The Weierstrass elliptic function can be defined in three closely related ways, each of which possesses certain advantages. One is as a function of a complex variable z and a lattice Λ in the complex plane. Another is in terms of z and two complex numbers ω_{1} and ω_{2} defining a pair of generators, or periods, for the lattice. The third is in terms of z and a modulus τ in the upper halfplane. This is related to the previous definition by τ = ω_{2}/ω_{1}, which by the conventional choice on the pair of periods is in the upper halfplane. Using this approach, for fixed z the Weierstrass functions become modular functions of τ.
In terms of the two periods, Weierstrass's elliptic function is an elliptic function with periods ω_{1} and ω_{2} defined as
℘
(
z
;
ω
1
,
ω
2
)
=
1
z
2
+
∑
n
2
+
m
2
≠
0
{
1
(
z
+
m
ω
1
+
n
ω
2
)
2
−
1
(
m
ω
1
+
n
ω
2
)
2
}
.
Then
Λ
=
{
m
ω
1
+
n
ω
2
:
m
,
n
∈
Z
}
are the points of the period lattice, so that
℘
(
z
;
Λ
)
=
℘
(
z
;
ω
1
,
ω
2
)
for any pair of generators of the lattice defines the Weierstrass function as a function of a complex variable and a lattice.
If
τ
is a complex number in the upper halfplane, then
℘
(
z
;
τ
)
=
℘
(
z
;
1
,
τ
)
=
1
z
2
+
∑
n
2
+
m
2
≠
0
{
1
(
z
+
m
+
n
τ
)
2
−
1
(
m
+
n
τ
)
2
}
.
The above sum is homogeneous of degree minus two, from which we may define the Weierstrass ℘ function for any pair of periods, as
℘
(
z
;
ω
1
,
ω
2
)
=
℘
(
z
ω
1
;
ω
2
ω
1
)
ω
1
2
.
We may compute ℘ very rapidly in terms of theta functions; because these converge so quickly, this is a more expeditious way of computing ℘ than the series we used to define it. The formula here is
℘
(
z
;
τ
)
=
π
2
ϑ
2
(
0
;
τ
)
ϑ
10
2
(
0
;
τ
)
ϑ
01
2
(
z
;
τ
)
ϑ
11
2
(
z
;
τ
)
−
π
2
3
[
ϑ
4
(
0
;
τ
)
+
ϑ
10
4
(
0
;
τ
)
]
There is a secondorder pole at each point of the period lattice (including the origin). With these definitions,
℘
(
z
)
is an even function and its derivative with respect to z, ℘′, is an odd function.
Further development of the theory of elliptic functions shows that Weierstrass's function is determined up to addition of a constant and multiplication by a nonzero constant by the position and type of the poles alone, amongst all meromorphic functions with the given period lattice.
In a deleted neighborhood of the origin, the Laurent series expansion of
℘
is
℘
(
z
;
ω
1
,
ω
2
)
=
z
−
2
+
1
20
g
2
z
2
+
1
28
g
3
z
4
+
O
(
z
6
)
where
g
2
=
60
∑
(
m
,
n
)
≠
(
0
,
0
)
(
m
ω
1
+
n
ω
2
)
−
4
g
3
=
140
∑
(
m
,
n
)
≠
(
0
,
0
)
(
m
ω
1
+
n
ω
2
)
−
6
.
The numbers g_{2} and g_{3} are known as the invariants. The summations after the coefficients 60 and 140 are the first two Eisenstein series, which are modular forms when considered as functions G_{4}(τ) and G_{6}(τ), respectively, of τ = ω_{2}/ω_{1} with Im(τ) > 0.
Note that g_{2} and g_{3} are homogeneous functions of degree −4 and −6; that is,
g
2
(
λ
ω
1
,
λ
ω
2
)
=
λ
−
4
g
2
(
ω
1
,
ω
2
)
g
3
(
λ
ω
1
,
λ
ω
2
)
=
λ
−
6
g
3
(
ω
1
,
ω
2
)
.
Thus, by convention, one frequently writes
g
2
and
g
3
in terms of the period ratio
τ
=
ω
2
/
ω
1
and take
τ
to lie in the upper halfplane. Thus,
g
2
(
τ
)
=
g
2
(
1
,
ω
2
/
ω
1
)
and
g
3
(
τ
)
=
g
3
(
1
,
ω
2
/
ω
1
)
.
