Derived Categories in Algebraic Geometry - Yujiro Kawamata

EMS Series of Congress Reports
EMS Congress Reports publishes volumes originating from conferences or seminars focusing on any field
of pure or applied mathematics. The individual volumes include an introduction into their subject and
review of the contributions in this context. Articles are required to undergo a refereeing process and are
accepted only if they contain a survey or significant results not published elsewhere in the literature.
Previously published:
Trends in Representation Theory of Algebras and Related Topics, Andrzej Skowroński (ed.)
K-Theory and Noncommutative Geometry, Guillermo Cortiñas et al. (eds.)
Classification of Algebraic Varieties, Carel Faber, Gerard van der Geer and Eduard Looijenga (eds.)
Surveys in Stochastic Processes, Jochen Blath, Peter Imkeller and Sylvie Rœlly (eds.)
Representations of Algebras and Related Topics, Andrzej Skowroński and Kunio Yamagata (eds.)
Contributions to Algebraic Geometry. Impanga Lecture Notes, Piotr Pragacz (ed.)
Geometry and Arithmetic, Carel Faber, Gavril Farkas and Robin de Jong (eds.)
Derived Categories in
Algebraic Geometry
Tokyo 2011
Yujiro Kawamata
Editor
Editor:
Yujiro Kawamata Department of Mathematics

Graduate School of Mathematical Sciences Faculty of Science
University of Tokyo King Abdulaziz University
Komaba, Meguro-ku P. O. Box 80257
Tokyo 153-8914, Japan Jeddah 21589, Saudi Arabia
E-mail: kawamata@ms.u-tokyo.ac.jp Department of Mathematics
Korea Institute for Advanced Study
85 Hoegiro, Dongdaemun-gu
Seoul 130-722, Republic of Korea
2010 Mathematics Subject Classification: Primary: 13D09, 14-02, 14-06, 14F05, 16E35, 18E30;
Secondary: 13D02, 13D10, 14A22, 14E08, 14E16, 14E30, 14J32, 14J33, 14J45, 14K05, 14L24, 14N35,
16E05, 16E40, 16F60, 18G10, 20G05, 20G15
Key words: Algebraic variety, derived category, triangulated category, Fourier–Mukai transform, cluster
algebra, birational geometry, semi-orthogonal decomposition, exceptional collection, minimal model,
flop, McKay correspondence, categorical action, abelian variety, non-commutative algebraic geometry,
mirror symmetry, Donaldson–Thomas theory
ISBN 978-3-03719-115-6
The Swiss National Library lists this publication in The Swiss Book, the Swiss national bibliography,
and the detailed bibliographic data are available on the Internet at http://www.helveticat.ch.
This work is subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation,
broadcasting, reproduction on microfilms or in other ways, and storage in data banks. For any kind of
use permission of the copyright owner must be obtained.
© 2012 European Mathematical Society
Contact address:
European Mathematical Society Publishing House
Seminar for Applied Mathematics
ETH-Zentrum SEW A27
CH-8092 Zürich
Switzerland
Phone: +41 (0)44 632 34 36
Email: info@ems-ph.org
Homepage: www.ems-ph.org
Typeset using the authors’ TEX files: I. Zimmermann, Freiburg

Printing and binding: Beltz Bad Langensalza GmbH, Bad Langensalza, Germany
∞ Printed on acid free paper
987654321
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Categorical representability and intermediate Jacobians of Fano threefolds

Marcello Bernardara and Michele Bolognesi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Fourier–Mukai functors: a survey

Alberto Canonaco and Paolo Stellari . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Flops and about: a guide

Sabin Cautis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
A note on derived categories of Fermat varieties

Akira Ishii and Kazushi Ueda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Homology of infinite loop spaces

Dmitry Kaledin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Cluster algebras and derived categories

Bernhard Keller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Some derived equivalences between noncommutative schemes and algebras

Izuru Mori . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Lagrangian-invariant sheaves and functors for abelian varieties

Alexander Polishchuk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Generic vanishing filtrations and perverse objects in derived categories

of coherent sheaves
Mihnea Popa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
The fundamental group is not a derived invariant

Christian Schnell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Introduction and open problems of Donaldson–Thomas theory

Yukinobu Toda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
vi Contents
Notes on formal deformations of abelian categories

Michel Van den Bergh . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
List of Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

Introduction
Triangulated categories are subjects of newly developing research areas which attract
young people from various fields of mathematics including abstract algebra, alge-
braic geometry, representation theory and theoretical physics. The derived category
of complexes of sheaves on a space is a triangulated category appearing in algebraic
geometry. An important observation is that the derived category has more hidden sym-
metries than the abelian category of sheaves. Namely there are non-isomorphic or even
non-birational algebraic varieties which have equivalent derived categories. One might
say that the derived categories provide a new concept of spaces. This aspect provides
a new insight in birational geometry. The homological mirror symmetry conjecture of
Kontsevich is another source of inspirations.
The concept of a triangulated category such as the derived category of sheaves
on an algebraic variety was invented by Grothendieck and Verdier in the 1960s as a
tool to express important results in algebraic geometry such as the duality theorem.
The study of derived categories of individual algebraic varieties started in Russia and
Japan simultaneously and independently in the 1970s. Beilinson and independently
Gelfand and Gelfand proved that the derived categories of projective spaces have very
nice explicit structures. Kapranov extended this result to more varieties of negative
Kodaira dimension. On the other hand, Mukai studied moduli spaces of sheaves on
abelian varieties, and discovered so-called Fourier–Mukai transforms. He proved that
a non-isomorphic pair of varieties may have equivalent derived categories, as in the
case of an abelian variety and its dual variety. He also proved similar results for K3
surfaces. The Fourier–Mukai transform is a new method for analyzing sheaves on
varieties by changing the slices of the derived category given by its t-structures. In the
1990s, Bondal and Orlov proved a reconstruction theorem saying that a Fano manifold
or canonically polarized manifold is recovered from its derived category. Moreover
they proved that some standard birational transformations such as simple blowing ups
along smooth centers correspond to semi-orthogonal decompositions of derived cate-
gories. This result suggested that there is a close relationship between the theory of
derived categories and the minimal model theory in the birational geometry, in a way
that varieties related by a flop have equivalent derived categories, and those related
by a divisorial contraction or a flip are described by a semi-orthogonal decomposi-
tion of derived categories. Bridgeland, King and Reid proved a generalized McKay
correspondence for three dimensional Gorenstein quotient singularities using the lan-
guage of derived categories. They found a new application of the derived categories
in showing that even the existence of crepant resolutions of such singularities can be
proved.
This book contains articles concerning more recent development of this rapidly de-
veloping field. These are mostly survey articles. The contribution by Bernardara and
viii Introduction
Bolognesi treats the conjectural correspondence by Orlov between the semi-orthogonal

decompositions of the derived categories and the direct sum decompositions of the
motives, which is reflected to the intermediate jacobians and the rationality question.
Canonaco and Stellari review results on the existence and the uniqueness of the kernels
for an equivalence functor between derived categories. Cautis surveys a new tech-
nique for proving a conjecture which says that the varieties connected by a flop have
equivalent derived categories. The case of stratified Atiyah or Mukai flops is confirmed
by using the representation theory of categorical Lie group actions, revealing an un-
expected relation to the knot homology groups. Ishii and Ueda give a nice concrete
description for the derived category of a Fermat variety by using an exceptional col-
lection on its quotient stack. Kaledin proves a simple expression for the homology of
a connected spectrum in terms of the associated special -space. Keller’s contribution
is an excellent survey on the relationship between the cluster algebras and the derived
categories by an authority of this field. Mori surveys non-commutative geometry from
the view point of the derived categories, where non-commutative schemes and non-
commutative algebras appear as incarnations of derived categories carrying the new
concept of spaces. Polishchuk’s paper treats the generalization of semi-homogeneous
vector bundles, which appeared in the context of the original Fourier–Mukai transform
for abelian varieties, to the Lagrangian invariant objects. Popa reviews a natural but
unexpected relationship between the generic vanishing theorems in birational geome-
try and the perverse coherent sheaves in the context of derived categories. The short
and elegant paper by Schnell proves that the fundamental group is not an invariant of
the derived category though the Hodge numbers are conjecturally derived invariants.
Toda surveys Donaldson–Thomas theory on the counting problem in terms of derived
categories, explaining relationships to the stability conditions and Gromov–Witten
invariants, and the invariance under flops. Van den Bergh treats a non-commutative
analogue of a classical result that the quadrics and projective line bundles are preserved
under deformations.
Tokyo, November 2012 Yujiro Kawamata

Categorical representability and intermediate
Jacobians of Fano threefolds
Marcello Bernardara and Michele Bolognesi
1 Introduction
These notes arise from the attempt to extend the results of [13] to a wider class of com-
plex threefolds with negative Kodaira dimension. If Y ! S is a conic bundle and S
is a rational surface, a semiorthogonal decomposition of Db .Y / by derived categories
of curves and exceptional objects gives a splitting of the intermediate Jacobian as the
direct sum of the Jacobians of the curves ([13], Theorem 1.1). This result is based
on the relation between fully faithful functors Db ./ ! Db .Y / (where is a smooth
projective curve) and algebraic cycles on Y . It turns out that the properties needed to
prove this result hold true also for certain threefolds other than conic bundles. One of
the aims of this article is to describe certain varieties satisfying these representability
assumptions. At the same time, semiorthogonal decompositions of rational conic bun-
dles over minimal rational surfaces are described in [13]. These turn out to be the most
recent examples in a quite extensive list of varieties (starting with [18]) of dimension
3 with negative Kodaira dimension admitting a semiorthogonal decomposition by ex-
ceptional objects and components which should somehow be related to the birational
properties. The possible interplays between derived categories and birational geometry
have been outlined in [19]. Recently, a challenging conjecture of Kuznetsov [49] has
added cubic fourfolds to the list.
In a generalization attempt, we define a new notion of representability based on
semiorthogonal decompositions, which we expect to carry useful geometrical insights
also in higher dimensions, and which allows to properly write down many of the ideas
which have been motivating these researches. Let X be a smooth projective variety of
dimension n. We define categorical representability in (co)dimension m for X , roughly
by requiring that the derived category Db .X / admits a semiorthogonal decomposition
by categories which can be fully faithfully embedded into derived categories of smooth
projective varieties of dimension bounded by m (resp. n m).
The idea of defining categorical representability comes from the classical theory
of algebraic cycles: various notions of representability of the group AiZ .X / of al-
gebraically trivial cycles of codimension i on X have appeared through the years in
the literature, and it seems interesting to understand their interactions with categorical

It is a pleasure to thank A. Iliev for his comments on a preliminary version of this paper. The authors
are grateful to the anonymous referee for pointing out an unprecise statement in an early version, and to
A. Chatzistamatiou and Ch.Vial for useful hints. The first named author was supported by the SFB/TR 45
‘Periods, moduli spaces, and arithmetic of algebraic varieties’.
2 M. Bernardara and M. Bolognesi
representability, as our examples suggest. Roughly speaking (for the actual defini-
tions see Section 2.2), weak representability for AiZ .X / is given by an algebraic map
J./ ! AiZ .X/ whose kernel is an algebraic group, for an algebraic curve . Working
with rational coefficients (that is, on AiQ ) gives the notion of rational representability.
Algebraic representability requires the existence of a quasi-universal regular isomor-
phism AiZ .X/ ! A onto an abelian variety A. Finally, if dim .X / D 2n C 1 is odd, A
is the algebraic representative of AnZ .X /, and if the principal polarization of A is “well
behaved“ with respect to this regular isomorphism we say that A carries an incidence
polarization.
The definition of categorical representability could seem rather disjoint from the
classical ones. It is nevertheless clear that rational representability is strongly related
to the structure of the motive h.X / of X . Grothendieck (or classical) motives were
introduced to give an algebraic universal description of cohomologies and cycles on
X. In particular one gets a big amount of information from a Chow–Künnneth de-
composition of the motive h.X /, which is, roughly, a decomposition whose summands
are strictly related to algebraic cycles of a given codimension. For example, if X is a
threefold, then rational representability of all the AiQ .X / is equivalent to the existence
of a specific Chow–Künneth decomposition [23]. A first point to note is then that
the existence of a fully faithful functor between the derived categories of two smooth
projective varieties should be reflected at a motivic level, as stated in the following
conjecture by Orlov.
Conjecture 1.1 ([64]). Let X and Y be smooth projective varieties and ˆ W Db .Y / !
Db .X/ be a fully faithful functor. Then the motive h.Y / is a direct summand of the
motive h.X/.
A clear link between categorical and rational representability should appear when
we consider the former in dimension 1. Note that being categorically representable
in dimension 1 is equivalent to the existence of a semiorthogonal decomposition by
exceptional objects and derived categories of curves. The motive of a curve splits into
two discrete and one abelian motives, the latter corresponding to the Jacobian up to
isogeny. Orlov’s conjecture would then imply that if X is categorically representable
in dimension 1, then its motive is a finite sum of abelian (corresponding to Jacobians of
curves) and discrete motives. This would give information about rational representabil-
ity for AiQ .X/. Being categorically representable in dimension 1 seems to be in fact a
very strong condition. For example a smooth cubic threefold is strongly representable
with incidence property but not categorically representable, otherwise we would have
the splitting of the intermediate Jacobian (see Corollary 3.10). Notice that in [50] the
study of the Abel–Jacobi map for some hypersurfaces and its link with categorical
constructions were already treated.
On the other hand, algebraic representability and the incidence property can have
deep interactions with categorical representability, and this is indeed the heart of the
proof of Theorem 1.1 in [13]. Consider a smooth projective threefold X and assume
it to be rationally representable, with h1 .X / D h5 .X / D 0 (i.e. X has discrete Picard
Categorical representability and intermediate Jacobians of Fano threefolds 3
group), and with A2Z .X / algebraically representable with the incidence property. The
arguments in [13] show that if X is categorically representable in dimension 1, then
the intermediate Jacobian J.X / splits into Jacobians of curves, namely of those curves
of positive genus appearing in the semiorthogonal decomposition. This result can then
be applied to a large class of complex threefolds with negative Kodaira dimension (see
a list in Remark 3.8).
We can then reasonably raise the following question, which also points out how
this new definition could be useful: is categorical representability in codimension 2
a necessary condition for rationality? This is true for complex surfaces, since any
rational smooth projective complex surface admits a full exceptional sequence. Re-
mark 3.12 shows that this is true for a wide class of complex threefolds with negative
Kodaira dimension, but we can only argue so far by a case by case analysis. In di-
mension 4, Kuznetsov’s conjecture about rationality of cubics ([49], Conjecture 1.1)
is clearly related to (and indeed stronger than) this question, while in [4] we state a
similar conjecture for another class of smooth projective fourfolds explicitly in terms
of categorical representability.
Notation. Any triangulated category is assumed to be essentially small. Given a

smooth projective variety X , we denote by X its Kodaira dimension, by Db .X / the
bounded derived category of coherent sheaves on it, by K0 .X / its Grothendieck group,
by CHdZ .X/ the Chow group of codimension d cycles modulo rational equivalence,
and by AdZ .X/ the subgroup of algebraically trivial cycles in CHdZ .X /. If X is pure
d -dimensional, and Y any smooth projective variety, we denote by Corr i .X; Y`/ WD
CHiCd
Q .X Y / the group of correspondences with rational coefficients. If X D Xj ,
L
with Xj connected, then Corr .X; Y / D
i
Corr i .Xj ; Y /.
2 Categorical and classical representabilities for smooth

projective varieties
2.1 Semiorthogonal decompositions and categorical representability. We start by
recalling some categorical definitions which are necessary to define representability.
Let K be a field and T a K-linear triangulated category. A full triangulated subcategory
A of T is called admissible if the embedding functor admits a left and a right adjoint.
Definition 2.1 ([17], [18]). A semiorthogonal decomposition of T is a sequence of

admissible1 subcategories A1 ; : : : ; Al of T such that
• HomT .Ai ; Aj / D 0 for all i > j and any Ai in Ai and Aj in Aj ;

1
Notice that some authors, as for example [66], do not require admissibility for the subcategories in the
definition of semiorthogonal decomposition.
• for all objects Ai in Ai and Aj in Aj , and for every object T of T, there is a chain
of morphisms 0 D Tl ! Tl1 ! ! T1 ! T0 D T such that the cone of
Tk ! Tk1 is an object of Ak for all k D 1; : : : ; l.
Such a decomposition will be written
T D hA1 ; : : : ; Al i:
Definition 2.2 ([16]). An object E of T is called exceptional if HomT .E; E/ D K, and

HomT .E; EŒi/ D 0 for all i ¤ 0. A collection fE1 ; : : : ; El g of exceptional objects is
called exceptional if HomT .Ej ; Ek Œi / D 0 for all j > k and for all integers i .
If E in T is an exceptional object, the triangulated category generated by E (that
is, the smallest full triangulated subcategory of T containing E) is equivalent to the
derived category of a point, seen as a smooth projective variety. The equivalence
Db .pt/ ! hEi T is indeed given by sending Opt to E. In the case where T is Db .X /
for a smooth projective variety X , given an exceptional collection fE1 ; : : : ; El g, there
is a semiorthogonal decomposition (see [18])
Db .X / D hA; E1 ; : : : ; El i;
where A is the full triangulated subcategory whose objects are all the A satisfying
Hom.Ei ; A/ D 0 for all i D 1; : : : ; l; and we denote by Ei the category generated
by Ei . We say that the exceptional sequence is full if the category A is trivial. More
generally, if A T is admissible, we have two semi-orthogonal decompositions
T D hA? ; Ai D hA;? Ai;
where A? and ? A are respectively the left and right orthogonal of A in T [17].
Definition 2.3. A triangulated category T is representable in dimension m if it admits
a semiorthogonal decomposition
T D hA1 ; : : : ; Al i;
and for all i D 1; : : : ; l there exists a smooth projective connected variety Yi with
dim Yi m, such that Ai is equivalent to an admissible subcategory of Db .Yi /.
Definition 2.4. Let X be a smooth projective variety of dimension n. We say that X
is categorically representable in dimension m (or equivalently in codimension n m)
if Db .X/ is representable in dimension m.
Remark 2.5. Suppose that X is not smooth. Then to define categorical representability
for it, we need to replace in Definition 2.4 the derived category Db .X / with another
z enjoying some “smoothness“ which would be called a cate-
triangulated category D,
gorical resolution of singularities.
In [46], Kuznetsov suggest a definition for which the resolution of singularities Dz

of D .X/ could be realized as an admissible subcategory of the derived category of
b
geometrical resolution of singularities Xz ! X , whenever X has rational singulari-

ties. The notion of categorical representability would naturally sit inside this kind of
approach (for example, all projective varieties would be categorically representable at
least in its dimension). Notice that anyway there is in general no unicity or minimality
of such a resolution.
Notice that any fully faithful functor F W Db .X / ! Db .Y / between the derived

categories of two smooth projective varieties X and Y is of Fourier–Mukai type [62],
[63], i.e. there is an object E in Db .X Y / (called kernel of F ) and F ./ is given by
pulling back a complex to Db .X Y /, tensoring with E and pushing-forward to Db .Y /.
It is moreover worth noting and recalling the following facts, which are well-known in
the derived categorical setting.
Remark 2.6 ([9]). The derived category of P n admits a full exceptional sequence.
Remark 2.7 ([59]). If is a smooth connected projective curve of positive genus, then
Db ./ has no proper admissible subcategory. Indeed any fully faithful functor A !
Db ./ is an equivalence, unless A is trivial. Then being categorically representable
in dimension 1 is equivalent to admit a semiorthogonal decomposition by exceptional
objects and derived categories of smooth projective curves.
Remark 2.8. If X and Yi are smooth projective and
Db .X / D hDb .Y1 /; : : : ; Db .Yk /i;
then
M
k
K0 .X / D K0 .Yi /
iD1
and the Riemann–Roch Theorem gives an isomorphism of Q-vector spaces
M
k
CHQ .X / D CHQ .Yi /:
iD1
Remark that the last isomorphism is in general not compatible with gradings.
Proposition 2.9 ([61]). Let X be smooth projective and Z X a smooth subvariety

of codimension d > 1. Denote by " W Xz ! X the blow up of X along Z. Then
Db .Xz / D h" Db .X /; Db .Z/1 ; : : : ; Db .Z/d 1 i;
where Db .Z/i is equivalent to Db .Z/ for all i D 1; : : : ; d 1.

2.2 Classical representabilities and motives. In general, it is a very deep and in-
teresting geometric problem to understand whether the group AiZ .X / of algebraically
trivial cycles of codimension i on X can carry a scheme structure. The notion of rep-
resentability of such groups has been introduced to tackle this problem. In this section
we outline a list of definitions of representabilities for the groups AiZ .X /. This is far
for being exhaustive, especially in the referencing. Indeed, giving a faithful list of all
contributions to these questions is out of the aim of these notes. Chow motives and
their properties could give, through Conjecture 1.1, a way to connect categorical and
classical representabilities. We also outline the basic facts needed to stress the possible
interplay between new and old definitions.
Let X as usual be a smooth projective variety over a field K. Recall that, if is a
curve, then J./ Š A1Z ./.
Definition 2.10. Let T be any nonsingular variety over K. An map f W T ! AiZ .X /
is an algebraic map if there exists a cycle class z in CHiZ .T X / such that f .t / is the
restriction of z to ft g X . In other words, f .t/ D q ..p t /:z/, where p and q denote,
respectively, the projections from T X to T and X . In this case, such a map will be
denoted by z .
Definition 2.11 ([15]). The group AiZ .X / is said to be weakly representable if there
exists a smooth projective curve , a class z of a cycle in CHiZ .X / and an algebraic
subgroup G J./ of the Jacobian variety of , such that, for any algebraically closed
field K, the induced algebraic map
z W J././ ' A1Z . / ! AiZ .X /
is surjective with kernel G./.
When working with coefficients in Q, we have the following definition.
Definition 2.12. The group AiQ .X / is rationally representable if there exists a cycle
z in CHiQ .X / giving a surjective algebraic map
z W A1Q ./ ! AiQ .X /:
The variety X is rationally representable if AiQ .X / is rationally representable for

all i .
Rational representability is a name that has been used several times in the literature,
so it might lead to some misunderstanding. We underline that Definition 2.12 is exactly
the one from ([23], page 5). In the complex case, we have also a stronger notion, which
is called the Abel–Jacobi property [15], which requires the existence of an isogeny
(i.e. a regular surjective morphism) AiZ .X / ! J i .X /, induced by a correspondence,
onto the i -th intermediate Jacobian J i .X / WD H i .X; R/=H i .X; Z/. The Abel–Jacobi
property implies weak representability for smooth projective varieties defined on C.
An even stronger notion is given by algebraic representability, which requires that
the group AiZ .X / is isomorphic via a regular map to an abelian variety.
Definition 2.13. Let A be an abelian variety. A group homomorphism g W AiZ .X / ! A

is a regular map if for every non-singular variety T and for every algebraic map
f W T ! Ai .X/Z , the composite map g B f is a morphism.
Definition 2.14 ([7], Definition 3.2.3). An abelian variety A is the algebraic represen-
tative of AiZ .X/ if there exists a quasi-universal regular map G W AiZ .X / ! A, i.e. for
all regular maps g from AiZ .X / to an abelian variety B, there is a unique morphism of
abelian varieties u W A ! B such that u B G D g. In this case we say that AiZ .X / is
algebraically representable.
The first examples of algebraic representatives are the Picard variety Pic0 .X / or
the Albanese variety Alb.X / if n D 1 or, respectively, n D dim .X /. We remark that
in these two cases the associated correspondences are those induced, respectively, by
the first Chern class of the Poincaré bundle on X Pic0 .X / and by the graph of the
natural Albanese map.
Definition 2.15 ([7], Definition 3.4.2). Let X be a smooth projective variety of odd
dimension 2nC1 and A the algebraic representative of AnC1 Z .X / via the canonical map
nC1
G W AZ .X/ ! A. A polarization of A with class A in Corr.A; A/ is the incidence
polarization with respect to X if for all algebraic maps f W T ! AnC1
Z .X / defined by
nC1
a cycle z in CHZ .X T /, we have
.G B f / A D .1/nC1 I.z/;
where I.z/ in Corr.T / is the composition of the correspondences z 2 Corr.T; X / and

z 2 Corr.X; T /.
For example, if C is a smooth projective curve, then the group A1Z .C / ' J.C /
carries an incidence polarization, namely the canonical polarization of the Jacobian.
Indeed, it is easy to check that the correspondence associated such polarization is the
opposite to the incidence polarization given by a Poincaré line bundle.
There are many complex threefolds X with negative Kodaira dimension, for which
A2Z .X/ is strongly represented by a generalized Prym variety with incidence polariza-
tion. For these threefolds, we will show how categorical representability in dimension
1 gives a splitting of the intermediate Jacobian. A list of the main cases will be given
in Section 3.2.
A more modern approach to representability questions has to take Chow motives
into account. Let us recall their basic definitions and notations. The category MK
of Chow motives over K with rational coefficients is defined as follows: an object of
MK is a triple .X; p; m/, where X is a variety, m an integer and p 2 Corr 0 .X; X / an
idempotent, called a projector. Morphisms from .X; p; m/ to .Y; q; n/ are given by
elements of Corr nm .X; Y / precomposed with p and composed with q.
There is a natural functor h from the category of smooth projective schemes to the
category of motives, defined by h.X / D .X; id; 0/, and, for any morphism W X ! Y ,
h./ being the correspondence given by the graph of . We write 1 WD .Spec K; id; 0/
for the unit motive and L WD .Spec K; id; 1/ for the Lefschetz motive, and M.i / WD
M ˝ Li . Moreover, we have Hom.Li ; h.X // D CHiQ .X / for all smooth projective
schemes X and all integers i .
If X is irreducible of dimension d , the embedding ˛ W pt ! X of the point defines
a motivic map 1 ! h.X /. We denote by h0 .X / its image and by h1 .X / the quotient
of h.X/ via h0 .X /. Similarly, Ld is a quotient of h.X /, and we denote it by h2d .X /.
Notice that both h0 .X / and h2d .X / split off the motive h.X / as direct summands.
In the case of smooth projective curves of positive genus there exists another factor
which corresponds to the Jacobian variety of the curve. Let C be a smooth projective
connected curve, let us define a motive h1 .C / such that we have a direct sum:
h.C / D h0 .C / ˚ h1 .C / ˚ h2 .C /:
The upshot is that the theory of the motives h1 .C / corresponds to that of Jacobian
varieties (up to isogeny), in fact we have
Hom.h1 .C /; h1 .C 0 // D Hom.J.C /; J.C 0 // ˝ Q:
In particular, the full subcategory of MK whose objects are direct summands of
the motive h1 .C / is equivalent to the category of abelian subvarieties of J.C / up to
isogeny. Such motives can be called abelian. We will say that a motive is discrete if it
is the direct sum of a finite number of Lefschetz motives.
Let S be a surface. Murre constructed [56] the motives hi .S /, defined by projectors
pi in CHiQ .S S/ for i D 1; 2; 3, and described a decomposition
h.S / D h0 .S / ˚ h1 .S / ˚ h2 .S / ˚ h3 .S / ˚ h4 .S /:
We have already remarked that h0 .S / D 1 and h4 .S / D L2 . Roughly speaking, the
submotive h1 .S / carries the Picard variety, the submotive h3 .S / the Albanese variety
and the submotive h2 .S / carries the Néron–Severi group, the Albanese kernel and the
transcendental cycles. If S is a smooth rational surface and K D K, x then h1 .S / and
h .S/ are trivial, while h .S / ' L , where is the rank of the Néron–Severi group. In
3 2
particular, the motive of S splits in a finite direct sum of (differently twisted) Lefschetz
motives.
In general dimension, it is conjectured [57] that if X is a smooth projective variety
of dimension d , there exist projectors pi in CHdQ .X X / defining motives hi .X /
L
such that h.X/ D 2d x
iD0 h .X /, and such that (over K) pi modulo (co)homological
i
equivalence is the usual Künneth component. For example, if K D C, we require

.pi / H .X; Q/ D H i .X; Q/. Such a decomposition is called a Chow–Künneth de-
composition (see [57], Definition 1.3.1) and should be thought of as a universal co-
homological theory. We have seen that the motive of any smooth projective curve or
surface admits a Chow–Künneth decomposition. This is true also for the motive of a
smooth uniruled complex threefold, thanks to [2]. In this case, while h1 still carries the
Picard variety, now it is straightforward to remark that it is h5 that carries the Albanese
kernel. We will call them accordingly.
The strict interplay between motives and representability for threefolds is shown
by Gorchinskiy and Guletskii. In this case, the rational representability of AiQ .X / for
i 2 is known ([55]). In [23] it is proved that A3Q .X / is rationally representable if
and only if the Chow motive of X has a given Chow–Künneth decomposition.
Theorem 2.16 ([23], Theorem 8). Let X be a smooth projective threefold. The group
A3Q .X/ is rationally representable if and only if the motive h.X / has the following
Chow–Künneth decomposition:
h.X/ Š 1 ˚ h1 .X / ˚ L˚b ˚ .h1 .J /.1// ˚ .L2 /˚b ˚ h5 .X / ˚ L3 ;
where h1 .X/ and h5 .X / are the Picard and Albanese motives respectively, b D
b 2 .X/ D b 4 .X / is the Betti number, and J is a certain abelian variety, which is
isogenous to the intermediate Jacobian J.X / if K D C.
3 Interactions between categorical and classical

representabilities
In this section, we will consider varieties defined over the complex numbers. This
restriction is not really necessary, since most of the constructions work over any alge-
braically closed field. Anyway, in the complex case, we can simplify our treatment by
dealing with intermediate Jacobians. Moreover, it will be more simple to list examples
without the need to make the choice of the base field explicit for any case.
3.1 Fully faithful functors and motives. At the end of the last section we have
seen that, in the case of threefolds, rational representability of A3Q .X / is equivalent
to the existence of some Chow–Künneth decomposition. The first step in relating
categorical and rational representability is exploiting an idea of Orlov about the motivic
decomposition which should be induced by a fully faithful functor between the derived
categories of smooth projective varieties. Assuming this conjecture we get that for
threefolds categorical representability in dimension 1 is a stronger notion than rational
representability.
Let us sketch Orlov’s idea [64]. If X and Y are smooth projective varieties of
dimension respectively n and m, and ˆ W Db .Y / ! Db .X / is a fully faithful functor,
then it is of Fourier–Mukai type [62], [63]. Let E in Db .X Y / be its kernel and F

in Db .X Y / the kernel of its right adjoint ‰, we have F ' E _ ˝ prX !X Œdim X
(see [53]). Consider e WD ch.E/Td.X / and f WD ch.F /Td.Y /, two mixed rational
cycles in CHQ .X Y /. We denote by ei (resp. fi ) the i -th codimensional component
of e (resp. f ), that is ei ; fi 2 CHiQ .X Y /. As correspondences they induce motivic
maps ei W h.Y / ! h.X /.i n/ and fj W h.X /.m j / ! h.Y /. The Grothendieck–
L
Riemann–Roch Theorem implies that f:e L WD nCmiD0 fnCmi ei D idh.Y / . This in turn
implies that h.Y / is a direct summand of i2Z h.X /.i /.
Conjecture 1.1. Let X and Y be smooth projective varieties and ˆ W Db .Y / ! Db .X /

be a fully faithful functor. Then the motive h.Y / is a direct summand of the motive
h.X/.
The conjecture is trivially true for Y a smooth point, in which case ˆ.Db .Y // is
generated by an exceptional object of Db .X /: then there is an integer d and a split
embedding Ld ! h.X / induced by the exceptional object. In [64], it is proven that
the conjecture holds if X and Y have the same dimension n and E is supported in
dimension n. This already covers some interesting examples: if X is a smooth blow-
up of Y , or if there is a standard flip from X to Y . Using the same methods as in
[13] we will show that if Y is a curve and X a rationally representable threefold with
h1 .X/ D h5 .X/ D 0, then h1 .Y / is a direct summand of h3 .X /.1/.
But let us first take a look to the simplest case, that is categorical representability
in dimension 0. In this case, we have that Chow groups are finite dimensional vector
spaces. Over the complex numbers this gives the discreteness of the motive.
Proposition 3.1. If a smooth projective complex variety X is categorically repre-
sentable in dimension 0, then the group K0 .X / is free of finite type and the motive
h.X/ is discrete.
Proof. Being representable in dimension 0 is equivalent to having a full exceptional
sequence fE1 ; : : : ; El g. Then the classes ŒEi , for i D 1; : : : l, are nontrivial (since
Ei is exceptional, we have .Ei ; Ei / D 1) and give a free system of generators of
K0 .X/, by the definition of semiorthogonal decomposition. Then K0 .X / ' Zl .
From this and Riemann–Roch, we get that CHQ .X / is a finite dimensional Q-vector
space. For a complex smooth projective variety, this is enough to split the motive into
Lefschetz motives ([40]). Notice anyway that, since the Riemann–Roch isomorphism
K0 .X/ ˝ Q ' CHQ .X / is not compatible with gradings, there is no canonical way
to obtain the decomposition of X explicitly from the base of K0 .X /, that is, from the
exceptional sequence.
A way more interesting case relates categorical representability in dimension 1 and
rational representability for threefolds. In this case, in light of Theorem 2.16, we have
a more specific conjecture.
Conjecture 3.2. If a smooth projective threefold X is categorically representable in
dimension 1, then it is rationally representable.
If X is a standard conic bundle over a rational surface and a smooth projective
curve, the Chow–Künneth decomposition of h.X / (see [58]) can be used to show that a
fully faithful functor Db ./ ! Db .X / gives h1 ./.1/ as a direct summand of h.X /.
In particular, this gives an isogeny between J./ and an abelian subvariety of J.X /,
and proves (up to codimensional shift for each direct summand of h./) Conjecture 1.1
in this case. The proof in [13] is based on the fact that the motive h.X / splits into a
discrete motive and in a unique abelian motive which corresponds to J.X /. Let us
make a first assumption:
(?) X is a smooth projective rationally representable threefold with h1 .X / D 0 and

h5 .X/ D 0.
Theorem 3.3. Suppose X satisfies .?/. If there is a smooth projective curve and a
fully faithful functor Db ./ ! Db .X /, then h1 ./.1/ is a direct summand of h.X /.
This gives an injective morphism J./Q ! J.X /Q , that is an isogeny between J./
and an abelian subvariety of J.X /.
Proof. We only need the case where g./ > 0, and we can use the same argument
as in [13], Lemma 4.2: since all but one summand of h.X / are discrete, the map
f:ejh1 ./ D idh1 ./ is given by f2 :e2 , which proves that h1 ./.1/ is a direct summand
of h3 .X/ D h1 .J /.1/.
Corollary 3.4. Suppose X satisfies .?/ and let fi gkiD1 be smooth projective curves of
positive genus. If Db .X / is categorically representable in dimension 1 by the categories
L
Db .i / and by exceptional objects, then J.X / is isogenous to kiD1 J.i /.
Proof. From Theorem
L 3.3 together with the semiorthogonality, we get an injective
morphism W J.i /Q ! J.X /Q , which has to be surjective by Remark 2.8, as
explained in the proof of Theorem 4.1 in [13].
Remark 3.5 (Threefolds satisfying .?/). By [23], [58] Fano threefolds, threefolds
fibered in Del Pezzo or Enriques surfaces over P 1 with discrete Picard group, and
standard conic bundles over rational surfaces satisfy .?/.
3.2 Reconstruction of the intermediate Jacobian. The aim of this section is to show
how, under appropriate hypotheses, categorical representability in dimension 1 for a
threefold X gives a splitting of the intermediate Jacobian J.X /. Notice that in the case
of curves the derived category carries the information about the principal polarization
of the Jacobian [11]. In the case of threefolds, we need first of all the hypotheses of
Theorem 3.3. As we will see, the crucial hypothesis that will allow us to recover also
the principal polarization is that the polarization on J.X / is an incidence polarization.
(\) X is a smooth projective rationally and algebraically representable threefold with
h1 .X/ D 0 and h5 .X / D 0 and the algebraic representative of A2Z .X / carries
an incidence polarization.
Theorem 3.6. Suppose X satisfies .\/. Let be smooth projective curve and
Db ./ ! Db .X / fully faithful. Then there is an injective morphism J./ ,! J.X /
preserving the principal polarization, that is J.X / D J./ ˚ A for some principally
polarized abelian variety A.
Proof. From Theorem 3.3 we get an isogeny. As in the proof of Proposition 4.4 in [13],
the incidence property shows that this isogeny is an injective morphism respecting the
principal polarizations.
Corollary 3.7. Suppose X satisfies .\/ and let fi gkiD1 be smooth projective curves
of positive genus. If Db .X / is categorically representable in dimension 1 by the
L
categories Db .i / and by exceptional objects, then J.X / is isomorphic to kiD1 J.i /
as principally polarized variety.
Remark 3.8 (Threefolds satisfying .\/). The assumptions of Theorem 3.6 seem rather
restrictive. Anyway, they are satisfied by a quite big class of smooth projective three-
folds with X < 0. The Chow–Künneth decomposition for the listed varieties is
provided by [58] for conic bundles and by [23] in any other case. In the following
list the references point out the most general results about strong representability and
incidence property. Giving an exhaustive list of all the results and contributors would
be out of reach (already in the cubic threefold case). We will consider Fano threefolds
with Picard number one only. The interested reader can find an exhaustive treatment
in [34].
1) Fano of index > 2: X is either P 3 or a smooth quadric.
2) Fano of index 2: X is a quartic double solid [68] , or a smooth cubic in P 4 [21],
or an intersection of two quadrics in P 5 [65], or a V5 (in the last case J.X / is
trivial).
3) Fano of index 1: X is a general sextic double solid [20], or a smooth quartic in P 4
[15], or an intersection of a cubic and a quadric in P 5 [15], or the intersection of
three quadrics in P 6 [7], or a V10 [52], [26], or a V12 [32] (J.X / is the Jacobian
of a genus 7 curve), or a V14 [31] (in which case the representability is related to
the birational map to a smooth cubic threefold), or a general V16 [27], [54], or a
general V18 [29], [34] (J.X / is the Jacobian of a genus 2 curve), or a V22 (and
the Jacobian is trivial).
4) Conic bundles: X ! S is a standard conic bundle over a rational surface [7],
[10], this is the case examined in [13].
5) Del Pezzo fibrations: X ! P 1 is a Del Pezzo fibration with 2 KX2 5 [35],
[36].
Notice that there are still some cases where it is not known (at least, to us) whether a
smooth projective threefold of negative Kodaira dimension satisfies .\/, as for example
if X is a Fano of index two and degree one, or a Del Pezzo fibration over P 1 of degree
one.
From the unicity of the splitting of the intermediate Jacobian we can easily infer
the following.
Corollary 3.9. Suppose X satisfies .\/ and is categorically representable in dimen-
sion 1, with semiorthogonal decomposition
Db .X / D hDb .1 /; : : : ; Db .k /; E1 ; : : : ; El i:

Then there is no fully faithful functor Db ./ ! Db .X / unless ' i for some
i 2 f1; : : : ; kg. Moreover, the semiorthogonal decomposition is essentially unique,
that is, any semiorthogonal decomposition of Db .X / by smooth projective curves and
exceptional objects is given precisely by the curves i and by l exceptional objects.
Corollary 3.10. Suppose X satisfies .\/, is a smooth projective curve of positive

genus and there is no splitting J.X / D J./˚A. Then there is no fully faithful functor
Db ./ ! Db .X /.
The assumptions of Corollary 3.10 are trivially satisfied if the threefold satisfying
.\/ has J.X/ D 0. A way more interesting case is when the intermediate Jacobian is
not trivial and there is no injective morphism J./ ! J.X / for any curve , in which
case the variety is not categorically representable in dimension < 2.
Remark 3.11 (Threefolds not categorically representable in dimension < 2). The
assumptions of Corollary 3.10 are satisfied by smooth threefolds with J.X / ¤ 0 for
all curves of positive genus in the following cases:
1) Fano varieties of index 2: X is a smooth cubic [21].
2) Fano varieties of index 1: for instance when X is a generic quartic threefold [51],
the intersection of three quadrics in P 7 [7], or a generic complete intersection
of type .3; 2/ in P 5 [7]. The case of a .3; 2/-complete intersection of Fermat
polynomials is described in [8].
3) Conic bundles: X is a standard conic bundle X ! P 2 degenerating along a

curve of degree 6 [7], or a non-rational standard conic bundle X ! S on a
Hirzebruch surface [67].
4) Del Pezzo fibrations: X ! P 1 non-rational of degree four [1].
There are some other cases of Fano threefolds of specific type satisfying geometric
assumptions. For a detailed treatment, see Chapter 8 of [34].
Notice that if X is a smooth cubic threefold, the equivalence class of a notable

admissible subcategory AX (the orthogonal complement of fOX ; OX .1/g) corresponds
to the isomorphism class of J.X / as principally polarized abelian variety [14]; the
proof is based on the reconstruction of the Fano variety and the techniques used there
are far away from the subject of this paper.
A natural question is if, under some hypotheses, one can give the inverse statement
L that is, suppose that X is a threefold satisfying either .?/ or
of Corollaries 3.4 and 3.7,
.\/, such that J.X / ' J.i /. Can one describe a semiorthogonal decomposition of
Db .X/ by exceptional objects and the categories Db .i /? Notice that a positive answer
for X implies a positive answer for all the smooth blow-ups of X .
Remark 3.12 (Threefolds with X < 0 categorically representable

L in dimension 1).
Let X be a threefold satisfying .?/ or .\/ and with J.X / D J.i /. Then if X is in
the following list (or is obtained by a finite number of smooth blow-ups from a variety
in the list) we have a semiorthogonal decomposition
Db .X / D hDb .1 /; : : : ; Db .k /; E1 ; : : : ; El i;
with Ei exceptional objects.

1) Threefolds with a full exceptional sequence: X is P 3 [9], or a smooth quadric
[37], or a P 1 -bundle over a rational surface or a P 2 -bundle over P 1 [61], or a V5
[60], or a V22 Fano threefold [41].
2) Fano threefolds without any full exceptional sequence: X is the complete inter-
section of two quadrics or a Fano threefold of type V18 , and J./ ' J.X / with
a genus 2 curve. The semiorthogonal decompositions are described in [18],
[44], and are strikingly related (as in the cases of V5 and V22 and of the cubic
and V14 ) by a correspondence in the moduli spaces, as described in [47]. X is a
V12 Fano threefold [42], or a V16 Fano threefold [44].
3) Conic bundles without any full exceptional sequence: X ! S is a rational conic
bundle over a minimal surface [13]. If the degeneration locus of X is either
empty or a cubic in P 2 , then X is a P 1 -bundle and is listed in 1).
4) Del Pezzo fibrations: X ! P 1 is a quadric fibration with at most simple de-
generations, in which case the hyperelliptic curve ! P 1 ramified along the
degeneration appears naturally as the orthogonal complement of an exceptional
sequence of Db .X / [48]. X ! P 1 is a rational Del Pezzo fibration of degree
four. In this case X is birational to a conic bundle over a Hirzebruch surface [1]
and the semiorthogonal decomposition is described in [4].
Notice that the first two items cover all classes of Fano threefolds with Picard number
one whose members are all rational.
4 Categorical representability and rationality: further

developments and open questions
This last section is dedicated to speculations and open questions about categorical
representability and rationality. The baby example of curves is full understood. A
smooth projective curve X over a field K is categorically representable in dimension 0
if and only if it is rational. Indeed, the only case where Db .X / has exceptional objects
is X D P 1 , and Db .X / D hOX ; OX .1/i.
Let us start with a trivial remark: the projective space P n over K is categorically
representable in dimension 0. Then if X is given by a finite number of smooth blow-ups
of P n , it is categorically representable in codimension 2. This is easily obtained from
Orlov’s blow-up formula (see Proposition 2.9). More generally, if a smooth projective
variety X of dimension 2 is categorically representable in codimension m, then any
finite chain of smooth blow-ups of X is categorically representable in codimension
min.2; m/.
One could naively wonder about the inverse statement: if X ! Y is a finite chain
of smooth blow-ups and X is categorically representable in codimension m, what can
we say about Y ? Unfortunately, triangulated categories do not have enough structure
to let us compare different semiorthogonal decomposition. For example, the theory of
mutations allows to do this only in a few very special cases.
In this section we present some more example to stress how the interaction between
categorical representability and rationality can be developed further, and we point out
some open question. We deal with surfaces in 4.1 and with threefolds in 4.2. Then
we will discuss in 4.3 how categorical representability for noncommutative varieties
plays an important role in this frame, to deal with varieties of dimension bigger than 3
in 4.4. Finally, we compare in 4.5 our methods with recent approaches to birationality
problems via derived categories. We will work over the field C for simplicity, even if
many problems and arguments do not depend on that.
4.1 Surfaces. If X is a smooth projective rational surface, then it is categorically

representable in codimension 2. Indeed, X is the blow-up in a finite number of smooth
points of a minimal rational surface, that is either P 2 or Fn . Are there any other
example of surfaces categorically representable in codimension 2? Notice that by
Proposition 3.1 such a surface would have a discrete motive, and even more: we would
have K0 .X/ D Zl . In particular, if K0 .X / is not free, then X is not categorically
representable in dimension 0.
In general, an interesting problem is to construct exceptional sequences on surfaces
with pg D q D 0, and to study their orthogonal complement. Notice that on such
surfaces any line bundle is an exceptional object, so we already have at least a length
one exceptional sequence. The main question is then to understand if it is possible to
find a somehow “maximal” one, that is, such that the orthogonal complement does not
contain exceptional objects. The length of such sequence should be bounded by the
rank of maximal free subgroup of K0 .X /.
Suppose for example that X is an Enriques surface: a (non-full) exceptional col-
lection of 10 vector bundles on X is described in [69]. Since K0 .X / is not free of finite
rank, we do not expect any full exceptional collection. The orthogonal complement
AX turns then out to be a very interesting object, related also to the geometry of some
singular quartic double solid [33]. Using a motivic trick, we can prove that, under
some assumption, a surface with pg D q D 0 is either categorically representable in
codimension 2 or not categorically representable in positive codimension.
Proposition 4.1. Let X be a surface with h.X / discrete. Then for any curve of
positive genus, there is no fully faithful functor Db ./ ! Db .X /.
Proof. Suppose there is such a curve and such a functor ˆ W Db ./ ! Db .X /. Let
E denote the kernel of ˆ (which has to be of Fourier–Mukai type) and F the kernel
of its adjoint.
L Consider the cycles e and f described in Section 3.1, and recall that
f:e D 3iD0 f3i :ei D idh./ . Restricting now to h1 ./ we would have that idh1 ./
would factor through a discrete motive, which is impossible.
Corollary 4.2. Let X be a surface with h.X / discrete and K0 .X / not free of finite
rank. Then X is not categorically representable in codimension > 0.
Remark 4.3 (Surfaces with pg D q D 0 not categorically representable in positive
codimension). Proposition 3.1 could be an interesting tool in the study of derived cate-
gories of surfaces with pg D q D 0: notice that many of them have torsion in H1 .X; Z/
(for an exhaustive treatment and referencing, see [6]). Anyway the discreteness of the
motive is a rather strong assumption, which for example implies the Bloch conjecture.
There are few cases where this is known.
1) X is an Enriques surface [22].
2) X is a Godeaux surface obtained as a quotient of a quintic by an action of
Z=5Z [24]. In this case in particular it is shown that the motive decomposes as
1 ˚ 9L ˚ L2 .
These observations lead to state some deep question about categorical representabil-
ity of surfaces.
Question 4.4. Let X be a smooth projective surface with pg D q D 0.
1) Is there a full exceptional sequence? Equivalently, is X categorically repre-
sentable in codimension 2? If not, can one describe a non full maximal (i.e. the
complement does not contain any exceptional object) exceptional sequence and
its complement?
2) If X is representable in codimension 2, is X rational?
4.2 Threefolds. Remark that there are examples of smooth projective non-rational
threefolds X which are categorically representable in codimension 2: just consider a
rank three vector bundle E on a curve C of positive genus and take X WD P .E/. In
Section 6.3 of [13] we provide a conic bundle example. Anyway, Corollary 3.7 some-
how suggests that categorical representability in codimension 2 should be a necessary
condition for rationality.
A reasonable idea is to restrict our attention to minimal threefolds with X < 0
(recall that this is a necessary condition for rationality), in particular to the ones we
expect to satisfy assumption .\/, in order to have interesting information about the
intermediate Jacobian from semiorthogonal decompositions. The three big families of
such threefolds are: Fano threefolds, conic bundles over rational surfaces and del Pezzo
fibrations over P 1 . Remarks 3.8 and 3.12 give a long list of examples either of rational
threefolds which are categorically representable in codimension 2 or non-rational ones

which cannot be categorically representable in codimension > 1.
Question 4.5. Let X be a smooth projective threefold with X < 0.
1) If X is rational, is X categorically representable in codimension 2?

2) Is X categorically representable in codimension 2 if and only if X is rational?
A positive answer to the second question is provided for standard conic bundles over
minimal surfaces [13], but itLseems to be quite a strong fact to hold in general: recall that
having a splitting J.X / ' J.i / is only a necessary condition for rationality, and
Corollary 3.7 shows that if X satisfies .\/, categorical representability in codimension
2 would give the splitting of the Jacobian.
Remark 3.12 provides a large list of rational threefolds categorically representable
in codimension 2. Is it possible to add examples to this list? In particular in the case of
Del Pezzo fibrations over P 1 only the quadric and the degree 4 fibration are described
respectively in [48] and [4].
A good way to understand these questions is by studying some special rational or
non-rational (that is non generic in their family) threefold. This forces to consider
non smooth ones, but we can use Kuznetsov’s theory of categorical resolution of sin-
gularities [46] and study the categorical resolution of Db .X /, as we pointed out in
Remark 2.5. For example, let X P 4 be nodal cubic threefold with a double point,
which is known to be rational.
Proposition 4.6. Let X P 4 be a cubic threefold with a double point and Xz ! X
the blow-up of the singular point. There exists a categorical resolution of singularities
z Db .X/
D z of Db .X / (in the sense of [46]) which is representable in codimension
two. Indeed there is a semiorthogonal decomposition
z D hDb ./; E1 ; : : : ; E3 i;
D
where Ei are exceptional objects and a complete intersection of a quadric and a

cubic in P 3 .
Proof. This is shown following step by step [49], Section 5, where the four dimensional
case is studied. Let us give a sketch of the proof. Let P be the singular point of X , and
W Xz ! X its blow-up. The exceptional locus ˛ W Q ,! Xz of

is a quadric surface.
The projection of P 4 to P 3 from the point P restricted to X gives the birational map
X Ü P 3 . The induced map W Xz ! P 3 is the blow-up of a smooth curve of
genus 4, given by the complete intersection of a cubic and a quadric surface. If we
write h WD OP 3 .1/ and H WD
OX .1/, we have that Q D 2h D, H D 3h D,
then h D H Q and D D 2H 3Q as in [49], Lemma 5.1. The canonical bundle
!Xz D 4h C D D 2H C Q can be calculated via the blow-up .
In order to describe a categorical resolution of singularities, we have to provide a
Lefschetz decomposition of Q with respect to the conormal bundle (for definitions and
details, see [46]). The conormal bundle of Q is OQ .h/ and the Lefschetz decomposition
with respect to it is
hA1 .h/; A0 i;
where A1 D hOQ i and A0 D hOQ ; S1 ; S2 i, with S1 and S2 the two spinor bun-
dles. Indeed, Q is even-dimensional and then has two non-isomorphic spinor bundles
giving the previous semiorthogonal decomposition [37]. The case where X is four-
dimensional, considered in [49] is slightly different.
z of Db .X / in the
We then get, by [46] a categorical resolution of singularities D
semiorthogonal decomposition:
z
Db .Xz / D h˛ OQ .h/; Di:
We then get
zX ; O z ; O z .H /i;
Db .Xz / D h˛ OQ .h/; A (4.1)
X X
where AzX is a categorical resolution of AX , as in [49], Lemma 5.8. The representability

z
of D relies then on the representability of A zX .
On the other side, applying the blow-up formula to W Xz ! P 3 (see Proposi-
tion 2.9), and choosing fOP 3 .3/; : : : ; OP 3 g as full exceptional sequence for Db .P 3 /,
we obtain
Db .Xz / D hˆDb ./; OXz .3h/; OXz .2h/; OXz .h/; OXz i;
where ˆ W Db ./ ! Db .Xz / is fully faithful. Now as in Lemma 5.3 of [49], if we

mutate OXz .3h/ and OXz .2h/ to the left with respect to ˆDb ./, and put B WD
hˆDb ./; OXz .h/i, we get
Db .Xz / D hOXz .3h C D/; OXz .2h C D/; B; OXz i: (4.2)
zX are equivalent, following exactly the same path of

Finally, one can show that B and A
mutations as in Section 5 of [49] to compare the decompositions (4.1) and (4.2).
Remark 4.7. As noted in [19], we could aim to some kind of minimal resolution of
singularities, where minimality has to be taken with respect to full and faithful functors.
Then one is lead to suspect that the category A zX is not minimal. A natural question to
ask for is if it possible to give a categorical resolution of singularities of AX equivalent
to Db ./.
Another special very interesting example is described in [33]: a singular double
solid X ! P 3 ramified along a quartic symmetroid. This threefold is non-rational
thanks to [3], because H 3 .X; Z/ has torsion. A rough account (skipping the details
about the resolution of singularities) of Ingalls and Kuznetsov’s result is the following:
if X 0 is the small resolution of X , there is an Enriques surface S and a semiorthogonal
decomposition
Db .X 0 / D hAS ; E1 ; E2 i; (4.3)
where Ei are exceptional objects and AS is the orthogonal complement in Db .S / of

10 exceptional vector bundles on S ([69]). By Corollary 4.2, the category AS is not
representable in dimension < 2. This is anyway not enough to show that X 0 cannot
be categorically representable in codimension > 1, because it does not exclude the
existence of other semiorthogonal decompositions.
Remark anyway that the lack of categorical representability of X 0 (and presumably
of X, thinking about the categorical resolution of singularities) should be based on
the lack of categorical representability of S, which relies on the presence of torsion
in K0 .S/ and in particular in H1 .S; Z/, and on the structure of the motive. On the
other side, the non-rationality of X is due to the presence of torsion in H 3 .X; Z/. The
relation between torsion in H 3 .X; Z/ and categorical representability needs a further
investigation, for example in the case recently described in [28].
4.3 Noncommutative varieties. The previous speculations and partial results give
rise to the hope of extending fruitfully the study of categorical representability to
higher dimensions and to the noncommutative setting. By the latter we mean, following
Kuznetsov ([48], Section 2), an algebraic variety Y with a sheaf B of OY -algebras of
finite type. Very roughly, the corresponding noncommutative variety Yx would have a
category of coherent sheaves Coh.Yx / D Coh.Y; B/ and a bounded derived category
Db .Yx / D Db .Y; B/. The examples which appear very naturally in our setting are the
cases where B is an Azumaya algebra or the even part of the Clifford algebra associated
to some quadratic form over Y .
Finally, if a triangulated category A has Serre functor such that SAm D Œn, for some
integers n and m, with m minimal with this property, we will call it a m n
-Calabi–Yau
category. If m D 1, these categories deserve the name of noncommutative Calabi–Yau
n-folds, even if they are not a priori given by the derived category of some Calabi–Yau
n-fold with a sheaf of algebras.
If S is any smooth projective variety, X ! S a Brauer–Severi variety of relative
dimension r, and A the associated Azumaya algebra in Br.S /, then (see [12])
Db .X / D hDb .S /; Db .S; A1 /; : : : ; Db .S; ArC1 /i:
The categorical representability of X would then rely on the categorical representability

of .S; A/, which is an interesting object in itself. For example, if Y is a generic cubic
fourfold containing a plane, there are a K3 surface S and an Azumaya algebra A such
that the categorical representability of .S; A/ is the subject of Kuznetsov’s conjecture
[49] about the rationality of cubic fourfolds.
If S is a smooth projective variety and Q ! S a quadric fibration of relative
dimension r, we can consider the sheaf B0 of the even parts of the Clifford algebra
associated to the quadratic form defining Q. There is a semiorthogonal decomposition:
Db .Q/ D hDb .S; B0 /; Db .S /1 ; : : : ; Db .S /r i;

where Db .S/i are equivalent to Db .S / [48]. The categorical representability of .S; B0 /

should then be a very important tool in studying birational properties of Q. This is
indeed the case for conic bundles over rational surfaces [13].
n
Finally, let A be an m -Calabi–Yau category. Such categories appear as orthogonal
complements of an exceptional sequence on Fano hypersurfaces in projective spaces,
see Corollary 4.3 in [43]. It is then natural to wonder about their representability.
For example, if X is a cubic or a quartic threefold, it follows from Remark 3.11 that
these orthogonal complements (which are, respectively, 53 - and 104
-Calabi–Yau) are not
representable in dimension 1.
n
Question 4.8. Let A be a m
-Calabi–Yau category.
1) Is A representable in some dimension?
2) If yes, is there an explicit lower bound for this dimension?
3) If m D 1, is A representable in dimension n if and only if there exist a smooth
n-dimensional variety X and an equivalence Db .X / ' A?
4.4 Higher dimensional varieties. Unfortunately, it looks like the techniques used
for threefolds in [13] hardly extend to dimensions bigger than 3. The examples and
supporting evidences provided so far lead anyway to suppose that categorical repre-
sentability can give useful information on the birational properties of any projective
variety. The main case is a challenging conjecture by Kuznetsov [49]. Let X P 5 be
a smooth cubic fourfold, then there is a semiorthogonal decomposition
Db .X / D hAX ; OX ; OX .1/; OX .2/i:
The category AX is 2-Calabi–Yau.
Conjecture 4.9 (Kuznetsov). The cubic fourfold X is rational if and only if AX '
Db .Y / for a smooth projective K3 surface Y .
This conjecture has been verified in [49] for singular cubics, Pfaffian cubics and
Hassett’s [25] examples. When X contains a plane P there is a way more explicit
construction: blowing up P we obtain a quadric bundle Xz ! P 2 of relative dimension
2, degenerating along a sextic. If the sextic is smooth, let S ! P 2 be the double cover,
which is a K3 surface. Then
AX ' Db .P 2 ; B0 / ' Db .S; A/;
where B0 is associated to the quadric fibration and A is an Azumaya algebra, obtained
lifting B0 to S. The questions about categorical representability of noncommutative
varieties arise then very naturally in this context. Notice that if AX is representable in
dimension 2, then we know something weaker than Kuznetsov conjecture: we would
have a smooth projective surface S 0 and a fully faithful functor AX ! Db .S 0 /. Point 3)
of Question 4.8 appears naturally in this context.
Question 4.10. One can then wonder if the Kuzentsov conjecture may be stated in
the following form: the cubic fourfold X is rational if and only if it is categorically
representable in codimension 2. An important check in this perspective is to show that
the 2-Calabi–Yau category AX is representable in dimension 2 if and only if there exist
a K3 surface Y and an equivalence Db .Y / ' AX .
Notice anyway that there could be a priori other semiorthogonal decompositions
not related to the one considered in the conjecture. A very deep question is then to un-
derstand if and under which conditions one has a canonical choice for a semiorthogonal
decomposition.
We can propose some more examples of fourfolds for which a Kuznetsov-type

conjecture seems natural: if X is the complete intersection of three quadrics Q1 ,
Q2 , Q3 in P 7 , then Homological Projective Duality ([45], [48]) gives an exceptional
sequence on X and its complement AX ' Db .P 2 ; B0 /, where B0 is the even Clifford
algebra associated to the family of quadrics generated by Q1 ; Q2 ; Q3 . Similarly, if
we consider two quadric fibrations Q1 ; Q2 ! P 1 of relative dimension 4 and their
complete intersection X , there is an exceptional sequence on X , and let AX be its
orthogonal complement. A relative version of Homological Projective Duality shows
that AX equivalent to Db .S; B0 /, where S is a P 1 -bundle over P 1 and B0 the even
Clifford algebra associated to the pencil of quadrics generated by Q1 and Q2 . It is
natural to wonder if representability in dimension 2 of the noncommutative varieties is
equivalent or is a necessary condition for rationality of X . A partial answer to the last
example is provided in [4].
Other examples in dimension 7 are provided in [30]. If X is a cubic sevenfold, there
is a distinguished subcategory AX of Db .X /, namely the orthogonal complement of the
exceptional sequence fOX ; : : : ; OX .5/g. This is a 3-Calabi–Yau category. Moreover
it can be shown [30] that AX cannot be equivalent to the derived category of a 3-
dimensional Calabi–Yau variety. It is also conjectured that AX is equivalent to the
orthogonal complement of an exceptional sequence in the derived category Db .Y / of
a Fano sevenfold Y of index 3, birationally equivalent to X .
4.5 Other approaches. Of course categorical representability is just one among dif-
ferent approaches to the study of birational geometry of a variety via derived categories.
Nevertheless there is some common ground.
First of all, Kuznetsov mentions in [49] the notion of Clemens–Griffiths component
of Db .X/, whose vanishing would be a necessary condition for rationality. It seems
reasonable to expect that categorical representability in codimension 2 implies the
vanishing of the Clemens–Griffiths component.
Another recent theory is based on Orlov spectra and their gaps [5]. Let us even
refrain from sketching a definition of it, but just notice that ([5], Conjecture 2) draws
a link between categorical representability and gaps in the Orlov spectrum (see, in
particular, [5], Corollary 1.11). Finally, conjectures based on homological mirror
symmetry are proposed in [38], [39], but we cannot state a precise relation with our
construction. A careful study of the example constructed in [28] would be a good

starting point.
References
[1] V. A. Alekseev, Rationality conditions for three-dimensional varieties with sheaf of del
pezzo surfaces of degree 4. Math. Notes 41 (1987), no. 5, 408–411.
[2] P. L. del Angel and S. Müller-Stach, Motives of uniruled threefolds. Compositio Math. 112
(1998), 1–16.
[3] M. Artin and D. Mumford, Some elementary examples of unirational varieties which are
not rational. Proc. London Math. Soc. 3 (1972), no. 25, 75–95.
[4] A. Auel, M. Bernardara, and M. Bolognesi, Fibrations in complete intersections of quadrics,
Clifford algebras, derived categories, and rationality problems. Preprint, arXiv:1109.6938
[math.AG].
[5] M. Ballard, D. Favero, and L. Katzarkov, Orlov spectra: bounds and gaps. Invent. Math.
189 (2012), no. 2, 359–430.
[6] I. Bauer, F. Catanese, and R. Pignatelli, Surfaces of general type with geometric genus zero:
a survey. Preprint, arXiv:1004.2583 [math.AG].
[7] A. Beauville, Variétés de Prym et jacobiennes intermédiaires. Ann. Sci. École Norm. Sup.
(4) 10 (1977), 309–391.
[8] A. Beauville, Non-rationality of the symmetric sextic Fano treefold. In Geometry and
arithmetic, EMS Ser. Congr. Rep., EMS Publ. House, Zurich 2012, 57–60.
[9] A. A. Beilinson, The derived category of coherent sheaves on P n . Selecta Math. Soviet. 34
(1984), no. 3, 233–237.
[10] M. Beltrametti, On the Chow group and the intermediate jacobian of a conic bundle. Ann.
Mat. Pura App. (4) 141 (1985), no. 4, 331–351.
[11] M. Bernardara, Fourier–Mukai transforms of curves and principal polarizations. C. R. Acad.
Sci. Paris Ser. I 345 (2007), 203–208.
[12] M. Bernardara, A semiorthogonal decomposition for Brauer Severi schemes. Math. Nachr.
282 (2009), no. 10, 1406–13.
[13] M. Bernardara and M. Bolognesi, Derived categories and rationality of conic bundles.
Preprint, arXiv:1010.2417 [math.AG].
[14] M. Bernardara, E. Macrì, S. Mehrotra, and P. Stellari, A categorical invariant for cubic
threefolds. Adv. Math. 229 (2012), no. 2, 770–803.
[15] S. Bloch and J. P. Murre, On the Chow group of certain types of Fano threefolds. Compositio
Math. 39 (1979), no. 1, 47–105.
[16] A. I. Bondal, Representations of associative algebras and coherent sheaves. Math. USSR
Izv. 34 (1990), no. 1, 23–42.
[17] A. I. Bondal and M. M. Kapranov, Representable functors, Serre functors and mutations.
Math. USSR Izv. 35 (1990), no. 3, 519–541.
[18] A. I. Bondal and D. O. Orlov, Semiorthogonal decomposition for algebraic varieties.

