veri cation and game theory - lsvbouyer/files/tuto-atva19.pdf · a broader sense: what is game...
TRANSCRIPT
![Page 1: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/1.jpg)
Verification and Game TheoryTutorial on Basic Game Theory
Patricia Bouyer
LSV, CNRS & ENS Paris-SaclayUniversite Paris-Saclay, Cachan, France
Thanks to:
my co-authors Nicolas Markey, Romain Brenguier,Michael Ummels, Nathan Thomasset
Stephane Le Roux for recent discussions on the subjectThomas Brihaye for some of the slides
![Page 2: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/2.jpg)
The tutorial in perspective
General objective of the research topic
Import game theory solutions to the verification field
Lift reasoning based on two-player zero-sum games to multiplayergames
two-player zero-sum games multiplayer non-zero-sum games
winning objective payoff function
winning strategy equilibria (various kinds)
von Neumann Theorem Nash Theorem
... ...
Focus of the tutorialGive basics of game theory
Discuss aspects that will be helpful for analyzing models useful forverification
2/62
![Page 3: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/3.jpg)
The tutorial in perspective
General objective of the research topic
Import game theory solutions to the verification field
Lift reasoning based on two-player zero-sum games to multiplayergames
two-player zero-sum games multiplayer non-zero-sum games
winning objective payoff function
winning strategy equilibria (various kinds)
von Neumann Theorem Nash Theorem
... ...
Focus of the tutorialGive basics of game theory
Discuss aspects that will be helpful for analyzing models useful forverification
2/62
![Page 4: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/4.jpg)
The tutorial in perspective
General objective of the research topic
Import game theory solutions to the verification field
Lift reasoning based on two-player zero-sum games to multiplayergames
two-player zero-sum games multiplayer non-zero-sum games
winning objective payoff function
winning strategy equilibria (various kinds)
von Neumann Theorem Nash Theorem
... ...
Focus of the tutorialGive basics of game theory
Discuss aspects that will be helpful for analyzing models useful forverification
2/62
![Page 5: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/5.jpg)
Outline
1 What is a game?Games we play for funA broader sense to the notion of game
2 Strategic games – Playing only once simultaneously(Strict) Domination and IterationStability: Nash equilibria
3 Extensive games – Playing several times sequentially
4 Repeated games – Playing the same game again and again
5 Conclusion
3/62
![Page 6: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/6.jpg)
Outline
1 What is a game?Games we play for funA broader sense to the notion of game
2 Strategic games – Playing only once simultaneously(Strict) Domination and IterationStability: Nash equilibria
3 Extensive games – Playing several times sequentially
4 Repeated games – Playing the same game again and again
5 Conclusion
4/62
![Page 7: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/7.jpg)
Games we play for fun
5/62
![Page 8: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/8.jpg)
These games can be classified
Number of players: 1 or 2 or 3 or . . .
1 ; Pacman, Candy Crush, Freecel...2 ; Chess, Tennis, Stratego, Four in a row, ...3 (or more) ; Poker, Monopoly,...
Type of interactions: simultaneous or sequential
Maximal length of a play: finite ou infinite
Type of information: perfect or imperfect
Presence of randomness: deterministic or probabilistic
Type of payoff: boolean or quantitative
6/62
![Page 9: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/9.jpg)
These games can be classified
Number of players: 1 or 2 or 3 or . . .
Type of interactions: simultaneous or sequential
simultaneous ; Rock-Paper-Scissor, Penalty,...sequential ; Chess, Stratego, ...
Maximal length of a play: finite ou infinite
Type of information: perfect or imperfect
Presence of randomness: deterministic or probabilistic
Type of payoff: boolean or quantitative
6/62
![Page 10: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/10.jpg)
These games can be classified
Number of players: 1 or 2 or 3 or . . .
Type of interactions: simultaneous or sequential
Maximal length of a play: finite ou infinite
finite ; Four in a row, Battleship,...infinite ; Tennis, Monopoly,...
Type of information: perfect or imperfect
Presence of randomness: deterministic or probabilistic
Type of payoff: boolean or quantitative
6/62
![Page 11: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/11.jpg)
These games can be classified
Number of players: 1 or 2 or 3 or . . .
Type of interactions: simultaneous or sequential
Maximal length of a play: finite ou infinite
Type of information: perfect or imperfect
perfect ; Four in a row, Chess,...imperfect ; Battleship, Poker, Stratego...
Presence of randomness: deterministic or probabilistic
Type of payoff: boolean or quantitative
6/62
![Page 12: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/12.jpg)
These games can be classified
Number of players: 1 or 2 or 3 or . . .
Type of interactions: simultaneous or sequential
Maximal length of a play: finite ou infinite
Type of information: perfect or imperfect
Presence of randomness: deterministic or probabilistic
deterministic ; Four in a row, Chess, Battleship,...probabilistic ; Monopoly, Poker,...
Type of payoff: boolean or quantitative
6/62
![Page 13: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/13.jpg)
These games can be classified
Number of players: 1 or 2 or 3 or . . .
Type of interactions: simultaneous or sequential
Maximal length of a play: finite ou infinite
Type of information: perfect or imperfect
Presence of randomness: deterministic or probabilistic
Type of payoff: boolean or quantitative
boolean ; Four in a row, Chess,...quantitative ; Poker,...
6/62
![Page 14: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/14.jpg)
These games can be classified
Number of players: 1 or 2 or 3 or . . .
Type of interactions: simultaneous or sequential
Maximal length of a play: finite ou infinite
Type of information: perfect or imperfect
Presence of randomness: deterministic or probabilistic
Type of payoff: boolean or quantitative
6/62
![Page 15: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/15.jpg)
Outline
1 What is a game?Games we play for funA broader sense to the notion of game
2 Strategic games – Playing only once simultaneously(Strict) Domination and IterationStability: Nash equilibria
3 Extensive games – Playing several times sequentially
4 Repeated games – Playing the same game again and again
5 Conclusion
7/62
![Page 16: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/16.jpg)
A broader sense: What is game theory?
[MSZ13] Maschler, Solan, Zamir. Game theory (Cambridge University Press)
Goal: Model and analyze (using mathematical tools)situations of interactive decision making
Ingredients
Several decision makers (called players)
All with different goals
The decision of each players impacts the outcome for all
Interactivity!
Wide range of applicability
“[...] it is a context-free mathematical toolbox”
Social science: e.g. social choice theory
Theoretical economics: e.g. models of markets, auctions
Political science: e.g. fair division
Biology: e.g. evolutionary biology
...
8/62
![Page 17: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/17.jpg)
A broader sense: What is game theory?
[MSZ13] Maschler, Solan, Zamir. Game theory (Cambridge University Press)
Goal: Model and analyze (using mathematical tools)situations of interactive decision making
Ingredients
Several decision makers (called players)
All with different goals
The decision of each players impacts the outcome for all
Interactivity!
Wide range of applicability
“[...] it is a context-free mathematical toolbox”
Social science: e.g. social choice theory
Theoretical economics: e.g. models of markets, auctions
Political science: e.g. fair division
Biology: e.g. evolutionary biology
...
8/62
![Page 18: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/18.jpg)
A broader sense: What is game theory?
[MSZ13] Maschler, Solan, Zamir. Game theory (Cambridge University Press)
Goal: Model and analyze (using mathematical tools)situations of interactive decision making
Ingredients
Several decision makers (called players)
All with different goals
The decision of each players impacts the outcome for all
Interactivity!
Wide range of applicability
“[...] it is a context-free mathematical toolbox”
Social science: e.g. social choice theory
Theoretical economics: e.g. models of markets, auctions
Political science: e.g. fair division
Biology: e.g. evolutionary biology
...
8/62
![Page 19: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/19.jpg)
A broader sense: What is game theory?
[MSZ13] Maschler, Solan, Zamir. Game theory (Cambridge University Press)
Goal: Model and analyze (using mathematical tools)situations of interactive decision making
Ingredients
Several decision makers (called players)
All with different goals
The decision of each players impacts the outcome for all
Interactivity!
Wide range of applicability
“[...] it is a context-free mathematical toolbox”
Social science: e.g. social choice theory
Theoretical economics: e.g. models of markets, auctions
Political science: e.g. fair division
Biology: e.g. evolutionary biology
...
8/62
![Page 20: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/20.jpg)
A broader sense: What is game theory?
[MSZ13] Maschler, Solan, Zamir. Game theory (Cambridge University Press)
Goal: Model and analyze (using mathematical tools)situations of interactive decision making
Ingredients
Several decision makers (called players)
All with different goals
The decision of each players impacts the outcome for all
Interactivity!
Wide range of applicability
“[...] it is a context-free mathematical toolbox”
Social science: e.g. social choice theory
Theoretical economics: e.g. models of markets, auctions
Political science: e.g. fair division
Biology: e.g. evolutionary biology
...
8/62
![Page 21: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/21.jpg)
A broader sense: What is game theory?
[MSZ13] Maschler, Solan, Zamir. Game theory (Cambridge University Press)
Goal: Model and analyze (using mathematical tools)situations of interactive decision making
Ingredients
Several decision makers (called players)
All with different goals
The decision of each players impacts the outcome for all
Interactivity!
Wide range of applicability
“[...] it is a context-free mathematical toolbox”
Social science: e.g. social choice theory
Theoretical economics: e.g. models of markets, auctions
Political science: e.g. fair division
Biology: e.g. evolutionary biology
...
8/62
![Page 22: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/22.jpg)
A broader sense: What is game theory?
[MSZ13] Maschler, Solan, Zamir. Game theory (Cambridge University Press)
Goal: Model and analyze (using mathematical tools)situations of interactive decision making
Ingredients
Several decision makers (called players)
All with different goals
The decision of each players impacts the outcome for all
Interactivity!
Wide range of applicability
“[...] it is a context-free mathematical toolbox”
Social science: e.g. social choice theory
Theoretical economics: e.g. models of markets, auctions
Political science: e.g. fair division
Biology: e.g. evolutionary biology
...
8/62
![Page 23: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/23.jpg)
The prisoner dilemma
Two suspects are arrested by the police. The police, havingseparated both prisoners, visit each of them to offer the same deal.
If one testifies (Defects) for the prosecution against the other andthe other remains silent (Cooperates), the betrayer goes free andthe silent accomplice receives the full 10-year sentence.
If both remain silent, both are sentenced to only 3 years in jail.
If each betrays the other, each receives a 5-year sentence.
How should the prisoners act?
Modelled as a matrix game
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
9/62
![Page 24: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/24.jpg)
The prisoner dilemma
Two suspects are arrested by the police. The police, havingseparated both prisoners, visit each of them to offer the same deal.
If one testifies (Defects) for the prosecution against the other andthe other remains silent (Cooperates), the betrayer goes free andthe silent accomplice receives the full 10-year sentence.
If both remain silent, both are sentenced to only 3 years in jail.
If each betrays the other, each receives a 5-year sentence.
How should the prisoners act?
Modelled as a matrix game
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
9/62
![Page 25: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/25.jpg)
The prisoner dilemma
Two suspects are arrested by the police. The police, havingseparated both prisoners, visit each of them to offer the same deal.
If one testifies (Defects) for the prosecution against the other andthe other remains silent (Cooperates), the betrayer goes free andthe silent accomplice receives the full 10-year sentence.
If both remain silent, both are sentenced to only 3 years in jail.
If each betrays the other, each receives a 5-year sentence.
How should the prisoners act?
Modelled as a matrix game
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
9/62
![Page 26: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/26.jpg)
Cournot competition
Two companies produce the same good, they compete on the amount ofoutput they produce, which they decide on independently of each otherand at the same time. The selling price is a commonly known decreasingfunction of the total amount produced.
Let ai denote the quantity produced by the i-th company.
