Rank, Determinants and Linear Equations
CSIR NET linear algebra: row reduction, rank inequalities, determinant rules, adjugate, matrix determinant lemma, consistency of Ax = b and solution counts.
- 19 min
On this page · 12 sections
Prerequisites
- Vector Spaces, Bases and Linear Transformations: rank–nullity and the matrix of a linear map. Rank of a matrix is the rank of the map \( x \mapsto Ax \).
- Elementary row operations and echelon form, as in any BSc course.
- The arithmetic of Numerical Ability is enough for the hand computations; the work is in choosing the shortcut.
The Hook: Three Equations, Two Pieces of Information
Rank of a Matrix
Row operations do not change the row space; column operations do not change the column space; both preserve rank. So reduce to echelon form and count pivots.
Rank inequalities
For \( A \) of size \( m \times n \) and \( B \) of size \( n \times p \):
| Statement | Note |
|---|---|
| \( \operatorname{rank} A \le \min(m, n) \) | full rank means equality |
| \( \operatorname{rank}(AB) \le \min(\operatorname{rank} A, \operatorname{rank} B) \) | multiplying never increases rank |
| \( \operatorname{rank}(AB) \ge \operatorname{rank} A + \operatorname{rank} B - n \) | Sylvester’s inequality; \( n \) is the shared dimension |
| \( \operatorname{rank}(A + B) \le \operatorname{rank} A + \operatorname{rank} B \) | same size \( A, B \) |
| \( \operatorname{rank}(PAQ) = \operatorname{rank} A \) | \( P, Q \) invertible |
| \( \operatorname{rank}(A^TA) = \operatorname{rank}(AA^T) = \operatorname{rank} A \) | real \( A \); for complex use \( A^*A \) |
| \( \operatorname{rank}(A^T) = \operatorname{rank} A \) | row rank = column rank |
| \( \operatorname{rank}(uv^T) = 1 \) for non-zero \( u, v \) | every rank-1 matrix has this form |
Spotting low rank without reducing.
- \( a_{ij} = i + j \) (any size \( \ge 2 \)): each row is \( (i)(1, 1, \dots) + (1, 2, 3, \dots) \), a combination of two fixed rows, so rank 2.
- \( a_{ij} = ij \): \( uu^T \) with \( u = (1, 2, \dots, n) \), so rank 1.
- \( a_{ij} = \sin(i + j) \): \( \sin i \cos j + \cos i \sin j \), rank at most 2 (exactly 2 for \( n \ge 2 \)).
- The \( n \times n \) all-ones matrix \( J \) has rank 1; \( J - I \) is invertible for \( n \ge 2 \).
Worked example: rank with a parameter
For \( A = \begin{pmatrix} a & 1 & 1 \\ 1 & a & 1 \\ 1 & 1 & a \end{pmatrix} \), \( A = (a - 1)I + J \), and \( J \) has eigenvalues \( 3, 0, 0 \), so
$$\det A = (a + 2)(a - 1)^2.$$- \( a \ne 1, -2 \): rank 3.
- \( a = 1 \): \( A = J \), rank 1.
- \( a = -2 \): \( \det A = 0 \) but the \( 2 \times 2 \) minor \( \begin{vmatrix} -2 & 1 \\ 1 & -2 \end{vmatrix} = 3 \ne 0 \), so rank 2.
