It is intended for PhD students, teachers and researchers who are interested in probability theory, statistics, and in their applications.The duration of each school is 13 days (it was 17 days up to 2005), and up to 70 participants can attend it. The aim is to provide, in three high-level courses, a comprehensive study of some fields in probability theory or Statistics. The lecturers are chosen by an international scientific board. The participants themselves also have the opportunity to give short lectures about their research work.Participants are lodged and work in the same building, a former seminary built in the 18th century in the city of Saint-Flour, at an altitude of 900 m. The pleasant surroundings facilitate scientific discussion and exchange.

We construct and study the Yang-Mills measure in two dimensions. According to the informal description given by the physicists, it is a probability measure on the space of connections modulo gauge transformations on a principal bundle with compact structure group. We are interested in the case where the base space of this bundle is a compact orientable surface.The construction of the measure in a discrete setting, where the base space of the fiber bundle is replaced by a graph traced on a surface, is quite well understood thanks to the work of E. Witten. In contrast, the continuum limit of this construction, which should allow to put a genuine manifold as base space, still remains problematic.This work presents a complete and unified approach of the discrete theory and of its continuum limit. We give a geometrically consistent definition of the Yang-Mills measure, under the form of a random holonomy along a wide, intrinsic and natural class of loops. This definition allows us to study combinatorial properties of the measure, like its Markovian behaviour under the surgery of surfaces, as well as properties specific to the continuous setting, for example, some of its microscopic properties. In particular, we clarify the links between the Yang-Mills measure and the white noise and show that there is a major difference between the Abelian and semi-simple theories. We prove that it is possible to construct a white noise using the measure as a starting point and vice versa in the Abelian case but we show a result of asymptotic independence in the semi-simple case which suggests that it is impossible to extract a white noise from the measure.

We investigate a relation between the Brownian motion on the unitary group and the most natural random walk on the symmetric group, based on Schur-Weyl duality. We use this relation to establish a convergent power series expansion for the expectation of a product of traces of powers of a random unitary matrix under the heat kernel measure. This expectation turns out to be the generating series of certain paths in the Cayley graph of the symmetric group. Using our expansion, we recover asymptotic results of Xu, Biane and Voiculescu. We give an interpretation of our main expansion in terms of random ramified coverings of a disk.

We prove that for a finite collection of real-valued functions $f_{1},...,f_{n}$ on the group of complex numbers of modulus 1 which are derivable with Lipschitz continuous derivative, the distribution of $(\tr f_{1},...,\tr f_{n})$ under the properly scaled heat kernel measure at a given time on the unitary group $\U(N)$ has Gaussian fluctuations as $N$ tends to infinity, with a covariance for which we give a formula and which is of order $N^{-1}$. In the limit where the time tends to infinity, we prove that this covariance converges to that obtained by P. Diaconis and S. Evans in a previous work on uniformly distributed unitary matrices. Finally, we discuss some combinatorial aspects of our results.Comment: 44 page