An extreme point result for convexity, concavity and monotonicity of parameterized linear equation solutions (Q1886525)

This paper addresses the classical equation \(Ax=b\), with \(n \times n\) matrix \(A=A(\theta)\) and \(n \times 1\) vector \(b=b(\theta)\) depending on an \(m\)-tuple of parameters \(\theta\) with components \(\theta_i\) entering in a rank-one manner. The focal point of this paper is the dependence of the solution of the linear matrix equation \(Ax=b\) on the parameters \(\theta=(\theta_1,\theta_2, \ldots,\theta_m)\) entering into \(A\) and \(b\). More specifically, the authors study the following problem: for the solution component \(x_i(\theta)\) and the parameter \(\theta_j\), determine if the first and second order partial derivatives of \(x_i\) with respect to \(\theta_j\) are of one sign for all \(\theta\) in a prescribed hypercube \(\Theta_r\) of radius \(r \geq 0\); that is, they determine which components enter the solution either monotonically, convexly or concavely. The authors provide a multilinear factorization of the derivaties of the solution variables \(x_i(\theta)\) and an extreme point condition which is necessary and sufficient for satisfaction of the desired one-sign derivative conditions over \(\Theta_r\). At the end of the paper the authors analyze some applications of convexity, concavity and monotonicity properties.