Galaxies in the Epoch of Reionization

Home , Big Bang nucleosynthesis, Inflation (cosmology), Lyman series, Reionization

Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology 2036

Galaxies in the epoch of reionization

Investigating the high-redshift galaxy population through simulations and observations

CHRISTIAN BINGGELI

ACTA UNIVERSITATIS UPSALIENSIS ISSN 1651-6214 ISBN 978-91-513-1196-8 UPPSALA urn:nbn:se:uu:diva-440032 2021 Dissertation presented at Uppsala University to be publicly examined in Polhemsalen, Ångströmlaboratoriet, Lägerhyddsvägen 1, Uppsala, Thursday, 10 June 2021 at 13:15 for the degree of Doctor of Philosophy. The examination will be conducted in English. Faculty examiner: Stephen Wilkins (University of Sussex).

Online defence: https://uu-se.zoom.us/j/68040940351

Abstract Binggeli, C. 2021. Galaxies in the epoch of reionization. Investigating the high-redshift galaxy population through simulations and observations. Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology 2036. 87 pp. Uppsala: Acta Universitatis Upsaliensis. ISBN 978-91-513-1196-8.

The cosmic reionization is the last major gas phase transition in cosmic history, yet it remains poorly understood. Current constraints indicate that early star-forming galaxies drove the reionization process through producing and releasing large numbers of ionizing photons into the intergalactic medium. However, our understanding of the ionizing escape fraction (fesc) and the general properties of high-redshift galaxies is still limited. In this thesis, simulated galaxies and observations are used to investigate epoch-of- reionization galaxies and to explore methods that can aid future investigations of such objects. Using simulations, we have shown that it may be possible to constrain fesc in epoch-of- reionization galaxies using quite simple diagnostics that should be observable with the upcoming James Webb Space Telescope (JWST). We also show that variations in star formation activity larger than those predicted in our simulations may lead to a possible degeneracy with high fesc. However, auxiliary information obtained with the JWST may allow us to disentangle variations in the star formation activity from high fesc. We compare galaxies from several simulations to the recently spectroscopically confirmed z=9.1096 galaxy MACS1149-JD1. We find that none of the simulations are able to reproduce the large Balmer break observed in MACS1149-JD1, and argue that unless it represents an outlier in the high-redshift galaxy population, this may indicate that the simulations fail to capture some key physics. Finally, we present ALMA observations of the z=7.6637 galaxy z7_GSD_3811. This object remains undetected in several commonly detected FIR emission lines and FIR dust emission. Using SED-fitting and by comparing our observations to models and low-redshift observations, we show that our non-detections could indicate that the object is poor in metals and dust. Our findings could help future observers to further constrain the nature of high-redshift galaxies and their role in reionization.

Keywords: galaxies: high-redshift, galaxies: ISM, galaxies: evolution, reionization, Lyman continuum

Christian Binggeli, Department of Physics and Astronomy, Observational Astronomy, 516, Uppsala University, SE-751 20 Uppsala, Sweden.

ISSN 1651-6214 ISBN 978-91-513-1196-8 urn:nbn:se:uu:diva-440032 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-440032) Till Linnéa

List of papers

This thesis is based on the following papers, which are referred to in the text by their Roman numerals.

I Zackrisson, E., Binggeli, C., Finlator, K.; Gnedin, N. Y., Paardekooper, J.-P., Shimizu, I., Inoue, A. K., Jensen, H., Micheva, G., Khochfar, S. and Dalla Vecchia, C. (2017) The spectral evolution of the ﬁrst galaxies. III. Simulated James Webb Space Telescope spectra of reionization-epoch galaxies with Lyman-continuum leakage The Astrophysical Journal 836, 78

II Binggeli, C., Zackrisson, E., Pelckmans, K., Cubo, R., Jensen, H. and Shimizu, I. (2018) Lyman continuum leakage versus quenching with the James Webb Space Telescope: The spectral signatures of quenched star formation activity in reionization-epoch galaxies Monthly Notices of the Royal Astronomical Society 479, 368-376

III Binggeli, C., Zackrisson E., Ma, X., Inoue A. K., Vikaeus, A., Hashimoto, T., Mawatari, K., Shimizu, I. and Ceverino D. (2019) Balmer breaks in simulated galaxies at z>6 Monthly Notices of the Royal Astronomical Society 489, 3827-3835

IV Binggeli, C., Inoue, A. K., Hashimoto, T, Toribio, M. C., Zackrisson, E., Ramstedt, S., Mawatari, K., Harikane, Y., Matsuo, H., Okamoto, T., Ota, K., Shimizu, I., Tamura, Y., Taniguchi, Y. and Umehata, H. (2021) A puzzling non-detection of [O III] and [C II]fromaz≈ 7.7 galaxy observed with ALMA Astronomy & Astrophysics 646, A26

Reprints were made with permission from the publishers.

Errata

I Zackrisson, E., Binggeli, C., Finlator, K.; Gnedin, N. Y., Paardekooper, J.-P., Shimizu, I., Inoue, A. K., Jensen, H., Micheva, G., Khochfar, S. and Dalla Vecchia, C. (2021) Erratum: The spectral evolution of the ﬁrst galaxies. III. Simulated James Webb Space Telescope spectra of reionization-epoch galaxies with Lyman-continuum leakage The Astrophysical Journal 908, 116

II Binggeli, C., Zackrisson, E., Pelckmans, K., Cubo, R., Jensen, H. and Shimizu, I. (2020) Erratum: Lyman continuum leakage versus quenching with the James Webb Space Telescope: The spectral signatures of quenched star formation activity in reionization-epoch galaxies Monthly Notices of the Royal Astronomical Society 496, 1766-1767

List of papers not included in the thesis

The following are publications to which I have contributed as author, but which are not included in the thesis.

1 Sugahara, Y., Inoue, A. K., Hashimoto, T. et al. (2021) Big Three Dragons: A [N II] 122 μm Constraint and New Dust- continuum Detection ofAz=7.15 Bright Lyman Break Galaxy with ALMA Submitted to The Astrophysical Journal

2 Vikaeus, A., Zackrisson, E. and Binggeli, C., (2020) The impact of star formation sampling effects on the spectra of lensed z > 6 galaxies detectable with JWST Monthly Notices of the Royal Astronomical Society 492, 1706-1712

3 Zackrisson, E., Majumdar, S., Mondal, R. et al. (2020) Bubble mapping with the Square Kilometer Array-I. Detecting galaxies with Euclid, JWST, WFIRST and ELT within ionized bubbles in the intergalactic medium at z> 6 Monthly Notices of the Royal Astronomical Society 493, 855-870

4 Giri, S. K., Zackrisson, E., Binggeli, C., Pelckmans, K. and Cubo, R. (2020) Identifying reionization-epoch galaxies with extreme levels of Lyman continuum leakage in James Webb Space Telescope surveys Monthly Notices of the Royal Astro nomical Society 491 5277-5286

5 Tamura, Y., Mawatari, K., Hashimoto, T. et al. (2019) Detection of the Far-infrared [O III] and Dust Emission in a Galaxy at Redshift 8.312: Early Metal Enrichment in the Heart of the Reionization Era The Astrophysical Journal 874, 27

6 Jensen, H., Zackrisson, E., Pelckmans, K., Binggeli, C.., Ausmees, K. and Lundholm, U., (2016) A Machine-learning Approach to Measuring the Escape of Ionizing Ra- diation from Galaxies in the Reionization Epoch The Astrophysical Journal 827, 5

Contents

1 Introduction ...... 13

2 Hydrogen and Lyman continuum radiation ...... 15

3 Cosmic reionization ...... 17 3.1 Constraints on reionization ...... 18 3.1.1 Constraints from quasars ...... 18 3.1.2 Constraints from the cosmic microwave background ... 20 3.1.3 Constraints from Lyman-α emitters ...... 22 3.2 Driving sources of cosmic reionization ...... 23

4 Properties of high-redshift star-forming galaxies ...... 25 4.1 Dust properties ...... 25 4.2 Star formation ...... 29 4.3 Far-infrared observations and the interstellar medium ...... 31

5 The Lyman continuum escape fraction ...... 36 5.1 Mechanisms of Lyman continuum leakage ...... 37 5.2 Observations of leaking Lyman continuum ...... 38 5.3 Indirect methods for constraining the escape fraction ...... 39 5.3.1 Emission-line constraints on the escape fraction ...... 39 5.3.2 Absorption-line constraints on the escape fraction ...... 41 5.3.3 Escape fractions from ionized proximity-zones and bubbles ...... 41 5.3.4 Zackrisson et al. 2013: Nebular emission features ...... 42

6 Synthetic spectra of high-redshift galaxies ...... 46 6.1 Population synthesis and model galaxy spectra ...... 46 6.2 Simulated galaxies ...... 47 6.3 Spectra of simulated galaxies ...... 49

7 Radio interferometry and interferometric imaging ...... 53

8 Summary of papers ...... 57 8.1 Paper I ...... 57 8.2 Paper II ...... 61 8.3 Paper III ...... 64 8.4 Paper IV ...... 67

9 Summary and outlook ...... 71 10 Contributions to included papers ...... 73

11 Svensk sammanfattning ...... 75

12 Acknowledgements ...... 79

References ...... 81 1. Introduction

This thesis is devoted to studying galaxies during a period known as the epoch of reionization (EoR). Cosmic reionization represents one of the major gas phase transitions that have occurred throughout the history of Universe. As will be discussed in this thesis, reionization is an important yet poorly understood chapter in cosmic history. The phase transition just prior to reionization was the recombination, during which the first neutral atoms formed, leading to a pre-galactic medium that consisted of neutral hydrogen, helium and trace amounts of lithium. During reionization, the process was reversed as hydrogen-ionizing photons from the first astrophysical sources were emitted into the neutral intergalactic medium (IGM). While helium also underwent a reionization, this thesis focuses on the reionization of hydrogen. The term ‘reionization’ will thus be used in the context of hydrogen reionization, unless otherwise stated. Over the years, observational studies with increasingly advanced telescopes covering a wide wavelength range have been aimed at understanding the reionization and its drivers. Meanwhile, on the theory side, cosmological simulations run on super-computers have provided the astronomy community with increasingly detailed models of the process (see reviews by e.g. Fan et al., 2006a; Stark, 2016; Dayal & Ferrara, 2018). Through these efforts, some facets of the reionization process, in particu- lar its timing and duration, have been fairly well constrained. Over the years, the emerging picture has become one of a reionization driven mainly by high- redshift star-forming galaxies. These galaxies thus were not only the progen- itors of the present-day galaxy population, but by driving reionization, they also affected the presence and distribution of neutral hydrogen in the IGM. As such, they likely had an impact on the subsequent formation of stars. How- ever, our picture of early star-forming objects is still incomplete. One part of the picture that yet has to be filled in relates to the fraction of ionizing photons released from these galaxies into the IGM (the escape fraction; fesc), and the role this played in the reionzation process. The main obstacle in determining the escape fraction of EoR galaxies is that we cannot observe the leaking LyC directly. The reason for this is that the LyC photons that leak from these galaxies are consumed in the neutral IGM during reionization, and therefore never reach us (Inoue & Iwata, 2008). One way to circumvent this obstacle is to observe ionizing emission from so-called low-redshift analogues in order to understand the physics behind the leakage of ionizing photons and link these to other parts of the electromagnetic spectrum. Another way is to use models and simulations to predict

13 observable features that can be linked to leakage of ionizing photons. Both of these avenues have their limitations in the sense that it is unclear to what degree low-redshift analogues or simulated galaxies are actually representative of the galaxies that likely drove reionization. Thus, understanding the fundamental properties of the high-redshift galaxy population, such as their star formation histories and dust properties is a piece of the puzzle of trying to form a complete picture of the evolution of the Universe. In coming years, the James Webb Space Telescope (JWST) will allow astronomers to study high- redshift galaxies at electromagnetic wavelengths hitherto unavailable for such distant and faint objects. In order to get the most out of the data the JWST will produce, it is crucial to have well-calibrated models that allow us to interpret the observations. The focus of this work is the high-redshift galaxies that likely drove the cosmic reionization, their properties and how we can use models in order to understand future observations. First, I discuss a method for indirectly assessing the escape fraction of ionizing photons from EoR galaxies. I test this method on spectra of simulated high-redshift galaxies with varying star formation histories, internal metallicity distributions and escape fractions. I also discuss problems with this approach related to possible variations in the star formation history of high-redshift galaxies. Related to this, I use cosmological simulations in order to understand recent observational results of the galaxy MACS1149-JD1, a redshift z ∼ 9.1 galaxy suggested to have experienced large variations in its star formation activity. I discuss to what extent contemporary simulations are able to reproduce these observations. Finally, I present ALMA observations of a z ∼ 7.7 star-forming galaxy (z7_GSD_3811) and constrain properties of the object using the observations.

14 2. Hydrogen and Lyman continuum radiation

During reionization, the neutral hydrogen first formed during recombination (Peebles, 1968) was reionized by highly energetic photons emitted from early astrophysical sources. In order to understand the reionization and the evolution of the IGM, we need to start with the most abundant element in the Universe, i.e. hydrogen. The ionization potential of the hydrogen atom is 13.6 eV. This means that ultraviolet (UV) photons with energies above 13.6 eV (photons with wavelengths shorter than 912 Å) are able to eject the single electron from the hydrogen atom and ionize it. This energy limit, above which photons will be able to ionize hydrogen, is called the Lyman limit, and radiation with energies above the Lyman limit is called Lyman-continuum (LyC) radiation (Draine, 2011). The reionization process thus requires astrophysical sources that are able to efficiently produce LyC in order to ionize the IGM. Several types of astrophysical objects produce LyC photons, but in the context of this thesis, the most important sources are stars in early star-forming galaxies. The stars that most efficiently produce LyC photons are hot and massive stars. The large ionizing flux from these stars can ionize the surrounding hydrogen (and other elements) in the interstellar medium (ISM) to form so- called H II regions. In such ionized regions, it is likely that the free electrons encounter another proton, and recombine back into neutral hydrogen. When this recombination occurs to the ground state of hydrogen directly, a new LyC photon is emitted. This will then likely be absorbed by another hydrogen atom in the surrounding ISM, leading to additional ionizations. If the amount of neutral hydrogen in the surrounding ISM is small enough, a fraction of the produced LyC photons can escape the ISM without encountering a hydrogen atom. If the recombination occurs to an excited state of hydrogen, a number of non-ionizing photons will be emitted as the electron cascades down through the discrete energy levels of hydrogen on its way to the ground state. The transition between an excited state of hydrogen and the ground state will produce a UV photon in the Lyman series. For example, the transition between the first excited state (2p) to the ground state (1s) leads to the emission of a Lyman-α (Lyα) photon, with a wavelength ≈ 1216 Å. The transitions from the second and third and higher excited states to the ground state are called Lyman-β, Lyman-γ and so on. Similarly, transitions from higher excited states to the first excited state of hydrogen lead to emission of photons in the Balmer series (mainly optical photons, usually denoted Hα,Hβ,Hγ and so on). Since H II regions form mainly around hot, massive and thus short-lived stars, the Lyman and Balmer emission lines trace the recent star formation

15 (e.g. Kennicutt, 1998; Kennicutt & Evans, 2012). While the transport of Lyα through the IGM becomes complicated due to its resonant nature (Hayes et al., 2010), the Balmer lines are excellent probes of star formation when the escape fraction is low. With the upcoming JWST, we should be able to observe UV and optical emission lines during the very earliest stages of galaxy formation and evolution. In paper I and paper II, we discuss a method for determining the amount of LyC radiation that escapes from high-redshift galaxies using Balmer lines in combination with the UV slopes (see section 4.1) of these objects.

16 3. Cosmic reionization

The current standard model of Big Bang cosmology is the Lambda cold dark matter (ΛCDM) model. A widely accepted extension to ΛCDM is cosmological inflation. According to this model, the Universe started in an extremely hot and dense state. Within the first fraction of a second, it is thought to have undergone a rapid expansion known as inflation (Guth, 1981), in which the volume of the Universe increased by a factor of ∼ 1078 (Guth & Kaiser, 2005). It is thought that this rapid expansion led to the homogeneous and isotropic distribution of matter which is reflected in large-scale galaxy surveys and observations of the cosmic microwave background radiation (CMBR). Af- ter inflation followed a period of more modest expansion and cooling. During this period, the Big Bang nucleosynthesis led to the formation of the first composite nuclei (Alpher et al., 1948). However, the Universe was still too hot for neutral atoms to form, and so it remained optically thick to photons, since these would scatter off free, charged particles in the hot, dense and ionized gas. At z ∼ 1100, the Universe had cooled to a temperature that allowed the first neutral atoms to form in the event known as recombination (Peebles, 1968). As a result of recombination, the photons could decouple from the baryonic matter and flow freely through the transparent Universe (see e.g. Barkana & Loeb, 2001). The light emitted at recombination is observed today as a nearly uniform background in all directions on the sky, and is known as the cosmic microwave background radiation (CMBR). While the post-recombination Universe allowed photons to travel freely, no sources of light had yet formed. Instead, the Universe was filled with a neutral gas. The formation of the first stars signified the end of this period known as the Dark Ages (see Rees, 1998), and the beginning of the cosmic dawn. The cosmic dawn culminated with the cosmic reionization, when the the first stars and galaxies produced LyC at a rate which could not be matched by the recombination rate in the IGM and thereby started to ionize the surrounding gas. Models suggest that in the early parts of reionization, only the gas in close proximity of the objects became ionized, leading to ‘bubbles’ of ionized gas. As a result of variations both in the gas density and the distribution of ionizing sources, models predict that reionization was patchy and inhomogeneous, with large variations in the neutral gas fraction (e.g. Furlanetto et al., 2004; Iliev et al., 2006). Over time, and as the number of ionizing sources increased, smaller ionized regions merged, leaving ever smaller patches and filaments of neutral gas that were slowly etched away over time. It has been argued that

17 this patchy nature of reionization is reflected in observations of high-redshift quasars which have showed that the neutral fraction of the IGM varies significantly between different lines of sight (e.g. Fan et al., 2006b; Willott et al., 2007; Becker et al., 2015; Barnett et al., 2017; Bosman et al., 2018; Eilers et al., 2018). An illustration showing the evolution of the Universe from the Big Bang to the current day is shown in figure 3.1. While this thesis focuses on the study of the galaxies present during reionization, there are also ways to study the IGM in the early Universe more directly. The ‘spin-flip’ 21 cm spectral line of neutral hydrogen provides a unique way to study the neutral hydrogen and its distribution in the pre- reionization and reionization era. This line, which has a wavelength of 21 cm, arises due to the difference in the energy of the two hyperfine levels in the ground state of hydrogen (Pritchard & Loeb, 2012). Due to cosmological redshift, the line is measured at meter wavelengths at the present day. Re- cently, the EDGES (Experiment to Detect the Global EoR Signature) team claimed to have detected the signal at z ∼ 17, and argued that the their measurement is consistent with the signal induced by early stars (Bowman et al., 2018). Other ongoing and planned experiments aimed at measuring the global 21 cm signal, such as the Large-Aperture Experiment to Detect the Dark Ages (LEDA), Radio Experiment for the Analysis of Cosmic Hydrogen (REACH), Shaped Antennas to measure the background RAdio Spectrum (SARAS3) and Probing Radio Intensity at high-Z from Marion (PRIZM), will likely be able to confirm this detection and provide additional constraints on the globally averaged signal. In addition, experiments such as the Low-Frequency Array (LOFAR), the Hydrogen Epoch of Reionization Array (HERA) and the future Square Kilometre Array (SKA) may be able to measure spatial fluctuations in the 21 cm signal.

3.1 Constraints on reionization In order to understand the driving mechanism behind the cosmic reionization, it is crucial to constrain the timing and duration of the process. Over the years, several different observational probes have been able to provide us with constraints on the evolution of the IGM in the EoR. In the following three sections, I will give an overview of current constraints and describe some of the ways in which these have been derived.

3.1.1 Constraints from quasars Quasars are an extremely bright type of active galactic nuclei (AGN). Their high brightness allows us to observe and use them as background sources to probe the IGM at very large distances. This makes them an extremely useful tool when trying to constrain the timing of cosmic reionization. One strong

18 rae ytepeec fsgicn mut fnurlhdoe,i be- in hydrogen, neutral of gaps, amounts dark significant with of troughs, presence absorption gas the the ionized in of by ‘spikes’ Patches created transmission IGM. to ionized lead inhomogenously will an in fraction neutral the that indication an at as done seen mostly wavelengths is was these it reionization fractions, cosmic at neutral absorption small relatively the Ly complete α for as quasar known occurs since is the and (1965), of Peterson trough, the & Gunn Gunn-Peterson of by side predicted originally blue redshifts feature, the At This on reionization. absorption approach intervening complete of we observe number as increasing the clouds to gas due occurs neutral This Ly 3.2). α the figure the sight, (see up of shifts making lines lines different absorption between of varies gas number intervening of depth Lyα optical the the as features Ly known α absorption is quasar features of redshifted absorption set the such of a side Lyα produce blue absorb will the will this on clouds redshift, the their As to quasar. corresponding The distant quasars. a toward redshift Ly high sight α the of from spectra increas- comes the redshifts the in Ly higher and trough at reionization IGM Gunn-Peterson the cosmic and in forest the gas of neutral of timing fraction the ing for evidence of piece motn vnsi h omchsoyaehglgtda h o ftefiue recom- figure: the of top the at highlighted (z Some are ( left. history bination the cosmic right toward the increases the in redshift to events the Universe and important right present-day the the toward increases and time left, cosmic the to inflation and 3.1. Figure rm,21) eoiain(z reionization 2013), Bromm,

BIG BANG α h rsneo opeeasrto ruh a eue ocntanthe constrain to used be can troughs absorption complete of presence The oeti h euto luso nevnn eta a ln h ieof line the along gas neutral intervening of clouds of result the is forest ∼ nilsrto hwn h vlto fteUies,fo h i Bang Big the from Universe, the of evolution the showing illustration An

INFLATION RECOMBINATION 10,tecsi anadfraino h rtsas(z stars ﬁrst the of formation and dawn cosmic the 1100),

FIRST STARS z ~ 1100 ~ z ∼

0–6 e .) n h nvretdy(z today Universe the and 3.1), see 6, – 10

z ~ 30 - 20 - 30 ~ z REDSHIFT z

∼ z ~ 10 - 6 - 10 ~ z 6. oet(e ac,19) While 1998). Rauch, (see forest msinln.Tecleto of collection The line. emission oeticessa ihrred- higher at increases forest

REIONIZATION z = 0 = z z ∼ twavelengths at

NOW = esatto start we 6 ) oethat Note 0). ∼ = 0–20; – 30 0). line. 19 tween. The number and length of these dark gaps provide information that can be used to constrain the neutral fraction of the IGM (e.g. Songaila & Cowie, 2002; Fan et al., 2006b; Mesinger, 2010). With a related method, McGreer et al. (2015) recently showed that the fraction of pixels showing no flux in high-redshift quasar spectra are consistent with a neutral fraction ≤ 6% at z = 5.9, which is also consistent with a reionization process that was mostly done at z ∼ 6. If the IGM neutral fraction is large, the absorption in the Gunn-Peterson trough will have a long tail, or wing, extending toward longer wavelengths. The shape of this feature, known as the Gunn-Peterson damping wing, will depend on the neutral fraction of hydrogen around a target quasar, and can thus be used to probe the neutral fraction using high-redshift quasars (Miralda- Escudé, 1998). This method becomes complex mainly due to the risk of find- ing dense clumps of neutral hydrogen in an otherwise ionized IGM. Such a clump in front of a targeted quasar would give a similar signature as a substantially neutral IGM. Nevertheless, this method has been used to constrain the neutral fraction of the IGM during reionization using both quasars and gamma-ray bursts (GRBs; e.g. Totani et al., 2006; Mortlock et al., 2011; Totani et al., 2016; Greig et al., 2017; Bañados et al., 2018; Davies et al., 2018; Greig et al., 2019; Wang et al., 2020). While there are still significant differences between the neutral fractions derived by different authors, even for the same targets, the overarching picture is consistent with a reionization process which was still ongoing at z ∼ 7 and which was mostly done at z ∼ 6. It is, however, important to note, that there are still open questions regarding the timing of reionization inferred from quasar spectra. While quasar constraints have been argued to indicate a reionization process which was mostly finished around z ∼ 6, scenarios in which large patches of neutral hydrogen exist all the way to z ∼ 5.5, and where reionization finishes at z = 5.3, have been argued to better explain the large variation and evolution in Gunn-Peterson troughs at z 5.5 (Kulkarni et al., 2019a).

