Transfer function approach to output specification in certain linear distributed parameter systems (Q1422545)

The authors consider the linear system: \(\partial w/\partial t= Aw+ bu\), \(y=\langle w,c\rangle\), where \(w\in{\mathcal W}\), \(u\in{\mathcal U}\). The (unbounded) operator \(A\) generates a strongly continuous semigroup \(S(t)\), \(t\geq 0\). The control \(u\), regarded as input, and the output \(y= \langle w,c\rangle\) are scalars. \(c\) is an element of the dual space \({\mathcal W}'\), but \(b\) is allowed to be in a space larger than \({\mathcal W}\). The input is generally in \(L^2[0,\infty)\). With zero initial state \(y(t)= c^*\int_0 S(t-\tau)bu(\tau)\,d\tau\). Following the common practice, the control theory is discussed in the ``frequency domain'' so that the initial time-like process is replaced by the Laplace (or sometimes Fourier) transform into the ``frequency domain'' process: \(\widehat Y(s)={\mathcal T}\widehat U(s)\). Multiplication is replaced by convolution (and vice versa), and the output to input relation in the frequency space is called the transfer function. The transfer function \({\mathcal T}\) obeys also the inverse relation: \(\widehat U(s)= {\mathcal T}^{-1}\widehat Y(s)\). The delay in time is reflected in the frequency space by exponentiation, thus right shift by \(T\) in time is reflected in the frequency space by multiplication by \(e^{-sT}\). All this is well known to experts in control theory, as well as in other related problems. The great Russian theorist A. G. Butkovskij claims that any scientific theory may be restated in the terminology of control theory. Here the authors return to a problem that D. L. Russell has studied on and off since the 1960s, when he published an important MRC Report on control of the one-dimensional wave. The system: \(\partial^2w/\partial t^2-\kappa\partial^2 w/\partial x^2= \delta_\xi U(t)\), where \(\delta_\xi\) is the Dirac delta applied at \(x=\xi\) with zero boundary conditions. Note that the Laplace transform was applied to the \(t\) variable only. It is easy to recognize Green's function, which in the elliptic equation produces sines and cosines, while in the hyperbolic case these trigonometric functions are replaced by hyperbolic sines and cosines. A more general theory is developed on the next few pages. The length is extended from \([0,L)\) to \([0,\infty)\). Some difficulties arise since zeros of the transfer function lead to poles of the Laplace transform of \(U(t)\) containing hyperbolic functions which lie on the imaginary axis. Four mutually exclusive cases are listed, and ``generalized exponential functions'' are introduced, accordingly. A problem of incorporation of the delayed action concludes this section. A fairly short section on arguments concerning positioning the end point for the simplified form of the homogeneous Euler-Bernoulli beam equation: \(\partial^2 w/\partial t^2+ \partial^4 w/\partial x^4= 0\) concludes this article.