This thesis extends the basic ordinary differential equations (ODE) course, specifically considering perturbations of ODEs. We introduce uniformly asympto- tic approximation and uniformly ordered approximation. We provide a perturba- tion-based method of computing derivatives of ODE solutions with respect to: an initial value, a parameter, and initial time. We present the method of averaging, error estimate, and a theorem about the existence and stability of a periodic so- lution to ODEs in periodic standard form. Furthermore, we apply the method of averaging to determine the period of a periodic solution of Duffing equation without forcing or damping. All the terms and methods of perturbation theory used in the thesis are accompanied with examples. 1

Ces travaux de thèse traitent de l'approximation en dimension finie de système de dimension infinie. La classe considérée est celle des systèmes hamiltoniens à ports. Nous étudions dans un premier temps les systèmes d'équations différentielles ordinaires. Sur la base d'un intégrateur énergétique, nous définissons une classe de dynamiques passives discrètes qui est invariante par interconnexion. Nous obtenons alors des conditions de stabilité (LMI) pour des dynamiques en réseau en présence de retards et d'incertitudes, et proposons une méthode de synthèse énergétique stabilisante. Ces développements ont été validés expérimentalement par la mise en oeuvre d'une commande énergétique sur un convertisseur de puissance (Buck). Nous étudions ensuite le formalisme hamiltonien en dimension infinie. Nous proposons une approximation qui combine une semi-discrétisation et un intégrateur énergétique. La composabilité mixte est étudiée et une méthode de synthèse IDA-PBC a été développée. L'ensemble des résultats obtenus sont illustrés numériquement dans le manuscrit. / This thesis work dealing with finite dimensional approximation of infinite dimension system. The class considered is that of Hamiltonian systems in ports. We study initially ordinary differential equations systems. Based on an energy integrator, we define a class of discrete passive dynamics is invariant interconnection. We obtain the stability conditions (LMI) for dynamic network in the presence of delays and uncertainties, and propose a method of stabilizing energy synthesis. These developments were experimentally validated by the implementation of an energy control a power converter (Buck). We then study the Hamiltonian formalism in infinite dimensions. We offer an approximation that combines a semi-discretization and an energy integrator. The mixed composability is studied and a method of synthesis IDA-PBC was developed. All the obtained results are numerically illustrated in the manuscript.

Le calcul des efforts hydrodynamiques de premier ordre sur un ou plusieurs corps perçant la surface libre est aujourd'hui bien maîtrisé, et plusieurs codes de calcul implémentant la méthode des singularités (dite BEM ou méthode d'élément frontière) ont été développés. Le cadre est la théorie linéarisée des écoulements potentiels à une surface libre. Dans ces codes BEM, les singularités utilisées ont la propriété intrinsèque de satisfaire à la fois l'équation de Laplace dans le domaine fluide ainsi que la condition linéarisée de surface libre. Ces singularités, dites fonctions de Green à surface libre, dans le domaine fréquentiel en profondeur infinie et sans vitesse d'avance constituent le point focal de cette thèse. Tout d'abord, les expressions mathématiques existantes pour la fonction de Green de surface libre sont examinées. Douze expressions différentes sont passées en revue et analysées. Plusieurs méthodes numériques existantes sont comparées par rapport à leur temps de calcul et leur précision. Ensuite, une série d'équations différentielles ordinaires (ODEs) pour les fonctions de Green de surface libre dans le domaine temporel et le domaine fréquentiel et leur gradient est établie. Ces ODEs peuvent être utilisées pour mieux comprendre les propriétés de la fonction de Green et peuvent constituer un moyen alternatif de calculer ces fonctions de Green et leurs dérivées. Cependant, il est difficile de résoudre numériquement ces ODEs à cause de l'existence d'une singularité à l'origine. Cette difficulté est éliminée en modifiant les ODEs par l'utilisation de nouvelles fonctions sans singularité. Les nouvelles ODEs sont ensuite écrites sous forme canonique en utilisant une nouvelle définition de la fonction vectorielle. La forme canonique peut être résolue avec les conditions initiales à l'origine puisque tous les termes impliqués sont finis. Une méthode d'expansion basée sur une série de fonctions logarithmiques et de polynômes ordinaires, très efficace pour les problèmes de basse fréquence, a également été développée pour obtenir des solutions analytiques. Enfin, la méthode basée sur les ODE pour calculer la fonction de Green est implémentée et un nouveau solveur BEM est obtenu. L'élimination des fréquences irrégulières est incluse. Le nouveau solveur est validé par comparaison des coefficients hydrodynamiques à des solutions analytiques pour une hémisphère, ainsi qu'à des résultats numériques obtenus avec un solveur commercial pour un chaland parallèlépipédique et le porte-conteneurs KCS. / The boundary element method (BEM) with constant panels is a common approach for wave-structure interaction problems. It is based on the linear potential-flow theory. It relies on the frequency-domain free-surface Green function, which is the focus of this thesis. First, the mathematical expressions and numerical methods for the frequency-domain free-surface Green function are investigated. Twelve different expressions are reviewed and analyzed. Several existing numerical methods are compared including their computational time and accuracies. Then, a series of ordinary differential equations (ODEs) for the time-domain and frequency-domain free-surface Green functions and their derivatives are derived. These ODEs can be used to better understand the properties of the Green function and can be an alternative way to calculate the Green functions and their derivatives. However, it is challenging to solve the ODEs for the frequency-domain Green function with initial conditions at the origin due to the singularity. This difficulty is removed by modifying the ODEs by using new functions free of singularity. The new ODEs are then transformed in their canonic form by using a novel definition of the vector functions. The canonic form can be solved with the initial conditions at the origin since all involved terms are finite. An expansion method based on series of logarithmic function together with ordinary polynomials which is very efficient for low frequency problems is also developed to obtain analytical solutions. Finally, the ODE-based method to calculate the Green function is implemented and an efficient BEM solver is obtained. The removal of irregular frequencies is included. The new solver is validated by comparison of hydrodynamic coefficients to analytical solutions for a heaving and surging hemisphere, and to numerical results obtained with a commercial solver for a box barge and the KCS container ship.