ProfitA1 (a1, a2) = a1
(α− β(a1 + a2)︸ ︷︷ ︸
selling price
)− γ a1︸︷︷︸
production cost
What should be the amount of the output to optimise the profit?
10/62
![Page 27: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/27.jpg)
Cournot competition
Two companies produce the same good, they compete on the amount ofoutput they produce, which they decide on independently of each otherand at the same time. The selling price is a commonly known decreasingfunction of the total amount produced.
Let ai denote the quantity produced by the i-th company.
ProfitA1 (a1, a2) = a1
(α− β(a1 + a2)︸ ︷︷ ︸
selling price
)− γ a1︸︷︷︸
production cost
What should be the amount of the output to optimise the profit?
10/62
![Page 28: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/28.jpg)
Cournot competition
Two companies produce the same good, they compete on the amount ofoutput they produce, which they decide on independently of each otherand at the same time. The selling price is a commonly known decreasingfunction of the total amount produced.
Let ai denote the quantity produced by the i-th company.
ProfitA1 (a1, a2) = a1
(α− β(a1 + a2)︸ ︷︷ ︸
selling price
)− γ a1︸︷︷︸
production cost
What should be the amount of the output to optimise the profit?
10/62
![Page 29: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/29.jpg)
Selling ice-cream on the beach...
Consider a beach that can be represented by a unit interval. Sun-tannedpeople are located uniformly on the beach. Everyone at the beachdreams of an ice-cream.
Two ice-cream sellers will settle on the beach.
Where should they build their stand in order to optimise their benefits ?
11/62
![Page 30: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/30.jpg)
Selling ice-cream on the beach...
Consider a beach that can be represented by a unit interval. Sun-tannedpeople are located uniformly on the beach. Everyone at the beachdreams of an ice-cream.
Two ice-cream sellers will settle on the beach.
Where should they build their stand in order to optimise their benefits ?
11/62
![Page 31: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/31.jpg)
The Nim game
The rules (simplified version)
Two players, turn-based games
Initially, there are 8 matches
On each turn, a player must remove 1 or 2 matches
The player removing the last match wins the game
Modelled as a game played on a graph
8 7 6 5 4 3 2 1 ,7 6 5 4 3 2 1 ,
12/62
![Page 32: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/32.jpg)
The Nim game
The rules (simplified version)
Two players, turn-based games
Initially, there are 8 matches
On each turn, a player must remove 1 or 2 matches
The player removing the last match wins the game
Modelled as a game played on a graph
8 7 6 5 4 3 2 1 ,7 6 5 4 3 2 1 ,
12/62
![Page 33: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/33.jpg)
Various models of games
Many models of games
Strategic games
Repeated games
Games played on graphs
Games played using equations
...
Many features
imperfect information
presence of randomness
continuous time
...
13/62
![Page 34: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/34.jpg)
Let us suppose that:
we have fixed a game,
we have identified an adequate model for this game.
The next natural question is:
What is a solution for this game?
14/62
![Page 35: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/35.jpg)
Outline
1 What is a game?Games we play for funA broader sense to the notion of game
2 Strategic games – Playing only once simultaneously(Strict) Domination and IterationStability: Nash equilibria
3 Extensive games – Playing several times sequentially
4 Repeated games – Playing the same game again and again
5 Conclusion
15/62
![Page 36: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/36.jpg)
Strategic games (aka matrix games, or one-shot games)
Strategic game
A strategic game G is a triple(
Agt,Σ, (gA)A∈Agt
)where:
Agt is the finite and non empty set of players,
Σ is a non empty set of actions,
gA : ΣAgt → R is the payoff function of player A ∈ Agt.
16/62
![Page 37: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/37.jpg)
Strategic games (aka matrix games, or one-shot games)
Strategic game
A strategic game G is a triple(
Agt,Σ, (gA)A∈Agt
)where:
Agt is the finite and non empty set of players,
Σ is a non empty set of actions,
gA : ΣAgt → R is the payoff function of player A ∈ Agt.
Example: Prisoner dilemma
Agt = {A1,A2},Σ = {C, D}
(gA1 , gA2 ) is given byC D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
16/62
![Page 38: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/38.jpg)
Hypotheses made in classical game theory
Hypotheses
The players are intelligent (i.e. they reason perfectly and quickly)
The players are rational (i.e. they want to maximise their payoff)
The players are selfish (i.e. they only care for their own payoff)
17/62
![Page 39: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/39.jpg)
Outline
1 What is a game?Games we play for funA broader sense to the notion of game
2 Strategic games – Playing only once simultaneously(Strict) Domination and IterationStability: Nash equilibria
3 Extensive games – Playing several times sequentially
4 Repeated games – Playing the same game again and again
5 Conclusion
18/62
![Page 40: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/40.jpg)
Optimality
Dominating profile
A profile b ∈ ΣAgt is dominating if
∀c ∈ ΣAgt ∀A ∈ Agt gA(c)≤ gA(b)
L R
T (0, 0) (2, 1)B (3, 2) (1, 2)
(B, L) is optimal!
19/62
![Page 41: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/41.jpg)
Optimality
Dominating profile
A profile b ∈ ΣAgt is dominating if
∀c ∈ ΣAgt ∀A ∈ Agt gA(c)≤ gA(b)
L R
T (0, 0) (2, 1)B (3, 2) (1, 2)
(B, L) is optimal!
19/62
![Page 42: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/42.jpg)
Strict domination
Stricly dominated action (or strategy)
An action bA ∈ Σ is strictly dominated by cA ∈ Σ for player A ∈ Agt if
∀a−A ∈ ΣAgt\{A} gA(bA, a−A)< gA(cA, a−A)
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
C is strictly dominated by D for player A1;
C is strictly dominated by D for player A2.
The only rational issue of the game is (D, D)whose payoff is (−5,−5).
(Even though this is sub-optimal)
20/62
![Page 43: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/43.jpg)
Strict domination
Stricly dominated action (or strategy)
An action bA ∈ Σ is strictly dominated by cA ∈ Σ for player A ∈ Agt if
∀a−A ∈ ΣAgt\{A} gA(bA, a−A)< gA(cA, a−A)
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
C is strictly dominated by D for player A1;
C is strictly dominated by D for player A2.
The only rational issue of the game is (D, D)whose payoff is (−5,−5).
(Even though this is sub-optimal)
20/62
![Page 44: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/44.jpg)
Strict domination
Stricly dominated action (or strategy)
An action bA ∈ Σ is strictly dominated by cA ∈ Σ for player A ∈ Agt if
∀a−A ∈ ΣAgt\{A} gA(bA, a−A)< gA(cA, a−A)
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
C is strictly dominated by D for player A1;
C is strictly dominated by D for player A2.
The only rational issue of the game is (D, D)whose payoff is (−5,−5).
(Even though this is sub-optimal)
20/62
![Page 45: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/45.jpg)
Strict domination
Stricly dominated action (or strategy)
An action bA ∈ Σ is strictly dominated by cA ∈ Σ for player A ∈ Agt if
∀a−A ∈ ΣAgt\{A} gA(bA, a−A)< gA(cA, a−A)
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
C is strictly dominated by D for player A1;
C is strictly dominated by D for player A2.
The only rational issue of the game is (D, D)whose payoff is (−5,−5).
(Even though this is sub-optimal)
20/62
![Page 46: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/46.jpg)
Strict domination
Stricly dominated action (or strategy)
An action bA ∈ Σ is strictly dominated by cA ∈ Σ for player A ∈ Agt if
∀a−A ∈ ΣAgt\{A} gA(bA, a−A)< gA(cA, a−A)
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
C is strictly dominated by D for player A1;
C is strictly dominated by D for player A2.
The only rational issue of the game is (D, D)whose payoff is (−5,−5).
(Even though this is sub-optimal)
20/62
![Page 47: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/47.jpg)
Domination - A finite variant of Cournot competition
L M H
L (4, 4) (2, 5) (1, 3)M (5, 2) (3, 3) (2, 1)H (3, 1) (1, 2) (0, 0)
The action H can be eliminated for both players.
As both players are rational and assume that their opponent is rational,we Iterate the Elimination of Strictly Dominated Strategies (IESDS).
The only rational issue of the game is (M, M)whose payoff is (3, 3).
21/62
![Page 48: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/48.jpg)
Domination - A finite variant of Cournot competition
L M H
L (4, 4) (2, 5) (1, 3)M (5, 2) (3, 3) (2, 1)H (3, 1) (1, 2) (0, 0)
Action H is strictly dominated by M for Player 1.The action H can be eliminated for both players.
As both players are rational and assume that their opponent is rational,we Iterate the Elimination of Strictly Dominated Strategies (IESDS).
The only rational issue of the game is (M, M)whose payoff is (3, 3).
21/62
![Page 49: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/49.jpg)
Domination - A finite variant of Cournot competition
L M H
L (4, 4) (2, 5) (1, 3)M (5, 2) (3, 3) (2, 1)H (3, 1) (1, 2) (0, 0)
Action H is strictly dominated by M for Player 2.The action H can be eliminated for both players.
As both players are rational and assume that their opponent is rational,we Iterate the Elimination of Strictly Dominated Strategies (IESDS).
The only rational issue of the game is (M, M)whose payoff is (3, 3).
21/62
![Page 50: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/50.jpg)
Domination - A finite variant of Cournot competition
L M H
L (4, 4) (2, 5) (1, 3)M (5, 2) (3, 3) (2, 1)H (3, 1) (1, 2) (0, 0)
The action H can be eliminated for both players.
As both players are rational and assume that their opponent is rational,we Iterate the Elimination of Strictly Dominated Strategies (IESDS).
The only rational issue of the game is (M, M)whose payoff is (3, 3).
21/62
![Page 51: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/51.jpg)
Domination - A finite variant of Cournot competition
L M H
L (4, 4) (2, 5) (1, 3)M (5, 2) (3, 3) (2, 1)H (3, 1) (1, 2) (0, 0)
The action H can be eliminated for both players.
As both players are rational and assume that their opponent is rational,we Iterate the Elimination of Strictly Dominated Strategies (IESDS).
The only rational issue of the game is (M, M)whose payoff is (3, 3).
21/62
![Page 52: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/52.jpg)
Domination - A finite variant of Cournot competition
L M HL (4, 4) (2, 5) (1, 3)M (5, 2) (3, 3) (2, 1)H (3, 1) (1, 2) (0, 0)
The action H can be eliminated for both players.
As both players are rational and assume that their opponent is rational,we Iterate the Elimination of Strictly Dominated Strategies (IESDS).
The only rational issue of the game is (M, M)whose payoff is (3, 3).
21/62
![Page 53: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/53.jpg)
Back to Cournot competition
ProfitA1 (a1, a2) = a1
(α− β(a1 + a2)︸ ︷︷ ︸
selling price
)− γ a1︸︷︷︸
production cost
ProfitA1(x,0)
ProfitA1(x,y0)
ProfitA1(x,y1)
α−γβ
x0x1
All actions in(α−γ
2β , α−γβ
]are strictly dominated
The IESDS converges to: (α− γ
3β,α− γ
3β
)The result is non trivial: the elimination process is infinite.
22/62
![Page 54: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/54.jpg)
Back to Cournot competition
1
βProfitA1 (x , y) = −x2 + x
(α− γβ− y
)
ProfitA1(x,0)
ProfitA1(x,y0)
ProfitA1(x,y1)
α−γβ
x0x1
All actions in(α−γ
2β , α−γβ
]are strictly dominated
The IESDS converges to: (α− γ
3β,α− γ
3β
)The result is non trivial: the elimination process is infinite.
