Math 213a (Fall 2021) Yum-Tong Siu 1

THETA FUNCTIONS OF JACOBI

Topics of Jacobian Theta Functions Covered. We will define the four

Jacobian theta functions ϑ1 , ϑ2 , ϑ3 , ϑ4 for the lattice Zπ + Zπτ by Fourier
series. Their periodicity factors are summarized in the table of periodicity
factors. The four identities which are analogues of sin2 θ + cos2 θ = 1 will
be derived. They are identities involving the squares of Jacobian theta func-
tions and their values at 0 which are called the Jacobian theta constants.
The three basic Jacobian elliptic functions will be expressed as quotients of
Jacobian theta functions. Their elliptic moduli will be expressed in terms of
the Jacobian theta constants.

One crucial ingredient in expressing the three basic Jacobian elliptic

functions as quotients of Jacobian theta functions is the identity ϑ01 (0) =
ϑ2 (0)ϑ3 (0)ϑ4 (0), which is the analogue of
 
d
sin θ = (cos θ)θ=0 ,
dθ θ=0

and which is proved by using the heat-equation-type differential equation

satisfied by the four Jacobian theta functions, proved by using their Fourier
series. At the end Jacobi’s theta transform is presented, which relates ϑ3 (z, τ )
to ϑ3 ( τz , − τ1 ) and is a very important and useful identity. For example, it play
a crucial role in the analytic approach to quadratic reciprocity and is used
in the proof of the two-squares theorem.

Theta Functions as Multiplicative Building Blocks. Any meromorphic

function f on the Riemann sphere C ∪ {∞} can be written uniquely as
Qp
(w − ak )
f (w) = Qk=1
q
`=1 (w − b` )

when {a1 , · · · , ap } and {b1 , · · · , bq } are disjoint. We consider each of w − aj

and w − bj a multiplicative building block which has a simple zero on C and
a pole at ∞. An elliptic function is a meromorphic function on C with two
R-linearly independent periods ω1 , ω2 , which is the same as a meromorphic
function on C /(Zω1 + Zω2 ). The question is whether there is a correspond-
ing object of a multiplicative building block for elliptic functions.
Math 213a (Fall 2021) Yum-Tong Siu 2

It is not possible to have an elliptic function with a simple zero and a

simple pole, because the derivative of its logarithm violates the condition
that the sum of the residues of an elliptic function over a fundamental par-
allelogram must be zero. Since we are interested in taking the quotient of
products of such multiplicative building blocks, we can relax the condition of
periodicity to allow a nowhere zero holomorphic factor for each period. To
make sure that the factor is nowhere zero holomorphic, the most natural way
is to make it the exponential function of a holomorphic function. This func-
tion in the exponent of the periodicity factor cannot be a constant, otherwise
the derivative of the logarithmic of the multiplicative building block would
be an elliptic function for which the sum of its residues over a fundamental
parallelogram is nonzero. The simplest situation is that the function in the
exponent of the periodicity factor is a polynomial of degree ≤ 1.

Construction of Theta Function by Recurrent Determination of the Co-

efficients of Its Fourier Series. Let us construct the simplest nontrivial
example f of this kind. We use π and πτ with Im τ > 0 as the two periods.
Assume that our multiplicative building block has period π and a periodicity
factor for πτ which is the exponent of a polynomial of degree ≤ 1. Since f
has period π, we can write f (w) as a function of e2iw .
∞
X
f (w) = cn e2niw .
n=−∞

For the other period πτ we have a periodicity factor

f (w + πτ ) = eaw+b f (w)

for some a, b ∈ C.

In order to make the induction process of the determination of the un-

known coefficients cn as simple as possible, we choose a = −2i and b =
−πτ i + πi so that eaw+b = −e−2iw−iπτ and

f (w + πτ ) = −e−2iw−iπτ f (w)
Math 213a (Fall 2021) Yum-Tong Siu 3

to solve for the undetermined coefficients cn . From

∞
X ∞
X
2ni(w+πτ ) −2iw−πτ i
cn e = −e cn e2niw
n=−∞ n=−∞
X∞ ∞
X
= −e−πτ i cn e2i(n−1)w = −e−πτ i cn+1 e2niw
n=−∞ n=−∞

we get the recurrent formula cn+1 = −e(2n+1)πτ i cn for the coefficients cn . We

choose the simplest initial condition c0 = 1. Let q = eπτ i . Since the sum
2
of 1, 3, · · · , 2n − 1 is n2 , it follows that cn = (−1)n q n . The reason for the
choice of a = −2i is to shift the exponent from e2niw to e2i(n−1)w so that
after changing the dummy index from n − 1 to n we end up with a formula
which relates cn+1 to cn . The choice of having −πiτ in b = −πiτ + π is to
change the factor e2niπτ i due to w 7→ w + πτ to e(2n+1)πτ i so that the sum
1, 3, · · · , 2n − 1 from the recurrent formula becomes the simpler expression
n2 . The choice of having π in b = −πiτ + π simply adds the factor (−1)n .

This way we arrive at the function

∞
2
X
(−1)n q n e2niw
n=−∞

and we denote it by ϑ(w) (also by ϑ(w|τ ) to emphasize its dependence on

τ ). It is the simplest Jacobi theta function. The transformation equations
for ϑ by translation by π and πτ are
ϑ (w + π) = ϑ (w) , ϑ(w + πτ ) = −e−2iw−iπτ ϑ(w).
We would like to remark why we use w and not z as the independent variable.
The reason is that the elliptic functions are defined as inverse functions and
so w which is frequently chosen as the dependent variable is used as the
independent variable.

