The Large Scale Structures: A Window On The Dark Components of The Universe

Springer Theses
Recognizing Outstanding Ph.D. Research
Stéphane Ilić
The Large Scale

Structures
A Window on the Dark
Components of the Universe
Springer Theses
Recognizing Outstanding Ph.D. Research
For further volumes:

http://www.springer.com/series/8790
Aims and Scope
The series ‘‘Springer Theses’’ brings together a selection of the very best Ph.D.
theses from around the world and across the physical sciences. Nominated and
endorsed by two recognized specialists, each published volume has been selected
for its scientific excellence and the high impact of its contents for the pertinent
field of research. For greater accessibility to non-specialists, the published versions
include an extended introduction, as well as a foreword by the student’s supervisor
explaining the special relevance of the work for the field. As a whole, the series
will provide a valuable resource both for newcomers to the research fields
described, and for other scientists seeking detailed background information on
special questions. Finally, it provides an accredited documentation of the valuable
contributions made by today’s younger generation of scientists.
Theses are accepted into the series by invited nomination only

and must fulfill all of the following criteria
• They must be written in good English.

• The topic should fall within the confines of Chemistry, Physics, Earth Sciences,
Engineering and related interdisciplinary fields such as Materials, Nanoscience,
Chemical Engineering, Complex Systems and Biophysics.
• The work reported in the thesis must represent a significant scientific advance.
• If the thesis includes previously published material, permission to reproduce this
must be gained from the respective copyright holder.
• They must have been examined and passed during the 12 months prior to
nomination.
• Each thesis should include a foreword by the supervisor outlining the signifi-
cance of its content.
• The theses should have a clearly defined structure including an introduction
accessible to scientists not expert in that particular field.
Stéphane Ilić
The Large Scale Structures

A Window on the Dark Components
of the Universe
Doctoral Thesis accepted by

the University of Paris-Sud, Orsay Cedex, France
123
Author Supervisor
Dr. Stéphane Ilić Prof. Mathieu Langer
Département de Physique Département de Physique
Institut d’Astrophysique Spatiale Institut d’Astrophysique Spatiale
Université Paris-Sud Université Paris-Sud
Orsay Cedex Orsay Cedex
France France
ISSN 2190-5053 ISSN 2190-5061 (electronic)

ISBN 978-3-319-07745-1 ISBN 978-3-319-07746-8 (eBook)
DOI 10.1007/978-3-319-07746-8
Springer Cham Heidelberg New York Dordrecht London
Library of Congress Control Number: 2014940758
Springer International Publishing Switzerland 2014

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief
excerpts in connection with reviews or scholarly analysis or material supplied specifically for the
purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the
work. Duplication of this publication or parts thereof is permitted only under the provisions of
the Copyright Law of the Publisher’s location, in its current version, and permission for use must
always be obtained from Springer. Permissions for use may be obtained through RightsLink at the
Copyright Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

‘‘When I reach for the edge of the universe, I do so
knowing that along some paths of cosmic discovery,
there are times when, at least for now, one must be
content to love the questions themselves.’’
Neil deGrasse Tyson

Dedicated to the ones no longer with us
to gaze upon this beautiful Universe
Supervisor’s Foreword
Modern cosmology is a young science. In less than a century of existence, it has

witnessed many twists and turns that have led to a tremendous global progress.
The parallel development of technological advances, observational efforts and
theoretical modelling has allowed the establishment of a standard cosmological
model, called LCDM. It is striking that we name the cosmological model after its
two most mysterious, but dominant, components. Indeed, on the one hand, CDM
stands for Cold Dark Matter, the existence of which we detect indirectly through
its many gravitational effects, but of which we do not know the intimate nature.
We know nevertheless that it makes up slightly more than a quarter of the total
energy budget of the Universe. Intensive experimental searches are dedicated to its
production in particle colliders, to its direct detection in underground laboratories
and indirect detection through astrophysical signatures. On the other hand, rep-
resenting 68 % of the energy budget, L stands for the famous cosmological
constant which is thought to be responsible for the accelerated expansion of the
Universe, evidenced in the late 1990s and crowned by the Nobel Prize in Physics
in 2011. Determining the intimate nature of the Dark Components of the Universe
is one of the most enthralling challenges of current cosmology. Many different
approaches have to be explored to pin down the exact amount and the properties of
these components. One such fruitful approach takes advantage of the structuration
of matter on large scales, affected both by Dark Energy and Dark Matter.
On the Dark Energy side, the measurement of the so-called integrated Sachs–
Wolfe (iSW) effect provides a way of testing both the cosmological model and the
properties of Dark Energy. It is essentially the net gain of energy that photons of
the Cosmic Microwave Background (CMB) acquire while travelling across time-
varying gravitational potentials. Before this thesis, the observational situation was
somewhat unclear: many different studies, in some cases using the same data,
reported iSW detections with statistical significances ranging from negligible to
above four standard deviations. To clarify the picture, in his thesis Stéphane Ilić
has adopted a strategy in three steps: (i) he developed the formalism and tools for
the cross-correlation of the Cosmic Infrared Background, tracing the gravitational
potentials underlying the distribution of galaxies, with the CMB, and showed
rigorously that the iSW is thus detectable, in principle, with a very high statistical
ix
x Supervisor’s Foreword
significance; (ii) he devised a complete stacking strategy to detect the iSW

imprinted in the CMB by the largest cosmic superstructures (super-clusters and
super-voids), including a thorough examination of statistical and selection effects,
and applied it successfully to both WMAP and Planck CMB data; (iii) he computed
the exact iSW effect expected from such superstructures within the framework of
General Relativity and the LCDM model, and showed that any possible discrep-
ancy reported so far between data and theory might be due rather to our poorer
understanding of the mildly nonlinear regime of structure formation than to a
failure of the LCDM model itself.
On the Dark Matter side, it is well known that the annihilation of Weakly
Interacting Massive Particles and the decay of light exotic particles may have left a
subtle imprint on the CMB, especially in its E-mode polarisation. On the one hand,
the related injection of energy into the Intergalactic Medium modifies the ioni-
sation state of the baryonic gas. This will be very difficult, if not impossible, to
detect, as the related signatures in the CMB appear on the largest scales where
uncertainties due to Cosmic Variance dominate. On the other hand, the thermal
state of the intergalactic gas is affected as well, in a potentially more significant
way. Stéphane showed how measurements of the intergalactic temperature can be
exploited to learn more about Dark Matter. Through an analysis of recent data
extracted from the Lyman alpha forest, Stéphane showed that powerful constraints
can be put both on the Dark Matter particle properties and on the redshift and
duration of the Second Reionisation of Helium.
Research on the exploitation of the distribution of matter in the Universe on
large scales is rapidly evolving, on both sides of theory and observation. New
results will be soon within reach, but this thesis sets a standard in the rigour and
thoroughness needed to understand properly what the large-scale structures tell us
about the Dark Sector of the cosmological model. It has been a great pleasure for
Marian Douspis and myself to guide Stéphane Ilić in this doctoral project, and his
results are a source of real satisfaction.
Orsay Cedex, April 2014 Prof. Mathieu Langer

Abstract
Dark Energy is one of the great mysteries of modern cosmology. It is an unknown

component supposedly filling the whole universe and responsible for the current
acceleration of the expansion of the Universe. Its study is a major focus of my
thesis: the way I choose to study and characterise this Dark Energy is based on the
large-scale structure of the Universe through a probe called the integrated Sachs–
Wolfe effect (iSW). This effect is theoretically detectable in the cosmic microwave
background (CMB): this light, which originated in the early Universe (380,000
years after the Big Bang), travelled through large cosmic structures (galaxies and
clusters) before reaching us, all of them underlain by gravitational potentials. The
acceleration of the expansion (and Dark Energy) has the effect of stretching and
‘‘flattening’’ these potentials during the crossing of photons. Depending on the
properties of the Dark Energy, these photons change frequency and should
therefore appear as colder or hotter spots in the CMB. The iSW effect has a direct
but very weak impact on the power spectrum of the temperature fluctuations of the
CMB. It therefore requires the use of external data to be detectable. A
conventional approach to this problem is to correlate the CMB with a tracer of
the distribution of matter in the Universe (usually galaxy surveys), and therefore
the underlying gravitational potential. This has been attempted numerous times
with surveys covering a large range of wavelengths, but the measured correlation
has yet to give a definitive and unambiguous result on the detection of the iSW
effect. This is mainly due to the shortcomings of current surveys that are not deep
enough and/or cover too small a fraction of the sky. A part of my thesis is devoted
to the correlation of the CMB with another diffuse background, namely the cosmic
infrared background (CIB), which is composed of the integrated emission of the
non-resolved distant star-forming galaxies. I was able to show that it is an
excellent tracer of the gravitational potentials, being free from many shortcomings
of current surveys. The results of my study show that the levels of significance for
the expected CIB-CMB correlation exceed those of current surveys, and compete
with those predicted for the future generation of very large surveys (Pan-STARRS,
LSST, Euclid). Next, my thesis focuses on the individual imprint in the CMB of
the largest structures in the Universe by the iSW effect. According to an article by
[1], the iSW effect was directly detected in the stacking of patches of the CMB at
the positions of superstructures. However, the high measured amplitude of the
effect seems to be at odds with predictions from the standard model of cosmology.
xi
xii Abstract
My work on the subject first involves revisiting this study with my own protocol,
completed and associated with a thorough variety of statistical tests to check the
significance of these results. The latter proved to be particularly difficult to assess
and subject to possible selection biases. I extended the use of this detection method
to other available catalogues of structures, more consequent and supposedly more
reliable in their detection algorithms. The results from one of these new catalogues
[2] suggest the presence of a signal at scales and with an amplitude more
consistent with theory, but at more moderate levels of significance than the
catalogue of Granett et al. At the same time, being a member of the HFI Core
Team of the Planck Collaboration, I also performed the same detection using data
from the Planck satellite. The results of the stacking approach raised a number of
questions concerning the nature of the expected signal. This led me to propose an
original theoretical prediction of the iSW effect produced by the superstructures
previously mentioned, through simulations based on general relativity and the
Lemaître–Tolman–Bondi metric. This allowed me to reproduce the structures of
Granett et al. and predict the exact full theoretical iSW effect of these structures. I
showed that the central amplitude of the measured signal is consistent with the
LCDM theory. However, the measured signal shows non-reproducible features
that are hardly compatible with my predictions. A different part of my thesis
focuses on a distant time in the history of the Universe, called Reionisation: the
transition from a neutral universe to a fully ionised state under the action of
the first stars and other ionising sources. This period has a significant influence on
the CMB and its statistical properties, in particular the power spectrum of its
polarisation fluctuations. In my case, I focused on the use of temperature
measurements of the intergalactic medium (IGM) during Reionisation in order to
investigate the possible contribution of the decay and annihilation of hypothetical
Dark Matter particles. Starting from a theoretical work based on several Dark
Matter models, I computed and compared predictions of the IGM temperature to
actual measurements, which allowed me to extract new, interesting and potentially
powerful constraints on key parameters of Dark Matter, as well as on signatures of
Reionisation itself.
References
1. B.R. Granett, M.C. Neyrinck, I. Szapudi, An imprint of superstructures on the microwave

background due to the integrated Sachs-Wolfe effect. APJ 683, L99–L102 (2008)
2. P.M. Sutter, G. Lavaux, B.D. Wandelt, D.H. Weinberg. A Public Void Catalog from the SDSS
DR7 Galaxy Redshift Surveys based on the Watershed Transform. APJ 761(1), 44 (2012).
ISSN 0004-637X
Acknowledgments
I wish to thank all the people I ever met in my life, for having played a part
(however small it may be) in making me the person I am today.
. . . This is what I would say if I had to summarise my acknowledgments in one
sentence only! However, a lot of people included in the aforementioned group
deserve much more than a single line of thanks, hence the following:
• First of all, I would like to thank again the members of my thesis committee for
accepting my invitation to evaluate my thesis and its defence, as well as for their
interesting remarks and questions. These will be of great use for the continuation
of some of the works that I presented! I would like to thank in particular my
referees, François-Xavier Désert and Martin Kunz (an old acquaintance!) for
their careful reading of my manuscript and their very relevant comments despite
the tight schedule that I imposed on them!
• The second group of people that I would like to thank are the ‘‘non-scientific’’
staff at the IAS with whom I interacted during the last few years: they all have
been always very kind and considerably helpful, and ever so patient with me and
my thousands of administrative questions and forms to fill! I give many thanks
to Véronique, Patricia, Hélène, Chloé, Nicole, Delphine and all the others!
• Thanks also to the many members of the MIC team and the rest of the lab for all
the different interactions that we have had, always enjoyable and very enriching.
Thanks to Alain, Vincent, Julien, Jean-Loup, Hervé, Guilaine, Jonathan, Lau-
rent, Rosario,. . . and many more, once again!
• Included in the aforementioned group, I am particularly grateful to those whom I
may plainly call my ‘‘close collaborators’’ but whom I much rather like to call
my friends: Nabila (ladies first), Alexandre (so long, and thanks for all the
chocolates), but also and foremost my two brilliant advisors Mathieu and
Marian (aka M&M): you undoubtedly made me the scientist I am today, and for
that I am forever grateful to you and all the more happy to be able to count you
as friends!
• If there is a group of people that made every day of my thesis so pleasant at the
lab, it is without a doubt the Ph.D. students and postdocs of the IAS, whose
comradeship, perpetual good mood and proneness to partying made the lab such
xiii
xiv Acknowledgments
an enjoyable place to live (and work, occasionally!). My thanks to the ‘‘elders’’

first, some of whom already had the time to leave and come back to the IAS:
Mister John Carter, Benjamin, the charming trio Aurélie, Léo and Anouck,
Nicolas, Romain, Olivier, Aikel, Cédric,... Let’s also mention the outrageous
Sébastien, a category by himself, to whom I may only associate the no-less
unforgettable Heddy: the two of them contributed to a proven 85 % drop in the
students’ productivity at the IAS, as well as a 1,526 % increase in morale! As for
those more contemporary to me, I salute my colleagues Julian, Sihane, Cédric,
Chloé, Paola, Vincent, Morgane, Joachim, Marco, David, Anna, Lapo and the
youngest ones too (JB and Co.)!
• Let me not forget the man (the legend!) with whom I have shared an office for
3 years: the one and only Fabien! Thanks for all the good times my friend,
I hope I didn’t disturb you too much on these afternoons when you’d try to beat
your high-scores! And thanks for your brilliant intellect too (a result of your
venerable age, I guess) which helped me so many times during the last 3 years!
• A word of thanks to my former classmates too, who cut their teeth with me
during the first years at the university already some time ago: Benjamin, Jérémy,
Marie-France, Vincent, Yann, Clélia, we may not have seen each other a lot
during the thesis, but each occasion was always memorable! And for the more
recent acquaintances made in M2: Marine, Sophie, Sandra, Laurent, David, Loïc
(although those last three belong to the previous group too) and the others,
thanks to you too!
• I give the warmest thanks to my ‘‘old’’ friends: Johan and Mikaël, Arthur,
Guillaume, Adrien, Alvi, Sandra. They knew me well before the thesis, during
and after it, always accepting me for the ‘‘egghead’’ that I am, and always there
and ready for a good time far away from work, the stars and the Universe.
Thanks guys (and girls)!
• Where would I be without my family, starting with my parents and grandparents
who, aside from allowing my very existence (!), always supported (and toler-
ated!) me throughout my studies and my life.
• I dedicate very special thanks to a certain lady dear to my heart (she’ll recognise
herself): for her, a single kiss, since words aren’t enough to express my gratitude
for her support, her kindness and her love during this quite (!) wearying
adventure that is a thesis.
• Last, but not least: who would I be without my two dear older brothers, for who
I have the utmost respect? Ivan and Alex, thank you from the bottom of my
heart for literally being there from the beginning and bringing me up to my feet,
for supporting me, being always there when needed, for all the good (the best!)
times and all the rest: you really are the best, guys, and I wouldn’t be there
without you!
Acknowledgments xv
And here I am now, at the end of these acknowledgments, where I should

probably give as a closing statement some of my impressions on this three-year
long human experience, and on the future… not so easy a task! As I try to avoid
the usual clichés, let me try and express it in the most complete, concise and honest
way possible:
. . . It was all really great! And thanks to a lot of wonderful people!
Thanks to you all.
Chilly-Mazarin, January 2014 Stéphane Ilić

Contents
1 Introduction: The Ingredients of a Good Cosmological Probe . . . . 1

1.1 The Cosmic Microwave Background . . . . . . . . . . . . . . . . . . . . 2
1.2 Large Scale Structures and Perturbation Theory . . . . . . . . . . . . 4
1.3 The (Current) Ruler of the Universe: Dark Energy . . . . . . . . . . 6
1.3.1 Models of Dark Energy . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.2 Probes of the Dark Energy . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Shedding Light on the Dark Energy: The Integrated
Sachs–Wolfe Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 9
1.5 A Glimpse at the Dark Components Through the Reionisation . .. 12
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 12
2 Unravelling the iSW Effect Through the Matter Distribution. . . . . 15

2.1 CMB Cross-Correlation with Tracers of Matter. . . . . . . . . . . . . 15
2.1.1 The Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.2 Theoretical Framework . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.3 Application to Detectability and Test Case. . . . . . . . . . . 19
2.1.4 State-of-the-Art of the iSW Detection . . . . . . . . . . . . . . 22
2.2 Optimising the Cross-Correlation for iSW Detection . . . . . . . . . 23
2.2.1 Context and Tools. . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.2 Simulations, Covariance Estimation and Parameter
Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.3 Tomography for iSW Studies . . . . . . . . . . . . . . . . . . . . 28
2.3 The Cosmic Infrared Background and the iSW Effect . . . . . . . . 32
2.3.1 The Blurry Red Light . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3.2 The CIB as an Alternative Tracer of Matter . . . . . . . . . . 34
2.3.3 Predictions on the CIB-CMB Cross-Correlation . . . . . . . 37
2.3.4 Accounting for Realistic Conditions . . . . . . . . . . . . . . . 42
2.3.5 Multi-frequency Joint Analysis . . . . . . . . . . . . . . . . . . . 45
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
xvii
xviii Contents
3 The Impact of Identified Superstructures in the CMB . . . . . . . . . . 51

3.1 A New Approach to the iSW Effect . . . . . . . . . . . . . . . . . . . . 51
3.1.1 Narrowing the View . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.1.2 The Stacking of Superstructures in the CMB . . . . . . . . . 52
3.1.3 The Gr08 Results: Protocol, Results and Caveats . . . . . . 53
3.2 My Revised and Improved Stacking Protocol . . . . . . . . . . . . . . 56
3.2.1 Data and Core of the Protocol . . . . . . . . . . . . . . . . . . . 56
3.2.2 Gr08 Catalogue: Preparation of the Maps . . . . . . . . . . . 58
3.2.3 Application to Gr08: Significance and Analysis . . . . . . . 64
3.2.4 Refining the Stacking Analysis . . . . . . . . . . . . . . . . . . . 66
3.3 Analysis of Additional Voids . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.3.1 The Catalogues of Pan et al. and Sutter et al. . . . . . . . . . 70
3.3.2 New Objects, New Results . . . . . . . . . . . . . . . . . . . . . . 71
3.3.3 Further Improvement of the Protocol. . . . . . . . . . . . . . . 74
3.4 Precautions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.4.1 Selection Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.4.2 Alignment and Overlap Effects. . . . . . . . . . . . . . . . . . . 78
3.4.3 Spurious Detections? . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.5 What Does KCDM Has to Say on the Matter? . . . . . . . . . . . . . 81
3.5.1 Too Small or too Large Signal? . . . . . . . . . . . . . . . . . . 81
3.5.2 Is 10 lK too Much?. . . . . . . . . . . . . . . . . . . . . . . . . 82
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4 Towards a Full Modelling of the iSW Effect . . . . . . . .......... 85

4.1 Computing the iSW Effect from a Structure:
A Step-by-Step Recipe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.1.1 Modelling an Inhomogeneity . . . . . . . . . . . . . . . . . . . . 86
4.1.2 Applying the LTB Metric to Cosmic Voids . . . . . . . . . . 87
4.1.3 ‘‘Lighting Up’’ the Voids . . . . . . . . . . . . . . . . . . . . . . 88
4.1.4 Numerical Implementation . . . . . . . . . . . . . . . . . . . . . . 92
4.2 The Predicted iSW Effect from Actual Structures . . . . . . . . . . . 94
4.2.1 Return to the Granett et al. Catalogue . . . . . . . . . . . . . . 94
4.2.2 First Computations and Predictions . . . . . . . . . . . . . . . . 96
4.2.3 Full Modelling of the iSW Profile. . . . . . . . . . . . . . . . . 98
4.3 Exploring the Limits of the LTB Landscape . . . . . . . . . . . . . . . 101
4.3.1 Assumptions and Their Consequences . . . . . . . . . . . . . . 101
4.3.2 Dependences on Initial Parameters . . . . . . . . . . . . . . . . 104
4.3.3 A Statistical Oddity? . . . . . . . . . . . . . . . . . . . . . . . . . . 105
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Contents xix
5 Studying Dark Matter Through the Lens of the Reionisation. . . . . 109

5.1 From the First to the Second Light of the Universe . . . . . . . . . . 109
5.1.1 The Reionisation of the Universe . . . . . . . . . . . . . . . . . 109
5.1.2 Physics of the Reionisation and Key Parameters . . . . . . . 110
5.1.3 Observational Constraints. . . . . . . . . . . . . . . . . . . . . . . 113
5.2 Contribution of the Decay and Annihilation of Dark Matter
to the EoR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.2.1 The Dark Matter Mystery . . . . . . . . . . . . . . . . . . . . . . 118
5.2.2 Dark Matter and Reionisation . . . . . . . . . . . . . . . . . . . . 119
5.2.3 Recombination Code and Customisation . . . . . . . . . . . . 121
5.2.4 Additional In-depth Analysis . . . . . . . . . . . . . . . . . . . . 123
5.2.5 Effects on the CMB Spectrum . . . . . . . . . . . . . . . . . . . 124
5.3 Constraints on Dark Matter from the Temperature
of the IGM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.3.1 Probing the IGM Temperature Evolution . . . . . . . . . . . . 128
5.3.2 Refining the Reionisation Model. . . . . . . . . . . . . . . . . . 129
5.3.3 Setting Up the Framework . . . . . . . . . . . . . . . . . . . . . . 130
5.3.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . 130
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6 Conclusions and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
Abbreviations
LCDM Lambda Cold Dark Matter

CMB Cosmic Microwave Background
DE Dark Energy
DM Dark Matter
EoR Epoch of Reionisation
FIR Far InfraRed
HEALPix Hierarchical Equal Area isoLatitude Pixelization
IGM Inter Galactic Medium
LRG Luminous Red Galaxies
LSS Large-Scale Structures
LTB Lemaître–Tolman–Bondi (metric)
QSO Quasi-Stellar Objects
RMS Root Mean Square
S/N Signal-to-Noise ratio
SDSS Sloan Digital Sky Survey
SLS Surface of Last Scattering
xxi
Chapter 1
Introduction: The Ingredients of a Good
Cosmological Probe
The state of our current understanding of the Universe may seem quite frustrating,
to say the least. On the one hand, cosmologists have accumulated in the last decades
a tremendous amount of data coming from a large number of ground- and space-
based instruments, scrutinising the whole range of the electromagnetic spectrum. The
results derived from theses experiments allowed to expand and refine our knowledge
in many fields of astrophysics and helped strengthen what is currently regarded as
the standard model of cosmology: the CDM paradigm. However, despite this ever-
growing wealth of data and the tighter and tighter agreement between theory and
observations, the intrinsic nature of 95 % of the energy content of the Universe still
eludes us!
Even more ironic is the fact this “known unknown” consists of the two very
elements that gave their names to the standard model of cosmology. First, the (Cold)
Dark Matter (CDM), whose existence was first postulated by Jan Oort in 1932 to
account for the orbital velocities of stars in the Milky Way and by Fritz Zwicky in
1933 to account for evidence of “missing mass” in the orbital velocities of galaxies
in clusters. This hypothetical type of matter is referred to as “Dark” to indicate one
of the few certainties that we have: its absence of emission/absorption of light at
any significant level. The nature of the DM is still uncertain, with models ranging
from axions, to supersymmetric particles, and even black holes. And secondly, the
so-called cosmological constant symbolised by the Greek letter , as it was fist
introduced by Albert Einstein himself—but for a wholly different purpose. It is
nowadays the preferred theory to explain the elusive “Dark Energy”, responsible for
the accelerated expansion of the Universe, and it accounts for 68.3 % [31] of the
Universe’s energy budget.
Despite the mystery surrounding these elements (often called the “Dark Com-
ponents” or “Dark Sector” of the Universe) and this rather hazy situation, progress
in the field of cosmology is far from being hampered by this apparent “obscurity”,
quite on the contrary: cosmologists are all the more hard at work to unravel these
fundamental components and the very nature of the Universe. However, rather than
focusing only on a never-ending race to improve the precision of our instruments,
S. Ilić, The Large Scale Structures, Springer Theses, 1

DOI: 10.1007/978-3-319-07746-8_1, © Springer International Publishing Switzerland 2014
2 1 Introduction: The Ingredients of a Good Cosmological Probe
a substantial effort in now also spent on the search for new and ingenious ways to
exploit and combine existing tools and datasets, find new and clever observables,
and turn them all into powerful probes of the Universe.
The majority of the work presented in this thesis revolves around one of such
probes, which calls for a unique combination of cosmological observables and is
aimed at the study of the Dark Energy: the so-called integrated Sachs–Wolfe effect.
In this introductory chapter, I will briefly present the origin and physical processes
involved in this particular effect, as well its related observables and possible probes.
1.1 The Cosmic Microwave Background
The existence of a background of photons roaming the Universe was speculated upon
as far back as the middle of the 20th century [3]: this relic of decoupled radiation
arises in the Big Bang scenario as a natural consequence of the expansion of the
Universe and its hot and dense past. It was almost two decades later that the Cosmic
Microwave Background (CMB) was finally discovered (Penzias and Wilson [26],
an achievement awarded by nothing less than the Nobel Prize) and immediately
interpreted [8] as this formidable source of cosmological information that it would
soon become.
We now know that the CMB is an almost isotropic radiation, matching a perfect
black-body spectrum at a temperature of T = 2.7260 ± 0.0013 K [14, 24]. Its
emission dates back to the epoch of recombination of the hydrogen and helium atoms
when the Universe was about 400,000 years old. Before this epoch, the Universe was
opaque to light due to the high density of free electrons which prevented the photons
from freely propagating by Compton scattering. Then, the fluctuations of the physical
fields at recombination (such as temperature, density and velocity) left their imprints
in the CMB in the form of anisotropies both in temperature and polarisation. A
crucial consequence is that the CMB radiation and its fluctuations represent therefore
an image, a snapshot of the Universe at that particular time. These fluctuations are
often called primordial (or primary) anisotropies, since they originate from physical
processes that occurred at recombination and even before—indirectly giving us a
glimpse of the otherwise inaccessible early Universe (numerous reviews of these
processes can be found in the literature, e.g. [16, 17]).
Establishing the tremendous potential of these anisotropies, cosmologists under-
took the task of measuring them with an ever-increasing precision. A myriad of
instruments were developed in the last three decades, from ground-based missions
(DACI, CBI, SPT, ACT,…) to balloons (BOOMERanG, MAXIMA, Archeops,…)
and finally three successive generations of satellites: in 1989, the COBE satellite
was the first to be launched to observe the CMB and succeeded in providing the
first angular power spectrum of CMB anisotropies. Then, the WMAP (2003–2011)
and Planck (2009–2013) satellites were launched and observed those small fluctu-
ations with unprecedented sensitivities and accuracy (cf. Fig. 1.1). The latest data
from Planck allows us to put the tightest constraints to date on many aspects of the
1.1 The Cosmic Microwave Background 3
Fig. 1.1 Top Map of the CMB temperature anisotropies (with 3 % of the sky replaced by a con-
strained Gaussian realization) as constructed by the Planck Collaboration [29] using a spectral
matching independent component analysis (SMICA). Bottom The angular power spectrum of the
CMB temperature anisotropies from Planck, well fit by a simple six-parameter CDM model. The
shaded area around the best-fit curve represents cosmic variance, including the sky cut used. The
error bars on individual points also include cosmic variance. Both figures borrowed from the Planck
Collaboration [30]
standard model of cosmology. Among these we find: the measurement of a highly

significant deviation from scale invariance of the primordial power spectrum, a revi-
sion of the density parameters of the Universe (some of them significantly different
from previous measurements), the confirmation of several large scale anomalies in
the CMB temperature distribution, new limits on the number and mass of neutrinos,
the measure of the gravitational lensing of CMB anisotropies at 25 σ , and no evidence
for non-Gaussian statistics of the CMB anisotropies.
Although the CMB signal has been almost unchanged since recombination, some
small alterations still arise due to several effects occurring on the path of these pho-
tons. Such late-time modifications of the CMB are called secondary anisotropies and
represent also a mine of cosmological information, since their features depend heav-
ily on the assumed cosmological model. Among others, different types of secondary
anisotropies may be generated when the photons go through high density areas such
as clusters of galaxies. In this case the high temperatures and velocities of high energy
electrons in the gas can be transferred to the photons via inverse Compton scattering,
producing characteristic, frequency dependent anisotropies on small scales called
thermal and kinetic Sunayev–Zel’dovich effect respectively. Another example, as a
fraction of the hydrogen in the Universe becomes ionised again during the epoch
of reionisation, the CMB photons will undergo Compton scattering again, smearing
out a part of the primary anisotropies; reionisation may happen globally or locally
around some sources, and is actually the focus of a chapter of the present thesis (cf.
Chap. 5). A description of these (and other) secondary anisotropies can be found in
reviews such as Aghanim et al. [2].
The central topic of my thesis concerns yet another type of secondary anisotropies,
due to the influence of the gravitational potentials on the streaming CMB photons.
More precisely, it concerns one of the effects that variations of these potentials may
have along the path of the photons, and that is commonly called the integrated Sachs-
Wolfe effect (iSW). The detailed principle of this process will be discussed first in
Sect. 1.4 and extensively throughout the whole manuscript; as we will see, the large
scale structures of the Universe play a crucial rôle in this effect.
1.2 Large Scale Structures and Perturbation Theory
In the context of secondary anisotropies, the CMB I just briefly described can be seen
as some sort of cosmic “embroidery” of a myriad of physical processes, all along the
way from its emission until it finally reaches us more than 13 billions years later.
As mentioned before, the presence of all kinds of structures (galaxies, clusters,
voids,…) on the line of sight is partly responsible for these secondary anisotropies.
Indeed, although the working assumption of a homogeneous and isotropic Universe
(the so-called cosmological principle) is successful in describing the overall behav-
iour of the expansion and its relation to the energy content (through the well-known
Friedmann–Lematre–Robertson–Walker, or FLRW metric) if we look at the Universe
closer, we see that isotropy and homogeneity are broken on small scales. Therefore,
one needs to form a more complex model in order to have the tools to describe more
of the details we observe, such as the structure of the matter distribution and, con-
sequently, the anisotropies in the CMB. This can be achieved by what is called a
first-order perturbative theory built on top of the zero-order homogeneous theory of
1.2 Large Scale Structures and Perturbation Theory 5
the FLRW model. This theory was first introduced by Lifshitz [21]; for more detailed
reviews, the interested reader can refer to Bertschinger [5] and Weinberg [40]. The
general idea is to start from the classical, unperturbed FLRW metric gμν :
g00 = −1 (1.1)
g0i = gi0 = 0 (1.2)
gi j = a 2 δi j (1.3)
with a(t) being the well-known scale factor of the Universe. From there, we introduce
the fact that the observed Universe is not perfectly homogeneous and isotropic: we
assume, however, at the simplest order that these inhomogeneities lead only to small
variations of the geometry. To do so, we define the perturbed geometry by:
∗
gμν = gμν + a 2 θμν (1.4)
with gμν being the FLRW metric above and with |θμν | √ 1. Similarly, a small
μ
perturbation is also added to the energy momentum tensor Tν . The Einstein equations
of general relativity put constraints on these additional terms, but a gauge degree
of freedom remains. One of the most commonly used gauge, called longitudinal (or
conformal), allows to rewrite the perturbed metric with only two additional functions,
φ and , both depending on space and time:
∗ −
g00 (→
x , t) = −1 − 2(− →
x , t) (1.5)
∗ − → −
→
g ( x , t) = g ( x , t) = 0
0i i0 (1.6)
gi∗ j (−
→
x , t) = a δi j (1 + 2φ(−
2 →
x , t)) (1.7)
The perturbations φ corresponds to the Newtonian potential, and is the perturba-

tion to the spatial curvature. Since the perturbations in the universe are small at the
times and scales of interest, we treat φ and as small quantities, dropping all terms
quadratic in them. These functions (called “Bardeen potentials”) will be of particular
importance in the Sect. 1.4 in order to introduce and define the iSW effect (see also
Durrer [10], for additional reference).
The perturbative theory, and the standard model in general, have to be capable of
predicting the behaviour of many observables, and in particular the distribution of
large scale structures in the Universe. In order to confront it to the data, a number of
large galaxy surveys have been performed since the middle of the 70s, starting with
astronomers of the Centre for Astrophysics (CfA) in Cambridge, who conducted the
first ever 3D survey with a statistically significant number of galaxies with redshifts
[18], with a mean redshift of 0.025, and a coverage of 0.2 square degree. This partic-
ular survey revealed for the first time the aforementioned breakdown of homogeneity
and isotropy in the local Universe, showing a “foamy” distribution of matter com-
posed of large voids and filaments of galaxies. Since then, tremendous progresses
have been accomplished on the coverage and depth of these surveys: one of the
most recent (and still active), the Sloan Digital Sky Survey (SDSS) [1] contains now
more than 200 millions objects, and more than 1 million galaxies with their spectra
resolved, both in the North and South hemispheres and with a mean redshift of 0.3.
The study of these surveys in themselves and their statistical properties provides
invaluable sources of cosmological information (for an overview, see e.g. Sodré [37]
and references therein). But for the purpose of my work, I was more interested in
their link to other cosmological observables, in particular to the CMB itself, and to
how the underlying cosmological model influences their relation and interaction.
1.3 The (Current) Ruler of the Universe: Dark Energy
The first conclusive proof of the existence of the most prominent member of the “Dark
sector” came in 1998, when Perlmutter et al. [28] and Riess et al. [34] evidenced
the acceleration of the expansion of the Universe through measurements of type
Ia supernovae luminosity distances (cf. example in Fig. 1.2). Since the consensus
at that time was in favour of a decelerated or linear expansion, these observations
prompted cosmologists to hypothesise the existence of a new form of unknown
energy responsible for this acceleration, which is henceforth referred to as the “Dark
Energy”.
1.3.1 Models of Dark Energy
From then on, a multitude of theoretical models has been proposed to find a phys-
ical explanation for this peculiar component of the Universe, and whose nature is
considered as one of the most important open questions in modern cosmology. The
simplest model which can explain most of the current observations is obtained by
adding a so-called cosmological constant to the Einstein equations: this constant
had been first introduced by Einstein in order to obtain a static Universe, which
proved later incompatible with the discovery of the Hubble expansion. In its modern
form, we can think of an interpretation of the cosmological constant as the energy of
vacuum, with an energy density constant in time and space and a negative equation
of state, which started dominating the energy budget of the Universe only at rela-
tively recent times (z ∼ 0.4) when the matter density was diluted enough. Despite
its success in reproducing the majority of equations and its status in the “standard
model” of cosmology, the cosmological constant is problematic, especially when
we try to estimate the vacuum energy quantitatively: its estimated value from the
standard model of particle physics is 10120 higher than its observed cosmological
value! This discrepancy of 120 orders of magnitude is possibly the worst prediction
of contemporary physics, and is known as the cosmological constant problem [39].
Since the identification of Dark Energy with the vacuum presents the afore-
mentioned difficulties, a range of possible alternatives is currently being explored
1.3 The (Current) Ruler of the Universe: Dark Energy 7
Fig. 1.2 Status of the measurements of the Hubble diagram of type Ia supernovae as reported in
Suzuki et al. [38]. The plot shows the distance modulus (related to the observed magnitude of the
supernova) as a function of the redshift. The solid line represents the best-fit cosmology for a flat
CDM universe for supernovae alone ( = 1 − m = 0.729 ± 0.014)
(although at the moment, every observation remains perfectly consistent with the
simplest model of the cosmological constant). The confrontation with these prob-
lems has led theorists to speculate that energy density of the Dark Energy may not
be a constant, but a dynamical quantity, which happens to be small today because
of the advanced age of the Universe. These “dynamical” models are usually called
quintessence. The most natural and popular way to realise a dynamical model is
to introduce one or more scalar fields which contributed to the total energy density
of the Universe. This class of models was introduced by Wetterich [41], Ratra and
Peebles [32]; for a review, see e.g. Linder [22].
All the theories mentioned so far are based on the introduction of some additional
component to the stress-energy tensor with certain properties which will induce the
desired accelerated behaviour of the expansion at late times. An interesting alternative
is to modify the geometrical part of the Einstein equation instead. These are the
so-called modified gravity theories, such as the f (R) models—perhaps one of the
simplest and most popular extensions of general relativity (for an introduction see
e.g. Amendola and Tsujikawa [4]). A more drastic approach is to assume that the
4D Universe we observe is in reality embedded in a higher dimensionality bulk,
whose extra dimensions are unobservable thanks to some particular mechanism. The
latter models are called braneworld, such as the Dvali–Gabadadze–Porrati (DGP)
model [11].
Finally, while a great effort is being made to find a solution to the Dark Energy
problem with the introduction of interesting new physics, it is still possible for current
observations to be explained by well known gravitational effects. This can happen
if at large scales we drop the homogeneity assumption on which the FLRW metric
is based: by doing so, non-linear gravitational effects arise which might be similar
to the effects of a homogeneous accelerating Universe [20]. This approach is often
called “backreaction” as the presence of the inhomogeneities acts on the background
evolution and changes it. On the other hand, several attempts exist to build models
which could satisfy all pieces of observational evidence: they are generally based
on the idea that if we happen to be in an underdense region, gravity will appear
locally weaker as matter is attracted to denser regions elsewhere. This concept can
be realised by models such as the Lemaître–Tolman–Bondi (LTB, see [13]), although
it was shown that these models have some trouble reproducing simultaneously all
the current cosmological observables (see e.g. [6]).
1.3.2 Probes of the Dark Energy
It is important to underline that the supernovae measurements mentioned earlier are

not able to prove the existence of the Dark Energy by themselves: they have to be
combined with other cosmological observations, most notably about the geometry
of the Universe. More generally, many other probes of the Dark Energy were devised
since its first discovery, allowing us to accumulate enough evidence to make it one
of the pillar of the current standard model of cosmology.
Among those probes, we find first the constraint on flatness we see in the cosmic
microwave background. As we know, the CMB radiation is not perfectly isotropic and
has small angular fluctuations. The angular power spectrum of these presents detailed,
observable features that depend on the cosmological parameters. In particular, the
total energy density of the Universe is measured by the position of the first peak
(cf. Fig. 1.1): this is because this scale corresponds to the angular dimension of
the cosmological horizon at the epoch of last scattering. Accurate measurements
available today conclude that the Universe is Euclidean, with a total energy density
close to critical ( total 1). However, at first glance, this is in clear contrast with the
sum of the “observed” components of the Universe (i.e. mainly matter in all its forms,
which amounts to m 0.3): the simplest solution is then to introduce heuristically
a new component with an energy density ∼ 0.7, i.e. the Dark Energy.
Additional evidence for Dark Energy comes from the observation of the large scale
structure in the Universe. The baryon acoustic oscillations (BAO), besides leaving
a large oscillatory pattern in the CMB angular power spectrum, also produce much
smaller oscillations in the matter power spectrum, corresponding to a single bump
in the matter angular correlation function at a scale of ∼150 Mpc. A measurement
of this bump in a catalogue of sources (e.g. luminous red galaxies) is an accurate
estimation of the angular distance to that characteristic scale, whose value again
indicates the presence of Dark Energy [12, 27]. More generally, statistics associated
1.3 The (Current) Ruler of the Universe: Dark Energy 9
to the distribution of galaxies (number counts, mass function,…) also allow to put
constraints on the features of the Dark Energy. Weak gravitational lensing (e.g. [25])
is a powerful technique to map the distribution of Dark Matter in the Universe by
observing the distortion of the images of background galaxies by the foreground
Dark Matter distribution. This distortion has been detected since 2000, its observation
putting constraints on the matter energy density—again pointing indirectly to a Dark
Energy dominated Universe. The deployment of a new generation of almost full-sky
lensing surveys such as Pan-STARRS, LSST or Euclid has the potential to make
weak lensing one of the most powerful methods to constrain Dark Energy in the near
future. Further evidence of Dark Energy comes from another gravitational effect on
the CMB of the distribution of matter, whose evolution is tightly influenced by Dark
Energy itself: the measurement of the integrated Sachs–Wolfe effect, which will be
the central point of my thesis and is described in the following section.
1.4 Shedding Light on the Dark Energy: The Integrated

