Sequential selection of random vectors under a sum constraint

From MaRDI portal





An i.i.d. sequence of coordinatewise nonnegative \(d\)-dimensional random vectors \(X_1,X_2,\dots, X_n\) is sequentially observed. When a certain random vector is observed it can be either selected or rejected but once made this decision is final. In each coordinate the sum of the selected vectors must not exceed one. The problem is to find a selection policy maximizing the expected number of selected vectors. Let \(\text{Opt}_n(\mu)\) denote the maximum (depending on the distribution \(\mu\) of \(X_1\)) and let \(\text{Proph}_n(\mu)\) denote the maximum a ``prophet (knowing the whole sequence in advance) can achieve. Clearly \(\text{Opt}_n(\mu)< \text{Proph}_n(\mu)\). The author studies the case of an absolutely continuous distribution \(\mu\). He obtains a selection policy (for which the expected number \(E_n(\mu)\) is achieved) being asymptotically optimal in the sense that \[ E_n(\mu)\sim \text{Opt}_n(\mu)\sim \text{Proph}_n(\mu)\sim n\mu(\Delta)\qquad (n\to\infty). \] Here, \(\mu(\Delta)\) denotes a positive constant depending on \(\mu\).











This page was built for publication: Sequential selection of random vectors under a sum constraint

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4819442)