On principal angles between subspaces of Euclidean space
Given two matrices \(X\) and \(Y\), with full column rank and common row dimension, the calculation of the cosines of the principal angles between the two subspaces generated by them can be efficiently carried out, as shown by \textit{A. Björck} and \textit{G. H. Golub} [Math. Comput. 27, 579-594 (1973; Zbl 0282.65031)], by means of the singular value decomposition of \(Q*T_x Q_y\), where \(Q_x\) and \(Q_y\) are the respective orthogonal factors in the \(QR\) decomposition of \(X\) and \(Y\). The present paper examines at first the numerical stability of the Björck-Golub algorithm and it is proved that it is stable in the framework of backward error analysis. After this, a new algorithm is proposed. Instead of directly computing \(Q_x\) and \(Q_y\), \(LU\) Gaussian factorizations of \(X\) and \(Y\) with complete (or partial) pivoting are calculated and the columns of unit lower factors are the new matrices whose columns are to be orthogonalized. This idea, which goes back to \textit{M. J. D. Powell} and \textit{J. K. Reid} [Inform. Process., Proc. IFIP Congr., Edinburgh 1968, No. 1, 122-126 (1969; Zbl 0154.47002)], improves the original Björck-Golub algorithm in the following way. First of all, it is backward stable too. Then, although both algorithms are sensitive to certain forms of ill-conditioning in \(X\) and \(Y\), the improved algorithm usually retains its accuracy properties, while the original one does not. This is shown with a minute analysis of a nice set of interesting numerical examples.
- Principal angles between subspaces in an A-based scalar product: Algorithms and perturbation estimates
- A canonical CS representation of a pair of subspaces
- Jordan's principal angles in complex vector spaces
- On the largest principal angle between random subspaces
- scientific article; zbMATH DE number 4027680
- On principal angles between subspaces in \(\mathbb{R}^n\)
- Canonical analysis relative to a closed subspace
- Product cosines of angles between subspaces
- Blade products and angles between subspaces
- Applications of CS decomposition in linear combinations of two orthogonal projectors
- Principal angles between subspaces in an A-based scalar product: Algorithms and perturbation estimates
- A canonical CS representation of a pair of subspaces
- scientific article; zbMATH DE number 988707 (Why is no real title available?)
- Jordan's principal angles in complex vector spaces
- Vector Fitting for Matrix-valued Rational Approximation
- scientific article; zbMATH DE number 4027680 (Why is no real title available?)
- Principal angles and approximation for quaternionic projections
- scientific article; zbMATH DE number 1159488 (Why is no real title available?)
- The Role of Principal Angles in Subspace Classification
- scientific article; zbMATH DE number 3992311 (Why is no real title available?)
- The Angle Between Complementary Subspaces
- Angles between subspaces and their tangents
- Matrices \(A\) such that \(AA^\dagger-A^\dagger A\) are nonsingular
- Euclidean Representation of Low-Rank Matrices and Its Geometric Properties
- On the smoothness of principal angles between subspaces and their application to angular values of dynamical systems
- Spectral norm posterior contraction in Bayesian sparse spiked covariance matrix model
- On the largest principal angle between random subspaces
- The many proofs of an identity on the norm of oblique projections
- On the angular metrics between linear subspaces
This page was built for publication: On principal angles between subspaces of Euclidean space
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2706249)
