Automorphism group of compact Riemann surfaces

Sergio Chaves

July 21, 2018

BIn this report we will study Aut(X), the automorphism group of a compact Riemann surface X
(CRS for short), The main purpose will be to prove the Hurwitz automorphism theorem which states
that Aut(X) is finite and its order is bounded by 84(g − 1) where g ≥ 2 is the genus of the Riemann
surface. We will show that this bound is sharp, and even there exists infinitely many values of g
where the equality is attached. Also, we will discuss further properties of the automorphisms of a
Riemann surface and some explicit examples of Aut(X).
The basic theory of Riemann surfaces is assumed and will be omitted; however, we will include a
few fundamental results that will be useful for the purposes of this document.

1 Preliminaries
In this section we will follow [Forster, 1991] for a review of important results in the theory of Riemann
surfaces.
Let f ∶ X → Y be a n-sheeted covering between compact Riemann surfaces. If x ∈ X is a ramification
point, then locally f (z) = z k for some charts U ⊆ X, V ⊆ Y around x, f (x) respectively and k > 1, we
write by νx = k the ramification index of f at x. Since each x is an isolated point and X is compact,
we have that X admits only finitely many ramification points. Therefore, the following expression

νf = ∑ νx − 1
x∈X

is well defined. The Riemann-Hurwitz theorem states

Theorem 1.1 (Riemann - Hurwitz Theorem). Let gx and gy be the genus of X and Y respectively.
Then
νf
gx − 1 = n(gy − 1) +
2
or equivalently,
χ(X) = nχ(Y ) − νf .

For any Riemann surface X, let M(X) denote the field of meromorphic maps on X. Recall that
any meromorphic function has as many zeros as poles.
Let D a divisor on X, write by L(D) = {f ∈ M(X) ∶ (f ) + D ≥ 0} and l(D) = dim L(D).

1
Theorem 1.2 (Riemann-Roch Theorem). Let X be a compact Riemann surface of genus g and K
be a canonical divisor of X. Then

l(D) − l(K − D) = deg(D) + 1 − g

An important consequence of the Riemann-Roch theorem is the existence of non-constant meromor-

phic functions.
Proposition 1.3. Let X be a CRS of genus g and let a ∈ X. There exists f ∈ M(X) that has a
pole of order at most g + 1 at a and it is elsewhere holomorphic.

Proof. Consider the divisor D = (g + 1) ⋅ a, the Riemann-Roch theorem implies that l(D) ≥ deg(D) +
1 − g = 2. Therefore, for f ∈ L(D), f has its only pole at a (otherwise (f ) + D ≱ 0) and it is of order
at most g + 1.

2 The automorphism group

Let X be a Riemann surface, the set Aut(X) = {f ∶ X → X biholomorphic } is a group under
composition and it is called the automorphism group of X. When X is simply connected, (not
necessarily compact) this group has an specific presentation. From the Uniformization theorem, X
is either isomorphic to the complex plane C, the upper half space H or the projective space P1
az + b
Proposition 2.1. Aut(C) = {f (z) = az + b ∶ a ≠ 0} and Aut(P1 ) = {f (z) = ∶ ad − bc ≠ 0} ≅
cz + d
SL(2; C)/ ± I = P SL(2, C).

Proof. Let us start with the case of X = P1 since any automorphism of C can be extended to an
automorphism of P1 sending ∞ to ∞. Let f ∈ Aut(P1 ), then f ∈ M(P1 ) and thus f is a rational
function f (z) = p(z)/q(z). Assume that p and q has no common zeros. Since f is injective, p can
only have exactly one zero (from the fundamental theorem of algebra) and thus p is a linear map;
hence, f has only one pole and thus q is also linear. That is,
az + b
f (z) =
cz + d
for some a, b, c, d ∈ C. The existence of f −1 is equivalent to ad − bc ≠ 0. Moreover, dividing by
√ a b
± ad − bc we may assume that ( ) ∈ SL(2, C)/{±I} = P SL(2, C) and so Aut(P1 ) ≅ P SL(2, C).
c d
Now, any automorphism f ∶ C → C can be extended to an automorphism f ∶ P1 → P1 by f (∞) = ∞.
If f (z) = az + b/cz + d then c = 0 and d ≠ 0 (otherwise f (∞) ≠ ∞) and so f (z) is a linear map.

Proposition 2.2. Aut(H) ≅ P SL(2, R).

a b
Proof. Let fA (z) = az+b
cz+d
with A = ( ) ∈ P SL(2, R). Now it is straightforward to check that
c d
im(fA (z)) = ∣cz+d∣
ad−bc
2 im(z) and so fA is a well defined map f ∶ H → H. We will show that any
f ∈ Aut(X) is of the form f (z) = az+b
cz+d
. Consider the map G ∶ C → C given by G(z) = i−z
i+z
, this

