Price of stability

From Wikipedia, the free encyclopedia
Jump to: navigation, search

In game theory, the price of stability (PoS) of a game is the ratio between the best objective function value of one of its equilibria and that of an optimal outcome. The PoS is relevant for games in which there is some objective authority that can influence the players a bit, and maybe help them converge to a good Nash equilibrium. When measuring how efficient a Nash equilibrium is in a specific game we often time also talk about the price of anarchy (PoA).

Examples[edit]

Another way of expressing PoS is:

 \text{PoS} = \frac {\text{value of best Nash equilibrium}} {\text{value of optimal solution}},\  \text{PoS} \geq 0.

In the following prisoner’s dilemma game, since there is a single equilibrium (B, R) we have PoS = PoA = 1/2.

Prisoner's Dilemma
Left Right
Top (2,2) (0,3)
Bottom (3,0) (1,1)

On this example which is a version of the battle of sexes game, there are two equilibrium points, (T, L) and (B, R), with values 3 and 15, respectively. The optimal value is 15. Thus, PoS = 1 while PoA = 1/5.

Left Right
Top (2,1) (0,0)
Bottom (0,0) (5,10)

Background and milestones[edit]

The price of stability was first studied by A. Schulzan and N. Moses and was so-called in the studies of E. Anshelevich. They showed that a pure strategy Nash equilibrium always exists and the price of stability of this game is at most the nth harmonic number in directed graphs. For undirected graphs Anshelevich and others presented a tight bound on the price of stability of 4/3 for a single source and two players case. Jian Li has proved that for undirected graphs with a distinguished destination to which all players must connect the price of stability of the Shapely network design game is O(\log n/\log\log n) where n is the number of players. On the other hand, the price of anarchy is about n in this game.

Network design games[edit]

Setup[edit]

Network design games have a very natural motivation for the Price of Stability. In these games, the Price of Anarchy can be much worse than the Price of Stability.

Consider the following game.

  • n players;
  • Each player i aims to connect s_i to t_i on a directed graph G = (V, E);
  • The strategies P_i for a player are all paths from s_i to t_i in G;
  • Each edge has a cost c_i;
  • 'Fair cost allocation': When n_e players choose edge e, the cost \textstyle d_e(n_e) = \frac{c_e}{n_e} is split equally among them;
  • The player cost is \textstyle C_i(S) = \sum_{e \in P_i} \frac{c_e}{n_e}
  • The social cost is the sum of the player costs: \textstyle SC(S) = \sum_i C_i(S) = \sum_{e \in S} n_e \frac{c_e}{n_e} = \sum_{e \in S} c_e
.
A network design game with \Omega(n) Price of Anarchy

Price of anarchy[edit]

The price of anarchy can be \Omega(n). Consider the following network design game.

Pathological Price of Stability game

Consider two different equilibria in this game. If everyone shares the 1 + \varepsilon edge, the social cost is 1 + \varepsilon. This equilibrium is indeed optimal. Note, however, that everyone sharing the n edge is a Nash equilibrium as well. Each agent has cost 1 at equilibrium, and switching to the other edge raises his cost to 1+\varepsilon.

Lower bound on price of stability[edit]

Here is a pathological game in the same spirit for the Price of Stability, instead. Consider n players, each originating from s_i and trying to connect to t. The cost of unlabeled edges is taken to be 0.

The optimal strategy is for everyone to share the 1+\varepsilon edge, yielding total social cost 1+ \varepsilon. However, there is a unique Nash for this game. Note that when at the optimum, each player is paying \textstyle \frac{1 + \varepsilon}{n}, and player 1 can decrease his cost by switching to the \textstyle \frac{1}{n}. edge. Once this has happened, it will be in player 2's interest to switch to the \textstyle \frac{1}{n-1} edge, and so on. Eventually, the agents will reach the Nash equilibrium of paying for their own edge. This allocation has social cost \textstyle 1 + \frac{1}{2} + \cdots + \frac{1}{n} = H_n, where H_n is the nth harmonic number, which is \Theta(\log n). Even though it is unbounded, the price of anarchy is exponentially better than the price of anarchy in this game.

Upper bound on price of stability[edit]

Note that by design, network design games are congestion games. Therefore, they admit a potential function \textstyle \Phi = \sum_e \sum_{i=1}^{n_e} \frac{c_e}{i}.

Theorem. Suppose there exist constants A and B such that for every strategy S,

 A \cdot SC(S) \leq \Phi(S) \leq B \cdot SC(S).

Then the price of stability is less than B/A

Proof. The global minimum NE of \Phi is a Nash equilibrium, so

 SC(NE) \leq 1/A \cdot \Phi(NE) \leq 1/A \cdot \Phi(OPT) \leq B/A \cdot SC(OPT).

Now recall that the social cost was defined as the sum of costs over edges, so

 \Phi(S) = \sum_{e \in S} \sum_{i=1}^{n_e} \frac{c_e}{i} = 
\sum_{e \in S} c_e H_{n_e} \leq \sum_{e \in S} c_e H_n = H_n \cdot SC(S).

We trivially have A = 1, and the computation above gives B = H_n, so we may invoke the theorem for an upper bound on the price of stability.

References[edit]

  1. Algorithmic Game Theory by N. Nisan, T. Roughgarden, E. Tardos, and V. Vazirani (eds), Cambridge University Press, 2007. ISBN 0521872820
  2. L. Agussurja and H. C. Lau. The Price of Stability in Selfish Scheduling Games. Web Intelligence and Agent Systems: An International Journal, 9:4, 2009.
  3. Jian Li. An O(\log n/\log\log n) upper bound on the price of stability for undirected Shapely network design games. Information Processing Letters 109 (15), 876-878, 2009.