Control and
Information Sciences
Edited by M.Thoma
72
A. Isidori
Springer-Verlag
Berlin Heidelberg New York Tokyo
Series Editor
M. Thoma
Advisory Board
A.V. Balakrishnan L. D. Davisson A. G. J. MacFarlane
H. Kwakernaak .1. L. Massey Ya Z. Tsypkin A. J. Viterbi
Author
Prof. Alberto Isidori
Dipartimento di Informatica e Sistemistica
Universit& di Roma ,,La Sapienza-
18 Via Eudossiana
00184 Rome (Italy)
I. Introduction .......................................... I
2. Distributions on a Manifold ........................... 5
3. Frobenius Theorem ..................................... 12
4. Invariant Distributions ............................... 24
5. Local Decompositions of Control Systems ............... 28
6. Local Reachability .................................... 33
7. Local Observability ................................ ... 47
I. I n t r o d u c t i o n
y = Cx
A11 A12
0 A22
whose elements o n the lower (n-d) rows and left d columns are vanishing.
Moreover, if t h i s subspace V is such that:
0
2
~2 = A 2 2 x 2
f
T
x I (T) = e x p ( A 1 1 T ) x I (0) + ~ e x p ( A 1 1 (T-T))AI2exp(A22T)dTx2(0) +
0
t
T
+ i e x p (At I ( T - T ) ) B l U (T) dT
0
x 2(T) = e x p ( A 2 2 T ) x 2(0)
x(T) = x(T) + v
S T = x(T) + V
r a n k ( B I A 1 1 B I ... A~;IBI) = d
x2 = A 2 2 x 2 + B2u
y = C2x 2
or, in o t h e r words,
C22exp(A22t)x 2 - 0 ~ x 2 = 0
Then, if two initial states are such that their difference does
not belong to W, t h e y m a y be d i s t i n g u i s h e d from e~ch other by the out-
put produced under zero input.
Again we m a y synthesize the above discussion with the following
considerations. Given a linear control system, let ~ be the largest
subspace of X s a t i s f y i n g (i) and (ii). Associated with W there is a
partition of X i n t o s u b s e t s of the f o r m
x+W
with the p r o p e r t y that each one of these subsets coincides with the
set of p o i n t s that are indistinguishable from a fixed point Qf X. Mo-
reover, these subsets have the structure ~f ~ d-dimensional flat sub-
manifold of X.
In the following sections of this chapter and in the following
chapter we shall deduce similar decompositions for nonlinear control
systems.
2. D i s t r i b u t i o n s on a Manifold
A(q) = span{Ti(q):i E I}
A I M A 2 : p ~ AI (p) ~ A2(P)
(*) In order to avoid confusions, we ,use the symbol span{-} to denote any B-linear
combination of elements of some ~-vector space (in particular, tangent vectors
at a point). The symbol sp{'} is used to denote a distribution (or a codistribu-
tion, see later).
6
= sp{_ 8 + --~ }
AI ~x I 8x 2
A 2 = sp{(1+Xl)~1 + ~x 2
Then we have
(A I N A2 ) (x) = {0} if xI ~ 0
(2.2) Remark. Recall that the s e t V(N) of a l l smooth vector fields de-
fined on N may be g i v e n the structure of a v e c t o r space over ~ and,
also, the structure of a m o d u l e over C~(N), the r i n g o f all smooth
real-valued functions defined o n N. The set M(A) defined before (which
is n o n - e m p t y because the zero element of V(N) belongs t o M(A) for any
4) is a s u b s p a c e of the v e c t o r space V(N) and a submodule of the m o d u l e
V(N). From this is it e a s i l y seen that the span of M(4) is c o n t a i n e d
in 4. []
smt(A) =4 sp M(A)
(2.3) Remark. N o t e that M(A) may not be the unique subspace of V(N),
or s u b m o d u l e of V(N), whose span coincides with smt(4). But if M' is
any o t h e r subspace of V(N), or submodule of V(N), with the property
that sp M' = smt(&), then M' c M(~).
4 = sp{x ~}
Then M(4) is the set of all vector fields of the form c(X)~x where
c(x) is a s m o o t h function defined on ~ which vanishes at x = 0 . C l e a r l y
is s m o o t h and coincides with smt(A). There are many submodules of
V(~) which span 4, for i n s t a n c e
M2 = {T 6 V ~ ) : T ( x ) = c ( x ) x 2 ~-~
~ and c 6 C~(~)}
8(p) 6 span{T(p):T 6 M}
8
then 8 belongs to M. []
d
(2.8) = ~ ci~ i
i=I
(2.10) Lemma. Let A 1 and A 2 be two smooth distributions with the pro-
perty that A 2 is nonsingular and At (p) C A2(p) at each point p of a
dense submanifold o f N. Then A I C A2"
(2.11) Lemma. Let A I and A 2 be two smooth distributions with the pro-
perty that A I is nonsingular, A I C A 2 and At(p) = A2(p) at each point
p of a dense submanifold of N. Then A 1 = A 2. []
We h a v e seen before that the intersection of t w o smooth distribu-
tions may fail to be smooth. However, around a regular point this
cannot happen, as w e see from the following result.
ql ,5, [ 2 C A ~ ['~1"t2] E A
~(q) = s p a n { e i (q) :i E I}
for a l l q C U.
In the same manner as w e did for distributions we may define the
10
!
Conversely, the annihilator o f ~, d e n o t e d ~-, is the d i s t r i b u t i o n
defined by the rule
A = sp{x ~8'
}
Then
Ai(x) = {0} if x ~ 0
AA(x) = T N if x : 0
x
x = 0. []
A I = s p { d x I - dx2}
A i2 = sp{dXl _ (I + x l ) d x 2}
d
e = [ ci~ i
i=I
ker(h~) : p ~ {v 6 T p N : h ~ v = 0}
(sp(dh)) = ker(h#) .
3. F r o b e n i u s Theorem
for all q C V.
There are two important consequences related t o the n o t i o n of
completely integrable distribution. First of all, observe that if t h e r e
exists a cubic coordinate chart (V,~), with coordinate functions
~I .... '~n ' such that (3.1) is s a t i s f i e d , then any slice o f V p a s s i n g
through any point p of V and defined by
m u s t vanish. From this one deduces immediately that also the last n-d
components of the v e c t o r
9e 3~
are vanishing. Since this vector represents locally the vector field
IT,e] one may conclude that [T,e] belongs to ~, i.e. that A is invo-
lutive.
We have seen that involutivity is a n e c e G s a ~ y condition for the
complete integrability of a d i s t r i b u t i o n . However, it c a n be p r o v e d
that this condition is a l s o sufficient, as it is s t a t e d below
for all q 6 U.
Moreover, let rd+1,...,Tn be a n y o t h e r set of v e c t o r fields with
the p r o p e r t y that
s p a n { T i ( p ) : i = I .... n} = T N. W i t h e a c h v e c t o r
' p Ti
field T i , i = 1,...,n, we a s s o c i a t e its f l o w #t and we c o n s i d e r the
mapping
14
F : C (0) >N
E
T2 ~n
(~1 ,~n) ~ T 1
: . . . . . . %.1 ~ 2 " ~n(p)
where Ce(O) = {~ ~ ~n:l~il < e, 1 i 3 n}.
If 8 is s u f f i c i e n t l y small, this mapping:
We show now that (i) and (ii) are true and, later, that both imply
the thesis.
~F
T1 (Ti-1 ~i CTn
= (~I)~ "'" ~i_1 )~ Yi o ~i ..... ~n(p) =
F,(%)0 = Ti (p)
where q = F(~).
If we are able to prove that for all q in a neighborhood of p, for Itl
small, and for any two vector fields T and @ belonging to A,
( ~ } , T o~St(q)_ E ~(q)
for i = 1,...,d.
Then, from a well known property of the Lie bracket we have
dVi =
dt (#@-t)~[ 8'Ti] ~St(q)
and, therefore,
d d
dVi = (~St)~[
dt j~1 lij (~t8 (q))] Tj oSt(q) = j~lliJ (~8t(q))vj(t)
16
[V1(t)...Vd(t)] = [VI(0)...Vd(0)]X(t)
[ T1o~(q)...TdO~(q)] =[ ( ~ ) ~ T 1 ( q ) . . . (~)~d(q)]X(t)
e
(~t),Tio#8_t(q) C span{Y I (q) ..... Tp(q) }
i.e.
~
(t),~io~@t(q) C A(q)
d
T = ~ ciT i
i=1
From (i) and (ii) the thesis follows easily. Actu~lly, (i) makes
it possible to consider on the neighborhood v = F(CE{0)) of p the
coordinate chart (V,F-I). By definition, the tangent vector (~)q at
tangent vectors
n,
j Ivij ( ~ x I _< i < d
~i(vj ) = ~ij
S = {x e A n : ~i(x) = ci , i = d + I ..... n}
x+V
)__~ +__!_~}
& = sp{ (exp x 2 ~x I ~x 2
v Vl ( )x x
for all x @ U, or
~I 2I
1 = (LT~ I) = (exp x2)~--~i + ~x 2
~2 ~2
0 = (LT~2) = (exp x 2 ) ~ I + ~x2
~2 = ~2 (x) = Xl - exp(x2)
19
x I = exp(x2) + constant O
n
X i = j=1
[ (LTi~J) ( % )
sP{T I T d} = s P { ~ I
. . . . . . . . . . 3~d }
on V, we must have
LT. ~j = 0
l
LT I LT 1
rank = d
~I ..... ~n"
Setting, as usual
-I
~ij(x) = (LTi~j)o ~ (x)
-I
Therefore, using just ~i(x) to denote the composite function ~jo9 (x) ,
one has
n ~j
LTi~ j (x) = k=~iTik(X)~x k
Setting
K(x)
(3.6) $---~
~ T(x) =
0(n-d)d
(3.7) Remark. There are alternative ways to describe the equation (3.6).
21
li = ~i+d
i 3
(-~-~ ~ j ( x ) - - ~ x ~i(x))e Im(T(x))
41C A 2 C ... C ~r
A collection A I , . . . , 4 r of d i s t r i b u t i o n s on N is s a i d to be in-
dependent if
s i = d i m ( g I + ... + A i _ I)
for i = 2 , . . . , r + I .
The following result is an a d d i t i o n a l extension of Frobenius
Theorem
r
--
(3.13) Di j=~1 A .]
jT~i
is i n v o l u t i v e .
4. I n v a r i a n t Distributions
(4.1) [f,~] a
n ~ I < i < d
~i = [ vij 8x -- --
j=1 j
fA (x) = A x
d d d
[ fA'T] = i=I~[ fA'CiTi] = i=I[Ci[ fA'Ti] + i=I~ (LfACi)Ti
fn(~d+1,...,~n )
n
f(q) = [ fi(~(q)) ( )
i=I
[ f, %] = I fi ~i'~j] =- ( ) e sp{~ ~ . ~ }
i=I i=I ~ j ~i 51 . . . . ~ d
26
~f.
From this we see that the c o e f f i c i e n t s ~ are such that
~f.
- 0
4.7) Lf~ C
This y i e l d s
(Lfw(p),v > = 0
for all q E U.
If ~ s a t i s f i e s these assumptions and if ~ is a l s o invariant under
f, then it is p o s s i b l e to find d more real-valued functions ~I .... '~d
defined on U w i t h the p r o p e r t y that, choosing as l o c a l coordinates on
U the functions ~i ' I ~ i ~ n, for e a c h q 6 U the v e c t o r field f is
represented by a vector f(~) of the form (4.4).
5. L o c a l Decompositions of C o n t r o l Systems
m
(5.1a) x = f(x) + ~ gi(x)u i
i=I
(5.2) Remark. One may define systems with the same structure as (5.1),
with the state evolving on some abstract manifold N (not n e c e s s a r i l y
diffeomorphic to an o p e n subset of l~n). In t h i s case, instead of (5.1),
which is an o r d i n a r y differential equation defined on an o p e n subset
of l~n,one should consider a description based upon an o r d i n a r y dif-
ferential equation defined on the abstract manifold N. The vector
fields f'q1'''''gm will be d e f i n e d on N and so the output functions
h l , . . . , h i. If we let p d e n o t e a point in N then, instead of (5.1), we
may use a description of the form
m
(5.3a) ~ = f(p) + [ gi(P)U i
i=I
m
(5.5a) ~I = fl (~I'~2) + [ gil (61'~2)ui
i=I
(5.5b) ~2 = f2(~2 )
m
(5.7a) %1 = fI(~I'~2) + ~ gi1(~1'~2)ui
i=I
30
(5.7c) Yi = hi (~2)
where (~1,~2) is a p a r t i t i o n of ~ a n d d i m ( ~ I) = d.
Proo]'. As b e f o r e , w e k n o w t h a t there e x i s t s , a r o u n d e a c h x E N, a
coordinate chart (U,~), w i t h coordinate functions ~I' "'''~n ' with
the p r o p e r t y that the v e c t o r fields f'g1' "'''gm are represented in the
form (4.4). M o r e o v e r , we h a v e assumed that
A C [ s p { d h I ..... dh}] !
~h
1
~j = 0
o
x (T) = ~ x )
%Tf being the flow of the v e c t o r field f, and let (~(T),~(T)) denote
the local c o o r d i n a t e s of x(T). We see that the set of points that
can be r e a c h e d at time T, s t a r t i n g from x , lies inside the set of
points w h o s e local c o o r d i n a t e s ~2 are equal (T) " This set is
to ~2
actually a 8lice of U p a s s i n g t h r o u g h the p o i n t x(T).
Thus, we see that locally the s y s t e m d i s p l a y s a b e h a v i o r s t r i c t l y
analogous to the one d e s c r i b e d in section I. Locally, the state space
may be p a r t i t i o n e d into s u b m a n i f o l d s (the slices of U), all of dimen-
sion d, and the p o i n t s reachable at time T, along t r a j e c t o r i e s that
stay in U for all t 6 [0,T], lie inside the slice p a s s i n g through the
point x(T) r e a c h e d under zero input.
The P r o p o s i t i o n (5.6) is useful in s t u d y i n g s t a t e - o u t p u t interac-
tions. S u p p o s e we take two initial states x a and x b b e l o n g i n g to U
with local c o o r d i n a t e s (~,~) and (~,~) such that
a b
~2 = ~2
a
coordinates ~2(t) and ~ (t) are solutions of the same differential
equation (the e q u a t i o n (5.7b)) with the same initial condition. If we
take into account also the (5.7c) we h a v e the e q u a l i t y
hi ltl) o h l (tl)
which holds for e v e r y input u. We m a y conclude that x a a n d x b are in-
distinguishable.
Again, we find that locally the state space may be p a r t i t i o n e d
into submanifolds (the slices of U), all of d i m e n s i o n d, a n d p a i r of
points of e a c h slice both produce the same output (i.e. are indistin-
guishable) under any input u which keeps the state trajectory evolving
on U.
In the n e x t sections we shall reach stronger conclusions, showing
that if we a d d to the h y p o t h e s e s contained in the P r o p o s i t i o n s (5.4)
and (5.6) the further assumption that the distribution ~ is "minimal"
(in the case of P r o p o s i t i o n (5.4)) or " m a x i m a l " (in the case of P r o -
position (5.6)), then from the decompositions (5.5) and (5.7) one m a y
obtain more informations about the set of states reachable f r o m x and,
respectively, indistinguishable f r o m x .
We conclude this section with a remark about a dual version of
Proposition (5.6).
7ij(~) = o
for all I < i < , I < j < d and all ~ in ~(U). But since
~h.
ij (6) =
6. L o c a l Reachabilit~
(i) A is i n v o l u t i v e
(ii) A contains the d i s t r i b u t i o n s p { g I .... ,gm }
(iii) A is i n v a r i a n t under the v e c t o r fields f,gl,...,g m
(~I' .... ~q I~ )
(6.2a) A0 =
q
(6.2b) A k = Ak_ I + [ [Ti,Ak_ I]
i=I
A k C (T I ..... TqlA )
Proof. Suppose U is an open set with the property that, for some k ,
Ak+(p) = Ak++1(p) for all p 6 U. Then, it is possible to show that
(~I,...,TqlA) (p) = Ak+(p) for all p 6 U. For, we already know from
Lemma (6.3) that ( T I , . . . , T q l A ) D Ak+ . Suppose the inclusion is
proper at some p 6 U and define a new d i s t r i b u t i o n A by setting
~(p) = Ak~(p) if p e U
for all q 6 U.
This fact may be p r o v e d by i n d u c t i o n using as N the subset of
N d e f i n e d in the p r o o f of L e m m a (6.4). Let d O denote the d i m e n s i o n of
A0 (which may depend on p but is c o n s t a n t locally around p). Since,
by assumption, A 0 is the span of some v e c t o r fields in the set
{TI,...,Tq}, there e x i s t e x a c t l y d O vector fields in this set that
span A 0 locally around p. Let d k denote the d i m e n s i o n of A k (constant
a r o u n d p) and suppose A k is spanned l o c a l l y a r o u n d p by d k vector
fields 8 1 , . . . , 8 d k of the form
As a c o n s e q u e n c e
37
(~1'''''Tq IA ) = An-1
and ( ~ 1 , . . . , ~ q l A ) is involutive.
(6.8b) R = (f,gl,...,gmlsP{f,gl,...,gm})
(a) P + sp{f} C R
8 i = (v r ..... [Vl,V0]]
8 i = Iv r ..... [vl,gj]]
R C p + sp{f}
R = P + sp{f}
~r+1 = 0
n-0
The last components of the vector field f are v a n i s h i n g because,
by c o n s t r u c t i o n , f 6 R. In the p a r t i c u l a r case where R = P also
the r-th component of f vanishes and the c o r r e s p o n d i n g equation for ~r
is
~r = 0
From the equation (6.12) we see that any trajectory x(t) e v o l v i n g
on the n e i g h b o r h o o d U actually belongs to an r-dimensional slice of U
41
P : t ~--+ ~r(t)
~r = p-1 (~r) = t
~r = I
By d e f i n i t i o n the d i s t r i b u t i o n R is the s m a l l e s t d i s t r i b u t i o n
w h i c h contains f , g l , . . . , g m and is i n v a r i a n t u n d e r f , g l , . . . , g m. Thus,
we may say that in the a s s o c i a t e d decomposition (6.12) the d i m e n s i o n
r is "minimal", in the sense that it is not p o s s i b l e to find another
set of local c o o r d i n a t e s ~i,...,~
'''''~n ' w i t h ~ strictly less than
r
r, w i t h the p r o p e r t y that the last n-r c o o r d i n a t e s remain constant
with the time. We shall now show that, from the p o i n t of v i e w of the
i n t e r a c t i o n b e t w e e n input and state, the d e c o m p o s i t i o n (6.12) has even
s t r o n g e r properties. Actually, we are going to prove that the states
reachable from the initial state x fill up at least an open subset of
the r - d i m e n s i o n a l slice of in w h i c h they are contained.
F : (_,)k ~ N
,#k I (x o)
(tl,...,tk) ' tk ..... tI
(6.16) : (S1,S)x...(Sk,e) -~ N
(tl,...,tk) i , F(tl,...,tk)
is an e m b e d d i n g .
Let M d e n o t e the image of the m a p p i n g (6.16) (which depends on
the p o i n t x). Consider the slice of U
o = {x e U : ~i(x) = ~i(x), r + I ~ i ~ n}
S
x
m
J
8j = f + [ giui
i=1
with u z3. E ~ for I .< i .< m . and. I < j < k, then for e small M is an
embedded submanifold of S o" T h i s implies, in p a r t i c u l a r , t h a t for e a c h
x
x6M
(6.17) T x M C R(x)
(6.18b) gi(x) e Tx M I ~ i ~ m
for all x 6 M. W e shall show that this contradicts the assumption k < r.
