京师数学前沿论坛 第三十讲
京师数学前沿论坛
报告题目(Title)：Online Trading: Secretary vs Prophet
报告人(Speaker)：陈旭瑾 研究员（中国科学院数学与系统科学研究院）
地点(Place)：后主楼1124
时间(Time)：2026年4月27日15:00-16:00
报告摘要
This talk explores an online trading model where an intermediary facilitates transactions between sequentially arriving sellers and buyers. In this setting, each seller holds a single item, and each buyer demands exactly one. The agents arrive in a uniformly random order and reveal their valuations (prices) only upon arrival. Without knowledge of future arrivals, the intermediary must make immediate, irrevocable decisions to either buy from a seller or sell to a buyer.
We investigate two distinct variants of this problem. First, we consider the secretary version, where the objective is to maximize social welfare without any prior knowledge of the underlying price distributions. In the single-seller setting, we establish a tight competitive ratio of 4e2e2+1≈3.523, strictly improving the previous state-of-the-art bound of 4.189 [Chen et al., DAM 2024].
Second, we study the prophet version, where the intermediary seeks to maximize their gain-from-trade, assuming prices are drawn from known i.i.d. distributions. For the single-seller case, we propose both adaptive and non-adaptive threshold algorithms that guarantee competitive ratios of at most 2. As the number of buyers grows large, these ratios asymptotically approach 1.58 and 1.76, respectively. Furthermore, for the balanced setting with n sellers and n buyers, we obtain a competitive ratio of 2n2−nn+4−n−1)(n+1, which is comparable to the previous bound of 2 established in a related trading-prophet setting [Correa et al., EC 2023].
(Joint work with Xiaodong Hu, Changjun Wang, and Qingjie Ye)
主讲人简介
陈旭瑾，2004年获香港大学博士学位，现为中国科学院数学与系统科学研究院研究员、中国科学院大学教授。主要研究兴趣是组合优化的理论和算法，包括网络优化、多面体组合、算法博弈论等。曾获中国青年科技奖、中国运筹学会青年科技奖、国家优秀青年基金。入选国家中青年科技创新领军人才计划。
