Four Lectures On Weierstrass Elliptic Function and Applications in Classical and Quantum Mechanics

Four Lectures on Weierstrass Elliptic Function and
Applications in Classical and Quantum Mechanics

Georgios Pastras1
1
NCSR “Demokritos”, Institute of Nuclear and Particle Physics
15310 Aghia Paraskevi, Attiki, Greece
arXiv:1706.07371v2 [math-ph] 6 Sep 2017
[email protected]
Abstract
In these four lectures, aiming at senior undergraduate and junior graduate Physics
and Mathematics students, basic elements of the theory of elliptic functions are pre-
sented. Simple applications in classical mechanics are discussed, including a point
particle in a cubic, sinusoidal or hyperbolic potential, as well as simple applications in
quantum mechanics, namely the n = 1 Lamé potential. These lectures were given at
the School of Applied Mathematics and Physical Sciences of the National Technical
University of Athens in May 2017.
1
Contents
Lecture 1: Weierstrass Elliptic Function 3
1.1 Prologue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Elliptic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 The Weierstrass Elliptic Function . . . . . . . . . . . . . . . . . . . . . 10
Lecture 2: Weierstrass Quasi-periodic Functions 17

2.1 Quasi-periodic Weierstrass Functions . . . . . . . . . . . . . . . . . . . 17
2.2 Expression of any Elliptic Function in Terms of Weierstrass Functions 20
2.3 Addition Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Lecture 3: Applications in Classical Mechanics 28

3.1 Real Solutions of Weierstrass Equation in the Real Domain . . . . . . 28
3.2 Point Particle in a Cubic Potential . . . . . . . . . . . . . . . . . . . . 32
3.3 The Simple Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 Point Particle in Hyperbolic Potential . . . . . . . . . . . . . . . . . . 40
Lecture 4: Applications in Quantum Mechanics 48

4.1 The n = 1 Lamé Potential . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2 The Bounded n = 1 Lamé Potential . . . . . . . . . . . . . . . . . . . 54
4.3 Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
A The Half-Periods for Real Moduli 59
B Problem Solutions 64
2
Lecture 1: Weierstrass Elliptic Function
1.1 Prologue
In the present lectures, basic elements of the theory of elliptic functions are presented
and simple applications in classical and quantum mechanics are discussed. The lec-
tures target an audience of senior undergraduate or junior graduate Physics and Math-
ematics students. For lectures 1 and 2, the audience is required to know basic complex
calculus including Cauchy’s residue theorem. For lecture 3, basic knowledge on clas-
sical mechanics is required. For lecture 4, the audience is required to be familiar with
basic quantum mechanics, including Schrödinger ’s equation and Bloch’s theorem.
The original constructions of elliptic functions are due to Weierstrass [1] and Jacobi
[2]. In these lectures, we focus on the former. Excellent pedagogical texts on the
subject of elliptic functions are the classic text by Watson and Whittaker[3] and the
more specialized text by Akhiezer [4]. Useful reference handbooks with many details
on transcendental functions including those used in these lectures are provided by
Bateman and Erdélyi, [5], which is freely available online, as well as the classical
reference by Abramowitz and Stegun [6].
On the applications in quantum mechanics we will meet the Lamé equation. His-
torically, this equation was studied by Lamé towards completely different applications
[7]. An excellent treatment of this class of ordinary differential equations in given by
Ince in [8]
The presenter took advantage of the experience acquired during his recent research
on classical string solutions and minimal surfaces to prepare these lectures. The
applications of elliptic functions in physics extend to many, much more interesting
directions.
1.2 Elliptic Functions

Basic Definitions
Assume a complex function of one complex variable f (z) obeying the property
f (z + 2ω1 ) = f (z) , f (z + 2ω2 ) = f (z) , (1.1)
for two complex numbers ω1 and ω2 , whose ratio is not purely real (thus, they cor-
respond to different directions on the complex plane). Then, the function f is called
doubly-periodic with periods 2ω1 and 2ω2 . A meromorphic, doubly-periodic function
is called an elliptic function.
The complex numbers 0, 2ω1 , 2ω2 and 2ω1 + 2ω2 define a parallelogram on the
complex plane. Knowing the values of the elliptic function within this parallelogram
3
completely determines the elliptic function, as a consequence of (1.1). However, in-
stead of 2ω1 and 2ω2 , one could use any pair of linear combinations with integer
coefficients of the latter, provided that their ratio is not real. In the general case the
aforementioned parallelogram can be divided to several identical cells. If 2ω1 and 2ω2
have been selected to be “minimal”, or in other words if there is no 2ω within the
parallelogram (boundaries included, vertices excepted), such that f (z + 2ω) = f (z),
then the parallelogram is called a fundamental period parallelogram. Two points z1
and z2 on the complex plane whose difference is an integer multiple of the periods
z2 − z1 = 2mω1 + 2nω2 , m, n ∈ Z, (1.2)
are called congruent to each other. For such points we will use the notation
z1 ∼ z2 . (1.3)
Obviously, by definition, the elliptic function at congruent points takes the same value,
z1 ∼ z2 ⇒ f (z1 ) = f (z2 ) . (1.4)
A parallelogram defined by the points z0 , z0 + 2ω1 , z0 + 2ω2 and z0 + 2ω1 + 2ω2 ,

for any z0 , is called a “cell ”. It is often useful to use the boundary of an arbitrary cell
instead of the fundamental period parallelogram to perform contour integrals, when
poles appear at the boundary of the latter.
Imz
2ω1 + 2ω2
2ω2 z0 + 2ω1 + 2ω2
z0 + 2ω2
2ω1
0 Rez
z0 z0 + 2ω1
Figure 1 – The fundamental period parallelogram is shaded with blue and an

arbitrary cell is shaded with red. The dashed lines define the period parallelograms.
4
Knowing the roots and poles of an elliptic function within a cell suffices to describe
all roots and poles of the elliptic function, as all other roots and poles are congruent
to the former. As such, a set of roots and poles congruent to those within a cell is
called an irreducible set of roots or poles, respectively.
Modular Transformations
Given two fundamental periods 2ω1 and 2ω2 , one can define two different fundamental
periods as,
ω1 0 = aω1 + bω2 , (1.5)

0
ω2 = cω1 + dω2 , (1.6)
where a, b, c, d ∈ Z. Any period in the lattice defined by ω1 0 and ω2 0 , 2ω = 2m0 ω1 0 +

2n0 ω2 0 is obviously a period of the old lattice, but is the opposite also true? In order
for the opposite statement to hold, the area of the fundamental period parallelogram
defined by the new periods 2ω1 0 and 2ω2 0 has to be equal to the area of the funda-
mental period parallelogram defined by the original ones 2ω1 and 2ω2 . The area of
the parallelogram defined by two complex numbers z1 and z2 is given by
A = |Im (z1 z̄2 )| . (1.7)
It is a matter of simple algebra to show that
Im ω1 0 ω̄20 = (ad − bc) Im (ω1 ω̄2 ) .

Thus, the new periods can generate the original lattice if

a b
= ±1, (1.8)

c d
or in other words if !
a b
∈ SL (2, Z) . (1.9)
c d
It is a direct consequence that an elliptic function necessarily obeys
f (z; ω1 , ω2 ) = f z; ω1 0 , ω2 0 ,

(1.10)
when ! !
ω1 0 ω1
=S , (1.11)
ω2 0 ω2
where S ∈ SL (2, Z).
5
Imz
2ω2 0
2ω1 0
2ω2
Rez
2ω1 2ω1 00
2ω2 00
Figure 2 – The green periods 2ω1 0 = 4ω1 − 2ω2 and 2ω2 0 = 2ω1 can generate the
whole original lattice. The red periods 2ω1 00 = 4ω1 + 2ω2 and 2ω2 00 = 2ω2 generate
only half of the original lattice. For example, there is no way to reach the red-
dotted vertex by a linear combination of 2ω1 00 and 2ω2 00 with integer coefficients.
The area of the parallelogram defined by 2ω1 0 and 2ω2 0 is equal to the area of the
parallelogram defined by the original periods, whereas the one defined by 2ω1 00 and
2ω2 00 has double this area.
Basic Properties of Elliptic Functions
Theorem 1.1. The sum of residues over an irreducible set of poles of an elliptic
function vanishes.
To demonstrate this, we use Cauchy’s residue theorem over the boundary of a cell.
I
X 1
Res (f, zi ) = f (z) dz
2πi ∂cell
zi ∈irreducible set of poles
Z z0 +2ω1 Z z0 +2ω1 +2ω2
1 1
= f (z) dz + f (z) dz
2πi z0 2πi z0 +2ω1
Z z0 +2ω2 Z z0
1 1
+ f (z) dz + f (z) dz.
2πi z0 +2ω1 +2ω2 2πi z0 +2ω2
6
Shifting z by 2ω1 in the second integral and by 2ω2 in the third, we yield
Z z0 +2ω1
X 1
Res (f, zi ) = [f (z) − f (z + 2ω2 )] dz
2πi z0
Z z0 +2ω2
1
− [f (z) − f (z + 2ω1 )] dz,
2πi z0
which vanishes as a consequence of f being a doubly periodic function. Therefore,

X
Res (f, zi ) = 0. (1.12)
Theorem 1.2. An elliptic function with an empty irreducible set of poles is a constant
function.
An elliptic function with no poles in a cell, necessarily has no poles at all, as a pole
outside a cell necessarily would have a congruent pole within the cell. Consequently,
such a function is not just meromorphic, but rather it is analytic. Furthermore, an
analytic function in a cell is necessarily bounded within the cell. A direct consequence
of property (1.1) is that an analytic elliptic function is bounded everywhere. But a
bounded analytic function is necessarily a constant function.
The number of roots of the equation
f (z) = z0 (1.13)
is the same for all z0 ∈ C. Before demonstrating this, we will review some properties
of Cauchy’s integral.
Consider a meromorphic function g, with a number of poles ui and roots wj with
multiplicities ri and sj respectively, within a region bounded by a closed contour C.
The Laurent series of the function g at the regime of a pole or a root z0 is
g (z) ' cm (z − z0 )m + cm+1 (z − z0 )m+1 + . . .
with cm 6= 0. In the case z0 is a root wj , then m = sj > 0, while in the case z0 is a

pole ui , then m = −ri < 0. The derivative of g in the regime of a pole or root is
g 0 (z) ' mcm (z − z0 )m−1 + (m + 1) cm+1 (z − z0 )m + . . .
Furthermore, assume an analytic function h. Its Laurent series at the regime of a pole
or root of g is trivially
h (z) ' h (z0 ) + h0 (z0 ) (z − z0 ) + . . .
7
It is a matter of simple algebra to show that the Laurent series of the function hg 0 /g
at the region of z0 is
g 0 (z)

mh (z0 ) 1
h (z) ' + mh (z0 ) + cm+1 m + 1 − 2 h (z0 ) + O (z − z0 )2 .
0
g (z) z − z0 cm
Thus, at any root or pole of the function g, the function hg 0 /g has a first order pole
with residue mh (z0 ). It is a direct consequence of Cauchy’s residue theorem that
g 0 (z) X
I
1 X
h (z) = sj h (wj ) − ri h (ui ). (1.14)
2πi C g (z)
j i
Let’s now return to the case of an elliptic function f . We would like to calculate
the contour integral of formula (1.14) with h (z) = 1, g (z) = f (z) − z0 and C being
the boundary of a cell, namely
f 0 (z)
I
1
I= .
2πi ∂cell f (z) − z0
The function f − z0 is trivially elliptic, while differentiating equations (1.1), one yields
f 0 (z + 2ω1 ) = f 0 (z) , f 0 (z + 2ω2 ) = f 0 (z) , (1.15)
implying that f 0 is also an elliptic function with the same periods as f . In an obvious
manner, the function f 0 (z)/ (f (z) − z0 ) is an elliptic function with the same periods
as f . A direct application of the fact that the sum of the residues of an elliptic function
over a cell vanishes (1.12) is
f 0 (z)
I
1
I= = 0.
2πi ∂cell f (z) − z0
Finally, using equation (1.14), we get

X X
sj − ri = 0. (1.16)
j i
Therefore,
Theorem 1.3. the number of roots of the equation f (z) = z0 in a cell is equal to the
number of poles of f in a cell (weighted by their multiplicity), independently of the
value of z0 .
This number is called the order of the elliptic function f .
Theorem 1.4. The order of a non-constant elliptic function cannot be equal to 1.
8
A non-constant elliptic function has necessarily at least one pole in a cell as a
consequence of theorem 1.2, thus its order is at least 1. However, an elliptic function
of order 1 necessarily has only a single first order pole in a cell. In such a case though,
the sum of the residues of the elliptic function in a cell equals to the residue at this
single pole, and, thus, it cannot vanish. This contradicts (1.12) and therefore an
elliptic function cannot be of order one. The lowest order elliptic functions are of
order 2. Such a function can have either a single second order pole in a cell or two
first order poles with opposite residues.
Theorem 1.5. The sum of the locations of an irreducible set of poles (weighted by
their multiplicity) is congruent to the sum of the locations of an irreducible set of roots
(also weighted by their multiplicity).
To demonstrate this, we will calculate Cauchy’s integral with h (z) = z, g (z) =
f (z) and as contour of integration C the boundary of a cell. The left hand side of
(1.14) equals
zf 0 (z) z0 +2ω1
zf 0 (z) z0 +2ω1 +2ω2
zf 0 (z)
I Z Z
1 1 1
I= = dz + dz
2πi ∂cell f (z) 2πi z0 f (z) 2πi z0 +2ω1 f (z)
z0 +2ω2
zf 0 (z) z0
zf 0 (z)
Z Z
1 1
+ dz + dz.
2πi z0 +2ω1 +2ω2 f (z) 2πi z0 +2ω2 f (z)
We shift z by 2ω1 in the second integral and by 2ω2 in the third one to yield
Z z0 +2ω1 0
zf (z) (z + 2ω2 ) f 0 (z + 2ω2 )

1
I= − dz
2πi z0 f (z) f 0 (z + 2ω2 )
Z z0 +2ω2 0
zf (z) (z + 2ω1 ) f 0 (z + 2ω1 )

1
− − dz.
2πi z0 f (z) f 0 (z + 2ω1 )
Using the periodicity properties of f , (1.1) and f 0 (1.15), we yield
Z z0 +2ω1 0
zf (z) (z + 2ω2 ) f 0 (z)

1
I= − dz
2πi z0 f (z) f 0 (z)
Z z0 +2ω2 0
zf (z) (z + 2ω1 ) f 0 (z)

1
− − dz
2πi z0 f (z) f 0 (z)
ω2 z0 +2ω1 f 0 (z) ω1 z0 +2ω2 f 0 (z)
Z Z
=− dz + dz
πi z0 f (z) πi z0 f (z)
ω2 ω1
= − (ln f (z0 + 2ω1 ) − ln f (z0 )) + (ln f (z0 + 2ω2 ) − ln f (z0 )) .
πi πi
Although z0 +2ω1 ∼ z0 ∼ z0 +2ω2 , due to the branch cut of the logarithmic function, in
general we have that ln f (z0 + 2ω1 ) − ln f (z0 ) = 2mπ and ln f (z0 + 2ω2 ) − ln f (z0 ) =
2nπ, with m, n ∈ Z. Thus,
I = 2mω1 + 2nω2 ∼ 0.
9
Applying property (1.14) we get
X X
I= ri ui − sj wj ,
i j
implying X X
rj uj ∼ s i wi , (1.17)
j i
which is the proof of theorem 1.5.
1.3 The Weierstrass Elliptic Function

Definition
As we showed in previous section, the lowest possible order of an elliptic function is 2.
One possibility for such an elliptic function is a function having a single second order
pole in each cell. It is actually quite easy to construct such a function. It suffices to
sum an infinite set of copies of the function 1/z 2 each one shifted by 2mω1 + 2nω2 for
all m, n ∈ Z. The usual convention includes the addition of a constant cancelling the
contributions of all these functions at z = 0 (except for the term with m = n = 0),
so that the Laurent series of the constructed function at the region of z = 0 has a
vanishing zeroth order term. Following these directions, we define,

1 X 1 1
℘ (z) := 2 + − . (1.18)
z
{m,n}6={0,0}
(z + 2mω1 + 2nω2 )2 (2mω1 + 2nω2 )2
By construction, this function is doubly periodic with fundamental periods equal to

2ω1 and 2ω2 .
℘ (z + 2ω1 ) = ℘ (z) , ℘ (z + 2ω2 ) = ℘ (z) . (1.19)
This function is called Weierstrass elliptic function.
Basic Properties
A direct consequence of the definition (1.18) is the fact that the Weierstrass elliptic
function is an even function
℘ (−z) = ℘ (z) . (1.20)
Let’s acquire the Laurent series of the Weierstrass elliptic function at the regime
of z = 0. It is easy to show that
∞
1 1 1 X z k
− = (k + 1) − .
(z + w)2 w2 w2
k=1
w
10
Consequently,
∞ ∞
1 X X (k + 1) (−1)k 1 k
X
℘ (z) = 2 + k+2
z = 2
+ ak z k ,
z (mω1 + nω2 ) z
k=1 {m,n}6={0,0} k=1
where X 1
ak = (k + 1) (−1)k .
{m,n}6={0,0}
(2mω1 + 2nω2 )k+2
The fact that ℘ is even implies that only the even indexed coefficients do not vanish,
a2l+1 = 0, (1.21)
X 1
a2l = (2l + 1) . (1.22)
{m,n}6={0,0}
(2mω1 + 2nω2 )2(l+1)
For reasons that will become apparent later, we define g2 and g3 so that
1 g2 2 g3 4 6

℘ (z) = + z + z + O z , (1.23)
z 2 20 28
implying that
X 1
g2 := 60 , (1.24)
{m,n}6={0,0}
(2mω1 + 2nω2 )4
X 1
g3 := 140 . (1.25)
{m,n}6={0,0}
(2mω1 + 2nω2 )6
Weierstrass Differential Equation

Directly differentiating equation (1.18), the derivative of Weierstrass function can be
expressed as
X 1
℘0 (z) = −2 3. (1.26)
m,n (z + mω1 + nω2 )
It follows that the derivative of Weierstrass elliptic function is an odd function
℘0 (−z) = −℘0 (z) . (1.27)
The Laurent series of ℘0 (z), ℘3 (z) and ℘02 (z) at the regime of z = 0 are
1 g2 g3
+ z2 + z4 + O z6 ,

℘ (z) = 2
z 20 28
0 2 g2 g3
℘ (z) = − 3 + z + z 3 + O z5 ,

z 10 7
1 3g 2 1 3g3
℘3 (z) = 6 + z2 ,

2
+ +O
z 20 z 28
4 2g2 1 4g3
℘02 (z) = 6 − z2 .

2
− +O
z 5 z 7
11
It is not difficult to show that there is a linear combination of the above, which is not
singular at z = 0 and furthermore it vanishes there. One can eliminate the sixth order
pole by taking an appropriate combination of ℘02 and ℘3 . This leaves a function with
a second order pole. Taking an appropriate combination of the latter combination
and ℘ allows to write down a function with no poles at z = 0. Trivially, adding
an appropriate constant results in a non-singular function vanishing at z = 0. The
appropriate combination turns out to be
℘02 (z) − 4℘3 (z) + g2 ℘ (z) + g3 = O z 2 .