22/62
![Page 55: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/55.jpg)
Back to Cournot competition
1
βProfitA1 (x , y) = −x2 + x
(α− γβ− y
)
ProfitA1(x,0)
ProfitA1(x,y0)
ProfitA1(x,y1)
α−γβ
x0x1
All actions in(α−γ
2β , α−γβ
]are strictly dominated
The IESDS converges to: (α− γ
3β,α− γ
3β
)The result is non trivial: the elimination process is infinite.
22/62
![Page 56: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/56.jpg)
Back to Cournot competition
1
βProfitA1 (x , y) = −x2 + x
(α− γβ− y
)
ProfitA1(x,0)
ProfitA1(x,y0)
ProfitA1(x,y1)
α−γβ
x0x1
All actions in(α−γ
2β , α−γβ
]are strictly dominated
The IESDS converges to: (α− γ
3β,α− γ
3β
)The result is non trivial: the elimination process is infinite.
22/62
![Page 57: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/57.jpg)
Back to Cournot competition
1
βProfitA1 (x , y) = −x2 + x
(α− γβ− y
)
ProfitA1(x,0)
ProfitA1(x,y0)
ProfitA1(x,y1)
α−γβ
x0x1
All actions in(α−γ
2β , α−γβ
]are strictly dominated
The IESDS converges to: (α− γ
3β,α− γ
3β
)
The result is non trivial: the elimination process is infinite.
22/62
![Page 58: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/58.jpg)
Back to Cournot competition
1
βProfitA1 (x , y) = −x2 + x
(α− γβ− y
)
ProfitA1(x,0)
ProfitA1(x,y0)
ProfitA1(x,y1)
α−γβ
x0x1
All actions in(α−γ
2β , α−γβ
]are strictly dominated
The IESDS converges to: (α− γ
3β,α− γ
3β
)The result is non trivial: the elimination process is infinite.
22/62
![Page 59: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/59.jpg)
Domination - Ice-cream sellers dilemma
The only strategies that are strictly dominated are the two borders...
23/62
![Page 60: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/60.jpg)
Summary
We have seen:
The notion of strictly dominated strategy:
+ allows to find rational issues of some games,Prisoner dilemma, Cournot competition
- not always easy to obtain the rational issue,Cournot competition
- very strong notion: rational issues are not always obtained.Ice-cream sellers dilemma
; We need another notion to determine rational issues.
24/62
![Page 61: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/61.jpg)
Summary
We have seen:
The notion of strictly dominated strategy:
+ allows to find rational issues of some games,Prisoner dilemma, Cournot competition
- not always easy to obtain the rational issue,Cournot competition
- very strong notion: rational issues are not always obtained.Ice-cream sellers dilemma
; We need another notion to determine rational issues.
24/62
![Page 62: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/62.jpg)
Summary
We have seen:
The notion of strictly dominated strategy:
+ allows to find rational issues of some games,Prisoner dilemma, Cournot competition
- not always easy to obtain the rational issue,Cournot competition
- very strong notion: rational issues are not always obtained.Ice-cream sellers dilemma
; We need another notion to determine rational issues.
24/62
![Page 63: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/63.jpg)
Summary
We have seen:
The notion of strictly dominated strategy:
+ allows to find rational issues of some games,Prisoner dilemma, Cournot competition
- not always easy to obtain the rational issue,Cournot competition
- very strong notion: rational issues are not always obtained.Ice-cream sellers dilemma
; We need another notion to determine rational issues.
24/62
![Page 64: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/64.jpg)
Outline
1 What is a game?Games we play for funA broader sense to the notion of game
2 Strategic games – Playing only once simultaneously(Strict) Domination and IterationStability: Nash equilibria
3 Extensive games – Playing several times sequentially
4 Repeated games – Playing the same game again and again
5 Conclusion
25/62
![Page 65: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/65.jpg)
Stability: the concept of Nash equilibria
[Nash50] Equilibrium Points in n-Person Games (1950).
Nash equilibrium
Let(
Agt,Σ, (gA)A∈Agt
)be a strategic game and b ∈ ΣAgt be a strategy
profile. We say that b is a Nash equilibrium iff
∀A ∈ Agt, ∀dA ∈ Σ s.t. gA(b−A, dA) ≤ gA(b)
A rational player should not deviate from the Nash equilibrium.
26/62
![Page 66: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/66.jpg)
Stability: the concept of Nash equilibria
[Nash50] Equilibrium Points in n-Person Games (1950).
Nash equilibrium
Let(
Agt,Σ, (gA)A∈Agt
)be a strategic game and b ∈ ΣAgt be a strategy
profile. We say that b is a Nash equilibrium iff
∀A ∈ Agt, ∀dA ∈ Σ s.t. gA(b−A, dA) ≤ gA(b)
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
(D, D) is the unique Nash equilibrium...
... even if (C, C) would be better for both prisoners
26/62
![Page 67: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/67.jpg)
Stability: the concept of Nash equilibria
[Nash50] Equilibrium Points in n-Person Games (1950).
Nash equilibrium
Let(
Agt,Σ, (gA)A∈Agt
)be a strategic game and b ∈ ΣAgt be a strategy
profile. We say that b is a Nash equilibrium iff
∀A ∈ Agt, ∀dA ∈ Σ s.t. gA(b−A, dA) ≤ gA(b)
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
(D, D) is the unique Nash equilibrium...
... even if (C, C) would be better for both prisoners
26/62
![Page 68: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/68.jpg)
Stability: the concept of Nash equilibria
[Nash50] Equilibrium Points in n-Person Games (1950).
Nash equilibrium
Let(
Agt,Σ, (gA)A∈Agt
)be a strategic game and b ∈ ΣAgt be a strategy
profile. We say that b is a Nash equilibrium iff
∀A ∈ Agt, ∀dA ∈ Σ s.t. gA(b−A, dA) ≤ gA(b)
L R
T (0, 0) (2, 1)B (3, 2) (1, 2)
R dominates L (but not strictly)
(B, R) is not a Nash equilibrium, but (T, R) is a Nash equilibrium
R might not be the best option...
(B, L) is optimal, hence a Nash equilibrium
26/62
![Page 69: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/69.jpg)
Back to IESDS
L M HL (4, 4) (2, 5) (1, 3)M (5, 2) (3, 3) (2, 1)H (3, 1) (1, 2) (0, 0)
The only rational issue (M, M) is a Nash equilibrium
General principle/result
No strictly dominated action can take part to a Nash equilibrium;
this is also the case in the IESDS process
A profile obtained by IESDS is a Nash equilibrium
27/62
![Page 70: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/70.jpg)
Back to IESDS
L M HL (4, 4) (2, 5) (1, 3)M (5, 2) (3, 3) (2, 1)H (3, 1) (1, 2) (0, 0)
The only rational issue (M, M) is a Nash equilibrium
General principle/result
No strictly dominated action can take part to a Nash equilibrium;
this is also the case in the IESDS process
A profile obtained by IESDS is a Nash equilibrium
27/62
![Page 71: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/71.jpg)
Back to IESDS
L M HL (4, 4) (2, 5) (1, 3)M (5, 2) (3, 3) (2, 1)H (3, 1) (1, 2) (0, 0)
The only rational issue (M, M) is a Nash equilibrium
General principle/result
No strictly dominated action can take part to a Nash equilibrium;this is also the case in the IESDS process
A profile obtained by IESDS is a Nash equilibrium
27/62
![Page 72: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/72.jpg)
Back to IESDS
L M HL (4, 4) (2, 5) (1, 3)M (5, 2) (3, 3) (2, 1)H (3, 1) (1, 2) (0, 0)
The only rational issue (M, M) is a Nash equilibrium
General principle/result
No strictly dominated action can take part to a Nash equilibrium;this is also the case in the IESDS process
A profile obtained by IESDS is a Nash equilibrium
27/62
![Page 73: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/73.jpg)
Do all the finite matrix games have a Nash equilibrium?
No!
The matching penny game
a b
a (1, 0) (0, 1)b (0, 1) (1, 0)
28/62
![Page 74: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/74.jpg)
Do all the finite matrix games have a Nash equilibrium?
No!
The matching penny game
a b
a (1, 0) (0, 1)b (0, 1) (1, 0)
28/62
![Page 75: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/75.jpg)
Do all the finite matrix games have a Nash equilibrium?
No!
The matching penny game
a b
a (1, 0) (0, 1)b (0, 1) (1, 0)
28/62
![Page 76: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/76.jpg)
Mixed strategies
Given E , we denote ∆(E ) the set of probability distributions over E .
Mixed strategy
If Σ is the of actions (or strategies), ∆(Σ) is the set of mixed strategies.
Expected payoff
Let σ = (σA1 , . . . , σAn) be a mixed strategy profile. Let A ∈ Agt:
gA(σ) =∑
b=(bA)A∈Agt∈ΣAgt
∏A∈Agt
σA(bA)
︸ ︷︷ ︸
probability of b
gA(b)
is the expected payoff of player A.
Mixed extension of game G‹G def=(
Agt,∆(Σ), (gA)A∈Agt
)is a game.
G has a mixed Nash equilibrium iff ‹G has a Nash equilibrium.
29/62
![Page 77: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/77.jpg)
Mixed strategies
Mixed strategy
If Σ is the of actions (or strategies), ∆(Σ) is the set of mixed strategies.
Expected payoff
Let σ = (σA1 , . . . , σAn) be a mixed strategy profile. Let A ∈ Agt:
gA(σ) =∑
b=(bA)A∈Agt∈ΣAgt
∏A∈Agt
σA(bA)
︸ ︷︷ ︸
probability of b
gA(b)
is the expected payoff of player A.
Mixed extension of game G‹G def=(
Agt,∆(Σ), (gA)A∈Agt
)is a game.
G has a mixed Nash equilibrium iff ‹G has a Nash equilibrium.
29/62
![Page 78: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/78.jpg)
Mixed strategies
Mixed strategy
If Σ is the of actions (or strategies), ∆(Σ) is the set of mixed strategies.
Expected payoff
Let σ = (σA1 , . . . , σAn) be a mixed strategy profile. Let A ∈ Agt:
gA(σ) =∑
b=(bA)A∈Agt∈ΣAgt
∏A∈Agt
σA(bA)
︸ ︷︷ ︸
probability of b
gA(b)
is the expected payoff of player A.
Mixed extension of game G‹G def=(
Agt,∆(Σ), (gA)A∈Agt
)is a game.
G has a mixed Nash equilibrium iff ‹G has a Nash equilibrium.
29/62
![Page 79: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/79.jpg)
Mixed strategies
Mixed strategy
If Σ is the of actions (or strategies), ∆(Σ) is the set of mixed strategies.
Expected payoff
Let σ = (σA1 , . . . , σAn) be a mixed strategy profile. Let A ∈ Agt:
gA(σ) =∑
b=(bA)A∈Agt∈ΣAgt
∏A∈Agt
σA(bA)
︸ ︷︷ ︸
probability of b
gA(b)
is the expected payoff of player A.
Mixed extension of game G‹G def=(
Agt,∆(Σ), (gA)A∈Agt
)is a game.
G has a mixed Nash equilibrium iff ‹G has a Nash equilibrium.
29/62
![Page 80: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/80.jpg)
Mixed strategies
Mixed strategy
If Σ is the of actions (or strategies), ∆(Σ) is the set of mixed strategies.
Expected payoff· · ·
Mixed extension of game G‹G def=(
Agt,∆(Σ), (gA)A∈Agt
)is a game.
G has a mixed Nash equilibrium iff ‹G has a Nash equilibrium.
29/62
![Page 81: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/81.jpg)
Nash equilibria in mixed strategies
[Nash50] Equilibrium Points in n-Person Games (1950).
a b
a (1, 0) (0, 1)b (0, 1) (1, 0)
The following profile is a Nash equilibrium in mixed strategies:
σA1 =1
2· a +
1
2· b and σA2 =
1
2· a +
1
2· b
whose expected payoff is ( 12 ,
12 ).
