The ``window problem'' for series of complex exponentials (Q1976466): Difference between revisions

Among the classical problems pertaining to sequences of exponentials \( e^{i\lambda_kt}\), \( \{\lambda_k\}\subset \mathbb C\), are the completeness and stability issues: for which \( \delta>0 \) are the exponentials complete in \(L^2[0,\delta]\); furthermore, if so, do they form a frame, basis etc? One can also ask about the sense in which an entire function can be recovered from its values at the points \( \lambda_k\). A problem that is dual to these in some sense is the so-called windowing problem which is: if one knows \(f(t) =\sum_k c_k e^{i\lambda_k t} \) over some interval \([0,\delta]\) can one recover the sequence \( {\mathbf c}_T= \{c_ke^{i\lambda_k T} \}\)? More precisely, under what conditions on \( \delta, T\) and \( \{\lambda_k\} \) is this possible and when is the operator \({\mathbb C}^T_\delta: f\mapsto \{c_ke^{i\lambda_k T} \} \) continuous from the span of the exponentials into \(\ell^2\)? The main result of the paper states that \({\mathbb C}^T_\delta \) is continuous provided the \({\lambda_k } \) satisfy a uniform separation condition, that is, there is a radial function \(\nu(r)\) such that at most \(\nu(r)\) of the \(\lambda_k\) lie in any disk of radius \(r\), and provided \(\delta\) and \(T\) are large enough. Explicit bounds on the norm of \({\mathbb C}^T_\delta \) are given in terms of \(\nu(r)\), \(\delta\) and \(T\). Not surprisingly the bounds boil down to estimates on the growth of certain entire functions that depend on these parameters. But an important part of the theory is an explicit construction of a basis of \(L^2(0,\delta)\) that is biorthogonal to the exponentials \( e^{i\lambda_kt}\). Applications to control theory including boundary control of heat flow and vibrational control with structural damping are provided. In these applications the \(\lambda_k\) are real although the general theory treats \(\lambda_k\) with possibly unbounded imaginary parts, in which case the estimates on \({\mathbb C}^T_\delta \) blow up as \(T\to 0\).