But, the derivative, as well as powers of an elliptic function are elliptic functions with
the same periods. Therefore, the function ℘02 (z) = 4℘3 (z) − g2 ℘ (z) − g3 is an elliptic
function with the same periods as ℘ (z). Since the latter has no pole at z = 0, it does
not have any pole at all, and, thus, it is an elliptic function with no poles. According
to theorem 1.2, elliptic functions with no poles are necessarily constants and since
℘02 (z) = 4℘3 (z) − g2 ℘ (z) − g3 vanishes at the origin, it vanishes everywhere. This
implies that Weierstrass elliptic function obeys the differential equation,
℘02 (z) = 4℘3 (z) − g2 ℘ (z) − g3 = 0. (1.28)
This differential equation is of great importance in the applications of Weierstrass

elliptic function in physics. For a physicist it is sometimes useful to even conceive this
differential equation as the definition of Weierstrass elliptic function.
It turns out that the Weierstrass elliptic function is the general solution of the
differential equation
2
dy
= 4y 3 − g2 y − g3 . (1.29)
dz
Performing the substitution y = ℘ (w), the equation (1.29) assumes the form
2
dw
= 1,
dz
which obviously has the solutions, w = ±z + z0 . This implies that y = ℘ (±z + z0 )

and since the Weierstrass elliptic function is even, the general solution of Weierstrass
equation (1.29) can be written in the form
y = ℘ (z + z0 ) . (1.30)
In the following, we will deduce an integral formula for the inverse function of ℘.
In order to do so, we define the function z (y) as
Z ∞
1
z (y) := p dt. (1.31)
y 3
4t − g2 t − g3
12
Differentiating with respect to z one gets
2
dy 1 dy
1=− p ⇒ = 4y 3 − g2 y − g3 .
dz 4y 3 − g2 y − g3 dz
We just showed that the general solution of this equation is
y = ℘ (z + z0 ) .
Since the integral in (1.31) converges, it should vanish at the limit y → ∞, or equiv-
alently, lim z (y) = 0. This implies that z = z0 is the position of a pole, or in other
y→∞
words it is congruent to z = 0. This means that
y = ℘ (z + 2mω1 + 2nω2 ) = ℘ (z)
Substituting the above into the equation (1.31) yields the integral formula for Weier-
strass elliptic function, Z ∞
1
z= p dt. (1.32)
℘(z) 4t3 − g2 t − g3
Once should wonder, how the above formula is consistent with the fact that ℘ is
an elliptic function, and, thus, all numbers congruent to each other should be mapped
to the same value of ℘. The answer to this question is that the integrable quantity in
(1.32) has branch cuts. Depending on the selection of the path from ℘ (z) to infinity
and more specifically depending on how many times the path encircles each branch
cut, one may result in any number congruent to z or −z. A more precise expression
of the integral formula is
Z ∞
1
p dt ∼ ±z. (1.33)
℘(z) 3
4t − g2 t − g3
The Roots of the Cubic Polynomial

We define the values of the Weierstrass elliptic function at the half-periods ω1 , ω2 and
ω3 := ω1 + ω2 as
e1 := ℘ (ω1 ) , e2 := ℘ (ω3 ) , e3 := ℘ (ω2 ) . (1.34)
The permutation between the indices of ω’s and e’s is introduced for notational reasons
that will become apparent later. The periodicity properties of ℘ combined with the
fact that the latter is an even function, imply that ℘ is stationary at the half periods.
For example,
℘0 (ω1 ) = −℘0 (−ω1 ) = −℘0 (2ω1 − ω1 ) = −℘0 (ω1 ) ,
13
implying that ℘0 (ω1 ) = 0. Similarly one can show that
℘0 (ω1 ) = ℘0 (ω2 ) = ℘0 (ω3 ) = 0. (1.35)
Substituting a half-period into Weierstrass equation (1.29), we yield
4e3i − g2 ei − g3 = 0. (1.36)
The derivative of ℘, as shown in equation (1.26) have a single third order pole in each
cell, congruent to z = 0. Thus, ℘0 is an elliptic function of order 3 and therefore it has
exactly three roots in each cell. Since ω1 , ω2 and ω3 all lie within the fundamental
period parallelogram, they cannot be congruent to each other, and, thus, there is no
other root within the latter. This also implies that ω1 , ω2 and ω3 are necessarily first
order roots of ℘0 . All other roots of ℘0 are congruent to those. Finally, when equation
(1.36) has a double root, the solution of the differential equation (1.29) cannot be an
elliptic function.
An implication of the above is the fact that the locations z = ω1 , z = ω2 and
z = ω3 are the only locations within the fundamental period parallelogram, where the
Laurent series of the function ℘ (z) − ℘ (z0 ) have a vanishing first order term at the
region of z0 . Consequently the equation ℘ (z) = f0 has a double root only when f0
equals any of the three roots e1 , e2 or e3 . Since ℘ is an order two elliptic function,
the complex numbers e1 , e2 and e3 are the only ones appearing only once in a cell,
whereas all other complex numbers appear twice.
Finally, equation (1.36) implies that ei are the three roots of the polynomial ap-
pearing in the right hand side of Weierstrass equation, namely
Q (t) := 4t3 − g2 t − g3 = 4 (t − e1 ) (t − e2 ) (t − e3 ) . (1.37)
This directly implies that ei obey
e1 + e2 + e3 = 0, (1.38)
g2
e1 e2 + e2 e3 + e3 e1 = − , (1.39)
4
g3
e1 e2 e3 = . (1.40)
4
Other Properties
The Weierstrass elliptic function obeys the homogeneity relation

2 g2 g3
℘ (z; g2 , g3 ) = µ ℘ µz; 4 , 6 . (1.41)
µ µ
For the specific case µ = i, the above relation assumes the form
℘ (z; g2 , g3 ) = −℘ (iz; g2 , −g3 ) . (1.42)
14
Finally, when two of the roots e1 , e2 and e3 coincide, the Weierstrass elliptic
function degenerates to a simply periodic function. Assuming that the moduli g2 and
g3 are real, then the existence of a double root implies that all roots are real. When
the double root is larger than the simple root, the Weierstrass elliptic function takes
the form
3e0
℘ z; 12e20 , −8e30 = e0 +

2 √
, (1.43)
sinh 3e0 z
whereas when the double root is smaller than the simple root, it takes the form
3e0
℘ z; 12e20 , 8e30 = −e0 +

2
√ . (1.44)
sin 3e0 z
If there is only one triple root, then it must be vanishing, since the three roots sum
to zero. In this case, the Weierstrass elliptic function degenerates to a function that
is not periodic at all, namely
1
℘ (z; 0, 0) = 2 . (1.45)
z
The proofs of the homogeneity relation, as well as the double root limits of the
Weierstrass elliptic function are left as an exercise for the audience.
15
Problems
Problem 1.1. Show, using the integral formula for the Weierstrass elliptic function
(1.32) that when g2 and g3 are real and all roots e1 , e2 and e3 are also real, the half-
period corresponding to the largest root e1 is congruent to a real number, whereas the
half-period corresponding to the smallest root e3 is congruent to a purely imaginary
number.
Then, show that when there is one real root and two complex ones, the half-period
corresponding to the real root e2 is congruent to both a real and a purely imaginary
number.
Problem 1.2. Show that at the limit two of the roots e1 , e2 and e3 coincide, the
Weierstrass elliptic function degenerates to a simply periodic function and can be
expressed in terms of trigonometric or hyperbolic functions as described by formulae
(1.43) and (1.44). Find the value of the unique period in terms of the double root. Also
show that at the limit all three roots e1 , e2 and e3 coincide, the Weierstrass elliptic
function degenerates to the non-periodic function of equation (1.45).
Problem 1.3. Use the definition (1.18) to deduce the homogeneity property of the
Weierstrass elliptic function (1.41).
16
Lecture 2: Weierstrass Quasi-periodic Functions
In the previous lecture, we used several times the fact that the derivative of an elliptic
function is also an elliptic function with the same periods. However, the opposite
statement is not correct; the indefinite integral of an elliptic function is not necessar-
ily an elliptic function. Such non-elliptic functions typically expose other interesting
quasi-periodicity properties. In the following we will study two such quasi-periodic
functions that are derived from Weierstrass elliptic function and have numerous im-
portant applications.
2.1 Quasi-periodic Weierstrass Functions

The Weierstrass ζ Function
The Weierstrass ζ function is defined as
dζ (z)
:= −℘ (z) , (2.1)
dz
furthermore satisfying the condition

1
lim ζ (z) − := 0, (2.2)
z→0 z
which fixes the integration constant.

Using the definition of Weierstrass ℘ function, we acquire

1 X 1 1 z
ζ (z) = + − + .
z z + 2mω1 + 2nω2 2mω1 + 2nω2 (2mω1 + 2nω2 )2
{m,n}6={0,0}
(2.3)
Weierstrass ζ function is an odd function.
ζ (−z) = −ζ (z) . (2.4)
Notice that the definition (2.1) coupled with the fact that ℘ is an even function implies
that ζ is an odd function up to a constant. The condition (2.2) fixes this constant to
zero, so that ζ is an odd function.
Quasi-periodicity of the Function ζ

Equation (2.3) implies that in each cell defined by the periods of the corresponding
℘ function, the function ζ has only a first order pole with residue equal to one. As
such, it cannot be an elliptic function with the same periods as ℘. Actually, it cannot
be an elliptic function with any periods, since it is not possible to define a cell where
17
the sum of the residues would vanish. The Weierstrass ζ function is quasi-periodic.
Its quasi-periodicity properties can be deduced from the periodicity of Weierstrass ℘
function. More specifically, integrating the relation ℘ (z + 2ωi ) = ℘ (z), we find
ζ (z + 2ωi ) = ζ (z) + c.
The above relation for z = −ωi yields ζ (ωi ) = ζ (−ωi ) + c. Since ζ is an odd function,
the above implies that c = 2ζ (ωi ), which in turn yields
ζ (z + 2ωi ) = ζ (z) + 2ζ (ωi ) . (2.5)
One can easily show inductively that
ζ (z + 2mω1 + 2nω2 ) = ζ (z) + 2mζ (ω1 ) + 2nζ (ω2 ) . (2.6)
The quantities ζ (ω1 ) and ζ (ω2 ) are related with an interesting property. Consider
the contour integral
I
1
I= ζ (z) dz
2πi ∂cell
Since ζ has only a first order pole with residue equal to one within a cell, Cauchy
residue theorem implies that
I = 1.
Performing the contour integral along the boundary of a cell, we get

Z z0 +2ω1 Z z0 +2ω1 +2ω2 Z z0 +2ω2 Z z0
2πiI = ζ (z) dz + ζ (z) dz + ζ (z) dz + ζ (z) dz.
z0 z0 +2ω1 z0 +2ω1 +2ω2 z0 +2ω2
Shifting z by 2ω1 in the second integral and by 2ω2 in the third, we yield
Z z0 +2ω1 Z z0 +2ω2
2πiI = (ζ (z) − ζ (z + 2ω2 )) dz − (ζ (z) − ζ (z + 2ω1 )) dz
z0 z0
Z z0 +2ω1 Z z0 +2ω2
= (ζ (z) − ζ (z) − 2ζ (ω2 )) dz − (ζ (z) − ζ (z) − 2ζ (ω1 )) dz
z0 z0
= −4ω1 ζ (ω2 ) + 4ω2 ζ (ω1 ) .
As a result, ζ (ωi ) and ζ (ωi ) are related as
πi
ω2 ζ (ω1 ) − ω1 ζ (ω2 ) = . (2.7)
2
18
The Weierstrass σ Function
Since Weierstrass elliptic function has a single second order pole in each cell, integrat-
ing it once resulted in a function (the Weierstrass ζ function) with a single first order
pole in each cell. Integrating once more would lead to a logarithmic singularity in
each cell. To avoid this, we define the next quasi-periodic function as the exponential
of the integral of the Weierstrass ζ function.
The Weierstrass σ function is defined as
σ 0 (z)
:= ζ (z) , (2.8)
σ (z)
together with the condition

σ (z)
lim := 1, (2.9)
z→0 z
which fixes the integration constant.

Integrating equation (2.3) term by term results in the following expression for σ
function.
2
Y z + 2mω1 + 2nω2 − 2mω1 z+2nω2 + 2(2mω z+2nω )2
σ (z) = z e 1 2 . (2.10)
2mω1 + 2nω2
{m,n}6={0,0}
This implies that σ is analytic. At the locations of the poles of ℘, it has first order
roots.
Similarly to the definition of ζ function, definition (2.8) implies that σ is odd up
to a constant, which is set to zero by the defining condition (2.9), so that σ is an odd
function,
σ (−z) = −σ (z) . (2.11)
Quasi-periodicity of the Function σ

The Weierstrass σ function is a quasi-periodic function. Its quasi-periodicity proper-
ties follow from the corresponding properties of the Weierstrass ζ function. Integrating
the equation (2.5), one yields
ln σ (z + 2ωi ) = ln σ (z) + 2ζ (ωi ) z + c,
or
σ (z + 2ωi ) = Ce2ζ(ωi )z σ (z) ,
where C = ec . Substituting z = −ωi , we get
σ (ωi ) = Ce−2ζ(ωi )ωi σ (−ωi ) = −Ce−2ζ(ωi )ωi σ (ωi ) ,
19
implying, C = −e2ζ(ωi )ωi . This means that the quasi-periodicity of the function σ is
σ (z + 2ωi ) = −e2ζ(ωi )(z+ωi ) σ (z) . (2.12)
Equation (2.12) can be used to prove inductively that
σ (z + 2mω1 + 2nω2 ) = (−1)m+n+mn e(2mζ(ω1 )+2nζ(ω2 ))(z+mω1 +nω2 ) σ (z) . (2.13)
2.2 Expression of any Elliptic Function in Terms of Weier-

strass Functions
The functions ℘, ζ or σ can be used to express any elliptic function with the same
periods. In the following, we will derive such expressions and deduce interesting
properties of a general elliptic function.
Expression of any Elliptic Function in Terms of ℘ and ℘0

Assume an elliptic function f (z). Then, it can be written as
f (z) + f (−z) f (z) − f (−z) 0
f (z) = + ℘ (z) .
2 2℘0 (z)
Both functions f (z)+f2

(−z)
and f (z)−f (−z)
2℘0 (z) are even. Thus, in order to express an ar-
bitrary elliptic function in terms of ℘ and ℘0 , it suffices to find an expression of an
arbitrary even elliptic function in terms of ℘ and ℘0 .
Assume an even elliptic function g (z). As it is even, any irreducible set of poles of
g (z) can be divided to a set of points ui with multiplicity ri and another set of points
congruent to −ui with equal multiplicities. In a similar manner, an irreducible set of
roots of g (z) can be divided to a set of points wi with multiplicities si and another
set of points congruent to −wi with equal multiplicities. Now consider the function
(℘ (z) − ℘ (ui ))ri
Q
h (z) := g (z) Qi .
(℘ (z) − ℘ (wj ))sj
j
This function is obviously an elliptic function. Furthermore, it has no poles since, the
poles of g (z) are cancelled by the roots of the product in the numerator, the poles
of the numerator are cancelled with the poles of the denominator, and the poles due
to the zero’s of the denominator are cancelled by the zeros of g (z). As a result, the
function h is constant by Liouville’s theorem and therefore,
(℘ (z) − ℘ (wj ))sj
Q
j
g (z) = C Q . (2.14)
(℘ (z) − ℘ (ui ))ri
i
20
Summarizing, any elliptic function can be written in terms of ℘ (z) and ℘0 (z) with
the same periods. This expression is rational in ℘ (z) and linear in ℘0 (z).
The above result has some interesting consequences. Consider two elliptic functions
f (z) and g (z) with the same periods. Then, they can be both expressed as functions
of ℘ and ℘0 with the same periods,
f (z) = F ℘ (z) , ℘0 (z) ,

g (z) = G ℘ (z) , ℘0 (z) .

Both functions F and G are rational in their first argument and linear in the second.
Bearing in mind that ℘ (z) and ℘0 (z) are also connected through the Weierstrass
equation,
℘02 (z) = 4℘3 (z) − g2 ℘ (z) − g3 = 0,
one can eliminate ℘ and ℘0 and result in an algebraic relation between f (z) and g (z).
This means that any pair of elliptic functions with the same periods are algebraically
connected. Two implications of the above sentence are
1. There is an algebraic relation between any elliptic function and its derivative.
2. There is an algebraic relation between any elliptic function and the same elliptic
function with shifted argument.
An algebraic relation between an elliptic function f and the same elliptic function
with shifted argument reads
n n
!
X X
k l
f (z + w) cl (w) f (z) = 0,
k=1 l=1
where cl (w) are unspecified functions of w. If one interchanges z and w, they will get
n n
!
X X
k l
f (z + w) cl (z) f (w) = 0,
k=1 l=1
implying that the unknown functions cl (w) are necessarily powers of f (w), so that
n
cl (w) f l (z) are symmetric polynomials in f (z) and f (w). This implies
P
the sums
l=1
that the above equations take the form of an algebraic relation between f (z), f (w)
and f (z + w), i.e. an algebraic addition theorem.
Expression of any Elliptic Function in Terms of ζ and its derivatives

Consider an elliptic function f (z). Assume that ui is an irreducible set of poles of
f (z), which have multiplicities ri . Furthermore, assume that the principal part of the
21
Laurent series at the region of a pole is given by
ci,ri ci,2 ci,1
0

f (z) ' + . . . + + + O (z − ui ) .
(z − ui )ri (z − ui )2 z − ui
Weierstrass ζ function has a single first order pole with residue equal to one in
locations congruent to z = 0. It is obvious that the n-th derivative of ζ has a single
(n + 1)-th order pole and the principal part of Laurent series in the regime of z = 0 is
dn ζ (z) (−1)n n!
0

' + O (z) .
dz n z n+1
It follows that the function
X dζ (z − ui )
g (z) := f (z) − ci,1 ζ (z − ui ) − ci,2 + ...
dz
i
!
(−1)ri −1 ci,ri dri −1 ζ (z − ui )
+
(ri − 1)! dz ri −1
has no poles. It is not obvious though whether g (z) is an elliptic function, since the
function ζ is not an elliptic function. However, all derivatives of ζ are elliptic functions,
and, thus,
X X
g (z + 2ωi ) − g (z) = ci,1 (ζ (z − ui ) − ζ (z + 2ωi − ui )) = −2ζ (ωi ) ci,1 = 0,
i i
P
since ci,1 is the sum of the residues of the elliptic function f (z) in a cell.
i
Thus, g (z) is an elliptic function with no poles, and, thus, by Liouville’s theorem,
it is a constant function. This means that the original elliptic function f (z) can be
written as
ri
XX (−1)j−1 ci,j dj−1 ζ (z − ui )
f (z) = C + . (2.15)
(j − 1)! dz j−1
i j=i
It follows that an elliptic function is completely determined by the principal parts

of its Laurent series at an irreducible set of poles, up to an additive constant.
Expression of any Elliptic Function in Terms of σ

Consider an elliptic function f (z) having an irreducible set of poles ui with multiplic-
ities ri and an irreducible set of roots wj with multiplicities sj . We remind the reader
P P
that s i wi ∼ rj uj . It is always possible to select the irreducible sets of poles and
i Pj P
roots so that s i wi = rj uj . In the following, we assume that we have made such
i j
a selection.
22
Assume the function
σ rι (z − ui )
Q
g (z) := f (z) Qi .
σ sj (z − wj )
j
Since the function σ has a unique first order root in each cell congruent to the origin,
it is obvious that g (z) has neither poles nor roots in a cell.
It turns out that the quasi-periodicity property (2.12) of function σ implies that
g (z) is an elliptic function. Indeed,
ri
σ rι (z − uι ) −e2ζ(ωk )(z−ui −ωk )
Q Q
g (z + 2ωk ) = f (z) Qi i
sj
σ sj (z − wj ) −e2ζ(ωk )(z−wj −ωk )
Q
j j
" ! !#
P P P P
ri − sj − ri ui − sj wj
P P
ri − sj 2ζ(ωk ) (z−ωk )
i j i j
= g (z) (−1) i j
e = g (z) .
Since g (z) is an elliptic function with no poles, it is a constant function and the
original elliptic function f (z) can be expressed as
Q sj
σ (z − wj )
j
f (z) = C Q rι . (2.16)
σ (z − uι )
i
This implies that an irreducible set of poles and roots of an elliptic function com-
pletely determines it, up to a multiplicative constant.
2.3 Addition Theorems

Addition Theorem for ℘
Above, we showed that there is an algebraic addition theorem for every elliptic func-
tion. Therefore such an addition theorem exists for Weierstrass elliptic function, too.
We define two functions of two complex variables c1 (z, w) and c2 (z, w) as
℘0 (z) = c1 (z, w) ℘ (z) + c2 (z, w) ,

℘0 (w) = c1 (z, w) ℘ (w) + c2 (z, w)
or in other words,
℘0 (z) − ℘0 (w)
c1 (z, w) := ,
℘ (z) − ℘ (w)
℘ (z) ℘0 (w) − ℘ (w) ℘0 (z)
c2 (z, w) := .
℘ (z) − ℘ (w)
23
We also define the function f (x) of one complex variable as
f (x) := ℘0 (x) − c1 (z, w) ℘ (x) − c2 (z, w) .
Clearly, the function f has only one third order pole in each cell congruent to x = 0.
Moreover, by definition it has two roots at x = z and x = w. Theorem 1.3 implies that
there is another first order root, which is not congruent to x = z or x = w. Theorem
1.5 implies that the position of this pole is congruent to x = −z − w,
f (−z − w) = 0.
The function g (x) of one complex variable defined as,
g (x) := ℘02 (x) − (c1 (z, w) ℘ (x) + c2 (z, w))2 ,
clearly vanishes everywhere f (x) vanishes, therefore,
g (z) = g (w) = g (−z − w) = 0.
Using Weierstrass differential equation (1.29), one can write the function g (x) as,
g (x) = 4℘3 (x) − c21 (z, w) ℘2 (x) − (2c1 (z, w) c2 (z, w) + g2 ) ℘ (x) − c22 (z, w) + g3 .