2
map induces a biholomorphism G ∶ H → D where D denotes the unit disc, and satisfying G(i) = 0.
Suppose that f ∈ Aut(X) is such that f (i) = i and consider the map g = G ○ f ○ G−1 ∶ D → D,
notice that g(0) = 0 and thus from the Schwartz g(z) = λz with λ = eiθ for some θ ∈ R. Now
cos(θ/2) sin(θ/2)
consider the matrix A = ( ) ∈ SL(2, R) and the linear map fA (z). This map
− sin(θ/2) cos(θ/2)
satisfies the condition of Schwarts lemma again and since fA (i) = i and fA′ (i) = eiθ the composite
G ○ fA ○ G−1 (z) = eiθ = G ○ f ○ G−1 and thus fA = f .
√
im(z0 ) √0
Finally let z0 ∈ H. Consider the matrix M = ( ) this matrix satisfies fM (i) =
0 1/ im(z0 )
1 z0 − fM (i)
im(z0 ) and the matrix N = ( ) induces a translation map in H that sends fM (i) to z0 .
0 1
Then the composite h = fN ○ fM = fN M is an automorphism of H that sends i to z0 .
Now we will prove that for any f ∈ Aut(X) there is a matrix A ∈ SL(2, R) such that f = fA . For such
f there is a z0 such that f (z0 ) = i, and from the above construction, there is an h ∈ Aut(X) such
that h(i) = z0 . The composite h ○ f is an automorphism of H that sends i to i and thus h ○ f = fB
for some B ∈ SL(2, R) using the earlier construction. Since f = h−1 ○ fB and h is also of the form fC
we get f = fC −1 ○ fB = fC −1 B = fA .

The automorphism group of simply connected Riemann surface is infinite; in particular, since any
compact Riemann surface of genus g = 0 is isomorphic to P1 we conclude that Aut(X) is an infinite
group for the case of genus zero.
Now let X be a simply connected Riemann surface and f ∈ Aut(X). Denote by FixX (f ) the set
of elements z ∈ Z such that f (z) = z, since f (z) = az+bcz+d
, any fixed point of f satisfies the equation
z = az+b
cz+d
, or equivalently, cz 2
+ (d − a)z − b = 0, therefore ∣ FixX (f )∣ ≤ 2 if f ≠ id. This is not a
coincidence, and actually for any Riemann surface X and any f ∈ Aut(X) the number of fixed
points of f is finite and it is bounded by the genus of X, as we will illustrate in the following result.
Theorem 2.3. Let X be a compact Riemann surface of genus g and f ∈ Aut(X). If f ≠ id then
∣ FixX (f )∣ ≤ 2g + 2.

Proof. Let p ∈ X such that f (p) ≠ p. From Proposition 1.3, there exist a meromorphic function
g ∶ X → P1 that has a pole of order r ≤ g + 1 at p and it is holomorphic in X ∖ {p}. Consider the
meromorphic map h(z) = g(z) − g(f (z)) and notice that for any x ∈ FixX (f ), h(x) = 0 and since
the zeroes of h are finite by compactness, we conclude that FixX (f ) is finite. Moreover, as g has
a single pole of order r at p, g ○ f has a single pole of order r at f −1 (p); thus h has two poles of
combined order at most 2r ≤ 2g + 2. Since any meromorphic function has the same number of zeros
and poles, and FixX (f ) is a subset of the zeros of h, we conclude that ∣ FixX (f )∣ ≤ 2g + 2.

3 Automorphism of complex tori

Let X be a compact Riemann surface of genus g = 1, recall that a consequence of the Abel’s
theorem [Forster, 1991, §20,21] is that X is isomorphic to a complex torus C/Λ. Observe that any
complex torus inherits a group structure under the addition, so the translation by elements x ∈ X
is an automorphism of X. This shows that for surfaces of genus g = 1, Aut(X) is infinite. However,

3
in this section we will describe explicitly the automorphism group of a complex torus X = C/Λ for
Λ a lattice in C.
Theorem 3.1. Let X be a compact Riemann surface of genus g = 1, from the uniformization
theorem, X ≅ C/Λ. Then Aut(X) is infinite and Aut(X) ≅ X × G where G is a cyclic group of order
either 2, 4 or 6.

Proof. We will proof this statement in several steps.

