This thesis is concerned with the development of novel numerical methods for solving nondifferentiable convex optimization problems in real Hilbert spaces and with the investigation of their asymptotic behavior. To this end, we are also making use of monotone operator theory as some of the provided algorithms are originally designed to solve monotone inclusion problems. After introducing basic notations and preliminary results in convex analysis, we derive two numerical methods based on different smoothing strategies for solving nondifferentiable convex optimization problems. The first approach, known as the double smoothing technique, solves the optimization problem with some given a priori accuracy by applying two regularizations to its conjugate dual problem. A special fast gradient method then solves the regularized dual problem such that an approximate primal solution can be reconstructed from it. The second approach affects the primal optimization problem directly by applying a single regularization to it and is capable of using variable smoothing parameters which lead to a more accurate approximation of the original problem as the iteration counter increases. We then derive and investigate different primal-dual methods in real Hilbert spaces. In general, one considerable advantage of primal-dual algorithms is that they are providing a complete splitting philosophy in that the resolvents, which arise in the iterative process, are only taken separately from each maximally monotone operator occurring in the problem description. We firstly analyze the forward-backward-forward algorithm of Combettes and Pesquet in terms of its convergence rate for the objective of a nondifferentiable convex optimization problem. Additionally, we propose accelerations of this method under the additional assumption that certain monotone operators occurring in the problem formulation are strongly monotone. Subsequently, we derive two Douglas–Rachford type primal-dual methods for solving monotone inclusion problems involving finite sums of linearly composed parallel sum type monotone operators. To prove their asymptotic convergence, we use a common product Hilbert space strategy by reformulating the corresponding inclusion problem reasonably such that the Douglas–Rachford algorithm can be applied to it. Finally, we propose two primal-dual algorithms relying on forward-backward and forward-backward-forward approaches for solving monotone inclusion problems involving parallel sums of linearly composed monotone operators. The last part of this thesis deals with different numerical experiments where we intend to compare our methods against algorithms from the literature. The problems which arise in this part are manifold and they reflect the importance of this field of research as convex optimization problems appear in lots of applications of interest.

Cette thèse traite de la théorie du risque en finance et en assurance. La mise en pratique du concept de comonotonie, la dépendance du risque au sens fort, est décrite pour identifier l’optimum de Pareto et les allocations individuellement rationnelles Pareto optimales, la tarification des options et la quantification des risques. De plus, il est démontré que l’aversion au risque monotone à gauche, un raffinement pertinent de l’aversion forte au risque, caractérise tout décideur à la Yaari, pour qui, l’assurance avec franchise est optimale. Le concept de comonotonie est introduit et discuté dans le chapitre 1. Dans le cas de risques multiples, on adopte l’idée qu’une forme naturelle pour les compagnies d’assurance de partager les risques est la Pareto optimalité risque par risque. De plus, l’optimum de Pareto et les allocations individuelles Pareto optimales sont caractérisées. Le chapitre 2 étudie l’application du concept de comonotonie dans la tarification des options et la quantification des risques. Une nouvelle variable de contrôle de la méthode de Monte Carlo est introduite et appliquée aux “basket options”, aux options asiatiques et à la TVaR. Finalement dans le chapitre 3, l’aversion au risque au sens fort est raffinée par l’introduction de l’aversion au risque monotone à gauche qui caractérise l’optimalité de l’assurance avec franchise dans le modèle de Yaari. De plus, il est montré que le calcul de la franchise s’effectue aisément. / This thesis deals with the risk theory in Finance and Insurance. Application of the Comonotonicity concept, the strongest risk dependence, is described for identifying the Pareto optima and Individually Rational Pareto optima allocations, option pricing and quantification of risk. Furthermore it is shown that the left monotone risk aversion, a meaningful refinement of strong risk aversion, characterizes Yaari’s decision makers for whom deductible insurance is optimal. The concept of Comonotonicity is introduced and discussed in Chapter 1. In case of multiple risks, the idea that a natural way for insurance companies to optimally share risks is risk by risk Pareto-optimality is adopted. Moreover, the Pareto optimal and individually Pareto optimal allocations are characterized. The Chapter 2 investigates the application of the Comonotonicity concept in option pricing and quantification of risk. A novel control variate Monte Carlo method is introduced and its application is explained for basket options, Asian options and TVaR. Finally in Chapter 3 the strong risk aversion is refined by introducing the left-monotone risk aversion which characterizes the optimality of deductible insurance within the Yaari’s model. More importantly, it is shown that the computation of the deductible is tractable.

