2

Jacobian elliptic functions of a complex variable

2.1 Introduction: the extension problem

In this chapter we define the Jacobian elliptic functions and establish their basic
properties; for the convenience of the reader a summary of those properties is
included at the end of the chapter.
In the last section of Chapter 1 we sought to make clear the possibility that the
Jacobian elliptic functions (arising out of our study of the simple pendulum and
as defined in (1.22) of that chapter) are much more than routine generalizations
of the circular functions. But to explore their suspected richness it is necessary
to move off the real line into the complex plane.
Our treatment is based on Abel’s original insight (1881), and is in three steps:
(i) extension of the definitions, (1.28), to the imaginary axis; (ii) derivation of
the addition formulae for the functions sn, cn and dn; (iii) formal extension to
the complex plane by means of the addition formulae. The discussion in Section
2.6 then shows that the extended functions are doubly periodic.
What appears to be unusual (perhaps new) in our treatment is: step (iv),
verification that the functions so extended are analytic except for poles (that is,
meromorphic) in the finite C plane. (Of course, there are other ways of obtaining
the functions sn, cn and dn as meromorphic functions of a complex variable,
but none of them is easy. As Whittaker & Watson (1927), p. 492, write: ‘unless
the theory of the theta functions is assumed, it is exceedingly difficult to show
that the integral formula

y
u= (1 − t 2 )−1/2 (1 − k 2 t 2 )−1/2 dt
0

defines y as a function of u, which is analytic except for simple poles. See

Hancock (1958).)

25

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 26 2 Jacobian elliptic functions of a complex variable

2.2 The half-angle formulae

The functions defined by Abel admit of an obvious extension to the imaginary
axis. However, we have chosen to work with the functions of Jacobi, one of
whose methods of extending them to the imaginary axis was to start with a
function, φ(u), like that in the Maclaurin expansion found in Exercises 1.9,
(1); so that sn(u) = φ(u), where φ(u) is a real, odd function. But then it is
plausible (by applying the formula twice) that sn(iu) = iφ(u) and one can start
with that as a definition of sn(iu). Another method is to use the integrals in
Exercise 1.5.1 and to replace the upper limit y by iy (see Bowman (1961),
Chapter IV for an outline, but note the quotation above from Whittaker and
Watson – the method relies on Jacobi’s ‘imaginary transformation’, which is
treated later on and which is described in detail in Whittaker & Watson (1927),
Chapter XXII). Copson (1935) treats the problem using the Weierstrass theory
of σ -functions, but that is essentially the theta functions approach. Another
interesting and unusual approach, based on properties of period-lattices, is to
be found in Neville (1944).
None of those methods is easy to justify (and we prefer to defer our treat-
ment of the theta functions to a later chapter); so we shall use the ‘half-angle
substitution’ method, introduced by Weierstrass to integrate rational functions
of sine and cosine.
To that end, set
ψ
t = tan (−π < ψ < π).
2
Then
1 − t2 2t dψ 2
cos ψ = , sin ψ = , = .
1 + t2 1 + t2 dt 1 + t2
It follows that
Q(t, k)
1 − k 2 sin2 ψ = ,
(1 + t 2 )2
where

Q(t, k) = 1 + 2(1 − 2k 2 )t 2 + t 4 . (2.1)

Note that Q(t, k) = [t 2 + (1 − 2k 2 )]2 + 4k 2 (1 − k 2 ), whence Q(t, k) > 0, for

−∞ < t < +∞, provided that 0 ≤ k < 1. If k = 1, then the result is true,
provided that −1 < t < 1. Equations (1.20) and (1.19) of Chapter 1 become

1
dt
K = K (k) = 2 , (2.2)
0 (Q(t, k))1/2

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.2 The half-angle formulae 27

t
dt
x =2 , (−2K < x < 2K ), (2.3)
0 (Q(t, k))1/2
respectively, where −∞ < t < ∞ when 0 ≤ k < 1, and where −1 < t < 1
when k = 1.
Note that
Q(t, 0) = (1 + t 2 )2 , Q(t, 1) = (1 − t 2 )2

and
 
1
Q t, √ = 1 + t 4.
2
On returning to the definitions in (1.28) in the case when u is real, we have,
for −2K < x < 2K ,
2t
sn(x, k) = sin ψ = ,
1 + t2
1 − t2
cn(x, k) = cos ψ = ,
1 + t2
(Q(t, k))1/2
dn(x, k) = (1 − k 2 sin2 ψ)1/2 = , (2.4)
1 + t2
where t is given as a function of x by inverting (2.3).
Note that the trigonometric functions are absent from our final formulae
(except implicitly, in the half-angle formulae). Whether or not that is a good
thing depends, to some extent, on one’s familiarity with the rigorous develop-
ment of the circular functions in analysis (see, for example, Apostol (1969) or,
in the spirit of our present treatment, the outline in Bowman (1961). In fact, the
elliptic functions may be introduced in other ways, which may also be used to
introduce the circular functions; see Weil (1975).
In two earlier papers, one of us (Eberlein (1954) and (1966)) has shown
how the half-angle approach to the circular functions offers an alternative to the
usual treatments. We now sketch the extension of those ideas to the functions
of Jacobi.
We start with (2.2) as the definition of K, taking K to be ∞ in the case when
k = 1. Set t = s −1 and then we see that

∞
1
dt ds
1/2
= , (0 ≤ k < 1).
1 (Q(t, k)) 0 (Q(s, k))1/2
Then (2.3) defines x as an odd, increasing function of t, with positive derivative
dx 2
= . (2.5)
dt (Q(t, k))1/2

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 28 2 Jacobian elliptic functions of a complex variable

We may then invert (2.3) to obtain t as an odd, increasing differentiable

function of x, (−2K < x < 2K ). The formulae (2.4) then define sn, cn and dn
for −2K < x < 2K . Note that sn is an odd function, and cn and dn are both
even functions of x.
When k = 1, K = ∞ and the definition is complete.
When k < 1, we follow the discussion in Chapter 1, Section 1.4, and use

2t
sn(±2K ) = lim sn(x) = lim = 0,
x→±2K t→±∞ 1 + t2
1 − t2
cn(±2K ) = lim cn(x) = lim = −1,
x→±2K t→±∞ 1 + t 2
1
(Q(t, k)) 2
dn(±2K ) = lim dn(x) = lim = 1. (2.6)
x→±2K t→±∞ (1 + t 2 )

Those formulae, originally defined for −2K < x < 2K , may be extended
to all real x by periodicity, that is by the formulae:

sn(x + 4K ) = sn x,
cn(x + 4K ) = cn x,
dn(x + 4K ) = dn x. (2.7)

As we shall see in (2.23), below, the function dn actually has the smaller period,
2K.
The formulae for differentiation follow from the chain rule in the range
−2K < x < 2K ; thus
 
d 2t
d dt 1 + t2 1 − t 2 (Q(t, k))1/2
sn (x) = = = cn (x) · dn (x). (2.8)
dx dx 1 + t2 1 + t2
dt

From the periodicity condition (2.7), it follows that (2.8) holds for all x other
than the odd multiples of 2K , k < 1. The validity at those points follows from
the following corollary of the mean value theorem.

Proposition 2.1 Let f (x) be continuous at x = a and differentiable in a neigh-

bourhood of a, with a deleted. Suppose that A = lim f  (x) exists. Then f  (a)
x→a
exists and equals A.
The treatment of cn  and dn  is similar.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.3 Extension to the imaginary axis 29

Exercises 2.2
2.2.1 Prove the corollary to the mean value theorem quoted in Proposition 2.1.
2.2.2 Obtain the formulae for cn  x and dn  x, using the method in (2.8).

