In this thesis, a bundle F →(M,ω) → B is said to be Lagrangian if (M,ω) is a 2n- dimensional symplectic manifold and the fibres are compact and connected Lagrangian submanifolds of (M,ω), i.e. ω |F = 0 for all F. This condition implies that the fibres and the base space are n-dimensional. Such bundles arise naturally in the study of a special class of dynamical systems in Hamiltonian mechanics, namely those called completely integrable Hamiltonian systems. A celebrated theorem due to Liouville [39], Mineur [46] and Arnol`d [2] provides a semi-global (i.e. in the neighbourhood of a fibre) symplectic classification of Lagrangian bundles, given by the existence of local action-angle coordinates. A proof of this theorem, due to Markus and Meyer [41] and Duistermaat [20], shows that the fibres and base space of a Lagrangian bundle are naturally integral affine manifolds, i.e. they admit atlases whose changes of coordinates can be extended to affine transformations of Rn which preserve the standard cocompact lattice Zn Rn. This thesis studies the problem of constructing Lagrangian bundles from the point of view of affinely at geometry. The first step to study this question is to construct topological universal Lagrangian bundles using the affine structure on the fibres. These bundles classify Lagrangian bundles topologically in the sense that every such bundle arises as the pullback of one universal bundle. However, not all bundles which are isomorphic to the pullback of a topological universal Lagrangian bundle are Lagrangian, as there exist further smooth and symplectic invariants. Even for bundles which admit local action-angle coordinates (these are classified up to isomorphism by topological universal Lagrangian bundles), there is a cohomological obstruction to the existence of an appropriate symplectic form on the total space, which has been studied by Dazord and Delzant in [18]. Such bundles are called almost Lagrangian. The second half of this thesis constructs the obstruction of Dazord and Delzant using the spectral sequence of a topological universal Lagrangian bundle. Moreover, this obstruction is shown to be related to a cohomological invariant associated to the integral affine geometry of the base space, called the radiance obstruction. In particular, it is shown that the integral a ne geometry of the base space of an almost Lagrangian bundle determines whether the bundle is, in fact, Lagrangian. New examples of (almost) Lagrangian bundles are provided to illustrate the theory developed.

Dans ce travail nous nous intéressons aux groupes cristallographiques, i.e. aux sous-groupes du groupe des transformations affines qui agissent proprement discontinûment et de façon cocompacte sur l’espace affine. Ce sont les groupes fondamentaux des variétés affines compactes et complètes. Nous classifions les groupes cristallographiques dont la partie linéaire préserve une forme hermitienne de signature (n,1). Grunewald et Margulis ont prouvé que ces groupes cristallographiques sont virtuellement résolubles (la conjecture d’Auslander affirme que c’est toujours le cas). Notre classification est effectuée pour n ≤ 3. Elle correspond à la classification, à revêtement fini près, des variétés Hermite-Lorentz plates, compactes et complètes en dimension complexe inférieure ou égale à4. Ce travail est inspiré par ceux menés par Bieberbach, puis Fried, et enfin Grunewald et Margulis sur les groupes cristallographiques dont la partie linéaire préserve une forme quadratique définie positive ou lorentzienne. En effectuant cette classification, nous avons été amené à étudier certains familles d’algèbres de Lie nilpotentes de dimension 8. Nous avons ensuite étendu cette classification à celle de toutes les algèbres de Lie 3-nilpotentes de dimension 8 ayant l’algèbre de Lie libre 3-nilpotente à 3générateurs pour quotient. Ce résultat peut être vu comme un pas dans la direction d’une classification des algèbres de Lie nilpotentes de dimension 8. Ensuite nous nous sommes demandé lesquelles de ces algèbres admettent une métrique pseudo-riemannienne plate et nous avons donné une réponse partielle. / In this work we deal with crystallographic groups, i.e. the subgroups of the group of affine transformations that act properly discontinuously and cocompactly on affine space. In otherwords they are the fundamental groups of compact and complete affine manifolds. In this thesis we classify such groups with the additional hypothesis that the linear part preserves a Hermitian form of signature (n,1). Grunewald and Margulis proved that such crystallographic groups are virtually solvable (the Auslander conjecture states that this is always true). Our classification is for n ≤ 3. It corresponds to a classification, up to finite covering, and for complex dimension at most 4, of flat compact complete Hermite-Lorentz manifolds. This is inspired by the works done by Bieberbach,then Fried, and finally Grunewald and Margulis who classified crystallographic groups whose line arpart preserves a positive definite or Lorentzian quadratic form. Making this classification we had to classify a family of 8-dimensional nilpotent Lie algebras. We then extended this classification toall the 8-dimensional 3-step nilpotent Lie algebras having the free 2-step nilpotent Lie algebra on 3generators as quotient. This result can be seen as a step in the direction of a general classification of nilpotent Lie algebras of dimension 8. We then wondered which of these Lie algebras admit flat pseudo-Riemannian metrics and gave a partial answer to this question.

Pour tout entier naturel impair d, on construit un domaine fondamental pour l'action sur l'espace affine de dimension 2d+1 de certains groupes de transformations affines libres non abéliens, discrets, agissant proprement et de partie linéaire Zariski-dense dans SO(d+1, d). Pour tout groupe de Lie semisimple réel non compact G, on construit ensuite un groupe de transformations affines de son algèbre de Lie g qui est libre non abélien, discret, agit proprement sur g et a sa partie linéaire Zariski-dense dans Ad G. Enfin, on donne quelques résultats sur le comportement local des fonctions harmoniques sur le triangle de Sierpinski, plus précisément de leur restriction à un bord du triangle. / For every odd positive integer d, we construct a fundamental domain for the action on the 2d+1-dimensional space of certain groups of affine transformations which are free, nonabelian, act properly discontinuously and have linear part Zariski-dense in SO(d+1,d). Next for every semisimple noncompact real Lie group G, we construct a group of affine transformations of its Lie algebra g which is free, nonabelian, acts properly discontinuously and has linear part Zariski-dense in Ad G. Finally, we give some results about the local behavior of harmonic functions on the Sierpinski triangle restricted to a side of the triangle.