Sachs–Wolfe Effect
As mentioned earlier, a particular type of temperature anisotropies is found in the

CMB due to the effect of the gravitational potentials on the streaming CMB photons.
The earliest effect to occur is called the “early” Sachs–Wolfe effect: it is caused by
the gravitational redshift occurring at the surface of last scattering—the (observer-
dependant) surface from which the CMB photons were originally emitted at the time
of photon decoupling. The effect is not constant across the sky due to differences
in the matter/energy density at the time of last scattering [35]. In the framework of
the linear, pertubative theory the effective temperature change of the photons can be
directly expressed as a function of the Bardeen potentials φ and at recombination.
However, the effect I am interested in alters the CMB power spectrum at a much
more recent epoch. When streaming through space from the last scattering surface,
the CMB photons undergo the effect of the local gravitational field at late times, which
can also be described using the Bardeen potentials. This effect can be integrated along
the line of sight and is expressed by the so-called Sachs–Wolfe equation:
η0
δTiSW = ˙
e−τ (η) (φ̇ − )[(η 0 − η)n̂, η] dη (1.8)
ηSLS
where φ and are the Bardeen gravitational potentials in the conformal gauge
described earlier, η is the conformal time (with ηSLS at the surface of last scattering
and η0 today), dots represent conformal-time derivatives, τ is the Thomson opti-
cal depth, and the e−τ (η) term accounts for the damping by Thomson scattering of
CMB photons off free electrons. Note that in the CDM paradigm, we consider
that the Universe is composed of perfect fluids (whatever they are) so that there is
no anisotropic stress (non-zero off-diagonal terms of the energy momentum tensor):
after some calculations in perturbation theory (see e.g. Durrer et al. [10]), it can be
shown that it leads to the following relation between Bardeen potentials:
φ = −, (1.9)
leading to an alternative formulation of the iSW effect very often found in the liter-
ature: η 0
δTiSW = 2 e−τ (η) φ̇[(η0 − η)n̂, η] dη. (1.10)
ηSLS
We understand here that the iSW effect can happen only if the gravitational potential
vary with time. The evolution of the potential is tightly linked to the evolution of the
density through the Poisson equation. In the framework of perturbation theory, we
can show that the evolution of density perturbations is controlled by a function D(a)
called the “growth function”, and that in turn the potential is proportional to D(a)/a.
In a flat, matter-dominated Universe, the linear evolution of density inhomogeneities
is proportional to a(t) (i.e. D(a) ∝ a, see e.g. [9, Chap. 7] for reference), so that
the linear evolution of the potentials is null: the iSW effect cannot occur if the
dominant fluid is composed of matter. However, in our best-fit cosmological model,
the presence of Dark Energy causes a decay of these potentials due to the accelerated
expansion of the Universe [19]: this is usually referred to as the “late” integrated
Sachs-Wolfe effect (iSW). Since the transition from a matter- to a DE-dominated
era occurred recently (in redshift) according to the standard model, new anisotropies
were formed in the CMB at recent times, when the horizon size was comparable with
its current size. Therefore the affected scales are generally the largest, at multipoles
π < 100 (cf. Fig. 1.1). The physical picture is very straightforward; as a CMB photon
falls into a gravitational potential well, it gains energy; as the photon climbs out of
a potential well, it loses energy. These effects exactly cancel if the potential is time
independent, but can result in a net kick if the potential evolves as the photon passes
through. The exact opposite phenomenon, i.e. a loss of energy, happens if the photon
crosses a cosmic void. It should be noted that the iSW effect is not exclusively a
signature of the standard model and its associated : it is interesting to remark that
in most modified gravity scenarii the gravitational potentials have a different time
evolution even during the matter era (and in a flat geometry), thus making this effect
a potential discriminant between different theories (as shown by Lue et al. [23]). This
also applies to other Dark Energy models, which all have their unique “version” of
the iSW effect.
In contrast to this “late” iSW effect, it is interesting to note that some “early” iSW
effect has occurred some time after recombination due to a non-negligible radiation
contribution in the energy budget, thus the gravitational potentials are decaying for
some time after the creation of the CMB. As the horizon size was much smaller then
than today, the small additional anisotropies were then produced at higher multipoles,
comparable with those of the first acoustic peak (π 200). However, the relative
signal is embedded in the primary CMB anisotropies, which are much larger and
make a direct measurement challenging. Of interesting note is also that at smaller
1.4 Shedding Light on the Dark Energy: The Integrated Sachs–Wolfe Effect 11
angular scales, the photons from the CMB may undergo some energy shift when
they cross a high density region, like a galaxy cluster. In this case linear theory
breaks down, and again the gravitational potentials are not constant even in a matter-
dominated era. This effect represents the non-linear part of the iSW, and is known as
Rees-Sciama [33]. Since it is sourced by non-linear regions, its contribution to the
CMB power spectrum typically peaks around π 100, and is always much smaller
than the primary CMB. However, a possible approach to the practical measurement
of this phenomenon has been proposed in the literature [36].
Probing the iSW effect As a consequence of what I mentioned above, a direct
measurement of the iSW contribution to the CMB power spectrum is made virtually
impossible by the embedding of the small iSW signal in the much larger primary
CMB anisotropies at the relevant multipoles (10 times at the very least for the lowest,
and increasingly more for higher πs). Furthermore, the total iSW signal is due to all
the density fluctuations, both positive and negative, along the line of sight: on small
scales, the individual temperature differences are small and they tend to cancel out.
The most significant iSW effect results from the coherent large scale potentials, but
unfortunately these scales are precisely where the CMB signal is mostly dominated
by cosmic variance.
However, the situation changed when Crittenden and Turok [7] presented a new
technique, which made it possible to extract this effect by cross-correlating the
observed CMB map with some tracer of the matter density field. The method is
based on the fact that the primary CMB anisotropies have been generated at the sur-
face of last scattering, and therefore are completely uncorrelated from the large scale
structure present in recent times; on the other hand, the iSW temperature correlates
with the density of galaxies, which should trace the potential wells and hills which
bring about the anisotropies. We can then extract the late iSW signal by measuring
the cross-correlation of some tracer of the large scale structure—typically a galaxy
survey—with the CMB.
Unfortunately the ability to detect the cross-correlation is limited because the
signal falls off on small scales. Not only is cosmic variance an important factor, but
there is also the fortuitous correlation that can happen between the galaxy surveys
and the CMB anisotropies produced at last scattering. Many groups have tried to
detect the late iSW effect in the past decade using this cross-correlation technique,
using successively the COBE, WMAP, and Planck data for the CMB, and a large
variety of galaxy catalogues observed in different regions of the electromagnetic
spectrum. However, as described in details in Sect. 2.1.4 of this manuscript, some
inconsistencies between studies appear, and a clear and definitive signal has yet to
be found.
During my Ph.D. thesis, I focused on a new and innovative tracer of the matter
density to be used in such cross-correlation methods, namely the Cosmic Infrared
Background. As I show in Chap. 2 this particular background presents all the required
characteristics of a good tracer for such studies, while having none of the shortcom-
ings of the current surveys. Another probe of the iSW effect was recently introduced
by Granett et al. [15], and is based on detecting the correlation in a localised way:
these authors considered the luminous red galaxies from the SDSS, and studied
regions in the sky containing superclusters and supervoids of scale ∼140 Mpc, and
then stacked the CMB signal at these positions on the sky. The result shows that
the CMB appears in average hotter at the locations of clusters and colder at those
of voids, at a large significance, but with peculiar features that I point out in Chap. 3.
This is an interesting result because it represents the first localisation of the iSW
signal, but it has the drawback of being dependent on the particular choice of super-
clusters and on the particular filter scale adopted for the signal detection—caveats
that I alleviated in my work, by performing a more thorough analysis and consid-
ering more recent (and much more consequent) alternative catalogues of structures,
all described in Chap. 3. These new results of mine prompted even more questions
regarding the nature of the iSW effect and its signature in such a context, which lead
me to devise a new approach to predicting the expected iSW effect from structures
in the Universe, detailed in Chap. 4. Interestingly enough, it makes use of the LTB
metric (mentioned earlier in the context of DE models) but in a totally different
context.
1.5 A Glimpse at the Dark Components Through

the Reionisation
In parallel to all my primary work related to the iSW effect, I also carried out a work
focusing on a particular epoch of the history of the Universe, namely the reionisation
(briefly mentioned earlier in this introduction). This era corresponds to the transition
of our Universe from a neutral to a fully ionised state, due to the apparition of the
first stars and other astrophysical sources of ionisation. This period has a significant
influence on the CMB and its statistical properties, in particular the power spectrum
of its polarisation fluctuations (a more detailed description of this epoch is reported
in Chap. 5). I took a particular interest during my thesis in the rôle that played
another actor of the Dark Sector in the framework of cosmic reionisation: the Dark
Matter (DM), whose possible influence has the potential of being traceable, and
used to constrain DM models. Indeed, as shown in detail in Chap. 5, temperature
measurements of the intergalactic medium (IGM) during reionisation can be used
to investigate the possible contribution of the decay and annihilation of hypothetical
Dark Matter particles, and therefore derive some potential constraints on their nature.
References
1. J.K. Adelman-McCarthy et al., The Sixth Data Release of the Sloan Digital Sky Survey. ApJS
175, 297–313 (2008)
2. N. Aghanim, S. Majumdar, J. Silk, Secondary anisotropies of the CMB. Rep. Prog. Phys. 71(6),
066902 (2008)
References 13
3. R.A. Alpher, H. Bethe, G. Gamow, The origin of chemical elements. Phys. Rev. 73, 803–804
(1948)
4. L. Amendola, S. Tsujikawa, Dark Energy: Theory and Observations (Cambridge University
Press, Cambridge, 2010)
5. E. Bertschinger, Cosmological dynamics. NASA STI/Recon. Tech. Rep. N 96, 22249 (1995)
6. P. Bull, T. Clifton, P.G. Ferreira, Kinematic Sunyaev-Zel’dovich effect as a test of general radial
inhomogeneity in Lemaître-Tolman-Bondi cosmology. Phys. Rev. D 85(2), 024002 (2012)
7. R.G. Crittenden, N. Turok, Looking for a cosmological constant with the Rees-Sciama effect.
Phys. Rev. Lett. 76, 575–578 (1996)
8. R.H. Dicke, P.J.E. Peebles, P.G. Roll, D.T. Wilkinson, Cosmic black-body radiation. ApJ 142,
414–419 (1965)
9. S. Dodelson, Modern Cosmology (Academic Press, San Diego, 2003)
10. R. Durrer, in Cosmological Perturbation Theory, ed. by E. Papantonopoulos. Lecture Notes in
Physics, vol. 653 (Springer, Berlin, 2004), p. 31
11. G. Dvali, G. Gabadadze, M. Porrati, 4D gravity on a brane in 5D Minkowski space. Phys. Lett.
B 485, 208–214 (2000)
12. D.J. Eisenstein et al., Detection of the Baryon acoustic peak in the large-scale correlation
function of SDSS luminous red galaxies. ApJ 633, 560–574 (2005)
13. K. Enqvist, Lemaître Tolman Bondi model and accelerating expansion. Gen. Relativ. Gravit.
40, 451–466 (2008)
14. D.J. Fixsen, The temperature of the cosmic microwave background. ApJ 707, 916–920 (2009)
background due to the integrated Sachs-Wolfe effect. APJ 683, L99–L102 (2008)
16. W. Hu, N. Sugiyama, J. Silk, The physics of microwave background anisotropies. Nature 386,
37–43 (1997)
17. W.T. Hu, Wandering in the Background: A Cosmic Microwave Background Explorer. PhD
thesis, University Of California, Berkeley, 1995
18. J. Huchra, M. Davis, D. Latham, J. Tonry, A survey of galaxy redshifts IV—the data. ApJS 52,
89–119 (1983)
19. L.A. Kofman, A.A. Starobinskii, Effect of the cosmological constant on largescale anisotropies
in the microwave background. Soviet Astron. Lett. 11, 271–274 (1985)
20. E.W. Kolb, S. Matarrese, A. Riotto, On cosmic acceleration without dark energy. New J. Phys.
8, 322 (2006)
21. E. Lifshitz, On the Gravitational stability of the expanding universe. J. Phys. (USSR), 10, 116
(1946)
22. E.V. Linder, The dynamics of quintessence, the quintessence of dynamics. Gen. Relativ. Gravit.
40, 329–356 (2008)
23. A. Lue, R. Scoccimarro, G. Starkman, Differentiating between modified gravity and dark
energy. Phys. Rev. D 69(4), 044005 (2004)
24. J.C. Mather et al., Measurement of the cosmic microwave background spectrum by the COBE
FIRAS instrument. ApJ 420, 439–444 (1994)
25. G. Meylan, et al., (eds.) Gravitational lensing: strong, weak and micro. A&A 452, 51–61 (2006)
26. A.A. Penzias, R.W. Wilson, A measurement of excess antenna temperature at 4080 Mc/s. ApJ
142, 419–421 (1965)
27. W.J. Percival et al., Measuring the Baryon acoustic oscillation scale using the Sloan digital sky
survey and 2dF galaxy redshift survey. MNRAS 381, 1053–1066 (2007)
28. S. Perlmutter et al., Measurements of Omega and Lambda from 42 high-redshift supernovae.
ApJ 517, 565–586 (1999)
29. Planck Collaboration, Planck 2013 Results XII, Component Separation. ArXiv:1303.5072
(2013a)
30. Planck Collaboration. Planck 2013 Results. I. Overview of Products and Scientific Results.
ArXiv:1303.5062 (2013b)
31. Planck Collaboration, Planck 2013 Results. XVI. Cosmological, Parameters. ArXiv:1303.5076
(2013c)
32. B. Ratra, P.J.E. Peebles, Cosmological consequences of a rolling homogeneous scalar field.
Phys. Rev. D 37, 3406–3427 (1988)
33. M.J. Rees, D.W. Sciama, Large-scale density inhomogeneities in the universe. Nature 217,
511–516 (1968)
34. A.G. Riess, et al, Observational evidence from supernovae for an accelerating universe and a
cosmological constant. AJ 116, 1009–1038 (1998)
35. R.K. Sachs, A.M. Wolfe, Perturbations of a cosmological model and angular variations of the
microwave background. ApJ 147, 73 (1967)
36. B.M. Schäfer, Mixed three-point correlation functions of the non-linear integrated Sachs-Wolfe
effect and their detectability. MNRAS 388, 1394–1402 (2008)
37. L. Sodré Jr, Cosmology with large galaxy redshift surveys, ed. by J. Alcaniz, et al. American
Institute of Physics Conference Series, vol. 1471, pp 22–26 (2012)
38. N. Suzuki, et al., The hubble space telescope cluster supernova survey. V. Improving the dark-
energy constraints above z > 1 and building an early-type-hosted supernova sample. ApJ 746,
85 (2012)
39. S. Weinberg, The cosmological constant problem. Rev. Mod. Phys. 61, 1–23 (1989)
40. S. Weinberg, Cosmology (Oxford University Press, London, 2008)
41. C. Wetterich, Cosmology and the fate of dilatation symmetry. Nucl. Phys. B 302, 668–696
(1988)
Chapter 2
Unravelling the iSW Effect Through
the Matter Distribution
2.1 CMB Cross-Correlation with Tracers of Matter
The end of the last chapter painted the iSW effect of CMB photons as a clever probe
that has the potential to independently prove the existence of Dark Energy in the
Universe. However, the faintness of this signal makes it almost impossible to detect,
at least using only the CMB as such. The story does not end here though, thanks to the
resourcefulness of cosmologists at finding ways to detect and exploit the iSW effect
more efficiently. Among the methods devised, one of them stands as the leading
technique in the literature: the cross-correlation of the CMB with the distribution of
matter in the Universe.
2.1.1 The Principle
Before going into the details of this approach, let us recall the main ideas it is based
on. As discussed in Chap. 1, the iSW effect felt by the CMB photons in a CDM
universe is a result of the stretching of the large scale potentials in the Universe,
caused by the acceleration of the expansion which is itself due to the presence of
Dark Energy. Of course, one has to remember that these gravitational potentials
originate from the presence of matter in the form of large Dark Matter halos in which
sit clusters of galaxies. Following this, it is reasonable to think that there exists
a certain degree of correlation between the distribution of matter and the resulting
pattern of the iSW effect that it generates in the CMB across the sky. The link between
these two elements is of course not trivial since, as its name emphasizes, the iSW
effect is an integrated effect: CMB photons that come from a particular direction
in sky have been affected by the matter distribution along the whole line of sight.
This is then further complicated by two competing phenomena: on the one hand the
iSW effect is expected in the CDM paradigm to be redshift-dependent with the
most recent structures in the Universe yielding a more pronounced effect as they are

16 2 Unravelling the iSW Effect Through the Matter Distribution
locally more DE-dominated. However, these very structures are also closer and closer
to us so that their number becomes limited by the available volume around us. At
some point, the stronger iSW effect of the closest large structures becomes balanced
by their increasing scarcity and therefore gets progressively harder to detect. We
therefore intuit already that there will be an optimal redshift for the detection of the
iSW effect, at some point between z = 0 and the start of the era in which Dark
Energy became cosmologically important.
Now that we have a grasp of the motivations behind the use of cross-correlation
and its associated features, we will have in the next subsection a closer look at the
theoretical tools used to describe this correlation.
2.1.2 Theoretical Framework
As seen in Sects. 1.2 and 1.4, the anisotropies that are generated by the iSW effect
are directly correlated to the distribution of matter through the evolution of the
gravitational potential σ̇. Exploiting this correlation and detecting the iSW effect can
be done via several approaches; I will focus here on the most widely used method
which is done in spherical harmonic space.
All cross-correlations methods suppose that we have first a map of the tempera-
ture of the CMB T (n̂) at hand, or rather, a map of the relative fluctuations of this
background:
T (n̂) − T̄
νT (n̂) = (2.1)
T̄
with T̄ the mean temperature of the CMB (with the latest studies indicating
T̄ = 2.7260 ± 0.0013, cf. [15]). On the other hand, we need a survey that traces
the distribution of galaxies, from which we derive a map of the projected galaxy
overdensity field:
N (n̂) − N̄
νg (n̂) = (2.2)
N̄
where N (n̂) is the number of galaxies in the pixel corresponding to the direction n̂ and
N̄ is the mean number of galaxies per pixel. Both maps are therefore in dimensionless
units—which is always welcomed to simplify calculations. Then, for the approach
that I consider here, we use the fact that any field can be decomposed into a series
of functions which form an orthonormal set, as do the spherical harmonic functions
Yδm (θ, φ). It follows that both the CMB temperature map (νT ) and the overdensity
map (νg ) can be decomposed into

ν X (θ, φ) = X
aδm Yδm (θ, φ), (2.3)
δ,m
2.1 CMB Cross-Correlation with Tracers of Matter 17
X (X = g, T ) are the spherical harmonic coefficients of the field resulting

where aδm
from its decomposition. The main quantity of interest for us is the 2-point galaxy-
temperature cross-correlation as a function of the multipole δ, whose estimator is:
1 1
g T ∗ g
Ĉ T g (δ) = Re aδm (aδm ) = Re aδm
T
(aδm )∗ , (2.4)
(2δ + 1) m (2δ + 1) m
It traces the degree of correlation between the two maps in harmonic space. We can
intuit that in the absence of Dark Energy and hence of the iSW effect, the correlation
between these two will be reduced to only fortuitous coincidences. But now that
we have defined its computation, let us take a more precise look at the expected
cross-correlation signal C T g (δ).
Firstly, the fluctuations of the CMB temperature νT (n̂) are known to be com-
posed of several contributions, often categorised into primordial and secondary
anisotropies. However, for the large scales that we consider the only ones that are
correlated to the distribution of matter in the Universe are the secondary anisotropies
generated through the late iSW effect. In the remainder of this section, I will asso-
ciate the notation νT (n̂) to these iSW fluctuations only. One way of expressing these
temperature fluctuations on a particular line of sight is written as the following
redshift-integral from the surface of last scattering (SLS) to us:
0
νT (n̂) = e−τ (z) (σ̇ − ˙ )[n̂, z] dz. (2.5)
z SLS
where the dot denotes here differentiation with respect to z (details about the equation
and its terms can be found in Sect. 1.4). Since the matter density is related to the
gravitational potentials σ and by the Poisson equation, these iSW temperature
fluctuations will be related to the observed galaxy density contrast, given by:
0
dN
νg (n̂) = bg (z) (z)νm (n̂, z) dz. (2.6)
dz
z SLS
In this expression, dN /dz is called the selection function of the survey (from which
the density map is derived) and represents simply the redshift distribution of galaxies
in the survey. More accurately, this function gives the number of galaxies contained
in a shell of width dz at redshift z; it is often normalised and then describes the
fraction of objects per redshift (an example of such selection function is shown in
Fig. 2.1). The term νm corresponds to the matter density perturbations, which are
related to the galaxy overdensities by a factor bg : indeed, although we correlate in
practice the CMB map with a galaxy map, we aim at probing in reality the correlation
with the underlying distribution of Dark Matter. This “galaxy bias” can theoretically
evolve in time and be a function of scale. However, it is generally assumed to be
Selection function
3.0
2.5
2.0
dN/dz
1.5
1.0
0.5
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Redshift z
Fig. 2.1 Redshift distributions of the two fictitious surveys considered (see text for details). These
selection functions follow a widely-used analytical expression that describes well such quantities:
d N /dz(z) = Az m exp −(z/z 0 )η , where A is chosen so that the integral of the function is normalized
to unity. The parameters m and η control the slope √of the rise and fall respectively, while z 0 fixes
the median redshift of the distribution z med (= z 0 2). Here I chose z 0 = 0.48287, m = 1.51964
and η = 2.34207 for the SDSS-like distribution (red solid curve), and increased z 0 to 0.7 for the
second one (blue dashed curve)
time and scale independent for simplicity. For our purposes, a time dependent bias
would be equivalent to changing the selection function of the survey. A possible scale
dependence of the bias is more problematic, but on the very large scales (>10 Mpc)
we are considering, the scale dependence is expected to be weak (see e.g. [8, 33]).
We are here interested in the correlation between νT and νg : their angular cross-
correlation function in real space is defined as:
C T g (τ) → ∼νT (n̂1 )νg (n̂2 ) (2.7)
with the average carried over all the pairs at the same angular distance τ = |n̂1 − n̂2 |.
As mentioned before, it is often preferred to work in the harmonic space and study
the cross-correlation spectrum C T g (δ) instead of the function C T g (τ). Those two
quantities are related through the Legendre polynomials Pδ :
∝
2δ + 1 Tg
C T g (τ) = Cδ Pδ [cos(τ)]. (2.8)
4π
l=2
After some calculations (for detailed steps, see e.g. [16]), it follows that the cross-
correlation power spectrum is given by:

Tg dk g
Cδ = 4π 2
(k)IδiSW (k)Iδ (k), (2.9)
k
where (k) is the almost scale invariant initial matter power spectrum 2 (k) →
4π k 3 P(k)/(2π )3 and the two integrands are respectively:

IδiSW (k) = −2 e−τ (z) (σ̇k − ˙ k ) jδ [kχ (z)]dz (2.10)

g dN
Iδ (k) = bg (z) (z)νm (k, z) jδ [kχ (z)]dz, (2.11)
dz
where σk , k and νm (k, z) are the Fourier components of the gravitational poten-
tials and matter perturbations for the wavenumber k, jδ (x) are the spherical Bessel
g
functions and χ is the comoving distance. The two integrands IδiSW and Iδ can then
be calculated for a given cosmological model using numerical codes that compute
all the relevant quantities needed (the details of the corresponding equations can be
found in [16]).
Of interesting note is that in the linear theory of perturbations (an approximation
which should be more than valid here considering the large scales involved) the
growth of the gravitational potentials is directly proportional to the ratio D(a)/a
(cf. Sect. 1.4). As a result, in a flat, matter-dominated universe the terms σ̇k and
˙ k in Eq. (2.10) would be equal to 0 and so would the integral—and the cross-
Tg
correlation spectrum Cδ itself. Therefore, we confirm here that in the absence of
Dark Energy (or more accurately, if the dominant component of the Universe were
pressureless matter) there would be no iSW effect and no correlation between the
CMB temperature and the distribution of matter.
2.1.3 Application to Detectability and Test Case

We reviewed in the previous section the analytic expression of the correlation between
the CMB anisotropies and the distribution of matter through their cross-correlation
power spectrum. The question now arises of the use of these theoretical predictions
for the study of the Dark Energy.
The first application of these theoretical tools is to give the possibility to predict
the detectability of the aforementioned correlation. Assuming a given cosmological
model and that both the CMB temperature and the galaxy maps behave as Gaussian
random fields, the covariance of the iSW cross-correlation signal (in the absence of
noise) can be calculated by:
Tg gg
Tg (Cδ )2 + Cδ CδT T
σ 2 [Cδ ] = , (2.12)
2δ + 1
where CδT T is the full CMB temperature-temperature power spectrum (and not only
gg
the secondary anisotropies generated by the iSW effect) and Cδ is the galaxy auto-
correlation function that can be calculated theoretically:

gg dk g g
Cδ = 4π 2
(k)Iδ (k)Iδ (k) (2.13)
k
(see Eq. (2.9) for a description of the terms). We can identify two sources of variance
Tg
in Eq. (2.12): the cosmic variance of the correlation itself (the (Cδ )2 term), and
the fortuitous coincidences that arise between the CMB temperature and the galaxy
gg
distribution (the Cδ CδT T term).
It follows then that the theoretical signal-to-noise ratio (S/N) of this iSW detection
for a given multipole range [δmin , δmax ] is:
2 δ Tg δ Tg
S max
(Cδ )2 max
(Cδ )2
= Tg
= (2δ + 1) Tg gg
. (2.14)
N
δ=δmin σ 2 [Cδ ] δ=δmin (Cδ )2 + Cδ CδT T
The cumulative character of the S/N is due to the fact that, in the ideal case we
consider here (Gaussian fields, full-sky maps), the power spectrum estimates at dif-
ferent multipoles are independent from one another (no off-diagonal terms in the
covariance matrix).1 For illustration purposes, I will consider two fictitious full-sky
surveys, one with a selection function (see Eq. 2.6) similar to the eighth data release
(DR8) of the Sloan Digital Sky Survey (SDSS, [4]) and one with a higher median
redshift (both are shown in Fig. 2.1). Assuming the best-fit CDM cosmology from
gg
Planck [38], I calculate the theoretical auto-correlation (Cδ ) and cross-correlation
Tg
(Cδ ) spectra for such surveys. To do so, I use a modified version of the cosmolog-
ical code CMBFAST [42] named CROSS_CMBFAST [12]. This code computes the
iSW-correlation power spectrum and the 2-point angular iSW-correlation function
for a given galaxy window function, and is limited to flat geometry. It includes dark
energy models specified by a constant equation of state or a linear parametrisation
in the scale factor and a constant sound speed. In addition to the auto- and cross-
correlation spectra, the computation also includes the usual CMBFAST outputs, i.e.
the CMB temperature and polarisation spectra.
I illustrate in Fig. 2.2 the resulting cross-correlation spectra and the ideal S/N
estimation defined in Eq. (2.14): as we can see, the signal peaks in the δ = 10–20
multipole range and quickly falls off at smaller scales. Similarly, the largest contri-
butions to the signal-to-noise ratio come mainly from the lowest multipoles, with the
total S/N quickly reaching a plateau: this shows that most of the significant signal
is below δ = 100, at that is it pointless to consider higher multipoles in cross-
correlation studies. It is important to notice here the crucial influence of the redshift
range (through the selection function) covered by the surveys considered, as already
intuited at the end of Sect. 2.1.1. Here, we can witness that the S/N is substantially
higher for the survey with the higher median redshift compared to the SDSS-like one
(4.5 vs. 3.3 for the cumulative S/N).
1 This is often not exactly true when working with real datasets (due to partial sky coverage,
non-Gaussian contamination, etc.) and has to be accounted for properly, e.g. using Monte-Carlo
simulations (see later in this chapter).
Cross spectrum S/N by multipole Cumulative S/N from l=2 to l max

10-7 5
1.000
-8 0.100
10
l(l+1)/2π CTg
3
l
Total S/N
(S/N) 2
0.010 2
10-9
0.001 1
10-10 0
1 10 100 1 10 100 1 10 100
Multipole l Multipole l Multipole l max
Fig. 2.2 Theoretical cross-correlation results for the SDSS-like survey (red solid curves) and a
second one with higher median redshift (blue dashed curves). Left panel theoretical angular cross
power spectra of the CMB-galaxy correlation. Middle panel contribution to the total squared S/N of
the signal as a function of the multipole δ. Right panel total cumulative S/N of the cross-correlation
signal for the multipole range δ = [2, δmax ] as a function of δmax
The above test cases give us an idea about the expected characteristics of the
cross-correlation in terms of the signal itself and its detectability; however, the pre-
dictive tools used here can be extended to any given survey (through the selection
function) and to a large variety of cosmologies (included in the CROSS_CMBFAST
code). We can now wonder: how can we use these tools to constrain cosmological
models and learn more about the Dark Energy? There are several tests that we can
think of and that are used in the literature; I will briefly present the most widely used
without diving too deep into calculations (a review of these methods can be found in
[14]). The first one is actually independent of the cosmology and does not assume
any kind of DE model (besides its existence itself): it consists in measuring the
cross-correlation on the data and checking how much it departs from a scenario with
no correlations at all, i.e. without Dark Energy. This is the so-called “null hypothe-
sis”, and basically consists in performing a χ 2 test on the measured correlation with
respect to a “null model”, i.e. a model with zero correlation. If the test shows a signif-
icant deviation from the null model, then it constitutes a proof of the existence of the
Dark Energy (assuming a flat Universe) but does not give any additional information
beyond that. In a second time, we can go further and try to check how the data fares
against the prediction from a given cosmological model: using a similar χ 2 test, we
can compare the measured correlation with the theoretical (non-zero) one predicted
beforehand using the framework that I described above. Although this test cannot
give a definitive answer on whether a given model is “the right one”, inversely it can
invalidate the model if the computed χ 2 is too high. A third and last approach, called
“amplitude fitting” (or “template matching”), combines both previous ideas: con-
sidering a cosmological model (often the fiducial one that we try to (in)validate),
we construct a “template” by multiplying its associated theoretical correlation
by an “amplitude factor”. The method consists then in computing the amplitude
(and associated error) that fits the best the measured correlation: a value close to 1
then indicates an agreement with the underlying model. If this amplitude simulta-
neously shows a sufficiently large S/N (i.e. a small associated error), it intrinsically
disproves the null hypothesis and strongly hints at the presence of Dark Energy.
We can already intuit here that the presence of sources of noise and partial sky
coverage will complicate the extraction of the iSW signal. The next paragraph will
be an overview of the current state of the detection of the iSW effect through cross-
correlation techniques.
2.1.4 State-of-the-Art of the iSW Detection
In the literature, many attempts have been made to detect an iSW signal through the
cross-correlation of the CMB with galaxy surveys, with varying degrees of success. I
do not intend to summarize here all the results so far, since such a type of compilation
has already been made in the past: for reference, in Table 1 of [14] the authors
present a “meta-analysis” of iSW detections (up to the publication of their work)
and their reported statistical significance. Over the last decade a large variety of
surveys has been explored, exploiting the whole spectrum of light: X-ray [9, XRB
survey]; optical [1, 3, SDSS], near infrared [23, 2MASS]; radio [10, NVSS]. The
cross-correlation studies also followed the evolution of CMB observations, from
the first detected anisotropies of the Cosmic Microwave Background Explorer [5,
COBE] to the succession of releases by the Wilkinson Microwave Anisotropy Probe
[43, WMAP] and the latest publication of the Planck satellite [37].
What strikes the most when reviewing the current results of the literature are their
wide diversity, as the reported significances range from negligible [40] to 4.4 σ [18].
Although these differences can be partly attributed to the specific features2 of each
survey (and CMB maps) and to the methods3 used by the authors of each work,
some puzzling discrepancies are still present. Indeed, a couple of analyses based of
very similar (if not identical) datasets have yielded contradictory conclusions on the
level of detection of an iSW signal (see [14], again for reference), while other works
reported a signal at odds with CDM expectations (1 σ excess in [18], 2 σ in [21]).
This intriguing situation may find its source in the characteristics of the current
(and past) generation of surveys, or rather, their shortcomings. Indeed, for each
of them, a trade-off has often been made between the deepness of the survey and
its coverage, resulting in the end in the observation of a small total volume of the
Universe. Unfortunately, this is a crucial point for the iSW studies and has a dramatic
impact on the potential S/N of the detection: considering a smaller volume results
2 In terms of the redshift range of the surveyed objects and the fraction of the sky covered by the
survey.
3 Not all of the cross-correlation works base their analysis in harmonic space, as some prefer working
in real space or using wavelets; although the tools and estimators are different, the different tests
(χ 2 , null hypothesis,…) remain however the same and similar to those described in Sect. 2.1.3.
simultaneously in a decrease of the theoretical signal and an increase of its variance.

Coupled to this is the intrinsic weakness of the signal, and the presence of galactic
foregrounds in both the CMB and the LSS maps that also mask crucial large scale
data and can introduce spurious correlations. Therefore, any method claiming to
detect the iSW effect may not be as thorough as required in accounting for missing
data. Ideally, any reported detection level should be independent of any assumption
or particular cosmology. Very recent works aimed at a better control of systematics
(see e.g. [20], for a thorough review of potential contaminants in SDSS), while others
made use of the latest releases of both CMB and LSS data [36]. In both cases, the
correlations were found to have much less discrepancies with CDM and a good
agreement with their expected amplitude (cf. “amplitude fitting” tests in Sect. 2.1.3),
but with relatively low significances (1.5 − 3 σ range).
It is quite established that current surveys are far from ideal: to further the point, [2]
showed that obtaining a near optimal iSW detection (at a ∼ 5 σ level) would require
an all-sky survey with about 10 million galaxies almost uniformly distributed within
0 < z < 1, with systematics below ∼0.1 % and systematic errors in redshift estimates
<0.05—again, far from the characteristics of currently available datasets. However,
the next generation of probes will address many of the current shortcomings: surveys
such as Euclid, Pan-STARRS or LSST will get much closer to the ideal requirements
for the detection of the iSW effect (see [13, 14] for forecasts) and will also—needless
to say—improve our knowledge of a vast range of cosmological topics. For the time
being, considering the landscape depicted above, there are a few possible orientations
that can be taken regarding iSW studies: the first is to make the most of the current
available data and has already been explored quite extensively, notably through the
combination of surveys in the context of LSS/CMB cross-correlation (see e.g. [18,
21]). However, some innovative ways of using the LSS and CMB are also explored,
such as the study of the impact of individual structures in the CMB that I will develop
in Chap. 3. A second possible line of research would be to anticipate the future release
of data from the next generation of surveys and develop optimised tools in advance,
in order to make the most of the improved accuracy and avoid potential biases and
caveats: this will be the topic of the next section of this chapter (Sect. 2.2). Finally,
considering the great time scales involved in the schedule of the aforementioned
surveys, an interesting approach would also be to look for alternative, promising, and
already available tracers of matter (and gravitational potentials) to cross-correlate to
the CMB: this will be the central point of the Sect. 2.3, through the use of the Cosmic
Infrared Background.
2.2 Optimising the Cross-Correlation for iSW Detection
The work that I will present in this section originates from a pre-thesis internship:
it aims at exploring, optimising and proof-testing a protocol that I devised for the
exploitation of the iSW effect, through CMB/LSS cross-correlation and in the context
of next-generation surveys. With this prospective work I have no pretension of giving
here a complete and exhaustive method for the analysis of these datasets, but I will
focus on a few crucial points that allowed me to point out some interesting features
and potential problems of this kind of analysis.
2.2.1 Context and Tools
Being entirely theoretical in nature, this study required me to simulate realistic

datasets and analyse them thoroughly—a task that required the use of accurate
numerical code and tools. The first requirement was to be able to predict, for a given
“next-gen-like” survey and a cosmology, the theoretical cross-correlation between
the CMB and these LSS, as well as their auto-correlation. If we assume that these
two fields are Gaussian in nature (which I will do), the only quantities that we need
to compute are their angular cross- and (respective) auto-correlation power spectra
in harmonic space. To do so I used the numerical code CROSS_CMBFAST that I
already mentioned in Sect. 2.1.3.
To develop a little further on this particular code, its specificities, and qualities, it
has the particular feature of adopting a top-down approach, starting from the primor-
dial spectrum of fluctuations. This differs from other similar studies where the starting
point is the present matter power spectrum (which is evolved backward in order to
find its correlation with the CMB). The approach of CROSS_CMBFAST unifies the
treatment of CMB and matter power spectra, and is more convenient for taking full
account of possible fluctuations in the Dark Energy. Moreover, since all perturbations
are evolved numerically with the CMBFAST code, it does not resort to the frequently
used approximate analytical expressions for the growth function (mentioned at the
end of Sect. 2.1.2), nor to the also commonly used “Limber approximation” for small
angles [28].
In order to thoroughly test my protocol (described in the next section) against “real-
life” situations, I also needed to work with realistic simulations of CMB and LSS
maps. For the creation and efficient exploitation of such maps I used the HEALPix
(Hierarchical Equal Area isoLatitude Pixelization) package [19]: this suite of codes
is based on a specific pixelization scheme of a spherical surface that is particularly
suited for fast and accurate statistical and astrophysical analysis of massive full-
sky data sets. It contains all the required tools to produce maps based on Gaussian
realisations of a given power spectrum, as well as perform a decomposition of a
map into its spherical harmonic coefficients (and consequently its angular power
spectrum).
2.2.2 Simulations, Covariance Estimation and Parameter

Recovery
The main idea that motivated my work presented here was to investigate a way to
recover as much information as possible on the Dark Energy from a single survey
2.2 Optimising the Cross-Correlation for iSW Detection 25
Selection function Cumulative S/N

1.0 8
0.8
6
Total S/N
0.6
dN/dz
4
0.4
2
0.2
0.0 0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 1 10 100
Redshift z Multipole l max
Fig. 2.3 Left panel normalised redshift distribution of my fiducial survey, following a similar
analytical form as the one mentioned in Fig. 2.1. The parameters chosen here are z 0 = 0.95,
m = 1.9 and η = 1.5. Right panel total cumulative S/N of the theoretical cross-correlation signal
for this survey as a function of δmax
with ideal characteristics (full-sky coverage, large redshift range, noiseless) and its
cross-correlation with the CMB. I started by fixing a reference cosmological model,
that I chose to be at that time the best-fit CDM model derived from the WMAP
7-year data [27]. In the meantime, I also fixed the characteristics of the mock survey
that I considered, through the choice of its selection function—the only quantity
needed to model a survey (in the linear regime) beside the cosmology; I illustrate
this survey in Fig. 2.3.
In this particular context, the information that we are trying to obtain from
CMB/LSS correlation becomes simply the density parameter of the Dark Energy.
From there, the protocol that I devised for the reconstruction is the following:
• I consider a CMB map and a galaxy overdensity map;
• I assume that the selection function of the galaxy survey has already been estimated
beforehand by other means;
• I use the tools of the HEALPix package to extract the cross-correlation power
Tg
spectrum Cδ (data) from the two “data” maps;
• I use the CROSS_CMBFAST code to compute the expected theoretical cross power
Tg
spectra Cδ ( ) for a large range of values (between 0.3 and 0.9);
• Using a χ 2 test, I look for the model that fits best the measured spectrum and keep
the corresponding as the reconstructed DE density parameter.
Although this protocol is fairly straightforward, there is a specific point that requires a
particular attention: the computation of the χ 2 , or more specifically the computation
of the covariance matrix involved in the calculation. Indeed, for a given measured
spectrum Cδdata , the χ 2 of a given model Cδtheo is given by:

χ2 = T
Cδdata − Cδtheo M−1 Cδdata − Cδtheo , (2.15)
M being the covariance matrix, each of its elements containing the covariance of
the cross-correlation power spectrum Cδdata between two multipoles, i.e.:

Mi j = Covar Cδdata
i
, Cδdata
j
. (2.16)
The tricky part here is that there are several methods for computing this matrix, that
differ quite substantially from each other. For the simplest approach, we suppose that
the survey considered is quite ideal and that there is no correlation between multi-
poles, which is generally true only for full-sky surveys. It follows that the covariance
matrix is diagonal and that an analytical approach is sufficient to compute its elements
using the theoretical covariance already shown in Eq. (2.12) of Sect. 2.1.3. However,
this approach is inadequate for more realistic scenarii, since galaxy surveys are often
plagued with partial sky coverage, and contaminants that may further complicate
the analysis. Although I do consider an ideal survey here, it is still necessary to go
beyond this simple case, especially from the perspective of the potential application
of the protocol to real datasets.
For these reasons, two alternative methods have been devised, often called “MC1”
and “MC2” in the literature (see e.g. [17] for an application of these approaches).
They are both based on Monte-Carlo techniques (hence the “MC”) and consist in
generating a very large number of as-accurate-as-possible maps (CMB and/or LSS)
of which we derive the covariance matrix of the cross-correlation. To be more precise,
let us start with the MC2 method: assuming a fiducial model, the principle of the
method is to generate N pairs4 of galaxy and CMB maps with the same characteristics
as the original ones and their expected cross-correlation. In practice, these maps are
obtained by generating Gaussian realisations of the theoretical power spectra derived
from the fiducial cosmology and the knowledge of the considered survey (for the
details of the calculations, see e.g. [17]). We then reproduce in these maps the known
features of the original maps, whether it is a partial sky coverage (with a mask) or
a known contaminant. From there, we compute the cross power spectrum for each
of these N pairs of maps, and then derive each element of the covariance matrix
according to:
1 k
N
Mi j = Covar Cδi , Cδ j = [Cδi − C δi ][Cδk j − C δ j ] (2.17)
N
k=1
where the Cδk are the cross power spectra of each simulated pair, and C δ is the
average of these power spectra over the N pairs. Now the only difference between
the two Monte-Carlo approaches is that in the MC1 method, only maps of the CMB
are generated. They are then correlated to the true galaxy map, and from these
correlations are computed the elements of the covariance matrix.
4 With N typically of the order of ten thousands.

