Computer Oriented Numerical Methods: formula sheet Typed from a student's handwritten revision notes (12 pages) so it can be searched and read on a phone. Where the handwritten version had a slip or left something out, the corrected form is given and marked Note. Each section links to a solved exam question that uses the formula.
1. Errors
Actual (true) value x, approximate value x′.
| Error | Formula |
|---|---|
| Absolute error | Eₐ = |x − x′| |
| Relative error | Eᵣ = |x − x′| / |x| |
| Percentage error | Eₚ = Eᵣ × 100 % |
Note: the absolute error is usually written with modulus bars; the sign of x − x′ only says whether x′ is an under- or over-estimate.
2. Roots of algebraic and transcendental equations
2.1 Descartes' rule of signs
For a polynomial f(x) = 0 of degree n:
- p = number of sign changes in f(x) → the number of positive real roots is p or p − 2, p − 4, …
- q = number of sign changes in f(−x) → the number of negative real roots is q or q − 2, …
- Number of complex roots is at least n − (p + q) (complex roots come in conjugate pairs).
2.2 Budan's theorem
Let V(c) = number of sign changes in the sequence f(c), f′(c), f″(c), …, f⁽ⁿ⁾(c). The number of real roots of f(x) = 0 in (a, b), counted with multiplicity, is
V(a) − V(b) − 2m, m = 0, 1, 2, …
| V(a) − V(b) | Real roots in (a, b) |
|---|---|
| 0 | none |
| 1 | exactly one |
| 2 | two or none |
| 3 | three or one |
(Even values give an even count, odd values an odd count.) Worked example: CONMP CET-1 2026, Q2(d).
2.3 Bracketing methods
Start with [a, b] where f(a) and f(b) have opposite signs.
| Method | Next approximation |
|---|---|
| Bisection | xₙ₊₁ = (a + b) / 2 |
| Regula falsi (false position) | xₙ₊₁ = [a f(b) − b f(a)] / [f(b) − f(a)] |
Which end to replace: keep the sub-interval where the sign changes. If f(xₙ₊₁) has the same sign as f(a), replace a by xₙ₊₁; otherwise replace b. (The handwritten shortcut "answer > 0 → replace the right end, answer < 0 → replace the left end" works when f(a) < 0 < f(b).)
2.4 Open-ended methods
| Method | Formula | Order of convergence |
|---|---|---|
| Newton-Raphson | xₙ₊₁ = xₙ − f(xₙ) / f′(xₙ), n = 0, 1, 2, … | 2 (quadratic) |
| Secant (method of chords) | xₙ₊₁ = xₙ − [(xₙ − xₙ₋₁) / (f(xₙ) − f(xₙ₋₁))] · f(xₙ), n ≥ 1 | ≈ 1.618 |
| Fixed-point iteration (successive approximations) | write f(x) = 0 as x = φ(x); xₙ₊₁ = φ(xₙ) | 1 (linear) |
Fixed-point iteration converges if |φ′(x)| < 1 for all x in the interval I containing the root, with x₀ ∈ I.
Worked secant example: CONMP CET-1 2025, Q3(a).
2.5 Newton-Raphson deductions (iterative formulas)
| To find | Iteration |
|---|---|
| 1/N (reciprocal, no division) | xₙ₊₁ = xₙ (2 − N xₙ) |
| √N | xₙ₊₁ = ½ (xₙ + N / xₙ) |
| 1/√N | xₙ₊₁ = ½ (xₙ + 1 / (N xₙ)) |
| ᵏ√N | xₙ₊₁ = (1/k) [(k − 1) xₙ + N / xₙᵏ⁻¹] |
Note: the 1/√N formula in the notes is the √ formula applied to 1/N and converges correctly. The form obtained directly from Newton's method on f(x) = 1/x² − N is xₙ₊₁ = ½ xₙ (3 − N xₙ²), which avoids division; both are accepted.
