Dynamical systems, 4
Mihaela Cistelecan, UT Cluj-Napoca
These notes were prepared using:
1. Hale, J.K., Kocak, H. Dynamics and Bifurcations. Springer-Verlag,
1991.
2. Hirsch, W., Smale, S, Devaney, R. Differential Equations, Dynamical
Systems and An Introduction to Chaos. elsevier, 2004
3. Diacu, F. An Introduction to Differential Equations: Order and Chaos.
W.H. Freeman and Company, 2000.
4. Khalil, H. Nonlinear Systems. Prentice Hall. 1996.
5. http://www.soe.ucsc.edu/classes/ams146/Spring05/Lectures.html
x Rn
(1)
1.1
Motivation
A. Linear Systems - Structural stability and bifurcation Since mathematical models are approximations of the reality, solving the associated
differential equations may not suffice for understanding the phenomenon it
describes; we might need additional information.
Example 1: If observations of experiments suggest as a model of a physical
phenomenon a certain eq., say
x + 2.1x + 0.76x = 0
(2)
then we might need to determine the behavior of the solutions for all equations of the type
(3)
x + bx + cx = 0
2
where b and c belong to some intervals that contain the coefficients in (2),
say b in (1.9, 2.3) and c in (0.6, 0.9). Since all measurements are approximate,
the value 2.1 could in truth be 2.05 or 2.17. If the solutions of all equations
given by (3), for b and c in the chosen intervals behave like the solution of
(2), then the results are trustworthy because the qualitative structure of the
phase-plane portrait remains the same. We then say that (3) is structurally
stable . In the phase plane, the system:
x = y
y = cx by
(4)
(5)
where c is a parameter.
Discussion The flows of system (30) depend
parameter c, as follows:
on the
1). c < 0 - the equilibrium
points are c
and
c
and the orbits are
(6)
If the rod of the pendulum is rigid, then one can interpret the application of
the torque as pushing the pendulum with constant force that is perpendicular
to the rod.
By choosing y1 = and y2 = the phase plane model is:
y 1 = y2
y 2 = y2 sin y1 + M
(7)
(8)
(9)
The eigenvalues
of the linearization of (9) about the equilibrium point at
the origin are i 1 2 . For < 0 the origin is asymptotically stable
because the real parts of the eigenvalues are negative. At = 0 the origin is
still stable, but for > 0 the real parts of the eigenvalues become positive
and thus the origin is unstable.
1.2
1
1
1.3
that a linear system may face bifurcations only in the following two cases:
a) The equilibrium of the linear system is nonhyperbolic and the system
(the coefficient matrix) undergoes small perturbations.
b) The linear system is parameter time-varying and under these variations
there exists a parameter configuration that produce a modification of the
hyperbolic equilibrium into a nonhyperbolic equilibrium. For example, we
may be given a linear system that contains parameters and the parameters
enter into the system as continuous functions. In this case it is unavoidable
to hit one of the nonhyperbolic systems if there are parameter values at which
the linear systems belong to two different hyperbolic equivalence classes.
Topological classifications of orbits
In the following we give two theorems (without proof) taken from literature concerning the topological classification of hyperbolic/ nonhyperbolic
systems.
Theorem T1: Hyperbolic linear systems:- Suppose that the eigenvalues of two matrices A and B have nonzero real parts. Then the two linear
systems x = Ax and x = Bx are topologically equivalent if and only if A
and B have the same number of eigenvalues with negative (and hence positive) real parts. Consequently, up to topological equivalence, there are three
distinct equivalence classes of hyperbolic planar linear systems with, for ex.,
the following representatives:
1
0
1
0
1
0
0
1!
0
1 !
0
1
(10)
Theorem T2: If a coefficient matrix A has at least one eigenvalue with zero
real part, then the planar linear system x = Ax is topologically equivalent to
precisely one of the following five linear systems with the indicated coefficient
matrices:
0
0
1
0
1
0
0
0
0
1
0
the zero matrix;
0 !
0
one negative and one zero eigenvalue;
0!
0
one positive and one zero eigenvalue;
0 !
1
two zero eigenvalues but one eigenvector;
0 !
1
two purely imaginary eigenvalues.
