.
Hardness Beyond 2 2 Games
Game Theory Course:
Jackson, Leyton-Brown & Shoham
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
Algorithms
Two example algorithms for finding NE
LCP (Linear Complementarity) formulation
[Lemke-Howson 64]
Support Enumeration Method
[Porter et al. 04]
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
Early History
1928 von Neumann: existence of Equilibrium in 2-player,
zero-sum games
proof uses Brouwers fixed point theorem;
led directly to algorithms:
Danzig 57: equivalent to LP duality
Khachiyan79: polynomial-time solvable
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
Early History
1928 von Neumann: existence of Equilibrium in 2-player,
zero-sum games
proof uses Brouwers fixed point theorem;
led directly to algorithms:
Danzig 57: equivalent to LP duality
Khachiyan79: polynomial-time solvable
1950 Nash: existence of Equilibrium in multiplayer,
general-sum games
proof also uses Brouwers fixed point theorem;
intense effort on equilibrium algorithms:
Kuhn 61, Mangasarian 64, Lemke-Howson 64, Rosenmller 71,
Wilson 71, Scarf 67, Eaves 72, Laan-Talman 79, Porter et al. 04,
all exponential in the worst case
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
The Lemke-Howson Algorithm
LCP (Linear Complementarity) formulation
u1 (aj1 , ak2 ) sk2 + r1j = U1
j A1
u2 (aj1 , ak2 ) sj1 + r2k = U2
k A2
kA2
jA1
sj1 = 1,
jA1
sj1
r1j
r1j
0,
sk2 = 1
kA2
j A1 , k A2
j A1 , k A2
sj1 = 0, r2j sj2 = 0
j A1 , k A2
0,
sk2
r2k
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
Support Enumeration Method: Porter et al. 2004
Step 1: Finding a NE with a specific support
p(ai )ui (ai , ai ) = vi
i {1, 2}, ai i
p(ai )ui (ai , ai ) vi
i {1, 2}, ai
/ i
a1 i
a1 i
pi (ai ) 0
pi (ai ) = 0
pi (ai ) = 1
i {1, 2}, ai i
i {1, 2}, ai
/ i
i {1, 2}
ai i
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
Support Enumeration Method: Porter et al. 2004
Step 2: Smart heuristic search through all sets of support
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
From Algorithms to Complexity Analysis
These algorithms have exponential worst-case time complexity.
So do all known others.
Can we do better?
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
From Algorithms to Complexity Analysis
Reminder of a (small part) of the complexity hierarchy.
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
From Algorithms to Complexity Analysis
So, is it NP-complete to find a Nash equilibrium?
Strictly speaking, no, since a solution is guaranteed to exist
However, it is NP-complete to find a tiny bit more info than a
Nash equilibrium; e.g., the following are NP-complete:
1. (Uniqueness) Given a game G, does there exist a unique equilibrium in G?
2. (Pareto optimality) Given a game G, does there exist a strictly Pareto efficient
equilibrium in G?
3. (Guaranteed payoff) Given a game G and a value v, does there exist an equilibrium in G
in which some player i obtains an expected payoff of at least v?
4. (Guaranteed social welfare) Given a game G, does there exist an equilibrium in which
the sum of agents utilities is at least k?
5. (Action inclusion) Given a game G and an action ai Ai for some player i N , does
there exist an equilibrium of G in which player i plays action ai with strictly positive
probability?
6. (Action exclusion) Given a game G and an action ai Ai for some player i N , does
there exist an equilibrium of G in which player i plays action ai with zero probability?
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
From Algorithms to Complexity Analysis
Still, finding even a single Nash equilibrium seems hard;
how do we capture that?
Enter PPAD (Polynomial Parity Arguments on Directed
graphs)
item (Papadimitriou 94)
At a high level:
FNP problems are constructive versions of NP problems (F stands
for Functional)
TFNP is a subclass of FNP for problems for which a solution is
guaranteed to exist (T stands for Total)
PPAD is a subclass of TFNP where the proofs are based on parity
arguments in directed graphs
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
From Algorithms to Complexity Analysis
Where is PPAD?
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
From Algorithms to Complexity Analysis
Where is PPAD?
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
Hardness beyond 2 2 games
From Algorithms to Complexity Analysis
The Complexity of the Nash Equilibrium
Theorem: Computing a Nash equilibrium is PPAD-complete
for games with 4 players;
[Daskalakis, Goldberg, Papadimitriou 05]
for games with 3 players; [Chen, Deng 05] &
[Daskalakis, Papadimitriou 05]
for games with 2 players. [Chen, Deng 06]
Game Theory Course: Jackson, Leyton-Brown & Shoham
Hardness Beyond 2 2 Games