Fig. 2.4 Normalised his-

tograms of the reconstructed
values of by the protocol
described in the text, using
either the MC1 (blue) or the
MC2 (green) method for the
computation of the covariance
matrix involved in the χ 2 test.
The true input value of is
indicated by the red vertical
line and is equal to the best-fit
value of the parameter from
WMAP7, = 0.734
The MC1 is the most widely used estimator in the literature: reasonably fast
to implement, it accounts for the cosmic variance and the accidental CMB/LSS
correlations, supposedly the primary source of error. However it does not account
for the variance in the density maps since only CMB maps are randomly generated.
The MC2 method alleviates these problems by also generating random density maps
(based on a fiducial cosmology and the selection function). However, this method is
more time demanding (more maps to generate) and requires a robust knowledge of the
features of the considered survey (including its selection function and systematics)
to accurately simulate realisations of it. Finally, a shortcoming common to both
approaches is that they are model dependent and could fail if the data model is
poorly understood (e.g. non-Gaussianity of the maps).
In the remainder of this section, I will focus on how these two MC methods fare
in the context of the objective of my protocol, i.e. the reconstruction of the
parameter. After fixing the input cosmology, I thoroughly tested each method by
repeating the following steps thousands of times:
• Simulation of pair of CMB/density maps with the correlation expected from the
chosen cosmology;
• Application of my protocol described earlier to search for the best-fitting ;
• Storage of the reconstructed value in a histogram.
A comparison of the recovered values for the two MC methods, along with the
input
input value (here = 0.734, WMAP7 best-fit), is shown in Fig. 2.4. The most
striking feature here is the bias of the MC1 method towards smaller values:
although the width of both distributions is quite large (with a noticeably larger trend
for MC1), the peak for the MC1 method is still clearly shifted (to ∼ 0.69), while
the MC2 method shows a much better agreement (peak at ∼ 0.74) with the input
value.
We can intuit an explanation for the observed difference between the two methods,
which revolves around the asymmetrical nature of the MC1 approach. In the ideal
case considered here (no noise, full-sky maps), there are but very little correlations
between multipoles: the covariance matrix is nearly diagonal, and the values of its
elements tend to the analytical expressions given by Eq. (2.12). Due to the intrinsic
Tg
weakness of the iSW signal (i.e. Cδ is relatively small), we can simplify this equation
even further and write: gg
Tg C CTT
σ 2 [Cδ ] ∼ δ δ . (2.18)
2δ + 1
In the MC2 method, the matrix is computed once and for all using thousands of
simulated maps, and is completely independent of the data considered. However,
in the MC1 approach, only one density map (the data) is used in the computation
of the covariance matrix. As a consequence, the resulting variance will be directly
proportional to the spectrum of this particular map, according to Eq. (2.18). Over
the thousands of repeated tests that we performed, the Gaussian realisations of this
spectrum will be sometimes lower, sometimes higher than the fiducial model, in equal
proportions. Then, as said above the “low” realisations will give a smaller covariance
matrix (i.e. smaller errors bars), pulling down the value preferred by the χ 2 test.
On the other hand, the “high” realisations will have larger error bars, and therefore
will not have the same “pull” towards high values. This asymmetry in the MC1
method is the most likely explanation for the over-abundance of smaller recovered
values when we repeat the test many times.
It stems from the previous analysis that the MC2 method should be the preferable
choice in order to avoid a possible bias on the covariance matrix. However the MC1
approach remains more time-efficient and does not require a deep knowledge of the
considered survey and its systematics, a point which is often not properly controlled.
In any case, whether we want to generate CMB or density maps, some form of model
dependence is always present and has to be accounted for in the interpretation of any
result. In the rest of Sect. 2.2, I will use the MC2 method for the covariance matrix,
bearing in mind its advantages (no bias in the recovered parameters) and limitations
(additional computation time and good knowledge of the considered survey required).
2.2.3 Tomography for iSW Studies
By design, the classic approach to the cross-correlation intrinsically misses part of

the information about the iSW effect itself, as it looks for an integrated signal along
the whole line of sight with no regard for its redshift dependence. Such loss of
information could prove critical, especially in the context of time-varying models
of Dark Energy (e.g. quintessence). But even for a cosmological constant, the iSW
signal recovered by CMB/LSS cross-correlation will have a redshift dependence due
to sheer volume effects (already discussed at the end of Sect. 2.1.1). In the case
of the mock survey that I used in the previous section, the large redshift coverage
and potential iSW S/N (illustrated in Fig. 2.3) might also be squandered by the
classic approach described previously. Now the question that I will develop in the
present section is the following: Could it be possible to modify my protocol of
Fig. 2.5 Illustration of the 1.0

redshift slicing of the ideal
survey (blue solid line) con- 0.8
sidered in Fig. 2.3. The selec-
tion function is divided using 0.6
dN/dz
“top-hat” functions, into five
redshift intervals (red dashed 0.4
lines): z ∈ [0, 0.5], [0.5, 1],
[1, 1.5], [1.5, 2] and [2, ∝] 0.2
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Redshift z
cross-correlation analysis in order to exploit as much as possible the full potential of

such surveys?
The main idea here is to take advantage of the deepness in redshift of the survey:
one of the ways that we can think of in order to do so is to “slice” the considered survey.
More precisely, instead of integrating the whole distribution of matter contained
within it to get one density map, I divided the survey into redshift bins and constructed
a density map for each. We intuit here already that some care has to be taken when
choosing the shape and the number of bins. The simplest way to slice a survey
would be to simply divide its selection function into redshift intervals (“top-hat”
cuts), as shown in Fig. 2.5. However, one has to remember that the redshift of survey
objects often have errors in their determination. Hu and Scranton [22] supposed that
photometric redshift estimates are Gaussian distributed with an RMS fluctuation
σ (z) that increases with redshift as σ (z) = σmax (1 + z)/(1 + z max ), i.e. the bin sizes
are chosen to increase proportionally to the error. The two free parameters z max and
σmax roughly correspond to the end of the redshift range of the survey and the redshift
estimate errors around z max , respectively. Then, a top-hat cut in the [z 1 , z 2 ] interval
in estimated redshift becomes a smooth overlapping distribution in actual redshift:

1 dN z1 − z z2 − z
n z 1 →z 2 (z) = (z) erfc √ − erfc √ , (2.19)
2 dz σ (z) 2 σ (z) 2
where d N /dz is the original, full selection function of the survey. An illustration of
this redshift slicing is shown in Fig. 2.6, which used the same redshift intervals as
the slicing shown in Fig. 2.5. Given this formalism, one could a priori divide a given
survey in as many slices as wanted.
Let us now refocus on our objective: the exploitation of a survey for CMB/LSS
correlations. We have now a different starting point than in the previous section, as we
have now as many density maps as there are redshift slices. From there, the general
progression of the protocol remains the same: if we suppose that we divided the
considered survey in N redshift bins, we first extract the N angular cross-correlation
power spectra of each density map with the CMB. In parallel to this, we compute
Fig. 2.6 Redshift slicing of 1.0

the same survey of Fig. 2.3
(blue solid line), using the 0.8
expression in Eq. (2.19)
for the redshift bins. The 0.6
dN/dz
five divisions (red dashed
lines) have the same bounds 0.4
(the z 1 and z 2 in Eq. (2.19)
expression) as the slices in 0.2
Fig. 2.5
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Redshift z
the expected theoretical cross-correlation of each slice using the knowledge on the
survey selection function coupled to the expression in Eq. (2.19). And finally, we
perform a χ 2 test on all the spectra simultaneously to determine the that best
fits all the data. However, one has to be careful once again in the calculation of the
covariance matrix: indeed, as it can be seen clearly in Fig. 2.6, there exists a non-
negligible overlap between the different redshift slices. This overlap is practically
unavoidable as it originates from the (always present) errors in redshift estimations.
It creates therefore correlations between the resulting slices of density maps, which
in turns induces a form of redundancy when cross-correlating all these partial maps
with the CMB. One can then end up with unrealistic results with underestimated
errors bars on the recovered parameters. To account for this “leakage” of the redshift
bins into one another, I proceeded with some additional steps in the computation
of the covariance matrix: when applying the MC2 method, after generating the N
partial density maps (and the CMB map), not only did I correlate each map with
the CMB, but I also correlated each possible pair of density maps. The end result
of this is a non-trivial covariance matrix, that is no longer diagonal even in the case
of an ideal survey, as the non-diagonal terms contain the correlations between the
overlapping redshift slices of the matter distribution.
Now in order to assess the impact of the redshift slicing on the efficiency of
the reconstruction, I subjected my revised protocol to the same repeated test
that I performed in the previous section. A particular point that I did not mention
in the previous paragraph is that the simulation of accurate data maps for this test
also has to nclude all the correlations between redshift slices: I therefore had to
extensively modify the CROSS_CMBFAST code so that it would compute not only
the theoretical cross-correlation between the CMB and density slices, but also the
one between each pair of slices. After that was taken care of, I tested my protocol and
explored several slicing strategies, focusing especially on the choice of the number
of slices: I compared five numbers of slices ranging from 0 to 10, and present the
results of the reconstruction in Fig. 2.7.
As we can see there, even a simple division of the survey into two redshift slices
yields an appreciable improvement of the reconstruction, with a significant
Fig. 2.7 Normalised histograms of the reconstructed values of by the protocol described in
the text, in the context of several slicing choices of the original survey. The number of slices are 0
(meaning the whole redshift range, blue histogram), 2 (purple), 3 (red), 5 (orange) and 10 (yellow),
with the histograms becoming thinner as the number of slices increases. The true input value of
is again the WMAP7 best-fit value = 0.734
tightening of the distribution: the standard deviation decreases from ∼ 0.120

to ∼ 0.093 (drop of ∼25 %). The width of the distribution gets even smaller
when I further increase the number of slices, although the improvement becomes
less and less noticeable, reaching a plateau at some point (particularly visible also in
Fig. 2.8). This shows that there is no need to divide the original survey into too many
redshift bins, as the information that we can recover on the parameter seems to
reach a limit.
One could possibly ignore this fact (as it has no harmful effect on the reconstruc-
tion) and still carelessly consider as many slices as possible, but another important
parameter has to be taken into account. As the number of redshift bins increases, so
does the number of “density” maps that we have to consider, especially when comput-
ing the covariance matrix. In more practical terms, the more maps we have, the more
auto- and cross-correlations we have to perform: one CMB maps and N redshift bins
yield N (auto) and N (N + 1)/2 (cross) spectra. This has two different effects, one of
them being the increase in computation time for the matrix, roughly proportional to
N 2 (times the number of simulations). The less obvious effect is that each underlying
auto- and cross-spectrum gets weaker as we further divide the survey: this has the
impeding consequence of requiring a higher number of simulated maps in order for
the computed covariance matrix to converge. This particular effect is illustrated in
Fig. 2.8, where I show how the width of the distribution of reconstructed evolves
with the number of simulated maps used for the covariance matrix computation. We
witness there that the number of simulations required for reaching a plateau does
indeed increase for larger numbers of slices, from only 2,000 for the original survey
to 2–3 times as much for five redshift slices. Combined to the observed convergence
of the recovered distribution, we understand that some form of balance has to be
Fig. 2.8 Standard deviation

of the distribution of recon-
structed as a function
of the number of sets of
simulated maps used in the
computation of the covari-
ance matrix. The same slicing
choices as in Fig. 2.7 (with
their corresponding colours)
are explored. The curves get
lower as the number of slices
increases
found between the choice of the number of slices (i.e. the precision of the constraints
on ) and the number of required simulations (i.e. the computation time needed).
In the end, in this section I only scratched the surface of the topic, as there are
many improvements to my protocol that I could consider for a future work. First, in
the context of real maps and real surveys, a lot of importance has to be given to the
often partial coverage of the sky that has some drastic and intricate effects that affect
the cross-correlation, as well as the presence of foregrounds and contaminants that
further complicates the interpretation of the signal. On the other hand one could be
interested in assessing the full constraining power of the iSW effect on cosmology,
without using any underlying model, using the potential of the future generation of
surveys. Also in the context of an optimal exploitation of future datasets, the exact
choice of redshift slices (shapes) would need a particular tailoring to be as adapted as
possible to each surveys (this kind of work has been explored recently by [24]). As
a final note, although single surveys are not appropriate for it as of now, the slicing
study that I performed may already be applicable to current data in the context of
the combination of surveys for iSW detection: indeed, considering several surveys
together with their own selection functions is quite similar to considering several
slices in redshift (sometimes overlapping) of the same underlying distribution of
matter, although some additional care is required to account for the calibration and
systematics of each survey.
2.3 The Cosmic Infrared Background and the iSW Effect
In Sect. 2.2, I briefly explored some methods of exploiting the potential of the next
generation of surveys, optimising the amount of information extracted while taking
care of the possible complications that may appear along the way.
However, no matter how promising these future missions may look, the time scales
involved in their schedules does not allow much work other than in the realm of
2.3 The Cosmic Infrared Background and the iSW Effect 33
simulations and predictions. As an example, one of the most exciting future satellite,
Euclid, has its most optimistic launch date scheduled for 2020! The ground-based
LSST also has a similar schedule. Other experiments may have a closer “due date”
(e.g. in the course of 2015 for DES) but still measured in terms of years. On the other
end of the spectrum, recent works have been aimed at exploiting as much as possible
the current generation of surveys, including attempts at combining several datasets
and revising their work at each new data release. In this context, I chose to sit in the
middle ground: I searched for an alternative tracer of matter that could be correlated
to the CMB and that would be both already available in some form, but also with a
better potential that the currently used datasets. Such alternatives have already been
explored in the literature, for example through the correlation between the iSW effect
and the thermal Sunyaev-Zel’dovich effect [44], the latter being the result of a boost
received by CMB photons from high energy cluster electrons by inverse Compton
scattering: a non-zero correlation is indeed expected between the two effects as
these electrons sit in the gravitational potentials that produce the iSW effect. An
interesting aspect of this approach is that it constitutes a “CMB-only” detection of
the iSW effect (although with some underlying assumptions). In a similar fashion,
the [36] explored for the first time the correlation of the CMB with the reconstructed
gravitational lensing potential extracted from the CMB data itself.
As for myself, using some of the assets and expertise available at the IAS, I set my
eyes on an original tracer of matter, which is one of the other few backgrounds of the
Universe: the Cosmic Infrared Background, whose characteristics and exploitation
for iSW studies I will describe in the next sections.
2.3.1 The Blurry Red Light
First discovered by [39], the Cosmic Infrared Background (CIB hereafter) is visible
roughly from 10 to 1,000 µm in wavelength and is one of the backgrounds present in
the Universe (see Fig. 2.9). This particular background, present in every survey that
covers infrared wavelengths, arises from the accumulated electromagnetic emissions
from star-forming galaxies distributed across a large redshift range. It finds its origin
in the smallest and farthest galaxies (which cannot be resolved by telescopes and
their finite resolution) and in the densest population of galaxies (whose observation
is confusion-limited), which thus appear as a blurry background. The earliest epoch
for the production of the CIB is thought to be when star formation first began at
the end of the Dark Ages (and the onset of the reionisation epoch); contributions
continued through the present epoch, including our current Dark Energy dominated
era. Similarly to the CMB, the CIB also features anisotropies (first detected and
discussed in [26, 29]) that are shaped by the distribution and clustering of these
infrared galaxies in the Universe.
Ever since its discovery, many efforts have been deployed to detect the CIB
and its anisotropies with increasing precision, as they contain a lot of information
about the star and galaxy formation histories, including their clustering processes.
Fig. 2.9 Schematic spectral energy distributions (SED) of the most important (by intensity) back-
grounds in the Universe, and their approximate brightness in nW· m−2 · sr−1 written in their respec-
tive zones. From right to left the radio background, the CMB, the CIB, the cosmic optical background
(COB), the UV background, the X-ray background (XRB) and the gamma-ray background (GRB).
Image courtesy of H. Dole
The most recent papers on the CIB anisotropies use sophisticated models which com-
pute the halo occupation distribution (HOD, see e.g. [11, 31]) and the Dark Matter
halos properties, in order to predict the power spectrum of these anisotropies (see
e.g. [32]).
2.3.2 The CIB as an Alternative Tracer of Matter
We can easily understand that the anisotropies of the CIB are underlined by the galaxy
density field and thus the matter density fluctuations: it is therefore reasonable to
expect that the CIB has a positive correlation with the CMB through the iSW effect.
We should also bear in mind that it contains contributions from galaxies over a very
large range of redshift (up to z ∼ 7, with a peak of emission around z ∼ 2), and that
it is contained in many past and current surveys (such as IRAS or the newly released
Planck, both covering IR frequencies) that have often a large coverage of the sky.
These two points are the reasons that motivated my choice of the CIB as a tracer of
matter, as they make the CIB a particularly suitable and promising candidate for the
iSW detection by cross-correlation with the CMB. However, the extraction of the
CIB from a survey remains a challenging and delicate task that is still ongoing to this
day: this led me to focus first on a theoretical study of the CIB-CMB cross-correlation
through the iSW effect, and an assessment of its detectability.
As mentioned before, the most recent works focused on the studies of CIB
anisotropies used sophisticated HOD to predict and describe them. Such models
are particularly useful when describing the small, non-linear scales of the CIB. In
the context of the iSW effect, we focus on much larger scales which is why I used
a simpler model for the CIB, similar to the description made by [25]. The general
definition of the CIB anisotropies at a given frequency ν and in a given direction n̂
can be then written as the following line-of-sight integral:
0
νTCIB (n̂, ν) = dz a(z) ν j (n̂, ν, z) (2.20)
z far
with ν j being the emissivity anisotropies of the CIB. The integration is made from
some initial time z far before star formation began to our location at z = 0. In their
work, Knox et al. hypothesized that the CIB anisotropies are direct tracers of the
matter density fluctuations ν = νρm /ρ̄m , up to a bias factor. Therefore, the previous
expression becomes an integral of the product between a mean far infrared (FIR)
emissivity and the matter density fluctuation field:
0
νTCIB (n̂, ν) = dz a(z) b j (ν, z) j̄(ν, z) ν(n̂, z). (2.21)
z far
Here the quantity b j (ν, z) is some form of bias that links the matter distribution and
the emissivity. It is frequency- and redshift-dependent and is here defined by:
ν j (n̂, ν, z)
= b j (ν, z) ν(n̂, z) (2.22)
j̄(ν, z)
with j̄(ν, z) being the mean emissivity per comoving unit volume at frequency ν as
a function of redshift z. We can observe here some similarities between Eq. (2.21)
and the expression in Eq. (2.6) of the galaxy density contrast for the usual surveys.
However the selection function has been replaced by the emissivity function, with an
additional scale factor as we no longer consider individual objects that we count on
the sky, but rather the light that the objects emit (which of course gets redshifted in the
time it takes to reach us). There is also here an added dependence on the frequency
through the emissivity: although overlapping, it is different populations of objects
that we observe when surveying different wavelengths. This emissivity function is
derived using an empirical, parametric model based on number counts of galaxies:
at the time of this study I used the work of [6], although a recent update has been
published since (see [7]).
500 200
SPIRE - 250 μm
SPIRE - 350 μm
Emissivity x scale factor

400 Planck - 350 μm
Emissivity (Jy/Mpc)
150 Planck - 550 μm

Planck - 850 μm
Planck - 1380 μm
300
100
200
50
100
0 0
0 2 4 6 8 0 2 4 6 8
Redshift Redshift
Fig. 2.10 Left panel emissivities as functions of redshift for six experiment/frequency pairs, com-
puted using the model of [7]. Notice the very slight difference between instruments at 350 µm
due to a difference in bandpass. Right panel same functions multiplied by the scale factor a(z), or
equivalently 1/(1 + z), for comparison purposes with selection functions of galaxy surveys
Examples of such emissivities are shown in Fig. 2.10 (left panel): it should be noted
that they not only depend on the frequency but also on the experiment considered,
because each instrument possesses a unique bandpass for each frequency and has its
own resolution. This results in the observation of a “unique” CIB map by every pair
of frequency and instrument, corresponding to a population of objects just as unique.
The depth in redshift of the CIB is also illustrated in this figure, with emissivity
functions that theoretically reach much higher redshifts that any current (and even
some future) surveys. We should however keep in mind that, according to Eq. (2.21),
these functions have to be multiplied by the scale factor (right panel of Fig. 2.10)
in order to be compared to the selection function of classic galaxy surveys. This
causes a decrease at high z but even then, they still cover a larger range of redshift
than typical galaxy surveys (see e.g. Figs. 2.1 and 2.3 for comparison). We also note
that Fig. 2.10 shows significant overlaps in redshift between the emissivity functions
at different frequencies which will consequently induce correlations between CIB
observations, similarly to those between redshift slices of a same survey (discussed
in Sect. 2.2.3). At this point, this overlap already indicates that a combined use of
CIB observations at several frequencies may not yield improvements in the detection
of the iSW effect.
A last element that needs to be pointed out is the previously mentioned linear bias5
b j (ν, z) present in Eq. (2.21) that I chose constant in redshift here: b j (ν, z) = b lin (ν).
To obtain it for each frequency that I considered, I computed the value of b lin that
gives the best agreement between my calculation of the linear CIB power spectrum
and those obtained from the Planck data [34]. This point will be discussed in more
detail in the next subsection.
5 This bias here represents our matter-emissivity bias in Eqs. (2.21) and (2.22) and should not be
confused with the widely used galaxy-Dark Matter bias, though ours does contain information about
how the emitting objects populate Dark Matter halos.
Now, starting from Eq. (2.21) we can use the same formalism as for the CMB-
galaxy correlation in order to derive the theoretical CMB-CIB cross-correlation at
any given frequency. Following the same steps from Eqs. (2.7) to (2.11), we end
up with similar expressions for both the cross- and auto-correlation spectrum of the
CIB, except that the galaxy bias is replaced by the emissivity one, and the mean
emissivity (times the scale factor) plays the rôle of the selection function. The next
section will be dedicated to the computation of the expected value of these modified
equations in several contexts.
2.3.3 Predictions on the CIB-CMB Cross-Correlation
2.3.3.1 Computing the Expected Correlations
To compute the expressions mentioned at the end of the previous section, I once
again adapted the CROSS_CMBFAST code to use it for this CIB study, performing
a number of modifications to suit this new context. In my customised version, for a
given cosmology and emissivity function j̄(ν, z) my code calculates the CIB-CMB
angular cross-correlation power spectrum, as well as the predicted auto-correlation
power spectrum of the CIB fluctuations.
In Fig. 2.11, I present predictions for the CIB-CMB cross-correlation, at several
FIR wavelengths and for different instruments, namely: IRAS at 100 µm, Herschel
SPIRE at 250, 350 and 500 µm and Planck HFI at 350, 550, 850, 1380, and 2097 µm.
We note that at 350 µm the SPIRE- and Planck-predicted spectra differ slightly from
each other, once again due to slight differences in wavelength bandwidth of the two
instruments (hence a difference in the emissivities used, see Fig. 2.10).
In a fashion similar to previous galaxy-iSW cross-correlations (see e.g. Fig. 2.2),
we note that the cross-correlation peaks around δ 10–30, and quickly vanishes at
higher multipoles. Comparing the signal at the different wavelengths shows that the
amplitude of the cross-correlation signal is maximum at a wavelength 250 µm.
This is not entirely surprising, since this wavelength roughly corresponds to the
maximum of the observed CIB spectral energy distribution (cf. Fig. 2.9).
To get some form of validation for my computed spectra, I compared my predic-
tions for CIB autocorrelation to the measurements of [34], taking the opportunity to
address a concern mentioned in the previous section—namely, the determination of
the linear emissivity bias. To obtain it at each frequency, we compute the value of
this bias that gives the best agreement between my linear CIB power spectrum and
the corresponding one obtained from the Planck data. I chose to fit the two spectra
in the range of multipoles δ ∈ [10, 50], where most of the iSW signal resides. This
is illustrated in Fig. 2.12 where I plotted the biased and non-biased CIB linear spec-
tra and compared them to the ones from [34] at their four frequencies. Overall, the
two sets of spectra show good agreement over the multipoles of interest; the spectra
deviate at higher δs (starting from 100) due to the rise of non-linearities that I did
not account for in my linear model—namely the small-scale correlations between
Fig. 2.11 Theoretical angular cross power spectrum of the CIB-CMB correlation calculated for
IRAS at 100 µm (left-hand panel), for Herschel SPIRE between 250 and 500 µm (central panel)
and Planck HFI between 350 and 2,097 µm (right panel). The linear bias, blin , is fixed here to 1 at
all frequencies in order to compare the non-biased CIB power spectra. The vertical dashed line on
each panel marks the upper limit of the multipoles used in our analysis: this choice comes from the
vanishing of the iSW signal (see Fig. 2.13) and the rise of non-linearities at higher δ
galaxies inside the same halos. The linear bias we obtain this way increases with
the wavelength: this is coherent with the fact that as we go deeper into the infrared,
the galaxies probed are more luminous at higher z. They reside in more massive and
rarer halos, and are therefore more biased.
It should be also noted that the results on the cross-correlation are also not exact at
the highest δs, as hinted at by the deviation observed in the auto-correlation spectrum.
Indeed, the non-linear counterpart to the iSW effect, called the Rees-Sciama effect,
contributes at those scales (see [41] for a discussion). However, in our case the linear
part of the iSW largely dominates at the peak observed in Fig. 2.11.
2.3.3.2 Detectability Assessment
Following the same progression as for the CMB-galaxy case, I then investigated the
detection level of the iSW effect using CMB-CIB cross-correlation by performing a
signal-to-noise ratio analysis. Using the power spectra that I computed in the previous
section, I can express for each given frequency ν the total signal-to-noise ratio of the
iSW detection as follows:
δ
S 2 max
[Cδcr (ν)]2
(ν) = (2δ + 1) cr (2.23)
N
δ=2
[Cδ (ν)]2 + CδCIB (ν) × CδCMB
where Cδcr (ν) is the CMB-CIB cross-correlation spectrum, and CδCIB (ν) the CIB
auto-correlation spectra, both at a given frequency ν, while CδCMB is the CMB
auto-correlation spectrum. The total (cumulative) signal-to-noise is summed over
Fig. 2.12 Angular power spectra of the CIB fluctuations at four frequencies of the Planck-HFI
instrument, as fitted by the Planck team (blue uppermost continuous line) and by our non-biased
models (dashed yellow line). For each frequency, we provide in red the linear bias which gives the
best agreement between the two models, and plot our models taking into account this bias (solid
yellow line). The data points correspond to measurements obtained by the Planck team [34]
multipoles between δ = 2 and δmax 100 where the signal has its major contribu-
tion (see Fig. 2.11).
In my analysis I considered first the ideal situation where the CIB and CMB
maps used for cross-correlation are noiseless and cover the whole sky. With these
assumptions we obtain the highest possible signal-to-noise ratio, the only limitation
being the cosmic variance. In Fig. 2.13 I present the predictions for the CIB-CMB
cross-correlation in the case of a full-sky CIB map, provided by the previously
mentioned instruments and frequencies.
With these optimistic assumptions, I obtained high levels of detection for the
CIB-CMB correlation which reach 7σ , on par with (if not better than) the most
promising surveys of the generation to come (for detailed S/N results, see Table 2.1
in Sect. 2.3.4). Interestingly, it should be mentioned that these results in the ideal
case are independent of the previously discussed linear bias, even if it boosts the
correlation signal. This can be understood from Eq. (2.23) where the linear bias can
be factorized from each term (one for Cδcr and a squared one for CδCIB ) and therefore
cancels out.
As evoked before, we see that the largest contribution to the S/N comes from
multipoles smaller than 50. On the other hand, the most interesting feature of these
Fig. 2.13 Left-hand panel cumulated S/N as a function of δmax (defined in Eq. 2.23) for the CMB-
CIB cross-correlation, at our chosen frequencies and instruments. Right-hand panel total S/N with
δmax = 100 as a function of frequency/wavelength. Dotted line is for IRAS, the dashed ones are
for SPIRE, and the dot-dashed ones for Planck
Table 2.1 Total signal-to-noise ratio of the CIB-CMB cross-correlation for four of the CIB fre-
quencies of Planck-HFI
CIB frequency (GHz) 857 545 353 217
CIB wavelength (µm) 350 550 850 1380
Ideal case, single correlation S/N 6.26 6.83 6.98 6.95
Joint S/N 7.12
Realistic case n◦ 1, single correlation S/N
( f sky = 0.75, f CMB = 0.01, 5.36 5.73 5.39 3.56
Afore. = 0.01)
Joint S/N 5.88
Realistic case n◦ 2, single correlation S/N
( f sky = 0.15, f CMB = 0.01, 2.40 2.56 2.41 1.59
Afore. = 0.01)
Joint S/N 2.63
The results are given for each frequency and for the joint cross-correlation, first for the ideal case
discussed in Sect. 2.3.3.2 and then for two more realistic cases
results is that contrary to what could be intuited from Fig. 2.11, the total S/N peaks
around 850 µm instead of 250 µm for the cross-correlation signal itself. The reason
for this is actually quite subtle: it comes from the shape of the “noise” term in the
S/N expression in Eq. (2.23), as a function of δ, namely:
[Nδ ]2 (ν) → ([Cδcr (ν)]2 + CδCIB (ν)CδCMB )/(2δ + 1). (2.24)

Fig. 2.14 “Signal” terms (left-hand panel, rescaled to unity) and “noise” terms (middle panel, same
rescaling factor as the “signal”) of the S/N as functions of δ (see text for details) for our chosen
frequencies and instruments. The quotient of the two terms, used in the calculation of the S/N itself,
is shown in the right-hand panel the main difference throughout the frequencies comes from the
shape of the “noise” term
For all the frequencies studied here, this “noise” has roughly the same amplitude
relatively to its corresponding “signal”:
[Sδ ]2 (ν) → [Cδcr (ν)]2 . (2.25)
This is illustrated in Fig. 2.14, where I plotted in the left panel all the [Sδ (ν)]2 terms
with their respective maximum rescaled to unity. In the middle panel, I applied the
same rescaling factor of each [Sδ (ν)]2 term to the corresponding [Nδ (ν)]2 term.
By doing this, it is possible to compare the results from all frequencies without
changing their associated signal-to-noise ratios. On the resulting graph, we see that
at δ = 100 the rescaled noise amplitude is roughly the same, while the signal has
the same shape at all frequencies, except for a small shift in δ. However there is
a major difference in the shape of the noise power spectrum from one frequency
to another: its slope changes depending on the frequency, with the steepest one for
Planck 850 µm. Therefore its amplitude goes down more quickly than the others as
δ approaches zero where coincidently the signal is strong, which then boosts the S/N
at the low multipoles, and the total S/N.
In light of these results, the optimal frequency for iSW detection appears to be
around 353 GHz/850 µm with a maximum S/N reaching 7σ . However in practice,
the CIB extraction at this frequency might prove challenging since the CMB becomes
dominant here, and increasingly so as we go down in frequency. Therefore the pos-
sible residuals in our extracted CIB map have to be accounted for, and other sources
of noise as well, which is the purpose of the next subsection.
2.3.4 Accounting for Realistic Conditions
The contaminants of the CIB and obstacles to its extraction are many: first the signal is
completely dominated on a large part of the sky by emissions from our own galaxy.
The contamination from this foreground in the galactic plane is several orders of
magnitude above the CIB level and prevents us from extracting the CIB, therefore
reducing the “usable” fraction of the sky by at least ∼25 %. Furthermore, the rest
of the sky is also quite polluted—from a CIB point-of-view—by these foregrounds
full of galactic dust. These will have to be removed from our maps although some
residuals might remain in the final CIB map used for the cross-correlation. There
may even be a significant CMB residual (especially at Planck frequencies) in this
map due to an imperfect separation of components, which could have a dramatic
impact on the cross-correlation and easily induce false detections.
In the light of these elements, it appeared clearly to me that I needed to carry a
more realistic study by including these possible sources of contamination and assess
their impact on the iSW detection. To account for these effects on the detectability of
the CIB-CMB cross-correlation, I used a more complete formulation of the signal-
to-noise ratio, by adding new elements to the noise term. The expression of the S/N
therefore becomes at a given frequency ν:
δ
S 2 max
(ν) = f sky (2δ + 1)
N
δ=2
[Cδcr (ν)]2
× (2.26)
[Cδcr (ν)+ Nδcr (ν)]2 + [CδCIB (ν)+ NδCIB (ν)][CδCMB + NδCMB ]
where f sky is the fraction of the sky common to the CMB and the CIB maps, and Nδcr ,
NδCIB and NδCMB are the noise contributions respectively in the cross, CIB and CMB
signal. Since the CMB is expected to be only variance-limited at the multipoles of
interest, I fixed here NδCMB = 0. However we still have to take into account the CIB
contamination.
To do so, I first break the CIB noise power spectrum into several independent
parts:
NδCIB (ν) = RδCMB (ν) + Rδfore. (ν) + Nδinstr. (ν) + Nδcorrel. (ν) (2.27)
where these four different terms represent, from left to right, the power spectra of the
CMB residual, the galactic foreground residuals, the instrumental noise and finally
the noise due to a correlation between residuals and the CIB (which appears when
autocorrelating the final CIB map).
I quantify the CMB residual in the CIB map as a fraction f CMB of the total CMB
map, which affects both the cross-correlation and CIB noise; this approach assumes
a “global” CMB contamination (on the whole sky) without any spatial dependence.
This consequently defines the noise in the cross signal:
Nδcr (ν) = f CMB (ν) × CδCMB (2.28)
and the following two contributions:
RδCMB (ν) = f CMB

2
(ν) × CδCMB (2.29)
Nδcorrel. (ν) = 2 f CMB (ν) × Cδcr (ν) (2.30)
In the literature, the power spectrum of foregrounds such as dust emissions are often
modelled as (and often found to be close to) power laws: here in my analysis I define
the spectrum of the foreground residuals with the following expression:
α
δ
Rδfore. (ν) = Afore. (ν) × Cδ=10
CIB
(ν) , (2.31)
10
This power law has an amplitude defined relatively to the real CIB signal through a
chosen constant Afore. , which defines the quantity:
Afore. (ν) = Rδ=10

fore.
(ν)/Cδ=10
CIB
(ν) , (2.32)
i.e. the ratio between the foreground residuals and the CIB spectrum at the multipole
δ = 10, approximatively where the cross-signal is at its maximum. The slope of
the spectrum α is fixed here for all frequencies; previous analysis of infrared maps
[30, 45] found it to be −3 for foregrounds at high galactic latitudes. Finally, the
instrumental noise power spectra Nδinstr. at each frequency are taken from the first
ten months of Planck data in [35], and extrapolated to the thirty months, i.e. the end
of the fourth Planck full-sky survey.
In this section I focused on four of the five previously described Planck HFI
frequencies, from 217 to 857 GHz. I chose to discard the fifth 143 GHz as the CMB
completely dominates the CIB signal there. I also put aside the IRAS frequency here
because of its weaker significance (even in the ideal case), and the SPIRE frequencies
since the instrument was not scheduled to ever cover very large regions of the sky
(i.e. f sky 1), dramatically decreasing its associated S/N—as it is proportional to
the square root of f sky in Eq. (2.26).
At this point, the framework that I devised for the S/N calculation has three input
parameters at each of the four frequencies: f sky , f CMB and Afore. , whose values can
be chosen freely. The next step would have been to explore this 3D parameter space
at each frequency and compute the corresponding S/N at each point. Considering the
very large number of possible combinations of parameters, it would not have been
practical to perform and display the complete results of such exploration. Therefore
I made the decision of fixing f sky to two values of interest:
• f sky = 0.75, which corresponds to an optimistic case where the only part of the
sky discarded is the galactic plane; this is an optimistic scenario in the sense that
there are other highly contaminated regions where from the component separation
techniques might not be able to extract the CIB.
Fig. 2.15 Total signal-to-noise ratio of the CIB-CMB cross-correlation at 353 GHz, as a function of
the CMB residuals (in percentage of the total CMB signal) and the foregrounds residuals (through
the parameter Afore. ). Left panel f sky = 0.75, the results go from less than 4 to more than 5, from
the brightest colored area to the darkest. Right panel f sky = 0.15, the S/N goes from slightly less
than 1.5 to more than 2, again from the brightest to the darkest area
• f sky = 0.15, which is a low estimate of the area of the sky where the current data
allow for an efficient CIB extraction. The methods currently employed are based
on the use of HI maps as a tracer of the galactic dust, though it only remains valid
for an HI column density lower than a specific threshold (see [34] for details on
these methods).
As far as the two other input parameters are concerned, I limited their ranges to
reasonable values, with f CMB ∈ [0, 0.1] and Afore. ∈ [0, 10].
Among the four Planck frequencies that I kept here, I started by studying the
effects of the noise for the frequency that gave the best S/N results in the ideal case
(850 µm/353 GHz)—having in mind that a significant drop in S/N at this wavelength
would not bode well for the rest of them. The corresponding results are presented in
Fig. 2.15 which shows the contour levels of the S/N in the ( f CMB , Afore. ) parameter
space. The influence of the CMB is clearly visible at this frequency, quickly reduc-
ing the S/N as its residual level increases. This effect is even more pronounced at
1,380 µm/217 GHz, where the S/N is typically twice as low as in the ideal case
(see Table 2.1), due to the fact that we get closer to the maximum of the SED of
the CMB. It makes this frequency far less significant for the iSW detection than in
the ideal case. The presence of instrumental noise—whose effect cannot be appre-
ciated with Fig. 2.15 alone—becomes significant at the two lowest frequencies (217
and 353 GHz), again reducing their value in the cross-correlation. As expected the
galactic foreground residuals also decrease the S/N, though their influence is roughly
the same at all frequencies as they are defined relatively to the CIB spectrum in this
analysis. Lastly, the biggest influence comesfrom the fraction of the sky through
the f sky parameter, as the total S/N scales as f sky . This makes it a crucial require-
ment for future applications to have the largest possible coverage to minimize this
effect—very similarly to the requirement of galaxy surveys.
Fig. 2.16 Total signal-to-noise ratio of the CIB-CMB cross-correlation at 545 GHz as a function
of the CMB residuals and the foregrounds residuals. Left panel f sky = 0.75, the results go from
slightly less than 4 to more than 5, from the brightest to the darkest area. Right panel f sky = 0.15,
the S/N goes from less than 2 to slightly more than 2.5
Taking all these remarks into account and after some exploration of the parame-
ter space, the optimal frequency that stands out in these more realistic scenarii is
545 GHz/550 µm. Indeed, it is weakly influenced by instrumental noise and CMB
residuals and also has a higher “original” S/N (in the ideal case) than the other
remaining frequency 857 GHz/350 µm. The S/N analysis at 545 GHz is presented in
Fig. 2.16: for the case of a large but realistic coverage, the S/N still reaches high and
promising values around 4.5 σ . Even in a more pessimistic scenario, the significance
of the detection stays around a 2.5 σ level, comparable with the constraints from
current galaxy surveys.
However, one may wonder here if it is really necessary to argue and determine
which is the most suited frequency for giving the best results for the CMB-CIB cross-
correlation. Indeed, in the same manner as I combined the constraints from redshift
slices in Sect. 2.2.3, it is reasonable to consider the possibility of combining the
cross-correlation signals from the CIB at several frequencies to improve the S/N—a
approach that I will explore in the last section of this chapter.
2.3.5 Multi-frequency Joint Analysis
Until now I have only considered a detection at a single CIB frequency and its
associated significance. In practice, when handling real data, we will most likely
have several maps of the CIB at different frequencies, hence as many cross spectra.
For example, in the case of Planck we should eventually be able to extract the CIB
at four different frequencies on a large fraction of the sky. This could potentially
allow to increase the total signal-to-noise ratio of the iSW detection by combining
the constraints from all available frequencies. However, the improvement brought
by this approach will be limited by the possible intrinsic correlations (and redundant
information) between the CIB maps at different frequencies—once again in the same
manner as the correlations between redshift slices of a galaxy survey limited the
amount of information available.
I expanded my previous S/N formalism to express the theoretical joint significance
of a set of n cross-correlations (i.e. CIB at n frequencies, each correlated to the same
CMB):
2
S
= X T M−1 X (2.33)
N Total
with X (X T ) being the column (row) vector of all the cross-correlations:

X T = X T (ν1 ) . . . X T (νn )
where X T (νi ) contains the cross-spectrum at the frequency νi , from δ = 2 to 100:
X T (νi ) = Cδ=2
cr
(νi ) . . . Cδ=100
cr
(νi )
The block matrix M is the covariance matrix, containing n × n blocks. Each one
of them represents the covariance of two cross-spectra at different CIB frequencies,
depending on the position of the block. At the ith line and jth column, the block
Mi j is written as: ⎛ ij ⎞
Mδ=2 0
⎜ .. ⎟
Mi j = ⎝ . ⎠
ij
0 Mδ=100
The diagonality of Mi j comes from the assumption that the different multipoles
are uncorrelated. This could prove no longer true for small fractions of the sky but
gives an upper bound on the S/N. In the noiseless case discussed in Sect. 2.3.3, the
elements of each block can be expressed as follows:
ij
Mδ = Covar(Ccr
δ (νi ), Cδ (νj ))
cr
Cδcr (νi )Cδcr (ν j ) + CδCMB CδcrCIB (νi , ν j )

=
2δ + 1
We can see here the dependence on the aforementioned possible correlation between
the CIB at frequency νi and the CIB at frequency ν j , through the cross-spectrum
CδcrCIB (νi , ν j ). To perform a more advanced analysis, it is easy to modify this expres-
sion to account for the possible sources of noise discussed in the previous section.
Once again, the large number of possible combinations of noise parameters makes
it unpractical to present a complete study of the joint correlation. Instead I focused on
a few particular cases, motivated by my previous findings. A summary of my results
on single and joint correlations is presented in Table 2.1. Going back first to the ideal
case, I quantified the impact of the joint detection. I found a relatively small gain, as
it increases the total S/N by a mere 0.15 compared to the maximum significance
of a single detection. This can be attributed to the high correlations between the
CIB at its different observed frequencies, which limits the usefulness of the joint
cross-correlation.
Considering now more realistic situations, with the presence of instrumental noise,
I choose to fix some of the free parameters, with f sky = 0.75 and f sky = 0.15. A
reasonable confidence in component separation techniques allows us to hope for
small enough residuals, so that we choose f CMB = 0.01 and Afore. = 0.01. In these
cases, the joint correlation has a limited interest (respectively a 0.15 and 0.07
gain for f sky = 0.75 and 0.15) due to the correlations in both the CIB signals but also
in the astrophysical noise contributions—CMB and dust—between frequencies.
In the end, in the context of my model, the joint analysis of several CIB frequencies
does not yield a significant improvement over the use of the best single frequency.
This answers our interrogation from the end of the previous section: looking for the
optimal frequency for the iSW detection is enough and justified here in order to focus
the effort in the right direction (especially for the tricky part of the process: the CIB
extraction). Nonetheless, the use of additional frequencies—if available—could have
a non-negligible impact in the presence of a source of uncorrelated noise (between
frequencies), as it would allow to get rid of most of it.
As a conclusion of this first ever investigation of the CIB-CMB cross-correlation,
I found very promising results on the iSW effect and its detectability under various
observational situations. Expected realistic significances range from ∼2.5 to 5.5
depending on the frequency, the levels of noise and the fraction of the sky available
for analysis: these show a great potential compared to even the most promising galaxy
surveys (cf. [2, 13]). The results of this work will be valuable in the forthcoming
years of analysis and exploitation of the Planck data. The formalism I developed
provides an accurate and flexible forecast of the expected results of the CIB-CMB
cross-correlation and allows to constrain the requirements for a significant iSW
detection.
References
1. J.K. Adelman-McCarthy et al., The sixth data release of the sloan digital sky survey. Astrophys.
J. Suppl. Ser. 175, 297–313 (2008)
2. N. Afshordi, Integrated Sachs-Wolfe effect in cross-correlation: the observer’s manual. Phys.
Rev. D 70(8), 083536 (2004)
3. M.A. Agüeros et al., Candidate isolated neutron stars and other optically blank X-ray fields
identified from the ROSAT all-sky and sloan digital sky surveys. Astron. J. 131, 1740–1749
(2006)
4. H. Aihara et al., The eighth data release of the sloan digital sky survey: first data from SDSS-III.
Astrophys. J. Suppl. Ser. 193, 29 (2011)
5. C.L. Bennett, G.F. Smoot, A. Kogut, COBE DMR maps of the microwave sky. In Bull. Am.
Astron. Soc. 22, 336 (1990)
6. M. Béthermin, H. Dole, G. Lagache, D. Le Borgne, A. Penin, Modeling the evolution of infrared
galaxies: a parametric backward evolution model. Astron. Astrophys. 529, A4 (2011)
7. M. Béthermin et al., A unified empirical model for infrared galaxy counts based on the observed
physical evolution of distant galaxies. Astrophys. J. 757, L23 (2012)
8. M. Blanton, R. Cen, J.P. Ostriker, M.A. Strauss, The physical origin of scale-dependent bias
in cosmological simulations. Astrophys. J. 522, 590–603 (1999)
9. E. Boldt, The cosmic X-ray background. Phys. Rep. 146, 215–257 (1987)
10. J.J. Condon et al., The NRAO VLA sky survey. Astron. J. 115, 1693–1716 (1998)
11. A. Cooray, R. Sheth, Halo models of large scale structure. Phys. Rep. 372, 1–129 (2002)
12. P.-S. Corasaniti, T. Giannantonio, A. Melchiorri, Constraining dark energy with cross-
correlated CMB and large scale structure data. Phys. Rev. D 71(12), 123521 (2005)
13. M. Douspis, P.G. Castro, C. Caprini, N. Aghanim, Optimising large galaxy surveys for ISW
detection. Astron. Astrophys. 485, 395–401 (2008)
14. F.-X. Dupé, A. Rassat, J.-L. Starck, M.J. Fadili, Measuring the integrated Sachs-Wolfe effect.
Astron. Astrophys. 534, A51 (2011)
15. D.J. Fixsen, The temperature of the cosmic microwave background. Astrophys. J. 707, 916–920
(2009)
16. J. Garriga, L. Pogosian, T. Vachaspati, Forecasting cosmic doomsday from CMB-LSS cross-
correlations. Phys. Rev. D 69(6), 063511 (2004)
17. T. Giannantonio et al., Combined analysis of the integrated Sachs-Wolfe effect and cosmolog-
ical implications. Phys. Rev. D 77(12), 123520 (2008)
18. T. Giannantonio, R. Crittenden, R. Nichol, A.J. Ross, The significance of the integrated Sachs-
Wolfe effect revisited. Mon. Not. R. Astron. Soc. 426, 2581–2599 (2012)
19. K.M. Górski et al., HEALPix: a framework for high-resolution discretization and fast analysis
of data distributed on the sphere. Astrophys. J. 622, 759–771 (2005)
20. C. Hernández-Monteagudo et al., The SDSS-III baryonic oscillation spectroscopic survey:
constraints on the integrated Sachs-Wolfe effect. Mon. Not. R. Astron. Soc. 438, 1724–1740
(2014)
21. S. Ho, C. Hirata, N. Padmanabhan, U. Seljak, N. Bahcall, Correlation of CMB with large-scale
structure. I. Integrated Sachs-Wolfe tomography and cosmological implications. Phys. Rev. D
78(4), 043519 (2008)
22. W. Hu, R. Scranton, Measuring dark energy clustering with CMB-galaxy correlations. Phys.
Rev. D 70(12), 123002 (2004)
23. T.H. Jarrett et al., 2MASS extended source catalog: overview and algorithms. Astron. J. 119,
2498–2531 (2000)
24. G. Jürgens, B.M. Schäfer, Integrated Sachs-Wolfe tomography with orthogonal polynomials.
Mon. Not. R. Astron. Soc. 425, 2589–2598 (2012)
25. L. Knox, A. Cooray, D. Eisenstein, Z. Haiman, Probing early structure formation with far-
infrared background correlations. Astrophys. J. 550, 7–20 (2001)
26. G. Lagache, J.L. Puget, Detection of the extra-galactic background fluctuations at 170 mu m.
Aston. Astrophys. 355, 17–22 (2000)
27. D. Larson et al., Seven-year Wilkinson microwave anisotropy probe (WMAP) observations:
power spectra and WMAP-derived parameters. Astrophys. J. Suppl. Ser. 192, 16 (2011)
28. D.N. Limber, The analysis of counts of the extragalactic nebulae in terms of a fluctuating
density field II. Astrophys. J. 119, 655 (1954)
29. H. Matsuhara et al., ISO deep far-infrared survey in the “Lockman Hole”. II. Power spectrum
analysis: evidence of a strong evolution in number counts. Astron. Astrophys. 361, 407–414
(2000)
30. M.-A. Miville-Deschênes, G. Lagache, J.-L. Puget, Power spectrum of the cosmic infrared
background at 60 and 100 µm with IRAS. Astron. Astrophys. 393, 749–756 (2002)
31. J.A. Peacock, R.E. Smith, Halo occupation numbers and galaxy bias. Mon. Not. R. Astron.
Soc. 318, 1144–1156 (2000)
32. A. Pénin, O. Doré, G. Lagache, M. Béthermin, Modeling the evolution of infrared galaxies:
clustering of galaxies in the cosmic infrared background. Astron. Astrophys. 537, A137 (2012)
33. W.J. Percival et al., The shape of the sloan digital sky survey data release 5 galaxy power
spectrum. Astrophys. J. 657, 645–663 (2007)
References 49
34. Planck Collaboration, Planck early results. XVIII. The power spectrum of cosmic infrared
background anisotropies. Astron. Astrophys. 536, A18 (2011a)
35. Planck Collaboration, Planck early results. VI. The high frequency instrument data processing.
Astron. Astrophys. 536, A6 (2011b)
36. Planck Collaboration, Planck 2013 results. XIX. The integrated Sachs-Wolfe effect (2013a).
ArXiv:1303.5079
37. Planck Collaboration, Planck 2013 results. I. Overview of products and scientific results
(2013b). ArXiv:1303.5062
38. Planck Collaboration, Planck 2013 results. XVI. Cosmological, parameters (2013c).
ArXiv:1303.5076
39. J.-L. Puget et al., Tentative detection of a cosmic far-infrared background with COBE. Astron.
Astrophys. 308, L5 (1996)
40. U. Sawangwit et al., Cross-correlating WMAP5 with 1.5 million LRGs: a new test for the ISW
effect. Mon. Not. R. Astron. Soc. 402, 2228–2244 (2010)
41. B.M. Schäfer, A.F. Kalovidouris, L. Heisenberg, Parameter estimation biases due to contri-
butions from the Rees-Sciama effect to the integrated Sachs-Wolfe spectrum. Mon. Not. R.
Astron. Soc. 416, 1302–1310 (2011)
42. U. Seljak, M. Zaldarriaga, A line-of-sight integration approach to cosmic microwave back-
ground anisotropies. Astrophys. J. 469, 437 (1996)
43. D.N. Spergel et al., First-year Wilkinson microwave anisotropy probe (WMAP) observations:
determination of cosmological parameters. Astrophys. J. Suppl. Ser. 148, 175–194 (2003)
44. N. Taburet, C. Hernández-Monteagudo, N. Aghanim, M. Douspis, R.A. Sunyaev, The ISW-tSZ
cross-correlation: integrated Sachs-Wolfe extraction out of pure cosmic microwave background
data. Mon. Not. R. Astron. Soc. 418, 2207–2218 (2011)
45. E.L. Wright, Angular power spectra of the COBE DIRBE maps. Astrophys. J. 496, 1 (1998)
Chapter 3
The Impact of Identified Superstructures
in the CMB
3.1 A New Approach to the iSW Effect
3.1.1 Narrowing the View
As of today, the cross-correlation of galaxy surveys with the CMB, extensively

discussed in the previous chapter, is still considered as the standard way to probe the
iSW effect. However, this method has yet to provide a definitive and unambiguous
detection despite its robustness. As I mentioned in Sect. 2.1.4, current galaxy surveys
are plagued by several shortcomings (small sky coverage and/or depth, not fully
controlled systematics and biases) making them less than ideal for iSW studies and
leading to inconsistent results across the literature.
In this clouded context, it is clear that cross-correlation studies could greatly
benefit from another way of evidencing the iSW effect, an alternative method that
could independently infirm or confirm its existence. We remember how in Chap. 2
we started from the cross-correlation of a single galaxy survey, projected on the 2D
sky, with the map of the CMB. Then later we considered to further exploit a survey
by taking advantage of its depth, decomposing it into slices of redshift. Following the
same progression, we can imagine pushing the idea even further and consider using
the full 3D information coming from a galaxy survey to “guide” us in our search for
the iSW effect.
One possible application of this idea would be to locate the individual objects in
the survey that are theoretically expected to leave the biggest imprint on the CMB,
i.e. focus on the largest superstructures in the Universe (superclusters or supervoids)
that are not virialised instead of considering the whole distribution of matter (and
all the complications that could go with its analysis). In the linear regime, the shift
in temperature experienced by photons (due to the iSW effect) is proportional to the
size of the gravitational potential that they cross, so that the biggest structures are
indeed expected to yield the biggest iSW effect.

52 3 The Impact of Identified Superstructures in the CMB
However, it would be foolish to try to detect the imprint of a single superstructure

on the CMB temperature map, as the small expected amplitude of the iSW effect
(a few μK, cf. below in Sect. 3.5) is completely drowned in the noise, which in this
case is none other than the primordial fluctuations of the CMB itself (of the order of
∗110 μK). However, this should not be a reason for abandoning the idea altogether:
we can still exploit this approach by considering many of these superstructures and
averaging the CMB over all their locations in the sky, in order to cancel the contri-
bution from the random CMB fluctuations while enhancing the final iSW signature.
It is already clear that the properties and the number of structures used will play a
crucial rôle here, as well as the other sources of contamination that can pollute and
bias the detection. The (historically) first application of this approach to real data
will be described in the following section, along with its results and conclusions.
3.1.2 The Stacking of Superstructures in the CMB
Firstly, the aforementioned approach requires to have a list of identified superstruc-

tures to work with, but no experiment has ever been dedicated to finding these specif-
ically. In their pioneering work, Granett et al. ([5] Gr08 thereafter) considered the
Data Release 6 (DR6) of the Sloan Digital Sky Survey (SDSS), more precisely the
catalogue of Luminous Red Galaxies (LRGs)—elliptical galaxies in massive galaxy
clusters thought to be good (but sparse) tracers of matter on large scales [3, 23].
The redshifts (estimated by photometry) of these galaxies span here z = 0.4–0.7 ap-
proximatively. On this sample of LRGs, Gr08 ran two numerical codes based on the
Voronoi tessellation, VOBOZ (VOronoi BOund Zones [13]) and ZOBOV (ZOnes
Bordering On Voidness [12]), respectively designed for void and cluster finding.
With their first run, they found a large number of candidate structures (631 voids
and 2,836 clusters), but with many spurious detections arising from discreteness
noise due to the high sensitivity of the algorithms. Therefore they chose to keep
only the most “significant” structures using a criterion based on the comparison of
the density contrasts of these identified objects with those of voids and clusters in a
uniform Poisson point sample. In the end, their final catalogue contains the 50 most
significant (according to their criterion) superclusters and 50 supervoids; it provides
several informations for each structure, including (but not limited to) the position
of its centre on the celestial sphere, its physical volume and its density contrast.
These clusters and voids have mean effective radii respectively around 38 and 109
Mpc, and mean redshifts of 0.50 and 0.53. This catalogue of 100 structures has been
made publicly available by the authors; It should be noted however that they never
released the full results of their algorithms, including the supposedly less significant
detections of structures.
Once the superstructures are identified, the immediate next step is to choose the
CMB map to work with: Gr08 used the ILC map from the WMAP five-year data, as
well as the raw maps from the Q, V and W frequency bands. The use of frequency
maps is particularly important to test the one of the expected features of the iSW
3.1 A New Approach to the iSW Effect 53
signal: its achromatic nature, as it does not depend on the energy of the photons
experiencing the effect.
From here, the protocol is relatively simple and can be summarised as follows:
• Retrieve the locations of the superstructures on the sky from the considered cata-
logue;
• In the CMB map, cut square patches of CMB at these locations (the patches have
to be large enough to include the entirety of each structure);
• Stack the patches together, i.e. compute the average of all these images.
This approach assumes that in each stacked patch, the corresponding superstructure
produces an iSW signal at the centre of the image, but well hidden behind a veil
of CMB fluctuations. However, since these primordial CMB anisotropies are uncor-
related to the position of structures, we expect their contamination to average out
when stacking a large enough number of patches, so that only the (mean) iSW signal
remains at the centre of the final image.
The output of this method as it was originally applied by Gr08 is reported in
Fig. 3.1, extracted from the original Gr08 article. It contains the results of the stacking
of three sets of structures: the 50 clusters of the Gr08 catalogue, the 50 voids, and
finally a combination of these 100 structures where the opposite of the patches is
used for the voids (to get a positive signal in the end). In each of the resulting images,
a signal appears quite clearly at the centre: a cold spot for the voids, and a hot spot
for the clusters (and the combination) as expected from the iSW effect that these
structures should produce. To obtain the results of Fig. 3.1, they also rotated the
CMB patches to align each structure’s major axis with the vertical direction: indeed,
such structures are often quite irregular in shape, and closer to ellipsoids rather than
to spheres. The purpose of this rotation is however purely cosmetic, in order to
visually enhance the signal in the images: the aperture photometry itself is invariant
by rotation of the patches.
3.1.3 The Gr08 Results: Protocol, Results and Caveats
Although these images appear quite convincing, a simple look will not tell us anything
about the amplitude of these apparent signals, or their significance. In order to obtain
such information, the standard procedure chosen by Gr08 was to apply a top-hat
filter on the stacked image, a procedure also called “aperture photometry”. For a
given aperture angle θ , this consists in averaging all the pixels in the image that are
contained within a disk of radius θ , and subtracting from it the mean of√the pixels
inside the surrounding ring of equal area, i.e. between a radius of θ and θ 2 (see the
black circles of Fig. 3.1 for illustration). This procedure guaranties that the obtained
value is not affected by a local fluctuation of temperature which could artificially
boost the measured signal. In their study, Gr08 reported photometry amplitudes of
−11.3, 7.9 and 9.6 µK for the voids, clusters, and combined signals respectively,
using a top-hat filter with a radius of 4→ . Moreover, they also reported that these
Fig. 3.1 Figure 1 of Granett et al. [5]. Stacked patches of the CMB (WMAP 5-year ILC) averaged at
the location of 50 supervoids (left) and 50 superclusters (centre), themselves identified in the SDSS
LRG catalogue. The stacked image for the combined sample is shown on the right. The patches
of CMB were rotated so as to align each structure’s major axis with the vertical direction. These
images are smoothed with a Gaussian kernel with FWHM 1.4→ . The inner circle (4→ radius) and
equal-area outer ring mark the extent of the compensated filter used in Gr08 analysis to highlight
the visible hot (for the clusters and the combination) and cold (for the voids) spots at the centre of
these images
amplitudes remain almost constant over the frequencies they considered (the Q, V,
and W band of WMAP), in agreement with expectations from an iSW signal.
Having such figures is a step in the right direction, but it is not sufficient to
determine if the observed signal is particularly exceptional or quite common: we
need therefore some reference to compare this numbers to. A way to proceed used in
Gr08 is to perform the same stacking protocol on random locations in the same CMB
map, drawing each time as many positions as the number of structures used in the
original stacked image. This procedure is repeated many (several thousands) times
while the aperture photometry of each “random stack” is computed. We end up with
a distribution of values that we expect to be close to Gaussian, since the dominant
signal comes from the primordial CMB anisotropies which constitutes a Gaussian
field itself.1 The photometry values previously obtained for the “true” stacking are
then compared to this distribution, and their significance can therefore be determined
(cf. an example in Fig. 3.10).
It should be noted that Gr08 also used an alternative approach to derive the dis-
tribution of random stacks: they kept the same fixed locations on sky—those of the
“real”, identified structures—and then repeated the stacking procedure many times
but with a randomly generated CMB (using a theoretical CMB power spectrum de-
rived from the fiducial cosmological model at the time). The new distribution did
not show any deviation from the one obtained previously; if it had, this would have
indicated that the measured signal in the real stacked images could be due (at least
partly) to a fortuitous correlation of the primordial CMB fluctuations for this partic-
ular arrangement of locations in the sky. Using the CMB map from the 5-year data
1 This is true as far as we can discern for now, since the current constraints on the presence of
primordial non-Gaussianities in the CMB are compatible with zero (see [16]).
3.1 A New Approach to the iSW Effect 55
release of WMAP, Gr08 found that for 50 randomly located patches of CMB, the
4→ -aperture photometry has a zero mean2 and a standard deviation of 3.1 μK, and
2.2 μK for 100 random positions. This led to a claim by Gr08 of a 2.6 σ signal in the
stacked image for their 50 voids, 3.7 σ for their 50 clusters, resulting in a combined
4.4 σ signal, which they attributed to the iSW effect of these structures.
Taken at face value, this result would effectively be one of the strongest detection
of the iSW effect to date. Gr08 also claimed that it showed some tension with respect
to the νCDM paradigm: they tried to estimate the expected iSW effect from this
kind of superstructures by measuring the signal that a large cosmological N-body
simulation (here the Millennium run [19]) produces, using ray-tracing techniques.
However, the volume of the simulation is only large enough for 1 or 2 supervoids
and superclusters only (with the same typical size as those of the Gr08 catalogue)
and therefore might not be representative of the variety of the 100 structures of the
Gr08 catalogue. The maximum iSW signal then gives 4.2 μK, about 2 σ lower than
measured in the stacked data. Although this estimate is quite crude, Gr08 noted
nonetheless at that time that this higher-than-expected results seemed to follow a
trend found in several previous iSW measurements (from cross-correlation studies)
that were also somewhat higher than predicted by νCDM (see e.g. [8]).
I take the opportunity from this last point to mention a few caveats in the Gr08
analysis and in their stacking procedure. First of all, the “iSW landscape” has quite
changed since the publication of their paper in 2008. Recent works such as [7]
and the dedicated iSW paper from the Planck cosmology results [17] have yielded
results with slightly lower significances but in better agreement with νCDM (all
within 1 σ ), including the results from new probes of the iSW effects such as the
iSW-lensing bispectrum. In this updated context, the Gr08 result is now the one that
stands out as inconsistent, casting some doubts on the iSW origin of the reported
signal. Moreover, confronting the stacking results to theory might require more than
a single set of N-body simulation; a thorough comparison between these and νCDM
predictions requires a more advanced work on the theoretical side. This particular
point will be discussed in detail in Sect. 3.5.
A few other issues can also be discussed regarding the Gr08 study: the 100 struc-
tures that they used were extracted from a much larger catalogue of candidate struc-
tures that they first identified. However this “proto-catalogue” has never been released
by the authors. I trust their claim that their selection of 100 structures was only based
on the properties of these objects and not linked to their measured impact on the
CMB. However, some elements in their study seem to point towards the possible
existence of selection biases, especially in the light of a particular test that they per-
formed: Gr08 tried to change the number of objects stacked and the aperture of the
top-hat filter, and observed that any deviation from their “golden numbers” (50 voids
and clusters, 4→ aperture) resulted in a decrease of the significance of the results (see
Table 3.1). This could be attributed to real physical effects, since it is reasonable to
believe that the measured signal can be influenced by the addition (dilution of the
2 Which is quite expected from the nature of the CMB fluctuations.

Table 3.1 Adapted from

N Aperture radius Photometry (μK) Significance (σ )
Table 1 of Granett et al. [5].
For the stacked signal that 30 4.0→ 11.1 4.0
they measured by photometry 50 4.0→ 9.6 4.4
and its significance, it shows 70 4.0→ 5.4 2.8
the dependence on the 50 3.0→ 8.4 3.4
number of structures N (i.e. N 50 3.5→ 9.3 4.0
voids and N clusters stacked) 50 4.0→ 9.6 4.4
and the aperture radius
50 4.5→ 9.2 4.4
50 5.0→ 7.8 3.8
mean iSW signal by lowly contributing objects) or the subtraction (increased noise
due to the limited number of objects) of structures in the stacking.
However, all of the aforementioned remarks and difficulties called for a more
in-depth analysis of the stacking procedure—a task that I chose to tackle. As I will
show in the next sections of this chapter, I even went beyond the Gr08 study and did
not confine myself to its sole catalogue and its intriguing results.
3.2 My Revised and Improved Stacking Protocol

At the time of this writing (mid-2013), the beautiful products of the Planck mission
have now been released: the first and foremost way to improve stacking studies will
be to use this new CMB data to improve and validate our stacking studies. I actually
conducted two separate works on the topic: one that used WMAP 7-year data (the
latest at that time) which led to the publication of a paper [10], and another inside
the Planck collaboration itself where I was the leader of the stacking section in the
iSW-dedicated cosmological paper [17]. Although the CMB data is different, the
protocol that I devised is largely the same for both new studies and the catalogues of
structures used are identical. I will therefore describe the general framework of my
method in the following, underlying any substantial difference when needed.
3.2.1 Data and Core of the Protocol

Planck data: For my Planck related work, I have made use of the foreground-cleaned
CMB map provided by the data processing centres (as described in the Planck com-
ponent separation paper [15]). In particular, I used the so-called “official” CMB
map which was cleaned using the SMICA component separation technique based on
spectral matching. In addition to this, I also performed my analyses on six foreground-
cleaned frequency maps (from 44–353 GHz) constructed by subtracting a linear com-
bination of internal templates using SEVEM, another component separation method
considered and used in the Planck collaboration ([15] contains a description of both
methods). We remind that the use of frequency maps is crucial in iSW studies as the
signal is expected to be achromatic. Also, to minimize the presence of foreground
3.2 My Revised and Improved Stacking Protocol 57
contamination in the maps, I used one of the official Planck mask, also described
in [15], which excludes regions with larger Galactic and point-source contamination
and keeps 72 % of the sky.
WMAP data: In my other stacking study, I used the maps of the CMB released
by the WMAP team after seven years of observation [11] in contrast to the five-year
data used by Gr08 in their study. Just as for the Planck data, I also took the individual
channel maps at the three frequencies that are the least contaminated by foregrounds
(the Q, V and W bands at 41, 61 and 94 GHz respectively). At the same time and
in order to assess the possible impact of foregrounds, I also did my analyses with
the foreground reduced maps released by the WMAP team in the same frequency
channels. A galactic mask was also used here: the KQ75 from the WMAP team, an
extended mask for temperature analysis removing about 22 % of the sky along the
galactic plane and around point sources.
My stacking procedure: The first protocol that I devised consisted in the follow-
ing steps. The CMB map considered is cleared of its the cosmological monopole and
dipole, accounting for the considered sky mask. Using as inputs the galactic longi-
tudes and latitudes of the structures studied, a code I wrote based on the HEALPix3
package cuts a patch in the CMB map centred around each structure. I chose the
patches to have a 6 arcmin/pixel resolution (small enough to oversample and keep
the details of the CMB maps used) and to be squares of 301×301 pixels, i.e. 30→ ×30→
patches (large enough to encompass any of the considered objects). Concurrently, the
code cuts identical square patches at the same locations in the associated mask map;
it computes the final stacked image as the average of all the CMB patches weighted
by their corresponding “mask patches”. The originality of my work is that I extract
two main products from the stacked images:
• The radial temperature profile starting from the centre of the image, by computing
the mean temperature of the pixels in rings of fixed width and increasing angular
radius. Considering the characteristics of my stacked images, it is calculated here
for 150 radii between 0→ and 15→ , with a width of 0.1→ ;
• The aperture photometry profile, using the previously mentioned top-hat filter
approach: At each chosen angle, we compute the photometry as the difference
between the mean temperature of the pixels inside the disk of the given angular
radius and the temperature of the pixels in the surrounding√ ring of same area. I
compute this profile for 150 angles between 0→ and (15/ 2) ∗ 10.6→ .
This differs from the original approach of Gr08 by adding a new dimension to the
results: as we will see, the temperature profile gives us a precious insight on the origin
and feature of any potential signal found in the photometry profile, which is itself
useful (among other things) for a better identification of the scale of these signals.
On a more technical note, the summation of square pixels contained inside a
disk can lead to calculation errors due to omitted fractions of pixels close to the
boundaries of the disk. To reduce these as much as possible, I upscaled the 301 × 301
stacked images into 1204 × 1204 images, each pixel of the original image being
divided into 16 sub-pixels with the same value. Statistical errors for these two profiles
are estimated by computing the standard deviation of each sample mean of pixels.
The stacked images themselves are useful mostly for illustration purposes, but I
mainly focused on the analysis of these two profiles in my work, looking for any
remarkable signal.
Of course, spotting a signal in the stacked images and/or their associated profiles
is not enough to conclude about a possible detection of the iSW effect. We have
to take great care in assessing the significance of our results: to do so I devised a
systematic way to compute the significance adopting a Monte Carlo approach. It is
quite similar to the Gr08 approach in the sense that I pick many sets (at least 10,000)
of N random positions on the sky (N being the number of structures used in the
stacking whose significance we want to assess). However, for each random set I
perform the same analyses as for the real structures, i.e. I produce not only a stacked
image but also compute its full radial temperature and photometry profiles. I store
all these profiles in memory and end up with thousands of temperature profiles and
photometry profiles. After this, for every angular size of the profiles, we compare
the results from the stack of real objects to the statistical distribution of results from
the random stacks. In practical terms, we calculate the signal-to-noise ratio (S/N) of
the data temperature and photometry profiles, at each angle considered.
With my protocol in place, I chose to apply it first on the catalogue of structures
from Gr08, using their original study of these objects as a basis for comparison. In
the following, I therefore use the catalogue (the 50 supervoids in particular) as a
“fiducial stack” for a series of tests to check the robustness of my method.
3.2.2 Gr08 Catalogue: Preparation of the Maps
Let us start this section with a picture: an example of a stacked image from the
aforementioned procedure is shown in Fig. 3.2. I obtained it using the catalogue of
50 voids from the Gr08 catalogue and performing the stacking in the WMAP V map.
Although a cold spot seems to appear quite clearly, we can note that the picture looks
a bit different from the original image of Gr08 (the left panel of Fig. 3.1). The main
reason for this is that Gr08 rotated all the CMB patches before stacking in order to
align the major axis of all the structures, as mentioned at the end of Sect. 3.1.2. Since
Gr08 did not release the information about the orientation of their objects, I could not
reproduce their work exactly. However this should not be a source of worry, as the
relevant quantities that I derive from these images (i.e. temperature and photometry
profiles) are rotationally invariant. Another difference between Figs. 3.1 and 3.2 is
that the latter was smoothed with a Gaussian kernel with FWHM 1.4→ for aesthetic
purposes, whereas the former is in its original resolution.
3.2.2.1 About Resolutions
This leads me to an important point: for a given experiment (both WMAP and Planck),
all the CMB maps that I use inherently have different resolutions. Indeed, the angular
15 40
10
20
Temperature (μK)
5
Degrees
0 0
-5
-20
-10
-15 -40
-15 -10 -5 0 5 10 15
Degrees
Fig. 3.2 Image resulting from my stacking procedure at the location of the 50 voids of Gr08, here
using the V band CMB map of WMAP. The cold spot reportedly due to the iSW effect is visible
roughly at the centre of the image with an angular radius of about 4→ (highlighted by the white
circle). Note the North–South temperature gradient, discussed in Sect. 3.2.2.3
resolution of a telescope is proportional to the observed wavelength: the CMB maps

obtained at higher frequencies will therefore have finer resolutions. Another conse-
quence of using maps at multiple frequencies is that each one of them will contain
different types and levels of foregrounds which may contaminate my stacking mea-
surements and produce inconsistent results across frequencies. Consequently, I chose
to first assess the impact of these properties of each map, using my fiducial stacking
as a basis.
In order to have a consistent stacking analysis through all the considered maps and
frequencies, I “standardised” these maps by smoothing them to the same resolution,
lower or equal to the lowest one of them. The beam sizes of the maps we considered
span from 30 to 12 arcminutes for the WMAP Q-V-W bands, and 28 to 4 arcminutes
for the Planck maps from 44 to 353 GHz. I chose therefore to smooth all the maps
using a common Gaussian kernel with a FWHM of 30 arcminutes.3 Figure 3.3 shows
the effect of such smoothing for the three WMAP maps through the temperature and
photometry profiles of the fiducial stacking. The stacks at each frequency, both raw
and smoothed, give roughly the same results with only percent-level differences es-
pecially for the photometry profiles, which are the most useful products here. The
degradation of the V and W maps to the lower resolution of the Q map naturally
smooths the measured profiles and reduces their dispersion around the results of the
Q band, which is the desired effect. We note identical behaviours when smoothing
the different frequency maps from Planck. Otherwise, this procedure does not mod-
ify significantly the amplitude and angular dependence of the profiles, so that we
can adopt this choice of a new common resolution for all maps and experiments,
facilitating the comparison of results across frequencies.
3 To be more precise, in order to bring the maps to a common resolution we need to smooth each
√
one of them by a Gaussian kernel with a FWHM equal to 30∼2 − R 2 with R the original resolution
of the considered map.
Fig. 3.3 First plot radial 15
Temperature (μK)
temperature profiles (top 10
panel) of the stacking of Gr08 5
voids, done on WMAP Q, V
0
and W maps (both in native V smoothed W smoothed
resolution and smoothed by -5
Q band V band W band
a 30.6 arcminutes kernel). -10
2
The differences in the profiles
ΔTV
0
between the smoothed and
-2
original maps are plotted
2
below the main plot (middle
ΔTW
0
V band, bottom W band).
-2
The width of the shaded
curves corresponds to the 2 4 6 8 10 12 14
Angular radius (o)
statistical errors on the profile 5
Photometry (μK)
measurements. Second plot

Same graphs and legend as 0
above but for the aperture
-5
photometry profiles of the
stacking of Gr08 voids -10
2
ΔTV
0
-2
2
ΔTW
0
-2
2 4 6 8 10
Aperture angle (o)
The corresponding photometry profiles using the Planck frequency maps are
shown in Fig. 3.4, and the equivalent of the stacked images of Fig. 3.1 with the official
Planck CMB map are displayed in Fig. 3.5. We keep in mind that the smoothing
procedure with a ∗30 arcminutes beam blurs the information and details contained
below this scale. Therefore we should not devote too much attention to any feature
in the profiles at angles lower than this value. Of important note is that the flux
measured is quite constant across all the considered frequencies, which (at this point
in the analysis) supports the idea that the signal originally pointed out by Gr08 is
indeed due to the iSW effect induced by structures.
3.2.2.2 Effects of Foregrounds
One other potential source of concern comes from the influence of foregrounds
present in the CMB maps, as they might mimic the expected iSW signal in the
stacked images (radio point sources for example could produce such effect with their
flat spectrum). The Planck data that I used (both the frequency maps and the output
from component separation) and the foreground-cleaned WMAP maps are supposed
Photometry (μK)
0
44 GHz
-5 70 GHz
100 GHz
143 GHz
217 GHz
-10 353 GHz
Photometry (μK)
5
-5
2 4 6 8 10
Aperture angle (o)
Fig. 3.4 Aperture photometry profiles measured in the stacked patches centred on the Gr08 super-
voids (top) and superclusters (bottom), using the SEVEM-cleaned Planck frequency maps. As we
can see, these are almost identical for all frequencies except for small differences at low angular
scales (<2→ ): these are most likely due to small-scale artefacts of foreground remaining after the
component separation. The interpretation of these profiles will be discussed later in Sect. 3.2.3
Clusters Voids (Clusters-Voids)

40
20
0 μK
-20
-40
-10o -5o 0o 5o 10o -10o -5o 0o 5o 10o -10o -5o 0o 5o 10o
Fig. 3.5 Stacked regions of the Planck CMB map (smoothed at 30 arcminutes) corresponding
to the locations of the superstructures identified by Gr08. From left to right we show the images
resulting from stacking of the 50 superclusters, the 50 supervoids, and the difference of both. The
superimposed white circles indicate the angular radius at which the signal-to-noise ratio is maximal
to be free of such contaminants. However, no component separation technique can

be considered as absolutely perfect: residuals always exist. To assess their possible
impact, I performed the stacking of the Gr08 voids first on raw and then foreground
cleaned CMB maps at all frequencies of WMAP. I then looked for differences be-
tween the two results, either in the amplitude or in the shape of the signal. Results
are illustrated in Fig. 3.6: we obtain systematic offsets of a few micro-kelvins in
the radial temperature profiles and less in the photometry profiles. This indicates
that the foreground removal (by the WMAP team) ends up removing an almost uni-
form background in my stacked images, which does not influence much the aperture
Fig. 3.6 For the Gr08 Temperature profiles

stacking, differences in
the temperature (top) and
0
photometry (bottom) profiles
ΔT [cleaned-raw] (μK)
between foreground cleaned
maps and raw maps, for the -1
three frequency bands con-
sidered. The width of the
-2
shaded curves corresponds Q band
to statistical errors on the
profile measurements. The -3 W band V band
quasi-flat offsets observed in
0 5 10 15
the temperature profiles do Angular radius (o)
not affect substantially the
photometry Photometry profiles
1.5
ΔT [cleaned-raw] (μK)
1
0.5
0
-0.5
-1
-1.5
2 4 6 8 10
Aperture angle (o)
photometry of the stacked image. It does not mean of course that these removed
foregrounds are uniform background across the sky, but rather that the number of
stacked patches is sufficient to smooth the contribution of the foregrounds to the
final stacked image. I still used the foreground cleaned maps for my analyses, so that
any measurement of a potential iSW signal would be as close as possible to its true
underlying amplitude. However, this test allowed to conclude that possible residuals
in the cleaned maps should not be of any concern since we expect them to have the
same “flat” behaviour in the stacked images as the foregrounds themselves.
3.2.2.3 Unforeseen Features
Another map-related problem showed up during my investigations: I observed that a

clear temperature gradient appeared in many of my stacked images (already visible
in Fig. 3.2), roughly on a North–South axis with hotter high latitudes. Foregrounds
as a possible source of this can be excluded, because the gradient appears in both
raw and foreground cleaned maps, and also—to a lesser extent—because we would
expect foreground contamination to be hotter close to the galactic plane. This feature
is not seen (or reported) in the work of Gr08, because of the rotation they performed
on the CMB patches (see the beginning of Sect. 3.2.2) which randomised and erased
this gradient.
Fig. 3.7 Orthographic

projection of the δ = 2–6
multipoles map extracted
from the foreground reduced
WMAP Q map, in galactic
coordinates. Only half of the
map is visible (centred on the
galactic North pole), with a
mask showing only the area
covered by the main SDSS.
A graticule grid has been
superposed with a 45→ step in
longitude and 30→ in latitude.
The border of the image cor-
responds to the galactic plane,
while the point at the very
bottom is the galactic centre
-42.6 42.6 microK
The choice of map (foreground-cleaned or not) did not influence this gradient,
ruling out the foregrounds as culprits. To have a better insight on the origin and feature
of this large scale gradient, I decomposed the CMB maps on the spherical harmonics
and to analyse the individual contribution of each multipole. This approach was
motivated by the methods used in other works in the literature dedicated to the study
of large scale anomalies in the CMB (e.g. [1, 18]). This idea proved fruitful as it
appears that the measured gradient is mainly caused by the first few multipoles of the
CMB maps we use, especially by the δ = 6 multipole map. In the region of the sky
covered by the SDSS (where all the superstructures we considered are located), these
multipoles combine to yield indeed a strong North–South gradient (see Fig. 3.7 for
WMAP) which will be present at some level in every patch of CMB, and therefore
in the final stacked image. As shown in Fig. 3.8, subtracting the contribution of these
multipoles does remove the gradient in the stacked image but has an almost negligible
effect in the photometry profile. Indeed, the removed contribution has a shape close
to simple tilted plane (see Fig. 3.8) which does not affect the aperture photometry.
In the present work, I used both maps with and without these few multipoles
removed and did not see any significant difference in the stacking results, except
in very particular cases (that I mention later in this chapter). This gradient should
therefore not be a source of particular worry in this context, although it should be
kept in mind as it always at least add an offset in the measured temperature profiles.
Original l=2-6 removed l=2-6 contribution

