Exchange Argument and Why Dimension Is Well Defined
proof, basis, dimension, exchange argument
1 Claim
The whole idea of dimension depends on one theorem:
different bases of the same finite-dimensional space must have the same size.
Theorem 1 (Theorem) Let \(W\) be a finite-dimensional subspace.
If \(v_1,\dots,v_k\) and \(w_1,\dots,w_m\) are both bases of \(W\), then
\[ k = m. \]
So the number of vectors in a basis of \(W\) is well defined.
2 Why This Proof Matters
Without this theorem, dimension would just be a guess depending on which basis you wrote down.
This proof matters because it also teaches the exchange style that later appears in:
- rank arguments
- basis reduction
- greedy comparisons between independent and spanning sets
3 Dependencies
- a basis both spans and is linearly independent
- every vector in \(W\) can be written as a linear combination of basis vectors
4 Strategy Before Details
The idea is to compare one basis against another using a counting principle:
- an independent list cannot be longer than a spanning list for the same space
Apply that principle twice, once in each direction.
5 Full Proof
Proof. We first prove a lemma.
5.1 Lemma
If \(u_1,\dots,u_r\) is a linearly independent list in \(W\) and \(s_1,\dots,s_t\) spans \(W\), then
\[ r \le t. \]
5.2 Proof of the lemma
We use induction on \(r\).
If \(r=0\), the statement is trivial.
Now assume it is true for independent lists of length \(r-1\), and suppose \(u_1,\dots,u_r\) is linearly independent while \(s_1,\dots,s_t\) spans \(W\).
Because the spanning list spans \(W\), we can write
\[ u_r = a_1 s_1 + \cdots + a_t s_t. \]
At least one coefficient is nonzero; relabel so that \(a_t \neq 0\). Then
\[ s_t = \frac{1}{a_t}u_r - \sum_{j=1}^{t-1}\frac{a_j}{a_t}s_j. \]
So the list
\[ s_1,\dots,s_{t-1},u_r \]
still spans \(W\).
Now \(u_1,\dots,u_{r-1}\) is linearly independent, and it lies in the span of
\[ s_1,\dots,s_{t-1},u_r. \]
By the induction hypothesis,
\[ r-1 \le t-1. \]
Therefore \(r \le t\), proving the lemma.
Now return to the theorem.
Since \(v_1,\dots,v_k\) is linearly independent and \(w_1,\dots,w_m\) spans \(W\), the lemma gives
\[ k \le m. \]
Since \(w_1,\dots,w_m\) is linearly independent and \(v_1,\dots,v_k\) spans \(W\), the same lemma gives
\[ m \le k. \]
Hence \(k=m\).
6 Step Annotations
- The key move is replacing one spanning vector by one independent vector without losing span.
- The induction counts how many independent directions can fit inside a spanning list.
- Running the argument in both directions forces equality.
7 Why The Assumptions Matter
- Finite dimensionality matters because the counting argument needs finite list lengths.
- The theorem compares bases of the
samesubspace. Different subspaces can of course have different basis sizes.
8 Common Failure Modes
- assuming the result is obvious because two bases “feel minimal”
- replacing a spanning vector without checking the new list still spans
- proving only one inequality and forgetting the reverse direction
9 Reusable Pattern
The exchange argument teaches a durable habit:
compare independence against spanning by swapping one vector at a time.
That same structure appears in rank proofs and basis simplification.
10 Where This Shows Up Again
- Subspaces, Basis, and Dimension
- rank and nullity arguments later in linear algebra
- low-dimensional approximation methods that rely on basis size as a real invariant
11 Exercises
- Show that any linearly independent list in a \(d\)-dimensional space has length at most \(d\).
- Show that any spanning list in a \(d\)-dimensional space has length at least \(d\).
- Apply the theorem to explain why the column space of a rank-\(r\) matrix cannot have a basis with more or fewer than \(r\) vectors.
12 Sources and Further Reading
- MIT 18.06SC: Basis and Dimension -
First pass- official source for the span, basis, and dimension chain. Checked2026-04-24. - Hefferon, Linear Algebra -
Second pass- stronger proof-based treatment of basis exchange ideas. Checked2026-04-24. - A Survey: Potential Dimensionality Reduction Methods -
Paper bridge- modern reminder that low-dimensional structure only makes sense because dimension is a real invariant. Checked2026-04-24.