No label defined (Q4522256)

The \(n\times n\)-matrix \(A=(a_{i,j})\) is bisymmetric if \(a_{i,j}=a_{j,i}=a_{n-j+1,n-i+1}\), \(i,j=1,\ldots ,n\). The set of all bisymmetric matrices is denoted by \(BSR^{n\times n}\). Set \(P_n=\{ A\in BSR^{n\times n}|x^TAx\geq 0, \forall x\in {\mathbb{R}}^n\}\). The following two problems are considered: Problem I. Given the \(n\times m\)-matrices \(X,B\), find \(A\in P_n\) such that \(AX=B\). Problem II. Given the \(n\times n\)-matrix \(A^*\), find \(\hat{A}\in S_E\) such that \(\|A^*-\hat{A}\|_F=\min _{A\in S_E}\|A^*-A\|_F\) where \(S_E\) is the solution set of Problem I and \(\|.\|_F\) stands for the Frobenius norm (the norm defined by the inner product \((A,B)=\text{tr}(B^TA)\)). The paper discusses the geometric construction of \(BSR^{n\times n}\) followed by the necessary and sufficient conditions for the solvability of Problem I and the proof that \(S_E\) is a closed convex set. Then the unicity of the solution to Problem II is proved and its expression is given. The paper includes an algorithm for finding this optimal solution and a numerical example.