Artificial vision is the problem of creating systems capable of processing visual information. A fundamental sub-problem of artificial vision is image segmentation, the problem of detecting a structure from a digital image. Examples of segmentation problems include the detection of a road from an aerial photograph or the determination of the boundaries of the brain's ventricles from medical imagery. The extraction of structures allows for subsequent higher-level cognitive tasks. One of them is shape comparison. For example, if the brain ventricles of a patient are segmented, can their shapes be used for diagnosis? That is to say, do the shapes of the extracted ventricles resemble more those of healthy patients or those of patients suffering from schizophrenia? This thesis deals with the problem of image segmentation and shape comparison in the mathematical framework of partial differential equations. The contribution of this thesis is threefold: 1. A technique for the segmentation of regions is proposed. A cost functional is defined for regions based on a non-parametric functional of the distribution of image intensities inside the region. This cost is constructed to favor regions that are homogeneous. Regions that are optimal with respect to that cost can be determined with limited user interaction. 2. The use of direction information is introduced for the segmentation of open curves and closed surfaces. A cost functional is defined for structures (curves or surfaces) by integrating a local, direction-dependent pattern detector along the structure. Optimal structures, corresponding to the best match with the pattern detector, can be determined using efficient algorithms. 3. A technique for shape comparison based on the Laplace equation is proposed. Given two surfaces, one-to-one correspondences are determined that allow for the characterization of local and global similarity measures. The local differences among shapes (resulting for example from a segmentation step) can be visualized for qualitative evaluation by a human expert. It can also be used for classifying shapes into, for example, normal and pathological classes.

We consider the numerical solution of discretised Hamilton-Jacobi-Bellman (HJB) equations with applications in finance. For the discrete linear complementarity problem arising in American option pricing, we study a policy iteration method. We show, analytically and numerically, that, in standard situations, the computational cost of this approach is comparable to that of European option pricing. We also characterise the shortcomings of policy iteration, providing a lower bound for the number of steps required when having inaccurate initial data. For discretised HJB equations with a finite control set, we propose a penalty approach. The accuracy of the penalty approximation is of first order in the penalty parameter, and we present a Newton-type iterative solver terminating after finitely many steps with a solution to the penalised equation. For discretised HJB equations and discretised HJB obstacle problems with compact control sets, we also introduce penalty approximations. In both cases, the approximation accuracy is of first order in the penalty parameter. We again design Newton-type methods for the solution of the penalised equations. For the penalised HJB equation, the iterative solver has monotone global convergence. For the penalised HJB obstacle problem, the iterative solver has local quadratic convergence. We carefully benchmark all our numerical schemes against current state-of-the-art techniques, demonstrating competitiveness.

Orientador: Bruto Max Pimentel Escobar / Co-orientador: / Banca:Andrey Yuryevich Mikhaylov / Banca: Edmundo Capelas de Oliveira / Resumo: A teoria de Hamilton-Jacobi geralmente é apresentada como uma extensão da teoria de Hamilton através das transformações canônicas. No entanto, o matemático Constantin Carathéodory mostrou que essa teoria, sua existência e validade, independem do formalismo hamiltoniano. Neste trabalho, apresentaremos a abordagem de Carathéodory para a teoria de Hamilton-Jacobi. Partindo desse procedimento, construiremos uma teoria de vínculos para que se possa resolver problemas com vínculos involutivos e não-involutivos. Para isso, analisaremos a integrabilidade das equações e introduziremos a operação dos parênteses generalizados que, no lugar do parênteses de Poisson, passará a descrever a dinâmica de sistemas vinculados. Mostraremos uma aplicação dessa teoria de vínculos no modelo BF da teoria de campos. Para finalizar, trataremos da Termodinâmica Axiomática de Carathéodory e também da teoria de Hamilton-Jacobi na Termodinâmica, o que é válido para ilustrar a grande abrangência desse formalismo / Abstract: The Hamilton-Jacobi theory is usually presented as an extension of the Hamilton's theory through the canonical transformations. However, the mathematician Constantin Carathéodory showed this theory, its existence and validity, is independent of the Hamiltonian formalism. In this work, we present the Caratheodory's approach to the Hamilton-Jacobi theory. From this procedure, we build a theory of constraints which can solve problems with involutive and non-involutive constraints. For this, we analyze the integrability of the equations and introduce the operation of the generalized brackets that, instead of Poisson brackets, will describe the dynamics of constrained systems. We show an application of this theory in BF model of the field theory. Finally, we will discuss the Carathéodory's Axiomatic Thermodynamics and also show the Hamilton-Jacobi theory in Thermodynamics, which is valid to illustrate the wide coverage of this formalism / Mestre

