Documente Academic
Documente Profesional
Documente Cultură
n
k=1
x
k
y
k
, norm: x =< x, x >
1/2
,
7. distance: d(x, y) = x y.
Theorem 1. Let x, y R
n
. Then we have
1. x > 0 and x = 0 0,
2. ax = |a|x for every real a,
3. < x, y > x y (Cauchy-Schwartz inequality),
4. x +y x +y (triangle inequality).
2 Rafael Wisniewski
Proof (of 4.).
x +y
2
=
n
i=1
(x
i
+y
i
)
2
=
n
i=1
(x
i
)
2
+ 2
n
i=1
x
i
y
i
+
n
i=1
(y
i
)
2
= x
2
+ 2 < x, y > +y
2
||x||
2
+ 2||x|| ||y|| +||y||
2
= (||x|| +||y||)
2
.
AF
A. Assume x S.
Then x A for some A F. A is open that is there exists B(x) A S.
Hence x is an interior point of S.
Theorem 3. Intersection of a nite collection of open sets is an open set.
Proof. Let S =
m
k=1
A
k
, where A
k
are open sets. Suppose x S then x A
k
for all k {1, ..., m}, but A
k
is open. There is a ball B(x, r
k
) A
k
. Choose
the least among r
k
s and denote it by r. It follows B(a, r) A
k
for all
k {1, ..., m}. Thus B(x, r) S.
Nonlinear Control. Supplementary Notes to Khalils Nonlinear Systems. 3
Denition 7. A collection of subsets of X is said to be topology in X if
has the properties
1. and X ,
2. V
i
, i {1, ..., n} V
1
... V
n
,
3. If {V
.
Members of a topological space are called open sets. Closed sets are com-
plements of open sets.
Denition 8. Let S R
n
, a R
n
(the point a is not necessary in S). The
point a is said to be adherent to S if every n-ball B(a) contains at least one
point of S.
Denition 9. If S R
n
and x R
n
, x is called an accumulation point of
S if every n-ball B(x) contains at least one point of S distinct from x.
Example 1. The set of numbers of the form
1
n
, n = 1, , 2, ... has 0 as an
accumulation point of {1,
1
2
,
1
3
, ...}
Theorem 4. A subset S R
n
is closed if and only if it contains all its
adherent points.
Proof. Let S be closed and assume that x is adherent to S. We will show
by a contradiction that x S indeed. Let x S, thus x R
n
S, which is
open. Hence there is an n-ball B(x) R
n
S thus x is not adherent and we
have a contradiction.
Conversely, assume S contains all adherent points. We shall show that
R
n
S is open. Let p R
n
S, hence p / S and p is by the assumption not
adherent to S. Therefore there exists an n-ball B(p) such that B(p) S = .
We conclude B(p) R
n
S and thus R
n
S is open, Thereby S is closed.
Denition 10. The set of all adherent points of the set S is called the closure
of S and is denoted by S.
Corollary 1. A set S is closed if and only if S = S.
Denition 11. The set of all accumulation points of a set S is called the
derived set of S and it is denoted by S
.
Note that S = S S
AF
A. If F is a collection of open sets then F is called an open
covering of S.
Theorem 6 (Heine-Borel). Let F be an open covering of a closed and
bounded set S R
n
. Then a nite subcollection of F also covers S
I shall provide you with some facts.
Lindelof Covering Theorem:
Assume S R
n
and let F be a covering of S. Then there exists a count-
able subcollection of F which also covers S.
Cantor Intersection Theorem:
Let {Q
1
, Q
2
, ...} be a countable collection of nonempty sets in R
n
such
that
1. Q
k+1
Q
k
for k = 1, 2, ...,
2. Each set Q
k
is closed and Q
1
is bounded.
Then the intersection
k=1
Q
k
is closed and nonempty.
Proof (of Heine-Borel Theorem). A countable collection of F, say {I
1
, I
2
, ...},
covers S. For m N consider a collection {S
m
}
mN
of sets
S
m
=
m
k=1
I
k
and a collection {R
m
}
mN
R
m
= R
n
S
m
.
Note that each S
m
is open and R
m
is closed. We shall also dene a collection
of {Q
m
}
mN
as follows
Q
1
= S
Q
m
= S R
m
.
