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.

Linear and orthogonal transformations

Linear and orthogonal transformations: definitions, matrix representation, kernel/range, rank–nullity, change of basis, and properties of orthogonal (length/angle-preserving) transformations with examples.

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 model and process signals, stresses, and motions using vectors; a linear transformation is the clean mathematical rule that turns one vector into another while respecting addition and scaling, so it models systems that are “proportional and additive.” Orthogonal transformations are the rigid motions (rotations/reflections) that preserve lengths and angles—essential in robotics, graphics, structural analysis, and signal processing for coordinate rotations and numerical stability.

Key ideas

  • Assumptions: V, W are vector spaces over the same field (here ℝ) with defined addition and scalar multiplication.

  • Consequences: T(0) = 0, T(−v) = −T(v), T(∑ αi vi) = ∑ αi T(vi).

  • Matrix representation of a linear transformation

    • Given finite-dimensional spaces with ordered bases B_V = {v1, …, vn} for V and B_W = {w1, …, wm} for W, there exists a unique matrix of T in these bases, [T]{B_W←B_V} ∈ ℝ^{m×n}, such that if [x]{B_V} is the coordinate column of x in B_V, then [T(x)]{B_W} = [T]{B_W←B_V} [x]_{B_V}.
    • Special case: With standard bases in ℝ^n and ℝ^m, T(x) = A x, where A is the m×n matrix whose j-th column is T(e_j).
  • Kernel, range, rank, nullity

    • Kernel (null space): ker T = {x ∈ V : T(x) = 0}. It is a subspace of V.
    • Range (image): range T = im T = {y ∈ W : ∃ x ∈ V with y = T(x)}. It is a subspace of W.
    • Rank: rank T = dim(im T).
    • Nullity: nullity T = dim(ker T).
    • Rank–Nullity Theorem (finite-dimensional V): dim V = rank T + nullity T.
  • Invertibility

    • A linear transformation T: V → V (same domain and codomain dimension) is invertible iff it is bijective iff its matrix (in any basis) is invertible (det ≠ 0) iff ker T = {0} and rank T = dim V.
    • The inverse, if it exists, is linear and its matrix is [T]^{-1} in the same basis.
  • Composition and change of basis

    • Composition: If S: W → U and T: V → W are linear, then S∘T is linear and [S∘T] = [S][T] with compatible sizes.
    • Change of basis: If P is the change-of-basis matrix from basis B to the standard basis (columns are basis vectors), and A is the matrix of a linear operator T in the standard basis, then the matrix of T in basis B is A_B = P^{-1} A P. Conversely, A = P A_B P^{-1}.
  • Inner product, norm, angle in ℝ^n

    • Dot product: For x, y ∈ ℝ^n, x·y = x^T y.
    • Norm (length): ∥x∥ = √(x·x).
    • Angle: For nonzero x, y, cos θ = (x·y)/(∥x∥∥y∥).
  • Orthogonal transformations and orthogonal matrices

    • A linear T: ℝ^n → ℝ^n is an orthogonal transformation if it preserves inner product: for all x, y, ⟨T(x), T(y)⟩ = ⟨x, y⟩ (using the standard dot product). Equivalent statements: ∥T(x)∥ = ∥x∥ for all x, and angles are preserved.
    • The matrix A of T in the standard basis is an orthogonal matrix: A^T A = I (equivalently, A^{-1} = A^T). Columns (and rows) of A form an orthonormal set.
    • Determinant: det A = ±1. If det A = +1, A is a proper rotation; if det A = −1, A is an improper rotation (a rotation plus a reflection).
    • Eigenvalues: For real orthogonal A, every eigenvalue λ satisfies |λ| = 1. Over ℝ, eigenvalues are either ±1 or occur in complex conjugate pairs e^{±iθ} corresponding to planar rotations.
    • Angle/axis descriptions
      • In ℝ^2: Any orthogonal A with det +1 is a rotation by θ: A = [ [cos θ, −sin θ], [sin θ, cos θ] ]. Any with det −1 is a reflection through a line through the origin.
      • In ℝ^3: det +1 gives a rotation about some unit axis u by angle θ (Rodrigues’ formula); det −1 can be a reflection in a plane or a roto-reflection.
  • Projections and orthogonality

    • The orthogonal projection onto a subspace S preserves distances only along S and shortens other components; it is linear but not orthogonal (in the sense A^T A = I) unless S = ℝ^n. If Q has orthonormal columns spanning S, the projector is P = Q Q^T; the reflection across S is R = 2P − I (Householder: for a hyperplane with unit normal n, H = I − 2 n n^T).
  • Geometric interpretation and diagrams (what to draw)

