Module 2 · GATE Engineering Mathematics

Linear Algebra

Updated 17 Nov 2026 · 3,950 words · illustrated
Updated 17 Nov 2026
Matrices, determinants, eigenvalues, eigenvectors, systems of equations.
Matrices · Determinants · Eigenvalues · Systems of Equations
🎯 Take Mock Test
Vector SpacesLinear MapsEigenvaluesMap ProjectionsSpatial PCAbasisprojectsdecomposesArt. 3: Linear Algebra underpins GIS analysisDef: Eigenvalue = scaling factor of eigenvectorvlymbooq
Linear Algebra Concepts Applied to Geographic Analysis for Competitive Exams — Shows how core linear algebra ideas drive map projections and spatial data reduction. · Open the full diagram →
Vector SpaceBasis VectorsLinear MapsEigen DecompositionCoordinate AxesTransformationMatrixPrincipalComponentsspansgeneratesyieldsArt. 3: Basis defines coordinate systemMechanism: Eigenvalues reveal variance axesvlymbooq
Vector Space Basis Linear Maps Eigen Decomposition Overview — Shows how vector spaces, bases, linear maps, and eigen decomposition form the core of spatial data analysis. · Open the full diagram →

Learning Objectives

  • Understand the fundamental concepts of Linear Algebra
  • Apply key formulas and techniques to solve problems
  • Practice with exam-level questions to build speed and accuracy

What is Linear Algebra?

Linear algebra is a foundational topic in engineering mathematics, covering vector spaces, matrices, determinants, systems of linear equations, eigenvalues, and eigenvectors. In GATE, linear algebra carries 5-8% weightage across all branches.

Key Concepts

1. Matrices: Types: row, column, square, diagonal, scalar, identity, zero, symmetric (Aᵀ = A), skew-symmetric (Aᵀ = -A), orthogonal (AAᵀ = I). Rank of a matrix = number of linearly independent rows/columns. Trace = sum of diagonal elements.

2. Determinants: |A| for square matrix. Properties: |AB| = |A||B|, |Aᵀ| = |A|, |A⁻¹| = 1/|A|, |kA| = kⁿ|A| for n×n matrix. Cramer's rule for solving linear systems using determinants.

3. Systems of Linear Equations: AX = B. If |A| ≠ 0, unique solution exists. If |A| = 0, either no solution or infinite solutions. Consistency: rank(A) = rank([A|B]) for consistency. Homogeneous systems: AX = 0 always has trivial solution X = 0; non-trivial solutions exist if |A| = 0.

4. Eigenvalues & Eigenvectors: Characteristic equation: |A - λI| = 0. Eigenvalues λ satisfy trace(A) = sum(λᵢ) and |A| = product(λᵢ). For symmetric matrices, eigenvectors are orthogonal. Spectral theorem: real symmetric matrices have real eigenvalues.

5. Cayley-Hamilton Theorem: Every square matrix satisfies its own characteristic equation. Useful for finding A⁻¹ and higher powers of A.

Solved Examples

Example 1 (Determinant — Properties)
If A is a 3×3 matrix with |A| = 5, find |3A|.

a) 15 b) 45 c) 135 d) 27
Solution: c) 135. |kA| = kⁿ|A| where n is the order. For 3×3 matrix, n=3. |3A| = 3³×|A| = 27×5 = 135.
Example 2 (System of Equations)
Find the value of k for which the system has infinite solutions: x + 2y + z = 3, 2x + 4y + 2z = 6, 3x + 6y + kz = 9.

a) 1 b) 2 c) 3 d) 4
Solution: c) 3. The second equation is 2 times the first, so it's redundant. The third: 3x+6y+kz=9. For infinite solutions, the third must be 3 times the first: 3(x+2y+z) = 3x+6y+3z = 9. So k = 3.
Example 3 (Eigenvalues)
Find the eigenvalues of the matrix A = [[3, 1], [1, 3]].

a) 2, 4 b) 1, 5 c) 3, 3 d) -2, 8
Solution: a) 2, 4. Characteristic equation: |A - λI| = 0 → |(3-λ) 1; 1 (3-λ)| = (3-λ)² - 1 = 0 → λ² - 6λ + 9 - 1 = 0 → λ² - 6λ + 8 = 0 → (λ-2)(λ-4) = 0 → λ = 2, 4. Verification: trace = 3+3 = 6 = 2+4. Determinant = 9-1 = 8 = 2×4.

