Cette thèse traite de la résolution du système de la Magnéto-Hydro-Dynamique (MHD) sur architectures massivement parallèles. Ce système est un système hyperbolique de lois de conservation. Pour des raisons de coût en termes de temps et d'espace, nous utilisons la méthode des volumes finis. Ces critères sont particulièrement importants dans le cas de la MHD, car les solutions obtenues peuvent présenter de nombreuses ondes de choc et être très turbulentes. L'approche d'un phénomène physique nécessite par conséquent de travailler sur un maillage fin entrainant une grande quantité de calcul. Afin de réduire les temps d'exécution des algorithmes proposés, nous proposons des méthodes d'optimisations pour l'exécution sur CPU telles que l'utilisation d'OpenMP pour une parallélisation automatique ou le parcours optimisé afin de bénéficier des effets de cache. Une implémentation sur architecture GPU à l'aide de la librairie OpenCL est également proposée. Dans le but de conserver une coalescence maximale des données en mémoire, nous proposons une méthode utilisant un splitting directionnel associé à une méthode de transposition optimisée pour les implémentations parallèle. Dans la dernière partie, nous présentons la librairie SCHNAPS. Ce solveur utilisant la méthode Galerkin Discontinu (GD) utilise des implémentations OpenCL et StarPU afin de profiter au maximum des avantages de la programmation hybride. / This thesis deals with the resolution of the Magneto-Hydro-Dynamic (MHD) system on massively parallel architectures. This problem is an hyperbolic system of conservation laws. For cost reasons in terms of time and space, we use the finite volume method. These criteria are particularly important in the case of MHD because the solutions obtained may have many shock waves and be very turbulent. The approach of a physical phenomenon requires working on a fine mesh which involves a large quantity of computations. In order to reduce the execution time of the proposed algorithms, we present several optimization methods for CPU execution such as the use of OpenMP for an automatic parallelization or an optimized way to browse a grid in order to benefit from cache effects. An implementation on GPU architecture using the OpenCL library is also available. To maintain a maximal coalescence of the data in memory, we propose a method using a directional splitting associated with an optimized transposition method for parallel implementations. In the last part, we present the SCHNAPS library. This solver using the Galerkin Disontinu (GD) method uses OpenCL and StarPU implementations in order to maximize the benefits of hybrid programming.

京都大学 / 0048 / 新制・課程博士 / 博士(農学) / 甲第21374号 / 農博第2298号 / 新制||農||1066(附属図書館) / 学位論文||H30||N5147(農学部図書室) / 京都大学大学院農学研究科地域環境科学専攻 / (主査)教授 村上 章, 教授 藤原 正幸, 教授 渦岡 良介 / 学位規則第4条第1項該当 / Doctor of Agricultural Science / Kyoto University / DFAM

In computational fluid dynamics, we often need to solve complex problems with high precision and efficiency. We propose a three-pronged approach to attain this goal. First, we use the discontinuous Galerkin spectral element method (DG-SEM) for its high accuracy. Second, we use graphics processing units (GPUs) to perform our computations to exploit available parallel computing power. Third, we implement a parallel adaptive mesh refinement (AMR) algorithm to efficiently use our computing power where it is most needed. We present a GPU DG-SEM solver with AMR and dynamic load balancing for the 2D wave equation. The DG-SEM is a higher-order method that splits a domain into elements and represents the solution within these elements as a truncated series of orthogonal polynomials. This approach combines the geometric flexibility of finite-element methods with the exponential convergence of spectral methods. GPUs provide a massively parallel architecture, achieving a higher throughput than traditional CPUs. They are relatively new as a platform in the scientific community, therefore most algorithms need to be adapted to that new architecture. We perform most of our computations in parallel on multiple GPUs. AMR selectively refines elements in the domain where the error is estimated to be higher than a prescribed tolerance, via two mechanisms: p-refinement increases the polynomial order within elements, and h-refinement splits elements into several smaller ones. This provides a higher accuracy in important flow regions and increases capabilities of modeling complex flows, while saving computing power in other parts of the domain. We use the mortar element method to retain the exponential convergence of high-order methods at the non-conforming interfaces created by AMR. We implement a parallel dynamic load balancing algorithm to even out the load imbalance caused by solving problems in parallel over multiple GPUs with AMR. We implement a space-filling curve-based repartitioning algorithm which ensures good locality and small interfaces. While the intense calculations of the high order approach suit the GPU architecture, programming of the highly dynamic adaptive algorithm on GPUs is the most challenging aspect of this work. The resulting solver is tested on up to 64 GPUs on HPC platforms, where it shows good strong and weak scaling characteristics. Several example problems of increasing complexity are performed, showing a reduction in computation time of up to 3× on GPUs vs CPUs, depending on the loading of the GPUs and other user-defined choices of parameters. AMR is shown to improve computation times by an order of magnitude or more.

