LA · whole subject on one page

Short notes: Linear Algebra

Every topic in revision form: the key ideas, formulas, the question types GATE asks, and the traps. Read it the night before.

1. 🧮 Matrices & Special Types

Medium ●●3 PYQs →

Short true/false statements: trace and transpose rules, invertibility, and what happens when a matrix satisfies A² = I or A³ = I.

  • Matrix operations and their rules. Addition and scalar multiplication work entry by entry. Multiplication is row times column and needs the inner sizes to match: (m×n)(n×p) = m×p. It is associative and distributive but not commutative: AB ≠ BA in general. Key rules: (AB)ᵀ = BᵀAᵀ, (AB)⁻¹ = B⁻¹A⁻¹ (order reverses), tr(AB) = tr(BA) even when AB and BA have different sizes, and (A + B)² = A² + AB + BA + B² (not A² + 2AB + B²).
    → Matrix multiplication is not commutative; transposes and inverses reverse the order; tr(AB) = tr(BA).
  • Special matrices. Symmetric: A = Aᵀ. Skew-symmetric: Aᵀ = −A (diagonal is 0). Orthogonal: AAᵀ = I (columns are orthonormal, det = ±1). Idempotent: A² = A. Involutory: A² = I. Nilpotent: Aᵏ = 0 for some k. Triangular / diagonal: zeros on one side of / off the diagonal. Every square matrix is a sum of a symmetric and a skew-symmetric part: A = ½(A + Aᵀ) + ½(A − Aᵀ).
    → Learn the defining equation of each type; it pins down the eigenvalues.
  • Inverse and adjoint. A⁻¹ exists exactly when det A ≠ 0 (A is non-singular). Then A⁻¹ = adj(A) / det(A), where adj(A) is the transpose of the cofactor matrix. Facts: A·adj(A) = det(A)·I, |adj A| = |A|ⁿ⁻¹, adj(AB) = adj(B)·adj(A), |A⁻¹| = 1/|A|. In practice, find A⁻¹ by Gauss–Jordan: row reduce [A | I] to [I | A⁻¹].
    → Invertible ⇔ det ≠ 0; A⁻¹ = adj A / det A; |adj A| = |A|ⁿ⁻¹.
  • Powers of a matrix. Look for a relation that closes the powers. If A² = cI then A¹³ = c⁶·A. If A³ = I then A⁸ = A². If A is idempotent then every power equals A; if nilpotent the powers eventually vanish. For eigenvalues, Aᵏ has eigenvalues λᵏ with the same eigenvectors. Never expand huge powers by multiplying.
    → Find the relation that closes the powers (A² = cI, A³ = I) or use λᵏ.

Formulas

Transpose of a product
(AB)ᵀ = BᵀAᵀ
Inverse of a product
(AB)⁻¹ = B⁻¹A⁻¹
Trace identity
tr(AB) = tr(BA); tr(A + B) = tr A + tr B
Adjoint
A · adj(A) = det(A) · I; A⁻¹ = adj(A) / det(A)
Determinant of the adjoint
|adj A| = |A|^(n−1)
Orthogonal matrix
AAᵀ = I, A⁻¹ = Aᵀ, det A = ±1
Symmetric + skew split
A = ½(A + Aᵀ) + ½(A − Aᵀ)
Invertible ⇔
det ≠ 0 ⇔ rank = n ⇔ columns independent ⇔ Ax = 0 only for x = 0

What GATE asks

  • ★★★True or false: properties of trace, transpose, inverse. Test each statement with a 2×2 example; remember AB ≠ BA but tr(AB) = tr(BA).
  • ★★★Powers like A⁸ when A³ = I. Reduce the exponent using the relation, then use det or eigenvalue facts.
  • ★★★Is a matrix invertible?. Equivalent to det ≠ 0, rank = n, or no zero eigenvalue.

Traps

(A + B)² ≠ A² + 2AB + B² because AB ≠ BA.
tr(AB) = tr(BA) is true, but tr(ABC) ≠ tr(ACB) in general (only cyclic shifts are allowed).
A² = 0 does not mean A = 0.
adj(A) is the transpose of the cofactor matrix; forgetting the transpose gives the wrong inverse.

2. 🔢 Determinants

High yield ●●●4 PYQs →

Row-operation rules, det(AB), det(kA) and special matrices (Vandermonde, rank 1, permuted rows) appear in almost every paper.