Convergence and Number of Zeros of Constructed Jacobian Theta Func-

tion. First of all we observe that the infinite series converges, because

|q| = eπτ i  = e−πIm τ < 1

 

and its exponent n2 grows fast enough as a function of n to handle the growth
of e2niw whose exponent depends only linearly on n. We now determine the
Math 213a (Fall 2021) Yum-Tong Siu 4

number of zeroes of this entire function ϑ in a fundamental parallelogram by

using the argument principle for a fundamental parallelogram and integrating
d log ϑ along its boundary. Since π is actually a period for ϑ, the integral of
d log ϑ along the line segments − [0, πτ ] and [π, π + πτ ] cancel out, because
of the difference in orientation.

The integral of d log ϑ along the line segments [0, π] and − [πτ, π + πτ ] is
equal to the length of [0, π] times 2i which is the negative of the derivative
of the exponent of the periodicity factor for πτ . Thus the net result of
integrating d log ϑ along the boundary of the fundamental parallelogram with
vertices 0, π, π + πτ, πτ is 2πi. According to the argument principle, we have
precisely one zero for ϑ in a fundamental parallelogram. However, at this
point we do not know where the zero is. The zero can be located in the
following way.

In terms of trigonometric functions ϑ(w) is given by

∞
2
X
ϑ(w) = 1 + 2 (−1)n q n cos 2nw,
n=1

2
because both (−1)n and q n are unchanged under n 7→ −n. It indicates that
perhaps by changing in some way the cosine function to the sine function,
one may say something about where the zero is from the vanishing of the sine
function at 0. The sine function can be obtained by translating the cosine
function by π2 , which is a half-period in the lattice Zπ + Zπτ of periods.
This motivates the consideration of applying half-period translations to ϑ(w).
There are three ways of applying half-period translations, by translating by
π πτ
2
, 2 and π2 + πτ2
. We end with three more Jacobian theta functions.

Translations by Half-Periods. By translating the Jacobi theta function

ϑ(w) by one of the three half-periods in a fundamental parallelogram we get
three other functions ϑ1 , ϑ2 , ϑ3 after we use a factor which is the exponential
of a polynomial of degree ≤ 1 to simplify the coefficient of the nonzero term
Math 213a (Fall 2021) Yum-Tong Siu 5

e2niw with the smallest |n|. We obtain the following three new theta functions.
    ∞
1 1 1 2
(−1)n q (n+ 2 ) sin(2n + 1)w,
X
ϑ1 (w) = −i exp iw + πiτ ϑ w + πτ = 2
4 2 n=0
∞
 π 1 2
q (n+ 2 ) cos(2n + 1)w,
X
ϑ2 (w) = ϑ1 w + =2
2 n=0
∞
 π X 2
ϑ3 (w) = ϑ w + =1+2 q n cos 2nw.
2 n=1

A fractional power q α means eαπiτ . Here we use trigonometric functions

instead of the exponential function in the series expansion to help us see the
origin as the zero in θ1 (w).

We denote our original function ϑ(w) by ϑ4 (w), because the zero of ϑ1 (w)
is at w = 0 and the numbering corresponds to the zero assuming the half-
period points in the counter-clockwise sense. These four functions are all
called the Jacobian theta functions.

Degenerate Theta Functions as Modulus Goes to Infinity. One advantage

of using the trigonometric functions in the series expansion of ϑν (w) is that
we can more readily see what the degenerate case τ = +∞ i corresponds to.
In the degenerate case τ = +∞ i the limit
1 −1/4
lim q ϑν (w)
τ →+∞i 2

is simply sin w when ν = 1 and is simply cos w when ν = 2. When τ = +∞ i,

the two other Jacobian theta functions ϑ3 (w) and ϑ4 (w) become identically
1.

The sine function plays two roles when the period lattice degenerates to
rank one. One role is that the sine function is a degenerate elliptic function
for the lattice Z2π. The other role is that the sine function is the limit of a
constant multiple of the first Jacobian theta function for the period lattice
degenerates to the lattice Zπ when τ = +∞ i. Note that the lattices in both
rôles are different. The reason for the difference of the two lattices is that
the nonzero factor  
1
−i exp iw + πiτ
4
Math 213a (Fall 2021) Yum-Tong Siu 6

has been added to simplify the expression for ϑ1 (w).

Table of Periodicity Factors of Four Jacobian Theta Functions. The peri-

odicity factors of the four Jacobian theta functions are given in the following
table.
ϑ1 (w) ϑ2 (w) ϑ3 (w) ϑ4 (w)
π −1 −1 1 1
πτ −N N N −N
Here
1 −2iw
N= e = e−2iw−πiτ .
q
The reason why π is not a period (without periodicity factors) is again be-
cause of the addition of the nonzero factor
 
1
−i exp iw + πiτ
4

to simplify the expression for ϑ1 (w). One use of the table of periodicity
factors is that we can readily see from it how to take the square of the
quotient of two Jacobian theta functions to get a doubly periodic function
with respect to the lattice Zπ + Zπτ .

Three Ways of Using Table of Periodicity Factors of Four Jacobian

Theta Functions. We will use the table of periodicity factors in three ways
in deriving the following formulas.

(i) The elliptic-function analogue of sin2 θ + cos2 θ = 1.

(ii) The addition formula for theta functions, especially the formula for
doubling the variable.

(iii) The formula expression the Jacobian elliptic functions as quotients of

the Jacobian theta functions (after rescaling by values of the Jacobian
theta functions at the origin, known as theta constants).