3.1.2 Constraints from the cosmic microwave background Another powerful probe of the timing of cosmic reionization is the CMBR. Since the CMB photons emitted at recombination may scatter off free electrons as they travel though space, they are sensitive to the density of free electrons in the IGM. An imprint left by this scattering is a damping of small-scale temperature anisotropies in the CMBR. Free electrons generated during reionization will lead to a larger optical depth of CMBR photons, through which the timing of the reionization process can be constrained (see e.g. Haiman & Knox, 1999). For example, an early reionization scenario leads to a larger optical depth compared to a late reionization scenario, and thus will also lead to a larger damping of small-scale anisotropies. In addition, the scattering of

20 3 PG1634+706 z=1.337 2 1 0 2400 2450 2500 2550 2600 2650 2700 2750 2800 2850

3 1422+231 z=3.635 2 1 0 4800 4900 5000 5100 5200 5300 5400 5500 5600 Flux (arb. units)

3 1030+0524 z=6.290 2 1 0 7600 7800 8000 8200 8400 8600 8800 Wavelength (Å) Figure 3.2. The Lyα forest in spectra of quasars observed at z = 1.337 (top), z = 3.635 (middle) and z = 6.290 (bottom). The spectra are shown at observed wavelengths. The Lyα forest lines, increasing with redshift can be seen on the blue (left) side of the redshifted quasar Lyα emission line, shown to the to the right in the spectra (at ≈ 2840 Å in the top panel). Image credit: Robert Carswell. Retrieved March 1, 2021 from Bob Carswell’s webpage, https://www.ast.cam.ac.uk/~rfc/zevol6e. jpg. Adapted with permission.

CMB photons on free electrons at reionization affects the polarization in the CMBR. By combining measurements of the polarization with measurements of the temperature anisotropies, the accuracy of optical depth measurements and thus constraints on reionization can be substantially improved (Zaldar- riaga et al., 1997; Eisenstein et al., 1999). The most recent measurements of CMB anisotropies, including both polarization and temperature maps, indicate a midpoint of reionization of zre = 7.7 ± 0.7 (Planck Collaboration et al., 2020). Taking into account the effect of photon scattering off electrons with a bulk velocity (the Kinetic Sunyaev- Zeldovich effect) allows for a better determination of the duration of reionization (Planck Collaboration et al., 2016). This does, however, require a model of the patchy reionization process. This was first done by Zahn et al. (2012), who found a value of Δz < 4 (95% confidence limit) using data from the South Pole Telescope (SPT). Here, Δz is defined as the redshift duration over which the Universe goes from 80% neutral to 1% neutral. More recently, Planck Col- laboration et al. (2016) found a value of Δz < 2.5 (95% confidence limit, with the same Δz definition as Zahn et al., 2012), using Planck data in combination with data from the Atacama Cosmology Telescope and the SPT. Their results also indicate that the Universe was less than 10% neutral at z 10.

21 3.1.3 Constraints from Lyman-α emitters A third way in which we can constrain the timing of cosmic reionization is through observations of Lyα emitters (LAE) at high redshifts. Since Lyα photons get absorbed by neutral hydrogen, two consequences arise from cosmic reionization. Firstly, we expect to observe a decrease in the number of LAE with increasing redshift at the EoR. Secondly, a patchy reionization process leads to large spatial variations in the neutral fraction, and creates large ionized regions through which Lyα is able to escape. Thus, patchy reionization should leave an imprint in the clustering of Lyα-emitting sources. Observations of Lyman-break galaxies (LBGs; high-redshift galaxies iden- tiﬁed via photometry using the rapid decline in ﬂux at 912 Å due to absorption of LyC in the IGM) have shown that the fraction of these that also exhibit Lyα emission declines at z 6 (Pentericci et al., 2014; Schenker et al., 2014; Fu- rusawa et al., 2016). Additional indications of an increased neutral fraction at z 6 come from the Lyα luminosity function. The Lyα luminosity function gives the space density of LAE as a function of Lyα luminosity. Similarly, the UV luminosity function gives the space density of UV-emitting galaxies per UV luminosity or magnitude. Several studies have found a redshift evolution of the Lyα luminosity function at z 6 (Ouchi et al., 2010; Kashikawa et al., 2011; Konno et al., 2014; Zheng et al., 2017; Ota et al., 2017; Konno et al., 2018). This evolution is not matched by a corresponding evolution in the UV luminosity function, indicating that Lyα becomes rarer at these redshifts. Both the decline in the fraction of LBGs which exhibit Lyα and the evolution of the Lyα luminosity function is consistent with an IGM which is increasingly neutral at z 6. One has to keep in mind that the possible evolution of galaxy properties may also lead to this effect. However, galaxies at lower redshift seem to follow the opposite trend, with stronger Lyα at increasing redshift, possibly due to the lower dust content (Hayes et al., 2011) and/or lower neutral hydrogen covering fractions (Jones et al., 2013) as we move toward earlier times. Several studies have used the clustering of LAE in combination with reionization models to constrain the neutral fraction (Ouchi et al., 2010; Sobacchi & Mesinger, 2015; Hutter et al., 2015; Ouchi et al., 2018). Recent measurements from Ouchi et al. (2018), using a sample of 2000 LAE at z ∼ 6−7, found good consistency with several other constraints on reionization, indicating a neutral fraction of 0.15 ± 15 at z ∼ 6.6. Slightly higher neutral fractions are inferred from the evolution of Lyα equivalent widths in Lyman-break galaxies at a similar redshift. For example, Mason et al. (2018); Hoag et al. (2019) found . 0.11 ∼ . 0.05 = . ± . neutral fractions of 0 59−0.15 at z 7 and 0 88−0.10 at z 7 6 0 6.

22 3.2 Driving sources of cosmic reionization There is still very much an ongoing debate regarding the details of reionization and its driving sources, ranging from star-forming galaxies to quasars to ex- otic sources of ionizing radiation. With the help of the constraints discussed in the previous sections, it is possible to infer which astrophysical sources likely drove the cosmic reionization. Over the years, backed by a large number of observational and theoretical studies, star-forming galaxies have become the main candidate for driving reionization (e.g. Faucher-Giguère et al., 2008; Ra- zoumov & Sommer-Larsen, 2010; Becker & Bolton, 2013; Robertson et al., 2015; Kulkarni et al., 2019b; Dayal et al., 2020). However, a galaxy-driven reionization hinges on the amount of LyC that escapes from EoR galaxies, which is still highly uncertain. A number of analyzes have shown that current constraints on reionization seem consistent with a purely galaxy-driven reionization if galaxies have average escape fractions somewhere around 10 – 30% (Finkelstein et al., 2012a; Sun & Furlanetto, 2016; Mitra et al., 2018; Naidu et al., 2020). There are, however, also studies that present scenarios where star-forming galaxies are able to drive the cosmic reionization with a substantially lower average escape fraction (< 5%; see e.g. Finkelstein et al., 2019). As discussed in section 5.2, there are now a substantial number of detections of leaking LyC from galaxies up to redshift ∼ 4, but whether galaxies in the EoR generally exhibit escape fractions fesc 10% is still somewhat unclear. There are also other open questions regarding the role of star-forming galaxies in reionization. For example, the shape of the UV luminosity function at the faint end, and therefore the contribution of objects too faint to be observed at present, is still uncertain (e.g. Stark, 2016). This is especially interesting considering the large number of theoretical studies that highlight the contribution of low mass galaxies to the total ionizing budget (Razoumov & Sommer- Larsen, 2010; Ferrara & Loeb, 2013; Wise et al., 2014; Paardekooper et al., 2015; Liu et al., 2016; Dayal et al., 2017, 2020). In addition to low-mass galaxies being more numerous than high-mass galaxies, simulations also indicate that the more shallow potential wells in low-mass objects may facilitate high escape fractions (Razoumov & Sommer-Larsen, 2010; Ferrara & Loeb, 2013; Wise et al., 2014; Paardekooper et al., 2015). In contrast, Sharma et al. (2016); Naidu et al. (2020) found that bright objects that are within current detection limits release the bulk of the LyC photons required for the cosmic reionization. Their result showed that the escape fraction may follow the opposite trend, with higher escape fractions in bright, massive objects. A reionization process primarily driven by massive galaxies seems, however, to be at odds with recent constraints from the Low Frequency Array (LOFAR; Mondal et al., 2020). There are also uncertainties regarding the properties of the high- redshift galaxy population in general. The escape fraction of ionizing photons and/or ionizing production efﬁciency may increase at higher redshifts, possibly complicating inferences made using observations at low redshifts (see e.g.

23 Inoue et al., 2006; Becker & Bolton, 2013; Stark et al., 2015; Bouwens et al., 2016). There is also an ongoing debate regarding the role of quasars and other types of AGN in the cosmic reionization. These objects are not only efﬁcient pro- ducers of ionizing radiation, but some evidence also suggests that they exhibit high ionizing escape fractions ( fesc 50%, (Cristiani et al., 2016; Grazian et al., 2018; Romano et al., 2019)). Often, the escape fraction of quasars is assumed to be 100%. However, there is still some uncertainty in these values, with some observations pointing toward lower escape fractions (Cowie et al., 2009; Micheva et al., 2017). One of the main arguments against an AGN- driven reionization is simply that the space density of these objects decreases at z 3 (e.g. Fan et al., 2004; Masters et al., 2012), which would make them too rare during EoR to produce the bulk of the ionizing photons. A recent detection of a faint population of AGN at z ∼ 4 − 6 by (Giallongo et al., 2015) re-sparked this debate, and a number of recent studies have found that AGN may have signiﬁcantly contributed to reionization (Boutsia et al., 2018; Giallongo et al., 2019; Grazian et al., 2020). However, other studies have been unable to verify these results, and the larger part of the literature still points toward a subdominant contribution to the reionization photon budget by AGN (Cowie et al., 2009; Willott et al., 2010; Kashikawa et al., 2015; Ricci et al., 2017; Onoue et al., 2017; McGreer et al., 2018; Parsa et al., 2018; Kulkarni et al., 2019b; Cowie et al., 2020; Shin et al., 2020; Shen et al., 2020; Kim et al., 2020). This scenario is also consistent with a number of recent theoretical studies which favor a galaxy-driven reionization process (e.g. Qin et al., 2017; Mitra et al., 2018; Hassan et al., 2018; Puchwein et al., 2019). While most of the discussion regarding the sources of reionization has focused on star-forming galaxies and AGN, it is likely that other sources, such as X-ray binaries and microquasars, also contributed to reionization. Whether this contribution was substantial, or only marginal is, however, still unclear (Tanaka et al., 2012; Madau & Fragos, 2017; Douna et al., 2018).

24 4. Properties of high-redshift star-forming galaxies

The key to forming a complete picture of the cosmic reionization lies in understanding the galaxies present in the high-redshift Universe. Part of the work toward this understanding has come down to pushing the sensitivity of spectroscopic observations in order to spectroscopically confirm galaxies at increasingly high redshifts. Such observations have provided successful spectroscopic detections all the way up to z ∼ 111, albeit with steeply decreasing numbers at the highest redshifts. In addition, a large number of both observational and theoretical studies have aimed to provide a more detailed account of the properties of high-redshift galaxies in general. In the following sections, some of these properties will be discussed. In figure 4.1 an example of a model high-redshift galaxy spectrum is shown. Some of the spectral features that have been used to constrain properties of the high-redshift galaxy population, and that are discussed in the following sections are indicated in this figure. Also shown in this figure are the wavelength coverages of some of the telescopes/instruments that have been extensively used to study these objects, such as the Hubble space telescope (HST), the Spitzer space telescope, the Very Large Telescope (VLT), the Keck observatory and the ALMA observatory. I also indicate the wavelength coverage of the upcoming JWST, which will provide photometric and spectroscopic capabilities covering a large wavelength range, from 0.6 μmto28μm with a significantly larger light-collecting area than current space-based facilities.

4.1 Dust properties In addition to dark matter, stars and gas, most galaxies contain some amount of interstellar dust. This dust resides in the ISM in the form of submicron-sized grains consisting of compounds of metals2 such as carbon, oxygen, magne- sium, silicon and iron (Draine, 2011). Since the elements that are used to build dust grains are metals, the dust content of galaxies is expected to follow the metal content of these objects. In addition to affecting the physics inside galaxies, dust has a strong impact on the light that we observe through

1GN-z11 is currently the most distant galaxy which has been spectroscopically conﬁrmed (Oesch et al., 2016; Jiang et al., 2020) 2Following common astronomical usage, all elements heavier than He are referred to as metals.

25 z=7

JWST ALMA (band 4-10) 2 HST [OIII] 88 m

VLT & Keck Spitzer/IRAC [CII] 158 m 1

[OIII] 4959, 5007 Å 0 H H [OII] 3727 Å

-1

log(flux/mJy) Rest-frame UV Continuum Rest-frame FIR dust-continuum -2

-3 104 105 106 107 Wavelength(Å) Figure 4.1. Synthetic spectrum from rest-frame UV to FIR wavelengths of a high- redshift galaxy (z=7). Some spectral features that are discussed in the text (such as the UV slope and Balmer emission lines) are indicated in the ﬁgure. Also indicated are the wavelength coverages of the upcoming JWST and a selection of facilities/instruments that have been extensively used to study high-redshift galaxies. The wavelength axis is shown in the observer frame, while wavelengths of individual lines are given in the rest-frame. The spectrum was generated using the BAGPIPES code (Carnall et al., 2018). effects of scattering and absorption. An incomplete understanding of the dust properties of high-redshift galaxies may make other observations and their interpretations prone to errors. Since dust extinguishes light more effectively at short wavelengths, interstellar dust reddens the spectra of background stars in galaxies. The strength of this effect at different wavelengths is commonly described by a so-called extinction or attenuation law. Often, the terms attenuation law and extinction law are used interchangeably. However, there is a fundamental difference in that an extinction law technically only considers absorption and scattering of light out of the line of sight, while attenuation considers absorption and scattering of light both out of and into the line of sight. Examples of extinction laws are the Milky way, Large magellanic cloud (LMC) and Small magellanic cloud (SMC) laws (e.g. Pei, 1992), which differ due to the grain size and com- position of the dust. On the other hand, the Calzetti law (Calzetti et al., 2000) is an attenuation law derived using observations of nearby starburst galaxies. The reddening will affect not only the continuum of observed galaxies but also line emission and the ratios of lines at different wavelengths. Since the relative ratios of certain emission lines are well constrained from theory (for example, the Balmer lines of hydrogen), the observed line ratios can be used to correct the spectrum of an observed galaxy for dust. In the absence of lines

26 that can be used for this correction, a feature which can also be used to constrain the dust content in star-forming galaxies is the spectral slope of the UV continuum (see Rest-frame UV continuum in figure 4.1). Since the intrinsic UV slope (in the absence of dust-reddening) mainly depends on the nature of the most massive and hot stars, it is also sensitive to the SFH and metallicity (and to some degree the escape fraction of ionizing photons; Wilkins et al., 2016). Translating measured UV slopes to dust attenuation thus requires some assumption regarding the intrinsic UV slope, either from simulations and models (e.g. Bouwens et al., 2014; Wilkins et al., 2016), or from empirical relations (e.g. Meurer et al., 1999). β Usually, the UV slope is parametrized with a power law fλ ∝ λ with a slope β (e.g. Calzetti et al., 1994), and slopes where β < −2 are referred to as ‘blue’ while slopes with β > −2 are referred to as ‘red’. This is since a UV continuum with β = −2 is flat when expressed in units of Jy, and thus has zero color in the AB magnitude system. For galaxies at high redshift, the UV slope is typically determined through HST imaging, sometimes in combination with ground-based imaging, using, for example, the connection: m − m β = − 1 2 − 2 (4.1) . (λ 1/λ 2) 2 5log c c λ 1 λ 2 Where m1, m2 are the measured magnitudes, and c , c are the central wavelengths of the two imaging bands, respectively (Ono et al., 2010). It is important to note that this can lead to large uncertainties in the derived UV slopes, since if the available filters used to derive the slope are relatively close in wavelength, the short leverage means that small differences in the fluxes can lead to large differences in the UV slope. Observations of large samples of galaxies at z 4 have highlighted two important features in the evolution of the UV slope. Firstly, there seems to be a correlation between UV slope and luminosity, such that galaxies with low luminosities exhibit the bluest UV slopes (Rogers et al., 2014; Bouwens et al., 2014). This correlation is also seen in theoretical studies (e.g. Dayal & Ferrara, 2012; Mancini et al., 2016). Secondly, there seems to be an evolution toward bluer UV slopes at higher redshifts (e.g. Finkelstein et al., 2012b; Bouwens et al., 2014). This redshift evolution is, however, relatively mild, and even at z ∼ 10, observed UV slopes seem to be consistent with, at least, some amount of dust (Wilkins et al., 2016). Nevertheless, the redshift and luminosity evolution of the UV slope have been interpreted as an effect of chemical maturity, and that low luminosity and high-redshift objects contain less metals and dust, and thus represent less evolved populations. This is also reflected in several theoretical studies of the UV luminosity function at the highest redshifts, where the amount of dust required to make observations and theory fit is lower than in the low-redshift Universe (e.g. Hutter et al., 2014; Khakhaleva-Li & Gnedin, 2016). Note however, that some simulations find good agreement with UV luminosity functions while simultaneously pre-

27 dicting significant amounts of dust already at z ∼ 8, especially at high stellar masses (e.g. Wilkins et al., 2018). As discussed in section 3.1.3, a gradually lower dust content with increasing redshift has also been suggested as an explanation for the increasing Lyα emission between z ∼ 0 and z ∼ 6. As light is absorbed by dust at optical and UV wavelengths, the dust is heated. This heat is then radiated away in the infrared part of the spectrum. In addition to providing an excellent tool for spectroscopic redshift determination and probing ISM properties, ALMA has been used as a powerful tool to study this thermal emission in EoR galaxies. Many observations of the FIR dust emission of galaxies at z 6 have only led to upper limits (see e.g. Matthee et al., 2019, for a compilation). This is consistent with the blue UV slopes observed by HST and seems to indicate that high-redshift star-forming galaxies in general contain less dust than nearby star-forming galaxies, and have FIR/UV ratios more consistent with local metal-poor dwarf galaxies (e.g. Maiolino et al., 2015; Matthee et al., 2019). However, a number of objects with significant dust emission also have been detected, indicating some scatter in the dust content of EoR galaxies (e.g. Watson et al., 2015; Laporte et al., 2017; Bowler et al., 2018; Tamura et al., 2019; Hashimoto et al., 2019). The interpretation of these results in terms of actual dust masses depends on assumptions regarding the dust temperature (as well as dust grain properties) in these objects (e.g. Behrens et al., 2018) which, while there are studies indicating higher dust temperatures at high-redshifts (Bakx et al., 2020, e.g.), are still uncertain. In addition, FIR observations in combination with UV observations provide a way to constrain the shape of the dust reddening law which best represents high-redshift galaxies trough the relation between the infrared excess (IRX) and the UV slope β (IRX-β relation). The shape of the reddening law at inter- mediate and high redshifts is still strongly debated, with some studies claiming consistency between observations and steep SMC-like relation (or possibly, an even steeper relation; Smit et al., 2018; Bouwens et al., 2016). Other studies find better consistency with the Meurer et al. (1999, also commonly called the Calzetti IRX-β relation) relation for local starburst galaxies (Koprowski et al., 2018; Hashimoto et al., 2019). Using a large sample of galaxies at z ∼ 4 − 6 from the ALPINE ALMA survey, Fudamoto et al. (2020) derive dust laws at z ∼ 4.5 and z ∼ 5.5 which are steeper than the SMC law. The translation between FIR dust continuum observations to IR luminosities often, however, relies on assumptions both regarding dust temperatures and grain properties. It is important to note that the inferred IRX-β relations are therefore also sensitive to such assumptions (e.g. Behrens et al., 2018). An interesting complication is the distribution of dust in galaxies. While dust-corrections of observations are often done assuming a so-called dust- screen, i.e. that dust affects all stars equally, we expect dust to be located preferentially around young stars (Charlot & Fall, 2000). This can occur since the birth-clouds of stellar clusters are dense, and are eventually dispersed by

28 supernovae and stellar winds as stars evolve. The timescale of this effect is expected to be relatively short, and birth-clouds seem to generally disperse within ∼ 10 Myr (e.g. Charlot & Fall, 2000; Chevance et al., 2020) once massive stars start to form. Recently, Katz et al. (2019b) argued that differential obscuration, with high dust attenuation for stars with ages up to ∼ 50 Myr, may be a possible explanation for the large Balmer breaks observed in certain high-redshift objects.

4.2 Star formation As discussed briefly in chapter 2, the stars responsible for producing the bulk of the ionizing photons in a galaxy are the young and massive stars. These do not only give rise to the emission lines we observe in the optical, but also to the majority of the rest-frame UV flux, and therefore, these features can be used to estimate the recent star formation activity in galaxies. A commonly used and reliable way to estimate the SFR at lower redshifts is through the use of the Hα line. However, observing this line in galaxies during EoR is not possible at present due to the lack of observational facilities with spectroscopic capabilities and sufficient sensitivity at relevant wavelengths, something which the JWST will remedy in the near future. The Lyα line also traces SFR and is observable at high redshifts, however, as discussed in chapter 2 and section 3.1.3, the usage of Lyα as an SFR tracer is hampered not only by the complex transport of Lyα through the ISM (Hayes et al., 2010), but also due to absorption in the IGM. The most common way to infer star formation rates for high-redshift galaxies is through their rest-frame UV continuum around 1500 Å. This relies on some assumptions regarding the relative distribution of stars of different masses (the initial mass function; IMF) and the timescales of variations in the SFR, which are assumed to be long ( 50 – 100 Myr), (Kennicutt & Evans, 2012; Madau & Dickinson, 2014). Since the temperature of the stars, and thus the hardness of the spectrum is affected by the metallicity of the stars, the above-mentioned calibrations also have a slight dependence on metallicity. In addition, since UV photons are efficiently absorbed by dust, UV SFRs only trace the star formation which is not obscured by dust. In order to trace the obscured star formation, one can probe the IR SFR by, for example, observing the FIR dust emission of high-redshift galaxies using ALMA (see section 4.3 and paper IV). Using such measurements of the SFRs in galaxies over a wide redshift range, it has been possible to show that the comoving3 star formation rate

3Comoving distances factor out the cosmic expansion, and thus do not change with the expansion of space. This is in contrast to the proper distance, which will change as the Universe expands. Comoving distances are often denoted by a lowercase ‘c’ in front of the unit, for example cMpc−3.