The fact that g (x) has the three roots x = z, x = w and x = −z − w, implies that
the third order polynomial
Q (P ) := 4P 3 − c21 (z, w) P 2 − (2c1 (z, w) c2 (z, w) + g2 ) P − c22 (z, w) + g3

(2.17)
has the roots P = ℘ (z), P = ℘ (w) and P = ℘ (−z − w),
Q (℘ (z)) = Q (℘ (w)) = Q (℘ (−z − w)) = 0.
In other words, the polynomial Q (P ) can be written as
Q (P ) = 4 (P − ℘ (z)) (P − ℘ (w)) (P − ℘ (−z − w)) .
Comparing the coefficients of the second order term of the polynomial Q (P ) in ex-
pressions (2.17) and (2.3), we find
c21 (z, w) = 4 (℘ (z) + ℘ (w) + ℘ (−z − w))
or
1 ℘0 (z) − ℘0 (w) 2

℘ (z + w) = −℘ (z) − ℘ (w) + , (2.18)
4 ℘ (z) − ℘ (w)
which is the desired addition theorem for the Weierstrass elliptic function. Although,
expression (2.18), which is the traditional form of the addition theorem, involves the
derivative of Weierstrass elliptic function, the latter can be eliminated with the use
of Weierstrass differential equation (1.29) resulting in an algebraic relation between
℘ (z), ℘ (w) and ℘ (z + w).
24
Pseudo-addition Theorems for ζ and σ
The functions ζ and σ are not elliptic, and, thus, they are not guaranteed to obey
algebraic addition theorems. However, the fact that any elliptic function can be writ-
ten as ratio of σ functions can be used to deduce pseudo-addition theorems for them.
Consider the function ℘ (z) − ℘ (w) as a function of z. This function, obviously has a
second order pole in each cell, congruent to z = 0. It also obviously has two roots con-
gruent to z = w and z = −w. Consequently, equation (2.16) implies that ℘ (z) − ℘ (w)
can be written as
σ (z − w) σ (z + w)
℘ (z) − ℘ (w) = A .
σ 2 (z)
Writing down the principal part of the Laurent series of the above relation at the
region of z = 0, we find
1 σ (−w) σ (w)
2
=A ,
z z2
implying that
1
A=− 2 .
σ (w)
The above results in the following pseudo-addition theorem for σ functions
σ (z − w) σ (z + w)
℘ (z) − ℘ (w) = − . (2.19)
σ 2 (z) σ 2 (w)
Differentiating equation (2.19) with respect to z and w, we arrive of the following

relations,
σ (z − w) σ (z + w)
℘0 (z) = − (ζ (z − w) + ζ (z + w) − 2ζ (z)) ,
σ 2 (z) σ 2 (w)
σ (z − w) σ (z + w)
−℘0 (w) = − (−ζ (z − w) + ζ (z + w) − 2ζ (w)) ,
σ 2 (z) σ 2 (w)
Adding them up, we find,
σ (z − w) σ (z + w)
℘0 (z) − ℘0 (w) = −2 (ζ (z + w) − ζ (z) − ζ (w)) ,
σ 2 (z) σ 2 (w)
which implies the following pseudo addition theorem for the ζ function
1 ℘0 (z) − ℘0 (w)
= ζ (z + w) − ζ (z) − ζ (w) . (2.20)
2 ℘ (z) − ℘ (w)
25
Problems
Problem 2.1. Use the definitions (2.1) and (2.8) to deduce the homogeneity properties
of the Weierstrass quasi-periodic functions

g2 g3
ζ (z; g2 , g3 ) = µζ µz; 4 , 6 , (2.21)
µ µ

1 g2 g3
σ (z; g2 , g3 ) = σ µz; 4 , 6 . (2.22)
µ µ µ
Problem 2.2. Prove the parity properties of Weierstrass quasi-periodic functions.

Namely,
• Show that the definition (2.1) together with the defining condition (2.2) imply
that ζ is an odd function.
• Show that the definition (2.8) together with the defining condition (2.9) imply
that σ is an odd function.
Problem 2.3. Prove the relations giving the Weierstrass quasi-periodic functions after
a shift of their argument by an arbitrary period in the lattice of the corresponding
Weierstrass elliptic function. Namely,
• Use equation (2.5) to prove (2.6).
• Use equation (2.12) to prove (2.13).
Problem 2.4. Use the addition theorem for Weierstrass elliptic function to show that
2e21 + e2 e3
℘ (z + ω1 ) = e1 + , (2.23)
℘ (z) − e1
2e2 + e1 e2
℘ (z + ω2 ) = e3 + 3 , (2.24)
℘ (z) − e3
2e2 + e3 e1
℘ (z + ω3 ) = e2 + 2 . (2.25)
℘ (z) − e2
Problem 2.5. Use the fact that every elliptic function can be written in terms of the ζ
function and its derivatives to deduce the pseudo-addition theorem for the ζ function.
Problem 2.6. Use the pseudo-addition theorem of the ζ function to deduce the addi-
tion theorem for the ℘ function.
26
Problem 2.7. Use the addition theorem of the ℘ function to deduce a duplication
formula. Then, Use the fact that ℘ (2z) can be considered an elliptic function with the
same periods as ℘ (z) to express it in terms of ζ (z) and its derivatives and result in
the same duplication formula. For this purpose, you will find the results of problem
2.4 useful.
27
Lecture 3: Applications in Classical Mechanics
The first applications of the Weierstrass functions in physics that we will face are prob-
lems in classical mechanics with one degree of freedom, where Weierstrass equation
(1.29) emerges as conservation of energy. In such problems, the moduli g2 and g3 are
connected to physical quantities, and, thus, they are real. Furthermore, the unknown
function, as well as the independent variable in Weierstrass equation represent some
physical quantity and they have to be real, too. It follows that we need to understand
the real solutions in the real domain of equation (1.29), in the case the moduli are
also real.
3.1 Real Solutions of Weierstrass Equation in the Real

Domain
The Weierstrass Elliptic Function with Real Moduli
As the general solution of equation (1.29) is given in terms of the Weierstrass elliptic
function, we need to study the special properties of the former with real moduli.
When the parameters g2 and g3 are real, there are two possible cases for the reality
of the three roots:
1. All three roots are real; conventionally we define them such that e1 > e2 > e3 .
In this case, we may select the fundamental half-periods so that ω1 is real and ω2
is purely imaginary. Then, they are given by the expressions,
Z +∞
dt
ω1 = p , (3.1)
e1 4 (t − e1 ) (t − e2 ) (t − e3 )
Z e3
dt
ω2 = i p . (3.2)
−∞ 4 (e1 − t) (e2 − t) (e3 − t)
The above expressions imply that at the limit the two larger roots coincide, the
real half-period diverges, whereas at the limit the two smaller roots coincide, the
imaginary half-period diverges.
2. There is one real root and two complex ones being complex conjugate to each
other; conventionally, we define e2 as the real one and e1 and e3 so that Ime1 > 0.
In this case, it is not possible to select the fundamental half-periods as in the case
of the three real roots, but we may select them so that they are complex conjugate
28
to each other. Then, they are given by the expressions,
Z +∞
dt
ω1 + ω2 = p , (3.3)
e2 4 (t − e1 ) (t − e2 ) (t − e3 )
Z e2
dt
ω1 − ω2 = i p . (3.4)
−∞ 4 (t − e1 ) (e2 − t) (t − e3 )
A proof of this statement is presented in the appendix.
The Locus of Complex Numbers z for Whom ℘ (z) Is Real

We return to the investigation for real solutions of equation (1.29) with real moduli g2
and g3 . Since the general solution of the latter is given in terms of Weierstrass elliptic
function, this investigation requires the specification of the locus of complex numbers
z for whom ℘ (z) is real. A preliminary observation that can be made is the fact that
the Weierstrass elliptic function assumes real values on the real and imaginary axes.
The definition of the elliptic function ℘ (1.18) implies that
℘ (z; ω1 , ω2 ) = ℘ (z̄; ω̄1 , ω̄2 ) .
The definitions of the moduli g2 and g3 , as given by equations (1.24) and (1.25) imply
that
g2 (ω̄1 , ω̄2 ) = g2 (ω1 , ω2 ), g3 (ω̄1 , ω̄2 ) = g3 (ω1 , ω2 )
and consequently,
℘ (z; g2 , g3 ) = ℘ (z̄; ḡ2 , ḡ3 ) .
Thus, when the moduli g2 and g3 are real, it holds that
℘ (z; g2 , g3 ) = ℘ (z̄; g2 , g3 ) . (3.5)
The above equation and the fact that ℘ is even imply that ℘ is real on the real
and imaginary axes. Let x ∈ R, then,
℘ (x; g2 , g3 ) = ℘ (x̄; g2 , g3 ) = ℘ (x; g2 , g3 ) , (3.6)

℘ (ix; g2 , g3 ) = ℘ (−ix; g2 , g3 ) = ℘ (ix; g2 , g3 ) . (3.7)
But is ℘ real on any other points not congruent to the real or imaginary axes? The
answer depends on the reality of the roots e1 , e2 and e3 . The function ℘ is an order
two elliptic function, and, thus, it assumes any real value (as well as any complex
value) twice in every cell. The only exception to this rule are the three roots ei , which
appear only once, since they correspond to double roots of the equation ℘ (z) = ei .
In figure 3, one cell of ℘ with real moduli is plotted for either three or one real root.
29
Imz Imz
−∞ e2 2ω2
2ω2
e3 e2 −∞
e2 +∞ e2 Rez
−∞ e2
Rez
+∞ e1 2ω1 +∞ 2ω1
three real roots one real root
Figure 3 – The values of ℘ on the real and imaginary axes
Dotted points are congruent to a period or a half-period. Grey dots at the boundary
of the plotted cell are congruent to a black dot at another point of the boundary, and
thus, they are not considered to be part of the cell. This holds for segments that
connect grey dots. Moreover, as the poles are second order with Laurent coefficient
equal to one, as one approaches a pole from the real axis, ℘ tends to +∞, while
when one approaches a pole from the imaginary axis, ℘ tends to −∞. Finally, as the
derivative vanishes only at the half periods, ℘ is monotonous at the segments between
consequent half-periods and poles.
Having the above into consideration, and following picture 3, it is clear that
• when there are three real roots, the function ℘ assumes all real values larger
than e1 twice in the segment [0, 2ω1 ] on the real axis; each value appears once
in [0, ω1 ] and once in [ω1 , 2ω1 ]. Similarly, it takes all real values smaller than e3
twice in the segment [0, 2ω2 ] on the imaginary axis.
• when there is only one real root, ℘ takes all real values larger than e2 twice in
the segment [−ω1 − ω2 , ω1 + ω2 ]. Similarly, it takes all real values smaller than
e2 twice in the segment [ω2 − ω1 , ω1 − ω2 ] on the imaginary axis.
Therefore in the case of one real root, the function ℘ assumes all real values exactly
twice in the cell of figure 3 at positions on the real and imaginary axes. Indeed, only e2
appears once, as it appears only at positions congruent to each other. Consequently,
the function ℘ cannot assume any real value at any other point within the cell, and,
thus all positions where ℘ is real on the complex plane are congruent to a point either
on the real or the imaginary axis.
30
In the case there are three roots, the function ℘ assumes all real values larger than
e1 or smaller than e3 exactly twice in the cell at positions on the real and imaginary
axes. This means that there are other positions within the cell, where ℘ is real and it
assumes all real values between e3 and e1 . We already know such a point, namely ω3 ,
where the function ℘ assumes the value e2 . It is a natural guess that ℘ is real on the
horizontal and vertical axes passing through ω3 . This is indeed true. Assuming that
x ∈ R,
℘ (ix + ω1 ) = ℘ (−ix + ω1 ) = ℘ (ix − ω1 ) = ℘ (ix − ω1 + 2ω1 ) = ℘ (ix + ω1 ) , (3.8)

℘ (x + ω2 ) = ℘ (x − ω2 ) = ℘ (x − ω2 + 2ω2 ) = ℘ (x + ω2 ) . (3.9)
Thus, ℘ assumes the values between e2 and e1 twice in the segment [ω1 , ω1 + 2ω2 ]
on the shifted imaginary axis and the values between e3 and e2 exactly twice in the
segment [ω2 , 2ω1 + ω2 ] on the shifted real axis. Thus, all real numbers appear twice
on the real, shifted real, imaginary and shifted imaginary axes except for the roots
that appear only once. Therefore, the function ℘ cannot be real at any other position
within the cell. All positions in the complex plane, where ℘ assumes real values have
to be congruent to point on these four segments. Figure 4 displays these positions.
Imz Imz
ω2 2ω1 − 2ω2
ω2 ω1 − ω2
0 ω1 2ω1 Rez 0 ω1 + ω2 2ω1 + 2ω2 Rez

Figure 4 – The locus of complex numbers z for whom ℘ (z) is real
Real Solutions of Weierstrass Equation in the Real Domain

It is now simple to find what are the real solutions in the real domain of the equation
2
dy
= 4y 3 − g2 y − g3 , (3.10)
dx
31
where g2 and g3 are real. We know that in the complex domain, the general solution
of this equation is
y = ℘ (x − z0 ) , (3.11)
where z0 ∈ C. In our problem x and y has to be real. However, z0 is a constant of
integration and has no physical meaning. It can assume any complex value, as long as
y is real for any real x. This is equivalent to selecting any line on the complex plane
that is parallel to the real axis. It is evident in figure 4 that when there is only one
real root, all such lines are congruent to the real axis itself. On the contrary, when
there are three real roots, there are two options, they are congruent to the real axis,
or the real axis shifted by ω2 . Consequently, the general real solution of (3.10) in the
real domain is
y = ℘ (x − x0 ) , (3.12)
where x0 ∈ R, when there is one real root and
y = ℘ (x − x0 ) or y = ℘ (x − x0 + ω2 ) , (3.13)
where x0 ∈ R, when there are three real roots; the appropriate choice depends on initial
conditions, namely whether the initial value of y lies within [e1 , ∞) or [e3 , e2 ].
3.2 Point Particle in a Cubic Potential

Problem Definition
Let’s now consider a point particle moving in one dimension under the influence of a
force that is a quadratic function of position. We select the origin of the coordinate
system as the position of extremal force and we select units, such that the mass of the
particle equals 2 and the coefficient of the quadratic term in the force equals 12. The
equation of motion is written as
2ẍ = F0 + 12x2 . (3.14)
This equation can be integrated once to the form of conservation of energy. Fixing
the integration constant so that the potential vanishes at x = 0 we get
ẋ2 + V (x) = E, V (x) = −F0 x − 4x3 (3.15)
or
ẋ2 = 4x3 + F0 x + E. (3.16)
Obviously, there is no local minimum of the potential when F0 > 0. In this case
all motions of the problem are scattering solutions evolving from +∞ to a minimum
x and back to +∞. On the contrary, when F0 < 0 there is a local maximum at
32
p
x = x0 ≡ −F0 /12 and a local minimum at x = −x0 , and, thus, a range of values
for the energy E, namely,
3
F0 2
|E| < E0 , E0 = − , (3.17)
3
for which the equation V (x) = E has three roots and consequently there are two
possible kinds of motion. One of them is a scattering solution evolving from +∞ to a
minimum x and back to +∞ and the other is an oscillating solution in the region of
the local minimum of the potential.
V (x) V (x)
E0
√ √ x x
− 3x0 −x0 x0 3x0
−E0
F0 < 0 F0 > 0
V (x) V (x)
E
E
x x
e3 e2 e1 e2
|E| < E0 |E| > E0
Figure 5 – The cubic potential for F0 < 0 (top left) and F0 > 0 top right. In
the case F0 < 0, there may exist one (bottom right) or two (bottom left) possible
motions depending on the value of the energy.
Problem Solution
In the language of classical mechanics, it now becomes obvious why the Weierstrass
differential equation has two independent real solutions when there are three real roots
33
and only one when there is one real root. The roots play the role of the extrema of
the motion, which are indeed the positions the velocity vanishes, as required for the
roots. Furthermore, the solution that always exists is the one defined on the real axis,
x = ℘ (t − t0 ; −F0 , −E) , (3.18)
which contains the pole, and, thus, it corresponds to the scattering solution. The
solution on the shifted real axis,
x = ℘ (t − t0 + ω2 ; −F0 , −E, ) , (3.19)
is bounded between e3 and e2 and corresponds to the oscillating solution in the region
of the local minimum. Finally, we would like to commend that from the point of
view of classical mechanics, it is natural that the Weierstrass elliptic function (or
more literally the solution of Weierstrass equation) has order equal to two. In every
position there are two possible initial conditions that correspond to the same energy,
depending on the direction of the initial velocity. This is mirrored in the fact that the
same real value appears in two non-congruent positions in every cell. This does not
apply only at the extrema of motion, where the appropriate initial velocity vanishes
and indeed these correspond to the roots of the Weierstrass function, which appear
only once in every cell.
Finally, let’s calculate the “time of flight” for scattering solutions and the period
of the oscillating solutions. The first is the distance between two consecutive poles on
the real axis, which obviously equals
Tscattering = 2ω1 . (3.20)
Similarly, the period of the oscillating solutions is the distance between two consecutive
appearances of the same root. This is also clearly
Toscillating = 2ω1 . (3.21)
Therefore, for the energies that an oscillating solution exists, the “time of flight” of
the scattering solution and the period of the oscillating solution with the same energy
are equal.
The Role of the Imaginary Period

In the above we found that the Weierstrass elliptic function naturally describes the
motion of a point particle in a cubic potential. The real period of Weierstrass function
(2ω1 in the case of three real roots and 2ω3 in the case of one real root) plays the role
of the “time of flight” or the period of the motion. Is there any physical meaning for
the imaginary period?
34
It is easy to answer this question using the homogeneity transformation (1.41).
This relation with µ = i implies that
℘ (iz; g2 , g3 ) = −℘ (z; g2 , −g3 ) . (3.22)
It is a direct consequence that ℘ (iz; g2 , g3 ) obeys the differential equation,

2
d℘ (iz; g2 , −g3 )
= −4℘3 (iz; g2 , −g3 ) + g2 ℘ (iz; g2 , −g3 ) − g3 .
dz
Selecting g2 = −F0 and g3 = E, we find that the function ℘ (iz; −F0 , −E) obeys the
differential equation,
2
d℘ (iz; −F0 , −E)
= −4℘3 (iz; −F0 , −E) − F0 ℘ (iz; −F0 , −E) − E
dz
and thus it is a solution to yet another one dimensional point particle problem, namely,
ẏ 2 + Ṽ (y) = −E, Ṽ (y) = y 3 + F0 y = −V (y) , (3.23)
which clearly is the problem of motion of a point particle in the inverted potential to
that of the initial problem, having the opposite energy.
V (x) Ṽ (y)
E
e3 e2 e1
x y
e3 e2 e1
E
Figure 6 – The original and inverted point particle problems
It is obvious that in this problem, the point particle is moving in the complementary
region than that of the initial problem. Furthermore, since the position of the particle
has to be real, it turns out that the unbounded motion of the point particle under the
influence of the potential Ṽ is given by
y = ℘ (i(t − t0 ) ; −F0 , −E) , (3.24)
35
while if there is a bounded one, it is given by
y = ℘ (i(t − t0 ) + ω1 ; −F0 , −E, ) . (3.25)
The physical meaning of the imaginary period is now obvious. The “time of flight”
of the unbounded motion, as well as the period of the bounded motion in this inverted
problem are given by the imaginary period of ℘ (z; −F0 , −E, ),
T̃scattering = T̃oscillating = −2iω2 . (3.26)
3.3 The Simple Pendulum

The Weierstrass elliptic function naturally describes a point particle in a cubic poten-
tial, due to the fact that the conservation of energy takes the form of the Weierstrass
equation (1.29). Its applications though are not limited to this problem. There are
several other problems with one degree of freedom that can be transformed to that of
a cubic potential with an appropriate coordinate transformation.
Problem Definition and Equivalence to Weierstrass Equation

One simple and important problem that can be transformed to a cubic potential
problem is the simple pendulum. The equation of motion reads
θ̈ = −ω 2 sin θ. (3.27)
It can be integrated once to take the form of energy conservation,

1 2
θ̇ + V (θ) = E, V (θ) = −ω 2 cos θ. (3.28)
2
The potential V (θ) is plotted in figure 7. The form of the potential indicates that
• there are no solutions for E < −ω 2 ,
• there are oscillating solutions for |E| < ω 2 ,
• there are continuously rotating solutions for E > ω 2 .
We perform the change of variable
E
− ω 2 cos θ = 2y + . (3.29)
3
Then, the conservation of energy takes the form
2 !
E2

E E
ẏ 2 = 4y 3 − 4
+ω y− − ω4 . (3.30)
3 3 3
36
V (x)
ω2
−ω 2
Figure 7 – The simple pendulum potential
This is the standard form of Weierstrass differential equation. The solution for y of
course should be real, but we should also ensure that

2y + E < ω 2 ,

(3.31)
3
so that the change of variable (3.29) leads to a real θ.
Problem Solution and Classification of Solutions

The roots of the cubic polynomial in the right hand side of equation (3.30) turn out
to acquire quite simple expressions,
E ω2 E ω2