Shortcut Techniques

  • For 2×2 matrices, eigenvalues can be found quickly: λ = (trace ± √(trace² - 4det))/2.
  • Check consistency of linear systems by comparing ranks of A and [A|B] — rank = number of non-zero rows in row-echelon form.
  • For orthogonal matrices, columns are unit vectors and mutually perpendicular — Aᵀ = A⁻¹.
  • For finding A⁻¹ of 2×2 matrix: swap diagonal elements, negate off-diagonal, divide by determinant.
  • Cayley-Hamilton theorem provides a quick way to compute A² or A⁻¹ without solving linear systems.
Pro Tip
In GATE, linear algebra questions are often formula-based. Master the properties of determinants, eigenvalues, and rank. Practice problems on consistency of linear systems — this is a frequently tested concept. Also, understand the geometric interpretation: determinant = volume scaling factor, eigenvectors = directions preserved by transformation.

Advanced Applications

Modern exams feature questions combining multiple quantitative concepts. Master these by practicing previous year papers.

Shortcut Methods

Percentages: Use fraction table (12.5%=1/8, 20%=1/5, 25%=1/4). SI/CI: For 2 years, CI-SI = P(r/100)^2. Time Work: Use LCM of time periods. Average Speed: 2ab/(a+b) for equal distances.

Pro Tip
Write key formulas on rough sheet immediately after the exam bell while memory is fresh.

Matrix Decompositions and Their Uses

Matrix decompositions factor a matrix into a product of simpler matrices that reveal structural properties such as rank, conditioning, and invariant subspaces. The most common decompositions in competitive examinations are LU, QR, and Cholesky. LU decomposition writes a square matrix A as A = LU where L is unit lower‑triangular and U is upper‑triangular; it exists for any non‑singular matrix after suitable row permutations (PA = LU). QR decomposition expresses A = QR with Q orthogonal (QᵀQ = I) and R upper‑triangular; it is numerically stable and underlies the Gram‑Schmidt process and the solution of least‑squares problems. Cholesky decomposition applies only to symmetric positive‑definite matrices, giving A = LLᵀ where L is lower‑triangular with positive diagonal entries; it halves the computational cost compared with LU. Each decomposition has a direct algorithmic counterpart: forward‑back substitution for LU, orthogonal transformations for QR, and a simple recurrence for Cholesky. In examinations, questions often ask to identify which decomposition is applicable given matrix properties, or to compute a specific factor when the matrix size is small (2×2 or 3×3). Understanding the existence conditions — non‑singular for LU, full column rank for QR, symmetric positive‑definite for Cholesky — prevents the common error of applying a decomposition to an unsuitable matrix.

Beyond the three classical factorizations, the Schur decomposition A = QTQᵀ (Q orthogonal, T upper‑triangular) guarantees that every real square matrix can be triangularised by an orthogonal similarity. The diagonal entries of T are the eigenvalues of A, and the columns of Q form an orthonormal basis of invariant subspaces. This result underpins the QR algorithm for eigenvalue computation. The spectral theorem for real symmetric matrices is a special case where T becomes diagonal. In practice, examiners may test the relationship between Schur form and the characteristic polynomial, or ask to deduce the eigenvalues of a block‑triangular matrix from its diagonal blocks. Mastery of these decompositions enables rapid solution of problems involving matrix powers, exponentials, and condition numbers without resorting to full diagonalisation.

Generalized Eigenvectors and Jordan Canonical Form

When a matrix does not possess a complete set of linearly independent eigenvectors, the Jordan canonical form provides the next best structural description. For each eigenvalue λ, the algebraic multiplicity is the exponent of (λ−λ) in the characteristic polynomial, while the geometric multiplicity is the dimension of the eigenspace ker(A−λI). If geometric multiplicity is less than algebraic multiplicity, the matrix is defective and generalized eigenvectors are required. A generalized eigenvector of rank k satisfies (A−λI)ᵏv = 0 but (A−λI)ᵏ⁻¹v ≠ 0. Chains of generalized eigenvectors form Jordan blocks Jₖ(λ) = λIₖ + Nₖ where Nₖ is the nilpotent matrix with ones on the super‑diagonal. The Jordan form J = P⁻¹AP is block‑diagonal with these blocks; P contains the eigenvectors and generalized eigenvectors as columns. The number of Jordan blocks for λ equals the geometric multiplicity, and the sizes of the blocks are determined by the sequence of dimensions of ker(A−λI)ʲ. In examinations, typical tasks include constructing the Jordan form of a 3×3 or 4×4 matrix given its characteristic and minimal polynomials, or determining the size of the largest Jordan block from the nilpotency index of (A−λI). A key shortcut: the minimal polynomial is the least common multiple of (λ−λ)ˢ where s is the size of the largest Jordan block for each eigenvalue. This allows rapid verification of whether a matrix is diagonalisable (minimal polynomial has only simple roots) without computing eigenvectors explicitly.