2.6 Lin-Bairstow method
Finds a quadratic factor x² + px + q of a polynomial a₀xⁿ + a₁xⁿ⁻¹ + … + aₙ, starting from guesses p₀, q₀.
- Synthetic division: bₖ = aₖ − p bₖ₋₁ − q bₖ₋₂ (b₋₁ = b₋₂ = 0).
- Second division: cₖ = bₖ − p cₖ₋₁ − q cₖ₋₂.
- Solve for the corrections: cₙ₋₂ Δp + cₙ₋₃ Δq = bₙ₋₁ and (cₙ₋₁ − bₙ₋₁) Δp + cₙ₋₂ Δq = bₙ.
- p ← p + Δp, q ← q + Δq; repeat until Δp, Δq are small.
Note: the handwritten page only states the goal; the steps above are the version used in the solved papers. Worked example: CONMP CET-2 2025, Q3(a).
3. Systems of linear equations
For a₁x + b₁y + c₁z = d₁, a₂x + b₂y + c₂z = d₂, a₃x + b₃y + c₃z = d₃ (rearranged so the system is diagonally dominant: |a₁| ≥ |b₁| + |c₁| and so on, which guarantees convergence).
Gauss-Jacobi (every new value uses only the previous iteration):
- xₖ₊₁ = (1/a₁) [d₁ − b₁yₖ − c₁zₖ]
- yₖ₊₁ = (1/b₂) [d₂ − a₂xₖ − c₂zₖ]
- zₖ₊₁ = (1/c₃) [d₃ − a₃xₖ − b₃yₖ]
Gauss-Seidel (uses each new value as soon as it is found):
- x₁ = (1/a₁) [d₁ − b₁y₀ − c₁z₀]
- y₁ = (1/b₂) [d₂ − a₂x₁ − c₂z₀]
- z₁ = (1/c₃) [d₃ − a₃x₁ − b₃y₁]
Gauss-Seidel typically converges about twice as fast as Gauss-Jacobi. Worked example: CONMP CET-1 2026, Q3(e).
4. Interpolation
4.1 Equally spaced arguments
Spacing h, p = (x − x₀) / h.
Newton's forward difference formula (use when x is near the beginning of the table):
y = y₀ + p Δy₀ + p(p − 1)/2! Δ²y₀ + p(p − 1)(p − 2)/3! Δ³y₀ + …
Newton's backward difference formula (x near the end; here p = (x − xₙ) / h and differences are taken at yₙ):
y = yₙ + p ∇yₙ + p(p + 1)/2! ∇²yₙ + p(p + 1)(p + 2)/3! ∇³yₙ + …
Note: the notes write y₀ in the backward formula; it is the last tabulated value yₙ.
4.2 Stirling's central difference formula
Use when x is near the middle of the table; choose x₀ as the tabulated value just below x, so −½ < p < ½ is ideal.
y = y₀ + p (Δy₀ + Δy₋₁)/2 + p²/2! Δ²y₋₁ + p(p² − 1²)/3! · (Δ³y₋₁ + Δ³y₋₂)/2 + p²(p² − 1²)/4! Δ⁴y₋₂ + p(p² − 1²)(p² − 2²)/5! · (Δ⁵y₋₂ + Δ⁵y₋₃)/2 + …
The differences used lie on the horizontal line through y₀ in the difference table: odd orders are averaged across the line (Δy₋₁ and Δy₀), even orders sit on it (Δ²y₋₁, Δ⁴y₋₂). Worked example: CONMP CET-2 2026, Q3(a).
