physics2u
PhysicsMathsChemistryBiology
Tier
⌕ Search ⌘K
Derivation

Invariance of Dimension of a Vector Space

D-074 Home PU-104 Threads symmetry Depends on —
Statement

Let \(V\) be a vector space over a field \(\mathbb{F}\) that is spanned by a finite set. Then any two bases of \(V\) are finite and have exactly the same number of elements. Consequently the integer \(\dim V\), defined as the cardinality of any basis, is a well-defined invariant of \(V\) that does not depend on the choice of basis.

Why it matters

Dimension is the single integer that classifies a finite-dimensional vector space up to isomorphism: two spaces over the same field are isomorphic if and only if they share a dimension. If the number of basis vectors could depend on which basis one chose, the whole apparatus of coordinates, rank, and degrees of freedom would collapse and "three-dimensional space" would carry no intrinsic meaning.

In physics this invariant underwrites counting arguments everywhere: the number of independent components of a tensor, the number of normal modes of a coupled system, the dimension of a Hilbert space of states, and the number of generators of a symmetry group. Each is a dimension, and each is meaningful only because it does not depend on the frame or coordinates used to compute it.

Assumptions
The field \(\mathbb{F}\) is a genuine field, so every nonzero scalar is invertible.Over a mere ring, e.g. \(\mathbb{Z}\), free modules can still have well-defined rank, but the pivoting step that divides by a leading coefficient fails and the elementary Steinitz exchange argument below no longer goes through unchanged.
\(V\) is finitely generated: some finite set spans \(V\).If \(V\) needs infinitely many vectors to span, e.g. the space of all polynomials, it is infinite-dimensional; the theorem still holds via a cardinality argument using Zorn's lemma, but the finite exchange counting below does not apply directly.
A basis is by definition both linearly independent and spanning.If one drops independence, a "basis" could be padded with redundant vectors and its size would be arbitrary; if one drops spanning, it could not express every vector. Both halves are needed for the exchange lemma to pin the count.
The vector-space axioms (associativity, distributivity, scalar action) hold exactly.Without distributivity the manipulation of linear combinations in the exchange step is invalid, and independence relations cannot be rearranged to solve for one vector in terms of the others.
Derivation

The engine is the Steinitz Exchange Lemma: if one set is linearly independent and another spans \(V\), then the independent set is no larger than the spanning set. Applying it twice, symmetrically, forces equality of the two basis sizes.

