Travail réalisé en cotutelle avec l'université Paris-Diderot et le Commissariat à l'Energie Atomique sous la direction de John Harnad et Bertrand Eynard. / Cette thèse traite des aspects géométriques et d'intégrabilité associés aux modèles de matrices aléatoires. Son but est de présenter diverses applications des modèles de matrices aléatoires allant de la géométrie algébrique aux équations aux dérivées partielles des systèmes intégrables. Ces différentes applications permettent en particulier de montrer en quoi les modèles de matrices possèdent une grande richesse d'un point de vue mathématique. Ainsi, cette thèse abordera d'abord l'étude de la jonction de deux intervalles du support de la densité des valeurs propres au voisinage d'un point singulier. On montrera plus précisément en quoi ce régime limite particulier aboutit aux équations universelles de la hiérarchie de Painlevé II des systèmes intégrables. Ensuite, l'approche des polynômes (bi)-orthogonaux, introduite par Mehta pour le calcul des fonctions de partition, permettra d'énoncer des problèmes de Riemann-Hilbert et d'isomonodromies associés aux modèles de matrices, faisant ainsi le lien avec la théorie de Jimbo-Miwa-Ueno. On montrera en particulier que le cas des modèles à deux matrices hermitiens se transpose à un cas dégénéré de la théorie isomonodromique de Jimbo-Miwa-Ueno qui sera alors généralisé. La méthode des équations de boucles avec ses notions centrales de courbe spectrale et de développement topologique permettra quant à elle de faire le lien avec les invariants symplectiques de géométrie algébrique introduits récemment par Eynard et Orantin. Ce dernier point fera également l'objet d'une généralisation aux modèles de matrices non-hermitien (beta quelconque) ouvrant ainsi la voie à la ``géométrie algébrique quantique'' et à la généralisation de ces invariants symplectiques pour des courbes ``quantiques''. Enfin, une dernière partie sera consacrée aux liens étroits entre les modèles de matrices et les problèmes de combinatoire. En particulier, l'accent sera mis sur les aspects géométriques de la théorie des cordes topologiques avec la construction explicite d'un modèle de matrices aléatoires donnant le dénombrement des invariants de Gromov-Witten pour les variétés de Calabi-Yau toriques de dimension complexe trois utilisées en théorie des cordes topologiques. L'étendue des domaines abordés étant très vaste, l'objectif de la thèse est de présenter de façon la plus simple possible chacun des domaines mentionnés précédemment et d'analyser en quoi les modèles de matrices peuvent apporter une aide précieuse dans leur résolution. Le fil conducteur étant les modèles matriciels, chaque partie a été conçue pour être abordable pour un spécialiste des modèles de matrices ne connaissant pas forcément tous les domaines d'application présentés ici. / This thesis deals with the geometric and integrable aspects associated with random matrix models. Its purpose is to provide various applications of random matrix theory, from algebraic geometry to partial differential equations of integrable systems. The variety of these applications shows why matrix models are important from a mathematical point of view. First, the thesis will focus on the study of the merging of two intervals of the eigenvalues density near a singular point. Specifically, we will show why this special limit gives universal equations from the Painlevé II hierarchy of integrable systems theory. Then, following the approach of (bi) orthogonal polynomials introduced by Mehta to compute partition functions, we will find Riemann-Hilbert and isomonodromic problems connected to matrix models, making the link with the theory of Jimbo, Miwa and Ueno. In particular, we will describe how the hermitian two-matrix models provide a degenerate case of Jimbo-Miwa-Ueno's theory that we will generalize in this context. Furthermore, the loop equations method, with its central notions of spectral curve and topological expansion, will lead to the symplectic invariants of algebraic geometry recently proposed by Eynard and Orantin. This last point will be generalized to the case of non-hermitian matrix models (arbitrary beta) paving the way to ``quantum algebraic geometry'' and to the generalization of symplectic invariants to ``quantum curves''. Finally, this set up will be applied to combinatorics in the context of topological string theory, with the explicit computation of an hermitian random matrix model enumerating the Gromov-Witten invariants of a toric Calabi-Yau threefold. Since the range of the applications encountered is large, we try to present every domain in a simple way and explain how random matrix models can bring new insights to those fields. The common element of the thesis being matrix models, each part has been written so that readers unfamiliar with the domains of application but familiar with matrix models should be able to understand it.

