arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2303.07544math.OC
收藏

线性二次平均场型 Stackelberg 随机微分博弈的闭环可解性

Closed-Loop Solvability of Linear Quadratic Mean-Field Type Stackelberg Stochastic Differential Games

Zixuan Li, Jingtao Shi

AI总结:

针对有限时域线性平均场型随机系统的二次 Stackelberg 博弈,本文利用耦合 Riccati 方程与平均场型倒向随机微分方程刻画跟随者和领导者的闭环最优策略,并讨论可解性及值函数表示。

摘要:

本文研究有限时间区间内,具有平均场型二次代价泛函的线性平均场型随机微分系统的 Stackelberg 随机微分博弈。状态方程中的系数和代价泛函中的权重矩阵均为确定性的。文中引入了要求独立于初始状态的闭环 Stackelberg 均衡策略。首先求解跟随者问题,这是一个随机线性二次最优控制问题。通过将原问题转化为一个最优控制已知的新问题,跟随者的闭环最优策略由两个耦合 Riccati 方程以及一个线性平均场型倒向随机微分方程刻画。随后,领导者转而求解一个平均场型正倒向随机微分方程的随机线性二次最优控制问题。领导者闭环最优策略存在的必要条件由两个耦合 Riccati 方程和一个线性平均场型倒向随机微分方程的存在性给出。在状态方程扩散项不包含跟随者控制过程的情形下,讨论了领导者优化问题的 Riccati 方程的可解性。此外,领导者的值函数通过两个倒向随机微分方程和两个 Lyapunov 方程表示。

英文摘要:

This paper is devoted to a Stackelberg stochastic differential game for a linear mean-field type stochastic differential system with a mean-field type quadratic cost functional in finite horizon. The coefficients in the state equation and weighting matrices in the cost functional are all deterministic. Closed-loop Stackelberg equilibrium strategies are introduced which require to be independent of initial states. Follower's problem is solved firstly, which is a stochastic linear quadratic optimal control problem. By converting the original problem into a new one whose optimal control is known, the closed-loop optimal strategy of the follower is characterized by two coupled Riccati equations as well as a linear mean-field type backward stochastic differential equation. Then the leader turns to solve a stochastic linear quadratic optimal control problem for a mean-field type forward-backward stochastic differential equation. Necessary conditions for the existence of closed-loop optimal strategies for the leader is given by the existence of two coupled Riccati equations with a linear mean-field type backward stochastic differential equation. The solvability of Riccati equations of leader's optimization problem is discussed in the case where the diffusion term of the state equation does not contain the control process of the follower. Moreover, leader's value function is expressed via two backward stochastic differential equations and two Lyapunov equations.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