Abstract: This talk investigates online demand fulfillment in networks with limited flexibility, arbitrary resources, and diverse request types, where irrevocable fulfillment decisions must be made in real-time. Integrating network revenue management (NRM) with re-optimization techniques, we first establish a fundamental connection between the non-degeneracy of the deterministic linear program (DLP) and the complete resource pooling (CRP) condition. Under this framework, we demonstrate that a probabilistic fulfillment policy with a re-solving heuristic, alongside a load deviation policy, can achieve a uniformly constant performance bound independent of the time horizon and resource capacities.

We then extend this analysis to highly asymmetric markets, where massive disparities in market shares render traditional performance metrics, such as the generalized chaining gap (GCG), less effective. To address this, we introduce a novel network condition termed -connectivity and prove it is both necessary and sufficient for maintaining bounded performance in asymmetric environments. Crucially, our findings reveal that classic structures like the 'long chain' may fail when low-demand markets are dispersed. Instead, we propose a 'hub-and-spoke' configuration embedded with a shorter long chain as an optimal network design strategy to mitigate -disconnectivity and ensure robust system performance.

主讲人简介：徐禛，利物浦大学管理学院讲师。他的研究处于运筹学、应用概率和机器学习的交叉领域，重点关注随机过程控制、动态规划以及针对各种静态和动态模型的数据驱动决策。他的研究成果已发表在顶级学术期刊上。包括Management Science、Manufacturing & Service Operations Management(MSOM）、Production and Operations Management(POM）以及Mathematics of Operations Research。