Technique: Building Jordan Chains from Nilpotent Powers
Let A be a 4×4 matrix with a single eigenvalue λ = 2 and characteristic polynomial (λ−2)⁴. Suppose (A−2I)² ≠ 0 but (A−2I)³ = 0. Determine the Jordan canonical form of A.
The nilpotency index of N = A−2I is 3, so the largest Jordan block has size 3. Since the matrix is 4×4, the remaining block must be of size 1. Therefore the Jordan form consists of one 3×3 block J₃(2) and one 1×1 block J₁(2). Explicitly, J = diag(J₃(2), [2]). The geometric multiplicity equals the number of blocks = 2, and the minimal polynomial is (λ−2)³.
Exam‑Day Strategy
When a problem gives only the characteristic polynomial and a condition on powers of (A−λI), first compute the nilpotency index to locate the largest Jordan block. Then use the matrix dimension to fill the remaining blocks. This avoids the longer route of finding explicit generalized eigenvectors.

Singular Value Decomposition and Low‑Rank Approximation

The singular value decomposition (SVD) factorises any real m×n matrix A as A = UΣVᵀ where U (m×m) and V (n×n) are orthogonal and Σ (m×n) is diagonal with non‑negative singular values σ₁ ≥ σ₂ ≥ … ≥ σᵣ > 0 (r = rank(A)). The columns of U are left singular vectors (eigenvectors of AAᵀ), columns of V are right singular vectors (eigenvectors of AᵀA), and σᵢ² are the non‑zero eigenvalues of both AAᵀ and AᵀA. SVD exists for every matrix, unlike eigendecomposition which requires square and diagonalizable matrices. A principal application is the best rank‑k approximation in the Frobenius or spectral norm: Aₖ = Σᵢ₌₁ᵏ σᵢ uᵢ vᵢᵀ, which minimises ‖A−B‖ among all matrices B of rank ≤ k (Eckart‑Young theorem). In examinations, questions may ask to compute the singular values of a 2×2 matrix analytically, to relate the condition number κ₂(A) = σ₁/σᵣ to numerical stability, or to determine the rank from the number of non‑zero singular values. A useful shortcut: for a 2×2 matrix A = [[a, b], [c, d]], the singular values are the square roots of the eigenvalues of AᵀA = [[a²+c², ab+cd], [ab+cd, b²+d²]], which can be obtained via the quadratic formula using trace and determinant of AᵀA. The nuclear norm ‖A‖* = Σ σᵢ and the operator norm ‖A‖₂ = σ₁ are also frequently tested. Understanding that the SVD provides an orthonormal basis for the four fundamental subspaces (column space, row space, null space, left null space) helps answer conceptual questions about rank‑deficiency and pseudoinverses.

Matrix Functions and the Matrix Exponential

Functions of a square matrix extend scalar functions via power series, diagonalisation, or the Jordan form. For a diagonalizable matrix A = PDP⁻¹, f(A) = Pf(D)P⁻¹ where f(D) applies f to each diagonal entry. For a defective matrix, the Jordan form yields f(J) where each Jordan block Jₖ(λ) is transformed into an upper‑triangular Toeplitz matrix whose entries are derivatives of f at λ divided by factorials: (f(Jₖ(λ)))ᵢⱼ = f⁽ʲ⁻ⁱ⁾(λ)/(j−i)! for j ≥ i. The matrix exponential eᴬ = Σₖ₌₀^∞ Aᵏ/k! is the most important matrix function; it solves the linear system x′ = Ax with solution x(t) = eᴬᵗ x(0). Properties include eᴬ⁺ᴮ = eᴬeᴮ when A and B commute, det(eᴬ) = etr(A), and (eᴬ)⁻¹ = e−ᴬ. For a 2×2 matrix with distinct eigenvalues λ₁, λ₂, eᴬ can be computed via the Lagrange interpolation formula eᴬ = (eλ₁−eλ₂)/(λ₁−λ₂) A + (λ₁eλ₂−λ₂eλ₁)/(λ₁−λ₂) I. When eigenvalues coincide, the formula involves the derivative: eᴬ = eλ I + eλ (A−λI). Examiners often ask to compute eᴬ for a given 2×2 or 3×3 matrix, or to use the exponential to solve a system of differential equations with constant coefficients. A practical shortcut: if A is nilpotent of index m (Aᵐ = 0), the series terminates at k = m−1, giving a polynomial in A. Recognising nilpotent blocks inside a Jordan form reduces the exponential to a finite sum.

