A family of matrix flows converging to normal matrices
Masaki Izumi
TL;DR
The paper develops a unified ODE-based framework for matrix and operator flows that converge to normal elements, encompassing the Aluthge flow and the Haagerup gradient flow as special cases. By imposing regularity conditions on a pair of functions $\varphi=(\varphi_1,\varphi_2)$ and exploiting a decomposition into diagonal and strictly upper-triangular components, it establishes global existence, unitary-invariant behavior, and Lyapunov-type convergence results, with exponential convergence under suitable hypotheses. It further extends the theory to bounded Hilbert space operators with unitarily invariant norms, yielding both convergence and non-convergence phenomena depending on the chosen ideal and regularity, and it highlights concrete dynamics (e.g., two-eigenvalue Haagerup flow) and potential limitations in infinite dimensions. The work connects finite-dimensional matrix dynamics to operator-algebraic contexts, offering a flexible template for analyzing similarity-orbit flows and their normal limiting behavior. Overall, it provides a robust, continuous analogue to iterated Aluthge transforms and clarifies when and how such flows stabilize to normal structures.
Abstract
The celebrated Antezana-Pujals-Stojanoff Theorem states that the iterated Aluthge transforms of an arbitrary matrix converge to a normal matrix. We introduce a family of matrix flows that share this convergence property by defining them through ordinary differential equations. The family includes a continuous analogue of the Aluthge transform, as well as a differential equation discussed by Haagerup in the context of II$_1$ factors. We also examine the same type of flows in the setting of Hilbert space operators equipped with unitarily invariant norms.
