The controllability of the Gurtin-Pipkin equation: a cosine operator approach (Q2493283)

The evolution of properties of materials with memory is generally modeled by integral equation of convolution type. The Gurtin-Pipkin (G-P) equation modeling the evolution of a thermal system with memory is \(\theta_t(t,x)=\int_0^tb (t-s)\Delta\theta(s,x)\,ds+ f(t,x)\), \(x\in\Omega\), \(t\geq 0\). The initial and boundary conditions are of Dirichlet type. Here \(b(0)>0\) is an essential condition for hyperbolicity of the system. The author simplifies this condition by assuming (without loss of generality) that \(b(0)=1\). This really amounts to a choice of time scale. While the author assumes that the special domain \(\Omega\) is simply connected, with a twice differentiable boundary, he admits that these assumptions are too strong. The use of results of \textit{R. Triggiani} [Appl. Math. Optimization 18, No.~3, 241--277 (1988; Zbl 0656.93011)] would allow the use of domains that are not connected. An another paper of \textit{I. Lasiecka} and \textit{R. Triggiani} [ibid. 7, 35--93 (1981; Zbl 0473.35022)] suggested a cosine operator approach in \(L_2\) control input in the setting of hyperbolic PDE equations. Let \(A\) be the operator generating an exponentially stable semigroup on space \(X=L_2(\Omega)\). Define \({\mathcal A}=i(-A)\), and the cosine operator \({\mathcal R}(t)_+=[e^{{\mathcal A}t}+e^{-{\mathcal A}t}]\). The sine operator is defined similarly: \({\mathcal R}(t)_-=[e^{{\mathcal A}t}-e^{-{\mathcal A}t}]\). Let \(D\) denote the Dirichlet map for Laplace operator, and let \(u\) denote the control restricted to the boundary. With \(\theta= Du\), the author introduces the Lasiecka-Triggiani variable \(\xi=\theta (t)-Du(t)\). The G-P equation becomes \(\xi_t=\int^t_b(t-s) A(\xi)ds-Du'(t)\). Now the author applies the cosine operator \({\mathcal R}(t)_+\) to both sides of this equality. This is manipulated into a classical form of Volterra integral equation. There is a large literature concerning solutions of Volterra equation with a bounded (operator-valued) kernel. After a routine manipulation it becomes clear that one can use the Banach fixed point theorem, showing the existence of a unique solution. To prove the controllability, the author introduces his version of the transfer function, generally relating the ratio of Fourier or Laplace transforms of output and input. The author's version consists in multiplying everything by \(e^{-tx}\), and then integrating from 0 to \(T\), rather then the usual \([0,\infty)\) favored by engineers. This change allows him to compute efficiently the reachable set at time \(T\). This in turn allows him to prove exact controllability of the G-P differential equation in finite time. Since the G-P equation is a favorite model in engineering calculations of materials with memory, this is an important result. Of course, this ``existence'' result is needed, but it offers us no clue to actual applications of the control \(u\). We need a follow up by somebody specifying what it is that the engineers wish to control, and proving the existence (and perhaps uniqueness) of an optimal control.