Les amas de galaxies sont les plus grandes structures gravitationnellement liées de l'Univers. Ils sont communément séparés en trois catégories, basées sur la distribution du gaz intra-amas. Ce gaz peut être très concentré vers le centre de l'amas, il peut être réparti dans l'amas de manière plutôt uniforme, ou encore il peut avoir une distribution légèrement piquée vers le centre dans un cas intermédiaire. Une autre distinction entre les trois catégories est l'interaction entre le trou noir supermassif se trouvant au centre de l'amas de galaxies et le gaz intra-amas environnant. Dans le cas de la première catégorie, lorsque le gaz est concentré au centre de l'amas, le trou noir est dit "actif". Il produit alors des jets, qui à leur tour injectent de l'énergie dans le gaz intra-amas sous forme d'ondes sonores, d'ondes de choc et de turbulence. Les amas de galaxies offrent donc une opportunité très intéressante pour étudier ce mécanisme d'échange d'énergie. Afin de mieux caractériser ces processus, il est essentiel d'avoir des méthodes robustes pour classifier les amas de galaxies selon les trois catégories. Il existe plusieurs propriétés pouvant être utilisées comme métriques de classification, mais celles-ci ne sont pas toujours en accord les unes avec les autres. Ces propriétés ont été étudiées pour des petits échantillons d'amas de galaxies, analysés de manière individuelle avec des méthodes traditionnelles. Cependant, avec le développement de puissants instruments d'observation tels que eROSITA, on s'attend à obtenir des échantillons contenant environ 100 000 amas de galaxies. Étant donné la taille de ces ensemble de données, il devient nécessaire d'avoir un moyen rapide, efficace et automatique pour les traiter. On a donc recours à l'apprentissage automatique pour accélérer l'analyse. Ce mémoire présente une analyse des propriétés du gaz intra-amas avec des méthodes d'apprentissage automatique. On se sert des simulations cosmologiques IllustrisTNG pour obtenir des images en rayons X d'amas de galaxies, à partir desquelles on construit notre ensemble de données. On s'intéresse à cinq propriétés du gaz intra-amas contenu dans les amas de galaxies, qui sont couramment utilisées comme métriques de classification: le temps de refroidissement central, la densité électronique centrale, l'excès d'entropie centrale, le paramètre de concentration de la brillance de surface et le paramètre de courbure du profil de densité. On explore les relations entre ces différentes métriques, puis on implémente un réseau de neurones qui vise à prédire leur valeur à partir d'une image en rayons X d'un amas de galaxies. Notre réseau atteint une pourcentage d'erreur moyen de 1.8% pour les prédictions de la métrique la plus performante, c'est-à-dire le temps de refroidissement central. Ensuite, afin d'estimer les incertitudes sur les résultats obtenus, on effectue une analyse probabiliste de nos prédictions à l'aide de la méthode de l'inférence sans vraisemblance. On utilise également une méthode de partitionnement de données qui rassemble les images en rayons-X en trois groupes distincts; on constate que ce regroupement corrèle fortement avec la division des mêmes images en utilisant le paramètre de concentration comme métrique de classification. L'ensemble de ce travail permet de conclure que le temps de refroidissement central et la concentration sont les métriques se prêtant le mieux à une analyse avec des méthodes d'apprentissage automatique, ainsi que de mettre en place les outils qui serviront à caractériser les futurs échantillons d'amas de galaxies. / Galaxy clusters are the largest gravitationally bound structures of the universe. They are commonly divided into three categories, based on the distribution of the intracluster gas. In one case, the gas is strongly concentrated towards the center of the cluster. In another case, it is rather uniformly dispersed through the cluster. In a third intermediate case, the distribution is slightly peaked towards the center. The three categories also differ by the interaction between the gas and the supermassive black hole located at the center of the cluster. In the first category, the black hole is said to be 'active' and it produces jets that heat up the intracluster gas through shock waves, sound waves and turbulence. The feedback mechanism from the black hole is not entirely understood, and galaxy clusters offer a valuable opportunity to study this energy transfer mechanism in more detail. Numerous properties can serve as classification metrics, but they are not always consistent with one another. Moreover, traditional methods used to extract those properties are time-consuming and have only been applied to small samples. With the advent of powerful X-ray observatories such as eROSITA, we expect to obtain large galaxy clusters datasets (~100 000). Given the size of the datasets and the number of parameters to consider, machine learning methods are needed to accelerate the data processing. This thesis presents an analysis of intracluster gas properties with machine learning techniques. We use the galaxy clusters from the IllustrisTNG cosmological simulations to create the X-ray images that make up our dataset. We study five properties of the hot gas in galaxy clusters that are commonly used as classification metrics; the central cooling time, the central electron density, the central entropy excess, the concentration of the surface brightness and the cuspiness parameter, which represents the slope of the density profile. We explore the correlations between the different metrics, and implement a neural network that predicts their values from an X-ray image of a galaxy cluster. The network achieves a mean percentage error of 1.8% on the central cooling time predictions, making it the best-performing metric. In order to get uncertainty estimates, we perform a probabilistic analysis of the network predictions using simulation-based inference. We also use a clustering approach that groups the X-ray images into three separate groups; we note that those groups are consistent with classification based on the concentration parameter. Our results show that the central cooling time and the concentration are the metrics that lend themselves the best to a machine learning analysis of galaxy cluster images. This project aims to lay the groundwork for characterizing future galaxy cluster observations.

