GATE Electronics (EC): Linear Algebra (Engineering Mathematics: rank, linear systems, eigenvalues, projections and least squares), previous-year questions solved

Every linear algebra question from the last five GATE Electronics and Communication papers, 10 in all: when Ax = b has no solution, rank and determinants with unknown entries, when three vectors stop being a basis, eigenvalues of outer-product and similar matrices, and the projections behind least squares. Each answer matches the official key and was recomputed by code in exact fractions, with random matrices for every 'always true' claim.

Questions from official GATE EC papers (IIT Kharagpur 2022, IIT Kanpur 2023, IISc Bengaluru 2024, IIT Roorkee 2025, IIT Guwahati 2026); solutions by Skill-Lync. · Updated 18 Nov 2026 · 10 questions

GATE 2022 · EC · Q.12MCQ · 1 markDeterminants, rank and systems of linear equations

Does Ax = b with A = [[1, -sqrt2, 3], [-1, sqrt2, -3]] and b = [1, 3] have one, many, none or two solutions?

Consider a system of linear equations , where
This system of equations admits ______.
  1. (A)a unique solution for
  2. (B)infinitely many solutions for
  3. (C)no solutions for
  4. (D)exactly two solutions for

Answer: (C) Official key: C

  1. Write the equations out: and .
  2. The second row of is exactly times the first row, so .
  3. Add the two equations: the left side cancels to and the right side is . That gives , which is impossible.
  4. Same thing in rank language: the augmented matrix has rank , which is greater than . A system is solvable only when the two ranks are equal, so there is no solution: option (C).

Trap: seeing rank unknowns and jumping to 'infinitely many solutions', option (B). That is true only when the system is consistent. Here the right side breaks the pattern of the rows: for the rank would stay and there would be infinitely many solutions. Option (D) is never possible, since a linear system has , or infinitely many solutions. The code computed the ranks exactly in sympy (1 and 2), found a least-squares residual of 2.828 (not zero), and checked that gives equal ranks (1).

Where this shows up in industry

Consistency of Ax = b is the question behind every overdetermined or rank-deficient measurement setup, for example a channel-estimation or calibration problem where two measurements are the same combination of the unknowns but give different readings. The residual of the least-squares fit is then the signal that the model, not the arithmetic, is wrong.

GATE 2024 · EC · Q.30NAT · 1 markDeterminants, rank and systems of linear equations

Value of alpha for which {[2 -3 alpha], [3 -1 3], [1 -5 7]} does not form a basis of R^3

Let and denote the set of real numbers and the three dimensional vector space over it, respectively. The value of for which the set of vectorsdoes not form a basis of is ______.

Answer: Official key: 5 to 5

  1. Three vectors form a basis of exactly when they are linearly independent, that is, when the determinant of the matrix with these vectors as rows is non-zero. They fail to be a basis exactly when that determinant is .
  2. Expand along the first row: .
  3. .
  4. Set it to zero: , so .
  5. Check: for the first vector is the average of the other two, , so the three are dependent.

Trap: a sign slip on the term. The cofactor of is , so the term is . Flipping that sign gives and ; the determinant at is , not . The code solved in sympy (root [5]), found rank at and rank at , found the dependency (coefficients of the three rows), and a -point scan of the smallest singular value over is lowest at .

Where this shows up in industry

Linear independence of a set of vectors is the condition for a set of training sequences, steering vectors or basis functions to be usable: if one is a combination of the others, the receiver cannot separate them. In array processing, two antennas whose steering vectors become dependent at some angle cannot resolve sources there.

GATE 2025 · EC · Q.11MCQ · 1 markDeterminants, rank and systems of linear equations

For which combinations of alpha, beta, gamma is the rank of the upper-triangular 4 x 4 matrix A at least three?

Consider the matrix below:For which of the following combinations of , , and , is the rank of at least three?
(i) and .
(ii) .
(iii) and .
(iv) .
  1. (A)Only (i), (iii), and (iv)
  2. (B)Only (iv)
  3. (C)Only (ii)
  4. (D)Only (i) and (iii)

Answer: (A) Official key: A

  1. The rank is the number of independent rows. The first two rows, and , are always independent (their leading block has determinant ), so the rank is at least . The question is whether rows 3 and 4 add at least one more.
  2. (i) , : row 3 is and row 4 is . They are non-zero and parallel to each other, so together they add exactly one. Rank . Yes.
  3. (ii) : rows 3 and 4 are zero. Rank . No.
  4. (iii) , : row 3 is , row 4 is zero. Rank . Yes.
  5. (iv) : the matrix is upper triangular with diagonal , all non-zero, so the determinant is and the rank is . Yes.
  6. The rank is at least three for (i), (iii) and (iv): option (A).