1. Any f ∈ Aut(X) is of the form f (z mod Λ) = az + b mod Λ and aΛ = Λ and a is a n-th root
of unity of some n.
Let π ∶ C → X be the projection map which is a covering map. Since f is unramified (from
the Riemann-Hurwitz theorem), f is a covering and so the composite f ○ π is a covering map.
Recall that C is the universal covering of X and thus there is a map F ∶ C → C that makes the
diagram
F
C C
π π
f
X X
commutative. Moreover, from uniqueness of the universal covering, F ∈ Aut(C) and from
Proposition 2.1, F (z) = az + b for some a, b ∈ C and a ≠ 0. The commutative of the above
diagram imply that aΛ ⊆ Λ and f (z mod Λ) = az + b mod Λ. The reverse inclusion follows
from considering f −1 . Now it only remains to show that ∣a∣ = 1. In fact, let λ ∈ Λ ∖ {0} of
minimal length, now since aΛ = Λ we get that ∣λ∣ ≤ ∣aλ∣ and thus ∣a∣ ≤ 1. The reverse inequality
holds by considering a−1 . Now notice that S = {γ ∈ Λ ∶ ∣γ∣ = ∣λ∣} is finite, from the above
remark we see that F permutes all the elements of S and thus F n ∣S is the identity map for
some n ≥ 1. This implies that an = 1 and so a is n-root of unity.
2. Aut(X) ≅ Autt (X) ⋊ Aut0 (X) where Autt (X) are the translation automorphisms, i.e., the
maps of the form f (z) = z + b for b ∈ X and Aut0 (X) are the automorphism of X fixing 0. It
is easy to see that both Autt (X) and Aut0 (X) are subgroups of Aut(X).
From the previous item, any f ∈ Aut(X) is of the form f (z) = az + b for some a, b with ∣a∣ = 1
and b ∈ X. Then the map Φ ∶ Aut(X) → Aut0 (X) given by Φ(f )(z) = f (z) − f (0) = az is
a surjective group homomorphism whose kernel is precisely Autt (X) and so it is a normal
subgroup of Aut(X). Also, since in general f ○ g ≠ g ○ f for f ∈ Autt (X) and g ∈ Aut0 (x) we
conclude that Aut(X) ≅ Autt (X) ⋊ Aut0 (X).
Finally, notice that the map b ↦ f (z) = z +b is an isomorphism of groups from X and Autt (X),
this implies that Aut(X) ≅ X ⋊ Aut0 (X).
3. Aut(X) is an infinite group; however, Aut0 (X) is finite and it is either isomorphic to Z/2Z,
Z/4Z or Z/6Z.
The first claim follows easily since there are infinitely many translation automorphism on X.
Now let f ∈ Aut0 (X), that is f (z) = az with aΛ = Λ and ∣a∣ = 1. If a = ± , f (z) is either the
identity map or the group inversion of X respectively, these maps always exist in Aut0 (X) and
so ∣ Aut0 (X)∣ ≥ 2. Assume now that a is not real and let λ ∈ Λ ∖ {0}, then λ and aλ generate
the lattice Λ and since a2 λ ∈ Λ, we get an equation a2 λ = maλ + bλ. Dividing out by λ we
obtain a quadratic equation with integers coefficients a2 − ma − b = 0. As ∣a∣ = 1 we conclude

4
 that a is a root of unit that satisfies a quadratic equation, that is, a is either a 3rd, 4th or
6th root of unit. In the first and later situation, we have Aut0 (X) ≅ Z/6Z, and in the second
situation Aut0 (X) ≅ Z/4Z. If there is no such complex number a, we obtain Aut0 (X) ≅ Z/2.
The last isomorphism has a geometric meaning: Recall that λ and aλ are generators of Λ of the
same length if a is non-real. If Aut0 (X) ≅ Z/4Z then a = ±i and thus λ and iλ form an angle of π/2
in the plane, that is, Λ is an square lattice. In the case of Aut0 (X) ≅ Z/6Z, the generators form an
angle of π/3 and thus Λ is an hexagonal lattice.

4 Hurwitz automorphism theorem

In 1878 Schwartz proved that for any Riemann surface X of genus g ≥ 2, the group Aut(X) is finite.
Later, in 1894 Hurwitz [Hurwitz, 1892] found an upper bound for the order of Aut(X) which depends
on the genus of X. We will recreate his proof that follows from the Riemann-Hurwitz theorem and a
combinatorial argument on the ramification index of the possible ramification points that can show
up in the formula.
Let X be a CRS and G be a finite group acting by automorphism of X, that is, the map lg (x) = g ⋅ x
is a biholomorphism. In the case of smooth manifolds or without assuming compactness, we need
extra conditions (properly discontinuous action for example) on the action to make the orbit space
X/G a smooth manifold; however, the rigidity of compact Riemann surfaces assures that we do not
need to assume extra conditions and actually X/G is a Riemannn surface and the projection map
π ∶ X → X/G is a branched covering.
Proposition 4.1. Consider a group action of a finite group G on a CRS X, for any x ∈ X, denote
by Gx = {g ∈ G ∶ g ⋅ x} the stablizer subgroup and S = {x ∈ X ∶ ∣Gx ∣ > 1} is finite

Proof. Since X is compact, we only need to show that S is discrete. We will show that X has no
accumulation points. In fact, let x ∈ X an accumulation point and xn → x a sequence of elements
xn ∈ S converging to x. For each xn there is a non trivial element gn ∈ Gxn . As G is finite, there
is a g ∈ G such that g = gn for infinitely many values of n; so we can assume that g = gn for every
n. Therefore, we obtain g ⋅ x = g limn xn = limn g ⋅ xn = limn xn = x. Therefore, we get an infinite set
with an accumulation point where the maps lg and le = id coincide. From the identitiy theorem in
complex analysis we obtain that g = e which is a contradiction.
Theorem 4.2. The complex structure of X induces a complex structure on the orbit space X/G
such that the projection map π ∶ X → X/G is a holomorphic map of degree ∣G∣ and for any x ∈ X,
the ramification index νx (π) = ∣Gx ∣.

Proof. It will follows from the existence of a neighborhood U ⊆ X around any x ∈ X satisfying the
following properties.
ˆ gU = U for any g ∈ Gx .
ˆ U ∩ gU = ∅ for any g ∉ Gx .
ˆ The induced map α ∶ U /Gx → X/G is a homeomorphism onto an open subset of X/G.
ˆ If y ∈ U and g ⋅ y = y for some g ∈ Gx , then y = x.

5
Let y ∈ X/G be a ramification point, then π −1 = {x1 , . . . , xs } are in the same G-orbit and so their
stabilizers are conjugate subgroups, thus they have the same order r. From the orbit stabilzer
theorem we get that s = ∣G∣/r. Now we can derive Hurwitz bound on the order of ∣G∣ and it is a
direct consequence of the Riemann-Hurwitz theorem.
Theorem 4.3 (Hurwitz,1893). Let G be a finite group acting by automorphism on a compact Rie-
mann surface X of genus g ≥ 2. Then ∣G∣ ≤ 84(g − 1).