0
(11)
Sketch of the proof for Theorem 1, first matrix The idea of the classification is to find the transformation function that transform a dynamics
x = Ax into its topologically equivalent dynamics x = Bx. This problem
is equivalent to the following one: for each linear system, find an ellipse encircling the origin such that each orbit, except the equilibrium point at the
origin, crosses the ellipse in the same direction and only at one point. This
is also equivalent to finding (or proving the existence) of two positive definite
symmetric matrices CA and CB such that:
AT CA + CA A = I; B T CB + CB B = I
(12)
(13)
Similarly,
i
d h T
x (t)CB x(t) = xT (t)[B T CB + CB B]x(t) = xT (t)x(t)
dt
(14)
!
a b
A=
(15)
c d
we know that the eigenvalues are the roots of the characteristic equation
which can be written:
2 (a + d) + (ad bc) = 0
(16)
2 trace(A) + det(A) = 0
(17)
or, equivalently:
The eigenvalues are given by:
q
1
= (tr(A) (tr(A))2 4det(A))
2
(18)
+ + = tr(A), + = det(A)
(19)
Note that:
Thus the point (tr(A), det(A)) defines the type of the equilibrium point,
as follow:
- if tr(A)2 4det(A) < 0 then the equilibrium is either a focus (tr(A) 6= 0)
or a center (tr(A) = 0); if tr(A) < 0 the focus is stable, otherwise it is
unstable;
- if tr(A)2 4det(A) 0 the equilibrium is a node; the node stability
depends on the sign of tr(A);
- if det(A) < 0 the equilibrium is a saddle.
Remarks: 1). The trace-determinant plane is a two-dimensional representation of a four dimensional space, since 2 2 matrices are determined by
four parameters, the entries of the matrix. Thus, there are infinitely many
different matrices corresponding to each point in the trace-determinant plane.
2). We can think of the trace-determinant plane as the analog of the bifurcation diagram for planar linear systems. A one-parameter family of linear
systems corresponds to a curve in the trace-determinant plane. When this
curve crosses the trace-axis, the positive determinant axis, or the parabola
tr(A)2 4det(A) = 0, the phase portrait of the linear system undergoes
trace
det
dynamics
|
|
|
|
|
<0
++
++
source
=0
+
0
bif.
>0
++
saddle
(20)
0
0 0
0
0
0 1
0 0
(21)
10
(22)
where:
tr A = a11 + a22 = 1 + 2 ; det A = a11 a22 a21 a12 = 1 2
(23)
!
1 0
A0 =
(24)
0 0
In this case, there is a neighborhood of A0 in R4 such that in this neighborhood the matrices that are topologically equivalent to the nonhyperbolic
matrix A0 satisfy the conditions tr A < 0 and det A = 0. The set of such
matrices is a three-dimensional surface S in R4 containing the matrix (24)
because a22 det A|A0 = 1 6= 0 (see IFT). All the matrices on one side of
the surface S are topologically equivalent. We construct a curve of matrices
which crosses this surface transversally at the point A0 . In other words, we
define a matrix A() depending on a scalar parameter such that A(0) = A0 .
As the parameter is varied from negative to positive values, one crosses the
surface S from the side with (tr A() < 0, det A() < 0) to the side with
(tr A() < 0, det A() > 0). Observe that on the first side, the matrices have
two real negative eigenvalues, and on the other side, they have one negative
and one positive eigenvalue. One such one-parameter matrix is:
A() =
1 0
0
(25)
In the first case the matrices have two real negative eigenvalues, and in
the other case they have one negative and one positive eigenvalue.
It follows that in the case of a nonhyperbolic equilibrium in the form (24)
the system x = Ax may switch from a stable node equilibrium to a saddle
equilibrium or vice-versa when passing through (24).
Case 2: Unfolding
!
0 1
A0 =
(26)
1 0
11
A() =
1
1
(27)
It means that for the nonhyperbolic equilibrium (26) the system x = Ax may
switch from a stable focus equilibrium to an unstable focus equilibrium or
vice-versa when passing through (26).
Case 3: Unfolding
!
0 1
A0 =
(28)
0 0
Since in this case the matrix A0 has tr A = 0 and det A = 0 the A0 unfolding
depends on two parameters. It was proved that bifurcations are given by the
two following two-parameters matrices:
A(1 , 2 ) =
1
1
2 1
, A(1 , 2 ) =
0 1
1 2
(29)
1.4
12
1. Saddle-node bifurcation
x = + x2 = F (, x)
(30)
x = x + x2 = F (, x)
(31)
x = + x x3 = F (, x)
(32)
x = x x3 = F (, x)
(33)
x = c + dx x3 = F (c, d, x)
(34)
2. Transcritical bifurcation
3. Hysteresis
4. Pitchfork bifurcation
5. Fold or cusp
A numerical procedure to draw the bifurcation diagram for a
scalar one-parameter system
A graphical method for depicting some of the important dynamical features in equations x = F (, x) depending on the parameter R is the
bifurcation diagram. This method consists of drawing curves on the (, x)plane, where the curves depict the equilibrium points for each value of the
parameter. More specifically, a point (0 , x0 ) lies on one of these curves if and
only if F (0 , x0 ) = 0. For instance, the bifurcation diagram of the saddlenode bifurcation x = + x2 is the parabola = x2 . Also, to represent the
stability types of these equilibria, we label stable equilibria with solid curves
and unstable equilibria with dotted curves.
