Sufficient conditions for the solvability of a Sylvester-like absolute value matrix equation (Q2213745)

The author considers real matrix equations of the form (*) \(AXB+C\left\vert X\right\vert D=E\), where \(A,B,C,D\) and \(E\) are given rectangular matrices of appropriate sizes and \(X\) is the unknown matrix. The absolute value of \(X\) is denoted by \(\left\vert X\right\vert \) and inequalities such as \(X\leq Y\) should be interpreted component-wise. The special case (**) \(Ax-C\left\vert x\right\vert =e\), where \(e\) and \(x\) are column matrices is closely related to a problem in linear programming. Conditions for the existence and the uniqueness of solutions for (**) have been studied in a series of papers, beginning with [\textit{J. Rohn}, Linear Algebra Appl. 126, 39--78 (1989; Zbl 0712.65029)]. In the present paper, the author shows how these results can be extended to the general case (*) using Kronecker products. If \(A\) is an \(m\times n\) matrix then \(\mathrm{vec}(A)\) is the vector \(mn\)-vector obtained by stacking the successive columns of \(A\) below one another. It is well known that for compatible matrices \(\mathrm{vec} (ABC)=(C^{T}\otimes A)\mathrm{vec}(B)\) (see, for example, [\textit{R. A. Horn} and \textit{C. R. Johnson}, Matrix analysis. 2nd ed. Cambridge: Cambridge University Press (2013; Zbl 1267.15001)]). Using this observation, the equation (*) can be rewritten as an equation of the form (**) in which \(x:=\mathrm{vec}(X)\). The following theorems are then proved. Theorem 1. The equation (*) has a unique solution for each \(E\) if any of the following is true: (i) the singular values satisfy \(\sigma_{\max} (C)\sigma_{\max}(D)<\sigma_{\min}(A)\sigma_{\min}(B)\); (ii) \(A\) and \(B\) are square and nonsingular and the spectral radii satisfy \[\rho(|A^{-1} C|)\rho(|B^{-1}D|)<1;\] (iii) \(A\) and \(B\) are square and the only solution to \(|AXB|\leq|C||X||D|\) is \(X=0\). Theorem 2. If \(C\) is square and nonsingular with \(0\neq C^{-1}E\geq0\) and \(\sigma_{\max}(A)\sigma _{\max}(B)<\sigma_{\min}(C)\), then \(AXB-C|X|=E\) has no solution.