Cette thèse est divisée en cinq parties portant sur les thèmes suivants: l’interprétation physique et algébrique de familles de fonctions orthogonales multivariées et leurs applications, les systèmes quantiques superintégrables en deux et trois dimensions faisant intervenir des opérateurs de réflexion, la caractérisation de familles de polynômes orthogonaux appartenant au tableau de Bannai-Ito et l’examen des structures algébriques qui leurs sont associées, l’étude de la relation entre le recouplage de représentations irréductibles d’algèbres et de superalgèbres et les systèmes superintégrables, ainsi que l’interprétation algébrique de familles de polynômes multi-orthogonaux matriciels. Dans la première partie, on développe l’interprétation physico-algébrique des familles de polynômes orthogonaux multivariés de Krawtchouk, de Meixner et de Charlier en tant qu’éléments de matrice des représentations unitaires des groupes SO(d+1), SO(d,1) et E(d) sur les états d’oscillateurs. On détermine les amplitudes de transition entre les états de l’oscillateur singulier associés aux bases cartésienne et polysphérique en termes des polynômes multivariés de Hahn. On examine les coefficients 9j de su(1,1) par le biais du système superintégrable générique sur la 3-sphère. On caractérise les polynômes de q-Krawtchouk comme éléments de matrices des «q-rotations» de U_q(sl_2). On conçoit un réseau de spin bidimensionnel qui permet le transfert parfait d’états quantiques à l’aide des polynômes de Krawtchouk à deux variables et on construit un modèle discret de l’oscillateur quantique dans le plan à l’aide des polynômes de Meixner bivariés. Dans la seconde partie, on étudie les systèmes superintégrables de type Dunkl, qui font intervenir des opérateurs de réflexion. On examine l’oscillateur de Dunkl en deux et trois dimensions, l’oscillateur singulier de Dunkl dans le plan et le système générique sur la 2-sphère avec réflexions. On démontre la superintégrabilité de chacun de ces systèmes. On obtient leurs constantes du mouvement, on détermine leurs algèbres de symétrie et leurs représentations, on donne leurs solutions exactes et on détaille leurs liens avec les polynômes orthogonaux du tableau de Bannai-Ito. Dans la troisième partie, on caractérise deux familles de polynômes du tableau de Bannai-Ito: les polynômes de Bannai-Ito complémentaires et les polynômes de Chihara. On montre également que les polynômes de Bannai-Ito sont les coefficients de Racah de la superalgèbre osp(1,2). On détermine l’algèbre de symétrie des polynômes duaux -1 de Hahn dans le cadre du problème de Clebsch-Gordan de osp(1,2). On propose une q - généralisation des polynômes de Bannai-Ito en examinant le problème de Racah pour la superalgèbre quantique osp_q(1,2). Finalement, on montre que la q -algèbre de Bannai-Ito sert d’algèbre de covariance à osp_q(1,2). Dans la quatrième partie, on détermine le lien entre le recouplage de représentations des algèbres su(1,1) et osp(1,2) et les systèmes superintégrables du deuxième ordre avec ou sans réflexions. On étudie également les représentations des algèbres de Racah-Wilson et de Bannai-Ito. On montre aussi que l’algèbre de Racah-Wilson sert d’algèbre de covariance quadratique à l’algèbre de Lie sl(2). Dans la cinquième partie, on construit deux familles explicites de polynômes d-orthogonaux basées sur su(2). On étudie les états cohérents et comprimés de l’oscillateur fini et on caractérise une famille de polynômes multi-orthogonaux matriciels. / This thesis is divided into five parts concerned with the following topics: the physical and algebraic interpretation of families of multivariate orthogonal functions and their applications, the study of superintegrable quantum systems in two and three dimensions involving reflection operators, the characterization of families of orthogonal polynomials of the Bannai-Ito scheme and the study of the algebraic structures associated to them, the investigation of the relationship between the recoupling of irreducible representations of algebras and superalgebras and superintegrable systems, as well as the algebraic interpretation of families of matrix multi-orthogonal polynomials. In the first part, we develop the physical and algebraic interpretation of the Krawtchouk, Meixner and Charlier families of multivariate orthogonal polynomials as matrix elements of unitary representations of the SO(d + 1), SO(d, 1) and E(d) groups on oscillator states. We determine the transition amplitudes between the states of the singular oscillator