Left and Right Singular Vectors
You'll learn how the columns of U and V in the SVD form orthonormal bases for the four fundamental subspaces of a matrix.
The Core Idea
When we decompose a matrix A (m × n) using SVD, we write A = UΣVᵀ, where:
- U is an m × m orthogonal matrix
- V is an n × n orthogonal matrix
- Σ contains the singular values
The columns of U and V aren't arbitrary—they're carefully chosen orthonormal vectors called left singular vectors (columns of U) and right singular vectors (columns of V). These vectors form orthonormal bases for the four fundamental subspaces we've studied throughout linear algebra.
How They Connect to the Four Subspaces
Think of A as a transformation. The singular vectors reveal:
-
Right singular vectors (V): The first r columns of V span the row space of A. The remaining columns span the null space of A.
-
Left singular vectors (U): The first r columns of U span the column space of A. The remaining columns span the left null space (null space of Aᵀ).
Here, r is the rank of A (the number of nonzero singular values).
Why This Matters
These singular vectors give us an orthonormal coordinate system perfectly aligned with how A acts. The right singular vectors in the row space get mapped by A to corresponding left singular vectors in the column space, scaled by the singular values. Everything in the null space gets mapped to zero.
This is more refined than arbitrary bases—we now have orthonormal bases that respect the geometry and action of A.
Key Takeaway: The columns of V provide orthonormal bases for the row space and null space, while the columns of U provide orthonormal bases for the column space and left null space—giving complete geometric insight into the matrix transformation.