Академический Документы
Профессиональный Документы
Культура Документы
529
530
stable steady state in the state space. The steady state works
as if it absorbs every neighboring state into itself. Such a
steady state is called an attractor, and the existence of attractors is the most significant property of a dissipative system
such as the circuit dynamics. On the other hand, a lossless circuit containing only inductors and capacitors is formulated as
an energy conservative system similar to classical mechanics.
In this case we have no attractors and the distinction between
transient state and steady state becomes difficult.
Examples of Steady State and Attractor
Among all possible states in the state space, some particular
states have a special property such that they are invariant in
some sense during the time evolution. They are candidates to
be attractors or steady states of the system. Equilibrium
point, periodic state, quasiperiodic state, and a more complicated state, called chaotic state, are typical invariant sets of
the system. An equilibrium point, also called a rest point, is a
single point in the state space which always rests at the same
point during the time evolution. This is the simplest steady
state and corresponds to a dc operating point in a real circuit.
Periodic state is a periodically repeated state with a definite
period or frequency. This state is commonly observed in
forced circuits driven by an ac voltage or current source. An
oscillator also produces a periodic state. Many biological
rhythms are also modeled as periodic states generated by biological oscillators. A state containing several distinct frequencies is called a quasiperiodic state. It appears sometimes in a
forced oscillator with periodic input signal. When the difference of free and driving frequencies is appropriately large,
both frequencies can survive and doubly periodic oscillation
becomes possible. In this case the time response is a beat or
quasi-periodic oscillation. Chaotic state is the most complicated state whose long-term time response looks like a noisy
or random nature. Later we will discuss this state more precisely. The concept of stability of the above invariant sets is
also important. Roughly speaking, a steady state is stable if
every neighboring state always stays in the neighborhood of
the steady state in future evolution. If a steady state satisfies
a stronger condition such that all neighbors approach the
steady state, then we say that the state is asymptotically stable. An attractor is an asymptotically stable steady state. In
a real system a physically observable state is an attractor. In
the theory of dynamical systems the above invariant sets are
called nonwandering sets. On the other hand, a transient
state corresponds to a wandering set. Figure 1 shows schematic diagram of states of dynamical systems and their bifurcations.
Role of System Parameters
A steady state of a circuit depends also on parameters contained in circuit dynamics. Associated with the change of parameters, the qualitative property of a steady state may
change at some particular value of the parameters. For example, the appearance of a couple of steady states, stability
change of a steady state, or the creation of a new type of
steady state, and so on, may occur under the variation of parameters. We may imagine the parameters as a controlling
device of the qualitative property of states. That is, by changing system parameters we can see a morphological process of
steady states, which is referred to as a bifurcation of state or
Transient states
Wandering
states
Periodic states
TB
PDB
NS
Quasiperiodic
states
PDB
Regular states
531
Complex
and
composed
states
Homoclinic
states
Chaotic states
Phase space
Attractor
Mutual
interaction
Stabilize or
destabilize
Selforganization
Synchronization
Tangent
period-doubling
Neimark-Sacker
Jump
hysteresis
Self-excited
oscillation
Nonlinear
resonance
Hopf
Tangent
Bifurcation
Parameter space
Parametric
excitation
Period doubling
Figure 2. Schematic diagram of typical nonlinear
phenomena: synchronization, self-excited oscillation, nonlinear resonance, and parametric excitation.
532
Equilibrium points
Limit cycles
Invariant sets
Phase portrait
Parameter family
of dynamical systems
Calculus, analysis
Bifurcation diagram
Dynamical
systems
Probability
Time response
Frequency spectrum
Lyapunov exponent
Fractal dimension of attractor
Topology
Global structure
of phase portrait
Basin of attractor
Geometry
Geometrical or Topological Method. Although nonlinear ordinary differential equations cannot generally be solved explicitly by quadrature, we can know the existence of a solution with a given initial condition, the uniqueness property of
the solution, extendability of the solution in long time interval, asymptotic property of solution, stability of the solution,
and so on. These properties depend upon the geometrical or
mainly topological property of dynamical systems. A qualitative approach is then directed to the study of phase portraits,
stability theory, and bifurcation processes.
Numerical Method or Simulation. Many numerical integration methods are now available. Combining these integration
methods and the qualitative approach, we can calculate any
type of steady state, stability condition, bifurcation condition,
statistical test condition for chaotic states, and so on. Newtons method and other root finding methods are effectively
used for the numerical computations.
References in This Section
The theory of dynamical systems, especially classical mechanics, has a long history and has developed many useful techniques to study the time evolution of state. During the early
part of the twentieth century, the theory of nonlinear oscillations arose in electrical and mechanical engineering and has
been developed also in parallel with that of dynamical systems. After discovering the chaotic state in many applied
fields, the nonlinear dynamics has become popular during the
last two decades. Many books and references are now available. We refer to only classical books (19) about nonlinear
oscillations in circuit dynamics and dynamical systems.
BASIC MATHEMATICAL FACTS
In this section we review minimal mathematical tools for understanding the circuit dynamics as a time-evolving process
called a dynamical system. We also mainly treat a smooth
system; that is, the functions or maps defining the system will
be differentiable as many times as we want. In the remainder
of the article the term dynamical system refers to a differentiable dynamical system or simply a smooth dynamical system.
Circuit Dynamics, State and State Equation
Every lumped electrical circuit obeys two basic physical laws:
(1) Kirchhoff s voltage and current laws and (2) the element
characteristics derived from the constitutive relation of circuit
element. Combining these two constraint relations and eliminating auxiliary variables, we can obtain a system of firstorder ordinary differential equations in normal form as the
state equation or circuit dynamics of the circuit:
x = f(t, x, )
(1)
iG
iG
iC
vC
(a)
i
iG = g(v)
iL
L
R
E
1
2
= . Rm
..
m
(2)
where A is an n n constant matrix and is a small parameter of real number. In this case we say Eq. (2) a quasilinear
system or a weakly nonlinear system.
x1
x2
x = . Rn ,
..
xn
533
iG = I
vC
vC = V
(b)
534
We assume that the capacitor and inductor have linear characteristics whereas the conductor G has a nonlinear characteristics with voltage controlled type [see Fig. 4(b)]. For convenience we assume the nonlinear characteristics as a cubic
polynomial. Then the constitutive relations are written as
dvC
,
dt
di
vL = L L ,
dt
iG = g(vG ) = IG G1 vG G2 v2G + G3 v3G ;
iC = C
(3)
(4)
1
1
i
g(v
)
L
C
C
v
C
f=
x= C ,
(5)
1
R
E
iL
vC iL +
L
L
L
This gives an autonomous vector field in two-dimensional
state space (vC, iL).
