This study uses the PIC (Particle-in-cell) method to simulate unsteady three-dimensional transport variables in argon plasma under low pressure and weak ionization between two planar electrodes suddenly biased by a negative voltage. Plasma has been widely used in etching, ion implantation, light source, and encountered in nuclear fusion, etc. Studying transport processes of plasmas therefore is important. This work ignores magnetic field, secondary electron emission, recombination between ions and electrons, and assumes a uniform distribution of the neutrals having velocity of a Maxwellian distribution. Accounting for elastic collisions between electrons and neutrals, ions and neutrals, and inelastic collisions resulting in ionization from impacting neutrals by electrons, and charge exchange between ions and neutrals, the computed results in this work quantitatively show non-isotropic pressures, shear stresses and heat conduction of the ions across the sheath to the surfaces suddenly biased by a dc negative voltage.

Electrodynamic tethers have a wide range of proposed applications in the fields of satellite propulsion and space plasma research. The fundamental purpose of this dissertation is to improve the understanding of the behavior of an electrodynamic tether (EDT) system in Earth's ionosphere. An electrodynamic tether system consists of two satellites connected by a long tether that generates current to produce either power or thrust via the system's electromagnetic interaction with the space environment. Previous electrodynamic tether investigations decouple the interaction between the tether and the constantly changing plasma environment. The limiting factor inhibiting the development of a full system model that has an accurate characterization of the tether/plasma interaction is that the understanding of that interaction is not well developed over a wide range of system parameters. The EDT system model developed in this study uses a high fidelity dynamics model that includes a tether current described by an analytical current collection model whose plasma parameters are determine by the International Reference Ionosphere. It is first shown that new instabilities are induced in the system dynamics under a basic analytical current model versus a constant current model. A 2-D3$v$ Particle-in-Cell (PIC) code has been developed to study the plasma dynamics near a positively charged EDT system end-body and their impact on the current collected. Simulations are run over a range of system parameters that occur throughout a LEO orbit. The azimuthal current structures observed during the TSS-1R mission are found to enhance the current collected by the satellite when the magnetic field is slightly off of perpendicular to the orbital velocity. When the in-plane component of the magnetic field becomes large, the electrons are not able to easily cross the field lines causing plasma lobes form above and below the satellite. The lobes limit the current arriving to the satellite and also cause an enhanced wake to develop. A high satellite bias causes a stable bow-shock structure to form in the ram region of the satellite, which limits the number of electrons entering the sheath region and thus limiting the current collected. Electron-neutral collisions are found to destabilize the bow-shock structure and remove its current limiting effects. Additionally, as the magnetization of the plasma is increased, the current becomes limited by the charged particle's inability to cross magnetic field lines. Analytical curve fits based on the simulation results are presented that characterize the dependence of the average current collected on the local magnetic field orientations, space plasma magnetization and satellite potential. The results from the PIC simulations characterizing the magnetic field's influence on the tether's current are incorporated into the system dynamics model to study the behavior of the EDT system over a range of inclinations. The magnetic field is found to limit the diurnal variations in the current collected by the system throughout its orbit. As the inclination of the system's orbit is increased, the impact of the magnetic field becomes more pronounced as its orientation sweeps through a larger range of angles. The impact of the magnetic field on the collected current is, therefore, found to limit the ability of an EDT system to boost the system's orbit as the orbit's inclination is increased. In summary, new system dynamics have been observed due to the previously unobserved behavior of the current over a range of end-body configurations. / Ph. D.

