Aa547 Lecture Lec12
Aa547 Lecture Lec12
Aa547 Lecture Lec12
Disclaimer: These notes have not been subjected to the usual scrutiny reserved for formal publications.
They may be distributed outside this class only with the permission of the Instructor.
12.1 Controllability
Let D = (U, X , Y, s, ρ) be a dynamical system. Let t0 < t1 . The input u[t0 ,t1 ] (·) steers x0 at t0 to x1 at t1 if
Definition 1.1 (Controllable) The system representation D is controllable on [t0 , t1 ] if for all (x0 , x1 ) ∈
X , there exists u[t0 ,t1 ] ∈ U which steers x0 at t0 to x1 at t1 .
+ u
v D y
−
Fs (·) x
+ u
v D y
−
Fo (·)
12-1
12-2 Lecture 12: Controllability (Reachability) and Observability
Applying Fs and Fo to our dynamical system D = (U, X , Y, s, r) (where s: state transition map, r: read
out map ) we get resulting systems Ds and Do , respectively, which are depicted in Figures 12.1 and 12.2,
respectively.
Assumption. (well-posedness) For all x0 , t0 , for all v(·), ∃! x(·) and y(·) for Ds and Do .
D is controllable on [t0 , t1 ]
⇐⇒ Ds is controllable on [t0 , t1 ] (1)
⇐⇒ Do is controllable on [t0 , t1 ] (2)
Proof. (1) (=⇒) Since D is controllable on [t0 , t1 ], given x0 and x1 ∈ X , ∃ũ[t0 ,t1 ] (·) steering x0 at t0 to x1 at
t1 . Define
ṽ(t) = ũ(t) + Fs (x(t))
= ũ(t) + Fs (s(t, t0 , x0 , ũ[t0 ,t] ))
Then ṽ(t) steers x0 at t0 to x1 at t1 by the well-posedness assumption.
(=⇒) for all x0 , x1 ∈ X , ∃ ṽ[t0 ,t1 ] that steers (x0 , t0 ) to (x1 , t1 ) on Ds . Consider ũ(t) generated by Ds : ũ(t)
steers x0 at t0 to x1 at t1 .
Key Take-Away. Roughly speaking, nonlinear memoryless state-feedback and output-feedback do not
affect controllability. The memeoryless assumption is crucial since it allows for us to use the same state
space for D, Ds , Do .
12.2 Observability
Definition 2.1 (Observable) The dynamical system D is called observable on [t0 , t1 ] if and only if,
given, D, for all inputs u[t0 ,t1 ] and for all corresponding outputs y[t0 ,t1 ] ∈ Y the state x0 at time t0 is
uniquely determined.
Of course, once x0 is calculated, from u[t0 ,t1 ] and the state transition map s we can calculate the state
trajectory x(·); indeed,
x(t) = s(t, t0 , x0 , u[t0 ,t1 ] ), ∀ t ∈ [t0 , t1 ]
Lecture 12: Controllability (Reachability) and Observability 12-3
Fact. D is observable on [t0 , t1 ] ⇐⇒ for each fixed u[t0 ,t1 ] the partial response map
is injective, that is the partial response map is a one to one map from X to Y.
Then for Df ,
η(t) = y(t) + Ff (u(t))
For Do , we make the uniqueness (well-posedness) Assumption 12.1.1.
+ u
v D y
−
Fo (·)
+
v D y
+
Ff (·)
Theorem 2.1 For the system Df and the system Do satisfying Assumption 12.1.1, we have
D is observable on [t0 , t1 ]
⇐⇒ Do is observable on [t0 , t1 ] (1)
⇐⇒ Df is observable on [t0 , t1 ] (2)
Remark. Memoryless state feedback may affect observability. For example, for a linear time-invariant
system representation R = [A, B, C, D], there may exist a linear state feedback Fs such that for some
states x0 and for some inputs u(·), the state trajectory remains, for all t, in the nullspace of C.
The reaching from the origin condition gives rise to the so-called reachability map.
Theorem 3.1 R = [A(·), B(·), C(·), D(·)] is completely controllable (CC) on [t0 , t1 ]
Proof.
x1 = Φ(t1 , t0 )x0 + Lr u
since x0 can be zero and x1 arbitrary, a necessary condition for CC is that R(Lr ) = Rn . But this is sufficient
too because if R(Lr ) = Rn , given x1 at t1 and x0 at t0 , ∃ u[t0 ,t1 ] such that
x1 − Φ(t1 , t0 )x0 = Lr u
Note: Lc surjective ⇐⇒ ∃ u[t0 ,t1 ] that steers arbitrary (x0 , t0 ) to arbitrary (x1 , t1 ).
Proof. If N (Lo ) = {0} then given y(·) and u(·), ∃! x0 such that
Z t
Lo x0 = y(t) − C(t)Φ(t, τ )B(τ )u(τ ) dτ − D(t)u(t)
t0
Thus, the key to giving constructive tests for controllability and observability is to give conditions under
which R(Lc ) = Rn and N (Lo ) = {0}.
12.4 Gramians
Remark. This is where the Finite Rank Operator Lemma (FROL) and adjoint map definition will play
a role.