    • Rotation in ℝ^2: Draw x- and y-axes, a vector v from the origin, and its image T(v) rotated anticlockwise by angle θ with an arc marking θ; label lengths ∥v∥ and ∥T(v)∥ equal, and show the angle at the origin.
    • Reflection in ℝ^2: Draw a line through the origin labeled ℓ (mirror line), a vector v, and its mirror image across ℓ, marking equal angles with ℓ.
  • Links to other topics

    • Eigenvalues and eigenvectors: For orthogonal A, eigenvalues have magnitude 1; reflections have eigenvalue −1 along the flipped direction, +1 along the mirror line/plane.
    • Diagonalisation and quadratic forms: Orthogonal matrices perform orthogonal similarity transforms that preserve x^T x and help diagonalise symmetric matrices (spectral theorem).
    • Rank of a matrix and systems of equations: Rank T equals the rank of its matrix; kernel solves A x = 0.

Formulas

  • rank(T) + nullity(T) = dim(V) — Rank–Nullity Theorem; dimensions are counts (dimensionless); V finite-dimensional.
  • A^T · A = I — Orthogonal matrix condition; A is square; entries dimensionless; applies to orthogonal transformations in standard inner product.
  • ∥T(x)∥ = ∥x∥ — Length preservation by orthogonal T; Euclidean norm; dimensionless in pure math.
  • T(x) · T(y) = x · y — Inner-product preservation; dot product; dimensionless.
  • det(A) = ±1 — Determinant of orthogonal matrix; applies to square orthogonal A; dimensionless.
  • A^{-1} = A^T — Inverse equals transpose for orthogonal A; dimensionless.
  • cos θ = trace(A)/2 — For a 2×2 rotation matrix A with det +1; dimensionless; applies when A is a pure rotation in ℝ^2. Also: for a 3×3 proper rotation A ∈ SO(3), cos θ = (trace(A) − 1)/2.
  • A(θ) = [[cos θ, −sin θ], [sin θ, cos θ]] — 2D rotation by angle θ (radians); entries dimensionless.
  • R_u(θ) = I·cos θ + (1 − cos θ)·u·u^T + [u]_×·sin θ — Rodrigues’ rotation (3D); u is a unit vector; [u]_× is the skew-symmetric cross-product matrix; dimensionless; applies to 3D rotations.
  • A_B = P^{-1} · A · P — Change of basis; P’s columns are basis vectors in standard coordinates; dimensionless.
  • columns(A) orthonormal ⇔ A^T · A = I — Characterisation of orthogonal matrices; dimensionless.

Worked examples

  1. Matrix, kernel, range, rank, nullity Given T: ℝ^3 → ℝ^2 defined by T(x) = A x with A = [ [1, 2, 0], [−1, 0, 3] ]. Find ker T, a basis for im T, rank, nullity.

Steps:

  1. Write T(x) = (1·x1 + 2·x2 + 0·x3, −1·x1 + 0·x2 + 3·x3).
  2. Kernel: Solve A x = 0.
    • Equations: x1 + 2x2 = 0; −x1 + 3x3 = 0.
    • From first: x1 = −2x2. From second: −(−2x2) + 3x3 = 0 ⇒ 2x2 + 3x3 = 0 ⇒ x2 = −(3/2)x3.
    • Then x1 = −2(−3/2 x3) = 3x3.
    • Let x3 = t. Then x = (3t, −1.5t, t) = t·(3, −3/2, 1).
    • Basis for ker T: {(3, −3/2, 1)}. It is common to clear fractions: multiply by 2 to get (6, −3, 2) as an equivalent basis vector. We can state ker T = span{(6, −3, 2)}.
  3. Image: columns of A span im T: c1 = (1, −1), c2 = (2, 0), c3 = (0, 3).
    • Check linear independence of columns in ℝ^2: At most 2 independent; pick c1 and c2.
    • Are c1 and c2 independent? Solve α(1, −1) + β(2, 0) = (0, 0): components give α + 2β = 0 and −α = 0 ⇒ α = 0 ⇒ β = 0. So yes, independent.
    • Therefore im T = span{(1, −1), (2, 0)} = ℝ^2.
  4. Rank: dim(im T) = 2.
  5. Nullity: dim(ker T) = dim(ℝ^3) − rank = 3 − 2 = 1 (also seen directly from the parametric solution).

Answer: ker T = span{(6, −3, 2)}, im T = ℝ^2, rank(T) = 2, nullity(T) = 1. Final check: A has full row rank 2, consistent.

