1
$\begingroup$

I have the following iteration method $$x_{n+1}=x_n-\frac{2f(x_n)f'(x_n)}{2[f'(x_n)]^2-f(x_n)f''(x_n)}\hspace{3mm} (Halley)$$ and I want to show that it has cubic convergence (I know that the solution has a multiplicity of one and $f''$ satisfies Lipschitz's condition). I somehow needed to show that $$\exists c\in\mathbb{R}:\hspace{1mm}|x_{n+1}-a|\leq c\cdot |x_n-a|^3\hspace{1mm}$$ where $a$ is the solution. I was stuck with this problem. But then I found a proof here (Proof). I thought I finally got it, but still some things didn't add up. First, the Wikipedia proof uses functions third derivative, I don't know if I can assume that it exists? Probably not. So if I can't assume that, I'm back to square one (because I can't use Taylor's theorem in my proof). Second, I don't know where I should use the facts that the solution a multiplicity of one and $f''$ satisfies Lipschitz's condition. Other than that the Wikipedia proof seemed pretty straightforward.

So, can anyone help me with that proof and/or give insight how to use my given assumptions?

$\endgroup$

1 Answer 1

1
$\begingroup$

You can assume anything you want if it helps you get the results you want. Later on you can try to make the assumptions more restrictive, but I always assume anything I need (for example differentiable instead of just continuous).

As to assuming that the third derivative exist, I think the usual proof of Newton's method assumes the second derivative exists so I see nothing wrong in assuming the third derivative exists here.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.