Analytic and Algebraic Geometry

Anilatmaja Aryasomayajula
Indranil Biswas
Archana S. Morye
A.J. Parameswaran Editors
Analytic
and Algebraic
Geometry
Analytic and Algebraic Geometry
Indranil Biswas Archana S. Morye
•
A.J. Parameswaran
Editors
Analytic and Algebraic

Geometry
123
Editors
Anilatmaja Aryasomayajula Archana S. Morye
Department of Mathematics School of Mathematics and Statistics
IISER Tirupati University of Hyderabad
Tirupati, Andhra Pradesh Hyderabad, Telangana
India India
Indranil Biswas A.J. Parameswaran

School of Mathematics School of Mathematics
Tata Institute of Fundamental Research Tata Institute of Fundamental Research
Mumbai, Maharashtra Mumbai, Maharashtra
India India
ISBN 978-981-10-5648-2 (eBook)

DOI 10.1007/978-981-10-5648-2
Library of Congress Control Number: 2017946031
This work is a co-publication with Hindustan Book Agency, New Delhi, licensed for sale in all countries
in electronic form only. Sold and distributed in print across the world by Hindustan Book Agency, P-19
Green Park Extension, New Delhi 110016, India. ISBN: 978-93-86279-64-4 © Hindustan Book
Agency 2017.
© Springer Nature Singapore Pte Ltd. 2017 and Hindustan Book Agency 2017
This work is subject to copyright. All rights are reserved by the Publishers, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publishers, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publishers nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publishers remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by Springer Nature

The registered company is Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore
Preface
For more than ten years, Indranil Biswas and A. J. Parameswaran of Tata In-
stitute of Fundamental Research, Mumbai are organizing an annual international
conference on the topics around Analytic and Algebraic Geometry. The first
part of the conference is held at the Tata Institute of Fundamental Research
and the second part in some other institute. These conferences are primarily in-
tended to facilitate interactions, between mathematicians in India and in other
countries, working in related areas. Another purpose of these conferences is to
have series of lectures by experts on topics of interest to mathematicians in
India. These annual conferences have initiated a large number of collaborative
works in Mathematics.
In 2015, a conference on Analytic and Algebraic Geometry was held at the
Tata Institute of Fundamental Research and the University of Hyderabad. This
volume is an outcome of that conference, although not all speakers submitted
their lecture material.
There are fifteen articles in this volume. The main purpose of the articles
is to introduce recent and advanced techniques in Analytic and Algebraic Ge-
ometry. We have attempted to give recent developments in the area to target
mainly young researchers, who are new to this area. We have also added some
research articles to give examples of how to use these techniques to prove new
results.
Our sincere thanks to the School of Mathematics, the University of Hyderabad
for their generous financial and logistic support for this conference.
Editors
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
About the Editors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Marco Antei
On the Bumpy Fundamental Group Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Heat Kernels, Bergman Kernels, and Cusp Forms . . . . . . . . . . . . . . . . . . . . . . . 19
Usha N. Bhosle
On a Conjecture of Butler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Indranil Biswas and Mahan Mj
A Survey of Low Dimensional (Quasi) Projective Groups . . . . . . . . . . . . . . . . 49
Niels Borne
Parabolic Sheaves and Logarithmic Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Emre Coskun
A Survey of Ulrich Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Ananyo Dan
Noether-Lefschetz Locus and a Special Case of the Variational
Hodge Conjecture: Using Elementary Techniques . . . . . . . . . . . . . . . . . . . . . 107
A. El Mazouni and D.S. Nagaraj
Tangent Bundle of P2 and Morphism from P2 to Gr(2, C4 ) . . . . . . . . . . . . . . 117
Michel Emsalem
Twisting by a Torsor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Viktoria Heu and Frank Loray
Hitchin Hamiltonians in Genus 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
Inder Kaur
Smoothness of Moduli Space of Stable Torsion-free Sheaves with Fixed
Determinant in Mixed Characteristic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Johan Martens
Group Compactifications and Moduli Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
viii Contents
Archana S. Morye
The Serre-Swan Theorem for Ringed Spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Georg Schumacher
An Extension Theorem for Hermitian Line Bundles . . . . . . . . . . . . . . . . . . . . . 225
Tathagata Sengupta
Elliptic Fibrations on Supersingular K3 Surface with Artin Invariant 1 in
Characteristic 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
About the Editors
Anilatmaja Aryasomayajula is assistant professor at Indian Institute of

Science Education and Research (IISER), Tirupati, India. He did his
Ph.D. from the Humboldt University of Berlin. A DST inspire fellow,
his areas of interest are automorphic forms, Hilbert modular forms, Arakelov
theory and Diophantine geometry.
Indranil Biswas is professor of mathematics at the Tata Institute of

Fundamental Research (TIFR), Mumbai, India. He works in algebraic
geometry, differential geometry, several complex variables and analytic
space, and deformation quantization. He has more than 400 publications to
his credit with many collaborators across the world. In 2006, the government
of India awarded him the Shanti Swarup Bhatnagar Prize in Mathematical
Sciences for his contributions to algebraic geometry, more precisely moduli
problems of vector bundles. He is a fellow of Indian Academy of Sciences and
Indian National Science Academy (INSA).
Archana S. Morye is a faculty in the University of Hyderabad, India. She did

her Ph.D. from Harish-Chandra Research Institute, Allahabad, India. Her research
area is algebraic geometry more specifically in vector bundles.
A.J. Parameswaran is professor of mathematics at the Tata Institute of

Fundamental Research (TIFR), Mumbai, India. He was the director of
Kerala School of Mathematics (KSOM), Kozhikode, India, for a short period of time.
He works in algebraic geometry, several complex variables and analytic
space. He is a fellow of Indian Academy of Sciences.
On the Bumpy Fundamental
Group Scheme
Marco Antei∗,†
Abstract
In this short paper we first recall the definition and the construction of the
fundamental group scheme of a scheme X in the known cases: when it is defined
over a field and when it is defined over a Dedekind scheme. It classifies all the
finite (or quasi-finite) fpqc torsors over X. When X is defined over a noetherian
regular scheme S of any dimension we do not know if such an object can be
constructed. This is why we introduce a new category, containing the fpqc
torsors, whose objects are torsors for a new topology. We prove that this new
category is cofiltered thus generating a fundamental group scheme over S, said
bumpy as it may not be flat in general. We prove that it is flat when S is a
Dedekind scheme, thus coinciding with the classical one.
Mathematics Subject Classification (2000). 14G99, 14L20, 14L30
1. Introduction
In his famous [SGA1] Alexander Grothendieck constructs the étale fundamental
group π ét (X, x) of a scheme X, endowed with a geometric point x : Spec(Ω) →
X. It is a pro-finite group classifying the étale covers over X. Grothendieck
elaborates for π ét (X, x) a specialization theory ([SGA1], Chapitre X) but it
seems he is not entirely satisfied of it. Indeed at the end of the aforementioned
tenth chapter he claims (here we freely translate from French1 ) that
∗ The author thanks the project TOFIGROU (ANR-13-PDOC-0015-01)

† Laboratoire J.A.Dieudonné, UMR CNRS-UNS No 7351 Université de Nice Sophia-
Antipolis, Parc Valrose, 06108 NICE Cedex 2, France.
E-mail: [email protected]
1 Une théorie satisfaisante de la spécialisation du groupe fondamental doit tenir compte
de la “composante continue” du “vrai” groupe fondamental, correspondant à la classification

des revtements principaux de groupe structural des groupes infinitésimaux; moyennant quoi
A. Aryasomayajula et al. (eds.), Analytic and Algebraic
Geometry, DOI 10.1007/978-981-10-5648-2_1
2 Marco Antei
a satisfactory specialization theory of the fundamental group should consider

the “continuous component” of the “true” fundamental group corresponding to
the classification of principal homogeneous spaces with infinitesimal structure
group; whereby one would have to expect that the “true” fundamental groups of
the geometric fibres of a smooth and proper morphism f : X → Y form a nice
local system over X, projective limit of finite and flat group schemes over X.
Though in a footnote in the same page he recalls us that2
this extremely seducing conjecture is unfortunately contradicted by an unpub-

lished example of M. Artin, even when the fibres of f are algebraic curves of
given genus g ≥ 2
the question of constructing a “true” fundamental group was neverthless very

interesting. It is only at the end of the seventies that Madhav Nori, in his PhD
thesis (see [No76] and [No82]), constructs it calling it “the fundamental group
scheme”. It will be called again “the true fundamental group” in [DeMi82] but,
at our knowledge, it will be the last time. In §2 we will recall his description,
while in §3 we will give some details about the construction of the fundamental
group scheme of a given scheme X defined over a Dedekind scheme. In §4 we
will finally suggest a new approach for higher dimensional base schemes: as
we do not know if a fundamental object classifying all finite (or quasi-finite)
fpqc torsors can be constructed, we introduce a new larger category whose
objects are torsors for a new topology that we will describe. We prove that this
new category is cofiltered thus generating a S-fundamental group scheme, said
bumpy, classifying all those new torsors. Though in general it may not be flat
(which explains its name) we prove in §5 that it is flat at least when S is a
Dedekind scheme, thus coinciding with the classical one.
2. Over a Field
The main references for the material contained in this sections are certainly
[No76] and [No82], but also [Sz09], Chapter 6. Other references will be provided
when necessary for more detailed results.
on serait en droit à s’attendre que les “vrais” groupes fondamentaux des fibres géométriques
d’un morphisme lisse et propre f : X → Y forment un joli système local sur X, limite
projective de schémas en groupes finis et plats sur X.
2 Cette conjecture extrêmement séduisante est malheureusement mise en défaut par un
exemple inédit de M. Artin, déjà lorsque les fibres de f sont des courbes algébriques de genre
donné g ≥ 2.
On the Bumpy Fundamental Group Scheme 3
2.1. Tannakian description. Let k be a perfect3 field, X a reduced

and connected scheme, proper over Spec(k), endowed with a k-rational point x.
A vector bundle V on X is said to be finite if there exist two polynomials f and
g with f = g with non-negative integral coefficients such that f (V ) g(V ),
where, when we evaluate the polynomials, we have replaced the sum by the
direct sum and the product by the tensor product. Now, let C be a smooth
and proper curve over k and W a vector bundle on C. Then W is said to be
semi-stable of degree 0 if deg(W ) = 0 and for any sub-vector bundle U ⊂ W
we have deg(U ) ≤ 0.
Definition 2.1. A vector bundle V on X is said to be Nori semi-stable4 if for
any curve C smooth and proper over k and every morphism j : C → X the
vector bundle j ∗ (V ) is semi-stable of degree 0.
When the characteristic of the field k is 0 then the category of finite vector
bundles is tannakian if we endow it with the trivial object OX , the tensor prod-
uct ⊗OX (of locally free sheaves) and the fibre functor x∗ (here x : Spec(k) → X
is the given rational point). When the characteristic of the field k is positive
the category of finite vector bundle is in general not even abelian, this is why
we need a bigger category containing the finite vector bundles. Nori proves that
every finite vector bundle is indeed Nori semi-stable (cf. Definition 2.1): Nori’s
intuition has been to understand that the the abelian hull of the category of
finite vector bundles inside the (abelian) category of Nori semi-stable vector
bundles would be a good candidate to obtain an interesting tannakian category.
He calls the objects of this category essentially finite vector bundles and the
category itself is denoted by EF (X), full subcategory of Coh(X), the category
of coherent sheaves over X: it is thus the abelian category generated by finite
vector bundles, their duals, their tensor products and their sub-quotients, full
sub-category of the category of Nori semi-stable vector bundles. The following
theorem summarize what we just said:
Theorem 2.2. The category EF (X) of essentially finite vector bundles en-
dowed with the trivial object OX , the tensor product ⊗OX and the fibre functor
x∗ is a tannakian category over k.
The duality theorem for tannakian categories (that we are tacitly assuming
to be neutral) says that to (EF (X), OX , ⊗OX , x∗ ) we can naturally associate
an affine k-group scheme that we denote by π(X, x). This is what we call the
3 In these notes we are using Nori’s setting, though it has been observed several times
that his tannakian construction would hold for any field k provided X satisfies the condition
H 0 (X, OX ) = k, which is automatically satisfied when k is perfect.
4 Of course this name has been given later; in Nori’s works those vector bundles did not
have a specific name.

4 Marco Antei
fundamental group scheme of X in x. Though in general a group scheme over

a field is pro-algebraic ([Wa79], §3.3) Nori proves that π(X, x) is pro-finite.
Furthermore (cf. for instance [DeMi82], Theorem 3.2) this theory provides a
pointed in x
π(X, x)-torsor X, ∈ X x (k) which is universal in the following
sense: let G be any finite k-group scheme and Y → X any G-torsor, pointed
in y ∈ Yx (k), then there exists a unique X-morphism X → Y of torsors (i.e.
commuting with the actions of π(X, x) and G) sending x
→ y. So we see that
all the finite torsors over X are classified by the fundamental group scheme,
which is thus the object dreamt by Grothendieck. When k is algebraically
closed then the group of rational points π(X, x)(k) of π(X, x) coincide with
the étale fundamental group π ét (X, x). We conclude this section with a recent
result, obtained in [BoVi] by Niels Borne and Angelo Vistoli who provide a
very nice, useful and surprisingly simple characterization for essentially finite
vector bundles:
Proposition 2.3. Let k, X and x like before then a vector bundle V on X
is essentially finite if and only if it is the kernel of a morphism between finite
vector bundles.
2.2. Description as pro-finite limit. There is a second construction

for the fundamental group scheme of a scheme X defined over a field k, again
due to Nori (cf.[No82]) and certainly inspired by the universal property satisfied
by the universal torsor over X, as recalled in 2.2. This new construction is easier
than the previous one, we do not use vector bundles or tannakian theory so it
is in general more understandable by a wider audience. The assumptions on X
can be weakened a lot and at some point some people started to believe that
the field k could be replaced by a discrete valuation ring or, more generally, by
a Dedekind scheme, but this will be the object of §3. In this section X will be a
reduced and conected scheme defined over any field k and x ∈ X(k) will be a k-
rational point. We consider the category F(X) where objects are finite torsors
(i.e. their structural group is a k-finite group scheme) over X, pointed over x
and morphisms are morphisms commuting with the actions of their structural
groups and sending a marked point to a marked point. We will often write
(Y, G, y) for a G-torsor Y → X pointed at y. Before stating the main result for
this section let us briefly recall the notion of cofiltered category: a category C
is cofiltered if it is non empty and if the following axioms are satisfied:
1. for any two objects A, B of C there exists an object C of C and two
morphisms cA : C → A and cB : C → B;
2. for any two objects A, B for any two morphisms c1 : A → B and c2 :
A → B there exists an object U of C and a morphism u : U → A such
that c1 ◦ u = c2 ◦ u.
We also observe that if the category C has a final object then C is cofiltered
if and only if for any triple of objects A, B, C and for any pair of morphisms
f : A → C, g : B → C there exists, in C, the fibre product A ×C B then the
category5 . This is a nice exercise left to the reader. We can now state the main
result of this section, due to Nori:
Theorem 2.4. Let X be a reduced and connected scheme over any field k and
let x ∈ X(k) be a k-rational point. Then the category F(X) is cofiltered.
Proof. The strategy is the following: we take three objects of F(X)

(Yi , Gi , yi ), i = 0, ..., 2 and two morphisms like in the following diagram
(Y1 , G1 , y1 ) (Y2 , G2 , y2 )
OOO oo
OOO ooo
OOO ooo
OO' wooo
(Y0 , G0 , y0 )
and we want to prove that this diagram has a fibre product in F(X). We chose
as a candidate the triple given by (Y3 , G3 , y3 ) := (Y1 ×Y0 Y2 , G1 ×G0 G2 , y1 ×y0 y2 )
and we need to prove that it is an object of F(X). It is not trivial to prove
that Y1 ×Y0 Y2 is faithfully flat over X and it is at this point that we use our
hypothesis on X, that we have assumed to be reduced and connected. Once we
prove that (Y3 , G3 , y3 ) is a torsor over X we are done.
The importance of this result is that we can now wonder whether or not we
can compute the projective limit of all the finite pointed torsors over X. But
since F(X) is cofiltered and all the morphisms in F(X) are affine then this limit
exists, what we obtain is again a universal triple (X̃, π(X, x), x̃); unsurprisingly
we call again the pro-finite k-group scheme π(X, x) the fundamental group
scheme of X in x and X̃ the universal π(X, x)-torsor (pointed at x̃). The fact
that no confusion can arise is a consequence of the following
Remark 2.5. Let k be a perfect field, X a reduced and connected scheme,

proper over Spec(k), endowed with a k-rational point such that H 0 (X, OX ) =
k then the tannakian construction introduced in §2.1 and the construction
described by the Theorem 2.4 give rise to the same fundamental group scheme
π(X, x).
5 Alternatively if C has a final object U , if there exist finite products (i.e. fibre products
of objects over U ) and if for any triple of objects A, B, C and for any pair of morphisms
f : A → C, g : B → C there exists, in C, a forth object D in C closing the diagram, then the
category C is cofiltered
6 Marco Antei
3. Over a Dedekind Scheme

As we pointed out in §2.2 the pro-finite construction is more likely to be ex-
tended to a much more general situation, for instance when we replace the base
field k by a Dedekind scheme S. First of all, to prevent any ambiguity, through-
out the whole paper by Dedekind scheme we mean a normal locally Noetherian
connected scheme of dimension 0 or 1 (e.g. the spectrum of Z, the spectrum of
a discrete valuation ring, the spectrum of a field, ... ). So we have a Dedekind
scheme S, a scheme X faithfully flat and of finite type over S, we fix a section
x ∈ X(S) and we wonder whether we can build a universal torsor dominating
all the finite torsors over X (i.e. torsors whose structural group is a finite and
flat S-group scheme). As the case dim(S) = 0 has already been considered we
only study here the case dim(S) = 1. The question is as before: is the category,
that we still denote by F(X), of finite pointed torsors over X cofiltered? A
first answer has been given in [Ga03], but the proof contains a mistake which
has been corrected in [AEG15]. We thus introduce here some ideas contained
in [AEG15], to which we refer the reader for the complete proofs of the results
stated in this section and much more. Whenever we have a S-scheme T , we will
denote by Tη its generic fibre. We first give an answer to the above question
(this is the subject of [AEG15], §4 and §5.1):
Theorem 3.1. Let S, X and x like before. Let us moreover assume that one
of the following assumptions is satisfied:
1. for every s ∈ S the fibre Xs is reduced;
2. for every z ∈ X\Xη the local ring OX,z is integrally closed.
Then the category F(X) is cofiltered.
Proof. The strategy being the same as for the field case, we take three objects
of F(X) (Yi , Gi , yi ), i = 0, ..., 2 and two morphisms as follows
(Y1 , G1 , y1 ) (Y2 , G2 , y2 )
OOO oo
OOO ooo
OOO ooo
OO' wooo
(Y0 , G0 , y0 )
and we want to prove that this diagram has a fibre product in F(X). Unfortu-
nately in general the triple given by (Y1 ×Y0 Y2 , G1 ×G0 G2 , y1 ×y0 y2 ) is not a
good candidate as G1 ×G0 G2 may easily be not flat (though finite) over S. But
we can hopefully do the following: we can consider (Y1 ×X Y2 , G1 ×S G2 , y1 ×S y2 )
(which is certainly a torsor), its generic fibre (which is a torsor over Xη ), the
natural closed immersion
Y1,η ×Y0,η Y2,η → Y1,η ×Xη Y2,η
(which is a morphism of torsors, we simply omitted the structural groups and

the marked points to ease the notation) and we need to hope that the Zariski
closure of Y1,η ×Y0,η Y2,η in Y1 ×X Y2 is a torsor under the action of the S-
finite and flat group scheme obtained as Zariski closure of G1,η ×G0,η G2,η in
G1 ×S G2 ; it turns out to be the good candidate we were looking for and this
follows from Lemma 3.2.
Lemma 3.2. Notations are those of Theorem 3.1. Let us moreover assume
that one of the assumptions given in Theorem 3.1 is satisfied. Let (Y, G, y) be
an object of F(X) and (T, H, t) an object of F(Xη ) contained in (Yη , Gη , yη )
(i.e. we are considering a morphism (T, H, t) → (Yη , Gη , yη ) whose induced
morphism on the group schemes H → Gη is a close immersion). Then the
triple (T , H, t) obtained as the Zariski closure of (T, H, t) in (Yη , Gη , yη ) is an
object of F(X).
Exactly like before we thus have a fundamental group scheme π(X, x) of X
in x and a universal π(X, x)-torsor over X dominating, by a unique morphism,
every finite pointed torsor over X.
Working over a base scheme with dimension > 0 gives us a lot of freedom
that we do not have when we are over a field: for example dealing with affine
S-group schemes we immediately notice that there are many group objects over
S whose fibres are finite group schemes. Of course there are finite and flat group
schemes, which we already considered, but quasi-finite and flat group schemes
(not necessarily finite) can also be considered. Finally, if we are brave enough,
we can also consider non flat quasi-finite group schemes, but for this we need
to wait until §4. We now conclude this section with a brief overview on the
“quasi-finite and flat” world: again we have a Dedekind scheme S, a scheme
X faithfully flat and of finite type over S, we fix a section x ∈ X(S) and we
wonder whether we can build a universal torsor dominating all the quasi-finite
torsors over X (i.e. torsors whose structural group is an affine quasi-finite and
flat S-group scheme). We now have a new, but similar, question: is the category,
that we denote by Qf (X), of quasi-finite pointed torsors over X cofiltered? At
the end of this section we will explain why it can be useful to work in this new
setting; first we state an existence result:
Theorem 3.3. Let S, X and x like before. Let us moreover assume that X is
integral and normal and that for each s ∈ S the fibre Xs is normal and integral.
Then the category Qf (X) is cofiltered.
8 Marco Antei
Proof. The proof is quite long and though the strategy is similar to that of
Theorem 3.1, the details are a bit more complicated and we refer the reader to
[AEG15], §5.2.
Again, not surprisingly, we have a fundamental group scheme π qf (X, x) of X

in x and a universal π qf (X, x)-torsor over X dominating, by a unique morphism,
every quasi-finite pointed torsor over X. The reason why π qf (X, x) can be useful
in many situations is the following: every finite torsor is, in particular, quasi-
finite, so there is a natural morphism
π qf (X, x) → π(X, x)
which is a schematically dominant morphism (i.e. the dual morphism on the

coordinates rings is injective) but in general not an isomorphism. So in some
sense π qf (X, x) is bigger than π(X, x) and, for a fixed point s ∈ S, the morphism
π(Xs , xs ) → π qf (X, x)s
is more likely to be an isomorphism than
π(Xs , xs ) → π(X, x)s
it can thus be easier for π qf (X, x), though not easy, to obtain information about
it from its fibres.
4. Any Dimension
We have seen that when the base scheme S has dimension 1 the whole picture
becomes more complicated. But there is at least a very useful property that we
used a lot: over a Dedekind domain a finitely generated module is flat if and
only if it is torsion free. This is no longer true when dim(S) ≥ 2 so to build
a fundamental group scheme classifying all the (quasi-)finite torsors might be
very hard. From now until the end of the paper, unless stated otherwise, S will
be a locally Noetherian regular scheme, thus in particular when dim(S) = 0
(resp. dim(S) = 1) then S is the spectrum of a field (resp. a Dedekind scheme).
What we are suggesting in the reminder of these short notes is a new approach
to study the existence of the fundamental group scheme for such a base scheme
S. As fqpc torsors can be difficult to study globally in this new setting we first
introduce new torsors, for a new Grothendieck topology, whose description can
be a bit cumbersome, but they will globally behave well. Usual fpqc torsors
will be a particular case. Let us do it by steps.
4.1. The bumpy topology. We are going to define a new Grothendieck

topology (see for instance [Vi05, §2.3] for a friendly introduction). Let f : Y →
X be a faithfully flat morphism of schemes. We recall that f is called a fpqc
morphism if every quasi-compact open subset of X is the image of a quasi-
compact open subset of Y . In [Vi05, Proposition 2.33] it is possible to find
many equivalent definitions for f to be a fpqc morphism. We first need to
introduce a new class of morphisms which will play the role of coverings in our
Grothendieck topology.
Definition 4.1. A S-morphism f : Y → X is called bumpy if for every s ∈ S
the induced morphism fs : Ys → Xs is fpqc whenever Xs is not empty.
Proposition 4.2. Let f : Y → X be an S-morphism. The following are true:
(i) If f is fpqc then it is bumpy.
(ii) If there is an open covering {Vi } of X such that the restriction f −1 (Vi ) →
Vi is bumpy then f is bumpy.
(iii) The composite of bumpy morphisms is bumpy.
(iv) Let i : U → X be any morphism of schemes. If f is bumpy then i∗ (f ) :
Y ×X U → U is bumpy.
(v) If f : Y → X is of finite presentation and for every s ∈ S the induced
morphism fs : Ys → Xs is faithfully flat then f is bumpy.
Proof. These are mainly easy consequences of the definition and [Vi05, Propo-
sition 2.35].
Definition 4.3. The bumpy topology on the category (Sch/S) is the topology
{Ui → U } are collections
in which the coverings of morphisms such that the
induced morphism i Ui → U is bumpy (here i Ui denotes the disjoint union).
It is indeed easy to verify that this defines a Grothendieck topology (see
[Vi05, Definition 2.24]). This gives rise to the bumpy site (Sch/S)bp . It will be
useful to observe that a single bumpy morphism is thus a covering.
Remark 4.4. Every S-group scheme G → S of finite type is bumpy. Indeed
here X = S and for all s ∈ S the restriction fs : Gs → Spec(k(s)) is fpqc.
We need a new definition of torsor whose difference will be clear from the
very beginning, that is even from the concept of trivial torsor.
Definition 4.5. A trivial torsor consists of a S-scheme Y with an action σ :
Y × G → Y together with an invariant arrow f : Y → X, such that there is a
G-equivariant X-morphism ϕ : X ×S G → Y with the property that for every
s ∈ S, ϕs : Xs ×k(s) Gs → Ys is an isomorphism such that fs ◦ ϕs = pr1 .
10 Marco Antei
Now we give an intuitive definition of a bumpy torsor and in Proposition

4.7 we will give a topological description in the bumpy site:
Definition 4.6. Let S be a scheme and G an affine S-group scheme. Let Y
be a S-scheme with an action σ : Y × G → Y and f : Y → X an affine bumpy
G-invariant S-morphism of schemes (i.e. f ◦ σ = f ◦ pr1 where pr1 : Y × G → Y
is the first projection). Then f : Y → X is said to be a bumpy G-torsor if for all
s ∈ S the canonical morphism Y ×S G → Y ×X Y is a fibrewise isomorphism,
that is the induced morphism (Y ×S G)s → (Y ×X Y )s is an isomorphism.
This is equivalent as saying that for an affine S-group scheme G, a S-scheme
Y with an action σ : Y × G → Y , an affine bumpy G-invariant S-morphism
f : Y → X is a G-torsor if and only if fs : Ys → Xs is a Gs -torsor for all s ∈ S.
Proposition 4.7. Let S be a scheme and G an affine S-group scheme. A S-
scheme Y with an action σ : Y × G → Y and a G-invariant S-morphism of
schemes Y → X is a bumpy G-torsor if and only if there exists a covering
{Ui → X} in the bumpy site (Sch/S)bp with the property that for each i the
arrow pr1 : Ui ×X Y → Ui is a trivial torsor (i.e. it is a torsor for the bumpy
topology).
Proof. On one direction we assume that Y → X is a G-torsor i.e. the canonical

morphism Y ×S G → Y ×X Y is a fibrewise isomorphism, that is for every s ∈ S
it induces an isomorphism Ys ×k(s) Gs → Ys ×Xs Ys . Since f : Y → X is itself a
covering in the bumpy site (Sch/S)bp then this implies that f is a G-torsor for
the bumpy topology. On the other direction let us assume that f : Y → X is a
G-torsor for the bumpy topology then there exists a covering {Ui → X} in the
bumpy site (Sch/S)bp such that for each i the arrow pr1 : Ui ×X Y → Ui is a
trivial torsor, which implies that for every s ∈ S, Ui,s ×Xs Ys is isomorphic to
Ui,s ×k(s) Gs . Now {Ui,s → Xs } is a fpqc covering so the previous isomorphism
says that the canonical induced morphism Ys ×k(s) Gs → Ys ×Xs Ys is an
isomorphism, which is enough to conclude.
Every G-invariant morphism Y → X which is a torsor for the fpqc topology

is also a torsor for the bumpy topology. Reciprocally a flat morphism Y → X
which is a bumpy torsor for a flat S-group scheme G is nothing but a fpqc
morphism (by the critère de platitude par fibres the morphism ϕ in Definition
4.5 becomes an isomorphism) so in particular a fpqc torsor. From now on we
will simply say torsor instead of bumpy torsor unless we will need to stress the
difference.
Furthermore, unless stated otherwise, by quasi-finite morphism we simply
mean a morphisms of finite type and with finite fibers. An affine quasi-finite (non
flat) group scheme over S can have a fiber at a closed point of order higher then
the generic fiber (or other closed points), whence the adjective bumpy which
stress the non-flatness of certain morphisms. Hereafter some examples of such

group schemes (in order to have non-trivial examples, S clearly cannot be the
spectrum of a field).
Example 4.8. Let R be a discrete valuation ring with fraction and residue
field respectively denoted by K and k. We assume the latter to be of positive
characteristic p. The letter π will denote an uniformising element. We set
R[x] R[x]
G := Spec and H := Spec ;
xp − x, πx xp , πx
they are quasi finite R-group schemes of finite type when provided with co-
multiplication, counit and coinverse given respectively by Δ(x) := x ⊗ 1 + 1 ⊗
x; ε(x) := 0; S(x) := −x. We immediately observe that neither G nor H are
S-flat: indeed x is, in both cases, a R-torsion element. Their generic fibers are
trivial K-group schemes while the special fibers Gs and Hs (of G and H respec-
tively) are isomorphic to (Z/pZ)k and αp,k respectively. When char(K) = p
then G can be easily recovered as the kernel of the morphism (Z/pZ)R → M1
(sending x
→ πx) where Mn := Spec xp −πR[x]
n(p−1) x is the finite and flat R-group
scheme defined in [Ma03, §3.2], so in particular G is finite and not just quasi-
finite. In a similar way H can be recovered as the kernel of the morphism
M1 → M2 (sending x
→ πx).
Remark 4.9. Let R be a discrete valuation ring of positive equal characteristic
p. Let Mn be as in Example 4.8. We define Hij := ker(ϕ : Mi → Mj )
(where ϕ# : x
→ π j−i x)). We observe, for instance, that there is a natural
group scheme morphism u : H12 → H13 whose corresponding morphism on
the coordinate rings is given by u# : xpR[x] R[t]
,π 2 x → tp ,πt , x
→ t. It is immediate
to observe that for all s ∈ Spec(R) the restriction us : H12,s → H13,s is an
isomorphism though u is not an isomorphism: indeed u# is not injective. This
phenomenon cannot happen in the flat schemes world.
Remark 4.10. Notations being as in Remark 4.9 we observe that H13 can be
both seen as a trivial H13 -torsor and as a trivial H12 -torsor. This phenomenon
is not entirily new as even in the easiest case where X = S is the spectrum
of a positive characteristic field then a μp -trivial torsor can also be seen as a
αp -trivial torsor.
These short examples show that it is easy and very natural to encounter
non flat quasi-finite group schemes.
4.2. The bumpy fundamental group scheme. Let S be any

scheme and X → S a faithfully flat morphism of schemes of finite type. We
assume the existence of a section x ∈ X(S). Even in this new setting by pointed
12 Marco Antei
torsor we will mean a G-torsor Y → X endowed with a section y ∈ Yx (S). Let

Z → X be a H-torsor pointed in z ∈ Zx (S); a morphism of pointed torsors from
Y → X to Z → X is the data of two morphisms α : G → H and β : Y → Z
where α is a morphism of group schemes, β(y) = z and the following diagram
commutes:
Y
σY
/ Y ×G
β β×α

/
Z σZ Z × H
where σY and σZ denote the actions of G on Y and H on Z respectively.

Though in §4.1 we gave a very general definition for non flat torsors we do
not need to work in that generality: our purpose is indeed to work in the
smallest possible cofiltered full subcategory of the category of bumpy pointed
torsors containing fpqc torsors; since we do not know, at this stage, whether the
category of (quasi-)finite fpqc torsors itself is cofiltered we suggest a different
candidate and we will prove that it is a cofiltered category. This is why from
now on we only consider the following situation:
Definition 4.11. Let G → S be a quasi finite, affine group scheme of finite
type and Y → X a bumpy G-torsor. Such a quasi-finite torsor will be often
denoted by (Y, G, y). We will denote by Qf (X, x) the category of all those quasi-
finite pointed torsors which also satisfy the following property: the canonical
fibrewise isomorphism Y ×S G → Y ×X Y is (globally) an isomorphism. The
full subcategory of finite pointed torsors (i.e. G → S is finite) will be denoted
by F(X, x).
The following theorem holds even without the stronger assumption on tor-
sors made in 4.11, but, again, we stress that not only we do not need that
generality but we also prefer to restrict as much as possible our category to get
closer to the category of fpqc torsors.
Theorem 4.12. Let X be a scheme and X → S a faithfully flat morphism of
finite type with the property that for all s ∈ S the fiber Xs is reduced and con-
nected. Let x ∈ X(S) be a section. Then the categories Qf (X, x) and F(X, x)
are cofiltered.
Proof. The proof for the two categories being essentially the same here we only
consider Qf (X, x). Since Qf (X, x) has a final object, it is sufficient to prove
that given three objects (Yi , Gi , yi ), i = 0, 1, 2 of Q(X, x) and two morphisms
ϕi : (Yi , Gi , yi ) → (Y0 , G0 , y0 ), i = 1, 2, there exists a forth object (Y3 , G3 , y3 )
and two morphisms ψi : (Y3 , G3 , y3 ) → (Yi , Gi , yi ), i = 1, 2 such that ϕ1 ◦ ψ1 =
ϕ2 ◦ ψ2 . We simply take the fiber product (Y3 , G3 , y3 ) := (Y2 ×Y0 Y1 , G2 ×G0
G1 , y2 ×y0 y1 ). We need to prove that it belongs to Qf (X, x). The assumption

that for all s ∈ S the fiber Xs is reduced and connected is required to ensure
that the category Q(Xs , xs ) of finite pointed (fpqc) torsors is cofiltered. For
this reason the fiber product Y3 is bumpy over X. It is now immediate to
observe that Y3 → X is a bumpy G3 -torsor following Proposition 4.7. This is
however not sufficient to conclude as we still need to prove that the fiberwise
isomorphism Y3 ×S G3 → Y3 ×X Y3 is actually an isomorphism: it is enough to
pull back over Y3 the square given by Y3 , Y2 , Y1 , Y0 and the conclusion follows
from the fact that Y2 , Y1 , Y0 belong to Qf (X, x).
Corollary 4.13. Let X → S be a surjective morphism of finite type with the

property that for all s ∈ S the fiber Xs is reduced and connected. Then there
exist a S-group scheme π Bqf (X, x) and a (pointed in xBqf over x) π Bqf (X, x)-
torsor X Bqf → X universal in the following sense: for every object (Y, G, y) of
Qf (X, x) there is a unique X-morphism of (pointed) torsors X Bqf → Y .
Proof. Very naturally from Theorem 4.12 we have, writing shortly,
(X Bqf , π Bqf (X, x), xBqf ) := lim(Yi , Gi , yi )

←−
i
the limit running through all the objects of Qf (X, x), which exists as the
morphisms are affine.
Definition 4.14. We call π Bqf (X, x) the bumpy fundamental group scheme.
In a similar way we obtain
(X B , π B (X, x), xB ) := lim(Yi , Gi , yi )

←−
i
the limit running through all the objects of F(X, x). This triple is universal in
the obvious sense already mentioned. The S-affine group scheme π B (X, x) will
be called the finite bumpy fundamental group scheme.
It is worth repeating that it is not known whether the fpqc pointed finite
(resp. quasi-finite) torsors form a cofiltered category for such a general base
scheme S, but all of them are already inside F(X, x) (resp. Qf (X, x)) and
so provided the fundamental group scheme π(X, x) classifying all the finite
pointed fpqc torsors (resp. the quasi-finite fundamental group scheme π qf (X, x)
classifying all the quasi-finite pointed fpqc torsors) exists, there would be a
natural morphism π B (X, x) → π(X, x) (resp. π Bqf (X, x) → π qf (X, x)).
14 Marco Antei
5. Over a Dedekind Scheme (Revisited)

This section is meant to show that our construction is not esoteric: indeed the
bumpy fundamental group scheme of a scheme X coincides with the already
existing ones, whenever comparable. It is immediate from the definition of
bumpy torsor that the bumpy fundamental group scheme of a scheme X defined
over a field is nothing but the fundamental group scheme of Nori, described in
these notes in §2. Here we show that if the base scheme S is a Dedekind scheme
then, when comparable, the bumpy fundamental group scheme of X coincides
with the “usual” fundamental group scheme described in §3. The reason is that
in this case we can prove (and we actually will in few lines) that every quasi-
finite bumpy torsor is preceded by a quasi-finite fpqc torsor. And, also, every
finite bumpy torsor is preceded by a finite fpqc torsor.
Theorem 5.1. Let S be a Dedekind scheme, X a faithfully flat S-scheme of

finite type and x ∈ X(S) a section. Let us moreover assume that one of the
following assumptions is satisfied:
1. for every s ∈ S the fibre Xs is reduced;
2. X is integral and normal and that for each s ∈ S the fibre Xs is normal
and integral
then if (1) is satisfied the natural morphism π B (X, x) → π(X, x) is an iso-

morphism and if moreover (2) is satisfied the natural morphism π Bqf (X, x) →
π qf (X, x) is an isomorphism too.
Proof. The proof is similar to (and can be deduced from) [AEG15], Proposition
4.2, Proposition 5.2, Proposition 5.5 and we leave to the reader all the details.
Here we only prove that π B (X, x) → π(X, x) is an isomorphism when S =
Spec(R) where R is a discrete valuation ring (this is thus a subcase of case
(1)): it is sufficient to prove that any object of F(X, x) is preceded by a finite
fpqc torsor, that means that if Y → X is a finite bumpy G-torsor then there
exist a finite and flat R-group scheme H and a finite fpqc H-torsor Z → X
and a morphism (of pointed torsors) Z → Y . So let Yη be, as usual, the generic
fibre of Y : it is a finite fpqc Gη -torsor over Xη . We claim that the Zariski
closure Yη of Yη in Y is a fpqc-torsor over X under the action of the R-finite
and flat group scheme obtained as the Zariski closure Gη of Gη in G. Following
[EGAIV.2] (2.8.3) we deduce an action Yη ×S Gη → Yη compatible with the
action of G on Y . Thus, in particular, the canonical morphism
u : Yη ×S Gη → Yη ×X Yη
complete the following diagram:
Yη ×S Gη
u / Yη ×X Yη
_ _
i j

Y ×S G
v / Y ×X Y
where v is an isomorphism, i and j are closed immersion so u is a closed

immersion too.
The huge difference between G and Gη is that G may not be R-flat, while
G is certainly R-flat. Thus according to [SGA3] Exposé V, Thorme 7.1, the
quotient Yη /Gη exists and it is represented by a scheme such that p : Yη →
Yη /Gη is faithfully flat and the morphism Y → X factors through λ : Yη /Gη →
X that we study below. First we observe that λ is separated: since p : Yη →
Yη /Gη is proper we just need to consider the following commutative diagram:
Δ2
/ Y η × X Yη
Yη
p p×p

Yη /Gη / Yη /Gη ×X Yη /Gη
Δ1
where Δ2 is a closed immersion as Yη → X is propre (by assumption, cf.

Definition 4.11, Y → X is propre). This diagram implies
Δ1 (Yη /Gη ) = Δ1 (p(Yη )) = (p × p)(Δ2 (Yη ))
which is closed in Yη /Gη ×X Yη /Gη because (p × p)Δ2 is propre. So finally λ

is separated ([Ha77], II, Corollary 4.2). But p is surjective, λ ◦ p is proper (it is
actually the composition of Y → X and the closed immersion Yη → Y ); hence
λ is propre too ([Li02], Ch 3, Prop 3.16 (f)). Furthermore we observe that for
every x ∈ X the set λ−1 (x) is finite so λ, which is proper and quasi-finite,
is in particular finite ([EGAIV.3] Théorème 8.11.1), hence affine. We did not
finish yet, as we want to prove that λ an isomorphism we first start by the
surjectivity: the image of λ in X contains the image of λ ◦ p which is dense in
X; furthermore, the properness of λ ensures that λ(Yη /Gη ) is closed in X. We
deduce from this the equality λ(Yη /Gη ) = X. In order to conclude that λ is
an isomorphism we need to observe that λs : (Yη /Gη )s → Xs is surjective too,
but Xs being reduced then λ# s : OXs → (fs )∗ (OYη /Gη ) is injective ([EGAI],
Corollaire 1.2.7); according to [WW80], Lemma 1.3, the morphism λ, which
is affine and is an isomorphism generically, is an isomorphism globally. This
proves that Yη → X is a fpqc finite Gη -torsor. It is of course pointed if Y is
pointed which concludes the proof once we set H := Gη and Z := Yη .
16 Marco Antei
In particular when S is a Dedekind scheme π B (X, x) and π Bqf (X, x) (for

X as in the statement of Theorem ) are S-flat. We do not know if this remains
true when S has higher dimension.
Acknowledgments
We would like to thank A. Aryasomayajula, I. Biswas, A. Morye, A.
Parameswaran and T. Sengupta for the wonderful meeting that they organ-
ised in both Tata Institute of Fundamental Research (TIFR, Mumbai) and
University of Hyderabad (UoH) and for inviting me to actively participate to
this event. We would also like to thank the referee for his very useful comments
and remarks.
References
[AEG15] M. Antei, M. Emsalem, C. Gasbarri, Sur l’existence du schéma en groupes
fondamental, arXiv:1504.05082v2 [math.AG]
[BoVi] N. Borne, A. Vistoli , The Nori fundamental gerbe of a fibered category,
Journal of Algebraic Geometry (to appear).
[DeMi82] P. Deligne, J. S. Milne, Tannakian Categories, in Hodge Cycles, Motives,
and Shimura Varieties, Lectures Notes in Mathematics 900, Springer-
Verlag, (1982), 101-227.
[SGA3] M. Demazure, A. Grothendieck Schémas en groupes. I: Propriétés
générales des schémas en groupes. Séminaire de Géométrie Algébrique
du Bois Marie 1962/64 (SGA 3). Lecture Notes in Mathematics, Vol. 151
Springer-Verlag, Berlin-New York 1970
[EGAI] A. Grothendieck, Éléments de géomérie algébrique. I. Le langage des
schémas. Publications Mathématiques de l’IHÉS, 4, (1960).
[EGAIV.2] A. Grothendieck, Éléments de géomérie algébrique. IV. Étude locale des
schémas et des morphismes de schémas. 2, Publications Mathématiques
de l’IHÉS, 24, (1965).
[EGAIV.3] A. Grothendieck, Éléments de géomérie algébrique. IV. Étude locale des
schémas et des morphismes de schémas. 3, Publications Mathématiques
de l’IHÉS, 28 (1966).
[Ga03] C. Gasbarri, Heights of vector bundles and the fundamental group scheme
of a curve. Duke Math. J. 117 (2003), no. 2, 287–311.
[SGA1] A. Grothendieck, Revêtements étales et groupe fondamental, Séminaire de
géométrie algébrique du Bois Marie, (1960-61).
[Ha77] R. Hartshorne, Algebraic Geometry, GTM, Springer Verlag (1977).
[Li02] Q. Liu, Algebraic geometry and arithmetic curves, Oxford Science Publi-
cations (2002).
[Ma03] S. Maugeais, Relèvement des revêtements p-cycliques des courbes ra-
tionnelles semi-stables, Math. Ann. 327 (2003), no. 2, 365–393.
[No76] M. V. Nori, On the representations of the fundamental group, Compos.
Math. 33 (1976), 29–42.
[No82] M. V. Nori, The fundamental group-scheme, Proc. Ind. Acad. Sci. (Math.
Sci.) 91 (1982), 73–122.
[Sz09] T. Szamuely, Galois groups and fundamental groups. Cambridge Studies
in Advanced Mathematics, 117. Cambridge University Press, Cambridge,
2009.
[Vi05] A. Vistoli, Grothendieck topologies, fibered categories and descent theory.
Fundamental algebraic geometry, 1–104, Math. Surveys Monogr., 123,
Amer. Math. Soc., Providence, RI, 2005.
[Wa79] W. C. Waterhouse, Introduction to affine group schemes. Graduate Texts
in Mathematics, 66. Springer–Verlag, New York–Berlin, 1979
[WW80] W.C. Waterhouse, B. Weisfeiler, One-dimensional affine group schemes,
Journal of Algebra, 66, 550-568 (1980).
Heat Kernels, Bergman Kernels, and
Cusp Forms
Anilatmaja Aryasomayajula∗
Abstract
In this article, we describe a geometric method to study cusp forms, which relies
on heat kernel and Bergman kernel analysis. This new approach of applying
techniques coming from analytic geometry is based on the micro-local analysis
of the heat kernel and the Bergman kernel from [3] and [2], respectively, using
which we derive sup-norm bounds for cusp forms of integral weight, half-integral
weight, and real weight associated to a Fuchsian subgroup of first kind.
Mathematics Subject Classification (2010). 11F11, 11F37, 30C40
1. Introduction
This is both a survey and a research article elucidating heat kernel and Bergman
kernel methods for studying cusp forms. We describe a geometric approach of
Bouche and Berman to study sup-norm bounds for sections of a positive line
bundle defined over a compact complex manifold. We then apply these methods
to study cusp forms associated to a Fuchsian subgroup of first kind, which
yields optimal results when the Fuchsian subgroup is cocompact. However, this
approach does not give optimal results when the Fuchsian subgroup is cofinite.
But an extension of the methods of Bouche to cuspidal neighborhoods should
allow one to derive optimal bounds for cusp forms, even when the Fuchsian
subgroup is cofinite.
∗ Anilatmaja Aryasomayajula, Department of Mathematics, IISER Tirupati, Karakambadi
Road, Tirupati-517507, India. E-mail: [email protected]
Geometry, DOI 10.1007/978-981-10-5648-2_2
20 Anilatmaja Aryasomayajula
1.1. Notation. Let C denote the complex plane. For z ∈ C, let x = Re(z)
and y = Im(z) denote the real and imaginary parts of z, respectively. Let
H = {z ∈ C| y = Im(z) > 0}
be the upper half-plane. Let Γ ⊂ PSL2 (R) be a Fuchsian subgroup of the first
kind acting by fractional linear transformations on H. We assume that Γ admits
no elliptic elements.
Let X be the quotient space Γ\H of genus g > 1. The quotient space X
admits the structure of a hyperbolic Riemann surface of finite volume. We allow
X to have genus g = 1, if X is not compact.
We denote the (1, 1)-form corresponding to the hyperbolic metric of X,
which is compatible with the complex structure on X and has constant negative
curvature equal to minus one, by μhyp (z). Locally, for z ∈ X, it is given by
i dz ∧ dz
μhyp (z) = · .
2 Im(z)2
Let μshyp (z) denote the rescaled hyperbolic metric μhyp (z)/ volhyp (X), which
measures the volume of X to be one.
For k ∈ R>0 , let ν denote the factor of automorphy of weight k with the
associated character being unitary. Let S k (Γ, ν) denote the complex vector
space of weight-k cusp forms with respect to Γ and ν. Let {f1 , . . . , fjk } denote
an orthonormal basis of S k (Γ, ν) with respect to the Petersson inner product.
Then, for z ∈ X, put
k ,ν

jk
BX (z) := y k |fi (z)|2 .
i=1
k ,ν k
When the associated character is trivial, we put BX (z) = BX (z).
k ,ν
1.2. Sup norm bounds for the function BX (z). Let the Fuch-
sian subgroup Γ be cocompact, i.e. X is a compact Riemann surface. With
notation as above, for k ∈ 12 Z (or 2Z), we have the following estimate
1 k
lim sup BX (z) = O(1), (1)
k z∈X k
where the implied constant is independent of Γ.

Furthermore, for a fixed k ∈ R>0 , let ν denote the factor of automorphy of
weight k with the associated character being unitary. Then, with notation as
above, we have the following estimate
1 nk ,ν n
lim sup B (z) = O(1), (2)
n z∈X nk X
where n ∈ Z, and the implied constant is independent of Γ.

Heat Kernels, Bergman Kernels, and Cusp Forms 21
Let Γ now be a cofinite subgroup, i.e., X is a noncompact hyperbolic Rie-

mann surface of finite volume, and let A be a compact subset of X. Then,
with notation as above, for k ∈ 12 Z (or 2Z) and z ∈ A, we have the following
estimate
1 k
lim B (z) = OA (1), (3)
k k X
where the implied constant depends on A.
Furthermore, for a fixed k ∈ R>0 , let ν denote the factor of automorphy of
weight k with the associated character being unitary. Let A be any compact
subset of X. Then, with notation as above, for any z ∈ A, we have the following
estimate
1 nk ,ν n
lim B (z) = OA (1), (4)
n nk X
where n ∈ Z and the implied constant depends on A.
Our estimates (1), (2), (3), and (4) are optimal. However, when X is non-
compact we cannot extend estimates (3) and (4) to the entire Riemann surface,
i.e., our method does not yield optimal estimates when A is equal to X. How-
ever, an extension of Bouche’s methods to cuspidal neighborhoods will enable
the extension of estimates (3) and (4) to X.
Lastly, our methods extend with notational changes to higher dimensions,
namely to Hilbert modular cusp forms and Siegel modular cusp forms.
1.3. Existing results on sup-norm bounds for the func-

k ,ν
tion BX (z). In [6], using heat kernel analysis, Jorgenson and Kramer
2
derived sup-norm bounds for the the Bergman kernel BX (z), associated to
any hyperbolic Riemann surface X (compact or noncompact of finite volume).
The bounds of Jorgenson and Kramer are optimal. Especially for the case
X = Y0 (N ), they derived
sup BY2 0 (N ) (z) = O(1),

z∈Y0 (N )
where the implied constant does not depend on the modular curve Y0 (N ).
In [5], extending their method from [6], Jorgenson, Kramer, and Friedman
k
derived sup-norm bounds for the the Bergman kernel BX (z), associated to
any hyperbolic Riemann surface X (compact or noncompact of finite volume).
When X is a compact hyperbolic Riemann surface, they showed that
k
sup BX (z) = O(k),
z∈X
where the implied constant is independent of the Riemann surface X. When

X is a noncompact hyperbolic Riemann surface of finite volume, they showed
that
3
k
sup BX (z) = O(k 2 ),
z∈X
where the implied constant is independent of the Riemann surface X. The

estimates of Jorgenson and Kramer are optimal, as shown in [6].
It is possible to extend the heat kernel analysis of Jorgenson and Kramer
to higher dimensions, namely to Hilbert modular cusp forms and Siegel mod-
ular cusp forms of both integral and half-integral weight. However, one has to
address certain non trivial convergence issues, while doing so.
For k ∈ 12 Z and N ∈ N, let f be any weight-k cusp form with respect to the
arithmetic subgroup Γ0 (4N ). Furthermore, let f be normalized with respect
to the Petersson inner-product. Then, in [4], Kiral has derived the following
estimate
1 1
sup y k |f (z)|2 = Ok,ε N 2 − 18 +ε ,
z∈Y0 (N )
for any ε > 0. Using above estimate, one can derive

1 1
sup BY0 (N ) (z) = Ok,ε N 2 − 18 +ε ,
z∈Y0 (N )
for any ε > 0.

k
For k ∈ R>0 with k > 2, the Bergman kernel BX (z) can be represented by
an infinite series, which is uniformly convergent in z ∈ X. Using which, Steiner
has extended the bounds of Jorgenson and Kramer to real weights.
Let Γ any subgroup of finite index in SL2 (Z), and for a fixed k ∈ R>0 with
k 1, let ν denote the factor of automorphy of weight k with the associated
character being unitary. Furthermore, let A be a compact subset of X. Then,
in [7], Steiner has derived the following estimates
k ,ν
sup BX (z) = OA (k),
z∈A
where the implied constant depends on the compact subset A; and

k ,ν 3
sup BX (z) = OX k 2 ,
z∈X
where the implied constant depends on X.

2. Heat Kernels and Bergman Kernels on

Compact Complex Manifolds
In this section, we recall the main results from [3] and [2], which we use in the
next section.
Let (M, ω) be a compact complex manifold of dimension n with a Hermitian
metric ω. Let L be a positive Hermitian holomorphic line bundle on M with
the Hermitian metric given by s(z)2L := e−φ(z) |s(z)|2 , where s ∈ L is any
section, and φ(z) is a real-valued function defined on M .
∗
For any k ∈ N, let k := (∂ +∂)2 denote the ∂-Laplacian acting on smooth
⊗k k
sections of the line bundle L . Let KM ,L (t; z, w) denote the smooth kernel of
2t
the operator e− k k . We refer the reader to p. 2 in [3], for the details regarding
k
the properties which uniquely characterize the heat kernel KM ,L (t; z, w). When
k
z = w, the heat kernel KM ,L (t; z, w) admits the following spectral expansion
2t k
k
KM ,L (t; z, w) = e− k λn ϕn (z) ⊗ ϕ∗n (w), (5)
n≥0
where {λkn }n≥0 denotes the set of eigenvalues of k (counted with multiplici-
ties), and {ϕn }n≥0 denotes a set of associated orthonormal eigenfunctions.
Let {si } denote an orthonormal basis of H 0 (M, L⊗k ). For any z ∈ M , the
Bergman kernel is given by

k
BM ,L (z) := si (z)2L⊗k . (6)
i
For any z ∈ M and t ∈ R>0 , from the spectral expansion of the heat kernel
k
KM ,L (t; z, w) described in equation (5), it is easy to see that
k k k k
BM ,L (t; z) ≤ KM ,L (t; z, z) and lim KM ,L (t; z, z) = BM ,L (t; z). (7)
t
For z ∈ M , let c1 (L)(z) := 2π i

∂∂φ(z) denote the first Chern form of the line
bundle L. Let α1 , . . . , αn denote the eigenvalues of ∂∂φ(z) at the point z ∈ M .
Then, with notation as above, from Theorem 1.1 in [3], for any z ∈ M and
t ∈ (0, k ε ), and for a given ε > 0 not depending on k, we have
1 k n
αj
lim n
K M ,L (t; z, z) = n
, (8)
k k j=1
(4π) sinh(αj t)
and the convergence of the above limit is uniform in z.

Using equations (7) and (8), in Theorem 2.1 in [3], Bouche derived the
following asymptotic estimate
1 k
lim B (z) = O detω c1 (L)(z) , (9)
k k n M ,L
where the implied constant does not depend on X, and the convergence of the
above limit is uniform in z ∈ X.
When M is a noncompact complex manifold, using micro-local analysis of
the Bergman kernel, in [2], Berman derived the following estimate
1 k
lim sup n
BM ,L (z) ≤ detω c1 (L)(z) .
k k
Furthermore, let A be any compact subset of M . Then, for any z ∈ A, from
the proof of Corollary 3.3 in [2], we have

1 k
lim n BM ,L (z) = OA detω c1 (L)(z) , (10)
k k
where the implied constant depends on the compact subset A.
3. Estimates of Cusp Forms

In this section, using results from previous section, we prove estimates (1), (2),
(3), and (4).
Let notation be as in Section 1. Let ΩX denote the cotangent bundle over
X. Then, for any k ∈ 2Z, cusp forms of weight k with respect to Γ are global
⊗k/2
section of the line bundle ΩX . Furthermore, recall that for any f ∈ ΩX , i.e.,
f a weight-2 cusp form, the Petersson metric on the line bundle ΩX is given by
f (z)2ΩX := y 2 |f (z)|2 . (11)
Let ωX denote the line bundle of cusp forms of weight 12 over X. Then, for any
k ∈ 12 Z, cusp forms of weight-k with respect to Γ are global section of the line
⊗2k
bundle ωX . Furthermore, recall that for any f ∈ ωX , i.e., f a weight- 12 cusp
form, the Petersson metric on the line bundle ωX is given by
f (z)2ωX := y 1/2 |f (z)|2 . (12)
Remark 1. For any z ∈ X and k ∈ 2Z, from the definition of the Bergman
k/2 ⊗k/2
kernel BX,ΩX (z) for the line bundle ΩX from equation (6), we have
k/2 k
BX,ΩX (z) = BX (z).
Similarly, for any z ∈ X and k ∈ 12 Z, from the definition of the Bergman kernel
⊗2k
BX,Ω
2k
X
(z) for the line bundle ωX from equation (6), we have
k
BX,ω
2k
X
(z) = BX (z). (13)
Theorem 2. Let Γ be cocompact, i.e., X is a compact hyperbolic Riemann

surface. Then, with notation as above, for k ∈ 12 Z (or 2Z), we have the following
estimate
1 k
lim sup B (z) = O(1),
k z∈X k X
where the implied constant is independent of Γ.
Proof. We refer the reader to Theorem 2 in [1] for the proof of the theorem. We
briefly describe the proof of the theorem for k ∈ 12 Z, and the case for k ∈ 2Z,
follows automatically with notational changes. For any z ∈ X, observe that
⊗2 1
c1 (ωX )(z) = μhyp (z),
4π
⊗2
⊗2
1
which shows that the line bundle ωX is positive, and detμhyp c1 (ωX )(z) = 4π .
Using equation (13), and applying estimate (9) to the complex manifold X with
its natural Hermitian metric μhyp and the line bundle ωk⊗2k , we find

1 k 1 2k ⊗2

lim BX (z) = lim BX,ω X
(z) = O det μhyp 1c (ω X )(z) = O(1).
k k k k
As the above limit convergences uniformly in z ∈ X, and as X is compact, we

have
1 k 1 k
sup lim B (z) = lim sup BX (z) = O(1),
z∈X k k X k z∈X k
which completes the proof of the theorem.

Corollary 3. Let Γ be cofinite, i.e., X is a noncompact hyperbolic Riemann
surface of finite volume, and let A be a compact subset of X. Then, with nota-
tion as above, for k ∈ 12 Z (or 2Z), and z ∈ A, we have the following estimate
1 k
lim B (z) = OA (1),
k k X
where the implied constant depends on A.
Proof. The proof of the theorem follows from estimate (10), and from similar
arguments as in Theorem 2.
Remark 4. For a fixed k ∈ R>0 , and let ωX,k,ν denote the line bundle of
weight-k cusp forms with the factor of automorphy ν, and associated character
being unitary. Then, for any n ∈ Z, cusp forms of weight-nk with respect to Γ
⊗n
and ν n are global section of the line bundle ωX,k,ν n . Furthermore, recall that
for any f ∈ ωX,k,ν , the Petersson metric on the line bundle ωX,k,ν is given by
f (z)2ωX,k,ν := y k |f (z)|2 . (14)
For any z ∈ X and n ∈ Z, from the definition of the Bergman kernel

⊗n
BX,ω
n
X,k,ν n
(z) for the line bundle ωX,k,ν n from equation (6), we have
n
nk ,ν
BX,ω
n
X,k,ν n
(z) = BX (z). (15)
Theorem 5. Let Γ be cocompact, i.e., X is a compact hyperbolic Riemann

surface. For a fixed k ∈ R>0 , let ν denote the factor of automorphy of weight k
with the associated character being unitary. Then, with notation as above, we
have the following estimate
1 nk ,ν n
lim sup B (z) = O(1),
n z∈X nk X
where n ∈ Z, and the implied constant is independent of Γ.
Proof. From equation (14), for any z ∈ X, observe that
i k
c1 ωX,k,ν (z) = − ∂∂ log y k |f (z)|2 = μhyp (z),
2π 4π
which shows that the line bundle ωX,k,ν is positive, and
k
detμhyp c1 (ωX,k,ν )(z) = 4π . Using equation (15), and applying estimate
(9) to the complex manifold X with its natural Hermitian metric μhyp and the
⊗n
line bundle ωX,k,ν n , we find

1 nk ,ν n 1 n 1
lim BX (z) = lim BX,ωX,k,ν n (z) = O detμhyp c1 ωX,k,ν (z) = O(1).
n nk n nk k
As the above limit convergences uniformly in z ∈ X, and as X is compact, we

have
1 nk ,ν n 1 nk ,ν n
sup lim B (z) = lim sup B (z) = O(1),
z∈X n nk X n z∈X nk X
which completes the proof of the theorem.

Corollary 6. Let Γ be cofinite, i.e., X is a noncompact hyperbolic Riemann

surface of finite volume. For a fixed k ∈ R>0 , let ν denote the factor of au-
tomorphy of weight k with the associated character being unitary. Let A be a
compact subset of X. Then, with notation as above, for any z ∈ A, we have the
following estimate
1 nk ,ν n
lim B (z) = OA (1),
nk nk X
where n ∈ Z, and the implied constant depends on A.
Proof. The proof of the theorem follows from estimate (10), and from similar
arguments as in Theorem 5.
Acknowledgements
Firstly, the author would like to thank the hospitality of the department of
Mathematics at University of Hyderabad, where the article was realized. The
author would like to thank J. Kramer and J. Jorgenson for introducing him to
the area of automorphic forms and heat kernels.
The author would also like to express his gratitude to T. Bouche for pro-
viding him with helpful references, and to Archana S. Morye, for many helpful
discussions and remarks. The author would also like to thank R. S. Steiner for
his communications.
The author acknowledges the support of INSPIRE research grant
DST/INSPIRE/04/2015/002263.
References
[1] A. Aryasomayajula, Heat kernel approach for sup-norm bounds for cusp forms of
integral and half-integral weight, Archiv der Mathematik 106, (2016), 165–173.
[2] R. J. Berman, Bergman kernels and local holomorphic Morse inequalities, Math.
Z. 248, (2004), 325–344.
[3] T. Bouche, Asymptotic results for Hermitian line bundles over complex mani-
folds: The heat kernel approach, Higher-dimensional complex varieties, 67–81,
de Gruyter, Berlin, 1996.
[4] E. M. Kiral, Bounds on Sup-norms of Half Integral Weight Modular Forms, Acta
Arithmetica, Vol. 165, (2014), 385-399.
[5] J. Friedman, J. Jorgenson, and J. Kramer, Uniform sup-norm bounds on average
for cusp forms of higher weights, To appear in Volume dedicated to the Memory
of Friedrich Hirzebruch, Springer-Verlag, Berlin.
[6] J. Jorgenson and J. Kramer, Bounding the sup-norm of automorphic forms,

GAFA 14, (2004), 1267–1277.
[7] R. S. Steiner, Uniform bounds on sup-norms of holomorphic forms of real weight,
Int. J. Number Theory 12, (2016), 1163–1185.
On a Conjecture of Butler
Usha N. Bhosle∗,†
1. Introduction
Syzygy bundles over smooth curves (as well as higher dimensional smooth va-
rieties) have been studied for several years now. Let L be a line bundle on a
smooth curve X. Given a subspace V of the space of sections of L which gener-
ates L, the kernel ML,V of the evaluation map V ⊗ OX → L is called a Syzygy
bundle or a Kernel bundle or a Lazarfeld bundle. These bundles have several
applications, applications to Syzygy problems, Greens conjectures, Minimal
Resolution conjectures, Theta functions, Picard bundles. They also play an
important role in Brill-Noether theory for higher ranks and coherent systems.
Eighteen years back, D.C. Butler made a conjecture about the semistability
of ML,V for general (L, V ) [15]. The conjecture was proved recently by Peter
Newstead, myself and Leticia Brambila-Paz [8].
In this article, we first present a short survey of previous work on the con-
jecture of Butler and related conjectures. Then we state our main results and
sketch the idea of the proofs. We end with a discussion of some applications of
Kernel bundles and generalisations of some results on Kernel bundles to nodal
and cuspidal curves.
2. Preliminaries
Let X be an integral smooth projective curve of genus g. Let E denote a vector
bundle of rank n on X and p : E → X the projection map. For x ∈ X, the
fibre Ex := p−1 x is a vector space of dimension n. We recall that a section s
∗ Department of Mathematics, Indian Institute of Science, Bangalore 560012, India.
† This article was written during author’s tenure as Raja Ramanna Fellow in Indian In-
stitute of Science, Bangalore. It was finalised during the visit of the author to Instituto de
Ciencias Matematicas under the Indo-European Mary Curie research program MODULI in
June-July 2015. The author thanks both the institutes for excellent working conditions.
Geometry, DOI 10.1007/978-981-10-5648-2_3
30 Usha N. Bhosle
of E is a map X → E such that p ◦ s = idX . Let H 0 (X, E) be the space of

sections of E, it is a finite dimensional vector space.
2.1. (semi)stability of vector bundles. Associated to any vector

bundle E, there is an integer d(E) called the degree of E. Denote by r(E) the
rank of E (r(E) = n). Define slope of E by
μ(E) := d(E)/r(E) .
A sub bundle F of E is a vector bundle F such that F ⊂ E with each fibre

Fx ⊂ Ex for all x ∈ X. A sub bundle F is called proper if F = E.
Definition 2.1. A vector bundle E is called semistable if for every sub bundle
F ⊂ E,
μ(F ) ≤ μ(E) .
A vector bundle F is called stable if for every proper sub bundle F of E,
μ(F ) < μ(E).
2.2. Syzygy Bundle. Let V ⊂ H 0 (X, E) be a vector subspace and X ×V

denote the trivial bundle with fibre isomorphic to V .
Definition 2.2. Define the evaluation map
evV : X × V → E
by (x, s)
→ s(x), x ∈ X, s ∈ V.
If this map is surjective, then E is said to be generated by V .
Definition 2.3. Let L denote a line bundle of degree d on X. Suppose that

the line bundle L is generated by a sub space V of its sections. The kernel of
the map evV , denoted by ML,V , is a vector bundle, called a Syzygy bundle (or
a kernel bundle or Lazarsfeld bundle).
The kernel bundle fits in an exact sequence
0 → ML,V → X × V → L → 0 .
If V = H 0 (X, L), the complete linear system, then ML,V is simply denoted by
ML .
We call the pair (L, V ) a general linear series of type (d, n) if L is a general
element of the Jacobian (of degree d) of X and V is a general subspace of
dimension n of H 0 (X, L).
On a Conjecture of Butler 31
It follows from the exact sequence that if the dimension of V is n + 1, then

the rank r(ML,V ) = n, and if the degree d(L) = d, then d(ML,V ) = −d.
The bundle ML,V has been extensively studied for many years because of
its several applications.
• As its name suggests, it has applications to Syzygy problems, Green’s

Conjectures, Minimal resolution conjecture, Ideal generation conjecture.
• It can be interpreted as the restriction of the cotangent bundle of P(V ).
• It has been studied in connection with Theta divisors,
• and Picard bundles.
• It is also studied in Brill-Noether Theory and coherent systems.
We shall discuss some of these applications in more detail later.
3. Some Known Results on ML

For various applications, the most important question is: Is ML,V stable or
semistable? A few words about why (semi)stability of ML,V is important: If
a bundle is not semistable, then it does not admit a theta divisor. Stability
(or semistability) of certain ML,V ensures that of Picard bundles. Stability of
ML,V has major implications in higher rank Brill-Noether theory also.
What does one know about the (semi)stability of ML,V ? The bundle ML
(i.e. the case V = H 0 (X, L)) is best understood. Recall that n = r(ML,V ) , d =
d(L).
3.1. The bundle ML . We list some interesting results on ML , the list

is not exhaustive.
• Paranjape K. and Ramanan S. [25] deal with the particular but important
case L = KX , the canonial bundle of X. They show that for any curve
X, ML is semistable and it is stable if X is not hyper elliptic (i.e. X is
not a double cover of P1 ).
• Ein and Lazarsfeld [17] proved that for any curve X, the vector bundle
ML is semistable if d ≥ 2g, stable if d > 2g.
Butler [15] generalised their results to higher ranks replacing L by a
vector bundle E of higher rank . He proved that for any curve X, the
vector bundle ME is semistable if μ(E) ≥ 2g and stable if μ(E) > 2g.
32 Usha N. Bhosle
• Camere [16] showed that for any curve X, ML is semistable if d ≥ 2g −

Cliff(X), where Cliff(X) is the Clifford index of X.
Let us recall what Clifford index is. For a line bundle N , let hi (X, N ) :=
dim. H i (X, N ).
Definition 3.1. Let g ≥ 2. Consider integers i, with 0 ≤ i ≤ g − 2,

satisfying the condition
d(N ) − i
h0 (X, N ) ≤ + 1,
2
for all line bundles N on X. The maximum such integer i is called the
Clifford index of X.
One has Cliff(X) ≥ 0 and Cliff(X) = 0 if and only if X is hyper elliptic.

For all curves, Cliff(X) ≤ [(g − 1)/2] and for a general curve, Cliff(X) =
[(g − 1)/2] [25, Remark 2.3]
• We say that L computes Cliff(X) if
Cliff(X) = deg(L) − 2r, h0 (L) ≥ 2, h1 (L) ≥ 2 , r = h0 (L) − 1.
Leticia Brambila-Paz [12, Theorem 1.3], Mistretta and Stoppino [24,

Corollary 5.5], independently proved the following interesting result. Over
any curve X, if L computes Clifford index, then ML is semistable and
stable if X is not hyper elliptic.
4. The Conjectures
Much less was known about ML,V if V is not a complete linear system, we list
a few conjectures about semistability and stability of ML,V .
4.1. The Semistability Conjecture. In [15], D. C. Butler made a

conjecture [15, Conjecture 2] for the more general case where L is replaced
by a semistable vector bundle E and V is any linear subspace of H 0 (E) that
generates E; for the case of a line bundle, this can be stated in the following
form (except that Butler restricts to the case g ≥ 3).
Conjecture 4.1. Semistability Conjecture

For a general curve X of genus g ≥ 1 and a general choice of (L, V ) the
vector bundle ML,V is semistable.
Remark 4.2. (1) A general curve means a curve belonging to a non-specified

Zariski open subset of the moduli space of curves.
(2) It follows from Brill-Noether theory that for a general choice of (L, V ), the
line bundle L on a general curve is generated by V if

g
d≥g+n− ,
n+1

g g
where n+1 denotes the integral part of n+1 .
Butler [15] proved that over a general curve X, ML is always semistable.

He also gave some precise conditions for stability of ML .
4.2. The Stability Conjectures. There are many variants of the sta-
bility conjecture, we consider two of them.
Conjecture 4.3. For a general curve X of genus g ≥ 3 and a general choice

of (L, V ) the vector bundle ML,V is stable.
Definition 4.4. A curve X is called a Petri curve if for any line bundle N on
X, the multiplication map
H 0 (X, N ) ⊗ H 0 (X, KX ⊗ N ∗ ) → H 0 (X, KX )
is injective. Here, N ∗ denotes the dual of N and KX is the canonical bundle

(dual of tangent bundle) of X.
Conjecture 4.5. For a Petri curve X of genus g ≥ 3 and a general choice of

(L, V ) the vector bundle ML,V is stable.
Petri curves form a (specified) Zariski open subset in the moduli space of
curves. Hence a Petri curve is a general curve, the converse is not true.
Thus, Conjecture 4.5 implies Conjecture 4.3 (not conversely).
In Conjectures 4.3 and 4.5, we have assumed that g ≥ 3. Why? What
happens if g ≤ 2 ?
4.3. Case of g ≤ 2. Note that any curve of g ≤ 2 is a Petri curve.

Conjectures 2 and 3 can fail for g ≤ 2.
1. Case g = 0.
For g = 0, X = P1 , the projective line. By a theorem of Grothendieck,
any vector bundle on P1 is isomorphic to a direct sum of line bundles, in
particular it is not stable for rank n ≥ 2.
34 Usha N. Bhosle
2. Case g = 1
For g = 1, the curve is an elliptic curve. By a result of Tu [32], there are
no stable vector bundles of rank n and degree d on an elliptic curve if n
and d are not coprime. If (n, d) = 1, then semistability is equivalent to
stability over any curve. For any elliptic curve, ML,V is stable if and only
if d ≥ n + 1 and (n, d) = 1 [19].
3. Case g = 2
For any curve of g = 2, ML,V is stable if and only if d ≥ n + 2, d = 2n [7].
∗
For d = 2n, E = ML,V is a vector bundle of rank n with degree 2n and
h (X, E) ≥ n + 1. Then by the Riemann-Roch theorem h1 (X, E) ≥ 1 so
0
that there is a nonzero homomorphism E → KX giving a contradiction

to stability.
5. Some Results on ML,V

A lot of work has been done on these conjectures for last seventeen years
by several authors. Different techniques have been employed to solve them,
including deformations of curves, classical Brill-Noether theory and coherent
systems for higher ranks. Many cases have been proved. We list here a few most
general results.
One of the earliest results is due to Ballico and Hein.
Theorem 5.1. [4, Proposition 1.6, Theorem 1.7 and Remark 1.10]
Let X be a general curve of genus g ≥ 2 and (L, V ) a general linear series
of type (d, n + 1) with n ≥ 3.

• If u = n+2g−1
and d ≥ n(u + 1) + 1, then ML,V is semistable.

• If u = n+2g−2
and d ≥ max{n(u + 2) + 1, 3n + 2},
then ML,V is stable.
Aprodu M., Farkas G., and Ortega A. have recently proved the following
general result.
Theorem 5.2. [1, Theorem 1.8].
of type (d, n + 1).
Then ML,V is semistable whenever

d
d+n ≥ 2g + 2n − 2.
n
It is more convenient to fix the genus g of the curve and the number n and
consider conditions on d in terms of g and n. Theorem 5.2 can be restated as
follows.
Restatement of Theorem 5.2 Suppose g = ns + t with 0 ≤ t ≤ n − 1. If
d ≥ max{g + n + min{n − t, t − 2}, g + n}, (5.1)
then ML,V is semistable.

Equivalence of the two statements is not obvious, it needs proof (see [8,
Proposition 2.5] for a proof).
Both the results, and many others, are proved using deformations of smooth
curves to reducible nodal curves. We applied the theory of coherent systems to
prove new results. We finally proved the Semistability Conjecture completely
last year.
5.1. Main Results of Usha Bhosle, Leticia Brambila-Paz,

Peter Newstead. Recall that on a Petri curve of genus g, a linear series
(L, V ) with d(L) = d and dimension V = n + 1 exists if and only if

g
d≥g+n− ,
n+1

g g
where n+1 denotes the integral part of n+1 .
Theorem 5.3. [Conjecture 4.1 is true] [8, Theorem 5.1]

of type (d, n + 1).
Then the vector bundle ML,V is semistable.
Theorem 5.4. [Conjectures 4.3 and 4.5]
Let X be a Petri curve of genus g ≥ 3 and (L, V ) is a general linear series
of type (d, n + 1). Then ML,V is stable for
1. n ≤ 4 [7, Theorems 7.1, 7.2, 7.3].
2. n ≥ 5 and g ≥ 2n − 4 [8, Theorem 6.1].
The proofs of these theorems use the following key result, which extends
the range of values of (g, n, d) for which Conjecture 4.5 holds.
Theorem 5.5. [8, Theorem 4.7]
Suppose that C is a Petri curve, g ≥ 2, n ≥ 2 and

g (n2 − n − 2)g
g+n− ≤d<g+n+ . (5.2)
n+1 2(n − 1)2
36 Usha N. Bhosle
Then, for the general linear series (L, V ) of type (d, n + 1), ML,V is stable.
When g is odd, the inequality < in the hypothesis can be replaced by ≤.
The following corollaries to the main theorems present short lists of possible
exceptions to Conjecture 4.5 for n = 5, 6, 7.
Corollary
g 5.6. [8, Corollary 6.3] Suppose that n = 5, g ≥ 3 and d ≥ g + 5 −
6 .
• If C and (L, V ) are general, then ML,V is stable.
• If C is Petri and (L, V ) is general, then ML,V is stable except possibly in

the following cases:
g = 3, d = 12
g = 4, d = 12, 13, 17, 18
g = 5, d = 13, 18.
Remark 5.7. To complete the proof of Conjecture 4.5 for n = 5, it is sufficient

to prove the following cases: g = 3, d = 12; g = 4, d = 12, 13; g = 5, d = 13.
Corollary
7 .

g = 3, d = 14
g = 4, d = 14, 15, 20, 21
g = 5, d = 14, 15, 16, 20, 21, 22
g = 6, d = 16, 22, 28.
• If C and (L, V ) are general, then ML,V is stable except possibly when
g = 3, d = 14; g = 4, d = 14, 15; g = 5, d = 14, 15, 16 or g = 6, d = 16.
Remark 5.9. To complete the proof of Conjecture 4.5 for n = 6, it is sufficient

to prove the cases g = 3, d = 14; g = 4, d = 14, 15; g = 5, d = 14, 15, 16;
g = 6, d = 16.
Corollary
8 .

g = 3, d = 16
g = 4, d = 16, 17, 23, 24

g = 5, d = 16, 17, 18, 23, 24, 25
g = 6, d = 17, 18, 19, 24, 25, 26, 31, 32, 33
g = 7, d = 18, 19, 25, 26, 32, 33.
• If C and (L, V ) are general, then ML,V is stable.
We could certainly continue with higher values of n, but these would become
increasingly complicated. However, we obtained some simple statements, which
lead to improved results for Conjecture 4.3, especially when n is prime.
Theorem 5.11. [8, Theorem 6.12]

Let n be a prime number, X a general curve of genus g ≥ 3 and (L, V ) a
general linear series of type (d, n + 1). Then ML,V is stable except possibly in
• n = 11, g = 13, 14, d = 33.
• n = 13, g = 15, 16, d = 39.
4(n−1)2
• n ≥ 17, n + 5 ≤ g ≤ gn , d = 3n, 4n, where gn = 3n−5 .
6. Coherent Systems
The proofs of our theorems use the theory of coherent systems. We briefly recall
the basic definitions and results on coherent systems.
Definition 6.1. A coherent system (E, V ) of type (r, d, k) on X is a pair

consisting of a vector bundle E of rank r and degree d and a linear subspace
V ⊂ H 0 (E) of dimension k.
The coherent system (E, V ) is said to be generated if the evaluation map

X × V → E is surjective.
For any real number α, there is a notion of α-stability of a coherent system.
Definition 6.2. For any real number α, a coherent system (E, V ) of type
(r, d, k) is α-stable if, for any proper coherent subsystem (E , V ) of type
(r , d , k ) of (E, V ),
d + αk d + αk
< .
r r
38 Usha N. Bhosle
6.1. Moduli spaces of coherent systems. There exist mod-

uli spaces G(α; r, d, k) of α-stable coherent systems of type (r, d, k). For
G(α; r, d, k), k > 0 to be nonempty, we must have α > 0.
There are finitely many critical values 0 = α0 < α1 < · · · < αL of α;
as α varies, the concept of α-stability remains constant between two consecu-
tive critical values. Denote by G0 (r, d, k) (resp. GL (r, d, k)) the moduli spaces
corresponding to 0 < α < α1 (resp. α > αL ).
G0 G1 GL
0 − − −− α1 − − −− α2 − − − − −− αL −−−−−
For (E, V ) ∈ G0 , the vector bundle E is semistable.