For, consider the d i s t r i b u t i o n ~ defined by setting
dim &(x) = k
m k+l
ek+ I = f + ~ giui
i=I
satisfies the c o n d i t i o n
0 k + 1 ( x ) ~ T_ M.
x .k+1
Let x = F(s~ ..... s~) be this p o i n t and ~t denote the flow of
8k+ I. Then the m a p p i n g
k+1
F' : (-s,e) ~ N
(t I ,t k tk+1) I , ~k+1 o F ( t l , . . . , t k )
,... , tk+ I
(s~-Sl)+...+(s~-s k) < T
m
8 = f +
i=Igiui
F~ : (0,E) ~ N
tl . , ~1 (x O)
t1
F2 : (s11'e)x(0'c) ~ N
(tl,t 2 ) I , ~22oQI~tl (x )
Fk : (s k - l , c ) x . . . x ( s kIk-l,)x(0,c)
_ ~ N
m k+1
field 8k+ I = f + [ giui s u c h t h a t 8k+1(x) ~ T_~. This makes it
i=I x
possible to d e f i n ethe n e x t m a p p i n g Fk+1" Note t h a t s2 > sk-1
i for
i = I,...,k-I and sk > 0
k
The p r o c e d u r e clearly stops at the s t a g e r, w h e n a mapping Fr is
defined
Fr : (s~-1'e)x'''(srr--I '6)x(0'6) ~ N
r
(t1'''''tr-1'tr) ' ' ~t ..... ~ (x)
r 1
ui(t) = u k for t 6 | t 1 + . . . + t k _ 1 , t 1 + t 2 + . . . + t k )
S ={x E U: (p+1=~p+1 ({~ (x)) ; (p+2 (x)=(p+2 (x) ,... , ~n (x) =~n (x) }
x,T
~n+l = 1
7. Local O b s e r v a b i l i t y
(i) A is i n v o l u t i v e
(ii) A is c o n t a i n e d in the d i s t r i b u t i o n sp{dhl,...,dhz}l
(iii) A is i n v a r i a n t under the v e c t o r fields f,gl,...,g m
(7.2a) ~0 = ~
q
(7.2b) ~k = ~k-1 + [ L~i~k-1
i=I
~k C (T I ..... Tq{fl )
49
~k = (TI' .... Tq I~ ) []
~i = d(Lvr...LvllJ)
for all q E U.
This may easily be proved by induction as for the c o r r e s p o n d i n g
statement in Remark (6.5). []
(7.7) Remark. From Lemmas (7.4),(7.6) and (2.11) one may also deduce
that if ~ is spanned by a set d l l , . . . , d l s of exact covector fields
50
and ( T 1 , . . . , T q l ~ ) I is involutive. []
m k
@k = f + [ giui
i=I
k
and let ~t denote the c o r r e s p o n d i n g flow. Then, the state reached at
time t k starting from x at time t = 0 under this input may be ex-
pressed as
x(tk ) = ~t
kk o...o~ItI (xo)
Yi(tk) = hi(X(tk))
o )k
FX : (-E,E ~
1
If two initial states x a and b are such that they produce two
identical outputs for any possible p i e c e w i s e constant input, we must
have
a
Fx xb
i (t1'''''tk) = Fi (t1'''''tk)
for all possible (tl,...,tk), with 0 ~ t i < ~ for I < i < k. From this
we deduce that
a
~F x ~FX b
1
(~t 1...~t k ) t 1 = . . . = t k = O (~t1.~.~tk)t1=...=tk=O
O
~F x
(~tl...~tk)t1=...=tk=Ol
= (Let "''L0khi(X))x
(ii) Now, remember that @j , j = I ..... k, depends on (u4 ..... u~) and
that the above equality must hold for all possible choices of
(u~ ..... u~) E ~ m By appropriately selecting these (u~ ..... u~) one
easily arrives at an equality of the form
(7.9)
(Lvl...Lvkhi)xa = (Lvl...Lvkhi)xb
m 1 m
(Lf~2) a + ~ (Lgi~2) aUi = (Lf~2)xb + ~ (LgiY2)xbU~
x i=I x i=I
I I
This, due to the arbitrariness of the ul,...,u m , implies that
(LvY2)Xa = (LvY2)Xb
where v is any vector in the set {f'gl' .... gm }. This procedure can be
iterated, by setting Y3 = L63"''L0kh" From the above equality one gets
m 2
(LvLfY3) a + m~ (LvLgi73) a u2i = (LvLfY3) b + ~ (LvLgz73) u
x i=I x x i=I . xb I
and, therefore,
for all Vl,V 2 belonging to the set {f,gl,...,gm }. Finally, one arrives
at (7.9).
(7.11) Ii = Lv ...Lvlh j
r
li(x') = li(x" )
~s+i(X") = ~s+i (X !
)
for I < i < n-s, i.e. must belong to the slice of U p a s s i n g through
x'. This, in v i e w of the results proved in (ii) completes the p r o o f
in the c a s e where x 6 U .
#
(iv) S u p p o s e x ~ U . Let x(x,T,u) denote the state reached at time t=T
under the action of the p i e c e w i s e constant input function u. If T is
sufficiently small, x(x,T,u) is still in U. Suppose x(x,T,u) 6 U .Then,
using the c o n c l u s i o n s of (iii), we deduce that in some neighborhood U'
of x' = x ( x , T , u ) , the states indistinguishable f r o m x' lie on the
slice of U p a s s i n g through x'.
Now, recall that the m a p p i n g
54
: xO ~ x(x,T,u)
(7.12) x(x,T,u) U
(7.13) R(x) N U* =
T(x1'x2) = -x2( )x + x I ( ~ ) 2 x
by
)T, 0 O, 0 0 0 0
ttXl,X2) = (x I c o s t - x 2 sint,x I sint-x2 cost)
8 (x 1,x 2) = -x2~(~-~i) x + X l e ( ~ 2 ) x
we define a n o t h e r v e c t o r f i e l d on S I , w h o s e f l o w is n o w g i v e n by
e
~t(z) = e3atz.
F r o m T a n d @ we m a y d e f i n e a vector field f on T 2 by s e t t i n g
f ( z l , z 2) = (T(Zl),@(z2))
ficient condition.
a p o i n t of N, and S is an i n t e g r a l m a n i f o l d of A p a s s i n g through p,
T.
then @ ti(p)
i should be a p o i n t of S for all v a l u e s of t i for w h i c h
T.
~tl(p)_ is defined. Thus, S s h o u l d include all points of N that can
l
be e x p r e s s e d in the form
rk T k - I m1
(I .6) o. (p)
@tk~_ 1 "'@t I
O : (-e,S) 'N
tl + ~T ~0 .T
~ t 1 ~ t ~ - t l (P)
i.e. s e t t i n g q = @T (p)
-t I
@m
( tl).0(q ) e A ( ~ I (q))
k
(1.9) T = [ ciT i
i=I
Proof. In this case, the set of all vector fields w h i c h belong to the
d i s t r i b u t i o n is i n v o l u t i v e and, as a c o n s e q u e n c e of Lemma I. (2.7),
l o c a l l y finitely generated. []
o : (-s,c) ~N
8 T
t P , ~TO~tO#~T(P)
T @
~* (~t)t = (~),T(#tO~_T(p)) =
= ( ~ ) , T (~@T(O (t)) )
~,(~t)t e A(o(t))
for all t E (-s,e). This shows that the smooth curve a lies on an in-
T
tegral s u b m a n i f o l d of A. Now,
let Pl = ~ T (p) and P2 = ~t(Pl )" Then
P2 and Pl are two points b e l o n g i n g to a m a x i m a l i n t e g r a l s u b m a n i f o l d
8 8
of 4, and the p r e v i o u s result shows that ~T(Pl ) and ~T(P2 ) again are
two points b e l o n g i n g to a m a x i m a l integral s u b m a n i f o l d of 4. Thus the
Theorem is p r o v e d for p o i n t s p1,p 2 such that P2 = ~Tt(Pl). If this is
not the case, u s i n g Lemma
(1.14) we can always find v e c t o r fields
T1 Tk
[1,...,Tk of A such that P2 = #t1'''#tk(Pl) and use the above re-
sult in o r d e r to prove the Theorem. []
m
(2.1) ~ = f(x) +
[ gi (x)u i
i=I
(a) is a s u b s e t of S
o
x
(b) c o n t a i n s an o p e n set of S o" []
X
and let LC(io) denote the set of all finite ~-linear combinations of
elements of i o. Then, it is p o s s i b l e to see t h a t i = LC(Lo). For, by
construction, every element of i o is an e l e m e n t of i b e c a u s e i,being
a subalgebra of V(N) which contains TI,...,T q , must contain every
vector f i e l d of the f o r m [ T ,k. [ T k_1' ,...,[ ~ i 2 , T i l ] ] ] . Therefore
&L = sp{T : ~ E L}
A c = sp{T : T e C}
(a) A C C p + sp{f} C R
(b) if x is a r e g u l a r point of A C , t h e n Ac(X)=(P+sp{f}) (x) = R(x).
m
= f + ~ qiui with u I ..... u m real numbers, are in A C (and therefore
i=I
tangent to
S o ), then any state trajectory of the control system (2.1)
x
passing t h r o u g h x at t = 0, due to the a c t i o n of a p i e c e w i s e c o n s t a n t
control, will stay in S
o
x
As a c o n s e q u e n c e of this we see that, when studying the b e h a v i o r
of a c o n t r o l system intialized at x 6 N, we m a y regard as a n a t u r a l
state space the submanifold S o of N i n s t e a d of the w h o l e N. S i n c e for
x
all x e S o the t a n g e n t v e c t o r s f ( x ) , ~ ( x ) ..... gm(X) are e l e m e n t s o f
the
x
tangent s p a c e to S at x, by t a k i n g the restrictions to S of the
o o
x x
original v e c t o r f i e l d s f , g l , . . . , gm~'~ one m a y d e f i n e a s'et of v e c t o r f i e l d s
f,gl,...,g m on S o a n d a control system evolving on S
o
x x
1 m
(2.8) x = f(x) + [ gi(x)uq
i=I
(a) is a s u b s e t of S
o
x
(b) c o n t a i n s an o p e n set of S
o"
x
(2.10) Remark. Note that, if Ac h a s the maximal integral manifolds pro-
66
(2.12) d i m Ac(X) = n
d i m Ac(X) = n
3. The O b s e r v a t i o n Space
(3.1) y = h(x)
TI,.-.,T q -
(3.4) Remark. The subspace S may be described as follows. Consider the
set
~S = sp{dl : I e S}
D0 = sp{dl : i 0}
(3.6) Remark. It is possible to prove that the d i s t r i b u t i o n ~0 is in-
v a r i a n t under the vector fields f, gl,...,g m. For, let I be any func-
tion in 0 and r a vector field in ~0" Then ( d l , ~ ) = 0 and {dLfl,Y)=0
because Lfl is again a function in 0. Therefore, from the equality
gl,...,gm.
If the d i s t r i b u t i o n ~ is smooth (e.g. when the c o d i s t r i b u t i o n ~0
is nonsingular) then using Lemma I. (4.8) one concludes that ~0 itself
is invariant under f,gl,...,g m.
y = h(x)
where h(x) is d e f i n e d as
h(0) = 0
!
ically on ~ a n d smt(Q~) = {0}. The maximal integral submanifold of
smt(~) passing through x is the p o i n t x itself.
At the point x = 0, w h i c h is n o t a regular point of ~0 ' we have
that U N S = {0} for all U, w h e r e a s we k n o w there are other poinEs of
U indistinguishable f r o m x = 0. []
(3.10) d i m ~ 0 ( x ) = n
d i m ~0(x) = n
for all x E N.
= Ax + Bu
y = CX
(4.1) f(x) = Ax
k-times
T I (x) = A k b i
T 2 (x) -- Ahbj
~I (x) = A k b i
Y2 (x) = A x
74
then
[ T I,T 2] = A k + I b
1
n-1
= span{Ax}+ ~. Im(AkB)
k=0
&C = sp{f} + P
V = Im(B)+Im(AB)+...+Im(An-IB)
= 0 -I 0 x +
1
0 0 I
v = {x 6 ~ 3 x 2 = x 3 = 0}
and that
p = sp{~}
{x E ~ n : x2 = 0 a n d s g n ( x 3) = s g n ( x ~ ) }
(ii) o
if x is such that both x 2 ~ 0 a n d x 3 ~ 0, then the m a x i m a l sub-
manifold passing through x is the surface
o o
{x e ]Rn : x 2 x 3 = x2x3}. []
76
For, note that functions of the form CiA'~x~ or ciAkb j are such that
k-times
k-times
n~(x) n-1
n ker(CA k)
k=0
m
= Ax + [ (Nix)u i
i=1
y=Cx
gi(x) = Nix I ~ i ~ m
In o r d e r to c o m p u t e the s u b a l g e b r a C we n o t e f i r s t t h a t any v e c t o r
field T in the set {f'gl ..... g m } has the f o r m T(x) = Tx, w h e r e T is an
nxn matrix. If we w a n t to take the Lie b r a c k e t of two v e c t o r fields
TI,T 2 of the f o r m
T 1(x) = T I x , T 2 (x) = T 2 x
we have
Mo = span{T1,...,Tr}
Mk = ~k*
for all k > k . The set of v e c t o r fields
L = {~ e v ~ n ) : T ( x ) =.Tx, T e Mk. }
= Ax + N x u
where x E ~ 3 and
0 1 0 0 0 I
A = -I 0 0 N = 0 0 0
0 0 0 -1 0 0
0 0
[A,N] = 0 0
0 -1
[N,[A,N]] = A
[A,[A,N]] = -N
Therefore, we have
x2 x3 0
(Ax,Nx, [ A,N] x) = -x I 0 x3
0 -x I -x 2
d i m Ac(X) = 0 if x = 0
d i m Ac(x) = 2 if x ~ 0
{x E IR3 : x 21 + x 22+ x 32 = ( x ~ ) 2 + ( x ~ ) 2 + ( x ~ ) 2 }
{x 6 ~ 3 : x I = x 2 = 0}
~I (x) = (Nx)
T2(x) = ([A,N]x)
T 3 (x) = (0 0 I)'
Accordingly, we g e t
8O
1 (x) =
#t x2
x1
~3t(x) = x2
t+ x3
F : (Zl,Z2,Z 3) I ~ ~I ~2 ~3 (xO)
Z1 z2 z 3
cos zltg z2
Zl = COS zltq z2 + u
z2 = -sin zI
z3 = 0
[]
I n-1 m
~(x) = n k e r (CNj I )
k=0 Jl ..... Jk =0 " " "Njk
The distribution ~0 = Q is n o n s i n g u l a r and its m a x i m a l integral
submanifolds have the f o r m x+W, where now
n-1 m
W = A N ker(CN~j ...Njk)
k=0 J1'" "''Jk =0 I
m
x2 = A22x2 + ~ Ni,22x2ui
i=I
y = C2x 2
The above equations are exactly of the form I. (5.7), this time ob-
tained by means of standard linear algebra arguments.
CHAPTER III
INPUT-OUTPUT MAPS AND REALIZATION THEORY
I. F l i e s s Functional Expansions
m
(1.1a) x = f(x) + ~ gi(x)ui
i=1
I = U Ik
k>O
(ik...il) (jh...jl) J ~ ( i k . . . i l J h . . . j I)
84
c : I i ,
Eik-'-ili0(t) = I d~ik'''d~ild~i 0
0
t0 (t) = t
t
~i(t) = lui(~)d~ for 1 < i < m
0
and
t t T
! d~i---d~i0
k
= ~d~i (~)Id~ik_1--'d~i0
!u k
0
(1.2) Example. Just for convenience, let us compute the first few
iterated integrals, in a case where m = I.
t t t
Id~o = t ; ]d~l = lu1(T)d~
0 0 0
t t t
Id~0d~0 - t2~ ; Id~0d~1 = I I~(@)d0dT
0 0 0 0
85
t t t t
0 0 0 0 0
(1.3) Lemma. Suppose there exist real numbers K > 0, M > 0 such that
the series
t
(1.6) y(t) = c(~) + m~ c(ik...i 0) i d ~ i k . .0d ~"i
k=0 iD'''''ik=O 0
t
f tk+1
d~ik''-d~i0 ! ~
0
y(t) = ya(t)_yb(t)
t2
c(g) + c(0)t + c(00)~T+ .... 0
c(~) = 0
k-times
m
( dd--t)t=0 = ~' c ( i ) u i ( 0 )
i=I
t = 0 and g e t
C ( l l i 0) = 0 I ! il,i 0 ! m
In the t h i r d d e r i v a t i v e , the c o n t r i b u t i o n of t e r m s
t t
(c(0i)Id~0d(i + c ( i 0 ) I d ~ i d ~ 0)
i=I
0 0
is
1 1 dui
i = I [ ~ c(0i) + ~ c(i0)] (-d~)t=0
du.
If this is zero for all l
(-~-)t=0 , then c(0i) = -2c(i0) which, toge-
ther w i t h the p r e v i o u s equality c(0i) = -c(i0) implies
c (H) = ~ (x )
(I .9)
c ( i k . . . i l i 0) = L L . .L l(x )
gi 0 gi I gi k
In view of this result and of Lemma (1.3), one may associate with
g 0 , g l , . . . , g m and I the functional
t
(1.10) v(t) = l(x ) + m[ L Lgil.. .Lg l(x) id~ik...d~ild~i0
k=0 io,...,ik=O gi 0 0
(1.11) Lemma. Let g 0 , g l , . . . , g m be as in the previous Lemma and let
11,...,I Z be r e a l - v a l u e d analytic functions defined on N. Moreover,
let y be a r e a l - v a l u e d analytic function defined on ~Z.
Let v1(t),...,vi(t) denote the functionals defined by'setting, in
(1.10), I = 11,...,i = I. The composition Y(v1(t),...,vi(t)) is again
a functional of the form (1.10), corresponding to the setting
I = y(ll,...,l~)-
Proof. We will only give a trace to the reader for the proof. Let
c1,c 2 denote the formal power series defined by setting, in (1.9),
I = 11 and r e s p e c t i v e l y I = 12 , and let v1(t),v2(t) denote the as-
sociated functionals (1.10). Then, it is immediately seen that with
the formal power series defined by setting I = ~i11 + ~212 , where ~I
and e2 are real numbers, there is a s s o c i a t e d the functional
~Iv1(t) + ~2v(t).
With a little work, it is also seen that with the formal power
series d e f i n e d by setting I = 1112 , there is associated the func-
tional v1(t)v2(t). We show only the very first computations needed
for that. For, consider the product
t t t
v I (t)v2(t) = (11+Lg0l I ; d~0+Lg I 1 IId~ 1+Lg0Lg01 lld~0d~0+" "" )
0 0 0
2+Lgo21do+glX2it d1*gogo21
0
t t
0 0
The factors that multiply it d~0 and it d~1 are clearly Lg011 ~2 and re-
0 0
spectively Lg11112. For the other three, we have
but also
t t t
0 0 0
so that the three terms in question give exactly
t
Lg0Lg01112 Id~0d~0
0
It is not difficult to set up a recursive formalism which makes it
possible to completely verify the claim.
If now ~ is any real-valued analytic function defined on ~, we
may take its Taylor series expansion at the origin and use recursively
the previous results in order to show that the composition
Y(v1(t),...,vz(t)) may be represented as a series like the (1.10)
with I replaced by the Taylor series expansion of (ll,...,IZ). []
At this point, it is easy to obtain the desired representation
of y(t) as a functional of the form (1.10).
(1.12) Theo~Gm. Suppose the inputs Ul,...,u m of the control system
(1.1) satisfy the constraint (1.5). If T is sufficiently small, then
for all 0 ~ t ~ T the j-th output yj (t) of the system (1.1) may be
expanded in following way
t
(1.13) yj (t)=hj(x ) + ~ m
~ L I
...Lg hj(x) d~ik.
k=0 i0'''''ik=0 gi0 0 ""d~i0
where g0 = f"
Proof. We first show that the j-th component of the solution of the
differential equation (1.1a) may be expressed as
t
m ...L x~ (x) Id~.
(1.14) xj (t)=xj (x)+ ~ ~ L Y~ J 0 ik ..d~i0
k=0 i0,...,ik=0 gi 0
and
t t
~t Id~id~ik_1"''d~io = ui(t)Id~ik_1'''d~io
0 0
Now, let fj_ and gij denote the j-th components of f and gi '
I < j < n, I < i < m and observe that
t
cx + L Lgikf j i aq0 =
k=0 i 0 , .... ik=O gi 0 ~ k
xj(0) = x~
c(~) = c x
3
cjAk+Ix O if i 0 = ... = ik = 0
c ( i k . . . i 0) = cjAkbio if i0 ~ iI = ... = i k = 0
0 elsewhere
c(9) = c x
3
o
e(i k. ..i 0) = c j N i k "Ni0x
where N O = A. []
92
2. V o l t e r r a Series Expansions
t Tk T2
SO = {t e l ~ : T >_ t ! 0}
the s e r i e s
S3
t ~k T2
(2.4) ~, ... W. (t,Tk ..... TI)U~(Tk)...
y(t) =w0(t)+k~=1 ii .... ik=1 0 ~k'"il
It
and P (x),...,Ptre(x) the vector fields
Then, there exist real numbers K > 0 and M > 0 such that the
condition (2.2) is satisfied. []
(2.9) Th6orem. Suppose the inputs u I ..... u m of the control system (1.1)
satisfy the c o n s t r a i n t (2.3). If T is s u f f i c i e n t l y small, then for all
0 ~ t ~ T the output yj(t) of the system (1.1) may be e x p a n d e d in the
94
form of a Volterra series, with kernels (2.8), where Qt(x) and P~(x)
are as in (2.6)-(2.7) and ~ = hj. []
nk nI n0
n0...n k (t-T k) ...(T2--T 1 ) T1
= Z C,
Wlk---il(t'tk'''''~l) n0 . .nk=0 Xk'''il nk! . ..nl,n
. . 0,
where
0
n0-..n k = Jl n +...+n k Wlk'''i I
ci k. . .i I nk nI n0
9(t-T k) . ..8(T2--T I) 8T I
t-xk=...=T2-~1=~1=0
y(t) = ~ e0 (d~O)n
n=O
(2.11) 0
= t
n0""nk I ( d ~ 0 ) ~ d ~ " nl n
c. . . . (d(0) d~il (d(0) 0
+k-Z-1il...!k=l n0!..nk=0 lk'''11 0
95
nO nI nk_ I nk
Lf LgiILf ...Lf LgikL f hj(x )
and so on.
The last step needed in order to prove Theorem (2.9) is to show
that the Taylor series expansions of the kernels (2.8), with Qt(x)
and Pt(x) defined as in (2.6), (2.7) for I = h i (t) coincide with the
expansions (2.1 2) .
This is only a routine computation, which may be carried out with
a little e f f o r t by keeping in m i n d the w e l l - k n o w n C a m p b e l l - B a k e r - H a u s -
dorff formula, which provides a Taylor series expansion of Pti (x) " A c -
cording to this formula it is possible to expand Pt(x) in the follow-
ing way
~n , ,tn
Pt(x) = (#ft), gio(~t(x)) = ~ aafgi~x'~.
n=0
96
and, therefore,
and so on.
3. Output Invariance
m
= f{x) + [ gi(x)ui
i=I
yj = hi(x) (j = 1 ..... ~ )
and let
yj(t;x;u I ..... u m)
Lgihj(x) = 0
for all x E N.
(3.4) L I .-.LTrhj(x) = 0
C(ik...i0) = 0
I d~ik'''d~i0
0
k=0
~ T c(ik'''i0 )I d~ik'''d~i0
i0'''''ik=0 0
in which the only nonzero coefficients are those with at least one of
the indexes i0,..,i k equal to i. The sum of this series is zero
b
for every input U l , . . . , u i _ 1 , v ,ui+1,...,u m. Therefore, according to
Lemma (1.7), all its coefficients must vanish, for all x 6 N. We con-
clude that (3.4) and, accordingly, (3.3) are satisfied for all x 6 N . []
~0 = sp(dX:X 6 0}
(dhj,g i ) (x ~ = 0
99
for all k ~ 0 and for all ik,...,i 0 6 I. From the above discussion we
conclude that the condition stated in Theorem (3.2) is equivalent to
the condition
(3.5) gi 6 n 0
3
Moreover, since
~ C sp{dhj} i
3
we see also that if (3.6) is true, we must have
Thus, we have seen that (3.5) implies (3.6) and this, in turn,
implies (3.7). We will show now that (3.7) implies (3.5) thus proving
that the three conditions are in fact equivalent.
For, observe that any vector field of the form [T,g i] with
T 6 {f,gl,...,gm} is by definition in the left-hand-side of (3.7).
Therefore, if (3.7) is true,
0 = <dhj,[ Y,gi]) = L ~ L g i h j - L g i L T h j
LgiLTh j = 0
100
i.e.
gi (sp{dLThj })
(3.8) gi (sp{dLTk'''LTIhj})
i) gi 6 ~ 0
3
iii) (f'gl ..... gmlSp{gi } > C n0.
J
L h(x) = 0
gi 3
L L ...L h.(x) = 0
gi T1 ~r 3
for all r ~ I and any choice of vector fields ~I' .... T r in the set
xl = f 1 ( x l , x 2 ) + ~ g k l ( X l , X 2 ) U k + g i ( x l , x 2 ) u i
k=1
k~i
m
x2 = f2(x2 ) + ~ gk2(X2)Uk
k=1
k~i
Yj = hj(x 2)
from which one sees that the input u i has no influence on the output
yj. D
(ii) gi A C (sp{dhj}) []
(i') ~ is i n v a r i a n t under f'gl' .... g i - 1 ' g i + 1 ' ' ' ' ' g m
Note that this implies that if there exists a distribution A with the
properties (i') and (ii) there exists another distribution ~ with the
properties (i) and (ii). []
puts Y J I ' ' ' ' ' Y J r has to be u n a f f e c t e d by a given set of inputs
m
~ = A x i=[Ibiu Z
yj = cjx j = I ..... Z
n-1
N ker (cjA k)
bi k=0
n-1 n-1
Im(Akbi ) C N k e r ( e j A k)
k=0 k=0
n-1
im(Akbi ) C ker(cj)
k=0
b i C V C ker(cj) .
4. L e f t - I n v e r t i b i l i t y
tion of i n v e r t i b i l i t y .
A system is left-invertible at x if w h e n e v e r u a and u b are two
different input functions
y ( t ; x ; u a) ~ y ( t ; x ; u b)
= f(x) + g(x)u
(4.1)
yj = hj(x) I < j <
k
(4.3) L Lfh. (x ) ~ 0
g 3
Proof. Suppose that u a and u b are two different analytic input func-
tions. Then, there exists an i n t e g e r r such that
dru a dru b
(4.5)
(d--~ -) t=o ~ (d-~- ) t:o
h ( x )
3
L L ...Lg hj(x )
gJ0 gJl Jk
where, in this case, 0 ~ j 0 ..... Jk ~ I and g0 = f' gl = g" From (4.4)
we have that the only possibly nonzero coefficients in the series are
those in which:
- either J0 = "'" = Jk = 0
ko-times
ko+ro+ 1
d yj =
k +r +1
dt o o t=O
ko i k Iro1!t ]
= ~ [ L ...Lg Lfhj(x ) d~jk'"d~j0 0
k=0 J0 ..... Jk =0 gJ0 Jk IdtrO+1 t=
Idrol!tI Id t1
r+------~
dt d~1 t=0 [ dt o t=0
k +r +I k +r +I
d o . o b, I
dtko+ro+ I
it=0 dtko+ro+l Jt= 0
k r Idrou b] ]
= LgLfhj ( x ) I I ~ ] - m ~ 0
[fat jt--0 [dt it=0J
This c o m p l e t e s the proof. []
LgLkhj (x) = 0
for all k > 0 and for all I < j < . This in turn implies that all
the c o e f f i c i e n t s of Fliess e x p a n s i o n of y(t) vanish but the ones in
which o n l y d i f f e r e n t i a t i o n s along the v e c t o r field f occurr. Under
these c i T c u m s t a n c e s we have
= k
y(t) =k~0Lkh(x)~.l
" =
L g L ~ h j ( x ) ~ 0
5. R e a l i z a t i o n Theory
= Ax + Bu
y=Cx
C exp(At)B = w(t)
107
h ( x O) = c(~)
L L ...L h(x ) = c ( i k . . . i l i 0)
gi 0 gi 1 gi k
If t h e s e conditions are satisfied, then it is c l e a r that the
dynamical system
m
x = g0(x) + ~ gi(x)ui
i=I
y = h (x)
m
(5.1) c = c(~) + [ X C (ik. i~) Z . . . . Z
"" v ik i0
k=0 i0...ik=0
108
d m
(5.2) p = p(@) + ~ [ P(ik...i0)zik...zi0
k=0 i 0 . . . i k = 0
istence of r e a l i z a t i o n s w i l l be c h a r a c t e r i z e d as a p r o p e r t y of the
formal p o w e r series w h i c h s p e c i f i e s the functional. We a s s o c i a t e w i t h
the formal p o w e r series c two integers, w h i c h will be called, fol-
lowing Fliess, the Hanke~ rank and the Lie rank of c. This is done
in the f o l l o w i n g manner. We use the given formal p o w e r series c to
define a m a p p i n g
F c : ~ ( Z ) ~ ~ Z < < Z ))
in the f o l l o w i n g way:
[Fc(Z3k
. . . . Zjo)] (ir-.-i 0) = c(i r...i 0 Jk...jo )
F c~(Z )) C ~R(( Z ))
Fc(L(Z)) Cm~<{Z ))
C ( i r . . . i 0 J k . . . J 0)
c : Z + ~I~
os
C ---- ~ ckzk
k=0
and the Hankel matrix associated with the mapping F c coincides with
the c l a s s i c a l Hankel matrix associated with the sequence c0,cl...
cO cI c2
cI c2 c3
S ~-
c
c2 c3 c4 O O I
I Q D
[]
ac(X )
PL(C) = dim Ac(X)-dim K(x ) = dim
ac(x ) n ~0i(x)
: L(Z) -~ V~R n)
by setting
m
Fc(p) = L (p) h(x) ...L h(x)z . . . . z.
+k~ [ L (p)Lgi0 gi k ik 10
0 io,...,ik=O ~
If we let v denote the value of the vector field ~(p) at x , the above
can be rewritten as
m
FC(p) = ( d h ( x ) , v )+ (dL ...L h(x ) ,v)z . . . . z.
k=0 i0,...,ik=0 gi 0 gi k zk 10
When p ranges over i(Z), the tangent vector v takes any value in
Ac(x). Moreover, the covectors dh(x ) . . dL ...Lg h(x),...
, . , gi0 ik
112
o
d i m A C(x ) - d i m A C ( x ) N ~0(x )
V
p_(c) = dim V-dim W N V = dim
W O----V
m
W = (ker C) N | n N k e r ( C N i r . ..Ni0 ~
r=0 i 0...ir=0
with N O = A. With the same kind of a r g u m e n t s one proves that the sub-
space V may be e x p r e s s e d as
113
V =
span{xO] + ~, m[ s p a n { N j k - - - N j x O}
k=0 j 0 . . . J k = 0 0
o
CN . . . . N N . . . . N. x
ir 10 ]k 30
(CNir'''Ni0) (Njk'''Nj0x)
m
= Ax + ~ N xu
i=I x m
y = Cx
h (x) = Cx
114
is a r e a l i z a t i o n of c.
a) 1.s = s
Note that from (a) and (b) we have that for all __ik-.-J0 @ I
(Zjk .. "z30
-s) (ir...i 0) = s(i r...i0jk...j 0)
Note also that since the ring ~< Z } is not commutative, the order in
which the p r o d u c t s are p e r f o r m e d is essential.
We leave to the reader the simple proof that the m a p F c previously
d e f i n e d becomes an ~( Z }-module m o r p h i s m when ~ ( ( Z ) } is e n d o w e d with
this k i n d of I~(Z > - m o d u l e structure. As a m a t t e r of fact, it is trivial
to check that F c(p) = p.c.
Now, c o n s i d e r the c a n o n i c a l f a c t o r i z a t i o n of F c
F
~<Z }
Ker F
c
m(Z >
X =
ker F
c
But X is also an ]R( Z ) - m o d u l e , so to e a c h of the indeterminates
z0,...,z m we m a y associate mappings
Mi:X~X
X ~--+Zi'X
H : X~]R
by taking
H X = [Qc(X)I (@)
o
x = P (I)
C
(5.9a) c(~) = Hx O
o
(5.9b) c(i k...i 0) = H M i k . . . M i 0 x
o
(5.10a) c = Fc(1) = QcPc(1) = Qc x
116
o
(5.10b) Fc(zik...zi0) = QcMik...Mi0x
t h e n we h a v e
x O) ...M. x O
Qc(Zi-Mik...Mi0 = QcMiMik 10
[ F c ( Z i k . . . z i 0 ) ] (~) = c ( z i k . . . z i 0 )
(k+1)
(5.12) 11c(ik...i0)11 1 <_ C(k+1) !r
m
(5.13) sl (@)s2(~) + ~ ~ sl ( i k . . . i 0 ) s 2 ( i k . . . i 0)
k=0 i0,...,ik=0
n ~ ~
cs14) w = exp xiP i) = 1 xiPi Ik
i=1 1 " i=I
where X l , . . . , x n are r e a l v a r i a b l e s .
The s e r i e s c which is to be r e a l i z e d a n d the s e r i e s w thus de-
fined are u s e d in o r d e r to c o n s t r u c t a set of a n a l y t i c functions of
Xl,.i.,Xn, defined in a n e i g h b o r h o o d of 0 a n d i n d e x e d by the e l e m e n t s
of I , in the f o l l o w i n g way
h. (x) = ( S (c),w)
ik...i 0 zx.k. . . . zl 0
h(0) = c(g)
hik...i0(0) = c(ik...i0)
and this shows that the set {h,g0,...,g m} together with the initial
state x = 0 is a realization of c.
To find the vector fields g0,...,g m one proceeds as follows.Since
the n series Fc(Pl),...,Fc(Pn) are ~ - l i n e a r independent, it is easily
seen that there exist n monomials ml,..~,m n in the set Z ~ with the
property that the (nxn) matrix of real numbers
[Fc(Pl)] (m I) .-.[Fc(Pn)] (m I)
(5.16)
From this, using linearity, one concludes that the above expression is
true also in the (general) case where Pi is an ~ - l i n e a r combination
of elements of Z .
Using this property, we conclude that the j-th row of the matrix
(5.16) coincides with the value at 0 of the differential of one of
the functions hik...i 0 , the one whose multiindex corresponds to the
monomial m..
3
Consider now the system of linear equations
119
gk (x) =
V~ <Smn(c)'w ) (Smnzk(c),w)
Lg k (S m i ( c ) , w ) = (Smizk(c),w)
6. U nniqueness of M i n i m a l Realizations
for all x 6 V b.
li(x) = Lv . . . L v l h j (x)
r
m
0i = f + [ -i
j=1 gjuj
defined on a n e i g h b o r h o o d (-e,c) n of 0.
F r o m a n y two m i n i m a l realizations,labeled "a" a n d "b" we w i l l
construct two of s u c h m a p p i n g s , denoted G a and G b (the same set of
u~'s b e i n g u s e d in b o t h G a a n d Gb).
3 a a
Recall that a minimal realization {fa,g~,...,gm,X } satisfies the
controllability rank condition at x a (Corollary (5.18)). From the p r o -
perties of 6 C a n d R (see R e m a r k II.(2.7), one deduces t h a t the d i s t r i -
bution R is n o n s i n g u l a r and n - d i m e n s i o n a l a r o u n d x a. Then, using the
same a r g u m e n t s as the o n e s u s e d in the p r o o f of T h e o r e m I. (6.15), it
is p o s s i b l e to see t h a t there exist a choice of u ~ ' s and an o p e n sub-
3
set W of (0,E) n such t h a t the r e s t r i c t i o n of G a to W is a d i f f e o m o r -
phism of W o n t o its image Ga(w).
a a a .a a~
(iii) It is n o t d i f f i c u l t to p r o v e t h a t if {f , g l , . . . , g m , n ,x ~ a n d
{fb, g ~ , . . . , g mb, h b ,x b } are t w o r e a l i z a t i o n s of the s a m e f o r m a l p o w e r
series c, then, for all 0 < t. < e, I < i < n, w i t h sufficiently small
6,
(6.4) H a o G a ( t l , . . . , t n ) = H b o G b ( t 1 , . . . , t n)
AS a m a t t e r of fact, if c is small t h e n G ( t l , . . . , t n) is a p o i n t of
UH , r e a c h e d from x under the p i e c e w i s e constant control defined by
F : Va ~ Vb
as
(6.5a) F = (Hb)-1 oH a
whose i n v e r s e m a y a l s o be e x p r e s s e d as
(6.5b) F -I = G a o ( G b) -I
m givia 0b fb ~ b
8a = fa + i=I
~ = +i I givi
eb b
Hao~aoGa(t1,...,t n) = Hbo~t oG (t 1,...,t n)
a a a b b b
(H),gioG (tl, .... t n) = ( H ) , g i G (t I ..... t n)
for all 0 < i < m. But these ones, in view of the definitions (6.5),
may be rewritten as
b
gi(x) = F ~gia oF-I (X) 0 < i _< m
(vi) Again, using the same arguments already used in (iii) one may
easily see that
i.e. that
m
(1 .I) = f(x) + [ gi(x)ui
i=I
m
(1.2) u i = Gi(X) +j=~l~iJ (X)Vj
m
(I 3) ~ = ~(x) + [ ~g i (x) vi
i=I
in w h i c h
% m
(1.4a) f(x) = f(x) + [ gi(x}~i(x)
i=I
m
(1.4b) gi(x) - j [ i g j (x) 6ji(x)
~u
(1.5a) f(x) = f(x) + g(x)~(x)
I
(1.6a) f(x) = f(x)-~(x)~- (x)a(x)
for all x E N.
A distribution A is said to be ZocalZy controZZed invari~nt if
for e a c h x 6 N there exists a neighborhood U of x w i t h the p r o p e r t y
that A is c o n t r o l l e d invariant on U. In v i e w of the p r e v i o u s defini-
tion, this requires the e x i s t e n c e of a f e e d b a c k pair (~,8) defined on
U such that (1.9) is true for all x 6 U.
The n o t i o n of l o c a l controlled invariance lends itself to a s i m p l e
geometric test. If w e set
126
G = s p { g I ..... gm }
(1.11a) [f,A] C A + G
m m
[ f,T] =[ f+ge,T] =[ f,T] + [ [ gj,T]~j + ~ (LT~ j)gj
j=1 j=1
m m m
m
[gj,Tl e [ [gi,A] + G
i=I
m
[f,T] e [f,A] + ~ [gi,A] + G
i=I
P i : U -~ ~ n n
127
(I . 1 3 ) -By(x)
Sx = Fi(x)y(x) I -< i -< m
l
y : U~V
y : Uo~V
for all x 6 U.
~2y = ~2y
~xi3x k ~XkDX i
one has
o
Expanding the d e r i v a t i v e s on both sides and e v a l u a t i n g them at x = x
one obtains
[ t"sFk"
~i;xO + F k ( x ) F i ( x ) ] Y =[ ( ~ ) x o +Fi(xO)Fk(xO)]yO
F : (x,y) ~-~
n n
A(x,y) = span{ (8 + ) + ~ ~ Fhk
i (x)Yk(~h), 1 _< i _< m}
i h=0 k=0
1 0 0
0 I 0
0 0 I
[,1 (x)y F 2 (x) y Fm(x)y
all (x,y) and F is nonsingular, one may easily conclude that the func-
tion
= F(x,y)
$~m+I ~m+l
8YI "'" $Yn
(1.15) , , .
~m+n ~m+n
~Yl "'" ~Yn
~ i ( x , y O) = 0
defined in a n e i g h b o r h o o d U c U' of x . M o r e o v e r
q i ( x o) = y O I < i < n
p = dim G - dim A N G
Given a n y x C N it is p o s s i b l e to f i n d a n e i g h b o r h o o d U of x a n d an
mm nonsingular m a t r i x B, w h o s e (i,j)-th element b.. is a s m o o t h real-
z3
valued function defined on U, s u c h that, for
the f o l l o w i n g is true
sp{gp+l ..... gm } C A
(1.18) [gi,A] C A + G
for all I < i < m and I < k < d. U s i n g the same a r g u m e n t s and setting
g0 = f
P
(1,11a')
[ go' ~k] j 0
(I .19)
for I < k < d, I < h < p, I < i < m. Then, it is easy to see that
m
(1.20) [ [ ghbhi,Tk] 6 A
h=1
m m m
[h=1
~ ghbhi'Tk] = -h=1
~ (LTkbhi)gh +j~1 bji[~j'Tk]
P m . P k - ~k = ~k
=- " "
~ (LTkbhi)gh + ~ b.. ~Ichjgh + i l
h=1 j=1 ] l h
m A
[ [ ghbhi,A] c A
h=l
B = BB
(1.21) k = 0
-Lrkah" +j ICkjaj + Ch0
m
(I .22)
[ go +h~ighah'~k ] ~ A
132
m P m P k P
[g0 + h=~ l g h a h " r k I = -h =['1 ( L k a.h ) g h +.j ! l a .j h ~ i. C h j ~ h. + h ~ i c ~ 0 g h + 6 k : ~k
a = Bfi
a
Tk = - ~ k 1 <_ k < d.
k k k
Cll .-. C l m c10
. . .
Fk = k k ck 1 < k < d
Cpl "'" C p m pO
0 ... 0 0
0 ... 0 0
~x k = I _<kid
the unknown vector y being m+1 dimensional. Since now the functions Fk
depend also on the c o o r d i n a t e s Xd+1,...,x n (with respect to which no
derivative of y is considered), in order to achieve uniqueness, the
o o
value of y must be specified, for a given x l , . . . , x d, at each Xd+1,...,xn.
For consistency, the last c o m p o n e n t of the initial value of the solu-
tion sought for the equations (1.23) must be set equal to zero, whereas
the last c o m p o n e n t of the initial value of the s o l u t i o n sought for the
equation (1.24) m u s t be set equal to I. In addition, the first m com-
ponents of the initial values of the solutions sought for each of the
equations (1.23) must be columns of a n o n s i n g u l a r mxm matrix, in order
to let B be nonsingular.
The s o l v a b i l i t y of an equation of the form (1.25) depends, as we
have seen, on the f u l f i l l m e n t of the i n t e g r a b i l i t y conditions (1.14).
This, in turn, is implied by (1.11). Consider the Jacobi identity
-[[gi,Tk],Th]+[[gi,Th],T k] =[ g i , [ T h , Tk]]
for any 0 <__ i <_ m. Using for [gi,Tk] and [gi,Th] the e x p r e s s i o n s given
by (1.11a') or (1.11b') and taking Tk = ~x-~ ' Th = ~ one easily ob-
tains
P k - 5k 3 ] P h - + 6h ~ ]
[j~ICjigj + i,~ h - [ X C4ig4 i,~ k = 0
j=1 ~
This yields
P ~cki P P h - ~k 3 .
- ~. ~x h gj + ~ c k +6 h) +|
j=1 j=1 Ji (~I cjgi i' ~ h l
P ~c~'i P h P k - 6h 3_~_]
j=1 j=1 3x i'gx k
and sP{~1 ..... gp> are direct summands and gl ..... ~p are linearly in-
dependent, the previous equality implies
ck . ~c h . P P
31 + 31 h k k h
3xh ~ +~icj~c~i= -~Ic9~c~i= = 0
134
m
(2.1a) x = f(x) + ~ gi(x)u i + p(x)w
i=I
(2.1b) y = h(x)
(i') [ f,~] C ~ + G
m
(2.3) x2 = f2(x2 ) + g i 2 ( x 2 ) u i
i=I
y = h(x2)
136
~0 = KI
(2.5)
m
Dk = ~k-1 + Lf(G L n~k_l ) + [ L (Gi n ~k_1)
i=1 gi
Proof. T h e f i r s t p a r t of the s t a t e m e n t is a t r i v i a l c o n s e q u e n c e of
the d e f i n i t i o n s . A s for the other, note first that f r o m the e q u a l i t y
~ k * + 1 = a k e we d e d u c e
L g i ( G i n ~k *) c ~ k *
we have
<L ~,T ) = 0
gi
(~,T) = 0
because T 6 ~k,
+ G. T h u s
(~,[gi,Y]) = 0
[ gi,] E ~k* + G
<L ~,~ ) = 0
gi
L (~L n ~ ) c ~l
gi
Suppose
~A D ~k
m ~ ~
~k+1 C ~k + Lf (~ ~ G) + ~ n ( A G I) C
i=I gi
/
C ~k*
and ~k% is the m a x i m a l element of I ( f , g ; K ) . []
138
L
(2.6) 3(K) = (~0+~1+...+~k+...)
on U. C o n s i d e r the distribution
d d
T" = [ ~ cij[ Yi,Tjl
i=I j=1
[gk,D] C D + G
[gk,[Ti,Tj]] C D + G
[gk,D] C D + G
D CK
D = J (K)
p E J (H)
m
~k = ~k-1 + L ~ (Gi ~ ~k-1 ) + ~ L~ (G N nk_1)
i=I gi
m m
L%~ = Lfe + ~ (Lgl~)~i + ~ (~'gi dei
f i=I " i=I
m m
L% ~ = ~. (Lgj~)Sji + ~ <~,gj )d~ji
gi j=1 j =I
m m
L~(G
If n ~k-1 ) +~i=IL~gi (G~ n ~k_1 ) C Lf(G ~ n ~k_1)+~i=iLgi (G N '~k-1)
3. Some Useful A l g o r i t h m s
(3.1a) LgjL~hi(x) = 0
Pi
(3.1b) LgjLf hi(x) ~ 0
tk (~(xO))
Yi(t) = ~ L~h i(x)~T. = h i
k=0
Pi Pi
(3.2) aij (x) = LgjLf hi(x) = {dLf hi(x),gj (x))
Pi +I Pi
(3.3) bi(x) = Lf hi(x) = <dLf hi(x),f(x))
(3.4) Lemma. Let (~,~) be any pair of feedback functions and let
= ~ + g~' gi (g~)i" Then
142
% Pi
aij(x) = L,~ L_ h i ( x )
gj
bi(x) = L~i+lhi(x)
Then
~(x) = A(x)e(x)+b(x)
m _k+l
L~k+1.n i (x) = L L hi (x) ~ = Lk+lhf i (x) + ~ L n ~ h (x)dj (x) = bf hi(x)
f j&1 gj r i
Z Pi
(3.7) = N n (sp{dL h i} )x
&sup i=I k=0
= = = dLf n i, T >
A c n
i=I k=O
P~nZ(sp{dL~hi))
(3.9b) (dLkhi,[~,T]> = 0
Pi Pi Pi +I
0 = L~(dLf hi,T )-<dL~Lff hi,T ) = <dL~ hi,r> = ( d b i , T >
0 = (d~ij,T)
O
Proof. Suppose t h a t the d i f f e r e n t i a l s are l i n e a r l y dependent at x
Then there exist real n u m b e r s Cik , 1 ~ i ~ Z, 0 --< k --< P i s u c h that
(3.12) ~ ~ CikdL~hi(x) = 0
i=I k = 0
Pi
C i k L f h i ix)
i=0 k=0
Z Pi ~ C aij (x)
< dl,gj > ix) =i=I~ c.lpi<dLf h i , g j > (x) =i=I zPi
Pi -1
k
~(x) = I [ cikLfh i (x)
i=0 k=0
145
(with the u n d e r s t a n d i n g that the above sum is exendend over all non-
negative k's) and observe that, if 0 ~ k ~ Pi-1, then (*)
-(dL~hi'[r f'gj] ) = ( d L k + l h )
f i'gj
i Pi -I Z
(dy,[f,gj]> (x) = - ~ ~ . - k+1
Cik%dLf ,gj ) = - [ c aij(x )
i=0 k=0 i=I i'Pi-1
been proved being equal to 0, the function X(x) is such that dy(x)=0
and the above e q u a l i t y implies again the linear dependence of the rows
of the m a t r i x A(x), i.e. a contradiction. Therefore Ci,Pi_I =
= ... = Cz,p_1 = 0 (for all Ci,Pi_1 defined, i.e. such that Pi ~ I).
(*)
k k )
- { dLfhi, [ f'gj ] > = < dLk+lh i' g j> --Lf< dLfh i 'gj
and the last term is zero because k < 01-1.
146
borhood o f x .
The algorithm consists of a f i n i t e number of i t e r a t i o n s , each one
defined as follows.
Iteration (k). Consider the sk xm matrix Ak(X) whose (i,j)-th
entry is ( d l k i ( x ) , g j ( x ) ) . Suppose that in a n e i g h b o r h o o d U k of the
point x the r a n k of Ak(X) is c o n s t a n t and equal to r k. Then it is
possible to f i n d rk rows of Ak(X) which, for all x in a n e i g h b o r h o o d
U'k C U k of x , are linearly independent. Let
Pkl
Pk =
Pk2
~k2 = sp{dl: I E Ak }
Sk+ 1
~kl + ~k2 = ~ sP{ dlk+1 }
j=1 'J
Ik+1,i"
This concludes the d e s c r i p t i o n of the algorithm. []
Sk*
A = n (sp{dlk,,i
i=O
[f + ge,A*] C 4"
sk
~k = ~ sp{dlkj}
j=1
sk
cidlkj(X)
i=I
149
(Cl...csk)Ak(X) = 0
m m
Dk + [ L (~k A G ) = ~k + ~ L~ (~k A G L)
i=0 g i i=O g i
NOw, take the Lie derivative of (3.20) along gi ' with ~ and 8
solutions of (3.18). As a result one obtains
L ~ = ((L%YI)Pk2 + (L~Y2)Pkl)dl k + y i P k 2 d L Ik + Y 2 P k l d L ~ l k
gi gi gi gi gi
PkiL~Ik = Pk1(dlk,g0 )= 0
go
PkIL~Ik = Pk1(dlk,~i ) = constant
gi
for all I <_ i < m. Thus, in the above expression we may replace Y2
with any arbitrary rk-row vector 72 of smooth functions. This makes
it possible to express L ~ in the form
gi
m Sk+1
~k+1 = ~k + [ L~ (~k n G i) = ~ k + ~ k 2 =~kl + ~ k 2 = 3~I sp{dlk+1,j}
i=0 gi '=
~k* = ~k*+1
L~ (~k* N G i) C ~k*
gi
and, therefore, we see from (3.21) that every entry of dL Ik. (due
gi
to the a r b i t r a r y n e s s of y4 ) is a c o v e c t o r field of ~k* . But, since
the entries of dlk. span ~k* ' this implies
L ~k. C ~k* 0 ~ i ~ m
gi
and thus ~k* is invariant under g 0 , g l , . . . , g m . ~k* being nonsingular
and therefore smooth, we may conclude that ~k* is invariant under the
new dynamics. []
~k
(3.22) r k = dim
~k N G ~
61 = r 0
62 = rl-rO
(3.23)
4. N o n i n t e r a c t i n g C o n t r o l
m
= f(x) + ~ gi(x)ui
i=I
= ~(x) + [ gi(x)vi
i=1
(4.1a) [~,Ai] C Ai
(g~)j e &i
for all j ~ I i. []
The e x i s t e n c e of a s o l u t i o n to t h i s p r o b l e m is c h a r a c t e r i z e d as
follows
L~ Lkhi(x) = 0
gj
I i = {mi_ I + I ..... m i} I ~ i ~
~(x) = A(x)~(x)
the rows of ~(x) are linearly independent and so are the Z rows of
A(x).
Qi
Ai* = k=0n (sp{dL~hi}) I
where Ai(x) and bi(x) denote the i-th rows of A(x) and b(x). The
scalar Yi(X) and the 1m,row vector 6i(x) are functions whose dif-
ferentials belong to (Ai): in particular, real numbers.
Considering the equations (4.4) all together, for all I ~ i ~ ,
one sees immediately that, thanks to the assumption on the rank of
A(x), there exists a pair of feedback functions (~,8) that makes
(4.3) satisfied simultaneously for all A i , i.e. that AI,...,A are
compatible locally controlled invariant distributions. In particular
if the right-hand-side of (4.4b) is chosen to be the i-th row of a
block diagonal matrix, one has that in the i-th row of A(x) 8(x),i.e.
in the i-th row of A(x), the only elements whose indexes belong to
the set I. are nonzero. This proves that a compatible B exists with
l
the property that
Pi
L~ L~ h i = 0
gj f
Pi
L ~ Lf h i = 0
gj
gj 6 A i
for all j ~ I i.
This p r o v e s that the L o c a l Single-Outputs Noninteracting Control
Problem is solved []
zi0 h i (x)
Zil L f h i (x)
~i(x) =
Pi
Z.
Lf h i (x)
iP i
~Zik (~ + m
Zik - ~x j=1 gjvj) = L ~ zik + j=1
[ L~gj ZikV j
(4.5) m
: + kh.v.
j=1 gj f i 3 = Lf h i = Zi,k+ I
156
m
~ik = i (x) +j~16iJ (x)vj
Again from the proof of Theorem (4.2), it is seen that Yi and ~i4J
(*)
depend only on zi0'''''Z&Pi As a matter of fact, Yi and ~ij may
be simply real numbers.
Finally, by definition, for all 1 < i <
(4.7) Yi = zi0
Yi = hi(~i )
wi th
Zil
fi(~i ) = gij(~i ) =
Z. 6
~Pi
Yi(~i ) ij(~i )
hi(~i ) = zi0
(4.9) Remark. The choice of Xi(~i) linear in ~i ' i.e. the choice
Pi
i(x) = ai0hi(x) + aiILfhi(x) + ... + a i p Lf hi(x)
i
(g~)j ~i
for all j I i. []
~j
N n (sp{dL~hj}) k
Ki = j~i k=0
i = sp{ ~ : 0 ~ k ~ pi } + s p { D z i + l , k :1 ~ k _< d}
= sp{-~-~7
~ : I < k < d}
DZ+I --+I,k
Di A ( [ Dj) = 0
j/i
D I + D 2 + ... + D + I = T M
Moreover,
h. = ~ D.
1 ji 3
Ki = Di + D~,+I
159
5. C o n t r o l l a b i l i t y Distributions
rU
(5.1) sp{gj: j e Ii } C A i C A (sp{dhj}) i
9~i
SO = A N G
(5.3) m
Sk = A ([ f,Sk_ I] + [ [ gj,Sk_ I] + G)
9=I
(~) This condition is indeed necessary in the Single-Outputs Noninteracting Control
Problem if B is nonsingular and all Oi! s are defined.
160
m m
([ f,S k ] + j=l
[ [ gj' Ski) D ([ f , S k _ 1 ] + j=1
[ { gJ 'Sk-1 ])
and t h e r e f o r e
Sk+l o Sk
(5.4) Remark. N o t e t h a t we m a y as w e l l r e p r e s e n t S k as
m
Sk = A A ([ f,Sk_1] + [ [gj,SkJ1 ] +G) + Sk_ I
3=I
or as
m
Sk = A ([ f,Sk_11 + [ g j , S k - I] + Sk_ I + G)
j=l
(5.6) S(A) = (S O + S 1 + . . . + S k + . . . )
m
Sk = A A ([ f,Sk_1] + ~ [ g j , S k _ I] + G)
j=l
161
m
[~,T] = [ f+ge,T] = [ f,T] + ~. ([ gj,~]ej - (LT~ j)gj)
j=l
m
Therefore
m m
[~'Sk-1] +j=1
[ [gj'Sk-1] + G C [ f,Sk_1] +j=1
[ [ gj' Sk-1] + G
(5.9a) [ f,A] C A +G
(5.10) S(a) = A
(5.11a) A0 = A A G
m
(5.11b) A k = [ ? , A k _ 11 + ~ [ ~ i , A k _ l ] + A0
i=1
Ak C A
for all k. This is true for k = 0 and, if true for some k = 0, the in-
v a r i a n c e of A u n d e r ~, 71' "" "'gm shows that Ak+ I C A. Therefore, one
has
m
= ~
Ak A ~ ([~,Ak_ I] + [ [gi' k-1 ] + G)
i=I
(5.12) A k = Sk
m
A n ([~,Sk_ll + X [gi' S k - 1 ] + G) + Sk-1 =
i=I
m
= [~'Sk-ll +i=I
~ [gi'Sk-l] + Sk-1
m
(5.13) Sk = [~,Sk_1] + X [~i,Sk_1] + Sk_ I
i=I
[ f,A] C A + G
q,
sp{gi: i 6 I} C S(A) C A
S(A) N G = sp{gi: i 6 I}
164
~0 = sp{gi: i I}
m
Ak = [f'Ak-1] +i:I
~ [gi'Ak-1] + Ak-1-
Ak C S (A)
Now recall (see Lemma I. (6.4)) that there exists a dense subset U with
the p r o p e r t y that at each x 6 U, ~(x) = ~k(X) for some integer k. Thus,
we have that
~(x) c S(A)(x)
Sk~ = Sk~+1
6. More On N o n i n t e r a c t i n ~ Control
~i = ~ i ( ~ i ) + ~ i ( ~ i ) v i
(6.1) ~#,+1 = f~,+l (~1 . . . . ' (9,+1) + [ gZ+l,j((1 ..... (+1)vj
j=1
Yi = h i ( ~ i )
166
C Ki C n (sp{dhj}) i
(gS) i
d = n - (pl+P2+...+p~+) > 0
K i* n (k~i
~ K k) = Kj n (k~jKk) = K I n K 2 . . . A KZ ~ {0}
C &i C n (sp{dhj}) , I ~ i ~
(gS)i
j~i
(iii) are nonsingular, independent and span the tangent space
(iv) are s i m u l t a n e o u s l y integrable. []
(6.3) Remark. In o t h e r words, this Lemma shows that the simpler state-
ment "A. n G is n o n s i n g u l a r and one-dimensional for all I < i < Z"
essentially replaces the statement "41,...,4 are c o m p a t i b l e and, for
some c o m p a t i b l e feedback, (gS)i 6 A i for all I < i _< ".
Pl
d L f h I (x)
A(x) = (g1(x)...gm(X))
PZ
d L f h (x)
ie.
G = [ (~i n G)
i=I
Z = [ dim(A i n G)
i=I
168
with
~i = cl({il ..... ~in. )
l
such that
~i = (g~)i
= A 2 + . . . + A ~ + sp{g I}
[gl,A2+...+A] C A 2 + . . . + A + sp{gl}
[f + ~ 1 ~ 1 , A 2 + . . . + A ] C A2+...+A
[g161,A2+...+A~] C A2+...+A ~
F o~1 61 0 ...0
c~2 0 62...0
(6.6) o~ =
~9.' 0 0 .6,%
= f + g~ = f + g l e l + . . . + g z ~ z
:u
gI(~i)
0
~2(2 )
~(~) = ..... ~(~) =
~(~) g~(~)
~i = fi(~i ) + gi(~i)vi
(6 .s)
Yi = hi(~i)
Proof. If the m a t r i x A(x) has rank for all x, then G also has rank
for all x (see p r o o f of L e m m a
(6.3)), K. is n o n s i n g u l a r (see Remark
(4.11)) l
and K~ N G also is n o n s i n g u l a r . For, the i n t e r s e c t i o n K ~
i A G
171
gi (x) c.
i=1 Z
Pj
which a n n i h i l a t e dhj (x) .... ,dLf hj (x), for all j ~ i. T h e c o e f f i c i e n t s
ci, .... c of this c o m b i n a t i o n m u s t be s o l u t i o n of the e q u a t i o n
a i + l , 1 (x) . . . a i + 1 , Z (x)
c
gi Ki I _< i < Z
so that
(6.11) K i N G = sp{ i}
because K. n G is o n e - d i m e n s i o n a l .
l
By d e f i n i t i o n
G n Ki C S ( K i) c K i
so that
G n S..(K
i) = G N K i
172
and
Di = S (Kj)
j~i
%
D i O sp{gj : j ~ i}
Therefore,
%
Di D ( ~ ' g l ..... gzlsp{gj : j / i} )
We will show now that also the reverse inclusion holds, so that
D i is actually the smallest distribution invariant under ~, 1,...,g
which contains sp{gj: j ~ i}. As a matter of fact, consider the se-
quence of distributions
'b
Si0 = sp{g i}
Z
Sik = [?,S i k_l ]+ [ lgj,Si,k- I] + Si,k_ I
' j=1
From (5.13), and (6.11 , we deduce that for some k
* ~ >
S(Ki) = (~'gl .... ,g}sp{g i }
tlsfied. Conditions (i) and (li) are satisfied by definition and con-
dition (iii) by assumption. Moreover, the fulfillment of (v) derives
from (6.12). This completes the proof. []
(6.15) Remark. From the proof of Theorem (6.10) it is seen that, when
rank A(x) = and S(KI),...,S(K i) are independent and span the tangent
space, then any feedback solving the Local Single-Outputs Noninterac-
ring Control Problem also solves the strong version of this problem. []
h I (x) = x I
h 2 (x) = x 3
dh I = (I 0 0)
dh 2 = (0 0 I)
we have
174
L g l h I = < d h l , g I } = XlX 3
L g 2 h I = XlX 3
Lglh 2 = I
Lg2h 2 = x 2
Then Pl = 0, P2 = 0 and
[ I x2
Since
det A(x) = X l X 3 ( X 2 - I)
U = {x 6 ~3: Xl ~ 0, x 2 ~ 0, x 2 ~ I}
the p r o b l e m in q u e s t i o n can be s o l v e d on U.
A feedback solving the p r o b l e m is f o u n d via the e q u a t i o n s (4.4).
Taking Yi = 0 and 6i = i-th row of the 2x2 i d e n t i t y matrix, these be-
c ome
A(x)~(x) = -b(x)
A(x) 6(x) = I
where
b(x) = ILfhlx
IIx+XlX31
Lfh 2 (x)
=
x~ - x 3
This yields
2 2 3
-I [ x2 + X l X 2 X 3 + X l X 3 - X l X 3
~(x) x1~3(x2_1)
t
:
3 2
-x 2 - x Ix 3 + x Ix 3 - x Ix 3
175
x2 -x I x 3
I
6(x) =
x I x 3 (x2-I)
-I XlX 3
51 = h I (x) = x I
52 = h 2 (x) = x 3
~1 = V l
~2 = v2
2
~3 = f(~) + ~ gi(~)vi
i=I
Yl = 51
Y2 = ~2
KI
=
(sp{dh 2}
)L = s p { ~ ~}
, ~x 2
XlX 3 XlX 3
2 2
x2x 3 x2x 3 cI I =
c2
I x2
xI
2
T = x2
[g1,T] 6 G, [g2,T] G
So that
Since
x 2 (I -x 2 )
[ f, T] = x2x I (x2-I)
-2x~
then
)L _~}
K2 = (sp{dh I} = s p { - ~ 2 , ~x 3
SO = K2 N G = sp{ }
3
Moreover,
so that
$I = K 2 n ([ f , S 0 ] + [g1,s0]+ [ g 2 , s 0 ] + G) = K 2 A G = S
S ( K 2) = K 2 n G = sp{ }
KIn K2 = s P { ~ x 2 }
whereas
S(KI) N S(K2) = 0.
CHAPTER V
EXACT LINEARIZATION METHODS
m
(I .la) = f(x) + [ gi(x)ui
i=I
(1 .lb) y = h(x)
m
(1.2) u i = ~i(X) + ~ ~ij(X)Vj
j=1
(I .3) y(t) = Q ( t , x O) + [
mft ki(t-T)ui(T)d~
i=I
0
wi th
ki(t-T ) = wi(t,T,x)
(1.5) L g i L ~ h j (x) = i n d e p e n d e n t of x
for all k > 0 a n d all I < j < , I < i < m. We m a y summarize this by
saying t h a t the input-dependent part of the r e s p o n s e of a n o n l i n e a r
system of the form (1.1) is l i n e a r in the input if and only if the
conditions (1.5) are satisfied.
In g e n e r a l , the c o n d i t i o n s (1.5) will n o t be satisfied for a
specific nonlinear system. If t h i s is the case, we m a y w i s h to h a v e
them s a t i s f i e d via feedback, thus setting a rather interesting syn-
thesis p r o b l e m . As usual, we c o u l d look at a g l o b a l problem, in w h i c h
a globally defined feedback is sought which solves the p r o b l e m for
180
-k-1
(I .7) T(s,x) =k~0Tk= (x)s
0 0 ... T o (x)
(~) Recall that f= f+g~ and gi = (g6)i (see IV.(l.4) and IV.(l.5)).
181
0o
K(s) = ~ Kk s-k-1
k=0
Vl = I PII
KI
VITo(X) = I S10(x) I
61 = r 0
YI (x) = P1h(x)
{I (x) = K11h(x)
LgY1(x) = SI(X )
Lg~1(x) = 0
LgY1 (x)
iSi_1(x
)
LgYi_1 (x) LgLfYi_1 (x)
LgLfYi_1(x)
(1.16)
LgLfYi-I LgLfYi-1
184
I61 ... 0 0
oo,
V.
1
0 0
"- I6i_I
0 ... 0 Pi
i Ki K~
K1 "'" i-I z
Lg1 (x)
Ii Si (x)
V, =
l
LgYi_ i (x) 0
LgLfYi_ I (x)
6 i = ri_ I - r i _ 2
Yi(x) = PiLfYi_l(X)
~i(x ) = KiY
i I (x)+...+Ki_IYi_
i i - I (x)
1(x)+KiLfYi_
= S i (x)
h Lg Yi (x)
LgYi (x) = 0
- Si_1
I_ LgLfYi-1
o
and the condition (I.16 holds for all 2 < i < k if and only if x is
a regular point of T i and the condition (1.13) holds for all I < i < k-1.
VI 0
V =
0 VI J
As a result, one obtains
LgP I h LgLfP1h
VILgh LgV I Lfh
0 LgLfK~h
VM I =
0 LgP1h
0 ViLgh
0 0
S1 LgLfX I
0 LgL fX I
0 SI
0 0
186
LgL fX I
and
i SI
rR
-I
I LgLf1
SI
=5<
LgLf I
SI
i.e. the condition (1.16) holds for i = 2. For higher values of k one
may proceed by induction. []
Ii LgY1(x)
LgXq[1(x)
LgLfyq_1(x)
Yq = PqLfXq_1(x)
and
i
- Sq_ I (x) I = Sq(X)
_ LgYq (x)
187
I Yq (x)
that S q = LgF
and r e c a l l is an rq_ 1 m matrix, of r a n k rq_ I at x O .
Then the e q u a t i o n s
(1.18b) [LgF(X)]B(x) = [I 0]
rq-1
on a s u i t a b l e neighborhood U of x are s o l v e d by a p a i r of s m o o t h
functions ~ a n d B.
S u f f i c i e n c y of (b): p r o o f t h a t the a b o v e feedback solves the p r o b l e m .
We show f i r s t t h a t
p Ki-1 I k
(I 19b) i i-I "'" K I L g L ^ h ( x ) = independent of x
(1.20a) L~y i = 0
(1.20d) L~ i = 0
g
(1 .21) K~ 1 k Ki 2 k-
...KIL%h = i...K2L~7I
f
= Ki 3 k-1-
i'''K3L~ Y2 = "'"
= KiL k-i+2 -
i ~ i-I
= -i1
If k < i
11 2 2 1 Ki =
i'" i''" k+l ~Yk
an identity matrix).
Thus, if i < q and k > i-1 we get from (1.21)
p iKi-1 I
i-I"''KIL~L ~ h = Pi'''Kk+2L~(Yk+Ig
k+2 - _3~i_k+1
.= ~j Yj;.
q
= ~_ %Y~_k+1
kk+ IgL ~~ - -
= L~ (~k+1 _311
'= K k3+ 1 j )
because the l a t t e r is e i t h e r i n d e p e n d e n t of x
(if k = 0) o r zero.
In o r d e r to c o m p l e t e the p r o o f of the s u f f i c i e n c y of (b), we n e e d
0nly to p r o v e t h a t the m a t r i x
PI
1
P2K1
(I .23) H =
Pq q-Kql
KqKq- I I
q q-1 "" "KI
L~L~h(x) = independent of x
(x) = B -I (x)
(x) = -8 -I (X)~(x)
and let ~
T k(x) = L~L~h(x)
~u
If the f e e d b a c k p a i r e a n d (3 is such as to m a k e Tk(X) independent
of x for all k (i.e. to s o l v e the p r o b l e m ) , then
k ~ ~ A ~ - % k-1
(I .24) L h = L h + Tk_1(~ + T k _ 2 L f e + . . . + T o L f
k+1 ~ ~ - ~ k-1
nf h = n ( ~ + ~ & ) n h + L f ( T k _ I ~ + . . . + T o L f ~) =
= L~
190
L Lkh = (L L k h ) ~ + %
g x ~ '1~ Tk-lLga- + %Tk-2LgLfC~+""
- ~ k-1-
"+ToLgLf a
orr
~. ~ -k-1
K(s) = TkS
k=0
k=0
Pi
Ki
i
Then, o n e has
Lf~i-1 = QiYi
, + Q i,,( ~ i - K iiY I - . . . - K i_IXi_1
i )
k+1 k+1
(2.3) ~k = ( ~ sp{dYi} + ~ sp{d~i})
i=1 i=1
k+1 k+1
(2.4) ( ~ sp{dYi } n X sp{dYi}) = 0
i=I i=I
k+l
(2.5) ak n G ~ = X sp{d~ i}
i=I
and ~k n G ~ is n o n s i n g u l a r around x O.
~0 = sp{d1} + sp{d~1}
which is c l e a r l y true b e c a u s e
Y1 = Vlh
iI
and V lis a n o n s i n g u l a r matrix. Moreover, by d e f i n i t i o n LgY1 = 0, i.e.
sp{d~1 } C G ~
which implies
But
sp{dY1 } n G ~ = 0
ILgY1 (x ) = IS 1(x ) = 0
~0 N G i = sp{d~1 }
sp{dY1 } n sp{d{1 } = 0
m k+1
i=~ILgi(~k n G ~) C ~ sp{d~j} C ~k
3=I
(because Lgi7 j = 0) and, therefore, that
k+1 k+1
= ~ sp{dy i} + ~ sp{d~ i} + sp{dYk+ 2} + sp{d~k+ 2}
i=I i=I
(the last equality being a consequence of the Remark (2.1)), and this
proves (2.3) for k+1.
Moreover, it is easily seen that
k+2 L
sp{d~i} C G
i=1
k+2
[ sp{d7 i} r? G i = 0
i=I
194
(2.7) Remark. Note, from the proof of Lemma (2.2), that an obvious
necessary condition for the existence of a solution to the Input-
Output Linearization Problem is that the sequence of codistributions
~k g e n e r a t e d by m e a n s of the A l g o r i t h m IV.(2.5) coincides with the
one generated by means of the (simpler) algorithm (2.6).
k+l k+l
(2.9) dim ~ = dim( ~ sp{dYi} ) = ~ 6.
i=1 i=I l
Proof. The first equality follows directly from (2.3),(2.4) and (2.5).
The second one is a c o n s e q u e n c e of the fact that the r k = 61+...+6k+i
rows of Sk+1(x) are linearly independent at each x in a n e i g h b o r h o o d
of x O. []
8 1 = rK(M0)
k+1
~i = ~ ( M k ) - ~(Mk-1) k ~ I
i=I
~q = ~q-1
%
i.e. that k = q-1. []
q-1 - I
= q
N sp{ dxi }J- A n sp{dxi }-
i=I i=I
P P
(2.14) dim( ~ sp{d~i(x) }) < ~ n i
i=I i=I
P
~p-1 F] G i = [ sp{d{i }
i=I
O
for all x a r o u n d x .
Differentiating (2.15) along f yields
k-1
(Lflk)d~ k + l k ( d L f ~ k) = ~ ((Lfli)d~ i + h i(dLf~i))
i=I
, + ,, - ,, k+1- ,, k+1
(Lflk)d{ k + lk(Qk+IdYk+1 Qk+IdYk+l-Qk+IK1 ~Y1-'''-Qk+IKk dYk) =
k-1
=i=I
[ ((Lf~ild~i+~iCQi+Id~i+1Q[Id~i1-Q~IK~IdC..-Q'~1~iIdYill
! i - , " "
This may be r e w r i t t e n as
k
(2.16) k(Qk+ldYk+l +Qk+ldYk+l ) = [ (PidYi + PidYi )
i=1
' = 0
IkQk+ I
pi = 0 I _< i _< k
I n"
k~k+l dYk+1 = PldYl+...+~kdYk
Ik (Qk+1 Qk+1 ) = 0
i = ~(x) + ~(x)v
y = h(x)
198
(2.18) [I~ 0] =
IiEIo
Eq
L%y i = 0
I
L~y i = E i
g
L~y i = 0
g
L~yq_1 = q
L~yq_I = 0
(2.19c)
~q-1 = ~q"
combining the (2.19) 's, one finds in the new coordinates a state
space description of the form
~. = Fz + Gv
fi = f ( z , q ) + g(z,n)v
y =Hz
with
Z = C o I ( C I ..... C q ' ~ l ..... ~q-1 )
and
0 0 0 ... 0 0 0 0 0 ... 0
. ~ . o . .
0 0 0 ... 0 0 0 0 0 ... 0
. , . . , o . .
o o o . . . o ~o o o..o
EI
E
q
G =
0
S'(x) = S(x)
3. Some A l g e b r a i c Properties
P2YI
2-
Hh = P3K2YI
q 1
2
p - 2
L Hh = (2-KIY1)
Kq 3 - 2
q "'" K3(Y2-KIY I)
0
0
L~Hh = Y3
Kq 4 - 3 3
q "'" K 4 (Y3-KIYI-K2Y 2)
202
Lq-IHh =
f Yq
. . q_IYq_1
and
~ +iHh = 0
0
- _K~+i+1 q+i+1
Yq+i+1 YI-'''-Kq Yq
it p o s s i b l e to get
E1
0
HL~h = 0
g
E2
HL L h= 2
-P3KIEI
_Kq 3 2
q "'K3KIEI
q-1
0
Eq
q+i
0
0
-K? +i+ 1E I -"" "-Kq+i+ IE
q q
Since El, .... Eq are rows of the matrix [ Iq 0], one easily under-
stands that the formal power series
co
(3.1) W(s) = H ~ (L Lkh)s -k-1
k=0 f
- -I
I61s 0 ... 0 0
16 s -2 ... 0 0
2
w31(s) 0 ... 0 0
o,
From the previous expression for HL~L~h,~ one also sees that in
the j-th block column of (3.1), I < j < q, the largest power of s
appearing in any o f f - d i a g o n a l element is -(j+1). As a c o n s e q u e n c e , o n e
may conclude that Wij(s)sJ is a strictly proper formal power series.
This p r o p e r t y will be i m m e d i a t e l y used in the following way. Set
I 0 ... 0
0 I ... 0
I I~ s -I 0 I
P1(s)W(s ) = I
0 W22(s)
-k-1
R(s) = R_1(x) +kk0Rk(X)S
w h o s e c o e f f i c i e n t s are m x m m a t r i c e s of smooth f u n c t i o n s d e f i n e d on
a n e i g h b o r h o o d U of x , and a b i p r o p e r formal power series
where
P(s) = Pq(S)...PI(S)H
n0
(3.9a) 61 = dim ]_
n o AG
n. hi_ 1
(3.9b) 6i+ I = dim l dim i > I
~i n G i ni_l n G~
4. L i n e a r m o d e l m a t c h i n g
4.2b) Yi = hi (~'x)
in which
f(~,x) =
I a(~,x)m
f(x) + [ gi(x)ci(~,x)
i=I
bi(~,x)
gi (~,x) = m
j~igj(x)dji (~'x)
hi(~,x) = hi(x)
C
t
(4.3) y(t) = Q[t,(~,x)) + j WM(t-T)V(T)dT
0
WM(t) being a ~ixed %p m a t r i x of functions of t, the impulse-re-
sponse m a t r i x of a p r e s p e c i f i e d linear model. As before, we seek
local solutions defined in a n e i g h b o r h o o d of the initial state. In
view of our earlier discussions, this yields the following formal
208
statement.
= f(x) + g(x)u
(4.5) z = Az + Bv
w = h(x) - Cz
~E = fE(xE ) + gE(xE)uE
w = hE(x E)
g (x, z) = I g(x) 0 I
(4.6a) E
fE(x'z) = I Azf(X)
0 B
k cAkB]
T~(x,z) = L E L k E h E (x, z) = [LgLfh(x)
g f
Then
oo co
= [K(s)
where
co
(4.7a) C I = PI C
210
(4.7b) CI = K~C
(4.7c) C i = PiCi_1A
W i t h the f u n c t i o n s X 1 ( x ) , . . . , y q ( X ) d e t e r m i n e d at e a c h n o n d e g e n e -
rate stage of the A l g o r i t h m we associate, as before, a matrix
r(x) = -71(x)
7q(X)
-C I
(4.8) D =
(4.10a) ~ = A~ + Bv
(4.10b) u = ~(X)-S(x)DA~-8(x)DBv
211
(4.11b) LgE(X)~(x) = Ir
q-1
y~(x) = PIhE(x)
-E I E
Y1 (x) = K1h (x)
~(x) = ~i(x) + 6i z
Ck B = 0
-E
L EYk(X) = 0
g
and the matrix
E
L EY1(x) Ii LgYI (x)
g
L EYk
g
E(x) LgYk (x)
cIB1
CkB
212
has a rank equal to the number 61+...+6 k of its rows, at (x,z O) (note
that z O is irrelevant) As a consequence, one may conclude that the
first i steps of the Structure Algorithm on the triplet(fE,gE,h E) may
be performed exactly in the same way as on the triplet(f,g,h). At each
of these steps, the same set of matrices Pk,K~ ..... K~ makes it pos-
sible to perform the required operations. In particular, since the
integers which characterize the structure at the infinity do not de-
pend on the choice of matrices in the Structure Algorithm, we see that
the first i entries in the structure at the infinity of (f,g,h) and
(fE,gE,hE) coincide. Now, let Vi+ I be the matrix determined at the
(i+1)-th stage of the A l g o r i t h m (1.15) and observe that
E E
L EY1(x ) LgY1(x) CIB Lgy I (x) CIB
g
"E(xE )
Vi+1 L Ey i = Vi+ I Lg~i(x) CiB = LgYi+ I Ci+IB
g
L _L _ ~ (xE) ngLf~ i (x) CiAB 0 Ci+IB
g~ f~ I
From this we see that the (i+1)-th entries in the structure at the
infinity of (f,g,h) and (fE,gE,hE) coincide if and only if
Ci+1B = 0
Sufficiency
one ends up with
FE (xE) =
of
E
(b). At the last nondegenerate
a matrix
YI (xE) ~1 (x)
ClZ
step of the algorithm
= (r(x) Dz)
YqE (xE ) yq (x) CqZ
(4.12b) L EFE(xE)BE(x E) = (I r 0)
g q-1
which correspond to the equations (1.18). If ~(x) and ~(x) are solu-
tions of (4.11), solutions ~E(xE) and ~E(xE) of (4.12) may be found as
213
~(x)-~(x)DAz
aE(x,z) =
~(x) -8(x)DB
8E(x,z) =
0
for all k > 0 around (x,z), for any z. In p a r t i c u l a r (see e.g. the
structure of (3.2)), the last ~ columns of these matrices vanish
for all k > 0.
The extended system (4.5) subject to the feedback thus defined
is d e s c r i b e d by equations of the form
(4.14) z = Az + B~
w = h(x) - Cz
where u and v represent new inputs. The response of this system con-
sists of a " z e r o - i n p u t " t e r m w0(t, (x,z)) and of a linear term in
alone, because, as w e observed, the last ~ columns of (4.13) are
vanishing. This means that, if u = 0, the response of such a system
consists of w0(t, (x,z)) alone.Equations (4.14) with u = 0 may be
interpreted as the composition of the original system (1.1) with the
dynamic feedback
-- Az + B ~
u = ~(x)-B(x)D(Az+Bv)
214
w = y - Cz
and, therefore,
t
y(t) = w0(t , (x,z))+ceAtz + I ceA(t-T)Bv(T)dT
0
C B = 0
1
i ) d+. i d+K i - )d =
= K I (Lgy I - .+Ki_ I (LgYi_ I ) (LgLfYi_ I
= (K C I + . . . + K i _ I C i _ I + K i C i A ) B = Ci B
~k
(5.1) r k = dim G~ k h 0
~k n
E
E ~k
(5.2) r k = dim E r] G EL k >_ 0
k
GE E E E E ]
= s p { g 1 ' ' ' ' ' g m ' g m + 1 ..... gm+p
QO
=E (sp{dh~ ..... dh~])
m+~
E E LfE(~ K A GEL)" [ L E(ak_1 A G El)
9k = ~k-1 + -I +
i=I gi
E
rk = r k
(5.4) rk = r k
E E}
Gu = sp{gl ..... gm
Gv s { E E }
= p gm+1,---,gm+u
and we n o t e that
(5.5) GE = Gu Gv
m+ U -i
(5.6) ~k+l = ~k + ~ T. ~.(~k n Gu)
i=0 gi
~k ~k
(5.7) rk = dim ~ = dim _~
~k n G ~k ~ Gu
r(x,z) = {o} T ~Z ~
(5.8) -~ D ~kE n ~u
Gv
for all k > 0. To this end, note first that the a s s u m p t i o n (5.4),
because of (5.7), m a y be r e w r i t t e n as
E GEL
~k + S~ ~k +
(5.9) dim( T ) = dim(--)
= G E~
U
E
(5.10) ~k + F = ~k +
E -k E + -i -L
~k + F = ~2k + F ~ ~k + G u = e k Gu (because r C Gu)
E -L
~k + G u ~kE + G E L
dim - - = dim (by (5.9))
-i
G GEL
U
d i m ~kE n G ~u = d i m ~kE n G E X
E ~ ~i = E n G E i (because G EL C ~ )
~k u ~k u "
(5.11) -L
Gu n +r=
au n +r) = S Lu n (~ +r) =
E
n ~k+F
Thus, we h a v e
m+u
= ~k + X L E(Cu~ n ~k + r) + r
i=0 gi
m+p (-L N E
= ~k + ~ L E G u ~k + F) + F by (5.11)
i=0 gi
m+p
(G EL n ~kE + ) + (see above)
=~k +~. ~E
i=0 gi
E m+p
E + F
= ~k + [ L E {GE L N ~2k ) + r = ~k+1 by (5.10)
i=0 gi
This shows that (5.10) holds also for k+1. Since (5.10) is true
for k = 0, the p r e v i o u s argument shows that it is true for all k > 0.
As a c o n s e q u e n c e , we h a v e also
219
E n -~ E n G EL
~k G u = ~k
for all k > 0 and this, since G El C ~ k implies (5.8) Note that (5.8)
-- V "
m a y be r e w r i t t e n as
EL
(5.8') G V C Rk + Gu
AE, = ~
EL,
k
is s u c h that
[fE,AE*] C A E* + GE
(5.12a) [ fE,AE*] C b E* +
u
Since E
~k* n Gu
- = ~k*
E N GEL is n o n s i n g u l a r around (x,z ) so is
bE* + Gu" A l s o b E* and Gu are n o n s i n g u l a r and therefore one m a y use
Lemma IV.(I.10) and d e d u c e the e x i s t e n c e of an m1 vector a(x,z) of
smooth functions, defined locally around (x,z), such that
(5.13a) [ fE + E
gi~i , A E*] C A E*
i=I
m
E E AE*
(5.13b) gm+i + ~ E I < i < P
j=1 gjYji -- --
c o n d i t i o n implies
m
(5.13c) [ gm+i
E + [ E
gjxj i,A E*] C A E* 1 < i <
j=1
m p m
.E fE E E E
x = + ~. giai + ~ (gm+i + I g j i )vi
i=1 i=I j=l
w -- h E (xE)
= Az + B Y
u = a(x,z) + y(x,z)v
(5.18a) ~ = A~ + Bv
I
0 ... 0 0 ... 0
1
0 ~ ... 0 0 ... 0
(5.20) W M(s)=C(sI-A)-IB =
... 0 ... 0
I
0 0 ... --~ 0 ... 0
s o
Proof. It is l e f t as an e x e r c i c e to the r e a d e r .
= (dt)0 = .. = (dtS_l) 0 = 0
(t
y(t) = ] WM(t-T)5(T)dw. []
t
(5.22) y(t) = Q(t, ({,x)) + I WM(t-~)v(T)dT
0
6. S t a t e - s p a c e linearization
= Fz + Gv
= f(z,~) + g(z,~)v
y = Hz
(6.1) F,(f+g~)oF-1(z) = Az
(6.2) F,(gS)ioF-1(z) = bi I ~ i ~ m
n-1 m
(6.3) [ ~ Im(Akbi ) : ~n
k=0i=l
m
= (f+g~) (x) + [ (gSi) (x)u i
i=1
and suppose x(t) 6 U for all t 6 [0,T] for some T > 0. If (6.1) and
(6.2) hold, then for all t 6 [0,T]
z(t) = F(x(t))
m
~ = Az + ~. b.u.
i=I i
(6.7) L g L n-1
f ~(x ) ~ 0
0 I 0 ... 0
0 0 1 .. 0
(6.8) T ( A + b k ) T -I = Tb=
0 0 0 ... 1
0 0 0 ... 0
= ~(x) = TF(x)
(x) = ~ (x)
F=~(q~)oE '-1(~) =
225
0 1 0 ... 0
0 0 I ... 0
#,(f+g~)oF-l(z) = . . , o .
0 0 0 ... 1
0 0 0 ... 0
F,(f(x)+g(x)a(x)) = AF(x)
that is
~z 1
~x ( f ( x ) + g ( x ) a ( x ) ) = z2(x)
SZn_ I
8--~(f(x)+g(x)a(x)) = Zn(X)
~Z
n
x-~--~--(f(x)+g(x)a(x)) = 0
~Zn_ I
- x-~-~-- q ( x ) 6 ( x ) = 0
~z
~X n g(x)(3 (X ] = I
226
~z.
(6.9) ~xl g(x) = LgZi(x) = 0 I _< i _< n-1
8Zn I
(6.10) ~x g(x) = LgZn(X) =
~(X)
(6.12) L f Z n (x) = ~(x)
for a l l x E U.
If one sets
Ln-1
Lg f ~(x) ~ 0
F ; x I ~ col(z1(x) ..... Z n ( X ) )
are c l e a r l y such t h a t
F,(f(x) 8(x)) = b
F,~(X) = AF(x)
rU
F,g(x) = b
f'(z) = Az
g'(z) = b
i~ i
F , ( a d ~ g ) (x) = (adf,g')o F(x)
for all 0 < i < n-1. The special f o r m of f' a n d g' is s u c h that
(ad~,g' = (-1)iAib
F,(g ad ~ .. a a_n-1~,
~ g} =
(6.7') L ~(x ) ~ 0
(adf -I g)
0
(6.15a) LgLf~ = .. . = L g L ~ = 0
is e q u i v a l e n t to the set of c o n d i t i o n s
(6.15b) L ~ = ... = L k ~ = 0
(ad~g) (adfg)
and b o t h imply
L[ f,g] m = L f L g m - L g L f ~ = -LgLf~
Suppose (6.15a) and (6.15b) true for some k and (6.16) true for
all i,j such that i+j = k+1. The latter yields, in p a r t i c u l a r ,
k+l k+l
L ik+1 ~ = (-I) LgLf
(aaf g)
So t h a t L g L k+1
f ~ = 0 if a n d o n l y if L. Ik+I , = 0. A s s u m e either one
of these conditions holds. Then [aef gj
229
2 2 + (_i) 3L 3
= (-I) LfL, ~k-1 ,Lf~ k Lf~
~aaf g) (ad_-Ig)
k+IL L L k+1 3L 3
= (-I) f g f ~ + (-I) (adk+1 L f ~
f g;
3
= (-1)3L(adk_Ig)Lf~ = - . .
and therefore that (6.16) is true for all i,j such that i+j = k+2.
From (6.15) and (6.16) the statement follows immediately. []
(6.17) Remark. We have proved, by the way, that either one of the two
equivalent sets of conditions (6.15) imply
L L~ = (-1)JL
(ad~g) r (ad~+Jg)~
for all i,j such that i+j < k+1. This fact will be used in the sequel.D
n-2
(6.19) A = sp{g, adfg,...,adf g}
is involutive in a neighborhood U of x .
230
(6.6") i ) (x) = 0
( d9, adfg
n-1
(6.7") (d~,adf g } (x ) / 0
i > = L g) L J9 =
<dL~, adfg (adf
= (-1)JL(adf+Jg)9 = (-1)J(dg,adf+Jg }
for all i+j < n-1. Therefore, using again (6.6") and (6.7") we h a v e
ad g>(x) = 0
for all i,j such that i+j < n-2 and all x E V and
d~(x )
dLfg(x )
(6.21) (g(x O) adfg(x)...ad~-Ig(x)) =
dL~-l~(x )
231
d L ~ - 1 ,g )(x o) <dLn-1
f ~,adfg ) (x) ... ( d L ~ - 1 9 , a d fn-1 g )(x o)
I
(6.22a) 6 (x)
L L n-1 (x)
g f
and
n
-Lf~(x)
(6.22b) e(x) =
LgL~-1~(x)
z (X)l = Lfi-1~(x)
= f(x) + g(x)u
y = ~(x)
n-1
F : x I )col(~(x),Lf~(x) ..... L f ~(x))
n-1 i i
A = 0 (sp{dLfg}) = 0
i=O
(6.24) z = F ( x ) = 0
(6.25) f(x ) + g ( x ) ~ ( x O) = 0
z 1 ( x ) = ~(x ) = 0
L fn ~ ( x ) (dL f -I ~(x o ) , f ( x o) )
(x ) . . . . . 0
L L n - 1 ~ ( x ) L L n-1
g f g f ~(x)
= f(x)
(6.26) f(x ) = c g ( x )
( d L ~ - 1 ~ ( x ) , f ( x ) )
~ ( x ) . . . . c
L g L ~ - I ? ( x )
as e x p e c t e d .
In this case, the i n i t i a l state x is n o t an e q u i l i b r i u m state
for the o r i g i n a l system, b u t an a m a y be f o u n d such that x O is an
equilibrium state for the s y s t e m
= f(x) + g(x)a(x)
In s u m m a r y , we h a v e the f o l l o w i n g result.
f(x ) 6 s p { g ( x ) } []
~ = (A + bk) z + bv []
235
G O = s p { g I ..... gm }
G i = Gi_ I + [ f , G i _ I]
k
Proof. S u p p o s e G i = sp{adfg. : 0 < k < i, I < j < m}. S u p p o s e that at
o k 1 ] kC -- --
x some v e c t o r s aaf g J 1 ' ' ' ' ' a d f gJr are l i n e a r l y i n d e p e n d e n t and
span G i(x ). T h e n on a n e i g h b o r h o o d U of x a n y v e c t o r f i e l d T in G.
r k i
may be w r i t t e n as =~=I [ c adfagj , w i t h c a e C ~ ( U ) . T h e n [ f,T] =
r k +I k k +I
=~=I
[ (c~adf~ gJa+(Lfc )adf~gj ) . Therefore, on U, G i + I = s p { a d f ~ gj ,
k
a k
adf g_j~: I <_ ~ <_ r}. Since, by c o n s t r u c t i o n , all adfgj , 0 _< k _< i+I
G i + I (x)
dim ~ - - - independent of x
l
(6.31a) ~0 = d i m G O
S.
(6.31b) V i = dim ~ i > I
Gi_ I
236
ui --> ~i+I
for all i > 0. Let v., denote the last nonzero element in the sequence
l
{~i : i ~ 0}. If
dim G = n
l
then
v 0 + v1+...+vi~ = n
and
v i + I _< v i . []
m 0 = vi
m o + m I = ~i~_I
m0 > 0
(6.35) d i m G.*
i -i = i m 0 + . . . + 2 m i _ 2 + m i _ 1
~0 = m
and
for all x 6 U, w h e r e
~, =~(~ ~ o)
o ) 6'=6(
Step (I): B y a s s u m p t i o n
d i m G. ~ = n
1
|
and dim G~_I_ = m 0 > 0. M o r e o v e r , Gi,_1 is a s s u m e d to be involutive.
Then, by Frobenius theorem, we k n o w that there exist a neighborhood
U I of x a n d m 0 f u n c t i o n s h01,...,h0m defined on U I, w h o s e differ-
0
entials span Gi~_1(x) at all x U I. In p a r t i c u l a r ,
for all I <__ j .< m ., I .< i . < m .0, 0 < e < i -I and all x U I. M o r e o v e r ,
the d i f f e r e n t i a l s dh01 (x), .... d h 0 m 0 ( X ) are linearly independent at all
xEU I.
We claim that the m 0 m m a t r i x
M0 = {m ij
(0) (x) } = {< d h 0 i , a d fig j (x)}
m0
i
< ~ e i d h 0 i , a d f g j )(x) = 0
i=I
m0
(6.39) < Z cidh0i,adfg j >(~) = o
i=I
m0
( y~ c i d h 0 i ( x ) , v } = 0
i=I
m 0
for all v 6 Gi~(x). S i n c e d i m Gi~ = n, then Z c i d h 0 i ( 2 ) m u s t be a zero
i=I
covector, but since dh01 ( x ) , . . . , d h 0 m 0 ( X ) are i n d e p e n d e n t , then
= 0.
c I = ... = Cm0
dim G~,_2 = 2m 0 + m I
We c l a i m t h a t d L f h 0 1 , . . . , d L f h 0 m 0 are s u c h that
(6.40) {dh01(x),...,dh0m0(X),dLfh01(x),...,dLfh0m0(X)}
m 0 ,
(i=I
~ (clidh0i + c 2 i d L f h 0 i ) ' a d ~ -Igj )(~) = 0
(6.43) {dh01 (x) ..... dh0m0 (x) ,dL~01 (x) .... ;dL~0m0 (x) ,d~ I (x) ..... ~hm I (x) }
for all I <_ j .< m,. I .< i .< m .I, 0 < ~ < i -2 and all x E U 2.
We c l a i m that the (m0+m I ) xm m a t r i x
I:i)
where M 0 is as b e f o r e and M I d e f i n e d as
MI = tmijr
(I) (x)} = { ( d h l i , a d f -Igj >(x)}
m0 , m I
<i=I
[ c0idh0i(x)'adf g j ( x ) > + ( i = i[ c l i d h l i ( x ) ' a d f -Igj(x) > = 0
m0 mI ,
(6.44) < ~ c0idLfh0i(~) + ~ Clidhli(~),ad~ -Igj(~)) = 0
i=I i=I
m0 mI
The c o v e c t o r [ c0idLfh0i(x) + [ Clidhli(x) annihilates, as we have
i=I i=I
seen before, all ad~gj(x), e < i~-2, I ~ j ~ m, but (6.44) tells us
that it also annihilates " i ~ I gj (x) . Thus,
aaf this covector annihilates
all vectors in Gi*_1.
From the previous discussion, we conclude that this covector must
belong to s p a n { d h 0 1 ( x ) , . . . , d h 0 m 0 ( X ) } , but this is a contradiction,
because the covectors (6.43) are linearly independent Therefore, the
c0i's and c11.'s of (6.44) must be zero.
Eventually, with this procedure we end up with a set of functions
i - i -I h
h01,...,h0m0,Lfh01,...,Lfh0m0,...,Lf lh01,...,L f 0m 0
i i -2 h
h11,...,hlm1,...,Lf -2h11,...,L f lm I
(6.45)
hi*-l,1,''',hi~-1,mi, i
i m 0 + (i - 1 ) m 1 + . . . + 2 m i , _ 2 + m i , _ l = n-m
- the vlxm m a t r i x
M0
Mi#_ I
where M. is m . x m and
l l
mj
(i) (x) = ( dhi,ad f -igj) (x)
~i(x)
Lf I ~i(x)
o
qualifies as a local diffeomorphism around x .
Moreover, by c o n s t r u c t i o n ,
(6.46) ( d L f ~ i , g j >Cx) = 0
is n o n s i n g u l a r at x = x O.
m
(6.48a) = f(x) + ~ gi(x)ui
i=1
243
6.48b) Y i = ~i (x) I ~ i ~ m
is s u c h that:
<1
Lf ~l(X)
A(x)a(x) = -
Lfm~m(X)
A(x) B(x) = I
0 1 0 ... 0 0 0
0 0 1 ... 0 0 0
... .
~i = vi
0 0 0 ... 0 I 0
0 0 0 ... 0 0 I
Yi = ( I 0 0 ... 0 0)~ i
= f (x)
y = h(x)
with y 6 ~.
Suppose there exists a coordinates transformation z = F(x) under
which the v e c t o r field f and the output map h become respectively
F ~ f o F -I (z) = A z + K(cz)
h o F -I (z) = c z
= (A+kc) ~ - k y + K (y)
e = ~ - z = ~ - F(x)
6 = (A + k c ) e
(7.1) hoF-1(z) = cz
(7.2) F ~ f o F -I (z) - A z = K ( c z )
n-1
(7.3) N k e r ( c A i) = {0}. []
i=0
0 0 ... 0 0
1 0 ... 0 0
(7.6) T (A+kc) T -I = c T -I = (0 0 . .. 0 1)
0 0 ... I 0
z = F(x) = TF(x)
K(y) = T(K(y)-ky)
where y E h(U).
246
0 0 ... 0 0
1 0 ... 0 0
. o ... .
0 0 ... 1 0
~z I
3 ~ f(x) = k I (z n(x))
~z 2
~x f(x) = z1(x)+k2(Zn(X))
(7.8)
~Z
n
~x f(x) = Zn_ 1 ( x ) + k n ( z n ( x ) )
~Zn_ I ~k n ~z n
L h(x) - 3x f(x) + ((~--~--)y=Zn)~-~ - f(x)
~k ~z
= Zn_2(x) + ( ( ~ - n) y = Z n ) ~ - X -n- f ( x ) + kn_ I (Zn (x))
247
where
~k 3k
- _ n + n
kn-1(Zn'Zn-1) ~z Zn-1 ~ z - kn(Zn) + kn-1(Zn)
n n
Proceeding in this way one obtains for each Lfi(x), for 2 < i < n-l,
an expression of the form
~h ~h
0 0 ... 0 I
~x
DLfh ~Lfh
O 0 ...
~x Sz
~~z
= 0 Q I F,(X)
~L~-lh aL~-lh
1 * ... *
~x ~z
= L -n-2h(x) = 0
LTh(X) = L~[Lfh(x) = ... TLf
LTLf-lh(x) = I
for all x 6 U.
As a matter of fact, one only needs to solve the set of equations
dh (x)
dLfh (x)
(7.9) ... T(X) =
n-2
dLf h(x)
dLf- lh (x)
248
(7.11) ~z = (I:
~-~ -adfi " "" (-1)
n-1 ~n-1~)o(Z)
aaf
-I
f(x) = (F,) (Az + K(cz))oF(x) =
lk+I 0 = [f,(-1)k#,(
aaf )o-I] =
= (-I)k+1~,(~Z-~k+----~)
0 - I
249
(-1)kL aafu
.k~h(x) ~h ~*(~z
= ~-x _ _ k ~ 1 ) o -1 (x) =
L =k~h(x) = 0
~t~/fU
(-1)n-IL ~n-1~h(x) = I
~f
LeLkfh (x) = 0
for all 0 < k < n-2 and (see also Remark (6.17)
LoL~-Ih(x) = I
dLf-lh (x )
250
By d e f i n i t i o n , the m a p p i n g # is such t h a t
so t h a t
(_i)k+1 ~ _ ~k+l
~Zk+ 2 = # , a a f ToF -I (z) = F~[ f,ad k T] oF -I (z) =
1
for i ~ k+2
z~TT+l =
3fk+2
= 1
3Zk+ 1
k 1 ( z n)
z I + k 2 ( z n)
F ~ f o F -I (X) =
Zn_ I + k n ( Z n)
L .k h = 0
aafT
n-1
L _n-1 h = (-1)
aaf T
we d e d u c e that
B h o F -I ~hoF -I
- 0
~z I -'- 8Zn_ I
and t h a t
~hoF -I
- I
~z
n
~x
(7.16) ~z i ~i(x(z))
that x(z ) = x , if a n d o n l y if
(7.17) [ T i , T j] = 0
A i = sp{Y i}
Then, the c o l l e c t i o n of d i s t r i b u t i o n s A 1 , . . . , A n is i n d e p e n d e n t , s p a n s
the t a n g e n t s p a c e a n d is s i m u l t a n e o u s l y i n t e g r a b l e b e c a u s e of (7.17)
(see T h e o r e m I. ( 3 . 1 2 ) ) . A s a c o n s e q u e n c e , we m a y f i n d a c o o r d i n a t e c h a r t
(U,~), such that z O = ~(x ) a n d
}
~i(x) = sp{ ( ~ i ) x
~9i
~--~. ci(~i) = I
The c o m p o s e d function
is c l e a r l y such t h a t ~(x ) = z a n d
~*~i(x) = ( ~ 7 ) o~(x)
l
-1
Thus x = ~ (z) s o l v e s the p . d . e . (7.16). []
(7.19) [adfT,adJT] = 0
k
[ T,adfT] = 0
for all k = 1 , 3 , . . . , 2 n - I . []
In s u m m a r y one m a y p r o c e e d as f o l l o w s in o r d e r to o b t a i n an ob-
server with linear (and s p e c t r a l l y assignable) error dynamics. If
condition (i) holds, one finds first the v e c t o r field solving the
equation (7.9). If a l s o c o n d i t i o n (ii) holds, one solves the p . d . e .
(7.11) and finds a function , d e f i n e d in a n e i g h b o r h o o d V of z ,
s u c h t h a t ~(z ) = x . T h e n one sets F = -I. E v e n t u a l l y , one c o m p u t e s
the m a p p i n g K as
k 1 ( z n) 0
k2(z n ) zI
K(Zn) = = F,foF-I(z) -
k n ( Z n) Zn- I
= (A+kc) ~ - k y + K (y)
Example. A t y p i c a l e x a m p l e o f a f u n c t i o n w h i c h is C ~ b u t n o t a n a l y t i c
is the function f: ~ ~ defined by
f(x) = 0 if x < 0
I
f(x) = exp(- ~) if x > 0 []
A mapping F: A ~ m is a c o l l e c t i o n (fl,...~fm) of f u n c t i o n s
fi: A ~. The mapping F is C if a l l fl~ s are C
Let U C ~n and V C ~n be o p e n sets. A mapping F: U ~ V is a d i f -
feomorphism if is b i j e c t i v e (i.e. one-to-one a n d onto) and both F and
F -I are o f c l a s s C ~. The jaaobian matrix o f F at a p o i n t x is the
matrix
~fl ~fl
~Xl "--
3F
~x
3f 3f
n n
~x I "'" ~x n
The value 8F a t a p o i n t
of ~-~ x = x o is sometimes denoted 8F
(~-x) o
x
Theorem. (Inverse function theorem). L e t A b e an o p e n set of ~n and
F: A ~ n a C ~ m a p p i n g If (~F)
~ xO is n o n s i n g u l a r at some x o C A, then
there exists an open neighborhood U of x in A s u c h t h a t V = F(U) is
open in ~ n and the r e s t r i c t i o n o f F to U is a d i f f e o m o r p h i s m o n t o V.
255
F = H - I o P k o G -I
F ( x , y ) = 0
~fl 5fl
~Yl " " "
9F ,oQ
~y
9f ~f
n n
~Yl "'" ~Yn
for all x E A
rank M ( x O) = k
M(x)G(x) = b(x)
for all x E U.
In o t h e r words, the equation
M(x)y = b(x)
set o f S I is an o p e n s e t o f S 2. T h e m a p p i n g F is an homeomorphism ~f
is a b i j e c t i o n and both continuous and open.
If F is a n h o m e o m o r p h i s m , the inverse mapping F -I is a l s o an
homeomorphism.
Two topological spaces S1,S 2 such that there is a n h o m e o m o r p h i s m
F:S I ~ S 2 a r e s a i d to be homeomorphic.
A subset U of a topological space is s a i d to b e closed if its
complement U in S is open. It is e a s y to s e e that the intersection of
any number of c l o s e d s e t s is c l o s e d , the union of any finite number of
closed sets is c l o s e d , and both S and ~ are closed.
If S O is a s u b s e t of a topological space S, there is a u n i q u e
open set, noted int(So) and called the interior o f S o , W h i c h is c o n -
tained in S O a n d c o n t a i n s any other open set contained in S o . A s a
matter of fact, int(So) is the u n i o n of all open sets c o n t a i n e d in S o .
Likewise, there is a u n i q u e closed set, noted cl(So) and called the
closure o f So, w h i c h c o n t a i n s S O a n d is c o n t a i n e d in a n y o t h e r closed
set which contains So. Actually, c l ( S o) is the intersection o f all
closed sets which contain So.
A subset o f S is s a i d to b e dense in S i f its c l o s u r e c o i n c i d e s
with S.
If S I and S 2 are topological spaces, then the c a r t e s i a n product
SIS 2 can be given a topology taking as a basis the c o l l e c t i o n of
all subsets of the form UIxU 2 , with U I a basic open set of SI and U2
a basic open set o f S 2. T h i s topology o n S I S 2 is s o m e t i m e s called
the product topology.
I f S is a t o p o l o g i c a l space and S 1 a subset o f S, then S I can
be given a topology taking as o p e n sets the subsets of the form
$I A U, w i t h U any open set in S. T h i s topology o n S I is s o m e t i m e s
called the subset topology.
L e t F: S I ~ S 2 be a continuous mapping of topological spaces,
and l e t F ( S I) d e n o t e the image of F. C l e a r l y , F ( S I) w i t h the subset
topology is a t o p o l o g i c a l space . Since F is c o n t i n u o u s , the inverse
image of a n y open s e t o f F ( S I) is an o p e n s e t o f S I. H o w e v e r , not all
open sets of S I are taken onto open sets ofF(St). In o t h e r w o r d s , t h e
mapping F': S I ~ F ( S I) d e f i n e d b y F' (p) = F(p) is c o n t i n u o u s but not
necessarily open. The s e t F ( S I) c a n b e g i v e n another topology, taking
as o p e n sets in F ( S I) the images of open sets in S I. It is e a s i l y
seen that this new topology, sometimes called the induced topology,
contains the subset topology (i.e. any set which is o p e n in the subset
topology is o p e n also in the induced topology), and that the m a p p i n g
F' is n o w o p e n . If F is an i n j e c t i o n , t h e n S I a n d F(SI) endowed with
258
3. S m o o t h manifolds
It is n o t p o s s i b l e that an o p e n subset U of ~ n be h o m e o m o r p h i c
to an o p e n subset V of ~ m , if n ~ m (Brouwer's theorem on invariance
of d o m a i n ) . Therefore, the dimension of a l o c a l l y Euclidean space is
a well-defined object.
A coordinate chart on a m a n i f o l d N is a p a i r ( U , ~ ) , where U is an
open set of N a n d ~ a homeomorohism of U o n t o an o p e n s e t of ~ n . s o m e -
times ~ is r e p r e s e n t e d as a set (~1;...,~n) , and ~i: U ~ is c a l l e d
the i - t h coordinate function. If p E U, the n - t u p l e of real numbers
(~l(p),...,~n(p)) is c a l l e d the set of local coordinates of p in the
coordinate chart (U,~). A coordinate chart (U,~) is c a l l e d a cubic
coordinate chart if ~(U) is an o p e n cube about the origin in ~ n . If
p E U and ~(p) = 0, then the coordinate chart is s a i d to be centered
at p.
Let (U,~) and (V,~) be two c o o r d i n a t e charts on a m a n i f o l d N,
with U N V ~ @. L e t (~I .... ,~n ) be the s e t of c o o r d i n a t e functions
associated with the m a p p i n g ~. The homeomorphism
-I
+o~ : ~(U nv) ~ ~(u n v)
-I
tion ~o~ c a n be r e p r e s e n t e d in the form
Yn Y n ( X l .... ,x n)
-I
and the inverse transformation ~o~ in the form
x = x(y)
X2
Jill :
:
3 ~ 4 1 ", _
. it--- -_
Y1
TwO coordinate charts (U,9) and (V,%) a r e C -Compatible if, w h e -
-I
never U n V ~ ~, the c o o r d i n a t e transformation o~ is a d i f f e o m o r -
phism, i.e. i f y(x) a n d x(y) are both C~ maps.
A C ~ atlas on a m a n i f o l d N is a c o l l e c t i o n A = {(Ui,~i)}iE I of
pairwise C~-compatible coordinate charts, with the p r o p e r t y that
U U. = N. A n a t l a s is complete if n o t properly contained in a n y o t h e r
i6I i
atlas
Example. A n y o p e n s e t U of ~ n is a s m o o t h m a n i f o l d , of dimension n.
For, consider the a t l a s A consisting of the (single) coordinate chart
(u, i d e n t i t y map on U) and let A ~ denote the unique complete atlas
containing A. In p a r t i c u l a r , A n is a s m o o t h manifold.
(x) = x
3
(x) = x
and the two charts are not compatible. Therefore the u n i q u e complete
atlas A which includes (R,~) and the unique complete atlas ~ which
includes dR,~) are different. This means that the same manifold N may
be considered as a s u b s t r a t e of two different objects (two smooth
manifolds), one arising with the a t l a s A and the o t h e r with the atlas
A~ . []
Example. L e t U b e a n o p e n s e t o f ~ m a n d l e t 1 1 , . . . , I m _ n be r e a l - v a l u e d
C~ functions defined on U. L e t N d e n o t e the (closed) subset of U on
which all functions 11,...,Im_ n vanish, i.e. let
211 211
~x I "'" ~x m
m-n m-n
~x I "'" ~x m
al I aX I
aXn+ 1 "'" ax m
~l ax
m-n m-n
~Xn+ 1 "'" ~Xm
ax I ax I
U' = U C~ N'
~' = r e s t r i c t i o n of ~ to U'
262
Example. L e t M a n d N b e s m o o t h m a n i f o l d s , of d i m e n s i o n m a n d n. Then
the cartesian product MN is a s m o o t h manifold. The topology of MN is
the p r o d u c t topology. If (U,9) and (V,) are coordinate charts of M
a n d N, the p a i r (UxV, (~,)) is a c o o r d i n a t e chart o f MxN. The dimension
of MN is c l e a r l y m+n.
An important example of this type of m a n i f o l d is the torus
T 2 = S I x S I, the cartesian product of two circles. []
= ~o -I : ~(U) -*:~
= o F o ~ -I
Yl f1(xl ..... x n )
y = = F(x)
~, = ~,oFo~ ,-1 =
= ,o-1ooFo -1 o~o~' -1 =
Definition. L e t N a n d M b e s m o o t h m a n i f o l d s , b o t h of d i m e n s i o n n.
A m a p p i n g F : N ~ M is & diffeomorphsim if F is b i j e c t i v e a n d b o t h F
a n d F -I are s m o o t h m a p p i n g s . T w o m a n i f o l d s N a n d M are diffeomorphic
if t h e r e exists a diffeomorphism F : N ~ M. []
~fl ~fl
Sx I "-- Sx n
~f ~f
m m
~x I "'" ~x n
4. S u b m a n i f o l d s
x1(t) = at-sin t
x2(t) = cos t
and, then,
i a - cos t
rank(F) = rank
- sin t
x2(t) = sin t
265
X2 X
2
\ -<~--~'~~X 1
a=! O<a<l
dx 1 2 cos2t
rank(F) = rank =
= I
dx 2
q-6 cos t
F(tl) = F ( t 2) ~ t I = t 2
xl (t) = C O S 2~t
x3(t) = t
-I
V = F' (U) , % = ~o (F')
U' = U N M'
5. T a n g e n t vectors
(al+bx) (q) = a l ( q ) + b x ( q )
for all q in a n e i g h b o r h o o d o f p.
Remark. The set TpN forms a v e c t o r space over the field ~ under the
rules of scalar m u l t i p l i c a t i o n and a d d i t i o n d e f i n e d in the f o l l o w i n g
way. If v l , v 2 are t a n g e n t vectors and Cl,C 2 real numbers, clv1+c2v2
is a new tangent v e c t o r w h i c h takes the f u n c t i o n I E Ca(p) into the
real n u m b e r
d e f i n e d in the f o l l o w i n g way
n
v : [ v i(a--~)
i=I i p
n n 3 (~io~ -I)
v(gi) = ) ~ivj(~'~j
j p(~i ) = ( [IVj
j 3xj )x=~(p) = v.l
because ~ i ~
-I(Xl,...,x n) = x i. Thus the real number v i coincides with
the value of v at ~i ' the i-th coordinate function.
A change of coordinates around p clearly induces a change of basis
in TpN. The computations involved are the following ones. Let (U,~)
and (V,) be coordinate charts around p. Let {( )p, .... ( ~3 }p} denote
the natural basis of T N induced by the coordinate chart (V,). Then
P
n 3 3 (~jo-1)
j=1 3Yi )Y= (P)
In other words
3(~jo -1)
3Y i
is the element on the j-th row and i-th column of the jacobian matrix
of the coordinate transformation
x = x{y)
n n
! )p
v = v i =
i=I ~i p i=I
then
ax 1 ~x I
v1 wI
3Yl " " aYn
~x ax
vn n n wn
3y I " " aYn
F~ : TpN -+ T F ( p ) M
(F~(v))(~) = V(~oF)
p(loF) = (~ (IF~-I))
(F~(~i). p) (I) = (~.)i x=~(P )
272
m ~ ~(~joFo~ -I)
j=1 ((~ )q (i)) ( ~xi )x=? (p)
In o t h e r w o r d s
m a(+joFo~ -I)
F. (~--~i)P = j=1 axi )x=~(p) a~j)q
-I
Now, recall that %oFo~ is an e x p r e s s i o n of F in local coordinates
Then, the q u a n t i t y
a(%joFo~ -I)
~x1
is the e l e m e n t on the j-th row and i-th column of the jacobian matrix
of the m a p p i n g expressing F in local coordinates. Using again
F I (X I ..... x n)
Fm(X I , .. ,x n)
-I
to d e n o t e ~oFo~ , one has simply
F% ( ~ i ) P ~ aF.
j=1 i
n m
v = v i w = w
i=I i p i=I i
then
aF I ~F I
wI vI
~Xl --- ~Xn
=
~F m ~F m
W m Vn
~x I " .. ~x n
273
(GoF). = G,F.
Example. The tangent vectors on I~n. Let ]Rn be e q u i p p e d with the "natur-
al" complete atlas already c o n s i d e r e d in p r e v i o u s e x a m p l e s (i.e. the
one i n c l u d i n g the chart ~n, i d e n t i t y map on ~ n ) ) . Then, if v is a
tangent v e c t o r at a p o i n t x and I a s m o o t h f u n c t i o n
n ~ n 3t
v(l) = i-~-Iv i ( ~ i ) x ( l ) = i=I[ (~
X ) V.l
SO, V(1) is just the value of the derivative of I along the direction
of the vector
col(v I ..... v n )
at the p o i n t x.
~' = (~1,...,~n)
n do.
o,(~t) t = ~ (~) (DTi)O(to)
o i=1 tO '
do I do n )
col ((-~-) to ..... (-a~) to
F(xl,x2) = (xl,x2,h(xl,x2))
Now,
3 3F 1 3 (3h ,
F , ( T X 1)x =1~i.=(-~i) (~-~i)F(x) = (Txl)F(x) + ~ i ) (-~3)F(x)
) = 3 + ~h (~x3)
o o o
This tangent space to F(~ 2) at some point (x~,x2,h(Xl,X2)) is the set
of tangent vectors whose expressions in local c o o r d i n a t e s are of the
form
V = B
3h 8h
~h ~h
~,~ being real numbers and ~ ,~ being e v a l u a t e d at x I = x~ and
o
x 2 = x 2. F r o m this point of view, we see that the notion of tangent
space to a 2-dimensional m a n i f o l d may be identified with the geometric
notion of tangent plane to a surface in a E u c l i d e a n space. []
One may define objects dual to the ones c o n s i d e r e d so far.
(clv1+c2v2,v)= c I ( v I ,v ) + c 2 (v 2 ,v )
is c a l l e d a dual basis,
If V a n d W are v e c t o r spaces, F : V ~ W a linear mapping and wEW,
v E V, the m a p p i n g F : W ~ V defined by
I, : TpN ~ Tl(p)~
natural basis
of T N i n d u c e d by the coordinate chart (U,~). The image
P
under l~ of a v e c t o r
n
v = ~ vi( )
i=I ~ p
is the v e c t o r
n
~, (v) = ( 2 ~ .~ vi) (~t) t
i=I 1
{ (dl)p,V)=
21
( 8x I "'"
81
~n )
vl)
vn
Remark. N o t e a l s o t h a t the v a l u e at I of a t a n g e n t v e c t o r v is e q u a l to
the v a l u e at v of the t a n g e n t c o v e c t o r (dl)p , i.e.
v(1) = ( (dl)p,V } []
8 ( ~ i o ~ -I) 8x.
< (d~i)p' ( 3 )p > = ( )p ( ~i) 8x . 8x.3 i3
, n ,
v = ~ vi(d~i) p ,
i=I
,
the real numbers v.],...,v n are such that
v i = <v ,( )p
n
v = [ v i(~-~i )p
i=1
v i = < (d~i)p,V) .
6. V e c t o r fields
n
f(q) = ~ fi(q) ( ~ ) q
i=1
n
,2)
f(q) = [ f~ (q) 'D--~i q
i=I l
(~i): P ~-+(~.)p
n
f(q) = [ fi (q) ~-~-~)
i=I (~i q
d
o~(~-{) t = f(o(t))
Moreover
d i
o~(~t)t = ~ ~(~i)~(t)
i=I
do
1
dt - fi(ol (t} ..... On(t))
d
to i n d i c a t e the image of (~)t under the d i f f e r e n t i a l ~, at t.
The following theorem contains all relevant informations about
the p r o p e r t i e s of integral curves of v e c t o r fields.
%:W~N
(i) ~(0,p) = p,
(ii) for e a c h p the m a p p i n g Op:Ip ~ N defined by
Op(t) = #(t,p)
is an i n t e g r a l curve of f,
(iii) if ~ : (tl,t 2) ~ N is a n o t h e r integral curve of f satisfying the
condition ~(0) = p, then (tl,t2) C Ip and the restriction of Op
to (tl,t 2) coincides with ~,
(iv) ~(S,~(t,p)) = (s+t,p) whenever both sides are defined,
(v) whenever #(t,p) is d e f i n e d , there exists an o p e n neighborhood U
of p s u c h that the m a p p i n g ~t: U ~ N defined by
Ct(q) = #(t,q)
~1 = ~-t
f(x) = (x 2 + I ) ( ~ x ) x
(t) = do (~x) x =
(~-~) (02(t)+i) (~x) x
so
d~ - o 2 + I
dt
0(t) = t g ( t + t g - 1 (x))
_ ~ < t + t g -I (x ) < ~_
2 2
Thus W is t h e set
W = { ( t , x O) : t e (-~-tg-1(x), ~ tg-1(x))}
(i.e. (Lfl) (p) is the value at I of the tangent vector f(p) at p). []
fl
) -
(Lfl) (x I .... ,Xn) = ~x I ... ~x n
fn
If fl,f2 are v e c t o r fields and I a real-valued function, we de-
note
LflLf 2 I = Lf 1 (Lf21)
The set of all smooth vector fields on a m a n i f o l d N is d e n o t e d by
the symbol V(N). This set is a Vector space o v e r ~ since if f, g are
vector fields and a,b are real numbers, their linear combination af+bg
is a v e c t o r field defined by
(af+bg) (p) = a f ( p ) + b g ( p )
(af+bg) (p) = a ( p ) f ( p ) + b ( p ) f ( p )
and this gives V(N) the structure of a module o v e r the ring, denoted
C~(N), of all smooth real-valued functions defined on N. The set V(N)
c a n be q i v e n , h o w e v e r , a more interesting algebraic structure in t h i s way.
Iv,w! = -[w,v]
283
The set V(N) forms a Lie algebra with the vector space structure
already discussed and a product [ .,.] defined in the following way.
If f and g are vector fields, [f,g] is a new vector field whose value
at p, a tangent vector in TpN, maps C~(p) into ~ according to the rule
In other words, [ f,g] (p) takes I into the real number (LfLgl) (p)-
+(LgLfl) (p). Note that one may write more simply
L[ f,g] I = LfLgl-LgLfl
Theorem. V(M) with the product [ f,g] thus defined is a Lie algebra. []
The product [ f,g] is called the Lie bracket of the two vector
fields f and g.
Remark. If f,g are smooth vector fields and I,X smooth real-valued
functions, then
Note that /,x,Lfy,Lgl are elements of C~(N), and g,f,[ f,g] elements of
V(N). []
The reader may easily find that the expression of [ f,g] in local
coordinates is given by the n - v e c t o r
~gn
~x I "'"
~gn
~x n ill ~f
n
i
~f
n
gl1
gn
=~f _
~g
~f
284
If, in p a r t i c u l a r , N = ~n and
f(x) = Ax , g(x) = Bx
then
Then also
be F-related if
Fef = goF
(Lfg) (p) = l i m l[
t ( ~ -f t ) ~ g (#~(p)) - g ( P )| " []
t~0
Lfg = adfg
0 k k-1
Lfg = g and Lfg = Lf(Lf g)
or
W(t) = (~ft)_ . g ( ~ ( p ) )
, d ~ (t), k k
- dt
- k-
~ t=0 = Lfg(p)
= a d f g(p)
tk
k=O
286
{ ~ , f } (p) = ( ~ ( p ) , f ( p ) )
= dl []
(~ft)*~(#tf(p))
The covector field thus defined is said to be ~ - r e l a t e d to ~.
1 f * {~(p))_m ]
lim ~[ ({t) ~ ( (p)
t~ 0
exists.
lim 1[ (~f)* ( f []
t ~t (p)) -m (p) ]
t~0
+ T ~f
q-----~xf) +m
~f ~f
B~I ~n n n
~Xn "'" ~n ~x I ..
L f ( ~ , g )= ( L f ~ , g ) + (~,Lfg)
and
Lfdl = d(Lfl)
o~
Note that l,y,Lfy,<~,f > are elements of C (N) and ~,Lf~,dl elements
of V (N).
B I B L I O G R A P H I C A L NOTES
Chapter I
C h a p t e r II
C h a p t e r III
C h a p t e r IV
C o n t r o l l e d i n v a r i a n t d i s t r i b u t i o n is the n o n l i n e a r v e r s i o n of
the n o t i o n of c o n t r o l l e d i n v a r i a n t subspace, i n t r o d u c e d i n d e p e n d e n t l y
by B a s i l e - M a r r o (1969) and W o n h a m - M o r s e (1970). For a c o m p r e h e n s i v e
p r e s e n t a t i o n of the theory of m u l t i v a r i b a l e linear control systems,
the reader is r e f e r r e d to the c l a s s i c a l treatise of W o n h a m (1979).
C o n t r o l l e d i n v a r i a n t d i s t r i b u t i o n s were i n t r o d u c e d i n d e p e n d e n t l y by
H i r s c h o r n (1981) and, in the more general form d e s c r i b e d here, by
Isidori et al. (1981a). The p r o o f of Lemma (1.10) may be found in
H i r s c h o r n (1981), Isidori et al. (1981b) and ~ i j m e i j e r (1981).
Lemma (3.6) is due to Claude (1982); the special case where the
m a t r i x A(x) has rank was dealt w i t h in Isidori et al. (1981a). The
a l g o r i t h m (3.17) has been s u g g e s t e d by K r e n e r (1985).
E a r l y results on n o n l i n e a r d e c o u p l i n g and n o n i n t e r a c t i n g con-
trol were given by S i n g h - R u g h (1972) and F r e u n d (1975). The p o s s i b i -
lity of solving d e c o u p l i n g p r o b l e m s in a d i f f e r e n t i a l - g e o m e t r i c
s e t t i n g was d e s c r i b e d by H i r s c h o r n (1981) and Isidori et al. (1981a).
The n o t i o n of c o n t r o l l a b i l i t y d i s t r i b u t i o n , the n o n l i n e a r version of
the one of c o n t r o l l a b i l i t y subspace, was i n t r o d u c e d by I s i d o r i - K r e n e r
(1982) and N i j m e i j e r (1982a). The solution of n o n i n t e r a c t i n g control
p r o b l e m s via c o n t r o l l a b i l i t y d i s t r i b u t i o n s is d e s c r i b e d in N i j m e i j e r
(1983).
C o n t r o l l e d i n v a r i a n c e for general n o n l i n e a r control systems (i.e.
systems where the control does not enter linearly) is s t u d i e d in
N i j m e i j e r - V a n der S c h a f t (1983). C o n t r o l l e d i n v a r i a n c e for d i s c r e t e -
time n o n l i n e a r s y s t e m is s t u d i e d in Grizzle (1985) and M o n a c o - N o r m a n d
Cyrot (1985).
Chapter V
Appendix
G. BASILE, G. M A R R O
(1969) C o n t r o l l e d and c o n d i t i o n e d i n v a r i a n t subspaces in linear systems
theory, J. Optimiz. Th. & Appl. 3, pp. 306-315.
D. BESTLE, M. ZEITZ
(1983) C a n o n i c a l form o b s e r v e r design for n o n - l i n e a r t i m e - v a r i a b l e
system, Int. J. Contr. 38, pp. 419-431.
W.A. BOOTHBY
(1975) "An I n t r o d u c t i o n to D i f f e r e n t i a b l e M a n i f o l d s and R i e m a n i a n n
Geometry", A c a d e m i c Press: N e w York.
R.W. B R O C K E T T
(1972a) S y s t e m theory on group m a n i f o l d s and c o s e t spaces, S I A M J.
Contr. 10, pp. 265-284.
(1972b) On the a l g e b r a i c s t r u c t u r e of b i l i n e a r systems, in "Theory
and A p p l i c a t i o n s of V a r i a b l e S t r u c t u r e Systems", R. M o h l e r and A.
R u b e r t i Eds., A c a d e m i c Press: New York, pp. 153-168.
(1976) V o l t e r r a series and g e o m e t r i c control theory, A u t o m a t i c a 12,
pp. 167-176.
(1978) F e e d b a c k i n v a r i a n t s for n o n l i n e a r systems, Preprints 6th IFAC
Congress, Helsinki, pp. 1115-1120.
P. B R U N O V S K Y
(1970) A c l a s s i f i c a t i o n of l i n e a r c o n t r o l l a b l e systems, Kybernetika
6, pp. 173-188.
W.L. CHOW
(1939) Uber systeme von l i n e a r e n p a r t i e l l e n d i f f e r e n t i a l g l e i c h u n g e n
ester ordnung, Math. Ann. 117, pp. 98-105.
D. CLAUDE
(1982) D e c o u p l i n g of n o n l i n e a r systems, Syst. & Contr. Lett. I,
pp. 242-248.
P. C R O U C H
(1981) D y n a m i c a l r e a l i z a t i o n s of finite V o l t e r r a Series, S I A M J.
Contr. Optimiz. 19, pp. 177-202.
D.L. E L L I O T T
(1970) A c o n s e q u e n c e of c o n t r o l l a b i l i t y , J. Diff. Eqs. 10, pp. 364-370.
M. FLIESS
(1974) M a t r i c e s de Hankel, J. Math. pures et appl 53, pp. 197-224.
(1981) F o n c t i o n n e l l e s c a u s a l e s non l i n e a i r e s et i n d e t e r m i n ~ e s non
c o m m u t a t i v e s , Bull. Soc. math. France 109, pp. 3-40.
(1983) R e a l i s a t i o n locale des s y s t e m e s non lineaires, a l g e b r e s de
Lie filtrees t r a n s i t i v e s et series g e n e r a t r i c e s non c o m m u t a t i v e s ,
Invent. math. 71, pp. 521-537.
M. FLIESS, M. LAMNABHI, F. L A M N A B H I - L A G A R R I G U E
(1983) An a l g e b r a i c a p p r o a c h to n o n l i n e a r f u n c t i o n a l expansions, IEEE
Trans. Circ. Syst. CAS-30, pp. 554-570.
E. F R E U N D
(1975) The structure of d e c o u p l e d n o n l i n e a r systems, Int. J. Contr.
21, pp. 651-659.
E.G. GILBERT
(1977) F u n c t i o n a l e x p a n s i o n for the r e s p o n s e of n o n l i n e a r d i f f e r e n t i a l
systems, IEEE Trans. Aut. Contr. AC-22, pp. 909-921.
J.W. GRIZZLE
(1985) C o n t r o l l e d i n v a r i a n c e for d i s c r e t e time n o n l i n e a r systems w i t h
an a p p l i c a t i o n to the d i s t u r b a n c e d e c o u p l i n g problem, to be published.
G.W. HAYNES, H. H E R M E S
(1970) N o n l i n e a r c o n t r o l l a b i l i t y via Lie theory, S I A M J. Contr. 8,
pp. 450-460.
R. H E R M A N N
(1962) The d i f f e r e n t i a l g e o m e t r y of foliations, J. Math. and Mech. 11,
pp. 302-316.
R. HERMANN, A.J. K R E N E R
(1977) N o n l i n e a r c o n t r o l l a b i l i t y and o b s e r v a b i l i t y , IEEE Trans. Aut.
Contr. AC-22, pp. 728-740.
294
R.M. H I R S C H O R N
(1979a) I n v e r t i b i l i t y of n o n l i n e a r c o n t r o l systems, S I A M J. Contr.
& Optimiz. 17, pp. 289-297.
(1979b) I n v e r t i b i l i t y of m u l t i v a r i a b l e n o n l i n e a r c o n t r o l systems,
IEEE Trans. Aut. Contr. AC-24, pp. 855-865.
(1981) ( A , B ) - i n v a r i a n t d i s t r i b u t i o n s and d i s t u r b a n c e d e c o u p l i n g of
n o n l i n e a r systems, S I A M J. Contr. & Optimiz. 19, pp. 1-19.
A. ISIDORI
(1973) D i r e c t c o n s t r u c t i o n of m i n i m a l b i l i n e a r r e a l i z a t i o n s from non-
linear i n p u t - o u t p u t maps, IEEE Trans. Aut. Contr. AC-18, pp. 626-631.
(1983) N o n l i n e a r feedback, structure at infinity and the i n p u t - o u t p u t
l i n e a r i z a t i o n problem, in " M a t h e m a t i c a l theorv of n e t w o r k s and
systems", P.A. F u h r m a n n Ed., S p r i n g e r Verlag: Berlin, DD. 473-493.
(1985) The m a t c h i n g of a p r e s c r i b e d linear i n p u t - o u t p u t b e h a v i o r in
a n o n l i n e a r system, IEEE Trans. Aut. Contr. AC-30, to appear.
A. ISIDORI, A. RUBERTI
(1984) On the s y n t h e s i s of linear i n p u t - o u t p u t r e s p o n s e s for n o n l i n e a r
systems, Syst. & Contr. Lett. 4, pp. 17-22.
B. J A K U B C Z Y K
(1980) E x i s t e n c e and u n i q u e n e s s of r e a l i z a t i o n s of n o n l i n e a r systems,
S I A M J. Contr. & Optimiz. 18, pp. 455-471.
B. JAKUBCZYK, W. R E S P O N D E K
(1980) On l i n e a r i z a t i o n of control systems, Bull. Acad. Polonaise
Sci. Set. Sci. Math. 28, pp. 517-522.
R.E. K A L M A N
(1972) K r o n e c k e r i n v a r i a n t s and feedback, in " O r d i n a r y d i f f e r e n t i a l
equations", C. W e i s s Ed., A c a d e m i c Press: New York, pp. 459-471.
A.J. KRENER
(1974) A g e n e r a l i z a t i o n of C h o w ' s T h e o r e m and the b a n g - b a n g T h e o r e m
to n o n l i n e a r control systems, S I A M J. Contr. 12, pp. 43-52.
295
C. LESJAK, A.J. K R E N E R
(1978) The e x i s t e n c e and u n i q u e n e s s of V o l t e r r a series for n o n l i n e a r
systems, IEEE Trans. Aut. Contr. AC-23, pp. 1090-1095.
C. L O B R Y
(1979) C o n t r S l a b i l i t ~ des syst~mes non lineaires, S I A M J. Contr. 8,
pp. 573-605.
M. M A L A B R E
(1984) S t r u c t u r e a l ' i n f i n i des triplets invariants: a p p l i c a t i o n a
la p u r s u i t e p a r f a i t e de modele, in "Analysis and O p t i m i z a t i o n of
Systems", A B e n s o u s s a n and J.L. Lions Eds., S p r i n g e r - V e r l a g : Berlin,
pp. 43-53.
R. M A R I N O
(1982) F e e d b a c k e q u i v a l e n c e of n o n l i n e a r systems with a p p l i c a t i o n s to
p o w e r systems equations, D.Sc. D i s s e r t a t i o n , W a s h i n g t o n University,
St. Louis.
S. MONACO, D. N O R M A N D - C Y R O T
(1983) The i m m e r s i o n under feedback of a m u l t i d i m e n s i o n a l d i s c r e t e -
time n o n l i n e a r s y s t e m into a linear system, Int. J. Contr. 38,
pp. 245-261.
(1985) I n v a r i a n t d i s t r i b u t i o n s for d i s c r e t e - t i m e n o n l i n e a r systems,
Syst. & Contr. Lett. 5, pp. 191-196.
T. N A G A N O
(1966) L i n e a r d i f f e r e n t i a l systems w i t h s i n g u l a r i t i e s and a p p l i c a t i o n s
to t r a n s i t i v e Lie algebras, J. Math. Soc. Japan 18, pp. 398-404.
296
H. N I J M E I J E R
(1981) C o n t r o l l e d i n v a r i a n c e for affine control systems, Int. J. Contr.
34, pp. 824-833.
(1982a) C o n t r o l l a b i l i t y d i s t r i b u t i o n s for n o n l i n e a r systems. Syst.
& Contr. Lett. 2, pp. 122-129.
(1982b) I n v e r t i b i l i t y of affine n o n l i n e a r control systems: a geometric
approach, Syst. & Contr. Lett. 2, pp. 163-168.
(1983) F e e d b a c k d e c o m p o s i t i o n of n o n l i n e a r control systems, IEEE Trans.
Aut. Contr. AC-28, pp. 861-862.
(1985) Right i n v e r t i b i l i t y of n o n l i n e a r control systems: a geometric
approach, to be published.
H. NIJMEIJER, J.M. S C H U M A C H E R
(1985) Zeros at i n f i n i t y for affine n o n l i n e a r control systems, IEEE
Trans. Aut. Contr. AC-30, to appear.
D. NORMAND-CYROT
(1983) Theorie et p r a t i q u e des syst~mes n o n l i n e a i r e s en temps discret,
Th6se d'Etat, U n i v e r s i t 6 de Paris Sud.
W. R E S P O N D E K
(1982) On d e c o m p o s i t i o n of n o n l i n e a r control systems, Syst. & Contr.
Lett. I, pp. 301-308.
W.J. RUGH
(1981) " N o n l i n e a r S y s t e m Theory: the V o l t e r r a - W i e n e r Approach", Johns
H o p k i n s Press: Baltimore.
(1983) A m e t h o d for c o n s t r u c t i n g m i n i m a l l i n e a r - a n a l y t i c r e a l i z a t i o n s
for p o l y n o m i a l systems, IEEE Trans. Aut. Contr. AC-28, pp. 1036-1043.
L.M. S I L V E R M A N
(1969) I n v e r s i o n of m u l t i v a r i a b l e linear systems, IEEE Trans. Aut.
Contr. AC-14, pp. 270-276.
R. S O M M E R
(1980) Control design for m u l t i v a r i a b l e n o n l i n e a r t i m e - v a r y i n g systems,
Int. J. Contr. 31, pp. 883-891.
E. S O N T A G
(1979) "Polynomial R e s p o n s e Maps", S p r i n g e r Verlag: Berlin,
297
R. SU
(1982) On the linear e q u i v a l e n t s of n o n l i n e a r systems, Syst. & Contr.
Lett. 2, pp. 48-52.
H. SuSSMANN
(1973) Orbits of families of v e c t o r fields and i n t e g r a b i l i t y of di-
stributions, Trans. A m e r i c a n Math. Soc. 180, pp. 171-188.
(1977) E x i s t e n c e and u n i q u e n e s s of m i n i m a l r e a l i z a t i o n s of n o n l i n e a r
systems, Math. Syst. Theory 10, pp. 263-284.
H. SUSSMANN, V. J U R D J E V I C
(1972) C o n t r o l l a b i l i t y of n o n l i n e a r systems, J. Diff. Eqs. 12,pp. 95-
116.
F.W. WARNER
(1970) "Foundations of d i f f e r e n t i a b l e m a n i f o l d s and Lie groups",Scott,
Foresman: Glenview.
M.H. W O N H A M
(1979) "Linear M u l t i v a r i a b l e Control: a G e o m e t r i c Approach", Berlin:
S p r i n g e r Verlag.