HOLOMORPHIC DIFFERENTIALS OF KLEIN FOUR COVERS

FRAUKE M. BLEHER AND NICHOLAS CAMACHO

Abstract. Let k be an algebraically closed field of characteristic two, and let G be isomorphic to Z/2×Z/2.
Suppose X is a smooth projective irreducible curve over k with a faithful G-action, and assume that the
arXiv:2206.04559v1 [math.AG] 9 Jun 2022

cover X → X/G is totally ramified, in the sense that it is ramified and every branch point is totally ramified.
We study to what extent the lower ramification groups of the closed points of X determine the isomorphism
types of the indecomposable kG-modules and the multiplicities with which they occur as direct summands of
the space H0 (X, ΩX/k ) of holomorphic differentials of X over k. In the case when X/G = P1k , we completely
determine the decomposition of H0 (X, ΩX/k ) into a direct sum of indecomposable kG-modules. Moreover,
we show that the isomorphism classes of indecomposable kG-modules that actually occur as direct summands
belong to an infinite list of non-isomorphic indecomposable kG-modules that contain modules of arbitrarily
large k-dimension. In particular, our results show that [14, Theorem 6.4] is incorrect.

1. Introduction

Suppose k is an algebraically closed field, and X is a smooth projective irreducible curve over k on which
a finite group G acts faithfully on the right. Let ΩX/k be the sheaf of relative differentials of X over k, and
let H0 (X, ΩX/k ) be the space of holomorphic differentials of X over k. It is a classical problem, posed by
Hecke in [10], to determine the indecomposable kG-modules that occur as direct summands of H0 (X, ΩX/k )
together with their multiplicities. If the characteristic of k does not divide the order of G, this was solved
by Chevalley and Weil in [5].
For the remainder of the paper, we assume that the characteristic of k is a prime p dividing #G. Many
authors have studied this problem and have made some progress by either making assumptions on the group
G or the ramification of the cover X → X/G. See [18, 11, 15, 16, 13, 12, 14, 2, 7] for a sample of previous
results. We will now briefly discuss two of these articles, as they are closely related to our results.
The authors of [16] studied the case when G is an elementary abelian p-group and X/G = P1k . They made
additional assumptions on the ramification of the G-cover X → P1k that ensured that the indecomposable
kG-modules that occur as direct summands of H0 (X, ΩX/k ) belong, up to isomorphism, to a list of |G| − 1
non-isomorphic indecomposable kG-modules that have pairwise distinct k-dimensions between 1 and |G| − 1.
The authors of [14] considered the case when G is a solvable group, X/G = P1k , and the G-cover X → P1k
decomposes into a tower of Artin-Schreier extensions and/or Kummer extensions that are in so-called “global
standard form.” Moreover, if G is abelian with Sylow p-subgroup Gp , they claimed in [14, Theorem 6.4] that
only a list of (|Gp |−1)·|G/Gp | non-isomorphic indecomposable kG-modules can occur as direct summands of
H0 (X, ΩX/k ) and that these kG-modules have k-dimensions between 1 and |Gp | − 1. However, this statement
is incorrect, as we will show below.
The goal of this paper is to study the smallest case when the set of isomorphism classes of indecomposable
kG-modules is infinite, which occurs when the characteristic of k is two and G is a Klein four group, i.e.

Department of Mathematics, University of Iowa, 14 MacLean Hall, Iowa City, IA 52242-1419, U.S.A.
E-mail address: [email protected], [email protected] .
Date: June 9, 2022.
2010 Mathematics Subject Classification. Primary 11G20; Secondary 20C20, 14H05, 14G17.
Both authors were supported in part by NSF Grant No. DMS-1801328.
1
G is isomorphic to Z/2 × Z/2. Note that the isomorphism classes of indecomposable kG-modules can still
be classified using certain one-parameter families (see §4 for a description). We assume that the Klein four
cover π : X → X/G is totally ramified, in the sense that it is ramified and every ramification point in X has
G as its inertia group. We study the following two questions:
(1) To what extent do the lower ramification groups of the closed points of X determine the pre-
cise kG-module structure of H0 (X, ΩX/k ), in the sense that they give the precise decomposition of
H0 (X, ΩX/k ) into a direct sum of indecomposable kG-modules?
(2) What can we say about the list of isomorphism classes of indecomposable kG-modules that actually
occur as direct summands of H0 (X, ΩX/k )?
Our main idea is to define infinitely many operators given by certain elements in the group ring kG (see
Notation 2.2) and to compare the filtrations defined by these on both the OU -G-sheaf π∗ ΩX/k and the kG-
module H0 (X, ΩX/k ) (see Propositions 2.4 and 2.5). We then use this relationship, which is governed by the
ramification data of the cover π, to determine information about the kG-module structure of H0 (X, ΩX/k ).
When the genus of X/G is arbitrary, we obtain restrictions on the indecomposable kG-modules and their
multiplicities that can occur as direct summands of H0 (X, ΩX/k ) (see Theorem 2.15). Moreover, we obtain
sufficient conditions that lead to the following complete answers to Questions (1) and (2) above (see Theorem
2.16 and Examples 2.18 and 2.19 for more details):

Theorem 1.1. Suppose π : X → U = X/G is a totally ramified Klein four cover. Moreover, assume that
for each branch point u ∈ U of π, there exists a tower of function fields k(U ) ⊂ k(Vu ) ⊂ k(X) such that
k(Vu ) = k(U )(yu ) and k(X) = k(Vu )(wu ) are degree 2 Artin-Schreier extensions satisfying the following two
conditions:
(i) If vu ∈ Vu lies above u and xu ∈ X lies above vu , then mu := −ordvu (yu ) and Mu := −ordxu (wu )
are positive odd integers such that mu ≤ Mu and the lower ramification groups at xu inside G have
jumps at mu and mu + 2(Mu − mu ), and
(ii) either mu = Mu and (g − 1)(yu ) and (g − 1)(wu ) lie in k × for all non-identity elements g ∈ G, or
mu < Mu and mu = 1.
Then the lower ramification groups of the closed points of X that ramify in the cover π fully determine the kG-
module structure of H0 (X, ΩX/k ). Moreover, the list of isomorphism classes of indecomposable kG-modules
that actually occur as direct summands of H0 (X, ΩX/k ) for such X is infinite.

Theorem 1.1 includes the case in [16] when p = 2 = n and extends it from U = P1k to an arbitrary smooth
projective curve U (see Example 2.17). As a consequence of Theorem 1.1 (and, more generally, Theorem
2.16), we will show in Examples 2.18 and 2.19 that [14, Theorem 6.4] is not correct.
Note that condition (i) in Theorem 1.1 is satisfied for every branch point u ∈ U (see Remark 2.7), but
condition (ii) may not be. In Theorem 2.16, we will extend condition (ii) to include one additional case,
which is a bit more technical to state but for which we can also fully determine the kG-module structure of
H0 (X, ΩX/k ) without any assumptions on the genus of X/G (see conditions (a) and (b) in Theorem 2.16). We
will show in Examples 2.18 and 2.19 that the list of isomorphism classes of indecomposable kG-modules that
occur as direct summands of H0 (X, ΩX/k ) in the situation of Theorem 1.1 (and, more generally, Theorem
2.16) for various X is infinite, and we write down the precise list. Moreover, we give examples in which
H0 (X, ΩX/k ) is a direct sum of an arbitrary finite number of non-isomorphic indecomposable kG-modules.
In the case when X/G = P1k , we obtain the following complete answers to Questions (1) and (2) above,
without any additional assumptions on the ramification behavior (see Theorem 3.7 and Example 3.9 for
more details):
2
Theorem 1.2. Suppose π : X → P1k = X/G is a totally ramified Klein four G-cover. Then the lower rami-
fication groups of the closed points of X that ramify in the cover π fully determine the kG-module structure
of H0 (X, ΩX/k ). More precisely, the isomorphism classes of the indecomposable kG-modules that actually
occur as direct summands and their multiplicities in H0 (X, ΩX/k ) can be described explicitly. Furthermore,
the list of these isomorphism classes is infinite and contains indecomposable kG-modules of arbitrarily large
finite k-dimension.

It is an interesting question whether the set of isomorphism classes of indecomposable modules that arises
in Theorem 1.2 changes when one drops the assumption that X/G is isomorphic to P1k . We will give a
sufficient criterion for this set to not change (see Remarks 2.20 and 3.10).

The paper is organized as follows. In §2, we relate filtrations of the OU -G-sheaf π∗ ΩX/k to filtrations of
the kG-module H0 (X, ΩX/k ) for arbitrary Klein four covers (see Propositions 2.4 and 2.5). We also provide
the detailed calculations needed for totally ramified Klein four covers in Propositions 2.11 and 2.13. We
use this in Theorem 2.15 to obtain restrictions on the indecomposable kG-modules and their multiplicities
that can occur as direct summands of H0 (X, ΩX/k ). In Theorem 2.16, we give sufficient conditions under
which Theorem 2.15 leads to a full determination of the kG-module structure of H0 (X, ΩX/k ). Moreover, we
show in Examples 2.18 and 2.19 that every member of the full list of possible indecomposable kG-modules
in Theorem 2.15 actually occurs as a direct summand of H0 (X, ΩX/k ) for various X. In §3, we focus on
the case when X/G = P1k . We determine in Theorem 3.7 the full kG-module structure of H0 (X, ΩX/k ) in
this case. Moreover, we show in Example 3.9 that every member of the full list of possible indecomposable
kG-modules in Theorem 3.7 actually occurs as a direct summand of H0 (X, ΩX/k ) for various X.
Part of this paper is the Ph.D. thesis of the second author under the supervision of the first (see [4]).

2. Klein four covers in characteristic two

Throughout this paper, we make the following assumptions.

Assumption 2.1. Let k be an algebraically closed field of characteristic 2, and let π : X → U be a Galois
cover of smooth projective curves over k with Galois group G isomorphic to Z/2 × Z/2. Let L = k(X) and
K = k(U ). Write L = K(y, z) where

(2.1) y 2 − y = p, z2 − z = q

for certain p, q ∈ K − k satisfying p − q 6= s2 − s for any s ∈ K. Write G = {1, σ, τ, στ } = hσ, τ i such that

(2.2) σ(y) = y, τ (y) = y + 1 and σ(z) = z + 1, τ (z) = z.