For a scalar system x = F (, x) depending on a scalar parameter, we can
parameterize the curve F ((t), x(t)) = 0 by t, satisfying (0) = 0 , x(0) = x0
and F (0 , x0 ) = 0. To represent this parameterized curve as an orbit of a
differential equation, we differentiate the eq. F ((t), x(t)) = 0 with respect
to t:
F
F
(, x) +
(, x)x = 0
(35)
x
x)
To satisfy this identity (,
must be a constant multiple of the vector
F
F
( x (, x), (, x)). If we choose the parametrization of the curve ((t), x(t))
in such a way that this constant is one, then we arrive at the following system:
= F
(, x)
x
F
x = (, x)
13
(36)
d
f (x)
dx
14
F
(0, 0)
x
6= 0
(37)
Then there are constants > 0 and > 0 and a continuous function
: { : kk < } R
such that
(0) = 0 and F (, ()) = 0 f or kk <
(38)
15
What if (0, 0) is a nonhyperbolic equilibrium ? Then, the parameter may change everything ... Therefore, we have to investigate the sign
of the remaining terms of the Taylor expansion.
Remember that the Taylor expansion of function F (, x) about (+, x+
) is as follows:
+ F
+
F ( + , x + ) = F (, x) + F
x
1 2F 2
1 2F 2
1 2F
+ 2 2 + + 2 x + higher order terms
2 x2
(39)
F
(0, 0)
x
= f 0 (0) 6= 0
(41)
Hence, the IFT implies that there exists a function () with (0) = 0
such that x = () and F (, ()) = 0. In order to establish the stability
behavior of the equilibrium () we compute the sign of the derivative:
F
(, ())
x
(42)
F
(0, 0)
x
=0
2F
(0, 0)
2x
= f 00 (0) 6= 0
(43)
x2
+ G(, x)
2
(44)
with a(0) = 0, b(0) = 0, c(0) = f 00 (0) 6= 0, and |G(, x)| < |x|2 .
It has been proved mathematically that the qualitative structure of the flow
of the perturbed equation x = F (, x) is determined from a single function
of the parameters , namely the function () corresponding to the extreme
value of F (, x). The extreme points of F correspond to the solutions x of the
equation F
(, x) = 0. The number of equilibrium points of eq. x = F (, x)
x
depends upon the extreme value () = F (, ()) of the function F , as
follows:
()f 00 (0) < 0 there are two hyperbolic equilibria near the origin
() = 0 implies that there is a nonhyperbolic eq. at the origin
()f 00 (0) > 0 there are no eq. points about the origin
Thus, even though there may be k components of the vector parameter , the
bifurcation behaviour of the perturbed eq. x = F (, x) depends on a single
parameter - the function () corresponding to the extreme value of F (, x).
Example: Study the orbit structure of the dynamical system x = 1 +
2 x + x2 , depending on two small parameters - = [1 2 ]T . The function
() for this example is as follows: (1 , 2 ) = 1 (1/4)22 (corresponds to
the minimum value of the function F (, x)). Thus, the bifurcations occur as
we cross the curve 1 = 22 /4 in the (1 , 2 )-plane. There are two equilibrium
points if 1 < 22 /4 and none if 1 > 22 /4. The same discussion results if we
study the roots of the equations F = 0. Why ?
Equilibria with cubic degeneracy:- Consider the perturbed system
x = F (, x), satisfying:
F (0, x) = f (x)
2F
(0, 0) = 0
2x
F
(0, 0)
x
3F
(0, 0)
3x
=0
= f 000 (0) 6= 0
(45)
x2
x3
+ d() + G(, x)
(46)
2
6
where a(0) = b(0) = c(0) = 0, d(0) = f 000 (0) 6= 0 and |G(, x)| < |x|3 .