For the first use, from the table of periodic factors ϑj (w)2 for 1 ≤ j ≤ 4
all the same period 1 and N 2 for π and πτ respectively so that quotients of
C-linear combinations of them are doubly periodic functions.
Math 213a (Fall 2021) Yum-Tong Siu 7

From the second use, ϑj (w)2 and ϑk (w+z)ϑk (w−z) have the same periods
1 and q12 e−4iw for π and πτ as functions of w for fixed z, so that quotients
of their C-linear combinations are doubly periodic functions. From the third
use, ϑϑ41 (w)
(w)
and ϑ2ϑ(w)ϑ 3 (w)
4 (w)
4 have the same periods −1 and 1 for π and πτ , so
that the quotient of the derivative of the former by the latter is a doubly
periodic function, because the derivative of a function with constant periodic
factors again has the same periodicity factors.

Terminology of Periodicity Factor. We are referring to the factor eaν w+bν

in
f (w + ων ) = eaν w+bν f (w)
as a periodicity factor when f (w) is a function on C and ω1 , ω2 are R-linearly
independent complex numbers. In the literature such a factor is also called
a factor of automorphy. We would like to comment on the terminology.
When we study a function f with two primitive periods ω1 , ω2 , it can be
considered as a function on the compact Riemann surface C /(Zω1 + Zω2 )
of genus 1. The question arises about what happens to functions with more
symmetry properties than having two primitive periods and how they should
correspondingly regarded as functions on a compact Riemann surface of genus
> 1.

According to the uniformization theorem which we will not discuss until

next term in Math 213b, unlike the case of a compact Riemann surface of
genus 1 whose universal cover is C, the universal cover of a compact Riemann
surface of genus > 1 is the open unit disk D. Unlike the case of a compact
Riemann surface of genus 1 which is the quotient C/L of C by a period lattice
L = Zω1 + Zω2 , a compact Riemann surface of genus > 1 is the quotient
D/Γ of the open unit disk D by the action of a discrete subgroup Γ of the
biholomorphism group Aut(D). For a formulation which works for compact
Riemann surfaces of all genus, we re-interpret a period ω for a function f as
invariance of the function f under the action of translation w 7→ w + ω in C.
The higher-genus analogue of a doubly periodic function is a function f on D
which is invariant under the action of every element of the discrete subgroup
Γ of Aut(D). Such a function is called an automorphic function with respect
to Γ. correspondingly regarded as functions on a compact Riemann surface
of genus > 1.

When the situation is like a theta function requiring a factor in the for-
Math 213a (Fall 2021) Yum-Tong Siu 8

mula for invariance, we call such a function f on D (or more generally a

complex manifold X) an automorphic form with respect to a discrete sub-
group Γ of the biholomorphism group or D (or of X) if f (γ(w)) = χγ (w)f (w)
for γ ∈ Γ, where χγ is a nowhere zero holomorphic function on D (or on X)
and is called a factor of automorphy. That is why in the literature a factor
such as eaν w+bν in
f (w + ων ) = eaν w+bν f (w)
is also called a factor of automorphy. We follow the practice of some older
textbooks such as the book A course of modern analysis by Whittaker and
Watson (Cambridge University Press 1973) and use the older terminology of
periodicity factor (line 5 from the bottom of p.463 of the book of Whittaker-
Watson).

Removing One Pole Order from Double Pole of Linear Combination to

Get Identity Similar to sin2 θ +cos2 θ = 1. In the degenerate case τ = +∞ i
after renormalization, the function ϑ1 (w) becomes sin w and ϑ2 (w) becomes
cos w, where as both ϑ3 and ϑ4 become 1. For trigonometric functions we have
the identity cos2 w+sin2 w ≡ 1. The question arises whether there already are
analogous identities among the Jacobian theta functions ϑ1 , ϑ2 , ϑ3 , ϑ4 before
we go to the degenerate case. The answer is affirmative and these identities
are derived by the technique of taking linear combinations to reduce one
order from the double pole of an elliptic function, thereby forcing the result
to be constant. For example, Two of these analogous algebraic relations are
ϑ2 (w)2 ϑ4 (0)2 − ϑ4 (w)2 ϑ2 (0)2 = −ϑ1 (w)2 ϑ3 (0)2 ,
ϑ3 (w)2 ϑ4 (0)2 − ϑ4 (w)2 ϑ3 (0)2 = −ϑ1 (w)2 ϑ2 (0)2 .

Derivation of Identity Similar to sin2 θ + cos2 θ = 1. We are going to derive

the first one. The derivation of the other and also the other similar identities
are completely analogous. First of all, from the table of periodicity factors
given above we note that
ϑ2 (w)2 ϑ4 (w)2
and
ϑ1 (w)2 ϑ1 (w)2
are both elliptic functions. They have at most double poles at the zeroes of
ϑ1 (w). We can reduce the order of the double poe by 1 by considering
a ϑ2 (w)2 + b ϑ4 (w)2
ϑ1 (w)2
Math 213a (Fall 2021) Yum-Tong Siu 9

so that a ϑ2 (w)2 + b ϑ4 (w)2 vanishes at the zeroes of ϑ1 (w).

The simplest choices for a and b are respectively ϑ4 (0)2 and −ϑ2 (0)2 .
Hence we know that
ϑ4 (0)2 ϑ2 (w)2 − ϑ2 (0)2 ϑ4 (w)2
ϑ1 (w)2
is an elliptic function having at most a simple pole in a fundamental paral-
lelogram and so must be a constant function. We still have to determine this
constant. We do this by evaluating the function at the point πτ 2
. At that
point ϑ4 (w) is zero. Moreover,
 πτ   
π + πτ π+πτ 1
π 
ϑ2 = ϑ1 = −iei( 2 )+ 4 πiτ ϑ4 + πτ
2 2 2
π+πτ 1 π+πτ 1 1
= −iei( 2 )+ 4 πiτ ϑ3 (πτ ) = −iei( 2 )+ 4 πiτ e−2iπτ ϑ3 (0) = q − 4 ϑ3 (0)
1

q
1
Likewise ϑ1 ( πτ
2
) = iq − 4 ϑ4 (0). Hence the constant is −ϑ3 (0)2 .