Nash Theorem [Nash50]
Any finite game admits mixed Nash equilibria.
30/62
![Page 82: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/82.jpg)
Nash equilibria in mixed strategies
[Nash50] Equilibrium Points in n-Person Games (1950).
a b
a (1, 0) (0, 1)b (0, 1) (1, 0)
The following profile is a Nash equilibrium in mixed strategies:
σA1 =1
2· a +
1
2· b and σA2 =
1
2· a +
1
2· b
whose expected payoff is ( 12 ,
12 ).
Nash Theorem [Nash50]
Any finite game admits mixed Nash equilibria.
30/62
![Page 83: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/83.jpg)
Best response
Best response
Let A ∈ Agt and a−A ∈ ΣAgt\{A} be a strategy profile for A’s opponents.
We say that bA ∈ Σ is a best response to a−A if
∀cA ∈ Σ gA(cA, a−A) ≤ gA(bA, a−A)
31/62
![Page 84: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/84.jpg)
Best response
Best response
Let A ∈ Agt and a−A ∈ ΣAgt\{A} be a strategy profile for A’s opponents.We say that bA ∈ Σ is a best response to a−A if
∀cA ∈ Σ gA(cA, a−A) ≤ gA(bA, a−A)
31/62
![Page 85: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/85.jpg)
Best response
Best response
Let A ∈ Agt and a−A ∈ ΣAgt\{A} be a strategy profile for A’s opponents.We say that bA ∈ Σ is a best response to a−A if
∀cA ∈ Σ gA(cA, a−A) ≤ gA(bA, a−A)
Example: Prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
A best response (for Prisoner 1) to C is
D.
31/62
![Page 86: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/86.jpg)
Best response
Best response
Let A ∈ Agt and a−A ∈ ΣAgt\{A} be a strategy profile for A’s opponents.We say that bA ∈ Σ is a best response to a−A if
∀cA ∈ Σ gA(cA, a−A) ≤ gA(bA, a−A)
Example: Prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
A best response (for Prisoner 1) to C is D.
31/62
![Page 87: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/87.jpg)
Best response
Best response
Let A ∈ Agt and a−A ∈ ΣAgt\{A} be a strategy profile for A’s opponents.We say that bA ∈ Σ is a best response to a−A if
∀cA ∈ Σ gA(cA, a−A) ≤ gA(bA, a−A)
Best response correspondence of Player A
BRA : ΣAgt\{A} → P(Σ)
a−A → {bA | bA is a best response to a−A}
Best response correspondence of the game
BR : ΣAgt → P(ΣAgt
)a→
∏A∈Agt
BRA(a−A)
31/62
![Page 88: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/88.jpg)
Best response
Best response
Let A ∈ Agt and a−A ∈ ΣAgt\{A} be a strategy profile for A’s opponents.We say that bA ∈ Σ is a best response to a−A if
∀cA ∈ Σ gA(cA, a−A) ≤ gA(bA, a−A)
Best response correspondence of Player A
BRA : ΣAgt\{A} → P(Σ)
a−A → {bA | bA is a best response to a−A}
Best response correspondence of the game
BR : ΣAgt → P(ΣAgt
)a→
∏A∈Agt
BRA(a−A)
31/62
![Page 89: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/89.jpg)
Best response and Nash equilibrium
Proposition
Let a be a strategy profile.
a is a Nash equilibrium if and only if a ∈ BR(a)
32/62
![Page 90: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/90.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
33/62
![Page 91: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/91.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
A strategy consists in giving a probability distribution over {T, B} (resp.{L, R}), that is, it consists in fixing the probability to play T (resp. L).
Assume
σA1 =1
4· T +
3
4· B and σA2 =
1
2· L +
1
2· R
the expected payoff is:
33/62
![Page 92: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/92.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
A strategy consists in giving a probability distribution over {T, B} (resp.{L, R}), that is, it consists in fixing the probability to play T (resp. L).
Assume
σA1 =1
4· T +
3
4· B and σA2 =
1
2· L +
1
2· R
the expected payoff is:
gA1
(1
4,
1
2
)=
7
8gA2
(1
4,
1
2
)= −7
8
33/62
![Page 93: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/93.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
In general, we have
σA1 = α · T + (1− α) · B and σA2 = β · L + (1− β) · R
whose expected payoff is:
33/62
![Page 94: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/94.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
In general, we have
σA1 = α · T + (1− α) · B and σA2 = β · L + (1− β) · R
whose expected payoff is:
gA1 (α, β) = α(3β − 2)− 2β + 2 = −gA2 (α, β)
33/62
![Page 95: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/95.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
gA1 (α, β) = α(3β − 2)− 2β + 2
BRA1 (β) =
{1} if 3β − 2 > 0
[0, 1] if 3β − 2 = 0
{0} if 3β − 2 < 0
α0 1 α0 1
33/62
![Page 96: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/96.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
gA1 (α, β) = α(3β − 2)− 2β + 2
BRA1 (β) =
{1} if 3β − 2 > 0
[0, 1] if 3β − 2 = 0
{0} if 3β − 2 < 0
α0 1 α0 1
33/62
![Page 97: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/97.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
gA1 (α, β) = α(3β − 2)− 2β + 2
BRA1 (β) =
{1} if 3β − 2 > 0
[0, 1] if 3β − 2 = 0
{0} if 3β − 2 < 0
α0 1 α0 1 α0 1
33/62
![Page 98: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/98.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
gA1 (α, β) = α(3β − 2)− 2β + 2
BRA1 (β) =
{1} if 3β − 2 > 0
[0, 1] if 3β − 2 = 0
{0} if 3β − 2 < 0
α0 1 α0 1 α0 1
33/62
![Page 99: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/99.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
BRA1 (β) =
{1} if 3β − 2 > 0
[0, 1] if 3β − 2 = 0
{0} if 3β − 2 < 0
BRA2 (α) =
{1} if 3α− 2 < 0
[0, 1] if 3α− 2 = 0
{0} if 3α− 2 > 0
α
β
0 123
23
33/62
![Page 100: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/100.jpg)
An example
L R
T (1,−1) (0, 0)B (0, 0) (2,−2)
Thus the following profile is an equilibrium in mixed strategies:
σA1 =2
3· T +
1
3· B and σA2 =
2
3· L +
1
3· R
whose expected payoff is: (2
3,−2
3
)
33/62
![Page 101: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/101.jpg)
Best response – Back to Cournot competition
gA1(x,y0) =−βx2+(α−βy0−γ)x
α−γβ −y0
α−γ2β −
y02 BRA1
(y0)=
{α−γ
2β −y0
2
}if α−γ
β ≥ y0
{0} otherwise
(x , y) ∈ BR(x , y) iff x ∈ BRA1 (y) and y ∈ BRA2 (x)
iff (x , y) =
(α− γ
3β,α− γ
3β
)
34/62
![Page 102: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/102.jpg)
Best response – Back to Cournot competition
gA1(x,y0) =−βx2+(α−βy0−γ)x
α−γβ −y0α−γ
2β −y02
BRA1(y0)=
{α−γ
2β −y0
2
}if α−γ
β ≥ y0
{0} otherwise
(x , y) ∈ BR(x , y) iff x ∈ BRA1 (y) and y ∈ BRA2 (x)
iff (x , y) =
(α− γ
3β,α− γ
3β
)
34/62
![Page 103: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/103.jpg)
Best response – Back to Cournot competition
gA1(x,y0) =−βx2+(α−βy0−γ)x
α−γβ −y0α−γ
2β −y02 BRA1
(y0)=
{α−γ
2β −y0
2
}if α−γ
β ≥ y0
{0} otherwise
(x , y) ∈ BR(x , y) iff x ∈ BRA1 (y) and y ∈ BRA2 (x)
iff (x , y) =
(α− γ
3β,α− γ
3β
)
34/62
![Page 104: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/104.jpg)
Best response – Back to Cournot competition
gA1(x,y0) =−βx2+(α−βy0−γ)x
α−γβ −y0α−γ
2β −y02 BRA1
(y0)=
{α−γ
2β −y0
2
}if α−γ
β ≥ y0
{0} otherwise
x
y
BRA1(y)
x
y
BRA2(x)
(x , y) ∈ BR(x , y) iff x ∈ BRA1 (y) and y ∈ BRA2 (x)
iff (x , y) =
(α− γ
3β,α− γ
3β
)
34/62
![Page 105: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/105.jpg)
Best response – Back to Cournot competition
gA1(x,y0) =−βx2+(α−βy0−γ)x
α−γβ −y0α−γ
2β −y02 BRA1
(y0)=
{α−γ
2β −y0
2
}if α−γ
β ≥ y0
{0} otherwise
x
y
(x , y) ∈ BR(x , y) iff x ∈ BRA1 (y) and y ∈ BRA2 (x)
iff (x , y) =
(α− γ
3β,α− γ
3β
)
34/62
![Page 106: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/106.jpg)
Best response – Back to Cournot competition
gA1(x,y0) =−βx2+(α−βy0−γ)x
α−γβ −y0α−γ
2β −y02 BRA1
(y0)=
{α−γ
2β −y0
2
}if α−γ
β ≥ y0
{0} otherwise
x
y
(x , y) ∈ BR(x , y) iff x ∈ BRA1 (y) and y ∈ BRA2 (x)
iff (x , y) =
(α− γ
3β,α− γ
3β
)
34/62
![Page 107: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/107.jpg)
Best response – Back to Cournot competition
gA1(x,y0) =−βx2+(α−βy0−γ)x
α−γβ −y0α−γ
2β −y02 BRA1
(y0)=
{α−γ
2β −y0
2
}if α−γ
β ≥ y0
{0} otherwise
x
y
(x , y) ∈ BR(x , y) iff x ∈ BRA1 (y) and y ∈ BRA2 (x)
iff (x , y) =
(α− γ
3β,α− γ
3β
)34/62
![Page 108: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/108.jpg)
Best response – Back to the ice-cream sellers dilemma
One can show that the only Nash equilibrium is:
35/62
![Page 109: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/109.jpg)
Best response – Back to the ice-cream sellers dilemma
One can show that the only Nash equilibrium is:
35/62
![Page 110: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/110.jpg)
Best response and Nash equilibrium
Proposition
Let a be a strategy profile.
a is a Nash equilibrium if and only if a ∈ BR(a)
Nash Theorem [Nash50]
Any finite game admits mixed Nash equilibria.
Key ingredient of the proof: Brouwer’s fixpoint theoremOr simply Kakutani’s fixpoint theorem
36/62
![Page 111: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/111.jpg)
Best response and Nash equilibrium
Proposition
Let a be a strategy profile.
a is a Nash equilibrium if and only if a ∈ BR(a)
Nash Theorem [Nash50]
Any finite game admits mixed Nash equilibria.
Key ingredient of the proof: Brouwer’s fixpoint theoremOr simply Kakutani’s fixpoint theorem
36/62
![Page 112: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/112.jpg)
Best response and Nash equilibrium
Proposition
Let a be a strategy profile.
a is a Nash equilibrium if and only if a ∈ BR(a)
Nash Theorem [Nash50]
Any finite game admits mixed Nash equilibria.
Key ingredient of the proof: Brouwer’s fixpoint theoremOr simply Kakutani’s fixpoint theorem
36/62
![Page 113: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/113.jpg)
Fixpoint theorems
Brouwer’s fixpoint theorem
Let X ⊆ Rn be a convex, compact and nonempty set. Then everycontinuous function f : X → X has a fixpoint.
Kakutani’s fixpoint theorem
Let X be a non-empty, compact and convex subset of Rn. Let f : X → 2X
be a set-valued function on X with a closed graph and the property thatf (x) is non-empty and convex for all x ∈ X . Then f has a fixpoint.