Proof. Let Y = X/G and g ′ denote the genus of Y . Y has only finitely many ramification points
y1 , . . . , yn and π −1 (yi ) = {xi1 , . . . , xisi }. From the Riemann-Hurwitz theorem we get
n si
2g − 2 = ∣G∣(2g ′ − 2) + ∑ ∑ νxij (π) − 1
i=1 j=1
n si
= ∣G∣(2g ′ − 2) + ∑ ∑ ri − 1
i=1 j=1
n
= ∣G∣(2g ′ − 2) + ∑ si (ri − 1)
i=1
n
∣G∣
= ∣G∣(2g ′ − 2) + ∑ (ri − 1)
i=1 ri
n
1
= ∣G∣ (2g ′ − 2 + ∑ 1 − ) = ∣G∣(2g ′ − 2 + R).
i=1 ri

Note that if R ≠ 0, R ≥ 1/2 since ri ≥ 2. Now we analyze the possible values of g ′ and R to derive
the desired bound.
ˆ Case g ′ ≥ 1. If R = 0, the formula turns into 2(g − 1) = ∣G∣(2g ′ − 2). Since g > 1 and ∣G∣ > 1 we
must have g ′ ≥ 2 and thus (2g ′ − 2) ≥ 2. So we obtain ∣G∣ ≤ g − 1. Now if R ≠ 0, we have that
R ≥ 1/2 and thus 2(g − 1) ≥ ∣G∣/2 and so ∣G∣ ≤ 4(g − 1).
ˆ Case g ′ = 0. The equation turns into 2g − 2 = ∣G∣(−2 + R); and since the left hand side is
positive, we must have R > 2. Now we need to look at the number of ramification points. If
n ≥ 5, then R ≥ n/2 ≥ 5/2 and so ∣G∣ ≤ 4(g − 1). Now assume n = 4. If ri = 2 for i = 1, 2, 3, 4
we get R = 2 which is impossible. So we must have at least r1 ≥ 3 and ri ≥ 2 for i = 2, 3, 4.
Then we get R ≥ 2/3 + 3(1/2) = 13/6, so we obtain ∣G∣ ≤ 12(g − 1). Now we look at n = 3,
again we must have at least r1 ≥ 3 and r2 , r3 ≥ 2. If (r1 , r2 , r3 ) = (3, 3, 3), (3, 2, 6) or (3, 4, 4)
then R = 2 which is impossible and these are the only cases where R = 2. So we need to
assume that R > 2. Since r1 ≥ 3 we get that (1 − 1/r2 ) + (1 − 1/r3 ) > 2 − 2/3 = 4/3. If we
let r2 ≥ 2 we get (1 − 1/r3 ) > 4/3 − 1/2 = 5/6, in this case we obtain r3 ≥ 7; finally we obtain
R ≥ 2/3 + 1/2 + 6/7 = 2 42
1
. Moreover, this is the minimum value that R can attach and so
∣G∣ ≤ 84(g − 1).
Hurwitz knew that its bound is sharp. In fact, in 1879 Klein [Klein, 1878] studied a Riemann surface
of genus 3 with exactly 184 automorphism. So he developed some criterion to determine when the
action of a finite group G, ∣G∣ attains the bound. However, he left open the question whether there
exists other Riemann surfaces that does attain the bound.
Definition 4.4. We say that a finite group G is a Hurwitz group if it acts by automorphism on a
CRS X of genus g and ∣G∣ = 84(g − 1).

6
Hurwtiz proved the following condition for a group G to be a Hurwitz group.
Proposition 4.5. G is a Hurwitz group if and only if it is generated by some elements x, y ∈ X that
satisfy the relations x2 = y 3 = (xy)7 = e (and possibly other relations).
A lower bound for ∣ Aut(X)∣ was discovered independently by Accola [Accola, 1968] and Maclachlan
[Maclachlan, 1969]. They constructed a group of order 8g + 8 acting on a Riemann surface of genus
g and showed that this is the minimum possible order of a finite group action on X. So we can
summarize
Theorem 4.6. Let X be a compact Riemann surface of genus g ≥ 2. Then 8g + 8 ≤ ∣ Aut(X)∣ ≤
84(g − 1).
Klein showed by explicit computation that if g = 2 , ∣ Aut(X)∣ ≤ 48. Later, Gordan showed that if
g = 4, ∣ Aut(X)∣ ≤ 120 and Wiman showed that for g = 5 and g = 6 we have ∣ Aut(X)∣ ≤ 192 and
∣ Aut(X)∣ ≤ 420 respectively. Therefore, for low-genus Riemann surfaces, the bound is not attained
except when g = 3 since the Klein-quartic does attain the bound. Moreover, Macbeath [Macbeath,
1961] showed that there are infinitely many values of g such that a Hurwitz group of order 84(g − 1)
is realized. On the other hand, in the same paper, Accola and Maclachaln constructed infinitely
many values of g with a compact Riemann surface of genus g having exaclty 8g + 8 automorphisms.

