On the Analysis of Backtrack Proceduresfor the Colouring of Random Graphs

Abstract

Backtrack search algorithms are procedures capable of deciding whether a decision problem has a solution or not through a sequence of trials and errors. Analysis of the performances of these procedures is a long-standing open problem in theoretical computer science. I present some statistical physics ideas and techniques to attack this problem. The approach is illustrated on the colouring of random graphs, and some current limitations and perspectives are presented.