We obtain fully computable constant free a posteriori error bounds on the broken energy seminorm of the error in nonconforming and discontinuous Galerkin finite element approximations of a linear second ore elliptic problem on meshes omprised of triangular elements. We do this for nonconforming finite element approximations of uniform arbitrary order as well as for non-uniform order symmetric interior penalty Galerkin, non-symmetric interior penalty Galerkin and ncomplete interior penalty Galerkin finite element approximations.

This thesis describes the formulation and application of an adaptive multigrid method for the efficient solution of nonlinear elliptic and parabolic partial differential equations and systems. A continuous Galerkin finite-element method is combined with locally adaptive mesh refinement and an optimal multigrid solver to achieve this efficiency. The novel contribution of this work lies in the manner in which these two techniques are combined. In particular the multigrid solver provides a natural and simple method of handling grid points that are not fully connected, so called hanging nodes. This allows for a straightforward adaptive gridding scheme that does not need to take any special measures to repair these hanging nodes for a standard element-by-element implementation of the finite element assembly process. Specifically, on each element, only the usual finite element basis functions are required, even in the vicinity of hanging nodes. Furthermore the standard multigrid full approximation scheme (FAS) may be applied with only minor modifications to account for the presence of the hanging nodes. A wide cross-section of nonlinear elliptic and parabolic problems are used to demonstrate the performance of the proposed algorithm, which is shown to provide optimal accuracy at an optimal computational cost.

La modélisation des réacteurs nucléaires repose sur la résolution de l'équation de Boltzmann linéaire. Pour la résolution spatiale de la forme stationnaire de cette équation, le solveur MINARET utilise la méthode des éléments finis discontinus sur un maillage triangulaire non structuré afin de pouvoir traiter des géométries complexes. Cependant, l'utilisation d'arêtes droites introduit une approximation de la géométrie. Autoriser l'existence d'arêtes courbes permet de coller parfaitement à la géométrie, et dans certains cas de diminuer le nombre de triangles du maillage. L'objectif principal de cette thèse est l'étude d'éléments finis sur des triangles possédant plusieurs bords courbes. Le choix des fonctions de base est un des points importants pour ce type d'éléments finis. Un résultat de convergence a été obtenu sous réserve que les triangles courbes ne soient pas trop éloignés des triangles droits associés. D'autre part, un solveur courbe a été développé pour traiter des triangles avec plusieurs bords courbes. Une autre partie de ce travail porte sur l'accélération de la convergence des calculs. En effet, la résolution du problème est itérative et peut converger très lentement. Une méthode d'accélération dite DSA (Diffusion Synthetic Acceleration) permet de diminuer le nombre d'itérations et le temps de calcul. L'opérateur de diffusion est utilisé comme un préconditionneur de l'opérateur de transport. La DSA a été mise en oeuvre en utilisant une technique issue des méthodes de pénalisation intérieure. Une analyse de Fourier en 1D et 2D permet de vérifier la stabilité du schéma pour des milieux périodiques avec de fortes hétérogénéités / To model the nuclear reactors, the stationnary linear Boltzmann equation is solved. After discretising the energy and the angular variables, the hyperbolic equation is numerically solved with the discontinuous finite element method. The MINARET code uses this method on a triangular unstructured mesh in order to deal with complex geometries (like containing arcs of circle). However, the meshes with straight edges only approximate such geometries. With curved edges, the mesh fits exactly to the geometry, and in some cases, the number of triangles decreases. The main task of this work is the study of finite elements on curved triangles with one or several curved edges. The choice of the basis functions is one of the main points for this kind of finite elements. We obtained a convergence result under the assumption that the curved triangles are not too deformed in comparison with the associated straight triangles. Furthermore, a code has been written to treat triangles with one, two or three curved edges. Another part of this work deals with the acceleration of transport calculations. Indeed, the problem is solved iteratively, and, in some cases, can converge really slowly. A DSA (Diffusion Synthetic Acceleration) method has been implemented using a technique from interior penalty methods. A Fourier analysis in 1D and 2D allows to estimate the acceleration for infinite periodical media, and to check the stability of the numerical scheme when strong heterogeneities exist

