A fast universal algorithm for quaternion singular value decomposition

DOI10.1016/J.AML.2026.110047MaRDI QIDQ7316092zbMATH OpenFDO

Authors Dong Zhang, Chuan Jiang, V. I. Vasil'ev, Tongsong Jiang

Publication date 19 August 2026

Published in Applied Mathematics Letters (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1016/J.AML.2026.110047

Mathematics Subject Classification ID

Matrices over special rings (quaternions, finite fields, etc.) (15B33) Numerical computation of eigenvalues and eigenvectors of matrices (65F15)

Review text

This paper addresses a critical flaw in the widely used QSVD algorithm proposed by Le Bihan and Mars -- namely, its failure when quaternion singular values are repeated. In such cases, the extracted singular vectors lose quaternion unitarity, leading to inaccurate matrix reconstruction.\N\NThe authors introduce a fast universal QSVD algorithm that retains the complex SVD stage but adds a lightweight correction step for invariant subspaces associated with repeated singular values. This correction enforces a proper paired orthonormal structure before the quaternion factors are recovered.\N\NThe proposed method is both efficient and robust -- it handles distinct and repeated singular values uniformly, guarantees unitary quaternion factors, and achieves reconstruction accuracy at the level of machine precision.\N\NNumerical experiments demonstrate that the new algorithm consistently outperforms existing QSVD implementations, including complex structure-preserving methods and the QTFM toolbox, in terms of computational speed while preserving high accuracy. This makes it a reliable tool for quaternion-based applications such as color image processing and vector-sensor signal analysis.

Cites work

• A complex structure-preserving algorithm for computing the singular value decomposition of a quaternion matrix and its applications
• A complex structure-preserving algorithm for the full rank decomposition of quaternion matrices and its applications
• A fast algorithm for the Schrödinger equation in quaternionic quantum mechanics
• A new complex structure-preserving method for QSVD
• An algebraic technique for total least squares problem in quaternionic quantum theory
• Approximation strategy based on the T-product for third-order quaternion tensors with application to color video compression
• Efficient quaternion CUR method for low-rank approximation to quaternion matrix
• Low-Rank Quaternion Approximation for Color Image Processing
• LSQR algorithm with structured preconditioner for the least squares problem in quaternionic quantum theory
• Quaternion algebras
• Quaternions and matrices of quaternions
• Real structure-preserving algorithms of Householder based transformations for quaternion matrices
• Robust quaternion matrix completion with applications to image inpainting.
• Singular value decomposition of quaternion matrices: a new tool for vector-sensor signal processing
• Structure Preserving Quaternion Biconjugate Gradient Method
• Vector Sparse Representation of Color Image Using Quaternion Matrix Analysis

Cited in

(1)

• An algorithm for the split quaternion total least squares problem and application in color image restoration

This page was built for publication: A fast universal algorithm for quaternion singular value decomposition

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7316092)