Underwater explosions (UNDEX) and a ship's vulnerability to them are problems of interest in early-stage ship design. A series of events occur sequentially in an UNDEX scenario in both the fluid and structural domains and these events happen over a wide range of time and spatial scales. Because of the complexity of the physics involved, it is a common practice to separate the description of UNDEX into early-time and late-time, and far-field and near-field. The research described in this dissertation is focused on the simulation of near-field and early-time UNDEX. It assembles a hybrid framework of algorithms to provide results while maintaining computational efficiency. These algorithms include Runge-Kutta, Discontinuous Galerkin, Level Set, Direct Ghost Fluid and Embedded Boundary methods. Computational fluid dynamics (CFD) solvers are developed using this framework of algorithms to demonstrate the computational methods and their ability to effectively and efficiently solve UNDEX problems. Contributions, made in the process of satisfying the objective of this research include: the derivation of eigenvectors of flux Jacobians and their application to the implementation of the slope limiter in the fluid discretization; the three-dimensional extension of Direct Ghost Fluid Method and its application to the multi-fluid treatment in UNDEX flows; the enforcement of an improved non-reflecting boundary condition and its application to UNDEX simulations; and an improvement to the projection-based embedded boundary method and its application to fluid-structure interaction simulations of UNDEX problems. / Doctor of Philosophy / Underwater explosions (UNDEX) and a ship's vulnerability to them are problems of interest in early-stage ship design. A series of events occur sequentially in an UNDEX scenario in both the fluid and structural domains and these events happen over a wide range of time and spatial scales. Because of the complexity of the physics involved, it is a common practice to separate the description of UNDEX into early-time and late-time, and far-field and near-field. The research described in this dissertation is focused on the simulation of near-field and early-time UNDEX. It assembles a hybrid framework of algorithms to provide results while maintaining computational efficiency. These algorithms include Runge-Kutta, Discontinuous Galerkin, Level Set, Direct Ghost Fluid and Embedded Boundary methods. Computational fluid dynamics (CFD) solvers are developed using this framework of algorithms to demonstrate these computational methods and their ability to effectively and efficiently solve UNDEX problems.

