Home Search Collections Journals About Contact us My IOPscience

A primer on elliptic functions with applications in classical mechanics

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

2009 Eur. J. Phys. 30 729

(http://iopscience.iop.org/0143-0807/30/4/007)

View the table of contents for this issue, or go to the journal homepage for more

Download details:
IP Address: 129.8.242.67
The article was downloaded on 26/03/2013 at 02:27

Please note that terms and conditions apply.

IOP PUBLISHING EUROPEAN JOURNAL OF PHYSICS
Eur. J. Phys. 30 (2009) 729–750 doi:10.1088/0143-0807/30/4/007

A primer on elliptic functions with

applications in classical mechanics
Alain J Brizard
Department of Chemistry and Physics, Saint Michael’s College, Colchester, VT 05439, USA

Received 2 December 2008, in final form 2 February 2009

Published 8 May 2009
Online at stacks.iop.org/EJP/30/729

Abstract
The Jacobi and Weierstrass elliptic functions used to be part of the standard
mathematical arsenal of physics students. They appear as solutions of many
important problems in classical mechanics: the motion of a planar pendulum
(Jacobi), the motion of a force-free asymmetric top (Jacobi), the motion of a
spherical pendulum (Weierstrass) and the motion of a heavy symmetric top
with one fixed point (Weierstrass). The planar pendulum can, in fact, be
used to highlight an important connection between the Jacobi and Weierstrass
elliptic functions. The easy access to mathematical software by physics
students suggests that they might reappear as useful mathematical tools in
the undergraduate curriculum.

1. Introduction

A long time ago, physics students were well trained in applications of elliptic functions in
solving a great variety of problems in classical mechanics. For example, Whittaker’s Treatise
on the Analytical Dynamics of Particles and Rigid Bodies [1] appears to implicitly assume
that the reader is fully conversant with the theory of elliptic functions.
Elliptic functions rapidly fell out the standard physics curriculum over the past fifty
years, however, and they are now only mentioned in passing in most standard (modern)
textbooks on classical mechanics [2–4]. The information on these mythical functions is now
relegated to mathematics textbooks [5, 6] and mathematical handbooks [7, 8] with notations
and conventions that are often contradictory or difficult to understand by physicists. The
purpose of this paper is thus to (re)introduce the Jacobi and Weierstrass elliptic functions to
a new generation of physics students through a series of standard problems found in classical
mechanics [9].
Many problems in classical mechanics involve calculating of one of the following integrals
[3]. First, the time integral
 x
dy
t (x) = ± √ (1)
x0 (2/m)[E − U (y)]
0143-0807/09/040729+22$30.00 
c 2009 IOP Publishing Ltd Printed in the UK 729
730 A J Brizard

arises in solutions x(t) of one-dimensional problems associated with motion of a particle of

mass m and constant total energy E in a time-independent potential U (x), where the initial
condition x0 may be chosen to correspond to a root of the turning-point equation E = U (x0 ).
Next, the orbit integral
 s
dσ
θ (s) = ±  (2)
s0 (2μ/ )[E − U (σ −1 ) − 2 σ 2 /2μ]
2

arises in solutions r(θ ) ≡ 1/s(θ ) of central-force problems involving the motion of a (fictitious)
particle of (reduced) mass μ, with constant total energy E and angular momentum  = μr 2 θ̇ ,
 potential U (r), where the initial condition s(0) = s0 is a turning point where
in a central
E = U s0−1 − 2 s02 2μ. Lastly, the time integral
 cos θ
du
t (θ ) = ±  (3)
cos θ0 (2/I1 )(1 − u2 )[E − V (u)]
arises in solutions θ (t) of rigid-body dynamics in the Lagrangian representation, where I1
denotes one principal component of the inertia tensor for a symmetric top (I1 = I2 = I3 ),
the effective potential V (cos θ ) contains terms associated with conserved angular momenta
associated with the ignorable Eulerian angles ψ and ϕ, and the initial condition θ (0) = θ0 is
a turning point where E = V (cos θ0 ).
Exact analytical solutions for these integrals exist only for certain potentials, in which
cases the inversions t (x) → x(t), θ (s) → s(θ ) and t (θ ) → θ (t) can be expressed in terms
of known functions. For example, exact solutions of the time integral (1) exist in terms of
trigonometric (or singly periodic) functions when the potential U (x) is a quadratic polynomial
in x. Trigonometric solutions of the orbit integral (2), on the other hand, exist for the Kepler
problem U (r) = −k/r and the isotropic harmonic oscillator U (r) = kr 2 /2. The purpose of
this paper is to explore exact analytic solutions of the integrals (1)–(3) expressed in terms of
doubly periodic functions called elliptic functions [6]. For example, exact solutions of the
orbit integral (2) for the central potential U (r) = kr n exist in terms of elliptic functions [1]
for n = ±6, ±4, 1 and −3. The time integral (3), on the other hand, has a solution in terms
of elliptic functions for the problem of the heavy symmetric top (of mass M) with one fixed
point (located at a distance h from the center of mass), where the gravitational potential is
V (cos θ ) = Mgh cos θ .

1.1. Doubly periodic elliptic functions

We now present a brief definition of elliptic functions in terms of their analytic (complex)
properties. A function F (z) is said to be doubly periodic, with periods η and η (where the
complex-valued ratio η /η has a positive-definite imaginary part), if F (z + mη + nη ) = F (z),
for m, n = 0, ±1, ±2, . . . (but not m = 0 = n). We note that, in the limit |η | → ∞ (assuming
that η is real), the function F (z) becomes singly periodic with period η. The fundamental
period–parallelogram of a doubly periodic function is defined by the quadrangle formed by
the four points (0, η, η , η + η ). Within this fundamental parallelogram, elliptic functions are
doubly periodic functions with two simple zeros and either a second-order pole (Weierstrass
elliptic function) or two first-order poles (Jacobi elliptic function). Note that there are no
elliptic functions of first order and that there are no multiply periodic functions with more than
two periods [6].
Elliptic functions y(x; a) are defined as solutions of the nonlinear ordinary differential
equation
 2
dy
= a4 y 4 + a3 y 3 + a2 y 2 + a1 y + a0 ,
dx
A primer on elliptic functions with applications in classical mechanics 731

where a ≡ (a0 , a1 , . . . , a4 ) are constant coefficients. This equation can be formally solved by
finding the inverse function
 y
ds
x(y; a) = x0 ±  ,
y0 (a) a4 s + a3 s + a2 s 2 + a1 s + a0
4 3

where y0 (a) is a root of the quartic polynomial a4 y 4 + a3 y 3 + a2 y 2 + a1 y + a0 and x(y0 ; a) = x0 .

Jacobi elliptic functions are defined in terms of the quartic polynomial (1−y 2 )(a +by 2 ), where
a and b are constants, while Weierstrass elliptic functions are defined in terms of the cubic
polynomial 4y 3 − g2 y − g3 , where g2 and g3 are constants. There is a connection between
the Jacobi and Weierstrass elliptic functions [6] that will be exploited later in section 3.1 (see
also appendices A and B).

1.2. Organization

The remainder of the paper is organized as follows. In section 2, we present the Jacobi
elliptic functions and discuss, first, the Seiffert spherical spiral [6, 10] as a mathematical
introduction of their doubly periodic nature. Next, we discuss exact solutions to the
physical problems of (i) the planar pendulum and (ii) Euler’s equations for a force-free
asymmetric top. In section 3, we present the Weierstrass elliptic functions and discuss exact
solutions to the physical problems of (i) the planar pendulum, (ii) the spherical pendulum and
(iii) the motion of a heavy symmetric top with one fixed point. The fact that the problem of the
planar pendulum is solved in terms of the Jacobi and Weierstrass elliptic functions highlights
an important connection between these functions. In section 4, we discuss one interesting
application of elliptic functions in terms of the travelling-wave solutions of nonlinear partial
differential equations. We summarize our work in section 5 and present mathematical details
in appendices A and B. Lastly, we note that the notation used for the Jacobi and Weierstrass
elliptic functions in our paper is partly based on the material presented elsewhere [7–9].

