Vector Spaces, Bases and Linear Transformations
CSIR NET linear algebra: vector spaces, subspaces, basis, dimension tables, rank-nullity, matrices of linear maps, counting over finite fields.
- 23 min
On this page · 12 sections
Prerequisites
- Matrix multiplication, row reduction and the idea of a field (\( \mathbb{Q} \), \( \mathbb{R} \), \( \mathbb{C} \), \( \mathbb{Z}_p \)). Nothing beyond a BSc course.
- Real Numbers, Sets and Countability: needed only for the remark that \( \mathbb{R} \) over \( \mathbb{Q} \) has an uncountable basis.
- Metric Spaces and Several Variables uses normed linear spaces; this page supplies the linear half of that idea.
The Hook: Two Functions Are Enough
Vector Spaces and Subspaces
The field matters. \( \mathbb{C} \) is a 1-dimensional space over \( \mathbb{C} \), a 2-dimensional space over \( \mathbb{R} \) (basis \( 1, i \)) and an infinite-dimensional space over \( \mathbb{Q} \). Always read which field the question names.
| Set | Subspace? | Why |
|---|---|---|
| \( \{ x \in \mathbb{R}^n : Ax = 0 \} \) | yes | kernel of a linear map |
| \( \{ x \in \mathbb{R}^n : Ax = b \} \), \( b \ne 0 \) | no | does not contain 0 (it is a translate of a subspace) |
| \( \{ (x, y) : xy = 0 \} \) (the two axes) | no | \( (1, 0) + (0, 1) \notin \) |
| first quadrant \( \{ x, y \ge 0 \} \) | no | not closed under \( -1 \cdot \) |
| invertible \( n \times n \) matrices | no | no 0; \( I + (-I) = 0 \) |
| singular \( n \times n \) matrices (\( n \ge 2 \)) | no | \( E_{11} + (I - E_{11}) = I \) |
| polynomials with \( p(1) = 0 \) | yes | kernel of evaluation at 1 |
| polynomials of degree exactly \( n \) | no | \( x^n + (-x^n) = 0 \) |
| \( \{ f \in C[0, 1] : f(0) = 2f(1) \} \) | yes | kernel of \( f \mapsto f(0) - 2f(1) \) |
Sums, intersections and unions
- \( U \cap W \) is always a subspace. \( U + W = \{ u + w \} \) is the smallest subspace containing both.
- \( U \cup W \) is a subspace iff \( U \subseteq W \) or \( W \subseteq U \).
- A vector space over an infinite field is never a finite union of proper subspaces.
- The sum is direct, \( V = U \oplus W \), when \( U + W = V \) and \( U \cap W = \{ 0 \} \); then every \( v \) is uniquely \( u + w \).
This gives the standard lower bound: in an \( n \)-dimensional space, \( \dim(U \cap W) \ge \dim U + \dim W - n \). Two 2-dimensional subspaces (planes through 0) of \( \mathbb{R}^3 \) always meet in at least a line.
Linear Independence, Basis and Dimension
The key fact behind “same size” is the replacement (exchange) lemma: if \( k \) vectors span \( V \), then any independent set in \( V \) has at most \( k \) vectors. Consequences you use every day, for \( \dim V = n \):
- any \( n + 1 \) vectors are dependent;
- any \( n \) independent vectors form a basis, and any \( n \) spanning vectors form a basis (you only need to check one condition when the count is right);
- every independent set extends to a basis, and every spanning set contains a basis;
- every subspace \( W \) has a complement \( U \) with \( V = W \oplus U \) (not unique).
Testing independence in \( F^n \): put the vectors as columns of a matrix. They are independent iff the rank equals the number of vectors. For \( n \) vectors in \( F^n \), iff the determinant is non-zero.