E
Q (y) = 4 y − y+ − y+ + . (3.32)
3 6 2 6 2
The three roots are real for all values of the energy constant E. They equal
E E ω2 E ω2
x1 := , x2 := − + , x3 := − − . (3.33)
3 6 2 6 2
We use the notation xi for the roots as given by equations (3.33) to reserve the notation
ei for the roots appropriately ordered.
As there are always three real roots, there are always two independent real solutions
of equation (3.30). They are
y = ℘ (t − t0 ; g2 (E) , g3 (E)) , (3.34)

y = ℘ (t − t0 + ω2 ; g2 (E) , g3 (E)) , (3.35)
37
where 2 !
E2

E E
g2 (E) = + ω4, g3 = −ω 4
. (3.36)
3 3 3
The ordering of the three roots xi depends on the value of the energy constant E, as
shown in figure 8. As the roots ei have to be appropriately ordered, the identification
xi
x1
−ω 2
E x2
ω2
x3
Figure 8 – The roots of the cubic polynomial (3.30) as function of the energy
constant E
of xi with ei depends on the value of the constant E. The appropriate assignments of

the three roots are summarized in table 1.
ordering of roots
E < −ω 2 e1 = x2 , e2 = x3 , e3 = x1
|E| < ω 2 e1 = x2 , e2 = x1 , e3 = x3
E > ω2 e1 = x1 , e2 = x2 , e3 = x3
Table 1 – The possible orderings of the roots xi
The unbounded solution (3.34) ranges from e1 to infinity, whereas the bounded
one (3.35) ranges between e3 and e2 . Using the appropriate assignment of roots for
each energy range, we find that the range of −ω 2 cos θ = 2y + E/3 is given in table 2.
Table 2 clearly implies that the unbounded solution never corresponds to a real θ.
The bounded solution corresponds to a real solution only when E > −ω 2 . Thus, as
38
unbounded range bounded range
2 2
E < −ω [ω , +∞) [E, −ω 2 ]
|E| < ω 2 [ω 2 , +∞) [−ω 2 , E]
E > ω2 [E, +∞) [−ω 2 , ω 2 ]
Table 2 – The range of −ω 2 cos θ
expected by the form of the potential, the pendulum problem has a real solution only
when E > −ω 2 and they are given by the bounded solution (3.35). There is naturally
a qualitative change of the form of the solutions at E = ω 2 , which is mirrored in the
ordering of the roots. The time evolution of θ is sketched in table 3. The period of
unbounded bounded
θ (0) θ (ω) θ (2ω) θ (0) θ (ω) θ (2ω)
E < −ω 2 – –
± π − arccos ωE2

|E| < ω 2 – 0 0
E > ω2 – 0 π 0
Table 3 – The range of the elliptic solutions of the simple pendulum equation
the oscillatory motions is

Toscillating = 4ω1 . (3.37)
while the period (or more literally the quasi-period) of the rotating motions is
Trotating = 2ω1 . (3.38)
The difference between the two expressions is due to the change of the topology of
the solution. A half-period of the solution corresponds to the transition from the
equilibrium position to the maximum displacement position. In the case of oscillatory
motion four such segments are required to complete a period, as after two segments
the pendulum is back at the equilibrium position but with inverted velocity. On the
contrary, in the case of rotating solution, the maximum displacement equals π, the
velocity is never inverted, and, thus, only two half-periods are required.
Since the solution is single valued for cos θ, in order to find an analytic expression
for θ, one has to match appropriate patches, so that the overall solution is everywhere
39
continuous and smooth. It is not difficult to show that selecting initial conditions, so
that θ (0) = 0 and θ̇ (0) > 0, the appropriate expression for the angle theta is
 j k
t
(−1) 2ω1 arccos 12 2℘ (t + ω2 ) + E , E < ω2,
ω 3
θ= j k (3.39)
(−1) ωt1 arccos 1 2℘ (t + ω ) + E + 2π t + 1 , E > ω 2 .
j k
ω2 2 3 2ω1 2
It is left as an exercise for the reader to verify that the above expressions are everywhere
continuous and smooth.
3.4 Point Particle in Hyperbolic Potential

Problems Definition and Equivalence to Weierstrass Equation
Now let’s consider the case of a point particle moving in one dimension under the
influence of a hyperbolic force. There are four such possible cases, namely,
ẍ = ω 2 sinh x, (3.40)
2
ẍ = −ω sinh x, (3.41)
ẍ = ω 2 cosh x, (3.42)
2
ẍ = −ω cosh x. (3.43)
We will group all those four cases and study them at once writing the equation of
motion as
ex + te−x
ẍ = −sω 2 . (3.44)
2
The parameters s and t take the values ±1. Appropriate selection of s and t results
in any of the four possible hyperbolic forces. Equation (3.44) can be integrated once
to yield
1 2 ω2 x
e − te−x .

ẍ + V (x) = E, V (x) = s (3.45)
2 2
The potential energy is plotted in figure 9 for all the four cases that we are studying.
Considering the form of the potential, we obtain a qualitative picture for the behaviour
of the solutions. In the case s = +1 and t = −1, we expect to find oscillating
solutions with energy E > m2 and no solutions for E < m2 . In the case s = −1 and
t = −1, we expect to find two different classes of solutions; for E < −m2 we expect
to find reflecting scattering solutions since the effective particle does not have enough
energy to overcome the potential barrier, whereas for E > −m2 , we expect to find
transmitting scattering solutions since the particle overcomes the potential barrier.
Finally, In the case t = +1, we expect to find reflecting scattering solutions for all
energies.
40
V (x)
ẍ = −ω 2 sinh x
m2 ẍ = −ω 2 cosh x
x
ẍ = ω 2 sinh x
−m2
ẍ = ω 2 cosh x
Figure 9 – The potential for the four cases of a hyperbolic force
Performing the change of variable
ω2 x E
−s e = 2y − , (3.46)
2 3
equation (3.45) takes the standard Weierstrass form
ω4 ω4

2 3 1 2 E 1 2
ẏ = 4y − E +t y+ E +t . (3.47)
3 4 3 9 8
The change of variable (3.46) transforms the problem of the motion of a particle under
the influence of a hyperbolic force to yet another one-dimensional problem, describing
the motion of a particle with zero energy under the influence of a cubic potential, which
has already been studied in section 3.2. Real solutions of this equation correspond to
real solutions of the initial variable x only when 2y − E3 has the same sign as s.
Four Problems Solved by the Same Expression

It is interesting to understand how the same equation can be used to describe a variety
of solutions that exhibit qualitatively different behaviour, as suggested by the form of
the potential in the four cases of hyperbolic forces under study.
Equation (3.47) is of the standard Weierstrass form (1.29) with a specific selection
for the constants g2 and g3 . Equation (3.47) is solved by
y = ℘ (t; g2 (E, t) , g3 (E, t)) , (3.48)

y = ℘ (t + ω2 ; g2 (E, t) , g3 (E, t)) , (3.49)
41
bearing in mind that the second solution is valid only when there are three real roots.
The coefficients g2 and g3 are given by
ω4 ω4

1 2 E 1 2
g2 (E, t) = E + t , g3 (E, t) = − E +t (3.50)
3 4 3 9 8
and the related cubic polynomial is
ω4 ω4

3 1 2 E 1 2
Q (x) = 4x − E +t x+ E +t . (3.51)
3 4 3 9 8
The roots of the cubic polynomial are easy to obtain noting that x = E/6 is one
of them. Thus, all three roots of Q (x) are
E E 1p 2
x1 = , x2,3 = − ± E + tω 4 . (3.52)
6 12 4
In the following, we use the notation xi for the roots of Q (x) as written in equations
(3.52) and reserve the notation ei for the ordered roots of Q (x) as described in section
3.1. The roots xi are plotted as functions of the energy constant E in figure 10.
xi xi
x1
−m2 E −m2 E
x2
m 2 m2
x3
cosh force sinh force
Figure 10 – The roots of the cubic polynomial (3.51) as function of the energy
constant E
The Weierstrass function allows for a unifying description of the elliptic solutions of
both sinh and cosh forces. Different classes of solutions simply correspond to different
ordering of the roots xi . Figure 10 suggests that there are four distinct cases for the
ordering of the three roots xi , which are summarized in table 4.
The unbounded solution ranges from e1 to infinity when there are three real roots
and from e2 to infinity when there is only one real root, whereas the bounded solution
42
reality of roots ordering of roots
t = +1 3 real roots e1 = x2 , e2 = x1 , e3 = x3
t = −1, E > ω 2 3 real roots e1 = x1 , e2 = x2 , e3 = x3
t = −1, |E| < ω 2 1 real, 2 complex roots e1 = x2 , e2 = x1 , e3 = x3
t = −1, E < −ω 2 3 real roots e1 = x2 , e2 = x3 , e3 = x1
Table 4 – The possible orderings of the roots xi
2
−s ω2 ex unbounded range bounded range
t = +1 2 (y − e2 ) [2 (e1 − e2 ) , +∞) [−2 (e2 − e3 ) , 0]
t = −1, E > ω 2 2 (y − e1 ) [0, +∞) [−2 (e1 − e3 ) , −2 (e1 − e2 )]
t = −1, |E| < ω 2 2 (y − e2 ) [0, +∞) –
t = −1, E < −ω 2 2 (y − e3 ) [2 (e1 − e3 ) , +∞) [0, 2 (e2 − e3 )]
2
Table 5 – The solutions for −s ω2 ex and their range
2
ranges from e3 to e2 . Then, using equation (3.46), −s ω2 ex = 2 (y − x1 ), we can explore
2
the range of −s ω2 ex in all cases. The results are summarized in table 5. In all cases,
2
the sign of −s ω2 ex does not alternate within its range. Consequently, each solution
corresponds to a real solution for exactly one value of the sign s.
In table 6, we sketch the evolution of each solution. In this table 2Ω stands for the
real period of y, which is equal to 2ω1 when there are three real roots and 2 (ω1 + ω2 )
when there is only one real root.
All solutions take the following form

2 E
x (t) = ln −s 2 2℘ (t + z0 ; g2 (E) , g3 (E)) − , (3.53)
ω 3
for all choices of the overall sign s and an appropriate choice of the complex integration
constant z0 . In particular:
• In the case s = −1 and t = +1, as expected from the form of the potential,
only reflecting solutions, coming from and going to the right are found for all
energies. In this case there are always three real roots and the solution is given
by the unbounded solution. If we select initial conditions, so that the particle is
at the minimum position at t = t0 , we need to select z0 = ω1 − t0 . The “time of
flight” equals T = 2ω1 .
43
unbounded bounded
x (0) x (Ω) x (2Ω) x (0) x (Ω) x (2Ω)
s = −1
4(e1 −e2 )
t = +1 +∞ ln ω2
+∞ –
t = −1, E > ω 2 +∞ −∞ +∞ –
t = −1, |E| < ω 2 +∞ −∞ +∞ –
t = −1, E < −ω 2 +∞ ln 4(e1ω−e
2
3)
+∞ −∞ 2 −e3 )
ln 4(em 2 −∞
s = +1
t = +1 – ln 4(e2ω−e
2
3)
−∞ ln 4(e2ω−e
2
3)
t = −1, E > ω 2 – ln 4(e1ω−e

2
2)
ln 4(e1ω−e
2
3)
ln 4(e1ω−e
2
2)
t = −1, |E| < ω 2 – –

t = −1, E < −ω 2 – –
Table 6 – The extrema of the motion
• In the case s = +1 and t = +1, as expected from the form of the potential,
only reflecting solutions, coming from and going to the left exist for all energies.
In this case there are always three real roots and the solution is given by the
bounded solution. Selecting initial conditions, so that the particle is at the
maximum position at t = t0 , we have to select z0 = ω2 − t0 . The “time of flight”
equals T = 2ω1 .
• In the case s = −1 and t = −1, the form of the potential suggests that there are
two possible cases depending on the energy.
– When E < −ω 2 , the particle does not have enough energy to overcome
the potential barrier. Therefore, there are reflecting solution coming from
and going to either of the two directions. In this case there are three real
roots. The particles coming from the right are described from the unbounded
solution, while the particles coming from the left are described from the
bounded solution. Selecting initial condition, so that the particle is at the
extremal position at t = t0 requires the selection z0 = ω1 −t0 for the particles
coming from the right and z0 = ω1 + ω2 − t0 for the particles coming from
the left. The “time of flight” in both cases equals T = 2ω1 .
– When E > −ω 2 , the particle has enough energy to overcome the potential
barrier. Therefore, there are two classes of transmitting solutions coming
from either direction. It is interesting though that the reality of the three
roots depends on whether the energy is smaller or larger than the critical
44
value ω 2 .
∗ When E < ω 2 , there is only one real root and both left-incoming and
right-incoming particles are described by the unbounded solution. The
“time of flight” in both cases equals T = ω3 . Selecting initial conditions
such that the particle at t = t0 lies at the origin, one should select
z0 = ω3 /2 − t0 for a particle coming from the right and z0 = 3ω3 /2 − t0
for a particle coming from the left.
∗ When E > ω 2 , there are three real roots. Apart from this, the situation
is similar to the case E < ω 2 , with the substitution of ω3 with ω1 .
• Finally, when s = +1 and t = −1, there are oscillatory solutions only when E >
ω 2 , as expected by the form of the potential. The period of the oscillation equals
T = 2ω1 . Selecting initial conditions so that the particle lies at its minimum
position at t = t0 yields z0 = −t0 .
45
Problems
Problem 3.1. Find the energies for which the solution for the cubic potential degen-
erates to a Weierstrass function with a double root, thus a simply periodic function.
Find the special expressions and the periods of the motion at these limits. Verify that
in one of these limits, the unique period is equal to the period of small oscillations in
the region of the local minimum.
Problem 3.2. In section 3.2, we showed that while the real period of the Weierstrass
elliptic function equals the “time of flight” or the period of the motion of a point
particle with energy E in a cubic potential, the absolute value of the imaginary period
equals the corresponding quantities for the motion of a point particle with energy −E
in the inverted potential. However, the inverted potential is identical to the initial
one if a coordinate reflection is performed. Thus, the absolute value of the imaginary
period should also equal the “time of flight” or period of the motion of a point particle
with energy −E in the initial potential. Verify this using expressions (3.1), (3.2), (3.3)
and (3.4)
Problem 3.3. In section 3.2, we showed that when a bounded oscillatory motion exists
in a cubic potential, its period equals the “time of flight” of the scattering solution for
the same energy. This happens due to a discrete symmetry of the conservation of
energy equation that exists when there are three real roots.
More specifically, find how the conservation of energy (3.16) is transformed under
the change of variables
(e3 − e1 ) (e3 − e2 )
x → e3 + .
y − e3
Then, find where the segments (−∞, e3 ], [e3 , e2 ], [e2 , e1 ] and [e1 , +∞) are mapped
through this coordinate transformation.
Finally, show that the above imply that the period of the oscillatory motion and the
“time of flight” of the scattering solution with the same energy are equal.
Problem 3.4. Find the energies for which the solution to the pendulum problem is
expressed in terms of the Weierstrass elliptic function with a double root. Find the
special expressions for the pendulum motion at these limits and verify that in one of
the limits the solution degenerates to the stable equilibrium and simultaneously the
period of motion equals the period of the small oscillations, 2π/ω.
Problem 3.5. For the motion of a particle in a hyperbolic potential, in the case
s = −1, t = −1 and E > −ω 2 , which corresponds to the case of transmitting scattering
solutions by a repulsive potential barrier, show that the motion is symmetric around
the instant t = 0, namely show that x (−t) = −x (t).
46
Obviously in the case of reflecting scattering solutions, E < −ω 2 , the symmetry
of the problem implies that x (−t) = x (t). What has to change in your previous
derivation in this case?
Problem 3.6. For the motion of a particle in a hyperbolic potential, in the case
s = +1, t = −1, which corresponds to oscillatory motion, the bounds of motion in
table 6 look asymmetric. However, the fact that the potential is even suggests that
they should be symmetric. Verify that they actually are. Furthermore, verify that
x (t + T /2) = −x (t), where T is the period of the oscillation.
Show that at the double root limit of the solution, the solution degenerates to the
equilibrium solution and that the period of motion tends to the period of the small
oscillations 2π/ω.
47
Lecture 4: Applications in Quantum Mechanics
It is well known that particles that move in a periodic potential
V (x + a) = V (x) (4.1)
accept as eigenfunctions Bloch wave solutions of the form
ψ (x) = eikx u (x) , (4.2)
where u (x) is periodic with the same period as the potential
u (x + a) = u (x) . (4.3)
The parameter k is a function of the energy. When k is real, these wavefunctions are
delta-function normalizable, while when k contains an imaginary part the wavefunc-
tions are exponentially diverging. This fact leads to the band structure of the periodic
potentials.
Although these are well known facts, it is not simple to find an analytically solvable
periodic potential that demonstrates the formation of an non-trivial band structure.
4.1 The n = 1 Lamé Potential

The n = 1 Lamé Potential and its Solutions
Let’s consider the periodic potential
V (x) = 2℘ (x) , (4.4)
where it is assumed that the moduli g2 and g3 are considered to be real. The
Schrödinger equation reads
d2 y
− + 2℘ (x) y = λy. (4.5)
dx2
Consider the functions
σ (x ± a) −ζ(±α)x
y± (x; a) = e . (4.6)
σ (x) σ (±a)
It is easy to verify by direct computation that these functions are both eigenfunctions
of the Schrödinger problem (4.5). Using the defining property of the Weierstrass ζ
and σ functions (2.1) and (2.8), we find
dy±
= (ζ (x ± a) − ζ (x) ∓ ζ (α)) y± ,
dx
d2 y± h 2
i
= (ζ (x ± a) − ζ (x) ∓ ζ (α)) − (℘ (x ± a) − ℘ (x)) y± .
dx2
48
Applying the addition theorems of Weierstrass ℘ and ζ functions (2.18) and (2.20),
the above relations can be simplified to the form
dy± 1 ℘0 (x) ∓ ℘0 (a)
= y± , (4.7)
dx 2 ℘ (x) − ℘ (a)
d2 y±
= (2℘ (x) + ℘ (a)) y± . (4.8)
dx2
The last equation implies that y± are eigenfunctions of problem (4.5), both corre-
sponding to the eigenvalue
λ = −℘ (a) . (4.9)
As long as the eigenfunction modulus a is not equal to any of the three half-periods,
the two σ functions appearing to the numerator of y± do not have roots at congruent
positions. As such the two wavefunctions are linearly independent and they provide
the general solution. When the modulus a equals any of the half-periods though, it
turns out that q
y± (x; ω1,2,3 ) = ℘ (x) − e1,3,2 . (4.10)
For those eigenvalues, the second linearly independent solution is given by
q
ỹ (x; ω1,2,3 ) = ℘ (x) − e1,3,2 (ζ (x + ω1,2,3 ) + e1,3,2 x) . (4.11)
Reality of the Solutions

We would like to study whether the eigenfunctions (4.6) are real or not. First, we
consider the case of three real roots. In this case, ℘ (a) will assume all real values if a
runs in the perimeter of the rectangle with corners located at 0, ω1 , ω2 and ω3 . Since
the Weierstrass elliptic function is of order two, for every point in the perimeter of
this rectangle there is another point in the fundamental period parallelogram, where
℘ assumes the same value. Due to ℘ being an even function, this point is congruent
to the opposite of the initial one. Therefore, the selection of the other point does not
correspond to new eigenfunctions, but simply corresponds to the reflection y+ ↔ y− .
As a result, we divide our analysis to four cases, one for each side of the rectangle with
corners at the origin and the half-periods. In the following, b is considered always real.
1. a lies in the segment [0, ω1 ]. Then, a = b and ℘ (a) > e1 . In this case, trivially,
the eigenfunctions (4.6) are real as,

y± (x; b) = y± x; b̄ = y± (x; b) . (4.12)
2. a lies in the segment [0, ω2 ]. Then, a = ib and ℘ (a) < e3 . In this case, trivially,
the eigenfunctions (4.6) are complex conjugate to each other as,

y± (x; ib) = y± x; ib = y± (x; −ib) = y∓ (x; ib) . (4.13)
49
3. a lies in the segment [ω2 , ω3 ]. Then, a = ω2 + b and e3 < ℘ (a) < e2 . In this case,

y± (x; ω2 + b) = y± x; ω2 + b = y± (x; −ω2 + b) .
We use the quasi-periodicity properties of functions ζ (2.5) and σ (2.12) to find

that
σ (x ∓ ω2 ± b) ∓ζ(−ω2 +b)x
y± (x; −ω2 + b) = e
σ (x) σ (∓ω2 ± b)
−σ (x ± ω2 ± b) e∓2ζ(ω2 )(x±ω2 ±b∓ω2 ) ∓(ζ(ω2 +b)−2ζ(ω2 ))x
= e
−σ (x) σ (±ω2 ± b) e∓2ζ(ω2 )(±ω2 ±b∓ω2 )
σ (x ± ω2 ± b) ∓ζ(ω2 +b)x
= e = y± (x; ω2 + b) ,
σ (x) σ (±ω2 ± b)
implying that
y± (x; ω2 + b) = y± (x; ω2 + b) , (4.14)
meaning that in this case the eigenfunctions (4.6) are real.
4. a lies in the segment [ω1 , ω3 ]. Then, a = ω1 + ib and e2 < ℘ (a) < e1 . In this
case,

y± (x; ω1 + ib) = y± x; ω1 + ib = y± (x; ω1 − ib) .
As in previous case, we use the quasi-periodicity properties of ζ and σ to find
σ (x ± ω1 ∓ ib) ∓ζ(ω1 −ib)x

y± (x; ω1 − ib) = e
σ (x) σ (±ω1 ∓ ib)
−σ (x ∓ ω1 ∓ ib) e±2ζ(ω1 )(x∓ω1 ∓ib±ω1 ) ∓(ζ(−ω1 −ib)+2ζ(ω1 ))x
= e
−σ (x) σ (∓ω1 ∓ ib) e±2ζ(ω1 )(∓ω1 ∓ib±ω1 )
σ (x ∓ ω1 ∓ ib) ±ζ(ω1 +ib)x
= e = y∓ (x; ω1 + ib) ,
σ (x) σ (∓ω1 ∓ ib)
meaning that
y± (x; ω1 + ib) = y∓ (x; ω1 + ib) (4.15)
or in other words, in this case, the eigenfunctions (4.6) are complex conjugate to
each other.
In the case of one real root, the situation is much simpler. ℘ (a) will assume all
real values if a runs in the union of two segments, one on the real axis with endpoints
0 and ω1 + ω2 and one in the imaginary axis with endpoints 0 and ω1 − ω2 . Similarly
to the case of three real roots, there are more points where ℘ assumes real values, but
their selection corresponds simply to the reflection y+ ↔ y− .
1. a lies in the segment [0, ω1 + ω2 ]. Then, a = b and ℘ (a) > e2 . This case is
identical to the first case above, and, thus, the eigenfunctions (4.6) are real.
50
2. a lies in the segment [0, ω1 − ω2 ]. Then, a = ib and ℘ (a) < e2 . This case
is identical to the second case above, and, thus, the eigenfunctions (4.6) are
complex conjugate to each other.
The Band Structure of the n = 1 Lamé Potential: Three Real Roots