Trap: reading (i) as rank-deficient because puts a zero on the diagonal, and so choosing (B) or (D) and dropping one of the cases. A zero on the diagonal of a triangular matrix lowers the rank by one only when the rest of that row does not supply a pivot; here supplies one in the last column. The code computed the rank exactly in sympy for several values in each case: (i) [3], (ii) [2], (iii) [3], (iv) [4].

Where this shows up in industry

Rank counts how many independent channels a matrix carries. For a MIMO link, the rank of the channel matrix is the number of independent data streams that can be sent at once, and a zero pivot like the one in case (i) is a lost spatial mode.

GATE 2026 · EC · Q.26MSQ · 1 markDeterminants, rank and systems of linear equations

For which (a, b) is det M non-zero, where M = [[2, 1, 1], [1, 3, 0], [-1, a, b]]?

Consider the matrix .
Which of the following options is/ are TRUE if ?
  1. (A) and
  2. (B) and
  3. (C) and
  4. (D) and

Answer: (B) and (D) Official key: B;D

  1. Expand along the second column or the first row. Along the first row: .
  2. .
  3. Test each option. (A): , singular. (B): . (C): , singular. (D): .
  4. The determinant is non-zero for (B) and (D).

Trap: ticking (A) and (C), which are the pairs that make the determinant zero; the question asks for . The code expanded the determinant in sympy (), evaluated it exactly for each option (A: 0, B: 6, C: 0, D: -23/2) and confirmed with floating-point determinants.

Where this shows up in industry

A non-zero determinant means the matrix can be inverted, which is the condition for a zero-forcing equaliser or a system-of-equations solver to exist. The pairs that make it zero, as in (A) and (C), are the parameter values where a design becomes singular and its inverse blows up.

GATE 2022 · EC · Q.37MCQ · 2 marksEigenvalues, eigenvectors and similar matrices

Eigenvalues of the 3 x 3 matrix A = alpha v1 v1^T + beta v2 v2^T for orthonormal 3 x 1 vectors v1, v2

Let be two non-zero real numbers and be two non-zero real vectors of size . Suppose that , , and . Let be the matrix given by:The eigenvalues of are ______.
  1. (A)
  2. (B)
  3. (C)
  4. (D)

Answer: (A) Official key: A

  1. Multiply by : . So is an eigenvalue with eigenvector .
  2. Multiply by : . So is an eigenvalue with eigenvector .
  3. is but built from only two directions, so its rank is . A vector orthogonal to both and gives , so is the third eigenvalue.
  4. The eigenvalues are : option (A). Check with the trace: , equal to .

Trap: treating the two terms like a sum or difference of numbers and writing , option (B). Take , and , : the exact eigenvalues are [0, 2, 3], not . The code tested all four options on 40 random orthonormal pairs with random non-zero of either sign (QR of a random matrix, seed fixed) and only (A) matched every time; that exact example has rank 2.

Where this shows up in industry

Writing a matrix as a sum of scaled outer products of orthonormal vectors is its spectral decomposition, the idea behind principal component analysis and the singular value decomposition used to find the dominant modes of a MIMO channel or a covariance matrix. The scale factors are the eigenvalues, and rank two means only two modes carry energy.

GATE 2023 · EC · Q.15MCQ · 1 markEigenvalues, eigenvectors and similar matrices

Eigenvalue and eigenvector sets of A when B = P^-1 A P has eigenvalues lambda_k and eigenvectors v_k

Let the sets of eigenvalues and eigenvectors of a matrix be and , respectively. For any invertible matrix , the sets of eigenvalues and eigenvectors of the matrix , where , respectively, are
  1. (A) and
  2. (B) and
  3. (C) and
  4. (D) and

Answer: (C) Official key: C

  1. means . Matrices related this way are called similar.
  2. Start from and replace : .
  3. Multiply both sides on the left by : .
  4. So is an eigenvector of with the same eigenvalue . The eigenvalues are unchanged and the eigenvectors become : option (C).

Trap: option (B), because the eigenvalues do not change under similarity but the eigenvectors do (they are transformed by ). Option (D) uses , which is the map in the other direction: it would be right for . The code built 30 random matrices and invertible , formed and found: eigenvalues of and agree, to a relative residual of at most 4.4e-14, while itself fails (residual up to 4.09) and fails (up to 3.98); scaling the eigenvalues by also fails.

Where this shows up in industry

A change of basis leaves the eigenvalues, and therefore the stability and the natural frequencies of a state-space model, unchanged; only the coordinates of the modes move. This is why a filter can be realised in different state-space forms (direct form, parallel, cascade) with the same poles.

