next up previous
Next: Newton-Raphson Method: Up: ratish-1 Previous: Modified Regular Falsi method:

Convergence of secant method:

Definition: Say, $ x_{n}=\xi + e_{n},\qquad x_{n+1}=\xi +
e_{n+1}$ where $ \xi$ is the root of $ f(x)=0$. $ e_{n}$, $ e_{n+1}$ are the errors at n$ ^{th}$ and (n+1)$ ^{th}$ iterations and $ x_n,\,
x_{n+1}$ are the approximations of $ \xi$ at $ n^{th}$, (n+1)$ ^{th}$, iterations. If $ e_{n+1}=k\, e^{p}_{n}$ where $ k$ is a constant, then the rate of convergence of the method by which $ \{x_{n}\}$ is generated is p.
Claim: Secant method has super linear convergence.
Proof: The iteration scheme for the secant method is given by

$\displaystyle x_{n+1}=\frac{f(x_{n})x_{n-1}-f(x_{n-1})x_{n}}{(f(x_{n})-f(x_{n-1}))}\qquad\qquad(i)$

$\displaystyle =x_{n}-\frac{f(x_{n})(x_{n}-x_{n-1})}{f(x_{n})-f(x_{n-1})}\quad\,\,\qquad\qquad(ii)$

Say $ f(\xi)=0$ and $ e_{n}=(\xi-x_{n})$ i.e the error in the n$ ^{th}$ iteration in estimating $ \xi$.

$\displaystyle \left. \begin{aligned}
x_{n+1}=e_{n+1}+\xi \\
x_{n}=e_{n}+\xi \\
x_{n-1}=e_{n-1}+\xi
\end{aligned} \qquad\qquad\qquad\qquad\right \}(iii)$

Using (iii) in (ii) we get

$\displaystyle e_{n+1}=\frac{e_{n-1}f(x_{n})-f(x_{n-1})e_{n}}{f(x_{n})-f(x_{n-1})}\qquad(iv)$

By Mean value Theorem, $ \exists \quad a\quad\eta_{n}$ in the interval $ x_{n}$ and $ \xi$ s.t.

$\displaystyle f'(\eta_{n})=\frac{f(x_{n})-f({\xi})}{x_{n}-\xi}$

$\displaystyle \because \qquad f(\xi)=0,\quad x_{n}-\xi=e_{n} $

We get

$\displaystyle f'(\eta_{n})=\frac{f(x_{n})}{e_{n}}$    i.e. $\displaystyle \quad f(x_{n})=e_{n}f'(\eta_{n})\qquad \qquad(v)$

Using (iii)above, we get

$\displaystyle f(x_{n-1})=e_{n-1}\, f'(\eta_{n-1})\qquad\qquad \quad\qquad(vi)$

% latex2html id marker 3302
$ \therefore$                 using (v), (vi), in (iv) we get

$\displaystyle e_{n+1}=e_n\, e_{n-1}\frac{f'(\xi_n)-f'(\xi_{n-1})}{f(x_n)-f(x_{n-1})}$

$\displaystyle i.e. \qquad\qquad e_{n+1} \propto e_n \,\, e_{n-1} \qquad\qquad (vii)$

By def of rate of convergence the method is of order p if

$\displaystyle \left. \begin{aligned}
\displaystyle {e_n \propto e^p_{n-1} }\\
...
...e {e_{n+1} \propto e^p_n}
\end{aligned} \qquad\qquad\qquad\qquad\right \}(viii)$

>From (vii) and (viii) we get

$\displaystyle e^p_n \,\,\propto \,\, e_n \,\,\,e_{n-1}$

i.e.

$\displaystyle e^p_n \,\,\propto \,\, e^p_{n-1}\,e_{n-1}$

i.e.

$\displaystyle \qquad\qquad \qquad\qquad e_n
\,\, \propto e_{n-1} \qquad\qquad \qquad \qquad(ix)$

From (viii), (ix) we get

$\displaystyle p=(p+1)/p$


i.e.

$\displaystyle \quad p^{2}-p-1=0$

$\displaystyle \Rightarrow
p=\frac{1\pm\sqrt{5}}{2}$

$\displaystyle \because \, p \, > \, 0,\quad p=1.618$

% latex2html id marker 3324
$\displaystyle \therefore \qquad e_{n+1}\,\,\,\propto\,\,\, e_{n}^{1.618}$

Hence the convergence is superlinear. Example:
Solve $ 5 \sin^{2}x-8\cos^{5}x=0$ for the root in the interval [0.5,1.5] by secant method.
    Secant Method    
         
0 0.5000000000 1.5000000000 0.8773435354 2.1035263538
1 1.5000000000 0.8773435354 0.4212051630 -4.2280626297
2 0.8773435354 0.4212051630 0.7258019447 0.3298732340
3 0.4212051630 0.7258019447 0.7037572265 0.0327354670
4 0.7258019447 0.7037572265 0.7013285756 -0.0005388701
5 0.7037572265 0.7013285756 0.7013679147 0.0000009526


next up previous
Next: Newton-Raphson Method: Up: ratish-1 Previous: Modified Regular Falsi method:
root 2006-02-07