First Year Engineering is coming soon

We're still writing and reviewing the notes for every subject, and we'll open it when every subject is ready. Meanwhile you can practise aptitude, ask Aria about any subject, or see the branches that are ready.

Eigenvalues and eigenvectors

Eigenvalues and eigenvectors: definitions, characteristic equation, eigenspaces, algebraic/geometric multiplicity, properties (trace/product), diagonalisation conditions, orthogonal eigenvectors for symmetric matrices, and standard exam-type computations.

Drafted with Aria, reviewed by the AiCanCode.org team · Updated . How we write and check our notes. Spotted an error? Use Give Feedback at the bottom of the page.

Why it matters

Engineers use eigenvalues and eigenvectors to understand natural modes of a system: vibration modes of structures, principal directions in stress/strain, stability of control systems, and diffusion/heat flow patterns. They also power data tools like Principal Component Analysis (PCA) in sensing and signal processing.

Key ideas

  • Eigenspace (definition): For an eigenvalue λ of A, the set E_λ = { x ∈ ℝⁿ (or ℂⁿ, as applicable) : (A − λI)·x = 0 } is a subspace (the null space of A − λI), called the eigenspace corresponding to λ. Its dimension is the geometric multiplicity of λ.
  • Characteristic polynomial and characteristic equation: For an n×n matrix A, the polynomial p_A(λ) = det(A − λI). A scalar λ is an eigenvalue if and only if p_A(λ) = 0. The equation det(A − λI) = 0 is the characteristic equation.
    • Derivation (what to write in exams): An eigenpair (λ, x ≠ 0) satisfies (A − λI)x = 0. A homogeneous system has a nontrivial solution iff the coefficient matrix is singular, i.e., det(A − λI) = 0. Conversely, if det(A − λI) = 0, then (A − λI) is singular and there exists x ≠ 0 with (A − λI)x = 0.
    • Cayley–Hamilton theorem (often examined): Every square matrix satisfies its own characteristic equation p_A(A) = 0. This allows computing powers of A and e^{tA} even when A is not diagonalizable.
  • Algebraic and geometric multiplicity:
    • Algebraic multiplicity of λ is its multiplicity as a root of p_A(λ) = 0.
    • Geometric multiplicity of λ is dim(E_λ) = nullity(A − λI).
    • Always 1 ≤ geometric multiplicity ≤ algebraic multiplicity.
  • Similarity and diagonalisation:
    • Similar matrices: A and B are similar if there exists an invertible P with B = P⁻¹AP. Similar matrices represent the same linear transformation under different bases and have the same eigenvalues (including algebraic multiplicities) and characteristic polynomial.
    • Diagonalisation (definition and criterion): A is diagonalizable if there exists an invertible P such that A = P D P⁻¹ where D is diagonal. A is diagonalizable over ℝ (or ℂ) if and only if the sum of geometric multiplicities equals n, i.e., there exists a basis of n linearly independent eigenvectors. Sufficient condition: if A has n distinct eigenvalues, then A is diagonalizable.
  • Orthogonal diagonalisation for real symmetric matrices:
    • If A = Aᵀ (real symmetric), then:
      • All eigenvalues are real.
      • Eigenvectors for distinct eigenvalues are orthogonal.
      • There exists an orthogonal matrix Q (QᵀQ = I) and diagonal Λ such that A = Q Λ Qᵀ. (Spectral theorem.) This is heavily used in principal axes of quadratic forms and in PCA. Details of quadratic forms are in the later topic “Diagonalisation and quadratic forms”.
  • Independence of eigenvectors for distinct eigenvalues (theorem): Eigenvectors corresponding to distinct eigenvalues are linearly independent.
    • Sketch proof: Suppose c₁x₁ + … + c_kx_k = 0 with A x_i = λ_i x_i, all λ_i distinct. Apply A, subtract λ_k times the original relation, and argue by induction that all c_i = 0.
  • Trace and determinant relations:
    • For any n×n matrix A over ℂ, the sum of eigenvalues counting algebraic multiplicities equals trace(A).
    • The product of eigenvalues equals det(A).
    • Reason: When A is diagonalizable, p_A(λ) = ∏(λ − λ_i), and expanding gives coefficients related to (−1)^{n−1}·trace(A) and (−1)^n·det(A). In general, the equalities hold by continuity of coefficients or using Schur triangular form.
  • Eigenvalues of triangular matrices: For upper or lower triangular matrices, the eigenvalues are exactly the diagonal entries (counted with multiplicity), since det(A − λI) is the product of diagonal entries (a_ii − λ).
  • Rayleigh quotient (definition): For real symmetric A and nonzero x, R(x) = (xᵀA x)/(xᵀx). It attains its minimum and maximum at the eigenvectors of the smallest and largest eigenvalues, respectively. Useful for estimating extreme eigenvalues.
  • Matrix functions via diagonalisation: If A = P D P⁻¹, then for integer k, A^k = P D^k P⁻¹; for the matrix exponential, e^{tA} = P e^{tD} P⁻¹, where e^{tD} is diagonal with entries e^{tλ_i}. This helps solve linear ODE systems x′ = A x.