Two Other Identities by Translation by Half Period. By replacing w

by w + π2 in the above two algebraic relations we get the following two new
algebraic relations

ϑ1 (w)2 ϑ4 (0)2 − ϑ3 (w)2 ϑ2 (0)2 = −ϑ2 (w)2 ϑ3 (0)2 ,

ϑ4 (w)2 ϑ4 (0)2 − ϑ3 (w)2 ϑ3 (0)2 = −ϑ2 (w)2 ϑ2 (0)2 .

We now put together these four identities similar to sin2 θ + cos2 θ = 1.

ϑ1 (w)2 ϑ4 (0)2 − ϑ3 (w)2 ϑ2 (0)2 = −ϑ2 (w)2 ϑ3 (0)2 ,

ϑ2 (w)2 ϑ4 (0)2 − ϑ4 (w)2 ϑ2 (0)2 = −ϑ1 (w)2 ϑ3 (0)2 ,
ϑ3 (w)2 ϑ4 (0)2 − ϑ4 (w)2 ϑ3 (0)2 = −ϑ1 (w)2 ϑ2 (0)2 ,
ϑ4 (w)2 ϑ4 (0)2 − ϑ3 (w)2 ϑ3 (0)2 = −ϑ2 (w)2 ϑ2 (0)2 .

Addition Theorem for Jacobian Theta Functions. ϑ3 (w + z)ϑ3 (w − z) as

a function of w (for fixed z) has periodicity factor 1 and
1 −2i(w+z) 1 −2i(w−z) 1
e e = 2 e−4iw
q q q
Math 213a (Fall 2021) Yum-Tong Siu 10

for π and πτ . For any a(z), b(z) ∈ C (which depends on z) the periodicity
factors of a(z)ϑ3 (w)2 +b(z)ϑ1 (w)2 are also 1 and q12 e−4iw for π and πτ . Hence
the quotient
a(z)ϑ3 (w)2 + b(z)ϑ1 (w)2
ϑ3 (w + z)ϑ3 (w − z)
is a doubly periodic function.

For fixed z we can choose the two z-dependent constants a(z) and b(z)
such that one of the two simple zeroes of the doubly periodic function

a(z)ϑ3 (w)2 + b(z)ϑ1 (w)2

ϑ3 (w + z)ϑ3 (w − z)

of w in a fundamental parallelogram is cancelled by a zero of a(z)ϑ3 (w)2 +

b(z)ϑ1 (w)2 in w so that at most one simple pole of the doubly periodic
function
a(z)ϑ3 (w)2 + b(z)ϑ1 (w)2
ϑ3 (w + z)ϑ3 (w − z)
of w in a fundamental parallelogram can remain. This means that the doubly
periodic function
a(z)ϑ3 (w)2 + b(z)ϑ1 (w)2
ϑ3 (w + z)ϑ3 (w − z)
of w is a z-dependent constant.

By replacing a(z) and b(z) by λ(z)a(z) and λ(z)b(z) for some z-dependent
constant λ(z), we can assume that

a(z)ϑ3 (w)2 + b(z)ϑ1 (w)2

≡1
ϑ3 (w + z)ϑ3 (w − z)

so that
a(z)ϑ3 (w)2 + b(z)ϑ1 (w)2 ≡ ϑ3 (w + z)ϑ3 (w − z)
as functions of w. To determine a(z), we set w = 0 to get

a(z)ϑ3 (0)2 = ϑ3 (z)2 ,

because ϑ1 (w) vanishes at w = 0 and ϑ3 (w) is an even function of w.

Math 213a (Fall 2021) Yum-Tong Siu 11

π πτ π πτ


To determine b(z), we set w = 2
+ 2
and use ϑ3 2
+ 2
to get
π πτ 2  π πτ   π πτ 
b(z)ϑ1 + = ϑ3 + + z ϑ3 + −z .
2 2 2 2 2 2
ϑ1 (z)2
Hence b(z) = ϑ1 (0)2
and

ϑ3 (w + z)ϑ3 (w − z)ϑ3 (0)2 = ϑ3 (w)2 ϑ3 (z)2 + ϑ1 (w)2 ϑ1 (z)2

when we use the relations among the four Jacobian theta functions by trans-
lations by half-periods. This is the addition theorem for Jacobian theta
functions. By translation by half-periods, we can get similar addition the-
orems for other Jacobian theta functions. From this we can get a formula
for ϑ3 (2w) by setting z = w. In a moment we will see that Jacobian elliptic
functions are quotients of Jacobian theta functions.

Reduction of Fundamental Parallelogram by One-Half. We introduce

the Jacobian theta functions ϑ1 , ϑ2 , ϑ3 , ϑ4 as multiplicative building blocks
for elliptic functions. The question arises, for example, how to express
the Jacobian elliptic function sn w in terms of the Jacobian theta functions
ϑ1 , ϑ2 , ϑ3 , ϑ4 . We are going to do this by showing that a certain quotient of
products of the Jacobian theta functions ϑ1 , ϑ2 , ϑ3 , ϑ4 satisfies the differential
equation which defines sn w.

The technique of getting such a differential equation is to construct an ex-

pression of the Jacobian theta functions which is not only an elliptic function
with only two simple poles in a fundamental parallelogram, but also admits
one half-period of the fundamental parallelogram as its period. This has the
effect of reducing by one-half the fundamental parallelogram so that there is
only a single simple pole in the new fundamental parallelogram, forcing the
constructed expression to be constant.

