We investigate cosmological constraints on K-mouflage models of modified gravity. We consider two scenarios: one where the background evolution is free to deviate from $\Lambda$CDM (K-mouflage) and another one which reproduces a $\Lambda$CDM expansion (K-mimic), implementing both of them into the EFTCAMB code. We discuss the main observational signatures of these models and we compare their cosmological predictions to different datasets, including CMB, CMB lensing, SNIa and different galaxy catalogues. We argue about the possibility of relieving the $H_0$ and weak lensing tensions within these models, finding that K-mouflage scenarios effectively ease the tension on the Hubble Constant. Our final 95\% C.L. bounds on the $\epsilon_{2,0}$ parameter that measures the overall departure from $\Lambda$CDM (corresponding to $\epsilon_{2,0}=0$) are $-0.04\leq \epsilon_{2,0} <0$ for K-mouflage and $0< \epsilon_{2,0} <0.002$ for K-mimic. In the former case the main constraining power comes from changes in the background expansion history, while in the latter case the model is strongly constrained by measurements of the amplitude of matter perturbations. The sensitivity of these cosmological constraints closely matches that of solar system probes. We show that these constraints could be significantly tightened with future ideal probes like CORE.

I compute the matter bispectrum of large-scale structure up to two loops in the Standard Perturbation Theory and up to three loops in the MPTbreeze renormalised perturbation theory, determining the contributing loop diagrams and evaluating them numerically. In the process I remove the leading divergences in the integrands, thus making them infrared-safe. By comparing the results to numerical simulations, I show that in the case of the Standard Perturbation Theory, the bispectrum at two loops is more accurate than at one loop, up to $k_{\textrm{max}} \sim 0.09 \, h/\textrm{Mpc}$ at $z=0$ and $k_{\textrm{max}} \sim 0.11 \, h/\textrm{Mpc}$ at $z=1$. The MPTbreeze can be employed to accurately model the matter bispectrum up to $k_{\textrm{max}} \sim 0.17 \, h/\textrm{Mpc}$ at $z=0$ and $k_{\textrm{max}} \sim 0.24 \, h/\textrm{Mpc}$ at $z=1$ using the results at three loops.

We evaluate for the first time the dark matter bispectrum of large-scale structure at two loops in the Standard Perturbation Theory and at three loops in the Renormalised Perturbation Theory (MPTbreeze formalism), removing in each case the leading divergences in the integrals in order to make them infrared-safe. We show that the Standard Perturbation Theory at two loops can be employed to model the matter bispectrum further into the quasi-nonlinear regime compared to one loop, up to $k_{\text{max}} \sim 0.1 \, h/{\text{Mpc}}$ at $z = 0$, but without reaching a high level of accuracy. In the case of the MPTbreeze method, we show that its bispectra decay at smaller and smaller scales with increasing loop order, but with smaller improvements. At three loops, this model predicts the bispectrum accurately up to scales $k_{\text{max}} \sim 0.17 \, h/{\text{Mpc}}$ at $z = 0$ and $k_{\text{max}} \sim 0.24 \, h/{\text{Mpc}}$ at $z = 1$.

We forecast constraints on primordial non-Gaussianity (PNG) and bias parameters from measurements of galaxy power spectrum and bispectrum in future radio continuum and optical surveys. In the galaxy bispectrum, we consider a comprehensive list of effects, including the bias expansion for non-Gaussian initial conditions up to second order, redshift space distortions, redshift uncertainties and theoretical errors. These effects are all combined in a single PNG forecast for the first time. Moreover, we improve the bispectrum modelling over previous forecasts, by accounting for trispectrum contributions. All effects have an impact on final predicted bounds, which varies with the type of survey. We find that the bispectrum can lead to improvements up to a factor $\sim 5$ over bounds based on the power spectrum alone, leading to significantly better constraints for local-type PNG, with respect to current limits from \textit{Planck}. Future radio and photometric surveys could obtain a measurement error of $\sigma(f_{\mathrm{NL}}^{\mathrm{loc}}) \approx 0.2$. In the case of equilateral PNG, galaxy bispectrum can improve upon present bounds only if significant improvements in the redshift determinations of future, large volume, photometric or radio surveys could be achieved. For orthogonal non-Gaussianity, expected constraints are generally comparable to current ones.

We study the matter bispectrum of large structure by comparing theoretical models (perturbation theories and halo models) to numerical simulations using shape and amplitude correlators. We show that among the perturbation theories at one loop the effective field theory of large scale structure extends the furthest into the non-linear regime. We analyse the one and two-loop bispectra in the renormalised perturbation theory and we show that there is a significant extension in the range where results are accurate when going to two loops. In the case of the halo model, we show that there are deficiencies in the modelling of the two-halo term at redshifts z>0 that worsen in the past. Based on this observation and on the shapes identified in the halo model, we build a simple `three-shape model' that provides a good fit to N-body simulations on all scales, at both low and high redshifts. We show that this model can be easily extended to local and equilateral primordial non-Gaussianity using the same shapes.