Diagram to draw (if asked to illustrate the idea): Draw the plane ℝ² with several arrows showing A acting on different vectors. Show one special direction v (an eigenvector), where Av lies on the same line as v (scaled by λ), while other general vectors u are rotated/sheared. Label v, Av, λ, and indicate “direction unchanged”.

Assumptions commonly required in statements:

  • A is an n×n matrix over ℝ or ℂ.
  • For orthogonal diagonalisation, A is real symmetric (A = Aᵀ).
  • For results using diagonalisation (like A^k), assume A is diagonalizable (i.e., has n linearly independent eigenvectors).

Formulas

  • E_λ = { x : (A − λ I) x = 0 } — Eigenspace (null space of A − λI); dimension is geometric multiplicity. Applies when λ is an eigenvalue.
  • trace(A) = sum of diagonal entries of A — Scalar equal to sum of eigenvalues counting algebraic multiplicities. Dimensionless in pure math. Applies to any square matrix.
  • det(A) = product of eigenvalues (with multiplicity) — Determinant equals product of eigenvalues; in particular, det(A) = 0 if and only if 0 is an eigenvalue (i.e., A is singular). Applies over ℂ; true for real matrices as well (complex roots come in conjugate pairs if A is real).
  • A = P D P⁻¹ — Diagonalisation; P: matrix of eigenvectors (columns), D: diagonal with eigenvalues. Applies if A is diagonalizable and P is invertible.
  • A^k = P D^k P⁻¹ — Power of a diagonalizable matrix; k ∈ ℤ (for k ≥ 0; for negative k also require A invertible). Useful to compute powers quickly.
  • e^{tA} = P e^{tD} P⁻¹ — Matrix exponential when A is diagonalizable; t has units of time in dynamical systems; e^{tD} is diagonal with entries e^{tλ_i}.
  • R(x) = (xᵀ A x) / (xᵀ x) — Rayleigh quotient; for real symmetric A and x ≠ 0. Gives bounds between min and max eigenvalues.

Worked examples

Example 1: Find eigenvalues, eigenvectors, and diagonalise A; then compute A⁴ Given A = [ [4, 2], [1, 3] ] (dimensionless).