By using the coordinate translation
vC = v + V,
iL = i + I,
iG = ig + I
1 = g1
L
,
C
g
3 = 3
C
L
,
C
k=R
C
,
L
B=e C
(12)
(13)
2
d 2y
3 dy
dy
+y=0
1 1
2
d
1 d
d
(14)
Both Eqs. (13) and (14) are expressed by the first-order form
as Eq. (11), hence they are equivalent.
2. A forced resonant circuit. Figure 5 shows another resonant circuit with a saturable nonlinear inductor driven by an
alternating voltage source E sin t. As shown in the figure,
the linear resistor R is placed in parallel with the linear capacitor C, so that the circuit is dissipative. With the notation
of Fig. 5, we have
(6)
dv
= i + g1 v g3 v 3
dt
di
= v Ri + e
L
dt
3 2 dx
d 2x
+x=0
1 1 3 x
d 2
1
d
G1 , G2 , G3 > 0
dvC
= iL iG = iL g(vC )
dt
di
L L = vC RiL + E
dt
v
dvC
+ C = iL
dt
R
(15)
d
+ vC = E sin t
dt
(7)
where we put
g1 = G1 + 2G2V 3G3V 2 ,
g3 = G3 ,
e = V RI + E
(8)
(16)
(9)
R
Cv,
y=
Li,
1
= t
LC
(10)
C
E sin t
dx
= y + 1 x 3 x3
d
dy
= x ky + B
d
iL
E0
(11)
Figure 5. Forced resonant circuit with a nonlinear saturable inductor.
v
dvC
1
= C +
f ()
dt
RC nC
(17)
d
1
E
= vC + sin t
dt
n
n
for the state variables (vC, ). This gives a nonautonomous
system. By eliminating vC, we have the following second-order
equation:
d
d 2
+ b1 + b2 2 + b3 3 = B cos
+k
d 2
d
(18)
where
1
,
k=
RC
= t tan1 k,
E
1 + k2
B=
n
a
bl = 2 l2 (l = 1, 2, 3),
n C
(19)
(20)
(21)
(22)
(23)
x(t0 ) = (t0 , x0 , ) = x0 ,
ant under any translation of time. Hence without loss of generality we can choose the initial instance t0 0.
Hence the questions arise. For a given initial value problem, does Eq. (1) has a solution for all t I R? If Eq. (1)
has a solution, is such a solution unique and does it extend to
the entire time interval R? The answer is the following theorem on the local existence and uniqueness of the solution of
Eq. (1).
Theorem 1. Suppose that in Eq. (1) the function f(t, x, ) is
differentiable in all variables t, x, , then there exists an interval I R containing t0 and the solution (21) also exists for
all initial conditions (t0, x0) I Rn. Moreover, this solution
is unique.
Remark 2. 1. The initial value problem (24) can be equivalently rewritten as the integral equation of the form:
x(t) = x0 +
f(s, x, )ds
(25)
t0
Existence and uniqueness property is then discussed by posing an appropriate condition on f. One of the sufficient conditions to guarantee the property is known as a local Lipschitz
condition. Because the differentiability is stronger than the
Lipschitz condition, we never worry about the existence and
uniqueness problem if f is differentiable. Note that the solution to Eq. (21) exists only in a short time interval I so that
the theorem asserts a local existence property.
2. Extendability of the solution to the entire time interval
R depends on f(t, x, ). Usually the function f(t, x, ) is defined in a bounded region of state space Rn. Starting with an
initial state in the region, the state may reach the boundary
of this region after a finite time, and the solution could no
longer be extended to rest in the region. The simplest example
of such behavior is a blow-up situation where a state approaches to infinity within a finite time. In most circuit applications, however, the solution can be extended to the entire
time interval R.
3. In circuit dynamics, under some particular connection of
elements, the normal form of the state equation (1) may break
at some points or in some subset in the state space as the
next example shows. This pathological situation occurs by
making an oversimplified model for a real physical circuit. It
can be remedied, however, by an appropriate normalization
technique, such as inserting stray reactance elements into
suitable positions of the circuit.
Example 2. In Eq. (7), if we remove the capacitor (i.e., C
0), then we have
= f(t, x(t), ),
535
tIR
(24)
i = g1 v + g3 v3
L
(26)
di
= v
dt
dv
=v
dt
(27)
536
or equivalently
dv
v
=
dt
L(g1 3g3 v2 )
Hence at the point v2 g1 /3g3, Eq. (27) or (28) becomes singular; that is, the circuit dynamics could not be defined. Note
that, instead of the inductors current i, the conductors voltage is used for describing Eq. (27). The inductor is connected
in series with the voltage-controlled conductor with noninvertible characteristics. Hence even if the element characteristics are differentiable, the state equation can never be described by the normal form of Eq. (1). The above points are
called impasse points and generally appear by making an
oversimplification of a mathematical model of the circuit. Indeed if we consider a small stray capacitance C in parallel
with the nonlinear conductor, then the state equation is written in the form of Eq. (4).
