Weighted tight frames of exponentials on a finite interval (Q1314860): Difference between revisions

Let \(I= [0,R]\), \(\Lambda\subseteq R\) be a discrete set, \(e_ \lambda(x)= e^{2\pi i\lambda x}\), and \(c_ \lambda>0\). The collection \(\{c_ \lambda\), \(e_ \lambda\), \(\lambda\in \Lambda\}\) is called a (weighted) frame for \(L^ 2(I)\), if there exist constants \(A,B>0\) so that \[ A\| f\|_ 2^ 2\leq \sum_{\lambda\in \Lambda} c_ \lambda^ 2| \langle f, e_ \lambda\rangle|^ 2\leq B\| f\|^ 2_ 2 \] for all \(f\in L^ 2(I)\). If \(A= B\), it is called a tight frame, and \(f\) has the unconditional expansion \(f= \sum_ \lambda c_ \lambda^ 2\widehat f(\lambda) e_ \lambda\) into a nonharmonic Fourier series, where \(\widehat f= \int f(x) e_{-\lambda} x dx= \langle f, e_ \lambda\rangle\) is the Fourier transform of \(f\). This paper gives a complete characterization of tight frames. Roughly speaking, the main result states that the sampling set \(\Lambda\) is the zero set of a function associated to the Nevanlinna class, and the weights \(c_ \lambda\) are values of a related function. Theorem: The following properties are equivalent: (A) \(\{c_ \lambda e_ \lambda\), \(\lambda\in \Lambda\}\) is a tight frame for \(L^ 2(I)\) and the coefficient map \(f\to (c_ \lambda\langle f,e_ \lambda\rangle)_{\lambda\in \Lambda}\) from \(L^ 2(I)\) into \(I^ 2(\Lambda)\) has finite codimension. (B) There exist two real-valued polynomials \(P\) and \(Q\) with no common factor, \(\deg P<\deg Q\leq m\) such that \(\text{Im }P/Q(x+ iy)\geq 0\) for \(y>0\) and \[ \Lambda= \left\{x\in \mathbb{R}: {P(x)\over Q(x)}=-\tan(\pi R x)\right\}. \] If \(S= P+iQ\) and \(F(x)=\overline{S(x)}+e^{-2\pi iR x} S(x)\), then \(c_ \lambda= 2\pi | S(\lambda)|/ | F'(\lambda)|\). This is a striking result and the proof is very ingenious and introduces new ideas to the theory of nonharmonic Fourier series. First, the frame problem is reformulated as an extension problem for positive-definite functions, which is then analyzed carefully. This makes the surprising appearance of functions in the Nevanlinna class (as in condition (B)) at least plausible, since they are used for the parametrization of solution to the moment problem and in the extension problem for positive-definite functions in other contexts as well.