In the last decade, the machine learning universe embraced deep neural networks (DNNs) wholeheartedly with the advent of neural architectures such as recurrent neural networks (RNNs), convolutional neural networks (CNNs), transformers, etc. These models have empowered many applications, such as ChatGPT, Imagen, etc., and have achieved state-of-the-art (SOTA) performance on many vision, speech, and language modeling tasks. However, SOTA performance comes with various issues, such as large model size, compute-intensive training, increased inference latency, higher working memory, etc. This thesis aims at improving the resource efficiency of neural architectures, i.e., significantly reducing the computational, storage, and energy consumption of a DNN without any significant loss in performance. Towards this goal, we explore novel neural architectures as well as training algorithms that allow low-capacity models to achieve near SOTA performance. We divide this thesis into two dimensions: \textit{Efficient Low Complexity Models}, and \textit{Input Hardness Adaptive Models}. Along the first dimension, i.e., \textit{Efficient Low Complexity Models}, we improve DNN performance by addressing instabilities in the existing architectures and training methods. We propose novel neural architectures inspired by ordinary differential equations (ODEs) to reinforce input signals and attend to salient feature regions. In addition, we show that carefully designed training schemes improve the performance of existing neural networks. We divide this exploration into two parts: \textsc{(a) Efficient Low Complexity RNNs.} We improve RNN resource efficiency by addressing poor gradients, noise amplifications, and BPTT training issues. First, we improve RNNs by solving ODEs that eliminate vanishing and exploding gradients during the training. To do so, we present Incremental Recurrent Neural Networks (iRNNs) that keep track of increments in the equilibrium surface. Next, we propose Time Adaptive RNNs that mitigate the noise propagation issue in RNNs by modulating the time constants in the ODE-based transition function. We empirically demonstrate the superiority of ODE-based neural architectures over existing RNNs. Finally, we propose Forward Propagation Through Time (FPTT) algorithm for training RNNs. We show that FPTT yields significant gains compared to the more conventional Backward Propagation Through Time (BPTT) scheme. \textsc{(b) Efficient Low Complexity CNNs.} Next, we improve CNN architectures by reducing their resource usage. They require greater depth to generate high-level features, resulting in computationally expensive models. We design a novel residual block, the Global layer, that constrains the input and output features by approximately solving partial differential equations (PDEs). It yields better receptive fields than traditional convolutional blocks and thus results in shallower networks. Further, we reduce the model footprint by enforcing a novel inductive bias that formulates the output of a residual block as a spatial interpolation between high-compute anchor pixels and low-compute cheaper pixels. This results in spatially interpolated convolutional blocks (SI-CNNs) that have better compute and performance trade-offs. Finally, we propose an algorithm that enforces various distributional constraints during training in order to achieve better generalization. We refer to this scheme as distributionally constrained learning (DCL). In the second dimension, i.e., \textit{Input Hardness Adaptive Models}, we introduce the notion of the hardness of any input relative to any architecture. In the first dimension, a neural network allocates the same resources, such as compute, storage, and working memory, for all the inputs. It inherently assumes that all examples are equally hard for a model. In this dimension, we challenge this assumption using input hardness as our reasoning that some inputs are relatively easy for a network to predict compared to others. Input hardness enables us to create selective classifiers wherein a low-capacity network handles simple inputs while abstaining from a prediction on the complex inputs. Next, we create hybrid models that route the hard inputs from the low-capacity abstaining network to a high-capacity expert model. We design various architectures that adhere to this hybrid inference style. Further, input hardness enables us to selectively distill the knowledge of a high-capacity model into a low-capacity model by cleverly discarding hard inputs during the distillation procedure. Finally, we conclude this thesis by sketching out various interesting future research directions that emerge as an extension of different ideas explored in this work.