Comparing to the flat potential, we would expect that when the eigenfunctions (4.6)
are complex conjugate to each other, they are delta function normalizable Bloch waves,
whereas when the eigenfunctions (4.6) are real, they are exponentially diverging non-
normalizable states. However, in order to explicitly show that, we need to find how
the eigenfunctions (4.6) transform under a shift of their argument by a period of the
potential. For this purpose, we need to write the eigenfunctions (4.6) in the form
(4.2). Let us consider the case of three real roots. Then, the period of the potential
equals 2ω1 and using the quasi-periodicity property of Weierstrass sigma function,
σ (x ± a + 2ω1 ) −e2ζ(ω1 )(x±a+ω1 ) σ (x ± a) ±2aζ(ω1 ) σ (x ± a)

= = e .
σ (x + 2ω1 ) σ (±a) −e2ζ(ω1 )(x+ω1 ) σ (x) σ (±a) σ (x) σ (±a)
Thus, we may write the eigenfunctions (4.6) as
y± (x; a) = u± (x; a) e±ik(a)x , (4.16)
where
σ (x ± a) ∓ aζ(ω 1) x aζ (ω1 ) − ω1 ζ (a)
u± (x; a) = e ω1 , ik (a) = (4.17)
σ (x) σ (a) ω1
and it holds that u± (x + 2ω1 ; a) = u± (x; a).
In order to discriminate non-normalizable states from Bloch waves, we need to
study the function
f (a) = aζ (ω1 ) − ω1 ζ (a) . (4.18)
This is clearly not an elliptic function, but rather a quasi-periodic function, since it
obeys,
f (a + 2ω1 ) = (a + 2ω1 ) ζ (ω1 ) − ω1 (ζ (a) + 2ζ (ω1 )) = f (a) , (4.19)

f (a + 2ω2 ) = (a + 2ω2 ) ζ (ω1 ) − ω1 (ζ (a) + 2ζ (ω2 )) = f (a) + iπ. (4.20)
However, its derivative equals
f 0 (a) = ζ (ω1 ) + ω1 ℘ (a) , (4.21)
which is clearly an order two elliptic function. Therefore, the function f (a) is station-
ary exactly twice in each cell. Furthermore, f 0 (a) is real, wherever ℘ (a) is real, thus,
in the space where the moduli a takes values.
51
The function f (a) takes the following values at the origin and the half-periods,
ω1
lim f (b) = − lim = −∞, (4.22)
b→0+ b→0+ b
ω1
lim f (ib) = − lim = +i∞, (4.23)
b→0+ b→0+ ib
f (ω1 ) = ω1 ζ (ω1 ) − ω1 ζ (ω1 ) = 0, (4.24)
π
f (ω2 ) = ω2 ζ (ω1 ) − ω1 ζ (ω2 ) = i , (4.25)
2
π
f (ω3 ) = (ω1 + ω2 ) ζ (ω1 ) − ω1 ζ (ω1 + ω2 ) = i . (4.26)
2
Since the derivative of f (a) is real at the perimeter of the rectangle with corners
the origin and the three half-periods, it follows that in the sides [0, ω1 ] and [ω2 , ω3 ]
only the real part of f (a) varies, whereas in the sides [0, ω2 ] and [ω1 , ω3 ] only the
imaginary part of f (a) varies.
Since the real part of f (a) is identical at a = ω2 and a = ω3 (it vanishes), the
mean value theorem implies that there is a point in the segment [ω2 , ω3 ] where the
derivative of f (a) vanishes. In every cell, there is another point, where the derivative
vanishes, which is congruent to the opposite of the above point, and, thus, it is not
congruent to any point of the perimeter of the rectangle with corners at the origin and
the half-periods. Since the derivative of f (a) is an order two elliptic function, there
is no other point in a cell, and, thus, in the aforementioned rectangle, where f (a) is
stationary.
The above, combined with the values of f (a) at the origin and the half-periods,
imply that
1. At the segment [0, ω1 ], f (a) is everywhere real. It is nowhere stationary in this
segment and therefore, it varies monotonically from −∞ at the origin to 0 at ω1 .
It follows that it vanishes nowhere except at a = ω1 .
2. At the segment [ω1 , ω3 ], f (a) is everywhere purely imaginary. It is nowhere
stationary in this segment and therefore, it varies monotonically from 0 at ω1 to
iπ/2 at ω3 .
3. At the segment [ω2 , ω3 ], f (a) has an imaginary part equal to iπ/2. The real
part vanishes at the endpoints of the segment, it reaches a minimum value at the
stationary point of f (a) and it vanishes nowhere expect of the endpoints, since
the derivative of f (a) vanishes only once.
4. At the segment [0, ω2 ], f (a) is everywhere purely imaginary. It is nowhere sta-
tionary in this segment and therefore, it varies monotonically from +i∞ at the
origin to iπ/2 at ω2 .
It follows that k (a) is purely imaginary, as required for Bloch waves, in the seg-
ments [0, ω2 ] and [ω1 , ω3 ] and nowhere else. Thus, the band structure of the n = 1
52
Lamé potential, in the case of three real roots, contains a finite “valence” band be-
tween the energies −e1 and −e2 and an infinite “conduction” band for energies larger
than −e3 .
The Band Structure of the n = 1 Lamé Potential: One Real Root

In the case of one real root, the period of the potential equals 2ω1 + 2ω2 . We write
the eigenfunctions (4.6) as
y± (x; a) = u± (x; a) e±ik(a)x , (4.27)
where
σ (x ± a) ∓ aζ(ω 1 +ω2 ) x aζ (ω1 + ω2 ) − (ω1 + ω2 ) ζ (a)
u± (x; a) = e ω1 +ω2 , ik (a) = (4.28)
σ (x) σ (a) ω1 + ω2
and u± (x + 2ω1 + 2ω2 ; a) = u± (x; a).

We now have to consider the function
f (a) = aζ (ω1 + ω2 ) − (ω1 + ω2 ) ζ (a) . (4.29)
Similarly to the case of three real roots, the function f (a) is quasi-periodic
f (a + 2ω1 ) = (a + 2ω1 ) ζ (ω1 + ω2 ) − (ω1 + ω2 ) (ζ (a) + 2ζ (ω1 )) = f (a) − iπ, (4.30)

f (a + 2ω2 ) = (a + 2ω2 ) ζ (ω1 + ω2 ) − (ω1 + ω2 ) (ζ (a) + 2ζ (ω2 )) = f (a) + iπ (4.31)
and its derivative equals
f 0 (a) = ζ (ω1 + ω2 ) + (ω1 + ω2 ) ℘ (a) , (4.32)
which is an order two elliptic function, implying that f (a) is stationary exactly once
in the union of the segments [0, ω1 + ω2 ] and [0, ω1 − ω2 ]. Furthermore, the derivative
f 0 (a) is real wherever ℘ (a) is real and therefore only the real part of f (a) varies on the
segment [0, ω1 + ω2 ], whereas only the imaginary part of f (a) varies on the segment
[0, ω1 − ω2 ]. We can easily verify that
ω1 + ω2
lim f (b) = − lim = −∞, (4.33)
b→0+ b→0+ b
ω1 + ω2
lim f (ib) = − lim = +i∞, (4.34)
b→0+ b→0+ ib
f (ω1 + ω2 ) = (ω1 + ω2 ) ζ (ω1 + ω2 ) − (ω1 + ω2 ) ζ (ω1 + ω2 ) = 0, (4.35)
f (ω1 − ω2 ) = (ω1 − ω2 ) ζ (ω1 + ω2 ) − (ω1 + ω2 ) ζ (ω1 − ω2 ) = −iπ, (4.36)
and, thus, f (a) is real on [0, ω1 + ω2 ] and purely imaginary on [0, ω1 − ω2 ].
53
The only question to be answered is whether the stationary point of f (a) belongs
in [0, ω1 + ω2 ] or [0, ω1 − ω2 ]. In the first case, there is another point on [0, ω1 + ω2 ],
where f (a) vanishes, and, thus, it can be considered purely imaginary, while in the
second case there is not. It turns out that the stationary point lies in [0, ω1 − ω2 ] and
therefore f (a) is purely imaginary everywhere on [0, ω1 − ω2 ] and nowhere else. Thus,
the band structure of the n = 1 Lamé potential, in the case of one real roots, contains
only an infinite “conduction” band for energies larger than −e2 .
4.2 The Bounded n = 1 Lamé Potential

The Bounded n = 1 Lamé Potential and its Band Structure
If three real roots exist, the whole process of finding the eigenstates and the band
structure can be repeated for the potential V = 2℘ (x + ω2 ). Unlike the potential of
the previous section, this one is bounded. In a trivial way, the functions
σ (x + ω2 ± a) ∓ζ(a)(x+ω2 )
ψ± (x; a) = y± (x + ω2 ; a) = e
σ (x + ω2 ) σ (a)
obey
d2 ψ± (x; a)
= (2℘ (x + ω2 ) + ℘ (a)) ψ± (x; a) , (4.37)
dx2
and, thus, they are eigenfunctions of the bounded n = 1 Lamé problem, both with an
eigenfunction equal to
λ = ℘ (a) . (4.38)
However, this asymmetric insertion of ω2 has deprived the eigenfunctions from
their nice reality properties. For example, for real a,
σ (x − ω2 + a) −ζ(a)(x−ω2 )
ψ̄+ (x; a) = e
σ (x − ω2 ) σ (a)
−σ (x + ω2 + a) e−2ζ(ω2 )(x+ω2 +a−ω2 ) −ζ(a)(x+ω2 ) 2ζ(a)ω2 (4.39)
= e e
−σ (x + ω2 ) σ (a) e−2ζ(ω2 )(x+ω2 −ω2 )
= e2(ω2 ζ(a)−aζ(ω2 )) ψ+ (x; a) ,
implying that ψ+ (x; a) is not real, but it has to be rotated by a constant complex
phase in order to become so. It is left as an exercise to the audience to show that the
eigenfunctions
σ (x + ω2 ± a) σ (ω2 ) −ζ(±a)x
ψ± (x; a) = e , (4.40)
σ (x + ω2 ) σ (ω2 ± a)
have reality properties similar to those of the eigenfunctions of the unbounded Lamé
potential (4.6). It is also left to the audience to verify that the band structure of the
bounded potential is identical to the band structure of the unbounded one.
54
It is quite interesting that the potentials V = 2℘ (x) and V = 2℘ (x + ω2 ) have the
same band structure. The two potentials are quite dissimilar functions, the unbounded
one having second order poles, whereas the bounded one being smooth, as shown in
figure 11.
V (x) V (x)
2e1 2e2
−e3 2ω1 4ω1 6ω1
−e2 2ω1 4ω1 6ω1
x x
2e2
−e1 −e2
2e3
three real roots 2℘ (x) one real root
2℘ (x + ω2 )
Figure 11 – The band structure of the n = 1 Lamé potential
Connection Between the Bounded and Unbounded n = 1 Lamé Prob-

lems
The fact that the two potentials are isospectral is not a coincidence. Assume the
function
℘0 (x)
W (x) = , (4.41)
2 (℘ (x) − e3 )
which we will call the “superpotential” and the operators
d d
A= + W (x) , A† = − + W (x) , (4.42)
dx dx
which we will call “annihilation” and “creation” operators respectively. Then,
d2
A† A = − + V1 (x) , V1 (x) = W 2 (x) − W 0 (x) , (4.43)
dx2
d2
AA† = − 2 + V2 (x) , V2 (x) = W 2 (x) + W 0 (x) . (4.44)
dx
It is a matter of algebra to show that for the specific superpotential given by (4.41),
it turns out that
V1 (x) = 2℘ (x + ω2 ) + e3 , V2 (x) = 2℘ (x) + e3 . (4.45)
55
In other words,
A† A = H̃ + e3 , AA† = H + e3 , (4.46)
where H is the Hamiltonian of the unbounded n = 1 Lamé problem and H̃n=1 is the
Hamiltonian of the bounded n = 1 Lamé problem.
Now consider an eigenfunction of the unbounded problem with energy E,

Hy = AA† − e3 y = Ey (4.47)
Then, the function ỹ = A† y is an eigenfunction of the bounded problem with the same
energy, since

H̃ ỹ = A† A − e3 A† y = A† AA† − e3 y = A† Hy = A† Ey = E ỹ. (4.48)
Action with the operator A† cannot change a Bloch wave to an exponentially

diverging function or the other way around. As a result, the two Hamiltonians are
isospectral. It is left as an exercise to the audience to show that indeed
A† y± (x; a) = cψ± (x; a) , (4.49)
where y± are given by equation (4.6), ψ± are given by equation (4.40) and c is a
constant.
4.3 Epilogue
The presentation of the n = 1 Lamé problem concludes this set of lectures on appli-
cations of Weierstrass elliptic and related functions in Physics. The applications of
elliptic functions in not constrained in the content of these lectures.
Many more problems in classical mechanics can be analytically solved in terms
of elliptic functions. We may refer to the point particle in a quartic potential, the
spherical pendulum and the symmetric top as three such examples. Furthermore, in
quantum mechanics, potentials of the form
V (x) = n (n + 1) ℘ (x) , (4.50)
where n ∈ Z can be analytically solved in terms of elliptic functions and present

amazing features. Such potentials have a richer band structure containing up to n
finite gaps in their spectrum.
The Lamé equation also appears in other problems of classical Physics, whenever
one expresses the Laplace operator in elliptical coordinates. Actually, this is the
historical reason for the study of this class of linear differential equations. For example,
elliptical functions will appear if one studies the heat diffusion on the surface of a
ellipsoid by rotation, such as the earth surface.
56
Furthermore, elliptic functions find applications in many fields of more modern
physics. Several solutions of very interesting integrable systems, such as the sine-
Gordon, the sinh-Gordon or the Korteweg-de Vries (KdV) equation can be analytically
expressed in terms of elliptic functions. Such solutions can be further used to construct
analytic string solutions in symmetric spaces, such as dS and AdS spaces, as well as
minimal surfaces in hyperbolic spaces. Elliptic functions also emerge naturally when
one calculates string scattering amplitudes for world-sheets that have the topology of
a torus.
Acknowledgements
I would like to thank the organizer of these lectures E. Papantonopoulos and the School
of Applied Mathematics and Physical Sciences of the National Technical University of
Athens for the hospitality.
The research of G.P. is funded by the “Post-doctoral researchers support” action
of the operational programme “human resources development, education and long
life learning, 2014-2020”, with priority axes 6, 8 and 9, implemented by the Greek
State Scholarship Foundation and co-funded by the European Social Fund - ESF and
National Resources of Greece.
57
Problems
Problem 4.1. Show that the special solution (4.11) is indeed an eigenfunction of the
n = 1 Lamé problem, corresponding to an eigenvalue equal to one of the three roots of
the respective Weierstrass elliptic function.
Problem 4.2. Show that the eigenfunctions (4.40) of the bounded n = 1 Lamé poten-
tial have the same reality properties as the eigenfunctions (4.6) of the unbounded one.
Problem 4.3. Find how the eigenfunctions (4.40) of the bounded n = 1 Lamé poten-
tial transform under a shift of their argument by the period of the potential. Once you
have accomplish that, deduce the band structure of the bounded n = 1 Lamé potential.
Problem 4.4. Using the addition theorem of the Weierstrass σ function (2.19), show
that the eigenstates y± , as given by equation (4.6) obey the following “‘normalization”
properties:
y+ y− = ℘ (x) − ℘ (a) , (4.51)

0 0 0
y+ y− − y+ y− = −℘ (a) . (4.52)
How are these properties modified for the eigenfunctions of the bounded n = 1
Lamé potential?
Problem 4.5. Show that the action of the creation operator A† on the eigenstates
of the unbounded n = 1 Lamé problem (4.6) results to the eigenstates of the bounded
n = 1 Lamé problem (4.40), namely
A† y± (x; a) = cψ± (x; a) (4.53)
How does the coefficient c compares with the results of problem 4.4?
58
The Half-Periods for Real Moduli
It is obvious that formulas (1.24) and (1.25) put certain restrictions on the fundamental
periods of Weierstrass elliptic function, if the moduli are real. Unfortunately, the
problem of inverting (1.24) and (1.25) is quite difficult.
For this purpose, we will apply the integral formula for Weierstrass elliptic function
(1.33), for the three half periods. In all cases, we select the integration path along the
real axes.
Following section 3.1, when the moduli g2 and g3 are real, the roots of the cubic
polynomial are either all real or there is one real and two complex ones, the latter
being complex conjugate to each other. In the case there are three real roots, we find
Z ∞
1
ω1 ∼ p dt ≡ x, (A.1)
e1 4 (t − e1 ) (t − e2 ) (t − e3 )
Z −∞
1
ω2 ∼ p dt ≡ y, (A.2)
e3 4 (t − e1 ) (t − e2 ) (t − e3 )
whereas when there is one real and two complex roots
Z ∞
1 0
ω3 ∼ r dt ≡ x , (A.3)
e2 2 2
4 (t − e2 ) (z − Ree1 ) + (Ime1 )
Z −∞
1 0
ω3 ∼ r dt ≡ y . (A.4)
e2
4 (t − e2 ) (z − Ree1 )2 + (Ime1 )2
The integrated function is real everywhere in the range of integration for the first
integral in both cases, whereas it is imaginary everywhere in the range of integration
for the second integral. Therefore,
x, x0 , iy, iy 0 ∈ R. (A.5)
The above imply that when there are three real roots the half period ω1 is congruent
to a real number and the half-period ω2 is congruent to an purely imaginary number,
x = (2m1 + 1) ω1 + 2n1 ω2 , (A.6)

y = 2m2 ω1 + (2n2 + 1) ω2 , (A.7)
whereas when there is only one real root the half period ω3 = ω1 + ω2 is congruent to
both a real and a purely imaginary number,
x0 = (2k1 + 1) ω1 + (2l1 + 1) ω2 , (A.8)

y 0 = (2k2 + 1) ω1 + (2l2 + 1) ω2 , (A.9)
59
where m1 , n1 , m2 , n2 ∈ Z.
The integrated quantity in (A.1) never changes sign in the whole integration range.
Consequently, as the integration limit approaches from infinity to e1 , the integral
monotonically changes from 0 to its final value. This means that x is not just a point
on the real axis that is congruent to the half-period ω1 , but there is no other such
position on the real axis between 0 and x. This implies that 2m1 + 1 and 2n1 are
relatively prime. In a similar manner, the above statement holds for x0 , as well as for
y and y 0 on the imaginary axis, and thus the pairs {2m2 , 2n2 + 1}, {2k1 + 1, 2l1 + 1}
and {2k2 + 1, 2l2 + 1} are also pairs of relatively prime numbers,
gcd (2m1 + 1, 2n1 ) = gcd (2m2 , 2n2 + 1) = 1

gcd (2k1 + 1, 2l1 + 1) = gcd (2k2 + 1, 2l2 + 1) = 1.
We may redefine the periods ω1 and ω2 with a modular transformation, as described

in section 1.2,
ω1 = aω1 0 + bω2 0 ,
ω2 = cω1 0 + dω2 0 ,
where ad − bc = 1. Let us select b = −2n1 and d = 2m1 + 1 in the case of three real
roots and b = −2l1 − 1 and d = 2k1 + 1 in the case of one real root. We find,
x = (ad − bc) ω1 0 ,
y = (2m2 a + (2n2 + 1) c) ω1 0 + (2m2 b + (2n2 + 1) d) ω2 0 ,
x0 = (ad − bc) ω1 0 ,
y 0 = ((2k2 + 1) a + (2l2 + 1) c) ω1 0 + ((2k2 + 1) b + (2l2 + 1) d) ω2 0 .
We managed to eliminate ω2 0 in the expressions for x and x0 . However, we should ask

whether there are modular transformations for the specific selections of b and d made
above. In other words, are there integer solutions for a and c to the equation
ad − bc = 1 (A.10)
for the specific selections of b and d made above? This equation is a linear Diophantine
equation of the form ax + by = c and it is known that such equations have integer
solutions, as long as c is a multiple of the greatest common divisor of a and b. In both
cases b and d are relatively prime, and, thus, their greatest common divisor is equal
to one. Therefore, in both cases, equation (A.10) does have solutions. Furthermore,
The parity of the selected b and d implies that in the case of three real roots, a will
be odd, while in the case of one real root a and c have to be of opposite parity. The
60
above imply that
x = ω1 0 , (A.11)
y = m2 0 ω1 0 + 2n2 0 + 1 ω2 0 ,

