Additive comparisons of stopping values and supremum values for finite stage multiparameter stochastic processes (Q549791): Difference between revisions

The author studies an optimal stopping problem for discrete time multiparameter processes with index set \(\mathbb{N}^d\). He studies ``additive prophet inequalities'' for processes \(\{X(z): z\in I\}\) with an index set \(I\) of the form \(I= \{z\in\mathbb{N}^d: z\leq t\}\), where \(t\in\mathbb{N}^d\) and the semi-order is the natural one. Assume that the random variables \(X(z)\) are independent and take values in \([0,1]\). The arm is to find the best possible bound \(\delta\) such that \[ D= E\Biggl(\sup_{z\in I} X(z)\Biggr)- V(\{X(z): z\in I\})\leq\delta. \] Here, \(V\) is the ``value'' of the process for a player without foresight. This player can only try to attain a maximal reward using ``tactics'': In each moment \(s\leq t\), he can only make use of the information given by \(\{X(z): z\leq s\}\) to decide if he wants to stop and get \(X(s)\), or to which of the direct sucessors of \(s\) he wants to proceed. For dimensional \(d=1\), \textit{T. P. Hill} and \textit{R. P. Kertz} [A survey of prophet inequalities in optimal stopping theory. Strategies for sequential search and selection in real time, Proc. Conf., Amherst/ MA (USA) 1990, Contemp. Math. 125, 191--207 (1992; Zbl 0794.60040)] determined \(\delta={1\over 4}\). Here, the author shows that the optimal \(\delta\) is determined by a nonlinear programming problem.