De nos jours, les centrales nucléaires produisent de l'énergie par des réactions de fission (division d'un noyau atomique lourd en plusieurs noyaux atomiques légers et neutrons). Une alternative serait d'utiliser plutôt la réaction de fusion de noyaux légers de deutérium et de tritium, isotopes de l'hydrogène. Toutefois, cette technique reste encore du domaine de la recherche en physique des plasmas. Les expériences effectuées dans ce domaine ont révélé que les réacteurs à configuration magnétique toroïdale, dite tokamak, sont les plus efficaces. Un mélange gazeux d'isotopes de l'hydrogène appelé plasma est confiné grâce à un champ magnétique produit par des bobines. Ce plasma doit être chauffé à une température très élevée afin que les réactions de fusion aboutissent. De même, un courant intense doit être maintenu dans le plasma afin d'obtenir une configuration magnétique qui permet de le confiner. Une des méthodes les plus attrayantes parmi les techniques connues pour générer du courant est basée sur l'injection d'ondes électromagnétiques dans le plasma à la fréquence proche de la résonance hybride. Cette méthode offre la possibilité de contrôler le profil de densité dans le plasma. Une analyse de type Full-Wave permet alors de modéliser la propagation et l'absorption de l'onde hybride à partir des équations de Maxwell. Le but de cette thèse est de développer une méthode numérique pour cette simulation Full-Wave. Le chapitre 2 présente les équations de propagation d'ondes en mettant en évidence les caractères physiques du plasma. Une approche variationnelle de type mixte augmentée est développée et une analyse mathématique de cette dernière est effectuée dans le chapitre 3. Dans le contexte de la géométrie d'un tokamak, le problème Full-Wave dépendant de trois paramètres peut être réduit en une série de problèmes à deux variables à l'aide de la transformation de Fourier, ce sera l'objet du chapitre 4. Dans le chapitre 5, la formulation variationnelle obtenue à partir du problème mode par mode est discrétisée en utilisant des éléments finis nodaux de type Taylor-Hood. Le chapitre 6 concerne les méthodes de résolution du système linéaire après discrétisation. À l'aide de différents diagnostics physiques présentés dans le chapitre 7, des résultats de la simulation Full-Wave obtenues à partir d'un code MATLAB sont présentées dans le chapitre 8. Enfin, dans le but de développer une version parallèle de la simulation, le chapitre 9 est consacré à une méthode de décomposition de domaine sans recouvrement associé au système Full-Wave. / In order to generate current in tokamak, we look at plasma heating by electromagnetic waves at the lower hybrid (LH) frequency. For this type of description, one use a ray tracing code but we consider a full-wave one, where dielectric properties are local.Our aim is to develop a finite element numerical method for the full-wave modeling and to apply a domain decomposition method. In this thesis, we have developped a finite element method in a cross section of the tokamak for Maxwell equations solving the time harmonic electric field and a nonoverlapping domain decom- position method for the mixed augmented variational formulation by taking continuity accross the interfaces as constraints

Cette thèse porte sur le développement d’estimateurs d'erreur a posteriori pour la résolution numérique par éléments finis de problèmes en électromagnétisme basse fréquence. On s’intéresse aux formulations en potentiels (A-φ et T-Ω) des équations de Maxwell en régime quasi-stationnaire, pour le cas harmonique ou temporel. L'enjeu consiste à développer des outils numériques mathématiquement robustes, exploitables dans un code de calcul industriel, notamment le Code_Carmel3D (EDF R&D), permettant d'estimer l'erreur de discrétisation spatio-temporelle et de pouvoir ainsi améliorer la précision des calculs. On prouve la fiabilité, assurant le contrôle de l’erreur. On prouve également dans certains cas l’efficacité locale, permettant de repérer les zones du maillage dans lesquelles l’erreur est la plus importante, et de mettre ainsi en œuvre des stratégies de raffinement adaptatif. L'équivalence globale entre l'erreur en norme énergétique et l'estimateur est en général assurée. Les estimateurs obtenus sont finalement utilisés pour des simulations physiques/industrielles par le Code_Carmel3D. / This thesis focus on the developement of a posteriori error estimators for the finite element numerical resolution of low frequency electromagnetic problems. We are interested in two potential formulations of the Maxwell's equations in the quasi-static approximation, known as A-φ et T-Ω formulations, for both harmonic and temporal regimes. The challenge consists in developing numerical tools mathematically robust, usable in an industrial code allowing the estimation of the spatio-temporal error discretisation and the improvement of the quality and the cost of the computation. We prove the reliability of the proposed error estimators, which ensures an upper bound for the error in the energy norm. In some cases we also prove the local efficicency of the estimators, which allows to detect the zones where the error is the highest, so that an adaptive remeshing process can be set up. Anyway, the global equivalence between the energy error norm and the estimator is derived. The developed error estimators are finally used for physical and industrial numerical simulations in Code_Carmel3D (EDF R&D).

