EE580, Fall 2023

Homework four
Due on November 14, 2023, 11:00PM. Upload your solution to Canvas.
Please write down name, student ID and course code on the first page of the solution.

Problem 1: (15points) 
−1 0 2
(P1) Is ẋ(t) =  0 0 0  x(t) marginally stable? asymptotically stable? unstable?
 0 0 0 
−1 0 2
(P2) Is ẋ(t) =  0 0 1  x(t) marginally stable? asymptotically stable? unstable?
 0 0 0
−1 0
(P3) Is ẋ(t) = −3t x(t) for t ≥ 0 marginally stable? asymptotically stable? unstable?
−e 0
Solution:
(P1) (4 points) Matrix A is diagonal and its eigenvalues are −1, 0 and 0. So the system is not asymptotically
stable. The minimal polynomial is either (λ + 1)λ or (λ + 1)λ2 . Since (A + I)A = 0, then (λ + 1)λ is the
minimal polynomial and 0 is its simple root. Hence, the system is marginally stable.
(P2) (4 points) Matrix A is diagonal and its eigenvalues are −1, 0 and 0. So the system is not asymptotically
stable. The minimal polynomial is either (λ + 1)λ or (λ + 1)λ2 . Since (A + I)A 6= 0 and (A + I)A2 = 0,
then (λ + 1)λ2 is the minimal polynomial and 0 is not its simple root. Hence, the system is unstable.
(P3) (7 points) Since ẋ1 (t) = −x1 (t), then we have the solution x1 (t) −t
R t = e−4τx1 (0). It renders that ẋ2 (t) =
−e x1 (t) = −e x1 (0). So we have the solution x2 (t) = x2 (0)+ 0 (−e x1 (0))dτ = 51 (e−5t −1)x1 (0)+
−3t −4t

x2 (0).
Choose x(0) = [1 0]T and then x(t) = [e−t 15 (e−5t −1)]T . Choose x(0) = [0 1]T and  then x(t) = [0 1] T
 .
e −t 0 e t 0
So the fundamental matrix X(t) = 1 −5t and its inverse is X(t)−1 = 1 t −4t ) 1 .
5 (e − 1) 1 5 (e − e
e−t+t0
 
−1 0
Therefore, the state transition matrix is Φ(t, t0 ) = X(t)X(t0 ) = 1 −5t+t0 .
5 (e − e−4t0 ) 1
One can see that every component of Φ(t, t0 ) is uniformly bounded, and then the system is marginally stable.
As t → +∞, the second row does not go to zero. Then the system is not asymptotically stable.

Problem 2: (10 points) Is the network BIBO stable? If not, find a bounded input that will excite an unbounded
output.
Solution: (6 points) The impendence of the capacitor is 1s . Note that the capacitor and inductor are in parallel. So
s· 1
the transfer function from u to y is ĝ(s) = s+s1 = s2s+1 . Then the impulse response is g(t) = cos t and it is not
s
absolutely integrable. So the system is not BIBO stable.
(4 points) Choose u(t) = sin t and its Laplace transform is û(s) = s21+1 . So ŷ(s) = ĝ(s)û(s) = (s2 +1)
s
2 =
jt −jt
1 1 1 −1 1 jt −1 1 −jt −1 te −te 1
4j ( (s−j)2 − (s+j)2 ). Note that L [ s−j ] = e and L [ s+j ] = e . So y(t) = L [ŷ(s)] = 4j = 2 t sin t
which is unbounded.

Problem 3: (20 points)

(P1) Let A be a symmetric and n × n matrix and λmin (A) (resp. λmax (A)) is the smallest (resp. largest) eigenvalue
of P . Show that λmin (A)kxk2 ≤ xT Ax ≤ λmax (A)kxk2 for any x.
(P2) Consider ẋ(t) = Ax(t) where all the eigenvalues of A have negative real parts. Use Lyapunov theorem to
show that the system is exponentially stable; i.e., there are c, α > 0 such that kx(t)k ≤ ce−αt kx(0)k for any
initial state x(0).
(P3) Given ẋ(t) = Ax(t), compare its marginal stability, asymptotic stability and exponential stability.
Hints for Part (P2): The following fact is known: if v̇(t) ≤ −av(t), then v(t) ≤ e−a(t−t0 ) v(t0 ) where v(t) is a
differentiable scalar signal.
Solution:
(P1) (7 points) Since A is symmetric, so is diagonalizable. That is, there is an orthogonal matrix P such that A =
P −1 ΛP = P T ΛP where Λ is diagonal and consists of the eigenvalues P of A, say λi . So xT Ax = xT P T ΛP x =
y Λy where y = P x. Since Λ is diagonal, λmin (A)kyk ≤ y Λy = ni=1 λi yi2 ≤ λmax (A)kyk2 .
T 2 T

