No label defined (Q4938233)

We study the following matrix approximation problem: Given a matrix \(A\in \mathbb{C}^{n\times n}\), a class \({\mathcal S}\) of structured matrices, and an integer \(r\geq 0\), find an approximant \(B= S+R\), where \(S\in{\mathcal S}\) and rank \(R\leq r\), providing the minimal possible value for \(\|A-B\|_F\). We first present a more detailed, experimental boundary element method motivation for this approximation problem. Then we confine ourselves to a model problem, simple enough to produce some theory yet not trivial. Taking scalar matrices as structured matrices we present an algorithm and prove its convergence for normal matrices. We could note that, rather strikingly, the greater the size the faster the convergence. For the entire collection see [Zbl 0924.00038].