The refined best-response correspondence in normal form games (Q2259419)

Let \(\Gamma=(I,S,u)\) be a finite \(n\)-player normal form game, where \(I=\{1,\dots,n\}\) is the set of players, \(S=S_1\times\dots\times S_n\) is the set of pure strategy profiles, and \(u:S\rightarrow \mathbb{R}^n\) is the payoff function. Let \(\Delta (S_i)\) denote the set of player \(i\)'s mixed strategies, and \(\Theta =\Delta (S_1)\times\dots\times\Delta (S_n)\). Further, for \(x\in \Theta\) let \(B_i(x)\subset S_i\) denote the set of pure strategy best responses to \(x\) for player \(i\), and \(B(x)= B_1(x)\times\dots\times B_n(x)\). The authors restrict the attention only to the set \({\mathcal G}^*\) of such games \(\Gamma\) in which the complement of the set of mixed-strategy profiles \(\Psi= \{x\in \Theta:\;B(x)\) is a singleton\(\}\) has Lebesgue measure \(0\). The basic notion in the paper is the set \(S_i(x)\) of pure \textit{refined best-responses} for the player \(i\in I\), when the strategy profile is \(x\in \Theta\). Its definition is the following: For \(x\in \Theta\), let \(\Omega (x)\) be the set of all sequences \(\bar{y}=(y_t)_{t=1}^\infty\) in \(\Psi\) convergent to \(x\). Then \(S_i(x)=\{s_i\in S_i:\) for some \(\bar{y}\in \Omega(x)\) \(B_i(y_t)=\{s_i\},\; t=1,2,\dots\}\). Now the \textit{refined best-response correspondence} (RBRC) is defined as \(\sigma:\Theta\rightarrow \Theta\) with \(\sigma (x)=(\sigma_1(x), \dots,\sigma_n(x))\), where \(\sigma_i(x)=\Delta(S_i(x))\). The authors give an example showing that RBRC can be very different from the standard \textit{best-response correspondence} (BRC). ``In two-player games, however, the RBRC of a given game is the same as the RBC of a slightly modified game. The modified game is derived from the original game by reducing the payoff by a small amount for all pure strategies that are \textit{weekly inferior}. Weekly inferior strategies, for two-player games, are either weekly dominated or are equivalent to a proper mixture of pure strategies. Fixed points of RBRC are not equivalent to any known Nash equilibrium refinement. A class of simple communication games demonstrates the usefulness and intuitive appeal of RBRC.'' (from the authors' abstract)