Dans cette thèse, l’étude de plusieurs modèles de systèmes magnétiques frustrés a été couverte. Leur racine commune est le modèle de la glace de spin, qui se transforme en modèle de la glace sur réseau kagome (kagome ice) et réseau en damier (square ice) à deux dimensions, et la chaîne d’Ising à une dimension. Ces modèles ont été particulièrement étudiés dans le contexte de transitions de phases avec un ordre magnétique induit par les contraintes du système : en effet, selon la perturbation envisagée, les contraintes topologiques sous-jacentes peuvent provoquer une transition de Kasteleyn dans le kagome ice, ou une transition de type vitreuse dans la square ice, due à l’émergence d’un ordre ferromagnétique dans une chaîne d’Ising induit seulement par des effets de taille fini. Dans tous les cas, une étude détaillée par simulations numériques de type Monte Carlo ont été comparées à des résultats théoriques pour déterminer les propriétés de ces transitions. Les contraintes topologiques du kagome ice ont requis le développement d’un algorithme de vers permettant aux simulations de ne pas quitter l’ensemble des états fondamentaux. Une revue poussée de la thermodynamique et de la réponse de la diffraction de neutrons sur kagome ice sous un champ magnétique planaire arbitraire, nous ont amené à une compréhension plus profonde de la transition de Kasteleyn, et à un modèle numérique capable de prédire les figures de diffraction de neutrons de matériau de kagome ice dans n’importe quelles conditions expérimentales. Sous certaines conditions, ce modèle a révélé des propriétés thermodynamiques quantifiées et devrait fournir un terreau fertile pour de futurs travaux sur les conséquences des contraintes et transitions de phases topologiques. Une étude combinée du square ice et de la chaîne d’Ising a mise en lumière l’apparition d’un ordre sur réseau potentiellement découplé de l’ordre ferromagnétique sous-jacent, et particulièrement pertinent pour les réseaux magnétiques artificiels obtenus par lithographie. / In this thesis a series of model frustrated magnets have been investigated. Their common parent is the spin ice model, which is transformed into the kagome ice and square ice models in two-dimensions, and an Ising spin chain model in one-dimension. These models have been examined with particular interest in the spin ordering transitions induced by constraints on the system: a topological constraint leads, under appropriate conditions, to the Kasteleyn transition in kagome ice and a lattice freezing transition is observed in square ice which is due to a ferromagnetic ordering transition in an Ising chain induced solely by finite size effects. In all cases detailed Monte Carlo computational simulations have been carried out and compared with theoretical expressions to determine the characteristics of these transitions. In order to correctly simulate the kagome ice model a loop update algorithm has been developed which is compatible with the topological constraints in the system and permits the simulation to remain strictly on the groundstate manifold within the appropriate topological sector of the phase space. A thorough survey of the thermodynamic and neutron scattering response of the kagome ice model influenced by an arbitrary in-plane field has led to a deeper understanding of the Kasteleyn transition, and a computational model that can predict neutron scattering patterns for kagome ice materials under any experimental conditions. This model has also been shown to exhibit quantised thermodynamic properties under appropriate conditions and should provide a fertile testing ground for future work on the consequences of topological constraints and topological phase transitions. A combined investigation into the square ice and Ising chain models has revealed ordering behaviour within the lattice that may be decoupled from underlying ferro- magnetic ordering and is particularly relevant to magnetic nanoarrays.

Model and variable selection have attracted considerable attention in areas of application where datasets usually contain thousands of variables. Variable selection is a critical step to reduce the dimension of high dimensional data by eliminating irrelevant variables. The general objective of variable selection is not only to obtain a set of cost-effective predictors selected but also to improve prediction and prediction variance. We have made several contributions to this issue through a range of advanced topics: providing a graphical view of Bayesian Variable Selection (BVS), recovering sparsity in multivariate nonparametric models and proposing a testing procedure for evaluating nonlinear interaction effect in a semiparametric model. To address the first topic, we propose a new Bayesian variable selection approach via the graphical model and the Ising model, which we refer to the ``Bayesian Ising Graphical Model'' (BIGM). There are several advantages of our BIGM: it is easy to (1) employ the single-site updating and cluster updating algorithm, both of which are suitable for problems with small sample sizes and a larger number of variables, (2) extend this approach to nonparametric regression models, and (3) incorporate graphical prior information. In the second topic, we propose a Nonnegative Garrote on a Kernel machine (NGK) to recover sparsity of input variables in smoothing functions. We model the smoothing function by a least squares kernel machine and construct a nonnegative garrote on the kernel model as the function of the similarity matrix. An efficient coordinate descent/backfitting algorithm is developed. The third topic involves a specific genetic pathway dataset in which the pathways interact with the environmental variables. We propose a semiparametric method to model the pathway-environment interaction. We then employ a restricted likelihood ratio test and a score test to evaluate the main pathway effect and the pathway-environment interaction. / Ph. D.