Continuous Dependence on Initial Condition
and Parameters at a Finite Time
Knowing that the solution Eq. (21) exists for any initial state
and parameters, we can regard the solution Eq. (21) as the
following function:
(t, , ) :
Rn Rm Rn ;
(x0 , )
(t, x0 , )
(29)
(30)
f(t, (t, x0 , ), )
+
(31)
(34)
(t,
x0 , ) + (t) = f(t, (t, x0 , ) + (t), )
= f(t, (t, x0 , ), )
+
f(t, (t, x0 , ), )
(t) +
x
(35)
F:
with the initial conditions
(36)
(t0 , x0 , )
= In
x0
(32)
(t0 , x0 , )
=0
(33)
x = gradF (x)
(37)
gradF (x) =
F
x
F
F
, ,
x1
xn
T
(38)
F
x
T
0
(39)
Hamiltonian System, Conservative System, or Lossless System. In classical mechanics we study mainly Hamiltonian
systems. In circuit application, a lossless circuit is described
by this type of equation. Let an energy function H be defined
as
R R R; (x, y)
H(x, y)
n
H:
x =
C1
H
y
H
y =
x
H dx H dy
H
dH
=
+
=
dt
x dt
y dt
x
T
(41)
H
y
H
y
H
x
F (v1 , v2 ) =
R R R; (x, y)
F (x, y)
n
(43)
H
y
F
x
y =
H
x
F
y
(l = 1, 2)
(45)
v2
C2
(49)
R2 R2 ;
(v1 , v2 )
1 (v1 , v2 ) = (v2 , v1 )
(50)
R2 R2 ;
(v1 , v2 )
(v1 , v2 ) = (v1 , v2 )
(51)
(v1 , v2 )
2 (v1 , v2 ) = (v2 , v1 )
(52)
0
1
1
,
0
1
0
0
,
1
2 =
0
1
1
0
(53)
ig2
v1
(48)
C1
and
g1
dF
F dv1
F dv2
=
+
dt
v1 dt
v2 dt
2
1 F
1 F
+
50
=
C1 v1
C2 v2
2 = 1 :
ig1
v2
(44)
g(v1 )dv1 +
Hence Eq. (48) is a kind of gradient system. In fact, F decreases along a trajectory, that is,
1 :
v1
F (v1 , v2 )
dv1
=
dt
v1
F (v1 , v2 )
dv
C2 2 =
dt
v2
=0
(42)
C1
Dissipative System. A dissipative system is a combined system of the above two systems. Let F be a dissipative scalar
function:
n
1
G(v1 v2 )2 +
2
x =
(46)
F:
dv1
= g(v1 ) G(v1 v2 )
dt
dv
C2 2 = g(v2 ) G(v2 v1 )
dt
(40)
537
g2
E1 = {(v1 , v2 ) R2 | v1 = v2 }
E2 = {(v1 , v2 ) R2 | v1 = v2 }
(54)
538
dv
= g(v),
v E1
dt
dv
= g(v) 2Gv,
v E2
C
dt
C
(55)
v E
1 2
1
v +
F () C sin t
2n C nC
n
(56)
where
F () =
0
(57)
dvC
H
1
=
=
f ()
dt
nC
d
H
1
E
=
= vC + sin t
dt
vC
n
n
(58)
1 2 1
1
y + c1 x2 + c3 x4 x(B0 + B cos )
2
2
4
(59)
dx
H
=
=y
d
y
dy
H
=
= c1 x c3 x3 + B0 + B cos
d
x
(60)
The qualitative, geometrical, or topological approach of nonlinear ordinary differential equations is a powerful tool for
understanding the nonlinear phenomena of circuit dynamics.
In this and the following sections we introduce some basic
examples from this approach. For now we consider an autonomous system
x = f(x, ),
(61)
x Rn , Rm
(65)
where x Rn is a state vector and Rm is a system parameter. Usually the terms state and phase have the same
meaning. Hence the state space Rn is also called the phase
space. In the two-dimensional case, we say the phase plane
instead of the state plane. Note that Eq. (65) defines the
phase velocity vector field at every point in the phase space.
The phase portrait of Eq. (65) is the set of all trajectories in
the phase space Rn. The phase portrait contains useful information of the behavior of trajectories. We see the number and
types of equilibrium points, their asymptotic behavior when
t , and so on. In practice, only typical trajectories are
illustrated in the portrait to show the behavior schematically.
Equilibrium Point and its Topological Classification
A point at which the phase velocity becomes zero is called an
equilibrium point. The point corresponds to a dc operating
point of a circuit. Hence an equilibrium point x0 Rn is given
by the relation
f(x0 , ) = 0
(62)
H
F
dx
=
d
y
x
H
F
dy
=
d
x
y
(67)
(63)
Example 4. 1. Consider Eq. (11) in Example 1(1). Equation
(66) is given by
(66)
f 1 (x0 , y0 ) = y0 + 1 x0 3 x30 = 0
(64)
f 2 (x0 , y0 ) = x0 ky0 + B = 0
(68)
(69)
k1 1 1
,
k3
k
k1 1
,
k3
(0, 0),
1
k1 1
,
k3
k
k1 1
k3
(70)
di1
= E 1 R1 i 1 v
dt
di
L 2 2 = E 2 R2 i 2 v
dt
dv
= i1 + i2 g(v)
C
dt
L1
(71)
g1 , g3 > 0
(72)
f 1 (i1 , i2 , v) = E1 R1 i1 v = 0
f 2 (i1 , i2 , v) = E2 R2 i2 v = 0
(73)
f 3 (i1 , i2 , v) = i1 + i2 g(v) = 0
Substituting the first and second equations into the third
equation, we have the following cubic function of v:
f (v) =
E1
E
+ 2
R1
R2
1
1
+
g1 v g3 v 3 = 0
R1
R2
(74)
iG
G
x(t) = x0 + (t)
L2
L1
i1
i2
R1
E1
R2
E2
(75)
Substituting Eq. (75) into Eq. (65), we have the linear variational equation as
(76)
1 2
D f(x0 , )( , ) +
2
(77)
It follows that the linear part A Df(x0, ) is a good approximation to the nonlinear function f( , ) near the equilibrium point x x0, and it is reasonable to expect that the
qualitative behavior of Eq. (65) near x x0 will be approximated by the behavior of Eq. (76). This is indeed the case if
the matrix A Df(x0, ) has no zero or pure imaginary eigenvalues. Hence we define an equilibrium point with this condition as a hyperbolic equilibrium point. That is, an equilibrium
point is hyperbolic if none of the eigenvalues of the matrix
A Df(x0, ) have zero real part. The HartmanGrobman
theorem shows that near a hyperbolic equilibrium point, the
nonlinear system of Eq. (65) has the same qualitative structure as the linear system of Eq. (76). That is, by a homeomorphism (continuous mapping with its inverse) h from an open
set U containing x0 of Eq. (65) onto an open set V containing
the origin of Eq. (76), trajectories of Eq. (65) in U map onto
trajectories of Eq. (76) while preserving their orientation by
time. Here qualitative structure, topological property, or
topological type has the same meaning.
