Enumerating All Nash Equilibria for Two-person Extensive Games

Slides:



Advertisements
Similar presentations
C&O 355 Mathematical Programming Fall 2010 Lecture 22 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A A.
Advertisements

Totally Unimodular Matrices
Bilinear Games: Polynomial Time Algorithms for Rank Based Subclasses Ruta Mehta Indian Institute of Technology, Bombay Joint work with Jugal Garg and Albert.
Computation of Nash Equilibrium Jugal Garg Georgios Piliouras.
Introduction to Algorithms
6.896: Topics in Algorithmic Game Theory Lecture 11 Constantinos Daskalakis.
C&O 355 Lecture 4 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A.
How should we define corner points? Under any reasonable definition, point x should be considered a corner point x What is a corner point?
Equilibrium Concepts in Two Player Games Kevin Byrnes Department of Applied Mathematics & Statistics.
The Structure of Polyhedra Gabriel Indik March 2006 CAS 746 – Advanced Topics in Combinatorial Optimization.
Basic Feasible Solutions: Recap MS&E 211. WILL FOLLOW A CELEBRATED INTELLECTUAL TEACHING TRADITION.
A Randomized Polynomial-Time Simplex Algorithm for Linear Programming Daniel A. Spielman, Yale Joint work with Jonathan Kelner, M.I.T.
CS38 Introduction to Algorithms Lecture 15 May 20, CS38 Lecture 15.
Totally Unimodular Matrices Lecture 11: Feb 23 Simplex Algorithm Elliposid Algorithm.
1 Computing Nash Equilibrium Presenter: Yishay Mansour.
Duality Dual problem Duality Theorem Complementary Slackness
Approximation Algorithms
Finite Mathematics & Its Applications, 10/e by Goldstein/Schneider/SiegelCopyright © 2010 Pearson Education, Inc. 1 of 99 Chapter 4 The Simplex Method.
UNIT II: The Basic Theory Zero-sum Games Nonzero-sum Games Nash Equilibrium: Properties and Problems Bargaining Games Bargaining and Negotiation Review.
Exponentially many steps for finding a NE in a bimatrix game Rahul Savani, Bernhard von Stengel (2004) Presentation: Angelina Vidali.
Computing equilibria in extensive form games Andrew Gilpin Advanced AI – April 7, 2005.
Chapter 4 The Simplex Method
Daniel Kroening and Ofer Strichman Decision Procedures An Algorithmic Point of View Deciding ILPs with Branch & Bound ILP References: ‘Integer Programming’
Approximating Minimum Bounded Degree Spanning Tree (MBDST) Mohit Singh and Lap Chi Lau “Approximating Minimum Bounded DegreeApproximating Minimum Bounded.
Simplex method (algebraic interpretation)
6.853: Topics in Algorithmic Game Theory Fall 2011 Constantinos Daskalakis Lecture 11.
Approximation Algorithms Department of Mathematics and Computer Science Drexel University.
Optimization - Lecture 4, Part 1 M. Pawan Kumar Slides available online
Algorithms for solving two-player normal form games
Approximation Algorithms Department of Mathematics and Computer Science Drexel University.
Hon Wai Leong, NUS (CS6234, Spring 2009) Page 1 Copyright © 2009 by Leong Hon Wai CS6234: Lecture 4  Linear Programming  LP and Simplex Algorithm [PS82]-Ch2.
OR Chapter 7. The Revised Simplex Method  Recall Theorem 3.1, same basis  same dictionary Entire dictionary can be constructed as long as we.
Lecture.6. Table of Contents Lp –rounding Dual Fitting LP-Duality.
Linear Programming Chapter 9. Interior Point Methods  Three major variants  Affine scaling algorithm - easy concept, good performance  Potential.
Parameterized Two-Player Nash Equilibrium Danny Hermelin, Chien-Chung Huang, Stefan Kratsch, and Magnus Wahlstrom..
Linear Programming Chap 2. The Geometry of LP  In the text, polyhedron is defined as P = { x  R n : Ax  b }. So some of our earlier results should.
Approximation Algorithms based on linear programming.
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design Advanced Algorithms Analysis and Design By Dr. Nazir Ahmad Zafar.
Submodularity Reading Group Matroid Polytopes, Polymatroid M. Pawan Kumar
Theory of Computational Complexity Probability and Computing Chapter Hikaru Inada Iwama and Ito lab M1.
Comp/Math 553: Algorithmic Game Theory
Nash Equilibrium: P or NP?
EMGT 6412/MATH 6665 Mathematical Programming Spring 2016
Lap Chi Lau we will only use slides 4 to 19
The Duality Theorem Primal P: Maximize
Chap 10. Sensitivity Analysis
Market Equilibrium Ruta Mehta.
Topics in Algorithms Lap Chi Lau.
The minimum cost flow problem
Computing equilibria in extensive form games
Communication Complexity as a Lower Bound for Learning in Games
Proving that a Valid Inequality is Facet-defining
Duality for linear programming.
Linear Programming Prof. Sweta Shah.
Chap 9. General LP problems: Duality and Infeasibility
Polyhedron Here, we derive a representation of polyhedron and see the properties of the generators. We also see how to identify the generators. The results.
Chap 3. The simplex method
Analysis of Algorithms
Polyhedron Here, we derive a representation of polyhedron and see the properties of the generators. We also see how to identify the generators. The results.
Chapter 4. Duality Theory
Chapter 5. The Duality Theorem
Advanced consistency methods Chapter 8
Quantum Foundations Lecture 3
Back to Cone Motivation: From the proof of Affine Minkowski, we can see that if we know generators of a polyhedral cone, they can be used to describe.
C&O 355 Lecture 3 N. Harvey Review of Lecture 2:
Proving that a Valid Inequality is Facet-defining
Chapter 2. Simplex method
Normal Form (Matrix) Games
Chapter 2. Simplex method
A Technique for Reducing Normal Form Games to Compute a Nash Equilibrium Vincent Conitzer and Tuomas Sandholm Carnegie Mellon University, Computer Science.
Presentation transcript:

Enumerating All Nash Equilibria for Two-person Extensive Games Huang Wan London School of Economics W.Huang2@lse.ac.uk 2019/2/24

Games in Extensive Form Games in extensive form are represented by trees, with players moves corresponding to tree edges and information sets corresponding to tree nodes. A player I has perfect recall iff all nodes in the same information set of player I have the same own former paths. There are two strategies L and R for player 2 and four strategies ac,ad,bc,bd for player 1. 2019/2/24

Sequence Form Hi - set of information sets of player i h - information set Ch - set of moves at h Si - set of sequences of player i Any sequence is either the empty sequence or uniquely given by its last move at the information set of that player. Bernhard introduced the sequence form for a game tree. The sequence form is analogous to an strategic form. The counterpart of a strategy of an strategic form is a sequence in the sequence form. The good thing for the sequence form is that the number of sequences of a player is polynomial in the size of the game tree, while the number of strategies in the strategic form is exponential in the size of the game tree. 2019/2/24

Example: sequence form The first decision point is a square (rectangle) in the game tree, which means the first move is made by the nature. There are 2 information sets for player 1. Sequences a and b are uniquely determined by the last moves at the left information set; c and d are determined by the right one. There is only one information set for the second player. Sequences L and R are given by the moves L and R from that information set. 2019/2/24

Property of realization plans ( von Stengel, 1996) Suppose x and y are realization plans for player 1 and player 2, respectively. Then A realization plan sets a probability to each sequence. 2019/2/24

Example: sequence form Here x and y are the realization plans for player 1 and 2, respectively. According to the property, we have Ex=e and Fy=f, where E and F are the constraint matrices. Each column of E and F corresponds to a sequence and each column corresponds to a information set. 2019/2/24

Linear Programs By the definition of best response, (x, y) is an equilibrium iff it is in the solution of the pair of linear programs (LPs): The objective function of the left LP means that x should be a best response for the given y; and the right one means y should be a best response for the given x. The subjective equations and inequalities ensure x and y are realization plans for the players. 2019/2/24

Linear Programming Duality Primal Dual P 1 P 2 On the right column of this form is the dual of the pair of the LP. The first row for player 1 and second for player 2. 2019/2/24

Complementary Slackness Theorem (von Stengel, 1996) The pair (x, y) of realization plans defines an equilibrium iff there are vectors u, v, such that In the last line there are the complementarity conditions. 2019/2/24

Removing Redundancy Terminal sequence - A sequence is terminal iff there is no move extending to a longer sequence. Independent sequence - A sequence of player 1 is called independent (in a solution of Ex=e) iff its realization probability is a free variable in that solution. Similarly for sequences of player 2. -- The independence of a sequence relates to which free variables are chosen for the solution of Ex=e. The next step is supposed to be, drawing two polyhedra, each for one player, according to the equations and inequalities in the theorem. And find out all the points in the polyhedra that satisfies the complementarity conditions. These points represents all the equilibria of the game. But before doing that, we’d like to eliminate some of the redundant variables as well as some redundant constraints. In order to do that, we observe the sequence of the game and define the terminal sequence and the independent sequent. A terminal sequence lead to a leaf. But not all sequences that lead to a leaf are terminal. Some non-terminal sequences can still lead to leaves. 2019/2/24

