solar mass and M˙ the mass accretion. This leads to a theoretical

7 8 1.2. ACTIVE GALACTIC NUCLEI

upper limit for the central luminosity, named after Arthur Stanley Eddington:

4πGMBH mpc 11 MBH Lmax

where mp is the mass of the proton, σT the Thomson cross section for interaction between electrons and protons. The wavelength spectral energy distribution of an AGN exhibits three characteristic features (Elvis et al. 1994) as shown in Fig 1.2:

Figure 1.1: Sketch of an AGN continuum spectrum of the nuclear region, without stellar contribution. Three diﬀerent bumps can be seen (Big Blue Bump in the middle, Infrared Bump on the left and X-ray ’Bump’ on the right). Figure from Manners (2002)

• Infrared Bump This feature consists of several components. Dominant is the emission from the hot and warm dust torus (red dashed line) and cooler dust from the host galaxy. The starburst activity in the host galaxy contributes to the far infrared (purple dotted line) followed by a steep decrease of the Infrared bump to the submillimeter (Chini et al. 1989). The local minimum at around 1m is given by the sublimation temperature of the dust around ∼ 1500 K. CHAPTER 1. INTRODUCTION 9

• Big Blue Bump At shorter waverlengths the minimum turns into the Big Blue Bump (blue dashed-tripple-dotted line). This bump comes from the thermal emission of hot gas (5 000K − 100 000 K) heated by viscous processes, in the accretion disk. The gap in the bump results from absorption of neutral hydrogen and therefore missing data. This optical/UV radiation is eﬃciently transfered into infrared emission and therefore powers the Infrared Bump. (Miley et al. 1985)

• X-ray ’Bump’ The ﬁnal feature in the AGN continuum is the high energy X-ray ’Bump’. The radiation in this bump is produced by the hot corona above the accretion disk (green dashed line) and reﬂection of the disk (blue dashed-dotted line).

The observational data of my thesis provide new constraints on the Infrared Bump (chapter 2), and the model part makes use of all three bumps (chapter 4).

1.2.1 Seyfert Galaxies

These galaxies, discovered by Seyfert (1943), contain a bright nucleus with strong emission lines from highly ionised gas (hydrogen, helium, nitrogen, oxygen). The Seyfert galaxies can be divided into two subclasses depending on the existence of broad and narrow emission lines (type 1) or only narrow lines (type 2) (Khachikian & Weedman 1974). The broad lines have velocities of 1 500−10 000 km s−1, may vary on short timescales and can be explained by Doppler broadening. These high velocities can be explained by gas clouds, orbiting the black hole at small distances. It is also possible that these lines are emitted from the accretion disc itself. However due to the extremly high resolution which is neccessary to resolve the innermost part, it is diﬃcult to observe the exact geometry of these objects. The narrow emission lines may by emitted by gas clouds further out. This is strengthened by the fact that the narrow lines are detected in all types of Seyfert galaxies, which implies that the emitting region is large. Breakthrough spectropolarimetric observations revealed, that some Seyfert 2 galaxies contain a hidden broad line region, leading to the AGN uniﬁcation scheme (section 1.2.3). 10 1.2. ACTIVE GALACTIC NUCLEI

1.2.2 Quasars & Radio Galaxies

These two object classes are the powerful radio-loud cousins of the Seyfert galaxies. On early optical images quasars appeared starlike, which gave these objects the name quasi stellar radio source. In quasars and radio galaxies large structures, the radio lobes, are prominent. Depending on the morphology and power of the radio lobes, the radio sources are subdivided into the FR I and FR II classes (Fanaroff & Riley 1974). We here consider only the edge-brightened powerful FR II sources. The radio emission is powered by synchrotron radiation of outflowing material reaching 95% of the speed of light. Assuming an accretion disc and perpendicular to this disc an outflowing jet producing the synchrotron radiation, it is possible to explain these two objects with the orientation of the jet to the line of sight. The quasars, where the jet points to the observer, represent the type 1 AGN and the Radio Galaxies, where the jet is perpendicular to the line of sight, can be classified as type 2 (Barthel 1989; Orr & Browne 1982). The isotropic radiation at meter wavelengths enables to select objects with the same radio power and to study orientation dependent effects. While for the local universe (z < 1) orientation dependencies are fairly well studied, the knowledge is poor at high redshift (z > 1).

1.2.3 Uniﬁcation

Figure 1.2 shows the actual picture of the unification model. It is believed that Type 1 and Type 2 galaxies are in essence the same, and they differ only by the angle at which they are observed. This unification scheme was proposed by Antonucci (1993). In Type 2 Seyferts the broad line component, which is produced by high velocity clouds in the vicinity of the black hole, is obscured by dust in a torus like structure. The obscuration then depends on the observing angle onto the dusty torus. The narrow emission lines are produced in clouds with lower rotation velocities and therefore larger distances to the black hole presumably in a bicone far outside the obscuring torus scale hight. Perpendicular to the accretion disk exists a jet of outflowing material, to remove the angular moment from the accretion disk. This outflow reaches speeds of up to 95% of the speed of light, and produces synchrotron radiation in the radio wavelength. In some Type 2 Seyfert galaxies the broad component can be observed in polarized light. In these objects the broad-line region is scattered by material in the bicone (dust or electrons), which allows us to see it indirectly. This effect was first discovered by Antonucci (1983) in the Type 2 Seyfert NGC 1068. CHAPTER 1. INTRODUCTION 11

Narrow Line Region

Broad Line Region Jet

Black Accretion Hole Disk

Obscuring Torus

Figure 1.2: This figure shows the unification scheme from Urry & Padovani (1995). In the center the black hole with its accreating disk is shown. Per- pendicular to this disk is a jet of outflowing material. Close to the black hole are clouds which produce the broad emission line features with speeds of up to 10 000 kms−1. The torus of dust shields the broad line region depending on the observing orientation. Further out is the narrow line region.

1.3 Galaxy Clustering

Clusters of nebulae (Zwicky 1937) have been recognised later as cluster of galaxies. When observed visually, clusters appear to be collections of galaxies held together by mutual gravitational attraction. However, their velocities are too large for 12 1.4. DUST RADIATION

them to remain gravitationally bound by their attractions, implying the presence of either an additional invisible mass component, or an additional force besides gravity. X-ray studies have revealed the presence of large amounts of intergalactic gas known as the intracluster medium (ICM). This gas is very hot, between 107K and 108K, and therefore emits X-rays in the form of bremsstrahlung and atomic line emission. The total mass of the gas is greater than that of the galaxies by roughly a factor of two. However, this is still not enough mass to keep the galaxies in the cluster. Since this gas is in approximate hydrostatic equilibrium with the overall cluster gravitational field, the total mass distribution can be determined. It turns out that the total mass deduced from this measurement is approximately six times larger than the mass of the galaxies or the hot gas. The missing component is called dark matter and its nature is unknown. In a typical cluster perhaps only 5% of the total mass is in the form of galaxies, maybe 10% in the form of hot X-ray emitting gas and the remainder is dark matter. Noteworthy, Brownstein & Moffat (2006) use a theory of modified gravity to explain X-ray cluster masses without dark matter. Clusters typically have the following properties:

• They contain 50 to 1,000 galaxies, hot X-ray emitting gas and large amounts of dark matter

14 15 • They have total masses of 10 to 10 M⊙ and sizes of 2 to 10 Mpc.

• The velocity distribution for the individual galaxies is about 800−1000 kms−1.

Well known galaxy clusters in the relatively nearby universe include the Virgo cluster, Hercules Cluster, and the Coma Cluster. A very large aggregation of galaxies known as the Great Attractor, dominated by the Norma cluster, is massive enough to affect the local expansion of the universe (Hubble flow). However, at large distances (redshift z > 1) cluster knowledge is very sparse, because of observational difficulties to discern cluster member galaxies from foreground galaxies. In chapter 3 a new pilot study to find high redshift (z ∼ 1.5) clusters is presented.

1.4 Dust radiation

This section describes the physical background of the dust emission. A single grain emits according to Kirchhoﬀ’s law, which can be derived from the assumption of harmonic oscillators in thermal equilibrium. In section 1.4.1 we discuss CHAPTER 1. INTRODUCTION 13 the thermal emission. The big grain radiation is discussed in section 1.4.2 and the very small grains in section 1.4.3.

1.4.1 The emissivity of dust

A dust particle in a radiation field has an equilibrium temperature derived from the condition that it absorbs as much energy as it emits. The ratio of the emission abs coefficient ǫν to the absorption coefficient Kν depends only on the frequency of the radiation field and temperature of the dust grain. Planck discovered that this dependence can be described by a function Bν(T ), named the Planck function. Astrophysical grains are usually not in local thermal equilibrium (LTE). But whenever it is heated by a photon the energy is distributed among the energy states. The distribution of the energy states derived by the Boltzmann function depends only on the dust temperature T . Therefore the grain emission only depends on T and is given, as in LTE, by

abs ǫν = Kν Bν(T ) . (1.3)

Then the emission into all directions equals 4πǫν. The value ǫν is called emissivity and can reference to a single particle, a unit volume or a unit mass. The dimension for the emissivity per unit volume is expressed in erg s−1 cm−3 Hz−1 ster−1.

1.4.2 The temperature of big grains

The dust temperature can be calculated with Equ. 1.3 and the dust temperature T from the balance between heating and cooling by

Kabs J dν = Kabs B (T) dν (1.4) ν ν ν ν where the left side of this equation describes the heating and the right side the cooling. Jν is the average of the radiation intensity over all directions. Equ. 1.4 neglects other heating or cooling mechanisms like collisions with gas particles. To account for these additional heating (cooling) one has to add them to Equ. abs 1.4. It is obvious that not only the absolute value of Kν determines the dust abs temperature, which means the grain is at high temperature if Kν is high at the absorption wavelength and low at the emission wavelength. For the simple case of abs a black-body the absorption coeﬃcient is Kν constant. This simpliﬁes equation 14 1.4. DUST RADIATION

1.4 to σ J dν = T 4 , (1.5) ν π with σ the radiation constant. The temperature of a dust black-body at distance r from a star with bolometric luminosity L is given by

L = 4σT 4 . (1.6) 4πr2

For the astrophysical example of our solar system, where L = L⊙, one ﬁnds:

r − 1 T = 279 ( ) 2 K , (1.7) AU

which is close to the surface temperature of the earth. This holds even if our planet emits like a perfect black body and the atmosphere reﬂects nearly 30% of the radiation from the sun.

1.4.3 Very small grains

When a very small dust particle absorbs an energetic photon its temperature ﬂuc- tuates. To compute these time variable temperatures their optical and thermal properties are needed. The optical behavior depends on the particle shape and

two dielectric functions ǫ1(ω) and ǫ2(ω) and the thermal behavior is determined from the speciﬁc heat (see Kr¨ugel 2008). For a large number of very small identical interstellar dust grains the chance for one of them to have a temperature in the interval T...T + dT is P (T )dT. Thus the normalised probability density is:

∞ P (T )dT = 1 . (1.8) 0

For big interstellar grains the temperature fluctuations are small, so that the probability density P (T ) approaches the δ-function δ(T ). The balance between emission and absorption is therefore given in equation 1.4. For very small particles we assume that the radiation fulfills Kirchhoff’s law at any time. In the case of spherical particles we can compute the average monochromatic emission per solid angle by ǫ = π a2 Qabs B (T )P (T ) dT (1.9) ν ν ν The emission of one of these grains is not constant over time, but the whole ensemble radiates at any frequency at a steady rate. The solution of 1.9 is reduced to find P (T ). A more detailed view of the emission of very small particles is CHAPTER 1. INTRODUCTION 15

Figure 1.3: Deﬁnition of the radiative intensity Iν

described by Guhathakurta & Draine (1989); Draine & Li (2001); Kr¨ugel (2008).

1.5 Basic Radiative transport

1.5.1 Deﬁnitions

The intensity Iν describes the radiation ﬁeld and is deﬁned as shown in Fig. 1.3 A surface element dσ at location r receives the power

dW = Iν(r,e) cos (θ)dσdΩdν , (1.10) where e is the direction, n is the unit vector of the surface element, θ is the angle between e and n, dΩ is the solid angle and dν is the width of the frequency interval. The dimension of the intensity is erg s−1 cm−2 Hz−1 ster−1. For extended astrophysical sources of solid angle ΩS the observed total ﬂux received at the observer is given by F obs = I (θ, φ)dΩ , (1.11) ν ν ΩS where dΩ= sin(θ)dθdφ (1.12) is the element of solid angle. For constant Iν(θ, φ) over the source Equ. 1.11 simpliﬁes to obs Fν = IνΩS . (1.13) 16 1.5. BASIC RADIATIVE TRANSPORT

The mean intensity J is the average over the total solid angle,

1 J = I(e)dΩ . (1.14) 4π 4π

The net ﬂux F through a unit surface is:

2π π F = I(θ, φ) cos (θ) sin (φ)dθdφ . (1.15) 0 0

1.5.2 The general transfer equation

The light changes its intensity by traveling through a dusty medium. The intensity is reduced by absorption and increased by the thermal emission of the dust and the scattering of photons out of and into the line of sight. The intensity change at position r into direction e can be written as

dI ν = −KextI (r,e)+ ǫ , (1.16) ds ν ν ν

ext abs sca where Kν = Kν + Kν is the extinction coeﬃcient per unit volume including the absorption and scattering. The emission and scattering of the dust at temperature T can be written as

sca abs Kν ′ ′ ′ ǫν = Kν Bν(T )+ pν(e ,e)Iν(e )dΩ . (1.17) 4π 4π

′ ′ The factor pν(e ,e) is the phase function for scattering from e into e. A funda-

mental quantity in the light propagation is the optical depth, τν. When the light

travels from a point P1 to P2 the optical depth is given by

P2 τ = Kextds , (1.18) ν ν P1 or ext dτν = Kν ds . (1.19)

For dusty media the extinction coeﬃcient is independent of temperature, which

simpliﬁes the integral in Equ. 1.18. With the column density Nd, the number of grains in a cylinder with a base of 1 cm2, the optical depth is given by

ext τν = NdCν , (1.20) CHAPTER 1. INTRODUCTION 17

ext where Cν is the extinction coeﬃcient for one grain.

1.5.3 Analytical solutions

It is possible to solve Equ. 1.16 for pure extinction without emission. This can be the light from an object, which is only decreased by the intervening dust. The solution is then −τ I(τν)= I(0)e , (1.21) where I(0) is the intensity of the source. The intensity I(τ) decreases exponential with the optical depth. The solution of Equ. 1.16 for pure emission without extinction (Kext = 0) is

s I(s)= I(0) + ǫ(s′) ds′ , (1.22) 0 where s is the path length and ǫ the emission coeﬃcient (Equ. 1.17). This situation is a good approximation for the sub-millimeter radiation of interstellar dust. The intensity for pure emission without a background source scales with the path length s. The more general case of background source emission plus dust emission and absorption is

τ −τ ′ τ−τ ′ ′ Iν(τ)= I(0)e + Sν(τ )e dτ , (1.23) 0 where S(τ) is the source function. For the simple case of a constant source function it is possible to solve the integral in equ. 1.23. We assume a dust cloud of uniform temperature T , the source function is given by the Planck function

Sν(τ)= Bν(T ) . (1.24)

This gives the solution for the intensity and ﬂux of the absorption plus emission of dust on a background source by

τ −τ Iν(τ)= I(0) e + Bν(T ) (1 − e ) (1.25) and τ −τ Fν(τ)= F (0) e + Bν(T ) (1 − e )Ω . (1.26)

If the source emission is known or neglectable this equation gives the possibility to get the dust temperature and optical depth and therefore the dust mass. The 18 1.5. BASIC RADIATIVE TRANSPORT

limiting solutions for Equ. 1.25 are

B(T )τ + I(0) : for τ << 1 Iν(τ)=  (1.27) B(T ) : for τ >> 1 .  For low optical depth and no background source the radiation intensity received at the observer is directly proportional to the optical depth. The intensity of the received dust emission can be used to investigate dust properties like the dust mass or the grain frequency dependence. For very high optical depth the intensity is given by the black-body radiation. This means all information about the dust, with the exception of the dust temperature, is lost. Chapter 2 Near- and mid infrared photometry

This part represents one of the developed papers during my PhD thesis. My contribution in this large collaboration of many coauthors was mainly:

• Photometry of the entire high redshift 3C sample (Tables 5.1 and 5.2 in appendix 3C sources).

• Developing a pipeline to extract the photometry from the Spitzer mosaic images.

• Interactive inspection, veriﬁcation and improvement of the spectral energy distribution of these faint sources.

Abstract: The 178 MHz-selected sample is unbiased with respect to orientation and therefore suited to study orientation-dependent effects in the most powerful active galactic nuclei (AGN). Quasar and radio galaxy subsamples matched in isotropic radio luminosity are compared. The quasars all have similar spectral energy distributions (SEDs), nearly constant in ν Fν through the rest 1.6–10 m range, consistent with a centrally heated dust distribution which outshines the host galaxy contribution. The radio galaxy SEDs show larger dispersion, but the mean radio galaxy SED declines from rest 1.6 to 3 m and then rises from 3 to 8 m. The radio galaxies are on average a factor 3–10 less luminous in this spectral range than the quasars. These characteristics are consistent with composite emission from a heavily reddened AGN plus starlight from the host galaxy. The mid- infrared colors and radio to mid-infrared spectral slopes of individual galaxies are also consistent with this picture. Individual galaxies show different amounts of extinction and host galaxy starlight, consistent with the orientation-dependent unified scheme.

