Organizers: Frédéric Bonnans, Cristopher Hermosilla, Hasnaa Zidani. Place: ENSTA ParisTech or École Polytechnique How to get ENSTA, see http://www.ensta-paristech.fr/en/getting-ensta-paristech
December 13, 2013 (Salle 2234, UMA) :
· 15:00 Soutenance de Zhiping RAO (Commands)
Hamilton-Jacobi-Bellman approach for optimal control problems with discontinuous coefficients.
Abstract: This thesis deals with the Dynamical Programming and Hamilton-Jacobi-Bellman approach for a general class of deterministic optimal control problems with discontinuous coefficients.The tools essentially used in this work are based on the control theory, the viscosity theory forPartial Differential Equations, the nonsmooth analysis and the dynamical systems.
The first part of the thesis is concerned with the state constrained problem of discontinuous trajectories driven by impulsive dynamical systems. A characterization result of the value function of this problem has been obtained. Another contribution of this part consists of the extension of the HJB approach for the problems with time-measurable dynamical systems and in presence of time-dependent state constraints.
The second part is devoted to the problem on stratified domain, which consists of a union of subdomains separated by several interfaces. One of the motivations of this work comes from the hybrid control problems. Here new transmission conditions on the interfaces have been obtained to ensure the uniqueness and the characterization of the value function.
The third part investigates the homogenization of Hamilton-Jacobi equations in the framework of state-discontinuous Hamiltonians. This work considers the singular perturbation of optimal control problem on a periodic stratified structure. The limit problem has been analyzed and the associatedHamilton-Jacobi equation has been established. This equation describes the limit behavior of the value function of the perturbed problem when the scale of periodicity tends to 0.
Keywords: Optimal control problems, Hamilton-Jacobi-Bellman equations, viscosity solutions,nonsmooth analysis, impulsive differential equations, state constraints, multi-domains, stratified dynamical system, transmission conditions, singular perturbation.
December 2, 2013 (Salle de Séminaire, UMA) :
· 14:00 Laurent EL GHAOUI (Berkeley Center for New Media, US)
Understanding large text corpora via sparse optimization.
November 13, 2013 (Amphi Becquerel, École Polytechnique) :
· 14:30 Soutenance de Xavier DUPUIS (Commands)
Contrôle optimal d’équations différentielles avec – ou sans – mémoire.
November 5, 2013 (Amphi Becquerel, École Polytechnique) :
· 14:30 Soutenance de Laurent PFEIFFER (Commands)
Analyse de sensibilité pour des problèmes de commande optimale ; commande optimale stochastique sous contrainte en probabilité.
September 17, 2013 (Salle de Séminaire, UMA) :
· 10:30 Claudia Sagastizábal (IMPA, Brazil)
An approximation scheme for finding equilibria in energy markets with risk aversion.
Abstract: We consider two models for stochastic equilibrium: one based on the variational equilibrium of a generalized Nash game, and the other on the mixed complementarity formulation. The models differ in how the agents interpret their own actions on the market. An important characteristic of our approach is that the agents hedge risk in the objective functions (on costs or profits) of their optimization problems, which has a clear economic interpretation. This feature is obviously desirable, but in the risk-averse case it leads to variational inequalities with set-valued operators – a class of problems for which no established software is currently available. To overcome this difficulty, we approximate the problem by a sequence of differentiable variational inequalities based on smoothing the nonsmooth risk measure (such as average (conditional) value-at-risk) in the agents’ problems. The resulting problems can be tackled by the PATH solver, for example. Convergence of this approximation scheme is shown. Finally, numerical results on a part of the real-life European gas network are presented, including the use of Dantzig-Wolfe decomposition in conjunction with the smoothing approach. Joint work with J. P. Luna and M. Solodov
· 12:00 Mikhail Solodov (IMPA, Brazil)
Some news on the convergence of augmented Lagrangian methods
Abstract: We discuss some recent results on convergence and rate of convergence of the classical augmented Lagrangian algorithm. First, we prove local primal-dual convergence under the sole assumption that the dual starting point is close to a multiplier satisfying the second-order sufficient optimality condition. In particular, no constraint qualifications of any kind are needed. Previous literature on the subject required, in addition, the linear independence constraint qualification and either the strict complementarity assumption or a stronger version of the second-order sufficient condition. Second, we show that when applied to optimization problems with complementarity constraints (MPCC), the augmented Lagrangian approach has theoretical global convergence guarantees that compare favorably to the alternatives, whether standard nonlinear programming solvers or those specially designed for this problem class. Moreover, we show that in practice the ALGENCAN implementation of the augmented Lagrangian method is a very good choice for MPCC if robustness and the quality of computed solutiuon is of primary concern.