Cette thèse est centrée autour de l’étude théorique et de l’analyse numérique des équations paraboliques non linéaires avec divers conditions aux limites. La première partie est consacrée aux équations paraboliques dégénérées mêlant des phénomènes non-linéaires de diffusion et de transport. Nous définissons des notions de solutions entropiques adaptées pour chacune des conditions aux limites (flux nul, Robin, Dirichlet). La difficulté principale dans l’étude de ces problèmes est due au manque de régularité du flux pariétal pour traiter les termes de bords. Ceci pose un problème pour la preuve d’unicité. Pour y remédier, nous tirons profit du fait que ces résultats de régularités sur le bord sont plus faciles à obtenir pour le problème stationnaire et particulièrement en dimension un d’espace. Ainsi par la méthode de comparaison "fort-faible" nous arrivons à déduire l’unicité avec le choix d’une fonction test non symétrique et en utilisant la théorie des semi-groupes non linéaires.L’existence de solution se démontre en deux étapes, combinant la méthode de régularisation parabolique et les approximations de Galerkin. Nous développons ensuite une approche directe en construisant des solutions approchées par un schéma de volumes finis implicite en temps. Dans les deux cas, on combine les estimations dans les espaces fonctionnels bien choisis avec des arguments de compacité faible ou forte et diverses astuces permettant de passer à la limite dans des termes non linéaires. Notamment, nous introduisons une nouvelle notion de solution appelée solution processus intégrale dont l’objectif, dans le cadre de notre étude, est de pallier à la difficulté de prouver la convergence vers une solution entropique d’un schéma volumes finis pour le problème de flux nul au bord.La deuxième partie de cette thèse traite d’un problème à frontière libre décrivant la propagation d’un front de combustion et l’évolution de la température dans un milieu hétérogène. Il s’agit d’un système d’équations couplées constitué de l’équation de la chaleur bidimensionnelle et d’une équation de type Hamilton-Jacobi. L’objectif de cette partie est de construire un schéma numérique pour ce problème en combinant des discrétisations du type éléments finis avec les différences finies. Ceci nous permet notamment de vérifier la convergence de la solution numérique vers une solution onde pour un temps long. Dans un premier temps, nous nous intéressons à l’étude d’un problème unidimensionnel. Très vite,nous nous heurtons à un problème de stabilité du schéma. Cela est dû au problème de prise en compte de la condition de Neumann au bord. Par une technique de changement d’inconnue et d’approximation nous remédions à ce problème. Ensuite, nous adaptons cette technique pour la résolution du problème bidimensionnel. A l’aide d’un changement de variables, nous obtenons un domaine fixe facile pour la discrétisation. La monotonie du schéma obtenu est prouvée sous une hypothèse supplémentaire de propagation monotone qui exige que la frontière libre se déplace dans les directions d’un cône prescrit à l’avance. / This thesis focuses on the theoretical study and numerical analysis of parabolic equations with boundary conditions.The first part is devoted to degenerate parabolic equation which combines features of a hyperbolic conser-vation law with those of a porous medium equation. We define suitable notions of entropy solutions foreach of the boundary conditions (zero-flux, Robin, Dirichlet). The main difficulty in these studies residesin the formulation of the adequate notion of entropy solution and in the proof of uniqueness. There isa technical difficulty due to the lack of regularity required to treat the boundaries terms. We take ad-vantage of the fact that boundary regularity results are easier to obtain for the stationary problem, inparticular in one space dimension. Thus, using strong-weak uniqueness approach we get the uniquenesswith the choice of a non-symmetric test function and using the nonlinear semigroup theory. The exis-tence of solution is proved in two steps, combining the method of parabolic regularization and Galerkinapproximations. Next, we develop a direct approach to construct approximate solutions by an implicitfinite volume scheme. In both cases, the estimates in the appropriately chosen functional spaces are com-bined with arguments of weak or strong compactness and various tricks to pass to the limit in nonlinearterms. In the appendix, we propose a result of existence of strong trace of a solution for the degenerateparabolic problem. In another appendix of independent interest, we introduce a new concept of solutioncalled integral process solution. We exploit it to overcome the difficulty of proving the convergence ofour finite volume scheme to an entropy solution for the zero-flux boundary problem.The second part of this thesis deals with a free boundary problem describing the propagation of a com-bustion front and the evolution of the temperature in a heterogeneous medium. So we have a coupledproblem consisting of the heat equation of bidimensional space and a Hamilton-Jacobi equation. The ob-jective is to construct a numerical scheme and to verify that the numerical solution converges to a wavesolution for a long time. Recall that an existence of wave solution for this problem was already proven inan analytical framework. At first, we focus on the study of a one-dimensional problem. Here, we face aproblem of stability of the scheme. This is due to a difficulty of taking into account the Neumann boun-dary condition. Through a technique of change of unknown, we can propose a monotone scheme. Wealso adapt this technique for solving two-dimensional problem. Using a change of variables, we obtaina fixed domain where the discretization becomes easy. The monotony of the scheme is proved under anadditional assumption of monotone propagation that requires the free boundary moves in the directionsof a cone given beforehand.