19 20 2.1. MOTIVATION

2.1 Motivation

When exploring the general evolution of galaxy populations across cosmic times, a particular challenge is to distinguish between black hole and star-forming activity. Star formation and obscuring dust go hand in hand, and black-hole-driven active galactic nuclei (AGN) are also surrounded by dust mainly distributed in a disk/torus-like geometry (Antonucci 1993). There is evidence that AGN mainly power the near- and mid-IR emission (NIR, ∼2 m; MIR, ∼10 m) from hot nuclear dust, while starbursts contribute mainly to the far-infrared (FIR, ∼60 m) luminosity (Rowan-Robinson 1995; Vernet et al. 2001; Schweitzer et al. 2006). Using the MIR/FIR luminosity ratio as an indicator for the relative AGN and starburst contributions, numerous studies have found an increase of AGN/starburst activity with total luminosity and redshift, but the validity of this trend is still under discussion because of selection effects on the various samples. More se- riously, an unfavorable AGN orientation could cause MIR obscuration (e.g. Pier & Krolik 1993), leading to a fundamental observational degeneracy: a low MIR/- FIR luminosity ratio can be due to either a high star-forming contribution or to an AGN in which the hot dust is obscured. The spectral energy distribution (SED) of an obscured AGN may thus mimic that of a starburst-powered source. While this degeneracy has now been widely examined at low/intermediate luminosity and redshift (z < 1), it has still to be explored for the most luminous sources at high redshift (z > 1). In order to assess galaxy and AGN evolution in the universe, we therefore need to understand this AGN/starburst degeneracy for a population of luminous high-redshift sources. A crucial step towards this is to study the orientation dependence of the NIR and MIR emission of high-redshift AGN. Orientation-dependent effects can only be tested and quantified with AGN samples having type 1 (unobscured) and type 2 (obscured) subsamples matched in isotropic emission. The clean AGN tracers — optical, [O III] λ5007 A,˚ NIR, and X-ray (<10 keV) — all fail to fulfill this requirement. The [O II] λ3727A˚ emission, while isotropic (Hes et al. 1993), is probably dominated by extended starbursts and shocks (Best 2000) rather than by the AGN. Therefore, the only feasible way is low-frequency (meter-wavelength) radio selection because the integrated emission from the radio lobes is optically thin and essentially isotropic. This makes radio-loud AGN particularly attractive for studying orientation-dependent properties at other wavelengths and, after sorting out the influence of radio jets/lobes on the emission, for generalizing conclusions about orientation-dependent effects CHAPTER 2. NEAR- AND MID INFRARED PHOTOMETRY 21 to the much larger population of radio-quiet AGN. The brightest low-frequency-selected AGN sample is the 3CR compilation Spinrad et al. (1985). The powerful double-lobed radio galaxies (henceforth sim- ply called radio galaxies) are supposed to be misaligned quasars Barthel (1989). Based on IRAS co-added scans and a few individual detections, Heckman et al. (1992, 1994) already noted an average MIR/FIR difference between 3CR quasars and radio galaxies. More comprehensive MIR and FIR spectrophotometry from ISO is in hand (as compiled by Siebenmorgen et al. (2004) and by Haas et al. (2004)) as well as from Spitzer (e.g., Shi et al. (2005); Haas et al. (2005); Ogle et al. (2006); Cleary et al. (2007), providing a basis to study the z < 1 3CR objects. These sources are, however, a factor of five less radio-luminous on average than the most powerful radio sources seen at higher redshift, and the lower indicated accretion power may reflect different source physics. The higher-luminosity population can be sampled by the 3CR sources at 1

H0 = 71 km/s/Mpc, Ωm = 0.27 and ΩΛ = 0.73.

2.2 Observations and Data

With the Spitzer Space Telescope (Werner et al. 2004), we have obtained the entire sample of 64 high-z 3CR sources using the instruments IRAC (3.6–8.0 m, Fazio et al. 2004), the IRS blue peak-up array (16 m, Houck et al. 2004) and MIPS (24 m, Rieke et al. 2004). Most observations are performed in our guaranteed time program (PID 40072; PI G. Fazio) with on-source exposure times 4×30 s (each IRAC band), 4×14 s (IRS), and 10×10 s (MIPS). A few sources have been observed in other programs, and we use the published photometry if available (e.g., PID 3329; PI D. Stern, Seymour et al. 2007). For IRAC, we used the basic calibrated data products (BCD, version S16) and co-added them to 0′′.869 pixels using the latest version of IRACProc (Schuster et al. 2006). This optimally handles the slightly under-sampled IRAC PSF in order to assure accurate point-source photometry. For IRS, we used the post- 22 2.2. OBSERVATIONS AND DATA

1047 3CR at z > 1 298 191 9 257

46 220.2 119 m, erg/s ) 469.1 10 68.2 µ 305.1

208.1

( rest 8 10 173 ν

L quasars galaxies ν Si absorption 1044 1045 ν Lν ( rest 178 MHz, erg/s )

Figure 2.1: Infrared versus radio luminosity of the 3CR sample at z > 1 prior to normalization. ’x’ symbols denote quasars; circles and squares denote radio galaxies. Superposed crosses indicate radio galaxies that show evidence of silicate absorption (§2.3.1). The vertical long-dashed lines mark the range of our luminosity-matched quasar and radio galaxy subsamples. The dotted lines indicate L8m/L178 MHz ratios of 1, 10, and 100. The radio galaxies are grouped into several SED classes in Fig. 2.3 and §2.3.1. The color-coding and symbols are: green circle (A), red circle (B), red square (C), blue square (D), blue circle (E). The two low-excitation radio galaxies 3C 68.1 and 3C 469.1 are labeled with their 3C numbers, as are sources outside the luminosity range of our analysis.

BCD pipeline product, version S16. For MIPS, we used custom routines to modify the version S17 BCD files to remove instrumental artifacts (e.g., residual images) before shifting and co-adding to create the final mosaics. All sources are well seen on the images in all filters. The sources were extracted and matched using the SExtractor tool Bertin & Arnouts (1996). We used sufficiently large apertures so that aperture corrections are small (<5%). The photometric errors are typically smaller than 10% but increase for faint sources; exceptions are 3C 225A and 3C 294, where nearby bright stars make the photometry uncertain in the shorter IRAC bands. As of 2008 April, 24 quasars and 38 radio galaxies have been observed, covering CHAPTER 2. NEAR- AND MID INFRARED PHOTOMETRY 23 the complete high-z 3CR sample with the exception of the quasar 3C 245 and the radio galaxy 3C 325. All 62 sources have IRAC measurements and are observed in at least one of the 16 and 24 m bands (54 sources at 16 m and 60 sources at 24 m). For the analysis it is desirable to compare rest frame SEDs with the same wavelength sampling. Depending on the redshift of our sources (1 1.5 × 10 erg/s, respectively. We have also excluded the quasar 3C 418 because of its flat radio spectrum (low-frequency spectral index

α178 ≈ 0), while all other sources have steep radio spectra (−1.1 . α178 . −0.6).

The resulting mean radio luminosities of the sample galaxies are L178 = (5.35 ± 2.53) × 1044 erg/s for quasars and (5.55 ± 2.34) × 1044 for radio galaxies.

While the quasar and radio galaxy distributions match very well in L178, a proper analysis of orientation-dependent eﬀects requires also that the individual SEDs are normalized by the radio luminosity, which serves as a tracer for the intrinsic AGN strength. Therefore, we have normalized each SED to the sample mean 178 MHz luminosity; after normalization each object has L178 = 5.4 × 44 10 erg/s. Because of the good L178 match of the samples, it turned out that 24 2.3. RESULTS AND DISCUSSION

the net eﬀect of this normalization on the results is small.

2.3 Results and Discussion

2.3.1 Radio galaxies as obscured quasars

The NIR–MIR SEDs of quasars are all very similar in shape, as shown in Fig- −1 ure 2.2. The SEDs can be described by a single power law Lν ∝ ν , consistent with previous results for lower-redshift objects (e.g., Elvis et al. (1994)). The dispersion of the SEDs is essentially caused by diﬀering ratios of MIR to radio luminosity. Some quasars exhibit small (10-20%) bumps around 5 m explainable by distinct hot dust components.1 The power law shape of the quasar SED can naturally be explained by the superposition of centrally-heated dust components with a radial temperature gradient (1500 K >T > 300 K) as has been found also in lower luminosity type-1 AGN (Ward et al. 1987; Barvainis 1987; Rowan- Robinson 1980). Any contribution of the quasar host galaxies to the NIR–MIR SEDs appears to be outshone (factor &5–10) by the AGN dust emission. In contrast to quasars, radio galaxies display a diversity of SED shapes leading to a 50% larger dispersion around their mean SED (Fig. 2.2). Despite the dispersion, nearly all radio galaxy SEDs show a decline from rest 1.6 mto3 m and a −1.9 rise from 3 m to 8 m (Lν ∝ ν ). In addition, the average radio galaxy SED is fainter by a factor of three at 8 m and a factor of eight at 2 m relative to the quasar SED. Unlike the quasars, hot (T > 750 K) dust emission is not seen in the 2 radio galaxy SEDs. Its absence can be explained by absorption (screen AV ≈ 50) of the central dust emission. The short wavelength (λ < 3 m) component can then be explained by emission from stars in the host galaxy. Extrapolation of the mean 3–8 m SED slopes towards longer wavelengths suggests that the radio galaxy and quasar SEDs meet each other at about 25–40 m, and beyond these wavelengths extinction may be no longer relevant. As noted above, the quasar NIR–MIR SED shapes are extremely homogeneous. This is reﬂected in the narrow range of the quasars’ NIR and MIR colors.

1Despite the similarity of the infrared SEDs, the quasar population is not homogeneous at optical wavelengths: there are quasars like 3C 186 with blue optical SED and 3C 68.1 with red optical SED, as listed in NED. In the orientation-based uniﬁed scheme, 3C 68.1 could be borderline so that the broad lines are detected, but most of the UV-optical continuum is absorbed. 2The reddening curve used is a compromise between the latest results for Milky Way reddening and earlier data (summarized by Indebetouw et al. (2005)): AV / AH / A3µm / A5µm / A8µm / A10µm = 1 / 0.184 / 0.070 / 0.037 / 0.028 / 0.040. CHAPTER 2. NEAR- AND MID INFRARED PHOTOMETRY 25

The color-color diagram shown in Figure 2.3 illustrates the diﬀering SED types. In this diagram, quasars populate a distinct locus (“E”), while radio galaxies show wider dispersion as mentioned above. According to their location in the color-color diagram, we have grouped the radio galaxies into ﬁve classes described below. Their SED shapes are illustrated in Figure 2.4.

A) Four sources at the high end of the 3m/1.6m ratio (above the dotted and dashed boxes in Fig. 2.3): basically, they have quasar-like SEDs, but the hottest dust emission at about 1–2 m appears to be absorbed (screen

AV ≈ 5) leading to a redder 3 m/1.6 m color compared to quasars.

B) The bulk of radio galaxies (20 sources) have declining 1–3 m SEDs with a steep 3–8 m rise. Their colors can be explained by a heavily reddened AGN plus an added component of starlight contributing at 1.6 m. If this

explanation is correct, the direction of the extinction vector AV becomes meaningless because host galaxy starlight will not be aﬀected by extinction near the nucleus. Instead, vertical position in the plot is determined by the relative contributions of starlight and AGN light, while horizontal position measures the amount of extinction (to the extent the underlying AGN SEDs

are the same). As noted above, AV ∼ 50 mag is required to explain the colors.

C) Three sources at the low end of the 3 m/1.6 m ratio (below the long dashed line in Fig. 2.3): Their SEDs show a very strong host galaxy contribution at 1.6 m, and starlight exceeds the dust luminosity even at wavelengths as long as 3.5 m. In principle, class C is similar to class B but with stronger host galaxy contribution.

D) Three sources immediately below the dotted box in Fig. 2.3: Their SEDs can be explained by a slightly reddened AGN (similar to class A) plus an added component of starlight contributing signiﬁcantly at 1.6–3 m.

E) Three sources with quasar-like SED colors (inside the dotted box in Fig. 2.3): Their SEDs overlap with the low luminosity end of the quasar SEDs. In the orientation-based uniﬁed scheme, these sources could be borderline so that most of the dust torus is visible but the broad line region and the UV-optical continuum are obscured.

While the rest 8–10 m range is poorly sampled, eight galaxies show declines in this range that could be caused by silicate absorption. (The MIPS-24 ﬁlter, 26 2.3. RESULTS AND DISCUSSION

50% transmission at 20.8–29.3 m, requires z . 1.8 for the silicate feature to fall into its range.) One of these sources (3C 469.1, z = 1.336) has an IRS spectrum

available. It shows a broad silicate absorption with optical depth τ9.7 ≈ 0.55

corresponding to AV ≈ 10, consistent with its position in Fig. 2.3. This supports the view that the SED declines in the other radio galaxies are due to silicate absorption, too. The photometric silicate absorption sources show a wide range of colors (Fig. 2.3), but only one galaxy (3C 469.1) is on the blue (right) side, where low-extinction sources reside. The important conclusion is that the silicate feature requires considerable extinction to be present in at least some of the radio galaxies, and this is largely independent of the SED class.3 If radio galaxies are misaligned quasars, as proposed in the unified scheme, reddening of individual galaxies should be correlated with their extinction. Fig- ure 2.5 shows that this is indeed the case. Quasars populate a distinct region of this diagram characterized by high MIR/radio and blue NIR–MIR colors. Most radio galaxies spread towards fainter MIR/radio and redder NIR/MIR. Under the reasonable assumption that the emission at 5–8 m is not affected by the host galaxy, de-reddening along the direction of the extinction vector can place each radio galaxy inside the region populated by quasars. Thus individual radio galaxies can be explained as reddened quasars, consistent with the orientation- dependent unified scheme.

The typical amount of radio galaxy reddening, AV ≈ 50 for an obscuring 22 −2 screen (Fig. 2.5), corresponds to a hydrogen column density NH ≈ 9 × 10 cm −22 −2 (for AV = 5.6 × 10 mag/cm — Seward (1999). This is close to but below 24 −2 the Compton-thick limit (NH = 10 cm ). Screen extinction is a simpliﬁcation, and one may expect a more complex geometry. If emitting dust particles are spatially mixed with the absorbing ones, the amount of dust has to be higher for the same observed reddening, typically by a factor 3–5.4 Thus there could very

3The photometric silicate absorption sources are 3C 68.2, 3C 222, 3C 249, 3C 250, 3C 266, 3C 305.1, 3C 324, and 3C 469.1. These galaxies lie in the redshift range 1.08 < z < 1.83, suggesting that in this range the broad band 16 m/24 m filter combination is able to register silicate absorption, if strong enough. For comparison, this redshift range contains 20 more radio galaxies with 16 and 24 m photometry available but without silicate absorption signatures in their broadband SEDs. Four of these sources (3C 13, 3C 266, 3C 267, 3C 356) have IRS spectra available, but significant silicate absorption is detected only in one of them (3C 267, z = 1.14, τ9.7 ≈ 0.2). At low redshift, the detection of silicate absorption appears not to be directly correlated with other absorption signatures, perhaps because of complex geometry and/or varying silicate dust abundance (Haas et al. 2005; Ogle et al. 2006; Cleary et al. 2007). A detailed analysis of the high-z 3CR spectra and the photometric detectability of silicate features will be presented elsewhere (Leipski et al. submitted.). 4 The transmission factors are exp(−τλ) and (1 − exp(−τλ))/τλ for the screen and the mixed case, respectively, with τλ = 0.916 × Aλ (Disney et al. 1989). CHAPTER 2. NEAR- AND MID INFRARED PHOTOMETRY 27 well be enough gas present to render the AGN Compton-thick. There is, unfortunately, no independent measurement of reddening for individual galaxies, nor is it certain that a Galactic reddening curve applies to AGN. Thus it is still an open question whether after de-reddening there will remain a difference in the 8 m/178 MHz ratio between radio galaxies and quasars. If such a difference remains, with quasars having a higher 8 m/178 MHz ratio than radio galaxies, then either our screen extinction premise is too simple or the MIR luminosity of quasars is enhanced by an additional — potentially non-thermal — contribution. Our Chandra X-ray observations and IRS spectra of the Si feature (Leipski et al. submitted) of a subset of the sample will provide independent estimates of the extinction towards the nuclei (Wilkes et al., in prep.). To summarize, while quasars exhibit a uniform SED shape which can be explained by a centrally heated dust distribution, radio galaxies show a diversity of SED shapes. In all cases, however, the radio galaxy SEDs are consistent with being intrinsically a quasar modified by absorption of the dust emission and addition of some amount of host galaxy starlight.