Moreover, if E is stable, then (E, V ) ∈ G0 (r, d, k) for any V .
We recall a few facts about the moduli space GL (n, d, n + 1) which are
crucial for our proofs.
6.2. The moduli space GL . Let X be a Petri curve. Let GL (n, d, n+1)
denote the moduli space of coherent systems of type (n, d, n + 1) for large α.
• dim GL (n, d, n + 1) = β(n, d, n + 1) := g − (n + 1)(n − d + g).
• GL (n, d, n + 1) = ∅ if and only if β(n, d, n + 1) ≥ 0 or equivalently

d ≥ g + n − n+1g
.
• GL (n, d, n + 1) is irreducible if β(n, d, n + 1) > 0

and finite if β(n, d, n + 1) = 0.
• There is a nonempty open subset of GL (n, d, n+1) consisting of generated

coherent systems. When β(n, d, n + 1) = 0, every element is generated.
Define an open subset U (n, d, n + 1) ⊂ GL (n, d, n + 1) by
U (n, d, n + 1) := {(E, V ) ∈ GL (n, d, n + 1)|E is stable}

= {(E, V )|E is stable and (E, V ) ∈ G(α; n, d, n + 1) for all α > 0}.
6.3. Reduction of problem to coherent systems. Note that the

pair (L, V ), that we started with in Section 2, is a generated coherent system
of type (1, d, n + 1). Dualising the sequence
(1) 0 → ML,V → X × V → L → 0
gives
(2) 0 → L∗ → X × V ∗ → E → 0, E = ML,V
∗
,
and from the comology exact sequence of the sequence (2), we see that V ∗ ⊂
H 0 (E) as H 0 (L∗ ) = 0. Note that L = detE. Thus (L, V ) gives a generated
coherent system (E, V ∗ ) of type (n, d, n + 1).
Conversely, starting with such (E, V ∗ ), we can recover (L, V ) if E is
semistable by dualising sequence (2). The semistability of E implies that
H 0 (E ∗ ) = 0 and so V ⊂ H 0 (L).
This is called the dual span construction. Conjecture 4.1 (respectively Con-
jecture 4.3) is equivalent to the existence of a generated coherent system (E, V ∗ )
of type (n, d, n+1) with E semistable (respectively E stable) on a general curve.
Conjecture 4.5 is equivalent to showing that U (n, d, n + 1) = ∅ where
U (n, d, n + 1) := {(E, V ) ∈ GL (n, d, n + 1)| E is stable} .
For : Intersecting U (n, d, n+1) with the open subset of GL (n, d, n+1) consisting
of generated coherent systems, we get an open subset consisting of generated
(E, V ∗ ) with E stable. Then by the dual span construction, E ∗ = ML,V for
some general (L, V ) proving conjecture 4.5.
6.4. Idea of Proof of Conjectures. We start with a simple but useful

lemma.
Lemma 6.3. On any smooth curve C, suppose that U (n, d, n + 1) = ∅. Then

U (n, d + cn, n + 1) = ∅ for any positive integer c.
Proof. Suppose that (E, W ) ∈ U (n, d, n+1). Let Lc be any effective line bundle
of degree c and let s be a non-zero section of Lc .The section s induces a linear
map H 0 (E) → H 0 (E ⊗ Lc ). Denote the image of W under this map by W ⊗ s.
Then (E ⊗ Lc , W ⊗ s) ∈ U (n, d + cn, n + 1).
An immediate consequence of the lemma is the following useful reduction.
Corollary 6.4. Let C is a Petri curve of genus g ≥ 3, n ≥ 2. Suppose that

ML ,V is stable for the general linear series (L , V ) of type (d , n + 1) for all
d with

g
g+n− ≤ d < g + n + b,
n+1

where b + n+1 g
> n − 1. Then ML,V is stable for the general linear series

(L, V ) of type (d, n + 1) for all d ≥ g + n − n+1g
.
Thus, for fixed g and n, we need to prove Conjecture 4.5 only for at most
n values of d.
40 Usha N. Bhosle
Let
U c := {(E, W ) ∈ GL (n, d, n + 1)|E not stable} ,
be the complement of U (n, d, n + 1). We need to show that either this set is
empty or it has dimension strictly less than the dimension of GL (n, d, n + 1).
If C is Petri, we can assume that (E, W ) is generated and h0 (E ∗ ) = 0. If E
is not stable, there exists a stable subbundle E1 of E with μ(E1 ) ≥ μ(E) and
hence an extension of coherent systems
0 → (E1 , V1 ) → (E, W ) → (E2 , V2 ) → 0 (6.1)
with E1 stable and (Ei , Vi ) of type (ni , di , ki ). Note that (E2 , V2 ) is generated
and h0 (E2∗ ) = 0. It follows that

1 g
k2 ≥ n2 + 1, 1 + g− ≤ μ(E2 ) ≤ μ(E) ≤ μ(E1 ). (6.2)
n2 n2 + 1
The hardest part of the proof is to estimate the dimensions of the spaces of
such coherent systems in several cases of (ni , di , ki ) satisfying (6.2).
This uses a lot of our work on coherent systems including the Ext groups for
coherent systems and the wall crossing formulas for moduli of coherent systems.
Finally it gives the following key theorem.
Theorem 6.5. Suppose that C is a Petri curve, g ≥ 2, n ≥ 2 and

g (n2 − n − 2)g
g+n− ≤d<g+n+ . (6.3)
n+1 2(n − 1)2
Then, for the general linear series (L, V ) of type (d, n + 1), ML,V is stable.
When g is odd, the inequality < in the hypothesis can be replaced by ≤.
Using this theorem and our other results as well as Theorems 5.1 and 5.2 , on
the conjectures, we solve Conjecture 4.1 completely and get many new results
on Conjectures 4.3 and 4.5. We compute complete lists of possible exceptions to
Conjectures 4.3 and 4.5 for n = 5, 6, 7. In principal, we can give such complete
lists for any given rank n, it gets more and more complicated as n increases.
7. Some Applications
∗
The bundle ML,V and its dual ML,V have been extensively studied over many
years, mainly because of their applications in syzygy problems. The Koszul
cohomology groups encode certain properties of ML,V and its exterior powers
(see, for example, [2, section 2.1]) and are fundamental in the study of syzygies
(see, among others, [2] for a general account, and [18] for the Minimal Res-
olution Conjecture). For an overview of early work in this area, see [20]. We
discuss a few of these below.
7.1. The Minimal Resolution Conjecture. The minimal resolu-

tion conjecture (MRC) for (smooth) curves can be reformulated in geometric
terms as follows [18]:
H 0 (X, Λi ML,V ⊗ ξ) = 0 ∀i = 0, · · · , n ,

for a general line bundle ξ with d(ξ) = g−1+ id
n (i.e., ML,V satisfies Raynaud’s
Property). Aprodu, Farkas and Ortega showed that if μ = d/n is an integer
then MRC holds for a general embedding X → Pn (i.e., given by a general
linear system (L, V )) of a general curve X and for all μ, n ≥ 1 [1].
The ideal generation conjecture (IGC) is a particular case of MRC. Let
X → Pn be the embedding given by the linear system (L, V ). Let Γ ⊂ X be a
(large enough) general set of points and γ the number of elements of Γ. Let IΓ
and IX denote the ideals of Γ and X in Pn . The IGC states that the quotient
ideal IΓ /IX is generated by minimal number of generators, precisely

g−1+γ
b1,u = max {d − n(du − n + 1 − g), 0} , u = 1 + .
d
The IGC holds for a generic curve X and a general embeding X → Pn [1].
7.2. Relation to cotangent bundle of Pn . Let L be a line bundle

of degree d on a smooth curve X generated by a subspace V of its sections.
Since V generates L, the restriction of the evaluation map (defined in section
2) to x × V, x ∈ X gives a surjection x × V → Lx , Lx being the fibre of L at
x. Thus for every x, we get a 1-dimensional quotient qx of V . This defines a
morphism
φV : X → P(V ) ,
where P(V ) is the projective space of 1-dimensional quotients of V . On P(V ),
there is an Euler sequence
0 → M → P(V ) × V → O(1) → 0
where O(1) is the tautological ample line bundle on P(V ) and the last but
one map is an evaluation map. Then M ∼ = ΩP(V ) ⊗ O(1), where ΩP(V ) is the
cotangent bundle of P(V ). Pulling back Euler sequence to X by φV ,
0 → φ∗V (M ) → X × V → L → 0 ,
showing that
φ∗V (M ) ∼
= ML,V .
42 Usha N. Bhosle
7.3. Koszul Cohomology and Green’s conjecture. If X is a

projective variety, L a line bundle and F is a coherent sheaf on X, then the
Koszul complex is
dp+1
→ Λp+1 H 0 (X, L) ⊗ H 0 (X, F ((q − 1)L)) → Λp H 0 (X, L) ⊗ H 0 (X, F (qL))
→ Λp−1 H 0 (X, L) ⊗ H 0 (X, F ((q + 1)L)) → .

The Koszul differential is defined by

p+1
dp+1 ((s1 ∧ · · · ∧ sp+1 ) ⊗ f ) = (s1 ∧ · · · ∧ ŝj ∧ · · · ∧ sp+1 ) ⊗ sj f ,
j=1
for sj ∈ h0 (X, L), f ∈ H 0 (F ((q − 1)L)).

The Koszul cohomology Kp,q (X, L, F ) is defined as the cohomology at the
middle of the above Koszul complex. One defines
Kp,q (X, L) := Kp,q (X, L, OX ) .
The Green’s Conjecture on syzygies of canonical curves says that
{Kp,1 (X, KX ) = 0, ∀p ≥ } if and only if {Clif f (X) ≥ g − 1 − } .
See [33, 34, 29, 30] for results on Green’s Conjecture.

The bundles ML have also been studied in connection with theta-divisors
[26, 27, 5, 23, 24]. If ML,V is not semistable, then it does not admit a theta
divisor (the converse is not true). One of the main questions about ML,V is
when it is stable or semistable; this has major implications in higher rank
Brill-Noether theory (see, for example, [9, section 11], [7, section 9], [21, 22]).
Applications to Brill-Noether theory and Picard bundles are discussed in the
next section in a more general set up.
8. Syzygy Bundles on Nodal and Cuspidal

Curves
Myself and Sanjay Kumar Singh generalised some of the results on Syzygy
bundles on smooth curves to nodal and cuspidal curves [11].
Let Y be an integral projective curve with at most ordinary nodes and cusps
as singularities defined over an algebraically closed field of characteristic 0. Let
g(Y ) denote the arithmetic genus of Y . Let KY denote the dualising sheaf of
Y . Since Y is a Gorenstein curve, KY is locally free.
Definition 8.1. For a torsion free sheaf E on Y generated by global sections,

define a torsion free sheaf D(E) on Y by
0 → D(E)∗ → H 0 (Y, E) ⊗ OY → E → 0. (8.1)
We note that D(E) is locally free if and only if E is locally free.

The following generalisation of a result by Paranjape and Ramanan in [25,
Corollary 3.5] holds over Y .
Theorem 8.2. Let g(Y ) ≥ 2.
1. The vector bundle D(KY ) is semistable.
2. If Y is nonhyperelliptic, then D(KY ) is stable.
The proof of Theorem 8.2 by Paranjape and Ramanan in case Y is a smooth

curve (of genus at least 2) crucially uses the fact that if W is a quotient vector
bundle of rank r of D(KY ) generated by sections on Y , then W is generated
by r + 1 independent sections. In case Y is nodal or cuspidal, W could be non
locally free and may not be generated by r + 1 sections. Hence the proof of [25]
fails to generalise to the singular case.
Unlike in the smooth case, a semistable torsionfree sheaf E on Y with
μ(E) ≥ 2g(Y ) may not be globally generated (i.e., generated by global sec-
tions). However if it is locally free then it is globally generated. We have the
following analogue of a result of Butler [14, Theorem 1.2].
Theorem 8.3. Let E be a torsionfree sheaf on Y with μ(E) ≥ 2g(Y ) such that
E is generated by global sections.
1. If E is semistable, then the torsionfree sheaf D(E) is semistable.
2. Suppose that E is stable. Then D(E) is stable if μ(E) > 2g(Y ).

For μ(E) = 2g(Y ), D(E) is stable except when Y is hyperelliptic or
E ⊃ KY .
We need a new strategy to prove Theorems 8.2 and 8.3. We first prove both
the theorems for lower genera and then use induction on (arithmetic) genus to
prove both of them together for all genera.
Theorem 8.4. 1. If Y is hyperelliptic, then
D(D(KY )) ∼
= D(KY ) ∼
= ⊕g(Y )−1 OY (1), h0 (Y, D(KY )) = 2g(Y ) − 2 .
Here OY (1) denotes the pull back to Y of OP1 (1).

44 Usha N. Bhosle
2. If Y is not hyperelliptic, then
D(D(KY )) ∼
= KY , h0 (Y, D(KY )) = g(Y ) .
3. If Y is not hyperelliptic, then the natural maps

m
⊗H 0 (Y, KY ) → H 0 (Y, KYm ), S m (H 0 (Y, KY )) → H 0 (Y, KYm ), m ≥ 2
are surjections.
Theorem 8.4(3) was proved for smooth curves Y by Noether and was inter-
preted as the canonical embedding of a smooth curve is projectively normal.
8.1. Applications to Brill-Noether Theory. We apply these the-

orems to study Brill-Noether subschemes for torsionfree sheaves with slope μ
at most 2 on Y with g(Y ) = g [11], generalising the results of V. Mercat over
smooth curves [21], [22]. The Brill-Noether subschemes for torsionfree sheaves
with slope μ less than 1 were studied by me [6]. Let U (n, d) be the mod-
uli space of semistable torsionfree sheaves of rank n and degree d on Y . Let
U (n, d) ⊂ U (n, d) be the open subvariety corresponding to locally free sheaves.
The superfix s will denote the subset corresponding to stable sheaves.
Let B(n, d, k) ⊂ U s (n, d) be the Brill-Noether locus of stable torsionfree
sheaves with at least k independent sections and B (n, d, k) = B(n, d, k) ∩
U (n, d). Denote by B̃(n, d, k) ⊂ U (n, d) the Brill-Noether locus of semistable
torsionfree sheaves E with h0 (grE) ≥ k, where grE denotes the associated
graded of E for stable filtration. Finally, B̃ (n, d, k) := B̃(n, d, k) ∩ U (n, d).
Any irreducible component of B(n, d, k) has dimension at least β(n, d, k) where
β(n, d, k) = n2 (g − 1) + 1 − k(k − d + n(g − 1)).
Moreover, no irreducible component of B (n, d, k) is contained in B (n, d, k+1).

We first consider the question of determining the conditions of nonemptiness.
d−n
Corollary 8.5. Let k > n + g .
1. B̃(n, d, k) = ∅ for 0 < d

n < 2.
2. If Y is nonhyperelliptic, then
B(n, 2n, k) = ∅ for (n, k) = (g − 1, g) and B(g − 1, 2g − 2, g) = {D(KY )}.
3. If Y is hyperelliptic, then B(n, 2n, k) = ∅ if (n, k) = (1, 2), B(1, 2, 2) =

{OY (1)}, the pull back of the hyperplane line bundle on P1 .
To investigate the singular set of B(n, d, k), we define the Petri map for a
torsionfree sheaf E as the map
h : H 0 (Y, E) ⊗ H 0 (Y, E ∗ ⊗ K) → (Ext1 (E, E))∗
dual to the natural map φ : Ext1 (E, E) → H 0 (Y, E)∗ ⊗H 1 (Y, E). If E is locally
free, the Petri map is the usual one given by multiplication of sections.
We prove that the Petri map h is injective for all vector bundles E ∈
B (n, d, k) − B (n, d, k + 1) except in the following case: Y is hyperelliptic,
d = 2, n = 1 and k = 2. We use this to show the following:
Let C denote a component of B(n, d, k) such that C is not contained in
B(n, d, k + 1). Then we show that dim C = β(n, d, k) and the regular locus
Creg of C is given by
Creg = C ∩ (B (n, d, k) − B (n, d, k + 1)).
Some explicit descriptions of Brill-Noether loci

We get explicit descriptions of Brill-Noether Loci in the cases l = (d−n)
g is a
positive integer and k = n + l. These descriptions enable us to determine their
geometric properties.
Theorem 8.6. Let l, n be integers, l > 0.
1. For n > (g − 2)l + 2l2 ,
B(n, n + gl, n + l) ∼
= U s (l, n + gl), B̃(n, n + gl, n + l) ∼
= U (l, n + gl)
In particular, B(n, n + gl, n + l) and B̃(n, n + gl, n + l) are irreducible,

seminormal and B̃(n, n + gl, n + l)reg = B(n, n + gl, n + l)reg = B (n, n +
gl, n + l).
2. For n > gl, there are rational maps B(n, n + gl, n + l) U s (l, n + gl)
and B̃(n, n + gl, n + l) U (l, n + gl) defined on B (n, n + gl, n + l) and
B̃ (n, n + gl, n + l) respectively. They induce isomorphisms
B (n, n + gl, n + l) ∼
= U s (l, n + gl), B̃ (n, n + gl, n + l) ∼

= U s (l, n + gl).
In particular B (n, n + gl, n + l) is irreducible and nonsingular. B̃ (n, n +

gl, n + l) is irreducible and normal.
Application to nonemptiness of B(n, d, k) and B̃(n, d, k):
Let B b (n, d, k) ⊂ B(n, d, k) be the subset consisting of torsionfree sheaves of
local type b at one node, locally free elsewhere, similarly B̃ b (n, d, k), 1 ≤ b ≤ n.
For a curve X of genus g the Brill-Noether number is denoted by βX (n, d, k),
it is the expected dimension of the Brill-Noether locus.
46 Usha N. Bhosle
Theorem 8.7. Let 1 < μ < 2.

d−2n
1. B n (n, d, k) is empty if k > n + g−1 .
If k ≤ n+ d−2n n n
g−1 , then B (n, d, k) and B̃(n, d, k) are nonempty. B (n, d, k)
has m irreducible components each of dimension βg−1 (n, d−n, k), m being
the number of nodes.
In particular B(n, d, k) is nonempty if k ≤ n + d−2n
g−1 .
2. If d ≤ n + g, then B (n, d, n) is nonempty and hence B (n, d, k) is

nonempty for k ≤ n.
B (n, d, n) is irreducible. B (n, d, 1) is irreducible of expected dimension.
3. The subset of B̃(n, d, k) corresponding to torsionfree sheaves which are of
type n at m ≥ 2 nodes or cusps and locally free at other nodes or cusps
is empty.
d
8.2. Application to Picard bundles. Let J (Y ) denote the com-
pactified Jacobian of Y i.e., the moduli space of torsion free sheaves of rank 1
and degree d on Y . Fix a line bundle of degree d + 1 on Y . Then Y is embedded
in J¯d (Y ) by the Abel map x
→ Ix ⊗ L for x ∈ Y where Ix denotes the ideal
d
sheaf of x. Let YL−1 denote the image of Y in J (Y ).
For d ≥ 2g(Y ) − 1, the direct image of the Poincare sheaf on J¯d (Y ) × Y
on J¯d (Y ) is a vector bundle called the Picard bundle Ed on the compactified
Jacobian. We apply Theorems 8.2 and 8.3 to study (semi)stability properties
of the restrictions of the Picard bundle Ed to the curve YL−1 generalising the
results of [17] over a smooth curve.
Proposition 8.8. Let g(Y ) ≥ 1.
(1) The restriction of the Picard bundle Ed to YL−1 is stable if d ≥ 2g(Y ) or if
g(Y ) = 1.
(2) For g(Y ) ≥ 2 and a general L (i.e. L not containing KY ), the restriction
of the Picard bundle E2g(Y )−1 to YL−1 is stable if Y is not hyperelliptic and it
is semistable if Y is hyperelliptic.
If L ⊇ KY , the restriction of the Picard bundle E2g(Y )−1 to YL−1 is semistable
(but non-stable) for g(Y ) ≥ 2.
References
[1] M. Aprodu, G. Farkas and A. Ortega, Minimal resolutions, Chow forms of K3
surfaces and Ulrich bundles, arXiv 1212.6248v4, to appear in J. Reine Angew.
Math.
[2] M. Aprodu and J. Nagel, Koszul Cohomology and Algebraic Geometry, Uni-
versity Lecture Series 62, American Mathematical Society, Providence RI
2010.
[3] E. Arbarello, M. Cornalba, P. A. Griffiths and J. Harris, Geometry of Algebraic
curves Volume I, Springer-Verlag, New York 1985.
[4] E. Ballico and G. Hein, On the stability of the restriction of T P n to projective
curves, Arch. Math. 71 (1998), 80–88.
[5] A. Beauville, Some stable vector bundles with reducible theta divisor,
Manuscripta Math. 110 (2003), 343–349.
[6] U. N. Bhosle, Brill-Noether theory on nodal curves. Internat. J. Math., Vol.
18, No. 10 (2007) 1133-1150.
[7] U. N. Bhosle, L. Brambila-Paz and P. E. Newstead, On coherent systems of
type (n, d, n + 1) on Petri curves, Manuscripta Math. 126 (2008), 409–441.
[8] U. N. Bhosle, L. Brambila-Paz and P. E. Newstead, On linear systems and a
conjecture of D.C. Butler. International Journal of Mathematics, Vol. 26, No.
1 (2015), (Published online).
[9] S. B. Bradlow, O. Garcia-Prada, V. Muñoz and P. E. Newstead, Coherent
systems and Brill-Noether theory, Internat. J. Math. 14, no. 7, (2003) 683–
733.
[10] S. B. Bradlow, O. Garcia-Prada, V. Mercat, V. Muñoz and P. E. Newstead,
Moduli spaces of coherent systems of small slope on algebraic curves, Comm.
in Algebra 37 (2009), no. 8, 2649–2678
[11] U. N. Bhosle, S. K. Singh, Brill-Noether loci and generated torsionfree sheaves
over nodal or cuspidal curves. Manuscripta math. 141, Issue 1, (2013), 241-
271.
[12] L. Brambila-Paz, Non-emptiness of moduli spaces of coherent systems, Inter-
nat. J. Math. 19 (2008), no. 7, 779 –799.
[13] L. Brambila-Paz and A. Ortega, Brill-Noether bundles and coherent systems
on special curves, Moduli Spaces and Vector Bundles, ed. L. Brambila-Paz et
al., LMS Lecture Notes Series 359, CUP, Cambridge 2009.
[14] D. C. Butler, Normal generation of vector bundles over a curve, J. Diff. Geom.
39 (1994) 1–34.
[15] D. C. Butler, Birational maps of moduli of Brill-Noether pairs, preprint,
arXiv:alg-geom/9705009.
[16] C. Camere, About the stability of the tangent bundle of Pn restricted to a
curve, C. R. Acad. Sci. Paris, Ser. I 346 (2008), 421–426.
[17] L. Ein and R. Lazarsfeld, Stability and restrictions of Picard bundles with
an application to the normal bundles of elliptic curves, Complex projective
geometry (Trieste 1989/Bergen 1989), ed. G. Ellingsrud, C. Peskine, G. Sac-
chiero and S. A. Stromme, LMS Lecture Notes Series 179, 149–156, CUP,
Cambridge 1992.
48 Usha N. Bhosle
[18] G. Farkas, M. Mustaţă and M. Popa, Divisors on Mg,g+1 and the minimal
resolution conjecture for points on canonical curves, Annales Sci. de L’École
Normale Supérieure 36 (2003), 553–581.
[19] H. Lange and P. E. Newstead, Coherent systems on elliptic curves. Internat.
J. Math. Vol. 16, No. 7 (2005) 787-805.
[20] R. Lazarsfeld, A sampling of vector bundle techniques in the study of linear
series, Proceedings of the first college on Riemann surfaces held in Trieste,
Italy, November 9 - December 18, 1987, ed. M. Cornalba et al., 500–559,
World Scientific Publishing Co. Teaneck, NJ 1989.
[21] V. Mercat, Le problème de Brill-Noether pour les fibrés stables de petite
pente, J. Reine Angew. Math. 506 (1999), 1–41.
[22] V. Mercat, Fibrés stables de pente 2, Bull. Lond. Math. Soc. 33 (2001) 535–
542.
[23] E. Mistretta, Stability of line bundles transforms on curves with respect to low
codimensional subspaces, J. Lond. Math. Soc. (2) 78 (2008), no. 1, 172–182.
[24] E. Mistretta and L. Stoppino, Linear series on curves: stability and Clifford
index, Internat. J. Math. 23 (2012), no. 12, doi: 10.1142/S0129167X12501212.
[25] K. Paranjape and S. Ramanan, On the canonical ring of a curve, Algebraic
Geometry and Commutative Algebra in Honor of Masayoshi Nagata (1987)
503–516.
[26] M. Popa, Generalized theta linear series on moduli spaces of vector bundles
on curves, arXiv 0712.3192.
[27] M. Popa, On the base locus of the generalised theta divisor, C. R. Acad. Sci.
Paris Sér. I Math. 329 (1999), no.6, 507–512.

[28] O. Schneider, Stabilité des fibrés p EL et condition de Raynaud, Ann. Fac.
Sci. Toulouse Math. (6) 14 (2005), no. 3, 515–525.
[29] M. Teixidor i Bigas, Green’s conjecture for the generic r-gonal curve of genus
g ≥ 3r − 7, Duke Math. J. 111 (2002), no.2, 195–222.
[30] M. Teixidor i Bigas, Syzygies using vector bundles, Trans. Amer. Math. Soc.
359 (2007), no.2, 897–908.
[31] M. Teixidor i Bigas, Existence of coherent systems. II, Internat. J. Math. 19
(2008), No.10, 1269–1283.
[32] L. Tu, Semistable vector bundles over elliptic curves, Adv. Math. 98 (1993)
1–26.
[33] C. Voisin, Green’s generic syzygy conjecture for curves of even genus lying on
a K3 surface, J. European Math. Soc. 4 (2002), 363–404.
[34] C. Voisin, Green’s canonical syzygy conjecture for generic curves of odd genus,
Compositio Math. 141 (2005), no.5, 1163–1190.
A Survey of Low Dimensional (Quasi)
Projective Groups
Indranil Biswas∗,‡ and Mahan Mj†,‡
Abstract
A brief survey of some recent results on projective and quasiprojective groups
of low cohomological dimensions was presented by the second author at the
conference in Hyderabad University in March 2015. This is a slightly expanded
version of the talk. Part of the survey involves joint work with H. Seshadri and
A. J. Parameswaran.
Mathematics Subject Classification (2010). 14P25, 57M05, 14F35, 20F65,

57M50, 57M05 (Primary); 14F35, 32J15 (Secondary)
Keywords. Kähler group, projective group, duality group, cohomological dimension,

holomorphically convex, Good complexification, splitting theorem, affine variety, fun-
damental group, Geometrization theorem, virtual Betti number
1. Motivational Questions
There are a couple of strands that have fed into work of several people, in
particular the authors, over the last few years.
1.1. Serre’s question and its low-dimensional versions. Our

starting point is the following question of Serre:
∗ School of Mathematics, Tata Institute of Fundamental Research, Homi Bhabha Road,
Bombay 400005, India. E-mail: [email protected]

† School of Mathematics, Tata Institute of Fundamental Research, Homi Bhabha Road,
Bombay 400005, India. E-mail: [email protected]

‡ The authors acknowledge the support of their respective J. C. Bose Fellowships.
Geometry, DOI 10.1007/978-981-10-5648-2_4
50 Indranil Biswas and Mahan Mj
Question 1.1 ([ABCKT] (Serre)). Which finitely presented groups can be

realized as fundamental groups of smooth complex projective varieties or more
generally of compact Kähler manifolds?
This question is the theme of the book [ABCKT]. Fundamental groups of
compact Kähler manifolds are usually referred to as Kähler groups. Similarly
fundamental groups of smooth complex projective (respectively, quasiprojec-
tive) varieties are referred to as projective (respectively, quasiprojective) groups.
Projective groups are of course Kähler. The converse is a well–known open
question.
A special case of Question 1.1 was posed by Donaldson and Goldman and
answered by Dimca and Suciu [DiSu]:
Question 1.2. Which Kähler groups appear as fundamental groups of closed
3-manifolds?
More generally, a standard theme in the theory of Kähler groups has been:
Question 1.3. Meta-question: Take your favorite class of groups. (Our fa-
vorite classes occur naturally in geometric group theory or low-dimensional
topology.) Which of them are Kähler/projective/quasiprojective?
Examples of such “favorite” classes include the following:
1. 3-manifold groups [DiSu, Ko1, BMS, Ko3, DPS, FrSu, BiMj2].
2. One-relator groups [BiMj3, Ko2].
3. More generally, groups of cohomological dimension 2 and 3. (See [BiMj1]
for preliminary results for cd2 groups.)
A quick look at the above list justifies the term “low-dimensional” in the
title of this paper.
1.2. Complexifications. The first part of Hilbert’s 16th problem may

be thought of as one of the starting points of real algebraic geometry. The main
problem can be summarized in the following question.
Question 1.4. Let X be a variety defined over the reals R. Let XR denote its
set of real points. What is the relationship between the topology of X and that
of XR ?
Hilbert’s problem deals with the dimension one case of this. Works of
Arnold [Ar], Gudkov [Gu], Viro [Vi1, Vi2] deserve special mention in ad-
dressing Hilbert’s 16th problem in dimension one. In dimension 2, Kharlamov
[Kh1, Kh2] made major contributions. The major thrust in dimension three
came from Kollár [Kol1, Kol2, Kol3, Kol4]. All these pieces of work dealt with
projective varieties, sometimes singular.
A Survey of Low Dimensional (Quasi) Projective Groups 51
Nash [Na] and Tognoli proved that any smooth closed connected mani-
fold may be realized as the real locus XR of a smooth projective variety. This
prompted Nash to ask:
Question 1.5. Can any smooth closed connected manifold may be realized as
the real locus XR of a rational variety?
Question 1.5 is often called the Nash Conjecture, though it is very likely
the shortest-lived conjecture in mathematical history lasting −38 years. It was
disproved in dimension 2 by Commessatti [Co] in 1914. It was disproved in
dimension 3 by Kollár in [Kol5].
We shall refer to the complex locus X of M = XR as a complexification
of M .
The class of quasiprojective or affine complexifications poses different ques-
tions. In this article we shall later have occasion to review a class of affine or
quasiprojective complexifications called minimal or good complexifications
[Ku, To, Mc, BiMj2, BMP2].
1.3. Cohomological Dimension. Fundamental groups of compact

Riemann surfaces are called surface groups. A group is said to be virtually
a surface group if some finite index subgroup of it is a surface group. The
following question is basic in the theory of low dimensional projective groups:
Question 1.6. Let G be a non-trivial Kähler/projective group of (integral co-

homological) dimension less than four. Is G the fundamental group of a closed
Riemann surface?
It is easy to see that uniform lattices in SU(n, 1) are projective Poincaré

duality groups of dimension 2n. Toledo, [To], has shown that non–uniform
lattices in SU(n, 1) are projective groups if and only if n > 2. These are
examples of 2n − 1 dimensional duality groups. Thus for every integer n ≥ 4,
there exist projective duality groups of dimension n. But a non–uniform lattice
in SU(2, 1) is not Kähler [To].
It was pointed out to us by Delzant, and a complete argument appears
already in a paper of Kotschick [Ko1], that the image of the Albanese map for
a compact Kähler manifold with fundamental group G of dimension less than
four is either a point or a smooth complex projective curve. As observed in
[BiMj3] it follows that if in addition G is also coherent (i.e., if every finitely
generated subgroup of G is finitely presented) and of cohomological dimension
2, possessing a finite index subgroup with positive first Betti number, then G
is in fact a surface group.
2. 3-manifold Groups
Question 1.2 was answered positively by Dimca and Suciu [DiSu]:
Theorem 2.1. Let G be the fundamental group of a closed 3-manifold. Then

G is Kähler if and only if G is finite.
This followed work of Reznikov and different proofs of it were given by

Kotschick [Ko1] and Biswas-Mj-Seshadri [BMS]. In fact, Kotschick [Ko1] and
Delzant (unpublished) have independently shown that if a Poincaré duality
group G of dimension three does not satisfy property T , then G is not Kähler.
In [BMS] with Seshadri we were interested in generalizing Theorem 2.1. We
were concerned mainly with the following general set-up:
i q
1 −→ N −→ G −→ Q −→ 1 (2.1)
is an exact sequence of finitely generated groups. Further suppose that Q is

infinite and not virtually cyclic. The group G is
1. either a Kähler group, i.e., the fundamental group of a compact Kähler

manifold,
2. or the fundamental group of a compact complex surface.
In [BMS], together with Seshadri we look at the restrictions that these

assumptions impose on the nature of G and Q.
It turns out that if in addition, Q is the fundamental group of a closed
3-manifold, then the existence of the exact sequence (2.1) with N finitely pre-
sented forces Q to be the fundamental group of a Seifert-fibered 3-manifold.
More precisely, let
1 −→ N −→ G −→ Q −→ 1
be an exact sequence of finitely presented groups, where Q is infinite and not

virtually cyclic, and is the fundamental group of some closed 3-manifold.
If G is Kähler, it is shown in [BMS] that Q contains as a finite index
subgroup either a finite index subgroup of the 3-dimensional Heisenberg group
or the fundamental group of the Cartesian product of a closed oriented surface
of positive genus and the circle. As a corollary of the above theorem of [BMS],
a new proof of Theorem 2.1 is obtained by setting N to be the trivial group.
If G is the fundamental group of a compact complex surface, it is shown in
[BMS] that Q must contain the fundamental group of a Seifert-fibered three
manifold as a finite index subgroup, and G contains as a finite index subgroup
the fundamental group of an elliptic fibration.
In [BMS], an example is given which shows that the relation of quasi-

isometry does not preserve Kähler groups. This gives a negative answer to
a question of Gromov (Problem on page 209 of [Gr93]) which asks whether
Kähler groups can be characterized by their asymptotic geometry.
The following technical Proposition is the starting point of the proof. Vari-
ous restrictions are imposed on Q and we deduce that G is virtually a surface
group. The above case (i) follows from [De10], [Bru03], case (ii)(a) from [DP10],
[CT89] and cases ii(b), (c) from the cut-Kähler Theorem of Delzant-Gromov
[DeGr].
Proposition 2.2. Let G be a Kähler group admitting a short exact sequence
1 −→ N −→ G −→ Q −→ 1
where N is finitely generated.

i) Then Q cannot be non-nilpotent solvable.
ii) Suppose in addition that Q satisfies one of the following:
a) Q admits a discrete faithful non-elementary action on Hn for some
n≥2
b) Q admits a discrete faithful non-elementary minimal action on a
simplicial tree with more than two ends.
c) Q admits a (strong-stable) cut R such that the intersection of all
conjugates of R is trivial
Then G is virtually a surface group.
As a consequence we have the following which we state for completeness:
Theorem 2.3 ([BMS]). Let
i q
1 −→ N −→ G −→ Q −→ 1
be the exact sequence (2.1) such that G is a Kähler group and Q is an infinite,
not virtually cyclic, fundamental group of some closed 3-manifold. Then there
exists a finite index subgroup Q of Q such that either Q is a finite index
subgroup of the 3-dimensional Heisenberg group or Q = π1 (Σ × S 1 ), where Σ
is a closed oriented surface of positive genus.
(The 3-dimensional Heisenberg group consists of the unipotent upper tri-
angular elements of GL(3, Z).)
The next theorem deals with the case that G is the fundamental group of
a compact complex surface.
Theorem 2.4 ([BMS]). Let

i q
1 −→ N −→ G −→ Q −→ 1
be the exact sequence (2.1) such that G is the fundamental group of a compact
complex surface and Q is an infinite, not virtually cyclic, fundamental group of
some closed 3-manifold. Then there exists a finite index subgroup Q of Q such
that Q is the fundamental group of a Seifert-fibered 3-manifold with hyperbolic
or flat base orbifold. Also there exists a finite index subgroup G of G such that
G is the fundamental group of an elliptic complex surface X which is a circle
bundle over a Seifert-fibered 3-manifold.
In Theorem 2.4, setting N to be the trivial group we conclude that Q is not
the fundamental group of a compact complex surface if Q is infinite and not
virtually cyclic.
Stronger results when X is of Kodaira Class VII, or admits an elliptic
fibration, are given in [BMS].
As a consequence of Theorems 2.3 and 2.4 we also get the following result,
the first part of which was proven by J. Hillman [Hi98] based on work of
Wall [Wa86]. The second part follows from Theorem 2.3 and the fact that the
product of the Heisenberg group with Z has Z ⊕ Z ⊕ Z as its abelianization.
Theorem 2.5. Let M be a closed orientable 3 manifold. Then
(i) M × S 1 admits a complex structure if and only if M is Seifert fibered.
(ii) M × S 1 admits a Kähler complex structure if and only if M = Σ × S 1
where Σ is a compact surface.
In [Ko3], Kotschick generalized Theorem 1.2 in a different direction, by
allowing compact 3-manifolds with boundary:
Theorem 2.6 ([Ko3]). Let G be the fundamental group of an arbitrary compact
three-manifold, possibly with boundary. Then G is a Kähler group, if and only
if it is either finite or the fundamental group of a closed orientable surface.
3. Quasiprojective 3 Manifold Groups

Quasiprojective 3-manifold groups were explored in [DPS, FrSu, Ko3]. In this
paper we characterize quasiprojective 3-manifold groups.
In this Section, we follow the convention that the 3-manifolds in question
have no spherical boundary components as capping such boundary com-
ponents off by 3-balls does not change the fundamental group. The following
Theorem answers Question 8.3 and Conjecture 8.4 of [FrSu] by Friedl and Suciu:
Theorem 3.1 ([BiMj2]). Let N be a compact 3-manifold (with or without

boundary). If π1 (N ) is a quasiprojective group, then N is either Seifert-fibered
or π1 (N ) is one of the following
• virtually free, or
• virtually a surface group.
Finer results leading to a complete characterization are given in [BiMj2].
The following theorem provides an answer to Question 8.1 of [FrSu] under mild
hypotheses.
Theorem 3.2 ([BiMj2]). Suppose A and B are groups, such that the free
product G = A ∗ B is a quasiprojective group. In addition suppose that both A
and B admit nontrivial finite index subgroups, and at least one of A, B has a
subgroup of index greater than 2. Then each of A, B are free products of cyclic
groups. In particular both A and B are quasiprojective groups.
4. One Relator Groups

In [BiMj3] we prove that infinite one–relator Kähler groups are precisely funda-
mental groups of (complex) one dimensional orbifolds with at most one cone–
point. One–relator groups have rational cohomological dimension two. This
provides a case where the answer to Question 1.6 is affirmative. A different
proof of the main result of [BiMj3] was given by Kotschick [Ko2] using l2 co-
homology.
In [Ar95], Arapura asks which one-relator groups are Kähler (see [Ar95, p.
12, Section J]). This question was also raised by Amorós. In [BiMj3] we prove:
Theorem 4.1. Let G be an infinite one-relator group. Then G is Kähler if
and only if it is isomorphic to
g n

a1 , b1 , · · · , ag , bg | [ai , bi ] ,
i=1
where g and n are some positive integers.

We show that each of the groups
n

g
a 1 , b1 , · · · , a g , bg | [ai , bi ] , g , n > 0,
i=1
can in fact be realized as the fundamental group of a smooth complex projective

variety.
We also prove the following closely related result:
Theorem 4.2. Let G be a Kähler group such that
• it is a coherent group of rational cohomological dimension two, and
• the virtual first Betti number of G is positive.
Then G is virtually a surface group.
We give an overview of the basic strategy of the proof from [BiMj3]:

It follows from the structure theory of one-relator groups that they are
described as iterated HNN (Higman–Neumann–Neumann) extensions. The
Kähler group G we are interested in therefore acts on the Bass-Serre tree T
associated to the HNN splitting. If T is not quasi-isometric to the real line,
it must be non-amenable and have infinitely many ends. It follows from a re-
finement of the theory of stable cuts of Delzant-Gromov [DeGr] using further
structure of one-relator groups that G is virtually a surface group in this case.
In case T is quasi-isometric to the real line, then G must be the map-
ping torus of a free group. These groups are known to be coherent [FH99]. A
simple cohomological dimension argument along with the structure of finitely
presented normal subgroups of cd 2 groups completes the proof in this case.
The torsion in G is finally handled by further structure theory of one-relator
groups.
5. Good Complexifications
A good complexification [Ku, To] of a closed smooth manifold M is defined
to be a smooth affine algebraic variety U over the real numbers such that M
is diffeomorphic to U (R) and the inclusion
U (R) −→ U (C)
is a homotopy equivalence [To], [Ku]. A good complexification comes naturally

equipped with a natural antiholomorphic involution A on U (C) whose fixed
point set is precisely the set of real points U (R). Kulkarni [Ku] and Totaro [To]
investigate the topology of good complexifications using characteristic classes
and Euler characteristic.
Theorem 5.1 ([Ku, To]). Let M admit a good complexification. Then χ(M ) ≥
0. Further, if χ(M ) > 0, then b2i+1 (M ) = 0 for all i.
It should be pointed out here that in the definition of a good complexifi-

cation, if one replaces “affine” or “quasiprojective” by Stein, then any closed
manifold M admits a Stein complexification.
5.1. 3 Manifolds Admitting a Good Complexification. In

[To], Totaro proves that all known examples of manifolds admitting metrics of
non-negative sectional curvature do admit good complexifications. He asks for
the converse. In [To, p. 69, 2nd para], Totaro asks the following question:
Question 5.2. If a closed smooth manifold M admits a good complexification,

does M also admit a metric of non-negative curvature?
In [BiMj2] we use the classification of quasiprojective 3-manifold groups to

give a positive answer to Question 5.2 for 3-manifolds.
Theorem 5.3 ([BiMj2]). A closed 3-manifold M admits a good complexifica-

tion if and only if one of the following hold:
1. M admits a flat metric,
2. M admits a metric of constant positive curvature,
3. M is covered by the (metric) product of a round S 2 and R.
Curiously, the proof of Theorem 5.3 in [BiMj2] is direct and there is virtually
no use of the method or results of [Ku, To, DPS, FrSu].
The main tools come from recent developments in 3-manifolds:
1. The Geometrization Theorem and its consequences (cf. [AFW]).
2. Largeness of 3-manifold groups [Ag, Wi, CLR, La].
The basic complex geometric tool is a theorem of Bauer, [Bau], regarding

existence of irrational pencils for quasiprojective varieties. It is a useful exis-
tence result in the same genre as the classical Castelnuovo-de Franchis Theorem
and a theorem of Gromov [Gr, ABCKT].
5.2. Splitting theorem for good complexifications. In

[BMP2], we prove a Cheeger-Gromoll type splitting theorem and initiate a
systematic study of fundamental groups of good complexifications:
Theorem 5.4. Let M be a closed manifold admitting a good complexification.

Then M has a finite-sheeted regular covering M1 satisfying the following:
1. M1 admits a fiber bundle structure with fiber N and base (S 1 )d . Here d

denotes the (real) Albanese dimension of M1 .
2. The first virtual Betti number vb1 (N ) = 0,
3. N admits a good complexification.

Gromov [Gr81] proves that if a closed smooth manifold M of dimension n

admits a metric of non-negative curvature, then there is an upper bound, that
depends only on n, on the sum of the Betti numbers of M . He further conjec-
tures, that bi (M ) ≤ bi ((S 1 )n ). Theorem 5.5 below furnishes positive evidence
towards a combination of Question 5.2 with this conjecture of Gromov by giv-
ing an affirmative answer for the first Betti number of manifolds admitting
good complexifications.
We shall say that a finitely presented group G is a good complexification
group if G can be realized as the fundamental group of a closed smooth man-
ifold admitting a good complexification (see also [ABCKT]). We deduce from
Theorem 5.4 the following critical restriction on good complexification groups:
Theorem 5.5. Let G be a good complexification group. Then there exists a
finite index subgroup G1 of G such that two following statements hold:
1. There is an exact sequence:
1 −→ H −→ G1 −→ Zk −→ 1,
where k can be zero.
2. The above H is a finitely presented good complexification group with
vb1 (H) = 0, where vb1 (H) denotes the virtual first Betti number of H.
Recall that for a group H, the virtual first Betti number vb1 (H) is the
supremum of first Betti numbers b1 (H1 ) as H1 runs over finite index subgroups
of H.
The following classes of groups are then ruled out as good complexification
groups:
1. Groups with infinite vb1 , in particular large groups.
2. Hyperbolic CAT(0) cubulated groups.
3. Solvable groups that are not virtually abelian.
4. 2- and 3-manifold groups that are not virtually abelian.
5. any group admitting a surjection onto any of the above.
Question 5.2 has an affirmative answer for 2-manifolds; this is probably
classical but follows also from [Ku, To]. An affirmative answer to Question 5.2
for 3-manifolds was given in [BiMj2]. As a consequence of the above restrictions,
Question 5.2 has an affirmative answer for 2 and 3-manifold. Thus a new self-
contained proof of the main Theorem of [BiMj2] on good complexifications is
obtained in [BMP2]. We also give a number of applications to low-dimensional
manifolds.
Theorem 5.6.
1. Question 5.2 has an affirmative answer for 2-manifolds.

2. Question 5.2 has an affirmative answer for 3-manifolds [BiMj2].
3. Let M be a closed simply connected 4-manifold admitting a symplectic
good complexification. Then M admits a metric of non-negative curva-
ture.
4. Let M be a closed 4-manifold admitting a good complexification. Further
suppose that π1 (M ) is infinite, torsion-free abelian. Then π1 (M ) is iso-
morphic to Zd , where d = 1, 2 or 4. Moreover, the manifold M admits a
finite-sheeted cover with a metric of non-negative curvature (i.e., Ques-
tion 5.2 has an affirmative answer up to finite-sheeted covering).
5. Let M be a closed 5-manifold admitting a good complexification. Further
suppose that π1 (M ) is infinite, torsion-free abelian. Then π1 (M ) is iso-
morphic to Zd , where d = 1, 2, 3 or 5. Further, if d = 2, 3 or 5, then M
admits a finite-sheeted cover M1 homeomorphic to S 2 × T 3 or S 3 × T 2 or
T 5 . In particular, M1 admits a metric of non-negative curvature.
In Theorem 5.6, a manifold M = XR is said to admit a symplectic good

complexification, if its cotangent bundle is symplectomorphic to X.
6. Homological Semistability in Cohomological

Dimension 2
In [BiMj1] we do some basic work on the “two-dimensional topology at infinity”
of projective groups.
A connected complex manifold M is called holomorphically convex if for ev-
ery sequence of points {xi }∞
i=1 of M without any accumulation point, there is a
holomorphic function f on M such that the sequence of nonnegative numbers
{|f (xi )|}∞
i=1 is unbounded. A natural subclass of projective groups is the class
of groups that can be realized as fundamental groups of smooth complex projec-
tive varieties, of dimension at least two, with holomorphically convex universal
covers. We shall call such groups holomorphically convex groups. It is easy
to see [BiMj1] that any holomorphically convex group is the fundamental group
of a smooth complex projective surface with holomorphically convex universal
cover. A conjecture of Shafarevich asserts that all smooth projective varieties
have holomorphically convex universal covers.
The homological semistability conjecture formulated by Geoghegan, [Gui,
Conjecture 5, Section 6.4], is equivalent to the statement that H 2 (G, ZG) is
free abelian for every one-ended finitely presented group [Ge, Section 13.7],
[GeMi1, GeMi2]. (Geoghegan’s conjecture was formulated originally as a ques-
tion in 1979 [Gui].) This conjecture has been established (in a stronger form)
for several special classes of groups arising naturally in the context of geo-
metric group theory: One-relator groups, free products of semistable groups
with amalgamation along infinite groups, extensions of infinite groups by in-
finite groups, (Gromov) hyperbolic groups, Coxeter groups, Artin groups and
so on [Mi1, Mi2, Mi3, MT1, MT2]. In [BiMj1] we establish this conjecture for
holomorphically convex groups.
Theorem 6.1. Let G = π1 (X) be a torsion-free holomorphically convex group.

Then H 2 (G, ZG) is a free abelian group.
A group is called linear if it is a subgroup of GL(n, C) for some n. If X

is a smooth complex projective variety such that π1 (X) is linear, then the
universal cover X is holomorphically convex [EKPR]. Therefore, Theorem 6.1
has the following corollary:
Corollary 6.2. If G is a linear torsion-free projective group, then H 2 (G, ZG)

is a free abelian group.
It is an open question whether the dualizing module of a duality group G is

a free abelian group or not [Br, p. 224]. It follows from Theorem 6.1 that this
is indeed the case if G is holomorphically convex of cohomological dimension
two.
Proposition 6.3. Let G be a holomorphically convex group of dimension two.

Then G is a duality group with free dualizing module.
The key ingredients in the proofs of Theorem 6.1, Corollary 6.2 and Propo-
sition 6.3 include
1. topology (especially second homotopy group) of smooth complex projec-

tive surfaces with holomorphically convex universal cover,
2. a spectral sequence argument for computing group cohomology with local

coefficients, which was inspired in part by an argument of Klingler [Kl],
3. homological group theory of duality, inverse duality and Poincaré duality

groups, and
4. a theorem of Eyssidieux, Katzarkov, Pantev and Ramachandran [EKPR]

showing that complex projective manifolds with linear fundamental group
have holomorphically convex universal cover.
References
[Ag] I. Agol, The virtual Haken conjecture. With an appendix by Agol, Daniel
Groves, and Jason Manning. Doc. Math. 18 (2013), 1045–1087
[ABCKT] J. Amorós, M. Burger, K. Corlette, D. Kotschick and D. Toledo, Funda-
mental groups of compact Kähler manifolds. Mathematical Surveys and
Monographs 44, American Mathematical Society, Providence, RI, 1996.
[Ar95] D. Arapura, Fundamental groups of smooth projective varieties. In: Cur-
rent topics in complex algebraic geometry (Berkeley, CA, 1992/93), 1–16,
Math. Sci. Res. Inst. Publ., vol. 28, Cambridge Univ. Press, 1995.
[ArNo] D. Arapura and M. Nori, Solvable fundamental groups of algebraic varieties
and Kähler manifolds. Compositio Math. 116 (1999), 173–188.
[Ar] V. I. Arnold, Distribution of ovals of the real plane algebraic curves, the
involutions of four-dimensional smooth manifolds, and the arithmetic of in-
tegral quadratic forms. Funkt. Anal. Prilozen 5 (1971), 1–9; English transl.,
Funct. Anal Appl. 5 (1972), 169–175
[AO] V. I. Arnold and O. A. Oleinik, The topology of real algebraic varieties.
Vestnik Moscov. Gos. Univ. Ser. 1 (1979), 7–17; English transl., Moscow
Univ. Math. Bull. 34 (1979), 5–17
[AFW] M. Aschenbrenner, S. Friedl and H. Wilton, 3-manifold groups. Preprint
arXiv:1205.0202v2
[Bas] H. Bass, Covering theory for graphs of groups. Jour. Pure and Appl. Alg.
89 (1993), 3–47.
[Bau] I. Bauer, Irrational pencils on non-compact algebraic manifolds. Internat.
J. Math. 8 (1997), 441–450.
[BiMj1] I. Biswas and M. Mj, H1 -semistability for projective groups. Math. Proc.
Cambridge Philos. Soc. 162 (2017), 89–100.
[BiMj2] I. Biswas and M. Mj, Quasiprojective three-manifold groups and complex-
ification of three-manifolds. Int. Math. Res. Notices 2015 no. 20, 10041–
10068.
[BiMj3] I. Biswas and M. Mj, One relator Kähler groups. Geom. Topol. 16 (2012),
no. 4, 2171–2186.
[BMP] I. Biswas, M. Mj and D. Pancholi, Homotopical Height. Internat. J. Math.
25 (2014), no. 13, 1450123, 43 pp.
[BMP2] I. Biswas, M. Mj and A. J. Parameswaran, A splitting theorem for good
complexifications. Ann. Inst. Fourier 66 (2016), 1965–1985.
[BMS] I. Biswas, M. Mj and H. Seshadri, Three manifold groups, Kähler groups
and complex surfaces. Commun. Contemp. Math. 14 (2012), no. 6,
1250038, 24 pp.
[Br] K. S. Brown, Cohomology of groups. Graduate Texts in Mathematics,
Springer-Verlag, 1982.
[Bru03] A. Brudnyi, Solvable quotients of Kähler groups. Michigan Math. J. 51

(2003), 477–490.
[CT89] J. A. Carlson and D. Toledo, Harmonic mappings of Kähler manifolds to
locally symmetric spaces. Publ. Math. I.H.E.S. 69 (1989), 173–201.
[Ca] F. Catanese, Fibred Kähler and quasi-projective groups. Adv. Geom.
(2003), 13–27.
[Co] A. Comessatti, Sulla connessione delle superfizie razionali reali, Annali di,
1914.
[CLR] D. Cooper, D. Long and A. Reid, Essential closed surfaces in bounded
3-manifolds, J. Amer. Math. Soc. 10 (1997), 553–563.
[De08] T. Delzant, Trees, Valuations and the Green-Lazarsfeld Set. Geom. Funct.
Anal. 18 (4) (2008), 1236–1250.
[De10] T. Delzant, L’invariant de Bieri-Neumann-Strebel des groupes fondamen-
taux des variétés kählériennes. Math. Annalen 348 (2010),119–125.
[DeGr] T. Delzant and M. Gromov, Cuts in Kähler groups, in: Infinite groups: ge-
ometric, combinatorial and dynamical aspects, ed. L. Bartholdi, pp. 31–55,
Progress in Mathematics, Vol. 248, Birkhäuser Verlag Basel/Switzerland,
2005.
[DP10] T. Delzant and P. Py, Kahler groups, real hyperbolic spaces and the
Cremona group. Compos. Math. 148 (2012), no. 1, 153–184.
[DPS] A. Dimca and S. Papadima and A. I. Suciu, Quasi-Kähler groups, 3-
manifold groups, and formality. Math. Zeit. 268 (2011), 169–186.
[DiSu] A. Dimca and A. Suciu, Which 3-manifold groups are Kähler groups? J.
Eur. Math. Soc. 11 (2009), 521–528.
[El] Y. Eliashberg, Topological characterization of stein manifolds of dimension
> 2. Internat. J. Math. 1, pages 29–46, 1990.
[EKPR] P. Eyssidieux, L. Katzarkov, T. Pantev and M. Ramachandran, Linear
Shafarevich conjecture. Ann. of Math. 176 (2012), 1545–1581.
[Fa] B. Farb, Relatively hyperbolic groups. Geom. Funct. Anal. 8 (1998),
810–840.
[FH99] M. Feighn and M. Handel, Mapping tori of free group automorphisms are
coherent. Ann. of Math. 149 (1999), 1061–1077.
[FrSu] S. Friedl and A. I. Suciu, Kähler groups, quasi-projective groups, and 3-
manifold groups, Jour. Lond. Math Soc. 89 (2014), 151–168.
[Fu] O. Fujino, On quasi-Albanese maps (preliminary version),
https://www.math.kyoto-u.ac.jp/∼fujino/quasi-albanese.pdf.
[Ge] R. Geoghegan, Topological Methods in Group Theory. Graduate Texts in
Mathematics 243, Springer, 2008.
[GeMi1] R. Geoghegan and M. L. Mihalik, Free abelian cohomology of groups and
ends of universal covers. Jour. Pure Appl. Alg. 36 (1985), 123–137.
[GeMi2] R. Geoghegan and M. L. Mihalik, A note on the vanishing of H n (G, ZG).

Jour. Pure Appl. Alg. 39 (1986), 301–304.
[Gr81] M. Gromov, Curvature, diameter and Betti numbers, Comment. Math.
Helvetici 56 (1981), 179–195.
[Gr85] M. Gromov, Hyperbolic Groups. in Essays in Group Theory, ed. Gersten,
75–263, MSRI Publ., vol. 8, Springer Verlag, 1985.
[Gr93] M. Gromov, Asymptotic Invariants of Infinite Groups. in Geometric Group
Theory, vol. 2; Lond. Math. Soc. Lecture Notes 182, Cambridge University
Press, 1993.
[Gr] M. Gromov, Sur le groupe fondamental d’une variete kählerienne. C. R.
Acad. Sci. Paris Ser. I Math. 308 (1989), 67–70.
[Gu] D. A. Gudkov, The topology of real projective algebraic varieties. Uspekhi
Mat. Nauk 29 (1974), 3–79; English transl., Russian Math. Surveys 29
(1974), 1–79.
[Gui] C. R. Guilbault, Ends, shapes, and boundaries in manifold topology and
geometric group theory. To appear in Springer Lecture Notes in Math. vol-
ume Topology and geometric group theory (Proceedings of the OSU special
year 2010-2011); arXiv:1210.6741
[Ha] A. E. Hatcher, A proof of the Smale conjecture, Diff(S 3 ) O(4). Ann. of
Math. 117 (1983), 553–607.
[He1] J. Hempel, 3-Manifolds, Ann. of Math. Studies, no. 86, Princeton Univer-
sity Press, Princeton, N. J., 1976.
[He2] J. Hempel, Residual finiteness for 3-manifolds, Ann. Math. Stud., vol. 111
(1987), 373–396.
[Hi98] J. Hillman, On 4-dimensional mapping tori and product geometries. J.
London Math. Soc. (2) 58 (1998), no. 1, 229–238.
[It] S. Iitaka, Logarithmic forms of algebraic varieties. Jour. Fac. Sci. Univ.
Tokyo 23 (1976), 525–544.
[Ka] M. Kapovich, On normal subgroups in the fundamental groups of complex
surfaces. Preprint arxiv:math/9808085.
[Kh1] V. M. Kharlamov, The maximum number of components of a surface of
degree 4 in RP 3 . Funkt. Anal. Prilozhen. 6 (1972), 101.
[Kh2] V. M. Kharlamov, Real algebraic surfaces, Proc. Internat. Congr. of Math-
ematicians, Helsinki, 1978, 421-428.
[Kl] B. Klingler, Kähler groups and duality. Preprint, arXiv:math 1005.2836.
[Kol1] J. Kollár, Real algebraic threefolds. I. Terminal singularities. Dedicated
to the memory of Fernando Serrano. Collect. Math. 49 (1998), no. 2-3,
335–360.
[Kol2] J. Kollár, Real algebraic threefolds. II. Minimal model program. J. Amer.
Math. Soc. 12 (1999), no. 1, 33–83.
[Kol3] J. Kollár, Real algebraic threefolds. III. Conic bundles. Algebraic geometry,
9. J. Math. Sci. (New York) 94 (1999), no. 1, 996–1020.
[Kol4] J. Kollár, Real algebraic threefolds. IV. Del Pezzo fibrations. Complex
analysis and algebraic geometry, 317346, de Gruyter, Berlin, 2000.
[Kol5] J. Kollár, The Nash conjecture for threefolds. Electron. Res. Announc.
Amer. Math. Soc. 4 (1998), 63–73
[Ko1] D. Kotschick, Three-manifolds and Kähler groups. Ann. Inst. Fourier 62
(2012), 1081–1090.
[Ko2] D. Kotschick, The deficiencies of Khler groups. J. Topol. 5 (2012), no. 3,
639–650.
[Ko3] D. Kotschick, Kählerian three-manifold groups. Math. Res. Lett. 20 (2013),
no. 3, 521–525.
[Ku] R. S. Kulkarni, On complexifications of differentiable manifolds. Invent.
Math. 44 (1978), 49–64.
[La] M. Lackenby, Some 3-manifolds and 3-orbifolds with large fundamental
group. Proc. Amer. Math. Soc. 135 (2007), 3393–3402.
[Lam] K. Lamotke, The topology of complex projective varieties after S. Lef-
schetz. Topology 20(1) (1981), 15–51.
[Mc] M. McLean, The growth rate of symplectic homology and affine varieties.
Geom. Funct. Anal. 22 (2012), 369–442.
[Mi1] M. L. Mihalik, Semistability at the end of a group extension. Trans. Amer.
Math. Soc. 277 (1983), 307–321.
[Mi2] M. L. Mihalik, Semistability at ∞, ∞–ended groups and group cohomol-
ogy. Trans. Amer. Math. Soc. 303 (1987), 479–485.
[Mi3] M. L. Mihalik: Semistability of Artin and Coxeter groups. Jour. Pure Appl.
Alg. 111 (1996), 205–211.
[MT1] M. L. Mihalik and S. T. Tschantz, One relator groups are semistable at
infinity. Topology 31 (1992), 801–804.
[MT2] M. L. Mihalik and S. T. Tschantz, Semistability of amalgamated products
and HNN-extensions. Mem. Amer. Math. Soc. 98 (1992), 471.
[NR08] T. Napier and M. Ramachandran, Filtered Ends, Proper Holomorphic
Mappings of Kähler Manifolds to Riemann Surfaces, and Kähler Groups.
Geom. Funct. Anal. 17 (5) (2008), 1621–1654.
[Na] J. Nash, Real algebraic manifolds. Ann. Math. 56 (1952), 405–421.
[NWY] J. Noguchi, J. Winkelmann and K. Yamanoi, The second main theorem
for holomorphic curves into semi-abelian varieties. Acta Math. 188 (2002),
129–161.
[Pe] P. Petersen, Riemannian Geometry. Graduate Texts in Mathematics, 171,
Springer Verlag, 2006.
[ScWa] P. Scott and C. T. C. Wall, Topological Methods in Group Theory, Ho-

mological group theory (C. T. C. Wall, ed.), London Math. Soc. Lecture
Notes Series, vol. 36, Cambridge Univ. Press, 1979.
[Sm] S. Smale, Diffeomorphisms of the 2-sphere. Proc. Amer. Math. Soc. 10
(1959), 621–626.
[To] D. Toledo, Examples of fundamental groups of compact Kähler manifolds.
Bull. London Math. Soc. 22 (1990), 339–343.
[To] B. Totaro, Complexifications of non-negatively curved manifolds. Jour.
Eur. Math. Soc. 5 (2003), 69–94.
[Vi1] O. Viro, Progress in the topology of real algebraic varieties over the last
six years. Uspekhi Mat. Nauk 41 (1986), 45–67; English Transl., Russian
Math. Surveys 41 (1986), 55–82.
[Vi2] O. Ya. Viro, From the sixteenth Hilbert problem to tropical geometry.
Japan. J. Math. 3 (2008), 185–214.
[Wa86] C. T. C. Wall, Geometric structures on compact complex surfaces. Topol-
ogy 25 (1986), 119–153.
[Wi] D. Wise, The structure of groups with a quasi-convex hierarchy, 189 pages,
preprint (2012).
Parabolic Sheaves and
Logarithmic Geometry
Niels Borne∗
1. Introduction
1.1. Motivation. The aim of this note is to give an introduction to the
notion of parabolic sheaves on logarithmic schemes, as first defined in my joint
work with Angelo Vistoli [BV12]. I will explain the examples we started from
in order to, hopefully, enlighten the rather formal definitions given in loc. cit.
I will conclude by a glimpse at subsequent developments.
1.2. The Context. Parabolic bundles were first introduced by V.Mehta

and C.Seshadri [MS80] on a compact Riemann surface X endowed with a set
D of marked points. In this initial formulation, a parabolic bundle is a vector
bundle endowed with a partial flag of the residual stalk at each marked point,
and also equipped with a real number in the range [0, 1[ called weight for
each component of the flags. The celebrated Mehta-Seshadri theorem states
that polystable parabolic vector bundles are in one-to-one correspondence with
unitary representations of the topological fundamental group π1 (X\D) of the
open Riemann surface.
C.Simpson, at the end of the 80’s, reformulated and then extended the
definition of a parabolic bundle to the situation where the data is a scheme X
endowed with a normal crossings divisor D. In this setup, a parabolic bundle
with weights in 1r Z can be seen as a decreasing sequence
E = E0 ⊃ E r1 ⊃ · · · ⊃ E r−1 ⊃ E1 = E(−D)
r
where we assume, for simplicity, that D has only one irreducible component.
This is sufficient for certain purposes, but it is also clear that this definition
is perfectible. For instance, one would like to be able to make the special locus
degenerate to fill up the whole space, but with the above setup one is unable
∗ Université Lille 1, Cité scientifique U.M.R. CNRS 8524, U.F.R. de Mathématiques 59
655 Villeneuve d’Ascq Cédex, France. E-mail: [email protected]
Geometry, DOI 10.1007/978-981-10-5648-2_5
68 Niels Borne
to restrict parabolic sheaves on (X, D) to (D, D), or more generally to restrict

these along a non flat morphism X → X.
There is an even more serious issue with the definition of a parabolic sheaf.
Namely, one usually works with a normal crossings divisor that contains several
irreducible components. It is straightforward to generalize Simpson’s definition
to take this into account, by using multi-indices. However, this obvious def-
inition gives satisfying results only in the case of a simple normal crossings
divisor. When one considers instead a general normal crossings divisor, where
components are allowed to self-intersect, the naive definition does not work,
in the sense that the algebraic analog of Mehta-Seshadri’s theorem does not
hold any longer. One needs to give a more local definition in order to take into
account the fact that branches of D separate étale locally, and this point was
the decisive fact that led us to introduce logarithmic geometry.
1.3. The Role of Logarithmic Geometry. Our definition of a

parabolic sheaf works for almost any logarithmic scheme (we only impose quasi-
integrality, which is a very mild assumption) and is functorial in the sense that
one can pull-back parabolic sheaves along a morphism of logarithmic schemes.
This solves the first issue mentioned above, but also enables, thanks to the
scope of logarithmic geometry, to consider parabolic sheaves in completely new
situations, for instance when X is allowed to be singular along D. We will
illustrate this fact in §3 in the special case when X is the quadratic cone.
Another application of our formalism is that, given a scheme X with a
general normal crossings divisor D, the algebraic analog of Mehta-Seshadri’s
theorem holds, if one uses the correct definition of a parabolic bundle. In order
to do so, one associates naturally to the pair (X, D) a logarithmic scheme, and
one then uses the definition for logarithmic schemes.
2. Parabolic Sheaves: Old and New

There are, of course, many possible higher-dimensional generalizations of
Mehta-Seshadri’s definition of a parabolic vector bundle on a marked algebraic
curve. But in order to be meaningful, the definition has to preserve the link
with the fundamental group of the complement of the divisor X\D. We start by
recalling this connection, and then proceed to give the modern interpretation
of parabolic vector bundles as ordinary vector bundles on certain orbifolds.
2.1. Parabolic Sheaves and Tamely Ramified Covers.

2.1.1. From representations of the fundamental group to parabolic
bundles. In this section, we consider a scheme X over a base field k and a
tamely ramified Galois cover p : Y → X with Galois group G. We assume for
Parabolic Sheaves and Logarithmic Geometry 69
convenience only that the ramification locus D, which is a Cartier divisor, is

irreducible, and we denote by r the ramification index : if E = (p∗ (D))red , then
p∗ (D) = r · E.
In this set-up, Mehta-Seshadri’s construction can be interpreted as the defi-
nition of a functor that associates to any k-linear representation ρ : G → GL(V )
a parabolic bundle (Eρ )· on (X, D). Before defining this functor, we switch to
a more convenient language.
2.1.2. From G-bundles to parabolic bundles. We keep the notations of

previous §2.1.1.
I.Biswas realized in [Bis97] that Mehta-Seshadri’s construction uses only
the structure of a G-sheaf on Y of Fρ = OY ⊗k V . Thus one can associate to
any G-vector bundle F on Y a parabolic bundle E· on (X, D). One obtains in
this way an equivalence between G-vector bundles on Y and parabolic bundles
with suitable weights on (X, D), that we will describe in full in the forthcoming
§2.1.3.
2.1.3. From stacky bundles to parabolic bundles. I.Biswas rightly calls

G-vector bundles on Y ”orbifold bundles”, they correspond namely to vector
bundles on the quotient stack [Y |G] by Galois descent along the étale mor-
phism Y → [Y |G]. The stack [Y |G] can indeed be seen as an orbifold since the
morphism π : [Y |G] → X is an isomorphism above U = X\D.
My contribution [Bor07] was at this stage to show that one can replace [Y |G]
by a canonically isomorphic stack built out only of X and the ramification data
(D, r).
Definition 2.1 ([AGV08, Cad07]). Let X be a scheme, L be an invertible
sheaf on X, endowed witha section s. For each positive integer r ∈ N∗ , one
defines the stack of roots r (L, s)/X as the stack classifying r-th roots of the
pair (L, s).
When X is instead equipped with an effective
Cartier divisor D, with canon-
ical section sD , we will write r D/X for r (OX (D), sD )/X.
Let DivX be the stack classifying pairs (L, s) on X. By definition of the
stack of roots, the following diagram is cartesian:
r
(L, s)/X DivX
·⊗r
(L,s)
X DivX
There is a natural morphism OX → DivX given by f →

(OX , f ), and since
this is clearly a Gm -torsor, the morphism OX → DivX identifies canonically
with A1 → [A1 |Gm ].
70 Niels Borne
Assume now that L is trivialized, that is we fix an isomorphism L OX

sending s to a function f . Then the cartesian diagrams
1 1
r
(L, s)/X A |μr A |Gm
x→xr
f
X A1 A1 |Gm

show that r
(L, s)/X → X is locally isomorphic to the quotient stack

R[x]
spec |μr → spec R .
xr − f
the notations of §2.1.1, the
The relevance of the stack of roots is that, with
equality p∗ (D) = r · E defines a morphism Y → r D/X. Moreover:
Lemma 2.2 (Abhyankar’s lemma). With notations of §2.1.1, if D/k is smooth,
then the morphism Y → r D/X induces an isomorphism [Y |G] → r D/X.
To explain our formulation of Biswas correspondence, we need a working
definition of parabolic sheaves.
Definition 2.3. With notations of Definition 2.1, a parabolic vector bundle on
X with weights in 1r Z is a functor:
op
1
E· : Z → Vec(X)
r
endowed with a functorial pseudo-periodicity isomorphism
E l +1 E l ⊗OX L∨
r r
satisfying the natural compatibility condition.

op
Here 1r Z is the opposite category of the one defined by the natural order,
Vec(X) is the category of vector bundles on X, and the compatibility condition
is that the morphism E l +1 → E l coincides with the morphism id ⊗s∨ : E l ⊗OX
r r r
L∨ → E l via the pseudo-periodicity isomorphism.
r
We will denote by Par r1 (X, (L, s)) the category of vector bundles on X with
weights in 1r Z with respect to (L, s).
We are now ready to relate stacky vector bundles to parabolic vector bun-
dles:
Theorem 2.4 ([Bor07]). With notationsof Definition 2.1, assume that s :
OX → L is a monomorphism, and let π : r (L, s)/X → X be the morphism to
r
the moduli space, and (M, t) be the universal r-th root of (L, s) on (L, s)/X.
Then the association

l ⊗−l

F → → π∗ F ⊗ M
r

defines an equivalence of categories Vec( r (L, s)/X) → Par r1 (X, (L, s)).
Following ideas of M.Nori, one can derive from Lemma 2.2 and Theorem
2.4 an algebraic version of Mehta-Seshadri’s theorem for a proper, reduced
scheme X over a field k, endowed with a simple normal crossings divisor D
(see [Bor09]).
Unfortunately, this strategy is bound to fail with a general normal crossings
divisor. The Fourier-like correspondence of Theorem 2.4 holds, but the corre-
sponding stack of roots is not relevant any longer, since Lemma 2.2 fails badly
if the divisor D is not smooth.
Indeed if a local equation of D is given by hk,
then the morphism [Y |G] → r D/X identifies locally with

R[x, y] R[x]
spec |μr × μr → spec |μr
x − h, y r − k
r x − hk
r
which is not an isomorphism. So the stack of roots has to be changed to take

into account the branches of D étale locally. Of course the definition of parabolic
bundles has to be modified as well, and we will have to use sheaves as coeffi-
cients. It turns out that logarithmic geometry provides the right framework to
perform these tasks.
2.2. Logarithmic Geometry.
2.2.1. Logarithmic schemes. We start by reviewing very briefly the classi-

cal theory of logarithmic schemes due to K.Kato (see [Kat89]).
Definition 2.5. Let X be a scheme. A pre-log structure (M, α) on X is a sheaf

M of commutative monoids on Xet and a morphism α : M → OX , where the
last sheaf is equipped with its multiplicative law. The pre-log structure (M, α)
is a log structure if the induced morphism α−1 OX ∗
→ OX∗
is an isomorphism.
A scheme equipped with a (pre-)log structure will be called a (pre-)log

scheme. The following example is of paramount importance for us.
Example 2.6. Let X be a scheme, and U ⊂ X an open subset. Then one

defines
MU = {f ∈ OX /f is invertible on U } .
The inclusion αU : MU → OX defines a log structure.
Morphisms of (pre-) log structures are defined in the natural way. To any
pre-log structure (M, α), one can associate a log-structure (M, α) → (M a , αa ).
The sheaf M a is obtained as the pushout of α−1 OX ∗
→ OX ∗
and α−1 OX
∗
→ M,
and the natural morphism (M, α) → (M , α ) is universal among morphisms
a a
from (M, α) to log-structures.

72 Niels Borne
Definition 2.7. Let X be a scheme, (M, α) a log structure on X, and P an

abstract monoid. A morphism P → M (X) is a Kato chart if the corresponding
morphism PX → M induces an isomorphism PX a
M.
Here is another way to formulate this notion: given an abstract monoid
P , and a scheme X, the data of P → OX (X) is equivalent to the data of a
morphism f : X → spec Z[P ]. Given such a data, and a log structure (M, α)
on X, then the additional data of a chart P → M (X) is equivalent to an
isomorphism of (M, α) with the pullback along f of the canonical log structure
on Z[P ] (that is, the log structure associated to the monoid morphism P →
Z[P ]).
Example 2.8 (The log point). Let k be a field. Then Ak1 = spec k[N] is endowed
with a natural log structure associated to the morphism N → k[N]. Pulling back
this log structure along the 0-section 0 : spec k → A1k , one gets a natural log
structure on spec k. One can check that it is given by the morphism k ∗ ⊕ N → k
defined by :
f, if n = 0
(f, n) → f · 0n =
0, if n = 0
Definition 2.9. Given a log-structure (M, α) on a scheme X, one defines
M
M= ∗ .
OX
This sheaf is important for several reasons. First of all, the support of M
is the locus where the log structure is not trivial. Moreover, if X is a smooth
scheme over a field k, endowed with a normal crossings divisor D, and U =
X\D, one can equip X with the log structure MU (Example 2.6). Then for any
geometric point x of X one has

M U x NC(x)
where C(x) is the set of local branches of D at x. This is precisely the sheaf of
coefficients we need to define parabolic sheaves.
2.2.2. Deligne-Faltings logarithmic schemes. In order to be able to give

the definition of parabolic sheaves on logarithmic schemes, we have to use
another interpretation of logarithmic geometry. To explain the following defi-
nition, let us start from a (Kato) log-structure (M, α) on a scheme X. Passing
∗
to the quotient by the action of OX one gets:
α
M OX
L
M DivX
∗
If we assume that (M, α) is quasi-integral (that is the action of OX on M is
free), then the diagram is cartesian, hence the data of L enables to reconstruct
(M, α).
Definition 2.10. A Deligne-Faltings log structure (A, L) on a scheme X is

the data of a sheaf of monoids A on Xet and of a morphism L : A → DivX
with trivial kernel.
Here by morphism, we mean a morphism of symmetric monoidal fibered

categories over Xet . The condition on the kernel (weaker than injectivity for
monoids!) is necessary to get back a Kato log structure (and not only a Kato
pre-log structure). One shows easily that Deligne-Faltings log structures on X
and (Kato) quasi-integral log structures on X are equivalent categories, through
the functor defined above.
Example 2.11. Let X be a scheme and ((L1 , s1 ), · · · , (Ln , sn ))be a family of

line bundles endowed with sections. This data is equivalent to the data of a
monoidal functor L0 : (Nr )X → DivX , and if we put A = (Nr )X / ker(L0 ), we
get a Deligne-Faltings log structure L : A → DivX .
Of course, there is nothing specific to Nr , and one could start from any
morphism L0 : P → Div(X), where P is an abstract monoid. We will say that
PX → PX / ker(L0 ) is a chart. The technical definition is as follows.
Definition 2.12. Let (A, L) be Deligne-Faltings log structure on a scheme X.

