HW 9, Fall 2021

 
A B
1. For a given LTI system (A ∈ Rn×n and D ∈ Rp×m ), we can say that any state z
C D
that can be reached from initial condition x(0) = 0 at a fixed finite time tf > 0 is a controllable
(or reachable) state (That is z is controllable if there exists a control u(τ ) such that x(tf ) = z
for a given tf and when x(0) = 0). We also can say that

(a) Show that the set of all the controllable (reachable) states is a subspace (called the
controllable subspace).
(b) If a state d ∈ Rn is controllable (reachable), then determine a control strategy that will
result in x(tf ) = d with x(0) = 0.
 
A B
2. Consider a system described by the state-space representation given by .
C 0

(a) Let z be a left eigenvector of A (that is z T A = λz T ) such that z T λI − A B = 0.

Prove that the state xf = z is unreachable that is, there exists no control u(τ ) which
will transfer x(0) = 0 to x(tf ) at tf . [Hint: write the solution equation x(tf ) = z and
show that z T (x(tf )) = z T z = 0. You may require to recall that z is a left eigenvector of
A with eigenvalue λ implies that z is also a left eigenvector of eA(tf −τ ) with eigenvalue
eλ(tf −τ ) .]

3. Consider an LTI system given by

ẋ = Ax + Bu
y = Cx,

where x ∈ R4 and
   
−1 4 0 0 2 0  
0 1 0 0 1 0 0 1 0 0
A= ,B =   , and C = .
0 0 −1 0  0 1 −1/3 −1/3 0 1
0 2 0 −1 1 0

You can use Matlab to answer the following questions.

(a) Find the Controllability matrix for this system and show that the system is not control-
lable. Determine the rank n1 of the controllability matrix.
 
Āc B̄c
(b) Determine a realization where Āc has dimension n1 × n1 such that G(s) =
C̄c D̄c
C(sI − A)−1 B + D = C̄c (sI − Āc )−1 B̄c + D̄c . Hint: Construct Q with the first n1 columns
formed by choosing n1 independent columns of the controllability matrix and let x = Qw
be a transformation.
(c) Determine a matrix X1 such that the reachable (controllable) subspace is the Range(X1 )
with columns of X1 independent.
(d) Determine all the eigenvalues of A and associated right eigenvectors.
 (e) For every left eigenvector and eigenvalue pair (λ, z) of A determine if the state z is
reachable or unreachable (uncontrollable). Hint: test if z T [λI − A B] = 0. Caution:
Note an eigenvalue can have more than one eigenvector.
 
A B
4. For a given LTI system (A ∈ Rn×n and D ∈ Rp×m ), a state v is unobservable if
C D
there exists a initial state x0 such that the output y(t) is identically zero (when the solution
evolves from this initial condition) when given that the input is identically zero.
(a) Show that the set of all the unobservable states is a subspace (called the unobservable
subspace).
(b) If a initial condition can be determined from y = Cx + Du then determine the initial
condition.
 
A B
5. Consider a system described by the state-space representation given by .
C 0
 
µI − A
(a) Let v be a right eigenvector of A (that is Av = µv) such that v = 0. Prove that
C
the state v is unobservable that is, the output y(t) from the initial condition x(0) = v is
not distinguishable from the output from the initial condition x(0) = 0.
6. Consider an LTI system given by
ẋ = Ax + Bu
y = Cx,
where x ∈ R4 and
   
−1 4 0 0 2 0  
0 1 0 0 1 0 0 1 0 0
A= ,B =   , and C = .
0 0 −1 0  0 1 −1/3 −1/3 0 1
0 2 0 −1 1 0
You can use Matlab to answer the following questions.
(a) Find the Observability matrix for this system and show that the system is not observable.
Determine the rank n2 of the observability matrix.
 
Āo B̄o
(b) Determine a realization where Āo has dimension n2 × n2 such that G(s) =
C̄o D̄o
C(sI − A)−1 B + D = C̄o (sI − Āo )−1 B̄o + D̄o . Hint: Construct the last n − n2 columns of
a invertible Q that form a basis for null space of the observability matrix. Let x = Qw
be a transformation.
(c) Determine a matrix X2 such that the unobservable is the Range(X2 ) with columns of
X2 independent.
(d) Determine all the eigenvalues of A and associated right eigenvectors.
(e) For every right eigenvector and eigenvalue
 pair (µ, z) of A determine if the state z is
µI − A
unobservable. Hint: test if v = 0. Caution: Note an eigenvalue can have more
C
than one eigenvector.