Applications to Linear Differential Equations and Stability Analysis

Linear systems of ordinary differential equations x′ = Ax with constant coefficient matrix A are solved by the matrix exponential x(t) = eᴬᵗ x₀. The long‑term behaviour is governed by the eigenvalues of A. If all eigenvalues have negative real parts, the origin is asymptotically stable and every solution decays to zero exponentially. If any eigenvalue has a positive real part, solutions grow unbounded and the system is unstable. Purely imaginary eigenvalues lead to marginal stability (oscillations) provided the Jordan blocks are of size one; a defective imaginary eigenvalue produces polynomial growth (e.g., t·sin ωt) and instability. The Lyapunov equation AᵀP + PA = −Q with Q positive‑definite has a unique positive‑definite solution P iff A is Hurwitz (all eigenvalues in the open left half‑plane); this provides an algebraic stability test without computing eigenvalues. In examinations, typical problems present a 2×2 or 3×3 matrix and ask to classify the equilibrium, to sketch phase portraits based on eigenvalue type (node, saddle, focus, centre), or to determine the range of a parameter for which the system is stable. The Routh‑Hurwitz criterion applied to the characteristic polynomial gives necessary and sufficient conditions for all roots to have negative real parts without solving the polynomial. For a cubic λ³ + a₁λ² + a₂λ + a₃, stability requires a₁>0, a₃>0, and a₁a₂ > a₃. Mastery of these criteria allows rapid parameter‑range questions that would otherwise require explicit eigenvalue computation.

  • Assuming LU decomposition exists without checking for zero pivots or the need for row permutations.
  • Confusing geometric multiplicity with algebraic multiplicity when deciding diagonalisability.
  • Applying the Eckart‑Young theorem to a matrix that is not real or not using the Frobenius norm as specified.
  • Using the scalar exponential formula eλt for each eigenvalue without accounting for Jordan blocks of size greater than one.
  • Neglecting the commutativity condition when combining matrix exponentials eᴬeᴮ = eA+B.
  • Misapplying the Routh‑Hurwitz array by omitting the first‑column sign‑change test for higher‑order polynomials.
  • Treating the singular values of a rectangular matrix as eigenvalues of the matrix itself rather than of AᵀA or AAᵀ.

Quadratic Forms and Definiteness Criteria

A quadratic form in n variables is a homogeneous polynomial of degree two, written as Q(x) = xᵀ A x where A is a real symmetric n×n matrix and x ∈ ℝⁿ. The classification of Q depends entirely on the eigenvalues of A. If all eigenvalues are positive, Q is positive‑definite; if all are non‑negative, Q is positive‑semidefinite. Negative‑definite and negative‑semidefinite are defined analogously. When eigenvalues have mixed signs the form is indefinite. Sylvester’s criterion provides an algebraic test without computing eigenvalues: a symmetric matrix is positive‑definite iff all its leading principal minors are strictly positive. For positive‑semidefiniteness every principal minor (not only the leading ones) must be non‑negative. The same minors with alternating signs characterise negative‑definiteness. In examinations the criterion is frequently applied to bordered Hessian matrices arising from constrained optimisation, where the sign of the last n‑m leading principal minors determines the nature of a stationary point under m equality constraints. The Cholesky factorisation A = LLᵀ exists exactly for positive‑definite matrices and yields a numerically stable way to solve linear systems and to sample from multivariate normal distributions. For a positive‑semidefinite matrix a pivoted Cholesky or an LDLᵀ factorisation with a diagonal D ≥ 0 can be used. The inertia of a symmetric matrix — the triple (π, ν, δ) counting positive, negative and zero eigenvalues — is invariant under congruence transformations (Sylvester’s law of inertia). This invariance underpins the classification of conic sections and the analysis of saddle points in optimisation.