5 Aut(X) is finite
In this section we will provide a skectch of several proofs of the fact that Aut(X) is a finite group
for X a compact Riemann surface of genus g ≥ 2. First, we will illustrate Schwartz proof using
Weierstrass points; then we will show how the Fuschian groups and Hyperbolic geometry implies
Hurwitz automorphism theorem and finally we will show that there is a faithful representation of
Aut(X) onto the isomorphism of the space of holomorphic 1-forms of X Ω1 (X). For further details
on the proofs the Farkas’s book [Farkas and Kra, 1992] is the followed reference.
From the rest of this section we will assume that X is a compact Riemann surface of genus g ≥ 2.

Weierstrass points

Let us start with the following motivation. Let X be a CRS and f ∈ M(X) a meromorphic function
with Ordp (f ) = −1 and holomorphic else where. Then f induces a biholomorphism X ≅ P1 . The
existence of such f is equivalent to l(D) > 1 where D = 1 ⋅ p is a divisor. Therefore, if X has genus
greater than zero, such function f can not exist.
Now assume X is a CRS of genus g > 0, we might ask now, for which values of n does not exist
f ∈ M(X) with Ordp (f ) = −n and holomorphic elsewhere? this is equivalent to ask l(Dn ) < 2 where
Dn = n ⋅ p is a divisor.
From the Riemann Roch theorem 1.2 we get

l(Dn ) − l(K − Dn ) = deg(Dn ) + 1 − g

If we want l(K −D), we may require deg(K −Dn ) < 0, or equivalently, n > 2g−2 since deg(K) = 2g−2.
Therefore, for n ≥ 2g − 1, l(Dn ) = n + 1 − g; moreover, if n ≥ 2g, l(Dn ) ≥ g + 1 ≥ 2 and so the required
function f exists.

7
Notice that l(Dj+1 ) ≤ l(Dj ) + 1 and then in the sequence

1 = l(D0 ) ≤ l(D1 ) ≤ ⋯l(D2g ) = g + 1

the increasing 2g-steps are of size at most 1 and so there exists integers 1 = n1 , i . . . , ng such that
l(Dni ) = l(Dni −1 ). In this case, there is no function f ∈ M(X) with Ordp (f ) = ni . These numbers
are called gap numbers of p. If n is a gap number of p, then l(Dn−1 )−l(Dn ) = 0. Using the Riemann-
Roch formula in each term, the equation turns into l(K − Dn−1 ) − l(K − Dn ) = 1, or equivalently,
there exists a holomorphic differential ω on X such that Ordp (ω) = n − 1.

Now let V = Ω1 (X) be the space of holomorphic differentials. Recall that dim V = g , and by
induction, there is a basis of V that locally is locally given by holomorphic functions ϕj , j = 1, . . . , g
satisfying
Ordp (ϕ1 ) < Ordp (ϕ2 ) < ⋯ < Ordp (ϕg )
(Let d1 = min{Ordp (ω) ∶ ω ∈ V } and ω1 = ϕ1 dz around p, write V = ω1 ⊕ V ′ and use induction on
V ′ ). And now define the Wronskian as

⎛ ϕ′1 ⋯ ϕg ⎞
⎜ ϕ ⋯ ϕ′g ⎟
W = det ⎜ ⋮1 ⋮
⎟
⎜ ⎟
⎝ϕ(g−1) ⋯ ϕ
(g−1) ⎠
1 g

So we will prove now that

Proposition 5.1. Ordp W = ∑gi=1 dj − j + 1 where dj = Ordp (ϕj ).

Proof. Write W = det(ϕ1 , . . . , ϕg ). Notice that det(f ϕ1 , . . . , f ϕg ) = f g (ϕ1 , . . . , ϕg ). We will proceed

now by induction on g. If g = 1, W = ϕ1 and thus Ordp (W ) = Ordp (ϕ1 ). Now supose that it holds
for k and we will prove it for k + 1. In fact,

det(ϕ1 , ϕ2 , . . . , ϕk+1 ) = ϕk+1

1 det(1, ϕ2 /ϕ1 , . . . , ϕk+1 /ϕ1 ) = ϕk+1
1 det((ϕ2 /ϕ1 )′ , . . . , (ϕk+1 /ϕ1 )′ )

And thus by induction we get

k+1 k+1 k+1
Ordp W = (k + 1)d1 + ∑ (dj − d1 − 1) − (j − 2) = (k + 1)d1 + ∑ dj − d1 − j + 1 = ∑ dj − j + 1
j=2 j=2 j=1

Notice that in fact, dj + 1 = nj the j-th gap number at p. So Ordp (W ) = ∑gj=1 nj − j. Combining this
result with the gaps points, we have then
Theorem 5.2. The following are equivalent.
1. Ordp W > 0.
2. W (p) = 0.
3. There exists a non-zero holomorphic differential ω ∈ Ω1 (X) such that Ordp (ω) ≥ g.
4. l(K − Dg ) > 0.

8
 5. l(Dg ) ≥ 2.
6. The gap numbers of p are {1, . . . , g}.
Definition 5.3. A point p ∈ X satisfying one (and all) of the above conditions is called a Weierstrass
point of X.
Since the Weierstrass points are isolated, they are finite. Call the set of Weierstrass points of X
W (X). Now we will find a lower and upper bound for ∣W (X)∣. First we need the following result.
g(g+1)
Proposition 5.4. W is a holomorphic m-differential where m = 2
and deg(W ) = g 3 − g.