; One can obtain twists or generalizations of Nash Theorem(ex: Nash-Glicksberg Theorem on compact sets of actions)
37/62
![Page 114: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/114.jpg)
Fixpoint theorems
Brouwer’s fixpoint theorem
Let X ⊆ Rn be a convex, compact and nonempty set. Then everycontinuous function f : X → X has a fixpoint.
Kakutani’s fixpoint theorem
Let X be a non-empty, compact and convex subset of Rn. Let f : X → 2X
be a set-valued function on X with a closed graph and the property thatf (x) is non-empty and convex for all x ∈ X . Then f has a fixpoint.
; One can obtain twists or generalizations of Nash Theorem(ex: Nash-Glicksberg Theorem on compact sets of actions)
37/62
![Page 115: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/115.jpg)
Outline
1 What is a game?Games we play for funA broader sense to the notion of game
2 Strategic games – Playing only once simultaneously(Strict) Domination and IterationStability: Nash equilibria
3 Extensive games – Playing several times sequentially
4 Repeated games – Playing the same game again and again
5 Conclusion
38/62
![Page 116: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/116.jpg)
A chocolate example
A firm A1 has a monopoly on the production of chocolate.
Another firm A2 would like to enter the market of chocolate.But entering the market has a cost !
If A2 enters the market, then A1 can share the clients or undersell.
A2
A1 10, 0
Not EnterEnter
5, 4 4,−1
Share Undersell
39/62
![Page 117: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/117.jpg)
A chocolate example
A firm A1 has a monopoly on the production of chocolate.
Another firm A2 would like to enter the market of chocolate.But entering the market has a cost !
If A2 enters the market, then A1 can share the clients or undersell.
A2
A1 10, 0
Not EnterEnter
5, 4 4,−1
Share Undersell
39/62
![Page 118: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/118.jpg)
A chocolate example
A firm A1 has a monopoly on the production of chocolate.
Another firm A2 would like to enter the market of chocolate.But entering the market has a cost !
If A2 enters the market, then A1 can share the clients or undersell.
A2
A1 10, 0
Not EnterEnter
5, 4 4,−1
Share Undersell
39/62
![Page 119: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/119.jpg)
A chocolate example
A firm A1 has a monopoly on the production of chocolate.
Another firm A2 would like to enter the market of chocolate.But entering the market has a cost !
If A2 enters the market, then A1 can share the clients or undersell.
A2
A1
Not Enter
10, 0
Not Enter
Enter
5, 4 4,−1
Share Undersell
39/62
![Page 120: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/120.jpg)
A chocolate example
A firm A1 has a monopoly on the production of chocolate.
Another firm A2 would like to enter the market of chocolate.But entering the market has a cost !
If A2 enters the market, then A1 can share the clients or undersell.
A2
A1 10, 0
Not EnterEnter
5, 4 4,−1
Share Undersell
39/62
![Page 121: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/121.jpg)
A chocolate example
A firm A1 has a monopoly on the production of chocolate.
Another firm A2 would like to enter the market of chocolate.But entering the market has a cost !
If A2 enters the market, then A1 can share the clients or undersell.
A2
A1 10, 0
Not EnterEnter
Share Undersell
5, 4 4,−1
Share Undersell
39/62
![Page 122: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/122.jpg)
A chocolate example
A firm A1 has a monopoly on the production of chocolate.
Another firm A2 would like to enter the market of chocolate.But entering the market has a cost !
If A2 enters the market, then A1 can share the clients or undersell.
A2
A1 10, 0
Not EnterEnter
5, 4 4,−1
Share Undersell
39/62
![Page 123: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/123.jpg)
Finite extensive games with perfect information
· · ·
......
...
40/62
![Page 124: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/124.jpg)
Finite extensive games with perfect information
· · ·
......
...
40/62
![Page 125: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/125.jpg)
Finite extensive games with perfect information
A1
A2 A2
A1 A3 A3
A3 A2 A1
· · ·
......
...
40/62
![Page 126: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/126.jpg)
Finite extensive games with perfect information
A1
A2 A2
A1 A3 A3
A3 A2 A1
7,4,2 2,0,0 4,2,1
5,3,8 2,7,8
3,5,38,7,6 1,9,1
· · ·
......
...
40/62
![Page 127: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/127.jpg)
Strategies
A1
A2 A2
A1 A3 A3
A3 A2 A1
7,4,2 2,0,0 4,2,1
5,3,8 2,7,8
3,5,38,7,6 1,9,1
· · ·
......
...
41/62
![Page 128: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/128.jpg)
Strategies
A1
A2 A2
A1 A3 A3
A3 A2 A1
7,4,2 2,0,0 4,2,1
5,3,8 2,7,8
3,5,38,7,6 1,9,1
· · ·
......
...
σA1 : strategy of A1,
41/62
![Page 129: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/129.jpg)
Strategies
A1
A2 A2
A1 A3 A3
A3 A2 A1
7,4,2 2,0,0 4,2,1
5,3,8 2,7,8
3,5,38,7,6 1,9,1
· · ·
......
...
σA1 : strategy of A1, σA2 : strategy of A2,
41/62
![Page 130: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/130.jpg)
Strategies
A1
A2 A2
A1 A3 A3
A3 A2 A1
7,4,2 2,0,0 4,2,1
5,3,8 2,7,8
3,5,38,7,6 1,9,1
· · ·
......
...
σA1 : strategy of A1, σA2 : strategy of A2, σA3 : strategy of A3.
41/62
![Page 131: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/131.jpg)
Strategies
A1
A2 A2
A1 A3 A3
A3 A2 A1
7,4,2 2,0,0 4,2,1
5,3,8 2,7,8
3,5,38,7,6 1,9,1
· · ·
......
...
Outcome (σA1 , σA2 , σA3 ) is the branch determined by the three strategies.
41/62
![Page 132: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/132.jpg)
Strategies
A1
A2 A2
A1 A3 A3
A3 A2 A1
7,4,2 2,0,0 4,2,1
5,3,8 2,7,8
3,5,38,7,6 1,9,1
· · ·
......
...
One could also have concurrent nodes, or stochastic nodes.One could also consider randomized strategies.
41/62
![Page 133: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/133.jpg)
Extensive games as strategic games
A finite extensive game can always beturned into a strategic game!
A2
A1 10, 0
E NE
5, 4 4,−1
S U
E NE
S (5, 4) (10, 0)U (4,−1) (10, 0)
; Notion of Nash equilibria applies
Nash equilibria
(S, E) whose payoff is (5, 4)
(U, NE) whose payoff is (10, 0)
42/62
![Page 134: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/134.jpg)
Extensive games as strategic games
A finite extensive game can always beturned into a strategic game!
A2
A1 10, 0
E NE
5, 4 4,−1
S U
E NE
S (5, 4) (10, 0)U (4,−1) (10, 0)
; Notion of Nash equilibria applies
Nash equilibria
(S, E) whose payoff is (5, 4)
(U, NE) whose payoff is (10, 0)
42/62
![Page 135: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/135.jpg)
Extensive games as strategic games
A finite extensive game can always beturned into a strategic game!
A2
A1 10, 0
E NE
5, 4 4,−1
S U
E NE
S (5, 4) (10, 0)U (4,−1) (10, 0)
; Notion of Nash equilibria applies
Nash equilibria
(S, E) whose payoff is (5, 4)
(U, NE) whose payoff is (10, 0)
42/62
![Page 136: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/136.jpg)
Extensive games as strategic games
A finite extensive game can always beturned into a strategic game!
A2
A1 10, 0
E NE
5, 4 4,−1
S U
E NE
S (5, 4) (10, 0)U (4,−1) (10, 0)
; Notion of Nash equilibria applies
Nash equilibria
(S, E) whose payoff is (5, 4)
(U, NE) whose payoff is (10, 0)
42/62
![Page 137: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/137.jpg)
Extensive games as strategic games
A finite extensive game can always beturned into a strategic game!
A2
A1 10, 0
E NE
5, 4 4,−1
S U
E NE
S (5, 4) (10, 0)U (4,−1) (10, 0)
; Notion of Nash equilibria applies
Nash equilibria
(S, E) whose payoff is (5, 4)
(U, NE) whose payoff is (10, 0)
42/62
![Page 138: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/138.jpg)
Extensive games as strategic games
A finite extensive game can always beturned into a strategic game!
A2
A1 10, 0
E NE
5, 4 4,−1
S U
E NE
S (5, 4) (10, 0)U (4,−1) (10, 0)
; Notion of Nash equilibria applies
Nash equilibria
(S, E) whose payoff is (5, 4)
(U, NE) whose payoff is (10, 0)
42/62
![Page 139: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/139.jpg)
Randomized strategies
[Kuhn57] Kuhn. Extensive games and the problem of information (Contribution to the Theory of Games).
[Aum64] Aumann. Mixed and behaviour strategies in infinite extensive games (Advances in Game Theory).
Mixed strategies: distribution over pure strategies
Behavior strategies: randomize at each step!
Kuhn’s Theorem for extensive games [Kuhn57]
Under a perfect recall hypothesis, mixed and behavior strategies coincidein finite extensive games.
Note: extends to infinite extensive games [Aum64]
Corollary
In a finite extensive game (with perfect information), there always existsa Nash equilibrium in behavior strategies.
43/62
![Page 140: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/140.jpg)
Randomized strategies
[Kuhn57] Kuhn. Extensive games and the problem of information (Contribution to the Theory of Games).
[Aum64] Aumann. Mixed and behaviour strategies in infinite extensive games (Advances in Game Theory).
Mixed strategies: distribution over pure strategies
Behavior strategies: randomize at each step!
Kuhn’s Theorem for extensive games [Kuhn57]
Under a perfect recall hypothesis, mixed and behavior strategies coincidein finite extensive games.
Note: extends to infinite extensive games [Aum64]
Corollary
In a finite extensive game (with perfect information), there always existsa Nash equilibrium in behavior strategies.
43/62
![Page 141: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/141.jpg)
Randomized strategies
[Kuhn57] Kuhn. Extensive games and the problem of information (Contribution to the Theory of Games).[Aum64] Aumann. Mixed and behaviour strategies in infinite extensive games (Advances in Game Theory).
Mixed strategies: distribution over pure strategies
Behavior strategies: randomize at each step!
Kuhn’s Theorem for extensive games [Kuhn57]
Under a perfect recall hypothesis, mixed and behavior strategies coincidein finite extensive games.Note: extends to infinite extensive games [Aum64]
Corollary
In a finite extensive game (with perfect information), there always existsa Nash equilibrium in behavior strategies.
43/62
![Page 142: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/142.jpg)
Randomized strategies
[Kuhn57] Kuhn. Extensive games and the problem of information (Contribution to the Theory of Games).[Aum64] Aumann. Mixed and behaviour strategies in infinite extensive games (Advances in Game Theory).
Mixed strategies: distribution over pure strategies
Behavior strategies: randomize at each step!
Kuhn’s Theorem for extensive games [Kuhn57]
Under a perfect recall hypothesis, mixed and behavior strategies coincidein finite extensive games.Note: extends to infinite extensive games [Aum64]
Corollary
In a finite extensive game (with perfect information), there always existsa Nash equilibrium in behavior strategies.
43/62
![Page 143: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/143.jpg)
Stackelberg competition
The Stackelberg leadership model is a strategic game in which the leaderfirm moves first and then the follower firms move sequentially.
Let ai denote the quantity produced by the i-th firm.
ProfitA1 (a1, a2) = a1
(α− β(a1 + a2)︸ ︷︷ ︸
selling price
)− γ a1︸︷︷︸
production cost
What should be the amount of the output to optimize the profit?
44/62
![Page 144: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/144.jpg)
Stackelberg competition
The Stackelberg leadership model is a strategic game in which the leaderfirm moves first and then the follower firms move sequentially.