Since P is orthogonal, then y T y = xT P T P x = xT x and kxk = kyk. Therefore, λmin (A)kxk2 ≤ xT Ax ≤

λmax (A)kxk2 .
(P2) (10 points) Pick a symmetric and positive definite matrix N . Since A is Hurwitz, then there is a unique,
symmetric and positive definite matrix M such that AT M + M A = −N .
Choose V (x) = xT M x and its derivative along the trajectory of ẋ(t) = Ax(t) is given by
V̇ (x(t)) = x(t)T (AT M + M A)x(t) = −x(t)T N x(t).
Since −x(t)T N x(t) ≤ −λmin (N )kx(t)k2 and x(t)T M x(t) ≤ λmax (M )kx(t)k2 , then we have
λmin (N )
V̇ (x(t)) ≤ −λmin (N )kx(t)k2 ≤ − V (x(t)).
λmax (M )
λmin (N )
− t
Therefore, we have V (x(t)) ≤ e λmax (M ) V (x(0)). Since λmin (M )kx(t)k2 ≤ V (x(t)) and V (x(0)) ≤
λmin (N )
2 , we have λ 2 ≤ e− λmax (M ) t λ 2
λ max (M )kx(0)k min (M )kx(t)k max (M )kx(0)k . This implies that kx(t)k ≤
λ (N )
q
λmax (M ) − 2λmin t
max (M ) kx(0)k.
λmin (M ) e
(P3) (3 points) Asymptotic stability implies marginal stability. Asymptotic stability is equivalent to exponential
stability.

Problem 4: (20 points) BIBO stability studies zero-state responses and asymptotic stability studies zero-input
responses. Input-to-output stability studies both responses. Consider a LTI system ẋ(t) = Ax(t) + Bu(t) and
y(t) = Cx(t) + Du(t). The system is IOS (input-to-output stable) if there exist positive scalar constants α, β
and γ such that ky(t)k ≤ max{βe−αt kx(0)k, γkuk[0,t] } for any initial state x(0) and any time t ≥ 0. Here,
kuk[0,t] = sups∈[0,t] ku(s)k, and the scalars α, β and γ are determined by A, B , C and D.
When C = I and D = 0, i.e., y(t) = x(t), IOS becomes input-to-state stability (ISS); i.e., kx(t)k ≤
max{βe−αt kx(0)k, γkuk[0,t] }.
(P1) If the LTI system is IOS, is the system BIBO stable? Justify your answer.
(P2) If the LTI system is ISS, is the system ẋ(t) = Ax(t) marginally stable? asymptotically stable? unstable? Justify
your answer.
(P3) It is known the following statement is true: the LTI system is IOS if all the eigenvalues of A have negative
real parts. This statement will be used in Problem 5. Now show a special case of the above statement: the LTI
system is IOS if all the eigenvalues of A have negative real parts and A is diagonalizable.
Hint for Part (P3): Start with the time-domain solution of the LTI system.
Solution: (P1) (5 points) Assume that x(0) = 0 and the input is bounded; i.e., supt≥0 ku(t)k ≤ α for some constant
α > 0. Since the system is IOS, then ky(t)k ≤ γkuk[0,t] ≤ γα for all t ≥ 0. So the output is bounded and the
system is IOS.
 (P2) (5 points) Since the LTI system is ISS, kx(t)k ≤ max{βe−αt kx(0)k, γkuk[0,t] }. For ẋ(t) = Ax(t), the input
is zero. Thus, kx(t)k ≤ βe−αt kx(0)k and the system is asymptotically stable.
(P3) (10 points) Since A is diagonalizable, there is a non-singular matrix P such that A = P −1 ΛP where Λ is
diagonal and includes all the eigenvalues of A. Then we have
Z t Z t
−1 Λt
At
kx(t)k = ke x(0) + e A(t−τ )
Bu(τ )dτ k = kP e P x(0) + P −1 eΛ(t−τ ) P Bu(τ )dτ k
0 0
Z t
≤ kP −1 kkeΛt kkP kkx(0)k + kP −1 kkeΛ(t−τ ) kkP Bkku(τ )kdτ
0
Z t
−1 λmax (A)t
≤ kP ke kP kkx(0)k + kP −1 kkeλmax (A)(t−τ ) kkP Bkku(τ )kdτ
0
−1 1
≤ kP kkP ke λmax (A)t
kx(0)k − kP −1 kkP Bkkuk[0,t] .
λmax (A)
As a result, we have
ky(t)k = kCx(t) + Du(t)k ≤ kCkkx(t)k + kDkku(t)k
≤ kCkkx(t)k + kDkkuk[0,t]
1
≤ kCkkP −1 kkP keλmax (A)t kx(0)k + (kDk − kCkkP −1 kkP Bk)kuk[0,t]
λmax (A)
1
≤ max{2kCkkP −1 kkP keλmax (A)t kx(0)k, 2(kDk − kCkkP −1 kkP Bk)kuk[0,t] }.
λmax (A)