associated to the Cartesian and polyspherical bases in terms of the multivariate Hahn polynomials. We examine the 9j coefficients of su(1,1) through the generic superintegrable system on the 3-sphere. We characterize the q-Krawtchouk polynomials as matrix elements of "q-rotations" of U_q(sl_2). We show how to design a two-dimensional spin network that allows perfect state transfer using the two-variable Krawtchouk polynomials and we construct a discrete model of the two-dimensional quantum oscillator using the two-variable Meixner polynomials. In the second part, we study superintegrable systems of Dunkl type, which involve reflections. We examine the Dunkl oscillator in two and three dimensions, the singular Dunkl oscillator in the plane and the generic system on the 2-sphere with reflections. We show that each of these systems is superintegrable. We obtain their constants of motion, we find their symmetry algebras as well as their representations, we give their exact solutions and we exhibit their relationship with the orthogonal polynomials of the Bannai-Ito scheme. In the third part, we characterize two families of polynomials belonging to the Bannai-Ito scheme: the complementary Bannai-Ito polynomials and the Chihara polynomials. We also show that the Bannai–Ito polynomials arise as Racah coefficients for the osp(1,2) superalgebra. We determine the symmetry algebra associated with the dual − 1 Hahn polynomials in the context of the Clebsch-Gordan problem for osp(1,2). We introduce a q -generalization of the Bannai-Ito polynomials by examining the Racah problem for the quantum superalgebra osp_q(1,2). Finally, we show that the q-deformed Bannai-Ito algebra serves as a covariance algebra for osp_q(1,2). In the fourth part, we determine the relationship between the recoupling of representations of the su(1,1) and osp(1,2) algebras and second-order superintegrable systems with or without reflections. We also study representations of Racah–Wilson and Bannai-Ito algebras. Moreover, we show that the Racah Wilson algebra serves as a quadratic covariance algebra for sl(2). In the fifth part, we explicitly construct two families of d-orthogonal polynomials based on su(2). We investigate the squeezed/coherent states of the finite oscillator and we characterize a family of matrix multi-orthogonal polynomials.

O principal objetivo deste trabalho 6 estudar o comportamento dos zeros de polinômios ortogonais e similares. Inicialmente, consideramos uma relação entre duas sequências ele polinômios ortogonais, onde as medidas associadas estão relacionadas entre si. Usamos esta relação para estudar as propriedades de monotonicidade dos zeros dos polinômios ortogonais relacionados a uma medida obtida através da generalização da medida associada a uma outra sequência de polinômios ortogonais. Apresentamos, como exemplos, os polinômios ortogonais obtidos a partir da generalização das medidas associadas aos polinômios de Jacobi, Laguerre e Charlier. Em urna segunda etapa, consideramos polinômios gerados por uma certa relação de recorrência de três termos com o objetivo de encontrar limitantes, em termos dos coeficientes da relação de recorrência, para as regiões onde os zeros estão localizados. Os zeros são estudados através do problema de autovalor associado a uma matriz de Hessenberg. Aplicações aos polinômios de Szegó, polinômios para-ortogonais e polinômios com coeficientes complexos não-nulos são consideradas. / The main purpose of this work is to study the behavior of the zeros of orthogonal and similar polynomials. Initially, we consider a relation between two sequences of orthogonal polynomials, where the associated measures are related to each other. We use this relation to study the monotonicity propertios of the zeros of orthogonal polynomials related with a measure obtained through a generalization of the measure associated with other sequence of orthogonal polynomials. As examples, we consider the orthogonal polynomials obtained in this way from the measures associated with the Jacobi, Laguerre and Charlier polynomials. We also consider the zeros of polynomials generated by a certain three term recurrence relation. Here, the main objective is to find bounds, in terms of the coefficients of the recurrence relation, for the regions where the zeros are located. The zeros are explored through an eigenvalue representation associated with a Hessenberg matrix. Applications to Szegõ polynomials, para-orthogonal polynomials anti polynomials with non-zero complex coefficients are considered.

