In this thesis we study anti–self–duality equations in four and eight dimensions on manifolds of special Riemannian holonomy, among these hyper–Kähler, Quaternion–Kähler and Spin(7)–manifolds. We first consider the octonionic anti–self–duality equations on manifolds with holonomy Spin(7). We construct explicit solutions to their symmetry reductions, the non–abelian Seiberg–Witten equations, with gauge group SU(2). These solutions are singular for flat and Eguchi–Hanson backgrounds, however we find a solution on a co–homogeneity one hyper–Kähler metric with a domain wall, and the solution is regular away from the wall. We then turn to Quaternion–Kähler four–manifolds, which are locally determined by one scalar function subject to Przanowski’s equation. Using twistorial methods we construct a Lax Pair for Przanowski’s equation, confirming its integrability. The Lee form of a compatible local complex structure gives rise to a conformally invariant differential operator, special cases of the associated generalised Laplace operator are the conformal Laplacian and the linearised Przanowski operator. Using recursion relations we construct a contour integral formula for perturbations of Przanowski’s function. Finally, we construct an algorithm to retrieve Przanowski’s function from twistor data. At last, we investigate the relationship between anti–self–dual Einstein metrics with non–null symmetry in neutral signature and pseudo–, para– and null–Kähler metrics. We classify real–analytic anti–self–dual null–Kähler metrics with a Killing vector that are conformally Einstein. This allows us to formulate a neutral signature version of Tod’s result, showing that around non-singular points all real–analytic anti–self–dual Einstein metrics with symmetry are conformally pseudo– or para–Kähler.

Usando uma transformação de dualidade generalizada, considerações de simetria e supondo que as superfície críticas sejam contínuas, obtivemos o dia grama de fase para sistemas de spins Z (N) bidimensionais e sistemas com invariança de calibre Z (N) a quatro dimensões. Caracterizamos as diversas fases dos sistemas de spins pelo valor esperado das potências dos operadores de ordem e desordem. No sistema com invariança de calibre, por outro lado, estas fases caracterizadas pelo comportamento do valor esperado das potências das alças de Wilson e de \'t Hooft. Obtivemos para ambos os sistemas fases moles em que no caso de spins 2D (calibre 4D) todas as potências dos parâmetros de ordem e desordem ( todas as potências das alças de Wilson e \'t Hooft) são nulas (exibem decaimento com o perímetro da alça). Enquanto no sistema com invariança de calibre todas as combinações de decaimento (área ou perímetro) das alças de Wilson e \'t Hooft são permitidas, as relações de comutação no sistema de spins proíbe a existência de fases em que tanto o parâmetro de ordem como o de desordem são não nulos (exceto quando estes operadores comutam). Apresentamos por completeza as relações de dualidade para sistemas de calibre Z (N) com campos de Higgs a três dimensões. / Using a generalized duality transformation, symetry considerations and assuming that criticality is continuous in the system?s parameters, we obtain the phase diagram for two-dimensional Z (N) spins system?s and four-dimensional gauge Z (N) system\'s. For spins system we characterize the various phases by the expectation value of powers of the order and disorder operators. For gauge systems, on the other hand, the characterization is via decay law of powers of Wilson and \'t Hooft loops. We obtain soft phases for both systems, with the folowing, behaviour: for spins system all powers of order and disorder parameters vanish, whereas for gauge systems all powers of Wilson and \'t Hooft loops decay like the perimeter. Whereas all combinations of area and perimeter decay are allowed for Wilson\'s and \'t Hooft\'s loops, the Z (N) commutation relations for spin systems forbid the simultaneous non-vanishing of order and disorder parameters (except when these operators commute). For completeness we include the duality relations for three-dimensional gauge plus Higgs Z(N) systems.

Usando uma transformação de dualidade generalizada, considerações de simetria e supondo que as superfície críticas sejam contínuas, obtivemos o dia grama de fase para sistemas de spins Z (N) bidimensionais e sistemas com invariança de calibre Z (N) a quatro dimensões. Caracterizamos as diversas fases dos sistemas de spins pelo valor esperado das potências dos operadores de ordem e desordem. No sistema com invariança de calibre, por outro lado, estas fases caracterizadas pelo comportamento do valor esperado das potências das alças de Wilson e de \'t Hooft. Obtivemos para ambos os sistemas fases moles em que no caso de spins 2D (calibre 4D) todas as potências dos parâmetros de ordem e desordem ( todas as potências das alças de Wilson e \'t Hooft) são nulas (exibem decaimento com o perímetro da alça). Enquanto no sistema com invariança de calibre todas as combinações de decaimento (área ou perímetro) das alças de Wilson e \'t Hooft são permitidas, as relações de comutação no sistema de spins proíbe a existência de fases em que tanto o parâmetro de ordem como o de desordem são não nulos (exceto quando estes operadores comutam). Apresentamos por completeza as relações de dualidade para sistemas de calibre Z (N) com campos de Higgs a três dimensões. / Using a generalized duality transformation, symetry considerations and assuming that criticality is continuous in the system?s parameters, we obtain the phase diagram for two-dimensional Z (N) spins system?s and four-dimensional gauge Z (N) system\'s. For spins system we characterize the various phases by the expectation value of powers of the order and disorder operators. For gauge systems, on the other hand, the characterization is via decay law of powers of Wilson and \'t Hooft loops. We obtain soft phases for both systems, with the folowing, behaviour: for spins system all powers of order and disorder parameters vanish, whereas for gauge systems all powers of Wilson and \'t Hooft loops decay like the perimeter. Whereas all combinations of area and perimeter decay are allowed for Wilson\'s and \'t Hooft\'s loops, the Z (N) commutation relations for spin systems forbid the simultaneous non-vanishing of order and disorder parameters (except when these operators commute). For completeness we include the duality relations for three-dimensional gauge plus Higgs Z(N) systems.