Preprint, arXiv:alg-geom/9506012.
[19] A. I. Bondal and D. O. Orlov, Derived categories of coherent sheaves. In Proc. International
congress of mathematicians, Beijing 2002, Vol. II, Higher Education Press, Beijing 2002,
47–56.
[20] G. Ceresa and A. Verra, The Abel-Jacobi isomorphism for the sextic double solid. Pacific
J. Math. 124 (1986), no. 1, 85–105.
[21] C. H. Clemens and P. Griffiths, The intermediate Jacobian of the cubic threefold. Ann. of
Math. 95 (1972), 281–356.
[22] K. Coombes, The K-cohomology of Enriques surfaces. Contemp. Math. 126 (1992), 47–57.
[23] S. Gorchinskiy and V. Guletskii, Motives and representability of algebraic cycles on three-
folds over a field. J. Algebraic Geom. 21 (2012), 347–373.
[24] V. Guletskii and C. Pedrini, The Chow motive of the Godeaux surface. In Algebraic geom-
etry, de Gruyter, Berlin 2002, 179–195.
[25] B. Hassett, Special cubic fourfolds. Compositio Math. 120 (2000), 1–23.
[26] A. Iliev, The Fano surface of the Gushel’ threefold. Compositio Math. 94 (1994), no. 1,
81–107.
[27] A. Iliev, The Sp3 -Grassmannian and duality for prime Fano threefolds of genus 9.
Manuscripta Math. 112 (2003), no. 1, 29–53.
[28] A. Iliev, L. Katzarkov, and V. Przyjalkowski, Double solids, categories and non-rationality.
[29] A. Iliev and L. Manivel, Prime Fano threefolds and integrable systems. Math. Ann. 339
(2007), no. 4, 937–955.
[30] A. Iliev and L. Manivel, On cubic hypersurfaces of dimension seven and eight. Proc. London
Math. Soc., to appear; preprint, arXiv:1102.3618 [math.AG].
[31] A. Iiev and D. Markushevich, The Abel-Jacobi map for cubic threefold and periods of fano
threefolds of degree 14. Doc. Math. 5 (2000), 23–47.
[32] A. Iiev and D. Markushevich, Elliptic curves and rank-2 vector bundles on the prime Fano
threefold of genus 7. Adv. Geom. 4 (2004), no. 3, 287–318.
[33] C. Ingalls and A. Kuznetsov, On nodal Enriques surfaces and quartic double solids. Preprint,
arXiv:1012.3530 [math.AG].
[34] V. A. Iskovskikh and Yu. G. Prokhorov, Fano varieties. Algebraic geometry V, Encyclopae-
dia Math. Sci. 47, Springer-Verlag, Berlin 1999.
[35] V. I. Kanev, Intermediate Jacobians of threefolds with a pencil of Del Pezzo surfaces and
generalized Prym varieties. C. R. Acad. Bulgare Sci. 36 (1983), no. 8, 1015–1017.
[36] V. I. Kanev, Intermediate Jacobians and Chow groups of three-folds witha pencil of del
Pezzo surfaces. Ann. Mat. Pura Appl. (4) 154 (1989), 13–48.
[37] M. M. Kapranov, On the derived categories of coherent sheaves on some homogeneous
spaces. Invent. Math. 92 (1988), 479–508.
[38] L. Katzarkov, Generalized homological mirror symmetry, superschemes and nonrationality.

In Special metrics and supersymmetry, AIP Conf. Proc., 1093, Amer. Inst. Phys., Melville,
NY, 2009, 92–131.
[39] L. Katzarkov, Generalized homological mirror symmetry and rationality questions. In
Cohomological and geometric approaches to rationality problems, Progr. Math. 282,
Birkhäuser, Boston 2010, 219–243.
[40] S.-I.Kimura, Surjectivity of the cycle map for Chow motives. Fields Inst. Commun. 56
(2009), 157–165
[41] A. Kuznetsov, An exception set of vector bundles on the varieties V22 . Vestnik Moskov.
Univ. Ser. I Mat. Mekh. 1996 (1996), no. 3, 41–44, 92; English transl. Moscow Univ. Math.
Bull. 51 (1996), no. 3, 35–37.
[42] A. Kuznetsov, Derived category of V12 Fano threefold. Math. Notes 78 (2005), no. 3–4,
537–550.
[43] A. Kuznetsov, Derived category of cubic and V14 threefold. Proc. V. A. Steklov Inst. Math.
246 (2004), 183–207.
[44] A. Kuznetsov, Hyperplane sections and derived categories. Izv. Ross. Akad. Nauk Ser. Mat.
70 (2006), no. 3, 23–128; English transl. Izv. Math. 70 (2006), no. 3, 447–547.
[45] A. Kuznetsov, Homological projective duality. Publ. Math. Inst. Hautes Études Sci. 105
(2007), 157–220.
[46] A. Kuznetsov, Lefschetz decompositions and categorical resolutions of singularities. Se-
lecta Math. (N.S.) 13 (2007), no. 4, 661–696.
[47] A. Kuznetsov, Derived categories of Fano threefolds. Proc. Steklov Inst. Math. 264 (2009),
no. 1, 110–122.
[48] A. Kuznetsov, Derived categories of quadric fibrations and intersections of quadrics. Adv.
Math. 218 (2008), no. 5, 1340–1369.
[49] A. Kuznetsov, Derived categories of cubic fourfolds. In Cohomological and geometric
approaches to rationality problems, Progr. Math. 282, Birkhäuser, Boston 2010, 219–243.
[50] A. Kuznetsov, L. Manivel, and D. Markushevich, Abel–Jacobi maps for hypersurfaces and
noncommutative Calabi–Yau’s. Commun. Contemp. Math. 12 (2010), no. 3, 373–416.
[51] M. Letizia, The Abel-Jacobi mapping for the quartic threefold. Invent. Math. 75 (1984),
no. 3, 477–492.
[52] D. Yu. Logachëv, Isogeny of the Abel-Jacobi mapping for a Fano threefold of genus six. In
Constructive algebraic geometry, Yaroslav. Gos. Ped. Inst., Yaroslavl’ 1982, 67–76, 143.
[53] S. Mukai, Duality between D.X / and D.Xy / and its application to Picard sheaves. Nagoya
Math. J. 81 (1981), 153–175.
[54] S. Mukai, Noncommutativizability of Brill-Noether theory and 3-dimensional Fano vari-
eties. SNugaku 49 (1997), no. 1, 1–24.
[55] J. P. Murre, Un résultat en théorie des cycles algébriques de codimension deux. C. R. Acad.
Sci. Paris Sér. I 296 (1983), 981–984.
[56] J. P. Murre, On the motive of an algebraic surface. J. Reine Angew. Math. 409 (1990),
190–204.
[57] J. P. Murre, On a conjectural filtration on the Chow groups of an algebraic variety. I: The
general conjectures and some examples. Indag. Math., New Ser. 4 (1993), no. 2, 177–188.
[58] J. Nagel and M. Saito, Relative Chow-Künneth decomposition for conic bundles and Prym
varieties. Internat. Math. Res. Notices 2009 (2009), 2978–3001.
[59] S. Okawa, Semiorthogonal decomposability of the derived category of a curve. Adv. Math.
228 (2011), no. 5, 2869–2873.
[60] D. O. Orlov, Exceptional set of vector bundles on the variety V5 , Vestnik Moskov. Univ.
Ser. I Mat. Mekh. 5 (1991), 69–71.
[61] D. O. Orlov, Projective bundles, monoidal transformations and derived categories of co-
herent sheaves. Russian Math. Izv. 41 (1993), 133–141.
[62] D. O. Orlov, On equivalences of derived categories and K3 surfaces. J. Math. Sci. 84 (1997),
1361–1381.
[63] D. O. Orlov, Derived categories of coherent sheaves and equivalences between them. Rus-
sian Math. Surveys 58 (2003), 511–591.
[64] D. O. Orlov, Derived categories of coherent sheaves and motives. Russian Math. Surveys
60 (2005), 1242–1244.
[65] M. Reid, The complete intersection of two or more quadrics. PhD Thesis, Cambridge, June
1972.
[66] R. Rouquier, Catégories dérivées et géométrie birationnelle. In Séminaire Bourbaki
2004/2005, Exposés 938–951, Astérisque 307, Société Mathématique de France, Paris
2006, 283–307.
[67] V.V. Shokurov, Prym varieties: theory and applications. Math. USSR-Izv. 23 (1984), 83–147.
[68] A. S. Tihomirov, The intermediate Jacobian of double P3 that is branched in a quartic. Izv.
Akad. Nauk SSSR Ser. Mat. 44 (1980), no. 6, 1329–1377, 1439 (in Russian).
[69] S. Zube, Exceptional vector bundles on Enriques surfaces. Math. Notes 61 (1997), no. 6,
693–699.
Fourier–Mukai functors: a survey
Alberto Canonaco and Paolo Stellari
1 Introduction
Fourier–Mukai functors are ubiquitous in geometric contexts and the general belief is
that they actually are the geometric functors. Essentially, all known exact functors are
of Fourier–Mukai type in the setting of proper schemes. This paper may be seen as an
attempt to survey some recent works addressing this expectation according to several
points of view.
Let us first recall the definition of this kind of functors. Assume that X1 and X2
are smooth projective varieties over a field k and denote by Db .Xi / WD Db .Coh.Xi //
the bounded derived category of coherent sheaves on Xi . Given E 2 Db .X1 X2 / we
define the exact functor Ê W Db .X1 / ! Db .X2 / as
L
Ê ./ WD R.p2 / .E ˝ p1 .//; (1.1)
where pi W X1 X2 ! Xi is the natural projection. An exact functor F W Db .X1 / !

Db .X2 / is a Fourier–Mukai functor (or of Fourier–Mukai type) if there exists E 2
Db .X1 X2 / and an isomorphism of exact functors F Š Ê . The complex E is called
a kernel of F. This definition will be extended to more general settings in the course
of the paper allowing Xi to be singular or considering supported derived categories.
One of the first examples of these functors appeared in Mukai’s seminal paper [40]
dating 1981. Mukai studied what he originally called a duality between the bounded
derived category Db .A/ of an abelian variety (or a complex torus) A and the one of its
y Such a duality is nothing but an equivalence
dual variety A.
y
F W Db .A/ ! Db .A/
realized as a Fourier–Mukai functor whose kernel is precisely the universal Picard sheaf
P 2 Coh.A A/. y In other words, the inverse of F sends a skyscraper sheaf Op (here
y on Ay to the degree 0 line bundle Lp 2 Pic0 .A/ parametrized
p is a closed point of A)
by p.
This discussion motivates the appearance of the word ‘Mukai’ in the name of these
functors. On the other hand, Mukai himself clarified why they should be thought of
as a sort of Fourier transforms. Indeed, the push forward along the projection is the
analogue of the integration while the Fourier–Mukai kernel is the same as the kernel
in a Fourier transform.

P.S. was partially supported by the MIUR of the Italian Government in the framework of the National
Research Project “Geometria algebrica e aritmetica, teorie coomologiche e teoria dei motivi” (PRIN 2008).
28 A. Canonaco and P. Stellari
A more precise historical reconstruction of the origins of the notion of Fourier–

Mukai functor should certainly point to the paper [50] where the notion of Fourier–Sato
transform was introduced (see also Section 3.7 in [28]). This is probably one of the
first attempts to ‘categorify’ the Fourier transform.
There are several possible directions along which to study these functors. In this
paper, we are interested in the very specific but important question already mentioned
at the beginning:
Are all exact functors between the bounded derived categories
of smooth projective varieties of Fourier–Mukai type?
This is certainly one of the main open problems in the literature concerning the special
geometric incarnation of the theory of derived categories. Our aim is to survey the
more recent approaches to it and, at the same time, to analyze other related questions
concerning, for example, the uniqueness of the Fourier–Mukai kernels. The relevance
of the question above cannot be overestimated. Indeed, once we know that an exact
functor is of Fourier–Mukai type and the base field is C, then we can study its action
on various cohomology groups and deform it along with the varieties. In Section 2 we
survey some of these issues.
The main problems we want to consider are listed in Section 3.1. The breakthroughs
in the theory are contained in [46] and, more recently, in [36], where new inputs from
the theory of dg-categories are taken into account. Namely,
(A) Orlov [46]: If F W Db .X1 / ! Db .X2 / is a fully faithful functor and X1 , X2 are
smooth projective varieties, then there exists a unique (up to isomorphism) E 2
Db .X1 X2 / and an isomorphism of exact functors F Š Ê (see Theorem 3.1).
(B) Lunts–Orlov [36]: The same holds when X1 and X2 are projective schemes and
we deal with the categories of perfect complexes on them (see Theorem 5.3).
These two results will provide the two leading references in this paper. They will be
explained in Sections 3 and 5 and, at the same time, we will study to which extent we
may expect that they can be extended and generalized. The examples that seem to be
encouraging in this direction are roughly the following (more precise statements are
given in the forthcoming sections):
(a) Toën [52]: Quasi-functors between dg-enhancements of the categories of perfect

complexes on projective schemes (see Theorem 5.8).
(b) Exact functors between the abelian categories of coherent sheaves on smooth
projective varieties (see Proposition 5.15 and [19]).
In both cases, one proves that these functors are of Fourier–Mukai type (in an appro-
priate sense) and that the kernel is unique (up to isomorphism). We also point to [6]
(and [48]) for results extending those in [52].
Fourier–Mukai functors: a survey 29
The fact that an optimistic point of view about extending (A) and (B) in full gener-
ality may be too much is discussed in Section 4.
During the exposition we will explain and list several open problems appearing
naturally in many geometric contexts. They will be presented all along the paper and,
in particular, in Section 6. Motivations are discussed in Section 2. Sections 3 and 5
deal with the main results and techniques now available in the literature. Of course,
we do not pretend to be exhaustive and complete in our presentation. For example,
other overviews on the subject but from completely different perspectives are in [1],
[21] (and, of course, in [23]).
Notation. In the paper, k is a field. Unless otherwise stated, all schemes are assumed to
be of finite type and separated over k; similarly, all additive (in particular, triangulated)
categories and all additive (in particular, exact) functors will be assumed to be k-linear.
An additive category will be called Hom-finite if the k-vector space Hom.A; B/ is finite
dimensional for any two objects A and B. If A is an abelian (or more generally an
exact) category, D.A/ denotes the derived category of A and Db .A/ its full subcategory
of bounded complexes. Unless stated otherwise, all functors are derived even if, for
simplicity, we use the same symbol for a functor and its derived version.
2 Motivations
In this section we would like to motivate the relevance of Fourier–Mukai functors
a bit more. We stress their appearance in moduli problems and we give indications
concerning the way they induce actions on various cohomologies. The reader interested
in an introduction about derived and triangulated categories in geometric contexts can
have a look at [23].
2.1 First properties and examples from moduli problems. There are several in-
stances where Fourier–Mukai functors appear. To make this clear, we discuss some
examples.
Example 2.1. Let X1 and X2 be smooth projective varieties.
(i) Given an object E 2 Db .X1 /, the functor F./ D E ˝ ./ is of Fourier–Mukai
type. Namely, its Fourier–Mukai kernel is the object E, where W X1 ! X1 X1
is the diagonal embedding.
A special example is provided by the Serre functor of Xi which is the exact equiv-
alence SXi ./ D ./ ˝ !Xi Œdim.Xi /, where !Xi is the dualizing sheaf of Xi . Hence
SXi is of Fourier–Mukai type. For later use, set SXi WD !Xi Œdim.Xi /.
(ii) For a given morphism f W X1 ! X2 , denote by f its graph. Then f
is a Fourier–Mukai functor with kernel Of . Analogously, one can show that f
is a Fourier–Mukai functor whose kernel is always Of , providing now a functor
Db .X2 / ! Db .X1 /.
We list here a number of useful properties.
Proposition 2.2. Let X1 and X2 be smooth projective varieties over k and let Ê be
a Fourier–Mukai functor.
(i) The left and right adjoints of Ê exist and are of Fourier–Mukai type with kernels
EL WD E _ ˝ p2 SX2 and ER WD E _ ˝ p1 SX1 respectively, where pi W X1 X2 ! Xi
is the projection.
(ii) The composition of two Fourier–Mukai functors is again of Fourier–Mukai type.
We leave it to the reader to explicitly determine a kernel in (ii) above.

Let us now see some more complicated but interesting examples. Indeed, soon after
[40], it was clear that Fourier–Mukai functors appear in many moduli problems. This
is the case of K3 surfaces (i.e. smooth, compact, complex simply connected surfaces
with trivial canonical bundle) and moduli spaces of stable sheaves on them. Following
[41], let X be a projective K3 surface and M a fine moduli space of stable sheaves
on X with topological invariants fixed in such a way that M is again a projective K3
surface. The universal family E 2 Coh.M X / associated to this moduli problem
provides an equivalence of Fourier–Mukai type
Ê W Db .M / ! Db .X /
sending a skyscraper sheaf to a stable sheaf on X . Most remarkably, it was observed

in [46] that all K3 surfaces Y such that Db .X / Š Db .Y / are actually isomorphic to
moduli spaces of stable sheaves on X .
In higher dimensions the interplay between Fourier–Mukai functors, geometric
problems and moduli interpretations of them have been extensively studied. There
are many occurrences in the context of birational geometry and in the more modern
theory of stability conditions due to Bridgeland. We refrain from discussing them in
this paper.
2.2 Action on (singular) cohomology. Having a description of an exact functor as a

Fourier–Mukai functor allows one to define an action on cohomologies and homologies
of various types. This may be very useful to describe the groups of autoequivalences
of the derived categories of smooth projective varieties, which are rather complicated
algebraic objects as soon as the variety has trivial canonical bundle.
The first highly non-trivial example we have in mind is the group of autoequiva-
lences of the derived category of a projective K3 surface X . This group has a very
complicated structure coming from the presence of the so called spherical objects
in Db .X/ (i.e. objects whose endomorphism graded algebra is isomorphic to the co-
homology of a 2-sphere). The idea proposed in [46] is to approach the analysis of
Aut.Db .X// by studying its action on singular cohomology.
To spell this out clearly, we start with some general remarks. Assume that X1
and X2 are smooth complex projective varieties and let Ê W Db .X1 / ! Db .X2 / be a
Fourier–Mukai functor with kernel E 2 Db .X1 X2 /. Then the induced morphism at
the level of Grothendieck groups is given by the morphism ˆK ŒE

W K.X1 / ! K.X2 /
defined by

ˆK
ŒE .e/ WD .p2 / .ŒE p1 .e//;
where pi W X1 X2 ! Xi is the natural projection.

Going further, for G 2 Db .Xi /, one can consider the Mukai vector
p
v.ŒG / WD ch.G / td.Xi /
of G . When the context is clear, we write v.G / instead of v.ŒG /. Now the morphism
ˆKŒE
W K.X1 / ! K.X2 / gives rise to a map ˆH v.ŒE/
W H .X1 ; Q/ ! H .X2 ; Q/ such
that

ˆH v.ŒE/ W b 7! .p2 / .v.ŒE/ p1 .b//:
The Grothendieck–Riemann–Roch Theorem shows that the following diagram com-

mutes:
ˆK
K.X1 /
ŒE
/ K.X2 / (2.1)
v./ v./
ˆH
H .X1 ; Q/
v.ŒE/
/ H .X2 ; Q/.
From now on, given a Fourier–Mukai functor Ê W Db .X1 / ! Db .X2 /, we denote

ˆH
v.ŒE/
by ˆH
E . The following is a fairly easy remark from [46].

Proposition 2.3. With the above assumptions, the morphism ˆH E W H .X1 ; Q/ !

H .X2 ; Q/ is an isomorphism of Q-vector spaces if Ê is an equivalence.
L For a positive integer n, one may take the Hodge decomposition H n .Xi ; C/ Š
pCqDn H .Xi /. A Fourier–Mukai equivalence does not preserve such a decom-
p;q
position as, in general, it does not preserve the grading of the cohomology rings.
Nevertheless, one has the following.
Proposition 2.4. If Ê is an equivalence, the morphism ˆH
E induces isomorphisms
M M
H p;q .X1 / Š H p;q .X2 /
pqDi pqDi
for all integers i .

The vector space H .Xi ; C/ can be endowed with some additional structure.
P L _
Pp j
Namely, for v D vj 2 j H .Xi ; C/, set v
j
WD 1 vj . Then, for all
v; w 2 H .Xi ; C/, we can define the Mukai pairing
Z
hv; wiXi WD exp.c1 .Xi /=2/:.v _ :w/:
Xi
Proposition 2.5. If Ê is an equivalence, then the morphism ˆH

E preserves the Mukai
pairing.
Before going back to specific examples, let us mention a property that will be
discussed later on in a different context. Here we assume that Ê ; ˆF W Db .X1 / !
Db .X2 / are Fourier–Mukai functors and not necessarily equivalences.
Lemma 2.6. If ˆH
E D ˆF , then v.ŒE/ D v.ŒF /.
H

Proof. The morphisms ˆHE and ˆF are induced by objects in H .X1 X2 ; Q/. Now
H
apply the Künneth decomposition for the cohomology of the product to get v.E/ D
v.F /.
In particular, this means that the ‘cohomological Fourier–Mukai kernel’ of coho-

mological Fourier–Mukai functors is always uniquely determined. Due to what we
will show in Section 4, one can speak about the action of a Fourier–Mukai functor,
being independent of the choice of the Fourier–Mukai kernel.
Assume now that X1 and X2 are projective K3 surfaces and take a Fourier–Mukai
equivalence Ê W Db .X1 / ! Db .X2 /. A remark by Mukai shows that ˆH E induces

an isomorphism of Z-modules H .X1 ; Z/ Š H .X2 ; Z/ in this case. The total
cohomology H .Xi ; Z/ endowed with the Mukai pairing and the Hodge structure
mentioned in Proposition 2.4, is called the Mukai lattice and denoted by Hz .Xi ; Z/.
Using the action of equivalences on cohomology and a bit of lattice theory, one can
prove the following.
Proposition 2.7 ([12], Proposition 5.3). Given a projective K3 surface X , the number
of isomorphism classes of K3 surfaces Y such that Db .X / Š Db .Y / is finite.
Nevertheless such a number can be arbitrarily large.
Proposition 2.8 ([43] and [51]). For any positive integer N , there exist non-isomorphic
K3 surfaces X1 ; : : : ; XN such that Db .Xi / Š Db .Xj / for i; j D 1; : : : ; N .
Two smooth projective varieties X1 and X2 such that Db .X1 / Š Db .X2 / are usually
called Fourier–Mukai partners. Notice that Proposition 2.7 is a special instance of the
following conjecture which is nothing but Conjecture 1.5 in [29].
Conjecture 2.9 (Kawamata). The number of Fourier–Mukai partners up to isomor-

phism of a smooth projective variety is finite.
Abelian varieties satisfy this prediction as well (see [45]). In [2], the authors provide
further evidence for it.
To give one more important application of the discussion in this section, we can go
back to the problem mentioned at the beginning of this section and use the structure of
Fourier–Mukai functors to get a (partial) description of the group of autoequivalences
of a K3 surface X . The following is the result of the papers [22], [24], [46].
Theorem 2.10. For a K3 surface X , there exists a surjective morphism
z .X; Z//
Aut.Db .X // ! OC .H
sending a Fourier–Mukai equivalence Ê to ˆH

E .
Here OC .H z .X; Z// is the group of Hodge isometries of the Mukai lattice preserving
the orientation of some 4-dimensional (real) vector subspace of H .X; R/.
2.3 Hochschild homology, cohomology and deformations. For many geometric

purposes, the cohomology theory one may want to consider is Hochschild coho-
mology (and homology). More precisely, assume that a Fourier–Mukai equivalence
Ê W Db .X1 / ! Db .X2 / between the bounded derived categories of the smooth com-
plex projective varieties X1 and X2 is given. Then one may want to study (first or-
der) deformations of Xi compatible with deformations of the Fourier–Mukai kernel
E 2 Db .X1 X2 /. To this end, we indeed have to study Hochschild cohomology and
homology and the corresponding actions of Ê .
If X is a smooth projective variety and !X is its dualizing sheaf, we define SX as in
Example 2.1, SX1 WD !X_ Œ dim.X / and S ˙1
WD ./ SX˙1 , where W X ,! X X
is the diagonal embedding. The i -th Hochschild homology and cohomology groups,
i 2 Z, are respectively (see, for example, [15])
1
HHi .X/ WD HomDb .XX/ .S Œi ; O / Š HomDb .X/ .OX Œi ; O /;
HHi .X/ WD HomDb .XX/ .O ; O Œi / Š HomDb .X/ . O ; OX Œi /:
L L
Set HH .X/ WD i HHi .X / and HH .X / WD i HHi .X /. The Hochschild–Kostant–
Rosenberg isomorphisms are graded isomorphisms
M
IHKR
X
W HH .X / ! H .X / WD Hi .X /;
i
M

IXHKR W HH .X / ! HT .X / WD HTi .X /;
i
L q L
where Hi .X/ WD qpDi H .X; X / and HT .X / WD
p i
pCqDi H .X; ^ TX /.
p q
One then defines the graded isomorphisms
IKX D .td.X /1=2 ^ .// B IHKR

X
;
IXK D .td.X /1=2 y.// B IXHKR :
From [14], [15], we get a functorial graded morphism .Ê /HH W HH .X1 / !
HH .X2 /. The following shows the compatibility between this action and the one
described in Section 2.2. It is based on [39].
Theorem 2.11 ([38], Theorem 1.2). Let X1 and X2 be smooth complex projective
varieties and let E 2 Db .X1 X2 /. Then the following diagram
.Ê /HH
HH .X1 / / HH .X2 /
X X
IK 1 IK 2
ˆH
H .X1 ; C/
E
/ H .X2 ; C/
commutes.
If Ê is an equivalence, then there exists also an action .Ê /HH on Hochschild
cohomology induced by the functor ÊP W Db .X1 X1 / ! Db .X2 X2 /, where
P Š EL Š ER is the kernel of the inverse of Ê , which sends OX1 to OX2 (see,
for example, [23], Remark 6.3).
Now the second Hochschild cohomolgy group controls first order deformations of a
smooth projective variety. Hence, given a Fourier–Mukai equivalence Ê W Db .X1 / !
Db .X2 / and combining the actions .Ê /HH , .Ê /HH and Theorem 2.11, one can control
first order deformations of X1 and X2 compatible with deformations of the Fourier–
Mukai functor Ê . This was done, for example, in [24].
Interesting recent developments are contained in [3], where the authors deal with
fully faithful Fourier–Mukai functors whose kernel is a (shift of a) sheaf.
3 The main problems and the first improvements

In this section we list the main problems that we want to address. The answers to them
which are available in the literature will be presented in Section 4. For the moment we
content ourselves with a discussion of a celebrated result of Orlov about Fourier–Mukai
functors. Various generalizations or attempts to weaken the hypotheses in this result
are discussed in this section as well.
3.1 The questions. Assume for the moment that all the varieties are smooth and
projective. The most important problems concerning Fourier–Mukai functors may be
summarized by the following two questions:
(1) Are all exact functors between the bounded derived categories of coherent sheaves
on smooth projective varieties of Fourier–Mukai type?
(2) Is the kernel of a Fourier–Mukai functor unique (up to isomorphism)?
A positive answer to the first one was conjectured in [8] as a consequence of a conjec-
ture about the possibility to lift all exact functors to the corresponding dg-enhancements.
In these terms, a positive or negative answer to the second one implies the uniqueness
or non-uniqueness of such dg-lifts.
We can now put these questions in a more general setting. Indeed, consider the
category ExFun.Db .X1 /; Db .X2 // of exact functors between Db .X1 / and Db .X2 /
(with morphisms the natural transformations compatible with shifts) and define the
functor
1 !X2
ˆX W Db .X1 X2 / ! ExFun.Db .X1 /; Db .X2 // (3.1)
by sending E 2 Db .X1 X2 / to the Fourier–Mukai functor Ê . Thus we can formulate
the following problems:
(Q1) Is ˆX 1 !X2 essentially surjective?

(Q2) Is ˆX 1 !X2 essentially injective?

(Q3) Is ˆX 1 !X2 faithful?

(Q4) Is ˆX 1 !X2 full?

(Q5) Does ExFun.Db .X1 /; Db .X2 // have a triangulated structure making ˆX 1 !X2

exact?
Clearly, (Q1) and (Q2) are precisely (1) and (2), respectively. Căldăraru provided a
negative answer to (Q3) in [13], Example 6.5 (see also [47]), while a negative answer
to (Q5) was expected already in [8], [52]. Nevertheless, in the seminal paper [46] a
positive answer to (1) and (2) has been provided under some additional assumption on
the exact functor. In the original formulation, it can be stated as follows:
Theorem 3.1 (Orlov). Let X1 and X2 be smooth projective varieties and let
F W Db .X1 / ! Db .X2 /
be an exact fully faithful functor admitting a left adjoint. Then there exists a unique
(up to isomorphism) E 2 Db .X1 X2 / such that F Š Ê .
A generalization to smooth stacks (actually obtained as global quotients) is con-
tained in [30]. In the rest of this section and as a preparation for a complete discussion
of (Q1)–(Q5) that will be carried out in Sections 4 and 5, we start discussing how one
may try to weaken the hypotheses of the above result.
3.2 Existence of adjoints. Of course, in purely categorical terms, the existence of

adjoints to a given functor is not automatic. In this section we will see a first approach,
due to Bondal and Van den Bergh, to make this straightforward in the geometric setting
we are dealing with.
P general setting where T is an Ext-finite triangulated
Let us start from the more
category. This means that n dimk Hom.A; BŒn/ < 1, for all A; B 2 T. Denote
by Vect-k the category of k-vector spaces. A contravariant functor H W T ! Vect-k is

cohomological if, given a distinguished triangle
A ! B ! C
in T, the sequence
H.C / ! H.B/ ! H.A/
in Vect-k. A cohomological contravariant functor H is said to be of finite type
is exactL
if dimk i H.AŒi / < 1, for all A 2 T.
Definition 3.2. The triangulated category T is (right) saturated if every cohomological
contravariant functor H of finite type is representable, i.e. there exists A 2 T and an
isomorphism of functors
H Š Hom.; A/:
Remark 3.3. (i) By the Yoneda Lemma, if a cohomological functor H is representable,
then the object representing it is unique (up to isomorphism).
(ii) In [10], the authors provide examples of ‘geometric’ categories which are not
saturated. Namely, if X is a smooth compact complex surface containing no compact
curves, then Db .X / is not saturated. Examples in higher dimensions are given in [42].
In the smooth proper case one has the following result.
Theorem 3.4 ([10], Theorem 1.1). Assume that X is a smooth proper scheme over k.
Then Db .X/ is saturated.
Now assume that X1 and X2 are smooth proper schemes. As an application of the
above theorem, we get the following well-known result.
Proposition 3.5. Any exact functor F W Db .X1 / ! Db .X2 / has left and right adjoints.
Proof. For any F 2 Db .X2 / the functor Hom.F./; F / is representable by a unique
E 2 Db .X1 / due to Theorem 3.4. Setting G.F / WD E, by the Yoneda Lemma we get a
functor G W Db .X2 / ! Db .X1 / which is right adjoint to F. Since Db .X1 / and Db .X2 /
have Serre functors, it is a very easy exercise to prove that F has also a left adjoint.
Observe that, due to Proposition 1.4 of [7], the right and left adjoints in the above
statement are automatically exact.
3.3 The algebricity assumption. In this section we show in which sense it is impor-
tant to work with algebraic varieties. In particular, we give examples of exact functors
between the bounded derived categories of coherent sheaves on smooth compact com-
plex manifolds which are not of Fourier–Mukai type.
For this, let X be a generic non-projective K3 surface. With this we mean a K3
surface X such that Pic.X / D 0. The following surprising result shows that the
abelian categories of coherent sheaves on those surfaces are not fine invariants (see,
for example, [37] for a brief account about coherent sheaves and Chern characters in
this setting).
Theorem 3.6 ([53]). Let X1 and X2 be generic non-projective K3 surfaces. Then there
exists an equivalence of abelian categories Coh.X1 / Š Coh.X2 /.
Remark 3.7. (i) In the case of smooth projective varieties X1 and X2 a result of Gabriel
(see Corollary 5.24 in [23] for an easy proof using Fourier–Mukai functors) asserts that
exactly the converse holds. Namely X1 Š X2 if and only if Coh.X1 / Š Coh.X2 /.
(ii) The above result was proved in [54] for the case of generic non-projective
complex tori as well.
Now take two non-isomorphic generic non-projective K3 surfaces X1 and X2 .
Theorem 3.6 implies that there exists an exact equivalence
F W Db .X1 / ! Db .X2 /:
One may then wonder whether all such equivalences are of Fourier–Mukai type.
Proposition 3.8. Let X1 and X2 be non-isomorphic generic non-projective K3 sur-
faces and let F W Db .X1 / ! Db .X2 / be the exact equivalence induced by an exact
equivalence Coh.X1 / Š Coh.X2 /. Then F is not of Fourier–Mukai type.
Proof. By assumption, F sends the minimal objects in Coh.X1 / to minimal objects in
Coh.X2 / (recall that an object in an abelian category is minimal if it does not admit
proper subobjects). In particular, following the same argument as in the proof of
Corollary 5.24 in [23], we get that F sends skyscraper sheaves to skyscraper sheaves.
Hence if F Š Ê , for some E 2 Db .X1 X2 /, then there should be an isomorphism
f W X1 ! X2 and a line bundle L 2 Pic.X2 / such that F Š .L ˝ .// B f (see, for
example, [23], Corollary 5.23). But this contradicts the assumption X1 6Š X2 .
3.4 Non fully faithful functors. Now we discuss how the fully faithfulness assump-
tion can be removed. We first discuss a generalization of Theorem 3.1 while later we
observe that the faithfulness assumption is redundant anyway. Indeed full functors turn
out to be automatically faithful.
3.4.1 Negative Hom’s and sheaves. We now see a way to reduce the assumptions on
the functor F, that, to our knowledge, is the best one available in the literature in the
context of smooth projective varieties. We will see later on how this has to be modified
for perfect complexes on singular (projective) varieties. Some details about the key
ingredients in the proof will be discussed in Section 4.
Theorem 3.9 ([19], Theorem 1.1). Let X1 and X2 be smooth projective varieties and
let F W Db .X1 / ! Db .X2 / be an exact functor such that, for any F ; G 2 Coh.X1 /,
HomDb .X2 / .F.F /; F.G /Œj / D 0 if j < 0: (3.2)
Then there exist E 2 Db .X1 X2 / and an isomorphism of functors F Š Ê . Moreover,

E is uniquely determined up to isomorphism.
A class of exact functors satisfying (3.2) is clearly provided by full functors. Un-
fortunately this is not a really interesting case, as in Section 3.4.2 we will show that,
in the present context, all full functors are actually automatically faithful.
Example 3.10. For a rather trivial example of a non-full exact functor satisfying (3.2),
we can consider id ˚ id W Db .X / ! Db .X /, where X is a smooth projective variety.
More generally, given a line bundle L 2 Pic.X /, we can take ˆ L ˚ ˆ L (see
Example 2.1).
Example 3.11. Notice that all exact functors Db .X1 / ! Db .X2 / obtained by deriving
an exact functor Coh.X1 / ! Coh.X2 / are examples of functors satisfying (3.2).
Remark 3.12. The original version of Theorem 3.9, stated in [19], deals with the more
general notion of twisted variety where condition (3.2) can be stated as well.
3.4.2 Full implies faithful. In this section we assume that k is algebraically closed
of characteristic 0. Let X1 and X2 be smooth projective varieties and assume that an
exact functor F W Db .X1 / ! Db .X2 / is full and such that F 6Š 0. By Theorem 3.9, F
is a Fourier–Mukai functor. So F Š Ê , for some E 2 Db .X1 X2 /.
There exists a very useful criterion to establish when a Fourier–Mukai functor
Ê W Db .X1 / ! Db .X2 / is fully faithful.
Theorem 3.13 ([9] and [11]). Under the assumptions above, Ê is fully faithful if and
only if
´
k if x1 D x2 and i D 0;
HomDb .X2 / .Ê .Ox1 /; Ê .Ox2 /Œi / Š
0 if x1 ¤ x2 or i 62 Œ0; dim.X1 /
for all closed points x1 ; x2 2 X1 :

Thus, because of this result and the fact that F is full, to show that the functor is
also faithful it is enough to prove that there are no closed points x 2 X1 such that
Hom.F.Ox /; F.Ox // D 0 or, in other words, such that F.Ox / Š 0:
To see this, take the left adjoint G W Db .X2 / ! Db .X1 / of F and consider the
composition G B F which is again a Fourier–Mukai functor (see Proposition 2.2), hence
isomorphic to ˆF for some F 2 Db .X1 X1 /: Assume that there are x1 ; x2 2 X1
such that F.Ox1 / 6Š 0 while F.Ox2 / Š 0: By [9] (see, in particular, Proposition 1.5
there) the Mukai vector v.ˆF .Ox1 // is not zero.
On the other hand, by Propositions 2.3, 2.4 and 2.5, the functor ˆF induces a

morphism ˆH F W H .X1 ; Q/ ! H .X1 ; Q/ such that
0 ¤ v.ˆF .Ox1 // D ˆH
F .v.Ox1 // D ˆF .v.Ox2 // D v.ˆF .Ox2 // D 0:
H
This contradiction proves that, if F were not faithful, then F.Ox / Š 0 for every closed
point x 2 X.
We claim that if this is true, then F Š 0. Indeed let G and H be the left and right
adjoints of F. Of course, G B F.Ox / Š 0, for all closed points x in X1 . In particular,
for all n 2 Z and any B 2 Db .X1 /, we have
0 D Hom.G B F.Ox /; BŒn/ Š Hom.Ox ; H B F.B/Œn/:
Therefore H B F.B/ Š 0, for all B 2 Db .X1 /. But now
0 D Hom.B; H B F.B// Š Hom.F.B/; F.B//:
Thus we would get F.B/ Š 0, for all B 2 Db .X1 / and so we proved the following
result.
Theorem 3.14. Let X1 and X2 be smooth projective varieties over an algebraically
closed field of characteristic 0 and assume that an exact functor F W Db .X1 / ! Db .X2 /
is full. If F 6Š 0, then F is faithful as well.
Remark 3.15. (i) Notice that in [16] a more general result is proved. In particular, the
target category can be any triangulated category while the source category can be the
category of perfect (supported) complexes on a noetherian scheme.
(ii) One may easily extend the proof above to the case of twisted varieties. For this
we just need to use the twisted version of the Chern character defined in [25] and again
apply Proposition 1.5 of [9]. We leave this to the reader.
4 The (partial) answers to (Q2)–(Q5)

We postpone for the moment the discussion about (Q1) which will be examined in Sec-
tion 5. The remaining problems can be studied in a unitary way explained here below.
4.1 Perfect complexes and good news. We start our discussion with a case where
all the above five questions have a positive answer. In particular, this implies that (in
the smooth case) interesting examples answering these questions negatively have to be
searched for in dimension greater than zero.
We begin by extending the setting explained in the previous section. In particular,
let X be a projective (not necessarily smooth) scheme over k. Denote by Perf .X / the
category of perfect complexes on X consisting of the objects in D.Qcoh.X // which
are quasi-isomorphic to bounded complexes of locally free sheaves of finite type over
X. Obviously, Perf .X / Db .X / and the equality holds if and only if X is regular.
The category Perf .X / coincides with the full subcategory of compact objects in
D.Qcoh.X//. Recall that an object A in a triangulated
L category T is compact if, for
each family of objects fXi gi2I T such that i Xi exists in T, the canonical map
M
Hom.A; Xi / ! Hom .A; ˚i Xi /
i
is an isomorphism.
In the singular setting we redefine the notion of Fourier–Mukai functors once

more since in general we cannot expect the Fourier–Mukai kernels of exact functors
Perf .X1 / ! Perf .X2 / to be objects in Perf .X1 X2 /, but rather in Db .X1 X2 /.
More precisely, one can show the following (see, for example, Lemma 4.3 in [18] for
the proof).
Lemma 4.1. Let X1 and X2 be projective schemes and let E 2 D.Qcoh.X1 X2 // be
an object such that Ê W D.Qcoh.X1 // ! D.Qcoh.X2 // (defined as in (1.1)) sends
Perf .X1 / to Db .X2 /. Then E 2 Db .X1 X2 /. Conversely, any E 2 Db .X1 X2 /
yields a Fourier–Mukai functor Ê W Perf .X1 / ! Db .X2 /.
Hence given two projective schemes X1 and X2 one can consider the functor
1 !X2
ˆX
W Db .X1 X2 / ! ExFun.Perf .X1 /; Db .X2 //
(which coincides with (3.1) in the smooth case) and for it one can again ask questions
(Q1)–(Q5).
Now, if X is a projective scheme over k, it is an easy exercise to show that every
exact functor F W Perf .Spec k/ D Db .Spec k/ ! Db .X / is of Fourier–Mukai type.
More precisely, there exists an isomorphism of exact functors F Š Ê , where
E WD F.OSpec k / 2 Db .X / D Db .Spec k X /:
Spec k!X
It is also straightforward to see that the functor ˆ is an equivalence of cate-
gories, so that all the above questions have a positive answer in this case.
If we exchange the role of X and Spec k above, the situation becomes slightly more
k
complicated but nevertheless it is not difficult to see that ˆX!Spec
is an equivalence
as well. Indeed, as an easy consequence of [49], Corollary 7.50 (see also Theorem 3.3
in [4]), there is an equivalence
Db .X / ! ExFun.Perf .X /B ; Db .Spec k//

k
and one can check that ˆX!Spec
is induced from this by the exact anti-equivalence
Perf .X/ Š Perf .X / sending F to F _ .
B
4.2 Non-uniqueness of Fourier–Mukai kernels. The aim of this section is to prove

that, even in the smooth case, (Q2) has a negative answer in general. First observe that
the functor ˆX 2 !X1 satisfies any of (Q1)–(Q5) if and only if ˆX1 !X2 does. To see this,

one identifies ˆX 2 !X1 with the opposite functor of ˆX1 !X2 under the equivalences

Db .X1 X2 / ! Db .X1 X2 /B (defined on the objects by E 7! E _ ˝ p1 !X1 Œd1 )
and ExFun.Db .X1 /; Db .X2 // ! ExFun.Db .X2 /; Db .X1 //B (defined on the objects
by F 7! F , the right adjoint of F). A key ingredient for this is Proposition 3.5. Here
we set di WD dim.Xi /.
For later use, we start studying the case of the projective line which provides a
positive result related to (Q2).
Lemma 4.2. If X1 or X2 is P 1 , then ˆX 1 !X2 is essentially injective.

Proof. As observed above, we can assume that X1 D P 1 . Since on P 1 P 1 there is a

resolution of the diagonal of the form
x0 x1 x1 x0
0 ! OP 1 P 1 .1; 1/ ! OP 1 P 1 ! O ! 0;
the argument in [19], Section 4.3, shows that, for every exact functor F W Db .P 1 / !
Db .X2 /, any object E in Db .P 1 X2 / such that F Š Ê is necessarily a convolution
of the complex
'WDx0 F.x1 /x1 F.x0 /
OP 1 .1/ F.OP 1 .1// ! OP 1 F.OP 1 /;
hence it is uniquely determined up to isomorphism as the cone of '.
As soon as the genus of the curve grows, the situation becomes more complicated
and, in a sense, more interesting. Indeed, we have the following result that is Theo-
rem 1.1 of [18].
Theorem 4.3. For every elliptic curve X over an algebraically closed field there exist
E1 ; E2 2 Db .X X / such that E1 6Š E2 but Ê1 Š Ê2 .
There is no space to explain the proof of this result in detail. Let us just mention
how the two kernels are defined. By Serre duality,
0 ¤ Hom.O ; O /_ Š Hom.O Œ1; O Œ1/;
where O D OX 2 Db .X X /. For 0 ¤ ˛ 2 Hom.O Œ1; O Œ1/, we set
E1 WD O ˚ O Œ1; E2 WD Cone.˛/:
It makes then perfect sense to pose the following.
Problem 4.4. Extend the non-uniqueness result in Theorem 4.3 to any curve of genus
1.
In [18] we provided our best approximation to the uniqueness of the Fourier–Mukai
kernels.
Theorem 4.5 ([18], Theorem 1.2). Let X1 and X2 be projective schemes and let
F W Perf .X1 / ! Db .X2 / be an exact functor. If F Š Ê for some E 2 Db .X1 X2 /,
then the cohomology sheaves of E are uniquely determined (up to isomorphism) by F.
Using the discussion in Section 2.2 we can derive the following straightforward
consequence from the above result. We will always assume that X1 and X2 are smooth
projective varieties.
Corollary 4.6. Let E1 ; E2 2 Db .X1 X2 / be such that Ê1 Š Ê2 W Db .X1 / !
Db .X2 /. Then ŒE1 D ŒE2 in K.X1 X2 / and so ˆK
E1 D Ê2 and Ê1 D Ê2 .
K H H
4.3 The remaining questions (Q3)–(Q5). Let us first consider the case of smooth
projective curves.
Proposition 4.7 ([18], Proposition 2.3). Set di WD dim.Xi /. If minfd1 ; d2 g D 1, then

ˆX 1 !X2 is neither faithful nor full.

Proof. We give a full proof only of the non-faithfulness, as it plays a role in the study of
(Q5) below. As above, we can assume that 1 D d1 d2 . Hence take a finite morphism
f W X1 ! P d2 and a finite and surjective (hence flat) morphism g W X2 ! P d2 . Then
F WD g B f W Coh.X1 / ! Coh.X2 / is an exact functor, which trivially extends
to an exact functor again denoted by F W Db .X1 / ! Db .X2 /. Clearly there exists
0 6Š E 2 Db .X1 X2 / such that F Š Ê (see Example 2.1 and Proposition 2.2).
Now observe that, by Serre duality,
HomDb .X1 X2 / .E; E/ Š HomDb .X1 X2 / .E; E ˝ !X1 X2 Œ1 C d2 /_ ;
so there exists 0 ¤ ˛ 2 HomDb .X1 X2 / .E; E ˝ !X1 X2 Œ1 C d2 /. Since !X1 X2 Š
p1 !X1 ˝ p2 !X2 , this induces, for any F 2 Coh.X1 /, a morphism
ˆ˛ .F / W Ê .F / Š F.F / ! Ê˝!X Œ1Cd2 .F / Š F.F ˝ !X1 / ˝ !X2 Œ1 C d2 :

1 X2
As F.F / and F.F ˝ !X1 / are objects of Coh.X2 /, it follows that ˆ˛ .F / D 0, whence
ˆ˛ D 0 because every object of Db .X1 / is isomorphic to the direct sum of its (shifted)
cohomology sheaves (since the abelian category Coh.X1 / is hereditary).
As for non-fullness, we prove it only when X1 D X2 D X is an elliptic curve
and k is algebraically closed. By Theorem 4.3 there exist E1 ; E2 2 Db .X X / with

E1 6Š E2 and an isomorphism W Ê1 ! Ê2 . Then we claim that there is no
morphism f W E1 ! E2 such that D ˆf X!X
. Indeed, assume that such an f exists.
Then it can be completed to a distinguished triangle
f
E1 / E2 / G;
for some G 2 Db .X1 X2 /. By assumption ˆG .A/ D 0, for all A 2 Db .X1 /. There-

fore ˆG Š 0, whence G Š 0 by Theorem 3.9. But then f would be an isomorphism,
contradicting the assumption E1 6Š E2 .
We finally recall how (Q5) is studied in [18]. For this we need a couple of easy
lemmas.
Lemma 4.8. Let T be a Hom-finite triangulated category and let f W A ! B be a

morphism of T. Then Cone.f / Š AŒ1 ˚ B if and only if f D 0.
Proof. The other implication being well-known, we assume that Cone.f / Š AŒ1˚B.
f
Applying the cohomological functor Hom.; B/ to the distinguished triangle A
!
B ! AŒ1 ˚ B ! AŒ1, one gets an exact sequence of finite dimensional k-vector

spaces
./Bf
Hom.AŒ1; B/ ! Hom.AŒ1 ˚ B; B/ ! Hom.B; B/ ! Hom.A; B/:
For dimension reasons, the last map must be 0, hence f D 0.
Lemma 4.9. Let F W T ! T0 be an exact functor between triangulated categories and

assume that T is Hom-finite. If F is essentially injective, then F is faithful, too.
Proof. Let f W A ! B be a morphism of T such that F.f / D 0. Then
F.Cone.f // Š Cone.F.f // Š F.A/Œ1 ˚ F.B/ Š F.AŒ1 ˚ B/
in T0 , whence Cone.f / Š AŒ1 ˚ B in T because F is essentially injective. It follows

from Lemma 4.8 that f D 0.
Recollecting the above results, we get the following.
Proposition 4.10 ([18], Corollary 2.7). If d1 ; d2 > 0 and X1 or X2 is P 1 , then there

is no triangulated structure on ExFun.Db .X1 /; Db .X2 // such that ˆX 1 !X2 is exact.

Proof. This follows from Lemma 4.9, since we know that in this case ˆX 1 !X2 is

essentially injective by Lemma 4.2, but not faithful by Proposition 4.7.
Notice that, as observed in [52], there is no natural triangulated structure on the

category ExFun.Db .X1 /; Db .X2 //. One can then pose the following question.
Problem 4.11. Understand whether there may be smooth projective varieties X1 and
X2 of positive dimension such that (Q5) has a positive answer.
5 Existence of Fourier–Mukai kernels and (Q1)

We are now ready to discuss the partial answers to (Q1) actually present in the literature.
As we have already observed, we need to impose rather strong conditions on the exact
functors in order to get nice results.
5.1 The non-smooth case. The idea of studying Fourier–Mukai functors between
triangulated categories associated to singular varieties explained in the baby examples
in Section 4.1 has been extensively analyzed in [36] using new ideas coming from
dg-categories. Let us start from the following result.
Proposition 5.1 ([36], Corollary 9.12). Let X1 and X2 be quasi-compact separated

schemes over k. Assume that X1 has enough locally free sheaves and let F W Perf .X1 / !
D.Qcoh.X2 // be a fully faithful exact functor that commutes with direct sums. Then
there is an E 2 D.Qcoh.X1 X2 // such that the functor Ê is fully faithful and
Ê .A/ Š F.A/ (5.1)
for any A 2 Perf .X1 /.
Needless to say, the existence of the isomorphism (5.1) is a rather weak condition
because, already in the smooth case, it may not extend to an isomorphism of functors.
To show that this is possible, consider the case of P 1 P 1 . Exactly as in Section 4.2,
observe that, by Serre duality,
0 ¤ Hom.O ; O /_ Š Hom.O Œ1; O ˝ !P 1 P 1 Œ1/:
Hence take a non-trivial ˛ W O Œ1 ! O ˝ !P 1 P 1 Œ1 Š !P˝2
1 Œ1 and consider
the objects
E1 WD O ˚ !P˝2 1 Œ1; E2 WD Cone.˛/:
Then one has the following easy result.
Lemma 5.2. For every A 2 Db .P 1 / we have Ê1 .A/ Š Ê2 .A/ but Ê1 6Š Ê2 .
Proof. The existence of an isomorphism Ê1 .A/ Š Ê2 .A/ for any A 2 Db .P 1 /
is obvious. The fact that Ê1 6Š Ê2 follows from the uniqueness of Fourier–Mukai
kernels for P 1 (see Lemma 4.2) and the fact that E1 6Š E2 .
On the other hand, putting some more hypotheses on the schemes, we get a global
isomorphism, as stated in the following theorem which is [36], Corollary 9.13. For a
scheme X, denote by T0 .OX / the maximal 0-dimensional torsion subsheaf of OX .
Theorem 5.3 (Lunts–Orlov). Let X1 be a projective scheme with T0 .OX1 / D 0 and
assume that X2 is a noetherian separated scheme over k. Given an exact fully faithful
functor F W Perf .X1 / ! Db .X2 /, there are an E 2 Db .X1 X2 / and an isomorphism
of exact functors Ê Š F.
Remark 5.4. The kernel turns out to be unique in perfect analogy with Theorem 3.1.
This is observed in [17], following a suggestion by Orlov.
There is another approach to the Fourier–Mukai functors in the non-smooth case
due to Ballard.
Theorem 5.5 ([4], Theorem 1.2). Let X1 and X2 be projective schemes such that
T0 .OX1 / D 0. If F W Perf .X1 / ! Perf .X2 / is a fully faithful exact functor with left
and right adjoints, then there are an E 2 Db .X1 X2 / and an isomorphism of exact
functors Ê Š F.
As remarked in [4], contrary to the smooth case, the existence of the adjoints is not
automatic at all. On the other hand, the proof of Theorem 5.5 differs from the one of
Theorem 5.3 as it does not make use of dg-categories and is closer to the spirit of the
one of Theorem 3.1.
5.2 Some ingredients in the proof of Theorem 5.3. A complete account of the details
of the proof of Theorem 5.3 is far beyond the scope of this paper. Nevertheless, there
are at least three main steps in it which we want to highlight as they provide sources
of interesting (and difficult) open problems.
5.2.1 Dg-categories. First one wants to find an object E 2 Db .X1 X2 / to compare

the functors F and Ê . This is done by passing to dg-enhancements and using a
celebrated result of Toën.
Recall that a dg-category is an additive category A such that, for all A; B 2
Ob.A/, the morphism spaces Hom.A; B/ are Z-graded k-modules with a differen-
tial d W Hom.A; B/ ! Hom.A; B/ of degree 1 compatible with the composition.
Given a dg-category A we denote by H 0 .A/ its homotopy category. The objects of
H .A/ are the same as those of A while the morphisms are obtained by taking the 0-th
0
cohomology H 0 .HomA .A; B// of the complex HomA .A; B/. If A is pre-triangulated
(see [31] for the definition), then H 0 .A/ has a natural structure of triangulated category.
A dg-functor F W A ! B is the datum of a map Ob.A/ ! Ob.B/ and of morphisms
of dg k-modules HomA .A; B/ ! HomB .F.A/; F.B//, for A; B 2 Ob.A/, which are
compatible with the composition and the units.
For a small dg-category A, one can consider the pre-triangulated dg-category
Mod-A of right dg A-modules. A right dg A-module is a dg-functor M W AB ! Mod-k,
where Mod-k is the dg-category of dg k-modules. The full dg-subcategory of acyclic
right dg-modules is denoted by Ac.A/, and H 0 .Ac.A// is a full triangulated sub-
category of the homotopy category H 0 .Mod-A/. Hence the derived category of the
dg-category A is the Verdier quotient
Ddg .A/ WD H 0 .Mod-A/=H 0 .Ac.A//:
According to [31], [52], given two dg-categories A and B, we denote by rep.A; B/

the full subcategory of the derived category Ddg .AB ˝ B/ of A-B-bimodules C such
that the functor ./ ˝A C W Ddg .A/ ! Ddg .B/ sends the representable A-modules to
objects which are isomorphic to representable B-modules. A quasi-functor is an object
in rep.A; B/ which is represented by a dg-functor A ! Mod-B whose essential image
consists of dg B-modules quasi-isomorphic to representable B-modules. Notice that a
quasi-functor M 2 rep.A; B/ defines a functor H 0 .M/ W H 0 .A/ ! H 0 .B/.
Given two pre-triangulated dg-categories A and B, a dg-lift of an exact functor
F W H 0 .A/ ! H 0 .B/ is a quasi-functor G 2 rep.A; B/ such that H 0 .G/ Š F.
An enhancement of a triangulated category T is a pair .A; ˛/, where A is a pre-
triangulated dg-category and ˛ W H 0 .A/ ! T is an exact equivalence. The enhance-
ment .A; ˛/ of T is unique if for any enhancement .B; ˇ/ of T there exists a quasi-
functor W A ! B such that H 0 . / W H 0 .A/ ! H 0 .B/ is an exact equivalence.
Example 5.6. For X a quasi-compact quasi-separated scheme, let Cdg .X / be the dg-
category of unbounded complexes of objects in Qcoh.X /. Denote by Acdg .X / the
full dg-subcategory of Cdg .X / consisting of acyclic complexes. Following [20], we

take the quotient Ddg .X / WD Cdg .X /=Acdg .X / which is again a dg-category. This
dg-category Ddg .X / is pre-triangulated and H 0 .Ddg .X // Š D.Qcoh.X // (see [31],
[52]). Therefore it is an enhancement of D.Qcoh.X //.
Consider then the full dg-subcategory Perf dg .X / whose objects are all the per-
fect complexes in D.Qcoh.X //. It turns out (see, for example, [36], Section 1) that
Perf dg .X/ is an enhancement of Perf .X /.
The following result answers positively a conjecture in [8]. The reader can have a
look at Section 9 in [36] for stronger statements.
Theorem 5.7 ([36], Theorem 7.9). The triangulated category Perf .X / on a quasi-
projective scheme X has a unique enhancement.
Given a functor F W Perf .X1 / ! DbCoh .Qcoh.X2 // as in the statement of Theo-
rem 5.3, Lunts and Orlov construct in a highly non-trivial way a quasi-functor
Fdg W Perf dg .X1 / ! Ddg .X2 /:
Now one can use the following.

Theorem 5.8 ([52], Theorem 8.9). Let X1 and X2 be quasi-compact and separated
schemes over k. Then we have a canonical quasi-equivalence

Ddg .X1 X2 /
! RHomc .Ddg .X1 /; Ddg .X2 //;
where RHomc denotes the dg-category formed by the direct sums preserving quasi-
functors (i.e. their homotopy functors do).
dg
Hence there are an E 2 Ddg .X1 X2 / and an isomorphism Fdg Š Ê and it
remains to show that F Š H 0 .Fdg / Š Ê .
5.2.2 Ample sequences. The projectivity assumption in the statement has a rather
important role. Indeed one needs to work with ample sequences according to the
following.
Definition 5.9. Given a Hom-finite abelian category A, a subset fPi gi2Z Ob.A/ is
an ample sequence if, for any B 2 Ob.A/, there exists an integer i.B/ such that, for
any i i.B/,
(1) the natural morphism HomA .Pi ; B/ ˝ Pi ! B is surjective;
(2) if j ¤ 0 then HomDb .A/ .Pi ; BŒj / D 0;
(3) HomA .B; Pi / D 0.
If X is a projective scheme and H is an ample line bundle on X , then one may
consider the set C (often identified with the corresponding full subcategory of Coh.X /)
consisting of objects of the form OX .iH /, where i is any integer.
Proposition 5.10 ([36], Proposition 9.2). If X is a projective scheme such that

T0 .OX / D 0, then C forms an ample sequence in the abelian category Coh.X /.
Notice that this is the place where the assumption about the maximal torsion sub-
sheaf plays a distinguished role. Thus there is space for further work:
Problem 5.11. Remove the assumption T0 .OX / D 0 and, in particular, find a way to
extend Theorem 5.3 when X1 is a 0-dimensional projective scheme.
At this point Lunts and Orlov show that the Fourier–Mukai functor Ê , with kernel
found in Section 5.2.1, and the given functor F are such that there is an isomorphism

1 W FjC
! Ê jC : (5.2)
Before discussing how this isomorphism can be extended, let us formulate the
following rather general problem.
Problem 5.12. Avoid the use of ample sequences and relax the projectivity assump-
tions.
Both Problem 5.11 and 5.12 are widely open but we believe that any improvement
in these directions may give new important impulses to the theory.
5.2.3 Convolutions. The extension of (5.2) is achieved in two steps. First the exten-
sion takes place on the level of sheaves. And for this one writes every perfect sheaf
(i.e. a coherent sheaf which is a perfect object as well) as a convolution of objects in
the ample sequence C on X1 described in the previous section.
Following [30], [46], recall that a bounded complex in a triangulated category T is
a sequence of objects and morphisms in T
dm dm1 d1
Am ! Am1
! ! A0 (5.3)
such that dj B dj C1 D 0 for 0 < j < m. A right convolution of (5.3) is an object A
together with a morphism d0 W A0 ! A such that there exists a diagram in T
dm dm1 d2 d1
Am= / Am1 / ? / A1 / A0
== }> D D ;; A ::
== ˚ }}}} DD ;; ::
DD ˚ ;; ˚ :
id == } DD ; d0 ::
}} "
Am o Cm1 o o C1 o A;
Œ1 Œ1 Œ1 Œ1
where the triangles with a ˚ are commutative and the others are distinguished.
Roughly speaking, in this part of the argument, we have A 2 Coh.X1 / \ Perf .X1 /
while Ai is a finite direct sum of objects in C, for all i . Unfortunately, to use convolu-
tions one needs to make assumptions on the functor F. The hypothesis in Theorem 5.3
that F is fully faithful goes exactly in this direction. Thus, if we want to substantially
improve Theorem 5.3, one has to address the following:
Problem 5.13. Avoid the use of convolutions.

