Sistemas diferenciais quadráticos planares estão presentes em muitas áreas da matemática aplicada. Embora mais de mil artigos tenham sido publicados sobre os sistemas quadráticos ainda resta muito a se conhecer sobre esses sistemas. Problemas clássicos, e em particular o XVI problema de Hilbert, estão ainda em aberto para essa família. Um dos objetivos dos pesquisadores contemporâneos é obter a classificação topológica completa dos sistemas quadráticos. Devido ao grande número de parâmetros (essa família possui doze parâmetros e, aplicando transformações afins e reescala do tempo, reduzimos esse número a cinco, sendo ainda um número grande para se trabalhar) usualmente subclasses são consideradas nas investigações realizadas. Quando características específicas são levadas em consideração, o número de parâmetros é reduzido e o estudo se torna possível. Nesta tese estudamos principalmente duas subfamílias de sistemas quadráticos: a primeira possuindo um nó triplo semielemental e a segunda possuindo uma selanó semi elemental finita e uma selanó semielemental infinita formada pela colisão de uma sela infinita com um nó infinito. Os diagramas de bifurcação para ambas as famílias são tridimensionais. A família tendo um nó triplo gera 28 retratos de fase topologicamente distintos, enquanto o fecho da família tendo as selasnós dentro do espaço de bifurcação de sua forma normal gera 417. Polinômios invariantes são usados para construir os conjuntos de bifurcação e os retratos de fase topologicamente distintos são representados no disco de Poincaré. Os conjuntos de bifurcação são a união de superfícies algébricas e superfícies cuja presença foi detectada numericamente. Ainda nesta tese, apresentamos todos os retratos de fase de um sistema diferencial conhecido como modelo do tipo SIS (sistema suscetívelinfectadosuscetível, muito comum na matemática aplicada) e a classificação dos sistemas quadráticos possuindo hipérboles invariantes. Ambos sistemas foram investigados usando de polinômios invariantes afins. / Planar quadratic differential systems occur in many areas of applied mathematics. Although more than one thousand papers have been written on these systems, a complete understanding of this family is still missing. Classical problems, and in particular Hilberts 16th problem, are still open for this family. One of the goals of recent researchers is the topological classification of quadratic systems. As this attempt is not possible in the whole class due to the large number of parameters (twelve, but, after affine transformations and time rescaling, we arrive at families with five parameters, which is still a large number), many subclasses are considered and studied. Specific characteristics are taken into account and this implies a decrease in the number of parameters, which makes possible the study. In this thesis we mainly study two subfamilies of quadratic systems: the first one possessing a finite semielemental triple node and the second one possessing a finite semielemental saddlenode and an infinite semielemental saddlenode formed by the collision of an infinite saddle with an infinite node. The bifurcation diagram for both families are tridimensional. The family having the triple node yields 28 topologically distinct phase portraits, whereas the closure of the family having the saddlenodes within the bifurcation space of its normal form yields 417. Invariant polynomials are used to construct the bifurcation sets and the phase portraits are represented on the Poincaré disk. The bifurcation sets are the union of algebraic surfaces and surfaces whose presence was detected numerically. Moreover, we also present the analysis of a differential system known as SIS model (this kind of systems are easily found in applied mathematics) and the complete classification of quadratic systems possessing invariant hyperbolas.

