Computer Methods in Applied Mechanics and Engineering 91 (1991) 1109-1121

North-Holland

The Hopf bifurcation and limit cycle by the

incremental harmonic balance method
S.L. Lau and S.W. Yuen
Department ofCivil and Structural Engineering, Hong Kong Polytechnic, Hong Kong

Received 24 October 1990

It is well recognized that periodic solutions and their stabilities under small perturbations play a very
im~rtant role in the study of nonlinear dynamic systems. These problems have been investigated by
various perturbation methods which are, however. subjected to the limitation of weak nonlinearity.
Moreover, a perturbation method cannot define the parameter range within which a correct solution or
a satisfactory approximation can be obtained by the method. This situation may lead to totally
incorrect results when the parameter is out of a certain range. By introducing a variable parameter the
incremental harmonic balance method, which can treat strongly nonlinear dynamic problems, can be
used to perform parametric studies on the Hopf bifurcation and limit cycle problems. This method is
ideally suited for parametric studies, giving solution diagrams. In this paper it is applied to study the
Van der Pal oscillator and coupled oscillators. Examples show that beyond a certain limit, the standard
perturbation methods will give incorrect results.

1. Introduction

Bifurcations are important phenomena which represent significant qualitative changes of

behaviour in a dynamical system when the parameters on which this system depends are
changed. In particular, the Hopf bifurcation is one of the representative examples. It is a study
of the appearance of periodic solutions of nonconservative system out of an equilibrium state
[l, 21. Some of its applications of the Hopf bifurcation theory can be found in [3]. The study of
bifurcation phenomena has received much attention recently because of its close relation with
the study of chaotic motion, as mentioned by Awrejcewicz and Mrozowski [4].
Periodic solutions play a very important role in the study of nonlinear dynamic systems
[5,6]. Particularly, periodic solutions together with their stabilities are especially significant in
the analysis not only of local bifurcations, but also of certain global properties in the dynamics
of the nonlinear system. In [7], it was stated that the classical nonlinear analysis is aimed at (i)
seeking all possible equilibrium states and periodic motion of the system, (ii) determining the
stability of these solutions, and (iii) finding the global domain of attraction in the state space
for each asymptotically stable equilibrium state or periodic motion. For the first two
objectives, various perturbation techniques were employed in classical nonlinear analyses.
Reviews of these methods can be found in [8,9].
However, these perturbation techniques are restricted to solve weakly nonlinear problems
only. Small parameters are essentially assumed so that perturbations can be carried out. More
importantly, it is still unknown whether the nonlinearity is weak enough for which the

00457825/91/$03.50 @ 1991 Elsevier Science Publishers B.V. All rights reserved

1110 S. L. Lau. S. W. Yuetl, The Hopf’ bifurcation and limit cycle

techniques can be applied adequately. A common mistake will be made when the parameter is
considered to be small, but actually it is not! Consequently, it will only result in an erroneous
solution.
Compared to the perturbation, the numerical integration (NI) method can treat strongly
nonlinear problems very well. However, the point to point integration makes it not efficient.
The convergence rate of getting a steady-state solution will be a problem of the method as the
rate will be extremely slow, especially at points of bifurcations or for the cases of low
damping.
It is also known that rhe harmonic balance (HB) method can be used in nonlinear dynamic
analyses. But the HB formation of a set of nonlinear algebraic equations in terms of the
Fourier series is a deficiency in the method. The formulation will be very tedious if more
harmonic terms are involved. For higher accuracy is desired, higher harmonic terms are
needed and the whole formulation are required to be reformed.
In comparison, the incremental harmonic balance (IHB) method can give accurate results
and without difficulty in increasing the accuracy of the results. Lau and Cheung [lo] originally
presented the analyses of periodic structural vibrations by this method and it has also been
successfully applied to various types of nonlinear dynamic problems. For examples, it was
extended to treat dynamic instability of nonlinear structural systems [ll, 121 and dry friction
damper problems [13]. With the solution expanded into a multiple Fourier series containing a
number of frequencies incommensurable with each other, the IHB method is also capable of
treating quasi-periodic structural vibrations [ 141. Recently, this method is further generalized
to deal with piecewise-linear systems [15].
The IHB method is simple and convenient for computer implementation. In this method
nonlinear differential equations are reduced to a set of linearized incremental algebraic
equations in terms of the Fourier coefficients. So, only linear equations have to be formed and
solved iteratively in each incremental step. Many equation solvers are readily available to
solve the resultant linear algebraic equations, such as the Gaussian elimination. The Fourier
increments are then updated until a convergent solution has been reached. Both stable and
unstable solutions can be sought directly by this algorithm.
The IHB method with a variable parameter is ideally suited to parametric studies for
obtaining solution diagrams. After obtaining the solution of a particular value of the
parameter, this parameter will then be incremented. The new solution may be sought again by
iteration using the previous solution as an approximation, then the neighbouring solution can
be found. By successful incrementation and iteration, the whole solution diagrams of a
dynamic system can easily be traced. To be more effective, an arc-length extrapolation
procedure can also be employed to predict the neighbouring state from some known ones; the
required number of iterations to converge can further be reduced. The stability of the
computed solutions is readily analyzed by the use of the Floquet theory with Hsu’s scheme,
which has been fully explained by Friedmann et al. [16]. Finally, the global domain of
attraction in the state space for each asymptotically stable equilibrium state or periodic motion
can also be sought, for example, by using simple cell mapping [7].
In this paper, the IHB method is applied to obtain solution diagrams, including the Hopf
Bifurcation. Two examples are presented. One is a study of single degree-of-freedom (DOF)
autonomous systems with Hopf bifurcation, using the well-known Van der Pol equation. A
very strong nonlinearity is assigned in the equation so that it is not always possible to be
 S.L. Lau, S.W. Yuen, The Hopf bifurcation and limit cycle 1111