Modern numerical ordinary differential equation initial-value problem (ODE-IVP) solvers compute a piecewise polynomial approximate solution to the mathematical problem. Evaluating the mathematical problem at this approximate solution defines the defect. Corless and Corliss proposed rigorous defect control of numerical ODE-IVP. This thesis automates rigorous defect control for explicit, first-order, nonlinear ODE-IVP. Defect control is residual-based backward error analysis for ODE, a special case of Wilkinson's backward error analysis. This thesis describes a complete software implementation of the Corless and Corliss algorithm and extensive numerical studies. Basic time-stepping software is adapted to defect control and implemented. Advances in software developed for validated computing applications and advances in programming languages supporting operator overloading enable the computation of a tight rigorous enclosure of the defect evaluated at the approximate solution with Taylor models. Rigorously bounding a norm of the defect, the Corless and Corliss algorithm controls to mathematical certainty the norm of the defect to be less than a user specified tolerance over the integration interval. The validated computing software used in this thesis happens to compute a rigorous supremum norm. The defect of an approximate solution to the mathematical problem is associated with a new problem, the perturbed reference problem. This approximate solution is often the product of a numerical procedure. Nonetheless, it solves exactly the new problem including all errors. Defect control accepts the approximate solution whenever the sup-norm of the defect is less than a user specified tolerance. A user must be satisfied that the new problem is an acceptable model. / Thesis / Master of Science (MSc) / Many processes in our daily lives evolve in time, even the weather. Scientists want to predict the future makeup of the process. To do so they build models to model physical reality. Scientists design algorithms to solve these models, and the algorithm implemented in this project was designed over 25 years ago. Recent advances in mathematics and software enabled this algorithm to be implemented. Scientific software implements mathematical algorithms, and sometimes there is more than one software solution to apply to the model. The software tools developed in this project enable scientists to objectively compare solution techniques. There are two forces at play; models and software solutions. This project build software to automate the construction of the exact solution of a nearby model. That's cool.

No description available.