A reconstrução de superfícies a partir de nuvens de pontos faz parte de um novo paradigma de modelagem em que modelos computacionais para objetos reais são reconstruídos a partir de dados amostrados sobre a superfície dos mesmos. O principal problema que surge nesse contexto é o fato de que não são conhecidas relações de conectividade entre os pontos que compõe a amostra. Os objetivos do presente trabalho são estudar métodos implícitos para a reconstrução de superfícies e propor algumas melhorias pouco exploradas por métodos já existentes. O uso de funções implícitas no contexto da reconstrução conduz a métodos mais robustos em relação a ruídos, no entanto, uma das principais desvantagens de tais métodos está na dificuldade de capturar detalhes finos e sharp features. Nesse sentido, o presente trabalho propõe o uso de abordagens adaptativas, tanto na poligonalização de superfícies quanto na aproximação de superfícies. Além disso, questões relativas à robustez das soluções locais e à qualidade da malha também são abordadas. Por fim, o método desenvolvido é acoplado aumsoftware traçador de raios afimde se obterumamaneira de modelar cenas tridimensionais utilizando nuvens de pontos, além dos objetos gráficos tradicionais. Os resultados apresentados mostram que muitas das soluções propostas oferecem um incremento à qualidade dos métodos de reconstrução anteriormente propostos / Surface reconstruction from point clouds is part of a new modeling paradigm in which computational models for real objects are reconstructed from data sampled from their surface. The main problem that arises in this context is the fact that there are no known connectivity relationships amongst the points that compose the sample. The objectives of the present work are to study implicit methods for surface reconstruction and to propose some improvements scarcely explored by previous work. The use of implicit functions in the context of surface reconstruction leads to less noise sensitive methods; however, one major drawback of such methods is the difficulty in capturing fine details and sharp features. Towards this, the present work proposes the use of adaptive approaches, not only in the polygonization but also in the surface approximation. Besides, robustness issues in local solutions and mesh quality are also tackled. Finally, the developed method is embedded in a ray tracer software in order to set a basis for modeling tridimensional scenes using point sets, in addition to traditional graphic objects. The presented results show that a great deal of the proposed solutions offer a quality increase to the reconstruction method previously proposed

[pt] Com diversas aplicações em matemática, física e finanças, Teoria das Matrizes Aleatórias (RMT) recentemente atraiu muita atenção. Enquanto o RMT Hermitiano é de especial importância na física por causa da Hermenticidade de operadores associados a observáveis em mecânica quântica, O RMT não-Hermitiano também atraiu uma atenção considerável, em particular porque eles podem ser usados como modelos para sistemas físicos dissipativos ou abertos. No entanto, devido à ausência de uma simetria simplificada, o estudo de matrizes aleatórias não-Hermitianas é, em geral, uma tarefa difícil. Um subconjunto especial de matrizes aleat órias não-Hermitianas, as chamadas matrizes aleatórias normais, são modelos interessantes a serem considerados, uma vez que oferecem mais simetria, tornando-as mais acessíveis às investigções analíticas. Por definição, uma matriz normal M é uma matriz quadrada que troca com seu adjunto Hermitiano. Nesta tese, amplicamos a derivação de estatísticas de valores extremos (EVS) de matrizes aleatórias Hermitianas, com base na abordagem de polinômios ortogonais, em matrizes aleatórias normais e em gases Coulomb 2D em geral. A força desta abordagem a sua compreensão física e intuitiva. Em primeiro lugar, essa abordagem fornece uma derivação alternativa de resultados na literatura. Precisamente falando, mostramos a convergência do autovalor redimensionado com o maior módulo de um conjunto de Ginibre para uma distribuição de Gumbel, bem como a universalidade para um potencial arbitrário radialmente simtérico que atenda certas condições. Em segundo lugar, mostra-se que esta abordagem pode ser generalizada para obter a convergência do autovalor com menor módulo e sua universalidade no limite interno finito do suporte do autovalor. Um aspecto interessante deste trabalho é o fato de que podemos usar técnicas padrão de matrizes aleatórias Hermitianas para obter o EVS de matrizes aleatórias não Hermitianas. / [en] With diverse applications in mathematics, physics, and finance, Random Matrix Theory (RMT) has recently attracted a great deal of attention. While Hermitian RMT is of special importance in physics because of the Hermiticity of operators associated with observables in quantum mechanics, non-Hermitian RMT has also attracted a considerable attention, in particular because they can be used as models for dissipative or open physical systems. However, due to the absence of a simplifying symmetry, the study of non-Hermitian random matrices is, in general, a diffcult task. A special subset of non-Hermitian random matrices, the so-called random normal matrices, are interesting models to