Regional gradient-constrained control problem. Approaches and simulations (Q1410226)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Regional gradient-constrained control problem. Approaches and simulations
scientific article

    Statements

    Regional gradient-constrained control problem. Approaches and simulations (English)
    0 references
    14 October 2003
    0 references
    The system under consideration is governed by a parabolic equation \[ \frac{\partial y}{\partial t} = Ay + Bu, \quad (y,t)\in \Omega\times ]0,T[ \] Here \(\Omega\subset \mathbb{R}^n\), \(A\) is a second-order elliptic operator with smooth coefficients, \(B:\,\mathbb{R}^m\to L_2(\Omega)\) is a linear bounded operator. Initial and zero Dirichlet boundary conditions are imposed. Given a subdomain \(\omega \subset\Omega\) and two functions \(\alpha\) and \(\beta\) from \(L((\omega))^n\), a control function \(u\in L_2(0,T,\mathbb{R}^m)\) is called admissible if for the respective solution \(y_u\) inequalities \[ \alpha(\cdot) \leq \nabla y_u(t=T, \cdot)| _{x\in\omega} \leq \beta(\cdot) \] hold true almost everywhere in \(\omega\). The existence of an admissible control is called \([\alpha(\cdot), \beta(\cdot)]\)-gradient controllable on \(\omega\). Two approaches to characterize the admissible control minimizing \(\| u\| ^2_{L_2}\) are considered. In the proof of the optimality criterion in terms of subdifferentials (Proposition~5), the authors use the equality \[ \sup_{a(\cdot)\leq v(\cdot)\leq b(\cdot)} \langle w, v \rangle_{L_2} = \sup_{0\leq \lambda\leq 1} \langle w, \lambda a + (1-\lambda)b \rangle_{L_2} \] which is, in general, a strict inequality. Therefore, the subdifferential approach is not proved. To develop a numerical minimization procedure for a reduced family of admissible controls of the type \(u(t,x)=\chi_D f(x)u(t)\), Lagrange multipliers are applied and an adjoint problem is derived. Here \(D\) and \(f\) are, respectively, a fixed subdomain of \(\bar{\Omega}\), and a function defined on \(D\). A class of such controls is called a strategic actuator if it includes an admissible control. Two numerical examples with pointwise actuators are presented. The first one is not illustrative, because the constraints are active only at a single point \(t=T\). In the second one, the method fails, so it is concluded that the actuator is not strategic.
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    parabolic system
    0 references
    gradient controllability
    0 references
    actuator
    0 references
    subdifferential
    0 references
    Lagrange multipliers
    0 references
    admissible controls
    0 references
    strategic actuator
    0 references
    0 references
    0 references