2. Jacobi elliptic functions

We begin our introduction of elliptic functions with the more familiar Jacobi elliptic functions
(sn z, cn z, dn z). First, the Jacobi elliptic function sn z ≡ sn(z|m), with modulus m < 1, is a
solution to the differential equation
 2
dy
= (1 − y 2 )(1 − my 2 ), (4)
dz
subject to the initial condition y(0) = 0. Next, the Jacobi elliptic function cn z ≡ cn(z|m),
with modulus m < 1 and complementary modulus m = 1 − m > 0, is a solution to the
differential equation
 2
dy
= (1 − y 2 )(my 2 + m ), (5)
dz
subject to the initial condition y(0) = 1. Lastly, the Jacobi elliptic function dn z ≡ dn(z|m),
with m = 1 − m > 0, is a solution to the differential equation
 2
dy
= (1 − y 2 )(y 2 − m ), (6)
dz
732 A J Brizard

1 1 1

2 4 6 8 2 4 6 8

-1 -1 2 4 6 8

Figure 1. From left to right, plots of sn(z|m), cn(z|m) and dn(z|m) from z = 0 to 4K(m) for
m = 4/5.

subject to the initial condition y(0) = 1. The differential equations (4)–(6) imply that the
Jacobi elliptic functions have the following derivatives with respect to z:
⎫
sn z = cn z dn z ⎬
cn z = −sn z dn z , (7)
⎭
dn z = −m cn z sn z
where the sign conventions satisfy the identities
sn2 (z|m) + cn2 (z|m) = 1 = dn2 (z|m) + m sn2 (z|m). (8)
We note that, for the special case m = 1, the solutions to the differential equations (4)–(6)
yield
⎫
sn(z|1) = tanh z ⎬
cn(z|1) = sech z , (9)
⎭
dn(z|1) = sech z
while for m = 0, they yield
⎫
sn(z|0) = sin z ⎬
cn(z|0) = cos z , (10)
⎭
dn(z|0) = 1
and both special cases trivially satisfy the identities (8).
The Jacobi elliptic functions (sn z, cn z, dn z) are doubly periodic functions of z, with
real-valued periods that are either 2K(dn z) or 4K (sn z and cn z) (see figure 1), where
 π/2  
dθ π 1 9
K ≡ K(m) = √ = 1 + m + m2 + · · · , (11)
0 1 − m sin2 θ 2 4 64
and purely imaginary periods that are either 2iK  (sn z) or 4iK  (cn z and dn z), where
 π/2
dθ
iK  (m) ≡ iK(m ) = i √ . (12)
0 1 − m sin2 θ
Figure 2 shows a plot of 2K(m)/π = 2K  (1 − m)/π from m = 0 to m = 1, with
K(0) = π/2 = K  (1) (the horizontal dashed line in figure 2) while K(m) = K  (1 − m) → ∞
as m → 1. For example, sn(z|0) = sin z has a real period of 4K(0) = 2π while
sn(z|1) = tanh z has an imaginary period of 2iK  (1) = iπ . Figure 3 shows the plots of
sn z and −i sn(iz) for m = 2/3, which exhibit both a real period (4K) and an imaginary period
(2iK  ). The fundamental period–parallelogram for the Jacobi elliptic functions is, therefore,
the rectangle with corners at (0, 4K, 4iK  , 4K + 4iK  ), where zeros occur for real values of z
(at 2K and 4K) while singularities occur for imaginary values of z (at iK  and 3iK  ).
A primer on elliptic functions with applications in classical mechanics 733

0.2 0.4 0.6 0.8 1

Figure 2. Plot of 2K(m)/π = 2K  (1 − m)/π from m = 0 to m = 1.

2 2

1 1

2 4 6 8 2 4 6
-1 -1

-2 -2

Figure 3. Plots of (left) sn(z|m) from z = 0 to z = 4K(m) for m = 2/3 and (right) −i sn(iz|m)
from z = 0 to z = 4K  (m) for m = 2/3.

The Jacobi elliptic functions for m > 1 (m < 0) are obtained from transformations of the
differential equations (4)–(6) in terms of the new variable m1/2 z. Hence, for m > 1, we find
⎫
sn(z|m) = m−1/2 sn(m1/2 z|m−1 )⎪ ⎬
cn(z|m) = dn(m1/2 z|m−1 ) , (13)
⎪
⎭
−1
dn(z|m) = cn(m z|m )
1/2

which satisfy identities similar to the identities (8).

2.1. Seiffert spherical spiral

A simple example that clearly exhibits the periodicity of the Jacobi elliptic functions sn z and
cn z is given by the Seiffert spherical spiral, [6, 10] defined as a periodic curve on the unit
sphere andconstructed as follows. First, we use the cylindrical metric ds 2 = dρ 2 +ρ 2 dϕ 2 +dz2
with z = 1 − ρ 2 , and the azimuthal angle ϕ(s) ≡ ks is parameterized by the arc length s
(assuming that the initial point of the curve is ρ = 0, ϕ = 0 and z = 1). Hence, we readily
obtain the differential equation
 2
dρ
= (1 − ρ 2 )(1 − k 2 ρ 2 ), (14)
ds
734 A J Brizard

y y y y

z z z z

x x x x

Figure 4. Three-dimensional plots of the Seiffert spherical spiral (16) on the surface of the unit
sphere for k = 0.15 from s = 0 to (from left to right) s = 2K(k 2 ), s = 4K(k 2 ), s = 6K(k 2 ) and
s = 8K(k 2 ).

Figure 5. Seiffert spherical spiral for k = 0.95 from s = 0 to s = 4K(k 2 ).

which leads (with 0 < m ≡ k 2 < 1) to the solutions

ρ(s) = sn(s|k 2 )
 , (15)
z(s) = 1 − ρ 2 (s) = cn(s|k 2 )

subject to the initial conditions ρ(0) = 0 and z(0) = 1. The Seiffert spherical spiral is
generated by plotting on the unit sphere the path of the unit vector


r(s) = sn(s|k 2 )[cos(ks) y] + cn(s|k 2 )
x + sin(ks) z (16)

as a function of s. Note that the identity sn2 s + cn2 s = 1 ensures that |r| = 1 for all values
of k. Figure 4 shows the Seiffert spherical spiral for k = 0.15; note that at each value
4nK(n = 1, 2, . . .), the orbit returns to the initial point at ρ = 0 and z = 1. The special
case k = 0 simply represents a great circle (r = sin s x + cos s
z) produced by the intersection
of the (x, z)-plane with the unit sphere. Figure 5 shows the complex periodic nature of the
Seiffert-spiral orbit for k = 0.95.
Lastly, the case k > 1 is handled with the identities (13), so that

ρ(s) = k −1 sn(ks|k −2 )
 , (17)
z(s) = 1 − ρ 2 (s) = dn(ks|k −2 )

and the Seiffert spherical spiral is now periodic with period 4k −1 K(k −2 ).
A primer on elliptic functions with applications in classical mechanics 735

Figure 6. Phase portrait (ϕ  versus ϕ) for the planar pendulum: the bounded libration orbits (19)
for
< 2 (inner curves) and the unbounded rotation orbits (20) for
> 2 (outer curves) are
separated by the separatrix orbits (21) for
= 2 (dashed curves).

