Relaxation oscillations are highly non-linear oscillations, which appear to feature many important biological phenomena such as heartbeat, neuronal activity, and population cycles of predator-prey type. They are characterized by repeated switching of slow and fast motions and occur naturally in singularly perturbed ordinary differential equations, which exhibit dynamics on different time scales. Traditionally, slow-fast systems and the related oscillatory phenomena -- such as relaxation oscillations -- have been studied by the method of the matched asymptotic expansions, techniques from non-standard analysis, and recently a more qualitative approach known as geometric singular perturbation theory. It turns out that relaxation oscillations can be found in a more general setting; in particular, in slow-fast systems, which are not written in the standard form. Systems in which separation into slow and fast variables is not given a priori, arise frequently in applications. Many of these systems include additionally various parameters of different orders of magnitude and complicated (non-polynomial) non-linearities. This poses several mathematical challenges, since the application of singular perturbation arguments is not at all straightforward. For that reason most of such systems have been studied only numerically guided by phase-space analysis arguments or analyzed in a rather non-rigorous way. It turns out that the main idea of singular perturbation approach can also be applied in such non-standard cases. This thesis is concerned with the application of concepts from geometric singular perturbation theory and geometric desingularization based on the blow-up method to the study of relaxation oscillations in slow-fast systems beyond the standard form. A detailed geometric analysis of oscillatory mechanisms in three mathematical models describing biochemical processes is presented. In all the three cases the aim is to detect the presence of an isolated periodic movement represented by a limit cycle. By using geometric arguments from the perspective of dynamical systems theory and geometric desingularization based on the blow-up method analytic proofs of the existence of limit cycles in the models are provided. This work shows -- in the context of non-trivial applications -- that the geometric approach, in particular the blow-up method, is valuable for the understanding of the dynamics of systems with no explicit splitting into slow and fast variables, and for systems depending singularly on several parameters.

Dans cette thèse, nous étudions les propriétés de synchronisation d'oscillateurs lents-rapides inspirés de la neuroendocrinologie et des neurosciences, en se concentrant sur les effets des phénomènes de type canard et bifurcations dynamiques sur le comportement collectif.Nous partons d'un système de dimension 4 qui représente les caractéristiques dynamiques qualitatives et quantitatives du profil de sécrétion de la neurohormone GnRH (gonadotropin releasing hormone) au cours d'un cycle ovarien. Ce modèle est constitué de deux oscillateurs de FitzHugh-Nagumo avec pour chacun des échelles de temps différentes. Le couplage unidirectionnel de l'oscillateur lent (représentant l'activité moyenne d'une population de neurones régulateurs) vers l'oscillateur rapide (représentant l'activité moyenne d'une population de neurones sécréteurs) donne une structure à trois échelles de temps. Le comportement de l'oscillateur rapide est caractérisé par une alternance entre un régime de type cycle de relaxation et un régime de quasi-stationnaire qui induit des transitions de type canard dans le modèle ; ces transitions ont un fort impact sur le modèle de sécrétion du système de dimension 4. Nous proposons un premier pas supplémentaire dans la modélisation multi-échelles (en espace) du système GnRH, c'est-à-dire que nous étendons le système original à 6 dimensions en considérant deux sous-populations distinctes de neurones sécréteurs recevant le même signal des neurones de régulation. Cette étape nous permet de enrichir les motifs possibles de sécrétion de GnRH tout en gardant un cadre dynamique compact et en préservant la séquence des événements neuro-sécréteurs capturés par le modèle de dimension 4, à la fois qualitativement et quantitativement.Une première analyse du modèle GnRH étendu à 6 dimensions est présentée dans le Chapitre 2, où nous montrons à l'aide d'un système minimal de dimension 5 l'existence de trajectoires de type canard dans des systèmes lents-rapides couplés présentant des points pseudo-stationnaires. Le couplage provoque la séparation des trajectoires correspondant à chaque sécréteur qui se retrouvent de chaque côté du canard maximal (associé soit à un point pseudo-stationnaire de type noeud soit à un pseudo-col). Nous explorons les rapports entre les canards en présence et le couplage, ainsi que leur impact sur les motifs de sécrétion collective du modèle de dimension 6. Nous identifions deux sources différentes de (dé)synchronisation due aux canards dans les événements sécrétoires, qui dépendent du type de point pseudo-stationnaire sous-jacent.Dans le Chapitre 3, nous proposons une modélisation possible des comportements complexes de sécrétion de GnRH qui ne sont pas capturés par le modèle de dimension 4, à savoir, une décharge avec 2 ``bosses'' et une désynchronisation partielle avant la décharge, en utilisant le modèle de dimension 6 précédemment construit. Pour obtenir une décharge avec deux bosses, il est essentiel d'utiliser des fonctions de couplage asymétriques dépendant du régulateur ainsi que d'introduire de l'hétérogénéité dans les sous-populations de sécréteurs. Pendant le régime pulsatile, il apparaît que le signal régulateur varie lentement et, ce faisant, provoque une bifurcation dynamique qui est responsable de la perte de synchronie dans le cas de sécréteurs non identiques et asymétriquement couplés. Nous introduisons des outils analytiques et numériques pour façonner et quantifier ces caractéristiques supplémentaires et les intégrer dans le profil complet de sécrétion. / This dissertation investigates synchronization properties of slow-fast oscillators inspired from neuroendocrinology and neuronal dynamics, focusing on the effects of canard phenomena and dynamic bifurcations on the collective behavior. We start from a 4-dimensional system which accounts for the qualitative and quantitative dynamical features of the secretion pattern of the neurohormone GnRH (gonadotropin releasing hormone) along a whole ovarian cycle. This model involves 2 FitzHugh-Nagumo oscillators with different timescales. Unidirectional coupling from the slow oscillator (representing the mean-field activity of a population of regulating neurons) to the fast oscillator (representing the mean-field activity of a population of the secreting neurons) gives a three timescale structure. The behavior of the fast oscillator is characterized by an alternation between a relaxation cycle and a quasi-stationary state which introduces canard-mediated transitions in the model; these transitions have a strong impact on the secretion pattern of the 4-dimensional system. We make a first step forward in multiscale modeling (in space) of the GnRH system, namely, we extend the original system to 6 dimensions by considering two distinct subpopulations of secreting neurons receiving the same signal from the regulating neurons. This step allows us to enrich further the GnRH secretion pattern while keeping a compact dynamic framework and preserving the sequence of neurosecretory events captured by the 4-dimensional model, both qualitatively and quantitatively. An initial analysis of the extended 6-dimensional GnRH model is presented in Chapter 2, where we prove using a 5D minimal model the existence of canard trajectories in coupled systems with folded singularities. Coupling causes separation of trajectories corresponding to each secretor by driving them to different sides of the maximal canard (associated with either a folded-node or a folded-saddle singularity). We explore the impact of the relationship between canard structures and coupling on the collective secretion pattern of the 6-dimensional model. We identify two different sources of canard-mediated (de)synchronization in the secretory events, which depend on the type of underlying folded singularity. In Chapter 3, we attempt to model complex behaviors of the GnRH secretion not captured by the 4-dimensional model, namely, a surge with 2 bumps and partial desynchronization before the surge, by using the 6-dimensional model previously constructed. Regulatory-dependent asymmetric coupling functions and heterogeneity in the secretor subpopulations are essential for obtaining such a 2-bump surge. During the pulsatile regime, we find that the slowly varying regulatory signal causes a dynamic bifurcation, which is responsible for loss of synchrony in asymmetrically coupled nonidentical secretors. We introduce analytic and numerical tools to shape and quantify the additional features embedded within the whole secretion pattern.