This thesis is concerned with the development of novel numerical methods for solving nondifferentiable convex optimization problems in real Hilbert spaces and with the investigation of their asymptotic behavior. To this end, we are also making use of monotone operator theory as some of the provided algorithms are originally designed to solve monotone inclusion problems. After introducing basic notations and preliminary results in convex analysis, we derive two numerical methods based on different smoothing strategies for solving nondifferentiable convex optimization problems. The first approach, known as the double smoothing technique, solves the optimization problem with some given a priori accuracy by applying two regularizations to its conjugate dual problem. A special fast gradient method then solves the regularized dual problem such that an approximate primal solution can be reconstructed from it. The second approach affects the primal optimization problem directly by applying a single regularization to it and is capable of using variable smoothing parameters which lead to a more accurate approximation of the original problem as the iteration counter increases. We then derive and investigate different primal-dual methods in real Hilbert spaces. In general, one considerable advantage of primal-dual algorithms is that they are providing a complete splitting philosophy in that the resolvents, which arise in the iterative process, are only taken separately from each maximally monotone operator occurring in the problem description. We firstly analyze the forward-backward-forward algorithm of Combettes and Pesquet in terms of its convergence rate for the objective of a nondifferentiable convex optimization problem. Additionally, we propose accelerations of this method under the additional assumption that certain monotone operators occurring in the problem formulation are strongly monotone. Subsequently, we derive two Douglas–Rachford type primal-dual methods for solving monotone inclusion problems involving finite sums of linearly composed parallel sum type monotone operators. To prove their asymptotic convergence, we use a common product Hilbert space strategy by reformulating the corresponding inclusion problem reasonably such that the Douglas–Rachford algorithm can be applied to it. Finally, we propose two primal-dual algorithms relying on forward-backward and forward-backward-forward approaches for solving monotone inclusion problems involving parallel sums of linearly composed monotone operators. The last part of this thesis deals with different numerical experiments where we intend to compare our methods against algorithms from the literature. The problems which arise in this part are manifold and they reflect the importance of this field of research as convex optimization problems appear in lots of applications of interest.