GATE 2024 · EC · Q.55MSQ · 2 marksEigenvalues, eigenvectors and similar matrices

Which vectors are eigenvectors of [[1, k], [2, 1]] with k a positive real number?

Consider the matrix , where is a positive real number. Which of the following vectors is/are eigenvector(s) of this matrix?
  1. (A)
  2. (B)
  3. (C)
  4. (D)

Answer: (A) and (B) Official key: A;B

  1. Characteristic equation: , so . Both are real because .
  2. For : gives , so . Taking : , option (B).
  3. For : , so and , option (A).
  4. Options (C) and (D): is parallel to only if . For both vectors this expression equals , which is non-zero for . So they are not eigenvectors.
  5. Eigenvectors: (A) and (B).

Trap: stopping after finding one eigenvector (usually the one) and ticking only (B); the matrix has two eigenvalues, so two eigenvector directions. Another is to put the in the first component, as in (C) and (D), which comes from mixing up which entry of the row to divide by. The code tested every option at eight values of from to (A and B always satisfy , with and at , that is ), and sympy gave the parallel-test value for (C) and (D) for all positive .

Where this shows up in industry

Eigenvalues and eigenvectors of a 2 x 2 state matrix give the natural modes of a second-order system: the eigenvalues set the growth or decay and the eigenvectors set the mix of the two states in each mode. A pair of real eigenvalues 1 plus or minus the square root of 2k, as here, with one positive, signals an unstable saddle-type mode.

GATE 2023 · EC · Q.11MCQ · 1 markVectors, projections and least squares

Value of alpha minimising the length of e in v1 = alpha v2 + e, with v1 = (1, 2, 0) and v2 = (2, 1, 3)

Let and be two vectors. The value of the coefficient in the expression , which minimizes the length of the error vector , is
  1. (A)
  2. (B)
  3. (C)
  4. (D)

Answer: (C) Official key: C

  1. The error is . Its length is smallest when is perpendicular to (the foot of the perpendicular from onto the line along ).
  2. Set : , so .
  3. and .
  4. , option (C).

Trap: putting the ratio upside down, , option (A); the minus signs in (B) and (D) have no source in the data, since both dot products are positive. The code found the exact fraction with Fractions, confirmed , and a -point scan of over put the minimum at (smallest length ); the derivative of in sympy gives .

Where this shows up in industry

This is the one-tap least-squares estimate: projecting a received vector onto a known pilot to get the channel gain, and the same projection is how a matched filter or a correlator picks out a known waveform from a measurement. The remaining error is orthogonal to the pilot, which is why it is the minimum-energy residual.

GATE 2023 · EC · Q.38MCQ · 2 marksVectors, projections and least squares

Trace of P = x x^T for an n x 1 real column vector x of length l = sqrt(x^T x)

Let be an real column vector with length . The trace of the matrix is
  1. (A)
  2. (B)
  3. (C)
  4. (D)

Answer: (A) Official key: A

  1. The diagonal entries of are .
  2. The trace is the sum of the diagonal entries: .
  3. By the definition of the length, . So : option (A).

Trap: option (C), . The length is the square root of the sum of squares, but the trace is the sum of squares itself. Example : and , not . The code checked against for random vectors of sizes to (largest difference 7.1e-15) and tested all four options on random vectors: only passes.

Where this shows up in industry

The matrix x x^T is the instantaneous correlation matrix of a signal vector, and its trace is the total signal power across the n channels or antennas, x^T x. Averaging x x^T over time gives the covariance matrix used in beamforming and in principal component analysis.

GATE 2025 · EC · Q.59NAT · 2 marksVectors, projections and least squares

Minimum over x of the 2-norm of a x - b, with a = (1, 1) and b = (0, 3 sqrt 2)

Consider the vectorsFor real-valued scalar variable , the value ofis ______ (rounded off to two decimal places).
denotes the Euclidean norm, i.e., for , .

Answer: Official key: 2.95 to 3.05

  1. runs along the line through the origin in the direction . The smallest distance from to that line is the length of the perpendicular from to the line.
  2. The best is the projection coefficient .
  3. The residual is , which is perpendicular to (its dot product with is ).
  4. Its length is .

Trap: giving the squared norm (the least-squares cost) instead of the norm, or taking and reporting . The code used numpy least squares (, minimum ), a -point scan of over (minimum ), and sympy (, value ); the squared cost is .

Where this shows up in industry

Minimising the norm of Ax minus b is the least-squares problem behind channel estimation, zero-forcing equalisers and linear prediction. The minimum value is the energy left over after the best fit, and the leftover is always perpendicular to the columns of A.

From GATE prep to a signal-processing or communication career

Talk to a counsellor.

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.

Talk to a career counsellor

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.