Using this result, we can classify topologically hyperbolic
equilibrium point. Let
() = det[In Df(x0 , )] = 0
(78)
(79)
iC
(t) = A (t)
539
(80)
540
(a) Rn = Eu Es ,
is called the orbit of Eq. (65) through the initial state x0. Asymptotic behavior in the future or in the past is also defined
as
A(Eu ) = Eu , A(Es ) = Es
(81)
(t, x0 , )
0 t
(t, x0 , )
(x0 ) = (Orb(x0 )) = lim(x0 ) =
(86)
They are called the limit set and the limit set of x0 or
Orb(x0), respectively. Now suppose that p Rn be a sink.
Then the set of all point x0 whose limit set is p is called the
basin of attraction or the domain of attraction of the attractor p:
{0 O, 1 O, , n O}
Basin(p) = {x Rn | lim(x) = p}
(82)
(87)
(88)
0 t
(83)
W u (x0 ) W s (x0 ) = x0
(84)
(85)
a2
2O
0O
a1
1O
Im
Im
541
y2
Im
Im
Re
2 1
Re
(a)
Re
(b)
y1
Re
(c)
(a)
(b)
() =
1
1
= 2
+ 1 = 0
x = x x3 (x y)
(92)
y = y y3 (y x)
x = y
(89)
y = x +
(1 x2 )y
where we put
where
= 1 > 0,
y2
v1 =
=3 3 >0
1
(90)
C
x,
g3
v2 =
C
y,
g3
y1
A = Df(x0 , y0 ) =
(a)
(b)
y2
y2
y1
g1
> 0,
C
G
>0
C
(93)
y2
y1
(c-1)
(91)
3x20
3y20
(c-2)
y1
Figure 10. Phase portrait of a sink. (a) Node, (b) focus, (c) degenerate focus.
(a)
(94)
y2
y2
y1
y1
(b)
542
2
0O
1O
1O
0O
0O
1O
0O
2O
1
2
0O
2O
1O
1O
0O
1O
1 1O
0
1
0O
3 x
0O
1O
All equilibria are hyperbolic and their types are easily calculated from Eq. (94). The phase portrait is illustrated in Fig.
15. Stable manifolds approaching four saddle points separate
the phase plane into four regions in which one sink is situated. That is, the phase plane is divided into four basins of
attractions whose boundaries are stable manifolds of the saddle points.
x Rn , Rm
(95)
where x Rn is a state vector and Rm is a system parameter. Suppose that Eq. (95) have a periodic solution (t, x0, )
with period L. The orbit
C = Orb(x0 ) = {x Rn | x = (t, x0 , ), t [0, L]}
(96)
forms a closed curve in the state space Rn. This is an invariant set of Eq. (95). That is,
(t, C, ) = C
(97)
A small perturbation or variation (t) from the periodic solution obeys the following variational equation:
(t) = A(t) (t)
(98)
where
A(t) = Df((t, x0 , ), ) =
f
((t, x0 , ), )
x
(99)
x
Figure 14. Surface of the dissipative function, Eq. (47). Each stationary point corresponds to the equilibrium point in Fig. 13.
tR
(100)
2O
Figure 16. Phase portrait of the van der Pol equation, Eq. (89), with
0.5 and 1. The closed curve C indicates a stable limit cycle.
(101)
y = x
(1 x2 + x4 )y
Equation (101) is the van der Pol equation with a hard characteristic. That is, the nonlinear characteristic is assumed to
543
be a fifth-order polynomial. Figure 17 shows the phase portrait for 0.2, 3.5. Two limit cycles C1 and C2, one of
which is stable and another unstable, exist and the origin is
a sink 0O in this case. Hence we have two attractors: a stable
limit cycle C1 and a stable equilibrium point 0O. The basin of
the latter equilibrium point is the region surrounded by the
unstable limit cycle cycle C2. The outer region of the unstable
limit cycle C2 is then the basin of the stable limit cycle C1.
According to the initial state we specify, the final steady state
becomes the sink 0O or the stable limit cycle C1. This shows
an example of the existence of multistable states. In the circuit, if the initial state is small, then the oscillatory state is
never realized. To observe an oscillatory state corresponding
to the stable limit cycle we must give an initial state large
enough to enter the basin of the limit cycle C1. This oscillatory
process is called a hard oscillation. On the other hand, the
process stated in Example 6, item 1, is called a soft oscillation.
Poincare Map for Autonomous Systems
The definition of a Poincare map, or first return map, for a
periodic solution is quite simple. Suppose that Eq. (95) has a
periodic solution (t, x0, ) through the point x0. We choose a
hypersurface Rn to intersect transversally the periodic
orbit at x0, Then for each point x1 sufficiently near x0,
the solution through x1 will return to again at a point x2
near x0 (see Fig. 18). The mapping x1 x2 is called the
Poincare map T. That is, T is defined as
T:
;
x1
x2 = T (x1 ) = ( , x1 , )
(102)
(103)
(104)
Rn
C1
(t, x1, )
0O
x2
x2 = ( (x1), x1, )
C2
x0
x1
x2 = (L, x0, )
(t, x0, )
Figure 18. Periodic orbit and Poincare map. Local cross section is
the hypersurface .
544
form
xk+1 = T (xk ),
xk ,
k = 1, 2, . . .
(105)
(106)
That is, x0 is a fixed point of T. If L is the period of the periodic solution, then the return time becomes (x0) L.
Stroboscopic Mapping: Poincare Map
for Periodic Nonautonomous Systems
x = f(t, x, ), x Rn , Rm
(107)
where x Rn is a state vector and Rm is a system parameter. We assume that f is periodic in t with period 2:
f(t, x, ) = f(t + 2, x, )
(108)
for all t R. Equation (107) describes a class of dynamic nonlinear circuits with a periodic forcing term. A nonlinear circuit with an ac operation is a typical example of this class.
Without loss of generality, we assume that the period of the
external forcing is 2. Suppose that Eq. (107) has a solution
x(t) (t, x0, ) with x(0) (0, x0, ) x0. This time we
have the periodic property equation, Eq. (108). Hence for every 2 instance the vector field equation, Eq. (107), returns
the same value so that a stroboscopic sampling of the solution
can be achieved under the fixed vector field. That is, we can
define the stroboscopic mapping as the Poincare map (see Fig.
19):
T:
R R ;
n
x0
x1 = T (x0 ) = (2, x0 , )
(109)
(110)
Rn
x0
(t, x0, )
Rn
(a) Rn = Eu Es ,
DT (Eu ) = Eu ,
x1
DT (Es ) = Es
t=0
(111)
n1 I}
(113)
where D and I denote the type of the fixed point and the subscript integer indicates the dimension of the unstable subspace: k dim Eu. Usually a completely stable fixed point 0D
is called a sink, a completely unstable fixed point nD is called
a source, and others are called saddles.
t = 2
Figure 19. Stroboscopic mapping. Poincare map for a periodic nonautonomous system.
(114)
d (t, x0 , )
f(t, (t, x0 , ), ) (t, x0 , )
=
dt
x0
x
x0
(0, x0 , )
= In
x0
(t, x0 , )
,
x0
(115)
(122)
This means that (L) has the unity multiplier with the eigenvector:
(0,
x0 , ) = f(x0 , )
(123)
(124)
(125)
W u (x0 ) W s (x0 ) = x0
(0) = In
(116)
T (x0 )
(2, x0 , )
=
= (2 )
x0
x0
(117)
L
f( , ( , x0 , ), )
d
trace
(118)
x
0
L
f( , ( , x0 , ), )
d > 0
= exp
trace
x
0
Thus T is an orientation preserving diffeomorphism.
3. In an autonomous system, Eq. (95), a periodic solution
always has at least one characteristic multiplier that is equal
to unity. Indeed, a periodic solution satisfies the relation
(t,
x0 , ) = f((t, x0 , ), )
(119)
Differentiating by t yields
(t,
x0 , ) =
(L,
x0 , ) = (L)(0,
x0 , ) = (0,
x0 , )
545
f((t, x0 , ), )
(t,
x0 , ) = A(t)(t,
x0 , ) (120)
x
x0 , )
(121)
(127)
(128)
(129)
546
x(k+1) = x(k) + h,
DF(x
(k)
)h = F(x
k = 0, 1, 2,
(k)
(130)
S2
1D
S1
(131)
2
Then we have:
(a) If 0 a2 1, 0 (1), 0 (1), then the hyperbolic
fixed point is 0D,
(b) If 0 a2, 0 (1), then the hyperbolic fixed point is
1D,
(c) If 0 a2, 0 (1), then the hyperbolic fixed point is
1I,
(d) If 1 a2, 0 (1), 0 (1), then the hyperbolic fixed
point is 2D.
These relations are illustrated in Fig. 20.
2. Periodic solutions of Duffings equation. Consider the following Duffings equation [cf. Eq. (20)]:
x = y
(132)
(1) = 0
(1) = 0
2D
1
0D
(133)
(134)
(135)
Substituting Eq. (135) into Eq. (134) and equating the same
power of , we have
a2
1D
Figure 21. Phase portrait of the Poincare map defined by Eq. (132).
Closed dotted curves indicate periodic solutions.
0:
x0 + 2 x0 = B cos t
1:
x1 + 2 x1 = x0 cx30
x2 + x2 =
1I
a1
3cx20 x1
(136)
x1
B
cos t
2 1
B
3cB3
sin
t
cos t
x1 (t) = 2
( 1)2
4(2 1)4
cB3
cos 3t
2
4( 1)3 (2 9)
x0 (t) =
(138)
0 : x0 + x0 = 0
(140)
(141)
where M0 and N0 are unknown coefficients. They are determined as follows: Substituting Eq. (141) into the second equation of Eq. (140), we have
!
3
x1 + x1 = M0 + a cr2 N0 sin t
4
!
3 2
+
a cr M0 N0 + b cos t
4
(142)
1
2
2
+ c(N0 3M0 )N0 sin 3t
4
1
+ c(3N02 M02 )M0 cos 3t
4
where r2 M02 N02. Equation (142) has a periodic solution if
and only if the following conditions are satisfied:
3
P(M0 , N0 ) = M0 + a cr2 N0 = 0
4
(143)
3
Q(M0 , N0 ) = a cr2 M0 N0 + b = 0
4
When a solution (M0, N0) of Eq. (143) is an isolate root, i.e.,
the Jacobian matrix:
P
M
0
Q
M0
P
N0
Q
N0
3
a cr2
4
2
+
r2 = b 2
(146)
(139)
(145)
(148)
0O
1.4
= 0
1.2
1.0
0.7
1O
0.8
0.6
0.4
0O
0.2
0
(144)
(147)
B =
b
1
c(N02 3M02 )N0 sin 3t
32
1
(137)
547
0.2
0.4
b
0.6
0.8
1.0
548
where
2.0
0O
b=
1.5
a1 = { (1) (2 ) + (2) (2 )} = 2
{1(1) (2 ) + 1(2) (2 )}
2 {2(1) (2 ) + 2(2) (2 )}
1.0
0.5
1O
+
2 {2(1) (2 ) + 2(2) (2 ) + 1(1) (2 )1(2) (2 )
1.0
r
1(2) (2 )1(1) (2 )} +
b = 1.0
0.5
0.5
0.3
0O
0.2
(1) = 1 a1 + a2 = 4 +
{ } + > 0
0.1
0
(155)
(1) = 1 + a1 + a2
2.0
1.0
=
2 {1(1) (2 )1(2) (2 ) 1(2) (2 )1(1) (2 )} +
3 { } +
3.0
a2 = 1 +
{1(1) (2 ) + 1(2) (2 )} +
(156)
where
the variational equation becomes
1(1) (2 )
+ =
{3c( (t))2 + }
(149)
(2)
(t) =
0(2) (t)
1(2) (t)
2 2(2) (t)
k(1) (0)
k(2) (0)
= k(1) (0) = 0
(150)
0(1) (0)
0(2) (0)
0(1) (0)
0(2) (0)
= 1,
= 0,
= 0,
= 1,
= k(2) (0) = 0
(k = 1, 2, . . .)
(151)
Substituting Eq. (150) into Eq. (149) and equating the same
power of , we have
0(1)
0(2)
1(1)
1(2)
+
+
+
+
0(1)
0(2)
1(1)
1(2)
=0
=
{3c( (t))2 0(1) + 0(1) }
(152)
=
{3c( (t))2 0(2) + 0(2) }
0(1) = cos t,
0(2) = sin t
t
3c( ( ))2 cos + sin sin(t )d
1(1) (t) =
1(2) (t) =
0
2
=
0
1(2) (2 ) =
1(2) (2 ) =
0
2
Q
M0
P
M0
P
N0
Q
N0
(157)
(1) =
2 det A +
3 { } +
a2 = 1 +
trace A +
3 { } +
(158)
where
P
M0
A=
Q
M0
P
N0
(159)
N0
trace A = 2 < 0
det A = a2 + 2 3acr2 +
(153)
(1) (2 )
() =
(1) (2 )
=0
1(1) (2 )
(2) (2 )
= 2 + a 1 + a 2 = 0
(2) (2 )
(154)
27 2 4
db2
c r =
16
dr2
(160)
Note that from the first equation of Eq. (160), we see that no
completely unstable type of periodic solution exists in Eq.