Removing Redundancy In our algorithm we only choose the realization plans of terminal sequences as free variables (thus only terminal sequences can be independent). We can do this because: Lemma 1 For every non-terminal sequence , there are some terminal sequences , s.t. Proof. (by contradiction) Therefore all the non-negativity constraints of the non-terminal sequences are redundant if all the non-negativity constraints of the terminal sequences are satisfied. 2019/2/24

Removing Redundancy Three kinds of sequences: Terminal sequences whose realization probabilities xI are independent Terminal sequences whose realization probabilities xD are dependent, which can be expressed as p2+P2xI Non-terminal sequences whose realization probabilities xN are dependent, which can be expressed as p1+P1xI, and for which the non-negativity constraints are redundant 2019/2/24

Removing Redundancy Realization plans for both players: Player 1 where p_1, p_2, q_1, q_2 are certain vectors and P_1,P_2, Q_1, Q_2 are certain matrices. I_m-k and I_n-l are the identity matrices. 2019/2/24

Removing Redundancy Primal and Dual after removing redundancy : Primal Dual P 1 P 2 2019/2/24

Removing Redundancy 2019/2/24

Removing Redundancy LCP after removing redundancy: Theorem 1 The pair (x,y) of realization plans defines an equilibrium iff there are vectors u,v, such that 2019/2/24

Removing Redundancy 2019/2/24 This theorem is analogues to the last one by bernhard (von Stengel 1996) 2019/2/24

Finding All Equilibria Consider the polyhedra Finding all equilibria of the game can be converted to an inspection of all vertices of and . Now we can move to the step drawing the polyhedra. The polyhedra, each for one player, are defined by the constrains after the redundancy removal. 2019/2/24

Finding All Equilibria Label of points Since redundant constraints will not be considered, we can re-index only the terminal sequences. Suppose there are s terminal sequences for player 1 and t for player 2. Let Labels are used in Lemke-Howson algorithm. We use the labels here to find out the points that satisfy all the complementarity conditions. 2019/2/24

Finding All Equilibria The labels of (xI,v) in D1 and (yI, u) in D2 are given by: Each label corresponds to a binding constraint. 2019/2/24

Finding All Equilibria Lemma 2 For a two-person extensive form game (A,B), a realization plan (x,y) is an equilibrium iff (x,y) is a pair of points in D1 and D2 that has full labels 1,…,s+t. This lemma is equivalent to Theorem 1. The pair of points that has full labels satisfies the complementarity conditions. 2019/2/24

Finding All Equilibria Lemma 3 Suppose y* is a point in D2 which has l* labels. Then (x,y*) is an equilibrium iff x is in the set of convex combination of vertices or points in the extreme rays which have all the rest labels in D1. Similarly for points x* in D1. 2019/2/24

2019/2/24

Finding All Equilibria Lemma 4 Suppose (x*,y) represents an equilibrium where y is a point in D2 and x* is in the convex combination of vertices and points in extreme rays or in the convex combination of points in extreme rays in D1, then there is a point x’, which is in the convex combination of certain vertices, such that (x’,y) represents the same equilibrium as (x*,y). 2019/2/24

2019/2/24

There are the other two cases when x There are the other two cases when x* is in the convex combination of some vertices and some points in the extreme rays, and the case when x* is in the convex combination of some points in the extreme rays. 2019/2/24

The Algorithm Enumerating the vertices of a polyhedron by linear inequalities: one recent method is lrs (Avis and Fukuda (1992) ). Convex Combination of full label pairs of vertices. 2019/2/24

References. D.Avis (2002), lrs: A revised implementation of the reverse search vertex enumeration algorithm. V.Chvatal (1983), Linear Programming. Freeman, New York. H.W.Kuhn (1961), An algorithm for equilibrium points in bimatrix games. Proc. National Academy of Sciences of the U.S.A. 47, 1657-1662 C.E.Lemk and J.T.Howson, Jr. (1964), Equilibrium points of bimatrix games. Journal of the Society for Industrial and Applied Mathematics 12, 778-780 B.von Stengel (1996), Efficient computation of behaviour strategies. Games and Economic Behaviour 14, 220-246 B.von Stengel (2002), Computing equilibria for two-person games. Handbook of Game Theory, Volume 3, 1725-1742 G.M.Ziegler (1995), Lectures on Polytopes. Graduate Texts in Mathematics, Vol. 152, Springer, New York. 2019/2/24