We have three intermediate fields between K and L, corresponding to three smooth projective curves V0 ,
V1 , V2 over k that are Galois covers of U of degree 2. Their function fields are given by

(2.3) k(V0 ) = K(y), k(V1 ) = K(z), k(V2 ) = K(y + z).

We have
k(V0 ) = Lhσi , k(V1 ) = Lhτ i , k(V2 ) = Lhστ i .

We need the following notation.

3
Notation 2.2. Define the following elements in the group ring kG:

f0 := 1,
f1 := σ − 1 and f1′ := τ − 1,
(a:b)
f2 = f2 := a(σ − 1) + b(τ − 1), for each closed point(a : b) ∈ P1k ,
f3 := (σ − 1)(τ − 1),
f4 := 0.

We view these elements as operators acting on k(X), on kG-modules, and on π∗ F for certain coherent
OX -G-modules F (see [2, §2] for more details). For i ∈ {0, 2, 3, 4} a superscript (i) denotes the kernel of the
action of fi , whereas a superscript (1) denotes the intersection of the kernels of the actions of f1 and f1′ .
(a:b)
Since the kernel of the action of f2 = f2 may be different for different choices of (a : b) in P1k , we often
add (a : b) to the superscript (2).

For every (a : b) ∈ P1k , we obtain a filtration

0 = kG(0) ⊂ kG(1) ⊂ kG(2),(a:b) ⊂ kG(3) ⊂ kG(4) = kG.

As (a : b) ranges over all closed points in P1k , these filtrations range over all possible composition series of
kG. Note that for different (a : b) ∈ P1k , the kG-modules kG(2),(a:b) are not isomorphic.
The main idea of this paper is to relate, for all (a : b) ∈ P1k , the filtration of the OU -G-sheaf π∗ ΩX/k given
by {(π∗ ΩX/k )(i) }4i=0 to the filtration of the kG-module H0 (X, ΩX/k ) given by {H0 (X, ΩX/k )(i) }4i=0 and to
use this relationship to determine as much as possible of the kG-module structure of H0 (X, ΩX/k ).
The following remark is important for this comparison.

Remark 2.3. By the normal basis theorem, k(X) is a free rank one module for the group ring k(U )G. Fix a
(a:b)
closed point (a : b) ∈ P1k , and define fei := fi for i ∈ {0, 3}, fe2 := f2 , and fe1 := f1 if b 6= 0 and fe1 := f1′ if
b = 0. Then, for 0 ≤ i ≤ 3, the map
k(X)(i+1)
(2.4) fei : → k(U )
k(X)(i)

given by multiplication by fei is a k(U )G-module isomorphism. In particular, if h1 , h2 , h3 , h4 ∈ k(X) are such
that, for 0 ≤ i ≤ 3, hi+1 ∈ k(X)(i+1) and fei hi+1 ∈ k × , then {h1 , . . . , hi+1 } is a k(U )-basis of k(X)(i+1) , and
ordu (fei hi+1 ) = 0 for all points u ∈ U .

The next result is proved using similar arguments as in the proof of [2, Prop. 4.1]; we will therefore omit
its proof.

Proposition 2.4. Under Assumption 2.1, fix a closed point (a : b) ∈ P1k . For 0 ≤ i ≤ 3, there exist divisors
−1
Di on U , which are uniquely determined by π∗ DX/U , such that there are isomorphisms of OU -G-modules

(π∗ ΩX/k )(i+1) /(π∗ ΩX/k )(i) ∼

= ΩU/k ⊗OU OU (Di ).
P
More precisely, writing Di = u∈U du,i u, for each u ∈ U , the coefficient du,i is given as follows: Let
h1 = h1 (u), h2 = h2 (u), h3 = h3 (u), and h4 = h4 (u) be elements of k(X) as in Remark 2.3. Then
−1
du,i = −min{ordu (λi+1 ) ; λ1 h1 + · · · + λi+1 hi+1 ∈ (π∗ DX/U )u for some λ1 , . . . , λi+1 ∈ k(U )}.

Moreover, we have the following for x ∈ X and u = π(x):

(i) If x is not a ramification point of π then du,i = 0.
4
 (ii) If x is totally ramified and if ordx (hj (u)) are pairwise distinct modulo 4 for j ∈ {1, 2, 3, 4} then
 
dx/u + ordx (hi+1 (u))
(2.5) du,i =
4
where dx/u is the different exponent of x over u.

The next result is proved using similar arguments as in the proof of [2, Lemma 4.2]; we will therefore omit
its proof.

Proposition 2.5. Under Assumption 2.1, fix a closed point (a : b) ∈ P1k , and let D0 , . . . , D3 be the divisors
from Proposition 2.4. Suppose D3 = 0 and deg(Di ) > 0 for 0 ≤ i ≤ 2. Then, for 0 ≤ i ≤ 3, there are
isomorphisms of kG-modules
H0 (X, ΩX/k )(i+1) /H0 (X, ΩX/k )(i) ∼
= H0 (U, (π∗ ΩX/k )(i+1) /(π∗ ΩX/k )(i) )
∼
= H0 (U, ΩU/k ⊗ΩU OU (Di )).

We now make the following assumption.

Assumption 2.6. The cover π : X → U in Assumption 2.1 is totally ramified, in the sense that π is ramified
and every branch point u ∈ U of π is totally ramified. In other words, if u ∈ U is a branch point of the cover
π then there exist sp,u , sq,u , sp+q,u ∈ K such that each of
ordu (p − (s2p,u − sp,u )), ordu (q − (s2q,u − sq,u )), ordu (p + q − (s2p+q,u − sp+q,u ))
is negative and odd, where p and q are as in (2.1).

This assumption enables us to find for each branch point u ∈ U , elements h1 (u), . . . , h4 (u) of k(X) as in
Remark 2.3 such that ordx (hj (u)) are pairwise distinct modulo 4 for j ∈ {1, 2, 3, 4}. Moreover, the formula
in (2.5) shows then that D3 = 0 and deg(Di ) > 0 for 0 ≤ i ≤ 2. In other words, Proposition 2.5 applies.
The following remark and lemma describe the basic cases we need to consider for each branch point u ∈ U .

Remark 2.7. Under Assumptions 2.1 and 2.6, let u ∈ U be a branch point, and let x ∈ X be the unique
point above it. Let mU,u denote the maximal ideal of the local ring OU,u . Then there exist positive odd
integers mu ≤ Mu such that the lower ramification groups at x are given as
G = Gx,0 = Gx,1 = · · · = Gx,mu > Gx,mu +1 = · · · = Gx,mu +2(Mu −mu ) > Gx,mu +2(Mu −mu )+1 = 1.
There exist sp,u , sq,u , sp+q,u ∈ K such that either
(i) mu = Mu and ordu (p−(s2p,u −sp,u )) = −mu = ordu (q−(s2q,u −sq,u )) = ordu (p+q−(s2p+q,u −sp+q,u )),
or
(ii) mu < Mu and either
(a) ordu (p − (s2p,u − sp,u )) = −mu , ordu (q − (s2q,u − sq,u )) = −Mu = ordu (p + q − (s2p+q,u − sp+q,u )),
or
(b) ordu (q − (s2q,u − sq,u )) = −mu , ordu (p − (s2p,u − sp,u )) = −Mu = ordu (p + q − (s2p+q,u − sp+q,u )),
or
(c) ordu (p + q − (s2p+q,u − sp+q,u )) = −mu , ordu (p − (s2p,u − sp,u )) = −Mu = ordu (q − (s2q,u − sq,u )).
In the cases (i) or (ii)(a), define yu := y − sp,u , pu := p− (s2p,u − sp,u ) and zu := z − sq,u , qu := q − (s2q,u − sq,u ),
in the case (ii)(b), define yu := z − sq,u , pu := q − (s2q,u − sq,u ) and zu := y − sp,u , qu := p − (s2p,u − sp,u ),
and in the case (ii)(c), define yu := y + z − sp+q,u , pu := p + q − (s2p+q,u − sp+q,u ) and zu := z − sq,u ,
qu := q − (s2q,u − sq,u ). In particular, this means
ordx (yu ) = −2mu , ordx (zu ) = −2Mu .
5
By [19, Corollary 3.10 and the proof of Theorem 3.11], there exist αu , βu ∈ K such that wu = zu + αu + βu yu
is an Artin-Schreier generator with the following properties:

ordx (wu ) = −mu − 2(Mu − mu ), ordx (βu ) = −2(Mu − mu ).

Moreover,
(σ − 1)(wu ) = 1, (τ − 1)(wu ) = βu in the situation of (i) or (ii)(a),
(τ − 1)(wu ) = 1, (σ − 1)(wu ) = βu in the situation of (ii)(b), and
(σ ◦ τ − 1)(wu ) = 1, (τ − 1)(wu ) = βu in the situation of (ii)(c).

Lemma 2.8. Assume the notation from Remark 2.7. Let πu be a uniformizer at u and write αu , βu , pu , qu
d(α ) P −(Mu −mu )/2 P
as Laurent series in k[[πu , πu−1 ]] as follows: αu = πu u i
i≥0 au,i πu , βu = πu
i
i≥0 bu,i πu , pu =
P P
πu−mu i≥0 pu,i πui , qu = πu−Mu i≥0 qu,i πui , where bu,0 , pu,0 , qu,0 ∈ k × . Then d(αu ) ≥ (−Mu + 1)/2,
X  
(2.6) qu,2j + pu,2i1 b2u,i2 = 0 for 0 ≤ j ≤ m4u ,
i1 +i2 =j

and bu,0 ∈ k − {0, 1} when mu = Mu . Moreover, for d(αu ) = (−Mu + 1)/2, we obtain
X    
(2.7) a2u,j = qu,2j+1 + pu,2i1 +1 b2u,i2 for 0 ≤ j ≤ 2m4u +3 − mu4+3 − 1.
i1 +i2 =j

Define
 
−⌊ mu4+3 ⌋−1
⌊ 2mu4 +3 ⌋X
 
(2.8) eu := πu(−Mu +1)/2 
α au,i πui 
i=0

and, for j ≥ 0, define

j
!
X
(2.9) βu (j) := πu−(Mu −mu )/2 bu,i πui and
i=0
(2.10) wu (j) := zu + α
eu + βu (j) yu .