GaugThe first half of the thesis concerns Abelian vortices and Yang-Mills theory. It is proved that the 5 types of vortices recently proposed by Manton are actually symmetry reductions of (anti-)self-dual Yang-Mills equations with suitable gauge groups and symmetry groups acting as isometries in a 4-manifold. As a consequence, the twistor integrability results of such vortices can be derived. It is presented a natural definition of their kinetic energy and thus the metric of the moduli space was calculated by the Samols' localisation method. Then, a modified version of the Abelian–Higgs model is proposed in such a way that spontaneous symmetry breaking and the Bogomolny argument still hold. The Painlevé test, when applied to its soliton equations, reveals a complete list of its integrable cases. The corresponding solutions are given in terms of third Painlevé transcendents and can be interpreted as original vortices on surfaces with conical singularity. The last two chapters present the following results in projective differential geometry and Hamiltonians of hydrodynamic-type systems. It is shown that the projective structures defined by the Painlevé equations are not metrisable unless either the corresponding equations admit first integrals quadratic in first derivatives or they define projectively flat structures. The corresponding first integrals can be derived from Killing vectors associated to the metrics that solve the metrisability problem. Secondly, it is given a complete set of necessary and sufficient conditions for an arbitrary affine connection in 2D to admit, locally, 0, 1, 2 or 3 Killing forms. These conditions are tensorial and simpler than the ones in previous literature. By defining suitable affine connections, it is shown that the problem of existence of Killing forms is equivalent to the conditions of the existence of Hamiltonian structures for hydrodynamic-type systems of two components.

Cette thèse s'inscrit dans l'étude mathématique de la percolation, qui regroupe une famille de modèles présentant une transition de phase. Des avancées majeures au cours des quinze dernières années, notamment l'invention du SLE et la preuve de l'invariance conforme de la percolation de Bernoulli critique, nous permettent aujourd'hui d'avoir une image très complète de la percolation de Bernoulli sur le réseau triangulaire. Cependant, de nombreuses questions demeurent ouvertes, et ont motivé notre travail.La première d'entre elle est l'universalité de la percolation plane, qui affirme que les propriétés macroscopiques de la percolation plane critique ne devraient pas dépendre du réseau sous-jacent à sa définition. Nous montrons, dans le cadre de la percolation Divide and Color, un résultat qui va dans le sens de cette universalité et identifions, dans ce contexte, des phénomènes macroscopiques indépendants du réseau microscopique. Une version plus faible d'universalité est donnée par la théorie de Russo-Seymour-Welsh (RSW), et sa validité est connue pour la percolation de Bernoulli (sans dépendance) sur les réseaux plans suffisamment symétriques. Nous étudions de nouveaux arguments de type RSW pour des modèles de percolation avec dépendance. La deuxième question que nous avons abordée est celle de l'absence d'une composante connexe ouverte infinie au point critique, une question importante du point de vue physique, puisqu'elle traduit la continuité de la transition de phase. Dans deux travaux en collaboration avec Hugo Duminil-Copin et Vladas Sidoravicius, nous montrons que la transition de phase est continue pour la percolation de Bernoulli sur le graphe Z^2x{0,...,k}, et pour la percolation FK sur le réseau carré avec paramètre q inférieur ou égal à 4. Enfin, la dernière question qui nous a guidés est la localité du point critique : la donnée des boules de grands rayons d'un graphe suffit-elle à identifier avec une bonne précision la valeur du point critique? Dans un travail en collaboration avec Sébastien Martineau, nous répondons de manière affirmative à cette question dans le cadre des graphes de Cayley de groupes abéliens. / This thesis is part of the mathematical study of percolation theory, which includes a family of models with a phase transition. Major advances in the 2000s, including the invention of SLE and the proof of conformal invariance of critical Bernoulli percolation, provide us with a very complete picture of the Bernoulli percolation process on the triangular lattice. Fortunately, many questions remain open, and motivated our work.The first of these is the universality of planar percolation, which states that the macroscopic properties of critical planar percolation should not depend on the underlying graph. We study this question in the framework of Divide and Color percolation, and prove in this context a result that goes in the direction of universality. A weaker universality statement is given by the theory of Russo-Seymour-Welsh (RSW), which is known to hold for planar Bernoulli percolation (without dependence) on sufficiently symmetric graphs. We study new RSW-type arguments for percolation models with dependence.The second question is the absence of an infinite cluster at the critical point, an important question from a physical point of view, equivalent to the continuity of the phase transition. In two different joint works with Hugo Duminil-Copin and Vladas Sidoravicius, we show that the phase transition is continuous for Bernoulli percolation on the graph Z^2 x {0,...,k} and for FK percolation on the square lattice with parameter q smaller than or equal to 4.Finally, the last question that guided us is the locality of the critical point: is it possible to determine with good accuracy the critical value for Bernoulli percolation on a graph if we know only the balls with large radii? Jointly with Sébastien Martineau, we answer positively to this question in the framework of Cayley graphs of abelian groups.