Problem 5: (20 points) Consider the following linearization of the system in Problem 6 of HW3: ẋ(t) = Ax(t) +
Bu(t) and y(t) = Cx(t) where
−1.7 −2.13 × 10−4
   
0 0
A =  696 2.9 2.4  , B =  0  , C = [0 1 0],
0 6.5 −19.5 −0.16
(P1) Is the control-free system ẋ(t) = Ax(t) marginally stable? asymptotically stable? unstable?
(P2) Now let u = Kx where K = [36206, 274, 28]. Is the closed-loop system ẋ(t) = (A + BK)x(t) marginally
stable? asymptotically stable? unstable?
(P3) Now consider the controller in Part (P2) and assume that it is corrupted by disturbance d(t). The closed-loop
system becomes ẋ(t) = (A + BK)x(t) + Bd(t) and y(t) = Cx(t). Is the closed-loop system IOS? Use the
general statement in (P3) of Problem 4.
(P4) Consider the disturbed closed-loop system in Part (P3). Assume that d(t) = 20 sin t. Use Matlab to plot ky(t)k
versus t for three different initial states [0, 2, 0]T , [0, 5, 0]T , and [0, 10, 0]T . Discuss the behavior of ky(t)k
when t is small and t is large, respectively.
Solution: (P1) (4 points) The eigenvalues of A are λ1 = −1.6728, λ2 = 3.5486 and λ3 = −20.1758. Since λ2 has
positive real part, the system is unstable.
(P2) (4 points) The eigenvalues of A + BK are λ1 = −1.0344, λ2 = −1.6721 and λ3 = −20.0735. Since all
the eigenvalues have negative real parts, the system is asymptotically stable.
(P3) (6 points) Since all the eigenvalues of A+BK have negative real parts, it follows from the general statement
in (P3) of Problem 4 that the system is IOS.
(P4) (6 points) The figure shows ky(t)k for three different initial states [0, 2, 0]T , [0, 5, 0]T , and [0, 10, 0]T . Before
4s, the trajectories exponentially decrease. After 4s, the trajectories oscillate within [−1, 1].

Problem 6: (15 points) Consider a linear system with state transition matrix Φ(t, τ ) for which
e cos(2t) e−2t sin(2t)
 t 
Φ(t, 0) = .
−et sin(2t) e−2t cos(2t)
(P1) Compute the state transition matrix Φ(t, t0 ).
(P2) Compute a matrix A(t) which corresponds to the given state transition matrix.
(P3) Compute the eigenvalues of A(t).
(P4) Classify stability of ẋ(t) = A(t)x(t).
Solution: (P1) (5 points)
Φ(t, t0 ) = Φ(t, 0)Φ(0, t0 ) = Φ(t, 0)Φ(t0 , 0)−1
 t−t
e 0 cos(2t) cos(2t0 ) + e2(t−t0 ) sin(2t) sin(2t0 ) e2(t−t0 ) cos(2t0 ) sin(2t) − et−t0 cos(2t) sin(2t0 )

= .
e2(t−t0 ) cos(2t) sin(2t0 ) − et−t0 cos(2t0 ) sin(2t) e2(t−t0 ) cos(2t) cos(2t0 ) + et−t0 sin(2t) sin(2t0 )
d
(P2) (5 points) Since Φ(t0 , t0 ) = I and Φ(t, t0 ) = A(t)Φ(t, t0 ), thus
dt
 1 3
− 2 + 2 cos(4t) 2 − 32 sin(4t)

d
A(t) = Φ(t, t0 )|t=t0 = .
dt −2 − 32 sin(4t) − 12 − 32 cos(4t)
√
7
(P3) (3 points) The characteristic polynomial of A(t) is s2 + s + 2 and its eigenvalues are − 12 ± j 2 .
(P4) (2 points) The system is unstable since Φ(t, t0 ) goes unbounded.