📐 The singular value decomposition (SVD) is one of the most important tools in applied linear algebra. This 1993 historical paper reconstructs how the idea emerged and evolved long before becoming a standard technique in scientific computing.
SVD expresses any matrix as the product of three parts: two orthogonal matrices and a diagonal matrix containing the singular values. This separates directions, scales, and transformations, even when a matrix is not square or has dependencies among its rows and columns.
The history connects contributions from mathematicians such as Beltrami, Jordan, Sylvester, Schmidt, and von Neumann. It also shows how approximation problems, linear equations, and numerical analysis drove the development of increasingly stable algorithms.
Today, SVD appears in image compression, dimensionality reduction, recommendation systems, least-squares solving, and machine learning. Its importance is not only an elegant formula: it can approximate a matrix while preserving its dominant patterns.
💡 Explanation in a nutshell#
Imagine a matrix as a transformation that deforms space. SVD finds the special axes that are rotated first, how much they are stretched, and how they are rotated again. It turns a complex operation into three simple steps and reveals which information matters most.
More information at the link 👇