Dimensions to know without thinking
| Space (over \( \mathbb{R} \) unless stated) | Dimension | Basis or reason |
|---|---|---|
| \( F^n \) | \( n \) | standard basis \( e_1, \dots, e_n \) |
| \( M_{m \times n}(F) \) | \( mn \) | matrix units \( E_{ij} \) |
| \( P_n \): polynomials of degree \( \le n \) | \( n + 1 \) | \( 1, x, \dots, x^n \) |
| \( \mathbb{C}^n \) over \( \mathbb{R} \) | \( 2n \) | \( e_j, ie_j \) |
| symmetric \( n \times n \) | \( \frac{n(n+1)}{2} \) | entries on and above the diagonal |
| skew-symmetric \( n \times n \) | \( \frac{n(n-1)}{2} \) | entries strictly above the diagonal |
| upper triangular \( n \times n \) | \( \frac{n(n+1)}{2} \) | same count as symmetric |
| diagonal \( n \times n \) | \( n \) | \( E_{11}, \dots, E_{nn} \) |
| trace-zero \( n \times n \) | \( n^2 - 1 \) | kernel of the onto map \( \operatorname{tr} \) |
| all row sums zero | \( n^2 - n \) | each row loses one free entry |
| Hermitian \( n \times n \) complex, over \( \mathbb{R} \) | \( n^2 \) | \( n \) real diagonal entries + \( 2 \cdot \frac{n(n-1)}{2} \) real numbers above |
| matrices commuting with \( \operatorname{diag}(d_1, \dots, d_n) \), \( d_i \) distinct | \( n \) | exactly the diagonal matrices |
| solutions of an order-\( n \) linear homogeneous ODE | \( n \) | fundamental set of solutions |
| \( \mathbb{R} \) over \( \mathbb{Q} \); \( C[a, b] \); \( F[x] \) | infinite | \( \mathbb{R} \) over \( \mathbb{Q} \) even needs an uncountable basis |
Since \( M_n(\mathbb{R}) = \text{Sym} \oplus \text{Skew} \) via \( A = \frac{A + A^T}{2} + \frac{A - A^T}{2} \), the two dimensions add up: \( \frac{n(n+1)}{2} + \frac{n(n-1)}{2} = n^2 \). Use the sum as a check.
Adding linear conditions. Each independent linear condition cuts the dimension by one. Symmetric \( 3 \times 3 \) matrices with trace zero: \( 6 - 1 = 5 \). Polynomials in \( P_4 \) vanishing at \( 1 \) and \( -1 \): \( 5 - 2 = 3 \). The conditions must be independent; “\( p(1) = 0 \) and \( 2p(1) = 0 \)” is one condition, not two.
Quotient spaces
For a subspace \( W \subseteq V \), the cosets \( v + W \) form the quotient space \( V/W \), and
$$\dim V/W = \dim V - \dim W \quad (\dim V < \infty).$$If \( v_1, \dots, v_k \) is a basis of \( W \) extended to a basis \( v_1, \dots, v_n \) of \( V \), the cosets of \( v_{k+1}, \dots, v_n \) form a basis of \( V/W \).
Counting over a finite field
Over \( F = \mathbb{F}_q \) (\( q \) a prime power), an \( n \)-dimensional space has exactly \( q^n \) vectors, so counting questions become arithmetic.
| Count | Formula | Example, \( q = 2 \), \( n = 3 \) |
|---|---|---|
| vectors in \( \mathbb{F}_q^n \) | \( q^n \) | 8 |
| ordered bases of \( \mathbb{F}_q^n \) \( = \lvert GL_n(\mathbb{F}_q) \rvert \) | \( \prod_{i=0}^{n-1} (q^n - q^i) \) | \( 7 \cdot 6 \cdot 4 = 168 \) |
| \( k \)-dimensional subspaces | \( \prod_{i=0}^{k-1} \frac{q^n - q^i}{q^k - q^i} \) | \( k = 1 \): 7; \( k = 2 \): 7 |
| 1-dimensional subspaces | \( \frac{q^n - 1}{q - 1} \) | 7 |
The reasoning for \( GL_n \): the first column can be any non-zero vector (\( q^n - 1 \) choices), the second anything outside the span of the first (\( q^n - q \)), and so on. A \( k \)-dimensional subspace is counted by its ordered bases, divided by the number of ordered bases of a \( k \)-dimensional space. So \( \lvert GL_2(\mathbb{F}_3) \rvert = (9 - 1)(9 - 3) = 48 \), and \( \mathbb{F}_3^3 \) has \( \frac{27 - 1}{3 - 1} = 13 \) lines through the origin.