Three Basic Jacobian Elliptic Functions as Quotients of Jacobian Theta

Functions. We now use this method to express sn w, cn w, dn w in terms of
quotients of Jacobian theta functions. We use the table of periodicity factors.
The quotients
ϑ1 (w) ϑ2 (w)ϑ3 (w)
and
ϑ4 (w) ϑ4 (w)2
Math 213a (Fall 2021) Yum-Tong Siu 12

both have periodicity factors −1 and 1 for the periods π and πτ . Their
derivatives would have the same periodicity factors. The derivative of

ϑ2 (w)ϑ3 (w)
ϑ4 (w)2

is more complicated. So we look at the derivative of

ϑ1 (w) ϑ01 (w)ϑ4 (w) − ϑ04 (w)ϑ1 (w)

which is .
ϑ4 (w) ϑ4 (w)2

To get rid of all the periodicity factors completely we take the quotient of

ϑ01 (w)ϑ4 (w) − ϑ04 (w)ϑ1 (w)

ϑ4 (w)2

by
ϑ2 (w)ϑ3 (w)
ϑ4 (w)2
and get the elliptic function

ϑ01 (w)ϑ4 (w) − ϑ04 (w)ϑ1 (w)

.
ϑ2 (w)ϑ3 (w)

Let us call this elliptic function ϕ. The only possible poles of ϕ are the zeroes
of ϑ2 (w)ϑ3 (w) and so they are at most simple poles at points congruent to
π
2
and π+πτ
2
. We claim that ϕ is constant.

The key point is to look at the function obtained by translating ϕ along

an appropriate half-period. The two Jacobian theta functions ϑ1 and ϑ4 in
the numerator are related by translation by the half-period πτ . The two
Jacobian theta functions ϑ2 and ϑ3 in the numerator are also related by
translation by the half-period πτ . This suggests that we should consider
the effect on ϕ of the translation by the half-period πτ . It turns out that
translation along the half-period 21 πτ leaves ϕ invariant. The verification is
simple and straightforward. This means the primitive periods of ϕ are π
and 21 πτ . In the fundamental parallelogram defined by these two primitive
periods ϕ has at most a simple pole at points congruent to π2 . Since the sum
of the residues of an elliptic function in a fundamental parallelogram must
be zero, we know that ϕ has no pole and must be a constant.
Math 213a (Fall 2021) Yum-Tong Siu 13

Evaluating ϕ at the origin, we conclude that the constant is

ϑ01 (0)ϑ4 (0)

.
ϑ2 (0)ϑ3 (0)

We have therefore the following differential equation for the Jacobian theta
functions
d ϑ1 (w) ϑ2 (w)ϑ3 (w)
=A ,
dw ϑ4 (w) ϑ4 (w)2
where the constant A is
ϑ01 (0)ϑ4 (0)
.
ϑ2 (0)ϑ3 (0)
Let
ϑ1 (w)
ξ(w) = .
ϑ4 (w)
From the two algebraic relations

ϑ2 (w)2 ϑ4 (0)2 − ϑ4 (w)2 ϑ2 (0)2 = −ϑ1 (w)2 ϑ3 (0)2 ,

ϑ3 (w)2 ϑ4 (0)2 − ϑ4 (w)2 ϑ3 (0)2 = −ϑ1 (w)2 ϑ2 (0)2

we have the following differential equation for some constant B.

 2
dξ
= B ϑ2 (0)2 − ξ 2 ϑ3 (0)2 ϑ3 (0)2 − ξ 2 ϑ2 (0)2 .



dw

We can reduce this to the form used to define sn w by letting

 
ϑ3 (0) w
x(w) = ξ
ϑ2 (0) ϑ3 (0)2

and get  2
dx
= C 2 1 − x2 1 − k 2 x2



dw
where C is a constant and
 2
ϑ2 (0)
k= .
ϑ3 (0)
Math 213a (Fall 2021) Yum-Tong Siu 14

dx
We claim that C is 1. For this purpose we have to show that dw
= 1 at
w = 0. Since  
w
ϑ
ϑ3 (0) 1 ϑ23
x=  
ϑ2 (0) ϑ w2
4 ϑ3

and ϑ1 (0) = 0, this is equivalent to the identity

ϑ01 (0) = ϑ2 (0)ϑ3 (0)ϑ4 (0).

This is the identity for the derivative of the first Jacobian theta functions at
the origin in terms of the values of the other Jacobian theta functions at the
origin. The value of a Jacobian theta function at the origin is also known as
a Jacobian theta constant. We will establish this identity below.

After the establishment of this identity we have

 
w
ϑ
ϑ3 (0) 1 ϑ3 (0)2
sn(w, k) =  .
ϑ2 (0) ϑ w
2 4 ϑ3 (0)

From the four algebraic identities

ϑ1 (w)2 ϑ4 (0)2 − ϑ3 (w)2 ϑ2 (0)2 = −ϑ2 (w)2 ϑ3 (0)2 ,

ϑ2 (w)2 ϑ4 (0)2 − ϑ4 (w)2 ϑ2 (0)2 = −ϑ1 (w)2 ϑ3 (0)2 ,
ϑ3 (w)2 ϑ4 (0)2 − ϑ4 (w)2 ϑ3 (0)2 = −ϑ1 (w)2 ϑ2 (0)2 ,
ϑ4 (w)2 ϑ4 (0)2 − ϑ3 (w)2 ϑ3 (0)2 = −ϑ2 (w)2 ϑ2 (0)2 .

we get
 
w w
ϑ4 (0) ϑ2 ( ϑ3 (0)2 ) ϑ4 (0) ϑ3 ϑ3 (0)2
cn(w, k) = and dn(w, k) = .
ϑ2 (0) ϑ4 ( ϑ3w(0)2 )

