On the relationship between the Pontryagin maximum principle and the Hamilton-Jacobi-Bellman equation in optimal control problems for fractional-order systems (Q6147956): Difference between revisions

The dynamics of the system are given by \[ (^C \! D^\alpha x)(\tau) = f(\tau, x(\tau), u(\tau)),\quad x(0) = x_0 \] where \(\tau \in[0, T]\) is time, \(x(\tau) \in \mathbb{R}^n\) is the state, \(u(\tau) \in U \subset \mathbb{R}^m\) is the control and \((^C \! D^\alpha x)(\tau)\) is the left Caputo fractional derivative. \[ (^C \! D^\alpha x)(\tau) = \frac{1}{\Gamma(1 - \alpha)} \int_0^\tau \frac{x(\xi) - x(0)}{(\tau - \xi)^\alpha}d\xi. \] The objective is to minimize \(J = \sigma (x(T)).\) Assuming an optimal control \(u^0(\tau)\) exists and \(x^0(\cdot) = x(\cdot \,; u^0(\cdot))\) is the corresponding optimal motion, Pontryagin's maximum principle is \[ \big\langle p(\tau), f(\tau, x^0(\ \tau), u^0(\tau)) \big\rangle = \max_{u \in U}\big\langle p(\tau), f(\tau, x^0(\tau), u) \big\rangle \tag{1} \] where \(p(\cdot)\) is the solution of the integral costate equation \[ p(\tau) = -\frac{\partial_x \sigma(x^0(T))}{\Gamma(\alpha)(T - \tau)^{1 - \alpha}} + \frac{1}{\Gamma(\alpha)} \int_\tau^T \frac{\partial_x f(\xi, x^0(\xi), u^0(\xi))^T p(\xi)}{(\xi - \tau)^{1 - \alpha} }d\xi. \] The \textit{optimal result functional} is \[ \varphi^0(t, w(\cdot)) = \inf_{u(\cdot) \in \mathcal{U}(t, T)} \sigma \big( x(T; t, w(\cdot), u(\cdot)) \big) \] where the \((t, w(\cdot))\) are feasible positions for the system and the dynamic programming approach leads to the Hamilton-Jacobi-Bellman equation \[ \partial_t^\alpha \varphi(t, w(\cdot)) + H(t, w(t), \nabla^\alpha \varphi(t, w(\cdot)) = 0 \tag{2} \] with Hamiltonian \(H(\tau, x, s) = \min_{u \in U}\langle s, f(\tau, x, u)\rangle\). Under various conditions on the system the author connects the maximum principle (1) and the dynamic programming approach (2) showing \[ p(\tau) = - \nabla^\alpha\varphi^0(\tau, x_\tau^0(\cdot)). \]