The Fourier series for
g
2
and
g
3
can be written in terms of the square of the nome
q
=
exp
(
i
π
τ
)
as
g
2
(
τ
)
=
4
3
π
4
[
1
+
240
∑
k
=
1
∞
σ
3
(
k
)
q
2
k
]
g
3
(
τ
)
=
8
27
π
6
[
1
−
504
∑
k
=
1
∞
σ
5
(
k
)
q
2
k
]
where
σ
a
(
k
)
is the divisor function. This formula may be rewritten in terms of Lambert series.
The invariants may be expressed in terms of Jacobi's theta functions. This method is very convenient for numerical calculation: the theta functions converge very quickly. In the notation of Abramowitz and Stegun, but denoting the primitive periods by
ω
1
,
ω
2
, the invariants satisfy
g
2
(
ω
1
,
ω
2
)
=
4
3
(
π
ω
1
)
4
(
a
8
−
a
4
b
4
+
b
8
)
=
2
3
(
π
ω
1
)
4
(
a
8
+
b
8
+
c
8
)
g
3
(
ω
1
,
ω
2
)
=
8
27
(
π
ω
1
)
6
(
a
12
−
3
2
a
8
b
4
−
3
2
a
4
b
8
+
b
12
)
where
a
=
θ
2
(
0
;
q
)
=
ϑ
10
(
0
;
τ
)
b
=
θ
3
(
0
;
q
)
=
ϑ
00
(
0
;
τ
)
c
=
θ
4
(
0
;
q
)
=
ϑ
01
(
0
;
τ
)
and
τ
=
ω
2
/
ω
1
is the period ratio,
q
=
e
π
i
τ
is the nome, and
θ
m
and
ϑ
m
n
are alternative notations.
If the invariants are g_{2} = 0, g_{3} = 1, then this is known as the equianharmonic case; g_{2} = 1, g_{3} = 0 is the lemniscatic case.
With this notation, the ℘ function satisfies the following differential equation:
[
℘
′
(
z
)
]
2
=
4
[
℘
(
z
)
]
3
−
g
2
℘
(
z
)
−
g
3
,
where dependence on
ω
1
and
ω
2
is suppressed.
This relation can be quickly verified by comparing the poles of both sides, for example, the pole at z = 0 of lhs is
[
℘
′
(
z
)
]
2

z
=
0
∼
4
z
6
−
24
z
2
∑
1
(
m
ω
1
+
n
ω
2
)
4
−
80
∑
1
(
m
ω
1
+
n
ω
2
)
6
while the pole at z = 0 of
[
℘
(
z
)
]
3

z
=
0
∼
1
z
6
+
9
z
2
∑
1
(
m
ω
1
+
n
ω
2
)
4
+
15
∑
1
(
m
ω
1
+
n
ω
2
)
6
.
Comparing these two yields the relation above.
The Weierstrass elliptic function can be given as the inverse of an elliptic integral. Let
u
=
∫
y
∞
d
s
4
s
3
−
g
2
s
−
g
3
.
Here, g_{2} and g_{3} are taken as constants. Then one has
y
=
℘
(
u
)
.
The above follows directly by integrating the differential equation.
The modular discriminant Δ is defined as the quotient by 16 of the discriminant of the righthand side of the above differential equation:
Δ
=
g
2
3
−
27
g
3
2
.
This is studied in its own right, as a cusp form, in modular form theory (that is, as a function of the period lattice).
Note that
Δ
=
(
2
π
)
12
η
24
where
η
is the Dedekind eta function.
The presence of 24 can be understood by connection with other occurrences, as in the eta function and the Leech lattice.
The discriminant is a modular form of weight 12. That is, under the action of the modular group, it transforms as
Δ
(
a
τ
+
b
c
τ
+
d
)
=
(
c
τ
+
d
)
12
Δ
(
τ
)
with τ being the halfperiod ratio, and a,b,c and d being integers, with ad − bc = 1.
For the Fourier coefficients of
Δ
, see Ramanujan tau function.
Consider the cubic polynomial equation 4t^{3} − g_{2}t − g_{3} = 0 with roots e_{1}, e_{2}, and e_{3}. Its discriminant is 16 times the modular discriminant Δ = g_{2}^{3} − 27g_{3}^{2}. If it is not zero, no two of these roots are equal. Since the quadratic term of this cubic polynomial is zero, the roots are related by the equation
e
1
+
e
2
+
e
3
=
0.
