Approximating Small and Large Amplitude

Periodic Orbits of the Oscillator

x• + 1 + x_ 2 x = 0

Tamás Kalmár-Nagy
Department of Aerospace Engineering,
Texas A&M University,
College Station, TX 77845, USA
Thomas Erneux
Université Libre de Bruxelles
Optique Nonlinéaire Théorique
Campus Plaine, C.P. 231
1050 Bruxelles, Belgium
May 27, 2006

Abstract
We show that this oscillator is conservative and derive scaling laws
for the period for small and large amplitude vibrations. We also con-
struct analytical approximations of the periodic orbits and observe
excellent agreement with numerical solutions.

1 Introduction
Beatty and Mickens [1] and later Mickens [5] investigated the nonlinear os-
cillator
x• + 1 + x_ 2 x = 0 (1)

1
Figure 1: The solution of (1) and (2) is represented in terms of z = x=A as a
function of t. The oscillations look nearly harmonic for A = 1 (broken line)
and square-wave-like for A = 5 (full line).

2
subject to the initial conditions
x (0) = A > 0 and x_ (0) = 0: (2)
This oscillator is characterized by a velocity-dependent sti¤ness coe¢ cient
and depends on only one parameter. Mickens pointed out that while a peri-
odic solution exists for all positive A and that the period T (A) approaches
zero for large A [5], simple approximations for the period are only de…ned
for a …nite range of initial amplitudes. For example, the …rst order harmonic
balance approximation results
p
T (A) = 4 A2 (3)
and even with more advanced techniques, such as the harmonic balance based
averaging of Chatterjee (see Eq. (28) of [2]), the problem persists. The
properties of the oscillator (1) contrasts with those of
x• + (1 + x)
_ x=0 (4)
which has been studied in the context of the Hopf bifurcation in relaxation
oscillations [7] and for Class B laser oscillations [4]. For (4) subject to (2),
the period of the oscillations increases monotonically with A and the output
consists of short and intense pulses separated by intervals where x is almost
zero. This is not the case of (1). As we shall demonstrate in this note, the
period of the oscillations decreases with A and the oscillations are reminiscent
of square waves for large A (see Figure 1).
Our objective is to determine approximations for the small and large A
limit of the solution of Eq. (1) and (2). To this end, we …rst show that the
oscillator is conservative and depends on a single well potential. We then
de…ne the period and determine its leading expression for small and large A.
Our results motivate analytical constructions for small and large amplitude
periodic solutions directly from Eq. (1).

2 Calculating the period

It will be mathematically convenient to reformulate Eq. (1) with the scaled
variable z = x=A so that the parameter A appears in the di¤erential equation
rather than the initial conditions. Speci…cally, we …nd
z• + 1 + A2 z_ 2 z = 0; with z (0) = 1; z_ (0) = 0; (5)

3
or equivalently, the following system of …rst-order di¤erential equations
dz dy
= y; = 1 + A2 y 2 z; (6)
dt dt
z (0) = 1 and y (0) = 0: (7)
By dividing the two equations in (6), we obtain an equation for the trajectory
y = y (z) given by
dy (1 + A2 y 2 ) z
= ; y(1) = 0: (8)
dz y
This equation is separable and admits the …rst integral

ln 1 + A2 y 2 + A2 z 2 = A2 : (9)

From (9) we can express 1 + A2 y 2 as

1 + A2 y 2 = exp(A2 (1 z 2 )) (10)

Di¤erentiating both sides and simplifying (y = z)

_ leads to the following
second-order di¤erential equation for z

z• + eA (1 z2 )
2
z=0 (11)

It is easy to show that

dV
z• + =0 (12)
dz
where V (z) is a single well potential de…ned by (V (1) = 0)
1 exp(A2 (1 z 2 ))
V (z) : (13)
2A2
The existence of the potential well provides a simple proof of the conserva-
tiveness [9] of the equivalent oscillators (11) and (1). Using (10), we express
p
y = z_ = A 1 exp(A2 (1 z 2 )) 1 (14)

which allows to de…ne the period as [5]

Z1
dz
T (A) = 4A p : (15)
exp(A2 (1 z 2 )) 1
0

4
The small A limit is obtained by expanding the exponential in (15) for small
A2 . Integrating the …rst two terms gives
A2
T (A) ' 2 : (16)
4
The large A limit is more delicate to obtain but the main contribution of the
integral comes for z close to 1. Introducing the new variable u de…ned by

z = 1 + A 2 u; (17)

equation (15) becomes

Z0
4 du
T (A) = p (18)
A e 2u u2 =A2 1
A2

In the A ! 1 limit this integral can be solved since [3]

Z0
du
lim p = : (19)
A!1 e 2u u2 =A2 1 2
A2

Therefore the leading asymptotic behavior of the period for large A is

2
T (A) ' . (20)
A
Figure 2 compares the exact numerical period with the two approximations
(16) and (20).

3 Approximations of Periodic Orbits

We now have enough information for proposing analytical constructions of

the periodic solutions for small and large A. For A 1

eA (1 z2 )
2
1 + A2 A2 z 2 (21)

and thus (11) can be written as

z• + 1 + A2 z A2 z 3 = 0 (22)

5
Figure 2: The numerically computed period T (A) (full line) is compared to
its small A and large A asymptotic approximations (broken lines) given by
(16) and (20), respectively.

