Beruflich Dokumente
Kultur Dokumente
Fall 2008
Goals
I
Linear Systems
Vector Norms
Definition
A (vector) norm on Rn is a function
k k : Rn R
x kxk
which for all x, y Rn and R satisfies
1. kxk 0, kxk = 0 x = 0,
2. kxk = ||kxk,
3. kx + yk kxk + kyk,
(triangle inequality).
Vector Norms
The most frequently used norms on Rn are given by
kxk2 =
n
X
!1/2
x2i
2-norm
i=1
n
X
!1/p
p
|xi |
p-norm.
i=1
-norm.
Vector Norms
Example
Let x = (1, 2, 3, 4)T . Then
kxk1 = 1 + 2 + 3 + 4 = 10,
kxk2 = 1 + 4 + 9 + 16 = 30 5.48,
kxk = max {1, 2, 3, 4} = 4.
Vector Norms
The boundaries of the unit balls defined by
{x Rn : kxkp 1}.
One can show the following useful inequalities:
Vector Norms
The boundaries of the unit balls defined by
{x Rn : kxkp 1}.
One can show the following useful inequalities:
I
Vector Norms
The boundaries of the unit balls defined by
{x Rn : kxkp 1}.
One can show the following useful inequalities:
I
x, y Rn .
Vector Norms
The boundaries of the unit balls defined by
{x Rn : kxkp 1}.
One can show the following useful inequalities:
I
x, y Rn .
Cauchy-Schwarz inequality,
xT y kxk2 kyk2
for all
x, y Rn .
Vector Norms
Theorem
Vector norms on Rn are equivalent, i.e. for every two vector norms k ka
and k kb on Rn there exist constants cab , Cab (depending on the vector
norms k ka and k kb , but not on x) such that
cab kxkb kxka Cab kxkb
x Rn .
Matrix Norms
Definition
A matrix norm on Rmn is a function
k k : Rmn R
A kAk,
which for all A, B Rmn and R satisfies
1. kAk 0, kAk = 0 A = 0 (zero matrix),
2. kAk = ||kAk,
3. kA + Bk kAk + kBk,
(triangle inequality).
1/2
n X
m
X
kAkF =
a2ij .
i=1 j=1
1/2
n X
m
X
kAkF =
a2ij .
i=1 j=1
10
10
10
10
Matrix Norms
kAxkp
kAxkp
= sup kAxkp = max
= max kAxkp
x6
=
0
kxkp
kxkp
kxkp =1
kxkp =1
11
Matrix Norms
kAxkp
kAxkp
= sup kAxkp = max
= max kAxkp
x6
=
0
kxkp
kxkp
kxkp =1
kxkp =1
kAxkp
.
kxkp
(1)
Note that on the left hand side in (1) the symbol k kp refers to the
p-matrix-norm, while on the right hand side in (1) the symbol k kp
refers to the p-vector-norm applied to the vectors Ax Rm and
x Rn , respectively.
11
Matrix Norms
For the most commonly used matrix-norms (1) with p = 1, p = 2, or
p = , there exist rather simple representations.
Let k kp be the matrix norm defined in (1), then
kAk1 = max
j=1,...,n
kAk = max
i=1,...,m
kAk2 =
m
X
i=1
n
X
j=1
q
max (AT A) (spectral norm).
12
Matrix Norms
Example
Let
1
A = 2
2
Then
3 6
4
2 .
1 1
kAk1 = max 5, 8, 9 = 9,
kAk = max 10, 8, 4 = 10,
p
kAk2 = max {3.07, 23.86, 49.06} 7.0045,
kxkF = 76 8.718.
13
Matrix Norms
Two important inequalities.
Theorem
For any A Rmn , B Rnk and x Rn , the following inequalities
hold.
kAxkp kAkp kxkp
and
kABkp kAkp kBkp
kIxkp
= 1.
kxkp
kIkF = n.
D. Leykekhman - MATH 3795 Introduction to Computational Mathematics
14
Error Analysis
I
Let
Ax = b
(2)
15
Error Analysis
I
Let
Ax = b
(2)
Let
(A + A)
x = b + b
nn
(3)
n
and b R represent
15
Error Analysis
I
Let
Ax = b
(2)
Let
(A + A)
x = b + b
nn
(3)
n
and b R represent
Use a representation
x
= x + x.
15
nn
or
x = A1 b.
16
nn
or
x = A1 b.
Take norms:
kxk = kA1 bk kA1 kkbk.
(4)
(5)
(6)
16
17
Let
Ax = b
(7)
Let
(A + A)(x + x) = b + b
(8)
kAk
kxkp
1 p (A) kAkpp
kAkp
kbk
+
kAkp
kbk
.
18
(9)
For n = 4,
H 1
H=
1
2
1
3
1
4
16
120
=
240
140
1
2
1
3
1
4
1
5
120
1200
2700
1680
1
3
1
4
1
5
1
6
1
.
i+j1
1
4
1
5
1
6
1
7
240
2700
6480
4200
140
1680
.
4200
2800
19
Example
We compute that the condition number of a Hilbert matrix grows very
fast with n. For n = 4
25
kH 1 k1 = 13620, 1 (H) = 28375,
12
25
=
kH 1 k = 13620, (H) = 28375,
12
kHk1 =
kHk
kHk2 1.5
kH 1 k2 1.03 104 ,
20
(H)
2.837500e + 004
9.436560e + 005
2.907028e + 007
9.851949e + 008
3.387279e + 010
1.099651e + 012
3.535372e + 013
kxex xk
kxex k
2.958744e 013
5.129452e 012
5.096734e 011
2.214796e 008
1.973904e 007
4.215144e 005
5.382182e 004
Sensitivity of the Solution
21
Error Analysis.
I
kbk
0.5 10m+1
kb|
22
Summary.
23