2.3 Extension to the imaginary axis

We make the following changes of variable, which at this stage are purely formal
(the justification for them will be given later). We write

x = iy, t = is
√
and observe that Q(is, k) = Q(s, k  ), where k  = 1 − k 2 is the complemen-
tary modulus. Thus (2.3) becomes (the following heuristic argument needs
justification in due course)

s
ds
y=2  1/2
, (2.9)
0 (Q(s, k ))

and, as suggested by that and by (2.2), we write

1
  ds
K = K (k ) = 2 . (2.10)
0 (Q(s, k  ))1/2
We are led accordingly to the following definitions. If −2K  < y < 2K  ,
define
2t 2s 2s/(1 + s 2 ) sn(y, k  )
sn(iy, k) = = i = i = i ,
1 + t2 1 − s2 (1 − s 2 )/(1 + s 2 ) cn(y, k  )
 −1
1 − t2 1 + s2 1 − s2 1
cn(iy, k) = = = = ,
1+t 2 1−s 2 1+s 2 cn(y, k  )
(Q(t, k))1/2 (Q(s, k  ))1/2
dn(iy, k) = =
1 + t2 1 − s2
(Q(s, k  ))1/2 /(1 + s 2 ) dn(y, k  )
= = , (2.11)
(1 − s 2 )/(1 + s 2 ) cn(y, k  )
provided that s = ±1(y = ±K  ).
The expression of the functions sn(iy, k) etc. in terms of functions in which
the argument is real is known as ‘Jacobi’s imaginary transformation’.
Now the last expression in each of the three formulae in (2.11) has a mean-
ing for all y ≡ K  (mod 2K  ). So we can now employ them to define sn, cn
and dn as functions defined on the imaginary axis and continuous, except for
poles at the odd multiples of iK  . Moreover, on the imaginary axis cn (z) and

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 30 2 Jacobian elliptic functions of a complex variable

dn (z) (as we may now denote them) have period 4iK  , whilst sn (z) has period
2iK  . That last result confirms the conjecture made at the end of Chapter 1,
Section 1.8.
Now we shall show that the differentiation formulae, (1.32), remain valid
for the functions in (2.11). When t = is, (s = ±1) and x = iy, (−2K  < y <
2K  ), y = ±K  , we have
d 2t d 2t
= dt 1 + t = dt 1 + t
d sn(iy, k) 2 2

d(iy) dx dy
dt ds
   
d 2t d 2t
dt 1 + t 2 dt 1 + 2t 2
= =
2/Q(s, k  )1/2 2/Q(t, k)1/2
1 − t 2 Q(t, k)1/2
= ·
1 + t2 1 + t2
= cn(iy, k) dn(iy, k).

The validity of that formula for all y ≡ K  (mod 2K  ) then follows, using the
periodicity and the corollary to the mean value theorem, Proposition 2.1. A
similar argument shows that the differentiation formulae for cn and dn also
remain valid in the pure imaginary case, when x = iy.

Exercises 2.3
2.3.1 Obtain the results for differentiating cn and dn.

2.4 The addition formulae

Euler and Legendre called the elliptic integrals ‘elliptic functions’ (rather
like calling arcsin x = sin−1 x a ‘circular function’), and their addition for-
mulae were derived by Euler in a dazzling tour de force. When k = 0 (as we
have seen), that amounts to studying addition formulae for arcsin or arccos,
instead of for sin and cos. (The treatment of the addition formula for arctan
in Hardy (1944), pp. 435–438, is reminiscent of Euler’s method.) Once one
inverts the integrals, in the manner first proposed by Gauss and Abel, matters
simplify.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.4 The addition formulae 31

There are many proofs of the addition formulae (see, for example,
Bowman (1961), p. 12; Copson (1935), pp. 387–389; Lawden (1989), p. 33;
and Whittaker & Watson (1927), pp. 494–498). We shall follow Abel’s method
(see Theorem 2.1, below), but first we shall try to make the forms of the addition
formulae plausible in the general case, 0 ≤ k ≤ 1, by interpolation between the
elementary limiting cases, k = 0 (the circular functions) and k = 1 (the hyper-
bolic functions). In one sense, the latter afford a more reliable guide, since the
imaginary poles are readily obtained.
We start on the real axis and we recall (see (1.29) and (1.30)) that
sn(u, 0) = sin u, sn(u, 1) = tanh u,
cn(u, 0) = cos u cn(u, 1) = sech u,
dn(u, 0) = 1, dn(u, 1) = sech u, (2.12)
for −∞ < u < ∞.
Now
cn(u + v, 0) = cos(u + v) = cos u cos v − sin u sin v
= cn(u, 0)cn(v, 0) − sn(u, 0)sn(v, 0) (2.13)
and
1
cn(u + v, 1) = sech(u + v) =
cosh(u + v)
1
= .
cosh u cosh v + sinh u sinh v
To obtain only those functions appearing in (2.12), we multiply numerator and
denominator in the last expression by sech u sech v to find
sech u · sech v
sech(u + v) = . (2.14)
1 + tanh u tanh v
Formula (2.14) is still unlike (2.13), but we can rewrite the latter as
cos u cos v − sin u sin v = cos u cos v(1 − tan u tan v).
That suggests that we multiply the numerator and denominator in (2.14) by
(1 − tanh u tanh v) and then obtain
sech u sech v − tanh u tanh v sech u sech v
sech(u + v) = . (2.15)
1 − tanh2 u tanh2 v
Now we appeal to symmetry in u and v to write
f (u, 1) f (v, 1) − sn(u, 1)sn(v, 1)g(u, 1)g(v, 1)
cn(u + v, 1) = ,
1 − sn 2 (u, 1)sn 2 (v, 1)

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 32 2 Jacobian elliptic functions of a complex variable

where f and g denote either cn or dn. On comparing that last equation with
(2.13) we see that we should take f = cn and g = dn and we may then write
cn(u + v, 0) = cn(u, 0)cn(v, 0) − sn(u, 0)sn(v, 0)dn(u, 0)dn(v, 0)
cn(u, 1)cn(v, 1) − sn(u, 1)sn(v, 1)dn(u, 1)dn(v, 1)
cn(u + v, 1) =
1 − sn 2 (u, 1)sn 2 (v, 1)
(2.16)
The equations in (2.16) lead us to the plausible conjecture that
cn(u) cn(v) − sn(u) sn(v) dn(u) dn(v)
cn(u + v) = , (2.17)
(u, v)
where:
(a) (u, v) is symmetric in u and v;
(b) (u, v) = 1 when k = 0;
(c) (u, v) = 1 − sn 2 u sn 2 v when k = 1.
To obtain a further condition on , put v = −u in (2.17). Then 1 =
cn(0) = (cn 2 u + sn 2 u dn 2 u)/(u, −u). That is,
(u, −u) = cn 2 u + sn 2 u dn 2 u = 1 − sn 2 u + sn 2 u(1 − k 2 sn 2 u)
and so finally
(d) (u, −u) = 1 − k 2 sn 4 u.
Clearly, one choice (and perhaps the simplest) for (u, v) that satisfies (a), (b),
(c) and (d) is
(u, v) = 1 − k 2 sn 2 u sn 2 v. (2.18)
So our final, informed conjecture is
cn(u) cn(v) − sn(u) sn(v) dn(u) dn(v)
cn(u + v) = = C(u, v), (2.19)
1 − k 2 sn 2 u sn 2 v
say.
For brevity, put:
s1 = sn (u), s2 = sn (v),
c1 = cn (u), c2 = cn (v),
d1 = dn (u), d2 = dn (v),
 = 1 − k 2 s12 s22 .
We can now guess the addition formulae for sn and dn from the identities
cn 2 + sn 2 = 1 and dn 2 + k 2 sn 2 = 1 (see (1.31)).
We start with the conjecture in (2.19):
cn(u + v) = C(u, v) = (c1 c2 − s1 s2 d1 d2 )/.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.4 The addition formulae 33

Then

sn 2 (u + v) = 1 − cn 2 (u + v) = 1 − C 2 (u, v)
 2  2  2
= 1 − k 2 s12 s22 − c1 s2 − s1 s2 d1 d2  .

We use the result in Exercise 2.4.1 below to obtain (again this is at present a
conjecture, a plausible inference from the cases k = 0, k = 1):
   
sn 2 (u + v) = c12 + s12 d22 c22 + s22 d12 − (c1 c2 − s1 s2 d1 d2 )2 2
 
= s12 c22 d22 + 2s1 s2 c1 c2 d1 d2 + s22 c12 d12 2
= (s1 c2 d2 + s2 c1 d1 )2 /2 .

So we have
sn(u + v) = ±(s1 c2 d2 + s2 c1 d1 )/.

We guess the sign by taking v = 0, and we see that we have to choose the ‘+’.
So finally we have the conjecture
sn(u) sn(v) dn(v) + sn(v) cn(u) dn(u)
sn(u + v) = = S(u, v). (2.20)
(u, v)
Having made the conjectures (2.19), (2.20), (2.21) (see below, Exercises
2.4.1 and 2.4.2), (in the spirit, we like to think of Polya (1954),) we shall
formulate them as a theorem, which will be the central result of this chapter.
Our proof follows that of Abel (see Bowman (1961), p.12).

