Quadratically constrained least squares and quadratic problems (Q1180757): Difference between revisions

For a given \(m\times n\) matrix \(A\) and a known vector \(b\) the authors consider the problem of finding a vector \(x\) such that \(\min\{\| Ax- b\|_ 2: \| x\|_ 2\leq \alpha\}\) with \(\alpha < \| A^ + b\|_ 2\) and a related problem: \(\min\{x^ TAx-2b^ Tx:\;\| x\|_ 2\leq \alpha\}\) with \(\alpha < \|(A-\sigma I)^ +b\|_ 2\) where \(\sigma\) denotes a lower bound for the spectrum of \(A\) provided \(A\) is a symmetric \(n\times n\) matrix. They propose an efficient iterative technique based on Gauss quadrature for computing the Lagrange multiplier associated with such quadratic problems. As a side-effect of the computation, this procedure enables to compute an approximate solution \(x\) without many additional computations. The procedure is well suited for this problem when the dimensions of \(A\) are large and the matrix is sparse. A thorough theoretical treatment of the procedure and some numerical results illustrating its efficiency are also given.