A New Polar Decomposition in a Scalar Product Space

Abstract: There are various definitions of right and left polar decompositions of an $m\times n$ matrix $F \in \mathbb{K}^{m\times n}$ (where $\mathbb{K}=\mathbb{C}$ or $\mathbb{R}$) with respect to bilinear or sesquilinear products defined by nonsingular matrices $M\in \mathbb{K}^{m\times m}$ and $N\in \mathbb{K}^{n\times n}$. The existence and uniqueness of such decompositions under various assumptions on $F$, $M$, and $N$ have been studied. Here we introduce a new form of right and left polar decompositions, $F=WS$ and $F=S'W'$, respectively, where the matrix $W$ has orthonormal columns ($W'$ has orthonormal rows) with respect to suitably defined scalar products which are functions of $M$, $N$, and $F$, and the matrix $S$ is selfadjoint with respect to the same suitably defined scalar products and has eigenvalues only in the open right half-plane. We show that our right and left decompositions exist and are unique for any nonsingular matrices $M$ and $N$ when the matrix $F$ satisfies $(F^{[M,N]})^{[N,M]}=F$ and $F^{[M,N]}F$ ($FF^{[M,N]}$, respectively) is nonsingular, where $F^{[M,N]}=N^{-1} F^# M$ with $F^#=F^T$ for real or complex bilinear forms and $F^#=\bar{F}^T$ for sesquilinear forms. When $M=N$, our results apply to nonsingular square matrices $F$. Our assumptions on $F$, $M$, and $N$ are in some respects weaker and in some respects stronger than those of previous work on polar decompositions.