Course contentsShow
← Mathematics
Lesson 2281 of 3,12147. Linear Algebra — Applications and Advanced TopicsFree lesson

Geometric Interpretation of SVD

Visualize how SVD decomposes a linear transformation into rotation, scaling, and rotation.

Geometric Interpretation of SVD

You'll learn how SVD breaks down any linear transformation into three simple geometric steps: rotate, scale, rotate.

The Core Idea

The Singular Value Decomposition tells us that any linear transformation can be viewed as a sequence of three geometric operations:

  1. First rotation (or reflection): The matrix V^T rotates the input space, aligning it with special directions
  2. Scaling along axes: The diagonal matrix Σ stretches or compresses along these aligned directions by the singular values
  3. Second rotation (or reflection): The matrix U rotates the scaled result into the output space

Think of it like this: imagine you're transforming an ellipse. SVD says "first, I'll rotate it so its axes align with the coordinate axes (V^T). Then I'll stretch it along those axes by specific amounts (Σ). Finally, I'll rotate the whole thing to its final orientation (U)."

Why This Matters

For a transformation A = UΣV^T:

  • V^T identifies the principal input directions (right singular vectors)
  • Σ reveals how much stretching happens in each direction (singular values)
  • U shows where these directions end up in the output (left singular vectors)

This decomposition reveals the "skeleton" of any linear map. A circle becomes an ellipse through SVD: the singular values are the ellipse's axis lengths, and U and V describe its orientation.

Unlike eigendecomposition (which only works for square matrices and can involve complex numbers), SVD works for any matrix and uses only real orthogonal transformations and positive scalings.

Key Takeaway: SVD geometrically decomposes any linear transformation into rotation → scaling → rotation, revealing the fundamental action of the matrix on space.