solved by the standard perturbation methods. The other one is a study of the solution diagram
of coupled oscillators. In this example, attention is drawn on the solution domain of the
equations. Incorrect solution will be obtained without the solution diagram.

2. IHB formulation

In this section, the IHB method is formulated to analyze n-DOF system vibrations. The
motion of nonlinear oscillators is generally governed by the following system of simultaneous
equations:

02iii+fi(u,,. . . ,u,,ci,,. . . ,li,*,q A)=% i=l,...,n, (I)

where a dot denotes differentiation with respect to the dimensionless time variable T, and
fi(u,, * * * , u,,, i,, . . * , tin,W, A) is a nonlinear function of the oscillation frequency w, the
parameter A, the dependent variable ui and its velocity ci,.
The new dimensionless time variable T is defined as

7 = wt ) (2)

in which t is the time variable. The parameter A is added to the governing equation on the
purpose for parametric studies.
The procedures of the IHB method for seeking periodic solutions is mainly divided into two
steps. The first step is a Newton-Raphson process; small increments (symbolized by A) are
added to the current solution (or a guessed solution at the beginning of the procedure) of (l),
i.e. wo, A,, and uio (i = 1,2, . . . , n), to get a neighbouring solution

o=w,,+Aw, A=A,,+Ah, U; = uio + Au.I ’ i=l,...,n. (3)

Expanding (1) by Taylor’s series about the initial state, the linearized incremental equations
yielded by ignoring all the nonlinear terms of small increments are

w;Aiii+i j=, ($)OAtii+l% ($),,“i

I

= - (&ii, + &) - ((~),j+2~&,i)A~-(~)0AA, i=l,..., n. (4)

This system of variational equations is equivalent to the linear matrix equation for the
increments:

w;AX+CAX+K.AX=R-(2w$,+Q).Ao-P-AA, (5)

where

AX = [AU,, AuZ, . . . , Au,]’ and X0 = [u ,,,, uZO,. . . , u,,J’ . (6)7 (7)

1112 S. L. Lau, S. W. Yuen, The Hopf bifurcation and limit cycle

The elements of the matrices K and C are given by

Kji=($),,,c;~=($-),,
I I
(a (9)

and the elements of the vectors Q, P and R are

(lo), (ll), (14

It should be noted that R in (5) is a corrective term to prevent the incrementation process
drifting away from the actual solution.
The second step of the IHB method is the Galerkin procedure. Assume that the unknown
u,, can be expanded into a truncated Fourier series:

c
uio = n’n,.“?. a;, cos 127+ c b;,sinmT, i=l,. . . ,n, (13)
nl=m,, rn?.

where the progressions n,,n,,... and m,,m,,... are positive integers. They may be
0, 1,2, . . . or 1,3,5, . . . etc., subject to the circumstance.
Accordingly, Au is expressed as

Au;= 2 Au;,, cos 11~+ 2 Ab;,,, sin rn7 i= 1 ,...,n. (14)

n=tl,,l72. tTl=m,. mz....