This dissertation consists of three studies of immersed finite element (IFE) methods for inter- face problems related to partial differential equations (PDEs) with discontinuous coefficients. These three topics together form a continuation of the research in IFE method including the extension to elasticity systems, new breakthroughs to higher degree IFE methods, and its application to inverse problems. First, we extend the current construction and analysis approach of IFE methods in the literature for scalar elliptic equations to elasticity systems in the vector format. In particular, we construct a group of low-degree IFE functions formed by linear, bilinear, and rotated Q1 polynomials to weakly satisfy the jump conditions of elasticity interface problems. Then we analyze the trace inequalities of these IFE functions and the approximation capabilities of the resulted IFE spaces. Based on these preparations, we develop a partially penalized IFE (PPIFE) scheme and prove its optimal convergence rates. Secondly, we discuss the limitations of the current approaches of IFE methods when we try to extend them to higher degree IFE methods. Then we develop a new framework to construct and analyze arbitrary p-th degree IFE methods. In this framework, each IFE function is the extension of a p-th degree polynomial from one subelement to the whole interface element by solving a local Cauchy problem on interface elements in which the jump conditions across the interface are employed as the boundary conditions. All the components in the analysis, including existence of IFE functions, the optimal approximation capabilities and the trace inequalities, are all reduced to key properties of the related discrete extension operator. We employ these results to show the optimal convergence of a discontinuous Galerkin IFE (DGIFE) method. In the last part, we apply the linear IFE methods in the literature together with the shape optimization technique to solve a group of interface inverse problems. In this algorithm, both the governing PDEs and the objective functional for interface inverse problems are discretized optimally by the IFE method regardless of the location of the interface in a chosen mesh. We derive the formulas for the gradients of the objective function in the optimization problem which can be implemented efficiently in the IFE framework through a discrete adjoint method. We demonstrate the properties of the proposed algorithm by applying it to three representative applications. / Doctor of Philosophy / Interface problems arise from many science and engineering applications modeling the transmission of some physical quantities between multiple materials. Mathematically, these multiple materials in general are modeled by partial differential equations (PDEs) with discontinuous parameters, which poses challenges to developing efficient and reliable numerical methods and the related theoretical error analysis. The main contributions of this dissertation is on the development of a special finite element method, the so called immersed finite element (IFE) method, to solve the interface problems on a mesh independent of the interface geometry which can be advantageous especially when the interface is moving. Specifically, this dissertation consists of three projects of IFE methods: elasticity interface problems, higher-order IFE methods and interface inverse problems, including their design, analysis, and application.

Problems governed by partial differential equations (PDEs) in deformable domains, t Rd; d = 2; 3; are of fundamental importance in science and engineering. They are of particular relevance in the design of many engineering systems e.g., aircrafts and bridges as well as to the analysis of several biological phenomena e.g., blood ow in arteries. However, developing numerical scheme for such problems is still very challenging even when the deformation of the boundary of domain is prescribed a priori. Possibility of excessive mesh distortion is one of the major challenge when solving such problems with numerical methods using boundary tted meshes. The arbitrary Lagrangian- Eulerian (ALE) approach is a way to overcome this difficulty. Numerical simulations of convection-dominated problems have for long been the subject to many researchers. Galerkin formulations, which yield the best approximations for differential equations with high diffusivity, tend to induce spurious oscillations in the numerical solution of convection dominated equations. Though such spurious oscillations can be avoided by adaptive meshing, which is computationally very expensive on ne grids. Alternatively, stabilization methods can be used to suppress the spurious oscillations. In this work, the considered equation is designed within the framework of ALE formulation. In the first part, Streamline Upwind Petrov-Galerkin (SUPG) finite element method with conservative ALE formulation is proposed. Further, the first order backward Euler and the second order Crank-Nicolson methods are used for the temporal discretization. It is shown that the stability of the semi-discrete (continuous in time) ALE-SUPG equation is independent of the mesh velocity, whereas the stability of the fully discrete problem is unconditionally stable for implicit Euler method and is only conditionally stable for Crank-Nicolson time discretization. Numerical results are presented to support the stability estimates and to show the influence of the SUPG stabilization parameter in a time-dependent domain. In the second part of this work, SUPG stabilization method with non-conservative ALE formulation is proposed. The implicit Euler, Crank-Nicolson and backward difference methods are used for the temporal discretization. At the discrete level in time, the ALE map influences the stability of the corresponding discrete scheme with different time discretizations, and it leads to schemes where conservative and non-conservative formulations are no longer equivalent. The stability of the fully discrete scheme, irrespective of the temporal discretization, is only conditionally stable. It is observed from numerical results that the Crank-Nicolson scheme induces high oscillations in the numerical solution compare to the implicit Euler and the backward difference time discretiza-tions. Moreover, the backward difference scheme is more sensitive to the stabilization parameter k than the other time discretizations. Further, the difference between the solutions obtained with the conservative and non-conservative ALE forms is significant when the deformation of domain is large, whereas it is negligible in domains with small deformation. Finally, the local projection stabilization (LPS) and the higher order dG time stepping scheme are studied for convection dominated problems. The analysis is based on the quadrature formula for approximating the integrals in time. We considered the exact integration in time, which is impractical to implement and the Radau quadrature in time, which can be used in practice. The stability and error estimates are shown for the mathematical basis of considered numerical scheme with both time integration methods. The numerical analysis reveals that the proposed stabilized scheme with exact integration in time is unconditionally stable, whereas Radau quadrature in time is conditionally stable with time-step restriction depending on the ALE map. The theoretical estimates are illustrated with appropriate numerical examples with distinct features. The second order dG(1) time discretization is unconditionally stable while Crank-Nicolson gives the conditional stable estimates only. The convergence order for dG(1) is two which supports the error estimate.