All in all, we get an isomorphism

2 W FjCoh.X1 /\Perf .X1 /
! Ê jCoh.X1 /\Perf .X1 / :
To produce the desired isomorphism

3 W F
! Ê
one argues by induction on the length of the interval to which the non-trivial coho-
mologies of an object F 2 Perf .X1 / belong.
Remark 5.14. The techniques used to get the extension 3 were improved in [17]
(see, in particular, Sections 3.2 and 3.3 of that paper). Indeed, we consider a wider
class of triangulated categories and we deal with extensions of natural transformations
rather than isomorphisms of functors. These ingredients play a role in the results of
Sections 5.3 and 5.4.
5.3 Exact functors between the abelian categories of coherent sheaves. As pointed
out in Example 3.11, if X1 and X2 are smooth projective varieties, then the functors
induced by exact functors from Coh.X1 / to Coh.X2 / satisfy (3.2), hence Theorem 3.9
holds for them. This suggests that questions analogous to (Q1)–(Q5) should be easier to
answer for exact functors between the abelian categories of coherent sheaves. Indeed,
for them one can prove the following result, improving Proposition 5.1 of [19].
As a matter of notation, if X1 and X2 are smooth projective varieties we denote
by K.X1 ; X2 / the full subcategory of Coh.X1 X2 / having as objects the sheaves E
which are flat over X1 and such that p2 jSupp.E/ W Supp.E/ ! X2 is a finite morphism.
Proposition 5.15. Let X1 , X2 be smooth projective varieties. If E is in Coh.X1 X2 /,
then the additive functor
‰E WD .p2 / .E ˝ p1 .// W Coh.X1 / ! Coh.X2 /
(where .p2 / and ˝ are not derived ) is exact if and only if E 2 K.X1 ; X2 /.
Moreover, if we denote by ExFun.Coh.X1 /; Coh.X2 // the category of exact func-
tors from Coh.X1 / to Coh.X2 /, the functor
X1 !X2
‰ W K.X1 ; X2 / ! ExFun.Coh.X1 /; Coh.X2 //
sending E 2 K.X1 ; X2 / to ‰E is an equivalence of categories.
Proof. We just stick to the second part of the statement and we invite the reader inter-
ested in a proof of the first part to have a look at [19].
X1 !X2
We sketch the proof that ‰ is essentially surjective (again, for more details
see [19]). Hence assume that F W Coh.X1 / ! Coh.X2 / is an exact functor. By Theo-
rem 3.9 there exists (unique up to isomorphism) E 2 Db .X1 X2 / such that the exten-
sion of F to the level of derived categories is isomorphic to Ê , and E 2 Coh.X1 X2 /
(to see that E is a sheaf, one can use, for example, [19]), Lemma 2.5. From the fact
that Ê .Coh.X1 // Coh.X2 / it is easy to deduce that F Š Ê jCoh.X1 / Š ‰E .
X1 !X2
L In order to demonstrate that ‰ is fully faithful, denoting by the sheaf
m0 .p2 / .p1 OX1 .mH // of graded algebras on X2 (H being an ample line bundle
on X1 ), we will use the relative version of the Serre correspondence between graded
-modules and sheaves on Proj Š X1 X2 . More precisely, denoting by gmod-
the category of graded -modules of finite type (meaning finitely generated in suf-
ficiently high degrees), one considers the associated sheaf functor H W gmod- !
Coh.X1 L X2 / and the functor G W Coh.X1 X2 / ! gmod- defined on objects by
G.E/ WD m2Z ‰E .OX1 .mH //. They satisfy H B G Š id and, moreover, an object
or a morphism of gmod- is sent to 0 by H if and only if it is 0 in sufficiently high
degrees.
Now, given E1 ; E2 2 K.X1 ; X2 /, morphisms in gmod- from G.E1 / to G.E2 / can
be identified with natural transformations from ‰E1 jC to ‰E2 jC , where C is the full
subcategory of Coh.X1 / with objects fOX1 .iH /gi2Z . By [17], Proposition 3.6 (applied
to the functors Ê1 and Ê2 ) such natural transformations correspond bijectively to
natural transformations from ‰E1 to ‰E2 . Therefore, in view of the properties of G
X1 !X2
and H mentioned above, the fully faithfulness of ‰ amounts to the following:
if ˛ W ‰E1 ! ‰E2 is a natural transformation such that ˛m WD ˛.OX1 .mH // D 0 for
m 0, then ˛m D 0 for every m 2 Z. Clearly to this purpose it is enough to show that
˛m D 0 implies ˛m1 D 0. To see this, take a monomorphism f W OX1 ..m 1/H / ,!
OX1 .mH / and just observe that in the commutative diagram
‰E .f /
‰E1 .OX1 ..m 1/H // 1
/ ‰E .OX1 .mH //
1
˛m1 ˛m D0
‰E .f /
‰E2 .OX1 ..m 1/H // 2
/ ‰E .OX1 .mH //
2
‰E2 .f / is a monomorphism, because ‰E2 is exact.
X1 !X2
In particular, this shows that for the functor ‰ questions (Q1)–(Q4) can be
answered positively. As for (Q5), notice that in general K.X1 ; X2 / is an additive but
not an abelian subcategory of Coh.X1 X2 /.
5.4 The supported case. In this section we want to show how Theorem 5.3 can
be extended both considering a more general categorical setting and weakening the
assumptions on the exact functor.
Indeed, let X be a separated scheme of finite type over k and let Z be a subscheme
of X which is proper over k. We denote by DZ .Qcoh.X // the derived category of
unbounded complexes of quasi-coherent sheaves on X with cohomologies supported
on Z. Using this, we can define the triangulated categories
DbZ .Qcoh.X // WD DZ .Qcoh.X // \ Db .Qcoh.X //

DbZ .X / WD DZ .Qcoh.X // \ Db .X /:
We also set
Perf Z .X / WD DZ .Qcoh.X // \ Perf .X /:
Example 5.16. These categories appear naturally studying the so called open Calabi–
Yau’s. Examples of them are local resolutions of An -singularities ([26], [27]) and
the total space tot.!P 2 / of the canonical bundle of P 2 ([5]). In the latter case, if
Z denotes the zero section of the projection tot.!P 2 / ! P 2 , the derived category
Perf Z .tot.!P 2 // D DbZ .tot.!P 2 // is a Calabi–Yau category of dimension 3 and may
be seen as an interesting example to test predictions about Mirror Symmetry and the
topology of the space of stability conditions according to Bridgeland’s definition (see
[5] for results in this direction). Moreover, as a consequence of [26], [27], all autoequiv-
alences of the supported derived categories of An -singularities are of Fourier–Mukai
type and the group of such autoequivalences can be explicitly described. See [17] for
more details.
The category DZ .Qcoh.X // is a full subcategory of D.Qcoh.X // and let
W DZ .Qcoh.X // ! D.Qcoh.X //
be the inclusion. This functor has a right adjoint
Š W D.Qcoh.X // ! DZ .Qcoh.X //; Š .E/ WD colim RHom.OnZ ; E/;

!n
where nZ is the n-th infinitesimal neighborhood of Z in X (see [35], Proposition 3.2.2).

Due to [35], Corollary 3.1.4, the functor Š sends bounded complexes to bounded
complexes and Š B Š id.
Now, let X1 and X2 be separated schemes of finite type over k containing, respec-
tively, two subschemes Z1 and Z2 which are proper over k. The following generalizes
the standard definition of Fourier–Mukai functor.
Definition 5.17. An exact functor
F W DZ1 .Qcoh.X1 // ! DZ2 .Qcoh.X2 //
is a Fourier–Mukai functor if there exists E 2 DZ1 Z2 .Qcoh.X1 X2 // and an

isomorphism of exact functors
F Š ˆsE WD Š .p2 / .. /E ˝ p1 ../// (5.4)
where pi W X1 X2 ! Xi is the projection.

An analogous definition can be given for functors defined between bounded derived
categories of quasi-coherent, coherent or perfect complexes. As always, the object E
is called Fourier–Mukai kernel. It should be noted that, contrary to the smooth non-
supported case, the Fourier–Mukai kernel cannot be assumed to be a bounded coherent
complex. This is clarified by the following example dealing with the identity functor.
Example 5.18. We want to show that a Fourier–Mukai kernel of the identity functor
id W DbZ .X/ ! DbZ .X / is
. /Š 2 DbZZ .Qcoh.X X //;

where, denoting by W X ! X X the diagonal embedding,
WD B B Š .OX /:
Indeed, according to [17], we have the following isomorphisms:
Hom.A; Š ˆ .B// Š Hom.A; .p2 / . B B Š .OX / ˝ p1 .B///
Š Hom..B/_ A; B B Š .OX //
Š Hom..B/_ ˝ A; OX /
Š Hom.A; Š B/ Š Hom.A; B/;
for any A; B 2 DbZ .X /. Here pi W X X ! X is the natural projection. For the
first and the fourth isomorphism we used the adjunction between and Š . The same
adjunction together with the one between and and the fact that is fully faithful
and .B/_ ˝ A has support in Z explains the third isomorphism.
Obviously . /Š does not belong to DbZZ .X X /. Suppose that there exists
E 2 DbZZ .X X / such that
ˆsE Š id W DbZ .X / ! DbZ .X /:

By Lemma 7.41 in [49], there exist n > 0 and En 2 Db .nZ nZ/ such that . /E Š
.in in / En , where in W nZ ! X is the embedding. For any Fn 2 Db .nZ/, we have
.in / Fn Š Ê ..in / Fn / Š .in / Ên ..in / .in / Fn /: (5.5)
Take now X D P k , Z D P k1 and Fn WD OnZ .m/, for m 2 Z. An easy

calculation shows that .in / .in / Fn Š OnZ .m/ ˚ OnZ .m n/Œ1. Hence to have
(5.5) verified, we should have either Ên .OnZ .m// D 0 or Ên .OnZ .m n// D 0.
But the following isomorphisms should hold at the same time
Ên .OnZ .m// ˚ Ên .OnZ .m n//Œ1 Š OnZ .m/;
Ên .OnZ .m C n// ˚ Ên .OnZ .m//Œ1 Š OnZ .m C n/:
If Ên .OnZ .m n// D 0, then from the second one we would have that OnZ .m/Œ1
is a direct summand of OnZ .m C n/ which is absurd. Thus Ên .OnZ .m// D 0. As
this holds for all m 2 Z, we get a contradiction.
Now let X1 be a quasi-projective scheme containing a projective subscheme Z1

such that OiZ1 2 Perf .X1 /, for all i > 0, and let X2 be a separated scheme of finite
type over k with a subscheme Z2 which is proper over k.
Remark 5.19. Notice that under these assumptions, and having fixed an ample divisor
H1 on X1 , the objects OjijZ1 .jH1 / are in Perf Z1 .X1 /, for all i; j 2 Z. Special cases
in which OiZ1 2 Perf .X1 / are when X1 D Z1 or X1 is smooth.
One can consider exact functors F W Perf Z1 .X1 / ! Perf Z2 .X2 / such that
(1) Hom.F.A/; F.B/Œk/ D 0, for any A; B 2 CohZ1 .X1 /\Perf Z1 .X1 / and
any integer k < 0;
./ (2) for all A 2 Perf Z1 .X1 / with trivial cohomologies in (strictly) positive
degrees, there is N 2 Z such that
Hom.F.A/; F.OjijZ1 .jH1 /// D 0;
for any i < N and any j
i , where H1 is an ample divisor on X1 .
Then we have the following.
Theorem 5.20 ([17], Theorem 1.1). Let X1 , X2 , Z1 and Z2 be as above and let
F W Perf Z1 .X1 / ! Perf Z2 .X2 /
be an exact functor.
If F satisfies ./, then there exist E 2 DbZ1 Z2 .Qcoh.X1 X2 // and an isomor-
phism of exact functors F Š ˆsE . Moreover, if Xi is smooth quasi-projective, for
i D 1; 2, and k is perfect, then E is unique up to isomorphism.
Back to Remark 5.19, the above theorem can be applied in at least two interesting
geometric contexts. If X1 D Z1 , then we get back (a generalization of) Theorem 5.3.
On the other hand, if X1 is smooth, then we can apply the above result to the au-
toequivalences of the categories described in Example 5.16 proving that they are all
of Fourier–Mukai type. As noticed in [17], if Xi D Zi , dim.X1 / > 0 and they are
smooth, then ./ is equivalent to (3.2). Thus, Theorem 5.20 recovers Theorem 3.9 as
well.
Remark 5.21. In the same vein as in [36], it is proved in [17], Theorem 1.2, that
Perf Z .X/ has a (strongly) unique dg-enhancement if X and Z have the same properties
as X1 and Z1 in Theorem 5.20 and T0 .OZ / D 0. See [36] for the definition of strongly
unique dg-enhancement which is not needed here.
6 More open problems

The list of problems mentioned in the above sections can be extended further. The main
sources are actually very concrete geometric settings where they appear naturally. We
try to list some of them below, although a complete clarification of their geometric
meaning goes far beyond the scope of this paper.
6.1 Does full imply essentially surjective? In Section 3.4.2 we have seen that a
full functor between the bounded derived categories of coherent sheaves on smooth
projective varieties is automatically faithful. Assume now that we are given an exact
endofunctor F W Db .X / ! Db .X /, where X is again a smooth projective variety. In
this section we want to discuss the following.
Conjecture 6.1. If F is full, then it is an autoequivalence.
Notice that we only need to show that F is essentially surjective.
Remark 6.2. The conjecture is true if !X is trivial, because in that case every fully
faithful exact endofunctor of Db .X / is an equivalence (see, for example, Corollary 7.8
in [23]).
The above conjecture is implied by another conjecture about admissible subcate-
gories that we want to explain here.
Given a triangulated category T and a strictly full triangulated subcategory S, we
say that S is left- (resp. right-) admissible in T if the inclusion functor W S ! T has
a left (resp. right) adjoint W T ! S (resp. Š W T ! S). If a subcategory is left and
right admissible, we say that it is admissible.
Remark 6.3. By [7], Proposition 1.6, an admissible subcategory S T is thick as
well.
We can use the notion of admissible subcategory to ‘decompose’ triangulated cat-
egories. More generally, one can give the following.
Definition 6.4. A semi-orthogonal decomposition of a triangulated category T is
given by a sequence of full triangulated subcategories A1 ; : : : ; An T such that
HomT .Ai ; Aj / D 0, for i > j and, for all K 2 T, there exists a chain of morphisms
in T
0 D Kn ! Kn1 ! ! K1 ! K0 D K
with Cone.Ki ! Ki1 / 2 Ai , for all i D 1; : : : ; n. We will denote such a decompo-
sition by T D hA1 ; : : : ; An i.
The easiest examples of semi-orthogonal decompositions are constructed via ex-
ceptional objects.
Definition 6.5. Assume that T is a k-linear triangulated category. An object E 2 T
is called exceptional if HomT .E; E/ Š k and HomT .E; EŒp/ D 0, for all p ¤ 0. A
sequence .E1 ; : : : ; Em / of objects in T is called an exceptional sequence if Ei is an
exceptional object, for all i , and HomT .Ei ; Ej Œp/ D 0, for all p and all i > j . An
exceptional sequence is full if it generates T.
Remark 6.6. If .E1 ; : : : ; Em / is a full exceptional sequence in T, then we get a semi-
orthogonal decomposition T D hE1 ; : : : ; En i, where for simplicity we write Ei for
the triangulated subcategory generated by Ei , which is equivalent to Db .Spec k/ and
is admissible in T.
Example 6.7. A celebrated result of Beilinson shows that Db .P n / has a full exceptional
sequence .OP n .n/; OP n .n C 1/; : : : ; OP n / (see, for example, [23]), Section 8.3.
For a triangulated subcategory S of a triangulated category T, we can define the
strictly full triangulated subcategories (i.e. full and closed under isomorphism)
S? WD fA 2 T W Hom.S; A/ D 0; for all S 2 Sg
called right orthogonal to S and its left orthogonal
?
S WD fA 2 T W Hom.A; S / D 0; for all S 2 Sg :
One can formulate the following conjecture due to A. Kuznetsov and contained in
[34].
Conjecture 6.8 (Noetherianity conjecture). Let X be a smooth projective variety and
assume that there exists a sequence
A1 A2 Ai Db .X /
of admissible subcategories. Then there is a positive integer N such that Ai D AN ,
for all i N .
Remark 6.9. Considering the strictly full triangulated subcategories Bi WD A? i , the
above conjecture can be equivalently reformulated in terms of stabilizing descending
chains.
Proposition 6.10. Conjecture 6.8 implies Conjecture 6.1.
Proof. The functor F is automatically faithful. Thus
I WD im F WD fE 2 Db .X / W E Š F.F / for some F 2 Db .X /g
is a strictly full triangulated subcategory of Db .X /. By Proposition 3.5, the functor F
has left and right adjoints and so I is admissible. Using the above notation, set J D I? .
Hence we have a semi-orthogonal decomposition
Db .X / D hJ; Ii:
As I Š Db .X /, we can think of F as an exact endofunctor of I. Hence, reasoning
as above we get a semi-orthogonal decomposition
Db .X / D hJ; J; Ii
Hence, given a positive integer n, repeating this argument n times we get that
An WD hJ; : : : ; Ji
„ ƒ‚ …
n times
is a strictly full admissible triangulated subcategory of Db .X /.

Since Ap Aq Db .X / whenever p q, by Conjecture 6.8, this sequence must
stabilize. Hence J D 0 and so F is essentially surjective.
Due to the following easy result, a full endofunctor is automatically an equivalence

when X has dimension at most 1.
Proposition 6.11. Conjecture 6.8 holds true when X is a smooth projective variety of
dimension smaller or equal to 1.
Proof. Obviously the conjecture is trivially true if Db .X / does not admit a non-trivial
semi-orthogonal decomposition and this is the case if dim.X / D 0.
If X is a curve of genus 1, Serre duality and Example 3.2 in [11] implies that Db .X /
cannot be decomposed. The same is true when X is a curve of genus g 2 due to
[44]. Thus the only case that has to be checked is X Š P 1 .
For this assume that Db .P 1 / D hA1 ; A2 i, where Ai is not trivial, for i D 1; 2 (i.e.
non-zero and not the whole category Db .P 1 /). It is clear that either A1 or A2 must
contain a locally free sheaf E. As on P 1 any locally free sheaf is the direct sum of line
bundles and Ai is thick (see Remark 6.3 (i)), there is j 2 Z such that OP 1 .j / 2 Ai ,
for i D 1 or i D 2. We assume i D 1 as the argument in the other case is similar.
Now A2 D ? A1 ? hOP 1 .j /i D hOP 1 .j C1/i. But hOP 1 .j C1/i Š Db .Spec k/
and so it does not contain proper thick subcategories. Thus A2 D hOP 1 .j C 1/i and
A1 D hOP 1 .j /i. Therefore, there cannot be non-stabilizing ascending chains of
admissible subcategories.
6.2 Splitting functors. Kuznetsov introduced in [33] the notion of splitting functor
as a natural generalization of fully faithful functor. The expectation was that, in this
context, one should get a representability result similar to Theorem 3.1. Let us clarify
the situation a bit more.
More precisely, given two triangulated categories T1 and T2 and an exact functor
F W T1 ! T2 , we can define the following full subcategories
ker F WD fA 2 T1 W F.A/ Š 0g; im F WD fA Š F.B/ W B 2 T1 g:
Remark 6.12. The subcategory ker F is always triangulated while im F, in general, is

not. It becomes triangulated if F is fully faithful.
Hence we can give the following.
Definition 6.13. An exact functor F W T1 ! T2 is right (respectively left) splitting
if ker F is a right (respectively left) admissible subcategory in T1 , the restriction of
F to .ker F/? (respectively ? .ker F/) is fully faithful, and the category im F is right
(respectively left) admissible in T2 .
An exact functor is splitting if it is both right and left splitting.
Remark 6.14. As observed in Lemma 3.2 of [33], a right (respectively left) splitting
functor F has a right (respectively left) adjoint functor FŠ (respectively F ).
We summarize the basic properties of these functors in the following.
Theorem 6.15 ([33], Theorem 3.3). Let F W T1 ! T2 be an exact functor. Then the
following conditions are equivalent:
(i) F is right splitting;
(ii) F has a right adjoint functor FŠ and the composition of the canonical morphism
of functors idT1 ! FŠ B F with F gives an isomorphism F Š F B FŠ B F;
(iii) F has a right adjoint functor FŠ , there are semi-orthogonal decompositions
T1 D him FŠ ; ker Fi; T2 D hker FŠ ; im Fi;
and the functors F and FŠ give quasi-inverse equivalences im FŠ Š im F;
(iv) There exists a triangulated category S and fully faithful functors G1 W S ! T1 ,
G2 W S ! T2 , such that G1 admits a left adjoint G1 , G2 admits a right adjoint
and F Š G2 B G1 .
Clearly, one can formulate analogous conditions for left splitting functors. The
main conjecture is now the following:
Conjecture 6.16 ([33], Conjecture 3.7). Let X1 and X2 be smooth projective varieties.
Then any exact splitting functor F W Db .X1 / ! Db .X2 / is of Fourier–Mukai type.
One may first wonder why the strategy outlined in Section 5.2 may not be applied
in this case. The main problem is that convolutions do not work for this kind of
functors. Alternatively, one would need to define an analogue of the ample sequence
in Section 5.2.2 for the subcategory S in part (iv) of Theorem 6.15. Hence, the solution
to Conjecture 6.16 is closely related to Problems 5.12 and 5.13.
Nevertheless, there are several instances in which the conjecture is verified. The
easiest one is when the category S mentioned in Theorem 6.15 (iv) is such that S Š
Db .Y /, for some smooth projective variety Y . Indeed, in this case, one reduces the
proof to Theorem 3.1 (using Proposition 2.2).
Moreover, it is not difficult to observe that, using the same type of arguments as in
the proof of Proposition 6.11, one can show the following (the zero-dimensional case
is trivial).
Proposition 6.17. Let either X1 or X2 be a smooth projective curve. Then any splitting
functor F W Db .X1 / ! Db .X2 / is of Fourier–Mukai type.
For less trivial situations where Conjecture 6.16 can be verified, one has to refer
to [32]. For this consider a full admissible subcategory W S ,! Db .X /, for a smooth
projective variety X. Thus we get the left and right adjoints W Db .X / ! S and
Š W Db .X/ ! S.
Take now the functors F1 WD B Š W Db .X / ! Db .X / and F2 WD B W Db .X / !
D .X/. It is not difficult to see (using, for example, Theorem 6.15 above) that F1 and
b
F2 are splitting functors. A non-trivial argument allows one to prove the following:
Theorem 6.18 ([32], Theorem 7.1). The functors F1 and F2 are of Fourier–Mukai type.
6.3 Relative Fourier–Mukai functors. In [33], Kuznetsov drove the attention to a

slightly more general version of the classical Fourier–Mukai functors. For sake of
simplicity, take a pair of smooth projective varieties X1 and X2 over the same smooth
projective variety S. To fix the notation, this means that, for i D 1; 2, there is a
morphism fi W Xi ! S. Clearly, one may want to relax the assumptions on Xi and S
but this is not in order here.
Definition 6.19. (i) A functor F W Db .X1 / ! Db .X2 / is S -linear if
F.A ˝ f1 .C // Š F.A/ ˝ f2 .C/;
for all A 2 Db .X1 / and for all C 2 Db .S /.

(ii) A strictly full subcategory S Db .Xi / is S -linear if for all C 2 S and all
A 2 Db .S/ we have fi .A/ ˝ C 2 S.
These functors have reasonable properties listed in the following proposition and
proved in [33] (see, in particular, Section 2.7 there).
Proposition 6.20. (i) If F is exact, S-linear and admits a right adjoint functor FŠ , then
FŠ is also S -linear.
(ii) If S Db .Xi / is a strictly full admissible S -linear subcategory, then its right
and left orthogonals are S-linear as well.
As pointed out in, for example, [32], [33], the relative functors play important roles
in various geometric situations. Thus it makes perfect sense to wonder whether the
machinery developed for Fourier–Mukai functors in the non-relative setting can be
applied.
It is clear that any full exact S-linear functor or rather any exact S -linear functor
F W Db .X1 / ! Db .X2 / satisfying (3.2) is of Fourier–Mukai type in view of Theo-
rem 3.9. In particular, there is a unique (up to isomorphism) E 2 Db .X1 X2 / and an
isomorphism F Š Ê .
On the other hand, we may consider the fibre product X1 S X2 and the closed
embedding i W X1 S X2 ,! X1 X2 .
Lemma 6.21 ([33], Lemma 2.32). If E 2 Db .X1 S X2 /, then the Fourier–Mukai
functor î E is S -linear.
It is not difficult to observe that the Fourier–Mukai kernel of an S -linear Fourier–
Mukai functor has to be set theoretically supported on the fibre product X1 S X2 .
The scheme theoretical point of view is more complicated to be dealt with and thus,
following [33], it makes sense to pose the following questions:
Question 6.22. (i) Given a full exact S-linear functor F W Db .X1 / ! Db .X2 /, do there
exist an E 2 Db .X1 S X2 / and an isomorphism of functors F Š î E ?
(ii) Is the choice of the Fourier–Mukai kernel E 2 Db .X1 S X2 / in (i) unique (up
to isomorphism)?
To our knowledge, no general answer to these problems is present in the literature.
Acknowledgements. The write-up of this paper started when P.S. was visiting the
University of Bonn which we thank for the warm hospitality and for the financial
support. The second author is also grateful to the organizers of the GCOE Conference
“Derived Categories 2011 Tokyo”, Y. Kawamata and Y. Toda, for the very stimulating
mathematical atmosphere during the conference. We are also grateful to Pawel Sosna
for comments on an early version of this paper. Pierre Schapira informed us about the
paper [50] and the notion of Fourier–Sato transform. David Ben-Zvi kindly brought
our attention to the results in [6] and [48]. We warmly thank both of them.
References
[1] B. Andreas and D. Hernández Ruipérez, Fourier Mukai transforms and applications to
string theory. Rev. R. Acad. Cienc. Exactas Fis. Nat. Ser. A Mat. 99 (2005), 29–77.
[2] M. Anel and B. Toën, Dénombrabilité des classes d’équivalences dérivées des variétés
algébriques. J. Algebraic Geom. 18 (2009), 257–277.
[3] D. Arinkin, J. Block, and T. Pantev, ?-Quantizations of Fourier–Mukai transforms. Preprint,
[4] M. R. Ballard, Equivalences of derived categories of sheaves on quasi-projective schemes.
[5] A. Bayer and E. Macrì, The space of stability conditions on the local projective plane. Duke
Math. J. 160 (2011), 263–322.
[6] D. Ben-Zvi, J. Francis, and D. Nadler, Integral transforms and Drinfeld centers in derived
algebraic geometry. J. Amer. Math. Soc. 23 (2010), 909–966.
[7] A. Bondal and M. Kapranov, Representable functors, Serre functors, and mutations. Math.
USSR-Izv. 35 (1990), 519–541.
[8] A. Bondal, M. Larsen, and V. Lunts, Grothendieck ring of pretriangulated categories. In-
ternat. Math. Res. Notices 29 (2004), 1461–1495.
[9] A. Bondal and D. Orlov, Semiorthogonal decomposition for algebraic varieties. Preprint,
arXiv:alg-geom/9506012.
[10] A. Bondal and M. Van den Bergh, Generators and representability of functors in commu-
tative and noncommutative geometry. Moscow Math. J. 3 (2003), 1–36.
[11] T. Bridgeland, Equivalences of triangulated categories and Fourier–Mukai transforms. Bull.
London Math. Soc. 31 (1999), 25–34.
[12] T. Bridgeland and A. Maciocia, Complex surfaces with equivalent derived categories. Math.
Z. 236 (2001), 677–697.
[13] A. Căldăraru, Derived categories of sheaves: a skimming. In Snowbird lectures in algebraic
geometry, Contemp. Math. 388, Amer. Math. Soc., Providence, RI, 2005, 43–75.
[14] A. Căldăraru, The Mukai pairing II: The Hochschild–Kostant–Rosenberg isomorphism.
Adv. Math. 194 (2005), 34–66.
[15] A. Căldăraru and S. Willerton, The Mukai pairing. I. A categorical approach. New York J.
Math. 16 (2010), 61–98.
[16] A. Canonaco, D. Orlov, and P. Stellari, Does full imply faithful? J. Noncommut. Geom. 7
(2013), no. 1, to appear; preprint, arXiv:1101.5931 [math.AG].
[17] A. Canonaco and P. Stellari, Fourier–Mukai functors in the supported case. Preprint,
[18] A. Canonaco and P. Stellari, Non-uniqueness of Fourier–Mukai kernels. Math. Z. 272
(2012), 577–588.
[19] A. Canonaco and P. Stellari, Twisted Fourier–Mukai functors. Adv. Math. 212 (2007),
484–503.
[20] V. Drinfeld, DG quotients of DG categories. J. Algebra, 272 (2004), 643–691.
[21] L. Hille and M. Van den Bergh, Fourier-Mukai transforms. In Handbook of tilting theory,
London Math. Soc. Lecture Note Ser. 332, Cambridge University Press, Cambridge 2007,
147–173.
[22] S. Hosono, B.H. Lian, K. Oguiso, and S.-T. Yau, Autoequivalences of derived category of
a K3 surface and monodromy transformations. J. Alg. Geom. 13 (2004), 513–545.
[23] D. Huybrechts, Fourier–Mukai transforms in algebraic geometry. Oxford Math. Monogr.,
Oxford Science Publications, Oxford 2006.
[24] D. Huybrechts, E. Macrì, and P. Stellari, Derived equivalences of K3 surfaces and orienta-
tion. Duke Math. J. 149 (2009), 461–507.
[25] D. Huybrechts and P. Stellari, Equivalences of twisted K3 surfaces. Math. Ann. 332 (2005),
901–936.
[26] A. Ishii and H. Uehara, Autoequivalences of derived categories on the minimal resolutions
of An -singularities on surfaces. J. Differential Geom. 71 (2005), 385–435.
[27] A. Ishii, K. Ueda, and H. Uehara, Stability conditions on An -singularities. J. Differential
Geom. 84 (2010), 87–126.
[28] M. Kashiwara and P. Schapira, Sheaves on manifolds. Grundlehren Math. Wiss. 292,
Springer-Verlag, Berlin 2004.
[29] Y. Kawamata, D-equivalence and K-equivalence. J. Differential Geom. 61 (2002), 147–171.
[30] Y. Kawamata, Equivalences of derived categories of sheaves on smooth stacks. Amer. J.
Math. 126 (2004), 1057–1083.
[31] B. Keller, On differential graded categories. In International Congress of Mathematicians,
Vol. II, Eur. Math. Soc., Zürich 2006, 151–190.
[32] A. Kuznetsov, Base change for semiorthogonal decompositions. Compositio Math. 147
(2011), 852–876.
[33] A. Kuznetsov, Homological projective duality. Publ. Math. Inst. Hautes Études Sci. 105
(2007), 157–220.
[34] A. Kuznetsov, Lefschetz decompositions and categorical resolutions of singularities. Se-
lecta Math. (N.S.) 13 (2008), 661–696.
[35] J. Lipman, Lectures on local cohomology and duality. In Local cohomology and its appli-
cations, Lecture Notes in Pure and Appl. Math. 226, Marcel Dekker, NY, 2001, 39–89.
[36] V. Lunts and D. Orlov, Uniqueness of enhancements for triangulated categories. J. Amer.
Math. Soc. 23 (2010), 853–908.
[37] E. Macrì and P. Stellari, Automorphisms and autoequivalences of generic analytic K3
surfaces. J. Geom. Phys. 58 (2008), 133–164.
[38] E. Macrì and P. Stellari, Infinitesimal derived Torelli theorem for K3 surfaces (Appendix
by S. Mehrotra). Internat. Math. Res. Notices 2009 (2009), 3190–3220.
[39] N. Markarian, Poincaré–Birkhoff–Witt isomorphism, Hochschild homology and Riemann–
Roch theorem. Preprint MPI 2001-52.
[40] S. Mukai, Duality between D.X / and D.Xy / with its applications to Picard sheaves. Nagoya
Math. J. 81 (1981), 153–175.
[41] S. Mukai, On the moduli space of bundles on K3 surfaces, I. In Vector bundles on algebraic
varieties, Tata Inst. Fund. Res. Studies in Math. 11, Tata Institute of Fundamental Research,
Bombay; Oxford University Press, New York 1987, 341–413.
[42] K. Oguiso, Bounded derived categories of very simple manifolds. Sci. China Math. 54
(2011), no. 8, 1741–1750.
[43] K. Oguiso, K3 surfaces via almost-primes. Math. Res. Lett. 9 (2002), 47–63.
[44] S. Okawa, Semi-orthogonal decomposability of the derived category of a curve. Adv. Math.
228 (2011), 2869–2873.
[45] D. Orlov, Derived categories of coherent sheaves on abelian varieties and equivalences
between them. Izv. Math. 66 (2002), 569–594.
[46] D. Orlov, Equivalences of derived categories and K3 surfaces. J. Math. Sci. 84 (1997),
1361–1381.
[47] D. Ploog, Group of autoequivalences of derived categories of smooth projectiove varieties.
PhD thesis, Berlin, 2004.
[48] A. Preygel, Thom–Sebastiani & duality for matrix factorizations. Preprint, arXiv:1101.5834
[math.AG].
[49] R. Rouquier, Dimensions of triangulated categories. J. K-theory 1 (2008), 193–258.
[50] M. Sato, T. Kawai, and M. Kashiwara, Microfunctions and pseudo-differential equations.
In Hyperfunctions and pseudo-differential equations (Proc. Conf., Katata, 1971; dedicated
to the memory of André Martineau), Lecture Notes in Math. 287, Springer-Verlag, Berlin
1973, 265–529.
[51] P. Stellari, Some remarks about the FM-partners of K3 surfaces with Picard number 1 and
2. Geom. Dedicata 108 (2004), 1–13.
[52] B. Toën, The homotopy theory of dg-categories and derived Morita theory. Invent. Math.
167 (2007), 615–667.
[53] M. Verbitsky, Coherent sheaves on general K3 surfaces and tori. Pure Appl. Math. Q. 4
(2008), 651–714.
[54] M. Verbitsky, Coherent sheaves on generic compact tori. In Algebraic structures and moduli
spaces, CRM Proc. Lecture Notes 38, Amer. Math. Soc., Providence, RI, 2004, 229–247.
Flops and about: a guide
Sabin Cautis
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2 Stratified flops of type A – the local model . . . . . . . . . . . . . . . . . . 63
2.1 Cotangent bundles to Grassmannians . . . . . . . . . . . . . . . . . . 63
2.2 Deformations of cotangent bundles . . . . . . . . . . . . . . . . . . . 64
3 Geometric categorical sl2 actions . . . . . . . . . . . . . . . . . . . . . . 65
3.1 Preliminary concepts . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.3 Some remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.4 Inducing equivalences . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4 Equivalences for the local model
L of stratified flops . . . . . . . . . . . . . . 70
4.1 Categorical actions on k D.T ? G.k; N // . . . . . . . . . . . . . . 70
4.2 The equivalence: an explicit description . . . . . . . . . . . . . . . . 72
4.3 The inverse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.4 The equivalence: stratified Atiyah flops . . . . . . . . . . . . . . . . 74
4.5 Equivalences and K-theory . . . . . . . . . . . . . . . . . . . . . . . 77
5 Geometric categorical slm actions . . . . . . . . . . . . . . . . . . . . . . 78
5.1 Some remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2 Braid group actions . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6 Twists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.1 Seidel–Thomas (spherical) twists . . . . . . . . . . . . . . . . . . . . 85
6.2 P n -twists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.3 Infinite twists and some geometry . . . . . . . . . . . . . . . . . . . 89
7 The general flop – a discussion . . . . . . . . . . . . . . . . . . . . . . . . 92
7.1 The Mukai flop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
7.2 The stratified Mukai flop of type A . . . . . . . . . . . . . . . . . . . 93
7.3 The stratified Mukai flop of type D . . . . . . . . . . . . . . . . . . . 94
7.4 Equivalences in type D . . . . . . . . . . . . . . . . . . . . . . . . . 97
8 Further topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

The author is thankful for the support received through NSF grant DMS-1101439 and the Alfred P.
Sloan foundation.
62 S. Cautis
1 Introduction
The stratified Mukai flop is an algebro-geometric construction relating two birational
varieties. There are three basic types of Mukai flops (A, D and E6;I/E6;II) named after
the Lie algebra to which they are related. Namikawa coined these terms in [Nam3]
where he showed that any two Springer resolutions of a nilpotent orbit closure are
connected by a series of such flops (this result also appears in [F]). There are also
deformations of these Mukai flops which we call Atiyah flops.
The local model of a Mukai flop of type A is a correspondence which relates
the cotangent bundles T ? G.k; N / and T ? G.N k; N / of Grassmannians. These
cotangent bundles are examples of Springer resolutions of the closure of nilpotent
orbits. In this case the orbit is that of a matrix X 2 GLN where X 2 D 0 and the rank
of X is min.k; N k/.
Stratified flops also control the birational geometry of moduli spaces of sheaves on
surfaces. In [M] Markman examined Brill–Noether type stratifications of the moduli
spaces of sheaves on a fixed K3 surface. He showed that two moduli spaces with Mukai
vectors related by certain involutions of the Mukai lattice are related by stratified Mukai
flops.
A basic question is when two birational varieties X and Y have isomorphic de-
rived categories of coherent sheaves D.X / and D.Y /. If D.X / Š D.Y / then we say
that X and Y are derived equivalent. One general conjecture implies that two vari-
eties related by a stratified flop are derived equivalent. In particular, T ? G.k; N / and
T ? G.N k; N / should be derived equivalent.
Constructing this equivalence is the motivating problem discussed in this paper. Of
course, T ? G.k; N / and T ? G.N k; N / are actually isomorphic, but this isomorphism
is not canonical. The derived equivalence described in Section 4 is canonical in the
sense that it also works in families. This means that for any vector bundle W over
some base one may consider the relative cotangent bundles T ? G.k; W / and T ? G.N
k; W /. These might not be isomorphic (for the same reason W and W _ might not be
isomorphic) but, nevertheless, are derived equivalent.
The case k D 1 goes back a few years to the work of Kawamata [K1] and Nami-
kawa [Nam1] who constructed equivalences D.T ? G.1; N // ! D.T ? G.N 1; N //.
Kawamata [K2] was also able to work out the case k D 2 and N D 4 and conjecture
explicit equivalences when k D 2 and N > 4.
To deal with arbitrary k and N we adopted a technique used by Chuang and Rouquier
[CR] in modular representation theory. The idea [CKL1] is to construct a categori-
cal sl2 action (defined in Section 3) on the union of all D.T ? G.k; N // where N is
fixed and k D 0; : : : ; N . Recall that given an sl2 representation one can construct an
action of its Weyl group which induces an isomorphism of weight spaces. Likewise,
for a categorical sl2 action one can construct an action of its braid group. In this
case of cotangent bundles to Grassmannians this braid group induces natural equiva-
lences D.T ? G.k; N // ! D.T ? G.N k; N // [CKL2], [CKL3]. In a sense, these
Flops and about: a guide 63
equivalences unify Seidel–Thomas twists [ST] and P n -twists [HT] into a more general
concept.
Categorical sl2 actions have a counterpart for any Kac–Moody Lie algebra g. When
g D sln one can define a categorical sln action on cotangent bundles to n-step partial
flag varieties. This induces an action of the braid group on n strands [CK3] on the
derived categories of these varieties which generalizes work of Khovanov–Thomas
[KT], Riche [Ric] and Bezrukavnikov–Riche [BR].
Stratified flops also show up in the geometry of the affine Grassmannian. More
precisely, the twisted products Gr Q Gr and Gr Q Gr of orbits in the affine Grass-
mannian of P GLN are related by stratified Mukai flops (see [CK2], Section 1). One
can construct a geometric categorical sl2 actions here which subsequently induces an
equivalence D.Gr Q Gr /
! D.Gr Q Gr /.
More generally, one can define a geometric categorical sln action on categories of
the form D.Gr 1 Q
Q Gr n / where 1 ; : : : ; n are fundamental weights. This ac-
tion induces a braid group action on these categories. In [CK1], [CK2], [C2] these braid
group actions were used to construct homological knot invariants such as Khovanov
homology.
This paper is, for the most part, a survey of topics related to flops, categorical Lie
algebra actions and derived equivalences. It is an expanded version of the talk given
at the conference “Derived categories” organized by Yujiro Kawamata and Yukinobu
Toda in Tokyo in January 2011.
Acknowledgements. I began working in this area with the paper [CK1] jointly written
with Joel Kamnitzer. In it we give an algebro-geometric construction of Khovanov knot
homology. Trying, at first somewhat unsuccessfully, to generalize this construction led
us, over several years, in several tangential directions. I would like to thank Joel for
many years of truly fantastic ideas and insights which he shared with great enthusiasm
and without which I am convinced none of this would have been possible.
I would also like to thank Yujiro Kawamata and Yukinobu Toda for inviting me
and organizing a very interesting, inspirational and enjoyable conference in Tokyo in
January 2011.
2 Stratified flops of type A – the local model

Here we recall the definitions of stratified Mukai flops and stratified Atiyah flops of
type A.
2.1 Cotangent bundles to Grassmannians. The local model for a stratified Mukai
flop of type A is based on cotangent bundles to Grassmannians. The cotangent bundle
T ? G.k; N / can be described very explicitly as
k N k
f.X; V / W X 2 End.C N /; 0 !
V ! C N ; X C N V and X V 0g (1)
64 S. Cautis
where G.k; N / denotes the Grassmannian of k-planes in C N . The arrows denote

inclusions and the superscripts indicate the codimension of the inclusion.
We will suppose from now on that 2k N . These cotangent bundles come
equipped with the affinization map
p.k; N / W T ? G.k; N / ! B.k; N /
where
B.k; N / WD fX 2 End.C N / W X 2 D 0 and dim.ker X / D N kg
and p is the map which forgets V . This map is birational since generically the rank of
X is k and one can recover V as the image of X. Likewise there is a projection map
p.N k; N / W T ? G.N k; N / ! B.k; N /
which is also birational since one can recover V as the kernel of X . The triple
T G.k; N / T G.N k; N / (2)
MMM o
MMM
ooooo
MMM oo
p.k;N / M& wooo p.N k;N /
B.k; N /
is the local model for a stratified Mukai flop of type A. When k D 1 this is the usual
(and better known) Mukai flop.
2.2 Deformations of cotangent bundles. The cotangent bundles above have a natural
E
one-parameter deformation T ? G.k; N / over A1 . This deformation can be described
explicitly as
f.X; V; x/ W X 2 End.C N /; 0 V C N ; x 2 C;
(3)
dim.V / D k; XC N V and .X x id/V 0g
where the map to A1 remembers x. The fibre over x D 0 is clearly just T ? G.k; N /.
These varieties also come equipped with the affinization map
z N/
Q N / W T ? G.k; N / ! B.k;
p.k;
z N / is the variety
where B.k;
f.X; x/ W X 2 End.C N /; x 2 C; X.X x id/ D 0 and dim.ker.X x id// D kg
and p.k;
Q N / forgets V . The map is again birational. In fact, it is an isomorphism if
x ¤ 0 because one can recover X as the kernel of .X x id/. The diagram
E
T G.k; N /
RRR
H
T G.N k; N /
k
RRRp.k;N
QR / Q k;Nk/kkk
p.N
k
RRR k kk
RRR kk
) ukkk
z N / Š B.N
B.k; z k; N /
z N/ Š
is the local model for a stratified Atiyah flop of type A. The isomorphism B.k;
z
B.N k; N / is given by .X; x/ 7! .X x id; x/. When k D 1 and N D 2 this is
the usual Atiyah flop (hence the terminology) where both deformed cotangent bundles
turn out to be isomorphic to the total space of the vector bundle OP 1 .1/ ˚ OP 1 .1/
over P 1 .
2.2.1 C -actions. There are compatible C -actions on T ? G.k; N / and its deforma-
tion defined by
t .X; V / D .t 2 X; V / and t .X; V; x/ D .t 2 X; V; t 2 x/
respectively. Also, notice that both T ? G.k; N / and its deformation carry a tautological
bundle, denoted V , whose fibre over .X; V / (or .X; V; x/) is V .
3 Geometric categorical sl 2 actions

The main tool used to construct derived equivalences between flops is the notion of a
geometric categorical sl2 action. The idea of using categorical sl2 actions originates
with Chuang and Rouquier’s remarkable proof of Broué’s abelian defect group con-
jecture for symmetric groups [CR]. They construct and use such an action to define
equivalences between blocks of representations of the symmetric group in positive
characteristic. We adapt their approach to categories of coherent sheaves.
3.1 Preliminary concepts. All varieties are defined over C. If X is a variety we

denote by D.X/ the bounded derived category of coherent sheaves on X . As usual,
we denote by Œ1 the cohomological shift in D.X / downwards by 1.
3.1.1 Fourier–Mukai transforms. An object P 2 D.X Y / whose support is

proper over Y induces a Fourier–Mukai (FM) functor ˆP W D.X / ! D.Y / via ./ 7!
2 .1 ./ ˝ P / (where every operation is derived). One says that P is the FM
kernel which induces ˆP . The right and left adjoints ˆR P and ˆP are induced by
L
PR WD P _ ˝ 2 !X Œdim.X / and PL WD P _ ˝ 1 !Y Œdim.Y / respectively.

If Q 2 D.Y Z/ then ˆQ ˆP Š ˆQP W D.X / ! D.Z/ where Q P D

13 .12 P ˝ 23 Q/ is the convolution product. So instead of talking about functors
and compositions we will speak of kernels and convolutions.
3.1.2 C -equivariance. If X carries a C -action then we will consider the bounded

derived category of C -equivariant coherent sheaves on X which, abusing notation,
we also denote by D.X /. The sheaf OX fi g denotes the structure sheaf of X shifted
with respect to the C -action so that if f 2 OX .U / is a local function then viewed as
a section f 0 2 OX fi g.U / we have t f 0 D t i .t f /. We denote by fi g the operation
of tensoring with OX fi g.
66 S. Cautis
Since D.X/ carries a grading fg its Grothendieck group is actually a ZŒq; q 1 -
module where q acts by twisting by f1g. We usually tensor the Grothendieck group
with C so that it becomes a CŒq; q 1 -module and denote it K.X /.
3.1.3 Convolution of complexes. Now consider a complex

d d d d
P WD Pm
! Pm1
!
! P1
! P0
where Pi 2 D.X / and d 2 D 0. If m D 1 one can just take the cone and obtain an
object in D.X/. If m > 1 one would like to take an iterated cone. This is commonly
called a right or a left convolution of P depending on whether you start the iterated
cone from the right end or from the left end. Do not confuse this convolution with the
convolution of kernels described above!
In general a right convolution is not guaranteed to exist or to be unique. This is
because the Cone operation is not functorial. However, under the following cohomo-
logical conditions
Hom.PiCkC1 Œk; Pi / D 0 and Hom.PiCkC2 Œk; Pi / D 0 for i 0; k 1
a unique right convolution exists. For details see [CKL3], Section 3.4.
3.2 Definition. Let us recall the definition of a geometric categorical sl2 action from
[CKL1]. To shorten notation we will write H ? .P r / for the (doubly) graded vector
space
CŒrfrg ˚ CŒr 2fr C 2g ˚ ˚ CŒr C 2fr 2g ˚ CŒrfrg:
By convention H ? .P 1 / is zero.
A geometric categorical sl2 action consists of the following data.
(i) A collection of smooth complex varieties Y ./ indexed by 2 Z and equipped

with C -actions.
(ii) Fourier–Mukai kernels
E .r/ ./ 2 D.Y. r/ Y . C r// and F .r/ ./ 2 D.Y . C r/ Y . r//
(which are C equivariant). We write E./ for E .1/ ./ and F ./ for F .1/ ./
while E .0/ ./ and F .0/ ./ are equal to the identity kernels O .
(iii) For each Y ./ a flat deformation Yz ./ ! A1 carrying a C -action compatible
with the action x 7! t 2 x (where t 2 C ) on the base A1 .
On this data we impose the following additional conditions.

(i) Y./ D ; for 0 or

0. Moreover, each (graded piece of the) Hom
space between two objects in D.Y .// is finite dimensional. In particular, this
means that if Y ./ ¤ ; then End.OY./ / D C I .
(ii) All E .r/ s and F .r/ s are sheaves (i.e. complexes supported in degree zero).
(iii) E .r/ ./ and F .r/ ./ are left and right adjoints of each other up to shift. More
precisely,
(a) E .r/ ./R D F .r/ ./Œrfrg and F .r/ ./L D E .r/ ./Œrfrg,
(b) E .r/ ./L D F .r/ ./Œrfrg and F .r/ ./R D E .r/ ./Œrfrg.
(iv) At the level of cohomology of complexes we have
H .E. C r/ E .r/ . 1// Š E .rC1/ ./ ˝C H ? .P r /:
(v) If 0 then
F . C 1/ E. C 1/ Š E. 1/ F . 1/ ˚ P
where H .P / Š O ˝C H ? .P 1 /.
Similarly, if 0 then
E. 1/ F . 1/ Š F . C 1/ E. C 1/ ˚ P 0
where H .P 0 / Š O ˝C H ? .P 1 /.
(vi) We have
H .i23 E. C 1/ i12 E. 1// Š E .2/ ./Œ1f1g ˚ E .2/ ./Œ2f3g
where i12 and i23 are the closed immersions
i12 W Y . 2/ Y ./ ! Y . 2/ Yz ./;
i23 W Y ./ Y . C 2/ ! Yz ./ Y . C 2/:
(vii) If 0 and k 1 then the image of supp.E .r/ . r// under the projection to
Y./ is not contained in the image of supp.E .rCk/ . r k// also under the
projection to Y ./. If 0 and k 1 then the image of supp.E .r/ . C r// in
Y./ is not contained in the image of supp.E .rCk/ . C r C k//.
At the level of Grothendieck groups E and F induce maps of C-vector spaces
E W K.Y. 1// ! K.Y . C 1// and F W K.Y . C 1// ! K.Y . 1//:
L
This gives an action of sl2 on K.Y .// where the weight spaces are K.Y .//. In
fact, everything is over CŒq; q 1 and we actually obtain a Uq .sl2 / representation. So
the above action should really be called a geometric categorical Uq .sl2 / action.
68 S. Cautis
3.3 Some remarks. The definition above is not necessarily the simplest but is tailored
so that it is easier to check on categories of coherent sheaves. Here are some remarks
about the relevance of conditions (i) through (vii) above.
Condition (i) is used to ensure that the Krull–Schmidt property holds (namely unique
decomposition into irreducibles). Condition (ii) is used to make sense of condition (iv).
Conditions (iv) and (v) are checked only at the level of cohomology. This is because
it is often possible to compute the cohomology of an object (like P in condition (v))
but difficult to show that the object is formal (i.e. the direct sum of its cohomology).
The role of the deformation Yz ./ ! A1 is actually related to this issue. We explain
this now.
The short exact sequence of tangent bundles
0 ! TY./ ! TYz ./ jY./ ! NY./=Yz ./ ! 0
gives us a connecting map b 2 H 1 .Y ./; TY./ f2g/ since NY./=Yz ./ Š OY./ f2g.
This is just the first order deformation corresponding to Yz ./ ! A1 and is uniquely
defined up to a non-zero multiple. Now, the Hochschild–Kostant–Rosenberg isomor-
phism states that M Vi
OY./ Š TY./
?
Œi
i
where W Y./ ! Y ./ Y ./ is the inclusion as the diagonal. This implies that
Hom. OY./ ; OY./ Œ2/

V
Š H 0 .Y ./; 2 TY./ / ˚ H 1 .Y ./; TY./ / ˚ H 2 .Y ./; OY./ /:
In particular, this means that b induces a map
ˇ W OY./ ! OY./ Œ2f2g: (4)
Unfortunately, in practice it is difficult to get your hands on such a map. The purpose
of the deformation Yz ./ is simply to yield ˇ.
Now consider the map
IˇI W E OY./ E ! E OY./ EŒ2f2g
where EE 2 D.Y.2/Y .C2//. The cohomology H 1 of both sides is E .2/ f1g.
The content of condition (vi) is that the map above induces an isomorphism on H 1 .
In turn, this allows you to conclude that E E equals E .2/ Œ1f1g ˚ E .2/ Œ1f1g on
the nose rather than at the level of cohomology. This is by a little trick that goes back
at least to Deligne. For more details see [CKL2].
Finally, condition (vii) is an annoying technical condition which is only ever used
once (namely in Lemma 4.6 of [CKL2] which is itself technical in nature). Though
unsightly, its main advantage is that it is very easy to check.
3.4 Inducing equivalences. We first explain why all this is related to constructing
0 Suppose one has an sl2 action on a vector space V . The action of
equivalences.
H WD 10 1 2 sl2 breaks up V into H -eigenspaces V ./ where H v D v if
v 2 V ./. Moreover, using the relation ŒE; F D H where

0 1 0 0
E WD 2 sl2 and F WD 2 sl2 ;
0 0 1 0
one can check that E W V ./ ! V . C 2/ and F W V ./ ! V . 2/.

If V ./ D 0 for 0 or
0 then this action integrates to an action of the Lie
group SL2 .C/. Here it is known that the reflection element

0 1
tD
1 0
induces an isomorphism of vector spaces V ./

! V ./. Moreover, if say 0,
we can write t as
t D F ./ F .C1/ E C F .C2/ E .2/ ˙ (5)
where E .k/ WD E k =kŠ and F .k/ WD F k =kŠ. Notice that the sum is finite since
V ./ D 0 for 0.
Now we try to imitate this construction with categories. We replace V ./ by the
category D.Y.//, the functors E .r/ and F .r/ by the kernels E .r/ and F .r/ and the
sum (5) describing t with a complex
‚ D Œ ! ‚s ! ‚s1 ! ! ‚1 ! ‚0
where
‚s WD F .Cs/ .s/ E .s/ . C s/Œsfsg 2 D.Y ./ Y .//: (6)
Again, this complex is finite since Y ./ is empty for 0 or
0. The differential
is given by the composition
F .CsC1/ E .sC1/ ! F .Cs/ F E E .s/ Œ2sfC2sg ! F .Cs/ E .s/ Œ1f1g
where the first map is the inclusion of F .CsC1/ and E .sC1/ into the lowest coho-
mological degrees of F .Cs/ F and E E .s/ respectively while the second map is
induced by the adjunction map F E ! O Œ C 2s C 1f 2s 1g (using that F
is the left adjoint of E up to a shift). The complex ‚ is sometimes called Rickard’s
complex.
Theorem 3.1 ([CKL3], Theorem 2.8). The complex ‚ has a unique right convolu-
tion T ./ 2 D.Y ./ Y .//. Moreover, ˆT ./ W D.Y .// ! D.Y .// is an
equivalence which categorifies the isomorphism t W K.Y .//
! K.Y .//.
70 S. Cautis
This theorem is proved in two steps. In the first step [CKL2] we prove that a
geometric categorical sl2 action induces a strong categorical sl2 action. Without
recalling the precise definition of the latter let us note that its most remarkable property
is an action of the nilHecke algebra on E’s.
More precisely, in [CKL2] we show that given a geometric categorical sl2 action
one can construct two types of maps
X W E ! EŒ2f2g and T W E E ! E EŒ2f2g
which satisfy the following relations:
(i) T 2 D 0 where T 2 End.E E/,
(ii) .I T /.T I /.I T / D .T I /.I T /.T I / where T I; I T 2 End.E E E/,
(iii) .XI /T T .IX / D I D .IX /T C T .XI / where XI; IX; T 2 End.E E/.
Recall that E and E E are just (complexes of) sheaves so X and T are maps of
(complexes of) sheaves. If instead we think of the functors induced by E and E E
then X and T are natural transformations of functors.
In the second step [CKL3] we show that in a strong categorical sl2 the complex ‚
has a unique right convolution which induces an equivalence. The role of the nilHecke
algebra is to help simplify expressions of the form ‚ F .r/ .
Of course, the second step no longer involves any geometry. In fact, a similar
result was proved in [CR]. However, their action was on abelian categories and it was
not clear how to extend it to triangulated categories. In the end, the proof we give in
[CKL3] is fairly different from that in [CR].
The maps X and T are examples of higher structure in the representation theory
of sl2 . The rôle of the nilHecke algebra in the (higher) representation theory of sl2
is studied in detail by Lauda in [L]. Subsequently, Khovanov–Lauda [KL1], [KL2],
[KL3] and Rouquier [Ro2] describe certain graded algebras now called quiver Hecke
algebras or KLR algebras which play the rôle for other Lie algebras (such as slm / that
the nilHecke plays for sl2 . We will discuss certain categorical slm actions in Section 5
although we do not make any further reference to these KLR algebras.
4 Equivalences for the local model of stratified flops

L
4.1 Categorical actions on k D.T ? G.k; N //. In this section we fix N and let
Y./ WD T ? G.k; N / and E

Yz ./ WD T ? G.k; N / where D N 2k: (7)
Consider the correspondences
W r ./ Y . r/ Y . C r/ D T ? G.k C r=2; N / T ? G.k r=2; N /

defined by
˚
W r ./ WD .X; V; V 0 / W X 2 End.C N /; dim.V / D k C 2r ; dim.V 0 / D k 2r ;
X X
0 V 0 V CN ; CN
! V 0 and V
!0 :
There are two natural projections 1 W .X; V; V 0 / 7! .X; V / and 2 W .X; V; V 0 / 7!

.X; V 0 / from W r ./ to Y . r/ and Y . C r/ respectively. Together they give us an
embedding
.1 ; 2 / W W r ./ Y . r/ Y . C r/:
On W r ./ there are two tautological bundles, namely V WD 1 .V / and V 0 WD

2 .V/ where the prime on the V 0 indicates that the vector bundle is the pullback of
the tautological bundle by the second projection. We also have natural inclusions
0 V 0 V CN
where C N denotes the trivial vector bundle on W r ./.