2.2. Planar pendulum

The problem of the planar pendulum [1, 2] is represented by the (normalized) differential
equation
 2
dϕ

= − sin2 ϕ, (18)
dτ 2
where 2ϕ denotes the angular deviation of the pendulum from the vertical, τ ≡ νt denotes the
dimensionless time (with ν 2 ≡ g/L for a pendulum of length L in a gravitational field with
acceleration magnitude g) and
≡ E/(MgL) denotes the normalized energy of the pendulum
(of mass M).
We transform the differential equation (18) into the Jacobi differential equation (4) with
the substitution y(τ ) ≡ m−1/2 sin ϕ (where m ≡
/2) and thus the solution of the planar
pendulum (for m < 1) is
ϕ(τ ) = sin−1 [m1/2 sn(τ |m)]. (19)
When
> 2 (m > 1), on the other hand, we use the identities (13) to obtain
ϕ(τ ) = sin−1 [sn(m1/2 τ |m−1 )]. (20)
We note that the libration motion (19) of the planar pendulum has a period 4K(m) while the
rotation motion (20) of the planar pendulum has a period 4m−1/2 K(m−1 ). In the limit m = 1,
the identities (9) yield the separatrix solution
ϕ(τ ) = sin−1 (tanh τ ), (21)
which is expressed in terms of singly periodic (hyperbolic) trigonometric functions (with an
imaginary period). Since ϕ → π/2 as τ → ∞, the period of the pendulum on the separatrix
orbit is infinite.
Figure 6 shows the phase portrait (ϕ  versus ϕ) of the pendulum orbits (19)–(21). Here,
for m < 1, the derivative of equation (19) yields the libration angular velocity
m1/2
ϕ = √ sn τ = m1/2 cn(τ |m), (22)
1 − m sn2 τ
where the identities (7) and (8) are used. For m > 1, on the other hand, the rotation angular
velocity is
ϕ  = m1/2 dn(m1/2 τ |m−1 ), (23)
736 A J Brizard

where the identity (13) is used. Note that, since the rotation angular velocity (23) does not
vanish (see figure 1), the rotation orbits in the phase portrait shown in figure 6 are generated
with angular velocities of both signs. According to equation (9), the angular velocity on the
separatrix orbit (m = 1) is ϕ  = sech τ , which implies that the pendulum’s angular velocity
approaches zero exponentially as the pendulum approaches ϕ = π/2. Each orbit in √ the phase
portrait shown in figure 6 corresponds to the initial conditions ϕ0 = 0 and ϕ0 = ±
/2 (only
√and is generated with√−2K(
/2) < τ < 2K(
/2)
one sign is chosen for the libration orbits)
(for the libration orbits with
< 2) or −2 2/
K(2/
) < τ < 2 2/
K(2/
) (for the rotation
orbits with
> 2). Note that the topology of the phase portrait for the planar pendulum is
represented as a cylinder since the angle ϕ = −π/2 is physically identical to ϕ = π/2.
We shall return to the planar pendulum later in section 3.1 to highlight the connection
between the Jacobi and Weierstrass elliptic functions (presented in appendix A).

2.3. Force-free asymmetric top

As a second physical example, we consider the Euler equations for a force-free asymmetric
top (with principal moments of inertia I1 > I2 > I3 ) [2]:
⎫
I1 ω̇1 = (I2 − I3 )ω2 ω3 ⎬
I2 ω̇2 = −(I1 − I3 )ω1 ω3 , (24)
⎭
I3 ω̇3 = (I1 − I2 )ω1 ω2
where the angular velocity ω = ω1 1 + ω2 2 + ω3
3 is decomposed in terms of its components
along the principal axes of inertia. The conservation laws of kinetic energy,
 
κ = 12 I1 ω12 + I2 ω22 + I3 ω32 ≡ 12 I0 20 , (25)
and (squared) angular momentum,
2 = I12 ω12 + I22 ω22 + I32 ω32 ≡ I02 20 , (26)
are used to define the parameters I0 ≡  /(2κ) and 0 ≡ 2κ/. These parameters can be used
2

to introduce the following definitions:


I0 (I0 − I3 ) 
ω1 (τ ) = − 0 1 − y 2 (τ )
I1 (I1 − I3 )

≡ −1 (I0 ) 1 − y 2 (τ ), (27)

I0 (I0 − I3 )
ω2 (τ ) = 0 y(τ )
I2 (I2 − I3 )
≡ 2 (I0 )y(τ ), (28)

I0 (I1 − I0 ) 
ω3 (τ ) = 0 1 − my 2 (τ )
I3 (I1 − I3 )

≡ 3 (I0 ) 1 − my 2 (τ ), (29)

where τ = [(I1 − I3 )1 3 /(I2 2 )]t is the dimensionless time used in equations (27)–(29)
and the modulus m is defined as
(I0 − I3 )(I1 − I2 )
m(I0 ) ≡ . (30)
(I2 − I3 )(I1 − I0 )
A primer on elliptic functions with applications in classical mechanics 737

Figure 7. Three-dimensional (ω1 , ω2 , ω3 ) plots of the orbit (31) for the asymmetric free top from
τ = 0 to (from left to right) τ = K(m), τ = 2K(m), τ = 3K(m) and τ = 4K(m).

By requiring that the modulus m be positive, the parameter I0 = 2 /2κ must satisfy
I3 < I0 < I1 and, hence, 0  m(I0 )  1 for I3  I0  I2 and m(I0 ) > 1 for I2 < I0 < I1
(with m → ∞ as I0 → I1 ).
When we substitute these expressions into the Euler equation (24) for ω2 ≡ 2 y(τ ), we
easily obtain the dimensionless Jacobi differential equation (4), which can now be integrated,
with the initial conditions (ω1 (0), ω2 (0), ω3 (0)) = (−1 , 0, 3 ), to yield [2]

(ω1 , ω2 , ω3 ) = (−1 cn τ, 2 sn τ, 3 dn τ ), (31)

whose orbit is shown in figure 7 where the 4K periodicity is clearly observed. The solution
of this problem is thus very elegantly expressed in terms of the Jacobi elliptic functions
(sn, cn, dn). Note also that the Jacobi identities (8) can be used to show that the solution (31)
preserves the constants of the motion (25) and (26).
The separatrix solution (m = 1) corresponds to the case when I0 = I2 for which 2 ≡ 0 ,
so that the separatrix solution is

I2 (I2 − I3 )
ω1 (τ ) = − 0 sech τ,
I1 (I1 − I3 )
ω2 (τ ) = 0 tanh τ,

I2 (I1 − I2 )
ω3 (τ ) = 0 sech τ.
I3 (I1 − I3 )

Lastly, the motion of a symmetric top (I1 = I2 = I3 ) corresponds to the limit m ≡ 0 (and
2 = 1 ). The Jacobian solution (31) for a symmetric top thus becomes (ω1 , ω2 , ω3 ) =
(−1 cos τ, 1 sin τ, 3 ), where τ ≡ (1 − I3 /I1 )3 t is now the dimensionless time.

3. Weierstrass elliptic functions

The Weierstrass elliptic function ℘ (z + γ ) ≡ ℘ (z + γ ; g2 , g3 ) is defined as the solution of the

differential equation
 2
dy
= 4y 3 − g2 y − g3
dz
≡ 4(y − e1 )(y − e2 )(y − e3 ), (32)
738 A J Brizard

2 4 6 8

Figure 8. Plots of ℘ (z) > 0 (solid lines) and ℘ (i z) < 0 (dashed lines) for g2 = 3 and g3 = 1/2
(with  > 0) showing the real period 2ω (upper graph) and the imaginary period 2ω (lower graph)
defined, respectively, by equations (34) and (35).

Table 1. Cubic roots (e1 , e2 , e3 ) and half-periods (ω1 , ω2 , ω3 ) for the Weierstrass elliptic function.

