Discontinuous Galerkin methods form a class of numerical methods to find a solution of partial differential equations by combining features of finite element and finite volume methods. Methods are defined using a weak form of a particular model problem, allowing for discontinuities in the discrete trial and test spaces. Using a discontinuous discrete space mesh provides proper flexibility and a compact discretisation pattern, allowing a multidomain and multiphysics simulation. Discontinuous Galerkin methods with a higher approximation polynomial order, the socalled p-version, performs better in terms of convergence rate, compared with the low order h-version with smaller element sizes and bigger mesh. However, the condition number of the Galerkin system grows subsequently. This causes surge in the amount of required storage, computational complexity and in the time required for computation. We use the following three approaches to keep the advantages and eliminate the disadvantages. The first approach will be a specific choice of basis functions which we call C1 polynomials. These ensure that the majority of integrals over the edge of the mesh elements disappears. This reduces the total number of non-zero elements in the resulting system. This decreases the computational complexity without loss in precision. This approach does not affect the number of iterations required by chosen Conjugate Gradients method when compared to the other choice of basis functions. It actually decreases the total number of algebraic operations performed. The second approach is the introduction of suitable preconditioners. In our case, the Additive two-layer Schwarz method, developed in [4], for the iterative Conjugate Gradients method is considered. This directly affects the spectral condition number of the system matrix and decreases the number of iterations required for the computation. This approach, however, increases the total number of algebraic operations and might require more operational time. To tackle the rise in the number of algebraic operations, we introduced a modified Additive two-layer non-overlapping Schwarz method with a Multigrid process. This using a fixed low-order approximation polynomial degree on a coarse grid. We show that this approach is spectrally equivalent to the first preconditioner, and requires less time for computation. The third approach is a development of an efficient mathematical framework for distributed data structure. This allows a high performance, massively parallel, implementation of the discontinuous Galerkin method. We demonstrate that it is possible to exploit properties of the system matrix and C1 polynomials as basis functions to optimize the parallel structures. The previously mentioned parallel data structure allows us to parallelize at the same time both the matrix-vector multiplication routines for the Conjugate Gradients method, as well as the preconditioner routines on the solver level. This minimizes the transfer ratio amongst the distributed system. Finally, we combined all three approaches and created a framework, which allowed us to successfully implement all of the above.

La prévention des risques industriels nécessite de simuler la dispersion turbulente de polluants. Cependant, les outils majoritairement utilisés à ce jour ne permettent pas de traiter les champs proches dans le cas de géométries complexes, et il est nécessaire d'utiliser les outils de CFD (“ Computational Fluid Dynamics ”) plus adaptés, mais plus coûteux. Afin de simuler les écoulements atmosphériques avec dispersion de polluants, les modèles CFD doivent modéliser correctement d'une part, les effets de flottabilité, et d'autre part les effets de la turbulence. Plusieurs approches existent, notamment dans la prise en compte des effets de flottabilité et la modélisation de la turbulence, et nécessitent des méthodes numériques adaptées aux spécificités mathématiques de chacune d'entre elles, ainsi que des schémas numériques précis pour ne pas polluer la modélisation. Une formulation d'ordre élevé en volumes finis, sur maillages non structurés, parallélisée, est proposée pour simuler les écoulements atmosphériques avec dispersion de polluants. L'utilisation de schémas d'ordre élevé doit permettre d'une part de réduire le nombre de cellules et diminuer les temps de simulation pour atteindre une précision donnée, et d'autre part de mieux contrôler la viscosité numérique des schémas en vue de simulations LES (Large Eddy Simulation), pour lesquelles la viscosité numérique des schémas peut masquer les effets de la modélisation. Deux schémas d'ordre élevé ont été étudiés et implémentés dans un solveur 3D Navier Stokes incompressible sur des maillages volumes finis non structurés. Nous avons développé un premier schéma d'ordre élevé, correspondant à un schéma Padé volumes finis, et nous avons étendu le schéma de reconstruction polynomiale de Carpentier (2000) aux écoulements incompressibles. Les