Made available in DSpace on 2014-12-17T15:15:00Z (GMT). No. of bitstreams: 1 CalistratoSCN.pdf: 479133 bytes, checksum: 26f581ebc6e3cf1ced0922a5e989272c (MD5) Previous issue date: 2001-11-07 / Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior / The present work investigates some consequences that arise from the use of a modifed lagrangean for the eletromagnetic feld in two diferent contexts: a spatially homogeneous and isotropic universe whose dynamics is driven by a magnetic feld plus a cosmological parameter A, and the problem of a static and charged point mass (charged black hole). In the cosmological case, three diferent general solutions were derived. The first, with a null cosmological parameter A, generalizes a particular solution obtained by Novello et al [gr-qc/9806076]. The second one admits a constant A and the third one allows A to be a time-dependent parameter that sustains a constant magnetic feld. The first two solutions are non-singular and exhibit in ationary periods. The third case studied shows an in ationary dynamics except for a short period of time. As for the problem of a charged point mass, the solutions of the Einstein-Maxwell equations are obtained and compared with the standard Reissner-Nordstrom solution. Contrary to what happens in the cosmological case, the physical singularity is not removed / No presente trabalho s?o investigadas algumas conseq??ncias da utiliza??o de uma nova lagrangeana para a eletrodin?mica em dois contextos: um universo espacialmente homog?neo e isotr?pico com campo magn?tico mais um par?metro cosmol?gico A e o problema da massa puntual caregada e est?tica (buraco negro carregado). No caso cosmol?gico, foram obtidas tr?s solu??es gerais: a primeira delas, para A=0, generaliza uma solu??o particular obrida por Novello et al [gr-qc/9806076]; a segunda admite um par?metro cosmol?gico constante e n?o-nulo e a terceira corresponde a um campo magn?tico constante sustentado por um A dependente do tempo. As duas primeiras solu??es s?o n?o-singulares e possuem per?odos inflacion?rios. A terceira solu??o apresenta uma din?mica inflacion?ria exceto por um curto intervalo de tempo. No contexto do problema da massa puntual carregada, a solu??o das equa??es de Einstein-Maxwell ? obtida e comparada com a solu??o padr?o de Reissner-Nordstr?m. Ao contr?rio do caso cosmol?gico, a singularidade f?isica n?o ? removida

<p>This dissertation is concerned with two of the largest questions that we can ask about the nature of physical reality: first, whether physical reality begin to exist and, second, what criteria would physical reality have to fulfill in order to have had a beginning? Philosophers of religion and theologians have previously addressed whether physical reality began to exist in the context of defending the Kal{\'a}m Cosmological Argument (KCA) for theism, that is, (P1) everything that begins to exist has a cause for its beginning to exist, (P2) physical reality began to exist, and, therefore, (C) physical reality has a cause for its beginning to exist. While the KCA has traditionally been used to argue for God's existence, the KCA does not mention God, has been rejected by historically significant Christian theologians such as Thomas Aquinas, and raises perennial philosophical questions -- about the nature and history of physical reality, the nature of time, the nature of causation, and so on -- that should be of interest to all philosophers and, perhaps, all humans. While I am not a religious person, I am interested in the questions raised by the KCA. In this dissertation, I articulate three necessary conditions that physical reality would need to fulfill in order to have had a beginning and argue that, given the current state of philosophical and scientific inquiry, we cannot determine whether physical reality began to exist.</p>

Experimental data suggest that the universe is homogeneous and isotropic on sufficiently large scales. An exact solution of the Einstein field equations exists for a homogeneous and isotropic universe, also known as a Friedmann-Lemaître-Robertson-Walker (FLRW) universe. However, this model is only a first approximation since we know that, locally, the universe has anisotropic and inhomogeneous structures such as galaxies and clusters of galaxies. In order to successfully introduce inhomogeneities and anisotropies to the model one uses perturbative methods. In cosmological perturbations the FLRW universe is considered the zeroth order term in a perturbation expansion and perturbation theory is used to derive higher order terms which one tries to match with observations. In this thesis I present a review of the main concepts of general relativity, discuss the 3+1 formalism which gives us the Einstein field equations in a useful form for the perturbative analysis, and lastly, I derive the first order scalar perturbations of the Einstein field equations.