(139). Hence the stability of the periodic solution is determined as follows:
(a) If det A 0, that is, db2 /dr2 0, then the periodic
solution is a completely stable type: 0D.
(b) If det A 0, that is, db2 /dr2 0, then the periodic
solution is a directly unstable type: 1D.
(161)
(162)
=
G(u cos t + v sin t, u sin t + v cos t,
, t) sin t
v(t)
=
G(u cos t + v sin t, u sin t + v cos t,
, t) cos t
(163)
2 0
G(u cos + v sin , u sin + v cos ,
, ) sin d
(164)
2
v(t)
=
2 0
G(u cos + v sin , u sin + v cos ,
, ) cos d
u(t)
0O
1O
1
S
2
R
N
2
Figure 25. Phase portrait of Eq. (165) with a c 1, 0.1, and
b 0.3.
!
3
u + a cr2 v = P(u, v)
u =
2
4
2
!
3
a cr2 u v + b = Q(u, v)
v =
2
4
2
(165)
where P(u, v) and Q(u, v) are given in Eq. (143). An equilibrium point of Eq. (165) gives a periodic solution Eq. (162).
Hence we have the correspondence between the equilibria of
Eq. (165) and the periodic solutions of Eq. (161). Moreover,
the phase portrait of Eq. (165) gives information about global
behavior of the solutions of Eq. (161). Figure 25 shows a
phase portrait of the case where three equilibria exist in Eq.
(165). We see two sinks R and N corresponding to the resonant and nonresonant solutions, respectively. We also illustrate a saddle S whose stable manifold forms the basin
boundary of two attractors.
References in This Section
The Poincare map stated in this section is discussed in any
books on nonlinear dynamics; for example, see Refs. 4, 9, and
12. Hyperbolicity of fixed point of the Poincare map is introduced in Refs. 7 and 12. The classification of the hyperbolic
fixed point is found in Refs. 1721. Various numerical methods are well treated in Refs. 22 and 23. Various nonlinear
resonancesthat is, subharmonic resonance and higher harmonic resonance as well as harmonic resonanceare treated
in the standard books on nonlinear oscillations (see Refs. 3
6). For the perturbation method stated in the last paragraph,
see Ref. 24. Practical applications of the averaging method
are found in Ref. 3.
BIFURCATIONS OF EQUILIBRIA AND PERIODIC STATES
0O
0 b = b1
549
b = b2
When the system parameter varies, the qualitative properties of the state space may change at 0. We may observe
the generation or extinction of a couple of equilibria or fixed
points, the branching of new equilibria or fixed or periodic
points, and the change of a topological type of equilibrium
550
The Hopf Bifurcation. This bifurcation is observed if a couple of characteristic roots becomes purely imaginary numbers
at 0. The stability of the equilibrium point changes and
a limit cycle appear or disappear after the bifurcation. Symbolically we have the following relation:
x Rn , Rm
(166)
where x Rn is a state vector and Rm is a system parameter. Suppose that x0 Rn is an equilibrium point of Eq.
(166):
f(x0 , ) = 0
(167)
(168)
() = det(In A())
= n + a1 n1 + + an1 + an = 0
(169)
d Re()
= 0
d =
(170)
(172)
kO
kO
+ LC(k+1 D)
k+2 O
+ LC(k D)
k+2 O
(k = 0, 1, . . . , n 2)
(k = 0, 1, . . . , n 2)
(173)
where LC(kD) denotes a limit cycle whose type of the corresponding fixed point of the Poincare map is kD. The first relation shows that before the bifurcation a k-dimensionally unstable hyperbolic equilibrium point exists, and after the
bifurcation the equilibrium point becomes a (k 2)-dimensionally unstable and k-dimensionally unstable limit cycle of
the type kD appears. If k 0, then a sink becomes two-dimensionally unstable and an orbitally stable limit cycle appears.
This type of Hopf bifurcation is called a supercritical type,
whereas the second relation shows a subcritical type (see Fig.
26). The bifurcation condition is then given by
( j) = det( jIn A()) = 0
(174)
k+1 O
(k = 0, 1, . . . , n 1)
(171)
( j) = ( j)2 + ja1 + a2 = 0
(175)
Hence we have
a1 = 0,
a2 = 2 > 0;
a2
(176)
(177)
551
k+ 1D
Equilibrium point
Equilibrium point
kO
k+ 2O
kO
Limit cycle
k+ 2O
Limit cycle
= 0
= 0
(a)
(b)
Hence we have
a1 a2 + a3 = 0,
a2 > 0;
= a2
(178)
(179)
Bifurcation of a Fixed Point
Hence we have
a4
> 0;
a2
a4
a2
(180)
(181)
() = det(In DT (x0 ))
(182)
= n + a1 n1 + + an1 + an = 0
0.7
1O
0O
0.5
0.5
1
b
00
Amplitude characteristics
1
0.4
0
0
0.5
0.3
t1
0.2
t2
0.1
3
0
0.6
0.5
0
1
Bifurcation diagram
Figure 27. Characteristic surface of Eq. (146) with amplitude characteristic curves and bifurcation diagram.
0.1
0.2
0.3
b
0.4
0.5
0.6
0.7
Figure 28. Bifurcation diagram of equilibria. Tangent bifurcation occurs on the curves t1 and t2, and the cusp point C is a degenerate
tangent bifurcation point.
552
Tangent Bifurcation of Fixed Point. Under the change of parameter , at 0 the generation or extinction of a couple
of fixed points occurs. The types of bifurcation are
?