ϑ3 (0) ϑ w
4 ϑ3 (0)2

Relation Between Lattices Z(4K) + Z(2iK 0 ) and Zπ + Zπτ . The modulus

 2
ϑ2 (0)
k=
ϑ3 (0)
Math 213a (Fall 2021) Yum-Tong Siu 15

is not the modulus of the lattice Zπ + Zπτ . Let us determine the values K
and K 0 for k in terms of τ and ϑ3 (0). We have
w
ϑ4 (0) ϑ2 ( ϑ3 (0)2 )
cn(w, k) = .
ϑ2 (0) ϑ4 ( ϑ2w(0) )
3

The periodicity factors of

ϑ2 ( ϑ3w(0)2 )
ϑ4 ( ϑ3w(0)2 )
are −1 for both πϑ3 (0)2 and πτ ϑ3 (0)2 .
We also know that cn w has periodicity factors −1 for both 2K and 2iK 0 .
In a fundamental parallelogram of 2KZ + 2iK 0 Z the function cn w has only
one zero. Also in the fundamental parallelogram of ϑ3 (0)2 (πZ + πτ Z) the
function  
w
ϑ2 ϑ3 (0)2
 
ϑ4 ϑ3w(0)2
has only one zero. Thus we know that πϑ3 (0)2 and πτ ϑ3 (0)2 generate the
same lattice as 2K and 2iK 0 . Consider the case of a purely imaginary τ
with Im τ sufficiently large. Then both ϑ2 (0) and ϑ3 (0) are positive and
ϑ2 (0) < ϑ3 (0). Thus
ϑ2 (0)2
k=
ϑ3 (0)2
is positive and less than 1.
Since both 2K and πϑ3 (0)2 are positive and both 2iK 0 and πτ ϑ3 (0)2 are
on the positive y-axis, from the fact that
2KZ + 2iK 0 Z = ϑ3 (0)2 (πZ + πτ Z)
we conclude that 2K = πϑ3 (0)2 and 2iK 0 = πτ ϑ3 (0)2 for this special case of
a purely imaginary τ with Im τ sufficiently large.
By analytic continuation we know that they must be the same for all
values of τ . The primitive periods of
 
w
ϑ3 (0) ϑ1 ϑ3 (0)2
sn(w, k) =  
ϑ2 (0) ϑ w
2 4 ϑ3 (0)
Math 213a (Fall 2021) Yum-Tong Siu 16

are 4K = 2π ϑ3 (0)2 and 2iK 0 = πτ ϑ3 (0)2 . When the period lattice Zπ + Zπτ
degenerates to the lattice Zπ, sn(w, k) degenerates to the function sin w with
period 2π.

Heat-Equation Type Differential Equations. One remaining task is to

verify the identity ϑ01 (0) = ϑ2 (0)ϑ3 (0)ϑ4 (0). This identity in the degenerate
case of τ = +∞i is simply the statement that the derivative of the sine
function at zero agrees with the value of the cosine function at zero. We will
establish the identity ϑ01 (0) = ϑ2 (0)ϑ3 (0)ϑ4 (0) by using the heat-equation-
type identity for the four elliptic theta functions. The differential equation
is an easy consequence of the infinite series expansion of the Jacobian theta
functions whose exponent for the n-th term has factor n2 in the coefficient for
τ and only the factor n in the coefficient of w. We will derive the differential
equation for ϑ4 and its derivation for the other three Jacobian theta functions
are completely analogous.
From
∞ ∞
n n2 2niw 2 πiτ +2niw
X X
ϑ4 (w|τ ) = (−1) q e = (−1)n en
n=−∞ n=−∞

it follows that
∞
∂ 2 ϑ4 (w|τ ) X 2 4 ∂ ϑ4 (w|τ )
2
= (−1)n+1 4n2 en πiτ +2niw = − .
∂w n=−∞
πi ∂τ
All four Jacobian theta functions ϑ1 , ϑ2 , ϑ3 , ϑ4 satisfy the heat-equation type
differential equation
πi ∂ 2 f ∂f
+ =0
4 ∂w2 ∂τ
in f.

Second Derivatives of Logarithms of Jacobian Theta Functions. We now

look at a relation among
 2   2   2   2 
∂ 0 ∂ ∂ ∂
2
ϑ1 (0), 2
ϑ2 (0), 2
ϑ3 (0), ϑ4 (0)
∂w ∂w ∂w ∂w2
by constructing a trivial theta function from the Jacobian theta functions.
This relation will be
ϑ000
1 (0) ϑ002 (0) ϑ003 (0) ϑ004 (0)
= + + ,
ϑ01 (0) ϑ2 (0) ϑ3 (0) ϑ4 (0)
Math 213a (Fall 2021) Yum-Tong Siu 17

which is an intermediate step needed to use the heat-equation type differential

equation
πi ∂ 2 f ∂f
2
+ =0
4 ∂w ∂τ
for f = ϑ1 , ϑ2 , ϑ3 , ϑ4 to derive the identity ϑ01 (0) = ϑ2 (0)ϑ3 (0)ϑ4 (0).