  • Definition and cofactor expansion. For 2×2, det = ad − bc. In general expand along any row or column: det A = Σ aᵢⱼ·Cᵢⱼ, where the cofactor Cᵢⱼ = (−1)ⁱ⁺ʲ·Mᵢⱼ and Mᵢⱼ is the minor (delete row i and column j). Expand along the row or column with the most zeros. The determinant is the signed volume scale factor of the map x ↦ Ax.
    → det = Σ entry · sign · minor; expand along the row or column with the most zeros.
  • Properties of determinants. Swap two rows → sign flips. Multiply a row by k → det × k (so det(kA) = kⁿ·det A). Add a multiple of one row to another → no change. det(Aᵀ) = det A. det(AB) = det A · det B. det(A⁻¹) = 1/det A. Two equal rows or a zero row → det 0. Triangular matrix → product of the diagonal. det(A + B) ≠ det A + det B.
    → Swap flips the sign; scaling a row scales det; adding rows does nothing; det(kA) = kⁿ det A.
  • Special determinants. Vandermonde: rows (1, x, x², …) give Π(xⱼ − xᵢ) over i < j. Rank 1 (an outer product uvᵀ) with n ≥ 2 → det 0. Block triangular → product of block determinants. A matrix whose rows are the same as another matrix but permuted has det = ± that det (sign of the permutation). The matrix J − I (all ones minus identity, n×n) has det (−1)ⁿ⁻¹(n − 1).
    → Spot Vandermonde, rank 1 or triangular shapes: the determinant is then a product or 0.

Formulas

2×2 determinant
|a b; c d| = ad − bc
Cofactor expansion
det A = Σⱼ aᵢⱼ · (−1)^(i+j) · Mᵢⱼ
Scalar multiple
det(kA) = kⁿ det(A) (n×n)
Product
det(AB) = det(A) det(B); det(Aᵏ) = det(A)ᵏ
Inverse and transpose
det(A⁻¹) = 1/det A; det(Aᵀ) = det A
Row swap
swapping two rows multiplies det by −1
Vandermonde
det[1 xᵢ xᵢ² …] = Π_{i<j} (xⱼ − xᵢ)
Triangular matrix
det = product of diagonal entries

What GATE asks

  • ★★★How does det change after row operations?. Swap flips sign, scaling multiplies, row addition keeps it; compare two similar-looking matrices.
  • ★★★Determinant of a structured matrix. Spot Vandermonde, rank 1, repeated rows or triangular form instead of expanding.
  • ★★★Product of eigenvalues = determinant. Compute |det| rather than the eigenvalues.

Traps

det(kA) = kⁿ det A, not k·det A.
det(A + B) is not det A + det B, and det(AB) = det A · det B is not a sum.
Row addition keeps det, but row scaling changes it: do not treat them alike.
A matrix with non-zero entries can still have det 0 (dependent rows).

3. 📐 Rank, Elimination & LU

High yield ●●●4 PYQs →

Gaussian elimination is the engine behind rank, solving, inverses and LU. Expect rank counting, rank inequalities and one LU question.

  • Row echelon form and Gaussian elimination. Use three row operations: swap two rows, scale a row, add a multiple of one row to another. Forward elimination makes zeros below each pivot (echelon form); continuing upward and scaling pivots to 1 gives the reduced row echelon form (RREF). Row operations never change the row space, the solution set of Ax = b, or the rank.
    → Elimination uses swap, scale and add; echelon form exposes the pivots.
  • Rank. The rank is the number of pivots = the number of independent rows = the number of independent columns (so rank A = rank Aᵀ ≤ min(m, n)). Rules: rank(AB) ≤ min(rank A, rank B), rank(A + B) ≤ rank A + rank B, rank(AᵀA) = rank A. A matrix uvᵀ (column times row) has rank 1. An n×n matrix is invertible exactly when rank = n.
    → Rank = number of pivots; rank(AB) ≤ min; rank(A + B) ≤ sum.
  • Rank–nullity theorem. For an m×n matrix, rank(A) + nullity(A) = n, where nullity = dimension of the null space {x : Ax = 0}. Free variables in the RREF = nullity. So a wide matrix (more columns than rows) always has a non-trivial null space.
    → rank + nullity = number of columns.
  • LU decomposition. If no row swaps are needed, A = LU with L lower triangular (1s on the diagonal, holding the elimination multipliers) and U upper triangular (the echelon form). To solve Ax = b: solve Ly = b by forward substitution, then Ux = y by back substitution. det A = product of the diagonal of U. A is singular exactly when some diagonal entry of U is 0. If a zero pivot appears, swap rows: PA = LU.
    → A = LU; solve Ly = b then Ux = y; det A = Π diag(U).