29 − − density (SFRD; Myr 1cMpc 3) was increasing with time at high redshifts, peaked at around z ∼ 2 and declined at z 2 (Madau & Dickinson, 2014). Ad- ditionally, observations have shown that the specific star formation rate (star formation rate over stellar mass; sSFR, usually expressed in units of Gyr−1) increases with increasing redshift (Madau & Dickinson, 2014; Salmon et al., 2015). The emerging picture of these results is that the Universe was much more active earlier in its evolution. Another way to try to understand the star formation activity at high redshifts is through simulations. Generally, simulations suggest that the SFR in galaxies was increasing over time during the EoR (e.g. Jaacks et al., 2012; Shimizu et al., 2014; Kimm et al., 2015; Katz et al., 2019b). There is, however, a significant difference between the variations in the SFR over time in different simulations. Some simulations suggest that star formation rates were quite smoothly increasing over time (Finlator et al., 2011). Generally, galaxies with higher masses tend to experience smoother star formation histories (SFHs; the SFR as a function of time), owing to their deeper gravitational potential wells. This means that feedback, such as stellar winds or supernova explosions will be less effective at dispersing and removing the gas available for star- formation. Furthermore, as massive galaxies are likely to contain a larger number of regions which are able to host star formation, galaxy-wide SFRs smooth out and become less stochastic in these objects (Hopkins et al., 2014; Kimm et al., 2015; Yajima et al., 2017). Many simulations produce large variations in the SFHs of EoR galaxies, especially in low mass galaxies (e.g. Trebitsch et al., 2017), but also in galaxies with stellar masses up to a couple of 108 M (Kimm & Cen, 2014; Kimm et al., 2015; Ma et al., 2015; Yajima et al., 2017; Ma et al., 2018). In paper II we showed that such fluctuations can lead to severe mis-interpretations of the spectral energy distributions (SEDs) of EoR galaxies if ignored. On the observational side, recent observations of the z = 9.1096 galaxy MACS1149-JD1 indicate a strong Balmer break in the spectrum of this galaxy (Hashimoto et al., 2018). Since this break is strongest in A type stars, it becomes more prominent as the more massive O and B type stars evolve off the main sequence and die. It is therefore associated with an aged population of stars (Wiklind et al., 2008). An old stellar population would not be unusual if observed in the low-redshift universe, but at z ∼ 9.1 it could have interesting implications for when the first stars formed in these types of objects. Hashimoto et al. (2018) suggest that the strong break is due to large variations in the star formation activity, and that MACS1149-JD1 has undergone a long passive period before a second burst of star formation occurred. In paper III, we showed that several different simulations are unable to reproduce a Balmer break of the size observed in MACS1149-JD1, even when accounting for different assumptions on the dust reddening, escape fraction or IMF. Similar features have been observed in other non-spectroscopically confirmed high-redshift galaxies. While these are more uncertain, this could indicate

30 that SFHs were signiﬁcantly more bursty than found in current simulations. However, Katz et al. (2019b) have argued that dense ISM conditions, with dust preferentially located around younger stars could also explain the strong Balmer break in MACS1149-JD1 without requiring large variations in the star formation activity. In addition, recent results by Stefanon et al. (2021) seem to indicate the Balmer breaks such as the one observed in MACS1149-JD1 are not very common.

4.3 Far-infrared observations and the interstellar medium In the high-redshift universe, optical lines commonly used to determine spectroscopic redshifts and derive galaxy properties, such as Hα, shift into wavelengths unavailable to current observational facilities with spectroscopic capabilities. Nonetheless, at the other end of the spectrum, ALMA has provided the high-redshift community with a way to probe EoR galaxies through observations of rest-frame FIR emission lines and dust-emission, since these redshift into sub-millimeter/millimeter wavelengths. As shown in figure 4.1, there are several emission lines that fall within the wavelength span of ALMA. To date, the two most commonly targeted lines are the [C II] 158 μm and [O III]88μm lines. Due to the difference in ionization energies of the species responsible for the formation of these lines, they arise in different parts of the ISM. While part of the [C II] emission is produced in H II regions, it is mainly generated in photo-dissociation regions (PDRs). PDRs are regions of gas in the ISM in which the chemical structure is strongly influenced by UV photons with energies lower than that required to ionize hydrogen. In these regions, the UV flux is sufficiently hard to photo-dissociate molecules and photo-ionize atomic species with ionization potentials lower than that of hydrogen, but not sufficiently hard to ionize hydrogen. Thus, these are regions where hydrogen is mostly neutral or in molecular form. The [O III] line, on the other hand, origi- nates in highly ionized gas in H II regions (e.g. Abel et al., 2005; Nagao et al., 2011; Vallini et al., 2015). The low energy required to excite the states responsible for the lines means that they are easily excited through collisions, and therefore provide important cooling in their respective environments. Their forbidden and long-lived nature, however, also means that they are sensitive to collisional de-excitation, making the lines sensitive to the density of the ISM (Draine, 2011). Many early attempts to detect the [C II] line from high-redshift galaxies led to non-detections (Kanekar et al., 2013; Ouchi et al., 2013; Ota et al., 2014; Maiolino et al., 2015; Schaerer et al., 2015). Several of the upper limits derived for the emission line from these objects placed them below the [C II]- SFR relation derived for local galaxies by De Looze et al. (2014). Since then,

31 several [C II] detections have been made in EoR galaxies, although a number of studies have continued to report non-detections (see compilations by e.g. Matthee et al., 2019; Harikane et al., 2020). While the [O III] line has been targeted in fewer high-redshift galaxies, it has been shown to be a reliable line for spectroscopic redshift confirmation, with a high detection rate all the way up to z ∼ 9.1 (Inoue et al., 2016; Laporte et al., 2017; Carniani et al., 2017; Hashimoto et al., 2018; Tamura et al., 2019; Hashimoto et al., 2019; Harikane et al., 2020). Over the years, the fact that a part of the high-redshift population exhibits relatively weak [C II] emission has led to a discussion on whether high-redshift galaxies follow a different [C II]-SFR relation than local galaxies. In a recent study of a compilation of galaxies at z ∼ 6 – 9, Harikane et al. (2020) derived a [C II]-SFR relation for high-redshift galaxies that is significantly steeper than the local relation by De Looze et al. (2014, see figure 4.2). On the other hand, other studies have found that if one accounts for galaxy sub-components and proper association of [C II] and UV components, high- redshift galaxies follow a [C II]-SFR relation which is consistent with the local one, albeit with significantly larger scatter (Carniani et al., 2018). Using a similar compilation as the one by Harikane et al. (2020), Matthee et al. (2019) argued that by using a consistent way to derive SFRs and [C II] upper limits, one can get better agreement between high-redshift galaxies and the local relation at high SFRs. However, they also found a possible deviation at low SFRs. Furthermore, using a large sample of z ∼ 4 − 6 galaxies from the recent ALPINE ALMA survey in combination with the compilation of Matthee et al. (2019) and other high-redshift observations, Schaerer et al. (2020) found a relation with only a slightly steeper slope and lower normalization than the local relation (see figure 4.2). Similar results are found in a recent study by Carniani et al. (2020), where the authors have re-analyzed several high-redshift [C II] datasets while compensating for surface brightness dimming due to extended [C II] emission. Carniani et al. (2020), however, have also highlighted that the dispersion in the high-redshift relation is significantly higher than that observed in local H II and Starburst galaxies, which indicates a broader range in ISM properties. There are still uncertainties in the derived relations, partly due to the number of [C II] undetected galaxies at high redshifts (especially with low SFRs), and partly due to assumptions used to derive SFRs in these objects, both from the UV and IR. A clear consensus on the [C II]-SFR relation at high redshifts has yet to be reached. Since the abundance of C+ in the ISM scales with the gas metallicity, the weak [C II] seen in some high-redhift galaxies has been argued be an effect of a low metallicity (e.g. Pentericci et al., 2016; Knudsen et al., 2016; Bradacˇ et al., 2017; Matthee et al., 2017; Harikane et al., 2018). Using cosmological simulations coupled with a UV radiative transfer code and a PDR modeling code, Vallini et al. (2015) showed that a low gas metallicity could, in princi-

32 (L ) [C II] L

-1 SFRUV+IR (M yr ) Figure 4.2. The [C II] 158 μm luminosity plotted versus the (UV+IR) SFR. The dark circles show z > 6 literature sample used in paper IV and the new upper-limit reported there. The local relation by De Looze et al. (2014) and its dispersion is shown by the yellow line and shaded region. The dashed and dash-dotted line shows the relations from the ALPINE ALMA survey, derived using 3σ and 6σ upper limits for non- detections, respectively. The dotted line shows the relation by Harikane et al. (2020).

ple, explain the low [C II] observed in some high-redshift galaxies. They also highlight that strong stellar feedback could lead to weak [C II] through the destruction of molecular clouds. However, using models calculated with the CLOUDY photoioniztion code, Harikane et al. (2020) in a later study argued that the gas metallicity does not significantly affect the [C II] luminosity. Even so, Harikane et al. (2020) highlight a secondary effect of a lower metallicity (specifically the stellar metallicity), which is a harder radiation field. This harder radiation may penetrate further into, and ionize a larger part of, the ISM, and therefore also regulate the [C II] luminosity. Such a scenario could lead to lower [C II] line strengths through the destruction of PDRs and growth of the highly ionized regions in which the [O III]88μm emission line forms, and thus to stronger [O III] emission and therefore, high [O III]-to-[C II] ratios. Of course, the [O III] line is also sensitive to the gas metallicity. Inoue et al. (2014) have argued that their models best reproduce observations when the [O III] luminosity increases with increasing metallicity at Z 0.2Z due to an increased abundance of oxygen in the ISM, but turns over, and starts decreasing with increasing metallicity at Z 0.2Z, possibly due to the change in the hardness of the stellar radiation field. Thus, the [C II] and [O III] lines are likely to some degree regulated by the combined effect of stellar and gas metallicity.

33 The scenario of a hard radiation field driven by low stellar metallicity (or a young stellar population, see below) would seem consistent with a number of recent studies that have found high [O III]-to-[C II] ratios in high-redshift objects (Inoue et al., 2016; Hashimoto et al., 2019; Harikane et al., 2020; Bakx et al., 2020). Recently, Harikane et al. (2020) found that the [O III]-to-[C II] ratios at high redshifts are systematically higher than those of local galaxies. Recent results from Carniani et al. (2020), who reported marginal detections in several previously [C II] un-detected objects by considering more extended [C II] emission, point toward [O III]-to-[C II] ratios that are still high, but closer to those observed in local galaxies. Alternatively, a low covering fraction of PDRs due to feedback could lead to low [C II] and high [O III]-to-[C II] ratios (Harikane et al., 2020). Both the scenario of a hard radiation field and lower PDR covering fractions could, in principle, facilitate the escape of LyC, since they would lead to lower covering fractions of neutral gas (Inoue et al., 2016; Harikane et al., 2020). Overall, both of these scenarios are also consistent with the apparent anti-correlation between [C II] and Lyα observed by Carniani et al. (2018); Harikane et al. (2018); Matthee et al. (2019); Harikane et al. (2020), since a lower neutral covering fraction also would facilitate the escape of Lyα. Furthermore, the Lyα line becomes stronger with the harder radiation produced by low-metallicity stars (Raiter et al., 2010). On the other hand, Schaerer et al. (2020) are not able to reproduce the [C II]-Lyα anti-correlation previously mentioned and argued that this is due to their larger dataset and a more consistent way to derive SFRs. It is important to highlight that there are still other mechanisms that could affect these lines. For example, a low C/O abundance ratio could also lead to lower [C II] emission and higher [O III]-to-[C II] ratio (Harikane et al., 2018, 2020; Arata et al., 2020). Since oxygen is formed mainly in core-collapse supernovae, while carbon has a contribution from type Ia supernovae as well as AGB stars, carbon forms on slower time-scales (Maiolino & Mannucci, 2019). In young and un-enriched stellar populations, the C/O abundance ratio is expected to be lower, and [O III]-to-[C II] ratios are expected to be higher (Harikane et al., 2020; Arata et al., 2020). The density of the ISM also affects how far ionizing photons will penetrate, and how often collisions that can potentially de-excite [C II] and [O III] occur, and will also affect the line strengths and ratio (Ferrara et al., 2019; Harikane et al., 2020). Since the ratio of these lines is sensitive to the ionizing flux from the nearby stars and possible feedback, the lines are also affected by a bursty star-formation history. High [O III]-to-[C II] line ratios and [C II] deficiencies would, for example, be expected for objects that are currently undergoing a burst, and have large populations of young stars (Pallottini et al., 2019; Katz et al., 2019a; Ferrara et al., 2019). Furthermore, the [C II] line forms in relatively cool regions, which means that heating from the CMBR and decreased contrast compared to the

34 surrounding CMBR can affect the line strength, though likely not by a large amount (Laporte et al., 2019; Harikane et al., 2020). As mentioned both in section 4.1 and in the above, FIR observations of the dust continuum can be used to probe the dust content of high-redshift galaxies. The FIR continuum traces the thermal re-emission from dust, and although several wavelength points, preferably straddling the peak of the thermal emission are required to constrain the shape of the thermal dust emission, even single observations can be used to get an understanding of the dust properties. Often, the interpretation of FIR continuum observations are done by assuming that the dust spectrum in the FIR to submillimeter is well represented by a modiﬁed black-body (also called a ‘grey body’) of the shape:

β Sν ∝ ν Bν (T) (4.2) Where ν is the frequency, β is a spectral index, the so-called dust-emissivity index (which depends on the dust grain properties) and Bν is the Planck function (e.g. De Breuck et al., 2003; Casey, 2012). The translation between observed FIR dust continuum and the total dust luminosity is thus done by modeling the dust emission as such a modified black-body curve with a specific temperature and dust-emissivity index. This curve can then be integrated in order to get the total dust luminosity. The larger the number of observed wavelengths, the larger the number of constraints on the shape and thus the temperature and β. In combination with the UV SED, the IR luminosity can be used to estimate how much light is absorbed (and thus heats) the dust. In- ferring dust masses from such a relation requires an additional assumption regarding the dust mass absorption coefficient. This coefficient also depends on the dust grain properties and uncertainties in the value of this coefficient can lead to a large difference in the estimated dust mass (e.g. Hashimoto et al., 2019; Inoue et al., 2020). In paper IV, we present observations of a z ∼ 7.7 galaxy that seems to exhibit weak [C II] and weak [O III]. While the nature of the object cannot be understood with the limited information currently available, one possible explanation for the weak [C II], if the model by Vallini et al. (2015) is indeed representative of these kinds of objects, could be a low metallicity (Z 0.1Z). By comparing our [O III] observations to local galaxies and by modeling the SED of the object, we showed that a low metallicity is consistent with the non-detection in [O III] and dust continuum as well.

35 5. The Lyman continuum escape fraction

As discussed in chapter 1 and 3, our current understanding of the cosmic reionization is that it relies on the production and escape of LyC photons from high-redshift star-forming galaxies. Since stars form in environments where gas densities are enhanced, any LyC photons produced are likely to be caught in the ISM surrounding the stars from which they are emitted. However, in order for star-forming galaxies to drive the reionization, some fraction of the LyC photons produced must make it into the IGM. As described in chapter 1, this fraction is known as the LyC escape fraction (often simply called the ‘escape fraction’; fesc). Escape of LyC photons can occur if for some reason, the density of neutral gas around the stars is low. The definition of the LyC escape fraction differs slightly between studies. From an observational standpoint, the escape fraction is defined as the observed number of ionizing photons over the intrinsic number of ionizing photons produced by the stellar population. Since the observed escape fraction depends on the observed flux of LyC photons, we do not measure the global escape fraction, but rather the fraction of LyC photons emitted toward us over the total number of produced LyC photons. The observed escape fraction can therefore differ from the actual global escape fraction. The term ‘escape fraction’ is often used both for observed and global escape fractions, which is also how it will be used throughout this thesis. It is, however, important to be aware of the distinction when we discuss simulated escape fractions, which are often global, and observed escape fractions, or escape fractions derived from line-of-sight diagnostics, which are not. In general, the observed escape fraction is calculated using a combination of the observed LyC flux and observed quantities that can be related to the intrinsic production rate of LyC photons via models. Since the LyC photons that get absorbed in the ISM of a galaxy will lead to the production of nebular recombination lines, models including these lines can be used to infer the number of photons that do not escape the ISM. In the local universe the Hα line can, for example, be used for this purpose. For objects where suitable lines are unavailable (for example due to the redshift) one can use the ratio of the observed Lyman break amplitude over the intrinsic Lyman break amplitude calculated from SED models to infer the escape fraction (Siana et al., 2007). The observed Lyman break amplitude and line fluxes also have to be corrected for dust effects in order to correctly calculate the escape fraction. For both of the discussed approaches, an additional correction also has to be made

36 for LyC absorption in the IGM and within the Milky Way. For local galaxies, the effect of absorption of LyC in the IGM is assumed to be negligible, since the probability to find neutral gas along the line of sight is small. For galaxies further away, however, remaining neutral hydrogen in the IGM may have a significant impact on the observed LyC flux. The optical depth of LyC photons in the Milky Way is, however, more important at very low redshift, since the Lyman limit of neutral gas in the Milky Way will be located close to the Lyman limit of the object that is being observed (e.g. Leitet et al., 2013). An additional complication at higher redshifts is the possible risk of non-ionizing emission from low-redshift interlopers which could be erroneously interpreted as LyC at higher redshifts. This can be circumvented using observations with sufficient angular resolution so as to rule out this effect (e.g. Vanzella et al., 2012).

5.1 Mechanisms of Lyman continuum leakage Leakage of LyC requires a sufficiently low neutral hydrogen column density in the ISM for a fraction of the photons to make it into the IGM without being absorbed. LyC leakage from galaxies can be divided into two main scenarios; the ionization-bounded nebula with holes, also known as the picket-fence model, and a density-bounded nebula, also known as the truncated Strömgren sphere (see figure 5.1). The Strömgren sphere (Strömgren, 1939) is the sphere around an ionizing source in which the photo-ionization rate and recombination rate is balanced. In other words, it is the sphere of gas that a given source of ionizing photons is able to ionize. In the case of an ionization-bounded nebula (shown in the left panel in figure 5.1), the photo-ionization rate is lower than what is required to completely ionize the gas surrounding the ionizing source, and thus the Strömgren sphere remains embedded inside a shell of neutral gas. For LyC leakage to occur in this scenario, there have to be channels where the neutral gas has been cleared and through which LyC photons can escape. If, however, the amount of gas around the stars is small compared to the ionizing flux, and recombination rate is lower than the photo-ionization rate, the result is a density-bounded nebula. Since all the neutral gas is ionized in this case, LyC photons are able to escape since the gas in their path is ionized. The ISM of real galaxies is, of course, more complex than is captured by either one of these two simple models. Off-center star formation, large density fluctuations or runaway stars can, for example, lead to more complicated leakage scenarios. Additional complications arise from the possibility that substantial amounts of LyC can be absorbed in dust (e.g. Inoue et al., 2001; Inoue, 2001).

37 HI REGION HII REGION STRÖMGREN RADIUS

LEAKING LYC

IONIZATION-BOUNDED DENSITY-BOUNDED NEBULA NEBULA

Figure 5.1. Schematic ﬁgure illustrating the two leakage scenarios: ionization- bounded nebula (left) and a density-bounded left (right). The scenarios show a central cluster stars embedded in gas. The ionized gas (H II region) is shown in blue, and the neutral gas (H I region) is shown in gray. In the density-bounded scenario, all of the hydrogen has been ionized by the stars, and the theoretical Strömgren sphere (marked by the dashed circle) has been truncated. In the ionization-bounded scenario leakage of LyC occurs in channels that have been cleared of gas. In this scenario, there is more than enough hydrogen to form a complete Strömgren sphere and neutral hydrogen remains outside the Strömgren radius.

5.2 Observations of leaking Lyman continuum Despite the importance of the reionization epoch and of understanding the processes behind LyC leakage, observations of leaking LyC from low-redshift and local galaxies were, until quite recently, few. Only fairly recently have the number of detections of leaking LyC increased drastically. There are now a substantial number of galaxies at z < 1 for which the escape fraction has been measured. Typically, the values found for these objects are fesc 10% (e.g. Bergvall et al., 2006; Leitet et al., 2013; Borthakur et al., 2014; Izotov et al., 2016a; Leitherer et al., 2016; Izotov et al., 2016b). There are, however, also some studies that report fesc > 40% (Izotov et al., 2018a,b). Interestingly, at higher redshifts (z ∼ 1 – 4) many individual LyC-detected objects exhibit relatively high escape fractions ( fesc 10%) compared to local galaxies (Mostardi et al., 2015; Vanzella et al., 2016; Shapley et al., 2016; Bian et al., 2017; Naidu et al., 2017; Vanzella et al., 2018; Steidel et al., 2018; Fletcher et al., 2019). At ﬁrst glance, this could be taken as an indication that galaxies in this redshift range have high enough escape fractions to explain galaxy driven reionization if we assume that these values are valid at

38 even higher redshifts. However, studies of galaxies at z ∼ 1 – 4 that use stack- ing methods to improve signal-to-noise ratios have primarily resulted in upper limits, and generally indicate average escape fractions fesc < 10% (e.g. Siana et al., 2007, 2010; Boutsia et al., 2011; Grazian et al., 2016, 2017; Japelj et al., 2017; Micheva et al., 2017; Matthee et al., 2017; Rutkowski et al., 2017; Naidu et al., 2018). This places these objects on the low end side of what is required for reionization. An exception to this is the study by Steidel et al. (2018), in which the authors found an average escape fraction of 9 ± 1% for galaxies at z ∼ 3. This result seems to hint at typical escape fractions that are more consistent with the values required for reionization.

5.3 Indirect methods for constraining the escape fraction Even though reionization was largely finished around z ∼ 6, there remained neutral hydrogen in the IGM at redshifts much lower than this. The effect of residual neutral hydrogen in the lines-of-sight toward objects increases at higher redshift, as discussed in section 5 and section 3.1.1. At z ∼ 4 this effect is already so large that it is difficult to find a line-of-sight through which leaking LyC can be observed (Inoue & Iwata, 2008). We must therefore turn to methods which allow us to estimate the escape fractions without directly observing any leaking LyC in order to constrain escape fractions at these redshifts. In the following, a few methods for indirect determination of the escape fraction will be discussed.