Let ai denote the quantity produced by the i-th firm.
ProfitA1 (a1, a2) = a1
(α− β(a1 + a2)︸ ︷︷ ︸
selling price
)− γ a1︸︷︷︸
production cost
What should be the amount of the output to optimize the profit?
44/62
![Page 145: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/145.jpg)
Stackelberg competition
The Stackelberg leadership model is a strategic game in which the leaderfirm moves first and then the follower firms move sequentially.
Let ai denote the quantity produced by the i-th firm.
ProfitA1 (a1, a2) = a1
(α− β(a1 + a2)︸ ︷︷ ︸
selling price
)− γ a1︸︷︷︸
production cost
What should be the amount of the output to optimize the profit?
44/62
![Page 146: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/146.jpg)
Cournot vs Stackelberg (simplified)
vs
ProfitA1 (a1, a2) = a1
(α− (a1 + a2)
)− γ a1
Nash equilibria
Cournot:(α−γ
3 , α−γ3
)with payoff
((α−γ)2
9 , (α−γ)2
9
).
Stackelberg:(α−γ
2 , α−γ4
)with payoff
((α−γ)2
8 , (α−γ)2
16
).
45/62
![Page 147: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/147.jpg)
Outline
1 What is a game?Games we play for funA broader sense to the notion of game
2 Strategic games – Playing only once simultaneously(Strict) Domination and IterationStability: Nash equilibria
3 Extensive games – Playing several times sequentially
4 Repeated games – Playing the same game again and again
5 Conclusion
46/62
![Page 148: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/148.jpg)
Back to the prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
The only Nash equilibrium is (D, D) whose payoff is (−5,−5)
However (C, C) whose payoff is (−3,−3) seems “more rational”
Strategic games are “one shot” games; there is no tomorrow, treason has no consequence
What would happen if the game was repeated again and again?
47/62
![Page 149: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/149.jpg)
Back to the prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
The only Nash equilibrium is (D, D) whose payoff is (−5,−5)
However (C, C) whose payoff is (−3,−3) seems “more rational”
Strategic games are “one shot” games; there is no tomorrow, treason has no consequence
What would happen if the game was repeated again and again?
47/62
![Page 150: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/150.jpg)
Back to the prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
The only Nash equilibrium is (D, D) whose payoff is (−5,−5)
However (C, C) whose payoff is (−3,−3) seems “more rational”
Strategic games are “one shot” games; there is no tomorrow, treason has no consequence
What would happen if the game was repeated again and again?
47/62
![Page 151: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/151.jpg)
Back to the prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
The only Nash equilibrium is (D, D) whose payoff is (−5,−5)
However (C, C) whose payoff is (−3,−3) seems “more rational”
Strategic games are “one shot” games; there is no tomorrow, treason has no consequence
What would happen if the game was repeated again and again?
47/62
![Page 152: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/152.jpg)
Prisoner dilemmaAs an extensive game with simultaneous moves
(−3,−3) (−10,0) (0,−10) (−5,−5)
C ,C
C,D
D,C
D,D
We need to define what will be the payoff in such a repeated game
48/62
![Page 153: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/153.jpg)
Prisoner dilemmaAs an extensive game with simultaneous moves
G
(−3,−3) (−10,0) (0,−10) (−5,−5)
C ,C
C,D
D,C
D,D
We need to define what will be the payoff in such a repeated game
48/62
![Page 154: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/154.jpg)
Prisoner dilemmaRepeated twice
C ,C
C,D
D,C
D,D
We need to define what will be the payoff in such a repeated game
48/62
![Page 155: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/155.jpg)
Prisoner dilemmaRepeated twice
C ,C
C,D
D,C
D,D
We need to define what will be the payoff in such a repeated game
48/62
![Page 156: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/156.jpg)
Prisoner dilemmaRepeated three times
C ,C
C,D
D,C
D,D
We need to define what will be the payoff in such a repeated game
48/62
![Page 157: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/157.jpg)
Prisoner dilemmaRepeated infinitely
......
......
......
......
......
......
......
......
C ,C
C,D
D,C
D,D
We need to define what will be the payoff in such a repeated game
48/62
![Page 158: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/158.jpg)
Prisoner dilemmaRepeated infinitely
......
......
......
......
......
......
......
......
C ,C
C,D
D,C
D,D
We need to define what will be the payoff in such a repeated game
48/62
![Page 159: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/159.jpg)
Repeated games
Given G = (Agt,Σ, (gA)A∈Agt) and t ∈ N,
at denotes the profile of actions played at the tth repetition of G .
A finitely repeated game denoted ΓT (where T ∈ N>0)
gTA (a1, . . . , aT ) =
1
T
T∑t=1
gA(at)
A discounted game denoted Γλ (where λ ∈ (0, 1))
gλA (a1, a2, . . .) =∞∑t=1
λt−1(1− λ)gA(at)
An infinitely repeated game denoted Γ∞
g∞A (a1, a2, . . .) = limT→∞
1
T
T∑t=1
gA(at)
49/62
![Page 160: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/160.jpg)
Repeated games
Given G = (Agt,Σ, (gA)A∈Agt) and t ∈ N,
at denotes the profile of actions played at the tth repetition of G .
A finitely repeated game denoted ΓT (where T ∈ N>0)
gTA (a1, . . . , aT ) =
1
T
T∑t=1
gA(at)
A discounted game denoted Γλ (where λ ∈ (0, 1))
gλA (a1, a2, . . .) =∞∑t=1
λt−1(1− λ)gA(at)
An infinitely repeated game denoted Γ∞
g∞A (a1, a2, . . .) = limT→∞
1
T
T∑t=1
gA(at)
49/62
![Page 161: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/161.jpg)
Repeated games
Given G = (Agt,Σ, (gA)A∈Agt) and t ∈ N,
at denotes the profile of actions played at the tth repetition of G .
A finitely repeated game denoted ΓT (where T ∈ N>0)
gTA (a1, . . . , aT ) =
1
T
T∑t=1
gA(at)
A discounted game denoted Γλ (where λ ∈ (0, 1))
gλA (a1, a2, . . .) =∞∑t=1
λt−1(1− λ)gA(at)
An infinitely repeated game denoted Γ∞
g∞A (a1, a2, . . .) = limT→∞
1
T
T∑t=1
gA(at)
49/62
![Page 162: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/162.jpg)
Repeated games
Given G = (Agt,Σ, (gA)A∈Agt) and t ∈ N,
at denotes the profile of actions played at the tth repetition of G .
A finitely repeated game denoted ΓT (where T ∈ N>0)
gTA (a1, . . . , aT ) =
1
T
T∑t=1
gA(at)
A discounted game denoted Γλ (where λ ∈ (0, 1))
gλA (a1, a2, . . .) =∞∑t=1
λt−1(1− λ)gA(at)
An infinitely repeated game denoted Γ∞
g∞A (a1, a2, . . .) = limT→∞
1
T
T∑t=1
gA(at)
49/62
![Page 163: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/163.jpg)
Equilibria in repeated games
RemarkSince repeated games are particular extensive games with perfectinformation, the notion of Nash equilibrium extends.
We denote by ET (resp. Eλ and E∞) the set of payoffs of the Nashequilibria in the game ΓT (resp. Γλ and Γ∞) in mixed strategies.
Two approaches to the study of (infinitely) repeated games
the compact approach:Study the equilibria of ΓT and observe what happens when T →∞Study the equilibria of Γλ and observe what happens when λ→ 1
the uniform approach:Study “directly” the equilibria of Γ∞
50/62
![Page 164: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/164.jpg)
Equilibria in repeated games
RemarkSince repeated games are particular extensive games with perfectinformation, the notion of Nash equilibrium extends.
We denote by ET (resp. Eλ and E∞) the set of payoffs of the Nashequilibria in the game ΓT (resp. Γλ and Γ∞) in mixed strategies.
Two approaches to the study of (infinitely) repeated games
the compact approach:Study the equilibria of ΓT and observe what happens when T →∞Study the equilibria of Γλ and observe what happens when λ→ 1
the uniform approach:Study “directly” the equilibria of Γ∞
50/62
![Page 165: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/165.jpg)
Equilibria in repeated games
RemarkSince repeated games are particular extensive games with perfectinformation, the notion of Nash equilibrium extends.
We denote by ET (resp. Eλ and E∞) the set of payoffs of the Nashequilibria in the game ΓT (resp. Γλ and Γ∞) in mixed strategies.
Two approaches to the study of (infinitely) repeated games
the compact approach:Study the equilibria of ΓT and observe what happens when T →∞Study the equilibria of Γλ and observe what happens when λ→ 1
the uniform approach:Study “directly” the equilibria of Γ∞
50/62
![Page 166: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/166.jpg)
Equilibria in repeated games
RemarkSince repeated games are particular extensive games with perfectinformation, the notion of Nash equilibrium extends.
We denote by ET (resp. Eλ and E∞) the set of payoffs of the Nashequilibria in the game ΓT (resp. Γλ and Γ∞) in mixed strategies.
Two approaches to the study of (infinitely) repeated games
the compact approach:Study the equilibria of ΓT and observe what happens when T →∞Study the equilibria of Γλ and observe what happens when λ→ 1
the uniform approach:Study “directly” the equilibria of Γ∞
50/62
![Page 167: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/167.jpg)
Equilibria in repeated games - Prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
One can prove that:
For all T ∈ N0: ET = {(−5,−5)}
For 35 ≤ λ < 1, (−3,−3) ∈ Eλ
(−3,−3) ∈ E∞
Grim-Trigger strategy: play C as long as everyone plays C; play D
otherwise
Payoff of main outcome: (−3,−3)
Payoff of any deviation (C, C) · · · (C, C)(D, C)(-, D)(-, D) · · · is < −3
; No profitable deviation
51/62
![Page 168: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/168.jpg)
Equilibria in repeated games - Prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
One can prove that:
For all T ∈ N0: ET = {(−5,−5)}
For 35 ≤ λ < 1, (−3,−3) ∈ Eλ
(−3,−3) ∈ E∞
Grim-Trigger strategy: play C as long as everyone plays C; play D
otherwise
Payoff of main outcome: (−3,−3)
Payoff of any deviation (C, C) · · · (C, C)(D, C)(-, D)(-, D) · · · is < −3
; No profitable deviation
51/62
![Page 169: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/169.jpg)
Equilibria in repeated games - Prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
One can prove that:
For all T ∈ N0: ET = {(−5,−5)}
For 35 ≤ λ < 1, (−3,−3) ∈ Eλ
(−3,−3) ∈ E∞
Grim-Trigger strategy: play C as long as everyone plays C; play D
otherwise
Payoff of main outcome: (−3,−3)
Payoff of any deviation (C, C) · · · (C, C)(D, C)(-, D)(-, D) · · · is < −3
; No profitable deviation
51/62
![Page 170: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/170.jpg)
Equilibria in repeated games - Prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
One can prove that:
For all T ∈ N0: ET = {(−5,−5)}
For 35 ≤ λ < 1, (−3,−3) ∈ Eλ
(−3,−3) ∈ E∞
Grim-Trigger strategy: play C as long as everyone plays C; play D
otherwise
Payoff of main outcome: (−3,−3)
Payoff of any deviation (C, C) · · · (C, C)(D, C)(-, D)(-, D) · · · is < −3
; No profitable deviation
51/62
![Page 171: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/171.jpg)
Nash equilibria in ΓT
If one repeats local Nash equilibria of the one-shot game Γ1, thenthis forms a Nash equilibrium in ΓT
Are there other equilibria?
A variant of the prisoner’s dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
The unique Nash equilibrium of Γ1 is (D, D) with payoff (1, 1)
Strategy profile in Γ2: play C in the first round and then D, unlessthe other player did not play as expected, in which case play P
Not so easy to compute the sets ET ...