In this study, four frequently used yield curve construction methods are evaulated on a set of metrics with the aim of determining which method is the most suitable for estimating yield curves from European zero rates. The included curve construction methods are Nelson-Siegel, Nelson-Siegel-Svensson, cubic spline interpolation and forward monotone convex spline interpolation. We let the methods construct yield curves on multiple sets of zero yields with different origins. It is found that while the interpolation methods show greater ability to adapt to variable market conditions as well as hedge arbitrary fixed income claims, they are outperformed by the parametric methods regarding the smoothness of the resulting yield curve as well as their sensitivity to noise and perturbations in the input rates. This apart from the Nelson-Siegel method's problem of capturing the behavior of underlying rates with a high curvature. The Nelson-Siegel-Svensson method did also exhibit instability issues when exposed to perturbations in the input rates. The Nelson-Siegel method and the forward monotone convex spline interpolation method emerge as most favorable in their respective categories. The ultimate selection between the two methods must however take the application at hand into consideration due to their fundamentally different characteristics. / I denna studie utvärderas fyra välanvända metode för att konstruera yieldkurvor på ett antal punkter. Detta med syfte att utröna vilken metod som är bäst lämpad för att estimera yieldkurvor på Europeiska nollkupongräntor. Metoderna som utvärderas är Nelson-Siegel, Nelson-Siegel-Svensson, cubic spline-interpolering samt forward monotone convex spline-interpolering. Vi låter metoderna estimera yieldkurvor på flera sammansättningar nollkupongräntor med olika ursprung. Vi ser att interpoleringsmetoderna uppvisar en större flexibilitet vad gäller att anpassa sig till förändrade marknadsförutsättningar samt att replikera godtyckliga ränteportföljer. När det gäller jämnhet av yieldkurvan och känsligheten för brus och störningar i de marknadsräntor som kurvan konstrueras utifrån så presterar de parametiska metoderna däremot avsevärt bättre. Detta bortsett från att Nelson-Siegel-metoden hade problem att fånga beteendet hos nollkupongräntor med hög kurvatur. Vidare hade Nelson-Siegel-Svensson-metoden problem med instabilitet när de underliggande marknadsrentorna utsattes för störningar. Nelson-Siegen-metoden samt foward monotone convex spline-interpolering visade sig vara bäst lämpade för att konstruera yieldkurvor på de Europeiska marknaderna av de utvärderade metoderna. Vilken metod av de två som slutligen bör användas behöver bedömas från fall till fall grundat i vilken tillämpning som avses.

This thesis presents the construction, the analysis and the verication of compact residual discretizations for the solution of conservation laws on unstructured meshes. The schemes considered belong to the class of residual distribution (RD) or fluctuation splitting (FS) schemes. The methodology presented relies on three main elements: design of compact linear first-order stable schemes for linear hyperbolic PDEs, a positivity preserving procedure mapping stable first-order linear schemes onto nonlinear second-order schemes with non-oscillatory shock capturing capabilities, and a conservative formulation enabling to extend the schemes to nonlinear CLs. These three design steps, and the underlying theoretical tools, are discussed in depth. The nonlinear RD schemes resulting from this construction are tested on a large set of problems involving the solution of scalar models, and systems of CLs. This extensive verification fills the gaps left open, where no theoretical analysis is possible. Numerical results are presented on the Euler equations of a perfect gas, on a two-phase flow model with highly nonlinear thermodynamics, and on the shallow-water equations. On irregular grids, the schemes proposed yield quite accurate and stable solutions even on very difficult computations. Direct comparisone show that these results are more accurate than the ones given by FV and WENO schemes. Moreover, our schemes have a compact nearest-neighbor stencil. This encourages to further develop our approach, toward the design of very high-order schemes, which would represent a very appealing alternative, both in terms of accuracy and efficiency, to now classical FV and ENO/WENO discretizations. These schemes might also be very competitive with respect to very high-order DG schemes.

Ce mémoire porte sur la simulation d'intervalles de crédibilité simultanés dans un contexte bayésien. Dans un premier temps, nous nous intéresserons à des données de précipitations et des fonctions basées sur ces données : la fonction de répartition empirique et la période de retour, une fonction non linéaire de la fonction de répartition. Nous exposerons différentes méthodes déjà connues pour obtenir des intervalles de confiance simultanés sur ces fonctions à l'aide d'une base polynomiale et nous présenterons une méthode de simulation d'intervalles de crédibilité simultanés. Nous nous placerons ensuite dans un contexte bayésien en explorant différents modèles de densité a priori. Pour le modèle le plus complexe, nous aurons besoin d'utiliser la simulation Monte-Carlo pour obtenir les intervalles de crédibilité simultanés a posteriori. Finalement, nous utiliserons une base non linéaire faisant appel à la transformation angulaire et aux splines monotones pour obtenir un intervalle de crédibilité simultané valide pour la période de retour. / This master's thesis addresses the problem of the simulation of simultaneous credible intervals in a Bayesian context. First, we will study precipation data and two functions based on these data : the empirical distribution function and the return period, a non-linear function of the empirical distribution. We will review different methods already known to obtain simultaneous confidence intervals of these functions with a polynomial basis and we will present a method to simulate simultaneous credible intervals. Second, we will explore some models of prior distributions and in the more complex one, we will need the Monte-Carlo method to simulate simultaneous posterior credible intervals. Finally, we will use a non-linear basis based on the angular transformation and on monotone splines to obtain valid simultaneous credible intervals for the return period.