Formulas

Rank bound
rank(A) ≤ min(m, n); rank A = rank Aᵀ
Product
rank(AB) ≤ min(rank A, rank B)
Sum
rank(A + B) ≤ rank A + rank B
Rank–nullity
rank(A) + nullity(A) = n (number of columns)
Outer product
rank(uvᵀ) = 1 (u, v ≠ 0)
LU solve
Ax = b → Ly = b, then Ux = y
LU determinant
det A = Π uᵢᵢ (L has 1s on its diagonal)
Gaussian elimination cost
≈ n³/3 multiplications

What GATE asks

  • ★★★Find the rank of a given matrix. Spot dependent rows (for example R3 = R1 + R2) before eliminating.
  • ★★★True or false: rank and determinant statements. rank(AB) ≤ min; rank(A + B) ≤ sum; counter-examples with diagonal matrices.
  • ★★★LU decomposition values and truth statements. Compute the multipliers, then reason about det L = 1 and singular ⇔ zero on the diagonal of U.

Traps

rank(AB) is not rank A × rank B.
Row operations change the column space, but not the row space or the rank.
L is unit lower triangular only in the Doolittle form; other conventions put the 1s on U.
LU without pivoting can fail even for invertible matrices (zero pivot): you then need PA = LU.

4. 🔗 Systems of Linear Equations

High yield ●●●5 PYQs →

Consistency tests, parameter questions ("for which k is there no / unique / infinite solutions") and homogeneous systems are the most predictable linear-algebra marks.

  • Consistency: the rank test. Write the system as Ax = b with A of size m×n and the augmented matrix [A | b]. Rouché–Capelli: the system is consistent exactly when rank A = rank [A | b]. Then if the rank equals n there is a unique solution, and if it is less than n there are infinitely many (n − rank free variables). If rank A < rank [A | b] there is no solution.
    → Consistent ⇔ rank A = rank [A | b]; then rank = n gives one solution, rank < n gives infinitely many.
  • Homogeneous systems Ax = 0. x = 0 is always a solution (the system is always consistent). A non-trivial solution exists exactly when rank A < n, that is, when the columns are dependent. For a square A this means det A = 0. If there are more unknowns than equations (n > m), non-trivial solutions always exist, and they form a space of dimension at least n − m.
    → Ax = 0 is always consistent; non-trivial solutions ⇔ rank < n.
  • Systems with a parameter. For a square system with a parameter k, compute det A as a function of k. For every k with det ≠ 0 the solution is unique. For each root of det = 0, row-reduce the augmented matrix at that value: if a row becomes [0 … 0 | non-zero] the system has no solution; if a row of zeros appears with consistent right-hand side it has infinitely many.
    → Find det(k) = 0, then test each root separately; det ≠ 0 means one solution.
  • Cramer's rule and the shape of the solution set. For a square system with det A ≠ 0, xᵢ = det(Aᵢ) / det(A), where Aᵢ is A with column i replaced by b. Over the reals the solution set of a consistent system is either a point, a line / plane / higher-dimensional flat, but never exactly two or a finite number greater than one: if x₁ and x₂ are solutions, so is every point on the line through them. The general solution is x = x_particular + (any solution of Ax = 0).
    → Cramer: xᵢ = det Aᵢ / det A; solution counts are 0, 1 or ∞, never 2.

Formulas

Rouché–Capelli
consistent ⇔ rank A = rank [A | b]
Unique solution
rank A = rank [A | b] = n
Infinitely many
rank A = rank [A | b] < n (n − rank free variables)
No solution
rank A < rank [A | b]
Homogeneous non-trivial solution
exists ⇔ rank A < n (⇔ det A = 0 for square A)
Cramer's rule
xᵢ = det(Aᵢ) / det(A)
General solution
x = x_p + x_h, where A x_h = 0
More unknowns than equations
m < n ⇒ Ax = 0 has a non-trivial solution

What GATE asks

  • ★★★For which k does the system have no / one / infinite solutions?. Find det as a function of k, then test each root separately by row reduction.
  • ★★★Number of solutions of a given system. Compare rank A and rank [A | b] with the number of unknowns.
  • ★★★Statements about Ax = 0 for an m×n matrix. Use rank–nullity: dim of solution space = n − rank ≥ n − m.

Traps

"Fewer equations than unknowns" does not make the system consistent: 0x + 0y = 1 has no solution.
Equal numbers of equations and unknowns does not guarantee a unique solution (det can be 0).
A real linear system never has exactly two solutions.
Always test the special k values separately: det = 0 only tells you the system is not unique.

5. 🎯 Eigenvalues & Eigenvectors

High yield ●●●14 PYQs →

The biggest linear-algebra topic: trace and determinant relations, powers, special matrices, Cayley–Hamilton and diagonalisation appear almost every year.

  • Eigenvalues, eigenvectors and the characteristic polynomial. A non-zero vector v is an eigenvector of A with eigenvalue λ if Av = λv: A only stretches v. Eigenvalues are the roots of the characteristic equation det(A − λI) = 0; for each λ, solve (A − λI)v = 0 for the eigenvectors. An n×n matrix has n eigenvalues counted with multiplicity (some may be complex). For 2×2: λ² − (trace)λ + det = 0.
    → Av = λv; eigenvalues are the roots of det(A − λI) = 0.
  • Trace, determinant and powers. Sum of eigenvalues = trace; product of eigenvalues = determinant. So A is singular exactly when 0 is an eigenvalue. Transformations: Aᵏ → λᵏ, A⁻¹ → 1/λ, A + cI → λ + c, cA → cλ, and Aᵀ has the same eigenvalues as A (but different eigenvectors). AB and BA have the same non-zero eigenvalues.
    → Σλ = trace, Πλ = det; Aᵏ, A⁻¹, A + cI map each λ the same way.
  • Eigenvalues of special matrices. Triangular or diagonal: the diagonal entries. Real symmetric: all real, with orthogonal eigenvectors. Skew-symmetric: purely imaginary or 0. Orthogonal: |λ| = 1. Idempotent (A² = A): 0 or 1. Involutory (A² = I): ±1. Nilpotent: all 0. Rank 1 (uvᵀ): one eigenvalue vᵀu, the rest 0. All-ones n×n: n once and 0 (n − 1 times). A real matrix has complex eigenvalues in conjugate pairs.
    → Read the eigenvalues from the matrix type: triangular, symmetric, orthogonal, idempotent, rank 1.
  • Cayley–Hamilton theorem. Every square matrix satisfies its own characteristic equation: if p(λ) = det(A − λI), then p(A) = 0. For 2×2: A² − (tr A)A + (det A)I = 0. Use it to reduce high powers of A to a low-degree polynomial in A and to find A⁻¹: from A² − tA + dI = 0 we get A⁻¹ = (tI − A)/d.
    → p(A) = 0; use it for the inverse and for high powers.
  • Diagonalisation and multiplicities. A is diagonalisable if A = PDP⁻¹ with D diagonal, which happens exactly when A has n linearly independent eigenvectors. Each eigenvalue has an algebraic multiplicity (its multiplicity as a root) and a geometric multiplicity (the dimension of its eigenspace, which is ≤ the algebraic one). Diagonalisable ⇔ geometric = algebraic for every eigenvalue. n distinct eigenvalues are enough; real symmetric matrices are always diagonalisable. Then Aᵏ = PDᵏP⁻¹.
    → Diagonalisable ⇔ n independent eigenvectors ⇔ geometric = algebraic for all λ.
  • Eigenvectors: independence and orthogonality. Eigenvectors for different eigenvalues are linearly independent. For a real symmetric matrix, eigenvectors for different eigenvalues are also orthogonal (dot product 0). Any non-zero multiple of an eigenvector is an eigenvector, and the eigenvectors of one eigenvalue (plus 0) form a subspace, the eigenspace. A matrix with only one eigenvector direction is not diagonalisable and has a repeated eigenvalue.
    → Different eigenvalues ⇒ independent eigenvectors; symmetric ⇒ orthogonal.