Linear Transformations
If \( V \) is finite-dimensional and \( T : V \to W \) is linear, then
$$\operatorname{rank} T + \operatorname{nullity} T = \dim V.$$What it forces, for finite-dimensional \( V, W \):
| Situation | Conclusion |
|---|---|
| \( \dim V > \dim W \) | \( T \) is not injective (\( \ker T \ne 0 \)) |
| \( \dim V < \dim W \) | \( T \) is not surjective |
| \( \dim V = \dim W \) | injective \( \iff \) surjective \( \iff \) bijective |
| \( T \) injective | \( T \) maps independent sets to independent sets |
| \( T \) surjective onto \( W \) | \( \operatorname{nullity} T = \dim V - \dim W \) |
Facts that go with it.
- A linear map is fixed by its values on a basis, and any choice of values extends to a unique linear map.
- \( V \cong W \iff \dim V = \dim W \) (finite-dimensional, same field). Every \( n \)-dimensional space is isomorphic to \( F^n \).
- \( \dim \mathcal{L}(V, W) = \dim V \cdot \dim W \). In particular the dual space \( V^* = \mathcal{L}(V, F) \) has \( \dim V^* = \dim V \).
- A non-zero linear functional \( f : V \to F \) is onto, so \( \ker f \) is a hyperplane of dimension \( n - 1 \).
- For a subspace \( W \) of \( V \), the annihilator \( W^0 = \{ f \in V^* : f(W) = 0 \} \) has \( \dim W^0 = n - \dim W \).
The matrix of a linear map
Fix ordered bases \( B = (v_1, \dots, v_n) \) of \( V \) and \( C = (w_1, \dots, w_m) \) of \( W \). The matrix \( [T]_{C \leftarrow B} \) is the \( m \times n \) matrix whose \( j \)-th column holds the \( C \)-coordinates of \( T(v_j) \). Then
$$[T(v)]_C = [T]_{C \leftarrow B}\,[v]_B, \qquad [S \circ T] = [S]\,[T].$$Rank, nullity, invertibility and (for \( V = W \), same basis) trace, determinant and eigenvalues of \( T \) can all be read from any matrix of \( T \).
Worked example: differentiation. \( D : P_3 \to P_3 \), basis \( 1, x, x^2, x^3 \). \( D(1) = 0 \), \( D(x) = 1 \), \( D(x^2) = 2x \), \( D(x^3) = 3x^2 \):
$$[D] = \begin{pmatrix} 0 & 1 & 0 & 0 \\ 0 & 0 & 2 & 0 \\ 0 & 0 & 0 & 3 \\ 0 & 0 & 0 & 0 \end{pmatrix}.$$Rank 3, nullity 1 (the constants), \( D^4 = 0 \) but \( D^3 \ne 0 \): \( D \) is nilpotent of index 4, its only eigenvalue is 0, and it is not diagonalisable. On \( P_n \) the same holds with index \( n + 1 \).
Worked example: transpose. \( T(X) = X^T \) on \( M_n(\mathbb{R}) \). Since \( T^2 = I \), the eigenvalues are \( \pm 1 \): symmetric matrices form the \( +1 \) eigenspace (dimension \( \frac{n(n+1)}{2} \)), skew-symmetric the \( -1 \) eigenspace (dimension \( \frac{n(n-1)}{2} \)). So \( T \) is diagonalisable, \( \operatorname{tr} T = \frac{n(n+1)}{2} - \frac{n(n-1)}{2} = n \) and \( \det T = (-1)^{n(n-1)/2} \).
