Title:Worst-Case Analysis of Decoupled Policies for Multi-Location Inventory Control Problems

Abstract:The difference in performance between centralized and decentralized control strategies crucially informs design choices in real-world control systems. Although computing and executing centralized control algorithms is often more costly than decentralized methods, their performance enhancements may far outweigh these costs. In this work, we study the value of centralization within the context of the well-known inventory control problem, where a planner seeks to identify optimal inventory levels that meet stochastic demand while minimizing ordering costs, holding costs, and shortage costs. We consider multilocation systems in which the inventories are coupled through a single ordering channel and the associated ordering cost function belongs to one of two classes of nonlinear cost functions that often arise in practical settings. For each of these classes, we derive constant-factor competitive ratios between the optimal coupled and decoupled policies and show they are almost tight. We then demonstrate that online algorithms also achieve tight competitive ratios for this problem. We conclude with numerical simulations that validate these results.