Computing optimal scalings by parametric network algorithms

From MaRDI portal





A symmetric scaling of a matrix \(A_{n\times n}\neq 0\) is a matrix of the form \(A'=XAX^{-1}\), where X is a nonnegative, nonsingular, diagonal matrix. Analogously, an asymmetric scaling of a matrix \(B_{m\times n}\) is a matrix of the form \(B'=XBY^{-1}.\) The authors consider two optimization problems. The first one is to find a scaling A' of a square matrix A, such that max\(| a'\!_{ij}|\) is minimal. The second is to find an asymmetric scaling B' of B, such that \(\max_{b'\!_{k\ell}\neq 0}| b'\!_{ij}/b'\!_{k\ell}|\) is minimal. The second problem is considered also if B is a square matrix. The authors show that these problems can be restated as parametric network optimization problems with one and two parameters in the goal function and the right-hand side, respectively. There exist well known algorithms solving these cases.











This page was built for publication: Computing optimal scalings by parametric network algorithms

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3691441)