The linear and constant coefficients (g_{2} and g_{3}, respectively) are related to the roots by the equations (see Elementary symmetric polynomial).
g
2
=
−
4
(
e
1
e
2
+
e
1
e
3
+
e
2
e
3
)
=
2
(
e
1
2
+
e
2
2
+
e
3
2
)
g
3
=
4
e
1
e
2
e
3
The roots e_{1}, e_{2}, and e_{3} of the equation
4
x
3
−
g
2
x
−
g
3
=
0
depend on τ and can be expressed in terms of theta functions. As before, let,
a
=
θ
2
(
0
;
e
π
i
τ
)
=
ϑ
10
(
0
;
τ
)
b
=
θ
3
(
0
;
e
π
i
τ
)
=
ϑ
00
(
0
;
τ
)
c
=
θ
4
(
0
;
e
π
i
τ
)
=
ϑ
01
(
0
;
τ
)
then
e
1
(
τ
)
=
π
2
3
(
b
4
+
c
4
)
e
2
(
τ
)
=
π
2
3
(
−
a
4
−
b
4
)
e
3
(
τ
)
=
π
2
3
(
a
4
−
c
4
)
Since
g
2
=
2
(
e
1
2
+
e
2
2
+
e
3
2
)
and
g
3
=
4
e
1
e
2
e
3
, then these can also be expressed as theta functions. In simplified form,
g
2
(
τ
)
=
2
3
π
4
(
a
8
+
b
8
+
c
8
)
g
3
(
τ
)
=
4
27
π
6
(
a
8
+
b
8
+
c
8
)
3
−
54
(
a
b
c
)
8
2
Δ
=
g
2
3
−
27
g
3
2
=
16
π
12
a
8
b
8
c
8
=
(
2
π
)
12
η
24
(
τ
)
Where
η
is the Dedekind eta function. In the case of real invariants, the sign of Δ = g_{2}^{3} − 27g_{3}^{2} determines the nature of the roots. If
Δ
>
0
, all three are real and it is conventional to name them so that
e
1
>
e
2
>
e
3
. If
Δ
<
0
, it is conventional to write
e
1
=
−
α
+
β
i
(where
α
≥
0
,
β
>
0
), whence
e
3
=
e
1
¯
, and
e
2
is real and nonnegative.
The halfperiods ω_{1}/2 and ω_{2}/2 of Weierstrass' elliptic function are related to the roots
℘
(
ω
1
2
)
=
e
1
℘
(
ω
2
2
)
=
e
2
℘
(
ω
3
2
)
=
e
3
where
ω
3
=
−
(
ω
1
+
ω
2
)
. Since the square of the derivative of Weierstrass' elliptic function equals the above cubic polynomial of the function's value,
℘
′
(
ω
i
2
)
2
=
℘
′
(
ω
i
2
)
=
0
for
i
=
1
,
2
,
3
. Conversely, if the function's value equals a root of the polynomial, the derivative is zero.
If g_{2} and g_{3} are real and Δ > 0, the e_{i} are all real, and
℘
(
)
is real on the perimeter of the rectangle with corners 0, ω_{3}, ω_{1} + ω_{3}, and ω_{1}. If the roots are ordered as above (e_{1} > e_{2} > e_{3}), then the first halfperiod is completely real
ω
1
2
=
∫
e
1
∞
d
z
4
z
3
−
g
2
z
−
g
3
whereas the third halfperiod is completely imaginary
ω
3
2
=
i
∫
−
e
3
∞
d
z
4
z
3
−
g
2
z
+
g
3
.
The Weierstrass elliptic functions have several properties that may be proved:
det
[
℘
(
z
)
℘
′
(
z
)
1
℘
(
y
)
℘
′
(
y
)
1
℘
(
z
+
y
)
−
℘
′
(
z
+
y
)
1
]
=
0
A symmetrical version of the same identity is
det
[
℘
(
u
)
℘
′
(
u
)
1
℘
(
v
)
℘
′
(
v
)
1
℘
(
w
)
℘
′
(
w
)
1
]
=
0
where
u
+
v
+
w
=
0.