Nesta tese exploramos uma classe de modelos de N-sítios que foram desenvolvidos para produzir grandes hierarquias de escalas. Usando a abordagem de desconstrução dimensional e correspondências apropriadas, esses modelos puramente quadridimensionais coincidem com modelos AdS5 no limite do contínuo, que corresponde ao limite com um grande núumero de grupos de gauge. Por outro lado, no limite em que há poucos grupos de gauge, tais teorias possuem os acoplamentos entre estados excitados e modos zero dos campos muito distintos dos acoplamentos em construções AdS5, resultando em uma rica fenomenologia a ser explorada no Large Hadron Collider (LHC). O problema da hierarquia do Modelo Padrão (MP) é resolvido se o campo de Higgs estão localizado no infravermelho, assim como nas teorias Randall-Sundrum. A hierarquia de massa dos férmions do MP e misturas são obtidas pelas diferentes localizações dos modos zero dos férmions no espaço dos sítios. Essa estrutura é empregada para abordar o problema da hierarquia sob uma nova perspectiva. Nós mostramos que uma descrição efetiva com poucos sítios de uma dimensão extra curva pode implementar os recentemente introduzidos modelos de relaxion, que são uma nova alternativa para explicar a estabilidade radiativa do setor escalar do MP através do mecanismo de relaxação cosmológico. Esses modelos requerem que o campo experimente grandes variações, que s~ao difíceis de serem geradas em um modelo ultravioleta consistente e de serem compatíveis com o espaço compacto do relaxion. Nós propomos um modelo de N-sítios que gera naturalmente essa grande constante de decaimento necessária para abordar esses problemas. No nosso modelo, a matriz de massa dos pseudo Bósons de Nambu-Goldstone (pBNGs), cujo modo zero faz o papel do relaxion, é idêntica à matriz obtida para uma linha de Wilson pBNG na desconstrução de AdS5. / this thesis we explore a class of N-site models that were developed to generate large-scale hierarchies. Using the dimensional deconstruction approach and appropriate matchings, these purely four-dimensional theories coincide with AdS5 models in the continuum limit, which corresponds to the limit with a large number of gauge groups. On the other hand, in the coarse lattice limit such theories have very distinct couplings of the excited states to zero mode fields compared with AdS5 constructions, resulting in a rich phenomenology to be explored at the Large Hadron Collider (LHC). The Standard Model (SM) hierarchy problem is solved if the Higgs field is infrared-localized as in Randall-Sundrum scenarios. The SM fermion mass hierarchy and mixings are obtained by different localizations of zero mode fermions in the theory space. This framework is employed to tackle the electroweak hierarchy problem from a new perspective. We show that an effective few site description of a warped extra dimension can implement the recently introduced relaxion models, which are a new alternative to explain the radiative stability of the SM scalar sector through the cosmological relaxation mechanism. These models require very large field excursions, which are difficult to generate in a consistent ultraviolet completion and to reconcile with the compact field space of the relaxion. We propose an N-site model that naturally generates the large decay constant needed to address these problems. In our model, the mass matrix of the pseudo-Nambu-Goldstone Bosons (pNGBs), whose zero mode plays the role of the relaxion field, is identical to the one obtained for a pNGBWilson line in the deconstruction of AdS5.