A chart for (A, L) is a morphism P → A(X) where P is an abstract monoid
and the associated sheaf morphism PX → A is a cokernel.
It is very often useful to fix additional conditions on P , like: P is finitely

generated. We will come back to this later on.
It has to be noticed that if (M, α) is a quasi-integral Kato log structure,
that admits a chart, the associated Deligne-Faltings log structure also admits
a chart, but the converse is not true, as will be clear when we describe the
refined stack of roots in §3.3. So the notion of a chart for a Deligne-Faltings
log structure is more flexible.
Despite of this, not all Deligne-Faltings log structures, even those of geo-
metric origin, admit global charts. For instance if X is a scheme, D is a nor-
mal crossings divisor, and U = X\D, then the Deligne-Faltings log structure
M U → DivX admits a global chart if and only if D is a simple normal crossings
divisor (and to get a Kato chart, we need the additional condition that each
irreducible component of D is a principal Cartier divisor). But charts do exist
étale locally.
In the sequel, we will only deal with Deligne-Faltings log structures that
admits charts étale locally, and to avoid technicalities we will mostly focus on
74 Niels Borne
the case when a global chart exists. Our aim will be, starting from the data of
a Deligne-Faltings log structure L : A → DivX , with a global chart P → A(X),
to define parabolic sheaves and refined stack of roots, so that the analog of
Theorem 2.4 holds.
2.3. Parabolic Sheaves on the Log Point.
2.3.1. Deligne-Faltings log structure of rank 1. Let us warm up with

the case P = N. In this situation, the data of the Deligne-Faltings log structure
boils down to the data of a monoidal morphism N → Div(X), that is, to the
data of a line bundle endowed with a section (L, s).
If we fix a positive integer r ∈ N∗ , then both the stack of roots and the
parabolic sheaves have been defined in section 2.1, and there is no way (and no
need !) to improve their definition further. A first question to address is: in the
extreme case when s = 0, is Theorem 2.4 still valid ? Of course, it is natural to
investigate this question when X = spec k, that is in the case of the log point.
2.3.2. The correspondence for the log point. We start by rephrasing

Example 2.6 in terms of Deligne-Faltings log structure. As it is clear from the
definition of the Kato log point as a restriction, the Deligne-Faltings log point
is the Deligne-Faltings log structure on spec k defined by the following chart
N → Div(spec k):
id
Ospec k −→ Ospec k , if n = 0
n → 0
Ospec k −
→ Ospec k , if n = 0
In others words, the Deligne-Faltings log point is defined by the pair (Ospec k , 0).
If we now apply Definition 2.3 in this situation, we see that a parabolic
bundle on the log point will be given by a diagram in the category of finite
dimensional vector spaces:
V0
φr−1 φ0
V r−1 V r1
r
φr−2 φ1
V r−2 V r2
r
···
such that the composition of r successive morphisms is 0. We have used the

opposite order of the one used in 2.3 for later convenience.
On the other hand, the stack of roots is, according to the remarks following
Definition 2.1:

k[x]
r
(Ospec k , 0)/ spec k spec |μr .
xr
So a vector
bundle on this stack of roots is a μr -equivariant vector bundle on
spec k[x] k[x]
xr , that is, a free xr -module M with a Z/r-graduation, compatible
with the natural Z/r-graduation of k[x]

xr .
To each such module M , one can associate a parabolic bundle as above
by setting V l = Ml and by defining φl : V l → V l+1 as the multiplication
r r r
by x. Since one can reverse this procedure, one has shown that Theorem 2.4
generalizes to the log point.
3. The Correspondence Stacky-parabolic for

Log Schemes
3.1. Toric Schemes. Let us fix some affine base S = spec R. In §2.1 we
started from a smooth S-scheme endowed with a normal crossings divisor. The
associated Kato log structure is étale locally a pull-back of the standard log
structure on An induced by:
Nn → Γ(OAn ) = R[x1 , · · · , xn ]
It we want to go beyond, we have to introduce new models. The simplest
thing we can do is to replace Nn by a general monoid. So we put S[P ] =
spec OS [P ] and consider the more general Kato log structure induced by:
P → Γ(OS[P ] ) = R[P ]
This is reasonable, of course, only if we fix some conditions on P . It makes
sense to assume that P is of finite type as a monoid and integral (i.e. the
cancellation law holds, equivalently, the morphism P → P gp is injective). When
both properties hold, we say that P is fine, and this enables to see S[P ] as a
toric scheme. The other condition that turns out to be very useful is that P is
saturated: for all x ∈ P gp and n ∈ N∗ , we have that nx ∈ P =⇒ x ∈ P .
As usual, we define P = DS (P gp ) as the diagonalizable group scheme asso-
ciated to the (finitely generated) abelian group P gp . This group acts naturally
on S[P ].
76 Niels Borne
Now we will introduce toric morphisms. In §2.1 our model were tamely
ramified covers, that by Abhyankar’s lemma are locally induced by Kummer
covers. This is the justification of the following definition.
Definition 3.1. A morphism P → Q of monoids is Kummer if it is injective

and for each q ∈ Q, there exists n ∈ N∗ such that nq ∈ P .
Under the assumptions that P and Q are fine, and P → Q is Kummer, one
then shows the existence of the generalized Kummer sequence
→ P → 0
0 → μQ/P → Q
where μQ/P is a finite S-group scheme.
Our generalized Kummer cover will be, of course, S[Q] → S[P ]. If one re-
on S[Q] to μQ/P , this morphism becomes μQ/P -invariant,
stricts the action of Q
and it is easy to check, using the fact that Q is saturated and P → Q Kummer,
that OS [P ] is exactly the ring of invariants of OS [Q] under the action of μQ/P
([Tal14, Proposition 2.2.7]). In stacky terms, the morphism

S[Q]|μQ/P → S[P ]

identifies S[P ] with the coarse moduli space of the stack S[Q]|μQ/P . This
stack will be the local model for the generalized stack of roots, but before
defining these, we turn to a concrete example.
3.2. Sample: The Quadratic Cone. Let us consider the following

extension of monoids. We set Q = 12 N2 and
P = {(α, β) ∈ Q/ α + β ∈ N}
Since Q is free of rank 2, S[Q] = A2S . We will use the exponential notation for
the morphism Q → Γ(OS [Q]) and thus write q → xq , and similarly for P . For
this specific example, we will write abusively x = x(1/2,0) and y = x(0,1/2) so
that Γ(OS [Q]) = R[x, y].
Since P is generated as a monoid by (1, 0), (0, 1) and (1/2, 1/2), we have
that Γ(OS [P ]) = R[x2 , y 2 , xy]. By denoting X = x2 , Y = y 2 and Z = xy, we see
that R[x2 , y 2 , xy] = R[X, Y, Z]/(XY −Z 2 ). So S[P ] is a closed subscheme of A3
that is called the quadratic cone. The point corresponding to X = Y = Z = 0
is clearly singular.
The Kummer sequence is also easy to compute: Qgp = 12 Z2 and P gp =
( 2 , 2 )Z ⊕ ( 12 , − 12 )Z, hence there is an exact sequence
1 1
2Z
1
+
0 → P gp → Qgp −
→ →0
Z
and passing to the associated diagonalizable groups

Δ
0 → μ2 −→ G2m → G2m → 0
where Δ is the composite of the embedding μ2 → Gm with the diagonal Gm →

G2m .
Since P , Q are fine and saturated, and P → Q is Kummer, OS [P ] is the
ring of invariants of OS [Q] under the action of μQ/P = μ2 . In other words
p : S[Q] → S[P ] identifies with the quotient morphism A2 → A2 /μ2 where μ2
acts on A2 by ζ · (x, y) = (ζx, ζy).
But p : S[Q] → S[P ] is also a standard example of finite morphism which is
not flat. The simplest way to see this is that smoothness descends fppf locally, so
p cannot be flat, which can also be checked directly by inspecting the lengths of
the fibers. So if we set R0 = R[x2 , y 2 , xy], then
R[x, y] is not a flat R0 -module.
Similarly, the morphism π : S[Q]|μ Q/P → S[P
] is finite but isn’t flat (else
p itself would be flat since S[Q] → S[Q]|μQ/P is a μ2 -torsor, hence is flat).
This fact has serious
consequences, indeed the direct image of a locally free sheaf
on S[Q]|μQ/P via π does not need to be locally free on S[P ] any longer (we
will see a concrete
example
in §3.5). Hence, even if we want to analyze vector
bundles on S[Q]|μQ/P with a correspondence of the type given Theorem 2.4,
we cannot use parabolic sheaves defined in terms of vector bundles on X only,
but we have to allow arbitrary (quasi-)coherent components for our parabolic
sheaves instead.
3.3. Generalized Stack of Roots. As explained §1.3, we need to

introduce a refined notion of stack of roots, in other words, we want to be able
to define stack of roots for logarithmic schemes. Prototypes of these generalized
stack of roots were first introduced by M.Olsson ([MO05]). For simplicity, we
will explain their definition only in the case where the data is a Deligne-Faltings
log structure on a scheme X endowed with a chart, or, which amounts to the
same, a monoidal morphism L0 : P → Div(X).
Definition 3.2. Let X be a scheme, L0 : P → Div(X)

a monoidal morphism,
and P → Q a Kummer morphism. The stack of roots Q L0 /X is the stack of
liftings of L0 to Q.
We can give an alternative, more concrete definition by using a reformu-

lation. We claim that the data of a monoidal morphism L0 : P → Div(X) is
equivalent to the data of a morphism X → S[P ]|P . Indeed, L0 determines
a monoidal morphism P gp → Pic(X), that is nothing else than a P-torsor T
on X. If we write L0 : p → (Lp , sp ), then OT = ⊕p∈P gp Lp , and from this
description it follows that we get in this way from the sections sp a monoidal
78 Niels Borne
morphism PT → OT , or which is the same a morphism T → S[P ], which is

clearly P -equivariant. So we have obtained a morphism X → S[P ]|P , and
we leave it to the reader that it is possible to go the other way round. It follows
from this description that the following diagram is cartesian:

Q
L0 /X S[Q]|Q

L0
X S[P ]|P
If moreover the Deligne-Faltings log-structure admits a Kato chart, we get

the following cartesian diagrams:

Q
L0 /X S[Q]|μQ/P S[P ]|P

L0
X S[P ] S[P ]|P
In particular, the generalized stack of roots is a tame Artin stack in the

sense of [AOV08].
3.4. Parabolic Sheaves on Log Schemes. We now turn to

parabolic sheaves. We follow the policy of the previous paragraph and we will
work with the following setup: X a scheme, P → Q a Kummer morphism of
monoids, L0 : P → Div(X) a monoidal morphism.
We wish to generalize Definition 2.3 to this situation. In this prototype, the
Kummer morphism is the inclusion N → 1r N, and the index-category of our
parabolic sheaves is the category associated to the poset 1r Z (we in fact used
the opposite order for historical reasons, but we will from now on deal only
with covariant functors).
Definition 3.3. Let P be a monoid. We will denote by P wt the monoidal
category whose objects are elements of P gp , and such that a morphism p → p
is given by an element p in P such that p = p + p .
We are now ready to give the definition of parabolic sheaves on log schemes:
Definition 3.4. Let X be a scheme, P → Q a Kummer morphism of monoids,
L0 : P → Div(X) a monoidal morphism. A parabolic bundle on (X, L0 ) with
weights in Q is a functor
E· : Qwt → Qcoh(X)
endowed with a functorial isomorphism for each (q, p) ∈ Qgp × P gp :
Eq+p Eq ⊗OX Lp
satisfying the natural compatibility conditions.
In this definition, we use the notation Qcoh for the category of quasi-
coherent sheaves, and L0 (p) = (Lp , sp ).
It turns out that this definition of a parabolic sheaf is the right one for
our purposes. We denote by ParQ/P (X) the category of parabolic sheaves on
(X, L0 ) with weights in Q. There is a natural functor Qcoh X → ParQ/P (X) ,
E → E · left adjoint of the functor E· →
E0 . Under the following equivalence, it
corresponds to the pull-back along π : Q P/X → X.
Theorem 3.5 ([BV12], Theorem 6.1). Let X be a scheme, P → Q a Kummer

morphism of fine and saturated monoids, L0 : P → Div(X) a monoidal mor-
phism. We denote by π : Q P/X → X the morphism from the corresponding
stack of roots to its moduli space, and by M0 : Q → Div( Q P/X), q → (Mq , tq )
the canonical extension of L0 . Then the association:
F → (q → π∗ (F ⊗ Mq ))

induces an equivalence of categories between Qcoh( Q P/X) and ParQ/P (X).
3.5. Parabolic Sheaves on the Quadratic Cone. We illustrate

now the use of Theorem 3.5 thanks to the example of the quadratic cone.
We keep the notations of §3.2, in particular for convenience we write R0
for R[x2 , y 2 , xy], seen as a sub-algebra of R[x, y]. This is the algebra of the
quadratic cone S[P ].
We first analyze the Deligne-Faltings log structure induced by P →
Div(S[P ]) on the quadratic cone. As we have seen, this data is equivalent
to the data of a monoidal functor P wt → Pic(S[P ]). It is easy to check that
this functor sends the fundamental domain
(0, 1) (1, 1)
( 12 , 21 )
(0, 0) (1, 0)
of P wt to the following diagram in R0 -mod (identified with Qcoh(S[P ])):

80 Niels Borne
x2
R0 R0
xy
y2 R0 y2
xy
x2
R0 R0
All invertible sheaves in this diagram are trivial because we have pulled back
the canonical log structure on [S[P ]/P gp ] on the Kato chart S[P ].
The shape of the previous diagram and the pseudo-periodicity isomor-
phisms show that a general parabolic sheaf with weights in Q = N2 on the
quadratic cone S[P ] is uniquely determined by the data of three R0 -modules
(M0 , Mx , My ) and eight morphisms (α0 , α0 , α1 , α1 , β0 , β0 , β1 , β1 ) between them
so that the following diagram commutes in R0 -mod:
α0 α0
M0 Mx M0
xy
β0 β1 β0
α1 α1
y2 My M0 My
xy
β0 β1 β0
α0 α0
M0 Mx M0
x2
It may seem awkward at first sight, but it is actually pretty easy to produce
natural and interesting examples of such parabolic sheaves. Let us start with
the simplest example, the parabolic sheaf OX · associated to the structure sheaf
OX . A straightforward computation using Theorem 3.5 shows that this sheaf
is given by the following diagram:
x x
R0 R1 R0
y y y
x x
R1 R0 R1
y y y
x x
R0 R1 R0
where R1 = xR0 + yR0 . Since R[x, y] = R0 ⊕ R1 , and R[x, y] is not R0 -

flat, it follows that R1 is not R0 -flat. This illustrates the need of considering
parabolic sheaves whose components are not locally free, even if we are only
interested in locally free parabolic sheaves (that is parabolic sheaves with locally
free counterparts on the stack of roots). Of course, for any R0 -module M ,

one can just apply the functor · ⊗R0 M to the above diagram, and get an
associated parabolic sheaf. This gives a concrete description of the functor
Qcoh(S[P ]) → ParQ/P (S[P ]).
But there are more interesting examples, of course. The simplest one not of
the previous type is obtained by twisting OX · :
R1 x R 0 x R 1
y y y
x x
R0 R1 R0
y y y
x x
R1 R0 R1
So set L· = OX [(α, β)]· for any (α, β) in Qgp = 12 Z2 such that α + β ∈ / Z (the
result is clearly independent of the choice of such a couple (α, β)). Although
the tensor product of parabolic sheaves is notoriously complicated, it is easy
to imagine that it behaves additively with respect to the twist so that L· is an
order 2 invertible parabolic sheaf on S[P ]. And this is indeed whathappens:
one checks that L· corresponds tothe order 2 invertible sheaf on Q L0 /S[P ]
given by the μ2 -torsor S[Q] → Q L0 /S[P ] (that we have already identified
with A2 → [A2 /μ2 ]).
It is also interesting to compare parabolic sheaves on the quadratic cone with
previously known cases, that is parabolic sheaves relative to a normal crossings
divisor. Let us now choose for P → Q the natural embedding 12 N2 → N2 , so that
S[Q] → S[P ] is the standard μ2 × μ2 -cover A2 → A2 given by (x, y) → (x2 , y 2 ).
Then a similar analysis shows that if R0 = R[x2 , y 2 ], a general parabolic sheaf
with weights in Q = N2 on the plane S[P ] is determined by the data of four R0 -
modules (M0 , Mx , My , Mxy ) and eight morphisms (α0 , α0 , α1 , α1 , β0 , β0 , β1 , β1 )
between them so that the following diagram commutes in R0 -mod:
α0 α0
M0 Mx M0
β0 β1 β0
α1 α1
y2 My Mxy My
β0 β1 β0
α0 α0
M0 Mx M0
2
x
So parabolic sheaves are still objects of combinatorial nature, but in their new
logarithmic version, they are able to take into account and reflect singularities
of the base space.
82 Niels Borne
3.6. Globalization. In this section, our set up was : X a scheme, P → Q

a Kummer morphism of monoids, L0 : P → Div(X) a monoidal morphism. But,
in general, we will meet Deligne-Faltings log structures that admit a chart only
étale locally (see §2.2.2). So if we start from a Deligne-Faltings log structure
A → DivX , we will have to replace the Kummer morphism of monoids P → Q
by
a Kummer morphism of sheaves of monoids A → B. There is a stack of roots
B
L/X obtained as a global version of Definition 3.2 (the idea of considering
stacks parametrizing extension of log structures is originally due to M.Olsson,
see [MO05]).
One can also extend the definition of parabolic bundles (Definition3.4) to
get a category ParB/A (X). Of course, the correspondence with Vec( B L/X),
that is, the analog of Theorem 3.5, still holds. All the details of these construc-
tions can be found in our original article [BV12].
4. Survey of Recent Work

In this last section, we very briefly report on more recent work.
4.1. The Infinite Root Stack. In [TV14], M.Talpo and A.Vistoli in-
vestigate further the relationship between logarithmic geometry and stack of
roots.
The interplay between logarithmic geometry and algebraic stacks has been
studied in depth by M.Olsson (see [Ols03]). The new idea of Talpo-Vistoli is to
introduce an infinite root stack that is, in some sense, simpler than the stacks
used by M.Olsson.
The definition is, roughly, as follows. Let us start from a fine and saturated
Deligne-Faltings log structure L : A → DivX on a scheme X. Then for each
n ∈ N∗ , we can consider the natural Kummer morphism A → n1 A, and the
1

corresponding stack of roots n A L/X. Going to the projective limit, we get
the infinite root stack, denoted by ∞ L/X.
Even if it is not an algebraic stack, the infinite root stack turns out to be
a very interesting object. It is locally the quotient of an affine scheme by a
diagonalizable group. Moreover, according to the first main result of [TV14],
the infinite root stack reflects faithfully the log structure it is associated with,
in the sense that it enables to reconstruct (X, L).
The infinite stack of roots is also useful to give a simple definition of a
quasi-coherent sheaf on a log scheme. The previous approach was to use the
Kummer étale ringed topos. But it turns out to be much more convenient to
define quasi-coherent sheaves on a log scheme as usual quasi-coherent sheaves
on the corresponding infinite root stack. This is the point of view advocated by
Talpo-Vistoli, which paves the way to the study of the K-theory of log schemes
(since the K-theory of algebraic stacks is fairly well understood). Moreover,

the second main result of [TV14] states that the correspondence between quasi-
coherent sheaves on the infinite root stack and parabolic sheaves with arbitrary
weights on the log scheme is still valid.
4.2. Moduli of Parabolic Sheaves. Let us end this survey by men-

tioning M.Talpo’s subsequent work (see [Tal14]) on moduli spaces of parabolic
sheaves on a fairly general log scheme.
In its simplest expression, when the denominators of the weights are
bounded, this study relies on one hand on the above mentioned correspon-
dence with stacky sheaves established in [BV12], and on the other hand on
F. Nironi’s construction of the moduli space of coherent sheaves on a Deligne-
Mumford stack (see [Nir08]). When the base log scheme is projective, fine and
saturated, and endowed with a global chart, the first main result of [Tal14]
shows the existence of a moduli stack of parabolic sheaves that is an Artin
stack (of finite type if one fixes the Hilbert polynomial).
The second main result of [Tal14] deals with the much more delicate sit-
uation when the denominators of the weights are allowed to grow arbitrarily.
It is then possible, partly thanks to [TV14], to proceed to the analysis of the
corresponding tower of moduli spaces.
Acknowledgements
I would like to express my gratitude to my co-author Angelo Vistoli, especially
for his invaluable help during the preparation of this manuscript.
References
[AGV08] Dan Abramovich, Tom Graber, and Angelo Vistoli. “Gromov-Witten theory
of Deligne-Mumford stacks”. In: Amer. J. Math. 130.5 (2008), pp. 1337-
1398. issn: 0002-9327. url: http://dx.doi.org/10.1353/ajm.0.0017.
[AOV08] Dan Abramovich, Martin Olsson, and Angelo Vistoli. “Tame stacks
in positive characteristic”. In: Ann. Inst. Fourier (Grenoble) 58.4
(2008), pp. 1057–1091. issn: 0373-0956. url: http://aif.cedram.
org/item?id=AIF 2008 58 4 1057 0.
[BV12] Niels Borne and Angelo Vistoli. “Parabolic sheaves on logarithmic
schemes”. In: Adv. Math. 231.3-4 (2012), pp. 1327–1363. issn: 00018708.
url: http://dx.doi.org/10.1016/j.aim.2012.06.015.
[Bis97] Indranil Biswas. “Parabolic bundles as orbifold bundles”. In:
Duke Math. J. 88.2 (1997), pp. 305–325. issn: 0012-7094. url:
http://dx.doi.org/10.1215/S0012-7094-97-08812-8.
84 Niels Borne
[Bor07] Niels Borne. “Fibrés paraboliques et champ des racines”. In: Int. Math.
Res. Not. IMRN 16 (2007), Art. ID rnm049, 38. issn: 10737928. url:
http://dx.doi.org/10.1093/imrn/rnm049.
[Bor09] Niels Borne. “Sur les représentations du groupe fondamental d’une
variété privée d’un diviseur à croisements normaux simples”. In: In-
diana Univ. Math. J. 58.1 (2009), pp. 137–180. issn: 0022-2518. url:
http://dx.doi.org/10.1512/iumj.2009.58.3734.
[Cad07] Charles Cadman. “Using stacks to impose tangency conditions on curves”.
In: Amer. J. Math. 129.2 (2007), pp. 405–427. issn: 00029327. url:
http://dx.doi.org/10.1353/ajm.2007.0007.
[Kat89] Kazuya Kato. “Logarithmic structures of Fontaine-Illusie”. In: Algebraic
analysis, geometry, and number theory (Baltimore, MD, 1988). Johns Hop-
kins Univ. Press, Baltimore, MD, 1989, pp. 191–224.
[MO05] Kenji Matsuki and Martin Olsson. “Kawamata-Viehweg vanishing as Ko-
daira vanishing for stacks”. In: Math. Res. Lett. 12.2-3 (2005), pp. 207–217.
issn: 1073-2780. url: http://dx.doi.org/10.4310/MRL.2005.v12.n2.a6.
[MS80] V. B. Mehta and C. S. Seshadri. “Moduli of vector bundles on curves with
parabolic structures”. In: Math. Ann. 248.3 (1980), pp. 205–239. issn: 0025-
5831. url: http://dx.doi.org/10.1007/BF01420526.
[Nir08] Fabio Nironi. Moduli Spaces of Semistable Sheaves on Projective Deligne-
Mumford Stacks. 2008. arXiv: arXiv:0811.1949 [math-AG].
[Ols03] Martin C. Olsson. “Logarithmic geometry and algebraic stacks”. In: Ann.
Sci. École Norm. Sup. (4) 36.5 (2003), pp. 747–791. issn: 00129593. url:
http://dx.doi.org/10.1016/j.ansens.2002.11.001.
[TV14] Mattia Talpo and Angelo Vistoli. Infinite root stacks and quasi-coherent
sheaves on logarithmic schemes. 2014. arXiv: arXiv:1410.1164 [math-AG].
[Tal14] Mattia Talpo. Moduli of parabolic sheaves on a polarized logarithmic
scheme. 2014. arXiv: arXiv:1410.2212 [math-AG].
A Survey of Ulrich Bundles
Emre Coskun∗,†
Abstract
The purpose of this article is to serve as an introduction to Ulrich bundles for

interested readers. We discuss the origins of the study of Ulrich bundles, their
elementary properties, and we give an exposition of the known results in their
classification on various kinds of smooth projective varieties. As an illustration
of a method recently used by Marta Casanellas and Robin Hartshorne for the
classification of Ulrich bundles on a cubic surface, we classify the stable Ulrich
bundles on a smooth quartic surface in P3 of Picard number 2 by first Chern
class.
1. Introduction and History

The study of Ulrich bundles originates in an article by Bernd Ulrich that ap-
peared in 1984. In [42], Ulrich investigated criteria for a local Cohen-Macaulay
ring to be Gorenstein. The second of these criteria involved finitely generated
modules M of positive rank over a local Cohen-Macaulay ring R. It can be
easily shown that, in this case, the minimal number of generators ν(M ) of
M is bounded above by the product of the multiplicity of R and the rank of
M . [42, Theorem 3.1] gives necessary conditions for R to be Gorenstein; one
condition is the existence of a finitely generated R-module M of positive rank
such that the minimal number of generators is greater than half of the product
mentioned above. In view of this result, Ulrich asked the following question
([42, p. 26]):
∗ Department of Mathematics, Middle East Technical University, 06800 Ankara, TURKEY
† The author was supported by TÜBİTAK scholarship 2232. He would also like to thank
the Tata Institute of Fundamental Research and the Institut des Hautes Études Scientifiques,
where he conducted research on Ulrich bundles on K3 surfaces.
Geometry, DOI 10.1007/978-981-10-5648-2_6
86 Emre Coskun
Let R be a local Cohen-Macaulay ring with positive dimension and

infinite residue class field. Does there exist a Cohen-Macaulay R-
module M with positive rank such that ν(M ) = e(R)rank(M )?
Three years later Joseph P. Brennan, Jürgen Herzog and Bernd Ulrich
investigated these modules, which they called ‘Maximally Generated Cohen-
Macaulay Modules’ (MGMCM) in [5].1 Specifically, they proved that a homo-
geneous, two-dimensional Cohen-Macaulay domain R with infinite residue class
field admits an MGMCM module. They also raised a number of questions on
the existence of MGMCM modules on various kinds of rings.
The term ‘Ulrich module’ as a synonym for a MGMCM was coined by Jörgen
Backelin and Jürgen Herzog in [3]. In that article, they proved the existence
of Ulrich modules on hypersurface rings. That result added to the previously
known existence results.
It was Arnaud Beauville who in [4], as far as we can ascertain, made a
systematic exposition of the relationship between Ulrich bundles and the de-
terminantal or Pfaffian representations of hypersurfaces; and he used these
results to reprove a number of classical results (plane curves, cubic surfaces
. . . ), as well as new results on threefolds and fourfolds.
In the meantime, there had been another line of inquiry that would lead to
the investigation of Ulrich bundles. The generalized Clifford algebra and its ma-
trix representations (see Section 2, (ii) for definitions) had been the subject of a
number of papers. For instance, in [27], Darrell Haile proved that the center of
the generalized Clifford algebra Cf of a binary cubic form f modulo the inter-
section of its three-dimensional representations was an Azumaya algebra over
its center, and that its center was the coordinate ring of an affine elliptic curve.
Three years later, Michel van den Bergh proved in [43] that the rd-dimensional
representations2 of the generalized Clifford algebra Cf of a binary form f of
degree d are in one-to-one correspondence with vector bundles on a plane curve
X associated to f whose pushforward under a linear projection X → P1 are
trivial. (Ulrich bundles are precisely the vector bundles satisfying an analogous
property; see Definition 3.2.) Rajesh Kulkarni used van den Bergh’s results to
prove an analogue of Haile’s result mentioned above; he proved in [34] that
the reduced Clifford algebra is Azumaya over its center; and the center is the
affine coordinate ring of the complement of a Θ-divisor in Picd+g−1 (X) for a
suitably defined plane curve X of genus g. Later on, using similar techniques,
the present author proved in [11] that the rd-dimensional representations of
1 The term ‘Ulrich module’ has been coined by Jürgen Herzog and Michael Kühl in the
article [30].
2 Van den Bergh proves that the matrix representations of C must have dimensions that
f
are multiples of d; the same result is also proved by Darrell Haile and Steven Tesser in [28].
A Survey of Ulrich Bundles 87
the generalized Clifford algebra Cf of a binary form have a fine moduli space
that is a nonempty open subset of the moduli space of vector bundles over a
suitable plane curve X.
This paper is intended as a survey of known results of Ulrich bundles. In
Section 2, we discuss the beginnings of Ulrich bundles; this section is divided
into three parts. In the first part (i), we discuss the relations between Ulrich
bundles and the problem of expressing the defining equation of a hypersurface
as the determinant or the Pfaffian of a matrix linear forms. In the second part
(ii), we discuss the relations between Ulrich bundles and the representation
theory of generalized Clifford algebras; the material in this part is due to van
den Bergh ([43]). In the third part (iii), we discuss the relations between Ulrich
bundles on del Pezzo surfaces and the Minimal Resolution Conjecture for a
finite collection of sufficiently many points lying on the del Pezzo surface. In
Section 3, we give two equivalent definitions of Ulrich bundles and collect some
elementary properties of Ulrich bundles that will be used later. In Section 4,
we outline known results in the classification of Ulrich bundles. This section
is arranged according to the dimension of the underlying variety. Finally, in
Section 5, we use a recent technique due to Casanellas and Hartshorne to
give a classification of Ulrich bundles according to the first Chern class on a
determinantal quartic surface in P3 with Picard number 2.
Notation and Conventions.

• We work over an algebraically closed base field k of characteristic 0.
• All varieties mentioned in the text are implicitly assumed to be smooth

and projective.
• X ⊂ PN denotes a variety, of degree d and dimension n. Note that X is

polarized by OX (H), where H denotes the hyperplane section of X. The
inclusion map is i : X → PN .
• We use the term vector bundle for a locally free sheaf.
• We denote OX (H)⊗t by simply OX (t) for t ∈ Z.
• Stability and semistability are used in the sense of Gieseker.
2. Origins
(i) How can we determine whether a hypersurface X ⊂ Pn+1 is given by a
determinantal or a Pfaffian form?
88 Emre Coskun
Theorem 2.1 ([4, Sections 1 and 2]). Assume that X ⊂ Pn+1 is a hypersurface.
(a) X is defined by an equation of the form det(M ) for a d × d matrix M of

linear forms if and only if there exists a short exact sequence
M
0 → OPn+1 (−1)d −→ OPdn+1 → i∗ L → 0
where L is a line bundle on X. (If one wants M to be symmetric, one

imposes the condition that L⊗2 ∼
= OX (d − 1).)
(b) X is defined by an equation of the form pf(M ) for a skew-symmetric 2d×2d

matrix M of linear forms if and only if there exists a short exact sequence
M
0 → OPn+1 (−1)2d −→ OP2dn+1 → i∗ E → 0
where E is a vector bundle of rank 2 on X with det(E) ∼

= OX (d − 1).
Note that the existence of these short exact sequences implies that
H i (X, L(t)) = 0 for t ∈ Z and 0 < i < n; a similar statement holds for E.
Note also that the Hilbert polynomials of L and E are given by dr t+n
n , where
r = 1, 2 is the rank of the corresponding vector bundle.
Definition 2.2 (First definition of Ulrich bundles). An Ulrich bundle on X

t ∈ Z and
is a vector bundle E of rank r on X such that H i (X, E(t)) = 0 for
0 < i < n, and whose Hilbert polynomial PE (t) is equal to dr t+n
n .
We now see that the representation of a hypersurface as a (symmetric)

determinantal or Pfaffian variety depends on the existence of Ulrich bundles
with certain properties.
The defining exact sequence above gives h0 (X, L) = d and h0 (X, L(−1)) =
0. Suppose now that L ∼ = OX (n0 ) for some n0 ∈ Z. It follows that n0 = 0
and L ∼ = X O . This forces the trivial case d = 1. Hence if Pic(X) is gener-
ated by H, X cannot be a determinantal variety unless it is linear. By the
Noether-Lefschetz and the Grothendieck-Lefschetz theorems, this leaves curves
and general surfaces of degree at most 3 as the nontrivial cases for determinantal
varieties (there are nongeneral surfaces of higher degree that are determinantal,
as we will see in Section 5).
Theorem 2.3 (Noether-Lefschetz, adapted statement from [24, Theorem (*)]).

If d ≥ 4, then a general surface X ⊂ P3 has Pic(X) generated by H.
Theorem 2.4 (Grothendieck-Lefschetz, adapted statement from [26, Corol-

laire 3.6]). Let X ⊂ PN and let H be a general hyperplane section. Then the
canonical morphism Pic(X) → Pic(X ∩ H) is an isomorphism if n > 3.
Remark 2.5. The Grothendieck-Lefschetz theorem implies in particular that,

if N > 3, the Picard group Pic(X) of the general hypersurface X ⊂ PN is
generated by the hyperplane section H.
Example 2.6. It is a classical result that a cubic surface X ⊂ P3 is deter-

minantal; and there are 72 different ways of writing its defining equation as
a 3 × 3 determinant of linear forms in four variables. We will see later that
the Ulrich line bundles on X are precisely the line bundles of the form OX (T )
where T is a twisted cubic on X; it is easily shown that there are 72 of them.
(ii) Let f (x1 , . . . , xn+1 ) be a form of degree d in n+1 variables x1 , . . . , xn+1 .

Define
Cf := k{x1 , . . . , xn+1 }/I,
where I ⊂ k{x1 , . . . , xn+1 } is the two-sided ideal generated by elements of the
form (α1 x1 + · · · + αn+1 xn+1 )d − f (α1 , . . . , αn+1 ) with α1 , . . . , αn+1 ∈ k. Cf is
called the generalized Clifford algebra of f .
We are interested in equivalence classes of matrix representations of Cf .
A matrix representation of Cf of dimension m is a k-algebra homomorphism
φ : Cf → Mm (k). Two matrix representations φ, ψ of Cf of the same dimension
m are called equivalent if there exists an invertible matrix A ∈ Mm (k) such that
ψ = AφA−1 . It turns out that m must be divisible by d; see [12, Proposition
2.3] and [28, Proposition 1.1]. We write m = rd.
Using f , we construct a hypersurface Xf ⊂ Pn+1 of degree d by the equation
wd = f (x1 , . . . , xn+1 ). Assume that f is chosen so that Xf is smooth. The
general linear projection π : Xf → Pn is a finite map of degree d.
Theorem 2.7 ([43, Proposition 1]). There is a one-to-one correspondence

between the equivalence classes of rd-dimensional matrix representations of
Cf and isomorphism classes of rank r vector bundles E on Xf such that
π∗ E ∼
= OPrdn .
Sketch. For a given representation φ : Cf → Mrd (k), let Ai = φ(xi ) ∈ Mrd (k)
for i = 1, . . . , n + 1. Using these matrices, we construct a ring homomorphism
S = k[x1 , . . . , xn+1 ]/(wd − f (x1 , . . . , xn+1 )) → Mrd (k[x1 , . . . , xn+1 ])
by sending xi to xi Ird and w to x1 A1 + · · · + xn+1 An+1 . This makes

k[x1 , . . . , xn+1 ]rd into a graded S-module; hence we obtain a coherent sheaf
F on X such that π∗ F ∼ = OPrdn . It turns out that this is a rank r vector bundle
on X.
Let us now assume that X ⊂ PN is any variety. Recall that we denote

n = dim X. Let π : X → Pn be a linear projection.
90 Emre Coskun
Definition 2.8 (Second definition of Ulrich bundles). An Ulrich bundle on X

is a vector bundle E of rank r on X such that π∗ E ∼
= OPrdk .
Remark 2.9. The two definitions of Ulrich bundles are equivalent. See Theorem
3.3.
(iii) The Minimal Resolution Conjecture (MRC) concerns the minimal free
resolutions of the ideal sheaves of collections Γ of points lying on subvarieties
Y ⊂ PN . The MRC was first studied by Anna Lorenzini in [36]; she consid-
ered the minimal resolutions of sufficiently general sets of points in PN . Gavril
Farkas, Mircea Mustaţǎ and Mihnea Popa generalized this conjecture in [22]
to sets of points that lie on varieties X ⊂ PN .
For a subvariety Y ⊂ PN , write
0 → ⊕ljk =1 OPN (−jk )bk,jk (Y ) → · · · ⊕lj1 =1 OPN (−j1 )b1,j1 (Y ) → IY |PN → 0
for the minimal free resolution of IY |PN . The number l + 1 is the Castelnuovo-
Mumford regularity reg(IY |PN ).
Definition 2.10 ([14, Definition 4.1]). Let Γ be a zero-dimensional subscheme
of Y . We say that Γ satisfies the MRC for Y if
bi+1,q−1 (Γ) · bi,q (Γ) = 0
for all i and whenever q ≥ reg(IY |PN ) + 1.
For general Γ consisting of sufficiently many points, the Betti diagram of Γ
contains the Betti diagram of Y together with two rows at the bottom. Also,
it was proved in [22, Theorem 1.2 (iv)] that the difference bi+1,q−1 (Γ) − bi,q (Γ)
is known. Therefore, knowing that the MRC holds would let us determine the
Betti diagram of Γ completely.
For a variety X ⊂ PN , define the Ulrich semigroup Ulr(X) to be the semi-
group consisting of the first Chern classes of Ulrich bundles on X. Since the
direct sum of two Ulrich bundles is again Ulrich by [12, Proposition 2.14], this
set is indeed a semigroup.
On a del Pezzo surface, this conjecture is related to the existence of Ulrich
bundles in the following way:
Proposition 2.11 ([14, Theorem 1.5]). Let Xd ⊂ Pd be a del Pezzo surface of
degree d. Suppose that for every generator Q of Ulr(Xd ), the MRC holds for a
general smooth curve in |Q|. Then for an effective divisor D ⊂ Xd which does
not lie on a hyperplane, the following are equivalent:
1. There exists an Ulrich bundle E of rank r on Xd with c1 (E) = D.
2. The degree of D is dr, and the MRC holds for a general smooth curve in
the linear system |D|.
Remark 2.12. In [14, Proposition 2.19], it was shown that the Ulrich line bun-
dles on Xd are the line bundles of the form OXd (Q), where Q ⊂ Xd is the
class of a rational normal curve; this also follows from [40, Theorem 1.1]. Using
[22, Corollary 1.8], it was shown in [14, Lemma 4.3] that a rational normal
curve Q ⊂ Xd satisfies the MRC. Now, if Ulr(Xd ) is generated by the classes
of the rational normal curves, the hypotheses of Proposition 2.11 are satisfied.
This is true for d = 3 by [6, Theorem 3.9], but false for d = 6, 7. This is a
consequence of [14, Proposition 3.7] that states that there is a rank-2 Ulrich
bundle E with c1 (E) = H + Q and the fact that for d = 6, 7 it is not possible
to write H + Q as the sum of two rational normal curves. For d = 4, 5 it is
possible to write H +Q as the sum of two rational normal curves; but we do not
yet know whether Ulr(Xd ) is generated by the classes of the rational normal
curves.
3. Properties of Ulrich Bundles

In this brief section, we recall certain properties of Ulrich bundles, some of
which will later be used.
We invite the reader to recall Definitions 2.2 and 2.8 now. We point out that
the definitions can be trivially generalized to any variety X ⊂ PN of dimension
n. For convenience, we restate the definitions here. Note that in the second
definition, π : X → Pn is any linear projection.
Definition 3.1 (First definition of Ulrich bundles). An Ulrich bundle on X
t ∈ Z and
is a vector bundle E of rank r on X such that H i (X, E(t)) = 0 for
0 < i < n, and whose Hilbert polynomial PE (t) is equal to dr t+n
n .
Definition 3.2 (Second definition of Ulrich bundles). An Ulrich bundle on X

is a vector bundle E of rank r on X such that π∗ E ∼
= OPrdn .
These definitions are equivalent by the following result.
Theorem 3.3. If n ≥ 2, the two definitions of Ulrich bundles are equivalent.
If X is a curve, a vector bundle E of rank r on X is Ulrich if and only if
deg(E) = r(d + g − 1) and h0 (X, E(−1)) = 0.
Proof. If X is a curve, the proof of [43, Lemma 2, (a)] can easily be generalized.
If n ≥ 2, the result is proved in [14, Proposition 2.3].
Remark 3.4. Definition 3.2 implies that, on a linear subvariety Pn ⊂ PN , the
only Ulrich bundles are the trivial bundles.
As is well known, the theory of moduli spaces of vector bundles requires
some sort of stability condition. The two main definitions of stability are
92 Emre Coskun
Gieseker stability (using the reduced Hilbert polynomial) and μ-stability or

slope-stability (using the slope). It turns out that Ulrich bundles have desir-
able stability properties.
Proposition 3.5 ([12, Proposition 2.11]). An Ulrich bundle is semistable.
Proposition 3.6 ([12, Corollary 2.16]). The Jordan-Hölder factors of an Ul-

rich bundle are again Ulrich.
Remark 3.7. The proofs of [12, Proposition 2.11] and [12, Corollary 2.16], even
though stated for hypersurfaces, are trivially generalized.
These propositions imply that any Ulrich bundle is obtained from a finite
collection of stable Ulrich bundles by consecutive extensions. Therefore, from
now on we will focus on stable Ulrich bundles.
4. Classification of Ulrich Bundles

4.1. Curves. Suppose that X ⊂ PN is a curve of degree d and genus g.
We use Theorem 3.3 for the classification of Ulrich bundles on X.
4.1.1. g = 0. Let E be an Ulrich bundle on X of rank r. Then E ∼ = L1 ⊕

· · · ⊕ Lr by the theorem of Grothendieck3 that states that any vector bundle on
P1 splits as a sum of line bundles ([25, Théorème 2.1]). Proposition 3.5 implies
that E is semistable and hence that the Li all have the same degree, say m.
We now get
deg(E) = rm = r(d − 1),
and hence m = d − 1 ≥ 0. It follows that E is simply the direct sum of r line
bundles of degree d − 1.
Conversely, we will show that a line bundle L of degree d − 1 on X is Ulrich;
it will then follow that direct sums of such line bundles are also Ulrich. Let
π : X → P1 be any linear projection. Write
π∗ L = F 1 ⊕ · · · ⊕ F d ,
! on P ⊂ P . Since the degree of L is d − 1,

1 N
where the Fi are line bundles
h0 (X, L(−1)) = 0 and hence h0 (P1 , Fi (−1)) = 0. This implies !that deg(Fi ) ≤
0. Since h0 (X, L) = d by the Riemann-Roch theorem, we have h0 (P1 , Fi ) = d
and this implies that deg(Fi ) = 0 and therefore Fi ∼
= OP1 , i.e. that L is Ulrich.
3 In fact, the content of this theorem was known before Grothendieck; see a note of
Winfried Scharlau at http://wwwmath.uni-muenster.de/u/scharlau/scharlau/grothendieck/

Grothendieck.pdf.
Since line bundles are completely classified by degree on a genus 0 curve,

we conclude that there is a unique Ulrich line bundle on X up to isomorphism,
and all Ulrich bundles on X are direct sums of copies of this line bundle.
4.1.2. g = 1. One can use Atiyah’s classification (see [2]) of vector bundles
on elliptic curves to classify Ulrich bundles on an elliptic curve.
Let X ⊂ PN be a curve of genus g = 1, and let E be a vector bundle of
rank r on X. By Theorem 3.3, E is Ulrich if and only if deg(E) = rd and
h0 (X, E(−1)) = 0. Since E and hence E(−1) is semistable by Proposition 3.5,
we can write
E(−1) = ⊕Ei ,
where the Ei are indecomposable vector bundles of rank ri , degree 0 and no

global sections. By [2, Theorem 5, (ii)] we can write Ei ∼
= Fri ⊗ Li , where Fri is
the unique indecomposable vector bundle on X of rank ri , degree 0 and nonzero
global sections, and Li is a line bundle of degree 0 determined uniquely by Ei .
The fact that h0 (X, Ei ) = 0 implies that Li OX . Summing up, we can write
E = ⊕(Fri ⊗ Li (1)),
where the Li are nontrivial line bundles of degree 0.

Conversely, it can be easily seen that any vector bundle E of the form
⊕(Fri ⊗ Li (1))
!
for various Fri and nontrivial line bundles Li of degree 0 has rank r = ri ,
degree rd and h0 (X, E(−1)) = 0. In other words, it is Ulrich.
4.1.3. g ≥ 2. Recall that there is a divisor Θ ⊂ Picd+g−1 (X), called the

theta-divisor, consisting of line bundles L on X such that H 0 (X, L(−1)) = 0.
It is therefore clear that Ulrich line bundles on X are precisely those line bundles
outside of the theta-divisor.
Let us now consider a given arbitrary rank r ≥ 2. In the case where X =
Xf ⊂ P2 is the smooth curve associated to a binary form f (u, v) that is not
divisible by a square, the existence of stable Ulrich bundles of rank r on X was
established by van den Bergh in [43, Theorem 4].
We would like to give a proof that there exist stable Ulrich bundles on X of
any given rank r ≥ 2 using a method that was first used by Marta Casanellas
and Robin Hartshorne in [6] to classify stable Ulrich bundles on a cubic surface.
Theorem 4.1. There exist stable Ulrich bundles of any rank r ≥ 2 on X.

94 Emre Coskun
Proof. Since Ulrich line bundles on X correspond to points outside a divisor in

Picd+g−1 (X), we can take two nonisomorphic Ulrich line bundles L1 and L2 .
By the Riemann-Roch theorem, we have
ext1 (L1 , L2 ) = h1 (X, L∨
1 ⊗ L2 )
∨
= −χ(L1 ⊗ L2 )
= g − 1 > 0.
Hence, there exist nonsplit extensions of L1 by L2 ; and these are simple by
[6, Lemma 4.2]. As Ulrich bundles, they are semistable by Proposition 3.5.
Let E be one such extension. The dimension of the moduli space of vector
bundles near [E] is
h1 (X, E ⊗ E ∨ ) = 1 − χ(E ⊗ E ∨ )
= 1 − (4(1 − g))
= 4(g − 1) + 1.
Now let us consider the modular dimension of strictly semistable Ulrich
bundles of rank 2. These are obtained by extensions of two Ulrich line bundles
L1 and L2 as above; and their dimension count is
g + g + ext1 (L1 , L2 ) − 1 = 3g − 2.
It is clear that 4(g − 1) + 1 > 3g − 2. Hence the dimension of Ulrich bundles
of rank 2 on X exceeds the dimension of the strictly semistable ones; and that
implies the existence of stable Ulrich bundles of rank 2.
For higher ranks r ≥ 3, we proceed similarly. Assume that there is a stable
rank r−1 Ulrich bundle F on X. Let L be an Ulrich line bundle on X. Note that
h0 (X, F ∨ ⊗ L) = 0 since F and L are both stable of the same slope and they
are not isomorphic; any nonzero homomorphism between them would have to
be an isomorphism by [31, Proposition 1.2.7]. Therefore by the Riemann-Roch
theorem, we have
ext1 (F, L) = h1 (X, F ∨ ⊗ L)
= −χ(F ∨ ⊗ L)
= (r − 1)(g − 1) > 0.
Just as before, there exist nonsplit extensions of F by L; and these are simple
by [6, Lemma 4.2]. If E is one such extension, then the dimension of the moduli
space of vector bundles near [E] is
h1 (X, E ⊗ E ∨ ) = 1 − χ(E ⊗ E ∨ )
= 1 − (r2 (1 − g))
= r2 (g − 1) + 1.
Next, we have to consider the modular dimension of strictly semistable Ulrich

bundles of rank r. There are various ways to obtain such bundles as consecutive
extensions of stable Ulrich bundles of strictly smaller rank, but by the discussion
that follows [6, Remark 4.6] it is enough to consider extensions of a stable Ulrich
bundle F of rank r − 1 and an Ulrich line bundle L. And their dimension count
is
(r − 1)2 (g − 1) + 1 + g + (r − 1)(g − 1) − 1.
Again, it is not too difficult to show that
r2 (g − 1) + 1 > (r − 1)2 (g − 1) + 1 + g + (r − 1)(g − 1) − 1,
establishing the existence of Ulrich bundles of rank r on X that are not strictly
semistable, hence stable.
Thus, we know the existence of stable Ulrich bundles on any rank on curves.
Since semistability and cohomology vanishing are open conditions, Ulrich bun-
dles are in one-to-one correspondence with the geometric points of a nonempty
open subset of MX (r, r(d + g − 1)).
4.2. Surfaces. Suppose now that X ⊂ P3 is a surface.
4.2.1. Cubic surfaces (d = 3). In this case, the Ulrich bundles of ranks
1 and 2 on X can be classified using the results in the article [20] by Daniele
Faenzi. Ulrich line bundles on X turn out to be OX (T ) for T a twisted cubic
in X. (Recall that in general, the Ulrich bundles on a del Pezzo surface are the
line bundles corresponding to the rational normal curves by [14, Proposition
2.19].) As there are 72 such classes, there are 72 Ulrich line bundles on X; and
hence there are 72 different representations of X as a determinantal variety,
agreeing with the classical result4 in [23]. One can also construct Ulrich bundles
of any rank on X: Casanellas and Hartshorne constructed rank r stable Ulrich
bundles on X with first Chern class rH in [6]; in particular, X can be seen to
be nontrivially Pfaffian since a stable Ulrich bundle of rank 2 cannot be written
as the direct sum of two Ulrich line bundles. Casanellas and Hartshorne also
classified the possible first Chern classes of stable Ulrich bundles. These turn out
to be sums of r twisted cubics satisfying a numerical condition; see [6, Theorem
1.1] for details.
4 For more information about the history of research on the geometry of cubic surfaces,
we invite the reader to consult the article [19] by Igor Dolgachev.

96 Emre Coskun
4.2.2. Quartic surfaces (d = 4). Results are limited.

According to [4, Proposition 7.6], the general surface X ⊂ P3 of degree d is
Pfaffian if and only if d ≤ 15. By Theorem 2.1, this implies in particular that
the general quartic surface X ⊂ P3 carries a rank 2 Ulrich bundle. However,
there are no Ulrich line bundles on X as Pic(X) is generated by H and any
Ulrich line bundle on X would have degree 6 by [15, Proposition 2.8].
Using the method of Casanellas and Hartshorne, and the existence of rank
2 simple Ulrich bundles with c1 = 3H on any smooth quartic ([13]), the present
author showed that for every even rank 2r, there exists a stable Ulrich bundle
of rank 2r and first Chern class 3rH on the general quartic surface X ⊂ P3 .
(See [15, Theorem 3.1].) This result was later generalized in [1, Theorem 0.5]
to any projective K3 surface.
It was proved by the present author, Rajesh Kulkarni and Yusuf Mustopa
in [13, Theorem 1.1] that any quartic surface is Pfaffian by constructing a 14-
dimensional family of simple Ulrich bundles of rank 2 and first Chern class 3H.
The idea is to construct this bundle E as in the short exact sequence
0 → E ∨ → OX
2
→ L → 0,
where L is a line bundle with suitable properties on a smooth curve C ∈ |3H|.
If the rank ρ of Pic(X) is 2, we will prove a limited result below in Section
5. We will show that there are Ulrich line bundles on X if and only if Pic(X)
has a basis consisting of H and C, where C.H = 6 and C 2 = 4. The Ulrich line
bundles are then OX (C) and OX (3H − C). In this case, we will also classify
the first Chern classes of stable Ulrich bundles of any rank r.
4.2.3. d ≥ 5. There are no results.
4.2.4. Del Pezzo Surfaces. The work of the present author, Rajesh Kulka-
rni, and Yusuf Mustopa about the existence of Ulrich bundles on del Pezzo
surfaces and its relation to the Minimal Resolution Conjecture has already
been mentioned in (iii), Section 2.
As mentioned in Remark 2.12, the work of Joan Pons-Llopis and Fabio
Tonini implies a classification of Ulrich line bundles on a del Pezzo surface
X ⊂ Pd of degree d. (See [40, Theorem 1.1].) For degrees up to 6, they also
construct, for any rank r ≥ 2, a family of simple ACM bundles of rank r
by considering consecutive extensions ACM line bundles of the form OX (Q)
where Q ⊂ X is a rational normal curve. Since these line bundles are Ulrich,
any vector bundle obtained by their extensions is also Ulrich. However, as the
authors point out, these vector bundles are not stable. (See [40, Proposition
5.7].)
Rosa M. Miró-Roig and Joan Pons-Llopis came up with similar results in
[38]. They prove that, for any del Pezzo surface X ⊂ Pd of degree d and for
any r ≥ 2, there exists an (r2 + 1)-dimensional family of simple Ulrich bundles

with c1 = rH.
For a given del Pezzo surface X ⊂ Pd , there exists no result classifying the
first Chern classes of stable Ulrich bundles completely.
4.3. Threefolds and higher dimensional varieties.
4.3.1. Cubic Hypersurfaces. The work of Casanellas and Hartshorne re-

lated to Ulrich bundles on a cubic surface X ⊂ P3 has been mentioned above.
They also constructed a rank 2 Ulrich bundle on any smooth cubic 3-fold in P4
and a rank 3 Ulrich bundle on the general cubic 3-fold in P4 . Recently, using
derived category techniques, Martı́ Lahoz, Emanuele Macrı̀ and Paolo Stellari
proved in [35, Theorem B] that there exist stable Ulrich bundles of any rank
r ≥ 2 on any smooth cubic 3-fold X ⊂ P4 .
4.3.2. Quintic Threefolds. In [10], Luca Chiantini and Carlo Madonna

proved the following result. Every stable rank 2 ACM bundle E on a gen-
eral quintic threefold X in P4 is infinitesimally rigid. In other words, E satisfies
H1 (E ∨ ⊗ E) = 0. This implies that the moduli space of Ulrich bundles of rank
2 on X is expected to be discrete.
Beauville proves in [4, Proposition 8.9] that there exists a rank 2 Ulrich
bundle on X. The number of such bundles must be finite, as is proved by a di-
mension counting argument. Let S5 denote the space of 10 × 10skew-symmetric
matrices of linear forms. The group GL(10) acts on S5 freely and properly; and
the Pfaffian map pf : S5 → |OP4 (5)| factors through S5 /GL(10). One can easily
calculate the dimensions of S5 /GL(10) and |OP4 (5)| to be the same. Then it
can be verified that (see [4, Appendix]) the map pf is generically surjective. It
follows that for a generic quintic threefold X ⊂ P4 , there must be finitely many
ways of expressing its defining equation as a Pfaffian. In other words, there are
finitely many rank 2 Ulrich bundles on X. However, as Beauville points out in
[4, 8.10], this number is not known.
4.3.3. Other Cases. We finish this section with a quick overview of recent
results.
• In [37], Rosa M. Miró-Roig proves that on nonsingular rational normal

scrolls (with exceptions) there exist families of simple Ulrich bundles of
arbitrarily high rank and dimension. Similarly, in [16], Laura Costa, Rosa
M. Miró-Roig and Joan Pons-Llopis prove that on Segre varieties (except
the quadric in P3 ) there exist families of simple Ulrich bundles of arbi-
trarily high rank and dimension.
98 Emre Coskun
• Write Gr(k, n) for the Grassmannian of k-dimensional linear subspaces

of Pn = P(V ) for an (n + 1)-dimensional k-vector space V . In [17], Laura
Costa and Rosa M. Miró-Roig classify the Ulrich bundles on the Grass-
mannian Gr(k, n) that are invariant under the action of GL(V ).
• In [32], Yeongrak Kim proves the existence of stable rank 2 Ulrich bun-
dles on rational surfaces with an anticanonical pencil, under additional
assumptions. To do this, he uses Lazarsfeld-Mukai bundles, which were
also used in [1].
• In a recent preprint, [18], Laura Costa and Rosa M. Miró-Roig classify

irreducible homogeneous Ulrich bundles on the flag manifold F(0, n−1, n).
• In two recent papers, [8] and [7] Gianfranco Casnati, Daniele Faenzi and
Francesco Malaspina study rank 2 ACM bundles on P1 ×P1 ×P1 and their
moduli spaces. Ulrich line bundles on P1 × P1 × P1 can be obtained by
[8, Lemma 2.4]; they are line bundles that correspond to permutations of
the triple (0, 1, 2), that is, the last six bundles in their list. Ulrich bundles
of rank 2 are have Chern classes that are listed in [8, Theorem A, (3)].
• In another preprint, [9], the same authors work on the classification of

rank 2 ACM bundles on P2 × P2 . From their work, it follows that there
are exactly two Ulrich line bundles on P2 × P2 ; namely the line bundles
corresponding to the pairs (0, 2) and (2, 0). The classification of the first
Chern classes of rank 2 Ulrich bundles on P2 ×P2 is given in [9, Proposition
4.2], [9, Theorem 5.7] and [9, Theorem 5.8].
• In [21], Daniele Faenzi and Francesco Malaspina prove that on a smooth

quartic scroll X ⊂ P5 , there exists a family of indecomposable Ulrich
bundles of rank 2r on X, parametrized by P1 . ([21, Theorem A]) They
also give a criterion to decide if a given ACM bundle is Ulrich on a smooth
surface scroll of degree at least 5. ([21, Theorem B])
5. Quartic Surfaces of Picard Number 2

In this section, we prove the results announced above in 4.2.2. Throughout
this section, X ⊂ P3 is a quartic - hence, K3 - surface. The Picard number
ρ = rank(Pic(X)) is assumed to be 2.
Proposition 5.1. (a) An Ulrich bundle E of rank r on X has degree 6r and

2
second Chern class c1 (E)
2 − 2r.
(b) Let E and F be Ulrich bundles on X of ranks r and s respectively, and with
first Chern classes C and D respectively. Then
χ(E ⊗ F ∨ ) = 6rs − C · D. (5.1)
Proof. (a) See [14, Proposition 2.10].
(b) This follows from [6, Proposition 2.12].
Remark 5.2. The slope of an Ulrich bundle is 6, and the reduced Hilbert poly-
nomial of an Ulrich bundle is 2(t + 2)(t + 1).
Remark 5.3. If OX (C) is an Ulrich line bundle on X, then C.H = 6 and C 2 = 4.
Lemma 5.4. Pic(X) admits a Z−basis containing the hyperplane class H.
Proof. Assume that A, B ∈ Pic(X) form a Z−basis for Pic(X), and let a, b ∈ Z
be such that H = aA + bB. We claim that a and b are relatively prime. Indeed,
suppose that they are both divisible by a prime p. Then for some D ∈ Pic(X),
we have 4 = H 2 = p2 D2 . This is impossible as D2 is even. Hence, the claim is
proved.
Let x and y be integers for which xa + yb = 1, and define H = yA − xB. It
is immediate that the transition matrix from {A, B} to {H, H } is unimodular,
and the result follows.
In light of Remark 5.3, we are interested in quartic K3 surfaces with Picard

lattice given by a basis {H, C}, where H 2 = C 2 = 4 and C.H = 6. The
following proposition implies that there exist quartics with this property.
Proposition 5.5. There exists a polarized K3 surface (X, H) of genus 3 and

a smooth, irreducible curve C on X satisfying C 2 = 4 and C.H = 6 such that
Pic(X) = ZH ⊕ ZC.
Proof. The result follows from [33, Theorem 1.1 (iv)] as the intersection num-
bers given above satisfy the required criteria.
Let us mention the following property of the lattice given above.
Proposition 5.6. In the Picard lattice given by the basis {H, C} with H 2 =
C 2 = 4 and C.H = 6, there do not exist divisor classes D with D2 = 0 or
D2 = ±2.
Proof. Write D = aH + bC. Then D2 = 4a2 + 12ab + 4b2 . Since this number is
a multiple of 4, it cannot be equal to ±2. By the quadratic formula, one sees
easily that there exist no integer solutions to 4a2 + 12ab + 4b2 = 0.
100 Emre Coskun
Assumption 1. The Picard lattice of X can be given as Pic(X) = ZH ⊕ ZC,

where H is the hyperplane class, C.H = 6 and C 2 = 4.
Proposition 5.7. Assume that Assumption 1 holds. Then one can take C to
be smooth and irreducible, and moreover one has h1 (OX (C)) = 0.
Proof. By a Riemann-Roch calculation, one finds h0 (OX (C)) = h1 (OX (C)) +

4 > 0, hence C can be assumed to be effective.
We claim that the complete linear system |C| has no fixed components.
Assume that D = aH + bC is such a fixed component. We assume, without
loss of generality, that D is integral. Then since C − D must be effective, we
have
(C − D).H = (−aH − (b − 1)C).H = −4a − 6b + 6 > 0
which is equivalent to 2a + 3b < 3. Since D is effective, we have
D.H = (aH + bC).H = 4a + 6b > 0.
Hence, we have that 2a + 3b = 1 or 2a + 3b = 2.

By [29, III, Ex. 5.3], pa (D) ≥ 0 and hence D2 = 4a2 + 12ab + 4b2 ≥ −2.
Assume first that 2a + 3b = 1. Substituting, we obtain 1 − 5b2 ≥ −2, which
implies b = 0 and hence a = 1/2, a contradiction.
Assume next that 2a + 3b = 2. Substituting, we obtain 4 − 5b2 ≥ −2, which
implies that b = 0 or b = 1. If b = 0, then D = H; which is a contradiction
since H cannot be a fixed component of any complete linear system. If b = 1,
then a = −1/2, a contradiction. Hence, the claim is proved.
The result now follows from [41, Proposition 2.6 (i)].
Proposition 5.8. Assume that Assumption 1 holds. The Ulrich line bundles
on X are OX (C) and OX (3H − C), and these line bundles are very ample.
Proof. By Proposition 5.7, we have that C is a smooth irreducible curve. Notice

also that C is nef. Since by Proposition 5.6 there are no divisors on X with
square 0 or ±2, by [33, Lemma 2.4(b)] it follows that OX (C) is very ample.
Note also that replacing C by 3H − C in the above arguments gives the same
results for OX (3H − C).
To prove that OX (C) is Ulrich, we need to prove that it has Hilbert poly-
nomial P (t) = 2(t + 2)(t + 1) and that it is ACM. The Hilbert polynomial is
easily computed. To prove that it is ACM, we use the following short exact
sequence:
0 → OX (C) → OX (C + tH) → ωtH (C) → 0,
where t ≥ 0 is an integer. Recall that, by Proposition 5.7 we have h1 (OX (C)) =
0. From the long exact sequence of the above short exact sequence, we obtain
h1 (OX (C + tH)) = h1 (ωtH (C)) = 0; where the second equality follows because
the degree of ωtH (C) is larger than 2g(tH) − 2.
Now recall that h1 (OX (3H − C)) = 0 as well. The argument in the above
paragraph gives h1 ((3 + t)H − C) = 0 for an integer t ≥ 0. Using Serre duality,
we have that h1 (C + tH) = 0 if t ≤ −3.
We claim that h1 (OX (C −H)) = 0. By Riemann-Roch, we have h0 (OX (C −
H)) = h1 (OX (C − H)) and hence it suffices to show that C − H is not effective.
If C − H were effective, since (C − H)2 = −4, there would be an irreducible
curve of negative self-intersection on X. This is impossible by Proposition 5.6.
Hence the claim is proved. Replacing C by 3H − C and applying the same
arguments gives h1 (OX (2H − C)) = h1 (OX (C − 2H)) = 0.
It remains to show that there are no other Ulrich line bundles on X. Con-
sider an Ulrich line bundle OX (D) where we write D = aH +bC. From H.D = 6
we obtain 4a + 6b = 6 and from D2 = 4 we obtain 4a2 + 12ab + 4b2 = 4. Since
a quadratic equation can have at most two solutions; and (a, b) = (0, 1) and
(a, b) = (3, −1), which correspond to OX (C) and OX (3H − C), are solutions,
it follows that these are the only solutions.
We now prove that the only linear determinantal quartics of Picard number
two are the ones considered in this section.
Proposition 5.9. If X is a quartic of Picard number two, and if there exists
an Ulrich line bundle on X, then Assumption 1 holds for X.
Proof. If L = OX (D) is an Ulrich line bundle, then the lattice M = H, D
is contained in Pic(X). By a well-known result in lattice theory, we have
[Pic(X) : M ]2 = d(M )/d(Pic(X)) where d(M ) denotes the discriminant of
M and similarly for Pic(X). We have d(M ) = −20, and hence d(Pic(X)) has
to divide −20 and the quotient must be a square. Therefore, the only cases are
d(Pic(X)) = −20 or d(Pic(X)) = −5. We prove the statement by showing that
the second case is not possible. Consider the matrix representation

4 a
Pic(X) =
a b
of Pic(X). We have 4b − a2 = −5. Note that b is even, so a2 ≡ 5 mod 8. But

this is impossible since 5 is not a square modulo 8.
From now on, we assume that Assumption 1 holds.
Proposition 5.8 guarantees the existence of Ulrich line bundles on X. Using
these, we can now attempt to classify all stable Ulrich bundles on X. We start
with the following lemma.
Lemma 5.10. Let E be a stable Ulrich bundle of rank r with c1 (E) = aH + bC.
Then we have 2a + 3b = 3r, and − 3r
5 ≤b≤ 5 .
3r
102 Emre Coskun
Proof. The first equality follows from the fact that, by Proposition 5.1, (iii), E
has degree 6r.
Since E and OX (C) are nonisomorphic stable Ulrich bundles with the
same reduced Hilbert polynomial, by [31, Proposition 1.2.7], we have h0 (E ∨ ⊗
OX (C)) = hom(E, OX (C)) = 0. Similarly, h2 (E ∨ ⊗ OX (C)) = 0. Hence,
χ(E ∨ ⊗ OX (C)) ≤ 0. By Proposition 5.1, (b), we have χ(E ∨ ⊗ OX (C)) =
−(aH + bC).C + 6r = −3r + 5b ≤ 0. Using the same argument with OX (C)
replaced by OX (3H − C) gives the inequality in the statement.
We now classify all stable Ulrich bundles on X by their first Chern class.
Theorem 5.11. Let r ≥ 2 be an integer. Denote Cr = {aH + bC ∈
Pic(X) | 2a + 3b = 3r, − 3r
5 ≤ b ≤ 5 }. Given D ∈ Pic(X), there exists a
3r
stable Ulrich bundle E of rank r on X with first Chern class c1 (E) = D if and
only if D ∈ Cr .
Proof. Lemma 5.10 gives one direction. For the other direction, we proceed by
induction on the rank. For r = 2, we have
χ(OX (C)∨ ⊗ OX (3H − C)) = −C.(3H − C) + 6 = −8
by Proposition 5.1, (iv). We have
h0 (OX (C)∨ ⊗ OX (3H − C)) = hom(OX (C), OX (3H − C)) = 0
since OX (C) and OX (3H −C) are nonisomorphic stable Ulrich bundles with the
same reduced Hilbert polynomial. Similarly, h2 (OX (C)∨ ⊗ OX (3H − C)) = 0.
Hence we obtain ext1 (OX (C), OX (3H − C)) = 8. Therefore, there exists a
nonsplit extension of OX (C) by OX (3H − C), which is a simple rank-2 Ulrich
bundle E with c1 (E) = 3H by [6, Lemma 4.2]. To prove that this bundle is
stable, we consider the modular dimension of the family of simple rank-2 Ulrich
bundles with first Chern class 3H. We claim that this dimension is 14. Let E
be one such bundle. By Proposition 5.1, (iv), we compute
χ(E ∨ ⊗ E) = −3H.3H + 24 = −12.
Since E is simple, we have
h0 (E ∨ ⊗ E) = hom(E, E) = 1.
Similarly, h2 (E ∨ ⊗ E) = 1. Therefore, h1 (E ∨ ⊗ E) = 14. The claim now follows

from [39, Theorem 0.1].
By Proposition 3.5, all Ulrich bundles are semistable; and by Remark 3.7
any strictly semistable Ulrich bundle of rank 2 must be an extension of Ulrich
line bundles. Now since ext1 (OX (C), OX (3H − C)) = 8, the extensions of
OX (C) by OX (3H −C) form a 7-dimensional family; similarly, the extensions of

OX (3H −C) by OX (C) form a 7-dimensional family. As the modular dimension
of the family of simple rank-2 Ulrich bundles with first Chern class 3H exceeds
the modular dimension of semistable Ulrich bundles with the same rank and
first Chern class, there exist stable rank-2 Ulrich bundles with first Chern class
3H.
Now assume that for some r ≥ 3, every class in Cr−1 is the first Chern class
of a stable Ulrich bundle of rank r − 1, and let aH + bC ∈ Cr be given. We
claim that at least one of aH + (b − 1)C or (a − 3)H + (b + 1)C is in Cr−1 . Since
2a + 3(b − 1) = 2(a − 3) + 3(b + 1) = 3(r − 1), it suffices to check that at least
one of b − 1 or b + 1 lies in the interval [− 35 (r − 1), 35 (r − 1)]. The inequalities
b − 1 ≤ 35 (r − 1) and b + 1 ≥ − 35 (r − 1) follow immediately from the fact that
− 35 r ≤ b ≤ 35 r. If b − 1 < − 35 (r − 1) and b + 1 > 35 (r − 1), it follows that
3r−8
5 < b < − 3r−8
5 , but this can only happen if r ≤ 2, which is contrary to the
assumption. This proves the claim.
Suppose aH + (b − 1)C ∈ Cr−1 . By our inductive hypothesis, there exists a
stable Ulrich bundle E of rank r − 1 with c1 (E) = aH + (b − 1)C. We have, by
Proposition 5.1, (iv),
χ(E ∨ ⊗ OX (C)) = −(aH + (b − 1)C).(C) + 6(r − 1)

= −3r + 5b − 2 < 0,
where we have used 2a + 3b = 3r to eliminate a. Since OX (C) and E are

nonisomorphic stable Ulrich bundles with the same reduced Hilbert polynomial,
h0 (E ∨ ⊗ OX (C)) = h2 (E ∨ ⊗ OX (C)) = 0. Hence ext1 (E, OX (C)) > 0, and there
exist nonsplit extensions of E by OX (C), and these are simple by [6, Lemma
4.2]. The case where (a − 3)H + (b + 1)C ∈ Cr−1 is handled similarly, by
considering χ(E ∨ ⊗ OX (3H − C)) and extensions of E by OX (3H − C).
We compute the modular dimension of simple, rank-r Ulrich bundles E with
first Chern class aH + bC by the expression h1 (E ∨ ⊗ E).
h1 (E ∨ ⊗ E) = 2 − χ(E ∨ ⊗ E)
= 2 + (aH + bC)2 − 6r2
= 2 + 3r2 − 5b2 ,
where we have used the equation 2a + 3b = 3r to eliminate a.

The proof will be complete once we prove that the modular dimension
of strictly semistable Ulrich bundles with fixed Jordan-Hölder type is strictly
smaller than the dimension given above. We make use of the observation that
it is enough to consider the extensions of a stable, rank-(r − 1) Ulrich bundle
F with first Chern class aH + (b − 1)C with the Ulrich line bundle OX (C); or
the extensions of a stable, rank-(r − 1) Ulrich bundle F with first Chern class
104 Emre Coskun
(a − 3)H + (b + 1)C with the Ulrich line bundle OX (3H − C). (We owe this
observation to the proof of [6, Theorem 4.3].) We will only consider the first
case, as the calculations for the second case are similar.
The extensions of F by OX (C) are parametrized by PExt(F, OX (C)). We
have
ext1 (F, OX (C)) = h1 (F ∨ ⊗ OX (C))

= −χ(F ∨ ⊗ OX (C))
= C.(aH + (b − 1)C) − 6(r − 1)
= 3r − 5b + 2,
where, as before, we have used the equation 2a + 3b = 3r to eliminate a.

Therefore, the modular dimension of the extensions of F by OX (C) is 3r−5b+1.
The Ulrich bundle F can be chosen from a family of dimension 2 + 3(r − 1)2 −
5(b − 1)2 . Adding the two dimensions 3r − 5b + 1 and 2 + 3(r − 1)2 − 5(b − 1)2 ,
we obtain that this number is smaller than 2 + 3r2 − 5b2 , which is the modular
dimension of simple, rank-r Ulrich bundles with the given first Chern class
aH + bC. Hence, there exist stable Ulrich bundles with rank r and first Chern
class aH + bC. This finishes the proof.
References
[1] Marian Aprodu, Gavril Farkas, and Angela Ortega, Minimal resolutions, Chow forms
and Ulrich bundles on K3 surfaces (2012), available at arXiv:1212.6248v4.
[2] M. F. Atiyah, Vector bundles over an elliptic curve, Proc. London Math. Soc. (3) 7
(1957), 414–452. MR0131423 (24 #A1274)
[3] Jörgen Backelin and Jürgen Herzog, On Ulrich-modules over hypersurface rings, Com-
mutative algebra (Berkeley, CA, 1987), Math. Sci. Res. Inst. Publ., vol. 15, Springer,
New York, 1989, pp. 63–68, DOI 10.1007/978-1-4612-3660-3 4, (to appear in print).
MR1015513 (90i:13006)
[4] Arnaud Beauville, Determinantal hypersurfaces, Michigan Math. J. 48 (2000), 39–64,
DOI 10.1307/mmj/1030132707. Dedicated to William Fulton on the occasion of his 60th
birthday. MR1786479 (2002b:14060)
[5] Joseph P. Brennan, Jürgen Herzog, and Bernd Ulrich, Maximally generated Cohen-
Macaulay modules, Math. Scand. 61 (1987), no. 2, 181–203. MR947472 (89j:13027)
[6] Marta Casanellas, Robin Hartshorne, Florian Geiss, and Frank-Olaf Schreyer, Sta-
ble Ulrich bundles, Internat. J. Math. 23 (2012), no. 8, 1250083, 50, DOI
10.1142/S0129167X12500838. MR2949221
[7] Gianfranco Casnati, Daniele Faenzi, and Francesco Malaspina, Moduli spaces of rank
two aCM bundles on the Segre product of three projective lines (2014), available at
arXiv:1404.1188v3.
[8] , Rank two aCM bundles on the del Pezzo threefold with Picard number 3, J.
Algebra 429 (2015), 413–446, DOI 10.1016/j.jalgebra.2015.02.008. MR3320630
[9] , Rank two aCM bundles on the del Pezzo fourfold of degree 6 and its general
hyperplane section (2015), available at arXiv:1503.02796v1.
[10] Luca Chiantini and Carlo Madonna, ACM bundles on a general quintic threefold,
Matematiche (Catania) 55 (2000), no. 2, 239–258 (2002). Dedicated to Silvio Greco
on the occasion of his 60th birthday (Catania, 2001). MR1984199 (2004f:14055)
[11] Emre Coskun, The fine moduli spaces of representations of Clifford algebras, Int.
Math. Res. Not. IMRN 15 (2011), 3524–3559, DOI 10.1093/imrn/rnq221. MR2822181
(2012j:14017)
[12] Emre Coskun, Rajesh S. Kulkarni, and Yusuf Mustopa, On representations of Clif-
ford algebras of ternary cubic forms, New trends in noncommutative algebra, Con-
temp. Math., vol. 562, Amer. Math. Soc., Providence, RI, 2012, pp. 91–99, DOI
10.1090/conm/562/11132, (to appear in print). MR2905555
[13] , Pfaffian quartic surfaces and representations of Clifford algebras, Doc. Math.
17 (2012), 1003–1028. MR3007683
[14] , The geometry of Ulrich bundles on del Pezzo surfaces, J. Algebra 375 (2013),
280–301, DOI 10.1016/j.jalgebra.2012.08.032. MR2998957
[15] Emre Coskun, Ulrich bundles on quartic surfaces with Picard number 1, C. R. Math.
Acad. Sci. Paris 351 (2013), no. 5-6, 221–224, DOI 10.1016/j.crma.2013.04.005 (English,
with English and French summaries). MR3089682
[16] Laura Costa, Rosa M. Miró-Roig, and Joan Pons-Llopis, The representation type of Segre
varieties, Adv. Math. 230 (2012), no. 4-6, 1995–2013, DOI 10.1016/j.aim.2012.03.034.
MR2927362
[17] L. Costa and R. M. Miró-Roig, GL(V )-invariant Ulrich bundles on Grassmannians,
Math. Ann. 361 (2015), no. 1-2, 443–457, DOI 10.1007/s00208-014-1076-9. MR3302625
[18] , Homogeneous Ulrich bundles on flag manifolds (2015), available at arXiv:
1506.03586v1.
[19] Igor V. Dolgachev, Luigi Cremona and cubic surfaces, Luigi Cremona (1830–1903) (Ital-
ian), Incontr. Studio, vol. 36, Istituto Lombardo di Scienze e Lettere, Milan, 2005, pp. 55–
70 (English, with Italian summary). MR2305952 (2008a:14002)
[20] Daniele Faenzi, Rank 2 arithmetically Cohen-Macaulay bundles on a nonsingular cu-
bic surface, J. Algebra 319 (2008), no. 1, 143–186, DOI 10.1016/j.jalgebra.2007.10.005.
MR2378065 (2009e:14065)
[21] Daniele Faenzi and Francesco Malaspina, Surfaces of minimal degree of tame and wild
representation type (2014), available at arXiv:1409.4892v2.
[22] Gavril Farkas, Mircea Mustaţǎ, and Mihnea Popa, Divisors on Mg,g+1 and the minimal
resolution conjecture for points on canonical curves, Ann. Sci. École Norm. Sup. (4) 36
(2003), no. 4, 553–581, DOI 10.1016/S0012-9593(03)00022-3 (English, with English and
French summaries). MR2013926 (2005b:14051)
[23] H. Grassmann, Die stereometrischen Gleichungen dritten Grades, und die
dadurch erzeugten Oberflächen, J. Reine Angew. Math. 49 (1855), 47–65, DOI
10.1515/crll.1855.49.47 (German). MR1578905
[24] Phillip Griffiths and Joe Harris, On the Noether-Lefschetz theorem and some re-
marks on codimension-two cycles, Math. Ann. 271 (1985), no. 1, 31–51, DOI
10.1007/BF01455794. MR779603 (87a:14030)
[25] A. Grothendieck, Sur la classification des fibrés holomorphes sur la sphère de Riemann,
Amer. J. Math. 79 (1957), 121–138 (French). MR0087176 (19,315b)
106 Emre Coskun
[26] Alexander Grothendieck, Cohomologie locale des faisceaux cohérents et théorèmes de

Lefschetz locaux et globaux (SGA 2), Documents Mathématiques (Paris) [Mathematical
Documents (Paris)], 4, Société Mathématique de France, Paris, 2005 (French). Séminaire
de Géométrie Algébrique du Bois Marie, 1962; Augmenté d’un exposé de Michèle Ray-
naud. [With an exposé by Michèle Raynaud]; With a preface and edited by Yves Laszlo;
Revised reprint of the 1968 French original. MR2171939 (2006f:14004)
[27] Darrell E. Haile, On the Clifford algebra of a binary cubic form, Amer. J. Math. 106
(1984), no. 6, 1269–1280, DOI 10.2307/2374394. MR765580 (86c:11028)
[28] Darrell Haile and Steven Tesser, On Azumaya algebras arising from Clifford algebras,
J. Algebra 116 (1988), no. 2, 372–384, DOI 10.1016/0021-8693(88)90224-4. MR953158
(89j:15044)
[29] Robin Hartshorne, Algebraic geometry, Springer-Verlag, New York-Heidelberg, 1977.
Graduate Texts in Mathematics, No. 52. MR0463157 (57 #3116)
[30] Jürgen Herzog and Michael Kühl, Maximal Cohen-Macaulay modules over Gorenstein
rings and Bourbaki-sequences, Commutative algebra and combinatorics (Kyoto, 1985),
Adv. Stud. Pure Math., vol. 11, North-Holland, Amsterdam, 1987, pp. 65–92. MR951197
(89h:13029)
[31] Daniel Huybrechts and Manfred Lehn, The geometry of moduli spaces of sheaves, 2nd
ed., Cambridge Mathematical Library, Cambridge University Press, Cambridge, 2010.
MR2665168 (2011e:14017)
[32] Yeongrak Kim, Ulrich bundles on rational surfaces with an anticanonical pencil (2014),
available at arXiv:1406.4359v2.
[33] Andreas Leopold Knutsen, Smooth curves on projective K3 surfaces, Math. Scand. 90
(2002), no. 2, 215–231. MR1895612 (2003c:14041)
[34] Rajesh S. Kulkarni, On the Clifford algebra of a binary form, Trans. Amer. Math.
Soc. 355 (2003), no. 8, 3181–3208, DOI 10.1090/S0002-9947-03-03293-8. MR1974681
(2004c:16025)
[35] Martı́ Lahoz, Emanuele Macrı̀, and Paolo Stellari, Arithmetically Cohen-Macaulay
bundles on cubic threefolds, Algebraic Geometry 2 (2015), no. 2, 231–269, DOI
10.14231/AG-2015-011.
[36] Anna Lorenzini, The minimal resolution conjecture, J. Algebra 156 (1993), no. 1, 5–35,
DOI 10.1006/jabr.1993.1060. MR1213782 (94g:13005)
[37] Rosa M. Miró-Roig, The representation type of rational normal scrolls, Rend. Circ. Mat.
Palermo (2) 62 (2013), no. 1, 153–164, DOI 10.1007/s12215-013-0113-y. MR3031575
[38] Rosa M. Miró-Roig and Joan Pons-Llopis, n-dimensional Fano varieties of wild
representation type, J. Pure Appl. Algebra 218 (2014), no. 10, 1867–1884, DOI
10.1016/j.jpaa.2014.02.011. MR3195414
[39] Shigeru Mukai, Symplectic structure of the moduli space of sheaves on an abelian or K3
surface, Invent. Math. 77 (1984), no. 1, 101–116, DOI 10.1007/BF01389137. MR751133
(85j:14016)
[40] Joan Pons-Llopis and Fabio Tonini, ACM bundles on del Pezzo surfaces, Matematiche
(Catania) 64 (2009), no. 2, 177–211. MR2800010 (2012f:14070)
[41] B. Saint-Donat, Projective models of K −3 surfaces, Amer. J. Math. 96 (1974), 602–639.
MR0364263 (51 #518)
[42] Bernd Ulrich, Gorenstein rings and modules with high numbers of generators, Math. Z.
188 (1984), no. 1, 23–32, DOI 10.1007/BF01163869. MR767359 (85m:13021)
[43] M. Van den Bergh, Linearisations of binary and ternary forms, J. Algebra 109 (1987),
no. 1, 172–183, DOI 10.1016/0021-8693(87)90171-2. MR898344 (88j:11020)
Noether-Lefschetz Locus and a
Special Case of the Variational
Hodge Conjecture: Using
Elementary Techniques
Ananyo Dan∗,†
Abstract
Fix integers n ≥ 1 and d such that nd > 2n + 2. The Noether-Lefschetz lo-
2n+1
2n hypersurfaces of degree d in P
cus NLd,n parametrizes smooth projective
satisfying the condition: H (X, C) H (X, Q) = Q. An irreducible compo-
n,n
nent of the Noether-Lefschetz locus is locally a Hodge locus.

