Numerical Analysis and Inversion
CSIR NET Earth Sciences Part C geophysics: quadrature, FDM and FEM, ODE solvers, pattern recognition, ill-posed problems, least squares, SVD and optimisation.
- 4
- Topics
- 77
- Min read
- 01 Numerical Differentiation, Integration, FDM and FEMStart here CSIR NET geophysics: finite-difference derivatives, trapezoid, Simpson and Gauss quadrature, FDM stability (CFL) and 1-D finite elements, with practice.
- 02 Initial Value Problems and Pattern Recognition CSIR NET geophysics: Euler, Heun and Runge-Kutta for IVPs, stability and stiffness, plus pattern recognition: features, k-NN, k-means and confusion matrices.
- 03 Well- and Ill-Posed Problems and Least Squares CSIR NET geophysics inversion: forward vs inverse problems, Hadamard well-posedness, non-uniqueness, condition number, least squares, weighting and damping.
- 04 Gradient and Direct Search, Generalized Inversion, SVD and Global Optimization CSIR NET geophysics: steepest descent, Gauss-Newton, Marquardt, simplex and grid search, SVD and generalized inverse, resolution, simulated annealing, GA.
- Numerical Analysis and Inversion Short Notes Revise The whole chapter on one page.
Chapter Overview
Numerical Analysis and Inversion is the third unit of the Geophysics section of Part C in CSIR NET Earth, Atmospheric, Ocean and Planetary Sciences (subject code 702). It is the computational backbone of the whole section. Gravity, magnetic, electrical, seismic and well-log interpretation all come down to two tasks: forward modelling (compute the data a given Earth model would produce) and inversion (recover a model from observed data). This unit supplies the tools for both.
The first half is numerical analysis as geophysicists use it: difference formulas and quadrature for sampled data, finite-difference and finite-element solution of the heat, wave and potential equations, time-stepping methods for initial value problems, and pattern recognition for sorting large volumes of data. The second half is inverse theory: why inverse problems are often ill-posed, how least squares and its weighted, damped and minimum-norm variants produce answers, and how gradient methods, direct search, the singular value decomposition and global optimisation find and judge those answers.
The notes assume an MSc-level course in numerical methods and in geophysical data analysis (texts such as Menke’s Geophysical Data Analysis: Discrete Inverse Theory and Press et al.’s Numerical Recipes cover the ground). Every numerical answer on these pages has been worked out step by step and is shown in full.
What You’ll Learn
By the end of this chapter you will be able to:
- Differentiate and integrate sampled data with stated error orders, and choose between trapezoidal, Simpson and Gauss quadrature
- Set up explicit and implicit finite-difference schemes, apply the \( r \le \frac{1}{2} \) and CFL stability limits, and run a 1-D finite-element calculation
- Take Euler, Heun and Runge–Kutta steps, judge order and stability, and recognise stiff problems
- Explain supervised and unsupervised classification, run k-means, compare distance measures and read a confusion matrix
- Test Hadamard’s conditions, solve least-squares, weighted, minimum-norm and damped problems, and judge fit with \( \chi^2 \)
- Use the SVD, the generalized inverse and resolution matrices, and choose between gradient, direct-search and global methods
Prerequisites
- Matrix algebra and eigenvalues: Linear Algebra (Mathematical Sciences)
- A first numerical-methods course: Numerical Analysis (Mathematical Sciences) covers root finding, interpolation and quadrature from the mathematician’s side
- Signals, convolution and deconvolution: Signal Processing, the first Geophysics unit
- The Part B background on seismic waves and Earth structure: Interior of the Earth, Deformation and Tectonics
Learning Path
The topics follow the order of the official syllabus lines:
graph TD
A["1. Numerical Differentiation,
Integration, FDM and FEM"] --> B["2. Initial Value Problems
and Pattern Recognition"]
A --> C["3. Well- and Ill-Posed Problems
and Least Squares"]
B --> C
C --> D["4. Gradient and Direct Search,
Generalized Inversion, SVD,
Global Optimization"]
D --> E["Quick Revision"]Pages 1 and 2 are forward modelling: computing what the Earth would do. Pages 3 and 4 are inversion: working back from data to the Earth. Page 2’s stability ideas reappear in page 1’s PDE schemes, and page 3’s damped least squares is the same algebra as page 4’s Marquardt method.