Die vorliegende Arbeit behandelt die kollektive Dynamik identischer Klasse-I-anregbarer Elemente. Diese können im Rahmen der nichtlinearen Dynamik als Systeme nahe einer Sattel-Knoten-Bifurkation auf einem invarianten Kreis beschrieben werden. Der Fokus der Arbeit liegt auf dem Studium aktiver Rotatoren als Prototypen solcher Elemente. In Teil eins der Arbeit besprechen wir das klassische Modell abstoßend gekoppelter aktiver Rotatoren von Shinomoto und Kuramoto und generalisieren es indem wir höhere Fourier-Moden in der internen Dynamik der Rotatoren berücksichtigen. Wir besprechen außerdem die mathematischen Methoden die wir zur Untersuchung des Aktive-Rotatoren-Modells verwenden. In Teil zwei untersuchen wir Existenz und Stabilität periodischer Zwei-Cluster-Lösungen für generalisierte aktive Rotatoren und beweisen anschließend die Existenz eines Kontinuums periodischer Lösungen für eine Klasse Watanabe-Strogatz-integrabler Systeme zu denen insbesondere das klassische Aktive-Rotatoren-Modell gehört und zeigen dass (i) das Kontinuum eine normal-anziehende invariante Mannigfaltigkeit bildet und (ii) eine der auftretenden periodischen Lösungen Splay-State-Dynamik besitzt. Danach entwickeln wir mit Hilfe der Averaging-Methode eine Störungstheorie für solche Systeme. Mit dieser können wir Rückschlüsse auf die asymptotische Dynamik des generalisierten Aktive-Rotatoren-Modells ziehen. Als Hauptergebnis stellen wir fest dass sowohl periodische Zwei-Cluster-Lösungen als auch Splay States robuste Lösungen für das Aktive-Rotatoren-Modell darstellen. Wir untersuchen außerdem einen "Stabilitätstransfer" zwischen diesen Lösungen durch sogenannte Broken-Symmetry States. In Teil drei untersuchen wir Ensembles gekoppelter Morris-Lecar-Neuronen und stellen fest, dass deren asymptotische Dynamik der der aktiven Rotatoren vergleichbar ist was nahelegt dass die Ergebnisse aus Teil zwei ein qualitatives Bild für solch kompliziertere und realistischere Neuronenmodelle liefern. / We study the collective dynamics of class I excitable elements, which can be described within the theory of nonlinear dynamics as systems close to a saddle-node bifurcation on an invariant circle. The focus of the thesis lies on the study of active rotators as a prototype for such elements. In part one of the thesis, we motivate the classic model of repulsively coupled active rotators by Shinomoto and Kuramoto and generalize it by considering higher-order Fourier modes in the on-site dynamics of the rotators. We also discuss the mathematical methods which our work relies on, in particular the concept of Watanabe-Strogatz (WS) integrability which allows to describe systems of identical angular variables in terms of Möbius transformations. In part two, we investigate the existence and stability of periodic two-cluster states for generalized active rotators and prove the existence of a continuum of periodic orbits for a class of WS-integrable systems which includes, in particular, the classic active rotator model. We show that (i) this continuum constitutes a normally attracting invariant manifold and that (ii) one of the solutions yields splay state dynamics. We then develop a perturbation theory for such systems, based on the averaging method. By this approach, we can deduce the asymptotic dynamics of the generalized active rotator model. As a main result, we find that periodic two-cluster states and splay states are robust periodic solutions for systems of identical active rotators. We also investigate a 'transfer of stability' between these solutions by means of so-called broken-symmetry states. In part three, we study ensembles of higher-dimensional class I excitable elements in the form of Morris-Lecar neurons and find the asymptotic dynamics of such systems to be similar to those of active rotators, which suggests that our results from part two yield a suitable qualitative description for more complicated and realistic neural models.

