[Math] Convergence Analysis of Regula Falsi method

numerical methodsrate-of-convergenceregula falsi

I was reading about the convergence analysis of the regula falsi method in the book A friendly introduction to numerical analysis by B. Bradie.

I got stuck in the last step where $e_n=\lambda e_{n-1}$ and $$\lambda= \frac{l f''(p)}{2f'(p)+l f''(p)}$$ which comes from equation (4). However, I think we should have one more term in the denominator $a_n-p$ or $b_n-p$. Why did the author not consider it? If we consider it, then we won't get $e_n=\lambda e_{n-1}$ as now we also have one $e_{n-1}$ in the denominator.

Convergence Analysis

Does the sequence of approximations generated by the method of false
position, $\{p_n\}$ converge to a root $p$? If so, what is the order
of convergence? To answer these questions, we need to examine the
associated sequence of errors, $\{e_n\}$, where $e_n = p_n – p$. The
sequence $\{p_n\}$ converges if and only if $|e_n| \to 0$ as $n \to
\infty$ and the order of convergencee is ddetermined by the asymptotic
relationship between $|e_n|$ and $|e_{n-1}|$.

The error sequence $\{e_n\}$ is governed by what is known as the
error evolution equation. To construct the error evolution equation, take the equation for $p_n$ and subtract $p$ from both sides. This
yields: $$p_n – p = b_n – p – f(b_n) \frac{b_n – a_n}{f(b_n) –
f(a_n)}$$ Next, approximate the function values $f(a_n)$ and $f(b_n)$
by the second degree Taylor polynomials $$f(a_n) \approx
f'(p)(a_n-p)+\frac{f''(p)}{2}(a_n-p)^2,$$ $$f(b_n) \approx
f'(p)(b_n-p)+\frac{f''(p)}{2}(b_n-p)^2$$

enter image description here

Best Answer

The removal of one term is due to the fact that one of the bounds will converge to the root while the other will not. Suppose $a\to p$. Then $l=b-p$ and $(b+a-2p)\to(b+p-2p)=(b-p)=l$. On the other hand, $a-p$ is written as $e_{n-1}$, the term $\lambda$ is being multiplied to.