Determinants
Rules
| Rule | Formula |
|---|---|
| Product | \( \det(AB) = \det A \det B \) |
| Transpose | \( \det A^T = \det A \) |
| Scalar | \( \det(cA) = c^n \det A \) |
| Inverse | \( \det A^{-1} = \frac{1}{\det A} \) |
| Triangular / block triangular | product of the diagonal entries / of the diagonal blocks’ determinants |
| Row swap | changes the sign |
| Adding a multiple of a row to another | no change |
| Eigenvalues | \( \det A = \lambda_1 \lambda_2 \cdots \lambda_n \) (with multiplicity, over \( \mathbb{C} \)) |
| Similar matrices | same determinant |
| Invertibility | \( A \) invertible \( \iff \det A \ne 0 \) |
Determinants to know
| Matrix | Determinant |
|---|---|
| Vandermonde \( (x_i^{j-1})_{i,j} \) | \( \prod_{i < j} (x_j - x_i) \) |
| \( aI_n + bJ_n \) | \( a^{n-1}(a + nb) \) |
| \( I_n + J_n \) (2 on the diagonal, 1 elsewhere) | \( n + 1 \) |
| \( J_n - I_n \) (0 on the diagonal, 1 elsewhere) | \( (-1)^{n-1}(n - 1) \) |
| \( I_n + uv^T \) (matrix determinant lemma) | \( 1 + v^Tu \) |
| \( A + uv^T \), \( A \) invertible | \( \det A \,(1 + v^TA^{-1}u) \) |
| odd-order real skew-symmetric | 0 |
| real orthogonal | \( \pm 1 \) |
| permutation matrix of \( \sigma \) | \( \operatorname{sgn}(\sigma) \) |
The \( aI + bJ \) formula comes from the eigenvalues: \( J \) has eigenvalues \( n \) (once) and \( 0 \) (\( n - 1 \) times), so \( aI + bJ \) has eigenvalues \( a + nb \) and \( a \) (\( n - 1 \) times). Eigenvalue arguments are usually the fastest route to a structured determinant; see Eigenvalues and the Cayley–Hamilton Theorem.
The adjugate
The adjugate \( \operatorname{adj} A \) is the transpose of the cofactor matrix, and \( A \operatorname{adj} A = \operatorname{adj} A \cdot A = (\det A) I \). For \( n \ge 2 \):
$$\boxed{\det(\operatorname{adj} A) = (\det A)^{n-1}}, \qquad \operatorname{adj}(\operatorname{adj} A) = (\det A)^{n-2} A, \qquad \operatorname{adj}(AB) = \operatorname{adj} B \operatorname{adj} A.$$$$\operatorname{rank}(\operatorname{adj} A) = \begin{cases} n & \operatorname{rank} A = n \\ 1 & \operatorname{rank} A = n - 1 \\ 0 & \operatorname{rank} A \le n - 2 \end{cases}$$The middle case: if \( \operatorname{rank} A = n - 1 \), some \( (n-1) \)-minor is non-zero so \( \operatorname{adj} A \ne 0 \), and \( A \operatorname{adj} A = 0 \) puts the columns of \( \operatorname{adj} A \) in the 1-dimensional \( \ker A \). If \( \operatorname{rank} A \le n - 2 \), every \( (n-1) \)-minor vanishes.
Systems of Linear Equations
Write the system as \( Ax = b \), \( A \) of size \( m \times n \) (\( m \) equations, \( n \) unknowns), with augmented matrix \( [A \mid b] \).
graph TD
A["Ax = b, A is m x n, r = rank A"] --> B{"rank A = rank of [A, b] ?"}
B -->|no| C["No solution (inconsistent)"]
B -->|yes| D{"r = n ?"}
D -->|yes| E["Exactly one solution"]
D -->|no| F["Infinitely many solutions
(n - r free parameters)"]| Fact | Reason |
|---|---|
| \( Ax = 0 \) always has \( x = 0 \) | homogeneous systems are consistent |
| \( Ax = 0 \) has a non-zero solution iff \( \operatorname{rank} A < n \) | \( \ker A \ne 0 \) |
| \( m < n \) (fewer equations than unknowns) \( \Rightarrow \) non-trivial solutions of \( Ax = 0 \) | \( \operatorname{rank} A \le m < n \) |
| square \( A \): unique solution for every \( b \) \( \iff \det A \ne 0 \) | then \( x = A^{-1}b \) |
| \( Ax = b \) solvable for every \( b \in F^m \) \( \iff \operatorname{rank} A = m \) | columns span \( F^m \) |
| over \( \mathbb{F}_q \), a consistent system has exactly \( q^{\,n - r} \) solutions | \( \lvert \ker A \rvert = q^{n - r} \) |
| over \( \mathbb{R} \) or \( \mathbb{C} \), the number of solutions is 0, 1 or infinite | never exactly 2 |
Cramer’s rule. If \( A \) is \( n \times n \) with \( \det A \ne 0 \), then \( x_i = \frac{\det A_i}{\det A} \), where \( A_i \) is \( A \) with column \( i \) replaced by \( b \). Useful for one unknown of a \( 3 \times 3 \) system; never for proving existence when \( \det A = 0 \).
