BEGINNINGS OF COMPACT RIEMANN SURFACE THEORY

These notes sketch how Siegel space and the symplectic group arise naturally
in the context of compact Riemann surfaces, and how the Jacobi theta function
creates the prime form to construct meromorphic functions and to help prove the
Abel–Jacobi Theorem.

1. Siegel space
Topologically, a compact Riemann surface X of genus g is a g-holed torus, i.e., a
sphere with g handles. Give it a marking consisting of a base point P0 and a canon-
ical homology basis A1 , . . . , Ag , B1 , . . . , Bg . Canonical means that the intersection
numbers of the paths are
Ai · Aj = Bi · Bj = 0, Ai · Bj = δi,j = −Bi · Aj .
(If two oriented paths cross with the direction from the first to the second be-
ing counterclockwise then the intersection number is 1.) Arrayed, these relations
produce the skew matrix, i.e.,
 
A1
 .. 
 . 
   
 Ag    0g 1g call
  · A1 · · · Ag B1 · · · Bg = = J.
 B1 
  −1g 0g
 . 
 .. 
Bg
The space of holomorphic 1-forms on X has dimension g as a complex vector
space. The dimension is nonobvious, proved most intuitively in an old book by Klein
using fluid flows and cut-and-paste techniques. Pick an ordered basis {ω1 , . . . , ωg }.
The period matrix for X—as marked—is
 
Z
 
U = 2 3ω~

4
A 5 
B 2g×g
 
A
where is the homology basis column vector from above and ω ~ is the row
B  
U1 R
vector ω~ = (ω1 , . . . , ωg ). Thus U = where [U1 ]i,j = Ai ωj and [U2 ]i,j =
R U2
ω .
Bi j
Riemann’s Bilinear Relations, assembled into matrix form, are
t
U t JU = 0 and (1/i) · U JU > 0.
The relations also show that U1 is invertible, so U can be normalized:
R right-multiply
by U1−1 , or, equivalently, pick a new basis ω
~ of 1-forms such that Ai ωj = δi,j . The
1
2 BEGINNINGS OF COMPACT RIEMANN SURFACE THEORY

normalized period matrix is  

1
= U U1−1
,
Ω
and Riemann’s Bilinear Relations become, in this case,
Ωt = Ω and Im(Ω) > 0.
That is, Ω ∈ Hg , where Hg is Siegel upper half-space of dimension g. So indeed
Siegel space arises naturally in the Riemann surface context.

2. The symplectic group

Next, consider a change of canonical homology bases. Let a new basis be
 0   
A A
=M , M ∈ GL2g (Z).
B0 B
 0 
A
· A0 B 0
 
The new basis is again canonical exactly when 0 = J, i.e.,
  B
A  
M · A B M t = J, i.e., M JM t = J. This is the defining condition for
B
the symplectic group Spg (Z), equivalent—though not quite trivially—to its usual
form,
M t JM = J.
Thus, the change of canonical homology basis group is Spg (Z) in a natural way.
Assembling the results so far also shows how the symplectic group Spg (Z) acts
on Siegel space Hg . Recovering the familiar formula in the 1-dimensional case,
where SL2 (Z) acts on the complex upper half-plane, requires
 compensating
 for the
z1
notational inconsistency that in one variable, vectors have their bottom
z2
coordinates normalized to 1, while period matrices are normalized
 up top.
 The
0 1 0 1
adjustment is to compose with the outer automorphism M 7→ M
1 0 1 0
of Spg (Z). That is, the action of M on period matrices is defined as
2 3 2 3
0 1 0 1
4 5 4 5
1 0 U10
U20 1 0
     
U1 U2 M
7 → 7→ . 7→
U2 U1 U10
U20
 
A B
If the symplectic matrix M has block structure M = and the normalized
C D
period matrices are Ω = U2 U1−1 and Ω0 = U20 U10−1 , then the relation between them
now works out to the familiar
Ω0 = (AΩ + B)(CΩ + D)−1 .

3. The Jacobian
The period matrix Ω has an associated lattice
L = LΩ = Zg + ΩZg ⊂ Cg .
The complex torus quotient Cg /L is the Jacobian of X, Jac(X). The Jacobian
inherits a natural abelian group structure from Cg .
 BEGINNINGS OF COMPACT RIEMANN SURFACE THEORY 3

The map from the Riemann surface to its Jacobian,

Z P
I : X −→ Jac(X), P 7→ ω
~,
P0
(recall that P0 is the base point of the marking) is well defined, as changing the
path of integration alters the integral by a period. It extends to a map from the
degree-0 divisor group on X,
X X Z Pi
I : D0 −→ Jac(X), ni Pi 7→ ni ω
~.
P0
An easy complex analysis argument shows that I(D` ) = 0, where D` is the divisor
subgroup linearly equivalent to 0, i.e., the divisors of meromorphic functions on X.
Abel’s Theorem says that I induces a group isomorphism,
I : D0 /D` −→Jac(X).
˜
Among other things, this shows which meromorphic functions exist on X: only
those whose divisors map to zero in the Jacobian under I. When g = 0 (recall that
g is the genus), Jac(X) = {0}, and Abel’s Theorem gives the well known fact that
the divisor class group on the Riemann sphere is trivial. When g = 1, the Jacobian
is Jac(X) = X and I is the identity map, so Abel’s Theorem gives the familiar
constraint on the possible zeros and poles of a meromophic function on the torus.