1
\[ B = \{v_1,\dots,v_n\}\ \text{spans } V, \qquad I = \{w_1,\dots,w_m\}\ \text{is linearly independent in } V. \]
Set up the two hypotheses of the exchange lemma; the goal is to show \(m \le n\). A
2
\[ w_1 = \sum_{i=1}^{n} a_i\, v_i, \qquad a_i \in \mathbb{F}. \]
Because \(B\) spans, \(w_1\) is a linear combination of the \(v_i\). A
3
\[ \text{some } a_k \neq 0 \quad(\text{else } w_1 = 0,\ \text{contradicting independence of } I). \]
A linearly independent set cannot contain the zero vector. A
4
\[ v_k = \frac{1}{a_k}\left( w_1 - \sum_{i\neq k} a_i\, v_i \right). \]
Solve for \(v_k\); legal precisely because \(a_k\neq 0\) is invertible in the field \(\mathbb{F}\). B
5
\[ B_1 = \{\,w_1,\ v_1,\dots,\widehat{v_k},\dots,v_n\,\}\ \text{spans } V. \]
Exchange: \(v_k\) is now expressible through \(w_1\) and the remaining \(v_i\), so anything the old set reached the new one reaches too. Still \(n\) vectors. B
6
\[ w_2 = b\, w_1 + \sum_{i\neq k} c_i\, v_i, \qquad \text{some } c_i \neq 0. \]
\(B_1\) spans, so \(w_2\) is a combination of it; some coefficient on a remaining \(v_i\) must be nonzero, otherwise \(w_2 = b\,w_1\), contradicting independence of \(\{w_1,w_2\}\). B
7
\[ \text{Induct: after } j \text{ steps},\ B_j = \{w_1,\dots,w_j\}\cup\{\text{some } n-j \text{ of the } v_i\}\ \text{spans } V. \]
At each stage a nonzero coefficient on a surviving \(v_i\) exists, by independence of the \(w\)'s, so we swap in \(w_{j+1}\) and remove one \(v_i\), keeping the count at \(n\). C
8
\[ \text{The } v_i \text{ cannot run out before the } w_i:\qquad m \le n. \]
If \(m>n\), after \(n\) swaps every \(v_i\) is gone yet \(w_{n+1}\) remains and would be a combination of \(\{w_1,\dots,w_n\}\), contradicting independence of \(I\). Hence \(m\le n\). This is the Exchange Lemma. C
9
\[ \text{Let } B \text{ and } B' \text{ be two bases}, \quad |B|=n,\ |B'|=n'. \]
Each basis is simultaneously independent and spanning, exactly the two roles the lemma needs. A
10
\[ B'\ \text{independent},\ B\ \text{spanning}\ \Rightarrow\ n' \le n; \qquad B\ \text{independent},\ B'\ \text{spanning}\ \Rightarrow\ n \le n'. \]
Apply the Exchange Lemma twice, swapping the roles of the two bases. B
11
\[ n' \le n \ \text{and}\ n \le n' \ \Longrightarrow\ n = n'. \]
Antisymmetry of \(\le\) on the integers. A
Result
\[ |B| = |B'| \equiv \dim V \quad\text{for every pair of bases } B, B'. \]

Reading. Every basis of a finitely generated vector space has the same number of elements. That common number is the dimension. It is an intrinsic property of \(V\): it survives any change of basis, any linear coordinate transformation, any choice of frame.

Units check. Dimension is a pure counting number: dimensionless and a non-negative integer. It is invariant under any invertible linear map \(T:V\to V\), which is the "unit" of a change of basis; both sides of \(n=n'\) are integers with no physical dimension attached.

Limiting cases
  • \(V=\{0\}\): the only basis is the empty set, \(\dim V = 0\); the exchange argument is vacuously satisfied.
  • \(V=\mathbb{F}^n\): the standard basis \(\{e_1,\dots,e_n\}\) gives \(\dim V = n\), and the theorem says every other basis also has exactly \(n\) vectors.
  • \(\mathbb{C}\) viewed over \(\mathbb{C}\) has dimension \(1\); the same set viewed over \(\mathbb{R}\) has dimension \(2\), so dimension is well-defined only once the field is fixed.
  • A single nonzero vector spans a line: \(\dim = 1\), and any other nonzero scalar multiple of it is an equally valid one-element basis.
Breaks when
  • Infinite-dimensional spaces. The finite swapping induction never terminates; one must instead invoke Zorn's lemma to guarantee a basis and compare cardinalities. The conclusion that all bases are equinumerous still holds, but the counting proof above does not.
  • Modules over a general ring. Over a ring that is not a field, division by a leading coefficient in Step 4 fails. For some pathological non-commutative rings the invariant-basis-number property genuinely fails and a free module can have bases of different sizes.
  • Changing the base field mid-argument. If the scalar field is enlarged or restricted, "linearly independent" changes meaning and the two counts refer to different structures; the equality \(n=n'\) holds only within a fixed \(\mathbb{F}\).
Failure modes
  • Confusing spanning with basis. Counting a spanning set, which may be too big, and calling that number the dimension; only an independent spanning set counts.
  • Forgetting independence of the exchanged set. In Step 6 the crucial point is that a nonzero coefficient sits on a remaining \(v_i\); assuming it sits on \(w_1\) breaks the induction.
  • Dividing by \(a_k\) without checking \(a_k\neq0\). Skipping Step 3 makes the pivot illegitimate.
  • Ignoring the base field. Reporting \(\dim\mathbb{C}=1\) when the problem intends \(\mathbb{R}\)-scalars, or vice versa.
  • Assuming the result for modules. Applying the counting proof to \(\mathbb{Z}\)-modules and concluding rank is always defined by this argument alone.
Discussion