5.3.1 Emission-line constraints on the escape fraction A method which has shown great success in ﬁnding nearby LyC leaking galaxies is selecting by high O32 ratios ([O III] 5007,4959 versus [O II] 3727). This method, suggested by Jaskot & Oey (2013), could in principle be used with a different combination of lines, but the aforementioned oxygen lines have been most extensively used. Since the ionization energy of neutral oxygen is very close to that of hydrogen, the [O III] and [O II] lines lines are formed in regions where hydrogen is ionized. In the density-bounded scenario, the out- ermost parts of the nebula from which the [O II] line is expected to originate, is truncated, and so the ratio of [O III]to[OII] becomes enhanced. Stasinska´ et al. (2015) showed that there are, however, other physical mechanisms that may lead to a similar effect on the O32 line ratio, and argued against using only high O32 as indicators of LyC leakage. High O32 ratios have been used to ﬁnd local LyC leakers with a great success (Izotov et al., 2016a,b; Izotov et al., 2018a,b). There are, however, observational and theoretical studies that have showed that high O32 ratios can be found in objects with low or no leakage of LyC (Naidu et al., 2018; Nakajima

39 et al., 2020; Barrow et al., 2020). In summary, these results indicate that a high O32 ratio may be a necessary, but not sufficient, condition for LyC leakage. Studying the shape of the Lyα profile has also been shown to be a viable way to constrain the escape fraction (Verhamme et al., 2015, 2017). Since paths that are optically thin to Lyα photons are also optically thin to LyC photons, the Lyα line strength is expected to correlate with the LyC escape fraction (e.g. Steidel et al., 2018). Using numerical simulations of Lyα radiative transfer, Verhamme et al. (2015, 2017) showed that scenarios in which LyC leakage is occurring should leave their imprint on the observed Lyα profiles, changing their shapes. The observed shapes depend on the leakage scenario and amount of leakage. However, since Lyα can be absorbed in the IGM as it travels toward us, the line shape may also be changed by effects not related to the physics in the host galaxy. This complication is likely to be more severe in the EoR where the IGM remains neutral to a larger degree than in the low redshift Universe (Verhamme et al., 2017). An interesting additional complication relates to the possible bias arising from only estimating the escape fraction of ionizing photons of galaxies that are observed in Lyα, which may not be representative of typical galaxies at high redshifts (e.g. Stark et al., 2015). Recently Katz et al. (2020) used their simulations of high-redshift galaxies with simulated LyC escape and an advanced post-processing method for nebular emission to study possible indicators of LyC leakage. Interestingly, they found that the galaxies that have high escape fractions often exhibit large O32 ratios, as suggested by Jaskot & Oey (2013). With the help of machine- learning methods, they showed that they were able to find LyC leakers using relatively few emission lines. They found that high [O III]88μmto[CII] 158μm line ratios can potentially be used to identify LyC leakers ( fesc > 10%). The accuracy of their method increases when they include a larger number of emission lines from several different species. Mas-Ribas et al. (2017) suggested using small-scale intensity mapping to detect galaxies leaking LyC. This method uses the fact that escaping ionizing radiation could lead to extended Lyα and Hα emission in the circum- galactic medium of star-forming galaxies. Thus, extended Hα or Lyα halos could be indicative of leaking LyC. An advantage of this method would be that one could use photometric narrow-band observations (the authors suggest the JWST/NIRCam F470N filter) to estimate the escape fraction. This does, however, limit the redshift range for which the method is applicable. Further- more, faint satellite galaxies around the target objects may change the surface brightness distribution and lead to misinterpretations. An interesting method which is not purely emission-line based, but relies on detailed modeling of the UV through NIR SED of galaxies in which stellar populations are resolved has been suggested by Choi et al. (2020). While this method is an interesting method for probing nearby galaxies, the spatial resolution required makes it unsuitable for high-redshift targets.

40 The method suggested by Zackrisson et al. (2013) also relies on measuring emission lines. Since this method is the one used in paper I and paper II, a more in-depth description is presented in section 5.3.4.

5.3.2 Absorption-line constraints on the escape fraction Jones et al. (2013) and Leethochawalit et al. (2016) have suggested a method for indirectly determining the escape fraction of galaxies using interstellar UV absorption lines of low ionization species that trace the neutral hydrogen (for example C+,Si+ and O). If the column density is low, this should leave an imprint in the form of residual ﬂux in the UV absorption lines of the aforementioned low-ionization species. This can then be used to estimate the escape fraction toward the observer. This method was tested with promising results by Chisholm et al. (2018); Gazagnes et al. (2018), although there are also studies that have found less convincing results (Vasei et al., 2016). One major downside of this method is that it requires relatively high spectral resolution and signal-to-noise levels in order to accurately probe absorption lines, which makes it unsuitable for studying faint objects in the EoR. Lyα absorption features in GRB afterglows have also been used to indirectly determine LyC escape fractions in GRB hosts. As the light from a GRB afterglow travels through the host ISM, intervening neutral hydrogen will lead to the absorption of photons at the wavelength of Lyα, making it appear in absorption. Using the shape of the absorption line, one can infer the neutral column density toward the GRB, which can then be translated to an escape fraction of the host galaxy (e.g. Totani et al., 2006; Chen et al., 2007; Tanvir et al., 2019; Vielfaure et al., 2020). However, in the EoR, the shape of the Lyα absorption line (absorption wing) also depends on the neutral fraction of the IGM (section 3.1.1). With accurate redshift determinations and models combining a neutral IGM and neutral gas in the host galaxy, one can disentangle these two cases (Patel et al., 2010). Using a sample of 140 GRBs at 1.6 < z < 6.7, Tanvir et al. (2019) derive an average escape fraction of only 0.5% for the GRB hosts. The current number of GRBs at z > 6 is, however, still low. As is the case with several of the methods mentioned in this section, this is something that the JWST may solve in the future. Note that the derived escape fraction values from the absorption-line methods discussed in this section are line-of-sight escape fractions rather than global escape fractions.

5.3.3 Escape fractions from ionized proximity-zones and bubbles Background high-redshift quasars may also be used to infer the escape fraction of LyC photons from galaxies in the foreground (Kakiichi et al., 2018). As discussed in section 3.1.1, the increasingly neutral IGM at EoR leads to

41 complete absorption of wavelengths shortward of the Lyman break. If the density of neutral hydrogen is lower in some foreground region, this may lead to transmission spikes in the Gunn-Peterson trough. Kakiichi et al. (2018) used the cross-correlation of the physical positions of foreground LBGs and such transmission spikes to find the mean Lyα transmitted flux at different distances from the foreground galaxies. By assuming that the correlation between the transmitted Lyα flux and physical distance is caused by ionization of the gas by the foreground LBGs and other undetected objects, Kakiichi et al. (2018) argued that one can infer the average escape fraction of ionizing photons of these objects. This is done by balancing the number of ionizing photons required to produce the observed correlation and the ones produced by the observed galaxies and possible fainter undetected objects. This requires assumptions regarding both the number of undetected fainter sources that may contribute to the total ionizing flux, and the number of ionizing photons per UV luminosity of the LBGs and the fainter sources. Applying this method to the quasar field J1148+5251, Kakiichi et al. (2018) inferred an average escape fraction of foreground galaxies of ≥ 8% at z = 6. One advantage of this method is that it allows for the determination of the average escape fraction from many different types of sources, albeit only in a single quasar sightline. A similar method has been suggested by Zackrisson et al. (2020). Instead of using transmission spikes to probe regions of lower neutral hydrogen density, Zackrisson et al. (2020) suggested that one could use the upcoming SKA to map large ionized bubbles around clusters of galaxies. Ignoring the effects of recombination within the bubbles, the size of the bubbles should be directly connected to the total number ionizing photons emitted within them. By combining the ‘bubble maps’ with rest-frame UV observations of the galaxies that are thought to lie inside the bubbles, one should be able to infer the mean escape fraction of the galaxies. As is the case with the Kakiichi et al. (2018) method, this requires some estimate of the number of produced ionizing photons per UV luminosity. This method is complicated by the possibility that undetected sources contribute a substantial fraction of the ionizing photons. An estimate of the number of undetected sources does, however, require some understanding of possible environmental biases in such large ionized bubbles.

5.3.4 Zackrisson et al. 2013: Nebular emission features In paper I and paper II we discuss a method for indirectly constraining the escape fraction of ionizing photons from EoR galaxies suggested by Zackris- son et al. (2013). This method is tailored for use with the JWST/NIRSpec spectrograph, which covers a wavelength range of approximately 0.6 to 5 μm (corresponding to 750 to 6250 Å in the rest-frame at z ∼ 7), and relies on the fact that LyC photons absorbed in the ISM are reprocessed to nebular emis-

42 sion. This means that the number of LyC photons that are absorbed inside the galaxy, and therefore the number of photons that escape into the IGM, leaves an imprint at longer wavelengths through nebular emission lines or nebular continuum emission. Assuming a dust-free ISM with a typical temperature, the strengths of lines such as Hα and Hβ should reflect the number of ionizing photons that are absorbed by the gas in the ISM. Of course, the strengths of these lines are therefore also affected by the LyC production efficiency. By simultaneously measuring the shape of the UV continuum, Zackrisson et al. (2013) argued that one should be able to gauge the production rate of LyC photons, since the slope of the UV continuum is expected to be blue for young stellar populations, and thus, to some degree traces LyC production efficiency. A similar method, using a combination of rest-frame UV observations and ALMA observations of the [O III]88μm emission line, has been used to estimate the escape fraction in a z ∼ 7 galaxy by Inoue et al. (2016).

Figure 5.2. UV/optical synthetic spectrum of a constant SFR model with an age of 10 Myr, a mass of 109 M and a metallicity of Z = 0.020. The impact of different escape fractions are highlighted by different color lines: 0.0 (red), 0.3 (yellow), 0.5 (green) and 0.7 (blue). Figure from Zackrisson et al. (2013).

Figure 5.2 shows the rest-frame UV/optical SED of an ionization-bounded, constant-SFR model with different escape fraction. This illustrates the effect that the escape fraction has on the nebular emission from a simple model galaxy. Zackrisson et al. (2013) suggested that the equivalent width (EW) of the Hβ line and the UV slope could be used to ﬁnd galaxies with high LyC escape fractions, and introduce the EW(Hβ) - β diagram shown in ﬁgure 5.3. This diagram shows the evolution of a constant SFR model with different escape fractions. Since the models of different escape fractions do not overlap in

43 Figure 5.3. The EW(Hβ) - β diagram for constant SFR models with different escape fractions: 0.0 (red), 0.3 (yellow), 0.5 (green) and 0.7 (blue). The ages of the models are marked with ﬁlled circles at 1, 10, 100, 700 Myr, from bottom left to top right. Solid lines indicate ionization-bounded models, and the dashed lines indicate density- bounded models. All models have a constant metallicity of Z = 0.020. Figure from Zackrisson et al. (2013).

this diagram, it should theoretically be possible to distinguish between objects with different escape fractions, albeit only at relatively high values. Variations in the star formation histories and internal metallicity distributions are likely to introduce a scatter into the EW(Hβ) - β diagram, and thus complicate the escape fraction estimates using this method. Zackrisson et al. (2013) argued one should be able to alleviate these complications by cali- brating the models using other spectral features available in JWST/NIRSpec observations. However, they showed that effects of dust can lead to additional complications. They argued that in the case of relatively simple dust geome- tries, there is a possibility that one could correct for dust using higher-order Balmer lines. However, for cases where dust is mixed into the H II region, and where part of the LyC is absorbed by dust, degeneracies remain that cannot be solved using only the rest-frame UV/optical SED. While the discussion in Zackrisson et al. (2013) and paper I is focused on the use of the Hβ line in combination with the UV slope, a high escape fraction will lead to weaker line emission overall. Thus, the method may also be used with a combination of many spectral spectral lines. This was done in the studies by Jensen et al. (2016) and Giri et al. (2020), where simulated galaxies were combined with machine learning in order to identify features linked to the escape fraction. However, since lines of heavier species have a strong

44 dependence on additional parameters, there is a risk of misinterpretation if the method is applied without this in mind. In paper I, we apply the Zackrisson et al. (2013) method to realistic simulated galaxies with varying internal metallicity distributions and varying star formation histories. We also model the effects of dust and discuss limitations of the method. In paper II, we discuss problems that may arise if the simulations used in paper I underestimate ﬂuctuations in the star-formation activity of high-redshift galaxies, and apply a classiﬁcation algorithm to determine if we may be able to identify such problematic cases using, for example, a combination of JWST/NIRCam and JWST/MIRI observations.

45 6. Synthetic spectra of high-redshift galaxies

Interpreting observations of galaxies at any redshift requires the ability to create models that capture the physical components and processes at work in these objects. Since stars are not resolved for a majority of galaxies (especially in the high-redshift Universe), we generally study the integrated light, which includes imprints of the SFH, stellar initial mass function (IMF), internal metallicity distribution and ISM conditions. The problem we have to solve when trying to interpret our observations is assessing which model parameters produce the SED which bears most resemblance to the observed one. Models can also be used to make predictions for future observations by predicting the observable signatures of a given set of parameters. In this way, models and simulations can provide a powerful tool in constraining what future observations may reveal.

6.1 Population synthesis and model galaxy spectra The simplest type of stellar population is the instantaneous-burst population, also known as a single stellar population (SSP). In an SSP, all stars are created in the same formation episode, meaning that they all have the same age, metallicity and abundance pattern. Such populations are the basic building blocks for more complex stellar populations. While there are different ways to create synthetic spectra for SSPs, a common approach is to combine three components; a stellar evolution model, which is used to calculate stellar effective temperatures and surface gravities at different ages for a set of metallicites and masses; a stellar atmosphere model, which provides spectra for a range of effective temperatures and surface gravities; and an initial mass function (IMF), which determines the stellar mass distribution within the population. More complex populations (composite stellar populations; CSPs) with arbi- trary SFHs can be modeled by expanding these into series of SSPs, assuming some SFR as a function of time (Bruzual & Charlot, 2003; Conroy et al., 2009; Fioc & Rocca-Volmerange, 2019). Examples of commonly used stellar population synthesis models that use the above mentioned method are the Galaxev models (Bruzual & Charlot, 2003), STARBURST99 (Leitherer et al., 1999; Vázquez & Leitherer, 2005), FSPS (Conroy et al., 2009), PEGASE (Fioc & Rocca-Volmerange, 1997, 2019) and the BPASS models for binary stellar evolution (Eldridge & Stanway, 2009). Several other population synthesis codes and galaxy modeling codes are built

46 upon these population synthesis models, and have been expanded to include additional effects (e.g. Zackrisson et al., 2011; Carnall et al., 2018; Boquien et al., 2019; Mawatari et al., 2020b). Many population synthesis and galaxy modeling codes include effects of nebular emission. In some cases, this is restricted to only nebular continuum, or nebular line emission, but several codes include both of these. One way to include nebular emission in a galaxy model is to couple the stellar spectrum directly to the nebula using a photoionization code such as CLOUDY (Fer- land et al., 1998; Ferland et al., 2013) or MAPPINGS (e.g. Binette et al., 1985; Sutherland & Dopita, 1993; Allen et al., 2008; Sutherland & Dopita, 2017). An alternative method is to calculate the total ionizing flux (or the luminosity of hydrogen emission lines) and then use pre-calculated line templates for different metallicities and ionization parameters in order to calculate the luminosities of the emission lines of interest (e.g. Boquien et al., 2019; Mawatari et al., 2020a,b). Finally, any realistic model of a galaxy needs to consider the effect that dust has on the emerging SED. While realistic dust distributions are likely to lead to more complex scenarios, a common simplifying assumption is that dust reddening occurs in a foreground screen, and that the shape of the dust-reddening curve is fixed. Thus, only the normalization of the curve changes as the dust amount varies. By requiring energy balance between the dust absorption and thermal re-emission, the thermal dust emission spectrum at IR wavelengths can be predicted. This can be done in several ways. Two common ways are to either assume that that the emerging spectrum follows some analytical function (e.g. a modified black body at FIR wavelengths, and a power law in the mid-infrared), or by assuming some empirical template for the dust emission (e.g. Boquien et al., 2019; Mawatari et al., 2020b). Many population synthesis and galaxy modeling codes can be used either to generate model galaxy spectra for a set of input parameters or, given input observations, derive the best-fitting model parameters (SED-fitting). Modern codes that can be used for both of these purposes in wavelengths ranging from the rest-frame UV to the FIR are, for example, the BAGPIPES code (Carnall et al., 2018), CIGALE (Boquien et al., 2019) and PANHIT (Mawatari et al., 2020a,b). In paper IV, we use the PANHIT code to infer galaxy parameters from a combination of ground-based, HST and ALMA observations.

6.2 Simulated galaxies There is a large number of cosmological simulations aimed at modeling galaxies and their role in the evolution of the Universe at different scales. Such simulations can be divided into different groups depending on the underlying theoretical technique upon which they are based. These groups include semi- numerical simulations, semi-analytical simulations and hydrodynamic simu-

47 lations. The simulations used in papers I–III belong exclusively to the third group. In this section, I will briefly introduce these simulations and some key differences, but an in-depth description of the simulations and details about specific implementations is beyond the scope of this thesis. Generally, hydrodynamic simulations fall into one of two categories: La- grangian, to which smoothed particle hydrodynamics (SPH) simulations belong, and Eulerian, to which adaptive mesh refinement (AMR) simulations belong. In SPH simulations, the fluids are discretized in particles, and thermo- dynamic properties are obtained by smoothing over neighboring particles. In AMR, fluids are discretized in space (cells), which are refined in areas where higher resolution is required. Each of these methods has its own strengths and weaknesses that depend on the numerical approach (see e.g. Springel, 2010; Kuhlen et al., 2012; Hopkins, 2015). In paper I, we use simulated galaxies from four different cosmological simulation suites: The simulations by Shimizu et al. (2014, ; S14), the Cosmic Reionization On Computers (CROC; Gnedin & Kaurov, 2014; Gnedin, 2014) simulations, the First Billion Years (FiBY; Paardekooper et al., 2013) simulations and the simulations by Finlator et al. (2013, ; F13). In paper II, we use only the S14 simulations. In paper III, we use galaxies from an updated, higher-resolution version of the S14 simulations (Shimizu et al., 2016, ; S16), the Feedback In Realistic Environments simulations (FIRE-2; Ma et al., 2018, 2019) and galaxies from the FirstLight project (Ceverino et al., 2017, 2018, 2019). The F13, FiBY, S14 and S16 simulations are all SPH simulations, and are based on the N-Body/SPH code GADGET (Springel et al., 2001; Springel, 2005). CROC and FirstLight are based on an AMR code called adaptive refinement tree (ART; e.g. Kravtsov et al., 1997; Kravtsov, 1999; Kravtsov et al., 2002). The FIRE-2 simulation used in paper III is run using the GIZMO code (Hopkins, 2015). This code uses a method known as a Meshless Finite-mass (MFM), which is aimed at combining the strengths of Lagrangian and Eule- rian methods (see Hopkins, 2015). While the implementations differ between simulations, these all include physics for e.g. simulating star formation, stellar feedback and chemical enrichment. In all of the simulations, the resolution is insufficient to resolve individual stars. The minimum resolvable stellar unit, or ‘star particle’, is essentially a cluster of stars with the same metallicity and abundance pattern. This means that there is a mass limit below which star formation becomes stochastic due to the limited number of star particles that make up a galaxy. In order to avoid possible issues associated with this type of stochastic star formation, we assign a minimum stellar mass to some of the simulations. For the CROC, FiBY and F13 simulations, we require M ≥ 107M, while for the S14 simulations, we require M ≥ 5 × 108M. In the case of the simulations used in paper III, we only extract galaxies with masses M ≥ 108M. However, this limitation does not come from the finite resolution of the simulations, but rather because we

48 want galaxies that have comparable stellar masses to MACS1149-JD1, which has a predicted stellar mass of ∼ 109M. In both the FIRE-2 and FirstLight simulations, galaxies are selected from so-called ‘zoom-in’ regions. This means that a simulation containing only dark-matter particles is first run in a large box. Then, regions of interest are selected and re-simulated with a higher resolution in the actual simulation. While the initial box-sizes of these kinds of simulations can be very large, the simulated high-resolution region from which simulated galaxies are extracted is generally substantially smaller than lower-resolution ‘full-box’ simulations. For details about each individual simulation, see the aforementioned references in this section. The differences in, for example, the choice of box size lead to differences in the number and masses of galaxies available in each snapshot of the simulation. Similarly, the choice of implementation of physical processes also lead to differences in the properties of the simulated galaxies. For example, the simulations produce galaxies with different average metallicities and SFHs. These differences are useful since they allow us to test if results are generic rather than limited to a specific model. The difference in the SFHs produced by the different simulations is most pronounced when comparing the FIRE-2 galaxies (and, to some degree, the FirstLight galaxies) to the galaxies of the other simulations. Like the other simulations, FIRE-2 produces galaxies that experience increasing SFR over time during the EoR. However, the FIRE-2 galaxies experience significantly larger variations in the star formation activity than the galaxies in the other simulations. This makes them an interesting test-case for the large variations in SFR suggested for MACS1149-JD1. Large variations in the SFR are also seen in a number of other simulations (Kimm & Cen, 2014; Kimm et al., 2015; Yajima et al., 2017; Trebitsch et al., 2017). One important difference between the suites used in paper I is that the FiBY simulations provide simulated escape fractions for all the galaxies. This means that there is better consistency between galaxy properties and LyC leakage. This is not the case for the other simulations, where the different values of the escape fractions are assumed.