?
k1 D + k D
k1 I
(183)
+ kI
= 1 + a1 + + an1 + an = 0
Period-Doubling Bifurcation. If a real characteristic multiplier passes through the point (1, 0) in the complex plane,
then the original fixed point changes its type and 2-periodic
points are branching. This bifurcation is called a period-doubling bifurcation. The types of bifurcation are
kD
kI
kI
k+1 I
+ 2 k D2
k1 I
+ 2 kD
k+1 D + 2 k D
k1 D +
kD
k2 D +
kI
k+2 D
kI
k2 D +
+ ICC
(186)
(191)
sin 2 + a1 sin = 0
Hence we have
a2 = 1,
2 < a1 < 2
(192)
(e j ) = e j3 + a1 e j2 + a2 e j + a3 = 0
(193)
2 < a3 a1 < 2
(194)
Second consider autonomous systems. In this case the characteristic equation has at least one unity multiplier. Thus we
factor the characteristic equation as
() = n + a1 n1 + + an1 + an = ( 1)A () = 0
(195)
where
A () = n1 + b1 n2 + + bn1 ,
bk = 1 +
k
ai (196)
i=1
A (e j ) = e j2 + b1 e j + b2
(187)
(197)
(198)
or, equivalently,
= e jn + a1 e j(n1) + + an1 e j + an = 0
cos 2 + a1 cos + a2 = 0,
ICC
(e j ) = det(e j In DT (x0 ))
That is,
a1 (a3 a1 ) + a2 = 1,
+ ICC
ICC
(190)
or equivalently
The NeimarkSacker Bifurcation. Similar to the Hopf bifurcation for equilibrium point, a fixed point becomes unstable
and there may appear an invariant closed curve of the Poincare map. Here the invariant closed curve C is a closed curve
in Rn such that T(C) C, which corresponds to doubly periodic oscillation in the original periodic nonautonomous system. This bifurcation indeed occurs if a pair of the characteristic multipliers and pass transversally through the unit
circle except for the points (1, 0) and (1, 0). The types of the
bifurcation are
k+2 D
(e j ) = e j2 + a1 e j + a2 = 0
(185)
2 k D2
kD
(189)
Note that in this case we need an additional inequality satisfying the condition: cos 1.
(184)
NS (x0 , ) = 0
() = det(In DT (x0 ))
kD
a2 + a1 = 0,
3 < a1 < 1
(199)
(188)
The numerical determination of the codimension one bifurcation value 0 and the location of the nonhyperbolic fixed
(200)
(201)
3 2
1 r u v = f (u, v)
u =
2
4
u + 1 r2 v + B = g(u, v)
v =
2
4
(202)
where
a1 = 3 r2 2
a2 = 2 + 1 3 r2 +
2( 1)
(203)
9
27 2 4
3
r = 2 + 1 r2
1 r2
16
4
4
(208)
where
r2 = u 2 + v 2 ,
553
x + (a + b cos 2t)x = 0
(209)
x = A(t)x
(210)
or, equivalently,
3 2
r u v = 0
4
3
u + 1 r2 v = B
4
1
(204)
0O
2
3 2
2
1 r
+ r2 = B2
4
1O
(205)
a1 = 0
a2 = 0
P
2O
f
u
g
u
3
f
2
2
1 4 (3u + v )
v
=
g 2
3
uv +
v
2
3
uv
2
3
1 (u2 + 3v2 )
4
(206)
v = 2 + a + a = 0
() = u
1
2
g
g
u
v
0.5
1.5
P
1
B 0.5
(207)
0.5
0
0
1
0
1
0.5
0.5
Figure 29. Characteristic surface of Eq. (205) and projected bifurcation diagram.
554
0.6
C1
C2
h2
0O
0.5
h1
t2
0.4
2O
0.3
h1
2O
0O+ 1O+ 2O
0O
t2
0.1
0O+ 1O+ 2O
0
0.8 0.6 0.4 0.2
A(t) =
(a + b cos 2t)
(211)
(t) =
1 (t)
2 (t)
1 (t)
2 (t)
(212)
R2 R2 ;
x0
x1 = ( )x0
(213)
(214)
where we put
m = 1 ( ) + 2 ( ) = 1 + 2
(215)
0 trace A ( ) d
=1
(216)
2) log 22, and (t) and (t) are periodic functions with
period 2.
(c) If m 2, then the origin is a nonhyperbolic fixed
pointthat is, a center-type fixed point. The general
solution is then a doubly periodic function. In the last
case the characteristic multipliers lie on the unit circle
in the complex plane:
1 = 2 = exp( j ),
j = 1,
= tan
4 m2
m
(218)
t1
(a)
where
x
x=
,
y
2O
t1
t3
0.2
0O+2x 1O
(217)
(219)
or, equivalently,
x = y
y = kx (a + b cos 2t)x x3
(220)
(221)
(222)
555
20
1I
1D
m = 10
15
m = 50
1I
20
10
10
m = 15
0
1
10
1
0
10
2
5
2
2
0
5
10
100
400
1D
0
1
15
where
a2 = (0) = det( ) = det(0)ek < 1
(223)
of Fig. 33(a). This is the typical parametric excitation phenomenon, also called the parametric resonance. Stable 2-periodic points 0D12 and 0D22 T(0D12) have the period 2 which is
the double period of the injected pumping signal. Traversing
the curve P2 from the region B to C, we observe that two directly unstable 2-periodic points branch off from the origin
and the origin itself becomes a completely stable fixed point.
Hence in the region C we have the phase portrait shown in
Fig. 33(b). On the tangent bifurcation curve T, 2-periodic
points 0D12 and 1D12, (and also 0D22 and 1D22) coalesce and disappear in the region A below the curve T.
References in This Section
Many books are available on bifurcation theory. We refer to
only a few of them: Refs. 79, 25, and 26. For the higherorder bifurcationsthat is, codimension two bifurcations
see Refs. 9, 26, and 27. Various numerical methods are stated
in Refs. 21 and 2628. Harmonic synchronization is analyzed
in Refs. 3, 6, and 9. Mathieus equation and more generally
the linear periodic differential equations are well surveyed in
1I
y
2
0D1
2
0D1
P1
P2
0D
1I
2
D
0 2
0D
T
c
0
2
1D1
2
1D2
0D
2
0D2
a
Figure 32. Bifurcation diagram of Eq. (219) with k 0.1. Curves
P1 and P2 indicate the period-doubling bifurcation of the origin, and
T denotes the tangent bifurcation of 2-periodic points.
(a)
(b)
Figure 33. Phase portrait of the time Poincare map of Eq. (219).
(a) k 0.1, a b 1.0; (b) k 0.1, a b 0.54.