The interaction of a high power, high frequency (HF) pump/electromagnetic (EM) wave transmitted from a ground-based station with the ionosphere, experiments which have been termed "ionospheric heating", produces secondary radiation known as stimulated electromagnetic emissions (SEEs). SEEs have been developed into powerful diagnostics yielding information such as electron temperature, ion species and hydrodynamic evolution of the modified ionospheric plasma. Classic SEEs which exist outside ±1 kHz of the pump wave frequency (ω0) have recently been classified into wideband SEEs (PW-WSEEs) and distinguished from narrowband SEEs (PW-NSEEs) which exist within ±1 kHz of ω0, where the "PW" prefix has been used to indicate that the frequency regimes in the aforementioned classification are relative to the pump wave (PW) frequency. The occurrence of SEEs near 2ω0 is known as second harmonic generation (SHG). SHG is longstanding and well-established in the field of Laser Plasma Interactions (LPI) where SHG has been harnessed to yield diagnostics such as the velocity of the critical region of the plasma, inference of the region in the plasma where the interaction that results in SHG occurs, plasma turbulence and density scale lengths. Past studies of ionospheric heating SHG were limited by the effective radiated power (ERP) available at ionospheric heating facilities and the frequency resolution of receivers/spectrum analyzers of the time. Experimental observations from these past studies reported either SEEs produced as a result of SHG in isolation or compared these SEEs with PW- WSEEs. Moreover, these experiments did not evaluate effects such as transmit ERP, tilt of the transmit antenna beam from the geomagnetic field (B0) and the offset of ω0 from harmonics of the electron gyrofrequency (ωce) on SEEs within a narrowband of twice the pump wave frequency produced as a result of SHG. Also, these studies did not attempt to draw from the knowledge-base on SHG from LPI. The novelty of the experimental observations in this dissertation is the juxtaposition of PW-NSEEs and second harmonic narrowband SEEs (SH-NSEEs), which are SEEs within kHz of 2ω0, measured at the same time. The heating experiments were all performed at HAARP using an O-mode polarized EM pump wave. Additionally, these measurements evaluate the effects on SHG of the transmit ERP, tilt of the transmit station antenna beam from the geomagnetic field (B0) and the offset of ω0 from nωce, n = 2, 3. The experimental observations show, for the first time, a clear association between PW-NSEEs and SH-NSEEs. This association is subsequently used, in conjunction with theories from LPI to propose the non-linear wave-mixing mechanisms responsible for the SH-NSEEs. As a prelude to harnessing the wealth of diagnostics that can be obtained from SHG, initial diagnostics of the velocity of the critical region and the interaction region where SHG occurs are determined using theories from LPI. With the association between PW-NSEEs and SH-NSEEs established, Particle- In-Cell (PIC) simulations are used to investigate the characteristics of a PW- NSEE herein referred to as the "SBS line", produced as a result of stimulated Brillouin scatter (SBS) instability in which the pump EM wave decays into a backscattered EM wave and an ion acoustic wave. The PIC simulations reveal that for high pump powers, the SBS line, which is intense at the onset of the heating experiment, is suppressed within 3 seconds due to the development of cavities in the ionospheric plasma (density) in which the pump wave depletes its energy in heating up electrons. Although, no PIC simulation results of SHG have been presented in this work, the association between PW-NSEEs and SH-NSEEs shown in this work is used to propose that similar mechanisms are responsible for the suppression the SBS line and its associated SH-NSEE for high pump powers. Results from ionospheric heating experiments presented in this dissertation show a rapid suppression of both the SBS line and its associated SH-NSEE for high pump powers. The attribution of the suppression of SH-NSEEs to the development of artificial field-aligned irregularities (AFAIs) in a past study fails to explain the rapid suppression in the experimental observations contained herein since the suppression occurs on a much faster timescale than the development of AFAIs. Thus, the PIC model results have led to a more feasible interpretation of the observed rapid suppression. To re-iterate, the contributions of this dissertation are as follows: 1. First observations of an SH-NSEE named "SH decay line" within 2ω0±30 Hz. The SH decay line occurs at the same transmit power as the SBS line within ω0±30 Hz and both of these SEEs are suppressed for ω0 ≈ 3ωce. Offset of the SH decay line from 2ω0 is twice the offset of the SBS line from ω0. 2. First experimental evaluation of the impact of B0 assessed by stepping the transmit beam offset from B0 and stepping ω0 near 2ωce shows contemporaneous SH-NSEEs and PW-NSEEs both ordered by the O+ ion cyclotron frequency. 