(A.12)
0 0
x = ω1 , (A.13)
y 0 = 2k2 0 + 1 ω1 0 + 2l2 0 ω2 0 ,

(A.14)
where m2 , n2 , k2 , l2 ∈ Z. In both cases, the new lattice has been formed so that the
basic period parallelogram has one side parallel to the real axis.
Let us now focus in the case of three real roots. As we commented above, the
Weierstrass function ranges between +∞ and e1 between the origin and the real half-
period x. The fact that ℘ is even means that it actually ranges between +∞ and
e1 in the whole period between −x and x and consequently in the whole real axis.
Periodicity implies that this holds for any shifted axis by any multiple of 2ω2 . If
|2n2 0 + 1| > 2, then the segment of the imaginary axis from 0 to y crosses such a line,
thus there is point on this segment where ℘ takes a value larger or at most equal to
e1 (see figure 12). But we have already stated that Weierstrass function is real on the
imaginary axis and changes monotonically from −∞ at z = 0 to e3 at z = y. Since
e3 < e1 , This is not possible. Therefore the only consistent possibility is n2 0 = 0 (or
n2 0 = −1) and
x = ω1 0 ,
y = m2 0 ω1 0 + ω2 0 .
A further modular transformation of the form
ω1 0 = ω1 00 ,
ω2 0 = m2 0 ω1 00 − ω2 00 ,
results in
x = ω1 00 , (A.15)
00
y = ω2 , (A.16)
meaning that when there are three real roots, the two fundamental half-periods can
be selected so that one of them is real and the other purely imaginary. The selection
n2 0 = −1 simply results in y = −ω2 00 . If such a selection is performed, then their
values are equal to x and y, which are given by the integral formulas (A.1) and (A.2).
In a similar manner, in the case of one real root, monotonicity of the Weierstrass
elliptic function on the imaginary axis between 0 and y 0 implies that l2 0 = 1 (or
61
Imz
z0
−2ω2 0
ω2 0
Rez
ω1 0 x
Figure 12 – An example of the inconsistency that appears when n2 0 6= 0. In

this example it is assumed that m2 0 = n2 0 = 1. It is evident that z0 is congruent
to a point on the real axis, thus, ℘ (z0 ) ∈ [e1 , +∞). At the same time z0 lies in
the imaginary axis between 0 and y, therefore ℘ (z0 ) ∈ (−∞, e3 ]. Obviously, since
e1 > e3 , these statements cannot be both true.
l2 0 = −1) and
x = ω1 0 ,
y = 2k2 0 + 1 ω1 0 + 2ω2 0 .

In this case, it is easy to show that there is no modular transformation that would
preserve the reality of ω1 simultaneously setting ω2 to an imaginary value. Such a
transformation would necessarily be of the form
ω1 0 = ω1 00 ,
ω2 0 = cω1 00 + ω2 00
and it would transform the half periods to the form,
x = ω1 00 ,
y = 2k2 0 + 2c + 1 ω1 00 + 2ω2 00 .

62
Clearly, the coefficient of ω1 00 cannot be set to zero by such a transformation. On the
contrary, in this case, one may perform the transformation
ω1 0 = ω1 00 + ω2 00 ,
ω2 0 = − k2 0 + 1 ω1 00 − k2 0 ω2 00 ,

to find
x = ω1 00 + ω2 00 , (A.17)
00 00
y = −ω1 + ω2 , (A.18)
meaning that ω1 and ω2 can be defined so that they are complex conjugate to each
other. If such a selection is performed, then the fundamental half-periods are given
by (x0 ± y 0 ) /2, where x0 and y 0 are given by the integral formulas (A.3) and (A.4)
respectively.
63
Problem Solutions
Problems on Weierstrass Elliptic Function
Problem 1.1 Solution
We may write the integral formula (1.33) in terms of the roots of the cubic poly-
nomial as,
Z ∞ Z ∞
1 1
±z ∼ p dt = p dt. (B.1)
℘(z) 4t3 − g2 t − g3 ℘(z) 4 (t − e1 ) (t − e2 ) (t − e3 )
It is obvious that the integrated function has branch cuts on the complex plane with
endpoints the three roots. We may select the branch cuts as the red lines in figure 13.
Imz Imz
e1
Rez Rez
e3 e2 e1 e2
e3

Figure 13 – The branch cuts and the integration paths used
Assume there are three real roots. We may now apply the integral formula (B.1)
using the right blue path of figure 13 as the integration path. Let ω1 be the half-period
corresponding to the largest root. Then, we get
Z +∞
1
ω1 ∼ p dt. (B.2)
e1 4 (t − e1 ) (t − e2 ) (t − e3 )
Clearly, the integrated function is everywhere real (and also positive). It vanishes at
infinity as t−3/2 and it diverges at the left boundary of the integration as (t − e1 )−1/2 ,
therefore the integral converges. Thus, we showed that the half-period associated with
the largest root is congruent to a real number.
Similarly in the case of the three real roots, we may apply the integral formula
(B.1) using the left blue path of figure 13 as the integration path. Let ω2 be the
64
half-period corresponding to the largest root. Then, we get
Z −∞
1
ω2 ∼ p dt
e3 4 (t − e1 ) (t − e2 ) (t − e3 )
Z e3 (B.3)
1
=i p dt.
−∞ 4 (e1 − t) (e2 − t) (e3 − t)
The integrated function is everywhere real and the integral converges for the same
reasons as in previous case. Thus, the half-period associated with the smallest root is
congruent to a purely imaginary number.
The situation is similar in the case of one real root. The only difference is that
we may use the two different integration paths for the same root. Let ω3 be the
half-period corresponding to the largest root. For the right blue path of figure 13, we
get
Z +∞
1
ω3 ∼ p dt
e2 4 (t − e1 ) (t − e2 ) (t − e3 )
Z +∞
1 (B.4)
= r h i dt,
e2
4 (t − Ree1 )2 + (Ime1 )2 (t − e2 )
whereas for the left blue path of figure 13 we get

Z −∞
1
ω3 ∼ r h i dt
e2
4 (t − Ree1 )2 + (Ime1 )2 (t − e2 )
Z e2 (B.5)
1
=i r h i dt.
−∞ 2 2
4 (t − Ree1 ) + (Ime1 ) (e2 − t)
Both integrated functions are real and both integrals converge, implying that the half-
period associated with the real root is congruent to both a real and a purely imaginary
number.
Suppose that the two larger roots coincide e1 = e2 ≡ e0 . Then, they are necessarily
positive, as the three roots have to sum up to zero, implying that e3 = −2e0 . In this
case the Weierstrass differential equation reads,
2
dy
= 4(y − e0 )2 (y + 2e0 ) .
dz
We perform the change of variable y = e0 + 3e0 /f 2 and then the Weierstrass equation
assumes the form,
f 02 = 3e0 1 + f 2 .

65
This one can be easily integrated to yield
√
f = ± sinh 3e0 z + c .
Returning to the initial variable y, we find,

3e0
y = e0 + 2 √ .
sinh 3e0 z + c
We know that by definition, the Weierstrass elliptic function should have a second
order pole at z = 0. This clearly sets the value of c to inπ, n ∈ Z and finally we have
3e0
℘ z; 12e20 , −8e30 = e0 +

√ . (B.6)
sinh2 3e0 z
In this case clearly the Weierstrass elliptic function has degenerated to a simply
periodic function. The only period is purely imaginary and it assumes the value
√
2ω2 = iπ/ 3e0 .
Similarly, if the two smaller roots coincide, we have e2 = e3 = −e0 and e1 = 2e0 ,
where e0 > 0. The Weierstrass differential equation reads,
2
dy
= 4(y + e0 )2 (y − 2e0 ) .
dz
We perform the change of variable, y = −e0 +3e0 /f 2 and then the Weierstrass equation
takes the form,
f 02 = 3e0 1 − f 2 .

This can be easily integrated to yield

√
f = ± sin 3e0 z + c .
Returning to the initial variable y, we find,

3e0
y = −e0 + 2
√ .
sin 3e0 z + c
Finally, requesting that the Weierstrass elliptic function has a second order pole at
z = 0 sets the constant c to nπ, n ∈ Z and we get,
3e0
℘ z; 12e20 , 8e30 = −e0 +

√
2
. (B.7)
sin 3e0 z
Once again, in this double root limit, the Weierstrass elliptic function has degenerated
to a simply periodic function. Its only period is real and assumes the value 2ω1 =
√
π/ 3e0 .
66
Finally, if all three roots coincide, they must be equal to zero, since their sum
should vanish. In this case the Weierstrass differential equation reads,
2
dy
= 4y 3 .
dz
This can trivially be integrated to acquire

1
y= .
(z + c)2
Once again, we demand that the Weierstrass ellitpic function has a double pole at
z = 0. Therefore c = 0 and
1
℘ (z; 0, 0) = 2 . (B.8)
z
To sum up, we found that whenever two roots coincide, the Weierstrass elliptic
function degenerates to a function that is not doubly but simply periodic. If all three
roots coincide, the Weierstrass elliptic function degenerates to a function that is not
periodic at all.

By definition (1.18) of the Weierstrass elliptic function, we have
1
℘ (z; ω1 , ω2 )
µ2

1 X 1 1
= + − ,
(µz)2 {m,n}6={0,0} (µz + 2mµω1 + 2nµω2 )2 (2mµω1 + 2nµω2 )2
or
1
℘ (z; ω1 , ω2 ) = ℘ (µz; µω1 , µω2 ) . (B.9)
µ2
Then, using the definition of the moduli g2 and g3 (1.24) and (1.25), we get
X 1 1
g2 (µω1 , µω2 ) = 60 4 = µ4 g2 (ω1 , ω2 ) ,
{m,n}6={0,0}
(2mµω1 + 2nµω2 )
X 1 1
g3 (µω1 , µω2 ) = 140 6 = µ6 g3 (ω1 , ω2 ) .
{m,n}6={0,0}
(2mµω1 + 2nµω2 )
The latter imply

2 g2 g3
℘ (z; g2 , g3 ) = µ ℘ µz; 4 , 6 , (B.10)
µ µ
which is the desired homogeneity property of the Weierstrass elliptic function.
67
Problems on Weierstrass Quasi-periodic Functions
Integrating the homogeneity property of Weierstrass elliptic function (1.41) and
using the definition (2.1) of the function ζ, we get

g2 g3
ζ (z; g2 , g3 ) = µζ µz; 4 , 6 + c.
µ µ
The defining condition (2.2), or in other words the fact that ζ is an odd function,
implies that c = 0, and, thus,

g2 g3
ζ (z; g2 , g3 ) = µζ µz; 4 , 6 . (B.11)
µ µ
Integrating once more and using the definition (2.8) we get

g2 g3
σ (z; g2 , g3 ) = cσ µz; 4 , 6 .
µ µ
The defining condition (2.9) implies that c = 1/µ. Therefore,

1 g2 g3
σ (z; g2 , g3 ) = σ µz; 4 , 6 . (B.12)
µ µ µ

The defining equations (2.1) and (2.2) imply that
Z z
1 1
ζ (z) = − ℘ (w) − 2 dw. (B.13)
z 0 w
Therefore,
Z −z
1 1
ζ (−z) = − − ℘ (w) − 2 dw
z 0 w
Z z Z z
1 0 1 0 1 1
℘ w − 0 2 dw0 ,
0

=− + ℘ −w − dw = − +
z (−w 0 )2 z w
0 0
where we made the change of the integration variable w = −w0 and we used the fact
that ℘ is an even function. The latter implies that
ζ (−z) = −ζ (z) , (B.14)
meaning that ζ is an odd function.
68
Similarly, equations (2.8) and (2.9) imply
Rz
σ (z) = ze 0 (ζ(w)− w1 )dw . (B.15)
Therefore,
R −z Rz 0 1

0 Rz
σ (−z) = −ze 0 (ζ(w)− w1 )dw = −ze− 0 ζ(−w )− −w0 dw
= −ze 0 (ζ(w0 )− w10 )dw0
or else
σ (−z) = −σ (z) , (B.16)
meaning that σ is an odd function.

First, we will prove the quasi-periodicity property
ζ (z + 2mω1 + 2nω2 ) = ζ (z) + 2mζ (ω1 ) + 2nζ (ω2 ) (B.17)
for n = 0, namely,
ζ (z + 2mω1 ) = ζ (z) + 2mζ (ω1 ) . (B.18)
This obviously holds for m = 0. Assuming that it holds for a given value m, we will
show that it does hold for m + 1,
ζ (z + 2 (m + 1) ω1 ) = ζ (z + 2mω1 ) + 2ζ (ω1 )
= ζ (z) + 2mζ (ω1 ) + 2ζ (ω1 ) = ζ (z) + 2 (m + 1) ζ (ω1 ) .
Similarly, one can show that (B.18) holds for m−1. Thus, inductively, equation (B.18)
holds for any m. Therefore, the quasi-periodicity property (B.17) holds for n = 0.
Assuming that it holds for a given value n, we will show that it holds for n + 1,
ζ (z + 2mω1 + 2 (n + 1) ω2 ) = ζ (z + 2mω1 + 2nω2 ) + 2ζ (ω2 )

= ζ (z) + 2mζ (ω1 ) + 2nζ (ω2 ) + 2ζ (ω2 ) = ζ (z) + 2mζ (ω1 ) + 2 (n + 1) ζ (ω2 ) .
Similarly, one can show that (B.17) holds for n − 1. Therefore, inductively, we have
shown that (B.17) holds for any m and n.
In a similar manner, we proceed to prove the quasi-periodicity property
σ (z + 2mω1 + 2nω2 ) (−1)m+n+mn e(2mζ(ω1 )+2nζ(ω2 ))(z+mω1 +nω2 ) σ (z) (B.19)
for n = 0, namely,
σ (z + 2mω1 ) = (−1)m e2mζ(ω1 )(z+mω1 ) σ (z) . (B.20)
69
This property obviously holds for m = 0. Assuming that it holds for a given value of
m, we will show that it does hold for m + 1,
σ (z + 2 (m + 1) ω1 ) = −e2ζ(ω1 )(z+2mω1 +mω1 ) σ (z + 2mω1 )

= −e2ζ(ω1 )(z+(2m+1)ω1 ) (−1)m e2mζ(ω1 )(z+mω1 ) σ (z)
2 +2m+1
= (−1)m+1 e2ζ(ω1 )((m+1)z+(m )ω1 ) σ (z)
= (−1)(m+1) e2(m+1)ζ(ω1 )(z+(m+1)ω1 ) σ (z) .
Similarly one can show that (B.20) holds for m − 1. Therefore, inductively, property
(B.20) holds for any m. This implies that the quasi periodicity property (B.19) holds
for n = 0. Assuming that it does hold for a given value of n, we will show that it does
hold for n + 1,
σ (z + 2mω1 + 2 (n + 1) ω2 ) = −e2ζ(ω2 )(z+2mω1 +2nω2 +ω2 ) σ (z + 2mω1 + 2nω2 )

= −e2ζ(ω2 )(z+2mω1 +2nω2 +ω2 ) (−1)m+n+mn e(2mζ(ω1 )+2nζ(ω2 ))(z+mω1 +nω2 ) σ (z)
2 +2n+1
= (−1)m+n+1+mn e2mζ(ω1 )(z+mω1 +nω2 ) e2ζ(ω2 )((n+1)z+(mn+2m)ω1 +(n )ω2 ) σ (z)
= (−1)m+n+1+mn e2mζ(ω1 )(z+mω1 +(n+1)ω2 ) e2(n+1)ζ(ω2 )(z+mω1 +(n+1)ω2 )

× em(2ζ(ω2 )ω1 −2ζ(ω1 )ω2 ) σ (z)
= (−1)m+n+1+mn e(2mζ(ω1 )+2(n+1)ζ(ω2 ))(z+mω1 +(n+1)ω2 ) e−iπm σ (z)
= (−1)m+(n+1)+m(n+1) e(2mζ(ω1 )+2(n+1)ζ(ω2 ))(z+mω1 +(n+1)ω2 ) σ (z) ,
where in the last step we used property (2.7). Similarly, one can show that (B.19)
holds for n − 1, and, thus, for any m and n.

The addition formula (2.18) implies that
2
1 ℘0 (z) − ℘0 (ω1 )

℘ (z + ω1 ) = −℘ (z) − ℘ (ω1 ) + .
2 ℘ (z) − ℘ (ω1 )
Since ℘ (ω1 ) = e1 and ℘ is stationary at all half-periods, it follows that

2
1 ℘0 (z)

℘ (z + ω1 ) = −℘ (z) − e1 + .
2 ℘ (z) − e1
Using Weierstrass differential equation (1.29), we get
(℘ (z) − e2 ) (℘ (z) − e3 )
℘ (z + ω1 ) = −℘ (z) − e1 + .
℘ (z) − e1
70
Finally, using the fact that the three roots sum to zero we find
e1 ℘ (z) + e21 + e2 e3 2e2 + e2 e3

℘ (z + ω1 ) = = e1 + 1 . (B.21)
℘ (z) − e1 ℘ (z) − e1
Similarly, it can be shown that
2e23 + e1 e2 2e22 + e3 e1
℘ (z + ω2 ) = e3 + , ℘ (z + ω3 ) = e2 + . (B.22)
℘ (z) − e3 ℘ (z) − e2

0 (z)−℘0 (w)
We will write the elliptic function 21 ℘℘(z)−℘(w) , being considered a function of
z, in terms of the Weierstrass function ζ and its derivatives, taking advantage of
the techniques analysed in section 2.2. In order to do, so we need an irreducible
set of poles of this function and the pricipal part of its Laurent series at each pole.
We have already shown in section 1.2 that given a meromorphic function f (z), the
function f 0 /f has only first order poles at the positions f has a root or a pole with
residues equal to the multiplicity of the root and the opposite of the multiplicity of
the pole respectively. We will apply that for the function ℘ (z) − ℘ (w). The latter
has obviously only a second order pole in a position congruent to z = 0 in each cell.
The function ℘ (z) − ℘ (w) obviously vanishes at z = w and z = −w, which in general
they are not congruent. Therefore, since it is a second order elliptic function, in each
cell it has only two first order roots congruent to z = w and z = −w. Consequently,
℘0 (z)
an irreducible set of poles of the function ℘(z)−℘(w) are z = 0, z = w and z = −w and
the corresponding Laurent series read,
℘0 (z) 2
= − + O z0 ,

℘ (z) − ℘ (w) z
℘0 (z) 1
+ O z0 ,

=
℘ (z) − ℘ (w) z−w
0
℘ (z) 1
+ O z0 .

=
℘ (z) − ℘ (w) z+w
0
℘ (w)
Similarly, the function − ℘(z)−℘(w) has poles only at the locations where the function
℘ (z) − ℘ (w) has roots. We have already shown that an irreducible set of roots of the
latter is z = w and z = −w. The corresponding Laurent series read,
℘0 (w) ℘0 (w) 1
+ O z0 = − + O z0 ,

− =− 0
℘ (z) − ℘ (w) ℘ (w) (z − w) z−w
0
℘ (w) 0
℘ (w) 1
+ O z0 = + O z0 .

− =− 0
℘ (z) − ℘ (w) ℘ (−w) (z + w) z+w
71
0 0
Combining the above information, it turns out that the function 21 ℘℘(z)−℘(w)
(z)−℘ (w)
has poles
only at z = 0 and z = −w and the corresponding Laurent series read,
1 ℘0 (z) − ℘0 (w) 1
= − + O z0 ,

2 ℘ (z) − ℘ (w) z
0 0
1 ℘ (z) − ℘ (w) 1
+ O z0 .

