On the optimality of threshold control in queues with model uncertainty (Q975793)

The paper studies a single-server queueing system in which arrival and departure processes are governed by simple point processes \(A_t\) and \(B_t\), respectively, i.e. \[ X_t=x_0+A_t-D_t, \] where \(x_0\) is an initial state and \(X_t\) is the state of the system at time \(t\). Let \(\mathcal{F}_t\) be the sigma-field generated by \(X_t\), and let \(A_t\) and \(D_t\) admit \(\mathcal{F}_t\)-predictable intensities \(\alpha_t\) and \(\beta_t\). The intensities \(\alpha_t\) and \(\beta_t\) satisfy the following capacity constraints: \[ 0\leq\alpha_t\leq z, \] \[ 0\leq\alpha_t\leq y \] for all values \(t\geq0\). Let \(\delta\) denote the discount factor, let \(\tilde{\rho}\) be the revenue obtained by selling one unit of output, let \(\tilde{c}\) be the cost of acquiring one unit of input, and let \(h\) be the unit holding cost. The problem is to find a control \(\{\beta_t,\alpha_t, t\geq0\}\) maximizing the discount value function \[ V(x_0,u)=\mathsf{E}_{x_0}\int_0^\infty\mathrm{e}^{-\delta t}(\tilde{\rho}\mathrm{d}D_t-\tilde{c}\mathrm{d}A_t-hX_t\mathrm{d}t), \] where \(\mathsf{E}_{x_0}\) denotes the conditional expectation given \(X_0=x_0\). The author prove that the threshold policy is optimal under the max-min robust model introduced in the paper by the authors, where uncertainty in the process is characterized by relative entropy. The model is a generalization of the standard concept of relative entropy to account for different levels of model uncertainty in the arrival and departure processes. The impact of uncertainty levels on the optimal threshold control is studied as well.