One question is to
ask under what choice of a Hodge class γ ∈ H n,n (X, C) H 2n (X, Q) does the
variational Hodge conjecture hold true? In this article we use methods coming
from commutative algebra and Hodge theory to give an affirmative answer in
the case γ is the class of a complete intersection subscheme in X of codimen-
sion n. Another problem studied in this article is: In the case n = 1 when is an
irreducible component of the Noether-Lefschetz locus nonreduced? Using the
theory of infinitesimal variation of Hodge structures of hypersurfaces in P3 , we
determine all non-reduced components with codimension less than or equal to
3d for d 0. Here again our primary tool is commutative algebra.
Mathematics Subject Classification (2000). 14C30, 14D07
Keywords. Variational Hodge conjecture, Noether-Lefschetz locus, Hodge theory,

non-reduced scheme, Néron-Severi group, Hilbert scheme
Notation 0.1. Throughout this article, X will denote a smooth " hypersurface
in P2n+1 . Denote by H n,n (X, Q) the intersection H n,n (X, C) H 2n (X, Q) and
HX the very ample line bundle on X.
∗ The author has been supported by the DFG under Grant KL-2244/2 − 1
† Humboldt Universität zu Berlin, Institut für Mathematik, Unter den Linden 6, Berlin
10099. E-mail: [email protected]
Geometry, DOI 10.1007/978-981-10-5648-2_7
108 Ananyo Dan
1. Introduction
It was first stated by M. Noether and later proved by S. Lefschetz that for a
general smooth surface X in P3 , the rank of the Néron-Severi group, denoted
NS(X) is of rank 1. We can then define the Noether-Lefschetz locus, denoted
NLd,1 , to be the space of smooth degree d surfaces in P3 with Picard rank
greater than 1. Using Lefschetz (1, 1)-theorem, one can see that NLd,1 is the
space of smooth degree d surfaces X such that H 1,1 (X, Q) = Q. Similarly, we
can define higher Noether-Lefschetz locus as follows: Let n > 1 and d another
integer such that nd > 2n + 2. Denote by NLd,n the space of smooth degree d
hypersurfaces X in P2n+1 such that H n,n (X, Q) = Q. The orbit of the action of
the monodromy group on a rational class is finite (see [CDK95]). Consequently,
NLd,n is an uncountable union of algebraic varieties (see [Voi03, §3.3] for more
details).
Let L be an irreducible component of NLd,n . Then L is locally the Hodge
locus corresponding to a Hodge class. In particular, consider the space of all
smooth degree d hypersurfaces in P2n+1 , denoted Ud,n . For X ∈ L general,
there exists γ ∈ H n,n (X, Q) and "an open (analytic) simply connected set U
in Ud,n containing X such that L U is the Hodge locus corresponding to γ,
denoted NLd,n (γ) (see [Voi02, §5.3] for more details).
Before we state the first main result in this article, we fix some notations.
Notation 1.1. Given a Hilbert polynomial P, of a subscheme Z, in P2n+1 ,
denote by HP the corresponding Hilbert scheme. Denote by Qd the Hilbert
polynomial of a degree d hypersurface in P2n+1 . The flag Hilbert scheme HP,Qd
parametrizes all pairs (Z, X), where Z ∈ HP , X is a smooth degree d hyper-
surface in P2n+1 containing Z.
For any n ≥ 1 we prove the following theorem which is a special case of the
variational Hodge conjecture:
Theorem 1.2. Let Z be a complete intersection subscheme in P2n+1 of codi-
mension n + 1. Assume that there exists a smooth hypersurface in P2n+1 , say
X, containing Z, of degree d > deg(Z). For the cohomology class γ = a[Z] ∈
H n,n (X, Q), a ∈ Q, γ remains of type (n, n) if and only if γ remains an alge-
braic cycle. Moreover, NLd,n (γ) (closure taken in Ud,n under Zariski topology)
is isomorphic to an irreducible component of pr2 HP,Qd which parametrizes all
smooth degree d hypersurfaces in P2n+1 containing a complete intersection sub-
cheme with Hilbert polynomial P , where P (resp. Qd ) is the Hilbert polynomial
of Z (resp. X).
In [Otw03], Otwinowska proves this statement for d 0. In this article,
we use a reduction of infinitesimal variation of Hodge structures in the case
of hypersurfaces in P3 to objects in commutative algebra (see [Voi03, §6] for
Using Elementary Techniques 109
further details). In [Ste87] Steenbrink proves a similar statement, but the proof
relies heavily on Bloch’s semiregularity map (introduced in [Blo72]). In this
article we achieve the same result using only commutative algebra.
Finally, we prove the following result which classifies certain big components
of NLd,1 into reduced and non-reduced schemes:
Theorem!1.3. Let d ≥ 5 and γ is a divisor in a smooth degree d surface of

r
the form
!r i=1 ai [Ci ] with Ci distinct integral curves for all i = 1, ..., r and
d > i=1 ai deg(Ci ) + 4. Then the following are true:
(i) If r = 1 and deg(C1 ) < 4 then NLd,1 (γ) (closure taken under Zariski
topology on Ud,1 ) is reduced. In particular, NLd,1 (γ) is an irreducible
component of pr2 (HP,Qd ), the space parametrizing all smooth degree d
surfaces containing a reduced curve with the same Hilbert polynomial as
C1 , which we denote by P .
(ii) Suppose that r > 1. For d 0, every irreducible component L of NLd,1

of codimension at most 3d is locally of the form NLd,1 (γ) with γ as above,
"r
deg(Ci ) ≤ 3 and NLd,1 (γ)red = i=1 NLd,1 ([Ci ])red . Moreover, NLd,1 (γ)
is non-reduced if and only if there exists a pair (i, j), i = j such that
Ci .Cj = ∅.
2. Proof of Theorem 1.2

2.1. The strategy of proof is to first understand the tangent space to the Hodge
locus TX (NL(γ)), which we do in Lemma 2.12. Then, in 2.16 we use this to
show the dimension of NL(γ) coincides with that of TX (NL(γ)).
Notation 2.2. Denote by Snk the degree k-graded piece of H 0 (OP2n+1 (k)).
Define Sn := ⊕k≥0 Snk . Let X be a smooth degree d hypersurface in P2n+1 ,
defined by an equation F . Denote by JF , the Jacobian ideal of F generated
∂
as an Sn -module by the partial derivatives of F with respect to ∂X i
for i =
1, ..., 2n + 1, where Xi are the coordinates of P 2n+1
. Define, RF := Sn /JF . For
k ≥ 0, let JFk (resp. RFk ) symbolize the degree k-graded piece of JF (resp. RF ).
2.3. We now recall some standard facts about Hodge locus. Let X be a smooth
projective hypersurface in P2n+1 of degree d. Recall, there is a natural morphism
from H n,n (X) to H n,n (X)prim , where H n,n (X)prim denotes the primitive co-
homology on H n,n (X) (see [Voi02, §6.2, 6.3] for more on this topic). Denote by
γprim the image of γ under this morphism. Using the Lefschetz decomposition
theorem, one can see that NLd,n (γ) coincides with NLd,n (γprim ) i.e., γ remians
of type (n, n) if and only if so does γprim .
110 Ananyo Dan
2.4. Now, KP2n+1 = OP2n+1 (−2n − 2), H 0 (KP2n+1 (2n + 2)) = H 0 (OP2n+1 ) ∼
=C
generated by
dX0 dX̂i dX2n+1

Ω := X0 ...X2n+1 (−1)i ∧ ... ∧ ∧ ... ∧ ,
i
X0 Xi X2n+1
where the Xi are homogeneous coordinates on P2n+1 . Recall, for the closed
immersion j : X → P2n+1 , denote by H 2n (X, Q)van , the kernel of the Gysin
morphism j∗ from H 2n (X, Q) to H 2n (P2n+1 , Q). Now, [Voi03, Theorem 6.5]
tells us that there is a surjective map,
αn+1 : H 0 (P2n+1 , OP2n+1 ((n + 1)d − 2n − 2)) → F n+1 H 2n+1 (P2n+1 \X, C)
∼
= F n H 2n (X, C)van
which sends a polynomial P to the resisue of the meromorphism form

P Ω/F n+1 , where F is the defining equation of X (see [Voi03, §6.1] for more).
Finally, [Voi03, Theorem 6.10] implies that αn+1 induces an isomorphism be-
(n+1)d−(2n+2)
tween RF and H n,n (X)prim .
2.5. We now recall a theorem due to Macaulay which will be used throughout
this article.
Notations/Definition 2.6. A sequence of homogeneous polynomials Gi ∈

Sndi , i = 0, ..., 2n + 1 with di > 0 is said to be regular if the Gi have no common
zero. Denote by IG the ideal in Sn generated by the polynomials Pi for i =
i
0, ..., 2n + 1. Denote by HG the quotient Sn /IG and by HG the degree i graded
piece in HG .
!2n+1
Theorem 2.7 (Macaulay). For N := i=0 di − 2n − 2, the rank of HG N
=1
N −k
and for every integer k, the pairing, HG × HG k
→ HG is perfect.
N
See [Voi03, Theorem 6.19] for the proof of the statement. Recall,
(n+1)d−(2n+2)
Theorem 2.8 ([Voi03, Theorem 6.17]). Denote by P ∈ Sn such
that αn+1 (P̄ ) = γ (where αn+1 is an in 2.4). Then, TX NLd,n (γ) is isomorphic
(n+2)d−(2n+2)
to the preimage of ker(.P̄ : RFd → RF ) under the natural quotient
morphism from Snd → Snd /JFd .
2.9. It is easy to see that for any γ ∈ H n,n (X, Q), NLd,n (γ ) = NLd,n (a γ )
for any a ∈ Q, non-zero. For the rest of this section, we assume γ = [Z], where
Z is as in the statement of the theorem, a complete intersection subscheme of
codimension n + 1 in P2n+1 .
Notation 2.10. Denote by N := (n + 1)d − (2n + 2). Since X is smooth,

the corresponding Jacobian ideal JF can be generated by a regular sequence
of 2n + 2 polynomials Gi of degree d − 1. Using Theorem 2.7, we see that
there exists a perfect pairing RFk × RF2N −k → RF2N for all k ≤ 2N and RF2N
is one dimensional complex vector space. Let P be as in Theorem 2.8. Denote
by T0 , the subspace of RFN which is the kernel under the multiplication map,
.P : RFN → RF2N and by T0 the preimage of T0 in SnN under the natural
projection map from SnN to RFN . Define T1 the subspace of Sn , a graded Sn -
module such that for all t ≥ 0, the t-graded piece of T1 , denoted T1,t , is the
largest subvector space of Snt such that T1,t ⊗SnN −t is contained in T0 for t < N ,
T1,N = T0 and T1,N +t = T0 ⊗ Snt for t > 0.
Remark 2.11. It follows from the perfect pairing above that dim SnN /T1,N =
dim RFN /T0 = 1. Using the definition of T1 , it follows,
Snk /T1,k × SnN −k /T1,N −k → SnN /T1,N
is a perfect pairing. Hence, dim Snd /T1,d = dim SnN −d /T1,N −d .
Lemma 2.12. The tangent space TX (NLd,n (γ)) coincides with T1,d .
Proof. Note that H ∈ T1,d if and only if H̄ ⊗ RFN −d is contained in T0 which
by definition is equivalent to P̄ H̄ ⊗ RFN −d = 0 in RF2N . Using the perfect
pairing 2.10 we can conclude that P̄ H̄ = 0 in RFN +d . This is equivalent to
H ∈ TX (NLd,n (γ)).
Setup 2.13. Suppose that Z is defined by n + 1 polynomials P0 , ..., Pn . Since
Z ⊂ X and deg Z < d, we can assume that there exist polynomials Q0 , ..., Qn
of degree d − deg Pi , respectively such that X is defined by a polynomial of the
form P0 Q0 + ... + Pn Qn . Let I be the ideal in Sn generated by P0 , ..., Pn and
Q0 , ..., Qn .
Proposition 2.14. The k-graded pieces, T1,k = Ik for all k ≤ N .
Proof. Denote by#Z1 the subschemes in P2n+1 , defined by Q0 = P1 = ... =

Pn = 0. Since Z Z1 is the intersection of X and {P1 = ... = Pn = 0}, then
[Z] = −[Z1 ] mod QHX n
in the cohomology group H n,n (X, Q). So, [Z]prim =
−[Z1 ]prim . Denote by Z2 the subvariety defined by Q0 = ... = Qn = 0. Pro-
ceeding similarly, we get [Z]prim = a[Z2 ]prim for some integer a.
A consequence of deformation theory is that Id (Z) ⊂ NLd,n ([Z]prim ). Since
Id (Z) is a vector space, it is contained in TX NLd,n ([Z]prim ). Therefore, Id (Z)
as well as Id (Z2 ) are both contained in T1 i.e., (P0 , ..., Pn , Q0 , ..., Qn ) ⊂ T1 .
Since X is smooth the sequence {P0 , ..., Pn , Q0 , ..., Qn } is a regular sequence.
Using Theorem 2.7 we can conclude that dim SnN /IN = 1, where IN denotes
112 Ananyo Dan
the degree N graded piece of I and

S/I|k × S/I|N −k → S/I|N
is perfect pairing. So, I is Gorenstein of socle degree N contained in T1 which
is Gorenstein of the same socle degree. So, T1,k = Ik for all k ≤ N .
Remark 2.15. The parameter space, say H of complete intersection sub-
schemes in P2n+1 of codimension n + 1, defined by n + 1 polynomials of degree
deg(Pi ), respectively is irreducible. In particular, it is an open subscheme of
deg P0 deg Pn
P(Sn ) × ... × P(Sn )
which is irreducible. Denote by R0 the Hilbert polynomial of Z as a subscheme
in P2n+1 . Consider the flag Hilbert scheme HR0 ,Qd and the projection map
pr1 which is the projection onto the first component. Since the generic fiber of
pr1 is isomorphic to P(Id (Z)) for the generic subscheme Z on pr1 HR0 ,Qd , it is
irreducible, where Id (Z) is the degree d graded piece of the ideal, I(Z), of Z.
So, there exists an unique irreducible component in HR0 ,Qd such that the image
under pr1 of this component coincides with H. For simplicity of notation, we
denote by HR0 ,Qd this irreducible component, since we are interested only in
this scheme.
2.16 (Proof of Theorem 1.2). Using basic deformation theory and Hodge the-
ory, we can conclude that pr2 (HR0 ,Qd ) is contained in NLd,n (γ). So,
codim pr2 (HR0 ,Qd ) ≥ codim NLd,n (γ) ≥ codim TX NLd,n (γ).
Now, there is a natural morphism, denoted p from T1,d to HQd which maps
F1 to the zero locus of F1 . Since every element of T1,d defines a hypersurface
containing a subscheme with Hilbert polynomial R0 , pr2 (HR0 ,Qd ) contains Im p.
Since the zero locus of a polynomial is invariant under multiplication by a scalar,
dim T1,d = dim Im p + 1.
Finally,
codim pr2 (HR0 ,Qd ) = dim P(H 0 (OP2n+1 (d))) − dim pr2 (HR0 ,Qd ) ≤
≤ (h0 (OP2n+1 (d))−1)−dim Im p ≤ h0 (OP2n+1 (d))−dim T1,d = codim TX NLd,n (γ)
where the last equality follows from Lemma 2.12. This proves Theorem 1.2.
Remark 2.17. The proof of Theorem 1.2 as given in 2.16 above infact proves
a part of Theorem 1.3(ii). In particular, let C be a complete intersection curve
in P3 , X a smooth degree d surface in P3 containing C and γ the cohomology
class of C. Then 2.16 directly implies, NLd,1 (γ) is reduced and parametrizes all
smooth degree d surfaces in P3 containing a complete intersection curve with
the Hilbert polynomial same as C. This is a part of Theorem 1.3(ii).
3. Proof of Theorem 1.3

3.1. If C1 is a complete intersection curve then reducedness of NLd,1 ([C1 ])
follows from 2.17.
If C1 is an integral curve, deg(C1 ) < 4 and C1 not complete intersection
then C1 is a twisted cubic. We then have,
Lemma 3.2. Let C1 be a twisted cubic, X a smooth degree d surface (d ≥ 5)
containing C1 and γ the cohomology class of C1 . Then, codim TX NLd,1 (γ) =
3d − 11.
Proof. Recall, the twisted cubic C1 is generated by 3 polynomials of degree 2
each. Suppose P is a polynomial in T1 such that P is not contained in I(C1 ).
Since the zero locus of the ideal, say I generated by I(C1 ) and P is non-empty,
JF (which is base point free) is not contained in I . Since the Jacobian ideal,
JF ⊂ T1 , we can show that there is a regular sequence in T1 consisting of 4
elements, two of which are the generators of I(C1 ), the third one is P and the
forth is an element in JFd−1 . Since, codim T1,2d−4 = 1, Theorem 2.7 implies
2 + 2 + deg(P ) + d − 1 − 4 ≥ 2d − 4. So, deg(P ) ≥ d − 3. By Proposition 2.14,
codim TX (NLd,1 (γ)) = codim T1,d = codim T1,d−4 =
= codim Id−4 (C1 ) = 3.(d − 4) + 1 = 3d − 11.

This proves the lemma.
3.3 (Proof of Theorem 1.3(i)). The only step remaining to prove is that for γ as
in Lemma 3.2, NLd,1 (γ) is reduced. Using basic deformation theory and Hodge
theory we can conclude that there exists an unique irreducible component H
of HP,Qd whose generic element is (C, X), where C is a twisted cubic contained
in X such that pr2 (H) is contained in NLd,1 (γ). It is easy to compute that
codim(pr2 (H)) = 3d − 11. So,
3d − 11 ≥ codim NLd,1 (γ) ≥ codim TX (NLd,1 (γ)) = 3d − 11.
So, NLd,1 (γ) is reduced and parametrizes smooth degree d surfaces containing
a twisted cubic. This finishes the proof of (i).
3.4. We now recall a result that will help us make the characterization of the
irreducible components of NLd,1 as in Theorem 1.3(ii).
Proposition 3.5 ([Dan14, Proposition 5.6]). Let d 0, γ be a Hodge class in a
smooth degree d surface in P3 such that codim NLd,1 (γ) ≤ 3d. Then there exists
!t
integral curves C1 , ..., Ct of degree at most 3 such that γ = i=1 ai [Ci ] + bHX
"t
for some integers ai , b and NL(γ)red is the same as i=1 NL([Ci ])red .
114 Ananyo Dan
Notations/Remark 3.6. We now come to the proof of the final part of the
theorem.
!t Suppose now that γ is as in the "above proposition i.e., of the form
t
i=1 ai [Ci ] + bHX such that NL(γ)red = i=1 NL([Ci ])red . Denote by P̄i the
element in RF2d−4 !r such that α!2 ( P̄ i ) = [C i ] prim for i = 1, ..., t.!Since α2 is a
r
linear map, α2 ( i=1 ai P̄i ) = i a i [C i ] prim . Denote by P̄ := i=1 ai P̄i . So,
[Ci ]
α2 (P̄ ) = γ. Denote by T1,d−4 the corresponding T1,d−4 in Notation 2.10 ob-
tained by replacing P (in the notation 2.10) by P̄i for i = 1, ..., r. Note that
codim TX NLd,1 (γ) = codim T1,d = codim T1,d−4 , where the last equality is due
"r [Ci ] !
to Remark 2.11. Note that, i=1 T1,d−4 ⊂ T1,d−4 because P̄ = a P̄ , so
"r #r i i i
i=1 ker P̄i ⊂ ker P̄ . Therefore, codim TX NLd,1 (γ) ≤ codim Id−4 ( i=1 Ci ).
Before we go to the last step of the proof we recall the following computa-
tion:
Lemma 3.7 ([Dan14, Lemma 3.6]). Let d ≥ 5!and C be an effective divisor

on a smooth degree d surface X of the form i ai Ci , where Ci are integral
curves with deg(C) + 4 ≤ d. Then, dim |C| = 0, where |C| is the linear system
associated to C.
3.8 (Proof of Theorem 1.3). Using Proposition 2.14 and Lemma 3.2, we have
[Ci ] "r [Ci ] #r
T1,d−4 = Id−4 (Ci ) for i = 1, ..., t. So, i=1 T1,d−4 = Id−4 ( i=1 Ci ) is contained
in TX NLd,1 (γ). Denote by Pi the Hilbert polynomial of Ci for i = 1, ..., t. Since
deg(Ci ) < 4, Theorem 1.3(i) implies that there exists an irreducible compo-
nent of HPi ,Qd such that its image under the natural projection morphism pr2
(onto the second component) is isomorphic to NLd,1 ([Ci ])red . So, there exists
an irreducible component, say Hγ of HP1 ,Qd ×HQd ... ×HQd HPt ,Qd such that
pr2 (Hγ )red = ∩ti=1 NLd,1 ([Ci ])red = NL(γ)red , where pr2 is the natural mor-
phism from Hγ to HQd and the second equality follows from Proposition 3.5.
Denote by Lγ := pr(Hγ ), where pr is the natural projection morphism from
Hγ to HP1 × ... × HPt . A generic t-tuple of curves in HP1 × ... × HPt does
!t
not intersect each other. Hence dim Lγ < i=1 dim HPi if and only " if for a
general t-tuple (C1 , ..., Ct ) in Lγ there exists i, j, i = j such that Ci Cj = ∅.
Lemma 3.7 implies that dim |Ci | = 0 for i = 1, ..., t. It is then easy to see
#t
that codim NLd,1 (γ) = codim ∩ti=1 NLd,1 ([Ci ]) = codim Id ( i=1 Ci ) − dim Lγ
#t † #t
(see [Dan14, Proposition 4.6]). If codim Id−4 ( i=1 Ci ) ≤ codim Id ( i=1 Ci ) −
!t
i=1 dim HPi then

$
t
† $
t
t
codim TX NLd,1 (γ) ≤ codim Id−4 Ci ≤ codim Id Ci − dim HPi
i=1 i=1 i=1

‡ $
t
≤ codim Id Ci − dim Lγ = codim NLd (γ),
i=1
where the first inequality follows from 3.6 and the inequality ‡ is strict if and
only i, j, i = j such that
" if for a general t-tuple (C1 , ..., Ct ) in Lγ there exists #
Ci Cj = ∅. Since d 0, using the Hilbert polynomial of Ci , the inequality
!t !t
† is equivalent to i=1 dim HPi ≤ 4 i=1 deg(Ci ). Since deg(Ci ) < 4 and Ci
is integral, it is easy to compute that dim HPi is infact equal to 4 deg(Ci ) for
i = 1, ..., t. This proves (ii). Hence, completes the proof of the theorem.
Acknowledgement
I would like to thank Prof. R. Kloosterman for reading the preliminary version
of this article and several helpful discussions.
References
[Blo72] S. Bloch. Semi-regularity and de-Rham cohomology. Inventiones Math.,
17:51–66, 1972.
[CDK95] E. Cattani, P. Deligne, and A. Kaplan. On the locus of hodge classes.
Journal of the American Mathematical Society, 8(2):483–506, 1995.
[Dan14] A. Dan. On a conjecture by griffiths and harris concerning certain
noether–lefschetz loci. Communications in Contemporary Mathematics,
page 1550002, 2014.
[Otw03] A. Otwinowska. Composantes de petite codimension du lieu de Noether-
Lefschez: un argument asymptotique en faveur de la conjecture de Hodge.
J. Alg. Geom., 12(2):307–320, 2003.
[Ste87] J. H. M. Steenbrink. Some remarks about the Hodge conjecture, Hodge
theory (Sant Cugat, 1985). Lecture Notes in Maths. -1246. Springer, Berlin,
1987.
[Voi02] C. Voisin. Hodge Theory and Complex Algebraic Geometry-I. Cambridge
studies in advanced mathematics-76. Cambridge University press, 2002.
[Voi03] C. Voisin. Hodge Theory and Complex Algebraic Geometry-II. Cambridge
studies in advanced mathematics-77. Cambridge University press, 2003.
Tangent Bundle of P2 and Morphism
from P2 to Gr(2, C4)
A. El Mazouni∗ and D.S. Nagaraj†
Abstract
In this note we study the image of P2 in Gr(2, C4 ) given by tangent bundle of
P2 . We show that there is component H of the Hibert scheme of surfaces in
Gr(2, C4 ) with no point of it corresponds to a smooth surface.
Mathematics Subject Classification (1991). 14F17
Keywords. Projective plane; Tangent bundle; Morphisms; Grassmannian.
1. Introduction
Let P2 denote the projective plane over the field of complex numbers C and
Gr(2, C4 ) Grassman variety of two dimensional quotients of the vector space
C4 .
The aim of this paper is to study the image of P2 by non constant morphisms
P → Gr(2, C4 ) obtained by tangent bundle TP2 of P2 . The bundle TP2 is
2
generated by sections and hence it is generated by four(= rank(TP2 ) + dim(P2 ))

independent global sections. Any set S of four independent generating sections
of TP2 defines a morphism
φS : P2 → Gr(2, C4 ),
such that the φ∗S (Q) = TP2 , where Q is the universal rank two quotient bundle
on Gr(2, C4 ) :
OGr(2,C
4
4 ) → Q → 0.
∗ Laboratoire de Mathématiques de Lens EA 2462 Faculté des Sciences Jean Perrin Rue
Jean Souvraz, SP18 F-62307 LENS Cedex France. E-mail: [email protected]

† Institute of Mathematical Sciences (HBNI) C.I.T. campus, Taramani, chennai 600113,
India. E-mail: [email protected]
Geometry, DOI 10.1007/978-981-10-5648-2_8
118 A. El Mazouni and D.S. Nagaraj
According to a result of Tango [4], if φ : P2 → Gr(2, C4 ) is an imbedding

then the pair of Chern classes (c1 (φ∗ (Q)), c2 (φ∗ (Q))) is equal to one of the
following pairs: (H, 0), (H, H 2 ), (2H, H 2 ) or (2H, 3H 2 ), where H is the ample
generator of H2 (P2 , Z).
Since c1 (TP2 ) = 3H and c2 (TP2 ) = 3H 2 , the morphism φS defined by a set
S of four independent generating sections of TP2 is not an imbedding. Thus,
it is natural to ask, does there exists a set S of four independent generating
sections of TP2 for which φS is generically injective? In this direction we have
the following (Theorem 3.6):
Theorem 1.1. For general choice of an ordered set S of four independent

generating sections of TP2 the morphism
φS : P2 → Gr(2, C4 )
is generically injective.
We also, show that in fact one can find an ordered set of generators S of
TP2 the morphism is an immersion i.e., the morphism induces an injection on
all the tangent spaces.
As by product of our result we obtain the following (Theorem 4.2):
Theorem 1.2. There is an irreducible component H of the Hilbert scheme of

surfaces in Gr(2, C4 ) no point which corresponds to a smooth surface.
2. The Tangent Bundle of P2

The tangent bundle of P2 fits in an exact sequence called the “Euler sequence”:
0 → OP2 → OP2 (1)3 → TPn → 0. (1)
This exact sequence together with the fact H1 (OPn ) = 0, implies that
dimH0 (TP2 ) = 8, where Hi denotes the i th sheaf cohomology group. Since
the rank two bundle TP2 on P2 is ample and generated by sections, a minimal
generating set of independent sections has cardinality four. Any set S of four
independent generators of TP2 gives to an exact sequence:
0 → ES → OP42 → TP2 → 0.
This in turn corresponds to a morphism φS : P2 → Gr(2, C4 ), where φS (x) =

{C4 = OP42 |x → TP2 |x → 0}.
Tangent Bundle of P2 and Morphism from P2 to Gr(2, C4 ) 119
3. The Main Result

Note that in the “Euler sequence” (1) the injective map
0 → OP2 → OP2 (1)3
is given by the section v = (X, Y, Z), where X, Y, Z is the standard basis of

H0 (OPn (1)). For a section vi of OPn (1)3 we denote wi the image section of TP2
under the surjection
OP2 (1)3 → TPn → 0
in (1). Let S̃ = (v1 , v2 , v3 , v4 ) be an ordered set of four linearly independent
sections of OP2 (1)3 and S = (w1 , w2 , w3 , w4 ) be the corresponding ordered set
of sections of TP2 . Clearly,the set S is generating set of independent sections of
TP2 if and only if S̃ ∪ {v} is a generating set of independent sections of OP2 (1)3 .
Lemma 3.1. Let v1 = (X, 0, 0), v2 = (0, Y, 0), v3 = (Y, Z, X), v4 = (Z, X, Y )
be four sections of OP2 (1)3 and S̃ = (vi |1 ≤ i ≤ 4). Then the ordered set S̃
with {v = (X, Y, Z)} is a generating set of independent sections of OP2 (1)3 .
Hence the corresponding ordered set of sections S = (w1 , w2 , w3 , w4 ) generate
TP2 , where wi is the image of vi under the map given in exact sequence (1).
Proof: Clearly, v generate a subspace of OP2 (1)3 dimension one at every point
of P2 . Hence wi , wj is not independent at a point p ∈ P2 if and only if the
section vij = vi ∧ vj ∧ v of OP2 (3) vanishes at p. Thus, if the six independent
sections {vi ∧ vj ∧ v|1 ≤ i < j ≤ 4} has no common zero implies S̃ with
{v = (X, Y, Z)} is a generating set of independent sections of OP2 (1)3 . Note
that v12 = XY Z, v13 = X(Z 2 − XY ), v14 = X(XZ − Y 2 ), v23 = Y (X 2 −
Y Z), v24 = Y (Y X − Z 2 ), v34 = 3XY Z − (X 3 + Y 3 + Z 3 ). It is easy to see
that the set {v12 , v13 , v14 , v23 , v24 , v34 } of sections of OP2 (3) has no common
zero in P2 and hence the ordered set S̃ with {v = (X, Y, Z)} is a generating
set of independent sections of OP2 (1)3 . Hence the corresponding ordered set of
sections S = (w1 , w2 , w3 , w4 ) generate TP2 . 2
Lemma 3.2. Let f : P2 → Pn be a non constant morphism and f ∗ (OPn (1)) =
OP2 (m). Assume that there exists a linear subspace W of codimension two such
that W ∩f (P2 ) consists of exactly m2 points. Then the morphism f is generically
injective.
Proof: Note as the morphism f is non constant f ∗ (OPn (1)) = OP2 (m) with
m > 0 and hence is ample. This means f is finite map. Set r = deg(f ), the
number of elements f −1 (f (x)) for a general x ∈ P2 . If d to be the degree
of f (P2 ) in Pn then it is easy to see that m2 = d.r. On the other hand the
assumption, W ∩ f (P2 ) consists of exactly m2 points, implies d ≥ m2 . Thus we
must have d = m2 and r = deg(f ) = 1. Thus f is generically injective. 2
Lemma 3.3. With the notations of Lemma(3.1), the surjection of vector bun-
dles on P2
OP42 → TP2
given by S defines a generically injective morphism
φS : P2 → Gr(2, C4 ).
Proof: Let p : Gr(2, C4 ) → P5 be the Pluker imbedding given by the determi-

nant of the universal quotient bundle. Then p ◦ φS is given by
(x; y; z) →
(xyz; x(z 2 − xy); x(xz − y 2 )); y(x2 − yz); y(xy − z 2 );
3xyz − (x3 + y 3 + z 3 )).
To prove the map φS is generically injective it is enough to prove the map p◦φS
is so. Set (Z0 , . . . , Z5 ) as the homogeneous coordinates of P5 and W be the
codimension two subspace of P5 defined by Z0 = 0 = Z5 . Then W ∩ p ◦ φS (P2 )
is equal to
{(0, −ω i , 1, 0, 0, 0); (0, 0, 0, −ω i , 1, 0); (0, ω i , 1, ω i , 1, 0)|1 ≤ i ≤ 3},
where ω is a primitive cube root of unity. Note that
(p ◦ φS )∗ (OP5 (1)) = OP2 (3).
Hence, the required result follows from Lemma(3.2). 2

Remark 3.4. We show (see Lemma 4.1) that p ◦ φS is an immersion. i.e., the
induced linear map on the tangent space at every point of P2 is injective and
one to one except finitely many points.
Next we recall the follwing [See, Lemma(3.13)[1]]:
Lemma 3.5. Let X and Y be two irreducible projective varieties. Let T be an
irreducible quasi-projective variety and t0 ∈ T be a point. Let
F :X ×T →Y
be a morphism. Assume that Ft := F |X×t : X → Y is finite for all t ∈ T and

Ft0 is a birational onto its image. Then there is an open subvariety U of T such
that t0 ∈ U and for t ∈ U the morphism Ft is birational onto its image.
Proof: For the sake of completeness we reproduce the proof here. Consider
the morphism G = F × IdT : X × T → Y × T. Then the assumption Ft is
finite implies the morphism G is finite and proper. Hence G = G∗ (OX×T ) is
coherent sheaf of OY ×T modules. Let Z ⊂ Y ×T be the subvariety on which the

sheaf G∗ (OX×S ) is supported. Then clearly the map p : Z → T, restriction of
the natural projection, is surjective. The section 1 ∈ OX×T gives an inclusion
of OZ in G. Let F = G/OZ . Let Z1 ⊂ Y × T be the subvariety on which
the sheaf F supported. Let q : Z1 → T be the natural projection and let
U = {t ∈ T |dimq −1 (t) < dim(X)} then we see that by semi continuity [See,
page 95, Exercise (3.22) [3]], U is an open subset and is non-empty as t0 ∈ U.
For t ∈ U the morphism Ft is an isomorphism on X × t − G−1 (q −1 (t). Since G
is finite G−1 (q −1 (t) is proper closed subset of X × t and hence the morphism
Ft is birational onto its image. This proves the Lemma. 2
Theorem 3.6. For a generic choice of an ordered set S of four independent
generating sections of TP2 the morphism
φS : P2 → Gr(2, C4 )
is generically injective.
Proof: It is easy to see that the ordered set of four sections S generating TP2 is an
irreducible quasi projective variety. In fact it is an open subvariety of the affine
space V 4 , where V = H0 (TP2 ). The theorem at once follows from Lemma(3.5),
if we show the existence of one S for which φS is generically injective. But the
existence of one such S follows from Lemma(3.3). 2
4. An Example
The result of the previous section can be used give an example of a component
of a Hilbert Scheme of Gr(2, C4 ) with out any point corresponding to a smooth
surface.
Lemma 4.1. The morphism p ◦ φS : P2 → P5 of Lemma(3.2) is an immersion
i.e., the induced linear map on the tangent space at every point of P2 is injective.
Moreover, p ◦ φS one to one except
S1 = {(1; 0; 0), (0; 1; 0), (0; 0; 1)} → (0; 0; 0; 0; 0; 1)
and
S2 = {(1; 1; 1), (ω; ω 2 ; 1), (ω 2 ; ω; 1)} → (1; 0; 0; 0; 0; 0),
where ω is a primitive cube root of unity.
Proof: Let X, Y, Z be the homogeneous coordinates functions on P2 and
Z0 , Z1 , Z2 , Z3 , Z4 , Z5 be the homogeneous coordinates functions on P5 . Clearly
under the morphism p ◦ φS : P2 → P5 the set S1 maps to (0; 0; 0; 0; 0; 1) and
the set S2 maps to (1; 0; 0; 0; 0; 0). Note that the lines X = 0, Y = 0, and
Z = 0 mapped to nodal cubics Z0 = Z1 = Z2 = Z33 + Z43 − Z3 Z4 Z5 = 0,
Z0 = Z3 = Z4 = Z13 + Z23 + Z1 Z2 Z5 = 0, and Z0 = Z1 − Z4 = Z2 + Z3 =
Z13 + Z23 − Z1 Z2 Z5 = 0 respectively. Thus we can conclude that the morphism
p ◦ φS is an immersion on these three lines. On the complement of these lines
the morphism p ◦ φS can be described as
(x, y) → (1/y − x, x/y − y, x − y/x, y − 1/x, 3 − x2 /y − y 2 /x − 1/xy)
from C2 − {xy = 0} → C4 . If (x, y) and (x1 , y1 ) maps to the same point then
we get the following equations:
1/y − x = 1/y1 − x1 (2)
x/y − y = x1 /y1 − y1 (3)
y/x − x = y1 /x1 − x1 (4)
1/x − y = 1/x1 − y1 . (5)

The equations 2 and 5 gives us
xy − 1 x1 y1 − 1 xy − 1 x 1 y1 − 1
= ; = .
y y1 x x1
Since xy = 0 and x1 y1 = 0 we see that either xy − 1 = 0 and (x, y) = (x1 , y1 )

or xy − 1 = 0 = x1 y1 − 1. Hence we get (p ◦ φS ) is one to one out side the
set {(1, 1), (ω, ω 2 ), (ω 2 , ω)} and this is mapped to (0, 0, 0, 0, 0). The assertion
about the immersion of the given morphism
C2 − {xy = 0} → C4
can be checked by looking at the two by two minors of the below jacobian
matrix of the morphism:

−1 1/y 1 + y/x2 1/x2 −2x/y + y 2 /x2 + 1/x2 y
2
−1/y 2 −x/y 2 − 1 −1/x 1 x2 /y 2 − 2y/x + 1/xy 2
Theorem 4.2. Let H be the irreducible component of the Hilbert scheme of

Gr(2, C4 ) containing the point corresponding to the image surface of the mor-
phism
φS : P2 → Gr(2, C4 )
of 3.3. Then no point of H corresponds to a smooth surface.
Proof: Let p : Gr(2, C4 ) → P5 be the Pluker imbedding. Since (p ◦

φS )∗ (OP5 (1)) = OP2 (3) and by Lemma 4.1 the morphism p◦φS is an imbedding
outside finite set of points. Moreover, general hyperplane section of (p ◦ φS )(P2 )
in P5 is smooth curve of genus one. If a point of the irreducible H of corresponds
to a smooth surface Y then it has to have the same cohomology class as that
of (p ◦ φS )(P2 ) namely (3, 6) ∈ H4 (Gr(2, C4 ), Z). Also, the general hyperplane
section of p(Y ) has to be a smooth curve of genus one. But according to the
classification of smooth surfaces of type (3, 6) in Gr(2, C4 ) (see, [2, Theorem
4.2]) we see that there are no smooth surface of type (3, 6) with hyperplane
section a smooth curve of genus one. This contradiction proves that no point
of H corresponds to a smooth surface. 2
Remark 4.3. The component H of the the Hilbert Scheme in Theorem 4.2 is
reduced irreducible of dimension 23. In fact computing the normal sheaf asso-
ciated to the morphism φS of Lemma 3.3 and counting the dimension of space
of all such morphisms we see that H is a reduced irreducible of dimension 23.
Acknowledgment
This work was supported in part by the Labex CEMPI (ANR-11-LABX-0007-
01). The second author thanks university D’artois, Lens and the University of
Lille. We thank Laytimi Fatima for help during the work.
References
[1] El Mazouni, A.; Laytimi, F.; Nagaraj, D. S. Morphisms from P2 to Gr(2, C4 ).
J. Ramanujan Math. Soc. 26 (2011), no. 3, 321332.
[2] Gross, Mark.; Surfaces of bidegree (3, n) in Gr(1, P3 ). Math. Z. 212 (1993), no.
1, 73106.
[3] R. Hartshorne: Algebraic Geometry. Graduate Texts in Mathematics, No. 52,
Springer-Verlag, New York, 1977.
[4] Hiroshi Tango: On (n-1)-dimensional projective spaces contained in the Grass-
mann variety Gr(n, 1). J. Math. Kyoto Univ. 14-3 (1974) 415-460.
Twisting by a Torsor
Michel Emsalem∗
Abstract
Twisting by a G-torsor an object endowed with an action of a group G is a
classical tool. For instance one finds in the paragraph 5.3 of the book [17] the
description of the ”opération de torsion” in a particular context. The aim of this
note is to give a formalization of this twisting operation as general as possible
in the algebraic geometric framework and to present a few applications. We will
focus in particular to the application to the problem of specialization of covers
addressed by P. Dèbes et al. in a series of papers.
Mathematics Subject Classification (2000). Primary: 14G99 14L20 14L30. Sec-

ondary: 11G99.
Keywords. torsors, group-schemes.
1. Introduction
Twisting by a G-torsor an object endowed with an action of G is a classical tool.
For instance one finds in the paragraph 5.3 of the book [17] the description of the
”opération de torsion” in a particular context. We refer to [3] for a formulation
in a topological framework. The aim of this note is to give a formalization of this
twisting operation as general as possible in the algebraic geometric framework
and to present a few applications. We will focus in particular to the application
to the problem of specialization of covers addressed by P. Dèbes and al. in a
series of papers, which was one of the motivations for writing these notes.
We begin with a section on torsors where we recall classical properties and
operations on torsors. It is the opportunity to set the notations. In section 3
we define the twisting operation by a torsor using the cocycle description of a
torsor. We state the main properties of this twisting operation in paragraph 4
∗ Michel Emsalem a reçu le soutien du Labex CEMPI (ANR-11-LABX-01)
Geometry, DOI 10.1007/978-981-10-5648-2_9
126 Michel Emsalem
and recall some classical examples. Paragraph 5 is devoted to the study of the
particular case where the twisted objects are themselves torsors. This will lead
to the situation we study extensively in the next paragraphs where we consider
(ramified) Galois covers that we twist with what we call arithmetic torsors. The
aim is to provide a test to know wether there are specializations of the given
Galois cover which are isomorphic as torsors to some given arithmetic torsor.
2. Torsors
2.1. Left Torsors and Bitorsors. The aim of this section is to set
notations in the description of torsors by cocycles. Let S be a scheme endowed
with some subcanonical Grothendieck topology, G → S a group scheme, X → S
a S-scheme and P a S-scheme endowed with a left action of G and a G-
equivariant S-morphism P → X.
Definition 2.1. P → X is a left G-torsor if there exists a covering (Ui )i∈I of
X trivializing the torsor P ; in other words there are sections si : Ui → P|Ui =
P ×X Ui for any i and these sections si induce trivializations θi : Ui ×S G →
P|Ui defined by θi (x, g) = gsi (x). When P → X is a G-torsor, the natural
morphism G ×S P → P ×X P given by the formula (g, p) → (g.p, p) (for any
g ∈ G(T ), p ∈ P (T ) where T is a S-scheme) is an isomorphism ([18], Prop.
4.43).
For i, j ∈ I, on the intersection Uij = Ui ×X Uj , one gets si |Uij = gij sj |Uij
for some gij ∈ G(Uij ) and the commutative diagram
θi |U
P|Uij o
ij
Uij ×S G
dJJ
JJ
JJ
JJ
θj |U
ij J
Uij ×S G
where the vertical map is defined by (x, g) → (x, ggij ) which is an isomorphism
of trivial torsors. The torsor P is obtained by gluing these trivial left torsors
over the Uij ’s.
The gij clearly satisfy the cocycle condition on Uijk :
gij gjk = gik .
Let us describe the group H = AutG (P ) of automorphisms of P as left G-

torsor : locally on each Ui via θi , sections of H over Ui are right multiplication
by some element gi ∈ G(Ui ). If the gi are the restriction to Ui of a global section
of H, they will make the following diagrams commutative :
Twisting by a Torsor 127
g / ggi .

ggij / ggij gj = ggi gij
One gets the relations
−1
gi = gij gj gij .
The X-group H is the inner form of G determined by the left G-torsor P ;
H acts on the right on P by the rule p.h = h−1 (p) (p ∈ P, h ∈ H), and the
actions of G and H commute. So P can be viewed as a (G, H)-bitorsor.
The link between left action of G and right action of H can be described
by isomorphisms ui : H|Ui → G|Ui defined in the following way
∀h ∈ H|Ui si h = ui (h)si .
One easily checks that
−1
∀h ∈ H|Uij ui (h) = gij uj (h)gij .
Remark 2.2. In order to avoid problems of representability it is convenient to
consider a more general situation and extend the notion of torsors to S-spaces
and not only S-schemes (cf. [13] for the definition of S-spaces). A scheme S
is still given with some subcanonical Grothendieck topology, and G → S is a
S-group space, i.e. a sheaf of groups over S. The definition 2.1 still holds in this
context where X is a S-scheme, P is a S-space endowed with a left action of G
and P → X a G-equivariant morphism. We will use this more general concept
of torsor in section 3, where, for an object ξ of some stack, we will consider
torsors under the sheaf of groups Aut(ξ) which is not necessarily representable.
2.2. Right Torsors. The same kind of description holds for a right H-
torsor P . One defines cocycles with values in H by si |Uij = sj |Uij hij , the hij ’s
satisfying the cocycle condition
hik = hjk hij .
The group G is the group of automorphisms of the right H-torsor, a global
section of G being given by local sections hi ∈ H(Ui ) satisfying
hi = h−1
ij hj hij .
One has isomorphisms vi : G|Ui → H|Ui satisfying gsi = si vi (g) and over
Uij ,
vi = h−1
ij vj hij .
If (gij , ui ) is the description of P as a left G-torsor, its description as right
H-torsor is given by (hij , u−1 −1
j ) where hij = uj (gij ).
128 Michel Emsalem
2.3. Contracted Product. Let P be a (G, H)-bitorsor on X, Y → X

a H-equivariant morphism, where Y is endowed with an left action of H. One
defines the contracted product P ∧H Y as the quotient of P ×X Y moded by
the equivalence relation (yh, z) = (y, hz), for h ∈ H. We will be interested in
the following two particular cases.
1. Y = H is a S-group space endowed with a morphism of S-group spaces

ϕ : H → H inducing the left action of H on H that we are considering.
In this case P ∧H H is naturally a right torsor under H . If P is described
as right torsor under H by the cocycle (hij ) with respect to the covering
(Ui )i∈I , P ∧H H is described by the cocycle (ϕ(hij )) as right H -torsor.
2. P be a (G, H)-bitorsor on X and Y = Q is a (H, K)-bitorsor on X

described by (gij , ui ) and (hij , vi ).
Lemma 2.3. In terms of cocycles the contracted product P ∧H Q is de-

scribed by (gij uj (hij ), ui vi ).
Proof. Let (Ui )i∈I be a trivializing covering for the two torsors with sec-
tions si : Ui → P and ti : Ui → Q; one denotes by (si , ti ) : Ui → P ∧H Q
the corresponding section. On Uij , one gets the equality
(si , ti ) = (gij sj , hij tj ) = (gij sj hij , tj ) = (gij uj (hij )sj , tj )

= gij uj (hij )(sj , tj ).
The second formula is obvious.
2.4. Inverse of a Torsor. Let P be a (G, H)-bitorsor. It is endowed

with a right action of G and a left action of H defined by the formulas
y g = g −1 y and h y = yh−1 .
Through these actions P is a (H, G)-bitorseur that one denotes by P 0 .
With the preceding notation, local sections si : Ui → P , and left cocycle gij
with values in G (resp. right cocycle hij with values in H ) describing P as a
left G-torsor (resp. as a right H-torsor) defined by the formulas
−1
si = gij sj = sj gij and si = sj hij = h−1
ij sj .
Let us denote as before ui : H|Ui → G|Ui the isomorphism given by si h =

ui (h)si . The description of P 0 in terms of cocycles is given by (gij
0
, u0i ), where
0 −1 −1 −1 0 −1
gij = hij = uj (gij ) and ui (g) = ui (g).
Lemma 2.3 shows that P ∧H P 0 G is the trivial (G, G)-bitorsor and

P ∧G P H is the trivial (H, H)-bitorsor: P 0 is an inverse of P for the
0
contracted product.
It is easy to check the following property.
Lemma 2.4. With the notation of Lemma 2.3, (P ∧H Q)0 Q0 ∧H P 0 .
3. Twisting by a Torsor
Suppose we are given a stack C → Schemes/S with Schemes/S endowed with
f pqc-topology, an affine S-group-scheme G locally of finite type and faithfully
flat. Consider the category C G (X) with objects ξ of C(X) endowed with a
morphism of sheaves ϕ : GX → Aut(ξ); a morphism from (ξ, ϕ) to (ξ , ϕ )
is a morphism f : ξ → ξ compatible with ϕ, ϕ , which means that for any
morphism of S-schemes λ : T → X and for any g ∈ G(T ) the following diagram
is commutative:
λ f
λ ξ / λ ξ
ϕ(g) ϕ (g)

λ ξ
λ f
/ λ ξ
Theorem 3.1. 1. To any right torsor P → X under GX is associated a

functor
ΦP : C G (X) → C(X)
and for any object ξ of C G (X) an isomorphism of Aut(ξ)-torsors
IsomC(X) (ξ, ΦP ξ) → P ∧GX Aut(ξ)
where P ∧GX Aut(ξ) is the contracted product of P with Aut(ξ) over GX

(see Remark 2.2).
2. The torsor P determines an inner form H of GX making P a (H, G)-

bitorsor and ΦP induces an equivalence of categories CG (X) CH (X).
H
3. If P is a (H, G)-bitorsor and Q a (K, H)-bitorsor, ΦQ ◦ ΦP ΦQ∧ P
.
4. Conversely if ξ and ξ are locally isomorphic objects of C(X), then P =

Isom(ξ, ξ ) is a right torsor under the X-group Aut(ξ) and the twisted
object ΦP (ξ) of ξ by P is canonically isomorphic to ξ .
130 Michel Emsalem
For an object ξ of CG (X), its image ΦP (ξ) will be referred to as the result
of the twisting operation by the torsor P .
Proof. 1. Let ui : Ui → X, i ∈ I be a trivializing covering of X for

the right torsor P and the corresponding cocycle gij ∈ G(Uij ) and
its image ϕ(gij ) = ḡij ∈ Aut(ξ|Uij ). The ḡij ’s define descent data for
the family of objects ui ξ and these descent data are effective. Thus
there exists an unique object ΦP (ξ) over X endowed with isomorphisms
θi : ui ξ → ui ΦP (ξ) making all diagrams commutative:
θi |Uij
uij ξ / ΦP (ξ)|U .
: ij
t tt
tt
tt
ḡij
tt θj |Uij
uij ξ
One checks that the object ΦP (ξ) does not depend on the trivializing
covering neither on the chosen representative gij .
Let f : (ξ1 , ϕ1 ) → (ξ2 , ϕ2 ) be a morphism between two objects of C G (X).
For all i, j, the following diagrams are commutative:
f|Uij
u∗i ξ1 |Uij / u∗i ξ2 |U
ij
ϕ1 (gij ) ϕ2 (gij )
f|Uij
u∗j ξ1 |U / u∗j ξ2 |U
ij ij
which means that the ΦP (f ) = (f|Ui )i define a morphism ΦP (ξ1 ) →

ΦP (ξ2 ). This defines a functor ΦP .
Let h̄ij be another cocycle with values in Aut(ξ) defining another object
Φ (ξ). A morphism λ : Φ (ξ) → ΦP (ξ) is the data of morphisms λi ∈
Hom(ui ξ, ui ξ) making the following diagrams commutative:
λi |Uij
uij ξ / uij ξ
h̄ij ḡij
λj |U
uij ξ
ij
/ uij ξ
In particular if hij is the trivial cocycle, Φ (ξ) = ξ and the preceding

diagrams resume to
λj |Uij ◦ λi −1
|Uij = ḡij
which means that the family (λi ) is a section of the torsor P ∧GX Aut(ξ),
which corresponds to the image ḡij of the cocycle gij by the morphism ϕ.
This shows a one to one correspondence between section of P ∧GX Aut(ξ)
on X and sections on X of Isom(ξ, ΦP ξ). The same holds over any T →
X, which proves the isomorphism of sheaves IsomC(X) (ξ, ΦP ξ) → P ∧GX
Aut(ξ).
2. The inner form H of GX is obtained by gluing u∗i G with u∗j G over Uij by
conjugation by gij . The fact that H acts on Φ(ξ) where ξ is an object of
CG (X) is more or less tautologic: suppose that we are given a section (gi )i
of H, which means sections gi ∈ G(Ui ) satisfying far all i, j
(∗) gj |Uij ◦ gij = gij ◦ gi |Uij
then we have the following commutative diagrams :

gi |Uij
u∗i ξ |Uij / u∗i ξ |U
ij
ϕ(gij ) ϕ(gij )

u∗j ξ |U / u∗j ξ |U
ij gj |U ij
ij
which associates to the section of H an automorphism of Φ(ξ).

To check that for a morphism f : ξ1 → ξ2 in CG (X) its image Φ(f )
commutes with the action of H, one can check it locally on the Ui where
it is obvious.
The fact that Φ is an equivalence is a consequence of point (3) which
0
implies that ΦP is a quasi-inverse of ΦP .
3. Point (3) is a consequence of the commutative diagram
ξ|Uij
θi
/ ΦP (ξ)|U ωi
/ ΦQ (ΦP (ξ))|U
u : ij
o7
ij
θj uu ωj ooo
uu h̄ij o
ḡij
uu ooo
uu ooo
ξ|Uij ΦP (ξ)|Uij
θj uu
u:
uu
uj (hij )
uu
uu
ξ|Uij
and of Lemma 2.3 which gives the description by cocycles of the contracted
product.
132 Michel Emsalem
4. In the other direction the first assertion is obvious.

Let Ui be a covering of X with isomorphisms λi : ui ξ → ui ξ . The cocycle
defining the torsor P = Isom(ξ, ξ ) is ḡij = λj −1
|Uij ◦ λi |Uij . The following
diagrams are commutative
λi |Uij
uij ξ / uij ξ
ḡij Id
λj |U
uij ξ
ij
/ uij ξ
which proves that ξ is obtained from ξ by the descent data ḡij , In other
words ξ = ΦP (ξ).
Remark 3.2. From the proof one clearly gets that if P is the trivial torsor in
point (2) H G and the equivalence of category ΦP is the identity of CG (X).
Remark 3.3. Given a right G-torsor π : P → X, π ∗ is a functor from CG (X) to
the category of G-equivariants objects of C(P ) (see [18], 3.8, for the definition
of G-equivariants objects). On the other hand, as explained in [18], Theorem
4.46, the G-torsor π : P → X defines an equivalence of categories FP between
G-equivariant objects of C(P ) and C(X). The functor ΦP of Theorem 3.1 is
ΦP = FP ◦ π ∗ .
4. Properties of Twisting by a Torsor

4.1. Moding by a Normal Subgroup.
Proposition 4.1. Suppose we are given a normal S-sub-group K of G. Then

the quotient P/K is naturally a G/K-torsor. Denote by Θ : CG/K (X) → CG (X)
the obvious functor. For any G-torsor P over X and any object ξ of CG/K (X),
ΦP/K (ξ) ΦP (Θ(ξ)).
Proof. The statement is a consequence of this simple remark: with the notation
of the proof of Theorem 3.1, if gij denotes the cocycle associated to the torsor
P and the covering Ui , the cocycle associated to the G/K-torsor P/K is θ(gij )
where θ : G → G/K is the canonical morphism.
4.2. Image by a Morphism of Stacks. Let us consider a functor

of S-stacks u : C1 → C2 over the category of S-schemes. Given an object ξ of
C1,G (X) one still denotes by u the morphism u : AutX (ξ) → AutX (u(ξ)). If ϕ :
GX → AutX (ξ) is the structural morphism of ξ, ψ = u ◦ ϕ : GX → AutX (u(ξ))
is the structural morphism of u(ξ) which is an object of C2,G (X). Let P → X
be a right G-torsor over X. One easily checks the following statement.
Proposition 4.2.
u(ΦP (ξ)) ΦP (u(ξ))
isomorphic in C2,H (X) where H = AutG (P ).
4.3. Base Change. Using notation of section 1, consider a morphism

f : X → X of S-schemes. Let P → X be a right G-torsor, the pull back
f P → X is a right G-torsor over X .
Proposition 4.3. For any object ξ of CG (X),

Φf P
(f (ξ)) f (ΦP (ξ))
Proof. If we are given effective descent data over X, i.e. objects ξi on the
opens Ui with isomorphisms αij : ξi |Uij → ξj |Uij satisfying descent conditions
and defining an object ξ on X, one can pull them by f : X → X getting
descent data for the objects f ξi on the opens f −1 Ui with isomorphisms f αij :
f ξi |f −1 Uij → f ξj |f −1 Uij which define the object f ∗ ξ. In particular the right
G-torsor P is defined by gluing together trivial right G-torsors GUi on the
opens Ui with descent data given by the cocycle gij ∈ G(Uij ) (which defines an
automorphism of the trivial right G-torsor GUij by left multiplication by gij ).
Pulling these data by f one gets the left multiplication by gij ◦ f over f −1 (Uij ).
Thus the torsor f ∗ P is defined by the cocycle (gij ◦ f ).
Fix a right G-torsor P . Let ξ be an object of CG (X); we will denote g the
image of an element g ∈ G by the structural morphism GX → Aut(ξ). The
twisted object ΦP (ξ) is defined by descent data provided by the gij ’s acting
on the ξUij ’s. On the other hand f (ΦP (ξ)) is defined by descent data f ∗ (ḡij )
on the f (ξUi ) = f (ξ)f −1 (Ui ) . The fact that GX → Aut(ξ) is a morphism of
sheaves implies that f ∗ (ḡij ) = gij ◦ f . This proves that f (ΦP (ξ)) is defined by
the local objects f (ξ)f −1 (Ui ) with descent data gij ◦ f . This is the definition

of Φf P (f (ξ)).
4.4. Classical Examples.

1. One knows that the isomorphism classes of vector bundles of rank n on a
scheme X are parametrized by H1 (X, GLn ). The pointed set H1 (X, GLn )
parametrizes also the GLn -torsors over X. The correspondence between
134 Michel Emsalem
vector bundles and GLn -torsors can be described in terms of twisting by

a torsor. In one direction to a vector bundle F of rank n corresponds the
⊕n
GLn -torsor P = Isom(OX , F). In the other direction to the GLn -torsor
⊕n
P corresponds the twisted vector bundle ΦP (OX ).
2. Let G → Spec(k) be an affine group scheme over a field k and consider

the category Repk (G) of representations of G on finite dimensional k-
vector spaces. It is a neutral tannakian category with the forgetful func-
tor i. For any k-scheme θ : X → Spec(k) one may consider the functor
θ∗ ◦ i : Repk (G) → Coh(X) with values in the category Coh(X) of co-
herent sheaves on X. There is a one to one correspondence between fiber
functors F : Repk (G) → Coh(X) and G-torsors : given a fiber functor
F : Repk (G) → Coh(X) one associates Isom⊗ (θ ◦ i, F ) which is a right
torsor under the automorphism group of the forgetful functor which is
G; in the other direction to a right G-torsor P → X one associates the
twisted by P of the forgetful functor ΦP (θ ◦ i). This correspondence in
the case of finite group schemes G underlies the tannakian construction
by Nori of the fundamental group scheme [16].
3. When X = Spec(k) one recovers the equivalence of categories between

the gerbe of fiber functors of the category Repk (G) and the gerbe BGk
of G-torsors over a k-scheme. More generally a gerbe G → Spec(k) over
a field k bound by a k-group scheme G is neutral if and only if it is
equivalent to the category BGk . Indeed if G is neutral and ξ0 is a section
ξ0 : Spec(k) → G, to any section ξ : X → G over a k-scheme X is
associated the G-torsor P = IsomX (ξ0 , ξ). In the other direction to any
G-torsor P → X corresponds the twisted section ΦP (ξ0 ).
4. In the paragraph 5.3 of ”Cohomologie Galoisienne” ([17]) J.-P. Serre de-

scribes the twisting operation in the category of H-sets, where H is a profi-
nite group. The groups G and the right G-torsors are respectively groups
and torsors in the category of H-sets. This leads to an interpretation of
the cohomology group H1 (H, G) in terms of G-torsors in the category of
H-sets, and in the situation of a subgroup G < G1 to an interpretation of
the fibers of the natural map of pointed sets H1 (H, G) → H1 (H, G1 ).
5. More generally coming back to the general situation of Theorem 3.1, con-
sider a morphism of S-group schemes u : G → G1 . It leads to a map
ũ : H1 (X, G) → H1 (X, G1 ) sending a G-torsor P to the contracted prod-
uct P ∧G G1 through u. On the other hand, given a right G-torsor P , u
induces a morphism of group schemes u : G = AutG P → G1 = AutG1 P1
where P1 = P ∧G G1 , and thus a morphism ũ : H1 (X, G ) → H1 (X, G1 ).
The following diagram is commutative:
H1 (X, G)
ũ / H1 (X, G1 )
O O

−∧G P −∧G1 P1
H1 (X, G ) / H1 (X, G )
ũ 1
where the vertical maps are bijections. Thus ũ−1 (P1 ) = ũ−1 (ũ(P )) is in
one to one correspondence by the left vertical map with the kernel of ũ ,

i.e. the set of right G -torsors Q such that Q ∧G G1 is the trivial G1 -
torsor. In other words the right G-torsors Q and P have same images by
ũ if and only if the image of Q ∧G P 0 by ũ is the trivial torsor.
5. Twisting a G-torsor
Consider the situation of Theorem 3.1 with C being the category of right G
torsors and let R be a right H-torsor over X. The category CH (X) contains
the category of (H, G)-bitorsors over X and if P is an element of CH (X), one
may consider the twisted object ΦR (P ) which is a right G-torsor over X.
Corollary 5.1. In the situation of Theorem 3.1, for any (H, G)-bitorsor P ,
ΦR (P ) R ∧H P
as right G-torsors and

IsomG (P, ΦR (P )) R
as right H-torsors.
Proof. This is a consequence of points (3) and (1) of Theorem 3.1.
Corollary 5.2. Let P and Q be right G-torsors.
IsomG (P, Q) ΦQ (P 0 )
as right AutG (P )-torsors.
Proof. Apply Corollary 5.1 to Q = ΦR (P ) = R ∧H P , where H = AutG (P ),

which is equivalent to R Q ∧G P 0 ΦQ (P 0 ).
Definition 5.3. The object ΦQ (P 0 ) will be referred to as P twisted by Q.

136 Michel Emsalem
Remark 5.4. In view of Lemma 2.4 considering P twisted by Q resumes to

considering Q twisted by P (they are inverse of each other).
Corollary 5.5. Let Gd be the trivial G-torsor. For any right G-torsor P ,
P 0 IsomG (P, Gd ).
Proof. This is an immediate consequence of Corollary 5.2 with Q = Gd .
Consider again a (K, G)-bitorsor Q and a (L, G)-bitorsor P . The aim of

what follows is to give a description is terms of cocycles and descent data of
the object ΦQ (P 0 ) = Q ∧G P 0 . The right G-torsor Q is defined by (gij , vi )
where gij ∈ GUij and vi : KUi → GUi ; the right G-torsor P is defined by

(gij , vi ) where gij

∈ GUij and vi : LUi → GUi . So the left G-torsor P 0 is
−1
defined by (gij , vi ) and as left K-torsor, Q is described by (vj−1 (gij ), vi−1 ).
The formula of the contracted product gives a description of the left K-torsor
−1
Q ∧G P 0 : (vj−1 (gij )vj−1 (gij ), vi−1 vi ). We state the result :
Proposition 5.6. Let Q be a (K, G)-bitorsor and P a (L, G)-bitorsor described
as right G-torsors by the cocycles (gij , vi ), where gij ∈ GUij and vi : KUi →

GUi , and (gij , vi ) where gij

∈ GUij and vi : LUi → GUi . The left K-torsor
ΦQ (P 0 ) is described by
−1
(vj−1 (gij gij ), vi−1 vi ).
In terms of descent data P 0 is obtained by gluing the trivial left G-torsors

−1
GUi over Uij by right multiplication by gij : precisely one has isomorphisms
of left G-torsors ϕi : P|Ui → GUi making the following diagrams commutative
0
over Uij :
0
P|U
ϕi
/ GUij
ij
EE
EE
E −1
ϕj EE
(gij )d
E"
GUij
where (g)d denotes the right multiplication by g. On the other hand by

definition of the twisting of a G-object by Q one has isomorphisms ψi :
ΦQ (P 0 )|Ui → P|U
0
i
which get into the commutative triangles on the Uij ’s :
ΦQ (P 0 )|Uij
ψi
/ P|U
0
JJJ ij
JJJ
J gij
ψj JJJ
$
0
P|U ij
0
where the vertical map is the action of gij on P|U ij
.
Finally the following diagrams are commutative over Uij :
ΦQ (P 0 )|Uij
ψi
/ P|U0 / GUij
ϕi
.
JJJ ij
DD
JJJ DD
J gij D −1
ψj JJJ ϕj DD
(gij )d
$ D !
0
P|U GUij
ij
EE
EE
E
ϕj EE
(gij )g
E"
GUij
We have proved the following statement :
Proposition 5.7. Let Q be a (K, G)-bitorsor and P a (L, G)-bitorsor described

as right G-torsors by the cocycles (gij , vi ), where gij ∈ GUij and vi : KUi →

GUi , and (gij , vi ) where gij

∈ GUij and vi : LUi → GUi . The left K-torsor
ΦQ (P 0 ) is obtained by gluing the GUi ’s over the Uij ’s by the morphisms
−1 −1
(gij )g ◦ (gij )d = (gij )d ◦ (gij )g
Remark 5.8. As P 0 is locally trivial, ΦQ (P 0 ) = Q ∧G P 0 is locally over X

isomorphic to Q.
Here is a consequence of Corollary 5.2 :
Corollary 5.9. Let u : U → X be a morphism of S-schemes. The U -scheme

u (ΦQ (P 0 )) has a section over U if and only if the torsors u P and u Q are
isomorphic over U .
Proof. This follows immediately from Corollary 5.2 and from the base change
formulas.
In the same way one gets the following.
Corollary 5.10. Let u : U → X be a morphism of S-schemes and Q → U a

right G-torsor. The U -scheme ΦQ (u∗ (P 0 )) has a section over U if and only if
the right torsors u∗ P and Q are isomorphic over U .
The next property deals with the quotient of G by a normal subgroup.

138 Michel Emsalem
Proposition 5.11. Let K → G be a normal subgroup of G; suppose we are

given a right G-torsor P1 → X, a right G-torsor P2 → X, a right G/K-torsor
R → X inserted in the commutative diagram with equivariant maps
P10 A P2 .
00AAA K }}

00 AAA K }
}
00 A ~}}}
G 0
0 R G
00
00 G/K
0
X
Suppose we are given a section s : X → IsomG/K,X (R, R) ΦR 0
G/K,X (R )
(which is a closed immersion). Denoting s(X) ∈ IsomG/K,X (R, R) the corre-
sponding element the following diagram is cartesian:
s(X)∗ P2 / s∗ ΦP2 (P10 ) / ΦP2 (P10 )

ΦK,R (P10 ) G,X G,X

R /X / ΦR 0
G/K,X (R )
s
This holds in particular for the section s corresponding to s(X) = IdR .
Proof. The diagram can be reformulated in
IsomK,R (P1 , s(X)∗ P2 ) / s∗ IsomG,X (P1 , P2 ) / IsomG,X (P1 , P2 )

R /X / IsomG/K,X (R, R)
s
The right square being cartesian one has to check that the large rectangle is
cartesian. The question is local and we may suppose that the torsors are trivial.
But IsomK,R (P1 , s(X)∗ P2 ) Isoms(X),R (P1 , P2 ) where the last term refers to
the set of X-isomorphisms from P1 to P2 inducing s(X) on R.
Given two right G-torsors P and Q over X Corollary 5.9 gives a tool allowing
to test for u : U → X whether u∗ P u∗ Q. In the following construction one
answers the question to know, given two morphisms u, v : U → X, whether
u∗ P v ∗ Q.
Consider the right G-torsors P1 = P ×S X → X ×S X and Q1 = X ×S Q →
X ×S X and the twisted object Q1 ∧G P10 .
Proposition 5.12. Let (u, v) : U → X ×S X; the G-torsors u∗ P and v ∗ Q are

isomorphic if and only if the fibre at (u, v) of Q1 ∧G P10 has a section over U .
Moreover if v : S → X is a section of the structural morphism X → S,
(1X × v)∗ (Q1 ∧G P10 ) (v ∗ Q ×S X) ∧G P 0 (P twisted by the fibre at v of Q).
A similar statement holds with (v × 1X ).
Proof. This comes from the cartesian diagrams
u∗ P / P ×S X and v∗ Q / X ×S Q

U
u×v
/ X ×S X U
u×v
/ X ×S X
and from commutation of base change with twisting operation. The proof of
second part of the statement uses the same tools.
In the particular case where P = Q one recovers the self-twisted cover
introduced in [5].
Corollary 5.13. With the above notation u∗ P and v ∗ P are isomorphic if and
only if the fibre at (u, v) of the self-twisted torsor P1 ∧G P10 has a section over U .
6. An Example of Computation: The Case of a

Galois Extension of Fields
Consider a Galois extension L of k of group G; then Y = Spec(L) → Spec(k)
is a right G-torsor. The left torsor Y 0 is Y 0 = Spec(L) where L is endowed
with the right action of G given by α σ = σ −1 .α. Then ΦY (Y 0 ) = Y ∧G Y 0 =
Spec(L ⊗k L)G where this denotes the fixed elements of L ⊗k L under the
diagonal action of G.
Let us explicit this diagonal action. The étale k-algebra L⊗k L is isomorphic
to L(G) on the following way:
α ⊗k β → (ασ(β))σ∈G .
Let τ ∈ G; the image of τ (α) ⊗k τ (β) under this isomorphism is
(τ (α)σ(τ (β))σ∈G = (τ (α(τ −1 στ )(β))))σ∈G . One deduces that the image of an
element (xσ )σ∈G by the diagonal action of τ ∈ G is (yσ )σ∈G where
yσ = τ (xτ −1 στ ).
The fixed elements under the action of G are the tuplets (xσ )σ∈G satisfying
∀σ ∀τ xσ = τ (xτ −1 στ ).
140 Michel Emsalem
Consider the partition G = ∪1≤i≤d Ci in conjugation classes and for every

i, a representative σi of Ci and Zi = Z(σi ) the centralizer of σi . From the
above condition one deduces that xσi ∈ LZi (where LZi denotes the field of
fixed elements under Zi ). For any other element σ ∈ Ci of the form σ = θσi θ−1
(θ is defined up to right multiplication by an element of Zi ) we have xσ =
θ(xθ−1 σθ ) = θ(xσi ) ∈ LZ(σ) = θ(LZi ). %
From that one deduces the inclusion (L ⊗k L)G ⊂ 1≤i≤d LZi and, consid-
ering the dimensions over k,

(L ⊗k L)G LZi .
1≤i≤d
So sections over k of IsomG (L, L) are in one to one correspondence with el-
ements of Z(G). If for instance G is abelian, all sections of IsomG (L, L) are
defined over k and IsomG (L, L)(k) = G.
Finally ΦY (Y 0 ) = Y ∧G Y 0 = Spec(L⊗k L)G is a right torsor under the inner
form H of G defined by the torsor Y . Suppose that this torsor Y is defined by
a morphism Φ : Gal(k̄/k) → G, H can be described by the action of Gal(k̄/k)
on its k̄-points i.e. the elements of the abstract group G by the formula
∀σ ∈ Gal(k̄/k) ∀g ∈ G σ g = Φ(σ)gΦ(σ)−1 .
7. Arithmetic Twisting
Let us give an affine S-group G. We will call an arithmetic torsor on f : X → S
a torsor coming by pull back by f from a torsor over S. In this section we will
consider the particular case of the twisting of an arithmetic G-torsor by a G-
torsor Q → X over X. In other words IsomG (PX , Q) = ΦQ (PX 0
) = Q ∧G PX 0
,
where P → S is a G-torsor over S. In this situation, P being locally trivial on
0
S, ΦQ (PX0
) is locally on S isomorphic to Q. We will say that it is a model of Q.
The following statements are immediate consequences of the properties of
the twisting operation.
Proposition 7.1. Let u : U → X be a S-morphism and s : U → S the

composed morphism s = f ◦ u. The following conditions are equivalent :
1. u∗ Q s∗ P ;
2. there exists a section U → u∗ (ΦQ (PX
0
)).
In the case U = Spec(k), where k is a ring, the preceding conditions are
equivalent to the following
3. ΦQ (PX
0
)(k) = ∅.
Proof. It is an immediate consequence of Corollary 5.2 and of base change

properties.
One can state consequences of Proposition 7.1 in particular situations.