Early Universe physics leaves distinct imprints on the Cosmic Microwave Background (CMB) and Large-Scale Structure (LSS). The current cosmological paradigm to explain the origin of the structures we see in the Universe today (CMB and LSS), named Inflation, says that the Universe went through a period of accelerated expansion. Density fluctuations that eventually have grown into the temperature fluctuations of the CMB and the galaxies and other structures we see in the LSS come from the quantization of the scalar field (inflaton) which provokes the accelerated expansion. The most simple inflationary model, which contains only one slowly-rolling scalar field with canonical kinetic term in the action, produces a power-spectrum (Fourier transform of the two-point correlation function) approximately scale invariant and an almost null bispectrum (Fourier transform of the three-point correlation function). This characteristic is called Gaussianity, once random fields that follow a normal distribution have all the odd moments null. Yet, more complex inflationary models (with more scalar fields and/or non-trivial kinetic terms in the action, etc) and possible alternatives to inflation have a non-vanishing bispectrum which can be parametrized by a non-linearity parameter f_NL, whose value differs from model to model. In this work we studied the basic ingredients to understand such statements and focused on the observational evidences of this parameters and how the current and upcoming galaxy surveys are able to impose constraints to the value of f_NL with a better accuracy, through the multi-tracer technique, than those obtained by means of CMB measurements. / A física do Universo primordial deixa sinais distintos na Radiação Cósmica de Fundo (CMB) e Estrutura em Larga Escala (LSS). O paradigma atual da cosmologia explica a origem das estruturas que vemos hoje (CMB e LSS) através da inflação, teoria que diz que o Universo passou por um período de expansão acelerada. As flutuações de densidade que eventualmente crescem, dando origem às flutuações de temperatura da CMB, às galáxias e outras estruturas que vemos na LSS, provém da quantização do campo escalar (inflaton) que provoca a tal expansão acelerada. O modelo inflacionário mais simples, o qual contém um único campo escalar nas condições de rolamento lento e termo cinético canônico da ação, possui o espectro de potências (transformada de Fourier da função de correlação de dois pontos) aproximadamente invariante de escala e o bispectro (transformada de Fourier da função de correlação de três pontos) aproximadamente nulo. Tal característica é conhecida por Gaussianidade, uma vez que campos aleatórios cuja distribuição é uma normal tem todas as funções de correlação de ordem ímpar nulas. Contudo, modelos inflacionários mais complexos (mais campos escalares, termos cinéticos não-triviais na ação, etc) e alternativas possíveis à inflação possuem um bispectro não nulo, o qual pode ser parametrizado através do parâmetro de não-linearidade f_NL, cujo valor difere de modelo para modelo. Neste trabalho estudamos os ingredientes básicos para entender tais afirmações e focamos nas evidências observacionais desse parâmetro e como os levantamentos de galáxias atuais e futuros podem impor restrições ao valor de f_NL com uma precisão maior, através da técnica de múltiplos traçadores, do que aquelas obtidas com medidas da CMB.