R(A∗ ) R(A)
A∗
0H 0m
N (A) N (A∗ )
H Fm
Figure 12.5: The orthogonal decomposition of the domain and the co-domain of a finite rank operator
A : H → F m and its associated bijections.
and
⊥
U = R(A∗ ) ⊕ N (A)
Moreover,
N (AA∗ ) = N (A∗ )
∗
R(AA ) = R(A)
∗
N (A A) = N (A)
∗
R(A A) = R(A∗ )
and
AA∗ |R(A) → R(A) is one-to-one and onto
A∗ A|R(A∗ ) → R(A∗ ) is one-to-one and onto
12-6 Lecture 12: Controllability (Reachability) and Observability
is
L∗r x = B ∗ (·)Φ∗ (t1 , ·)x
R t1
Indeed, Lr : u[t0 ,t1 ] 7→ t0
Φ(t1 , τ )B(τ )u(τ )dτ so that
where
Z t1
hz, Lc uiRn = z∗ Φ(t1 , τ )B(τ )u(τ )dτ
t0
Z t1
= (B ∗ (τ )Φ∗ (t1 , τ )z)∗ u(τ )dτ
t0
= hB (τ )Φ∗ (t1 , τ )z, uiPC([t0 ,t1 ])
∗
so that
L∗r z = B ∗ (·)Φ∗ (t1 , ·)z
The adjoint L∗c is derived similarly:
L∗c z = B ∗ (·)Φ∗ (t0 , ·)z
Note that
R(Lr ) = Φ(t1 , t0 )R(Lc )
Indeed, let x ∈ Lr then
Z t1
x = Lr u = Φ(t1 , τ )B(τ )u(τ ) dτ
t0
Proposition 4.1
CC on [t0 , t1 ] ⇐⇒ R(Lc ) = Rn
⇐⇒ R(Lc L∗c ) = Rn
Definition 4.1 (Reachability and Controllability Gramian) The rechability Gramian is given by
the symmetric positive semi-definite matrix
Z t1
Wr [t0 , t1 ] = Φ(t1 , τ )B(τ )B ∗ (τ )Φ(t1 , τ ) dτ ∈ Rn×n
t0
Lecture 12: Controllability (Reachability) and Observability 12-7
CC ⇐⇒ rank(Wc ) = n
Lo x0 = C(·)Φ(·, t0 )x0
and Z t1
L∗o y(·) = Φ∗ (τ, t0 )C ∗ (τ )y(τ ) dτ
t0
and Z t1
L∗o Lo = Φ(τ, t0 )C ∗ (τ )C(τ )Φ(τ, t0 ) dτ
t0
Proposition 4.2
CO ⇐⇒ rank(W0 ) = n
The notions of controllability (and observability) described in the preceding sections really focus on a black
and white question of whether a system satisfies a property, but it is more interesting to consider how to
divide the state, input, and output spaces into subspaces that have certain properties. This helps with the
design process (for example, designing a controller that targets particular states).
Consider
An input u(·) transfers the state x(t0 ) = x0 at time t0 to the state x(t1 ) = x1 at time t1 such that
Z t1
x1 = Φ(t1 , t0 )x0 + Φ(t1 , τ )B(τ )u(τ ) dτ
t0
12-8 Lecture 12: Controllability (Reachability) and Observability
Definition 5.1 (Reachable Subspace) Given two times t1 > t0 ≥ 0, the reachable or controllable-
from-the-origin on [t0 , t1 ] subspace R[t0 , t1 ] consists of all states x1 for which there exists an input u :
[t0 , t1 ] → Rm that transfers the state from x(t0 ) = 0 to x(t1 ) = x1 —i.e.
Z t1
R[t0 , t1 ] = x1 ∈ Rn | ∃u(·), x1 = Φ(t1 , τ )B(τ )u(τ ) dτ
t0
Definition 5.2 (Controllable Subspace) Given two times t1 > t0 ≥ 0, the controllable or controllable-
to-the-origin on [t0 , t1 ] subspace C[t0 , t1 ] consists of all the states x0 for which there exists an input
u : [t0 , t1 ] → Rm that transfers the state from x(t0 ) = x0 to x(t1 ) = 0—i.e.
Z t1
n
C[t0 , t1 ] = x0 ∈ R | ∃u(t), 0 = Φ(t1 , t0 )x0 + Φ(t1 , τ )B(τ )u(τ ) dτ
t0
The matrices C(·) and D(·) play no role in these definitions; therefore, one often simply talks about the
reachable or controllable subspaces of the system
ẋ = A(t)x + B(t)u
Remark. Recall the notion of surjectivity. We will see controllability (reachability) relates to surjectiv-
ity.
Determining the reachable subspace amounts to finding for which vectors x1 ∈ Rn the equation
Z t1
x1 = Φ(t1 , τ )B(τ )u(τ ) dτ
t0
has a solution u(·). Similarily, determining the controllable subspace amounts to finding for which vectors
x1 ∈ Rn the equation Z t1
0 = Φ(t0 , t1 )x0 + Φ(t1 , τ )B(τ )u(τ ) dτ (12.3)
t0
hence finding the controllable subspace is the same as characterizing for which x1 the above has a solution
v(·) ≡ −u(·).
Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.
Alternative Proxies: