Graph decompositions : treelength and pursuit-evasion games ; Décomposition de graphes : longueur arborescente et jeux de poursuite

Graph decompositions are a powerful tool aimed to divide a graph into several parts, called bags, connected in a tree-like or a path-like fashion, depending on whether we consider a tree-decomposition or a path-decomposition. They can be used to solve some NP-hard problems in linear time, as long as the maximum size of the bags (i.e. the width) is bounded. This has motivated the study of treewidth (minimum width out of all tree-decompositions of a graph) over the past 30 years. Nevertheless, there are still a lot of open questions, such as the computational complexity of treewidth in planar graphs. To answer this question, it could be interesting to study the length of these decompositions. This measure corresponds to the maximum diameter among the bags of a decomposition, and it was shown that there is a relationship between treelength and treewidth in planar graphs.In chapter 2, we study the treelength of several simple classes of planar graphs, such as series-parallel graphs. We give an infinite list of forbidden isometric subgraphs for series-parallel graphs of treelength 2. This list is then used to decide in polynomial time if a series-parallel graph has treelength at most 2 and, in case of a positive answer, to output a decomposition of length at most 2.In chapter 3, we focus on the pathlength, and we give a characterization for trees and cycles. We also study the pathlength of outerplanar graphs for which we design an approximation algorithm with additive ratio 1.Finally, in chapter 4, we study the Hunters and Rabbit game, an algorithmic variant of path-decomposition defined as a pursuit-evasion game. In this game, a group of hunters hunts an invisible rabbit which is forced to move to a neighboring position at every step. We are interested in the minimum number h(G) of hunters needed to catch the rabbit, independently of its moves along a graph G. This game has already been studied for some bipartite graphs (meshes, trees, hypercubes.). However, it remains open questions in a lot of graph classes, ...