Uma maneira de resolver problemas de programação linear de grande escala é explorar a relaxação lagrangeana das restrições \"difíceis\'\' e utilizar métodos de subgradientes. Populares por fornecerem rapidamente boas aproximações de soluções duais, eles não produzem diretamente as soluções primais. Para obtê-las com custo computacional adequado, pode-se construir seqüências ergódicas ou utilizar uma técnica proposta recentemente, denominada algoritmo do volume. As propriedades teóricas de convergência não foram bem estabelecidas nesse algoritmo, mas pequenas modificações permitem a demonstração da convergência dual. Destacam-se como adaptações o algoritmo do volume revisado, um método de feixes específico, e o algoritmo do volume incorporado ao método de variação do alvo. Este trabalho foi baseado no estudo desses algoritmos e de todos os conceitos envolvidos, em especial, análise convexa e otimização não diferenciável. Estudamos as principais diferenças teóricas desses métodos e realizamos comparações numéricas com problemas lineares e lineares inteiros, em particular, o corte máximo em grafos. / One way to solve large-scale linear programming problems is to exploit the Lagrangian relaxation of the difficult constraints and use subgradient methods. Such methods are popular as they give good approximations of dual solutions. Unfortunately, they do not directly yield primal solutions. Two alternatives to obtain primal solutions under reasonable computational cost are the construction of ergodic sequences and the use of the recently developed volume algorithm. While the convergence of ergodic sequences is well understood, the convergence properties of the volume algorithm is not well established in the original paper. This lead to some modifications of the original method to ease the proof of dual convergence. Three alternatives are the revised volume algorithm, a special case of the bundle method, and the volume algorithm incorporated by the variable target value method. The aim of this work is to study such algorithms and all related concepts, especially convex analysis and nondifferentiable optimization. We analysed the main theoretical differences among the methods and performed numerical experiments with linear and integer problems, in particular, the maximum cut problem on graphs.

Uma maneira de resolver problemas de programação linear de grande escala é explorar a relaxação lagrangeana das restrições \"difíceis\'\' e utilizar métodos de subgradientes. Populares por fornecerem rapidamente boas aproximações de soluções duais, eles não produzem diretamente as soluções primais. Para obtê-las com custo computacional adequado, pode-se construir seqüências ergódicas ou utilizar uma técnica proposta recentemente, denominada algoritmo do volume. As propriedades teóricas de convergência não foram bem estabelecidas nesse algoritmo, mas pequenas modificações permitem a demonstração da convergência dual. Destacam-se como adaptações o algoritmo do volume revisado, um método de feixes específico, e o algoritmo do volume incorporado ao método de variação do alvo. Este trabalho foi baseado no estudo desses algoritmos e de todos os conceitos envolvidos, em especial, análise convexa e otimização não diferenciável. Estudamos as principais diferenças teóricas desses métodos e realizamos comparações numéricas com problemas lineares e lineares inteiros, em particular, o corte máximo em grafos. / One way to solve large-scale linear programming problems is to exploit the Lagrangian relaxation of the difficult constraints and use subgradient methods. Such methods are popular as they give good approximations of dual solutions. Unfortunately, they do not directly yield primal solutions. Two alternatives to obtain primal solutions under reasonable computational cost are the construction of ergodic sequences and the use of the recently developed volume algorithm. While the convergence of ergodic sequences is well understood, the convergence properties of the volume algorithm is not well established in the original paper. This lead to some modifications of the original method to ease the proof of dual convergence. Three alternatives are the revised volume algorithm, a special case of the bundle method, and the volume algorithm incorporated by the variable target value method. The aim of this work is to study such algorithms and all related concepts, especially convex analysis and nondifferentiable optimization. We analysed the main theoretical differences among the methods and performed numerical experiments with linear and integer problems, in particular, the maximum cut problem on graphs.

Cette thèse est composée de deux parties, chacune portant sur un sous-domaine de l'optimisation combinatoire a priori distant de l'autre. Le premier thème de recherche abordé est la programmation biniveau stochastique. Se cachent derrière ce terme deux sujets de recherche relativement peu étudiés conjointement, à savoir d'un côté la programmation stochastique, et de l'autre la programmation biniveau. La programmation mathématique (PM) regroupe un ensemble de méthodes de modélisation et de résolution, pouvant être utilisées pour traiter des problèmes pratiques que se posent des décideurs. La programmation stochastique et la programmation biniveau sont deux sous-domaines de la PM, permettant chacun de modéliser un aspect particulier de ces problèmes pratiques. Nous élaborons un modèle mathématique issu d'un problème appliqué, où les aspects biniveau et stochastique sont tous deux sollicités, puis procédons à une série de transformations du modèle. Une méthode de résolution est proposée pour le PM résultant. Nous démontrons alors théoriquement et vérifions expérimentalement la convergence de cette méthode. Cet algorithme peut être utilisé pour résoudre d'autres programmes biniveaux que celui qui est proposé.Le second thème de recherche de cette thèse s'intitule "problèmes de coupe et de couverture partielles dans les graphes". Les problèmes de coupe et de couverture sont parmi les problèmes de graphe les plus étudiés du point de vue complexité et algorithmique. Nous considérons certains de ces problèmes dans une variante partielle, c'est-à-dire que la propriété de coupe ou de couverture dont il est question doit être vérifiée partiellement, selon un paramètre donné, et non plus complètement comme c'est le cas pour les problèmes originels. Précisément, les problèmes étudiés sont le problème de multicoupe partielle, de coupe multiterminale partielle, et de l'ensemble dominant partiel. Les versions sommets des ces problèmes sont également considérés. Notons que les problèmes en variante partielle généralisent les problèmes non partiels. Nous donnons des algorithmes exacts lorsque cela est possible, prouvons la NP-difficulté de certaines variantes, et fournissons des algorithmes approchés dans des cas assez généraux. / This thesis is composed of two parts, each part belonging to a sub-domain of combinatorial optimization a priori distant from the other. The first research subject is stochastic bilevel programming. This term regroups two research subject rarely studied together, namely stochastic programming on the one hand, and bilevel programming on the other hand. Mathematical Programming (MP) is a set of modelisation and resolution methods, that can be used to tackle practical problems and help take decisions. Stochastic programming and bilevel programming are two sub-domains of MP, each one of them being able to model a specific aspect of these practical problems. Starting from a practical problem, we design a mathematical model where the bilevel and stochastic aspects are used together, then apply a series of transformations to this model. A resolution method is proposed for the resulting MP. We then theoretically prove and numerically verify that this method converges. This algorithm can be used to solve other bilevel programs than the ones we study.The second research subject in this thesis is called "partial cut and cover problems in graphs". Cut and cover problems are among the most studied from the complexity and algorithmical point of view. We consider some of these problems in a partial variant, which means that the cut or cover property that is looked into must be verified partially, according to a given parameter, and not completely, as it was the case with the original problems. More precisely, the problems that we study are the partial multicut, the partial multiterminal cut, and the partial dominating set. Versions of these problems were vertices are