Contribution to uncertainties evaluation for fast reactors neutronic cross sections

The thesis has been motivated by a wish to increase the uncertainty knowledge on nuclear data, for safety criteria. It aims the cross sections required by core calculation for sodium fast reactors (SFR), and new tools to evaluate its.The main objective of this work is to provide new tools in order to create coherent evaluated files, with reliable and mastered uncertainties. To answer those problematic, several methods have been implemented within the CONRAD code, which is developed at CEA of Cadarache. After a summary of all the elements required to understand the evaluation world, stochastic methods are presented in order to solve the Bayesian inference. They give the evaluator more information about probability density and they also can be used as validation tools. The algorithms have been successfully tested, despite long calculation time. Then, microscopic constraints have been implemented in CONRAD. They are defined as new information that should be taken into account during the evaluation process. An algorithm has been developed in order to solve, for example, continuity issues between two energy domains, with the Lagrange multiplier formalism. Another method is given by using a marginalization procedure, in order to either complete an existing evaluation with new covariance or add systematic uncertainty on an experiment described by two theories. The algorithms are well performed along examples, such the ²³⁸U total cross section. The last parts focus on the integral data feedback, using methods of integral data assimilation to reduce the uncertainties on cross sections. This work ends with uncertainty reduction on key nuclear reactions, such the capture and fission cross sections of ²³⁸U and ²³⁹Pu, thanks to PROFIL and PROFIL-2 experiments in Phenix and the Jezebel benchmark. (author)