For conciseness, the functions u0 and Au can be rewritten as

u;o= Ta;,, 7
i=l,...,n, Aui = T Aai i= 1,. ,n, (15) (16)

where

T=[cosn,~,~~~n~~~ ,..., sinm,r,sinm,r, *I7 (17)

ajo = [a:,,,, &,, . . . , by’,, bym2,. . .I’ , (18)
Aai = [Au,,,, Auin2, . . . , Abim,, Abi,,,, . . .I’ . (19)
Therefore,

X,=Y-A,,, AX=Y-AA, w%(21)

where

Y=L’ ,’ *.: ,.I. (22)

A,, = [a:,,ai,,. . . ,a&l’ , AA = [Aa:, Aa:, . . . , Aa:]’ . (23) (24)

 S.L. Lau, S.W. Yuen, The Hopf bifurcation and limit cycle 1113

With these notations, we have

ito= I’A, , &=iiA,. (25)

By substituting (20), (21) and (25) into (5), and applying the Galerkin procedure for one
period, we obtain
2Tr

6 AA’ Y’[,;Y+ CY+EY]d7 AA

I
0
27l 27

=&AA’ Y’ . R dr - Y’ . [Q + 2w,,Y - A,,] dr - Aw - Y’.Pdr.Ah , (26)

I 1
0 J
or simply

kAA=r+qAw+pAh, (27)
where the matrix
2n

k=
I
Y’+;Y+CY+K.Y]dr (28)
0

and the vectors

r= ~~~~rt.~.A~-Yt.~~,d~,
0
277

q = 7
0
[2001;‘I’Ao - Y’Q] dT, P=
-

0
I Y'P d7 .

3. Numerical examples

In this paper, analyses of the limit cycles with bifurcation problems are carried out by the
IHB method. One is the Van der Pol oscillator of great interest, which is of single
degree-of-freedom system (n = 1). The equation involved is the simplest in form and Hopf
bifurcation will be examined by performing a parametric study. The other one is analysis of
the steady state behaviours of two coupled oscillators (n = 2). The solution diagram is also
found to be very useful in solving such kind of nonlinear vibration problems.

3.1. Van der Pol oscillator

The Van der Pol equation is known to have the occurrence of the Hopf bifurcation. The
1114 S. L. Lau. S. W. Yuen, The Hopf bifurcation and limit cycle

typical non-dimensional form of the equation being considered is

o’ii - &( A - u2)ozi+ u = 0, (32)

where a dot denotes the differentiation with respect to the dimensionless time variable T and a
variable parameter A is introduced to investigate the Hopf bifurcation. Clearly, (32) is a
special case of (1) with y1= 1. A number of papers have already discussed this equation for the
case A = 1.0. For instance, Hassard et al. [3] used this equation to study a resistor-inductor-
capacitor (RIC) electric circuit. Burton [9], Minorsky [18] and Nayfeh and Mook [19] used it
as illustrative examples. The value of E is essentially assumed to be very small so that the
perturbation procedure can be applied.
However, it is not possible to identify the range of E within which the perturbation methods
can give correct solutions for this equation. It is because the methods itself do not provide a
criterion to check whether E is sufficiently small to give correct solutions. By comparing the
results computed by Davis [23] using the NI method, Burton [9] shows that the perturbation
methods will give incorrect results when E is as large as 1.0. When E tends to 1.0, relaxation
oscillations take place in which the corresponding motion is jerky, in other words, the motion
consists of slowly varying stretches followed by abrupt changes. The nonlinearity may be
regarded as very strong when the value of E is greater than 1.0 and in rare cases the equation
can be solved by the standard perturbation methods. As an illustration we choose E equal to
5.0.
As it is a symmetric oscillator, we consider a solution with m odd harmonic terms up to
cos(2m + 1)~ and sin(2m + l)~, as follows:
2m+l

U(T)= C (a, cosm+ b, sinn~). (33)

n= 1.3.

Referring to (27),

kAA=r+qAw+pAh, (34)

where k (28) is a 2m x 2m coefficient matrix which corresponds to the Fourier coefficient

vector

A= [a,,~3, . . . , Use+,, b,, b,, . . . , b2,,,+Jf.

