Skip to content

Numerical Analysis Formula Sheet and Short Notes

One-page CSIR NET numerical analysis revision: convergence orders, Newton-Raphson, Gauss-Seidel, interpolation, quadrature errors, Euler and Runge-Kutta.

  • 8 min read
  • Updated
On this page · 7 sections
  1. The paper in one table
  2. Roots of equations
  3. Linear systems
  4. Finite differences and interpolation
  5. Differentiation and integration
  6. Initial value problems
  7. Part C checklist

Every method, formula and error term from Iteration, Newton-Raphson and Rate of Convergence, Gauss Elimination and Gauss-Seidel, Finite Differences and Interpolation, Numerical Differentiation and Integration and Picard, Euler, Modified Euler and Runge-Kutta, on one page.

The paper in one table

Per the NTA information bulletin for December 2026 (Mathematical Sciences, code 704):

PartGivenAttemptCorrectWrongMaximum
B4025+3−0.7575
C6020+4.75095

Part C: one or more options may be correct; marks only if you mark all correct options and no wrong one. No partial credit, no negative marking. Details on the Mathematical Sciences overview.

Roots of equations

$$\boxed{\frac{\lvert e_{n+1} \rvert}{\lvert e_n \rvert^p} \to C} \qquad p \approx \frac{\log\lvert e_{n+1}/e_n \rvert}{\log\lvert e_n/e_{n-1} \rvert}$$
MethodIterationOrder
Bisectionhalve the bracket; \( \lvert c_n - \alpha \rvert \le \frac{b - a}{2^n} \)linear (bound halves)
Fixed point\( x_{n+1} = g(x_n) \)1 with \( C = \lvert g'(\alpha) \rvert \); \( p \) if \( g' = \dots = g^{(p-1)} = 0 \) at \( \alpha \)
Regula falsisecant line, keep the sign changelinear
Secant\( x_n - f_n\frac{x_n - x_{n-1}}{f_n - f_{n-1}} \)\( \frac{1 + \sqrt 5}{2} \approx 1.618 \)
Newton-Raphson\( x_n - \frac{f(x_n)}{f'(x_n)} \)2, \( C = \lvert \frac{f''(\alpha)}{2f'(\alpha)} \rvert \)
Newton, multiplicity \( m \)same1, \( C = 1 - \frac{1}{m} \)
Modified Newton\( x_n - m\frac{f(x_n)}{f'(x_n)} \)2
  • Contraction theorem: \( g([a, b]) \subset [a, b] \), \( \lvert g' \rvert \le L < 1 \) \( \Rightarrow \) unique fixed point, convergence from every \( x_0 \), \( \lvert x_n - \alpha \rvert \le \frac{L^n}{1 - L}\lvert x_1 - x_0 \rvert \).
  • Sign of \( g'(\alpha) \): positive, monotone approach; negative, alternating.
  • Newton recipes: \( \sqrt N \): \( \frac{1}{2}(x + \frac{N}{x}) \); \( \frac{1}{N} \): \( x(2 - Nx) \); \( N^{1/k} \): \( \frac{1}{k}((k - 1)x + \frac{N}{x^{k-1}}) \).
  • Fourier: start where \( f(x_0)f''(x_0) > 0 \) (with \( f', f'' \) of fixed sign) for monotone convergence.

Linear systems

ItemFact
Gauss elimination\( m_{ik} = \frac{a_{ik}}{a_{kk}} \); \( \sim \frac{n^3}{3} \) multiplications; back substitution \( \sim \frac{n^2}{2} \); Gauss-Jordan \( \sim \frac{n^3}{2} \)
Determinantproduct of pivots \( \times (-1)^{\text{row swaps}} \)
Partial pivotinglargest \( \lvert a_{ik} \rvert \) in the column; all \( \lvert m_{ik} \rvert \le 1 \)
\( A = LU \)exists, unique (unit \( L \)) when all leading principal minors \( \ne 0 \); in general \( PA = LU \)
Cholesky\( A = LL^T \) for symmetric positive definite \( A \)
Condition number\( \kappa = \lVert A \rVert\lVert A^{-1} \rVert \); \( \frac{\lVert \delta x \rVert}{\lVert x \rVert} \le \kappa\frac{\lVert \delta b \rVert}{\lVert b \rVert} \)
$$\boxed{T_J = -D^{-1}(L + U)} \qquad \boxed{T_{GS} = -(D + L)^{-1}U} \qquad \text{converge for all } x^{(0)} \iff \rho(T) < 1$$
Condition on \( A \)JacobiGauss-Seidel
Strictly diagonally dominantconvergesconverges
Symmetric positive definitenot guaranteedconverges
Tridiagonal / 2 by 2\( \rho(T_{GS}) = \rho(T_J)^2 \)