Les fuites non détectées dans les réseaux d'eau potable sont responsables en moyenne de la perte de 30% de l'eau transportée. Il s'avère donc primordial de pouvoir contrôler ces fuites. Pour atteindre cet objectif, la modélisation de l'écoulement de l'eau dans les conduites en tenant compte des fuites a été formulée de différente manière. La première formulation est un système d'équations différentielles ordinaires représentant des fuites constantes, réparties uniformément le long des conduites. Le système peut s'avérer être numériquement raide lorsque des organes hydrauliques sont rajoutés. Deux méthodes implicites ont été proposées pour sa résolution : la méthode de Rosenbrock et la méthode de Gear. Les résultats obtenus montrent que le débit varie linéairement le long des conduites et que les pertes en eau par unité de longueur sont identiques sur chaque conduite. La seconde formulation prend en compte la relation entre les fuites et la pression. Un système de deux équations aux dérivées partielles a été proposé. L'EDP de transport-diffusion-réaction, contenant l'opérateur du p-Laplacien, est résolue par une méthode à pas fractionnaires. Deux méthodes ont été testées. Dans la première, la réaction est couplée avec la diffusion et dans la seconde, elle est couplée avec le transport. Les résultats indiquent que les pertes en eau ne sont pas réparties de façon homogène sur le réseau. Cette formulation décrit de manière plus réaliste les réseaux d'eau potable. Enfin, le problème du contrôle du volume des fuites par action sur la pression a été étudié. Pour cela, un problème d'optimisation est résolu sous la contrainte que la pression doit être minimale pour réduire les fuites et être suffisante pour garantir un bon service aux consommateurs. Les résultats trouvés confirment que la réduction de la pression permet de réduire le volume des fuites de façon significative et que le choix de l'emplacement du ou des points de contrôle est primordial pour optimiser cette réduction. / Leakage represents a large part, in average more than 30%, of the water supplied. Consequently, it is important to control leakage in Water Distribution System (WDS). For this purpose different methods, which take leakage into account, are proposed to model the hydraulics of WDS. The first formulation considers constant leakage in a network and leads to an ordinary differential equation. It turns out to be a hydraulic stiff problem due to valve and pump operations. This equation is solved using two methods: the first one is a generalised Runge-Kutta method and the second one the Gear method. The results show that the flow rate varies linearly along a pipe and that the water loss per unit of length is identical for each pipe. Magnitude of inertia terms has also been studied. The second formulation takes pressure-dependent leakage into account. We propose to introduce partial differential equations in order to predict more accurately hydraulic flows in WDS. Thus, the physical advection-diffusion-reaction model is presented. A nonlinear operator, called p-Laplacian, related to the diffusion is included into the model. Two resolutions of this model based on a splitting method are detailed. The results confirm that losses vary nonlinearly with pressure. Finally, the leakage-control problem is studied. For this purpose, we solve an optimisation problem with the objective to minimize the distributed volume in order to reduce leakage. The condition of sufficient pressure to satisfy consumers is imposed in this optimisation. The results prove that pressure control significantly reduces leakage and that the emplacement of the valve is important to optimise this reduction.

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Nesta dissertação é apresentada uma modelagem analítica para o processo evolucionário formulado pela Teoria da Evolução por Endossimbiose representado através de uma sucessão de estágios envolvendo diferentes interações ecológicas e metábolicas entre populações de bactérias considerando tanto a dinâmica populacional como os processos produtivos dessas populações. Para tal abordagem é feito uso do sistema de equações diferenciais conhecido como sistema de Volterra-Hamilton bem como de determinados conceitos geométricos envolvendo a Teoria KCC e a Geometria Projetiva. Os principais cálculos foram realizados pelo pacote de programação algébrica FINSLER, aplicado sobre o MAPLE. / This work presents an analytical approach for modeling the evolutionary process formulated by the Serial Endosymbiosis Theory represented by a succession of stages involving different metabolic and ecological interactions among populations of bacteria considering both the population dynamics and production processes of these populations. In such approach we make use of systems of differential equations known as Volterra-Hamilton systems as well as some geometric concepts involving the KCC Theory and the Projective Geometry. The main calculations were performed by the computer algebra software FINSLER based on MAPLE.

在人口統計領域中，早期習慣將人口變化視為時間的函數，企圖以Deterministic Function來刻劃，例如：1798年Malthus提出的Malthusian Growth Model ；1825年Gompertz提出的Gompertz Model以及1838年Verhulst主張以Logistic Function描述人口成長。而近年來則是傾向於逐項分析各種因素的隨機性模型，例如：1983年Holford加入世代的APC模型；1992年Lee 和Carter提出的Lee-Carter死亡率模型以及2003年Renshaw與Haberman提出改善Lee-Carter死亡率模型的Reduction Factor模型。 人口變化主要分成自然增加與社會增加，而自然增加是為出生扣掉死亡，社會增加則為移入扣掉移出。首先，本文先不考慮遷移的部分，各別以出生與死亡人口的變化為研究對象，視其變化為一隨時間變動的動態系統，以常微分方程來刻劃。由台灣地區人口統計資料顯示，出生率或死亡率都有逐年下降的趨勢，而且隨著時間而變化加劇的傾向，使得以往使用的模型不易捕捉變化，因此我們提出「二次逼近法」，從出生、死亡人數對時間的變化率與曲度利用數值分析的方式來估計出生與死亡數，進而從中找出在此動態系統背後隱藏的規則。而後再同時考慮其他各種變項，以偏微分方程來刻劃，最後即可建立台灣地區人口變化模型。 / In early population statistics, the population changes were regarded as a function of time so that people tended to describe the variations by deterministic functions. For instance, Malthus proposed the Malthusian Growth Model in 1798; Gompertz presented Gompertz Model in 1825; Verhulst advocated using logistic function to describe an increase in population. In recent years, people tend to use the stochastic forecast method to analyse every factor term by term. For instance, the Age-Period-Cohort (APC) Model which was proposed by Holford in 1983; Lee and Carter proposed the Lee-Carter Mortality Model in 2003; and Renshaw and Haberman proposed the Reduction Factor Model in 2003 that improve the Lee-Carter Mortality Model. The population changes equal to nature and social increase, where the nature increase is the difference between birth and death population, and the social increase is the difference between immigrants and emigrants. First, we focus on natural increase rather than social increase. Moreover, we use ordinary differential equation to decribe the variation as a dynamic system over time. From the data obtained from the Ministry of Interior Taiwan, we know that the fertility and mortality has been decreasing, and the change is getting more violent year by year. Under the consideration that previous models are not able to accurately present the changes of birth and death, we proposed "second-order (or parabola) approximation method." From the variation rates and curvatures of birth and death population, we estimated the population size. Furthermore, we want to find the rule in the dynamic system. Later we will consider other factors simultaneously, and describe them by partial differential equation. Finally, the population model is constructed.

