## Abstract

Bidiagonal reduction is the preliminary stage for the fastest stable algorithms for computing the singular value decomposition (SVD) now available. However, the best-known error bounds on bidiagonal reduction methods on any matrix are of the form A + δA = UBV^{T}, ∥δA∥_{2} ≤ ε M f(m,n)∥A∥_{2} where B is bidiagonal, U and V are orthogonal, ε_{M} is machine precision, and f(m,n) is a modestly growing function of the dimensions of A. A preprocessing technique analyzed by Higham [Linear Algebra Appl., 309 (2000), pp. 153-174] uses orthogonal factorization with column pivoting to obtain the factorization A = Q (C^{T} 0) P^{T}, where Q is orthogonal, C is lower triangular, and P is permutation matrix. Bidiagonal reduction is applied to the resulting matrix C. To do that reduction, a new Givens-based bidiagonalization algorithm is proposed that produces a bidiagonal matrix B that satisfies C + δC = U(B + δB)V^{T} where δB is bounded componentwise and δC satisfies a columnwise bound (based upon the growth of the lower right corner of C) with U and V orthogonal to nearly working precision. Once we have that reduction, there is a good menu of algorithms that obtain the singular values of the bidiagonal matrix B to relative accuracy, thus obtaining an SVD of C that can be much more accurate than that obtained from standard bidiagonal reduction procedures. The additional operations required over the standard bidiagonal reduction algorithm of Golub and Kahan [J. Soc. Indust. Appl. Math. Ser. B Numer. Anal., 2 (1965), pp. 205-224] are those for using Givens rotations instead of Householder transformations to compute the matrix V, and 2n^{3}/3 flops to compute column norms.

Original language | English (US) |
---|---|

Pages (from-to) | 761-798 |

Number of pages | 38 |

Journal | SIAM Journal on Matrix Analysis and Applications |

Volume | 23 |

Issue number | 3 |

DOIs | |

State | Published - Jan 1 2002 |

## All Science Journal Classification (ASJC) codes

- Analysis