Proof. Let {ω1 , . . . , ωg } be a basis for Ω1 (X). Choose a chart U around p such that ωj = ϕj (z)dz
and let w = f (z) a holomorphic change of coordinates given by the choosing of another chart
V . In this chart we have ωj = ψ(w)dw and thus in the common domain we have the relation
ψ(w) = ψ(f (z))f ′ (z) = ϕj (z). Therefore, we obtain

det(ϕ1 , . . . , ϕg ) = det(ψ1 (f )f ′ , ψ2 (f )f ′ , . . . , ψg (f )f ′

by elemental operations in the last matrix, we obtain

′
⎛ ψ 1 (f )f ⋯ ψg (f )f ′ ⎞
′ ′ 2
⎜ ψ (f )(f ) ⋯ ψg′ (f )(f ′ )2 ⎟
W = det ⎜ 1 ⋮ ⋮
⎟
⎜ ⎟
⎝ψ (g−1) (f )(f ′ )g ⋯
(g−1)
ψg (f )(f )′ g⎠
1

And finally we obtain det(ϕ1 , . . . , ϕg ) = (f ′ )

g(g+1) g(g+1)
2 det[ψ1 , . . . , ψg ](f ) and so W is a m = 2
differential, and its degree is (2g − 2)m = g 3 − g.

So we have g 3 − g = deg(W ) = ∑p∈X Ordp (W ) and thus we obtain:

Corollary 5.5. The number of Weierstrass points ∣W (X)∣ ≤ g 3 − g.
For our purposes, we need to find a lower bound. To do that, we need to look at the sequence of
non-gap numbers of p. That is, let m1 , . . . , mg denote the numbers that are not gap numbers in the
sequence {1, . . . , 2g}. Notice that mg = 2g and for each mi there exist a meromorphic function fi
such that Ordp (fi ) = mi . It is straightforward to check the following properties of this sequence
ˆ For each 0 < i < g, mi + mg−i ≥ 2g.
ˆ If m1 = 2, then mi = 2i and mi + mg−i = 2g for all i.
ˆ if m1 > 2 there is a j, 1 < j < g such that mj + mg−j > 2.
From these properties we get
Proposition 5.6. ∑g−1
j=1 mj ≥ g(g − 1) with equality if and only if m1 = 2.

As a corollary we obtain
Corollary 5.7. ∣W (X)∣ ≥ 2g + 2.

9
Proof. Let p ∈ X, n1 < . . . < ng be the gap numbers at p and m1 < ⋯ < mg the non-gap numbers at
p. Then from Proposition 5.1 we have
g 2g g g
Ordp (W ) = ∑ nj − j = ∑ j − ∑ mj − ∑ j
j=1 j=1 j=1 j=1
2g−1 g−1
= ∑ j − ∑ mj
j=g+1 j=1
3 g(g − 1)
≤ g(g − 1) − g(g − 1) =
2 2
Since deg(W ) = g 3 − g there are at least deg(W )/ Ordp (W ) = 2g + 2 Weierstrass points.

Let F ∶ X → Y a holomorphic map between CRS. Define F ∗ ∶ Div(Y ) → Div(X) by setting

F ∗ D(p) = νp (F )D(F (p)) where νp (F ) is the ramification index of F at p. Let f ∈ L(D) , notice
that Ordp F ∗ (f ) = Ordp (f (F )) = νp (F ) Ordp (f ) ≥ −νp (F )D(p) = −F ∗ D(p), that is, f ∈ L(F ∗ D).
So if F is an isomorphism, we have L(D) ≅ L(F ∗ D). Now we can prove
Proposition 5.8. Let T ∈ Aut(X)be a an automorphism of a CRS of genus g ≥ 2. If p ∈ X is a
Weierstrass point, so is T (p).

Proof. If p is a Weierstrass point, the divisor D = g ⋅ p satisfies l(D) ≥ 2. Notice that T ∗ D = g ⋅ T (p)
and by the above Remark we obtain that l(T ∗ D) = l(D) ≥ 2 and thus T (p) is also a Weierstrass
point.

Now we will see that the Weierstrass points have a particular characterization when X is a hyper-
elliptic Riemann-Surface.
Definition 5.9. We say that X is hyperelliptic if there is a two-sheeted branched covering f ∶ X →
P1 .
The existence of the covering f is equivalent to the existence of a divisor D with l(D) ≤ 2 and
deg(D) = 2. In particular, from the Riemann-Hurwitz theorem, the map f must have 2g + 2 branch
points in P1 .
The following result relates the Weierstrass points on a hypereliptic Riemann surface and the auto-
morphism of X.
Proposition 5.10. Let X be a compact Riemann surface. then
1. X has at least 2g +2 Weierstrass points and it has exaclty 2g +2 if and only if X is hyperelliptic.
2. X is hyperelliptic if and only if there is an automorphism J ∈ Aut(X) such that J 2 = id and
the fixed points of J are exactly the Weierstrass points.
3. If X is hyperelliptic, for any T ∈ Aut(X), T ≠ I, J, T has at most 4 fixed points.
Now we can sketch Schwartz arguing.
Theorem 5.11. ∣ Aut(X)∣ is finite for g ≥ 2.