Vectors r, q and p can be calculated corresponding to (29), (30) and (31), respectively. There
are totally 2m + 2 unknowns, including the frequency of the limit cycle w and the parameter A.
But we have only 2m equations. As far as only the limit cycle is concerned, A is taken to be
the control parameter (i.e. AA = 0) while Aw should be regarded as an unknown. Therefore,
one of the Fourier coefficient has to be fixed (e.g. a,. = 0 or bj = 0) to take the number of
equations equal to that of the unknowns. It will not cause any effects because this system is
autonomous but it only represents a shift of u on the time axis. In the following, we will set b,
equal to zero.
With the column corresponding to coefficient b, replaced by q in k and Ab, replaced by Am,
 S.L. Lau, S.W. Yuen. The Hopf bifurcation and limit cycle 1115

we have

k’ AA’ = r , (36)

where k’ and AA’ are the corresponding matrix k and vector AA after such a replacement,
respectively. Equation (36) can be solved by any equation solver and a better approximation
can be obtained by iterations. The iteration process is then carried out with updated k’ and r
each time until a given tolerance has been reached. This in fact is a Newton-Raphson process
of seeking solutions.
After obtaining solution for a particular A, it is then added by an increment AA (known as
the control increment). This is known as an augmentation process. The iteration process is
again repeated using the previous solution as an approximation until the solution for the
neighbouring A is sought. By this means, the neighbouring solution can also be sought. By
successive use of augmentation and iteration process, a solution diagram may easily be traced.
Alternatively, the frequency w, or one of the Fourier coefficients ai and bi, may also be used
as the control increment in the above augmentation process. The choice of the control
increment depends on the rate of the variation of the increments; usually the one with the
greatest increment (absolute values) will be chosen as the control increment. This is extremely
useful when a solution diagram is very sensitive to the variation of the parameter A or the like.
The number of the steps to trace the solution diagram will be reduced.
To be more efficient in tracing the solution diagram, an incremental arc-length extrapola-
tion method (see [17]) can be used to predict the next solution from some known states. With
such prediction, the number of iterations required for convergent solution will be greatly
reduced. Further, it is also useful in solving the convergence problem at peaks of the solution
diagram. A solution diagram for this parametric study can be done by computer with this
technique and the IHB method can be installed in the same program.
Referring to (33)-( 35)) m = 40 is assigned for the present example; the highest harmonic
terms are cos 807 and sin 807, respectively. The bifurcation diagram for (21) is shown in Fig. 1,
from which it can be clearly shown that the bifurcation occurs at A = 0. For A < 0, a trivia1
solution can be obtained and it represents a stable equilibrium state. For A > 0, there exist two
solutions for a given A: a trivia1 solution corresponding to unstable equilibrium state and a
non-trivia1 solution corresponding to a stable limit cycle oscillation.
The limit cycles for different values of parameter A are plotted in Fig. 2. This graph shows
that the size and shape of the limit cycles vary with the value of parameter A.
For comparison, the IHB method shows exactly the same results with that by the numerical
integration (using the Runge-Kutta method) for A = 1.0 (Fig. 2). In addition, the time history
for this case is also given in Fig. 3.

3.2. Coupled oscillators

Investigators are becoming more interested in studying the dynamics of coupled nonlinear
oscillators. Applications of coupled nonlinear oscillators to biological problems have been
discussed by Pavlidis [20]. The objective of this example is to emphasise the range of the small
parameter beyond which the standard perturbation method will surely give incorrect solutions.
The system of equations considered in this sub-section is expressed in the following as
1116 S. L. Lau, S. W. Yuen, The Hopf bifurcation and limit cycle

2.1 -l
2.0 -
1.9 -
1.8 -
1.7 -
1.6 -
1.5 -
1.4 -
1.3 -
1.2 -
1.1 -
1.0 -

0.9 -
0.8 -
0.7 -
0.6 -
0.5 -
0.4 -
0.3 -
0.2 -
0.1 -
0.0 I I I I I I I 1 I
0.0 0.2 0.4 0.6 0.8 1.0

The Parameter A

Fig. 1. Solution diagram of the Van der Pol oscillator (E = 5.0).

8.0 -
7.0 -
6.0 -
5.0 -
4.0 -
3.0 -
.3 2.0 -
x
.?Y 1.0 -
"
0 .
3 0.0 -.------.

-1.6-
aJ
E -2.0 -
-3.0 -
-4.0 -
~. numerical
-5.0 - integration
-6.0 -
- by IHB method
-7.0 -
-8.0 --
-4.0 0.0 2.0 4.0

The variable u

Fig. 2. Limit cycles of the Van der Pol oscillator (E = 5.0); + direct numerical integration, -1HB method with
40 harmonic terms.
 S.L. Lau, S.W. Yuen, The Hopf bifurcation and limit cycle 1117

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 16 18 2.0

The Dimensionless Time Variable

Fig. 3. Time history of the Van der Pol oscillator (E = 5.0 and h = 1.0).

o*ii, + U, - A(-2j.$oti, + .y~U:U,) = 0, w*ii, + 9u, - h(-2/_&*0ti* + y*u;> = 0,

(37)

where

pu, = -0.1 , ~1.~

= 0.5, ‘yl = 0.1 and y2 = 0.08.