6.3 Spectra of simulated galaxies A straight-forward way to obtain realistic stellar spectra for simulated star- forming galaxies of the type discussed here is by associating each star-particle with an SSP spectrum of a given mass, age and metallicity. Then, in order to create the spectrum for the whole galaxy (not accounting for dust and gas), one simply has to sum up all the star particles. This is the method used for creating the spectra in papers I–III, although we also account for the nebular emission associated with each SSP. The SSP spectra used in papers I–III come from the population synthesis code YGGDRASIL (Zackrisson et al., 2011). YG-

49 GDRASIL is set up to handle input from several other population synthesis codes in order to generate SSPs and CSPs consisting of several different input populations. Throughout papers I–III, we have used different SSP spectra for population I and II stars from the STARBURST99 population synthesis models, generated using both Geneva and Padova-AGB evolutionary tracks (Leitherer et al., 1999; Vázquez & Leitherer, 2005). We also use SSP spectra calculated using the BPASS v.2.0 binary evolution models (Eldridge & Stanway, 2009; Stanway et al., 2016). The metallicity range used is Z = 0.001–0.040, Z = 0.0004 – 0.050 and Z = 0.001–0.030, for the Geneva, Padova-AGB and BPASS v.2.0 models, respectively. For papers I–II, we extend these models to lower metallicities (Z = 10−7 –10−5) using the Raiter et al. (2010) models for population III and extremely metal-poor stars. While it is possible to simulate different SFHs directly within YGGDRASIL by re-weighting the SSP time steps, the procedure used throughout our work is to generate a grid of models spanning over metallicity and age. We then interpolate this grid in log(age) and log(Z) in order to find the spectrum of each star-particle in the simulation. Note that all YGGDRASIL spectra have been re-scaled to the Kroupa universal stellar initial mass function1 (Kroupa, 2001). YGGDRASIL is also able to calculate a corresponding nebular emission spectrum for each SSP using the photo-ionization code CLOUDY (Ferland et al., 1998; Ferland et al., 2013). This is done assuming a spherically symmet- ric nebula with constant density (100 cm−3) and allowing nebular conditions to adapt to the incoming ionizing spectrum (Zackrisson et al., 2011, 2013). In this procedure, the nebular and stellar metallicity and abundance ratios are assumed to be the same. The nebular spectra are calculated under the assumption that no LyC photons are escaping. Different escape fractions are instead simulated by re-scaling the nebular spectra to account for different gas covering fractions ( fcov) and assuming that fesc = 1 − fcov. This assumes that the nebula is ionization bounded, and that the leakage occurs through holes in the gas (see section 5.1 and figure 5.1). In order to calculate the effects that dust has on the synthetic spectra, some recipe or measure of the dust amount is required. This recipe is different for the different simulations used throughout papers I–III. For the S14 and S16 simulations, a galaxy wide extinction at 1500 Å (AUV) is provided for each simulated galaxy. This extinction is based on the metallicity, total gas mass and size of each of the galaxies, and has been calibrated against the observed UV luminosity function at z = 7 (Shimizu et al., 2014). For the F13, CROC and FiBY galaxies, the dust-recipe presented in Finlator et al. (2006) is used. This recipe gives the B − V color excess, E(B − V), as a combination of a fixed component based on the overall stellar metallicity (Z), and a Gaussian

1Note that there was an error in the BPASS models used for paper I and paper II. This mistake led to a ﬂatter high-mass slope of the assumed IMF, resulting in an increased number of massive stars and boosted ionizing ﬂux (see paper I erratum and paper II erratum).

50 scatter δE with variance equal to one-half of the ﬁxed (metallicity dependent) component:

E(B − V)=9.0Z0.9 + δE (6.1)

The output of these two recipes are then used to redden the spectra using either of two extinction curves (LMC, SMC) or the Calzetti attenuation law. The latter has been implemented in two different versions. The first version is the standard law presented in Calzetti et al. (2000) where E(B − V)stellar = 0.44E(B − V)nebular. The second version uses the same attenuation for the nebular and stellar component E(B − V)stellar = E(B − V)nebular, following Erb et al. (e.g. 2006). In the former of these, the effect of dust is stronger on the nebular component of the spectrum. The effects of the different reddening- laws on the emerging spectra are shown in figure 6.1. Two major assumptions are made in this procedure: Firstly, we assume the dust is situated outside the H II regions, and that no LyC is absorbed by dust. Secondly, we also assume that dust is distributed evenly across the galaxy, and that all star-particles receive the same amount of extinction or attenuation. As discussed in section 4.1, there may be age-dependent effects that are thus not accounted for. To emulate such an effect in our code, we have implemented dust-recipes where an age-dependent dust amount is calculated for each individual star-particle, for example, following Bergvall et al. (2016). The effect of this recipe on the EW(Hβ) - β diagram is similar to the effect obtained by using the Calzetti law with different scaling for the stellar and nebular part. There are a number of simulations which have more advanced implementations for dust effects on high-redshift galaxy spectra. For example, the simulations presented in Wilkins et al. (2017, 2018); Katz et al. (2019b); Ma et al. (2019) all use recipes in which the dust content at different locations in the simulation is coupled to the gas properties. This can either be used for a dust radiation-transfer calculation (Ma et al., 2019), or to calculate the optical depth of dust toward star particles along a line of sight, and assuming some extinction- or attenuation law to account for dust reddening (Wilkins et al., 2017, 2018; Katz et al., 2019b). Such a recipe automatically accounts for differential obscuration and possible variations in dust-reddening with stellar age. In paper III, we use a version of the FIRE-2 simulations presented in Ma et al. (2019). The spectra of these galaxies are calculated both using our code and the YGGDRASIL population synthesis modes, and using the method where dust effects are calculated using the radiative transfer code SKIRT. This does lead to an age-dependence of the dust extinction, but to a significantly milder degree than what is argued to explain the strong Balmer break in MACS1149-JD1 by Katz et al. (2019b). This may be a consequence of the enhanced feedback effects in the FIRE-2 simulations, which lead to a ‘blown-up’ and porous ISM, thereby decreasing the dust density around star-forming regions. Indeed, in the simulations by Katz et al. (2019b), a strong Balmer break due to stronger

51 ing rasaedie yds xicin aoiyo h on tr n pre- and Balmer stars large young the the where of majority scenario H a the associated extinction, In the dust sumably by (2019b). driven scenario al. are the et breaks and (2018) Katz al. by et Hashimoto suggested by star-formation suggested large as with scenarios ﬂuctuations between activity distinguish to possible be would it which compact, communi- are personal that Katz, 2019). (H. objects feedback 2, in the September of cation, seen strength the only on is entirely depends stars young of obscuration with galaxy S14 a for calculated were spectra The I–III. papers out 6.1. Figure ae,tedr-rylnsso h pcr eutn rmtesadr Calzetti standard the bottom-right from to resulting top-left spectra the the From show lines (E(B law spectrum. dark-gray (dust-free) the intrinsic panel, the shows line 52 scenario a from is distinguished it ﬂuctuations. reliably lines, SFR be emission large the can with of scenario strength a parame- and such different SED the whether of of unclear number shape the the affect to can due that However, ters weak bright. very emission FIR to remain the lead should as vicinity should lines such scenario the lines dust-unaffected a in while such emission, emission dust, line the nebular UV/optical reaching to before reprocessed stars the is of radiation ionizing the that E(B Log (flux / erg s-1 Å-1)

nitrsigqeto eadn h aueo AS19J1i whether is MACS1149-JD1 of nature the regarding question interesting An 10 − YGGDRASIL V) nebular − ytei pcr oprn ifrn utrdeiglw sdthrough- used laws dust-reddening different comparing spectra Synthetic V) stellar Clet*,teLClw n h M law. SMC the and law, LMC the (Calzetti*), Sswith SSPs = 0.44E(B II BPASS ein r evl bcrdb ut Assuming dust. by obscured heavily are regions − V) Rest-frame wavelength (Å) nebular .. oes nalpnl,telight-gray the panels, all In models. v.2.0 ,teClet a hr E(B where law Calzetti the ), A − UV V) ∼ stellar us- 2 = 7. Radio interferometry and interferometric imaging

As we probe galaxies at increasing redshifts, their spectra shift toward longer wavelengths. At z > 6, FIR emission lines and dust emission shift into sub- millimeter and millimeter wavelengths. While this is actually useful in terms of atmospheric transmission, which is generally better at sub-millimeter wavelengths than at FIR wavelengths, the longer wavelengths require larger telescopes to maintain angular resolution. This is due to the relationship between angular resolution of a telescope and observed wavelength, which is θ ∝ λ/D, where λ is the wavelength and D is the aperture diameter. A way around this problem is to combine the signal from several telescopes, and thus ‘simulate’ a telescope with a diameter that is given by the separation of the telescopes rather than their individual diameters. The ALMA interferometer used in paper IV for example, consists of 66 antennae with diameters of 7–12 meters separated over distances up to 16 kilometers. Such long baselines allow ALMA to produce images with resolutions of ≈ 20 milliarcseconds at 200 GHz. Observations with millimeter and sub-millimeter interferometers such as ALMA have become increasingly popular in the high-redshift community. This is partly due to the difﬁculty of securing spectroscopic redshifts from the rest-frame UV and partly due to the fact that facilities such as ALMA can target bright FIR emission lines (such as [C II] 158 μm, [O III]88μm) and dust continuum at a high angular resolution. This means that these facilities provide us with several constraints on the high-redshift galaxy population that are currently unavailable at other wavelength ranges (see sections 4.1 and 4.2). As an example; until recently, the record for the most distant emission line from a galaxy was held by an [O III]88μm emission line in MACS1149- JD1 observed with ALMA. The combination of a high angular resolution and spectroscopic capabilities provided by interferometric observations with facilities such as ALMA also allows us to study the kinematics of high-redshift objects (e.g. Hashimoto et al., 2019; Matthee et al., 2019). The antennae in an interferometer are sampling the electromagnetic wave as it reaches the ground, but at different locations, and therefore with different delays. The signals from the antennae are then used to digitally form the interference pattern by multiplying and time-averaging the signals in a corre- lator. This means that the response of the interferometer is not an image in the conventional sense, but something known as the complex visibility V(u,v), where u and v are the projected distances (or baselines; measured in units of

53 the observed wavelength) between the antennae in the East-West and North- South direction, respectively, as seen from the source (e.g. Wilson et al., 2009; Thompson et al., 2017; Remijan et al., 2020). The plane spanned by the u and v directions is referred to as the (u,v) plane. The Cittert-Zernike theorem (van Cittert, 1934; Zernike, 1938) states that the complex visibility in the (u,v) plane is the 2-dimensional Fourier transform of the sky brightness distribution (I(l,m)) in the image plane: V(u,v)= I(l,m)e−2πi(ul+vm)dldm (7.1) Where l and m are direction cosines in the East-West and North-South directions in the image plane, respectively. Thus, the inverse Fourier transform of the complex visibility in the (u,v) plane is simply the sky brightness distribution in the image plane, or the image of the sky. I(l,m)= V(u,v)e2πi(ul+vm)dudv (7.2) However, since a real interferometer consists of a limited number of antennae, not all (u,v) distances will be sampled. Thus, the integral in equation 7.2 will actually become a summation. The sampling of the (u,v)-plane (the (u,v) coverage) depends not only on the number and placement of the antennae, but also on the observation time, since the projected baselines seen from the source change as the Earth rotates. The effect of the location observed in the (u,v)-plane (the (u,v)-coverage) and maximum baseline on the resulting (dirty) image can be seen in ﬁgure 7.1. The task of imaging data with, for example, ALMA is thus to calculate this inverse Fourier transform given the observed visibilities. The data is ﬁrst gridded into a regular grid in the (u,v) plane, and weighted using some weighing function. The Fourier transform is then calculated to produce an image commonly called the dirty image. This image is the convolution of the real sky brightness distribution and the so- called dirty beam, or point spread function (PSF).

ID(l,m)=bD(l,m) ∗ I(l,m) (7.3) Where ID is the dirty image and bD is the dirty beam. In order to retrieve I(l,m), the dirty image needs to be deconvolved with the dirty beam. A common way in which this is done is by using the clean algorithm (Högbom, 1974), which is an iterative procedure. First, bright peaks in the dirty image are identified. Secondly, a model for the bright peaks, containing a fraction of their flux is created. In this model, each bright peak is represented by a point source. The model is then convolved with the dirty beam, and subtracted from the dirty image. The process is then repeated with the remaining bright peaks in the image until some stopping criterion (for example, a target residual peak intensity) is reached. This way, the model is built up iteratively. The final model, consisting of a number of point-sources, is then convolved with a

54 clean beam, a Gaussian with full-width-at-half-maximum equal to the central component of the dirty beam, to produce the clean image. It is important to note that there are many different outcomes to this procedure that depend on the chosen imaging strategy. During the imaging process, it is possible to adjust the relative weighting applied to the visibilities in the (u,v) plane, and thereby adjust the resulting beam shape (Briggs et al., 1999; Thompson et al., 2017; Remijan et al., 2020). The applied weight is commonly split into a density weight, which can be used to offset for the variation in the density of sampled points in the (u,v) plane, and a taper weight, which is most commonly used to de-emphasize long baselines. An often used type of taper weight is a Gaussian in the (u,v) plane. Such a taper weight results in a larger beam. Density weighting can be done in different ways, but two common weighting schemes are natural and uniform weighting. Natural weighting means that each (u,v) sample is weighted in the same way. Since the (u,v) coverage is generally better at points close to the origin in the (u,v) plane, this will give more emphasis to short baselines. As a result, natural weighting leads to lower spatial resolution, but better point-source sensitivity. Uniform weighting, on the other hand, means that the weight of each visibility will be inversely proportional to the sample density at that (u,v) point. Since the sample density will generally be higher close to the origin in the (u,v) plane, this will lead to a lower emphasis of short baselines, leading to higher spatial resolution, but lower sensitivity. Normally, the ﬁnal image from the above procedure still contains an effect of the response of the individual antennae, or primary beam. For parts of the image close to the center, this effect and the required correction are usually small, but grow as you move away from the center of the image. The data used in paper IV was reduced using the standard pipeline within the Common Astronomy Software Application (CASA) software package. Imaging and de- convolution was done with the CASA task clean, using a natural weighting and various different tapering schemes.

55 SOURCE MODEL DIRTY IMAGE (Nant = 4, Tint = 3 h)

DIRTY IMAGE (Nant = 4, Tint = 6 h) DIRTY IMAGE (Nant = 10, Tint = 6 h)

Figure 7.1. The effect of observation time, UV coverage and baseline length on the resulting dirty image. The top-left image shows a simple Gaussian source model. The top-right image shows the dirty image resulting from observing this source for two hours using an array consisting of four antennae positioned in a Y-pattern. The bottom-left image shows the dirty image obtained after six hours of observation time. The image at the bottom right shows the same source observed with an array consisting of 10 antennae placed in the same sort of pattern, and an observation time of 6 hours. In this panel, the maximum baseline is longer (since the array is extended along the arms of the Y-pattern), and therefore the angular resolution is higher. Figure created with APSYNSIM (Marti-Vidal, 2017, https: //github.com/onsala-space-observatory/APSYNSIM).

56 8. Summary of papers

8.1 Paper I As discussed in chapter 5, the partially neutral IGM during the EoR makes it practically impossible to directly detect any leaking LyC from the galaxies that drove the reionization. For this reason, we must rely on indirect methods to assess the escape fraction of galaxies in this epoch. In paper I, we use simulated galaxies to test the method for identifying EoR LyC leakers with high escape fractions proposed by Zackrisson et al. (2013). The simulated galaxies used in paper I come from four different simulation suites (CROC, S14, F13 and FiBY; see section 6.2), extracted at z = 7. Due to the limited number of galaxies in the relatively small volume of the FiBY simulation, we combine galaxies at z = 6, 7, 8. As discussed in section 6.2, we set a limit on the minimum allowed mass of the simulated galaxies, which leads to a total of 874, 406, 106 and 16 galaxies from the CROC, S14, F13 and FiBY simulations, respectively. For the ﬁducial set of spectra, SSP spectra with nebular emission from YGGDRASIL are used. These are calculated using STARBURST99 models with Geneva stellar evolutionary tracks. To test assumptions regarding the stellar evolutionary models, additional sets of spectra are generated with STAR- BURST99 Padova-AGB evolutionary tracks and BPASS v.2.0 models. Effects of dust reddening are calculated using the recipes and extinction/attenuation laws discussed in section 6.3. Figure 8.1 shows the EW(Hβ) - β diagram for dust-free simulated galaxies from the four simulation suites with varying escape fractions (panels a, b and c). According to the method suggested in Zackrisson et al. (2013), galaxies with different escape fractions should form distinct groups in this diagram. While there is a substantial internal scatter (due to varying star formation histories and internal metallicity distributions), the galaxies with different escape fractions remain relatively well separated. Given the distributions shown in the diagram, it should be possible to identify galaxies with fesc 50% using only the UV slope and EW(Hβ). The CROC galaxies exhibit the smallest spread of any of the simulations, mainly due to the smoother star formation histories of these simulations compared to the others. For clarity, the FiBY simulations are shown in the same panel as the F13 simulations. Note that while the FiBY simulations have escape fractions that are predicted in the simulations, these are binned into fesc = 0.0, 0.5, 0.7, 0.9 depending on their escape fraction. As ﬁgure 8.1 shows, FiBY galaxies are generally caught in stages of extremely high LyC leakage (around 90%), or in stages where there is essentially zero

57 Figure 8.1. EW(Hβ) - β diagram for dust-free simulated galaxies. Panels a and b show the S14 and CROC galaxies, respectively. Panel c shows the F13 galaxies (dots) and FiBY galaxies (squares). Panel d shows the CROC galaxies at z = 7 (light circles) and z = 9 (dots). The colors represent different escape fractions: fesc = 0.0 (red), fesc = 0.5 (yellow), fesc = 0.7 (green) and fesc = 0.9 (blue). Figure from paper I.

58 Figure 8.2. EW(Hβ) - β diagram for simulated galaxies from the S14 simulations with effects of dust. Panels a, b and c show the positions of galaxies resulting from dust-reddening using the LMC, SMC and Calzetti (E(B − V)stellar = 0.44E(B − V)nebular) reddening laws, respectively. Panel d shows the positions of the galaxies resulting from the modiﬁed Calzetti law, where E(B − V)stellar = E(B − V)nebular. The colors represent different escape fractions: fesc = 0.0 (red), fesc = 0.5 (yellow), fesc = 0.7 (green) and fesc = 0.9 (blue). Figure from paper I.

leakage of LyC photons. Panel d in ﬁgure 8.1 shows the redshift evolution of the CROC galaxies between z = 9 and z = 7 in the EW(Hβ) - β diagram. The marginal difference in the distributions at these redshifts shows that there is relatively little evolution in the EW(Hβ) - β diagram over this period spanning across ∼ 200 Myr of cosmic time. Figure 8.2 shows the effects of dust-reddening on the S14 galaxies in the EW(Hβ) - β diagram. While the S14 simulations predict relatively modest dust reddening AV = 0.2–0.4, the effect on the EW(Hβ) - β diagram is substantial. For the simulations not shown in this diagram, the dust amounts are slightly smaller, and thus dust effects are slightly milder. As shown in the diagram, dust effects dominate the UV slope, which substantially reduces the usefulness of the UV slope as a probe of the escape fraction. In the scenarios shown in panel a, b and d, one should still be able to infer the escape fraction

59 using only EW(Hβ), since essentially all objects with EW(Hβ) 30 Å have fesc 50%. While the distribution of objects in the EW(Hβ) - β diagram in itself provides a constraint if we have a large observed sample, inferring the escape fraction of individual objects can be problematic if we have no handle on the most representative dust-reddening law. This can be seen when comparing the Calzetti attenuation law (panel c) to any of the other reddening laws. The Calzetti attenuation law will shift galaxies toward smaller EW(Hβ), this may lead to a situation where a non-leaking galaxy shifts to a position in the EW(Hβ) - β diagram that would be associated with a higher escape fraction if one assumes any of the other reddening laws. For redshifts up to z ∼ 7.2, Hα should fall within JWST/NIRSpec wavelengths, allowing for dust-corrections using the Hα and Hβ line ratios. At higher redshifts, additional observations using JWST/MIRI would be required to obtain measurements of Hα. While both Hγ and Hβ should fall within JWST/NIRSpec wavelengths at these redshifts, the modest amounts of dust in the majority of galaxies in combinations with relatively weak rest-frame equivalent widths of Hγ ( 10 Å for the STAR- BURST99 Geneva models with zero escape fraction and Calzetti attenuation) will likely make dust corrections using these lines problematic. The BPASS models lead to slightly stronger lines. This is due to the increased ionizing flux as an effect of binary evolution and stellar rotation (the increase in EW(Hβ) is a factor of ∼ 1.61). Without prior knowledge of the most representative stellar evolutionary model, this shift slightly increases the risk of mis-classifying galaxies in terms of their escape fractions. In theory, it should be possible to alleviate this risk by checking for mis-calibrations in the stellar evolutionary models using the e.g. distribution of galaxies at the largest equivalent widths of Hβ or possibly Hα at z < 7.2. The positions of the galaxies in the EW(Hβ) - β diagram calculated using Padova-AGB stellar evolutionary tracks are not qualitatively different than when using the Geneva tracks. Our results indicate that the method of Zackrisson et al. (2013) seems to hold, despite variations in the SFHs and internal metallicity distributions predicted by simulations. This means that one should be able to find galaxies with fesc 50% using relatively simple diagnostics, such as the UV slope and EW(Hβ). We find that uncertainties associated with how to account for dust- reddening effects and uncertainties regarding the choice of stellar evolution model may lead to systematic uncertainties in the inferred fesc of individual galaxies. Auxiliary information obtained both at z > 6 and at lower redshift may help to constrain such uncertainties.