Jun 03, 2013 (Salle de Séminaire, UMA) :
- 14:00 Florence Hubert (Aix-Marseille Université)
Modelisation de la croissance metastatique d’un cancer et des traitements anti-cancereux.
May 24, 2013 (Salle de Séminaire, UMA) :
- 14:00 Franco Rampazzo (Università degli Studi di Padova, Italy)
(Sup + Bolza)-control problems as dynamic differential games.
The whole approach will be hopefully utilized to get necessary conditions for a curve to be a minimizer.
April 22, 2013 (Salle de Séminaire, UMA) :
- 14:00 Alexandre Chapoutot (LIE, ENSTA)
Calcul d’ensemble de trajectoires de systèmes hybrides par des méthodes numériques garanties.
Abstract: Cet exposé s’intéresse au problème d’atteignabilité dans les systèmes hybrides, c’est-à-dire aux systèmes composés d’éléments évoluant continûment dans le temps et d’éléments évoluant de manière discrète dans le temps, pour lesquels nous voulons connaître l’ensemble des trajectoires pour un ensemble de conditions initiales. Cette problématique est abordée suivant une vision informaticienne issue du domaine de la vérification formelle. Nous présentons une approche fondée sur les méthodes d’intégration numérique garantie pour résoudre des équations différentielles ordinaires et sur l’interpolation polynomiale garantie pour calculer les instants d’apparition d’événements de zero-crosssing. Après avoir présenté le contexte de ce travail, nous aborderons les principales caractéristiques des outils utilisés qui seront illustrés à l’aide d’exemples.
- 15:30 Srinivas Sridharan (UMA, ENSTA)
An overview of scalable control and estimation using max-plus methods.
Abstract: Conventional approaches to solving optimal control problems using dynamic programming require the solution of Hamilton Jacobi equations. Numerical solution techniques suffer from the curse of dimensionality and are thus intractable in higher dimensions. This talk will provide a gentle introduction to a promising class of control synthesis methods which are more scalable than grid based methods. Further the application of these ‘max-plus’ methods to an intriguing range of problems in estimation and control of linear, non-linear and multi-sensor problems will be described.
March 11, 2013 (Salle de Séminaire, UMA) :
- 14:00 Eladio Ocaña (Instituto de Matemática y Ciencias Afines, Lima)
Explicit solution for singular calculus of variations in infinite horizon.
Abstract: We consider a one dimensional infinite horizon calculus of variations problem (P), where the integrand is linear with respect to the velocity. The Euler-Lagrange equation, when defined, is not a differential equation as usual, but reduces to an algebraic (or transcendental) equation $C(x)=0 $. Thus this first order optimality condition is not informative for optimal solutions with initial condition $x_0 $ such that $C(x_0) \neq 0 $. To problem (P) we associate an auxiliary calculus of variations problem whose solutions connect as quickly as possible the initial conditions to some constant solutions. Then we deduce the optimality of these curves, called MRAPs (Most Rapid Approach Path), for (P). According to the optimality criterium we consider, we have to assume a classical transversality condition. We observe that (P) possesses the Turnpike property, the Turnpike set being given by the preceding particular constant solutions of the auxiliary problem.
- 15:30 Adriano Festa (Imperial College London)
Pursuit-Evasion Games with Multi-Pursuer: a decomposition approach.
Abstract: In a pursuit-evasion game, each pursuer attempts to minimize the distance between himself (P) and the evader (E) and capture it in the shortest time, whereas the evader tries to maximize the same distance to escape from being captured. In this talk, we deal with PE games with an evader and n pursuers. It is easy to imagine that the dimension of the space where the agents are moving increases with the number of pursuers, making the problem hard to solve. We will use the well-known PDE approach and an original decomposition technique to get over the big dimensionality contained in the problem. Simulation results will show the effectiveness of the proposed strategy of resolution as well as the limitations of it.
February 19, 2013 (Salle de Séminaire, UMA) :
- 14:00 Bean-San Goh (Curtin University, Sarawak Malaysia)
Optimal control of single species biological population.
- 15:30 Frédéric Bonnans (Commands & CMAP)
Stabilité des arcs singuliers pour un problème de commande optimale perturbé.
January 7, 2013 (Salle de Séminaire, UMA) :
- 14:00 Kristian Debrabant (University of Southern Denmark)
Convergence of one-step methods for SDEs.
- 15:30 Fabien Crauste (Université Claude Bernard Lyon 1)