Finite frame varieties: Nonsingular points, tangent spaces, and explicit local parameterizations (Q650088): Difference between revisions

This paper is concerned with real and complex \((\mu,S)\)-frames, finite frames having a given frame operator \(S\) and consisting of vectors \(\{f_1, f_2, \dots, f_N\}\) having specified lengths \(\{\mu_1, \mu_2, \dots, \mu_N\}\), respectively. In terms of the synthesis operator \(F=[f_1 f_2 \cdots f_N]\), a \((\mu,S)\)-frame satisfies the quadratic equations \(FF^*=S\) and \(\mathrm{diag}(F^* F)=\mu\). From a geometric viewpoint, the condition involving \(S\) means that \(F\) is in an ellipsoidal, deformed Stiefel manifold, while the norm condition means that it is in a product of spheres. The intersection of these two manifolds forms a variety with possible singular points. The intersection of these two matrix manifolds forms a real algebraic variety with possible singular points. The paper characterizes the singular points in this variety as orthodecomposable frames. This result is the generalization of earlier work with Dykema on spherical tight frames [\textit{K. Dykema} and \textit{N. Strawn}, ``Manifold structure of spaces of spherical tight frames'', Int. J. Pure Appl. Math. 28, No. 2, 217--256 (2006; Zbl 1134.42019)]. A finite frame is orthodecomposable if it can be partitioned into subsequences which are spanning for mutually orthogonal subspaces. If a \((\mu,S)\)-frame is not orthodecomposable, then it is at a non-singular point in the variety. An explicit, real-analytic parameterization for the neighborhood of any non-singular point is constructed. The method of construction begins with a suitable permutation of the frame vectors, assigning the order inductively so that the last \(d\) vectors form a basis \(B\) for the Hilbert space. The parameterization is realized by perturbing the first \(N-d\) vectors in a norm-preserving way and by compensating the change in the frame operator with the remaining \(d\) vectors in \(B\). There is some residual freedom in the choice of the remaining vectors which amounts to choosing the entries below the subdiagonal of the synthesis operator of \(B\). The construction is explained in detail for \((\mu, S)\)-frames in real Hilbert spaces. The appendix of the paper explains a similar strategy for complex Hilbert spaces.