1using the corrected BPASS models, see paper I erratum

60 8.2 Paper II The aim of paper II is to study the impact of a recent decline in the start- formation activity on the applicability of methods for indirectly estimating the escape fraction of ionizing photons using equivalent-width measurements of Balmer lines (paper I; Zackrisson et al., 2013). In paper I, we used four different simulation suites to test if the method holds when applied to realistic simulated galaxies despite possible variations in metallicities and SFHs. How- ever, a number of simulations predict variations in the star-formation activity that are larger than those predicted by the four simulations used in paper I (Kimm & Cen, 2014; Kimm et al., 2015; Ma et al., 2015; Trebitsch et al., 2017; Yajima et al., 2017; Ma et al., 2018). In addition, large variations in the star-formation activity have been suggested to explain the recent observations of MACS1149-JD1 (Hashimoto et al., 2018). If one were to catch a galaxy in a state of decreased star formation (in this paper, we refer to such objects as quenched), it is likely that there exists a time-span in which the ionizing UV emission has decreased substantially while the shape of UV/optical continuum has remained relatively unchanged due to the different types of stars that dominate these parts of the spectrum. Such a scenario could lead to relatively small EW(Hβ) in combination with blue UV slopes, which could be misinterpreted as LyC leakage. An additional aim of this study is to assess whether auxiliary information obtained from JWST observations can be used to distinguish between scenarios of high LyC leakage and scenarios where star formation has been quenched. In order to test the aforementioned scenario, we use simulated galaxies from the S14 simulations and YGGDRASIL SSPs generated with BPASS v.2.0 stellar models. The escape fraction of the galaxies is allowed to vary from 0 to 1 in steps of 0.05. As the fiducial dust-reddening law, we use a modified Calzetti law, where E(B − V)stellar = E(B − V)nebular. However, to test the assumption of this reddening law, additional models with the standard Calzetti law and the SMC law are also generated. In addition to the S14 simulations, mock galaxies that have experienced a recent drop in the SFR are created. The size and timescale of the drop is selected in such a way that it roughly overlaps with the range of the variations seen in the simulations by Kimm et al. (2015); Ma et al. (2015); Ma et al. (2018). Figure 8.3 shows the star formation histories and cumulative stellar masses of three simulated galaxies and three mock galaxies that have experienced different decreases in the SFR. The effect of a decline in the SFR is a decrease in the number of LyC photons produced by the stellar population, and therefore a decrease of the strength of nebular emission features. However, the stellar continuum at 4000 Å remains relatively constant. The net effect is a decrease in EW(Hβ). Since the UV continuum is dominated by stars with longer lifetimes, the evolution of the UV slope is significantly slower. For mock galaxies that have experienced a factor 10 decline in the SFR, the equivalent width has dropped

61 ) ) 1 1 − − yr yr 0 0 10 M M 10 -1 R( 10 F S SFR ( 0.2 0.3 0.4 0.5 0.6 0.7 0.2 0.3 0.4 0.5 0.6 0.7 Cosmic time (Gyr) Cosmic time (Gyr) ) ) 9 10 9 10 8 10 8 10 10 7 10 7 0.2 0.3 0.4 0.5 0.6 0.7 0.2 0.3 0.4 0.5 0.6 0.7 Stellar mass ( M Stellar mass ( M Cosmic time (Gyr) Cosmic time (Gyr)

Figure 8.3. SFHs (top) and total stellar mass (bottom) versus cosmic time. The left panel shows three example simulated S14 galaxies, and the right panel shows three example mock-galaxies that have experienced a factor of 5 (yellow), 10 (red) and 100 (black) quenching for 50 Myr. The SFR and stellar mass are calculated in 10 Myr bins. Figure from paper II. by several factors already after 10 Myr, while the average UV slope has only changed by ∼ 0.2 within 50 Myr after the drop in the SFR. This is well within the spread of reddening experienced by the galaxies due to dust if we assume a Calzetti reddening law. This scenario would therefore not be distinguish- able from a normal star forming galaxy with a slightly higher dust content by observing only the UV slope. Thus, these quenched galaxies would not be dis- tinguishable from LyC leaking galaxies in the EW(Hβ) − β diagram used in paper I. Examples of three cases of this degeneracy can be seen in figure 8.4. As can be seen in this figure there are, however, some subtle differences in the spectra, for example around the Balmer break or longward of rest-frame ∼ 7000 Å. To assess whether the spectra contain sufficient information to identify such quenched galaxies, we generate mock photometric JWST/NIRCam and MIRI fluxes of the normal and quenched galaxies. We then feed these into a classi- fication algorithm (linear discriminant analysis; LDA) and train the algorithm to identify the galaxies as belonging to one of the two classes ‘quenched’ or ‘normal’. Allowing for quenching factors of 5,10 and 100, and quenching durations of 10 – 100 Myr in steps of 10 Myr, we find that about 80% of the quenched galaxies are correctly identified as quenched, and about 85% of the normal galaxies are correctly identified as normal. In figure 8.5, the true positive rate (fraction of quenched galaxies correctly identified as quenched) are shown as a function of quenching duration and quenching strength. It is clear that the problematic cases are those that experience relatively small drops in the star-formation activity, since these more closely resemble the normal simulated galaxies. However, it is also clear that cases with short quenching durations are problematic. At about 20 Myr after the drop in SFR, only about 80% of the galaxies that have undergone a drop

62 4 Star Forming, fesc ≥ 50% a) Quenched, fesc =0% 3 ) 2 1250

/F b) λ

F 1 ( log 0 c)

-1

2000 3000 4000 5000 6000 7000 8000 9000 10000 11000 Rest Frame Wavelength (A)˚

Figure 8.4. Synthetic spectra of simulated S14 galaxies with fesc ≥ 50% (black), and quenched mock-galaxies with zero escape fraction (red, dashed). The spectra in a) show a simulated galaxy with an escape fraction fesc = 50% and a visual attenuation AV ≈ 0.41 mag together with a factor 5 quenched mock-galaxy with AV ≈ 0.15 mag. The spectra in b) show a simulated galaxy with fesc = 70%, AV ≈ 0.49 mag and a factor 10 quenched mock-galaxy with AV ≈ 0.21 mag. The spectra in c) show a simulated galaxy with fesc = 90%, AV ≈ 0.51 mag along with a factor 100 quenched mock-galaxy with AV ≈ 0.09 mag. The spectra have been normalized at 1250 Å, and the spectra for a) and b) have been shifted upward for clarity. For all these mock galaxies, the quenching has been ongoing for 40 Myr. Figure from paper II. of a factor 100 in the SFR are correctly identiﬁed. The corresponding number for a factor 10 drop is about 70%. For galaxies that have undergone a drop in the SFR of a factor 10-100 for 50-100 Myr, the true positive rate is 85%.

Figure 8.5. True positive rate (TPR, or recall; the fraction of quenched galaxies that are correctly identiﬁed as quenched) for different quenching fractors: 5 (yellow), 10 (red) and 100 (black) as a function of quenching timescale. Figure from paper II.

63 When testing the method with varying dust recipes, we find that unless we know the ‘true’ reddening law (or most representative reddening law), dust can pose a problem, lowering the accuracy of the method significantly. For example, if the true reddening law is a Calzetti-like law, and we assume that it is more like an SMC law, only 64% of the normal galaxies will be identified as normal, compared to 85% if the true law is used. In principle, if there is a subpopulation of galaxies that experience variations in the SFR as large as 10-100 over time-scales of 50-100 Myr or larger, for example such as the galaxy MACS1149-JD1 at z = 9.1096, we should be able to statistically identify such a population using the JWST, and in this way assess to what degree this may pose a problem for the Zackrisson et al. (2013) method. However, scenarios of more modest quenching may be more difficult to identify using the types of observations discussed here. We also argue that a high-priority task for the JWST should be to characterize dust in EoR galaxies.

8.3 Paper III In paper III, we investigate to what degree contemporary simulations produce star-formation activity variations large enough to reproduce the observed Balmer break of MACS1149-JD1 (hereafter JD1) at z = 9.1096. This galaxy, which has an estimated stellar mass of ∼ 109 M, has been suggested to have experienced a large drop in the SFR and a period of between ∼ 100 to 200 Myr of low or no star formation. While substantial variations in star-formation activity are seen in some simulations, it is unclear whether these are large enough, and occur over timescales that are long enough to reproduce the SED of JD1. The reason that this type of SFH has been suggested for JD1 is that the galaxy appears red in the Spitzer/IRAC 3.6 μm and 4.5 μm bands. While this may be explained by strong [O III] 5007 Å emission within the 4.5 μm band at lower redshifts, the spectroscopic redshift of JD1 (z = 9.1096) places this line outside the 4.5 μm band. Instead, the explanation suggested for the red IRAC color is a large Balmer break. In this paper, we use simulated galaxies from the S16, FIRE-2 and First- Light simulations to investigate whether the simulated galaxies are able to reproduce the IRAC observations of JD1. In addition, we predict Balmer break distributions for the simulated z ≈ 7 – 9 galaxies that may be tested with future spectroscopic JWST observations. The simulated galaxies from the FIRE-2 simulations have been post-processed in the dust radiative transfer code SKIRT. In order to test the effect of the typical assumption of a dust screen we use a second set of FIRE-2 galaxies, with a ﬁxed extinction in the V band of 0.5. Note that this relatively large amount of dust extinction is unlikely to be representative for the majority of the high-redshift galaxy population.

64 Figure 8.6. Balmer break strength distributions (Fν (4200Å)/Fν (3500Å)) for the S16 galaxies (black), FIRE-2 galaxies with YGGDRASIL SEDs and ﬁxed extinction in the V band of 0.5 (green) and FIRE-2 galaxies that have been post-processed in SKIRT (blue). The panels show z = 9 (top), z = 8 (middle) and z = 7 (bottom). All galaxies have stellar masses M ≥ 108M. The total number of galaxies at each redshift (z = 9; 8; 7) is 182, 513 and 1277 for the S16 simulation, and 150, 152 and 187 for the FIRE-2 simulation. Figure from paper III.

65 Figure 8.7. Balmer break strength distributions as probed by Spitzer/IRAC 4.5/3.6 micron flux-ratio for simulated galaxies at z = 9.11. The left panel shows all galaxies with stellar masses M ≥ 108M, while the right panel shows galaxies with stellar masses M ≥ 5 × 108M, i.e. closer to the predicted mass of JD1. The distributions show the S16 galaxies (black), FIRE-2 galaxies with YGGDRASIL SEDs and fixed extinction in the V band of 0.5 (green) and FIRE-2 galaxies that have been post-processed in SKIRT (blue). The dashed yellow line and yellow area show the Spitzer/IRAC flux ratio and error measured for JD1. Figure from paper III.

The predicted Balmer break distributions as probed by the flux at 4200 Å and 3500 Å rest-frame (Fν (4200Å)/Fν (3500Å))atz = 9 – 7 are shown in figure 8.6. In general, the FIRE-2 galaxies produce a larger spread in the predicted Balmer break distributions, due to the burstier nature of these objects. The larger variation in the star formation activity means that we are more likely to catch galaxies in low-SFR phases. As shown in the figure, there seems to be a slight trend toward an increasing fraction of galaxies with large Balmer breaks (> 2) with decreasing redshift among the FIRE-2 galaxies. Looking closer at the FIRE-2 galaxies that have been post-processed in SKIRT, we find that dust has an insignificant effect on the Balmer break distributions. Both the dust-free and dusty FIRE-2+SKIRT distributions are comparable to the ones obtained without dust using the YGGDRASIL models. The distribution obtained from the FirstLight simulation is slightly narrower and centered at lower mean value than those obtained using FIRE-2. Part of this can be explained by the difference in the SED modelling. We also test assumptions regarding the escape fraction of ionizing photons and the stellar IMF, but find that neither of these are likely to lead to significant increases in Balmer breaks unless we employ a scenario with a very bottom-heavy IMF. Figure 8.7 shows the Balmer break strength distributions of the simulated galaxies as measured by Spitzer/IRAC, along with the measured Balmer break in JD1. While the FIRE-2 simulations produce objects that exhibit stronger Balmer breaks than the S16 simulations, neither of the simulations produce

66 galaxies with Balmer breaks comparable to JD1. This is the case even if we assume a stellar mass which is up to a factor of 10 lower than that suggested for JD1. While stronger dust effects, LyC escape or more bottom-heavy IMFs may lead to stronger Balmer breaks, we ﬁnd no likely scenario which would push the simulations to produce Balmer breaks of the size seen in JD1. We conclude that objects such as JD1 are uncommon in the simulations used here, and unless it represents an outlier in the high-redshift galaxy population, there could be some physical process that the simulations fail to capture. As JD1 represents a single measurement, and given the size of the Spitzer/IRAC error- bars, future observations of high-redshift objects should provide a clearer picture of the consistency between simulated galaxies and the real high-redshift population.

8.4 Paper IV In paper IV we present results from ALMA observations targeting the [O III] 88 μm and [C II] 158 μm FIR emission lines and FIR dust-continuum in the z = 7.6637 ± 0.0011 galaxy z7_GSD_3811. This galaxy has been observed using both HST and the VLT within the Cosmic Assembly Near-infrared Deep Extra-galactic Legacy Survey (CANDELS; Grogin et al., 2011; Koekemoer et al., 2011). The spectroscopic redshift has been determined via Lyα from follow-up observations with Keck/MOSFIRE (Song et al., 2016). At the redshift of this galaxy, the [O III]88μm and [C II] 158 μm emission lines fall within ALMA bands 8 and 6, respectively. This makes it a suitable target for studying the ISM via these lines and their relative strengths. In addition, a number of objects with similar UV magnitudes have been detected in [C II] and/or [O III] (Pentericci et al., 2016; Carniani et al., 2018; Inoue et al., 2016; Tamura et al., 2019; Bakx et al., 2020). We combine observations performed with ALMA during April, August and September of 2018 (PI: A. Inoue) with archival data taken during March and May of 2016 (PI: S. Finkelstein). We find that z7_GSD_3811 is not detected in either of the targeted emission lines nor in the FIR dust-continuum. In order to analyze the implications of the non-detection, we derive upper limits to the line and continuum luminosities. For line luminosities, we use a line-width of 100 km s−1, based on earlier measurements of the lines, as our fiducial value. For the dust continuum, we calculate a total IR dust luminosity by assuming that the dust can be modeled as a modified black-body with a temperature of Td = 45 K and dust emissivity index of β = 1.5. This black-body curve is integrated over 8 – 1000 μm to obtain the total IR dust luminosity and IR SFR. Figure 8.8 shows the [C II] luminosity versus the UV+IR SFR. For com- parison, we also show other high-redshift galaxies and [C II]-SFR relations from the literature. The upper limit on the [C II] line places z7_GSD_3811

67 Figure 8.8. [C II] luminosity versus the UV+IR SFR. The yellow star and arrow show the position of our [C II] upper limit of z7_GSD_3811 for the fiducial line-width of 100 km s−1. The yellow dash and arrow show the upper limit obtained by assuming a line-width of 400 km s−1. The dark-gray circles show z > 6 galaxies from the compilation of Matthee et al. (2019), where upper-limits are calculated assuming a line-width of 100 km s−1. The galaxies from this compilations that have alse been detected in [O III] are indicated by their names and a lighter gray color. The orange dashed line and the pink line show the local relation for local HII/starburst galaxies and metal-poor dwarf galaxies by De Looze et al. (2014). The shaded regions around these relations show the 1σ dispersion in the relations. The black dotted and dash- dotted line show the relations derived by Schaerer et al. (2020) calculated using 3σ and 6σ upper limits for non-detections, respectively. The upper limits in the figure are at the 3σ level. Figure from paper IV. below the local relationships for HII/Starburst galaxies and metal-poor dwarf galaxies by De Looze et al. (2014) as well as the relation by Schaerer et al. (2020). We note, however, that our upper limit could, in principle, be consistent with either relation given the scatter around these. We use the [C II]- SFR-Metallicity relation by Vallini et al. (2015) to derive an upper limit to the metallicity, and find Z 0.07 Z for our fiducial line-width. Our [O III] upper limit places z7_GSD_3811 below the [O III] to UV lumi- / nosity ratio (L[O III] LUV) of earlier galaxies that have been detected in [O III] (see figure 8.9). Compared to SXDF-NB1106-2 and MACS0416_Y1, the / ∼ ∼ L[O III] LUV ratio of z7_GSD_3811 is a factor of 5 and 12 times lower, / respectively. We also compare the L[O III] LUV ratio to those of nearby low- metallicity dwarf galaxies. We find that the oxygen abundance of local ob- / jects with L[O III] LUV below our upper limit is 10% of the solar value. This suggests that a possible explanation for the relatively weak [O III] and [C II]

68 Figure 8.9. Left: The [O III] over UV luminosity ratio for z7_GSD_3811 derived from our upper limit assuming a line-width of 100km s−1 (yellow star), and assuming − a line-width of 400 km s 1 (yellow dash). The dark gray circles show [O III]-detected galaxies at (z > 7). The light gray squares show three recent detections at z ∼ 6. The UV luminosities have been corrected for lensing magnification, but errors in the lensing magnification are not included in the error-bars. Figure from paper IV. of z7_GSD_3811 could be a low metallicity. We use observations from CAN- DELS and our upper limits from ALMA to model the SED of the galaxy using the PANHIT code. Within the parameter space that we explore, we find three best-fitting models that are statistically indistinguishable (see figure 8.10). Two of these models contain older stellar populations (0.29 Gyr), one with a high metallicity (Z = 0.02) and one with a low metallicity (Z = 0.0004). The third model also has a low metallicity (Z = 0.0001) but is substantially younger (0.002 Gyr). Neither of the three models contain any dust, nor do we find any other models with substantial amounts of dust that are able to match our observations. The fact that we find high-metallicity solutions among the best-fitting models suggests that low metallicity is not an exclusive explanation to the observed SED. If we compare the SED-fitting solutions to the S16 simulations, we see that galaxies with masses comparable to those predicted by the SED-fitting (∼ 109 M) and approximately solar metallicities are rare in these simulations. On the other hand, so are objects with Z = 0.0001 – 0.0004. The metallicity values obtained from our upper limits on the [C II] and [O III] lines (Z 0.1Z ≈ 0.0014)2 show better consistency with the simulated galaxies of S16.

2Using the Asplund et al. (2009) solar metallicity

69 Dust Continuum

Jy) 100 Used phot 75 Upper limit 50 1 Young 25 High Z 0 Jy) Flux density ( Flux density

Low Z 500 1000 1500

0.1 )

-2 Emission line(s) Flux density ( Flux density cm

-1 1 0.8

erg s 0.6

0.01 -18 0.4 0.2 0

1 2 3 5 6 Flux (10 500 1000 1500

Observed Wavelength (m)

Figure 8.10. PANHIT SEDs of the three best-fitting models from SED-fitting. The yellow and pink lines show the models with ages of 0.29 Gyr and metallicities of Z = 0.0004 and Z = 0.02, respectively. The blue line shows the young solution with an age of 0.002 Gyr and metallicity of Z = 0.0001. The main panel shows the SED in the rest-frame UV and optical, with upper limits and measurements from HST and VLT/HAWK-I. The top-right panel shows our upper limits on the dust continuum fluxes, where the values for the three models completely overlap (crosses). The bottom-right panel shows our upper limit on the [O III] luminosity. Upper limits are at the 3σ level. Figure from paper IV.

We conclude that a likely explanation for our observations is that z7_GSD_3811 contains relatively little dust and metals. Given the limited number of constraints from the UV and ALMA data we are, however, unable to rule out other explanations and provide detailed constraints on the nature of z7_GSD_3811. We argue that future observations with the JWST may help in this regard.

70 9. Summary and outlook

The cosmic reionization required early astronomical objects that were able to produce and release large amounts of ionizing radiation into the IGM. While current constraints on the reionization process are consistent with the bulk of ionizing photons coming from star-forming galaxies, our understanding of these objects still contains substantial gaps. As discussed in chapter 4, metallicities, SFHs and dust properties of these objects are still uncertain, and observations that might allow us to probe these are currently difficult to perform. Furthermore, the escape fraction of ionizing photons from the early galaxy population remains highly uncertain. While substantial escape fractions are required to allow a sufficient number of LyC photons to escape from these objects, the metallicity and SFHs affect the production efficiency of such photons. Constraining these properties is therefore crucial to understand the role that early star-forming galaxies played in the reionization. In this thesis, I have presented my work aimed at investigating EoR galaxies and exploring methods that will allow us to better do so in the future using both simulated galaxies and observations. Using simulations, we have shown that relatively simple diagnostics that are observable with the JWST may allow us to identify EoR galaxies with high LyC escape fractions. We have also investigated possible complications in the escape fraction es- timation that could arise from variations in star formation activity. Our results indicate that such variations can lead to spectral features that are to some degree degenerate with a high escape fraction. We show that if these variations are large, and occur over long time-scales, the shape of the SED contains enough information to disentangle variations in star formation activity from high escape fractions. We have also shown that contemporary simulations struggle to reproduce the large Balmer break of a recently observed z = 9.1096 galaxy. While this could indicate that this object is an outlier in the high-redshift galaxy population, we highlight that this could mean that the simulations may be missing some key physics. Objects with similar spectral features exist, albeit without spectroscopic redshift confirmations, meaning that the JWST may clarify the situation both by providing spectroscopic redshifts, and by observing Balmer breaks in a wider range of objects. Finally, we present results from ALMA observations of the z = 7.6637 galaxy z7_GSD_3811. Similar to a number of observations at high redshifts, the object is undetected in [C II] 158 μm FIR emission line and FIR dust emission. However, unlike earlier observations of high-redshift galaxies in

71 the literature, the galaxy is also undetected in the [O III]88μm FIR emission line. Using SED modeling and by comparing our observations to low-redshift observations and theoretical studies, we show that the non-detections could imply that the galaxy is poor in metals and dust. The non-detection of [O III] could also indicate that there is a larger spread in galaxy properties than indicated by earlier [O III] observations. Our results from modeling this galaxy should enable future observers to further constrain the nature of the object. The coming years will likely be very exciting for the high-redshift community. With the JWST, it will be possible for the first time to observe the rest-frame optical spectrum of many of the galaxies that drove the cosmic reionization. These observations will allow us to obtain better constraints on metallicities, SFHs and dust properties in objects such as z7_GSD_3811 discussed in paper IV. In combination with further ALMA observations of FIR emission lines and dust continuum, these constraints will likely improve even more. Future JWST and ALMA observations should also allow us to calibrate simulations, and to rule out unrepresentative simulation scenarios. Using methods such as the one discussed in paper II, for example, we may be able to constrain on how common large SFR fluctuations are in the high-redshift Universe. Fur- thermore, by comparing the distributions of observed spectral features, such as the Balmer break, it may be possible to determine how well current simulations capture the physics of real EoR galaxies. Another interesting question in this regard is whether observations can be used to distinguish between scenarios where large Balmer breaks are driven by fluctuations in star-formation activity, and scenarios where large Balmer breaks are driven by dust in dense environments around young stars. Armed with well-calibrated models, it may be possible to use methods such as the one presented in paper I (and other methods presented in section 5.3) to estimate the escape fraction of ionizing photons from EoR galaxies. Doing this would be a great step towards forming a complete understanding of the role these galaxies played in the cosmic reionization, and how the Universe evolved from the cosmic dark ages to the present day.

72 10. Contributions to included papers

Paper I Zackrisson, E., Binggeli, C., Finlator, K.; Gnedin, N. Y., Paardekooper, J.-P., Shimizu, I., Inoue, A. K., Jensen, H., Micheva, G., Khochfar, S. and Dalla Vecchia, C. (2017) The spectral evolution of the ﬁrst galaxies. III. Simulated James Webb Space Telescope spectra of reionization-epoch galaxies with Lyman-continuum leakage The Astrophysical Journal, 836, 78

The project was led by E. Zackrisson, who also deﬁned the scope of the project, with input from myself and co-authors. Co-authors ran the simulations and provided simulated galaxies. E. Zackrisson and I wrote the code for generating synthetic spectra for the simulated galaxies. I performed the data analysis (calculated synthetic spectra for the simulated galaxies, calculated spectral features and compared results for different model parameters) using codes written by myself and E. Zackrisson. I also created ﬁgures 1, 5, 6, 7 and 8 in the paper. The interpretation of the results was done E. Zackrisson and myself. I participated in, and contributed to, the discussion of the results and writing the manuscript.

Paper II Binggeli, C., Zackrisson, E., Pelckmans, K., Cubo, R., Jensen, H. and Shimizu, I. (2018) Lyman continuum leakage versus quenching with the James Webb Space Tele- scope: The spectral signatures of quenched star formation activity in reionization-epoch galaxies Monthly Notices of the Royal Astronomical Society 479, 368-376

The scope of the project was formulated by E. Zackrisson and myself. I also led the project. I. Shimizu ran the simulations and provided simulated galaxies. I generated the mock galaxies, calculated the synthetic spectra and mock JWST/NIRSpec and MIRI observations using codes developed by myself and E. Zackrisson. I did the analysis, including the LDA classiﬁcation and interpretation of the results with input from co-authors. I led the process of writing the manuscript with input from co-authors.

73 Paper III Binggeli, C., Zackrisson E., Ma, X., Inoue A. K., Vikaeus, A., Hashimoto, T., Mawatari, K., Shimizu, I., Ceverino D. (2019) Balmer breaks in simulated galaxies at z>6 Monthly Notices of the Royal Astronomical Society 489, 3827-3835

The scope of the project was formulated by myself and E. Zackrisson with input from co-authors. I led the project. Co-authors ran the simulations and provided simulated galaxies. I calculated the simulated galaxy spectra based on YGGDRASIL models. X. Ma ran the SKIRT radiative transfer calculations and calculated the FIRE-2+SKIRT spectra, and D. Ceverino calculated the First- Light spectra. A. Vikaeus calculated spectra with varying IMF slopes and wrote the corresponding part of the paper. I calculated the mock Spitzer/IRAC observations and did the analysis of the simulated galaxies, their Balmer break distributions and interpretation of the results with input from co-authors. E. Zackrisson wrote the abstract, and I wrote the rest of the manuscript with input from co-authors.