Proposition 7.2. Let S = Spec(R) where R is an Henselian ring whose
generic point, special point, field of fractions and residue field will be denoted
η, s, k and κ respectively. Let X → S be a smooth S-scheme, G → S an
étale group scheme, P → S a right G-torsor, Q → X a right GX -torsor. One
assumes there exists a special point x : Spec(κ) → X and a κ-rational point
y : Spec(κ) → ΦQ (PX 0
) over x.
Then there exist generalisations x̃ : S → X of x and these x̃ satisfy
x̃∗ Q P . In particular x̃∗η Q Pη .
Proof. The existence of generalizations x̃ : S → X of x is due to the fact that
X → S is smooth using Hensel Lemma (see [15], Chapter 1, section 4, Ex. 4.13).
On the other hand ΦQ (PX 0
) is étale over X as PX → X is étale and the section
x̃ lifts to a section ỹ : S → ΦQ (PX 0
) : indeed the following diagram is cartesian
x∗ ΦQ (Px0 ) / x̃∗ ΦQ (P 0 )
X

Spec(κ) /S
s
where the vertical maps are étale; the sections of the left vertical map which
correspond to κ-points of ΦQ (PX 0
) over x are in one to one correspondence with
sections of the right vertical map which correspond to S-points of ΦQ (PX 0
) over
x̃ ([15], Chapter 1, section 4, Proposition 4.4). Finally according to Proposition
7.1 the S-points of ΦQ (PX 0
) over x̃ correspond to isomorphisms x̃∗ Q P . By
restriction to the generic fiber one deduces an isomorphism x̃∗η Q Pη .
Proposition 7.3. Let S = Spec(R) where R is an henselian ring whose we
will denote the generic and special points η and s, k the field of fractions of
R and κ its residue field. Let X → S be a smooth S-scheme, G → S an affine
smooth group scheme, P → S a right G-torsor, Q → X a right GX -torsor.
One assumes there exists a κ-rational special point y : Spec(κ) → ΦQ (PX0
).
∗
Then there exists a S-point x̃ : S → X such that x̃ Q P . In particular
x̃∗η Q Pη .
Proof. The proof is similar to that of preceding Proposition. The scheme
ΦQ (PX0
) is smooth over X and thus over S and Hensel’s Lemma insures the
existence of a section ỹ : S → ΦQ (PX
0
) specializing at y. If x̃ : S → X is its
image in X, according to Proposition 7.1, x̃∗ Q P .
142 Michel Emsalem
When the residue field κ of R is finite, one may apply Lang Weil estimates
for the number of κ-rational points of a κ-variety ([12]) and apply Propositions
7.2 and 7.3 to insure the existence of S-points x̃ : S → X such that x̃∗ Q P .
In order to do this we will have to check whether the special fiber of ΦQ (PX 0
)
has κ-rational geometrically irreducible components, a question that we will
address in section 9.
8. Twisting a Ramified Cover

A ramified cover F : Y → X is a finite faithfully flat morphism whose restriction
to some dense open U ⊂ X is étale. Let S be a scheme, X, Y be S-schemes, G →
S an étale goup scheme and F : Y → X be a S-morphism which is a ramified
cover endowed with a compatible action of GX on Y which makes the restriction
of F to U a left G-torsor. Let P → S be a right G-torsor. According to Theorem
3.1 the twisted object ΦPX (Y ) is well defined. The restriction of ΦPX (Y ) to U is
isomophic to ΦPU (Q), where Q → U is the left G-torsor restriction to U of F :
Y → X and is in particular étale. As P is locally trivial for the étale topology,
F̃ : Ỹ = ΦPX (Y ) → X is étale locally on S isomorphic to Y → X. In particular
ΦPX (Y ) → X is finite flat ( [10] Proposition 2.7.1). So F̃ : ΦPX (Y ) → X is a
ramified cover. As a consequence of Corollary 7.1 one gets the following result.
Proposition 8.1. There exists a ramified cover F̃ : Ỹ → X étale locally

isomorphic on S to F : Y → X such that for any S-scheme t : T → S there
exists a T -point y : T → Ỹ over a point x ∈ U (T ) if and only if x∗ Y t∗ P 0
as GT -left torsors.
In the particular case where S = Spec(k) of a field k one obtains the

following consequence :
Corollary 8.2. Let F : Y → X be a ramified cover over k endowed with a com-

patible action of GX on Y which makes the restriction of F to the complement of
the branch locus of f a left G-torsor. Let P → Spec(k) be a right G-torsor. Then
there exists a model F̃ : Ỹ → X over k of Fk̄ : Yk̄ → Xk̄ satisfying the following
property : for any extension k of k and for all unramified x ∈ X(k ) the fiber of
F at x is isomorphic to the G-torsor Pk0 if and only if the fibre at x of F̃ has a
k -rational point. This statement applies in particular to Galois ramified covers.
Proof. It is a consequence of Corollary 7.1 applied to the G-torsor Q obtained

from F : Y → X by removing the branch locus from X. The model F̃ : Ỹ → X
is the unique finite cover of X whose restriction to the complement of the branch
0
locus of F is isomorphic to ΦPX (Q) (recall that ΦPX (Q) (ΦQ (PX 0 0
)) ).
Proposition 8.3. Let S = Spec(R) where R is a discrete valuation ring, η

and s the generic and special points, k the field of fractions of R and κ its
residue field. Let X → S be a proper S-scheme which we assume to be normal
and connected, G be a constant finite group, P → S be a right-G-torsor for
the étale topology, Z → Xη be a Galois ramified cover of group G, with Z
normal. We assume that the normalization F : Y → X of X in Z → Xη is
étale outside a closed S-subscheme D = X. There exists a model F̃ : Ỹ → X
(in the étale local sense over S), such that
1. if there exists a k-rational point y ∈ Ỹη (k) over an unramified point xη ∈
(X \ D)(k) the fiber at xη of Z → Xη is isomorphic to the G-torsor Pη0 .
2. if moreover the unique extension x ∈ X(R) of xη does not meet D,
x∗ Y P 0 .
3. one supposes here that R Henselian and that X is smooth over S; if Ỹs
has a κ-rational point v ∈ Ỹs (κ) over a point u ∈ Xs (κ) not belonging to
D, there exists a section x ∈ (X \ D)(R) extending u such that x∗ Y P 0 .
Proof. As a consequence of the hypotheses the generic fiber of Y → X is

isomorphic to Z → Xη . The action of G on Z extends in an action of G on Y due
to the normality of Y . Denote U = X \ D, Q = F −1 (U ), by hypothesis Q → U
is étale. As U is normal, the same is true for Q ([10], Corollaire 6.5.4). The same
arguments show that Q ×X\D Q and Q ×S G are normal. From that it follows
that the isomorphism f : Qη ×k G Qη ×Xη \Dη Qη and its inverse g = f −1
extend in morphisms f¯ : Q×S G → Q×X\D Q and ḡ : Q×X\D Q → Q×S G over
U and the restrictions of ḡ◦f¯ and f¯◦ḡ to the generic fibers are the identity. Then
f¯ and ḡ are isomorphisms inverses from each other and Q → U is a G-torsor.
One can consider the twisted object ΦPU (Q) → U which is locally isomor-
phic to Q → U for the étale topology and thus étale. Moreover ΦPU (Q) → U
is the restriction to U of Ỹ = ΦPX (Y ) → X.
1. the point y ∈ Ỹη (k) belongs to ΦPU (Q)(k) over xη ∈ U (k). The conclusion
follows from 7.1.
2. the restriction to U of Ỹ → X is finite étale and the unique section
x̃ ∈ U (R) lifts to a section ỹ ∈ Ỹ (R) and belongs in fact to ΦPU (Q)(R).
The conclusion follows from Proposition 7.1.
3. As U → S and ΦPU (Q) → S are smooth the point v ∈ Ỹs (κ) extends in a
section y ∈ ΦPU (Q)(R) over a section x ∈ U (R) which satisfy according
to Proposition 7.1 x∗ Q x∗ Y P 0 .
An example of situation where Proposition 8.3 apply is given by the

following statement.
144 Michel Emsalem
Corollary 8.4. Let S = Spec(R) where R complete discrete valuation ring,

η and s the generic and special points, k the field of fractions of R and κ
its residue field. Let X → S be a smooth proper relative curve over S, G a
finite constant group, P → S a right G-torsor for the étale topology, Z → Xη
a Galois ramified cover of group G. One assumes that the normalization
F : Y → X of X in Z → Xη has no vertical ramification. Then conclusions of
Proposition 8.3 hold.
Proof. Under these hypotheses the morphism F : Y → X is flat and defines
a ramified cover along a divisor D whose components are the closure in X of
the branch points of the cover Z → Xη . One can apply Proposition 8.3.
Remark 8.5. When the center Z(G) of G is trivial, it follows from [2],
Propsition 2.3 that if the residue characteristic does not divide the order of
G and the distinct branch points do not meet on the special fiber, the cover
F : Y → X has no vertical ramification. One may apply 8.4 in this situation.
Let us end this section by a statement which illustrates how Lang-Weil’s
estimates (in this instance Riemann hypothesis in function fields) can be used
in this context.
Proposition 8.6. Let R be a Henselian discrete valuation ring with finite
residue field κ and with fraction field k, X → S = Spec(R) be a smooth proper
R-curve with H0 (X, OX ) = R, f : Y → X a ramified cover with no vertical ram-
ification (finite étale over some open U ⊂ X which surjects onto S), G → S an
étale finite group scheme acting on Y over X such that the restriction Q → U
of f to U is a left G-torsor. Assume Y → S to be smooth and H0 (Y, OY ) = R.
Then there exists a constant C > 0 depending on the degree d of the cover
f , the genus g of the fibers of Y → S and the number r of the branch points,
such that for any finite integral ring extension R ⊂ R of residue field κ , with
[κ : κ] ≥ C and any right G-torsor P → Spec(R ), there exist unramified
R -points x : Spec(R ) → X such that x∗ Y P 0 as left G-torsors.
Proof. It follows from the hypothesis that Y → S has smooth and geo-
metrically connected fibers and thus geometrically irreducible fibers. As
Ỹ = ΦPX (Y ) → S is étale locally isomorphic to Y → S, Ỹ → S has geomet-
rically irreducible fibers. In particular the special fiber Ỹs → Xs is a smooth
ramified cover with less than rd ramification points and Ỹs is geometrically
irreducible. For any finite extension of the residue field κ ⊂ κ , the number
Nκ of κ -points on Ỹs satisfies the inequality

|Nκ − (|κ | + 1)| ≤ 2g |κ |.
So for |κ | large enough (depending on g, r, d) Nκ > rd, and there exist
unramified κ -points v on Ỹs . Let U be the complement of the branch locus
and Q = f −1 (U ); then Q → U is a left G-torsor and v extends in a R -point

y : Spec(R ) → ΦPU Q for any discrete valuation ring extension R of residue
field κ . Let x : Spec(R ) → U ⊂ X the image of y. As in point (3) of
Proposition 8.3, x∗ Y P 0 .
9. Specialization of a Cover
Let k be a field and X → Spec(k) a proper k-sheme. Let F : Y → X be a
ramified cover and U ⊂ X a dense open subscheme such that the restriction
Q = F −1 (U ) → U of F above U is finite étale. We assume X to be geometrically
normal and geometrically connected and Y geometrically normal (geometrically
normal resumes to normal if the base field is perfect, see [1], Lemma 10.151.1 ).
The open subscheme U is obviously geometrically normal. On the other hand,
as for any field extension k ⊂ k , Xk is the normalization of itself in Uk → Xk ,
U is also geometrically connected. As F is faithfully flat, it is open, and Q is a
dense open subscheme of Y . This is true for any base field extension, and thus if
Q is geometrically connected, Y is geometrically connected. On the other hand,
for any field extension k ⊂ k , Yk is the normalization of Xk in Qk → Xk ,
and thus, if Y is geometrically connected, Q is geometrically connected.
Describing finite étale covers of U in terms of morphisms π1 (U, x̄) → Sd ,
one sees that Q is geometrically connected if and only if there isn’t a non trivial
finite field extension k → L such that Q → U factors through Q → UL → U .
In particular if Q is not geometrically connected Q(k) = ∅.
In the case F : Y → X is Galois of group G, then FU : Q = F −1 (U ) → U
is a left G-torsor under the constant group G. The above remarks can easily
be formulated in terms of morphisms of étale fundamental groups. The étale
cover Q → U is described by a surjective morphism Φ : π1 (U, x̄) → G (where
x̄ refers to a geometric point of U ). This morphism Φ inserts in the following
commutative diagram where the vertical maps are surjective
1 / π1 (Uk̄ , x̄) / π1 (U, x̄) u / Gal(k̄/k) /1 (9.1)

ϕ Φ

1 / Ḡ /G v / Gal(L/k) /1
where L is the scalar extension in the covering. This means that the étale cover
Q → U factors through Q → UL → U . The condition L = k (or equivalently
Ḡ = G) is equivalent to the condition that Q is geometrically connected or
equivalently that Y is geometrically connected. We conclude that a necessary
condition for Q to have a k-rational point is that Q is geometrically connected.
146 Michel Emsalem
Coming back to the problem to know, given a left G-torsor P = Spec(K) →

Spec(k) (where K is a finite étale k-algebra), whether the specialization of Q →
X at some rational point x ∈ U (k) is isomorphic to P → Spec(k), Proposition
8.1 gives an answer in terms of k-rational unramified point in Ỹ . In what follows
we won’t assume Y to be geometrically connected. To apply Proposition 8.1 one
has to be able to describe the connected components of Ỹ and to check whether
they are geometrically connected. This is the aim of this more technical section.
In what follows we will assume that the following condition is satisfied :
() the quotient of the left G-torsor P = Spec(K) → Spec(k) by Ḡ is
isomorphic to the G/Ḡ-left torsor Spec(L) → Spec(k)
It is an obvious necessary condition for Q → X to having a rational
specialization isomorphic to P → Spec(k). Suppose that P → Spec(k) is
described by the morphism Ψ : Gal(k̄/k) → G whose kernel Gal(k̄/N ) defines
a Galois extension N of k. If H = Gal(N/k) one may suppose L ⊂ N ,
L = N Ḡ∩H and ḠH = G.
We have seen that a necessary and sufficient condition for the problem
to have an affirmative answer is that Q0 ∧G PU has a k-rational point y; if
x ∈ X(k) is the image of y, the fiber of Q at x will be isomorphic to P . We are
in the situation of Proposition 5.11: the G-torsors Q and PU have a common
quotient RU . The twisted object R0 ∧G/Ḡ R = IsomG/Ḡ (R, R) is a trivial
torsor whose sections over k are in one to one correspondence with Z(G/Ḡ).
It follows that if X has k-rational points (otherwise the question is empty) the
sections X → IsomG/Ḡ (RX , RX ) are themselves in one to one correspondence
with Z(G/Ḡ). Let s be such a section. Let us recall the diagram
s∗ IsomG (Q, PU ) / IsomG (Q, PU ) (9.2)

X o Z [ ] _ 0 IsomG/Ḡ (RU , RU )
s
and if one pulls it at the rational point x ∈ U (k) one gets
x∗ (s)∗ IsomG (x∗ Q, P ) / IsomG (x∗ Q, P ) (9.3)

Spec(k) o \ ] _ a 0 IsomG/Ḡ (R, R)
x∗ (s)
The sections Spec(k) → IsomG (x∗ Q, P ) induce sections x∗ (s) : Spec(k) →

IsomG/Ḡ (R, R) for some s corresponding to some element of Z(G/Ḡ)
and for a given s are in one to one correspondence with sections

Spec(k) → x∗ (s)∗ IsomG (x∗ Q, P ), in other words k-rational points of
s∗ IsomG (Q, PU ) above x. We state the result:
Proposition 9.1. Assume that condition () is fulfilled. With the preceding
notation let {sγ }γ∈Z(G/Ḡ) be the set of sections Spec(k) → IsomG (R, R). Then
for all γ ∈ Z(G/Ḡ), s∗γ (Q ∧G PU0 ) is geometrically connected. Moreover there
exists an unramified rational point x ∈ X(k) such that the fiber at x of Y is
isomorphic to the G-torsor P if and only if there exists γ ∈ Z(G/Ḡ) such that
s∗γ (Q ∧G PU0 )(k) = ∅.
Proof. The only thing to prove is that the s∗γ (Q ∧G PX0

)’s are geometrically
connected. Recall the cartesian diagram of Proposition 5.11.
IsomḠ,XL (Q, (1X × γ)∗ PX ) / s∗ IsomG (Q, PX ) / IsomG (Q, PX )

/XoZ [
XL ] _ 0 IsomG/Ḡ (XL , XL )
sγ
(9.4)
But (1X × γ)∗ PX (γ ∗ P )X where γ ∗ P → Spec(L) is a Ḡ-torsor. And
IsomḠ,XL (Q, (1X × γ)∗ PX ) Q ∧Ḡ (γ ∗ P )0XL is isomorphic over Xk̄ to Q ×L k̄
which is connected as L is the extension of scalars contained in Q.
Remark 9.2. Instead of sections over k one can more generally consider
sections over a finite extension of k. For instance if one works over L, sections
Spec(L) → IsomG (R, R) are in one to one correspondence with G/Ḡ (the inner
form of G/Ḡ induced by the torsor R = Spec(L) → Spec(k) split over L and
is thus isomorphic to G/Ḡ). Thus over L, IsomG (QL , PXL ) = QL ∧G PX 0
L
is a
∗
disjoint union of the open closed sγ IsomG (QL , PXL ) with γ running in G/Ḡ.
The situation reduces to that of the regular G-cover QL → XL .
When k is a pseudo algebraically closed field (abbreviated by PAC) the
conclusion of Proposition 9.1 always holds. Recall that a field k is a PAC field
if every geometrically irreducible variety defined over k has k-rational points
(see [11]). As in Proposition 9.1 the components s∗γ (Q ∧G PU0 ) are geometrically
irreducible, one gets the following corollary.
Corollary 9.3. Assume that condition () is fulfilled. Then if k is a PAC
field there are infinitely many unramified rational points x ∈ X(k) such that
the fiber at x of Q is isomorphic to the G-torsor P .
One can generalize Proposition 9.1 to the situation of schemes over a dis-
crete valuation ring instead of a field. To avoid confusion in the notation, let us
148 Michel Emsalem
call A this discrete valuation ring; we denote as usual k its field of fractions and
κ is residue field. Let X → Spec(A) be a faithfully flat and proper A-scheme
that we will assume to be integral and normal. Let Z → Xη be a Galois ramified
cover of the generic fiber of Galois group G and geometric Galois group Ḡ < G.
Let us assume that the normalized f : Y → X of X in Z is flat1 . Assume also
that f : Y → X has no vertical ramification (this means that its restriction to
the special fiber is a ramified cover). Let Spec(L) → Spec(k) be the extension
of scalars in the Galois cover Z → Xη (L is a Galois extension of k of group
G/Ḡ) and AL the integral closure of A in L that we assume étale over A. We
suppose that there exists an AL -point x : Spec(AL ) → XAL . Let P → Spec(A)
be an étale G-torsor such that the generic fiber Pη → Spec(k) factors through
Pη → Spec(L) → Spec(k). We are in the situation of Proposition 5.11.
Proposition 9.4. Under these hypothesis
1. R = Spec(AL ) → Spec(A) is an étale G/Ḡ-torsor and there are
factorizations P → Spec(AL ) → Spec(A) and Y → XAL → X;
2. IsomG/Ḡ (XAL , XAL ) = IsomG/Ḡ (R, R) = H0 (Spec(A), IsomG/Ḡ (R, R))
IsomG/Ḡ (Rη , Rη ) Z(G/Ḡ);
3. for all γ ∈ Z(G/Ḡ), if we denote by sγ ∈ IsomG/Ḡ (XAL , XAL ) =
0 0
ΦRX (RX ) the corresponding element, s∗γ (ΦPX (Y )) has geometrically
connected fibers.
Proof. The first assertion comes from the normality of P and Y . In the second
assertion the first equality comes from the existence of a point Spec(AL ) → X.
Other equalities are clear. As L is the constant field extension in fη : Yη → Xη ,
one gets H0 (Yη , OYη ) = L. On the other hand H0 (Yη , OYη ) H0 (Y, OY ) ⊗A k
(cf. [14], section 5, Ex. 1.16, p. 174). As H0 (Y, OY ) has no A–torsion, so
AL ⊂ H0 (Y, OY ) ⊂ H0 (Y, OY ) ⊗A k = L and as H0 (Y, OY ) is a finite A-algebra
by Serre’s theorem, it is the integral closure of A in L: H0 (Y, OY ) = AL .
It follows from this fact that Y → XAL has geometrically connected fibers.
As in the proof of Proposition 9.1 one uses Proposition 5.11 to show that
0
s∗γ (ΦPX (Y )) → X is étale locally over Spec(A) isomorphic to Y → Spec(AL )
and thus has geometrically connected fibers.
In the situation of Proposition 9.4, let us give an example of application of

the Lang-Weil estimates.
Proposition 9.5. Under the hypothesis of Proposition 9.4, assume moreover
that X and Y are smooth relative curves over Spec(A) and that A is Henselian
1 This will be always the case for a regular scheme X of relative dimension 1.
with finite residue field κ. There is a constant C > 0 depending on the degree
of the covering Y → X, the genus of the fibers of Y and the number of
branch points, such that for any discrete valuation ring extension A of A
whose residue field κ satisfies [κ : κ] ≥ C there are unramified A -points
x : Spec(A ) → X such that x∗ Y PA as left G-torsors.
Proof. The proof is similar to the proof of Proposition 8.6 taking advantage
of the fact that Y → X is smooth and A is Henselian. One uses Lang-Weil
estimates [12] to insure the existence of κ -unramified points on the special
0
fiber of s∗γ (ΦPX (Y )) for [κ : κ] large enough.
The viewpoint of [6] is different : for a Galois extension N of k the authors

ask if there are specializations of the étale covering Q of U at unramified points
x ∈ X(k) isomorphic to a disjoint union of (G : H) copies of N forgetting the
action of G. So they have to consider all G-torsors over Spec(k) associated to
injective morphisms H → G whose composition with the canonical surjection
G → G/Ḡ is surjective. One is lead to look for k-rational points on a family of
schemes indexed by embedding H → G whose composition with the canonical
surjection G → G/Ḡ is surjective. Each of these schemes is the twisted object
Q ∧ G PX 0
where P runs among the above mentioned G-torsors. The answer to
the question can be formulated as follows.
Proposition 9.6. Let N be a Galois extension of group H. Consider the set
J of embeddings j : H → G whose composition with the canonical surjection
G/Ḡ is surjective, modulo automorphisms of G fixing H. For each j ∈ J
denote by Kj the finite étale k-algebra such that Pj = Spec(Kj ) → Spec(k) is
the G-torsor contracted product of the H-torsor Spec(N ) → Spec(k) with the
embedding j : H → G. The following conditions are equivalent :
1. There exists x ∈ X(k) such that the fiber at x of Q is isomorphic to a
disjoint union of (G : H)-copies of Spec(N ) → Spec(k).
2. There exists j ∈ J such that (Q ∧G Pj,X
0
)(k) = ∅.
3. There exists j ∈ J and γ ∈ Z(G/Ḡ) such that s∗γ (Q ∧G Pj,X
0
)(k) = ∅.
In [6], section 3.2 the authors consider more generally the case of non neces-
sarily Galois covers. Let F : Y → X be a connected ramified cover of degree n
étale above a connected dense open U ⊂ X, Φ : π1 (U, x̄) → Sn the correspond-
ing morphism, G < Sn the image of Φ, Z → F −1 (U ) → U the Galois closure of
F −1 (U ) → U , corresponding to the surjective morphism φ : π1 (U, x̄) → G and
Q → U the associated Sn -torsor which is the contracted product of Z by Sn
via the inclusion G < Sn . Let L be the extension of scalars in Z → U . We are
also given an extension of étale k-algebras N of k of degree n corresponding
to a morphism Ψ : Gal(k̄/k) → Sn and let H be the image of Ψ which is the
150 Michel Emsalem
Galois group of the Galois closure N of N over k (which can be viewed as

the compositum in k̄ of the Galois closures of the components of N ). We call
P → Spec(k) the Sn -torsor associated to the morphism Ψ. The Sn -torsor Q is
the contracted product Z ∧G Sn and splits in (Sn : G) connected components
isomorphic to Z. In the same way the Sn -torsor P is the contracted product
Spec(N )∧H Sn and has (Sn : H) connected components isomorphic to Spec(N ).
The Sn -torsor Q → U restriction to U of Y → X has the following
description, x being a k-rational point of U and sx the corresponding section,
defined up to conjugation by an element of π1 (Uk̄ , x̄) :
q _
sx
1 / π1 (Uk̄ , x̄) / π1 (U, x̄) / Gal(k̄/k) /1 (9.5)
u
ϕ φ Λ

1 / Ḡ /G / G/Ḡ /1
v
ν

Sn
where v ◦ φ ◦ sx = Λ. The fiber of Q at x is described by
Gal(k̄/k)
sx
/ π1 (U, x̄) φ
/G ν / Sn (9.6)
We write Φ = ν ◦ φ and Φx = ν ◦ φ ◦ sx . On the other hand the Sn -torsor

P has the following description :
Gal(k̄/k)
ψ
/H μ
/ Sn (9.7)
Let us denote Ψ = μ ◦ ψ. The specialization of Y → X at x is isomorphic

to Spec(N ) → Spec(k) if and only if up to conjugation by an element of Sn ,
ν ◦ φ ◦ sx = μ ◦ ψ. Identifying H and G with their images in Sn this implies
the existence of an embedding η : H → G such that
ν◦η =μ and η ◦ ψ = φ ◦ sx .
Conversely suppose that there is an embedding η : H → G such that

ν ◦ η = μ and such that the G-torsor P = Spec(N ) ∧H G → Spec(k) through
the embedding η : H → G is isomorphic to the G-torsor x∗ Z described by the
morphism φ ◦ sx : Gal(k̄/k) → G. There exists an element ω ∈ G such that
∀γ ∈ Gal(k̄/k) ω(φ ◦ sx (γ))ω −1 = η ◦ ψ(γ)

which implies
∀γ ∈ Gal(k̄/k) ν(ω)(ν ◦ φ ◦ sx (γ))ν(ω)−1 = ν ◦ η ◦ ψ(γ) = μ ◦ ψ(γ)
and thus up to conjugation by an element of Sn , ν ◦ φ ◦ sx = μ ◦ ψ which

means that the specialization of Y → X at x is isomorphic as cover to N .
One can consider the twisted object Z ∧G PU and the existence of
0
k-rationals points on this k-scheme over a point x ∈ X(k) is equivalent to the

existence of an isomorphism of G-torsors x∗ Z P . One gets the following
statement which is a reformulation of the ”Twisting Lemma” 3.4 of [6]:
Proposition 9.7. With the preceding notation the finite étale k-schemes
x∗ Y → Spec(k) and Spec(N ) → Spec(k) are isomorphic if and only if there
exists an embedding η : H → G such that
1. ν ◦ η = μ;
2. the fiber at x of Z ∧G PU
0
has a k-rational point where
P = Spec(N ) ∧H G → Spec(k) through the embedding η : H → G.
Acknowledgments
The author would like to thank the referee for her/his thorough reading of the
article and her/his many valuable suggestions which have improved the text.
References
[1] AA. VV., Stacks Project, version 94a58fd
[2] S. Beckmann, On extensions of number fields obtained by specializing branched
coverings, Journal fr die reine und angewandte Mathematik, 419 (1991) p. 27 - 54.
[3] L. Breen, Notes on 1- and 2-gerbes, dans ”Towards Higher Categories”,
J.C. Baez and J.P. May (édit.), The IMA Volumes in Mathematics and its
Applications, Tome 152, 193-235, Springer (2009), arXiv:math/0611317.
[4] P. Dèbes, Galois covers with prescribed fibers: the Beckmann-Black problem,
Ann. Scuola Norm. Sup. Pisa, Cl. Sci. 4, vol. 28, 1999, 273-286.
[5] P. Dèbes, On the Malle conjecture and the self-twisted cover, arXiv:1404.4074.
[6] P. Dèbes and F. Legrand, Twisted covers and specializations. Galois-Teichmller
theory and arithmetic geometry, 141162, Adv. Stud. Pure Math., 63, Math. Soc.
Japan, Tokyo, (2012).
[7] P. Dèbes and F. Legrand, Specialization results in Galois theory. Transactions
A. M. S., vol. 365, no 10, 2013, 5259-5275
152 Michel Emsalem
[8] P. Dèbes and N. Ghazi, Galois covers and the Hilbert-Grunwald property. Ann.
Inst. Fourier, vol. 62, no 3, 2012, 989-1013.
[9] P. Dèbes and N. Ghazi, Specializations of Galois covers of the line. Alexandru
Myller Mathematical Seminar, Proceedings of the Centennial Conference, V.
Barbu and O. Carja, dds, American Institute of Physics, vol. 1329, 2011, 98-108,
[10] A. Grothendieck, Éléments de géomérie algébrique. IV. Étude locale des schémas
et des morphismes de schémas. 2, Publications Mathématiques de l’IHÉS, 24,
(1965).
[11] M. Fried, M. Jarden, Field arithmetic., Ergebnisse der Mathematik und ihrer
Grenzgebiete. 3. Folge 11, Springer Verlag.
[12] S. Lang, A. Weil, Numbers of points of varieties in finite fields, American
Journal of Mathematics, Vol. 76, 4, 1954, pp. 819-827.
[13] G. Laumon, L. Moret-Bailly, Champs algébriques, Ergebnisse der Mathematik
und ihrer Grenzgebiete : 3. Folge; 39 (2000), Springer Verlag.
[14] Q. Liu, Algebraic geometry and arithmetic curves, Oxford Science Publications
(2002)
[15] J. S. Milne, Étale Cohomology, Princeton University Press (1980F).
[16] M. V. Nori, The fundamental group-scheme, Proc. Indian Acad. Sci. (Math.
Sci.), Vol. 91, Number 2, (1982), p. 73-122.
[17] J. -P. Serre, Cohomologie galoisienne, Lecture Notes in Mathematics, Vol. 5,
Springer-Verlag, Berlin-New York, 5-th edition, 1994.
[18] A. Vistoli, Notes on Grothendieck Topologies, Fibred Categories and Descent
Theory, in Grothendieck’s FGA explained, Math. Surveys and Monographs of
the AMS, 123 (2005).
Hitchin Hamiltonians in Genus 2
Viktoria Heu∗,† and Frank Loray‡,§
Abstract
We give an explicit expression of the Hitchin Hamiltonian system for rank two
vector bundles with trivial determinant bundle over a curve of genus two.
Mathematics Subject Classification (1991). Primary 14H60; Secondary 34Mxx,

32G34, 14Q10
Keywords. Vector Bundles, Moduli Spaces, Higgs Bundles, Kummer Surface
1. Introduction
We are interested in rank two vector bundles E → X with trivial determi-
nant bundle det(E) = OX over a Riemann surface X of genus 2. The moduli
space MN R of semistable such vector bundles up to S-equivalence has been
constructed by Narasimhan and Ramanan in [16]. If E ∈ MN R is stable (and
is therefore the unique vector bundle S-equivalent to E), the cotangent space
of MN R at E is canonically isomorphic to the moduli space of trace free holo-
morphic Higgs fields θ : E → E ⊗ Ω1X on E:
T∨
E MN R Higgs(X)|E .
Since MN R P3 and the locus of semistable but non stable bundles (up to
S-equivalence) there is given by a singular quartic hypersurface, we have
T∨ MN R Higgs(X)
∗ IRMA, 7 rue René-Descartes, 67084 Strasbourg Cedex, France.
† The first author is supported by the ANR grants ANR-13-BS01-0001-01 and ANR-13-
JS01-0002-01.
‡ IRMAR, Université de Rennes I, Campus de Beaulieu, 35042 Rennes Cedex, France.
§ The second author is supported by CNRS and the ANR grant ANR-16-CE40-0008.
Geometry, DOI 10.1007/978-981-10-5648-2_10
154 Viktoria Heu and Frank Loray
in restriction to a Zariski open subset of MN R , where Higgs(X) denotes the

moduli space of tracefree holomorphic Higgs bundles (E, θ).
In [13], Hitchin considered the map
& '
Higgs(X) → H0 (X, Ω1X ⊗ Ω1X )
Hitch : (1.1)
(E, θ) → det(θ)
and established that it defines an algebraically completely integrable Hamil-
tonian system: the Liouville form on MN R induces a symplectic structure on
Higgs(X) and any set of (three) generators of H0 (X, Ω1X ⊗ Ω1X ) commutes for
the induced Poisson structure. Moreover, fibers of the Hitchin map are open
sets of abelian varieties whose compactification is given by the Jacobian of the
spectral curve. A broad field of applications has been deduced from the various
algebraic and geometric properties of the Hitchin system and its generalizations
since then.
Of course the Hitchin system in [13] is defined in a more general setting,
but in the present paper, we focus on the special case as above (rank 2 vector
bundles with trivial determinant over curves of genus 2) and announce results
of a forthcoming paper [12]:
• We describe the moduli space Bun(X) of (not necessarily semistable)
vector bundles E equivariant under the hyperelliptic involution ι on X.
On the categorical quotient we have a birational morphism
∼
Bun(X) MN R .
• It is well-known that there is no universal bundles on a Zariski-open

subset of MN R . Yet from the dictionary between equivariant bundles
and parabolic bundles on the quotient established in [3] (see also [2]) we
obtain a rational two-cover
2:1
Bun(X/ι) Bun(X)
and we construct a universal bundle over affine charts of Bun(X/ι) which
can be identified with the universal bundle in [4] obtained from different
methods.
• We deduce a universal family of Higgs bundles on affine charts of
Higgs(X/ι) := T∨ Bun(X/ι). Note that in restriction to the stable lo-
cus, Higgs(X/ι) → Bun(X/ι) is a principal C3 -bundle.
• These explicit universal families allow us to calculate the determinant
map on Higgs(X/ι) explicitly, which by construction factors through the
Hitchin map. We deduce an explicit expression of the Hitchin map (1.1)
completing partial results in [7].
Hitchin Hamiltonians in Genus 2 155
2. The Narasimhan-Ramanan Moduli Space

MN R
Let us first briefly recall the classical Narasimhan-Ramanan construction. Let
E → X be a semistable rank 2 vector bundle with trivial determinant bundle
over a Riemann surface X of genus 2. The subset
CE := {L ∈ Pic1 (X) | dim H0 (X, E ⊗ L) > 0}
of Pic1 (X) defines a divisor DE on Pic1 (X) which is linearly equivalent to

the divisor 2Θ on Pic1 (X), where Θ denotes the theta-divisor defined by the
canonical embedding of X in Pic1 (X). In that way, we associate to the vector
bundle E an element DE of the Narasimhan-Ramanan moduli space
MN R := PH0 (Pic1 (X), OPic1 (X) (2Θ)) P3 .
For each smooth (analytic or algebraic) family E → X × T of semistable rank 2

vector bundle with trivial determinant bundle on X, the Narasimhan-Ramanan
classifying map
T → MN R ; t → DE|X×{t}
is an (analytic or algebraic) morphism. Moreover, if DE = DE , then the vector
bundles E and E are S-equivalent:
• either E is stable and then E = E
• or E is strictly semistable (i.e. semistable but not stable) and then there
are line subbundles L and L of degree 0 of E and E respectively, such
that L = L or L = L⊗−1
The strictly semistable locus in MN R , which we shall denote by Kum(X)

is thus defined by an embedding
& '
Pic0 (X)/ι → MN R
.
L mod ι → DL⊕L⊗−1 = L · Θ + ι∗ L · Θ
Note that if ι denotes the hyperelliptic involution on X, then ι∗ L = L⊗−1 by

a classical argument that will be recalled in Section 2.1.
2.1. Straightforward Coordinates on MN R . Any compact con-

nected Riemann surface X of genus 2 can be embedded into P1 × P1 and is
given, in a convenient affine chart, by an equation of the form
X : y 2 = F (x) with F (x) = x(x − 1)(x − r)(x − s)(x − t). (2.1)

The hyperelliptic involution on X then writes ι : (x, y) → (x, −y) and the
induced projection on the Riemann sphere P1 is given by π : (x, y) → x.
Denote by
W := {w0 , w1 , wr , ws , wt , w∞ }
the six Weierstrass points on X invariant under the hyperelliptic involution,

given by wi = (i, 0) for i = ∞ and w∞ = (∞, ∞). We will write W for π(W ).
Recall that the rational map
& '
X2 Pic2 (X)
{P, Q} → [P ] + [Q]
is surjective. More precisely, it is a blow-up of the canonical divisor
KX ∼ [P ] + [ι(P )] for all P ∈ X.
Moreover, Pic2 (X) Pic1 (X); D → D − [w∞ ] is an isomorphism. Global

sections of OPic1 (2Θ) thus correspond bijectively to symmetric meromorphic
functions on X × X with polar divisor at most 2Δ + 2∞1 + 2∞2 , where
Δ := {(P, ι(P )) | P ∈ X} and ∞i := {(P1 , P2 ) | Pi = w∞ }. Any set
of (four) generators of the vector space H0 (Pic1 (X), OPic1 (2Θ)) can be ex-
pressed as such meromorphic functions, the simplest one being certainly the set
{1, Sum, Prod, Diag} of functions in (P1 , P2 ) ∈ X × X given, for Pi = (xi , yi ),
by
1 : (P1 , P2 ) → 1
Sum : (P1 , P2 ) → x1 + x2
Prod : (P1 , P2 ) → x1 x2 ,
2
y2 −y1
Diag : (P1 , P2 ) → x2 −x1 − (x1 + x2 )3 + (1 + r + s + t)(x1 + x2 )2 +
+x1 x2 (x1 + x2 ) − (r + s + t + rs + st + tr)(x1 + x2 )
(2.2)
We obtain coordinates on MN R = PH0 (Pic1 (X), OPic1 (2Θ)) P3 , where we
identify a point (v0 : v1 : v2 : v3 ) with the push-forward DE on Pic1 (X) of the
zero-divisor of the meromorphic function
v0 · 1 + v1 · Sum + v2 · Prod + v3 Diag (2.3)
on X (2) .
For example if E = L ⊕ L−1 , where
L = OX ([Q1 ] + [Q2 ] − 2[w∞ ]) = OX ([Q1 ] − [ι(Q2 )])

and Q1 , Q2 ∈ X, we can calculate the (unique up to scalar) meromorphic

function as in (2.3) whose 0-divisor corresponds to DE = L · Θ + ι∗ L · Θ and
we obtain
(v0 : v1 : v2 : v3 ) = (−Diag(Q1 , Q2 ) : Prod(Q1 , Q2 ) : −Sum(Q1 , Q2 ) : 1).

(2.4)
The strictly semistable locus Kum(X) in MN R is parametrized by
{Q1 , Q2 } ∈ X (2) according to formula (2.4). We deduce an equation for
Kum(X) in our coordinates (v0 : v1 : v2 : v3 ) of MN R by eliminating vari-
ables (xi , yi ) = Qi ∈ X in equation (2.4). We get
Kum(X) :
0= (v22 − 4v1 v3 )v02 +

+ v12 − v1 v3 (r + s + t) − v3 (v1 + v2 )(rs + st + tr)
2
−v3 (v2 + v3 )rst) + 4v12 v32 rst − 4v1 v33 (rs + st + tr)rst
−4v1 v3 (v2 + v3 )(v2 + v3 (r + s + t))rst − 2v0 v12 (v2 + 2v3 )
−2v0 v1 v3 (v2 + 2v1 )(r + s + t) − 2v0 v3 (v22 + v1 (v2 − 2v3 ))
(rs + st + tr) + 2v0 v3 (2v1 v3 − v2 (v2 + v3 ))rst
(2.5)
Since the strictly semistable locus Kum(X) is a quartic with sixteen conic
singularities it is usually referred to as the Kummer surface in the context of
MN R .
Let OX (τ ) be a 2-torsion line bundle on X, i.e. OX (2τ ) OX . Then
τ ∼ [wi ] − [wj ] with wi , wj ∈ W (2.6)
and the group of 2-torsion line bundles on X with respect to the tensor prod-
4
uct is isomorphic to (Z/2Z) . If E is a rank two vector bundle with trivial
determinant bundle over X, then its twist E ⊗ OX (τ ) also has the trivial de-
terminant line bundle. Moreover, by construction of the Narasimhan-Ramanan
moduli space, the action of the group of 2-torsion line bundles by twist is linear
and free on MN R and preserves Kum(X). By formula (2.4) we can explicitly
calculate the coordinates (v0 : v1 : v2 : v3 ) of the trivial bundle E0 = OX ⊕ OX
and its twists
Eτ := E0 ⊗ OX (τ ).
The trivial bundle for example is given by E0 : (1 : 0 : 0 : 0). Note that
these sixteen bundles Eτ correspond to the sixteen singularities of the Kummer
surface Kum(X).
The fact that we know the action (by permutation) of the 2-torsion group
on the set of bundles Eτ and we also know the coordinates of these bundles
in the Narasimhan-Ramanan moduli space is sufficient to calculate explicitly

the linear action of the 2-torsion group on MN R : for any τ as in (2.6), there
is a matrix Mτ ∈ SL4 C such that if the image of E under the Narasimhan-
Ramanan classifying map is given by (v0 : v1 : v2 : v3 ), then E = E ⊗ OX (τ )
is given by (v0 : v1 : v2 : v3 ) with
⎛ ⎞ ⎛ ⎞
v0 v0
⎜v1 ⎟ ⎜v 1 ⎟
⎜ ⎟ = Mτ ·⎜ ⎟
⎝v 2 ⎠ ⎝v 2 ⎠ .
v3 v3
The equivalence classes in PGL4 C of these matrices (with respect to a set of

generators of the 2-torsion group) then are given by the following:
⎛ ⎞
0 rs + st + rt + rst rst 0
⎜0 0 0 rst ⎟
M[w0 ]−[w∞ ] ∼⎜
⎝1
⎟
0 0 −(rs + st + rt + rst)⎠
0 1 0 0
⎛ ⎞
1 r + s + t + rst rs + st + rt 0
⎜−1 −1 0 rs + st + rt ⎟
M[w1 ]−[w∞ ] ∼⎜
⎝1
⎟
0 −1 −(r + s + t + rst)⎠
0 1 1 1
⎛ ⎞
r2 r2 (1 + s + t) + st r2 (s + t + st) 0
⎜−r −r2 0 r2 (s + t + st) ⎟
M[wr ]−[w∞ ] ∼⎜
⎝1
⎟
0 −r2 −r2 (1 + s + t) − st⎠
0 1 r r2
⎛ ⎞
s2 s2 (1 + r + t) + rt s2 (r + t + rt) 0
⎜−s −s2 0 s2 (r + t + rt) ⎟
M[ws ]−[w∞ ] ∼⎜
⎝1
⎟
0 −s2 −s (1 + r + t) − rt⎠
2
0 1 s s2
2.2. Nice Coordinates on MN R . A quick calculation shows that

the character of the representation
& 4
'
(Z/2Z) → SL4 C
ρ:
OX (τ ) → Mτ
introduced above is the regular one : it vanishes on all elements of the group
exept OX . Hence ρ is conjugated for example to the regular representation
/τ given by
ρ. : OX (τ ) → M
⎛ ⎞ ⎛ ⎞
0 0 1 0 0 1 0 0
⎜0 0 0 1⎟ ⎜−1 0 0 0⎟
/[w ]−[w ]
M = ⎜ ⎟ , /[w ]−[w ]
M = ⎜ ⎟
0 ∞ ⎝1 0 0 0⎠ 1 ∞ ⎝0 0 0 −1⎠
0 1 0 0 0 0 1 0
⎛ ⎞ ⎛ ⎞
1 0 0 0 0 1 0 0
⎜0 −1 0 0⎟ ⎜1 0 0 0⎟
/[w ]−[w ]
M = ⎜ ⎟ , /[w ]−[w ]
M = ⎜ ⎟
r ∞ ⎝0 0 −1 0⎠ s ∞ ⎝0 0 0 −1⎠
0 0 0 1 0 0 −1 0
Any conjugation matrix M ∈ SL4 C such that ρ. = M ρM −1 is given, up to a

scalar, as follows: Choose square-roots ω0 , ω1 , ωr , ωs such that
ω02 = F (0), ω12 = −F (1), ωr2 = F (r), ωs2 = F (s),
where F (x) is given in (2.1) and F (x) is its derivative with respect to x. Then
⎛ ⎞ ⎛ ⎞
a b c d 1 1 0 −ω0
⎜−b a d −c⎟ ⎜0 ω 1 0 0 ⎟
M =⎜
⎝c
⎟·⎜ ⎟, (2.7)
d a b ⎠ ⎝0 ω 0 ω0 ω0 ⎠
d −c −b a 0 0 0 ω0 ω1
where a = rst(r − s)ω1 + tωr ωs − rt(r − 1)ωs − stω1 ωr
b = −st(s − 1)ωr + rtω1 ωs

c = t(r − s)ω0 ω1 − t(r − 1)ω0 ωs
d = −t(r − 1)(s − 1)(r − s)ω0 + t(s − 1)ω0 ωr .
After the coordinate-change (v0 : v1 : v2 : v3 ) → (u0 : u1 : u2 : u3 ) on MN R
defined by
⎛ ⎞ ⎛ ⎞
u0 v0
⎜ u1 ⎟ ⎜v1 ⎟
⎜ ⎟ = M · ⎜ ⎟,
⎝ u2 ⎠ ⎝v2 ⎠
u3 v3
the action of the 2-torsion group is then normalized to ρ.. In particular, the
equation of the Kummer surface with respect to the coordinates (u0 : u1 :
u2 : u3 ) is invariant under double-transpositions and double-changes of signs.
Calculation shows
Kum (X) :
0 = (u40 + u41 + u42 + u43 )
rs − rt + r − s st + t − 2s 2 2 (2.8)
−8 u 0 u1 u2 u3 − 2 (u0 u3 + u21 u22 )
t(s − 1) t(s − 1)
2r − t 2 2 2r − s − 1 2 2
−2 (u1 u3 + u20 u22 ) + 2 (u2 u3 + u20 u21 ).
t s−1
In summary, the straightforward coordinates (v0 : v1 : v2 : v3 ) of MN R
introduced in the previous section have the advantage that
• a given divisor DE on Pic1 (X) linearly equivalent to 2Θ can rather easily

be expressed in terms of (v0 : v1 : v2 : v3 ),
• and we are going to use this property when we describe the universal
family on a 2-cover of MN R ,
whereas the new coordinates (u0 : u1 : u2 : u3 ) of MN R defined above have the

advantages that
• the action of the 2-torsion group is simply expressed by double-

transpositions and double-changes of signs of (u0 : u1 : u2 : u3 ) and
• the equation of the Kummer surface is rather symmetric. As pointed out

in [7], the classical line geometry for Kummer surfaces in P3 is related
to certain symmetries of the Hitchin Hamiltonians. For this geometrical
reason, the explicit Hitchin Hamiltonians we are going to establish have
a much simpler expression with respect to (dual) coordinates (u0 : u1 :
u2 : u3 ) when compared to (v0 : v1 : v2 : v3 ).
• Moreover, the five ui -polynomials in (2.8) invariant under the action of

the 2-torsion group define a natural map MN R → P4 . The image is
a quartic hyper surface [5, Proposition 10.2.7] and can be seen as the
coarse moduli space of semistable P1 -bundles over X.
3. How to Construct a Bundle from a Point

in MN R
Given a stable rank 2 vector bundle E with trivial determinant bundle on
X, the Narasimhan-Ramanan divisor DE ∈ |2Θ| can be seen as space of line
subbundles L of E of degree −1. Whilst we know that for any D ∈ |2Θ| there is
a semistable vector bundle E with DE = D, it is not obvious how to construct
it. We provide such a construction by considering the moduli space of rank

2 vector bundles E with trivial determinant bundle on X equivariant under
the hyperelliptic involution. For the present exposition however, we restrict
our attention to the space Bun(X) of rank 2 vector bundles E with trivial
determinant bundle on X such that
• E is stable but off the odd Gunning planes, which means that no line
subbundle L ⊂ E is isomorphic to OX (−[wi ]) for some wi ∈ W , or
• E is strictly semistable but undecomposable, or
• E = L ⊕ ι∗ L where L = OX ([P ] − [Q]) satisfies P, Q ∈ W , or
• E is an odd Gunning bundle, i.e given by the unique non-trivial extension
0 −→ OX ([wi ]) −→ E −→ OX (−[wi ]) −→ 0
for a Weierstrass point wi ∈ W .
We construct Bun(X) as an algebraic stack whose categorical quotient is bira-

tional to the Narasimhan-Ramanan moduli space
1:1
Bun(X) MN R .
For convenience of notation let us for now denote by Bun(X) the set of
vector bundles E as in the above list, before we put an algebraic structure on
Bun(X). We use the fact that any bundle E ∈ Bun(X) is equivariant under
the hyperelliptic involution:
Proposition 3.1 Let E be a vector bundle in Bun(X). Then there is a bundle

isomorphism h such that the following diagram commutes
∼ / ι∗ E ∼ / E.
E h ι∗ h >
idE
and such that for each Weierstrass point wi ∈ W , the induced automorphism
of the Weierstrass fibre
h|Ewi : Ewi → ι∗ Ewi Ewi
possesses two opposite eigenvalues +1 and −1.

Now, hyperelliptic descent [3]

π∗ E = (E + , p+ ) ⊕ (E − , p− )
produces two rank 2 vector bundles E ± with determinant bundle det(E ± ) =

OP1 (−3) over the Riemann sphere, each endowed with a natural quasi-parabolic
structure p± with support W = π(W ). Moreover,
Proposition 3.2 Consider (E, h) as in Proposition 3.1. Denote by p+ and p−
the quasi-parabolic structure with support W on E induced by the +1 and −1
eigendirections of h respectively. Then
±
(E, p± ) = elm+ ∗ ±
W (π (E , p )),
where elm+W denotes the composition of six positive elementary transformations,

one over each Weierstrass point, given by the corresponding quasi-parabolic
direction of π ∗ (E ± , p± ).
In convenient local coordinates (ζ, Y ) ∈ U ×C2 of E → X near a Weierstrass

∗
point wi : {ζ = 0}, the map elm+W ◦π can be understood as follows:
0
p|wi : {Y ∈ VectC ( 10 )} (ζ, Y )
h / −ζ, 10 −1 Y
O O
elm+
p̂ elm+
p̂

p̂|wi : {Y ∈ VectC ( 01 )} ζ, Y = ζ, 0 / −ζ, Y
1 0 h
1
ζ
Y
O O
π∗ π∗

ζ, Y = ζ 2 , Y / ζ, Y
id
p|π(wi ) : {Y ∈ VectC ( 01 )}
Let μ be a real number in [0, 1]. Denote by Bunμ (X/ι) the moduli space
of pairs (E, p), where E is a rank 2 vector bundle of degree −3 over P1 and
p is a quasi-parabolic structure with support W such that (E, p) is a sta-
ble parabolic bundle if to each quasi-parabolic direction p|wi we associate the
parabolic weight μ. For each choice of μ, this moduli space is either empty or
birational to P3 [15]. Moreover, for any μ ∈ [0, 1], the map
OP1 (−3) ⊗ elm+
W
is a canonical birational isomorphism between Bunμ (X/ι) and Bun1−μ (X/ι).

Note further that for μ = 15 , the space Bunμ (X/ι) is precisely the moduli
space of those quasi-parabolic bundles (E, p), where E is a vector bundle on
P1 and p is a quasi parabolic structure with support W on E such that
• E = OP1 (−1) ⊕ OP1 (−2),

• the quasi-parabolic directions pw are all disjoint from the total space of
i
the destabilizing subbundle OP1 (−1) ⊂ E and
• the quasi-parabolic directions pw are not all contained in the total space
i
of a same subbundle OP1 (−2) → E
Consider the following affine chart (R, S, T ) ∈ C3 of Bun 15 (X/ι), which we
shall call the canonical chart : Recall that E = OP1 (−1) ⊕ OP1 (−2). Let σ1 be
a meromorphic section of some line subbundle OP1 (−2) → E with only one
(double) pole over x = ∞. Let σ−1 be a meromorphic section of the unique
line subbundle OP1 (−1) ⊂ E with only one pole over x = ∞. In the total
space of E restricted to P1 \ {∞}, we consider coordinates (x, ( zz12 )) given by
(x, z1 σ−1 + z2 σ1 ). To (R, S, T ) ∈ C3 we then associate the following normalized
quasi-parabolic structure on E:
x= 0 x= 1 x = r x= s = t x = ∞
x
0 1 R S T . (3.1)
OP1 (−1)
1 1 1 1 1
Here the first line indicates the Weierstrass point wi we are considering,
whereas the second line defines a generator of the corresponding quasi-parabolic
direction. Note that (3.1) already defines a universal quasi-parabolic bundle
∗
W ◦π is well-defined and algebraic
over the canonical chart. The lifting map elm+
and provides a universal rank 2 vector bundle with trivial determinant bundle
over the canonical chart of Bun 15 (X/ι).
Proposition 3.3 The Narasimhan-Ramanan classifying map C3(R,S,T )

MN R is explicitely given by (R, S, T ) → (v0 : v1 : v2 : v3 ) where
v0 = s2 t2 (r2 − 1)(s − t)R − r2 t2 (s2 − 1)(r − t)S + s2 r2 (t2 − 1)(r − s)T +
+t2 (t − 1)(r2 − s2 )RS − s2 (s − 1)(r2 − t2 )RT + r2 (r − 1)(s2 − t2 )ST
v1 = rst [((r − 1)(s − t)R − (s − 1)(r − t)S + (t − 1)(r − s)T +

+(t − 1)(r − s)RS − (s − 1)(r − t)RT + (r − 1)(s − t)ST ]
v2 = −st(r2 − 1)(s − t)R + rt(s2 − 1)(r − t)S − rs(t2 − 1)(r − s)T −

−t(t − 1)(r2 − s2 )RS + s(s − 1)(r2 − t2 )RT − r(r − 1)(s2 − t2 )ST
v3 = st(r − 1)(s − t)R − rt(s − 1)(r − t)S + sr(t − 1)(r − s)T +

+t(t − 1)(r − s)RS − s(s − 1)(r − t)RT + r(r − 1)(s − t)ST
The indeterminacy points
(R, S, T ) = (0, 0, 0), (1, 1, 1) and (r, s, t)
of this map correspond to the odd Gunning bundles E[w1 ] , E[w0 ] and E[w∞ ] re-
spectively. Conversely, a generic point (v0 : v1 : v2 : v3 ) ∈ MN R has precisely
two preimages in C3(R,S,T ) given by
r(t − 1)(v0 + rv1 − r(s + t + st)v3 )T

R=
t(r − 1)(v0 + tv1 − t(r + s + rs)v3 ) − (r − t)(v0 + v1 − (rs + st + rt)v3 )T
s(t − 1)(v0 + sv1 − s(r + t + rt)v3 )T
S= ,
t(s − 1)(v0 + tv1 − t(r + s + rs)v3 ) − (s − t)(v0 + v1 − (rs + st + rt)v3 )T
where T is any solution of aT 2 + btT + ct2 = 0 with
a = (v1 + v2 t + v3 t2 )(v0 + v1 − (rs + st + rt)v3 )

b = −(1 + t)(v0 v2 + v12 + tv1 v3 ) − 2(v0 v1 + tv0 v3 + tv1 v2 )
+(rs + st + rt)(tv1 + v2 + tv3 )v3 + (r + s + rs)(v1 + t2 v2 + t2 v3 )v3
c = (v1 + v2 + v3 )(v0 + tv1 − t(r + s + rs)v3 ).
The discriminant of this polynomial leads again to the equation (2.5) of the
Kummer surface.
By construction, Bun 15 (X/ι) is covered by affine charts similar to the canon-

ical chart, where we just permute the role of the Weierstrass points in (3.1).
The (birational) transition maps between affine charts are obvious. The Galois-
involution OP1 (−3) ⊗ elm+ W is given in the canonical chart by the birational
. . .
map (R, S, T ) → (R, S, T ), where
. = λ(R, S, T ) · (s − t) + (t − 1)S − (s − 1)T

R
−t(s − 1)S + s(t − 1)T + (s − t)ST
(r − t) + (t − 1)R − (r − 1)T
S. = λ(R, S, T ) ·
−t(r − 1)R + r(t − 1)T + (r − t)RT
(r − s) + (s − 1)R − (r − 1)S
T. = λ(R, S, T ) ·
−s(r − 1)R + r(s − 1)S + (r − s)RS
t(r − s)RS − s(r − t)RT + r(s − t)ST

and λ(R, S, T ) = .
(s − t)R − (r − t)S + (r − s)T
For the Galois-involution to be everywhere well defined, we need to consider
the smooth (non separated) scheme Bun(X/ι) obtained by canonically glu-
ing Bun 15 (X/ι) and Bun 45 (X/ι). From an exhaustive case-by case study, one
can show that Bun(X) corresponds precisely to the isomorphism classes of
the Galois-involution on Bun(X/ι). In terms of parabolic bundles, the Galois
∗
involution is given by elm+W ◦π . In other words,
∗ 2:1
W ◦π : Bun(X/ι) −→ Bun(X) is an algebraic
Proposition 3.4 The map elm+
2 cover.
Moreover, the lift of the Kummer surface in MN R defines a dual Weddle

surface in Bun 15 (X/ι) P3 ⊂ Bun(X/ι) which is given with respect to the
canonical chart by the equation
Wed(X) :
0 = ((s − t)R + (t − r)S + (r − s)T )RST + t((r − 1)S − (s − 1)R)RS
+r((s − 1)T − (t − 1)S)ST + s((t − 1)R − (r − 1)T )RT
−t(r − s)RS − r(s − t)ST − s(t − r)RT.
4. Application to Higgs Bundles

Let E again be a rank 2 vector bundle over X. By definition, the moduli space of
tracefree Higgs fields on E is given by H0 (X, sl(E) ⊗ Ω1X ), where sl(E) denotes
the vector bundle of trace-free endomorphisms of E. By Serre duality, we have
H0 (X, sl(E) ⊗ Ω1X ) H1 (X, sl(E)∨ ⊗ Ω1X )∨ .
If det(E) = OX , then sl(E)∨ = sl(E). If E is stable, then sl(E) possesses no

non-trivial global sections and then H1 (X, sl(E)∨ ⊗ Ω1X ) canonically identifies
with Higgs(X) := T ∨ MN R . We can calculate explicitly the Hitchin map
& '
Higgs(X) → H0 (X, Ω1X ⊗ Ω1X )
Hitch :
(E, θ) → det(θ)
from the following idea : The complement in Bun(X) of the image of the Weddle
surface is embedded into M \ Kum(X) (we obtain all stable bundles except
those on the odd Gunning planes). Since we have a universal vector bundle in
each affine chart of the two-cover Bun(X/ι) of Bun(X), we can expect to find
a universal family of Higgs bundles there as well. Then we calculate a Hitchin
map for Bun(X/ι) and push it down to MN R .
More precisely, we will calculate the Hitchin map in the following steps :
• Provided that H0 (P1 , sl(E, p)) = {0}, we have a canonical isomorphism
T(E,p) Bun(X/ι) = H1 (P1 , sl(E, p)),
where sl(E, p)) denotes trace free endomorphisms of E leaving p invari-

ant. We work out how the hyperelliptic descent φ : elm+ ◦ π ∗ defines an
algebraic 2-cover
φ
Higgs(X/ι) := T∨ Bun(X/ι) −→ T∨ Bun(X)
• The Liouville form on Bun(X/ι) is given with respect to coordinates

(R, S, T ) of the canonical chart by dR + dS + dT. We work out Serre
duality for the generators
∂ ∂ ∂
, , ∈ T Bun(X/ι)
∂R ∂S ∂T
and deduce an explicit universal Higgs bundle on an affine chart of
Higgs(X/ι).
• We calculate the determinant map
Higgs(X/ι) −→ H0 (P1 , Ω1P1 ⊗ Ω1P1 (W )) H0 (X, Ω1X ⊗ Ω1X )
and show that it factors through the Hitchin map. Then we deduce the
explicit Hitchin map from the formulas in Proposition 3.3.
4.1. Hyperelliptic Descent, Again. One can show that if E is a

stable rank 2 vector bundle with trivial determinant on X and h is a lift of the
hyperlliptic involution as in Proposition 3.1, then any trace free Higgs field θ
on E is h-equivariant, that is, the following diagram commutes :
E
θ / E ⊗ Ω1
O X
h ι∗ h
ι∗ θ/ ∗
ι∗ E ι E ⊗ Ω1X .
Hyperelliptic decent of the pair (E, θ) then produces two triples (E, θ, p), where
θ : E → E ⊗ Ω1P1 (W )
is a logarithmic Higgs field with at most apparent singularities over W : the

residue at any wi ∈ W is either zero or conjugated to

0 0
Reswi (θ) ∼
1 0
such that the quasi-parabolic pw corresponds precisely to the 0-eigendirection

i
of the residue. Conversely, consider a logarithmic Higgs field
θ. ∈ H0 (P1 , sl(E, p
.) ⊗ Ω1P1 (W ))
that lies in the image of the canonical embedding
H0 (P1 , sl(E, p) ⊗ Ω1P1 ) → H0 (P1 , sl(E, p) ⊗ Ω1P1 (W )).

In other words, θ. has only trivial residues:

. 0 0
Reswi (θ) ∼ .
0 0
Then the logarithmic Higgs field θ obtained from .

θ by applying the meromor-
phic gauge transformation
OP1 (−3) ⊗ elm+

W
has (at most) apparent singularities over each Weierstrass point, and the quasi-
parabolic structure p obtained from p . corresponds to 0-eigendirections of θ.
Note that here we have to choose a meromorphic section σ : P1 → OP1 (−3).
As long as we are not on the Kummer, respectively Weddle surface, we obtain
an isomorphism
T∨ 1 1 .))∨
p) Bun(X/ι) = H (P , sl(E, p
Serre / H0 (P1 , sl(E, p
.) ⊗ Ω1P1 )
(E, ∼
∼ OP1 (−3)⊗elm+
W

∼ H0 (P1 , sl(E , p) ⊗ Ω1P1 (W ))apparent
∼ elm+
W ◦π
∗

∨ o ∼
T∨
E Bun(X) = H (X, sl(E))
1
H0 (X, sl(E) ⊗ Ω1X )
Serre
and deduce an algebraic 2-cover

2:1
Higgs(X/ι) −→ Higgs(X).
4.2. Universal Higgs Bundles. Serre duality gives us a perfect pair-

ing

H1 (P1 , sl(E, p)) × H0 (P1 , sl2 (E , p) ⊗ Ω1P1 (W ))apparent → C
·, · : .
(φ, θ) → Res(trace(φ · θ))
Let (E, p) be an element of Bun(X/ι) given with respect to the canonical chart
by (R0 , S0 , T0 ) ∈ C3 . The vector field ∂R
∂
∈ T(R0 ,S0 ,T0 ) Bun(X/ι) is given in
H (P , sl(E, p)) by the cocycle
1 1

0 1
φ01 :=
0 0
with respect to trivialization charts U0 × C2 with U0 := P1 \ {r} U1 × C

1 ζand
2
with U1 := Dε (r) of sl(E). Indeed, if we consider exp(ζφ) = 0 1 as applied

from the left over U0 ∩ U1 ⊂ U1 to U0 ∩ U1 ⊂ U0 for a quasi-parabolic vector

bundle (E, p) with parabolic structure normalized as in (3.1), we obtain the
quasi parabolic structure corresponding to (R + ζ, S, T ). The dual basis in
H0 (P1 , sl2 (E , p) ⊗ Ω1P1 (W ))apparent with respect to ·, · of the basis

∂ ∂ ∂
, ,
∂R ∂S ∂T
of T(R0 ,S0 ,T0 ) Bun(X/ι) then is given by (θr , θs , θt ) with

0 0 dx R −R dx −R R2 dx
θr := + +
1−R 0 x R −R x−1 −1 R x−r

0 0 dx S −S dx −S S 2 dx
θs := + +
1−S 0 x S −S x−1 −1 S x−s

0 0 dx T −T dx −T T 2 dx
θt := + + .
1−T 0 x T −T x−1 −1 T x−t
We obtain the universal Higgs bundle .

θ = OP1 (−3) ⊗ elm+
W (θ) defined by
θ = c r θ r + cs θ s + ct θ t (4.1)
on the canonical chart (R, S, T, cr , cs , ct ) ∈ C6 of T∨ Bun(X/ι). Recall that all

other charts are obtained up to Galois involution by permuting the role of the
Weierstrass points.
Corollary 4.1 The Liouville form on Bun(X/ι), given with respect to the
canonical chart (R, S, T ) ∈ C3 by
dR + dS + dT
defines a holomorphic symplectic 2-form on Higgs(X/ι) given with respect to

the canonical chart (R, S, T, cr , cs , ct ) ∈ C6 of T∨ Bun(X/ι) by
dR ∧ dcr + dS ∧ dcs + dT ∧ dct .
4.3. The Hitchin Fibration. The determinant map

0
C6 → H0 (P1 , Ω1P1 ⊗ Ω1P1 (−6))
(dx)⊗2 ,
→
det(θ) = (h2 x2 + h1 x + h0 ) x(x−1)(x−r)(x−s)(x−t)
(R, S, T, cr , cs , ct )
(4.2)
where θ is the Higgs bundle in (4.1) is given by
h0 = (cr (R − 1) + cs (S − 1) + ct (T − 1)) (cr st(R − r)R + cs rt(S − s)S

+ct rs(T − t)T )
h1 = +cr (cr (s + t)(r + 1) + cs s(t + 1) + ct t(s + 1)) R2 − cr 2 (t + s) R3

+cs (cs (r + t)(s + 1) + cr r(t + 1) + ct t(r + 1)) S 2 − cs 2 (t + r ) S 3
+ct (ct (r + s)(t + 1) + cr r(s + 1) + cs s(r + 1)) T 2 − ct 2 (r + s) T 3
−cr cs (t(R − 1 + S − 1) + r(S − s) + s(R − r))RS
−cr ct (s(R − 1 + T − 1) + r(T − t) + t(R − r))RT
−cs ct (r(S − 1 + T − 1) + s(T − t) + t(S − s))ST
− (ct t(r + s) + cr r (s + t) + cs s(r + t)) (cr R + cs S + ct T )
h2 = (cr (R − 1)R + cs (S − 1)S + ct (T − 1)T ) (cr (R − r) + cs (S − s)

+ct (T − t))
Table 1: Explicit Hitchin Hamiltonians for the canonical coordinates (R, S, T )

on Bun(X/ι)
It is easy to check that the Hitchin Hamiltonians h0 , h1 , h2 do Poisson-

commute as expected : for any f, g ∈ {h0 , h1 , h2 }, we have
∂f ∂g ∂f ∂g
− =0
i=r,s,t
∂p i ∂qi ∂q i ∂pi
in Darboux notation (pr , ps , pt , qr , qs , qt ) := (R, S, T, cr , cs , ct ).

Since the determinant is invariant under meromorphic gauge transforma-
tions (and in particular elementary transformations), we can immediately de-
duce the Hitchin map

T∨ Bun(X/ι) → H0 (P1 , Ω1P1 ⊗ Ω1P1 )
Hitch : = (h2 x2 + h1 x + h0 ) 2 (dx)⊗2 ,
(R, S, T, cr , cs , ct ) → det(θ) σ ·x(x−1)(x−r)(x−s)(x−t)
where σ is our previously chosen meromorphic section σ : P1 → OP1 (−3) and

θ. is the universal Higgs bundle in the canonical chart of T∨ Bun(X/ι).
More importantly, again since the determinant map does is invariant un-
der meromorphic Gauge transformations, the map in (4.2) factors through the
Hitchin map Higgs(X) T ∨ Bun(X) → H0 (X, Ω1X ⊗ Ω1X ) by construction.
Consider the natural rational map φ∗ : T∨ MN R T∨ Bun(X/ι) induced by
the map φ : Bun(X/ι) MN R stated explicitly with respect to the canonical
chart in Proposition 3.3. The general
section
crdR+ cs dS + ct dT then lifts
to a general section μ0 d vv03 + μ1 d vv13 + μ2 d vv23 . Moreover, from the ex-
plicit coordinate change (v0 : v1 : v2 : v3 ) ↔ (u0 : u1 : u2 : u3 ) to the nice
coordinates, given in (2.7), we know how to identify general sections

u0 u1 u2 v0 v1 v2
η0 d + η1 d + η2 d = μ0 d + μ1 d + μ2 d .
u3 u3 u3 v3 v3 v3
The Hamiltonians h0 , h1 , h2 of the Hitchin map on MN R then can be explicitly
deduced from (4.2). We get

T∨ MN R → H0 (X, Ω1X ⊗ Ω1X )
Hitch : (dx)⊗2 ,
((u0 : u1 : u2 : u3 ), η0 , η1 , η2 ) → (h2 x2 + h1 x + h0 ) x(x−1)(x−r)(x−s)(x−t)
where
⎧ 2
⎪
⎪ rst· η0 (u20 − u23 ) + η1 (u0 u1 + u2 u3 ) + η2 (u0 u2 + u1 u3 )
⎪
⎪ 2
⎪
⎨
1
−st· η0 (u0 u1 − u2 u3 ) + η1 (u21 + u23 ) + η2 (u0 u3 + u1 u2 )
h0 = 4u4
·
3 ⎪
⎪ +4rs· (η0 u0 + η1 u1 )2 u23
⎪
⎪
⎪
⎩ 2
−rt· η0 (u20 + u23 ) + η1 (u0 u1 + u2 u3 ) + η2 (u0 u2 − u1 u3 )
⎧
2
⎪
⎪ t· u0 + u21 + u22 + u23 (η02 + η12 + η22 )u23 + (η0 u0 + η1 u1 + η2 u2 )2
⎪
⎪
⎪
⎪
2
⎪
⎪ +st· u0 − u21 + u22 − u23 (η02 − η12 + η22 )u23 − (η0 u0 + η1 u1 + η2 u2 )2
⎪
⎪
⎪
⎨ +4r·
1
(u0 u2 − u1 u3 ) u3 [η0 η2 u3 + (η0 u0 + η1 u1 + η2 u2 )η1 ]
h1 = 4u4
·
3 ⎪
⎪ +4sr· (u0 u2 + u1 u3 ) u3 [η0 η2 u3 − (η0 u0 + η1 u1 + η2 u2 )η1 ]
⎪
⎪
⎪
⎪
⎪
⎪ +4s· (u0 u3 + u1 u2 ) u3 [η1 η2 u3 − (η0 u0 + η1 u1 + η2 u2 )η0 ]
⎪
⎪
⎪
⎩ +4rt· (u u + u u ) u [η η u − (η u + η u + η u )η ]
0 1 2 3 3 0 1 3 0 0 1 1 2 2 2
⎧ 2
⎪
⎪ s· η0 (u0 u2 + u1 u3 ) + η1 (u0 u3 + u1 u2 ) + η2 (u22 − u23 )
⎪
⎪
⎨ −1· η0 (u0 u2 − u1 u3 ) + η1 (u0 u3 + u1 u2 ) + η2 (u2 + u2 ) 2
⎪
1 2 3
h2 = 4u4
· 2
3 ⎪
⎪ −t· η0 (u0 u1 + u3 u3 ) − η2 (u0 u3 − u1 u2 ) + η1 (u2 + u23 )
2
⎪
⎪
⎪
⎩
+4r· (η1 u1 + η2 u2 )2 u23
Table 2: Explicit Hitchin Hamiltonians for the coordinates (u0 : u1 : u2 : u3 ) of

MN R .
Note that in [7], B. van Geemen and E. Previato conjectured a projective

version of explicit Hitchin Hamiltonians, which has been confirmed in [6]. These
Hamiltonians H1 , . . . H6 can be seen as evaluations, up to functions in the base,
of the explicit Hitchin map at the Weierstrass points. More precisely, if we
denote
h(x) := h2 x2 + h1 x + h0 ,
where hi for i ∈ {0, 1, 2} then

4h(0) 4h(s)
H1 = rst H4 = s(s−1)(s−r)(s−t)
4h(t) 4h(r)
H2 = − t(t−1)(t−r)(t−s) H5 = r(r−1)(r−s)(r−t)
4h(1)
H3 = (r−1)(s−1)(t−1) H6 = 0.
References
[1] M. F. Atiyah, Complex analytic connections in fibre bundles, Trans. Amer. Math.
Soc. 85 (1957) 181–207.
[2] I. Biswas, V. Heu On the logarithmic connections over curves, J. of the Ramanu-
jan Math. Soc., Volume 28A, Issue SPL, July - Special Issue (2013), 21–40.
[3] I. Biswas, Parabolic bundles as orbifold bundles, Duke Math. J. 88 (1997), no.
2, 305–325.
[4] M. Bolognesi, A conic bundle degenerating on the Kummer surface, Math. Z.
261 (2009), no. 1, 149–168.
[5] I. Dolgachev, Classical Algebraic Geometry: a modern point of view, Cambridge
University Press, Cambridge, 2012.
[6] K. Gawȩdzki and P. Tran-Ngoc-Bich, Self-Duality of the SL2 Hitchin Integrable
System at Genus 2 Communications in Mathematical Physics 196 (1998), no. 3,
641–670.
[7] B. van Geemen and E. Previato, On the Hitchin system Duke Mathematical
Journal 85 (1996), no. 3, 659-683.
[8] W. M. Goldman, Ergodic theory on moduli spaces, Ann. of Math. (2) 146 (1997)
475–507.
[9] R.C. Gunning, Lectures on vector bundles over Riemann surfaces, University
of Tokyo Press, Tokyo, Princeton University Press, Princeton, ISBN 0691079986
(1967), v+243 pp.
[10] R. C. Gunning, Special coordinate coverings of Riemann surfaces, Math. Ann.
170 (1967) 67–86.
[11] R. C. Gunning : Analytic structures on the space of flat vector bundles over a
compact Riemann surface, Princeton University, Princeton, New Jersey 08540
[12] V. Heu and F. Loray, Flat rank 2 vector bundles over genus 2 curves, to appear
in Memoirs of the AMS (2017). arXiv 1401.2449.
[13] N. Hitchin, Stable bundles and integrable systems, Duke Math. J. 54 (1987), no.
1, 91–114.
[14] J. C. Hurtubise, Integrable systems and algebraic surfaces. Duke Math. J. 83
(1996), no. 1, 19–50.
[15] F. Loray and M.-H. Saito, Lagrangian fibration in duality on moduli space of
rank two logarithmic connections over the projective line, Int. Math. Res. Not.
IMRN 2015, no. 4, 995–1043.
[16] M.S. Narasimhan and S. Ramanan, Moduli of vector bundles on a compact
Riemann surface, The Annals of Mathematics (2) 89 (1969) 14–51.
[17] A. Weil, Généralisation des fonctions abéliennes, Jour. Math. Pure Appl., 17
(1938) 47–87.
Smoothness of Moduli Space of Stable
Torsion-free Sheaves with Fixed
Determinant in Mixed Characteristic
Inder Kaur
Abstract
Let R be a complete discrete valuation ring with fraction field of characteristic 0
and algebraically closed residue field of characteristic p > 0. Let XR → Spec(R)
be a smooth projective morphism of relative dimension 1. We prove that, given
a line bundle LR the moduli space of Gieseker stable torsion-free sheaves of
rank r ≥ 2 over XR , with determinant LR , is smooth over R.
1. Introduction
Notation 1.1. Let R be a complete discrete valuation ring with maximal ideal
m. Denote by K its fraction field of characteristic 0 and by k its residue field
of characteristic p > 0. Assume k is algebraically closed. Let XR → Spec(R)
be a smooth fibred surface and Xk its special fibre. Fix a line bundle LR on
XR . Let P be a fixed Hilbert polynomial. Throughout this note, semistability
always refers to Gieseker semistability (see [6, Definition 1.2.4]).
In [8, Theorem 0.2], Langer proves that the moduli functor of
semi(stable)torsion-free sheaves with fixed Hilbert polynomial P on XR is
uniformly (universally) corepresented by an R-scheme MXR (P ) (respectively
s
MX R
(P )). Recall the definition of the moduli functor of flat families of
(semi)stable torsion free sheaves with fixed Hilbert poynomial P and deter-
minant LR on XR (see Definition 2.2). We denote this functor by MsXR ,LR . In
this note we prove the following:
Theorem 1.2 (see Proposition 2.3, Remark 2.4 and Theorem 4.5). We have
the following:
1. The moduli functor MXR ,LR is uniformly corepresented by a projective
R-scheme of finite type denoted MR,LR . The open subfunctor MsXR ,LR for
Geometry, DOI 10.1007/978-981-10-5648-2_11
174 Inder Kaur
stable sheaves is universally corepresented by a R-scheme of finite type,

s
denoted MR,L R
.
s
2. The morphism MR,LR
→ Spec(R) is smooth.
Part 1 is proven analogously to [2, Theorem 3.1]. For part 2, we prove that the
s
deformation functor at a point in the moduli space MR,L R
is unobstructed (see
Theorem 3.19).
Note that Theorem 1.2 is proven by Langer in the case when R is a k-algebra
(see [7, Proposition 3.4]). However, the proof does not generalize to our setup.
This is because it relies on [1, Proposition 1], the proof of which does not hold
in mixed characteristic. The main difficulty is that even in the case of vector
bundles it uses the structure of R as a k-algebra in a fundamental way (see [1,
Section 3]). We use the same philosophy as [1, Proposition 1] (of using Cech
cohomology) but take a more direct approach since we are working on a family
of curves.
The setup is as follows: in §2 we recall the basic definitions and results
needed for this note. We also prove the existence of the moduli space of stable
torsion free sheaves with fixed determinant over Spec(R). In §3 we show that
s
the deformation functor at a point in the moduli space MR,L R
is unobstructed.
Finally in §4 we prove that this moduli space is smooth over Spec(R).
2. Basic Definitions and Results

Keep Notations 1.1
In this section we define the moduli functor of (semi)stable sheaves with fixed
determinant. We prove that it is uniformly corepresented by an R-scheme of
finite type.
Definition 2.1. Let XR → Spec(R) be as in Notation 1.1.
1. Let MXR / Spec(R) (P ) (as in [2, Theorem 3.1]) of pure Gieseker semistable
sheaves. For simplicity we will denote this functor by MR and the cor-
responding moduli space by MR . Denote by PicXR the moduli functor
for line bundles. By assumption XR → Spec(R) is flat, projective with
integral fibres, therefore by [3, Theorem 9.4.8] the functor PicXR is rep-
resentable. We denote this moduli space by Pic(XR ).
2. By assumption XR is smooth over R. By [6, Theorem 2.1.10], every co-

herent sheaf E on XR admits a locally free resolution
0 → En → En−1 → · · · → E0 → E → 0.
Smoothness of Moduli Space of Stable Torsion-free Sheaves 175
i
Then det(E) := ⊗det(Ei )(−1) .
Therefore we can define a natural transformation det : MR → PicXR .
This induces a morphism between the schemes corepresenting these func-
tors MR → Pic(XR ).
Now we define the moduli functor for families of pure Gieseker semistable
sheaves with fixed determinant.
Definition 2.2. Let XR → Spec(R) be a smooth, projective morphism and LR

a line bundle on XR . For P a fixed Hilbert polynomial, we define the moduli
functor MXR ,LR (P ), denoted MR,LR for simplicity, on XR of sheaves with
fixed determinant LR . Let MXR ,LR : (Sch/R)◦ → (Sets) be such that for an
R-scheme T ,
⎧ ⎫
⎨S- equivalence classes of families of pure Gieseker ⎬
MXR ,LR (T ) := semistable sheaves F on XT with the property that /∼
⎩ ∗ ∗ ⎭
det(F) πX R
LR ⊗ π T Q, for some line bundle Q on T
where πXR : XT → XR and πT : XT → T are the natural projection maps and

F ∼ F , if and only if there exists a line bundle L on T , such that F F ⊗πT∗ L.
We denote by MsXR ,LR the subfunctor for the stable sheaves.
s
We note that the moduli space MR,LR
is a projective R-scheme.
Proposition 2.3. The functor MsR,LR is universally corepresented by a R-

s
scheme of finite type. We denote this scheme by MR,LR
.
Proof. We know from the proof of [2, Theorem 3.1], there exists a subset of the
Quot scheme denoted Rs , such that MRs is a universal categorical quotient of
this subset by the action of a certain general linear group. Let α : Rs → MRs
denote this quotient.
The natural transformation MsR → PicXR which induces the determinant
morphism det : MRs → Pic(XR ). By composing the morphism det with α we
obtain, a morphism detRs : Rs → MRs → Pic(XR ).
Let RsLR := det−1 Rs (LR ) denote the fibre of the map detRs at the point cor-
responding to LR and let NR,LR := det−1 (LR ). Let MR,L s
R
be a universal
s
categorical quotient of RLR by GL(V ). By definition of categorical quotient,
there exists a unique morphism from φsLR : MR,L s
R
→ NR,LR . Since the quo-
s s s s
tients R → MR and RLR → MR,LR are PGL(V )-bundles in the fppf topology
(see [9, Lemma 6.3]), it implies φsLR is an isomorphism. Therefore, we have the
following diagram,
176 Inder Kaur
s
MR,L NR,LR - Ms
R R
det
? [LR-] ?
Spec(R) Pic(XR )
Finally by [6, Theorem 4.3.1] we conclude that the functor MsR,LR is universally
s
corepresented by the R-scheme MR,L R
.
Remark 2.4. Note that the functor MR,LR is corepresented by a projective

R-scheme, denoted MR,LR of finite type. Recall the proof of [2, Theorem 3.1].
Since XR is smooth, using [6, Theorem 2.1.10], we can define a morphism det :
QuotXR (H, P ) → Pic(XR ) mapping a coherent sheaf on XR to its determinant

bundle. Denote by A the (scheme-theoretic) intersection of det −1 (LR ) and Q,
where Q as in the proof of [2, Theorem 3.1]. Then the statement follows after
replacing Q by A in the proof of [2, Theorem 3.1].
3. Deformation of Moduli Spaces with Fixed

Determinant
s
Keep Notations 1.1. We have seen in the proof of Proposition 2.3 how MR,L R
s
can be considered as the fiber of the determinant morphism det : MR →
Pic(XR ) over the point corresponding to LR . Using the trace map (see Def-
inition 3.13), we relate the obstruction theory of the deformation functor at
a point in the moduli space MRs to the obstruction theory of the deformation
functor at a point in the moduli space Pic(XR ). We use this (see Theorem 3.19)
s
to show that the deformation functor at a point in the moduli space MR,L R
is
unobstructed.
We begin by recalling some basic definitions.
Notation 3.1. We denote by Art/R the category of local artinian R-algebras

with residue field k. Denote by Xk := XR ×Spec(R) Spec(k) and XA :=
XR ×Spec(R) Spec(A). Let [Fk ] denote a closed point of MRs . As MRs → Spec(R)
is a morphism of finite type, the closed points of the moduli space MRs are k-
points. Since k is algebraically closed, by [2, Theorem 3.1] we have a bijection
θ(k) : MR (k) → HomR (k, MR ).
Therefore to a closed point of MRs say [Fk ], we can associate a Gieseker

stable sheaf Fk on the curve Xk . Since the curve Xk is smooth, the torsion-free
sheaf is in fact locally free.
We define a covariant functor at the point [Fk ] in MRs .

