Exact and approximate expressions for the period of anharmonic oscillators

Paolo Amore∗
Facultad de Ciencias, Universidad de Colima,
Bernal Díaz del Castillo 340, Colima, Colima,
Mexico.

Francisco M. Fernández†
INIFTA (Conicet, UNLP), Blvd. 113 y 64 S/N,
Sucursal 4, Casilla de Correo 16,
arXiv:math-ph/0409034v1 15 Sep 2004

1900 La Plata, Argentina

In this paper we present a straightforward systematic method for the exact and approximate
calculation of integrals that appear in formulas for the period of anharmonic oscillators and other
problems of interest in classical mechanics.

PACS numbers: 45.10.Db, 04.25.-g

I. INTRODUCTION

The discussion of periodic motion in one dimension is important in most introductory courses on classical mechanics.
Several problems can be solved exactly, but in most cases one has to resort to approximate solutions. Simple but
sufficiently accurate approximate solutions for such problems are very important in understanding many features
of classical mechanics. In addition to it, in some cases one is simply satisfied with accurate numerical results, and
expressions suitable for computation are most welcome.
The purpose of this paper is the discussion of the exact and approximate calculation of the period of a particle that
moves in one dimension under the effect of an anharmonic potential.

II. PERIODIC MOTION IN ONE DIMENSION

Consider a particle of mass m moving in one dimension under a potential–energy function V (x). Without loss of
generality we assume that V (x) has a minimum at x = 0; more precisely, we assume that V (0) = 0, V ′ (0) = 0, and
V ′′ (0) > 0, where the prime indicates differentiation with respect to x. Following the standard notation in classical
mechanics, we use a dot to indicate differentiation with respect to time, for example: v = ẋ.
¿From the equation of motion

mẍ = −V ′ (x) (1)

we easily obtain an integral of the motion

mẋ2
E= + V (x) (2)
2
which is the total energy. The motion of the particle is restricted to the interval x− < x < x+ , where the turning
points x± satisfy V (x± ) = E; that is to say, ẋ = 0 at those points.
 2

It is well–known that the period of the motion is given by

√ Z x+
dx
I
T = dt = 2m p (3)
x− E − V (x)

from which we obtain the frequency Ω = 2π/T .

We can simplify the equations of motion by the introduction of a dimensionless time τ = ω0 t, where ω0 is an
arbitrary frequency. If we define

E V (x)
E= , U (x) = (4)
mω02 mω02

then we obtain the equations of motion of a particle of unit mass; for example:

ẋ2
E= + U (x) (5)
2

and
√ Z x+
2 dx
T = p . (6)
ω0 x− E − U (x)

It is worth noticing that equation (5) is not dimensionless because E and U (x) have units of length squared. In
order to get a truly dimensionless equation we should define a dimensionless coordinate q = x/L, where L has units
of length. Thus E/ mω02 L2 and V (Lq)/ mω02 L2 are the dimensionless counterparts of the total and potential

energies, respectively. In this paper, however, we have opted for equations that are similar to those often found in
current literature.
For example, if

v2 x2 v4 x4
V (x) = + (7)
2 4
p
we may choose ω0 = v2 /m so that

x2 λx4
U (x) = + , (8)
2 4

where λ = v4 /v2 .

III. THE MAIN INTEGRAL

It follows from the discussion above that the period is proportional to an integral of the form
Z x+
dx
I= p , (9)
x− Q(x)

were Q(x) exhibits simple zeros at x− and x+ and is positive definite for all x− < x < x+ . That is to say, we can
write

Q(x) = (x+ − x)(x − x− )R(x) (10)

where R(x) > 0 for all x− ≤ x ≤ x+ .

 3

The reason for rewriting our problem in this somewhat abstract way is that the integral (9) applies to problems
other than the period of a motion in one dimension. We will mention some examples later on.
In order to develop suitable exact and approximate expressions for the integral (9) we define the reference function
ω2
Q0 (x) = (x+ − x) (x − x− ) (11)
2
that satisfies the appropriate boundary conditions at the turning points. It is clear that Q0 (x) is the function that
would appear in the treatment of a harmonic oscillator. Then we rewrite (9) as
Z x+
dx
I= p p , (12)
x− Q0 (x) 1 + ∆(x)
where
Q(x) − Q0 (x) 2R(x) − ω 2
∆(x) ≡ = . (13)
Q0 (x) ω2
The change of variables
x+ + x− x+ − x−
x= + cos θ (14)
2 2
makes the integral (12) much simpler:
√ Z π
2 dθ
I= √ . (15)
ω 0 1+∆
This equation leads to an exact expression for the period, which in most cases one has to calculate numerically. In
order to derive simple analytical formulas we expand
 
∞
1 X −1/2
√ =   ∆j (16)
1+∆ j=0 j
 
a
where   = a!/[b!(a − b)!] is a combinatorial number. Notice that this series converges for all x such that |∆| < 1.
b
We thus obtain a series for the integral (15):
√
 