This is the well-known Du¢ ng equation and its solution can be found in
closed form in terms of Jacobi elliptic functions [8]. Substituting a solution
of the form (a1 , a2 nonzero)

z (t) = a1 sn (u; m) ; u = a2 t + b (23)

into (22) yields

1 A2 + a22 mcn2 (u; m) + dn2 (u; m) + A2 a21 sn2 (u; m) = 0: (24)

Using the identities

sn2 (u; m) + cn2 (u; m) = 1 (25)

msn2 (u; m) + dn2 (u; m) = 1 (26)

results
a22 (1 + m) 1 A2 + A2 a21 2a22 m sn2 (u; m) = 0: (27)
Equating the coe¢ cients of this equation with zero results in two algebraic
equations. Two additional equations are provided by the initial conditions

6
(5) and therefore the four unknowns a1 , a2 , m, b can be solved for as

a1 = 1
p
a2 = 1 + A2 =2
m = A 2 = 2 + A2
b = sn 1 (1; m) : (28)

Figure 3 shows the periodic solution of (11) for A = 0:5 and the corresponding
approximation (23, 28).

Figure 3: The periodic solution of (11) for A = 0:5 (full line) is compared
with its asymptotic approximation (dots) given by (23, 28).

The analysis of the period for large A indicates that the solution spends
most of time near the slow manifold z = 1 and with the time scale t =
O(A 1 ): This suggests to introduce the new time s = At and z = 1 + A 2 u
into Eqs. (6). The A2 y 2 term in the equation for y also motivates the change
of variable y = A 1 v: For large A, the equation for y then reduces to
dv
= (1 + v 2 ); v(0) = 0 (29)
ds
which admits the solution
v= tan(s): (30)

7
Figure 4: The periodic solution of (11) for A = 5 (full line) is compared with
its asymptotic approximation (dots) given by (32) and (33).

This expression becomes unbounded as s ! =2 where the fast jump tran-

sitions between z = 1 occur: As a consequence the half-period in s is the
time interval between two successive jumps and equals : Equivalently, the
half-period in t equals =A; as we have found from the integral de…nition of
the period. The expression for u is then obtained by integrating the …rst
equation in (6). It transforms as du=ds = v, u(0) = 0 and leads to the
expression
u = ln j sec(s)j (31)
For A = 5; the approximation of the slow manifolds near z = 1 are given
by

z = 1 + A 2 ln j sec(2 t=T0 )j T0 =4 < t < 3T0 =4 (32)

z = 1 A 2 ln j sec(2 t=T0 )j 3T0 =4 < t < 5T0 =4 (33)

where T0 = 2 =A ' 1:3. Figure 4 shows excellent agreement between the

numerical solution and the approximation.

8
4 Discussion
In summary, our analysis of Equations (1) and (2) demonstrated that for
small amplitude oscillations a good analytical approximation can be given
in terms of the Jacobi elliptic function. Further, the period quadratically
depends on the vibration amplitude for small A. For large A we showed that
the oscillations are square-wave-like, switching periodically between the slow
manifolds x = A: Moreover the period scales like A 1 . Numerical solutions
for A = 0:5 and A = 5 compare quantitatively well with the analytical
approximations.

5 Acknowledgments
TK-N would like to thank Prof. Adonios Karpetis for the insightful discus-
sions. TE is supported by the Fonds National de la Recherche Scienti…que
(Belgium).

References
[1] J. Beatty, R. E. Mickens, A qualitative study of the solutions to the
di¤erential equation x• + (1 + x_ 2 ) x = 0, Journal of Sound and Vibration
283 (2005) 475–477
[2] A. Chatterjee, Harmonic balance based averaging: approximate realiza-
tions of an asymptotic technique, Nonlinear Dynamics 32 (2003) 323–343.
[3] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Prod-
ucts, Academic Press, Inc (London), 4th Edition 1980
[4] T. Erneux, S. M. Baer and P. Mandel, Subharmonic bifurcation and
bistability of periodic solutions in a periodically modulated laser, Phys.
Rev. A 35, 1165-1171 (1987)
[5] R. E. Mickens, Investigation of the properties of the period for the non-
linear oscillator x• + (1 + x_ 2 ) x = 0, Journal of Sound and Vibration 292
(2006) 1031–1035
[6] R. E. Mickens, Oscillations in Planar Dynamic Systems, World Scienti…c,
Singapore, 1996

9
[7] S. M. Baer and T. Erneux, Singular Hopf bifurcation to relaxation oscil-
lations, SIAM J. Appl. Math. 46, 721-739 (1986)

[8] R. H. Rand, Topics in nonlinear dynamics with computer algebra. Com-

putation in Education, vol. 1. Langhorne, PA: Gordon and Breach, Sci-
ence Publishers, 1994

[9] S. H. Strogatz, Nonlinear Dynamics and Chaos, Addison-Wesley Pub.

Comp., Reading, 1994

10