Notes on accuracy and stability of algorithms in numerical linear algebra (Q4938346)

The effects of rounding errors on algorithms in numerical linear algebra have been much-studied for over fifty years, since the appearance of the first digital computers. The subject continues to occupy researchers, for several reasons. First, not everything is known about established algorithms. Second, new algorithms are continually being derived, and their behaviour in finite precision arithmetic needs to be understood. Third, new error analysis techniques lead to different ways of looking at and comparing algorithms, requiring a reassessment of conventional wisdom. The main purpose of these notes is to describe some up to date results in rounding error analysis in a way accessible to non-experts. We have chosen to analyze several practically important algorithms that are not thoroughly treated in numerical linear algebra textbooks. Section 3 considers block \(\text{LDL}^{\text{T}}\) factorization and \textit{J. O. Aasen's} method for symmetric indefinite systems [BIT 11, 233-242 (1971; Zbl 0242.65032)], Section 4 QR factorization and the solution of constrained least squares problems, and Section 5 Jacobi's method for the singular value decomposition. These notes can be regarded as a supplement to the author's book: Accuracy and stability of numerical algorithms (1996; Zbl 0847.65010). They include mathematical problems and algorithms not treated therein. For the entire collection see [Zbl 0920.00043].