Paper IV Binggeli, C., Inoue, A. K., Hashimoto, T, Toribio, M. C., Zackrisson, E., Ramstedt, S., Mawatari, K., Harikane, Y., Matsuo, H., Okamoto, T., Ota, K., Shimizu, I., Tamura, Y., Taniguchi, Y., Umehata, H. (2021) A puzzling non-detection of [O III] and [C II]fromaz≈ 7.7 galaxy observed with ALMA Astronomy & Astrophysics, 646, A26

The co-authors (PI A. Inoue) wrote the observing proposal. The observations were reduced using an automatic pipeline. I combined the data from different observation epochs and the archival data and did pre-imaging checks (pipeline reduction output, noise expectations and calibrator variability). I did the imaging of the data and following analysis of the resulting images and image cubes, performed the SED-ﬁtting and the analysis of the results with input from co-authors. I wrote the manuscript with input from co-authors.

74 11. Svensk sammanfattning

Denna avhandling handlar om den kosmiska återjoniseringen och de tidiga stjärnbildande galaxerna som vi tror drev denna process. Vår nuvarande för- klaringsmodell av Universums utveckling innebär att Universum började i ett hett och kompakt tillstånd. Att Universum var hett och kompakt innebar att materian som fyllde tidiga Universum var joniserad, dvs. att den bestod av fria joner och elektroner. I takt med att Universum expanderade och kyldes av bör- jade de första atomerna bildas, och Universum blev neutralt. Så småningom ledde fluktuationer i densitet till att gasen började samlas i kompakta regioner ur vilka de första stjärnorna och galaxerna bildades. Dessa tidiga objekt pro- ducerade joniserande fotoner (s.k. LyC-fotoner) och började att jonisera den neutrala gasen i omgivningen i den process som kallas den kosmiska återjoni- seringen. Genom observationer har man lyckats fastställa att den kosmiska återjo- niseringen skedde kring rödförskjutning (z) 6 till 10, och att universum var mestadels joniserat kring z ∼ 6. Mycket pekar på att återjoniseringen drevs av de första stjärnbildande galaxerna, men det finns fortfarande stora luckor i vår förståelse av dessa objekt. Vi vet till exempel väldigt lite om hur det interstellära mediet i tidiga galaxer såg ut, och om den joniserande strålningen som stjärnorna i galaxerna bilda- ed verkligen kunde ta sig ut. Läckage av joniserande fotoner har observerats i lokala universum och upp till z ∼ 4. Dock blir det mycket svårt att observera läckande joniserande fotoner från galaxer under återjoniseringen, då dessa ab- sorberas i det delvis neutrala intergalaktiska mediet. Huruvida en galax läcker joniserande strålning lämnar dock ett avtryck i den icke-joniserande delen av dess spektrum. Detta sker eftersom att den joniserande strålning som absorbe- ras i galaxen (och därmed inte läcker ut) leder till starka emissionslinjer i galaxens spektrum. Det innebär att man kan använda galaxens icke-joniserande spektrum för att förstå dess joniserande spektrum. Vidare så finns det fortfarande osäkerheter kring hur metallrika de tidiga galaxerna var, hur mycket stoft de innehöll och hur stjärnbildningen i tidiga galaxer såg ut. Både observationer och simuleringar verkar peka mot ett relativt metallfattigt och stoftfritt tidigt universum. Samtidigt finns det observationer som skiljer sig något från den bilden, och indikerar att det kan finnas en sprid- ning i tidiga galaxers egenskaper. De flesta simuleringar indikerar att mängden stjärnbildning ökade över tid i tidiga Universum, men huruvida denna ökning var jämn eller stötvis är fortfarande oklart. Vissa simuleringar förutspår stora variationer i stjärnbildning över tid, medan andra pekar på något mer jämna

75 ökningar. Observationer av en galax vid z = 9.1096 (MACS1149-JD1) verkar överensstämma med mycket stora variationer i galaxens stjärnbildning över tid. Denna typ av stjärnbildningshistorik kan ha implikationer för när sjärn- bildningen påbörjades i dessa galaxer, och om de är vanliga i tidiga Univer- sum är det viktigt att vi att vi känner till och förstår detta för att undvika att misstolka observationer. Inom några år kommer James Webb-teleskopet (JWST) att skickas ut i rym- den, och kommer då att ge astronomer möjligheten att observera de tidiga galaxerna vid hittills otillgängliga våglängder. Samtidigt har radiointerfero- metrar såsom Atacama Large Millimeter/Submillimeter Array (ALMA) fram- gångsrikt observerat stoft och emissionslinjer i galaxer vid hög rödföskjutning vid längre våglängder. I denna avhandling undersöker jag bland annat en metod som kan användas för att identifiera galaxer med stort läckage av joniserande fotoner (‘escape fraction’; fesc). Denna metod använder sig av ekvivalentbredden av vätelinjen Balmer beta; (EW(Hβ)) som avges i joniserad gas i stjärnbildande galaxer, och lutningen av spektrumet vid ultravioletta (UV) våglängder för att identifiera galaxer som har höga fesc. Jag undersöker huruvida denna metod håller när den tillämpas på realistiska simulerade galaxer med variationer i intern metallhalt och stjärnbildningshistorik. Våra resultat visar att man bör kunna använda EW(Hβ) och UV-lutningen för att identifiera objekt som har fesc 50%. Stoft kan komplicera situationen något, då UV-lutningen totalt domineras av stoft även för små stoftmängder. Om man har ett betydande urval av observerade galaxer, bör fördelningen av dessa kunna användas för att få en viss förståelse för representativa stoftegenskaper. Dock kan problem uppstå om man försöker bestämma fesc för individuella objekt, då vissa stoftrecept innebär att emissionslinjer påverkas mer än kontinuum. Detta kan ‘härma’ signalen som ges av stort LyC-läckage. I sådana fall finns det en risk att man felbedömer fesc hos individuella galaxer om man inte kan korrigera för stoftet. Vi testar även antaganden om stjärnmodellering och visar att potentiella problem kan uppstå om man inte har god förståelse kring vilka stjärnmodeller som är mest representativa. Jag diskuterar även på vilka sätt stora variationer i stjärnbildningshistorik kan leda till problem när man försöker fastställa fesc via EW(Hβ) och UV- lutningen. Då emissionslinjer såsom Hβ beror på den joniserande strålningen, beror de också på mängden massiva och heta stjärnor (med mycket korta livstider), eftersom dessa dominerar den joniserande strålningen. Å andra sidan domineras UV-lutningen av något svalare och mindre massiva stjärnor med längre livstider. Detta innebär att om en galax slutar bilda stjärnor, eller bildar färre stjärnor under en period, så kommer emissionslinjerna att ‘reagera’ tidigare än UV-lutningen. Detta kan leda till situationer där EW(Hβ) avtar, medan UV-lutningen förblir densamma, vilket påminner om signaturen för stort LyC- läckage. Vi använder oss av simulerade galaxer och modeller av galaxer som har haft en stor minskning i stjärnbildning för att testa huruvida detta kan or-

76 saka problem för ovan nämnda metod och om det finns sätt att kringgå dessa problem. Våra resultat visar att galaxer som har haft en stor minskning i stjärn- bildning har signaturer som liknar dem från galaxer med stort LyC-läckage, och att EW(Hβ) och UV-lutningen inte kan användas för skilja dessa två sce- narier åt. Vi beräknar fotometriska ljusstyrkor i utvalda JWST/NIRCam- och MIRI-filter och använder klassificeringsalgoritmen ‘Linear Discriminant Ana- lysis’ (LDA) för att förstå om galaxernas spektrum innehåller information som kan användas för att särskilja galaxer med stort LyC-läckage och galaxer som har haft en stor minskning i stjärnbildning. Våra resultat visar att om minsk- ningen i stjärnbildning har pågått länge nog ( 50 miljoner år) eller är stor nog (faktor 10 – 100) så kan man klassificera 85% av galaxerna korrekt. Vi visar även att stoft kan vara ett betydande problem om man inte har en god förståelse för stoftegenskaper i tidiga galaxer. Som nämndes ovan indikerar observationer av galaxen MACS1149-JD1 att den har haft stora variationer i stjärnbildning under sin livstid. Modellering av objektet visar att galaxen kan ha bildat majoriteten av sina stjärnor för ca 200 miljoner år sedan, och att den sedan har haft en lång (ca 100-200 miljoner år) passiv period, då inga eller få stjärnor bildades. Signaturen av denna stjärn- bildningshistorik är ett stort så kallat Balmersprång; en intensitetsförändring i spektrumet vid en våglängd av ungefär 4000 Ångström. Vi jämför galaxer från olika simuleringar med denna observation för att förstå huruvida simuleringar kan reproducera Balmersprånget i detta objekt. Vi använder bland annat simuleringar som leder till galaxer med stora variationer i stjärnbildning över tid, men hittar inga objekt som har ett Balmersprång som är jämförbart med MACS1149-JD1. Våra resultat visar att de simuleringar som leder till störst variation i stjärnbildningen också leder till störst variation i Balmer- språng. Vi testar antaganden kring stoft och dess fördelning, läckage av LyC och olika antaganden kring stjärnors massfördelningar, men hittar ingen ef- fekt som rimligtvis kan leda till Balmersprång av storleken som observerades i MACS1149-JD1. Våra slutsatser är att MACS1149-JD1 antingen är ett av- vikande objekt bland tidiga galaxer, eller att våra simuleringar och modeller saknar någon ingrediens som behövs för att kunna reproducera egenskaper hos den tidiga galaxpopulationen. Framtida observationer med JWST kommer att kunna användas för att fastställa huruvida MACS1149-JD1 är en typisk galax, och därmed kommer vi eventuellt också att kunna avgöra om våra simuleringar och modeller behöver anpassas. Slutligen diskuterar jag resultat från nya ALMA-observationer av galaxen z7_GSD_3811 vid z = 7.6637. Denna galax observerades med ALMA band 6 och band 8 för att mäta styrkan på två långvågiga infraröda emissionslinjer ([O III]88μm och [C II] 158 μm) samt kontinuumemission från stoft kring 88 samt 158 μm. Varken linjer eller kontinuumemission från stoft detektera- des. Vi beräknade övre gränser för styrkor på emissionslinjer och kontinuumemission. Vi använder oss av dessa samt tidigare observationer med Hubble- teleskopet (HST) samt Very Large Telescope (VLT) vid European Southern

77 Observatory för att analysera objektet. Våra övre gränser på emissionslinjerna är konsistenta med en relativt låg metallhalt ( 0.1Z). Att stoftemissionen inte är detekterad indikerar att galaxen troligtvis innehåller relativt lite stoft. Genom att modellera spektrumet och jämföra modellen med våra övre grän- ser samt tidigare VLT- och HST-observationer visar vi att en låg metallhalt och liten stoftmängd kan förklara observationerna. Vi betonar dock att vissa modeller med höga metallhalter (∼ Z) och mycket låga stoftmängder tek- niskt sett också är konsistenta med observationerna. Vi noterar dock att det är oklart hur realistiska låga stoftmängder och höga metallhalter är, då vi förvän- tar oss ett samband mellan stoft och metallhalt. Framtida observationer med JWST och vidare observationer med ALMA kommer sannolikt ge en bättre förståelse för galaxens egenskaper.

78 12. Acknowledgements

First of all, I would like to express my deepest gratitude to my supervisors Erik Zackrisson, Sofia Ramstedt and Martin Sahlén, for always being available when I have needed help and guidance. Erik, you have been the most patient and understanding supervisor one could wish for. Thanks for always being there when I have felt confused or when I have needed advice. You have taught me so much over these years, and I count myself very lucky to have had you as a supervisor. Sofia, I am sincerely grateful that you accepted to be my co-supervisor. In addition to being a generally great person, your help and advice have truly been invaluable. Your encouragement has always come at the right time and has really helped me when I have needed it the most. Martin, thank you for always having an open door when I have needed input, advice and guidance. I have very much enjoyed our conversations, whether they be about cosmology, statistics or anything else. I would also like to thank my mentor, Johan Söderström. I really value the discussions we have had over the years, and I truly appreciate that you took on the role as my mentor. I would like to thank Akio Inoue for giving me the opportunity to come and work with you briefly in Osaka in 2018, and for giving me the opportunity to work with data from ALMA. Your input, knowledge and expertise have been so helpful. In this regard, I would also like to thank Takuya Hashimoto. Of course, a big thanks to the both of you and to Satoshi Yamanaka for making my stay in Osaka enjoyable, and for introducing me to Shabu-shabu. Thank you to the staff at the ALMA regional center in Onsala for hosting me briefly in the summer of 2019. A special thanks to Carmen Toribio, my go-to ALMA support scientist! Your help with everything related to ALMA has been invaluable, and I have really enjoyed all the conversations we have had, both in person and over Zoom. I would like to thank my colleagues at the astronomy division not only for the warm and welcoming atmosphere, but also for all the interesting lunch discussions, Monday fikas and pubs! There are so many of you at the division that I would like to thank for providing great advice, sharing knowledge/wisdom and for being amazing colleagues. My time at the division has left me with many fond memories with most of you, and I will take these with me to what- ever comes next. I am also very grateful to the people at IT and administration. Not only have you made my time as a PhD student so much easier, but I have also enjoyed our lunch- and fika-time discussions. It has been great working with you all.

79 A big thanks to all the PhD students and post-docs that I have shared my time in Uppsala with. I have really appreciated the open, accepting and supportive atmosphere among us, and I will miss our Monday PhD meetings and social events. A very special thanks to Alexis, you have been the best, most understanding and supportive ofﬁce mate ever; James, for being a great and supportive friend, for always listening to my moaning/whining; Soﬁe, for advice, beer, jokes and good times; Alvin, for fun and interesting conversations about everything from science to hiking; Lisa, for your positive attitude and being so welcoming when I was just a master student; Anton, for being a great fellow high-redshift PhD student; Sara L., Samuel, Ansgar, Terese, Matías, Miora and the new kids Axel, Emelie, Arief and Linn for contributing to the great and warm atmosphere. Also thanks to Thomas for all the beer, brewing and great and fun discussions on just about everything. I would like to thank my parents, Thomas and Waltraud, for always support- ing me, even when I left Skåne for Uppsala. Thanks to my brother Alexander, for always being there for me and for being the best brother and friend ever. A big thanks to all my Uppsala and climbing friends, especially Mikael, Måns, Jennie and Jonathan, for all the great times climbing, drinking beer, playing board games and for being great friends. A very special thanks to Stefan, for all the long walks, for the support, encouragement and for all the interesting discussions about statistics, physics, music, programming, beer and everything under the sun. You are an amazing friend. Thank you. Finally, I would like to thank Linnéa, to whom this thesis is dedicated. I don’t think anything I could write here would make you justice. You are the most supportive and amazing partner one could wish for, and this thesis would probably not exist if it wasn’t for you. Thank you.

80 References

Abel, N. P., Ferland, G. J., Shaw, G., & van Hoof, P. A. M. 2005, ApJS, 161, 65 Allen, M. G., Groves, B. A., Dopita, M. A., Sutherland, R. S., & Kewley, L. J. 2008, ApJS, 178, 20 Alpher, R. A., Bethe, H., & Gamow, G. 1948, Physical Review, 73, 803 Arata, S., Yajima, H., Nagamine, K., Abe, M., & Khochfar, S. 2020, MNRAS, 498, 5541 Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009, ARA&A, 47, 481 Bañados, E., Venemans, B. P., Mazzucchelli, C., et al. 2018, Nature, 553, 473 Bakx, T. J. L. C., Tamura, Y., Hashimoto, T., et al. 2020, MNRAS Barkana, R. & Loeb, A. 2001, Phys. Rep., 349, 125 Barnett, R., Warren, S. J., Becker, G. D., et al. 2017, A&A, 601, A16 Barrow, K. S. S., Robertson, B. E., Ellis, R. S., et al. 2020, ApJ, 902, L39 Becker, G. D. & Bolton, J. S. 2013, MNRAS, 436, 1023 Becker, G. D., Bolton, J. S., Madau, P., et al. 2015, MNRAS, 447, 3402 Behrens, C., Pallottini, A., Ferrara, A., Gallerani, S., & Vallini, L. 2018, MNRAS, 477, 552 Bergvall, N., Marquart, T., Way, M. J., et al. 2016, A&A, 587, A72 Bergvall, N., Zackrisson, E., Andersson, B.-G., et al. 2006, A&A, 448, 513 Bian, F., Fan, X., McGreer, I., Cai, Z., & Jiang, L. 2017, ApJ, 837, L12 Binette, L., Dopita, M. A., & Tuohy, I. R. 1985, ApJ, 297, 476 Boquien, M., Burgarella, D., Roehlly, Y., et al. 2019, A&A, 622, A103 Borthakur, S., Heckman, T. M., Leitherer, C., & Overzier, R. A. 2014, Science, 346, 216 Bosman, S. E. I., Fan, X., Jiang, L., et al. 2018, MNRAS, 479, 1055 Boutsia, K., Grazian, A., Giallongo, E., Fiore, F., & Civano, F. 2018, ApJ, 869, 20 Boutsia, K., Grazian, A., Giallongo, E., et al. 2011, ApJ, 736, 41 Bouwens, R. J., Aravena, M., Decarli, R., et al. 2016, ApJ, 833, 72 Bouwens, R. J., Illingworth, G. D., Oesch, P. A., et al. 2014, ApJ, 793, 115 Bouwens, R. J., Smit, R., Labbé, I., et al. 2016, ApJ, 831, 176 Bowler, R. A. A., Bourne, N., Dunlop, J. S., McLure, R. J., & McLeod, D. J. 2018, MNRAS, 481, 1631 Bowman, J. D., Rogers, A. E. E., Monsalve, R. A., Mozdzen, T. J., & Mahesh, N. 2018, Nature, 555, 67 Bradac,ˇ M., Garcia-Appadoo, D., Huang, K.-H., et al. 2017, ApJ, 836, L2 Briggs, D. S., Schwab, F. R., & Sramek, R. A. 1999, in Astronomical Society of the Paciﬁc Conference Series, Vol. 180, Synthesis Imaging in Radio Astronomy II, ed. G. B. Taylor, C. L. Carilli, & R. A. Perley, 127 Bromm, V. 2013, Reports on Progress in Physics, 76, 112901 Bruzual, G. & Charlot, S. 2003, MNRAS, 344, 1000 Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682

81 Calzetti, D., Kinney, A. L., & Storchi-Bergmann, T. 1994, ApJ, 429, 582 Carnall, A. C., McLure, R. J., Dunlop, J. S., & Davé, R. 2018, MNRAS, 480, 4379 Carniani, S., Ferrara, A., Maiolino, R., et al. 2020, MNRAS, 499, 5136 Carniani, S., Maiolino, R., Amorin, R., et al. 2018, MNRAS, 478, 1170 Carniani, S., Maiolino, R., Pallottini, A., et al. 2017, A&A, 605, A42 Casey, C. M. 2012, MNRAS, 425, 3094 Ceverino, D., Glover, S. C. O., & Klessen, R. S. 2017, MNRAS, 470, 2791 Ceverino, D., Klessen, R. S., & Glover, S. C. O. 2018, MNRAS, 480, 4842 Ceverino, D., Klessen, R. S., & Glover, S. C. O. 2019, MNRAS, 484, 1366 Charlot, S. & Fall, S. M. 2000, ApJ, 539, 718 Chen, H.-W., Prochaska, J. X., & Gnedin, N. Y. 2007, ApJ, 667, L125 Chevance, M., Kruijssen, J. M. D., Vazquez-Semadeni, E., et al. 2020, Space Sci. Rev., 216, 50 Chisholm, J., Gazagnes, S., Schaerer, D., et al. 2018, A&A, 616, A30 Choi, Y., Dalcanton, J. J., Williams, B. F., et al. 2020, ApJ, 902, 54 Conroy, C., Gunn, J. E., & White, M. 2009, ApJ, 699, 486 Cowie, L. L., Barger, A. J., Bauer, F. E., & González-López, J. 2020, ApJ, 891, 69 Cowie, L. L., Barger, A. J., & Trouille, L. 2009, ApJ, 692, 1476 Cristiani, S., Serrano, L. M., Fontanot, F., Vanzella, E., & Monaco, P. 2016, MNRAS, 462, 2478 Davies, F. B., Hennawi, J. F., Bañados, E., et al. 2018, ApJ, 864, 142 Dayal, P., Choudhury, T. R., Bromm, V., & Pacucci, F. 2017, ApJ, 836, 16 Dayal, P. & Ferrara, A. 2012, MNRAS, 421, 2568 Dayal, P. & Ferrara, A. 2018, Phys. Rep., 780, 1 Dayal, P., Volonteri, M., Choudhury, T. R., et al. 2020, MNRAS, 495, 3065 De Breuck, C., Neri, R., Morganti, R., et al. 2003, A&A, 401, 911 De Looze, I., Cormier, D., Lebouteiller, V., et al. 2014, A&A, 568, A62 Douna, V. M., Pellizza, L. J., Laurent, P., & Mirabel, I. F. 2018, MNRAS, 474, 3488 Draine, B. T. 2011, Physics of the Interstellar and Intergalactic Medium Eilers, A.-C., Davies, F. B., & Hennawi, J. F. 2018, ApJ, 864, 53 Eisenstein, D. J., Hu, W., & Tegmark, M. 1999, ApJ, 518, 2 Eldridge, J. J. & Stanway, E. R. 2009, MNRAS, 400, 1019 Erb, D. K., Steidel, C. C., Shapley, A. E., et al. 2006, ApJ, 647, 128 Fan, X., Carilli, C. L., & Keating, B. 2006a, ARA&A, 44, 415 Fan, X., Hennawi, J. F., Richards, G. T., et al. 2004, AJ, 128, 515 Fan, X., Strauss, M. A., Becker, R. H., et al. 2006b, AJ, 132, 117 Faucher-Giguère, C.-A., Lidz, A., Hernquist, L., & Zaldarriaga, M. 2008, ApJ, 688, 85 Ferland, G. J., Korista, K. T., Verner, D. A., et al. 1998, PASP, 110, 761 Ferland, G. J., Porter, R. L., van Hoof, P. A. M., et al. 2013, Revista Mexicana de Astronomia y Astroﬁsica, 49, 137 Ferrara, A. & Loeb, A. 2013, MNRAS, 431, 2826 Ferrara, A., Vallini, L., Pallottini, A., et al. 2019, MNRAS, 489, 1 Finkelstein, S. L., D’Aloisio, A., Paardekooper, J.-P., et al. 2019, ApJ, 879, 36 Finkelstein, S. L., Papovich, C., Ryan, R. E., et al. 2012a, ApJ, 758, 93 Finkelstein, S. L., Papovich, C., Salmon, B., et al. 2012b, ApJ, 756, 164 Finlator, K., Davé, R., Papovich, C., & Hernquist, L. 2006, ApJ, 639, 672