The hook system, solved
Reduce \( [A \mid b] \):
$$\left(\begin{array}{ccc|c} 1 & 1 & 1 & 1 \\ 1 & 2 & 4 & k \\ 1 & 4 & 10 & k^2 \end{array}\right) \to \left(\begin{array}{ccc|c} 1 & 1 & 1 & 1 \\ 0 & 1 & 3 & k - 1 \\ 0 & 3 & 9 & k^2 - 1 \end{array}\right) \to \left(\begin{array}{ccc|c} 1 & 1 & 1 & 1 \\ 0 & 1 & 3 & k - 1 \\ 0 & 0 & 0 & (k - 1)(k - 2) \end{array}\right).$$\( \operatorname{rank} A = 2 \) always. The system is consistent iff \( (k - 1)(k - 2) = 0 \), and then it has infinitely many solutions (one free parameter). There is no \( k \) giving a unique solution.
Invertibility: the equivalent conditions
For \( A \in M_n(F) \), the following are equivalent; a Part C option that is any one of them can be traded for any other.
- \( A \) is invertible; 2. \( \det A \ne 0 \); 3. \( \operatorname{rank} A = n \); 4. \( Ax = 0 \) has only \( x = 0 \); 5. \( Ax = b \) is solvable for every \( b \); 6. the columns (rows) form a basis of \( F^n \); 7. 0 is not an eigenvalue; 8. \( A \) is a product of elementary matrices.
Memory Tricks
- R-R-N for systems. Rank A vs Rank \( [A \mid b] \) decides existence; N minus rank counts free parameters.
- One fewer for the adjugate. \( \det(\operatorname{adj} A) = (\det A)^{n - 1} \), \( \operatorname{adj}(\operatorname{adj} A) = (\det A)^{n-2} A \): the exponent drops by one each time.
- Sylvester subtracts the middle. \( \operatorname{rank}(AB) \ge \operatorname{rank} A + \operatorname{rank} B - n \), with \( n \) the size the product sums over.
- aI + bJ: one odd eigenvalue. \( a + nb \) once, \( a \) for the rest. Determinant \( a^{n-1}(a + nb) \).
- Rank-one plus identity. \( \det(I + uv^T) = 1 + v^Tu \): the only interesting eigenvalue is \( 1 + v^Tu \).
Don’t Confuse
| This | Is not this | The difference |
|---|---|---|
| \( \det(cA) = c^n \det A \) | \( c \det A \) | each of the \( n \) rows is scaled |
| \( \operatorname{rank} A = \operatorname{rank}[A \mid b] \) (consistent) | \( \operatorname{rank} A = n \) (unique) | consistency vs uniqueness |
| \( \det(\operatorname{adj} A) = (\det A)^{n-1} \) | \( (\det A)^n \) | exponent is \( n - 1 \) |
| rank of \( A^TA \) equals rank of \( A \) over \( \mathbb{R} \) | over \( \mathbb{C} \) or \( \mathbb{Z}_p \) | needs \( x^Tx = 0 \Rightarrow x = 0 \); over \( \mathbb{C} \) use \( A^*A \) |
| invertible over \( \mathbb{Q} \) | invertible over \( \mathbb{Z} \) or \( \mathbb{Z}_p \) | over \( \mathbb{Z} \) need \( \det = \pm 1 \); over \( \mathbb{Z}_p \) need \( p \nmid \det \) |
| homogeneous: always consistent | non-homogeneous: may be inconsistent | \( 0 \) always solves \( Ax = 0 \) |
| row operations preserve row space | they preserve column space | they change the column space (but not its dimension) |
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 rank of \( uv^T \) for non-zero \( u, v \in \mathbb{R}^n \)?