consider, since they offer more symmetry, thus making them more amenable to analytical investigations. By definition, a normal matrix M is a square matrix which commutes with its Hermitian adjoint, i.e., (M, M (1)). In this thesis, we present a novel derivation of extreme value statistics (EVS) of Hermitian random matrices, namely the approach of orthogonal polynomials, to normal random matrices and 2D Coulomb gases in general. The strength of this approach is its physical and intuitive understanding. Firstly, this approach provides an alternative derivation of results in the literature. Precisely speaking, we show convergence of the rescaled eigenvalue with largest modulus of a Ginibre ensemble to a Gumbel distribution, as well as universality for an arbitrary radially symmetric potential which meets certain conditions. Secondly, it is shown that this approach can be generalised to obtain convergence of the eigenvalue with smallest modulus and its universality at the finite inner edge of the eigenvalue support. One interesting aspect of this work is the fact that we can use standard techniques from Hermitian random matrices to obtain the EVS of non-Hermitian random matrices.

The primary objective of this study is to develop new computational methods for robust design optimization (RDO) and reliability-based design optimization (RBDO) of high-dimensional, complex engineering systems. Four major research directions, all anchored in polynomial dimensional decomposition (PDD), have been defined to meet the objective. They involve: (1) development of new sensitivity analysis methods for RDO and RBDO; (2) development of novel optimization methods for solving RDO problems; (3) development of novel optimization methods for solving RBDO problems; and (4) development of a novel scheme and formulation to solve stochastic design optimization problems with both distributional and structural design parameters. The major achievements are as follows. Firstly, three new computational methods were developed for calculating design sensitivities of statistical moments and reliability of high-dimensional complex systems subject to random inputs. The first method represents a novel integration of PDD of a multivariate stochastic response function and score functions, leading to analytical expressions of design sensitivities of the first two moments. The second and third methods, relevant to probability distribution or reliability analysis, exploit two distinct combinations built on PDD: the PDD-SPA method, entailing the saddlepoint approximation (SPA) and score functions; and the PDD-MCS method, utilizing the embedded Monte Carlo simulation (MCS) of the PDD approximation and score functions. For all three methods developed, both the statistical moments or failure probabilities and their design sensitivities are both determined concurrently from a single stochastic analysis or simulation. Secondly, four new methods were developed for RDO of complex engineering systems. The methods involve PDD of a high-dimensional stochastic response for statistical moment analysis, a novel integration of PDD and score functions for calculating the second-moment sensitivities with respect to the design variables, and standard gradient-based optimization algorithms. The methods, depending on how statistical moment and sensitivity analyses are dovetailed with an optimization algorithm, encompass direct, single-step, sequential, and multi-point single-step design processes. Thirdly, two new methods were developed for RBDO of complex engineering systems. The methods involve an adaptive-sparse polynomial dimensional decomposition (AS-PDD) of a high-dimensional stochastic response for reliability analysis, a novel integration of AS-PDD and score functions for calculating the sensitivities of the failure probability with respect to design variables, and standard gradient-based optimization algorithms, resulting in a multi-point, single-step design process. The two methods, depending on how the failure probability and its design sensitivities are evaluated, exploit two distinct combinations built on AS-PDD: the AS-PDD-SPA method, entailing SPA and score functions; and the AS-PDD-MCS method, utilizing the embedded MCS of the AS-PDD approximation and score functions. In addition, a new method, named as the augmented PDD method, was developed for RDO and RBDO subject to mixed design variables, comprising both distributional and structural design variables. The method comprises a new augmented PDD of a high-dimensional stochastic response for statistical moment and reliability analyses; an integration of the augmented PDD, score functions, and finite-difference approximation for calculating the sensitivities of the first two moments and the failure probability with respect to distributional and structural design variables; and standard gradient-based optimization algorithms, leading to a multi-point, single-step design process. The innovative formulations of statistical moment and reliability analysis, design sensitivity analysis, and optimization algorithms have achieved not only highly accurate but also computationally efficient design solutions. Therefore, these new methods are capable of performing industrial-scale design optimization with numerous design variables.