Now define kernels
E .r/ ./ 2 D.Y. r/ Y . C r//

and
F .r/ ./ 2 D.Y. C r/ Y . r//
by
E .r/ ./ WD OW r ./ ˝ det.C N =V /r ˝ det.V 0 /r fr.N r/=2g; (8)

0
F .r/
./ WD OW r ./ ˝ det.V =V / fr.N C r/=2g: (9)
In [CKL3] (although most of the hard work is done in [CKL1]) we prove the
following:
Theorem 4.1 ([CKL3], Theorem 6.1). The varieties Y ./ and their deformations Yz ./
defined in (7) together with the functors E .r/ ./ and F .r/ ./ from (8) and (9) define
a geometric categorical sl2 action.
So, as a consequence of Theorem 3.1, this gives us an equivalence

ˆT .k;N / W D.T ? G.k; N //
! D.T ? G.N k; N //:
In fact, one can show that T .k; N / is a sheaf ([CKL2], Proposition 6.6). In the next
section we identify T .k; N / more explicitly.
72 S. Cautis
4.2 The equivalence: an explicit description
4.2.1 Some geometry. Using the above notation Y ./ WD T ? G.k; N / where D
N 2k recall that the stratified Mukai flop is summarized by the diagram
p.k;N / p.N k;N /
Y ./ ! B.k; N / Y ./:
Now consider the fibre product
Z.k; N / WD Y ./ B.k;N / Y ./
k V N k N
D f0 C W X C N V; XC N V 0 ; X V 0; X V 0 0g:
N k V0 k
Since p.k; N / and p.N k; N / are semi-small, Z.k; N / is equidimensional of dimen-
sion 2k.N k/. It consists of .k C1/ irreducible components Zs .k; N / (s D 0; : : : ; k/
where
Zs .k; N / WD p.k; N /1 .B.k s; N // B.ks;N / p.N k; N /1 .B.k s; N //:
The component Zs .k; N / can be described more directly as
f.X; V; V 0 / 2 Z.k; N / W dim.ker X / N k C s and dim.V \ V 0 / k sg:
It is helpful to keep in mind the following. Any two components Zs .k; N / and
Zs 0 .k; N / intersect in a divisor if js s 0 j D 1 but their intersection has strictly higher
codimension if js s 0 j > 1.
Now, since span.V; V 0 / ker X , it follows that dim.ker X / C dim.V \ V 0 / N
on Z.k; N /. We define the open subscheme
Z o .k; N /
WD f.X; V; V 0 / 2 Z.k; N / W N C 1 dim.ker X / C dim.V \ V 0 /g Z.k; N /
and Zso .k; N / WD Zs .k; N / \ Z o .k; N /.
Theorem 4.2 ([C1], Theorem 3.8). There exists a C -equivariant line bundle L.k; N /
on Z o .k; N / such that T .k; N / Š i j L.k; N / where i and j are the natural inclu-
sions
j i
Z o .k; N / !
Z.k; N / ! Y ./ Y ./:
Note that the map j in Theorem (4.2) is an open immersion. Whenever we have
an open immersion in this paper j denotes the non-derived push-forward. This is the
only case in this paper when a functor is not derived.
The line bundle L.k; N / is uniquely determined by its restriction to each Zso .k; N /.
One has that L.k; N /jZso .k;N / is isomorphic to
OZso .k;N / .ŒDs;C
o
.k; N // ˝ det.C N =V /s ˝ det.V 0 /s fk.N k/ .k s/2 C sg
where Ds;C
o
.k; N / is the divisor Zso .k; N /\ZsC1 .k; N / and V; V 0 are the tautological
bundles on Zso .k; N / Y ./Y ./ pulled back from Y ./ and Y ./ respectively.
4.2.2 Why is Theorem 4.2 believable? Recall that T .k; N / is the right convolution
of the complex
F .Ck/ E .k/ Œkfkg ! ! F .C1/ EŒ1f1g ! F ./ : (10)
Now, for s D 0; 1; : : : ; k, one can show ([CKL3], Proposition 6.3) that F .Cs/ E .s/ is
a sheaf supported exactly on Zs .k; N / (actually, one can identify this sheaf explicitly).
It then follows
S quite easily that the convolution in (10) above is also a sheaf supported
exactly on s Zs .k; N / D Z.k; N /.
So we just need to identify this sheaf. To do this we first argue that T .k; N / is (the
push-forward by a closed embedding of) a Cohen–Macaulay sheaf. This is done by
identifying the kernel which is the inverse of T .k; N / (see Section 4.3) and showing
by the same argument above that it is also a sheaf. Formal non-sense says that the
inverse kernel is just T .k; N /_ tensored with some line bundle and a shift. This means
T .k; N /_ is a sheaf and hence T .k; N / is (the push-forward of) a Cohen–Macaulay
sheaf.
Finally, any Cohen–Macaulay sheaf is uniquely determined by its restriction to an
open subset of codimension at least two. The last step is to identify the restriction
of T .k; N / to Z o .k; N / which is codimension two inside Z.k; N / (we do this in
[C1]). The advantage of Z o .k; N / over Z.k; N / is that two components Zso .k; N /
and Zso0 .k; N / in Z o .k; N / intersect in a Cartier divisor if js s 0 j D 1 and are disjoint
if js s 0 j > 1. So Z o .k; N / avoids all the nastier singularities of Z.k; N /.
4.3 The inverse. The inverse T .k; N /1 of T .k; N / is given by its left (or equiva-
lently its right) adjoint. This is equal to the left convolution of the complex
.‚0 /L ! .‚1 /L ! ! .‚s1 /L ! .‚s /L !
where it is easy to check ([C1], Section 5.1) that
.‚s /L Š F .s/ E .N 2kCs/ Œsfsg:
It then follows that T .k; N /1 is again a sheaf which is the push-forward of a line
bundle from an open subset of Z.k; N / [C1], Theorem 5.3. Perhaps a little more
surprising:
Proposition 4.3 ([CKL4], Corollary 7.5). The kernels T .k; N /1 and T .N k; N /
are related by
T .k; N /1 Š T .N k; N / ˝ det.V /1 ˝ det.V 0 /1 fN 2kg 2 D.Y ./ Y .//:
This isomorphism is something which is special to the example of cotangent bundles

of Grassmannians. In other words, such a relation is not a formal consequence of having
a geometric categorical sl2 action.
74 S. Cautis
4.4 The equivalence: stratified Atiyah flops. Recall the notation

E
Yz ./ WD T ? G.k; N /
which comes equipped with a map Yz ./ ! A1 . Similar to the above description of
T .k; N / we now proceed to describe a kernel Tz .k; N / which induces an equivalence
ˆTz .k;N / W D.Yz .//
! D.Yz .//.
4.4.1 Some geometry. Recall that the stratified Atiyah flop is summarized by the
diagram
Q
p.k;N / Q k/
p.N
Yz ./ ! B.k;
z N / Š B.N
z k; N / Yz ./:
Once again we can consider the fibre product
z N / WD Yz ./
Z.k; Yz ./
z
B.k/
n k V N k
D 0 0 C N W X C N V; .X x id/C N V 0 ;
N k V k
o
.X x id/V 0; X V 0 0
z N / is
which deforms the old fibre product Z.k; N /. However, unlike Z.k; N /, Z.k;
z N / is naturally a subscheme of Yz ./ A1 Yz ./
now irreducible. Notice that Z.k;
(where the second projection Yz ./ ! A1 maps .X; V; x/ 7! x).
Next, as before, we can define an open subscheme Zz o .k; N / as follows:
z N / W N C1 dim.ker.X x id//Cdim.V \V 0 /g Z.k;
f.X; V; V 0 ; x/ 2 Z.k; z N /:
Notice that if x ¤ 0 then V and V 0 are uniquely determined by X as the kernels of

.X x id/ and X respectively. This means that Z z o .k; N / contains all the fibres over
0
x ¤ 0 since when x ¤ 0 we have V \ V D 0.
Inside Zz o .k; N / we have Zso .k; N /, the components of the central fibre. One can
check that these are Cartier divisors.
Theorem 4.4 ([C1], Theorem 4.1). Consider the line bundle on Z z o .k; N / given by
k
X
z N / WD O z o sC1
L.k; ŒZso .k; N / fk.N 2k/g
Z .k;N /
sD0
2
and let
z N / 2 D.Yz ./ A1 Yz .//
Tz .k; N / WD iQ jQ L.k;
where iQ and jQ are the natural inclusions
jQ i Q
z o .k; N / !
Z z N/ !
Z.k; Yz ./ A1 Yz ./:
Then ˆTz .k;N / W D.Yz .//
! D.Yz .// is an equivalence and the restriction of
Tz .k; N / to Y./ Y ./ is T .k; N /.
The map jQ above is an open immersion so again Q

L j denotes the non-derived push-
forward. There is no categorical sl2 action on D.Yz .// so to prove Theorem 4.4
one guesses the expression for Tz .k; N / and shows that it restricts to an equivalence
over each fibre of the map to A1 . Hence Theorem 4.4 is essentially a corollary of
Theorem 4.2.
z N / contains two components

4.4.2 The case k D 1. The central fibre of Z.1;
Z0 .1; N / and Z1 .1; N / while
Tz .1; N / Š iQ jQ OZz o .1;N / .ŒZ1o .1; N //fN 2g:
Now, one can show (see Lemma 5.1 of [C1]) that
Xk
iQ jQ OZz o .k;N / . sŒZso .k; N // Š OZ.k;N
z
_ 0
/ ˝ det.C =V / ˝ det.V /f2kg (11)
N
sD1
for any k. When k D 1 this means that

Tz .1; N / Š OZ.1;N
z
_ 0
/ ˝ det.C =V / ˝ det.V /fN g:
N
In particular this implies:

z N / induce equiv-
Corollary 4.5. The fibre product correspondences Z.1; N / and Z.1;
alences

ÔZ.1;N / W D.Y .1//
!D.Y.N 1// and ÔZ.1;N
z /
W D.Yz .1//
!D.Yz .N 1//:

This corollary was originally proved by Kawamata [K1] and Namikawa [Nam1].
Namikawa [Nam1], Section 2, also shows that the correspondence
W II
zz II 2
1
zzz II
II
z
}z
z I$
Y .1/ Y .N 1/
where
1 N 2 1
W D f.X; V; V 0 / W X 2 End.C N /; 0 !
V ! V 0 !
C N ; XC N V; X V 0 D 0g
does not induce an equivalence. This correspondence is natural since it is isomorphic
to the blowup of the zero section of Y .1/ D T ? G.1; N / and that of Y .N 1/ D
T ? G.N 1; N /.
From the point of view of categorical sl2 actions, ÔW is not an equivalence
because OW is equal to F .N 1/ .0/ 2 D.Y .1/ Y .N 1// (up to tensoring by a line
bundle). This means that the composition .OW /L OW is equal to
E .N 1/ .0/ F .N 1/ .0/ Š O ˚ F E ˝C H ? .P 1 / (12)
up to tensoring by a line bundle (this is clearly not equal to O ). Note that relation (12)
above is a formal consequence of having a categorical sl2 action ([CKL3], Lemma 4.2).
76 S. Cautis
4.4.3 The case k D 2. The argument above, namely twisting by the line bundle in
(11), does not work here to imply that ÔZ.2;N
z /
is an equivalence. Perhaps even more
surprising is Namikawa’s result [Nam2] that
ÔZ.2;4/ W D.Y .2; 4// ! D.Y .2; 4// and ÔZ.2;4/
z
W D.Yz .2; 4// ! D.Yz .2; 4//
are not equivalences.
In [K2] Kawamata tried to tweak the kernel OZ.2;N z / to obtain an equivalence.
z
He defined functors ‰ and ˆ as follows. Inside Y .2; N / there are two natural strata:
namely the locus where X D 0 (isomorphic to G.2; N /) and the locus where rank X
1 (the locus rank X 2 is the whole central fibre Y .2; N /). Kawamata blows up the
first locus and then the strict transform of the second locus to obtain
f f
Yz 00 .2; N / ! Yz 0 .2; N / ! Yz .2; N /:
1 2
Inside Yz 00 .2; N / we denote by E1 the exceptional divisor of f1 and by E2 the strict

transform of the exceptional divisor of f2 . Warning: our labeling of divisors does not
match precisely that in [K2].
Kawamata then blows up Yz .N 2; N / in the same way to obtain Yz 00 .N 2; N / and
identifies this smooth variety with Yz 00 .2; N /. To summarize, we arrive at the following
commutative diagram:
Yz 00 .2; N / Š Yz 00 .N 2; N / (13)
ll TTTT
f llll TTTfT C
lll TTTT
ll TTT)
ulll 2
z
Y .2; N / Ro 1
z
Z.2; N / / Yz .N 2; N /
RRR kk k
RRR
RRR kk kkkkk
RRR kk
R) ukkkk
z N /.
B.2;
z N / D Yz .2; N /
Notice that Z.2; Yz .N 2; N / so the map exists by the
z
B.2;N /
universal property of fibre products. The functors ‰ and ˆ are then defined by
‰./ WD f .f C ./ ˝ OYz 00 .2;N / .ŒE2 //; (14)
ˆ./ WD fC .f ./ ˝ OYz 00 .2;N / ..2N 5/ŒE2 C .N 3/ŒE1 //: (15)
If we ignore the fg shift for convenience we have:
Proposition 4.6 ([C1], Proposition 5.7). The functor induced by the kernel
z N / ˝ det.C N =V / ˝ det.V 0 /_
iQ jQ L.2;
is an isomorphism and together with its adjoint is equal to Kawamata’s functors
‰; ˆ W D.Yz .2; N // ! D.Yz .N 2; N //:
4.5 Equivalences and K-theory. Namikawa’s proof that ÔZ.2;4/ z

is not an equiv-
alence is via an impressive calculation of Hom spaces which implies that it is not
fully-faithful. It would be interesting to have a more conceptual explanation of this
fact. Moreover, it is not known if ÔZ.k;N
z /
fails to be an equivalence for any k ¤ 1
and N 2k (although one expects this is the case).
Namikawa also shows ([Nam2], Theorem 2.6 and Theorem 2.7) that on K-theory
we do have isomorphisms

ŒÔZ.k;N / W K.Y .//
! K.Y .// and ŒÔZ.k;N
z /
W K.Yz .//
! K.Yz .//

where K.X/ denotes the usual Grothendieck group of coherent sheaves on a variety X .
This fact is a consequence of specialization in K-theory. The argument is as follows.
For simplicity let us ignore the C -action for a moment. We follow the notation of
Chriss and Ginzburg [CG]. Suppose Yz ! A1 is a flat family with central fibre Y and
denote Yz WD Yz n Y . Then (see Section 5.3 in [CG]) they describe a specialization
map in K-theory
lim W K.Yz / ! K.Y /:
t!0
More precisely, they show that given any sheaf P on Yz there exists a sheaf P on Yz
such that
• P restricts to P on Yz ,
• P has no subsheaves supported on Y .
Moreover, they show that for any two such sheaves P1 and P2 their restriction to K.Y /
are the same. This restriction is by definition lim t!0 P .
Now let us apply this to Yz WD Yz ./ A1 Yz ./ where P D OZ.k;N
z / jYz . Then
P D OZ.k;N
z / is a possible choice for the extension which means that
lim P D ŒOZ.k;N / :
t!0
On the other hand, Z.k; z N / restricted to Yz is actually the graph of an isomorphism

z z
Y ./ ! Y ./ where the indicates the complement of the central fibre. This
immediately implies that ŒP is invertible in K-theory, i.e. Œ.P /L ŒP Š ŒO z .
Now, in Theorem 5.3.9 of [CG] they also show that the specialization map is com-
patible with convolution. Since lim t!0 ŒO z D ŒO this implies that Œ.OZ.k;N / /L
ŒOZ.k;N / Š ŒO which means that ŒÔZ.k;N / W K.Y .// ! K.Y .// is invertible.
It then follows that ŒÔZ.k;N
z /
is also invertible.
Since the restriction of the line bundle L.k; z N / from Theorem 4.4 to the general
fibre is trivial this also means that in K-theory
ŒT .k; N / D ŒOZ.k;N / 2 K.Y ./ Y .//:

78 S. Cautis
But when k D 2; N D 4 we know that T .k; N / induces an equivalence while OZ.k;N /

does not. In particular, there is no natural specialization map at the level of categories
which lifts the one on K-theory. This is unfortunate as it makes it difficult to use
deformations to construct and prove derived equivalences.
Remark 4.7. Maulik and Okounkov have been working on a more functorial spe-
cialization map in K-theory. It is possible their work can help define a reasonable
specialization map at the level of derived categories.
5 Geometric categorical sl m actions

One can define the concept of a geometric categorical g action for any simply laced
Kac–Moody Lie algebra g [CK3], Section 2.2.2. Let us summarize the definition when
g D slm .
The weight lattice of slm is denoted X and h WD X ˝Z C Š Am1 . We denote
the simple and fundamental roots of slm by ˛i ; ƒi 2 X where i D 1; : : : ; m 1.
The data of a geometric categorical slm action consists of
(i) a collection of smooth complex varieties Y ./ where 2 X equipped with
C -actions;
(ii) Fourier–Mukai kernels
Ei.r/ ./ 2 D.Y ./ Y . C r˛i // and Fi.r/ ./ 2 D.Y . C r˛i / Y .//
(which are C equivariant);
(iii) for each Y ./ a flat deformation Yz ./ ! h carrying a C -action compatible
with the action x 7! t 2 x (where t 2 C ) on the base h.
Remark 5.1. Unfortunately, the indexing of E .r/ and F .r/ here is slightly different than
the convention used in [CKL1], [CKL2], [CKL3] when g D sl2 . In that convention,
the above Ei.r/ ./ should be denoted Ei.r/ . C 2r ˛i /. We use the notation here because
it is more convenient. This is also the convention adopted in [CK3], [CKL4] and
subsequent papers. Sometimes we just write Ei.r/ and Fi.r/ when the weight is obvious
or irrelevant.
On this data we impose the following conditions.
(i) The spaces fY . C r˛i / W r 2 Zg together with deformations Yz . C r˛i /
restricted to span.˛i / Š A1 and with kernels fEi.r1 / .Cr2 ˛i /; Fi.r1 / .Cr2 ˛i / W
r1 ; r2 2 Zg generate a geometric categorical sl2 action.
(ii) If ji j j D 1 then
Ei Ej Ei Š Ei.2/ Ej ˚ Ej Ei.2/
while if ji j j > 1 then Ei Ej Š Ej Ei .
(iii) If i ¤ j then Fj Ei Š Ei Fj .
(iv) The sheaf Ei deforms over ˛i? h to some
Ezi 2 D.Yz ./j˛? ˛? Yz . C ˛i /j˛? /:

i i i
(v) If ji j j D 1 then one can show formally from the relations above that there
exists a unique non-zero map Tij W Ei Ej Œ1 ! Ej Ei whose cone we denote
Tij
Eij WD Cone Ei Ej Œ1 ! Ej Ei 2 D.Y ./ Y . C ˛i C ˛j //:
Then Eij deforms over B WD .˛i C ˛j /? h to some
Ezij 2 D.Yz ./jB B Yz . C ˛i C ˛j /jB /:
5.1 Some remarks. The first condition above summarizes conditions (i)–(vii) in Sec-
tion 2.2.2 of [CK3] while the last four are conditions (viii)–(xi) in [CK3].
Conditions (ii) and (iii) are just categorical versions of the standard Uq .slm / rela-
tions
1 2
Ei Ej Ei D Ei Ej C Ej Ei2 if ji j j D 1;
2
Ei Ej D Ej Ei if ji j j > 1;
and
Ei Fj D Fj Ei if i ¤ j:
To explain the content of (iv) and (v) recall that the deformation Yz ./ ! h induces
a map
ˇv W OY./ ! OY./ Œ2f2g
for any v 2 h by restricting Yz ./ to span.v/ and then applying the same construction
used to obtain the map in (4). The content of (iv) is that for any v 2 ˛i? h both maps
ˇv I W OY./ Ei ! OY./ Ei Œ2f2g;

Iˇv W Ei OY.C˛i / ! Ei OY.C˛i / Œ2f2g
are equal to zero.

To see why this is the case consider a deformation Yz ./ ! A1 and denote by
i W Y./ ! Yz ./ the inclusion of the fibre over 0 2 A1 . Then for any A 2 D.Y .//
there is a natural distinguished triangle
AŒ1f2g ! i i A ! A (16)
80 S. Cautis
z for some
which induces a connecting map ˛ W AŒ1 ! AŒ1f2g. But if A D i A
z z
A 2 D.Y .// (i.e A deforms) then
z Š i .A
i i A Š i i i A z ˝ i i OY./ Š A ˚ AŒ1f2g:
z ˝ i OY./ / Š i A
This means that ˛ D 0. On the other hand, one can check that ˛ is the same as the map
ˇI W OY./ A ! OY./ AŒ2f2g which means that ˇI D 0.
In condition (v) the object Eij should be thought of as the E for the root ˛i C ˛j .
Then the content of (v) is analogous to that of (iv), namely, it states that for any
v 2 .˛i C ˛j /? h both maps
ˇv I W OY./ Eij ! OY./ Eij Œ2f2g;

Iˇv W Eij OY.C˛i C˛j / ! Eij OY.C˛i C˛j / Œ2f2g
are zero.
Condition (v) was included in [CK3] because in practice Eij is a natural kernel
supported on the union of supp.Ei Ej / and supp.Ej Ei / and one can write down the
deformation Ezij fairly easily and explicitly. Moreover, from a geometric standpoint, it
is interesting to see these deformations. However, the rôle of (v) in proving the braid
relation in [CK3] is quite minimal. Namely, it is used to show in Lemma 4.9 of [CK3]
that the map
Tij I
Ei Ej Ei Œ1f1g ! Ej Ei Ei
induces an isomorphism between the summands Ej Ei.2/ Œ1f1g on either side. How-
ever, it turns out one can check this directly without the need of (v) (see [C2]). Thus,
condition (v) is essentially redundant.
5.2 Braid group actions. First, recall some basic facts about the weight lattice X
of slm . The weight lattice X comes equipped with a symmetric bilinear pairing h; i.
Under this pairing we have h˛i ; ƒj i D ıij and h˛i ; ˛j i equals 2; 1 or 0 depending
on whether i D j , ji j j D 1 or ji j j > 1 respectively.
The Weyl group of slm is the symmetric group Sm generated by s1 ; : : : ; sm1 with
the usual relations si2 D 1, si sj D sj si if ji j j > 1 and si sj si D sj si sj if ji j j D 1.
It acts on the weight lattice X via
si ./ WD h; ˛i i˛i :
Recall that our motivation behind studying geometric categorical sl2 actions is that
they induce equivalences (Theorem 3.1). A geometric categorical slm action contains
m 1 different sl2 actions, generated by Ei and Fi for i D 1; : : : ; m 1. Thus, for
each i one can write down a complex ‚ just like the one in (6) but where
.h;˛i iCs/
‚s WD Fi Ei.s/ Œsfsg 2 D.Y ./ Y .si //:
Here we are assuming that h; ˛i i 0 (there is a similar complex if h; ˛i i 0). These
complexes have unique right convolutions, denoted Ti ./ 2 D.Y ./Y .si //, which
induce equivalences. In particular, this means that any two varieties in the same Weyl
orbit are derived equivalent.
But having a geometric categorical slm action is more than having m 1 geometric
categorical sl2 actions. This extra structure leads to the following result:
Theorem 5.2 ([CK3], Theorem 2.10). The kernels Ti satisfy the braid group relations.
Namely, Ti Tj Š Tj Ti if ji j j > 1 and Ti Tj Ti Š Tj Ti Tj if ji j j D 1.
This gives an action of the braid group Bm on D.tY .// compatible with the action
of the Weyl group on the weight lattice.
The key step in the proof of Theorem 5.2 lies in proving that if ji j j D 1 then
Eij Ti Š Ti Ej (see Corollary 5.4 in[CK3]. This implies that Tij Ti Š Ti Tj where
Tij is the equivalence build out of Eij and Fij (in other words, this is the equivalence
induced by the sl2 action corresponding to the root ˛i C ˛j ). It then follows by a
similar argument that Tj Eij Š Ti Tj which means Tj Tij Š Ti Tj . Thus
Tj1 Ti Tj Š Tij Š Ti Tj Ti 1
and we are done.
5.3 Examples. We now describe some examples of geometric categorical slm actions.
5.3.1 Example: resolution of Kleinian singularities. Consider the standard diago-

nal embedding of G WD Z=mZ inside SL2 .C/ and let W Y ! C 2 =G be the minimal
resolution. Recall that the exceptional fibre 1 .0/ contains a chain of m 1 P 1 s
which we label E1 ; : : : ; Em1 .
We can construct a geometric categorical slm action as follows. We let Y .0/ WD Y ,
Y./ WD pt for a root of slm and Y ./ WD ; for all other 2 X. The action of C
on Y comes from the scaling action on C 2 . We define Ei .0/ W D.Y / ! D.pt/ using
the kernel OEi .1/ 2 D.Y pt/ and similarly with Ei .˛i /, Fi .0/ and Fi .˛i /
(all other Ei ; Fi we need to define are functors D.pt/ ! D.pt/ which we take to
be the identity). The deformation Yz of Y is the standard deformation which may be
constructed by thinking of C 2 =G as a Slodowy slice or by deforming the polynomial
defining the singularity C 2 =G.
Fact. This defines a geometric categorical slm action.
Let us check condition (ii) for having a geometric categorical slm action (all other
conditions are immediate or follow along the same lines). Suppose ji j j D 1 so that
Ei and Ej intersect in a point. Then condition (ii) states that
Ei .˛j / Ej .0/ Ei .˛i / Š Ej .˛i / Ei.2/ .˛i / ˚ Ei.2/ .˛i C ˛j / Ej .˛i /:

82 S. Cautis
Now Y.˛i C ˛j / D ; while Ei.2/ .˛i / D Opt D Ei .˛j /. So we see that this is
equivalent to the fact that the composition
Ei .˛i / Ej .0/
D.pt/ ! D.Y / ! D.pt/
is the identity. Since the first functor is given by tensoring with OEi .1/ and the
second functor is Ext .OEj .1/Œ1; / this condition corresponds to the fact that
Ext` .OEj .1/; OEi .1// is zero unless ` D 1 in which case it is one-dimensional.
5.3.2 Example: flag varieties. The following generalizes the geometric categorical
sl2 action on cotangent bundles to Grassmannians from Section 4.1.
Fix m N and consider the variety Flm .C N / of m-step flags in C N . This variety
has many connected components, which are indexed by the possible dimensions of the
spaces in the flags. In particular, let
C.m; N / WD f D .1 ; : : : ; m / 2 N N W 1 C C m D N g:
For 2 C.m; N /, we can consider the variety of m-steps flags where the jumps are
given by :
Fl .C N / WD f0 D V0 V1 Vm D C N W dim Vi =Vi1 D i g:
Let Y./ WD T ? Fl .C N / (if 62 C.m; N / we take Y ./ D ;). These will be our
varieties for the geometric categorical slm action. We regard each as a weight for
slm via the identification of the weight lattice of slm with the quotient Zm =.1; ; 1/.
By convention the simple root ˛i equals .0; : : : ; 0; 1; 1; 0; : : : ; 0/ where the 1 is in
position i.
We will use the following description of the cotangent bundle to the partial flag
varieties (this generalizes the description for Grassmannians in (1)):
Y./ WD f.X; V / W X 2 End.C N /; V 2 Fl .C N /; X Vi Vi1 g
This description suggests the following deformation Yz ./ ! C m of Y ./:
f.X; V; x/ W X 2 End.C N /; V 2 Fl .C N /; x 2 C m ; X Vi Vi ; X jVi =Vi 1 D xi idg:
We will restrict our deformation over the locus f.x1 ; : : : ; xn / 2 C m W xm D 0g
which we identify with h, the Cartan for slm .
We define an action of C on Yz ./ by t .X; V; x/ D .t 2 X; V; t 2 x/. Restricting
to Y./ D T ? Fl .C N / this corresponds to a trivial action on the base and a scaling of
the fibres.
To construct the kernels Ei.r/ we consider correspondences Wir ./ analogous to
W r ./ defined in Section 4.1. More specifically, let , i , r be such that 2 C.m; N /
and C r˛i 2 C.m; N / (i.e. i r). Then we define
Wir ./ WD f.X; V; V 0 / W .X; V / 2 Y ./; .X; V 0 / 2 Y . C r˛i /;
Vj D Vj0 for j ¤ i; and Vi0 Vi g:
From this correspondence we define
Ei.r/ ./ WD OWir ./ ˝ det.ViC1 =Vi /r ˝ det.Vi0 =Vi1 /r fr.i r/g
where, abusing notation, Vi denotes the vector bundle on Y ./ whose fibre over
.X; V / 2 Y./ is Vi . This sheaf belongs to D.Y ./ Y . C r˛i //. Similarly,
we define
Fi.r/ ./ WD OWir ./ ˝ det.Vi0 =Vi /i C1 i Cr friC1 g 2 D.Y . C r˛i / Y .//:
Note that now we regard Wir ./ as a subvariety of Y .Cr˛i /Y ./ which means that
Vi Vi0 (since, by convention, the prime indicates pullback from the second factor).
Theorem 5.3 ([CK3], Theorem 3.1). The datum above defines a geometric categorical
slm action on D.T ? Flm .C N //.
As a corollary of Theorems 5.2 and 5.3 we obtain:
Corollary 5.4 ([CK3], Theorem 3.2). There is an action of the braid group Bm on
the derived category of coherent sheaves on T ? Flm .C N / which is compatible with the
action of Sm on the set of connected components C.m; N /.
Although the construction of each kernel
Ti 2 D.T ? Fl .C N / T ? Flsi .C N //
via a categorical slm action is elaborate, Ti has a concrete description like the one in
Theorem 4.2 (which is just the special case m D 2). In particular, Ti is a Cohen–
Macaulay sheaf supported on the variety
Zi ./ WD f.X; V; V 0 / W X 2 End.C N /; V 2 Fl .C N /; V 0 2 Flsi .C N /

X Vj Vj 1 ; X Vj0 Vj01 and Vj D Vj0 if j ¤ ig:
5.3.3 Special cases of interest. If N D d m for some integer d and we choose

D .d; : : : ; d / then si D for all i . Thus we obtain an action of the braid group
on D.T ? Fl .C N // which is a connected variety.
Furthermore, if d D 1 (i.e. N D m) then T ? Fl .C N / is the cotangent bundle of
the full flag variety of C N . An action of the braid group on the full flag variety was
also constructed by Khovanov–Thomas [KT] and by Riche [Ric], Bezrukavnikov–
Riche [BR]. In this case, the generators of the braid group act by Seidel–Thomas
(a.k.a. spherical) twists (see Section 6.1 below). Their work served as motivation for
constructing the braid group actions above between more general partial flag varieties.
84 S. Cautis
5.3.4 Affine braid groups.. The (extended) affine braid group of slm has generators
Ti and i for i D 1; : : : ; m 1 and the followingrelations:
• Ti Tj D Tj Ti if ji j j > 1 and Ti Tj Ti D Tj Ti Tj if ji j j D 1,
• Ti j D j Ti if i ¤ j ,
1 1
• Ti D i1 iC1 i Ti1 i for all i ,
• i j D j i for all i , j .
In [CKL4] we show that the braid group action on D.T ? Flm .C N // extends to an
affine braid group action as follows:
Corollary 5.5 ([CKL4], Theorem 7.2). The kernels Ti together with i WD det.Vi /
induce an action of the (extended) affine braid group of slm on the non-equivariant
derived category of coherent sheaves on T ? Flm .C N /.
In the above corollary i is the kernel inducing the functor which is tensoring with
the line bundle det.Vi /. Note that this extended action is only possible on the non-
equivariant categories as, perhaps a bit surprisingly, the equivariant shifts fg do not
work out. Note that the majority of the content in Corollary 5.5 is captured in the
relation Ti D i1 1 1
1 iC1 i Ti i which generalizes the result in Proposition 4.3.
Again, when m D N , this affine braid group action on the full flag variety was
constructed first in [KT], [Ric], [BR].
5.3.5 Example: quiver varieties. The varieties in the two examples above (Sec-
tions 5.3.1 and 5.3.2) are special cases of Nakajima quiver varieties of type Am1
[Nak1], [Nak2]. In Theorem 3.2 of [CKL4] we construct a geometric categorical slm
action on derived categories of coherent sheaves on arbitrary Nakajima quiver varieties
of type Am1 (in fact we do this for any simply laced Kac–Moody Lie algebra g). This
action recovers the two examples above as special cases.
Note that the geometry involved in constructing the geometric categorical actions
on arbitrary Nakajima quiver varieties is precisely the geometry of cotangent bundles on
Grassmannians from Section 2.1. In particular, the generators of the braid group actions
induce derived equivalences between varieties, such as T ? Fl .C N / and T ? Flsi .C N /,
which are related by stratified Mukai flops. In fact, inspired by work of Nakajima,
many of the proofs in [CKL4] reduce the problem to the case of cotangent bundles to
Grassmannians.
These quiver varieties are also equipped with natural deformations. These de-
formations are related to each other via the geometry of stratified Atiyah flops from
Section 2.2.
6 Twists
One of the first techniques for constructing derived autoequivalences was that of spher-
ical twists as defined by Seidel and Thomas in [ST]. This notion was generalized by
various authors (Horja [Ho], Anno [An], and Rouquier [Ro1]) to twists in spherical
functors (a relative version). Spherical objects were also generalized to P -objects by
Huybrechts and Thomas in [HT]. We briefly discuss their work here.
6.1 Seidel–Thomas (spherical) twists. First recall the definition of a spherical func-
tor. Let X; Y be varieties (for convenience, we ignore the C -action in this section).
Then a FM kernel P 2 D.X Y / is spherical if:
• PR Š PL Œk for some k.
• OX ! PR P Š PL P Œk ! OX Œk is a distinguished triangle in
D.X X /. Both maps here are the adjunction maps.
The induced map ˆP W D.X / ! D.Y / is called a spherical functor. Define
adj
TP WD Cone.P PR ! OY / 2 D.Y Y /
where adj is the natural adjunction map. The induced functor ˆTP W D.Y / ! D.Y /
is called a spherical twist.
Remark 6.1. The second condition above is sometimes replaced by PR P Š
OX ˚ OX Œk (which is a priori stronger). The right hand side then resem-
bles the cohomology of a sphere. There is also a mirror side to this story where the
twist TP is often induced by monodromy around a singularity whose vanishing cycle
is a sphere. This explains the terminology “spherical functor”.
Theorem 6.2 ([ST], [Ho], [An], [Ro1]). If P 2 D.X Y / is a spherical kernel then
ˆTP W D.Y / ! D.Y / is a derived autoequivalence.
If X is just a point then P 2 D.Y / is referred to as a spherical object. In this case
the setup above recovers the construction from [ST].
On the other hand if k D 2 then a spherical functor is just a special case of a
geometric categorical sl2 action. To see this we take Y ./ D ; if 62 f2; 0; 2g while
Y .2/ WD X; Y .0/ WD Y and Y .2/ WD X:
We then define
E.1/ WD P 2 D.X Y /; E.1/ WD PR Œ1 2 D.Y X /;
F .1/ WD P 2 D.X Y /; F .1/ WD PR Œ1 2 D.Y X /:
It turns out this case is simple enough that we do not need the deformations Yz ./. Now
one can easily check that the geometric categorical relations on the Es and F s defined
86 S. Cautis
above are equivalent to the fact that P is a spherical functor. Furthermore, the complex
‚ from Section 3.4 becomes
ŒF EŒ1 ! OY Š ŒP PR ! OY
which means that T Š TP and ˆTP is an autoequivalence by Theorem 3.1.

This explains how (geometric) categorical sl2 action induce spherical twists when
k D 2. Now recall that sl2 is the Lie algebra defined by the Cartan datum consisting of
the 1 1 matrix .2/. The matrix .k/ for k > 2 also defines a (generalized) Lie algebra.
Geometric categorical actions of this Lie algebra explain the existence of spherical
twists for any k 2. In practice these algebras are similar to sl2 and the equivalences
they induce are also very similar (just like spherical twists for k D 2 and for k > 2 are
defined by the same basic construction).
6.2 P n -twists. In [HT] Huybrechts and Thomas define P 2 D.Y / to be a P n -object

if P Š P ˝ !Y and Ext .P ; P / Š H .P n ; C/ as a graded ring. They then prove:
Proposition 6.3 ([HT], Lemma 2.1 and Proposition 2.6). Let P 2 D.Y / be a P n
object and h 2 Ext2 .P ; P / a generator. Then inside D.Y Y / the complex
h_ ididh tr
.P _ P /Œ2 ! P _ P !
OY
has a unique right convolution TP 2 D.Y Y / which induces an autoequivalence

ˆTP W D.Y / ! D.Y / called a P n -twist.
Remark 6.4. This is an analogue of a spherical object with k D 2. One can clearly
replace H .P n ; C/ above with a ring where the degree jumps are some arbitrary k 2.
Everything in this section works in this greater generality but for exposition purposes
we restrict to the case k D 2.
In analogy with spherical functors one can define a P n functor as follows. A FM

kernel P 2 D.X Y / is a P n kernel if
• PR Š PL Œ2n,
• H .PR P / Š OX ˝C H .P n ; C/ where H ./ denotes the cohomology

sheaves,
• there exists a map ˇ W OY ! OY Œ2 in D.Y Y / so that
IˇI W PR . OY / P ! PR . OY / P Œ2
induces an isomorphism (at the level of cohomology) between n summands

OX on either side.
Note that the third condition is the analogue of the fact that Ext .P ; P / is iso-
morphic to H .P n ; C/ as a ring (rather than as a vector space). One could replace
this condition with a ring condition on PR P but the language above seems more
convenient.
Remark 6.5. Nick Addington recently gave a similar definition of a P n functor in

Section 3 of [Ad]. Although his definition is slightly more general the key ideas and
properties are the same.
Then the analogue of Proposition 6.3 is:
Proposition 6.6. Let P 2 D.X Y / be a P n functor and suppose HH1 .X / D 0 (where

HH denotes Hochschild cohomology). Then inside D.Y Y / there is a complex
ˇI I I Iˇ adj
.P PR /Œ2 ! .P PR / ! OY (17)
that has a unique right convolution TP 2 D.Y Y /. This kernel induces an autoe-
quivalence ˆTP W D.Y / ! D.Y /.
Proof. The proof of Proposition 6.3 given in [HT] generalizes directly. The only tricky
point is to show that the convolution is unique. Using Section 3.1.3 it suffices to check
that Hom..P PR /Œ2; OY / D 0. Now
Hom..P PR /Œ2; OY Œ1/ Š Hom.P ; P Œ1/

Š Hom.PL P ; OX Œ1/
Š Hom. OX ˚ OX Œ2; OX Œ1/
Š HH1 .X / ˚ HH1 .X /:
Since HH1 .X/ D 0 the result follows.
Remark 6.7. I would like to thank Nick Addington for pointing out that one should
add the condition HH1 .X / D 0 as part of the hypothesis in Proposition 6.6. However,
I suspect that the convolution of (17) is unique even if HH1 .X / ¤ 0 (this condition is
sufficient but not necessary). However, the proof involves a lot of diagram chasings so
we leave it up to the reader as an exercise/conjecture.
It turns out P n functors are also closely related to categorical sl2 actions. To see
this consider a geometric categorical sl2 action where Y ./ D ; for > nC1. Denote
Y WD Y.n 1/ and X WD Y .n C 1/ and let P WD F .n 1/ 2 D.X Y /. Furthermore,
let ˇ W OY ! OY Œ2 be the map defined using the deformations Yz ./ of Y ./
as in Section 3.3.
We claim that P D F .n 1/ is a P n -kernel. The first two conditions are easy
consequences of conditions (iii) and (v) in Section 3.2. The last condition is harder to
see but essentially follows from condition (vi).
88 S. Cautis
Now, we have
T .n 1/ Š Cone.F .n/ EŒ1 ! F .n1/ / 2 D.Y .n 1/ Y .n C 1//
since F .nCs/ E .s/ D 0 for s > 1. And likewise
T .n C 1/ Š Cone.E .n/ F Œ1 ! E .n1/ / 2 D.Y .n C 1/ Y .n 1//:
Thus T .n C 1/ T .n 1/ 2 D.Y.n 1/ Y .n 1// is given by the right convolution

of a complex
E .n/ F F .n1/
E .n/ F F .n/ EŒ2 ! ! E .n1/ F .n1/
˚E .n1/ F .n/ EŒ1
which simplifies to
F E ˝C H ? .P n1 /Œ1
F E ˝C H ? .P n /Œ2 ! ! id ˚ F E ˝ H ? .P n2 /
˚F E ˝C H ? .P n1 /Œ1
where we use the convention for H ? .P n / from Section 3.2 (i.e. symmetric with respect
to degree zero). Note that this simplification uses some basic commutation relations
between Es and F s (Lemma 4.2 of [CKL3]) which follow formally from the relations
in Section 3.2.
It is not too hard to check that the second map is surjective on summands of the
form F E while the first map is injective on n such summands. It follows that this
complex is homotopic to one of the form
F EŒn 2 ! F EŒn ! id:
The second map in this complex is unique and hence must be the adjunction map (up
to a non-zero multiple). The first map is a little harder to deduce but it turns out to be
equal to ˇI I I Iˇ where ˇ is the map in equation (4). We conclude that:
Proposition 6.8. Given a geometric categorical sl2 action with Y ./ D ; for > nC1
it follows that T .n C 1/ T .n 1/ 2 D.Y.n 1/ Y .n 1// is isomorphic to the
unique right convolution of
ˇI I I Iˇ adj
F EŒn 2 ! F EŒn ! id: (18)
Moreover, the sl2 action induces a P n kernel P WD F .n 1/ 2 D.X Y / where

Y D Y.n 1/ and X D Y .n C 1/ such that the induced P n twist TP is isomorphic to
T .n C 1/ T .n 1/.
Remark 6.9. Notice that unlike the spherical functor case, a P n functor does not
induce a geometric categorical sl2 action because it does not give us the spaces Y ./
for ¤ n 1; n C 1.
One can imagine trying to compute T .n C 1/ T .n 1/ even if Y ./ is not

empty for > n C 1. Indeed, one can probably obtain some reasonable expressions
for these kernels in terms on complexes where all the terms are of the form F .s/ E .s/
for some s.
One could also try to define a Grassmannian G.k; N / object P (generalizing P n
objects). The data for this should contain spaces Y; X1 ; : : : ; Xk together with kernels
Pi 2 D.Xi Y / for i D 1; : : : ; k. This is because, imitating the construction above,
one needs spaces
Y WD Y .n 1/; X1 WD Y .n C 1/; : : : ; Xk WD Y .n 1 C 2k/
together with kernels F .i/ 2 D.Y.n 1 C 2i / Y .n 1// for i D 1; : : : ; k.
6.3 Infinite twists and some geometry. When n D 1 Proposition 6.8 states that, in
a geometric categorical sl2 action, if Y ./ D ; for > 2 then T 2 2 D.Y .0/ Y .0//
is the right convolution of the complex (18). Hence T 2 is given by the adjoint which
is the left convolution of the complex
adj ˇI I I Iˇ
id ! F EŒ1 ! F EŒ3:
More generally, if you look at T 2` 2 D.Y .0/ Y .0// then a little bit of work shows
that it is given as the unique left convolution of
adj ˇI I I Iˇ ˇI I CI Iˇ ˇI I I Iˇ
id ! F EŒ1 ! ! F EŒ2`1 ! F EŒ2`C1:
If we let ` ! 1 then this complex converges to
adj
id ! F EŒ1 ! F EŒ3 ! ! F EŒ2` 1 ! F EŒ2` C 1 ! (19)
where the differentials after the left hand adjunction map alternate between .ˇI I
I Iˇ/ and .ˇI I I Iˇ/. Here we say that a sequence of complexes converges
if it eventually stabilizes in any given degree (see, for instance, Section 3 in [Roz] for
more details). We denote the left convolution of (19) by T 1 .
The object T 1 lives naturally in D .Y .0/ Y .0// which is the bounded above
derived category of coherent sheaves. This might seem strange since the complex (19)
is bounded below. However, F E is some bounded complex and F EŒ2` 1, when
you perform the left convolution, is shifted by Œ2` 1 ` D Œ` 1 so as ` ! 1 this
is shifted lower and lower in cohomology which explains why it belongs to D and
not D C .
Now consider the geometric categorical sl2 action on Y .0/ WD T ? P 1 where Y .2/ D
Y.2/ are points and E 2 D.Y .0/ pt/ is given by the twisted zero section OP 1 .1/
and the same with F 2 D.pt Y .0//. The map ˇ W E ! EŒ2 is the unique map in
Ext2T ? P 1 .OP 1 .1/; OP 1 .1//. Now
F E Š OP1 .1/ OP1 .1/ 2 D.T ? P 1 T ? P 1 /:

90 S. Cautis
This means that T 1 2 D .T ? P 1 T ? P 1 / is a complex whose cohomology

H i .T 1 / is
• isomorphic to OP1 .1/ OP1 .1/ if i < 0,
• a sheaf supported on P 1 P 1 [ T ? P 1 T ? P 1 if i D 0,
• 0 if i > 0.
Recall the map p.1; 2/ W T ? P 1 ! B.1; 2/. In this case B.1; 2/ is just the quadric
Q C 3 and p.1; 2/ (or p for short) is the map which collapses the zero section
inside T ? P 1 to a point. The composition p p does not preserve the bounded derived
category since Q is singular but it does preserve the bounded above derived category.
Proposition 6.10. The composition p p W D .T ? P 1 / ! D .T ? P 1 / is induced by

the kernel T 1 2 D .T ? P 1 T ? P 1 /.
Proof. The pushforward and pullback maps from T ? P 1 to Q are given by the graph
p of p which is the kernel Op 2 D.T ? P 1 Q/. We denote by K the convolution
Op Op 2 D .T ? P 1 T ? P 1 / which induces p p .
The adjoint map p p ./ ! ./ corresponds to a natural map W K ! O . Now,
since p OP 1 .1/ D 0 it follows that
I adj
K T 1 Š Cone.K id ! K F E/ Š K:
So applying this to we get a map K ! T 1 . Repeating this and taking the limit we
obtain a morphism O W K ! T 1 . We would like to show that Cone./ O D 0. To do
this we show that it acts by zero on any object in D .T ? P 1 /.
Let M 2 D .T ? P 1 / and consider the exact triangle
adj
p p M ! M ! Cone.adj/:
Notice that p Cone.adj/ D 0 so it suffices to show that ˆCone./

O .N / for any N where
either N D p N 0 for some N 0 2 D .Q/ or p N D 0.
If N D p N 0 then
0 0 0
O .p N / Š Cone.ˆK .p N / ! ˆT 1 .N //:
ˆCone./ (20)
Now ˆK .p N 0 / Š p p p N 0 Š p N 0 and ˆT 1 .p N 0 / Š p N 0 since, by a

straightforward calculation, Ê .p N 0 / D 0. It is not hard to see that the map in (20)
0
above induces an isomorphism and hence ˆCone./ O .p N / D 0.
On the other hand, suppose p N D 0 then N . Since the fibres of p are at most
one-dimensional this means p H i .N / D 0 for any i and so we can assume N is a
sheaf. But then by Lemma 6.11 below N is a direct sum of OP 1 .1/. So it suffices to
O .OP 1 .1// D 0.
show that ˆCone./
To see this we check that ˆK .OP 1 .1// D 0 and ˆT 1 .OP 1 .1// D 0. The first
follows since p OP 1 .1/ D 0. On the other hand, it is a standard exercise to check
that ˆT 1 .OP 1 .1// Š OP 1 .1/Œ1. This means that
ˆT 2` .OP 1 .1// Š OP 1 .1//Œ2`
and hence, ˆT 1 .OP 1 .1// D 0. This completes the proof.

Lemma 6.11. If M is a coherent sheaf on T ? P 1 and p M D 0 then M Š OP 1 .1/˚` .
Proof. Suppose p M D 0. Then M is set theoretically supported on P 1 because p is
an isomorphism away from P 1 . Now any sheaf of P 1 is a direct sum of structure sheaf
and line bundles. So if M were scheme theoretically supported on P 1 T ? P 1 then
the result would follow because OP 1 .1/ is the only sheaf with vanishing cohomology.
More generally, consider the short exact sequence
0 ! M00 ! M ! M0 ! 0 (21)
where M0 is the quotient of M scheme theoretically supported on P 1 (equivalently,

M00 is the part of M killed by f where f is the local equation defining P 1 T ? P 1 ).
It would suffice to show p M0 D 0 since then p M00 D 0 and we can proceed up
induction to conclude that M00 and M 0 are both direct sums of OP 1 .1/ and then the
same must be true of M since Ext1 .OP 1 .1/; OP 1 .1// D 0.
We now prove that p M0 D 0. From the long exact sequence induced by (21) and
the fact that Q is affine it suffices to show that H 0 .M0 / D 0. On the other hand, take
the standard short exact sequence
0 ! OT ? P 1 .P 1 / ! OT ? P 1 ! OP 1 ! 0
and tensor it with M. Then H 0 .M ˝ OP 1 / Š M0 so from the long exact sequence it

suffices to show that H 1 .OT ? P 1 .P 1 / ˝ M/ D 0. But OT ? P 1 .P 1 / Š OP 1 .2/
where W T ? P 1 ! P 1 is the standard projection. Then, by the projection formula
.OT ? P 1 .P 1 / ˝ M/ Š . OP 1 .2/ ˝ M/ Š OP 1 .2/ ˝ .M/:
Now M on P 1 has no cohomology so it must be isomorphic to OP 1 .1/˚` for some

`. Subsequently OP 1 .2/ ˝ .M/ Š OP 1 .1/˚` has no higher cohomology and hence
H 1 .OT ? P 1 .P 1 / ˝ M/ D 0.
Remark 6.12. There are two points worth noting here. First is that the limit T 1
is a projector (i.e. T 1 T 1 Š T 1 ) even though T ` is invertible for any `.
Secondly, the kernel T 1 , which is defined using a formal construction not involving
p, has a geometric description as the kernel inducing p p .
More generally, one can consider the cotangent bundle to the full flag variety
T ? Fl.C N / where there is a straightforward generalization of the discussion above
92 S. Cautis
(which was the case N D 2). From Section 5.3.2, one can construct a categorical slN
action so that D.T ? Fl.C N // corresponds to the zero weight space. This induces, a
braid group action on D.T ? Fl.C N // generated by kernels Ti 1 WD Cone.id ! Fi Ei /
for i D 1; : : : ; N 1.
Then the arguments above can be used to show that Ti 1 WD lim`!1 Ti 2` is well
defined. Moreover, Ti 1 is isomorphic to the kernel which induces pi pi where pi
is the projection from T ? Fl.C N / given by forgetting Vi .
6.3.1 Non-generalizations. Consider again the situation in Proposition 6.8 where

n 0 is now arbitrary. This time T 2` 2 D.Y .n 1/ Y .n 1// is isomorphic to
the left convolution of the complex
id ! F EŒn ! F EŒn C 2 ! F EŒ3n C 2 ! F EŒ3n C 4

! ! F EŒ.2` 1/.n C 1/ 1 ! F EŒ.2` 1/.n C 1/ C 1
P after the left-most adjunction map, the maps alternate between .ˇI I I Iˇ/
where,
and niD0 .ˇ i I I C I Iˇ ni /. This complex also has an obvious limit as ` ! 1
which we denote T 1 2 D .Y .n 1/ Y .n 1//. This is completely analogous to
the case n D 1 discussed above.
Now suppose Y ./ D T ? G.k; n C 1/ (where D n C 1 2k) so that Y .n 1/ D
T P . As before we have the map p.1; nC1/ W T ? P n ! B.1; n C 1/ which collapses
? n
the zero section. However, if n > 1, then T 1 6Š K where K is the kernel inducing
the map
p.1; n C 1/ p.1; n C 1/ W D .T ? P n / ! D .T ? P n /:
It turns out K is a stronger projection than T 1 . In other words,
K T 1 Š K Š T 1 K 2 D .T ? P n T ? P n /:
The argument used to prove Proposition 6.10 fails because the kernel of the map
p.1; n C 1/ is now larger (and more complicated) than the kernel of the map T 1 .
Geometrically, this difference seems to be related to the fact that the singular and
intersection cohomologies of B.1; n C 1/ are the same if n D 1 but different for
n > 1. More precisely, K is akin to singular and T 1 to intersection cohomology.
In [C2] we use T 1 (rather than the geometric kernel K) to categorify Reshetikhin–
Turaev knot invariants. This suggests that T 1 is at least as natural as K. However,
this also begs the obvious question: what is the geometric interpretation of the kernel
T 1 2 D .T ? P n T ? P n / when n > 1?
7 The general flop – a discussion

7.1 The Mukai flop. We recall the definition of a general Mukai flop (see for example
[Nam1], Section 4). Let Y be a smooth variety of dimension 2n which contains a
subvariety X Y isomorphic to P n so that NX=Y Š T ? P n . If Y is holomorphic

symplectic then this second condition is automatically satisfied.
Then one can blow up Y and then blow down to obtain another variety Y C which
also contains a subvariety X C Š P n with NX C =Y C Š T ? P n . Moreover, there exists
maps Y ! Yx and Y C ! Yx to a common (singular) variety Yx which collapses X and
X C to a point but are isomorphisms away from X and X C . To summarize, we have
the following diagram:
Y Yx Y C
HH
1 vvv HH2
vv HH
vv HH
zvv H$
Y II Y
II uuu
II u
II uu
II uuu
$ zu
Yx .
Of course, this generalizes our example above where Y D T ? P n . As before, Y Yx Y C
is a variety with two equidimensional components. Namikawa proves:
Proposition 7.1 ([Nam1], Section 4). There exists an isomorphism
2 1 W D.Y /

! D.Y C /:
Namikawa first checks this isomorphism for the local case of Y D T ? P n . He
then uses the fact that the formal neighbourhoods of X Y and P n T ? P n are
isomorphic to prove the more general case above.
7.2 The stratified Mukai flop of type A. An abstract definition of a stratified Mukai
flop was first discussed by Markman in [M] while studying the geometry of the moduli
spaces of sheaves on K3 surfaces. The idea is to imitate the geometry of the situation
in diagram (2). More specifically, there is a filtration
T ? G.k; N / T ? G.k; N /1 T ? G.k; N /k
where T ? G.k; N /i is the subvariety
k N k
f.X; V / W X 2 End.C N /; 0 !
V ! C N ; X C N V
and X V 0; dim.ker X / N k C i g
and the projection map p.k; N / W T ? G.k; N / ! B.k; N / restricted to T ? G.k; N /i n

T ? G.k; N /iC1 is a G.i; N k C i /-fibration. In this paper we will use the definition
from [FW], which is similar to that in [M] but fits better for our discussion.
Let Y and Y C be two smooth varieties equipped with two collections of closed
subvarieties
Xk X1 Y and XkC X1C Y C :
94 S. Cautis
p pC
Now assume there are two birational maps Y ! Yx Y C and denote by f W Y Ü
C
Y the induced birational map. Then this data describes a stratified Mukai flop of type
AN;k (where 2k N ) if the following conditions hold.
! Y C n X1C .
• f induces an isomorphism Y n X1
• p.Xj / D p C .XjC / for j D 1; : : : ; k. We write Sj WD p.Xj /.
• Sk is smooth and pjXk W Xk ! Sk is isomorphic to the projection map
G.k; W / ! Sk where W is some N -dimensional vector bundle on Sk and
G.k; W / denotes the relative Grassmannian of k-planes. Moreover, the normal
bundle NXk =Y is isomorphic to the relative cotangent bundle TX?k =Sk . The same
thing holds for p C jX C W XkC ! Sk with W replaced by W _ .
k
• If k D 1 this should be the usual Mukai flop. If k 2 let BlXk Y , BlX C Y C

k
and BlSk Yx denote the blowups of Y; Y C and Yx in Xk ; XkC and Sk respectively.
Then the proper transforms of all Xj and XjC together with the birational maps
BlXk Y ! BlSk Yx BlX C Y C must describe a stratified Mukai flop of type
k
AN 2;k1 .
Remark 7.2. Given just the AN;k contraction p W Y ! Yx it follows (just like for the
usual Mukai flop) that the corresponding stratified Mukai flop exists. For a proof see,
for instance, Proposition 2.1 of [FW].
Of course, one would like an analogue of Proposition 7.1 in the spirit of Theorem 4.2.
This theorem would identify an open subset inside Y Yx Y C which is an analogue of
Z o .k; N / and a line bundle on it so that the pushforward of this line bundle is a kernel
which induces a derived equivalence D.Y / ! D.Y C /.
One possible approach to proving this equivalence is to deform to the normal cone.
This means looking at Y A1 and blowing up Xk f0g Y (and likewise with Y C ).
It is shown in Section 5 of [FW] that the degeneration of Y Yx Y C to the normal cone
BlXk f0g .Y A1 / A1 BlX C f0g .Y C A1 /

k
breaks up into the correspondence Z.k; N / (Section 4.2.1) for the local version of
the stratified Mukai flop and into the correspondence BlXk Y BlS Yx BlX C Y C for a
k k
stratified Mukai flop of type AN 2;k1 . Then using Theorem 4.2 and induction one
can imagine proving the equivalence on the special fibre (the fibre over 0 2 A1 ). Since
a kernel inducing an equivalence fibre-wise is an open condition this would then imply
the equivalence on the general fibre too.
7.3 The stratified Mukai flop of type D. We now briefly discuss the stratified Mukai
flop of type D2mC1 .
7.3.1 The local model. Fix a symmetric, non-degenerate bilinear form h; i on C 2N .
Denote by IG.k; 2N / the isotropic Grassmannian parametrizing isotropic k-planes in
C 2N . When k D N it turns out IG.N; 2N / has two components denoted IG.N; 2N /
and IG.N; 2N /C . Two isotropic planes V; V 0 C 2N belong to the same component
if and only if dim.V \ V 0 / N mod 2.
The cotangent bundles of IG.N; 2N /˙ can be described as
T ? IG.N; 2N /˙ D f.X; V / 2 so.2N / IG.N; 2N /˙ W X.C 2N / V; X.V / 0g
where X 2 so.2N / is a skew-symmetric matrix meaning that hXv; wi D hv; Xwi.
Now consider the map
ip W T ? IG.N; 2N / ! IB.N; 2N /
given by forgetting V where IB.N; 2N / WD fX 2 so.2N / W X 2 D 0g. A general point
X 2 IB.N; 2N / has dim.ker X / equal to N or N 1 depending on whether N is even
or odd. This essentially comes down to the fact that a skew-symmetric matrix of size
N has rank at most N 1 if N is odd but can have full rank if N is even.
So there are two cases to consider. If N is even then IB.N; 2N / has two com-
ponents. Two general points X1 ; X2 2 IB.N; 2N / lie in the same component if and
only if X1 .C 2N / \ X2 .C 2N / is even. Then one component has a resolution given by
T ? IG.N; 2N / and the other component has a resolution given by T ? IG.N; 2N /C .
If N is odd then IB.N; 2N / only has one component. A resolution of this component
is the variety
2
V ? ! C 2N ; X.C 2N / V; X.V ? / 0g
f.X; V / W 0 ! V ! (22)
where X 2 so.2N / and V 2 IG.N 1; 2N /. On the other hand, there is a natural
map from
N 1 1 1 N 1
f.X; V; V 0 / W 0 ! V !
V0!
V ? ! C 2N ; X.C 2N / V; X.V ? / 0g (23)
to (22). This map is everywhere 2 W 1 since the fibres are all isomorphic to IG.1; 2/
using the restriction of h; i to V ? =V and IG.1; 2/ is the disjoint union of two points.
Forgetting V and V ? we get a generically one-to-one map from (23) to the two con-
nected components in T ? IG.N; 2N /. Thus if N is odd we get the following diagram
(in analogy with (2)):
T IG.N; 2N / T IG.N; 2N /C (24)

PPP nn
PPP nnn
PPP nnn
PPP n C
ip
( vnnn ip
IB.N; 2N /.
This is the local model for the stratified Mukai flop of type D2mC1 where N D 2mC1.
We equip everything with the C -action acting on the fibres of T ? IG.N; 2N /˙ just
like in the case of T ? G.k; N /.
96 S. Cautis
Remark 7.3. The type A Grassmannian G.k; N / corresponds to the minuscule GL.N /
representation ƒk .C N /. In the case of DN there are three minuscule representations.
One of them corresponds to IG.1; 2N / while the other two correspond to IG.N; 2N /
and IG.N; 2N /C . When N is even the latter two representations are self dual but when
N is odd they are dual to each other. This is the representation theoretic manifestation
of the dichotomy above.
0 IExample.
Let us briefly examine IB.1; 2/ and IB.2;4/. We
fix the bilinear form
I 0 on C 2N
and write a general element of so.2N / as A B where A; B; C; D are
C D
N N matrices. The condition that it be skew-symmetric translates into
A C Dt D 0 and B C Bt D 0 D C C C t
while the condition that it squares to zero is equivalent to
A2 C BC D 0; AB D BAt and CA D At C:
Now, if N D 1 then the first condition implies that B D 0 D C and D D A and the
second condition says A D 0 so that IB.1; 2/ consists of just a point.
If N D 2 then an elementary calculation (which we omit) that IB.2; 4/ has
0shows
two possible types of solutions. The first is of the form A0 A where A is a 2 2
matrix with det.A/ D 0 D tr.A/ (i.e. a 2-dimensional quadric cone). The second
solution is of the form 0 1
u 0 0 x
B 0 u x 0 C
B C
@ 0 y u 0 A
y 0 0 u
where u2 D xy. So IB.2; 4/ is the union of two 2-dimensional quadric cones which
intersect only at their apex.
7.3.2 The general model. Fix N D 2m C 1 from now on. The varieties
T ? IG.N; 2N /˙
have a natural filtration
T ? IG.N; 2N /˙ T ? IG.N; 2N /˙ ˙
1 T IG.N; 2N /m
?
where T ? IG.N; 2N /˙ i corresponds to the locus where dim.ker X / N C 1 C

2i. As before, we denote the image of T ? IG.N; 2N /˙ i by Si . The subvariety
T ? IG.N; 2N /˙
i n T ?
IG.N; 2N /˙
iC1 consists of the locus where dim.ker X / D N C
1 C 2i and hence is isomorphic to
˚ N 12i 2iC1 2iC1 ? N 12i
.X; V; V 0 / W 0 ! V 0 ! V ! V 0 ! C 2N ;
?
X.C 2N / D V 0 ; X.V 0 / D 0
since V 0 can be recovered as X.C 2N /. Restricting h; i to V 0 ? =V 0 we find that

V =V 0 is isotropic inside V 0 ? =V 0 . Thus the restriction of ip˙ to T ? IG.N; 2N /˙ i n
T ? IG.N; 2N /˙ iC1 is a IG.2i C 1; 4i C 2/ ˙
-fibration onto its image S i n S i1 .
Motivated by this structure, one can define a stratified Mukai flop of type D2mC1
just like in the type A case. In other words, one has subvarieties

Xm X1 Y C
and Xm X1C Y C
ip ipC
and maps Y ! Yx Y C satisfying the same conditions as before. The difference
is that W is now a rank 2N vector bundle equipped with a fibre-wise non-degenerate,
symmetric bilinear form and ip˙ jXm ! Sm is IG.N; W /˙ ! Sm which is the relative
isotropic Grassmannian.
7.4 Equivalences in type D. Once again we can consider the fibre product
IZ.N / WD T ? IG.N; 2N / IB.N;2N / T ? IG.N; 2N /C
but, as before, we cannot expect OIZ.N / to induce an equivalence. On the other hand,
IZ.N / is made up of m C 1 components IZ0 .N /; : : : ; IZm .N / of dimension N.N 1/
where
IZs .N / D f.X; V; V 0 / W dim.ker X / N C 2s C 1 and dim.V \ V 0 / 2.m s/;

X.C 2N / V; X.C 2N / V 0 ; X.V / 0; X.V 0 / 0g
inside T ? IG.N; 2N / T ? IG.N; 2N /C . One can define open subvarieties IZos .N /

IZs .N / by imposing the additional condition
dim.ker X / C dim.V \ V 0 / 2N C 2
just as we did to define Zso .k; N /. It is not difficult to check that the open subvariety
inside IZos .N / \ IZosC1 .N / given by the condition
dim.ker X / D N C 2s C 3 and dim.V \ V 0 / D 2.m s/
is co-dimension one inside IZs .N / and IZsC1 .N /. This is completely analogous to

the situation in type A. So one should strongly expect an analogue of Theorem 4.2.
Conjecture 7.4. There exists a C -equivariant line bundles L.N / on IZo .N / such
that i j L.N / induces an equivalence
D.T ? IG.N; 2N / /

! D.T ? IG.N; 2N /C /:
Here i and j are the natural inclusions

j i
IZo .N / ! T ? IG.N; 2N / T ? IG.N; 2N /C :
IZ.N / !
98 S. Cautis
Since the cohomology of IG.k; 2N / is fairly different than that of G.k; N / it does
not seem possible to construct a categorical sl2 action on cotangent bundles to isotropic
Grassmannians. Nevertheless, one can imagine that some sort of action still exists. Fi-
nally, note that T ? IG.N; 2N /˙ also have natural one-parameter deformations, defined
just like in the type A case. This leads to a stratified Atiyah flop of type D. One can
also conjecture and study derived equivalences in this case.
Remark 7.5. There are also stratified flops of type E which show up naturally in the
birational geometry of resolutions of nilpotent orbit closures. See, for instance, [CF]
for a description of these. Most questions mentioned above in the case of type D flops
also remain valid for type E.
8 Further topics
Deformation quantization. The category of D-modules on G.k; N / can be deformed
to the category of coherent sheaves on T ? G.k; N /. The specialization map from D-
modules to coherent sheaves is given by taking the associated graded. Now consider
the open subset
j W U ,! G.k; N / G.N k; N /
defined as the locus .V; V 0 / where V \V 0 D 0. It turns out that the push-forward j OU
of the D-module OU is a D-module on G.k; N / G.N k; N / which induces an
equivalence Dmod.G.k; N // ! Dmod.G.N k; N //. In [CDK] we check that the
associated kernel of this equivalence is actually the kernel T .k; N / 2 D.T ? G.k; N /
T ? G.N k; N //.
Remark 8.1. Calculating the associated graded of a D-module is quite difficult in

general. In [CDK] we compute the associated
F graded by first constructing a categorical
sl2 action on categories of D-modules on k G.k; N / and then showing that it agrees,
via the associated graded map, with the categorical sl2 action on coherent sheaves on
F
k T G.k; N /. So this approach does not really give an entirely different proof that
?
T .k; N / is invertible.
The category of D-modules on G.k; N / is an example of a deformation quantization

of the category of coherent sheaves on T ? G.k; N /. But deformation quantizations also
exists for category of coherent sheaves on quiver varieties, for instance. Understanding
how these categorical Lie algebra actions and the corresponding equivalences deform
to these deformation quantizations is a little explored but interesting problem.
Flops as moduli spaces. Bridgeland [B] describes a way to construct the Atiyah flop
Y C of a 3-fold p W Y ! Yx as the moduli of perverse coherent sheaves on Y (the
definition of these perverse sheaves uses the map p). Then the universal family over
the product Y Y C induces the derived equivalence D.Y / ! D.Y C /.
Can you generalize this result to other Atiyah or Mukai flops? This question seems
D
difficult (but also interesting) in part because, as we saw in Section 4.4.3, the auto-
equivalence of D.T ? G.2; 4// is not induced by the structure sheaf of the natural fibre
product.
References
[Ad] N. Addington, New derived symmetries of some Hyperkähler varieties. Preprint,
arXiv:1112.0487v1 [math.AG].
[An] R. Anno, Spherical functors. Preprint, arXiv:0711.4409 [math.CT].
[B] T. Bridgeland, Flops and derived categories. Invent. Math. 147 (2002) no. 3, 613–632.
[BR] R. Bezrukavnikov and S. Riche, Affine braid group actions on derived categories of
Springer resolutions. Preprint, arXiv:1101.3702 [math.RT].
[C1] S. Cautis, Equivalences and stratified flops. Compositio Math. 148 (2012), no. 1,
185–208.
[C2] S. Cautis, Clasp technology to knot homology via the affine Grassmannian. Preprint,
arXiv:1207.2074v1 [math.RT].
[CDK] S. Cautis, C. Dodd and J. Kamnitzer, Categorical actions on quiver varieties: from
D-modules to coherent sheaves. In preparation.
[CK1] S. Cautis and J. Kamnitzer, Knot homology via derived categories of coherent sheaves
I, sl(2) case. Duke Math. J. 142 (2008), no. 3, 511–588.
[CK2] S. Cautis and J. Kamnitzer, Knot homology via derived categories of coherent sheaves
II, sl(m) case. Invent. Math. 174 (2008), no. 1, 165–232.
[CK3] S. Cautis and J. Kamnitzer, Braiding via geometric Lie algebra actions. Compositio
Math. 148 (2012), no. 2, 464–506.
[CKL1] S. Cautis, J. Kamnitzer, and A. Licata, Categorical geometric skew Howe duality.
Invent. Math. 180 (2010), no. 1, 111–159.
[CKL2] S. Cautis, J. Kamnitzer, and A. Licata, Coherent sheaves and categorical sl2 actions.
Duke Math. J. 154 (2010), no. 1, 135–179.
[CKL3] S. Cautis, J. Kamnitzer, and A. Licata, Derived equivalences for cotangent bundles
of Grassmannians via categorical sl2 actions. J. Reine Angew. Math., to appear,
Doi 10.1515/CRELLE.2011.184; preprint, arXiv:0902.1797 [math.AG].
[CKL4] S. Cautis, J. Kamnitzer, and A. Licata, Coherent sheaves on quiver varieties and cate-
gorification. Preprint, arXiv:1104.0352 [math.AG].
[CF] P. E. Chaput and B. Fu, On stratified Mukai flops. Math. Res. Lett. 14 (2007), no. 6,
1055–1067.
[CG] N. Chriss and V. Ginzburg, Representation theory and complex geometry. Birkäuser,
Boston 1997.
[CR] J. Chuang and R. Rouquier, Derived equivalences for symmetric groups and sl2 -
categorification. Ann. of Math. 167 (2008), no. 1, 245–298.
100 S. Cautis
[F] B. Fu, Extremal contractions, stratified Mukai flops and Springer maps. Adv. Math. 213
(2007), 165–182.
[FW] B. Fu and C.-L. Wang, Motivic and quantum invariance under stratified Mukai flops.
J. Differential Geom. 80 (2008), 261–280.
[Ho] R. P. Horja, Derived category automorphisms from mirror symmetry. Duke Math. J.
127 (2005), 1–34.
[HT] D. Huybrechts and R. Thomas, P -objects and autoequivalences of derived categories.
Math. Res. Lett. 13 (2006), no. 1, 87–98.
[K1] Y. Kawamata, D-equivalence and K-equivalence. J. Differential Geom. 61 (2002),
147–171.
[K2] Y. Kawamata, Derived equivalence for stratified Mukai flop on G.2; 4/. In Mirror
symmetry. V, AMS/IP Stud. Adv. Math. 38, Amer. Math. Soc., Providence, RI, 2006,
285–294.
[KL1] M. Khovanov and A. Lauda, A diagrammatic approach to categorification of quantum
groups I. Represent. Theory 13 (2009), 309–347.
groups II. Trans. Amer. Math. Soc. 363 (2011), 2685–2700.
groups III. Quantum Topol. 1, Issue 1 (2010), 1–92.
[KT] M. Khovanov and R. Thomas, Braid cobordisms, triangulated categories, and flag
varieties. Homology, Homotopy, Appl. 9 (2007), 19–94.
[L] A. Lauda, A categorification of quantum sl2 . Adv. Math. 225 (2010), no. 6, 3327–3424.
[M] E. Markman, Brill-Noether duality for moduli spaces of sheaves on K3 surfaces. J.
Algebraic Geom. 10 (2001), 623–694.
[Nak1] H. Nakajima, Quiver varieties and Kac–Moody algebras. Duke Math. J. 91 (1998),
no. 3, 515–560.
[Nak2] H. Nakajima, Quiver varieties and finite-dimensional representations of quantum affine
algebras. J. Amer. Math. Soc. 14 (2001), no.1, 145–238.
[Nam1] Y. Namikawa, Mukai flops and derived categories. J. Reine Angew. Math. 560 (2003),
65–76.
[Nam2] Y. Namikawa, Mukai flops and derived categories II. In Algebraic structures and moduli
spaces, CRM Proc. Lecture Notes 38, Amer. Math. Soc., Providence, RI, 2004, 149–
175.
[Nam3] Y. Namikawa, Birational geometry of symplectic resolutions of nilpotent orbits. In Mod-
uli spaces and arithmetic geometry, Adv. Stud. Pure Math. 45, Mathematical Society
of Japan, Tokyo, 2006, 75–116.
[Ric] S. Riche, Geometric braid group action on derived category of coherent sheaves. Rep-
resent. Theory 12 (2008), 131–169.
[Ro1] R. Rouquier, Categorification of sl2 and braid groups. In Trends in representation the-
ory of algebras and related topics, Contemp. Math. 406, Amer. Math. Soc., Providence,
RI, 2006, 137–167.
[Ro2] R. Rouquier, 2-Kac–Moody algebras. Preprint, arXiv:0812.5023 [math.RT].

