Numerical linear algebra

About the book

Lloyd N. Trefethen's Numerical Linear Algebra treats matrices as objects that computers must actually manipulate, where rounding, conditioning and cost shape which method is worth using. The subject covers the core operations of scientific computing: solving linear systems, least-squares problems, eigenvalue problems and the singular value decomposition.

The presentation is lecture-style, combining mathematical argument with attention to practical behavior. It explains the reasoning behind stable algorithms and the geometry that makes them work. Readers encounter both classical factorizations and iterative methods, with an emphasis on understanding the procedures.

Its scope sits between a pure linear algebra course and a software manual. The aim is fluency in the ideas that numerical libraries implement, so that a reader can judge when a computed answer deserves trust.

How to read it

Approach the book as a sequence of arguments to follow with pencil and paper. Work through the derivations of factorizations and the conditioning estimates instead of skimming for formulas. The density rewards slow reading; a chapter may need a second pass before its point settles.

Keep a computational outlet nearby. Implementing small examples, or testing stated bounds on matrices you construct, makes the discussion of stability and error concrete. The prose is spare and direct, so difficulty comes from the mathematics rather than from obscurity of style.

Who it suits

This suits students and practitioners in mathematics, engineering and computational science who already know basic linear algebra and want to understand the numerical methods behind software they use. Programmers moving into scientific computing will find the treatment of stability and cost especially useful. Readers seeking a first course in matrices should look elsewhere first.

Reading considerations

The book assumes comfort with matrix notation and some calculus-level reasoning, and it does not function as a reference for every algorithm. Later printings and editions may differ in details and supplementary material, so check which printing you hold if you rely on specific exercises or page references.