You are on page 1of 2

Computational Statistics

Homework 1
Eirik Aalvik Stranden Uni: eas2270
January 29, 2015

Show/Find the convergence rate of the secant-method

So we consider a function g(x), where x is a root such that g 0 (x ) = 0. First, Newton with
Taylor expansion gives
g 0 (xt )
(1)
xt+1 = xt 00 t
g (x )
Further we approximate g 00 (xt ) with finite differences(backward)
g 00 (xt )

g 0 (xt ) g 0 (xt1
xt xt1

(2)

Then we plug 2 into 1 which gives the secant-method


xt+1 = xt g 0 (xt )

xt xt1
g 0 (xt1

g 0 (xt )

(3)

Further we define xt+1 x = et+1 which is the error of the xt+1 iterate. This is equivalent for
the et and et1 error. If we insert these expressions for the error in 3 we get
et+1 + x = et + x

g 0 (xt )
(et + x et1 x )
g 0 (xt ) g 0 (xt1 )

(4)

Then if we subtract x we obtain


et+1 = et

g 0 (xt )
(et et1 )
g 0 (xt1

g 0 (xt )

(5)

Further collect the terms by making a common denominator:


et+1 =

et g 0 (xt ) et g 0 (xt1 ) g 0 (xt )et + g 0 (xt ) et1


g 0 (xt ) g 0 (xt1 )

This simplifies to

(6)

et1 g 0 (xt ) et g 0 (xt1 )


g 0 (xt ) g 0 (xt1

et+1 =

(7)

Then we can use the Mean Value Theorem which says that there exist a t [xt , x ] such that
0
t
0
(x )
g 00 ( t ) = g (xxt)g
. But here g 0 (x ) = 0 and we have et defined earlier. By using the same for
x
00 t1
g ( ) one gets by inserting this in 7
et+1 =

et1 et g 00 ( t ) et et1 g 00 ( t1 )
g 0 (xt ) g 0 (xt1 )

(8)

By pulling et et1 out of the expression one obtains


et+1 = et et1

g 00 ( t ) g 00 ( t1 )
g 0 (xt ) g 0 (xt1 )

(9)

If we call the fraction in equation 9 for M t we obviously have M t M when t inf because
this converges to a constant. This means that et+1 et et1 or
et+1 M t et et1

(10)

We also know that for

|et+1 |
=C
x |et |
for a constant C then it has order . So we have
lim

lim |et+1 | = C lim |et |

(11)

lim |et | = C lim |et1 |

(12)

which gives us
lim |et1 | = (

1
1
lim |et | )
C x

(13)

Inserting equations 11 and 13 in 10 we get


lim |et+1 | lim |M t et et1 | lim |M t | lim |et | lim |et1 |

(14)

This gives us
1

limx |et | limx |et |


CC
=
t

limx |e |
M
Since the right side is a positive constant and et 0 we need 1 +

(15)
1

=0

And the solution of 2 1 = 0 is 1.62. Which means the order is 1.62.


I have used the textbook Givens and Hoeting (2005) Computational statistics and the web page
http://nptel.ac.in/courses/122104019/numerical-analysis/Rathish-kumar/ratish-1/f3node5.html
in this work.
2

You might also like