(g3 , ) e1 e2 e3 ω1 ω2 ≡ ω1 + ω3 ω3
(−, −) a − ib a + ib −2a < −1 | | + i/2 | | − i/2 −i
(−, +) d>0 c−d >0 −c < 0 |ω | −iω + |ω | −iω
(+, +) c>0 d −c <0 −d < 0 ω ω + ω ω
(+, −) 2a > 1 −a − ib −a + ib  /2 +  −/2 + 

subject to the initial condition y(0) = ℘ (γ ). Here, (e1 , e2 , e3 ) denote the roots of the cubic
polynomial 4y 3 − g2 y − g3 (such that e1 + e2 + e3 = 0), and the invariants g2 and g3 are defined
in terms of the cubic roots as [6, 8]
⎫
g2 = −4(e1 e2 + e2 e3 + e3 e1 )⎪
⎬
 
= 2 e12 + e22 + e32 . (33)
⎪
⎭
g3 = 4e1 e2 e3
The applications of Weierstrass elliptic functions are analysed in terms of four different
cases (see table 1) based on the signs of (g3 , ) = [(−, −), (−, +), (+, −), (+, +)], where
 = g23 − 27g32 is the modular discriminant.
Figure 8 shows that, for  > 0, ℘ (z) has different periods 2ω and 2ω along the real and
imaginary axes, respectively, with the half-periods ω and ω defined as
 ∞
ds
ω(g2 , g3 ) =  , (34)
e1 4s − g2 s − g3
3
 e3
ds
ω (g2 , g3 ) = i  . (35)
−∞ |4s − g2 s − g3 |
3

Figure 9(a) shows

 theplots of ω(g2 , g3 ) and ω (g2 , g3 ) for g2 = 3 as functions of 0 < g3 < 1,
with  = 27 1 − g32 > 0. Note that for g3 = 0 (with e1 = −e3 and e2 = 0), we find that
ω ≡ iω while |ω | approaches infinity as g3 approaches 1 (or  approaches zero). Explicit
A primer on elliptic functions with applications in classical mechanics 739

Figure 9. Plots of (a) ω and Im(ω ) for 0 < g3 < 1 and g2 = 3 (i.e.,  > 0) and (b)  and Im( )
for g3 > 1 and g2 = 3 (i.e.,  < 0). Note that, as  → 0 (i.e., g3 → 1 when g2 = 3), both
Im(ω ) and Im( ) become infinite while (g2 , 1) = ω(g2 , 1). Lastly, ω (g2 , 0) = iω(g2 , 0) and
both (,  ) decrease to zero as g3 becomes infinite.

calculations of the Weierstrass half-periods (34)–(35) in terms of Jacobian quarter-periods K

and K  are given in appendix B.
For  < 0, on the other hand, ℘ (z) has different periods 2 and 2 along the real and
imaginary axes, respectively, with the half-periods  and  defined as
 ∞
ds
(g2 , g3 ) =  , (36)
e1 4s − g2 s − g3
3
 e1
ds
 (g2 , g3 ) = i  . (37)
−∞ |4s − g2 s − g3 |
3

Figure 9(b) shows

 the plots of (g2 , g3 ) and  (g2 , g3 ) for g2 = 3 as functions of g3 > 1,
with  = 27 1 − g32 < 0. Note that ω(g2 , 1) = (g2 , 1), | | approaches infinity as g3
approaches 1, and that both  and  approach zero as g3 approaches infinity.
Table 1 shows the cubic roots ei = (e1 , e2 , e3 ) and the half-periods ωi = (ω1 , ω2 , ω3 ),1
which satisfy the following definitions: [6] ℘ (ωi ) ≡ ei and ℘ (z + 2ωi ) ≡ ℘ (z), while, for
i = j = k, we have the identity [6]
(ei − ej )(ei − ek )
℘ (z + ωi ) = ei + , (38)
℘ (z) − ei
so that ℘ (ωi + ωj ) = ek . Figure 10 shows the plots of ℘ (z + ω2 ) and ℘ (z + ω3 ) for one
complete period from z = 0 to 2ω1 , which obey the identity (38). The singular behaviour
predicted by equation (38) at z = ωi , on the other hand, is shown in figure 8 for i = 1 and
j, k = 1. Additional properties of the Weierstrass elliptic function are that (i) it is an even-
parity function ℘ (−z) = ℘ (z), and (ii) under a change of sign g3 > 0 → g3 = −|g3 | < 0
(with fixed g2 and, thus, fixed discriminant ), the Weierstrass elliptic function satisfies the
identity
℘ (z; g2 , g3 ) ≡ −℘ (iz; g2 , |g3 |). (39)
1 The reader should be warned that only the case (g , ) = (+, +) follows the standard convention [8]. The
3
convention presented in table 1 was developed from a careful study of particle orbits in a cubic potential [9].
740 A J Brizard

1 2

-0.5

-1

Figure 10. Plots of ℘ (z + ω2 ) (solid line) and ℘ (z + ω3 ) (dashed line) for g2 = 3 and g3 = 1/2
(with  > 0) over one complete period from 0 to 2ω1 . Note that ℘ (ωj ) = ej for j = 2 or 3 and
℘ (ωi + ωj ) = ek , for i = 1 and (j, k) = (2, 3) or (3, 2).

The g3 -inversion identity (39) is used in table 1 for writing the following transformations
ei+ (g3 > 0) → ei− (g3 < 0) for the Weierstrass roots (for fixed g2 and ):
⎛ −⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ +⎞
e1 ℘ (ω1− ; g2 , g3 ) ℘ (iω1− ; g2 , |g3 |) ℘ (ω3+ ; g2 , |g3 |) e3
⎝e2− ⎠ ≡ ⎝℘ (ω2− ; g2 , g3 ) ⎠ ≡ −⎝℘ (iω2− ; g2 , |g3 |) ⎠ ≡ −⎝℘ (ω2+ ; g2 , |g3 |) ⎠ ≡ − ⎝e2+ ⎠ ,
e3− ℘ (ω3− ; g2 , g3 ) ℘ (iω3− ; g2 , |g3 |) ℘ (ω1+ ; g2 , |g3 |) e1+
(40)
which makes use of the transformation
 − − −  
ω1 , ω2 , ω3 ≡ −i ω3+ , ω2+ , ω1+ . (41)
The transformations (40)–(41) were previously introduced in order to present a uniform
solution of the problem of a particle moving in a cubic potential in terms of the Weierstrass
elliptic function for all values of energy and for all bounded and unbounded orbits [9]. They
also play a fundamental role in discussing the Weierstrass solution of the planar-pendulum
problem (see equations (46) and (47)).
Lastly, we note that, in contrast to the simple rectangular form of the fundamental
period–parallelogram of the Jacobi elliptic functions, the fundamental period–parallelogram
(0, ω1 , ω2 = ω1 + ω3 , ω3 ) of the Weierstrass elliptic function changes shape depending on the
signs of (g3 , ) in table 1.

3.1. Planar pendulum

For our first example, we return to the planar-pendulum problem solved in section 2.2 in terms
of the Jacobi elliptic functions. By writing y = 2 sin2 ϕ (i.e., 0 < y < 2), we transform
equation (18) into the cubic-potential equation
(y  )2 = 2y(2 − y)(
− y), (42)
with turning points at y = 0, 2 and
. Physical motion is possible only when the right side of
equation (42) is positive. Hence, the motion is periodic between y = 0 and y =
for
< 2,
while the motion is periodic between y = 0 and y = 2 for
> 2. We recover the standard
Weierstrass differential equation (32) by setting
y(τ ) = 2℘ (τ + γ ) + μ, (43)
A primer on elliptic functions with applications in classical mechanics 741
√
Table 2. Weierstrass roots (e1 , e2 , e3 ) and Jacobi parameters κ = e1 − e3 and m =
(e2 − e3 )/(e1 − e3 ) for the planar-pendulum problem.

