Sie sind auf Seite 1von 2

Linear controllability

November 10, 2014


Controllability notion refers to state transfer in the state-space. In a linear system, this
property assures arbitrary pole-placement using linear state-feedback control law. We begin
this brief write-up with the definition of controllability for a linear system.

Definition 0.1 Let


x = Ax + bu

(1)

be a SISO system, where x IRn . The pair (A, b) is completely controllable if and only if for
each x1 , x2 IRn and for each T 0, there exists an admissible input u : [0, T ] IR with
the property that if x(t) is the solution of (1) with x(0) = x1 , then x(T ) = x2 .
We next present the result by Kalman, Ho and Narendra (1963) that provides a method to
verify the linear controllability property for a given pair (A, b).
Theorem 0.1 The continuous-time system (1) is completely controllable if and only if the
controllability matrix

C(A, b) = [b Ab An1 b]
has rank n.
Proof: For t > 0 define P (t) =

Rt
0

eA bb eA d , P (t) is termed the controllability gram-

mian. We first show that C(A, b) is invertible P (t) is positive definite for all t > 0. Since
Rt
Rt

P (t) is positive semi-definite (x P (t)x = 0 x eA bb eA xd = 0 (x eA b)(x eA b) d =


R t A 2
|x e b| d 0) we show that C(A, b) is invertible P (t) is invertible for all t > 0.
0

Let C(A, b) be not invertible. Then there exists a non-zero vector v IRn such that
v C(A, b) = 0 = v [b Ab An1 b] = 0. Now by Cayley-Hamilton theorem
An = 1 I + 2 A + + n An1
1

Note that the notion of controllability is not tied to the nature of the system (linear/nonlinear).

where, the i s are not identically equal to zero. On pre and post multiplying the above by
v and b respectively, we have

v An b = 1 v b + 2 v Ab + + n v An1 b

or v Ak b = 0 k 0 which can be rewritten as z(t) = v eAt b = 0 t 0. Thus


Rt

v eAt bb eA t v = 0 t [0, T ] or 0 v (eA bb eA )vd = 0, thereby proving P (t) is not

invertible.

Conversely, let P (t) be not invertible. Then there exists a non-zero vector p IRn such
that
p eAt b = 0, t [0, T ].

(2)

Differentiating (2) with respect to time t, (n 1) times and then evaluating at t = 0, we


get p Ab = 0, p A2 b = 0, . . . , p An1 b = 0. Rearranging, we have p [b Ab An1 b] = 0.
Hence the claim that C(A, b) is not invertible. It now follows that C(A, b) is invertible if
and only if P (t) is invertible.
The controllability property is exhibited by constructing an u which steers x1 to x2 . One
such control is
u(t) = b eA

(T t)

P (T )1 (eAT x1 x2 )

that results in the solution of (1) satisfying x(T ) = x2 .


Finally, suppose that C(A, b) is not invertible, which implies that P (t) is not invertible.
Thus there exists a non-zero vector q IRn such that q eAt b = 0 for t [0, T ]. Let

x1 = eAT q, then
Z

x1 +
eA(T ) bu( )d
0
Z T
q x(T ) = q q
q eA(T ) bu )d.
AT

x(T ) = e

Since the second term in the above is zero, we have q (x(T ) q) = 0 which further implies
that x(t) cannot be steered to the origin in time T .
2