52/62
![Page 172: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/172.jpg)
Nash equilibria in ΓT
If one repeats local Nash equilibria of the one-shot game Γ1, thenthis forms a Nash equilibrium in ΓT
Are there other equilibria?
A variant of the prisoner’s dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
The unique Nash equilibrium of Γ1 is (D, D) with payoff (1, 1)
Strategy profile in Γ2: play C in the first round and then D, unlessthe other player did not play as expected, in which case play P
Not so easy to compute the sets ET ...
52/62
![Page 173: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/173.jpg)
Nash equilibria in ΓT
If one repeats local Nash equilibria of the one-shot game Γ1, thenthis forms a Nash equilibrium in ΓT
Are there other equilibria?
A variant of the prisoner’s dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
The unique Nash equilibrium of Γ1 is (D, D) with payoff (1, 1)
Strategy profile in Γ2: play C in the first round and then D, unlessthe other player did not play as expected, in which case play P
Not so easy to compute the sets ET ...
52/62
![Page 174: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/174.jpg)
Nash equilibria in ΓT
If one repeats local Nash equilibria of the one-shot game Γ1, thenthis forms a Nash equilibrium in ΓT
Are there other equilibria?
A variant of the prisoner’s dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
The unique Nash equilibrium of Γ1 is (D, D) with payoff (1, 1)
Strategy profile in Γ2: play C in the first round and then D, unlessthe other player did not play as expected, in which case play P
Not so easy to compute the sets ET ...
52/62
![Page 175: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/175.jpg)
Nash equilibria in ΓT
If one repeats local Nash equilibria of the one-shot game Γ1, thenthis forms a Nash equilibrium in ΓT
Are there other equilibria?
A variant of the prisoner’s dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
The unique Nash equilibrium of Γ1 is (D, D) with payoff (1, 1)
Strategy profile in Γ2: play C in the first round and then D, unlessthe other player did not play as expected, in which case play P
Not so easy to compute the sets ET ...
52/62
![Page 176: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/176.jpg)
Nash equilibria in ΓT
If one repeats local Nash equilibria of the one-shot game Γ1, thenthis forms a Nash equilibrium in ΓT
Are there other equilibria?
A variant of the prisoner’s dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
The unique Nash equilibrium of Γ1 is (D, D) with payoff (1, 1)
Strategy profile in Γ2: play C in the first round and then D, unlessthe other player did not play as expected, in which case play P
Total-payoff of main outcome: (3, 3)No profitable deviation at the second round, since D is dominatingWhat if a player plays D instead of C at the first round? Then, at thesecond round, he will be punished by P. He would then get at most3− 1 = 2. Not profitable.
Not so easy to compute the sets ET ...
52/62
![Page 177: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/177.jpg)
Nash equilibria in ΓT
If one repeats local Nash equilibria of the one-shot game Γ1, thenthis forms a Nash equilibrium in ΓT
Are there other equilibria?
A variant of the prisoner’s dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
The unique Nash equilibrium of Γ1 is (D, D) with payoff (1, 1)
Strategy profile in Γ2: play C in the first round and then D, unlessthe other player did not play as expected, in which case play P
Total-payoff of main outcome: (3, 3)No profitable deviation at the second round, since D is dominatingWhat if a player plays D instead of C at the first round? Then, at thesecond round, he will be punished by P. He would then get at most3− 1 = 2. Not profitable.
Not so easy to compute the sets ET ...
52/62
![Page 178: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/178.jpg)
Nash equilibria in ΓT
If one repeats local Nash equilibria of the one-shot game Γ1, thenthis forms a Nash equilibrium in ΓT
Are there other equilibria?
A variant of the prisoner’s dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
The unique Nash equilibrium of Γ1 is (D, D) with payoff (1, 1)
Strategy profile in Γ2: play C in the first round and then D, unlessthe other player did not play as expected, in which case play P
Total-payoff of main outcome: (3, 3)No profitable deviation at the second round, since D is dominatingWhat if a player plays D instead of C at the first round? Then, at thesecond round, he will be punished by P. He would then get at most3− 1 = 2. Not profitable.
Not so easy to compute the sets ET ...
52/62
![Page 179: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/179.jpg)
Nash equilibria in ΓT
If one repeats local Nash equilibria of the one-shot game Γ1, thenthis forms a Nash equilibrium in ΓT
Are there other equilibria?
A variant of the prisoner’s dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
The unique Nash equilibrium of Γ1 is (D, D) with payoff (1, 1)
Strategy profile in Γ2: play C in the first round and then D, unlessthe other player did not play as expected, in which case play P
This is a fresh Nash equilibrium (with payoff (1.5, 1.5))!
Not so easy to compute the sets ET ...
52/62
![Page 180: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/180.jpg)
Nash equilibria in ΓT
If one repeats local Nash equilibria of the one-shot game Γ1, thenthis forms a Nash equilibrium in ΓT
Are there other equilibria?
A variant of the prisoner’s dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
The unique Nash equilibrium of Γ1 is (D, D) with payoff (1, 1)
Strategy profile in Γ2: play C in the first round and then D, unlessthe other player did not play as expected, in which case play P
This is a fresh Nash equilibrium (with payoff (1.5, 1.5))!
Not so easy to compute the sets ET ...
52/62
![Page 181: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/181.jpg)
The uniform approach I
[vNeu28] von Neumann. Zur Theorie der Gesellschaftspiele (Mathematische Annalen)
Minmax level of Player A
Let G = (Agt,Σ, (gA)A∈Agt) be a strategic game.The Minmax level of Player A denoted vA is defined by:
vA = minπ−A∈(∆(Σ)Agt\{A})
maxbA∈∆(Σ)
gA(bA, π−A).
vA = smallest payoff that A can ensure against Agt \ {A}, orsmallest payoff that Agt \ {A} can impose to Player A [vNeu28]
It is realized by an element π−A ∈ ∆(Σ)Agt\{A}.
π−A is the punishment strategy of coalition Agt \ {A}
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
vA1 = minβ
maxα
gA1 (α, β) = −5 = vA2
53/62
![Page 182: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/182.jpg)
The uniform approach I
[vNeu28] von Neumann. Zur Theorie der Gesellschaftspiele (Mathematische Annalen)
Minmax level of Player A
Let G = (Agt,Σ, (gA)A∈Agt) be a strategic game.The Minmax level of Player A denoted vA is defined by:
vA = minπ−A∈(∆(Σ)Agt\{A})
maxbA∈∆(Σ)
gA(bA, π−A).
vA = smallest payoff that A can ensure against Agt \ {A}, orsmallest payoff that Agt \ {A} can impose to Player A [vNeu28]
It is realized by an element π−A ∈ ∆(Σ)Agt\{A}.
π−A is the punishment strategy of coalition Agt \ {A}
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
vA1 = minβ
maxα
gA1 (α, β) = −5 = vA2
53/62
![Page 183: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/183.jpg)
The uniform approach I
[vNeu28] von Neumann. Zur Theorie der Gesellschaftspiele (Mathematische Annalen)
Minmax level of Player A
Let G = (Agt,Σ, (gA)A∈Agt) be a strategic game.The Minmax level of Player A denoted vA is defined by:
vA = minπ−A∈(∆(Σ)Agt\{A})
maxbA∈∆(Σ)
gA(bA, π−A).
vA = smallest payoff that A can ensure against Agt \ {A}, orsmallest payoff that Agt \ {A} can impose to Player A [vNeu28]
It is realized by an element π−A ∈ ∆(Σ)Agt\{A}.
π−A is the punishment strategy of coalition Agt \ {A}
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
vA1 = minβ
maxα
gA1 (α, β) = −5 = vA2
53/62
![Page 184: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/184.jpg)
The uniform approach II
The set E
Edef= {(xA)A∈Agt | ∀A ∈ Agt, xA ≥ vA}︸ ︷︷ ︸
Individually rational
∩Conv(g(ΣAgt
))︸ ︷︷ ︸
Feasible
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
-3-5-10
-3
-5
-10
54/62
![Page 185: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/185.jpg)
The uniform approach II
The set E
Edef= {(xA)A∈Agt | ∀A ∈ Agt, xA ≥ vA}︸ ︷︷ ︸
Individually rational
∩Conv(g(ΣAgt
))︸ ︷︷ ︸
Feasible
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
-3-5-10
-3
-5
-10
54/62
![Page 186: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/186.jpg)
The uniform approach II
The set E
Edef= {(xA)A∈Agt | ∀A ∈ Agt, xA ≥ vA}︸ ︷︷ ︸
Individually rational
∩Conv(g(ΣAgt
))︸ ︷︷ ︸
Feasible
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
-3-5-10
-3
-5
-10g(ΣAgt
)54/62
![Page 187: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/187.jpg)
The uniform approach II
The set E
Edef= {(xA)A∈Agt | ∀A ∈ Agt, xA ≥ vA}︸ ︷︷ ︸
Individually rational
∩Conv(g(ΣAgt
))︸ ︷︷ ︸
Feasible
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
-3-5-10
-3
-5
-10Conv(g(ΣAgt
))54/62
![Page 188: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/188.jpg)
The uniform approach II
The set E
Edef= {(xA)A∈Agt | ∀A ∈ Agt, xA ≥ vA}︸ ︷︷ ︸
Individually rational
∩Conv(g(ΣAgt
))︸ ︷︷ ︸
Feasible
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
-3-5-10
-3
-5
-10E = {(xA)A∈Agt | ∀A ∈ Agt, xA ≥ −5} ∩ Conv(g(ΣAgt
))54/62
![Page 189: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/189.jpg)
The uniform approach II
The set E
Edef= {(xA)A∈Agt | ∀A ∈ Agt, xA ≥ vA}︸ ︷︷ ︸
Individually rational
∩Conv(g(ΣAgt
))︸ ︷︷ ︸
Feasible
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
E
-3-5-10
-3
-5
-10
54/62
![Page 190: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/190.jpg)
Results on the uniform approach
[AS76] Aumann,Shapley. Long-term competition – A game theoretic analysis (Essays on Game Theory, 1994)[Rub77] Rubinstein. Equilibrium in supergames (Research Memorandum)
Folk Theorem [AS76,Rub77]
E = E∞
Testing that a strategy profile is a NE in Γ∞ seems very difficult...
... but computing E∞ is simple!
The proofs build “simple” equilibria basedon the concept of punishment.
55/62
![Page 191: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/191.jpg)
Results on the uniform approach
[AS76] Aumann,Shapley. Long-term competition – A game theoretic analysis (Essays on Game Theory, 1994)[Rub77] Rubinstein. Equilibrium in supergames (Research Memorandum)
Folk Theorem [AS76,Rub77]
E = E∞
Testing that a strategy profile is a NE in Γ∞ seems very difficult...
... but computing E∞ is simple!
The proofs build “simple” equilibria basedon the concept of punishment.
55/62
![Page 192: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/192.jpg)
Results on the uniform approach
[AS76] Aumann,Shapley. Long-term competition – A game theoretic analysis (Essays on Game Theory, 1994)[Rub77] Rubinstein. Equilibrium in supergames (Research Memorandum)
Folk Theorem [AS76,Rub77]
E = E∞
Testing that a strategy profile is a NE in Γ∞ seems very difficult...
... but computing E∞ is simple!
The proofs build “simple” equilibria basedon the concept of punishment.
55/62
![Page 193: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/193.jpg)
Results on the uniform approach
[AS76] Aumann,Shapley. Long-term competition – A game theoretic analysis (Essays on Game Theory, 1994)[Rub77] Rubinstein. Equilibrium in supergames (Research Memorandum)
Folk Theorem [AS76,Rub77]
E = E∞
Testing that a strategy profile is a NE in Γ∞ seems very difficult...
... but computing E∞ is simple!