2.3.2 Evolutionary eﬀects and non-thermal contributions

Studying powerful 3CR sources at z < 1, Ogle et al. (2006) found evidence for a population of accretion-inefficient radio galaxies, in which the jet/lobe may be powered by extraction of rotational black hole energy. These sources, mainly optically-classified low-excitation radio galaxies (LERGs), have a 15 m luminos- 43 ity below 8 × 10 erg/s and a luminosity ratio L15m /L178MHz < 10. In contrast, with the reasonable assumption that L8m . L15m, all our z > 1 sources have 44 observed MIR luminosity L8m > 5 × 10 erg/s, which is expected to be even higher after de-reddening. Also, the two LERGs (3C 68.2, 3C 469.1) in our sample show a high luminosity ratio L8m /L178MHz > 10 comparable to that of quasars (Fig. 2.1). From this, our data do not support the existence of an accretion- inefficient population among the powerful 3CR sources at z > 1. A possible explanation for the deficit of optical high-excitation line luminosity (for instance [O III] λ5007 A)˚ in our two LERGs may be extinction of the narrow-line region. On the other hand, some of our radio galaxies with very strong host contribution (plotted as squares in Fig. 2.5) are expected after de-reddening to lie at the low end of the L8m/L178MHz distribution. Hence compared with the strength of both the host and the radio lobes, these galaxies are relatively weak in the MIR and may represent a population at the beginning of a different evolutionary state. 28 2.4. CONCLUSIONS

Some authors have attributed the excess emission of quasars compared to radio galaxies to non-thermal emission from synchrotron jets. For example, Cleary et al. (2007) fitted the SEDs and spectra of 3CR sources at 0.5 1, it would show up in Fig. 2.5 as an offset by about a factor of two between de-reddened radio galaxies and quasars. This conclusion is, however, dependent on both the reddening law and on the radiative transfer and thus the geometry of the emitting region. In order to draw definite conclusions about any MIR luminosity excess, detailed radiative transfer modeling is required (see chapter 4). Spherically symmetric models are wholly inadequate for this purpose. In an inclined disk-like system, some fraction of the MIR emission is likely to have very little obscuration while the bulk of the MIR emission is heavily obscured, and no spherical model can properly account for this geometry. All we can say at the moment is that our data appear consistent with a simple thermal interpretation and show no evidence for a non-thermal component.

2.4 Conclusions

The 3CR sample at 1

1) Quasars all have similar energy distributions in the rest frame 1.6–10 m range, and their ratio of MIR to radio luminosity is also nearly constant. This is consistent with results seen previously in lower-redshift samples.

2) The rest frame 1.6–10 m SEDs of radio galaxies can be explained as reddened quasars, consistent with orientation-dependent uniﬁcation. Various amounts of extinction of the AGN emission combined with addition of host galaxy starlight can explain the diversity of radio galaxy SEDs.

3) If the extinction is suﬃciently large, there is no need to invoke a beamed synchrotron contribution to explain the MIR luminosity diﬀerence between CHAPTER 2. NEAR- AND MID INFRARED PHOTOMETRY 29

quasars and radio galaxies. The actual amount of extinction has to be derived from additional observations.

4) The above results hold also for splitting our sample in redshift and luminosity; within our sample we do not ﬁnd any trends with redshift or luminosity.

5) At rest frame 8 m, quasars are 3 times more luminous than radio galaxies. If this diﬀerence applies also to high-redshift, radio-quiet AGN, then MIR (24 m) surveys are strongly biased in favor of type-1 and against type-2 AGN. This will make it very diﬃcult to resolve the AGN/starburst degeneracy with only broadband SEDs, and spectral line diagnosis will be required.

While our near-mid-IR SEDs provide a fundamental set of high luminosity AGN templates, we expect to derive more stringent conclusions from proper two- dimensional radiative transfer modeling in combination with Spitzer MIR spectra, Chandra X-ray observations, and Herschel far-IR/sub-mm data. 30 2.4. CONCLUSIONS

Mean SEDs

quasar 102 erg/s ]

44 radio galaxy [ 10 ν

L 1 quasar reddened by AV = 50 ν 10 host galaxy

1 10 λ rest [ µm ]

Figure 2.2: Rest frame quasar and radio galaxy SEDs normalized by rest 178 MHz 44 luminosity (L178 = 5.4×10 erg/s). Symbols connected with thick blue and red lines show the mean SEDs for quasars and radio galaxies, respectively. The thin dotted blue and red lines indicate the dispersion (upper and lower quartiles) around the mean SEDs; the mean ratio of upper/lower quartiles are 2.3 (quasars) and 3.4 (radio galaxies). The radio galaxy SED can be explained by the sum (black long-dashed line) of a reddened quasar (blue long-dashed line) and starlight from the host galaxy (thin black solid line). The long-dashed lines have been shifted slightly to make them visible in the plot. The diﬀerence between radio galaxies and reddened quasars at 10 m may be due to the silicate absorption feature which may escape detection in our broad band photometry. CHAPTER 2. NEAR- AND MID INFRARED PHOTOMETRY 31

A V

V = 5 469.1 4 267 A

m (rest) E µ B 1 68.2 D

m / F 1.6 C µ quasars galaxies

F 3 Si absorption 0.3 0.5 0.8 F 5 µm / F 8 µm (rest)

Figure 2.3: NIR/MIR color-color diagram. The radio galaxies are grouped into ﬁve classes labeled A–E as explained in §2.3.1. Symbol color-coding is the same as in Figure 2.1. Radio galaxies with a photometric signature for silicate absorption are additionally marked with an underlying plus. The two sources 3C 267 and 3C 469.1 with spectroscopically-detected silicate absorption are labeled, as are the two low-excitation radio galaxies 3C 68.1 and 3C 469.1. The error bar in the upper left corner represents a color rms of 15%. The AV arrow indicates screen extinction with the reddening law given in §2.3.1. 32 2.4. CONCLUSIONS

Diversity of SEDs

quasar A 102 erg/s ] 44 E B D Si absorption [ 10 ν L 1 host galaxy ν 10 C

1 10 λ rest [ µm ]

Figure 2.4: Mean SED of each radio galaxy class as identiﬁed in §2.3.1. The SEDs have been normalized to the mean 178 MHz luminosity. The mean quasar SED is also shown for comparison. The dispersion around each SED (measured as mean ratio of upper/lower quartile) is: 2.3 (quasars), 2.0 (A), 3.2 (B), 2.4 (C), 1.7 (D), 1.5 (E), and 4.8 (silicate absorption). CHAPTER 2. NEAR- AND MID INFRARED PHOTOMETRY 33

10−3

= 50 A V V −4 10 267 469.1

68.2

10−5 m / F 178 MHz (rest)

µ quasars galaxies Si absorption F 8 0.3 0.5 0.8 F 5 µm / F 8 µm (rest)

Figure 2.5: Infrared-radio color-color diagram. The dotted line marks the region occupied by quasars. The color-coding and symbols of radio galaxies correspond to those in Figures 2.1 and 2.3. The two sources 3C 267 and 3C 469.1 with spectroscopically-detected silicate absorption are labeled as well as the two low- excitation radio galaxies 3C 68.1 and 3C 469.1. The error bar in the upper left corner represents an rms of 15%. The AV arrow indicates screen extinction with the reddening law given in §2.3.1. 34 2.4. CONCLUSIONS Chapter 3 The cluster search

This part represents one of the developed papers during my PhD thesis. My contribution in this large collaboration of many coauthors was mainly:

• Developing a pipeline to extract the photometry of all objects in the Spitzer mosaic images.

• Develop an interactive data tool to provide photometric redshifts for all objects in the center and control ﬁelds (Fig 3.1).

• Interactive inspection, veriﬁcation and improvements of the SEDs of the cluster candiates by eye.

• Technical data analysis (cluster properties, plots).

Abstract: Observations of the z = 1.53 quasar 3C 270.1 with the Spitzer Space Telescope at 3.6–24 m and with the 6.5-m MMT in the z′- and Y -bands allow detection of potential cluster members via photometric redshifts. Compared with nearby control ﬁelds, there is an excess of ∼11 extremely red objects (EROs) at 1.33 ≤ zphot ≤ 1.73, consistent with a proto-cluster around the quasar. The spectral energy distributions (SEDs) of 3/4 of the EROs are better ﬁtted with passive elliptical galaxies than with dust-reddened starbursts, and of four sources well- detected on an archival HST snapshot image, all have undisturbed morphologies.

3.1 Motivation

Galaxy groups and clusters are important for the understanding of large scale structures and their evolution. As a few thousands of clusters are known in the

35 36 3.1. MOTIVATION

Figure 3.1: Outcome of the photometric fitting routine. The left column shows the two dimensinal fitting space of redshift and luminsosity. This helps to check the quality of the fitting procedure. The right column shows the logarithmic plot of the data, with the wavelenth on the x-axis and νLν on the y-axis. The top left small box contains information like the redshift, χ2, position and distance to the 3C source. The small pictures on the left shows a small region around the object to check for double sources or other peculiarities. CHAPTER 3. THE CLUSTER SEARCH 37 local universe up to redshift one, the main problems are that the flux which reaches the earth is dramatically weaker for distant clusters at redshift higher than one. The other fact is that it is challenging to discriminate the distant cluster members from the foreground galaxies. The most distant clusters till now have been found via X-Ray observations (Mullis et al. 2005; Bremer et al. 2006; Stanford et al. 2006; Fang et al. 2007). The second approach to find clusters at high redshift is to search for over densities in wide area optical-infrared surveys. A few of these surveys are the Spitzer-IRAC shallow survey, the UKIDSS ultra deep survey, the HIRCOCS/COSMOS, and the Spitzer First look Survey. A different way is to look around high redshift radio sources. For the local universe it is known, that these objects exist in massive environments. Therefore they should trace over density regions, in other words galaxy groups or clusters. The advantage of this method is that the redshift is known and this limits the observational effort. For very high redshifted radio sources (2 < z < 5) this method has revealed an excess of Lyman break galaxies (Kurk et al. 2000). These Lyman galaxies show no real concentration, which is consistent with a cluster that is forming but not yet virialised (Intema et al. 2006). There is evidence that the high redshift radio sources have an over density of EROs (Best (2000)). But it is not known if these objects are passive ellipticals or dust reddened star forming galaxies. So the question, if these giant ellipticals are already formed or still in evolution, is not answered till now. If these ellipticals already exist, the formation must be at larger redshifts. On the other hand if these EROs are star-forming Galaxies we probe the epoch in which they form. With our large Spitzer maps (> 5’) we should be able to detect the whole cluster with sizes of about 1’ - 2’ at that redshift range and our MIPS images should in principle identify the dusty star-forming galaxies.

3.2 Clustering around 3C 270.1

3.2.1 Observations and data

We analyse Spitzer maps of 3C 270.1 taken with the Infrared Array Camera (IRAC) and Multiband Imaging Photometer (MIPS) on the Spitzer space telescope. The exposure times on the sources were 4 × 30s for each IRAC band, and 10 × 10s for the MIPS bands. The size of the maps is about 4′ × 4′ for the IRAC camera. In addition we also obtain a comparison ﬁeld northeast of the quasar for the 3.6m and 5.8m band. For the 4.5m and 8.0 m we get the data for a ﬁeld 38 3.2. CLUSTERING AROUND 3C 270.1

Southwest of the quasar. The center of the comparison fields are shifted of about 6′ from the quasar. The 3σ detection limit of the maps is 4Jy for the first two IRAC bands (3.6m,4.5m),10 Jy for the last two IRAC bands (5.8m,8.0m) and 100Jy for the 24m MIPS band. We also obtain z’ and Y-band images at the 6.8m MMT using MegaCam (30’ FOV). These filters encompasses the 4000 A break at the redshift of the quasar. The sensitivity for the both MMT filter is better than 1 Jy with exposure times of 40 minutes for the z’ filter and 90 minutes for the Y-Band. The reduction of the images was achieved with interactive analysis tools. IRAC mosaics were corrected for residual images from previous observations by making object masked, median-stacked co-adds with all the science frames and the subtract them from the science frames before final co-add.

3.2.2 Results

Cluster galaxy candidates

Cluster galaxies around 3C 270.1 should lie at the redshift of the quasar. We

determined photometric redshifts zphot by ﬁtting the observed spectral energy distributions (SEDs) of the 184 galaxies detected at z′ with two basic templates, an elliptical galaxy and a dusty starburst galaxy.1 For the elliptical galaxy, we used NGC 221, which has a strong 4000 A˚ break. The NGC 221 spectral template from Kinney et al. (1996), covering the rest-frame wavelength range 3000-8000 A,˚ was smoothly extrapolated to longer wavelengths by a 4000 K black-body. For the starburst galaxy template, we used the ultra-luminous infrared galaxy Arp 220 with photometry from the NASA Extragalactic Database (NED) and the Sloan Digital Sky Survey (SDSS, DR6) and mid-IR spectra from Spitzer/IRS. We also tested the dust-reddened star-forming galaxy template M 82, but the results were

similar to those for Arp 220 (differences in zphot less than ∆z = 0.1), and therefore we present only the Arp 220 results. The accuracy of photometric redshifts can be estimated, for instance, from the Spitzer Wide-Area Infrared Extragalactic Survey (SWIRE), which has spectroscopic redshifts available for many of its sources. In SWIRE, 7 filters (5 optical and 2 IRAC) were sufficient to discriminate between 8 templates (ranging from

blue to red galaxy types) and to determine zphot with an rms of ∆z/(1+z) = 3.5% (Rowan-Robinson et al. 2008). We have fewer ﬁlters than SWIRE, but our task

1We also tried type-1/-2 AGN templates, but none of the SEDs is consistent with such templates at redshift z ∼ 1.5 except 3C 270.1 itself. CHAPTER 3. THE CLUSTER SEARCH 39 is easier because we only have to determine whether the photometry is consistent with z = 1.53 or not. Furthermore, we consider only extremely red sources and only two templates and therefore suggest that we can reach accuracy similar to that of SWIRE. At z = 1.5, an accuracy (rms) of about 4% corresponds to ∆z = 0.1. This is much larger than the expected redshift dispersion within a cluster (∆z < 0.01). While the SEDs can provide cluster galaxy candidates within appropriate redshift bins around zQSO, confirmation of a cluster will re- quire spectroscopic redshifts of at least a sample of the candidates. (We have been granted NIR multiobject spectroscopy with MMIRS at the MMT to look for redshifted Hα emission, but observation have still to be carried out) Figures 3.2 and 3.3 show examples of SEDs and template fits to cluster galaxy candidates. The most striking SED features are the steep rise from z′- to Y -band and beyond and the decline between 3.6 and 8.0 m. The slopes of these features determine the redshift. We performed the template fits over a grid in redshift (dz = 0.01) and intensity. For a red SED with F (3.6)/F (z′) ≈ 30, the chi square contour plots exhibit a sharp minimum, suggesting that for a given template the achievable accuracy of zphot is about ∆z = 0.1. As illustrated in Fig. 3.3, even for SEDs with fewer than six data points the accuracy of zphot should be not worse than ∆z < 0.20. Therefore, we have chosen as cluster galaxy candidates the 29 objects for which either the elliptical or the starburst template yields

1.33 < zphot < 1.73. The basic conclusions on the clustering of EROs around 3C270.1 remain unchanged for other ranges |∆z| between 0.15 and 0.25. Usually the 0.9–8.0 m SEDs can be fitted very well with both the elliptical and the starburst template. Both templates are extremely red, and therefore no other galaxy type is likely to fit the SEDs. The examples in Figures 3.2 and 3.3 illustrate the ability to identify the two galaxy types at z ∼ 1.53. The resulting fit parameters for the 29 cluster candidates are listed in Table 3.1. The photometric redshifts differ systematically for the elliptical and starburst templates. On average, zell is higher by 0.27 ± 0.1 than zSB. Among the 29 cluster candidates, the elliptical template yields better fits (smaller chi square values) for 22 sources (75%, Table 3.1). All of these sources have lie in the redshift bin 1.33 < zSB < 1.73. Among the remaining 7 cluster candidates, those

3 sources which favor the starburst ﬁt lie in the redshift bin 1.33 < zSB < 1.73, 2 2 too. These sources have χSB < 0.3χell (object numbers 3, 5 and 29 in Table 3.1). The remaining 4 sources (object numbers 11, 14, 17 and 28) are not detected at 5.8 and 8.0 m, and they do not clearly favor the starburst template, so that – as ellipticals – they are cluster galaxy candidates. 40 3.2. CLUSTERING AROUND 3C 270.1

−16 10 elliptical ] 2 [ W / m ν F 10−17 ν

10−16 dusty starburst ] 2 [ W / m ν F 10−17 ν

1 10 Wavelength [ µm ]

Figure 3.2: Observed spectral energy distributions for two good examples of the 29 cluster candidates. The IRAC and MMT data are marked with filled circles and 1σ error bars. The MIPS 24 m data point is marked with a filled circle, too; in the upper panel (elliptical source) it is a 3σ upper limit, in the lower panel (dusty starburst source) it is a 2σ detection also visible on the map. The horizontal bar indicates the 24 m pass band for comparison with the silicate absorption feature. HST photometry is marked with an open circle; it is not available for the dusty starburst source. Solid lines show the elliptical galaxy (NGC 221) fit for the source in the upper panel and the dusty starburst (Arp 220) fit for the galaxy in the lower panel. Dotted lines show the alternative template for each galaxy. The upper and lower panel shows object numbers 23 and 3, respectively, as listed in Table 3.1). CHAPTER 3. THE CLUSTER SEARCH 41

10−17 z = 1.73 χ2= 0.16

] 2 z = 1.53 χ2= 0.07 z = 1.33 2

[ W / m χ

ν = 0.16 F ν 10−18

z = 1.73 χ2= 0.28

] −17 2 10 z = 1.53 χ2= 0.04 [ W / m ν

ν z = 1.33 χ2= 0.26

10−18 1 10 Wavelength [ µm ]

Figure 3.3: Observed spectral energy distributions for two cluster candidates with poorer data quality. The IRAC and MMT data are marked with filled circles and 1σ error bars. The sources are detected only in the z′-band (in the lower panel also in the Y -band) and at 3.6 and 4.5 m, but the upper limits at 5.8 and 8.0 m help to constrain the redshift fits. As in Fig. 3.2, the upper panel shows a source (object 11) which is preferably fit by the elliptical template and the lower panel one (29) fit by the starburst template. The dotted and dashed lines show the preferred templates at different redshifts (indicated in the figure), showing that the accuracy of the photometric redshifts should be dz . 0.2. 42 3.2. CLUSTERING AROUND 3C 270.1

Quasar field 102

28 color cand.