[Roz] L. Rozansky, An infinite torus braid yeilds a categorified Jones-Wenzl projector.
Preprint, arXiv:1005.3266v1 [math.GT].
[ST] P. Seidel and R. Thomas, Braid group actions on derived categories of coherent sheaves.
Duke Math J. 108, (2001), 37–108.
A note on derived categories of Fermat varieties
Akira Ishii and Kazushi Ueda
1 Introduction
The Fermat variety of degree m and dimension n is defined by
Xm
n
D fŒx1 W W xnC2 2 P nC1 j x1m C C xnC2
m
D 0g;
which can be constructed from Xm r

for r < n by taking direct products, blow-ups,
quotients by cyclic groups and blow-downs, see Theorem I in [SK79]. It follows that
any Fermat variety can be constructed inductively from ‘Fermat points’ Xm 0
and Fermat
curves Xm1
. The number of Fq -rational points on Xm n
can be counted by Jacobi sum,
which motivated Weil to propose his conjectures [Wei49]. Fermat varieties have also
provided a testing ground for the Hodge conjecture and the Tate conjecture [Shi79b],
[Shi79a], [Ran81].
The Fermat variety Xm n
has a natural action of the abelian group
Gm
n
D fŒdiag.a1 ; : : : ; anC2 / 2 PGLnC2 .C/ j a1d D D anC2
d
D 1g;
and let Xnm D ŒXm

n
=Gm
n
be the quotient stack. We prove the following in this paper:
Theorem 1.1. Xnm has a full strong exceptional collection consisting of invertible
sheaves.
King conjectured that a smooth complete toric variety has a full strong exceptional
collection consisting of invertible sheaves [Kin97]. It is true for toric weak Fano
stacks in dimension two [BH09], [IU], but fails for toric surfaces [HP06] and higher-
dimensional toric Fano varieties [Efi10]. The existence of full exceptional collections
on toric varieties consisting of sheaves is proved by Kawamata [Kaw06]. The case
n D 1 in Theorem 1.1 is contained in [GL87], which gives a full strong exceptional
collection of invertible sheaves on any rational orbifold curve. The existence of a
full exceptional collection on Xnm follows from [Orl09], Theorem 2.5, and [FU11],
Theorem 1.2. The existence of a full exceptional collection consisting of sheaves
follows from the fact that Xnm can be obtained from P n by iterated root constructions.
On the other hand, if m is greater than n C 1, then the canonical divisor of Xm
n
is free,
so that D coh Xm has no exceptional object at all.
b n
Let D b coh Xnm be the pretriangulated dg category defined in Section 3 whose

cohomology category is equivalent to the derived category D b coh Xnm of coherent

Supported by Grant-in-Aid for Scientific Research (No.18540034).

Supported by Grant-in-Aid for Young Scientists (No.20740037).
104 A. Ishii and K. Ueda
sheaves on Xnm . In other words, D b coh Xnm gives an enhancement of D b coh Xnm in the
sense of Bondal and Kapranov [BK90]. Let further D b coh Xm n
be the enhancement of
D coh Xm defined similarly. There is a coherent action of the group Gm
b n n_
of characters
of Gm on D coh Xm , and one can show the following:
n b n
Theorem 1.2. The pretriangulated dg category D b coh Xm

n
is equivalent to the idem-
potent completion of the orbit category of D coh Xm with respect to the coherent
b n
action of the group Gm

n_
.
The idea to use dg categories to study the triangulated structure on the orbit category
is due to Keller [Kel05]. Theorems 1.1 and 1.2 show that one can study the derived
category of a Fermat variety in terms of a coherent action of a finite abelian group on an
enhancement of a triangulated category which has a full strong exceptional collection.
2 A full strong exceptional collection on Xnm

We prove Theorem 1.1 in this section. The stack X D Xnm is the closed substack of the
toric stack P D Œ.C nC2 n f0g/=K; where
K D f.˛1 ; : : : ; ˛nC2 / 2 .C /nC2 j ˛1m D D ˛nC2

m
g:
An element aE of the group
c =.mxEi cE/nC2
L D ZxE1 ˚ ˚ ZxEnC2 ˚ ZE iD1
of characters of K gives an invertible sheaf OP .Ea/ on P , which restricts to an invertible

sheaf OX .E
a/ on X. The canonical bundles are given by !P D OP .xE1 xEnC2 / and
!X D OX .xE1 xEnC2 C cE/: Since P is a toric Fano stack of Picard number one,
the cohomology H i .OP .E a// of an invertible sheaf is non-trivial only if i D 0; n C 1.
The long exact sequence associated with
0 ! OP .E
a cE/ ! OP .E
a/ ! OX .E
a/ ! 0
implies that H i .OX .E

a// is zero if i ¤ 0; n. Serre duality gives
a/; OX / D .H 0 .OX .!E C aE ///_

Extn .OX .E
where !E D xE1 xEnC2 C cE: Any aE 2 L can be presented uniquely as
aE D a1 xE1 C C anC2 xEnC2 C e cE
where 0 ai m 1 and e 2 Z. One has Extn .OX .E

a/; OX / ¤ 0 if and only if
!E C aE D .a1 1/xE1 C C .anC2 1/xEnC2 C .e C 1/E

c
can be presented in such a way that all the coefficients are positive. This happens if
Derived categories of Fermat varieties 105
• 0 e n and at least n e C 1 of ai are greater than or equal to one.

Define a finite subset of L by
A D fai1 xEi1 C C aik xEik C e cE j 0 k n; 1 i1 < < ik n C 2;

0 e n kg;
where 0 < ai` m 1 for ` D 1; : : : ; k. One can see that Exti .OX .E E D0
a/; OX .b//
E
for aE ; b 2 A and i ¤ 0, so that .OX .E a//aE 2A is a strong exceptional collection.
Let pi D Œx1 W W xnC2 be the point such that xi D 1 and xj D 0 for i ¤ j . An
exact sequence
M M X
0 ! OP ! OP .xEk / ! OP xEk C xEl ! ! OP xEk ! Opi ! 0
k¤i i¤k;` k¤i
of OP -modules restricts to an exact sequence

M M X
0 ! OX ! OX .xEk / ! OX xEk C xEl ! ! OX xEk ! 0 (2.1)
k¤i i¤k;` k¤i
of OX -modules. By using these sequences and their translates, the fact that one can
construct any invertible sheaf from .OX .E
a//aE 2A by taking shifts and cones is reduced
to the following lemma:
Lemma 2.1. Let B be the subset of L containing A satisfying the following for any
0 i n and any aE 2 L:
P
• Let P be the power set of f1; : : : ; n C 2g n fi g. If P
B contains aE C k2K xEk for
any K 2 P except K0 , then B also contains aE C k2K0 xEk .
Then B coincides with the whole of L.
Proof. We use the following two extreme cases:
P
(1) If B contains aE C k2K xEk for all P subset K strictly contained in f1; : : : ;
n C 2g n fig, then B also contains aE C k¤i xEk .
P
(2) If B contains aE C k2K xEk for any non-empty subset K f1; : : : ; n C 2g n fi g,
then B also contains aE .
If we replace cE by mxEi for suitable i , we can write
A D fa1 xE1 C C anC2 xEnC2 j 0 ai m; at least two of ai ’s are zerog:
We show that vE WD a1 xE1 C C anC2 xEnC2 2 B for all .a1 ; : : : ; anC2 / 2 ZnC2 .
We first consider the case ai 0 for all i . In this case, we may assume either
0 ai m for all i or 0 < ai for all i . In the former case, we may assume at most
P
one of ai ’s is zero and we can show vE 2 B by the induction on i ai , using (1). Then
the same
P induction also shows the latter case.
If i ai m.n C 2/, then we Pmay assume ai 0 for all i and obtain vE 2 B.
Finally, the reverse induction on i ai and (2) prove vE 2 B for all .a1 ; : : : ; anC2 /.
Let T be the full triangulated subcategory of D b coh X generated by .OX .E a//aE 2A .

Since the condition that Hom.OX .E a/; E/ D 0 for a coherent sheaf E and any aE 2 L
implies E Š 0, the right semiorthogonal complement T ? is trivial so that the collection
.OX .E
a//aE 2A is full.
Remark 2.2. Let m D .m1 ; : : : ; mnC2 / be a sequence of positive integers and Xm D

ŒZm =Km be the quotient stack of
˚ mnC2
Zm D .x1 ; : : : ; xnC2 / 2 C nC2 n f0g j x1m1 C C xnC2 D0
by ˚ mnC2
Km D .˛1 ; : : : ; ˛nC2 / 2 .C /nC2 j ˛1m1 D D ˛nC2 :
Then it is straightforward to generalize the proof of Theorem 1.1 above to show that
Xm has a full strong exceptional collection consisting of invertible sheaves.
3 Coherent actions and orbit categories

Let G be a group and C be a category. A coherent action .Fg ; ˛g;h / of G on C consists
of endofunctors Fg W C ! C for each g 2 G together with natural isomorphisms
˛g;h W Fg B Fh ! Fgh of endofunctors for g; h 2 G such that
• Fe is the identity functor of C ,
• ˛g;h is the identity if g D e or h D e, and
• the diagram
˛g;h B Fk
Fg B Fh B Fk / Fgh B Fk
Fg B ˛h;k ˛gh;k
˛g;hk
Fg B Fhk / Fghk
is commutative.
For a coherent action .Fg ; ˛g;h / of a finite group G on an additive category C , the
corresponding orbit category C =G is defined as follows:
• An object of C =G is an object of C .
• The space of morphisms from x to y is given by

M
HomC=G .x; y/ D HomC .x; Fg .y//:
g2G
• The composition of ' 2 HomC .x; Fg .y// and 2 HomC .y; Fh .z// is given
by
' Fg . / ˛g;h .z/
x!
Fg .y/ ! Fg B Fh .z/ ! Fgh .z/:
Now we recall the notion of enhanced triangulated categories by Bondal and Kapranov
[BK90]. See [Kel06] and references therein for basic definitions and results on dg
categories. A dg category D is an additive category such that the spaces of morphisms
are cochain complexes, the identity morphisms are cocycles, and the compositions
satisfy the Leibniz rule. The cohomology category H 0 .D / has the same objects as D
and the spaces of morphisms are zero-th cohomology groups. A twisted complex over
D is a collection f.Ei /i2Z ; .qij W Ei ! Ej /i;j 2Z g of objects Ei of D and morphisms
qij of degree i Pj C 1 such that Ei D 0 for almost all i 2 Z and the Maurer–Cartan
equation dqij C k qkj qik D 0 is satisfied for any i; j 2 Z. We always assume that
a twisted complex is one-sided in the sense that qij D 0 for i j . Twisted complexes
form a dg category T w.D /, equipped with a full and faithful functor W D ! T w.D /
sending an object E to the twisted complex f.Ei /i2Z ; .qij /i;j 2Z g such that Ei D 0
for i ¤ 0, E0 D E and qij D 0 for any i; j 2 Z. A dg module over a dg category
D is a dg functor from D to the dg category of chain complexes of C-vector spaces.
Dg modules over a dg category D form a dg category mod.D /, and D has the Yoneda
embedding into mod.D /. The idempotent completion of D is defined as the smallest
full subcategory of mod.D / containing D and closed under direct summands. A twisted
complex X defines a dg functor homT w.D / ../; X /; which gives an object of mod.D /,
and D is said to be pretriangulated if this functor is always representable by an object
of D . The cohomology category of a pretriangulated dg category is triangulated, and
an enhanced triangulated category is a triple .T ; D ; W H 0 .D / ! T / consisting
of a triangulated category T together with a pretriangulated dg category D and an
equivalence W H 0 .D / ! T of triangulated categories.
Let W G X ! X be an action of a finite group G on an algebraic variety
X. We write the group law of G as G W G G ! G. A G-linearization of a
line bundle W L ! X on X is an action † W G L ! L of G on L such that
.†.g; l// D .g; .l// for any g 2 G and l 2 L. The space H 0 .X; L/ of sections of
L has a natural G-action defined by .g s/.x/ D †.g; s. .g 1 ; x// for s 2 H 0 .X; L/,
g 2 G and x 2 X . In terms of the invertible sheaf L associated with L, a G-
linearization is an isomorphism W L ! p2 L on G X satisfying the cocycle
condition

.p23 / B Œ.idG / D .G idX / W Œ B .idG / L ! Œp2 B p23 L
on G G X (see e.g. §3 of [MFK94]). A G-equivariant coherent sheaf on X is a pair

! p2 E satisfying the
.E; / of a coherent sheaf on X and a G-linearization W E
cocycle condition above. It follows from the definition that a G-equivariant coherent
sheaf on X is equivalent to a coherent sheaf on the quotient stack ŒX=G.
Let X be a smooth projective
S variety and G be a finite group acting on X . Choose
an affine open covering X D 2ƒ U such that ƒ has a G-action satisfying g.U / D
Ug for
Sg 2 G and 2 ƒ. Such a covering can be constructed from any affine covering
X D 2„ V by setting ƒ D G „ and U D gV for D .g; /. We introduce a
dg category vectG X as follows:
• An object of vectG X is a G-equivariant locally-free sheaf on X .
• The space of morphisms between two objects E and F is the Čech complex;
M
homivectG X .E; F / D
.U0 \ \ Ui ; E _ ˝ F /:
0 ;:::;i 2ƒ
The Čech complex has the G-action coming from the G-linearizations of E and F ,
which sends s 2
.U0 \ \Ui ; E _ ˝F / to g s 2
.Ug0 \ \Ugi ; E _ ˝F /:
We do not take the space of G-invariants in the definition of the spaces of morphisms
in vectG X, so that the isomorphism class of an object does not depend on the choice
of a G-linearization. Let D b coh X be the idempotent completion of the dg category
T w.vectG X/ consisting of twisted complexes over vect G X . The following lemma
shows that D b coh X gives an enhancement of D b coh X :
Lemma 3.1. Let G be a finite group acting on a smooth projective variety X . Then any
object in D b coh X is a direct summand of a complex of G-linearizable locally-free
sheaves.
P
Proof. Let D be an ample divisor on X . Then g2G g D is a G-invariant ample
divisor on X, and the corresponding invertible sheaf L is G-linearizable. For any
coherent sheaf E, there is a surjection
'0 W .L˝.n0 / /˚k0 ! E
for sufficiently large n0 and k0 . Let E1 D ker '0 be the kernel of this morphism. Then
there is a surjection
'1 W .L˝.n1 / /˚k1 ! E1
for sufficiently large n1 and k1 , and one can set E2 D ker '1 . By repeating this process,
one obtains a distinguished triangle
ŒC1
EkC1 Œk ! F ! E ! EkC1 Œk C 1
where
n ˚mk 'k ˝.n / ˚mk1 'k1 '0 ˚k0 o
F D L˝.nk / ! L k1 ! ! L˝.n0 /
for any k 0. Since X is smooth, the homological dimension of coh X is equal to the
dimension of X , and this triangle splits for k > dim X . It follows that any coherent
sheaf is a direct summand of a complex of G-linearizable locally-free sheaves, and
Lemma 3.1 is proved.
Let vect G X be the subcategory of vect G X with the same set of objects and G-
invariant morphisms;
homvectG X .E; F / D .homvectG X .E; F //G :
The existence of a G-equivariant ample line bundle shows that any G-equivariant
coherent sheaf has a resolution by a bounded complex of G-equivariant locally-free
sheaves. It follows that D b cohG X WD T w.vect G X / gives an enhancement of the
derived category D b cohG X of G-equivariant coherent sheaves on X .
Now assume that G is abelian and let G _ D Hom.G; C / be the group of characters
of G. There is a coherent action .F ; ˛; / of G _ on vect G X defined as follows:
• For a G-equivariant locally-free sheaf E and a character 2 G _ , the G-
equivariant locally-free sheaf F .E/ D E ˝ is given by changing the G-
linearization of E by . To be more precise, let V be a vector space and
W G ! GL.V / be a representation. Then the coherent sheaf E ˝ V is the
functor sending an open set U X to the vector space E.U / ˝ V , and the G-
linearization is given by ˝ W .E ˝ V / ! p2 .E ˝ V / where is considered
as a GL.V /-valued function on G X constant along X .
• For a morphism 2 hom.E; F / of G-equivariant locally-free sheaves (i.e. an
element of the Čech complex), the morphism F . / 2 hom.E ˝ ; F ˝ / is
defined in the obvious way as ˝ id.
• For two characters ; 2 G _ , the natural isomorphism ˛; W F B F
! F˝
comes from the structure of a tensor category on the category of representations
of G (i.e. the natural isomorphism .A ˝ B/ ˝ C ! A ˝ .B ˝ C / for any
G-linear spaces).
This coherent G _ -action on vectG X induces a coherent G _ -action on the G-invariant
part vectG X, which in turn induces a coherent G _ -action on the category D b cohG X
of twisted complexes over vectG X . It is clear that the orbit category vect G X=G _
is equivalent to vectG X , so that the orbit category D b cohG X=G _ is equivalent to
T w.vectG X/. Since D b coh X is the idempotent completion of T w.vect G X /, Theo-
rem 1.2 is proved.
References
[BH09] L. Borisov and Z. Hua, On the conjecture of King for smooth toric Deligne-Mumford
stacks. Adv. Math. 221 (2009), no. 1, 277–301.
[BK90] A. I. Bondal and M. M. Kapranov, Enhanced triangulated categories. Mat. Sb. 181
(1990), no. 5, 669–683.
[Efi10] A. I. Efimov, Maximal lengths of exceptional collections of line bundles. Preprint,
[FU11] M. Futaki and K. Ueda, Homological mirror symmetry for Brieskorn-Pham singular-
ities. Selecta Math. (N.S.) 17 (2011), no. 2, 435–452.
[GL87] W. Geigle and H. Lenzing, A class of weighted projective curves arising in representa-
tion theory of finite-dimensional algebras. In Singularities, representation of algebras,
and vector bundles (Lambrecht, 1985), Lecture Notes in Math. 1273, Springer-Verlag,
Berlin 1987, 265–297.
[HP06] L. Hille and M. Perling, A counterexample to King’s conjecture. Compositio Math.
142 (2006), no. 6, 1507–1521.
[IU] A. Ishii and K. Ueda, Dimer models and exceptional collections. Preprint,
[Kaw06] Y. Kawamata, Derived categories of toric varieties. Michigan Math. J. 54 (2006),
no. 3, 517–535.
[Kel05] B. Keller, On triangulated orbit categories. Doc. Math. 10 (2005), 551–581.
[Kel06] B. Keller, On differential graded categories. In International Congress of Mathemati-
cians, Vol. II, Eur. Math. Soc., Zürich 2006, 151–190.
[Kin97] A. King, Tilting bundles on some rational surfaces. Preprint 1997, available at
http://www.maths.bath.ac.uk/~masadk/papers/
[MFK94] D. Mumford, J. Fogarty, and F. Kirwan, Geometric invariant theory. Third ed., Ergeb.
Math. Grenzgeb. (2) 34, Springer-Verlag, Berlin 1994.
[Orl09] D. Orlov, Derived categories of coherent sheaves and triangulated categories of sin-
gularities. In Algebra, arithmetic, and geometry: in honor of Yu. I. Manin, Vol. II,
Progr. Math. 270, Birkhäuser, Boston, MA, 2009, 503–531.
[Ran81] Z. Ran, Cycles on Fermat hypersurfaces. Compositio Math. 42 (1980/81), no. 1,
mbox121–142.
[Shi79a] T. Shioda, The Hodge conjecture and the Tate conjecture for Fermat varieties. Proc.
Japan Acad. Ser. A Math. Sci. 55 (1979), no. 3, 111–114.
[Shi79b] T. Shioda, The Hodge conjecture for Fermat varieties. Math. Ann. 245 (1979), no. 2,
175–184.
[SK79] T. Shioda and T. Katsura, On Fermat varieties. Tôhoku Math. J. (2) 31 (1979), no. 1,
97–115.
[Wei49] A. Weil, Numbers of solutions of equations in finite fields. Bull. Amer. Math. Soc. 55
(1949), 497–508.
Homology of infinite loop spaces
Dmitry Kaledin
Introduction
A spectrum X q is a sequence of pointed topological spaces X0 ; X1 ; : : : and homotopy
equivalences Xn Š XnC1 , n 0 (we tacitly assume that all the topological spaces
in consideration are nice enough, e.g. having homotopy type of a CW complex). A
spectrum X q is connected if all its components Xn , n 0 are connected. Homology
H q .X q ; Z/ of a spectrum X q with coefficients in a ring R is given by
z qCn .Xn ; R/;
H q .X q ; R/ D lim H
n
!
where H z q .; R/ denotes reduced homology of a pointed topological space, and the
limit is taken with respect to maps †Xn ! XnC1 adjoint to the structure maps Xn !
XnC1 . For any n; i 0, we then have a natural map
ziCn .Xn ; R/ ! Hi .X q ; R/:
H
If the spectrum X q is connected, this map is an isomorphism for i < n.

The forgetful functor sending a spectrum X q to its component X0 is conservative on
the category of connected spectra, so that up to a homotopy equivalence, a connected
spectrum X q can be reconstructed from a pointed topological space X0 equipped with an
additional structure. This structure is usually called an “infinite loop space structure”,
and it can be described in several ultimately equivalent ways, mostly discovered in the
early 1970es and sometimes called “machines” (see [A] for an all-time great overview
of the subject). One of these machines is that of G. Segal [S], where a connected
spectrum is constructed from a so-called special -space. This turned out to be very
useful, since e.g. in algebraic K-theory, the relevant -space often can be obtained
almost for free.
The goal of this short note is to give a simple expression for the homology of a
connected spectrum X q in terms of the associated special -space. We state right away
that the expression is not new, and it is due to T. Pirashvili – namely, it can be deduced
rather directly from Proposition 2.2 of [P2], and for Eilenberg–Maclane spectra, the
results goes back at least to [JP] (see the end of Section 3 for more details). All the
basic ideas behind the proof are also definitely due to Pirashvili. However, the result
itself is never stated explicitly in the general corpus of Pirashvili’s work, and while well

Partially supported by AG Laboratory SU-HSE, RF government grant, ag. 11.G34.31.0023, the RFBR
grant 09-01-00242 and the Science Foundation of the SU-HSE award No. 10-09-0015.
112 D. Kaledin
known to experts, is not universally known. So, a short and self-contained independent
proof might be useful. This is what the present paper aims to provide.
The paper consists of three parts: in Section 1, we recall the details of the Segal
machine in a convenient form, in Section 2, we build a homological counterpart of the
theory, and finally in Section 3, we state and prove our results, and sketch an alternative
approach using Proposition 2.2 in [P2].
Acknowledgements. The paper owes its existence to J. Peter May who explained to
me that the result is not known to everybody. I am also grateful to S. Prontsev for
useful discussions, and I am extremely grateful to T. Pirashvili who explained to me
the exact status of the result and kindly provided all the references. I am grateful to the
referee for useful suggestions.
A note on notation. For the convenience of the reader, here is a brief comparison
between our notation and that of Pirashvili. In [P2] and elsewhere, our C is . Our
functor T is denoted L in [PB]. Our t is t .
1 Recollection on the Segal machine

We start by briefly recalling Segal’s approach to infinite loop spaces and rephrasing it
in a language that suits our goal.
Denote by C the category of finite pointed sets. For any integer n 0, denote by
ŒnC 2 C the set with n unmarked elements (plus one distinguished element o 2 ŒnC ).
0
Alternatively, C is equivalent to the category C of finite sets and partially defined
maps between them – that is, a map from S1 to S2 is given by a diagram

S1 !
S ! S2 (1.1)
0
with injective . The equivalence W C ! C adds a distinguished element o to a
0
set S 2 C , and for any f W S1 ! S2 represented by a diagram (1.1), .f / sends
S1 n .S / to this added distinguished element in S2 .
0
For any injective map W S1 ! S2 of finite sets, let # W S2 ! S1 be the map in C
represented by the diagram
id
S2 !
S1
! S1 :
Definition 1.1. (i) A -space is a functor X W C ! TopC from C to the category of
compactly generated pointed topological spaces. Say that a -space X is normalized
if X.Œ0C / is the one-point set pt.
0
(ii) A -space is special and for any S1 ; S2 2 C
ìf it is normalized,` with the natural
embeddings 1 W S1 ! S1 S2 , 2 W S2 ! S1 S2 , the map
` X..#1 //X..#2 //
X..S1 S2 // ! X..S1 // X..S2 //
is a homotopy equivalence.
Homology of infinite loop spaces 113
Remark 1.2. Sometimes it is convenient to relax the normalization condition on spe-

cial -spaces by only requiring that X.Œ0C / is contractible. However, the stronger
condition is harmless: replacing X.ŒnC /, n 1, with the homotopy fiber of the map
X.Œn/C ! X.Œ0C / corresponding to the unique map ŒnC ! Œ0C , one can always
achieve X.Œ0C / D pt.
The category of -spaces is denoted C TopC . For any two -spaces X1 ; X2 2

C TopC , we define X1 _ X2 , X1 X2 and X1 ^ X2 pointwise.
C
Let C TopC C TopC be the full subcategory spanned by normalized -spaces.
Then the forgetful functor
C
U W C TopC ! TopC ; U.X / D X.Œ1C /
has a left-adjoint
C
T W TopC ! C TopC :
Explicitly, for a pointed topological space X , the -space T.X / is given by

_
T.X /.ŒnC / D X D X ^ ŒnC : (1.2)
s2ŒnC nfog
The adjunction map id ! U B T is an isomorphism, so that T is a full embedding, and

the adjunction map W T B U ! id can be described as follows: for any X 2 C TopC ,
ŒnC 2 C , we have
_ _
D X.is / W T.U.X //.ŒnC / D X.Œ1C / ! X.ŒnC /;
s2ŒnC nfog s2ŒnC nfog
where is W Œ1C ! ŒnC is the embedding onto the subset fs; og ŒnC .
Let be, as usual, the category of finite non-empty totally ordered sets, with
Œn 2 denoting the set of integers from 0 to n, and let S W opp ! Sets be the standard
simplicial circle – that is, the simplicial set obtained by gluing together the two ends of
the standard 1-simplex. The glued ends give a natural distinguished element in S.Œn/,
Œn 2 , so that S is actually a pointed simplicial set. Moreover, S.Œn/ Š ŒnC is a
finite set for any Œn 2 opp , so that S can be interpreted as a functor W opp ! C .
Recall that for any simplicial topological space X W opp ! TopC , we have its
geometric realization Real.X / 2 TopC , and this construction is functorial in X and
compatible with products and colimits. For any simplicial abelian group M , denote
by N q .M / the corresponding standard complex with terms Nn .M / D M.Œn/ and
differential given by the alternating sum of the face maps. Then for any ring R, the
reduced singular chain complex Cz q .Real.X /; R/ is naturally quasiisomorphic to the
total complex of a bicomplex
N q .Cz q .X; R//: (1.3)
114 D. Kaledin
The geometric realization Real.S / of the simplicial circle is homeomorphic to the

1-sphere S 1 . For any simplicial pointed topological space X W opp ! TopC , the
realization
Real.X ^ S /
of the pointwise smash product X ^ S is homeomorphic to the suspension †X D
S 1 ^ X.
Definition 1.3. The geometric realization Real.X / of a -space X is given by
Real.X / D Real. X /:
In particular, for any X 2 TopC , we have T.X / Š S ^ X , so that Real.T.X // Š

†X.
Now consider the product C 2
D C C . Let
1 ;
2 W C 2
! C be the pro-
jections onto the first resp. second factor, and let ˇ W C ! C be the smash-product
2
functor, ˇ.Œn1 C Œn2 C / D Œn1 C ^ Œn2 C Š Œn1 n2 C . Denote by C 2

TopC the
category of functors from C to TopC , and let
2
U1 ; Real1 W C
2
TopC ! C TopC ; T1 W C TopC ! C
2
TopC
be the functors obtained by applying U resp. Real resp. T fiberwise over fibers of
the projection
1 W C
2
! C (in particular, U1 Š i1 , where i1 W C ! C 2
is the
embedding onto C Œ1C C ). For any normalized -space X W C ! TopC , let
2
BX D Real1 .ˇ X /;
and let
†.X / D Real1 .T1 .X //:
Note that for any ŒnC , we have
†.X/.ŒnC / D Real1 .T1 .X //.ŒnC / D Real.T.X.ŒnC /// Š †X.ŒnC /;
and in particular, U.†.X // Š †U.X /. Moreover, since ˇ B i1 Š id, we have

U1 .ˇ X/ Š X, so that we obtain a natural adjunction map
W T1 .X / Š T1 .U1 .ˇ X // ! ˇ X (1.4)
and its realization

X D Real1 . / W †.X / ! BX: (1.5)
Segal, then, proved the following.
Proposition 1.4. Assume given a special -space X . Then
(i) the -space BX is also special, and

(ii) the natural map

U.X / ! U.BX /
adjoint to the map
U.X / W †U.X / Š U.†.X // ! U.BX / (1.6)
is a homotopy equivalence.
By (i), the functor B can be iterated, so that every special -space X gives rise
to a sequence of special -spaces B n X ; by (ii), the sequence U.B n X / with the maps
U.B n X / then naturally forms a spectrum. We will denote this spectrum by EX q .
2 Homology of -spaces
Fix once and for all a commutative ring R, and consider the category Fun.C ; R/ of
functors from C to the category R-mod of R-modules. This is an abelian category
with enough injectives and projectives. We equip it with pointwise tensor product, and
we denote by D.C ; R/ its derived category. An obvious set of projective generators
is given by representable functors Tn ,
Tn .ŒmC / D R.C .ŒnC ; ŒmC /;
since by Yoneda, we have Hom.Tn ; E/ Š E.ŒnC / for any E 2 Fun.C ; R/. Let
T 2 Fun.C ; R/ be the functor given by
M
T .ŒnC / D RŒŒnC D R;
s2ŒnC nfog
that is, the reduced span functor. We have an obvious direct sum decomposition T1 Š
T0 ˚ T .
Consider the functor T W R-mod ! Fun.C ; R/ given by
T.M / D T ˝R M
for any R-module M . This is consistent with previous notation, in the sense that for any
X 2 TopC with reduced singular chain complex Cz q .X; R/, (1.2) immediately gives a
canonical isomorphism
z q .X; R// Š Cz q .T.X /; R/:

T.C (2.1)
The functor T W R-mod ! Fun.C ; R/ is exact, and it has a right and a left-adjoint
R; Q W Fun.C ; R/ ! R-mod.
116 D. Kaledin
Lemma 2.1. For any E 2 Fun.C ; R/, we have a canonical decomposition
E.Œ1C / Š R.E/ ˚ E.Œ0C /:
The functor R is exact, the functor T is fully faithful, and its extension T W D.R-mod/ !
D.C ; R/ is also fully faithful.
Proof. The decomposition is induced by the decomposition T1 Š T ˚ T0 . Exactness
of R follows; to see that the embedding T is fully faithful, note that R B T Š id both on
the abelian and on the derived category level.
Definition 2.2. The homology H q .E/ of a functor E 2 Fun.C ; R/ is given by
q
H q .E/ D L Q.E/;
the derived functors of the functor Q left-adjoint to the full embedding T W R-mod !
Fun.C ; R/.
Explicitly, homology can be expressed as
H q .E/ D Tor q C .t; E/;

opp
where t W C ! R-mod is given by t .ŒnC / D HomR .T .ŒnC /; R/, and Tor q is taken
over the small category C in the usual way, see e.g. [K], Section 1.1. To compute it, it
suffices to find a projective resolution of the functor t . One very elegant way to do it was
discovered by Pirashvili and Jibladze, and it leads to the so-called cube construction
of MacLane (see [LP], or a slightly less computational exposition in Section 3.3 of
[K]). Whatever resolution Q q one fixes, one immediately obtains a canonical way to
generalize homology to complexes: for any complex E q in Fun.C ; R/, we obtain a
complex
Q q .E q / D Q q ˝C E q (2.2)
of R-modules whose homology we denote by H q .E q /. If the complex E q D E is
concentrated in degree 0, we have H q .E q / Š H q .E/
The following lemma is the crucial result of the theory.
Lemma 2.3. Assume given two functors E1 ; E2 2 Fun.C ; R/ such that E1 .Œ0C / D
E2 .Œ0C / D 0. Then H q .E1 ˝R E2 / D 0.
Proof. This is [PB], Lemma 2; I give a proof for the convenience of the reader.
Consider the product C C , let
1 ;
2 W C C ! C be the projections
onto the first resp. second factor, and let 1 ; 2 W C ! C C be the embeddings
sending ŒnC to ŒnC Œ0C resp. Œ0C ŒnC . Moreover, let m W C C ! C be
the coproduct functor, and let ı W C ! C C be the diagonal embedding. Then
i is right-adjoint to i , i D 1; 2, and m is right-adjoint to ı. We obviously have
m T Š
1 T ˚
2 T;
and this decomposition induces an isomorphism
m B T Š .
1 B T / ˚ .
2 B T /:
By adjunction, we obtain a functorial isomorphism

q q q
L Q.ı E/ Š L Q.1 E/ ˚ L Q.2 E/
for any functor E W C C ! R-mod. Take E D E1 R E2 , and note that by

assumption, 1 E D 2 E D 0, while ı E Š E1 ˝R E2 .
Definition 2.4. The homology H q .X; R/ of a -space X is given by
H q .X; R/ D H q .Cz q .X; R//;
where Cz q .X; R/ is a complex in Fun.C ; R/ obtained by taking pointwise the reduced

singular chain complex Cz q .; R/.
Lemma 2.5. For any X 2 TopC we have a canonical isomorphism
z q .X; R/ Š H q .T.X /; R/:
H
Proof. By virtue of the quasiisomorphism (2.1), this immediately follows from the last
claim of Lemma 2.1: we have
q q
Tor q C .t; T / Š HomR .Ext C .T; T /; R/ Š HomR .Ext R .R; R//;
and the right-hand side is R in degree 0 and 0 in higher degrees, so that for any complex
C q of R-modules, the groups
H q .T.C q // Š Tor q C .t; C q ˝ T /
coincide with the homology groups of the complex C q itself.

Lemma 2.3 has the following implication for the homology of -spaces. For any
n 0, let n W C ! C be the functor given by n .ŒmC / D ŒnC ^ ŒmC . Assume
given two pointed finite sets ŒnC ; Œn0 C 2 C , identify ŒnC _ Œn0 C Š Œn C n0 C , and
consider the natural maps ŒnC ! ŒnC _ Œn0 C Š Œn C n0 C , Œn0 C ! ŒnC _ Œn0 C Š
Œn C n0 C . These maps then induce maps
W n ! nCn0 ; 0 W n0 ! nCn0 ;
and for any -space X , we obtain natural maps
W n X ! nCn0 X; 0 W n0 X ! nCn0 X
and
0
# W nCn0 X ! n X; # W nCn0 X ! n0 X:
118 D. Kaledin
Corollary 2.6. Assume that the -space X is special, Then the natural map
_ 0 W n X _ n0 X ! nCn0 X
of -spaces induces an isomorphism of homology H q .; R/.

Proof. For any two pointed topological spaces X1 , X2 , we have a cofiber sequence
X1 _ X2 ! X1 X2 ! X1 ^ X2 :
Since X is special, the natural map

0#
#
nCn0 X ! n X n0 X
is a pointwise homotopy equivalence. Therefore the sequence

0#
_0 # ^
n X _ n0 X ! nCn0 X ! n X ^ n0 X
is a pointwise cofiber sequence, and it suffices to prove that
H q .n X ^ n0 X; R/ D 0:
This immediately follows from Lemma 2.3 and the Künneth formula.
3 Stabilization
We can now formulate and prove the main result of the paper. For any special -space
X, let
X W T.U.X // ! X
be the adjunction map, and let X W †.X / ! BX be as in (1.5).
Lemma 3.1. For any special -space X , the diagram
†.X /
†.T.U.X /// Š T.U.†.X /// / †.X /
T.U.X // X
BX
T.U.BX // / BX
is commutative.
Proof. By (1.2), we have
T1 .T.Y //.Œn1 C Œn2 C / Š Œn2 C ^ T.Y /.Œn1 C / Š Œn2 C ^ Œn1 C ^ Y

Š Œn1 n2 C ^ Y Š ˇ T.Y /.Œn1 C Œn2 C /
for any Y 2 TopC , Œn1 C ; Œn2 C 2 C , so that T1 .T.Y // Š ˇ T.Y /. Taking Y D

U.X/, we obtain a natural commutative diagram
T1 .T.U.X // / T1 .X /

ˇ T.U.X // / ˇ X:
Applying Real1 , we get the claim.

Taking homology H q .; R/ and using Lemma 2.5, we obtain a commutative dia-
gram
Hz q .U.X /; R/ / H q .X; R/

z qC1 .U.BX /; R/
H / H q .BX; R/;
C1
and passing to the limit, we get a natural map
H q .EX q ; R/ ! lim H qCn .U.B n X /; R/: (3.1)

n
!
Here is, then, our main result.

Theorem 3.2. Assume given a special -space X , and let EX q be the corresponding
spectrum. Then the natural map (3.1) factors through an isomorphism
H q .EX q ; R/ Š H q .X; R/;
where the right-hand side is as in Definition 2.4.

The proof is a combination of the following two results.
Lemma 3.3. Assume given a special -space X . Then the map X of (1.5) induces
an isomorphism
H q .†.X /; R/ Š H q .BX; R/:
Proof. Combining (1.3) and (2.2), we see that for any X 2 C 2
TopC , the homol-
ogy H q .Real1 .X /; R/ can be computed by the total complex of the triple complex

Q q .N q .C q .X/; R//. This gives rise to a convergent spectral sequence
Hi .in X; R/ ) HiCn

.Real1 .X /; R/;
where in W C ! C 2
is the embedding onto C ŒnC . We conclude that to prove the
lemma, it suffices to prove that for every n 0, the map
H q .in T1 .X /; R/ ! H q .in ˇ X; R/
120 D. Kaledin
induced by the map of (1.4) is an isomorphism. By (1.2), we have

_
in T1 .X / Š ŒnC ^ X Š X
s2ŒnC nfog
and by definition,
in ˇ X Š n X;
so that the statement immediately follows by induction on n from Corollary 2.6.
Lemma 3.4. Assume given a special -space X , and assume that U.X / is n-connected
for some n 1. Then the natural map
zi .U.X /; R/ Š Hi .T.U.X //; R/ ! Hi .X; R/
H
induces by the map X is an isomorphism for i < 2n.
Proof. Since U.X / is n-connected, H0 .U.X /; R/ Š R and Hi .U.X /; R/ D 0 when

0 < i < n. Then by Definition 1.1 (ii), X.ŒnC / is homotopy-equivalent to X.Œ1C /n D
U.X/n for any n 1, and then the Künneth formula immediately implies that the map
Cz q .T.U.X //; R/ ! Cz q .X; R/
induced by X is a quasiisomorphism in homological degrees less than 2n. Applying

q
L Q, we get the claim.
Proof of Theorem 3.2. By Lemma 3.3, the natural map
H q .X; R/ ! lim H qCn .U.B n X /; R/

n
!
is an isomorphism, and by Lemma 3.4, (3.1) is also an isomorphism – in fact, in each

homological degree, it becomes an isomorphism at some finite step in the inductive
sequence.
To finish the paper, let us explain how Theorem 3.2 can be deduced from the work
of T. Pirashvili mentioned in the introduction. Note that any abelian group can be
treated as a pointed set, by taking 0 as the distinguished point and forgetting the rest
of the group structure. Thus a functor F W C ! opp Ab from C to the category of
simplicial abelian groups can be treated as a pointed simplicial -set. Then even if such
F is not special in the sense of Definition 1.1 (ii), the map (1.6) is still well-defined,
so that the sequence B n F , n 0 forms a pre-spectrum. One denotes by
st q .F / the
homotopy groups of the corresponding spectrum. Then Proposition 2.2 of [P2] claims
that there exists a natural isomorphism
q .F / Š Tor q C .t; F /

st
(to be precise, Proposition 2.2 in [P2] is stated only for functors to constant simplicial
groups, but generalization to arbitrary ones is immediate). Pirashvili’s proof of this
fact also uses Lemma 2.3, but it is in fact simpler since working with an arbitrary F
gives more latitude. Then to deduce Theorem 3.2, one has to take F D C q .X; R/, and
show that
H q .EX q ; R/ Š
st
q .F /:
This is also rather straightforward. So in a nutshell, Pirashvili’s proof is ultimately
simpler but relies on some context, while our proof is longer but elementary and self-
contained.
References
[A] J. F. Adams, Infinite loop spaces. Ann. of Math. Stud. 90, Princeton University Press and
University of Tokyo Press, Princeton, NJ, 1978.
[K] D. Kaledin, Non-commutative Hodge-to-de Rham degeneration via the method of Deligne-
Illusie. Pure Appl. Math. Q. 4 (2008), 785–875.
[JP] M. Jibladze and T. Pirashvili, Cohomology of algebraic theories. J. Algebra 137 (1991),
253–296.
[LP] J.-L. Loday and T. Pirashvili, Mac Lane (co)homology. Chapter 13 in J.-L. Loday, Cyclic
homology, second ed., Grundlehren Math. Wiss. 301, Springer-Verlag, Berlin 1998.
[P1] T. Pirashvili, Polynomial approximation of ext and tor groups in functor categories. Comm.
Algebra 21 (1993), 1705–1719.
[P2] T. Pirashvili, Hodge decomposition for higher order Hochschild homology. Ann. Sci. Ec.
Norm. Sup. (4) 33 (2000), 151–179.
[PB] T. Pirashvili and B. Richter, Robinson-Whitehouse complex and stable homotopy. Topology
39 (2000), 525–530.
[S] G. Segal, Categories and cohomology theories. Topology 13 (1974), 293–312.
Cluster algebras and derived categories
Bernhard Keller
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
2 Description and first examples of cluster algebras . . . . . . . . . . . . . . 126
2.1 Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
2.2 First example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
2.3 Cluster algebras of rank 2 . . . . . . . . . . . . . . . . . . . . . . . . 127
3 Cluster algebras associated with quivers . . . . . . . . . . . . . . . . . . . 127
3.1 Quiver mutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.2 Seed mutation, cluster algebras . . . . . . . . . . . . . . . . . . . . . 129
3.3 Cluster algebras associated with valued quivers . . . . . . . . . . . . 131
4 Cluster algebras of geometric type . . . . . . . . . . . . . . . . . . . . . . 134
4.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.2 Example: Planes in a vector space . . . . . . . . . . . . . . . . . . . 135
4.3 Example: The Grassmannian Gr.3; 6/ . . . . . . . . . . . . . . . . . 137
4.4 Example: Rectangular matrices . . . . . . . . . . . . . . . . . . . . . 138
4.5 Finite generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.6 Factoriality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5 General cluster algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.1 Parametrization of seeds by the n-regular tree . . . . . . . . . . . . . 141
5.2 Principal coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.3 Principal coefficients: c-vectors . . . . . . . . . . . . . . . . . . . . 142
5.4 Principal coefficients: F -polynomials and g-vectors . . . . . . . . . . 144
5.5 Tropical duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
5.6 Product formulas for c-matrices and g-matrices . . . . . . . . . . . . 147
5.7 Cluster algebras with coefficients in a semifield . . . . . . . . . . . . 148
5.8 The separation formulas . . . . . . . . . . . . . . . . . . . . . . . . 150
6 Quantum cluster algebras and quantum dilogarithms . . . . . . . . . . . . . 150
6.1 The quantum dilogarithm . . . . . . . . . . . . . . . . . . . . . . . . 150
6.2 Quantum mutations and quantum cluster algebras . . . . . . . . . . . 151
6.3 Fock–Goncharov’s separation formula . . . . . . . . . . . . . . . . . 153
6.4 The quantum separation formula . . . . . . . . . . . . . . . . . . . . 154
7 Categorification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
7.1 Mutation of quivers with potential . . . . . . . . . . . . . . . . . . . 157
7.2 Ginzburg algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
7.3 Derived categories of dg algebras . . . . . . . . . . . . . . . . . . . . 161
7.4 The derived category of the Ginzburg algebra . . . . . . . . . . . . . 162
124 B. Keller
7.5 Derived equivalences from mutations . . . . . . . . . . . . . . . . . . 164

7.6 Torsion subcategories and intermediate t -structures . . . . . . . . . . 166
7.7 Patterns of tilts and decategorification . . . . . . . . . . . . . . . . . 168
7.8 Reign of the tropics . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
7.9 Rigid objects and cluster monomials . . . . . . . . . . . . . . . . . . 172
7.10 Proof of decategorification . . . . . . . . . . . . . . . . . . . . . . . 172
7.11 Proof of the quantum dilogarithm identities . . . . . . . . . . . . . . 174
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
1 Introduction
Cluster algebras, invented [45] by Sergey Fomin and Andrei Zelevinsky around the
year 2000, are commutative algebras whose generators and relations are constructed in
a recursive manner. Among these algebras, there are the algebras of homogeneous co-
ordinates on the Grassmannians, on the flag varieties and on many other varieties which
play an important role in geometry and representation theory. Fomin and Zelevinsky’s
main aim was to set up a combinatorial framework for the study of the so-called canon-
ical bases which these algebras possess [79], [98] and which are closely related to the
notion of total positivity [99], [41] in the associated varieties. It has rapidly turned
out that the combinatorics of cluster algebras also appear in many other subjects, for
example in
• Poisson geometry [61], [62], [63], [64], [10], …;
• discrete dynamical systems [27], [48], [75], [80], [85], [82], [92], …;
• higher Teichmüller spaces [35], [36], [37], [38], [39], …;
• combinatorics and in particular the study of polyhedra like the Stasheff associa-
hedra [21], [22], [42], [43], [72], [91], [103], [105], [104], [131], …;
• commutative and non commutative algebraic geometry and in particular the study
of stability conditions in the sense of Bridgeland [11], Calabi–Yau algebras [65],
[76], Donaldson–Thomas invariants in geometry [77], [90], [89], [124], [138],
… and in string theory [1], [2], [14], [15], [16], [50], [51], [52], …;
• in the representation theory of quivers and finite-dimensional algebras, cf. for
example the survey articles [3], [5], [60], [85], [96], [127] [126], [128], …
as well as in mirror symmetry [68], KP solitons [88], hyperbolic 3-manifolds [110],
… . We refer to the introductory articles [41], [47], [145], [146], [147], [148], and to
the cluster algebras portal [40] for more information on cluster algebras and their links
with other subjects in mathematics (and physics).
In these notes, we give a concise introduction to cluster algebras and survey their
(additive) categorification via derived categories of Ginzburg dg (D differential graded)
algebras. We prepare the ground for the formal definition of cluster algebras by giving
Cluster algebras and derived categories 125
an approximate description and the first examples in Section 2. In Section 3, we

introduce the central construction of quiver mutation and define the cluster algebra
associated with a quiver and, more generally, with a valued quiver (Section 3.3). We
extend the definition to that of cluster algebras of geometric type and present several
examples in Section 4. Here we also review results on ring-theoretic properties of
cluster algebras (finite generation and factoriality). In Section 5, we give the general
definition of cluster algebras with coefficients in an arbitrary semifield. In this general
framework, the symmetry between cluster variables and coefficients becomes apparent,
for example in the separation formulas in Theorem 5.7 but also, at the ‘tropical level’, in
the duality Theorem 5.4. In Section 6, we present the construction of quantum cluster
algebras and its link with the quantum dilogarithm function. We show how cluster
algebras allow one to construct identities between products of quantum dilogarithm
series. This establishes the link to Donaldson–Thomas theory, as we will see for
example in Section 7.11.
In Section 7, we turn to the (additive) categorification of cluster algebras. In Sec-
tion 5 of [85], the reader will find a gentle introduction to this subject along the lines
of the historical development. We will not repeat this here but restrict ourselves to a
description of the most recent framework, which applies to arbitrary symmetric clus-
ter algebras (of geometric type). The basic idea is to lift the cluster variables in the
cluster algebra associated with a quiver Q to suitable representations of Q. These
representations have to satisfy certain relations, which are encoded in a potential on
the quiver. We review quivers with potentials and their mutations following Derksen–
Weyman–Zelevinsky [25] in Section 7.1. A conceptual framework for the study of the
representations of a quiver with potential is provided by the derived category of the
associated Ginzburg dg algebra (Section 7.4). Here mutations of quivers with potential
yield equivalences between derived categories of Ginzburg algebras (Section 7.5). In
fact, each mutation canonically lifts to two equivalences. Thus, in trying to compose the
categorical lifts of N mutations, we are forced to choose between 2N possibilities. The
canonical choice was discovered by Nagao [107] and is presented in Section 7.7. The
framework thus created allows for the categorification of all the data associated with
a commutative cluster algebra (Theorem 7.9). A recent extension to quantum cluster
algebras (under suitable technical assumptions) is due to Efimov [30]. Surprisingly, the
combinatorial data determine the categorical data to a very large extent (Sections 7.8
and 7.9). We end by linking our formulation of the ‘decategorification Theorem’ 7.9
to the statements available in the literature (Section 7.10) and by proving Theorem 6.5
on quantum dilogarithm identities (Section 7.11).
This introductory survey leaves out a number of important recent developments,
notably monoidal categorification, as developed by Hernandez–Leclerc [71], [96] and
Nakajima [112], the theory of cluster algebras associated with marked surfaces [43],
[44], [105], [104], [20] … and recent progress on the links between (quantum) cluster
algebras and Lie theory [56], [57], [54], … .
126 B. Keller
Acknowledgment. This survey expands on a talk given by the author at the GCOE
Conference ‘Derived Categories 2011 Tokyo’, where Theorem 6.5 was presented. He
deeply thanks the organizers for their invitation and their kind hospitality. He is very
grateful to B. Leclerc for correcting an error in a previous version of Example 4.3 and
to H. Nakajima for a helpful conversation on the results of [30].
2 Description and first examples of cluster algebras

2.1 Description. A cluster algebra is a commutative Q-algebra endowed with a set
of distinguished generators (the cluster variables) grouped into overlapping subsets
(the clusters) of constant cardinality (the rank) which are constructed recursively via
mutation from an initial cluster. The set of cluster variables can be finite or infinite.
Theorem 2.1 ([46]). The cluster algebras having only a finite number of cluster vari-
ables are parametrized by the finite root systems.
Thus, the classification is analogous to the one of semi-simple complex Lie algebras.
We will make the theorem more precise in Section 3 below (for simply laced root
systems).
2.2 First example. In order to illustrate the description and the theorem, we present
[148] the cluster algebra AA2 associated with the root system A2 . By definition, it is
generated as a Q-algebra by the cluster variables xm , m 2 Z, submitted to the exchange
relations
xm1 xmC1 D 1 C xm ; m 2 Z:
Its clusters are by definition the pairs of consecutive cluster variables fxm ; xmC1 g,
m 2 Z. The initial cluster is fx1 ; x2 g and two clusters are linked by a mutation if and
only if they share exactly one variable.
The exchange relations allow one to write each cluster variable as a rational expres-
sion in the initial variables x1 , x2 and thus to identify the algebra AA2 with a subalgebra
of the field Q.x1 ; x2 /. In order to make this subalgebra explicit, let us compute the
cluster variables xm for m 3. We have:
1 C x2
x3 D ; (1)
x1
1 C x3 x1 C 1 C x2
x4 D D ; (2)
x2 x1 x2
1 C x4 x1 x2 C x1 C 1 C x2 1 C x2 1 C x1
x5 D D D : (3)
x3 x1 x2 x1 x2
Notice that, contrary to what one might expect, the denominator in (3) remains a
monomial! In fact, each cluster variable in an arbitrary cluster algebra is a Laurent
polynomial, cf. Theorem 3.1 below. Let us continue the computation:

1 C x5 x2 C 1 C x1 x1 C 1 C x2
x6 D D D x1 ; (4)
x4 x2 x1 x2
1 C x1
x7 D .1 C x1 / D x2 : (5)
x2
It is now clear that the sequence of the xm , m 2 Z, is 5-periodic and that the number
of cluster variables is indeed finite and equal to 5. In addition to the two initial vari-
ables x1 and x2 , we have three non initial variables x3 , x4 and x5 . By examining
their denominators we see that they are in natural bijection with the positive roots ˛1 ,
˛1 C ˛2 , ˛2 of the root system A2 . This generalizes to an arbitrary Dynkin diagram,
cf. Theorem 3.1.
2.3 Cluster algebras of rank 2. To each pair of positive integers .b; c/, there is
associated a cluster algebra A.b;c/ . It is defined in analogy with AA2 by replacing the
exchange relations with
´
x b C 1 if m is odd,
xm1 xmC1 D m
xmc
C 1 if m is even.
The algebra A.b;c/ has only a finite number of cluster variables if and only if we have
bc 3. In other words, if and only if the matrix

2 b
c 2
is the Cartan matrix of a finite root system ˆ of rank 2. The reader is invited to check
that in this case, the non initial cluster variables are still in natural bijection with the
positive roots of ˆ.
3 Cluster algebras associated with quivers

3.1 Quiver mutation. A quiver is an oriented graph, i.e. a quadruple
Q D .Q0 ; Q1 ; s; t /
formed by a set of vertices Q0 , a set of arrows Q1 and two maps s and t from Q1 to Q0
which send an arrow ˛ respectively to its source s.˛/ and its target t .˛/. In practice,
a quiver is given by a picture as in the following example:
$ //
3 ^= ˛ 5 /6
==
==
QW ==
ˇ
/
1 /2o 4.