Orientador: Geraldo Nunes Silva / Banca: Iguer Luis Domini dos Santos / Banca: Marko Antonio Rojas Medar / Resumo: Neste trabalho caracterizaremos sistemas dinâmicos na forma dos chamados domínios estratificados. Bressan e Hong[9] foram os primeiros a definir e trabalhar em domínios estratificados. Grosso modo, estes são uma coleção de domínios disjuntos, cada um tendo sua própria dinâmica; mas não se exige que seus domínios sejam proximamente suaves e nem wedged. Estes termos foram introduzidos por P. Wolenski e R. Barnard em[10]. Primeiramente, estabeleceremos condições Hamiltonianos para caracterizar invariância fraca e forte para sistemas não Lipschitz em domínios estratificados. Depois, estudamos condições Hamiltonianas para sistemas fracamente e fortemente decrescentes e apresentamos condições que garantem a estabilidade assintótica global para uma dinâmica estratificada e finalmente apresentamos o problema tipo Mayer em domínios estratificados em que mostramos que a função valor e a única solução semicontínua inferior de uma generalização adequada da clássica equação Hamilton-Jacobi-Bellman, para a dinâmica estratificada / Abstract: In this work will characterize dynamical systems in the form of the so-called strati ed domain. Bressan and Hong [9] were the rst to de ne and work in strati ed domains. Roughly speaking, these are a collection of disjoint domains, each having its own dynamics; but not requiring that their domains are proximally smooth and not wedged. These terms were introduced by P. Wolenski and R. Barnard in [10]. At rst, we will establish Hamiltonian conditions to characterize weak and strong invariance for systems non-Lipschitz in strati ed domains. Secondly, we study the Hamiltonian conditions for systems weakly and strongly de- creasing and present conditions that guarantee global asymptotic stability for a strati ed dynamics and nally we present the problem Mayer type in strati ed domains where we show that the value function is the unique lower semicontinuous solution of an appropriate generalization of the classical Hamilton-Jacobi-Bellman equation for strati ed dynamics / Mestre