82 Finlator, K., Muñoz, J. A., Oppenheimer, B. D., et al. 2013, MNRAS, 436, 1818 Finlator, K., Oppenheimer, B. D., & Davé, R. 2011, MNRAS, 410, 1703 Fioc, M. & Rocca-Volmerange, B. 1997, A&A, 500, 507 Fioc, M. & Rocca-Volmerange, B. 2019, A&A, 623, A143 Fletcher, T. J., Tang, M., Robertson, B. E., et al. 2019, ApJ, 878, 87 Fudamoto, Y., Oesch, P. A., Faisst, A., et al. 2020, A&A, 643, A4 Furlanetto, S. R., Zaldarriaga, M., & Hernquist, L. 2004, ApJ, 613, 1 Furusawa, H., Kashikawa, N., Kobayashi, M. A. R., et al. 2016, ApJ, 822, 46 Gazagnes, S., Chisholm, J., Schaerer, D., et al. 2018, A&A, 616, A29 Giallongo, E., Grazian, A., Fiore, F., et al. 2015, A&A, 578, A83 Giallongo, E., Grazian, A., Fiore, F., et al. 2019, ApJ, 884, 19 Giri, S. K., Zackrisson, E., Binggeli, C., Pelckmans, K., & Cubo, R. 2020, MNRAS, 491, 5277 Gnedin, N. Y. 2014, ApJ, 793, 29 Gnedin, N. Y. & Kaurov, A. A. 2014, ApJ, 793, 30 Grazian, A., Giallongo, E., Boutsia, K., et al. 2018, A&A, 613, A44 Grazian, A., Giallongo, E., Fiore, F., et al. 2020, ApJ, 897, 94 Grazian, A., Giallongo, E., Gerbasi, R., et al. 2016, A&A, 585, A48 Grazian, A., Giallongo, E., Paris, D., et al. 2017, A&A, 602, A18 Greig, B., Mesinger, A., & Bañados, E. 2019, MNRAS, 484, 5094 Greig, B., Mesinger, A., Haiman, Z., & Simcoe, R. A. 2017, MNRAS, 466, 4239 Grogin, N. A., Kocevski, D. D., Faber, S. M., et al. 2011, ApJS, 197, 35 Gunn, J. E. & Peterson, B. A. 1965, ApJ, 142, 1633 Guth, A. H. 1981, Phys. Rev. D, 23, 347 Guth, A. H. & Kaiser, D. I. 2005, Science, 307, 884 Haiman, Z. & Knox, L. 1999, in Astronomical Society of the Paciﬁc Conference Series, Vol. 181, Microwave Foregrounds, ed. A. de Oliveira-Costa & M. Tegmark, 227 Harikane, Y., Ouchi, M., Inoue, A. K., et al. 2020, ApJ, 896, 93 Harikane, Y., Ouchi, M., Shibuya, T., et al. 2018, ApJ, 859, 84 Hashimoto, T., Inoue, A. K., Mawatari, K., et al. 2019, PASJ, 71, 71 Hashimoto, T., Laporte, N., Mawatari, K., et al. 2018, Nature, 557, 392 Hassan, S., Davé, R., Mitra, S., et al. 2018, MNRAS, 473, 227 Hayes, M., Schaerer, D., Östlin, G., et al. 2011, ApJ, 730, 8 Hayes, M., Östlin, G., Schaerer, D., et al. 2010, Nature, 464, 562 Hoag, A., Bradac,ˇ M., Huang, K., et al. 2019, ApJ, 878, 12 Högbom, J. A. 1974, A&AS, 15, 417 Hopkins, P. F. 2015, MNRAS, 450, 53 Hopkins, P. F., Kereš, D., Oñorbe, J., et al. 2014, MNRAS, 445, 581 Hutter, A., Dayal, P., & Müller, V. 2015, MNRAS, 450, 4025 Hutter, A., Dayal, P., Partl, A. M., & Müller, V. 2014, MNRAS, 441, 2861 Iliev, I. T., Mellema, G., Pen, U. L., et al. 2006, MNRAS, 369, 1625 Inoue, A. K. 2001, AJ, 122, 1788 Inoue, A. K., Hashimoto, T., Chihara, H., & Koike, C. 2020, MNRAS, 495, 1577 Inoue, A. K., Hirashita, H., & Kamaya, H. 2001, ApJ, 555, 613 Inoue, A. K. & Iwata, I. 2008, MNRAS, 387, 1681 Inoue, A. K., Iwata, I., & Deharveng, J.-M. 2006, MNRAS, 371, L1

83 Inoue, A. K., Shimizu, I., Tamura, Y., et al. 2014, ApJ, 780, L18 Inoue, A. K., Tamura, Y., Matsuo, H., et al. 2016, Science, 352, 1559 Izotov, Y. I., Orlitová, I., Schaerer, D., et al. 2016a, Nature, 529, 178 Izotov, Y. I., Schaerer, D., Thuan, T. X., et al. 2016b, MNRAS, 461, 3683 Izotov, Y. I., Schaerer, D., Worseck, G., et al. 2018a, MNRAS, 474, 4514 Izotov, Y. I., Worseck, G., Schaerer, D., et al. 2018b, MNRAS, 478, 4851 Jaacks, J., Nagamine, K., & Choi, J. H. 2012, MNRAS, 427, 403 Japelj, J., Vanzella, E., Fontanot, F., et al. 2017, MNRAS, 468, 389 Jaskot, A. E. & Oey, M. S. 2013, ApJ, 766, 91 Jensen, H., Zackrisson, E., Pelckmans, K., et al. 2016, ApJ, 827, 5 Jiang, L., Kashikawa, N., Wang, S., et al. 2020, Nature Astronomy Jones, T. A., Ellis, R. S., Schenker, M. A., & Stark, D. P. 2013, ApJ, 779, 52 Kakiichi, K., Ellis, R. S., Laporte, N., et al. 2018, MNRAS, 479, 43 Kanekar, N., Wagg, J., Chary, R. R., & Carilli, C. L. 2013, ApJ, 771, L20 Kashikawa, N., Ishizaki, Y., Willott, C. J., et al. 2015, ApJ, 798, 28 Kashikawa, N., Shimasaku, K., Matsuda, Y., et al. 2011, ApJ, 734, 119 Katz, H., Galligan, T. P., Kimm, T., et al. 2019a, MNRAS, 487, 5902 Katz, H., Laporte, N., Ellis, R. S., Devriendt, J., & Slyz, A. 2019b, MNRAS, 484, 4054 Katz, H., Durovˇ cíková,ˇ D., Kimm, T., et al. 2020, MNRAS, 498, 164 Kennicutt, Robert C., J. 1998, ARA&A, 36, 189 Kennicutt, R. C. & Evans, N. J. 2012, ARA&A, 50, 531 Khakhaleva-Li, Z. & Gnedin, N. Y. 2016, ApJ, 820, 133 Kim, Y., Im, M., Jeon, Y., et al. 2020, ApJ, 904, 111 Kimm, T. & Cen, R. 2014, ApJ, 788, 121 Kimm, T., Cen, R., Devriendt, J., Dubois, Y., & Slyz, A. 2015, MNRAS, 451, 2900 Knudsen, K. K., Richard, J., Kneib, J.-P., et al. 2016, MNRAS, 462, L6 Koekemoer, A. M., Faber, S. M., Ferguson, H. C., et al. 2011, ApJS, 197, 36 Konno, A., Ouchi, M., Ono, Y., et al. 2014, ApJ, 797, 16 Konno, A., Ouchi, M., Shibuya, T., et al. 2018, PASJ, 70, S16 Koprowski, M. P., Coppin, K. E. K., Geach, J. E., et al. 2018, MNRAS, 479, 4355 Kravtsov, A. V. 1999, Ph.D. Thesis, 5564 Kravtsov, A. V., Klypin, A., & Hoffman, Y. 2002, ApJ, 571, 563 Kravtsov, A. V., Klypin, A. A., & Khokhlov, A. M. 1997, ApJS, 111, 73 Kroupa, P. 2001, MNRAS, 322, 231 Kuhlen, M., Vogelsberger, M., & Angulo, R. 2012, Physics of the Dark Universe, 1, 50 Kulkarni, G., Keating, L. C., Haehnelt, M. G., et al. 2019a, MNRAS, 485, L24 Kulkarni, G., Worseck, G., & Hennawi, J. F. 2019b, MNRAS, 488, 1035 Laporte, N., Ellis, R. S., Boone, F., et al. 2017, ApJ, 837, L21 Laporte, N., Katz, H., Ellis, R. S., et al. 2019, MNRAS, 487, L81 Leethochawalit, N., Jones, T. A., Ellis, R. S., Stark, D. P., & Zitrin, A. 2016, ApJ, 831, 152 Leitet, E., Bergvall, N., Hayes, M., Linné, S., & Zackrisson, E. 2013, A&A, 553, A106 Leitherer, C., Hernandez, S., Lee, J. C., & Oey, M. S. 2016, ApJ, 823, 64

84 Leitherer, C., Schaerer, D., Goldader, J. D., et al. 1999, ApJ Supplement Series, 123, 3 Liu, C., Mutch, S. J., Angel, P. W., et al. 2016, MNRAS, 462, 235 Ma, X., Hayward, C. C., Casey, C. M., et al. 2019, MNRAS, 487, 1844 Ma, X., Hopkins, P. F., Garrison-Kimmel, S., et al. 2018, MNRAS, 478, 1694 Ma, X., Kasen, D., Hopkins, P. F., et al. 2015, MNRAS, 453, 960 Madau, P. & Dickinson, M. 2014, ARA&A, 52, 415 Madau, P. & Fragos, T. 2017, ApJ, 840, 39 Maiolino, R., Carniani, S., Fontana, A., et al. 2015, MNRAS, 452, 54 Maiolino, R. & Mannucci, F. 2019, A&A Rev., 27, 3 Mancini, M., Schneider, R., Graziani, L., et al. 2016, MNRAS, 462, 3130 Marti-Vidal, I. 2017, arXiv e-prints, arXiv:1706.00936 Mas-Ribas, L., Hennawi, J. F., Dijkstra, M., et al. 2017, ApJ, 846, 11 Mason, C. A., Treu, T., Dijkstra, M., et al. 2018, ApJ, 856, 2 Masters, D., Capak, P., Salvato, M., et al. 2012, ApJ, 755, 169 Matthee, J., Sobral, D., Best, P., et al. 2017, MNRAS, 465, 3637 Matthee, J., Sobral, D., Boogaard, L. A., et al. 2019, ApJ, 881, 124 Matthee, J., Sobral, D., Boone, F., et al. 2017, ApJ, 851, 145 Mawatari, K., Inoue, A. K., Hashimoto, T., et al. 2020a, ApJ, 889, 137 Mawatari, K., Inoue, A. K., Yamanaka, S., Hashimoto, T., & Tamura, Y. 2020b, in Panchromatic Modelling with Next Generation Facilities, ed. M. Boquien, E. Lusso, C. Gruppioni, & P. Tissera, Vol. 341, 285–286 McGreer, I. D., Fan, X., Jiang, L., & Cai, Z. 2018, AJ, 155, 131 McGreer, I. D., Mesinger, A., & D’Odorico, V. 2015, MNRAS, 447, 499 Mesinger, A. 2010, MNRAS, 407, 1328 Meurer, G. R., Heckman, T. M., & Calzetti, D. 1999, ApJ, 521, 64 Micheva, G., Iwata, I., & Inoue, A. K. 2017, MNRAS, 465, 302 Micheva, G., Iwata, I., Inoue, A. K., et al. 2017, MNRAS, 465, 316 Miralda-Escudé, J. 1998, ApJ, 501, 15 Mitra, S., Choudhury, T. R., & Ferrara, A. 2018, MNRAS, 473, 1416 Mondal, R., Fialkov, A., Fling, C., et al. 2020, MNRAS, 498, 4178 Mortlock, D. J., Warren, S. J., Venemans, B. P., et al. 2011, Nature, 474, 616 Mostardi, R. E., Shapley, A. E., Steidel, C. C., et al. 2015, ApJ, 810, 107 Nagao, T., Maiolino, R., Marconi, A., & Matsuhara, H. 2011, A&A, 526, A149 Naidu, R. P., Forrest, B., Oesch, P. A., Tran, K.-V. H., & Holden, B. P. 2018, MNRAS Naidu, R. P., Oesch, P. A., Reddy, N., et al. 2017, ApJ, 847, 12 Naidu, R. P., Tacchella, S., Mason, C. A., et al. 2020, ApJ, 892, 109 Nakajima, K., Ellis, R. S., Robertson, B. E., Tang, M., & Stark, D. P. 2020, ApJ, 889, 161 Oesch, P. A., Brammer, G., van Dokkum, P. G., et al. 2016, ApJ, 819, 129 Ono, Y., Ouchi, M., Shimasaku, K., et al. 2010, ApJ, 724, 1524 Onoue, M., Kashikawa, N., Willott, C. J., et al. 2017, ApJL, 847, L15 Ota, K., Iye, M., Kashikawa, N., et al. 2017, ApJ, 844, 85 Ota, K., Walter, F., Ohta, K., et al. 2014, ApJ, 792, 34 Ouchi, M., Ellis, R., Ono, Y., et al. 2013, ApJ, 778, 102 Ouchi, M., Harikane, Y., Shibuya, T., et al. 2018, PASJ, 70, S13 Ouchi, M., Shimasaku, K., Furusawa, H., et al. 2010, ApJ, 723, 869

85 Paardekooper, J.-P., Khochfar, S., & Dalla Vecchia, C. 2013, MNRAS, 429, L94 Paardekooper, J.-P., Khochfar, S., & Dalla Vecchia, C. 2015, MNRAS, 451, 2544 Pallottini, A., Ferrara, A., Decataldo, D., et al. 2019, MNRAS, 487, 1689 Parsa, S., Dunlop, J. S., & McLure, R. J. 2018, MNRAS, 474, 2904 Patel, M., Warren, S. J., Mortlock, D. J., & Fynbo, J. P. U. 2010, A&A, 512, L3 Peebles, P. J. E. 1968, ApJ, 153, 1 Pei, Y. C. 1992, ApJ, 395, 130 Pentericci, L., Carniani, S., Castellano, M., et al. 2016, ApJ, 829, L11 Pentericci, L., Vanzella, E., Fontana, A., et al. 2014, ApJ, 793, 113 Planck Collaboration, Adam, R., Aghanim, N., et al. 2016, A&A, 596, A108 Planck Collaboration, Aghanim, N., Akrami, Y., et al. 2020, A&A, 641, A6 Pritchard, J. R. & Loeb, A. 2012, Reports on Progress in Physics, 75, 086901 Puchwein, E., Haardt, F., Haehnelt, M. G., & Madau, P. 2019, MNRAS, 485, 47 Qin, Y., Mutch, S. J., Poole, G. B., et al. 2017, MNRAS, 472, 2009 Raiter, A., Schaerer, D., & Fosbury, R. A. E. 2010, A&A, 523, A64 Rauch, M. 1998, ARA&A, 36, 267 Razoumov, A. O. & Sommer-Larsen, J. 2010, ApJ, 710, 1239 Rees, M. J. 1998, Proceedings of the National Academy of Science, 95, 47 Remijan, A., Biggs, A., Cortes, P., et al. 2020, ALMA Technical Handbook, ALMA Doc. 8.3, ver. 1.0 Ricci, F., Marchesi, S., Shankar, F., La Franca, F., & Civano, F. 2017, MNRAS, 465, 1915 Robertson, B. E., Ellis, R. S., Furlanetto, S. R., & Dunlop, J. S. 2015, ApJ, 802, L19 Rogers, A. B., McLure, R. J., Dunlop, J. S., et al. 2014, MNRAS, 440, 3714 Romano, M., Grazian, A., Giallongo, E., et al. 2019, A&A, 632, A45 Rutkowski, M. J., Scarlata, C., Henry, A., et al. 2017, ApJL, 841, L27 Salmon, B., Papovich, C., Finkelstein, S. L., et al. 2015, ApJ, 799, 183 Schaerer, D., Boone, F., Zamojski, M., et al. 2015, A&A, 574, A19 Schaerer, D., Ginolﬁ, M., Béthermin, M., et al. 2020, A&A, 643, A3 Schenker, M. A., Ellis, R. S., Konidaris, N. P., & Stark, D. P. 2014, ApJ, 795, 20 Shapley, A. E., Steidel, C. C., Strom, A. L., et al. 2016, ApJL, 826, L24 Sharma, M., Theuns, T., Frenk, C., et al. 2016, MNRAS, 458, L94 Shen, X., Hopkins, P. F., Faucher-Giguère, C.-A., et al. 2020, MNRAS, 495, 3252 Shimizu, I., Inoue, A. K., Okamoto, T., & Yoshida, N. 2014, MNRAS, 440, 731 Shimizu, I., Inoue, A. K., Okamoto, T., & Yoshida, N. 2016, MNRAS, 461, 3563 Shin, S., Im, M., Kim, Y., et al. 2020, ApJ, 893, 45 Siana, B., Teplitz, H. I., Colbert, J., et al. 2007, ApJ, 668, 62 Siana, B., Teplitz, H. I., Ferguson, H. C., et al. 2010, ApJ, 723, 241 Smit, R., Bouwens, R. J., Carniani, S., et al. 2018, Nature, 553, 178 Sobacchi, E. & Mesinger, A. 2015, MNRAS, 453, 1843 Song, M., Finkelstein, S. L., Livermore, R. C., et al. 2016, ApJ, 826, 113 Songaila, A. & Cowie, L. L. 2002, AJ, 123, 2183 Springel, V. 2005, MNRAS, 364, 1105 Springel, V. 2010, MNRAS, 401, 791 Springel, V., Yoshida, N., & White, S. D. M. 2001, New Astronomy, 6, 79 Stanway, E. R., Eldridge, J. J., & Becker, G. D. 2016, MNRAS, 456, 485 Stark, D. P. 2016, ARA&A, 54, 761

86 Stark, D. P., Walth, G., Charlot, S., et al. 2015, MNRAS, 454, 1393 Stasinska,´ G., Izotov, Y., Morisset, C., & Guseva, N. 2015, A&A, 576, A83 Stefanon, M., Bouwens, R. J., Labbé, I., et al. 2021, arXiv e-prints, arXiv:2103.06279 Steidel, C. C., Bogosavljevic,´ M., Shapley, A. E., et al. 2018, ApJ, 869, 123 Strömgren, B. 1939, ApJ, 89, 526 Sun, G. & Furlanetto, S. R. 2016, MNRAS, 460, 417 Sutherland, R. S. & Dopita, M. A. 1993, ApJS, 88, 253 Sutherland, R. S. & Dopita, M. A. 2017, ApJS, 229, 34 Tamura, Y., Mawatari, K., Hashimoto, T., et al. 2019, ApJ, 874, 27 Tanaka, T., Perna, R., & Haiman, Z. 2012, MNRAS, 425, 2974 Tanvir, N. R., Fynbo, J. P. U., de Ugarte Postigo, A., et al. 2019, MNRAS, 483, 5380 Thompson, A. R., Moran, J. M., & Swenson, George W., J. 2017, Interferometry and Synthesis in Radio Astronomy, 3rd Edition Totani, T., Aoki, K., Hattori, T., & Kawai, N. 2016, PASJ, 68, 15 Totani, T., Kawai, N., Kosugi, G., et al. 2006, PASJ, 58, 485 Trebitsch, M., Blaizot, J., Rosdahl, J., Devriendt, J., & Slyz, A. 2017, MNRAS, 470, 224 Vallini, L., Gallerani, S., Ferrara, A., Pallottini, A., & Yue, B. 2015, ApJ, 813, 36 van Cittert, P. 1934, Physica, 1, 201 Vanzella, E., de Barros, S., Vasei, K., et al. 2016, ApJ, 825, 41 Vanzella, E., Guo, Y., Giavalisco, M., et al. 2012, ApJ, 751, 70 Vanzella, E., Nonino, M., Cupani, G., et al. 2018, MNRAS, 476, L15 Vasei, K., Siana, B., Shapley, A. E., et al. 2016, ApJ, 831, 38 Vázquez, G. A. & Leitherer, C. 2005, ApJ, 621, 695 Verhamme, A., Orlitová, I., Schaerer, D., & Hayes, M. 2015, A&A, 578, A7 Verhamme, A., Orlitová, I., Schaerer, D., et al. 2017, A&A, 597, A13 Vielfaure, J. B., Vergani, S. D., Japelj, J., et al. 2020, A&A, 641, A30 Wang, F., Davies, F. B., Yang, J., et al. 2020, ApJ, 896, 23 Watson, D., Christensen, L., Knudsen, K. K., et al. 2015, Nature, 519, 327 Wiklind, T., Dickinson, M., Ferguson, H. C., et al. 2008, ApJ, 676, 781 Wilkins, S. M., Bouwens, R. J., Oesch, P. A., et al. 2016, MNRAS, 455, 659 Wilkins, S. M., Feng, Y., Di Matteo, T., et al. 2018, MNRAS, 473, 5363 Wilkins, S. M., Feng, Y., Di Matteo, T., et al. 2017, MNRAS, 469, 2517 Willott, C. J., Delorme, P., Omont, A., et al. 2007, AJ, 134, 2435 Willott, C. J., Delorme, P., Reylé, C., et al. 2010, AJ, 139, 906 Wilson, T. L., Rohlfs, K., & Hüttemeister, S. 2009, Tools of Radio Astronomy Wise, J. H., Demchenko, V. G., Halicek, M. T., et al. 2014, MNRAS, 442, 2560 Yajima, H., Nagamine, K., Zhu, Q., Khochfar, S., & Dalla Vecchia, C. 2017, ApJ, 846, 30 Zackrisson, E., Inoue, A. K., & Jensen, H. 2013, ApJ, 777, 39 Zackrisson, E., Majumdar, S., Mondal, R., et al. 2020, MNRAS, 493, 855 Zackrisson, E., Rydberg, C.-E., Schaerer, D., Östlin, G., & Tuli, M. 2011, ApJ, 740, 13 Zahn, O., Reichardt, C. L., Shaw, L., et al. 2012, ApJ, 756, 65 Zaldarriaga, M., Spergel, D. N., & Seljak, U. 1997, ApJ, 488, 1 Zernike, F. 1938, Physica, 5, 785 Zheng, Z.-Y., Wang, J., Rhoads, J., et al. 2017, ApJ, 842, L22

87 Acta Universitatis Upsaliensis Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology 2036 Editor: The Dean of the Faculty of Science and Technology

A doctoral dissertation from the Faculty of Science and Technology, Uppsala University, is usually a summary of a number of papers. A few copies of the complete dissertation are kept at major Swedish research libraries, while the summary alone is distributed internationally through the series Digital Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology. (Prior to January, 2005, the series was published under the title “Comprehensive Summaries of Uppsala Dissertations from the Faculty of Science and Technology”.)

ACTA UNIVERSITATIS UPSALIENSIS Distribution: publications.uu.se UPPSALA urn:nbn:se:uu:diva-440032 2021