Note that Q
1
is bounded and closed and each Q
m
is closed. We proof the the-
orem by a contradiction and assume that each Q
m
is nonempty. Use Cantor
Intersection Theorem to conclude that
m=1
Q
m
is closed and nonempty.
Let x
m=1
Q
m
then x S and x R
n
S
m
for all m N. In
other words for all m N we have x / S
m
, but S
m=1
S
m
. This is a
contradiction.
Denition 14. A set S in R
n
is said to be compact if every open covering
of S contains a nite subcover, i.e. a nite subcollection which also covers S.
Theorem 7. Let S be a subset in R
n
. The following are equivalent:
1. S is compact.
Nonlinear Control. Supplementary Notes to Khalils Nonlinear Systems. 5
2. S is closed and bounded.
3. Every innite subset of S has an accumulation point in S.
Denition 15. A metric space is a nonempty set M together with a function
d : M M R satisfying the following. For any x, y, z M
1. d(x, x) = 0;
2. d(x, y) > 0 if x = y;
3. d(x, y) = d(y, x);
4. d(x, y) d(x, z) +d(z, y)
The function d is called the metric.
Denition 16. Let (S, d
s
) and (T, d
T
) be metric spaces and let f : S T
be a function. We say that f is continuous at p S if
> 0 > 0 such that d
T
(f(x), f(p)) < whenever d
s
(x, p) < . (1)
Denition 17. Let (S, d
s
) and (T, d
T
) be metric spaces and let f : S T
be a function. We say that f is continuous on S if it is continuous at each
point p S.
I shall provide you with some useful facts about continuous functions
Condition (1) means that f(B
S
(p, )) B
T
(f(p), ), where B
S
, B
T
are
balls in S and T respectively.
If (S, d
S
), (T, d
T
), (U, d
U
) are metric spaces and f : S T is continuous
at p and g : f(S) U is continuous at f(p) then h = g f (h(x) =
g(f(x))) is continuous at p.
Denition 18. Let f : S T be a function. The preimage of a subset
Y T is a subset of S dened by
f
1
(Y ) = {x S| f(x) Y }.
Theorem 8. Let f : S T be a function between metric spaces (S, d
S
) and
(T, d
t
). Then f is continuous if and only if for every open set Y in T, f
1
(Y )
is open in S.
Proof. Assume rst that f is continuous on S and let y be open in T. Choose
any point p f
1
(Y ). We shall prove that p is an interior point of f
1
(Y ).
Let y = f(p). Since Y is open there exists a ball B
T
(y, ) Y , but f is
continuous thus there is a ball B
S
(p, ) such that f(B
S
(p, )) B
T
(y, ). We
conclude
B
S
(p, ) f
1
(f(B
S
(p, )) f
1
(B
T
(y, )) f
1
(Y ).
Therefore p is an interior point of f
1
(Y ).
6 Rafael Wisniewski
Conversely, assume f
1
(Y ) is open for any open subsets Y in T. Choose
p S and let y = f(p). We shall prove that f is continuous at p. For any
> 0 the ball B
T
(y, ) is open in T and by the assumption f
1
(B
T
(y, ) is
open in S. But p f
1
(B
T
(y, ) therefore there is > 0 such that B
S
(p, )
f
1
(B
T
(y, )) or in other words f(B
S
(p, )) B
T
(y, ). This means that f
is continuous at p.
Theorem 9. Let (S, d
S
) and (T, d
T
) be metric spaces. If f : S T is
continuous on a compact subset X of S, then the image f(X) is a compact
subset of T.
Proof. Let F be an open covering of f(X). This means f(X)
AF
A.
The set f
1
(A) is open and {f
1
(A)}
AF
covers X, since X f
1
(f(X)).
Since X is compact there is a nite subcollection, say f
1
(A
1
), ..., f
1
(A
1
)
covering X. We conclude that
f(X) f(f
1
(A
1
)...f
1
(A
n
)) = f(f
1
(A
1
))...f(f
1
(A
n
)) = A
1
...A
n
.
(U) =
r0
C
r
(U).
The functions in C
i
1
and
i
1
for all i {1, ..., n}, where
i
is the projection
on the ith factor, are C
cover M,
2. for any , the neighborhoods (U
) and (U
) are compatible,
3. any coordinate neighborhood (V, ) compatible with every (U
) U
is itself in U. A smooth manifold is a topological manifold together with
a dierentiable structure.