3. First experimental observations of suppression of SBS line and SH decay line for high pump powers, which unlike a past study cannot be attributed to AFAIs. 4. First PIC simulation investigation of suppression of SBS line observed during high pump power ionospheric heating, revealing depletion of pump energy in heating electrons in cavities created in the plasma (density) as the mechanism responsible for the suppression. Broadening of SBS line observed in ionospheric heating with high power is also observed in PIC simulation results. This work has laid the foundations to develop SHG into powerful ionospheric diagnostics. / Doctor of Philosophy / When a high power, high frequency radio wave is injected from a ground-based transmit station into the ionosphere, a region of Earth's atmosphere containing charged particles in addition some neutral atoms and molecules, the frequency spectrum measured at a location removed from the transmit station shows emissions at other frequencies in addition to an emission at the transmit frequency. The emissions at these other frequencies are known as stimulated electromagnetic emissions (SEEs). The frequency offsets of SEEs contain information such as the average kinetic energy associated with random motion of electrons, a parameter known as electron temperature and the ion species present in the region of the ionosphere the radio wave is injected into. The occurrence of SEEs near twice the pump wave frequency is known as second harmonic generation. This dissertation presents experimental observations that compare SEEs which exist within ±1 kHz of the transmit frequency with SEEs which exist within a similar frequency range of twice the transmit frequency unlike past studies. This dissertation also investigates effects of varying the transmit frequency, power and the direction of the transmit station antenna beam relative to the local direction of the magnetic field of the Earth. These new studies reveal, for the first time, a similarity in characteristics of the SEEs near the transmit frequency and two times the transmit frequency. This similarity is used in conjunction with theories from studies of Laser Plasma Interaction (LPI), which have corollaries with high power radio wave-ionosphere interaction, to propose the processes that underlie the occurrence of SEEs near twice the transmit frequency. Methods from LPI have also been used for the first time to obtain measurements of some parameters of the ionosphere. High power radio wave-ionosphere interaction experiments are very expensive and moreover, direct measurement of ionospheric parameters/processes require radar facilities which may not be available or sounding rockets or satellites which increase the cost of experiments. Computer simulations offer a facile and an inexpensive means to investigate SEEs and processes internal to the ionosphere. Computer simulations have been used for the first time in this dissertation to investigate the mechanisms responsible for the characteristics of SEEs near the transmit frequency for low and high transmit powers. Since an association has been established in this dissertation between SEEs near the transmit frequency and SEEs near twice the transmit frequency, the mechanisms responsible for the characteristics for the SEEs near the transmit frequency for high transmit power, have been proposed to be the same mechanisms responsible for the characteristics of SEEs near twice the transmit frequency for a similar transmit power regime. The experimental results, computer simulation results and the corollaries drawn between high power radio wave-ionosphere interaction and LPI detailed in this dissertation have opened new doors to develop SEEs near twice the transmit frequency into a powerful tool to study the ionosphere.

A new particle-in-cell algorithm was developed for plasma simulations involving complex boundary conditions. The new algorithm is based on the three-dimensional immersed finite element method which is developed in this thesis, and a modified legacy particle-in-cell code. The model also applies a new meshing technique that separates the field solution mesh from the particle pushing mesh in order to increase the computational eciency of the model. The new simulation model is used in two applications of great importance to the development of ion propulsion technology: the ion optics performance and the interaction between spacecraft and the ion thruster. The first application is ion optics simulations. Simulations are performed to investigate ion optics plasma flow for a whole subscale NEXT ion optics. The operating conditions modeled cover the entire cross-over to perveance limit range. The results of the ion optics simulations demonstrated good agreement with the available experimental data. The second application is ion thruster plume simulations. Simulations are performed to investigate ion thruster plume - spacecraft interactions for the Dawn spacecraft. Plume induced contaminations on the solar array are studied for a variety of ion thruster configurations including multiple thruster firings. / Ph. D.

