Resumen,Sistemas Lineales

Kevin Lopez Preciado

10 de marzo de 2015

I.- Introduction

(t2 , t1 , (t0 , t0 , x0 , u), u).

d) Smoothness: The functions and r are
continuous functions of t,t t0 .
The most prominent example of a dynamical
system in state-space form is the finitedimensional linear system(FDLS), described
by the ordinary differential equation:

The most innovative aspect of modern system theory is undoubtedly the prevalence of
state-space models for dynamical systems.
The paper introduce the concepts of controllability and observability for linear systems formally.
They first prove that for such systems it is possible to locate the closed-loop poles using state
feedback if and only if the system is controllable, then show that is is possible to build a
state reconstructor with arbitrary error dynamics provided the system is observable.
And then consider qualitative aspects of inputoutput descriptions versus state-space descriptions and find that minimality of the state space is equivalent to controllability (reachability)
and observability.

x = A(t)x + B(t)u and y = C(t)x

With X=Rn , U = Rm and Y = Rp
The solution at the ordinary differential
equation is:
x(t) = (t, t0 )x(t0 +

Rt
t0

(t, )B( )u( )d

Where the transition matrix (t, ) is defined as the solution of the matrix differential
equation (t, ) = A(t)(t, ), (, ) = I. The
transition matrix satisfies the composition
law (t2 , tl )(tl , t0 ) = (t2 , to). Of particular
importance in practice are the so-called
II.-Dynamical Systems
stationary dynamical systems in which 4 and
The Dynamical systems are defined here:
y commute with the shift operator. Stationary
Let U, Y, and X be normed linear spaces, linear finite-dimensional systems (SFDLS) are
and let U,Y denote the space of continuous described by the differential equation:
functions defined on R = (- a, a) with values
in U, Y, respectively.
x = Ax + Bu and y = Cx which corresU is termed the input space, Y the output ponds to a system with transfer function
space, U the set of input values, Y the set of matrix G(s) = C(Is A)l B, where s is a
output values, and X the state space.
complex variable.
Let t0 R and let = t R|t t0 .Consider
the maps : R X U X and
III.-Some fundamental properties of
r : R X U Y termed the state evolution
State-Space Models.
map and read-out map, respectively.
The concept of controllability refers to
Definition 1: A dynamical system transferring an arbitrary initial state to a
is
a
quintuple
{U, Y, X, , r}
satisf- desired trajectory. This desired trajectory is
ying the following axioms, for all often an equilibrium point. Assuming this
to be the case and take the zero element, to
u1 , u2 U, x0 X, t0 , t1 , t2 R, t0 t1 t2 .
a)
Causality:
(tl , t0 , x0 , ul )
= represent this equilibrium.
(tl , t0 , x0 , u2 )wheneveru1 (t) = u2 (t)f ort0 Let t0 be an element of the real line. The
state space of a dynamical system is said to
t t1 .
be reachable a t0 if, given any x X,there
b ) Consistency:(t0 , t0 , x0 , u) = xO .
c) Semi-group property: (t2 , t0 , x0 , u) = exists a t1 t0 and a u U such that
1

and controllability and that observability

implies zero-input observability, which in turn
implies irreducibility.
It should also be noted that reachability,
controllability, connectedness, observability,
irreducibility, and reconstructibility are preserved under (output) feedback, but that
zero-input observabi1ity and zero-input reconstructibility are in general not unless the
system is again linear and finite dimensional.
Definition 5: A dynamical system is said to be
uniformly controllable if there exists a T > 0
and a continuous function : R R such
that for all xR0 X and t0 R there exists a
t +T
u U with t00
k u(t) k u2 dt (k x0 k)
such that (t0 + T, t0 , x0 , u) = 0
The states pace of a dynamical system
is said to be uniformly reachable if there
exists a T > 0 and a continuous function
: R R such that for all x0 X
and
R t0 t0 R there exists a u U with
t0 T parallelu(t) ku2 dt (k x0 k) such that
(t0 , t0 T, 0, u) = x0 .
A dynamical system is said to be uniformly
zero-input observable if there exists a T > 0
and a monotone increasing function : R R
with (0) = 0 such that for all x0 , x1 X and
t0 R

x = (t0 , t1 , 0, u). A dynamical system is

said to be controllable at t0 if, given any
x0 X,there exists a t1 t0 and a u U such
that (tl , t0 , xo, u) = 0. The state space of a
dynamical system is said t0 be connected if,
given any x0 , x1 X, there exist t0 tl and a
u U such that (t1 , t0 , xo, u) = x1 .