Theorem 2.1 Let u, v ∈ (−∞, +∞). Then:

(a) sn(u+v)=S(u, v) ≡ [sn(u) cn(v) dn(v) + sn(v) cn(u) dn(u)]/(u, v),

(b) cn(u+v)=C(u, v) ≡ [cn(u) cn(v) − sn(u) sn(v) dn(u) dn(v)]/(u, v),
(c) dn(u+v)=D(u, v) ≡ [dn(u) dn(v)−k 2 sn(u) sn(v) cn(u) cn(v)]/(u, v),

where
(u, v) = 1 − k 2 sn 2 u sn 2 v.

Proof We shall use the abbreviated notation introduced above:

S = (s1 c2 d2 + s2 c1 d1 )/.

Partial differentiation with respect to u yields

∂S     
2 =  c1 d1 c2 d2 − s1 s2 d12 + k 2 c12 + (s1 c2 d2 + s2 c1 d1 ) 2k 2 s1 c1 d1 s22
∂u
     
= c1 d1 c2 d2  + 2k 2 s12 s22 − s1 s2  d12 + k 2 c12 − 2k 2 s22 c12 d12
      
= c1 d1 c2 d2  + 2k 2 s12 s22 − s1 s2 d12  − k 2 s22 c12 + k 2 c12  − s22 d12

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 34 2 Jacobian elliptic functions of a complex variable

   
= c1 d1 c2 d2 1 + k 2 s12 s22 − s1 s2 d12 d22 + k 2 c12 c22
 
= (c1 c2 − s1 s2 d1 d2 ) d1 d2 − k 2 s1 s2 c1 c2 .

Hence ∂∂uS = C D, which is symmetric in u, v. Since S is symmetric in u, v we

conclude that
∂S ∂S
= CD = .
∂v ∂u
Let us now introduce new variables z = u + v, w = u − v to transform that
∂S
last equation into ∂w = 0, whence S = f (z) = f (u + v). Take v = 0. Then
f (u) = S(u, 0) = sn(u) and so S = sn(u + v).

It is possible to use a similar argument to obtain (b) and (c) (left as an

exercise!), but it is simpler to argue as follows. By Exercise 2.4.3 (see below),
we have
cn 2 (u + v) = 1 − sn 2 (u + v) = 1 − S 2 (u, v) = C 2 (u, v).

Hence cn(u + v) = ±C(u, v). Now

∂S ∂
CD = = sn(u + v) = cn(u + v)dn(u + v).
∂u ∂u
We conclude that dn(u + v) = ±D(u, v) whenever cn(u + v) = 0. Now
dn > 0 and D > 0 and so we must have dn(u + v) = D(u, v) if u + v ≡
K (mod 2K ). On taking limits, we obtain dn(u + v) = D(u, v) for all u, v.
But then C D = cn(u + v) · D, whence C(u, v) = cn(u + v).

Remark It is tempting to try to determine the sign in cn(u + v) = ±C(u, v)

by putting v = 0, whence cn(u + 0) = C(u, 0). It is reasonable to guess that
one should always take the positive sign, but the justification for that argument
requires that the function cn is an analytic function (except for poles) and that
will occupy us in the next section.

To see something of the difficulty, consider the following example. Let

f (x) = exp(−1/x 2 ), x = 0, and f (x) = 0 when x = 0. Put g(x) = f (x) when
x ≤ 0 and g(x) = − f (x) when x > 0. Then f, g ∈ C ∞ (−∞, ∞), f 2 =
g 2 , f (x) = g(x) for x ≤ 0, but f (x) = g(x) when x > 0.

Exercises 2.4
2.4.1 Show that

(u, v) = cn 2 u + sn 2 u dn 2 v = cn 2 v + sn 2 v dn 2 u
= dn 2 u + k 2 sn 2 u cn 2 v = dn 2 v + k 2 sn 2 v cn 2 u.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.5 Extension to the complex plane 35

2.4.2 From the identity dn 2 u + k 2 sn 2 u = 1, and by using a similar argument

to that used for sn(u + v), derive the conjecture
dn(u) dn(v) − k 2 sn(u) sn(v) cn(u) cn(v)
dn(u + v) = = D(u, v). (2.21)
(u, v)
Show that (given the foregoing)

C 2 + S2 = 1

and that
D 2 + k 2 S 2 = 1.

2.4.3 Show that (on the basis of the conjectures)

 1/2  1/2  1/2  1/2 
D(u, v) = 1 − k 2 s12 1 − k 2 s22 ± k 2 s1 s2 1 − s12 1 − s22 

and that |s j | ≤ 1, 0 ≤ k < 1, and |s j | < 1 when k = 1 ( j = 1, 2). Con-

clude that D(u, v) > 0.
2.4.4 Obtain formulae for sn(2u), cn(2u), dn(2u) in terms of sn (u), cn (u),
dn (u) and also sn 2 u, cn 2 u, dn 2 u in terms of sn(2u), cn(2u), dn(2u).
2.4.5 Show that
√
K 1 K k K √
sn = √ , cn = √ , dn = k.
2 (1 + k  ) 2 (1 + k  ) 2
2.4.6 Prove that
cn(u + v) − dn(u + v) cn(u)dn(v) − dn(u)cn(v)
= .
sn(u + v) sn(u) − sn(v)
2.4.7 Use the previous question to show that:
(i) if u 1 + u 2 + u 3 + u 4 = 2K , then

(cn(u 1 )dn(u 2 ) − dn(u 1 )cn(u 2 ))(cn(u 3 )dn(u 4 ) − dn(u 3 )cn(u 4 ))

= k 2 (sn(u 1 ) − sn(u 2 ))(sn(u 3 ) − sn(u 4 )),

(ii) if u 1 + u 2 + u 3 + u 4 = 0, then

(cn(u 1 )dn(u 2 ) − dn(u 1 )cn(u 2 ))(sn(u 3 ) − sn(u 4 ))

+ (cn(u 3 )dn(u 4 ) − dn(u 3 )cn(u 4 ))(sn(u 1 ) − sn(u 2 )) = 0.

2.5 Extension to the complex plane

In the course of the proof of Theorem 2.1, we proved that ∂∂uS = C D =
∂S
∂v
. Now ∂C
∂u
∂
= ∂u cn(u + v) = −sn(u + v)dn(u + v) = −S D, whence, by

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 36 2 Jacobian elliptic functions of a complex variable

symmetry, ∂C∂v
= −S D. In a similar manner, we obtain ∂D
∂u
= −k 2 SC = ∂D
∂v
.
It is convenient to summarize those results as
∂S ∂S
= CD = ,
∂u ∂v
∂C ∂D
= −S D = ,
∂u ∂v
∂D ∂D
= −k 2 SC = , (−∞ < u, v < +∞). (2.22)
∂u ∂v
The first of the equations in (2.22) was derived by differentiating explicitly,
and it is clear that the remaining ones can be derived in a similar manner. Since
the differentiation formulae for sn, cn and dn remain valid on the imaginary
axis, it follows that Equations (2.22) remain valid when u and v are pure imagi-
nary, provided only that none of u, v and u + v is of the form (2n + 1)iK  (n ∈ Z)
– to avoid the poles. But then the argument used in the proof of Theorem 2.1 (a)
goes through again and so the addition formulae, (2.19), (2.20) and (2.21), are
valid when u, v are pure imaginary, subject to the obvious restrictions, as above.
Finally, it is clear that the terms in (2.22) are defined and the results remain valid
when u is real and v is pure imaginary and not equal to (2n + 1)iK  , n ∈ Z. So,
following Abel, we are led to the possibility of defining sn(x + iy), cn(x + iy)
and dn(x + iy) as follows:

sn(x + iy) = S(x, iy) (−∞ < x, y < +∞),

cn(x + iy) = C(x, iy),
dn(x + iy) = D(x, iy) (y = (2n + 1)K  , n ∈ Z). (2.23)

When x = 0 or y = 0, (2.23) reduces to our earlier definitions. In general,

consider the possible definition

sn(x + iy) = S(x, iy)

= [sn(x) cn(iy) dn(iy) + sn(iy) cn (x) dn (x)]/(1 − k 2 sn 2 x sn 2 iy)

sn(x, k)dn(y, k  ) sn(y, k  )cn(x, k)dn(x, k)
+ i
cn 2 (y, k  ) cn(y, k  )
=
1 + k sn (x, k)sn (y, k )/cn 2 (y, k  ).
2 2 2 

sn(x, k)dn(y, k  ) + i sn(y, k  )cn(y, k  )cn(x, k)dn(x, k)

= .
cn 2 (y, k  ) + k 2 sn 2 (x, k)sn 2 (y, k  )

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.5 Extension to the complex plane 37

