Home / Courses / IIT JAM Physics / Practice / Numerical methods: roots, integration and ODEs

Numerical methods: roots, integration and ODEs

Mathematical methods · 13 questions and templates for IIT JAM Physics, in order from warm-up to top rank. Check each answer, then read the full solution.

Loading questions… (they need JavaScript; nothing you type is sent anywhere)

Written questions with solutions

The same questions as a list, so you can print them or read without the practice box. Templates (fresh numbers) appear only in the practice box above.

  1. MCQWarm-up

    Which root-finding method needs the derivative f′(x) at every step?
    1. Bisection
    2. Secant
    3. Newton-Raphson
    4. Regula falsi (false position)
    Answer and solution

    Answer: C

    Newton-Raphson uses xn+1=xn−f(xn)/f′(xn). The secant method replaces f′ by the slope through the last two points; bisection and false position use only values of f at the ends of a bracket.

    Why people get this wrong: The secant method looks like Newton's but needs no derivative.

  2. NATStandard

    The secant method is applied to f(x)=x2−3 with x0=1 and x1=2. The next iterate x2 is ______ (three decimal places).
    Answer and solution

    Answer: 1.667 (±0.002)

    f(1)=−2, f(2)=1. x2=x1−f(x1)x1−x0f(x1)−f(x0)=2−1⋅11−(−2)=2−13≈1.667. (The root is 3≈1.732.)

    Why people get this wrong: Using f(x0)−f(x1) in the denominator flips the correction and gives 2.333.

  3. MCQStandard

    The forward difference f(x+h)−f(x)h approximates f′(x) with truncation error of order
    1. h
    2. h2
    3. h3
    4. 1 (it does not improve as h→0)
    Answer and solution

    Answer: A

    Taylor: f(x+h)=f(x)+hf′(x)+h22f′′(x)+…, so the forward difference is f′(x)+h2f′′(x)+…: error O(h). The central difference cancels the h term and is O(h2).

    Why people get this wrong: Mixing up the forward difference with the central one, which is second order.

  4. MCQExam level

    For a smooth integrand on small panels, the error of the composite midpoint rule compared with that of the composite trapezoidal rule (same panels) is about
    1. the same size and sign
    2. half the size and of opposite sign
    3. twice the size and of the same sign
    4. zero for every cubic
    Answer and solution

    Answer: B

    On one panel of width h: trapezoidal error =−h312f′′, midpoint error =+h324f′′. So the midpoint rule is about twice as accurate and errs the other way; this is why Simpson's rule, S=23M+13T, cancels both leading errors.

    Why people get this wrong: Both rules are exact only for straight lines; neither integrates every cubic exactly.

  5. NATExam level

    The composite midpoint rule with two panels (h=1) estimates ∫02x3dx as ______ (one decimal place).
    Answer and solution

    Answer: 3.5 (±0.01)

    Midpoints 0.5 and 1.5: h[f(0.5)+f(1.5)]=0.125+3.375=3.5. The exact value is 4; the trapezoidal rule gives 12[0+2⋅1+8]=5. Errors −0.5 and +1: half the size and opposite sign.

    Why people get this wrong: Using the end points of the panels instead of their midpoints.

  6. MSQExam level

    Which statements about solving Ax→=b→ (A dense, n×n) are true?
    1. Partial pivoting (choosing the largest available pivot) improves numerical stability
    2. Gaussian elimination takes of order n3 operations
    3. Once A=LU is known, each new right-hand side costs only of order n2 operations
    4. Computing A−1 and then A−1b→ is the most efficient way to solve one system
    Answer and solution

    Answer: A, B, C

    Small pivots amplify rounding errors; swapping rows to use the largest one keeps multipliers at most 1. Elimination costs about n3/3 multiplications. With L and U stored, forward and back substitution cost about n2 each. Forming A−1 costs about three times more than one elimination and is usually less accurate.

    Why people get this wrong: Writing x→=A−1b→ is fine on paper but wasteful in code.

  7. MSQExam level

    Which statements about polynomial interpolation are true?
    1. High-degree interpolation of 1/(1+25x2) at equally spaced points on [−1,1] oscillates wildly near the ends
    2. Clustering the nodes near the ends (Chebyshev points) tames that oscillation
    3. A cubic spline has continuous first and second derivatives at the nodes
    4. Raising the degree of an equally spaced interpolant always improves its accuracy
    Answer and solution

    Answer: A, B, C

    This is Runge's phenomenon: the error near ±1 grows with the degree for equally spaced nodes. Chebyshev nodes, cos⁡(2k+1)π2n+2, minimise the node polynomial and make the interpolant converge for smooth functions. Cubic splines join cubic pieces with matching value, slope and curvature. The last statement is exactly what Runge's example disproves.

    Why people get this wrong: "More points, more accuracy" fails for equally spaced polynomial interpolation.

  8. NATHard

    One step of the implicit (backward) Euler method yn+1=yn+hf(yn+1) with h=0.1 is applied to dy/dt=−50y, y(0)=1. The result y1 is ______ (three decimal places).
    Answer and solution

    Answer: 0.167 (±0.002)

    y1=y0−5y1, so y1=y01+50h=16≈0.167. It decays, as the true solution does (e−5≈0.007). Explicit Euler with the same step gives y1=(1−5)y0=−4 and blows up: backward Euler is stable for every h>0 on decaying problems.

    Why people get this wrong: Evaluating f at the old point (explicit Euler) gives −4.

  9. MCQHard

    The iteration xn+1=xn2+23 has fixed points x=1 and x=2. Started from x0=1.5, it
    1. converges to 2
    2. oscillates between 1 and 2
    3. runs off to infinity
    4. converges to 1
    Answer and solution

    Answer: D

    With g(x)=(x2+2)/3, g′(x)=2x/3: |g′(1)|=23<1 (attracting) and |g′(2)|=43>1 (repelling). From 1.5: x1=1.417, x2=1.336, ... decreasing towards 1. (Starting above 2 it would run off to infinity.)

    Why people get this wrong: Being a fixed point is not enough; the iteration converges only to points where |g′|<1.

  10. NATHard

    The heat equation ut=κuxx with κ=1.0×10−4 m2/s is solved by the explicit scheme (forward in time, central in space) with grid spacing Δx=0.01 m. It is stable for κΔt/Δx2≤12. The largest stable time step is ______ s (one decimal place).
    Answer and solution

    Answer: 0.5 s (±0.01)

    Δtmax⁡=Δx22κ=10−42×10−4=0.5 s. Halving Δx would cut the allowed step by four, which is why implicit schemes (e.g. Crank-Nicolson) are used for fine grids.

    Why people get this wrong: Using Δx instead of Δx2 gives 50 s, a hundred times too large.

← All topics · Timed mock test →