15o 40
o
10
20
5o
0o 0 μK
-5o
-20
-10o
o
-15 -40
-10o -5o 0o 5o 10o -10o -5o 0o 5o 10o -10o -5o 0o 5o 10o
15
10
Temperature (μK)
5
0
-5
-10
0 5 10 15
Angle (o)
Fig. 3.8 The three images represent the stacking of the Gr08 voids done (from left to right) on the
cleaned WMAP V map, on the same map without its δ = [2, 6] multipoles, and on these multipoles
only. The temperature (dotted curves) and photometry (solid) profiles shown on the rightmost plot
are obtained from the first (black curves) and second (red) stacked image. The temperature offset
induced by the removed multipoles does not affect the photometry
3.2.3 Application to Gr08: Significance and Analysis
In all the stacking results shown so far, a signal appears in the photometry profile with
a maximum (in absolute value) at an angular scale of about 3.5→ for the Gr08 voids
(see Figs. 3.3 and 3.4) and around 4.5→ for the clusters (see Fig. 3.4), the preferred
size changing only very slightly between frequencies. These figures mark the first
departure from the Gr08 results as my analysis highlights a respectively smaller and
larger scale for the voids and the clusters. Before trying to further interpret those
results, one must remember that these profiles by themselves are not enough: we
need to compute their significance.
I described the procedure for such a computation at the end of Sect. 3.2.1: I applied
it to assess the significance of the stacking of the Gr08 structures, as illustrated in
Fig. 3.9 (using there the CMB of Planck). For this particular example, I used more
than 15,000 sets of 50/100 random positions, enough to sample the distribution of
temperature and photometry profiles. This is verified in Fig. 3.10 where the histogram
of the “random” photometry values at a given angular size follows closely a Gaussian
distribution.
Using the same catalogue as Gr08, I find results in reasonable agreement with their
original study (see Fig. 3.9), with only slightly different amplitude (of order ∗ μK)
and significance, and small shifts in the preferred scales (<1→ ). These differences can
be imputed to our use of different CMB maps (WMAP 7 and Planck vs. WMAP 5)
Temperature profile (μK)

15
10
1σ
5 mean
1σ
0
-5
-10 Clusters
Voids
0 5 10 15
Angular radius (o)
10
Photometry (μK)
5
1σ
0 mean
1σ
-5
-10
10
Photometry (μK)
8 o
6
4
2 1σ
0 mean
1σ
-2
2 4 6 8 10
Aperture angle (o)
Fig. 3.9 Temperature (top) and photometry (middle and bottom) profiles of the stacked patches
(from the Planck CMB map) at the location of the 50 supervoids and 50 superclusters of Gr08.
The lower panel shows the combined photometry profile (i.e., the average cluster profile minus
the average void profile). The significance is represented by 1, 2, 3 σ level curves. These curves
represent the dispersion of the thousands of stacks of 50 CMB patches chosen at random positions
(for illustration, on the top panel, we represent in grey a few hundreds of those random profiles)
At θ = 3.74o
Occurences
-10 -5 0 5 10
Photometry (μK)
Fig. 3.10 Distribution of photometry values for a single aperture (here 3.74→ ) from 14,000 stacks
of 50 random positions. The mean, 1, 2 and 3 σ values are marked the same way as in Fig. 3.9.
The blue long-dashed line on the left shows the value obtained from the fiducial stacking of the 50
voids. The best fitting Gaussian (red solid curved) follows closely the distribution
which may have different levels of foreground residuals, and to a lesser extent to light
differences in the practical application of the stacking procedure (profile calculations,
significance estimation, etc.). While we can argue its cosmological origin, the signal
is clearly not peculiar to WMAP 5 CMB maps: as mentioned before, it is essentially
identical across frequencies as expected of the iSW effect. However, an important
feature shows up in the temperature profile of the stacked image of the 50 voids and
its significance. Indeed, in the top panel of Fig. 3.9 we see that the central cold spot
of the signal (below 3.5→ ) does not particularly stand out compared to random stacks
(1–2σ significance only). On the other hand, I measured around the spot a wide hot
ring with a constant significance (around 1.5 σ ) at scales between 3.5→ and 10→ , also
clearly visible in the stacked images (see e.g. Fig. 3.5). This ring even has the effect
of rising the mean temperature of the stacked image in the region delimited by a
circle of ∗9→ radius: this has a visible impact in the photometry profile at higher
scales, with a significance rising again around 9→ .
Interpreting it in the light of the iSW effect, this would imply the presence of
much larger overdensities surrounding the already large supervoids. Considering the
filamentary structure of the Universe (where voids occupy the majority of the vol-
ume), this situation is unlikely and the source of this hot ring remains unknown. This
peculiarity leads us to question whether the measured central cold spot—physically
interpreted as an iSW signal—is really remarkable: it might as well be due to ran-
dom fluctuations of the CMB, of which the significance in the photometry profile
was coincidentally strengthened by a surrounding hot region in the stacked image. In
a similar line of thought, the temperature profiles from the stacked clusters does not
present such peculiarity, although we note that its most prominent photometry signal
is quite wide and has a lower significance. But more importantly, the photometry
profile peaks at angular scales more than twice as large as those of the underlying
clusters (which have a mean radius of 38 Mpc and a mean angular scale of ∗1.7→ ).
Reaching any definitive conclusions on these particular results (and the scale of
the measured signals) would require a complementary rigorous investigation through
theory and/or numerical simulations of the iSW effect expected from such sets of
superstructures with their large distribution of sizes. This is beyond the scope of this
chapter, but this problem is treated in Chap. 4 of this thesis. However, we can try to
account for the size of each structure inside the stacking procedure itself, which is
the purpose of the next subsection.
3.2.4 Refining the Stacking Analysis
A quick look at Fig. 3.11 is worth a thousand word: in all the stacking that I per-
formed so far, I actually mixed together structures with a relatively large variety of
angular sizes on the sky. Combined with the fact that these objects have also different
physical sizes and redshifts (therefore different iSW signatures), it is hard to draw
any conclusion from the size of the signal measured by photometry. To account for
this, I revised my stacking protocol in order to include a rescaling of the structures
Fig. 3.11 Histogram of the Gr08 voids Gr08 clusters

angular sizes of the 50 voids
15
(left panel) and 50 clusters
No of objects
(right) of the Gr08 catalogue.
Note that the x axis is not the 10
same in both panels
5
0
2 4 6 8 1 1.5 2 2.5 3
Angular radius (o) Angular radius (o)
according to their effective radii. In practical terms, this meant cutting each CMB
patch with the same number of pixels but with a tailored resolution so that the cor-
responding object always occupies the same surface on the image (hence also in the
final stacked image).
The resolution of the extracted patches now depends on the size of the correspond-
ing structures: I chose to cut square patches with a side several times4 the size of
the effective radius of the structure it contains. Naturally, I also adapted my protocol
for the estimation of the significance: each random patch is first subjected to the
same rescaling as the one done on the identified void patches. We can guess that this
will have the effect of mixing different scales in the original CMB map, and will
most likely amplify the variance of these random stacks compared to the previous
estimation.
I begin with the fiducial stacking of the voids of Gr08: a comparison between the
stacked images and profiles is illustrated in Fig. 3.12. The signal which was identified
above still appears after rescaling, with the best significance at scales between 0.7
and 0.9 times the void effective radii.5 Such angles are not entirely surprising, as we
could expect the signal to show a size smaller than the objects themselves. We intuit
that the iSW effect should fade close to the border of the voids; combined to the
irregular geometry of the objects, it is conceivable that it would make the cold spot
noticeably smaller than the underlying structure. However, a strong disagreement is
found in the rescaled stacking of the superclusters, where the photometry prefers a
scale that is 2.6 times the effective radius of the structures! This mismatch could be
a result of an underestimation of the structure extent by the ZOBOV and VOBOZ
algorithms (as already suggested by Gr08) or because larger potential hills and valleys
(responsible for the iSW effect and mainly generated by the Dark Matter) underlie
the detected superstructures (composed of baryonic matter). Nevertheless, the factor
4 This was done in anticipation of a potentially large scale of the stacked signal, especially in the
light of the “non-rescaled” results from the Gr08 clusters in Sect. 3.2.3.
5 It should be noted that these results differ slightly from those reported my published paper [10],
the reason being that I made a mistake in the calculation of the physical radius of the Gr08 voids
(by a factor of ∗0.7). The purple points in Fig. 3.13 were affected as well. This does not alter
fundamentally my conclusions and I intend to submit an erratum to the editor of the paper very
shortly.
Without rescaling With rescaling

15o 2 40
o
10
1 20
5o
0o 0 0 μK
-5o
-1 -20
-10o
-15o -2 -40
-15o-10o -5o 0o 5o 10o 15o -2 -1 0 1 2
x void radius
15
Temperature (μK)
10
1σ 1σ
5 mean mean
0 1σ 1σ
-5
0 5 10 0.0 0.5 1.0 1.5 2.0

Angular radius (o) (x void radius)
5
1σ 1σ
Photometry (μK)
0 mean mean
1σ 1σ
-5
-10
2 4 6 8 10 0.2 0.4 0.6 0.8 1.0 1.2 1.4

Aperture angle (o) (x void radius)
Fig. 3.12 Top Stacked images of Gr08 voids, without (left) and with (right) a rescaling of the
CMB patches proportional to the void sizes. Middle and bottom Respectively the temperature
and photometry profiles for the original (left) and rescaled (right) stacking of Gr08 voids (same
conventions as Fig. 3.9)
of 2.6 for the case of superclusters seems hard to explain entirely with this argument.
Another culprit could be found in the LRG subsample of the SDSS itself that was
used for the identification of these structures; these LRGs are known to be biased
tracers of matter (see e.g. [22]), so that the structures identified could be substantially
larger than they appear. In any case, reconciling these measured signals with their
associated structures would imply that some important corrections are needed in the
characterising of these objects.
I also noted that the significance of these signals is found to be slightly lower
(∗0.5 less in the S/N), although this is partly due to the increased variance of the
signal (cf. the wider significance contours in the profiles) induced by the rescaling
as expected above. But it is also a consequence of the lower amplitude measured for
the signal, at odds with our expectations of the rescaling procedure. This could be
250 o
10
50
o
o
Pan et al.
10
o
40 8
15
20
o
30 6
200 o
Granett et al. voids 8
o
20 4
10
0.00 0.02 0.04 0.06 0.08 0.10
Effective radius (Mpc)
lrgdim o o
150 10 lrgbright 6
o
8
o
100 4
bright2 o
6
o
dim2 bright1 4
o
50 2
dim1 o
2
o
10
o
8
0
0.0 0.2 0.4 0.6
Redshift
Fig. 3.13 Effective radii (in Mpc) as a function of redshift for all the voids used in this work. The
two catalogues of Granett et al. and Pan et al., and the six subsamples of Sutter et al. are delimited
by boxes labelled by their names. The black curves are here to indicate the angular size (in degrees)
of a structure in the redshift-radius plane
a further hint that random CMB fluctuations actually contribute notably to the signal
seen in the stacked image. As a take-away message, the use of the rescaling certainly
shows that it is a crucial step to reveal possible inconsistencies in the results from the
stacking of structures. But in spite of all these comments, the Gr08 catalogue still
yield results that remain at a high significance. One of the possible ways to gather
some insight on those results would be to look for alternative objects to study, instead
of being confined to the analysis of those only 100 structures.
3.3 Analysis of Additional Voids
Luckily, a few other works dedicated to the identification of structures were published
since the original Gr08 paper, with publicly available catalogues. In this section, I
describe these new datasets and perform a similar stacking analysis with these new
objects.
3.3.1 The Catalogues of Pan et al. and Sutter et al.
Two catalogues in particular held my attention: both are based on the same survey
for the identification of structures, here the SDSS DR7 release. Of important note is
that these two catalogues contain only voids, but these are actually more adapted for
stacking studies. Indeed, emission of astrophysical origin is less likely to contaminate
the iSW signal from these objects.
The first catalogue that I considered was published by Pan et al. [14] (Pan12
thereafter), who used the VoidFinder algorithm as described by Hoyle and Vogeley
[9] to identify and catalogue 1,055 voids with redshifts lower than z = 0.1. They
provide for each void all the information that I need for my stacking studies and
even more: the position on the sky, the physical radius (defined as the radius of the
maximal sphere enclosing the void), an effective radius (as voids are often found
to be elliptical) defined as the radius of a sphere of the same volume, its physical
distance to us, its volume and mean density contrast. The largest void is just over
47 Mpc in effective radius while their median effective radius is of about 25 Mpc.
Some are both very close to us and relatively large (more than 30 Mpc in radius)
resulting in large angular sizes on the sky, up to 15→ and above.
The second (and most recent) catalogue that I studied was released by Sutter et al.
[21] (Sut12 thereafter). Using their own modified version of the already mentioned
void finding algorithm ZOBOV, they took particular care of accounting for the effects
of the survey boundary and masks. I should mention here that since its first release in
July 2012, the catalogue has been subjected to regular updates (contrary to the Pan12
catalogue) and modifications reflecting improvements in the detection algorithm,
bug corrections, and inclusion of additional void data. In the latest version of their
catalogue (at the time of this writing, mid-2013), they found a total of 1,495 voids
which they divided into six redshift subsamples: four extracted from the main SDSS
(dim1, dim2, bright1 and bright2) and two from the SDSS LRG sample (lrgdim and
lrgbright). The redshifts of these voids span z ∗ [0, 0.4] while their sizes range
approximatively from 5 to 150 Mpc. This catalogue stands out by the amount of
information provided about its voids: position of the barycentre, redshift, effective
radius, locations of member galaxies, one-dimensional radial profiles of stacked
voids, two-dimensional density projections and other statistical information about
their distribution.
As a summary, we plot the properties of the voids from these two catalogues
(including also the 50 voids of Gr08) in Fig. 3.13, including their redshifts, effective
radii and corresponding angular sizes. More details about all the subsamples can be
found in Table 3.2. It is now time to take a closer look at these catalogues and see if
they can give us any precious results and knowledge about the iSW effect.
3.3 Analysis of Additional Voids 71
Table 3.2 Summary of all the void catalogues considered in the stacking procedure
(Sub)sample name Redshift range Number of voids
Sut12 0.003 < z < 0.43 1495
dim1 0.003 < z < 0.048 218
dim2 0.05 < z < 0.1 419
bright1 0.1 < z < 0.15 341
bright2 0.15 < z < 0.20 176
lrgdim 0.16 < z < 0.35 291
lrgbright 0.36 < z < 0.43 50
Gr08 0.43 < z < 0.69 50
Pan12 0.009 < z < 0.1 1055
3.3.2 New Objects, New Results
For the study of these two new catalogues, I kept the same protocol that I used for the
Gr08 catalogue, i.e. the same method of cutting and stacking the patches of CMB, the
same computation of the temperature and photometry profiles, as well as the same
estimation of their significance and the same rescaling procedure.
As a first (and relatively crude) approach, I tried to stack these voids “as such”
by subsamples, i.e. stacking the voids in each of the six samples of Sut12, and not
trying any division of the Pan12 catalogue. Results for the photometry profiles and
their significance are shown in Fig. 3.14.
The stacking of the 1,055 voids of Pan et al. gives a faint signal below the 1 σ
level at about 1→ which does not allow any interpretation. Such lack of results may
have been expected and could be due to the high number of voids in the sample and
their very wide distribution of angular sizes. I considered the idea of dividing this
catalogue into subsamples based on redshift, radius, and/or angular sizes; however
after several attempts, it did not yield any significant result, and I faced a number
of issues regarding possible a posteriori selection effects and the hassle of finding
appropriate bins of size for the subsamples. Indeed, a smaller number of structures
implies a narrower range of sizes and redshifts, but it also greatly reduces the signal-
to-noise ratio.
This issue was partially addressed for the Sut12 catalogue when I chose to analyse
separately the six redshift subsamples—a simple division based on redshift but in
principle free from selection effects. But as it can be seen in Fig. 3.14, only one of
the subsamples (dim1) yields a negative signal in the photometry with a significance
higher than 1 σ , the other profiles either spanning between 0–1 σ or being entirely
positive (lrgbright). An explanation for this apparent lack of significant results may
be found again by considering the dispersion in the angular radii of the voids. Even if
the subsamples from Sut12 contain significantly more objects compared to the Gr08
sample (and therefore should strengthen the signal), their sizes on the sky are also
more scattered (as can be seen in Fig. 3.13). Mixing such a variety of void sizes will
3
2
Photometry (μK)
1 1σ 1σ
mean mean
0 1σ
-1 1σ
-2
-3
0 1 2 3 4 5 6 7 8 9 10
Aperture angle (o)
3
2
Photometry (μK)
1σ 1σ
1σ
1
mean mean 1σ
mean
1σ mean
0
1σ 1σ
-1 1σ
-2 dim1 dim2 bright1

6
bright2 lrgdim lrgbright
Photometry (μK)
4
1σ
2 1σ 1σ
mean mean 1σ
mean
1σ mean
0
1σ 1σ
-2
0 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10
Aperture angle (o)
Fig. 3.14 Top panel Photometry profiles from the stacking of Pan12 voids (same conventions as
Fig. 3.9). Bottom panel Photometry profiles from the stacking of Sut12 voids and its six subsamples
(same conventions)
necessarily dilute the associated iSW signal over the same range of scales, and might
drastically reduce its significance.
This preliminary analysis strongly suggests that it is necessary to take into account
the size of each individual void in the stacking procedure in order to improve the
significance of the results, i.e. use the rescaling procedure that I already performed
with the Gr08 catalogue. As we have seen earlier, if the rescaling process does not
at least increase the absolute amplitude of any previously detected signal, then any
subsequent significance estimation is very unlikely be higher than without rescaling.
Indeed, since the variance will also necessarily increase, this will in turn further
decrease the S/N. Therefore I chose to first produce an overview of the rescaled
photometry profiles for Sut12 and Pan12 in Fig. 3.15.
Again, no signal of particular importance arises from this new analysis of Pan12
voids, with amplitudes which do not depart much from the original stacking except
at very small angular sizes, in all likelihood due to random fluctuations and unrelated
to the underlying voids. Concerning the Sut12, signals seem to arise in several of
the rescaled profiles, especially around a scale equal to 0.5–0.55 times the voids
effective radii, with a clear departure from the results without rescaling for some of
them, e.g. the lrgdim subsample. However, some of the other subsamples (bright1,
lrgbright) do not apparently benefit from the rescaling procedure. I identified the two
Rescaling of Pan et al.

2
Photometry (μK)
1
0
-1
-2
0.0 0.5 1.0 1.5 2.0
Rescaling of Sutter et al.

lrgbrig
6 ht
Photometry (μK)
4
bright2
2 bright1 dim2
lrgdim
0 dim1
-2
-4
1
bright2 lrgbright
Estimated SNR
bright1 dim2
lrgdim
0 dim1
-1
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
Aperture (x void radius)
Fig. 3.15 Summary of the photometry profiles extracted from the rescaled stacks of Pan12 (top) and
Sut12 (middle), performed in the WMAP V band cleaned map. The coloured dashed lines indicate
for each sample the lowest amplitude measured in the original stacked image (without rescaling the
voids). These allow to roughly estimate if the rescaling procedure did improve the detection
√ of any
previously detected signal. The bottom panel shows the previous profiles multiplied by Nv , with
Nv the respective number of voids in each subsample. They are then normalized to the strongest
signal (lrgdim). These curves provide an√estimate of their potential significance as the noise in the
stacked image is expected to scale as 1/ Nv approximately
most promising subsamples, i.e. dim1 and lrgdim, and evaluated the significance of
their rescaled profiles. The results, shown in Fig. 3.16, are close to the expectations
from Fig. 3.15: the dim1 subsample yields a signal at ∗0.52 times the void effective
radius, with a significance similar (albeit a bit smaller) to the the results without
rescaling (around 1.36 σ ). This is coherent with the fact that the amplitude of the
signal remained almost at the same level (illustrated by the corresponding dashed
line in Fig. 3.15) whereas the rescaling procedure slightly increased the variance
of the random stacks used in the S/N estimation. Remembering that we are still
looking for an iSW signal, I can note that the apparently significant signals at small
aperture angles (∗0.15× and 0.3×) are not constant across frequencies and therefore
are probably not related to an iSW effect. Regarding the lrgdim subsample, it gives
a ∗2.35 σ signal around 0.57 times the void radius, a clear improvement over the
non-rescaled results which was expected considering the stronger amplitude of the
rescaled signal. For both subsamples, we note that the amplitude of the highlighted
Fig. 3.16 Photometry profiles

and significance for the
Photometry (μK)
rescaled stacks of two Sutter 1σ
dim1
et al. subsamples, dim1 and mean
lrgdim (same conventions as
1σ
Fig. 3.9)
Photometry (μK)
1σ
lrgdim
mean
1σ
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

signal stays remarkably constant across frequencies, whereas as mentioned before

the other potentially significant features of the profiles are not achromatic.
Interestingly, the highlighted scales show up consistently around (if not slightly
above) half the effective radius of the considered voids, but this will be discussed
more extensively in the following subsection. For now, I noted that the rescaling
procedure was very fruitful in enhancing the photometry signals, and highlighted a
seemingly consistent scale for several subsamples. Inspired by these observations, I
wondered if the data from the Sut12 catalogue may be used in a more efficient way
to further explore this hint of a signal.
3.3.3 Further Improvement of the Protocol
From the previous subsection, I gathered that in some of the Sut12 subsamples, an
iSW-like signal seems to appear around 0.5–0.6 times the voids effective radii, and
that it is especially significant in the lrgdim sample. A possible explanation may
come from the presence in this particular subsample of some of the largest voids
in the whole Sut12 catalogue (as can be seen in Fig. 3.13) which are expected to
yield the strongest iSW effect according to linear theory. We could argue that the
lrgbright sample also contains many large voids but does not show such significant
signal: however this could be explained by its small number of objects (50) and the
consequently high level of noise from primordial CMB fluctuations in the stacked
image.
This indicates that instead of considering each subsample separately, a better
approach may be to combine them all and stack the voids starting from the largest
ones. Indeed, in theory the noise should scale as usual roughly as the inverse square
root of the number of stacked voids, but the stacked iSW signal is also expected to drop
at some point due to the addition of smaller and less contributing voids. By starting
from the largest voids, we intend to select the supposedly largest iSW contributions
in order to keep the stacked signal from dropping too fast and effectively to boost
the S/N of the detection. I carried out this analysis on the 1,495 voids of Sut12,
first focusing on the whole photometry profiles and increasing progressively the
number of stacked voids. As expected, a negative signal consistently appears around
an aperture of 0.54 times the voids effective radii. As intuited before, its amplitude
gradually decreases as we include smaller and smaller voids in the stacking. In order
to estimate the significance of this signal, I focused on the value of the photometry
at this particular aperture scale: in the top half of Fig. 3.17, we show these values
as a function of the increasing number of stacked voids. Similarly to the previous
section, we estimate the significance of these values by repeating the analysis many
times after randomly shifting the stacked positions. Therefore we can compute the
S/N of these results, shown on the bottom half of Fig. 3.17. We note once again that
the photometry is stable across frequencies and consistently negative for practically
any number of stacked voids, but the shape of this curve and its significance are
hard to interpret. The significance first rises up to ∗2.3 σ for the first 200 stacked
voids, a behaviour that would be expected from an iSW signal that progressively
takes over the CMB noise. After this, the S/N quickly decreases and then oscillates
between ∗1–2 σ before dropping after stacking more than 1,300 voids. Although
this significance appears to vary quite significantly, the stability of the signal itself
(always negative and at the same scale) may indicate that this variability is due to
random CMB fluctuations.
The scale of the detected signal clearly differs from the large sizes that were
obtained with the Gr08 catalogue. What is encouraging here is that a physical expla-
nation can be found for these smaller scales in the geometry of the voids from these
subsamples. Indeed as noted by Sut12, the majority of them present a shape similar
to a prolate ellipsoid with an ellipticity close to 2. Since the orientation of these voids
is a priori random, we can intuit that stacking such ellipsoids (and their associated
iSW signature) will eventually give a circular signal, with a smaller typical scale,
closer to half the major axis of the ellipsoids.
Following this, I selected two particular numbers of voids with a high signifi-
cance observed in Fig. 3.17, namely 231 and 983 voids, and performed the same
photometry/significance analysis as usual. The results are shown in Fig. 3.18: both
profiles highlight a scale again equal to 0.54 times the voids effective radii where
the signal reaches significances equal to 2.38 σ and 2.20 σ for the 231 and 983 voids
respectively. Despite the bigger number of voids, the small drop in significance may
again be caused by CMB fluctuations and/or the inclusion of small voids. Another
notable feature appears in both photometry profiles, namely one hot significant sig-
nal at smaller scales (∗0.2×) and one at larger scales (>0.9×). The former and
its associated scale seem to coincide with an angular size close to 1→ . It is directly
related to the curious fact that the minimal temperature spots do not appear at the
centre of the stacked image (see Fig. 3.18). Were the stacks centred on the coldest
spot, I would not have obtained this positive signal in the photometry profiles at
small angles. The origin of this hot central spot is undetermined, but could be due to
several contributions of which the background CMB fluctuations, the irregular shape
Signal (μK) at 0.54x

0 mean
1σ
-2
-4
-6
-8
2.5
2.0
SNR at 0.54x
1.5
1.0
0.5
0.0
0 200 400 600 800 1000 1200 1400
Number of stacked voids
Fig. 3.17 Top Photometry values of the stacked and rescaled voids from the Sutter et al. catalogue,
at an aperture of 0.54 times their effective radii, as a function of the number of voids (sorted
by decreasing size). Significance contours are computed using random stacks produced from the
WMAP V band cleaned map. Legend is identical to previous plots of photometry profiles (see
Fig. 3.9). Bottom For the stacking performed in the WMAP V map, S/N of the above photometry
values computed using the significance contours
4
Photometry (μK)
2
231 voids
1σ Stack for 231 voids

0 mean 2 30
1σ
-2 1 20
-4 10 μK
3 0
Photometry (μK)
2 0
983 voids
1 1σ
-1
-10
0 mean
1σ -2
-1
-2 -1 0 1 2
-2 x void radius
-3
0.2 0.4 0.6 0.8 1.0 1.2 1.4
Fig. 3.18 On the left, photometry profiles and significance of the rescaled stacking of the 231 (top)
and 983 (bottom) largest voids from the Sutter et al. catalogue (same conventions as Fig. 3.9). On
the right, stacked images of the 231 largest voids with a rescaling of the CMB patches proportional
to the void sizes
of the underlying voids, and the possible mismatch between the position of the void
barycentres and their most underdense zones. Concerning the signal at larger scales,
what comes to mind is the possible influence of large scale fluctuations through the
low multipoles of the CMB, already glimpsed at in Sect. 3.2.2.3 with the large North–
South gradient. Thus I redid the stacking of the same sets of voids on new maps with
a few more low multipoles removed (from δ = 2 to 20). The results then showed
that these multipoles indeed have a non-negligible contribution to our photometry
profiles: removing them reduces noticeably the measured amplitude at large scales
but keeps almost intact the rest of the signal. Although it does not account for the
entirety of the large scale signal, it does reduce its significance to less remarkable
levels.
In summary, contrary to the Gr08 catalogue, the rescaling process had positive
results on the much larger catalogue of Sut12, highlighting a particular scale around
half the void sizes in all the tests performed, in better agreement with our intuitive
arguments. Although the maximum observed significance only reaches around 2.5 σ
and the signal depends quite significantly on the number of stacked voids and their
size, the persistent nature of the signal seems to bolster the case for iSW detection.
Lastly, the Pan12 catalogue did not benefit from the rescaling process, and I wish
to mention that I also explored the same incremental approach used with the Sut12
catalogue but it did not yield any significant results. I hypothesise that this is caused
by the narrow range of small sizes of these voids, whose faint iSW signal is likely
to be dominated by CMB fluctuations.
To further elaborate on the possible conclusions that we can deduce from the
present work, there are a couple of effects that can make the interpretation of the
results even more difficult and that we should keep in mind. I will mention a few of
them in the next section.
3.4 Precautions
3.4.1 Selection Effects
The results from the previous section highlight the difficulty of getting a clear de-
tection of a signal, at least when using the various catalogues as such. With this in
mind, one could be tempted to amplify the signals hinted at by isolating the voids
that contribute most in the stacking of a particular sample. I experimented with this
idea, working with one of the best S/N from the results of the previous section: the
rescaled stacking of the lrgdim subsample. From the 291 voids of the original set, I
kept only the half (146) that contributes the most to the minimum of the photometry
profile at the scale of 0.56 times the voids radii. The resulting photometry profile is
plotted in Fig. 3.19 and indeed shows a dramatic increase in the amplitude of the sig-
nal. From there, the crux of the matter is to assess the significance of this new result.
When using the same procedure as usual (many stacks of 146 random positions), I
obtained a surprisingly high S/N, above 11 σ . This value clearly overestimates the
real significance, as it ignores the selection that we performed on the sample. It is
therefore required that I revise our protocol as follows. I first generate many sets of
291 random positions. Then for each set, I select and stack only the half that con-
tributes most to the photometry profile at the same scale of 0.56 times the voids radii.
Once I draw enough such random stacks, I keep the rest of the procedure identical.
The corrected significance (Fig. 3.19) drops down to a level of ∗2 comparable with
Classical σ estimation Accounting for selection

1σ
0 mean
1σ 1σ
n
Photometry (μK)
me1aσ
-5
-10
-15 10σ
11σ
-20 12σ
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4
Aperture (x void radius) Aperture (x void radius)
Fig. 3.19 Photometry profile for the rescaled stacking of half the Sutter et al. lrgdim subsample
(146 voids out of 291), chosen so that the amplitude at 0.56 times the voids radii is the strongest
(see text for details). Left the significance is estimated with stacks of 146 random positions. Right
the significance accounts for the selection effect. The difference between the two significances is
obvious and very pronounced
(even lower than) the initial S/N in the original stack (Fig. 3.16). Therefore, such an
a posteriori selection cannot be used to improve the S/N of the final stacked signal.
Actually, after playing a while and taking a closer look at these selection effects,
I noticed that with only one set of a few hundred random positions, one can obtain a
strong—but completely artificial—signal at almost any desired scale (Fig. 3.20) by
selecting the appropriate half of it. This further illustrates, and warns us about the risk
of a posteriori selections, and does put into perspective any apparently significant
signal we obtained. The results from the new catalogues can be considered safe, as
the only form of selection comes from the division of the Sut12 voids into redshift
subsamples already performed a priori by the authors. The Gr08 results are probably
also safe, although their 50 voids come originally from a ten times larger sample
and were selected according to a density contrast criterium. This selection was also
made by the authors prior to looking at CMB data. It is not clear however whether
that selection might have helped increase the S/N artificially. Of further interest is
the observation made by the Gr08 that either increasing or decreasing by a few
tens the amount of voids in their selected sample does make the significance drop
(mentioned in Sect. 3.1.3). This may either indicate a possible selection effect—thus
casting doubt on the iSW nature of the signal—or be an expected effect due to the
addition of noise-dominated voids (when increasing the number of voids) or the
deletion of contributing voids (when decreasing it).
3.4.2 Alignment and Overlap Effects
When interpreting my results, another important issue arises due to the number and
location of the voids studied. In principle, each one of them leaves an imprint in the
CMB temperature. These hundreds of voids are confined in the area covered by the
SDSS. Since the angular size of these objects often exceeds several degrees, they are
3.4 Precautions 79
Fig. 3.20 Illustration of 2o signal 3o signal

selection effects: from a single o
5
set of around 200 random
positions, we are able to 0o
construct a false “iSW-like”
signal at any desired scale -5o
(here for angular radii of 2,
3, 4 and 5→ ) by selecting 40
and stacking each time the 4o signal 5o signal
o
appropriate half of the same 5 20
set
0o 0 μK
o
-5 -20
-40
-5o 0o 5o -5o 0o 5o
bound to fill this area and overlap. As a consequence, the stacking of a chosen set of
voids contains in fact contributions from many others. Some structures may also be
very close to each other on the sky, further complicating the interpretation of signals.
I devised two ways of quantifying these issues:
• First, for each individual sample, we compute both the total area covered by all
the voids and the area where at least 2 voids overlap; we then compute the ratio of
these quantities, which represents the “self-contamination” of each sample.
• Then, for every possible pair of samples, we compute the fraction of the area of the
first sample shared by the second sample area—a measure of the contamination
between samples.
The results of this study showed that every sample presents moderate to high conta-
mination from other samples and self-contaminates itself quite strongly. The conta-
mination is less pronounced for the Gr08 sample due to the small number of voids.
Still, many voids from other catalogues will contribute to the stacking of the Gr08
voids, making it further difficult to interpret its relatively high significance results.
This is true for every catalogue of voids identified in the same area: the level of
overlap is very high and it would be difficult to determine which are the objects that
produce the actual iSW signal. Moreover, the proximity of the voids may artificially
amplify some detections due to the possibility of repeated stacking on the location
of a significant signal/void.
Both aspects of this study showed that contribution from other samples is indeed
frequent in the stacking of these voids. Supposedly, we can expect the contribution
to cancel out for large numbers of stacked patches, although this argument may be
weakened by the fact that voids represent a much larger fraction of the volume of the
Universe than large overdensities (as shown by the large scale N-body simulations
in the literature). In any case, the probability of measuring a false signal (due to a
fortuitous event) may be heightened by the overlapping of so many voids. An accurate
simulation of the expected signature from such a collection of overlapping structures
will help quantify these contaminations and their effects on a possible iSW effect
detection.
3.4.3 Spurious Detections?
I would also like to stress that the superstructure data sets should be considered with
caution. A quick comparison between the Sut12 and the Pan12 catalogues reveals
that although they share a common interval in redshift (between z = 0 and 0.1),
the structures identified by the two catalogues are very different. Whether it is by
their radius, redshift, or position in the sky, the two sets of structures are quite
hard to match: this mismatch may very well be due only to the different void finding
algorithms used in the two works, although the fact remains that they were both based
on the same survey (SDSS DR7) and should have more pronounced similarities.
On another note, during my work on the stacking of these superstructures, several
updates of the Sut12 catalogue were released. The changes were stated as having
no impact on the conclusions of the associated paper [21]. However, a detailed
examination showed that modifications ranged from minor to more consequent ones.
Numerous additional small voids were detected thanks to an improved void finding
algorithm; quite a few voids were removed from one version to the other, and many
others have seen modifications in their redshifts, sizes and positions on sky, up to
the point where one of the subsample was almost completely different (lrgbright).
These updates did consequently have a significant impact on the stacking of the
voids, especially through the inclusion of the new small voids. For instance, with the
July version, my procedure yielded a ∗2.5 σ detection of a negative signal at about
2.4→ in the bright2 subsample prior rescaling, while in the latest version of the same
subsample the photometry shows a positive excess with almost 2 σ significance.
Conversely, rescaling the voids improved the detection of a negative photometric
decrement only with the latest version of the catalogue. While the present status of
the catalogue should be considered as robust (according to a private communication
I had with P. M. Sutter), the fact mentioned above makes it rather difficult to interpret
without ambiguity the signals that I obtained.
The next-generation of galaxy surveys (such as LSST, Pan-STARRS or Euclid)
will help alleviating some of these concerns: thanks to a better control of the system-
atics and observational biases, the use of these future surveys will hopefully allow
a better identification of the structures confined within their boundaries. Combined
with the much larger volumes surveyed by these future missions (i.e. with improved
depth and sky coverage), the upcoming catalogues of structures will contain a large
number of trustworthy objects. The stacking studies will certainly benefit from such
new catalogues, and improve both the significance and our confidence in the results.
The depth in redshift will also help investigating the influence of alignments such as
those that I described in the previous subsection. Moreover, the increased precision
of these surveys will allow to derive additional information on the structures such as
their density/velocity profiles. This will in turn enable further cosmological tests and
3.4 Precautions 81
studies, such as the Alcock-Paczynski test (based on the geometry of the structures,
see e.g. [20]), or the exploitation of mass profiles to constraint the matter growth
rate.
3.5 What Does CDM Has to Say on the Matter?
Up to this point in the present chapter, we have mainly looked at the results from
the stacking procedure as such, trying to estimate their significance, in hope of
finding “some kind of signal”. Indeed, we only mentioned briefly the iSW signal
expected from the kind of structures studied here, and made few links with theoretical
calculations. In this last section, I assess where my results stand with respect to several
works in the current literature that tried various approaches to tackle the prediction
of the expected iSW contribution from superstructures.
3.5.1 Too Small or too Large Signal?
Before getting to the expected amplitude of the signal, let us focus on the size of the
signature itself first. As I mentioned in the previous sections of this chapter, there
seems to be a gap on this particular point between the stacking results of the Gr08
catalogue and the Sut12 one. The former yields a signal for the superclusters which
has a size significantly larger than the underlying structures (2.6 times their mean
angular radius), and a more reasonable scale for the supervoids (with a maximum of
significance around ×0.9 their size). On the other hand, the latter produce a signal
with a typical scale of about half the size of the voids, which may seem more natural
using some simple arguments about the iSW effect and the geometry of the structures
(cf. Sect. 3.3.3 for more details).
Nonetheless, a recent study by Cai et al. [2] comforts these intuitions: they con-
structed large N-body simulations following a νCDM cosmology and identified
the voids in the manner of Sut12 and Gr08. They subsequently computed the iSW
signature resulting from the stacking of their simulated voids which highlighted an
optimal scale around 0.6 times the effective radius of the voids for the analysis by
aperture photometry. The similarity between this predicted scale and the signal I
observed in the Sut12 void subsamples is encouraging. In contrast, using a different
approach [6] generated matter density maps (from both Gaussian and N-body sim-
ulations) that followed the redshift distribution of Gr08, from which they derived
full-sky iSW maps. After performing a stacking in these maps at the locations of
the density peaks/troughs, they found that the scales highlighted by aperture pho-
tometry ranged from 1→ to 20→ , with a maximum around 7→ —seemingly in better
agreement with the Gr08 results, in the sense that it explains why the highlighted
scale tends towards higher values. These somewhat contradictory predictions does
not allow to conclude on the agreement of my results with theory, although none of
the aforementioned works is really dedicated to compute the exact expected signal
from such structures—a task that will be at the centre of the next chapter.
3.5.2 Is ∼10 μK too Much?
The question of the supposed size of the iSW signal seems not solved yet. However,
both the aforementioned studies concur on one point: the amplitude of the Gr08
signal (about 9.6 μK) is significantly higher that what is expected. Using their N-body
simulations of a νCDM universe, Cai et al. [2] predict a photometry signal barely at
the level of 0.1 μK, while the Gaussian simulations of [6] yield a maximum signal
of ∗2 μK. They are joined in their conclusions by the work of [4] who developed a
theoretical model that goes beyond the linear approximation in order to compute the
expected signal of the Gr08 objects, only to find a 3 σ discrepancy between predicted
(around 2.2 μK) and measured signals.
Up to now, the predictions from the literature seems therefore to reach a consensus
on this matter; however, all of the results from these works are quite hard to directly
compare to any of the stacking results from a particular catalogue. Indeed, these works
consider the general features of the Gr08 structures (mean size, mean redshift, etc.)
without considering the properties of each individual structure, their own expected
signature, the influence of their relative locations on the sky, etc. This complicates
the comparisons between these predictions and the results from an actual catalogue
such as the Sut12 one. In this context, I felt that it would be necessary to be able
to compute an exact modelling of the impact of single superstructures on the CMB
temperature, which is precisely the the aim of the next chapter.
References
1. C.L. Bennett et al., Seven-year Wilkinson microwave anisotropy probe (WMAP) observations:
are there cosmic microwave background anomalies? Astrophys. J. Suppl. Ser. 192, 17 (2011)
2. Y.-C. Cai, M.C. Neyrinck, I. Szapudi, S. Cole, C.S. Frenk, A possible cold imprint of voids on
the microwave background radiation (2013). ArXiv:1301.6136
3. D.J. Eisenstein et al., Spectroscopic target selection for the sloan digital sky survey: the lumi-
nous red galaxy sample. Astron. J. 122, 2267–2280 (2001)
4. S. Flender, S. Hotchkiss, S. Nadathur, The stacked ISW signal of rare superstructures in νCDM.
J. Cosmol. Astropart. Phys. 2, 013 (2013)
background due to the integrated Sachs-Wolfe effect. Astrophys. J. 683, L99–L102 (2008)
6. C. Hernández-Monteagudo, R.E. Smith, On the signature of z ∗ 0.6 superclusters and voids
in the Integrated Sachs-Wolfe effect. Mon. Not. R. Astron. Soc. 435, 1094–1107 (2013)
7. C. Hernández-Monteagudo et al., The SDSS-III Baryonic oscillation spectroscopic survey:
constraints on the integrated Sachs-Wolfe effect. Mon. Not. R. Astron. Soc. 438, 1724–1740
(2014)
References 83
8. S. Ho, C. Hirata, N. Padmanabhan, U. Seljak, N. Bahcall, Correlation of CMB with large-scale

structure. I. Integrated Sachs-Wolfe tomography and cosmological implications. Phys. Rev. D
78(4), 043519 (2008)
9. F. Hoyle, M.S. Vogeley, Voids in the point source catalogue survey and the updated Zwicky
catalog. Astrophys. J. 566, 641–651 (2002)
10. S. Ilić, M. Langer, M. Douspis, Detecting the integrated Sachs-Wolfe effect with stacked voids.
Astron. Astrophys. 556, A51 (2013)
11. N. Jarosik et al., Seven-year Wilkinson microwave anisotropy probe (WMAP) observations:
sky maps, systematic errors, and basic results. Astrophys. J. Suppl. Ser. 192(2), 14 (2011).
ISSN 0067–0049
12. M.C. Neyrinck, ZOBOV: a parameter-free void-finding algorithm. Mon. Not. R. Astron. Soc.
386, 2101–2109 (2008)
13. M.C. Neyrinck, N.Y. Gnedin, A.J.S. Hamilton, VOBOZ: an almost-parameter-free halo-finding
algorithm. Mon. Not. R. Astron. Soc. 356, 1222–1232 (2005)
14. D.C. Pan, M.S. Vogeley, F. Hoyle, Y.-Y. Choi, C. Park, Cosmic voids in sloan digital sky survey
data release 7. Mon. Not. R. Astron. Soc. 421, 926–934 (2012)
15. Planck Collaboration. Planck 2013 Results. XII. Component separation (2013a).
ArXiv:1303.5072
16. Planck Collaboration. Planck 2013 Results. XXIV. Constraints on primordial non-gaussianity
(2013b). ArXiv:1303.5084
17. Planck Collaboration. Planck 2013 Results. XIX. The integrated Sachs-Wolfe effect (2013c).
ArXiv:1303.5079
18. Planck Collaboration et al. Planck 2013 Results. XXIII. Isotropy and statistics of the CMB
(2013). ArXiv:1303.5083
19. V. Springel et al., Simulations of the formation, evolution and clustering of galaxies and quasars.
Nature 435, 629–636 (2005)
20. P.M. Sutter, G. Lavaux, B.D. Wandelt, D.H. Weinberg, A first application of the Alcock-
Paczynski test to stacked cosmic voids. Astrophys. J. 761, 187 (2012)
21. P.M. Sutter, G. Lavaux, B.D. Wandelt, D.H. Weinberg, A public void catalog from the SDSS
DR7 galaxy redshift surveys based on the watershed transform. Astrophys. J. 761(1), 44 (2012).
ISSN 0004–637X
22. M. Tegmark et al., Cosmological constraints from the SDSS luminous red galaxies. Phys. Rev.
D 74(12), 123507 (2006)
23. D.A. Wake et al., The 2dF-SDSS LRG and QSO survey: evolution of the clustering of luminous
red galaxies since z = 0.6. Mon. Not. R. Astron. Soc. 387, 1045–1062 (2008)
Chapter 4
Towards a Full Modelling
of the iSW Effect
As I explored the possible impact of individual structures on the CMB in the previous
chapter, it appeared that evidencing the existence and characteristics of the Dark
Energy with this method was challenging. The stacking studies that I presented there
yielded promising, and interesting results—yet somewhat perplexing and uncertain,
which did not allow to draw any definitive conclusions on the matter.
One of the missing elements to this puzzle is a clear idea, if not an exact prediction
of the iSW signal expected from structures such as the ones that I considered, in the
framework of the CDM paradigm. Indeed, the theoretical studies performed so
far in the literature and their predictions (see Sect. 3.5 of Chap. 3) often resort to
simulations, approximations and the use of the linear regime of the evolution of
structures. None of them addressed the problem from that particular angle, i.e. the
exact modelling of objects similar to the observed ones and the computation of their
associated iSW signal. The motivation for such endeavour arises quite clearly in
the light of the previously discussed results, and is actually twofold. These “exact”
predictions would first allow to verify the validity of current claims of a detected
iSW signal, such as the results of [11], especially when we consider the multiple
discrepancies that are associated to it (size, amplitude, …) and which I pointed out in
the previous chapter. The new results that I presented, using more recent catalogues
of structures, would also benefit from a validation from these kind of predictions.
Secondly, such predictive work would constitute a formidable tool for exploring the
range of iSW signals that can be generated by potential structures (with realistic
features) in the Universe. Then, it naturally follows that it could also be used to
determine the requirements for a significant iSW detection through the means of
stacking, in terms of sample size, the objects characteristics, positions, etc.
The structure of this chapter will revolve around these two axes of thought,
preceded by a description of the process of elaborating a predictive tool and
framework.