Topics in This Chapter
| Topic | What it covers | Time |
|---|---|---|
| Numerical Differentiation, Integration, FDM and FEM | difference formulas and errors, Richardson extrapolation, round-off, trapezoidal and Simpson rules, Gauss–Legendre quadrature, FTCS, implicit and Crank–Nicolson schemes, von Neumann stability, CFL condition, Laplace stencil, Lax theorem, weak form, Galerkin method, stiffness matrix, FDM versus FEM | 35 min |
| Initial Value Problems and Pattern Recognition | IVPs and systems, shooting, Euler, Heun, RK4, error scaling, adaptive steps, multistep methods, stability regions, stiffness; features, supervised and unsupervised learning, distances, k-NN, Bayes, neural networks, k-means, confusion matrix, geophysical applications | 35 min |
| Well- and Ill-Posed Problems and Least Squares | forward versus inverse, Hadamard’s conditions, non-uniqueness, downward continuation, over-, under- and mixed-determined problems, condition number, normal equations, line fitting, covariance, weighted LS, L1 versus L2, \( \chi^2 \), minimum norm, damped LS, linearisation | 35 min |
| Gradient and Direct Search, Generalized Inversion, SVD and Global Optimization | gradient and Hessian, steepest descent, Newton, Gauss–Newton, Levenberg–Marquardt, conjugate gradients, grid and simplex search, SVD, null spaces, generalized inverse, resolution and covariance, truncation, Monte Carlo, simulated annealing, genetic algorithms | 40 min |
| Quick Revision | every formula and trap on one page | 10 min |
Official Syllabus Lines and Where They Are Covered
| Official topic line | Covered in |
|---|---|
| Numerical differentiation/integration, FEM, FDM, quadrature | Numerical Differentiation, Integration, FDM and FEM |
| Initial value problems, pattern recognition | Initial Value Problems and Pattern Recognition |
| Well- and ill-posed problems, least squares | Well- and Ill-Posed Problems and Least Squares |
| Gradient and direct search methods, generalized inversion, SVD, global optimization | Gradient and Direct Search, Generalized Inversion, SVD and Global Optimization |
How CSIR NET Asks It
This unit is examined in Part C of the Earth Sciences paper. Per the NTA information bulletin for December 2026:
| Part C (Earth Sciences) | |
|---|---|
| Questions given | 80 |
| Questions to attempt | 25 |
| Marks per correct answer | 4 |
| Marks per wrong answer | −1.32 |
| Maximum | 100 marks |
Expect three kinds of question from this unit:
- Short numericals: one Runge–Kutta or Gauss–Newton step, a Simpson’s-rule integral of a short table, the largest stable time step, a straight-line least-squares fit, a Metropolis acceptance probability.
- Property recall in disguise: the order of a formula, the degree of precision of a rule, which condition of Hadamard fails, what a resolution matrix of the identity means.
- Statement sets: “which of the following are correct” about FDM and FEM, stability, the SVD and null spaces, or local versus global methods. These reward knowing the exact conditions (A-stable, rank, trace of \( R \)) rather than slogans.
There is no official breakdown of question counts by unit; nothing on these pages should be read as one.
Quick Reference
| Concept | Formula or fact |
|---|---|
| Central difference | \( \frac{f_{i+1} - f_{i-1}}{2h} \), error \( O(h^2) \) |
| Simpson’s 1/3 rule | \( \frac{h}{3}(f_0 + 4f_1 + f_2) \), precision 3 |
| \( n \)-point Gauss | exact to degree \( 2n - 1 \) |
| Explicit heat / 1-D wave | \( r \le \frac{1}{2} \) / \( v\Delta t/\Delta x \le 1 \) |
| RK4 | order 4, four evaluations per step |
| Euler stability | \( h \le 2/\lvert\lambda\rvert \) |
| Least squares | \( (G^{\mathsf T}G)^{-1}G^{\mathsf T}\mathbf{d} \) |
| Minimum norm | \( G^{\mathsf T}(GG^{\mathsf T})^{-1}\mathbf{d} \) |
| Damped / Marquardt | \( (G^{\mathsf T}G + \varepsilon^2 I)^{-1}G^{\mathsf T}\mathbf{d} \) |
| Generalized inverse | \( V_p\Lambda_p^{-1}U_p^{\mathsf T} \); \( R = V_pV_p^{\mathsf T} \) |
| Simulated annealing | accept uphill with \( e^{-\Delta E/T} \) |
Found an error? See About for how to report it.
Start Learning
Begin with the first topic: