Beruflich Dokumente
Kultur Dokumente
Com S 477/577
Oct 5, 2004
A sequence xi R converges to if for each > 0, there exists an integer N () such that
|xl | > for all l N (). The Cauchy convergence criterion states that a sequence xi R
is convergent if and only if for each > 0 there exists an N () such that |xl xm | < for all
l, m N ().
Let a sequence xi R be generated by an iteration function , that is,
xi+1 = (xi ),
i = 0, 1, 2, . . .
Let be a fixed point of , that is, = (). Suppose that the sequence {xi } is generated in the
neighborhood of . The corresponding iteration method is said to be of at least pth order if there
exists a neighborhood N () of such that for all x0 N () the generated sequence xi+1 = (xi ),
i = 0, 1, . . ., satisfies
|xi+1 | C|xi |p ,
where C < 1 if p = 1. In the case of first order convergence, for instance, we have
|xi | C|xi1 | C 2 |xi2 | C i |x0 |.
Since C < 1, it follows that
lim |xi | = lim C i |x0 | = 0,
(p) ()
(xi )p + O (xi )p+1 .
p!
Because () = , we obtain
xi+1
(p) ()
=
.
i (xi )p
p!
lim
( x)
( x)
xi 0
x i+1
x i +2
xi
x i +2
x i+1
In the below we study the convergence rates of several root finding methods introduced before.
with
f () = 0.
Suppose f is sufficiently continuously differentiable in some neighborhood N (). In the nondegenerate case, f () 6= 0. So we have
() = ,
2
() = 1
2
f (x)
x=
= 0,
since f () = 0,
2
2
For clarity of analysis we let xi = bi for all i. We make some simplification assumptions for the
discussion of the convergence behavior: f exists and for some k the following conditions hold: (a)
xk < ak ; (b) f (xk ) < 0 and f (ak ) > 0; (c) f (x) 0 for all x [xk , ak ].
x k +1
xk
ak
(a)
(b)
Under these assumptions, either f (xk+1 ) = 0 or f (xk+1 )f (xk ) > 0 and consequently
xi < xi+1 < ai+1 = ai .
To see this, use the remainder formula for polynomial interpolation at xk and ak :
f (x) p(x) = (x xk )(x ak )
f ()
2
for x [xk , ak ] and a suitable [xk , ak ]. Under condition (c), the above equation implies that
f (x) p(x) 0. In particular, f (xk+1 ) p(xk+1 ) 0, which in turn implies that f (xk+1 ) 0 since
p(xk+1 ) = 0.
Unless f (xk+1 ) = 0, in which case the iteration stops at xk+1 , we can see that conditions (a),
(b), and (c) hold for all i k. Therefore ai = ak = a and
xi+1 =
af (xi ) xi f (a)
f (xi ) f (a)
for all i k. Furthermore, {xi } for i k form a monotone increasing sequence bounded by a. So
limi xi = exists. Consequently,
f () 0,
f (a) > 0,
and
af () f (a)
,
f () f (a)
which gives
( a)f () = 0.
But < a since f () 0 < f (a). Hence f () = 0 and {xi } converges to a zero of f .
The above discussion enables us to look at the order of convergence through the iteration
function
af (x) xf (a)
xi+1 = (xi ),
where
(x) =
.
f (x) f (a)
3
=
=
f (a)
= f (1 ),
a
f (xi )
= f (2 ),
xi
< 1 < a;
(1)
xi < 2 < .
(2)
f (xi )
,
f [xi1 , xi ]
i = 0, 1, . . .
(3)
To determine the local convergence rate, without loss of generality we assume that the sequence
{xi } is in a small enough neighborhood of the zero and that f is twice differentiable. Subtract
from both sides of (3):
f (xi )
xi+1 = (xi )
f [xi1 , xi ]
f (xi )
f [xi , ]
f (xi ) f ()
=
= (xi ) 1
,
since f [xi , ] =
f [xi1 , xi ]
xi
xi
f [xi1 , xi ] f [xi , ]
= (xi )(xi1 )
(xi1 )f [xi1 , xi ]
f [xi1 , xi , ]
= (xi )(xi1 )
.
f [xi1 , xi ]
From error estimation of polynomial interpolation, we learned that
f [xi1 , xi ] = f (1 ),
1
f (2 ),
f [xi1 , xi , ] =
2
4
1 I[xi1 , xi ];
2 I[xi1 , xi , ],
(4)
where I[xi1 , xi ] is the smallest interval containing xi1 and xi , and I[xi1 , xi , ] the smallest
interval containing xi1 , xi ,
n
If is a simple zero, that is, f () 6= 0, there exists a bound M and an interval J = x |x|
o
for some > 0 such that
1 f (2 )
(5)
2 f (1 ) M,
for any 1 , 2 J.
Let ei = M |xi | and e0 , e1 < min{1, M }. By induction and using (4) and (5) we can easily
show that
ei ei1
ei+1 = M |xi+1 | M
M = ei ei1 ,
M M
and
|ei | min{1, M },
for i = 1, 2, . . ..
Let q = (1 + 5)/2 be the root of the equation z 2 z 1 = 0. Then we have
i
ei K q ,
where K = max{e0 ,
i = 0, 1, 2, . . .
q
e1 } < 1. This is because (by induction)
i
ei+1 ei ei1 K q K q
i1
= Kq
i1 (q+1)
= Kq
i1 q 2
= Kq
i+1
Thus the secant method converges at least as well as a method of order p = 1+2 5 = 1.618 . . ..
One-step secant requires one additional function evaluation. But one-step Newton requires two
(f and f ). Therefore two secant steps are as expensive as single Newton step. But two secant
steps has a convergence order of (1.618)2 2.618. This explains why in practice the secant method
always dominates Newtons method with numerical derivatives.
References
[1] J. Stoer and R. Bulirsch. Introduction to Numerical Analysis. Springer-Verlag New York, Inc.,
2nd edition, 1993.
[2] M. Erdmann. Lecture notes for 16-811 Mathematical Fundamentals for Robotics. The Robotics
Institute, Carnegie Mellon University, 1998.
[3] W. H. Press, et al. Numerical Recipes in C++: The Art of Scientific Computing. Cambridge
University Press, 2nd edition, 2002.