Dans cette thèse, on s'intéresse à l'étude d'existence de solutions pour les processus de rafle. Ce problème prend la forme d'une inclusion différentielle contrainte avec des cônes normaux qui apparaissent naturellement dans nombreuses applications telles que le mouvement de foule, l'élastoplasticité, les mécaniques, les circuits électroniques, etc. L'objective de ce travail est de rapprocher deux importantes classes d'inclusions différentielles. D'une part, nous établissons quelques résultats d'existence de tube-solutions pour des processus de rafle à des ensembles uniformément prox-réguliers. D'autre part, nous présentons des résultats d'existence de solutions monotone par rapport à un préordre pour un système mixte d'inclusions différentielles projetées. De plus, nous montrons l'existence d'un point-selle pour notre système et nous fournissons deux exemples d'applications. / In this thesis, we were interested in the study of the existence of solutions for sweeping processes. This problem takes the form of a constrained differential inclusion involving normal cones which appears naturally in many applications such as crowd motion, elastoplasticity, mechanics, electrical circuit, etc.The aim of this work is to bring together two classes of differential inclusions. On one hand, we establish some existence results of solutions-tube for sweeping processes with uniformly prox-regular sets. On the other hand, we present existence results of monotone solutions with respect to a preorder for a mixed system of projected differential inclusions. In addition, we show that our system has a saddle-point and we provide two examples of applications.

Distributed optimization over multi-agent networks has become an increasingly popular research topic as it incorporates many applications from various areas such as consensus optimization, distributed control, network resource allocation, large scale machine learning, etc. Parallel distributed solution algorithms are highly desirable as they are more scalable, more robust against agent failure, align more naturally with either underlying agent network topology or big-data parallel computing framework. In this dissertation, we consider a multi-agent optimization formulation where the global objective function is the summation of individual local objective functions with respect to local agents' decision variables of different dimensions, and the constraints include both local private constraints and shared coupling constraints. Employing and extending tools from the monotone operator theory (including resolvent operator, operator splitting, etc.) and fixed point iteration of nonexpansive, averaged operators, a series of distributed solution approaches are proposed, which are all iterative algorithms that rely on parallel agent level local updates and inter-agent coordination. Some of the algorithms require synchronizations across all agents for information exchange during each iteration while others allow asynchrony and delays. The algorithms' convergence to an optimal solution if one exists are established by first characterizing them as fixed point iterations of certain averaged operators under certain carefully designed norms, then showing that the fixed point sets of these averaged operators are exactly the optimal solution set of the original multi-agent optimization problem. The effectiveness and performances of the proposed algorithms are demonstrated and compared through several numerical examples.<br>

Neste trabalho consideramos três tópicos principais. Nos dois primeiros generalizamos alguns dos resultados clássicos da Teoria da Confiabilidade na otimização dos procedimentos de burn-in e de políticas de garantia, respectivamente, sob o modelo de tempo de vida geral, quando um sistema coerente é observado ao nível de seus componentes, e estendemos os conceitos de intensidade de falha na forma de banheira e do modelo de falha geral através da definiçâo de processos progressivamente mensuráveis sob a pré-t-história completa dos componentes do sistema. Uma regra de parada monótona é usada na metodologia de otimizaçâo proposta. No terceiro tópico modelamos os custos de garantia descontados por reparo mínimo de um sistema coerente ao nível de seus componentes, propomos o estimador martingal do custo esperado para um período de garantia fixado e provamos as suas propriedades assintóticas mediante o Teorema do Limite Central para Martingais. / In this work we consider three main topics. In the first two, we generalize some classical results on Reliability Theory related to the optimization in burn-in procedures and warranty policies, using the general lifetime model of a coherent system observed on the component level and extending the definitions of bathtub shaped failure rate and general failure model to progressively measurable processes under the complete pre-t-history. A monotone stopping rule is applied within the proposed methodology. In the third topic, we define the discounted warranty cost process for a coherent system minimally repaired on the component level and we propose a martingale estimator to the expected warranty cost for a fixed period and setting its asymptotic properties by means of Martingale Central Limit Theorem.