The present thesis considers the development and analysis of arbitrary Lagrangian-Eulerian discontinuous Galerkin (ALE-DG) methods with time-dependent approximation spaces for conservation laws and the Hamilton-Jacobi equations. Fundamentals about conservation laws, Hamilton-Jacobi equations and discontinuous Galerkin methods are presented. In particular, issues in the development of discontinuous Galerkin (DG) methods for the Hamilton-Jacobi equations are discussed. The development of the ALE-DG methods based on the assumption that the distribution of the grid points is explicitly given for an upcoming time level. This assumption allows to construct a time-dependent local affine linear mapping to a reference cell and a time-dependent finite element test function space. In addition, a version of Reynolds’ transport theorem can be proven. For the fully-discrete ALE-DG method for nonlinear scalar conservation laws the geometric conservation law and a local maximum principle are proven. Furthermore, conditions for slope limiters are stated. These conditions ensure the total variation stability of the method. In addition, entropy stability is discussed. For the corresponding semi-discrete ALE-DG method, error estimates are proven. If a piecewise $\mathcal{P}^{k}$ polynomial approximation space is used on the reference cell, the sub-optimal $\left(k+\frac{1}{2}\right)$ convergence for monotone fuxes and the optimal $(k+1)$ convergence for an upwind flux are proven in the $\mathrm{L}^{2}$-norm. The capability of the method is shown by numerical examples for nonlinear conservation laws. Likewise, for the semi-discrete ALE-DG method for nonlinear Hamilton-Jacobi equations, error estimates are proven. In the one dimensional case the optimal $\left(k+1\right)$ convergence and in the two dimensional case the sub-optimal $\left(k+\frac{1}{2}\right)$ convergence are proven in the $\mathrm{L}^{2}$-norm, if a piecewise $\mathcal{P}^{k}$ polynomial approximation space is used on the reference cell. For the fullydiscrete method, the geometric conservation is proven and for the piecewise constant forward Euler step the convergence of the method to the unique physical relevant solution is discussed. / Die vorliegende Arbeit beschäftigt sich mit der Entwicklung und Analyse von arbitrar Lagrangian-Eulerian discontinuous Galerkin (ALE-DG) Methoden mit zeitabhängigen Testfunktionen Räumen für Erhaltungs- und Hamilton-Jacobi Gleichungen. Grundlagen über Erhaltungsgleichungen, Hamilton-Jacobi Gleichungen und discontinuous Galerkin Methoden werden präsentiert. Insbesondere werden Probleme bei der Entwicklung von discontinuous Galerkin Methoden für die Hamilton-Jacobi Gleichungen untersucht. Die Entwicklung der ALE-DG Methode basiert auf der Annahme, dass die Verteilung der Gitterpunkte zu einem kommenden Zeitpunkt explizit gegeben ist. Diese Annahme ermöglicht die Konstruktion einer zeitabhängigen lokal affin-linearen Abbildung auf eine Referenzzelle und eines zeitabhängigen Testfunktionen Raums. Zusätzlich kann eine Version des Reynolds’schen Transportsatzes gezeigt werden. Für die vollständig diskretisierte ALE-DG Methode für nichtlineare Erhaltungsgleichungen werden der geometrischen Erhaltungssatz und ein lokales Maximumprinzip bewiesen. Des Weiteren werden Bedingungen für Limiter angegeben. Diese Bedingungen sichern die Stabilität der Methode im Sinne der totalen Variation. Zusätzlich wird die Entropie-Stabilität der Methode diskutiert. Für die zugehörige semi-diskretisierte ALE-DG Methode werden Fehlerabschätzungen gezeigt. Wenn auf der Referenzzelle ein Testfunktionen Raum, der stückweise Polynome vom Grad $k$ enthält verwendet wird, kann für einen monotonen Fluss die suboptimale Konvergenzordnung $\left(k+\frac{1}{2}\right)$ und für einen upwind Fluss die optimale Konvergenzordnung $\left(k+1\right)$ in der $\mathrm{L}^{2}$-Norm gezeigt werden. Die Leistungsfähigkeit der Methode wird anhand numerischer Beispiele für nichtlineare Erhaltungsgleichungen untersucht. Ebenso werden für die semi-diskretisierte ALE-DG Methode für nichtlineare Hamilton-Jacobi Gleichungen Fehlerabschätzungen gezeigt. Wenn auf der Referenzzelle ein Testfunktionen Raum, der stückweise Polynome vom Grad k enthält verwendet wird, kann im eindimensionalen Fall die optimale Konvergenzordnung $\left(k+1\right)$ und im zweidimensionalen Fall die suboptimale Konvergenzordnung $\left(k+\frac{1}{2}\right)$ in der $\mathrm{L}^{2}$-Norm gezeigt werden. Für die vollständig diskretisierte ALE-DG Methode werden der geometrischen Erhaltungssatz bewiesen und für die stückweise konstante explizite Euler Diskretisierung wird die Konvergenz gegen die eindeutige physikalisch relevante Lösung diskutiert.

