Skip to the content ForoHackers
Log inJoin

Numerical analysis: start here

Numerical analysis
#635 · Ophuscado · · Report

Numerical analysis designs and studies algorithms that compute approximate answers to continuous problems: solving equations, integrating, fitting curves, simulating physics. Computers use finite precision, so every result carries error. The field teaches you to control that error and to know when an answer can be trusted.

Key concepts

  • Floating point: numbers stored with limited precision, so rounding, cancellation and overflow are part of every computation.
  • Conditioning and stability: a problem can be sensitive to small changes in its input; an algorithm can amplify rounding errors.
  • Root finding: bisection, Newton's method and the secant method, and how fast each converges.
  • Linear systems: direct methods such as LU and iterative methods for large sparse systems.
  • Integration and differential equations: quadrature rules, and Euler and Runge-Kutta steps with their error orders.

Vocabulary

  • Machine epsilon: the gap between 1 and the next representable number, a measure of precision.
  • Truncation and rounding error: the error of the formula, and of the finite arithmetic.
  • Convergence order: how quickly the error shrinks per step.
  • Stiff problem: a differential equation whose fast and slow parts force tiny steps in simple methods.

What to study, in order

  1. Explore floating point: add 0.1 repeatedly, and find where cancellation hurts.
  2. Implement bisection and Newton's method, and plot the error per iteration.
  3. Solve linear systems with elimination and pivoting; see what happens without pivoting.
  4. Integrate numerically with the trapezoid and Simpson rules and measure the error order.
  5. Solve an ordinary differential equation with Euler and Runge-Kutta, and compare with the exact answer.

Related tags

#linear-algebra #simulation #computational-science #optimization #julia

Log in or join to reply.