2 by 2 \( \begin{pmatrix} a & b \\ c & d \end{pmatrix} \): \( \rho(T_J) = \sqrt{\lvert \frac{bc}{ad} \rvert} \), \( \rho(T_{GS}) = \lvert \frac{bc}{ad} \rvert \). Counter-examples: \( \begin{pmatrix} 1 & 2 & -2 \\ 1 & 1 & 1 \\ 2 & 2 & 1 \end{pmatrix} \) (Jacobi yes, GS no); \( \begin{pmatrix} 2 & -1 & 1 \\ 2 & 2 & 2 \\ -1 & -1 & 2 \end{pmatrix} \) (Jacobi no, GS yes).

Finite differences and interpolation

$$\Delta = E - 1, \quad \nabla = 1 - E^{-1}, \quad \delta = E^{1/2} - E^{-1/2}, \quad E = e^{hD}, \quad \Delta - \nabla = \Delta\nabla = \delta^2, \quad \mu^2 = 1 + \tfrac{\delta^2}{4}$$
  • Degree-\( n \) polynomial: \( \Delta^n f = a_n\,n!\,h^n \), \( \Delta^{n+1} f = 0 \).
  • Error \( \varepsilon \) in one entry: \( m \)-th differences carry \( \varepsilon \times \) binomial coefficients with alternating signs.
  • Newton forward \( p(x_0 + sh) = \sum \binom{s}{k}\Delta^k f_0 \); backward \( p(x_n + sh) = \sum \frac{s(s+1)\cdots(s+k-1)}{k!}\nabla^k f_n \).
  • Divided differences: symmetric, \( f[x_0, \dots, x_k] = \frac{f^{(k)}(\xi)}{k!} = \frac{\Delta^k f_0}{k!\,h^k} \), \( f[x_0, x_0] = f'(x_0) \).
  • Lagrange: \( L_i = \prod_{j \ne i}\frac{x - x_j}{x_i - x_j} \), \( \sum L_i = 1 \), \( \sum x_i^k L_i = x^k \) for \( k \le n \).
$$\boxed{f(x) - p(x) = \frac{f^{(n+1)}(\xi)}{(n+1)!}\prod_{i=0}^n (x - x_i)} \qquad \text{linear: } \le \frac{h^2}{8}\max\lvert f'' \rvert$$
InterpolantDataDegree / smoothnessError
Lagrange / Newton\( n + 1 \) valuesdegree \( \le n \)\( \frac{f^{(n+1)}}{(n+1)!}\prod(x - x_i) \)
Hermite\( n + 1 \) values and slopesdegree \( \le 2n + 1 \)\( \frac{f^{(2n+2)}}{(2n+2)!}\prod(x - x_i)^2 \)
Cubic spline\( n + 1 \) values + 2 end conditionspiecewise cubic, \( C^2 \)\( O(h^4) \) (clamped, smooth \( f \))
  • Two-point cubic Hermite on \( [0, 1] \): \( h_{00} = 2t^3 - 3t^2 + 1 \), \( h_{10} = t^3 - 2t^2 + t \), \( h_{01} = -2t^3 + 3t^2 \), \( h_{11} = t^3 - t^2 \).
  • Spline count: \( 4n \) unknowns \( = 2n \) (values) \( + 2(n - 1) \) (\( C^1, C^2 \)) \( + 2 \) (ends). Natural: \( S'' = 0 \) at both ends; clamped: \( S' \) given.
  • Moments (equal \( h \)): \( M_{i-1} + 4M_i + M_{i+1} = \frac{6}{h^2}(f_{i-1} - 2f_i + f_{i+1}) \).
  • Runge: \( \frac{1}{1 + 25x^2} \), equally spaced nodes, error grows with degree; Chebyshev nodes minimise \( \max\lvert \omega \rvert = 2^{-n} \).

Differentiation and integration

FormulaError
\( \frac{f(x+h) - f(x)}{h} \)\( -\frac{h}{2}f'' \)
\( \frac{f(x+h) - f(x-h)}{2h} \)\( -\frac{h^2}{6}f''' \)
\( \frac{f(x+h) - 2f(x) + f(x-h)}{h^2} \)\( -\frac{h^2}{12}f^{(4)} \)
\( hf'_0 = \Delta - \frac{\Delta^2}{2} + \frac{\Delta^3}{3} - \cdots \)from \( hD = \log(1 + \Delta) \)
Richardson (even powers)\( \frac{4A(h/2) - A(h)}{3} \)

Best step with rounding \( \varepsilon \): forward \( \sim\sqrt\varepsilon \), central \( \sim\varepsilon^{1/3} \).

RuleFormulaError \( \int - Q \)Precision
Midpoint (width \( h \))\( hf(m) \)\( +\frac{h^3}{24}f'' \)1
Trapezoid\( \frac{h}{2}(f_0 + f_1) \)\( -\frac{h^3}{12}f'' \)1
Simpson 1/3\( \frac{h}{3}(f_0 + 4f_1 + f_2) \)\( -\frac{h^5}{90}f^{(4)} \)3
Simpson 3/8\( \frac{3h}{8}(f_0 + 3f_1 + 3f_2 + f_3) \)\( -\frac{3h^5}{80}f^{(4)} \)3
Boole\( \frac{2h}{45}(7, 32, 12, 32, 7) \)\( -\frac{8h^7}{945}f^{(6)} \)5
Weddle\( \frac{3h}{10}(1, 5, 1, 6, 1, 5, 1) \)5
Gauss-Legendre, \( n \) pointszeros of \( P_n \)\( 2n - 1 \)
  • Composite trapezoid: \( -\frac{(b - a)h^2}{12}f'' \); composite Simpson (\( n \) even): \( -\frac{(b - a)h^4}{180}f^{(4)} \).
  • Gauss: 2-point \( \pm\frac{1}{\sqrt 3} \), weights 1; 3-point \( 0, \pm\sqrt{\frac{3}{5}} \), weights \( \frac{8}{9}, \frac{5}{9}, \frac{5}{9} \); on \( [a, b] \) multiply by \( \frac{b - a}{2} \).
  • Convex \( f \): trapezoid over, midpoint under; Simpson \( = \frac{2M + T}{3} \); Romberg step 1 \( = \) Simpson.

Initial value problems

MethodFormulaOrderReal stability
Picard\( y_{n+1}(x) = y_0 + \int_{x_0}^x f(t, y_n(t))\,dt \)error \( \le \frac{ML^n\lvert x - x_0 \rvert^{n+1}}{(n+1)!} \)
Euler\( y_n + hf(x_n, y_n) \)1\( -2 < h\lambda < 0 \)
Heun (modified Euler)\( y_n + \frac{h}{2}[f_n + f(x_{n+1}, y_n + hf_n)] \)2\( -2 < h\lambda < 0 \)
Midpoint\( y_n + hf(x_n + \frac{h}{2}, y_n + \frac{h}{2}f_n) \)2\( -2 < h\lambda < 0 \)
RK4\( y_n + \frac{1}{6}(k_1 + 2k_2 + 2k_3 + k_4) \)4about \( -2.785 < h\lambda < 0 \)
Backward Euler\( y_n + hf(x_{n+1}, y_{n+1}) \)1all \( h\lambda < 0 \)
  • RK2 family: \( a + b = 1 \), \( b\alpha = b\beta = \frac{1}{2} \) (Heun \( b = \frac{1}{2} \); midpoint \( b = 1 \); Ralston \( b = \frac{3}{4} \)).
  • RK4 stages: \( k_1 = hf(x_n, y_n) \), \( k_2 = hf(x_n + \frac{h}{2}, y_n + \frac{k_1}{2}) \), \( k_3 = hf(x_n + \frac{h}{2}, y_n + \frac{k_2}{2}) \), \( k_4 = hf(x_n + h, y_n + k_3) \).
  • Order \( p \): local \( O(h^{p+1}) \), global \( O(h^p) \). Methods as quadrature: Euler = rectangle, Heun = trapezoid, RK4 = Simpson when \( f = f(x) \).
  • “Modified Euler” means Heun in many books and the midpoint method in others: use the formula given.

Benchmark \( y' = x + y \), \( y(0) = 1 \), \( h = 0.1 \), exact \( 2e^x - x - 1 \):

\( y(0.1) \)\( y(0.2) \)
Euler1.11.22
Heun1.111.24205
RK41.11034171.2428051
Exact1.11034181.2428055

Part C checklist

Back to the Numerical Analysis chapter overview. For the analysis behind the error terms, see Real Analysis.

Advertisement
Advertisement