On a deception game with three boxes (Q687056)

This paper considers the following two-person zero-sum game \(\Gamma_ n\) where \(n\geq 2\) is an integer. Players 1 and 2 are both informed that \(n\) numbers have been drawn, each uniformly and independently from the interval \([-1,1]\). Only player 2 is further informed the actual outcome of the numbers. These numbers are then dropped into \(n\) numbered boxes, one in each box. Player 2 closes the boxes, and labels the lids. At least \(n-1\) of the lid-labels must equal to the contained numbers, but one of the lids may be labelled arbitrarily with a number from \([-1,1]\). Player 1 then looks at the labels, chooses a box, and receives from player 2 an amount equal to the selected box's actual contents. The interesting feature here concerns the usefulness of certain information given a piece of it has been manipulated for deception. Obviously, by ignoring the labels completely, player 1 can guarantee himself an expected payoff of 0 by choosing any fixed box. Thus \(v_ n\geq 0\), where \(v_ n\) is the value of \(\Gamma_ n\). The question whether \(v_ n=0\) is tantamount to asking whether the falsification of one out of \(n\) pieces of information is enough to render the information completely useless to him. It is easy to see \(v_ 2=0\), and to show \(v_ n>0\) for \(n\geq 4\). So falsifying one out of two pieces of information can make the information useless. When there are four or more pieces of information, falsifying only one cannot make the information useless. What is then the situation for precisely three pieces of information? \textit{Mark Thompson} discussed but not solved the question whether \(v_ 3=0\) in his Harvard undergraduate thesis in 1970. Subsequently, \textit{J. Spencer} [Am. Math. Monthly 80, 416-417 (1973; Zbl 0268.90068)] posed this question explicitly as an open problem. The present paper constructs two strategies for player 2 that can restrict the expected payoff obtained by player 1 to at most zero, thus implying \(v_ 3=0\).