GATE Mechanical: Numerical Methods (Engineering Mathematics), previous-year questions solved
Numerical methods is the smallest Engineering Mathematics section in GATE Mechanical, about one quick 1-mark question a paper, and the easiest to score: the five keyed papers from 2022 to 2026 set 5, all on Newton-Raphson, the trapezoidal and Simpson's rules or a time-stepping scheme. Each is solved here with the shortcut that skips most of the arithmetic.
Questions from official GATE ME papers (IIT Kharagpur 2022, IISc Bengaluru 2024, IIT Roorkee 2025, IIT Guwahati 2026); solutions by Skill-Lync. · Updated 16 Nov 2026 · 5 questions
Newton-Raphson method for solving equations is based on which series?
- (A)Taylor series
- (B)Fourier series
- (C)Laurent series
- (D)Power series
Answer: (A) Official key: A
- Expand about the current guess and keep the first two terms of the Taylor series: .
- Set this straight-line approximation to zero and solve for the step: , so That is the whole method: option (A).
- The dropped Taylor term also explains how fast it converges. The error obeys , so the number of correct digits roughly doubles every step once you are close.
Check on Newton's own example, from : the iterates are , , , with errors , , . From the second step on, each new error is to within 1%, the constant at the root .
Why not (D)? A Taylor series is a power series, but a specific one: its coefficients are the derivatives of at the expansion point, and that is what puts in the formula. Fourier series use sines and cosines; Laurent series allow negative powers around a singularity.
Every nonlinear FEA solve (plasticity, contact, large deflection) is a Newton-Raphson loop: the solver linearises the equilibrium equations with the tangent stiffness matrix, solves for a correction and repeats until the residual force is below tolerance. When a run 'fails to converge' in ANSYS or Abaqus, it is these iterations that stopped shrinking the residual, often because the load step was too large for the straight-line approximation to hold.
Trapezoidal rule with four intervals for ∫ dx/(1 + x) from 0 to 4: what is m − n?
If represents numerically evaluated value of the above integral using Trapezoidal rule by considering four equal subintervals in the range of , then is
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- Exact: .
- Trapezoidal rule, , nodes with : (exactly ).
- : option (A).
You can fix the sign before calculating anything. is convex (), so every trapezoid's top edge, a chord, lies above the curve and the rule overestimates: . That rules out (B), which is . Simpson's rule on the same four strips gives , an error of only , but the question asks for the trapezoidal rule.
The trapezoidal rule reappears in implicit structural dynamics as the Newmark average-acceleration method (, ): it assumes the acceleration over a time step is the average of its end values, which is a trapezoid in time. It is unconditionally stable for linear problems, which is why transient FEA solvers can take large time steps with it, and why accuracy, not stability, decides the step size.
Simpson's rule with 10 intervals for ∫(4x² + 2x + 6) dx from 1 to 2: what is the percentage error?
- (A)2.5
- (B)3.5
- (C)1.2
- (D)0
Answer: (D) Official key: D
- Simpson's rule fits a parabola through each pair of strips. The integrand here is itself a quadratic, so every parabola matches it exactly and the rule has no error at all. In fact Simpson's error term contains , which is zero for any polynomial up to degree 3.
- So and : option (D). No arithmetic needed.
Check: , and Simpson's rule with in exact fractions gives the same . The trapezoidal rule with the same 10 strips would give an error of , still nowhere near the non-zero options: they are there for anyone who assumes a numerical method must have some error.
Finite element codes use the same idea with Gauss quadrature: an n-point Gauss rule integrates polynomials up to degree 2n − 1 exactly, so element stiffness matrices are computed exactly with very few points. Deliberately using fewer points (reduced integration) makes elements cheaper and avoids locking, but can let zero-energy 'hourglass' modes appear, which is why Abaqus and LS-DYNA offer hourglass control for reduced-integration elements.
Simpson's one-third rule on five tabulated values from x = 0 to 1: what is the integral?
| 0 | 0.25 | 0.5 | 0.75 | 1.0 | |
| 0.9 | 2.0 | 1.5 | 1.8 | 0.4 |
Answer: 1.62 Official key: 1.61 to 1.65
- Five points give four strips of width , an even number, so Simpson's one-third rule applies directly.
- Weights : odd-numbered interior points get 4, even-numbered interior points get 2.
- .
Traps: swapping the weights (2 on and 4 on ) gives , and the trapezoidal rule gives . Both are outside the key range of 1.61 to 1.65.
This is how measured data gets integrated: the work per cycle from an engine's pressure-volume indicator data, the drag force from a row of pressure taps, or the energy under a load-displacement curve from a test rig. Simpson's one-third rule needs an even number of strips; with an odd number, a common fix is Simpson's three-eighths rule on the last three strips.
Trapezoidal (Crank-Nicolson) scheme for dy/dt = −y: above which time step does it oscillate?
- (A)
- (B)
- (C)
- (D)
Answer: (D) Official key: D
- Solve the scheme for : , so each step multiplies by
- The exact solution never changes sign. The numerical one does whenever : then alternates . when , that is . So : option (D).
- Note for every : the scheme never blows up (it is unconditionally stable), it just stops looking physical. At , .
Trap: the explicit (forward) Euler scheme starts oscillating at and blows up at . Mixing up the two schemes gives (B). The fully implicit (backward) Euler scheme, , never oscillates for this equation.
This scheme is Crank-Nicolson, the trapezoidal rule in time, and it is second-order accurate. CFD codes use it for transient flow, and the oscillation you see here is a known problem on stiff or sharply changing solutions. That is why OpenFOAM's CrankNicolson time scheme takes an off-centring coefficient between 0 (backward Euler) and 1 (pure Crank-Nicolson), and many tutorials use 0.9: a little damping for a small loss of accuracy.