Proof. Write W (X) the (finite) set of Weierstrass points of X.

10
 ˆ If T ∈ Aut(X), T (W (X)) = W (X). That is, T induces a permutation of the Weierstrass points
of X by Proposition 5.8.
ˆ If X is not hyperelliptic, from Theorem 2.3, the only automorphism that fixes all the Weier-
strass points is the identity map.
ˆ If X is hyperelliptic, any automorphism different from J and id fixes at most 4 < 2g + 2
Weierstrass points.
ˆ Let S(W (X)) denote the group of permuatations of the set W (X). The induced map Φ ∶
Aut(X) → S(W (X)) is injective unless X is hyperelliptic. In the latter case ker(Φ) = {id, J}.
ˆ Φ is a group homomorphism of Aut(X) onto a finite group and it has finite kernel. Thus
Aut(X) is finite.

Hyperbolic geometry

Recall that from the uniformization theorem, for a compact Riemann surface X of genus g ≥ 2, the
upper half plane H is its universal covering and we can realize X as the quotient H/Λ where Λ is a
subgroup of Aut(H) ≅ P SL(2, R) that acts properly discontinuous on H. Poincare [Poincaré, 1882]
in 1882 showed that a subgroup G ⊆ Aut(H) acts properly discontinuous on H if and only if G is a
discrete subgroup; and he introduced the notion of Fuschian group which is a discrete subgroup of
P SL(2, R). Poincare introduced this notion while he was studying the Hyperbolic plane modeled as
the upper half plane H. Even though he did not make a direct connection with the automorphism
group of compact Riemann surfaces, in 1945 Siegel [Siegel, 1945] proved the finiteness of Aut(X)
and Hurwitz bound by using the geometry of the Hyperbolic space and relating Fuschian groups
with Aut(X).
Theorem 5.12. Aut(X) is finite and ∣ Aut(X)∣ ≤ 84(g − 1)∣.

Proof. 1. (Siegel) For any Fuschian subgroup G ⊆ P SL(2, R), Area(H/G) ≥ π/21.
2. Realize X = H/Λ where Λ is a discrete subgroup of P SL(2, R).
3. The normalizer subgroup Γ = N (Λ) of Λ in P SL(2, R) is discrete.
4. Aut(X) ≅ Γ/Λ .
5. Consider the Riemann surface Y = H/Γ and the canonical map π ∶ X → Y is a covering map
of degree [Γ ∶ Λ].
Area(X)
6. Area(Y )
= [Γ ∶ Λ].
7. From Gauss-Bonnet theorem Area(X) = −2πχ(X) = 2π(2g − 2).
8. Finally we get
Area(X) 2π(2g − 2)
∣Aut(X)∣ = [Γ ∶ Λ] = ≤ = 84(g − 1)
Area(Y ) π/21

11
Representation of Aut(X) on the space of holomorphic 1-forms

Recall that Ω1 is a finite dimensional vector space of dimension g. Let ω ∈ Ω1 and T ∈ Aut(X). Let
x ∈ X and choose chart around x such that locally ω = f (z)dz where f is a holomorphic map. Also
choose a chart around T (x) such that T −1 have the form h(z). Consider the action of Aut(X) on
Ω1 given by T ⋅ ω = (T −1 )∗ ω. Locally T ⋅ ω = f (h(z))h′ (z)dz. We have then
Proposition 5.13. 1. For T1 , T2 ∈ Aut(X) and ω ∈ Ω1 , (T1 ) ⋅ (T2 ⋅ ω) = (T1 ○ T2 ) ⋅ ω.
2. The induced map T ∶ Ω1 → Ω1 is a C-linear isomorphism.
Theorem 5.14. The representation of Aut(X) on GL(Ω1 ) is faithful.

Proof. Let T ∈ Aut(X) not the identity. Assume that X is not hyperelliptic, then there is a
Weierstrass point x ∈ X such that T (x) ≠ x. Then consider a non-zero ω ∈ Ω1 such that ordx (ω) ≥ g.
If T ⋅ ω = ω, then ω has at least g zeros at x and g zeros at T (x); thus ω has at least 2g > (2g − 2)
zeros. That is, ω = 0, which is a contradiction. So T can not induce the identity map on Ω1 .
Now assume that X is hyperelliptic. It can be shown that X can be realized locally as

w2 = (z − x1 )⋯(z − x2g−2 )
j−1
where x1 , . . . , x2g−2 are distinct complex numbers. A basis for Ω1 is given by the forms ωj = z w dz
for j = 1, . . . , g. Now let T ∈ Aut(X), if T ≠ id, J the proof for non-hyperelliptic Riemann surfaces is
still valid. Now assume T = J, then J is represented by the involution (z, w) ↦ (z, −w) and clearly
it does not acts as the identity on the forms ωj .

The trace of the matrix representing an automorphism T in GL(V ) can be explictly described, the
result is known as the Eichler trace formula and it depends on the fixed points of T . (For full
reference on the details of the statement and its proof see [Farkas and Kra, 1992, Ch V.2])
A nice consequence of the Eichler trace formula is the Lefschetz fixed point formula.
Proposition 5.15. Let t be the number of fixed points of T . Then

Tr(T ) + Tr(T ) = t

Notice that the matrix T + T̄ is the representation of the automorphism into the space of Harmonic
differentials of X.