Case (g3 , )
e3 e2 e1 γ = ω3 Half-period ω1 κ m
(a) (+, +) 0<
<1 −μ/2 μ−1 1 − μ/2 ω ω 1
/2
(b) (−, +) 1<
<2 −μ/2 μ−1 1 − μ/2 −iω |ω | 1
/2
√
(c) (−, +) 2<
<4 −μ/2 1 − μ/2 μ−1 −iω |ω |
/2 2/

√
(d) (+, +) 4<
−μ/2 1 − μ/2 μ−1 ω ω
/2 2/

where μ ≡ (
+ 2)/3 and the constant γ is determined from the initial condition y(0). The
Weierstrass invariants g2 and g3 are
g2 = 1 + 3(μ − 1)2 and g3 = μ(μ − 1)(μ − 2),
and the modular discriminant is  =
(2 −
)  0. 2 2

The Weierstrass solution of the planar pendulum is discussed in terms of the four cases
summarized in table 2, where the root (−μ/2) corresponds to the turning point y = 0, the
root (1 − μ/2) corresponds to the turning point y = 2 and the root (μ − 1) corresponds to the
turning point y =
. Using the initial conditions y(0) = 0 and y  (0) = 0, i.e., γ ≡ ω3 so that
℘ (γ ) = e3 , the Weierstrass solutions for cases (a) and (d) are expressed as
y(τ ) = 2℘ (τ + ω ) + μ, (44)
where ω3 = ω and the period of oscillation is 2ω1 = 2ω (see (g3 , ) = (+, +) in table 1).
For cases (b) and (c), the Weierstrass solutions are expressed as
y(τ ) = 2℘ (τ − iω) + μ, (45)

where ω3 = −iω and the period of oscillation is the period of oscillation 2ω1 = 2|ω | (see
(g3 , ) = (−, +) in table 1). As expected, when
→ 2 (i.e.,  → 0), the period 2|ω |
approaches infinity as we approach the pendulum’s separatrix (see figure 9).
Each pair of cases (a)–(b) and (c)–(d) in table 2 satisfies the transformations (40)–(41).
Consider, for example, the pair (a)–(b): in case (a), we denote the initial angle as ϕ0 (with
ϕ0 = 0), where 0 < ϕ0 < π/4, so that 0 <
= 2 sin2 ϕ0 < 1, while in case (b), we denote
the initial angle as ϕ 0 (with ϕ 0 = 0), where π/4 < ϕ 0 < π/2, so that 1 <
≡ 2 sin2 ϕ 0 < 2.
We now introduce the transformation
     
ϕ0 ϕ0 π/2 − ϕ0
→ ≡ , (46)


2−

which generates the g3 -inversion transformation (g2 , g3 , ) → (g 2 , g 3 , ) = (g2 , −g3 , )

on the Weierstrass invariants for the planar pendulum. From the transformation (46),
we obtain μ → μ = 2 − μ, and thus (e1 , e2 , e3 ) → (e1 , e2 , e3 ) = −(e3 , e2 , e1 ) and
(ω1 , ω2 , ω3 ) → (ω1 , ω2 , ω3 ) = −i(ω3 , ω2 , ω1 ), in exact agreement with the transformations
(40)–(41). Next, the solution y(τ ; ω1 ) = 2℘ (τ +ω3 ; g 2 , g 3 )+μ, which has the real half-period
ω1 = |ω |, is expressed as
y(τ ; ω1 ) = 2℘ (τ − iω1 ; g2 , −g3 ) + (2 − μ)
= 2 − (2℘ (iτ + ω1 ; g2 , g3 ) + μ) ≡ 2 − y(iτ ; −iω3 ), (47)

where the solution y(iτ ; −iω3 ), which has the real half-period −iω3 = |ω |, is a function of
imaginary time iτ . Because the normalized time τ ≡ (g/L)1/2 t involves the gravitational
acceleration g, we obtain an imaginary time if we invert gravity’s direction (g → g ≡ −g), so
that τ → τ ≡ iτ and y(iτ ; −iω3 ) ≡ y(τ ; |ω |). The physical interpretation of the imaginary
742 A J Brizard

0.5 1 1.5 2

Figure 11. Upper half of the phase portrait (y  versus y) for the planar pendulum represented in
terms of the Weierstrass elliptic solutions (44) and (45). The separatrix orbit (dashed curve with

= 2) separates the libration orbits (a) and (b) (inner curves with turning points 0 and
< 2) from
the rotation orbits (c) and (d) (outer curves with turning points 0 and 2).

half-period ω of the planar pendulum is, therefore, that its magnitude |ω | is the real half-period
of the inverted planar pendulum (or pendulum with imaginary time) [1].
Figure 11 shows the phase portrait (y  = 2℘  versus y = 2℘ + μ) for the planar pendulum
represented in terms of the Weierstrass elliptic solutions (44) and (45). Here, the orbits are
labelled with
= (0.75, 1.5, 2, 2.5, 5), from the inner curve (
= 0.75) to the outer curve
(
= 5), and each orbit is generated with −ω1 < τ < ω1 . We note that the transformation
y = 2 sin2 ϕ has replaced the Jacobian phase portrait (figure 6), whose nontrivial topology is
represented as an infinite cylinder, with the Weierstrassian phase portrait (figure 11), whose
trivial topology is represented as an infinite strip (because y is limited to the range 0  y  2
and −∞ < y  < ∞).
Lastly, the solution of the planar pendulum in terms of the Jacobi elliptic function sn(z|m)
and the Weierstrass elliptic function ℘ (z+γ ) suggests a close connection between these elliptic
functions. A derivation of the relations between the Jacobi and Weierstrass elliptic functions
is presented in appendix A (see equations (A.5)–(A.7)). Using equation (A.6), for example,
we indeed find the general solution for the planar pendulum for all values of normalized
energy
:



sn2 (τ |
/2) (
< 2)
2℘ (τ + ω3 ) + μ = 2mκ sn (κτ |m) =
2 2
√ , (48)
2 sn2 (
/2τ |2/
) (
> 2)
A primer on elliptic functions with applications in classical mechanics 743

√
where κ = e1 − e3 and m = (e2 − e3 )/(e1 − e3 ) (see table 2). Appendix B shows that
the Weierstrass half-periods ω and ω are related to the Jacobian quarter-periods K and K  as
ω ≡ K(m) and ω = iK  .

3.2. Spherical pendulum

Our second example is the spherical pendulum (whose pendulum bob moves on the surface
of a unit sphere), which is described here in terms of the cylindrical coordinates (ρ, ϕ, z) [1].
The energy equation for a spherical pendulum of unit mass and unit length is

ν 2 = 12 (ρ̇ 2 + ρ 2 ϕ̇ 2 + ż2 ) + ν 2 z,
√
where
denotes the normalized total energy (ν 2 ≡ g). By substituting ρ(z) = 1 − z2
(where −1  z  1) and the angular momentum conservation law  ≡ ρ 2 ϕ̇/ν, we obtain the
differential equation
(z )2 = 2(
− z)(1 − z2 ) − 2
≡ 2(z − z1 )(z − z2 )(z − z3 ), (49)
 −1
where z (τ ) ≡ ν ż and z1 + z2 + z3 =
. Because the right side of equation (49) is negative
at z = ±1, the highest root z1 (> 1 > z2 > z3 > −1) of the cubic polynomial is unphysical
(since the cylindrical radius ρ then becomes imaginary). The periodic motion of the spherical
pendulum is therefore bounded between z3 < z < z2 .
The differential equation (49) can be transformed into the standard differential
equation (32) by setting z(τ ) = 2℘ (τ + γ ) + μ, where μ ≡
/3, the constant γ is
determined from the initial condition z(0), and the invariants g2 and g3 are g2 = 1 + 3μ2
and g3 = 2 /4 + μ(μ2 − 1). If we choose the initial condition z(0) = z3 > −1, then the
constant γ is γ = ω3 = ω and the solution of the spherical pendulum problem for the z
(and ρ) coordinate is

z(τ ) = 2℘ (τ + ω3 ) + μ ≡ 1 − ρ 2 (τ ). (50)
The motion is periodic with half-period ω1 = ω so that z(ω1 ) = 2℘ (ω1 + ω3 ) + μ =
2℘ (ω2 ) + μ ≡ z2 < 1 as expected. Note that we recover the planar pendulum (discussed in
section 3.1) as a special case of the spherical pendulum with  ≡ 0 and a suitable change of
definition for (z,
, μ).
The solution for the azimuthal angle ϕ(τ ) is obtained from the angular-momentum
conservation law ϕ  (τ ) = /ρ 2 (τ ), which is integrated as [1]
 τ
ds
ϕ(τ ) = 
0 1 − [2℘ (s + ω3 ) + μ]
2
 τ
 ds
≡− , (51)
4 0 [℘ (s + ω3 ) − ℘ (κ)][℘ (s + ω3 ) − ℘ (λ)]
where we used the initial condition ϕ(0) = 0 and the imaginary constants κ and λ are defined
by the relations ℘ (κ) = −(1 + μ)/2 and ℘ (λ) = (1 − μ)/2 corresponding to z = −1 < z3
and z = +1 > z2 , respectively. These constants also yield the relations ℘  (κ) = i/2 = ℘  (λ),
obtained from the Weierstrass differential equation (32) for z = κ and λ. These relations allow
us to write equation (51) as
  
i τ ℘  (λ) ℘  (κ)
ϕ(τ ) = ds − , (52)
2 0 ℘ (s + ω3 ) − ℘ (λ) ℘ (s + ω3 ) − ℘ (κ)
where we used the identity ℘ (λ) − ℘ (κ) = 1.
744 A J Brizard

The integral (52) can be solved exactly in terms of the quasi-periodic functions ζ (τ ) and
σ (τ ) associated with the Weierstrass elliptic function ℘ (τ ) [6]: ℘ (τ ) ≡ −dζ (τ )/dτ and
ζ (τ ) ≡ σ  (τ )/σ (τ ). Using the identity
℘  (λ)
≡ ζ (s − λ) − ζ (s + λ) + 2ζ (λ)
℘ (s) − ℘ (λ)
 
d σ (s − λ)
= ln + 2ζ (λ),
ds σ (s + λ)
we find the standard solution [1] for the azimuthal motion of the spherical pendulum
   
σ (τ + ω3 + λ)σ (ω3 − λ) σ (τ + ω3 − κ)σ (ω3 + κ)
e2iϕ(τ ) = e2τ [ζ (κ)−ζ (λ)] × , (53)
σ (ω3 + λ)σ (τ + ω3 − λ) σ (ω3 − κ)σ (τ + ω3 + κ)
where we can easily verify that ϕ(0) = 0. This solution (presented in Whittaker’s book
[1]) can be studied numerically with Mathematica since its library of functions contains the
Weierstrass elliptic function ℘ (τ ) and its associated quasi-periodic functions ζ (τ ) and σ (τ ).
It turns out that we may further simplify the standard solution (53) by recognizing that,
since −1 < z3  ℘ (τ + ω3 )  z2 < +1 for real values of τ , there must be imaginary numbers
iα and iβ (where α and β are real-valued constants) such that κ ≡ ω3 + iα and λ ≡ ω3 + iβ.
Using these substitutions, the solution (53) becomes
   
σ (τ + 2ω3 + iβ)σ (−iβ) σ (τ − iα)σ (2ω3 + iα)
e2iϕ(τ )
=e 2τ [ζ (κ)−ζ (λ)]
× . (54)
σ (2ω3 + iβ)σ (τ − iβ) σ (−iα)σ (τ + 2ω3 + iα)
Next, using the identity [6] σ (τ + 2ω3 ) ≡ − exp[2η3 (τ + ω3 )]σ (τ ), where η3 ≡ ζ (ω3 ), we
now find
σ (τ + 2ω3 + iβ) σ (τ + iβ)
= −e2η3 (τ +iβ)
σ (τ − iβ) σ (τ − iβ)
and
σ (−iβ) σ (−iβ)
= −e−2iη3 β ≡ e−2iη3 β ,
σ (2ω3 + iβ) σ (iβ)
where we used the fact that σ (τ ) is an odd function of τ , so that σ (−iβ) = −σ (iβ). Hence,
we obtain
 
σ (τ + 2ω3 + iβ)σ (−iβ) σ (τ + iβ)
≡ −e2η3 τ ,
σ (2ω3 + iβ)σ (τ − iβ) σ (τ − iβ)
and
 
σ (τ − iα)σ (2ω3 + iα) σ (τ − iα)
≡ −e−2η3 τ .
σ (−iα)σ (τ + 2ω3 + iα) σ (τ + iα)
We combine these relations to obtain the simplified solution
 
2τ [ζ (κ)−ζ (λ)] σ (τ + iβ)σ (τ − iα)
e2iϕ(τ )
≡e . (55)
σ (τ − iβ)σ (τ + iα)
Lastly, we note that the ratio σ (τ + iβ)/σ (τ − iβ) must have unit modulus for real τ -values,
so that we may write
   β
σ (τ + iβ)
ln =i ζ (τ + is) ds
σ (τ − iβ) −β
 β
= 2i Re [ζ (τ + is)] ds.
0
A primer on elliptic functions with applications in classical mechanics 745

Hence, the solution (53) for the azimuthal angle ϕ(τ ) can finally be given in a more compact
form as
 β
ϕ(τ ) = Re[ζ (τ + is)] ds + iτ [ζ (ω3 + iβ) − ζ (ω3 + iα)]
α
 β 
≡ Re [ζ (τ + is) + τ ℘ (ω3 + is)] ds , (56)
α

which is now expressed only in terms of the quasi-periodic function ζ (τ ) and ℘ (τ ) = −ζ  (τ ).

After a full period 2ω1 , when the (ρ, z)-coordinates return to their initial values, the azimuthal
angle has changed by an amount ϕ ≡ ϕ(τ + 2ω1 ) − ϕ(τ ) expressed as
 β
ϕ = 2ω1 ℘ (ω3 + is) ds + 2η1 (β − α), (57)
α
where we used the identity [6] ζ (τ + 2ω1 + is) = ζ (τ + is) + 2η1 and η1 ≡ ζ (ω1 ).
The problem of the spherical pendulum, expressed in terms of the cylindrical coordinates
(ρ, ϕ, z), is thus exactly (and compactly) solved by equations (50) and (56) in terms of the
Weierstrass elliptic function ℘ (τ + γ ) and its associated function ζ (τ ).

3.3. Heavy symmetric top with one fixed point

Our final example is provided by the motion of a symmetric top (I1 = I2 = I3 ) with one fixed
point described in terms of the energy equation
 
1 (pϕ − pψ cos θ )2
E= 2 2
I1 θ̇ + I3 3 + + Mgh cos θ ,
2 I1 sin2 θ
where E is the total energy of the symmetric top (with total mass M and principal
moments of inertial I1 = I2 = I3 ), 3 denotes the constant component of the angular
velocity and the angular momenta pϕ and pψ associated with the ignorable Eulerian
angles ϕ and ψ are  constants of motion. By defining the dimensionless parameters

= E − 12 I3 32 (Mgh), (a, b) = (pϕ /I1 ν, pψ /I1 ν), where ν 2 = Mgh/(2I1 ), the
differential equation for u = cos θ is obtained from equation (58) as
(u )2 = 4(1 − u2 )(
− u) − (a − bu)2
≡ 4(u − u1 )(u − u2 )(u − u3 ), (59)
where the prime denotes a derivative with respect to the dimensionless time τ = νt and
u3 < u2 < u1 are the three roots of the cubic polynomial on the right side of equation (59).
Since the right side of equation (59) is negative at u = ±1, we conclude that −1 < u3 < u2 < 1
and u1 > 1 (which is unphysical for u = cos θ ). The physical motion is therefore periodic in
θ and is bounded between u3 ≡ cos θ3 and u2 ≡ cos θ2 (or θ2 < θ (τ ) < θ3 ).
By using the change of integration variable u = ℘ + μ, where μ = (4
+ b2 )/12, the
differential equation (59) becomes the standard differential equation (32) for the Weierstrass
elliptic function, where the invariants

g2 = 2(2 − ab + 6μ2 )
(60)
g3 = (a 2 + b2 ) + 2μ(4μ2 − 4 − ab)
and the discriminant  = g23 − 27g32 depend on the three constants of the motion (
, a, b) for
the problem. Hence, the solution is expressed in terms of the Weierstrass elliptic function as
u(τ ) ≡ cos θ (τ ) = ℘ (τ + γ ) + μ, (61)
746 A J Brizard

where γ is determined from the initial condition θ (0). Assuming that −1 < u3 = e3 + μ <
u2 = e2 + μ < 1 < u1 = e1 + μ, we choose u(0) = e3 + μ = u3 (i.e., −1 − μ < e3 < 1 − μ) so
that γ ≡ ω and, hence, at the half-period τ = ω, we find u(ω) = ℘ (ω + ω ) + μ = e2 + μ = u2
as expected.
The solution (61) for the Eulerian angle θ (τ ) can now be used to integrate the differential
equations for the remaining Euler angles ϕ and ψ:
a − b cos θ (τ )
ϕ  (τ ) =
1 − cos2 θ (τ )
 
i ℘  (κ) ℘  (λ)
≡ − , (62)
2 ℘ (τ + ω3 ) − ℘ (κ) ℘ (τ + ω3 ) − ℘ (λ)
and, defining ψ  ≡ (3 /ν − b) + χ  ,
b − a cos θ (τ )
χ  (τ ) =
1 − cos2 θ (τ )
 
i ℘  (−κ) ℘  (λ)
≡ − , (63)
2 ℘ (τ + ω3 ) − ℘ (−κ) ℘ (τ + ω3 ) − ℘ (λ)
where ℘ (κ) = 1 − μ and ℘ (λ) = −(1 + μ), with ℘  (κ) = i(a − b) and ℘  (λ) = i(a + b).
Note that the sign of ϕ  in equation (62) depends on the sign of a − b cos θ2 < a − b cos θ <
a − b cos θ3 . If a > b cos θ2 (or a < b cos θ3 ), ϕ  does not change sign as θ bounces between
θ2 and θ3 and the motion in ϕ involves monotonic azimuthal precession. If a = b cos θ2 (or
a = b cos θ3 ), ϕ  vanishes at θ = θ2 (or θ = θ3 ) and the motion in ϕ exhibits a cusp at that
angle (since both θ  and ϕ  vanish). If a < b cos θ2 , ϕ  vanishes at an angle θ2 < θ0 < θ3 and
the motion in ϕ exhibits retrograde motion between θ0 < θ < θ3 .
Lastly, since equation (62) is the same as equation (52), its solution is identical to
equation (56) (even if the constants κ and λ are different). This same solution can
also be applied to the solution for equation (63), where equation (62) is transformed into
equation (63) by performing the change (a, b) → (b, a) and noting that ℘ (τ ) has even parity,
i.e., ℘ (−κ) = ℘ (κ), while ℘  (τ ) has odd parity, i.e., ℘  (−κ) = −℘  (κ).

4. Other applications of elliptic functions in physics

The previous two sections showed problems of classical mechanics, involving the orbital
motions of particles and rigid bodies, that can be solved explicitly in terms of Jacobi and
Weierstrass elliptic functions. More problems of classical mechanics with solutions expressed
in terms of the Jacobi and Weierstrass elliptic functions can be found in Whittaker’s textbook
[1] as well as more recent textbooks [9, 11]. These special functions have many more
applications outside of classical mechanics as well [11], including particle orbits and light
paths in general relativity [12] and solutions of cosmological models [13]. These applications
unfortunately involve advanced topics that fall well outside the purpose of the present work.
One interesting application worthy of discussion, however, deals with exact solutions of
the Korteweg–de Vries (KdV) equation [14]
∂u ∂u ∂ 3 u
+u + = 0, (64)
∂t ∂x ∂x 3
which describes the nonlinear evolution of the field u(x, t). This nonlinear equation appears
in many areas of physics [15] and is a member of an important class of nonlinear partial
differential equations that possesses soliton solutions [16–18].
A travelling-wave solution of the Korteweg–de Vries (KdV) equation (64) is a function of
the form u(x, t) = v(ξ ), where ξ = κ(x − ct) denotes the wave phase (with constants κ and
A primer on elliptic functions with applications in classical mechanics 747

c to be determined). Substituting this travelling solution into the KdV equation, we obtain an
ordinary differential equation for v(ξ ):
(v − c)v  + κ 2 v  = 0,
which can be integrated with respect to ξ to yield
κ 2 v  = α + cv − 12 v 2 , (65)

where α is a constant of integration. If we multiply equation (65) with v and integrate again
with respect to ξ , we obtain
κ2  2 c 1
(v ) = (αv + β) + v 2 − v 3 , (66)
2 2 6
where β is a second constant of integration. It is now immediately clear that v(ξ ) ≡ A℘ (ξ )+B
can be expressed in terms of elliptic functions (where A ≡ −12κ 2 , B ≡ c) because the right
side of equation (66) involves a cubic polynomial in v. The travelling-wave solution of the KdV
equation (64) is
u(x, t) = A℘[κ(x − ct) + γ ] + B,
where the constant γ is determined from the initial condition u(x, 0) = u0 (x).
Using the relation between the Weierstrass and Jacobi elliptic functions (see appendix A),
the travelling-wave solution to the KdV equation (64) may also be expressed as [17, 19]
u(x, t) = a cn2 [κ(x − ct) + γ |m] + b,
√ √
where m = (r3 − r2 )/(r3 − r1 ), a = r3 − r2 , b = r2 and κ = (r3 − r1 )/6; here,
r1 < r2 < r3 are the roots of the cubic polynomial on the right side of equation (66). This
second representation is known as the periodic cnoidal-wave solution of the KdV equation
(64).
Lastly, we note that for the special case α = 0 = β in equation
√ (66), for which r3 = 3c
and r1 = 0 = r2 , then we find m = 1, a = 3c, b = 0, κ = c/2, and the travelling-wave
solution becomes
 
c
u(x, t) = 3c sech2 (x − ct) ,
2
which describes the well-known localized soliton solution of the KdV equation (64).

5. Summary

We presented a brief introduction of the Jacobi and Weierstrass elliptic functions with
applications in classical mechanics. The problem of the planar pendulum was used to establish
a connection (equation (48)) between the Jacobi and Weierstrass elliptic functions. The double
periodicity of the elliptic functions was easily observed in each of the problems of classical
mechanics discussed in sections 2 and 3. We also briefly discussed applications of the
elliptic functions in section 4 in other areas of physics (e.g., travelling-wave solutions of the
Korteweg–de Vries equation).
Because of their similarity with trigonometric functions sin z and cos z, physicists are
most familiar with the Jacobi elliptic functions sn(z|m) and cn(z|m). This familiarity is
further increased by the fact that the problems of the planar pendulum and the force-free
asymmetric top are solved simply and elegantly in terms of the Jacobi elliptic functions (as
shown in section 2).
Physicists are usually less familiar with the Weierstrass elliptic function ℘ (z; g2 , g3 ),
introduced in section 3. The complexity of the Weierstrass elliptic function is partly due to
748 A J Brizard

the fact that the form of its fundamental period–parallelogram depends on the invariants g2
and g3 (and the modular discriminant  = g23 − 27g32 ). This is in contrast to the fundamental
period–parallelogram of the Jacobi elliptic function which remains rectangular for all values
of m = 1. The greatest advantage of the Weierstrass elliptic function, however, is that the
function ℘ (z) itself and its derivative ℘  (z) can be used to represent any doubly periodic
function f (z) ≡ A(℘)℘  + B(℘), where A and B are arbitrary functions. This simplicity was
further demonstrated by the solution of the planar pendulum in terms of a single expression
(48) for all values of energy.
While the introduction of the Jacobi and Weierstrass elliptic functions in the undergraduate
curriculum remains difficult, it is hoped that the present primer, combined with the easy access
to mathematical software, will facilitate their dissemination.

Appendix A. Relation between Weierstrass and Jacobi elliptic functions

In this appendix, we explore the connection between the Jacobi and Weierstrass elliptic
functions. For this purpose, we begin with the differential equation for the Weierstrass elliptic
function
 2
dy
= 4y 3 − g2 y − g3 , (A.1)
dx
and introduce the transformation
y(x) = αs p (κx) + β, (A.2)
where p is an integer (= 0, 1) and (α, β, κ) are constants to be determined. Under the
transformation (A.2), equation (A.1) becomes (with z = κx)
 2
ds 4α 12β s 2−p s 2−2p
= 2 2 s 2+p + 2 2 s 2 + (12β 2 − g2 ) 2 2 + (4β 3 − g2 β − g3 ) 2 2 2 . (A.3)
dz p κ p κ αp κ α p κ
The constants (α, β, κ) and the integer p = 0, 1 are chosen such that the right side of
equation (A.3) has the Jacobi form (4). For p = 2, we recover the Jacobi form (4) if
κ2
β = −(m + 1) (A.4)
3
is a root of the cubic polynomial 4β 3 − g2 β − g3 (i.e., β = e1 , e2 , or e3 ) and
12β 2 − g2
α = mκ 2 = .
4κ 2
For p = −2, on the other hand, we recover the Jacobi form (4) if β, given by equation (A.4),
is again a root of the cubic polynomial 4β 3 − g2 β − g3 (i.e., β = e1 , e2 , or e3 ) and
12β 2 − g2
α = κ2 = .
4mκ 2
Hence, the Jacobi elliptic function s(z) is related to the Weierstrass elliptic function y(x) in
the case of p = ±2.
An application
√ of the first transformation (p = 2) shows that for m = (e2 − e3 )/(e1 − e3 )
and κ = e1 − e3 , we find α = e2 − e3 and β = e3 , and we obtain the relation
℘ (x + ω2 ; g2 , g3 ) = e3 + (e2 − e3 )cn2 (κx|m) (A.5)
or
℘ (x + ω3 ; g2 , g3 ) = e3 + (e2 − e3 ) sn2 (κx|m) , (A.6)
A primer on elliptic functions with applications in classical mechanics 749

which oscillates between e2 = ℘ (ω2 ) = ℘ (ω + ω3 ) and e3 = ℘ (ω + ω2 ) = ℘ (ω3 ) (see

figure 10), where ω = K(m)/κ. Relation (A.6) plays a crucial role in expressing
the Weierstrassian solution of the planar pendulum in terms of its Jacobian solution in
equation (48).
An application of the second transformation (p = −2) shows that for the same m and κ
as the first transformation, we find α = e1 − e3 and β = e3 , and we obtain the relation
e1 − e3
℘ (x; g2 , g3 ) = e3 + , (A.7)
sn2 (κx|m)
which has singularities at x = 0 and 2K(m)/κ ≡ 2ω and a minimum (e1 ) at x = K(m)/κ ≡ ω
(see the upper plot in figure 8). We note that the relation (A.7) is equivalent to the property
(38) of the Weierstrass elliptic function:
(e1 − e3 )(e2 − e3 )
℘ (x) = e3 + ,
℘ (x + ω3 ) − e3
when substituting equation (A.6) into the denominator of equation (A.7).

Appendix B. Mathematical details

In this appendix, where we assume (g3 , ) = (+, +), we show how the half-periods ω and
ω of the Weierstrass elliptic function, defined in equations (34) and (35), are related to the
quarter periods K and K  of the Jacobi elliptic function.
First, we start with the half-period (34). By introducing the change of variable [20]
s = e3 + (e1 − e3 ) csc2 ψ, equation (34) transforms into
 π/2
dψ
ω(g2 , g3 ) = 
0 (e1 − e3 ) − (e2 − e3 ) sin2 ψ
K(m)
≡√ , (B.1)
e1 − e3
where the modulus m of the Jacobian quarter period is defined by the relation m =
(e2 − e3 )/(e1 − e3 ).
Next, we look at the half-period (35). By introducing the change of variable s =
e1 − (e1 − e3 ) csc2 ψ, we readily obtain
 π/2
dψ
ω (g2 , g3 ) = i 
0 (e 1 − e3 ) − (e1 − e2 ) sin2 ψ

iK(m )
≡√ , (B.2)
e1 − e3
where m = 1 − m = (e1 − e2 )/(e1 − e3 ).
Note that the relations (B.1)–(B.2) between the Weierstrass half-periods (ω, ω ) and the
Jacobi quarter-periods (K, K  ) hold because the relation (A.6) between the Weierstrass and
Jacobi elliptic functions involves the square of the Jacobi elliptic function, which reduces the
latter’s period by half (e.g., sin2 ϕ has a period of π ).
These relations played an important role in the Weierstrass solution of the planar pendulum
in section 3.1. Under the transformation
→
= 2 −
generated by the transformation
ϕ0 → ϕ 0 ≡ π/2−ϕ0 , the Jacobian modulus m ≡
/2 (in the case (a) of table 2) transforms into
the Jacobian modulus m = 1−
/2 ≡ m (in the case (b) of table 2), which is the complementary
750 A J Brizard

modulus of case (a). According to equations (B.1) and (B.2), since ω1 ≡ ω = K(m) and
ω3 ≡ ω = iK  (m) = iK(m ) in case (a), in case (b) we find

ω1 = K(m) = K  (m) ≡ −iω3
, (B.3)
ω3 = −iK  (m) = −iK(m) ≡ −iω1
exactly in agreement with the transformation (46).

References
[1] Whittaker E T 1937 A Treatise on the Analytical Dynamics of Particles and Rigid Bodies 4th edn (New York:
Dover)
[2] Landau L D and Lifshitz E M 1976 Mechanics (Amsterdam: Elsevier)
[3] Taylor J R 2005 Classical Mechanics (Sausalito, CA: University Science Books)
[4] Goldstein H, Poole C and Safko J 2002 Classical Mechanics 3rd edn (San Francisco: Addison-Wesley)
[5] Greenhill A G 1959 The Applications of Elliptic Functions (New York: Dover)
[6] Whittaker E T and Watson G N 1963 A Course of Modern Analysis 4th edn (Cambridge: Cambridge University
Press)
[7] Milne-Thomson L M 1965 Jacobi elliptic functions and theta functions Handbook of Mathematical Functions
ed M Abramowitz and I A Stegun (New York: Dover) chapter 16
[8] Southard T H 1965 Weierstrass elliptic and related functions Handbook of Mathematical Functions ed M
Abramowitz and I A Stegun (New York: Dover) chapter 18
[9] For a shorter version of the present paper (with additional material), see Brizard A J 2009 An Introduction to
Lagrangian Mechanics (Singapore: World Scientific) appendix B
[10] Erdös P 2000 Am. J. Phys. 68 888
[11] For a recent textbook on elliptic functions with applications in physics, see Armitage J V and Eberlein W 2006
Elliptic Functions (Cambridge: Cambridge University Press)
[12] Hartle J B 2003 Gravity: An Introduction to Einstein’s General Relativity (San Francisco: Addison-Wesley)
[13] Ryden B 2003 Introduction to Cosmology (San Francisco: Addison-Wesley)
[14] Gratton J and Delellis R 1989 Am. J. Phys. 57 683
[15] Giambo S, Pantano P and Tucci P 1984 Am. J. Phys. 52 238
[16] Gardner C S, Greene J M, Kruskal M D and Miura R M 1967 Phys. Rev. Lett. 19 1095
[17] Olver P J 1986 Applications of Lie Groups to Differential Equations (New York: Springer) section 3.2
[18] Degasperis A 1998 Am. J. Phys. 66 486
[19] Cervero J M 1986 Am. J. Phys. 54 35
[20] Critchfield C L 1989 J. Math. Phys. 30 295