Generation of finite tight frames by Householder transformations (Q2491237)

Two theorems are proved. Theorem 1.1. Let \(V:=[a_1,\dots,a_p] \subset\mathbb{H}^n\), where \(\mathbb{H}\) is either \(\mathbb{R}\) or \(\mathbb{C}\). Suppose that \(\lambda_1\geq \lambda_2 \geq\cdots\geq\lambda_n\) are all the eigenvalues of \(VV^*\). Then for any vectors \(a_{p+1},a_{p+2},\dots, a_m\), the condition number of the matrix \({\mathcal A}:=[a_1, \dots,a_m]\) satisfies the inequality \[ c({\mathcal A}):=\frac{\lambda_{\max}} {\lambda_{\min}}\geq \frac{\lambda_1}{\lambda_{n-k}}, \] where \(k:= m-p\) (if \(n-k\leq 0\), then set \(\lambda_{n-k}:=\lambda_1)\). It follows that at most \(m:=n-1\) vectors \(a_{p+1},\dots,a_{p+n-1}\) are needed to make \({\mathcal A}\) a tight frame matrix (TFM). Theorem 1.2. Under the notations of Theorem 1.1, suppose that \(a_1,\dots,a_p\) are with norm 1. Denote by \(d\) the smallest integer greater than or equal to \(\lambda_1+ 1\), then one can find vectors \(a_{p+1},\dots,a_{nd}\) with norm 1 such that \({\mathcal A}:=[a_1,\dots,a_{nd}]\) is an equinormed TFM. A fast algorithm is provided for generating TFM by using Hausholder transformations, and two elaborated examples are also presented.