Then
 mu 
(2.11) ordx (wu (j)) ≥ −2Mu + 4j + 4 for 0 ≤ j ≤ 4 − 1,

and
 mu 
(2.12) ordx (wu (j)) = −mu − 2(Mu − mu ) for j ≥ 4 .

Proof. We have

wu2 − wu = zu2 − zu + βu2 (yu2 − yu ) + α2u − αu + (βu2 − βu )yu

= qu + βu2 pu + α2u − αu + (βu2 − βu )yu .

Let V ∈ {V0 , V1 , V2 } be the degree two cover of U such that yu ∈ k(V ), and let v ∈ V be below x and above
u. Note that
ordx (wu ) = ordv (wu2 − wu ),
ordv (qu + βu2 pu ) ≥ −2Mu is even, ordv (α2u − αu ) is either negative and divisible by 4 or non-negative
and even, and ordv (yu ) = −mu . Hence, to obtain ordv (wu2 − wu ) = −mu − 2(Mu − mu ), we must have
ordv (βu2 − βu ) = −2(Mu − mu ), in addition to ordv (qu + βu2 pu + (α2u − αu )) ≥ −mu − 2(Mu − mu ) + 1. The
6
former condition implies that bu,0 6= 1 in the case when mu = Mu . To analyze the latter condition, we use
the Laurent series expansions of αu , βu , pu , qu in k[[πu , πu−1 ]]. We obtain
!
X X
qu + βu2 pu = qu,i + pu,i1 b2u,i2 πu−Mu +i
i≥0 i1 +2i2 =i
 
⌊ m4u ⌋
X X
= qu,2j + pu,2i1 b2u,i2  πu−Mu +2j
j=0 i1 +i2 =j
 
⌊ m4u
X⌋ X
+ qu,2j+1 + pu,2i1 +1 b2u,i2  πu−Mu +2j+1
j=0 i1 +i2 =j
!
X X
+ qu,i + pu,i1 b2u,i2 πu−Mu +i
i≥2⌊ m4u ⌋+2 i1 +2i2 =i

where
−Mu +2⌊ m4u ⌋+2 m 
ordv (πu ) = −2Mu + 4 u
4 + 4 ≥ −2Mu + (mu − 3) + 4 = −mu − 2(Mu − mu ) + 1.

For 0 ≤ j ≤ ⌊ m4u ⌋, we have that ordv (πu−Mu +2j ) = −2Mu + 4j is not divisible by 4 and ordv (πu−Mu +2j+1 ) =
−2Mu + 4j + 2 is divisible by 4. Considering j = 0, we see in particular that d(αu ) ≥ (−Mu + 1)/2. Since
(−M +1)/2
ordv (πu u ) = −Mu + 1 ≥ −mu − 2(Mu − mu ) + 1, we moreover obtain equations (2.6) and (2.7). Note
that if mu ≡ 3 mod 4 then ⌊ 2m4u +3 ⌋ − ⌊ mu4+3 ⌋ − 1 = ⌊ m4u ⌋. On the other hand, if mu ≡ 1 mod 4 then
−Mu +2⌊ m4u ⌋+1 m 
ordv (πu ) = −2Mu + 4 4
u
+ 2 = −2Mu + (mu − 1) + 2 = −mu − 2(Mu − mu ) + 1

which means that equation (2.7) only follows for 0 ≤ j ≤ ⌊ m4u ⌋ − 1 = ⌊ 2m4u +3 ⌋ − ⌊ mu4+3 ⌋ − 1.
Fix now j ≥ 0 and consider αeu , βu (j) and wu (j) as in (2.8), (2.9) and (2.10). We have

wu (j)2 − wu (j) = qu + βu (j)2 pu + α

e2u − α
eu + (βu (j)2 − βu (j))yu .

Since bu,0 6∈ {0, 1} when mu = Mu , we have ordv (βu (j)2 − βu (j)) = −2(Mu − mu ). Moreover, equations
(2.6) and (2.7) show that for j ≤ ⌊ m4u ⌋ − 1,

ordv (qu + βu (j)2 pu + α

e2u − α
eu ) ≥ ordv (πu−Mu +2j+2 ) = −2Mu + 4j + 4.

Since (
 mu 
−mu − 2(Mu − mu ) − 1 if mu ≡ 1 mod 4,
−2Mu + 4 4 −1 +4=
−mu − 2(Mu − mu ) − 3 if mu ≡ 3 mod 4,
we obtain equations (2.11) and (2.12). 

The following equation gives a useful connection between the upper bounds of j in (2.6) and (2.7):
    (  mu 
2mu + 3 mu + 3 if mu ≡ 1 mod 4, and
(2.13) − =  m4u 
4 4 4 + 1 if mu ≡ 3 mod 4.

Remark 2.9. Assume the notation from Remark 2.7 and Lemma 2.8. Using (2.9) and (2.10), define
 
 

(2.14) βeu := βu m4u and w eu := wu m4u .

We have
eu ) = −mu − 2(Mu − mu )
ordx (w and ordx (βeu ) = −2(Mu − mu ).
7
Moreover,
(σ − 1)(weu ) = 1, eu ) = βeu
(τ − 1)(w in the situation of (i) or (ii)(a),
(τ − 1)(weu ) = 1, eu ) = βeu
(σ − 1)(w in the situation of (ii)(b), and
(σ ◦ τ − 1)(w eu ) = βeu
eu ) = 1, (τ − 1)(w in the situation of (ii)(c).

In other words, we can replace βu by βeu and wu by w

eu .

We use the following additional notation, based on Remark 2.7 and Lemma 2.8.

Notation 2.10. Assume the notation from Remark 2.7 and Lemma 2.8.
(i) If we are in the situation of Remark 2.7(i), we define λu := bu,0 . Additionally, if 1 ≤ ordu (βu − λu ) ≤
 mu 
4 , we define δu := ord (β − λu ), and otherwise we define δu := 0. In particular, λu ∈ k − {0, 1}
 u u
and δu ∈ {0, 1, . . . , m4u }. Moreover, if δu = 0 then βeu = λu and weu = zu + αeu + λu yu .
(ii) If we are in the situation of Remark 2.7(ii), we define δu := −1. Moreover, we define


 ∞ if we are in the situation of Remark 2.7(ii)(a),
λu := 0 if we are in the situation of Remark 2.7(ii)(b),

 1 if we are in the situation of Remark 2.7(ii)(c).

For a branch point u ∈ U and a point (a : b) ∈ P1k , we define



 (a : b) if u is as in Remark 2.7(i) or 2.7(ii)(a),
(2.15) (c : d) := (b : a) if u is as in Remark 2.7(ii)(b),

 (a : a + b) if u is as in Remark 2.7(ii)(c).

Proposition 2.11. Under Assumptions 2.1 and 2.6, let u ∈ U be a branch point, and let x ∈ X be a point
above u. Fix a point (a : b) ∈ P1k and let (c : d) be as in (2.15). Let βeu and w
eu be as in (2.14), and define

h1 (u) := 1,
( ) h2 (u) := (c + dβeu )yu + dw
eu ,
yu if δu = 0 and c = dλu
h3 (u) := 1 , h4 (u) := w
eu yu .
c+dβeu w
eu otherwise

Then h1 = h1 (u), h2 = h2 (u), h3 = h3 (u), and h4 = h4 (u) are elements of k(X) as in Remark 2.3,
and ordx (hj (u)) are pairwise distinct modulo 4 for j ∈ {1, 2, 3, 4}. Moreover, the divisors D0 , . . . , D3 from
Proposition 2.4 satisfy D3 = 0 and deg(Di ) > 0 for 0 ≤ i ≤ 2. More precisely, for 0 ≤ i ≤ 2, the coefficient
du,i of u in Di is given as follows:
 
3mu + 3 M u − mu
du,0 = + ,
4 2
  2m +3 


u
4 if δu = 0 and c = dλu ,
  mu +3  + δ

if δu > 0 and c = dλu ,
du,1 =  mu4+3  Muu −mu

 + if δu = −1 and d = 0,

  mu4+3  2

4 otherwise,
  m +3 


u
4 if δu = 0 and c = dλu ,
  2mu +3  − δ

if δu > 0 and c = dλu ,
u
du,2 =  2m4u +3 

 if δu = −1 and d = 0,

  2m4u +3  Mu −mu
4 + 2 otherwise.
8
Proof. A straightforward computation shows that fej hj = 0 for 1 ≤ j ≤ 4. We have
(
b if b 6= 0,
fe0 h1 = 1, fe1 h2 =
a if b = 0,
(
d if δu = 0 and c = dλu ,
fe2 h3 = fe3 h4 = 1,
1 otherwise,

which implies that fei hi+1 ∈ k × for 0 ≤ i ≤ 3. Moreover, we have

ordx (h1 ) = 0,


 −mu − 2(Mu − mu ) if δu = 0 and c = dλu ,

 −2m + 4δ − 2(M − m )
u u u u if δu > 0 and c = dλu ,
ordx (h2 ) =
 −2mu
 if δu = −1 and d = 0,


−2mu − 2(Mu − mu ) otherwise,


 −2mu − 2(Mu − mu ) if δu = 0 and c = dλu ,

 −m − 4δ − 2(M − m )
u u u u if δu > 0 and c = dλu ,
ordx (h3 ) =

 −m u − 2(M u − m u ) if δu = −1 and d = 0,


−mu otherwise,
ordx (h4 ) = −3mu − 2(Mu − mu ).

Since Mu − mu is even, this shows that ordx (hj ) are pairwise distinct modulo 4 for j ∈ {1, 2, 3, 4}. Using

dx/u = 3(mu + 1) + 2(Mu − mu ),

the formulas for du,i , 0 ≤ i ≤ 3, given in Proposition 2.11 now follow from Proposition 2.4(ii). In particular,
we obtain D3 = 0 and deg(Di ) > 0 for 0 ≤ i ≤ 2. 

Notation 2.12. Define Ubr ⊂ U to be the set of branch points of the cover π : X → U , and define
Λbr := {λu ; u ∈ Ubr }. We organize Ubr into 3 disjoint subsets:

B1 := {u ∈ Ubr ; δu = 0},
B2 := {u ∈ Ubr ; δu ≥ 1},
B3 := {u ∈ Ubr ; δu = −1}.

