Subspaces, Basis, and Dimension
subspace, basis, dimension, column space, linear independence
1 Role
This page turns linear combinations into actual spaces you can reason about.
It explains the three ideas that keep traveling together in linear algebra:
- a
subspaceis a linear world closed under the operations we care about - a
basisis a minimal set of directions that builds that world - the
dimensioncounts how many independent directions that world really has
2 First-Pass Promise
Read this page after Matrices and Linear Maps.
If you stop here, you should still understand:
- what makes a subset a subspace
- what a basis does
- what dimension is counting
- why column space is one of the most important subspaces in applications
3 Why It Matters
These ideas matter because many later problems are really questions about the right subspace:
- in least squares, the fitted vector must lie in the column space of the feature matrix
- in signal processing, a signal model may live in a low-dimensional subspace
- in dimensionality reduction, we search for a smaller subspace that still captures useful variation
- in numerical methods, the useful part of a computation is often restricted to a smaller space of directions
So this page gives you the language for where the action lives.
4 Prerequisite Recall
- the span of vectors is the set of all their linear combinations
- a matrix-vector product \(Ax\) lies in the span of the columns of \(A\)
- a linear map sends linear combinations to linear combinations
5 Intuition
A subspace is a place where linear algebra can happen without leaving the room.
If you add two vectors in the room, you stay in the room. If you scale a vector in the room, you stay in the room. That closure is what makes the set useful.
But once you know a subspace exists, you do not want to carry every vector in it around. You want a short generating list with no redundancy. That is what a basis gives you.
Dimension then answers the natural next question:
How many genuinely different directions does this space have?
So the flow is:
subspace -> basis -> dimension
A basis tells you how to build every vector in the subspace, and dimension tells you how many independent knobs you actually need.
6 Formal Core
Definition 1 (Definition) A subset \(W \subseteq \mathbb{R}^n\) is a subspace if:
- \(0 \in W\)
- whenever \(u,v \in W\), then \(u+v \in W\)
- whenever \(u \in W\) and \(c \in \mathbb{R}\), then \(cu \in W\)
So a subspace is closed under vector addition and scalar multiplication.
Definition 2 (Basis) A list of vectors \(v_1,\dots,v_k\) is a basis for a subspace \(W\) if:
- the vectors span \(W\)
- the vectors are linearly independent
A basis is therefore a spanning set with no redundancy.
Theorem 1 (Key Statement) Every finite-dimensional subspace has a basis, and any two bases of the same finite-dimensional subspace have the same number of vectors.
That common number is called the dimension of the subspace:
\[ \dim(W) = \text{number of vectors in any basis of } W. \]
7 Worked Example
Consider the set
\[ W = \left\{ \begin{bmatrix} x \\ y \\ z \end{bmatrix} \in \mathbb{R}^3 : z = x + y \right\}. \]
This is the plane through the origin given by the equation \(z=x+y\).
First rewrite a general vector in \(W\):
\[ \begin{bmatrix} x \\ y \\ z \end{bmatrix} = \begin{bmatrix} x \\ y \\ x+y \end{bmatrix} = x \begin{bmatrix} 1 \\ 0 \\ 1 \end{bmatrix} + y \begin{bmatrix} 0 \\ 1 \\ 1 \end{bmatrix}. \]
So every vector in \(W\) lies in the span of
\[ v_1 = \begin{bmatrix} 1 \\ 0 \\ 1 \end{bmatrix}, \qquad v_2 = \begin{bmatrix} 0 \\ 1 \\ 1 \end{bmatrix}. \]
That already shows
\[ W = \operatorname{span}\{v_1,v_2\}. \]
Now check independence. If
\[ a v_1 + b v_2 = 0, \]
then
\[ a \begin{bmatrix} 1 \\ 0 \\ 1 \end{bmatrix} + b \begin{bmatrix} 0 \\ 1 \\ 1 \end{bmatrix} = \begin{bmatrix} a \\ b \\ a+b \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \\ 0 \end{bmatrix}, \]
which forces \(a=0\) and \(b=0\).
So \(v_1\) and \(v_2\) are linearly independent, which means they form a basis for \(W\).