Also
℘
(
z
+
y
)
=
1
4
{
℘
′
(
z
)
−
℘
′
(
y
)
℘
(
z
)
−
℘
(
y
)
}
2
−
℘
(
z
)
−
℘
(
y
)
and the duplication formula
℘
(
2
z
)
=
1
4
{
℘
″
(
z
)
℘
′
(
z
)
}
2
−
2
℘
(
z
)
,
unless 2z is a period.
If
ω
1
=
1
, much of the above theory becomes simpler; it is then conventional to write
τ
for
ω
2
. For a fixed τ in the upper halfplane, so that the imaginary part of τ is positive, we define the Weierstrass ℘ function by
℘
(
z
;
τ
)
=
1
z
2
+
∑
(
m
,
n
)
≠
(
0
,
0
)
1
(
z
+
m
+
n
τ
)
2
−
1
(
m
+
n
τ
)
2
.
The sum extends over the lattice {n+mτ : n and m in Z} with the origin omitted. Here we regard τ as fixed and ℘ as a function of z; fixing z and letting τ vary leads into the area of elliptic modular functions.
℘ is a meromorphic function in the complex plane with a double pole at each lattice point. It is doubly periodic with periods 1 and τ; this means that ℘ satisfies
℘
(
z
+
1
)
=
℘
(
z
+
τ
)
=
℘
(
z
)
.
The above sum is homogeneous of degree minus two, and if c is any nonzero complex number,
℘
(
c
z
;
c
τ
)
=
℘
(
z
;
τ
)
/
c
2
from which we may define the Weierstrass ℘ function for any pair of periods. We also may take the derivative (of course, with respect to z) and obtain a function algebraically related to ℘ by
℘
′
2
=
4
℘
3
−
g
2
℘
−
g
3
where
g
2
and
g
3
depend only on τ, being modular forms. The equation
Y
2
=
4
X
3
−
g
2
X
−
g
3
defines an elliptic curve, and we see that
(
℘
,
℘
′
)
is a parametrization of that curve. The totality of meromorphic doubly periodic functions with given periods defines an algebraic function field associated to that curve. It can be shown that this field is
C
(
℘
,
℘
′
)
,
so that all such functions are rational functions in the Weierstrass function and its derivative.
One can wrap a single period parallelogram into a torus, or donutshaped Riemann surface, and regard the elliptic functions associated to a given pair of periods to be functions defined on that Riemann surface.
℘ can also be expressed in terms of theta functions; because these converge very rapidly, this is a more expeditious way of computing ℘ than the series used to define it.
℘
(
z
;
τ
)
=
π
2
ϑ
2
(
0
;
τ
)
ϑ
10
2
(
0
;
τ
)
ϑ
01
2
(
z
;
τ
)
ϑ
11
2
(
z
;
τ
)
+
e
2
(
τ
)
.
The function ℘ has two zeros (modulo periods) and the function ℘′ has three. The zeros of ℘′ are easy to find: since ℘′ is an odd function they must be at the halfperiod points. On the other hand, it is very difficult to express the zeros of ℘ by closed formula, except for special values of the modulus (e.g. when the period lattice is the Gaussian integers). An expression was found, by Zagier and Eichler.
The Weierstrass theory also includes the Weierstrass zeta function, which is an indefinite integral of ℘ and not doubly periodic, and a theta function called the Weierstrass sigma function, of which his zetafunction is the logderivative. The sigmafunction has zeros at all the period points (only), and can be expressed in terms of Jacobi's functions. This gives one way to convert between Weierstrass and Jacobi notations.
The Weierstrass sigmafunction is an entire function; it played the role of 'typical' function in a theory of random entire functions of J. E. Littlewood.
For numerical work, it is often convenient to calculate the Weierstrass elliptic function in terms of Jacobi's elliptic functions. The basic relations are
℘
(
z
)
=
e
3
+
e
1
−
e
3
sn
2
w
=
e
2
+
(
e
1
−
e
3
)
dn
2
w
sn
2
w
=
e
1
+
(
e
1
−
e
3
)
cn
2
w
sn
2
w
where e_{1–3} are the three roots described above and where the modulus k of the Jacobi functions equals
k
≡
e
2
−
e
3
e
1
−
e
3
and their argument w equals
w
≡
z
e
1
−
e
3
.