6 Final Remarks
In this final section we include the Klein quartic that was the first example of a Riemann surface
that attains the Hurwitz bound. In fact, this surface has been widely studied and has helped in the
development of further studies in several branches of mathematics. This curve is fascinating since
it can be constructed as a tiled platonic hyperbolic polygon and its automorphism group can be
computed just by visual arguments as Klein showed in his first construction. The Klein quartic has
books, websites, videos and even a real sculpture just for itself. A nice reference for this section and
an overview of this Riemann surface see [Levy, 2001]

12
Consider the Projective curve X in P2 given by the equation

X 3 Y + Y 3 Z + Z 3 X = 0.

From the degree-genus formula we get g = (d−2)(d−1)/2 = 3. And so from the Hurwitz automorphism
theorem we get ∣ Aut(X)∣ ≤ 84(g − 1) = 168. Now we will describe explicitly the automorphism of X.
Let ξ be a primitive 7th-root of unit and f ∶ P2 → P2 be the map given by f ([x ∶ y ∶ z]) = [ξx, ξ 2 y, ξ 4 z].
Clearly f (X) = X and f defines an automorphism of order 7.
Now the cyclic permutation g([x ∶ y ∶ z]) = [z ∶ x ∶ y]) defines an automorphism of order 3 of X.
Finally, consider the map h induced by the matrix

ξ − ξ6 ξ2 − ξ5 ξ4 − ξ3⎞
i ⎛ 2
√ ⎜ξ − ξ 5 ξ4 − ξ3 ξ − ξ6 ⎟
7 ⎝ξ 4 − ξ 2 ξ − ξ6 ξ2 − ξ5⎠

it is straightforward to check that that h defines an automorphism of order 2. Now notice that
g ○ f ○ g −1 = f 4 and h ○ g ○ h−1 = g 2 and consider the subgroup G of Aut(X) generated by f, g, h.
We can check that the elements 49 elements of the form f m ○ h ○ f n are all distinct and thus < f >
can not be normal in G. Notice that 2 ∗ 3 ∗ 7 = 42 divides ∣B∣, we get that ∣G∣ = 42, 84, 126 or 168.
From the Sylow’s theorem, we can see that in the first three cases < f > is a normal subgroup of G;
therefore ∣G∣ = 168 and thus G attains the maximum possible order for a genus 3 surface.
Also, this curve can be realized as the quotient H/Γ where Γ is a discrete subgroup of Aut(H).
Similar to the case of tori, which is constructed as identifying the fundamental region given by the
lattice (and usually it is a polygon), a fundamental region for the Klein quartic has the form

Figure 1: Fundamental region for the Klein Quartic

Gluing together the same labeled edges and vertices, we obtain a representation of the genus 3
surface in R3 .

13
 Figure 2: The Klein quartic in the space

Remark:
The figures are Copyright ©Greg Egan, http://www.gregegan.net/SCIENCE/KleinQuarticKleinQuartic.html.
We finish this document with the following table, which show the genus where there exist a sur-
face (and even how many up to biholomorphism) that realizes the maximum possible order of the
automorphism group.

Group Genus Number of surfaces

PSL(2,7) 3 1
PSL(2,8) 7 1
PSL(2,13) 14 3
PSL(2,27) 118 1
PSL(2,29) 146 3
PSL(2,41) 411 3
PSL(2,43) 474 3
J1 2091 7
PSL(2,71) 2131 3
PSL(2,83) 3404 3
PSL(2,97) 5433 3
J2 7201 5
PSL(2,113) 8589 3
PSL(2,125) 11626 1

Table 1: Simple Hurwitz groups of order less than 106

Here P SL(2, m) denotes the group P SL(2, Z/mZ) and J1 , J2 denote the first two Janko Groups.

14
References
[Accola, 1968] Accola, R. D. (1968). On the number of automorphisms of a closed riemann surface.
Transactions of the American Mathematical Society, pages 398–408.
[Farkas and Kra, 1992] Farkas, H. M. and Kra, I. (1992). Riemann surfaces. In Riemann surfaces,
pages 9–31. Springer.
[Forster, 1991] Forster, O. (1991). Lectures on riemann surfaces, volume 81 of graduate texts in
mathematics.
[Hurwitz, 1892] Hurwitz, A. (1892). Über algebraische gebilde mit eindeutigen transformationen in
sich. Mathematische Annalen, 41(3):403–442.
[Klein, 1878] Klein, F. (1878). Ueber die transformation siebenter ordnung der elliptischen functio-
nen. Mathematische Annalen, 14(3):428–471.
[Levy, 2001] Levy, S. (2001). The eightfold way: the beauty of Klein’s quartic curve, volume 35.
Cambridge University Press.
[Macbeath, 1961] Macbeath, A. M. (1961). On a theorem of hurwitz. Glasgow Mathematical Journal,
5(2):90–96.
[Maclachlan, 1969] Maclachlan, C. (1969). A bound for the number of automorphisms of a compact
riemann surface. Journal of the London Mathematical Society, 1(1):265–272.
[Poincaré, 1882] Poincaré, H. (1882). Théorie des groupes fuchsiens. Acta mathematica, 1(1):1–62.
[Siegel, 1945] Siegel, C. L. (1945). Some remarks on discontinuous groups. Annals of mathematics,
pages 708–718.

15