Abstract : This PhD thesis presents several aspects of the interplay between high energy physics and cosmology. We study in details the predictions of a class of theories of particle physics beyond the Standard Model concerning the formation of topological defects (and more especially cosmic strings). Within supersymmetric Grand Unified Theories (GUT), we define a class of models that take into account phenomenological constraints from particle physics and cosmology. These models have to be in agreement with proton decay measurements, and have to explain the neutrinos oscillation and the matter/antimatter asymmetry. A phase of hybrid inflation is well motivated in this framework and is also necessary to solve several major problems of the standard model of cosmology. In this framework, the genericity of cosmic string formation is shown. These strings are stable and generically form during the symmetry breaking that ends the (last) hybrid inflationary phase. Therefore, their mass is of the GUT scale, and they can have an important cosmological role. As experimental data show a weak or even negligible role, we can constrain strongly theoretical models.
We have analysed the prediction of string contribution to the Cosmic Microwave Background (CMB) for the two standard models of supersymmetric hybrid inflation. We have shown that they are both in agreement with CMB data, contrary to what was previously thought. The strong limit on the allowed cosmic strings contribution can be translated into constraints on all parameters of the inflationary models : couplings constants, energy scales of inflation. These constraints show that both models suffer from the fine tuning of their superpotential coupling constant, and this tends to reduce the naturalness and the attractivity of these models.
Constraining inflationary physics can also be done through the extraction of parameters from futur CMB experiments measuring polarisation. Several statistical tools exist to estimate cosmological parameters and the errors on these estimations. We have implemented one of these tools in a numerical code and studied analytically and numerically the effect of weak lensing on such analysis. It allows us to validate the usual approximation consisting on neglecting the weak lensing effect on the non-gaussianity (or the four point correlation function) of the CMB signal.