[EN] In this thesis, invariant-set theory is used to study the stability and feasibility of constrained scenario-based predictive controllers for Markov-jump linear systems. In the underlying optimisation problem of the predictive controllers technique, considers all possible future realisations of certain variables (uncertainty, disturbances, operating mode) or just a subset of those. Two different scenarios denoted as not risky and risky are studied. In the former, the trajectories of the system with initial states belonging to certain invariant sets, converge (in mean square sense) to the origin or an invariant neighbourhood of it with 100% probability. In such cases, the conditions that scenario trees must meet in order to guarantee stability and feasibility of the optimisation problem are analysed. Afterwards, the scenario-based predictive controller for Markov-jump linear systems under hard constraints and no disturbances is formulated. A study is presented for risky scenarios to determine sequence-dependent controllable sets, for which there exists a control law such that the system can be driven to the origin only for a particular realisation of uncertainty, disturbances, etc. A control law (optimal for disturbances-free systems and suboptimal for disturbed systems) able to steer the system to the origin with a probability less than 100% (denoted as reliability bound), is proposed for states belonging to those regions. Note that closed-loop unstable systems have zero reliability bound. Hence, an algorithm to determine the mean-time to failure is developed. In this context, failure means a violation in the constraints of the process' states and/or inputs in a future time. / [ES] La presente tesis emplea la teoría de conjuntos invariantes para el estudio de estabilidad y factibilidad de controladores predictivos basados en escenarios para sistemas lineales con saltos markovianos sujetos a restricciones. En el problema de optimización subyacente a la técnica de controladores predictivos, se consideran bien sea todas las posibles realizaciones futuras de una variable (incertidumbres, perturbaciones, modo de funcionamiento) o solo un subconjunto de estas. Se estudian dos escenarios diferentes, denotados como: a) escenarios no arriesgados y b) escenarios arriesgados, entendiéndose como no arriesgados, aquellos en donde las trayectorias del sistema con estados iniciales pertenecientes a ciertos conjuntos invariantes, convergen --en media-- al origen o a una vecindad invariante de este con un 100% de probabilidad. Para estos casos, se presenta un análisis de las condiciones que deben cumplir los árboles de escenarios para garantizar estabilidad --en media-- y factibilidad del problema de optimización. Luego se formula el control predictivo basado en escenarios para sistema lineales con saltos markovianos sujeto a restricciones y en ausencia de perturbaciones. En presencia de escenarios arriesgados, se propone el cálculo de conjuntos controlables dependientes de secuencias para los cuales existen una ley de control tal que el sistema puede ser conducido al origen, solo para una realización en particular de la incertidumbre, perturbaciones, etc. Para estados pertenecientes a estos conjuntos, se propone una ley de control (óptima para el caso de sistemas libres de perturbaciones y, subóptima para sistemas perturbados) capaz de dirigir el sistema al origen con una probabilidad menor al 100%, dicha probabilidad es denotada como cota de confiabilidad. Sistemas inestables en lazo cerrado tienen cota de confiabilidad igual a cero, por consiguiente se diseña un algoritmo que determina el tiempo medio para fallar. En este contexto, un fallo se entiende como la violación de las restricciones en los estados y/o entradas del proceso en algún instante de tiempo futuro. / [CA] La present tesi empra la teoria de conjunts invariants per a l'estudi d'estabili-tat i factibilidad de controladors predictius basats en escenaris per a sistemes lineals amb salts markovians subjectes a restriccions. En el problema d'optimit-zació subjacent a la tècnica de controladors predictius, es consideren bé siga totes les possibles realitzacions futures d'una variable (incerteses, pertorbacions, modes de funcionament) o només un subconjunt d'aquestes. S'estudien dos escenaris diferents, denotats com a escenaris no arriscats i arriscats, entenent-se com no arriscats, aquells on les trajectòries del sistema amb estats inicials pertanyents a certs conjunts invariants, convergeixen --en mitjana-- a l'origen o a un veïnatge invariant d'est amb un 100% de probabilitat. Per a aquests casos, es presenta una anàlisi de les condicions que han de complir els arbres d'escenaris per a garantir estabilitat --en mitjana-- i factibilidad del problema d'optimització. Després es formula el control predictiu basat en escenaris per a sistema lineals amb salts markovians subjecte a restriccions i en absència de pertorbacions. En presència d'escenaris arriscats, es proposa el càlcul de conjunts controlables dependents de seqüències per als quals existeix una llei de control tal que el sistema pot ser conduït a l'origen, solament per a una realització en particular de l'incertesa, pertorbacions, etc. Per a estats pertanyents a aquests conjunts, es proposa una llei de control (òptima per al cas de sistemes lliures de pertorbacions i, subóptima per a sistemes pertorbats) capaç de dirigir el sistema cap a l'origen amb una probabilitat menor del 100%, aquesta probabilitat és denotada com a cota de confiabilitat. Sistemes inestables en llaç tancat tenen cota de confiabilitat igual a zero, per tant es dissenya un algoritme que determina el temps mitjà per a fallar. En aquest context, una fallada s'entén com la violació de les restriccions en els estats i/o entrades del procés en algun instant de temps futur. / Hernández Mejías, MA. (2016). CONTROL PREDICTIVO BASADO EN ESCENARIOS PARA SISTEMAS LINEALES CON SALTOS MARKOVIANOS [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/68512