Fixed-point iteration

Level UniversityDifficulty ★★★★★Method⌖ Open in the map

What is it?

Rewrite the equation as x=g(x)x = g(x) and iterate xk+1=g(xk)x_{k+1} = g(x_k). If gg is a contraction (∣g′∣≤q<1|g'| \le q < 1), Banach's theorem guarantees a unique fixed point and linear convergence with ratio qq.

Formulas

∣g(x)−g(y)∣≤q ∣x−y∣, q<1  ⟹  ∣xk−x∗∣≤qk1−q ∣x1−x0∣|g(x) - g(y)| \le q\,|x - y|,\ q < 1 \implies |x_k - x^\ast| \le \frac{q^k}{1 - q}\,|x_1 - x_0|

Where it shows up in computing

  • Scientific computing★★★★★frequentScientific computing and algorithms

    Jacobi and Gauss–Seidel solvers, and many self-consistent schemes, are fixed-point iterations.

Where it shows up in AI

  • Reinforcement learning★★★★★fundamentalAI and machine learning

    Value iteration converges because the Bellman operator is a γ\gamma-contraction.

Where is it used?

Computing topics reachable from here, through the chain of ideas that leads to them:

This page has the essentials. A fuller treatment (intuition, formal definition, worked example) is on the way.

↑ ↓ to navigate · ↵ · Esc