Formulas

Definition
Av = λv, v ≠ 0
Characteristic equation
det(A − λI) = 0
Sum and product
Σλᵢ = trace(A); Πλᵢ = det(A)
2×2 shortcut
λ² − (trace)λ + (det) = 0
Function of a matrix
eigenvalues of Aᵏ: λᵏ; A⁻¹: 1/λ; A + cI: λ + c
Rank 1 matrix
uvᵀ has non-zero eigenvalue vᵀu
Cayley–Hamilton (2×2)
A² − (tr A)A + (det A)I = 0
Diagonalisation
A = PDP⁻¹; Aᵏ = PDᵏP⁻¹
Multiplicities
1 ≤ geometric ≤ algebraic; diagonalisable ⇔ equal for all λ

What GATE asks

  • ★★★Product or sum of eigenvalues. Use det and trace; do not solve the characteristic polynomial.
  • ★★★Eigenvalues of Aᵏ, A⁻¹, A + cI. Find the eigenvalues of A first (often from A² = cI), then transform them.
  • ★★★Which vectors are eigenvectors of a given matrix?. Multiply each option by the matrix and see whether the result is a multiple of the same vector.
  • ★★★Statements about symmetric / rank-1 / real matrices. Recall: symmetric ⇒ real eigenvalues and orthogonal eigenvectors; complex roots come in pairs.

Traps

det(A) is the product of eigenvalues; a matrix with a zero row has eigenvalue 0 but not necessarily a zero diagonal.
Aᵀ has the same eigenvalues as A but not the same eigenvectors.
Repeated eigenvalues do not make a matrix non-diagonalisable by themselves: compare geometric and algebraic multiplicity.
Eigenvalues of A + B are not the sums of eigenvalues of A and B.

6. 🧭 Vector Spaces

Medium ●●1 PYQs →

Dimension counting: basis, span, independence and the four fundamental subspaces of a matrix, usually in one or two short questions.

  • Vector spaces and subspaces. A vector space is a set closed under addition and scalar multiplication (Rⁿ is the standard example). A subspace is a non-empty subset that is itself closed under both operations. Quick test: it must contain the zero vector and be closed under u + v and c·u. Lines and planes through the origin are subspaces; those not through the origin are not.
    → Subspace = contains 0 and is closed under addition and scalar multiplication.
  • Span and linear independence. The span of vectors is the set of all their linear combinations. Vectors are linearly independent if the only combination equal to 0 is the trivial one. Test: put them as columns of a matrix; independent ⇔ rank = number of vectors. More than n vectors in Rⁿ are always dependent, and any set containing 0 is dependent.
    → Independent ⇔ rank = number of vectors.
  • Basis and dimension. A basis is a set of independent vectors that spans the space. Every basis has the same size, the dimension. In Rⁿ any n independent vectors form a basis. For subspaces U and W: dim(U + W) = dim U + dim W − dim(U ∩ W). So two 4-dimensional subspaces of a 6-dimensional space meet in dimension at least 2.
    → dim(U ∩ W) ≥ dim U + dim W − n.
  • The four fundamental subspaces. For an m×n matrix A of rank r: the column space (dimension r, inside Rᵐ), the row space (dimension r, inside Rⁿ), the null space {x : Ax = 0} (dimension n − r) and the left null space (dimension m − r). The row space and the null space are orthogonal complements in Rⁿ. Pivot columns of A give a basis for the column space; the non-zero rows of the RREF give a basis for the row space.
    → Column = row = r; null = n − r; left null = m − r.

Formulas

Subspace test
0 ∈ U; u, v ∈ U ⇒ u + v ∈ U; u ∈ U, c ∈ R ⇒ cu ∈ U
Independence test
columns independent ⇔ rank = number of columns
Dimension of Rⁿ
n (any n independent vectors form a basis)
Sum of subspaces
dim(U + W) = dim U + dim W − dim(U ∩ W)
Column space / row space
both have dimension rank(A)
Null space
dimension = n − rank(A)
Left null space
dimension = m − rank(A)
Orthogonality
row space ⊥ null space; column space ⊥ left null space

What GATE asks

  • ★★★Smallest or largest dimension of an intersection of subspaces. Use dim(U + W) ≤ n and the dimension formula.
  • ★★★Is this set a subspace / a basis?. Check the zero vector and closure; for a basis check independence and the count.
  • ★★★Dimension of null / column space from the rank. null = n − rank, column = rank.

Traps

A set not containing the zero vector is never a subspace (a plane x + y = 1 fails).
The union of two subspaces is usually not a subspace; their sum is.
Null-space dimension counts columns, not rows: nullity = n − rank.
A spanning set can be dependent; a basis is spanning and independent.