128 B. Keller
An arrow ˛ whose source and target coincide is a loop; a 2-cycle is a pair of distinct
arrows ˇ and such that s.ˇ/ D t . / and t .ˇ/ D s./. Similarly, one defines n-cycles
for any positive integer n. A vertex i of a quiver is a source (respectively a sink) if
there is no arrow with target i (respectively with source i ).
By convention, in the sequel, by a quiver we always mean a finite quiver without
loops nor 2-cycles whose set of vertices is the set of integers from 1 to n for some
n 1. Up to an isomorphism fixing the vertices such a quiver Q is given by the skew-
symmetric matrix B D BQ whose coefficient bij is the difference between the number
of arrows from i to j and the number of arrows from j to i for all 1 i; j n.
Conversely, each skew-symmetric matrix B with integer coefficients comes from a
quiver.
Let Q be a quiver and k a vertex of Q. The mutation k .Q/ is the quiver obtained
from Q as follows:
ˇ
1) for each subquiver i /k ˛ / j , we add a new arrow Œ˛ˇ W i ! j ;
2) we reverse all arrows with source or target k;
3) we remove the arrows in a maximal set of pairwise disjoint 2-cycles.
For example, if k is a source or a sink of Q, then the mutation at k simply reverses all
the arrows incident with k. In general, if B is the skew-symmetric matrix associated
with Q and B 0 the one associated with k .Q/, we have
´
0 bij if i D k or j D k;
bij D (6)
bij C sgn.bik / max.0; bik bkj / else.
This is the matrix mutation rule for skew-symmetric (more generally: skew-symmetri-
zable) matrices introduced by Fomin–Zelevinsky in [45], cf. also [49].
One checks easily that k is an involution. For example, the quivers
1 1
E 222 Y333
22 33
22 and
33 (7)
2 3
2 o 3 2 3
are linked by a mutation at the vertex 1. Notice that these quivers are drastically
different: The first one is a cycle, the second one the Hasse diagram of a linearly
ordered set.
Two quivers are mutation equivalent if they are linked by a finite sequence of
mutations. For example, it is an easy exercise to check that any two orientations of
a tree are mutation equivalent. Using the quiver mutation applet [83] or the package
[106] one can check that the following three quivers are mutation equivalent.
E12 10 \\\- 5 46
222 !! 6 } 5 jUUU
! ~} 3Q$
1 6 $$
o 7 m\\\ 6 x
|xx
F 2 33 F 3 22 Z55
33 22 7$ xx< 2
$$ (8)
o o 8 RRR( lll5 4 X1 > 10 E
F 4 22 E 5 22 F 6 22 9! 1 8 UUU* || E"
22 22 22 !! 9 1
3
7 o 8o 9o 10 2
The common mutation class of these quivers contains 5739 quivers (up to isomor-
phism). The mutation class of ‘most’ quivers is infinite. The classification of the
quivers having a finite mutation class was achieved by Felikson–Shapiro–Tumarkin
[34], [33]: in addition to the quivers associated with triangulations of surfaces (with
boundary and marked points, cf. [43]) the list contains 11 exceptional quivers, the
largest of which is in the mutation class of the quivers (8).
3.2 Seed mutation, cluster algebras. Let n 1 be an integer and F the field
Q.x1 ; : : : ; xn / generated by n indeterminates x1 ; : : : ; xn . A seed (more precisely: X -
seed) is a pair .R; u/, where R is a quiver and u a sequence u1 ; : : : ; un of elements
of F which freely generate the field F . If .R; u/ is a seed and k a vertex of R, the
mutation k .R; u/ is the seed .R0 ; u0 /, where R0 D k .R/ and u0 is obtained from u
by replacing the element uk by the element u0k defined by the exchange relation
Y Y
u0k uk D u t.˛/ C us.˛/ ; (9)
s.˛/Dk t.˛/Dk
where the sums range over all arrows of R with source k respectively target k. Notice
that, if B is the skew-symmetric matrix associated with R, we can rewrite the exchange
relation as Y Œb Y Œb
u0k uk D ui i k C C ui i k C ; (10)
i i
where, for a real number x, we write ŒxC for max.x; 0/. One checks that 2k .R; u/ D
.R; u/. For example, the mutations of the seed
.1 /2 / 3, fx1 ; x2 ; x3 g /
with respect to the vertices 1 and 2 are the seeds
.1o 2 / 3, f 1Cx2 ; x2 ; x3 g /
x1
(11)
and
&
.1o 2o 3, fx1 ; x1xCx
2
3
; x3 g /: (12)
130 B. Keller
Let us fix a quiver Q. The initial seed of Q is .Q; fx1 ; : : : ; xn g/. A cluster
associated with Q is a sequence u which appears in a seed .R; u/ obtained from the
initial seed by iterated mutation. The cluster variables are the elements of the clusters.
The cluster algebra AQ is the Q-subalgebra of F generated by the cluster variables.
Clearly, if .R; u/ is a seed associated with Q, the natural isomorphism

Q.u1 ; : : : ; un /
! Q.x1 ; : : : ; xn /
induces an isomorphism of AR onto AQ which preserves the cluster variables and the
clusters. Thus, the cluster algebra AQ is an invariant of the mutation class of Q. It is
useful to introduce a combinatorial object which encodes the recursive construction of
the seeds: the exchange graph. By definition, its vertices are the isomorphism classes of
seeds (isomorphisms of seeds renumber the vertices and the variables simultaneously)
and its edges correspond to mutations. For example, the exchange graph obtained from
the quiver Q W 1 /2 / 3 is the 1-skeleton of the Stasheff associahedron [137]:
0123P
7654
qqqq 3$$ PPPPP
q $$ PPP
qqq PPP
qqq $$ PP
B* $$ B,
**
** $$ ,,
* $
B-
-- 0123
7654 B
-- MMMB M 1? B 1
MM ?? uuuuu 11
B: B? 0123
7654
:: ?? 2 tt B
:: tt
:: B ttt
t
:: tt
:: ttt
: ttt
t
B
Here the vertex 1 corresponds to the initial seed and the vertices 2 and 3 to the seeds (11)
and (12). For analogous polytopes associated with the other Dynkin diagrams, we refer
to [22].
Let Q be a connected quiver. If its underlying graph is a simply laced Dynkin
diagram , we say that Q is a Dynkin quiver of type .
Theorem 3.1 ([46]). a) Each cluster variable of AQ is a Laurent polynomial with
integer coefficients [45].
b) The cluster algebra AQ has only a finite number of cluster variables if and only
if Q is mutation equivalent to a Dynkin quiver R. In this case, the underlying graph
of R is unique up to isomorphism and is called the cluster type of Q.
c) If Q is a Dynkin quiver of type , then the non initial cluster variables of AQ
are in bijection with the positive roots of the root system ˆ of ; more precisely, if ˛1 ,
…, ˛n are the simple roots, then for each positive root ˛ D d1 ˛1 C C dn ˛n , there
is a unique non initial cluster variable X˛ whose denominator is x1d1 : : : xndn .
Statement a) is usually referred to as the Laurent phenomenon. A cluster monomial

is a product of non negative powers of cluster variables belonging to the same cluster.
The construction of a ‘canonical basis’ of the cluster algebra AQ is an important and
largely open problem, cf. for example [45], [135], [29], [18], [56], [104], [93], [94],
[70]. It is expected that such a basis should contain all cluster monomials. Whence the
following conjecture.
Conjecture 3.2 ([45]). The cluster monomials are linearly independent over the
field Q.
The conjecture was recently proved in [19] using the additive categorification of
[118] and techniques from [17], [20]. It is expected to hold more generally for cluster
algebras associated with valued quivers, cf. Section 3.3 below. It is shown for a certain
class of valued quivers by L. Demonet [23], [24]. For special classes of quivers, a
basis containing the cluster monomials is known: If Q is a Dynkin quiver, one knows
[13] that the cluster monomials form a basis of AQ . If Q is acyclic, i.e. does not have
any oriented cycles, then Geiss–Leclerc–Schröer [53] show the existence of a ‘generic
basis’ containing the cluster monomials.
Conjecture 3.3 ([46]). The cluster variables are Laurent polynomials with non nega-
tive integer coefficients in the variables of each cluster.
For quivers with two vertices, an explicit and manifestly positive formula for the
cluster variables is given in [97]. The technique of monoidal categorification devel-
oped by Leclerc [95] and Hernandez–Leclerc [71] has recently allowed to prove this
conjecture first for the quivers of type An and D4 , cf. [71], and then for each bipartite
quiver [112], i.e. a quiver where each vertex is a source or a sink. The positivity of
all cluster variables with respect to the initial seed of an acyclic quiver is shown by
Fan Qin [120] and by Nakajima [112] (Appendix). This is also proved by Efimov
[30], who moreover shows the positivity of all cluster variables belonging to an acyclic
seed with respect to the initial variables of an arbitrary quiver. Efimov combines the
techniques of [89] with those of [107]. A proof of the full conjecture for acyclic quivers
using Nakajima quiver varieties is due to Kimura–Qin [87]. The conjecture has been
shown in a combinatorial way by Musiker–Schiffler–Williams [105] for all the quivers
associated with triangulations of surfaces (with boundary and marked points) and by
Di Francesco–Kedem [27] for the quivers and the cluster variables associated with the
T -system of type A, with respect to the initial cluster.
We refer to [47] and [49] for numerous other conjectures on cluster algebras and to
[26], cf. also [107] and [119], [118], for the solution of a large number of them using
additive categorification.
3.3 Cluster algebras associated with valued quivers. A valued quiver is a quiver
Q endowed with a function v W Q1 ! N 2 such that
a) there are no loops in Q,
132 B. Keller
b) there is at most one arrow between any two vertices of Q and

c) there is a function d W Q0 ! N such that d.i / is strictly positive for all vertices
i and, for each arrow ˛ W i ! j , we have
d.i / v.˛/1 D v.˛/2 d.j /;
where v.˛/ D .v.˛/1 ; v.˛/2 /.
For example, we have the valued quivers (we omit the labels .1; 1/ from our pictures)
.1;2/ .2;1/
BÅ3 W 1 /2 /3 and CÅ3 W 1 /2 / 3,
where possible functions d are given by d.1/ D d.2/ D 2, d.3/ D 1 respectively

d.1/ D d.2/ D 1, d.3/ D 2. A valued quiver .Q; v/ is equally valued if we have
v.˛/1 D v.˛/2 for each arrow ˛. If Q is an ordinary quiver without loops nor 2-cycles,
the associated valued quiver is the equally valued quiver which has an arrow ˛ W i ! j
if there is at least one arrow i ! j in Q and where v.˛/ D .m; m/, where m is the
number of arrows from i to j in Q. For example, the equally valued quiver
.2;2/
/2 //
1 corresponds to the Kronecker quiver 1 2:
In this way, the ordinary quivers without loops nor 2-cycles correspond bijectively to
the equally valued quivers (up to isomorphism fixing the vertices). Let Q be a valued
quiver with vertex set I . We associate an integer matrix B D .bij /i;j 2I with it as
follows 8
ˆ
<0 if there is no arrow between i and j ;
bij D v.˛/1 if there is an arrow ˛ W i ! j I
:̂
v.˛/2 if there is an arrow ˛ W j ! i:
If D is the diagonal I I -matrix with diagonal entries di i D d.i /, i 2 I , then the
matrix DB is skew-symmetric. The existence of such a matrix D means that the matrix
B is skew-symmetrizable. It is easy to check that in this way, we obtain a bijection
between the skew-symmetrizable I I -matrices B and the valued quivers with vertex
set I (up to isomorphism fixing the vertices). Using this bijection, we define the
mutation of valued quivers using Fomin–Zelevinsky’s matrix mutation rule (6). For
example, the mutation at 2 transforms the valued quiver
C27 2 [7
777 777

2;1 3;2 into 1;2 2;3
77 77
77 77

1 3 1 6;2 / 3.
We extend the notion of an (X -)seed .R; u/ by now allowing the first component R
to be any valued quiver and we extend the construction of seed mutation by using the
rule (10), where B is the skew-symmetrizable matrix associated with R. For example
the mutations of the seed
.1;2/
.1 /2 / 3, fx1 ; x2 ; x3 g /
at the vertices 1 and 2 are the seeds

.1;2/
.1o 2 / 3, f 1Cx2 ; x2 ; x3 g /
x1
and
.1;2/
& x1 Cx32
.1o 2o 3, fx1 ; x2
; x3 g /:
.2;1/
Given a valued quiver Q, we define its associated clusters, cluster variables, cluster
monomials, the cluster algebra AQ and the exchange graph in complete analogy with
the constructions in Section 3.2. For example, the exchange graph of the above quiv-
ers BÅ3 and CÅ3 is the 3rd cyclohedron [22], with 4 quadrilateral, 4 pentagonal and 4
hexagonal faces:
Let .Q; v/ be a valued quiver with vertex set I D Q0 . Its associated Cartan matrix
is the Cartan companion, [46] of the skew-symmetrizable matrix B associated with Q.
Explicitly, it is the I I -matrix C whose coefficient cij vanishes if there are no arrows
between i and j , equals 2 if i D j , equals v.˛/1 if there is an arrow ˛ W i ! j and
equals v.˛/2 if there is an arrow ˛ W j ! i . Thus, the Cartan matrix associated with
the above valued quiver BÅ2 equals

2 2
:
1 2
Fomin–Zelevinsky have shown in [46] that the analogue of Theorem 3.1 holds for
valued quivers. In particular, the Laurent phenomenon holds and the cluster algebra
associated with a valued quiver Q has only finitely many cluster variables iff Q is
mutation-equivalent to a valued quiver whose associated Cartan matrix corresponds to
a finite root system.
134 B. Keller
For valued quivers, the Independence Conjecture 3.2 is open except for the valued
quivers treated by Demonet [23], [24]. The Positivity Conjecture 3.3 is open except in
rank two, where it was shown by Dupont in [28].
4 Cluster algebras of geometric type

We will slightly generalize the definition of Section 3 in order to obtain the class of
‘skew-symmetrizable cluster algebras of geometric type’. This class contains many
algebras of geometric origin which are equipped with ‘dual semi-canonical bases’
[100]. The construction of a large part of such a basis in [58] is one of the most
remarkable applications of cluster algebras so far.
We refer to Section 5.7 for the definition of the ‘skew-symmetrizable cluster alge-
bras with coefficients in a semi-field’, which constitute so far the most general class
considered.
4.1 Definition. Let 1 n m be integers. Let Q z be an ice quiver of type .n; m/,
i.e. a quiver with m vertices and which does not have any arrows between vertices i ,
j which are both strictly greater than n. The principal part of Qz is the full subquiver
Q whose vertices are 1, …, n (a subquiver is full if, with any two vertices, it contains
all the arrows linking them). The vertices n C 1, …, m are called the frozen vertices.
The cluster algebra associated with the ice quiver Qz
AQz Q.x1 ; : : : ; xm /
is defined in the same manner as the cluster algebra associated with a quiver (Section 3)
but
• only mutations with respect to non frozen vertices are allowed and no arrows
between frozen vertices are added in the mutations;
• the variables xnC1 , …, xm , which belong to all clusters, are called coefficients
rather than cluster variables;
• the cluster type of the ice quiver is that of its principal part (if it is defined).
Notice that the datum of Q z is equivalent to that of the integer m n-matrix Bz whose
coefficient bij is the difference of the number of arrows from i to j minus the number
of arrows from j to i for all 1 i m and all 1 j n. The top n n part B
of Bz is called its principal part. In complete analogy, one defines the cluster algebra
associated with a valued ice quiver respectively with an integer m n-matrix whose
principal part is skew-symmetrizable.
We have the following sharpening of the Laurent phenomenon proved in Proposi-
tion 11.2 of [46].
Theorem 4.1 ([46]). Each cluster variable in AQz is a Laurent polynomial in the initial
variables x1 , …, xn with coefficients in ZŒxnC1 ; : : : ; xm .
Often one considers localizations of AQz obtained by inverting some or all of the
coefficients. If K is an extension field of Q and A a commutative K-algebra without
zero divisors, a cluster structure of type Qz on A is given by an isomorphism ' from
AQz ˝Q K onto A. Such an isomorphism is determined by the images of the coefficients
and of the initial cluster variables '.xi /, 1 i m. We call the datum of Q z and
of the '.xi / an initial seed for A. The following proposition is a reformulation of
Proposition 11.1 of [46], cf. also Proposition 1 of [133]:
Proposition 4.2. Let X be a rational quasi-affine irreducible algebraic variety over

C. Let Qz be an ice quiver of type .m; n/. Assume that we are given a regular function
'c on X for each coefficient c D xi , n < i m, and a regular function 'x on X for
each cluster variable x of AQz such that
a) the dimension of X equals m;
b) the functions 'x and 'c generate the coordinate algebra CŒX ;
c) the correspondence x 7! 'x , c 7! 'c takes each exchange relation of AQz to an

equality in CŒX .
Then the correspondence x 7! 'x , c 7! 'c extends to an algebra isomorphism

' W AQz ˝Q C ! CŒX so that CŒX carries a cluster algebra structure of type
z
Q with initial seed 'xi , 1 i m.
4.2 Example: Planes in a vector space. Let n 1 be an integer. Let A be the

algebra of polynomial functions on the cone over the Grassmannian of planes in C nC3 .
This algebra is generated by the Plücker coordinates xij , 1 i < j n C 3, subject
to the Plücker relations: for each quadruple of integers i < j < k < l between 1 and
n C 3, we have
xik xj l D xij xkl C xj k xil : (13)
Notice that the monomials in this relation are naturally associated with the diagonals
and the sides of the square
i= j
O === O
O == O
O == O

l k.
The idea is to interpret this relation as an exchange relation in a cluster algebra with
coefficients. In order to describe this algebra, let us consider, in the euclidean plane, a
regular polygon P whose vertices are numbered from 1 to n C 3. Consider the variable
xij as associated with the segment Œij which links the vertices i and j .
136 B. Keller
Proposition 4.3 ([46], Example 12.6). The algebra A has a cluster algebra structure
such that
- the coefficients are the variables xij associated with the sides of P ;
- the cluster variables are the variables xij associated with the diagonals of P ;
- the clusters are the n-tuples of cluster variables corresponding to diagonals

which form a triangulation of P .
Moreover, the exchange relations are exactly the Plücker relations and the cluster type
is An .
A triangulation of P determines an initial seed for the cluster algebra and the
z
exchange relations satisfied by the initial cluster variables determine the ice quiver Q.
For example, one can check that in the following picture, the triangulation and the ice
quiver (whose frozen vertices are in boxes) correspond to each other.
1O
ooo 44OOO
16 12
U,,
ooo 44 OO ,,
6 44 2
44 / 04 gO 02 / 23

44 56 O woo O
4
03
? ???
5 NNN 3
NNN ooooo
o 45 34
4
The hypotheses of Proposition 4.2 are straightforward to check in this example. Many
other (homogeneous) coordinate algebras of classical algebraic varieties admit cluster
algebra structures (or ‘upper cluster algebra structures’) and in particular the Grass-
mannians [133], cf. Section 4.3 below, and the double Bruhat cells [9]. Some of
these algebras have only finitely many cluster variables and thus a well-defined cluster
type. Here is a list of some examples of varieties and their cluster type extracted from
[47], where N denotes a maximal unipotent subgroup of the corresponding reductive
algebraic group:
Gr2;nC3 Gr3;6 Gr3;7 Gr3;8

An D4 E6 E8
SL3 =N SL4 =N SL5 =N Sp4 =N SL2 SL3

A1 A3 D6 B2 A1 D4
A theorem analogous to Proposition 4.3 for ‘reduced double Bruhat cells’ is due to
Yang and Zelevinsky [141]. They thus obtain a cluster algebra (with principal coeffi-
cients) with an explicit description of the cluster variables for each Dynkin diagram.
4.3 Example: The Grassmannian Gr.3; 6/. Let us consider the cone X over the
Plücker embedding of the variety Gr.3; 6/ of 3-dimensional subspaces in 6-dimensional
complex space C 6 , considered as a space of rows. The Plücker coordinates of the
subspace generated by the rows of a complex 3 6-matrix are the 3 3-minors of the
matrix, i.e. the determinants D.j/ of the 3 3-submatrices formed by the columns with
indices in a 3-element subset j of f1; : : : ; 6g. It is a particular case of Scott’s theorem
[133], cf. also Example 10.3 of [59], that the algebra CŒX admits a cluster algebra
structure of the type
123
??
?
124_? / 125 / 126
?? _??
?? ??
?? ??
?? ??
?? ??

134_? / 145_? / 156
?? ??
?? ??
?? ??
?? ??
? ?
234 / 345 / 456
whose initial seed is given by the minors D.j/ associated with the vertices j of this
quiver (frozen vertices appear in boxes). If we mutate the principal part of this quiver
at the vertex 124, we obtain a Dynkin quiver of type D4 , which is thus the cluster type
of this cluster algebra. It admits 4 C 12 D 16 cluster variables. As shown in [133],
fourteen among these are minors and the remaining two are
X1 D jP1 ^ Q1 ; P2 ^ Q2 ; P3 ^ Q3 j and X2 D jQ1 ^ P2 ; Q2 ^ P3 ; Q3 ^ P1 j;
where we denote the columns of our matrix by P1 , Q1 , P2 , Q2 , P3 , Q3 (in this order)

and write jj for the determinant. In this cluster algebra, we have the remarkable identity
[142]
jP1 P2 Q2 jjP2 P3 Q3 jjP3 P1 Q1 j jP1 P2 Q1 jjP2 P3 Q2 jjP3 P1 Q3 j D jP1 P2 P3 j X1 ;

(14)
which we can rewrite as
D.134/D.356/D.125/ C D.123/D.345/D.156/ D D.135/X1 :
This is in fact an exchange relation in our cluster algebra (many thanks to B. Leclerc
for pointing this out): Indeed, if we successively mutate the initial seed at the vertices
124 and 145, we obtain the cluster
D.135/; D.125/; D.356/; D.134/

138 B. Keller
(exercise: compute the corresponding quiver!) and if we now mutate at the variable
D.135/, we obtain X1 and the exchange relation (14). This relation appears implicitly
in [66] and finding a suitable generalization to higher dimensions would be of interest
in view of Zagier’s conjecture [143].
4.4 Example: Rectangular matrices. Polynomial algebras admit many interesting

cluster algebra structures. As a representative example, let us consider such a structure
on the algebra A of polynomial functions on the space of complex 4 5-matrices. For
1 i 4 and 1 j 5, let D.ij / be the determinant of the largest square submatrix
of a 4 5-matrix whose upper left corner is the .i; j /-coefficient. Then the algebra A
admits a cluster structure of type Qz
11 o 12 o 13 o 14 o 15
O ?? O ?? O ?? O ?? O
?? ?? ?? ??
?? ?? ?? ??

21 o 22O ?o 23O ?o 24O ?o 25
O ?? ?? ?? ?? O
?? ?? ?? ??
?? ?? ?? ??

31 o 32O ?o 33O ?o 34O ?o 35 O
O ?? ?? ?? ??
?? ?? ?? ??
?? ?? ?? ??

41 o 42 o 43 o 44 o 45
whose initial seed is formed by the functions D.ij / associated with the vertices of
the quiver Q. z This is a particular case of a theorem of Geiss–Leclerc–Schröer [57].
Perhaps the most remarkable fact is that iterated mutations of the initial seed still
produce polynomials in the matrix coefficients (and not fractions). Geiss–Leclerc–
Schröer’s proof of this fact in [57] is ultimately based on Lusztig’s results [100]. They
sketch a more elementary approach in Section 7.3 of [55], cf. Section 4.6 below. It
is not obvious either that the cluster variables generate the polynomial ring. To prove
it, we first notice that the variables x25 , x35 , x42 , x43 , x44 , x45 already belong to the
initial seed. Now, following [57], we consider the sequence of mutations at the vertices
45; 44; 43; 42; 35; 25I 34; 33; 32; 24I 23; 22I 45; 44; 43; 35I 34; 33I 45; 44:
The sequence naturally splits into ‘hooks’, which we have separated by semicolons.
The cluster variables which appear successively under this sequence of mutations are
x34 ; x33 ; x32 ; : : : ; x24 ; : : : ; x23 ; x22 ; : : : ;
where we have only indicated those variables associated with mutations at the vertices
of the lower right rim: 25, 35, 42, 43, 44, 45. So we see that in fact all the functions
xij are cluster variables.
4.5 Finite generation. In general, cluster algebras are not finitely generated as alge-
bras. For example, consider the cluster algebra AQ associated with the quiver
@ 2 ?
@ ????????
????
???

oo
1 3:
Let us show, following [102], that AQ is not even Noetherian. Indeed, up to isomor-
phism, the quiver Q is invariant under mutations. Hence all exchange relations are of
the form
uk u0k D u2i C uj2
for three pairwise distinct indices i, j and k. It follows that AQ admits a grading
such that all cluster variables have degree 1. Since Q is not mutation-equivalent to
a Dynkin quiver, by Theorem 3.1, there are infinitely many cluster variables and by
Conjecture 3.2, proved in [19], they are linearly independent over the field Q, which is
the degree 0 part of AQ . But a positively graded commutative algebra whose degree
1 part is not a finitely generated module over its degree 0 part cannot be Noetherian.
Many more examples are provided by the following theorem.
Theorem 4.4 (Theorem 1.24 of [9]). If Q is a valued quiver with three vertices, the
cluster algebra AQ is finitely generated over the rationals if and only if Q is mutation-
equivalent to an acyclic valued quiver.
For an acyclic valued quiver with n vertices, the cluster algebra AQ admits a set of
2n generators. More precisely, we have the following theorem.
Theorem 4.5 (Corollary 1.21 of [9]). If Q is acyclic with n vertices, the cluster algebra
AQ is generated over the rationals by the initial variables x1 , …, xn and the cluster
variables xj0 , 1 j n, obtained by mutating the initial seed at each vertex j .
Moreover, by Corollary 1.21 of [9], if Q is acyclic, the generators x1 , …, xn , x10 ,

…, xn0 together with the exchange relations between xj and xj0 , 1 j n, form a
presentation of AQ and the monomials in the generators not containing any product
xj xj0 form a Q-basis.
The class of ‘locally ayclic’ cluster algebras is introduced in [102]. It contains
all ayclic cluster algebras. As shown in [102], each locally acyclic cluster algebra is
finitely generated, integrally closed and locally a complete intersection.
4.6 Factoriality. In general, cluster algebras need not be factorial, even when the
exchange matrix is of full rank. The following example, based on an idea of P. Lampe,
is given in [55]. Let Q be the generalized Kronecker quiver
//
1 /2
140 B. Keller
and x10 the cluster variable obtained by mutating the inital seed at the vertex 1. Then
we have
x1 x10 D 1 C x23 D .1 C x2 /.1 x2 C x22 /
and one can show that these are essentially different factorizations of the product x1 x10
in AQ , cf. Proposition 6.3 of [55].
Now let Q z be a valued ice quiver of type .n; m/ and let n p m be an integer.
Let P be the polynomial ring ZŒxnC1 ; : : : ; xm and L its localization at xnC1 , …, xp .
Let
A D AQz ˝P L
be the localization of the cluster algebra AQz at xnC1 , …, xp . Notice that the invertible
elements of L are the Laurent monomials in xnC1 , …, xp multiplied by ˙1.
Theorem 4.6 ([55]). a) The invertible elements of A are those of L.
b) Each cluster variable of A is irreducible and two cluster variables are associate
iff they are equal.
As an application, let us show that the cluster algebra associated with a Dynkin
quiver of type A3 is not factorial. Indeed, consider the cluster algebra A associated
with the quiver
QW 1 /2 / 3:
Let x10 and x30 be the cluster variables obtained from the initial seed by mutating
respectively at the vertices 1 and 3. We have
1 C x2 1 C x2
x10 D and x30 D
x1 x3
and therefore
x10 x1 D x30 x3 :
Since x1 , x10 , x3 , x30 are pairwise distinct cluster variables, it follows from the theorem
that these are essentially distinct factorizations.
Despite these examples, many cluster algebras appearing ‘in nature’ are in fact
factorial. The following theorem often allows to check this.
Theorem 4.7 ([55]). As above, let A be the cluster algebra associated with a valued
ice quiver of type .n; m/ localized at a subset xnC1 , …, xp of the set of coefficients. Let
y and z be disjoint clusters and U A a subalgebra which is factorial and contains
y, z and the localized coefficient algebra L. Then A equals U and an element x
of the ambient field Q.x1 ; : : : ; xm / belongs to A iff it is a Laurent polynomial with
coefficients in L both in y and in z.
As a prototypical example, consider the ice quiver
zW 1
Q /2 /3 / 4 :
We will parametrise its coefficient and its cluster variables by the vertices of the fol-
lowing quiver:
04
= BBB
||| !
=| 03 BB =| 13 BB
|| B! || B!
02 12 22
|= BBB |= BBB |= BBB
|| ! || ! || !
01 11 21 31 :
Namely, to a vertex ij , we associate a cluster variable xi;j in such a way that x0;j
equals xj , 1 j 4, and each ‘mesh’ gives rise to an exchange relation: We have
xi;1 xiC1;1 D xi;2 C 1 for 0 i 2
and
xi;j xiC1;j D xi;j C1 xiC1;j 1 C 1
for all vertices ij among 02, 03, 12. Then the set of cluster variables is the set of the
xi;j , where ij runs through the vertices other than 04. The variables at the bottom are
1 C x2 x1 C x3 x2 C x4
x0;1 D x1 ; x1;1 D ; x2;1 D ; x3;1 D :
x1 x2 x3
They are algebraically independent and the polynomial ring
U D ZŒx0;1 ; x1;1 ; x2;1 ; x3;1
contains the disjoint clusters y D fx1 ; x2 ; x3 g and z D fx1;3 ; x2;2 ; x3;1 g appearing on
the left and the right rim. We see from the theorem that U equals the cluster algebra
and that an element of the ambient field belongs to the cluster algebra iff it is a Laurent
polynomial with coefficients in ZŒx4 both in y and in z. We refer to Section 7.3 of
[55] for more elaborate examples arising as coordinate algebras of unipotent cells in
Kac–Moody groups.
5 General cluster algebras

5.1 Parametrization of seeds by the n-regular tree. Let us introduce a convenient
parametrization of the seeds in the mutation class of a given initial seed. Let 1 n m
be integers and Q z a valued ice quiver of type .n; m/. Let X D fx1 ; : : : ; xm g be the
z X / the initial seed. Let Tn be the n-regular tree: Its edges are
initial cluster and .Q;
labeled by the integers 1, …, n such that the n edges emanating from each vertex carry
different labels, cf. Figure 1. Let t0 be a vertex of Tn . To each vertex t of Tn we
z /; X.t // such that at t D t0 , we have the initial seed and whenever
associate a seed .Q.t
t is linked to t by an edge labeled k, the seeds associated with t and t 0 are related by
0
the mutation at k. We write xi .t /, 1 i n, for the cluster variables in the seed X.t /.
If Bz is the m n-matrix associated with Q, z we write B.t
z / for the matrix associated
z
with Q.t/.
142 B. Keller
•
5 •
4
• • 1 •
• 3
5
4 •
• 1 •
2
2
3 •
• •
5
t0 4
• 1 •
3
2 •
• 4 •
5
5
• 1 •
3 •
2 • 4
• • 1 •
3
2 •
•
Figure 1. A picture, up to depth 4, of the 5-regular tree.
5.2 Principal coefficients. Let n 1 be an integer and Q a valued quiver with n

vertices. Let B be the associated skew-symmetrizable integer n n-matrix. In the next
subsections, following [49], we will define data associated with Q which are relevant
z whose
for all cluster algebras with coefficients associated with valued ice quivers Q
principal part is Q. This will become apparent from a general formula expressing the
cluster variables in terms of these data, cf. Section 5.7.
5.3 Principal coefficients: c-vectors. Let Qpr be the principal extension of Q, i.e.
the valued quiver obtained from Q by adding new vertices n C 1, …, 2n and new
arrows i C n ! i, 1 i n, for each vertex i of Q. For example, if we have
10 20
QW 1 /2; then Qpr W

1 / 2,
where we write i 0 for i C n. The cluster algebra with principal coefficients associated
with Q is the cluster algebra associated with Qpr . We write Bpr for the corresponding
integer 2n n-matrix. It is obtained from B by appending an n n identity matrix at
the bottom:
B
Bpr D :
In
For a vertex t of the n-regular tree, the matrix of c-vectors C.t / is by definition the
n n-matrix appearing in the bottom part of Bpr .t /, so that we have

B.t /
Bpr .t / D :
C.t /
Its columns are the c-vectors at t. When necessary, we will denote the matrix C.t / by
C.B; t0 ; t/ to clarify its dependence on B and the sequence of mutations linking t0 to
t. For example, if we successively mutate the quiver Qpr associated with Q W 1 ! 2
at the vertices 1, 2, 1, …, we obtain the sequence
10 20 1O 0 @ 20 10 _@ 2O 0 10 _@ ? 20 10 @ ? 20 10 @ 20
@@ @~@~ @~@~ @~@~ @~
7! @ 7! ~ ~ @ 7! ~~ @ 7! ~~ 7! ~~~@@ ; (15)
@
()*+
/.-,
1 /2 1 o /.-, ()*+
2 ()*+
/.-,
1 /2 1 o /.-, ()*+
2 ()*+
/.-,
1 /2 1o 2
which yields the sequence of matrices of c-vectors

1 0 1 1 0 1 0 1 0 1 0 1
; ; ; ; ; : (16)
0 1 0 1 1 1 1 0 1 0 1 0
Notice that in total we find 6 distinct c-vectors and that these are in natural bijection
with the (positive and negative) roots of the root system corresponding to the underlying
graph A2 of the quiver Q: We simply map a c-vector with components c1 and c2 to
the root c1 ˛1 C c2 ˛2 , where ˛1 and ˛2 are the simple roots.
˛X22
22 F
22
2
o B2 / ˛1
222

22

As shown in [123], cf. also [136], this bijection generalizes to all cluster-finite cluster
algebras. In particular, we see that in these examples, each c-vector is non zero and has
all its components of the same sign. This is conjectured to be true in full generality:
Main Conjecture 5.1 ([49]). Each c-vector associated with a valued quiver is non
zero and has either all components non negative or all components non positive.
For equally valued quivers, this conjecture follows from the results of [26], which
are based on categorification using decorated representations of quivers with potential,
cf. below. Two different proofs were given in [118] and, up to a technical extra hy-
pothesis which is most probably superfluous, in [107]. In the case of valued quivers,
the conjecture is open in general, but known to be true in many important cases thanks
to the work of Demonet [23]. The determination of the c-vectors for general quivers
seems to be an open problem. A non acyclic example is computed in [111].
144 B. Keller
5.4 Principal coefficients: F -polynomials and g-vectors. We keep the above no-
tations Q, B, Qpr and Bpr . By the sharpened Laurent phenomenon (Theorem 4.1),
each cluster variable of the cluster algebra A.Qpr / associated with Qpr is a Laurent
polynomial in x1 , …, xn with coefficients in ZŒxnC1 ; : : : ; x2n . In other words, for
each vertex t of the n-regular tree and each 1 j n, the cluster variable xj .t /
belongs to the ring
ZŒx1˙1 ; : : : ; xn˙1 ; xnC1 ; : : : ; x2n :
The F -polynomial
Fj .t / 2 ZŒxnC1 ; : : : ; x2n
is by definition the specialization of xj .t / at x1 D 1, x2 D 1, …, xn D 1.
To define the g-vectors, let us endow the ring
ZŒx1˙1 ; : : : ; xn˙1 ; xnC1 ; : : : ; x2n
with the Zn -grading such that
deg.xj / D ej and deg.xnCj / D Bej for 1 j n:
For each vertex t of the n-regular tree and each 1 j n, the cluster variable xj .t /
of A.Qpr / is in fact homogeneous for this grading (Proposition 6.1 of [49]). Its degree
is by definition the g-vector gj .t /. The matrix of g-vectors G.t / has as its columns the
vectors gj .t/. When necessary, we will denote this matrix by G.B; t0 ; t / to clarify its
dependence on B and the sequence of mutations linking t0 to t .
For example, if B is associated with Q W 1 ! 2 and we mutate along the path
1 2 1 2 1
t0 t1 t2 t3 t4 t5
in the 2-regular tree, then, in addition to the g-vectors g1 .t0 / D e1 and g2 .t0 / D e2
and the F -polynomials F1 .t0 / D F2 .t0 / D 1 associated with the initial variables,
we successively find the following cluster variables in A.Bpr / and the corresponding
F -polynomials and g-vectors:
x2 C x3
x1 .t1 / D ; F1 .t1 / D 1 C x3 ; g1 .t1 / D e2 e1 ;
x1
x2 C x3 C x1 x3 x4
x2 .t2 / D ; F2 .t2 / D 1 C x3 C x3 x4 ; g2 .t2 / D e1 ;
x1 x2
1 C x1 x4
x1 .t3 / D ; F1 .t3 / D 1 C x4 ; g1 .t3 / D e2 ;
x2
x2 .t4 / D x1 ; F2 .t4 / D 1; g2 .t4 / D e1 ;
x1 .t5 / D x2 ; F1 .t5 / D 1; g1 .t5 / D e2 :
The associated G-matrices are

1 0 1 0 1 1 0 1 0 1 0 1
; ; ; ; ; : (17)
0 1 1 1 1 0 1 0 1 0 1 0
If we let ˛1 and ˛2 be the simple roots of the root system A2 , then clearly the linear map
which takes e1 to ˛1 and e2 to ˛1 C ˛2 yields a bijection from the set of the g-vectors
to the set of almost positive roots, i.e. the union of the set of positive roots with the set
of opposites of the simple roots, cf. Figure 2.
e2 e1 e2 ˛2 ˛1 C ˛ 2
e1 • e1 ˛1 • ˛1
e2 ˛2
Figure 2. g-vectors and almost positive roots for A2 .
This statement generalizes to an acyclic (equally valued) quiver Q as follows:

For two vertices i; j of Q, let pij be the number of paths from i to j (i.e. formal
compositions of 0 arrows). Let ˛1 , …, ˛n be the simple roots of the root system
corresponding to the underlying graph of Q. The following theorem is a consequence
of the results of [12].
P
Theorem 5.2. The linear map taking ej to niD1 pij ˛i , 1 j n, is a bijection from
the set of g-vectors of Q to the union of the set of real ( positive) Schur roots with the
set of negative simple roots.
5.5 Tropical duality. Let Q be a valued quiver, B the associated skew-symmetrizable

nn-matrix and D a diagonal integer nn-matrix with strictly positive diagonal entries
such that the transpose .DB/T of DB equals DB. The opposite valued quiver Qop
corresponds to the matrix B. For example, the opposite valued quiver of
.1;2/ .2;1/
BÅ3 W 1 /2 /3 BÅ3 W 1 o 2o
op
is 3,
which is in fact mutation equivalent to BÅ3 (we mutate at 1 and 3).

Theorem 5.3 ([116]). Suppose that the main Conjecture 5.1 holds for Q. Then for
each vertex t of the n-regular tree, we have
G.t /T D C.t / D D; (18)
C.t/1 D C.Q.t /op ; t; t0 / and G.t /1 D G.Q.t /op ; t; t0 /: (19)

To check the equality (18) in the example of the quiver 1 ! 2, the reader may
inspect the C - and G-matrices given in (16) and (17). The equalities (19) are given
146 B. Keller
in Theorem 1.2 of [116]. The equality (18) is equation (3.11) from [116], cf. also
Proposition 3.2 of [115]. For skew-symmetric matrices B, it was first proved using
Plamondon’s results [118] in Proposition 4.1 of [114] by T. Nakanishi, who had dis-
covered the statement by combining in Corollary 6.10 and 6.11 of [82].
Let v W Q1 ! N 2 denote the valuation of the valued quiver Q, cf. Section 3.3. Fol-
lowing [37], we define the Langlands dual Q_ as the valued quiver whose underlying
oriented graph equals that of Q and whose valuation v _ is defined by reversing the
valuation of Q: For each arrow ˛, we put
v _ .˛/ D .v.˛/2 ; v.˛/1 /: (20)
The corresponding skew-symmetrizable matrix B _ equals B T . For example, if Q is
the valued quiver
.1;2/
BÅ3 W 1 /2 / 3,
_
then Q is
.2;1/
CÅ3 W 1 /2 / 3.
Theorem 5.4 (Theorem 1.2 of [116]). Suppose that the Main Conjecture 5.1 holds for
Q. Then for each vertex t of the n-regular tree, we have
G.Q; t0 ; t /T D C.Q_ ; t0 ; t /1 :
For example, if we successively mutate the principal extension of the above valued
quiver CÅ3 at the vertices 1, 2, 3, 1, 2, 3, we find the valued quiver
4 _? 5 6
?? _??? oo ooo O
?? ???ooo
?? ooo
?

?? oo
1;2 1;2
??
ooooo??? ??
??

wooo o ?
1 /2 2;1 /3
and hence the C -matrix

2 3
1 1 0
C.CÅ3 ; t0 ; t / D 41 0 25 :
1 0 1
On the other hand, if successively mutate the initial seed of the principal extension of
BÅ3 at 1, 2, 3, 1, 2, 3, we find the cluster
1 2
x1 .t / D .x C x1 x 5 /;
x2 3
1
x2 .t / D .x 2 x 2 x4 x52 C 2x12 x2 x4 x52 x6 C C x34 x4 /;
x1 x22 x32 1 2
1
x3 .t / D .x 2 C x1 x5 C x1 x2 x5 x6 /
x2 x3 3
and thus the G-matrix

2 3
0 1 0
G.BÅ3 ; t0 ; t / D 41 1 15 :
2 2 1
This is indeed the inverse transpose of C.BÅ3 ; t0 ; t /. This was to be expected by theo-
rem 5.4 since the main conjecture holds for CÅ3 by Demonet’s work [24], [23].
5.6 Product formulas for c-matrices and g-matrices. We will give a key ingredient
for the proof of theorem 5.4 which is also useful in the investigation of quantum cluster
algebras (Section 6.2). Let Q be a valued quiver, B the associated skew-symmetri-
zable nn-matrix and D a diagonal integer nn-matrix with strictly positive diagonal
entries such that the transpose .DB/T of DB equals DB. 1 k n be an integer.
Choose a sign " equal to 1 or 1. Let F" D Fk;" .Q/ be the n n-matrix which differs
from the identity matrix only in its k-th row, whose coefficients are given by
´
1 if j D kI
.F" /kj D
Œ"bkj C if j ¤ k:
Let E" D Ek;" .Q/ be the n n-matrix which differs from the identity matrix only in
its k-th column, whose coefficients are given by
´
1 if i D kI
.E" /ik D
Œ"bik C if i ¤ k:
Notice that both E" and F" square to the identity matrix. Parts a) to d) of the following
lemma become natural in the categorical picture to be developed in Section 7, cf.
Corollary 7.5. Part e) seems harder to interpret.
Lemma 5.5. a) We have E" k .B/ D BF" and E"T DF" D D.
b) We have Ek;" .k .Q// D Ek;" .Q/1 and Fk;" .k .Q// D Fk;" .Q/1 .
c) For 1 k n, let Tk D Ek;" .k .Q//Ek;" .Q/. Then for two vertices i , j , the
matrices Ti and Tj satisfy the braid relation associated with the full valued subquiver
whose vertices are i and j , i.e. we have
Ti Tj Ti : : : D Tj Ti Tj : : : ; (21)
„ ƒ‚ … „ ƒ‚ …
m factors m factors
where the number of factors m equals 2, 3, 4 or 6 depending on whether jbij bj i j equals

0, 1, 2 or 3.
d) We have Ek;" .Qop / D Ek;" .Q/, Fk;" .Qop / D Fk;" .Q/.
e) We have Ek;" .Q_ /T D Fk;" .Q/.
148 B. Keller
Now let
i1 i2 i3 iN
t0 t1 t2 ::: tN .
be a path in the n-regular tree, let "s be the sign of the c-vector C.ts1 /es and let
Eis ;"s .ts / resp. Fis ;"s .ts / be the matrix E"s resp. F"s associated with the quiver Q.ts1 /
and the vertex is , 1 s N .
Theorem 5.6 ([116]). If the Main Conjecture 5.1 holds for Q, we have
G.tN / D Ei1 ;"1 .t1 / : : : EiN ;"N .tN / and C.tN / D Fi1 ;"1 .t1 / : : : FiN ;"N .tN /:
5.7 Cluster algebras with coefficients in a semifield. A semifield is an abelian group

P endowed with an additional binary operation ˚ W P P ! P which is commutative,
associative and distributive with respect to the group law of P . For example, the tropical
semifield Trop.u1 ; : : : ; un / is the free (multiplicative) abelian group generated by the
indeterminates ui endowed with the operation ˚ defined by
Y Y Y
l m min.l ;m /
uii ˚ ui i D ui i i :
Clearly, it is isomorphic to Zntrop , where Ztrop is the abelian group Z endowed with the
operation ˚ defined by x ˚ y D min.x; y/. It is shown in Lemma 2.1.6 of [8] that the
universal semifield Qsf .x1 ; : : : ; xn / on given indeterminates x1 , …, xn is the closure,
in Q.x1 ; : : : ; xn /, of the set fx1 ; : : : ; xn g under multiplication, division and addition.
Notice that this closure contains polynomials whose coefficients are not all positive;
for example, the polynomial
x3 C 1
x2 x C 1 D
xC1
belongs to Qsf .x/. The abelian group underlying a semifield P is torsion-free. Indeed,
if an element x satisfies x m D 1, then
x m ˚ x m1 ˚ ˚ x 1 ˚ x m1 ˚ ˚ x
xD D D 1:
x m1 ˚ x m2 ˚ ˚ 1 x m1 ˚ x m2 ˚ ˚ 1
Thus, the group ring ZP is integral.
Let us fix a semifield P and an integer n 1. A Y -seed of rank n with values in P
is a pair .Q; Y / formed by a valued quiver Q with n vertices and by a sequence Y D
.y1 ; : : : ; yn / of elements of P . Let B be the skew-symmetrizable matrix corresponding
to Q. If .Q; Y / is a Y -seed and k a vertex of Q, the mutated Y -seed k .Q; Y / is the
Y -seed .Q0 ; Y 0 / where Q0 D k .Q/ and, for 1 j n, we have
´
0 yk1 if j D k;
yj D Œbkj C bkj
(22)
yj yk .1 ˚ yk / if j ¤ k.
One checks that 2k .Q; Y / D .Q; Y /. For example, the following Y -seeds are related
by a mutation at the vertex 1:
y1 / y2 1=y1 o y2 =.1 ˚ y11 /

O | O O
|||
||
~|
|
y3 / y4 , / y4 ,
y3 .1 ˚ y1 /
where we write the element yi in place of the vertex i .

Let QP be the fraction field of the group ring ZP and F any field obtained from
QP by adjoining n indeterminates. A seed with coefficients in P is a triple .Q; Y; X /,
where .Q; Y / is a Y -seed of rank n with values in P and X is a sequence .x1 ; : : : ; xn /
of elements of F which freely generate the field F . If .Q; Y; X / is a seed and k a
vertex of Q, the mutation k .Q; Y; X / is the seed formed by the mutation k .Q; Y /
and the sequence X 0 with xj0 D xj for j ¤ k and xk0 defined by the exchange relation
Y Œbi k C
Y Œbi k C
xk0 xk .1 ˚ yk / D yk xi C xi : (23)
i i
A seed pattern is the datum, for each vertex t of the n-regular tree, of a seed
.Q.t/; Y.t/; X.t // such that if t and t 0 are linked by an edge labeled k, then the seeds
corresponding to t and t 0 are linked by the mutation at k. The cluster algebra is the
ZP -subalgebra of the field F generated by the cluster variables.
We recover the cluster algebra of geometric type associated with an m n-matrix
Bz as follows: We let B be the principal part of B; z we define the semifield P to be the
tropical semifield Trop.xnC1 ; : : : ; xm / and the initial Y -variables to be
Y
m
b
yj D xi ij ; 1 j n:
iDnC1
As a simple example of a cluster algebra of ‘non geometric’ type, consider the case
where n D 1, P D Qsf .y1 ; y2 / and Q W 1 ! 2. Then the sequence of mutations
1 2 1 2 1
t0 t1 t2 t3 t4 t5
starting from the initial seed .1 ! 2; fx1 ; x2 g; fy1 ; y2 g/ yields

1 y1 y2 y1 C x2
y1 .t1 / D ; y2 .t1 / D
; x1 .t1 / D ;
y1 1 C y1 x1 .1 C y1 /
y2 1 C y1 x1 y1 y2 C x2 C y1
y1 .t2 / D ; y2 .t2 / D ; x2 .t2 / D ;
1 C y1 C y1 y2 y1 y2 x1 x2 .1 C y1 C y2 y1 /
1 C y1 C y1 y2 1 x1 y2 C 1
y1 .t3 / D ; y2 .t3 / D ; x1 .t3 / D ;
y2 y1 .1 C y2 / x2 .1 C y2 /
150 B. Keller
1
y1 .t4 / D ; y2 .t4 / D y1 .1 C y2 /; x2 .t4 / D x1 ;
y2
y1 .t5 / D y2 ; y2 .t5 / D y1 ; x1 .t5 / D x2 :
5.8 The separation formulas. Let a seed pattern be given and let us write .Q; Y; X /
for the initial seed .Q.t0 /; Y .t0 /; X.t0 // associated with the chosen root t0 of the n-
regular tree. Let us write cij .t / for the coefficients of the c-matrix C.t / and gij .t / for
those of the g-matrix G.t / associated with a vertex t of the n-regular tree. Recall that
Fj .t / 2 ZŒxnC1 ; : : : ; x2n ; 1 j n;
are the F -polynomials at the vertex t . By construction, they belong to the universal
semifield Qsf .xnC1 ; : : : ; x2n / and thus it makes sense to consider their evaluations
Fj .t /.y1 ; : : : ; yn /
at the elements y1 ; : : : ; yn of P and more generally, at an n-tuple of elements of any

semifield.
Theorem 5.7 (Proposition 3.13 and Corollary 6.3 of [49]). For each vertex t of the
n-regular tree and each 1 j n, we have
c .t/ c .t/
Y
yj .t / D y11j : : : ynnj Fi .t /.y1 ; : : : ; yn /bij .t/ ; (24)
i
g .t/ g .t/ Fj .t /.yO1 ; : : : ; yOn /
xj .t / D x1 1j : : : xn nj ; (25)
Fj .t /.y1 ; : : : ; yn /
Q b
where yOl D yj i xi i l , 1 l n.
6 Quantum cluster algebras and quantum dilogarithms

6.1 The quantum dilogarithm. Let q 1=2 be an indeterminate. We will denote its
square by q. The (exponential of) the quantum dilogarithm series is
2
q 1=2 q n =2
E.y/ D Eq .y/ D 1 C y C C n C :
q1 .q 1/.q n1 1/ : : : .q 1/
It is a series in the indeterminate y with coefficients in the field Q.q 1=2 /. It is related
to the classical dilogarithm
X1 Z x
xn log.1 y/
Li2 .x/ D D dy; jxj < 1;
nD1
n2
0 y
by the asymptotic expansion

Li2 .y/
Eq .y/ exp
log.q/
when q goes to 1 . An easy computation shows that we have the functional equation
.1 C q 1=2 y/ E.y/ D E.qy/: (26)
The quantum dilogarithm is related to the classical q-exponential function by the sub-
1=2
stitution y 7! qq1 y. Therefore, as discovered by Schützenberger [132], if y1 and y2
are two indeterminates which q-commute, i.e. y1 y2 D qy2 y1 , then we have
E.y1 C y2 / D E.y2 /E.y1 /: (27)
In 1993, Faddeev, Kashaev and Volkov [32], [31] discovered that (26) and (27) together
imply the pentagon identity:
y1 y2 D qy2 y1 H) E.y1 /E.y2 / D E.y2 /E.q 1=2 y1 y2 /E.y1 /; (28)
cf. [140] for a recent account. Their main result states that this identity implies the
classical five-term identity

x xy y xy
L.x/ C L.y/ L.xy/ D L CL
1 xy 1 xy
for the Rogers dilogarithm
L.x/ D Li2 .x/ C log.1 x/ log.x/=2:
We refer to [114], [78] for more information on the many recent developments around
this subject and to [144] for more information on the dilogarithm function.
6.2 Quantum mutations and quantum cluster algebras. We will construct quantum
cluster algebras following Berenstein–Zelevinsky [10]. Quantum cluster algebras are
certain non commutative deformations of cluster algebras of geometric type. Let 1
n m be integers, Bz an integer m n-matrix with skew-symmetrizable principal part
B and ƒ a skew-symmetric integer m m-matrix. Let Q z and Q be the associated
valued ice quivers. Recall from Section 4 that the datum of Bz gives rise to a cluster
z is a compatible pair, i.e. we have
algebra of geometric type. Let us assume that .ƒ; B/
Bz T ƒ D ŒD 0; (29)
where D is a diagonal n n-matrix whose diagonal coefficients are strictly positive

integers. This will ensure that ƒ gives rise to a (non commutative) deformation of the
z We first need to define the mutation of compatible
cluster algebra associated with B.
152 B. Keller
pairs: Let 1 k n be an integer and choose a sign " equal to 1 or 1. In the
notations of Section 5.6, let F" be the n n-matrix Fk;" .Q/ and E" the m m-matrix
z The mutation k .B;
Ek;" .Q/. z ƒ/ is defined to be the compatible pair .Bz 0 ; ƒ0 / with
z "
Bz 0 D E" BF and ƒ0 D E"T ƒE" :
One checks that Bz 0 equals k .B/

z and that .ƒ0 ; Bz 0 / does not depend on the choice of "
and is again a compatible pair (with the same matrix D). One checks that mutation
of compatible pairs is an involution. Thus, given a compatible pair .B; z ƒ/, we can
z /; ƒ.t // to each vertex t of the n-regular tree such that
assign a compatible pair .B.t
the given pair is assigned to t0 and, whenever t and t 0 are linked by an edge labeled k,
the corresponding pairs are related by the mutation at k.
The quantum affine space Aƒ associated with ƒ is by definition the ZŒq ˙1=2 -
algebra generated by all symbols x ˛ , ˛ 2 N m , subject to the relations
1 T ƒˇ
x˛ xˇ D q 2 ˛ x ˛Cˇ :
The quantum torus Tƒ is defined similarly on generators x ˛ , ˛ 2 Zm . One checks that

the underlying ZŒq ˙1=2 -module of Aƒ resp. Tƒ is free on the basis formed by the x ˛ ,
˛ 2 N m resp. ˛ 2 Zm . The completed quantum affine space A y ƒ is the completion
˙1=2
of Aƒ with respect to the kernel of the projection AQ ! ZŒq . The algebras Aƒ
and Tƒ are Ore domains (cf. the Appendix to [10]) and so have a field of fractions Fƒ
whose elements are given by right fractions (or left fractions).
z ƒ; X /, where X is the sequence of the xi D x ei .
The initial quantum seed is .B;
Its mutation at k, where 1 k n, is .Bz 0 ; ƒ0 ; X 0 /, where the sequence X 0 is formed
by the xi , i ¤ k, and by the element xk0 defined by the quantum exchange relation
xk0 D x EC ek C x E ek : (30)
By part (3) of Proposition 4.7 of [10], there is a unique morphism of ZŒq ˙1=2 -algebras
#k W Aƒ0 ! Tƒ
taking xi to xi0 , 1 i m; moreover it is injective and induces an isomorphism

#k W Fƒ0
! Fƒ :
One checks that mutation of quantum seeds is an involution. Thus, with each vertex
z /; ƒ.t /; X.t // such that
t of the n-regular tree, one can associate a quantum seed .B.t
the initial quantum seed is associated with t0 and seeds with vertices t and t 0 linked by
an edge labeled k are related by a quantum mutation. The quantum cluster variables
are the xj .t/, 1 j n, associated with the vertices t of the n-regular tree. The
quantum cluster algebra is the ZŒq 1=2 -subalgebra of Fƒ generated by the quantum
cluster variables. We have the quantum Laurent phenomenon:
Theorem 6.1 (Corollary 5.2 of [10]). The quantum cluster variables are contained in
the quantum torus Tƒ .
We refer to [10], [67], [54] for examples of quantum cluster algebras. The exchange
graph of quantum seeds associated with .B;z ƒ/ is defined in analogy with the exchange
z cf. Section 3.2. The specialization map
graph of (classical) seeds associated with B,
ZŒq ˙1=2 ! Z
taking q 1=2 to 1 yields a morphism of ZŒq ˙1=2 -modules
Tƒ ! ZŒx1˙1 ; : : : ; xm
˙1

which takes quantum cluster variables to classical ones and induces a map from the
quantum exchange graph to the classical exchange graph.
Theorem 6.2 (Theorem 6.1 of [10]). The specialization at q 1=2 D 1 yields an isomor-
phism from the quantum to the classical exchange graph.
6.3 Fock–Goncharov’s separation formula. Recall that the numbers di are the co-
efficients of the diagonal matrix D appearing in the compatibility condition (29). We
z ƒ; X /.
consider the mutation at k of a given initial quantum seed .B;
Lemma 6.3 ([37]). We have the separation formulas
#k D Ad0 .Eq dk .yk // B 'k;C D Ad0 .Eq dk .yk1 //1 B 'k; ; (31)
where the right adjoint action Ad0 .u/ takes an element v to u1 vu, we put
z
yk D x Bek
and 'k;" W Tƒ0 ! Tƒ is the unique morphism of ZŒq ˙1=2 -algebras taking x ˛ to x E" ˛ .
Thus, we have separated the mutation isomorphism into a ‘tropical’ part and a
‘transcendental’ part. Notice that in order to give meaning to the formulas (31), we
need to embed the quantum tori into suitable localizations of completions of quantum
affine space. Using formula (26) one then checks the claim. Of course, one would like
to iterate this formula. The iteration should be meaningful in (at least) two ways:
(1) the product of the appearing power series should have a meaning, i.e. all the
series should live in a common completion of quantum affine space;
(2) the composition of the ‘tropical parts’ should have a meaning from the point of
view of ‘tropical’ cluster theory, as we have seen it in Sections 5.3 and 5.4.
In order to obtain both, it is essential to choose the sign ˙ in each factor carefully. This
can be achieved using the Main Conjecture 5.1.
154 B. Keller
6.4 The quantum separation formula. To simplify the notations, let us assume from
z ƒ/ is unitally compatible, i.e. equation (29) holds with D the n n-
now on that .B;
identity matrix. Let i D .i1 ; : : : ; iN / be a sequence of vertices in f1; : : : ; ng. Consider
the corresponding path in the n-regular tree
i1 i2 i3 iN
t0 t1 t2 ::: tN :
It yields a chain of mutation isomorphisms between the associated quantum tori:
#i #i #i #i

Tƒ o 1
Tƒ.t1 / o 2
Tƒ.t2 / o 3
::: o N
Tƒ.tN / :
Let us write ˆ.i/ for the composition of these isomorphisms. We would like to write
down a separation formula for ˆ.i/ which generalizes (31). We need some more
notation: For 1 s N , let ˇs be the c-vector C.ts1 /eis and let "s be the common
sign of the components of ˇs (cf. Section 5.4). For a vector ˛ in Zn , let us write E.˛/
z
for E.y ˛ /, where y ˛ D x B˛ .
Theorem 6.4 ([107]). Put
E.i/ D E."N ˇN /"N : : : E."1 ˇ1 /"1 ; (32)

'.i/ D 'i1 ;"1 B B 'iN ;"N : (33)
Then we have
ˆ.i/ D Ad0 .E.i// B '; (34)
the isomorphism ' sends x ˛ to x G.tN /˛ , where G.tN / is the g-matrix at tN (Section 5.4),
and Ad0 .E.i// acts on x ei by multiplication with the quantum F -polynomial of [139].
Notice that by construction all the vectors "s ˇs have non negative components so
that all the series E."s ˇs /"s belong to the same completion of quantum affine space. If
we replace the right adjoint action of E.i/ by the multiplication with the quantum F -
polynomials, we obtain Tran’s formula (Theorem 6.1 of [139]), which is the quantum
analogue of Fomin–Zelevinsky’s [49] separation formula (25). The theorem is due,
in a different language, to Nagao [107] (cf. also Theorem 5.1 in [30]). Alternatively,
using Theorem 5.6 and Tran’s formula, it is not hard to prove the analogous theorem
for arbitrary valued quivers Q for which the Main Conjecture 5.1 holds.
Let us keep the notations from Theorem 6.4. It is not hard to check that there is a
unique Q.q 1=2 /-algebra embedding
AB ! Tƒ
z
taking an element x ˛ to x B˛ (if D is not the identity matrix, it is an embedding
ADB ! Tƒ ). Thus, by construction, the product E.i/ lies in a completed quantum
y B and independent of the choice of the non principal

affine subspace isomorphic to A
part in B. For example, we can always choose Bz D Bpr , cf. Section 5.3 and
z

0 I
ƒD :
I BT
Theorem 6.5 ([81], [109]). a) If C.tN / is a permutation matrix, then E.i/ D 1.
b) If the opposite matrix C.tN / is a permutation matrix, then E.i/ 2 A y B is
Kontsevich–Soibelman’s non commutative Donaldson–Thomas invariant [90] associ-
ated with the quiver corresponding to B (when this invariant is defined, cf. Section 7.11).
Remark 6.6. One can sharpen part a) as follows: Let i and i 0 be two sequences of
vertices in f1; : : : ; ng and let t and t 0 be the end points of the corresponding paths in
the n-regular tree. Suppose that we have P C.t / D C.t 0 / for a permutation matrix
P . We will show in Section 7.11 that we then have E.i/ D E.i 0 /. Thus, if Q admits
some sequence i such that C.t / is a permutation matrix, then the series E.i/ 2 A y B is
independent of the choice of the sequence i with this property. We then call this series
the combinatorial DT-invariant associated with Q.
We will give a proof of the theorem and the remark in Section 7.11, cf. also Theo-
rem 3.5 in [78]. Let us illustrate the theorem on the example of the mutation sequence
i D .1; 2; 1; 2; 1/ of the quiver AÅ2 W 1 ! 2. We have computed the sequence of
c-matrices C.ts /, 1 s 5, in equation (16). We obtain
ˇ1 D e1 ; ˇ2 D e1 C e2 ; ˇ3 D e2 ; ˇ4 D e1 ; ˇ5 D e2 :
Since C.t5 / is the matrix of the transposition e1 $ e2 , part a) of the theorem yields
the identity
E.e2 /1 E.e1 /1 E.e2 /E.e1 C e2 /E.e1 / D 1;
which is of course equivalent to the pentagon identity (28). Since C.t3 / is the opposite
of the transposition matrix, we find that Kontsevich–Soibelman’s DT invariant equals
E.e2 /E.e1 C e2 /E.e1 /
for the quiver AÅ2 , as is well-known, cf. Example 2) in Section 6.4 of [90]. This example
can be generalized to any Dynkin quiver, which yields a family of quantum dilogarithm
identities due to Reineke [125], cf. also Corollary 1.7 in [81] and [121], [122]. Namely,
let be a simply laced Dynkin diagram and let Q be an alternating quiver (i.e. each
vertex is a source or a sink) whose underlying graph is . Let iC be the sequence of
sources of Q and i its sequence of sinks (in any order). Let
i D iC i iC : : : ;
„ ƒ‚ …
h factors
where h is the Coxeter number of and let i 0 D i iC . Let i .t0 / be the final vertex in
the path in the regular tree which starts at t0 and runs through the sequence of mutations i
156 B. Keller
starting at the leftmost vertex in the sequence. Then one can show that both C.i .t0 //
and C.i 0 .t0 // are permutation matrices and so the Kontsevich–Soibelman invariant
associated with Q equals
E.i/ D E.i 0 /;
which is Reineke’s identity associated with Q. One can further generalize this class as
follows: Let and 0 be two simply laced Dynkin diagrams and Å and Å 0 alternating
0
quivers with underlying graphs and . Let Q be the square product Å Å 0 as
defined in Section 3.3 of [82]. For example, the square product of the quivers
AÅ4 W 1 o 2 /3o 4,
pp 4
ppp
Å5 W 1 o
D 2 / 3 xpgN
NNN
NN
5.
is depicted in Figure 3. Let iC be the sequence of all source-sinks of Å Å 0 (i.e.