1.2 Tangent Vectors
Denition 25. Directional derivative D
a
is the direction a = (a
1
, a
2
, ..., a
n
)
is
D
a
(f)(p) = lim
h0
f(p +ha) f(p)
h
=
n
i=1
a
i
f
x
i
(p).
8 Rafael Wisniewski
In particular if we denote the ith coordinate by x
i
, i.e. x
i
=
i
then
D
a
(x
i
) =
n
j=1
a
i
ij
= a
i
.
Another way of dening the directional derivative is by specifying a curve
with a tangent vector corresponding to a:
s : (, )
U M, > 0.
In local coordinates specied by a coordinate neighborhood :
U U R
n
we have
s(t) = (x
1
(t), ..., x
n
(t)).
If we additionally require that s(0) = p and s(0) = a then
D
a
(f)(p) = lim
h0
f(s(h)) f(p)
h
=
df(s(t)
dt
=
f
x
i
dx
i
(t)
dt
=
a
i
f
x
i
.
The choice of the path s was not unique, we have still some undesired
exibility.
Denition 26. Given p U, C
(U, p).
Let us try to compute it
D
<s>p
(f) =
d
dt
f(s(t))|
t=0
=
ds(t)
dt
|
t=0
f
x
i
(p)
Nonlinear Control. Supplementary Notes to Khalils Nonlinear Systems. 9
If we dene
x
i
p
: C
(U, p) R by
x
i
p
(f) =
f
x
i
(p) and let
ds(t)
dt
|
t=0
= a,
then
D
<s>p
(f) =
a
i
x
i
p
(f).
We observe that
x
i
p
is a basis of the tangent space T
p
(U), whereas a
i
is the
coordinate. We identify T
p
(U) with the Euclidean space R
n
by
a
i
x
i
p
(a
1
, ..., a
n
).
We dene the disjoint union of all tangent space in U,
T(U) =
_
pU
T
p
(U).
We have a map T(U) U R
n
given by
n
i=1
a
i
x
i
p
(p, (a
1
, ..., a
n
)).
Denition 28. The tangent bundle is the projection : T(U) U,
_
n
i=1
a
i
x
i
p
_
= p.
1.3 Vector Fields
A smooth vector eld on U R
n
is a map : U T(U) R
2n
, which is
smooth as a map from the subset U R
n
to R
2n
and satises = id. We
shall denote totality of smooth vector elds on U as (U).
Let us consider a simple example
: (x
1
, x
2
) ((x
1
, x
2
), (x
2
1
x
2
, x
2
))
or shortly
(x
1
, x
2
) (x
2
1
x
2
, x
2
).
Notice that in particular
x
i
p
: (x
1
, ..., x
i
, ..., x
n
) (0, ..., 1, ..., 0) is a
vector eld. Every element (U) can be written
=
n
i=1
a
i
x
i
, where a
i
C
x
1
x
2
x
2
. We see that can be treated as a
derivative, e.g for f = x
1
x
2
we have (f) = x
2
1
x
2
2
x
2
x
1
.
10 Rafael Wisniewski
Denition 29. A derivative of C
(U) C
(U)
such that for all f, g C
(U).
Proof. For (U) we have =
a
i
x
i
, and (f) =
a
i f
x
i
C
(U),
but partial derivatives are derivatives.
1.4 Dierential Equation Is Born
A 1-parameter group of dieomorphisms of a manifold M is a smooth map
: R M M
such that
1. for each t R the map
t
: M M dened by
t
(q) = (t, q) is a
dieomorphism of M onto itself,
2.
0
(p) = p p M,
3. for all t, s R we have
t+s
=
t
s
.
Given a 1-parameter group of dieomorphisms of M we dene a vector eld
on M as follows. For every smooth real valued function f let
q
(f) = lim
h0
f(
h
(q)) f(q)
h
.
This vector is said to generate the group .
Given any smooth curve
t s(t) M
it is convenient to dene the velocity vector
ds
dt
T
c(t)
(M)
by the identity
ds
dt
(f) = lim
h0
f(s(t +h)) f(s(t))
h
.
Now let be a one parameter group of dieomorphisms, generated by the
vector eld . Then for each xed q the curve
t
t
(q)
Nonlinear Control. Supplementary Notes to Khalils Nonlinear Systems. 11
satises the dierential equation
d
t
(q)
dt
=
t
(q)
with initial condition
0
(q) = q. This is true since
d
t
(q)
dt
(f) = lim
h0
f(
t+h
(q)) f(
t
(q))
h
= lim
h0
f(
h
(p)) f(p)
h
=
p
(f),
where p =
t
(q).
2 Foundation of Dynamical Systems
2.1 Planar Linear Systems
The most important class of systems of dierential equations are linear sys-
tems. I shall only consider planar case in this note. This shall give you enough
intuition to work with other dimensions on your own.
Consider an autonomous planar system
x = ax +by
y = cx +dy,
where a, b, c and d are constants. We may abbreviate this system by using
the coecient matrix A where
A =
_
a b
c d
_
.
Then the linear system may be written
X = AX
Note that the origin is always an equilibrium point of a linear system. To
nd other equilibria, we must solve the linear system of algebraic equations
AX = 0.
This system has a nonzero solution if and only if det A = 0, but we have a
unique equilibrium point (0, 0) if detA = 0.
Denition 30. A nonzero vector v is called an eigenvector of A if AV = V
for some . The constant is called an eigenvalue of A.
Notice also that if V is an eigenvector of A then so is its nonzero scalar
multiple
A(aV ) = aAV = (aV ).
12 Rafael Wisniewski
Proposition 2. Suppose that V is an eigenvector for the matrix A with as-
socited eigenvalue . Then the function X(t) = e
t
V is a solution of the
system
X = AX.
Suppose we have two distinct real eigenvalues
1
and
2
with eigenvectors
V
1
and V
2
. Thus V
1
and V
2
are linearly independent and form a basis for R
2
.
For a given point Z
0
we may nd a unique pair of real number a and b such
that
Z
0
= aV
1
+bV
2
.
Proposition 3. Suppose A has a pair of real eigenvalues
1
=
2
and asso-
ciated eigenvectors V
1
and V
2
. Then the general solution of the linear system
X = AX is given by
X(t) = ae
1
t
V
1
+be
2
t
V
2
.
Proof.
X(t) =
1
(ae
1
t
V
1
) +
2
(be
2
t
V
2
) = A(ae
1
t
V
1
) +A(be
2
t
V
2
) = AX(t).
1
0
0
2
_
.
It follows that the solution is of the form
X(t) = ae
1
t
_
1
0
_
+be
2
t
_
0
1
_
Nonlinear Control. Supplementary Notes to Khalils Nonlinear Systems. 13
Since
1
< 0, the straight line solutions of the form ae
1
t
(1, 0) lie on the
x-axis and tend to (0, 0) This axis is called the stable line. Since
2
> 0, the
solutions be
2
t
(0, 1) lie on the y-axis and tend away from (0, 0) as t .
This axis is called the unstable line. Try to draw what is happening if a, b = 0.
An equilibrium with
1
< 0 and
2
> 0 will be called a saddle point.
We shall now assume that
1
<
2
< 0. Unlike the saddle case, all the
solutions tend to (0, 0) as t . The question is how they approach the
origin. We write
x(t) = ae
1
t
y(t) = be
2
t
and compute
dy
dx
=
dy/dt
dx/dt
=
2
be
2
t
1
ae
1
t
=
2
b
1
a
e
(
2
1
)t
Since
2
1
> 0, it follows that the slopes approach (provided
b = 0). Thus these solutions tend to the origin tangentially to the y-axis. An
equilibrium for which the matrix A has distinct positive eigenvalues will be
called a sink.
The last example is a source. It happens if the eigenvalues satisfy 0 <
1
<
2
. Our vector elds may be regarded as the negative of the previous
example. The general solution and phase portrait remain the same, except
that all solutions now tend away from the point (0, 0) along the same paths.
2.3 Complex Eigenvalues
Consider a system
X = AX, where A =
_
0
0
_
, and = 0. The char-
acteristic polynomial is
2
+
2
= 0, so the eigenvalues are now imaginary
numbers i. We therefore solve
_
i
i
_ _
x
y
_
=
_
0
0
_
or ix = y. Thus we nd complex eigenvector (1, i). This situation has not
been covered by our theory. We can remedy this with help of the Eulers
formula
e
it
= cos t +i sint.
We have
X(t) = e
it
_
1
i
_
=
_
cos t +i sint
i(cos t +i sint)
_
=
_
cos t +i sint
sint +i cos t
_
.
14 Rafael Wisniewski
We break X(t) into its real and imaginary parts, we have
X(t) = X
re
(t) +iX
im
(t),
where
X
re
(t) =
_
cos t
sint
_
, X
im
(t) =
_
sint
cos t
_
.
But now we see that both X
re
(t) and X
im
(t) are real solutions of the
original system
X
re
(t)+i
X
im
(t) =
X(t) = AX(t) = A(X
re
(t)+iX
im
(t)) = AX
re
(t)+iAX
im
(t).
Moreover since X
re
(0) =
_
1
0
_
and X
im
(0) =
_
0
1
_
, the linear combination of
these solutions
X(t) = c
1
X
re
(t) +c
2
X
im
(t),
where c
1
and c
2
are arbitrary constants provides a solution to any initial
value problem.
We claim that this is the general solution to
X(t) = AX(t). Suppose that
Y [t] =
_
u(t)
v(t)
_
is another solution of this equation. Consider a complex function f(t) =
(u(t) +iv(t))e
it
. Dierentiating this equation yields
f(t) = 0. Thus
u(t) +iv(t) = Ce
it
= C(cos t i sint),
where C is a constant. From this it follows that Y (t) is linear combination
of X
re
(t) and X
im
(t). Each solution is aperiodic solution with period 2/.
The solutions lie on circles centered at the origin. Try to draw them.
More generally, consider
X = AX(t), where
A =
_
_
and , = 0. The characteristic equation is one
2
2 +
2
+
2
, so the
eigenvalues are = i. The eigenvector associated to +i is determined
by the equation
( ( +i)x +y = 0.
We conclude that (1, i) is the eigenvector. We have solutions of the form
X(t) = e
(+i)t
_
1
i
_
= e
t
_
cos t
sint
_
+ie
t
_
sint
cos t
_
= X
re
(t) +iX
im
(t).
Nonlinear Control. Supplementary Notes to Khalils Nonlinear Systems. 15
Thus we nd the general solution
X(t) = c
1
e
t
_
cos t
sint
_
+c
2
e
t
_
sint
cos t
_
.
When < 0 we call the equilibrium a spiral sink, if > 0 the equilibrium
point is called spiral source.
2.4 Repeated Eigenvalues
The only remaining case occur when A has repeated eigenvalues. We consider
the coecient matrix
A =
_
1
0
_
.
There are two eigenvalues equal , but there is only one linearly independent
eigenvector given by (1, 0). We have a solution
X
1
(t) = e
t
_
1
0
_
.
To nd the other solution we write
x = x +y
y = y.
Thus if y = 0 we must have
y(t) = e
t
.
Therefore the dierential equation for x(t) reads
x(t) = x(t) +e
t.
This is non-autonomous, rst order dierential equation and a solution is of
the form
x(t) = e
t
+te
t
for some constants and . Inserting this guess into the dierential equation
shows that in fact = . Hence, the solution of the system may be written
e
t
_
1
0
_
+e
t
_
t
1
_
.
Note, that, if < 0, each term in this solution tends to 0 as t .
16 Rafael Wisniewski
2.5 Changing Coordinates
Any 2 by 2 matrix that is in one of the following forms
_
0
0
_
,
_
_
,
_
1
0
_
,
is said to be canonical.
We can consider nonsingular linear map T : R
2
R
2
. Now, instead of
considering a linear system
X = AX, suppose we consider a dierent system
Y = (T
1
AT)Y.
Note that if Y (t) is a solution of this new system, then X(t) = TY (t)
solves
X = AX. Indeed, we have
d
dt
(TY ) = T
Y (t) = T(T
1
AT)Y = A(TY (t)).
The important fact is that we always can nd a linear map that converts
a given linear system to T
1
AT in canonical form.
3 Stability
Theorem 13 (Existence theorem for ordinary dierential equations).
Suppose U R
n
be an open set and I
U.
Then for each x U there exists > 0 and a neighborhood U
of x,
U
U, such that:
1. For each a = (a
1
, ..., a
n
) U
and mapping I
, we write them
x
i
(t, a
1
, ..., a
2
), i = 1, ..., n, in which case they are of class C
r
in all
variables and thus determine a C
r
map of I
U.
Nonlinear Control. Supplementary Notes to Khalils Nonlinear Systems. 17
Suppose U R
n
is an open set with 0 U. For a vector eld (U)
such that 0 is an isolated singularity (0) = 0. We consider an autonomous
system
x = (x). (4)
Denition 31. The equilibrium point x = 0 of (4) is
1. stable if for all > 0 there is () > 0 such that x(0) < x(t) <
for all t > 0;
2. unstable if it is not stable;
3. asymptotically stable if it is stable and there exists > 0 such that
x(0) < lim
t
x(t) = 0.
Theorem 14. Consider system (4) with x = 0 an equilibrium point. Let
V C
1
(U) satises
1. V (0) = 0 and V (x) > 0 for x U {0},
2. v = dv() = (v) 0 for all x U.
Then x = 0 is stable. Moreover if
V (x) < 0 for x U {0} then x is
asymptotically stable.
Example 2. Consider a system
x
1
= x
1
+x
1
x
2
x
2
= x
2
.
Take V (x) =
1
2
(x
2
1
+x
2
2
) then
V = x
1
(x
1
+x
1
x
2
) x
2
2
= x
2
1
x
2
2
+x
1
x
1
x
2
.
If we consider x B(0, r); in particular x
1
< r, then
V x
2
1
x
2
2
+r|x
1
||x
2
| = |x|
T
P|x|,
where P =
_
1
r
2
r
2
1
_
. Determinant of p is 1
r
2
4
> 0. Hence we conclude
that P is positive denite for r < 2 and the system is asymptotically stable.
Theorem 15. Consider system (4) with x = 0 an equilibrium point. Let
V C
1
(R
n
) satises
1. V (0) = 0 and V (x) > 0 for all x R
n
{0},
2. x v(x) (v is radially unbounded),
3.
V (x) < 0, for x = 0.
Then x = 0 is globally asymptotically stable.
18 Rafael Wisniewski
Example 3. Euler equations for a rotating rigid spacecraft are given by
J
1
1
= (J
2
J
3
)
2
3
+u
1
,
J
2
2
= (J
3
J
1
)
3
1
+u
2
,
J
3
3
= (J
1
J
2
)
1
2
+u
3
,
where u
1
, u
2
, u
3
are torques about the principal axes, J
1
, J
2
, J
3
are principal
moments of inertia, and
1
,
2
,
3
are angular velocities.
If u
1
= u
2
= u
3
= 0 the origin is stable. Let the Lyapunov function be
kinetic energy of the spacecraft
V (
1
,
2
,
3
) =
1
2
(J
1
2
1
+J
2
2
2
+J
3
2
3
).
We check its derivative
V (
1
,
2
,
3
) = J
1
1
1
+J
2
2
2
+J
3
3
3
= (J
2
J
3
)
1
3
+ (J
3
J
1
)
1
3
+ (J
1
J
2
)
1
3
= 0
and conclude that the origin is indeed stable.
If we implement a simple control u
i
= k
i
i
, k
i
> 0, i = 1, 2, 3 we see
that the closed loop system is globally asymptotically stable:
V =
1
u
1
+
2
u
2
+
3
u
3
= k
1
2
1
k
2
2
2
k
3
2
3
,
which is negative denite. Notice also that
V (
1
,
2
,
3
)
1
2
min{J
1
, J
2
, J
3
}(
2
1
+
2
2
+
2
3
) =
1
2
min{J
1
, J
2
, J
3
}
2
.
Thus V is radially unbounded.
Check of stability of a linear system x = Ax, as you would expect, is
particularly simple. Dene a Lyapunov function
V = x
T
Px,
where P is a positive denite function, then
V = x
T
(A
T
P +PA)x = x
T
Qx,
with A
T
P +PA+Q = 0. This analysis motivates the following proposition.
Proposition 4. A linear system x = Ax is asymptotically stable if and only
if for any B such that (A, B) is controllable there exists positive denite
solution to the following Lyapunov equation
A
T
P +PA+BB
T
= 0.
References