86 4 Towards a Full Modelling of the iSW Effect
4.1 Computing the iSW Effect from a Structure:

A Step-by-Step Recipe
4.1.1 Modelling an Inhomogeneity
The need for a precise description of a single structure and its evolution arises clearly
from the introduction to this chapter. Similarly to the approach adopted in the previous
chapter, I will focus on the study of cosmic voids, as they were at the core of my
stacking studies and constitute a considerable amount of material and results, to
which I will compare the present theoretical study. However, all the tools that I will
discuss and develop here can be applied to overdensities as well—a task that I will
perform in the near future.
We can find a good starting point by assuming a spherical symmetry for the voids
that we are trying to describe. This constitutes a reasonable statement as voids are
found to become more and more spherical as they evolve [12], this spherical form
additionally being a stable one [28]. This assumption also works as a simplification
to the current problem, without being too strong. Considering an isolated void—i.e.
here an underdense region with spherical symmetry—my choice for its description
then turned naturally to a particular metric of General Relativity (GR), namely the
so-called Lemaître–Tolman–Bondi (LTB) metric.1 It is the most general metric of
GR for pressureless dust with spherical symmetry. The usual expression for this
metric (adopting natural units, i.e. c = 1) is the following:
ds 2 = −dt 2 + A2 (r, t)dr 2 + B 2 (r, t)dσ2 (4.1)

where dσ2 = dθ2 + sin2 θdφ2 is the standard metric on the 2-sphere. A(r, t) and
B(r, t) are two almost arbitrary functions, as the Einstein equations impose a relation
between them (see e.g. [9], for a derivation):
B,r 2
A2 (r, t) = (4.2)
1 + 2E(r )
where E(r ) is a truly arbitrary function (as per usual GR conventions, the “,r ” and
“,t ” subscripts indicate derivation with respect to the space and the time coordinates
respectively). The function B(r, t) is often renamed to R(r, t), after which the metric
is finally written as:
R,r 2
ds 2 = −dt 2 + dr 2 + R 2 (r, t)dσ2 . (4.3)
1 + 2E(r )
1 It should be noted for the sake of historical accuracy that this metric might as well be called the
Lemaître metric only, as he was the first to find this particular solution to the Einstein equations in
1933 [17] Tolman found it again independently but only in 1934 [30], and it was later investigated
by Bondi in 1947 [4]. However, to avoid a possible confusion with the FLRW metric, Tolman’s
name (and sometimes Bondi’s) is often associated to Lemaître.
4.1 Computing the iSW Effect from a Structure: A Step-by-Step Recipe 87
The evolution of the function at the core of the metric, R(r, t), is then derived from
the Einstein equations yielding:
2G M(r ) 1
R,t 2 = 2E(r ) + − R 2 (4.4)
R 3
where M(r ) is a new arbitrary function. A physical interpretation can be found for
each function involved in this metric: the central function R(r, t) acts as an analogue
to the angular distance of the FLRW metric. The arbitrary function of space M(r )
can be assimilated to a sort of “mass function” as it is directly related to the T00
element of the stress–energy tensor (see [4], for more details), while on the other
hand the E(r ) function can be interpreted as a “curvature profile”. A last arbitrary
function appears when integrating Eq. (4.4) but it can be safely fixed to any value,
as the equation is also invariant with respect to a reparametrisation of the spatial
coordinate r .
This metric has sparked a lot of interest in the literature in the last decade, as
it was considered as an alternative to the classical FLRW description of our whole
Universe (and not only a localised structure), without resorting to the addition of a
cosmological constant [2, 3, 6, 7, 10]. Indeed, if we place ourselves at the centre of
a carefully tailored over/underdensity (and the associated LTB metric), it is possible
to fit many of the current cosmological observables (CMB, BAO, supernovae, …)
without the need for invoking the “” of the standard model. However, this theory
is not without caveats, as it fails to reproduce—alone—some observables within
reasonable error margins (most notably constraints from the kinetic SZ effect, see
e.g [5] ). It constitutes nonetheless a valuable tool to explore the limitations of the
FLRW model and the possible influence on cosmological observations of a localised
inhomogeneity in our vicinity.
4.1.2 Applying the LTB Metric to Cosmic Voids
My use of the LTB metric here concerns less ambitious scales, with only one structure
considered at a time. The last missing step here—besides fixing a few initial para-
meters and a gauge degree of freedom—is the choice of the features of the void that
we want to model and study. This step is often done through the choice of a density
profile ρ(r ) at a given time in the history of the structure. Indeed, the density in the
LTB metric is related to the aforementioned function M(r ) of Sect. 4.1.1 as follows:
M,r (r )
4πρ(r ) = . (4.5)
R 2 R,r
The last degree of freedom of the LTB, the other arbitrary function E(r ), is then
fixed by the combination of the chosen initial conditions and Eq. (4.4). The choice
of this density profile is of course of no small matter, and will be discussed in more
detail further on.
This LTB description of cosmic voids has been already explored multiple times in
the past literature (among many others: [18, 20]) to study several theoretical problems
of voids, such as their evolution, stability, etc. My present work is also not the first
interested in the topic of the exact iSW effect experienced by CMB photons when
crossing cosmic voids [1, 22, 26]. It should be noted however that some of these
works may not have fully considered the issue of the background that these voids
are sitting in (often considered to be a classical FLRW background) and may have
neglected the discontinuities that arise then at the border of the structures [13, 14, 27].
In my work, I assumed that the objects considered are compensated, meaning that
these cosmic voids are surrounded by overdense “shells”—an assumption backed
up by observations (see [25] for a recent exemple). In such case, the conditions of
junction (the so-called Darmois–Israel conditions, [8, 15, 16]) between the LTB
metric describing the voids and the FLRW metric of the background can be reduced
to two simple conditions (see e.g. [32] for details):
• the void needs to be exactly compensated, i.e. its mean density should be equal to
the density of the background;
• the density of the void should be continuous as it reconnects to the background
metric.
Once these conditions are satisfied, we have all the elements in place to compute
the full history of an underdensity through the evolution of its metric, and follow
(among other variables) its density profile as a function of time. An example is shown
in Fig. 4.1 where I computed the evolution of a void with an arbitrary profile and an
initial density contrast of δ = 10−3 at its centre, from recombination (z ∗ 1100) all
the way to z = 0. As expected, the void gets deeper as time progresses, and matter
builds up at its edges, creating a gradually denser and thinner shell. This phenomenon
can be simply understood using only Newtonian physics: imagining the void as the
sum of infinitesimally thin shells of growing radius, it appears clearly that the outer
shells will expand slower than the inner ones as they contain more matter inside
of them, and therefore are slowed by a stronger gravitational pull (accordingly to
the Gauss theorem). The catching-up of the inner shells is then responsible for the
apparition of the surrounding overdensity.
4.1.3 “Lighting Up” the Voids
While simulating the history of such cosmic voids and exploiting the results could
be the subject of a dedicated study, I am more interested here in the effect of the
evolution of these structures on the CMB, i.e. on crossing photons. Tracking the
path of a photon, as well as its energy, through such voids requires to solve the
geodesic equations of GR for this photon in the corresponding LTB metric at a
chosen time/redshift. Then, we compare the energy of this photon to another one
travelling simultaneously in the FLRW background without any interruption: the
Fig. 4.1 Left panel initial density profile (normalised to the mean density in the universe) of a
compensated void at z = 1100 as a function of the LTB radial coordinate r (normalised to the void
radius). This arbitrary profile was used as an input in my LTB framework. Right panel evolution of
the density profile of the same void from z = 1100 (blue curve, almost flat) to z = 0 (red curve,
the deepest), as computed by my code solving the LTB equations
difference between the two then yields an estimate of the iSW effect caused by the
considered void.
For a radial photon that crosses the void through its diameter, the geodesic equa-
tions are fairly simple. To obtain its equation of motion, we can even start from
Eq. (4.3), and equate the line element ds 2 to zero (also eliminating the dσ2 term)
which gives the following equation:
R,r 2
dt 2 = dr 2 (4.6)
1 + 2E
and then: √
dr 1 + 2E
=± . (4.7)
dt R,r
The “±” in Eq. (4.7) depends on whether the photon is travelling towards (−) or
away (+) from the centre of the void.2 The evolution of the energy of the photon is
obtained through the geodesic equation for the time coordinate:
d 2t α β
0 dx dx
= −ν αβ (4.8)
dλ2 dλ dλ
After a few intermediate steps, we obtain the differential equation of the energy
= dt/dλ of the photon:
d R,r t
=− (4.9)
dt R,r
2 We are able to deduce this from the fact that the term after ± in Eq. (4.7) is always positive: the
angular distance R(r, t) is always a growing function of r (no shell of matter crosses another one)
hence R,r > 0.
Fig. 4.2 Relative temperature shift of a photon diametrally crossing the void of Fig. 4.1 (at some
late redshift) with respect to a background photon. The shift is plotted as a function of the photon’s
distance to the centre and was computed by solving the geodesic equations of a diametral photon
in the LTB metric
Once the complete evolution of the LTB metric—and its associated function R(r, t)—
has been computed, we have all the required elements at hand to compute the path
of the photon. As an example, I performed this computation for a photon crossing
the void described in Fig. 4.1 at some late redshift (z < 1), and illustrate in Fig. 4.2
the relative difference in its energy with respect to that of a background photon.
While the influence of the cosmic void is visible along the path of the photon (with
a maximal energy difference of ∗0.1 %), the final difference is in fact tiny (with
δT /T ∗ 10−5 − 10−6 ) as expected from the weak nature of the iSW effect.
Diametral photons give us a valuable insight on the maximum magnitude of the
iSW effect produced by the considered void (since it corresponds to the maximum
distance a photon can travel under the influence of the void evolution). However,
a complete and thorough analysis requires the full “iSW profile” of the void, i.e.
including non-diametral photons with various impact parameters, so as to allow for
a rigorous comparison with observations of the CMB and stacking results. For this
task, we invoke once more the same geodesic equations for the photons, although
now more complex due to the added degree3 of freedom. We derive then from these
a series of new equations of motion for the crossing photon:
3 Here, the angular coordinate θ: it is indeed unnecessary to use all three space coordinates thanks
to the spherical symmetry of the problem. There always exists a plane, intersecting the centre of
the metric, that contains all the motion of the photon.
10
1.0
Temperature shift (mK)

5
0.5
0.0 0
-0.5
-5
-1.0
-10
-1 0 1 2
r/rv
Fig. 4.3 Paths of 24 photons (thick coloured solid lines) crossing a void with a 130-Mpc physical
radius at z = 0.44 (N.B.: these are the properties of one of the Gr08 voids), plotted in comoving
coordinates normalised by the void comoving radius. The colour of the paths indicates the temper-
ature shift experienced by the crossing photons compared to background photons. Notice here the
scale in mK: the final iSW effect (a few µK) is imperceptible here. The large and smaller black
circles indicate respectively the radii of the void with or without its compensating overdense shell.
The black dashed lines represent the same wavefront of photons at different times of its propagation.
The red arrows indicate the direction of propagation of the photons
dr kr
= (4.10)
dt kt
dθ kθ
= t (4.11)
dt k
dk t 1 R,r t R,r r 2
=− (k ) + R,t R(k θ )2 (4.12)
dt k t 1 + 2E

dk r 1 E,r R,rr (1 + 2E)R θ 2 2R,r t r
= t − (k ) +
r 2
(k ) − k (4.13)
dt k 1 + 2E R,r R,r R,r

dk θ 2k θ R,r k r
=− R,t + t (4.14)
dt R k
with k χ = dχ/dλ (χ = t, r, θ). From this set of equations, we are able to trace
the path and follow the energy of any photon crossing the considered void, allowing
to reconstruct its iSW profile. An example of such reconstruction is illustrated in
Fig. 4.3, where I represented the motion of photons crossing a void, as well as their
temperature shift along the way (compared to a background photon evolving simul-
taneously). The “bending” of the photon trajectories is clearly visible here, as well
as their temperature shift along the way, although the final residual shift (i.e. iSW
effect) is tiny and is not visible. At this point of the study remains only the question
of the tools used for solving the aforementioned equations and getting these results,
which will be discussed in the next subsection.
4.1.4 Numerical Implementation
To perform all the required calculations of this work, I implemented the equations
described previously in two distinct codes that I wrote in the Interactive Data Lan-
guage (IDL).
The first is dedicated to the computation of the LTB metric, its evolution and all its
relevant functions, and including a precise control over the initial conditions (density
profile, etc.) as well as the choice for the value of the cosmological parameters. I
coded my own fourth-order Runge–Kutta algorithm (with an additional feature in
the form of an adaptive step finding procedure) for solving the differential equations
that govern the main function of the metric, R(r, t). Such numerical method for
solving equations implies a discretisation of R(r, t) that needs to be carefully chosen
in order to sample the function correctly over the relevant range of time and space
coordinates. An emphasis on later times is important for t as the considered voids will
often be located at low redshifts. On the other hand, a uniform sampling for r proved
to be sufficient to obtain a satisfying degree of precision: after this, the problem
becomes equivalent to solving the evolutions of independent, infinitesimal shells
of matter enclosing a fixed amount of matter. However, some extra care had to be
taken to check for any “shell-crossing” (i.e. one inner shell catching up another outer
one) that may happen for very underdense voids, that have a weak gravitational pull
on the shells. It should be noted however that in a real situation, such crossings will
probably not occur due to the presence of pressure forces not accounted for in the LTB
metric—at least when baryons are considered.4 The second code I devised solves
the geodesic equations for either a single diametral photon, or the full iSW profile.
It needs as an input the previously computed R(r, t) to perform its calculations. The
one and only free input parameter here is the redshift at which the considered void
is located.
In the interests of accuracy, I performed several consistency tests for both codes.
I first tested my codes against a “null case”, i.e. by reproducing a void with a zero-
density contrast profile that should yield results equivalent to the FLRW metric: in
terms of density evolution, as well as its effect on photon propagation. Secondly, I
studied the effect of refining the spatial grid of all the relevant functions in my codes
by a factor of 10 to check the stability of the results when considering realistic voids.
My computed R(r, t) function (the core function of the LTB metric) agrees with
its analytical expected value in the “flat” case (which is simply R(r, t) = (a(t)/a0 )r )
within a relative error of only 10−14 −10−13 . Then, choosing as a reference the values
of R(r, t) for the finest grid of r , the use of a ×10 coarser grid resulted in an average
relative change of 10−8 (10−4 at most). Regarding the computed propagation of
4 Strictly speaking, in a DM-only scenario, nothing would stop a shell from crossing another one: it
would then be probably pulled back as it will then experience an additional gravitational pull from
the shell it just passed, and so on. In reality, with the presence of baryons, such situation would
most likely lead to a virialisation process and the formation of structures at the border of the void.
In my case, it is actually more of a computing problem as the crossing of these infinitesimal shells
produces a local infinite density that crashes the numerical code.
Diametral photon (Slightly) Off-centre photon

104 104
Temperature shift (µK)
102 102
100 100
10-2 10-2
10-4 10-4
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
r/rv r/rv
Fig. 4.4 Left panel temperature shift (upper black curve) experienced by a photon crossing a void
along a diameter (with properties as in Fig. 4.3) as a function of its distance to the void centre in
terms of the comoving coordinate r normalised by the void comoving radius. This computation
is performed with a “fine” spatial grid (see text) and acts as a reference. The differences/errors
introduced by the use of a ×10 coarser grid are illustrated by the lower blue curve. As the y-axis
is in a logarithmic scale, negative parts of both curves are indicated by dotted portions. The red
arrow indicates the direction of propagation of the photon. Right panel same legend for a non-radial
photon, passing close to the void centre
photons in a “flat void”, the results agree with the predictions of the FLRW metric
within a margin of 10−10 , both for their path (straight line) and energy (→ 1 + z).
Finally, concerning the stability of the results, I used one of the Gr08 voids (that I
reproduced with the LTB metric) as a test case: the difference in the results between
the two choices of grid sampling are illustrated in Fig. 4.4. Here, the temperature shift
of crossing photons (central and off-centre) along their propagation is shown for the
finest sampling scheme. The difference in the results between the two sampling
methods (original minus finer) is then overplotted on the same graph. Generally,
the differences are below the percent level, with a maximum of a few percents in
the non-central case. An examination of Fig. 4.4 seems to indicate first that sudden
differences appear near the edges of the voids (r/r V = ±1), most likely caused by
the density profile of the void, whose derivative is not continuous at the void edge (see
Sect. 4.2.1 for more details on the choice of profile). Small differences add up along
the path of the photons, with a more pronounced effect near the centre of the void: for
diametral photons, this probably originates once again from a discontinuity at r = 0.
For off-centre photons, the culprit most likely lies in the stiffness of the differential
equation governing the propagation of the photons. Indeed, near the centre of the
voids, the angular coordinate θ found in Eqs. (4.12) through (4.11) begins to vary
very quickly, perturbing the numerical resolution of these equations.
Several of the aforementioned issues could be attenuated by introducing cus-
tomised, non-uniform grid for the radial coordinate r (with a finer sampling at the
edges for example), as well as using numerical solvers specialised in stiff differential
equations—all of which I intend to implement in the near future. For the present
work, these errors are acceptable: they affect the predicted iSW signal in a relatively
small way (1 µK and below), especially compared to the errors present in the stacking
results (most notably due to the primordial CMB contamination) that I try to repro-
duce theoretically. As seen in the previous chapter, the way the temperature profiles
are measured on the stacked images introduces errors with similar amplitude, which
are themselves several times smaller than the influence of the variance of the CMB.
4.2 The Predicted iSW Effect from Actual Structures
As I mentioned, the main goal of the present chapter is to clarify some of the most
puzzling results that I obtained when studying the impact of superstructures on the
CMB temperature. One of the most important aspects that needed attention is the
intriguing results that arose from my study of the [11] catalogue of structures, where
the features of the stacked signal (amplitude, scale, …) showed inconsistencies with
the expected signal, that is predicted in the literature so far.
Multiple interrogations arose from this study: is the apparent discrepancy with
respect to these CDM predictions a sign of new physics? Or is it only due to the
fact that the current predictions, often based on N-body simulations or linear theory
calculations (see previous chapter), are unsuited for the problem? Or even another
possibility: could it be that the data has been incorrectly interpreted as a signature of
the iSW effect?
My purpose in this section will be to answer these questions (and more) consid-
ering an approach not yet examined in the past literature, that is, not searching to
compute the expected amplitude of the iSW signal directly, but rather reproducing
first the considered structures with the help of the LTB metric, and then deducing
their iSW signature.
4.2.1 Return to the Granett et al. Catalogue
The first step to reproducing the objects of this catalogue is to consider all the
information that it provides that is relevant to the problem at hand. Examining the data
provided by the authors, I realised that the exploitable content was hardly exhaustive
and that some assumptions had to be made. For each object, the catalogue provides:
• the redshift of the lowest-density galaxy in the void,
• the volume of the void,
• three different measures of its density contrast, either calculated from the mean of
all its Voronoi cells (see Sect. 3.1.2 for details), from the mean of its underdense
cells exclusively, or finally the density contrast of the single most underdense cell.
4.2 The Predicted iSW Effect from Actual Structures 95
I first assumed (reasonably) that the lowest-density galaxy is located near the centre
of the void, and therefore considered the given redshift as the “redshift of the void”.
Then, I converted the volume of the void to an effective radius (the radius of a sphere
with the same volume) that I could work with in the LTB framework. Finally, I used
two of the density contrast values: I considered again that the most underdense cell
would be near the centre of the void, and consequently identified its density contrast
as the one that should be at the centre of the void. Lastly, I kept the mean of the
underdense cells and associated it with the expected mean density contrast of the
void but only in its underdense part, i.e. excluding the compensation (the overdense
region) that forms at the edges of the voids in the LTB model.
Due to the non-analytic nature of the solutions of the equations governing the
evolution of the LTB metric, reproducing the Gr08 voids with identical features at
the same redshift required a trial-and-error approach. One of the first requirements
was to find some adapted expression for the modelling of the voids in terms of their
initial density profile. In my work, accounting for the few constraints extractable
from the Gr08 catalogue, I chose a simple model for a void by dividing it into two
regions:
• an inner part, exclusively underdense and with a density contrast strictly increasing
along the radial coordinate, from some initial (negative) value δ0 at the centre to
δ = 0 at the border of the region, at a radius of r = R1 ,
• an outer region, modelling the initial overdense compensation and extending from
r = R1 to r = R2 .
To model these two regions, I used the following mathematical expressions respec-
tively:
δ(r ) = Ar n + B(r − R1 ) + C (4.15)

∼
δ(r ) = A (r − R1 )(r − R2 ) (4.16)
The four parameters of Eq. (4.15), i.e. A, B, C, and n, allow me to fix the values
of δ and its derivative at r = 0 and r = R1 , giving enough control to reproduce a
large spectrum of density profiles. In Eq. (4.16) however, the parameter A∼ is fixed
by the fact that we require the void to be exactly compensated. In order to mimic
the features of the actual Gr08 voids, an exploration of these initial parameters was
necessary: eventually I was able to match the available data within a error margin of
0.1–0.5 %. I also checked that the resulting profiles did not yield unrealistic features
for the voids at the initial time (z = 1,100), e.g. too large or too deep fluctuations
that would be inconsistent with the Gaussian theory of structure formation.
Two other points require a particular attention here, the first one being that there
are no constraints in the Gr08 data on the width of the (possible) compensation that
(supposedly) surrounds the voids. I chose an initial width that yielded at the later times
of interest (z = 0.4−0.7) an overdense shell roughly a third of the void radius at these
times. The absence of precise measurements of such profiles in the literature forces us
to rely on more theoretical studies (see e.g.[24] Figs. 4 and 5), that agree reasonably
with my choice. The second point of interest comes from the Gr08 data itself, more
particularly from the reported values of mean density contrasts in underdense cells.
I found these values to be somewhat small (δ ∈ [−0.48, −0.25]) which, when
coupled to the bigger reported contrasts at the centre (δ ∈ [−0.87, −0.64]), has
some implication on the shape of the density profiles. It requires them to rise quickly
as a function of the radius, whereas they are expected to be rather flat near the void
centre. Whether this indicates an issue of the data, or a unsuitable assumption on my
part, remains an open question.
As a last note, I should point out that my choice of expression for the void profiles
(with two distinct regions) results in a continuous function with, however, a discon-
tinuity in its derivative at r = R1 . This may have some small effects on the results
(as hinted at in Sect. 4.1.4) but should however be no reason for concern, as the δ(r )
function appears mainly in an integrated form through the “mass function” M(r )
(mentioned in Sect. 4.1.1) in the LTB equations.
With these final remarks, my framework is then ready to provide its first predictions
regarding the iSW effect produced by actual cosmic voids, or more precisely those
of the Gr08 catalogue.
4.2.2 First Computations and Predictions
With all the required tools at hand, I was able to reproduce all the voids of the Gr08
catalogue, with nearly identical properties (radius, depth, etc.) at their redshift of
observation. The first task that I chose to perform was to calculate the theoretical iSW
effect associated to these voids using diametral photons—a fairly fast computation. I
obtained for each void a value of the expected shift in temperature, and then averaged
these 50 values to obtain the mean iSW effect that should be observed near the centre
of the stacked image of the CMB patches corresponding to the location of these 50
voids.
The final value that I derived from this calculation is equal to δTiSW ∝
−11.44 µK. Interestingly, when compared to the temperature decrement observed
in the stacked image of the previous chapter (see e.g. the red dashed line of Fig. 3.8),
which peaks around −10 to −13 µK near the centre of the void, the predicted value
appears quite close. On this point, my findings seem to differ from the trend observed
in other theoretical results in the literature that predict smaller amplitude, as men-
tioned in Sect. 3.5. However, it would be premature to conclude anything yet as I
only considered diametral photons at this point and not the full profile and shape
of the predicted iSW effect (remember that the shape and size of the signal were
of particular importance in the stacking studies). Nonetheless, I took advantage of
the computed contributions of each of the 50 voids considered and investigated the
potential correlations between the amplitude of the predicted signal and the features
of the considered voids: here, their radius, redshift, and mean density contrast (in
their underdense region). I illustrate the results of these tests in Fig. 4.5. Overall, we
cannot find here any obvious correlations, except for a clear one between the void
sizes and their associated iSW amplitudes, as expected from our initial idea about
ISW signal (µK) 0 0
ISW signal (µK)

-10 -10
-20 -20
-30 -30
-40 -40
-50 -50
-60 -60
0 50 100 150 200 0.40 0.45 0.50 0.55 0.60 0.65 0.70
Void radius (Mpc) Void redshift
0 0
ISW signal (µK)
ISW signal (µK)

-10 -10
-20 -20
-30 -30
-40 -40
-50 -50
-60 -60
-0.50 -0.45 -0.40 -0.35 -0.30 -0.25 -0.90 -0.85 -0.80 -0.75 -0.70 -0.65 -0.60
Mean density contrast Density contrast at centre
Fig. 4.5 Total theoretical temperature shift experienced by a photon as a function of one of four
features of the reproduced Gr08 void that it crossed diametrically: the void physical radius (top left),
redshift (top right), mean density contrast (bottom left) and density contrast at the centre (bottom
right). In each plot, the red dashed line corresponds to the average temperature shift for the 50 voids
the behaviour of the iSW effect: the larger the void, the longer the photon path, and
the larger the associated iSW effect. One crucial point here is the fact that only a fifth
(around 10) of the voids contribute for the majority of the predicted signal: these
also happen to be the largest ones in the sample, as is expected. We find them again
in the iSW-redshift graph, at the highest redshifts of the sample. This indicates a
correlation between the size and distance of these voids that we already noticed and
discussed in Chap. 3 when studying the Gr08 catalogue.
To come back to these ∗10 most contributing voids, they appear to make up for
at least half of the total iSW signal of the sample, as shown in Fig. 4.6. Starting
from here, I can identify these most (theoretically) contributing voids in the Gr08
catalogue, then go back to the data, and observe the effect of ignoring these voids
in the stacking process. This supposedly should have a significant impact on the
measured signal, but as I show in Fig. 4.7 by comparing temperatures profiles, we
observe almost no modification of the resulting iSW flux. This constitutes yet another
counterintuitive fact regarding the analysis of the Gr08 voids and associated results.
I will push the analysis even further in the next subsection by considering the full
theoretical iSW profiles of these voids.
Fig. 4.6 For the reproduced 0

Gr08 voids, mean temperature
Mean ISW signal (µK)

shift (experienced by diame- -2
tral photons) as a function of
-4
the number of voids consid-
ered in the average, starting -6
from the smallest void. The
black dashed line marks the -8
point where the iSW contri-
bution of the 10 biggest and -10
most contributing voids start -12
to add to the average 0 10 20 30 40 50
Number of voids considered
Fig. 4.7 Temperature profiles

of the images resulting from 5
Temperature (µK)
the stacking of the CMB

patches (from WMAP V map)
at the location of the 50 Gr08 0
voids (black dashed curve)
and at the location of only -5
the 40 smallest voids (red
solid curve). No significant -10
difference can be reported
0 5 10 15
Angle (o)
4.2.3 Full Modelling of the iSW Profile
As mentioned earlier, the set of tools that I devised allows for the computation of
the complete radial profile of the theoretical iSW temperature shift caused by any
void. Applying these tools to the reproduced Gr08 voids, I was able to derive the
iSW profile of each of those 50 objects, as well as their mean, which should tell
us what to expect from the results of the stacking of the same voids. In Fig. 4.8, I
plotted the 50 individual profiles that I computed, the mean profile (red dashed line)
and for the sake of comparison the stacked temperature profile of the Gr08 voids
extracted in the previous chapter (blue line). There are several things to notice here:
once again, the profiles of ∗10 voids seem to stand out significantly from the rest,
as observed with the radial photons results. Interestingly enough, the mean signal
that arises from the combination of all theoretical profiles follows quite closely the
measured stacked temperature profile at the lowest angles (from 0◦ to ∗3◦ ). However
we also observe for larger scales a clear departure from the theoretical prediction. The
signal becomes even positive with the presence of a “hot shell” (already discussed
in the previous chapter) which does not find any equivalent or logical explanation
in my computations. More generally the LTB voids that I simulated do not produce
a positive iSW signal near the edges, although they all feature an overdense shell.
10
0
iSW shift (µK)

-10
-20
-30
-40
-50
-60
0 2 4 6 8 10 12
Angle (o)
Fig. 4.8 For the 50 simulated voids tailored to match the Gr08 ones, theoretical temperature shift
(thin black solid curves) as a function of the angular distance to the void centre in the sky. The red
dashed line indicates the mean of these 50 profiles. For the sake of comparison, the temperature
profile of the stacked image computed from the actual CMB data is plotted in blue (long dashes)
This may originate from the compensated nature of the void: no matter the distance
from the centre of the void, the mean density contrast of the region enclosed within
this radius is always lower or equal to 0, causing the photons to constantly experience
the effect of an underdensity, even when it is in the locally overdense shell of the
void.
This point may prove to be critical because I showed in Chap. 3 that the hot signal
surrounding the voids in the stacked image of Gr08 voids was responsible for the high
significance of this detection. If it turns out that this hot signal is not physical (in the
sense that it cannot be reproduced theoretically), it may prove to be therefore a simple
statistical fluke, forcing us to reconsider the significance of the Gr08 stacking results.
But before concluding anything, we have to remember that in reality, all these
signals come from different locations in the Universe and projected onto the celestial
sphere. This may give rise to distortions in the mean iSW signal due to the proximity
of some of these voids in the sky. To deal with this possible issue, I reconstructed the
“theoretical iSW map” associated with my 50 theoretical voids (shown in Fig. 4.9),
i.e. the temperature shift due to the iSW effect associated to each point in the sky,
accounting for all the voids present on the corresponding line of sight. Then, I used
the exact same tools that I have used in the previous chapter to perform stacking
analyses on real CMB maps: a computation of the stacked image, as well as its
temperature and photometry profile. The resulting profiles are plotted in Fig. 4.10. I
also show there the results the same analysis with the inclusion of a rescaling of the
patches prior to the final stacking, again similarly to what I did in Chap. 3. No huge
surprise here as far as the temperature profile is concerned: we still see no sign of a
positive signal, which is quite expected from the combination of the strictly negative
profiles of Fig. 4.8. However, some small changes are visible in the amplitude of
the signal: firstly, the whole profile is shifted towards negative values compared to
the mean profile of Fig. 4.8. This is illustrated by the value of the temperature shift
at the largest angles which no longer tends towards 0 (∗−4 µK instead). Secondly,
-60.0 0.2 microK
Fig. 4.9 Orthographic projection of the map of the total theoretical temperature shift caused by the
50 voids of Gr08 as simulated using my LTB framework. Only half of the map is visible (centred
on the galactic North pole) and a graticule grid has been superposed with a 45◦ step in longitude
and 30◦ in latitude
Raw stacking Rescaled stacking

0 0
Temperature (µK)
Temperature (µK)
-5 -5
-10 -10
-15 -15
0 5 10 15 0 1 2 3
Angle (o) Angle (x void radius)
Fig. 4.10 Temperature (dashed lines) and photometry (solid lines) profiles of the images obtained
from the stacking procedure performed on the map of Fig. 4.9 at the location of the Gr08 voids,
with (right panel) or without (left panel) rescaling the patches prior to stacking
the signal at the centre of the stack is itself slightly colder (by 2 or 3 µK), even
accounting for the previous shift. Both of these effects are consequences of the
proximity of the voids on the sky and the introduced overlaps when projecting their
all-negative theoretical iSW profile. We note that similar observations can be made
on the rescaled profile.
The analysis of the photometry profile yields some interesting results: the max-
imum (absolute) value of the signal is found around a 4◦ scale (around one void
radius for the rescaled profile) with an amplitude of ∗5 µK, more than twice as low
as the measured Gr08 value. This is again expected from my previous remarks on the
absence of a “hot signal” at the edges of my predicted profiles. Based on the stacking
results of Chap. 3, such signal would have a significance around 1.5 σ. Regarding the
scale of the highlighted signal, my findings appear somewhat closer to the features
of the measured photometry profile for Gr08 voids (see Fig. 3.12 for reference), as
all of my simulated voids consistently produces as signal with the same extent as the
void itself. As a consequence, these results differ from my conclusions using other
catalogues of voids, as well as from predictions from other works in the literature
(detailed in Sect. 3.5) which all found a signal with a smaller typical scale (typically
∗0.6 times the voids radii). My assumptions of spherical symmetry and the choice
of using the effective radius (computed from the void volume) to characterise my
simulations of the Gr08 voids may have played a rôle in this inconsistency. Indeed,
modelling a void this way could lead to an overestimate of its actual projected size
on the sky, especially for ellipsoidal voids aligned with the line of sight, as they
would therefore appear much more smaller than a sphere with equivalent volume.
The arbitrary size that I fixed for the width of the compensating shell surrounding my
simulated voids, which could artificially “expand” the size of the signal on the sky.
Some of these hypotheses will be explored in the next and last section of this chapter.
In the mean time, the results of the present section proved however that although all of
its features cannot be reproduced (and might even be unphysical), the actual stacked
signal of the Gr08 voids and its amplitude are not entirely irreproducible and at odds
with the CDM model, as the original authors claimed so.
4.3 Exploring the Limits of the LTB Landscape
4.3.1 Assumptions and Their Consequences
In the previous section I showed an example of application of the LTB framework

that I devised: I reproduced the features of the Gr08 voids as closely as possible, and
subsequently computed their associated iSW profiles. However, given the relatively
scarce information available about these voids, some assumptions on their nature had
to be made to be able to work with them. This includes in particular the choice of
working with spherically symmetric and compensated voids (whose radii I derived
from their volume given in the Gr08 catalogue), as well as the arbitrary choice of
the size of the compensated shell (and even its existence) surrounding each simu-
lated void. The geometry of the simulated void cannot be changed without drastic
modifications of the protocol, including a change in the choice of metric. However,
an exploration of the initial parameters of the LTB model can teach us how our
assumptions affect the computed results.
Using one of the reproduced Gr08 void as a basis, I studied the rôle of its compen-
sation in the resulting iSW effect by varying its width at the initial time of the LTB
simulation (in my case, z ∗ 1100). As illustrated by the density profiles in Fig. 4.11,
Fig. 4.11 Top panel effect 0

of a change in the width of
the void compensation shell

on the predicted iSW profile, -5
for a void with a radius of Initial compensation width
∗130 Mpc at z = 0.44. The multiplied by :
original profile (solid black x0.5
-10 x0.75
curve) is plotted against the x1.5
x2
profiles resulting from a
multiplication of the initial
-15
width of the compensation
as indicated on the graph
(coloured curves). Bottom 0 2 4 6 8 10 12
panel density contrast profiles Angle (o)
of the corresponding voids
as a function of the physical 1.0
radius, computed using my
LTB framework and plotted at 0.5
Density contrast
z = 0.44
0.0
-0.5
-1.0
0 50 100 150 200 250
Physical radius (Mpc)
these changes do not affect in any way the evolution of the inner part of the void
(i.e. its underdense part) because of its spherical symmetry. However, they evidently
change the total size of the void, hence extending the range of its associated iSW
effect accordingly (Fig. 4.11).
What are then the consequences of these remarks on our previous results and
observations? They show that the arbitrary assumption about the size of compensating
shell does have some influence on the shape and amplitude of the predicted iSW
signal. A larger (thinner) shell leads to a wider (narrower) signal, adding a longer
“tail” to the profile (or shortening it), but I found that this only translated into small
changes (less than 0.5◦ ) in the scale highlighted by aperture photometry. The size of
the shell however has the effect of noticeably modifying the amplitude of iSW signal,
with the largest compensating shells yielding the strongest signal. This phenomenon
can be intuitively understood by considering the limit case where the compensated
shell has been “stretched” infinitely far from the underdense part of the voids. From
the point of view of a crossing photon, the structure would become quasi-equivalent
to a non-compensated voids, which are know to yield a more pronounced iSW effect
than their compensated counterparts [23, 33]. It is then reasonable to believe that
4.3 Exploring the Limits of the LTB Landscape 103
Fig. 4.12 Top panel effect of