Worked example: a commutator map. \( T(X) = AX - XA \) with \( A = \operatorname{diag}(a_1, \dots, a_n) \). On matrix units, \( T(E_{ij}) = (a_i - a_j)E_{ij} \). So \( T \) is diagonal in the basis \( \{ E_{ij} \} \), its kernel is spanned by the \( E_{ij} \) with \( a_i = a_j \), and if the \( a_i \) are distinct the kernel is exactly the diagonal matrices (dimension \( n \)) and \( \operatorname{rank} T = n^2 - n \).
Projections and nilpotent maps
- A projection satisfies \( P^2 = P \). Then \( V = \operatorname{im} P \oplus \ker P \), \( P \) acts as the identity on its image, and \( I - P \) is the complementary projection.
- A nilpotent map satisfies \( T^k = 0 \) for some \( k \). On an \( n \)-dimensional space the index is at most \( n \), so \( T^n = 0 \).
- If \( T^2 = 0 \) then \( \operatorname{im} T \subseteq \ker T \), so \( \operatorname{rank} T \le \operatorname{nullity} T \) and \( \operatorname{rank} T \le \frac{n}{2} \).
- If \( T \) is nilpotent, \( I + T \) is invertible: \( (I + T)^{-1} = I - T + T^2 - \dots \pm T^{k-1} \).
graph TD
A["Is the set a subspace?"] --> B{"Defined by linear equations
with zero right side?"}
B -->|yes| C["Subspace: dim = ambient dim
minus independent conditions"]
B -->|no| D{"Contains 0 and closed
under + and scaling?"}
D -->|no| E["Not a subspace: show one failure"]
D -->|yes| F["Subspace: find a basis directly"]Memory Tricks
- Sym + Skew = Square. \( \frac{n(n+1)}{2} + \frac{n(n-1)}{2} = n^2 \). Upper triangular has the symmetric count.
- Rank–nullity is a see-saw. The dimension of \( V \) is fixed; whatever the image gains, the kernel loses.
- Right count, one check. \( n \) vectors in an \( n \)-dimensional space: independence alone or spanning alone is enough.
- GL counts columns. \( q^n - 1, q^n - q, q^n - q^2, \dots \): each new column avoids the span of the earlier ones.
- Zero on the right means subspace. Homogeneous linear conditions give subspaces; a constant on the right gives a translate.
Don’t Confuse
| This | Is not this | The difference |
|---|---|---|
| \( U + W \) | \( U \cup W \) | the sum is always a subspace; the union only when one contains the other |
| \( \dim(U + W) \) | \( \dim U + \dim W \) | subtract \( \dim(U \cap W) \); equality only for direct sums |
| Hermitian matrices over \( \mathbb{R} \) (dim \( n^2 \)) | a complex subspace | not closed under multiplication by \( i \) |
| injective \( \iff \) surjective | true in infinite dimensions | needs \( \dim V = \dim W < \infty \) |
| \( \mathbb{C} \) over \( \mathbb{R} \) (dim 2) | \( \mathbb{C} \) over \( \mathbb{C} \) (dim 1) | the field changes the dimension |
| complement of a subspace | set complement | a complement \( U \) with \( V = W \oplus U \) is a subspace and is not unique |
| polynomials of degree \( \le n \) | of degree exactly \( n \) | only the first is a subspace |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper. In Level 3, as in Part C, one or more options may be correct.
Level 1: Foundation
- What is the dimension of the space of real symmetric \( 4 \times 4 \) matrices?
- What is \( \dim_{\mathbb{R}} \mathbb{C}^3 \)?
- A linear map \( T : \mathbb{R}^5 \to \mathbb{R}^3 \) is onto. What is its nullity?
Answers
- 10. \( \frac{n(n+1)}{2} = \frac{4 \cdot 5}{2} \).
- 6. Each complex coordinate contributes two real ones.
- 2. Rank 3 (onto), so nullity \( = 5 - 3 \).