In other words, u ∈ B1 ∪ B2 is as in the situation of Remark 2.7 (i), and u ∈ B3 is as in the situation of
Remark 2.7 (ii). Moreover, for i ∈ {1, 2, 3} and λ ∈ k ∪ {∞}, define

Bi,λ := {u ∈ Bi ; λu = λ}.

The next result is a direct consequence of Propositions 2.5 and 2.11.

Proposition 2.13. Under Assumptions 2.1 and 2.6, fix a point (a : b) ∈ P1k , and let M = H0 (X, ΩX/k ). For
0 ≤ i ≤ 3, the non-negative integer

ri := dimk (M (i+1) /M (i) ) − g(U ) + 1

9
is given as follows:

X  3mu + 3  Mu − mu 
r0 = + ,
4 2
u∈Ubr

 P  2mu +3  P  

 mu +3
 u∈B1,λ +
u∈B2,λ + δu

 P
4
 mu +3  4
r1 = + u∈Ubr ,λu 6=λ 4 if λ = a/b 6∈ {0, 1, ∞},

  mu +3 
 mu +3 
 P

+ Mu −mu
+
P
if λ = a/b ∈ {0, 1, ∞},
u∈B3,λ 4 2 u∈Ubr −B3,λ 4

 P  mu +3  P  2mu +3 


 u∈B1,λ u∈B2,λ + − δu

 P
4
 2mu +3  4


Mu −mu
r2 = + u∈Ubr ,λu 6=λ 4 + 2 if λ = a/b 6∈ {0, 1, ∞},

  2mu +3   2mu +3 

 P

+
P
+ Mu −mu
if λ = a/b ∈ {0, 1, ∞},
u∈B3,λ 4 u∈Ubr −B3,λ 4 2

r3 = 1.

We now use Proposition 2.13 to determine as much as possible about the indecomposable direct kG-module
summands of H0 (X, ΩX/k ). We make the following definition, using Notation 4.1 for the indecomposable
kG-modules (see also Remark 4.2).

Definition 2.14. Let M = H0 (X, ΩX/k ) and write

ε1
M ε2
M ε3
M
M= N2ℓh ,λh ⊕ M2mi +1,1 ⊕ M2nj +1,2 ⊕ k ⊕ε4 ⊕ kG⊕ε5
h=1 i=1 j=1

where ε1 , ε2 , ε3 , ε4 , ε5 ∈ Z+ ∪ {0}, ℓh , mi , nj ∈ Z+ , λh ∈ k ∪ {∞} for 1 ≤ h ≤ ε1 , 1 ≤ i ≤ ε2 , 1 ≤ j ≤ ε3 .

Define ΛM := {λh ; 1 ≤ h ≤ ε1 }, and for λ ∈ ΛM , define ε1,λ := #{h ; 1 ≤ h ≤ ε1 , λh = λ}.

Theorem 2.15. Under Assumptions 2.1 and 2.6, let M = H0 (X, ΩX/k ) be as in Definition 2.14. Then
(i) ΛM ⊆ Λbr and, if λ ∈ ΛM , then
 P  2mu +3   mu +3 
P
 − + u∈B2,λ δu if λ 6∈ {0, 1, ∞},
u∈B1,λ 4 4
ε1,λ = P
 Mu −mu
if λ ∈ {0, 1, ∞},
u∈B3,λ 2

!
X  mu + 3 
(ii) ε2 = − 1,
4
u∈Ubr
X  3mu + 3   2mu + 3 
(iii) ε3 + ε4 = − ,
4 4
u∈Ubr
(iv) ε5 = g(U ).
Suppose there exists λ ∈ Λbr − ΛM . Then B2,λ ∪ B3,λ = ∅, and the points u ∈ B1,λ must all satisfy mu = 1.

Proof. Let M = H0 (X, ΩX/k ) be as in Definition 2.14, and fix (a : b) ∈ P1k . We see from Proposition 2.13
and the second row of Table 4.1 that

1 = r3 = dimk (M (4) /M (3) ) − g(U ) + 1 = ε5 − g(U ) + 1

10
which shows part (iv) of Theorem 2.15. Using this and the remainder of Table 4.1, we obtain the following
values of ri = dimk (M (i+1) /M (i) ) − ǫ5 + 1 for 0 ≤ i ≤ 2:
ε1
X ε2
X ε3
X
r0 = ℓh + mi + (nj + 1) + ε4 + 1,
h=1 i=1 j=1

r1 = ε1,λ + ε2 + 1,
ε1
X ε2
X ε3
X
r2 = −ε1,λ + ℓh + mi + nj + 1,
h=1 i=1 j=1

where λ = a/b ∈ k ∪ {∞}.

If λ = a/b does not lie in Λbr ∪ ΛM then the equations for r0 − r2 above and in Proposition 2.13 show
part (iii) of Theorem 2.15.
If λ = a/b does not lie in Λbr ∪ ΛM then the equations for r1 above and in Proposition 2.13 show
X  mu + 3 
(2.16) ε2 + 1 =
4
u∈Ubr

proving part (ii) of Theorem 2.15.

Using (2.16) and the equation for r1 in Proposition 2.13, we then obtain for λ = a/b 6∈ {0, 1, ∞} that
X  2mu + 3   mu + 3  X
(2.17) ε1,λ = − + δu .
4 4
u∈B1,λ u∈B2,λ

If λ = a/b ∈ {0, 1, ∞} then (2.16) and the equation for r1 in Proposition 2.13 imply that
X M u − mu
(2.18) ε1,λ = .
2
u∈B3,λ

In particular, (2.17) and (2.18) imply that ΛM ⊆ Λbr . Moreover, if λ ∈ Λbr − ΛM , then we see that
B2,λ ∪ B3,λ = ∅, and the points u ∈ B1,λ must all satisfy mu = 1. This proves part (i) and the last statement
of Theorem 2.15. 

Theorem 2.16. Under Assumptions 2.1 and 2.6, let M = H0 (X, ΩX/k ) be as in Definition 2.14, and suppose
that for all u ∈ Ubr , one of the following conditions holds:
(a) u ∈ B1 , or
   
(b) u ∈ B2 and δu = 2m4u +3 − mu4+3 , or
(c) u ∈ B3,λ , for λ ∈ {0, 1, ∞}, and mu = 1.
Then ℓh = 1 = mi for all 1 ≤ h ≤ ε1 , 1 ≤ i ≤ ε2 , and ε3 = 0. We obtain
M ⊕a
M∼ = N2,λ λ ⊕ M3,1⊕b
⊕ k ⊕c ⊕ kG⊕d
λ∈Λbr

where
X  2m  m 

u +3 u +3 Mu −mu
aλ = 4 − 4 + 2 ,
u∈Ubr ,λu =λ
P  mu +3 
P    

b= u∈Ubr 4 − 1, c = u∈Ubr 3m4u +3 − 2m4u +3 , and d = g(U ). Moreover, if u ∈ Ubr satisfies
 
condition (b) above then we must have mu ≡ 1 mod 4 and δu = m4u .

Proof. Suppose λ = a/b 6∈ Λbr . By Proposition 2.13, we obtain

X  2mu + 3   mu + 3  Mu − mu 
(2.19) r2 − r1 = − + .
4 4 2
u∈Ubr
11
By Theorem 2.15 and the assumptions (a)-(c) in Theorem 2.16, it follows that the right side of (2.19) equals
ε1 . Using the equations for r1 , r2 from the proof of Theorem 2.15, we see that
ε1
X ε2
X ε3
X
(2.20) ℓh + (mi − 1) + nj = r2 − r1 .
h=1 i=1 j=1

Hence we obtain that

ε1
X ε2
X ε3
X
(ℓh − 1) + (mi − 1) + nj = 0.
h=1 i=1 j=1
This implies ℓh = 1 = mi for all 1 ≤ h ≤ ε1 , 1 ≤ i ≤ ε2 , and ε3 = 0. In case (b), equation (2.13) and the
   
inequality δu ≤ m4u shows that we must have mu ≡ 1 mod 4 and δu = m4u . This completes the proof of
Theorem 2.16. 

Example 2.17. Suppose the field k(X) is generated over k(U ) by an element T ∈ k(X) such that T 4 −T = ω
for an element ω ∈ k(U ) satisfying the following property: For each u ∈ U , there exists t(u) ∈ k(U ) such
that ordu (ω − (t(u)4 − t(u))) is either (i) non-negative or (ii) negative and odd. Moreover, assume that (ii)
occurs for at least one point u ∈ U .
This case was studied in [4, Section 3] and the precise decomposition of H0 (X, ΩX/k ) into a direct sum
of indecomposable kG-modules was determined. In particular, this case generalizes the main result of [16]
when p = 2 = n from U = P1k to an arbitrary smooth projective curve U .
We claim that this case implies that Ubr = B1 and Λbr = {α} where hαi = F× 4 , so that it is a subcase
of Theorem 2.16. To see this, let y = T − T and z = αT − α T . Then y − y = ω and z 2 − z = α2 ω.
2 2 2 2

Therefore, if u ∈ U is such that ordu (ω − (t(u)4 − t(u))) = −mu is negative and odd, then sp,u := t(u)2 − t(u),
sq,u := αt(u)2 − α2 t(u) and sp+q,u := α2 t(u)2 − αt(u) satisfy
ordu (ω − (s2p,u − sp,u )) = −mu = ordu (α2 ω − (s2q,u − sp,u )) = ordu (αω − (s2p+q,u − sp+q,u )).
In other words, letting p = ω, q = α2 ω, we see that the cover π : X → U satisfies Assumption 2.6 (see
also [17, Proposition 3.7.10]). Moreover, if u ∈ U is a branch point and x ∈ X lies above it, then, letting
yu = y − sp,u and zu = z − sq,u , we see that wu = zu + αyu satisfies ordx (wu ) = −mu . Therefore, we obtain
Ubr = B1 and Λbr = {α}. By Theorem 2.16, it follows that
H0 (X, ΩX/k ) ∼ ⊕a
= N2,α ⊕ M3,1⊕b
⊕ k ⊕c ⊕ kG⊕d
P  2m +3   m +3 
P  m +3 
P    

where a = u∈Ubr
u
4 − u
4 ,b= u∈Ubr
u
4 − 1, c = u∈Ubr 3m4u +3 − 2m4u +3 , and
d = g(U ).