When 0 < k < 1, that last expression is defined except when both cn(y, k  ) = 0
and sn(x, k) = 0. It is clear from Equations (2.4), (2.6) and (2.7) that that is
equivalent to x = 2m K and y = (2n + 1)K  . Consider then the open, connected
set

= C − {x + iy ∈ C|x = 2m K , y = (2n + 1)K  }. (2.24)

Our discussion suggests that we should extend our definition of sn(x + iy) =
S(x, iy) to  in the following way. Let 0 < k < 1 and x + iy ∈ .
Define

sn(x + iy) = S(x, iy)

sn(x, k)dn(y, k  ) + i sn(y, k  )cn(y, k  )cn(x, k)dn(x, k)
= .
cn 2 (y, k  ) + k 2 sn 2 (x, k)sn 2 (y, k  )
(2.25)

Clearly f (x, y) = S(x, iy) ∈ C 1 (). By similar arguments we extend the def-
inition of cn(x + iy) and dn(x + iy) to  as

cn(x + iy) = C(x, iy)

cn(x, k)cn(y, k  ) − i sn(x, k)dn(x, k)sn(y, k  )dn(y, k  )
= ,
cn 2 (y, k  ) + k 2 sn 2 (x, k)sn 2 (y, k  )
(2.26)
dn(x + iy) = D(x, iy)
dn(x, k)cn(y, k  )dn(y, k  ) − ik 2 sn(x, k)cn(x, k)sn(y, k  )
= .
cn 2 (y, k  ) + k 2 sn 2 (x, k)sn 2 (y, k  )
(2.27)

In (2.22) we saw that, when y = (2n + 1)K  ,

∂ f (x, y) 1 ∂
= CD = f (x, y).
∂x i ∂y
Since all our functions are in C 1 (), we may appeal to continuity to
obtain
∂ 1 ∂
f (x, y) = C D = f (x, y) (2.28)
∂x i ∂y
for z = x + iy ∈ , where f (x, y) = S(x, iy) = sn(x + iy). But (2.28) is a
form of the Cauchy–Riemann equations (put f (x, y) = u(x, y) + iv(x, y) and
note that (2.28) implies ∂∂ux = ∂v , ∂u = − ∂∂vx ) and so sn z = sn(x + iy) is ana-
∂y ∂y
lytic in . Similar arguments applied to cn and dn then yield the central result
of this chapter, as follows.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 38 2 Jacobian elliptic functions of a complex variable

Theorem 2.2 The functions sn (z), cn (z), dn (z) defined by (2.25), (2.26,)
(2.27), where z = x + iy, are analytic in  and for z ∈  we have:

d
sn (z) = cn (z) dn (z),
dz
d
cn (z) = −sn (z) dn (z),
dz
d
dn (z) = −k 2 sn (z) dn (z). (2.29)
dz
We have not yet established the nature of the exceptional points, z = 2m K +
i(2n + 1)K  , m, n ∈ Z, but we expect them to be simple poles, because of
(2.11).
Our next theorem shows that the addition formulae (2.19), (2.20), (2.21)
hold for all complex numbers u, v, u + v in . That will enable us to apply
those formulae to obtain results like

sn(u + K + iK  ) = k −1 dn(u)/cn(u)

(see below), by appealing to the addition formulae.

Theorem 2.3 The addition formulae (2.19), (2.20), (2.21) remain valid for
u, v ∈ C, provided that u, v and u + v are in .

Proof As in the proof of Theorem 2.1, the result

∂ ∂
S(u, v) = C(u, v)D(u, v) = S(u, v)
∂u ∂v
implies that S(u, v) = sn(u + v). The results C(u, v) = cn(u + v) and
D(u, v) = dn(u + v) follow similarly.

It is possible to prove Theorem 2.3 by appealing to a fundamental theorem

of complex analysis, to which we shall frequently have recourse in the sequel
and which we state here as:

Theorem 2.4 (The Uniqueness Theorem of Complex Analysis).

Let f (z), g(z) be analytic (holomorphic) in a domain G (that is, a connected
open subset of C) and suppose that f (z) = g(z) in a set of points H having a
limit point in G. Then f (z) = g(z) identically in G.

For the proof and further background, we refer the reader to Nevanlinna &
Paatero (1969), pp. 139−140.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.6 Periodic properties 39

Exercises 2.5
2.5.1 Show that:

sn(K + iK  ) = k −1 ,
cn(K + iK  ) = −ik  k −1 ,
dn(K + iK  ) = 0. (2.30)

2.5.2 Prove Theorem 2.3 using Theorem 2.4.

2.5.3 Show that the formulae (2.11), sn(iy, k) = i sn(y, k  )/cn(y, k  ), etc.,
remain valid for y ∈ C.

2.6 Periodic properties associated with K , K + iK  and iK 

In this and the following sections we follow Whittaker and Watson (1927)
pp. 500–505. However, it should be emphasized that those authors base their
treatment on the theta functions (which we postpone to a later chapter),
whereas our treatment is based essentially on that of Abel and on the addition
formulae.
We recall that sn (K ) = 1, cn (K ) = 0 and dn (K ) = k  (see 1.37) and we
take v = K in the addition theorem, Theorem 2.1, as extended to C in Theorem
2.3, to obtain
sn(u + K ) = cn (u)/dn (u),
cn(u + K ) = −k  sn (u)/dn (u), (u ∈ C), (2.31)
dn(u + K ) = k  /dn (u).
Iteration of the result in (2.31) yields
sn(u + 2K ) = −sn (u)
cn(u + 2K ) = −cn (u)
dn(u + 2K ) = dn (u)
and finally
sn(u + 4K ) = sn (u)
cn(u + 4K ) = cn (u)
dn(u + 4K ) = dn(u + 2K ) = dn (u);
from which we see that sn and cn each have period 4K, whereas dn has the
smaller period 2K.
Note that those last results follow from the ‘Uniqueness Theorem’, Theorem
2.4, and the results we already have for u ∈ R in (2.7)

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 40 2 Jacobian elliptic functions of a complex variable

We turn to periodicity in K + iK  . We take v = K + iK  , u ∈ C, in Theo-

rems 2.1 and 2.3 and use the results of (2.30) to obtain:

sn(u + K + iK  ) = k −1 dn (u)/cn (u),

cn(u + K + iK  ) = −ik  k −1 /cn (u),
dn(u + K + iK  ) = ik  sn (u)/cn (u). (2.32)

As before, iteration yields:

sn(u + 2K + 2iK  ) = −sn (u),
cn(u + 2K + 2iK  ) = cn (u),
dn(u + 2K + 2iK  ) = −dn (u), (2.33)

and, finally:
sn(u + 4K + 4iK  ) = sn (u),
cn(u + 4K + 4iK  ) = cn (u),
dn(u + 4K + 4iK  ) = dn (u). (2.34)

Hence the functions sn (u) and dn (u) have period 4K + 4iK  , but cn (u) has
the smaller period 2K + 2iK  .
We turn finally to periodicity with respect to iK  .
By the addition theorem, again, we have:
sn(u + iK  ) = sn(u − K + K + iK  )
= k −1 dn(u − K )/cn(u − K )
= k −1 /sn(u), (2.35)

by (2.31). Similarly:
cn(u + iK  ) = −ik −1 dn (u)/sn (u)
dn(u + iK  ) = −i cn (u)/sn (u) (2.36)

and, as before, iteration yields:

sn(u + 2iK  ) = sn (u),
cn(u + 2iK  ) = −cn (u),
dn(n + 2iK  ) = −dn (u). (2.37)

Finally:
sn(u + 4iK  ) = sn(u + 2iK  ) = sn (u),
cn(u + 4iK  ) = cn (u),
dn(n + 4iK  ) = dn (u), (2.38)

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.6 Periodic properties 41

and we see that the functions cn (u) and dn(u) have period 4iK  , while sn (u)
has the smaller period 2iK  .
Once again the Uniqueness Theorem 2.3 may be used to obtain those results.

Theorem 2.5 The results of this section, together with (2.12), yield the identity
conjectured on the basis of dynamical considerations in Section 1.8, namely:

sn(±ix + K , k) = 1/dn(x, k  ),

and
k −1 dn(x − K  , k  ) = 1/dn(x, k  ).

Proof We have

sn(±i x + K , k) = cn(±i x, k)/dn(±i x, k)

cn(±x, k  )−1 cn(±x, k  )
=
dn(±x, k  )
= 1/dn(x, k  ).

Again

k −1 dn(x − K  , k  ) = k −1 dn(−x + K  , k  )
= k −1 k/dn(−x, k  )
= 1/dn(x, k  ).

The points m K + niK  , m, n ∈ Z, form a pattern of rectangles in the complex

plane, called a lattice. The lattice is generated as a Z-module by the basis
vectors (K , iK  ). We have seen that the sub-lattice of points of the form z =
4m K + 4niK  is a period lattice for the functions sn, cn and dn and that each
of the functions has a different sub-lattice of periods, given by the scheme:
sn (u) 4K 2iK 
cn (u) 4K 2K + 2iK 
dn (u) 2K 4iK  (2.39)

The fundamental parallelogram of each of those sub-lattices has area a half

that of the lattice with basis 4K , 4iK  .
The function cn (u) for example, takes the same value, cn (u 0 ), at all points
u ∈  congruent, modulo its period lattice, to the point u 0 interior to the paral-
lelogram generated by (4K , 2K + 2iK  ). A similar result holds for each of the
functions sn and dn, modulo its period parallelogram.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 42 2 Jacobian elliptic functions of a complex variable

We have not yet described the behaviour of the Jacobian elliptic functions
at points congruent to iK  (though it will come as no surprise to learn that
they have poles there), but we shall see later that the properties of the Jacobian
elliptic functions derived here are characteristic of elliptic functions in general
and that the notion of a period lattice plays a fundamental role in their definition
and the derivation of their properties.
In his book The White Knight1 , A. L. Taylor suggests that, in Through
the Looking Glass, Lewis Carroll had the double periodicity of the Jacobian
elliptic functions in mind when he pictured the chess-board world that Alice
found there – a sort of two-dimensional time represented in terms of a
(square, K = K  ) period lattice, in which space and time have been inter-
changed.

Exercises 2.6
2.6.1 Prove the periodicity properties with respect to K and iK  , using Theorem
2.4.
2.6.2 Obtain the formulae:
sn(u + 2m K + 2niK  ) = (−1)m sn(u)
cn(u + 2m K + 2niK  ) = (−1)m+n cn(u)
dn(u + 2m K + 2niK  ) = (−1)n dn(u)
2.6.3 Prove that:
K 1 iK  i
sn = √ , sn =√ ,
2 1+k  2 k
√ √
K k iK  1+k
cn = √ 
, cn = √ ,
2 (1 + k ) 2 k

K iK
dn = k  , dn = (1 + k),
2 2
and that:
K + iK  √ √
sn = ( 1 + k + i 1 − k)/ (2k),
2
(K + iK  ) √
cn = (1 − i) k  / (2k),
2
K + iK  √ √ √
dn = (kk  ){ 1 + k  − i 1 − k  }/ 2k.
2

1 Edinburgh, Oliver and Boyd, 1952, pp. 89–93.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.7 Jacobian elliptic functions 43

2.7 The behaviour of the Jacobian elliptic functions near

the origin and near iK 
We have, for u ∈ ,
d
sn(u) = cn(u) dn(u)
du
and
d3
sn(u) = 4k 2 sn 2 (u) cn(u) dn(u) − cn(u) dn(u)(dn 2 (u) + k 2 cn 2 (u)).
du 3
Since sn(u) is an odd function of u, Maclaurin’s Theorem (see Exercises 1.9,
for the case when u ∈ R) yields, for small values2 of |u|,
1
sn(u) = u − (1 + k 2 )u 3 + O(u 5 ). (2.40)
6
In like manner (recalling that cn, dn are even functions)
1
cn(u) = 1 − u 2 + O(u 4 ), (2.41)
2
1
dn(u) = 1 − k 2 u 2 + O(u 4 ). (2.42)
2
The identification of all the zeros of sn and of those of cn and dn is given in
Exercise 2.7.1.
It follows, using (2.36), that:
sn(u + iK  ) = k −1 /sn(u)
−1
1 1
1 − (1 + k 2 )u 2 + O(u 4 )
=
ku 6
1 1 + k2
= + u + O(u 3 ). (2.43)
ku 6k
A similar argument yields:
i 2k 2 − 1
cn(u + iK  ) = + + iu + O(u 3 ), (2.44)
ku 6k
i 2 − k2
dn(u + iK  ) = − + iu + O(u 3 ). (2.45)
u 6
It follows that at the point iK  the functions sn(u), cn(u), dn(u) have simple
poles with residues k −1 , −ik −1 , −i, respectively.

2 The symbol O denotes ‘of the order of’ and f = O(g) as x → x0 means that | f (x)/g(x)| < K
as x → x0 , where K is independent of x for x close enough to x0 . Similarly f = o(g) means
f (x)
that lim g(x) = 0. The notation is due to Bachmann and Landau, see Whittaker and Watson
x→u
(1927) p. 11.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 44 2 Jacobian elliptic functions of a complex variable

Note that
1
sn(u + 2K + iK  ) = −sn(u + iK  ) = − + O(u)
ku
and so, at the point 2K + iK  , sn has a simple pole with residue −k −1 .
Similarly,

cn(u + 2K + iK  ) = −cn(u + iK  ),

whence, at 2K + iK  , cn has a simple pole with residue ik −1 . Finally,

dn(u + 3iK  ) = dn(u + iK  + 2iK  ) = −dn(u + iK  ),

whence, at the point 3iK  , dn has a simple pole with residue i.

The change in sign of the residues shows that the sum of the residues at the
poles in the corresponding period parallelogram is 0 and is a special case of
a general theorem about the poles of doubly periodic functions in general that
will be proved in Chapter 3.

Exercises 2.7
2.7.1 (The zeros of sn, cn, dn). Show that (2.25) implies that sn(x + iy) = 0 if
and only if simultaneously sn(x, k) = 0 and sn(y, k  ) = 0; that is, if and
only if x + iy ≡ 0(mod 2K , 2iK  ). So that identifies all the zeros of sn,
modulo periods.
(That the zeros are simple follows from
d
sn(u)|u=0 = cn (0) dn (0) = 1 = 0
du
and the periodicity properties proved above.)
Use a similar argument to show that the zeros of cn and dn are simple
and occur at the following points:

cn (z) = 0 if and only if z ≡ K (mod 2K , 2iK  )

dn (z) = 0 if and only if z ≡ K + iK  (mod 2K , 2iK  ).

2.7.2 Define the parameter q by

iK 
q = eπ iτ , τ= , K > 0, K  > 0.
K
(Note that Im(τ ) > 0.) Verify that the mappings k → K and k →

K are increasing and decreasing, respectively and hence show that the

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.7 Jacobian elliptic functions 45

mapping k → q is a one-to-one and strictly increasing mapping of the

open interval (0, 1) onto (0, 1). Prove that

lim q = 0, lim q = 1.
k→0+ k→1−

(The significance of the parameter q will emerge in Section 2.9.)

General description of the functions sn, cn, dn

The investigations of this section may be summarized, as follows (u denotes a
complex variable).

1. The function sn(u) is a doubly periodic function of the complex variable

u with periods 4K , 2iK  . It is analytic (holomorphic) except at the points
congruent to iK  and 2K + iK  · (mod 4K , 2iK  ). Those excluded points
are simple poles of sn(u), the residues at the first set all being k −1 and the
residues at the second set −k −1 . The function has a simple zero at all points
congruent to 0(mod 2K , 2i K  )
If 0 < k 2 < 1, then K and K  are real and sn(u) is real for u ∈ R and is
pure imaginary when u is pure imaginary.
(We shall see later that sn(u) is a unique function in the sense that it is
the only function satisfying the foregoing description. For if F(u) were a
second such function then sn(u) − F(u) would be a doubly periodic function
without singularities, and we shall show in Chapter 3 that such a function
must be a constant, and that constant must be 0.)
2. The function cn(u) is a doubly periodic function of u with periods 4K and
2K + 2iK  . It is analytic except at points congruent to iK  or to 2K +
iK  (mod 4K , 2K + 2iK  ), where it has simple poles with residues −ik −1
(at the first set) and +ik −1 , (at the second set), respectively. The function cn
has a simple zero at all points congruent to K (mod 2K , 2iK  ).
3. The function dn (u) is a doubly periodic function of u with periods 2K
and 4iK  . It is analytic at all points except those congruent to iK  or to
3iK  · (mod 2K , 4iK  ) where it has simple poles. The residues at the first
set are −i and at the second set +i. The function has a simple zero at all
points congruent to K + iK  (mod 2K , 2iK  )