Level 2: Part B Style
Practice Q1. Let \( W = \{ A \in M_3(\mathbb{R}) : A = A^T, \ \operatorname{tr} A = 0 \} \). Then \( \dim W \) is:
(a) 4 (b) 5 (c) 6 (d) 8
Answer: (b)
Symmetric \( 3 \times 3 \) matrices form a 6-dimensional space. The trace is a non-zero linear functional on it (\( \operatorname{tr} E_{11} = 1 \)), so its kernel has dimension \( 6 - 1 = 5 \). Option (d) is the trace-zero space without symmetry (\( 9 - 1 \)).
Practice Q2. Let \( U \) and \( W \) be subspaces of \( \mathbb{R}^7 \) with \( \dim U = 4 \) and \( \dim W = 5 \). The smallest possible value of \( \dim(U \cap W) \) is:
(a) 0 (b) 1 (c) 2 (d) 3
Answer: (c)
\( \dim(U \cap W) = \dim U + \dim W - \dim(U + W) \ge 4 + 5 - 7 = 2 \). It is attained: take \( U = \operatorname{span}(e_1, \dots, e_4) \), \( W = \operatorname{span}(e_3, \dots, e_7) \), which meet in \( \operatorname{span}(e_3, e_4) \).
Practice Q3. Let \( P_3 \) be the real polynomials of degree at most 3 and \( T(p) = p' + p'' \). The rank of \( T \) is:
(a) 1 (b) 2 (c) 3 (d) 4
Answer: (c)
\( T(1) = 0 \), \( T(x) = 1 \), \( T(x^2) = 2x + 2 \), \( T(x^3) = 3x^2 + 6x \). The last three have degrees 0, 1, 2, so they are independent: rank 3, nullity 1 (the constants). Structurally, \( T = D(I + D) \) and \( I + D \) is invertible because \( D \) is nilpotent, so \( \operatorname{rank} T = \operatorname{rank} D = 3 \).
Practice Q4. The number of invertible \( 2 \times 2 \) matrices over \( \mathbb{Z}_3 \) is:
(a) 24 (b) 36 (c) 48 (d) 81
Answer: (c)
First column: any non-zero vector of \( \mathbb{Z}_3^2 \), \( 9 - 1 = 8 \) choices. Second column: outside the line spanned by the first, \( 9 - 3 = 6 \) choices. \( 8 \times 6 = 48 \). (81 is the number of all \( 2 \times 2 \) matrices.)
Level 3: Part C Style
Practice Q5. Let \( A = \begin{pmatrix} 1 & 0 \\ 0 & 2 \end{pmatrix} \) and \( T : M_2(\mathbb{R}) \to M_2(\mathbb{R}) \), \( T(X) = AX - XA \). Which statements are true?
(a) \( T \) is linear. (b) \( \operatorname{rank} T = 2 \). (c) \( \ker T \) is the space of diagonal matrices. (d) \( T \) is nilpotent.
Answer: (a), (b) and (c)
- (a) True. \( T(aX + bY) = A(aX + bY) - (aX + bY)A = aT(X) + bT(Y) \).
- (b) True. \( T(E_{ij}) = (a_i - a_j)E_{ij} \) with \( a_1 = 1, a_2 = 2 \): \( T(E_{11}) = T(E_{22}) = 0 \), \( T(E_{12}) = -E_{12} \), \( T(E_{21}) = E_{21} \). The image is spanned by \( E_{12}, E_{21} \).
- (c) True. The kernel is spanned by \( E_{11}, E_{22} \): exactly the diagonal matrices (dimension 2, consistent with rank–nullity \( 2 + 2 = 4 \)).
- (d) False. \( T \) has eigenvalues \( 0, 0, -1, 1 \); a nilpotent map has only the eigenvalue 0. Indeed \( T^k(E_{12}) = (-1)^k E_{12} \ne 0 \).
Practice Q6. Let \( V \) be the real polynomials of degree at most 4 and \( W = \{ p \in V : p(1) = p(-1) = 0 \} \). Which statements are true?