The next two examples (Examples 2.18 and 2.19) show that the list of isomorphism classes of indecom-
posable kG-modules that actually occur, for various X, as direct summands of H0 (X, ΩX/k ) in the situation
of Theorem 2.16 is infinite and given as follows:
(2.21) {N2,λ ; λ ∈ k ∪ {∞}} ∪ {M3,1 , k, kG} .
Moreover, we use these examples to show that the statement of [14, Theorem 6.4] is incorrect.
Notice that kG will occur with multiplicity at least 1 as soon as the genus of U is bigger than 0. For
this reason, we restrict ourselves to examples when U = P1k to illustrate that the other indecomposable
kG-modules in the set (2.21) all do occur as direct summands of H0 (X, ΩX/k ) for various X.

Example 2.18. Let U = P1k with function field k(U ) = k(t), and let λ ∈ k − {0, 1}. We first provide
an example in which H0 (X, ΩX/k ) contains simultaneously as direct summands the 4 non-isomorphic 2-
dimensional kG-modules N2,∞ , N2,0 , N2,1 and N2,λ . Moreover, we will show that k(X) is a global standard
function field, in the sense of [14, Definitions 2.1 and 3.1]. Hence, this example contradicts [14, Theorem 6.4]
12
which states that at most 3 non-isomorphic indecomposable kG-modules that have k-dimensions 1, 2 and 3,
respectively, should occur as direct summands of H0 (X, ΩX/k ).
Let α ∈ k − {λ} satisfy the equation α2 + α + 1 = 0, i.e. F4 = {0, 1, α, α2 }. Let p and q in (2.1) be given
by
1
p = ,
t(t − 1)3 (t − α)3 (t − λ)3
α
q = ,
t3 (t − 1)(t − α)3 (t − λ)3
which implies
α2
p+q = .
t3 (t − 1)3 (t − α)(t − λ)3
For µ ∈ k ∪ {∞}, let uµ be the corresponding closed point in P1k with uniformizer πuµ = t − µ for µ ∈ k and
πu∞ = t−1 . Then
orduµ (p), orduµ (q), orduµ (p + q) ∈ {−1, −3} if µ ∈ {0, 1, α, λ},
and orduµ (p), orduµ (q), orduµ (p + q) are non-negative for all µ ∈ k − {0, 1, α, λ}. In particular, Ubr =
{u0 , u1 , uα , uλ }, and muµ = 1, Muµ = 3 for µ ∈ {0, 1, α}, and muλ = Muλ = 3. Define w := z + α2 t−1t y.
Then  
2 (t − 1)2 2 t−1 (t − 1)(t − α)
w −w = α 2
−α y= y.
t t t2
Hence, if xµ ∈ X lies above uµ , for µ ∈ k ∪ {∞}, we obtain


 −5 if µ = 0,
ordxµ (w) = −3 if µ = λ,

 −1 if µ ∈ {1, α},

and ordxµ (w) is non-negative for all µ ∈ k − {0, 1, α, λ}. In other words, k(X)/k(t) is a global standard
function field, in the sense of [14, Definitions 2.1 and 3.1]. By Notations 2.10 and 2.12, it follows that
u0 ∈ B3,∞ , u1 ∈ B3,0 , uα ∈ B3,1 , and uλ ∈ B2,λ .
Since muµ = 1 for µ ∈ {0, 1, α}, we have that u0 , u1 , uα satisfy condition (c) of Theorem 2.16, whereas uλ
satisfies condition (a) of this theorem. Hence this theorem shows that
H0 (X, ΩX/k ) ∼ ⊕3
= N2,∞ ⊕ N2,0 ⊕ N2,1 ⊕ N2,λ ⊕ M3,1 ⊕ k
showing that there are 6 non-isomorphic indecomposable kG-modules that occur as direct summands of
H0 (X, ΩX/k ), including 4 non-isomorphic indecomposable kG-modules of k-dimension 2.

Example 2.19. Let U = P1k with function field k(U ) = k(t). Given a positive integer n, we next provide an
example in which there are n+2 non-isomorphic indecomposable kG-modules that occur as direct summands
of H0 (X, ΩX/k ). Moreover, we will show that k(X) is a global standard function field, in the sense of [14,
Definitions 2.1 and 3.1]. As in Example 2.18, we obtain a contradiction to [14, Theorem 6.4] provided n ≥ 2.
Let λ1 , . . . , λn ∈ k − {0, 1} be n distinct elements, and let p and q in (2.1) be given by
1
p = ,
(t − λ1 )5 · · · (t − λn )5
t2
q = .
(t − λ1 )5 · · · (t − λn )5
For 1 ≤ i ≤ n, let ui be the point in A1k ⊂ P1k corresponding to λi with uniformizer πui = t − λi . Then
ordui (p) = −5 = ordui (q) = ordui (p + q).
13
This means that Ubr = {u1 , . . . , un }, and mui = 5 = Mui for all 1 ≤ i ≤ n. Define w := z + t y. Then

w2 − w = t(t − 1)y.

Since none of the ui correspond to either 0 or 1 in k, we obtain, for 1 ≤ i ≤ n, that ordxi (w) = −5 when
xi ∈ X lies above ui . In other words, k(X)/k(t) is a global standard function field, in the sense of [14,
Definitions 2.1 and 3.1].
Let i ∈ {1, . . . , n}. Using Remark
j k itj followsk that βui = λi + πui . Therefore, in Notation 2.10, we have
2.7,
2mui +3 mui +3
λui = λi and δui = 1. Since 4 − 4 = 1, all ui satisfy condition (b) of Theorem 2.16. Hence
this theorem shows that
⊕(2n−1)
H0 (X, ΩX/k ) ∼
= N2,λ1 ⊕ · · · ⊕ N2,λn ⊕ M3,1 ⊕ k ⊕n .

In particular, there are n + 2 non-isomorphic indecomposable kG-modules that occur as direct summands of
H0 (X, ΩX/k ), including n non-isomorphic indecomposable kG-modules of k-dimension 2.

Theorem 1.1 follows from Remark 2.7, Theorem 2.16 and Example 2.18. Note that u ∈ Ubr satisfies
condition (ii) in Theorem 1.1 if and only if u satisfies either condition (a) or condition (c) in Theorem 2.16.

Remark 2.20. In all cases that are not covered by conditions (a)-(c) in Theorem 2.16, it is tougher to find
the actual values of ℓh , mi , nj for 1 ≤ h ≤ ε1 , 1 ≤ i ≤ ε2 , 1 ≤ j ≤ ε3 , in the notation of Definition 2.14.
However, here is a situation in which we can say more. Namely, using the notation of Theorem 2.15, suppose
ε1
X X  2m  m 

u +3 u +3 Mu −mu
(2.22) ℓh ≥ 4 − 4 + 2 .
h=1 u∈Ubr

Using the same arguments as for equations (2.19) and (2.20) in the proof of Theorem 2.16, we then obtain
that
Xε2 Xε3
(mi − 1) + nj ≤ 0
i=1 j=1

which implies that we must have mi = 1 for all 1 ≤ i ≤ ε2 , and ε3 = 0. In particular, we must then have
X  3mu + 3   2mu + 3 
ε4 = − .
4 4
u∈Ubr

In §3, we will show that if U = P1k then the inequality in (2.22) is satisfied and actually an equality, and we
will give the precise description of the kG-module structure of H0 (X, ΩX/k ) in this case (see Theorem 3.7).

3. Klein four covers of the projective line

Suppose U = P1k and π : X → P1k is a G-cover satisfying Assumptions 2.1 and 2.6. In this section, we
determine the precise kG-module structure of H0 (X, ΩX/k ).
We use the following notation.
Write the function field K = k(U ) = k(t) for a variable t. For µ ∈ k ∪ {∞}, let uµ be the corresponding
closed point in P1k with uniformizer
(
t−1 if µ = ∞,
(3.1) πuµ =
t − µ if µ 6= ∞.

For simplicity, we will write ∞ instead of u∞ in what follows. If f is any rational function in K = k(t),
then we can use its partial fraction decomposition to bring f into standard form for all the primes of k(t)
14
(as described by Hasse in [9]). More precisely, since k is algebraically closed of characteristic 2, there exists
a rational function sf ∈ K = k(t) such that
Nf (t)
(3.2) f − (s2f − sf ) = Q mf,i
i f,i (t)
D
where Df,i (t) = t − af,i for pairwise distinct af,i ∈ k, mf,i are positive odd integers, and the polynomial
Nf (t) is relatively prime to the denominator. Because of Assumption 2.6 and Remark 2.7, we can change
the variable t to be able to make the following assumptions throughout this section.

Assumption 3.1. Let U = P1k with function field K = k(U ) = k(t) = k(t−1 ). Let p, q ∈ K − k be as in
(2.1), and define

(3.3) r := p + q.

There exist n ≥ 0 distinct elements λ1 , . . . , λn ∈ k − {0} with corresponding closed points u1 , . . . , un ∈

A1k − {0} ⊂ P1k such that the following is true:
For f ∈ {p, q, r}, there exists a rational function sf in K such that
f1 (t−1 )
(3.4) f − (s2f − sf ) =
f2 (t−1 )
where f1 (t−1 ), f2 (t−1 ) are relatively prime polynomials in k[t−1 ], and there exist positive odd integers
mf,∞ , mf,u1 , . . . , mf,un such that

(3.5) f2 (t−1 ) = (t−1 )mf,∞ (t−1 − λ−1

1 )
mf,u1
· · · (t−1 − λ−1
n )
mf,un
.

Remark 3.2. Under Assumptions 2.1 and 3.1, the finite set of branch points is given as

Ubr = {∞, u1 , . . . , un }.