=
2 ℘ (z) − ℘ (w) z+w
The above is the necessary information in order to write the elliptic function
1 ℘0 (z)−℘0 (w)
2 ℘(z)−℘(w) in terms of function ζ and its derivatives. Applying formula (2.15), we
get
1 ℘0 (z) − ℘0 (w)
= ζ (z + w) − ζ (z) + c (w) .
2 ℘ (z) − ℘ (w)
Interchanging z and w, we get in a similar manner
1 ℘0 (z) − ℘0 (w)
= ζ (z + w) − ζ (w) + c (z) ,
2 ℘ (z) − ℘ (w)
which implies that
1 ℘0 (z) − ℘0 (w)
= ζ (z + w) − ζ (z) − ζ (w) + c.
2 ℘ (z) − ℘ (w)
Finally, the fact that ℘0 and ζ are odd functions, whereas ℘ is an even function implies
that c = 0, which results in the desired pseudo-addition formula
1 ℘0 (z) − ℘0 (w)
= ζ (z + w) − ζ (z) − ζ (w) . (B.23)
2 ℘ (z) − ℘ (w)

We differentiate the pseudo-addition theorem for Weierstrass ζ function (2.20) with
respect to z and w. We get
1 ℘00 (z) (℘ (z) − ℘ (w)) − ℘0 (z) (℘0 (z) − ℘0 (w))
−℘ (z + w) + ℘ (z) = ,
2 (℘ (z) − ℘ (w))2
1 ℘00 (w) (℘ (z) − ℘ (w)) − ℘0 (w) (℘0 (z) − ℘0 (w))
−℘ (z + w) + ℘ (w) = − .
2 (℘ (z) − ℘ (w))2
Adding the above two equations yields
1 (℘00 (z) − ℘00 (w)) (℘ (z) − ℘ (w)) − (℘0 (z) − ℘0 (w))2

−2℘ (z + w) + ℘ (z) + ℘ (w) =
2 (℘ (z) − ℘ (w))2
1 ℘00 (z) − ℘00 (w) 1 ℘0 (z) − ℘0 (w) 2

= − .
2 ℘ (z) − ℘ (w) 2 ℘ (z) − ℘ (w)
72
The second derivative of Weierstrass elliptic function can easily be expressed in
terms of the Weierstrass elliptic function itself. Differentiating the Weierstrass differ-
ential equation, we get
g2
℘00 = 6℘2 − , (B.24)
2
which implies that
℘00 (z) − ℘00 (w) = 6 ℘2 (z) − ℘2 (w) .

Therefore, we conclude that
℘2 (z) − ℘2 (w) 1 ℘0 (z) − ℘0 (w) 2

−2℘ (z + w) + ℘ (z) + ℘ (w) = 3 −
℘ (z) − ℘ (w) 2 ℘ (z) − ℘ (w)
and finally,
1 ℘0 (z) − ℘0 (w) 2

℘ (z + w) = −℘ (z) − ℘ (w) − , (B.25)
4 ℘ (z) − ℘ (w)
which the addition theorem for the Weierstrass elliptic function (2.18).

The addition formula for Weierstrass elliptic function states that
2
1 ℘0 (z) − ℘0 (w)

℘ (z + w) = −℘ (z) − ℘ (w) + .
2 ℘ (z) − ℘ (w)
The duplication formula requires taking the limit w → z. At this limit the fraction
in the right hand side of the addition formula becomes indeterminate. It is simple
though to use Hospital’s rule to find
2
1 ℘00 (z)

℘ (2z) = −2℘ (z) + . (B.26)
2 ℘0 (z)
One can eliminate the second derivative of ℘. For this purpose we have to differ-
entiate the Weierstrass equation to yield
g2
℘00 (z) = 6℘2 (z) − . (B.27)
2
Then, the duplication formula assumes the form
2
6℘2 (z) − g22
℘ (2z) = −2℘ (z) +
4 (4℘3 (z) − g2 ℘ (z) − g3 )
or
g22
1 3g2 ℘2 (z) + 9g3 ℘ (z) + 4
℘ (2z) = ℘ (z) + . (B.28)
4 4℘02 (z)
73
Let’s now derive the duplication formula expressing ℘ (2z) in terms of ζ (z) and
its derivatives. Let 2ω1 and 2ω2 be the two fundamental periods of ℘ (z). Then the
function ℘ (2z) is an elliptic function with fundamental periods equal to ω1 and ω2 .
This obviously means that it is also an elliptic function with periods 2ω1 and 2ω2 ,
but in the parallelogram defined by the latter, there exist four fundamental period
parallelograms of ℘ (2z). Therefore, the function ℘ (2z), as an elliptic function with
periods 2ω1 and 2ω2 is an order 8 elliptic function with four double poles at positions
congruent to z = 0, z = ω1 , z = ω2 and, z = ω3 . The principal part of the Laurent
series of ℘ (2z) at each of these poles reads
1 1
2

℘ (2z) = + O (z − z 0 ) .
4 (z − z0 )2
We have the necessary data to express ℘ (2z) in terms of the function ζ (z) and its
derivatives. Formula (2.15) implies that
1 1 1 1
℘ (2z) = C − ζ 0 (z) − ζ 0 (z − ω1 ) − ζ 0 (z − ω2 ) − ζ 0 (z − ω1 − ω2 )
4 4 4 4
1 1 1 1
= C + ℘ (z) + ℘ (z − ω1 ) + ℘ (z − ω2 ) + ℘ (z − ω3 ) .
4 4 4 4
Finding the Laurent series of the above equation at the region of z = 0, we get
1 2
1 2

+ O z = C + + e1 + e2 + e3 + O z ,
4z 4 4z 4
implying that C = 0, since the three roots e1 , e2 and e3 sum to zero. Using the results
of problem 2.4, we get
2e21 + e2 e3 2e23 + e1 e2 2e22 + e3 e1

1 1 1 1
℘ (2z) = ℘ (z)+ e1 + + e3 + + e2 + .
4 4 ℘ (z) − e1 4 ℘ (z) − e3 4 ℘ (z) − e2
After some algebra and using the fact that the three roots sum to zero, we get
2 e21 + e22 + e23 + e1 e2 + e2 e3 + e3 e1 ℘2 (z)

1
℘ (2z) = ℘ (z) +
4 ℘02 (z)
2 e31 + e32 + e33 + 3e1 e2 e3 ℘ (z) e21 e22 + e22 e23 + e23 e21

+ + .
℘02 (z) ℘02 (z)
The fact that the three roots sum to zero can be used to calculate the coefficients
of the above expression. We find,
(e1 + e2 + e3 )2 = e21 + e22 + e23 + 2 (e1 e2 + e2 e3 + e3 e1 ) = 0,

which with the help of equation (1.39) implies

g2
e21 + e22 + e23 = . (B.29)
2
74
Furthermore,
e21 + e22 + e23 (e1 + e2 + e3 )

= e31 + e32 + e33 + e21 e2 + e22 e3 + e23 e1 + e1 e22 + e2 e23 + e3 e21 = 0,

(e1 + e2 + e3 )3
= e31 + e32 + e33 + 3 e21 e2 + e22 e3 + e23 e1 + e1 e22 + e2 e23 + e3 e21 + 6e1 e2 e3 = 0.

Taking into account equation (1.40), the last two equations imply that
3g3
e31 + e32 + e33 = 3e1 e2 e3 = . (B.30)
4
Finally,
(e1 e2 + e2 e3 + e3 e1 )2 = e21 e22 + e22 e23 + e23 e21 + 2e1 e2 e3 (e1 + e2 + e3 ) ,
implying that
g22
e21 e22 + e22 e23 + e23 e21 = (e1 e2 + e2 e3 + e3 e1 )2 = . (B.31)
16
Putting everything together, we get
g22
1 3g2 ℘2 (z) + 9g3 ℘ (z) + 4
℘ (2z) = ℘ (z) + , (B.32)
4 4℘02 (z)
which is the desired duplication formula.
75
Problems on Classical Mechanics Applications
When the cubic polynomial has a double root e0 , the third root necessarily equals
−2e0 , as the three roots sum to zero. Therefore, the cubic polynomial in such a case
equals,
Q (x) = 4(x − e0 )2 (x + 2e0 ) = 4x3 − 12e20 x + 8e30 . (B.33)
Thus, a cubic polynomial with a double root obeys
g2 = 12e20 , g3 = −8e30 , (B.34)
or
g2 > 0, g23 − 27g32 = 0. (B.35)
When g3 > 0, the double root is negative, and, thus, the two smaller roots coincide,
whereas when g3 < 0, the double root is positive meaning that the two larger roots
coincide.
In the case of the motion of a particle in a cubic potential, the solution is given by
x = ℘ (t − t0 ; −F0 , −E) , (B.36)

x = ℘ (t − t0 + ω2 ; −F0 , −E) , (B.37)
where the second solution exists only when there are three real roots. The moduli
take the values g2 = −F0 and g3 = −E and therefore a double root exists when
3
F0 2
F0 < 0, E=± − . (B.38)
3
These values correspond to the case the potential that possesses a local minimum and
the energies equal the limiting values for the existence of a bounded motion.
When E = +(−F0 /3)3/2 , the double root equals e0 = (−F0 /12)1/2 > 0. Therefore,
the real period diverges, while the imaginary one assumes the value
1
3F0 − 4

2ω2 = iπ − . (B.39)
4
Formula (1.43) suggests that in this case,
3 ! 1 1
− 3F4 0 2

F0 2 F0 2
℘ z; −F0 , − − = − + , (B.40)
3 12 2
1
3F0 4
sinh − 4 z
76
implying that the unbounded and bounded motions degenerate to
1 1
− 3F4 0 2

F0 2
x= − + , (B.41)
12 2 3F0 4
1
sinh − 4 (t − t0 )
1 1
− 3F4 0 2

F0 2
x= − − , (B.42)
12 2
1
3F0 4 π
cosh − 4 (t − t0 ) + i 2
respectively. These motions describe a particle having exactly the energy of the local
maximum, in the first case coming from the right and in the second coming from the
left. The particle arrives at the position of the unstable equilibrium in infinite time.
When E = −(−F0 /3)3/2 the double root equals e0 = −(−F0 /12)1/2 > 0. The
imaginary period diverges, while the real one assumes the value
1
3F0 − 4

2ω1 = π − . (B.43)
4
Formula (1.44) suggests that in this case,
3 ! 1 1
− 3F4 0 2

F0 2 F0 2
℘ z; −F0 , + − =− − + , (B.44)
3 12 1
sin2 − 3F4 0 4 z
implying that the unbounded and bounded motions degenerate to
1 1
− 3F4 0 2

F0 2
x= − + , (B.45)
12 2
1
3F0 4
sin − 4 (t − t0 )
1 1 1
− 3F4 0 2

F0 2 F0 2
x= − + lim = − , (B.46)
12 x→∞ 2
1
3F0 4 12
sin − 4 (t − t0 ) + ix
respectively. The form of the limit of the bounded motion is also physically expected.
The bounded motion ranges between e3 and e2 . Since these two roots coincide, the
bounded motion necessarily degenerates to a constant. The unbounded motion de-
scribes a point particle having exactly the energy of the local minimum, in the first
case coming from the right and getting reflected by the potential barrier, while the
bounded motion describes a point particle resting at the local minimum equilibrium
position.
77
The period of motion in this case assumes a finite value. The Taylor series of the
potential at the region of the local minimum is
3 1 !2 1 !3
F0 2 p F0 2 F0 2
V (x) = − − + 2 −3F0 x + − −4 x+ − . (B.47)
3 12 12
Since the mass of the point particle has been assumed to equal to 2, the period of the
small oscillations at the region of the local minimum is equal to
2π
Tsmall = 1 = 2ω1 , (B.48)
(−12F0 ) 4
as expected.

The cubic polynomial associated with the problem of a point particle with energy
E is
Q (x) = 4x3 + F0 x + E.
Let e1 , e2 and e3 be its three roots, appropriately ordered, as described in section 3.1.
They obviously obey
Q (ei ) = 4e3i + F0 ei + E = 0.
The cubic polynomial associated with the problem of a point particle with energy
−E is
R (x) = 4x3 + F0 x − E.
This polynomial obeys
R (−ei ) = −4e3i − F0 ei − E = −Q (ei ) = 0. (B.49)
In other words, the roots of the problem with inverted energy are just the opposite
of the roots of the initial problem. It is obvious that the appropriate ordering of the
roots of the inverted energy problem is
e1 (−E) = −e3 (E) , e2 (−E) = −e2 (E) , e3 (−E) = −e1 (E) . (B.50)
In the case the three roots are real, the above is obvious, since e1 > e2 > e3 ⇒ −e1 <
−e2 < −e3 . In the case there is only one real root, this is necessarily −e2 , since e2 is
real, while −e1 has a negative imaginary part, since e1 has positive imaginary part.
In the case of three real roots, we use formula (3.1) to find that
Z +∞
dt
2ω1 (−E) = p
−e3 (t + e1 ) (t + e2 ) (t + e3 )
dt0
Z e3
(B.51)
= p
0 0
(e1 − t ) (e2 − t ) (e3 − t )0
−∞
= −2iω2 (E) = |2ω2 (E)| ,
78
where we defined t0 = −t. Similarly, in the case of one real root, we use formula (3.3)
to find
Z +∞
dt
2ω1 (−E) + 2ω2 (−E) = p
−e2 (t + e1 ) (t + e2 ) (t + e3 )
dt0
Z e2
(B.52)
= p
0 0 0
(t − e2 ) (e2 − t ) (t − e3 )
−∞
= −2i (2ω1 (E) − 2ω2 (E)) = |2ω1 (E) − 2ω2 (E)| .
This completes the proof that the absolute value of the imaginary period corresponds
in all cases to the “time of flight” or period of oscillations for a point particle with
opposite energy.

The conservation of energy in the cubic potential problem reads
ẋ2 − 4x3 − F0 x = E
or in terms of the roots
ẋ2 = 4 (x − e1 ) (x − e2 ) (x − e3 ) . (B.53)
(e3 −e1 )(e3 −e2 )
We perform the change of variable x = e3 + y−e3 . It is a matter of simple
algebra to find that
(e3 − e1 ) (e3 − e2 )
ẋ = − ẏ,
(y − e3 )2
(e3 − e1 ) (y − e2 ) (e3 − e2 ) (y − e1 ) (e3 − e1 ) (e3 − e2 )
x − e1 = , x − e2 = , x − e3 = .
y − e3 y − e3 y − e3
Thus, in terms of the new variable y, the conservation of energy is written as
(e3 − e1 )2 (e3 − e2 )2 2 (e3 − e1 ) (y − e2 ) (e3 − e2 ) (y − e1 ) (e3 − e1 ) (e3 − e2 )

4 ẏ = 4
(y − e3 ) y − e3 y − e3 y − e3
or
ẏ 2 = 4 (y − e1 ) (y − e2 ) (y − e3 ) . (B.54)
Therefore, the conservation of energy is invariant under this transformation. We
could say that the problem has a Z2 symmetry, since performing the change of variable
twice leads to the initial variable,
(e3 − e1 ) (e3 − e2 ) (e3 − e1 ) (e3 − e2 )

x → e3 + → e3 + = x. (B.55)
x − e3 e3 + (e3 −ex−e
1 )(e3 −e2 )
3
− e3
79
It is trivial to show that
lim x = e3 , lim x = −∞, lim x = +∞,

y→−∞ y→e− y→e+
3 3
lim x = e1 , lim x = e2 , lim x = e3 .

y→e2 y→e1 y→+∞
As a consequence, the following intervals of x are mapped to intervals of y as
(−∞, e3 ] → (−∞, e3 ] ,
[e3 , e2 ] → [e1 , +∞) ,
[e2 , e1 ] → [e2 , e1 ] ,
[e1 , +∞) → [e3 , e2 ] .
That means that the bounded motion in one problem is mapped to the unbounded
motion of the other. Since the problems are identical, the two solutions, oscillatory
and scattering, are also identical,
xoscillatory (t) = yoscillatory (t) = f (t) , (B.56)

xscattering (t) = yscattering (t) = g (t) , (B.57)
which implies
(e3 − e1 ) (e3 − e2 ) (e3 − e1 ) (e3 − e2 )
f (t) = e3 + , g (t) = e3 + . (B.58)
g (t) − e3 f (t) − e3
Without knowing the exact form of the oscillatory and scattering motions, we know
that they are connected in a specific way. Let T be the period of the oscillatory
motion. The above connection implies that
f (0) = e3 , g (0) → +∞,

T T
f = e2 , g = e1 ,
2 2
f (T ) = e3 , g (T ) → +∞.
Thus, it turns out that the period of the oscillatory motion and the “time of flight”
of the scattering motion are equal.

It is evident in figure 8 that there are two energies where a double root appears,
namely, E = ±ω 2 .
For E = ω 2 , the two larger roots coincide,
ω2
e1 = e2 = e0 = .
3
80
Following the outcomes of problem 1.2, the real period of the Weierstrass elliptic
function diverges, whereas the imaginary one assumes the value
π
2ω2 = i
ω
and the Weierstrass elliptic function is expressed in terms of hyperbolic functions as
4ω 4 8ω 6 ω2 ω2

2℘ z; ,− = + .
3 27 3 sinh2 ωz
The pendulum solution is expressed in terms of the bounded real solution in the real
domain of Weierstrass equation, namely the Weierstrass elliptic function on the real
axis shifted by ω2 ,
4ω 4 8ω 6 ω2 ω2 ω2 ω2

2℘ t + ω2 ; ,− = + = − .
sinh2 ωt + i π2 cosh2 ωt

3 27 3 3
Finally, equation (3.29) implies that the degenerate solution is written as

2
cos θ = −1 + . (B.59)
cosh2 ωt
For E = −ω 2 , the two smaller roots coincide.
ω2
e2 = e3 = −e0 = − .
3
Therefore, the imaginary period diverges, while the real one assumes the value
π π
2ω1 = √ = .
3e0 ω
As in problem 3.1, the bounded real solution in the real domain of Weierstrass equation
degenerates to a constant being equal to the double root,
4ω 4 8ω 6 ω2

2℘ t + ω2 ; , =− . (B.60)
3 27 3
Finally, equation (3.29) implies

cos θ = 1. (B.61)
This solution describes the pendulum lying at the stable equilibrium position. Consid-
ering this solution as the limit of the oscillatory pendulum motion when the amplitude
of the oscillation goes to zero, we conclude that the limit of the period of the oscillatory
motion at zero amplitude is
2π
Toscillating = 4ω1 = , (B.62)
ω
81
as expected.

In the case E > ω 2 , it holds that e1 = x1 , e2 = x2 and e3 = x3 . We start using
the addition formula for Weierstrass elliptic function (2.6)

2 E
x (−t) = ln 2 2℘ (ω1 /2 − t) −
ω 3
" !#
1 ℘ (ω1 ) − ℘0 (−ω1 /2 − t) 2
0
2
= ln −2℘ (ω1 ) − 2℘ (−ω1 /2 − t) + 2 − 2e1
ω2 2 ℘ (ω1 ) − ℘ (−ω1 /2 − t)
" 2 !#
℘0 (ω1 /2 + t)

2 1
= ln −4e1 − 2℘ (ω1 /2 + t) + .
ω2 2 e1 − ℘ (ω1 /2 + t)
Then, we use the Weierstrass differential equation (1.29) to substitute the derivative
of Weierstrass function

2 (℘ (ω1 /2 + t) − e2 ) (℘ (ω1 /2 + t) − e3 )
x (−t) = ln −4e1 − 2℘ (ω1 /2 + t) + 2
ω2 ℘ (ω1 /2 + t) − e1
(−e1 − e2 − e3 ) ℘ (ω1 /2 + t) + 2e21 + e2 e3

2
= ln 2 2
ω ℘ (ω1 /2 + t) − e1
2e21 + e2 e3

2
= ln 2 2 .
ω ℘ (ω1 /2 + t) − e1
Using the specific form of the roots, we find
2e21 + e2 e3 = 2x21 + x2 x3
2
E E 1p 2 E 1p 2
=2 + − + E − ω4 − − E − ω4
6 12 4 12 4
E2 E2 E 2 − ω4 ω4
= + − = ,
18 144 16 16
which implies that finally, we get
ω2

2 E
x (−t) = ln = − ln 2 2℘ (ω1 /2 + t) − = −x (t) .
4 (℘ (ω1 /2 + t) − e1 ) ω 3
(B.63)
2
The case E < ω is identical with the permutation e1 ↔ e2 , ω1 ↔ ω3 .
The essential difference between the transmitting solutions and the reflecting solu-
tions is that in the former case the time of flight equals the real half-period, while in
the latter equals the whole real period. Therefore, in the case of reflecting solutions,
82
the situation is much simpler, since,

2 E 2 E
x (−t) = ln 2 2℘ (ω1 − t) − = ln 2 2℘ (−ω1 + t) −
ω 3 ω 3

2 E
= ln 2 2℘ (ω1 + t) − = x (t) , (B.64)
ω 3
as expected.

The extrema of motion can be read from table 6. They are ln 4(e1ω−e
2
2)
and ln 4(e1ω−e
2
3)
.
It is a matter of simple algebra to show that
(e1 − e3 ) (e1 − e2 ) = e21 − (e2 + e3 ) e1 + e2 e3 = 2e21 + e2 e3 . (B.65)
In problem 3.5 we found that 2e21 + e2 e3 = ω16 . Therefore, ln 4(e1ω−e = − ln 4(e1ω−e

4 2) 3)
2 2
and consequently the extrema of motion are opposite.