Definition 3.2. We define the deformation functor D[Fk ] : Art/R → (Sets),
such that for A ∈ Art/R
⎧ ⎫
⎨ coherent sheaves FA with Hilbert polynomial P ⎬
D[Fk ] (A) := on XA flat overA such that its pull-back to Xk
⎩ ⎭
is isomorphic to Fk .
Similarly, we define a covariant functor at the point [det(Fk )] of the moduli
space Pic(XR ).
Definition 3.3. Let D[det(Fk )] : Art/R → (Sets) be a covariant functor such
that for A ∈ Art/R
⎧ ⎫
⎨ coherent sheaves FA with Hilbert polynomial the ⎬
D[det(Fk )] (A) := same as det(Fk ) on XA flat overA such that its
⎩ ⎭
pull-back to Xk is isomorphic to det(Fk ).
The following theorem gives the obstruction theories of D[Fk ] and D[det(Fk )] .
Using this we prove the following corollary.
Remark 3.4. By [5, Theorem 7.3]] the functors D[Fk ] and D[det(Fk )]
have obstruction theories in the groups H 2 (HomXk (Fk , Fk ) ⊗k I)
and H 2 (HomXk (det(Fk ), det(Fk )) ⊗k I) respectively. For Xk a curve,
by Grothendieck vanishing theorem, H 2 (HomXk (Fk , Fk ) ⊗k I) and
H 2 (HomXk (det(Fk ), det(Fk )) ⊗k I) vanish. Therefore, D[Fk ] and D[det(Fk )] are
unobstructed.
Now we define a natural transformation between the two deformation functors.
Definition 3.5. By assumption Fk is a locally-free OXR module. Moreover, by
[5, Exercise 7.1] any coherent sheaf FA on XA satisfying the property FA ⊗OXA
OXk Fk is a locally free OXA -module. Therefore, the notion of determinant
is well-defined for any coherent sheaf on XA which pulls back to Fk .
We define a natural transformation Det : D[Fk ] → D[det(Fk )] such that for
A ∈ Art/R,
DetA : D[Fk ] (A) → D[det(Fk )] (A), EA → det(EA ).
Using this we define a deformation functor at a point in the moduli space
s
MR,L R
.
Definition 3.6. Let LR be as in Notation 1.1. For A a R-algebra, denote by
LA the pullback p∗A LR under the natural morphism pA : XA → XR .
We define a functor D[Fk ],[detFk ] : Art/R → (Sets), such that for A ∈ Art/R.
D[Fk ],[detFk ] (A) := Det−1
A (LA ).
178 Inder Kaur
3.7. Group action on the torsors: By [5, Theorem 7.3], the set D[Fk ] (A ) (respec-
tively D[det(Fk )] (A )) is a torsor under the action of H 1 (HomXk (Fk , Fk ) ⊗k I)
(respectively H 1 (HomXk (det(Fk ), det(Fk )) ⊗k I)).
Since Xk is noetherian, we can identify the sheaf cohomology
H 1 (Xk , Hom(Fk , Fk ) ⊗k I) with the Cech cohomology Ȟ 1 (U, Hom(Fk , Fk ) ⊗k
I), where U is an affine open covering of Xk . Then an element, say ξ of
the cohomology group H 1 (Hom(Fk , Fk ) ⊗k I) can be seen as a collection
of elements {φij } ∈ Γ(Ui ∩ Uj , Hom(Fk , Fk )) satisfying the cocycle condi-
tion i.e. for any i, j, k, we have φik |Uijk = φjk |Uijk + φij |Uijk . Since I is a
k-vector space, Ȟ 1 (U, Hom(Fk , Fk ) ⊗k I) Ȟ 1 (U, Hom(Fk , Fk )) ⊗k I. There-
fore, {φij }i,j is of the form {φij ⊗ a}i,j for a ∈ I not depending on i, j and
φij ∈ Γ(Ui ∩ Uj , Hom(Fk , Fk )) satisfying φik |Uijk = φjk |Uijk + φij |Uijk .
Let FA be an extension of FA on XA i.e an element of D[Fk ] (A ). Since
it is locally free, there exists a covering U = {Ui } of XA by such that FA |Ui
is OUi -free. Denote by U := {Ui } the cover of Xk where Ui := Ui ∩ Xk . We
know from the proof of [5, Theorem 7.3] that FA (ξ) is given by a collection of
sheaves Fi := FA |Ui and isomorphisms φij : Fi |Ui ∩Uj → Fj |Ui ∩Uj such that
Id+((φ
ij ⊗a)◦π)
φii = Id, φij : Fi |Ui ∩Uj = FA |Ui ∩Uj −−−−−−−−−−→ FA |Ui ∩Uj = Fj |Ui ∩Uj
where φij , a are as above and π is the natural restriction morphism FA → Fk .
Then by [4, Ex. II.1.22], FA (ξ) glues to a sheaf if the morphisms {φij } satisfy
the cocycle condition. In the following lemma we prove that this is indeed the
case.
Lemma 3.8. Let Fi and φij be as above. The morphisms {φij } satisfy the
cocycle condition i.e. for any i, j, k φik = φjk ◦ φij .
Proof. It suffices to prove this equality for the basis elements, say si1 , . . . , sir
generating Fi |Ui ∩Uj ∩Uk . For any basis element sit ,
φjk ◦ φij (sit ) = φjk (Id + (a ⊗ φij ))(π(sit ))

= φjk (π(sit ) + aφij (π(sit ))) = (Id + a ⊗ φjk )(π(sit )) + aφij (π(sit ))
= π(sit ) + aφij (π(sit )) + aφjk (π(sit )) + 0
because a2 = 0 in A . Since φik = φij + φjk , we have
φjk ◦ φij (sit ) = π(sit ) + a(φik (π(sit )) = φik (sit ).
This shows that {φij }i,j satisfy the cocycle condition.

Using this we conclude that FA (ξ), obtained by glueing the sheaves Fi along
the isomorphism φij is a sheaf.
Similarly, an element say ξ in H 1 (HomXk (det(Fk ), det(Fk )) ⊗k I) acts on
an element in D[det(Fk )] (A ), say det(FA ) to produce a line bundle det(FA )(ξ )
given by a family of sheaves {Li := LA |Ui } and isomorphisms
Id+((φ
ij ⊗a)◦π)
φij : Li |Uij −−−−−−−−−−→ Lj |Uij
where φij ∈ Γ(Ui ∩ Uj , Hom(det(Fk ), det(Fk )) ⊗k I) is the col-

lection of sections corresponding to ξ given by the ismomorphism
H 1 (HomXk (det(Fk ), det(Fk )) ⊗k I) Ȟ 1 (U, Hom(det(Fk ), det(Fk ))) ⊗k I.
Again by Lemma 3.8, det(FA )(ξ ) is a sheaf.
Definition 3.9. We have the following definitions.
1. We define a map
φ1 : H 1 (HomXk (Fk , Fk ) ⊗k I) → D[Fk ] (A ), ξ → FA (ξ)
which uniquely associates an extension FA (ξ) of FA (using Lemma 3.8)
to an element ξ of H 1 (HomXk (Fk , Fk ) ⊗k I).
2. Replacing FA by det(FA ) and starting with det(FA ) we as-

sociate an extension say (det(FA ))(ξ ) to an element ξ of
H 1 (HomXk (det(Fk ), det(Fk )) ⊗k I). Hence we define a map
φ2 : H 1 (HomXk (det(Fk ), det(Fk ))⊗k I) → Ddet(Fk ) (A ), ξ → det(FA

)(ξ )
Remark 3.10. Note that by Corollary 3.4, there exist surjective morphisms
r1 : DFk (A ) DFk (A) and r2 : Ddet(Fk ) (A ) Ddet(Fk ) (A). By [5, Theorem
7.3], r1−1 (FA ) = Im(φ1 ), r2−1 (det(FA )) = Im(φ2 ).
The following lemma tells us that taking the determinant commutes with
glueing of the sheaf.
Lemma 3.11. The determinant of the sheaf FA (ξ) is the line bundle obtained
by glueing {det(Fi )} along the isomorphisms
φij : det(Fi )|Ui ∩Uj → det(Fj )|Ui ∩Uj , s1 ∧ ... ∧ s(i) i i

(i)
r → φij (s1 ) ∧ ... ∧ φij (sr )
where si1 , ..., sir are the basis elements of Fi |Ui ∩Uj .
180 Inder Kaur
Proof. By Lemma 3.8 for all t = 1, . . . , r, we have φik (sit ) = φjk (sit ) ◦ φij (sit ).
Then,
φjk ◦ φij (si1 ∧ · · · ∧ sir ) = φjk (φij (si1 ) ∧ · · · ∧ φij (sir ))

= (φjk ◦ φij (si1 )) ∧ · · · ∧ (φjk ◦ φij (sir ))
= φik (s1 ) ∧ · · · ∧ φik (sir )
= φik (si1 ∧ · · · ∧ sir )
Hence the morphisms {φij } satisfy the cocycle condition i.e φik = φjk ◦ φjk .
By Lemma 3.8, there exist isomorphisms ψi : FA (ξ)|Ui Fi satisfying
ψj |Uij = φij ◦ ψi |Uij . We define ψ i : det(FA (ξ))|Ui det(Fi ) as follows. Let
si1 , . . . , sir be the basis of FA (ξ)|Ui . Then ψ i (si1 ∧· · ·∧sir ) := ψi (si1 )∧· · ·∧ψi (sir ).
Therefore
φij ◦ ψ i (si1 ∧ · · · ∧ sir ) = φij (ψi (si1 ) ∧ · · · ∧ ψi (sir ))

= φij (ψi (si1 )) ∧ · · · ∧ φij (ψi (sir ))
= ψj (si1 ) ∧ · · · ∧ ψj (sir )
= ψ j (s1 ∧ · · · ∧ sir )
Then by the uniqueness of glueing mentioned in [4, Ex. II.1.22], {det(Fi )} glues
along the isomorphisms {φij }i,j to det(FA )(ξ).
3.12. We relate the obstruction theory of D[Fk ] to that of D[det(Fk )] by relating
the action of the group H 1 (HomXk (Fk , Fk ) ⊗k I) on the vector bundle to the
action of the group H 1 (HomXk (det(Fk ), det(Fk )) ⊗k I) on the determinant of
the vector bundle. This relation is given by the trace map which we recall here.
Definition 3.13. Let U be an affine open set on which Fk is free, generated
by sections say s1 , ..., sr (for r = rk(Fk )). Recall the map,
trU : Hom(Fk , Fk )(U ) → Hom(det(Fk ), det(Fk ))(U ), (∗)

φ → trU (φ) := (s1 ∧ ... ∧ sr → s1 ∧ .. ∧ φ(sj ) ∧ ... ∧ sr ).
j
Let U := {Ui } be a small enough open cover of Xk such that Fk is free on each
Ui . Then the trace map is given by
tr : HomXk (Fk , Fk ) → HomXk (det(Fk ), det(Fk ))

such that tr|Ui = trUi for any affine open set Ui of Xk .
Remark 3.14. Note that the morphism trU is OXk linear. Let f ∈ OXk (U ).
Then

trU (f φ) = s1 ∧ ... ∧ sr → s1 ∧ .. ∧ f φ(sj ) ∧ ... ∧ sr
j

= f (s1 ∧ .. ∧ φ(sj ) ∧ ... ∧ sr )
j

= f s1 ∧ .. ∧ φ(sj ) ∧ ... ∧ sr
j
= f trU (φ).
Lemma 3.15. The morphism tr is surjective.

Proof. It suffices to prove surjectivity on the level of stalks. Let x ∈ Xk be a
closed point. Consider the induced morphism
trx : HomXk (Fk,x , Fk,x ) → HomXk (det(Fk,x ), det(Fk,x ))
and basis s1 , ..., sr ∈ Fk,x . Since the map trx is OXk ,x linear and
HomOXk (det(Fk,x ), det(Fk,x )) ∼
= OXk ,x , it suffices to show that Id ∈ Im(trx ).
Let φ ∈ HomXk (Fk,x , Fk,x ) defined as φ(si ) = si for i = 1 and 0 otherwise.
This concludes the proof.
We can define the trace map cohomologically as follows:
Definition 3.16. Let U := {Ui } be a small enough open affine cover of
Xk such that Fk is free on each Ui . Using [4, III. Theorem 4.5] we define
Čech cocycle C p (U, Hom(Fk , Fk )) (resp C p (U, Hom(det(Fk ), det(Fk ))), such
that the corresponding Čech cohomology coincides with the sheaf cohomology
H i (Xk , Hom(Fk , Fk )) (resp H i (Xk , Hom(det(Fk ), det(Fk )))). The morphism
(∗) of Definition induces a morphism on cohomologies
tri : H i (Xk , Hom(Fk , Fk )) → H i (Xk , Hom(det(Fk ), det(Fk ))) ∼

= H i (Xk , OXk ).
As a corollary to Lemma 3.15 we have:

Corollary 3.17. The morphism induced on cohomology
tr1 : H 1 (Xk , HomXk (Fk , Fk )) → H 1 (Xk , HomXk (det(Fk ), det(Fk ))
is surjective.
Proof. Consider the short exact sequence,
tr
0 → ker tr → HomXk (Fk , Fk ) −
→ HomXk (det(Fk ), det(Fk )) → 0.
182 Inder Kaur
We get the following terms in the associated long exact sequence,

tr1
... → H 1 (Xk , HomXk (Fk , Fk )) −−→ H 1 (Xk , HomXk (det(Fk ), det(Fk )))
→ H 2 (ker tr) → ...
Since Xk is a curve, by Grothendieck’s vanishing theorem, H 2 (ker(tr)) = 0.
Therefore, the morphism tr1 is surjective.
The following proposition tells us that the determinant map ’commutes’
with the trace map.
Proposition 3.18. Notation as in 3.7. Let
detij : Γ(Ui ∩ Uj , Hom(Fi , Fi )) → Γ(Ui ∩ Uj , Hom(det(Fi ), det(Fi )))
be a morphism defined by
φij ∈ Γ(Ui ∩Uj , Hom(Fi , Fi )) → detij (φij ) := (si1 ∧...∧sir → φij (si1 )∧...∧φij (sir ))
where si1 , ..., sir are the basis elements of Fi |Ui ∩Uj . Then for any pair i = j, we
have
det ◦(Id + (φij ⊗ a) ◦ π) = Id + (trUij (φij ) ⊗ a) ◦ π.
ij
In other words, the following diagram is commutative:

φ1 -
H 1 (HomXk (Fk , Fk ) ⊗k I) D[Fk ] (A )
tr1 ⊗ Id DetA
? ?
φ2
H 1 (HomXk (det(Fk ), det(Fk )) ⊗k I) - D[det(Fk )] (A )
Proof. Let si1 , ..., sir be the sections generating Fi |Ui ∩Uj .
Any section of Hom(det(Fi ), det(Fi )) is (uniquely) defined by the image of
si1 ∧ ... ∧ sir . Hence it suffices to prove
(detij ◦ (Id + (φij ⊗ a) ◦ π))(si1 ∧ ... ∧ sir ) = (Id + (trUij (φij ) ⊗ a) ◦ π)(si1 ∧ ... ∧ sir ).
For 1 ≤ t ≤ r, (Id + (φij ⊗ a) ◦ π)(sit ) = sit + aφij (π(sit )) and since I.mA =
0, at = 0 for t > 1. Hence,
(detij ◦(Id+(φij ⊗a)◦π))(si1 ∧...∧sir ) = (si1 +aφij (π(si1 )))∧...∧(sir +aφij (π(sir ))) =

= si1 ∧...∧sir +a si1 ∧...∧φij (π(sik ))∧...∧sir = (Id+(trUij (φij )⊗a)◦π)(si1 ∧...∧sir ).
k
This completes the proof of the proposition.
We end this section with the following theorem.
Theorem 3.19. The functor D[Fk ],[det(Fk )] is unobstructed.
Proof. Let A A be a small extension in Art/R and φ1 , φ2 be as in Definition

3.9. Recall the surjective morphisms r1 , r2 from Remark 3.10. Then we have
the following diagram.
ψ
D[Fk ],[det(Fk )] (A ) - D[Fk ],[det(Fk )] (A)
∩ ∩

? ?
φ1 - r - D[F ] (A)
H 1 (HomXk (Fk , Fk ) ⊗k I) D[Fk ] (A ) k
tr1 ⊗ Id DetA DetA

? ? ?
φ2 - r-
H 1 (HomXk ((Fk ), det(Fk )) ⊗k I) D(Fk ) (A ) D[det(Fk )] (A)
where the upper right square and the lower right square are commutative by
definition and the lower left square is commutative by Proposition 3.18. To
prove that D[Fk ],[det(Fk )] is unobstructed, we need to show that ψ is surjective.
Let LA be the unique pull-back of LR under the morphism XA → XR and FA
be an element in D[Fk ],[det(Fk )] (A). Since D[Fk ],[det(Fk )] (A ) = det(LA ) where
LA is π ∗ LR for π : XA → XR , we need to prove there exists a sheaf FA on
XA with determinant LA which is an extension of FA .
By definition r2 (LA ) = LA . Since φ1 and φ2 are injective,
r1−1 (FA ) = Im(φ1 ) and r2−1 (LA ) = Im(φ2 ). Therefore, there exists t ∈
H 1 (HomXk (det(Fk ), det(Fk )) ⊗k I) such that φ2 (t) = LA . By Corollary 3.17,
tr1 ⊗ Id is surjective. Hence there exists t ∈ H 1 (HomXk (Fk , Fk ) ⊗k I) such
that tr1 ⊗ Id(t ) = t. Denote by FA := φ1 (t ). By commutativity of the lower
left square, det(FA ) = LA . This concludes the proof of the theorem.
4. Main Results
In Theorem 3.19, we showed that the deformation functor D[Fk ],[det(Fk )] is un-
s
obstructed for any closed point [Fk ] of the moduli space MR,L R
. In this section
we prove that this functor is in fact prorepresented by the completion of the
local ring at the point [Fk ] (see Proposition 4.4). Using this we prove that the
s
moduli space MR,L R
of pure stable sheaves with fixed determinant LR over
XR , is smooth over Spec(R).
184 Inder Kaur
Notation 4.1. Keep Notations 1.1 and 3.1. Let [Fk ] be a k-rational point
s
of MR,L R
and denote by Λ := ÔMR,[F
s , the completion of the local ring
k]
OMRs ,[Fk ] . Under the determinant morphism det : MRs → PicXR , the line bundle
det(Fk ) is a k-point of Pic(XR ). Denote by Λ := ÔPic(XR ),[det(Fk )] and by
Λ := ÔMR,L s ,[Fk ] .
R
Definition 4.2. By ÔM s ,[Fk ] we denote the covariant functor

R
Hom(Λ , −) : Art/R → Sets, A → HomR−alg (Λ , A)

We define the functors ÔPic(XR ),[det(Fk )] and ÔM s ,[Fk ] similarly.
R,LR
Lemma 4.3. The deformation functor D[Fk ] (resp. D[Lk ] ) are pro-
representable by ÔMR,[F
s
]
(resp. ÔPicXR ,[det(Fk )] ).
k
Proof. Recall from the proof of [2, Theorem 3.1], that for m sufficiently large,
Rs is the open subset of Quot(H; P ) where H := OXR (−m)P (m) parametrizing
stable quotients. By [9, Lemma 6.3], φ : Rs → MRs is an etale PGL(V )-principal
bundle. Therefore, ÔRs ,[Fk ] ∼
= ÔMRs ,[Fk ] .
Denote by Q := Quot(H; P ) and by DQ,[Fk ] the deformation functor corre-
sponding to the Quot-scheme at the point [Fk ]. Recall that for any local Artin
ring A, Pic(Spec(A)) = 0, hence D[Fk ] = DQ,[Fk ] . Since the functor Quot is
representable, the deformation functor DQ,[Fk ] is pro-representable by ÔQ,[Fk ]
i.e.,
DQ,[Fk ] ∼
= ÔQ,[Fk ] ∼= ÔRs ,[Fk ] ,
where the second isomorphism follows from the fact that Rs is an open subset
of Q. Therefore, D[Fk ] is isomorphic to ÔM s ,[Fk ] .
R
Using the same argument we can show that D[det(Fk )] ∼

= ÔPicXR ,[det(Fk )] .
This proves the lemma.
Using this lemma we prove the following proposition.
Proposition 4.4. The deformation functor D[Fk ],[det(Fk )] is pro-represented by
the completion of the local ring OMR,L
s ,[Fk ] .
R
Proof. By Lemma 4.3, D[Fk ] (respectively D[det(Fk )] ) is pro-represented by

ÔMRs ,[Fk ] (respectively ÔPic(XR ),[det(Fk )] ). We have a natural transformation
det : ÔM s → ÔPic(XR ),[det(Fk )]

R,[Fk ]
induced by the determinant morphism, det : MRs → Pic(XR ) localized at the

point [Fk ]. Let A ∈ Art/R and LA be the pullback of the line bundle LR under
the morphism XA → XR . Recall the natural transformation DetA defined in

Definition 3.5. We have the following commutative diagram
∼ -
D[Fk ] (A) ÔM s (A)
R,[Fk ]
DetA detA
? ?
∼
D[det(Fk )] (A) - ÔPic(XR ),[det(Fk )] (A)
σ
Hence the deformation functor D[Fk ],[det(Fk )] (A) ∼ = detA −1 (φLA ), where
φLA := σ(LA ). Therefore to prove that D[Fk ],[det(Fk )] is pro-represented by
ÔM s ,[Fk ] , we need to show that for any A ∈ Art/R,
R,LR
detA −1 (φLA ) ∼
= HomR (Λ, A). (1)
∼
By Lemma 4.3, D[det(Fk )] (A) − → HomR (Λ , A). Hence for a fixed element LA ∈
D[det(Fk )] (A), the corresponding morphism from Spec(A) → Spec(Λ ) is unique
and this is the morphism φLA . This implies the commutativity of the following
diagram
Spec(A) - Spec(Λ )

? ?
Spec(R) - Spec(Λ )
where the morphism Spec(R) → Spec(Λ ) is the morphism corresponding to

the line bundle LR . Then the bijection in (1) follows from the property of fibre
product and the following diagram.
Spec(A)
-
Spec(Λ) - Spec(Λ )

? ?
-
Spec(R) - Spec(Λ )
Since A was arbitrary, (1) holds for any A ∈ Art/R. Hence D[Fk ],[det(Fk ] is
pro-represented by ÔMR,L
s ,[Fk ] .
R
186 Inder Kaur
Using this we prove the following theorem.

s
Theorem 4.5. The morphism MR,LR
→ Spec(R) is smooth.
s
Proof. Since the scheme MR,L R
is noetherian and smoothness is an open con-
s
dition, it suffices to check that the morphism MR,L R
→ Spec(R) is smooth
s
at closed points. Let [Fk ] be a closed point of MR,L R
. Since the morphism
s
MR,L R
→ Spec(R) is of finite type, to prove that it is smooth at the point [Fk ],
we need to show that the functor ÔM s ,[Fk ] is unobstructed.
R,LR
By Proposition 4.4, the completion of the local ring OMR,L

s ,[Fk ] pro-
R
represents the functor D[Fk ],[det(Fk )] , i.e ÔM s ,[Fk ] D[Fk ],[det(Fk ] . By The-
R,LR
orem 3.19, the deformation functor D[Fk ],[det(Fk )] is unobstructed. Hence the
functor ÔM s ,[Fk ] is unobstructed. This implies ÔMR,L s ,[Fk ] is unobstructed.
R,LR R
s
Hence, the morphism MR,LR
→ Spec(R) is smooth at the point [Fk ].
Acknowledgements
The author thanks Prof. A. Langer for a discussion during the conference ’Top-
ics in characteristic p > 0 and p-adic Geometry ’. The author is grateful to the
Berlin Mathematical School for financial support.
References
[1] V. Artamkin. On deformation of sheaves. Math USSR Izv, 32:663–668, 1989.
[2] H. Esnault and A. Langer. On a positive equicharacteristic variant of the p-
curvature conjecture. Documenta Math. J., 18:23–50, 2013.
[3] B. Fantechi, L. Göttsche, L. Illusie, S. L. Kleiman, N. Nitsure, and A. Vistoli. Fun-
damental algebraic geometry. Grothendieck’s FGA explained,Mathematical Sur-
veys and Monographs, volume 123. Amer. Math. Soc, 2005.
[4] R. Hartshorne. Algebraic Geometry, volume 52. Graduate texts in Math, Springer
Verlag, 1977.
[5] R. Hartshorne. Deformation Theory, volume 257. Graduate texts in Math,
Springer Verlag, 2010.
[6] D. Huybrechts and M. Lehn. The geometry of moduli spaces of sheaves, volume 31.
Aspects of Mathematics, Vieweg, Braunshweig, 1997.
[7] A. Langer. Castenuovo-mumford regularity. Duke Math. J., 124:571–586, 2004.
[8] A. Langer. Semistable sheaves in positive characteristic. Ann of Math, 159:251–
276, 2004.
[9] M. Maruyama. Moduli of stable sheaves II. J.Math.Kyoto Univ, 18:557–614, 1978.
Group Compactifications and
Moduli Spaces
Johan Martens∗,†
Abstract
We give a summary of joint work with Michael Thaddeus that realizes toroidal
compactifcations of split reductive groups as moduli spaces of framed bundles on
chains of rational curves. We include an extension of this work that covers Artin
stacks with good moduli spaces. We discuss, for complex groups, the symplectic
counterpart of these compactifications, and conclude with some open problems
about the moduli problem concerned.
1. Introduction
1.1. The basic problem we want to discuss in this note is the following: given
a reductive group G (which, as an affine variety, is not complete), how can you
think about completions of G? For good measure, we want to restrict ourselves
to completions that are well-behaved, which as a minimum will require G × G-
equivariance. For G = P GL(n) this is a very classical question, going back to
19th century enumerative geometry. We will make no attempt to relate the
history of this, but would like to point out the early paper [Sem51] by Semple,
who was likely the first to think of these completions as (in modern language)
moduli spaces, which will be central theme in our discussion.
In general one can think of this question as a special case of spherical em-
beddings (thinking of G as (G × G)/G), and use the Luna-Vust theory that
classifies these [LV83], which was developed from the early 1980s. Around the
same time, in another seminal work [DCP83, DCP85], De Concini and Pro-
cesi gave a description of embeddings of symmetric varieties, and in particular
showed that every adjoint group (a reductive group with trivial center, hence
necessarily semi-simple) has a canonical smooth completion, which is known as
∗ School of Mathematics and Maxwell Institute, The University of Edinburgh, Peter
Guthrie Tait Road, Edinburgh EH9 3FD, United Kingdom. E-mail: [email protected]
† The author was partially supported by EPSRC grant EP/N029828/1.
Geometry, DOI 10.1007/978-981-10-5648-2_12
188 Johan Martens
the wonderful compactification. For a general description of equivariant com-

pletions of reductive groups, see [Tim03].
1.2. An impetus for looking for completions of G as moduli spaces came from
the search of a universal compactification of the moduli space of G-principal
bundles on curves, as advocated by Sheshadri. Indeed, if X is a complete curve
with a single nodal singularity whose normalization X. is connected, then any
G-bundle on X corresponds to a G-bundle on X,. equipped with an isomorphism
between the fibres over the exceptional points of X̃. This isomorphism can be
changed by the action of G, which indicates that any suitable compactification
of the moduli problem of bundles-on-curves lying over the Deligne-Mumford
compactification of the moduli space of curves would give rise to a compactifi-
cation of the group G itself. Motivated by this, Kausz developed a particular
modular compactification of GL(n) [Kau00].
1.3. In [MT16], Thaddeus and the author introduced a general moduli prob-
lem of framed bundle chains. This is represented by a stack, such that any
(torsion-free stacky simplicial) fan with support in the positive Weyl chamber
of G determines a stability condition on this stack. Imposing this condition
cuts out a substack, whose coarse moduli space is the toroidal embedding of
the group G determined by the fan. All toroidal embeddings of G can be un-
derstood this way.
We will review this work here, indicating how these stacks can be described
as global quotients of reductive monoids using work of Vinberg, and show how
the simplicial condition can be relaxed, to give Artin moduli stacks with good
moduli spaces (in the sense of Alper [Alp13]).
Next we show how semi-projective toroidal embeddings of complex groups
can be understood in real symplectic geometry as non-abelian cuts of the cotan-
gent bundle of a compact form of the group. Finally we conclude with a list
of open problems and questions related to the moduli problem we described.
The aim of this note is to provide a readable introduction, for more technical
details we refer to [MT16, MT12a, MT12b].
1.4. General Setup. Throughout we will let G be a connected split

reductive group over an arbitrary field k (though at some points we will restrict
to k algebraically closed, or even complex), with a chosen split maximal torus
T , center ZG , character lattice V = Hom(T, Gm ) and co-character lattice Λ =
Hom(Gm , T ). Furthermore we will choose a Borel subgroup B, corresponding
to positive simple roots αi and positive Weyl chamber Λ+ Q ⊂ Λ ⊗Z Q.
For our purposes a G-principal bundle will be a G-torsor that is locally
trivial in the étale topology. For technical reasons we ask these to be ratio-
Group Compactifications and Moduli Spaces 189
nally trivial (that is, trivial on the generic point of each component) — this is
automatically satisfied if k is algebraically closed or of characteristic zero.
A polyhedron will be a subset of a vector space cut out by a finite number
of linear inequalities – so in particular, it need not be bounded. Finally we shall
say that a variety is semi-projective if it is projective over an affine variety.
2. The Moduli Problem

We want to consider the moduli problem of framed bundle chains.
2.1. The objects we want to study are G-principal bundles over chains of
projective lines meeting in nodal singularities. The chains are connected, but
their length is a priori arbitrary (though having at least one component) and
allowed to vary in families. We do not want the chain to ‘flip’ in a family though.
To this end, we ask for a consistent choice of endpoints p+ and p− in the chain
(which are encoded, for each family, by two sections of the chain, taking values
in the extremal components over any closed point).
p+ p−
Figure 1. A chain of lines
Each such family of chains of projective lines comes with a canonical action
of the multiplicative group Gm [MT16, Theorem 1.6]. We ask for all our G-
bundles to be equivariant with respect to this Gm -action (i.e. to be equipped
with the lift of this action to the total space of the bundle, commuting with
the G-action). Finally, we want the bundles to be framed over p+ and p− , in a
Gm -invariant way. That is, the sections p+ and p− are lifted to the total space
of the bundle, and these sections are invariant under the action of Gm .
2.2. Before we can proceed, we need to discuss a description of the

isomorphism classes of bundle chains. Recall that the classical theorem of
Birkhoff-Grothendieck-Harder [Gro57, Har68] classifies isomorphism classes of
G-principal bundles on a single projective line: they are given by elements of
Λ/W . Naively, but not incorrectly, one can view this theorem as follows: write
P1 by stereographic projection as A1 ∪ A1 (where the two copies of the affine
line are glued together along A1 ∩ A1 ∼ = Gm ). A G-principal bundle on P1
trivialises on each of the A s, and is determined by a single clutching function
1
A1 ∩ A1 → G. The theorem can be interpreted as saying that this clutching

function can always be taken to be a group homomorphism ρ : Gm → G, which
190 Johan Martens
then can be conjugated to take values in T ⊂ G, and hence gives an element

of Λ. Finally the action of W gives rise to isomorphic bundles.
We need a generalisation of this in two directions: we want to work with
chains of P1 s, not just single copies, and we want to account for the Gm -
equivariance. This generalisation was given in [MT12b]. To summarise it, we
can again first look at a single P1 . By using the Gm -fixed points for stereo-
graphic projection, we can again think of P1 as covered by two copies of A1 ,
equipped with linear Gm -actions of weights 1 and −1 respectively. Any Gm -
equivariant G-principal bundle on P1 again trivializes on each A1 , and for each
the Gm -equivariance can just be characterised by indicating how Gm acts on
the fibre over p+ or p− . This is given by two elements ρ+ , ρ− ∈ Λ, and it is
straightforward to see that the only way to glue two such bundles on A1 to
get a bundle on P1 is to use ρ = ρ+ − ρ− as clutching function. Finally, the
diagonal action of W gives again rise to isomorphic bundles, and hence we see
that the isomorphism classes of Gm -equivariant G-principal bundles are given
by Λ2 /W .
Moreover, for chains this story goes through line by line, and finally we
obtain that the isomorphism classes of Gm -equivariant G-principal bundles on
a chain of P1 s of length n is given by assigning an element of Λ to each Gm -
fixed point of the chain, and this assignment is again unique up to the diagonal
action of W . So we end up with Λn+1 /W .
2.3. Now if we go back to our moduli problem of framed bundle chains,

we see that, since our bundles are framed at p+ and p− (and the framing is
preserved by Gm ), the co-characters are necessarily zero there, and hence the
only (potentially) non-trivially co-characters are assigned to the nodes . So for
a single framed bundle chain over a chain of lenght n+1, the isomorphism class
of the bundle (forgetting the framing) is given by an element of Λn /W , where
the Weyl group W acts diagonally. We shall refer to this data as the splitting
type of the bundle.
We shall begin by making the restriction that we only consider bundles
where all the elements in Λn making up its splitting type lie in the same Weyl
chamber (this condition is preserved by the action of W of course). It is shown
in [MT16, §3] that this moduli problem determines a smooth Artin stack (non-
separated and of infinite type). Moreover, the group G×G acts on this stack, by
changing the framing over p+ and p− . Finally, this stack contains G × BGm as
the open substack corresponding to those framed bundle chain where all chains
have length one. To see the last statement, remark that any Gm -equivariant
framed bundle on a single projective line is necessarily trivial, and hence the
only thing that can vary are the framings. These can be changed by G × G,
but as we only consider these up to isomorphisms, we can quotient out the
diagonal G. This gives a family parametrized by G ∼ = (G × G)/G, but as the
Gm -equivariance is built into the moduli problem, we really have the gerbe
G × BGm .
2.4. There is a general process of rigidification [ACV03, §5.1] that we can

use to remove the Gm from the automorphism (see [MT16, p. 76]). We will
refer to the rigidified stack as M. Note that in [MT16] this notation is used for
the open substack where furthermore all vectors in a splitting type are linearly
independent. As we want to consider non-simplical fans below, we shall not
make this extra restriction here (this is the only instance where our notation
will deviate from that of [MT16]). This stack M is still G × G-equivariant, and
honestly contains G as an open substack.
3. The Stability Condition

3.1. There is no hope for the stack M to have a coarse (or good) moduli
space. We will however now formulate stability conditions which cut out open
sub-stacks that do have these.
Begin by choosing a fan Σ (in the sense of toric varieties) with support
in the positive Weyl chamber of G. In fact we want a little bit more: we also
want a choice of an non-zero element of Λ on each ray, which we will refer to
as a ray vector, i.e. we want to have a (torsion-free) stacky fan in the sense
of [BCS05, FMN10]. Strictly speaking stacky fans have to be simplicial, and
[MT16] was restricted to this case as well; we will show below in Section 5 that
this can be relaxed to arbitrary fans, but for now we shall restrict ourselves
to the simplicial case. Finally we need to make a choice of ordering of the ray
vectors β1 , . . . , βn . Absolutely any ordering is fine, see Section 7.1 below for
more comments about this.
3.2. With such a choice of stacky fan and ordering of the ray vectors, we
can now give the corresponding stability condition: a framed bundle chain is
stable if its isomorphism class is determined by a tuple of co-characters that
all lie in the same Weyl chamber (which can be chosen to be the postive Weyl
chamber), who moreover are all distinct ray vectors of a single cone in the fan,
and occur in the chosen order from p+ to p− . See Figure 2 for an illustration.
Imposing the stability condition cuts out an open substack MG (Σ) ⊂ M.
The main result of [MT16] is now the following:
Theorem 3.1. The stack MG (Σ) is a smooth, separated tame stack (hence
Deligne-Mumford if k has characteristic zero), which is proper if the support of
Σ is all of the Weyl chamber. The coarse moduli space of MG (Σ) is the toroidal
spherical embedding of (G × G)/G given by the uncolored fan wΣ, where w is
the longest element in W .
192 Johan Martens
β1 β2
wβ1 β3
wβ3
wβ2
Σ-stable bundle chains include Σ-unstable bundle chains include
β1
β2 wβ1 β2 β1 β2
β1 wβ3 β2
β3 wβ2 β1 β3 wβ3
β3
Figure 2. Illustration of a stability condition determined by a stacky fan
Note that though the stability condition was dependent on the choice of
ordering of the ray vectors (β1 , . . . , βn ) in the stacky fan, for different choices
the resulting stacks are canonically isomorphic.
4. The Cox-Vinberg Quotient Construction

In this section we will show that the stacks MG (Σ) discussed above can actually
be obtained as quotients of a quasi-affine variety by a torus. This quasi-affine
variety is an open subvariety of a reductive monoid, and if the fan Σ is polar
(i.e. coming from a polytope) such that W Σ has convex support, it can be
understood as a (semi-) stable locus in the sense of Geometric Invariant Theory
(which guarantees that the corresponding coarse moduli space is projective over
an affine, and indeed projective if the fan W Σ was complete).
4.1. The Cox Construction Revisited. The first ingredient that

we need is the classical Cox constructions for toric varieties. Let us briefly recall
how this goes: suppose X is a toric variety for a torus T , given by a fan Σ with
N rays. As before we assume ray vectors βi ∈ Λ chosen; if one is only interested
in toric varieties one can simply chose the first non-zero elements of Λ along
the rays of Σ for these.
The standard presentation of the Cox construction assumes that the βi
generate Λ ⊗Z Q (this is always the case if X, or equivalently Σ, is complete).
In this case we have a short exact sequence (where Gβ := GN m)
(β1 ,...,βN )
1 L Gβ T 1. (1)
Remark that this sequence only depends on the ray(vector)s of Σ, not on
the full fan structure. Cox now defines an open subvariety A0β inside Aβ := AN
as the union of the Uσ , where for every cone σ of Σ we define
Uσ = {(x1 , . . . , xN ) ∈ Aβ | xi = 0 if βi is not a ray vector of σ}
(so A0β does depend on all the data in the fan).
Cox shows that the toric variety X can be understood as the categorical
quotient A0β /L, which is a geometric quotient when Σ is simplicial. Moreover, if
Σ is the normal fan to a polyhedron (which corresponds to X being projective
over an affine variety), then A0β /L can be understood as the semi-stable locus
in the sense of geometric invariant theory (in fact stable if Σ is simplicial).
Finally, if Σ is simplicial, then also for the corresponding toric orbifolds we
have X ∼ = [A0β /K].
4.2. Now, it turns out that the condition that the βi generate Λ ⊗ Q can be
relaxed in an easy way: even if this is not the case, i.e. when the sequence (1)
is no longer exact on the right (for instance in the extreme case where X = T ,
where there are no βi at all), we still have that

X∼ = A0 × T /Gβ ,
β
and all the above statements (about geometric vs categorical quotients, orb-
ifolds, and GIT) remain valid in this generality. Whereas Aβ was Spec of the
Cox ring of X, Aβ × T is Spec of the equivariant Cox ring – using the equiv-
ariance in this way is exactly what compensates for the lack of compactness.
So this generalizes the Cox construction to all toric varieties, at no addi-
tional cost, but more importantly for us this variation on the Cox constrution
also has a non-abelian generalization. For this we need an extra ingredient: the
Vinberg monoid (a.k.a. enveloping semigroup) of G.
194 Johan Martens
4.3. The Vinberg Monoid. The Vinberg monoid SG , introduced in

[Vin95], is a reductive monoid (i.e. a monoid object in the category of (affine)
algebraic varieties, whose group of units is a reductive group) that one can
canonically associate to a reductive group G. Its group of units is Genh =
(G×T )/ZG , and it comes with a canonical monoid morphism to the affine space
A = SG //(G × G), which can be understood as the toric variety for the group
T /ZG corresponding to the positive Weyl chamber (note that this is indeed
an affine space, as the primitive elements along the rays are the fundamental
co-weights, which indeed generate the co-weight lattice).
At least for k algebraically closed of characteristic zero it can be constructed
explicitly, as Spec of a subring of the ring of regular functions of Genh . By the
algebraic Peter-Weyl theorem, the latter can be decomposed as
1
k[Genh ] = k[Genh ](λ,μ) ,
+
(λ,μ)∈VG
enh
where k[Genh ](λ,μ) denotes the isotypical components of type (λ, μ) (which one
can think of as the matrix coefficient for the corresponding highest weight
representations), and the character lattice of Genh can be described as
2 3
VGenh = (λ, μ) ∈ VG2 | μ − λ = mi αi , with all mi ∈ Z
i
(here the αi are the positive simple roots of G). In this language we have
1
k[SG ] = k[Genh ](λ,μ) ,
+
(λ,μ)∈VG ∩QG
enh
!
where QG is the cone consisting of those (λ, μ) such that μ−λ = i mi αi where
all mi ≥ 0. A description of SG in arbitrary characteristic (for k algebraically
closed) was given in [Rit01].
Vinberg showed (for G semisimple) that SG satisfies a universal property in
the category of reductive monoids [Vin95, Theorem 5]: essentially it says that
any reductive monoid S satisfying some natural conditions is obtained from
SG , where G = [S × , S × ] (here S × is the group of units of G), as the base
change of SG → A by a toric morphism of an affine toric variety to A. From
our point of view it is useful to think of SG as a (universal) degeneration of G
(the generic fibres of SG → A are indeed isomorphic to G).
Vinberg studied SG extensively; in particular he introduced a certain open
0
subvariety SG (not a submonoid) of SG , and showed that the geometric quotient
0
SG /T is the wonderful compactification of Gad = G/Z of De Concini-Procesi
0
[DCP83]. It was shown in [MT16, Theorem 5.4] that SG can also be understood
as a GIT-stable set for a suitable linearization of the action of T on SG .
4.4. The Cox-Vinberg Hybrid. It turns out that the Cox construc-
tion (or rather the variation thereof sketched above) and the Vinberg con-
struction of the wonderful compactification of Gad can be hybridized in a very
natural way. Indeed, as all of the βi are in the positive Weyl chamber of G, the
morphisms βi : Gm → T naturally induce monoid morphisms A1 → A. These
combine into a monoid morphism Aβ → A, and we can now look at the reduc-
tive monoid SG,β arising as the fibred product Aβ ×A SG . Inside of SG,β we can
now again look at an open subvariety SG,β 0
:= A0β ×A SG0
. If Σ is a polar fan this
can again be understood to be a GIT semi-stable locus [MT16, Theorem 8.1]
(stable if Σ is simplicial). In [MT16, proof of Theorem 6.4] a family of framed
bundle chains was constructed over Aβ ×A SG 0
, which induces an isomorphism
of stacks
0
SG,β /Gβ ∼= MG (Σ).
4.5. In some sense one could think of this Cox-Vinberg monoid SG,β as
playing a similar rol to MG (Σ) as the quot-scheme plays to the moduli space
of bundles on a smooth projective curve. There is a crucial difference however:
we don’t know what moduli problem these SG,β represent, and in fact are only
able to construct a family of framed bundle chains over the open subvariety
Aβ ×A SG 0
. It is very tempting to search for a modular interpretation of all of
the monoid; see Section 7.2 below.
5. The Non-Simplicial/Artin Case

5.1. As mentioned above, the treatment in [MT16] was restricted to stability
conditions given by simplicial (stacky) fans, which lead to moduli stacks that
are smooth Deligne-Mumford or tame. We will show here that with little extra
effort one can remove the simplicial condition, leading to moduli stacks that are
Artin but still have so-called good moduli spaces, in the style of Alper [Alp13].
To this end we shall from now on understand the notion of a stacky fan Σ to
mean an arbitrary fan (i.e. not necessarily simplicial), together with a non-zero
ray vector βi ∈ ΛG on each of its rays, as well as an ordering of these ray
vectors.
The approach we shall take is to generalise the Cox-Vinberg construction.
This is not the only possible path, but in the style of this note we prefer it as
it extends the results in the most economical fashion. As we want to use the
gluing results of good moduli spaces from [Alp13], we restrict ourselves now to
working over algebraically closed fields of characteristic zero. Presumably the
gluing can be extended to the setting of adequate moduli spaces, as in [Alp14],
in which case the restriction on characteristic is not necessary.
196 Johan Martens
Recall also that the construction of the Vinberg monoid over algebraically
closed fields of positive characteristic was done by Rittatore in [Rit01] using
the Luna-Vust theory of spherical embeddings. It is widely assumed by the
experts that the Vinberg monoid also exists for split reductive groups over
arbitrary fields, though we are not aware of any place where this is written
down. The theory of spherical embeddings has been extended by Huruguen
[Hur11] to more general settings (including the case of split reductive groups
over arbitrary fields); presumably the combination of the work of Huruguen
and Rittatore would extend the notion of Vinberg monoid to arbitrary fields,
which should also extend the discussion below.
5.2. The notion of a good moduli space [Alp13] for an Artin stack generalizes
the relationship between the stack [X ss /G] and the GIT-quotient X//G for the
linearized action of a reductive group on a (semi-)projective variety – here X ss
stands for the semi-stable locus in X. For a general Artin stack X a good moduli
space is an algebraic space Y , together with a morphism φ : X → Y such that
the push-forward functor on quasi-coherent sheaves for φ is exact, and such
that the induced morphism on sheaves OY → φ∗ OX is an isomorphism. If a
good moduli space exists it is unique, in the sense that φ is universal for maps
from X to algebraic spaces.
Of importance for us is that good moduli spaces can be glued, in the fol-
lowing sense [Alp13, §7.8]: if φ : X → Y is an Artin stack with a good moduli
space, an open substack U ⊂ Y is said to be saturated if φ−1 (φ(U)) = U. We
now have
Proposition 5.1 ([Alp13, Proposition 7.9]). Suppose X is an Artin stack cov-

ered by open substacks Ui such that for each i there exist a good moduli space
φi : Ui → Yi , with Yi a scheme. Then there exist a good moduli space φ : X → Y
and open sub-algebraic spaces Ỹi ⊂ Y such that Ỹi ∼ = Yi and φ−1 (Ỹi ) = Ui if
and only if for each i, j, Ui ∩ Uj is saturated for φ : Ui → Yi .
5.3. We want to apply Proposition 5.1 to the open covering of MG (Σ) given
by the cones σ in a stacky fan Σ (of course we want to think of these as stacky
fans in their own right, so we remember the ray vectors βi ). We shall refer to
these open substacks as MG (σ). The criterion will be easy to verify, since we
know, by the results on the GIT construction in [Alp13], that they hold for
those stability conditions that give rise to semi-projective good moduli spaces.
5.4. The conditions under which the stability condition that cuts out
MG (Σ) can be understood as GIT-(semi-)stability for the Cox-Vinberg monoid
were described in [MT16, §8]. Note that we are applying GIT here to all semi-
projective schemes, i.e. schemes projective over an affine. These include all affine
and projective varieties, and more precisely are exactly the schemes that can
be generated as Proj of a graded k-algebra. We shall say that a fan is normal
if it is determined by some polyhedron (with the ray vectors of the fan the
outward normal vectors of the polyhedron).
Theorem 5.2 ([MT16, Theorem 8.1]). If Σ is a normal fan such that W Σ has
convex support then the stack MG (Σ) is isomorphic to the stack [SG,β
ss
/Gm ],
ss 0
where SG,β is the GIT-semistable locus, which equals SG,β .
Observe that strictly speaking in [MT16] only the case of simplicial fans was
discussed (and hence no properly semi-stable points were present); the proof
of the above theorem, however, goes through without change in the general
case. Also note that the condition for a fan Σ with support in the postive
Weyl chamber to be normal such that W Σ has convex support can also be
stated as saying that Σ is the normal fan to a polyhedron P in the positive
Weyl chamber such that W P is a polytope which meets the walls of all Weyl
chambers perpendicularly.
Lemma 5.3. Any (stacky) cone σ with support in the positive Weyl chamber
of G can be included in a normal fan Σ such that W Σ has convex support.
Proof. First, remark that without loss of generality, we can assume that σ has
dimension equal to the rank of G. Let P. be the polyhedron (in VR ) that is
obtained by taking −σ ∨ (where σ ∨ is the dual cone to σ), and shifting it by
a non-zero-vector v contained in σ ∨ . The issue is that P. may not meet the
walls of the Weyl chambers perpendicularly. In order to correct this, let Q be
the polyhedron given by taking the cone whose outward normal vectors are
the simple positive roots of G (which will not be strongly convex if G is not
semi-simple), and shifting it by (1 + )v. If we now take > 0 small enough, the
polyhedron given by the intersection P = P. ∩ Q will now meet all of the walls
of the Weyl chamber perpendicularly, and its polar fan (wich is independent
of the choices of v and under the stated conditions) will be contained in the
positive Weyl chamber in ΛR , and it will contain the original cone σ.
5.5. We can now conclude:

Theorem 5.4. For any stacky fan Σ as above (possibly non-simplicial), we
have that MG (Σ) has a good moduli space, which is a scheme that as a
(toroidal) spherical variety is given by the uncolored fan wΣ.
Proof. We need to verify that the conditions of Proposition 5.1 hold for the
substacks MG (σ), for any (stacky) cone σ in Σ. So firstly, we need to establish
that each MG (σ) has a good moduli space. In light of [Alp13, Remark 6.2],
. for some
this boils down to verifying that each MG (σ) is saturated in MG (Σ),
198 Johan Martens
stacky fan Σ . containing σ such that MG (Σ) . has a good moduli space. By
.
Lemma 5.3 we can always find such a Σ, and moreover assume MG (Σ) . (or
rather its good moduli space) is obtained as a GIT-quotient; hence we can
apply the general results of GIT there. In particular, we know that in the good
moduli space orbits need to be identified if their closures meet in the semi-stable
0
locus, in casu SG,β .
We can now use the Cox-Vinberg construction to show that MG (σ) will
be saturated in MG (Σ), . which means that no two orbits that do not lie in
Uσ ×A SG0 . will have closures that intersect in S 0 . Indeed,
. in Σ
for some cone σ G,β
again using [Alp13, Remark 6.2], observing that the corresponding result holds
for toric varieties (since any toric variety corresponding to a single cone can
itself be obtained as a GIT-quotient – something not true for arbitrary toroidal
spherical varieties corresponding to a single cone), and hence using the variation
on the Cox construction described above in Section 4.2 we find that if orbits
in A0β × T intersect, they have to be contained in Uσ × T for some cone σ . in
.
Σ. As all orbits in T itself are closed, we find moreover that the corresponding
orbit closures intersect in Uσ .
0
We can now go back to SG,β : if orbit closures intersect there, so will their
images in Aβ , and hence also the corresponding orbits in A0β × T . Using the
0
toric statement, we find that4the original orbits had to be contained in some

Uσ ×A SG 0
. Hence [Uσ ×A SG0 . and therefore each
Gβ ] is saturated in MG (Σ),
MG (σ) has a good moduli space.
The fact that for any σ, σ in Σ we have that MG (σ ∩ σ ) is saturated in
MG (σ) follows from a similar reasoning using a Σ . as above that contains σ

(note that it is not necessary that σ be contained in Σ, . the only fact we use is
that σ ∩ σ is a sub-cone of σ).
Finally the reasoning identifying the good moduli space of MG (Σ) as the
spherical embedding corresponding to wΣ is exactly the same as [MT16, Propo-
sition 10.1].
6. Non-abelian Symplectic Cutting

6.1. In this section we will change viewpoint, and think about those embed-
dings MG (Σ) (for k = C) whose good moduli space is semi-projective in terms
of (real) symplectic geometry.
To this end, recall that the classical (abelian) symplectic cut procedure,
introduced by Lerman [Ler95], takes a Hamiltonian T -space M (for T a compact
connected torus) and a polyhedron P ⊂ t∗ (where t = Lie(T )), the normal
vectors to all of whose facets lie in the lattice Hom((1), T ) ⊂ t∗ . It returns a new
Hamiltonian T -space MP , whose image under the moment map is μT (MP ) =
μT (M ) ∩ P , and such that the pre-images of the interior of P in M and MP

are T -equivariantly symplectomorphic.
Symplectic cutting can be understood from two points of view, either as a
global quotient
MP = (M × CN )//U (1)N ,
(where // denotes symplectic reduction here), or as a local surgery, by taking the
pre-image of P under μT , and collapsing the circle-actions determined by the
normal vectors on the pre-images of the facets of P . Note that in particular the
global quotient construction allows the cutting to be interpreted in Kähler, or
even algebro-geometric terms. On the other hand, for the local surgery picture
one just needs a map to a polytope such that the projection to the normal
direction for each facet is a moment-map in a neighbourhood of that facet
(which is strictly less than asking for a full T -moment map).
6.2. Symplectic cutting is an elementary but very useful operation, and it

is natural to try to generalise it to non-abelian group actions. A number of
suggestions for this are in the literature (e.g. [Wei01, Par09]), but in some
sense the most natural one is due to Woodward [Woo96] and further developed
by Meinrenken [Mei98]. This operation starts from a Hamiltonian K-space M ,
for K a compact connected Lie group. It is useful to compose the moment map
μK : M → k∗ with the quotient map under the co-adjoint action k∗ → t∗+ to the
positive Weyl chamber in the dual of the Lie algebra of a maximal torus of K
– we shall denote this composition Φ. If P is now a polyhedron in t∗+ such that
if it meets any facet of t∗+ in Φ(M ), it does so perpendicularly, one can again
apply the recipe for the local surgery for abelian cutting (since the components
of Φ given by the outward normal vectors will all be smooth on neighborhoods
of the pre-images of the corresponding faces), and apply it to this map Φ. The
result will be a Hamiltonian K-space MP , with Φ(MP ) = Φ(M ) ∩ P .
One difference with the abelian set up is that a global (Kähler) quotient
counter part to this construction seems to be missing: indeed, Woodward ap-
plies his cut to a coadjoint orbit (which in particular is Kähler), to obtain a
symplectic space that cannot carry any compatible Kähler structure [Woo98].
This seems to rule out any possibility of interpreting this operation in complex
algebraic geometry.
We show in [MT12a] however that this is not the case, as long as the poly-
tope P has the property that all of its outward normal vectors lie in the pos-
itive Weyl chamber t+ (and hence P determines a fan Σ as we have used
above). In this case a global quotient construction to the operation of Wood-
ward and Meinrenken does exist. It begins by taking the symplectic reduction
(M ×SG )//K, where SG is the Vinberg monoid for the complexification G = KC
(one can canonically interpret SG as a (stratified) symplectic space by choos-
200 Johan Martens
ing a faithful representation V of SG to embed it into the affine space End(V )

equipped with the Hermitian Kähler structure – the symplectic structure this
endows SG with is independent of the choice of V ). This spaces comes with a
natural morphism to A, and the data of the polytope now allows us to take the
fibered product with Aβ over A. Finally the symplectic cut of Woodward and
Meinrenken can now be constructed as a global quotient as
MP ∼
= (((M × SG ) //K) ×A Aβ ) //ξ U (1)N
(see [MT12a, Corollary 4]). Note that the fibered product only needs the ray
vectors of Σ, but the level at which the final symplectic reduction is performed
uses all the data of P . If M was a complex semi-projective variety to begin
with, and P had all outward normal vectors in the positive Weyl chamber, this
construction can therefore be understood to give a new semi-projective variety.
6.3. For details of the proof of this we refer to [MT12a], but we will briefly
mention the key ideas behind this: firstly, it is well-known that any suitably
intrinsic operation X one wants to apply to a Hamiltonian K-space M can
equivalently be done by first applying it to T ∗ K, and then taking the quotient
of the carthesian product of the resulting space and M – symbolically we could
write this as
X(M ) ∼ = (M × X (T ∗ K)) //K.
This is the case for symplectic implosion [GJS02] where the implosion of T ∗ K
is dubbed the universal implosion (and trivially for X = id or symplectic reduc-
tion by K), but also for symplectic cutting. Indeed, in the abelian symplectic
cut one has that the universal symplectic cut of T ∗ T is just the symplectic toric
space corresponding to the polyhedron P .
In fact, this latter point of view is just the symplectic counter part of the
alternative approach to the Cox construction that was outlined above. The stan-
dard symplectic counterpart to the Cox construction is known as the Delzant
construction [Del88], which essentially says that any compact toric manifold (or
one whose normal fan satisfies the condition that (1) is exact) can be obtained
as the symplectic reduction of CN , where N is the number of facets of P . This
again suffers from the same disadvantages as the standard Cox construction
when one wants to consider more general toric symplectic spaces (with proper
moment maps).
Whereas in the algebro-geometric picture this was remedied by looking
at equivariant Cox rings, in the symplectic picture one can just replace the
symplectic reduction of CN by the symplectic cut of T ∗ T with respect to P
(one can think of this as the universal cut). Moreover, we find that by doing
so we not only generalize the validity of the Delzant construction, but also
allow for a non-abelian generalisation: indeed, it is shown in [MT12a, Theorem
3] that the universal non-abelian cuts T ∗ KP correspond exactly to the spaces

MG (Σ) discussed above, and one can then use the Cox-Vinberg construction
for the latter.
The second idea, crucial in establishing this correspondence between T ∗ KP
and MG (Σ), is to use a section of the moment map for SG . The use of such a
section of the moment map was already implicit in [Del88] (present whenever
a square root showed up in the defining formulas), and can still be done in the
non-abelian setting, by using the polar decomposition of complex matrices in
combination with an (arbitrary) faithful representation of SG .
7. Open Questions
Finally, we will list here a number of open questions regarding the compactifi-
cations discussed above.
7.1. Ordering of Weights and Shuffling Functors. As dis-

cussed above, the stability condition requires an ordering of the ray vectors
βi . Different orderings will cut out different substacks MG (Σ), but it turns out
these are all isomorphic, as can easily be seen from the Cox-Vinberg construc-
tion. This implies that there is a sort of shuffling functor that associates with
every family of bundle chains that are stable for one choice of ordering, a new
family (all of whose chains have the same length as before) that is stable with
respect to another choice of ordering, such that the ray vectors in the splitting
type of a bundle chain over a closed point just get shuffled around.
Other circumstantial evidence comes from thinking about affine toric vari-
eties: all of these have modular interpretations by the description above, but
at the same time we also know that all of affine toric varieties are canonically
abelian monoids, extending the group operation of the torus. It is therefore
natural to ask how this monoidal structure can be understood from the mod-
ular point of view. The most obvious candidate is given by gluing two chains
together, attaching p+ from the first to p− of the second. If one has a shuf-
fling functor available, one can then shuffle the chain to have the ray vectors
in the correct order, and contract any component of the chain whose two as-
sociated ray vectors are identical. This hypothetical shuffling operation should
correspond to the monoidal operation on the affine toric variety.
In spite of this evidence we do not know how to describe such a shuffling
functor directly however.
7.2. Modular Interpretations of the Vinberg Monoid. Re-

lated to the gluing of chains is the question wether the Vinberg monoid itself
represents a moduli problem. As an embedding of Genh , SG is not toroidal, so
202 Johan Martens
doesn’t fall under the framework discussed above. The stack of all framed bun-
dle chains has a monoidal structure by gluing chains together, but the condition
that all elements of the splitting type lie in the same Weyl chamber typically
gets lost while doing this, so even the construction of M in [MT16] needs to
be expanded to accommodate this.
7.3. Stratifications of the Stack M. The stability condition de-

scribed above is introduced in a seemingly ad-hoc way, motivated by the
Birkhoff-Grothendieck-Harder theorem and its generalization, and the use of
fans in toric and spherical geometry. Recently, however, more systematic stud-
ies have been made of notions of (in-)stability on stacks, by Heinloth [Hei16]
and Halpern-Leistner [HL14]. One can therefore ask if it is possible to recover
the notions of stability we have introduced from a general Θ-stratification fol-
lowing Halpern-Leistner. Note that since the method above also gives rise to
non-projective complete good moduli spaces, at least some of the methods for
constructing stability functions in [HL14] would have to be extended.
7.4. Beyond Toroidal. The work described above is restricted to moduli

stacks whose good moduli spaces are toroidal embeddings of G. The theory
of spherical varieties also gives many other embeddings, classified by colored
cones. Can the modular interpretation be extended to these as well?
7.5. Gauge-theoretic Description. In Section 6 above a symplectic

viewpoint on the compactifications of G was given in terms of non-abelian
cutting. This does not directly come with a modular interpretation however.
It is a seminal result in the theory of moduli spaces of bundles on projective
curves that they can also be interpreted (as stratified real smooth spaces) as
moduli spaces of flat connections, for the compact form of the structure group,
on the underlying surface. This is a special case of the Hitchin-Kobayashi corre-
spondence, in this setting due to Donaldson [Don83], but essentially originating
with the work of Narasimhan-Seshadri [NS64]. Given that the spaces we con-
sider can be interpreted as moduli spaces of G-bundles in algebraic geometry,
it is natural to ask if a similar interpretation exists, in gauge-theoretic, or at
least differential-geometric terms?
An intruiging hint comes from the work of Mundet i Riera and Tian
[MiRT09], who give a compactification of the moduli space of so-called twisted
holomorphic maps – equivariant maps of a U (1)-principal bundle over a Rie-
mann surface into a symplectic manifold with Hamiltonian U (1) action. In
their compactification they see that, when a Riemann surface degenerates to
one with nodal singularities, they have to allow the images of the various com-
ponents to be disjoint of each other, but connected by paths of broken flow
lines for the U (1) moment map. The moduli problem we discuss above can
in some sense be interpreted as being given by twisted holomorphic maps

of P1 with two marked points into BG, and it is tantalizing to try to inter-
pret the chains of projective lines similarly as (U (1)-orbits of) broken flow
lines.
7.6. Compactification of Universal Moduli Space. As we in-

dicated above, a historical motivation for studying group compactifications is
the link with a compactification of the universal moduli space of bundles on
curves (where both are allowed to vary), lying over the Deligne-Mumford com-
pactification of the moduli space of curves. This has been an open problem for
several decades now; in the case of vector bundles various constructions are
available in the literature (e.g. [Pan96, Sch04, NS99, Gie84]), and in particular
for the case of line bundles the literature is vast, but for general principal bun-
dles much less is known – for recent approaches to this based on compactifying
loop groups, see the work of Solis [Sol12, Sol13].
Note that compactifying the group is not sufficient to obtain a compacti-
fication of these moduli spaces, but any such compactification does give rise
to a compactification of the group, by taking a fixed stable bundle on a nodal
curve whose normalisation is connected, and creating a family of bundles on
the nodal curve out of this parametrised by the group, by changing the gluing
at the node. One can therefore ask if a compacticiation of the universal moduli
space of bundles on curves exists, such that the corresponding compactifica-
tions of the group (with attached modular interpretation) are those described
above.
7.7. Link with Work of Kausz. As mentioned in the introduction,

Kausz introduced a particular equivariant compactification of GL(n), and gave
it a modular interpretation (in terms of “generalized isomorphisms” and “back-
and-forth morphisms”, see [Kau00, §5]), motivated by the study of degenera-
tions of moduli spaces of vector bundles on smooth curves. As a variety it can
be understood by embedding GL(n) into P GL(n + 1), and then taking the
closure in the wonderful compactication of the latter. As a spherical variety
this is toroidal, and hence the variety can also be understood as a moduli space
of framed bundle chains (see [MT16, §12]).
It would be interesting to see if the corresponding morphism can be un-
derstood in terms of the two moduli problems. In other work of Kausz some
ideas similar to those described above are present (e.g. moduli spaces of
stable maps into BGL(r) in [Kau10], or in particular chains of projective
lines, dubbed zollstöcke, in [Kau08]). Neverthless we have not been able to
find a link between the two modular interpretations of the compactifcation
of GL(r).
204 Johan Martens
References
[ACV03] Dan Abramovich, Alessio Corti, and Angelo Vistoli. Twisted bundles and
admissible covers. Comm. Algebra, 31(8):3547–3618, 2003. Special issue in
honor of Steven L. Kleiman. doi:10.1081/AGB-120022434.
[Alp13] Jarod Alper. Good moduli spaces for Artin stacks. Ann. Inst. Fourier
(Grenoble), 63(6):2349–2402, 2013. doi:10.5802/aif.2833.
[Alp14] Jarod Alper. Adequate moduli spaces and geometrically reductive group
schemes. Algebr. Geom., 1(4):489–531, 2014. doi:10.14231/AG-2014-022.
[BCS05] Lev A. Borisov, Linda Chen, and Gregory G. Smith. The orbifold Chow
ring of toric Deligne-Mumford stacks. J. Amer. Math. Soc., 18(1):193–215
(electronic), 2005. doi:10.1090/S0894-0347-04-00471-0.
[DCP83] C. De Concini and C. Procesi. Complete symmetric varieties. In Invariant
theory (Montecatini, 1982), volume 996 of Lecture Notes in Math., pages
1–44. Springer, Berlin, 1983. doi:10.1007/BFb0063234.
[DCP85] C. De Concini and C. Procesi. Complete symmetric varieties. II. Intersec-
tion theory. In Algebraic groups and related topics (Kyoto/Nagoya, 1983),
volume 6 of Adv. Stud. Pure Math., pages 481–513. North-Holland, Ams-
terdam, 1985.
[Del88] Thomas Delzant. Hamiltoniens périodiques et images convexes de
l’application moment. Bull. Soc. Math. France, 116(3):315–339, 1988.
[Don83] S. K. Donaldson. A new proof of a theorem of Narasimhan and Seshadri. J.
Differential Geom., 18(2):269–277, 1983. URL: http://projecteuclid.org/
euclid.jdg/1214437664.
[FMN10] Barbara Fantechi, Etienne Mann, and Fabio Nironi. Smooth toric
Deligne-Mumford stacks. J. Reine Angew. Math., 648:201–244, 2010.
doi:10.1515/CRELLE.2010.084.
[Gie84] D. Gieseker. A degeneration of the moduli space of stable bundles. J.
Differential Geom., 19(1):173–206, 1984. URL: http://projecteuclid.org/
euclid.jdg/1214438427.
[GJS02] Victor Guillemin, Lisa Jeffrey, and Reyer Sjamaar. Symplectic implosion.
Transform. Groups, 7(2):155–184, 2002. doi:10.1007/s00031-002-0009-y.
[Gro57] A. Grothendieck. Sur la classification des fibrés holomorphes sur la sphère
de Riemann. Amer. J. Math., 79:121–138, 1957. doi:10.2307/2372388.
[Har68] Günter Harder. Halbeinfache Gruppenschemata über vollständigen Kur-
ven. Invent. Math., 6:107–149, 1968. doi:10.1007/BF01425451.
[Hei16] Jochen Heinloth. Hilbert-Mumford stability on algebraic stacks and appli-
cations to G-bundles on curves, 2016. arXiv:1609.06058.
[HL14] Daniel Halpern-Leistner. On the structure of instability in moduli theory,
2014. arXiv:1411.0627.
[Hur11] Mathieu Huruguen. Toric varieties and spherical embeddings over an ar-
bitrary field. J. Algebra, 342:212–234, 2011. doi:10.1016/j.jalgebra.2011.
05.031.
[Kau00] Ivan Kausz. A modular compactification of the general linear group. Doc.
Math., 5:553–594 (electronic), 2000.
[Kau08] Ivan Kausz. A cyclic operad in the category of artin stacks and gravita-
tional correlators, 2008. arXiv:0802.3675.
[Kau10] Ivan Kausz. Stable maps into the classifying space of the general linear
group. In Teichmüller theory and moduli problem, volume 10 of Ramanu-
jan Math. Soc. Lect. Notes Ser., pages 437–449. Ramanujan Math. Soc.,
Mysore, 2010.
[Ler95] Eugene Lerman. Symplectic cuts. Math. Res. Lett., 2(3):247–258, 1995.
doi:10.4310/MRL.1995.v2.n3.a2.
[LV83] D. Luna and Th. Vust. Plongements d’espaces homogènes. Comment.
Math. Helv., 58(2):186–245, 1983. doi:10.1007/BF02564633.
[Mei98] Eckhard Meinrenken. Symplectic surgery and the Spinc -Dirac operator.
Adv. Math., 134(2):240–277, 1998. doi:10.1006/aima.1997.1701.
[MiRT09] I. Mundet i Riera and G. Tian. A compactification of the moduli
space of twisted holomorphic maps. Adv. Math., 222(4):1117–1196, 2009.
doi:10.1016/j.aim.2009.05.019.
[MT12a] Johan Martens and Michael Thaddeus. On non-Abelian symplectic cut-
ting. Transform. Groups, 17(4):1059–1084, 2012. doi:10.1007/s00031-012-
9202-9.
[MT12b] Johan Martens and Michael Thaddeus. Variations on a theme of
grothendieck. 2012. arXiv:1210.8161.
[MT16] Johan Martens and Michael Thaddeus. Compactifications of reductive
groups as moduli stacks of bundles. Compos. Math., 152(1):62–98, 2016.
doi:10.1112/S0010437X15007484.
[NS64] M. S. Narasimhan and C. S. Seshadri. Holomorphic vector bundles on
a compact Riemann surface. Math. Ann., 155:69–80, 1964. doi:10.1007/
BF01350891.
[NS99] D. S. Nagaraj and C. S. Seshadri. Degenerations of the moduli spaces
of vector bundles on curves. II. Generalized Gieseker moduli spaces.
Proc. Indian Acad. Sci. Math. Sci., 109(2):165–201, 1999. doi:10.1007/
BF02841533.
[Pan96] Rahul Pandharipande. A compactification over M g of the universal moduli
space of slope-semistable vector bundles. J. Amer. Math. Soc., 9(2):425–
471, 1996. doi:10.1090/S0894-0347-96-00173-7.
[Par09] Paul-Émile Paradan. Formal geometric quantization. Ann. Inst. Fourier
(Grenoble), 59(1):199–238, 2009. doi:10.5802/aif.2429.
206 Johan Martens
[Rit01] Alvaro Rittatore. Very flat reductive monoids. Publ. Mat. Urug., 9:93–121
(2002), 2001.
[Sch04] Alexander Schmitt. The Hilbert compactification of the universal moduli
space of semistable vector bundles over smooth curves. J. Differential
Geom., 66(2):169–209, 2004. URL: http://projecteuclid.org/euclid.jdg/
1102538609.
[Sem51] John G. Semple. The variety whose points represent complete collineations
of Sr on Sr . Univ. Roma. Ist. Naz. Alta Mat. Rend. Mat. e Appl. (5),
10:201–208, 1951.
[Sol12] Pablo Solis. A wonderful embedding of the loop group, 2012.
arXiv:1208.1590.
[Sol13] Pablo Solis. A complete degeneration of the moduli of G-bundles on a
curve, 2013. arXiv:1311.6847.
[Tim03] D. A. Timashëv. Equivariant compactifications of reductive groups. Mat.
Sb., 194(4):119–146, 2003. doi:10.1070/SM2003v194n04ABEH000731.
[Vin95] E. B. Vinberg. On reductive algebraic semigroups. In Lie groups and Lie
algebras: E. B. Dynkin’s Seminar, volume 169 of Amer. Math. Soc. Transl.
Ser. 2, pages 145–182. Amer. Math. Soc., Providence, RI, 1995.
[Wei01] Jonathan Weitsman. Non-abelian symplectic cuts and the geo-
metric quantization of noncompact manifolds. Lett. Math. Phys.,
56(1):31–40, 2001. EuroConférence Moshé Flato 2000, Part I (Dijon).
doi:10.1023/A:1010907708197.
[Woo96] Chris Woodward. The classification of transversal multiplicity-free group
actions. Ann. Global Anal. Geom., 14(1):3–42, 1996. doi:10.1007/
BF00128193.
[Woo98] Chris Woodward. Multiplicity-free Hamiltonian actions need not be
Kähler. Invent. Math., 131(2):311–319, 1998. doi:10.1007/s002220050206.
The Serre-Swan Theorem for
Ringed Spaces
Archana S. Morye∗
Abstract
In this article we prove that if every locally free sheaf of bounded rank over a
ringed space X is acyclic and generated by finitely many global sections, then
the category of locally free sheaves of bounded rank over X is equivalent to
the category of finitely generated projective modules over the ring of its global
sections. This result is a generalization of the classical results of Serre for affine
schemes, and of Swan for paracompact topological spaces.
Mathematics Subject Classification (2000). Primary: 14F05; Secondary: 16D40,

18F20.
Keywords. Serre-Swan Theorem, vector bundles, projective modules
Introduction
We generalize two classical results of Serre and Swan on the relation between lo-
cally free sheaves and projective modules by emphasizing the axiomatic aspect
of the problem in the article [Mor2013].
For any ringed space (X, OX ), let OX -mod denote the category of OX -
modules, and Lfb(X) the full subcategory of OX -mod consisting of locally free
OX -modules of bounded rank. Let A = Γ(X, OX ), and Fgp(A) the category
of finitely generated projective A-modules. We will say that the Serre-Swan
Theorem holds for a ringed space (X, OX ) if for every F in Lfb(X), Γ(X, F)
is in Fgp(A), and the canonical functor Γ(X, •) : Lfb(X) → Fgp(Γ(X, OX ))
is an equivalence of categories. Let the sheaf of OX -morphisms from F to G be
denoted by HomOX (F, G). A full abelian subcategory C of OX -mod is called
∗ School of Mathematics and Statistics, University of Hyderabad, Hyderabad 500 046,
India. E-mail: [email protected]
Geometry, DOI 10.1007/978-981-10-5648-2_13
208 Archana S. Morye
an admissible subcategory if HomOX (F, G) belongs to C for every pair of sheaves

F in Lfb(X) and G in C, every sheaf in C is acyclic, and generated by global
sections, and Lfb(X) is a subcategory of C.
Theorem 1. [Mor2013, Theorem 2.1] Let (X, OX ) be a locally ringed space.
Then the Serre-Swan Theorem holds for a locally ringed space (X, OX ) if every
locally free OX -module of bounded rank is finitely generated by global sections,
and if the category OX -mod contains an admissible subcategory.
After reading the work presented in [Mor2011, Chapter2], Profes-
sor J. Oesterlé kindly pointed out that the hypothesis in [Mor2013, Theorem
2.1], of the existence of an admissible category can be replaced by the weaker
hypothesis that locally free sheaves of bounded rank are acyclic. We present a
proof of this remark in this article.
In Section 1, we recall some background material about sheaves. In order
to make the article self contained we recall definitions, facts, and proofs about
sheaves which are well known. In the second section we prove the main theorem
of the article.
Theorem 2. Let (X, OX ) be a locally ringed space such that each locally free
OX -module of bounded rank is acyclic and generated by finitely many global
sections, then the Serre-Swan theorem holds for the ringed space (X, OX ).
1. Theory of Sheaves
We briefly recall definitions and properties of sheaves of finite type, sheaves
of finite presentation, coherent sheaves, and locally free sheaves. Results and
proofs in this section are mainly taken from [Gro60].
1.1. The Sheaf of Morphisms of OX -modules. We will assume

that all rings are commutative unless otherwise mentioned. Let (X, OX ) be a
ringed space. For any ringed space (X, OX ), let OX -mod denote the category
of OX -modules. For any ring A, let A-mod denote the category of A-modules.
Let F and G be OX -modules. We will denote the Γ(X, OX )-module of ho-
momorphism of OX -modules F and G by HomOX (F, G). If u ∈ HomOX (F, G),
and U is an open subset of X, then uU : F(U ) → G(U ) denotes the homo-
morphism of OX (U )-modules induced by u. For every open subset U of X,
define
HomOX (F, G) (U ) = HomOX |U (F|U , G|U ).
If U ⊂ V , we have a canonical restriction map,

HomOX (F, G) (V ) −→ HomOX (F, G) (U ).
The Serre-Swan Theorem for Ringed Spaces 209
This defines a sheaf HomOX (F, G) called the sheaf of OX -morphisms from
F to G. There is a canonical structure of an OX -module on HomOX (F, G)
defined as follows. Let U be an open subset of X, and let u : F|U → G|U be
a morphism of OX |U -modules. Let f ∈ OX (U ). Define f u : F|U → G|U by
setting, for every open V ⊂ U ,
(f u)V : F(V ) → G(V ),
to be (f u)V (s) = f |V · uV (s), s ∈ F(V ).
Definition 1.1. For every point x ∈ X, there exists a canonical homomor-

phism of OX,x -modules
φx : (HomOX (F, G))x −→ HomOX,x (Fx , Gx )
defined as follows. Let α ∈ (HomOX (F, G))x , choose an open neighborhood U

of x, and a morphism of OX |U -modules, u : F|U → G|U , such that α equals the
germ of u at x. For each point y ∈ U , let uy : Fy → Gy denote the OX,y -module
homomorphism induced by u (stalk homomorphism). Define
φx (α) = ux .
It follows from this definition that φx (α) is independent of the choices of U and
u.
Note that the canonical homomorphism φx is functorial. The canonical ho-

momorphism φx is in general neither injective nor surjective.
Remark 1.2. Let F be a presheaf, and G a sheaf on a topological space X.
For each point x ∈ X, let αx : Fx → Gx be a map. Suppose that for every
x ∈ X, every open neighborhood U of x, and for every section s ∈ Γ(U, F),
there exist an open neighborhood V of x in U , and a section t ∈ Γ(V, G), such
that αy ((s)y ) = (t)y for all y ∈ V . Then, there exists a unique morphism of
presheaves u : F → G such that (u)x = αx for all x ∈ X.
We denote by C op the opposite category of a category C.
Proposition 1.3. Let (X, OX ) be a ringed space. For every OX -module G, the
functor
HomOX (•, G) : OX -modop −→ Γ(X, OX )-mod
is left exact. In particular, the functor
HomOX (•, G) : OX -modop −→ OX -mod
is left exact.
u v
Proof. We have to show that if F → F → F → 0 is an exact sequence of
OX -modules, then the sequence
HomOX (v,G) HomOX (u,G)
0 / HomO (F , G) / HomOX (F, G) / HomO (F , G)
X X
is exact. Let w ∈ HomOX (F , G) be such that HomOX (v, G)(w) = 0, that is,
for every x ∈ X, and θ ∈ Fx , wx ◦ vx (θ) = 0. Let θ ∈ Fx . Then, there exists
θ ∈ Fx such that vx (θ) = θ . Therefore, wx (θ ) = wx (vx (θ)) = 0. Hence,
wx = 0 for all x ∈ X, that is, w = 0. Hence, HomOX (v, G) is injective. Since
HomOX (•, G) is a functor,
HomOX (u, G) ◦ HomOX (v, G) = HomOX (v ◦ u, G) = 0.
This implies that Im(HomOX (v, G)) ⊂ Ker(HomOX (u, G)).