This thesis develops the methodology for solving initial boundary value problems with the use of summation-by-parts discretizations. The combination of high orders of accuracy and a systematic approach to construct provably stable boundary and interface procedures makes this methodology especially suitable for scientific computations with high demands on efficiency and robustness. Most classes of high order methods can be applied in a way that satisfies a summation-by-parts rule. These include, but are not limited to, finite difference, spectral and nodal discontinuous Galerkin methods. In the first part of this thesis, the summation-by-parts methodology is extended to the time domain, enabling fully discrete formulations with superior stability properties. The resulting time discretization technique is closely related to fully implicit Runge-Kutta methods, and may alternatively be formulated as either a global method or as a family of multi-stage methods. Both first and second order derivatives in time are considered. In the latter case also including mixed initial and boundary conditions (i.e. conditions involving derivatives in both space and time). The second part of the thesis deals with summation-by-parts discretizations on multi-block and hybrid meshes. A new formulation of general multi-block couplings in several dimensions is presented and analyzed. It collects all multi-block, multi-element and  hybrid summation-by-parts schemes into a single compact framework. The new framework includes a generalized description of non-conforming interfaces based on so called summation-by-parts preserving interpolation operators, for which a new theoretical accuracy result is presented.

We analyze a number of systems of evolution equations that arise in the study of physical chemistry. First we discuss the well-posedness of a system of mixing compressible barotropic multicomponent flows. We discuss the regularity of these variational solutions, their existence and uniqueness, and we analyze the emergence of a novel type of entropy that is derived for the system of equations. Next we present a numerical scheme, in the form of a discontinuous Galerkin (DG) finite element method, to model this compressible barotropic multifluid. We find that the DG method provides stable and accurate solutions to our system, and that further, these solutions are energy consistent; which is to say that they satisfy the classical entropy of the system in addition to an additional integral inequality. We discuss the initial-boundary problem and the existence of weak entropy at the boundaries. Next we extend these results to include more complicated transport properties (i.e. mass diffusion), where exotic acoustic and chemical inlets are explicitly shown. We continue by developing a mixed method discontinuous Galerkin finite element method to model quantum hydrodynamic fluids, which emerge in the study of chemical and molecular dynamics. These solutions are solved in the conservation form, or Eulerian frame, and show a notable scale invariance which makes them particularly attractive for high dimensional calculations. Finally we implement a wide class of chemical reactors using an adapted discontinuous Galerkin finite element scheme, where reaction terms are analytically integrated locally in time. We show that these solutions, both in stationary and in flow reactors, show remarkable stability, accuracy and consistency. / text

Cette thèse porte sur l’étude d’une méthode de type Galerkin discontinu en domaine temporel (GDDT), afin de résoudre numériquement les équations de Maxwell instationnaires sur des maillages hybrides tétraédriques/hexaédriques en 3D (triangulaires/quadrangulaires en 2D) et non-conformes, que l’on note méthode GDDT-PpQk. Comme dans différents travaux déjà réalisés sur plusieurs méthodes hybrides (par exemple des combinaisons entre des méthodes Volumes Finis et Différences Finies, Éléments Finis et Différences Finies, etc.), notre objectif principal est de mailler des objets ayant une géométrie complexe à l’aide de tétraèdres, pour obtenir une précision optimale, et de mailler le reste du domaine (le vide environnant) à l’aide d’hexaèdres impliquant un gain en terme de mémoire et de temps de calcul. Dans la méthode GDDT considérée, nous utilisons des schémas de discrétisation spatiale basés sur une interpolation polynomiale nodale, d’ordre arbitraire, pour approximer le champ électromagnétique. Nous utilisons un flux centré pour approcher les intégrales de surface et un schéma d’intégration en temps de type saute-mouton d’ordre deux ou d’ordre