1
vertices .u; v/ such that u is a source in the full subquiver p2 .v/ and v a sink the full
BK /Bo BK /B

B X1 / B 1o B X1 / B 1
11 11 11 11
1 1 1 1
Bo BO /Bo BO

BO /Bo BO /B

Bo B /Bo B
Figure 3. The quiver AÅ4 D

Å5 .
subquiver p11 .u/, where the pi are the projections) and let i be the sequence of all
sink-sources. Let
i D iC i iC : : : and i 0 D i iC i : : : ;
„ ƒ‚ … „ ƒ‚ …
h factors h0 factors
where h is the Coxeter number of and h that of 0 . Again one can show that
0
both C.i .t0 // and C.i 0 .t0 // are permutation matrices and so the Kontsevich–
Soibelman invariant associated with Å Å 0 equals
E.i/ D E.i 0 /:
In physics, a related method for computing this invariant is the mutation method de-
veloped and applied in [2].
7 Categorification
The setup we will describe uses triangulated 3-Calabi–Yau categories (derived cat-
egories of Ginzburg dg algebras). It is due to Kontsevich–Soibelman [89] and Na-
gao [107]. It is closely related to that of Plamondon [119], who uses triangulated 2-
Calabi–Yau categories (cluster categories). Both build on work by Derksen–Weyman–
Zelevinsky on quivers with potentials [25], who first proved a statement equivalent to
the main theorem 7.9 using decorated representations of quivers with potentials [26].
7.1 Mutation of quivers with potential. We follow Derksen–Weyman–Zelevinsky’s

b
fundamental article [25]. Let Q be a finite quiver. Let C Q be the completed path
algebra, i.e. the completion of the path algebra at the ideal generated by the arrows of
b
Q. Thus, C Q is a topological algebra and the paths of Q form a topological basis so
b
that the underlying vector space of C Q is
Y
kp
p path
and the multiplication is induced from the composition of paths (we compose paths in
the same way as we compose morphisms). The continuous zeroth Hochschild homology
b b b
of C Q is the vector space HH0 .C Q/ obtained as the quotient of C Q by the closure
of the subspace generated by all commutators. It admits a topological basis formed
by the cycles of Q, i.e. the orbits of paths p D .i j˛m j : : : j˛1 ji / of any length m 0
with identical source and target under the action of the cyclic group of order m. In
b
particular, the space HH0 .C Q/ is a product of copies of C indexed by the vertices if
Q does not have oriented cycles. For each arrow a of Q, the cyclic derivative with
respect to a is the unique continuous C-linear map
b
@a W HH0 .C Q/ ! C Q b
which takes the class of a path p to the sum
X
vu
pDuav
taken over all decompositions of p as a concatenation of paths u, a, v, where u and v

b
are of length 0. A potential on Q is an element W of HH0 .C Q/ whose expansion
in the basis of cycles does not involve cycles of length 1. A potential is reduced if
it does not involve cycles of length 2. The Jacobian algebra J.Q; W / associated
to a quiver Q with potential W is the quotient of the completed path algebra by the
closure of the 2-sided ideal generated by the cyclic derivatives of the elements of W .
If the potential W is reduced and the Jacobian algebra J.Q; W / is finite-dimensional,
its quiver is isomorphic to Q.
158 B. Keller
As typical examples, we may consider the quiver Q
2 (35)
@ ===
b ==a
==

1o c 3
with the potential W D abc or with the potential W D .abc/2 .

In order to define the mutation of a quiver with potential .Q; W / at a vertex k,
we need to recall the construction of a reduced quiver with potential from an arbitrary
quiver with potential.
Two quivers with potential .Q; W / and .Q0 ; W 0 / are right equivalent if Q0 D Q00
and there exists a C-algebra b 0 b
Qisomorphism ' W kQ ! kQ such that ' induces the
identity on the subalgebra Q0 C and the induced map in topological Hochschild
homology takes W to W 0 . A quiver with potential .Q; W / is trivial if W isQ a (possibly
infinite) linear combination of 2-cycles and J.Q; W / is isomorphic to Q0 C. If
.Q; W / and .Q0 ; W 0 / are two quivers with potential such that the sets of vertices of Q
and Q0 coincide, their direct sum .Q; W / ˚ .Q0 ; W 0 / is defined as the pair consisting
of the quiver with the same vertex set, with set of arrows the disjoint union of those of
Q and Q0 , and with the potential equal to the sum W ˚ W 0 .
Theorem 7.1 ([25], Theorem 4.6 and Proposition 4.5). Any quiver with potential
.Q; W / is right equivalent to the direct sum of a reduced one .Qred ; Wred / and a
trivial one .Qtriv ; Wtriv /, both unique up to right equivalence. Moreover, the inclusion
induces an isomorphism from J.Qred ; Wred / onto J.Q; W /.
The quiver with potential .Qred ; Wred / is the reduced part of .Q; W /.
We can now define the mutation of a quiver with potential. Let .Q; W / be a quiver
with potential such that Q does not have loops. Let k be a vertex of Q not lying on
a 2-cycle. The mutation i .Q; W / is defined as the reduced part of the quiver with
potential Q i .Q; W / D .Q0 ; W 0 /, which is defined as follows:
a) (i) To obtain Q0 from Q, add a new arrow Œ˛ˇ for each pair of arrows ˛ W k !j
and ˇ W i ! k of Q and
(ii) replace each arrow with source or target i by a new arrow with s. / D
t./ and t . / D s. /.
b) Put W 0 D ŒW C , where
(i) ŒW is obtained from W by replacing, in a representative of W without
cycles passing through k, each occurrence of ˛ˇ by Œ˛ˇ, for each pair of
arrows ˛ W i ! k and ˇ W i ! k of Q;
(ii) is the sum of the cycles Œ˛ˇˇ ˛ taken over all pairs of arrows ˛ W k ! j
and ˇ W i ! k of Q.
Then k is not contained in a 2-cycle of k .Q; W / and k .k .Q; W // is right equivalent
to .Q; W / (Theorem 5.7 of [25]). As examples, consider the mutation at 2 of the cyclic
quiver (35) endowed with the potential W D abc and with W 0 D .abc/2 . For
W D abc, the mutated quiver with potential is the acyclic quiver
2 ^=
b ===a
== (36)
=
1 3
with the zero potential. But for W D .abc/2 , the mutated quiver with potential is
2 ^=
b ===a
== (37)
=
o
e /
1 3
c
with the potential ecec C eb a .

The general construction implies that if neither Q nor the quiver Q0 in .Q0 ; W 0 / D
k .Q; W / have loops or 2-cycles, then Q and Q0 are linked by the quiver mutation
rule (cf. Proposition 7.1 of [25]) . Thus, if we want to ‘extend’ this rule to quivers
with potentials, it is important to ensure that no 2-cycles appear during the mutation
b
process.
Let Q be a finite quiver. A continuous quotient of HH0 .C Q/ is linear surjection
b
q W HH0 .C Q/ ! V such that for some N
0, all potentials involving only cycles
b
of length > N lie in the kernel of q. A polynomial function HH0 .C Q/ ! C is
b
the composition of a continuous quotient HH0 .C Q/ ! V with a polynomial map
b
V ! C . A hypersurface in HH0 .C Q/ is the set of zeroes of a non zero polynomial
function.
Theorem 7.2 ([25], Corollary 7.4). Let Q be a finite quiver without loops nor 2-cycles.
b
There is a countable union of hypersurfaces C HH0 .C Q/ such that for each W not
belonging to C , no 2-cycles appear in any iterated mutation of .Q; W /.
A potential W not belonging to C is called generic. So if Q is a quiver without

loops nor 2-cycles and W a generic potential, we can indefinitely mutate the quiver
with potential .Q; W / and the mutation of the underlying quivers is given by the quiver
mutation rule. Notice that the potential W D .abc/2 on the quiver (35) is not generic,
which is compatible with the appearance of a 2-cycle in (37).
7.2 Ginzburg algebras. Let Q be a finite quiver and W a potential on Q (cf. Sec-
tion 7.1). Let be the Ginzburg [65] dg algebra of .Q; W /. It is constructed as
x be the graded quiver with the same vertices as Q and whose arrows are
follows: Let Q
160 B. Keller
• the arrows of Q (they all have degree 0),

• an arrow a W j ! i of degree 1 for each arrow a W i ! j of Q,
• a loop ti W i ! i of degree 2 for each vertex i of Q.
The underlying graded algebra of .Q; W / is the completion of the graded path algebra
CQ x in the category of graded vector spaces with respect to the ideal generated by the
arrows x Thus, the n-th component of .Q; W / consists of elements of the form
of Q.
P

p p p, where p runs over all paths of degree n. The differential of .Q; W / is the
unique continuous linear endomorphism homogeneous of degree 1 which satisfies the
Leibniz rule
d.uv/ D .du/v C .1/p udv;
for all homogeneous u of degree p and all v, and takes the following values on the
x
arrows of Q:
• da D 0 for each arrow a of Q,
• d.a / D @a W for each arrow a of Q,
P
• d.ti / D ei . a Œa; a /ei for each vertex i of Q, where ei is the lazy path at i
and the sum runs over the set of arrows of Q.
One checks that d 2 D 0. For example, for the cyclic quiver 35 with the potential
x is
W D abc, the graded quiver Q
t2

2 =^====
@
b ====a

====

b a =
:1o
c
t1 /3d t3
c
and the differential is given by
d.a / D bc; d.b / D ca; d.c / D ab; d.t1 / D cc b b; : : : :
The Ginzburg algebra should be viewed as a refined version of the Jacobian algebra
J.Q; W /. It is concentrated in (cohomological) degrees 0 and H 0 ./ is isomorphic
to J.Q; W /. Two right equivalent quivers with potential have isomorphic Ginzburg
algebras (Lemma 2.9 of [86]). If .Q; W / D .Qtriv ; Wtriv / ˚ .Qred ; Wred / is the direct
sum of a trivial and a reduced quiver with potential, then the projection from Q onto
Qred induces a quasi-isomorphism .Q; W / ! .Qred ; Wred / (Lemma 2.10 of [86]).
7.3 Derived categories of dg algebras. Let us recall the construction of the derived
category D.A/ of a dg (D differential graded) algebra A: A (right) dg module M over
A is a graded A-module equipped with a differential d such that
d.ma/ D d.m/a C .1/jmj md.a/;
where m in M is homogeneous of degree jmj, and a 2 A.

Given two dg A-modules M and N , we define the morphism complex to be the
graded C-vector spaceQHomA .M; N / whose i -th component HomAi .M; N / is the sub-
space of the product j 2Z HomC .M j ; N j Ci / consisting of the morphisms f such
that
f .ma/ D f .m/a;
for all m in M and all a in A, together with the differential d given by
d.f / D f B dM .1/jf j dN B f
for a homogeneous morphism f of degree jf j.

The category C .A/ of dg A-modules is the category whose objects are the dg A-
modules, and whose morphisms are the 0-cycles of the morphism complexes. This is
an abelian category and a Frobenius category for the conflations which are split exact
as sequences of graded A-modules. Its stable category H .A/ is called the homotopy
category of dg A-modules, which is equivalently defined as the category whose objects
are the dg A-modules and whose morphism spaces are the 0-th homology groups of the
morphism complexes. The homotopy category H .A/ is a triangulated category whose
suspension functor † is the shift of dg modules M 7! †M D M Œ1. The derived
category D.A/ of dg A-modules is the localization of H .A/ at the full subcategory of
acyclic dg A-modules. A short exact sequence
0 /M /N /L /0
in C.A/ yields a triangle
M /N /L / †M
in D.A/. A dg A-module P is cofibrant if

s
HomC.A/ .P; L/ ! HomC.A/ .P; M /
is surjective for each quasi-isomorphism s W L ! M which is surjective in each com-

ponent. We use the term “cofibrant” since these are actually the objects which are
cofibrant for a certain structure of Quillen model category on the category C .A/, cf.
Theorem 3.2 in [84]. For an explicit description of the cofibrant dg A-modules, cf.
Proposition 2.13 of [86].
162 B. Keller
The derived category D.A/ admits arbitrary (set-indexed) coproducts. An object

P of D.A/ is compact or if the functor HomD.A/ .P; ‹/ W D.A/ ! D.C/ commutes
with arbitrary coproducts. For example, the functor
HomD.A/ .A; ‹/ D H 0 .‹/
commutes with arbitrary sums and so the free A-module of rank 1 is compact. An
arbitrary object of D.A/ is compact iff it is perfect, i.e. if it belongs to the closure of A
under left and right shifts, extensions and passage to direct factors. The perfect derived
category per.A/ D.A/ is the full subcategory on the perfect objects.
7.4 The derived category of the Ginzburg algebra. As in Section 7.2, let Q be
a finite quiver. Assume that the vertex set of Q is f1; : : : ; ng. Let W be a reduced
potential on Q. Let be the Ginzburg dg algebra of the opposite quiver with potential
.Qop ; W op /. Let D./ be the derived category of and per./ the perfect derived
category. By Lemma 2.17 of [86], the category per./ is a Krull–Schmidt category,
i.e. each object decomposes into a finite direct sum of indecomposable objects and
each indecomposable object has a local endomorphism algebra. In particular, the free
module of rank one 2 per./ decomposes into the indecomposable summands
Pi D ei associated with the vertices i of Q. The Grothendieck group K0 .per.// is
free on the basis formed by the classes ŒPi , 1 i n.
Now let D fd ./ the finite-dimensional derived category of , i.e. the full subcate-
gory of D./ formed by the dg modules whose homology is of finite total dimension.
An object M belongs to D fd ./ if and only if for each object P of per./, the space
HomD./ .P; †i M / vanishes for almost all i 2 Z and is finite-dimensional for all
i 2 Z. The category D fd ./ is in fact contained in per./. It is triangulated and
has finite-dimensional morphism spaces. More precisely, for L and M in D fd ./, the
spaces HomD./ .L; †i M / are finite-dimensional for all i 2 Z and vanish for all but
finitely many i 2 Z. Thus, the Grothendieck group K0 .D fd .// carries a well-defined
Euler form: X
hL; M i D .1/p dim HomD./ .L; †p M /:
p2Z
x op -representation Si concentrated at the vertex i yields

For a vertex i of Q, the simple Q
a simple dg -module still denoted by Si . The Si generate the triangulated category
D fd ./. The Grothendieck group K0 .D fd .// is free on the basis given by the classes
ŒSi , 1 i n. We have a well-defined non degenerate pairing
h ; i W K0 .per.// K0 .D fd .// ! Z
given again as a Euler form

X
hP; M i D .1/p dim HomD./ .P; †p M /:
p2Z
We have
hPi ; Sj i D ıij ; 1 i; j n;
so that the basis of the ŒSj , 1 j n, is dual to that of the ŒPi , 1 i n.
Let j be a vertex of Q. It follows from the cofibrant resolution of Sj given at
the beginning of the proof of Lemma 3.12 in [86] that the image of the class of Sj in
K0 .per.// equals
X X X
ŒPj ŒPs.˛/ C ŒP t.˛/ ŒPj D bij ŒPi ;
˛Wt.˛/Dj ˇ Ws.ˇ /Dj i
where the source and target maps refer to Qop and B D .bij / is the antisymmetric
matrix associated with the quiver Q. Thus, the matrix of the linear map
K0 .D fd .// ! K0 .per.//
in the bases of the ŒSj and the ŒPi is B. It follows that we have
hSi ; Sj i D bij
so that B is the matrix of the Euler form h ; i on K0 .D fd .// in the basis of the ŒSi ,
1 i n.
The category D fd ./ is 3-Calabi–Yau, by which we mean that we have bifunctorial
isomorphisms

DHom.X; Y / ! Hom.Y; †3 X /;
where D is the duality functor HomC .‹; C/ and † the shift functor. The simple modules
Si are 3-spherical objects in D fd ./, i.e. we have an isomorphism
Ext .Si ; Si / Š H .S 3 ; C/;
where the left-hand side denotes the graded vector space whose p-th component is
HomD./ .Si ; †p Si / and the right-hand side is the (singular) cohomology of the 3-
sphere with complex coefficients. The spherical objects Si yield the Seidel–Thomas
[134] twist functors twSi . These are autoequivalences of D./ such that each object
X fits into a triangle
RHom.Si ; X / ˝k Si ! X ! twSi .X / ! †RHom.Si ; X / ˝k Si :
By [134], the twist functors give rise to a (weak) action on D./ of the braid group
associated with Q, i.e. the group with generators i , i 2 Q0 , and relations i j D j i
if i and j are not linked by an arrow and
i j i D j i j
if there is exactly one arrow between i and j (no relation if there are two or more
arrows).
164 B. Keller
The category D./ admits a natural t-structure whose truncation functors are those
of the natural t -structure on the category of complexes of vector spaces (because is
concentrated in degrees 0). Thus, we have an induced natural t -structure on D fd ./.
Its heart A is canonically equivalent to the category nil.J.Q; W // of finite-dimensional
right modules over J.Q; W / where all sufficiently long paths act by 0. In particular,
the inclusion of A into D fd ./ induces an isomorphism in the Grothendieck groups

K0 .A/
! K0 .D fd .//:
The cluster category is the triangle quotient
C D per./=D fd ./: (38)
For acyclic quivers Q, Amiot [4] has shown that it is equivalent to the cluster category
CQop (we pass to the opposite because is associated with Qop ) in the sense of [6]. For
arbitrary quivers, there is also a close link between per./ and C : For a triangulated
category T and an object X of T , let us denote by prT .X / the subcategory of X -
presentable objects of T , i.e. the objects Y which occur in a triangle
X 00 ! X 0 ! Y ! †X 00 ; (39)
where X 00 and X 0 belong to the closure add.X / of X under taking (finite) direct sums
and direct summands.
Proposition 7.3 (Proposition 2.10 of [119]). The projection per./ ! C induces a

C-linear equivalence

prper./ ./
! prC ./:
Plamondon also relates the extension groups computed in the two categories (Propo-
sition 2.19 of [119]).
7.5 Derived equivalences from mutations. As in the preceding section, let Q be

a finite quiver without loops nor 2-cycles with vertex set f1; : : : ; ng and let W be
a generic potential on Q. Let denote the Ginzburg algebra associated with the
opposite quiver with potential .Qop ; W op /. Let k be a vertex of Q and 0 the Ginzburg
algebra associated with the opposite of the mutated quiver with potential k .Q; W /.
Put Pi D ei and Pi0 D ei 0 , 1 i n.
Theorem 7.4 ([86]). There are two canonical equivalences
ˆ˙ W D. 0 / ! D./
related by an isomorphism

twSk B ˆ
! ˆC :
Both ˆC and ˆ send Pi0 to Pi for i ¤ k and the images of Pk0 under the two functors
fit into triangles
L
/ / ˆ .P 0 / / †Pk
Pk k!i Pi k (40)
and L
†1 Pk / ˆC .P 0 / / j !k Pj / Pk , (41)
k
where the sums are taken over the arrows in Qop .

The intrinsic characterizations of the subcategories per./ and D fd ./ show that
the equivalences ˆ˙ induce equivalences
per. 0 / ! per./ and D fd . 0 / ! D fd ./;
and thus isomorphisms in the associated Grothendieck groups. By the triangles (40)
and (41), we get the first statement of the following corollary; the second one follows by
passage to the duals. We use the matrices E" and F" associated with Q in Section 6.2
(remember however that is the Ginzburg algebra of the opposite of .Q; W /).
Corollary 7.5. Let " be equal to 1 or 1. Under the assumptions of the theorem, the
matrix of the induced isomorphism
K0 .ˆ" / W K0 .per. 0 // ! K0 .per.//
in the bases ŒPj0 and ŒPi is E" and the matrix of the induced isomorphism
K0 .ˆ" / W K0 .D fd . 0 // ! K0 .per.//
in the bases ŒSj0 and ŒSi is F" .

Let A0 be the heart of the canonical t -structure on D fd . 0 /. The equivalences ˆ˙
send A0 onto the hearts ˙ k
.A/ of two new t-structures. These can be described in terms
of A and the subcategory addSk as follows (cf. Figure 4): Let Sk? be the right orthogonal
subcategory of Sk in A, whose objects are the M with Hom.Sk ; M / D 0. Then C k
.A/
is formed by the objects X of D fd ./ such that the object H 0 .X / belongs to Sk? , the
object H 1 .X/ belongs to addSk and H p .X / vanishes for all p ¤ 0; 1. Similarly, the
subcategory k
.A/ is formed by the objects X such that the object H 0 .X / belongs
to the left orthogonal subcategory ?Sk , the object H 1 .X / belongs to add.Sk / and
H p .X/ vanishes for all p ¤ 1; 0. The subcategory C k
.A/ is the right mutation of
A and k
.A/ is its left mutation.
By construction, we have
C
twSk .
k .A// D k .A/:
Since the categories A and ˙ .A/ are hearts of bounded, non degenerate t -structures
on D fd ./, their Grothendieck groups identify canonically with that of D fd ./. They
166 B. Keller
A A
Sk Sk? †1 Sk †Sk ?

Sk Sk
C
k .A/
k .A/
Figure 4. Right and left mutation of a heart.
are endowed with canonical bases given by the simples. Those of A identify with the
simples Si , i 2 Q0 , of nil.J.Q; W //. The simples of C k
.A/ are †1 Sk , the simples
Si of A such that Ext1 .Sk ; Si / vanishes and the objects twSk .Si / where Ext1 .Sk ; Si / is
of dimension 1. By applying tw1
Sk to these objects we obtain the simples of k .A/.
7.6 Torsion subcategories and intermediate t-structures. In order to investigate

the effect on hearts of suitable compositions of the equivalences ˆ˙ of Theorem 7.4,
let us recall the construction of ‘tilted hearts’, a variation on a construction of [69]. Let
D be a triangulated category (for example the category D fd ./). Let .D0 ; D0 / be
a bounded non degenerate t -structure on D and A its heart. A torsion pair on A is a
pair .T ; F / of full subcategories such that
a) we have Hom.T; F / D 0 for all T 2 T and F 2 F and
b) for each object M of A, there is a short exact sequence
0 / MT /M / MF /0
with MT in T and M F in F .
In this case, the subcategories T and F determine each other: we have F D T ? and
T D ? T , where the orthogonal subcategories are taken in A.
For two full subcategories U and V of D, let us write U?V for the full subcategory
whose objects X occur in triangles
U ! X ! V ! †X
with U in U and V in V . Let .T ; F / be a torsion par in A. Then the subcategory

D0 ? .†1 F / is the left aisle of a new t-structure, whose heart A.F ; †1 T / equals
F ?†1 T . It is called the right tilt of A at .T ; F /. Dually, the subcategory .†F /?D0
is the right aisle of a new t -structure on D, whose heart A.†F ; T / equals .†F / ? T .
It is called the left tilt of A at .T ; F /. The right tilt A.F ; †1 T / admits the torsion
pair .F ; †1 T / and its left tilt with respect to this pair equals the original category
A D T ? F . Similarly, the left tilt A.†F ; T / admits the torsion pair .†F ; T / and
we recover A as its right tilt with respect to this pair.
Clearly, the left aisle D0 ? .†1 F / is an intermediate left aisle, i.e. we have
D0 D0 ? .†1 F / D1 :
It is not hard to check that each intermediate left aisle is of this form. Dually, each
right aisle between D1 and D0 is of the form .†F / ? D0 .
Of course, in the situation of Section 7.5, the heart C k
.A/ is the right tilt of A with
?
respect to .addSk ; Sk / and k .A/ is its left tilt. The following lemma will allow us
to iterate such mutations. The point is that given a suitable torsion pair in a right tilt
A0 of A, either the left tilt or the right tilt of A0 with respect to this pair is again a right
tilt of A, cf. Figure 5.
T F †1 T
A
A0
A00
0
T F0 †1 T 0
T F †1 T
A
A0
00
A
†1 F 0 T 0
F0
Figure 5. Composition of a right tilt with a right tilt (top) resp. a left tilt (bottom).
Lemma 7.6 (Bridgeland [107]). Let .T ; F / be a torsion pair in A and .T 0 ; F 0 / a

torsion pair in A0 D A.F ; †1 T /.
a) If T 0 F , then the right tilt A00 D A0 .F 0 ; †1 T 0 / equals the right tilt
A.T ? T 0 ; F \ F 0 /:
168 B. Keller
b) If F 0 †1 T , then the left tilt A00 D A0 .†T 0 ; F 0 / equals the right tilt
A.†F 0 ? T ; T 0 \ †1 T /:
The lemma is not hard to check. The following easy lemma is a key point for the
Main Conjecture 5.1:
Lemma 7.7. Let A0 D A.F ; †1 T / be the right tilt of A with respect to a torsion
pair .T ; F /. Then each simple object of A0 either lies in A or in †1 A.
Indeed, let S be the given simple object. Since .F ; †1 T / forms a torsion pair in
0
A , we have the exact sequence
0 / SF /S / 1 T /0
S†
1
where SF belongs to F A and S † T to †1 F †1 A. Since S is simple, we
1
either have SF
! S or S
!S † T .
Now let .Q; W / and .Q0 ; W 0 / be two quivers with reduced potentials and let and
0
be the associated Ginzburg dg algebras. Suppose that
ˆ W D. 0 / ! D./
is a triangle equivalence. Let .D0 ; D0 / be the natural t -structure on D fd ./ and let
0 0
A be its heart. Similarly, let .D0 ; D0 / be the natural t -structure on D fd . 0 / and let
0 p
A be its heart. Let us denote by HA , p 2 Z, the homology functors with respect to
the natural t -structure on D fd ./.
Proposition 7.8. The following are equivalent:
(i) the subcategory ˆ.A0 / D fd ./ is the right tilt of A with respect to a torsion
pair .T ; F /;
(ii) the object ˆ. 0 / is .†1 /-presentable (cf. the end of Section 7.4);
0
(iii) we have D0 ˆ.D0 / D1 .
If these conditions hold, then the torsion subcategory T of (i) is formed by the finite-
dimensional quotients of the object HA 1
.ˆ. 0 //.
7.7 Patterns of tilts and decategorification. As in Section 7.5, let Q be a finite

quiver without loops nor 2-cycles with vertex set f1; : : : ; ng and let W be a generic
potential on Q. Let denote the Ginzburg algebra associated with the opposite quiver
with potential .Qop ; W op /.
As we have seen in Section 7.1, we can indefinitely mutate .Q; W /. Thus, with each
vertex t of the n-regular tree, we can associate a quiver with potential .Q.t /; W .t //
such that .Q; W / is associated with t0 and, whenever t and t 0 are linked by an edge
labeled k, the corresponding quivers with potential are linked by a mutation. We write
.t/ for the Ginzburg dg algebra associated with the opposite of .Q.t /; W .t //.
Now we will use induction on the distance of a vertex t of n-regular tree from the
root t0 to define a triangle equivalence
ˆ.t / W D..t // / D./
such that ˆ.t/ satisfies the equivalent conditions of Proposition 7.8. The construction
follows an idea of Bridgeland [107]. By definition, ˆ.t0 / is the identity. Now suppose
that ˆ.t/ has been defined and that t 0 is linked to t by an edge labeled k and is at greater
distance from t0 than t . Let A.t / D fd ./ be the image under ˆ.t / of the heart of
the natural t-structure of D fd ..t // and let Si .t /, 1 i n, be the simple objects of
A.t/. By assumption, the subcategory A.t / is the right tilt of A D A.t0 / with respect
to some torsion theory .T .t /; F .t //. Thus, by Lemma 7.7, the simple object Sk .t /
either lies in F .t / A or in †1 T .t / †1 A. In the first case, we put
ˆ.t 0 / D ˆ.t / B ˆk;C

and in the second case
ˆ.t 0 / D ˆ.t / B ˆk; :
Then in the first case, as we have seen in Section 7.5, A.t 0 / is the right tilt of A.t / with
respect the torsion pair
.add.Sk .t //; Sk .t /? /
and in the second case, it is the left tilt with respect to
.? Sk .t /; add.Sk .t ///:
In both cases, Lemma 7.6 shows that A.t 0 / is again a right tilt of A and so ˆ.t 0 / again
satisfies the conditions of Proposition 7.8.
Notice that at the same time, this construction produces a sign ".e/ for each edge
e W t ! t 0 of the n-regular tree. For each vertex t of Tn , and for 1 i n, let Ti .t /
be the image of ei .t / under ˆ.t /.
Theorem 7.9. Let t be a vertex of the n-regular tree and let 1 j n.
a) The j -th column of the c-matrix C.t / contains the coordinates of ŒSj .t / in the
basis ŒS1 , …, ŒSn of K0 .D fd .//.
b) The object Sj .t / lies in A or in †1 A. Therefore, each c-vector is non zero and
has either all its components non negative or all its components non positive (i.e.
the Main Conjecture 5.1 holds for Q).
c) The j -th column of the g-matrix G.t / contains the coordinates of ŒTj .t / in the
basis ŒP1 , …, ŒPn of K0 .per.//.
170 B. Keller
d) The (left) J.Q; W /-module HA1

.Tj .t // is finite-dimensional and the F -polyno-
mial Fj .t / equals
X
.Gre .HA 1
.Tj .t //// y e ;
e
where e runs through N , Gre denotes the variety of submodules whose quotient
n
has dimension vector e, is the Euler characteristic (with respect to singular

cohomology with rational coefficients) and
Y
n
e
y D
e
xnCi
i
:
iD1
To make sure that our conventions are coherent, let us consider the example of the
quiver Q W 1 ! 2 and the vertex t linked to t0 by the mutation at 1. We have to consider
the Ginzburg algebra associated with Qop W 2 ! 1 and perform a right mutation at
the vertex 1. We get the exchange triangle
†1 P1 ! T1 .t / ! P2 ! P1 : (42)
Thus, the class of T1 .t / in K0 .per.// equals ŒP1 C ŒP2 , which does correspond
to the g-vector
1
g1 .t / D :
1
The new simple modules are S1 .t / D †1 S1 and S2 .t / given by the universal extension
S2 ! S2 .t / ! S1 ! †S2 :
So in K0 .D fd .//, we have ŒS1 .t / D ŒS1 and ŒS2 .t / D ŒS1 C ŒS2 , which does
correspond to the c-matrix
1 1
C.t / D :
0 1
Using the exchange triangle (42), we easily check that Hom.†1 P1 ; T1 .t // D C and
Hom.†1 P2 ; T1 .t // D 0 so that the module HA 1
.T1 .t // is the simple at the vertex 1.
The associated generating series of Euler characteristics is indeed equal to
F1 .t / D 1 C y e1 :
Notice that since each Sj .t / belongs to A or †1 A (Lemma 7.7), part a) implies
part b). Thanks to parts a) and c), the duality between the bases formed by the ŒTi .t /
and the ŒSj .t/ corresponds to the first part of the tropical duality theorem 5.5.
Parts a), b) and c) are proved in Nagao’s [107], and part d) is proved there under
an additional technical assumption. Parts a), b), c) and d) follow from the results
of Plamondon [118], cf. Section 7.10 (and when H 0 ./ is finite-dimensional from
[117]). Using his dictionary between objects of the cluster category and decorated
representations, the theorem also can also be deduced from the results of [26].
For acyclic quivers Q, part d) was extended to the quantum case by Qin [120] and
(for prime powers q) by Rupel [130], [129], who also obtained an analogous result for
acyclic valued quivers. Under certain technical assumptions, Efimov [30] has recently
been able to extend part d) to the quantum case for arbitrary quivers (without loops
nor 2-cycles). He mainly builds on the work of Kontsevich–Soibelman [90], [89] and
Nagao [107].
7.8 Reign of the tropics. The following theorem and its corollary are the basis of the
‘tropicalization method’ which is used in applications of cluster algebras to discrete
dynamical systems and to dilogarithm identities, cf. [113], [73], [74], [114], [115].
Theorem 7.10 ([119]). Let 0 and 00 be two Ginzburg dg algebras and let
ˆ0 W D. 0 / ! D./ and ˆ00 W D. 00 / ! D./ be triangle equivalences satisfy-
ing the conditions of Proposition 7.8. For 1 j n, let us write Sj0 for the image of
the j -th simple module under ˆ0 and Pj0 for the image of the module ej 0 . Similarly,
we define Sj00 and Pj00 . Suppose that for each 1 j n, we have ŒSj0 D ŒSj00 in
K0 .D fd .//. Then for each 1 j n, we have Pj0 Š Pj00 and Sj0 Š Sj00 :
We first notice that for each 1 j n, we have the equality ŒPj0 D ŒPj00 in
K0 .per.//. Indeed, this follows from the duality between the bases ŒPi0 and ŒSj0 in
K0 .per./ and K0 .D fd .//. Now the first isomorphism follows from the fact, proved
in Section 3.1 of [118], that an object X of pr./ which is rigid, i.e. Hom.X; †X / D 0,
is determined by its class in K0 .per.//. The objects Si0 are the simples of the abelian
subcategory of D./ formed by the objects U such that Hom.Pj0 ; †p U / vanishes for
all p ¤ 0 and all 1 j n. Among these simples, Si0 is the only one which receives
a non zero morphism from Pi0 . Now it is clear that the isomorphisms for the Pj0 imply
those for the Sj0 .
Corollary 7.11. Let t and t 0 be vertices of the n-regular tree such that there is a
permutation
of f1; : : : ; ng with C.t 0 / D P C.t /, where P is the permutation
matrix associated with
. Then we have G.t / D P G.t 0 /, the permutation
yields
an isomorphism Q.t /
! Q.t 0 / and for each 1 j n, we have
a) Tj .t 0 / D T.j / .t / and Sj .t 0 / D S.j / .t /;
b) Fj .t 0 / D F.j / .t /;
c) xj .t 0 / D x.j / .t / and yj .t 0 / D y.j / .t /.
In particular, the seeds associated with t and t 0 are isomorphic via
.
To prove the corollary, we apply the theorem to the equivalences ˆ.t / and ˆ.t 0 /.
We immediately obtain part a). This implies the statement on the g-matrices and the
quivers. By Theorem 7.9, it also implies parts b) and c).
172 B. Keller
A proof of the corollary based on the study of stability conditions on D fd ./ is

given in Section 4.2 of [107]. It can also be deduced from the results of [26].
7.9 Rigid objects and cluster monomials. Let Q z be an ice quiver (equally valued).
Let be the Ginzburg dg algebra associated with the opposite of .Q;z W /, where W is
a generic potential.
For each object L of pr.†1 / such that H 1 .L/ is finite-dimensional, we define a
Laurent polynomial X
X.L/ D .Gre .HA 1
.L/// yO e ;
e
Qm bQi l Q e
where yOl D iD1 xi , 1 l n, and yO e D l yOl l . By part c) of Theorem 7.9 and
by the separation formula of Theorem 5.7, when L D Ti .t / for some 1 i n and
some vertex t of the n-regular tree, then X.L/ equals the cluster variable xi .t /. It is
not hard to check that for two objects L and L0 of pr./, we have
X.L ˚ L0 / D X.L/X.L0 /:
So if we apply the map L 7! X.L/ to direct sums of objects Ti .t /, 1 i n, for a
fixed vertex t , we recover the cluster monomials associated with t .
Let us call a rigid object L of pr.†1 / reachable if it there is a vertex t of the
n-regular tree such that L is a direct sum of copies of the objects Ti .t /, 1 i n.
Theorem 7.12 ([118], [19]). a) If L1 , …, LN are pairwise non-isomorphic reachable
rigid objects, then the Laurent polynomials X.L1 /, …, X.LN / are linearly indepen-
dent.
b) The map L 7! X.L/ induces a bijection from the set of isomorphism classes of
reachable rigid objects onto the set of cluster monomials.
The surjectivity in b) is proved by Plamondon [118]. The linear independence in
a), and hence the injectivity in b), is proved in [19].
7.10 Proof of decategorification. We will sketch a proof of Theorem 7.9. We prove

a) and b) simultaneously by induction on the distance of t from t0 . For t D t0 , there is
nothing to prove. Now assume we have proved the claim for t and that t 0 is at greater
distance from t0 and linked to t by an edge labeled k. Then the coefficients of the
c-matrix at t 0 can be computed as
´
0 cij .t / if j D k;
cij .t / D
cij .t / C cik .t /Œ"bkj .t /C C Œ"cik .t /C bkj .t / else;
where 1 i; j n and " is any sign, cf. Proposition 5.8 of [45] and formula (3.3) in
[115]. We know that bkj .t / equals the number of arrows from k to j in Q.t / minus
the number of arrows from j to k in Q.t /. Thus, we have
bkj .t / D dim Ext1 .Sk .t /; Sj .t // dim Ext1 .Sj .t /; Sk .t //:
By the induction hypothesis, the coordinates of ŒSk .t / in the basis of the ŒSi are the
components cik .t /, 1 i n, of the c-vector C.t /ek . By Lemma 7.7, they are all
of the same sign. Let us choose " equal to this sign. Then the formula for the cij .t 0 /
simplifies as follows:
´
0 cij .t / if j D k;
cij .t / D
cij .t / C cik .t /Œ"bkj .t /C else.
Now assume that " D 1. This means that Sk .t / lies in A and that Sk .t 0 / is †1 Sk .t /.
Let us put m D bkj .t /. If we have m 0, then the space Ext1 .Sk .t /; Sj .t // vanishes
and we have Sj .t 0 / D Sj .t /. If we have m 0, then we get m D Ext1 .Sk .t /; Sj .t //
and the object Sj .t 0 / is constructed as a universal extension:
.†1 Sk .t //m ! Sj .t / ! Sj .t 0 / ! Sk .t /m :
In both cases, the formula for cij .t 0 / gives the correct multiplicity of ŒSi in ŒSj .t 0 /.
Now suppose that " D 1. Then Sk .t / belongs to †1 A.t / and Sk .t 0 / is †Sk .t /. We
put m D bkj .t / D bj k .t /. If we have m 0, then the space Ext1 .Sj .t /; Sk .t // van-
ishes and Sj .t 0 / D Sj .t /. If we have m 0, then we get m D dim Ext1 .Sj .t /; Sk .t //
and Sj .t 0 / is constructed as a universal extension
Sk .t /m ! Sj .t 0 / ! Sj .t / ! †Sk .t /m :
Again, in both cases, the formula for cij .t 0 / gives the correct multiplicity of ŒSi in
ŒSj .t 0 /.
We get part c) as a consequence: Indeed, by part b) the Main Conjecture 5.1 holds
for Q and so we have G.t /T C.t / D I for all vertices t of the n-regular tree, by the
tropical duality theorem 5.3. On the other hand, the basis of the ŒPi .t / is dual to that
of the ŒSj .t/. Clearly, this implies c).
We deduce d) from Plamondon’s results [119]. Indeed, both Tj .t / and †1 belong
to pr.†1 /. Thus, by Proposition 7.3, we have
HA
1
.Tj .t// D Homper./ .†1 ; Tj .t //

! HomC .
.†1 /;
.Tj .t ///;
where C D per./=D fd ./ is the cluster category and
the projection functor. Let
us omit this functor from the notations. Since Tj .t / 2 C is obtained by iterated
mutation from , it belongs to Plamondon’s category D C and so the space
HomC .Tj .t /; †/
is finite-dimensional. By Proposition 2.16 of [119], this space is in duality with
HomC .†1 ; Tj .t //
which therefore also finite-dimensional. So we find that HA 1
.Tj .t // is finite-dimen-
sional and in duality with HomC .Tj .t /; †/. Now let P 7! P _ denote the canonical
equivalence

per./op
! per. op /; P 7! P _ D RHom .P; /:
174 B. Keller
It induces an equivalence C
op
! C op still denoted by the same symbol. We have

! HomC op .†/_ ; Tj .t /_ / D HomC op .†1 ; Tj .t /_ /:
HomC .Tj .t /; †/
Notice that op D .Qop ; W op /op / D .Q; W /. So we get that the left J.Q; W /-
module HA 1
.Tj .t // is in duality with the right J.Q; W /-module
HomC op ..†/_ ; Tj .t /_ / D HomC op .†1 ; Tj0 .t //;
where Tj0 .t/ denotes the object obtained from in C op by the sequence of mutations
linking t0 to t. Thus, the Grassmannian of e-dimensional quotients of HA 1
.Tj .t //
identifies with the Grassmannian of e-dimensional submodules of the above J.Q; W /-
module. The generating series of their Euler characteristics is the F -polynomial asso-
ciated with Tj0 .t / in Def. 3.14 of [118] and it equals the F -polynomial Fj .t /, as shown
in Section 4.2 of [118].
7.11 Proof of the quantum dilogarithm identities. We will sketch a proof of Theo-
rem 6.5. We start with part a). We prove the stronger statement given in Remark 6.6. So
suppose that, in the notations of the remark, we have P C.t / D C.t 0 / for the permuta-
tion matrix P D P associated with a permutation
of f1; : : : ; ng. By Corollary 7.11,
we find that the seeds associated with t and t 0 are isomorphic via
in any cluster
algebra associated with a matrix Bz whose principal part B corresponds to Q. Now by
Theorem 6.2, we find that the quantum seeds associated with t and t 0 are isomorphic
via
in any quantum cluster algebra associated with a compatible pair .B; z ƒ/, where
the principal part of Bz is the given matrix B. Thus, in the notations of Theorem 6.4,
we have ˆ.i/ D ˆ.i 0 /. Now by the duality theorem 5.3, we also have P G.t / D G.t 0 /
for the same permutation
. By the equality (34) in Theorem 6.4, we obtain
Ad0 .E.i// D Ad0 .E.i 0 //:
Now let us choose Bz D Bpr . We find that the power series E.i/E.i 0 /1 in the variables
x1 , … , xn commutes with the variables xnCi , 1 i n. Now by our choice of
Bz D Bpr , we have
xnCi xj D q ıij xj xnCi
for all 1 i; j n. This implies that for any power series f .x1 ; : : : ; xn /, we have
1
xnCi f .x1 ; : : : ; xn /xnCi D f .x1 ; : : : ; qxi ; : : : ; xn /:
So a power series in x1 , …, xn which commutes with all the xnCi , 1 i n, must

be constant. Since the constant term of E.i/E.i 0 /1 is 1, we find E.i/E.i 0 /1 D 1 as
claimed.
For part b), we have to invoke Donaldson–Thomas theory in its form pioneered
by Kontsevich–Soibelman [90], [89]. This theory is not yet completely developed for
formal potentials1 and therefore, for the moment, does not apply to arbitrary quivers
(cf. [7], [101], [108] for recent progress on special classes). However, in its final form,
the theory should yield the following: Let A y Q denote the completed quantum affine
space associated with Q. Let A be the category of finite-dimensional (hence nilpotent)
right modules over the completed Jacobian algebra J.Q; W / of the quiver Q endowed
with a generic potential. Let T1 and T2 be torsion subcategories of A. Following the
explanation after Remark 21 on page 90 of [90] we define T1 to be constructibly less
than or equal to T2 if T1 is contained in T2 and for each dimension vector d , the subset
of the variety of (contravariant) representations of J.Q; W / with dimension vector d
formed by the points corresponding to modules in T1? \ T2 is constructible. In this
case, following [90] we write
T1 constr T2 : (43)
What the fully fledged version of Kontsevich–Soibelman’s theory should yield is a
DT-character on A, i.e. the datum of an element AT1 ;T2 of the group A y for each pair
Q
of torsion theories T1 , T2 satisfying (43) such that the following hold:
a) whenever we have three torsion theories T1 , T2 and T3 such that
T1 constr T2 ; T2 constr T3 and T1 constr T3 ;
we have
AT1 ;T2 AT2 ;T3 D AT1 ;T3 I (44)
b) if we have T2 D T1 ?add.L/, where L is a module in T1? satisfying Hom.L; L/ D
C and Ext1 .L; L/ D 0, we have
AT1 ;T2 D E.x ˛ /; (45)
where ˛ is the dimension vector of L.
The non commutative DT invariant associated with Q and the given DT-character is
then the power series
DTQ D A0;A 2 A y Q: (46)
y
Via the duality functor HomC .‹; C/ and the canonical isomorphism A
op
!A y Qop
Q
y Q is mapped
taking x ˛ to x ˛ , a DT-character for Q yields one for Qop and DTQ 2 A
to DTQop 2 A y Qop .
Now assume that we have a quiver Q whose non commutative DT-invariant is
defined, i.e. it admits a DT-character. Then this also holds for Qop . Suppose that
we are in the situation of part b) of Theorem 6.5 so that C.tN / is a permutation
matrix. Then by Theorem 7.10, the simples of A.tN / lie in †1 A and so we must
have A.tN / D †1 A and T .tN / D A. Now the torsion subcategories
f0g D T .t0 /; T .t1 /; : : : ; T .tN / D A
form a sequence such that for each 1 s N , we either have
1
cf. Section 4.8 in [30] and the discussion in Section 7.1 of [89].
176 B. Keller
(1) Sis .ts1 / 2 A and then
T .ts1 / constr T .ts / and T .ts / D T .ts1 / ? add.Sis .ts1 //
or
(2) Sis .ts1 / 2 †1 A and then
T .ts / constr T .ts1 / and T .ts1 / D T .ts / ? add.†Sis .ts1 //;
depending on the sign of the c-vector ˇs D C.ts1 /eis , which is just the (signed)
dimension vector of Sis .ts1 /, by Theorem 7.9. By induction on s, one now proves
that
E."1 ˇ1 /"1 : : : E."s ˇs /"s D A0;T .ts / :
For s D N , we obtain the equality
E."1 ˇ1 /"1 : : : E."s ˇN /"N D A0;A D DTQop

y Qop . Its image under the canonical isomorphism A
in A y op y op
Q ! AQ is the claimed
y Q.
equality in A
References
[1] M. Alim, S. Cecotti, C. Córdova, S. Espahbodi, A. Rastogi, and C. Vafa, BPS quivers and
spectra of complete N D 2 quantum field theories. Preprint, arXiv:1109.4941 [hep-th].
[2] M. Alim, S. Cecotti, C. Córdova, S. Espahbodi, A. Rastogi, and C. Vafa, N D 2 quantum
field theories and their BPS quivers. Preprint, arXiv:1112.3984 [hep-th].
[3] C. Amiot, On generalized cluster categories. In Representations of algebras and related
topics (ICRA XIV), EMS Ser. Congr. Rep., Eur. Math. Soc., Zürich 2011, 1–53.
[4] C. Amiot, Cluster categories for algebras of global dimension 2 and quivers with potential.
Ann. Inst. Fourier 59 (2009), no. 6, 2525–2590.
[5] A. Bakke Buan and R. Marsh, Cluster-tilting theory. In Trends in representation theory
of algebras and related topics, Contemp. Math. 406, Amer. Math. Soc., Providence, RI,
2006, 1–30.
[6] A. Bakke Buan, R. J. Marsh, M. Reineke, I. Reiten, and G. Todorov, Tilting theory and
cluster combinatorics. Adv. Math. 204 (2006), no. 2, 572–618.
[7] K. Behrend, J. Bryan, and B. Szendroi, Motivic degree zero Donaldson–Thomas
invariants. Invent. Math., to appear, Doi 10.1007/s00222-012-0408-1; preprint,
[8] A. Berenstein, S. Fomin, and A. Zelevinsky, Parametrizations of canonical bases and
totally positive matrices. Adv. Math. 122 (1996), no. 1, 49–149.
[9] A. Berenstein, S. Fomin, and A. Zelevinsky, Cluster algebras. III. Upper bounds and
double Bruhat cells. Duke Math. J. 126 (2005), no. 1, 1–52.
[10] A. Berenstein and A. Zelevinsky, Quantum cluster algebras. Adv. Math. 195 (2005), no. 2,
405–455.
[11] T. Bridgeland, Stability conditions on triangulated categories. Ann. of Math. (2) 166
(2007), no. 2, 317–345.
[12] P. Caldero and B. Keller, From triangulated categories to cluster algebras. II. Ann. Sci.
École Norm. Sup. (4) 39 (2006), no. 6, 983–1009.
[13] P. Caldero and B. Keller, From triangulated categories to cluster algebras. Invent. Math.
172 (2008), 169–211.
[14] S. Cecotti, C. Córdova, and C. Vafa, Braids, walls and mirrors. Preprint, arXiv:1110.2115
[hep-th].
[15] S. Cecotti, A. Neitzke, and C. Vafa, R-twisting and 4d /2d -correspondences. Preprint,
arXiv:1006.3435 [physics.hep-th].
[16] S. Cecotti and C. Vafa, Classification of complete N D 2 supersymmetric theories in 4
dimensions. Preprint, arXiv:1103.5832 [hep-th].
[17] G. Cerulli Irelli, Positivity in skew-symmetric cluster algebras of finite type. Preprint,
arXiv:1102.3050 [math.RA].
[18] G. Cerulli Irelli, Cluster algebras of type A.1/
2 . Algebr. Represent. Theory 15 (2012), no. 5,
977–1021.
[19] G. Cerulli Irelli, B. Keller, D. Labardini-Fragoso, and P.-G. Plamondon, On cluster mono-
mials. In preparation.
[20] G. Cerulli Irelli and D. Labardini-Fragoso, Quivers with potentials associated to tri-
angulated surfaces, part III: Tagged triangulations and cluster monomials. Preprint,
arXiv:1108.1774 [math.RT].
[21] F. Chapoton, Enumerative properties of generalized associahedra. Sém. Lothar. Combin.
51 (2004/05), Art. B51b, 16 pp. (electronic).
[22] F. Chapoton, S. Fomin, and A. Zelevinsky, Polytopal realizations of generalized associa-
hedra. Canad. Math. Bull. 45 (2002), no. 4, 537–566.
[23] L. Demonet, Mutations of group species with potentials and their representations. Appli-
cations to cluster algebras. Preprint, arXiv:1003.5078 [math.RT].
[24] L. Demonet, Categorification of skew-symmetrizable cluster algebras. Algebr. Represent.
Theory 14 (2011), no. 6, 1087–1162.
[25] H. Derksen, J. Weyman, and A. Zelevinsky, Quivers with potentials and their representa-
tions I: Mutations. Selecta Math. (N.S.) 14 (2008), 59–119.
[26] H. Derksen, J. Weyman, and A. Zelevinsky, Quivers with potentials and their representa-
tions II: Applications to cluster algebras. J. Amer. Math. Soc. 23 (2010), 749–790.
[27] P. Di Francesco and R. Kedem, Q-systems as cluster algebras. II. Cartan matrix of finite
type and the polynomial property. Lett. Math. Phys. 89 (2009), no. 3, 183–216.
[28] G. Dupont, Positivity in coefficient-free rank two cluster algebras. Electron. J. Combin.
16 (2009), no. 1, Research Paper 98, 11.
[29] G. Dupont, Generic variables in acyclic cluster algebras. J. Pure Appl. Algebra 215 (2011),
no. 4, 628–641.
178 B. Keller
[30] A. Efimov, Quantum cluster variables via vanishing cycles. Preprint, arXiv:1112.3601
[math.AG].
[31] L. Faddeev and A. Yu. Volkov, Abelian current algebra and the Virasoro algebra on the
lattice. Phys. Lett. B 315 (1993), no. 3–4, 311–318.
[32] L. D. Faddeev and R. M. Kashaev, Quantum dilogarithm. Modern Phys. Lett. A 9 (1994),
no. 5, 427–434.
[33] A. Felikson, M. Shapiro, and P. Tumarkin, Cluster algebras of finite mutation type via
unfoldings. Internat. Math. Res. Notices 2012 (2012), 1768–1804.
[34] A. Felikson, M. Shapiro, and P. Tumarkin, Skew-symmetric cluster algebras of finite
mutation type. J. Eur. Math. Soc. 14 (2012), no. 4, 1135–1180.
[35] V. V. Fock and A. B. Goncharov, Moduli spaces of local systems and higher Teichmüller
theory. Publ. Math. Inst. Hautes Études Sci. 103 (2006), 1–211.
[36] V. V. Fock and A. B. Goncharov, Dual Teichmüller and lamination spaces. In Handbook
of Teichmüller theory, Vol. I, IRMA Lect. Math. Theor. Phys. 11, Eur. Math. Soc., Zürich
2007, 647–684.
[37] V. V. Fock and A. B. Goncharov, . Ann. Sci. Éc. Norm. Supér. (4) 42 (2009), no. 6, 865–930.
[38] V. V. Fock and A. B. Goncharov, Cluster ensembles, quantization and the dilogarithm. II.
The intertwiner. In Algebra, arithmetic, and geometry: in honor of Yu. I. Manin, Vol. I,
Progr. Math. 269, Birkhäuser, Boston, MA, 2009, 655–673.
[39] V. V. Fock and A. B. Goncharov, The quantum dilogarithm and representations of quantum
cluster varieties. Invent. Math. 175 (2009), no. 2, 223–286.
[40] S. Fomin, Cluster algebras portal. www.math.lsa.umich.edu/~fomin/cluster.html
[41] S. Fomin, Total positivity and cluster algebras. In Proceedings of the International
Congress of Mathematicians, Vol. II, Hindustan Book Agency, New Delhi 2010, 125–145.
[42] S. Fomin and N. Reading, Generalized cluster complexes and Coxeter combinatorics.
Internat. Math. Res. Notices 2005 (2005), no. 44, 2709–2757.
[43] S. Fomin, M. Shapiro, and D. Thurston, Cluster algebras and triangulated surfaces. I.
Cluster complexes. Acta Math. 201 (2008), no. 1, 83–146.
[44] S. Fomin and D. Thurston, Cluster algebras and triangulated surfaces. Part II: Lambda
lengths. Preprint, arXiv:1210.5569 [math.GT].
[45] S. Fomin and A. Zelevinsky, Cluster algebras. I. Foundations. J. Amer. Math. Soc. 15
(2002), no. 2, 497–529.
[46] S. Fomin and A. Zelevinsky, Cluster algebras. II. Finite type classification. Invent. Math.
154 (2003), no. 1, 63–121.
[47] S. Fomin and A. Zelevinsky, Cluster algebras: notes for the CDM-03 conference. In
Current developments in mathematics 2003, International Press, Somerville, MA, 2003,
1–34.
[48] S. Fomin and A. Zelevinsky, Y -systems and generalized associahedra. Ann. of Math. (2)
158 (2003), no. 3, 977–1018.
[49] S. Fomin and A. Zelevinsky, Cluster algebras IV: coefficients. Compositio Math. 143
(2007), 112–164.
[50] D. Gaiotto, G. W. Moore, and A. Neitzke, Framed BPS States. Preprint, arXiv:1006.0146
[physics.hep-th].
[51] D. Gaiotto, G. W. Moore, and A. Neitzke, Wall-crossing, Hitchin systems, and the WKB
approximation. Preprint, arXiv:0907.3987 [physics.hep-th].
[52] D. Gaiotto, G. W. Moore, and A. Neitzke, Four-dimensional wall-crossing via three-
dimensional field theory. Comm. Math. Phys. 299 (2010), no. 1, 163–224.
[53] C. Geiß, B. Leclerc, and J. Schröer, Cluster algebra structures and semicanonical bases
for unipotent groups. Preprint, arXiv:math/0703039 [math.RT].
[54] C. Geiß, B. Leclerc, and J. Schröer, Cluster structures on quantum coordinate rings. Selecta
Math., to appear; preprint, arXiv:1104.0531 [math.RT].
[55] C. Geiß, B. Leclerc, and J. Schröer, Factorial cluster algebras. Preprint, arXiv:1110.1199
[math.RT].
[56] C. Geiß, B. Leclerc, and J. Schröer, Generic bases for cluster algebras and the chamber
ansatz. J. Amer. Math. Soc. 25 (2012), 21–76.
[57] C. Geiß, B. Leclerc, and J. Schröer, Kac-Moody groups and cluster algebras. Adv. Math.
228 (2011), no. 1, 329–433.
[58] C. Geiß, B. Leclerc, and J. Schröer, Rigid modules over preprojective algebras. Invent.
Math. 165 (2006), no. 3, 589–632.
[59] C. Geiß, B. Leclerc, and J. Schröer, Partial flag varieties and preprojective algebras. Ann.
Inst. Fourier (Grenoble) 58 (2008), no. 3, 825–876.
[60] C. Geiß, B. Leclerc, and J. Schröer, Preprojective algebras and cluster algebras. In Trends
in representation theory of algebras and related topics (ICRA XII), EMS Ser. Congr.
Rep., Eur. Math. Soc., Zürich 2008, 253–283.
[61] M. Gekhtman, M. Shapiro, and A. Vainshtein, Cluster algebras and Poisson geometry.
Mosc. Math. J. 3 (2003), no. 3, 899–934, 1199.
[62] M. Gekhtman, M. Shapiro, and A. Vainshtein, Cluster algebras and Weil-Petersson forms.
Duke Math. J. 127 (2005), no. 2, 291–311.
[63] M. Gekhtman, M. Shapiro, and A. Vainshtein, On the properties of the exchange graph of
a cluster algebra. Math. Res. Lett. 15 (2008), no. 2, 321–330.
[64] M. Gekhtman, M. Shapiro, and A. Vainshtein, Cluster algebras and Poisson geometry.
Math. Surveys Monogr. 167, Amer. Math. Soc., Providence, RI, 2010.
[65] V. Ginzburg, Calabi–Yau algebras. Preprint, arXiv:math/0612139v3 [math.AG].
[66] A. B. Goncharov, The classical trilogarithm, algebraic K-theory of fields and Dedekind
zeta functions. Bull. Amer. Math. Soc. (N.S.) 24 (1991), no. 1, 155–162.
[67] J. Grabowski and S. Launois, Quantum cluster algebra structures on quantum grassman-
nians and their quantum Schubert cells: the finite-type cases. Internat. Math. Res. Notices
2011 (2011), no. 10, 2230–2262.
[68] M. Gross, P. Hacking, and S. Keel, Mirror symmetry for log Calabi-Yau surfaces I.
[69] D. Happel, I. Reiten, and S. O. Smalø, Tilting in abelian categories and quasitilted algebras.
Mem. Amer. Math. Soc. 120 (1996), no. 575.
180 B. Keller
[70] D. Hernandez and B. Leclerc, Quantum Grothendieck rings and derived Hall algebras.
Preprint, arXiv:1109.0862 [math.QA].
[71] D. Hernandez and B. Leclerc, Cluster algebras and quantum affine algebras. Duke Math.
J. 154 (2010), no. 2, 265–341.
[72] C. Ingalls and H. Thomas, Noncrossing partitions and representations of quivers. Com-
positio Math. 145 (2009), no. 6, 1533–1562.
[73] R. Inoue, O. Iyama, B. Keller, A. Kuniba, and T. Nakanishi, Periodicities of T and Y -
systems, dilogarithm identities, and cluster algebras I: type Br . Publ. Res. Inst. Math.
Sci., to appear; preprint, arXiv:1001.1880 [math.QA].
[74] R. Inoue, O. Iyama, B. Keller, A. Kuniba, and T. Nakanishi, Periodicities of T and Y -
systems, dilogarithm identities, and cluster algebras II: types Cr , F4 , and G2 . Preprint,
arXiv:1001.1881 [math.QA].
[75] R. Inoue, O. Iyama, A. Kuniba, T. Nakanishi, and J. Suzuki, Periodicities of T -systems
and Y -systems. Nagoya Math. J. 197 (2010), 59–174.
[76] O. Iyama and I. Reiten, Fomin-Zelevinsky mutation and tilting modules over Calabi-Yau
algebras. Amer. J. Math. 130 (2008), no. 4, 1087–1149.
[77] D. Joyce and Y. Song, A theory of generalized Donaldson-Thomas invariants. Preprint,
arXiv:0810.5645v3 [math.AG].
[78] R. M. Kashaev and T. Nakanishi, Classical and quantum dilogarithm identities. SIGMA
Symmetry Integrability Geom. Methods Appl. 7 (2011), Paper 102.
[79] M. Kashiwara, Bases cristallines. C. R. Acad. Sci. Paris Sér. I Math. 311 (1990), no. 6,
277–280.
[80] R. Kedem, Q-systems as cluster algebras. J. Phys. A 41 (2008), no. 19, 194011, 14.
[81] B. Keller, Cluster theory and quantum dilogarithm identities. In Representations of al-
gebras and related topics (ICRA XIV), EMS Ser. Congr. Rep., Eur. Math. Soc., Zürich
2011, 85–116.
[82] B. Keller, The periodicity conjecture for pairs of Dynkin diagrams. Ann. of Math., to
appear; preprint, arXiv:1001.1531 [math.RT] .
[83] B. Keller, Quiver mutation in Java. Java applet available at the author’s home page,
http://www.math.jussieu.fr/~keller/quivermutation/.
[84] B. Keller, On differential graded categories. In International Congress of Mathematicians,
Vol. II, Eur. Math. Soc., Zürich 2006, 151–190.
[85] B. Keller, Cluster algebras, quiver representations and triangulated categories. In Triangu-
lated categories (Thorsten Holm, Peter Jørgensen, and Raphaël Rouquier, eds.), London
Math. Soc. Lecture Note Ser. 375, Cambridge University Press, Cambridge 2010, 76–160.
[86] B. Keller and D. Yang, Derived equivalences from mutations of quivers with potential.
Adv. Math. 26 (2011), 2118–2168.
[87] Y. Kimura and F. Qin, Quiver varieties and quantum cluster algebras. In preparation.
[88] Y. Kodama and L. Williams, KP solitons, total positivity and cluster algebras. Proc.
Natl. Acad. Sci., to appear, Doi 10.1073/pnas.1102627108; preprint, arXiv:1105.4170
[math.CO].
[89] M. Kontsevich and Y. Soibelman, Cohomological Hall algebra, exponential Hodge struc-
tures and motivic Donaldson-Thomas invariants. Preprint, arXiv:1006.2706 [math.AG].
[90] M. Kontsevich and Y. Soibelman, Stability structures, Donaldson-Thomas invariants and
cluster transformations. Preprint, arXiv:0811.2435 [math.AG].
[91] C. Krattenthaler, The F -triangle of the generalised cluster complex. In Topics in discrete
mathematics, Algorithms Combin. 26, Springer-Verlag, Berlin 2006, 93–126.
[92] A. Kuniba, T. Nakanishi, and J. Suzuki, T-systems and Y-systems in integrable systems.
J. Physics A 44 (2011), no. 10, 103001.
[93] P. Lampe, A quantum cluster algebra of Kronecker type and the dual canonical basis.
Internat. Math. Res. Notices 2011 (2011), no. 13, 2970–3005.
[94] P. Lampe, Quantum cluster algebras of type A and the dual canonical basis. Preprint,
[95] B. Leclerc, Algèbres affines quantiques et algèbres amassées. Notes from a talk at the
Algebra Seminar at the Institut Henri Poincaré, January 14, 2008.
[96] B. Leclerc, Cluster algebras and representation theory. In Proceedings of the Interna-
tional Congress of Mathematicians, Vol. IV, Hindustan Book Agency, New Delhi 2010,
2471–2488.
[97] K. Lee and R. Schiffler, A combinatorial formula for rank 2 cluster variables. J. Alge-
braic Combin., to appear, Doi 10.1007/s10801-012-0359-z; preprint, arXiv:1106.0952
[math.CO].
[98] G. Lusztig, Canonical bases arising from quantized enveloping algebras. J. Amer. Math.
Soc. 3 (1990), no. 2, 447–498.
[99] G. Lusztig, Total positivity in reductive groups. In Lie theory and geometry, Progr. Math.
123, Birkhäuser, Boston, MA, 1994, 531–568.
[100] G. Lusztig, Semicanonical bases arising from enveloping algebras. Adv. Math. 151 (2000),
no. 2, 129–139.
[101] S. Mozgovoy, On the motivic Donaldson-Thomas invariants of quivers with potentials.
[102] G. Muller, Locally acyclic cluster algebras. Preprint, arXiv:1111.4468 [math.AG].
[103] G. Musiker, A graph theoretic expansion formula for cluster algebras of classical type.
Ann. Combin. 15 (2011), 147–184.
[104] G. Musiker, R. Schiffler, and L. Williams, Bases for cluster algebras from surfaces. Com-
posito Math., to appear; preprint, arXiv:1110.4364 [math.RT].
[105] G. Musiker, R. Schiffler, and L. Williams, Positivity for cluster algebras from surfaces.
Adv. Math. 227 (2011), no. 6, 2241–2308.
[106] G. Musiker and C. Stump, A compendium on the cluster algebra and quiver package in
Sage. Sém. Lothar. Combin. 65 (2010/12), Art. B65d.
[107] K. Nagao, Donaldson–Thomas theory and cluster algebras. Preprint, arXiv:1002.4884
[math.AG].
[108] K. Nagao, Wall-crossing of the motivic Donaldson-Thomas invariants. Preprint,
182 B. Keller
[109] K. Nagao, Quantum dilogarithm identities. RIMS Kokyuroku Bessatsu B28 (2011),
164–170.
[110] K. Nagao, Y. Terashima, and M. Yamazaki, Hyperbolic 3-manifolds and cluster algebras.
Preprint, arXiv:1112.3106 [math.GT].
[111] A. Nájera Chávez, On the c-vectors and g-vectors of the Markov cluster algebra. Preprint,
arXiv:1112.3578 [math.CO].
[112] H. Nakajima, Quiver varieties and cluster algebras. Kyoto J. Math. 51 (2011), no. 1,
71–126.
[113] T. Nakanishi, Dilogarithm identities for conformal field theories and cluster algebras:
simply laced case. Nagoya Math. J. 202 (2011), 23–43.
[114] T. Nakanishi, Periodicities in cluster algebras and dilogarithm identities. In Representa-
tions of algebras and related topics (ICRA XIV), EMS Ser. Congr. Rep., Eur. Math. Soc.,
Zürich 2011, 407–443.
[115] T. Nakanishi, Tropicalization method in cluster algebras. In Tropical geometry and inte-
grable systems, Contemp. Math. 580, Amer. Math. Soc., Providence, R.I., 2012, 95–115.
[116] T. Nakanishi and A. Zelevinsky, On tropical dualities in cluster algebras. In Algebraic
groups and quantum groups, Contemp. Math. 565, Amer. Math. Soc., Providence, RI,
2012, 217–226.
[117] Y. Palu, Cluster characters for 2-Calabi-Yau triangulated categories. Ann. Inst. Fourier
(Grenoble) 58 (2008), no. 6, 2221–2248.
[118] P.-G. Plamondon, Cluster algebras via cluster categories with infinite-dimensional mor-
phism spaces. Compositio Math. 147 (2011), no. 6, 1921–1934.
[119] P.-G. Plamondon, Cluster characters for cluster categories with infinite-dimensional mor-
phism spaces. Adv. Math. 227 (2011), no. 1, 1–39.
[120] F. Qin, Quantum cluster variables via Serre polynomials. J. Reine Angew. Math. 668
(2012), 149–190.
[121] A. King and Y. Qiu, Exchange graphs of acyclic Calabi–Yau categories. Preprint,
[122] Y. Qiu, Stability conditions and quantum dilogarithm identities for Dynkin quivers.
[123] N. Reading and D. Speyer, Combinatorial frameworks for cluster algebras. Preprint,
arXiv:1111.2652 [math.CO].
[124] M. Reineke, Cohomology of quiver moduli, functional equations, and integrality of
Donaldson–Thomas type invariants. Compositio Math. 147 (2011), 943–964.
[125] M. Reineke, Poisson automorphisms and quiver moduli. J. Inst. Math. Jussieu 9 (2010),
no. 3, 653–667.
[126] I. Reiten, Tilting theory and cluster algebras. Preprint, arXiv:1012.6014 [math.RT].
[127] I. Reiten, Cluster categories. In Proceedings of the International Congress of Mathemati-
cians, Vol. I, Hindustan Book Agency, New Delhi 2010, 558–594.
[128] C. M. Ringel, Some remarks concerning tilting modules and tilted algebras. Origin. Rel-
evance. Future. In Handbook of tilting theory, London Math. Soc. Lecture Note Ser. 332,
Cambridge University Press, Cambridge 2007, 413–472.
[129] D. Rupel, Quantum cluster characters. Preprint, arXiv:1109.6694 [math.QA].
[130] D. Rupel, On a quantum analog of the Caldero-Chapoton formula. Internat. Math. Res.
Notices 2010 (2010), 1–30, 10.1093/imrn/rnq192.
[131] R. Schiffler, On cluster algebras arising from unpunctured surfaces. II. Adv. Math. 223
(2010), no. 6, 1885–1923.
[132] M. Paul Schützenberger, Une interprétation de certaines solutions de l’équation fonction-
nelle: F .x C y/ D F .x/F .y/. C. R. Acad. Sci. Paris 236 (1953), 352–353.
[133] J. S. Scott, Grassmannians and cluster algebras. Proc. London Math. Soc. (3) 92 (2006),
no. 2, 345–380.
[134] P. Seidel and R. Thomas, Braid group actions on derived categories of coherent sheaves.
Duke Math. J. 108 (2001), no. 1, 37–108.
[135] P. Sherman and A. Zelevinsky, Positivity and canonical bases in rank 2 cluster algebras
of finite and affine types. Mosc. Math. J. 4 (2004), no. 4, 947–974, 982.
[136] D. Speyer and H. Thomas, Acyclic cluster algebras revisited. Preprint, arXiv:1203.0277
[math.RT].
[137] J. D. Stasheff, Homotopy associativity of H -spaces. I, II. Trans. Amer. Math. Soc. 108
(1963), 275-292; ibid. 108 (1963), 293–312.
[138] B. Szendrői, Non-commutative Donaldson-Thomas invariants and the conifold. Geom.
Topol. 12 (2008), no. 2, 1171–1202.
[139] T. Tran, F-polynomials in quantum cluster algebras. Algebr. Represent. Theory 14 (2011),
1025–1061,
[140] A. Yu. Volkov, Pentagon identity revisited I. Preprint, arXiv:1104.2267 [math.QA].
[141] S.-W. Yang and A. Zelevinsky, Cluster algebras of finite type via Coxeter elements and
principal minors. Transform. Groups 13 (2008), no. 3–4, 855–895.
[142] D. Zagier, Private communication, October 2011.
[143] D. Zagier, Polylogarithms, Dedekind zeta functions and the algebraic K-theory of fields.
In Arithmetic algebraic geometry (Texel, 1989), Progr. Math. 89, Birkhäuser, Boston,
MA, 1991, 391–430.
[144] D. Zagier, The dilogarithm function. In Frontiers in number theory, physics, and geometry.
II, Springer-Verlag, Berlin 2007, 3–65.
[145] A. Zelevinsky, Cluster algebras: notes for 2004 IMCC (Chonju, Korea, August 2004).
Preprint, arXiv:math/0407414 [math.RT].
[146] A. Zelevinsky, From Littlewood-Richardson coefficients to cluster algebras in three lec-
tures. In Symmetric functions 2001: surveys of developments and perspectives, NATO Sci.
Ser. II Math. Phys. Chem. 74, Kluwer Academic Publishers, Dordrecht 2002, 253–273.
[147] A. Zelevinsky, Cluster algebras: origins, results and conjectures. In Advances in algebra
towards millennium problems, SAS International Publications, Delhi 2005, 85–105.
[148] A. Zelevinsky, What is a cluster algebra? Notices Amer. Math. Soc. 54 (2007), no. 11,
1494–1495.
Some derived equivalences between noncommutative
schemes and algebras
Izuru Mori
1 Motivation and notation