The proofs build “simple” equilibria basedon the concept of punishment.
55/62
![Page 194: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/194.jpg)
Results on the uniform approach
[AS76] Aumann,Shapley. Long-term competition – A game theoretic analysis (Essays on Game Theory, 1994)[Rub77] Rubinstein. Equilibrium in supergames (Research Memorandum)
Folk Theorem [AS76,Rub77]
E = E∞
Testing that a strategy profile is a NE in Γ∞ seems very difficult...
... but computing E∞ is simple!
The proofs build “simple” equilibria basedon the concept of punishment.
55/62
![Page 195: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/195.jpg)
Proof idea of the Folk Theorem
Pick a target payoff vector u = (uA)A∈Agt ∈ E
Let a1a2 · · · ∈(ΣAgt
)Nbe s.t. for every A ∈ Agt,
limT→+∞
1
T
T∑t=1
gA(at) = uA
This is the principal plan (which is a pure profile)
For every A ∈ Agt, let π−A ∈ ∆(Σ) be the punishment strategy
The following profile is a Nash equilibrium with payoff u:
play along u as long as noone deviates
if player A is the first player deviating from this plan, then all playersof Agt \ {A} switch to π−A
56/62
![Page 196: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/196.jpg)
Proof idea of the Folk Theorem
Pick a target payoff vector u = (uA)A∈Agt ∈ E
Let a1a2 · · · ∈(ΣAgt
)Nbe s.t. for every A ∈ Agt,
limT→+∞
1
T
T∑t=1
gA(at) = uA
This is the principal plan (which is a pure profile)
For every A ∈ Agt, let π−A ∈ ∆(Σ) be the punishment strategy
The following profile is a Nash equilibrium with payoff u:
play along u as long as noone deviates
if player A is the first player deviating from this plan, then all playersof Agt \ {A} switch to π−A
56/62
![Page 197: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/197.jpg)
Proof idea of the Folk Theorem
Pick a target payoff vector u = (uA)A∈Agt ∈ E
Let a1a2 · · · ∈(ΣAgt
)Nbe s.t. for every A ∈ Agt,
limT→+∞
1
T
T∑t=1
gA(at) = uA
This is the principal plan (which is a pure profile)
For every A ∈ Agt, let π−A ∈ ∆(Σ) be the punishment strategy
The following profile is a Nash equilibrium with payoff u:
play along u as long as noone deviates
if player A is the first player deviating from this plan, then all playersof Agt \ {A} switch to π−A
56/62
![Page 198: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/198.jpg)
Proof idea of the Folk Theorem
Pick a target payoff vector u = (uA)A∈Agt ∈ E
Let a1a2 · · · ∈(ΣAgt
)Nbe s.t. for every A ∈ Agt,
limT→+∞
1
T
T∑t=1
gA(at) = uA
This is the principal plan (which is a pure profile)
For every A ∈ Agt, let π−A ∈ ∆(Σ) be the punishment strategy
The following profile is a Nash equilibrium with payoff u:
play along u as long as noone deviates
if player A is the first player deviating from this plan, then all playersof Agt \ {A} switch to π−A
56/62
![Page 199: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/199.jpg)
Proof idea of the Folk Theorem
Pick a target payoff vector u = (uA)A∈Agt ∈ E
Let a1a2 · · · ∈(ΣAgt
)Nbe s.t. for every A ∈ Agt,
limT→+∞
1
T
T∑t=1
gA(at) = uA
This is the principal plan (which is a pure profile)
For every A ∈ Agt, let π−A ∈ ∆(Σ) be the punishment strategy
The following profile is a Nash equilibrium with payoff u:
play along u as long as noone deviates
if player A is the first player deviating from this plan, then all playersof Agt \ {A} switch to π−A
56/62
![Page 200: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/200.jpg)
Proof idea of the Folk Theorem
Pick a target payoff vector u = (uA)A∈Agt ∈ E
Let a1a2 · · · ∈(ΣAgt
)Nbe s.t. for every A ∈ Agt,
limT→+∞
1
T
T∑t=1
gA(at) = uA
This is the principal plan (which is a pure profile)
For every A ∈ Agt, let π−A ∈ ∆(Σ) be the punishment strategy
The following profile is a Nash equilibrium with payoff u:
play along u as long as noone deviates
if player A is the first player deviating from this plan, then all playersof Agt \ {A} switch to π−A
56/62
![Page 201: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/201.jpg)
Example: the prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
E = E∞
-3-5-10
-3
-5
-10
57/62
![Page 202: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/202.jpg)
Example: the prisoner dilemma
C D
C (−3,−3) (−10, 0)D (0,−10) (−5,−5)
E = E∞
-3-5-10
-3
-5
-10
E1
57/62
![Page 203: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/203.jpg)
Example: the variant of the prisoner dilemma
C D P
C (2, 2) (0, 3) (−2,−1)D (3, 0) (1, 1) (−1,−1)P (−1,−2) (−1,−1) (−3,−3)
We have that vA1 = vA2 = −1 and E1 = {(1, 1)}.
E1E = E∞
58/62
![Page 204: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/204.jpg)
The compact approach
[Tom06] Tomala. Theorie des jeux : Introduction a la theorie des jeux repetes, chapter “Jeux repetes”[BK87] Benoit, Krishna. Nash equilibria of finitely repeated games (Int. Journal of Game Theory)[Sor86] Sorin. On repeated games with complete information (Math. of Operations Research)
Link between ET and E∞ [BK87]
Given G = (Agt,Σ, (gA)A∈Agt) satisfying some condition (easy to test),a
we have:ET
T→∞−−−−→ E∞
aFor every A ∈ Agt, there is b ∈ E1 s.t. gTA (b) > vA.
Note: The prisoner dilemma does not satisfy the above condition
Link between Eλ and E∞ [Sor86]
Given G = (Agt,Σ, (gA)A∈Agt) satisfying some condition (easy to test),a
we have:Eλ
λ→1−−−→ E∞
aTwo players, or there is x ∈ E∞ s.t. xA > vA for every A ∈ Agt.
Note: The prisoner dilemma satisfies the above condition
59/62
![Page 205: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/205.jpg)
The compact approach
[Tom06] Tomala. Theorie des jeux : Introduction a la theorie des jeux repetes, chapter “Jeux repetes”[BK87] Benoit, Krishna. Nash equilibria of finitely repeated games (Int. Journal of Game Theory)[Sor86] Sorin. On repeated games with complete information (Math. of Operations Research)
Link between ET and E∞ [BK87]
Given G = (Agt,Σ, (gA)A∈Agt) satisfying some condition (easy to test),a
we have:ET
T→∞−−−−→ E∞
aFor every A ∈ Agt, there is b ∈ E1 s.t. gTA (b) > vA.
Note: The prisoner dilemma does not satisfy the above condition
Link between Eλ and E∞ [Sor86]
Given G = (Agt,Σ, (gA)A∈Agt) satisfying some condition (easy to test),a
we have:Eλ
λ→1−−−→ E∞
aTwo players, or there is x ∈ E∞ s.t. xA > vA for every A ∈ Agt.
Note: The prisoner dilemma satisfies the above condition
59/62
![Page 206: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/206.jpg)
Outline
1 What is a game?Games we play for funA broader sense to the notion of game
2 Strategic games – Playing only once simultaneously(Strict) Domination and IterationStability: Nash equilibria
3 Extensive games – Playing several times sequentially
4 Repeated games – Playing the same game again and again
5 Conclusion
60/62
![Page 207: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/207.jpg)
Conclusion
[LL10] Le Treust, Lasaulce. A repeated game formulation of energy-efficient decentralized power control (IEEE Trans. on Wireless Communications).[LeT11] Le Treust. Theorie de l’information, jeux repetes avec observation imparfaite et reseaux de communication decentralises (PhD Thesis)
Content of the tutorialBasic results on strategic games
Extension to extensive games
The special case of repeated games:
includes temporal aspectsincludes notions and mechanisms that will be used in models forverificationhas already interesting applications to the modelling of wirelesscommunications in general, and more specifically to distributedpower control problems [LL10]
61/62
![Page 208: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/208.jpg)
Conclusion
[LL10] Le Treust, Lasaulce. A repeated game formulation of energy-efficient decentralized power control (IEEE Trans. on Wireless Communications).[LeT11] Le Treust. Theorie de l’information, jeux repetes avec observation imparfaite et reseaux de communication decentralises (PhD Thesis)
Content of the tutorialBasic results on strategic games
Extension to extensive games
The special case of repeated games:
includes temporal aspectsincludes notions and mechanisms that will be used in models forverificationhas already interesting applications to the modelling of wirelesscommunications in general, and more specifically to distributedpower control problems [LL10]
61/62
![Page 209: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/209.jpg)
Conclusion
[LL10] Le Treust, Lasaulce. A repeated game formulation of energy-efficient decentralized power control (IEEE Trans. on Wireless Communications).[LeT11] Le Treust. Theorie de l’information, jeux repetes avec observation imparfaite et reseaux de communication decentralises (PhD Thesis)
Content of the tutorialBasic results on strategic games
Extension to extensive games
The special case of repeated games:
includes temporal aspectsincludes notions and mechanisms that will be used in models forverificationhas already interesting applications to the modelling of wirelesscommunications in general, and more specifically to distributedpower control problems [LL10]
61/62
![Page 210: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/210.jpg)
Conclusion
[LL10] Le Treust, Lasaulce. A repeated game formulation of energy-efficient decentralized power control (IEEE Trans. on Wireless Communications).[LeT11] Le Treust. Theorie de l’information, jeux repetes avec observation imparfaite et reseaux de communication decentralises (PhD Thesis)
Content of the tutorialBasic results on strategic games
Extension to extensive games
The special case of repeated games:
includes temporal aspectsincludes notions and mechanisms that will be used in models forverificationhas already interesting applications to the modelling of wirelesscommunications in general, and more specifically to distributedpower control problems [LL10]
61/62
![Page 211: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/211.jpg)
Conclusion
[LL10] Le Treust, Lasaulce. A repeated game formulation of energy-efficient decentralized power control (IEEE Trans. on Wireless Communications).[LeT11] Le Treust. Theorie de l’information, jeux repetes avec observation imparfaite et reseaux de communication decentralises (PhD Thesis)
Content of the tutorialBasic results on strategic games
Extension to extensive games
The special case of repeated games:
includes temporal aspectsincludes notions and mechanisms that will be used in models forverificationhas already interesting applications to the modelling of wirelesscommunications in general, and more specifically to distributedpower control problems [LL10]
61/62
![Page 212: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/212.jpg)
Conclusion
[LL10] Le Treust, Lasaulce. A repeated game formulation of energy-efficient decentralized power control (IEEE Trans. on Wireless Communications).[LeT11] Le Treust. Theorie de l’information, jeux repetes avec observation imparfaite et reseaux de communication decentralises (PhD Thesis)
Content of the tutorialBasic results on strategic games
Extension to extensive games
The special case of repeated games:
includes temporal aspectsincludes notions and mechanisms that will be used in models forverificationhas already interesting applications to the modelling of wirelesscommunications in general, and more specifically to distributedpower control problems [LL10]
61/62
![Page 213: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/213.jpg)
What’s next?
Talk on Thursday!
Why game theory for verification?
Which games? How can we treat them?
Discussion
62/62
![Page 214: Veri cation and Game Theory - LSVbouyer/files/tuto-atva19.pdf · A broader sense: What is game theory? [MSZ13]Maschler, Solan, Zamir. Game theory (Cambridge University Press) Goal:](https://reader034.vdocuments.mx/reader034/viewer/2022042116/5e940f580deacd40d104987f/html5/thumbnails/214.jpg)
What’s next?
Talk on Thursday!
Why game theory for verification?
Which games? How can we treat them?
Discussion
62/62