This theorem is the reason "dimension" is a legitimate word. Its proof is entirely combinatorial: no inner product, no topology, no notion of length enters. That purity matters, because it means dimension is preserved by any isomorphism, since an isomorphism carries a basis to a basis; therefore two finite-dimensional spaces are isomorphic precisely when their dimensions agree. Dimension is thus a complete isomorphism invariant for finite-dimensional vector spaces.

The invariance is what makes the rank of a linear map well-defined, and with it the rank–nullity theorem, the notion of degrees of freedom of a mechanical system, and the counting of independent components of a tensor. When a physicist says a rigid body has six degrees of freedom, or that a spin-\(j\) representation is \((2j+1)\)-dimensional, the statement is coordinate-free only because of this invariance.

The Steinitz exchange is really the statement that independence and spanning are dual size constraints: independent sets are "small" (at most \(n\)) and spanning sets are "large" (at least \(n\)), and a basis is exactly where the two bounds meet. This min–max flavour recurs throughout linear algebra, from the definition of rank as both a maximal independent count and a minimal spanning count, to matroid theory where the exchange property is taken as an axiom.

The deeper structural reason is that a finitely generated free module over a commutative ring, with a field as the special case, has the invariant-basis-number property, provable by tensoring down to a residue field or by taking determinants: an invertible change-of-basis matrix must be square, forcing the two counts equal. Over non-commutative rings this can fail spectacularly, since there exist rings \(R\) with \(R \cong R^2\) as modules, which is exactly why the field hypothesis, or at least commutativity together with IBN, is load-bearing rather than cosmetic.

Common misconceptions. Dimension is not "how big" a space is in any metric sense: \(\mathbb{R}\) and a bounded open interval have the same cardinality of points, yet dimension counts independent directions, not points. Also, a basis of \(n\) vectors does not require the vectors to be orthogonal or normalized; orthogonality is extra structure irrelevant to the count.

Worked examples

Example 1 — Two different-looking bases of \(\mathbb{R}^2\) have the same size.

1
\[ B = \{(1,0),(0,1)\}, \qquad B' = \{(1,1),(1,-1)\}. \]
Two candidate bases of \(V=\mathbb{R}^2\); we verify \(B'\) is a basis and count. A
2
\[ \det\begin{pmatrix}1&1\\1&-1\end{pmatrix} = (1)(-1)-(1)(1) = -2 \neq 0. \]
Nonzero determinant \(\Rightarrow B'\) is independent and spanning, hence a basis. B
3
\[ |B| = 2, \qquad |B'| = 2. \]
Count elements; the invariance theorem guarantees these must match, and they do. A
\[ \dim \mathbb{R}^2 = 2. \]

Reading. A rotated and sheared basis has exactly as many vectors as the standard one. Units. Pure integer, dimensionless.

Example 2 — The exchange lemma bounds an independent set.

1
\[ V=\mathbb{R}^3,\quad B=\{e_1,e_2,e_3\}\ \text{spans},\quad I=\{w_1,w_2\},\ w_1=(1,2,0),\ w_2=(0,1,1). \]
Independent set \(I\) and spanning set \(B\); the lemma predicts \(|I|\le|B|\), i.e. \(2\le 3\). A
2
\[ w_1 = 1\cdot e_1 + 2\cdot e_2 + 0\cdot e_3,\qquad a_1=1\neq0. \]
Pivot on \(e_1\) and swap it out for \(w_1\). New spanning set \(\{w_1,e_2,e_3\}\). B
3
\[ w_2 = 0\cdot w_1 + 1\cdot e_2 + 1\cdot e_3,\qquad \text{coeff on } e_3 =1\neq0. \]
Nonzero coefficient on a surviving \(e_i\); swap \(e_3\) for \(w_2\). New set \(\{w_1,e_2,w_2\}\), still 3 vectors, spans \(V\). B
4
\[ \text{both } w\text{'s inserted, one } e_i\ (\text{namely } e_2)\ \text{remains}:\quad m=2 \le n=3.\ \checkmark \]
The independent set fit inside the spanning set with room to spare, as the lemma requires. C
\[ |I| = 2 \ \le\ 3 = |B|. \]

