Jacobi Elliptic Functions from a Dynamical Systems Point of View

Author(s): Kenneth R. Meyer

Source: The American Mathematical Monthly, Vol. 108, No. 8 (Oct., 2001), pp. 729-737
Published by: Taylor & Francis, Ltd. on behalf of the Mathematical Association of America
Stable URL: https://www.jstor.org/stable/2695616
Accessed: 08-05-2020 01:53 UTC

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/2695616?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Taylor & Francis, Ltd., Mathematical Association of America are collaborating with
JSTOR to digitize, preserve and extend access to The American Mathematical Monthly

This content downloaded from 186.85.131.131 on Fri, 08 May 2020 01:53:28 UTC
All use subject to https://about.jstor.org/terms
 Jacobi Elliptic Functions
from a Dynamical Systems Point of View

Kenneth R. Meyer

1. INTRODUCTION. The theory of Jacobi elliptic functions arose in an attempt to

integrate certain algebraic expressions, but soon found many applications to geome-
try, mechanics, physics, and engineering. The theory of these important functions is
vast, but an interesting introduction with some applications can be given even at the
advanced undergraduate level. These functions satisfy a simple system of differen-
tial equations, which can be analyzed using some of the basic theory of differential
equations. Many of their basic properties come as immediate applications of the fun-
damental theorems on existence, uniqueness, and continuous dependence of solutions
on initial conditions, and thus this system of equations serves as an excellent example
in a course in differential equations.
The Jacobi elliptic functions are important examples of doubly periodic meromor-
phic functions, but the dynamical systems approach that we present considers them as
real valued functions only. However, there are many applications of these functions in
the real domain.
My first attempts [10, pp. 8-10] to give a dynamical systems approach used the
second order differential equation presented in Section 2.4, but a far better approach is
found in the brief introduction found in Hille's classic book [9, pp. 66-74] on the an-
alytic theory of differential equations. It is Hille's definition that I give in Section 2.1.
A very nice dynamical systems introduction using this same definition can be found
in [4, pp. 445-449].
In Section 3, we sketch how to solve a class of differential equations that includes
the pendulum equation and the undamped Duffing equation.
Over the years in courses at the undergraduate and graduate level, I have used the
Jacobi elliptic functions as an example of the power of the geometric approach to
the theory of ordinary differential equations. Rarely do I present all the material in
Section 3 and often break up much of the material into a series of problems; but always
I present the solution of the pendulum equation.

2. THE JACOBI ELLIPTIC FUNCTIONS. For a moment, think of the various

definitions of sine and cosine that you have encountered over the years. For many of
us, they were first defined as ratios of sides of a right triangle or as coordinates of a
point on the unit circle. Some of us were subjected to a rigorous and laborious analytic
definition as found in the classic texts by Hardy [8, pp. 447-486] or Whittaker and
Watson [11, pp. 579-590]. A specialist in differential equations might define the sine
function as the solution of the harmonic oscillator satisfying x (0) = 0, x (0) = 1 [1],
or better still define x (t) = cos t, y (t) = sin t as the solution of a system of first order
equations x = y, y =-x. It is this latter approach that we take to define the Jacobi
elliptic functions.

2.1. The system definition. Let k be a number in (0, 1), and let t denote a real vari-
able that we interpret as time. The Jacobi elliptic functions sn(t, k), cn(t, k), dn(t, k)

October 2001] JACOBI ELLIPTIC FUNCTIONS 729

This content downloaded from 186.85.131.131 on Fri, 08 May 2020 01:53:28 UTC
All use subject to https://about.jstor.org/terms
are defined as the solutions of the system of differential equations

= yz

y=-zx (1)

= -k2xy

that satisfy the initial conditions

sn(0,k) = x(0) = 0, cn(0, k) =y(0) = 1, dn(0,k) = z(0) = 1. (2)

The dots in (1) denote differentiation with respect to t. The parameter k is known as
the modulus and satisfies 0 < k < 1; the complementary modulus is K = 1 .-k
One speaks of these functions by pronouncing the letters as " 's 'n' of 't' and 'k'
These functions have also been denoted by sin am(t, k), cos am(t, k), delta am(t, k)
and called sine amplitude, cosine amplitude and delta amplitude.
The equations (1) are real analytic in the variables t, x, y, z and the parameter k,
so the basic existence theory of ordinary differential equations ensures that the Jacobi
elliptic functions are smooth or even real analytic functions of t and k; see [9, pp. 48-
56, 90-96], [6, pp. 18-27], or [7, p. 10, 93-113]. The definition immediately gives the
derivatives for the functions, namely

d
- sn(t, k) = cn(t, k) dn(t, k),
dt
d
__cn(t, k) = - dn(t, k) sn(t, k), (3)
dt

- dn(t, k) = -k2 sn(t, k) cn(t, k).

dt

The following theorem is an interesting application of the theorem on the continu-

ous dependence of solutions of a differential equation on parameters.

Proposition 2.1. As k approaches O from the right we have

sn(t, k) -* sin(t), cn(t, k) -> cos(t), dn(t, k) -> 1, (4)

and as k approaches 1 from the left we have

sn(t, k) -> tanh(t), cn(t, k) -. sech(t), dn(t, k) -. sech(t). (5)

The convergence is uniform on compact sets.

Proof. When k = 0, the equations (1) become x = yz, y = -zx, z = 0, and the so-
lutions satisfying x(0) = 0, y(O) = 1, z(0) = 1 are (sin(t), cos(t), 1). The solutions
of system (1) are continuous in the parameter k for t in a compact set. The lim-
its follow from the theorem on continuous dependence of solutions on parameters;
[9, pp. 90-96], [6, pp. 25-27], or [7, p. 94]. This proves (4), and (5) follows in a simi-
lar manner. U

2.2. The integrals. Many of the basic facts about the Jacobi functions are a result of
the special properties of equations (1).

730 ?g THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 108

This content downloaded from 186.85.131.131 on Fri, 08 May 2020 01:53:28 UTC
All use subject to https://about.jstor.org/terms
Proposition 2.2. The equations (1) admit the two functions

I = x2 + y2 J = k2x2 + z2

as integrals, i.e., the functions I and J are constant along solutions of (1).

Proof

-I (x(t), y(t)) = 2xx + 2yy = 2x(yz) + 2y(-zx) , 0

dt

so I is constant along solutions. In the same way, dJ/dt _ O.

The existence of these two integrals imposes geometric restrictions on the solu-
tions. In particular, the following corollaries follow at once from the geometry and the
continuation theorem of differential equations.

Corollary 2.1. The functions sn(t, k), cn(t, k), and dn(t, k) are periodic in t. In par-
ticular, sn(t, k), cn(t, k), and dn(t, k) are defined and real analytic for all t E R.

Proof. The values of the integrals on (sn(t, k), cn(t, k), dn(t, k)) are I = J = 1. The
equation I = 1 defines a right circular cylinder centered on the z axis, and J = 1
defines a right elliptic cylinder centered on the y axis. These two cylinders intersect in
two closed curves C and C' on which z > 0 and z < 0, respectively; see Figure 1. The
solution (sn(t, k), cn(t, k), dn(t, k)) starts in C and so remains in C for all t. Since C is
bounded, this solution can be continued for all t E R by the continuation theorem for
differential equations; see [4, p. 3], [6, pp. 16-17], or [7, pp. 12-13]. Since there are
no equilibrium points on C, the solution must traverse all of C and hence is periodic;
we investigate the period in Proposition 2.4.

iz

Figure 1. The intersection of I = 1 and J = 1.

October 2001] JACOBI ELLIPTIC FUNCTIONS 731

This content downloaded from 186.85.131.131 on Fri, 08 May 2020 01:53:28 UTC
All use subject to https://about.jstor.org/terms
Corollary 2.2. Forfixed k, 0 < k < 1, and all t E R the identities

sn2(t, k) + cn2(t, k) 1, i2 sn2(t, k) ? dn2(t, k)-1,

hold, and the inequalities

-1 < sn(t, k) < 1, -1 < cn(t, k) < 1, K < dn(t, k) < 19

are satisfied.

Proof. The identities are restatement of the equations I = J = 1. The first identity
implies the first two inequalities.
The proof of Corollary 2.1 shows that dn(t, k) > 0. This fact and the second identity
imply the last inequality. U

2.3. Symmetries. Several symmetry properties of the system (1) imply some of the
symmetry properties of the Jacobi elliptic functions.

Proposition 2.3. If (x(t), y(t), z(t)) is a solution of (1), then so are (-x(-t), y(-t),
z(-t)), (x 0 (-)-y(-t), z(-t)), and (x (-t), y(-t) ,-z (-t)).

Proof. Let ( (t), ij(t), (t)) = (-x(-t), y(-t), z(-t)); then

(t) = x*(-t) = y(-t)z(-t) = (t0 ),

= -y(-t) = z(-t)x(-t) = -OW0t
(t) = -z(-t) = k2x(-t)y(-t) = -k24 (t) r (t).

Thus, (t (t), r1 (t), 9 (t)) is a solution also. The other cases follow in the
.

Proposition 2.3 says that taking a solution, reversing time, and reflecting through
any coordinate plane gives another solution. Such symmetries are known as time-
reversing symmetries.
The following corollaries illustrate how the uniqueness theorem for differential
equations can be used to derive symmetries of the solutions from symmetries of the
equations.

Corollary 2.3. For fixed k, 0 < k < 1, sn(t, k) is an odd function of t; cn(t, k) and
dn(t, k) are even functions of t.

Proof. By definition, (sn(t, k), cn(t, k), dn(t, k)) is a solution of (1) and hence by
Proposition 2.3 so is (- sn(-t, k), cn(-t, k), dn(-t, k)), but these two solutions both
satisfy the initial condition (0, 1, 1). Thus, the basic uniqueness theorem for ordinary
differential equations ensures that they are identical; see [4, pp. 1-4], [6, pp. 18-24],
or [7, pp. 31-34]. A

Consider the solution (x(t), y(t), z(t)) = (sn(t, k), cn(t, k), dn(t, k)) of equa-
tion (1) and refer to Figure 1. It starts at (0, 1, 1) and moves into the first octant
(x > 0, y > 0, z > 0) where sn(t, k) increases and cn(t, k) and dn(t, k) decrease.
Let K > 0 be the time that cn(t, k) takes to decrease to zero, i.e., cn(K, k) = 0 and
cn(t, k) > 0 for 0 < t < K. From Corollary 2.2 we have

732 (A THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 108

This content downloaded from 186.85.131.131 on Fri, 08 May 2020 01:53:28 UTC
All use subject to https://about.jstor.org/terms
 sn(O, k) = 0, sn(K, k) = 1, 0 < sn(t, k) < 1 for 0 < t < K,

cn(0,k) = 1, cn(K,k) =0, 0 < cn(t,k) < 1 for0 < t < K, (6)

dn(0, k) = 1, dn(K, k)= K, K< dn(t, k) < 1 for 0 < t < K.

Proposition 2.4. As functions of t, sn(t, k) and dn(t, k) are even about K and cn(t, k)
is odd about K, i.e., forfixed k, 0 < k < 1, and all t E R

sn(K + t, k)-sn(K -t, k),

cn(K + t, k) -cn(K -t, k), (7)

dn(K + t, k) _dn(K -t, k).

Thus, sn(t, k) and cn(t, k) are 4K periodic in t and dn(t, k) is 2K periodic in t.

Proof. Since equation (1) is time-independent, (sn(K + t, k), cn(K + t, k),

dn(K + t, k)) is a solution of (1), and by Proposition 2.3, so is (sn(K - t, k),
- cn(K - t, k), dn(K - t, k)), but these two solutions both satisfy the initial con-
dition (1, 0, K). Thus, the basic uniqueness theorem for ordinary differential equations
ensures that they are identical and therefore (7) follows; see [4, pp. 1-4], [6, pp. 18-
24], or [7, pp. 31-37].
By (7) and Corollary 2.3, sn(t + K, k) _ sn(-t + K, k) -sn(t - K, k) or
sn(t + 2K, k) _- sn(t, k), from which it follows that sn(t, k) is 4K periodic in t.
The other cases are similar. U

Proposition 2.4 says that sn(t, k) and cn(t, k) have the same symmetries with re-
spect to K as sin t and cos t have with respect to 7r/2.

2.4. Other differential equation definitions. The Jacobi elliptic functions satisfy
many other important functional equations. Here are some of the classical differen-
tial equations that are important in the theory.

Proposition 2.5. The functions x = sn(t, k), y = cn(t, k), and z = dn(t, k) satisfy
thefirst order equations

x2 = (1 -x2)(1 -k2x2), x(0) = 0, x(0) = 1,

y2 = ( y2) (K2 + k2y2), y(O) = 1, (0) = 0? (8)

2 =(1 -Z2)(Z2 K 2), z(0) = 1, Z(0) = 0.

Proof. From (1) and Corollary 2.2 we have

=z ? y* = -xz2 - k2xy2 = -x(I - k2X2) - k2X(I - X2) = -(1 + k 2)x + 2k x2

(9)
Thus, sn(t, k) satisfies (9). But (9) has an integral L = x2 + (1 + k2)x2 - k2x4, which
is equal to 1 on sn(t, k) since sn(0, k) = 0, sn(0, k) = 1. Rearranging the equation
L = 1 shows that sn(t, k) satisfies the first equation in (8). There is a similar argument
for the cn(t, k) and dn(t, k) equations. U

Corollary 2.4. sn(t, k) is concave down for 0 < t < 2K and concave up for -2K <
t <0.

October 2001] JACOBI ELLIPTIC FUNCTIONS 733

This content downloaded from 186.85.131.131 on Fri, 08 May 2020 01:53:28 UTC
All use subject to https://about.jstor.org/terms
Proof. Since sn(t, k) is positive for 0 < t < 2K and 0 < k < 1, the function
(1 + k2)x - 2k2x3 evaluated at x = sn(t, k) is positive when 0 < t < 2K. Hence,
(9) ensures that x is negative when 0 < t < 2K. U

At this point we have all the basic qualitative features

sketch their graphs. Software packages such as Maple
Jacobi functions and nice graphical routines; see Figure 2.

0
K 2K 3K 4K

cn

Figure 2. The graphs of sn(t, k), cn(t, k) and dn(t, k) for k = .95.

2.5. The integral definition. The equations in (8) are "solvable up to quadrature".
For 0 < t < K we have x > 0 and hence

dx - (1-x2)(1-k2x2).
dt

This is a separable equation, so

fsn(t,k) dx

I (1-x2)(1-k2x2)
Jo -t.X) I X2 (10)t(10
This classical definition of sn(t, k) is found in
defining sin t by

jsint dx
Jo X

The integral in (10) converges as the upper limit te

K- dx (1
K (1-x2)(-k2x2) 1)

In this integral make the sub

7zr/2 du
K= f 1 (12)
Jo +/1-k2 si2 u

Since the integral (12) doesn't have a singularity, we have:

734 ?A THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 108

This content downloaded from 186.85.131.131 on Fri, 08 May 2020 01:53:28 UTC
All use subject to https://about.jstor.org/terms
Proposition 2.6. The integral in (12) defines K as an even, positive real analytic
function of k for -1 < k < 1. The function K is increasing on 0 < k < 1, since
dK/dk > OforO <k < 1. Moreover, K(O) =.Tr/2 and K - +0oc ask -> 1-.

There are similar integral formulas for cn and dn.

3. APPLICATIONS. The references [2], [3], [5], and [11] give many applications
of Jacobi elliptic functions in mathematics, physics, and engineering. Here are some
applications that are appropriate for a course in differential equations.

3.1. The Pendulum Equation. The equation for the simple pendulum with all con-
stants set to 1 is

+[sin0=0. (13)

It admits the energy integ

21 =1 02 + (1- COS 0) = 102 ? 2sin2 2. (14)

2 2 2

Set y = sin0/2 and 2y = 1 -y2in (14) to get

y2 = (I - y2)( _ y2). (15)

Case A: I = 0 corresponds to the downward equilibrium position; y (t) 0(t) _ 0

is the only solution.

Case B: 0 < I < 1 corresponds to oscillatory solutions where the pendulum

swings back and forth. Set I = k2 and note that y = k sn(t - , k) satis-
fies (15) for any constant r, so the solution of the pendulum equation in
this case is 0(t) = 2 arcsin k sn(t - -, k), where I = k2.
Case C: I = 1 corresponds to the upward equilibrium position and to the solu-
tions that are asymptotic to the upward equilibrium position. 0 (t) -r is
the upward equilibrium solution. For the asymptotic solutions note that
y = -4 tanh(t - -r) satisfies (15), so the solution of the pendulum equation
is 0 (t) = -42 arcsin tanh(t -r).
Case D: I > 1 corresponds to circulating orbits, where the pendulum's energy is
high enough to carry the pendulum over the top. Set I = k-2 and note that
y = sn(t/k, k) satisfies (15), so the solution of the pendulum equation is
0(t) = ?2 arcsin sn((t -r) /k, k). In this case, 0 increases (or decreases)
forever, so one must switch branches of the arcsine function so that 0
increases (or decreases) continuously.

3.2. Elliptic integrals. A careful reading of the chapter on integration techniques in

a standard calculus book shows that if R is a rational function of X and x and if X
is linear or quadratic in x, then any integral of the form

J R(x, X)dx

October 2001] JACOBI ELLIPTIC FUNCTIONS 735

This content downloaded from 186.85.131.131 on Fri, 08 May 2020 01:53:28 UTC
All use subject to https://about.jstor.org/terms
can be integrated using elementary functions. By completing the square, using trigono-
metric substitutions, partial fractions, etc., one expresses the integral in terms of the
trigonometric functions, logarithms, and exponentials.
The theory of elliptic integrals investigates integrals of the same form, where X is
now a cubic or quartic in x. There are reduction methods to reduce any integral of
this form to either elementary integrals or to what are known as elliptic integrals of
the first, second, and third kinds. Jacobi elliptic functions can be used to evaluate any
integral of the first kind, i.e., any integral of the form

a dx (16)

A complete discussion with all the degenerate cases can be found in [2, pp. 4-15],
[3, pp. 86-98], and [5, pp. 31-42]. Since the development is lengthy, we give only a
brief hint of the theory.
When X is a quartic, one can find constants p and q such that the change of variables
x = (p + qy)/(l + y), dx = (q - p)dy/(I + y)2 reduces (16) to

I=(qp) dy

where Y is a quadratic in y2. When X is a cubic and a is a real root of X, the change
of variables x = y2 + a, dx = 2ydy effects a similar reduction.
In both cases the problem is reduced to integrating an integral of the first kind (16)
where X is a quadratic in x2 that can be factored. Then elementary tricks reduce the
integral to a standard form such as (10). Here is an example. Let a and b be constants
with 0 < a < b. Then

fx du -xa dv
Jo ^/(a2 -u2)(b2 -u2)

1I xla dv
1 Jx/a d v ~ ~(k = alb) (17)
b1 /(1-v2)(1-k2v2) (=/)(7

I b sn- (x/a, a/b)

b

3.3. Systems with quadratic or cubic forces. If f (x) is either a quadratic or cubic
polynomial in x, then any differential equation of the form

x~+f(x)=0 (18)
is solvable in terms of the Jacobi elliptic functions.
Equation (18) admits the integral

I = x2 + X(x), where X(x) = 2 f (x) dx (19)

as a constant of motion, so the phase portrait can be obtain

of I. By setting I = c, a constant, (19) becomes a separable equation. Thus,

736 ( THE MATHEMATICAL ASSOCIATION OF AMERICA [Monthly 108

This content downloaded from 186.85.131.131 on Fri, 08 May 2020 01:53:28 UTC
All use subject to https://about.jstor.org/terms
 J dx _
[ dX ._ = t (20)
c - X(x)

and since f (x) is either a quadratic or a cubic, c - X (x) is either a cubic or a quartic.
Therefore (18) can be integrated by the methods outlined in the previous section.
For example, consider the undamped Duffing equation

3x +x -2x3 = 0

with integral I = X2 + X2 - X4. Let I = c, 0 < c < 1/4, and seek a solution satisfying
x (0) = 0, x > 0 so that (20) becomes

fx dx

Jo X4-X2+C

Since 0 < c < 1/4, the polynomial factors to give 4 - x2 + C = (a2 - X2)(b2 - X2)
with 0 < a < b. Thus, (17) shows that x(t) = a sn(bt, a/b).

REFERENCES

1. R. P. Agnew, Views and approximations on differential equations, Amer. Math. Monthly 60 (1953) 1-6.
2. A. L. Baker, Elliptic Functions, John Wiley and Sons, New York, 1890.
3. F. Bowman, Introduction to Elliptic Functions with Applications, Dover Publ., New York, 1961.
4. C. Chicone, Ordinary Differential Equations with Applications, Springer, New York, 1999.
5. A. G. Greenhill, The Applications of Elliptic Functions, Macmillan and Co., London, 1892.
6. J. K. Hale, Ordinary Differential Equations, Wiley-Interscience, New York, 1969.
7. P. Hartman, Ordinary Differential Equations, Wiley, New York, 1964.
8. G. H. Hardy, A Course of Pure Mathematics, Cambridge University Press, Cambridge, 1908.
9. E. Hille, Lectures on Ordinary Differential Equations, Addison-Wesley Publ. Co., Reading MA, 1969.
10. K. R. Meyer and G. R. Hall, Introduction to Hamiltonian Dynamical Systems and the N-body Problem,
Springer-Verlag, New York, 1992.
11. E. T. Whittaker and G. N. Watson, A Course of Modern Analysis, Cambridge University Press, Cam-
bridge, 1902.

KEN MEYER was born and bred in Cincinnati, obtained his Ph.D. from the University of Cincinnati, and
spent the majority of his career as a professor at the University of Cincinnati. One of his hobbies is collecting
fountain pens of Cincinnati origin: John Hollands, Picks, Stars, and Weidlichs. However, he did spend five
snowy years getting a degree in engineering physics from Cornell University and another five frigid years as
an Associate Professor at the University of Minnesota. But his best years were at RIAS and Brown University
learning differential equations at the feet of Joe LaSalle, Jack Hale, and Solomon Lefschetz.
Department of Mathematical Sciences, University of Cincinnati, Cincinnati, Ohio 45221-0025
[email protected]

This content downloaded from 186.85.131.131 on Fri, 08 May 2020 01:53:28 UTC
All use subject to https://about.jstor.org/terms