Avec les dernières augmentations de l'intensité maximale de laser réalisable grâce à de courtes impulsions à haute puissance (gamme femtoseconde) un intérêt a surgi dans les sources de plasma laser potentiels. Les lasers sont utilisés en radiographie proton, allumage rapide, hadronthérapie, la production de radioisotopes et de laboratoire astrophysique. Au cours de l'interaction laser-cible, les ions sont accélérés par des processus physiques différents, en fonction de la zone de la cible. Tous ces mécanismes ont un point commun: les ions sont accélérés par des champs électriques intenses, qui se produisent en raison de la séparation de forte charge induite par l'interaction de l'impulsion laser avec la cible, directement ou indirectement. Deux principales sources distinctes pour le déplacement de charge peuvent être mis en évidence. Le premier est le gradient de charge provoquée par l'action directe de la force ponderomotive de laser sur les électrons dans la surface avant de la cible, qui est la prémisse pour le processus d'accélération des radiations de pression (RPA). Une deuxième source peut être identifiée comme provenant du rayonnement laser qui est transformée en énergie cinétique d'une population d'électrons relativistes chaud (~ quelques MeV). Les électrons chauds se déplacent et font recirculer à travers la cible et forment un nuage d'électrons relativistes à la sortie de la cible dans le vide. Ce nuage, qui se prolonge pour plusieurs longueurs de Debye, crée un champ électrique extrêmement intense longitudinal, la plupart du temps dirigé le long de la surface normale, ce qui, par conséquent, est la cause de l'accélération d'ions efficace, qui conduit à l'accélération cible normale gaine (TNSA) processus . Le mécanisme TNSA permet d'utiliser des géométries différentes cibles afin de parvenir à une meilleure focalisation des faisceaux de particules de l'ordre de plusieurs dizaines de microns, avec des densités d'énergie élevées. Les électrons chauds sont produits par l'irradiation d'une feuille solide avec une impulsion laser intense; ces électrons sont transportés à travers la cible, la formation d'un champ électrostatique fort, normal à la surface cible. Protons et les ions chargés positivement de la surface arrière de la cible sont accélérés par ce domaine jusqu'à ce que la charge de l'électron est compensée. La densité d'électrons chauds et la température dans le vide arrière dépendent des propriétés géométriques et de composition cibles tels que la courbure de la cible, les structures de mise au point d'impulsion et de microstructure pour l'accélération de protons améliorée. Au cours de ma première année, j'ai étudié les effets de la géométrie de la cible sur le proton et l'ion énergie et la distribution angulaire afin d'optimiser les faisceaux de particules laser accéléré au moyen de deux dimensions (2D) particule-in-cell (PIC) simulations de l'interaction de l'ultra-court impulsions laser avec plusieurs cibles microstructurées. Également au cours de cette année, je l'ai étudié la théorie derrière les modèles utilisés. / With the latest increases in maximum laser intensity achievable through short pulses at high power (femtosecond range) an interest has arisen in potential laser plasma sources. Lasers are used in proton radiography, rapid ignition, hadrontherapy, production of radioisotopes and astrophysical laboratory. During the laser-target interaction, the ions are accelerated by different physical processes, depending on the area of ​​the target. All these mechanisms have one thing in common: the ions are accelerated by intense electric fields, which occur due to the separation of high charge induced by the interaction of the laser pulse with the target, directly or indirectly. Two main distinct sources for charge displacement can be identified. The first is the charge gradient caused by the direct action of the laser ponderomotive force on the electrons in the front surface of the target, which is the premise for the pressure ramping acceleration (RPA) process. A second source can be identified as coming from the laser radiation which is transformed into kinetic energy of a hot relativistic electron population (~ a few MeV). The hot electrons move and recirculate through the target and form a cloud of relativistic electrons at the exit of the target in a vacuum. This cloud, which extends for several lengths of Debye, creates an extremely intense longitudinal electric field, mostly directed along the normal surface, which is therefore the cause of effective ion acceleration, which leads to the normal target sheath acceleration (TNSA) process. The TNSA mechanism makes it possible to use different target geometries in order to obtain a better focusing of the beams of particles on the order of several tens of microns, with high energy densities. Hot electrons are produced by irradiating a solid sheet with an intense laser pulse; these electrons are transported through the target, forming a strong electrostatic field, normal to the target surface. Protons and positively charged ions from the back surface of the target are accelerated by this domain until the charge of the electron is compensated. The density of hot electrons and the temperature in the back vacuum depend on the target geometric and compositional properties such as target curvature, pulse and microstructure tuning structures for enhanced proton acceleration. In my first year I studied the effects of target geometry on the proton and energy ion and angular distribution in order to optimize the accelerated laser particle beams by means of two-dimensional (2D) particle -in-cell (PIC) simulations of the interaction of ultra-short laser pulses with several microstructured targets. Also during this year, I studied the theory behind the models used.