Nesta tese exploramos uma classe de modelos de N-sítios que foram desenvolvidos para produzir grandes hierarquias de escalas. Usando a abordagem de desconstrução dimensional e correspondências apropriadas, esses modelos puramente quadridimensionais coincidem com modelos AdS5 no limite do contínuo, que corresponde ao limite com um grande núumero de grupos de gauge. Por outro lado, no limite em que há poucos grupos de gauge, tais teorias possuem os acoplamentos entre estados excitados e modos zero dos campos muito distintos dos acoplamentos em construções AdS5, resultando em uma rica fenomenologia a ser explorada no Large Hadron Collider (LHC). O problema da hierarquia do Modelo Padrão (MP) é resolvido se o campo de Higgs estão localizado no infravermelho, assim como nas teorias Randall-Sundrum. A hierarquia de massa dos férmions do MP e misturas são obtidas pelas diferentes localizações dos modos zero dos férmions no espaço dos sítios. Essa estrutura é empregada para abordar o problema da hierarquia sob uma nova perspectiva. Nós mostramos que uma descrição efetiva com poucos sítios de uma dimensão extra curva pode implementar os recentemente introduzidos modelos de relaxion, que são uma nova alternativa para explicar a estabilidade radiativa do setor escalar do MP através do mecanismo de relaxação cosmológico. Esses modelos requerem que o campo experimente grandes variações, que s~ao difíceis de serem geradas em um modelo ultravioleta consistente e de serem compatíveis com o espaço compacto do relaxion. Nós propomos um modelo de N-sítios que gera naturalmente essa grande constante de decaimento necessária para abordar esses problemas. No nosso modelo, a matriz de massa dos pseudo Bósons de Nambu-Goldstone (pBNGs), cujo modo zero faz o papel do relaxion, é idêntica à matriz obtida para uma linha de Wilson pBNG na desconstrução de AdS5. / this thesis we explore a class of N-site models that were developed to generate large-scale hierarchies. Using the dimensional deconstruction approach and appropriate matchings, these purely four-dimensional theories coincide with AdS5 models in the continuum limit, which corresponds to the limit with a large number of gauge groups. On the other hand, in the coarse lattice limit such theories have very distinct couplings of the excited states to zero mode fields compared with AdS5 constructions, resulting in a rich phenomenology to be explored at the Large Hadron Collider (LHC). The Standard Model (SM) hierarchy problem is solved if the Higgs field is infrared-localized as in Randall-Sundrum scenarios. The SM fermion mass hierarchy and mixings are obtained by different localizations of zero mode fermions in the theory space. This framework is employed to tackle the electroweak hierarchy problem from a new perspective. We show that an effective few site description of a warped extra dimension can implement the recently introduced relaxion models, which are a new alternative to explain the radiative stability of the SM scalar sector through the cosmological relaxation mechanism. These models require very large field excursions, which are difficult to generate in a consistent ultraviolet completion and to reconcile with the compact field space of the relaxion. We propose an N-site model that naturally generates the large decay constant needed to address these problems. In our model, the mass matrix of the pseudo-Nambu-Goldstone Bosons (pNGBs), whose zero mode plays the role of the relaxion field, is identical to the one obtained for a pNGBWilson line in the deconstruction of AdS5.

Early Universe physics leaves distinct imprints on the Cosmic Microwave Background (CMB) and Large-Scale Structure (LSS). The current cosmological paradigm to explain the origin of the structures we see in the Universe today (CMB and LSS), named Inflation, says that the Universe went through a period of accelerated expansion. Density fluctuations that eventually have grown into the temperature fluctuations of the CMB and the galaxies and other structures we see in the LSS come from the quantization of the scalar field (inflaton) which provokes the accelerated expansion. The most simple inflationary model, which contains only one slowly-rolling scalar field with canonical kinetic term in the action, produces a power-spectrum (Fourier transform of the two-point correlation function) approximately scale invariant and an almost null bispectrum (Fourier transform of the three-point correlation function). This characteristic is called Gaussianity, once random fields that follow a normal distribution have all the odd moments null. Yet, more complex inflationary models (with more scalar fields and/or non-trivial kinetic terms in the action, etc) and possible alternatives to inflation have a non-vanishing bispectrum which can be parametrized by a non-linearity parameter f_NL, whose value differs from model to model. In this work we studied the basic ingredients to understand such statements and focused on the observational evidences of this parameters and how the current and upcoming galaxy surveys are able to impose constraints to the value of f_NL with a better accuracy, through the multi-tracer technique, than those obtained by means of CMB measurements. / A física do Universo primordial deixa sinais distintos na Radiação Cósmica de Fundo (CMB) e Estrutura em Larga Escala (LSS). O paradigma atual da cosmologia explica a origem das estruturas que vemos hoje (CMB e LSS) através da inflação, teoria que diz que o Universo passou por um período de expansão acelerada. As flutuações de densidade que eventualmente crescem, dando origem às flutuações de temperatura da CMB, às galáxias e outras estruturas que vemos na LSS, provém da quantização do campo escalar (inflaton) que provoca a tal expansão acelerada. O modelo inflacionário mais simples, o qual contém um único campo escalar nas condições de rolamento lento e termo cinético canônico da ação, possui o espectro de potências (transformada de Fourier da função de correlação de dois pontos) aproximadamente invariante de escala e o bispectro (transformada de Fourier da função de correlação de três pontos) aproximadamente nulo. Tal característica é conhecida por Gaussianidade, uma vez que campos aleatórios cuja distribuição é uma normal tem todas as funções de correlação de ordem ímpar nulas. Contudo, modelos inflacionários mais complexos (mais campos escalares, termos cinéticos não-triviais na ação, etc) e alternativas possíveis à inflação possuem um bispectro não nulo, o qual pode ser parametrizado através do parâmetro de não-linearidade f_NL, cujo valor difere de modelo para modelo. Neste trabalho estudamos os ingredientes básicos para entender tais afirmações e focamos nas evidências observacionais desse parâmetro e como os levantamentos de galáxias atuais e futuros podem impor restrições ao valor de f_NL com uma precisão maior, através da técnica de múltiplos traçadores, do que aquelas obtidas com medidas da CMB.