Moreover, if u ∈ Ubr then {mu , Mu } = {mp,u , mq,u , mr,u }, and if mu < Mu then precisely two of mp,u , mq,u ,
mr,u are equal to Mu . For f ∈ {p, q, r}, suppose df is the degree of f1 (t−1 ) in t−1 , and write

f1 (t−1 ) = cf,df + · · · + cf,1 t−df +1 + cf,0 t−df

in k[t−1 ] where cf,0 6= 0. Since t−1 is not a factor of f1 (t−1 ), we also have that cf,df 6= 0. By multiplying f
in (3.4) and (3.5) by a non-zero scalar and an appropriate power of t, we can write
mf,u1 m
cf,df tdf + · · · + cf,1 t + cf,0
f − (s2f − sf ) = tmf,∞ +mf,u1 +···+mf,un −df λ1 · · · λn f,un .
(t − λ1 )mf,u1 · · · (t − λn )mf,un
Since 0 6∈ Ubr , it follows that
n
X
(3.6) df ≤ mf,∞ + mf,ui for f ∈ {p, q, r}.
i=1

Remark 3.3. Let y, z, p, q, r, sp , sq , sr be as in Assumption 3.1, where r = p + q. Define



 ye := y − sp , pe := p − (s2p − sp ),
(3.7) ze := z − sq , qe := q − (s2q − sq ),

 y] + z := y + z − sr , re := r − (s2r − sr ).
Fix u ∈ Ubr . By Remarks 2.7 and 3.2, we obtain the following values for mu and Mu :
• If mp,u = mq,u = mr,u or mp,u < mq,u = mr,u , then mu = mp,u and Mu = mq,u = mr,u , and

yu = ye, pu = pe, zu = ze, qu = qe.

15
 • If mq,u < mp,u = mr,u , then mu = mq,u and Mu = mp,u = mr,u , and

yu = ze, pu = qe, zu = ye, qu = pe.

• If mr,u < mp,u = mq,u , then mu = mr,u and Mu = mp,u = mq,u , and

yu = y]
+ z, pu = re, zu = ze, qu = qe.

Remark 3.4. For each u ∈ U = P1k , let πu be the corresponding uniformizer, as defined in (3.1). Then the
differentials of the uniformizers are given as
(
t−2 dt if u = ∞,
(3.8) dπu =
dt if u 6= ∞.

The stalk of ΩU/k at u is given as

(ΩU/k )u = OU,u dπu = {h dt ; h ∈ k(U ), ordu (h) ≥ −ku }

where
(
−2 if u = ∞,
(3.9) ku =
0 6 ∞.
if u =

It follows from the proof of [8, Proposition IV.2.3] that there is a natural isomorphism of OX -G-modules
−1
ΩX/k = π ∗ ΩU/k ⊗OX DX/U .

In particular, we obtain that if xu ∈ X lies above u ∈ U then the stalk at xu equals

−1
(ΩX/k )xu = (π ∗ ΩU/k )xu ⊗OX,xu (DX/U )xu
= {f dt ; f ∈ k(X), ordxu (f ) ≥ −exu /u ku − dxu /u }

where exu /u is the ramification index and dxu /u is the different exponent. We have exu /u = 1 and dxu /u = 0
if u 6∈ Ubr . Moreover, we have

(3.10) exu /u = 4 and dxu /u = 3(mu + 1) + 2(Mu − mu ) for u ∈ Ubr and xu ∈ X above u.

We obtain the following natural identifications as kG-modules:

\
H0 (X, ΩX/k ) = (ΩX/k )x
x∈X
= {f dt ; f ∈ k(X), ordx (f ) ≥ −ex/π(x)kπ(x) − dx/π(x) for all x ∈ X}.

By the Riemann-Hurwitz formula, we have

1 X 1 X
(3.11) g(X) = 1 + 4(g(U ) − 1) + dxu /u = −3 + dxu /u
2 2
u∈Ubr u∈Ubr

Definition 3.5. Fix u ∈ Ubr . Let yu , zu be as in Remark 3.3, let αu , βu , λu , δu be as in Remark 2.7, let
wu (j), for j ≥ 0, be as in Lemma 2.8, and let πu , ku be as in Remark 3.4.
     
(a) Define µu,1 := mu4+3 , µu,2 := 2m4u +3 , µu,3 := 3m4u +3 , and νu := Mu −m 2
u
. Moreover define
(
0 if u = ∞,
s(u) :=
1 if u 6= ∞.
16
 (b) Define
−i1 (u)
fu,1,i1 (u) := πu , s(u) ≤ i1 (u) ≤ µu,3 + νu + ku ,
−i2 (u)
fu,2,i2 (u) := πu yu , s(u) ≤ i2 (u) ≤ µu,1 + νu + ku ,
−i (u)
fu,3,i3 (u) := πu 3 zu , s(u) ≤ i3 (u) ≤ µu,1 + ku ,
−i (u)
fu,3,i3 (u) := πu 3 wu (i3 (u) − µu,1 − ku − 1) , µu,1 + ku + 1 ≤ i3 (u) ≤ µu,2 + ku ,
and define

Bu,1 := {fu,1,i1 (u) ; s(u) ≤ i1 (u) ≤ µu,3 + νu + ku },

Bu,2 := {fu,2,i2 (u) ; s(u) ≤ i2 (u) ≤ µu,1 + νu + ku },
Bu,3 := {fu,3,i3 (u) ; s(u) ≤ i3 (u) ≤ µu,2 + ku }.

(c) For each u ∈ Ubr , define Bu = Bu,1 ∪ Bu,2 ∪ Bu,3 . Define

[
B := Bu .
u∈Ubr

Lemma 3.6. Under Assumptions 2.1 and 3.1, a k-basis for H0 (X, ΩX/k ) is given by {f dt ; f ∈ B} where
B is as in Definition 3.5.

Proof. Fix u ∈ Ubr and xu ∈ X above u. By (3.10), we have the following equalities of positive integers:
j k j k
dxu /u dxu /u −2mu
4 = µu,3 + νu , 4 = µu,1 + νu ,
(3.12) j k j k
dxu /u −2Mu dxu /u −mu −2(Mu −mu )
4 = µu,1 , 4 = µu,2 .

Recall the following equation from (2.13):

(  
mu
if mu ≡ 1 mod 4, and
(3.13) µu,2 − µu,1 =  m4u 
4 +1 if mu ≡ 3 mod 4.

Suppose bu ∈ Bu . Using (3.12), (3.13) and Lemma 2.8, we see that ordxu (bu ) ≥ −exu /u ku − dxu /u .
Suppose now that u′ ∈ Ubr − {u}. Since yu , zu ∈ {e y , ze, y]
+ z}, we obtain that both ordxu′ (yu ) and ordxu′ (zu )
are greater than or equal to −2Mu′ . Moreover, for a ∈ Z, it follows from (3.1) and Assumption 3.1 that
ordx∞ (πua ) = −4a and ordxu′ (πua ) = 0 = ordxu′ (π∞ a
) when u 6= ∞ and u′ 6∈ {u, ∞}. Using this, we obtain
that if u′ ∈ Ubr − {u} then ordxu′ (bu ) ≥ −exu′ /u′ ku′ − dxu′ /u′ .
Finally, suppose that x ∈ X with π(x) 6∈ Ubr . Since yu , zu ∈ {e y, ze, y]
+ z}, we obtain from (3.6) that both
ordx (yu ) and ordx (zu ) are greater than or equal to 0. Using this, we see that ordx (bu ) ≥ 0.
Therefore, f dt ∈ H0 (X, ΩX/k ) for all f ∈ B. Since k(X) = k(U )[y, z] = k(t)[e y , ze] has a k(t)-basis given
by {1, ye, ze, yeze} and since
y]
+ z = ye + ze + (sp + sq − sr )
by (3.7), we obtain that the elements in B are k-linearly independent.
Using that mu is a positive odd integer, in addition to (3.10) and (3.12), we see that
1
µu,3 + νu + µu,1 + νu + µu,2 = 2 dxu /u .

Therefore, we obtain (
1
2 dxu /u − 3 if u = ∞,
#Bu = 1
2 dxu /u if u 6= ∞,
which implies by (3.11) that #B = g(X). This completes the proof of Lemma 3.6. 
17
 The next theorem gives the precise decomposition of H0 (X, ΩX/k ) into a direct sum of indecomposable
kG-modules, where we use Notation 4.1 for the indecomposable kG-modules. Moreover, we define N0,λ to
be the zero module for all λ ∈ k ∪ {∞}.

Theorem 3.7. Under Assumptions 2.6 and 3.1, we have an isomorphism of kG-modules
M  ⊕a ⊕au,2

H0 (X, ΩX/k ) ∼
= N2ℓuu,1
,λu ⊕ N 2(ℓu −1),λu
⊕b
⊕ M3,1 ⊕ k ⊕c
u∈Ubr
P  mu +3 
P  3mu +3   2mu +3 

where b = u∈Ubr 4 − 1, c = u∈Ubr 4 − 4 , and ℓu , au,1 , au,2 are given as follows
for u ∈ Ubr :
   
(i) If δu = 0 then ℓu = 1, au,1 = 2m4u +3 − mu4+3 and au,2 = 0.
(ii) If δu ≥ 1 then ℓu ≥ 1 and 1 ≤ au,1 ≤ δu are given by
 2mu +3   mu +3 
4 − 4 = (ℓu − 1)δu + au,1
and au,2 = δu − au,1 .
(iii) If δu = −1 then ℓu ≥ 1 and 1 ≤ au,1 ≤ Mu −m
2
u
are given by
 2mu +3   mu +3  Mu −mu
4 − 4 + 2 = (ℓu − 1) Mu −m
2
u
+ au,1
Mu −mu
and au,2 = 2 − au,1 .

