ECTS Credits: 4
Duration: 36 hours
Person(s) in charge:Yannick PRIVAT, Chargé de Recherche UPMC, email@example.com
Prerequisites: Linear algebra, differential calculus. Knowledge in linear numerical analysis would be a plus.
Goal: provide with a basic knowledge allowing, given a practical problem, to choose the adapted algoritms
Program and contents:
Analysis (well-posedness of the problem). First and second order optimality conditions, use of convexity. General classes of algorithms: steepest descent, conjugate gradient, quasi-Newton methods. Line search algorithms for choosing an adequate step-length. Global convergence and asymptotic convergence of the algorithms.
Analysis (well-posedness of the problem), optimality conditions. Equality or inequality constraints. Feasible directions. Lagrange and Kuhn and Tucker theorems. Convex problems. Gradient projection and penalty methods. Lagrange-Newton method. The Lagrangian, saddle points and duality. Uzawa method and extensions.
Description and operational vocabulary
Gestion des contenus