Therefore
\[ \dim(W)=2. \]
This example shows the whole story at once:
- the plane is a subspace because it is built from linear combinations through the origin
- a basis gives a minimal generating set
- dimension counts the number of independent directions, not the number of coordinates in the ambient space
8 Computation Lens
In practice, these questions usually look like one of the following:
Is this set a subspace?Check whether it contains \(0\) and is closed under addition and scaling.What is a basis?Find a spanning set, then remove redundant directions.What is the dimension?Count the vectors in a basis.
For matrices, this viewpoint becomes especially concrete.
If \(A\) is an \(m \times n\) matrix, then the column space
\[ \operatorname{col}(A) = \{Ax : x \in \mathbb{R}^n\} \]
is a subspace of \(\mathbb{R}^m\).
Its basis is an independent set of columns that still spans all outputs of the map \(x \mapsto Ax\), and its dimension is the number of independent column directions.
9 Application Lens
In linear regression, the prediction vector \(X\beta\) must lie in the column space of the design matrix \(X\).
That means the fitted outputs are constrained to a specific subspace determined by the feature columns.
So when we later solve least squares, we are not just “finding a best fit” in vague terms. We are finding the best approximation to the target vector inside a concrete subspace.
That is why subspace language matters so much: it tells you exactly where the model is allowed to live.
10 Stop Here For First Pass
If you can now explain:
- what makes a set a subspace
- why a basis must both span and be independent
- why dimension counts independent directions rather than ambient coordinates
- why column space is the subspace behind regression outputs
then this page has done its main job.
11 Go Deeper
If you want more after the main page:
Proof: Exchange Argument and Why Dimension Is Well DefinedApplication: Low-Dimensional Subspace ModelsVisual intuition: Computation Lab: Basis and Column Space GeometryPractice: Exercises: Subspaces, Basis, and Dimension
12 Optional Paper Bridge
13 Optional After First Pass
If you want more practice before moving on:
- test whether translated planes or lines are subspaces
- describe a subspace both by equations and by a spanning set
- compare a large spanning set with a smaller basis for the same space
- continue to Orthogonality and Least Squares when you want approximation inside a subspace
14 Common Mistakes
- forgetting that a subspace must contain the zero vector
- thinking every line or plane in space is a subspace, even if it does not pass through the origin
- confusing a spanning set with a basis
- thinking dimension means the number of coordinates written down
- forgetting that different bases can describe the same subspace
15 Exercises
Decide whether the set
\[ \left\{ \begin{bmatrix} x \\ y \\ z \end{bmatrix} : x+y+z=0 \right\} \]
is a subspace of \(\mathbb{R}^3\).
Find a basis and the dimension of
\[ \operatorname{span} \left\{ \begin{bmatrix} 1 \\ 0 \\ 1 \end{bmatrix}, \begin{bmatrix} 0 \\ 1 \\ 1 \end{bmatrix} \right\}. \]
Explain why the set
\[ \left\{ \begin{bmatrix} x \\ y \end{bmatrix} : x+y=1 \right\} \]
is not a subspace, even though it looks like a line.
16 Sources and Further Reading
- MIT 18.06SC: Basis and Dimension -
First pass- official source that keeps span, basis, and dimension tightly connected. Checked2026-04-24. - Stanford Math 51 schedule -
First pass- current course sequence showingspan -> subspace -> dimension -> basis -> projection -> regressionas a coherent spine. Checked2026-04-24. - Hefferon, Linear Algebra -
Second pass- strong self-study text for more exercises and proof-level depth on subspaces and basis. Checked2026-04-24. - Deep learning, transformers and graph neural networks: a linear algebra perspective -
Second pass- current survey showing why low-dimensional structure and matrix viewpoints matter in AI. Checked2026-04-24. - A Survey: Potential Dimensionality Reduction Methods -
Paper bridge- bridge toward how low-dimensional subspaces motivate practical reduction methods. Checked2026-04-24.
Sources checked online on 2026-04-24:
- MIT 18.06SC Basis and Dimension
- Stanford Math 51 schedule
- Stanford Math 51H topic summary
- Hefferon, Linear Algebra
- Springer Numerical Algorithms survey on linear algebra perspectives in deep learning
- arXiv survey on dimensionality reduction methods