Inventory planning in a deterministic environment: Continuous time model with concave costs (Q795707)

The authors study an inventory model where the instantaneous demand, 0(t), is known for 0\(\leq t\leq T\). To balance demand there is instantaneous replenishment of goods. Any ordering policy, v(t), must satisfy two conditions: (i) m(t)\(\leq v(t)\leq M(t)\), 0\(\leq t\leq T\); and (ii) \(y(t)=y_ 0+\int^{t}_{0}(v(s)-0(s))ds\geq 0\), 0\(\leq t\leq T\), where \(y_ 0=y(0)\) is the inventory on hand before any ordering is done. An optimal ordering policy, \(v^*\), is a policy satisfying these constraints which minimizes (1) \(\int^{T}_{0}c(t,v(t))+f(t,y(t))dt.\) The c and f represent (respectively) the cost of ordering v(t) amount of goods at time t and the inventory cost of having y(t) amount of goods at time t. Both functions are assumed concave and non-decreasing in v(t) for fixed t and integrable in t for fixed v and y. If \(v(t)=m(t)\) satisfies (ii), it clearly is optimal. The authors study the structure of \(v^*\) when this is not the case. The major result is: If c or f is strictly convex almost everywhere, then there exists a \(v^*\) which satisfies (2) \(y^*(t)(M(t)-v^*(t))(v^*(t)-m(t))=0\) a.e., 0\(\leq t\leq T\), i.e. for almost all t either \(v^*(t)=m(t)\), \(v^*(t)=M(t)\) or \(y^*(t)=0\) (the inventory level is zero). For general f and c, the authors show that the minimal value for (1) can be approximated within \(\epsilon\) by an ordering policy satisfying (2). One should note, however, that many of the proofs are somewhat sketchy, with parts left unproven which the authors seem to consider self-evident, but which would appear to require some explanation.