Home
 
Attendees
Conference program
Registration
Location
Hotel information
Links
 
 
Previous editions
2002


    

Session MA10 - Problèmes à deux niveaux I / Bilevel Problems I

Day Monday, May 05, 2003
Room Cogeco
President Gilles Savard

Presentations

10:30 Enumeration of All Extreme Equilibria in Game Theory: Bimatrix and Polymatrix Games
  Slim Belhaiza, École Polytechnique de Montréal, GERAD et Mathématiques et génie industriel, C.P. 6079, Succ. Centre-ville, Montréal, Québec, Canada, H3C 3A7
Pierre Hansen, HEC Montréal, GERAD et Méthodes quantitatives de gestion, 3000, ch. de la Côte-Sainte-Catherine, Montréal, Québec, Canada, H3T 2A7
Charles Audet, École Polytechnique de Montréal, GERAD et Mathématiques et génie industriel, C.P. 6079, Succ. Centre-ville, Montréal, Québec, Canada, H3C 3A7
Bernhard Von Stengel, London School of Economics, London, U.K.

This work is a study of a particular form of games in Game Theory: Bimatrix and Polymatrix Games. Compared to extensive or dynamic games, decision making is simultaneous, static and unique in bimatrix and polymatrix games. Complete Enumeration of Extreme Equilibria helps identifying all the equilibria sets that could be obtained for such a class of games. The purpose of this paper is to present a new mixed 0-1 linear formulation of bimatrix and polymatrix games and a new algorithm that enumerates all there extreme equilibria. The paper is organized as follows. Bimatrix Games and Polymatrix Games are introduced in Sections 2 and 3. New formulations of bimatrix and polymatrix games are proposed in Section 4. Elimination of dominated strategies is clarified in Section 5. In Section 6, E3-MIP algorithm for enumeration of extreme equilibria is presented and illustrated on some examples. Results of E3-MIP and EEE on randomly generated bimatrix games are compared and discussed, then E3-MIP results on randomly generated polymatrix games are presented in Section 7.


10:55 Heuristics for Toll Setting Problem
  Sophie Dewez, Université Libre de Bruxelles, Institut de statistique et de recherche opérationnelle, CP 210/01, boulevard du Triomphe, 1050 Bruxelles, Belgique
Martine Labbé, Université Libre de Bruxelles, Institut de statistique et de recherche opérationnelle, CP 210/01, boulevard du Triomphe, 1050 Bruxelles, Belgique
Patrice Marcotte, Université de Montréal, C.R.T. et Informatique et recherche opérationnelle, C.P. 6128, Succ. Centre-ville, Montréal, Québec, Canada, H3C 3J7
Gilles Savard, École Polytechnique de Montréal, GERAD et Mathématiques et génie industriel, C.P. 6079, Succ. Centre-ville, Montréal, Québec, Canada, H3C 3A7

We consider the problem of determining a set of optimal prices on a subset of all thearcs of a highway network in order to maximize the revenue of the government or private firms. We can formulate this problem as a bilinear bilevel problem. We develop heuristics to solve the problem and present computational results.


11:20 Dantzig-Wolfe and Benders Decomposition of Equilibrium Models
  David J. Fuller, University of Waterloo, Management Sciences, University of Waterloo, Waterloo, Ontario, Canada, N2L 3G1

The talk will explain natural extensions of Dantzig-Wolfe and Benders decomposition methods from the realm of optimization models to equilibrium models, specifically to equilibrium models that are represented as variational inequality problems. Conditions for convergence of the algorithms are met by many economic equilibrium models.


11:45 A Cuts Algorithm for the Linear Bilevel Programming Problem
  Walid Zghal, École Polytechnique Montréal, GERAD et Mathématiques et génie industriel, C.P. 6079, Succ. Centre-ville, Montréal, Québec, Canada, H3C 3A7
Charles Audet, École Polytechnique de Montréal, GERAD et Mathématiques et génie industriel, C.P. 6079, Succ. Centre-ville, Montréal, Québec, Canada, H3C 3A7
Gilles Savard, École Polytechnique de Montréal, GERAD et Mathématiques et génie industriel, C.P. 6079, Succ. Centre-ville, Montréal, Québec, Canada, H3C 3A7

We propose an algorithm which uses Gomory cuts to solve the Bilevel Linear Problem BLP. First, we present the mathematical model of BLP and its properties. We emphasize the links between Bilevel Linear Programming and Linear Mixed 0-1 Programming. Next, we propose a set of valid cuts to the BLP. These cuts are generated from the mixed reformulation for the BLP. We examine their depths. Then a new algorithm Al(coupes) for Bilevel Linear Programming is introduced. Our algorithm includes two steps. The first one generates Gomory cuts to reduce the feasible region. The second step is a Branch-and-Bound procedure. We study the different features of the algorithm, in particular the cut selection criterion and the branching criterion. We also propose a set of tests and procedures that streamline the algorithm. Finally, we feature the optimal algorithmic criteria and evaluate the effectiveness of the cuts through experimental results. Our algorithm Al(coupes) outperforms the CPLEX resolution of mixed reformulation when tested on a series of randomly generated test problems.