quatre. Après avoir introduit le contexte historique et physique des équations de Maxwell, nous présentons les étapes détaillées de la méthode GDDT-PpQk. Nous réalisons ensuite une analyse de stabilité L2 théorique, en montrant que cette méthode conserve une énergie discrète et en exhibant une condition suffisante de stabilité de type CFL sur le pas de temps, ainsi que l’analyse de convergence en h (théorique également), conduisant à un estimateur d’erreur a-priori. Ensuite, nous menons une étude numérique complète en 2D (ondes TMz), pour différents cas tests, des maillages hybrides et non-conformes, et pour des milieux de propagation homogènes ou hétérogènes. Nous faisons enfin de même pour la mise en oeuvre en 3D, avec des simulations réalistes, comme par exemple la propagation d’une onde électromagnétique dans un modèle hétérogène de tête humaine. Nous montrons alors la cohérence entre les résultats mathématiques et numériques de cette méthode GDDT-PpQk, ainsi que ses apports en termes de précision et de temps de calcul. / This thesis is concerned with the study of a Discontinuous Galerkin Time-Domain method (DGTD), for the numerical resolution of the unsteady Maxwell equations on hybrid tetrahedral/hexahedral in 3D (triangular/quadrangular in 2D) and non-conforming meshes, denoted by DGTD-PpQk method. Like in several studies on various hybrid time domain methods (such as a combination of Finite Volume with Finite Difference methods, or Finite Element with Finite Difference, etc.), our general objective is to mesh objects with complex geometry by tetrahedra for high precision and mesh the surrounding space by square elements for simplicity and speed. In the discretization scheme of the DGTD method considered here, the electromagnetic field components are approximated by a high order nodal polynomial, using a centered approximation for the surface integrals. Time integration of the associated semi-discrete equations is achieved by a second or fourth order Leap-Frog scheme. After introducing the historical and physical context of Maxwell equations, we present the details of the DGTD-PpQk method. We prove the L2 stability of this method by establishing the conservation of a discrete analog of the electromagnetic energy and a sufficient CFL-like stability condition is exhibited. The theoritical convergence of the scheme is also studied, this leads to a-priori error estimate that takes into account the hybrid nature of the mesh. Afterward, we perform a complete numerical study in 2D (TMz waves), for several test problems, on hybrid and non-conforming meshes, and for homogeneous or heterogeneous media. We do the same for the 3D implementation, with more realistic simulations, for example the propagation in a heterogeneous human head model. We show the consistency between the mathematical and numerical results of this DGTD-PpQk method, and its contribution in terms of accuracy and CPU time.

De récentes analyses statistiques de données satellitales et de bouées dérivantes ont révélé le caractère hautement hétérogène et intermittent de la déformation de la banquise Arctique, démontrant de ce fait que le schéma rhéologique visco-plastique utilisé traditionnellement en modélisation climatique et opérationnelle ne simule pas adéquatement le comportement dynamique des glaces ainsi que les efforts mécaniques en leur sein.Un cadre rhéologique alternatif, baptisé "Maxwell-Élasto-Fragile" (Maxwell-EB) est donc développé dans le but de reproduire correctement la dérive et la déformation des glaces dans les modèles continus de la banquise à l'échelle régionale et globale. Le modèle se base en partie sur un cade de modélisation élasto-fragile utilisé pour les roches et la glace. Un terme de relaxation visqueuse est ajouté à la relation constitutive d'élasticité linéaire ainsi qu'une viscosité effective, ou "apparente", laquelle évolue en fonction du niveau d'endommagement local du matériel simulé, comme son module d'élasticité. Ce cadre rhéologique permet la dissipation partielle des contraintes internes par le biais de déformations permanentes, possiblement grandes, le long de failles (ou "leads") lorsque le matériel est fortement endommagé ainsi que la conservation de la mémoire des contraintes associées aux déformations élastiques dans les zones où le matériel reste relativement peu endommagé.The schéma numérique du modèle Maxwell-EB est basé sur des méthodes de calcul variationnel et par éléments finis. Une représentation Eulérienne des équations du mouvement est utilisée et des méthodes dites Galerkin discontinues sont implémentées pour le traitement des