Reading. Any independent set in \(\mathbb{R}^3\) has at most 3 vectors; this bound, applied both ways to two bases, is what forces \(\dim=3\). Units. Counting integers.

Problems
  1. Show that \(\{(1,2),(3,6)\}\) is not a basis of \(\mathbb{R}^2\), and state the dimension of its span.
    Solution Since \((3,6)=3\cdot(1,2)\), the set is linearly dependent and cannot be a basis, because a basis must be independent. The span is the line \(\{t(1,2):t\in\mathbb{R}\}\), which has the one-element basis \(\{(1,2)\}\); hence \(\dim(\text{span})=1\), not \(2\). The two vectors, though two in number, do not certify dimension \(2\) because they are not independent.
  2. Compute \(\dim_{\mathbb{R}}\mathbb{C}\) and \(\dim_{\mathbb{C}}\mathbb{C}\), and explain the difference.
    Solution Over \(\mathbb{C}\): the single vector \(\{1\}\) spans \(\mathbb{C}\) since every \(z=z\cdot1\), and it is independent, so \(\dim_{\mathbb{C}}\mathbb{C}=1\). Over \(\mathbb{R}\): the set \(\{1,i\}\) is required, since \(z=x\cdot1+y\cdot i\) with \(x,y\in\mathbb{R}\), and \(\{1,i\}\) is \(\mathbb{R}\)-independent; so \(\dim_{\mathbb{R}}\mathbb{C}=2\). Dimension is well-defined only after fixing the scalar field; enlarging the field to \(\mathbb{C}\) halves the count here.
  3. Let \(P_2\) be real polynomials of degree \(\le 2\). Exhibit two different bases and confirm they have equal size.
    Solution Standard basis \(B=\{1,x,x^2\}\), size 3. Alternative \(B'=\{1,\ 1+x,\ 1+x+x^2\}\). To see \(B'\) is a basis, note the change-of-basis matrix whose columns express \(B'\) in terms of \(B\) is \(\begin{pmatrix}1&1&1\\0&1&1\\0&0&1\end{pmatrix}\), which is upper-triangular with determinant \(1\neq0\), hence invertible. So \(B'\) is independent and spanning. Both have size \(3\), so \(\dim P_2 = 3\), consistent with invariance.
  4. Suppose \(V\) is spanned by \(4\) vectors. What is the largest possible size of a linearly independent subset of \(V\)? Justify by the exchange lemma.
    Solution By the Steinitz exchange lemma, any independent set has size at most that of any spanning set. A spanning set of size \(4\) therefore caps every independent set at \(4\) vectors. So the largest possible independent subset has \(\le 4\) elements; if in addition \(V\) contains an independent set of exactly \(4\), then \(\dim V=4\). In general \(\dim V\le 4\).
  5. A physical system's configuration space is a real vector space. Two engineers coordinatize it linearly, one using \(6\) coordinates and one claiming \(5\). Show that one of them is wrong.
    Solution Each linear coordinate system corresponds to a basis of the configuration space, since \(k\) independent coordinates correspond to a basis of size \(k\). By invariance of dimension every basis has the same cardinality, so the number of independent coordinates is forced. Thus \(6\neq5\) is impossible for two genuine (independent and complete) coordinate systems on the same space. Either the \(5\)-coordinate set fails to span, missing a degree of freedom, or the \(6\)-coordinate set is dependent, carrying a redundant coordinate. The invariant \(\dim V\), the true number of degrees of freedom, settles the dispute.