Course contentsShow
← Machine Learning and Deep Learning
Lesson 12 of 3,5381. Mathematical Foundations for Machine LearningFree lesson

Column Space and Null Space

Understand fundamental subspaces: column space (range) and null space (kernel) of matrices.

Column Space and Null Space

You'll learn what column space and null space tell us about a matrix's behavior and why they matter for understanding data transformations.

The Core Idea

A matrix represents a transformation. When you multiply a matrix by different vectors, two fundamental questions arise:

  1. Column Space (Range): What are all possible outputs this matrix can produce?
  2. Null Space (Kernel): Which input vectors get completely squashed to zero?

Column Space

The column space of a matrix is the span of its column vectors—every linear combination you can make from those columns. Think of it as "all destinations reachable" by the matrix transformation.

Example: If matrix A has three columns but they all lie in a 2D plane, the column space is 2-dimensional. No matter what vector you multiply A by, the output stays within that plane.

Why it matters: The dimension of the column space (called the rank) tells you how much "information capacity" the matrix has. Low rank = information loss.

Null Space

The null space contains all vectors that the matrix sends to the zero vector. These are inputs that "disappear" under the transformation.

Example: Imagine a projection matrix that flattens 3D space onto a 2D plane. Any vector perpendicular to that plane gets mapped to zero—those vectors form the null space.

Why it matters: A non-trivial null space means the transformation loses information. Multiple different inputs produce the same output (or zero), making the transformation non-invertible.

The Connection

Together, these subspaces partition your space: the column space shows where you can go, the null space shows what gets lost. Their dimensions always satisfy: dimension = rank + nullity (rank-nullity theorem).

Key Takeaway: Column space reveals all possible outputs of a matrix; null space reveals which inputs vanish to zero—both are essential for understanding information flow in linear transformations.