Review of Matrix Theory - I: Dr. Radhakant Padhi

Lecture – 13
Review of Matrix Theory – I
Dr. Radhakant Padhi

Asst. Professor
Dept. of Aerospace Engineering
Indian Institute of Science - Bangalore
Definition and Examples
z Definition: Matrix is a collection of elements
(numbers) arranged in rows and columns.
z Examples:
⎡1 ⎤
⎡1 2 3⎤
A1×3 = [1 2 3] , B3×1 = ⎢⎢ 2 ⎥⎥ , C2×3 =⎢ ⎥ ,
⎢⎣ 3 ⎥⎦ ⎣4 5 6⎦
⎡1 2⎤ ⎡1 4 6⎤
D3×2 = ⎢⎢3 4 ⎥⎥ , E3×3 = ⎢⎢ 4 2 5 ⎥⎥
⎢⎣5 6 ⎥⎦ ⎢⎣ 6 5 3⎥⎦
ADVANCED CONTROL SYSTEM DESIGN 2

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Definitions
z Symmetric matrix: A = AT
z Singular matrix: A =0
z Inverse of a matrix: B is inverse of A iff AB = BA = I

A−1 = adj ( A ) / A
Orthogonal matrix: AA = A A = I
T T
z
⎡cos θ − sin θ ⎤
• Example: T (θ ) = ⎢ ⎥
⎣ sin θ cos θ ⎦
• Result: Columns of an orthogonal matrix are orthonormal.

Eigenvalues and Eigenvectors
Matrices also act as linear operators with
“stretching” and “rotation” operations.
⎡ y1 ⎤ ⎡ x1 ⎤
⎡ 1⎤
y Y = AX ⎢ y ⎥ = A ⎢x ⎥
Q=⎢ ⎥
⎣y 2 ⎦ ⎣ 2⎦ ⎣ 2⎦
⎡ x1 ⎤
P=⎢ ⎥
⎣ x2 ⎦

Eigenvalues and Eigenvectors
z Question: Can we find a direction (vector), along which
the matrix will act only as a stretching operator?
z Answer: If such a solution exists, then
AX = λ X ⇒ ( λ I − A) X = 0
z For nontrivial solution, λI − A = 0
z Utility: Stability and control, Model reduction, Principal
component analysis etc.

Terminology Definition Properties of
Eigenvalues
Positive
X AX > 0 ∀X ≠ 0
T
λi > 0, ∀i
definite A > 0
Positive semi
definite A ≥ 0 X T AX ≥ 0 ∀X ≠ 0 λi ≥ 0, ∀i
Negative
X T AX < 0 ∀X ≠ 0 λi < 0, ∀i
definite A < 0
Negative semi
definite A ≤ 0
X T
AX ≤ 0 ∀X ≠ 0 λi ≤ 0, ∀i

Eigenvalues and Eigenvectors:
Some useful properties
z If λ1 , λ2 , … , λn are eigenvalues of An×n then
for any positive integer m, λ 1m , λ 2m , … , λ nm
are eigenvalues of Am
z If A is a nonsingular matrix with eigenvalues

λ1 , λ2 , … , λn , then λ1−1, λ−21........λ−n1 are
−1
eigenvalues of A
z For triangular matrix, the eigenvalues are
the diagonal elements

Eigenvalues and Eigenvectors:
Some useful properties
z If a An×n matrix is symmetric, its eigenvalues are all
REAL. Moreover, it has n linearly-independent
eigenvectors.
z If An×n has n real eigenvalues and n real orthogonal

eigenvectors, then the matrix is symmetric
z AT A and AAT are always positive semi definite.
z If A is a positive definite symmetric matrix, then every

principal sub-matrix of A is also symmetric and positive
definite. In particular, the diagonal elements of A are
positive.
Generalized Eigenvectors
If an eigenvalue is repeated p times, there may or may not
be p linearly independent eigenvectors corresponding to it.
In case linearly independent eigenvalues cannot be found,
generalized eigenvectors are the next option.
Example:
Suppose A3×3 has eigenvalues λ1 , λ2 , λ2 .
Then eigenvectors V1 ,V2 and generalized
eigenvector V3 can be found as follows:
(a) ( A − λ1 I )V1 = 0
(b) ( A − λ2 I )V2 = 0
(c) ( A − λ3 I )V3 = V2
Vector Norm
Vector norm is a “real valued function”
with the following properties:
(a) X > 0 and X = 0 only if X = 0

