A scaling characterization of nc-rank via unbounded gradient flow

Abstract

Given a tuple of $n \times n$ complex matrices ${\cal A} = (A_1,A_2,\ldots, A_m)$, the linear symbolic matrix $A = A_1x_1 + A_2x_2 + \cdots + A_m x_m$ is nonsingular in the noncommutative sense if and only if the completely positive operators $T_{\cal A} (X) = \sum_{i=1}^m A_i X A_i^{\dagger}$ and $T_{\cal A}^*(X) = \sum_{i=1}^m A_i^{\dagger} X A_i$ can be scaled to be doubly stochastic: For every $\epsilon > 0$ there are $g,h \in GL(n,\mathbb{C})$ such that $\|T_{g^{\dagger}{\cal A}h}(I)- I\| < \epsilon$, $\| T^*_{g^\dagger{\cal A}h}(I) - I\| < \epsilon$. In this paper, we show a refinement: The noncommutative corank of $A$ is equal to one-half of the minimum residual $\|T_{g^{\dagger}{\cal A}h}(I) - I\|_1 + \|T^*_{g^{\dagger}{\cal A}h}(I) - I\|_1$ over all possible scalings $g^{\dagger}{\cal A}h$, where $\|\cdot \|_1$ is the trace norm. To show this, we interpret the residuals as gradients of a convex function on symmetric space $GL(n,\mathbb{C})/U_n$, and establish a general duality relation of the minimum gradient-norm of a lower-unbounded convex function $f$ on $GL(n,\mathbb{C})/U_n$ with an invariant Finsler metric, by utilizing the unbounded gradient flow of $f$ at infinity.