There are two separate state reconstruction problems. One refers to deducing the
present state from past output observations
and the other refers to deducing the present
state from future output observations.
Definition 3: Let t0 R. A dynamical system
is said to be zero input observable at t0 if
knowledge of the output y(t0 , xo, 0) for t t0
uniquely determines x0 . A dynamical system
is said to be observable at t0 if for all x X
and u U knowledge of the observed output
y(t0 , x0 , u) for t t0 determines x0 uniquely.
The state space of a dynamical system is
said to be irreducible at to if for all x0 X
there exists a u U such that knowledge
of the output y(t0 , x0 , u) for t t0 uniquely
determines x0 .
The difference between irreducibility and
observability is that in the former case the
u which yields the initial state x0 may be a
function of x0 itself whereas in the latter case R
t0 +T
k y(t, (t, t0 , x0 , 0), 0)
any u will do.
t0
y(t, (t, t0 , x1 , 0), 0) k2 dt (k x0 x1 k).
Definition 4: Let to be an element of the
real line. Then the state of a dynamical sys- A dynamical system is said to be unitem is said to be zero input reconstructible at formly zero-input recmlstructtle if there
t0 if knowledge of the output corresponding to exist.s a T > 0 and a monotone non-increasing
an input u = 0 for t t0 uniquely determines function : R R with (0) = 0 such that
x0 X. Thus the output due to some initial for all x0 , x1 R and t0 R
state at t1 = nf is observed and the
R
present state x0 is to be reconstructed. The tt0T k y(t, (t, t0 T, x0 , 0), 0)
0
state of dynamical system is said to be re- y(t, (t, t0 T, x1 , 0), 0) k2 dt
constructible at to if for all u U, knowledge (k (t0 , t0 T, x0 , 0 (t0 , t0 T, x, 0) k).
of the output for t I to uniquely determines
x0 X.
Definition 6: A dynamical system in stateNote that connectedness implies reachability
2