propriétés numériques des différents schémas implémentés dans le même code de calcul sont étudiées sur différents cas tests bi-dimensionnels (calcul de flux convectifs et diffusifs sur une solution a-priori, convection d'une tâche gaussienne, décroissance d'un vortex de Taylor et cavité entraînée) et tri-dimensionnel (écoulement autour d'un obstacle cubique). Une attention particulière a été portée à l'étude de la précision et du traitement des conditions limites. L'implémentation proposée du schéma polynomial permet d'approcher, pour un maillage identique, les temps de simulation obtenus avec un schéma décentré classique d'ordre 2, mais avec une précision supérieure. Le schéma compact donne la meilleure précision. En utilisant une méthode de Jacobi sans calcul implicite de la matrice pour calculer le gradient, le temps de simulation devient intéressant uniquement lorsque la précision requise est importante. Une alternative est la résolution du système linéaire par une méthode multigrille algébrique. Cette méthode diminue considérablement le temps de calcul du gradient et le schéma Padé devient performant même pour des maillages grossiers. Enfin, pour réduire les temps de simulation, la parallélisation des schémas d'ordre élevé est réalisée par une décomposition en sous domaines. L'assemblage des flux s'effectue naturellement et différents solveurs proposés par les librairies PETSC et HYPRE (solveur multigrille algébrique et méthode de Krylov préconditionnée) permettent de résoudre les systèmes linéaires issus de notre problème. / The prevention of industrial risks requires simulating turbulent dispersion of pollutants. However, the tools mostly used so far do not allow near fields treated in the case of complex geometries, and it is necessary to utilize the tools of CFD (Computational Fluid Dynamics ") more suitable but more expensive. To simulate atmospheric flows with dispersion of pollutants, the CFD models must correctly model the one hand, the effects of buoyancy, and secondly the effects of turbulence. Several approaches exist, including taking into account the effects of buoyancy and turbulence modeling, and require numerical methods adapted to the specific mathematics of each, and accurate numerical schemes to avoid pollution modeling. A formulation of high order finite volume on unstructured meshes, parallelized, is proposed to simulate the atmospheric flows with dispersion of pollutants. The use of high order schemes allow one hand to reduce the number of cells and decrease the simulation time to achieve a given accuracy, and secondly to better control the viscosity numerical schemes for simulation LES (Large Eddy Simulation), for which the numerical viscosity patterns may mask the effects of modeling. Two high-order schemes have been studied and implemented in a 3D Navier Stokes solver on unstructured mesh finite volume. We developed the first high-order scheme, corresponding to a Padé finite volume scheme, and we have extended the scheme of reconstruction polynomial Carpentier (2000) for incompressible flows. The numerical properties of the various schemes implemented in the same computer code are studied different two-dimensional test cases (calculation of diffusive and convective flow on a solution a priori, a task Gaussian convection, decay of a vortex of Taylor and driven cavity) and tri-dimensional (flow past an obstacle cubic). Particular attention has been paid to the study of the accuracy and treatment of boundary conditions. The implementation of the polynomial allows to obtain quasi identical simulation time compared to a classical upwind scheme of order 2, but with higher accuracy. The compact layout gives the best accuracy. Using a Jacobi method without calculation implied matrix to calculate the gradient, the simulation time becomes interesting only when the required accuracy is important. An alternative is the resolution of linear system by an algebraic multigrid method. This method significantly reduces the computation time of the gradient and the Padé scheme is effective even for coarse meshes. Finally, to reduce simulation time, the parallelization schemes of high order is achieved by a decomposition into subdomains. The assembly flow occurs naturally and different solvers provided by PETSc libraries and HYORE (algebraic multigrid solver and preconditioned Krylov method) used to solve linear systems from our problem. The work was to identify and determine the parameters that lead to lowest time resolution simulation. Various tests of speed-up and scale-up were used to determine the most effective and optimal parameters for solving linear systems in parallel from our problem. The results of this work have been the subject of a communication in an international conference "Parallel CFD 2008" and an article submitted to "International Journal for Numerical Methods in Fluids" (Analysis of high-order finite volume schemes for the incompressible Navier Stokes equations)