∞
2  −1/2  π j
X Z
I= Ij , Ij = ∆ dθ. (17)
j=0
ω j 0

In this way we can derive approximate expressions for the integral (15) by means of the partial sums:
N
X
I (N ) = Ij . (18)
j=0

IV. THE DUFFING OSCILLATOR

The potential–energy function (8) gives rise to the Duffing oscillator. Since it is parity invariant (U (−x) = U (x))
then x+ = −x− = A is the amplitude of the oscillations. According to the general discussion of the preceding section,
it follows from
 

1 λ 2
2 22


Q(x) = E − U (x) = A − x + A +x (19)
2 4
 4

that

1 λ 2
A + x2


R(x) = + (20)
2 4

and
λA2
1 + λA2 − ω 2 − 2 sin2 θ
∆= (21)
ω2

where we have substituted x = A cos θ. We conclude that the period depends on the dimensionless parameter ρ = λA2
that is the ratio of v4 A4 and v2 A2 both having units of energy.
√
If we choose ω = 1 + ρ then we obtain an already known suitable compact expression for the integral1
√ Z π
2 dθ ρ
I= √ p , ξ= . (22)
1+ρ 0 2
1 − ξ sin θ 2ρ +2

This equation yields the series

2
√

∞
2π X −1/2
I=√   ξj (23)
1 + ρ j=0 j

that converges for all |ξ| < 1; that is to say, for all ρ > −2/3 or ρ < −2.
√
When λ < 0 the potential exhibits two barriers of height 1/(−4λ) at x = ±1/ −λ and therefore the amplitude of the
√
periodic motion cannot be greater than AL = 1/ −λ. In other words, there is periodic motion if ρ > ρL = λA2L = −1.
The series (23) does not converge for −1 < ρ < −2/3 and the analytical expressions that we may derive from it will
not be valid for all the values of the energy that give rise to periodic motion. Can we improve this approach?. The
answer is ”yes ” as we will see below.
Let RM and Rm be the maximum and minimum values of R(x) in the interval [x− , x+ ] and ∆M and ∆m the
corresponding values of ∆(x). Since R(x) is positive definite we know that RM ≥ R(x) ≥ Rm > 0. If we choose the
value of the adjustable parameter ω so that ∆M = −∆m we obtain

ωb2 = RM + Rm > 0 (24)

and

2R(x) − RM − Rm
∆b (x) = . (25)
RM + Rm

The subscript b indicates that this particular value of ω ”balances” the maximum and minimum values of ∆(x).
Notice that |∆(x)| < 1 for all x− ≤ x ≤ x+ because ∆M = (RM − Rm ) / (RM + Rm ) < 1.
For the particular case of the Duffing oscillator we have Rm = R(0) = 1/2 + ρ/4 and RM = R(±A) = 1/2 + ρ/2 so
that

4 + 3ρ
ωb2 = (26)
4

and

ρ
∆b (θ) = cos(2θ). (27)
4 + 3ρ
 5

Thus the integral becomes

√ Z π
2 2 dθ ρ
I=√ p , ξ= (28)
4 + 3ρ 0 1 + ξ cos(2θ) 4 + 3ρ

that gives rise to the series

√
  
∞
2 2π X −1/2 −1/2
I=√ (−1)j    ξ 2j . (29)
4 + 3ρ j=0 j 2j

which converges for all |ξ| < 1; that is to say, for all ρ > −1 or ρ < −2. In this way we may obtain simple analytical
expressions for the period valid for all values of the energy consistent with periodic motion.
According to equation (6) the period is given by
√
2
T = I (30)
ω0
and equation (29) enables us to derive simple analytical approximate expressions for it. For concreteness and simplicity
we choose ω0 = 1 in what follows. For example, the first two approximations are
4π
T (0) = √ (31)
4 + 3ρ
and
π 147ρ2 + 384ρ + 256


(1)
T = 5/2
. (32)
4 (4 + 3ρ)
These expressions are expected to be accurate for small values of ρ, and in fact they give the exact result for ρ = 0.
However, they are also accurate for extremely great values of ρ. Notice that
Z π
√ dθ
lim ρT = 4 p ≈ 7.4162987 (33)
ρ→∞ 0 3 + cos(2θ)
A straightforward calculation shows that
√ (0) 4π
lim ρT = √ ≈ 7.26 (34a)
ρ→∞ 3
√
√ (1) 49 3π
lim ρT = ≈ 7.406. (34b)
ρ→∞ 36
We conclude that such simple analytical expressions are sufficiently accurate for most purposes and that one easily
improves them by straightforward addition of more terms of the series (29).

V. QUADRATIC–CUBIC OSCILLATOR

Parity–invariant oscillators exhibit symmetric turning points; if the potential is nonsymmetric so are the turning
points. The simplest example is
v2 2 v3 3
V (x) = x + x . (35)
2 3
p
If we again choose ω0 = v2 /m then we obtain

x2 λ v3
U (x) = + x3 , λ = . (36)
2 3 v2
 6

The potential–energy function U (x) shows a barrier of height U (xM ) = 1/ 6λ2 at xM = −1/λ, and the turning

points satisfy x+ > 0 > x− .