The solution of optimal control problems through the Hamilton-Jacobi-Bellman (HJB) equation offers guaranteed satisfaction of both the necessary and sufficient conditions for optimality. However, finding an exact solution to the HJB equation is a near impossible task for many optimal control problems. This thesis presents an approximation method for solving finite-horizon optimal control problems involving nonlinear dynamical systems. The method uses finite-order approximations of the partial derivatives of the cost-to-go function, and successive higher-order differentiations of the HJB equation. Natural byproducts of the proposed method provide sensitivities of the controls to changes in the initial states, which can be used to approximate the solution to neighboring optimal control problems. For highly nonlinear problems, the method is modified to calculate control sensitivities about a nominal trajectory. In this framework, the method is shown to provide accurate control sensitivities at much lower orders of approximation. Several numerical examples are presented to illustrate both applications of the approximation method.

No description available.

The method of separation of variables facilitates the integration of the Hamilton-Jacobi equation by reducing its solution to a series of quadratures in the separable coordinates. The case in which the metric tensor is diagonal in the separable coordinates, that is, orthogonal separability, is fundamental. Recent theory by Benenti has established a concise geometric (coordinate-independent) characterisation of orthogonal separability of the Hamilton-Jacobi equation on a pseudoRiemannian manifold. It generalises an approach initiated by Eisenhart and developed by Kalnins and Miller. Benenti has shown that the orthogonal separability of a system via a point transformation is equivalent to the existence of a Killing tensor with real simple eigen values and orthogonally integrable eigenvectors. Applying a moving frame formalism, we develop a method that produces the orthogonal separable coordinates for low dimensional Hamiltonian systems. The method is applied to a two dimensional Riemannian manifold of arbitrary curvature. As an illustration, we investigate Euclidean 2-space, and the two dimensional surfaces of constant curvature, recovering known results. Using our formalism, we also derive the known superseparable potentials for Euclidean 2-space. Some of the original results presented in this thesis were announced in [8, 9, 10].

The method of separation of variables facilitates the integration of the Hamilton-Jacobi equation by reducing its solution to a series of quadratures in the separable coordinates. The case in which the metric tensor is diagonal in the separable coordinates, that is, orthogonal separability, is fundamental. Recent theory by Benenti has established a concise geometric (coordinate-independent) characterisation of orthogonal separability of the Hamilton-Jacobi equation on a pseudoRiemannian manifold. It generalises an approach initiated by Eisenhart and developed by Kalnins and Miller. Benenti has shown that the orthogonal separability of a system via a point transformation is equivalent to the existence of a Killing tensor with real simple eigen values and orthogonally integrable eigenvectors. Applying a moving frame formalism, we develop a method that produces the orthogonal separable coordinates for low dimensional Hamiltonian systems. The method is applied to a two dimensional Riemannian manifold of arbitrary curvature. As an illustration, we investigate Euclidean 2-space, and the two dimensional surfaces of constant curvature, recovering known results. Using our formalism, we also derive the known superseparable potentials for Euclidean 2-space. Some of the original results presented in this thesis were announced in [8, 9, 10].

The solution of optimal control problems through the Hamilton-Jacobi-Bellman (HJB) equation offers guaranteed satisfaction of both the necessary and sufficient conditions for optimality. However, finding an exact solution to the HJB equation is a near impossible task for many optimal control problems. This thesis presents an approximation method for solving finite-horizon optimal control problems involving nonlinear dynamical systems. The method uses finite-order approximations of the partial derivatives of the cost-to-go function, and successive higher-order differentiations of the HJB equation. Natural byproducts of the proposed method provide sensitivities of the controls to changes in the initial states, which can be used to approximate the solution to neighboring optimal control problems. For highly nonlinear problems, the method is modified to calculate control sensitivities about a nominal trajectory. In this framework, the method is shown to provide accurate control sensitivities at much lower orders of approximation. Several numerical examples are presented to illustrate both applications of the approximation method.