Computational science uses computers as an instrument of discovery: you turn a question about nature or engineering into a mathematical model, solve it numerically, and compare the result with measurement. It sits beside theory and experiment, and it is how we study things too large, small, fast, slow or dangerous to test directly, from weather to bridges.
Key concepts
- Modelling: choose what to include and what to ignore, and state the equations and assumptions plainly.
- Discretisation: turn continuous problems into finite ones with grids, time steps and finite elements.
- Error and stability: round-off, truncation and conditioning decide whether a numerical answer can be trusted.
- Verification and validation: check that you solved the equations right, then that they describe reality.
- Performance: vectorisation, parallelism and memory layout often matter more than clever algorithms.
Vocabulary
- Convergence: the numerical answer approaching the exact one as the grid or step shrinks.
- Stiff problem: one where stability, not accuracy, forces tiny time steps.
- Monte Carlo: estimating a quantity by random sampling.
- Reproducibility: another person, or you later, getting the same result from the same inputs.
What to study, in order
- Calculus, linear algebra and a programming language with numerical arrays, such as Python with NumPy or Julia.
- Solve an ordinary differential equation with Euler's method, then a better one, and plot the error against step size.
- Study floating-point arithmetic by finding where a formula loses digits.
- Solve a heat or wave equation on a grid, and test convergence against a known solution.
- Profile your code, parallelise the slow part, and write down how you validated it.
Related tags
#numerical-analysis #simulation #hpc #research-software #linear-algebra