Last Step in Proof of the Identity ϑ01 (0) = ϑ2 (0)ϑ3 (0)ϑ4 (0). A trivial
theta functions means a holomorphic function on C without zeroes whose
periodicity factors are exponential functions of polynomials of degree ≤ 1,
which automatically means that it is the exponential function of a polynomial
of degree ≤ 2, because if the difference of an entire function and its translate
by any element of a maximum-rank lattice is a polynomial of degree ≤ 1,
then the entire function must be a polynomial of degree ≤ 2.
This trivial theta function is
ϑ1 (w)ϑ2 (w)ϑ3 (w)ϑ4 (w)
f (w) = .
ϑ1 (2w)
The function has no zero and no pole, because the four zeroes of ϑ1 (2w)
in the parallelogram spanned by π and πτ are precisely the four zeroes of
ϑ1 (w)ϑ2 (w)ϑ3 (w)ϑ4 (w). Now f (w) is an entire function with periodicity
factors 1 for π and q 2 N 2 for πτ .
We are going to use the fact that f (w) has no zero to conclude that it is
a trivial theta function in the sense that f (w) = exp(aw2 + bw + c) for some
complex numbers a, b, c. Choose a branch log f (w) on C.
Since f (w + π) = f (w) and f (w + πτ ) = q 2 N 2 f (w) with N being a linear
d2
function of w, it follows that the entire function dw 2 log f (w) has periods π

and πτ and thus must be constant. We can conclude by integrating and

exponentiating that f (w) = exp(aw2 + b w + c) for some complex numbers
a, b, c.
Now we use the fact that f (w + π) = f (w) to conclude that
exp(a(w + π)2 + b(w + π) + c) = exp(aw2 + bw + c)
from which we get exp(2awπ + π 2 + bπ) = 1 for all w and hence a = 0. Thus
f (w) = exp(bw + c) and
d2
log f
dw2
Math 213a (Fall 2021) Yum-Tong Siu 18

is identically zero. In particular,

d2
 
log f (0)
dw2
is zero.

From the evenness of ϑν (w) we have the vanishing of ϑ0ν (0) for 2 ≤ ν ≤ 4.
Hence for 2 ≤ ν ≤ 4,
ϑ00ν (0)
 2 
d
log ϑν (0) = .
dw2 ϑν (0)
Let
ϑ1 (w) = α(w + β w3 + · · · )
be the power series expansion of the odd entire function ϑ1 (w). Let
ϑ1 (w)
g(w) = .
ϑ1 (2w)
From
α (w + βw3 + · · · )
g(w) =
α 2w + β (2w)3 + · · ·

1 + βw2 + · · · 1
1 − 3βw2 + · · ·


= 2
=
2 (1 + 4βw + · · · ) 2
it follows that
g 00 (0) (g 0 (0))2 ϑ000
 2 
d 1 (0)
log g (0) = − = −6β = − .
dw 2 g(0) (g(0)) 2
ϑ01 (0)
Since
d2 d2
 
ϑ1 (w)ϑ2 (w)ϑ3 (w)ϑ4 (w)
0≡ log f (w) = log
dw2 dw2 ϑ1 (2w)
2 2 2
d d d d2
= log g + log ϑ2 + log ϑ 3 + log ϑ4 ,
dw2 dw2 dw2 dw2
it follows from evaluation at w = 0 that
ϑ000
1 (0) ϑ002 (0) ϑ003 (0) ϑ004 (0)
= + + .
ϑ01 (0) ϑ2 (0) ϑ3 (0) ϑ4 (0)
Math 213a (Fall 2021) Yum-Tong Siu 19

Using the heat-equation type differential equation, we obtain from it

1 dϑ01 (0|τ ) 1 dϑ2 (0|τ ) 1 dϑ3 (0|τ ) 1 dϑ4 (0|τ )
0 = + + .
ϑ1 (0|τ ) dτ ϑ2 (0|τ ) dτ ϑ3 (0|τ ) dτ ϑ4 (0|τ ) dτ
Integrating with respect to τ we get
ϑ01 (0, q) = Eϑ2 (0, q)ϑ3 (0, q)ϑ4 (0, q),
where E is a constant independent of q.

From the infinite series expansions of the Jacobian theta functions we

have
1 1
lim q − 4 ϑ01 (0, q) = 2, lim q − 4 ϑ2 (0, q) = 2,
q→0 q→0
lim ϑ3 (0, q) = 1, lim ϑ4 (0, q) = 1.
q→0 q→0

Hence E = 1 and ϑ01 (0) = ϑ2 (0)ϑ3 (0)ϑ4 (0). This proof of the identity ϑ01 (0) =
ϑ2 (0)ϑ3 (0)ϑ4 (0) is due to C. A. Stewart, which is used as Example 21 on p.490
of the book A Course of Modern Analysis by Whittaker and Watson

Jacobi’s Theta Transform. Jacobi’s theta transform relates the values of the
four Jacobian theta functions at (w, τ ) to those at (wτ 0 , τ 0 ), where τ 0 = − τ1 .
The four identities are
1 0 w2
ϑ1 (w, τ ) = −i(−iτ )− 2 eiτ π ϑ1 (wτ 0 , τ 0 ),
1 0 w2
ϑ2 (w, τ ) = (−iτ )− 2 eiτ π ϑ4 (wτ 0 , τ 0 ),
1 0 w2
ϑ3 (w, τ ) = (−iτ )− 2 eiτ π ϑ3 (wτ 0 , τ 0 ),
1 0 w2
ϑ4 (w, τ ) = (−iτ )− 2 eiτ π ϑ2 (wτ 0 , τ 0 ).
Since the other three Jacobian theta functions are obtained by translating
ϕ3 (w, τ ) by half-periods, the four identities can be easily generated by trans-
lation by half-periods when one of them is known.

Jacobi’s Theta Transform for ϑ3 . The one that is most commonly used is
the third one, which we are going to prove in the following form suppressing
τ 0. r
−iw2
   
τ w 1
ϑ3 (w, τ ) = exp ϑ3 ,− .
i πτ τ τ
Math 213a (Fall 2021) Yum-Tong Siu 20

We now prove it.

Proof of Jacobi’s theta transform for
ϑ3 . The lattice for ϑ3 (w, τ ) is w ∈

Z π + Z πτ and the lattice for ϑ3 wτ , − τ1 is wτ ∈ Z π + Z π τ1 which is the same
as w ∈ Z π + Z πτ . The function ϑ3 (w, τ ) has a single zero
at w = 12 π + 12 πτ
in the fundamental parallelogram. The function ϑ3 wτ , − τ1 has a single zero
at wτ = 12 π + 12 π τ1 in the fundamental parallelogram which is the same as a
single zero at w = 12 π + 21 πτ in the fundamental parallelogram. We have to
consider the transformation laws. We have ϑ3 (w + π, τ ) = ϑ3 (w, τ ) and

ϑ3 (w + πτ, τ ) = ϑ3 (w, τ ) exp (−2iw − iπτ ) .

The second one can be written also as

ϑ3 (w − πτ, τ ) = ϑ3 (w, τ ) exp(2iw − iπτ )

after we replace w by w − πτ . Hence

   
w+π 1 w  π 1
ϑ3 ,− = ϑ3 − − ,−
τ τ τ τ τ
   
w 1 w 1
= ϑ3 ,− exp 2i + iπ .
τ τ τ τ
In order to cancel the factor
 
w 1
exp 2i + iπ ,
τ τ
we use
−iw2
   
w 1
exp ϑ3 ,− .
πτ τ τ
We have to check also the transformation when w is replaced by w + πτ . In
that case    
w + πτ 1 w 1
ϑ3 ,− = ϑ3 ,−
τ τ τ τ
and
−i(w + πτ )2
   
w + πτ 1
exp ϑ3 ,−
πτ τ τ
2
   
−iw w 1
= exp (−2iw − iπτ ) exp ϑ3 ,− .
πτ τ τ
Math 213a (Fall 2021) Yum-Tong Siu 21

Hence we conclude that

−iw2
   
w 1
f (τ )ϑ3 (w, τ ) = exp ϑ3 ,−
πτ τ τ

for some constant f (τ ) which may depend on τ . To determine f (τ ), we use

the heat equation

πi ∂ 2 ∂
ϑ3 (w, τ ) + ϑ3 (w, τ ) = 0.
4 ∂w2 ∂τ
More precisely we consider the first-order differential operator

πi ∂ 2 

∂
L= 2
+
4 ∂w w=0 ∂τ


in the variable τ . Since

−iw2
   
w 1
f (τ )ϑ3 (w, τ ) − exp ϑ3 −
πτ τ τ

is identically zero, the function f (τ ) satisfies the ordinary differential equa-

tion it follows that
−iw2
    
w 1
L f (τ )ϑ3 (w, τ ) − exp ϑ3 ,− =
πτ τ τ
−iw2
    
1 w 1
f (τ )ϑ3 (w, τ ) − exp ϑ3 ,−
2τ πτ τ τ

with τ as the variable.

p τ We explicitly compute the case when the function f (τ ) is replaced by

i
and verify that the above differential equation is satisfied when f (τ ) is
replaced by τi . In the verification we use the fact that
p

πi ∂ 2 ∂
2
ϑ3 (w, τ ) + ϑ3 (w, τ ) = 0
4 ∂w ∂τ
when w = 0. We also verify that when τ = i and w = 0 we have
r
−iw2
   
τ w 1
ϑ3 (w, τ ) − exp ϑ3 ,− = 0.
i πτ τ τ
Math 213a (Fall 2021) Yum-Tong Siu 22

Relation Between Jacobian Theta Function and Weierstrass Sigma Func-

tion. The relation between the Weierstrass sigma function σ(w) for the pe-
riod lattice Zω1 + Zω2 and the first Jacobian theta function ϑ1 (w, τ ) is given
by the formula  
w ω2 η1ωw2
σ(w) = ϑ1 ,τ e 1 ,
ω1 ϑ03 (0)
where the constant η1 satisfies
ω1
σ(w + ω1 ) = −σ(w)e2η1 (w+ 2 ) .
This is checked by observing that the two entire functions on both sides
have a single zero at 0 in a fundamental parallelogram and have the same
periodicity factors. Moreover, they have the same derivative at 0.

Jacobi’s Fundamental Formulas for Jacobian Theta Functions. The

elliptic-function analogue of the identity sin2 θ + cos2 θ = 1 is a special case
of general Jacobi’s fundamental formulas for Jacobian theta functions which
are proved by using by using the following special orthogonal symmetric real
4 × 4 matrix.  
1 1 1 1
1 1 1 −1 −1
A=  
2 1 −1 1 −1
1 −1 −1 1
The matrix A satisfies the property that
4 4
X
2 X
n0j = (nj )2
j=1 j=1

holds for integers n1 , n2 , n3 , n4 and n01 , n02 , n03 , n04 related by

   0
n1 n1
n2  n02 
  = A 0.
n3  n3 
n4 n04
Moreover, when n1 , n2 , n3 , n4 are either all even integers or all odd integers,
then also n01 , n02 , n03 , n04 are all even integers or all odd integers. As a conse-
quence,
Y4 Y4 4
Y 4
Y
ϑ2 (wν ) + ϑ3 (wν ) = ϑ2 (wν0 ) + ϑ3 (wν0 )
ν=1 ν=1 ν=1 ν=1
Math 213a (Fall 2021) Yum-Tong Siu 23

holds if
w10
   
w1
w20  w2 
 0 = A .
w3  w3 
0
w4 w4
Q4
This is obtained by expanding ν=1 ϑ2 (wν ) as the product of four infinite
series of exponential terms and the other products of four Jacobian theta
functions in the equation. The details are left to a homework assignment.