<p>This thesis consists of four papers on the following topics in analysis and probability: analysis on Wiener space, asymptotic properties of orthogonal polynomials, and convergence rates in the central limit theorem. The first paper gives lower bounds on the constants in the Meyer inequality from the Malliavin calculus. It is shown that both constants grow at least like <i>(p-1)</i>^-1 or like <i>p</i> when <i>p</i> approaches 1 or ∞ respectively. This agrees with known upper bounds. In the second paper, an extremal problem on Wiener chaos motivates an investigation of the <i>L</i>^p-norms of Hermite polynomials. This is followed up by similar computations for Charlier polynomials in the third paper. In both cases, the <i>L</i>^p-norms present a peculiar behaviour with certain threshold values of p, where the growth rate and the dominating intervals undergo a rapid change. The fourth paper analyzes a connection between probability and numerical analysis. More precisely, known estimates on the convergence rate of finite difference equations are "translated" into results on convergence rates of certain functionals in the central limit theorem. These are also extended, using interpolation of Banach spaces as a main tool. Besov spaces play a central role in the emerging results.</p>

This thesis consists of four papers on the following topics in analysis and probability: analysis on Wiener space, asymptotic properties of orthogonal polynomials, and convergence rates in the central limit theorem. The first paper gives lower bounds on the constants in the Meyer inequality from the Malliavin calculus. It is shown that both constants grow at least like (p-1)-1 or like p when p approaches 1 or ∞ respectively. This agrees with known upper bounds. In the second paper, an extremal problem on Wiener chaos motivates an investigation of the Lp-norms of Hermite polynomials. This is followed up by similar computations for Charlier polynomials in the third paper. In both cases, the Lp-norms present a peculiar behaviour with certain threshold values of p, where the growth rate and the dominating intervals undergo a rapid change. The fourth paper analyzes a connection between probability and numerical analysis. More precisely, known estimates on the convergence rate of finite difference equations are "translated" into results on convergence rates of certain functionals in the central limit theorem. These are also extended, using interpolation of Banach spaces as a main tool. Besov spaces play a central role in the emerging results.

In this thesis, a general pseudospectral formulation for a class of Sturm-Liouville eigenvalue problems is consructed. It is shown that almost all, regular or singular, Sturm-Liouville eigenvalue problems in the Schr&ouml / dinger form may be transformed into a more tractable form. This tractable form will be called here a weighted equation of hypergeometric type with a perturbation (WEHTP) since the non-weighted and unperturbed part of it is known as the equation of hypergeometric type (EHT). It is well known that the EHT has polynomial solutions which form a basis for the Hilbert space of square integrable functions. Pseudospectral methods based on this natural expansion basis are constructed to approximate the eigenvalues of WEHTP, and hence the energy eigenvalues of the Schr&ouml / dinger equation. Exemplary computations are performed to support the convergence numerically.