Proof. Fix u ∈ Ubr , and use Remark 3.3 and Definition 3.5. Define
σu := σ, τu := τ in the situation of Remark 2.7 (i) or (ii)(a),
σu := τ, τu := σ in the situation of Remark 2.7 (ii)(b), and
σu := σ ◦ τ, τu := τ in the situation of Remark 2.7 (ii)(c).
Recall from Lemma 2.8 that for 0 ≤ j ≤ µu,2 − µu,1 − 1, we have
j
!
X
−νu i
βu (j) = πu bu,i πu and wu (j) = zu + α
eu + βu (j) yu
i=0

where πu is as in (3.1). By (2.2) and Remark 2.7, we have the following G-actions on the elements of
Bu = Bu,1 ∪ Bu,2 ∪ Bu,3 :
σu − 1 τu − 1
fu,1,i1 (u) 0 0 for s(u) ≤ i1 (u) ≤ µu,3 + νu + ku
fu,2,i2 (u) 0 fu,1,i2 (u) for s(u) ≤ i2 (u) ≤ µu,1 + νu + ku
fu,3,i3 (u) fu,1,i3 (u) 0 for s(u) ≤ i3 (u) ≤ µu,1 + ku
i3 (u)−µu,1 −ku −1
X
fu,3,i3 (u) fu,1,i3 (u) bu,i fu,1,i3 (u)+νu −i for µu,1 + ku + 1 ≤ i3 (u) ≤ µu,2 + ku .
i=0
Define the following, pairwise disjoint subsets of Bu :
Bu,3,1,s1 (u) = {fu,1,s1 (u) , fu,2,s1 (u) , fu,3,s1 (u) } for s(u) ≤ s1 (u) ≤ µu,1 + ku ,
Bu,1,s2 (u) = {fu,1,s2 (u) } for µu,2 + νu + ku + 1 ≤ s2 (u) ≤ µu,3 + νu + ku .
Considering the actions of σu − 1 and τu − 1 on these subbases, we see that the k-span of each Bu,3,1,s1 (u)
gives a kG-module isomorphic to M3,1 , and the k-span of each Bu,1,s2 (u) gives the trivial kG-module k. It
remains to analyze the actions of σu − 1 and τu − 1 on the remaining (ordered) subbasis Bλu of Bu :
(3.14) Bλu := {fu,1,s3 (u) ; µu,1 + ku + 1 ≤ s3 (u) ≤ µu,2 + νu + ku }
∪ {fu,2,s4 (u) ; µu,1 + ku + 1 ≤ s4 (u) ≤ µu,1 + νu + ku }
∪ {fu,3,s5 (u) ; µu,1 + ku + 1 ≤ s5 (u) ≤ µu,2 + ku }.
18
Define nu := µu,2 − µu,1 + νu , so that Bλu has 2nu elements. We have nu = 0 if and only if νu = 0 and
mu = 1. In this case, δu = 0 since ⌊ m4u ⌋ = 0. Since au,1 = 0 by part (i) of the theorem, it follows correctly
that there are no direct summands of H0 (X, ΩX/k ) of the form N2ℓu ,λu associated to u.
Suppose from now on that nu ≥ 1. It follows that σu − 1 and τu − 1 act on the ordered basis Bλu from
(3.14) as the following square matrices:
! !
0nu Cλu 0nu Dλu
σu − 1 ←→ and τu − 1 ←→ ,
0nu 0nu 0nu 0nu
where 0nu is the nu × nu zero matrix and Cλu and Dλu are nu × nu matrices with (i, j)-entries (for 1 ≤
i, j ≤ nu )
(
1 if j = νu + i and and 1 ≤ i ≤ nu − νu ,
(Cλu )i,j =
0 otherwise,


 1 if i = j and 1 ≤ i ≤ νu ,
(Dλu )i,j = bu,j−i if νu + 1 ≤ i ≤ j ≤ nu ,


0 otherwise.
Suppose first that we are in parts (i) or (ii) of the theorem; in other words, we are in the situation of
Remark 2.7 (i). Then mu = Mu and νu = 0, which means that Cλu is the nu × nu identity matrix. By
Notation 2.10 and Equation (2.13), it follows that 0 ≤ δu ≤ nu . If δu = 0 or δu = nu then (Dλu − λu ) is
the zero matrix. If 0 < δu < nu then (Dλu − λu ) is a non-zero nilpotent matrix that is upper triangular.
Moreover, its entries along the diagonal strips (i, d + i) are all zero, for 0 ≤ d < δu and 1 ≤ i ≤ nu − d, and
its entries along the diagonal strip (i, δu + i) all equal bu,δu 6= 0, for 1 ≤ i ≤ nu − δu . In all cases, this implies
that (Dλu − λu )ℓu = 0 and (Dλu − λu )ℓu −1 6= 0. Moreover, if δu > 0 then, for i ≥ 0, the rank of the matrix
(Dλu − λu )i equals nu − i δu if i δu < nu and 0 otherwise. Hence, we obtain that Dλu has au,1 Jordan blocks
with eigenvalue λu of size ℓu and δu − au,1 Jordan blocks with eigenvalue λu of size ℓu − 1. Using Notation
⊕a ⊕au,2
4.1, it follows that the k-span of Bλu gives a kG-module isomorphic to N2ℓuu,1 ,λu ⊕ N2(ℓu −1),λu .
Suppose next that we are in part (iii) of the theorem; in other words, we are in the situation of Remark
2.7 (ii). Then mu < Mu and νu ≥ 1. This means that Cλu is a non-zero nilpotent upper triangular
matrix whose only non-zero entries are equal to 1 and occur precisely along the diagonal strip (i, νu + i),
for 1 ≤ i ≤ nu − νu . On the other hand, Dλu is an upper triangular matrix with non-zero diagonal entries.
Define C eλu := D−1 Cλu . Then C eλu has νu zero columns, followed by the first nu − νu columns of D−1 . Since
λu λu
Dλ−1 is upper triangular, we see that (Ceλu − 0)ℓu = 0 and (C eλu − 0)ℓu −1 6= 0. Moreover, for i ≥ 0, the
u

rank of the matrix (C eλu − 0)i equals nu − i νu if i νu < nu and 0 otherwise. Hence, we obtain that C eλu
has au,1 Jordan blocks with eigenvalue 0 of size ℓu and νu − au,1 Jordan blocks with eigenvalue 0 of size
ℓu − 1. Using Notation 4.1 and Remark 4.3, it follows that the k-span of Bλu gives a kG-module isomorphic
⊕a ⊕au,2
to N2ℓuu,1
,λu ⊕ N2(ℓu −1),λu . This completes the proof of Theorem 3.7. 

Remark 3.8. Let u ∈ Ubr . Then we have ℓu = 1 in Theorem 3.7 in the following cases:
(i) If δu = 0 then always ℓu = 1.
 
(ii) If δu ≥ 1 then ℓu = 1 if and only if mu ≡ 1 mod 4 and δu = m4u . This follows from Equation
 
(2.13) since 1 ≤ δu ≤ m4u .
   
(iii) If δu = −1 then ℓu = 1 if and only if mu = 1. This follows since 2m4u +3 − mu4+3 ≥ 0.
In particular, these cases coincide, for U = P1k , with the cases covered by Theorem 2.16.

The next example shows that the list of isomorphism classes of indecomposable kG-modules that actually
occur as direct summands of H0 (X, ΩX/k ) in the situation of Theorem 3.7 for various X is infinite and given
19
as follows:

(3.15) N2d,λ ; d ∈ Z+ , λ ∈ k ∪ {∞} ∪ {M3,1 , k} .

In particular, this list contains indecomposable kG-modules of arbitrarily large k-dimension.

Example 3.9. Let U = P1k with function field k(U ) = k(t). Given a positive integer d, we provide a family of
examples such that there are at least two non-isomorphic indecomposable kG-modules whose k-dimensions
are 2d and 2(d + 1), respectively, and that occur as direct summands of H0 (X, ΩX/k ). All these examples
are Harbater-Katz-Gabber G-covers, in the sense of [3, §4.B], i.e. there is exactly one branch point and this
unique branch point is totally ramified. In the examples below, we have have chosen the branch point to
be at ∞. However, if we change the variable t to e e of P1 such that
t := t−1 , we obtain examples of covers X k
e e
the places in k(X) above ∞ are unramified. In particular, k(X) is then a global standard function field, in
the sense of [14, Definitions 2.1 and 3.1]. Since H0 (X,e ΩX/k ) ∼= H0 (X, ΩX/k ) as kG-modules, we obtain a
contradiction to [14, Theorem 6.4], similarly to Examples 2.18 and 2.19.
To cover all modules given in the list (3.15), we need to consider two main cases. Let λ ∈ k − {0, 1}.
(1) Let p, q in Assumption 3.1 be given by

p = t8d+3 ,
q = λ2 t8d+3 (1 + t−4 ),

which implies that r = p + q = (1 + λ2 + λ2 t−4 ) t8d+3 . Then

ord∞ (p) = −(8d + 3) = ord∞ (q) = ord∞ (r),

and ordu (p), ordu (q), ordu (r) ≥ 0 for all other u ∈ P1k . This means that Ubr = {∞}, and m∞ =
8d + 3 = M∞ . If w := z + λ(1 + t−2 ) y then ordx∞ (w) = −(8d + 3) when x∞ ∈ X lies above ∞.
2
Using Remark 2.7, it follows that β∞ = λ(1 + π∞ ). Therefore, in Notation 2.10, we have λ∞ = λ
and δ∞ = 2. Since
 2m∞ +3   m∞ +3 
4 − 4 = 4d + 2 − (2d + 1) = 2d + 1,

we obtain by part (ii) of Theorem 3.7 that ℓ∞ = d + 1 and a∞,1 = 1 = a∞,2 . Therefore,
⊕(2d)
H0 (X, ΩX/k ) ∼
= N2(d+1),λ ⊕ N2d,λ ⊕ M3,1 ⊕ k ⊕(2d+1) .

