Backward stability of iterations for computing the polar decomposition
The authors prove backward stability of a general iteration for computing the polar decomposition, under two assumptions: (a) each iterate is obtained from the previous one in a mixed backward-forward stable way in floating point arithmetic; (b) no singular value of an iterate significantly decreases relative to the largest singular value from one iteration to the next, which is a condition on the iteration function.NEWLINENEWLINEThe analysis is generally applicable since it makes no direct reference to acceleration parameters or implementation details of the iteration. It is used to prove backward stability of the QR-based dynamically weighted Halley algorithm under the assumption that column pivoting and either row pivoting or row sorting are used in the QR factorization. The backward error bound involves a growth factor that can be exponentially large in \(n\) but is known to be small in practice. It is shown that the algorithm can be rarely unstable without pivoting.NEWLINENEWLINEIn addition, the authors prove in a short and simple way that the scaled Newton iteration is backward stable; give insight into why the scaled inverse Newton iteration is not backward stable; and show that the (scaled) Newton-Schulz iteration is backward stable if the starting matrix has 2-norm safely less than \(\sqrt{3}\) but can be unstable if the norm is close to \(\sqrt{3}\) (which is the boundary of the region of convergence).
- A New Scaling for Newton's Iteration for the Polar Decomposition and its Backward Stability
- New numerical methods for computing the polar decomposition
- Optimizing Halley's Iteration for Computing the Matrix Polar Decomposition
- On iterative algorithms for the polar decomposition of a matrix and the matrix sign function
- Stable and efficient computation of generalized polar decompositions
- On the backward stability of Gauss-Jordan elimination
- On Grioli's minimum property and its relation to Cauchy's polar decomposition
- Principal component projection with low-degree polynomials
- Improving backward stability of Sakurai-Sugiura method with balancing technique in polynomial eigenvalue problem.
- The minimization of matrix logarithms: on a fundamental property of the unitary polar factor
- Computing Fundamental Matrix Decompositions Accurately via the Matrix Sign Function in Two Iterations: The Power of Zolotarev's Functions
- On the matrix square root via geometric optimization
- Note on ``A new scaling for Newton's iteration for the polar decomposition and its backward stability by R. Byers and H. Xu
- A New Scaling for Newton's Iteration for the Polar Decomposition and its Backward Stability
- A backward stable algorithm for computing the CS decomposition via the polar decomposition
- Stable and efficient computation of generalized polar decompositions
- The geometric mean of two matrices from a computational viewpoint.
- A Structure-Preserving Divide-and-Conquer Method for Pseudosymmetric Matrices
- Pseudospectral shattering, the sign function, and diagonalization in nearly matrix multiplication time
- A multilinear Nyström algorithm for low-rank approximation of tensors in Tucker format
- On iterative algorithms for the polar decomposition of a matrix and the matrix sign function
- Computing accurate eigenvalues using a mixed-precision Jacobi algorithm
- Accuracy and stability of CUR decompositions with oversampling
- Halving the size of skew-symmetric eigenvalue problems via the polar decomposition
This page was built for publication: Backward stability of iterations for computing the polar decomposition
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2910965)
