Annotated
Annotated
Root-finding: Fixed-point
iteration
Peng Zhang
2025-02-14
Today’s plan
• Root-finding: fixed point iteration
• f(x) = 3x - 2, f(x) = x2 + 2x - 3
• f(x) = x2 + ecos(x) - 3
f(xn)
iteratively improve our guess: xn+1 = xn − , for n ≥ 0
• f′(xn)
~ def
=
g(xN)
Root finding
• Recall Newton’s method:
f(xn)
iteratively improve our guess: xn+1 = xn − , for n ≥ 0
• f′(xn)
X.
,
Xz , . .
x = 5+ x x
All four iterations have the property:
-
equations, we get α =± 5.
d= E(x 5)()x E
+
=
Motivational example
2
• Consider solving the equation x − 5 = 0 for the root
x* = 5 = 2.2361...
x2
• If xn converges to α, then
lim Xn + lim
1 =
g(Xn)
n n-
>
-
9([MXn) (1 :
g is continuous)
=> d =
g(x)
Motivational example
• All the four iterations have the form xn+1 = g(xn)
for a continuous function g.
• If xn converges to α, then
α = g(α)
• Geometrically, fixed points lie at the intersection of the line y = x and the
graph of y = g(x).
= X -
M
(Xo ,
g(Xo))
Y g(X)
= ·
(Xi , XI)
X5 X4
Sufficient Condition for Convergence
iteration : X =
g(Xn) , #2 o
• Theorem. Assume that g(x), g′(x) are continuous and | g′(x) | < 1 in
some interval [x* − δ, x* + δ] centered about a fixed point x* of g. If x0
is in this interval, then the fixed point iteration xn+1 = g(xn), n ≥ 0
converges to x*.
[ 2
x* *S
X
Sufficient Condition for Convergence
ll -
2x(x = 0xX)
• Theorem. Assume that g(x), g′(x) are continuous and | g′(x) | < 1 in
some interval [x* − δ, x* + δ] centered about a fixed point x* of g. If x0
[5-8 55 S
is in this interval, then the fixed point iteration xn+1 = g(xn), n ≥ 0
.
converges to x*. [1 :
g(x) =
5+ X -
x2
g((x) =
1 -
2x
*
X =
5 = 2 .
2361 ...
19'125) 1 =
11 5 -4)
-
=
> 1
Xo
[2 :
g(x) = g'(x =
-
*
(g(x))) =>
(14) = X +
(5 3)
,
+
13 :
g(x) = 1 + x =
jxz UK & .
-
55)
g'(x) = 1 -
EX .
19ix , / <
1, = 11 -
Ex)4
Sufficient Condition for Convergence
=> -
11 -
EX() = 0 < X 5
,
[E -
1 , F +]
Xo 2 5
Theorem. Assume that g(x), g′(x) are continuous and | g′(x) | < 1 in
=
•
.
If Xn+ 1 =
9(Xn) = g(X ) *
+ g'(n)(Xn -
X
*
) where In is
,
1 :
*
X is a fixed *
between Xn ,
X
*
= x +
g'(n)(Xn *
X )
point)
-
=> Xn + X
g((n)(Xn *
-
~ -
-
X
)
def en + def
= en
Sufficient Condition for Convergence
-
= en =
9 (n) en
consider
lem 1
+
=
19 n))-len
Xo[x* 8
#
n = 0
.
,
x*+ 5)
(e
=>
. ) max Igia)) ·
1eol IX.-x 1- (X
*
.
-
**
ce[X* 8 .
*
x+ 8]
-
-1
*
=> X , e[X -
8 x
*
+ S]
,
In . if Xne[X
*
general -
8 .
*** 5)
=>
/en + / = max 1g)1 ·
Jen
ce[X * 6 .
x +
*
6]
-
def M M <
/
=> Lentil a
M **, e.
~
-
> constant
=> **
As n >
-
a
,
M 12 . >
-
0
.
=>
len1 0
limit, al
In the lim
n> -
- (en)
=
9'(x *
)
Another Sufficient Condition for Convergence
XI contraction
=
the iteration xn+1 = g(xn) converges to x* from any x0.
g,x
X
not contraction
Rough idea :
(ax =
Xn) = 19(Xn) -
g(Xn 1)) -
*
G(Xn -
Xa ) ,
n21
"
=> (Xn +
-
Xn) = ( (X ,
-
Xo)
See the next page Theorem 4.5.2 for a rigorous proof for this theorm.
(b) Next consider . In this case, |φ′2(x)| = /(x + 6)3/2,
which is less than 1 for x > 21/3 − 6 ≈ −4.74. Since we know that there
is a fixed point x* somewhere in the interval [1, 2], the distance from x0
= 1.5 to this fixed point is at most 0.5; hence the interval [x* − 0.5, x* +
0.5] ⊂ [0.5, 2.5] contains x0 and is an interval on which |φ′2(x)| < 1.
Therefore convergence is guaranteed. One also could establish
convergence by noting that φ2 maps the interval [1, 2] that contains a
fixed point to the interval [1, ] ⊂ [1, 2]. Therefore, by the
discussion after theorem 4.5.1, the iteration must converge.
(c) Finally consider . In this case |φ′3(x)| = (x2/4) ×
. It can be checked that this is less than 1 for x less than
about 1.6, but if we know only that there is a fixed point in the interval
[1, 2], we cannot conclude that fixed point iteration applied to φ3 will
converge. Since we know from running fixed point iteration with φ2,
however, that the fixed point x* is at about 1.06, we can again conclude
that in the interval [x* − 0.5, x* + 0.5], which contains the initial guess
x0 = 1.5, we have |φ′3(x)| < 1; hence the fixed point iteration will
converge to x*.
Actually, φ′(x*) need not exist in order to have convergence. The iteration
(4.21) will converge to a fixed point x* if φ is a contraction; that is, if there
exists a constant L < 1 such that for all x and y
Theorem 4.5.2. If φ is a contraction (on all of R), then it has a unique fixed
point x* and the iteration xk+1 = φ(xk) converges to x* from any x0.
If k and j are both greater than or equal to some positive integer N, we will
have
and this quantity goes to 0 as N → ∞. This shows that the sequence xk, k = 0,
1, . . . is a Cauchy sequence and hence converges.
To show that xk converges to a fixed point of φ, we first note that the fact
that φ is a contraction implies that it is continuous. Hence φ(limk→∞ xk) =
limk→∞ φ(xk). Letting x* denote limk→∞ xk, we have
so that the limit of the sequence xk is indeed a fixed point. Finally, if y* is also
a fixed point, then since φ is a contraction, we must have
should be <=
|x* − y*| = |φ(x*) − φ(y*)| − L|x* − y*|,
where L < 1. This can hold only if y* = x*, so the fixed point is unique.
Theorem 4.5.2 could be modified to assume only that φ is a contraction on
some interval [a, b], where φ maps [a, b] into itself. The conclusion would
then be that there is a unique fixed point x* in [a, b] and that if fixed point
iteration is started with an initial guess x0 ∈ [a, b], then it converges to x*.