Kronecker Product, Vectorization, and Matrix Equations

The Kronecker product A ⊗ B of an m×n matrix A and a p×q matrix B is the mp×nq block matrix whose (i,j) block equals aᵢⱼ B. It satisfies (A ⊗ B)(C ⊗ D) = (AC) ⊗ (BD) whenever the ordinary products are defined, and (A ⊗ B)ᵀ = Aᵀ ⊗ Bᵀ. Vectorisation vec(X) stacks the columns of X into a single column vector. The fundamental identity vec(AXB) = (Bᵀ ⊗ A) vec(X) converts matrix equations into standard linear systems. For the Sylvester equation AX + XB = C, vectorisation gives (I ⊗ A + Bᵀ ⊗ I) vec(X) = vec(C). The coefficient matrix is a sum of two Kronecker products; its eigenvalues are λᵢ(A) + λⱼ(B). Hence the Sylvester equation has a unique solution iff λᵢ(A) + λⱼ(B) ≠ 0 for all i,j. The discrete‑time Lyapunov equation X − A X Aᵀ = Q and the continuous‑time Lyapunov equation A X + X Aᵀ = −Q are special cases. In control theory the controllability Gramian solves the Lyapunov equation and its positive‑definiteness is equivalent to controllability. The Kronecker sum A ⊕ B = A ⊗ I + I ⊗ B appears naturally when differentiating matrix exponentials: d/dt eA t = A eA t = eA t A. The vec‑operator also linearises the Riccati equation, enabling Newton‑type iterations that operate on vectorised unknowns. Examiners often ask to rewrite a given matrix equation in vec‑form, to state the solvability condition in terms of eigenvalue sums, or to compute the dimension of the solution space when the coefficient matrix is singular.

Worked Example: Sylvester Equation Solvability
Let A = [[2, 0], [0, -1]] and B = [[-2, 0], [0, 3]]. Determine whether the Sylvester equation AX + XB = C has a unique solution for every 2×2 matrix C.
The eigenvalues of A are 2 and -1. The eigenvalues of B are -2 and 3. The sums λᵢ(A) + λⱼ(B) are: 2+(-2)=0, 2+3=5, -1+(-2)=-3, -1+3=2. Because one sum equals zero, the coefficient matrix I ⊗ A + Bᵀ ⊗ I is singular. Hence the Sylvester equation does not have a unique solution for all C; it is solvable only when vec(C) lies in the column space of that coefficient matrix.
Exam‑Day Strategy
When a matrix equation appears, first identify its type (Sylvester, Lyapunov, Stein, Riccati). Write the vectorised form immediately; the Kronecker‑sum eigenvalue condition is a one‑line check for uniqueness. If the problem asks for a parameter range guaranteeing a unique solution, compute the eigenvalue sums symbolically and enforce non‑zero inequalities.

Perturbation Bounds for Eigenvalues and Eigenvectors

Let A be a diagonalizable matrix with eigenvalue decomposition A = V Λ V⁻¹ and let à = A + E be a perturbation. The Bauer–Fike theorem states that for any eigenvalue μ of à there exists an eigenvalue λ of A such that |μ − λ| ≤ κ(V) ‖E‖, where κ(V) = ‖V‖ ‖V⁻¹‖ is the condition number of the eigenvector matrix in the chosen norm. For normal matrices (including Hermitian and unitary) κ(V)=1, so eigenvalues are perfectly conditioned: each perturbed eigenvalue lies within ‖E‖ of some original eigenvalue. Weyl’s inequalities sharpen this for Hermitian matrices: if eigenvalues are ordered λ₁ ≤ … ≤ λₙ and μ₁ ≤ … ≤ μₙ, then |μᵢ − λᵢ| ≤ ‖E‖₂ for all i. Eigenvector perturbation is governed by the Davis–Kahan sin Θ theorem: if an invariant subspace of A is separated from the rest of the spectrum by a gap δ > 0, then the sine of the angle between the true and perturbed subspaces is bounded by ‖E‖₂ / δ. For a simple eigenvalue λ with normalised eigenvector x, the first‑order change in the eigenvector is (λI − A)⁺ E x, where ⁺ denotes the Moore–Penrose inverse. In examinations, typical questions give a symmetric matrix with a known eigenvalue gap and ask for a bound on the eigenvector rotation caused by a perturbation of given norm. Recognising that the bound scales inversely with the spectral gap is crucial; clustered eigenvalues are highly sensitive even to tiny perturbations.