- If \( A \) is \( 3 \times 3 \) with \( \det A = 5 \), find \( \det(2A) \).
- Does \( Ax = 0 \) with \( A \) of size \( 3 \times 5 \) always have a non-zero solution?
Answers
- 1. Every column is a multiple of \( u \).
- 40. \( 2^3 \cdot 5 \).
- Yes. \( \operatorname{rank} A \le 3 < 5 \), so \( \dim \ker A \ge 2 \).
Level 2: Part B Style
Practice Q1. The rank of the \( 4 \times 4 \) matrix \( A = (a_{ij}) \) with \( a_{ij} = i + j \) is:
(a) 1 (b) 2 (c) 3 (d) 4
Answer: (b)
Row \( i \) is \( i \cdot (1, 1, 1, 1) + (1, 2, 3, 4) \), so all rows lie in the span of two independent vectors: rank \( \le 2 \). The first two rows \( (2, 3, 4, 5) \) and \( (3, 4, 5, 6) \) are not proportional, so rank \( = 2 \).
Practice Q2. The system \( x + y + z = 1 \), \( x + 2y + 4z = k \), \( x + 4y + 10z = k^2 \) has a solution if and only if:
(a) \( k = 1 \) (b) \( k \in \{ 1, 2 \} \) (c) \( k \notin \{ 1, 2 \} \) (d) \( k \) is any real number
Answer: (b)
Row reduction (worked above) leaves the last row \( (0 \ 0 \ 0 \mid (k - 1)(k - 2)) \). Consistency needs this entry to vanish. For both values the solution is not unique, because \( \operatorname{rank} A = 2 < 3 \).
Practice Q3. Let \( A \) be a \( 4 \times 4 \) real matrix with \( \det A = 3 \). Then \( \det(\operatorname{adj} A) \) equals:
(a) 3 (b) 9 (c) 27 (d) 81
Answer: (c)
\( \det(\operatorname{adj} A) = (\det A)^{n - 1} = 3^3 = 27 \). (Taking determinants of \( A \operatorname{adj} A = 3I \) gives \( 3 \det(\operatorname{adj} A) = 3^4 \).)
Practice Q4. The determinant of the \( 5 \times 5 \) matrix with every diagonal entry 2 and every off-diagonal entry 1 is:
(a) 1 (b) 5 (c) 6 (d) 32
Answer: (c)
The matrix is \( I + J \). \( J \) has eigenvalues \( 5, 0, 0, 0, 0 \), so \( I + J \) has eigenvalues \( 6, 1, 1, 1, 1 \) and determinant 6. (32 is \( 2^5 \), the trap of multiplying the diagonal.)
Level 3: Part C Style
Practice Q5. Let \( A \) be a \( 5 \times 5 \) real matrix of rank 3. Which statements are true?
(a) \( \operatorname{adj} A = 0 \). (b) \( \operatorname{rank}(A^2) \ge 1 \). (c) \( Ax = b \) has a solution for every \( b \in \mathbb{R}^5 \). (d) \( \{ x \in \mathbb{R}^5 : A^Tx = 0 \} \) has dimension 2.
Answer: (a), (b) and (d)
- (a) True. \( \operatorname{rank} A = 3 \le n - 2 \), so every \( 4 \times 4 \) minor is zero and every cofactor vanishes.
- (b) True. Sylvester: \( \operatorname{rank}(A^2) \ge 3 + 3 - 5 = 1 \). (The bound is attained: the nilpotent matrix \( J_3(0) \oplus J_2(0) \) has rank \( 2 + 1 = 3 \) and its square has rank \( 1 + 0 = 1 \).)
- (c) False. The column space has dimension \( 3 < 5 \); any \( b \) outside it gives no solution.
- (d) True. \( \operatorname{rank} A^T = 3 \), so \( \dim \ker A^T = 5 - 3 = 2 \).
