I am not sure if I understand the following problem correctly nor if I am on the right path. Here it is, with my idea for a proof:
Consider the fixed point iteration method $x_{k+1}=g(x_k)$, $k= 0,1,...$ for solving the nonlinear equation $f(x) = 0$. Consider choosing an iteration function of the form:
$g(x) =x−af(x)−b(f(x))^2−c(f(x))^3$ where a, b and c are parameters to be determined. Find expressions for the parameters a, b and c such that the iteration method is of fourth order.
Idea: If $r$ is a root such that $f(r)=0$ then $g'(r)=0$, $g''(r)=0$, $g'''(r)=0$ but $g^{IV}(r)\neq 0$. As such, $g'(r) =1−af'(r)−2b(f(r))f'(r)−3c(f(r))^2f'(r)=0$ which means $a=\frac{1}{f'(r)}$.
Is this the right approach to find a? Where can I take it next? Thanks and regards,
The basic fixed-point iteration to find roots of $f$ has the form $$ g(x)=x-a(x)f(x)\implies g'(r)=1-a(r)f'(r). $$ So to get $g'(r)=0$, one needs $a(r)=1/f'(r)$. As $r$ is unknown, one can only set $a(x)=1/f'(x)+b(x)f(x)$ to get the correct value at $x=r$ while retaining the freedom to further adapt the formula. In consequence $$ g(x)=x-\frac{f(x)}{f'(x)}-b(x)f(x)^2\\ g'(x)=\frac{f(x)f''(x)}{f'(x)^2}-b'(x)f(x)^2-2b(x)f'(x)f(x) $$ This has a double root at $x=r$ if $$ \frac{f''(r)}{f'(r)^2}-2b(r)f'(r)=0 \implies b(x)=\frac{f''(x)}{2f'(x)^3}+c(x)f(x). $$ Now do a similar computation to determine the form of $c(x)$.
One could alternatively start with a formula with known 4th convergence order, $$ \tilde g(x)=x+p\frac{(1/f)^{(p-1)}}{(1/f)^{(p)}},~~~p=4-1=3, $$ treat the fraction as a division of Taylor series in $f(x)$ or $s=f(x)/f'(x)$ and carry it out to get a truncated Taylor series expansion for the quotient.
Another related approach is to directly employ the Taylor expansion of $f$, we want $f(g(x))=O(f(x)^4)$. So \begin{align} f(g(x))&=f(x)-f'(x)(af+bf^2+cf^3)+\tfrac12f''(x)(af+bf^2+cf^3)^2-\tfrac16f'''(x)(af+bf^2+cf^3)^3+O(f(x)^4) \\ &=(1-af'(x))f(x)+(-bf'(x)+\tfrac12a^2f''(x))f(x)^2 \\&\quad+(-cf'(x)+abf''(x)-\tfrac16a^3f'''(x))f(x)^3+O(f(x)^4) \end{align} also gives easy-to-solve equations if one wants the coefficients to be zero.