(a) \( \dim W = 3 \). (b) \( \{ x^2 - 1, \ x(x^2 - 1), \ x^2(x^2 - 1) \} \) is a basis of \( W \). (c) \( \dim V/W = 3 \). (d) The even polynomials in \( W \) form a 1-dimensional subspace.
Answer: (a) and (b)
- (a) True. \( \dim V = 5 \). The functionals \( p \mapsto p(1) \) and \( p \mapsto p(-1) \) are independent (on \( 1 \) and \( x \) they give the independent vectors \( (1, 1) \), \( (1, -1) \)), so \( \dim W = 5 - 2 = 3 \).
- (b) True. Every \( p \in W \) is \( (x^2 - 1)q(x) \) with \( \deg q \le 2 \), and \( q \) ranges over \( P_2 \) with basis \( 1, x, x^2 \). The three products have different degrees, so they are independent.
- (c) False. \( \dim V/W = 5 - 3 = 2 \).
- (d) False. Even polynomials in \( V \) are spanned by \( 1, x^2, x^4 \); for them \( p(-1) = p(1) \), so only one condition remains and the even part of \( W \) has dimension \( 3 - 1 = 2 \) (basis \( x^2 - 1 \), \( x^4 - 1 \)).
Practice Q7. Let \( T : \mathbb{R}^4 \to \mathbb{R}^4 \) be linear with \( T^2 = 0 \). Which statements are necessarily true?
(a) \( \operatorname{rank} T \le 2 \). (b) \( \operatorname{im} T \subseteq \ker T \). (c) \( T \) is diagonalisable. (d) \( I + T \) is invertible.
Answer: (a), (b) and (d)
- (b) True. If \( y = T(x) \) then \( T(y) = T^2(x) = 0 \).
- (a) True. From (b), \( \operatorname{rank} T \le \operatorname{nullity} T = 4 - \operatorname{rank} T \), so \( \operatorname{rank} T \le 2 \).
- (c) False in general. The only eigenvalue is 0, so a diagonalisable \( T \) would be the zero map. \( T = E_{12} \) (sending \( e_2 \mapsto e_1 \), everything else to 0) satisfies \( T^2 = 0 \) and is not diagonalisable. (\( T = 0 \) is diagonalisable, which is why the option fails as “necessarily”, not always.)
- (d) True. \( (I + T)(I - T) = I - T^2 = I \).
Quick Revision Table
| Idea | Fact |
|---|---|
| Subspace | non-empty, closed under linear combinations; kernels are subspaces |
| Union | subspace iff one contains the other |
| Sum formula | \( \dim(U + W) = \dim U + \dim W - \dim(U \cap W) \) |
| Basis with the right count | independence alone or spanning alone suffices |
| Standard dimensions | Sym \( \frac{n(n+1)}{2} \), Skew \( \frac{n(n-1)}{2} \), trace-zero \( n^2 - 1 \), \( P_n \): \( n + 1 \), Hermitian over \( \mathbb{R} \): \( n^2 \) |
| Quotient | \( \dim V/W = \dim V - \dim W \) |
| Finite fields | \( \lvert GL_n(\mathbb{F}_q) \rvert = \prod (q^n - q^i) \); lines: \( \frac{q^n - 1}{q - 1} \) |
| Rank–nullity | rank + nullity = dim of the domain |
| Equal finite dimensions | injective \( \iff \) surjective |
| Matrix of \( T \) | column \( j \) = coordinates of \( T(v_j) \) |
| \( T^2 = 0 \) | \( \operatorname{im} T \subseteq \ker T \), rank \( \le n/2 \) |
| Nilpotent \( N \) | \( I + N \) invertible; \( N^n = 0 \) |
Key Takeaways
What’s Next
- Rank, Determinants and Linear Equations: computing the rank of a matrix and deciding when \( Ax = b \) is solvable.
- Eigenvalues and the Cayley–Hamilton Theorem: the invariants of a map from a space to itself.
- Linear Algebra chapter overview and the Linear Algebra Quick Revision.
- Normed linear spaces, where linear algebra meets analysis: Metric Spaces and Several Variables.