EE61209: Linear Systems and Control

Assignment 1
Amit Kumar(24EE62R17)
October 21, 2024

Exercise 1
Prove the following :

 
A1 0
1. If G = , then λi is an eigenvalue of G if and only if it
0 A2
is an eigenvalue of A1 or A2 .
Solution:
Let λ be an eigenvalue of G. By definition, there exists a non-zero vector
v such that:
Gv = λv.
Since G is block diagonal, we can write the vector v as:
 
v
v= 1 ,
v2

where v1 and v2 are vectors of appropriate dimensions corresponding to

the sizes of A1 and A2 .
Using the block structure of G, the equation Gv = λv becomes:
      
A1 0 v1 A1 v 1 v
= =λ 1 .
0 A2 v 2 A2 v 2 v2

From the equation above, we get two separate conditions:

A1 v1 = λv1 and A2 v2 = λv2 .

This implies that: - If v1 ̸= 0, then λ is an eigenvalue of A1 . - If v2 ̸= 0,

then λ is an eigenvalue of A2 .
Thus, the eigenvalues of G are exactly the **union of the eigenvalues** of
A1 and A2 .

1
  
A1 Q
2. If G = , then λi is an eigenvalue of G if and only if it is an
0 A2
eigenvalue of A1 or A2 .
A1 , A2 and Q are block matrices of appropriate dimensions.
Solution:
Let λ be an eigenvalue of G. By definition, there exists a non-zero vector
v such that:
Gv = λv.
We can write the vector v as:
 
v1
v= ,
v2

where v1 and v2 are vectors of appropriate dimensions.

Using the block structure of G, the equation Gv = λv becomes:
      
A1 Q v 1 A1 v1 + Qv2 v
= =λ 1 .
0 A2 v 2 A2 v 2 v2

From the second row of the equation:

A2 v2 = λv2 .

This implies that if v2 ̸= 0, then λ is an eigenvalue of A2 .

From the first row:
A1 v1 + Qv2 = λv1 .
If v2 = 0, this simplifies to:

A1 v1 = λv1 ,

which means that λ is an eigenvalue of A1 .

We have shown that the eigenvalues of the upper block triangular matrix:
 
A1 Q
G=
0 A2

are the eigenvalues of either A1 or A2 . The off-diagonal block Q does not

affect the eigenvalues.

Exercise 2
Consider the system:  
    u1
y1 1 2 2  
= u2
y2 2 1 2
u3

2
Tasks:
1. Compute the Singular Value Decomposition (SVD) of the system
matrix.
Solution:  
  1 2
1 2 2
A= , AT = 2 1
2 1 2
  2 2
9 8
AAT =
8 9
The eigenvalues of AAT are found by solving the characteristic equation:

det AAT − λI = 0.

This gives the eigenvalues:

λ1 = 17, λ2 = 1.

The singular values are the square roots of the eigenvalues:

√ √
σ1 = 17 ≈ 4.12, σ2 = 1 = 1.

Thus matrix Σ is:  

4.12 0 0
Σ= .
0 1 0
The left singular vectors are the eigenvectors of AAT .

• For λ1 = 17:
solving AAT u1 = λ1 u1 , we get
 
1 1
u1 = √ .
2 1

• For λ2 = 1:

solving AAT u2 = λ2 u2 , we get

 
1 1
u2 = √ .
2 −1

Thus, the matrix U is:  

1 1 1
U=√ .
2 1 −1
The right singular vectors are the normalized eigenvectors of AT A.

3
    
1 2   5 4 6
1 2 2
AT A= 2 1. = 4 5 6.
2 1 2
2 2 6 6 8

The eigenvalues of AT A are found by solving the characteristic equation:

det AT A − λI = 0.

This gives the eigenvalues:

λ1 = 17, λ2 = 1, λ3 = 0

The left singular vectors are the eigenvectors of AAT .

• For λ1 = 17:  
0.5144
v1 = 0.5144 .
0.6860
• For λ2 = 1:  
0.7071
v2 = −0.7071 .
0
• For λ3 = 0:  
0.4851
v3 =  0.4851  .
−0.7276

Thus, the matrix V is:

 
0.5144 0.7071 0.4851
V = 0.5144 −0.7071 0.4851  .
0.6860 0 −0.7276

The SVD of A is:

A = U ΣV T
 ⊤
   0.5144 0.7071 0.4851
1 1 4.12 0 0 
A= 0.5144 −0.7071 0.4851 
1 −1 0 1 0
0.6860 0 −0.7276

2. Find the maximal achievable ∥y∥2 over all inputs such that:

u21 + u22 + u23 = 1.

4
 Also, determine the direction of the corresponding output signal
with maximal gain.
Solution:
Given:
u21 + u22 + u23 = 1.
Therefore, ∥u∥2 = 1
The maximal achievable output ∥y∥2 is along to the largest singular value
that is along σ1

∥y∥2 = σ1 ∥u∥2 = 4.12 × 1 = 4.12.

The direction of the corresponding output signal is along the first column
of matrix U:  
0.7071
y ∝ u1 = .
0.7071

3. Find the maximal achievable ∥y∥2 over all inputs such that:

u21 + u22 + u23 ≤ 2.

Solution:
Given:
u21 + u22 + u23 ≤ 2.
√
Therefore, maximum value of ∥u∥2 = 2
The maximal achievable output ∥y∥2 is along to the largest singular value
that is along σ1
√
∥y∥2 = σ1 ∥u∥2 = 4.12 × 2 = 5.826.

Exercise 3
 
1+α 1−α
Consider the plant P = with α ∈ [0, 1]. Consider the
−1 + α −1 − α  
1 0
feedback loop with proportional controller C = k . Don’t use SVD of P
0 1
for points 2, 3, and 4.

5
Tasks:
   
r u
1. Let G : → . Find G.
d e

Solution:
   
u r
=G
e d
C(1 + P C)−1 −C(1 + P C)−1 P
   
G11 G12
G= =
(1 + P C)−1 −(1 + P C)−1 P G21 G22
     
1+α 1−α 1 0 k 0
where, P = and C = k =
−1 + α −1 − α 0 1 0 k

On calculating elements of matrix G we get,

" #
k(k+αk−1) k−αk+1
 2αk−1 1

4αk2 −1 4αk
− 2αk 2αk
G11 = −k+αk+1 G12 = −1 2αk+1
4αk − k+αk+1
4αk 2αk − 2αk
k+αk−1 k−αk+1
   2αk−1 1

4αk2 4αk2 − 2αk2 2αk2
G21 = −k+αk+1 G22 = −1
4αk2 − k+αk+1
4αk2 2αk2
2αk+1
− 2αk2

2. With α ̸= 0, show that as gain k → ∞, e → 0 and u = P −1 r − d.

Solution:
   1+α 1−α

1+α 1−α
P = , P −1 = −1+α
4α 4α
−1 + α −1 − α 4α − 1+α
4α

    
u G11 G12 r
= (1)
e G21 G22 d

k+αk−1 k−αk+1
 
G11 = 4αk
lim −k+αk+1 4αk
k→∞ 4αk − k+αk+1
4αk
 1+α 1−α

= 4α 4α
−1+α
4α − 1+α
4α
−1
=P

 2αk−1 1

− 2αk 2αk
G12 = lim −1 2αk+1
k→∞ 2αk − 2αk
 
−1 0
=
0 −1
= −I

6
 So from the equation (1) u can be written as,

u = G11 r + G12 d = P −1 r − Id
= P −1 r − d (2)

k+αk−1 k−αk+1
 
G21 = 4αk2
lim −k+αk+1 4αk2
k+αk+1
k→∞ 4αk2 − 2
 1+α 1−α
4αk
= 4αk
lim −1+α 4αk
k→∞ 4αk − 1+α
4αk
=0

 2αk−1 1

− 2αk2 2αk2
G22 = lim −1 2αk+1
k→∞ 2αk2 − 2αk2
−1 1
 
= lim k 2αk2
−1 −1
k→∞ 2αk2 − 2αk
=0

So from equation (1) e can be written as,

e = G21 r + G22 d = 0r + 0d
=0 (3)

So, from equation (2) and (3) we can se that as k → ∞, e → 0 and

u = P −1 r − d
3. For a finite k, if α = 0, ∥e∥ is zero only if r1 = r2 = 0 and d1 = −d2 .
(For a vector y, yi is its ith component).
Solution:

2
   
k(k+αk−1)
 4αk 2 −1 − k4αk
(α−1)
2 −1 
k(α−4αk+1)
 4αk 2 −1
k(α−1)
− 4αk 2 −1
G11 = 2 G12 =  k(α−1)

k (α−1) k(k+αk+1)  k(α+4αk+1 
4αk 2 −1 − 4αk2 −1 4αk 2 −1 − 4αk2 −1

 
k+αk+1 k(α−1) 
α−4αk+1 α−1

2 − 4αk2 −1  2 −1 − 4αk 2 −1
G21 =  4αk −1 G22 =  4αk

α−1 α+4αk+1

k(α−1)
− k+αk+1 − 4αk2 −1

4αk 2 −1 4αk 2 −1 4αk 2 −1

7
 For some finite k and α=0 we have,
−k 2
   
−k(k − 1) −k −k
G11 = G12 =
k2 k(k + 1) k k

   
1−k −k −1 −1
G21 = G22 =
k k+1 1 1
From equation (1),
e = G21 r + G22 d

     
1−k −k r1 −1 −1 d1
= +
k k+1 r2 1 1 d2

 
(1 − k)r1 − kr2 − d1 − d2
=
kr1 + (k + 1)r2 + d1 + d2

p
∥e∥ = ((1 − k)r1 − kr2 − d1 − d2 )2 + (kr1 + (k + 1)r2 + d1 + d2 )2

from above equation for ∥e∥ to be zero for some finite k,

(1 − k)r1 − kr2 − d1 − d2 = 0
kr1 + (k + 1)r2 + d1 + d2 = 0
solving above both equation we get r1 = r2 = 0 and d1 + d2 =0 which gives
d1 = −d2 .
Therefore, ∥e∥ =0 only if r1 = r2 = 0 and d1 = −d2
4. If k → ∞, find a condition under which ∥e∥ is bounded.
Solution: From the part, ∥e∥ can be written as,
p
∥e∥ = ((1 − k)r1 − kr2 − d1 − d2 )2 + (kr1 + (k + 1)r2 + d1 + d2 )2

p
= (r1 − kr1 − kr2 − d1 − d2 )2 + (kr1 + kr2 + r2 + d1 + d2 )2
as k → ∞, ∥e∥ → ∞
for ∥e∥ to be bounded it must be independent of k as it makes it infinite.
Therefore,

−kr1 − kr2 = 0
kr1 + kr2 = 0
Solving above equations, we get r1 = −r2
For ∥e∥ to be bounded r1 = −r2

8
  
1 0
5. Assuming d = 0, and controller as C = k , show that even
0 1
in the presence of the controller, the feedback loop effectively
acts as a open loop system. (Hint: Use SVD).
Solution:
The plant matrix is given by:
 
1+α 1−α
P =
−1 + α −1 − α
 
1 + α −1 + α
PT =
1 − α −1 − α
2 + 2α2 2α2 − 2
     
1+α 1−α 1 + α −1 + α
PPT = . =
−1 + α −1 − α 1 − α −1 − α 2α2 − 2 2 + 2α2

Solving for the eigen values of above matrix we get,

λ1 = 4, λ2 = 4α2 .
The singular values are the square roots of the eigenvalues:
p p
σ1 = λ1 = 2, σ2 = λ2 = 2α.
Thus matrix Σ is:  
2 0
Σ= .
0 2α
The normalized left singular vectors are the eigenbasis of P P T .
• For λ1 = 4:
 
1 −1
u1 = √ .
2 1
• For λ2 = 4α:
 
1 1
u2 = √ .
2 1
Thus, the matrix U is:  
1 −1 1
U=√ .
2 1 1
The normalized right singular vectors are the eigenbasis of P T P .
   
1+α −1 + α 1+α 1−α
PTP = .
1−α −1 − α −1 + α −1 − α

2(1 + α2 ) 2(1 − α2 )
 
= .
2(1 − α2 ) 2(1 + α2 )

9
 Eigenvalues of the above matrix is given by:

λ1 = 4, λ2 = 4α2

The left singular vectors are the eigenvectors of P T P .

• For λ1 = 4:  
1 1
v1 = √ .
2 1
• For λ2 = 4α:
 
1 −1
v2 = √
2 1
Thus, the matrix V is:  
1 1 −1
V =√ .
2 1 1
Therfore SVD of matrix P is given by
      ⊤
T 1 −1 1 2 0 1 1 −1
U ΣV = √ √
2 1 1 0 2α 2 1 1

6. Can you explain points 2, 3, and 4 using SVD of P and P −1 . Are

there any more insights that one can gather?
Solution:
SVD of P is :       ⊤
T 1 −1 1 2 0 1 1 −1
P = UΣV = √ √
2 1 1 0 2α 2 1 1
 1+α 1−α

P −1 = −1+α 4α 4α
−1−α
 4α 4α 
1 1+α 1−α
= 4α
−1 + α −1 − α
1
= 4α P
1
= 4α U ΣV T
    1   ⊤
1 −1 1 2α 0 1 1 −1
= √ √
2 1 1 0 12 2 1 1

10