Givens Rotation Algorithm — Geometric Interpretation in 3D

A visual note on using plane rotations to zero selected entries and gradually build a triangular form

The Givens rotation algorithm applies a sequence of rotations in coordinate planes to eliminate selected entries of a matrix one at a time. Each step acts only in a two-dimensional coordinate plane, so the transformation is local and targeted rather than global.

In QR factorization, these rotations are used to turn a matrix into an upper triangular form. Geometrically, each rotation reorients the columns just enough to zero one chosen entry while preserving lengths and orthogonality properties.

Explore more on GraphMath Linear Algebra

Givens rotation algorithm visualized as solid cubes

This animation shows the full three-rotation Givens process for a 3D QR factorization. Each rotation acts in one coordinate plane, eliminates one selected entry and moves the matrix step by step toward upper triangular form.

Animated geometric visualization of the full Givens rotation algorithm using solid cubes, showing three plane rotations that transform a 3D matrix toward upper triangular form. — The solid-cube version presents the whole algorithm using cubes with coordinates in the interval [−1, 1]. The sequence shows all three rotations and the corresponding geometric motion of the column frame.

Givens rotation algorithm visualized as semi-transparent cubes (same matrix)

This animation shows the same full three-rotation process with a different visual representation. Semi-transparent cubes with coordinates in the interval [0, 1] make the destination of the vectors easier to visualize.

Animated geometric visualization of the full Givens rotation algorithm using semi-transparent cubes, matrix entries and three coordinate-plane rotations. — The semi-transparent version emphasizes the same complete QR process from another viewpoint: each rotation removes a selected component while the matrix display tracks the evolving Q and R factors.

Concept

A Givens rotation is an orthogonal transformation that mixes only two coordinates and leaves the others unchanged. In matrix language, it is almost the identity matrix, except for a 2×2 rotation block.

This makes the algorithm especially useful when one wants to eliminate entries one at a time, or when the matrix is sparse and it is better not to disturb more structure than necessary.

Structure

Suppose we want to eliminate an entry below the diagonal. We choose a rotation in the coordinate plane determined by the two relevant rows so that the target entry becomes zero after multiplication. Repeating that process systematically transforms the matrix toward upper triangular form.

In QR factorization, the product of all these Givens rotations gives the orthogonal factor, while the transformed matrix becomes the triangular factor.

Key equations

A = Q R

G(i, j, θ)ᵀ G(i, j, θ) = I

G acts only in the (i, j) coordinate plane

G A eliminates one selected entry of A

R is obtained after a sequence of Givens rotations

Q is the accumulated orthogonal factor

Each step is small, but the accumulated effect produces the full QR factorization.

Query phrases

Givens rotation algorithm geometric interpretation
how Givens rotations build QR decomposition
Givens rotation in 3D
plane rotation to zero a matrix entry
Givens rotation visualization

References

Related chapter: GraphMath — Givens rotations