556
(a)
(b)
(c)
(224)
y = x +
{(1 y2 )y cx3 + B cos t}
3
3
1 r2 u cr2 v
2
4
4
3 2
3 2
cr u + 1 r v + B
v =
2
4
4
u =
(225)
2( 1)
(226)
3
1 r2
4
2
3 2
r2 = B2
+ cr
4
(227)
3
t2
h1 t
1
2
1O
1O2
t1
h2
Q
0
1
(a)
h1
R
t2
SL
1O1
(b)
Figure 34. Schematic diagram of a saddle to saddle orbit. (a) A saddle connection orbit, or heteroclinic orbit; (b) a separatrix loop, or
homoclinic orbit.
t3
0
557
(a)
(b)
Homoclinic Point
Now we consider the phase portrait of the Poincare map. We
focus our attention to the behavior of invariant manifolds
stated in Remark 5(1) [see Eq. (124)] and its related property.
For simplicity consider a two-dimensional periodic nonautonomous system defined by Eq. (107):
x = f(t, x, ),
x R2 , Rm
(228)
R2 R2 ;
x0 = (x0 , y0 )
x1 = (x1 , y1 )
= T (x0 ) = (2, x0 , )
(230)
(x) = Q
(231)
(229)
h1
t2
P
t1
SL
1D
c
a
d
Figure 39. Schematic diagram of a horseshoe map on the small rectangle abcd near homoclinic points. The region comes back to the
curved rectangle abcd after some finite iteration of the Poincare
map T.
558
C
H0
H1
H2
1D
x = y
y = 0.02y x3 + 0.3 cos t 0.08
H2
H1
(233)
indicated as
H2 = T 2 (H0 ),
H1 = T 1 (H0 ),
H0 , H1 = T (H0 ),
H2 = T (H0 )
2
Figure 42(a)(c) shows the images T(R), T 2(R), and T 3(R), respectively. We see that the number of intersections,
Horseshoe Map
A horseshoe map on a rectangular region contains a complex
invariant set. This is a typical chaotic state. Hence we summarize briefly some of the properties of the map. Let a map
T have a directly unstable fixed point 1D and let its and
branches intersect each other, forming homoclinic points as
schematically illustrated in Fig. 41. Then the map from the
rectangular region R ABCD into the plane becomes a horseshoe map. In the figure, images of the homoclinic point H0 are
#{R T k (R)} = 2k
(k = 1, 2, . . .)
(234)
increases as 2k while forming the vertically very narrow rectangles. And the positively invariant set becomes
R = R
T k (R) CantorSet I
(235)
k=1
y
A
R = R
C
T k (R) I CantorSet
(236)
k=1
D
B
1D
A
A
B C
(a)
Figure 40. Phase portrait of the Poincare map defined by Eq. (233).
D
(b)
(c)
R
= R
T k (R)
k=1
2S
k=1
(237)
559
1S
1S
2S
1. R
has countably many periodic points of T with an
arbitrarily high period, and these periodic points are all
of the saddle type.
2. R
has uncountably many nonperiodic point of T.
3. R
set R
is chaotic. Note that this chaotic invariant set R
is
unstable because of the above properties 1 and 3. To observe
R
as an attractor, there exists another mechanism to encapsulate this invariant set in some bounded region of the phase
plane. Mathematically, this problem is not yet solved completely. In circuit dynamics, however, this mechanism may be
achieved by the dissipative property of the circuit.
(238)
1I
2k
+ 2 0 D2
k+1
k = 0, 1, 2, . . .
(b)
S
2S
(c)
S
2
(239)
In many systems with weak dissipation, this bifurcation occurs successively until k tends to infinity under the finite
change of parameters. The universality of this cascade of bifurcations is studied by Feigenbaum. We illustrate this cascade by the following example.
(d)
Figure 43. Phase portrait of the Poincare map defined by Eq. (238).
Homoclinic points appear when k becomes small. (a) k 0.1, (b) k
0.05, (c) k 0.005, (d) k 0.
0D
(a)
y = 0.1y x3 + B cos t + B0
(240)
Figure 44 shows the bifurcation diagram of a fixed point corresponding to a nonresonant oscillation. P and T denote the
period doubling and tangent bifurcation curves, respectively,
on which these bifurcations appear. The superscript k indicates the k-periodic point, and the subscript shows the numk
ber for distinct curves. On the curve labeled P12 the bifurcation process of Eq. (239) occurs. These curves accumulate on
just the inner region of the curve P18 so that in the shaded
region we see a chaotic state. Phase portraits of the perioddoubling cascade are shown in Fig. 45. Stable 2-periodic
points exist in Fig. 45(a), which bifurcate into 4-periodic
points in Fig. 45(b). Chaotic states separated into four groups
appear in Fig. 45(c). They gather as two parts in Fig. 45(d)
and finally coalesce into one big attractor in Fig. 45(e). The
attractor grows until it touches the branch of the directly
unstable fixed point D. After intersecting, the chaotic state
loses its attractivity and the attractor disappears, although
an unstable chaotic state exists.
Lyapunov Exponent to Measure a Chaotic State
To determine whether an attractor is chaotic or not, we have
a conventional method of evaluating the mean value of the
560
0.15
P1
P12
0.1
P18
I
I22 S 4
2
S44
B0
S22
T12
0.05
S14
I12
S34
S12
P14
(b)
(a)
T22
P22
0.0
T32
0
0.2
0.1
0.3
0.4
B
D
Figure 44. Bifurcation diagram of Eq. (240). Period-doubling cascade
appears on a fixed point corresponding to a nonresonant periodic solution.
Ch41
I12
Ch42
Ch44
2
Ch43
k 2k
I2
(d)
(c)
(241)
y
Ch21
I
Ch22
(242)
D
Ch
Ch
x
(f)
(e)
Figure 45. Phase portraits of the Poincare map defined by Eq. (240)
with B0 0.075: (a) B 0.15, (b) B 0.185, (c) B 0.195, (d) B
0.197, (e) B 0.199, (f) B 0.217.
0.08
Chaotic attractor
0.06
0.04
P4
0.02
0
0.02
0.04
0.06
0.10
P2
P
4
0D
0D
0.12
0D
P8
P
E
0D
0.14 0.16
B
0.18
0.2
0.22
561
562
NONLINEAR EQUATIONS
HIROSHI KAWAKAMI
The University of Tokushima