space form is said to be exponentially stable if u(t) satisfies

there exist, constants M, > 0 such that
Rt
y(t) t0 C(t)(t, )u( )d = C(t)(t, t0 )x0 ,
k (t1 , t0 , x0 , 0) k M e(t1 t0 ) k x0 k Hx0
where t0 t t1 .
for all x0 X and t1 t0 .
It is immediately clear from the above that
the finite dimensional linear system (FDLS) is
controllable at t0 in some finite time interval
IV.- Finite-Dimensional Linear Sys- [t0 , t1 ] ifm and only ifn the linear noperator
tems: Algebraic Conditions for Con- F : C2 (t0 , t1 ) R is onto R and is
observable in the interval [t0 , t1 ] if and only
trollability and Observability
if the linear operator H : Rn C2p (t0 , t1 ) is
Definition 7: Let V be an inner product one-one on Rn .Consequently controllabi1ity
space and let S be a subspace of V. Then on [t0 , t1 ] requires that, R(F ) = Rn and obthe orthogonal complement of X, denoted servability on [t0 , t1 ] requires that M(H) = (0).
by S , is defined as S = v V |hv, si = 0
for all L
s S. If S is a closed subspace then Theorem 1: The finite-dimensional linear
system (FDIS)is controllable at t0 if and
V =S
S.
Let L be a linear operator from V1 into V2 only if detW (t0 , t1 ) 6= 0 for some t1 t0 ,
with V1 and V2 inner product spaces. Then its state space is reachable at t0 if and only
the null space M(L) and the range space R(L) if detW (t1 , t0 ) 6= 0 for some t1 t0 ,
of L are the subspaces of V1 and V2 defined and its state space is connected if and only
if det W (t1 , tl ) 6= 0 for some tl , tl . It is
respectively by
observable (zero-input observable, irreducible)
at t0 if and only if det M (t0 , tl ) 6= 0 for
M(L) = {v1 V1 |Lv1 = 0}
some t1 t0 and its state is reconstructible
R(L) = {v2 V2 |v2 = Lv2 , some v1 V1 } (zero-input, reconstructible) at t0 if and only
if det M (t1 , t0 ) 6= 0 for some t1 t0 .
Lemna 1 : Let L be a linear operator from V1
Rt
into V2 with V1 and/or V2 finite dimensional. W (t0 , t1 ) , t01 (t0 , )B( )B 0 ( )0 (t0 , )d
Then L exists,M(L) = R(L ) , R(L) =
Rt
M(L ) , M(L) = M(L L) and R(L) = M (t0 , t1 ) , t01 0 (, t0 )C 0 ( )C( )(, t0 )d
R(LL )
Theorem 2: The finite- dimensional linear
Turning to the question of controllability system (FDLS)is uniformly controllable (uniand observability,for the finite-dimensional formly reachable) if and only if for some
linear system (FDLS) that the state x0 Rn T > 0 there exists an 1 > 0 such that
at t0 will be transferred to the state x1 Rn W (t0 , t0 + T ) 1 I for all t0 R;it is uniat tl by the continuous control u(t) if and only formly observable (uniformly reconstructible)
if and only if for some T > 0 there exists an
if:
2 > 0 such that M (t0 , t0 + T ) 2 I for all
R t1
(t0 , t1 )x1 x0 = t0 (t0 , )B( )u( )d , F u t0 R.
Similarly it follows that the output. on [t0 , t1 ]
due to the initial state x at t0 and the control
3

turns out to be equivalent to controllability.

In the multi-input case the first results in
this direction are due to Langenhop and
Popov. They proved that, given an arbitrary
polynomial r(s) with coefficients in the field
of real or complex numbers, there exists a
matrix K (possibly complex) such that det(Is
- A +BK) = r( s) if and only if the system is
controllable.

Theorem 3: The stationary finite-dimensional

linear system (SFDLS) is controllable (reachable, connected) at t0 if and only if rank
{B, AB, A2 B, ..., An1 B} = n. It is observable
(reconstructible, irreducible) at t0 if and only
if rank {C 0 , A0 C 0 , ..., (A0 )n1 C 0 } = n.
Silverman, Meadows and Chang have developed algebraic conditions which are applicable
to linear time-varying dynamical systems.
These results require that the matrices A(t),
B(t), and C(t) be sufficiently smooth.

Theorem 4: There exists a real m n matrix L

such that det(Is-A-BL)=sn +rn1 sn1 +...+r0
V.-Pole Allocation State Reconstruc- for arbitrary real coefficients {r0 , rl , .., rnl }
if and only if the system x = Ax + Bu is
tion and Close-Loop Regulation
controllable; if and only if the (nmXn) matrix
The problems of regulation and state re- [B : AB : : An1 B] is of rank n.
construction discussed above will be solvedn
the same class of dynamical systems, i.e., we Theorem 5: There exists an (nXp) mawill only use stationary finite-dimensional trix H such that the error dynamics of
linear systems to achieve a possible design the state reconstructor are governed by
e = (A HC) where e = x
x with det(Is - A
procedure.
Consider first the question of regulation under + HC) =an + rnl snl + .. + rOfor arbitrary
state feedback and assume that the feedback real coefficients {r0 , r1 , ..., rnl if and only
-Kx is being applied to the system. The if the system x = Ax + Bu, y = Cx is
closed-loop response is then governed by the reconstructible (observable); i.e., if and only if
the (npXn) matrix [C 0 : A0 C 0 : .. : (A0 )n1 C 0 ]
dynamical equations:
is of full rank n.
x = (A BK)x + Bu, y = Cx.