This example has been studied by Xu and Lau [21] using an asymptotic method. The notations
of (37) have remained the same as in [21], except A which has been used instead of E. The
system of equations (37) is a particular case of (1) in which n = 2 and the functions f, and f,
are given by

f, = u1 - h(0.2 wzi, + 0.1 U$,) = 0 ) f* = 9U, + h(l.O Wti, - 0.08 U;) = 0.

(3% (39)

This system of equations is solved by the IHB method with harmonic terms cos T,
cos 37, . . . ) cos 97, sin 7, sin 37, . . . , sin 97. The solution diagram is shown in Fig. 4. In
the figure, limit cycles ‘1’ and ‘2’ are used to denote the limit cycles for U, and u2, respectively.
It can be observed from this figure that periodic solutions can only be found for A < 0.092
(approximately). For A in [O.O,0.0921, two limit cycle motions co-exist at a particular A. The
limit cycles for the case A = 0.08 are plotted in Figs. 5 and 6. The solution comes to its tip at
A = 0.092; and only one limit cycle will be found.
1118 S.L. Lau. S.W. Yuen, The Hopf bifurcation and limit cycle

1.000 -\

0.998 -

0.996 -

0.994 -
3
0.992-
5
.-
: 0.990-
0 0.988-
>
% 0.986 -

0.970 4 I I / / I / / / ,

0.00 0.02 0.04 0.06 0.08 0 10

The Parameter A

Fig. 4. Solution diagram of the coupled oscillators.

For A > 0.092, the result has been computed by the NI method; the motion for the case
A = 0.1 is computed and shown in Fig. 7. It seems to be periodic at the first sight, however, it
can be shown to be non-periodic, as in Fig. 8. The motion in fact does not repeat itself after

10

9-
8-

7-
G-
5-
N 4
.3
_ 3 1
.3- 2-
: l-
‘L= 0 - .. _
.-
-l-
;:
zi -2-
>
a> -3 -
= -4-
-5 -
-6 -
-7 -
-a -
-9 -
- 10 I I I I, I , I , I / I I , I , , , ,

- 10 -8 -6 -4 -2 0 2 4 6 8 10

The variables u ,u
t 2

Fig. 5. Limit cycles of the coupled oscillators (A = 0.08, o = 0.993).

 S.L. Lau, S.W. Yuen, The Hopf bifurcation and limit cycle 1119

10

9
a
7

1
z
.- 0 ----.
Y
;I -1 t --.----.-..------..-.-.---....-.
a, -2
>
-3
(II
.3- 2 I

-lo-/ , , , , I I , I I i I 1 I I I I I I I I
-10 ‘-8 -6 -4 -2 0 2 4 6 a 10

The variables u , u
1 2

Fig. 6. Limit cycles of the coupled oscillators (A = 0.08, o = 0.978).

-5
-6
-7

-10 -8 -6 -4 -2 0 2 4 6 a 1c

The variables u ,u
1 2
Fig. 7. Limit cycles of the coupled oscillators (A = 0.10) by the numerical integration.

each period and many different trajectories can be observed. Further, if this result is used as
initial values for the IHB calculation, no successful iteration can be established. It is therefore
concluded that no periodic solution will exist for A > 0.092.
1120 S. L. Lau, S.W. Yuen, The Hopf bifurcation and limit cycle

7.40
7.39 -
7.38 -
7.37 -
7.36
7.35 -
7.34 -
- 7.33 -_
I
7.32 -
>\
5 7.31 L
0
5 7.30 -
> 7.29 -
ar
," 7.28 -
7.27 -
7.26 -
7.25 -
7.24 -
7.23 -
7.22 -
7.21 -
7.20 1 I I # I , I I I
1.00 1.02 1.04 1.06 1.08 1.10

The variable u
1

Fig. 8. Detail of Fig. 7.