The functions cn and dn, like sn, are unique with respect to the doubly
periodic functions having those properties, respectively. See Figures 2.1, 2.2
and 2.3.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 46 2 Jacobian elliptic functions of a complex variable

u2

3K′

2K′

K′

u3
−K 0 K 2K 3K 4K

Figure 2.1 Relief of sn (u) for k = 0.8 (u = u 1 + iu 2 )

u1

3K′

2K ′

K′

u1
0 K 2K 3K 4K 5K
−K′
−K

Figure 2.2 Relief of cn (u) for k = 0.8 (u = u 1 + iu 2 )

2.8 Glaisher’s notation, the lemniscatic integral and the

lemniscate functions
As in the case of the circular functions (k = 0) it is convenient to have a notation
for the reciprocals and quotients of our three functions, as will already have
occurred to the reader in reading formulae like (2.32). The standard notation
is due to Glaisher (1881) (who also introduced the notation s1 , c1 , d1 etc. used
above, see Glaisher, 1881). One denotes the reciprocals by reversing the order
of the letters, used to express the function concerned, thus:
1 1 1
ns(u) = , nc(u) = , nd(u) = .
sn(u) cn(u) dn(u)

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.8 Glaisher’s notation: lemniscate functions 47

u1

4K ′

3K ′

2K ′

K′

u2

−K′
2K 3K
−2K′
K

Figure 2.3 Relief of dn (u) for k = 0.8 (u = u 1 + iu 2 )

The quotients are denoted by writing, in order, the first letter of the function in
the numerator and of the function in the denominator, thus:
sn(u) sn(u) cn(u)
sc(u) = , sd(u) = , cd(u) = ,
cn(u) dn(u) dn(u)
cn(u) dn(u) dn(u)
cs(u) = , ds(u) = , dc(u) = .
sn(u) sn(u) cn(u)
We return to the problem of the rectification of the arc of the lemniscate,
considered in Miscellaneous Exercises 1.9.7.
The equation of the curve introduced there may also be written in terms of
polar coordinates (r, θ ) in the form

r 2 = cos 2θ

(more generally, r 2 = a 2 cos 2θ, where the fixed points are now Q(−a, 0),
R(a, 0); so that the distance between them is 2a and we take the constant
represented by the product of the distance d(P, Q)d(P, R) to be a).
Let s denote the arc length from θ = π2 to θ (see Figure 2.4).
Then ds 2 = dr 2 + r 2 dθ 2 becomes
dr 2
ds 2 = ,
(1 − r 4 )

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 48 2 Jacobian elliptic functions of a complex variable

s
(r, q)

(−1, 1) q
(l, 0)

Figure 2.4 The lemniscate, r 2 = cos 2θ

and so we are led to consider the integral

r
φ= (1 − r 4 )−1/2 dr.
0

The integral is reminiscent of the integral for arcsin r (when r 4 is replaced by

r 2 ) and we also recall (on comparison with (1.16), see Exercise 1.8.7) that it is
Jacobi’s normal form with k = i. So instead of writing x = arcsin φ, we write
x = sin lemn φ and we have the relation
 
1
sin lemn φ = cos lemn −ϕ
2
obtained from

1
1
1
ϕ= (1 − r 4 )−1/2 dr, ω= (1 − r 4 )−1/2 dr. (2.46)
x 2 0

Those lemniscate functions were the first defined by the inversion of an

integral3 and they may be represented in terms of elliptic functions. As already
noted they lead to the integral 0 (1 − r 4 )−1/2 dr, which is Jacobi’s normal form
r

with k = i, √
but they may also be expressed in terms of Jacobi functions with
modulus 1/ 2, as follows. √ √
In Exercise 2.8.1, take k = 1/ 2 (also k  = 1/ 2) to obtain

sd(x,1/√2)  −1/2
1
x= 1 − t4 dt
0 4
√ √
√
sd(x,1/ 2)/ 2
= 2 (1 − r 4 )−1/2 dr
0

3 Gauss (1799) Werke III p. 404. The idea of investigating the lemniscate functions by inversion
of the integral occurred to Gauss on January 8th 1797.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.8 Glaisher’s notation: lemniscate functions 49

√ √
on putting t = r / 2. Now let x = K (1/ 2) and u = r 4 to obtain4 :
 
1 √
1
K √ = 2 (1 − r 4 )−1/2 dr
2 0

1
1
= √ u −3/4 (1 − u)−1/2 du
23 0
     
− 32 1 1 3
=2   
4 2 4
 2
1 1
= π −1/2  · (2.47)
4 4
The formula of Exercise 2.8.1 also shows that
√ √
sin lemn φ = 2−1/2 sd(φ 2, 1/ 2)
and similarly
√ √
cos lemn φ = cn(φ 2, 1/ 2).
The number (1/2)ω in (2.46) is the smallest positive value of φ for which
√ √
cn(φ 2, 1/ 2) = 0;
√
so that ω = 2K 0 , the number K 0 being the value of K associated with the
modulus k = √12 = k  . The result in (2.47) expresses K 0 in terms of the gamma
K 0
π ii
function. Note that K 0 = K 0 and so q0 = e K0 = e−π .
The ideas of this section played an important role in the development of the
theory of elliptic functions and especially in the inversion of elliptic integrals;
for further details see Siegel (1969), Chapter 1.

Exercises 2.8
2.8.1 Show that:

sd(x)
(a) x = (1 − k 2 t 2 )−1/2 (1 + k 2 t 2 )−1/2 dt;
0
sin x
(Hint: how would you show that (1 − t 2 )−1/2 dt = x?)
0

4 We have used the definition of the beta function

1
(ρ)(ε)
B(ρ, ε) = = x ρ−1 (1 − x)ε−1 d x
(ρ + ε) 0
and the result (z)(1 − z) = π/sin π z from the theory of the gamma functions. See
Whittaker & Watson (1927) p. 253 or Bateman et al. (1953).

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 50 2 Jacobian elliptic functions of a complex variable

sc(x)
(b) x = (1 + t 2 )−1/2 (1 + k 2 t 2 )−1/2 dt
0

∞
(c) x = (t 2 + 1)−1/2 (t 2 + k 2 )−1/2 dt
cs(x)

1
(d) x = (1 − t 2 )−1/2 (1 − k 2 t 2 )−1/2 dt
cd(x)

∞
(e) x = (t 2 − 1)−1/2 (t 2 − k 2 )−1/2 dt
ns(x)

nd(x)
(f) x = (t 2 − 1)−1/2 (1 − k 2 t 2 )−1/2 dt.
1

 1/2
2.8.2 Fagnano’s5 substitution t → r = 2t 2 /(1 + t 4 ) (cf. (2.6)) effects a
monotone mapping of the interval 0 ≤ t ≤ 1 on to the interval 0 ≤ r ≤ 1.
Show that
dr dt
= 21/2 .
(1 − r 4 )1/2 (1 + t 4 )1/2
Hence
 
1  
1
1 1 −1/2
K √ =2 Q t, √ dt = 2 (1 + t 4 )−1/2 dt
2 0 2 0

1
= 21/2 (1 − r 4 )−1/2 dr
0

(as in (2.47)).
2.8.3 Show that
1 − cos lemn2 φ
sin lemn2 φ = ·
1 + cos lemn2 φ

2.8.4 Show that the length of one loop of the lemniscate r 2 = a 2 sin 2θ is

a  
2a 2 dr √ 1
= a 2K √ .
0 (a 4 − r 4 ) 2

5 Il Marchese Giulio Carlo de Toschi di Fagnano, another of the founders of the theory, who, like
Maclaurin and Legendre, encountered elliptic integrals in connection with the problem of
rectifying an arc of an ellipse. See also Chapter 9.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.9 Fourier series and sums of two squares 51

If s denotes the length of the arc measured from the pole to the point
(r, θ ), 0 < θ < 14 π, show that
 
√ 1
r = a cn(K − s 2/a) k = √ .
2

2.9 Fourier series and sums of two squares

Consider a function f (z) of period 2π analytic in the strip a < Im(z) <
b (−∞ ≤ a < b ≤ +∞). The mapping z → ζ = eiz maps the strip in the z
plane into the annulus e−b < |ζ | < e−a in the ζ -plane. Since f (z) has period
2π, F(ζ ) = f (z) is well defined and analytic in the annulus and so it possesses a
Laurent expansion there (see, for example Whittaker & Watson (1927), Section
5.6, p. 100), namely:
∞
F(ζ ) = Cn ζ n ,
n=−∞

where


1
Cn = F(ζ )ζ −n−1 dζ, (e−b < r < e−a ).
2πi
|ζ |=r

By changing the variable back to z, the coefficients Cn give

∞
f (z) = Cn einz , (a < Im(z) < b) (2.48)
n=−∞

where

z 0 +2π
1
Cn = f (z)e−inz dz, (2.49)
2π z0

z 0 being any point of the strip a < Im(z) < b and the integration is along any
path from z 0 to z 0 + 2π that remains entirely in the strip. When f (z) is an entire
function (that is a = −∞, b = +∞) the Fourier development given by (2.48)
is valid everywhere.  
π
As functions of z = 2K u, the functions sn (u), cn (u) and dn (u) are ana-
lytic in the strip
 π  π
|Im(z)| < K  = Im(τ ),
2K 2
where τ = iK  /K , and have period π or 2π (depending on the periodicity
properties of sn (u), cn (u) and dn (u) with respect to K).