We study the matter bispectrum of large-scale structure by comparing the predictions of different perturbative and phenomenological models with the full three-dimensional bispectrum from $N$-body simulations estimated using modal methods. We show that among the perturbative approaches, effective field theory succeeds in extending the range of validity furthest on intermediate scales, at the cost of free additional parameters. By studying the halo model, we show that although it is satisfactory in the deeply non-linear regime, it predicts a deficit of power on intermediate scales, worsening at redshifts $z>0$. By comparison with the $N$-body bispectrum on those scales, we show that there is a significant squeezed component underestimated in the halo model. On the basis of these results, we propose a new three-shape model, based on the tree-level, squeezed and constant bispectrum shapes we identified in the halo model; after calibration this fits the simulations on all scales and redshifts of interest. We extend this model further to primordial non-Gaussianity of the local and equilateral types by showing that the same shapes can be used to describe the additional non-Gaussian component in the matter bispectrum. This method provides a HALOFIT-like prototype of the bispectrum that could be used to describe and test parameter dependencies and should be relevant for the bispectrum of weak gravitational lensing and wider applications.

We study the matter bispectrum of the large-scale structure by comparing different perturbative and phenomenological models with measurements from $N$-body simulations obtained with a modal bispectrum estimator. Using shape and amplitude correlators, we directly compare simulated data with theoretical models over the full three-dimensional domain of the bispectrum, for different redshifts and scales. We review and investigate the main perturbative methods in the literature that predict the one-loop bispectrum: standard perturbation theory, effective field theory, resummed Lagrangian and renormalised perturbation theory, calculating the latter also at two loops for some triangle configurations. We find that effective field theory (EFT) succeeds in extending the range of validity furthest into the mildly nonlinear regime, albeit at the price of free extra parameters requiring calibration on simulations. For the more phenomenological halo model, we confirm that despite its validity in the deeply nonlinear regime it has a deficit of power on intermediate scales, which worsens at higher redshifts; this issue is ameliorated, but not solved, by combined halo-perturbative models. We show from simulations that in this transition region there is a strong squeezed bispectrum component that is significantly underestimated in the halo model at earlier redshifts. We thus propose a phenomenological method for alleviating this deficit, which we develop into a simple phenomenological "three-shape" benchmark model based on the three fundamental shapes we have obtained from studying the halo model. When calibrated on the simulations, this three-shape benchmark model accurately describes the bispectrum on all scales and redshifts considered, providing a prototype bispectrum HALOFIT-like methodology that could be used to describe and test parameter dependencies.

We improve predictions of the cosmic microwave background (CMB) power spectrum induced by cosmic strings by using source terms obtained from Nambu-Goto network simulations in an expanding universe. We use three high-resolution cosmic string simulations that cover the entire period from recombination until late-time $\Lambda$ domination to calculate unequal time correlators (UETCs) for scalar, vector and tensor components of the cosmic-string energy-momentum tensor. We calculate the CMB angular power spectrum from strings in two ways. First, to aid comparison with previous work, we fit our simulated UETCs to those obtained from different parameter combinations from the unconnected segment model and then calculate the CMB power spectra using these parameters to represent the string network. Second and more accurately, we decompose the UETCs into their corresponding eigenvalues and eigenvectors and input them directly into an Einstein-Boltzmann solver to calculate the power spectrum for each of the three simulation time periods. We combine the three simulations together, using each of them in its relevant redshift range and we obtain overall power spectra in temperature and polarisation channels. Finally, we use the power spectra obtained with the latest Planck and BICEP2 likelihoods to obtain constraints on the cosmic string tension.

We use three domain wall simulations from the radiation era to the late time dark energy domination era based on the PRS algorithm to calculate the energy-momentum tensor components of domain wall networks in an expanding universe. Unequal time correlators in the radiation, matter and cosmological constant epochs are calculated using the scaling regime of each of the simulations. The CMB power spectrum of a network of domain walls is determined. The first ever quantitative constraint for the domain wall surface tension is obtained using a Markov chain Monte Carlo method; an energy scale of domain walls of 0.93 MeV, which is close but below the Zel'dovich bound, is determined.

Cosmic strings generate vector and tensor modes in the B-channel of polarization, as well as the usual temperature power spectrum and E-mode polarization spectrum. We use the power spectrum obtained from high-resolution Nambu-Goto cosmic string simulations together with the Planck and BICEP2 likelihoods to explore the degeneracies appearing between cosmic strings and other cosmological parameters in different inflationary scenarios, as well as the constraints that can be imposed on cosmic strings in each of these situations. In standard $\Lambda$CDM, the Planck likelihood yields an upper limit $G\mu<1.49 \times 10^{-7}$ (95% confidence). We also analyse the possibility of explaining the BB power spectrum signal recently detected by the BICEP2 probe. We find that cosmic strings alone are able to explain only part of the B-mode polarization signal. Apart from the standard $\Lambda$CDM model, we look at the following non-minimal parameters: the running of the spectral index, non-zero tensor-to-scalar ratio, additional degrees of freedom ($N_{eff}$) and sterile neutrinos. We find that in both Planck and BICEP2 scenarios adding $N_{eff}$ induces degeneracies between cosmic strings and $N_{eff}$ and other $\Lambda$CDM parameters. With $N_{eff}$ a larger contribution from cosmic strings is allowed, even favoured, but after combining with large-scale structure data, such as BAOs, strings remain strongly constrained.