  1. Characteristic polynomial and eigenvalues
  • Formula: det(A − λI) = 0.
  • Compute A − λI = [ [4 − λ, 2], [1, 3 − λ] ].
  • det(A − λI) = (4 − λ)(3 − λ) − (2)(1) = (12 − 7λ + λ²) − 2 = λ² − 7λ + 10.
  • Solve λ² − 7λ + 10 = 0 ⇒ (λ − 5)(λ − 2) = 0 ⇒ eigenvalues λ₁ = 5, λ₂ = 2.
  • Check with trace and determinant:
    • trace(A) = 4 + 3 = 7 = 5 + 2 ✔
    • det(A) = 4·3 − 2·1 = 10 = 5·2 ✔
  1. Eigenvectors
  • For λ = 5: Solve (A − 5I)x = 0.
    • A − 5I = [ [−1, 2], [1, −2] ]. Row-reduce: equations are −x + 2y = 0 ⇒ x = 2y.
    • Take y = 1 ⇒ x = 2. Eigenvector v₁ = [2, 1]ᵀ.
  • For λ = 2: Solve (A − 2I)x = 0.
    • A − 2I = [ [2, 2], [1, 1] ]. Equation 2x + 2y = 0 ⇒ x = −y.
    • Take y = 1 ⇒ x = −1. Eigenvector v₂ = [−1, 1]ᵀ.
  • The eigenvectors are linearly independent (distinct eigenvalues), so A is diagonalizable.
  1. Diagonalisation and A⁴
  • Form P with eigenvectors as columns: P = [ [2, −1], [1, 1] ]. D = diag(5, 2).
  • Compute P⁻¹ (2×2 inverse): det(P) = 2·1 − (−1)·1 = 3. So P⁻¹ = (1/3) [ [1, 1], [−1, 2] ].
  • Then A = P D P⁻¹.
  • For A⁴: A⁴ = P D⁴ P⁻¹ with D⁴ = diag(5⁴, 2⁴) = diag(625, 16).
  • Compute D⁴ P⁻¹ first: D⁴ P⁻¹ = (1/3) [ [625, 625], [−16, 32] ].
  • Then A⁴ = P · (D⁴ P⁻¹) = (1/3) [ [2, −1], [1, 1] ] · [ [625, 625], [−16, 32] ]. Multiply:
    • First row: [ 2·625 + (−1)(−16), 2·625 + (−1)·32 ] = [ 1250 + 16, 1250 − 32 ] = [ 1266, 1218 ].
    • Second row: [ 1·625 + 1·(−16), 1·625 + 1·32 ] = [ 609, 657 ].
  • Now scale by 1/3:
    • A⁴ = [ [1266/3, 1218/3], [609/3, 657/3] ] = [ [422, 406], [203, 219] ].
  • Final answer: A⁴ = [[422, 406], [203, 219]].

Example 2: Repeated eigenvalue and geometric multiplicity (non-diagonalizable case) Given A = [ [1, 1, 0], [0, 1, 0], [0, 0, 2] ].

  1. Eigenvalues (triangular matrix)
  • For upper triangular matrices, eigenvalues are diagonal entries: 1, 1, 2.
  • Algebraic multiplicities: λ = 1 (multiplicity 2), λ = 2 (multiplicity 1).
  1. Eigenvectors and eigenspaces
  • For λ = 1: Solve (A − I)x = 0.
    • A − I = [ [0, 1, 0], [0, 0, 0], [0, 0, 1] ]. Equations: y = 0 and z = 1·z = 0 ⇒ z = 0. x is free.
    • General eigenvector: [x, 0, 0]ᵀ = x·[1, 0, 0]ᵀ. So E₁ = span{ [1, 0, 0]ᵀ }, geometric multiplicity = 1 (less than algebraic multiplicity 2).
  • For λ = 2: Solve (A − 2I)x = 0.
    • A − 2I = [ [−1, 1, 0], [0, −1, 0], [0, 0, 0] ]. Equations: −x + y = 0 ⇒ y = x; and −y = 0 ⇒ y = 0 ⇒ x = 0. z is free ⇒ take z = 1.
    • Eigenvector v = [0, 0, 1]ᵀ. So E₂ = span{ [0, 0, 1]ᵀ }.
  1. Conclusion about diagonalisation
  • Total number of linearly independent eigenvectors = 1 (for λ = 1) + 1 (for λ = 2) = 2 < 3.
  • Hence A is not diagonalizable. This demonstrates geometric multiplicity < algebraic multiplicity leads to non-diagonalizability.

Example 3: Real symmetric matrix — orthogonal eigenvectors and orthogonal diagonalisation Given A = [ [6, 2], [2, 3] ], which is real symmetric (A = Aᵀ).