Submitted by Biblioteca do Instituto de Física (bif@ndc.uff.br) on 2017-05-26T19:15:31Z No. of bitstreams: 1 tese-mestrado.pdf: 17362957 bytes, checksum: 024ecd9a310935efc222a31e2a31dca0 (MD5) / Made available in DSpace on 2017-05-26T19:15:31Z (GMT). No. of bitstreams: 1 tese-mestrado.pdf: 17362957 bytes, checksum: 024ecd9a310935efc222a31e2a31dca0 (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Nesta Dissertação são estudadas as condições em que um pulso eletromagnético (semente) é amplificado e comprimido através da interação não linear de três ondas de um plasma. Para isto, foi empregada uma abordagem analítica em que os três modos, fonte-plasma-semente, variam lentamente no tempo e no espaço. Durante a interação dos três modos, a amplitude da fonte é completamente depletada, enquanto que a onda de plasma e o pulso semente se amplificam. Durante este processo, o pulso semente se dividirá num pulso principal, o qual diminui sua largura proporcionalmente com o tempo de interação, e em vários outros pulsos secundários e de largura constante. Observa-se que a saturação do pulso semente ocorre antes que as instabilidades dos tipos lamentação, quebra de onda e espalhamento Raman para frente ocorram. Fazendo uso do código computacional XOOPIC e de acordo com as condições iniciais impostas ao sistema, obtem-se a potência final do pulso semente e a e ciência do processo. / The conditions under which an electromagnetic (seed) pulse is amplified and compressed by nonlinear three wave interaction in a plasma are studied. An analytical approach is used considering the slowly varying, in space and time, envelope amplitudes of the three wave modos (pump-plasma-seed). It is noted that the amplitude of the pump wave is completely depleted during the interaction while the plasma wave and the seed pulse are amplified. During the interaction, the seed pulse is divided into a main pulse, which decreases its width during the interaction, and in many other secondary pulses with constant width. The process of the saturation of the ampli ed seed pulse occurs before the filamentation, plasma wavebreaking and Raman forward scattering instabilities take place. Using the particle simulation code XOOPIC and according to the prescribed initial conditions of the system, the nal power of the seed pulse and the e ficiency of the process are obtained.

L'objectif de cette thèse est l'étude de la reconnexion magnétique dans les plasmas non-collisionels et relativistes. De tels plasmas sont présents dans divers objets astrophysiques (MQs, AGNs, GRBs...), où la reconnexion pourrait expliquer la production de particules et de radiation de haute énergie, un chauffage, ou des jets. Une compréhension fondamentale de la reconnexion n'est cependant toujours pas acquise, en particulier dans les plasmas relativistes ion-électron. Nous présentons d'abord les bases de la reconnexion magnétique. Nous démontrons des résultats particuliers à la physique des plasmas relativistes, concernant par exemple la distribution de Maxwell-Jüttner. Ensuite, nous réalisons une étude détaillée de l'outil numérique utilisé : les simulations particle-in-cell (PIC). Le fait que le plasma réel contienne beaucoup plus de particules que le plasma PIC a des conséquences importantes (collisionalité, relaxation, bruit) que nous décrivons. Enfin, nous étudions la reconnexion magnétique dans les plasmas ion-électron et relativistes à l'aide de simulations PIC. Nous soulignons des points spécifiques : loi d'Ohm (l'inertie de bulk dominante), zone de diffusion, taux de reconnexion (et sa normalisation relativiste). Les ions et les électrons produisent des lois de puissance, avec un index qui dépend de la vitesse d'Alfvén et de la magnétisation, et qui peut être plus dur que dans le cas des chocs non-collisionels. De plus, les ions peuvent avoir plus ou moins d'énergie que les électrons selon la valeur du champ guide. Ces résultats fournissent une base solide à des modèles d'objets astrophysiques qui, jusque là, supposaient a priori ces résultats. / The purpose of this thesis is to study magnetic reconnection in collisionless and relativistic plasmas. Such plasmas can be encountered in various astrophysical objects (microquasars, AGNs, GRBs...), where reconnection could explain high-energy particle and photon production, plasma heating, or transient large-scale outflows. However, a first principle understanding of reconnection is still lacking, especially in relativistic ion-electron plasmas. We first present the basis of reconnection physics. We derive results relevant to relativistic plasma physics, including properties of the Maxwell-Jüttner distribution. Then, we provide a detailed study of our numerical tool, particle-in-cell simulations (PIC). The fact that the real plasma contains far less particles than the PIC plasma has important consequences concerning relaxation times or noise, that we describe. Finally, we study relativistic reconnection in ion-electron plasmas with PIC simulations. We stress outstanding properties: Ohm's law (dominated by bulk inertia), structure of the diffusion zone, energy content of the outflows (thermally dominated), reconnection rate (and its relativistic normalization). Ions and electrons produce power law distributions, with indexes that depend on the inflow Alfvén speed and on the magnetization of the corresponding species. They can be harder than those produced by collisionless shocks. Also, ions can get more or less energy than the electrons, depending on the guide field strength. These results provide a solid ground for astrophysical models that, up to now, assumed with no prior justification the existence of such distributions or of such ion/electron energy repartition.

