Derivation: fixed-point convergence and order

Expanding ϕ\phi by Taylor around the fixed point yields the method's error equation and proves both the ϕ(α)<1|\phi'(\alpha)|<1 criterion and the order theorem.

Taylor of φ around the fixed point

  1. Let ek=xkαe_k=x_k-\alpha. Expand ϕ(xk)\phi(x_k) by Taylor around α\alpha and use ϕ(α)=α\phi(\alpha)=\alpha:

    xk+1=ϕ(xk)=α+ϕ(α)ek+ϕ(α)2ek2+ϕ(α)6ek3+x_{k+1}=\phi(x_k)=\alpha+\phi'(\alpha)\,e_k+\frac{\phi''(\alpha)}{2}\,e_k^2+\frac{\phi'''(\alpha)}{6}\,e_k^3+\cdots
  2. Subtracting α\alpha leaves the error equation of the fixed-point iteration:

    ek+1=ϕ(α)ek+ϕ(α)2ek2+ϕ(α)6ek3+e_{k+1}=\phi'(\alpha)\,e_k+\frac{\phi''(\alpha)}{2}\,e_k^2+\frac{\phi'''(\alpha)}{6}\,e_k^3+\cdots
  3. If ϕ(α)0\phi'(\alpha)\ne 0, the dominant term is linear: ek+1ϕ(α)eke_{k+1}\approx\phi'(\alpha)e_k. Errors contract if ϕ(α)<1|\phi'(\alpha)|<1 (linear convergence with factor ϕ(α)|\phi'(\alpha)|) and grow if ϕ(α)>1|\phi'(\alpha)|>1: that is the local convergence criterion.

  4. If moreover ϕ(α)=ϕ(α)==ϕ(p1)(α)=0\phi'(\alpha)=\phi''(\alpha)=\dots=\phi^{(p-1)}(\alpha)=0 and ϕ(p)(α)0\phi^{(p)}(\alpha)\ne 0, all earlier terms vanish and the first surviving one fixes the order: this is the order theorem for a fixed-point method.

    ek+1=ϕ(p)(α)p!ekp+O(ekp+1)e_{k+1}=\frac{\phi^{(p)}(\alpha)}{p!}\,e_k^p+\mathcal{O}\bigl(e_k^{p+1}\bigr)
  5. Immediate application: for Newton, ϕ=xff\phi=x-\frac{f}{f'} and ϕ=ff(f)2\phi'=\frac{f f''}{(f')^2}, which vanishes at a simple root (f(α)=0f(\alpha)=0). Since in general ϕ(α)0\phi''(\alpha)\ne 0, Newton has order 2, agreeing with the direct proof.