101 F 3.6 / z' 100

29 SED cand. −1 other 10

Control field 1 102

17 color cand.

101 F 3.6 / z' 100

16 SED cand. −1 other 10 100 101 102 103 F 3.6 [ µJy ]

Figure 3.4: Color-magnitude diagram F (3.6)/F (z′) versus F (3.6), for the quasar ﬁeld (upper panel) and control ﬁeld 1 (lower panel). Sources with zphot = 1.53 ± 0.20 (SED determined candidates marked with circles) concentrate in a distinct color range 15 < F (3.6)/F (z′) < 52 indicated by the horizontal dotted lines. The number of color determined candidates is also given. CHAPTER 3. THE CLUSTER SEARCH 43

The cluster galaxy candidates occupy a limited range of z′ − [3.6] color as shown in Figure 3.4. While the candidates were determined from multi-band SED ﬁtting (usually 4–6 bands), selecting a color range of 15 < F (3.6)/F (z′) < 52 would have given nearly the same sample. The sources redder than this color range are probably at larger redshift (z > 1.8), while bluer sources are consistent with being either foreground objects or unreddened star-forming galaxies at the quasar redshift. Our EROs with F (3.6)/F (z′) > 15 (corresponding to z′ −[3.6] > 5 mag in the Vega system) are likely to also obey the standard ERO deﬁnition of R − K > 5 mag (Wilson et al. (2004)).

Sky distribution and surface density

Figure 3.5 shows the sky distribution of the cluster candidates. Spreading over a diameter of more than 2′, they form at most a loose concentration around the quasar, indicating a proto-cluster rather than a virialised, concentrated system. The peak density appears to lie ∼30′′ to the east of 3C 270.1, but it is difficult to be precise with so few galaxies. The 3C 270.1 proto-cluster is larger in angular size than the X-ray clusters at z = 1.45 and z = 1.22 found by Mullis et al. (2005) and Bremer et al. (2006). These clusters have an extent of less than 1′ in both their galaxy distributions and their X-ray sizes. We have also analyzed the two comparison fields taken with IRAC and covered by the MegaCam z′ image (but not the SWIRC Y image). Details of the analysis are described in Appendix 3.2.2. The surface density of possible z = 1.53 ± 0.20 galaxies is less than about 18 objects in each control field of area 15.2 arcmin2 (1.2 per arcmin2). Thus, in the redshift range z = 1.53 ± 0.20, the central field surrounding the quasar shows an excess of at least about 29 − 18 = 11 sources, i.e., 60% over the comparison fields. Figure 3.6 shows the radial surface density plot of the 29 EROs in the redshift range z = 1.53 ± 0.20. The surface density peaks inside the central 50′′ radius and declines steadily with increasing distance down to the surface density of the control fields. This provides further evidence that there is an excess of zphot ≈ 1.53 galaxies near 3C 270.1. The radial overdensity is also present when using circles around the centroid east of the quasar position. Existing X-ray data on 3C 270.1 are inconclusive about the presence of a cluster. The quasar itself is clearly detected as a point source in our 10 ks Chandra data (Wilkes et al., in preparation). Weak extended X-ray emission (<20′′) is also present, but most of the counts come from the position of the southern radio lobe 44 3.2. CLUSTERING AROUND 3C 270.1 45' 30" o

33 IRAC 43' 30" o 33 DEC J2000

HST 41' 30" o 33

12h 20m 45s 12h 20m 35s 12h 20m 25s RA J2000

Figure 3.5: Sky distribution of the 29 candidate cluster galaxies around the quasar 3C 270.1 (marked with a cross). The triangle shows the apparent centroid of the cluster galaxy distribution. The star in the southwest marks the starburst candidate whose SED is shown in Fig. 3.2. The solid lines surround the areas covered by IRAC and HST frames. The dotted circles of radius 50′′, 100′′, and 150′′ outline the areas considered in Fig. 3.6, they are centered around the quasar. At z = 1.53, 50′′ corresponds to 427 co-moving kpc. CHAPTER 3. THE CLUSTER SEARCH 45

4 2 3C270.1 field

2 Number / arcmin

1 control fields

0 0 50 100 150 Distance [arcsec]

Figure 3.6: Surface density of the 29 cluster galaxy candidates versus projected distance from the quasar 3C270.1 (solid line through fat dots with Poisson error bars). The radial bins centered around the quasar are outlined in Fig. 3.5, and the surface density of the outermost annulus has been corrected for the area not covered by IRAC. For comparison, the long-dashed histogram shows the surface density of cluster galaxy candidates versus projected distance from centroid (the triangle in Fig. 3.5). The dotted line indicates the mean surface density in the two control ﬁelds. 46 3.2. CLUSTERING AROUND 3C 270.1

Table 3.1: Fit parameters for the 29 cluster galaxy candidates in the quasar ﬁeld.

[1] [2] 2 [3] 2 Object RA J2000 Dec J2000 n zell χell zSB χSB 1 12 20 27.90 33 43 11.5 3 1.47 0.43 1.16 0.53 2 12 20 28.72 33 41 51.9 6 1.50 0.06 1.20 0.22 3 12 20 29.20 33 42 10.5 6 1.55 0.51 1.35 0.15 4 12 20 30.06 33 43 18.1 6 1.47 0.19 1.00 0.85 5 12 20 30.66 33 44 50.0 4 1.90 0.17 1.60 0.03 6 12 20 30.73 33 44 39.4 6 1.53 0.12 1.24 0.35 7 12 20 31.18 33 43 14.1 6 1.63 0.04 1.47 0.31 8 12 20 32.49 33 43 20.3 5 1.50 0.29 1.34 0.39 9 12 20 33.50 33 42 25.4 5 1.35 0.10 1.11 0.38 10 12 20 33.57 33 44 08.3 4 1.42 0.13 1.16 0.18 11 12 20 33.85 33 45 37.5 3 1.53 0.07 1.22 0.03 12 12 20 33.90 33 42 34.6 4 1.36 0.08 1.17 0.30 13 12 20 34.79 33 41 55.4 5 1.47 0.10 1.07 0.27 14 12 20 34.95 33 42 41.0 4 1.45 0.01 1.20 0.01 15 12 20 35.20 33 42 09.6 5 1.35 0.12 0.96 0.43 16 12 20 35.88 33 43 32.2 4 1.59 0.07 1.28 0.10 17 12 20 36.50 33 43 30.6 4 1.40 0.19 1.20 0.10 18 12 20 37.70 33 41 18.2 5 1.54 0.34 1.33 0.37 19 12 20 38.52 33 43 04.2 4 1.53 0.05 1.25 0.08 20 12 20 38.78 33 43 50.2 6 1.68 0.32 1.38 0.53 21 12 20 39.16 33 44 31.9 4 1.50 0.01 1.27 0.03 22 12 20 40.10 33 43 20.5 6 1.50 0.23 1.36 0.35 23 12 20 40.49 33 43 21.9 6 1.56 0.13 1.04 0.98 24 12 20 40.55 33 43 17.3 4 1.63 0.03 1.55 0.06 25 12 20 40.62 33 43 00.2 4 1.48 0.01 1.25 0.02 26 12 20 43.31 33 42 51.2 5 1.50 0.44 1.32 0.66 27 12 20 43.32 33 44 19.8 5 1.69 0.01 1.37 0.04 28 12 20 44.17 33 44 03.9 4 1.56 0.11 1.27 0.07 29 12 20 45.13 33 43 18.9 4 1.84 0.26 1.53 0.04

[1]Number of data points (detections) used to calculate χ2 of the fit. [2]Redshift of the fit using the elliptical template NGC 221 [3]Redshift of the fit using the starburst template Arp 220

or from between the northern radio lobe and the quasar. ROSAT data (19.3 ks, 1993 May) show a strong detection of 3C 270.1, but at ROSAT resolution not only is there no way to separate cluster gas emission from quasar emission, but the emission is also heavily blended with that of unrelated QSO B1223+338B located 0.′8 away.2 We found no XMM observations. Longer Chandra exposures

2Because of its lower redshift, z = 1.038, this QSO does not aﬀect the clustering evidence CHAPTER 3. THE CLUSTER SEARCH 47 will be needed for a deﬁnitive detection of any X-ray emission from cluster gas.

Comparison with the control ﬁelds

The quasar field has photometry in six filters, two MMT and four IRAC bands. But the two control fields have photometry in only three filters, in z′ and two IRAC bands: 3.6 and 5.8 m in control field 1, 4.5 and 8.0 m in control field 2. This makes the uncertainties on the photometric redshifts larger, and therefore comparison within the same redshift bin appears not to be straightforward. Fur- thermore, in the central field, the source detection rate at 4.5 m is lower than at 3.6 m, so that for a fair comparison the number of candidates in control field 2 has to be corrected for. In order to facilitate the comparison, we used the colors z′ − [3.6] and z′ − [4.5] and counted the number of candidates in the respective color bins. Control field 1: The colors of the elliptical and starburst template, when redshifted to z = 1.53, are F (3.6)/F (z′) = 25 and 39, respectively. In the color range 15 < F (3.6)/F (z′) < 52 we found 28 and 17 sources for the quasar and the control field, respectively (Fig. 3.4). With regard to the low number statistics, these values are consistent with the 29 and 16 sources in the quasar and the control field, having 1.33 < zphot < 1.73 (denoted as SED candidates in Fig. 3.4). This leaves an excess of 11 colour– and 13 SED–selected sources in the quasar field. Control field 2: The colors of the elliptical and starburst template, redshifted to z = 1.53, are F (4.5)/F (z′) = 24 and 43, respectively. Guided by these “ex- pectation values” and the actual z′ − [4.5] distribution of the candidates in the quasar field, we selected a color range of 13 < F (4.5)/F (z′) < 54 (Fig. 3.9, top). Now we considered the fields, where the 4.5 m source list was created using the single-mode SExtractor option, i.e., not making use of the 3.6 m information in the quasar field. We found 24 and 14 color-selected candidates for the quasar and the control field, respectively (Fig. 3.9, middle and bottom). These values are consistent with the corresponding numbers for SED-selected candidates of 23 and 12 in the quasar and the control field, respectively. Because the quasar field contains more cluster galaxy candidates, 29 via color and 29 via zphot (Fig. 3.9, top), we have scaled up the number of candidates in the control field by the factors 29/24 for the color-candidates and 29/23 for the SED-candidates. This results in 17 colour- and 15 SED-selected candidates. This leaves an excess of 12 (= 29 − 17) colour- and 14 (= 29 − 15) SED-selected sources in the quasar field. 48 3.2. CLUSTERING AROUND 3C 270.1

Quasar field 102

29 color cand.

101 F 4.5 / z' 100

29 SED cand. −1 other 10

Quasar field 2 10 4.5µm confined

24 color cand.

101 F 4.5 / z' 100

23 SED cand. −1 other 10

Control field 2 102

14 color cand.

101 F 4.5 / z' 100

12 SED cand. −1 other 10 100 101 102 103 F 4.5 [ µJy ]

Figure 3.7: Color-magnitude diagram F (4.5)/F (z′) versus F (4.5) of the quasar field and control field 2. top: quasar field with all sources detected at 4.5 m using SExtractor in double-image mode at 3.6 & 4.5 m. middle: quasar field restricted to those sources detected at 4.5 m using SExtractor in single-image mode, as was done for the control field 2. bottom: control field 2. Sources with zphot = 1.53 ± 0.20 (SED determined candidates marked with circles) concentrate in a distinct color range 13 < F (4.5)/F (z′) < 54 indicated by the horizontal dotted lines. The number of color determined candidates is also given. CHAPTER 3. THE CLUSTER SEARCH 49

Combining the control ﬁeld counts we estimate an excess of 11–12 colour- and 13–14 SED-selected candidate cluster galaxies in the quasar ﬁeld.

Nature of cluster galaxy candidates

The cluster galaxy candidates have an absolute rest-frame magnitude H ≈ K < −23.8 (AB system). An L⋆ galaxy at z ∼ 1.5 has H ≈ K = −23.6 (as determined from observations at 1

1.0 1.0 RA/Dec=12h20m33.50s/33o42'25.4" RA/Dec=12h20m33.90s/33o42'34.6"

0.5 0.5

0.0 0.0

arcsec

V V E E -0.5 -0.5

V V N N -1.0 -1.0 1.0-1.0 -0.5 0.0 0.5 1.01.0-1.0 -0.5 0.0 0.5 1.0 RA/Dec=12h20m34.95s/33o42'41.0" RA/Dec=12h20m35.20s/33o42'09.6"

0.5 0.5

0.0 0.0

arcsec

V V E E -0.5 -0.5

V V N N -1.0 -1.0 -1.0 -0.5 0.0 0.5 1.0-1.0 -0.5 0.0 0.5 1.0 arcsec arcsec

Figure 3.8: HST F702W images of four cluster galaxy candidates (object numbers 9, 12, 14 and 15 in Table 3.1). The contours are linearly spaced in steps of 10% of the peak ﬂux value (0%, 10%, ... 90%). Arrows indicate the orientation of each panel.

with the elliptical template. They have a regular shape, and none of them shows a peculiar morphology. This argues against starburst galaxy pairs and in favor of an evolved elliptical population. The remaining ﬁve of the 9 HST-detected sources are too faint to draw stringent morphological conclusions. The HST F702W photometry is of limited use because of the short exposure time, but six of the nine detected sources appear to show a rest-frame UV excess above the elliptical template (Fig. 3.3). This indicates some level of ongoing star formation activity. However, at rest wavelengths around 1 m this activity appears to be outshone by large numbers of already-evolved stars. CHAPTER 3. THE CLUSTER SEARCH 51

Having so many as ∼11 galaxies brighter than L⋆ would be very high for a local cluster, but passive evolution will cause these galaxies to become fainter by something like 3 mag3 by z = 0 to become comparable to today’s cluster ellipticals. Thus the photometric data are compatible with the existence of a cluster.

3.3 No evidence for clustering around 3C 437

In addition to the published results on 3C 270.1, I have analysed 3C 437 yielding interesting results.

3.3.1 Observations and data

We obtained Spitzer maps of 3C 437, a radio galaxy at redshift z = 1.48 using IRAC and MIPS on the Spitzer Space Telescope. The exposure times on the sources are the same as for the Quasar 3C 270.1 (120s for each IRAC band. and 100s for the MIPS bands). The image distribution of sources on the sky is similar to the image distribution of 3C 270.1. We got IRAC (3.6m,4.5m,5.8m,8.0m) images of 4′ × 4′ on the Radio Galaxy and two other control fields south west (IRAC 3.6m,5.8m) and north east (IRAC 4.5m,8.0m). The 3σ detection limit of the maps is 4Jy for the first two IRAC bands (3.6m,4.5m),10 Jy for the last two IRAC bands (5.8m,8.0m) and 100Jy for the 24m MIPS band. We also obtained HAWKI H band images at the VLT4 and MEGACAM z’ band images at the MMT. The sensitivity for the VLT HAWKI image is better than 5 Jy with an exposure time of 60 minutes. This is sufficient, because at redshift z ∼ 1.5 the ERO SEDs peak in the H-band. The sensitivity for the MMT MEGACAM z’ band image is better than 1 Jy with an exposure time of 60 minutes. The reduction of the images Spitzer Images was achieved identical to that of 3C 270.1, while the HAWKI reduction was done with the ESO pipeline.

3.3.2 Results

We found no evidence for clustering aroung the Radio Galaxy 3C 437. This is demonstrated in the color-magnitude diagram in Fig. 3.9. Also the inclusion of the H-band data does not provide any hint in favour of a cluster (or proto-cluster)

3Passive evolution was estimated via a Starburst99 model (V´azquez & Leitherer 2005) with an instantaneous burst age 1 Gyr before z = 1.53 and thus an age of 10 Gyr at z = 0. 4Vera Hoﬀmeister & Rolf Chini 52 3.3. NO EVIDENCE FOR CLUSTERING AROUND 3C 437

3C 437 center field

102

101 F 3.6 / z' 100

10−1 100 101 102 103 104 F 3.6 [ µJy ]

3C 437 left comparison field

102

101 F 3.6 / z' 100

10−1 100 101 102 103 104 F 3.6 [ µJy ]

Figure 3.9: Color-magnitude diagram F (3.6)/F (z′) versus F (3.6) of the quasar field and control field. top: quasar field with all sources detected at 3.6 m bottom: control field with all sources detected at 3.6 m. The dotted lines mark the range of EROs having a redshift similar to that of 3C 437. Note that there is no excess of such objects in the cluster field compared to the left field CHAPTER 3. THE CLUSTER SEARCH 53 around this radio galaxy. If 3C 437 were actually surrounded by a cluster (as 3C 270.1) the one explanation could be the limited sensitivity leading to the fact that we only see the tip of the ice berg of the galaxies in the environment of the radio source. Maybe the cluster is in the state of growing and the existing satelite galaxies are small and not luminous enough to detect them. This is consistent with the predictions that galaxy clusters should form between redshift z = 1.5 and z = 2.0. The case of 3C 437 reveals that some of the huge and massive radio sources, which would lie in the cluster potential, show no sign of a large number of massive clustered galaxies in their environment.

3.3.3 Preliminary Conclusion

Given that we have suitable data for two massive radio sources 3C 270.1 and 3C 437, both at redshift z ∼ 1.5, a formal statistical extrapolation would yield, that at this redshift range only 50 % of massive radio sources are surrounded by a cluster or protocluster of EROs. This is roughly consistent with the high redshift decline of clustering predicted by cosmological models. Obviously, the statistics is too sparse and larger samples should be investigated, including spectroscopic veriﬁcation of the redshift of the cluster candidates. Nevertheless, the pilot studies demonstrate the sucsess of the photometric method to identify cluster candidates around high redshift radio sources, the cluster signposts. 54 3.3. NO EVIDENCE FOR CLUSTERING AROUND 3C 437 Chapter 4

Parallel 3D radiative transport

This chapter describes the model part of my thesis.

• Containing a new developed parallel 3D Monte Carlo method with a speed up factor of ∼ 100.

• Careful benchmarking of the new method with existing codes for 1D and 2D conﬁgurations

• Expansion of the Benchmarks to the third dimension, which is now possible with this speed up

• Explaining the average 2 − 16m SEDs and spectra of high redshift AGN with a model consisting of a central power source and three dust components.

4.1 Theory

4.1.1 Monte Carlo method

One approach to solve the radiative transfer equation in an inhomogeneous dusty medium is the Monte Carlo (MC) method. ∞ In our MC procedure the bolometric luminosity L = 0 Lνdν of the heating source is divided into N monochromatic photon packages of equal energy ǫ =

L/N. The spectral energy distribution of the source is divided into 1 ≤ j⋆ ≤ m⋆ bins. The width of the emitted photon package (in the following called photons)

55 56 4.1. THEORY

Figure 4.1: Illustration of our model grid and a typical photon path in the x,z- plane. A photon package is emitted at the heating source (i), interacts with the dust, and can be either absorbed (ii) or scattered (iii). It is also possible that more than one event occurs in a grid cell (iv). The line style illustrates the frequency change during absorption events. Finally the photon leaves the model space (v). CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 57 is set by its energy and the frequency can be calculated from

1 νj⋆ (j⋆ − 2 ) L = Lνdν . (4.1) m⋆ o

Photon packets with high frequency photons have a smaller number of photons than packets with low frequency photons. It is useful to set up a second grid of frequencies to account for the dust re-emission at long wavelengths and low temperatures. In the radiative transfer problem it is necessary to treat the interaction between photons and dust particles. This is described by the cross-sections abs sca for absorption κν and scattering κν . The model space is set up as a Cartesian grid in which each cube can be divided into sub-cubes of volume Vi and constant density ρi (Fig. 4.1), where i is the index of the sub-cube. This separation of cubes allows a ﬁner sampling wherever required: For example close to the dust evaporation zone or at places where the optical depth of a cube is large. The trajectory of the photons is illustrated in Fig. 4.1. Photons are ini- tially emitted by the source at frequency νj⋆ and then traced through the model (Fig.4.1(i)). The direction of a photon is randomly chosen so that 0 ≤ φ < 2π and −1 ≤ cos θ ≤ 1. The distance from the entry point of a cell to the exit point in travel direction of the photon is li. In the MC method the interaction of photons with the dust in a sub-cube is probabilistic and can be determined with a uniformly distributed random number 0 ≤ z ≤ 1 (Witt (1977), Lucy (1999)) and the optical depth abs sca τi(ν)=(κν + κν ) ρi li . (4.2)

Photons leave the cell if τi ≤ − log z and otherwise they interact. If the photon leave the cell it enters along its travel direction a neighboring cell or reaches the border of the model space. We use a new random number in the new cell to determine the interaction probability of photons with the dust 1. The photon ′ interacts with the dust after it has traveled a distance li, given by

′ − log z li = abs sca , (4.3) ρ(κν + κν )

′ The travel distance li deﬁnes the point in the cell where interaction takes place and photons are either scattered (Fig. 4.1(ii)) or absorbed (Fig. 4.1(iii)). As indicated in Fig. 4.1(iv), it is possible that multiple scattering or absorption

1We conﬁrmed that the interaction probability calculated with a new random number is in agreement with the treatment of a travel distance determined by one random number (see also Lucy (1999)). 58 4.1. THEORY

events occur in one cell. The probability of scattering is given by the albedo sca sca abs Aν = κν / (κν + κν ) and the chance of an absorption event is 1 − Aν. When a photon scatters it keeps its frequency, but changes its travel route as given by

the phase function. We use the asymmetry factor gν to approximate anisotropic ′sca sca scattering (Eq. 2.7 in Kr¨ugel 2008): κν = (1 − gν)κν . For isotropic scattering g equals 0. If the photon is absorbed a new photon is immediately emitted, with a new direction and usually a diﬀerent frequency. The emitted photon package has the same energy as the absorbed one. The absorbed photons heat the dust in the cell and the temperature can be calculated by the number of absorbed

photons. After k absorptions the grains reach the temperature Tk,

abs kǫ κν Bν(Tk) dν = , (4.4) 4πρiVi

where Bν(T ) is the Planck function. The dust emits a photon given by the minimum frequency ν′ computed from

ν′ ∞ abs dBν(Tk) abs dBν(Tk) κν dν ≥ z κν dν . (4.5) 0 dTk 0 dTk

In the Monte Carlo method one follows all N photon packages through the dust cloud until they reach the outer boundary of the model. The dust temperature of a cell is calculated iteratively. Each absorption event increases the temperature of the dust which implies a change of the radiation ﬁeld unless local thermal equilibrium is reached. As in the method by Lucy (1999) convergence is usually reached after 3-5 iterations. The number of photons, which are needed to calculate the temperature, can be decreased by an optimization algorithm developed by Lucy (1999). In his treatment every photon package which crosses a cell accounts to a temperature increase. However, as the interaction takes place in every cell this procedure is computational expensive. An iteration free MC method is developed by Bjorkman & Wood (2001). Other MC methods and optimizations exists citep(min09,bae08,gor01).

4.1.2 Parallelization

Microprocessors based on a single central processing unit (CPU), like the IntelR TM PentiumR or the AMDR Opteron , increased their performance during more than two decades. The actual speed of these microprocessors are giga ﬂoating- CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 59 point operations per second (GFLOPS), which means they are able to calculate one billion mathematical operations per second. The increase of the number of ﬂoating point operations per CPU slowed down during the last years (Fig. 4.2). This is mainly caused by the increased energy consumption and heat dissipation for higher clock speeds. However, to further increase the FLOPS of the CPUs nearly all vendors developed models where multiprocessing units are used in each chip. This switch has a tremendous impact on the code developer (Sutter & Larus 2005).

Most of the existing codes are written as sequential programs (von Neumann 1945). These sequential codes do not scale with multi processing units, which yields to a small speed increase with the new microprocessors. Without the speed up it is not possible to implement new features into these codes.

The solution to this challenge are parallel codes which use the power of the multi processing units. A sequential code segment which is executed on diﬀerent cores of the multiprocessor is called thread. In order to get good performance these threads should act indepently. Every interaction of threads usually wastes time, for instance, if one thread ﬁnished the calculation but has to wait for the solution of the other thread to continue.

Figure 4.2: Floating-Point operations per second for the central processing unit (CPU-blue) and NVIDIAs graphics processing unit (GPU-green). 60 4.1. THEORY

Figure 4.3: Show the CUDA scheme take from CUDA manual. The executes serial code and organize the exectution on the graphics cards.

4.1.3 Pseudo random number generator in parallel

In this section we discuss the pseudo random number generator for parallel ap- plications. One of the major properties of Monte Carlo methods is the huge dependence on random numbers. Therefore it is necessary to have a random number generator, which can produce random numbers as fast as possible. Another important fact is the needed unique sequence of numbers for every thread, in parallel running systems. This unique sequence can be created serially. But this is slow for Monte Carlo simulations because of the large amount of these random numbers needed. In order to get the best performance increase from the parallelization we use a special random number generator. It is called parallel Mersenne Twister (Matsumoto & Nishimura (2000)) This random number generator is able to produce unique sequences for each parallel thread. We also performed several tests in order to check the randomness of this random number generator. All of our tests were passed successfully, and the performance scaling is nearly the number of parallel threads. This random number generator fulﬁlls all the conditions needed for fast parallel Monte Carlo simulations.

TM 4.1.4 CUDA parallelization on graphic cards

This section contains a short overview of the data parallelism and program structure of CUDA. Fig. 4.3 and 4.4 illustrates the concept of a cuda project. The CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 61

Figure 4.4: Show the separation of grids into blocks. Ever thread in block can access the same shared memory while threads in diﬀerent blocks need to be independent of each other

computing system consists of a host, the traditional CPU, and one or more de- vices for massive parallel computing (GPUs). The CUDA program consists of phases which are executed either on the CPU or on the GPU. The code parts without parallel structure are executed on the host and the other on the device. The first phase of a project initializes all the needed data on the host and than copy the data which is needed for parallelization into the device (Fig. 4.3). The second phase is the execution of the threads, called kernels, on the device. The huge parallel power of the graphic cards is utilised by an auto scaled parallel running of threads. The scheme of the parallel running threads is shown in Fig. 4.4. The threads are grouped as blocks which have access to the same fast shared memory. The blocks are grouped into a grid. Threads from different blocks can only interact through the slow global memory of the device. After finishing phase two the device copies the data during phase three back to the host. Depending on the project the host may analyse this data and either repeats phase one or finishes the project. Further details are discribed in NVIDIA (2009).

4.2 Imaging

The Monte Carlo simulation was speeded up for axisymmetric problems. In such a symmetry the calculation of one octant is suﬃcient. For the visualization process the complete cube is necessary. Reﬂecting and moving the octant coordinates 62 4.2. IMAGING

Figure 4.5: This ﬁgure shows the geometrical situation for parallel projection. The MC cube has the coordinate system ex,ey,ez and the detector ex′ plane has the coordinate system e e ′ ′ ′ z z ex′ ,ey′ and the pixels x , y . The (0, 0) ′ ′ ′ center of pixel x 0 = (x0, y0) is at position x0. The vector ez′ ′ ey deﬁnes the observing direction.

(0, 0, 0) ey

(x∗, y∗, z∗) leads to the new cube coordinates (x, y, z):

∗ ( i mod2) ∗ 1 ( i mod2) x = n +(−1) 20 x + [1+(−1) 20 ] x 2 ∗ ( i mod2) ∗ 1 ( i mod2) y = n +(−1) 21 y + [1+(−1) 21 ] y 2 ∗ ( i mod2) ∗ 1 ( i mod2) z = n +(−1) 22 z + [1+(−1) 22 ] z 2

∗ ∗ ∗ Where i = [0..7] is a running number over all octants, nx, ny, nz are the number

of grids in the octant and nx, ny, nz are the number of grids in the cube. At astronomical distances the MC cube can be displayed using parallel projection. We set the image at position (D, θ, φ). Where D is the distance from the center of the cube to the image plane and (θ, φ) are angles in spherical coordi-

nates (Fig. 4.6c). Vector ez is the normal surface of the image plane and ex′ ,ey′ are the normal unit vectors for the x’,y’ axis.

ez = (sin θ cos φ, sin θ sin φ, cos θ) (4.6)

ex = (cos θ cos φ, cos θ sin φ, − sin θ) (4.7)

ey =(− sin φ, cos φ, 0) . (4.8)

The required size of the image is derived by projection of the vertices of the ′ ′ cube (Fig. 4.6). It consists of nx × ny quadratic pixels and was at position

x0 = [x0, y0, z0], behind the cube in observing direction. The position x0 has CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 63

Figure 4.6: The left im- O ez ez′ age shows the line of sight l and the position of the ob- θ server O and the intersection of l with the z-plane of ex φ ey the cube. The upper right S Image shows the deﬁnition of the spherical coordinates. The lower right Image shows l the projection of the cube.

′ ′ ′ ′ the pixel coordinates [x0, y0] (Fig. 4.5). A vector xx ,y in cube coordinates is represented in pixel coordinates (x′, y′) of the image by:

′ ′ ′ ′ ′ ′ xx ,y = x0 +(x − x0)ex +(y − y0)ey. (4.9)

The ray tracer follows the line of sight l from the observer O in direction ez′ of each pixel through the 3D grid. The vector form of the linear equation for l is:

l = O + aez, (4.10) where a is a scaling factor. For the xy-plane (z = c = const.) we calculate the intersection point between l and the z-plane, with the scaling factor

c − (0, 0, 1)O axy = . (4.11) (0, 0, 1)ez

Analogously we calculate the intersection points with the x,y-plane. The point S, where the line of sight hits the border of the cube, is derived with Eq. 4.10. The distance between the entry point S into the cube and the center of the cell is l = (O − S)2 (4.12) as shown in Fig. 4.6. Taking the emission and absorption of every cell along this line, we compute the ﬂux received by the observer. For simplicity we neglect foreground absorption. The emission is calculated according to Kirchhoﬀ’s law. A dust grain in a 64 4.2. IMAGING

radiation ﬁeld absorbs as much energy as it emits. For local thermal equilibrium

(LTE) the emissivity ǫν per unit mass is

abs ǫν = Kν Bν(T ), (4.13)

abs with temperature T , the absorption cross section Kν per unit mass and the

Planck function Bν(T ). The intensity of a cell with dust mass mdust is

ǫνmdust Iν(0) = , (4.14) Ap

where Ap is the projected source area on the detector plane.

The light, which reaches the observer, is weakened by dust, located between the cell and the border of the cube. For pure extinction

−τν Iν = Iν(0)e (4.15)

where τν is the optical depth at frequency ν. It is calculated with the distance l (eqn. 4.12) and the dust density ρ(x) distribution

l abs τν = Kν ρ (x) ds. (4.16) 0

The ﬂux density is

Fν = IνΩ (4.17)

with solid angle Ω, which is for large distances D:

A Ω= source (4.18) D2

where the source area Asource = Ap for parallel projection. Therefore the flux density is Kabsm B (T )e−τν F = ν dust ν . (4.19) ν D2 for each cell. Adding the flux densities from the cells along the line of sight l we compute the flux density for each detector pixel.

Fx′,y′ = Fν (4.20) x,y,z∈l CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 65

Figure 4.7: 10 micron image of our solar system up to Jupiter at a distance of 1 parsec. This face on view is generated with ray-tracer and 64k × 64k pixels.

4.2.1 Solar system

The solar system is an excellent environment to test the correct behavior of our raytracing technic. We set the solar system at a distance of 1 parsec and simulate the observed flux at 10 micron. The flux can be determined analytically with a black body radiation at the different temperatures given with Eq. 1.7. The simulated fluxes in the image 4.7 are correct within 1% accuracy. The difference is caused by the number of observing pixels in comparison to the model pixel. More observing pixels correspond to a better accuracy. It is useful to give a lower limit of 10 × 10 observing pixels per model pixel to guarantee the 1% level correctness.

4.3 Benchmark test

To test our code we use several benchmarks tests, which are available in the literature. All test cases assume local thermal equilibrium with the radiation ﬁeld. For the simplest 1 dimensional case we use the test from Ivezic et al. (1997). For the 2 dimensional case we use the disk geometry introduced by Pascucci et al. (2004). And for the 3 dimensional case we extend the Pascucci et al. (2004) disk by spiral structure, which may be caused ba a proto-planet. We show that we can reproduce the temperature distributions and the SEDs for all existing benchmark cases in much less time.

4.3.1 Spherical symmetry (1D)

This benchmark is completely described in Ivezic et al. (1997). They assume a central point source which is embedded in a spherical symmetric dust envelope. The innermost volume is free of dust. The source radiates as a black body with a given temperature T⋆. The one dimensional dust density distribution is assumed 66 4.3. BENCHMARK TEST

0 10 100

−1 10 10−1

tot −2 10 10−2 tot / F λ F λ / F λ −3

10−3 F 10 λ

−4 10 10−4 −5

10 10−5 1 10 100 5 0 -5 diff. [%] 1 10 100 wavelength [µm]

Figure 4.8: upper panel: SEDs from DUSTY as reference code for 4 diﬀerent optical depth of 1(magenta), 10(blue), 100(orange), 1000(green). The colored lines shows the SED calculated with our code, while the black line represents the reference code. Note for the higher optical depth 100 and 1000 we use a larger number of iterations. lower panel: The diﬀerence of our code to the solution of Ivezic et al. (1997) is less than 5%.

to decrease with a power law r−p. They combine the parameters, dust density, envelope size, opacity in one parameter, the optical depth, which is ﬁxed at 1 m. To minimize the number of analytic functions they choose.

qabs = qsca = 1 (4.21)

for λ ≤ 1m, and 1 1 q = ,q = (4.22) abs λ sca λ4 for λ > 1m. They argue that the value of 1 m is chosen, because this is the typical astrophysical grain size. These quantities fully describe the benchmark problem. We compare our code with four diﬀerent values of optical depth, at

temperature T⋆ = 2500K, and a constant density distribution p = 0. The outer

radius is ﬁxed to 1000×rinner, with rinner given by the radiation ﬁeld (Table 4.1). It is calculated with equation 4 in Ivezic et al. (1997) and references therein. We performed Monte Carlo simulations for this one dimensional test case with 2 × 108 photon packages per iteration, on a Cartesian grid with 106 cells. The CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 67

Table 4.1: Values for rinner for four diﬀerent optical depth, router = 1000 rinner, L⋆ = L⊙ and T⋆ = 2500K. The values are calculated with Equ. 4 in Ivezic et al. (1997) τ 1 10 100 1000 12 rinner [ ×10 cm] 3.14 3.15 3.20 3.52 number of iterations depends on the optical depth for, τ = (1, 10, 100, 1000) the number of iterations is (3, 3, 4, 5). The time needed for the highest optical depth conﬁguration is 78 min. The temperature distributions are compared with the results of the dusty code for the same optical depth. We reproduced the temperature distribution of the dusty code, within an error smaller than one percent. With this temperature distribution we generated the SEDs shown in

Fig. 4.8. We plot on the y-axis λFλ divided by the total flux Ftot = λ Fλdλ and the wavelength in micron on the x-axis. Each curve represents a different optical depth. The SEDs generated with our code and the one from the reference code DUSTY are in good agreement. The error is below 5 percent for optical depth τ < 100. It is largest on the boundaries of the simulated wavelength range, where the flux Fλ is low compared to the total flux. This is caused by the statistical behavior of the Monte Carlo method. It increases for the highest optical depth case of τ = 1000, but is still below 5 percent. The larger deviation is caused by the limited number of grid cells.

4.3.2 Disk geometry (2D)

This benchmark is completely described in Pascucci et al. (2004). They consider the general astrophysical case of a point source surrounded by a circumstellar dust disk. The disk is composed of spherical dust grains, with absorption and scattering properties of 0.12 m silicate grains. (see Fig. 4.10). The absorption and scattering eﬃciencies are taken from their web-page.2 With an inner region r < rinner free of dust, they adopt the following density distribution

r − π × z −1.0 4 h(r)2 ρ(r, z)= ρ0 × ( ) × e (4.23) rd with r 1.125 h(r)= zd × ( ) . (4.24) rd

2http://www.mpia.de/PSF/PSFpages/RT/benchmark.html 68 4.3. BENCHMARK TEST

] -2

[Wm λ F λ

-10 difference [%]

-20

wavelength [µm] wavelength [µm]

Figure 4.9: left column: face on disc at an inclination angle of 12.5◦ right column: edge on disc at an inclination angle of 75◦ upper panel: SEDs from (Pascucci et al. 2004) as reference code for 4 diﬀerent mid-plane optical depth of τ = 1(purple), 10(green), 100(blue). The symbols shows the SED calculated with our code. Note for the higher optical depth of 100 we use a larger number of iterations. lower panel: Diﬀerence of our code with to the solution of Pascucci et al. (2004).

These equations describe a disk which is similar to that of Chiang & Goldreich (1997, 1999). The complete list of model parameters is shown in Table 4.2. We performed Monte Carlo simulations with 2 × 108 photon packages per iteration, 3×106 grid cells and 3 iterations for the optical depth of τ = 1, 10 and 4 iterations for τ = 100. The calculated SEDs and relative differences to the reference code are shown in Fig. 4.9 for two different inclinations(face on and edge on) and four different optical depth values. The results of our three dimensional Monte Carlo codes agree with the benchmark to better than 10 percent. The largest deviation arises at the 9.8 micron feature for the highest optical depth case. This can be explained by the fact that for this case the grid arrangement plays an increasing role. Our code uses a Cartesian grid which is not as ideal as spherical grids for axis symmetric problems. We also want to note that our code reproduces the slope CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 69

Table 4.2: Model parameter space as in Pascucci et al. (2004) Symbol Meaning Value M⋆ Stellar Mass 1 M⊙ R⋆ Stellar Radius 1 R⊙ T⋆ Stellar eﬀective temperature 5800K Rout Outer disk radius 1000 AU Rin Inner disk radius 1 AU zd Disk height 125 AU a Grain radius 0.12 m −3 ρg Grain density 3.6 gcm τv Optical depth at 550 nm 0.1, 1, 10, 100 at long wavelength correctly. This slope only depends on the dust properties.

4.3.3 Dust properties

We use an analytic function for the dust properties, which can be described as follows for λ ≤ 0.01m, 1 q = q = (4.25) abs sca λ3 for 0.01m ≤ λ ≤ 0.5m,

qabs = qsca = 1 (4.26) for 0.5m ≤ λ ≤ 10m, and

1 1 q = ,q = (4.27) abs λ sca λ4 for λ ≥ 10m 1 q = 10−23ac2 ,q = 0 (4.28) abs (10−4λ)2 sca and at 10 micron we add a Gaussian with a maximum value of 0.1 and a deviation σ = 1, where a is the dust grain radius, λ the wavelength in micron. The dust extinction coeﬃcient is plotted in Fig. 4.10 against the dust properties of Pascucci et al. (2004). We expand the extinction up to 0.01m.

4.3.4 Spiral expansion of disk structure (3D)

From theoretical considerations one may suggest that in the process of planet formation a circumstellar dust disk will show spiral structures in the dust density distribution. Hydrodynamical simulations of proto-planetary disks show spiral structures, produced by the orbiting planet (FARGO). Therefore we extend the 70 4.3. BENCHMARK TEST

Figure 4.10: Extinction efficiency for used in the 2D and 3D configurations. The solid line shows the extinction coefficient for astronomical silicate grains as taken from Pascucci et al. (2004). The dashed line is the extinction coefficient described by equations 4.25-4.28

2 dimensional case by adding a spiral structure to the dust density distribution. The dust density is described with Eqn. 4.23 and decreased by a factor of 10−5, in the spiral: R r(φ)= out × φ (4.29) 2πnloop

Where nloop is the number of loop in the spiral and Rout is size of the disk. This benchmark configuration with the simplified spiral dust geometry shows weak differences in the SEDs compared to the two dimensional disk. (Fig 4.13) Which is the different dust emission peak, caused by the missing of warm dust in the spiral structure. For small optical depth up to 100 this shifts the peak to right for a spiral configuration.

4.3.5 Clumpy Torus geometry (3D)

Another astronomical source which is suitable to be calculated in three dimensions is the environment of an AGN. In the uniﬁed scheme (Antonucci 1993) these objects are likely surrounded by a dust torus, which might be composed of dust clumps (Nenkova et al. 2002). We use this as a second benchmark conﬁguration CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 71

Figure 4.11: Density structure for the 3D spiral test case. The values are normalized to the maximum density. The density decrease from cyan(minimum) to black (maximum)

Figure 4.12: Left: modeled 11 m ﬂux at a distance of 50 pc. Right: Prediction for the planed European Extremly Large Teleskop (EELT) 72 4.3. BENCHMARK TEST

] -2

[Wm λ F λ

-10 difference [%]

-20

wavelength [µm] wavelength [µm]

Figure 4.13: left column: face on disc at an inclination angle of 12.5◦ right column: edge on disc at an inclination angle of 75◦ upper panel: SEDs from (Pascucci et al. 2004) as reference code for 4 diﬀerent mid-plane optical depth of τ = 1(purple), 10(green), 100(blue). The symbols shows the SED calculated with our code. Note for the higher optical depth of 100 we use a larger number of iterations. lower panel: only a small diﬀerence at long wavelengt of the 3D spiral expansion to the solution of Pascucci et al. (2004). CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 73

Figure 4.14: Density distribution for the 3D torus structure. On the left is the face on view. While the right shows the edge on view. Values are normalize to maximum density and the density decrease from red (maximum) to black (minimum). Even in the edge on view it is possible to see the hot inner dust.

for the three dimensional radiative dust transfer. We assume clumps of the same size with a constant density ρclump. These clumps are distributed randomly within a half opening angle of 35 degrees around the mid-plane. We use the random number generator ’suggested’ by numerical recipes, which is very fast and can be easily reproduced.

In+1 = 1664525 ∗ In + 1013904223 (4.30)

In+1 ,with I0 = 0. The ﬂoating point random number fn+1 is calculated by f = 232 . To check if the random number generator is working correctly, we list the ﬁrst

4 values I1 − I4 in Table 4.3. The value ρ0 is given in Table 4.4, and r is the

Table 4.3: 4 values I1 − I4 of the random number generator for I0 = 0 i1 i2 i3 i4 HEX 3C6EF35F 47502932 D1CCF6E9 AAF95334 DEC 1013904223 1196435762 3519870697 2868466484 ﬂoat 0.236068 0.278567 0.819534 0.667867

distance of the clump center to the AGN. We distribute Nclump = 1 000, 2 500 and 5 000 of these clumps in an area within a half opening angle of 35◦ degree between rinner and router. The density map is shown in Fig 4.14 This dust distribution can 74 4.3. BENCHMARK TEST

1.000

0.100 tot / F λ F λ 0.010

5000 clumps 2500 clumps 1000 clumps 0.001 1 10 wavelength [µm]

Figure 4.15: Phi averaged SEDs for two diﬀerent inclination angles (face on (upper) to edge on (lower)). The simulation contains 108 photons, with 106 grid cells. The dust is distributed in 1 000, 2 500 and 5 000 clumps with an half opening angle of 35◦ around the mid-plane, with an inner radius of 0.2 pc and an outer radius of 5 pc. The optical depth at 550nm at to randomly chosen lines is τ550nm = 115, 230 and 385. Note that the Si feature is in weak emission for the edge on case with the lowest number of clumps. CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 75

Table 4.4: Parameter set of the three benchmark models Symbol ModelI ModelII ModelIII Description Nclump 1 000 2 500 5 000 Number of clumps 11 LAGN 10 L⊙ Luminosity of the AGN 9 DAGN 10 pc Distance of the AGN τV 115 230 383 Visual optical depth 3 3 3 MDust 1 × 10 M⊙ 2 × 10 M⊙ 4 × 10 M⊙ Dust mass rin 0.2pc Inner radius rout 1.5pc Outerradius be reproduced by the given shift random number generator where

I r = r + 0+i (r − r ) (4.31) i inner 232 outer inner I φ = 2 π 0+i (4.32) 232 pi I θ = + (2 0+i − 1) θ . (4.33) 2 232 open

We performed Monte Carlo simulations for 3 different numbers of clouds. Now we briefly discuss the outcome of this benchmark configuration. What we found is that for a clump distribution, which allows to see the central power engine, it is possible to see hot silicate without any disturbing cloud in the line of sight (blue solid line in Fig. 4.15 ). If we increase the number of clumps up to 2 500 (5 000), we shield the central part of the structure (solid magentat (red) line in Fig. 4.15). This means hot dust in front of cold clumps, which yields to silicate absorption. The cloud distribution with most of the silicate emission near the central region, is also visible in the 10 m re-emission image (Fig. 4.16)

4.4 Modeling average spectra & SEDs

This section of the thesis models the average SEDs and spectra of the powerful high redshift (z > 1.0) 3C sources. The used spectral energy distributions are already discussed in chapter 2 as well as the average SED. The average IRS spectra from the 20 most luminous objects are taken from Leipski et al. (2010). The addition of the 19 − 38m spectra, which are redshifted from the restframe 9−16m in the source system, enables us to study the 10m silicate feature (Fig. 4.19). The assumed dust size distribution is equal to that of Mathis et al. (1977) for a composition of 2 grain types, silicate and amorphous carbon. The extinction −5 curve (Fig. 4.18) is calculated with a silicate abundance of NSi/NH = 3.1 × 10 76 4.4. MODELING AVERAGE SPECTRA & SEDS

Figure 4.16: Image of the dust re-emission at 10 micron at an inclination of 90 degrees. It is possible to see the shielding of individual clouds as well as the three dimension structure of illuminated spheres. In direction to the source the cloud surface is hot. In contradiction is the surface opposite to the heating source cold.

−4 and a carbon abundance of NC /NH = 2.0×10 . MIDI observations of the nearby Seyfert 2 galaxy NGC 1068 (Jaffe et al. 2004) show hints of a two component structured dust torus. It consists of one hot component with temperatures of T ∼ 800K in the central region of the black hole, with distances of about 0.7pc and one warm component with temperatures of T ∼ 320K and a size of 3.4×2.1pc (Jaffe et al. 2004). This leads to the dust configuration in our model. Other MIDI observations (Tristram et al. 2007) suggest a clumpy dust torus of the Circinius

galaxy. Therefore we distributed Nclump spherical dust clumps, with a size of 0.5pc and constant dust density, in the torus. These clumps fill a disk like structure with a half-opening angle of 35◦. The clump distribution is shown in Fig. 4.17. In order to reproduce the hot and cold dust components we inserted a gap with no clumps between 1.5pc and 3.5pc. We performed the radiative transfer Monte Carlo simulation for this three dimensional dust configuration with 109 photon packages and 106 grid cells. The best fitting inclination angle i = 15◦ (i = 75◦) for the face on (edge on) view is shown in Fig. 4.19. The silicate feature in Fig. 4.19 is still to strong. It can be reduced by foreground dust extinction in the host galaxy. We therefore used our dust model (Fig. 4.18) to “redden” Fig. 4.19. The result is shown in Fig. 4.20. The emerged model fits the average SED and CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 77

0 size [pc] 25 0 size [pc] 25 0 size [pc] 25 0 size [pc] 25

Figure 4.17: Face on (left) and edge on (right) view of the clumpy density distribution for the best ﬁtted torus model. Size is 25 × 25 parsec. Inner gap of 1 parsec. Half opening angle of 35 degrees

] 10 dust /g 2

104

103

2 Extinction coefficient [cm 10 10−3 10−2 10−1 100 101 102 wavelength [µm]

Figure 4.18: Extinction curve for dust composition of 2 grain species silicate and carbon. The size distribution is equal to that of Mathis et al. (1977) with the −5 silicate abundance of NSi/NH = 3.1 × 10 and carbon abundance of NC /NH = 2.0 × 10−4 78 4.4. MODELING AVERAGE SPECTRA & SEDS

Table 4.5: Parameter for the ﬁtting torus model Symbol Values Description Nclump 5 000 Number of clumps 11 LAGN 10 L⊙ Luminosity of the AGN 9 DAGN 10 pc Distance of the AGN sx × sy × sz 25pc×25pc×12.5pc Size of the torus τV ∼ 100 Visual optical depth 3 MDust 10 M⊙ Dust mass

spectra of the high redshift radio sources very well. It is important to note that a single model is able to explain average SED and spectra for both types of radio sources, which is in agreement with the uniﬁed scheme. CHAPTER 4. PARALLEL 3D RADIATIVE TRANSPORT 79

20 erg/s) 44 (10

ν 10 L ν

RG QSO 0 0 5 10 15 rest wavelength (µm)

Figure 4.19: Model of the average spectra and SED. The colored shapes represents our data from 2 − 8m the average SED and from 9 − 16m the average spectra (Leipski et al. 2010). The color represents the type of radio source red for radio galaxies and blue for quasars. The black solid line represents our model prediction. The error bars are calculated from the diﬀerent viewing angles.

