Uniqueness of optimal strategies in Captain Lotto games (Q1742138): Difference between revisions

The author studies a continuous \textit{Captain Lotto game} \(\Lambda_{c_A,c_B}(a,b)\), being some specific two-person zero-sum game with the following structure: There are two players \(A\) and \(B\). The game is described by four real numbers \(a, c_A, b\) and \(c_B\) such that \(0\leq a\leq c_A\) and \(0\leq b\leq c_B\) (\(\infty\) is possible for \(c_A\) and \(c_B\)). Player \(A\) chooses (a distribution of) a nonnegative random variable \(X\) with expectation \(E(X)=a\) and values bounded from above by \(c_A\), \(0\leq X\leq c_A\). Simultaneously Player \(B\) chooses (a distribution of) a nonnegative random variable \(Y\) with expectation \(E(Y)=b\) and values bounded from above by \(c_B\), \(0\leq Y\leq c_B\). The choices of \(X\) and \(Y\) are independent. The payoff function (from \(B\) to \(A\)) is defined by \(H(X,Y) = P(X>Y) - P(X<Y)\) where \(P\) denotes probability. The results of this paper are closely to the ones of \textit{S. Hart} in [Int. J. Game Theory 45, No. 1--2, 37--61 (2016; Zbl 1388.91004)] where he found optimal and \(\varepsilon\)-optimal strategies for both players in the Captain Lotto game \(\Lambda_{c_A,c_B}(a,b)\), in all cases of parameters \(c_A, c_B, a\) and \(b\). The author of the present paper shows for the game \(\Lambda_{c_A,c_B}(a,b)\) that \begin{itemize} \item[(1)] the optimal strategies for both players found by Hart are unique; \item[(2)] in some cases there is no optimal strategies for players \(A\) or \(B\); \item[(3)] any \(\varepsilon\)-optimal strategy for player \(A\) (when \(\varepsilon \rightarrow 0\)) can be fully characterized by its limit points. \end{itemize}