Cette thèse est consacrée à l'étude de la dynamique des systèmes complexes. Nous construisons des réseaux couplés à partir de multiples instances de systèmes dynamiques déterministes, donnés par des équations différentielles ordinaires ou des équations aux dérivées partielles de type parabolique, qui décrivent un problème d'évolution. Nous étudions le lien entre la dynamique interne à chaque nœud du réseau, les éléments de la topologie du graphe portant ce réseau, et sa dynamique globale. Nous recherchons les conditions de couplage qui favorisent une dynamique globale particulière à l'échelle du réseau, et étudions l'impact des interactions sur les bifurcations identifiées sur chaque nœud. Nous considérons en particulier des réseaux couplés de systèmes de réaction-diffusion, dont nous étudions le comportement asymptotique, en recherchant des régions positivement invariantes, et en démontrant l'existence d'attracteurs exponentiels de dimension fractale finie, à partir d'estimations d'énergie qui révèlent la nature dissipative de ces réseaux de systèmes de réaction-diffusion. Ces questions sont étudiées dans le cadre de quelques applications. En particulier, nous considérons un modèle mathématique pour l'étude géographique des réactions comportementales d'individus, au sein d'une population en situation de catastrophe. Nous présentons les éléments de modélisation associés, ainsi que son étude mathématique, avec une analyse de la stabilité des équilibres et de leurs bifurcations. Nous établissons l'importance capitale des chemins d'évacuation dans les réseaux complexes construits à partir de ce modèle, pour atteindre l'équilibre attendu de retour au comportement du quotidien pour l'ensemble de la population considérée, tout en évitant une propagation du comportement de panique. D'autre part, la recherche de solutions périodiques émergentes dans les réseaux d'oscillateurs nous amène à considérer des réseaux complexes de systèmes hamiltoniens pour lesquels nous construisons des perturbations polynomiales qui provoquent l'apparition de cycles limites, problématique liée au XVIème problème de Hilbert. / This thesis is devoted to the study of the dynamics of complex systems. We consider coupled networks built with multiple instances of deterministicdynamical systems, defined by ordinary differential equations or partial differential equations of parabolic type, which describe an evolution problem.We study the link between the internal dynamics of each node in the network, its topology, and its global dynamics. We analyze the coupling conditions which favor a particular dynamics at the network's scale, and study the impact of the interactions on the bifurcations identified on each node. In particular, we consider coupled networks of reaction-diffusion systems; we analyze their asymptotic behavior by searching positively invariant regions, and proving the existence of exponential attractors of finite fractal dimension, derived from energy estimates which suggest the dissipative nature of those networks of reaction-diffusion systems.Our framework includes the study of multiple applications. Among them, we consider a mathematical model for the geographical analysis of behavioral reactions of individuals facing a catastrophic event. We present the modeling choices that led to the study of this evolution problem, and its mathematical study, with a stability and bifurcation analysis of the equilibria. We highlight the decisive role of evacuation paths in coupled networks built from this model, in order to reach the expected equilibrium corresponding to a global return of all individuals to the daily behavior, avoiding a propagation of panic. Furthermore, the research of emergent periodic solutions in complex networks of oscillators brings us to consider coupled networks of hamiltonian systems, for which we construct polynomial perturbationswhich provoke the emergence of limit cycles, question which is related to the sixteenth Hilbert's problem.