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 52 2 Jacobian elliptic functions of a complex variable

The function sn (u) is an odd function of u and so has a Fourier sine series
∞
sn(u) = bn sin nx, x ∈ R,
n=1

where

π
π ibn = sn(u) · exp(nix)dx.
−π

Similarly, cn(u) and dn (u) are even functions and possess Fourier cosine expan-
sions. For reasons which will become apparent and which are dictated by elegant
applications to the theory of numbers, we choose to derive the series for dn (u).
(The other series are given as exercises.) π 
We recall that dn (u) is an even function of z = 2K u, with periods π
and 2π τ , where τ = iK  /K , and that it has a pole at z = π τ /2 with residue
−iπ/(2K ). Hence we may write
∞ ∞
b0
dn(u) = an e2niz = + bn cos 2nz,
n=−∞ 2 n=1

where a−n = an and

π
−1
bn = 2an = π e−2niz dn(u) dz. (2.50)
−π

To evaluate the integrals in (2.50) we appeal to the residue theorem and we

consider the integral


exp(−2niz) · dn(u) · dz,
C

where C denotes the boundary of the parallelogram shown in Figure 2.5.

Inside C, the integrand has simple poles at π τ /2 and at −π + π τ /2,
π −niπ τ
 −n at each (inπ iτterms of the variable z) being −i 2K e
the residue
π
=
−i 2K q (using q = e ). Hence

 
2π 2
e−2niz · dn(u) · dz = 2πi(sum of the residues) = q −n .
K
C

The integrals along the left- and right-hand boundaries cancel by periodic-
ity. For the upper boundary, write z = t − π + π τ (t goes from π to −π )

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.9 Fourier series and sums of two squares 53

−2 p + pt pt

pt
•−π + ⎯
2 pt
⎯
2

-p 0 p

Figure 2.5 The contour C.

to obtain

πτ
π  
−2niz −2niπ τ −2nit 2K
− e · dn(u) · dz = e e · dn t dt
−2π+πτ −π π

π
−2n
=q e−2niz · dn(u) · dz.
−π

Hence

π
(1 + q −2n ) e−2niz dn(u) dz
−π

 
2π 2
= e−2niz dn(u) dz = q −n .
K
C

But now we have obtained

π
1
bn = e−2niz dn(u) dz
π
−π

2π q −n 2π q n
= = .
K 1 + q −2n K 1 + q 2n
So finally we have
∞
π 2π qn
dn(u) = + cos 2nz, (2.51)
2K K n=1
1 + q 2n

in the strip |Im(z)| < π2 Im(τ ).

In his Presidential address to the London Branch of the Mathematical Asso-
ciation (Mathematical Gazette, XXXIV, 1950), Davenport urged teachers to

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 54 2 Jacobian elliptic functions of a complex variable

stress ‘the different ways of looking at each topic, and the curious connections
between various topics’. One of the most surprising and curious connections is
that between the theory of numbers and the theory of elliptic functions, which
emerges if we take u = 0 and then multiply both sides of (2.51) by 2K /π . We
obtain
∞ ∞ ∞
2K qm
= 1+4 = 1 + 4 qm (−1)l q 2ml
π m=1
1 + q 2m
m=1 l=0
∞ ∞
= 1+4 (−1)l q (2l+1)m .
m=1 l=0

Now write n = (2l + 1)m; d = 2l + 1, whence l = (d − 1)/2. Then d runs

through the odd divisors of n and we have
⎡ ⎤
∞
2K ⎣ (−1) d−1
=1+4 2 ⎦q n .
π n=1 d|n
d odd

Now define
0 if m is even
χ (m) =
(−1)(m−1)/2 if m is odd
and write
δ(n) = χ (d).
d|n

Clearly, χ (m) = 1 if m is odd and ≡ 1(mod 4) and χ (m) = −1 if m is odd and

≡ 3(mod 4). Whence
δ(n) = [number of divisors of n ≡ 1(mod 4)]
− [number of divisors of n ≡ 3(mod 4)].
So finally we obtain
 
∞
2K
= 1+4 χ (d) q n
π n=1 d|n
∞
= 1+4 δ(n)q n . (2.52)
n=1

Denote by r (n) the number of representations of n in the form

n = a 2 + b2 (a, b ∈ Z),
where we count as distinct even those representations which differ only in
respect of the sign or the order of the integers a, b. (Thus r (5) = 8, since

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.9 Fourier series and sums of two squares 55

5 = (±1)2 + (±2)2 = (±2)2 + (±1)2 ). In Chapter 4 we shall prove Jacobi’s

Theorem
∞
2K
=1+ r (n)q n , (2.53)
π n=1

from which (together with Exercise 2.7.2 and (2.52)) we obtain:

Theorem 2.6 Let n be an integer greater than or equal to 1, then r (n) = 4δ(n).

Theorem 2.6 is Theorem 278 in Hardy and Wright (1979). It was first proved
by Jacobi using elliptic functions, but it is equivalent to one stated by Gauss
(see Hardy and Wright (1979), p. 243).

Corollary 2.1 Every prime p of the form 4m + 1 is representable as the sum

of two squares in an essentially unique way6 .

That a prime p of the form 4m + 3 is never the sum of two squares also
follows, but that result is trivial, since the square of an integer is congruent to
0 or 1(mod 4) and so the sum of two squares cannot be congruent to 3, mod 4.

Exercises 2.9
2.9.1 Show, using a similar argument to that used to obtain (2.51), that in the
same strip
∞
2π q n+1/2 sin(2n + 1)z
sn(u) = , (2.54)
Kk n=0
1 − q 2n+1

∞
2π q n+1/2 cos(2n + 1)z
cn(u) = . (2.55)
Kk n=0
1 + q 2n−1

(Hint: the details of the calculation are given in pp. 510–511 of Whittaker
and Watson (1927)).
2.9.2 By writing z + π/2 in place of z in the results obtained (including Exer-
cise 2.9.1) show that, if
2K z π
u= , |Im(z)| < Im(τ ),
π 2

6 See the earlier remark; ‘essentially unique’ means that we regard representations differing only
in the signs of a or b or in the order of a and b as the same.

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 56 2 Jacobian elliptic functions of a complex variable

then
∞
2π q n+1/2 cos(2n + 1)z
cd(u) = (−1)n ,
Kk n=0
1 − q 2n+1
∞
2π q n+1/2 sin(2n + 1)z
sd(u) = (−1)n ,
K kk  n=0
1 + q 2n+1
∞
π 2π q n cos 2nz
nd(u) = 
+ (−1)n .
2K k K k n=1
1 + q 2n

2.9.3 Verify that r (n) = 4δ(n) for n = 1, 2, . . . , 10, by explicit calculations.

2.9.4 Interchange the order of summation in the expression obtained from
(2.51) to obtain
∞
2K q 2l+1
=1+4 (−1)l .
π l=0
l − q 2l+1

2.10 Summary of basic properties of the Jacobian

elliptic functions
The numbers in parentheses below indicate the appropriate references in the
text.

Basic notation Let the modulus k, 0 < k < 1, be given and let ψ = 2 arctan t

(2.1) Q(t, k) = 1 + 2(1 − 2k 2 )t 2 + t 4 = [t 2 + (1 − 2k 2 )]2 + 4k 2 (1 − k 2 ),

1
π/2
dt dψ
(2.2) K =2 = ,
0 Q(t, k)1/2 0 (1 − k 2 sin2 ψ)1/2

t
ds
(2.3) x =2 , −2K < x < 2K .
0 Q(s, k)1/2

Complementary modulus In terms of the modulus, k, define

1
dt
k  = (1 − k 2 )1/2 , K  = K (k  ) = 2  1/2
·
0 Q(t, k )

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.10 Summary of basic properties of the Jacobian elliptic functions 57

Definition of the Jacobian elliptic functions on (−2K , 2K ) Let t be defined

as a function of x by inverting (2.3)
2t
sn(x, k) = = sin ψ,
1 + t2
1 − t2
(2.4) cn(x, k) = = cos ψ,
1 + t2
Q(t, k)1/2
dn(x, k) = = (1 − k 2 sin2 ψ)1/2 ,
1 + t2

(2.6) sn(±2K ) = 0, cn(±2K ) = −1, dn(±2K ) = 1.

