Basis Images Determine a Linear Map
proof, basis, linear map, matrix representation
1 Claim
We want the theorem behind the slogan
a matrix is determined by what it does to the basis vectors.
Theorem 1 (Theorem) Let \(e_1,\dots,e_n\) be a basis of \(\mathbb{R}^n\), and let \(w_1,\dots,w_n \in \mathbb{R}^m\).
Then there exists a unique linear map \(T : \mathbb{R}^n \to \mathbb{R}^m\) such that
\[ T(e_j) = w_j \qquad \text{for } j=1,\dots,n. \]
2 Why This Proof Matters
This proof is the real reason matrices work.
It tells you:
- why the columns of a matrix are enough to define the whole linear map
- why linear maps are fully controlled by basis data
- why “define the map on a basis and extend by linearity” is a legitimate construction
3 Dependencies
- every vector in \(\mathbb{R}^n\) has a unique expansion in a basis
- a linear map must preserve linear combinations
4 Strategy Before Details
There are two parts.
Existence: define the map by expanding an arbitrary vector in the basis and replacing each basis vector by its target image.Uniqueness: any linear map with the same basis images must agree on every basis expansion.
5 Full Proof
Proof. Let \(x \in \mathbb{R}^n\) be arbitrary. Because \(e_1,\dots,e_n\) is a basis, there are unique scalars \(a_1,\dots,a_n\) such that
\[ x = a_1 e_1 + \cdots + a_n e_n. \]
Define
\[ T(x) = a_1 w_1 + \cdots + a_n w_n. \]
This is well defined because the coefficients \(a_1,\dots,a_n\) are unique.
Now check linearity. Let
\[ x = \sum_{j=1}^n a_j e_j, \qquad y = \sum_{j=1}^n b_j e_j. \]
Then
\[ x+y = \sum_{j=1}^n (a_j+b_j)e_j, \]
so
\[ T(x+y) = \sum_{j=1}^n (a_j+b_j)w_j = \sum_{j=1}^n a_j w_j + \sum_{j=1}^n b_j w_j = T(x)+T(y). \]
Similarly, for any scalar \(c\),
\[ cx = \sum_{j=1}^n (ca_j)e_j, \]
and therefore
\[ T(cx) = \sum_{j=1}^n (ca_j)w_j = c \sum_{j=1}^n a_j w_j = cT(x). \]
So \(T\) is linear.
Also, if we apply the definition to a basis vector \(e_i\), whose coefficient expansion has \(1\) in the \(i\)th position and \(0\) elsewhere, we get
\[ T(e_i) = w_i. \]
This proves existence.
For uniqueness, let \(S : \mathbb{R}^n \to \mathbb{R}^m\) be another linear map with \(S(e_j)=w_j\) for all \(j\). If
\[ x = a_1 e_1 + \cdots + a_n e_n, \]
then by linearity,
\[ S(x) = a_1 S(e_1) + \cdots + a_n S(e_n) = a_1 w_1 + \cdots + a_n w_n = T(x). \]
So \(S(x)=T(x)\) for every \(x\), hence \(S=T\).
6 Step Annotations
- The basis matters because it gives a unique coordinate expansion.
- Existence works by replacing coordinates in the domain basis with prescribed target vectors.
- Uniqueness works because every vector is built from the basis, and linearity forces the map to respect that construction.
7 Why The Assumptions Matter
- If the list is not a basis, coordinates may fail to exist or fail to be unique, and the construction can break.
- Linearity is essential. Nonlinear maps are not determined by basis images alone.
8 Common Failure Modes
- forgetting to justify why the definition is well defined
- assuming the result also holds for arbitrary generating sets
- using linearity before proving the constructed map is linear
9 Reusable Pattern
This proof teaches the construction:
define on a basis -> extend by linearity -> use uniqueness of coordinates
That same pattern appears in linear operator design, Fourier-style expansions, and later basis changes.
10 Where This Shows Up Again
- Matrices and Linear Maps
- any argument that constructs a matrix from column images
- operator definitions in control, numerics, and ML layers
11 Exercises
- Use the theorem to build the matrix of a linear map once the images of \(e_1,e_2,e_3\) are given.
- Explain why the theorem fails if the chosen list of vectors is linearly dependent.
- Adapt the proof from the standard basis to an arbitrary basis.
12 Sources and Further Reading
- MIT 18.06SC: Linear Transformations and their Matrices -
First pass- official source for the basis-image viewpoint behind matrices. Checked2026-04-24. - Hefferon, Linear Algebra -
Second pass- stronger proof-level treatment of linear maps and coordinate representation. Checked2026-04-24. - Deep learning, transformers and graph neural networks: a linear algebra perspective -
Paper bridge- current survey showing why learned operators still reduce to matrix maps after coordinates are chosen. Checked2026-04-24.