This thesis presents how performance data from hardware accelerators can be included in event logs. It extends the capabilities of trace-based performance analysis to also monitor and record data from this novel parallelization layer. The increasing awareness to power consumption of computing devices has led to an interest in hybrid computing architectures as well. High-end computers, workstations, and mobile devices start to employ hardware accelerators to offload computationally intense and parallel tasks, while at the same time retaining a highly efficient scalar compute unit for non-parallel tasks. This execution pattern is typically asynchronous so that the scalar unit can resume other work while the hardware accelerator is busy. Performance analysis tools provided by the hardware accelerator vendors cover the situation of one host using one device very well. Yet, they do not address the needs of the high performance computing community. This thesis investigates ways to extend existing methods for recording events from highly parallel applications to also cover scenarios in which hardware accelerators aid these applications. After introducing a generic approach that is suitable for any API based acceleration paradigm, the thesis derives a suggestion for a generic performance API for hardware accelerators and its implementation with NVIDIA CUPTI. In a next step the visualization of event logs containing data from execution streams on different levels of parallelism is discussed. In order to overcome the limitations of classic performance profiles and timeline displays, a graph-based visualization using Parallel Performance Flow Graphs (PPFGs) is introduced. This novel technical approach is using program states in order to display similarities and differences between the potentially very large number of event streams and, thus, enables a fast way to spot load imbalances. The thesis concludes with the in-depth analysis of a case-study of PIConGPU---a highly parallel, multi-hybrid plasma physics simulation---that benefited greatly from the developed performance analysis methods. / Diese Dissertation zeigt, wie der Ablauf von Anwendungsteilen, die auf Hardwarebeschleuniger ausgelagert wurden, als Programmspur mit aufgezeichnet werden kann. Damit wird die bekannte Technik der Leistungsanalyse von Anwendungen mittels Programmspuren so erweitert, dass auch diese neue Parallelitätsebene mit erfasst wird. Die Beschränkungen von Computersystemen bezüglich der elektrischen Leistungsaufnahme hat zu einer steigenden Anzahl von hybriden Computerarchitekturen geführt. Sowohl Hochleistungsrechner, aber auch Arbeitsplatzcomputer und mobile Endgeräte nutzen heute Hardwarebeschleuniger um rechenintensive, parallele Programmteile auszulagern und so den skalaren Hauptprozessor zu entlasten und nur für nicht parallele Programmteile zu verwenden. Dieses Ausführungsschema ist typischerweise asynchron: der Skalarprozessor kann, während der Hardwarebeschleuniger rechnet, selbst weiterarbeiten. Die Leistungsanalyse-Werkzeuge der Hersteller von Hardwarebeschleunigern decken den Standardfall (ein Host-System mit einem Hardwarebeschleuniger) sehr gut ab, scheitern aber an einer Unterstützung von hochparallelen Rechnersystemen. Die vorliegende Dissertation untersucht, in wie weit auch multi-hybride Anwendungen die Aktivität von Hardwarebeschleunigern aufzeichnen können. Dazu wird die vorhandene Methode zur Erzeugung von Programmspuren für hochparallele Anwendungen entsprechend erweitert. In dieser Untersuchung wird zuerst eine allgemeine Methodik entwickelt, mit der sich für jede API-gestützte Hardwarebeschleunigung eine Programmspur erstellen lässt. Darauf aufbauend wird eine eigene Programmierschnittstelle entwickelt, die es ermöglicht weitere leistungsrelevante Daten aufzuzeichnen. Die Umsetzung dieser Schnittstelle wird am Beispiel von NVIDIA CUPTI darstellt. Ein weiterer Teil der Arbeit beschäftigt sich mit der Darstellung von Programmspuren, welche Aufzeichnungen von den unterschiedlichen Parallelitätsebenen enthalten. Um die Einschränkungen klassischer Leistungsprofile oder Zeitachsendarstellungen zu überwinden, wird mit den parallelen Programmablaufgraphen (PPFGs) eine neue graphenbasisierte Darstellungsform eingeführt. Dieser neuartige Ansatz zeigt eine Programmspur als eine Folge von Programmzuständen mit gemeinsamen und unterchiedlichen Abläufen. So können divergierendes Programmverhalten und Lastimbalancen deutlich einfacher lokalisiert werden. Die Arbeit schließt mit der detaillierten Analyse von PIConGPU -- einer multi-hybriden Simulation aus der Plasmaphysik --, die in großem Maße von den in dieser Arbeit entwickelten Analysemöglichkeiten profiert hat.