Cette thèse traite des aspects géométriques et d'intégrabilité associés aux modèles de matrices aléatoires. Son but est de présenter diverses applications des modèles de matrices aléatoires allant de la géométrie algébrique aux équations aux dérivées partielles des systèmes intégrables. Ces différentes applications permettent en particulier de montrer en quoi les modèles de matrices possèdent une grande richesse d'un point de vue mathématique. Ainsi, cette thèse abordera d'abord l'étude de la jonction de deux intervalles du support de la densité des valeurs propres au voisinage d'un point singulier. On montrera plus précisément en quoi ce régime limite particulier aboutit aux équations universelles de la hiérarchie de Painlevé II des systèmes intégrables. Ensuite, l'approche des polynômes (bi)-orthogonaux, introduite par Mehta pour le calcul des fonctions de partition, permettra d'énoncer des problèmes de Riemann-Hilbert et d'isomonodromies associés aux modèles de matrices, faisant ainsi le lien avec la théorie de Jimbo-Miwa-Ueno. On montrera en particulier que le cas des modèles à deux matrices hermitiens se transpose à un cas dégénéré de la théorie isomonodromique de Jimbo-Miwa-Ueno qui sera alors généralisé. La méthode des équations de boucles avec ses notions centrales de courbe spectrale et de développement topologique permettra quant à elle de faire le lien avec les invariants symplectiques de géométrie algébrique introduits récemment par Eynard et Orantin. Ce dernier point fera également l'objet d'une généralisation aux modèles de matrices non-hermitien (beta quelconque) ouvrant ainsi la voie à la ``géométrie algébrique quantique'' et à la généralisation de ces invariants symplectiques pour des courbes ``quantiques''. Enfin, une dernière partie sera consacrée aux liens étroits entre les modèles de matrices et les problèmes de combinatoire. En particulier, l'accent sera mis sur les aspects géométriques de la théorie des cordes topologiques avec la construction explicite d'un modèle de matrices aléatoires donnant le dénombrement des invariants de Gromov-Witten pour les variétés de Calabi-Yau toriques de dimension complexe trois utilisées en théorie des cordes topologiques. L'étendue des domaines abordés étant très vaste, l'objectif de la thèse est de présenter de façon la plus simple possible chacun des domaines mentionnés précédemment et d'analyser en quoi les modèles de matrices peuvent apporter une aide précieuse dans leur résolution. Le fil conducteur étant les modèles matriciels, chaque partie a été conçue pour être abordable pour un spécialiste des modèles de matrices ne connaissant pas forcément tous les domaines d'application présentés ici. / This thesis deals with the geometric and integrable aspects associated with random matrix models. Its purpose is to provide various applications of random matrix theory, from algebraic geometry to partial differential equations of integrable systems. The variety of these applications shows why matrix models are important from a mathematical point of view. First, the thesis will focus on the study of the merging of two intervals of the eigenvalues density near a singular point. Specifically, we will show why this special limit gives universal equations from the Painlevé II hierarchy of integrable systems theory. Then, following the approach of (bi) orthogonal polynomials introduced by Mehta to compute partition functions, we will find Riemann-Hilbert and isomonodromic problems connected to matrix models, making the link with the theory of Jimbo, Miwa and Ueno. In particular, we will describe how the hermitian two-matrix models provide a degenerate case of Jimbo-Miwa-Ueno's theory that we will generalize in this context. Furthermore, the loop equations method, with its central notions of spectral curve and topological expansion, will lead to the symplectic invariants of algebraic geometry recently proposed by Eynard and Orantin. This last point will be generalized to the case of non-hermitian matrix models (arbitrary beta) paving the way to ``quantum algebraic geometry'' and to the generalization of symplectic invariants to ``quantum curves''. Finally, this set up will be applied to combinatorics in the context of topological string theory, with the explicit computation of an hermitian random matrix model enumerating the Gromov-Witten invariants of a toric Calabi-Yau threefold. Since the range of the applications encountered is large, we try to present every domain in a simple way and explain how random matrix models can bring new insights to those fields. The common element of the thesis being matrix models, each part has been written so that readers unfamiliar with the domains of application but familiar with matrix models should be able to understand it. / Travail réalisé en cotutelle avec l'université Paris-Diderot et le Commissariat à l'Energie Atomique sous la direction de John Harnad et Bertrand Eynard.