Sur une droite, le transport optimal ne pose pas de difficultés. Récemment, ce constat a été utilisé pour traiter des problèmes plus généraux. En effet, on a remarqué qu'une habile désintégration permet souvent de se ramener à la dimension un, ce qui permet d'utiliser les méthodes afférentes pour obtenir un premier résultat, que l'on fait ensuite évoluer pour gagner en précision.Je montre ici l'efficacité de cette approche, en revenant sur deux problèmes déjà résolus partiellement de cette manière, et en complétant la réponse qui en avait été donnée.Le premier problème concerne le calcul de l'application de Yann Brenier. En effet, Guillaume Carlier, Alfred Galichon et Filippo Santambrogio ont prouvé que celle-ci peut être obtenue grâce à une équation différentielle, pour laquelle une condition initiale est donnée par le réarrangement de Knothe--Rosenblatt (lui-même défini via une succession de transformations unidimensionnelles). Ils n'ont cependant traité que des mesures finales discrètes ; j'étends leur résultat aux cas continus. L'équation de Monge--Ampère, une fois dérivée, donne une EDP pour le potentiel de Kantorovitch; mais pour obtenir une condition initiale, il faut utiliser le théorème des fonctions implicites de Nash--Moser.Le chapitre 1 rappelle quelques résultats essentiels de la théorie du transport optimal, et le chapitre 2 est consacré au théorème de Nash--Moser. J'expose ensuite mes propres résultats dans le chapitre 3, et leur implémentation numérique dans le chapitre 4.Enfin, le dernier chapitre est consacré à l'algorithme IDT, développé par François Pitié, Anil C. Kokaram et Rozenn Dahyot. Celui-ci construit une application de transport suffisamment proche de celle de M. Brenier pour convenir à la plupart des applications. Une interprétation en est proposée en termes de flot de gradients dans l'espace des probabilités, avec pour fonctionnelle la distance de Wasserstein projetée. Je démontre aussi l'équivalence de celle-ci avec la distance usuelle de Wasserstein. / In dimension one, optimal transportation is rather straightforward. The easiness with which a solution can be obtained in that setting has recently been used to tackle more general situations, each time thanks to the same method. First, disintegrate your problem to go back to the unidimensional case, and apply the available 1D methods to get a first result; then, improve it gradually using some evolution process.This dissertation explores that direction more thoroughly. Looking back at two problems only partially solved this way, I show how this viewpoint in fact allows to go even further.The first of these two problems concerns the computation of Yann Brenier's optimal map. Guillaume Carlier, Alfred Galichon, and Filippo Santambrogio found a new way to obtain it, thanks to an differential equation for which an initial condition is given by the Knothe--Rosenblatt rearrangement. (The latter is precisely defined by a series of unidimensional transformations.) However, they only dealt with discrete target measures; I~generalize their approach to a continuous setting. By differentiation, the Monge--Ampère equation readily gives a PDE satisfied by the Kantorovich potential; but to get a proper initial condition, it is necessary to use the Nash--Moser version of the implicit function theorem.The basics of optimal transport are recalled in the first chapter, and the Nash--Moser theory is exposed in chapter 2. My results are presented in chapter 3, and numerical experiments in chapter 4.The last chapter deals with the IDT algorithm, devised by François Pitié, Anil C. Kokaram, and Rozenn Dahyot. It builds a transport map that seems close enough to the optimal map for most applications. A complete mathematical understanding of the procedure is, however, still lacking. An interpretation as a gradient flow in the space of probability measures is proposed, with the sliced Wasserstein distance as the functional. I also prove the equivalence between the sliced and usual Wasserstein distances.

In dimension one, optimal transportation is rather straightforward. The easiness with which a solution can be obtained in that setting has recently been used to tackle more general situations, each time thanks to the same method. First, disintegrate your problem to go back to the unidimensional case, and apply the available 1D methods to get a first result; then, improve it gradually using some evolution process.This dissertation explores that direction more thoroughly. Looking back at two problems only partially solved this way, I show how this viewpoint in fact allows to go even further.The first of these two problems concerns the computation of Yann Brenier's optimal map. Guillaume Carlier, Alfred Galichon, and Filippo Santambrogio found a new way to obtain it, thanks to an differential equation for which an initial condition is given by the Knothe--Rosenblatt rearrangement. (The latter is precisely defined by a series of unidimensional transformations.) However, they only dealt with discrete target measures; I~generalize their approach to a continuous setting. By differentiation, the Monge--Ampère equation readily gives a PDE satisfied by the Kantorovich potential; but to get a proper initial condition, it is necessary to use the Nash--Moser version of the implicit function theorem.The basics of optimal transport are recalled in the first chapter, and the Nash--Moser theory is exposed in chapter 2. My results are presented in chapter 3, and numerical experiments in chapter 4.The last chapter deals with the IDT algorithm, devised by François Pitié, Anil C. Kokaram, and Rozenn Dahyot. It builds a transport map that seems close enough to the optimal map for most applications. A complete mathematical understanding of the procedure is, however, still lacking. An interpretation as a gradient flow in the space of probability measures is proposed, with the sliced Wasserstein distance as the functional. I also prove the equivalence between the sliced and usual Wasserstein distances.