4. Concluding remarks

The present method is straight-forward, efficient and reliable for treating strongly nonlinear
systems. By introducing a parameter A in the IHB formulation, it can be used to obtain
solution diagram, including phenomena of bifurcations and critical points for nonlinear
dynamic systems.
The advantage of perturbation methods is able to give analytical solutions. However,
generally speaking, these methods do not provide a definite range of parameter (E in the first
example or h in the second one), in which they give reliable results.
The application of the IHB method to parametric studies can further be improved by
knowing that the determinant of the coefficient matrix, k (23, will become singular at
bifurcation points. By using this property, Leung and Fung [22] developed a method of
parametric unfolding for constructing the bifurcation diagram.

References

fl] E. Hopf, Abzweigung einer periodischen Losung von einer stationaren Losung eines ~ifferentiaisystems, Ber.
Math-Phys., Sachs. Akad. Wiss. Leipsig 94 (1942) 3-22; English translation in [2].
[2] J.E. Marsden and M. McCracken, The Hopf bifurcation and its applications, Applied Mathematical Sciences
(Springer, New York, 1976) III + 408~~.
[3] B.D. Hassard, N.D. Kazarinoff and Y.-H. Wan, Theory and applications of Hopf bifurcation, London
Mathematical Society Lecture Notes Series 41 (1981).
[4] J. Awrejcewicz and J. Mrozowski, Bifurcations and chaos of a particular Van der Pol-Duffmg oscillator, J.
Sound and Vibration 132 (1989) 89-100.
 S.L. Lau, S.W. Yuen, The Hopf bifurcation and limit cycle 1121

[5] M. Lieberinan and A. Lichtenberg, Regular and Stochastic Motion (Springer, Berlin, 1983).
[6] J. Guckenheimer and P. Holmes, Dynamical systems and bifurcations of vector fields, in, Nonlinear
Oscillations (Springer, Berlin, 1983).
[7] J. Xu, R.S. Guttalu and C.S. Hsu, Domains of attraction for multiple limit cycles of coupled Van der Po1
equations by simple cell mapping, Internat. J. Non-Linear Mech. 20 (1985) 507-517.
[8] A.H. Neyfeh, Perturbation Methods (Wiley, New York, 1973).
[9] T.D. Burton, Non-linear oscillator limit cycle analysis using a time transformation approach, Internat. J.
Non-Linear Mech. 17 (1982) 7-19.
[lo] S.L. Lau and Y.K. Cheung, Amplitude incremental variational principle for nonlinear vibration of elastic
system, ASME J. Appl. Mech. 48 (1981) 959-964.
[ll] S.L. Lau, Y.K. Cheung and S.Y. Wu, A variable parameter incrementation method for dynamic instability of
linear and nonlinear system, ASME J. Appl. Mech. 49 (1982) 849-853.
[12] C. Pierre and E.H. Dowell, A study of dynamic instability of plates by an extended incremental harmonic
balance method, ASME J. Appl. Mech. 52 (1985) 693-697.
[13] C. Pierre, A.A. Ferri and E.H. Dowell, Multi-harmonic analysis of dry friction damped systems using an
incremental harmonic balance method, ASME J. Appl. Mech. 52 (1985) 658-694.
[14] S.L. Lau, Y.K. Cheung and S.Y. Wu, Incremental harmonic balance method with multiple time scales for
aperiodic vibration of nonlinear systems, ASME J. Appl. Mech. 50 (1983) 871-876.
[15] S.L. Lau and W.-S. Zhang, Nonlinear vibrations of piecewise-linear systems by incremental harmonic balance
method, ASME J. Appl. Mech. (accepted).
[16] P. Friedmann, C.E. Hammond and T.H. Woo, Efficient numerical treatment of periodic systems with
application to stability problems, Internat. J. Numer. Methods Engrg. 11 (1977) 1117-1136.
[17] S.L. Lau, Y.K. Cheung and S.Y. Wu, Nonlinear vibration of thin elastic plates, Part II: Internal resonance,
ASME J. Appl. Mech. 49 (1982) 849-853.
[18] N. Minorsky, Nonlinear Oscillations (Van Nostrand, New York, 1967).
[19] A.H. Nayfeh and D.T. Mook, Nonlinear oscillations (Wiley/Interscience, New York, 1979).
[20] T. Pavlidis, Biological Oscillators: Their mathematical analysis (Academic Press, New York, 1973).
1211 Z. Xu and S.L. Lau, Asymptotic method for nonlinear autonomous system with two degrees of freedom,
Proc. Asia Vibration Conference ‘89 (November, 1989) 69-74.
[22] A.Y.T. Leung and T.C. Fung, Construction of chaotic regions, J. Sound and Vibration 131 (1989) 445-455.
[23] H.T. Davis, Introduction to nonlinear differential and integration equations (Dover, New York, 1962).