User:IssaRice/Minus notation in game theory

From Machinelearning
Jump to: navigation, search

In game theory, s_{-i} is a shorthand for (s_1,\ldots,s_{i-1},s_{i+1},\ldots,s_n) \in S_1 \times \cdots \times S_{i-1} \times S_{i+1} \times \cdots \times S_n. In other words, s_{-i} is a tuple of strategies of all the players other than player i.

One then writes things like v_i(s_i, s_{-i}) for the payoff of the strategy profile where each player i chooses s_i.

But to my mind, there is a type error here, because v_i \colon S_1 \times \cdots \times S_n \to \mathbf R is supposed to take the strategies in order, whereas in the previous paragraph, for each player, their own strategy appears first. To give an example, in the case of n=2, each v_i should take (s_1,s_2) \in S_1\times S_2, but the notation (s_i, s_{-i}) for player 2 is (s_2,s_1), which is reversed. Type error!

I think usually this is not a problem, because the reader can mentally put the s_i in the right spot, or one can assume that no two strategies (even across different players) are the same (e.g. the strategies can be "marked" with numbers to be distinguishable),[1] so that one can pass a set of strategies like \{s_1, \ldots ,s_n\} and there would be no confusion even if the ordering is different.

Here are two more ideas for dealing with this notation:

  • One can change each v_i to have domain S_i\times S_1 \times \cdots \times S_{i-1} \times S_{i+1} \times \cdots \times S_n. This is sort of inelegant because now one cannot pass s \in S_1\times \cdots \times S_n to an arbitrary v_i. (One can only pass it to v_1.)
  • One can define a function g_i for each player that is supposed to "put s_i back in the right spot". Formally, g_i\colon S_i \times (S_1 \times \cdots \times S_{i-1} \times S_{i+1} \times \cdots \times S_n) \to S_1 \times \cdots \times S_n is defined by g(x,y) = (y_{\leq i-1}, x, y_{\geq i+1}). Now one can write v_i(g_i(s_i, s_{-i})) (if the ith strategy needs to be put in the correct place) and v_i(s) (if the strategies are already in order). Note that the altered v_i suggested in the previous bullet is, in this notation, v_i \circ g_i.


  1. To be even more precise, given finite sets of strategies S_1,\ldots,S_n, one can start with S_1 and replace each strategy s_i with (i, s_i), to number them. Then for S_2, one starts the numbering where S_1 left off, and so on.