processus d'advection.Une première série de simulations idéalisées et sans advection est présentée, lesquelles démontrent que la rhéologie Maxwell-Élasto-Fragile reproduit les caractéristiques principales du comportement mécanique de la banquise, c'est-à-dire la localisation spatiale, l'anisotropie et l'intermittence de la déformation ainsi que les lois d'échelle qui en découlent. La représentation adéquate de ces propriétés de la déformation se traduit par la présence de très forts gradients au sein des champs de contrainte, de déformation et du niveau d'endommagement simulés par le modèle. Des tests visant à évaluer la diffusion numérique découlant de l'advection de ces gradients extrêmes ainsi qu'à identifier certaines contraintes numériques du modèle sont ensuite présentés. De premières simulations en grandes déformations, incluant les processus d'advection, sont réalisées, lesquelles permettent une comparaison aux résultats d'une expérience de Couette annulaire sur de la glace fabriquée en laboratoire. Le modèle reproduit en partie le comportement mécanique observé. Par ailleurs, les différences entre les résultats des simulations et ceux obtenus en laboratoire permettent d'identifier certaines limitations, numériques et physiques, du modèle en grandes déformations. Finalement, le modèle rhéologique est utilisé pour modéliser la dérive et la déformation des glaces à l'échelle de la banquise Arctique. Des simulations idéalisées de l'écoulement de glace dans un chenal étroit sont présentées. Le modèle simule une propagation localisée de l'endommagement, définissant des failles en forme d'arche, et la formation de ponts de glace stables. / In recent years, analyses of available ice buoy and satellite data have revealed the strong heterogeneity and intermittency of the deformation of sea ice and have demonstrated that the viscous-plastic rheology widely used in current climate models and operational modelling platforms does not simulate adequately the drift, deformation and mechanical stresses within the ice pack.A new alternative rheological framework named ''Maxwell-Elasto-Brittle” (Maxwell-EB) is therefore developed in the view of reproducing more accurately the drift and deformation of the ice cover in continuum sea ice models at regional to global scales. The model builds on an elasto-brittle framework used for ice and rocks. A viscous-like relaxation term is added to a linear-elastic constitutive relationship together with an effective viscosity that evolves with the local level of damage of the material, like its elastic modulus. This framework allows for part of the internal stress to dissipate in large, permanent deformations along the faults/leads once the material is highly damaged while retaining the memory of small, elastic deformations over undamaged areas. A healing mechanism is also introduced, counterbalancing the effects of damaging over large time scales.The numerical scheme for the Maxwell-EB model is based on finite elements and variational methods. The equations of motion are cast in the Eulerian frame and discontinuous Galerkin methods are implemented to handle advective processes.Idealized simulations without advection are first presented. These demonstrate that the Maxwell-EB rheological framework reproduces the main characteristics of sea ice mechanics and deformation : the strain localization, the anisotropy and intermittency of deformation and the associated scaling laws. The successful representation of these properties translates into very large gradients within all simulated fields. Idealized numerical experiments are conducted to evaluate the amount of numerical diffusion associated with the advection of these extreme gradients in the model and investigate other limitations of the numerical scheme. First large-deformation simulations are carried in the context of a Couette flow experiment, which allow a comparison with the result of a similar laboratory experiment performed on fresh-water ice. The model reproduces part of the mechanical behaviour observed in the laboratory. Comparison of the numerical and experimental results allow identifying some numerical and physical limitations of the model in the context of large-deformation and laboratory-scale simulations. Finally, the Maxwell-EB framework is implemented in the context of modelling the drift and deformation of sea ice on geophysical scales. Idealized simulations of the flow of sea ice through a narrow channel are presented. The model simulates the propagation of damage along arch-like features and successfully reproduces the formation of stable ice bridges.