The fractal-like finite element method (FFEM) is developed to compute stress intensity factors (SIFs) for isotropic homogeneous and bi-material V-notched plates. The method is semi-analytical, because analytical expressions of the displacement fields are used as global interpolation functions (GIFs) to carry out a transformation of the nodal displacements within a singular region to a small set of generalised coordinates. The concept of the GIFs in reducing the number of unknowns is similar to the concept of the local interpolation functions of a finite element. Therefore, the singularity at a notch-tip is modelled accurately in the FFEM using a few unknowns, leading to reduction of the computational cost.The analytical expressions of displacements and stresses around a notch tip are derived for different cases of notch problems: in-plane (modes I and II) conditions and out-of-plane (mode III) conditions for isotropic and bi-material notches. These expressions, which are eigenfunction series expansions, are then incorporated into the FFEM to carry out the transformation of the displacements of the singular nodes and to compute the notch SIFs directly without the need for post-processing. Different numerical examples of notch problems are presented and results are compared to available published results and solutions obtained by using other numerical methods.A strain energy approach (SEA) is also developed to extract the notch SIFs from finite element (FE) solutions. The approach is based on the strain energy of a control volume around the notch-tip. The strain energy may be computed using commercial FE packages, which are only capable of computing SIFs for crack problems and not for notch problems. Therefore, this approach is a strong tool for enabling analysts to compute notch SIFs using current commercial FE packages. This approach is developed for comparison of the FFEM results for notch problems where available published results are scarce especially for the bi-material notch cases.A very good agreement between the SEA results and the FFEM results is illustrated. In addition, the accuracy of the results of both procedures is shown to be very good compared to the available results in the literature. Therefore, the FFEM as a stand-alone procedure and the SEA as a post-processing technique, developed in this research, are proved to be very accurate and reliable numerical tools for computing the SIFs of a general notch in isotropic homogeneous and bi-material plates.

This thesis investigates the high-order hierarchical finite element method, also known as the finite element p-version, as a computationally-efficient technique for generating numerical solutions to the cardiac monodomain equation. We first present it as a uniform-order method, and through an a priori error bound we explain why the associated cardiac cell model must be thought of as a PDE and approximated to high-order in order to obtain the accuracy that the p-version is capable of. We perform simulations demonstrating that the achieved error agrees very well with the a priori error bound. Further, in terms of solution accuracy for time taken to solve the linear system that arises in the finite element discretisation, it is more efficient that the state-of-the-art piecewise linear finite element method. We show that piecewise linear FEM actually introduces quite significant amounts of error into the numerical approximations, particularly in the direction perpendicular to the cardiac fibres with physiological conductivity values, and that without resorting to extremely fine meshes with elements considerably smaller than 70 micrometres, we can not use it to obtain high-accuracy solutions. In contrast, the p-version can produce extremely high accuracy solutions on meshes with elements around 300 micrometres in diameter with these conductivities. Noting that most of the numerical error is due to under-resolving the wave-front in the transmembrane potential, we also construct an adaptive high-order scheme which controls the error locally in each element by adjusting the finite element polynomial basis degree using an analytically-derived a posteriori error estimation procedure. This naturally tracks the location of the wave-front, concentrating computational effort where it is needed most and increasing computational efficiency. The scheme can be controlled by a user-defined error tolerance parameter, which sets the target error within each element as a proportion of the local magnitude of the solution as measured in the H^1 norm. This numerical scheme is tested on a variety of problems in one, two and three dimensions, and is shown to provide excellent error control properties and to be likely capable of boosting efficiency in cardiac simulation by an order of magnitude. The thesis amounts to a proof-of-concept of the increased efficiency in solving the linear system using adaptive high-order finite elements when performing single-thread cardiac simulation, and indicates that the performance of the method should be investigated in parallel, where it can also be expected to provide considerable improvement. In general, the selection of a suitable preconditioner is key to ensuring efficiency; we make use of a variety of different possibilities, including one which can be expected to scale very well in parallel, meaning that this is an excellent candidate method for increasing the efficiency of cardiac simulation using high-performance computing facilities.