Les Oscillations Baryoniques Acoustiques (BAOs) correspondent au phénomène d'ondes acoustiques dans le plasma baryon-photon avant la recombinaison. Ces oscillations impriment une échelle standard, correspondant à l'horizon sonique, qui peut être observée dans les grandes structures de l'Univers. De par cette propriété d'étalon standard, les BAOs permettent d'étudier la relation distance-redshift dans les catalogues de galaxies et fournissent un outil très prometteur pour étudier les propriétés de l'énergie sombre. Les BAOs peuvent être étudiés à partir des statistiques du second ordre (la fonction de corrélation ou le spectre de puissance) dans la distribution des galaxies. Dans cette thèse, on se restreint au cas de la fonction de corrélation. Les BAOs apparaissent dans la fonction de corrélation comme un faible pic à l'échelle de l'horizon sonique. Il y a deux applications principales des BAOs: la détection des BAOs et la contraintes des paramètres cosmologiques grâce à la propriété d'étalon standard. La détection du pic BAO à une échelle proche de l'échelle attendue permet de confirmer le modèle cosmologique actuel. Quant à la contrainte des paramètres cosmologiques, permettant d'étudier l'énergie sombre, il s'agit d'un objectif majeur de la cosmologie moderne. Dans cette thèse, on s'intéresse à différents problèmes statistiques concernant l'étude de la fonction de corrélation dans la distribution des galaxies, avec un accent particulier sur l'étude des BAOs. Dans la première partie, on fait une étude à la fois théorique et pratique du biais dû à la contrainte intégrale dans les estimateurs de la fonction de corrélation. On montre que ce biais est très faible pour les relevés de galaxies actuels. Dans la seconde partie, on s'intéresse à la détection des BAOs. On montre les limitations de la méthode de détection classique, et on propose une nouvelle méthode plus rigoureuse. Notre méthode permet notamment de prendre en compte la dépendance de la matrice de la covariance de l'estimateur en fonction du modèle. Enfin dans la troisième partie, on étudie à nouveau la question la dépendance de la matrice de la covariance, cette fois pour la contrainte des paramètres cosmologiques. On estime une matrice de covariance dépendant du modèle et l'on compare nos contraintes avec les contraintes obtenues en faisant l'approximation habituelle d'une matrice de covariance constante. On montre que l'effet de cette approximation est relativement faible et diminue lorsque les relevés de galaxies grandissent. / Baryon Acoustic Oscillations (BAOs) correspond to the acoustic phenomenon in the baryon-photon plasma before recombination. BAOs imprint a particular scale, corresponding to the sound horizon, that can be observed in large-scale structures of the Universe. Using this standard ruler property, BAOs can be used to probe the distance-redshift relation in galaxy catalogues, thus providing a very promising tool to study dark energy properties. BAOs can be studied from the second order statistics (the correlation function or the power spectrum) in the distribution of galaxies. In this thesis we restrict to the case of the correlation function. BAOs appear in the correlation function as a small localized bump at the scale of the sound horizon in comoving coordinates. There are two major applications of BAO study: BAO detection and cosmological parameter constraints using the standard ruler property. The detection of BAOs at the expected scale enables to confirm the current cosmological model. As for cosmological parameter constraints, enabling the study of dark energy, it is a major goal of modern cosmology. In this thesis we tackle different statistical problems concerning the correlation function analysis in the galaxy distribution, with a focus on the study of BAOs. In the first part, we make both a theoretical and practical study of the bias due to the integral constraints in correlation function estimators. We show that this bias is very small for current galaxy surveys. In the second part we study the BAO detection. We show the limitations of the classical detection method and propose a new method, which is more rigorous. In particular our method enables to take into account the model-dependence of the covariance matrix of the estimators. In the third part, we focus again on the model-dependence of the covariance matrix, but this time for parameter constraints. We estimate a model-dependent covariance matrix and compare our constraints with constraints obtained by making the usual approximation of a constant covariance matrix. We show that the effect of this approximation is quite small for current galaxy surveys and decreases with the survey size.