20 erg/s) 44 (10

ν 10 L ν

RG QSO 0 0 5 10 15 rest wavelength (µm)

Figure 4.20: Model of average spectra and SED. The colored shapes represents our data from 2 − 8m the average SED and from 9 − 16m the average spectra (Leipski et al. 2010). The color represents the type of radio source red for radio galaxies and blue for quasars. The black solid line represents our model prediction. The error bars are calculated from the diﬀerent viewing angles. 80 4.4. MODELING AVERAGE SPECTRA & SEDS Chapter 5

Summary and Outlook

This thesis studied for the first time the near- and mid-infrared emission of the most powerful high-redshift 3CR radio sources. The aim was to quantify any orientation-dependent differences between type 1 and type 2 AGNs by means of new observations and models, and in addition to determine the cluster properties at the so far almost unexplored high redshift regime. The observational part was supervised at the University of Bochum and the model part at ESO Garching. In collaboration with the Spitzer IRAC team at the CfA, 3 − 24m spectral energy distributions (SED) of 64 3CR radio galaxies and quasars at z > 1 have been obtained. In addition, 19 − 38m spectra of the 20 brightest objects at 1 1 theoretical models predict a drastic decline of the cluster space density. However, finding clusters at these high redshifts has turned out to be difficult. Therefore we used massive radio sources as signposts of galaxy clusters in the early universe. The Spitzer IRAC maps are deep and large enough to look for clusters around the 3CR sources. In our pilot study, combining the IRAC data with deep MMT z’ band data, we found an overdensity of extremly red objects around the quasar 3C 270.1 at z = 1.53 (Haas, Willner, Heymann et al. 2009). On the other hand using similar data, the radio galaxy 3C 437 at redshift z = 1.48 does not show any sign of clustering. While these two cases indicate a clustering decline between redshift z = 1 and z = 1.5, the number statistics is

81 82

obviously too poor. Nevertheless these pilot studies demonstrate that the method might be successful, when applied to larger samples for providing observational constraints for the theoretical models. Further z’ band observations of our sample are currently carried out with the 6.5m Magellan Telescope, Las Campanas, and at the 10m Gran Telescopio Canarias (GTC), La Palma. After a careful benchmarking in existing one dimensional and two dimensional codes, I have implemented an efficient 3D Monte Carlo radiative transfer code with the aim to model the obtained SEDs and spectra. While in general the required computer power is not available, I found a solution using the parallel capabilities of graphic cards. For an arbitrarily given dust distribution and primary source(s) the model spectra are derived in two steps: First calculate the dust temperature and then the dust emission images and SEDs. The code passes the one dimensional Ivezic et al. (1997) and two dimensional Pascucci et al. (2004) benchmark tests. The speed up by more than a factor of ten allows us for the first time to set up a three dimensional benchmark test (Heymann et al. in prep.). Application of the model to both the Spitzer Infrared and Chandra X-ray data: Because an essential part of the sample has also been observed with Chan- dra (PI Belinda Wilkes), we have both the information on the power source and the reprocessed dust emission. Thus we are able to properly model any dust geometry in the most powerful AGN, and to derive not only the orientation of the torus with respect to the line of sight but also the filling factor of the clumpy dust torus and the host galaxy contribution (Heymann et al. in prep.).

Outlook: We want to use our Spitzer NIR/MIR data of the most powerful high redshift radio sources to expand the knowledge of their environment. The needed additional optical/NIR are obtained at the Magellan Telescope as well as at the GTC. Our method is able to find clusters at high redshift z > 1.0 with low observational costs in comparison to the deep multiwavelengths surveys. The cluster properties at high redshift provide constraints on cosmological model parameters. With the granted Herschel observations of our group, we are able to further increase our dust torus model to explain the MIR/FIR dust emission. This new data in the MIR/FIR set of the powerful high redshift radio sources will give fascinating new challenges to the model part. The high performance increase with the help of graphic cards could be used to model a moderate number of template SEDs for various type of quasars and radio sources. The speed improvement also increases the requirements on the code in CHAPTER 5. SUMMARY AND OUTLOOK 83 terms of spatial resolution, to resolve the hot inner dust (Spitzer and the cold dust further out (Herschel). Furthermore, the speed improvement by a factor of ∼ 100 gives for the first time the possibility to model the from the dust torus scattered photons to verify the unification scheme. 84 ACKNOWLEDGEMENT 85

Acknowledgment

I would like to thank my supervisors Priv.-Doz. Dr. Martin Haas and Dr. Ralf Siebenmorgen for giving me the possibility to work on these two fascinating topics as well as the friendly advices and discussions. I would like to thank Prof. Dr. Rolf Chini, who made this thesis possible, for fascinating lectures on star formation to come back to the ’smaller’ scales. Furthermore, I thank Priv.-Doz. Dr. Dominik Bomans for helpful discussions on extragalactic topics. In addition I would like to thank all the other people, which helped in various ways to ﬁnish this thesis.

• My oﬃce mates, Lu Feng, Ulf Seemann, Frank Karsten Broogard, during the time at ESO Garching and Ingo Steiner, Michael Ramolla, Mathias D¨orr, during the time at the Bochum University

• The information technology oﬃcer, Tim Falkenbach in Bochum as well as the information technology people at ESO for constant advices and help.

• All the kind people at the Astronomische Institut Bochum and ESO / MPIA Garching for the lovely environment.

• All my friends for making my social life outside the institutes.

I will also thank my wife for the support during the time and reading the manuskript as well as my family for the support during my study and education.

Thank you all very much! 86 ACKNOWLEDGEMENT Bibliography

Antonucci, R. 1993, ARA&A, 31, 473

Antonucci, R. R. J. 1983, Nature, 303, 158

Antonucci, R. R. J. & Miller, J. S. 1984, in Bulletin of the American Astronomical Society, Vol. 16, Bulletin of the American Astronomical Society, 957–+

Barthel, P. D. 1989, ApJ, 336, 606

Barvainis, R. 1987, ApJ, 320, 537

Bertin, E. & Arnouts, S. 1996, A&AS, 117, 393

Best, P. N. 2000, MNRAS, 317, 720

Bjorkman, J. E. & Wood, K. 2001, ApJ, 554, 615

Bremer, M. N., Valtchanov, I., Willis, J., et al. 2006, MNRAS, 371, 1427

Brownstein, J. R. & Moﬀat, J. W. 2006, MNRAS, 367, 527

Chiang, E. I. & Goldreich, P. 1997, ApJ, 490, 368

Chiang, E. I. & Goldreich, P. 1999, ApJ, 519, 279

Chini, R., Kreysa, E., & Biermann, P. L. 1989, A&A, 219, 87

Cleary, K., Lawrence, C. R., Marshall, J. A., Hao, L., & Meier, D. 2007, ApJ, 660, 117

De Propris, R., Stanford, S. A., Eisenhardt, P. R., Holden, B. P., & Rosati, P. 2007, AJ, 133, 2209

Disney, M., Davies, J., & Phillipps, S. 1989, MNRAS, 239, 939

87 88 BIBLIOGRAPHY

Draine, B. T. & Li, A. 2001, ApJ, 551, 807

Elvis, M., Wilkes, B. J., McDowell, J. C., et al. 1994, ApJS, 95, 1

Fanaroﬀ, B. L. & Riley, J. M. 1974, MNRAS, 167, 31P

Fang, T., Gerke, B. F., Davis, D. S., et al. 2007, ApJ, 660, L27

Fazio, G. G., Hora, J. L., Allen, L. E., et al. 2004, ApJS, 154, 10

Guhathakurta, P. & Draine, B. T. 1989, ApJ, 345, 230

Haas, M., M¨uller, S. A. H., Bertoldi, F., et al. 2004, A&A, 424, 531

Haas, M., Siebenmorgen, R., Schulz, B., Kr¨ugel, E., & Chini, R. 2005, A&A, 442, L39

Haas, M., Willner, S. P., Heymann, F., et al. 2008, ApJ, 688, 122

Haas, M., Willner, S. P., Heymann, F., et al. 2009, ApJ, 695, 724

Heckman, T. M., Chambers, K. C., & Postman, M. 1992, ApJ, 391, 39

Heckman, T. M., O’Dea, C. P., Baum, S. A., & Laurikainen, E. 1994, ApJ, 428, 65

Hes, R., Barthel, P. D., & Fosbury, R. A. E. 1993, Nature, 362, 326

Houck, J. R., Roellig, T. L., van Cleve, J., et al. 2004, ApJS, 154, 18

Indebetouw, R., Mathis, J. S., Babler, B. L., et al. 2005, ApJ, 619, 931

Intema, H. T., Venemans, B. P., Kurk, J. D., et al. 2006, A&A, 456, 433

Ivezic, Z., Groenewegen, M. A. T., Men’shchikov, A., & Szczerba, R. 1997, MN- RAS, 291, 121

Jaﬀe, W., Meisenheimer, K., R¨ottgering, H. J. A., et al. 2004, Nature, 429, 47

Khachikian, E. Y. & Weedman, D. W. 1974, ApJ, 192, 581

Kinney, A. L., Calzetti, D., Bohlin, R. C., et al. 1996, ApJ, 467, 38

Kr¨ugel, E. 2008, An introduction to the physics of interstellar dust (Springer)

Kurk, J. D., R¨ottgering, H. J. A., Pentericci, L., et al. 2000, A&A, 358, L1 BIBLIOGRAPHY 89

Leipski, C., Haas, M., Willner, S. P., et al. 2010, submitted

Lucy, L. B. 1999, A&A, 344, 282

Manners, J. C. 2002, PhD thesis, The University of Edinburgh

Mathis, J. S., Rumpl, W., & Nordsieck, K. H. 1977, ApJ, 217, 425

Matsumoto, M. & Nishimura, T. 2000, in Monte Carlo and Quasi-Monte Carlo Methods 1998 (Springer), 59–69

Miley, G. K., Neugebauer, G., & Soifer, B. T. 1985, ApJ, 293, L11

Mullis, C. R., Rosati, P., Lamer, G., et al. 2005, ApJ, 623, L85

Nenkova, M., Ivezi´c, Z.,ˇ & Elitzur, M. 2002, ApJ, 570, L9

NVIDIA. 2009, CUDA programming guide 2.3

Ogle, P., Whysong, D., & Antonucci, R. 2006, ApJ, 647, 161

Orr, M. J. L. & Browne, I. W. A. 1982, MNRAS, 200, 1067

Pascucci, I., Wolf, S., Steinacker, J., et al. 2004, A&A, 417, 793

Pier, E. A. & Krolik, J. H. 1993, ApJ, 418, 673

Pozzetti, L. & Mannucci, F. 2000, MNRAS, 317, L17

Rieke, G. H., Young, E. T., Engelbracht, C. W., et al. 2004, ApJS, 154, 25

Rowan-Robinson, M. 1980, ApJS, 44, 403

Rowan-Robinson, M. 1995, MNRAS, 272, 737

Rowan-Robinson, M., Babbedge, T., Oliver, S., et al. 2008, MNRAS, 386, 697

Schuster, M. T., Marengo, M., & Patten, B. M. 2006, in Presented at the Soci- ety of Photo-Optical Instrumentation Engineers (SPIE) Conference, Vol. 6270, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series

Schweitzer, M., Lutz, D., Sturm, E., et al. 2006, ApJ, 649, 79

Seward, F. D. 1999, Allen’s astrophysical quantities, ed. Cox, A. N. (Springer), 183

Seyfert, C. K. 1943, ApJ, 97, 28 90 BIBLIOGRAPHY

Seymour, N., Stern, D., De Breuck, C., et al. 2007, ApJS, 171, 353

Shi, Y., Rieke, G. H., Hines, D. C., et al. 2005, ApJ, 629, 88

Siebenmorgen, R., Kr¨ugel, E., & Spoon, H. W. W. 2004, A&A, 414, 123

Spinrad, H., Marr, J., Aguilar, L., & Djorgovski, S. 1985, PASP, 97, 932

Stanford, S. A., Romer, A. K., Sabirli, K., et al. 2006, ApJ, 646, L13

Stern, D., Chary, R., Eisenhardt, P. R. M., & Moustakas, L. A. 2006, AJ, 132, 1405

Sutter, H. & Larus, J. 2005, Queue, 3, 54

Tristram, K. R. W., Meisenheimer, K., Jaﬀe, W., et al. 2007, A&A, 474, 837

Urry, C. M. & Padovani, P. 1995, PASP, 107, 803

V´azquez, G. A. & Leitherer, C. 2005, ApJ, 621, 695

Vernet, J., Fosbury, R. A. E., Villar-Mart´ın, M., et al. 2001, A&A, 366, 7

von Neumann, J. 1945, U.S. Army Ordnace Department and University of Penn- sylvania

Ward, M., Elvis, M., Fabbiano, G., et al. 1987, ApJ, 315, 74

Werner, M. W., Roellig, T. L., Low, F. J., et al. 2004, ApJS, 154, 1

Wilson, G., Huang, J., P´erez-Gonz´alez, P. G., et al. 2004, ApJS, 154, 107

Witt, A. N. 1977, ApJS, 35, 1

Zwicky, F. 1937, ApJ, 86, 217 3C SOURCES 91

3C sources

Table 5.1: High redshift z > 1 3C sources a Object Name RA DEC Type z 3C 002 00h06m22.6s -00d04m25s QSO 1.037 3C 009 00h20m25.3s +15d40m53s QSO 2.009 3C 013 00h34m14.5s +39d24m17s G 1.351 3C 014 00h36m06.5s +18d37m59s QSO 1.469 3C 036 01h17m59.5s +45d36m22s G 1.301 3C 043 01h29m59.8s +23d38m20s QSO 1.459 3C 065 02h23m43.2s +40d00m52s G 1.176 3C 068.1 02h32m28.9s +34d23m47s QSO 1.238 3C 068.2 02h34m23.8s +31d34m17s G 1.575 3C 119 04h32m36.5s +41d38m28s G 1.023 3C 124 04h41m59.1s +01d21m02s G 1.083 3C 173 07h02m20.6s +37d57m22s G 1.035 3C 181 07h28m10.3s +14d37m36s QSO 1.382 3C 186 07h44m17.4s +37d53m17s QSO 1.067 3C 190 08h01m33.5s +14d14m42s QSO 1.195 3C 191 08h04m47.9s +10d15m23s QSO 1.956 3C 194 08h10m03.6s +42d28m04s G 1.184 3C 204 08h37m44.9s +65d13m35s QSO 1.112 3C 205 08h39m06.4s +57d54m17s QSO 1.534 3C 208 08h53m08.8s +13d52m55s QSO 1.110 3C 208.1 08h54m39.3s +14d05m53s G 1.020 3C 210 08h58m09.9s +27d50m52s G 1.169 3C 212 08h58m41.5s +14d09m44s QSO 1.048 3C 220.2 09h30m33.5s +36d01m24s QSO 1.157 3C 222 09h36m32.0s +04d22m10s G 1.339 3C 225A 09h42m08.5s +13d51m54s G 1.565 3C 230 09h51m58.8s -00d01m27s G 1.487 3C 238 10h11m00.4s +06d24m40s G 1.405 3C 239 10h11m45.4s +46d28m20s G 1.781 3C 241 10h21m54.5s +21d59m30s G 1.617 3C 245 10h42m44.6s +12d03m31s QSO 1.028 3C 249 11h02m03.8s -01d16m17s QSO 1.554 3C 250 11h08m52.1s +25d00m55s G — 3C 252 11h11m33.1s +35d40m42s G 1.100 3C 255 11h19m25.2s -03d02m52s QSO 1.355 3C 256 11h20m43.0s +23d27m55s G 1.819 3C 257 11h23m09.2s +05d30m19s G 2.474 3C 266 11h45m43.4s +49d46m08s G 1.275 3C 267 11h49m56.5s +12d47m19s G 1.140 92 3C SOURCES

Table 5.2: High redshift z > 1 3C sources b Object Name RA DEC Type z 3C 268.4 12h09m13.6s +43d39m21s QSO 1.397 3C 270.1 12h20m33.9s +33d43m12s QSO 1.532 3C 280.1 13h00m33.3s +40d09m08s QSO 1.671 3C 287 13h30m37.7s +25d09m11s QSO 1.055 3C 294 14h06m44.0s +34d11m25s G 1.779 3C 297 14h17m24.0s -04d00m48s G 1.406 3C 298 14h19m08.2s +06d28m35s QSO 1.437 3C 300.1 14h28m31.3s -01d24m08s G 1.159 3C 305.1 14h47m09.5s +76d56m22s G 1.132 3C 318 15h20m05.4s +20d16m06s G 1.574 3C 322 15h35m01.2s +55d36m53s G 1.681 3C 324 15h49m48.9s +21d25m38s G 1.206 3C 325 15h49m58.6s +62d41m21s G 1.135 3C 326.1 15h56m10.1s +20d04m20s G 1.825 3C 356 17h24m19.0s +50d57m40s G 1.079 3C 368 18h05m06.3s +11d01m33s G 1.131 3C 418 20h38m37.0s +51d19m13s QSO 1.686 3C 432 21h22m46.2s +17d04m38s QSO 1.785 3C 437 21h47m25.1s +15d20m37s G 1.480 3C 454.1 22h50m32.9s +71d29m19s G 1.841 3C 454.0 22h51m34.7s +18d48m40s QSO 1.757 3C 469.1 23h55m23.3s +79d55m20s G 1.336 3C 470 23h58m35.3s +44d04m39s G 1.653 4C 13.66 18h01m39.0s +13d51m23s G 1.45 4C 16.49 17h34m42.6s +16d00m31s QSO 1.296 List of Figures

1.1 Sketch of an AGN continuum spectrum of the nuclear region, without stellar contribution. Three diﬀerent bumps can be seen (Big Blue Bump in the middle, Infrared Bump on the left and X-ray ’Bump’ on the right). Figure from Manners (2002) ...... 8

1.2 This figure shows the unification scheme from Urry & Padovani (1995). In the center the black hole with its accreating disk is shown. Perpendicular to this disk is a jet of outflowing material. Close to the black hole are clouds which produce the broad emission line features with speeds of up to 10 000 kms−1. The torus of dust shields the broad line region depending on the observing orientation. Further out is the narrow line region...... 11

1.3 Deﬁnition of the radiative intensity Iν ...... 15

2.1 Infrared versus radio luminosity of the 3CR sample at z > 1 prior to normalization. ’x’ symbols denote quasars; circles and squares denote radio galaxies. Superposed crosses indicate radio galaxies that show evidence of silicate absorption (§2.3.1). The vertical long-dashed lines mark the range of our luminosity-matched quasar and radio galaxy subsamples. The dotted lines indicate

L8m/L178 MHz ratios of 1, 10, and 100. The radio galaxies are grouped into several SED classes in Fig. 2.3 and §2.3.1. The color- coding and symbols are: green circle (A), red circle (B), red square (C), blue square (D), blue circle (E). The two low-excitation radio galaxies 3C 68.1 and 3C 469.1 are labeled with their 3C numbers, as are sources outside the luminosity range of our analysis. . . . 22

93 94 LIST OF FIGURES

2.2 Rest frame quasar and radio galaxy SEDs normalized by rest 178 MHz 44 luminosity (L178 = 5.4×10 erg/s). Symbols connected with thick blue and red lines show the mean SEDs for quasars and radio galaxies, respectively. The thin dotted blue and red lines indicate the dispersion (upper and lower quartiles) around the mean SEDs; the mean ratio of upper/lower quartiles are 2.3 (quasars) and 3.4 (radio galaxies). The radio galaxy SED can be explained by the sum (black long-dashed line) of a reddened quasar (blue long-dashed line) and starlight from the host galaxy (thin black solid line). The long-dashed lines have been shifted slightly to make them visible in the plot. The diﬀerence between radio galaxies and reddened quasars at 10 m may be due to the silicate absorption feature which may escape detection in our broad band photometry. 30 2.3 NIR/MIR color-color diagram. The radio galaxies are grouped into ﬁve classes labeled A–E as explained in §2.3.1. Symbol color- coding is the same as in Figure 2.1. Radio galaxies with a photometric signature for silicate absorption are additionally marked with an underlying plus. The two sources 3C 267 and 3C 469.1 with spectroscopically-detected silicate absorption are labeled, as are the two low-excitation radio galaxies 3C 68.1 and 3C 469.1. The error bar in the upper left corner represents a color rms of 15%.