Now let g ∈ Ker(HomOX (u, G)). Define f x : F x → Gx as follows. For every
θ ∈ F x , there exists θ ∈ Fx such that vx (θ) = θ . Define, f x (θ ) = gx (θ).

Since g ∈ Ker(HomOX (u, G)), f x is well defined. Now it is easy to check the
condition in Remark 1.2, which gives a morphism f : F → G. For all x ∈ X,
and θ ∈ F x ,
(HomOX (v, G)(f ))x (θ) = fx ◦ vx (θ) = f x ◦ vx (θ) = gx (θ),
hence, HomOX (v, G)(f ) = g. Thus, Ker(HomOX (u, G)) ⊂ Im(HomOX (v, G)).
This proves that, HomOX (•, G) is left exact. The second statement follows from
the first by putting an arbitrary open subset of X in the place of X.
(I)
Let I be an arbitrary index set. Consider the direct sum OX = ⊕i∈I OX ,
(I)
and for each i ∈ I, let hi : OX → OX denote the canonical injective morphism
(I)
from the ith factor OX to OX . Then, for every OX -module F, we have a
homomorphism of A-modules,
(I)

φF : HomOX (OX , F) −→ Γ(X, F)I = Γ(X, F),
i∈I
u −→ ((u ◦ hi )X (1))i∈I .
The map φF is an isomorphism, whose inverse
(I)
ψF : Γ(X, F)I −→ HomOX (OX , F)
!
is given by ψF (s)U (f ) = i∈I fi si |U for s = (si )i∈I ∈ Γ(X, F)I , U ⊂ X
(I) (I)
open, and f = (fi )i∈I ∈ OX (U ). (Since f ∈ OX (U ), the above sum is a finite
sum, hence welldefined.)
Thus, we get
HomOX (OX , •) ∼
(I)
= Γ(X, •)I . (1)
(I)
We say that the morphism of OX -modules u = ψF (s) : OX → F is defined by
the family of sections s = (si )i∈I of F on X.
Remark 1.4. If I is a finite set, then by (1),
I
HomOX (OX , G) ∼
= Γ(X, G)I .
Let U be an open subset of X. Then, the above isomorphism of Γ(X, OX )-

modules gives an isomorphism of OX -modules
HomOX (OX
I
, G)(U ) = HomOX |U (OX
I
|U , G|U ) ∼
= Γ(U, G)I = G(U )I .
Thus, HomOX (OX

I
, G) ∼
= G I for every finite set I.
1.2. Sheaves of Finite Type.

Definition 1.5. An OX -module F is said to be of finite type if for every point
x ∈ X, there exists an open neighborhood U of x such that F|U is generated
by a finite family of sections of F on U .
The following properties of sheaves of finite type are easy to verify.
Proposition 1.6. Let (X, OX ) be a ringed space.
1. If u : F → G is a surjective morphism of OX -modules, and if F is of

finite type, then so is G. Thus, every quotient sheaf of a sheaf of finite
type is also of finite type.
2. The direct sum and the tensor product (over OX ) of a finite family of
sheaves of finite type are also of finite type.
Proposition 1.7. Let F be an OX -module of finite type. Let x ∈ X, and let

s1 , s2 , . . . , sn be sections of F on an open neighborhood U of x, such that the
n
family (si )x i=1 generates Fx . Then, there exists an open neighborhood V of
n
x in U , such that the family (si )y i=1 generates Fy for all y ∈ V .
Proof. Since F is of finite type, there exist an open neighborhood

p U1 of x in U ,
and sections t1 , . . . , tp ∈ Γ(U1 , F) such that the family (ti )y i=1 generates Fy
n
for all y ∈ U1 . Since the family (sj )x j=1 also generate Fx , there exist germs
θij ∈ OX,x such that

n
(ti )x = θij (sj )x (i = 1, . . . , p).
j=1
As the family (θij )i,j is finite, there exist an open neighborhood U2 of x in U1 ,

and sections fij ∈ Γ(U2 , OX ) such that θij = (fij )x for all i, j, and

n
(ti )x = (fij )x (sj )x (i = 1, . . . , p).
j=1
Therefore, there exists an open neighborhood V of x in U such that

n
(ti )|V = (fij )|V (sj )|V (i = 1, . . . , p),
j=1
!
hence, (ti )y = j (fij )y (sj )y for all y ∈ V . Therefore, (sj )y generates Fy for
all y ∈ V .
Let F be a sheaf of abelian groups on a topological space X. Recall that
the support of F is the set
Supp(F) = {x ∈ X | Fx = 0}.
Note that the support of a sheaf need not be a closed subset of X.

Corollary 1.8. If F is an OX -module of finite type, then Supp(F) is a closed
subset of X.
Proof. Let x ∈ X\Supp(F). Then, Fx = 0, hence the germ 0x of the zero section
0 ∈ Γ(X, F) generates Fx . Thus, by Proposition 1.7, there exists an open
neighborhood V of x such that 0y generates Fy , for all y ∈ V , that is, Fy = 0
for all y ∈ V . Therefore, V is an open neighborhood of x in X\Supp(F).
Corollary 1.9. Let F be an OX -module of finite type, and let u : F → G be a
morphism of OX -modules. Let x ∈ X and suppose that the stalk homomorphism
ux : Fx → Gx equals 0. Then, there exists an open neighborhood U of x such
that uy = 0 for all y ∈ V .
Proof. Let H = Im(u). Then H is an OX -submodule of G, and u induces an

exact sequence of OX -modules F
u /H / 0 . Since u is surjective, and
F is of finite type, by (1) of Proposition 1.6, H is of finite type. Since
u y = uy : Fy → Hy = Im(uy )
for all y ∈ X, we have Hy = 0 if and only if uy = 0. Thus,
Supp(H) = {y ∈ X | uy = 0}.
By Corollary 1.8, since x ∈ X\Supp(H), there exists an open neighborhood U

of x in X\Supp(H). We have uy = 0 for all y ∈ U .
Corollary 1.10. Let F and G be OX -modules of finite type. Let u : F → G

be a morphism of OX -modules, and x ∈ X. If ux : Fx → Gx is surjective,
then there exists an open neighborhood U of x, such that u|U : F|U → G|U is
surjective.
Proof. Since G is of finite type, Coker(u) = G/Im(u) is also of finite type
(Proposition 1.6, (1)). Hence, by Corollary 1.8, F = Supp(Coker(u)) is a closed
subset of X. If ux is surjective, the (Coker(u))x = Coker(ux ) = 0, hence x ∈
X\F . Therefore, U = X\F is an open neighborhood of x such that u|U :
F|U → G|U is surjective.
Corollary 1.11. Let F be an OX -module of finite type. Then, for every OX -
module G, and for every point x ∈ X, the canonical homomorphism of OX,x -
modules (Definition 1.1)
is injective.
Proof. Let α ∈ (HomOX (F, G))x , and suppose that φx (α) = 0. Let U and u be
as in Definition 1.1. Then ux = φx (α) = 0, so by Corollary 1.9, there exists an
open neighborhood V of x in U such that uy = 0 for all y ∈ V . We must show
that u|V : F|V → G|V is equals 0, which will imply that α = 0. Let W be an
open subset of V . Consider the homomorphism

u|V W = uW : F(W ) → G(W ).
Let s ∈ F(W ). Then, (uW (s))y = uy ((s)y ) = 0 for all y ∈ W . Since G is a

sheaf, this implies that uW (s) = 0. Therefore, uW = 0 for all W open in V ,
hence u|V = 0.
1.3. Sheaves of Finite Presentation.

Definition 1.12. An OX -module F is said to be is of finite presentation if for
every point x in X, there exist an open neighborhood U of x, and an exact
sequence of OX |U -modules
p u q v
OX |U → OX |U → F|U → 0
Proposition 1.13. Let F be an OX -module of finite presentation, and let

x ∈ X. Then, for every OX -module G, the canonical homomorphism of OX,x -
modules (Definition 1.1)
is an isomorphism.
Proof. Let U be an open neighborhood of x, on which there exists an exact

sequence of OX |U -modules
p uq v
OX | U → OX |U → F|U → 0.
Since the proposition is local with respect to x, we will replace X by U , so
X = U . By Proposition 1.3, HomOX (•, G) is a left exact functor, so we get an
exact sequence of OX -modules
q p
0 → HomOX (F, G) → HomOX (OX , G) → HomOX (OX , G).
This induces an exact sequence of OX,x -modules
q p
0 → (HomOX (F, G))x → (HomOX (OX , G))x → (HomOX (OX , G))x .
Since
p
OX,x
ux
/ OX,x
/ Fx
q / 0 is an exact sequence of OX,x -
vx
modules, and since HomOX,x (•, Gx ) is a left exact functor, we get an exact
sequence
q p
0 → HomOX,x (Fx , Gx ) → HomOX,x (OX,x , Gx ) → HomOX,x (OX,x , Gx ).
Since the canonical homomorphism φx is functorial, we have a commutative
diagram
0 / (HomOX (F, G)) / (HomOX (Oq , G)) / (HomOX (Op , G))

x X x X x
φx φx φx

0 / HomOX,x (Fx , Gx ) / HomOX,x (OX,x
q
, Gx ) / HomOX,x (OX,x
p
, Gx )
By Remark 1.4 there exists a canonical isomorphism HomOX (OX q

, G) ∼
= Gq.
Thus, the modules in the second column of the above diagram are identifies
with Gxq and HomOX,x (OX,x q
, Gx ) ∼
= Gxq respectively, and the second vertical
arrow is an isomorphism. For the same reasons, the third vertical arrow is also
an isomorphism. Since the rows of the above diagram are exact, it follows that
the first vertical arrows also is an isomorphism.
Corollary 1.14. Let F and G be OX -modules of finite presentation, and let
x ∈ X. Suppose that f : Fx → Gx is an isomorphism of OX,x -modules. Then
there exist an open neighborhood V of x, and an isomorphism of OX |V -modules
u : F|V → G|V such that ux = f .
Proof. Let g : Gx → Fx be the inverse of f . By Proposition 1.13, there exist
an open neighborhood U of x, and sections u ∈ Γ(U, HomOX (F, G)) and v ∈
Γ(U, HomOX (G, F)) such that ux = f and vx = g. Because
(u ◦ v)x = ux ◦ vx = f ◦ g = 1Gx and (v ◦ u)x = vx ◦ ux = g ◦ f = 1Fx
Proposition 1.13 again implies that the germs, at x, of u ◦ v and v ◦ u equal

those of the corresponding identity morphisms. Therefore, there exist an open
neighborhood U of x in V such that
(u ◦ v)|U = 1G|U and (v ◦ u)|U = 1F |U .
Thus, u|U : F|U → G|U is an isomorphism.

Remark 1.15. [Gro60, Chap. 0, 5.2.7, pp. 46-47] Let F and G be OX -modules
of finite presentation. Let x ∈ X, and suppose that the OX,x -modules Fx and
Gx are isomorphic. Then, there exists an open neighborhood U of x such that
the OX |U -modules F|U and G|U are isomorphic. This immediately follows from
Corollary 1.14.
Lemma 1.16. Let (X, OX ) be a ringed space, and let F be an OX -module
of finite presentation. Let x ∈ X, and suppose that Fx is a free OX,x -module.
Then there exist an open neighborhood U of x, and n ∈ N, such that F|U is
isomorphic to OXn
|U .
Proof. Since Fx is a free OX,x -module, there exists an integer n ∈ N such that
OX,x -modules OX,x
n
and Fx are isomorphic. All locally free sheaves of finite type
are of finite presentation, hence OXn
is of finite presentation. From Remark 1.15
there exists an open neighborhood U of x such that F|U ∼ = OX n
|U .
1.4. Locally Free Sheaves.

Definition 1.17. Let (X, OX ) be a ringed space. We say that an OX -module
F is locally free if for every point x ∈ X, there exist an open neighborhood U
of x, and a set I, such that F|U ∼
(I)
= OX |U as an OX |U -module.
If A is a nonzero ring, and M a free A-module, any two bases have the
same cardinality [Bou89, Corollary to Proposition 3, p. 294], and this common
cardinality is called the rank of M .
Remark 1.18. Let X be a ringed space such that Supp(OX ) = X. Let F be
a locally free OX -module, and let x ∈ X. Then, Fx is a free OX,x -module.
Since OX,x = 0, the above paragraph implies that the rank of Fx (as a free
OX,x -module) is well defined, we denote it by rkx (F). The definition implies
that there exist an open neighborhood U of x, such that
rky (F) = rkx (F), for all y ∈ U.
Proposition 1.19. Let (X, OX ) be a ringed space. Let F be an OX -module of

finite presentation, and let x ∈ X. Suppose that Fx is a free OX,x -module of
rank n. Then, n is finite, and there exists an open neighborhood U of x, such
that F|U is a locally free (OX |U )-module of rank n.
Proof. Since Fx is a finitely generated free OX,x -module, the rank n of Fx is

finite. Let u : Fx → OX,x
n
be an isomorphism of OX,x -modules. Since F and
OX are of finite presentation, Corollary 1.14 implies that there exist an open
n
neighborhood U of x, and an isomorphism of (OX |U )-modules φ : F|U → OX n

|U
such that φx = u.
Let (X, OX ) be a locally ringed space, and let x ∈ X. If F is any OX -
module, we define,
F(x) = Fx ⊗OX,x k(x) = Fx /mX,x Fx ,
where, mX,x is the maximal ideal in OX,x , and k(x) the residue field of X at
x. We call F(x) the fibre of F at x, it is a k(x)-vector space. The rank of F at
x is defined by
rkx (F) = dimk(x) F(x).
If rkx (F) = n is independent of x, we say that F is of rank n. If rkx (F) is
finite for all x ∈ X, we say that F is of finite rank. If {rkx (F) | x ∈ X} < ∞,
then we call F is of bounded rank.
Remark 1.20. Let (X, OX ) be a locally ringed space, and let F be a locally
∼ O(I) .
free OX -module. For x ∈ X, then there exists a set I such that Fx = X,x
Hence, we get,
F(x) = Fx ⊗OX,x k(x) ∼

(I)
= OX,x ⊗OX,x k(x) = k(x)(I) .
This implies that, dimk(x) F(x) equals the rank of the free OX,x -module Fx .
Hence the definition of the rank of F as defined in Remark 1.18 is the same as
that of the definition of the rank of F as an OX -module over a locally ringed
space (X, OX ).
Let Lfb(X) denotes the full subcategory of OX -mod consisting of locally
free OX -modules of bounded rank.
Proposition 1.21. Let (X, OX ) be a locally ringed space, and let F be an
OX -module of finite type. Then, the following are equivalent:
1. F is locally free.
2. F is of finite presentation, and for every point x ∈ X, Fx is a free OX,x -
module.
3. For every point x ∈ X, Fx is a free OX,x -module, and the function X →
N, x → rkx (F) is locally constant.
Proof. For any ringed space (1) implies (2) is clearly true. By Proposition 1.19,
(2) implies (1) for any ringed space.
By Remark 1.18 and Remark 1.20 (1) implies (3). Therefore, it remains to
prove (3) implies (1). Let z ∈ X, and let n = rkz (F). By hypothesis, there
exists an open neighborhood U of z, such that rkz (F) = n for all x ∈ U .
Let {θ1 , . . . , θn } be a basis of the free OX,z -module Fz . Replacing U by a
smaller set, we may assume there exist sections s1 , . . . , sn ∈ Γ(U, F) such that
(si )z = θi . Let u : OX n
|U → F|U be the morphism of OX |U -modules defined
by the si . Then, uz : OX,z n
→ Fz is an isomorphism. By Corollary 1.10, there
exists an open neighborhood V of z in U such that u|U : OX n
|U → F|U is a
surjective morphism. Therefore, for all x ∈ V , ux : OX,x n
→ Fx is a surjective.
Since Fx is free of rank n, there exists an isomorphism φx : Fx → OX,x n
.
The map φx ◦ ux : OX,x n
→ OX,x
n
is a surjective endomorphism of OX,xn
. By
[Eis95, Corollary 4.4, p. 120], φx ◦ ux is an isomorphism. Therefore, ux is an
isomorphism for all x ∈ V , hence u|U : OX n
|U → F|U is an isomorphism. This
shows that F is locally free.
2. The Serre-Swan Theorem

In this section we will prove the main theorem of this article. We will use results
from Section 1 to prove the theorem.
2.1. The Functors Γ(X, •) and S. Let (X, OX ) be a ringed space

and A denote the ring Γ(X, OX ) of global sections of OX .
We have a canonical functor
Γ(X, •) : OX -mod −→ A-mod,
defined by, Γ(X, •)(F) = Γ(X, F), and for φ ∈ HomOX (F, G),
Γ(X, •)(φ) = φX : Γ(X, F) → Γ(X, G).
Recall that Γ(X, •) is a left exact functor.

Let M be an A-module. Define a presheaf P(M ) on X by P(M )(U ) =
M ⊗A OX (U ) for every open set U of X. Note that, OX (U ) is canonically an
A-module. Indeed if f ∈ A, and s ∈ OX (U ), then f · s = f |U s ∈ OX (U ).
Therefore, P(M )(U ) is an OX (U )-module. We will denote by S(M ) the sheaf
associated to the presheaf P(M ). Similarly for a homomorphism u : M → N ,
and for every U ⊂ X open, define (P(M )(u))U = u ⊗ 1OX (U ) . This morphism
of presheaves induces a morphism of sheaves S(u) : S(M ) → S(N ). Thus, we
get a functor
S : A-mod −→ OX -mod. (2)
Since for every x in X, the functor • ⊗A OX,x is right exact, the functor S is
right exact.
Remark 2.1. If P is a finitely generated projective A-module, then S(P ) =

P(P ), that is, P(P ) is a sheaf. This follows from the fact that, the tensor prod-
uct by P over A commutes with projective limits, since P is finitely generated
projective module. Hence, the canonical A-linear map h → h ⊗ 1A from P to
Γ(X, S(P )) is an isomorphism.
Proposition 2.2. For any ringed space (X, OX ), the functor S is a left adjoint
of Γ(X, •).
Proof. Let F be an OX -module, and M an A-module. Let u : M → Γ(X, F)
be a homomorphism of A-modules. Consider a homomorphism of presheaves,
such that for every open subset U of X,
λu : M ⊗A OX (U ) −→ F (U )
is given by λu (m ⊗A f ) = f · u(m)|U , for m ∈ M , and f ∈ OX (U ). Let λF ,M (u)

be the morphism of sheaves associated to λu . Thus, we get a map,
λF ,M : HomA (M, Γ(X, F)) −→ HomOX (S(M ), F). (3)

We will prove that λ : A-modop × OX -mod → Set, given by M × F → λF ,M
is an adjunction between Γ(X, •) and S, where Set denotes the category of
sets. So we have to check λ is functorial in F and M , and λF ,M is a bijection.
We must show that for every g : M → M and ψ : F → F , the diagram
λF ,M
HomA (M, Γ(X, F)) / HomOX (S(M ), F) (4)
Hom(g,Γ(X,ψ)) Hom(S(g),ψ)

HomA (M , Γ(X, F )) / HomO (S(M ), F )
X
λF ,M
commutes. Let u : M → Γ(X, F), then
Hom(S(g), ψ) ◦ λF ,M (u) = ψ ◦ λF ,M (u) ◦ S(g) : S(M ) → F , (5)
and
λF ,M ◦ Hom(g, ψ)(u) = λF ,M (Γ(X, ψ) ◦ u ◦ g) : S(M ) → F . (6)
It is enough to check that, this two morphism agrees on the presheaf P(M ).
Let U be an open subset of X, and m ⊗ s ∈ P(M )(U ) = M ⊗ OX (U ). Then,

ψU ◦ λF ,M (u) ◦ S(g)(m ⊗ s) = ψU ◦ λu (g(m ) ⊗ s) = ψU s · u(g(m ))|U

= s · Γ(X, ψ)(u ◦ g(m )) |U
since ψ is a sheaf morphism. Hence, ψU ◦ λF ,M (u) ◦ S(g)(m ⊗ s) =

λF ,M (Γ(X, ψ)◦u◦g)(m ⊗ s). Thus, the two morphisms (5) and (6) agree, that
is, the diagram (4) commutes. Hence, λ is functorial in F and M . It remains
to check that λF ,M is a bijection. Define
φF ,M : HomOX (S(M ), F) −→ HomA (M, Γ(X, F)) (7)
by φF ,M (u)(m) = uX (m ⊗ 1) for u : S(M ) → F, and m ∈ M . Now it

is easy to show that λF ,M ◦ φF ,M = 1HomOX (S(M ),F ) , and φF ,M ◦ λF ,M =
1HomA (M,Γ(X,F )) . Hence, λF ,M is a bijection.
Definition 2.3. Recall that an OX -module F is said to be generated by global
sections if there is a family of sections (si )i∈I in Γ(X, F) such that for each
x ∈ X, the images of si in the stalk Fx generate that stalk as an OX,x -module.
We will say that F is finitely generated by global sections if a finite family of
global sections (si )i∈I exists with the above property.
Proposition 2.4. Let (X, OX ) be a ringed space, and an OX -module F
generated by global sections. Let w : S(Γ(X, F)) → F be such that
λF ,Γ(X,F ) (1Γ(X,F ) ) = w (that is, w is the counit morphism of F with respect
to the adjunction λ (3)). Then w is surjective.
Proof. Let M = Γ(X, F). Since w is counit morphism of F with respect to the
adjunction λ, wx : M ⊗A OX,x → Fx is such that
n
n
wx si ⊗A ai = ai · (si )x ,
i=1 i=1
for x ∈ X, ai ∈ OX,x , and si ∈ M , i = 1, . . . , n. Since w is a morphism of

sheaves to prove that w is surjective, it is enough to prove that wx is surjective
for every x in X.
The sheaf F is generated by global sections, therefore any germ β belongs
to Fx can be written as

n
β= ai · (si )x , for some ai ∈ OX,x , and si ∈ M, i = 1, . . . , n.
i=1
!n
Define α = i=1 si ⊗A ai . Then
n

n
wx (α) = wx si ⊗A ai = ai · (si )x = β.
i=1 i=1
Therefore, wx is surjective.
Remark 2.5. The functor Γ(X, •) is in general not fully faithful. Let X be a
compact Riemann surface, and let OX be a sheaf of holomorphic functions over
X. Then Γ(X, OX ) = C. Let L be a line bundle on X, with negative degree.
Then Γ(X, L) = 0, so is End(Γ(X, L)). On the other hand 1L : L → L and
0 : L → L are two distinct morphism of L, hence, EndOX (L) → EndC (Γ(X, L))
is not injective, and Γ(X, •) is not fully faithful. But under certain conditions
restriction of Γ(X, •) to a subcategory of OX -mod is fully faithful.
Proposition 2.6. [Mor2013, Proposition 2.2] Let (X, OX ) be a ringed space,
and let C be a full abelian subcategory of OX -mod, such that OX belongs to
C. Suppose that every sheaf in C is generated by global sections. Then Γ(X, •) :
C → Γ(X, OX )-mod is fully faithful.
2.2. Main Theorem. Let (X, OX ) be a ringed space, and A denote the
ring Γ(X, OX ). Let Fgp(A) denote the full subcategory of A-mod consisting
of finitely generated projective A-modules.
Recall that, we say that the Serre-Swan Theorem holds for a ringed space
(X, OX ) if Γ(X, F) is a finitely generated projective module for every sheaf F
in Lfb(X), and the functor
Γ(X, •) : Lfb(X) → Fgp(Γ(X, OX ))
is an equivalence of categories. In this subsection we will prove that, if each

locally free OX -module of bounded rank is acyclic and generated by finitely
many global sections, then the Serre-Swan theorem holds for the ringed space
(X, OX ). To prove the theorem we will need some preliminary results.
The following lemma is from [Mor2013], with a slight change in the hypoth-
esis.
Lemma 2.7. Let (X, OX ) be a locally ringed space, and let A denote the ring
Γ(X, OX ). Then for every finitely generated projective A-module P , the sheaf
S(P ) is a locally free OX -module of bounded rank.
Proof. It is given that, an A-module P is finitely generated and projective, hence
it is of finite presentation. Therefore, we get an exact sequence of A-modules
Ap → Aq → P → 0, for some p, q ∈ N.
Since the functor S is right exact, and S(Am ) ∼

= OX
m
for all m ∈ N, we get an
exact sequence of OX -modules
p q
OX → OX → S(P ) → 0.
This shows that S(P ) is of finite presentation. Since for every x ∈ X, OX,x is a
local ring, and P is a finitely generated projective module P ⊗A OX,x is a free
OX,x -module of finite rank. We denote the rank rkx (S(P )) of the sheaf S(P )
at every point x in X by nx . Therefore, P ⊗A OX,x is isomorphic to OX,x
nx
. Now
by Lemma 1.16 S(P ) is a locally free OX -module. Also the family of integers
(nx )x∈X is bounded above by q, so the sheaf S(P ) is of bounded rank.
Proposition 2.8. Let (X, OX ) be a locally ringed space, and let A denote the
ring Γ(X, OX ). The functor S : Fgp(A) → Lfb(X) is fully faithful.
Proof. By Lemma 2.7, if P is in Fgp(A) then S(P ) is a locally free OX -module
of bounded rank. Let P and Q be two finitely generated projective A-modules.
The map S : HomA (P, Q) → HomOX (S(P ), S(Q)), u → S(u) is obtained by
composing the two canonical isomorphisms
HomA (P, Q) → HomA (P, Γ(X, S(Q))) → HomOX (S(P ), S(Q))
(second isomorphism is given by the adjunction (3)).
Lemma 2.9. Let (X, OX ) be a locally ringed space. Let F and G be two locally
free OX -modules of bounded rank. If u : F → G is a surjective homomorphism,
ker(u) is a locally free OX -module of bounded rank.
u
Proof. We have the exact sequence 0 → ker(u) → F → G → 0. By choosing
local frames for F and G, we can see that the above exact sequence locally
splits. Hence, ker(u) is of finite type. By Corollary 1.10, the rank of the sheaf
ker(u) is locally constant, hence is locally free of bounded rank (Proposition
1.21).
Now we are ready for giving proof of the main theorem.
Theorem 2.10. Let (X, OX ) be a locally ringed space and A = Γ(X, OX )
its ring of global sections. Assume that each locally free OX -module of bounded
rank is acyclic, and generated by finitely many global sections. Then the functor
S defines an equivalence of categories from the category Fgp(A) to the category
Lfb(X). A quasi-inverse is the canonical functor Γ(X, •).
Proof. We already know by Proposition 2.8 and Remark 2.1 that, the functor
S is fully faithful, and that Γ(X, S(P )) is canonically isomorphic to P when
P is a finitely generated projective A-module. Hence, it suffices to show that
the functor S is essentially surjective, that is, each locally free OX -module F
of bounded rank is isomorphic to S(P ) for some P in Fgp(A).
Since F is finitely generated by global sections, there exists a surjective
morphism u : OX n
→ F for some n ∈ N. Hence, by Lemma 2.9, ker(u) is in
Lfb(X), so is acyclic. Thus, the map Γ(X, u) : An → Γ(X, F) is surjective.
For all x ∈ X, HomOX,x (Fx , OX,x
n
) → HomOX,x (Fx , Fx ) → 0 is surjective,
this follows from the fact that Fx is free. Since F is of finite presentation by
Proposition 1.13,
(HomOX (F, OX
n
))x → (HomOX (F, F))x → 0 (8)
is exact, that is, Hom(F, u) is a surjective homomorphism between

HomOX (F, OX n
) and HomOX (F, F). Also we have that HomOX (F, OX n
) and
HomOX (F, F) are locally free sheaves of bounded rank and hence are acyclic
by the hypothesis. Applying Γ(X, •) to (8) we get the homomorphism
Hom(F, u) : HomOX (F, OX

n
) → HomOX (F, F)
is surjective. Hence there exists v : F → OX n

such that u ◦ v = 1F . We then
have Γ(X, u) ◦ Γ(X, v) = 1Γ(X,F ) . This prove that Γ(X, u) has a section, and
therefore P = Γ(X, F) is a projective A-module of finite rank.
Let w : S(P ) → F be such that λF ,Γ(X,F ) (1Γ(X,F ) ) = w. Then by Proposi-
tion 2.4, w is surjective, since F is generated by global section. By Lemma 2.9,
G = ker(w) is a locally free OX -module of bounded rank. Since Γ(X, •) is left ex-
act we get that ker(Γ(X, w)) = Γ(X, G). But Γ(X, w) : Γ(X, F) ⊗ A → Γ(X, F)
is by definition 1Γ(X,F ) ⊗ 1A , hence an isomorphism. Thus, ker(Γ(X, w)) = 0 =
Γ(X, G). But by hypothesis G is generated by global section, hence G = 0. This
proves that F is isomorphic to S(P ).
Remark 2.11. In Theorem 2.10, we can replace the assumption that every ob-
ject in Lfb(X) is acyclic, with the weaker hypothesis that for every surjective
homomorphism u : F → G, where F and G belongs to Lfb(X), the homo-
morphism Γ(X, u) : Γ(X, F) → Γ(X, G) is surjective. That this hypothesis is
indeed weaker follows from Lemma 2.9
Remark 2.12. By [For67, Satz 6.7 and Satz 6.8] or [Mor2013, Corollary 3.7],
if (X, OX ) be a finite dimensional connected Stein space then the Serre-Swan
theorem holds of X. Every non-compact connected Riemann surface is a Stein
space [GR79, Chapter V, §5, p. 134], hence the Serre-Swan Theorem holds
for non-compact Riemann surfaces. Note that every vector bundle over a non-
compact connected Riemann surface (X, OX ) is holomorphically trivial [For91,
Chapter 3, §30, Theorem 30.4, p. 229]. Let A = Γ(X, OX ). Therefore, if F is
in Lfb(X), then Γ(X, F) ∼ = An for some n ∈ N ∪ {0}. Conversely, let P be a
finitely generated projective A-module. Then, S(P ) is in Lfb(X), and hence,
Γ(X, S(P )) ∼=P ∼ = An for some n. This implies that every projective A-module
is free.
Note that every projective module over local ring or polynomial ring is free.
But the ring of holomorphic functions over non-compact connected Riemann
surface, is different in nature from these rings. This ring has many maximal
ideals [NN2004, p. 133], and is not Nötherian ring [NN2004, Remark p. 132].
But still every projective module over A is free.
On the other hand, for compact Riemann surfaces the Serre-Swan theorem
does not hold. Let (X, OX ) be a compact Riemann surface, and let L be a
line bundle over X of negative degree. Then the zero OX -module 0 and L are
non-isomorphic vector bundles over X. But, Γ(X, 0) and Γ(X, L) are 0, that
is isomorphic Γ(X, OX ) = C-modules. Hence, the functor Γ(X, •) : Lfb(X) →
A-mod is not faithful.
Acknowledgment
The author would like to thank Prof. N. Raghavendra for suggesting the prob-
lem and for helping with it. The author is grateful to Prof. Joseph Oesterlé,
who helped through his remarks on the first draft of her thesis. This article is
the result of the remark which he made.
References
[Bou89] N. Bourbaki, Algebra I: Chapters 1-3, Elements of Mathematics, Trans-
lated from the French, Springer-Verlag, 1989.
[Eis95] D. Eisenbud, Commutative algebra with a view towards algebraic geometry,
Springer, New York, 1995.
[For67] O. Forster, Zur Theorie der Steinschen Algebren und Moduln, Math. Z.
97(1967), 376–405.
[For91] O. Forster, Lectures on Riemann Surfaces, Translated from the German by
Bruce Gilligan, Graduate Texts in Mathematics, no. 81, Springer-Verlag,
New York, 1991.
[GR79] H. Grauert and R. Remmert, Theory of Stein spaces, Translated from the
German by Alan Huckleberry, Springer, Berlin, 1979.
[Gro60] A. Grothendieck, Éléments de géométrie algébrique. I. Le langage des
schémas, Inst. Hautes Études Sci. Publ. Math. No. 4 (1960), 228 pp.
[Mor2011] A. Morye, On the Serre-Swan theorem, and on vector bundles over real
abelian varieties, Thesis, Homi Bhabha National Institute.
[Mor2013] A. Morye, Note on the Serre-Swan Theorem, Mathematische Nachrichten,
286(2013), Issue 2-3, 272–278.
[NN2004] R. Narasimhan and Y. Nievergelt, Complex analysis in one variable, Sec-
ond edition, Birkhäuser, 2004.
An Extension Theorem for Hermitian
Line Bundles
Georg Schumacher∗
Abstract
We prove a general extension theorem for holomorphic line bundles on reduced
complex spaces, equipped with singular hermitian metrics, whose curvature
currents can be extended as positive, closed currents. The result has applications
to various moduli theoretic situations.
1. Introduction
Our aim is to prove an extension theorem for holomorphic line bundles L that
are defined on the complement Y of an analytic subset A ⊂ Y in a reduced,
complex space of pure dimension n say. We do not have to assume that the given
space is projective. However, if we do, an essential step is to solve the extension
problem for complex analytic hypersurfaces H ⊂ Y . For this situation classical
results are known: By the Remmert-Stein theorem [R-St] the closure of H is
an analytic subset of Y , if the dimension of A is smaller than n − 1 everywhere,
and Shiffman’s theorem [Sh1] states that it is sufficient to require that the
2n − 3-dimensional Hausdorff measure of A is equal to zero. More general is
Bishops theorem [B], which only requires that for suitable neighborhoods U of
points of A the set H ∩ (U \A) has finite volume.
An analytic aspect of the extension problem for complex hypersurfaces H
is given by the induced current of integration [H ], which is a closed positive
current of degree (1, 1). Here we assume that Y is smooth. Now, if an extension
of the current of integration [H ] to Y as a positive closed current T is known
to exist, then Siu’s theorem [Siu2] implies that the locus, where the Lelong
numbers of T are positive, yields an extension of the hypersurface H to Y .
∗ Fachbereich Mathematik und Informatik, Philipps-Universität Marburg, Lahnberge,
Hans-Meerwein-Straße, D-35032 Marburg, Germany.

Geometry, DOI 10.1007/978-981-10-5648-2_14
226 Georg Schumacher
In this sense a natural generalization of the extension problem for analytic

hypersurfaces concerns positive, closed (1, 1)-currents on complex manifolds.
Various results are known: Shiffman proved in [Sh2] that any semi-positive
closed (1, 1)-form of class C ∞ can be extended as such into an analytic subset
of codimension greater or equal to two. Skoda’s theorem [Sk] implies that any
positive closed (1, 1)-current T on Y can be extended as such, if it is locally
of finite mass near points of A. We also mention the extension theorems for
positive currents by El Mir [EM] and Sibony [Si].
The above reasoning suggests considering the curvature current of the given
line bundle that is defined on a reduced not necessarily projective space. We
denote by h a singular hermitian metric on L over the reduced space Y .
Before turning to singular hermitian metrics and curvature currents, we
consider the smooth case: For a complex manifold Y containing an analytic
subset A of codimension at least two with complement Y , and a holomorphic
line bundle (L , h ) on Y with smooth hermitian metric h of semi-positive
curvature Shiffman shows the extension property in [Sh2]. The question is about
how to treat loci A of codimension one and singularities of Y . We show the
following theorem.
Theorem 1. Let Y be a reduced complex space and A ⊂ Y a closed ana-

lytic subset. Let (L , h ) be a holomorphic line bundle on Y = Y \A equipped
with a singular hermitian metric whose curvature current ωY = 2πc1 (L , h )
is positive. We assume that there exists a desingularization of Y such that the
pull-back of ωY extends as a positive, closed (1, 1)-current.
Then there exists a modification τ : Z → Y , which is an isomorphism over
Y such that (L , h ) extends to Z as a holomorphic line bundle equipped with
a singular hermitian metric, whose curvature form is a positive current.
Even though the hermitian metric h is being extended, in general the cur-
vature current of the extension is different from the given extension of the
curvature current of h .
The above extension theorem applies to moduli theoretic situations, where
certain determinant line bundles exist, whose curvature forms are natural Käh-
ler forms on the respective moduli spaces. Typical examples are moduli spaces
of canonically polarized manifolds (including Riemann surfaces of genus greater
than one) and moduli spaces of stable vector bundles on projective varieties.
The Kähler-Einstein and Hermite-Einstein metrics resp. on the fibers induce
distinguished Kähler metrics on the moduli spaces, namely the Weil-Petersson
metrics and L2 metrics resp. These distinguished Kähler forms are known to
be equal to the curvature forms of the Quillen metrics on certain determinant
line bundles up to a numerical constant (cf. [F-S, Sch], [B-S1, B-S2]). Also for
classical Hurwitz spaces this construction was carried out [A-B-S].
An Extension Theorem for Hermitian Line Bundles 227
On the other hand, compactifications of moduli spaces in the category of

complex spaces were known to exist in the category of algebraic spaces (cf.
[M-F-K]), and a natural question is to extend these positive holomorphic line
bundles to suitable compactifications. In the above mentioned cases we showed
explicitly that the Weil-Petersson forms could be extended as positive currents.
Our aim is to give a detailed proof of the extension theorem (cf. also [Sch]).
We begin with some comments on Theorem 1. By definition positive cur-
rents on reduced complex spaces are currents on the normalization. (If the
given metric is of class C ∞ , positivity of the curvature current corresponds to
semi-positivity of the curvature form by definition as usual.) In this sense we
will prove the theorem first under the extra hypothesis that Y is normal. If
Y is just a reduced complex space, we pull back the line bundle to the nor-
malization, and extend it. It is possible to descend the line bundle to a proper
modification space over Z (given by a finite map) that is biholomorphic over
Y by Theorem 2 below.
The first essential step is to treat the smooth divisorial part of A as far as
it is contained in the regular locus of Y and use a monodromy argument (cf.
Proposition 3). Both, the holomorphic line bundle and the singular hermitian
metric are extended into this locus.
Closed, positive (1, 1)-currents on normal complex spaces in general do not
possess ∂∂-potentials (unless these are already given as curvature currents of
singular hermitian metrics). Also, potentials are required, in order to use a pull-
back of such a current under a holomorphic map. A desingularization of Y is
necessary in order to apply the above mentioned methods, and the extension of
the curvature current to a desingularization is the weakest possible assumption
and most easy to check, hence suitable for moduli theoretic applications.
Descending both the holomorphic line bundle and the singular hermitian
metric from a desingularization poses an extra difficulty: The push-forward of
the curvature current need not possess a ∂∂-potential and the direct image of
the invertible sheaf may only be invertible after applying Hironaka’s flattening
theorem. This is taken into account in the proof of the main theorem.
2. Main Theorem for Complex Manifolds

Grauert’s Oka principle states that on a Stein space any topologically trivial
holomorphic vector bundle is holomorphically trivial. This fact contains the
statement that on non-compact Riemann surfaces all holomorphic line bundles
are trivial. We will need the following classical fact.
Lemma 1. All holomorphic line bundles on Δ∗ × Δk are trivial, where Δ and

Δ∗ resp. denote the unit disk and the punctured unit disk resp.
We mention that line bundles on the product of two punctured disks are not
necessarily trivial, and Lemma 1 only provides local extensions of holomorphic
line bundles from Δ∗ × Δ∗ to Δ2 \{(0, 0)}. (Once a line bundle is extended to
Δ2 \{(0, 0)} as a trivial line bundle, it can obviously be extended to Δ2 .)
We will consider holomorphic line bundles L equipped with a √ possibly singu-
lar hermitian metric h on a normal complex space. We will call −1Θ(L, h) =
2πc1 (L, h) the (real) curvature current. Since the curvature current possesses
locally a ∂∂-potential that is (locally) integrable, pull-backs of the curvature
current under holomorphic maps are well-defined.
Our result will be proven under the further assumption that X is normal.
The general case will be treated in the last section.
Note that on normal complex spaces we use the notion of positivity for
(1, 1)-currents that possess local ∂∂-potentials.
The core of the Theorem is the following version for manifolds and smooth
hypersurfaces. We will identify holomorphic line bundles and invertible sheaves
and use the same letter for both.
Proposition 1. Let Y be a complex manifold and let D ⊂ Y be a closed

analytic smooth hypersurface. Let (L , h ) be a holomorphic line bundle on Y =
Y \D with a singular hermitian metric, whose curvature current is positive and
extends to Y as a positive closed current ω. Then (L , h ) can be extended to Y
as a holomorphic line bundle (L, h) equipped with a singular hermitian metric,
whose curvature current is positive. The latter current differs from ω by the
sum over currents of integration over the components of D with non-negative
coefficients smaller than 2π.
Observe that we do not assume that D is connected.

We state a well-known fact.
Lemma 2. Let Y a complex manifold and L1 and L2 holomorphic line bundles

on Y , equipped with singular hermitian metrics h1 and h2 . Let A ⊂ Y be a
closed analytic subset, whose codimension is at least 2 in any point. Assume
that there exists an isomorphism α : (L1 , h1 )|Y \A → (L2 , h2 )|Y \A. Then α
can be extended to Y in a unique way.
Proof. The bundle L1 ⊗ L−1 2 possesses a nowhere vanishing section over Y \A

that can be extended holomorphically to Y with no zeroes. Locally with re-
spect to Y the hermitian metrics hj are of the form e−ϕj for locally integrable
functions ϕj , j = 1, 2, whose difference must be pluriharmonic.
Corollary 1. The statement of Proposition 1 holds, if D is a simple normal

crossings divisor.
!k
Proof of Corollary 1. Let D = j=1 Dj be the decomposition, where the Dj
#
are the smooth irreducible components of D. Let Y = Y \ j=1 Dj . We apply
Proposition 1 to Yk ⊂ Yk−1 and extend (L , h ) to Yk−1 . In the process the
current ω|Yk−1 is being replaced by (ω + γk [Dk ])|Yk−1 , 0 ≤ γk < 2, which
obviously extends to Y0 = Y . By induction over we arrive at = 0, which
comprises the claim.
Proof of Proposition 1. Let {Uj } be an open covering of Y such that the set
D ∩ Ui consists of the zeroes of a holomorphic function zi on Ui . Since all
holomorphic line bundles on the product of a polydisk and a punctured disk
are trivial by Lemma 1, we can chose the sets {Ui } such that the line bundle
L extends to such Uj as a holomorphic line bundle. So L possesses nowhere
vanishing sections over Uj = Uj \D. Hence L is given by a cocycle gij
∈
∗
OY (Uij ), where Uij = Ui ∩ Uj and Uij = Uij ∩ Y . If necessary, we will replace
{Ui } by a finer covering.
We fix the notation first. All quantities that carry an apostrophe exist on
the complement of the set A. Let ei be generating sections of L over Ui so
that for any section σ of L the equation σ |Ui = σi ei holds. With transition

functions gij over Uij we have
ej = ei gij

, and σi = gij

σj .
The hermitian metric h is defined on Ui by the norms of the generators

ei 2h = hi so that σ 2h |Ui = |σi |2 hi .

Now over Uij
2
|gij | = hj /hi
holds. Later we will change the generating sections of L |Ui to e.i , and use the
analogous relations.
The following monodromy argument is essential for the extension of L .
Claim. There exist plurisubharmonic functions ψi on the open subsets Ui ⊂ Y
and holomorphic functions ϕi on Ui such that
hi · |ϕi |−2 = e−ψi |Ui , (1)
where the ψi − ψj are pluriharmonic functions on Uij .

Proof of the claim. Let √
ω|Ui = −1∂∂(ψi0 )
for some plurisubharmonic functions ψi0 on Ui . Now
0
log(eψi hi )
is pluriharmonic on Ui . For a suitable number βi ∈ R and some holomorphic

function fi on Ui we have
0
log(eψi hi ) + βi log |zi | = fi + fi . (2)
We write
βi = γi + 2ki
for 0 ≤ γi < 2 and some integer ki . We set
ψi = ψi0 + γi log |zi |. (3)
These functions are clearly plurisubharmonic on Ui , and γi log |zi | contributes

as an analytic singularity to ψi0 . Set

ϕi = zi−ki efi ∈ O∗ (Ui ).
With these functions (1) holds.

The idea of the proof is to introduce new local generators for the given bun-
dle L over the spaces Ui = Ui \A. This amounts to change the local extensions
to the sets Ui of the line bundle L .
The new generators are
−1
e.i = ei ϕi .
In these bundle coordinates the hermitian metric h on L is given by
.
hi = .
ei 2 = ei 2 |ϕi |−2 = hi · |ϕi |−2
so that (1) reads

. 0
hi = |zi |−γi e−ψi |Ui = e−ψi |Ui . (4)
The transformed transition functions are

g.ij = ϕi · gij

· (ϕj )−1
yielding 5 5γ i
5 zi 5
| =.
hj /.
hi = |zj |γi −γj 55 55 · eψi −ψj .
0 0
2
|.
gij (5)
zj
Since the function zi /zj is holomorphic and nowhere vanishing on Uij , and
since the function ψi0 − ψj0 is pluriharmonic on Uij , the function
5 5γi
5 zi 5
5 5 · eψi0 −ψj0
5 zj 5
is of class C ∞ on all of Uij with no zeroes.


Now −2 < γi − γj < 2, and g.ij is holomorphic on Uij . So (5) implies

that g.ij ∈ Lloc (Uij ). Hence g.ij can be extended holomorphically to Uij , which
2
implies γi ≥ γj , and by symmetry
γi = γj (6)
(whenever D ∩ Uij = ∅). Accordingly (5) now reads

5 5γ i
5 zi 5
|. 2 . .
gij | = hj /hi = 55 55 · eψi −ψj ,
0 0
(7)
zj

and the transition functions g.ij can be extended holomorphically to all of Uij
with no zeroes. So a line bundle L exists.
The functions ψi are plurisubharmonic, and the quantity
√
ω = −1∂∂ψi = ω + π γν [Dν ]
ν
is a well-defined positive current on Y because of (3) and (6). Its restriction to

Y equals ω|Y , where the sum is taken over all connected components of D.
Now we can define
. 0
hi = e−ψi = e−ψi |zi |−γi
on all of Ui . It defines a positive, singular, hermitian metric on L. This shows
the theorem in the special case.
Observe that the numbers βj = γj + 2kj only depend on the connected
component of the smooth hypersurface D.
The above argument also contains the extension property for lower dimen-
sional sets, which is already known for metrics of class C ∞ (cf. Introduction).
Proposition 2. Let Y be a complex manifold and let A ⊂ Y be a closed analytic
subset, whose codimension is at least 2 everywhere. Let (L , h ) be a holomorphic
line bundle on Y = Y \A with a singular hermitian metric, whose curvature
form extends as a closed positive current. Then (L , h ) can be extended to Y
as a holomorphic line bundle (L, h) equipped with a singular hermitian metric
in a unique way.
Note that the extension property of the curvature current need not be as-
sumed.
Proof. By Lemma 2 we may use a stratification of A and assume that A is
smooth of codimension k ≥ 2. The first step is to produce for any point a ∈ A
a neighborhood U (a), into which (L , h )|(U (a)\A) can be extended. We write
A ∩ U (a) as a complete intersection of n − k smooth hypersurfaces, and proceed
like in the proof of Corollary 1. Because of the assumption on the codimension

of A we have trivial monodromy, and for an open covering of U (a)\A the
numbers βi from the proof of Proposition 1 vanish.
Altogether, we find a covering {Ui } of Y such that the line bundle L is
holomorphically trivial on all Ui := Ui \A. In terms of our standard notation,
again the hi = h|Ui can be treated like functions such that the − log hi are
plurisubharmonic functions, which can be extended as such to Ui in a unique
way. Again the monodromy argument from the proof of Proposition 1 is not
needed – only the holomorphic functions fi can occur, and these are used to
change the trivialization over the sets Ui . Finally, the transition functions for
L with respect to {Ui } extend holomorphically (in a unique way) to {Ui } (with
no zeroes) defining an extension (L, h) of (L , h ).
Proposition 3. Let Y be a complex manifold and let A ⊂ Y be a closed analytic
subset. Let (L , h ) be a holomorphic line bundle on Y = Y \A with a singular
hermitian metric, whose curvature current is positive, and extends to Y as a
positive closed current ω. Then (L , h ) can be extended to Y as a holomorphic
line bundle (L, h) equipped with a singular hermitian metric, whose
! curvature
current is positive. The curvature current of h is equal to ω + π γj [Aj ], where
the Aj are the divisorial components of A, and 0 ≤ γi < 2.
Proof. Let A1 ⊂ A be the divisorial part of A, and A2 the union of components
in higher codimension. Denote by A0 the singular set of A1 . We remove the
set A0 and A2 from Y and set Y = Y \(A0 ∪ A2 ). We apply Proposition 1
and extend (L , h ) to Y . Observe that ω|Y is altered like in the proof of
Corollary 1. We finally apply Proposition 2.
3. Main Theorem for Normal Spaces

Proof of Theorem 1 for normal spaces. Let μ : X → Y be a desingularization
of Y such that the pull-back of the curvature tensor of (L , h ) extends to a
positive closed current ω on X. Denote by Y ⊂ Y be the regular part. The
given current ω gives rise to a current ω on Y . We apply Proposition 3 and
extend the line bundle (L , h ) to Y with
! a resulting bundle (L , h ). The
curvature current of h equals ω + π γj [Aj ∩ Y ], where the Aj are the
codimension 1 components of A. Let A .j ⊂ X be the proper transforms of the
! .j ],
Aj on the manifold X. We replace ω with the positive current ω + π γj [A
and we are dealing with the line bundle (L , h ) on Y ⊂ Y , whose curvature
current extends to a desingularization as a positive current.
Altogether we can assume without loss of generality that the set A is con-
tained in the singular set of Y . We proceed with the proof under this additional
assumption and use the original notation.
By Proposition 3 there exists a holomorphic line bundle L on X, equipped

with a singular hermitian metric h whose restriction to μ−1 (Y ) is equal to the
pull-back of (L , h ). We consider the sheaf S = μ∗ (L). Since Y is normal, the
fibers of μ are connected so that μ∗ (L)|Y = L .
We apply Hironaka’s flattening theorem to the coherent sheaf S of rank 1:
There exists a modification τ : W → Y given by a sequence of blow-ups with
smooth centers located over A with the property that L = τ ∗ S/torsion is an
invertible sheaf.
Claim 1. Without loss of generality we can assume the existence of a holo-
morphic map κ : X → W such that τ ◦ κ = μ.
X
}
κ }
} μ
~}
W
τ /Y
Proof of the Claim. The pair (μ, τ ) defines a bimeromorphic map from X to
/ be a desingularization of its graph X ×Y W :
W . Let W
τ /X
/
W

μ μ

W
τ /Y
Now, since the fibers of τ. are connected, τ.∗ τ.∗ L = L, and also the curvature
current of τ.∗ h pushed forward to X equals the original curvature current of h.
Hence we may replace X by W / , the line bundle L by τ.∗ L, the map μ by μ ◦ τ.,
and we set κ = μ ..
We return to the original situation with the additional map κ.
Claim 2. There is an (injective) morphism of invertible OX -modules
→ L.
κ∗ L
Proof of Claim 2. The canonical morphism κ∗ τ ∗ μ∗ L = μ∗ μ∗ L → L factors

over κ∗ (τ ∗ μ∗ L/torsion) = κ∗ L.
Now we turn to the sheaf L. in L can

The image of the invertible sheaf κ∗ L
be identified with L(−D), where D denotes an effective divisor on X, and the
injection is the multiplication ·σ by a canonical section σ of OX (D). It follows
from the construction that D ⊂ μ−1 (A).
Let D = 0. We know that L agrees with L over the complement Y of A in
Y . Because of Claim 2 the line bundle L is trivial on the fibers of κ, and since
the curvature of h is semi-positive, the singular hermitian metric h descends
to W as a (singular) metric The hermitian line bundle (L,

h on L. h) on W
satisfies the claim of Theorem 1.
If D = 0, the hermitian metric on L does not yield a (singular) hermitian
metric on κ∗ L so that zeroes must be taken into account.
We consider τ ∗ μ∗ (L(D)). The multiplication with σ induces a morphism
τ μ∗ L → τ ∗ μ∗ (L(D)). Modulo torsion we get
∗
·σ / τ ∗ μ∗ (L(D))/torsion ,

L
and apply κ∗ (observe that torsion elements are mapped to torsion elements):
·σ / (μ∗ μ∗ (L(D)))/torsion

L(−D) = κ∗ L (8)
·σ ν
·σ

L / L(D)
where ν is the natural morphism, which exists, because L(D) is torsion-free. We

can see from the diagram that the morphism ν is induced by the multiplication
by σ.
We apply the Hironaka flattening process to τ ∗ μ∗ (L(D)): There exists a
modification λ : Z → W whose exceptional set is mapped to A under τ ◦ λ such
that

= λ∗ τ ∗ μ∗ (L(D))/torsion
L
is an invertible OZ -module.
on Z, where τ ◦λ : Z → Y is a modification
Claim 3. The invertible sheaf L

with center contained in A possesses a singular hermitian metric
h that satisfies
the statement of Theorem 1.

and L
we get an injection of in-
Proof of Claim 3. From the definitions of L
vertible sheaves
·σ /

λ∗ L
L
that is induced by the multiplication with σ.
We apply the construction of Claim 1 by letting V be a desingularization
of Z ×Y X, and get
V
α /X .
}}
κ }}
ρ } μ
~ }}}
λ / τ /Y
Z W
From (8) we have a commutative diagram of invertible sheaves on Z
= α∗ κ∗ (τ ∗ μ∗ L/torsion)
ρ∗ λ∗ L
·σ / ρ∗ (λ∗ τ ∗ μ (L(D))/torsion = ρ∗ L

,
∗
·σ ·σ
·σ

α∗ L / α∗ (L(D))
(9)

∗
which implies that we can identify the invertible sheaves α L and ρ L as ∗
subsheaves of α∗ (L(D)) on V .
We equip α∗ L with the pull-back of h. Like in the case D = 0 we see from

that α∗ h descends to L
(since h has semi-positive curvature and
α∗ L = ρ∗ L

is trivial on the fibers of ρ). Over Z\λ−1 τ −1 (A) the hermitian line
since ρ∗ L

bundle (L, h) can be identified with (L , h ).
We mention that by Proposition 3 and Theorem 1 we can trace the growth
of the singular hermitian metric: The given extension of the curvature current
of h is only being changed by adding non-negative multiples of currents of
integration located at the boundary with coefficients smaller than 2π, whereas
the pullback of the given singular metric h is just being extended.
4. Reduced Spaces
Although our extension theorem is based upon analytic methods that are in
principle restricted to normal complex spaces, the theorem holds for arbitrary
reduced complex spaces.
Definition 1. Let X be a reduced complex space and L a holomorphic line
bundle. Then a singular hermitian metric h on L is a singular hermitian metric
on the pull-back of L to the normalization of X, and its curvature is a current
on the normalization of X.
Theorem 2. Let Y be a reduced complex space, and A ⊂ Y a closed analytic
subset. Let L be an invertible sheaf on Y \A, which possesses a holomorphic
extension to the normalization of Y as an invertible sheaf. Then there exists
a reduced complex space Z together with a finite map Z → Y , which is an
isomorphism over Y \A such that L possesses an extension as an invertible
sheaf to Z.
Proof. Denote by ν : Y → Y the normalization of Y . The presheaf
U → {σ ∈ (ν∗ OY )(U ); σ|U \A ∈ OY (U \A)}

defines a coherent OY -module, the so-called gap sheaf
OY [A]ν∗ OY
on Y (cf. [Siu1, Proposition 2]). It carries the structure of an OY -algebra.

According to Houzel [H, Prop. 5 and Prop. 2] it follows that it is an OY -algebra
of finite presentation, and hence its analytic spectrum provides a complex space
Z over Y (cf. also Forster [Fo, Satz 1]).
References
[A-B-S] Axelsson, Reynir; Biswas, Indranil; Schumacher, Georg: Kähler structure on
Hurwitz spaces. Manuscr. math. 147, 63–79 (2015).
[B] Bishop, Errett: Conditions for the analyticity of certain sets. Michigan
Math. J. 11, 289–304 (1964).
[B-S1] Biswas, Indranil; Schumacher, Georg: Coupled vortex equations and moduli:
Deformation theoretic approach and Kähler geometry. Math. Ann. 343,
825–851 (2009).
[B-S2] Biswas, Indranil; Schumacher, Georg: The Weil-Petersson current for moduli
of vector bundles and applications to orbifolds. Ann. Fac. Sci. Toulouse
Math. XXV, 895–917 (2016).
[EM] El Mir, Hassine: Sur le prolongement des courants positifs ferés. Acta Math.
153, 1–45 (1984).
[Fo] Forster, Otto: Zur Theorie der Steinschen Algebren und Moduln. Math. Z.
97, 376–405 (1967).
[F-S] Fujiki, Akira; Schumacher, Georg: The moduli space of extremal compact
Kähler manifolds and generalized Weil-Petersson metrics. Publ. Res. Inst.
Math. Sci. 26, 101–183 (1990).
[G] Griffiths, Phillip: Two theorems on extensions of holomorphic mappings.
Invent. Math. 14, 27–62 (1971).
[H] Houzel, Christian: Géometrie analytique locale, II. Théorie des morphismes
finis. Séminaire Cartan, 13e année, 1969/61, 19.
[M-F-K] Mumford, David; Fogarty, John; Kirwan, Frances: Geometric invariant the-
ory. 3rd enl. ed. Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge.
34. Berlin: Springer-Verlag, (1993).
[R-St] Remmert, Reinhold; Stein, Karl: Über die wesentlichen Singularitäten ana-
lytischer Mengen. Math. Ann. 126, 263–306 (1953).
[Sch] Schumacher, Georg: Positivity of relative canonical bundles and applica-
tions. Invent. math. 190, 1–56 (2012) and Erratum to: Positivity of relative
canonical bundles and applications. Invent. math. 192, 253–255 (2013).
[Sh1] Shiffman, Bernard: On the removal of singularities of analytic sets. Michigan

Math. J. 15, 111–120 (1968).
[Sh2] Shiffman, Bernard: Extension of positive line bundles and meromorphic
maps. Invent. math. 15, 332–347 (1972).
[Si] Sibony, Nessim: Quelques problemes de prolongement de courants en anal-
yse complexe. Duke Math. J. 52, 157–197 (1985).
[Siu1] Siu, Yum-Tong: Absolute gap-sheaves and extensions of coherent analytic
sheaves. Trans. Am. Math. Soc. 141, 361–376 (1969).
[Siu2] Siu, Yum-Tong: Analyticity of sets associated to Lelong numbers and the
extension of closed positive currents. Invent. Math. 27, 53–156 (1974).
[Sk] Skoda, Henri: Prolongement des courants positifs fermes de masse finie.
Invent. Math. 66, 361–376 (1982).
Elliptic Fibrations on Supersingular K3
Surface with Artin Invariant 1 in
Characteristic 3
Tathagata Sengupta∗,†
Abstract
We describe elliptic models with section on the Shioda supersingular K3 surface
X of Artin invariant 1 over an algebraically closed field of characteristic 3. We
compute elliptic parameters and Weierstrass equations for the fifty two different
fibrations, and analyze some of the reducible fibers and Mordell-Weil lattices.
1. Introduction
1.1. Definitions and Examples.
Definition 1. K3 surfaces are smooth algebraic surfaces X for which the
canonical divisor KX ∼
= OX and H 1 (X, OX ) = 0.
Definition 2. Supersingular surfaces (in the sense of Shioda) are those for
which all cycles in the 2−dimensional étale cohomology groups are algebraic.
The Néron-Severi group, the group of divisors modulo algebraic equiva-
lence, denoted by N S(X), is a lattice of rank, say ρ. The number ρ is called
the rank of the K3. In characteristic 0, we have the inequality ρ ≤ b2 − 2,
where b2 is the second Betti number. In characteristic p however, we can only
say ρ ≤ b2 . In the case of K3 surfaces, b2 = 22. Thus, ρ ≤ 22. According
to the definition, (Shioda) supersingular K3 surfaces are the ones for which
ρ = 22. The existence of such surfaces is a phenomenon particular to positive
∗ Department of Mathematics, University of Hyderabad, Hyderabad, India.
† The author thanks his thesis advisor Prof. Abhinav Kumar for his precious guidance.
And also the mathematics departments of Brandeis University, Massachusetts Institute of

Technology and University of Hyderabad for invaluable help during the period of this project.
Geometry, DOI 10.1007/978-981-10-5648-2_15
240 Tathagata Sengupta
characteristic. We now list some of the basic properties of K3 surfaces. The

geometric genus, pg = dim H 2 (X, OX ) = h0,2 = 1, because, by Serre duality,
h2 (OX ) = h0 (X, KX ⊗ OX ) = h0 (OX )). Its arithmetic genus, pa = 1. Hence,
the topological Euler characteristic χ(OX ) = h0 (OX ) − h1 (OX ) + h2 (OX ) = 2.
We also have that c21 (X) = 0, since c21 (X) = KX · KX . Using Noether’s formula
c21 (X)+c2 (X) = 12χ(OX ), we get the second Chern number, c2 (X) = 24. Since
b1 (X) = 2h1 (OX ), we get that b1 (X) = 0. Thus, the Betti numbers must be
b0 = 1, b1 = 0, b2 = 22, b3 = 0, b4 = 1
Some of the best known examples of supersingular K3 surfaces are:
1. Fermat Quartic. The Fermat quartic is a hypersurface in P3 given by the
equation
x4 + y 4 + z 4 + w 4 = 0
It is a supersingular K3 surface over fields of characteristic 3 mod 4.
2. Kummer Surfaces. The Kummer surface of the product of 2 supersingular
elliptic curves over fields of odd characteristic is a supersingular K3 sur-
face. We describe in detail the Kummer surface associated to the product
of the supersingular elliptic curve y 2 = x3 − x with itself, over a field of
characteristic 3.
3. Double sextic. The double sextic is a double cover of the projective plane
P2 , branched over a sextic. A K3 surface with polarization of degree 2
can be written as a double cover of P2 , branched along a sextic. The
supersingular K3 surface of Artin invariant 1 in characteristic 3 admits
such a model.
Many of the ideas underlying this work are based on the work of Tetsuji
Shioda [Sh1], [Sh2] Noam Elkies, Matthias Schuett and Abhinav Kumar [Ku1],
[Ku2], among others. The author is grateful to all the above mathematicians
for their amazing contributions and insights.
1.2. Elliptic Fibrations. Let k be an algebraically closed field, and let

C be a smooth curve over k.
Definition 3. An elliptic surface X over C is a smooth projective surface
X with an elliptic fibration over C, that is, there is a surjective morphism
f : X → C such that all but finitely many fibers are smooth curves of genus
1, no fiber contains an exceptional curve of the first kind (that is X → C is
relatively minimal ). Moreover, f has a section, that is, a smooth morphism
s : C → X such that f ◦ s = idC . We also require that X has at least one
singular fiber, so that it is not isomorphic to the product E × C for some
elliptic curve E after base-change by a finite etale map C → C.
Elliptic Fibrations on Supersingular K3 Surface with Artin Invariant1 241
We can use the zero section to describe the Weierstrass form. The generic
fiber E can then be regarded as an elliptic curve over the function field k(C).
The general Weirstrass form looks like
y 2 + a1 (t)xy + a3 (t)y = x3 + a2 (t)x2 + a4 (t)x + a6 (t), with ai (t) ∈ k(t)
which, in characteristic 3, can be simplified to
y 2 = x3 + a2 (t)x2 + a4 (t)x + a6 (t)
The sections form an abelian group E(k(C)). The zero section will be de-
noted by O throughout this paper. The sections are in a natural one-to-one
correspondence with the k(C)-rational points on E.
The Kodaira-Néron model gives a way to associate an elliptic surface X → C
over the ground field k, given the generic fiber E over k(C). The first step is
to remove all the points from C at which the Weierstrass form is singular,
and then filling in suitable singular fibers. This process is formally described
by Tate’s algorithm which we use extensively. It is a well-known fact that the
Kodaira-Néron model is unique, given the Weierstrass form.
1.2.1. Singular fibers. The singular fibers can be read off from the Weier-
strass form using Tate’s algorithm. All the irreducible components of a singular
fiber are rational curves. If a singular fiber is irreducible, then it is a rational
curve with a node or a cusp. In Kodaira’s classification of singular fibers, these
are denoted by I1 and II respectively. If a singular fiber is reducible, then every
component is a rational curve with self-intersection −2 on the surface.
1.2.2. Elliptic K3 surfaces. In this case, the base curve is P1 . The degrees
of the coefficients ai are restricted by the condition deg(ai (t)) ≤ 2i, and some
deg(ai (t)) > i, or else it gives a rational elliptic surface. At t = ∞, we change
variables by s = 1t , and the coefficients become ai (s) = s2i ai ( 1s ). Thus we can
homogenise the coefficients to polynomials of degree 2i in two variables, s and
t. The discriminant becomes then a homogeneous polynomial of degree 24 (in
s and t). Thus, the Weierstrass form of a elliptic K3 surface over P1 can be
seen as a hypersurface in the weighted projective space P[1, 1, 4, 6].
1.2.3. Neron-Severi lattice and Mordell-Weil group.
Definition 4. A key invariant of a K3 surface X is its Néron-Severi lattice

N S(X) = N Sk̄ (X). This is the usual Néron-Severi group of divisors defined
over k̄ modulo algebraic equivalence, equipped with the symmetric integer-
valued bilinear form induced from the intersection pairing on the divisors. For
K3 surfaces, the notions of algebraic equivalence, numerical equivalence and
linear equivalence coincide, which implies that the Picard group of a K3 surface
is naturally isomorphic to the Néron-Severi group.
For a K3 surface, this is an abelian group and the pairing is even, that
is, C · C ∈ 2Z for all C ∈ N S(X). By the Hodge index theorem, the pairing
is non-degenerate of signature (1, ρ − 1), where ρ is the rank of N S(X) (also
known as the rank of the K3). In our case, ρ = 22, in other words, X is a
supersingular K3 surface. By Artin [Ar], the discriminant of the Neron-Severi
lattice is −p2σ , where p is the characteristic of the base field k, and σ is an
invariant for the surface, known as the Artin invariant. It is also known that
1 ≤ σ ≤ 10. In this paper, we will largely deal with the supersingular K3
surface over characteristic 3, with Artin invariant 1. Note the definite article
here, since there can be only one, upto isomorphism, as shown in [Og].
If X → C is an elliptic fibration with a zero section, N S(X) contains two
special classes, the fiber F (preimage of any point of P1 under f ), and the
image of the zero section O. The intersection pairing they satisfy is F · F = 0,
O · O = −2 and F · O = 1. Hence the sublattice U generated by F and O is
isomorphic to the hyperbolic plane (U). Conversely, any copy of U in N S(X)
describes X as an elliptic surface. One of the generators or its negative is
effective, and has two independent sections, whose ratio gives the map to P1 .
Definition 5. The essential lattice is the orthogonal complement of the copy

of U in N S(X), and is denoted by N Sess . It is a positive definite lattice.
Definition 6. The Mordell-Weil group of the surface f : X → C is the

(abelian) group of sections from C → X. This is also naturally identified with
the k(C)-rational points of E, denoted by E(k(C)), where E is the generic fiber
for the elliptic fibration defined by f : X → C.
According to teh Mordell-Weil theorem [Si], E(k(C)) is a finitely generated

group.
Definition 7. The trivial lattice T is the lattice generated by the classes O, F

and the Fν,i , where O is the zero section, F is the class of the fiber, and Fν,i is
the i-th non-identity component of the reducible fiber at ν. Here, the identity
component of a reducible fiber is the component intersecting the zero section.
Thus,
! if mν is the number of components of the fiber Fν , then the rank(T ) =
2 + ν∈P (mν − 1), where P is the set of points of P1 (k) where the fibers are
reducible. The Shioda-Tate formula gives the following relation between the
Mordell-Weil group and the Neron-Severi lattice:
E(k(C)) ∼
= N S(X)/T
Let R ⊂ N Sess be the root lattice of N Sess , that is the sublattice spanned
by the vectors of norm 2 (known as roots). Then N Sess /R can be canonically
identified with the Mordell-Weil group E(k(C)). R itself is a direct sum of root
lattices An (n ≥ 1), Dn n ≥ 4, E6 , E7 or E8 , with each factor indicating a
reducible fiber of the corresponding type.
1.2.4. Dynkin diagrams. The Dynkin diagrams corresponding to the root
systems are given in Figure 1.1. Here the nodes indicate roots (rational curves
with self-intersection −2), and two nodes are connected by an edge if and only
if the corresponding rational curves intersect, in which case the intersection
number is 1.
• A1
• • A2
• • An
•
• • • D4
•
• • • • D5
•
• • • Dn
•
• • • • • E6
•
• • • • • • E7
•
• • • • • • • E8
Figure 1.1. Dynkin diagrams
The determinants of the negative-definite lattices defined by the respective
Dynkin diagrams are:
det(An ) = (−1)n (n + 1), det(Dn ) = (−1)n 4 (n ≥ 4)
det(E6 ) = 3, det(E7 ) = −2, det(E8 ) = 1
1.2.5. Height Pairing. The height pairing formula between two sections P
and Q is given by

P, Q = χ(X) + P · O + Q · O − P · Q − contrν (P, Q)
ν
where χ(X) is the Euler characteristic of the surface (= 2 in the case of K3

surfaces), and the correction terms contrν (P, Q) depend on the components of
the fibers that are met by P and Q. More specifically, let Aν be the intersection
matrix of the non-identity components of the fibers Fν , ν ∈ P1
Aν = (Θν,i , Θν,j )1≤i,j≤mν −1
Then if P meets Θν,i , and Q meets Θν,j , then the local contribution term is
given by
contrν (P, Q) = 0 if ij = 0, or = −(A−1
ν )i,j if ij = 0
We specialize P = Q to get the height of a single section P . Thus,

h(P ) = P, P = 2χ(X) + 2P · O − contrν (P, P )
ν
The contribution terms are given in the following table. The conventions
followed are the following:
1. The components of the An fibers are numbered cyclically by Θ0 ,

Θ1 ,· · · ,Θn , where Θ0 is the identity component.
2. At additive fibers, only the simple components are named.

3. The simple components of the Dn fibers are numbered as Θ1 for the
near component, and Θ2 , Θ3 for the far components (with respect to the
identity component).
Dynkin Diagram An−1 E6 E7 Dn+4

i(n−i) 4 3
i=j n 3 2 1, i = 1
1 + n4 , i = 2, 3
i(n−j) 2 1
i<j n 3 - 2, i = 1
1 n
2 + 4, i = 2
As described in [Sh2], we compute the determinant of the Neron-Severi

using
(−1)rank(E(k(C))) disc(T (X))disc(M W L(X))

disc(N S(X)) =
|E(k(C))tor |2
where the discriminant of the Mordell-Weil Lattice (the free part generated by
non-torsion sections) is calculated using the height pairing between the sections.
(E(k(C)))tor is the group of torsion sections.
2. Elliptic Fibrations for the Supersingular K3

Surface with σ = 1 over Char 3
2.1. Introduction. Our goal is to use 2-neighbour and 3-neighbour con-
structions in root lattices to obtain all possible elliptic fibrations of the super-
singular surface of Artin invariant 1 over characteristic 3. Many of the ideas
underlying the work in this paper are based on the works of Tetsuji Shioda
[Sh1], [Sh2] Noam Elkies, Matthias Schuett and Abhinav Kumar [Ku1], [Ku2],
among others. From the works of Rudakov-Shafarevich and Ichiro Shimada we
know descriptions of the abstract Néron-Severi lattices for other characteristics
and Artin invariants. The methods used in this paper work more generally for
those cases as well, although the computations can become more cumbersome,
albeit finite. Ideally one should be able to design a computer program which
could compute these fibrations in a reasonable time.
A few definitions:
Definition 8. A N iemeier lattice is the orthogonal complement of a copy

of the hyperbolic plane U in the even unimodular lattice of signature (1, 25),
II(1,25) . There are 24 such lattices, which have been enumerated by Hans-Volker
Niemeier, and except one, all of them have root sublattices. These are all the
unimodular even positive definite lattices of rank 24.
Definition 9. A root in a lattice is an element of norm −2.
Definition 10. A root lattice is one that is generated by its set of roots.
The only Niemeier lattice that does not have any roots is the famous Leech
lattice. All the other Niemeiers have finite index root sublattices. The entire
list of root sublattices of the Niemeier lattices is as follows
A24 12 8 6 4 6 4 2 2
1 , A 2 , A 3 , A 4 , A 5 D 4 , D 4 , A 6 , A 7 D5 ,
A38 , A29 D6 , D64 , E64 , A11 D7 E6 , A212 , D83 , A15 D9 ,

A17 E7 , D10 E72 , D12
2
, A24 , D16 E8 , E83 , D24
Here we prove that there is a finite list of possible elliptic fibrations on
the supersingular K3 over a field of characteristic 3 which we assume to be
algebraically closed. We describe the singular fiber, using the Neimeier lattice
description. We go on to describe the fibrations, and give Weierstrass equations
for these models.
Lemma 1. There is a unique embedding of A2 into each of the root lattices R,

upto the action of the automorphism group
Proof. From the general theory of Weyl group actions on root lattices, we know
that the action is simply transitive. Thus, if D and D are two irreducible
Dynkin diagrams giving a basis for the corresponding root lattice (as described
in one of the earlier sections), then there is an element σ ∈ W (R) such that
σ(D) = D . Based on this, we want to prove that given any copy of A2 within
a root lattice, it can be extended to the complete Dynkin diagram, such that
the copy of A2 forms
1. the first 2 roots of the An diagram, in case of an An root lattice,
2. the first 2 roots of the long tail of the Dn (n > 4) diagram, in case of the
Dn root lattice, (note that any embedding of A2 is equivalent in a D4 )
3. the 2 roots of the short tail of the E6 diagram, in case of the E6 lattice,
4. the 2 roots of the tail of length 2 (the other 2 tails being of length 1 and
3, all attached to the one single node of valency 3), in case of the E7
lattice, and
5. the first 2 roots of the longest tail of the E8 diagram, in case of the E8
lattice.
In the case of An , let us denote the nodes by [a1 , a2 , a3 , a4 , · · · an ]. If our A2
maps to {a2 , a3 }, we can extend the chain of nodes given by [a2 , a3 , a4 , · · · an ]
by the root −(a1 +a2 +a3 +· · · an ). This clearly gives us a new An diagram, with
{a2 , a3 } as the first 2 roots. If A2 maps to {a3 , a4 }, we can extend the chain of
nodes given by [a3 , a4 , a5 , · · · an ] by the roots {−(a2 + a3 + · · · an ), −a1 }. This
gives us a new An diagram, with {a3 , a4 } as the first 2 roots. If A2 maps to
{a4 , a5 }, we can extend the chain of nodes given by [a4 , a5 , · · · an ] by the roots
{−(a3 + · · · an ), −a2 , −a1 }. This gives us a new An diagram, with {a3 , a4 } as
the first 2 roots. We can do this with any embedding of A2 into An .
In the case of Dn , given by [a1 , a2 , a3 , · · · , an−1 , an ], where an is connected
to an−2 and no other root. If A2 maps to {a2 , a3 }, we can extend the straight
chain given by [a2 , a3 , · · · , an ] to a Dn , using −(a1 + a2 + · · · + an−2 + an ) and
−(a1 + a2 + · · · + an−2 + an−1 ). The other cases are very similar to this.
In the case of E6 , let us denote the diagram by [a1 , a2 , a3 , a4 , a5 , a6 ], where
[a1 , a2 , a3 , a4 , a5 ] gives an A5 , and a6 is connected to a3 . If our A2 maps to
{a1 , a2 }, we can extend the chain [a1 , a2 , a3 , a6 ] by the roots −(a1 + 2a2 + 2a3 +
a4 + a6 ) and −a5 , which gives an E6 with the A2 now mapping to the shortest
tail of an E6 diagram.
We can do similar extensions to other embeddings of A2 , and also similarly
in the case of the other En ’s.
Theorem 1. There are 52 possible elliptic fibrations on the supersingular K3
surface of Artin invariant 1 over an algebraically closed field of characteristic 3
Remark 1. Although we assume the ground field to be algebraically closed, in
our calculations we have to go only upto F9 .
Proof. Let X/k be the supersingular K3 of Artin invariant 1, k is algebraically

closed field of characteristic 3. The Néron-Severi lattice of X, N S(X)(= N ) is
an invariant of X. From the classification of even non-degenerate lattices, we
know that N S(X) ∼ = U ⊕ A22 ⊕ E82 , where U is the hyperbolic plane. This is
a 22 dimensional negative definite lattice of signature (1, 21) and discriminant
−32 . Using well-known results from lattice theory, we know that N can be
embedded into the even unimodular lattice of signature (1, 25), II(1,25) . Now,
as we described before, the choice of the class of a fiber and a zero section
on X determines a copy of the hyperbolic plane U in N S(X), and vice-versa.
Describing embeddings U → N S(X) is equivalent to describing the embeddings
N S(X)⊥ → U ⊥ , where the orthogonal complements are taken in the ambient
II(1,25) . We know that the U ⊥ ’s are by definition the Niemeier lattices. Now,
according to a result due to Nikulin, let M → L be a primitive embedding
of non-degenerate even lattices, and suppose that L is unimodular. Then the
discriminant forms qM ⊥ ∼ = −qM . Conversely if M1 and M2 are non-degenrate
even lattices which satisfy qM1 ∼ = −qM2 , then there is a primitive embedding
of M1 → L such that M1⊥ ∼ = M2 . In our case, let M = U ⊕ A22 ⊕ E82 . Thus,
the discriminant of M is −9. Using the above result from Nikulin [N1],[N2] the
discriminant of M ⊥ is 9, and it is of rank 4. Thus, M ⊥ ∼ = A22 , using to the
classification of lattices. We thus need to figure out the number of embeddings
of A22 in each of the Niemeier lattices, upto isomorphism.
Now, using the previous lemma, we can describe the orthogonal complement
of A2 in each of the root systems, and successively for the second copy of A2 .
The uniqueness of these embeddings follows from the preceding lemma. The
list of the orthogonal complements are given in the table below.
The following table gives the orthogonal complement of A22 in each of the
different Niemeier lattices. This calculation is due to Noam Elkies and Matthias
Schuett.
Root System D A⊥
2 ⊂D (A22 )⊥ ⊂ D
A2 0 0
A3 0 0
An for n > 3 An−3 An−6 for n > 6 (0 for A5 , A6 )
D4 0 0
D5 A21 0
D6 A3 0
Dn for n > 6 Dn−3 Dn−6 for n > 9 (0 for D7 , A21 for D8 , A3 for D9 )
E6 A22 A2
E7 A5 A2
E8 E6 A22
Thus the lattices that can contain A22 are An (n > 4), Dn (n > 5), and all of the
En ’s.
Let N be a Niemeier lattice, and N the orthogonal complement of A22 in
N . Following is the list of Roots(N ) and Roots(N ). Borrowing notation from
Noam Elkies, m : stands for the lattice obtained by extracting A22 from the
m-th factor of Roots(N ), and mn : for the lattice obtained by extracting A2 ’s
from the m-th and n-th factor.
Roots(N ) Roots(N ) rk(M W ) Roots(N ) Roots(N ) rk(M W )

A12
2 A2 10
0 A212 1 : A6 A12 2
A83 A63 2 11 : A29 2
A64 A21 A44 2 D83 1 : A21 D82 2
D46 D44 4 11 : D52 D8 2
D4 A45 12 : A2 A35 3 D9 A15 1 : A3 A15 2
2 : D4 A35 1 2 : A9 D9 2
22 : A22 D4 A25 2 12 : D6 A12 2
A46 1 : A36 2 E7 A17 1 : A2 A17 1
11 : A23 A26 2 2 : E7 A11 2
D52 A27 11 : A41 A27 2 12 : A5 A14 1
2 : A1 D52 A7 2 E72 D10 1 : A2 E7 D10 1
2 2
12 : A1 A4 D5 A7 2 11 : A5 D10 0
22 : A24 D52 2 2 : D4 E72 2
A38 1 : A2 A28 2 12 : A5 D7 E7 1
11 : A25 A8 2 2
D12 1 : D6 D12 2
D64 1 : D63 2 11 : D92 2
11 : A23 D62 2 E83 1 : A22 E82 0
D6 A29 1 : A29 2 11 : E62 E8 0
2 : A3 D6 A9 2 E8 D16 1 : A22 D16 0
12 : A3 A6 A9 2 2 : E8 D10 2
22 : A26 D6 2 12 : E6 D13 1
E64 1 : A2 E63 0 A24 A18 2
4 2
11 : A2 E6 0 D24 D18 2
E6 D7 A11 1 : A2 D7 A11 0
2 : E6 A11 3
12 : A22 D4 A11 1
3 : E6 D7 A8 2
13 : A22 D7 A8 1
23 : D4 E6 A8 2
Now we begin writing down the elliptic models.