Matrix Norms, Condition Numbers, and Numerical Stability

A matrix norm ‖·‖ is a vector norm on ℝm×n satisfying sub‑multiplicativity ‖AB‖ ≤ ‖A‖ ‖B‖ whenever the product is defined. The induced (operator) norm ‖A‖ₚ = maxx≠0 ‖Ax‖ₚ / ‖x‖ₚ equals the largest singular value for p=2, the maximum absolute column sum for p=1, and the maximum absolute row sum for p=∞. The Frobenius norm ‖A‖_F = √(Σ aᵢⱼ²) = √(trace(AᵀA)) is unitarily invariant but not an induced norm. The condition number of a nonsingular matrix A with respect to inversion is κ(A) = ‖A‖ ‖A⁻¹‖. For the 2‑norm κ₂(A) = σ₁/σₙ, the ratio of largest to smallest singular values. A problem is ill‑conditioned when κ is large; the relative error in the solution of Ax = b can be amplified by up to κ(A) times the relative error in the data. Backward stability of an algorithm means the computed solution x̂ satisfies (A+ΔA)x̂ = b with ‖ΔA‖/‖A‖ = O(ε_machine). Gaussian elimination with partial pivoting is backward stable for most matrices, but the growth factor can be exponential for pathological cases (e.g., Wilkinson’s matrix). The QR factorisation via Householder reflections is unconditionally backward stable and yields an orthogonal Q, making it the preferred method for least‑squares problems. In exams, you may be asked to compute κ₁, κ₂, or κ_∞ for a given matrix, to compare the sensitivity of solving linear systems versus computing eigenvalues, or to explain why forming the normal equations AᵀA x = Aᵀb squares the condition number and should be avoided when κ₂(A) is moderate.

Schur Decomposition and the Real Schur Form

Every square complex matrix A admits a Schur decomposition A = Q T Q* where Q is unitary and T is upper triangular with the eigenvalues of A on its diagonal. The columns of Q are a Schur basis; the first k columns span an invariant subspace corresponding to the first k diagonal entries of T. For real matrices a real Schur form exists: A = Q T Qᵀ with Q orthogonal and T block upper triangular where each diagonal block is either a 1×1 real eigenvalue or a 2×2 block [[a, b], [c, d]] with complex conjugate eigenvalues a ± i√(bc−ad). The real Schur form is computed by the QR algorithm with implicit double shifts and is the backbone of modern eigensolvers (e.g., LAPACK’s ?gees). The ordered Schur form allows the eigenvalues to appear in any prescribed order along the diagonal, which is essential for solving Sylvester and Riccati equations via the Bartels–Stewart algorithm. The separation between two invariant subspaces is quantified by sep(T₁₁, T₂₂) = min‖X‖_F=1 ‖T₁₁ X − X T₂₂‖_F; this quantity appears in perturbation bounds for invariant subspaces. In examinations, you may be asked to construct the real Schur form of a 3×3 matrix with one real and a pair of complex eigenvalues, to explain how the QR algorithm converges to the Schur form, or to use the ordered Schur form to decouple a large Sylvester equation into smaller independent subsystems.

  • Confusing the condition number of the eigenvector matrix κ(V) with the eigenvalue condition number; for normal matrices they coincide but for non‑normal matrices κ(V) can be arbitrarily large.
  • Applying Weyl’s inequalities to non‑Hermitian matrices; they hold only for Hermitian (or symmetric) perturbations.
  • Assuming the Kronecker sum eigenvalue condition λᵢ(A)+λⱼ(B)≠0 is sufficient for the Sylvester equation when A or B are defective; the condition remains necessary and sufficient regardless of diagonalizability.
  • Using the Frobenius norm as an induced norm in backward‑error analysis; it is not sub‑multiplicative with respect to the vector 2‑norm.
  • Neglecting the growth factor in Gaussian elimination when arguing numerical stability; partial pivoting bounds the growth factor by 2n−1 but does not eliminate it.
  • Treating the real Schur form as a diagonalisation; the 2×2 blocks cannot be further reduced by orthogonal transformations.
  • Forgetting that the separation sep(T₁₁,T₂₂) governs the sensitivity of invariant subspaces, not the eigenvalue gaps alone.

These advanced topics complete the core toolkit required for high‑level linear algebra problems in competitive examinations.

Practice Questions

Calculus →