For a locally compact group G, the papers [13] and [7] have many results about G-invariant subsets of G-modules, and the relationship between G-module maps, L1(G)-module maps and M(G)-module maps. In both papers, the results were given for one specific module action. In this thesis we extended many of their results to arbitrary Banach G-modules. In addition, we give detailed proofs of most of the results found in the first section of the paper [21].

For a locally compact group G, the papers [13] and [7] have many results about G-invariant subsets of G-modules, and the relationship between G-module maps, L1(G)-module maps and M(G)-module maps. In both papers, the results were given for one specific module action. In this thesis we extended many of their results to arbitrary Banach G-modules. In addition, we give detailed proofs of most of the results found in the first section of the paper [21].

This thesis is devoted to the study of operators on discrete structures. The operators are supposed to be self-adjoint and obey a certain translation invariance property. The discrete structures are given as Cayley graphs via finitely generated groups. Here, sofic groups and amenable groups are in the center of our considerations. Note that every finitely generated amenable group is sofic. We investigate the spectrum of a discrete self-adjoint operator by studying a sequence of finite dimensional analogues of these operators. In the setting of amenable groups we obtain these approximating operators by restricting the operator in question to finite subsets Qn , n ∈ N. These finite dimensional operators are self-adjoint and therefore admit a well-defined normalized eigenvalue counting function. The limit of the normalized eigenvalue counting functions when |Qn | → ∞ (if it exists) is called the integrated density of states (IDS). It is a distribution function of a probability measure encoding the distribution of the spectrum of the operator in question on the real axis. In this thesis, we prove the existence of the IDS in various geometric settings and for different types of operators. The models we consider include deterministic as well as random situations. Depending on the specific setting, we prove existence of the IDS as a weak limit of distribution functions or even as a uniform limit. Moreover, in certain situations we are able to express the IDS via a semi-explicit formula using the trace of the spectral projection of the original operator. This is sometimes referred to as the validity of the Pastur-Shubin trace formula. In the most general geometric setting we study, the operators are defined on Cayley graphs of sofic groups. Here we prove weak convergence of the eigenvalue counting functions and verify the validity of the Pastur-Shubin trace formula for random and non-random operators . These results apply to operators which not necessarily bounded or of finite hopping range. The methods are based on resolvent techniques. This theory is established without having an ergodic theorem for sofic groups at hand. Note that ergodic theory is the usual tool used in the proof of convergence results of this type. Specifying to operators on amenable groups we are able to prove stronger results. In the discrete case, we show that the IDS exists uniformly for a certain class of finite hopping range operators. This is obtained by using a Banach space-valued ergodic theorem. We show that this applies to eigenvalue counting functions, which implies their convergence with respect to the Banach space norm, in this case the supremum norm. Thus, the heart of this theory is the verification of the Banach space-valued ergodic theorem. Proceeding in two steps we first prove this result for so-called ST-amenable groups. Then, using results from the theory of ε-quasi tilings, we prove a version of the Banach space-valued ergodic theorem which is valid for all amenable groups. Focusing on random operators on amenable groups, we prove uniform existence of the IDS without the assumption that the operator needs to be of finite hopping range or bounded. Moreover, we verify the Pastur-Shubin trace formula. Here we present different techniques. First we show uniform convergence of the normalized eigenvalue counting functions adapting the technique of the Banach space-valued ergodic theorem from the deterministic setting. In a second approach we use weak convergence of the eigenvalue counting functions and additionally obtain control over the convergence at the jumps of the IDS. These ingredients are applied to verify uniform existence of the IDS. In both situations we employ results from the theory of large deviations, in order to deal with long-range interactions.

Dans la veine du programme d'Erlangen de Klein, travaux d'E. Cartan, M. Gromov, et d'autres, ce travail se trouve à cheval, entre la géométrie et les actions de groupes. Le thème global serait de comprendre les groupes d'isométries des variétés pseudo-riemanniennes. Plus précisément, suivant une "conjecture vague" de Gromov, classifier les variétés pseudo-riemanniennes dont le groupe d'isométries agit non-proprement, i.e. que son action ne préserve pas de métrique riemannienne auxiliaire?Plusieurs travaux ont été accomplis dans le cas des métriques lorentziennes (i.e. de signature (- +...+)). En revanche, le cas pseudo-riemannien général semble hors de portée.Les structures Hermite-Lorentz se trouvent entre le cas lorentzien et le premier cas pseudo-riemannien général, i.e. de signature (- - +…+). De plus, elle se définit sur des variétés complexes, et promet une extra-rigidité. Plus précisément, une structure Hermite-Lorentz sur une variété complexe consiste en une métrique pseudo-riemannienne de signature (- - +…+) qui est hermitienne au sens qu'elle est invariante par la structure presque complexe. Par analogie au cas hermitien classique, on définit naturellement une notion de métrique Kähler-Lorentz.Comme exemple, on a l'espace de Minkowski complexe ; dans un certain sens, on a un temps de dimension 1 complexe (du point de vue réel, le temps est 2-dimensionnel). On a également l'espace de Sitter et anti de Sitter complexes. Ils ont une courbure holomorphe constante, et généralisent dans ce sens les espaces projectifs et hyperboliques complexes.Cette thèse porte sur les variétés Hermite-Lorentz homogènes. En plus des exemples cités, il y a deux autres espaces symétriques, qui peuvent naturellement jouer le rôle de complexification des espaces de Sitter et anti de Sitter réels.Le résultat principal de la thèse est un théorème de rigidité de ces espaces symétriques : tout espace Hermite-Lorentz homogène à isotropie irréductible est l'un des cinq espaces symétriques précédents. D'autres résultats concernent le cas où l'on remplace l'hypothèse d'irréductibilité par le fait que le groupe d'isométries soit semi-simple. / In the vein of Klein's Erlangen program, the research works of E. Cartan, M.Gromov and others, this work straddles between geometry and group actions. The overall theme is to understand the isometry groups of pseudo-Riemannian manifolds. Precisely, following a "vague conjecture" of Gromov, our aim is to classify Pseudo-Riemannian manifolds whose isometry group act’s not properly, i.e that it’s action does not preserve any auxiliary Riemannian metric. Several studies have been made in the case of the Lorentzian metrics (i.e of signature (- + .. +)). However, general pseudo-Riemannian case seems out of reach. The Hermite-Lorentz structures are between the Lorentzian case and the former general pseudo-Riemannian, i.e of signature (- -+ ... +). In addition, it’s defined on complex manifolds, and promises an extra-rigidity. More specifically, a Hermite-Lorentz structure on a complex manifold is a pseudo-Riemannian metric of signature (- -+ ... +), which is Hermitian in the sense that it’s invariant under the almost complex structure. By analogy with the classical Hermitian case, we naturally define a notion of Kähler-Lorentz metric. We cite as example the complex Minkowski space in where, in a sense, we have a one-dimensional complex time (the real point of view, the time is two-dimensional). We cite also the de Sitter and Anti de Sitter complex spaces. They have a constant holomorphic curvature, and generalize in this direction the projective and complex hyperbolic spaces.This thesis focuses on the Hermite-Lorentz homogeneous spaces. In addition with given examples, two other symmetric spaces can naturally play the role of complexification of the de Sitter and anti de Sitter real spaces.The main result of the thesis is a rigidity theorem of these symmetric spaces: any space Hermite-Lorentz isotropy irreducible homogeneous is one of the five previous symmetric spaces. Other results concern the case where we replace the irreducible hypothesis by the fact that the isometry group is semisimple.