In der vorliegenden Arbeit betrachten wir ein System nichtautonomer gewöhnlicher Differentialgleichungen, das aus zwei gekoppelten Teilsystemen besteht. Die Teilsysteme bestehen aus langsamen bzw. schnellen Variablen, wobei die Zeitskalierung durch Multiplikation der rechten Seite eines Teilsystems mit einem kleinen Faktor erzeugt wird. Das Ziel unserer Untersuchungen besteht im Nachweis der Existenz einer Integralmannigfaltigkeit, mit deren Hilfe die schnellen Variablen eliminiert werden können. Dabei verzichten wir auf die übliche Annahme einer Dichotomiebedingung und ersetzen diese durch die Hinzunahme eines zusätzlichen Steuervektors. Wir beweisen, dass unter gewissen Voraussetzungen über die rechten Seiten der Teilsysteme ein eindeutiger Steuervektor existiert, der die Existenz der gewünschten Integralmannigfaltigkeit impliziert. Das Prinzip des Nachweises einer solchen beschränkten Integralmannigfaltigkeit basiert auf dem Zusammenkleben von anziehenden und abstossenden invarianten Mannigfaltigkeiten. In der Arbeit wird die Glattheit dieser Mannigfaltigkeit sowie deren asymptotische Entwicklung nach dem kleinen Parameter untersucht. / This work is devoted to nonautonomous slow-fast systems of ordinary differential equation without dichotomy. We are interested in the existence of a slow integral manifold in order to eliminate the fast variables. The peculiarity of the problem under consideration is that the right hand side of the system depends on some parameter vector which can be considered as a control to be determined in order to guarantee the existence of an integral manifold consisting of canard trajectories. We call the vector function as gluing function. We prove that under some conditions on the right hand side of the system there exists a unique gluing function such that the system has a slow integral manifold. We investigate the problems of asymptotic expansions of the integral manifold and the gluing function, and study their smoothness.

Relaxation oscillations are highly non-linear oscillations, which appear to feature many important biological phenomena such as heartbeat, neuronal activity, and population cycles of predator-prey type. They are characterized by repeated switching of slow and fast motions and occur naturally in singularly perturbed ordinary differential equations, which exhibit dynamics on different time scales. Traditionally, slow-fast systems and the related oscillatory phenomena -- such as relaxation oscillations -- have been studied by the method of the matched asymptotic expansions, techniques from non-standard analysis, and recently a more qualitative approach known as geometric singular perturbation theory. It turns out that relaxation oscillations can be found in a more general setting; in particular, in slow-fast systems, which are not written in the standard form. Systems in which separation into slow and fast variables is not given a priori, arise frequently in applications. Many of these systems include additionally various parameters of different orders of magnitude and complicated (non-polynomial) non-linearities. This poses several mathematical challenges, since the application of singular perturbation arguments is not at all straightforward. For that reason most of such systems have been studied only numerically guided by phase-space analysis arguments or analyzed in a rather non-rigorous way. It turns out that the main idea of singular perturbation approach can also be applied in such non-standard cases. This thesis is concerned with the application of concepts from geometric singular perturbation theory and geometric desingularization based on the blow-up method to the study of relaxation oscillations in slow-fast systems beyond the standard form. A detailed geometric analysis of oscillatory mechanisms in three mathematical models describing biochemical processes is presented. In all the three cases the aim is to detect the presence of an isolated periodic movement represented by a limit cycle. By using geometric arguments from the perspective of dynamical systems theory and geometric desingularization based on the blow-up method analytic proofs of the existence of limit cycles in the models are provided. This work shows -- in the context of non-trivial applications -- that the geometric approach, in particular the blow-up method, is valuable for the understanding of the dynamics of systems with no explicit splitting into slow and fast variables, and for systems depending singularly on several parameters.