De nombreux corps du système solaire possèdent des enveloppes liquides internes, comme par exemple les noyaux métalliques des planètes telluriques et les océans profonds des satellites de glace de Jupiter et Saturne, dans lesquelles se produisent des courants de convection. La modélisation de la dynamique de ces enveloppes est cruciale pour comprendre la génération des champs magnétiques planétaires (pour les noyaux) et pour mieux déterminer l’habitabilité potentielle des satellites joviens. La convection dans ces enveloppes est généralement produite par la combinaison d’au moins deux sources de flottabilité : une source thermique et une source solutale. Une telle situation est plus complexe qu’un régime de convection purement thermique ou purement solutale, d’une part en raison de l’existence d’un couplage thermochimique lorsqu’un processus de fusion ou de cristallisation se produit à l’une des frontières de l’enveloppe, et d’autre part à cause de la forte différence de diffusivité moléculaire entre les champs thermique et compositionnel qui permet potentiellement le développement d’instabilités double-diffusives. Classiquement, ces complexités ont été ignorées dans les simulations numériques de la dynamo terrestre ; les champs thermique et compositionnel ayant été combinés en une seule variable nommée « codensité ». Cette approche est sans doute simpliste mais permet d’esquiver une difficulté technique liée à la description du champ compositionnel dont la très faible diffusivité nécessite de recourir à des méthodes numériques adaptées. Cette thèse présente d’abord l’implémentation d’une méthode semi-Lagrangienne du type « particle-in-cell » dans un code de dynamo pré-existant, permettant ainsi de traiter de manière plus réaliste le champ de composition dans les enveloppes liquides internes des planètes. Les optimisations réalisées sont détaillées ainsi que les résultats de tests sur des cas de benchmark qui valident cet outil. Une comparaison avec des méthodes Eulériennes est également présentée. Une première exploration de la physique de la convection compositionnelle et thermochimique en rotation dans la limite d’un nombre de Prandtl compositionnel infini est ensuite conduite dans le contexte du noyau liquide terrestre. Il est montré que la dynamique convective est très différente de celle de la convection thermique pure. Notamment, les matériaux légers injectés à la frontière graine/noyau liquide sont capables d’atteindre la frontière noyau/manteau et de s’y accumuler pour former une couche chimiquement stratifiée, dont l’existence a été évoquée théoriquement mais qui n’a jamais pu être produite dans de précédentes simulations. Enfin, la dynamique double-diffusive des couches stratifiées est également discutée, et de premières simulations de « salt fingers » sont présentées. / Numerous planetary bodies contain internal liquid layers in which convective currents are generated by the combination of buoyancy sources of thermal and compositional origin. The strong difference between the thermal and chemical molecular diffusivities and the possibility of thermo-chemical coupling at melting or freezing boundaries create a convective regime that is much more complex than pure thermal convection, partly due to the potential occurrence of double-diffusive instabilities. Traditionally, numerical simulations have modeled the dynamics of the liquid part of planetary cores in a more simplistic way by neglecting the diffusivity difference and combining both fields into one single variable, an approximation that is convenient but maybe not relevant. However, distinguishing both fields and dealing with a large or infinite diffusivity ratio makes it compulsory to use numerical methods that minimize numerical diffusion as much as possible. In this thesis, I adapted a semi-Lagrangian particle-in-cell method into a pre-existing dynamo code to describe the weakly diffusive compositional field. I optimized the code for massively parallel computing and validated it on two different benchmarks. I compared the particle-in-cell method to Eulerian schemes and showed that its advantages extend beyond its lower numerical dissipation. Using this new tool, I performed first numerical simulations of rotating pure compositional and thermochemical convection in the limit of null chemical diffusivity. I explored the physics of pure compositional convection and addressed questions related to the existence and the dynamics of a stratified layer below the Earth’s core mantle boundary. In particular, I showed that the stratification could potentially be of chemical origin and proposed some mechanisms to explain its formation. In the case of a thermally stratified layer, I performed a scaling analysis of fingering instabilities, wrote the first steps of a linear stability analysis and ran a few simulations of fingering instabilities in the rotating case. The potential effects of the magnetic field and the coupling of thermochemical boundary conditions in planetary cores are finally discussed in this thesis.