The AV arrow indicates screen extinction with the reddening law given in §2.3.1...... 31 2.4 Mean SED of each radio galaxy class as identiﬁed in §2.3.1. The SEDs have been normalized to the mean 178 MHz luminosity. The mean quasar SED is also shown for comparison. The dispersion around each SED (measured as mean ratio of upper/lower quartile) is: 2.3 (quasars), 2.0 (A), 3.2 (B), 2.4 (C), 1.7 (D), 1.5 (E), and 4.8(silicateabsorption)...... 32 2.5 Infrared-radio color-color diagram. The dotted line marks the region occupied by quasars. The color-coding and symbols of radio galaxies correspond to those in Figures 2.1 and 2.3. The two sources 3C 267 and 3C 469.1 with spectroscopically-detected silicate absorption are labeled as well as the two low-excitation radio galaxies 3C 68.1 and 3C 469.1. The error bar in the upper left

corner represents an rms of 15%. The AV arrow indicates screen extinction with the reddening law given in §2.3.1...... 33 LIST OF FIGURES 95

3.1 Outcome of the photometric fitting routine. The left column shows the two dimensinal fitting space of redshift and luminsosity. This helps to check the quality of the fitting procedure. The right column shows the logarithmic plot of the data, with the wavelenth on

the x-axis and νLν on the y-axis. The top left small box contains information like the redshift, χ2, position and distance to the 3C source. The small pictures on the left shows a small region around the object to check for double sources or other peculiarities. . . . 36

3.2 Observed spectral energy distributions for two good examples of the 29 cluster candidates. The IRAC and MMT data are marked with filled circles and 1σ error bars. The MIPS 24 m data point is marked with a filled circle, too; in the upper panel (elliptical source) it is a 3σ upper limit, in the lower panel (dusty starburst source) it is a 2σ detection also visible on the map. The horizontal bar indicates the 24 m pass band for comparison with the silicate absorption feature. HST photometry is marked with an open circle; it is not available for the dusty starburst source. Solid lines show the elliptical galaxy (NGC 221) fit for the source in the upper panel and the dusty starburst (Arp 220) fit for the galaxy in the lower panel. Dotted lines show the alternative template for each galaxy. The upper and lower panel shows object numbers 23 and 3, respectively, as listed in Table 3.1)...... 40

3.3 Observed spectral energy distributions for two cluster candidates with poorer data quality. The IRAC and MMT data are marked with filled circles and 1σ error bars. The sources are detected only in the z′-band (in the lower panel also in the Y -band) and at 3.6 and 4.5 m, but the upper limits at 5.8 and 8.0 m help to constrain the redshift fits. As in Fig. 3.2, the upper panel shows a source (object 11) which is preferably fit by the elliptical template and the lower panel one (29) fit by the starburst template. The dotted and dashed lines show the preferred templates at different redshifts (indicated in the figure), showing that the accuracy of the photometric redshifts should be dz . 0.2...... 41 96 LIST OF FIGURES

3.4 Color-magnitude diagram F (3.6)/F (z′) versus F (3.6), for the quasar ﬁeld (upper panel) and control ﬁeld 1 (lower panel). Sources with

zphot = 1.53 ± 0.20 (SED determined candidates marked with circles) concentrate in a distinct color range 15 < F (3.6)/F (z′) < 52 indicated by the horizontal dotted lines. The number of color determined candidates is also given...... 42 3.5 Sky distribution of the 29 candidate cluster galaxies around the quasar 3C 270.1 (marked with a cross). The triangle shows the apparent centroid of the cluster galaxy distribution. The star in the southwest marks the starburst candidate whose SED is shown in Fig. 3.2. The solid lines surround the areas covered by IRAC and HST frames. The dotted circles of radius 50′′, 100′′, and 150′′ outline the areas considered in Fig. 3.6, they are centered around the quasar. At z = 1.53, 50′′ corresponds to 427 co-moving kpc. . 44 3.6 Surface density of the 29 cluster galaxy candidates versus projected distance from the quasar 3C270.1 (solid line through fat dots with Poisson error bars). The radial bins centered around the quasar are outlined in Fig. 3.5, and the surface density of the outermost annulus has been corrected for the area not covered by IRAC. For comparison, the long-dashed histogram shows the surface density of cluster galaxy candidates versus projected distance from centroid (the triangle in Fig. 3.5). The dotted line indicates the mean surface density in the two control fields...... 45 3.7 Color-magnitude diagram F (4.5)/F (z′) versus F (4.5) of the quasar field and control field 2. top: quasar field with all sources detected at 4.5 m using SExtractor in double-image mode at 3.6 & 4.5 m. middle: quasar field restricted to those sources detected at 4.5 m using SExtractor in single-image mode, as was done for the control

field 2. bottom: control field 2. Sources with zphot = 1.53 ± 0.20 (SED determined candidates marked with circles) concentrate in a distinct color range 13 < F (4.5)/F (z′) < 54 indicated by the horizontal dotted lines. The number of color determined candidates is alsogiven...... 48 3.8 HST F702W images of four cluster galaxy candidates (object numbers 9, 12, 14 and 15 in Table 3.1). The contours are linearly spaced in steps of 10% of the peak flux value (0%, 10%, ... 90%). Arrows indicate the orientation of each panel...... 50 LIST OF FIGURES 97

3.9 Color-magnitude diagram F (3.6)/F (z′) versus F (3.6) of the quasar field and control field. top: quasar field with all sources detected at 3.6 m bottom: control field with all sources detected at 3.6 m. The dotted lines mark the range of EROs having a redshift similar to that of 3C 437. Note that there is no excess of such objects in the cluster field compared to the left field ...... 52

4.1 Illustration of our model grid and a typical photon path in the x,z-plane. A photon package is emitted at the heating source (i), interacts with the dust, and can be either absorbed (ii) or scattered (iii). It is also possible that more than one event occurs in a grid cell (iv). The line style illustrates the frequency change during absorption events. Finally the photon leaves the model space (v). 56

4.2 Floating-Point operations per second for the central processing unit (CPU-blue) and NVIDIAs graphics processing unit (GPU- green)...... 59

4.3 Show the CUDA scheme take from CUDA manual. The executes serial code and organize the exectution on the graphics cards. . . 60

4.4 Show the separation of grids into blocks. Ever thread in block can access the same shared memory while threads in diﬀerent blocks needtobeindependentofeachother ...... 61

4.5 This ﬁgure shows the geometrical situation for parallel projection.

The MC cube has the coordinate system ex,ey,ez and the detector ′ ′ plane has the coordinate system ex′ ,ey′ and the pixels x , y . The ′ ′ ′ ′ center of pixel x 0 =(x0, y0) is at position x0. The vector ez deﬁnes theobservingdirection...... 62

4.6 The left image shows the line of sight l and the position of the observer O and the intersection of l with the z-plane of the cube. The upper right Image shows the deﬁnition of the spherical coordinates. The lower right Image shows the projection of the cube...... 63

4.7 10 micron image of our solar system up to Jupiter at a distance of 1 parsec. This face on view is generated with ray-tracer and 64k × 64kpixels...... 65 98 LIST OF FIGURES

4.8 upper panel: SEDs from DUSTY as reference code for 4 different optical depth of 1(magenta), 10(blue), 100(orange), 1000(green). The colored lines shows the SED calculated with our code, while the black line represents the reference code. Note for the higher optical depth 100 and 1000 we use a larger number of iterations. lower panel: The difference of our code to the solution of Ivezic etal.(1997)islessthan5%...... 66 4.9 left column: face on disc at an inclination angle of 12.5◦ right column: edge on disc at an inclination angle of 75◦ upper panel: SEDs from (Pascucci et al. 2004) as reference code for 4 different mid-plane optical depth of τ = 1(purple), 10(green), 100(blue). The symbols shows the SED calculated with our code. Note for the higher optical depth of 100 we use a larger number of iterations. lower panel: Difference of our code with to the solution of Pascucci etal.(2004)...... 68 4.10 Extinction efficiency for used in the 2D and 3D configurations. The solid line shows the extinction coefficient for astronomical silicate grains as taken from Pascucci et al. (2004). The dashed line is the extinction coefficient described by equations 4.25-4.28 ...... 70 4.11 Density structure for the 3D spiral test case. The values are normalized to the maximum density. The density decrease from cyan(minimum)toblack(maximum) ...... 71 4.12 Left: modeled 11 m flux at a distance of 50 pc. Right: Prediction for the planed European Extremly Large Teleskop (EELT) . . . . 71 4.13 left column: face on disc at an inclination angle of 12.5◦ right column: edge on disc at an inclination angle of 75◦ upper panel: SEDs from (Pascucci et al. 2004) as reference code for 4 different mid-plane optical depth of τ = 1(purple), 10(green), 100(blue). The symbols shows the SED calculated with our code. Note for the higher optical depth of 100 we use a larger number of iterations. lower panel: only a small difference at long wavelengt of the 3D spiral expansion to the solution of Pascucci et al. (2004)...... 72 4.14 Density distribution for the 3D torus structure. On the left is the face on view. While the right shows the edge on view. Values are normalize to maximum density and the density decrease from red (maximum) to black (minimum). Even in the edge on view it is possibletoseethehotinnerdust...... 73 LIST OF FIGURES 99

4.15 Phi averaged SEDs for two diﬀerent inclination angles (face on (upper) to edge on (lower)). The simulation contains 108 photons, with 106 grid cells. The dust is distributed in 1 000, 2 500 and 5 000 clumps with an half opening angle of 35◦ around the mid- plane, with an inner radius of 0.2 pc and an outer radius of 5 pc. The optical depth at 550nm at to randomly chosen lines is

τ550nm = 115, 230 and 385. Note that the Si feature is in weak emission for the edge on case with the lowest number of clumps. 74 4.16 Image of the dust re-emission at 10 micron at an inclination of 90 degrees. It is possible to see the shielding of individual clouds as well as the three dimension structure of illuminated spheres. In direction to the source the cloud surface is hot. In contradiction is thesurfaceoppositetotheheatingsourcecold...... 76 4.17 Face on (left) and edge on (right) view of the clumpy density distribution for the best fitted torus model. Size is 25 × 25 parsec. Inner gap of 1 parsec. Half opening angle of 35 degrees ...... 77 4.18 Extinction curve for dust composition of 2 grain species silicate and carbon. The size distribution is equal to that of Mathis et al. −5 (1977) with the silicate abundance of NSi/NH = 3.1 × 10 and −4 carbon abundance of NC /NH = 2.0 × 10 ...... 77 4.19 Model of the average spectra and SED. The colored shapes represents our data from 2 − 8m the average SED and from 9 − 16m the average spectra (Leipski et al. 2010). The color represents the type of radio source red for radio galaxies and blue for quasars. The black solid line represents our model prediction. The error bars are calculated from the different viewing angles...... 79 4.20 Model of average spectra and SED. The colored shapes represents our data from 2 − 8m the average SED and from 9 − 16m the average spectra (Leipski et al. 2010). The color represents the type of radio source red for radio galaxies and blue for quasars. The black solid line represents our model prediction. The error bars are calculated from the different viewing angles...... 79 100 LIST OF ACRONYMS

3C Third Cambridge Catalogue of radio sources 3CR Third Cambridge Catalogue of radio sources revisioned AGN Active Galactic Nucleus BCD Basic Calibrated Data CPU Central Processing Unit COSMOS Cosmological Evolution Survey CUDA Compute Unified Device Architecture EELT European Extremly Large Teleskope ERO Extremly red object ESO European Southern Observatories FLOPS Floatingpoint Operations per Second FIR Far Infrared FR I / II Fanaroff and Riley I / II GPU Graphics Processing Unit ICM Intra Cluster Medium IRAC Infrared Array Camera IRS Infrared spectrograph HAWKI High Acuity Wide field K-band Imager (VLT) HST Hubble Space Telescope LERG Low-Excitation Radio Galaxy LTE Local Thermal Equilibrium ΛCDM Lambda Cold Dark Matter MC Monte Carlo MIDI MID-infrared Interferometric instrument MIPS Multiband Imaging Photometer MIR Mid Infrared MMIRS The MMT and Magellan Infrared Spectroraph MMT Multi Mirror Teleskope MPIA Max Planck Institute for Astronomy NGC New General Catalogue NASA National Aeronautics and Space Administration NED NASA Extragalactic Database NIR Near Infrared PAH Polycyclic Aromatic Hydrocarbons PSF Point Spread Function QSO Quasi stellar Object Quasar Quasi stellar radio source ROSAT Roentgen Satellite SDSS Sloan Digital Sky Survey SED Spectral energy distribution SWIRC SAO Widefield InfraRed Camera (MMT) SWIRE Spitzer Wide-Area Infrared Extragalactic Survey ULIRG Ultra Luminous Infrared Galaxy UKIDSS The UKIRT Infrared Deep Sky Survey UKIRT United Kingdom Infrared Telescope UV Ultra Violett VLT Very Large Telescope XMM X-ray multi mirror CURRICULUM VITAE 101

Curriculum vitae

Personal Profile Name: Frank Heymann Date of birth: 21.10.1982 Place of birth: Leipzig, Germany Nationality: German Address: Ruhr Universit¨at Bochum Universit¨atsstrae 150 44801 Bochum, Germany Telephone: +49 234 3223452 E-Mail: [email protected] Education since Sep. 2007 University Bochum / ESO PhD fellowship PhD thesis “Observations and models of high-redshift radio glaxies and quasars from the 3rd Cambridge Catalogue” supervised by: Dr. habil. Martin Haas / Dr. Ralf Siebenmorgen 2001 − 2007 University Leipzig: Physics, Astronomy, Computer Science Diploma/Master Physics, grade: very good Gravitational micro - lensing and the variability of quasars supervised by Prof. Helmut Meusinger 1993 − 2001 Wilhelm Ostwald Gymnasium Leipzig, Germany Abitur (Intensive courses: Mathematics, Physics, German) Work Experience 2010 University Bochum 2008 − 2010 ESO 2007 − 2008 University Bochum Extra astronomical Activity 2003 − 2006 Work student at Mercedes Benz VGmbH NDL Leipzig Internet presentation 2001 − 2007 Alternative military service Emergency service by the German Red Cross 102 CURRICULUM VITAE DECLARATION 103

Declaration

I hereby certify that this PhD thesis has been composed by myself and is based on my own work, unless stated otherwise. This work has not been submitted for any other degree.

Ich versichere hiermit, dass ich die vorliegende Dissertation eigenständig ange- fertigt und ausschließlich die angegebenen Quellen und Hilfsmittel verwendet habe. Diese Arbeit ist in dieser oder ähnlicher Form noch keiner Prüfungsbehörde vorgelegen.

Bochum, Germany, April 30, 2010

Frank Heymann