  1. Checking orthogonality and identifying a 2D rotation Given A = [ [0, −1], [1, 0] ]. Verify A is orthogonal, find det A, and identify the rotation angle/direction.

Steps:

  1. Compute A^T A.
    • A^T = [ [0, 1], [−1, 0] ].
    • A^T A = [ [0, 1], [−1, 0] ] · [ [0, −1], [1, 0] ] = [ [1, 0], [0, 1] ] = I.
    • Therefore A is orthogonal.
  2. Determinant: det A = (0·0 − (−1)·1) = 1.
    • det = +1 ⇒ a proper rotation (no reflection).
  3. Identify angle θ: Compare with rotation matrix [[cos θ, −sin θ], [sin θ, cos θ]].
    • Here cos θ = 0 and sin θ = 1 ⇒ θ = π/2 radians (90°) anticlockwise.

Answer: A is orthogonal with det +1; it is a rotation by θ = π/2 anticlockwise about the origin.

Diagram to draw: x- and y-axes, a vector on the x-axis (e.g., v = (1, 0)) and its image T(v) = (0, 1) on the y-axis, with an arc showing 90° CCW.

  1. Change of basis for a linear operator on ℝ^2 Let T: ℝ^2 → ℝ^2 be T(x) = A x with A = [ [3, 1], [0, 2] ]. Let B = {b1, b2} be a new basis with b1 = (1, 1), b2 = (1, −1). Find the matrix A_B of T in basis B.

Steps:

  1. Form the change-of-basis matrix P from B to the standard basis: columns are b1, b2.
    • P = [ [1, 1], [1, −1] ].
  2. Compute P^{-1}.
    • det P = (1·(−1) − 1·1) = −2.
    • P^{-1} = (−1/2) · [ [−1, −1], [−1, 1] ] = (1/2) · [ [1, 1], [1, −1] ].
  3. Compute A_B = P^{-1} A P.
    • First A P = [ [3, 1], [0, 2] ] · [ [1, 1], [1, −1] ] = [ [4, 2], [2, −2] ].
    • Then A_B = P^{-1} (A P) = (1/2) · [ [1, 1], [1, −1] ] · [ [4, 2], [2, −2] ].
    • Multiply: [ [1, 1], [1, −1] ] · [ [4, 2], [2, −2] ] = [ [6, 0], [2, 4] ].
    • Now scale by 1/2: A_B = [ [3, 0], [1, 2] ].

Answer: In basis B, [T]_B = [ [3, 0], [1, 2] ]. (Note: This basis diagonalises the symmetric part only partially; here A_B becomes upper triangular with the eigenvalues 3 and 2 on the diagonal.)

Common mistakes

  • Calling any length-reducing map “orthogonal”: Orthogonal means length and angle preserving (A^T A = I), not just “related to perpendicularity.” Projections are not orthogonal matrices unless they are the identity.
  • Assuming every square matrix is diagonalisable by an orthogonal matrix: Only symmetric real matrices are guaranteed orthogonally diagonalisable (spectral theorem), not arbitrary matrices.
  • Dropping transpose in orthogonality checks: Checking A A = I instead of A^T A = I.
  • Using degrees inside trig functions without specifying: Rotation matrices use cos θ, sin θ with θ in radians in calculus/analysis formulas.

For your exam

  • Derivations: Rank–Nullity Theorem statement/application; proof that A^T A = I ⇒ T preserves dot product and norm; equivalence “columns orthonormal ⇔ A^T A = I.” Step-by-step algebra is expected.
  • Numericals: Given T by action on basis vectors, form its matrix; compute ker T and im T; verify orthogonality of a given matrix; find rotation angle from a 2×2 orthogonal matrix; perform change of basis A_B = P^{-1} A P.
  • Diagrams: 2D rotation/reflection—axes, original vector, image vector, rotation angle or mirror line labeled; equal lengths marked.
  • Full-marks answers include: Correct definitions with conditions; clean matrix computations with justified steps; explicit bases for subspaces; rank/nullity numbers; if orthogonality is claimed, show A^T A = I; if rotation angle is asked, show how cos θ and sin θ are read off consistently.

Quick check

  1. If columns of A ∈ ℝ^{n×n} are orthonormal, what is A^T A and what is det A?
  2. For T: ℝ^4 → ℝ^3 with rank 2, what is nullity(T)?
  3. Given A = [ [cos θ, −sin θ], [sin θ, cos θ] ] with θ ≠ 0, π, what are its eigenvalues over ℂ and what is det A?

Answers: 1) A^T A = I; det A = ±1. 2) Nullity = 4 − 2 = 2. 3) Eigenvalues e^{±iθ}; det A = 1.

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

Stuck on something here?