3. Fibration 1 : A11 , A2 , D7
Weierstrass equation: y 2 = x3 + 2(t3 + 1)x2 + t6 x
This fibration corresponds to the fiber type of A11 A2 D7 . In the diagram
below, we have the A11 fiber at t = 0, the A2 fiber at t = 1, and the D7 fiber
at t = ∞, given by the nodes ai ’s, bj ’s and ck ’s respectively. The zero section is
given by the node O, and the torsion section by T . These nodes represent the
roots of N S(X), which correspond smooth rational curves of self-intersection
−2.
jjjj T ==
j jjjj ==
a6 E == c6 D c7
www E == D zz
a7 a5 == c5
==
==
a8 a4 b1 @ b2 == c4
@ ==
==
a9 a3 b0 ==c3
==
a10 a2 c2==D=
zz D=
a11 G a1 c0 c1
GG yy rrrr
a0 TTT rr
TTTT rrrrr
TT r
O
The trivial lattice (that is, the sub-lattice of N S(X) spanned by the classes
of the zero section and the irreducible components of the fibers) has rank
2 + 11 + 2 + 7 = 22. Thus the M W -rank is 0. We have a 4-torsion section,
given by (t3 , 0). The discriminant of the trivial lattice is 12 · 3 · 4 = 144. Thus,
the trivial lattice together with the 4-torsion section has signature (1, 21) and
absolute discriminant 32 , and is thus equal to the full N S(X). Below are the
positions of the reducible fibers, and some of the sections.
Position Kodaira-Néron type

t=0 A11
t=1 A2
t=∞ D7
section type equation

4-torsion (t3 , 0)
4. Fibration 2 : D4 , D4 , D4 , D4
Weierstrass equation: y 2 = x3 − t2 (t − 1)2 (t + 1)2 x
This is the Kummer surface associated to the product of the supersingular
elliptic curve y 2 = x3 − x with itself, over characteristic 3. This fibration cor-
responds to the fiber type of D44 , with M W -rank 4, and full 2-torsion sections.
In the diagram below, we see the D4 fibers at t = 0,1,−1 and ∞,given by the
nodes Rij , (1 ≤ i ≤ 4, 0 ≤ j ≤ 4), the zero section O, and one of the non-torsion
sections U . The nodes represent the roots in the Néron-Severi lattice, which
correspond to smooth rational curves of self-intersection −2.
gg U ZSSZSZSZZZZZZZ
ggggggggg SSS Z
SSS ZZZZZZZZZZZZ
gggg ggg SS ZZZZZZZ
gg
R11 I R13 R21 I R23 R31 I R33 R41 I R43
I vv I vv I vv I vv
R10 I R20 I R30 I R40 I
vv I vv I vv I vv I
R12 R14 WWWWR R R R
k 34 R d
42 dddd R44
kkk dddddddddddd
WW22WWW 249 32
WWWWW 99 k k
WWWWW 9 kk dddd
W dkkdkdkdddddd
O
The trivial lattice (that is, the sub-lattice of N S(X) spanned by the classes
of the zero section and the irreducible components of the fibers) has rank
2 + 4 · 4 = 18, and discriminant −16. The heights of the 4 non-torsion sections
are 2,1,1 and 2. We see that the sublattice generated by the trivial lattice, the
torsion sections and the non-torsion sections has signature (1, 21) and absolute
discriminant 32 . Consequently it must be all of N S(X), which we know to be
of the same rank and discriminant.

t=0 D4
t=1 D4
t = −1 D4
t=∞ D4

2-torsion (0, 0)
(t(t − 1)(t + 1), 0)
(−t(t − 1)(t + 1), 0)
5. Fibration 3 : A6 , A6 , A6
Weierstrass equation: y 2 = x3 + (t4 − t + 1)x2 + t2 (t − 1)(1 + t − t2 )x + t4 (t − 1)2
The original model over Q was given by Tate as an example of a singular
K3 surface over Q, that is one with rank 20. The reduction of Tate’s equation
in characteristic 3 gives the above equation, and has fiber configuration
A6 , A6 , A6
Since it also has a non-torsion section, (0, t2 (t−1)), thus the rank of the surface
over characteristic 3 is at least 21. It is known that any such surface is actually
of rank 22, and is hence supersingular.
Fiber configuration
A6 , A6 , A6
6. Fibration 4 : A1 , A7 , D5 , D5
To obtain the Weierstrass equation for this fibration, we use a 2-neighbor step
from Fibration 1. We compute the explicitly the space of sections of the line
bundle O(F ), where F = 2O + 2a0 + a1 + a11 + b0 + c0 is the class of the D5
fiber we are considering. The space of the sections is 2 dimensional and the
ratio of two linearly independent global sections will be an elliptic parameter
for X, for which the class of the fiber will be F . Any global section has a pole
of order at most 2 along O, the zero section of Fibration 1. Also it has at most
a double pole along a0 , which is the identity component of the t = 0 fiber of
Fibration 1, a simple pole along each of a1 and a11 and along b0 and c0 , the
identity components of the fibers at t = 1 and t = ∞ respectively. We deduce
that a global section must have the form
a0 + a1 t + a2 t2 + a3 t3 + a4 t4 + bx
w=
t2 (t − 1)
Since 1 is a global section, we can subtract multiples of t2 (t − 1) from the

numerator, and assume that a3 = 0. With this constraint, we obtain a 1-
dimensional space of sections, the generator of which will give us the parameter
for the base of the Fibration 4. To ensure a unique choice of the generator, we
scale x such that b = 1.
The next step is to obtain further conditions on the ai ’s by looking at the
order of vanishing of the section at various non-identity components (as pre-
scribed above). For this purpose, we employ successive blow-ups at the singular
points, and follow Tate’s algorithm to obtain conditions. For example, at t = 0,
the first blow-up maps x → tx and y → ty, which ensures a0 = 0 for w to
have a pole of order at most 1 at the near leaf of the A11 fiber. Similarly, the
second blow-up maps x → t2 x, which means a1 = 0. At t = ∞, we change
coordinates,and define u = 1t , such that u = 0. Thus, the original equation
becomes
y 2 = x3 + 2(u + u4 )x2 + u2 x
and the section becomes
a2 u2 + a4 + xu3
w=
u(1 − u)
Again, blowing-up once gives a4 = 0. At t = 1, we again change coordinates,

setting u = t − 1, such that u = 0. Now, there is one extra step that needs to be
taken care of. The singularity of the curve at t = 1 is at (1, 0). Hence, we map
x → 1 + x to map shift the singularity to (0, 0). Thus, the original equation
changes to
y 2 = x3 + 2((u + 1)3 + 1)x2 + (u + 1)6 x
and the section changes to
x + a 2 t2 x + a2 (u + 1)2
w= =
t2 (t − 1) (u + 1)2 u
x−t2
The first blow-up maps x → 1 + ux, giving a2 = −1. Thus w = t2 (t−1) , which
means
x = wt2 (t − 1) + t2
We now replace x by the above in the right side of the original equation, and
simplify by absorbing square factors from the right into the y 2 term on the left.
We get a quartic in t, with coefficients in F3 (w), which can be converted into
the following Weierstrass form, using the recipe described in the Appendix.
y 2 = x3 + (t3 + t − 1)x2 − t(t − 1)(t + 1)2 x + t2 (t − 1)2 (t + 1)4

t=0 A1
t=1 A7
t = −1 D5
t=∞ D5
This fibration has M W -rank of 2.

non-torsion (t(t − 1)(t + 1)2 , 0)
7. Fibration 5: E7 , A2 , D10
We use a 2-neighbor step from Fibration 1, and explicitly compute the 2-
dimensional space of sections of O(F ), where F = a1 + a2 + 2a0 + 2O + 2c0 +
2c2 + 2c3 + 2c4 + 2c5 + c6 + c7 is the new D10 fiber considered. The new section
is
a 0 + a 1 t + a 3 t3 + a 4 t4 + x
w=
t2
By blowing up at t = 0, we get a0 = a1 = 0. Similarly, subsequent blow-ups at
t = ∞ give a3 = a4 = 0. Thus,
x
w= 2
t
Substituting x = wt2 in the original equation, dividing by suitable factors and
converting the resulting quartic into the Weierstrass form, we get
y 2 = x3 − t3 x2 + t3 x

t=0 E7
t=1 A2
t = −1 D10
The trivial lattice is of rank 2 + 7 + 2 + 10 = 21. Thus, the M W rank is 1.

We also have a 2-torsion section given by (0, 0). The only non-torsion section is
given by (1, 1). The height pairing formula requires that the height of the non-
torsion section is 5/2, and it intersects the A2 fiber at the identity component,
and the D10 fiber at the near leaf.

2-torsion (0, 0)
non-torsion (1, 1)
8. Fibration 6: D6 , D6 , D6
We use a 2-neighbor construction from Fibration 5 and explicitly compute the
2-dimensional space of sections of O(F ). Here F = a1 + a3 + 2a0 + 2O + b0 + c0 ,
where a0 is the identity component of the D10 fiber, and a1 the near leaf, b0
is the identity component of the A2 fiber, and c0 that of the E7 fiber. We
reparametrized the original equation to get the D10 fiber at t = 0, the A2 at
t = 1 and the E7 at t = ∞. The original equation now becomes
y 2 = x3 − tx2 + t5 x
And the new section is

a 0 + a 1 t + a 2 t2 + a 4 t4 + x
w=
t2 (t − 1)
Consecutive blow-ups at t = 0 give a0 = a1 = 0. Similar calculations at t = ∞

give a4 = 0, and similarly the conditions on the degree of poles of the section
at t = 1 demands that a2 = 1. Thus,
x + t2
w=
t2 (t − 1)
which means
x = wt2 (t − 1) − t2
Replacing x by the above, and simplifying, we get the required Weierstrass
equation
y 2 = x3 + t(1 − t2 )x2 − t2 (t3 − 1)x

t=0 D6
t=1 D6
t=∞ D6
The M W -rank of this surface is 2, and it has full 2-torsion, as given by the
equations below.

2-torsion (0, 0)
(t3 + t2 + t, 0)
(t − t2 , 0)
9. Fibration 7 : D8 , A1 , A1 , D8
We use a 2-neighbor step from Fibration 6. We explicitly compute the 2-
dimensional space of sections of the line bundle O(F ), where F = a1 + a3 +
2a2 + 2a0 + 2O + 2b0 + 2b2 + b1 + b3 is the new fiber we consider. Here, a0 and
b0 are the identity components of the D6 fibers at t = 0 and t = 1 respectively.
Whereas a1 and b1 are the near components, and a2 , a3 , b2 , b3 are the double
components.
The original equation is given by
y 2 = x3 + t(1 − t2 )x2 − t2 (t3 − 1)x

2 3
and the new section by a0 +a1 t+a 2 t +a3 t +x
t2 (t+1)2 . Blow-ups at t = 0 force a0 = a1 = 0.
Blow-ups at t = 1 give a2 = a3 = −1. Thus,
x − t2 − t3
w=
t2 (t + 1)2
. Substituting
x = wt2 (t + 1)2 + t2 + t3
in the original equation, and simplifying as usual, we get
y 2 = x3 + t(t2 + 1)x2 − t3 (t + 1)2 x + t5 (t + 1)2
This can further be simplified to
y 2 = x3 + (t3 + t)x2 + t4 x

t=0 D8
t=1 A1
t = −1 A1
t=∞ D8
The M W -rank of this surface is 2, and it has full 2-torsion, given by the
following equations.

2-torsion (0, 0)
(−t, 0)
(−t3 , 0)
10. Fibration 8 : E6 , D7 , A5
We use a 2-neighbor construction from Fibration 4. We compute the space of
sections of the line bundle O(F ), where F = 2O + 3a0 + 2a1 + 2a7 + a2 + a6
is the new E6 fiber we are considering. Here a0 is the identity component of
the A7 fiber at t = 0, a1 and a7 are the near leaves, and a2 and a6 are the
components intersecting the near leaves. The original equation is
y 2 = x3 − (t − 1)(t2 + t − 1)x2 + (t − 1)2 (t + 1)4 x + t2 (t + 1)2 (t − 1)4
The new section is given by

a 0 + a 1 t + a 2 t2 + a 3 t3 + x
w=
t3 (t − 1)
At t = 0, the singularity is at x = −1. Hence, we translate x → x − 1 and blow-

up to get a0 = 1. Replacing x by tx in the original equation, we are required to
translate x → x − t. The second blow-up maps x → t2 x, and forces a1 = 1 to
give a pole of order 2. Similarly, after translating x → x − t2 , the third blow-up
gives a2 = 1. Thus,
1 + t + t2 + a3 t3 + x
w=
t3 (t − 1)
At t = 1, we define u = t − 1, so that
1 + (u + 1) + (u + 1)2 + a3 (u + 1)3 + x
w=
u(u + 1)3
Since the singularity is at x = 0, the first blow-up x → ux, to give a3 = 0.
Thus,
x + t2 + t + 1
w=
t3 (t − 1)
whence
x = wt3 (t − 1) − t2 − t − 1
Simplifying, as usual,
y 2 = x3 + tx2 − t3 (t + 1)2 x + t5 (t + 1)4
(It further simplifies to x3 − tx2 + t12 − t9 + t6 ).

t=0 D7
t = −1 A5
t=∞ E6
This fibration has M W -rank 2. The non-torsion sections are given below.
non-torsion (0, t3 (t3 + 1))
11. Fibration 9 : A3 , A9 , D6
We use a 2-neighbor construction on Fibration 1. The original equation is
y 2 = x3 − (t3 + 1)x2 + t6 x
The new D6 fiber is given by F = 2O + a0 + b0 + 2c0 + 2c1 + c2 + c3 , where the
ai ’s are the nodes from the A11 fiber, the bj ’s from the A2 fiber, and ck ’s are
3 4
from D7 . The new section is given by a0 +a1 t+a 3 t +a4 t +x
t(t−1)
At t = 0, we blow-up once, sending x → tx, and get a0 = 0. For t = ∞, we

replace u = 1t , to get
a1 u3 + a3 u + a4 + x
w=
u2 (1 − u)
The first blow-up, sending x → ux, gives a pole of order 2 automatically. For
the second blow-up, we need to translate x → x − u, to shift the singularity
to (0, 0). The second blow-up thus sends x → u2 x − u, and we get a4 = 0.
The third blow-up doesn’t need a translation, and sends x → u3 x − u, to yield
a3 = 1. A similar calculation at t = 1 shows that a1 = a3 . Thus,
x + t3
w=
t(t − 1)
whence
x = wt(t − 1) − t3 − t
Substituting the above in the original equation, and simplifying as usual, we
get the following new equation
y 2 = x3 + (t3 − t − 1)x2 + t5 x

t=0 A9
t=1 A3
t=∞ D6
This configuration has M W -rank 2, and has a 2-torsion section given by

(0, 0)

2-torsion (0, 0)
12. Fibration 10 : E7 , D4 , E7
y 2 = x3 + (t3 − t − 1)x2 + (t3 − 1)(t2 + t)x − t2 (t − 1)4
The original fibers are A9 at t = 0 (roots: a0 , a1 , · · · ), A3 at t = 1 (roots:

b0 , b1 , b2 ), and D6 at t = ∞ (roots: c0 , c1 , · · · ). The new E7 fiber is given by
F = 2O + 4a0 + 3a1 + 3a2 + 2a3 + 2a4 + a5 + a6 and the new section by
2 3
w = a0 +a1 t+at24t +a3 t +x
At t = 0, the singularity is at (0, 0). Hence the blow-up maps x → tx, which
gives a0 = 0. For the second blow-up we need to translate x to x + t. Thus, the
blow up sends x → t2 x + t, giving a1 = −1. The third blow-up does not require
a translation, and maps x → t3 x + t, which means a2 = 0. Similarly, we find
a4 = 0. Thus,
x−t
w= 4
t
that is,
x = wt4 + t
Substituting the above in the original equation, simplifying as usual, and trans-
forming the resulting quartic into the Weierstrass form, we get the new equation
y 2 = x3 + t3 (t + 1)2 x

t=0 E7
t = −1 D4
t=∞ E7
This fibration has M W -rank of 2 and a 2-torsion section

2-torsion (0, 0)
non-torsion (t + 1, t3 + 1)
(t3 (t + 1), −t3 (t3 + 1))
Note that the second non-torsion section is the sum of the first non-torsion and
the torsion section.
13. Fibration 11 : A15 , A3

y 2 = x3 + (t3 + t)x2 + t4 x
The original fibers are D8 at t = 0 (roots: a0 , a1 , · · · ), D8 at t = ∞ (roots:

b0 , b1 , b2 ), A1 at t = 1 (roots: c0 , c1 , · · · ), and A1 at t = −1 (roots: d0 , d1 , · · · ).
The new A15 fiber is given by F = O + P + a0 + a1 + a3 + a4 + a5 + a6 + a7 +
b0 + b1 + b3 + b4 + b5 + b6 + b7 , where P is the 2-torsion section (0, 0).
The new section is
a 0 + a 2 t2 y
w= +
t xt
The singularity at t = 0 is at (0, 0). The first blow-up sends x → tx and y → ty,
giving a pole of order 1. Continuing this for both t = 0 and t = ∞, we get the
new equation
y 2 = x3 + (t4 + 1)x2 + (t4 − 1)x + t4 − 1

t=0 A3
t=∞ A15
This fibration has M W -rank 2, and a 4-torsion section described below.

4-torsion (−1, t2 )
14. Fibration 12 : A11 , E6

We use the 2-neighbor construction on fibration 11. The original equation is
y 2 = x3 + (t4 + 1)x2 + t4 (1 − t4 )x + t8 (1 − t4 )
The original fibers are A15 at t = 0 (roots: a0 , a1 , · · · ), and A3 at t = ∞ (roots:

b0 , b1 , b2 , · · · ). The new E6 fiber is given by F = 2O+3a0 +2a1 +2a2 +a3 +a4 +b0 ,
2 4
and the new section w = a0 +a1 t+at23t +a4 t +x
The singularity at t = 0 is (0, 0), and none of the first three blow-ups require
a translation. Thus we get a0 = a1 = a2 = 0. At t = ∞, we replace t by u1 . We
need to translate x → x − 1. Thus, the first blow-up maps x → ux,a nd implies
that a4 = 1. Thus,
x + t4
w=
t3
whence
x = wt3 − t4
Substituting the above in the original equation, simplifying and converting to
the Weierstrass form, we get
y 2 = x 3 + x 2 + t4 x + t8

t=0 A11
t=∞ E6
This fibration has M W -rank 3, and a 3-torsion section, as given below.

3-torsion (0, t4 )
non-torsion (t , t + t2 )
2 4
(−t2 , t2 − t4 )
(t4 , t6 )
15. Fibration 13 : D6 , D12

We use the 2-neighbor construction on Fibration 11. The original equation is
y 2 = x3 + (t4 + 1)x2 + t4 (1 − t4 )x + t8 (1 − t4 )
The original fibers are A15 at t = 0 (roots: a0 , a1 , · · · ), A3 at t = ∞ (roots:

b0 , b1 , b2 , · · · ). The new D6 fiber is given by F = 2O+2a0 +a1 +a2 +2b0 +b1 +b2 ,
3 4
and the new section is w = a0 +a1 t+at32t +a4 t +x
Like in the previous calculation at t = 0, a0 = a1 = 0. At t = ∞, replace t
by u1 and translate x → x − 1. the first blow-up x → ux gives a4 = 1, and the
second blow-up (no translation needed) x → u2 x gives a3 = 0. Thus,
x + t4
w=
t2
Replacing
x = wt2 − t4
in the original equation, simplifying and converting to the Weierstrass equation,

we get
y 2 = x3 − (t3 + t)x2 + t6 x

t=0 D12
t=∞ D6
This has M W -rank of 2.

2-torsion (0, 0)

y 2 = x3 − tx2 − t4 (t + 1)x + t6 (t + 1)2
The original fibers are D7 at t = 0 (roots: a0 , a1 , · · · ), A5 at t = 1 (roots:

b0 , b1 , b2 , · · · ), and E6 at t = ∞ (roots: c0 , c1 , · · · ). The new A12 fiber is given
by O+P +a0 +a1 +a3 +a4 +a5 +a6 +b0 +b1 +b2 +b3 +b4 , where P is the section
y+t3 (t+1)
(0, t3 (t + 1)). The new section is w = a(t) t + b(t) xt . Since we want a pole
of order 1 at t = 0 and also at t = ∞, we get that deg(a(t)) = 2. Subtracting
suitable multiples of t from the fraction, we get a1 = 0. Also, deg(b(t)) = 0,
and by rescaling, we can assume b = 1. Thus,
a0 + a2 t2 y + t3 (t + 1)
w= +
t xt
As usual, blow-ups at t = 0 and t = ∞ gives both a0 = a2 = 0. Thus,
y + t3 (t + 1)
w=
tx
that is,
y = wtx − t3 (t + 1)
Substituting the above in the original equation, cancelling common factors,
converting to the Weierstrass form and simplifying, we get
y 2 = x3 + (t4 − t3 + 1)x2 − t(t − 1)2 x + t2

t=0 A6
t=∞ A12
This again has M W -rank 2.

non-torsion (0, t)
17. Fibration 15 : A11 , E7

y 2 = x3 + (t4 − t + 1)x2 − t5 (t − 1)2 x + t10

The original fibers are A12 at t = 0 (roots: a0 , a1 , · · · ), and A6 at t = ∞ (roots:

b0 , b1 , b2 , · · · ). The new E7 fiber is given by 2O + 4a0 + 3a1 + 3a2 + 2a3 + 2a4 +
2 3
a5 + a6 . The new section is w = a0 +a1 t+at24t +a3 t +x
At t = 0, the singularity is already at (0, 0). Hence we do not need to
translate, and can blow-up, sending x → tx. This gives a0 = 0. Similarly,
subsequent blow-ups give a1 = a2 = a3 = 0. Thus,
x
w=
t4
that is,
x = wt4
Substituting the above in the original equation, dividing by suitable powers of
t till we get a quartic in t, and then converting into Weierstrass form, we get
y 2 = x3 + (1 − t)x2 + t2 (1 − t)(1 + t)2 x + t4 (t3 + t2 + 1)

t=0 A11
t=∞ E7
This has M W -rank 1. The non-torsion section is described below.

non-torsion (t2 , t3 )
18. Fibration 16 : A5 , A5 , A5 , D4
y 2 = x3 − (t3 + t − 1)x2 + t3 (1 − t2 )x + t6 (t − 1)2

b0 , b1 ), D5 at t = −1 (roots: c0 , c1 , · · · ), and D5 at t = ∞ (roots: d0 , d1 , · · · ).
The new D4 fiber is given by 2O + a0 + b0 + c0 + d0 . The new section is
2 4
w = a0 +at(t−1)(t+1)
1 t+a2 t +a4 t +x
At t = 0, we blow-up once to get a0 = 0. At t = 1, we get a1 + a2 + a4 = 0.

Similarly, at t = −1, we get −a1 + a2 + a4 = 1. At t = ∞, blowing-up once
gives a4 = 0. Solving all of the above conditions, we get a1 = 1, a2 = −1. Thus,
t − t2 + x
w=
t(t2 − 1)
Substituting
x = wt(t2 − 1) + t2 − t
in the original equation, we get, after simplifying,
y 2 = x3 + x2 − t2 (t4 − 1)x + t4 (t − 1)2 (t + 1)2

t=0 A5
t=1 A5
t = −1 A5
t=∞ D4
This fibration has M W -rank 1. The non-torsion section is given below.

non-torsion (0, t2 (t2 − 1))
19. Fibration 17 : A8 , A2 , A8
y 2 = x3 + x2 − t2 (t4 − 1)x + t4 (t − 1)2 (t + 1)2

b0 , b1 , · · · ), A5 at t = −1 (roots: c0 , c1 , · · · ), and D4 at t = ∞ (roots: d0 , d1 , · · · ).
The new A8 fiber is given by O + P + a0 + a1 + · · · + b0 + b1 + · · · , where P =
2 2
(t −1)
(0, t2 (t2 − 1)) is a non-zero section. The new section is w = at(t−1) 0 +a1 t
+ y+t
xt(t−1) .
y
At t = 0, we choose the branch of A5 fiber such that x = 1. Thus, a0 = 1.
Similarly, at t = 1, we replace t by u + 1, and follow the same logic to get
a1 = 0. Thus,
−1 y + t2 (t2 − 1)
w= +
t(t − 1) xt(t − 1)
whereby
y = wtx(t − 1) + x − t2 (t2 − 1)
We substitute the above in the original equation, and follow the method de-
scribed in the Appendix, followed by renaming of variables, dividing by suitable
squares of polynomials in t, and then converting to the Weierstrass form to get
y 2 = x3 + (t + 1)4 x2 − t4 (t + 1)2 (t − 1)x + t8 (t + 1)2


t=0 A8
t = −1 A2
t=∞ A8
This has M W -rank 2.

non-torsion (0, t4 (t + 1))
20. Fibration 18 : D10 , E8

We do a 2-neighbor construction on Fibration 15. The original equation is
y 2 = x3 + (1 − t)x2 + t2 (1 − t)(1 + t)2 x + t4 (t3 + t2 + 1)
The original fibers are A11 at t = 0 (roots: a0 , a1 , · · · ), E7 at t = ∞ (roots:

b0 , b1 , · · · ). The new E8 fiber is given by F = 2O + a0 + 3b0 + 4b1 + 5b2 + 6b3 +
2 3 4
3b4 + 4b5 + 2b6 . The new section is w = a0 +a2 t +at3 t +a4 t +x
By blowing-up once at t = 0, we get a0 = 0. At t = ∞, we replace t by u1 as
usual, and use subsequent blow-ups to get a2 = −1, a3 = 0, and a4 = 0. Thus,
−t2 + x
w=
t
that is,
x = wt + t2
t, and converting the resulting quartic into its Weierstrass form, we get
y 2 = x3 − tx2 + t9

t=0 D10
t=∞ E8
This fibration has M W -rank 2.

y 2 = x3 − t3 x2 + t3 x
The original fibers are E7 at t = 0 (roots: a0 , a1 , · · · ), A2 at t = 1 (roots:

b0 , b1 , b2 ), and D10 at t = ∞ (roots: c0 , c1 , · · · ). The new A9 fiber is given by
F = O + P + a0 + a1 + a2 + · · · + b0 , where P = (0, 0) is the 2-torsion section.
y
The new section is w = at(t−1)0 +a1 t
+ xt(t−1)
Following the usual procedure of blowing up at t = 0 and t = 1, we get
both a0 = a1 = 0. Thus,
y
w=
xt(t − 1)
which means
y = wt(t − 1)x
polynomials of t, and renaming variables, we get
y 2 = x3 + (t4 + 1)x2 − t2 (t2 − 1)x + t4

t=0 A9
t=∞ A9
This fibration has M W -rank 2. It also has a 5-torsion section described

below.

5-torsion (0, t2 )
non-torsion (−1, it) (defined over F9 )

y 2 = x3 − t3 x2 + t3 x
The original fibers are E7 at t = 0 (roots: a0 , a1 , · · · ), A2 at t = 1 (roots:

b0 , b1 , b2 ), and D10 at t = ∞ (roots: c0 , c1 , · · · ). The new A17 fiber is given by
O + P + a0 + a1 + · · · + c0 + c1 + · · · , where P = (0, 0) is the 2-torsion section.
2
y
The new section is w = a0 +a
t
2t
+ xt . Following the usual procedure of blowing
up at t = 0 and t = ∞, we get both a0 = a1 = 0. Thus,
y
w=
xt
which means
y = wtx
Substituting the above in the original equation, completing squares, and divid-
ing by suitable powers of polynomials of t, and renaming variables, we get
y 2 = x 3 + t 4 x2 + t 2 x + 1
t=0 A2
t=∞ A17
This fibration has M W -rank of 1. There is a 3-torsion section described
below.
3-torsion (0, 1)
non-torsion (t, t3 + 1)
23. Fibration 21 : A7 , A7 , A1 , A1 , A1 , A1
We use the 2-neighbor construction on the Fibration 2. The original equation
is
y 2 = x3 − t2 (t − 1)2 (t + 1)2
The original fibers are D4 at t = 0 (roots: a0 , a1 , · · · ), D4 at t = 1 (roots:
b0 , b1 , b2 ), D4 at t = −1 (roots: c0 , c1 , · · · ), and D4 at t = ∞ (roots: d0 , d1 , · · · ).
The new A7 fiber is given by O + P + a0 + a1 + a2 + d0 + d1 + d2 , where
2
y
P = (0, 0) is the 2-torsion section. The new section is w = a0 +a t
2t
+ xt . Thus,
y = (wt − a0 − a2 t2 )x. Following the usual procedure of blowing up at t = 0
and t = ∞, we get both a0 = a1 = 0. Thus,
y
w=
xt
which means
y = wtx
polynomials of t, renaming variables, and converting into Weierstrass form, we
get
y 2 = x3 + (t4 + 1)x2 − x − (t4 + 1)

t=0 A7
t=1 A1
t = −1 A1
t=i A1
t = −i A1
t=∞ A7
This has M W -rank rank 2, and full 2-torsion .

2-torsion (1, 0)
(−1, 0)
(−t4 − 1, 0)
24. Fibration 22 : A11 , A2 , A2 , D4

y 2 = x3 + (t4 + 1)x2 + t4 x

b0 , b1 , b2 ), A1 at t = 1 (roots: c0 , c1 ),A1 at t = −1 (roots: d0 , d1 ), A1 at t = i
(roots: e0 , e1 ), A1 at t = −i (roots: f0 , f1 ). The new D4 fiber is given by
2 4
F = 2O + a0 + b0 + c0 + d0 . The new section is w = a0 +a1 t+a 2 t +a4 t +x
t(t2 −1) .
Blowing-up at t = 0, we get a0 = 0. At t = 1, we replace t by u + 1, and
shift x to x − 1 (so that the singularity is at (0, 0)), and then blow-up, to get
a1 + a2 + a4 = 1. At t = −1, we replace t by u − 1, shift x to x − 1, and blow-up
to get −a1 + a2 + a4 = 1. At t = ∞, blowing-up gives a4 = 0. Solving the above
equations, we get
a1 = 0, a2 = 1, a4 = 0
Thus,
t2 + x
w=
t(t2 − 1)
Substituting
x = wt(t2 − 1) − t2
in the original equation, dividing by suitable powers of polynomials of t, re-
naming variables, and converting into Weierstrass form, we get
y 2 = x 3 + x 2 − t6 x

t=0 A11
t=i A2
t = −i A2
t=∞ D4
This has M W -rank 1, and also a 4-torsion section.

4-torsion (it3 , it3 ) (defined over F9 )
25. Fibration 23 : D6 , A3 , A3 , D6
y 2 = x3 + (t4 + 1)x2 + t4 x

(roots: e0 , e1 ), and A1 at t = −i (roots: f0 , f1 ). The new D6 fiber is given by
3 4
2O + 2a0 + a1 + a2 + 2b0 + b1 + b2 . The new section is w = a0 +a1 t+at32t +a4 t +x .
Blowing-up at t = 0, we get a0 = 0 = a1 . At t = ∞, blowing-up, after the usual
change of variables, gives a3 = a4 = 0. Thus,
x
w=
t2
Substituting
x = wt2
y 2 = x3 + t(t2 + 1)x2 − t4 x − t5 (t2 + 1)

t=0 D6
t=1 A3
t = −1 A3
t=∞ D6
This has M W -rank of 2, and full 2-torsion.

2-torsion (t2 , 0)
(−t2 , 0)
(−t3 − t, 0)
26. Fibration 24 : D8 , D5 , D5
y 2 = x3 + (t4 + 1)x2 + t4 x

(roots: e0 , e1 ), and A1 at t = −i (roots: f0 , f1 ). The new D5 fiber is given by
2 3
2O + 2a0 + a1 + a2 + c0 + d0 . The new section is w = a0 +a1 t+a 2 t +a3 t +x
t2 (t2 −1) .
Like in the previous case, blowing-up at t = 0 gives a0 = 0 = a1 . Blowing-
up at t = 1 gives a2 + a3 = 1, and blowing up at t = −1 gives a2 − a3 = 1.
Solving which, we get a2 = 1, a3 = 0. Thus,
t2 + x
w=
t2 (t2 − 1)
Substituting
x = wt2 (t2 − 1) − t2
y 2 = x3 + (t − t3 )x2 − t4 (t + 1)x + t8

t=0 D8
t = −1 D5
t=∞ D5

2-torsion (t3 , 0)
non-torsion (0, t4 )
(t − t, −t4 + t3 − t2 )
2
27. Fibration 25 : A4 , D5 , A1 , A1 , A7
y 2 = x3 + (t4 + 1)x2 + t4 x

(roots: e0 , e1 ), and A1 at t = −i (roots: f0 , f1 ). The new A7 fiber is given by
F = O+P +a0 +· · ·+c0 +c1 . The new section is w = at(t−1) 0 +a1 t y
+ xt(t−1) .Successive
blowing-ups at t = 0 and t = 1 give a0 = 1 and a1 = −1. Thus,
1−t y
w= +
t(t − 1) xt(t − 1)
Substituting
y = (wt(t − 1) − 1 + t)x
y 2 = x3 + (t4 + t3 + 1)x2 + (t − 1)4 (t + 1)(t2 + 1)x + (t − 1)4 (t + 1)2 (t2 + 1)2
which simplifies to
y 2 = x3 + (t4 + t3 + 1)x2 − (t4 + t3 − t2 + t + 1)x;

t=0 D5
t = −1 A4
t=i A1
t = −i A1
t=∞ A7

2-torsion (0, 0)
28. Fibration 26 : A5 , E7 , D7
y 2 = x3 + t(t2 + 1)x2 − t4 x − t5 (t2 + 1)


b0 , b1 , b2 , b3 ), A3 at t = −1 (roots: c0 , c1 , c2 , c3 ),D6 at t = ∞ (roots: d0 , d1 , · · · ).
The new D7 fiber is given by a1 + a3 + 2a2 + 2a0 + 2O + 2b0 + b1 + b2 . The new
2 3
section is w = a0 +a1 t+a 2 t +a3 t +x
t2 (t−1)2 .
Successive blowing-ups at t = 0 give a0 = a1 = 0. Blowing-up at t = 1
after the usual change of variables u = t − 1, and translating x → x + 1, gives
a2 + a3 = −1. Blowing-up again, after translating x → x − u gives a2 = −1.
Hence, a3 = 0. Thus,
x − t2
w= 2
t (t − 1)2
Substituting
x = wt2 (t − 1)2 + t2
in the original equation, dividing by suitable powers of polynomials of t, and
renaming variables, we get
y 2 = x3 − t3 x2 − t3 x + t6

t=0 E7
t=1 A5
t=∞ D7
This has M W -rank 1. We describe the non-torsion section below.

29. Fibration 27 : A5 , A5 , D7
y 2 = x3 + t3 x2 + t3 (t2 − 1)x + t5 (t − 1)2
The original fibers are A5 at t = 0 (roots: a0 , a1 , · · · ), E7 at t = 1 (roots:

b0 , b1 , · · · ), D7 at t = ∞ (roots: c0 , c1 , · · · ). The new D7 fiber is given by
3 4
2O+2a0 +a1 +a2 +2c0 +c1 +2c2 +c3 . The new section is w = a0 +a1 t+at32t +a4 t +x .
Successive blowing-ups at t = 0 give a0 = a1 = 0. Similarly, blowing up at
t = ∞ gives a3 = a4 = 0. Thus,
x
w=
t2
Substituting
x = wt2
naming variables, we get
y 2 = x3 + (t3 + 1)x2 − (t3 − 1)x
t=0 A5
t=1 A5
t=∞ D7
This has a M W -rank of 3. It also has full 2 torsion.
2-torsion (0, 0)
(1, 0)
(1 − t3 , 0)

y 2 = x3 + x2 − (t6 + t2 )x − t4 (t4 − t2 − 1)
The original fibers are A11 at t = 0 (roots: a0 , a1 , · · · ), D4 at t = ∞ (roots:
b0 , b1 , b2 , b3 , b4 ), A2 at t = 1 (roots: c0 , c1 , c2 ), and A2 at t = −1 (roots:
d0 , d1 , d2 ). The new A14 fiber is given by O + P + a0 + · · · + c0 + c1 , where P =
y+t2 (t2 −1)
a +a t+
(0, t2 (t2 − 1)) is a non-zero section. The new section is w = 0 1 t(t−1) x .
Successive blowing-ups at t = 0 and t = 1 give a0 = 1 and a1 = 0. Thus,
1 y + t2 (t2 − 1)
w= +
t(t − 1) xt(t − 1)
Substituting
y = (wt(t − 1) − 1)x − t2 (t2 − 1)
in the original equation, dividing by suitable powers of polynomials in t, re-
y 2 = x3 + (t3 + 1)(t − 1)x2 − t6 (t − 1)2 x − t6 (t6 + 1)
t=0 A5
t=∞ A14
31. Fibration 29 : A5 , A5 , D10

y 2 = x3 + x2 − (t6 + t2 )x − t4 (t4 − t2 − 1)
The original fibers are A11 at t = 0 (roots: a0 , a1 , · · · ), D4 at t = ∞
(roots: b0 , b1 , b2 , b3 , b4 ), A2 at t = 1 (roots: c0 , c1 , c2 ), and A2 at t = −1
(roots: d0 , d1 , d2 ). The new A5 fiber is given by O + P + c0 + c1 + d0 + d1 ,
where P = (0, t2 (t2 − 1)) is a non-zero section. The new section is w =
a0 +a1 t y+t2 (t2 −1)
(t+1)(t−1) + x(t+1)(t−1) .
Successive blowing-ups at t = 0 and t = 1 give a0 = 1 and a1 = 0. Thus,
1 y + t2 (t2 − 1)
w= +
(t + 1)(t − 1) x(t + 1)(t − 1)
Substituting
y = (w(t + 1)(t − 1) − 1)x − t2 (t2 − 1)
y 2 = x3 + (t3 + 1)(t − 1)x2 − t6 (t − 1)2 x − t6 (t6 − 1)
t=0 A5
t=1 D10
t=∞ A5
The trivial lattice for this fibration is of rank 22, hence the M W -rank is 0.
We also have full 2-torsion (descriptions given below). The absolute discrimi-
nant of the trivial lattice is 6 · 6 · 4 = 144. Then since we have full 2-torsion,
we find, from the Shioda-Tate formula that the full lattice is indeed the full
N S(X),since it has absolute discriminant 32 , and signature (1, 21).
2-torsion (t4 − t3 , 0)
(−t4 + t3 , 0)
(−t + t3 − t + 1, 0)
4
32. Fibration 30 : A3 , A9 , A6
y 2 = x3 + x2 − (t6 + t2 )x − t4 (t4 − t2 − 1)
The original fibers are A11 at t = 0 (roots: a0 , a1 , · · · ), D4 at t = ∞ (roots:

b0 , b1 , b2 , b3 , b4 ), A2 at t = 1 (roots: c0 , c1 , c2 ), and A2 at t = −1 (roots:
d0 , d1 , d2 ). The new A6 fiber is given by O +P +c0 +c1 +b0 +b1 +b2 , where P =
2 2
+a2 t2 (t −1)
(0, t2 (t2 −1)) is a non-zero section. The new section is w = a0 t−1 + y+t
x(t−1) .
Successive blowing-ups at t = 1 and t = ∞ give a0 = 1 and a1 = 0. Thus,
1 y + t2 (t2 − 1)
w= +
t−1 x(t − 1)
Substituting
y = (w(t − 1) − 1)x − t2 (t2 − 1)
y 2 = x3 + (t4 + t − 1)x2 + t2 (t3 + t2 + 1)x + t4 (t − 1)2

t=0 A3
t=1 A9
t=∞ A6

non-torsion (0, t2 (t − 1))
33. Fibration 31 : D6 , A12

y 2 = x3 + (t3 − t2 )x2 − t3 (t + 1)x + t4
The original fibers are D5 at t = 0 (roots: a0 , a1 , · · · ), D5 at t = −1 (roots:

b0 , b1 , · · · ), and D8 at t = ∞ (roots: c0 , c1 , · · · ). The new A12 fiber is given by
F = O + P + a0 + a1 + · · · + c0 + c1 + · · · , where P = (0, t2 ) is a non-zero
2 2
section. The new section is w = a0 +a t
2t
+ y+t xt .
Successive blowing-ups at t = 0 and t = ∞ give a0 = 0 and a2 = 0. Thus,
y + t2
w=
tx
Substituting
y = wtx − t2

y 2 = x3 + (t4 + t3 + t)x2 + (−t4 + t2 + t)x + t4 − t3 − t2 + t − 1
t=0 D6
t=∞ A12
34. Fibration 32 : A5 , A2 , A5 , A5
y 2 = x3 + (t4 − t3 + t)x2 + t4 (t − 1)(t2 − t − 1)x + t4 (t − 1)2 (t2 − t − 1)2
b0 , b1 , · · · ), A7 at t = ∞ (roots: c0 , c1 , · · · ), A1 at t = i − 1 (roots: d0 , d1 ), and
A1 at t = −i − 1 (roots: e0 , e1 ). The new A5 fiber is given by F = O + P + b0 +
b1 + c0 + c1 , where P = (0, t2 (t − 1)(t2 − t − 1)) is a non-zero section. The new
2 2
+a2 t2 −t−1)
section is w = a0 t−1 + y+t (t−1)(t
x(t−1) .
Successive blowing-ups at t = 1 and t = ∞ give
a0 = 0 and a2 = −1
. Thus,
−t2 y + t2 (t − 1)(t2 − t − 1)
w= +
t−1 x(t − 1)
polynomials in t, renaming variables, and converting into Weierstrass form, we
get
y 2 = x3 + (t4 − t3 − t + 1)x2 + (t8 + t7 + t6 )x + t9 + t6
t=0 A5
t=1 A5
t = −1 A5
t=∞ A2
non-torsion (−t − 1, it)
(t2 , (t − 1)t2 (t + 1)2 )
35. Fibration 33 : A8 , D4 , E6
y 2 = x3 + (t4 + t3 − t + 1)x2 + (t5 − t3 − t2 − t)x + t6 + t5 + t3 + t2

b0 , b1 , · · · ), and A3 at t = ∞ (roots: c0 , c1 , c2 , c3 ). The new E6 fiber is
given by F = 2O + 3a0 + 2a1 + 2a2 + a9 + a8 + b0 . The new section is
2 4
w = a0 +a1 t+at23t +a4 t +x . Successive blowing-ups at t = 0 give a0 = 0, a1 =
1, a2 = −1 and blowing up at t = 1 gives a4 = 0. Thus
t − t2 + x
w=
t3
Substituting
x = wt3 + t2 − t
y 2 = x3 + (t + 1)x2 − (t5 + t4 )x + t8

t=0 A8
t = −1 D4
t=∞ E6

36. Fibration 34 : A8 , A2 , A2 , D7
y 2 = x3 + (t4 + t3 + t)x2 − (t5 + t3 )x + t6 − t5
The original fibers are D6 at t = 0 (roots: a0 , a1 , · · · ), A12 at t = ∞ (roots:

b0 , b1 , · · · ). The new D7 fiber is given by 2O + 2a0 + 2a1 + a2 + a3 + 2b0 + b1 + b2 .
3 4
The new section is w = a0 +a1 t+at32t +a4 t +x . Successive blowing-ups at t = 0
give a0 = 0, a1 = 0 and blowing up at t = ∞ gives a3 = a4 = 0. Thus
x
w=
t2
Substituting
x = wt2
y 2 = x3 + (t3 + 1)x2 + t3 x + t6

t=0 A8
t = −1 − i A2
t = −1 + i A2
t=∞ D7

non-torsion (−1, t3 )
37. Fibration 35 : D9 , D9
y 2 = x3 + (t4 + t3 + t)x2 − (t5 + t3 )x + t6 − t5
The original fibers are D6 at t = 0 (roots: a0 , a1 , · · · ), and A12 at t = ∞ (roots:

b0 , b1 , · · · ). The new D9 fiber is given by 2O + 2a0 + 2a2 + 2a3 + 2a4 + a5 +
3 4
a6 + 2b0 + b1 + b12 . The new section is w = a0 +a1 t+at32t +a4 t +x .
Successive blowing-ups at t = 0 give a0 = 0, a1 = 0 and blowing up at
t = ∞ gives a3 = 1, a4 = 0. Thus
t3 + x
w=
t2
Substituting
x = wt2 − t3
y 2 = x3 + (t3 + t)x2 + (t6 − t5 )x + t8

t=0 D9
t=∞ D9

38. Fibration 36 : D13 , E6

y 2 = x3 + (t3 + t)x2 + (t6 − t5 )x + t8
The original fibers are D9 at t = 0 (roots: a0 , a1 , · · · ), and D9 at t = ∞ (roots:

b0 , b1 , · · · ). The new D13 fiber is given by 2O + 2a0 + 2a1 + 2a3 + · · · + 2b0 +
3 4
2b1 + b2 + b3 . The new section is w = a0 +a1 t+at32t +a4 t +x .
Successive blowing-ups at t = 0 and t = ∞ give a0 = a1 = a3 = a4 = 0.
Thus
x
w= 2
t
Substituting
x = wt2 − t3
y 2 = x3 − tx2 + t12

t=0 D13
t=∞ E6

39. Fibration 37 : D18

y 2 = x3 + (t3 + t)x2 + (t6 − t5 )x + t8
The original fibers are D9 at t = 0 (roots: a0 , a1 , · · · ), D9 at t = ∞ (roots:

b0 , b1 , · · · ). The new D18 fiber is given by F = 2O + 2a0 + 2a1 + 2a3 + · · · +
3 4
2b0 + 2b1 + 2b3 + · · · . The new section is w = a0 +a1 t+at32t +a4 t +x .
Successive blowing-ups at t = 0 and t = ∞ give a0 = a3 = a4 = 0, a1 = 1.

Thus
x
w= 2
t
Substituting
x = wt2 − t3
y 2 = x3 − (t4 + t3 + t)x2 + (t7 + t6 )x + t12 − t11
t=0 D18
This again has M W -rank 2.
40. Fibration 38 : E6 , E6 , E8
y 2 = x3 + (−t3 + t)x2 − t4 (t + 1)x + t8
The original fibers are D8 at t = 0 (roots: a0 , a1 , · · · ), D5 at t = −1 (roots:
b0 , b1 , · · · ), and D5 at t = ∞ (roots: c0 , c1 , · · · ). The new E8 fiber is given by
F = 2O + 4a0 2+ 3a31 + 6a2 + 5a3 + 4a4 + 3a5 + 2a6 + a7 . The new section is
w = a0 +a1 t+at24t +a3 t +x .
Successive blowing-ups at give a0 = a1 = a2 = a3 = 0. Thus
x
w= 4
t
Substituting
x = wt4
y 2 = x3 + t3 (t + 1)4
t=∞ E8
E6
E6
This pseudo-elliptic fibration has M W -rank 0, since the trivial lattice is
already of rank 22. The absolute discriminant of the trivial lattice is already
32 . Thus, the trivial lattice is itself the full N S(X), with signature (1, 21) and
the prescribed absolute discriminant 32 .
41. Fibration 39 : A9 , D9
y 2 = x3 − (t3 + 1)x2 + t6 x

b0 , b1 , b2 ), and D7 at t = ∞ (roots: c0 , c1 , · · · ). The new D9 fiber is given
by F = 2O + a0 + b0 + a2 + 2c0 + 2c1 + 2c3 + · · · . The new section is w =
a0 +a1 t+a3 t3 +a4 t4 +x
t(t−1) .
Blowing-up at t = 0 gives a0 = 0. Blowing-up at t = 1 requires translating
x → x+1, and gives a1 +a3 +a4 = −1. Blowing-up at t = ∞ gives a3 = a4 = 0.
Thus, a1 = −1. Thus
x−t
w=
t(t − 1)
Substituting
x = wt(t − 1) + t
y 2 = x3 + (t3 − t − 1)x2 + (t5 + t3 )x + t7 − t6

t=0 A9
t=∞ D9

non-torsion (t , it ) (defined over F9 )
2 2
42. Fibration 40 : A2 , A2 , D16

y 2 = x3 − tx2 + t5 x

b0 , b1 , b2 ), and E7 at t = ∞ (roots: c0 , c1 , · · · ). The new D16 fiber is given by
F = 2O + 2a0 3+ 2a41 + 2a3 + · · · + 2c0 + 2c1 + 2c3 + · · · . The new section is
w = a0 +a1 t+at32t +a4 t +x .
Blowing-up at t = 0 and t = ∞ gives a0 = a3 = a4 = 0, and a1 = −1. Thus
−t + x
w=
t2
Substituting
x = wt2 + t

y 2 = x3 + (t4 + t)x2 + t8 x

t=0 D16
t=1 A2
t=∞ A2
Here the trivial lattice is of rank 22, hence the M W -rank is 0. The absolute
discriminant of the trivial lattice is 3·3·4. And the fibration has a clear 2-torsion
section, (0, 0). Thus, the trivial lattice, together with the 2-torsion section gives
a lattice of signature (1, 21) and absolute discriminant 32 . Hence it must be the
full N S(X).

2-torsion (0, 0)
43. Fibration 41 : A2 , A2 , A2 , A2 , A2 , A2 , A2 , A2 ,
A2 , A 2
The equation of this surface is due to Ito.
y 2 = x3 + t10 + t2
The fibers are defined over the base P1 (F9 ), with an A2 fiber at each point
of the projective line.

t=0 A2
t=1 A2
t = −1 A2
t=i A2
t=1+i A2
t = −1 + i A2
t = −i A2
t=1−i A2
t = −1 − i A2
t=∞ A2
For this pseudo-elliptic surface, the trivial lattice is already of rank 22, and
of absolute discriminant 310 . The torsion group is of order 34 , as described
below. Together, the trivial lattice and the 3-torsion sections account for a
lattice of signature (1, 21) and absolute discriminant 32 . Hence this gives the
full N S(X).
3-torsion (t , t(t4 − 1))
2
(t2 , −t(t4 − 1))

(−t2 , t(t4 + 1))
(−t2 , −t(t4 + 1))
(t6 , t(t8 − 1))
(t6 , −t(t8 − 1))
( t12 , t5 + t12 )
( t12 , −t5 − t12 )
44. Fibration 42 : D4 , A2 , A2 , A5 , A5
y 2 = x3 + t10 + t2
The original fibers are A2 ’s at all the points of P1 (F9 ). The new A5 fiber we
consider is F = O + P + a0 + a1 + b0 + b1 , where a0 and b0 are the identity
components of the fibers at t = 0 and t = 1 respectively, a1 and b1 are non-
identity components of the same, and P is the 3-torsion section (t2 , −t(t4 − 1)).
y+t(t4 −1)
Our new section is w = at(t−1)
0 +a1 t
+ (x−t 2 )t(t−1) .
Blowing-up successively at t = 0 and t = 1 gives a0 = a1 = 0. Substituting

the same in the above definition of w, and replacing
y = wt(t − 1)(x − t2 ) − t(t4 − 1)
in the original equation, converting to the Weierstrass form after completing

squares, renaming variables, and then simplifying, we get
y 2 = x3 + (t3 + 1)x2 + (t3 − t6 )x

t=0 A5
t=1 A5
t = −1 − i A2
t = −1 + i A2
t=∞ D4
This has M W -rank 2. This also has a 2-torsion section, which does not
come from a 4-torsion section.

2-torsion (0, 0)
non-torsion (−1, t3 )
45. Fibration 43 : A4 , A4 , D5 , D5
y 2 = x3 + (t3 + 1)x2 + (t3 − t6 )x
The original fibers are A5 at t = 0 (Roots: a0 , a1 , · · · ), A5 at t = 1 (Roots:

b0 , b1 , · · · ), A2 at t = −i − 1, A2 at t = −i + 1, and D4 at t = ∞. The new D5
fiber we consider is F = O + 2a0 + a1 + a5 + b0 , where a0 and b0 are the identity
components of the fibers at t = 0 and t = 1 respectively, ai ’s are non-identity
2 4
components of the t = 0 fiber. Our new section is w = a0 +a1 t+a 2 t +a4 t +x
t2 (t−1) .
Blowing-up successively at t = 0 and t = 1 gives a0 = a1 = a2 = a4 = 0.
Thus,
x
w= 2
t (t − 1)
Substituting
x = wt2 (t − 1)
y 2 = x3 − (t3 + t)x2 − t2 (t2 + t − 1)x + t4 (t2 + t − 1)2


t=0 D5
t=1+i A4
t=1−i A4
t=∞ D5

non-torsion (0, t2 (t2 + t − 1))
(t2 + t − 1, t(t + 1)(t2 + t − 1))
y 2 = x3 + (t3 + 1)x2 + (t3 − t6 )x
The original fibers are A5 at t = 0, A5 at t = 1, A2 at t = −i − 1 (Roots:

c0 , c1 , c2 ), A2 at t = −i + 1 (Roots: d0 , d1 , d2 ), and D4 at t = ∞. The new A3
fiber we consider is F = O + P + c0 + d0 , where c0 and d0 are the identity
components of the fibers at t = −i − 1 and t = i − 1 respectively, and P is the
y
2-torsion section (0, 0). Our new section is w = ta20−t−1 +a1 t
+ x(t2 −t−1) .
Blowing-up successively at t = 0 and t = 1 gives a0 = a1 = 0. Thus,
y
w=
x(t2 − t − 1)
Substituting
y = wx(t2 − t − 1)
y 2 = x3 − (t4 + 1)x2 + x + t8 − t4

t=0 A3
t=1 A3
t = −1 A3
t=i A3
t = −i A3
t=∞ A3

non-torsion (t2 , t(t2 − 1))
47. Fibration 45 : A18

This was calculated by Schuett in [Sc]
y 2 = x3 + (t4 + t3 + 1)x2 + (−t3 − t2 + t)x + t2 + t + 1
is the supersingular K3 with an A18 fiber.

t=∞ A18

(0, t − 1)
48. Fibration 46 : A5 , A5 , A8
y 2 = x3 + (t3 + 1)x2 + (t3 − t6 )x
The original fibers are A5 at t = 0, A5 at t = 1, A2 at t = −i − 1 (Roots:

c0 , c1 , c2 ), A2 at t = −i + 1 (Roots: d0 , d1 , d2 ), and D4 at t = ∞ (Roots:
e0 , e1 , e2 , e3 , e4 ). The new A8 fiber we consider is F = O + a0 + a1 + a2 + a3 +
P + e0 + e1 + e2 , where a0 and e0 are the identity components of the fibers at
t = 0 and t = ∞ respectively, and P is the 2-torsion section (0, 0). Our new
y y
section is w = a0 +a t
1t
+ xt . Our new section is w = a0 +a
t
1t
+ xt .
Blowing-up successively at t = 0 and t = ∞ gives a0 = 1, a1 = 0. Thus,
1 y
w= +
t xt
Substituting in the original equation, converting to the Weierstrass form after
completing squares, renaming variables, and then simplifying, we get
y 2 = x3 + (t4 + t)x2 − t5 x + t6 − t3 − 1

t=i+1 A5
t = −i + 1 A5
t=∞ A8
y 2 = x3 − (t4 + 1)x2 + (t4 − 1)x + t8 + t4 + 1
The original fibers are A3 ’s at t = 0, t = 1, t = −1, t = i, t = −i and at t = ∞.
The new A4 fiber we consider is F = O + P + a0 + b0 + b1 , where a0 and b0
are the identity components of the fibers at t = 0 and t = 1 respectively, b1 is
a non-identity fiber at t = 1, and P is the section (0, t4 − 1). Our new section
4
−1
is w = at(t−1)
0 +a1 t
+ y+t
xt(t−1) .
Blowing-up successively at t = 0 and t = 1 gives a0 = a1 = 1. Thus,
1+t y + t4 − 1
w= +
t(t − 1) xt(t − 1)
y 2 = x3 + (−t4 + t3 − t − 1)x2 + (t7 − t6 + t5 )x − t10 − t9 + t8 − t7 − t6 − t5
t=i A1
t = −i A1
t=0 A4
t=i+1 A4
t = −i + 1 A4
t=∞ A4
50. Fibration 48 : A2 , A2 , E8 , E8
We use the 3-neighbor construction on the fiber configuration E6 , E6 , E8 . The
original equation is
y 2 = x3 + t4 ∗ (t − 1)5
The original fibers are E6 at t = 0 (Roots: p0 , p1 , p2 , p3 , p4 , p5 , p6 where p0 , p4 , p6

are the simple components and p2 is the component with multiplicity 3), E6 at
t = ∞ (Roots: q0 , q1 , q2 , q3 , q4 , q5 , q6 where q0 , q4 , q6 are the simple components
and q2 is the component with multiplicity 3), E8 at t = 1 (Roots: r0 , r1 , r2 , . . . ).
The new E8 fiber we consider is F = 3O + 4p0 + 5p1 + 6p2 + 4p3 + 2p4 + 3p5 +
2 2 3 5 6
2q0 + q1 . Our new section is w = y+(a0 +a1 t+a2 t )x+b0 +bt41 t+b2 t +b3 t +b5 t +b6 t .
Blowing-up successively at t = 0 and t = ∞ gives a0 = 0, a1 = 0, a2 =
0, b0 = 0, b1 = 0, b2 = 0, b3 = i, b5 = 0, b6 = 0. Thus,
y + it3
w=
t4
y 2 = x3 + (t2 + 1)2 (t3 + t2 + 1)
This is a pseudo-elliptic fibration with M W -rank 0. Also since the determi-
nant obtained from the fiber components is already −9, hence it doesn’t have
torsion sections either.
51. Fibration 49 : A2 , A2 , A2 , A2 , E6 , E6
We use the 3-neighbor construction on the fiber configuration E6 , E6 , E8 . The
y 2 = x3 + t4 ∗ (t − 1)5
The original fibers are E6 at t = 0 (Roots: p0 , p1 , p2 , p3 , p4 , p5 , p6 where p0 , p4 , p6
are the simple components and p2 is the component with multiplicity 3), E6 at
t = ∞ (Roots: q0 , q1 , q2 , q3 , q4 , q5 , q6 where q0 , q4 , q6 are the simple components
and q2 is the component with multiplicity 3), E8 at t = 1 (Roots: r0 , r1 , r2 , . . . ,
where r0 is the simple component and r1 is the multiplicity 2 component inter-
secting r0 ). The new E6 fiber we consider is F = 3O+2p0 +p1 +2q0 +q1 +2r0 +r1 .
2 2 3 5 6
Our new section is w = y+(a0 +a1 t+a2 t )x+bt02 +b 1 t+b2 t +b3 t +b5 t +b6 t
(t−1)2 .
Blowing-up successively at t = 0 and t = ∞ gives a0 = 0, a1 = 0, a2 =
0, b0 = 0, b1 = 0, b2 = 0, b3 = 0, b5 = 0, b6 = 0. Thus,
y
w=
t2 (t − 1)2
Substituting y = wt2 (t − 1)2 in the original equation, converting to the Weier-
strass form after completing squares, renaming variables, and then simplifying,
we get
y 2 = x3 + t4 (t2 + 1)2
This is a pseudo-elliptic fibration with M W -rank 0. It has 3-torsion sections.

3-torsion (0, t2 (t2 + 1))
(0, −t2 (t2 + 1))
(−t2 , t2 (t2 − 1))
(−t2 , −t2 (t2 − 1))
52. Fibration 50 : A2 , E6 , E6 , E6
We use a 2-neighbor construction on the fiber configuration A5 , D7 , E6 . The
y 2 = x3 + (t4 − t)x2 + t6
The original fibers are A5 at t = 0 (Roots: p0 , p1 , p2 , p3 , . . . where p0 intersects
O, the zero section), D7 at t = ∞ (Roots: q0 , q1 , q2 , . . . where q0 intersects O),
E6 at t = 1. The new E6 fiber we consider is F = 2O + 3p0 + 2p1 + p2 + 2p5 +
2 4
p4 + q0 . Our new section is w = a0 +a1 t+at23t +a4 t +x .
Blowing-up successively at t = 0 and t = ∞ gives a0 = a1 = a2 = a4 = 0.
Thus,
x
w= 3
t
Substituting x = wt3 in the original equation, converting to the Weierstrass
form after completing squares, renaming variables, and then simplifying, we
get
y 2 = x3 + (t2 + 1)t4 (t − 1)2
This is another pseudo-elliptic fibration with MW rank 0.
53. Fibration 51 : A3 , A3 , A6 , A6
We use a 2-neighbor construction on the fiber configuration A8 , A2 , A2 , D7 . The
y 2 = x3 + (t3 + 1)x2 + t3 x + t6
The original fibers are A8 at t = 0 (Roots: p0 , p1 , p2 , p3 , . . . where p0 intersects

O, the zero section), A2 at t = −i − 1 (Roots: q0 , q1 , q2 where q0 intersects O),
A2 at t = i − 1 (Roots: r0 , r1 , r2 , where r0 intersects O), and D7 at t = ∞
(Roots:s0 , s1 , . . . ). The new A3 fiber we consider is F = O + P + p0 + q0 , where
3
a0 +a1 t+ y+t
x+1
P is the section (−1, t3 ). Our new section is w = t(t+i+1) .
Blowing-up successively at t = 0 and t = −i − 1 gives a0 = 0, a1 = i − 1.

Thus,
3
(i − 1)t + y+t
x+1
w=
t(t + i + 1)
y 2 = x3 +(−it4 −t3 +(i−1)t+1)x2 −t2 (t−i)(t3 +it2 +t−i−1)x+t4 (t−i)2 ((1−i)t2 −i)
t=0 A6
t=1 A3
t=i A6
t=∞ A3
This has MW rank 2.
54. Fibration 52 : A6 , A6 , D6
We use a 2-neighbor construction on the fiber configuration A3 , A3 , A6 , A6 . The
y2 = x3 + (−it4 − (i + 1)t3 − t − 1)x2 + (−t6 − it5 + t4 − (i + 1)t3 + −it2 )x
+(1 − i)t8 + (1 − i)t7 − it5 − (i + 1)t4
The original fibers are A3 at t = 0 (Roots: p0 , p1 , p2 , p3 where p0 intersects
O, the zero section), A3 at t = ∞ (Roots: q0 , q1 , q2 , q3 where q0 intersects
O), A6 at t = i, and A6 at t = 1. The new D6 fiber we consider is F =
3 4
2O + 2p0 + p1 + p3 + 2q0 + q1 + q3 . Our new section is w = a0 +a1 t+at32t +a4 t +x .
Blowing-up successively at t = 0 and t = ∞ gives a0 = a1 = a3 = a4 = 0.
Thus,
x
w= 2
t
Substituting x = wt2 in the original equation, converting to the Weierstrass
form after completing squares, renaming variables, and then simplifying, we
get
y2 = x3 − (it3 + it) ∗ x2 + (t8 − it7 + (i + 1)t6 + (−i + 1)t4 + it3 − it2
+(i − 1)t − 1)x + (it12 + it11 + (i + 1)t10 − it9 − it8 + t7 − t6
+(i + 1)t5 + (i − 1)t4 + it2 + i + 1)
This has MW rank 2.
55. Appendix A
Here we describe the method of 2-neighbor and 3-neighbor construction, which
I learnt from my advisor Prof. Abhinav Kumar. For a more detailed description,
we refer to [Ku1].
2-neighbor and 3-neighbor construction
X is an elliptic K3 surface, defined over the base field k of characteristic
3, whose minimal model is given by y 2 = x3 + a2 (t)x2 + a4 (t)x + a6 (t) (ai ∈
k[t], deg ai ≤ 2i). X admits an elliptic fibration over P1 , with generic fiber
E/k(t). O denotes the zero section and P, Q etc. are other sections. We want
to compute the global sections of OX (a1 O + a2 P + a3 Q + · · · + C). We know
that OX (P ) is a line bundle of degree 1, where P is a section. Riemann-Roch
implies h0 (nP ) = n, with sections
1 generating H 0 (OX (P ))
12 , x generating H 0 (OX (2P ))

13 , 1.x, y generating H 0 (OX (3P ))
etc.
For the 2-neighbor construction, we often consider D = 2O. The basis for
Γ(OX (D)) is in this case given by 1 and x. If F1 , an elliptic divisor, is given by
F1 = D +C, then we have 2 linearly independent global sections to Γ(OX (F1 )),
given by 1 and a(t)+b(t)x for some a(t), b(t). The new elliptic parameter, which
we denote by w throughout in this paper, is given by the ratio of these two
sections. Thus w = a(t)+b(t)x. Here, a(t) is a rational function with numerator
of degree at most 4, and b(t) has a constant numerator, which we assume to be
1 by rescaling x. The denominators of a(t) and b(t) are polynomials giving the
correct order of poles at the singular fibers. Then we follow Tate’s algorithm
of successive blow-ups to pin down the coefficients of a(t), b(t) so that they
give the correct order of poles at the respective irreducible components of the
fibers. Ultimately we write x = (w − b(t))/a(t). This when substituted into the
original equation gives
y 2 = h(t, u)
where h, after suitable change of coordinates to absorb square factors into y 2 ,
is a cubic or quartic polynomial in t.
The other case when D = O +P is similar, and has been explained in [Ku1].
In some cases we need to use the 3-neighbor construction, in which our
divisor D is of the form D = 3O + G or 2O + P + G and so on. In this case
the basis for the space of global sections is given by 1, x, y. Thus we look at
sections of the form w = a(t) + b(t)x + c(t)y, where a(t) is a rational function
with numerator a polynomial of degree 6, the numerator of b(t) is one of degree
2, and that of c(t) is a constant which we assume to be 1 by rescaling y. All

these rational functions have denominators giving the correct order of poles
at the singular fibers. We then use successive blow-ups, as usual, to pin down
w. Next step is to write y in terms of x, t and w, and replace in the original
equation. This we then simplify to get an equation in x and t of total degree 3,
with coefficients in k[w]. This we then convert into a Weierstrass form, using a
rational point on the curve.
The method of conversion of the cubic or quartic into the Weierstrass form
can be read off from [Ku1]
Acknowledgements
I thank Abhinav Kumar and Noam Elkies for many helpful discussions and
suggestions. The computer algebra systems PARI/gp and Maxima were used
in the calculations for this paper. I thank the developers of these programs.
References
[Ar] M. Artin, Supersingular K3 surfaces, Ann. Sci. Ecole Norm. Sup. (4) 7 (1974),
543-567 (1975).
[Bo] R. Borcherds, Automorphism groups of Lorentzian lattices, J. Algebra 111
(1987). 133–153.
[Co] J. H. Conway, Three lectures on exceptional groups, in Finite Simple Groups,
pp. 215–247, Academic Press, New York, 1971.
[I] H. Ito, The Mordell-Weil groups of unirational quasi-elliptic surfaces in char-
acteristic 3, Math. Z. 211 (1992), 1–40.
[Ku1] A. Kumar, Elliptic fibrations on a generic Jacobian Kummer surface, (2011)
arXiv:1105.1715v2 [math.AG].
[Ku2] A. Kumar, K3 surfaces associated with curves of genus two, Int. Math. Res.
Not. IMRN 2008, no. 6, Art. ID rnm165, 26 pp.
[KS] M. Kuwata and T. Shioda, Elliptic parameters and defining equations for ellip-
tic fibrations on a Kummer surface, Algebraic geometry in East Asia – Hanoi
2005, 177–215, Adv. Stud. Pure Math. 50, Math. Soc. Japan, Tokyo, 2008.
[N1] V. V. Nikulin, Finite groups of automorphisms of Kahlerian surfaces of type
K3, Uspehi Mat. Nauk 31 (1976), no. 2(188), 223-224.
[N2] V. V. Nikulin, Integral symmetric bilinear forms and some of their applica-
tions, Math. USSR Izv. 14 (1980), 103-167.
[Ni] K. Nishiyama, The Jacobian fibrations on some K3 surfaces and their Mordell-
Weil groups, Japan. J. Math. (New Series) 22 (1996), no. 2, 293–347
[O] K. Oguiso, On Jacobian fibrations on the Kummer surfaces of the product of

nonisogenous elliptic curves, J. Math. Soc. Japan 41 (1989), no. 4, 651–680.
[Og] A. Ogus, Supersingular K3 crystals, Asterisque, 64 (1979), 3–86.
[PSS] I. Piatetski-Shapiro and I. R. Shafarevich, A Torelli theorem for algebraic sur-
faces of type K3 , Math. USSR Izv. 5 (1971), 547–587.
[Sc] M. Schuett, K3 surfaces with Picard rank 20, (2010) arXiv:0804.1558v3
[math.NT].
[Sh1] T. Shioda, Classical Kummer surfaces and Mordell-Weil lattices, Algebraic
geometry, 213–221, Contemp. Math. 422, Amer. Math. Soc., Providence, RI,
2007.
[Sh2] T. Shioda, On the Mordell-Weil lattices, Comment. Math. Univ. St. Paul. 39
(1990), no. 2, 211–240.
[Si] J. H. Silverman, Advanced topics in the arithmetic of elliptic curves, Graduate
Texts in Mathematics, 151. New York: Springer-Verlag, (1994).
[St] H. Sterk, Finiteness results for algebraic K3 surfaces, Math. Z. 189 (1985),
no. 4, 507–513.
[V] Vinberg, Some arithmetical discrete groups in Lobachevskii spaces, in Proc.
Internat. Colloq. on Discrete Subgroups of Lie Groups and Applications to
Moduli (Bombay 1973), pp. 323–348. Oxford University Press, Bombay, 1975.

Analytic and Algebraic Geometry

Uploaded by

Copyright:

Available Formats

Analytic and Algebraic Geometry

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Analytic and Algebraic Geometry

Uploaded by

Copyright:

Available Formats

What are the main topics discussed in the document?

What are the main topics discussed in the document?

What are some of the references cited in the document?

What are some of the references cited in the document?

Anilatmaja Aryasomayajula

Analytic and Algebraic

Indranil Biswas A.J. Parameswaran

ISBN 978-981-10-5648-2 (eBook)

Library of Congress Control Number: 2017946031

This Springer imprint is published by Springer Nature

Anilatmaja Aryasomayajula is assistant professor at Indian Institute of

Indranil Biswas is professor of mathematics at the Tata Institute of

Archana S. Morye is a faculty in the University of Hyderabad, India. She did

A.J. Parameswaran is professor of mathematics at the Tata Institute of

Mathematics Subject Classiﬁcation (2000). 14G99, 14L20, 14L30

∗ The author thanks the project TOFIGROU (ANR-13-PDOC-0015-01)

de la “composante continue” du “vrai” groupe fondamental, correspondant à la classiﬁcation

a satisfactory specialization theory of the fundamental group should consider

Though in a footnote in the same page he recalls us that2

this extremely seducing conjecture is unfortunately contradicted by an unpub-

the question of constructing a “true” fundamental group was neverthless very

2.1. Tannakian description. Let k be a perfect3 ﬁeld, X a reduced

have a speciﬁc name.

fundamental group scheme of X in x. Though in general a group scheme over

2.2. Description as pro-ﬁnite limit. There is a second construction

Proof. The strategy is the following: we take three objects of F(X)

Remark 2.5. Let k be a perfect ﬁeld, X a reduced and connected scheme,

3. Over a Dedekind Scheme

1. for every s ∈ S the ﬁbre Xs is reduced;

2. for every z ∈ X\Xη the local ring OX,z is integrally closed.

Then the category F(X) is coﬁltered.

Y1,η ×Y0,η Y2,η → Y1,η ×Xη Y2,η

(which is a morphism of torsors, we simply omitted the structural groups and

Again, not surprisingly, we have a fundamental group scheme π qf (X, x) of X

which is a schematically dominant morphism (i.e. the dual morphism on the

π(Xs , xs ) → π qf (X, x)s

is more likely to be an isomorphism than

π(Xs , xs ) → π(X, x)s

4.1. The bumpy topology. We are going to deﬁne a new Grothendieck

Now we give an intuitive deﬁnition of a bumpy torsor and in Proposition

Proof. On one direction we assume that Y → X is a G-torsor i.e. the canonical

Every G-invariant morphism Y → X which is a torsor for the fpqc topology

stress the non-ﬂatness of certain morphisms. Hereafter some examples of such

4.2. The bumpy fundamental group scheme. Let S be any

torsor we will mean a G-torsor Y → X endowed with a section y ∈ Yx (S). Let

where σY and σZ denote the actions of G on Y and H on Z respectively.

G1 , y2 ×y0 y1 ). We need to prove that it belongs to Qf (X, x). The assumption

Corollary 4.13. Let X → S be a surjective morphism of ﬁnite type with the

Proof. Very naturally from Theorem 4.12 we have, writing shortly,

(X Bqf , π Bqf (X, x), xBqf ) := lim(Yi , Gi , yi )

(X B , π B (X, x), xB ) := lim(Yi , Gi , yi )

5. Over a Dedekind Scheme (Revisited)

Theorem 5.1. Let S be a Dedekind scheme, X a faithfully ﬂat S-scheme of

1. for every s ∈ S the ﬁbre Xs is reduced;

then if (1) is satisﬁed the natural morphism π B (X, x) → π(X, x) is an iso-

complete the following diagram:

where v is an isomorphism, i and j are closed immersion so u is a closed

where Δ2 is a closed immersion as Yη → X is propre (by assumption, cf.

Δ1 (Yη /Gη ) = Δ1 (p(Yη )) = (p × p)(Δ2 (Yη ))

f (z)2ΩX := y 2 |f (z)|2 . (11)

f (z)2ωX := y 1/2 |f (z)|2 . (12)

f (z)2ωX,k,ν := y k |f (z)|2 . (14)