Id: arxiv/2608.11808 Url: https://arxiv.org/abs/2608.11808 Tags: math.OC Score: 0 Title: Data-Driven optimal control via Koopman operators and Hamilton-Jacobi-Bellman equations Domain: math.OC Source: arxiv Authors: Guuyuan Chen, Gaosheng Zhu Reviews: 0 Subject: math.OC Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-08-12 Page: https://delphi.science/r/arxiv/2608.11808 This paper presents a data-driven stable manifold (DD-SM) method, which integrates Koopman operator representation learning with the geometric stable manifold approach to Hamilton-Jacobi-Bellman (HJB) equations, enabling end-to-end optimal feedback control synthesis from raw trajectory data without prior knowledge of system dynamics. We construct an augmented control system under a unified symmetric subspace decomposition (SSD) and extended dynamic mode decomposition (EDMD) framework for joint approximation of the drift field, control matrix and their spatial derivatives, and derive…