VI.- Stabilizability and State Reconstructibility for Time-Varyng Systems

A representative feature of the response

of this closed-loop system is given by the
location of its poles, i.e., by the zeros of
det(Is-A+BK).The question thus arises of
under what conditions is it possible to assign
the poles of a dynamical system arbitrarily by
suitably choosing the feedback gain matrix K.
More precisely, given an arbitrary polynomial:

The control:
u(t) = B 0 (t)(t0 , t)W 1 (t0 , t1 )x0 ,
t t1

t0

will transfer the state of a controllable linear finitedimensional system from state x0 at
n
n1
time t0 to state 0 at time tl . Implemented in a
r(s) = s + rn1 s
+ ... + r0
feedback form, this would lead to the feedback
with real coefficients, when does there control law:
exist, for given matrices A and B, at least one
(mXn) matrix K such that det (Is - A + B K u = B 0 (t)W l (t, t1 )x
) = r ( s )? The possibility of pole assignment
4

which, since limtt1 k W 1 (t, t1 ) k= , A much more interesting realization is the minimal realization for which the state-space X
calls for unbounded feedback gains.
has as few elements as possible.This realizaTheorem 6: Assume that the system tion is always reachable and irreducible.
(FDLS) is uniformly controllable. Then Difficulty with time varying systems arises
there exists a bounded continuous ma- from the fact that usually the above equivalentrix G(t) such that the feedback system ce class idea will result in a state space which
x = A(t)x + B(t)u; u = G(t)x is exponen- is itself time varying.
tially stable.
Theorem 7: Assume that the system (FDLS)
is uniformly observable. Then there exists a
bounded continuous matrix H(t) such that the VIII.- Minimal Realizations of Linear
Systems
system:
x = A(t)
x + B(t)u H(t)(
y y)
It is clear that the input-output
With y = C(t)
x is a state reconstructor such
that the error equation e = (A(t) H(t)C(t))e system (LS) is realized by the statespace model (FDLS) if and only if
is exponentially stable.
Theorem 8: Assume that the system (FDLS) w(t, ) = C(t)(t, )B( ), f or all t t0
is uniformly controllable and uniformly obser- Furthermore the system (LS) has a
linear
realization
vable. Then there exist bounded continuous finite-dimensional
matrices G(t) and H(t) such that the closed- (FDLS) if and only if there exist continuous matrices G and H such that
loop system
w(t, ) = H(t)G( ), f or all t .
Definition : Assume that the state-space model
x = A(t)x + B(t)u
(FDLS) is a realization of the input-output
system (LS). Then it is said to be a minimal
y = C(t)x
realization of (IS) if every other realization of
the type (FDLS) has a state space of greater
x
= A(t)
x B(t)u H(t)(y y)
equal dimension.
Theorem 9: The linear system (LS) is reay = C(t)
x
lizable by a state-space model (SFDLS) if
and only if: 1)the kernel w(t, ) is separau = G(t)
x is exponentially stable.
ble, i.e., there exist matrices H and G such
Input-Output Descriptions of Dyna- that H(t)G( ) = w(t, ) for t ; and 2)
w(t, ) = w(t , 0)

mical Systems

Theorem 1O: A realization A, B, C is minimal

if and only if the system x = Ax + Bu; y = Cx
is observable and has a reachable state space.
Moreover a minimal realization always exists
and any two minimal realizations A1, B1, C1
and A2, B2, C2 are related via the similarity
transformation A2 =T AT 1 , B2 = TBI, C2
= CIT l for some invertible matrix T.

There are two possible descriptions of dynamical systems. One is the so-called state-space
description.The other is the input-output description.
The problem of realization is to construct a dynamical system in state-space form such that it
generates the same input-out,put pairs as the
dynamical system in input-output form.
5

IX. Stability
Definition: Let G be a dynamical system in
input-output, form and assume that Go = 0.
Then it is said to be Lp input output stable,
1 p nf, if there exists a constant K < nf
such that all u U, u Lp , yield Gu Lp ,
and k Gu kL p K k u kLp .
Theorem 11: Assume that the system
(FDLS) is uniformly reachable and uniformly
observable. Then exponential stability and
Lp -input-output stability ,1 p nf, are
equivalent.