Moreover, it has been proven that an arbitrary two-parameter perturbation
of the cubic degeneracy, under certain reasonable conditions, can be reduced
to the special two-parameter perturbation on the following form:
F (, x) = a() + b()x + c()
F (, x) = 1 + 2 x + c()
x2
x3
+ d() + G(, x)
2
6
(47)
We find the bifurcation values of the parameters, that is, the values of =
(1 , 2 ) for which the function F (, x) has multiple zeroes. To accomplish
this we must solve the following two eq:
2
(48)
(49)
These equations are the parametric representation of a cusp in the (1 , 2 )plane which near the origin approximately coincides with the curve
9
32 = f 000 (0)21
8
(50)
Therefore, the bifurcations are determined only by the constant and the
linear terms of the Taylor expansion of the vector field. The term c()x2 has
little influence on the bifurcations of (47) because this term does not enter
into the first approximation to the cusp in the (1 , 2 )-plane.
Example: Fold or Cusp
x = c + dx x3 = F (c, d, x)
(51)
The vector field (51) is the most general perturbation of the function
x with lower order terms (because the term involving x2 can always be
3
19
F (c, d, x)
x
=0
(52)
It follows:
c + dx x3 = 0 and d 3x2 = 0
(53)
The objective is to determine all values of c and d for which these two
equations can have some common solution x. It follows that
d = 3x2 and c = 2x3
(54)
(55)
Pitchfork bifurcation
x 1 = + x21
x 2 = x2
(56)
x 1 = x1 x31
x 2 = x2
(57)
Vertical bifurcation
x 1 = x1 + x2
x 2 = x1 + x2
Poincare-Andronov-Hopf bifurcation
x 1 = x2 + x1 ( x21 x22 )
x 2 = x1 + x2 ( x21 x22 )
(58)
(59)
(60)
Home-works
1. Prove that the system (4) is structurally stable in the whole parameter
plane except for the b-axis and the positive part of the c-axis.
2. Using Lemma 1 verify that
a). for x = x the equilibrium point at 0 is asymptotically stable;
b). for x = x2 the equilibrium point at 0 is unstable;
c). for x = x x3 the equilibrium points at -1 and 1 are asymptotically
stable, and 0 is unstable.
3. Prove that () = is the function whose existence is guaranteed
by the IFT for the function F (, x) = + (1 + )x + x2 .
4. Apply the IFT to show that there is a unique solution of the equation
+ (1 )y + y 3 = 0 near (, y) = (1, 1).
5. Show that the conditions of the IFT are satisfied for the following
functions: a). + sin x, b). 1/2 + cos(/6 + x), c). sin + tan x.
6. Consider an RLC electric circuit which has a resistor, an inductor and
a capacitor all connected in series. The circuit is characterized by q, the total
charge in the plates of the capacitor and by i, the intensity of the current,
which are connected by the relation q = i:
q +
R
1
q +
q=0
L
LC
(61)
Study the type of the equilibrium and prove that the charge in plates will
eventually tend to zero.
7. The model describing the motion of water in a pipe is
k
x
(a) x = m
2g
x
L
(62)
2g
x
L
(63)
if frictions are neglected. Study the structural stability in cases (a) and (b).
8. Prove the topological equivalence for the systems x = A1 x, x = A2 x,
x = A3 x, x R2 , where:
A1 =
1 0
0 2
; A2 =
; A3 =
1
0
(64)
x =
1 0
0 2
x; x =
1 2
2 1
x;
(65)
10. For the following matrices, discuss the transition from one topological
equivalence class to another as the scalar parameter is varied:
(a)
, (b)
1
2
2 + 2
(66)
1 0
0 1
(a) A0 =
(b) A0 =
(c) A0 =
1 0
0 1
1 0
0 2
A1 =
A1 =
A1 =
1 0
0 1
1 0
0 1
(67)
1 1
0 1
(68)
!
(69)
(70)
(72)
Lab-works
x =
2 0
0 2
x; x =
2 1
0 2
x;
(73)
Draw the phase portraits of these two linear systems and compare.
3. Consider the following five coefficient matrices A of the linear systems
x = Ax:
!
!
!
1 0
1 1
1 0
,
,
0
1
0 1
0 2
!
!
(74)
0
2
1 1
0 2
1 3
(a) For each A above, determine a symmetric matrix CA satisfying the
matrix equation AT CA + CA A = I.
(b) Draw the level set {x : xT CA x = 1} for each of your CA .
4. Obtain the bifurcation curves of the following one and two-parameter
perturbations of the cubic vector field f (x) = x3 and sketch some representative phase portraits:
x = 1 + 1 + 21 x x3
x = 1 + 22 + 2 x x3
23
(75)