Practice Q6. Let \( A = \begin{pmatrix} 1 & 1 & 0 \\ 0 & 1 & 1 \\ 1 & 0 & 1 \end{pmatrix} \). Which statements are true?
(a) \( A \) is invertible as a real matrix. (b) \( A \) is invertible as a matrix over \( \mathbb{Z}_2 \). (c) \( Ax = 0 \) has exactly 2 solutions \( x \in \mathbb{Z}_2^3 \). (d) \( A^{-1} \) (over \( \mathbb{R} \)) has integer entries.
Answer: (a) and (c)
Expanding along the first row: \( \det A = 1(1 - 0) - 1(0 - 1) + 0 = 2 \).
- (a) True. \( \det A = 2 \ne 0 \).
- (b) False. \( 2 \equiv 0 \pmod 2 \). Indeed the three rows add to \( (2, 2, 2) \equiv 0 \).
- (c) True. Over \( \mathbb{Z}_2 \) the first two rows are independent, so the rank is 2 and the kernel has \( 2^{3 - 2} = 2 \) elements: \( (0, 0, 0) \) and \( (1, 1, 1) \).
- (d) False. \( A^{-1} = \frac{1}{2}\operatorname{adj} A \) and \( \operatorname{adj} A = \begin{pmatrix} 1 & -1 & 1 \\ 1 & 1 & -1 \\ -1 & 1 & 1 \end{pmatrix} \), so the entries are \( \pm \frac{1}{2} \). An integer matrix has an integer inverse iff its determinant is \( \pm 1 \).
Practice Q7. Let \( u = (1, 1, 1, 1)^T \), \( v = (1, 2, 3, 4)^T \) and \( A = I_4 + uv^T \). Which statements are true?
(a) \( \det A = 11 \). (b) \( A \) is invertible. (c) \( \operatorname{rank}(A - I) = 1 \). (d) \( \operatorname{tr} A = 10 \).
Answer: (a), (b) and (c)
- (a) True. Matrix determinant lemma: \( \det(I + uv^T) = 1 + v^Tu = 1 + (1 + 2 + 3 + 4) = 11 \).
- (b) True. \( \det A \ne 0 \).
- (c) True. \( A - I = uv^T \) has rank 1.
- (d) False. \( \operatorname{tr} A = 4 + \operatorname{tr}(uv^T) = 4 + v^Tu = 14 \). (10 is \( v^Tu \) alone.)
Quick Revision Table
| Idea | Fact |
|---|---|
| Rank | dim of column space = dim of row space = largest non-zero minor |
| Product | \( \operatorname{rank} A + \operatorname{rank} B - n \le \operatorname{rank}(AB) \le \min \) |
| Real \( A^TA \) | same rank as \( A \) |
| det rules | multiplicative, \( \det(cA) = c^n\det A \), \( \det A^T = \det A \) |
| \( aI + bJ \) | \( \det = a^{n-1}(a + nb) \) |
| Rank-one update | \( \det(I + uv^T) = 1 + v^Tu \) |
| Adjugate | \( \det \operatorname{adj} A = (\det A)^{n-1} \); rank \( n, 1, 0 \) |
| Odd skew-symmetric real | \( \det = 0 \) |
| Consistency | \( \operatorname{rank} A = \operatorname{rank}[A \mid b] \) |
| Number of free parameters | \( n - \operatorname{rank} A \) |
| Over \( \mathbb{F}_q \) | \( q^{n - r} \) solutions when consistent |
| Integer inverse | iff \( \det = \pm 1 \) |
Key Takeaways
What’s Next
- Eigenvalues and the Cayley–Hamilton Theorem: \( \det(A - \lambda I) = 0 \) turns this page’s determinants into eigenvalues.
- Vector Spaces, Bases and Linear Transformations for rank–nullity, the engine behind the solution count.
- Inner Product Spaces and Orthonormal Bases: when \( Ax = b \) has no solution, least squares finds the best approximate one.
- Linear Algebra Quick Revision.