Library / Advanced Mathematics
Singular value decomposition, or SVD, factors a matrix into orthogonal directions and nonnegative scale factors. It is one of the most useful structured factorizations in linear algebra.
The SVD of a matrix writes it as U Sigma V^T, where U and V
are orthogonal matrices and Sigma is diagonal with nonnegative entries called singular
values.
This factorization separates direction from scale in a way that is stable and computationally useful even for non-square matrices.
SVD reveals dominant directions, rank structure, low-rank approximations, and geometric stretching behavior. That is why it appears in numerical linear algebra, optimization, data analysis, and machine learning.
Geometrically, SVD says that a matrix can be viewed as a rotation, followed by axis-aligned stretching, followed by another rotation. That picture is often easier to understand than a raw matrix of entries.
SVD is useful for pseudoinverses, compression, noise filtering, principal subspace analysis, and stable numerical algorithms. It is one of the workhorse decompositions of applied mathematics.
SVD belongs in the same cluster as eigenvalues, matrix exponentials, and tensor structure. It is not the same as eigen-decomposition, but it plays a similar role in making operator behavior more interpretable.
Rather than treating a matrix as an opaque block of coefficients, SVD exposes the important directions and scales that govern its behavior. That is why it remains one of the most valuable decompositions in practical mathematics.