The periodicity property x (T − t) = −x (t) can be proved easily repeating all the
steps of problem 3.5 with the substitution t → −t.
The double root limit, as one can easily see in figure 10 corresponds to E = ω 2 .
In this limit the two smaller roots coincide to the value e2 = e3 = −ω 2 /12. The
oscillatory motion is described by the bounded solution. As shown in problem 3.1, in
this limit, the bounded solution degenerates to a constant equal to the value of the
double root. Consequently,
2
ω2

2 ω
x (t) = ln − 2 2 − = ln 1 = 0, (B.66)
ω 12 3
which is indeed the equilibrium position. The real period degenerates to the value
π 2π
T = 2ω1 = √ = , (B.67)
−3e2 ω
as expected.
83
Problems on Quantum Mechanics Applications
The special solution is
q
ỹ (x; ω1,2,3 ) = ℘ (x) − e1,3,2 (ζ (x + ω1,2,3 ) + e1,3,2 x)
(B.68)
= y± (x; ω1,2,3 ) (ζ (x + ω1,2,3 ) + e1,3,2 x) .
The first and second derivatives of y± (x; ω1,2,3 ) are given by equations (4.7) and (4.8).
These equations for the solution modulus being equal to any of the half-periods read
dy± (x; ω1,2,3 ) 1 ℘0 (x)

= y± (x; ω1,2,3 ) , (B.69)
dx 2 ℘ (x) − e1,3,2
d2 y± (x; ω1,2,3 )
= (2℘ (x) + e1,3,2 ) y± (x; ω1,2,3 ) . (B.70)
dx2
Therefore, we may express the first and second derivatives of ỹ (x; ω1,2,3 ) as,
dỹ (x; ω1,2,3 ) dy± (x; ω1,2,3 )

= (ζ (x + ω1,2,3 ) + e1,3,2 x)
dx dx
+ y± (x; ω1,2,3 ) (−℘ (x + ω1,2,3 ) + e1,3,2 )
and
d2 ỹ (x; ω1,2,3 ) d2 y± (x; ω1,2,3 )
= (ζ (x + ω1,2,3 ) + e1,3,2 x)
dx2 dx2
dy± (x; ω1,2,3 )
+2 (−℘ (x + ω1,2,3 ) + e1,3,2 )
dx
− y± (x; ω1,2,3 ) ℘0 (x + ω1,2,3 )

(B.71)
= (2℘ (x) + e1,3,2 ) (ζ (x + ω1,2,3 ) + e1,3,2 x) y± (x; ω1,2,3 )
℘0 (x)
− (℘ (x + ω1,2,3 ) − e1,3,2 ) y± (x; ω1,2,3 )
℘ (x) − e1,3,2
− ℘0 (x + ω1,2,3 ) y± (x; ω1,2,3 ) .
The Weierstrass elliptic function, with an argument shifted by a half-period can

be calculated with the use of the addition theorem. This has been done in problem
2.4,
(e1,3,2 − e3,2,1 ) (e1,3,2 − e2,1,3 )
℘ (x + ω1,2,3 ) = e1,3,2 + . (B.72)
℘ (x) − e1,3,2
It is a direct consequence that
(e1,3,2 − e3,2,1 ) (e1,3,2 − e2,1,3 )

℘0 (x + ω1,2,3 ) = − 2 ℘0 (x) . (B.73)
(℘ (x) − e1,3,2 )
84
Combining the above two relations, we find
℘0 (x)
(℘ (x + ω1,2,3 ) − e1,3,2 ) + ℘0 (x + ω1,2,3 )
℘ (x) − e1,3,2
℘0 (x) (e1,3,2 − e3,2,1 ) (e1,3,2 − e2,1,3 )
=
℘ (x) − e1,3,2 ℘ (x) − e1,3,2
(e1,3,2 − e3,2,1 ) (e1,3,2 − e2,1,3 ) 0
− ℘ (x) = 0
(℘ (x) − e1,3,2 )2
The latter implies that equation (B.71) is written as
d2 ỹ (x; ω1,2,3 )
= (2℘ (x) + e1,3,2 ) (ζ (x + ω1,2,3 ) + e1,3,2 x) y± (x; ω1,2,3 )
dx2 (B.74)
= (2℘ (x) + e1,3,2 ) ỹ (x; ω1,2,3 ) ,
meaning that indeed the functions ỹ (x; ω1,2,3 ) are eigenfunctions of the n = 1 Lamé
problem with eigenvalues
λ = −e1,3,2 . (B.75)

We follow the derivation for the unbounded potential. In the following, b is con-
sidered real. For a in the segment [0, ω1 ], a = b. Then,
σ (x − ω2 ± b) σ (−ω2 ) −ζ(±b)x
ψ± (x; b) = e
σ (x − ω2 ) σ (−ω2 ± b)
σ (x + ω2 ± b) e−2ζ(ω2 )(x±b) σ (ω2 )
= e−ζ(±b)x
σ (x + ω2 ) e−2ζ(ω2 )x σ (+ω2 ± b) e−2ζ(ω2 )(±b)
= ψ± (x; b) .
For a in the segment [0, ω2 ], a = ib,
σ (x − ω2 ∓ ib) σ (−ω2 ) −ζ(∓ib)x
ψ± (x; ib) = e
σ (x − ω2 ) σ (−ω2 ∓ ib)
σ (x + ω2 ∓ ib) e−2ζ(ω2 )(x∓ib) σ (ω2 )
= −2ζ(ω )x −2ζ(ω )(∓ib)
e−ζ(∓ib)x
σ (x + ω2 ) e 2 σ (ω2 ∓ ib) e 2
= ψ∓ (x; ib) .
For a in the segment [ω2 , ω3 ], a = ω2 + b,
σ (x − ω2 ∓ ω2 ± b) σ (−ω2 ) −ζ(∓ω2 ±b)x
ψ± (x; ω2 + b) = e
σ (x − ω2 ) σ (−ω2 ∓ ω2 ± b)
−σ (x + ω2 ± ω2 ± b) e−(2±2)ζ(ω2 )(x±b) σ (ω2 )
= e±2ζ(ω2 )x e−ζ(±ω2 ±b)x
−σ (x + ω2 ) e−2ζ(ω2 )x σ (ω2 ± ω2 ± b) e−(2±2)ζ(ω2 )(±b)
= ψ± (x; ω2 + b) .
85
Finally, for a in the segment [ω1 , ω3 ], a = ω1 + ib and
σ (x − ω2 ± ω1 ∓ ib) σ (−ω2 ) −ζ(±ω1 ∓ib)x

ψ± (x; ω1 + ib) = e
σ (x − ω2 ) σ (−ω2 ± ω1 ∓ ib)
σ (x + ω2 ∓ ω1 ∓ ib) e2(±ζ(ω1 )−ζ(ω2 ))(x∓ib) σ (ω2 )
= −2ζ(ω )x 2(±ζ(ω )−ζ(ω ))(∓ib)
e∓2ζ(ω1 )x e−ζ(∓ω1 ∓ib)x
σ (x + ω2 ) e 2 σ (ω2 ∓ ω1 ∓ ib) e 2 2
= ψ∓ (x; ω1 + ib) ,
which concludes the derivation of the reality properties of the eigenfunctions of the
bounded n = 1 Lamé problem.

Following the derivation in the case of the unbounded potential presented in section
4.1, we have
σ (x + 2ω1 + ω2 ± a) σ (ω2 ) −e2ζ(ω1 )(x+ω1 +ω2 ±a) σ (x + ω2 ± a) σ (ω2 )

=
σ (x + 2ω1 + ω2 ) σ (ω2 ± a) −e2ζ(ω1 )(x+ω1 +ω2 ) σ (x + ω2 ) σ (ω2 ± a)
σ (x + ω2 ± a) σ (ω2 )
= e±2aζ(ω1 ) .
σ (x + ω2 ) σ (ω2 ± a)
Thus, we can write the eigenfunctions (4.40) of the bounded problem in the form of a
Bloch wave
ψ± (x; a) = u± (x; a) e±ik(a)x , (B.76)
where
σ (x + ω2 ± a) σ (ω2 ) ∓ aζ(ω 1) x aζ (ω1 ) − ω1 ζ (a)

u± (x; a) = e ω1 , ik (a) = , (B.77)
σ (x + ω2 ) σ (ω2 ± a) ω1
with u± (x + 2ω1 ; a) = u± (x; a). The function k (a) is identical to the one in the un-
bounded potential. Therefore, the band structure of the bounded potential is identical
to the band structure of the unbounded problem.

By direct computation starting from the expressions (4.6), we find
σ (x + a) σ (x − a) −ζ(a)x −ζ(−a)x σ (x + a) σ (x − a)
y+ y− = 2
e e =− .
σ (x) σ (a) σ (−a) σ 2 (x) σ 2 (a)
Using the pseudo-addition formula for the Weierstrass σ function (2.19), we get
y+ y− = ℘ (x) − ℘ (a) . (B.78)
86
Similarly, using equation (4.7),
0
1 ℘ (x) − ℘0 (a) 1 ℘0 (x) + ℘0 (a)

0 0
y+ y− − y+ y− = − y+ y−
2 ℘ (x) − ℘ (a) 2 ℘ (x) − ℘ (a)
(B.79)
℘0 (a)
=− (℘ (x) − ℘ (a)) = −℘0 (a) .
℘ (x) − ℘ (a)
We repeat for the eigenfunctions (4.40) of the bounded n = 1 Lamé potential
σ (x + ω2 + a) σ (x + ω2 − a) σ 2 (ω2 ) −ζ(a)x −ζ(−a)x
ψ+ ψ− = e e
σ 2 (x + ω2 ) σ (ω2 + a) σ (ω2 − a)
σ (x + ω2 + a) σ (x + ω2 − a) σ 2 (ω2 ) σ 2 (a)
= (B.80)
σ 2 (x + ω2 ) σ 2 (a) σ (ω2 + a) σ (ω2 − a)
℘ (x + ω2 ) − ℘ (a)
=
e3 − ℘ (a)
and
1 ℘0 (x + ω2 ) − ℘0 (a) 1 ℘0 (x + ω2 ) + ℘0 (a)

ψ+ 0 ψ− − ψ+ ψ− 0 = − ψ+ ψ−
2 ℘ (x + ω2 ) − ℘ (a) 2 ℘ (x + ω2 ) − ℘ (a)
(B.81)
℘0 (a) ℘ (x + ω2 ) − ℘ (a) ℘0 (a)
=− =− .
℘ (x + ω2 ) − ℘ (a) e3 − ℘ (a) e3 − ℘ (a)
Therefore the “normalization” of the eigenfunctions of the bounded potential dif-
fer from the “normalization” of the eigenfunctions of the unbounded potential by a
p
factor of e3 − ℘ (a). Of course this is just a constant and it could be included to
the definition of the bounded eigenfunctions. However, such an inclusion would mess
the reality properties of the eigenfunctions, as the reality of this factor depends on the
value of ℘ (a).

Using the formula (4.7) and the definition of the superpotential (4.41), we find

† d σ (x ± a) −ζ(±a)x
A y± (x; a) = − + W (x) e
dx σ (x) σ (±a)
1 ℘ (x) ∓ ℘0 (a)
0
℘0 (x)

σ (x ± a) −ζ(±a)x
=− − e
2 ℘ (x) − ℘ (a) ℘ (x) − e3 σ (x) σ (±a)
≡ f± (x, a) y± (x; a) .
The Lamé eigenfunctions themselves are not elliptic functions. However, it is trivial
that the prefactor f± (x, a) is an elliptic function both as a function of x or a.
Using the pseudo-addition formula for the Weierstrass ζ function (2.20), we find
f± (x, a) = ζ (x) + ζ (±a) − ζ (x ± a) + ζ (x + ω2 ) − ζ (x) − ζ (ω2 )

= ζ (±a) − ζ (x ± a) + ζ (x + ω2 ) − ζ (ω2 ) .
87
Remembering that f± (x, a) as a function of x is an elliptic function, we may write
it as a ratio of Weierstrass σ functions. f± (x, a) is clearly a second order elliptic
function having two first order poles at x = −ω2 and x = ∓a. It also obviously has a
zero at x = 0 since,
f± (0, a) = ζ (±a) − ζ (±a) + ζ (ω2 ) − ζ (ω2 ) = 0.
Theorem 1.5 implies that the other zero should be congruent to x = ∓a − ω2 . Indeed,
f± (−ω2 ∓ a, a) = ζ (±a) − ζ (−ω2 ) + ζ (∓a) − ζ (ω2 ) = 0.
The sum of the above poles equals the sum of the zeros, and, thus, we may write
f± (x, a) as,
σ (x) σ (x + ω2 ± a)
f± (x, a) = C .
σ (x + ω2 ) σ (x ± a)
σ(−ω2 ±a)
Requiring that f± (x, a) has residue equal to one at x = −ω2 yields C = σ(−ω2 )σ(±a) ,
and, thus,
σ (−ω2 ± a) σ (x) σ (x + ω2 ± a)
f± (x, a) = .
σ (−ω2 ) σ (±a) σ (x + ω2 ) σ (x ± a)
Therefore,
σ (−ω2 ± a) σ (x) σ (x + ω2 ± a) σ (x ± a) −ζ(±a)x

A† y± (x; a) = e
σ (−ω2 ) σ (±a) σ (x + ω2 ) σ (x ± a) σ (x) σ (±a)
σ (−ω2 ± a) σ (x + ω2 ± a) −ζ(±a)x
= e
σ (−ω2 ) σ 2 (±a) σ (x + ω2 ) (B.82)
σ (ω2 ± a) σ (−ω2 ± a) σ (x + ω2 ± a) σ (ω2 ) −ζ(±a)x
=− e
σ 2 (ω2 ) σ 2 (±a) σ (x + ω2 ) σ (ω2 ± a)
= (℘ (a) − e3 ) ψ± (x; a) .
Indeed the action of the creation operator on the eigenfunctions of the unbounded
n = 1 Lamé problem yields the eigenfunctions of the bounded problem multiplied
with a constant. This constant is the same constant appearing in the “normalization”
properties of the eigenfunctions in problem 4.4.
88
References
[1] K. Weierstrass, Zur Theorie der Elliptischen Funktionen, Mathematische Werke,
Bd 2, Teubner, Berlin, pp. 245-255 (1894).
[2] C. G. J. Jacobi, “Fundamenta Nova Theoriae Functionum Ellipticarum”,
Königsberg, Germany: Regiomonti, Sumtibus fratrum Borntraeger (1829)
[3] E. T. Whittaker and G. N. Watson, “A Course of Modern Analysis”, 4th ed.
Cambridge, England: Cambridge University Press, ISBN 1-438-51390-9 (1990)
[4] N. I. Akhiezer, “Elements of the Theory of Elliptic Functions”, Translations of
Mathematical Monographs, Volume 79, American Mathematical Society, ISBN
0-8218-4532-2 (1990)
[5] H. Bateman, A. Erdélyi, “Higher Transcendental Functions”, Volumes 1, 2, 3,
ISBN 0-486-44614-X, ISBN 0-486-44615-8, ISBN 0-486-44616-6 (1953)
[6] M. Abramowitz and I. Stegun, “Handbook of Mathematical Functions with Formu-
las, Graphs, and Mathematical Tables”, United States Department of Commerce,
National Bureau of Standards (NBS) (1964)
[7] G. Lamé, “Sur les Surfaces Isothermes dans les Corps Homognes en Équilibre de
Température”, Journal de mathématiques pures et appliques, 2: 147188. Available
at Gallica (1837)
[8] E. L. Ince, Ordinary Differential Equations, Dover Publications, New York, ISBN
978-0-486-60349-0 (1944)
89

Four Lectures On Weierstrass Elliptic Function and Applications in Classical and Quantum Mechanics

Uploaded by

Copyright:

Available Formats

Four Lectures On Weierstrass Elliptic Function and Applications in Classical and Quantum Mechanics

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Four Lectures On Weierstrass Elliptic Function and Applications in Classical and Quantum Mechanics

Uploaded by

Copyright:

Available Formats

Four Lectures on Weierstrass Elliptic Function and

Applications in Classical and Quantum Mechanics

Lecture 2: Weierstrass Quasi-periodic Functions 17

Lecture 3: Applications in Classical Mechanics 28

Lecture 4: Applications in Quantum Mechanics 48

A The Half-Periods for Real Moduli 59

1.2 Elliptic Functions

f (z + 2ω1 ) = f (z) , f (z + 2ω2 ) = f (z) , (1.1)

z2 − z1 = 2mω1 + 2nω2 , m, n ∈ Z, (1.2)

z1 ∼ z2 ⇒ f (z1 ) = f (z2 ) . (1.4)

A parallelogram defined by the points z0 , z0 + 2ω1 , z0 + 2ω2 and z0 + 2ω1 + 2ω2 ,

2ω2 z0 + 2ω1 + 2ω2

Figure 1 – The fundamental period parallelogram is shaded with blue and an

ω1 0 = aω1 + bω2 , (1.5)

where a, b, c, d ∈ Z. Any period in the lattice defined by ω1 0 and ω2 0 , 2ω = 2m0 ω1 0 +

A = |Im (z1 z̄2 )| . (1.7)

It is a matter of simple algebra to show that

Im ω1 0 ω̄20 = (ad − bc) Im (ω1 ω̄2 ) .

Thus, the new periods can generate the original lattice if

Basic Properties of Elliptic Functions

which vanishes as a consequence of f being a doubly periodic function. Therefore,

g (z) ' cm (z − z0 )m + cm+1 (z − z0 )m+1 + . . .

with cm 6= 0. In the case z0 is a root wj , then m = sj > 0, while in the case z0 is a

g 0 (z) ' mcm (z − z0 )m−1 + (m + 1) cm+1 (z − z0 )m + . . .

h (z) ' h (z0 ) + h0 (z0 ) (z − z0 ) + . . .

f 0 (z + 2ω1 ) = f 0 (z) , f 0 (z + 2ω2 ) = f 0 (z) , (1.15)

Finally, using equation (1.14), we get

This number is called the order of the elliptic function f .

Theorem 1.4. The order of a non-constant elliptic function cannot be equal to 1.

which is the proof of theorem 1.5.

1.3 The Weierstrass Elliptic Function

By construction, this function is doubly periodic with fundamental periods equal to

This function is called Weierstrass elliptic function.

Weierstrass Differential Equation

It follows that the derivative of Weierstrass elliptic function is an odd function

℘0 (−z) = −℘0 (z) . (1.27)

℘02 (z) − 4℘3 (z) + g2 ℘ (z) + g3 = O z 2 .

℘02 (z) = 4℘3 (z) − g2 ℘ (z) − g3 = 0. (1.28)

This differential equation is of great importance in the applications of Weierstrass

which obviously has the solutions, w = ±z + z0 . This implies that y = ℘ (±z + z0 )

We just showed that the general solution of this equation is

y = ℘ (z + 2mω1 + 2nω2 ) = ℘ (z)

The Roots of the Cubic Polynomial

e1 := ℘ (ω1 ) , e2 := ℘ (ω3 ) , e3 := ℘ (ω2 ) . (1.34)

℘0 (ω1 ) = −℘0 (−ω1 ) = −℘0 (2ω1 − ω1 ) = −℘0 (ω1 ) ,

℘0 (ω1 ) = ℘0 (ω2 ) = ℘0 (ω3 ) = 0. (1.35)

Substituting a half-period into Weierstrass equation (1.29), we yield

Q (t) := 4t3 − g2 t − g3 = 4 (t − e1 ) (t − e2 ) (t − e3 ) . (1.37)

This directly implies that ei obey

℘ (z; g2 , g3 ) = −℘ (iz; g2 , −g3 ) . (1.42)

2.1 Quasi-periodic Weierstrass Functions

which fixes the integration constant.

ζ (−z) = −ζ (z) . (2.4)

Quasi-periodicity of the Function ζ

ζ (z + 2ωi ) = ζ (z) + 2ζ (ωi ) . (2.5)

One can easily show inductively that

ζ (z + 2mω1 + 2nω2 ) = ζ (z) + 2mζ (ω1 ) + 2nζ (ω2 ) . (2.6)

Performing the contour integral along the boundary of a cell, we get