0
a change of the void density

parameters on the predicted -5
iSW profile, for a void with
a radius of ∗130 Mpc at -10
z = 0.44. The original profile
(solid black curve) is plotted -15
against two pairs of profiles
-20
(blue–cyan and orange–red):
each pair corresponds to voids -25
with the same central den-
sity contrast, but different -30
mean density. Bottom panel 0 2 4 6 8
Angle (o)
density contrast profiles of
the corresponding voids as
1.5
a function of the physical
radius, computed using my
1.0
LTB framework and plotted at
Density contrast
z = 0.44. Colour corre-

spondence dotted = blue, 0.5
dashed = cyan, dot-
dashed = orange, dot-dot- 0.0
dot-dashed = red
-0.5
-1.0
0 50 100 150
Physical radius (Mpc)
in the intermediate cases, a void with a larger and more diluted compensating shell
would produce an iSW effect with higher amplitude.
This test therefore shows that my results on the simulated Gr08 voids may indeed
depend of the initial assumptions about the void density profiles. However, on the
one hand the scale of the signal (as highlighted by aperture photometry) remains
largely unchanged (as it can be intuited from the profiles Fig. 4.11), therefore not
affecting my conclusions on the matter. On the other hand, if we were to try to
boost the predicted signal so that the photometry would reach the level associated
with the actual Gr08 voids, it would require an unrealistically large compensation
(several times the size of the underdense part of the void). This stresses once again
the peculiarities of the [11] results, notably the presence and rôle of the measured
“hot ring” in the measured signal: so far in my study, I found no way of reproducing
this very intriguing feature of the temperature profile.
At this point in the study, it would be interesting to relax some of the constraints
I put on my LTB models (i.e. the reproduction of the Gr08 voids features) and to
conduct a broader exploration of its parameters, which might give us a better insight
on the variety of obtainable iSW signal.
4.3.2 Dependences on Initial Parameters
Beside their sizes, two of the main features of the Gr08 voids that I managed to
reproduce accurately were the measured mean and central density contrasts of these
objects (as I defined them in Sect. 4.2.1). Whether simply to explore the freedom of
my LTB framework, or to consider the possibility that my assumptions on the density
were somewhat biased, I decided to study the effect of these particular parameters
on the predicted iSW profile. Some of the results of this exploration are shown in
Fig. 4.12: one constant here is that any decrease in the density (central and/or mean)
of the underdense part of the void results in an amplified iSW signal, and also a natural
increase in the density of the compensation. However, one of the most interesting
feature of these results is the presence of a small “hot” signal at the edges of the iSW
profile for the voids with a very deep density contrast (δ ∗ 0.9). This is a unique case
so far in the present chapter, most probably because the density contrasts of the Gr08
voids were not very deep (as already mentioned in Sect. 4.2.1). Indeed, this quantity
seems to be fundamental in determining whether the iSW signature of a void will
possess such positive signal, as pointed out by other works in the literature. In [31]
for example, the author also remarked that the theoretical signature of a compensated
void in the CMB was linked to its current evolutionary phase: a linear (δ ∈ 1) or
quasi-linear (δ ∝ [−0.9, −0.1]) void yields an entirely cold spot, while a non-linear
void (δ ∝ −1) shows a cold spot surrounded by a hot ring. In my case, it is likely that
the deepest voids in Fig. 4.12 is nearing the non-linear regime, hence the apparition
of the small hot signal at its edges.
One important note, considering the signal that I am trying to reproduce: the hot
feature of the predicted iSW signal is confined to the edge of the void and appears
quite narrow. Reconsidering our initial choices for the Gr08 density contrasts, it might
be possible to raise the amplitude of this positive signal at the edges by exploring
the LTB parameter space for deeper voids, but even then the resulting positive signal
would never be as large in spatial extent as the “hot ring” measured in the stacking
of the actual Gr08 voids, which is nearly as large as the mean radius of the voids
themselves. This would indeed imply a compensating shell with roughly the same
dimensions, but with a drastically lower density which would not reproduce a positive
iSW signal. As a last argument, if we suppose that the density contrasts of the Gr08
voids are indeed actually deeper that what I considered and are (at least partially) the
reason for such a high signal, this would actually be problematic with respect to the
standard cosmological picture. Indeed, as [19] showed in their work, the probability
of obtaining within the SDSS volume 50 supervoids with the required features, i.e.
the same size as the Gr08 ones (∗100 − 150 Mpc) and such low density contrast, is
negligibly small.
Adding these few remarks with the findings of the previous subsection, this explo-
ration of the LTB model demonstrates that the features of the signal measured by
Gr08 (attributed to the iSW effect) are hard to reproduce theoretically without making
some important concessions to the CDM paradigm, or even unreasonable assump-
tions on the voids features.
4.3 Exploring the Limits of the LTB Landscape 105
Fig. 4.13 Mean temperature 10

profile (solid red line) and 1 σ
contours (red dashed lines)
5
iSW shift (µK)

obtained from the stacking of 0
the Gr08 voids on the map of
Fig. 4.9 with a randomly gen-
-5
erated primordial CMB added -10
to it. The actual temperature
profile from the real CMB
-15
data is plotted in blue (long -20
dashes) -25
0 2 4 6 8 10 12
Angle (o)
4.3.3 A Statistical Oddity?
At this point in the study, I already explored quite thoroughly the LTB framework that
I devised, and may have reached its limits in terms of how closely it can reproduce
the Gr08 signal while staying within realistic assumptions.
To close this chapter, instead of looking further into the theoretical side of the
problem, it might be a good opportunity to change our approach and look back at
the data itself. Indeed, an important point that was stressed multiple times in the last
chapter was the influence of the CMB on the stacking results and the necessity to
account for its variance. We know that in the measured temperature and photometry
profiles, a portion of the signal is purely due to the random contamination by the
primordial CMB itself—a contribution that can be estimated using stacked images
at random position in the sky. A question comes then naturally to the mind: if we
account for this contamination, is it possible to reconcile the LTB predictions with
the measured signal?
The results shown in Fig. 4.13 aim at answering this question. Similarly to the
results of Fig. 4.10, I redid the stacking at the location of the voids in my simulated
LTB map (cf. Fig. 4.9), but this time adding on top of it a simulated primordial CMB
(using the latest CDM parameters from Planck). I repeated this procedure many
times and computed the temperature profile of the stacked image each time. Then I
plotted in Fig. 4.13 the mean and 1 σ limits of this distribution of profiles (in a fashion
similar to what I did back in Chap. 3), as well as the actual measured temperature
profile of the stacked Gr08 voids. We notice that the mean profile of the distribution
is naturally identical to the mean iSW signal from the 50 Gr08 voids (red dashed
curve of Fig. 4.8). If we compare now the measured profile to the contours, we find a
reasonable agreement between them. The signal lies within the 1 σ limit for most of
it, except for a small portion of the positive part of the signal which is still below the
2 σ threshold. We note for the sake of completeness that an additional error should
be accounted for: the numerical uncertainty mentioned in Sect. 4.1.4, that we safely
ignored here due to its relatively low value (∗1 µK) compared to the primordial
CMB contamination.
This result seems like a step in the right direction, tightening the agreement
between theory and measurements, although it does not solve the issue of the ∗10
Gr08 voids out of the 50 that I showed to be the most contributing ones in theory (cf.
end of Sect. 4.2.2). But it also points out another caveat concerning the whole study:
the number of voids in the Gr08 sample remains particularly low to allow us to draw
any definitive conclusions! As shown in Fig. 4.13, the variance of the CMB gives so
much latitude that a large variety of theoretical predictions would also fit reasonably
the data.
In the near future, I intend to assemble all the work and results discussed in this
chapter into an article, as well as extend the study to much larger samples of voids
such as those I used in the previous chapter, from [21] and [29]. Exploiting the latter
might prove fruitful as its catalogue contains a significantly larger amount of data on
the detected voids, including information on their density profiles.
References
1. C. Baccigalupi, L. Amendola, F. Occhionero, Imprints of primordial voids on the cosmic

microwave background. MNRAS 288, 387–396 (1997)
2. T. Biswas, R. Mansouri, A. Notari, Non-linear structure formation and ‘apparent’ acceleration:
an investigation. JCAP 12, 017 (2007)
3. T. Biswas, A. Notari, W. Valkenburg, Testing the void against cosmological data: fitting CMB,
BAO, SN and H0 . JCAP 11, 030 (2010)
4. H. Bondi, Spherically symmetrical models in general relativity. MNRAS 107, 410 (1947)
5. P. Bull, T. Clifton, P.G. Ferreira, Kinematic Sunyaev–Zel’dovich effect as a test of general radial
inhomogeneity in Lemaître–Tolman–Bondi cosmology. Phys. Rev. D 85(2), 024002 (2012)
6. M.-N. Célérier, Do we really see a cosmological constant in the supernovae data? A&A 353,
63–71 (2000)
7. C. Clarkson, M. Regis, The cosmic microwave background in an inhomogeneous universe.
JCAP 2, 013 (2011)
8. G. Darmois. Les équations de la gravitation einsteinienne. (Gauthier-Villars, US, 1927)
9. K. Enqvist, Lemaître Tolman Bondi model and accelerating expansion. Gen. Relativ. Gravit.
40, 451–466 (2008)
10. S. February, J. Larena, M. Smith, C. Clarkson, Rendering dark energy void. MNRAS 405,
2231–2242 (2010)
background due to the integrated Sachs–Wolfe effect. ApJ 683, L99–L102 (2008)
12. V. Icke, Voids and filaments. MNRAS 206, 1P–3P (1984)
13. K.T. Inoue, J. Silk, Local voids as the origin of large-angle cosmic microwave background
anomalies. I. ApJ 648, 23–30 (2006)
14. K.T. Inoue, J. Silk, Local voids as the origin of large-angle cosmic microwave background
anomalies: the effect of a cosmological constant. ApJ 664, 650–659 (2007)
15. W. Israel, Singular hypersurfaces and thin shells in general relativity. Nuovo Cimento B Ser.
44, 1–14 (1966)
16. W. Israel, Singular hypersurfaces and thin shells in general relativity. Nuovo Cimento B Ser.
48, 463–463 (1967)
17. G. Lemaître, L’Univers en expansion. Annales de la Societe Scietifique de Bruxelles 53, 51
(1933)
18. A. Meszaros, Voids in the Tolman–Bondi universe. Ap&SS 207, 5–15 (1993)
References 107
19. S. Nadathur, S. Hotchkiss, S. Sarkar, The integrated Sachs–Wolfe imprint of cosmic super-
structures: a problem for CDM. JCAP 6, 042 (2012)
20. D.W. Olson, J. Silk, Primordial inhomogeneities in the expanding universe. II—General fea-
tures of spherical models at late times. ApJ 233, 395–401 (1979)
21. D.C. Pan, M.S. Vogeley, F. Hoyle, Y.-Y. Choi, C. Park, Cosmic voids in Sloan Digital Sky
Survey Data Release 7. MNRAS 421, 926–934 (2012)
22. M. Panek, Cosmic background radiation anisotropies from cosmic structures—models based
on the Tolman solution. ApJ 388, 225–233 (1992)
23. P. Pápai, I. Szapudi, Cosmological density fluctuations on 100 Mpc scales and their ISW effect.
ApJ 725, 2078–2086 (2010)
24. S.G. Patiri, J. Betancort-Rijo, F. Prada, On an analytical framework for voids: their abundances,
density profiles and local mass functions. MNRAS 368, 1132–1144 (2006)
25. A. Pisani, G. Lavaux, P. M. Sutter, B.D. Wandelt, Real-space density profile reconstruction of
stacked voids. (2013) ArXiv:1306.3052
26. A. Rakić, S. Räsänen, D.J. Schwarz, The microwave sky and the local Rees–Sciama effect.
MNRAS 369, L27–L31 (2006)
27. N. Sakai, N. Sugiyama, J. Yokoyama, Effect of void network on cosmic microwave background
anisotropy. ApJ 510, 1–10 (1999)
28. H. Sato, K. Maeda, The expansion law of the void in the expanding universe. Progress Theoret.
Phys. 70, 119–127 (1983)
DR7 galaxy redshift surveys based on the watershed transform. ApJ 761(1), 44 (2012). ISSN:
0004–637X
30. R.C. Tolman, Effect of inhomogeneity on cosmological models. Proc. Natl. Acad. Sci. 20,
169–176 (1934)
31. S.L. Vadas. in The Signatures of Voids and the CMBR, in H. Böhringer, G.E. Morfill, J.E.
Trümper. Seventeeth Texas Symposium on Relativistic Astrophysics and Cosmology, vol. 759
(Annals of the New York Academy of Sciences, 1995), p. 710
32. R.A. Vanderveld, É.É. Flanagan, I. Wasserman, Luminosity distance in “Swiss cheese” cos-
mology with randomized voids. I. Single void size. Phys. Rev. D 78(8), 083511 (2008)
33. H. Zhan, Rees–Sciama effect and impact of foreground structures on galaxy redshifts. ApJ
740, 26 (2011)
Chapter 5
Studying Dark Matter Through the Lens
of the Reionisation
Stars are not eternal: they are known to go through a series of phases, from their births
in gaseous nebulae to their often cataclysmic ends and final transformation into stellar
remnants. It follows through that, if we go back in time far enough, there was an era
where the Universe was too young to have any stars yet, but was old and cool enough
to be transparent to photons. Without any visible light to illuminate the Universe, this
particular era was naturally nicknamed “the Dark Ages” by cosmologists, illustrating
our lack of observations of it. The physical processes that occurred through this period
are a hot topic of research, as they laid the foundations for the birth of the first stars
and thus the beginning of another era that will be the focus of this chapter: the
reionisation of the Universe. In parallel to all the primary work that I described in the
previous chapters, I also took a particular interest during my thesis to the rôle that
played in this era one less obvious actor present in the Universe, namely: the Dark
Matter, whose possible influence I will describe in the next few sections.
5.1 From the First to the Second Light of the Universe
5.1.1 The Reionisation of the Universe
As mentioned in Chap. 1, about 400,000 years after the Big Bang, the Universe’s
density decreased enough so that its temperature fell below 3,000 K, allowing ions
and electrons to (re)combine into neutral hydrogen and helium with only a negligi-
ble fraction of heavier elements. Immediately afterwards, photons decoupled from
baryons and the Universe became transparent, leaving a relic signature known as
the cosmic microwave background radiation, making it effectively the “first light”
to roam the Universe. As mentioned in the introduction of this chapter, this event
ushered the Universe into a period of darkness, the so-called Dark Ages. They ended
about 400 million years later when the first galaxies formed and started emitting

110 5 Studying Dark Matter Through the Lens of the Reionisation
ionising radiation, lighting up the Universe for the first time since recombination and
thus starting the era known as the epoch of reionisation (EoR).
Initially during the EoR, the intergalactic medium (IGM) was mostly composed
of neutral hydrogen (HI) and helium (HeI) except in regions surrounding the first
ionising objects (the so-called Strömgren spheres), such as the first generation of stars
(the so-called population III stars). As this reionisation progressed, these regions of
ionised hydrogen (HII) and helium (HeII) evolved and expanded: very schematically,
after a sufficient number of UV-radiation emitting objects formed the temperature
and the ionised fraction of the gas in the Universe increased rapidly. Eventually, all
ionised regions percolated and permeated to fill the whole Universe, thus ending the
EoR and leaving only traces of neutral matter in the IGM.
The current observational constraints (discussed in Sect. 5.1.3) allow us to roughly
situate the whole EoR in a redshift range between z ∗ 6 (according to Lyman-θ
observations) and z ∗ 15 ( see e.g. [8]). However, the details of the reionisation
history are still poorly constrained yet to be clarified: it requires a knowledge of
fundamental issues in cosmology, galaxy formation, quasars, the physics of very
metal poor stars and radiative transfer in a clumpy medium. I will describe in the
next section the basics of the mechanisms of the reionisation, and mention some
key equations and parameters used in the study of the EoR. However, as it does not
constitute my primary topic of research, I will not delve much into the details of the
era although substantial theoretical and observational efforts are currently dedicated
to understanding the physical processes that trigger the EoR and govern its evolution.
5.1.2 Physics of the Reionisation and Key Parameters
The starting point is given by a series of equations called the Saha equations, which
describe the degree of ionisation of a plasma as a function of the temperature, density,
and ionisation energies of its atoms, assuming a local thermal equilibrium. In our
case, they allow us to determine the populations of the ionisation states for the atoms
that populate the Universe during the Dark Ages and the EoR. We then consider
the balance that takes places in any situation between two competing processes: the
recombination of the ions with the free electrons in the medium, against the ionisa-
tion of atoms by photons (photoionisation). This all results in an equality between
probabilities: the one of an atom to be ionised (which in turn involve the photon den-
sity and photoionisation cross-section), and the one of an atom to capture an electron
on a certain level (which depends on the speed and density of free electrons, and the
capture cross-section). For a population of particles (e.g. atoms) at a given level k
(i.e. a given degree of ionisation), the whole equilibrium can be put in the form of
an equation:
nk ne aj = n jbj, (5.1)
j j
5.1 From the First to the Second Light of the Universe 111
where n e is the free electrons density, and n X is is the density of particles at the X level.
The a j and b j coefficients are respectively the recombination and photoionisation
coefficients and are determined from the aforementioned probabilities. The sums are
made over all the possible levels of the particle. As the lifetime of excited levels
are often very short compared to the interval between two photoionisations, we
can
safely consider that ionisations happen only from the fundamental level, i.e.
j b j = n f b f where the subscript f indicates the fundamental. If we also set
j n
a = j a j , Eq. (5.1) reduces to:
nk ne a = n f b f . (5.2)
As a convention, most of the reionisation-related works in the literature choose to

divide all the densities by the total density of hydrogen in all its forms, n H , and work
instead with “ionised fractions”:
ni
n H xk xe a = x f b f with xi = (5.3)
nH
If this equilibrium between recombination and ionisation is disrupted, we get the

equation for the evolution of the ionised fraction xk :
d xk
= n H xk xe a − x f b f (5.4)
dt
These equations govern the evolution of the population of neutral (HI) and ionised
hydrogen (HII), as well as neutral (HeI), singly (HeII) and doubly ionised helium
(HeIII). After the recombination at z ∗ 1100, they lead initially to an almost-neutral
Universe around z ∗ 100, with some residuals of ionised matter due the very low
free electron density and associated probability of capture. The key parameter whose
evolution we want to study here is the “total ionised fraction” xe , i.e. the (spatial
averaged) ratio of the density of free electrons over n H , which is equal to the sum of
the ionised fraction of hydrogen and helium—or more precisely:
xe = xHII + xHeII + 2 × xHeIII . (5.5)
where the factor 2 in front of xHeIII represents the two electrons that an atom of
helium releases when fully ionised. When the reionisation process starts later in the
era, this ionised fraction is thought to have sharply increased over a short period
of time. In current cosmological codes and models not precisely dedicated to the
study of the EoR, it is simply modelled as a step function for xe . At the beginning of
the reionisation, it starts from its residual value from recombination determined by
the equations mentioned earlier (more details about the precise calculation of these
residuals can be found in [34]). After this, xe is made to increase sharply until it
reaches a final value slightly higher that one: indeed, it corresponds to a Universe
with all the hydrogen ionised and the helium singly ionised, hence in Eq. (5.5) we
have xHeIII = 0, xHII = 1 and xHeII equal to the He/H number ratio often written as
f He .1 Thus the final value of xe is equal to 1 + f He .
This particular choice of a simple scenario is motivated by the lack of constraints
on the detailed reionisation history; however, the current consensus over the available
observations places the “middle” of reionisation (i.e. the redshift at which xe is equal
to half its maximum value) around z ∗ 11.4 [30] and its end around z ∗ 6 (from
Lyman-θ data, cf. Sect. 5.1.3.1). Lastly, at some point later in the history of the
Universe, thought to be around z ∗ 3 − 3.5, the deposition of energy in the IGM
reaches a sufficient level for the second ionisation of helium (from HeII to HeIII)
to occur, which further raises the total ionised fraction to xe = 1 + 2 f He (since
xHeIII = xHII = 1 and xHeII = 0 then). This particular phase will be discussed again
later in this chapter (see Sect. 5.3.2).
The other important variable in the history of the reionisation is the temperature
of the intergalactic gas as a function of time, as it impacts the interaction between
particles and the distribution of matter in general. Before recombination, its evo-
lution is straightforward: the matter temperature is identical to the temperature of
the photons it is coupled with, i.e. the one of a black body undergoing an adiabatic
cooling due to the expansion of the universe:
Tmat (z) = Tφ (z) = T0 (1 + z), (5.6)
with T0 the temperature of photons (i.e. the CMB) today. This is equivalent to the
following equation for the evolution of the temperature:
dTmat Tmat
= (5.7)
dz 1+z
After decoupling of the baryonic matter and photons, without any source of energy,
only two processes are competing: the adiabatic cooling of matter (alone), and its
interaction with the background photons—either cooling or heating, depending on
their relative temperature. Without delving too much into details, the equation for
the evolution of matter temperature can be then expressed as follows:
dTmat 8 aρT xe Tφ4 Tmat − Tφ 2Tmat (z)

= + . (5.8)
dz 3 m e c 1 + xe + f He H (z)(1 + z) 1+z
where ρT is the Thomson cross-section, a is the so-called radiation constant (derived

from the Stefan-Boltzmann constant), m e is the mass of the electron and c is the
speed of light. The first term of Eq. 5.8 represents the photon-baryon interaction,
while the second term corresponds to the adiabatic dilation of matter: the factor 2
that appears compared to Eq. (5.7) is due the transition of matter from a relativistic
1 This ratio is equal to Y P /(K (1 − Y P )), with K √ 3.9715 the ratio of the helium-4 atomic mass
to the hydrogen mass, and Y P the primordial helium abundance (the latest value of this parameter
was determined by [30], to be 0.24771 ± 0.00014) which yields f He ∗ 0.08.
to a non-relativistic regime after decoupling. Just as Eq. (5.4) for the evolution of xe ,
the previous equation covers the basic evolution of the matter temperature, to which
heating or cooling from others processes can be then added, which will be especially
the case during reionisation. The probing and the use of the IGM temperature in the
context of the EoR will be the central point of Sect. 5.3.
5.1.3 Observational Constraints
To date, the majority of observations related to the EoR provide weak and model
dependent constraints on the reionisation history. However, there are currently a
number of observations which impose strong constraints on the broad picture of the
reionisation.
5.1.3.1 Lyman-α Observations
One of the existing probes uses an absorption phenomenon seen in the spectra of back-
ground quasi-stellar objects (QSOs, or quasars). Releasing extraordinary amounts of
energy, QSOs are even detectable as far back as the epoch of reionisation and even a
little beyond. As their light travels through the Universe, it interacts with and excites
atoms along the line of sight, and in particular atoms of neutral hydrogen. This inter-
action produces an absorption line in the spectra of the QSO at the wavelength of
Lyman-θ transition of hydrogen (i.e. 1215.67 Å). However, due to the expansion of
the Universe, this absorption line becomes redshifted as it reaches us, by an amount
proportional to the redshift of the considered cloud of neutral hydrogen. As the light
of QSOs goes though multiple HI clouds at various redshifts, it produces a series of
absorption lines known as the Lyman-θ forest, with each individual cloud leaving
its fingerprint at a different position in the observed spectrum (see Fig. 5.1). The
large cross-section for the Lyman-θ absorption makes this effect a very powerful
technique for studying gas in the intergalactic medium over a large range of redshift:
indeed, even for low levels of neutral hydrogen, absorption is highly likely as the
associated cross-section scales as the neutral fraction xHI times 105 !
As we observe more and more distant quasars, the density of absorbing lines
in their spectra increases with redshift (see Fig. 5.1). In fact, at redshifts above
4, the density of the absorption features becomes so high that it is hard to define
them as separate absorption features. Instead, one sees only the flux in-between the
absorption minima which looks as if they were emission rather than absorption lines.
As a consequence, the best approach to exploit these spectra is to extract from them a
quantity known as the “optical depth” for absorption of Lyman-θ photons. We know
the theoretical expression of this optical depth, which involves the proper number
density of neutral hydrogen n HI along the line of sight of the quasar (for details, see
e.g. [42]). This allows us to extract information on the density of hydrogen in the
IGM, or as it is more frequently used, on the neutral fraction of hydrogen xHI defined
Fig. 5.1 Comparison of the spectra of two quasars at very different redshifts, 3C 273 at z = 0.158
and 1422+2309 at z = 3.62, both rescaled to the rest frame wavelength of their Lyman-θ emission
line—the strong and broad emission peak in the shown spectra. It is almost chopped in half by
the onset of the Lyman-θ forest in the high-redshift quasar. At low redshift, 3C 273 shows only
a handful (but distinctly more than zero) Lyman alpha absorbers. On the other hand, hundreds of
lines can be identified in the spectrum of 1422+2309. Figure borrowed from Bill Keel http://www.
astr.ua.edu/keel/agn/forest.html
as xHI = n HI /n H , where n H is the total number density of hydrogen (both neutral

and ionised).
So far, observations of Lyman-θ forests up to redshifts of ∗6 have shown the
neutral fraction to be of the order of 10−4 meaning that the Universe is highly
ionised at least below z ∗ 6. However, an important discovery was made using high
resolution spectroscopy of high redshift Sloan Digital Sky Survey (SDSS) quasars
(reported in [11, 12]). The SDSS discovered about 19 QSOs with redshifts around 6:
the spectra of some of these QSO exhibit an absence of the expected sharp Lyman-θ
spectral lines—or rather, the density of the lines is so high that we observe instead
a trough in these spectra. This feature indicates that the quasar’s light has travelled
through a large, spread out region of neutral hydrogen, and is commonly called a
Gunn-Peterson trough [15] which increases the aforementioned Lyman-θ optical
depth. From the SDSS data, it was shown for the first time that this optical depth
significantly increases around z ∗ 6.3: although the precise interpretation of this
rise is subject to debate, all authors agree that this is a sign of an increase in the
Universes neutral fraction when going towards high redshifts, marking the tail end
of the reionisation process. This explains why quasars below a certain redshift do
not show the Gunn-Peterson trough (though they do show the Lyman-alpha forest),
while quasars emitting light prior to the end of reionisation will feature a Gunn-
Peterson trough. To summarize, the main conclusion from the Lyman-θ optical depth
measurements is that it provides us with a lower limit for the end of the EoR, with a
highly ionised Universe at redshifts below 6, and an increase of its neutral fraction
at about z = 6.3.
5.1.3.2 Thomson Scattering and the CMB
Another way to constrain the reionisation era comes from the study of CMB
anisotropies. It is known that the Universe has indeed recombined and became largely
neutral at z √ 1100. If recombination had been absent or substantially incomplete,
the resulting high density of free electrons would imply that photons could not escape
Thomson scattering until the density of the Universe dropped much further. This scat-
tering would inevitably destroy the correlations at subhorizon angular scales seen in
the CMB data (see e.g. [37]). However, a similar scattering must have occurred due
to the reionisation of the Universe and the reintroduction of free electrons into the
IGM.
In order to calculate the effect of reionisation on CMB photons, a function often
defined is the visibility function:
g(π) = −δ̇ e−δ (π) , (5.9)

where π → dt/a is the conformal time, a is the scale factor of the Universe
and δ̇ is the derivative of the optical depthwith respect to π. The optical depth for
π π
Thomson scattering is given by δ (π) = − π 0 dπ δ̇ = π 0 dπ a(π) n e ρT , where π0
is the present time and n e is the electron density. The visibility function gives the
probability density that a photon had scattered out of the line of sight between π
and π + dπ. The influence of reionisation on the CMB temperature fluctuations is
obtained by integrating Eq. 5.9 along each line of sight to estimate the temperature
fluctuation suppression due to the scattering. The suppression probability turns out
to be roughly proportional to 1 − e−δ [40]. Since the amount of suppression in the
measured power spectrum is small, the optical depth for Thomson scattering must be
small too. The left hand panel in Fig. 5.2 shows the impact on the CMB temperature
fluctuations power spectrum of increasing the value of δ . The right hand panel shows
the reionisation history of the Universe assumed in the left panel.
While temperature anisotropies on small scales are erased, polarisation
anisotropies of the CMB are actually introduced because of reionisation, most pre-
cisely in the E modes (see [21], and references therein). By looking at the CMB
anisotropies observed, and comparing with what they would look like had reionisa-
tion not taken place, the electron column density at the time of reionisation can be
determined. With this, the age of the universe when reionisation occurred can then
be calculated, provided a model is assumed for the history of xe (z). Furthermore,
the process of reionisation is the only one that can produce a large scale E mode
(a) (b)
Fig. 5.2 Left hand panel a Influence of reionisation on the CMB temperature angular power spec-
trum. Reionisation damps anisotropy power under the horizon (diffusion length) at last scattering.
The models here are fully ionised out to a reionisation redshift z i . Notice that with high optical depth,
fluctuations at intermediate scales are regenerated as the fully ionised (long-dashed) model shows.
This figure is taken from Wayne Hu’s Ph.D. thesis [19]. Right panel b shows the assumed reion-
isation history used, with a uniform and sudden reionisation model at the reionisation redshift z i
polarisation signal, a “bump” in the power spectrum at low multipoles. Such large
scale correlation in the E-mode has been measured by the WMAP team (see [2], for
the final results of the mission), and its existence, as well as its characteristics give
a strong indication that the Universe became ionised around redshift z ∗ 10. The
argument in essence is mostly geometric, namely it has to do with the scale of the
E-mode power spectrum as well as the line of sight distance to the onset of the reion-
isation front along a given direction (see [40], more details). Unfortunately however,
the large cosmic variance at large scales limits the amount of possible information
one can extract from the large scale bump shape, and more detailed constraints on
reionisation are hard to obtain (see e.g. [22]).
In the end, the current best constraints on reionisation from CMB data come from
the measurement of the optical depth δ for the Thomson scattering mentioned earlier,
found to be around 0.089 + 0.012/−0.014 (found in [30], from the combination of
WMAP and Planck constraints). This value can in turn be used to constraint the
global reionisation history through the integral:
z dec
cH−1 dz
δ= ρT n e 0 , (5.10)
(1 + z) m (1 + z)3 + σ
0
where z dec is the decoupling redshift. This formula can be applied for the optical
depth along each line of sight but allows also an estimation for the mean electron
density, i.e. the mean reionisation history of the Universe. An important point to
notice here is that, in order to turn δ into a measurement of the reionisation redshift,
one needs a model for xe as a function of redshift. Hence, one has to be careful
when using the reionisation redshift given by CMB papers as in most cases a gradual
(step-like) reionisation is assumed (see previous Sect. 5.1.2). Sudden reionisation
gives a one to one correspondence between the measured optical depth and the
reionisation redshift: here, the WMAP measurement for the optical depth δ implies
z i = 11.0 ± 1.4. However, sudden reionisation is very unlikely and most models
predict a more gradual evolution of the electron density as a function of redshift (see
the related perspective discussed in Chap. 6).
5.1.3.3 The 21 cm Line as a Probe of the EoR
Even with the quasar data roughly in agreement with the CMB anisotropy data,
there are still a number of questions, especially concerning the energy sources of
reionisation and the effects on, and rôle of, structure formation during reionisation.
The last probe that I will mention in this section, the 21-cm line in hydrogen, is
potentially a mean of studying this period, as well as the Dark Ages that preceded
reionisation. The 21-cm line occurs in neutral hydrogen, due to differences in the
ground energy between the parallel and anti-parallel spin states of the electron and
proton. This transition is forbidden, meaning it occurs extremely rarely. The transition
is also highly temperature dependent, meaning that as objects form in the Dark Ages
and emit Lyman-alpha photons that are absorbed and re-emitted by surrounding
neutral hydrogen, it will produce a 21-cm line signal in that hydrogen (for reference,
see [42]). By studying 21-cm line emission, it will be possible to learn more about the
early structures that formed. While there are currently no results, there are a a number
of telescopes dedicated to measure this faint radiation. In the short term, these consist
of: The Low Frequency Array (LOFAR), the Murchison Widefield Array (MWA),
Precision Array to Probe Epoch of Reionization (PAPER) and Giant Metrewave
Radio Telescope (GMRT), while, on a somewhat longer time scales the Square
Kilometer Array (SKA). One of the most challenging tasks in studying the EoR is to
extract and identify the cosmological signal from the data and interpret it correctly.
This is because the detectable signal in the frequency range relevant to the EoR is
composed of a number of components: the cosmological EoR signal, extragalactic
and galactic foregrounds, ionospheric distortions, instrumental response and noise,
each with its own physical origin and statistical properties.
5.2 Contribution of the Decay and Annihilation

of Dark Matter to the EoR
The EoR is a watershed epoch in the history of the Universe. Prior to it, the formation
and evolution of structure was dominated by Dark Matter alone, while baryonic
matter played a marginal rôle. The EoR marks the transition to an era in which the
rôle of cosmic gas in the formation and evolution of structure became prominent
and, on small scales, even dominant. However, to this day we still do not know
precisely what are the sources of energies responsible for the reionisation of the
Universe. Quasars, population III stars and dwarf early galaxies are very often cited
as candidates, but other sources may have played a rôle in the process. Despite its
“secondary rôle” in the history of the EoR, the Dark Matter may potentially have
played a rôle, during the Dark Ages and the earliest stages of the reionisation epoch,
as well as after the EoR ended. This idea will be the topic of the two sections, present
and Sect. 5.3, of this chapter, with a particular focus on what we could learn on DM
thanks to the study of the EoR.
5.2.1 The Dark Matter Mystery
The nature of Dark Matter (DM) is one of the crucial open questions in cosmology.
Originally hypothesized to account for discrepancies between calculations of the
mass of galaxies from velocity measurements, Dark Matter’s existence has been
since inferred from a myriad of other gravitational effects on visible matter and by the
gravitational lensing of background radiation. Originally hypothesized to account for
discrepancies between the mass of large astronomical objects determined from their
gravitational effects and the mass calculated from the visible matter they contain
(stars, gas, and dust). It was first postulated by Jan Oort in 1932 to account for
the orbital velocities of stars in the Milky Way and by Fritz Zwicky in 1933 to
account for evidence of missing mass in the orbital velocities of galaxies in clusters.
Although Dark Matter is now estimated to constitute 84.5 % of the total matter in
the universe and 26.8 % of the total energy content of the Universe (according to the
latest results of [29], in the standard model of cosmology) any direct detection has yet
to be confirmed (cf. the DAMA/LIBRA experiment, COGENT, XENON, CDMS,
EDELWEISS, etc.) mainly due to its absence of light emission or absorption at any
significant level, and its supposed weak interaction with ordinary matter.
In the standard model of cosmology, DM particles are defined as “cold” particles,
because of their negligible free-streaming length (i.e. the length below which Dark
Matter fluctuations are suppressed). The most famous alternative model to CDM is
called warm Dark Matter (WDM), where particles have longer free-streaming length.
Recently, there has been a lot of interest regained in these WDM theories, as they
could alleviate some of the caveats of the CDM theories (see e.g. [3], for a review),
such as the so-called substructure crisis.
At present, there is no definitive evidence which allows us to exclude one of the
two scenarios and even the properties (mass, lifetime, etc.) of CDM and/or WDM
particles are substantially unknown. From an observational point of view, one of the
most direct ways to detect DM particles, and maybe distinguish between the existing
CDM and WDM models, is represented by particle decays and annihilation. Indeed,
depending on the considered DM model, a fraction of DM particles is expected to
decay or annihilate, the rate of these processes generally depending on their density,
5.2 Contribution of the Decay and Annihilation of Dark Matter to the EoR 119
mass, and cross-section. These involve the emission of photons (although not directly,
but through a cascade of products) at wavelengths depending in the particle mass, so
that it would be theoretically possible to distinguish DM models using observations
of these photons. However, at the moment, constraints on the radiation emitted by
such particle decays and annihilation are loose and no definite detection has been
made yet, although a large number of experiments are now involved in the present and
future of this search, such as the Fermi Gamma-ray Space Telescope, the High Energy
Stereoscopic System (HESS), the Cherenkov Telescope Array, the Alpha Magnetic
Spectrometer (AMS), the General Antiparticle Spectrometer (GAPS) and PAMELA
experiment. For a review of recent results on the topic, see e.g. [27, 31, 36].
5.2.2 Dark Matter and Reionisation
Of particular interest to me here is that it has also been pointed out in the literature
(see [24], and references therein) that photons eventually due to particle decays or
annihilations can be sources of partial early reionisation and heating of the inter-
galactic medium: these ideas will be at the centre of the work that I will present in
the rest of this chapter, revolving around the possible influence of DM on the reioni-
sation history and parameters, as well as the potential constraining power of the EoR
on various DM models and properties. In this work, I did not pretend to present a
complete overview of DM candidates. Instead, I wanted to give a global description
of the effects of standard DM candidates, aiming to point out the differences among
the considered DM particles and their relations to the cosmic reionisation. There-
fore I considered only two types of Dark Matter among the most popular models,
each time making the assumption that the DM is composed of one single species of
particles:
• First, a light kind of Dark Matter (abbreviated LDM, [4, 18]) whose mass does not
exceed 100 MeV, or else its disintegration products would contain easily detectable
muons that are incompatible with observations. The axino is a representative of
such light Dark Matter particles.
• Another type of Dark Matter, with a mass greater that 30 GeV, that I will call
“heavy Dark Matter” (hDM) and that includes particles such as the often quoted
neutralinos.
As mentioned earlier, these DM particles can potentially contribute to the reionisation
of the Universe through two processes: spontaneous decay and self-annihilation.
Both produce new elements whose energy will contribute to three main processes:
the ionisation, the excitation and the heating of the IGM. In my work, I considered
two different scenarii of energy injection due to Dark Matter:
• The decay of the LDM particles, based on the assumption that if these are light
enough, their lifetime will be short enough for a non-vanishing part of its population
to decay over the history of the Universe—but long enough for Dark Matter to
be still present today. According to existing models, the DM could decay into a
variety of products (photons, electron-positron pairs, etc) which can then in turn
inject energy into the IGM (e.g. by photoionisation) and therefore contribute to
the reionisation of the Universe. If we simplify the problem by assuming that the
only products are photons, then the photon emission rate is simply given by an
exponential law:
dn n0
= e(t0 −t (z))/δ , (5.11)
dt δ
with n 0 the current density of the considered DM particles, δ their mean lifetime,
t0 and t (z) the times passed from the Big Bang to now and to the redshift z
respectively. Following [24], I defined δ as a function of the particle mass m L D M
as δ = 4 × 1026 s (m L D M /MeV)−1 , meaning that larger DM particles naturally
have a lower lifetime.
• The annihilation of hDM particles with their anti-particles (possibly themselves) ;
we rule out a possible decay of these particles, because if they could, their lifetime
would then be very short due to their massive nature and therefore they could not
be a viable DM candidate. Assuming the same hypothesis as for the LDM decay
products, the photon emission rate is given by:
dn
= n 20 (1 + z)3 < ρv > C, (5.12)
dt
where <ρv> is the annihilation cross-section—typically of the order of
10−24 cm3 · s−1 in optimistic cases, 10−26 for more conservative models—averaged
over the temperature. The variable C stands for the “clumping factor”, defined as
C = ∼n 2 /∼n2 with n being the local matter density. It characterises the tendency
of matter to aggregate, which boosts the number of interactions, hence the poten-
tial annihilations and the number of produced photons. In the literature, this factor
is often chosen to be equal to unity, due to a lack of detailed understanding of the
dynamics of DM and the processes involved in its clumping. Moreover, on global
scales in the young Universe, this assumption should not be too far from reality.
In both of these scenarii, the injection rate of energy per baryon is written simply as:
dn E φ
λD M = (5.13)
dt n b
where E φ is the energy of the emitted photons (in theory, half the mass of the DM
particle in case of decay, or the whole mass for annihilations) and n b the current
number density of baryons today. Part of this released energy will ionise the hydrogen
and helium atoms, while another part will heat the IGM. In order to estimate these
different fractions, we use the approximation of [7] (itself based on the work of [35]).
Qualitatively this approximation states that initially, when the IGM is still neutral
(xe = 0), the energy is distributed equally into the heating and the two ionisation
processes (of HI and HeI). Conversely, when the universe is fully ionised (xe = 1),
all of the energy goes into the heating of the IGM; and in between, the different
fractions evolve linearly with xe . Once I set up the framework, the next step for me
was to compute the evolution of the key parameters of the EoR in the context of these
scenarii in order to have a first assessment of the effect of DM on the history of the
IGM.
5.2.3 Recombination Code and Customisation
In order to test and exploit the different models of DM and associated reionisa-
tion histories, my main tool was the numerical RECFAST created by [34] whose
main purpose is to compute the history of the IGM through the ionised fractions of
hydrogen and helium, as well as the temperature of the IGM. Initially, the code does
not account for any reionisation model and computes the aforementioned quanti-
ties starting from a time before the recombination (around z ∗ 8000) and goes to
z = 0. I therefore modified the code to suit my needs by including additional terms
to the evolution equations, corresponding to sources of reionisation. More precisely,
I modified the three equations that govern the evolution of xHII , xHeII and Tmat (see
Sect. 5.1.2 for definitions of these parameters), adding the following terms:

d xHII λ D M 1 − xHII
−α = E (5.14)
dz E th, H 3 (1 + f He )

d xHeII λ D M 1 − xHeII
−α = E (5.15)
dz E th, H e 3 (1 + f He )

