Asymptotic analysis of generalized orthogonal flows

DOI10.1016/J.LAA.2025.03.018MaRDI QIDQ6995502zbMATH OpenFDO

Authors Yueh-Cheng Kuo, Huey-Er Lin, Shih-Feng Shieh

Publication date 22 April 2025

Published in Linear Algebra and its Applications (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1016/J.LAA.2025.03.018

zbMATH Keywords

generalized orthogonal flowminimal variationOja-like floworthogonal floworthogonal iterationrate of convergence

Mathematics Subject Classification ID

Eigenvalues, singular values, and eigenvectors (15A18) Initial value problems, existence, uniqueness, continuous dependence and continuation of solutions to ordinary differential equations (34A12) Nonlinear ordinary differential equations and systems (34A34) Theoretical approximation of solutions to ordinary differential equations (34A45) Asymptotic properties of solutions to ordinary differential equations (34D05)

Review text

The initial-value problem \N\[\N\dot Y=BY+Y\left(-\frac{1}{2}Y^T(B^T+B)Y+K\right),\quad Y(0)=Y_0 \N\]\Nis studied, where \(Y=Y(t)\in \mathbb{R}^{n\times k}\), \(n\ge k\), \(Y_0\) is an orthonormal matrix, and \(K\equiv K(t,Y)\in \mathbb{R}^{k\times k}\) is skew-symmetric. For a general matrix function \(K\) it is proved that the solution \(Y(t)\) stays in the Stiefel manifold. The existence of solution and its uniqueness is proved as well. Moreover \(Y(t)=\hat Y(t)Q(t)\), where \({\hat Y}\) satisfies \N\[\N\dot {\hat Y}=B\hat Y+\hat Y\left(-\frac{1}{2}\hat Y^T(B^T+B)\hat Y+K_{\hat Y}\right),\quad \hat Y(0)=Y_0, \N\]\Nand \N\[\N(K_{\hat Y})_{ij}= {1}/{2}(\hat Y^T(B^T+B)\hat Y)_{ij} \N\] \Nif \(i>j\), \( (K_{\hat Y})_{ij}=0 \) if \(i=j\), \N\[\N(K_{\hat Y})_{ij}=- {1}/{2}(\hat Y^T(B^T+B)\hat Y)_{ij} \N\]\Nif \(i<j\), and \(Q\) is a unique solution of the problem \N\[\N\dot Q(t)=-K_{\hat Y}Q+QK(t,\hat YQ),\quad Q(0)=I, \N\]\Nsuch that \(Q^T(t)Q(t)=I\). A generalized orthogonal flow with minimal variation, identical to an Oja-like flow is constructed and analysis of asymptotic behavior and the rate of convergence of this flow is performed.

Cites work

• A simplified neuron model as a principal component analyzer
• A structure preserving flow for computing Hamiltonian matrix exponential
• Hermitian matrices depending on three parameters: Coalescing eigenvalues
• scientific article; zbMATH DE number 6159604 (Why is no real title available?)
• Isospectral Flows and Abstract Matrix Factorizations
• Linear algebra algorithms as dynamical systems
• Matrix Analysis
• Matrix groups for undergraduates.
• Numerical computation of an analytic singular value decomposition of a matrix valued function
• On Smooth Decompositions of Matrices
• On the Continuous Realization of Iterative Processes
• Smoothness and periodicity of some matrix decompositions
• Stability and Convergence of Principal Component Learning Algorithms
• Structure-Preserving Flows of Symplectic Matrix Pairs
• SVD, joint-MVD, Berry phase, and generic loss of rank for a matrix valued function of 2 parameters
• The asymptotic analysis of the structure-preserving doubling algorithms
• The orthogonal flows for orthogonal iteration
• Time-asymptotic dynamics of Hermitian Riccati differential equations

Cited in

(1)

• Generalized gentlest ascent dynamics methods for high-index saddle points

This page was built for publication: Asymptotic analysis of generalized orthogonal flows

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6995502)