This thesis presents how performance data from hardware accelerators can be included in event logs. It extends the capabilities of trace-based performance analysis to also monitor and record data from this novel parallelization layer. The increasing awareness to power consumption of computing devices has led to an interest in hybrid computing architectures as well. High-end computers, workstations, and mobile devices start to employ hardware accelerators to offload computationally intense and parallel tasks, while at the same time retaining a highly efficient scalar compute unit for non-parallel tasks. This execution pattern is typically asynchronous so that the scalar unit can resume other work while the hardware accelerator is busy. Performance analysis tools provided by the hardware accelerator vendors cover the situation of one host using one device very well. Yet, they do not address the needs of the high performance computing community. This thesis investigates ways to extend existing methods for recording events from highly parallel applications to also cover scenarios in which hardware accelerators aid these applications. After introducing a generic approach that is suitable for any API based acceleration paradigm, the thesis derives a suggestion for a generic performance API for hardware accelerators and its implementation with NVIDIA CUPTI. In a next step the visualization of event logs containing data from execution streams on different levels of parallelism is discussed. In order to overcome the limitations of classic performance profiles and timeline displays, a graph-based visualization using Parallel Performance Flow Graphs (PPFGs) is introduced. This novel technical approach is using program states in order to display similarities and differences between the potentially very large number of event streams and, thus, enables a fast way to spot load imbalances. The thesis concludes with the in-depth analysis of a case-study of PIConGPU---a highly parallel, multi-hybrid plasma physics simulation---that benefited greatly from the developed performance analysis methods. / Diese Dissertation zeigt, wie der Ablauf von Anwendungsteilen, die auf Hardwarebeschleuniger ausgelagert wurden, als Programmspur mit aufgezeichnet werden kann. Damit wird die bekannte Technik der Leistungsanalyse von Anwendungen mittels Programmspuren so erweitert, dass auch diese neue Parallelitätsebene mit erfasst wird. Die Beschränkungen von Computersystemen bezüglich der elektrischen Leistungsaufnahme hat zu einer steigenden Anzahl von hybriden Computerarchitekturen geführt. Sowohl Hochleistungsrechner, aber auch Arbeitsplatzcomputer und mobile Endgeräte nutzen heute Hardwarebeschleuniger um rechenintensive, parallele Programmteile auszulagern und so den skalaren Hauptprozessor zu entlasten und nur für nicht parallele Programmteile zu verwenden. Dieses Ausführungsschema ist typischerweise asynchron: der Skalarprozessor kann, während der Hardwarebeschleuniger rechnet, selbst weiterarbeiten. Die Leistungsanalyse-Werkzeuge der Hersteller von Hardwarebeschleunigern decken den Standardfall (ein Host-System mit einem Hardwarebeschleuniger) sehr gut ab, scheitern aber an einer Unterstützung von hochparallelen Rechnersystemen. Die vorliegende Dissertation untersucht, in wie weit auch multi-hybride Anwendungen die Aktivität von Hardwarebeschleunigern aufzeichnen können. Dazu wird die vorhandene Methode zur Erzeugung von Programmspuren für hochparallele Anwendungen entsprechend erweitert. In dieser Untersuchung wird zuerst eine allgemeine Methodik entwickelt, mit der sich für jede API-gestützte Hardwarebeschleunigung eine Programmspur erstellen lässt. Darauf aufbauend wird eine eigene Programmierschnittstelle entwickelt, die es ermöglicht weitere leistungsrelevante Daten aufzuzeichnen. Die Umsetzung dieser Schnittstelle wird am Beispiel von NVIDIA CUPTI darstellt. Ein weiterer Teil der Arbeit beschäftigt sich mit der Darstellung von Programmspuren, welche Aufzeichnungen von den unterschiedlichen Parallelitätsebenen enthalten. Um die Einschränkungen klassischer Leistungsprofile oder Zeitachsendarstellungen zu überwinden, wird mit den parallelen Programmablaufgraphen (PPFGs) eine neue graphenbasisierte Darstellungsform eingeführt. Dieser neuartige Ansatz zeigt eine Programmspur als eine Folge von Programmzuständen mit gemeinsamen und unterchiedlichen Abläufen. So können divergierendes Programmverhalten und Lastimbalancen deutlich einfacher lokalisiert werden. Die Arbeit schließt mit der detaillierten Analyse von PIConGPU -- einer multi-hybriden Simulation aus der Plasmaphysik --, die in großem Maße von den in dieser Arbeit entwickelten Analysemöglichkeiten profiert hat.