dTmat 2 λ D M 1 + 2 xHII + f He (1 + 2 xHeII )
−α = E, (5.16)
dz 3 kB 3 (1 + f He )
where E th, H = 13.6 eV (E th, H e = 24.6 eV) is the ionisation energy of hydrogen
(helium) atoms, k B the Boltzmann constant and E → [H (z)(1 + z)]−1 . Aside from
these three equations, I also added the computation of another important variable of
the EoR, the optical depth δ which is one of the most accessible observable in the
data, that I already mentioned and defined in Sect. 5.1.3.2.
After applying this modification to the RECFAST code, I first tried and succeeded
in recovering the results of a previous and similar study by [24] for a few cases of
DM models. My results are presented in Fig. 5.3.
Three masses of LDM are considered for decay: according to the results, these
particles may be considered as a potential source of significant reionisation, especially
for the highest considered mass (10 MeV). Indeed, the ionised fraction xe (see Eq. 5.5)
reaches a value of 0.8 even without any other source of reionisation: we remember
that the maximum value for xe is around 1.2 (and corresponds to xH = xHe = 1)
which really makes the contribution of LDM non-negligible. We note however that
the value of xe is ten times lower at a redshift of z = 6, a time when reionisation
is supposed to be already finished according to current observations: this acts as a
105 105
Tmat (K) 104 104
Tmat (K)
103 103
102 102
101 101
10-1 10-1
10-2 10-2
-3
10 10-3
τ
τ
-4
10 10-4
10-5 10-5
0.1000 0.1000
xe
xe
0.0100 0.0100
0.0010 0.0010
0.0001 0.0001
1 10 100 1000 1 10 100 1000
(1+z) (1+z)
Fig. 5.3 Ionised fraction (bottom panels), Thomson optical depth (central panels) and matter
temperature (upper panels) as a function of redshift as computed by my modified RECFAST code.
Left panels Results for the decaying LDM of masses 1 (thick dotted line), 5 (dashed) and 10 MeV
(solid). The thin solid line represents the case without any reionisation source. Right panels Results
for the annihilation of hDM ∼ρ v = 2 × 10−26 (thick dashed line) and 10−24 cm3 s−1 (solid). In
both the cases the particle mass is 100 GeV. The thin solid line is the same as in the left panels
reminder for the need of other, astrophysical sources of reionisation. Naturally, the
effect of the LDM decay fades rapidly when considering lower masses: its impact on
the IGM temperature is nonetheless important even for these cases, with a temperature
significantly higher as soon as z = 20 − 30. On the other end, the annihilation of
hDM is clearly insufficient to provide a significant reionisation of the Universe or
heating of the IGM at z = 0, no matter the annihilation cross-section considered.
However, the effect of hDM on the ionised fraction, although weak, occurs much
earlier in the history of the Universe than for the LDM—as soon as z = 600 − 700—
due to the simple fact that the annihilation rate depends on the square of the matter
density (which is much higher at earlier redshifts) whereas decay rates are only
linear in density. This difference has repercussions on the Thomson optical depth,
as it is (roughly) related to the integral of the ionised fraction: indeed, we observe
in Fig. 5.3 that the most influencing model of LDM reaches only a optical depth of
0.01 at z = 1,000 while the optimistic model of hDM gets close 0.05 (this has to be
put into perspective of the measured value of 0.08).
In any case, these first results allow us to see that DM particles, even in the most
optimistic cases, cannot have reionised the Universe by themselves. However, can we
determine if they contributed to this process significantly and especially, if they had
a measurable impact? Tentative answers to these questions lie in the next section,
where I performed a more detailed analysis of the rôle of DM within the known
paradigm of the EoR.
5.2.4 Additional In-depth Analysis
As discussed in Sect. 5.1.3, we know from observation that the reionisation of the
Universe occurred around a redshift of z = 10 and was mostly finished before
z = 6.5. In order to account for this knowledge in my study, I added in my version of
RECFAST a classic (cf. end of Sect. 5.1.3.2) arbitrary, step-like reionisation, starting
around z = 20 from whichever value xe reached at that time, and ending around
z = 6. Thus, by adding simultaneously the effect of DM particles, one can observe
its relative impact on the reionisation history. At the same time, an original part of my
work is that I also decided to include a similarly shaped increase of the temperature of
the IGM up to a few 104 K—a value motivated by the knowledge of the temperature
within Strömgren spheres and the thermal history of the IGM (cf. [39]). However,
unlike the ionised fraction that stays at its maximum value, the temperature of the
IGM keeps on evolving after the end of reionisation (if only because of the adiabatic
cooling). An interesting fact: although it remains quite arbitrary, my simple step-like
model for the temperature proved to be quite pertinent, as it happened to match quite
accurately more sophisticated models that I found in the literature and aimed at a
more physical modelling of the sources of heating in the EoR (see again [39]).
To further improve the relevance and accuracy of my code, I also included an
additional sophistication to it: I considered a departure from the simplistic assumption
about the clumping factor C involved in the DM annihilation process. Simply fixing it
to unity is reasonable only for high redshifts, when the Universe is still homogeneous
(z ∝ 10), but it becomes invalid for low redshifts with the onset of the virialisation
of structures and the apparition of the “cosmic web”. Therefore, I devised instead
a more coherent formulation of this factor based on two observations: the fact that
the apparition of the first structures can be dated below z = 60, and the mean matter
density of the Universe which is known to evolve proportionally to (1 + z)3 . The
resulting expression that I chose is the following:
3
61
C =1 for z > 60, C= for 0 < z < 60, (5.17)
1+z
a form similar to tho one used in e.g. [9]. Although not very complex, this new
expression is still physically consistent: in the linear regime at high redshifts, the
clumping factor stays equal or very close to 1, and increases accordingly with the
apparition of structures over the course of the Universe. For the sake of completeness,
I wish to mention that I also explored two other sophistications, in the form of two
additional sources of reionisation: namely the annihilation of LDM annihilation
and the excitation of hydrogen by Lyman-θ photons. However, I found that both
processes have in the end a negligible impact on each key observable (the ionised
fraction, optical depth and IGM temperature).
Let us now carry on to the results of this extended analysis: I will focus here each
time on the most optimistic models of DM in terms of impact on the reionisation
(witnessed in the previous section). We start with the decay of 10 MeV LDM particles,
shown in Fig. 5.4.
Here again, LDM alone is not enough to fully reionise the Universe; what is worse
is that the impact of the DM is completely overshadowed by the sudden, “astrophys-
ical” reionisation, except for a redshift range between 10 and 100. Although small,
this difference (of the order of 0.01) may imply a surplus of free electrons in the IGM
which may have had consequences on the evolution of the intergalactic gas, as the
ionised fraction is still multiplied by a factor of 100 around z = 20. Concerning the
optical depth results, we note first that the observable value of δ (obtained through
CMB studies) corresponds roughly to the plateau observed in the middle panel of
Fig. 5.4. We see that the impact of DM here is quite negligible, with only tenuous
differences at high redshift when the step-like reionisation is added. In contrast to this
mildly exciting results, the evolution of the IGM temperature is much more affected
by DM particles: I will not go into much detail into the implication of this impact as
it will be the focus of the last section of this chapter. Still, we note that DM particles
alone can yield enough heating to bring the IGM temperature to a level similar to (or
even higher than) the step-like heating. We can also see very distinct phases in the
temperature evolution for the full (step + LDM) scenario, with a cooling of the IGM
until z = 100 where it is overcome by the DM heating, then the step corresponding
to the ignition of astrophysical sources, and finally after z = 6, the competition
between the adiabatic cooling and the DM heating again.
The situation is somewhat different for the results with the annihilation of hDM
particles: its impact on the history of the ionised fraction remains weak, but its
integrated effect on the optical depth and its plateau is more significant than in the
LDM case (see Fig. 5.5). It could even reach detectable levels, although we have to
remember that I considered here an optimistic annihilation cross-section. Even more
interesting is the impact on the IGM temperature, which is boosted compared to
the previous section thanks to the change in the clumping factor. This has the effect
of bringing it to similar levels as the lone step-like reionisation and the heating of
LDM decays, making it a suitable candidate for further exploration, as we will see
in Sect. 5.3.
The various effects on the optical depth that we have observed, although not very
important, should have nonetheless an effect on the visibility function (mentioned in
Sect. 5.1.3.2) and therefore on the CMB and its fluctuations, as we will see briefly
on the next and last subsection.
5.2.5 Effects on the CMB Spectrum
The first effect that is dependent on the reionisation history is a deviation of the
spectral energy distribution of the CMB from a perfect black body spectrum. This
deviation is caused by the sum of interactions between free electrons and CMB
photons along their paths; it can be quantified by the Compton parameter y, defined as:
Fig. 5.4 Ionised fraction (top), optical depth (middle) and IGM temperature as a function of redshift,
in a scenario without any reionisation (thin black line), with a step-like reionisation only (thick
black), and the same two scenarii with the addition of the decay of LDM particles (thin and thick
red)
Fig. 5.5 Same legend as Fig. 5.4 for the optical depth and IGM temperature in the context of the
addition of annihilation of hDM particles

z=1100
ρT k B xe (z)n b (z)
y= (Tmat − Tφ )dz (5.18)
m e c2 (1 + z)H (z)
z=0
using the same notations as previous equations. The current constraints on this para-
meter are only in the form of an upper bound found to be around 2.5 × 10−5 by
the FIRAS instrument of the COBE satellite [25]. This is a useful parameter for
eliminating models that may predict a too large modification of the CMB spectrum.
In my case, among all the reionisation models that I tested, the maximum value of y
Fig. 5.6 Temperature-temperature (top panel), polarisation-polarisation (central panel) and

temperature-polarisation (bottom panel) spectra. Thick lines indicate the CMB spectrum derived
assuming the WMAP 3 value of the Thomson optical depth δ e ∗ 0.09 and a step-like reionisation
model (consistent with the 3-year WMAP data); thin lines indicate the CMB spectrum derived
assuming no reionisation. Dashed (solid) lines indicate the CMB spectrum obtained (without) tak-
ing into account the decays of 10 MeV LDM particles. The two thick lines, solid and dashed, appear
superimposed, because the contribution of decaying particles (the dashed line) is completely hidden
by the stronger effect of a sudden reionisation. Open circles in all the panels indicate the WMAP 3
data
that I found was equal to 10−7 . Although it does not validate these models, it is still
a viability criterium for these theories.
As mentioned earlier, the reionisation history has a direct effect on the power
spectrum of the CMB anisotropies. Some of our models did have a somewhat signif-
icant impact on the ionisation fraction, with respect to the value due to relic electrons,
already at high redshift. This fact should therefore leave some imprint on the CMB
spectrum. To check whether these effects are measurable, I simulated the expected
CMB spectrum in the case we take into account DM decays. This has been done by
implementing our modified version of RECFAST in the cosmological code CAMB
that I already mentioned in the previous chapters.
Figure 5.6 shows the temperature-temperature (TT), temperature-polarisation
(TE) and polarisation-polarisation (EE) spectra, in the case of 10 MeV decaying
LDM (i.e. the particle for which we obtained the maximum contribution to the
reionisation among the considered cases), compared with the results of the WMAP
3 mission. In all cases, the contribution due to DM decays alone is negligible:
their effect is tiny and concentrated at low multipoles. Knowing that there existed
other sources of reionisation besides DM decays, their influence on the CMB is

indistinguishable from a scenario without any DM.
Because 10 MeV LDM particles produce the highest ionisation fraction among
the considered models, the effects on the CMB spectra due to other species of DM
particles will be far more negligible. One the other end, we remember from the
previous section the promises shown by the impact of DM particles on the IGM
temperature: the clear influence of DM there made me reconsider my approach. In
the light of this results, it would be wise to tackle the problem from a different
angle, maybe using the temperature of the IGM as a way to detect and constrain the
presence of DM through it heating of the medium: the purpose of the last section of
this chapter.
5.3 Constraints on Dark Matter from the Temperature

of the IGM
The potential of the IGM temperature for constraining DM models is very promising,
as shown in the results that I mentioned previously. However, no matter how encour-
aging it looks, this particular method—just as any other one—will require robust and
precise measurements of the IGM temperature to compare our predictions to.
5.3.1 Probing the IGM Temperature Evolution
Due to its low density, the intergalactic medium cooling time is long and retains some
memory of when and how it was last heated. Hence, measuring the IGM temperature
at a certain redshift (typically up to z ∗ 6) allows us to reconstruct, under certain
assumptions, its thermal history up to the reionisation phase where the IGM has been
substantially heated.
Such measurements have been carried out by a number of authors: they are
obtained using high resolution data of the forest of intergalactic Lyman θ absorption
lines observed in the spectra of bright quasars. Indeed, the widths of these Lyman θ
absorption lines are sensitive to the temperature of the IGM through a combination of
thermal (Doppler) broadening and pressure (Jeans) smoothing of the underlying gas
distribution (e.g. [16, 28]), in addition to broadening from peculiar motions and the
Hubble flow (e.g. [17, 38]). Consequently, using a statistics sensitive to the thermal
broadening kernel combined with an accurate model for measurement calibration
(typically high-resolution hydrodynamical simulations of the IGM), various authors
have placed constraints on the thermal evolution of the IGM. In my work, I used the
recent measurements of the IGM temperature by [1], [6] and [14] alongside earlier,
less constraining IGM temperature data [32, 33, 41].
5.3 Constraints on Dark Matter from the Temperature of the IGM 129
5.3.2 Refining the Reionisation Model
In the previous section of this chapter, I illustrated how the impact of Dark Matter
can be clearly seen in the history of the IGM temperature. Since measurements of
this temperature are available, a simple comparison of this data with the evolution
predicted by my previous calculations should be enough to validate or exclude the
various DM models considered. However, I was faced with an additional difficulty:
indeed, the aforementioned measurements only constrain the IGM temperature in
a range of redshifts comprised between z ∗ 0 and z ∗ 6, with most of the data
in z ∗ [2, 4.5]. The reason why this is problematic is that this period overlaps a
particular phase that is poorly constrained, namely the second ionisation of helium
(briefly mentioned in Sect. 5.1.2).
Indeed, in the current picture for the evolution of the baryons, there are thought
to be two reionisation events which turned the neutral gas in the IGM into an com-
pletely ionised medium. The first reionisation event that I already discussed happened
where neutral hydrogen and neutral helium were ionised by early galaxies. The sec-
ond reionisation event is expected instead to have been driven by quasars at lower
redshifts, which produce a hard ionising spectrum that can reionise singly ionised
helium into HeIII by z ∗ 3 [13, 23, 26].
As a consequence, we expect naturally that photo-heating during both of these
reionisation events leaves a “footprint” on the thermal state of the IGM, while I
only considered and implemented the first one in the numerical code that I described
in the previous section. To rectify this caveat, I added another “step-like” increase
in temperature and ionised fraction, similar to the the previous reionisation event.
However, as this second event is much less constrained, I did not fix the characteristics
of this step, and ended up with three new free parameters: the redshift and width of
the HeII ionisation, and the amplitude of the IGM temperature boost. In order to limit
the number of free parameters in my model, I fixed the amplitude (at ∗2.5 × 104
K) of the first temperature boost associated to the reionisation of hydrogen, justified
by related works in the literature (cf. [20, 39]). It should be noted that adding the
HeII temperature step does result in a more rigorous modelling of the reionisation
history, but this additional freedom may introduce some degeneracies with the DM
parameters.
As a last sophistication, I also included an additional heating term in the evo-
lution of the IGM temperature after the first reionisation event: in addition to the
processes already considered—the adiabatic cooling of matter, its interaction with
CMB photons, and the impact of DM—I added the contribution of the photo-heating
due to the ionising background of sources (see [5], for details), as the population of
objects that appeared during the Dark Ages and the EoR (the first stars, galaxies and
quasars) are logically expected to have had an influence during the entire duration
of the reionisation.
5.3.3 Setting Up the Framework
With this completed model, I focused back on the use of data for constraining DM
models. Exploring the possible DM models “by hand” and comparing each time their
prediction with the available data would have been very fastidious, time-consuming
and not representative. After compiling all the measurements of the IGM temperature,
I therefore devised a protocol based on an MCMC analysis of the parameter space
of my model, which I limited to five parameters:
• Two parameters for the Dark Matter, considering the same models as in the previous
section: the mass m L D M and the life-time δ L D M for Light DM decay models, and
the mass m h D M and annihilation cross-section ρh D M for annihilating Heavy DM
models. I ran several MC chains and derived constraints for both models separately;
• Three parameters for the less-constrained second reionisation of helium, already
He and width νz He , and the
mentioned in the previous section: its redshift z reion reion
amplitude of the IGM temperature boost νTreion .He
The rest of the parameters of the reionisation parameters are fixed, namely here the
redshift of the HI/HeI ionisation (taken to be z = 10.4, the WMAP7 best-fit value)
and the amplitude of its temperature boost chosen to be νT = 2.5 × 104 K. Finally,
since the heating induced by DM prior to the HI/HeI reionisation is washed out by
astrophysical sources (and inaccessible by our datasets), I thus switched on that DM
contribution only for redshift lower than the end of the first reionisation (z ∗ 10).
5.3.4 Results and Discussion
I present here the results of this analysis, although it should be kept in mind that
these are still work in progress. Using the combined data of several MCMC runs, I
show here first in Fig. 5.7 the evolution of the IGM temperature as predicted by the
LDM and hDM models of Sect. 5.2.2 with the best-fit parameters of each model, i.e.
the sets of parameters that give the best agreement between the model and the data
measurements—also included in the graph.
Once again, the contrast between the impact of both DM models can be seen
here: the hDM annihilations tend to have a more pronounced influence at earlier
times when the matter density—and consequently the annihilation rate—was higher.
On the other hand, their impact at low redshifts is much less visible than in the case
of LDM decays, whose rate does not depend on density and is not affected by the
expansion of the Universe. For intermediate redshifts, around the second reionisation
of helium, the two models are not significantly different: as a consequence, the most
extreme points of data (in terms of redshift) will have a particular importance in our
analysis.
When comparing the β2 of both best-fit models with respect to the data points,
we find a slight advantage for the LDM model (both models have a β2 around 60):
Fig. 5.7 Evolution of the IGM temperature as predicted by the best fit models for the LDM decay
(solid line) and for hDM annihilation (dashed) when compared to the data points of various work
of the literature (crosses of various colours)
the relative matching of the two with respect to the data is indeed balanced as the
hDM model fits more tightly the points at the lowest redshifts, while the LDM model
performs better at higher z. However, it should be noted that both DM models are
actually quite bad fits to the data (having reduced β2 approaching 2), although the
models themselves are not entirely to blame. Indeed, a quick look at the data points
used here shows some contradictions between the values of the IGM temperature
obtained by the various studies, over the same range of redshifts (e.g. the difference
between the [41] and [1] points). These discrepancies may be due to a difference in
the quality of the source data used to derive these temperatures, as the most recent
works [1, 6, 14] seem to have more coherent and tighter constraints. We can actually
already witness in Fig. 5.7 that the best fits are strongly driven by these points,
especially those of [1]. A possibility for a future revision of this work would be to
discard some of the oldest datasets to keep only the (seemingly) most robust ones.
Another important test would be to compare the previous β2 (from the DM models)
to the value of the β2 yielded by a simple model without any influence of DM,2 to see
how it fares against the data and to show if the inclusion of DM is really necessary.
2Of course, this simple model would have two less parameters (the DM ones), i.e. two less degrees
of freedom that will have to be accounted for when comparing it to the β2 of the DM models.
4 6 8
26
log10(τLDM)
25.95
25.9
4 6 8 25.9 25.95 26
3.8 3.8
zreion
He
3.6 3.6
3.4 3.4
4 6 8 25.9 25.95 26 3.4 3.6 3.8
1.5 1.5 1.5

Δ zreion
He
1 1 1
0.5 0.5 0.5
0 0 0
4 6 8 25.9 25.95 26 3.4 3.6 3.8 0 1
3
Δ Treion / 10
6 6 6 6
He
4 4 4 4
2 2 2 2
4 6 8 25.9 25.95 26 3.4 3.6 3.8 0 1 2 4 6
zHe Δ zHe Δ THe / 10
3
log10(mLDM) log10(τLDM)
reion reion reion
Fig. 5.8 For LDM decay models, marginalized posterior distributions and 2-D contour plots show-
ing the ranges of and correlations between the parameters and the 68 and 95 % confidence limits.
The units of m L D M , δ L D M and νTreion
He are respectively eV, s, and K
Moving on to the rest of the analysis, the results of the MCMC exploration for
the LDM and hDM models are illustrated respectively in Figs. 5.8 and 5.9 (again,
these are preliminary results of a work in progress). In both models of DM, we
see that the mass of the considered DM particles remains largely unconstrained;
however in the Light DM decaying case, the temperature data seem to favour a tightly
defined life-time around 1025.93 s, corresponding to ∗2.7 × 109 Gyrs. According
to the same model, the reionisation of HeII is found to have preferably occurred
He
at z reion √ 3.58 and rapidly (νz reion
He √ 1). Annihilating DM models seem more
compatible with an extended HeII reionisation νz reion
He √ 2.2 occurring slightly later
(z reion √ 3.48). However, the annihilation cross-section of hDM particles is loosely

He
constrained. Interestingly, in both cases, the energy deposited by DM is sufficient to

He √
allow a temperature increase associated to HeII reionisation photoheating, νTreion
6 − 7 × 10 K, slightly lower than that of theoretical estimates (√ 1 − 4 × 104 K,
3
e.g. [5]) but consistent with current estimates [1]. As mentioned earlier, additional
work is currently underway to assess the need of a non-zero DM contribution, as
only a comparison of likelihoods with and without DM will determine whether
present data prefer or not the addition of this non-astrophysical source of energy.
6 8 10
)
10 HDM
−22
log (σ
−24
−26
6 8 10 −26 −24 −22
3.6 3.6
reion
zHe
3.5 3.5
3.4 3.4
6 8 10 −26 −24 −22 3.4 3.5 3.6
2.4 2.4 2.4

reion
2.2 2.2 2.2

Δ zHe
2 2 2
1.8 1.8 1.8
6 8 10 −26 −24 −22 3.4 3.5 3.6 2 2.5
Δ THe / 103
8 8 8 8
reion
7 7 7 7
6 6 6 6
6 8 10 −26 −24 −22 3.4 3.5 3.6 2 2.5 6 7 8
zHe Δ zHe Δ THe / 10
3
log (m ) log (σ )
10 HDM 10 HDM reion reion reion
Fig. 5.9 Similar content and legend as Fig. 5.8 in the case of hDM annihilation models. The units
of m h D M and ρh D M are respectively eV and cm3 .s1
Furthermore, I am also considering and running tests with a more sophisticated (and
redshift dependent) modelling of the fraction of the DM energy that goes into heating
(inspired by Evoli [10], whose author I am in contact with). As an example, the energy
deposition in a fully ionised medium is far from trivial and instantaneous, as the mean
free path of photon becomes noticeably longer.
References
1. G.D. Becker, J.S. Bolton, M.G. Haehnelt, W.L.W. Sargent, Detection of extended He II reion-
ization in the temperature evolution of the intergalactic medium. Mon. Not. R. Astron. Soc.
410, 1096–1112 (2011)
2. C.L. Bennett et al., Nine-year Wilkinson microwave anisotropy probe (WMAP) observations:
final maps and results. Astrophys. J. Suppl. Ser. 208, 20 (2013)
3. P.L. Biermann, H.J. de Vega, N.G. Sanchez, Towards the Chalonge Meudon workshop 2013.
Highlights and conclusions of the Chalonge Meudon workshop 2012: warm dark matter galaxy
formation in agreement with observations. ArXiv:1305.7452 (2013)
4. C. Bœhm, P. Fayet, R. Schaeffer, Constraining dark matter candidates from structure formation.
Phys. Lett. B 518, 8–14 (2001)
5. J.S. Bolton, S.P. Oh, S.R. Furlanetto, Photoheating and the fate of hard photons during the
reionization of HeII by quasars. Mon. Not. R. Astron. Soc. 395, 736–752 (2009)
6. J.S. Bolton et al., Improved measurements of the intergalactic medium temperature around
quasars: possible evidence for the initial stages of He II reionization at z ∗ 6. Mon. Not. R.
Astron. Soc. 419, 2880–2892 (2012)
7. X. Chen, M. Kamionkowski, Particle decays during the cosmic dark ages. Phys. Rev. D 70(4),
043502 (2004)
8. T.R. Choudhury, A. Ferrara, Updating reionization scenarios after recent data. Mon. Not. R.
Astron. Soc. 371, L55–L59 (2006)
9. L. Chuzhoy, Impact of dark matter annihilation on the high-redshift intergalactic medium.
Astronphys. J. 679, L65–L68 (2008)
10. C. Evoli, M. Valdés, A. Ferrara, N. Yoshida, Energy deposition by weakly interacting massive
particles: a comprehensive study. Mon. Not. R. Astron. Soc. 422, 420–433 (2012)
11. X. Fan, et al, A survey of z>5.7 quasars in the sloan digital sky survey. II. Discovery of three
additional quasars at z>6. Astron. J. 125, 1649–1659 (2003)
12. X. Fan et al., A survey of z>5.7 quasars in the sloan digital sky survey. IV. Discovery of seven
additional quasars. Astron. J. 131, 1203–1209 (2006)
13. S.R. Furlanetto, S.P. Oh, The history and morphology of helium reionization. Astrophys. J.
681, 1–17 (2008)
14. A. Garzilli, J.S. Bolton, T.-S. Kim, S. Leach, M. Viel, The intergalactic medium thermal history
at redshift z = 1.7-3.2 from the Lyθ forest: a comparison of measurements using wavelets and
the flux distribution. Mon. Not. R. Astron. Soc. 424, 1723–1736 (2012)
15. J.E. Gunn, B.A. Peterson, On the density of neutral hydrogen in intergalactic space. Astrophys.
J. 142, 1633–1641 (1965)
16. M.G. Haehnelt, M. Steinmetz, Probing the thermal history of the intergalactic medium with
Lyalpha absorption lines. Mon. Not. R. Astron. Soc. 298, L21–L24 (1998)
17. L. Hernquist, N. Katz, D.H. Weinberg, J. Miralda-Escudé, The Lyman-Alpha forest in the cold
dark matter model. Astrophys. J. 457, L51 (1996)
18. D. Hooper, L.-T. Wang, Possible evidence for axino dark matter in the galactic bulge. Phys.
Rev. D 70(6), 063506 (2004)
19. W.T. Hu, Wandering in the background: a cosmic microwave background explorer. Ph.D. thesis
(University Of California, Berkeley, 1995)
20. L. Hui, Z. Haiman, The thermal memory of reionization history. Astrophys. J. 596, 9–18 (2003)
21. M. Kaplinghat et al., Probing the reionization history of the universe using the cosmic
microwave background polarization. Astrophys. J. 583, 24–32 (2003)
22. A. Lewis, J. Weller, R. Battye, The cosmic microwave background and the ionization history
of the Universe. Mon. Not. R. Astron. Soc. 373, 561–570 (2006)
23. P. Madau, F. Haardt, and M.J. Rees, Radiative transfer in a clumpy universe. III. The nature of
cosmological ionizing sources. Astrophys. J. 514, 648–659 (1999)
24. M. Mapelli, A. Ferrara, E. Pierpaoli, Impact of dark matter decays and annihilations on reion-
ization. Mon. Not. R. Astron. Soc. 369, 1719–1724 (2006)
25. J.C. Mather et al., Measurement of the cosmic microwave background spectrum by the COBE
FIRAS instrument. Astrophys. J. 420, 439–444 (1994)
26. M. McQuinn et al., He II reionization and its effect on the intergalactic medium. Astrophys. J.
694, 842–866 (2009)
27. C. Muñoz, Indirect dark matter searches and models. Nucl. Instrum. Methods Phys. Res. A
692, 13–19 (2012)
28. M.S. Peeples, D.H. Weinberg, R. Davé, M.A. Fardal, N. Katz, Pressure support versus thermal
broadening in the Lyman θ forest - I. Effects of the equation of state on longitudinal structure.
Mon. Not. R. Astron. Soc. 404, 1281–1294 (2010)
29. Planck Collaboration. Planck 2013 results. I. Overview of products and scientific results
(2013a). ArXiv:1303.5062
30. Planck Collaboration. Planck 2013 results. XVI. Cosmological, parameters (2013b)
ArXiv:1303.5076
References 135
31. T.A. Porter, R.P. Johnson, P.W. Graham, Dark matter searches with astroparticle data. Ann.
Rev. Astron. Astrophys. 49, 155–194 (2011)
32. M. Ricotti, N.Y. Gnedin, J.M. Shull, The evolution of the effective equation of state of the
intergalactic medium. Astrophys. J. 534, 41–56 (2000)
33. J. Schaye, T. Theuns, M. Rauch, G. Efstathiou, W.L.W. Sargent, The thermal history of the
intergalactic medium. Mon. Not. R. Astron. Soc. 318, 817–826 (2000)
34. S. Seager, D.D. Sasselov, D. Scott, A new calculation of the recombination epoch. Astrophys.
J. 523, L1–L5 (1999)
35. J.M. Shull, M.E. van Steenberg, X-ray secondary heating and ionization in quasar emission-line
clouds. Astrophys. J. 298, 268–274 (1985)
36. L.E. Strigari, Galactic searches for dark matter. Phys. Rep. 531, 1–88 (2013)
37. N. Sugiyama, Cosmic background anisotropies in cold dark matter cosmology. Astrophys. J.
Suppl. 100, 281 (1995)
38. T. Theuns, J. Schaye, M.G. Haehnelt, Broadening of QSO Lyθ forest absorbers. Mon. Not. R.
Astron. Soc. 315, 600–610 (2000)
39. P. Valageas, J. Silk, The reheating and reionization history of the universe. Astron. Astrophys.
347, 1–20 (1999)
40. M. Zaldarriaga, Polarization of the microwave background in reionized models. Phys. Rev. D
55, 1822–1829 (1997)
41. M. Zaldarriaga, L. Hui, M. Tegmark, Constraints from the Lyθ forest power spectrum. Astro-
phys. J. 557, 519–526 (2001)
42. S. Zaroubi, The epoch of reionization. In T. Wiklind, B. Mobasher, V. Bromm (eds) Astro-
physics and Space Science Library, vol 396, p 45 (2013)
Chapter 6
Conclusions and Perspectives
In this thesis, I explored the potential of new and innovative probes of the nature of
Dark Energy, by combining large scale structure data and a type of late time CMB
anisotropies known as the integrated Sachs-Wolfe effect. This peculiar feature can
arise only in special circumstances, and in particular in the presence of Dark Energy.
In Chap. 2, I have described the cross-correlation technique between the CMB
and tracers of matter commonly used to evidence the iSW effect, and shown how the
measurement of this effect is a method to constrain cosmology in general and Dark
Energy in particular. I insisted on the crucial rôle of the features of the galaxy surveys
usually considered as such tracers. Starting from there I presented a series of tools
aimed at fully exploiting the constraining power of the future large scale surveys. I
extensively tested my framework on simulated data whose analysis yielded promising
results. In the same chapter, I investigated the cross-correlation between the CMB
and the Cosmic Infrared Background, an alternative and original tracer of the matter
distribution. I studied the detectability of this correlation under various observational
situations, after calculating the theoretical angular power spectrum of the CMB-CIB
cross-correlation at several frequencies and for different instruments. Developing an
advanced S/N analysis which included the main sources of noise, both instrumental
and astrophysical, and all their possible correlations, I pointed out the most promising
frequencies, and obtained very encouraging results (with significance as high as 7 σ
in the most ideal case, ∼5 σ for more realistic scenarii) especially when compared
to the current constraints from classical galaxy-CMB correlations. The results of
this work will be valuable in the forthcoming years of analysis and exploitation of
the Planck data. I actually recently started applying my formalism to CIB maps
extracted from the Planck data at several frequencies and cleaned using data from
the Parkes Galactic All-Sky Survey [2]. These resulting CIB maps currently cover
approximatively 10 % of the sky and will be improved in the upcoming months
by the dedicated group of the Planck Collaboration. I already managed to perform
some very simple, model-independent correlation tests (in real and harmonic space)
with the CMB from Planck. I found a positive correlation each time but with only a
∼1 σ significance, assessed by confronting the results to correlations between these

138 6 Conclusions and Perspectives
CIB maps and 1,000 random Gaussian realisations of the CMB. Although it may
seem disappointing at first, many further tests have yet to be performed on this CIB
data, as it will be progressively improved in the near future. Techniques need to
be developed to reduce the contamination from dust, in particular to account for the
spatial variations of its properties (e.g. its spectral index). When the data will be clean
enough, an interesting prospect would be to use the multiple observed frequencies
of the CIB to reconstruct the contributions from different redshift bands, in order
to obtain several decorrelated CIB maps corresponding to these redshift slices. The
resulting independent CIB maps could then be individually correlated with the CMB,
with the promise of a signal with even higher significance.
I took quite a different approach to the iSW effect and its detection in Chap. 3,
namely through the impact that individual structures in the Universe should have on
the CMB temperature. I first revisited the stacking of structures in CMB maps of a
previous work by Granett et al. [1] which claimed a very significant signal, at odds
with σCDMpredictions. I thereby devised a new complete protocol for the stacking
procedure, from a careful choice of maps to a rigorous estimation of the significance.
Although I did not find any strong difference, my tools allowed me to discover some
peculiar features in these results. I then extended the analysis to two more recent and
consequent void catalogues, one of whom [3] hinted at significant signals (although
not as strong as the previous Gr08 results) with a preferred scale in the signal (∼half
the mean size of the voids used in the stacking). The amplitude and scale of this signal
were more coherent with our expectations about the iSW effect compared to the Gr08
results and their peculiarities. As an important note, I also discussed extensively the
robustness of the detection and the validity of the catalogues of structures used.
One of the main conclusions of my work is that, in regard to all the results that I
obtained and the objects and catalogues studied, it would be premature to either claim
a detection of an iSW-like signal and/or claim an oddity with respect to σCDM. The
oddity in question may very well be found in the data itself, especially in the case of
the Gr08 catalogue and its odd inconsistencies. In parallel, as member of the Core
Team of the HFI Planck Collaboration, I undertook the task of reiterating the stacking
analysis on the Gr08 and other catalogues with the new Planck data, and I led the
corresponding section of the paper released in March 2013 along with the other
cosmological papers. In the upcoming years, I intend to extend the exploration of the
stacking methods using the upcoming catalogues that will be derived from the next
generation of galaxy surveys; not only for the iSW detection, but also for further
cosmological studies, such as Alcock-Paczyński test, another promising probe of
DE, based on the measurement of void shapes in surveys. As another objective, I
plan to personally undertake all the steps of a stacking study, and in particular the
identification of structures inside a survey using cluster and void finding algorithms,
which would give me a unique insight on some of the interrogations that arose during
my previous analyses, as well as on possible improvements.
Directly along the lines of some of the results and questions that arose from the
stacking analysis, I decided that it is necessary to reconsider and revise the intuitions
one can have about the iSW effect and its relation to the properties of the structures
that generate it. Requiring an accurate prediction of the impact of such objects on the
6 Conclusions and Perspectives 139
CMB temperature, I worked extensively on a theoretical framework (presented in

Chap. 4) to predict the thermal impact in the CMB of individual structures, based on
the modelling of each structure as a spherical object with the use of the LTB metric.
I was able to reproduce all the physical characteristics of the Gr08 structures and
derive the temperature profile of their full (not only linear) theoretical iSW effect.
Although I found that the predicted mean shift in temperature was comparable to the
level measured with the Gr08 void sample, I also discovered several discrepancies,
most notably concerning the relative contribution of each void in the sample and the
impossibility to reproduce a peculiar hot ring observed in the stacking of the actual
Gr08 voids. I also derived many interesting results from a more general exploration of
my LTB framework, from which I gained a valuable insight on the relations between
the void characteristics and its resulting iSW signature. My framework can be now
applied to any possible case and be used from now on as a benchmark for computing
the expected iSW signal produced by a structure. As a work in progress, I intend to
refine and extend this analysis very soon to the other catalogues of voids mentioned
in Chap. 3. In the longer term, I plan to go further by relaxing the assumptions of
spherical symmetry and consider axisymmetric dust solutions, possibly including
pressure effects. This would be a necessary step towards accurate predictions for
existing structures in the Universe, and their use for constraining cosmological mod-
els, as it allows for a much better account of the geometries (often ellipsoidal) of
these objects.
In parallel to all these works dedicated to the study of the iSW effect, I also turned
to another type of cosmological information encoded in the large scale structure of
the Universe. In the fifth and final chapter of my thesis, I discussed a part of my
Ph.D. work that I dedicated to the study of the epoch of reionisation—or rather, to
the study of Dark Matter trough its potential influence on the EoR. Indeed, I inves-
tigated the possible contribution from the annihilation and decay of DM particles to
the reionisation history, studying several models of DM particles of various masses
and cross-sections. I focused mainly on their impacts on the temperature of the IGM
before, during and after reionisation, taking great care in studying and modelling
the energy deposition of these particles, considering all the possible mechanisms
and associated efficiencies. I then compiled all the recent measurements of the IGM
temperature and confronted them to my theoretical predictions. Using MCMC sim-
ulations, I computed constraints on the properties of several chosen flagship? DM
models allowed by the data. The preliminary results of this work in progress are
promising, both in themselves and also when combined with other DM constraints
where they might break degeneracies. This is however not the only work that I am
currently conducting on the EoR. In the context of CMB experiments, only a few cos-
mological parameters are extracted from the data with the use of current numerical
codes. These assume a fairly simple step-like modelling of the reionisation history
(mentioned in Chap. 5) that I found to be too crude, especially with the advent of
more precise CMB experiments. I therefore set out to develop a new, more complete
parametrisation of the reionisation history, which can account for a broader range
of scenarii: as a notable example, a two-stage, asymmetrical reionisation which is
relevant in terms of the physics of the processes involved. I developed and I started
140 6 Conclusions and Perspectives
applying my new parametrisation in the context of the Planck Reionisation Working

Group; the results of this ongoing work will contribute to the next release of Planck
papers scheduled in the next few months.
To conclude, the work presented in my thesis can be summarised as a series of very
encouraging successes to better probe and understand some of the most fundamental
properties of the Universe through the observations of its large scale structure and
the secondary anisotropies created in the cosmic microwave background.
References
1. B.R. Granett, M.C. Neyrinck, I. Szapudi, An imprint of superstructures on the microwave back-
ground due to the integrated Sachs-Wolfe effect. ApJ 683, L99–L102 (2008)
2. P.M.W. Kalberla, et al., GASS: the Parkes Galactic all-sky survey. II. Stray-radiation correction
and second data release. A&A 521, A17 (2010)
DR7 galaxy redshift surveys based on the watershed transform. ApJ 761(1), 44 (2012). ISSN
0004–637X

The Large Scale Structures: A Window On The Dark Components of The Universe

Uploaded by

Copyright:

Available Formats

The Large Scale Structures: A Window On The Dark Components of The Universe

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Large Scale Structures: A Window On The Dark Components of The Universe

Uploaded by

Copyright:

Available Formats

Springer Theses

Recognizing Outstanding Ph.D. Research

The Large Scale

Recognizing Outstanding Ph.D. Research

For further volumes:

Theses are accepted into the series by invited nomination only

• They must be written in good English.

The Large Scale Structures

Doctoral Thesis accepted by

ISSN 2190-5053 ISSN 2190-5061 (electronic)

Library of Congress Control Number: 2014940758

Springer International Publishing Switzerland 2014

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Neil deGrasse Tyson

Modern cosmology is a young science. In less than a century of existence, it has

significance; (ii) he devised a complete stacking strategy to detect the iSW

Orsay Cedex, April 2014 Prof. Mathieu Langer

Dark Energy is one of the great mysteries of modern cosmology. It is an unknown

1. B.R. Granett, M.C. Neyrinck, I. Szapudi, An imprint of superstructures on the microwave

an enjoyable place to live (and work, occasionally!). My thanks to the ‘‘elders’’

And here I am now, at the end of these acknowledgments, where I should

Thanks to you all.

Chilly-Mazarin, January 2014 Stéphane Ilić

1 Introduction: The Ingredients of a Good Cosmological Probe . . . . 1

2 Unravelling the iSW Effect Through the Matter Distribution. . . . . 15

3 The Impact of Identified Superstructures in the CMB . . . . . . . . . . 51

4 Towards a Full Modelling of the iSW Effect . . . . . . . .......... 85

5 Studying Dark Matter Through the Lens of the Reionisation. . . . . 109

6 Conclusions and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

LCDM Lambda Cold Dark Matter

S. Ilić, The Large Scale Structures, Springer Theses, 1

1.1 The Cosmic Microwave Background

standard model of cosmology. Among these we find: the measurement of a highly

1.2 Large Scale Structures and Perturbation Theory

The perturbations φ corresponds to the Newtonian potential, and  is the perturba-

1.3 The (Current) Ruler of the Universe: Dark Energy

1.3.1 Models of Dark Energy

1.3.2 Probes of the Dark Energy

It is important to underline that the supernovae measurements mentioned earlier are

1.4 Shedding Light on the Dark Energy: The Integrated

As mentioned earlier, a particular type of temperature anisotropies is found in the

1.5 A Glimpse at the Dark Components Through

2.1 CMB Cross-Correlation with Tracers of Matter

2.1.1 The Principle

S. Ilić, The Large Scale Structures, Springer Theses, 15

2.1.2 Theoretical Framework

X (X = g, T ) are the spherical harmonic coefficients of the field resulting

C T g (τ) → ∼νT (n̂1 )νg (n̂2 ) (2.7)

2.1.3 Application to Detectability and Test Case

Cross spectrum S/N by multipole Cumulative S/N from l=2 to l max

2.1.4 State-of-the-Art of the iSW Detection

simultaneously in a decrease of the theoretical signal and an increase of its variance.

2.2 Optimising the Cross-Correlation for iSW Detection

2.2.1 Context and Tools

Being entirely theoretical in nature, this study required me to simulate realistic

2.2.2 Simulations, Covariance Estimation and Parameter

Selection function Cumulative S/N

4 With N typically of the order of ten thousands.

The perturbations φ corresponds to the Newtonian potential, and is the perturba-

C T g (τ) → ∼νT (n̂1 )νg (n̂2 ) (2.7)

tightening of the distribution: the standard deviation decreases from ∼ 0.120