Extend all the functions to (−∞, +∞) with period 4K. The extension of sn, cn
and dn are respectively odd, even and even functions.

Further notation Let u, v be real:

 = (u, v) = 1 − k 2 sn 2 u sn 2 v,
S(u, v) = [(sn(u))(cn (v))(dn (v)) + (sn (v))(cn (u))(dn(u))]/,
C(u, v) = [(cn(u))(cn (v)) − (sn(u))(sn (v))(dn(u))(dn (v))]/,
D(u, v) = [(dn(u))(dn (v)) − k 2 (sn(u))(sn (v))(cn(u))(cn (v))]/.

Extension to the complex plane

sn(x + iy) = S(x, iy),
cn(x + iy) = C(x, iy), x and y real, y = (2n + 1)K  ,
(2.23) dn(x + iy) = D(x, iy).

Addition formulae (Theorem 2.1)

sn(u + v) = S(u, v),

cn(u + v) = C(u, v), u and v complex,
dn(u + v) = D(u, v).

Differentiation formulae
sn  z = cn(z) dn(z),
cn  z = −sn(z) dn(z),
(2.29) dn  z = −k 2 sn(z) dn(z).

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 58 2 Jacobian elliptic functions of a complex variable

Periodicity Let u in the formulae below be complex

sn(u + K ) = cn(u)/dn(u),
cn(u + K ) = −k  sn(u)/dn(u),
(2.31) dn(u + K ) = k  /dn(u), .

sn(u + 2K ) = −sn(u),
cn(u + 2K ) = −cn(u),
dn(u + 2K ) = dn(u),

(2.35) sn(u + iK  ) = k −1 /sn(u),

cn(u + iK  ) = −ik −1 dn(u)/sn(u),

(2.36) dn(u + iK  ) = −icn(u)/sn (u),
sn(u + 2iK  ) = sn(u),
cn(u + 2iK  ) = −cn(u),
(2.37) dn(u + 2iK  ) = −dn(u).

Periods
sn : 4K , 2iK  ; cn : 4K , 2K + 2iK  , 4iK  ; dn : 2K , 4iK  .

Poles and residues The functions are analytic except for simple poles at
u ≡ iK  mod (2K , 2iK  ); sn, cn and dn have residues k −1 , −ik −1 and −i
respectively at iK  .

Zeros mod(2K , 2iK  ) (Exercise 2.7.1)

sn : u ≡ 0; cn : u ≡ K ; dn : u ≡ K + iK  .

Formulae after change of variable

u = (2K /π )z; τ = iK  /K ; q = exp(π iτ ) = exp(−π K  /K ).

(a) Periods

sn : 2π, πτ ; cn : 2π, π + πτ, 2π τ ; dn : π , 2π τ.

(b) Poles are simple and lie at z ≡ 12 πτ mod (π, π τ ).

(c) Zeros mod(π, πτ )
π π πτ
sn : z ≡ 0; cn : z ≡ ; dn : z ≡ + .
2 2 2

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.10 Summary of basic properties of the Jacobian elliptic functions 59

2.10 Miscellaneous exercises

2.10.1 Show that, if  = 1 − k 2 s12 s22 , where s1 = sn(u), s2 = sn (v), etc.,
then:
(a) sn(u + v) + sn(u − v) = 2s1 c2 d2 /;
(b) sn(u + v) − sn(u − v) = 2s2 c1 d1 /;
(c) cn(u + v) + cn(u − v) = 2c1 c2 /;
(d) cn(u + v) − cn(u − v) = −2s1 s2 d1 d2 /;
(e) dn(u + v) + dn(u − v) = 2d1 d2 /;
(f) dn(u + v) − dn(u − v) = −2k 2 s1 s2 c1 c2 /.

Check that those results agree with what you would expect in the cases
k = 0, k = 1.
2.10.2 In the notation of Exercise 2.10.1, prove that:

(a) sn(u + v)sn(u − v) = (s12 − s22 )/;

(b) sn(u + v)cn(u − v) = (1 − s12 − s22 + k 2 s12 s22 )/;
(c) dn(u + v)dn(u − v) = (1 − k 2 s12 − k 2 s22 + k 2 s12 s22 )/.

2.10.3 Let cn(u + v) = C1 , cn(u − v) = C2 , dn(u + v) = D1 and dn(u − v)

= D2 . Prove that
C2 − C1 1 D2 − D1
sn(u) sn (v) = = 2
D2 + D1 k C2 + C1
and hence
2(C1 D2 + C2 D1 )
1 − k 2 sn 2 u sn 2 v = ,
(C1 + C2 )(D1 + D2 )
C1 D2 + C2 D1
cn(u) cn (v) = ,
D1 + D2
C 1 D2 + C 2 D1
dn(u) dn (v) = .
C1 + C2
By putting u = α + β and v = α − β, deduce that
cn(2β) − cn(2α)
sn(α + β)sn(α − β) = ,
dn(2β) + dn(2α)
cn(2α) dn(2β) + dn(2α) cn(2β)
cn(α + β)cn(α − β) = ,
dn(2α) + dn(2β)
cn(2α) dn(2β) + dn(2α) cn(2β)
dn(α + β)dn(α − β) = .
cn(2α) + cn(2β)

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 60 2 Jacobian elliptic functions of a complex variable

2.10.4 Prove the following:

√
1  i 1  (1 + k) 1
(a) sn iK = √ , cn iK = √ , dn iK  = (1 + k);
2 k 2 k 2
(b) sn 1 (K + iK  ) = (√1 + k + i√1 − k)/√2k;
2
1 √ √
cn (K + iK  ) = (1 − i) k  / 2k;
2
1 √ √ √ √
dn (K + iK  ) = kk  ( 1 + k  − i 1 − k  )/ 2k.
2
2.10.5 For u, v ∈ R, prove that:
  
 
sn u + 1 iK  = √1 = constant
 2  k
   
 1 
 sn K + iv 
 2  1
   = √ ;
 1 k
 cn K + iv 

2
   
 1 
 dn u + iK  
 2  √
   = k.
 1
 cn u + iK  
 
2

2.10.6 By considering eπ iu/K cn(u) dn(u)

sn(u)
du round the rectangle, C, with
C
vertices at ±K , ±K + 2iK  , suitably indented, show that

K
cn(u) dn(u) πK
eπ iu/K du = π i tanh .
−K sn(u) 2K
2.10.7 Prove that
(1 ± cn(u + v))(1 ± cn(u − v))
(cn u ± cn v)2
= .
(1 − k 2 sn 2 u sn 2 v)
2.10.8 Prove that
cn 2 u + cn 2 v
1 + cn(u + v)cn(u − v) = .
1 − k 2 sn 2 u sn 2 v
2.10.9 Let am(u) be defined by the formula

am(u)
u= (1 − k 2 sin2 θ )−1/2 dθ.
0

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010
 2.10 Summary of basic properties of the Jacobian elliptic functions 61

Verify that
d
sn(u) = sin(am(u)), cn(u) = cos(am(u)), dn(u) =
(am(u)).
du
2.10.10 Obtain the following Fourier expansions, valid throughout |Im(z)| <
πIm(τ ), except at the poles of the first term on the right-hand side of
the respective expansions.
∞
π 2π sin(2n + 1)z
ds(u) = cosec z − q 2n+1 ,
2K K n=0
1 + q 2n+1
∞
π 2π sin 2nz
cs(u) = cot z − 2n
q ,
2K K n=1
1 + q 2n
∞
π 2π
dc(u) = sec z + (−1)n q 2n+1 .
2K K n=0

2.10.11 Show that if k is so small that k4 and higher powers of k may be

neglected, then
1
sn(u) = sin u − k 2 cos u(u − sin u cos u),
4
for small values of u.
2.10.12 Prove that, for 0 > Im(z) > −πIm(τ ),
∞
π 2π q 2n+1 sin(2n + 1)z
ns(u) = cosec z + .
2K K n=0
1 − q 2n+1

Cambridge Published
https://doi.org/10.1017/CBO9780511617867.004 Books Online ©Cambridge
online by Cambridge University
University Press Press, 2010