We examine the covariant properties of generalized models of two-field inflation, with non-canonical kinetic terms and a possibly non-trivial field metric. We demonstrate that kinetic-term derivatives and covariant field derivatives do commute in a proper covariant framework, which was not realized before in the literature. We also define a set of generalized slow-roll parameters, using a unified notation. Within this framework, we study the most general class of models that allows for well-defined adiabatic and entropic sound speeds, which we identify as the models with parallel momentum and field velocity vectors. For these models we write the exact cubic action in terms of the adiabatic and isocurvature perturbations. We thus provide the tool to calculate the exact non-Gaussianity beyond slow-roll and at any scale for these generalized models. We illustrate our general results by considering their long-wavelength limit, as well as with the example of two-field DBI inflation.

Inflationary predictions for the power spectrum of the curvature perturbation have been verified to an excellent degree, leaving many models compatible with observations. In this thesis we studied third-order correlations, that might allow one to further distinguish between inflationary models. From all the possible extensions of the standard inflationary model, we chose to study two-field models with canonical kinetic terms and flat field space. The new feature is the presence of the so-called isocurvature perturbation. Its interplay with the adiabatic perturbation outside the horizon gives birth to non-linearities characteristic of multiple-field models. In this context, we established the second-order gauge-invariant form of the adiabatic and isocurvature perturbation and found the third-order action that describes their interactions. Furthermore, we built on and elaborated the long-wavelength formalism in order to acquire an expression for the parameter of non-Gaussianity fNL as a function of the potential of the fields. We next used this formula to study analytically, within the slow-roll hypothesis, general classes of potentials and verified our results numerically for the exact theory. From this study, we deduced general conclusions about the properties of fNL, its magnitude depending on the characteristics of the field trajectory and the isocurvature component, as well as its dependence on the magnitude and relative size of the three momenta of which the three-point correlator is a function.

We examine the momentum dependence of the bispectrum of two-field inflationary models within the long-wavelength formalism. We determine the sources of scale dependence in the expression for the parameter of non-Gaussianity fNL and study two types of variation of the momentum triangle: changing its size and changing its shape. We introduce two spectral indices that quantify the possible types of momentum dependence of the local type fNL and illustrate our results with examples.

We study the second-order gauge-invariant adiabatic and isocurvature perturbations in terms of the scalar fields present during inflation, along with the related fully non-linear space gradient of these quantities. We discuss the relation with other perturbation quantities defined in the literature. We also construct the exact cubic action of the second-order perturbations (beyond any slow-roll or super-horizon approximations and including tensor perturbations), both in the uniform energy density gauge and the flat gauge in order to settle various gauge-related issues. We thus provide the tool to calculate the exact non-Gaussianity beyond slow-roll and at any scale.

We use the long-wavelength formalism to compute the bispectral non-Gaussianity produced in two-field inflation. We find an exact result that is used as the basis of numerical studies, and an explicit analytical slow-roll expression for several classes of potentials that gives insight into the origin and importance of the various contributions to fNL. We also discuss the momentum dependence of fNL. Based on these results we find a simple model that produces a relatively large non-Gaussianity. We show that the long-wavelength formalism is a viable alternative to the standard delta-N formalism, and can be preferable to it in certain situations.

We consider the effect on the propagation of light of inhomogeneities with sizes of order 10 Mpc or larger. The Universe is approximated through a variation of the Swiss-cheese model. The spherical inhomogeneities are void-like, with central underdensities surrounded by compensating overdense shells. We study the propagation of light in this background, assuming that the source and the observer occupy random positions, so that each beam travels through several inhomogeneities at random angles. The distribution of luminosity distances for sources with the same redshift is asymmetric, with a peak at a value larger than the average one. The width of the distribution and the location of the maximum increase with increasing redshift and length scale of the inhomogeneities. We compute the induced dispersion and bias on cosmological parameters derived from the supernova data. They are too small to explain the perceived acceleration without dark energy, even when the length scale of the inhomogeneities is comparable to the horizon distance. Moreover, the dispersion and bias induced by gravitational lensing at the scales of galaxies or clusters of galaxies are larger by at least an order of magnitude.

We extend the covariant analysis of the brane cosmological evolution in order to take into account, apart from a general matter content and an induced-gravity term on the brane, a Gauss-Bonnet term in the bulk. The gravitational effect of the bulk matter on the brane evolution can be described in terms of the total bulk mass as measured by a bulk observer at the location of the brane. This mass appears in the effective Friedmann equation through a term characterized as generalized dark radiation that induces mirage effects in the evolution. We discuss the normal and self-accelerating branches of the combined system. We also derive the Raychaudhuri equation that can be used in order to determine if the cosmological evolution is accelerating.

We study the form of the luminosity distance as a function of redshift in the presence of large scale inhomogeneities, with sizes of order 10 Mpc or larger. We approximate the Universe through the Swiss-cheese model, with each spherical region described by the Tolman-Bondi metric. We study the propagation of light beams in this background, assuming that the locations of the source and the observer are random. We derive the optical equations for the evolution of the beam area and shear. Through their integration we determine the configurations that can lead to an increase of the luminosity distance relative to the homogeneous cosmology. We find that this can be achieved if the Universe is composed of spherical void-like regions, with matter concentrated near their surface. For inhomogeneities consistent with the observed large scale structure, the relative increase of the luminosity distance is of the order of a few percent at redshifts near 1, and falls short of explaining the substantial increase required by the supernova data. On the other hand, the effect we describe is important for the correct determination of the energy content of the Universe from observations.

The acceleration parameter defined through the local volume expansion is negative for a pressureless, irrotational fluid with positive energy density. In the presence of inhomogeneities or anisotropies the volume expansion rate results from averaging over various directions. On the other hand, the observation of light from a certain source in the sky provides information on the expansion along the direction to that source. If there are preferred directions in the underlying geometry one can define several expansion parameters. We provide such definitions for the case of the Tolman-Bondi metric. We then examine the effect of a localized inhomogeneity on the surrounding cosmological fluid. Our framework is similar in spirit to the model of spherical collapse. For an observer in the vicinity of a central overdensity, the perceived local evolution is consistent with acceleration in the direction towards the center of the overdensity, and deceleration perpendicularly to it. A negative mass leads to deceleration along the radial direction, and acceleration perpendicularly to it. If the observer is located at the center of an overdensity the null geodesics are radial. The form of the luminosity distance as a function of the redshift is consistent with acceleration for a certain range of redshifts.