4. Theta functions and Abel’s Theorem

Let e(x) = e2πix for x ∈ C. I will prove part of Abel’s Theorem using the Jacobi
theta function,
X 1
ϑ(z, Ω) = e( nt Ωn + nt z), z ∈ Cg , Ω ∈ Hg .
g
2
n∈Z
The theta series has absolute convergence properties justifying rearrangement. It
is quasi-periodic with respect to the lattice LΩ , meaning
ϑ(z + m, Ω) = ϑ(z, Ω), m ∈ Zg ,
but
1
ϑ(z + Ωm, Ω) = ϑ(z, Ω) · e(− mt Ωm − mt z), m ∈ Zg .
2
These are easy to show. For the first identity, note that 12 nt Ωn + nt (z + m) differs
from the original exponent by an integer; for the second, compute
1 t 1 1
n Ωn + nt (z + Ωm) = (n + m)t Ω(n + m) + (n + m)t z − mt Ωm − mt z,
2 2 2
and the result follows by summing over n + m rather than n.
Suppose that ε ∈ Cg satisfies ϑ(ε) = 0. (From now on Ω is fixed, so ϑ will only
receive one argument.) Define for any pair of points x, y ∈ X,
Z y
Eε (x, y) = ϑ(ε + ω
~ ).
x
This is the prime form on X. (The name “prime form” will become clear soon.) It
is locally well-defined once points x0 , y0 ∈ X and a path from x0 to y0 are chosen
and then the points are perturbed slightly. If moving the points around X back
to themselves changes the path between them by an Ak , the prime form is left
invariant since the integral changes by an integer vector. But if the path changes
4 BEGINNINGS OF COMPACT RIEMANN SURFACE THEORY

by a Bk , the integral changes by the kth column of Ω, i.e., by Ωek (where ek is

a standard basis vector). Quasi-periodicity shows that the effect is to multiply
Eε (x, y) by the factor
Z y Z y
1 1
e(− etk Ωek − etk (ε + ~ )) = e(− Ωk,k − εk −
ω ωk ),
2 x 2 x
where the integral is along the original path.
Pd
Returning to Abel’s Theorem, we are given a degree-0 divisor δ = i=1 Zi −
Pd P ZiR R P
i=1 Pi such that I(δ) = 0 in Jac(X), i.e., ( P0 − P0i )~
ω ∈ LΩ . Choose specific
paths from P0 to the Zi and from P0 to the Pi . Having done so, change one of the
paths so that even in the full space Cg ,
X Z Zi Z Pi
( − )~
ω = 0.
P0 P0
The goal is to find a meromorphic function f on X with zeros Zi and poles Pi .
Such a function is Qd
Eε (Zi , y)
f (y) = Qi=1
d
.
i=1 Eε (Pi , y)
This clearly has zeros at the Zi and poles at the Pi , and by some technicalities that
I am skipping, ε can be chosen so that f has no other zeros or poles.
R y To ensure that f is well defined, make the convention that in Eε (Zi , y) = ϑ(ε +
Zi
ω
~ ), the path of integration proceeds from Zi to P0 as specified a moment ago,
and then from P0 on to y. Similarly for Eε (Pi , y), using the same path from P0
to y. Under this convention, f (y) depends only on the choice of path from P0 to
y. As before, altering this path by an Ak leaves each term of f (y) invariant. And
altering by a Bk multiplies f (y) by a net factor of
X 1 Z y Z y
1
e( (− Ωk,k − εk − ωk + Ωk,k + εk + ωk ))
2 Zi 2 Pi
which simplifies to
XZ Zi Z Pi
e( ( − )ωk ) = e(0) = 1,
P0 P0
by how the paths from P0 to the Zi and the Pi were specified. So indeed f makes
sense and has the required divisor.
The function Eε (x, y) is called the prime form on X because, as just shown, it
is the basic building block for all meromorphic functions. By analogy, the function
E(x, y) = x − y is the prime form on the Riemann sphere since any meromorphic
function there takes the form
Q
E(Zi , y)
f (y) = c Q .
E(Pi , y)