(b) α X = α X
(c) X +Y ≤ X + Y ∀X , Y

Vector Norm
X 1 = x1 + x2 + + xn (l1 norm)
X 2 (
= x1 + x2 +
2 2
+ xn )
2 1/ 2
(l2 norm)
=( x )
3 1/ 3
+ x2 + + xn
3 3
X 3 1 (l3 norm)
X p (
= x1 + x2 +
p p
+ xn )
p 1/ p
(l p norm)
X ∞ ( ∞
= x1 + x2 +
∞
+ xn )
∞ 1/ ∞
= max xi
i
(l∞ norm)

Matrix/Operator/Induced Norm
Definition:
AX
A = max = max ( A X )
X ≠0 X X =1
Properties:
(a) A > 0 and A = 0 only if A = 0
(b) α A = α A
(c) A+ B ≤ A + B
(d) AB ≤ A B
z 1-Norm
n
A 1 = max ∑ aij : Largest of the absolute column sums
1≤ j ≤ n
i =1
z 2-Norm
A 2 = σ max ( A ) : Largest Singular Value
z ∞ -Norm
n
A ∞ = max ∑ aij : Largest of the absolute row sums
1≤ i ≤ n
j =1

z Frobenius Norm
• Holds good for non-square matrices as well
• Used frequently in neural network and adaptive
control literature
1/ 2
⎛ m
2 ⎞
n
Am×n = ⎜ ∑∑ aij ⎟
⎝ i =1 j =1 ⎠
F

Spectral Radius
For An×n with eigenvalues λ1 , λ2 , … , λn
the spectral radius ρ ( A) is defined as
ρ ( A ) = max λ i
1≤ i ≤ n
A Result:
A 2 = ⎡⎣ ρ ( AT A ) ⎤⎦
1/ 2
If A is symmetric, then
A 2 = ⎡⎣ ρ ( A A ) ⎤⎦ = ⎡⎣ ρ ( A ) ⎤⎦ = ⎡( ρ ( A ) ) ⎤
1/ 2
= ρ ( A)
1/ 2 1/ 2 2
T 2
⎣ ⎦
Least Square Solutions
System: AX = b where A ∈ R m×n , X ∈ Rn , b ∈ Rm
(
Case 1: m = n and A ≠ 0 )
(No. of equations = No. of variables)
−1
Unique solution: X =A b

Least square solutions
Case 2: (m < n ) (under constrained problem)
(No. of equations < No. of variables)
In this case, there are infinitely many solutions. One way to
get a meaningful solution is to formulate the following
optimization problem:
Minimize J = X 2
, Subject to AX = b
( AA )
−1
+
Solution X = A b, where A = A + T T
( right pseudo inverse )
This solution WILL satisfy the equation AX = b exactly.

Least square solutions
Case 2: (m > n) (over constrained problem)
(No. of equations > No. of variables)
In this case, there is no solution. However, one way to get a
meaningful (error minimizing) solution is to formulate the
following optimization problem:
Minimize: J = AX − b 2
Solution: X = A b, where A = ( A A ) AT
−1
+ + T
( left pseudo inverse )
This solution need not satisfy the equation AX = b exactly.
Generalized/Pseudo Inverse
A = ( A A) A
+ T −1 T
• Left pseudo inverse:
A = A ( AA )
+ T T −1
• Right pseudo inverse:
• Properties: (a) A A+ A = A
(b) A+ A A + = A+
(c ) (A A ) + T
= A A+
( A A)
+ T
(d ) = A+ A
(e) A+ = A−1 , if A is square and A ≠ 0


Review of Matrix Theory - I: Dr. Radhakant Padhi

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Review of Matrix Theory - I: Dr. Radhakant Padhi

Hochgeladen von

Copyright:

Verfügbare Formate

Lecture – 13

Review of Matrix Theory – I

Dr. Radhakant Padhi

ADVANCED CONTROL SYSTEM DESIGN 2

z Inverse of a matrix: B is inverse of A iff AB = BA = I

ADVANCED CONTROL SYSTEM DESIGN 3

ADVANCED CONTROL SYSTEM DESIGN 4

ADVANCED CONTROL SYSTEM DESIGN 5

ADVANCED CONTROL SYSTEM DESIGN 6

z If A is a nonsingular matrix with eigenvalues

ADVANCED CONTROL SYSTEM DESIGN 7

z If An×n has n real eigenvalues and n real orthogonal

z AT A and AAT are always positive semi definite.

z If A is a positive definite symmetric matrix, then every

(a) X > 0 and X = 0 only if X = 0

ADVANCED CONTROL SYSTEM DESIGN 10

ADVANCED CONTROL SYSTEM DESIGN 11

ADVANCED CONTROL SYSTEM DESIGN 13

ADVANCED CONTROL SYSTEM DESIGN 14

ADVANCED CONTROL SYSTEM DESIGN 16

ADVANCED CONTROL SYSTEM DESIGN 17

ADVANCED CONTROL SYSTEM DESIGN 19

Das könnte Ihnen auch gefallen