(2) Define
f = t8d−5 , g = λ2 t8d−1 , and h = t8d−1 (λ2 + t−4 )
which implies that the sum of any two of these polynomials equals the third one. Then

ord∞ (f ) = −(8d − 5) and ord∞ (g) = −(8d − 1) = ord∞ (h),

and ordu (f ), ordu (g), ordu (h) ≥ 0 for all other u ∈ P1k .
(a) Let p, q in Assumption 3.1 be given by p = f and q = g. Then Ubr = {∞}, and m∞ = 8d − 5
and M∞ = 8d − 1. If w := z + λt2 y then ordx∞ (w) = −(8d + 3) when x∞ ∈ X lies above ∞.
−2
Using Remark 2.7, it follows that β∞ = λ π∞ . Therefore, in Notation 2.10, we have δ∞ = −1
and λ∞ = ∞. Since
 2m∞ +3   m∞ +3  M∞ −m∞
4 − 4 + 2 = 4d − 2 − (2d − 1) + 2 = 2d + 1,

we obtain by part (iii) of Theorem 3.7 that ℓ∞ = d + 1 and au∞ ,1 = 1 = au∞ ,2 . Therefore,
⊕(2d−2)
H0 (X, ΩX/k ) ∼
= N2(d+1),∞ ⊕ N2d,∞ ⊕ M3,1 ⊕ k ⊕(2d−1) .
20
 (b) If we define p, q in Assumption 3.1 by p = g and q = f , then everything stays the same as in
part (a) except that we now have w = y + λt2 z and λ∞ = 0. Hence we obtain
⊕(2d−2)
H0 (X, ΩX/k ) ∼
= N2(d+1),0 ⊕ N2d,0 ⊕ M3,1 ⊕ k ⊕(2d−1) .
(c) If we define p, q in Assumption 3.1 by p = h and q = g, then everything stays the same as in
part (a) except that we now have w = z + λt2 (y + z) and λ∞ = 1. Hence we obtain
⊕(2d−2)
H0 (X, ΩX/k ) ∼
= N2(d+1),1 ⊕ N2d,1 ⊕ M3,1 ⊕ k ⊕(2d−1) .

Remark 3.10. It follows from Remark 2.20 and Example 3.9 that the list (3.15) is a complete list of iso-
morphism classes of indecomposable kG-modules that actually occur as direct summands of H0 (X, ΩX/k ) as
soon as the inequality in (2.22) is satisfied, with no restriction on the genus of U = X/G.

4. Appendix: Characteristic two Klein four representation theory

Let k be an algebraically closed field of characteristic 2 and let G = hσ, τ i be isomorphic to Z/2 × Z/2.

Notation 4.1. We define the following infinite list of kG-modules:

(a) Let k denote the trivial simple kG-module, and let kG denote the free rank one kG-module.
(b) For each positive integer n and each λ ∈ k ∪ {∞}, let N2n,λ be the kG-module of k-dimension 2n
such that σ − 1 and τ − 1 act as the following matrices with respect to a suitable k-basis of N2n,λ :
• if λ ∈ k then
! !
0n 1n 0n Jn (λ)
σ − 1 ←→ and τ − 1 ←→ ,
0n 0n 0n 0n
• and if λ = ∞ then
! !
0n Jn (0) 0n 1n
σ − 1 ←→ and τ − 1 ←→ ,
0n 0n 0n 0n
where 0n (resp. 1n ) is the n × n zero matrix (resp. the n × n identity matrix), and Jn (λ) is the
upper triangular Jordan block of size n with eigenvalue λ.
(c) For every positive integer n, let M2n+1,1 and M2n+1,2 be two indecomposable kG-modules of k-
dimension 2n + 1 such that σ − 1 and τ − 1 act as the following matrices with respect to a suitable
k-basis of M ∈ {M2n+1,1 , M2n+1,2 }:
• if M = M2n+1,1 then
   
0n 1n 0n×1 0n 0n×1 1n
   
σ − 1 ←→  0n 0n 0n×1  and τ − 1 ←→  0n 0n×1 0n  ,
01×n 01×n 0 01×n 0 01×n
• and if M = M2n+1,2 then
   
0n 0n×1 1n 0n 0n×1 01×n
   
σ − 1 ←→  01×n 0 01×n  and τ − 1 ←→  01×n 0 1n  ,
0n 0n×1 0n 0n 0n×1 0n
where 0n×1 (resp. 01×n ) is the zero column vector (resp. zero row vector) of length n.

Remark 4.2. Each indecomposable kG-module is isomorphic to one of the modules in the list
 
{k, kG} ∪ N2n,λ ; n ∈ Z+ , λ ∈ k ∪ {∞} ∪ M2n+1,1 , M2n+1,2 ; n ∈ Z+
and no two such modules are isomorphic to each other (see, for example, [1] or [6]).
21
Remark 4.3. For each positive integer n, there is the following connection between N2n,∞ , N2n,0 and N2n,1 .
Let (ρ1 , ρ2 ) be an ordered pair of two distinct elements of {σ, τ, σ ◦ τ }, i.e. (ρ1 , ρ2 ) is an ordered pair of
generators of G. Consider the kG-module M of k-dimension 2n with actions by ρ1 − 1 and ρ2 − 1 given as
the following matrices with respect to a suitable k-basis of M :
! !
0n Jn (0) 0n 1n
ρ1 − 1 ←→ and ρ2 − 1 ←→ ,
0n 0n 0n 0n
Since the product of these two matrices is the zero matrix, it follows that ρ1 ◦ ρ2 − 1 acts as
!
0n Jn (1)
ρ1 ◦ ρ2 − 1 ←→ .
0n 0n
We obtain:
(a) If (ρ1 , ρ2 ) = (σ, τ ) then M ∼ = N2n,∞ .
(b) If (ρ1 , ρ2 ) = (τ, σ) then M ∼ = N2n,0 .
(c) If (ρ1 , ρ2 ) = (σ ◦ τ, τ ) then M ∼= N2n,1 .

Remark 4.4. A similar phenomenon to the one described in Remark 4.3 occurs for all parameters in k ∪ {∞}.
More precisely, fix a positive integer n and fix λ ∈ k ∪ {∞}. For any permutation ξ of the set {σ, τ, σ ◦ τ },
ξ
define N2n,λ to be the kG-module with the same underlying k-vector space as N2n,λ but on which each g ∈ G
acts as ξ(g). Let ξ1 be the permutation interchanging σ and τ , and let ξ2 be the permutation interchanging
τ and σ ◦ τ . Identifying 01 = ∞, ∞1
= 0 and 1 + ∞ = ∞, we obtain
ξ1
N2n,λ ∼
= N2n, λ1
ξ2
and N2n,λ ∼
= N2n,1+λ .
Let Sλ be the set of all µ ∈ k ∪ {∞} for which there exists a permutation ξ of {σ, τ, σ ◦ τ } such that
ξ
N2n,λ ∼
= N2n,µ . Then
 
1 1 λ 1+λ
Sλ = λ, , 1 + λ, , , .
λ 1+λ 1+λ λ
This set contains precisely 6 elements for all λ ∈ k − F4 . Moreover,

S0 = {0, 1, ∞} = S1 = S∞ ,

and, if F×
4 = hαi, then
Sα = {α, 1 + α} = S1+α .

Fix (a : b) ∈ P1k . Table 4.1 provides, for every indecomposable kG-module T , the k-dimensions of the
subquotients T (i+1) /T (i) for 0 ≤ i ≤ 3, as defined in Notation 2.2.

Table 4.1. Dimensions of successive subquotients of indecomposable kG-modules.

T dimk T (4) /T (3) dimk T (3) /T (2) dimk T (2) /T (1) dimk T (1) /T (0)
k 0 0 0 1
kG 1 1 1 1
n − 1 if λ = a/b 1 if λ = a/b
N2n,λ 0 n
n if λ 6= a/b 0 if λ 6= a/b
M2n+1,1 0 n 1 n
M2n+1,2 0 n 0 n+1

22
 References
[1] V. A. Bašev. Representations of the group Z2 × Z2 in a field of characteristic 2. Dokl. Akad. Nauk SSSR, 141:1015–1018,
1961.
[2] F. M. Bleher, T. Chinburg, and A. Kontogeorgis. Galois structure of the holomorphic differentials of curves. J. Number
Theory, 216:1–68, 2020.
[3] F. M. Bleher, T. Chinburg, B. Poonen, and P. Symonds. Automorphisms of Harbater-Katz-Gabber curves. Math. Ann.,
368(1-2):811–836, 2017.
[4] N. Camacho. Holomorphic Differentials of Klein Four Covers. ProQuest LLC, Ann Arbor, MI, 2020. Thesis (Ph.D.)–The
University of Iowa.
[5] C. Chevalley, A. Weil, and E. Hecke. Über das Verhalten der Integrale 1. Gattung bei Automorphismen des Funktio-
nenkörpers. Abh. Math. Sem. Univ. Hamburg, 10(1):358–361, 1934.
[6] S. B. Conlon. Certain representation algebras. J. Austral. Math. Soc., 5:83–99, 1965.
[7] J. Garnek. p-group Galois covers of curves in characteristic p. Preprint, arXiv:2110.10789, 2022.
[8] R. Hartshorne. Algebraic geometry. Springer-Verlag, New York, 1977. Graduate Texts in Mathematics, No. 52.
[9] H. Hasse. Theorie der relativ-zyklischen algebraischen Funktionenkörper, insbesondere bei endlichem Konstantenkörper.
J. Reine Angew. Math., 172:37–54, 1935.
[10] E. Hecke. Über ein Fundamentalproblem aus der Theorie der elliptischen Modulfunktionen. Abh. Math. Sem. Univ. Ham-
burg, 6(1):235–257, 1928.
[11] E. Kani. The Galois-module structure of the space of holomorphic differentials of a curve. J. Reine Angew. Math., 367:187–
206, 1986.
[12] S. Karanikolopoulos and A. Kontogeorgis. Representation of cyclic groups in positive characteristic and Weierstrass semi-
groups. J. Number Theory, 133(1):158–175, 2013.
[13] B. Köck. Galois structure of Zariski cohomology for weakly ramified covers of curves. Amer. J. Math., 126(5):1085–1107,
2004.
[14] S. Marques and K. Ward. Holomorphic differentials of certain solvable covers of the projective line over a perfect field.
Math. Nachr., 291(13):2057–2083, 2018.
[15] S. Nakajima. Galois module structure of cohomology groups for tamely ramified coverings of algebraic varieties. J. Number
Theory, 22(1):115–123, 1986.
[16] M. Rzedowski-Calderón, G. Villa-Salvador, and M. L. Madan. Galois module structure of holomorphic differentials in
characteristic p. Arch. Math. (Basel), 66(2):150–156, 1996.
[17] H. Stichtenoth. Algebraic function fields and codes, volume 254 of Graduate Texts in Mathematics. Springer-Verlag, Berlin,
second edition, 2009.
[18] R. C. Valentini and M. L. Madan. Automorphisms and holomorphic differentials in characteristic p. J. Number Theory,
13(1):106–115, 1981.
[19] Q. Wu and R. Scheidler. The ramification groups and different of a compositum of Artin-Schreier extensions. Int. J. Number
Theory, 6(7):1541–1564, 2010.

23