  1. Eigenvalues
  • det(A − λI) = det( [ [6 − λ, 2], [2, 3 − λ] ] ) = (6 − λ)(3 − λ) − 4 = λ² − 9λ + 14.
  • Solve λ² − 9λ + 14 = 0 ⇒ (λ − 7)(λ − 2) = 0 ⇒ eigenvalues 7 and 2 (both real, as expected for symmetric A).
  • Check: trace = 6 + 3 = 9 = 7 + 2 ✔; det = 6·3 − 2·2 = 18 − 4 = 14 = 7·2 ✔.
  1. Eigenvectors (choose orthonormal)
  • For λ = 7: (A − 7I) = [ [−1, 2], [2, −4] ]. Equation −x + 2y = 0 ⇒ x = 2y ⇒ pick y = 1 ⇒ v₁ = [2, 1]ᵀ.
    • Normalize: ‖v₁‖ = √(4 + 1) = √5 ⇒ u₁ = (1/√5)[2, 1]ᵀ.
  • For λ = 2: (A − 2I) = [ [4, 2], [2, 1] ]. Equation 4x + 2y = 0 ⇒ x = −(1/2)y ⇒ pick y = 2 ⇒ v₂ = [−1, 2]ᵀ.
    • Normalize: ‖v₂‖ = √(1 + 4) = √5 ⇒ u₂ = (1/√5)[−1, 2]ᵀ.
  • Check orthogonality: u₁·u₂ ∝ [2, 1]·[−1, 2] = −2 + 2 = 0 ✔.
  1. Orthogonal diagonalisation and use
  • Let Q = [ u₁ u₂ ] = (1/√5)[ [2, −1], [1, 2] ], and Λ = diag(7, 2).
  • Then A = Q Λ Qᵀ with QᵀQ = I. This is directly usable to compute, for example, e^{tA} = Q diag(e^{7t}, e^{2t}) Qᵀ for solving x′ = A x.

Common mistakes

  • Mixing up algebraic multiplicity and geometric multiplicity, or assuming they are always equal.
  • Assuming “distinct eigenvalues” is necessary for diagonalisation; it is sufficient but not necessary. Repeated eigenvalues can still yield enough independent eigenvectors.
  • For triangular matrices, not recognizing that eigenvalues are the diagonal entries (counted with multiplicity).
  • Arithmetic slips in determinants (sign errors) and in solving (A − λI)x = 0 (row-reduction mistakes). Always verify with trace and determinant relations.
  • Forgetting to check diagonalizability before attempting A = P D P⁻¹; or trying to form P⁻¹ when P is singular (columns not independent).
  • In symmetric cases, not taking advantage of orthogonality; mixing non-orthonormal eigenvectors leads to QᵀQ ≠ I and messy inverses.

For your exam

  • Derivations/proofs commonly asked:
    • Show λ is an eigenvalue iff det(A − λI) = 0 (include nontrivial solution argument).
    • Prove that eigenvectors corresponding to distinct eigenvalues are linearly independent.
    • Show sum of eigenvalues equals trace(A) and product equals det(A) (state for general case; give proof for diagonalizable case or outline using Schur form).
    • State and prove (or outline) that a real symmetric matrix has real eigenvalues and orthogonal eigenvectors (full spectral theorem statement; detailed proof may be deferred to linear/orthogonal transformations topic).
  • Computational numericals:
    • Find eigenvalues and eigenvectors for given 2×2 or 3×3 matrices.
    • Decide if a matrix is diagonalizable; if so, find P, D and compute A^k or e^{tA} (basic t values).
    • Handle repeated eigenvalues; compute geometric multiplicity and conclude about diagonalisation.
  • Diagrams: If asked to “illustrate eigenvectors,” draw the plane with a special invariant line; label v and Av on the same line.
  • Full-marks answers include: correct setup det(A − λI) = 0; clean, correct algebra; clearly stated eigenvectors as spans; a final conclusion about diagonalisation with justification; and quick trace/det checks.

Quick check

  1. Define eigenvalue and eigenvector of A. What equation must they satisfy?
  2. What are the eigenvalues of an upper triangular 3×3 matrix with diagonal entries 2, −1, 2?
  3. A has eigenvalues 1, 1, 4. Give two quick checks using trace and determinant for your computed eigenvalues of any 3×3 A.

Answers: (1) λ is an eigenvalue and x ≠ 0 an eigenvector if A x = λ x, equivalently (A − λI)x = 0. (2) 2, −1, 2 (diagonal entries, with multiplicity). (3) trace(A) should equal 1 + 1 + 4 = 6; det(A) should equal 1·1·4 = 4.

Finished this topic? Mark it so your progress, study plan and readiness keep up.

Stuck on something here?