In dieser Arbeit studieren wir das ekpyrotische Szenario, welches ein kosmologisches Modell des frühen Universums ist. Dieses Modell erklärt mit Hilfe einer kontrahierenden ekpyrotischen Phase die "Anfangsbedingungen" des Universums. Das bedeutet, dass der konventionelle "Urknall" durch einem Rückprall ersetzt wird. In dieser Arbeit versuchen wir Unstimmigkeiten zwischen den Vorhersagen der ekpyrotischen Modelle und den Messungen der Kosmologischen Hintergrundstrahlung des Planck Satelliten zu lösen. Den Planck Messungen zufolge sind die ursprünglichen adiabatischen Fluktuationen fast skaleninvariant und gaußverteilt. Während der ekpyrotischen Phase werden typischer Weise Flutuationen mit nicht-Gaußschen Korrekturen erzeugt. Wir schlagen zwei Ansätze vor, um diese Unstimmigkeit zu beheben. In dem nicht-minimalen entropischen Mechanismus werden fast skaleninvariante entropische Fluktuationen mit Hilfe einer nicht-minimalen kinetischen Kopplung zwischen zwei Skalarfeldern erzeugt. Wir werden zeigen, dass die nicht-Gaußschen Korrekturen während der ekpyrotischen Phase genau Null sind. Dies führt zu insgesamt kleinen nicht-Gaußschen Korrekturen nach der Umwandlung von entropischen zu adiabatischen Fluktuationen. Im Folgendem werden wir eine kinetische Umwandlung untersuchen, die nach einem nicht-singulären Rückprall stattfindet. Das Wachstum der entropischen Fluktuationen während des Rückpralls hat zur Folge, dass die möglichen nicht-Gaußschen Korrekturen, die zur Zeit der ekpyrotischen Phase erzeugt wurden, während des Rückpralls unterdrückt werden. Im letzten Teil der Arbeit gehen wir ein gravierendes Problem des inflationären Paradigmas an, welches "slow-roll eternal inflation" genannt wird. Wir schlagen ein Modell vor, das Ideen von Inflation und Ekpyrosis verbindet. Während der Konflation expandiert das Universum beschleunigt. Die adiabatischen Fluktuationen verhalten sich jedoch wie bei ekpyrotischen Modellen und wird "slow-roll eternal inflation" verhindert. / In this thesis, we study the ekpyrotic scenario, which is a cosmological model of the early universe. In this model the ``initial conditions'' of the universe are determined by a contracting ekpyrotic phase, which means that the conventional ``Big Bang'' is replaced by a bounce. The following thesis addresses the tension between ekpyrotic predictions and the observations of the Cosmic Microwave Background radiation by the Planck team. According to the Planck data, the primordial curvature fluctuations are nearly scale-invariant and Gaussian. However, during ekpyrosis, the fluctuations have typically sizable non-Gaussian signatures. In this thesis, we propose two approaches in order to resolve the tension with observations. In the non-minimal entropic mechanism, nearly scale-invariant entropy perturbations are created due to a non-minimal kinetic coupling between two scalar fields. We will show that the non-Gaussian corrections during ekpyrosis are precisely zero leading to overall small non-Gaussian signatures after the conversion process from entropy perturbations to curvature perturbations. In the following, we will consider a kinetic conversion phase, which takes place after a non-singular bounce. Due to the growth of entropy perturbations during the bounce phase, the possibly large non-Gaussian corrections created during the ekpyrotic phase become suppressed during the bounce. The last part of this thesis addresses a major problem of the inflationary paradigm: Due to large adiabatic fluctuations, slow-roll eternal inflation creates infinitely many physically distinct pocket universes. We propose a model in the framework of scalar-tensor theories, which conflated ideas of both inflation and ekpyrosis. During conflation, the universe undergoes accelerated expansion, but there are no large adiabatic fluctuations like during ekpyrosis resulting in the absence of slow-roll eternal inflation.