Optimisation non Differentiable: Methodes de FaisceauxC.LemarechalauthorWe define a class of descent methods to minimize a nondifferentiable function. These methods are based on a representation of the objective which combines a quadratic approximation and the usual approximation by a piecewise linear function. Hence, they realize a synthesis between quasi-Newton methods and cutting plane methods. In addition, they have the particularity of requiring no sophisticated line search. They are also presented in ref. [7] (in English).2006Springer Berlin HeidelbergBook Section