4.3 Unequally spaced arguments
Lagrange's interpolation formula (four points shown):
y = y₀ (x − x₁)(x − x₂)(x − x₃) / [(x₀ − x₁)(x₀ − x₂)(x₀ − x₃)]
- y₁ (x − x₀)(x − x₂)(x − x₃) / [(x₁ − x₀)(x₁ − x₂)(x₁ − x₃)]
- y₂ (x − x₀)(x − x₁)(x − x₃) / [(x₂ − x₀)(x₂ − x₁)(x₂ − x₃)]
- y₃ (x − x₀)(x − x₁)(x − x₂) / [(x₃ − x₀)(x₃ − x₁)(x₃ − x₂)]
Newton's divided difference formula:
y = y₀ + (x − x₀)[x₀, x₁] + (x − x₀)(x − x₁)[x₀, x₁, x₂] + (x − x₀)(x − x₁)(x − x₂)[x₀, x₁, x₂, x₃] + …
where [x₀, x₁] = (y₁ − y₀)/(x₁ − x₀), [x₀, x₁, x₂] = ([x₁, x₂] − [x₀, x₁])/(x₂ − x₀), and so on.
Note: the notes write Δy₀, Δ²y₀ here; for unequal spacing they must be the divided differences above. Worked examples: Lagrange, end-sem 2025 Q2.1 · divided differences, end-sem 2025 Q2.2.
5. Numerical integration
I = ∫ₐᵇ y dx with h = (b − a)/n and ordinates y₀, y₁, …, yₙ.
| Rule | Condition on n | Formula |
|---|---|---|
| Trapezoidal | any n | (h/2) [(y₀ + yₙ) + 2(y₁ + y₂ + … + yₙ₋₁)] |
| Simpson's 1/3 (parabolic rule) | n even (n = 2m) | (h/3) [(y₀ + yₙ) + 4(y₁ + y₃ + … + yₙ₋₁) + 2(y₂ + y₄ + … + yₙ₋₂)] |
| Simpson's 3/8 | n a multiple of 3 (n = 3m) | (3h/8) [(y₀ + yₙ) + 3(y₁ + y₂ + y₄ + y₅ + …) + 2(y₃ + y₆ + …)] |
In Simpson's 3/8 rule the ordinates whose index is a multiple of 3 get weight 2, all other interior ordinates get weight 3. Worked example: CONMP CET-2 2025, Q2(b).
6. Numerical solution of ordinary differential equations
Initial value problem dy/dx = f(x, y), y(x₀) = y₀, step size h, xₙ₊₁ = xₙ + h.
Euler's method: yₙ₊₁ = yₙ + h f(xₙ, yₙ), n = 0, 1, 2, …
Modified Euler / improved Euler / Heun's / predictor-corrector method:
- Predictor: yᵖₙ₊₁ = yₙ + h f(xₙ, yₙ)
- Corrector: yₙ₊₁ = yₙ + (h/2) [f(xₙ, yₙ) + f(xₙ₊₁, yᵖₙ₊₁)] (repeat the corrector until two values agree, if asked)
Runge-Kutta method of order 2:
- Identify f(x, y), x₀, y₀ and h.
- k₁ = h f(x₀, y₀)
- k₂ = h f(x₀ + h, y₀ + k₁)
- k = ½ (k₁ + k₂)
- y₁ = y₀ + k, x₁ = x₀ + h
Runge-Kutta method of order 4:
- Identify f(x, y), x₀, y₀ and h.
- k₁ = h f(x₀, y₀)
- k₂ = h f(x₀ + h/2, y₀ + k₁/2)
- k₃ = h f(x₀ + h/2, y₀ + k₂/2)
- k₄ = h f(x₀ + h, y₀ + k₃)
- k = (1/6) [k₁ + 2k₂ + 2k₃ + k₄]
- y₁ = y₀ + k, x₁ = x₀ + h
7. Curve fitting (least squares)
Straight line y = ax + b (n data points):
- Σy = a Σx + n b
- Σxy = a Σx² + b Σx
Second-degree parabola y = ax² + bx + c:
- Σy = a Σx² + b Σx + n c
- Σxy = a Σx³ + b Σx² + c Σx
- Σx²y = a Σx⁴ + b Σx³ + c Σx²
Build a table of x, y, x², x³, x⁴, xy, x²y, add the columns and solve the normal equations. Worked examples: straight line, end-sem 2025 Q4.4 · parabola, end-sem 2025 Q5.6.