If we write

Q(x) = E − U (x) = (x − x− )(x+ − x)(b0 + b1 x) (37)

then we obtain
x+ x− λ x+ + x−
b0 = − , b1 = =− (38)
2(x2+ + x+ x− + x2− ) 3 2(x2+ + x+ x− + x2− )
and
3 x− + x+
λ=− . (39)
2 x2+ + x+ x− + x2+

Since U (−x, −λ) = U (x, λ) we consider only the case λ > 0 without loss of generality; therefore x− + x+ < 0 because
x2+ + x+ x− + x2− > 0. Taking into account that b0 > 0 and b1 > 0 we conclude that Rm and RM take place at the
turning points; therefore,

x2+ + 4x+ x− + x2−

ωb2 = R(x+ ) + R(x− ) = −
. (40)
2 x2+ + x+ x− + x2−

Is ωb real for all values of E below the barrier?. In order to answer this question notice that the third root x3 of Q(x)
is smaller than x− and is given by
b0 x+ x−
x3 = − =− < 0. (41)
b1 x+ + x−
Therefore

x2+ + 4x+ x− + x2−

2
= (x+ + x− ) + 2x+ x−

= (x+ + x− ) (x+ + x− − 2x3 ) < 0 (42)

because x− − x3 > 0 and x+ − x3 > 0.

Finally, after the change of variables (14) the function ∆(θ) takes a particularly simple form:

(x2+ − x2− )
∆b (θ) = ξ cos θ, ξ = (43)
x2+ + 4x+ x− + x2−

where x2+ − x2− < 0 because 0 < x+ < −x− .

The resulting integral
√ Z π
2 dθ
I= √ (44)
ωb 0 1 + ξ cos θ
gives rise to the series
√
  
∞
2π X −1/2 −1/2
I= (−1)j    ξ 2j . (45)
ωb j=0 j 2j

which is similar to the one derived above for the Duffing oscillator and converges for all |ξ| < 1.
 7

When E = U (xM ) then x3 = x− (remember that λ > 0) and ξ = 1. We appreciate that the series (45) converges
for all values of the energy for which there is periodic motion.
Equation (44) gives us a simple and exact expression for the period of the anharmonic oscillator (36) that requires
numerical integration to obtain results for a given set of potential parameters. On the other hand, equation (45)
provides approximate analytical expressions that one makes as accurate as desired by simply adding a sufficiently
large number of terms. The choice of one or another depends on the particular application.
Following a different procedure Apostol2 derived the exact expression for the period of the quadratic–cubic oscillator

π/2
r
3 4 dα x+ − x−
Z
T = √ p , k2 = . (46)
2λ ω0 x+ − x3 0
2
1 − k 2 sin α x+ − x3

The expansion of this equation in powers of k 2 is also convergent for all values of the energy consistent with periodic
motion because k 2 < 1.

VI. CONCLUSIONS

In this paper we present a straightforward systematic procedure for constructing exact and approximate expressions
for the period of anharmonic oscillators. The recipe is simple: first, we factor the function Q(x) and obtain the turning
points and the function R(x) as in equation (10). Second, we obtain the maximum and minimum values of R(x) in
the interval between the turning points which determine the optimum value of ω. Thus we are left with an exact
expression for the period that we may use in numerical applications. In addition to it, we may expand this exact
expression in a Taylor series in order to obtain partial sums that become analytical expressions for the period of
increasing accuracy. These partial sums converge to the exact result for all values of the energy that give rise to
periodic motion.
The method proposed in this paper is not restricted to the period of anharmonic oscillators with polynomial
potentials. We may, for example, expand a given arbitrary potential U (x) about its minimum to any desired degree
and then apply the approach developed above. Moreover, some other problems have been expressed in terms of
integrals of the form (9), such as, for example, the deflection of light by a massive body or the precession of a planet
orbiting around a star3 . Recently, we have already applied a variant of present approach to such problems4,5 .
There is a wide range of interesting applications for present method and for that reason we believe that it is suitable
for teaching in advanced undergraduate courses on classical mechanics.

∗
Electronic address: [email protected]
†
Electronic address: [email protected]
1
A. H. Nayfeh, Introduction to Perturbation Techniques (John Wiley & Sons, New York, 1981).
2
B. F. Apostol, J. Theor. Phys. 86, 1 (2003).
3
S. Weinberg, Gravitation and Cosmology (John Wiley & Sons, New York, 1972).
4
P. Amore and R. A. Sáenz, The Period of a Classical Oscillator, math-ph/0405030.
 8

5
P. Amore, A. Aranda, F. M. Fernández, and R. Sáenz, Systematic Perturbation of Integrals with Applications to Physics,
math-ph/0407014.