Throughout this paper, we fix an algebraically closed field k, and assume that all
algebraic varieties, algebras, and categories are over k unless otherwise stated. For an
algebraic variety X , we denote by Qcoh X the category of quasi-coherent sheaves on
X and by coh X the category of coherent sheaves on X . For an algebra R, we denote by
Mod R the category of right R-modules and by mod R the category of finitely presented
right R-modules. For an abelian category C , we denote by D.C / the derived category
of C and by D b .C / the bounded derived category of C .
Let X be a smooth algebraic variety. If there is a commutative algebra R such
that coh X Š mod R, then X is an affine variety (by definition). If X is not an affine
variety, is it still possible to find an algebra R such that D b .coh X / Š D b .mod R/?
Presumably, this is hardly ever the case if we stick for R to be commutative. However,
it is sometimes possible to find a noncommutative algebra R such that D b .coh X / Š
D b .mod R/. For example, if DL b
.coh X / has a full strong exceptional sequence
fE1 ; : : : ; En g, then R D EndX . niD1 Ei / is a noncommutative algebra such that
D b .coh X/ Š D b .mod R/ by [5]. More generally, if E is a tilting generator and
R D EndX .E/, then D b .coh X / Š D b .mod R/ by [18]. Another interesting example
is a McKay correspondence. Suppose that a finite subgroup G SL.2; k/ naturally
A
acts on A2 and on S D kŒx; y. If A2 =G is the minimal resolution of the Kleinian
singularity A2 =G and S G is the skew group algebra of S by G, which is a non-
A
commutative algebra, then D b .coh A2 =G/ Š D b .mod.S G// by [7]. In this survey
paper, we will show that similar derived equivalences exist between noncommutative
schemes and noncommutative algebras.
Now we define noncommutative schemes. If R is a commutative algebra and
X D Spec R, then Qcoh X Š Mod R and coh X Š mod R. For this reason, if R is not
necessarily commutative, then we define the noncommutative affine scheme associated
to R by SpecncL R WD Mod R and specnc R WD mod R.
Let A D i2N Ai be an N-graded algebra. We denote by GrMod A the cate-
gory of graded right A-modules and by grmod A the category of finitely presented
graded right A-modules. Morphisms in these categories are right A-modules homo-
morphisms preserving L degrees. For M 2 GrMod A and n 2 Z, we define the degree
truncation Mn WD in Mi 2 GrMod A, and the degree shift M.n/ 2 GrMod A by

This work was supported by Grant-in-Aid for Scientific Research (C) 22540044.
186 I. Mori
M.n/i D MnCi for all i 2 Z. For M; N 2 GrMod A, we denote HomA .M; N / WD

HomGrMod A .M; N /, and define a graded vector space
M
ExtAi .M; N / WD ExtAi .M; N.n//:
n2Z
We say that an element m 2 M is torsion if mAn D 0 for some n 2 N. We say that

M is torsion-free if 0 2 M is the only torsion element of M . The torsion submodule
of M is defined to be the smallest submodule .M / M such that M= .M / is
torsion-free. We say that M is torsion if .M / D M . If Tors A GrMod A is the
full subcategory consisting of torsion modules, then Tors A is a dense subcategory of
GrMod A, that is, for every short exact sequence 0 ! L ! M ! N ! 0 in GrMod A,
M 2 Tors A if and only if L; N 2 Tors A, so we can form the quotient category
GrMod A= Tors A (cf. [3]). Moreover, if A is a graded right coherent, that is, if every
finitely generated homogeneous right ideal in A is finitely presented, then grmod A is
an abelian category, so we can also form the quotient category grmod A= tors A where
tors A is the full subcategory of grmod A consisting of torsion modules (cf. [13]). If
A is commutative and finitely generated in degree 1, that is, A Š kŒx1 ; : : : ; xn =I
with deg xi D 1 for all i and I C kŒx1 ; : : : ; xn is a homogeneous ideal, and X D
Proj A, then Qcoh X Š GrMod A= Tors A and coh X Š grmod A= tors A by [14]. For
this reason, if A is not necessarily commutative, then we define the noncommutative
projective scheme associated to A by Projnc A WD GrMod A= Tors A and projnc A WD
grmod A= tors A as in [3]. We often write M.n/ for the image of M.n/ 2 GrMod A
in Projnc A. For M; N 2 Projnc A, we denote HomA .M; N / WD HomProjnc A .M; N /,
and define a graded vector space
M
ExtiA .M; N / WD Ext iA .M; N .n//
n2Z
as before.
2 AS-regular algebras
L
For the rest of this paper, unless otherwise stated, let A D i2N Ai be a finitely
generated N-graded algebra, that is, A D khx1 ; : : : ; xn i=I where khx1 ; : : : ; xn i is the
free associative algebra on the finite set of generators fx1 ; : : : ; xn g with deg xi 2 N,
and I C khx1 ; : : : ; xn i is a homogeneous two-sided ideal.
Noncommutative algebraic geometry was motivated by the classification of AS-
regular algebras defined below. We say that an N-graded algebra A is connected over
R if R D A0 . In this terminology, A is connected graded in the usual sense if and only
if A is connected over k.
Definition 2.1 ([1]). A graded algebra S connected over k is called an AS-regular
algebra of dimension d and of Gorenstein parameter ` if
Some derived equivalences between noncommutative schemes and algebras 187
(1) gldim S D d < 1, and

´
0 if i ¤ d;
(2) ExtiS .k; S / Š as graded vector spaces.
k.`/ if i D d
The original definition of AS-regularity given in [1] requires the finiteness of GKdi-
mension, denoted by GKdim S < 1. In this paper, we drop this condition in the
definition of AS-regularity.
Since commutative AS-regular algebras are exactly weighted commutative polyno-
mial algebras, and projective spaces are projective schemes associated to commutative
polynomial algebras generated in degree 1, it may be reasonable to define a (weighted)
quantum projective space by Projnc S for some AS-regular algebra S .
The only AS-regular algebra of dimension 0 is k where the Gorenstein parameter
is 0, and the only AS-regular algebra of dimension 1 is kŒx where the Gorenstein
parameter is ` D deg x 2 N C . AS-regular algebras of dimension 2 were classified by
Zhang as below.
Theorem 2.2 ([20]). A graded algebra A connected over k is an AS-regular algebra

of dimension 2 and of Gorenstein parameter ` if and only if
ı X
n
A Š khx1 ; : : : ; xn i xi .xnC1i /
iD1
where n 2, deg x1 deg xn such that deg xi C deg xnC1i D ` for all i , and
2 Autk khx1 ; ; xn i is a graded algebra automorphism of khx1 ; : : : ; xn i.
Note that thePcondition deg xi C deg xnC1i D ` for all i is equivalent to saying
that the relation niD1 xi .xnC1i / is homogeneous of degree `.
AS-regular algebras of dimension 3 of finite GKdimension were classified using
commutative algebraic geometry by Artin–Tate–Van den Bergh [2] and Stephenson
[16], [17], which can be thought of as a starting point of noncommutative algebraic
geometry.
There is a way of constructing a new AS-regular algebra from old. Let A be a graded
algebra and 2 Autk A a graded algebra automorphism. A k-linear map ı W A ! A
of degree m 2 N C is called a graded -derivation if
ı.ab/ D .a/ı.b/ C ı.a/b
for all a; b 2 A. A graded Ore extension of A is a graded algebra AŒxI ; ı where

AŒxI ; ı D AŒx as a graded vector space with deg x D m and with the multiplication
induced by the formula xa WD .a/x C ı.a/. Clearly, AŒxI id; 0 Š AŒx as graded
algebras. Note that if D id, then -derivation is a usual derivation. We denote
AŒxI WD AŒxI ; 0 and AŒxI ı WD AŒxI id; ı.
188 I. Mori
Lemma 2.3 ([15]). If S is a right noetherian AS-regular algebra of dimension d and of

Gorenstein parameter `, then the graded Ore extension S ŒxI ; ı is a right noetherian
AS-regular algebra of dimension d C 1 and of Gorenstein parameter ` C deg x.
This produces many examples of noetherian AS-regular algebras. For example,
every iterated graded Ore extension of k
kŒx1 I 1 ; ı1 Œxn I n ; ın
is a noetherianAS-regular algebra of dimension n and of Gorenstein parameter deg x1 C
C deg xn .
Example 2.4. If 2 Aut k kŒx is a graded algebra automorphism defined by .x/ D
˛x where ˛ 2 k n f0g, then kŒxŒyI Š khx; yi=.yx ˛xy/. On the other hand, if
ı W kŒx ! kŒx is a k-linear map defined by ı.x i / D ix iC1 , then ı is a derivation and
kŒxŒyI ı Š khx; yi=.yx xy x 2 /. It is known that every AS-regular algebra of
dimension 2 and of finite GKdimension generated in degree 1 is isomorphic to one of
the above algebras.
It is known that every commutative projective space is derived equivalent to a finite
dimensional algebra of finite global dimension, called a Beilinson algebra. We can
similarly prove the following noncommutative generalization.
Theorem 2.5 ([12]). If S is a graded right coherent AS-regular algebra of dimension
d 1 and of Gorenstein parameter `, then f; .1/; : : : ; .` 1/g is a full strong
exceptional sequence for D b .projnc S/ where .i / is the image of S.i / 2 grmod S in
L
projnc S . Moreover, R WD End . `1iD0 .i // is a finite dimensional algebra of global
dimension d 1 such that D b .projnc S/ Š D b .specnc R/.
One of the purposes of this paper is to extend the above theorem.
3 Fano algebras
In [6], Bondal and Polishchuk introduced a notion of geometric exceptional sequence,
and showed that if a smooth projective variety X has a full geometric exceptional
sequence, then X is Fano, so it is interesting to find a class of smooth Fano varieties
which are derived equivalent to finite dimensional algebras of finite global dimension.
On the other hand, Minamoto recently found a class of finite dimensional algebras of
finite global dimension, called (quasi-)Fano algebras, which are derived equivalent to
smooth noncommutative projective schemes.
Recall that if L is an ample line bundle on a scheme X , then
M M
A WD .X; L˝i / Š HomX .OX ; L˝i /
i2N i2N
is a graded algebra such that X Š Proj A. We extend the notion of ampleness to a

noncommutative setting, which satisfies a similar property.
Definition 3.1 ([3]). Let C be an abelian category, O 2 C an object, and s 2 Aut C

an autoequivalence. We say that the pair .O; s/ is ample for C if
L
(1) for any M 2 C , there are `1 ; : : : ; `p 2 N such that piD1 s ì O ! M is an
epimorphism C , and
(2) for any epimorphism M ! N in C , there is n0 2 N such that, for every n n0 ,

the induced map HomC .s n O; M/ ! HomC .s n O; N / is surjective.
Very roughly speaking, .O; s/ is an ample pair for C if s n O is a projective generator

for C for n 0. The following result is the motivation of the above definition. We say
that a k-linear category C is Hom-finite if dimk HomC .M; N / < 1 for all M; N 2 C .
Note that if X is a projective variety, then coh X is Hom-finite.
Theorem 3.2 ([3], [13]). If C is a Hom-finite k-linear abelian category and .O; s/ is
an ample pair for C , then
M M
A WD HomC .s i O; O/ Š HomC .O; s i O/
i2N i2N
is a graded right coherent algebra such that C Š projnc A.
Unfortunately, we do not have a notion of ampleness for a derived category in

general yet, however, we will define one for a derived category of a finite dimensional
algebra following [8]. Let R be a finite dimensional algebra. The derived category
D b .mod R/ has the standard t -structure .D 0 ; D 0 / where
D 0 WD fM 2 D b .mod R/ j hi .M / D 0 for all i < 0g;

D 0 WD fM 2 D b .mod R/ j hi .M / D 0 for all i > 0g:
For an autoequivalence s 2 Autk D b .mod R/, we define the following new subcate-
gories of D b .mod R/:
D s;0 WD fM 2 D b .mod R/ j s i .M / 2 D 0 for all i 0g;

D s;0 WD fM 2 D b .mod R/ j s i .M / 2 D 0 for all i 0g:
Definition 3.3 ([8]). Let R be a finite dimensional algebra and s 2 Autk D b .mod R/
an autoequivalence.
(1) We say that s is quasi-ample if s i .R/ 2 D 0 \ D 0 Š mod R for all i 0.
(2) We say that s is ample if s is quasi-ample and .D s;0 ; D s;0 / is a t -structure for
D b .mod R/.
In fact, if s is ample for D b .mod R/, then the pair .R; s/ is ample for the heart of
the t-structure H D D s;0 \ D s;0 by [8], justifying the above definition.
190 I. Mori
Recall that if X is a smooth projective variety of dimension d , and !X is the

L
canonical sheaf on X, then ˝X !X Œd 2 Aut k D b .coh X / is the Serre functor. We
1
say that X is Fano if !X is ample. We use these facts to define a Fano algebra.
Let R be a finite dimensional algebra of global dimension d . Since R can be viewed
as an R-R bimodule, the k-vector space dual R WD Homk .R; k/ has a structure
of an R-R bimodule. We define the canonical module of R by !R WD R Œd 2
D b .mod R/. This is not a module but a complex, however, ˝LR !R Œd D ˝LR
R 2 Aut k D b .mod R/ is the Serre functor, so !R plays the role of the canonical
1
sheaf in algebraic geometry. The inverse of the canonical module is given by !R Š
RHomR .!R ; R/.
Definition 3.4 ([8]). A finite dimensional algebra R is called a (quasi-)Fano algebra
of dimension d if
(1) gldim R D d < 1, and
1
(2) ˝LR !R 2 Autk D b .mod R/ is (quasi-)ample.
If X is a smooth Fano variety of dimension d having a full geometric exceptional
L
sequence fE0 ; : : : ; Ed g, then R D EndX . diD0 Ei / is a Fano algebra of dimension d
by [10], so X is derived equivalent to a Fano algebra. We will see that Fano algebras
are also interesting algebras to study in representation theory of finite dimensional
algebras.
A quiver Q consists of a set of vertices Q0 and the set of arrows Q1 between vertices.
For an arrow ˛ 2 Q1 , we write s.˛/; t .˛/ 2 Q0 for the starting and terminating vertices
of ˛. A path of length n is a sequence of arrows ˛1 ˛2 ˛n such that t .˛i / D s.˛iC1 /
for i D 1; : : : ; n 1. A vertex is viewed as a path of length 0. A path algebra kQ of
a quiver Q is a vector space spanned by paths where the multiplication is given by the
concatenation of paths.
Theorem 3.5 ([8]). Let Q be a finite acyclic quiver. Then the path algebra R D kQ
is a Fano algebra if and only if Q is of infinite representation type.
x D .Q
Let Q D .Q0 ; Q1 / be a quiver. The double Q x0 ; Q
x 1 / of Q is defined by
x 0 D Q0 ;
Q
x 1 D f˛ W i ! j; ˛ W j ! i j ˛ 2 Q1 g:
Q
The preprojective algebra of Q is defined by

X
…Q WD k Q=.x ˛˛ ˛ ˛/:
˛2Q1
We now extend the notion of preprojective algebra to a quiver with relations. If R is

a finite dimensional algebra of global dimension d , then we define the preprojective
algebra of R by …R WD TR .Ext dR .R ; R//, the tensor algebra of the R-R bimodule
ExtdR .R ; R/ over R. We endow the grading on …R by that on the tensor algebra.
1 1
Note that if R is a quasi-Fano algebra, then !R Š ExtdR .R ; R/, so …R Š TR .!R /.
If R D kQ=I is a path algebra of a quiver Q with relations I C kQ, then we write
….Q; I / WD …R. By this notation, it is known that ….Q; 0/ Š …Q (cf. [8]), so
….Q; I / is a generalization of the usual preprojective algebra of Q defined above.
As we mentioned earlier, a Fano algebra is derived equivalent to a noncommutative
projective scheme.
Theorem 3.6 ([8]). If R is a Fano algebra, then …R is graded right coherent and
D b .specnc R/ Š D b .projnc …R/.
4 Generalization
In order to combine results in the previous two sections, we generalize the notion of
AS-regularity.
Definition 4.1 ([9]). A graded algebra S connected over R is called an AS-regular
algebra over R of dimension d and of Gorenstein parameter ` if
(1) gldim S D d < 1, gldim R < 1, and
´
0 if i ¤ d;
(2) ExtS .R; S / Š
i

as graded right and left R-modules.
R .`/ if i D d;
Of course, an AS-regular algebra over k is a usual AS-regular algebra as we defined
earlier. L
We construct a few new graded algebras from old. Let A D i2Z Ai be a Z-graded
algebra. For r 2 N C , we define the r-th Veronese algebra of A by
M
A.r/ WD Ari ;
i2Z
and the r-th quasi-Veronese algebra of A by

0 1
Air AirC1 AirCr1
MB B Air1 Air AirCr2 C
C
AŒr WD B :: :: :: :: C
@ : : : : A
i2Z
AirrC1 AirrC2 Air
For a graded algebra automorphism 2 Aut k A, we define a new graded algebra A
where A D A as a graded vector space with a new multiplication a b D a i .b/
for every a 2 Ai ; b 2 Aj . It is hardly ever the case that GrMod A.r/ Š GrMod A for
r 2, however, it is always the case that GrMod AŒr Š GrMod A for any r 2 N C by
[10], and GrMod A Š GrMod A for any 2 Autk A by [19].
If S is an AS-regular algebra over R of Gorenstein parameter `, then we define the
Beilinson algebra of S by rS WD .S Œ` /0 . Our first main result is below.
192 I. Mori
Theorem 4.2 ([9]). If S is an AS-regular algebra over R of dimension d 1 and of

Gorenstein parameter `, then
(1) rS is a quasi-Fano algebra of dimension d 1,
(2) S Œ` Š ….rS/ for some graded algebra automorphism 2 Aut k ….rS /,
(3) GrMod S Š GrMod ….rS/, and
(4) D.Projnc S/ Š D.Specnc rS/.
Moreover, if S is graded right coherent, then rS is Fano and
D b .projnc S/ Š D b .specnc rS /:
Note that if S is not connected over k, then f; .1/; : : : ; .` 1/g is no longer a
(full strong) exceptional sequence for D b .projnc S /.
We say that S is symmetric if 2 Aut k ….rS / appearing in the above theorem is
the identity so that S Œ` Š ….rS/. Then we can characterize the preprojective algebra
of a quasi-Fano algebra as follows.
Theorem 4.3 ([9]). Symmetric AS-regular algebras over R of dimension d 1 and
of Gorenstein parameter 1 are exactly preprojective algebras of quasi-Fano algebras
of dimension d 1.
5 McKay correspondence
Let G be a group acting on an algebra R. The fixed subalgebra of R by G is defined
by RG WD fr 2 R j g.r/ D r for all g 2 Gg R, and the skew group algebra of
R by G is defined by R G D R ˝k kG as a vector space with the multiplication
.r g/.s h/ D rg.s/ gh.
1
Let G GL.n; k/ be a finite subgroup, fi gNiD0 the set of irreducible represen-
tations of G, and the natural n-dimensional representation of G by the inclusion.
L 1 mij
Define mij 2 N by the irreducible decomposition j ˝ D N iD0 i . The McKay
quiver QG of G is the quiver whose set of vertices is .QG /0 D f0; 1; : : : ; N 1g and
the number of arrows from i to j is mij .
The derived equivalences in the theorem below are called a McKay correspondence,
which are essentially proved in [7]. (We arrange the statements of the results for the
purpose of this paper.)
Theorem 5.1. If G SL.2; k/ is a finite subgroup naturally acting on A2 and S D
kŒx; y, then the following hold.
(1) A2 =G Š Spec S G .
xD
(2) There is a quiver Q whose underlying graph is extended Dynkin such that Q
QG and specnc …Q Š specnc .S G/ Š specnc EndS G .S /.
(3) (McKay correspondence)
A C
D b .coh A2 =G/ Š D b .coh Spec S G /
Š D b .specnc …Q/
Š D b .specnc .S G//
Š D b .specnc EndS G .S //
A
where A2 =G is the minimal resolution of the Kleinian singularity A2 =G.
We will show that analogous results hold for an AS-regular algebra. If S D
khx1 ; : : : ; xn i=IS is an AS-regular algebra of Gorenstein parameter ` with deg xi D
ai 2 N C such that gcd.a1 ; : : : ; an / D 1, then the cyclic subgroup G D hgi
GL.n; k/ of order ` where 2 k is a primitive `-th root of unity and
0 a 1
1 0 0
B 0 a2 0 C
B C
gDB : :: :: C 2 GL.n; k/
@ :: : : 0 A
0 0 an
naturally acts on S. Every irreducible representation of G is 1-dimensional, and we

may define i by i .g/ D i 2 GL.1; k/ D k n f0g. Then the McKay quiver QG of
G is given by
.QG /0 D f0; 1; : : : ; ` 1g;

.QG /1 D f˛ij W i ! i C aj .mod `/ j 0 i ` 1; 1 j ng:
e
We further define the reduced McKay quiver QG of G by
e
.QG /0 D .QG /0 ;
e/
.Q G 1 D f˛ 2 .QG /1 j s.˛/ < t .˛/g:
Our second main result is as follows.

Theorem 5.2 ([11]). Let S and G be as above. If gldim S 2, then the following
hold.
e e
(1) projnc ….QG ; IeS / Š projnc .S G/ where IeS C k QG is a two-sided ideal
naturally induced by IS .
e
(2) If S is symmetric, then ….QG ; IS / Š S G.
194 I. Mori
(3) If S is noetherian and generated in degree 1, or S is commutative such that

gcd.ai ; `/ D 1 for all i , then S G Š EndS G .S /, so that projnc ….QG ; IeS / Še
projnc .S G/ Š projnc EndS G .S /.
Moreover, the following are equivalent.
(1) gldim S D 2.
e e
(2) IeS D 0 so that projnc …QG Š projnc .S G/ where …QG is the usual prepro-
jective algebra of QG .e
e
(3) QG D QG .
Since projnc .S G/ is a “smooth” noncommutative projective scheme which is often
“birationally equivalent” to projnc S G by [11], we may view projnc .S G/ as a “resolu-
C
tion” projnc S G of projnc S G . Moreover, if S is commutative such that gcd.ai ; `/ D 1
for all i , then cohŒAn n f.0; ; 0/g=Gm Š projnc S G where Gm D k n f0g acts on
An n f0; : : : ; 0g by .x1 ; : : : ; xn / D .a1 x1 ; : : : ; an xn / and ŒAn n f.0; ; 0/g=Gm
is the quotient stack by [4] and [11], so the above theorem looks like a McKay corre-
spondence.
6 Examples
In this last section, we will give three concrete examples. Let S be an AS-regular
algebra and G GL.n; k/ the corresponding subgroup as in the previous section. It
was shown in [20] that if gldim S D 2, then the following are equivalent.
(1) GKdim S < 1.
(2) S is noetherian.
(3) G SL.2; k/.
e
(4) The underlying graph of QG is extended Dynkin.
Example 6.1. If S D kŒx; y with deg x D 1, deg y D 3, then S is a noetherian
AS-regular algebra of dimension
D E2 and of Gorenstein parameter 4. The corresponding
0
cyclic subgroup is G D 0 3 SL.2; k/ where is a primitive 4-th root of unity.
Since S is commutative, S G Š EndS G .S /. The reduced McKay quiver QG and e
the McKay quiver QG of G are
/1 /
0 0O o 1O
and
/
3o 2 3o 2.
e
Since GKdim S D 2 < 1, the underlying graph of QG is extended Dynkin. Since
gldim S D 2, QG D QG . e
Example 6.2. If S D khx; y; zi=.xz C y 2 C zx/ with deg x D 1, deg y D 2,
deg z D 3, then S is a graded coherent (but not noetherian) AS-regular algebra of
dimension
2 and of Gorenstein parameter 4. The corresponding cyclic subgroup is
0 0
GD 0 2 0 GL.3; k/ where is a primitive 4-th root of unity. The reduced
e
0 0 3
McKay quiver QG and the McKay quiver QG of G are
/1 /
0= 0O ^=o == @ 1O
== ====
=
=== and =====
==/
3o 2 3o 2.
e
Since GKdim S D 1, the underlying graph of QG is no longer extended Dynkin.
Since gldim S D 2, QG D QG . e
Example 6.3. If S D khx; yi=.x 2 y yx 2 ; xy 2 y 2 x/ with deg x D deg y D 1, then
S is a noetherian AS-regular algebra of dimension
D E 3 and of Gorenstein parameter 4.
The corresponding cyclic subgroup is G D 0 0 GL.2; k/ where is a primitive
4-th root of unity. Since S is generated in degree 1, S G Š EndS G .S /. The reduced
e
McKay quiver QG and the McKay quiver QG of G are
// //
0 1 0O O 1
and
oo o
3 2 3o 2.
e
The underlying graph of QG is not extended Dynkin. Since gldim S ¤ 2, QG ¤ QG . e
References
[1] M. Artin and W. Schelter, Graded algebras of global dimension 3. Adv. Math. 66 (1987),
171–216.
[2] M. Artin, J. Tate, and M. Van den Bergh, Some algebras associated to automorphisms of
elliptic curves. In The Grothendieck Festschrift, Vol. I, Progr. Math. 86, Birkhäuser, Boston,
MA, 1990, 33–85.
[3] M. Artin and J. J. Zhang, Noncommutative projective schemes. Adv. Math. 109 (1994),
228–287.
[4] D. Auroux, L. Katzarkov, and D. Orlov, Mirror symmetry for weighted projective planes
and their noncommutative deformations. Ann. of Math. 167 (2008), 867–943.
196 I. Mori
[5] A. I. Bondal, Representation of associative algebras and coherent sheaves. Math. USSR Izv.
34 (1990), 23–42.
[6] A. I. Bondal and A. E. Polishchuk, Homological properties of associative algebras: The
method of helices. Russian Acad. Sci. Izv. Math. 42 (1994), 219–260.
[7] M. Kapranov and E. Vasserot, Kleinian singularities, derived categories and Hall algebras.
Math. Ann. 316 (2000), 565–576.
[8] H. Minamoto, Ampleness of two-sided tilting complexes. Internat. Math. Res. Notices 2012
(2012), no. 1, 67–101.
[9] H. Minamoto and I. Mori, The structure of AS-Gorenstein algebras. Adv. Math. 226 (2011),
4061–4095.
[10] I. Mori, B-construction and C-construction. Comm. Algebra, to appear.
[11] I. Mori, McKay type correspondence for AS-regular algebras. Preprint.
[12] D. Orlov, Derived categories of coherent sheaves and triangulated categories of singulari-
ties. In Algebra, arithmetic, and geometry: in honor of Yu. I. Manin, Vol. II, Progr. Math.
270, Birkhäuser, Boston, MA, 2009, 503–531.
[13] A. Polishchuk, Noncommutative proj and coherent algebras. Math. Res. Lett. 12 (2005),
63–74.
[14] J. P. Serre, Faisceaux algébriques cohérents. Ann. of Math. 61 (1955), 197–278.
[15] S. P. Smith, Non-commutative algebraic geometry. Lecture notes, University of Washing-
ton, 1994.
[16] D. R. Stephenson, Artin-Schelter regular algebras of global dimension three. J. Algebra
183 (1996), 55–73.
[17] D. R. Stephenson, Algebras associated to elliptic curves. Trans. Amer. Math. Soc. 349
(1997), 2317–2340.
[18] Y. Toda and H. Uehara, Tilting generators via ample line bundles. Adv. Math. 223 (2010),
1–29.
[19] J. J. Zhang, Twisted graded algebras and equivalences of graded categories. Proc. Lond.
Math. Soc. 72 (1996), 281–311.
[20] J. J. Zhang, Non-noetherian regular rings of dimension 2. Proc. Amer. Math. Soc. 126
(1998), 1546–1653.
Lagrangian-invariant sheaves and functors for abelian
varieties
Alexander Polishchuk
Introduction
Derived categories of coherent sheaves on smooth projective varieties have been playing
an increasingly significant role in modern algebraic geometry. They serve as a moti-
vation and a testing ground for constructions in noncommutative geometry of smooth
proper dg-categories (see [10]) and have interesting links to representation theory, mir-
ror symmetry and other areas (see [2] and references therein). One of the first results
that revealed the importance of derived categories was the equivalence of the derived
categories D b .A/ ' D b .A/y for an abelian variety A and its dual Ay established by
Mukai [13]. The present paper continues the line of study of [18], where it was shown
that the group of autoequivalences of D b .A/ is closely related to the group U.A A/y
y
of “symplectic” automorphisms of A A (the term “symplectic” is explained below).
Namely, we study endofunctors of D b .A/ related to symplectic automorphisms of
A Ay in the bigger category AbQ of abelian varieties up to isogeny. More generally,
we study a natural class of LI-functors between derived categories of abelian varieties
governed by Lagrangian correspondences between the corresponding varieties A A. y
Our motivation is twofold. On the one hand, such functors could be useful for the
Manin’s real multiplication program (see [11]) if one tries to approach it using the
categories of holomorphic bundles on noncommutative tori described in [24]. On the
other hand, as will be shown in a sequel to this paper, using these functors one can
realize the Bridgeland stability space of an abelian surface A as a homogeneous space
for certain covering of the group U.A A; y R/ of R-points of the algebraic group of
y
symplectic automorphisms of A A (more precisely, one should consider the connected
component of the stability space described in [3]).
As in [20], with every abelian variety A we associate a “symplectic” object: the
abelian variety XA D A Ay together with a skew-symmetric biextension E D p14
P˝
1 y
p23 P of XA XA (where P is the Poincaré line bundle on A A). There is an analog
of the Schrödinger representation of the Heisenberg group in this setting obtained
by considering the action of a certain Heisenberg extension H of XA on D b .A/ by
translations and tensoring with line bundles in Pic0 .A/ (see Section 2.1). One can
show that under some technical assumptions, the stabilizer of an endosimple object of
D b .A/ is always an isotropic subgroup Y XA with respect to the biextension E (see
Remark 2.3.3). We study the case when this subgroup is Lagrangian, i.e., when we have

Supported in part by the NSF grant DMS-1001364.
198 A. Polishchuk
a duality between Y and XA =Y (we show that this is equivalent to Y having the maximal
possible dimension dim A D dim XA =2). Our first main result is the description of the
subcategory of invariants in D b .A/ of a Lagrangian abelian subvariety Y XA (see
Theorem 2.4.5). More precisely, to define this subcategory of invariants we need to
choose a line bundle ˛ on Y that gives a lifting of Y to the Heisenberg groupoid H of
A (see Section 2.2). Then we can define the subcategory of .Y; ˛/-invariant objects in
D b .A/. We show that this subcategory is generated by a single endosimple coherent
sheaf SY;˛ and that every .Y; ˛/-invariant coherent sheaf is a direct sum of several copies
of SY;˛ . We call such sheaves Lagrangian-invariant (LI-sheaves for short). Note that
the case when the projection Y ! A is an isogeny corresponds to semihomogeneous
vector bundles on A (see [14]). However, our condition of .Y; ˛/-invariance is stronger
than the condition of invariance under all functors corresponding to points in Y .k/.
Namely, the condition for a sheaf F 2 Coh.A/ to be .Y; ˛/-invariant is given by a
certain isomorphism over A Y (see Section 2.4). This accounts for the fact that
nontrivial extensions of SY;˛ with itself are not .Y; ˛/-invariant (see Example 2.4.1).
Note that finite Heisenberg groups appear in our picture as well. Namely, for a
pair of Lagrangian subvarieties .Y; Z/ in XA equipped with liftings to the Heisenberg
groupoid, we have a natural group extension G of the intersection Y \ Z by Gm ,
which is a Heisenberg group in the case when Y \ Z is finite (this construction goes
back to [20]). Generalizing Mumford’s theory of theta groups we show that the space

HomD b .A/ .SY;˛ ; SZ;ˇ / is an irreducible representation of the Heisenberg group G (see
Corollary 3.2.12).
For a pair of abelian varieties A and B we consider Lagrangian correspondences
from XA to XB , i.e., Lagrangian abelian subvarieties L XA XB with respect to
the difference of biextensions E pulled-back from XA XA and XB XB . More
precisely, we consider pairs .L; ˛/, where ˛ is a line bundle on L that gives a lifting
of L to the Heisenberg groupoid of XA XB . As above, with each pair .L; ˛/ we
associate the LI-sheaf SL;˛ on A B. Then we define the corresponding LI-functor
ˆL;˛ W D b .A/ ! D b .B/ as the Fourier–Mukai functor associated with SL;˛ . Our
main result is the calculation of the composition of functors of this form. We do this
in two steps. First, we introduce generalized Lagrangians in XA by allowing for Y
not to be connected and for the homomorphism Y ! XA to have finite kernel. The
condition for Y to be Lagrangian is formulated in terms of duality between Y and the
complex ŒY ! XA . The context for such a duality is provided by the category of orbi-
abelian varieties studied in Section 4 of [23]. The theory of invariants of Lagrangians in
D b .A/ carries over to generalized Lagrangians. We define composition of generalized
Lagrangian correspondences (which is again generalized Lagrangian) and show that
the composition of LI-functors is compatible with the composition of correspondences
up to a shift (see Theorem 3.2.14). The second step consists of the study of the relation
between the LI-sheaf SY;˛ associated with a generalized Lagrangian .Y; ˛/ for XA
and the corresponding usual Lagrangian obtained by passing to the image in XA of
the connected component of zero Y0 Y (see Proposition 2.4.7). The final result
(in characteristic zero) is that the composition of two LI-functors decomposes as an
Lagrangian-invariant sheaves and functors for abelian varieties 199
explicit direct sum of LI-functors. In the case B D A we obtain an explicit description

of a certain semiring of endofunctors of D b .A/ in terms of the group U.XA ; Q/ of
symplectic automorphisms of XA in AbQ . Considering these endofunctors up to a
natural action of XA .k/ we get a simpler picture: the resulting monoid has a group
completion that can be identified with an explicit central extension of U.XA ; Q/ by
Z Q . The Z-part of this central extension is described by essentially the same
2-cocycle as in the picture with autoequivalences of D b .A/ (see [18], Section 4). The
Q -part of the central extension is a new feature of the picture with endofunctors, and we
show that it can be nontrivial in the case of elliptic curves with complex multiplication.
The paper is organized as follows. In Section 1 we gather necessary results about
finite Heisenberg group schemes and twisted equivariant sheaves, along with some other
background facts. In Section 2 we review the symplectic setting from [20]. The main
object here is the Heisenberg groupoid H associated with a symplectically self-dual
abelian variety (equipped with some extra data). We define generalized isotropic (resp.,
Lagrangian) pairs and the corresponding representations of H. The main result of this
section is Theorem 2.4.5 describing the subcategories of invariants of a generalized
Lagrangian pair in standard representations of H. Another important technical result
is Proposition 2.4.7 giving a relation between invariants for a generalized Lagrangian
i W Y ! X and the associated Lagrangian i.Y0 / X . In Section 3 we study LI-kernels
and LI-functors. First, we define generalized Lagrangian correspondences (or g-Lag-
correspondences) and show that they can be composed under some assumptions. Then
in 3.2 we prove the compatibility between the composition of g-Lag-correspondences
and the composition of the corresponding LI-functors up to a shift (see Theorems 3.2.11
and Theorem 3.2.14). Finally in 3.3 assuming that the characteristic is zero we describe
a semiring of endofunctors of D b .A/ built from LI-functors and the related central
extension of the group U.XA ; Q/.
Notation and conventions. We consider schemes over a fixed algebraically closed

ground field k. We say that an object F of a k-linear category C is endosimple if
HomC .F; F / D k. For a scheme X we denote by D b .X / the bounded derived category
of coherent sheaves on X . We say that an object F 2 D b .X / is cohomologically pure if
H such that F ' H Œn for n 2 Z. For F1 ; F2 2 D b .X / we
there exists a coherent sheafL

set HomD b .X/ .F1 ; F2 / D i HomD b .X/ .F1 ; F2 Œi /. We denote by AbQ the category
of abelian varieties up to an isogeny (i.e., the localization of the category of abelian
varieties over k with respect to the class of isogenies). For a finite group scheme
G by a representation of G on a vector space we always mean a finite-dimensional
representation. For a group scheme G we denote by G0 the connected component of
the neutral element in G with the reduced scheme structure, and we set 0 .G/ D G=G0 .
All our biextensions are biextensions by Gm . Most of the facts about biextensions we
use are summarized in [22], II.10. We always normalize the Poincaré line bundle
P D PA on A Ay by requiring it to be trivial over f0g Ay and A f0g (thus making it
a biextension of A A).y For a line bundle L over a commutative group scheme X we
1
denote by ƒ.L/ the line bundle on X X given by ƒ.L/x;x 0 D LxCx 0 ˝ L1 x ˝ Lx 0 .
200 A. Polishchuk
1 Preliminaries
1.1 Finite Heisenberg group schemes. In this section we recall some facts about
Heisenberg group schemes and their representations of weight one following Sec-
tion 12.2 of [22] (see also Chapter V in [12] and [21], Sections 1, 2).
A finite Heisenberg group scheme G is a central extension of a finite commutative
group scheme K by Gm , such that the commutator pairing K K ! Gm is perfect,
i.e., induces an isomorphism K ' K , where K is the Cartier dual to K.
In general, when G is a central extension of some group scheme K by Gm we say
that a finite-dimensional representation V of G is of weight one if Gm G acts on
V via the identity character. All weight-1 representations of a finite Heisenberg group
scheme are direct sums of several copies of the unique irreducible representation of
weight one, called Schrödinger representation. Its dimension is equal to jKj1=2 , where
jKj is the order of K.
Sometimes it is convenient to describe a Heisenberg extension G in terms of the
line bundle L over K associated with G viewed as a Gm -torsor over K. The group law
on G corresponds to an isomorphism
LxCy ' Lx ˝ Ly : (1.1.1)
of line bundles on K K. A representation of G of weight one on a finite dimensional

vector space V can be described as a morphism L ˝ V ! OK ˝ V of bundles over K
satisfying the natural compatibility with (1.1.1). We denote by G-rep1 the category of
weight-1 representations of G.
Any subgroup I K, isotropic with respect to the commutator pairing, can be
lifted to a subgroup of G. Equivalently, there is a trivialization of L over I , compatible
with (1.1.1). Any maximal isotropic subgroup of K has order jKj1=2 . The Schrödinger
representation can be realized explicitly as follows. The space H 0 .K; L/ has natural
commuting left and right G-actions of weight one: the corresponding isomorphisms
L ˝ H 0 .K; L/ ' OK ˝ H 0 .K; L/
are obtained from (1.1.1) by taking the push-forward with respect to the projection
K K ! K to one of the factors. Now if I K is a maximal isotropic subgroup
then we can lift it to a subgroup of G and take invariants H 0 .K; L/I with respect to the
right action. This becomes an irreducible G-representation of weight one. Trivializing
the line bundle L over K we can identify H 0 .K; L/I with the subspace of functions
f on K satisfying certain quasi-periodicity with respect to translations by I , which
is the customary description of the Schrödinger representation. If V is a Schrödinger
representation of G and I K is a maximal isotropic subgroup with a lifting to G
then V I is one-dimensional. Hence, if W is any weight-1 representation of G then we
have an isomorphism of G-representations (depending on a trivialization of V I ):
W ' V ˝ W I:
Note that if V is a Schrödinger representation of G then V is a Schrödinger

representation of G op (using the isomorphism G ! G op W g 7! g 1 we can also view
V as a representation of G of weight 1). For the entire space H 0 .K; L/ we have an
isomorphism of G G op -modules
H 0 .K; L/ ' V ˝ V : (1.1.2)
Often when we talk about an isotropic subgroup in K lifted to G we just say that I is
an isotropic subgroup in G. If I is such a subgroup then the normalizer subgroup N.I /
contains Gm and N.I /=Gm D I ? , the orthogonal complement to I with respect to the
commutator pairing. The quotient N.I /=I is a Heisenberg extension of I ? =I . If V is
a Schrödinger representation of G then the space V I of I -invariants is a Schrödinger
representation of N.I /=I . Viewed as a representation of I , V is a direct sum of
dim.V I / copies of the regular representation H 0 .I; O/ ' H 0 .I ; O/. Hence, in the
case when I is reduced we get an isomorphism
M
V ' VI ˝
2I
of I -representations.
1.2 1-cocycles with values in Picard stacks and twisted equivariant sheaves. Let
us recall briefly some definitions from [23], Section 3.4. Let X be a scheme, G a group
scheme acting on X . A 1-cocycle of G with values in the Picard stack Pic.X / of line
bundles on X is a line bundle L on G X equipped with an isomorphism
Lg1 g2 ;x ' Lg1 ;g2 x ˝ Lg2 ;x (1.2.1)
on G G X such that the following diagram of isomorphisms on G G G X

is commutative:
Lg1 g2 g3 ;x / Lg1 ;g2 g3 x ˝ Lg2 g3 ;x
(1.2.2)

Lg1 g2 ;g3 x ˝ Lg3 ;x / Lg1 ;g2 g3 x ˝ Lg2 ;g3 x ˝ Lg3 ;x
(cf. [23], (3.4.2)). Given such a 1-cocycle L we can consider L-twisted G-equivariant
sheaves on X as sheaves equipped with an isomorphism
Fgx ' Lg;x ˝ Fx (1.2.3)

202 A. Polishchuk
on G X subject to the commutativity of the diagram

Fg1 g2 x / Lg1 ;g2 x ˝ Fg2 x
(1.2.4)

Lg1 g2 ;x ˝ Fx / Lg1 ;g2 x ˝ Lg2 ;x ˝ Fx
on G G X (cf. Section 3.4 of [23]). We denote the category of L-twisted G-

equivariant coherent sheaves on X by CohH L
.X /. Similarly, we define the category
DHL
.X/ of L-twisted G-equivariant objects in the derived category D b .X / as objects
F 2 D b .X/ equipped with an isomorphism (1.2.3) in D b .G X /, subject to the same
compatibilities.
For example, if G ! K is Heisenberg group extension then the underlying Gm -
torsor L over K has a structure of a 1-cocycle of K with values in the Picard stack
Pic.k/ of 1-dimensional vector spaces, and the category G-rep1 can be identified with
the category of L-twisted K-equivariant coherent sheaves on Spec.k/.
Lemma 1.2.1. Let G be a central extension of a finite commutative group scheme K
x D G=C !
by Gm with the center C Gm , such that the induced central extension G
K=C is a Heisenberg group. Let pGx W G ! G x and pK W G ! K be the projections,
x Then there exists an equivalence
and let V be a Schrödinger representation of G.
ˆ W G- rep1 ! C - rep (1.2.5)

such that ˆ.pG
x V / is the trivial one-dimensional representation of C , and for a char-
acter W K ! Gm we have
ˆ. W / ' jC ˝ ˆ.W /; (1.2.6)
where W G ! G is the automorphism sending g to .pK .g// g.
Proof. By Lemma 3.5.3 (i) of [23], the extension of commutative groups

0 ! C ! K ! K=C ! 0 (1.2.7)
gives rise to a biextension B of K=C C . Furthermore, if M z is the 1-cocycle
of K with values in Pic.k/ corresponding to the extension G ! K, then we have
Mz ' M0 , where M0 is the 1-cocycle of K=C with values in Pic.k/ corresponding
x ! K=C . We can extend M0 to a 1-cocycle M of K=C with values
to the extension G

in Pic.C / by the formula
Mx;c D Bx;c ˝ .M0 /x ;
where x 2 K=C , c 2 C . Then one has a natural isomorphism of k-algebras
z
! kŒKM ;
˛ W p OC ŒK=C M

where p W C ! Spec.k/ is the projection to the point (see [23], Proposition 3.5.4,
for a more general result). Here for a group H acting trivially on a scheme X and a
1-cocycle L of H with values in Pic.X / we denote by OX ŒH L the twisted group
algebra of H over X (see [23], (3.4.9)). The equivalence between the categories of
modules induced by ˛ can be interpreted as an equivalence
M z
CohK=C .C / ' CohK
M
.Spec.k//:
As we have seen above, the category on the right can be identified with G-rep1 . On
M
the other hand, CohK=C .C / is equivalent to the category of weight-1 representations
of the finite Heisenberg group scheme Gx over C , where Gx is the extension of the
constant finite group scheme .K=C /C by Gm associated with the 1-cocycle M. Since
the obstacle to the existence of a Schrödinger representation is an element of the
Brauer group Br.C / D 0 (see [21], Section 2), the latter category is equivalent
to Coh.C / ' C - rep. Thus, we obtain the required equivalence (1.2.5).
It is easy to check that ˛ is compatible with natural homomorphisms of both sides
to kŒK=C M0 . Note that modules over kŒK=C M0 are exactly weight-1 representa-
tions of G.x This compatibility implies that p V corresponds to the trivial represen-
x
G
tation of C under the equivalence ˆ. The compatibility (1.2.6) with automorphisms
W G ! G follows from the fact that the whole construction is functorial with respect
to isomorphisms G ! G 0 , identical on Gm .
We will use the following description of the categories of twisted equivariant sheaves
associated with some transitive actions.
Proposition 1.2.2. Let H be a group scheme acting transitively on X with finite
stabilizers, and let Lh;x be a 1-cocycle for the action of H on X .
(i) For a point x 2 X there is a natural central extension
1 ! Gm ! Gx ! Hx ! 1
of the stabilizer subgroup Hx of x with the underlying line bundle LjHx fxg . Consider
the line bundle L.x/ WD LjH fxg over H , and let .x/ W H ! X be the map sending h
to hx. Then L.x/ has a natural .idH .x// L-twisted H -equivariant structure with
respect to the regular left action of H on itself. The right action of Hx on H extends
to an action of Gx on L.x/ respecting the above twisted H -equivariant structure.
The map F 7! F jx extends to an equivalence
fibx W CohH
L
.X / ! Gx - rep1 (1.2.8)
of L-twisted H -equivariant coherent sheaves on X with the category of weight-1
representations of Gx .
(ii) For an element h 2 H the isomorphism Hx ! Hhx given by the conjugation
by h extends to an isomorphism of central extensions
h W Gx ! Ghx : (1.2.9)
204 A. Polishchuk
Furthermore, there is a natural isomorphism of functors
Lh;x ˝ fibx ' h B fibhx ; (1.2.10)
where h W Ggx - rep1 ! Gx - rep1 is the equivalence induced by h (Lh;x is just a
one-dimensional vector space one has to insert to make the isomorphism canonical).
(iii) Now let f W X ! X be an automorphism commuting with the H -action. Let
M be a line bundle on X together with an isomorphism of cocycles
1
f W Mx ˝ Mhx ˝ Lh;x ! Lh;f .x/
(so M is a coboundary for L ˝ .id f / L1 ). Then the map F 7! f F ˝ M extends

to an autoequivalence f;M of CohHL
.X /, so that the following diagram of functors is
commutative up to an isomorphism:
L
CohH .X /
fibx
/ Gx - rep1
f;M .˛f / (1.2.11)

L
CohH .X / / Gx - rep1 ,
fibx
where ˛f is the composition

1
h
Nf 0
Gx ! Gf .x/ ! Gx
for some h0 2 H such that h0 x D f .x/. Here the homomorphism Nf is given by the
restriction of f to Hx fxg.
Proof. (i) Since the embedding of the point x into X is an Hx -morphism, the re-
striction of LjHx fxg has a natural structure of a 1-cocycle of Hx with values in the
Picard groupoid of the point, hence it defines a central extension of Hx by Gm . The
.idH .x// L-twisted H -equivariant structure on L.x/ is given by the isomorphism
L.x/h0 h ' Lh0 ;.x/.h/ ˝ L.x/h ;
where h; h0 2 H , obtained from (1.2.1) by the restriction to H H fxg. Restricting

(1.2.1) further to H Hx fxg we obtain an isomorphism
L.x/hh0 ' L.x/h ˝ Lh0 ;x ; (1.2.12)
where h 2 H , h0 2 Hx . This isomorphism can be interpreted as a right action of

Gx on L.x/. The fact that this action commutes with the twisted left H -equivariant
structure on L.x/ follows from the commutativity of diagram (1.2.2).
For an L-twisted H -equivariant coherent sheaf F on X we can restrict the structure

isomorphism
Fhx ' Lh;x ˝ Fx
to Hx fxg H X and get an isomorphism
a.h0 ;x/
Fx ! Lh0 ;x ˝ Fx (1.2.13)
for h0 2 Hx satisfying the natural compatibility condition on Hx Hx . Thus, we can

view Fx as a weight-1 representation of Gx . This gives the functor fibx .
Conversely, starting with a representation W of Gx of weight 1 we can construct an
L-twisted H -equivariant sheaf on X as follows. Consider the sheaf Fz D W ˝ L.x/
on H . Viewing W as a right module over Gx of weight 1 (via the inversion map on
Gx ) and taking the tensor product of this action with the right action of Gx on L.x/, we
obtain a right action of Hx on Fz . Hence, Fz descends to a sheaf F on H=Hx ' X . Now
the .idH .x// L-twisted H -equivariant structure on L.x/ induces an L-twisted H -
equivariant structure on F .
(ii) The cocycle structure on L gives an isomorphism
Lhh0 h1 ;hx ˝ Lh;x ' Lhh0 ;x
for h; h0 2 H . Combining it with the isomorphism
Lh;h0 x ˝ Lh0 ;x ' Lhh0 ;x
we get an isomorphism
Lhh0 h1 ;hx ˝ .Lh;x ˝ L1

h;h0 x / ' Lh0 ;x
for fixed h 2 H , compatible with the cocycle structures in .h0 ; x/. The restriction to
h0 2 Hx gives an isomorphism
Lhh0 h1 ;hx ' Lh0 ;x ; (1.2.14)
compatible with the central extension structures, so we obtain the required isomorphism
(1.2.9).
Now (1.2.10) is given by the isomorphism

W Lh;x ˝ Fx
! Fhx
which is a part of the L-twisted H -equivariant structure on F . To check the compatibil-

ity of with the Gx -action via the isomorphism (1.2.9), let us apply the commutativity
of (1.2.4) for the pairs .h; h0 / and .hh0 h1 ; h/, where h0 2 Hx . We get the following
206 A. Polishchuk
commutative diagram (in which we skipped the tensor product signs for brevity):
1 D a.hh0 h1 ;hx/

Lh;h0 x Fh0 x / Fhh x / F.hh h1 /hx / Lhh h1 ;hx Fhx
0 0 0
a.h0 ;x/

/ Lhh ;x Fx D / L.hh h1 /h;x Fx / Lhh h1 ;hx Lh;x Fx ,
Lh;h0 x Lh0 ;x Fx 0 0 0
where a.; / are the maps (1.2.13) inducing the action of Gx (resp., Ghx ) on Fx (resp.,
Fhx ). The required compatibility follows from the fact that the composition of arrows
in the bottom row corresponds to the isomorphism (1.2.14).
(iii) The functoriality of the construction gives the commutative diagram
fibf .x/
L
CohH .X / / Gf .x/ - rep1
f;M Nf

L
CohH .X / / Gx - rep1 .
fibx
The diagram (1.2.11) follows by applying the isomorphism of functors established in

(ii).
Lemma 1.2.3. In the situation of Proposition 1.2.2 (i) the bundle
V D .x/ .LjH fxg /
on X has a natural L-twisted H -equivariant structure and a commuting right Gx -

action (where Gx acts trivially on X ). Assume in addition that Hx is commutative
and Gx is a Heisenberg extension. Let I Gx be a maximal isotropic subgroup.
Then the bundle V I of I -invariants is a simple object in CohH
L
.X / (i.e., fibx .V I / is a
Schrödinger representation). Also, one has the following isomorphism of sheaves on
H compatible with the right Hx -action:
.x/ V I ' V ˝ LjH fxg ; (1.2.15)

op
where V is a Schrödinger representation of Gx , viewed as a representation of Gx of
weight 1.
Proof. The right Gx -action on V D .x/ .L.x// and the L-twisted H -equivariant
structure are induced by the right Gx -action on L.x/ and the .idH .x// L-twisted
H -equivariant structure, respectively. In the case when Gx is a Heisenberg extension
of Hx and I Gx is a maximal isotropic subgroup, we have an isomorphism
fibx .V I / ' H 0 .Hx ; LjHx fxg /I

compatible with the left Gx -action. But the right-hand side is a standard model for the
Schrödinger representation of G (see Section 1.1). Hence, V I is simple in CohH L
.X /.

Let V be a Schrödinger representation of Gx . Viewing V as a Schrödinger repre-
op
sentation of Gx , we obtain an isomorphism
V ' V ˝ VI ;
compatible with L-twisted H -equivariant structures and the right Gx -actions. Note
also that we have a natural isomorphism H X H ' H Hx (where in the first product
we use the map .x/ W H ! X). This gives an isomorphism
.x/ V D .x/ .x/ .L.x// ' pH m .L.x//;
where m W H Hx ! H is given by the group law and pH W H Hx ! H is the

projection. Taking into account the isomorphism (1.2.12) we get
.x/ V D H 0 .Hx ; L.x/jHx / ˝ L.x/:
Hence, we obtain an isomorphism
V ˝ .x/ V I ' .x/ V ' H 0 .Hx ; L.x/jHx / ˝ L.x/ ' V ˝ V ˝ L.x/

op
compatible with Gx Gx -actions, which leads to (1.2.15).
Example 1.2.4. Let A be an abelian variety, L a line bundle on A (trivialized at zero)
K A a finite subgroup such that L .K/ D 0, where L W A ! Ay is the homomor-
phism associated with A. Consider the biextension ƒ.L/ D m L ˝ p1 L1 ˝ p2 L1
on A A. The restriction ƒ.L/jAK is trivial, hence, ƒ.L/ descends to a biextension
B of A A=K. Now set
L D p1 L ˝ B:
It is easy to see that this line bundle on A A=K has a structure of a 1-cocycle with
respect to the action of A on A=K. Hence, we are in the situation of Proposition 1.2.2
with H D A and X D A=K. Taking x D 0 2 A=K we obtain a central extension
G of K by Gm with the underlying Gm -torsor LjK and the right G-action on .L/,
where W A ! A=K is the projection. One can check that if K D K.L/ D ker.L /
then G ' G.L/op , the opposite of the Mumford group of L (see Section 12.2 of [22]).
This explains the seeming discrepancy between (1.2.15) and [22], Proposition 12.7.
1.3 Index of a symmetric isogeny. Recall that with every line bundle L on an abelian
variety A one associates a symmetric homomorphism
L W A ! AyW a 7! ta L ˝ L1 2 Pic0 .A/:
Furthermore, this induces an isomorphism

yC
NS.A/Q ! Hom.A; A/ (1.3.1)
Q
208 A. Polishchuk
of the Neron–Severi group with the space of symmetric homomorphisms tensored with
Q. In the case when L is an isogeny there exists a unique integer i.L/, 0 i.L/
g D dim A, such that
H i .A; L/ D 0 for i ¤ i.L/; H i.L/ .A; L/ ¤ 0:
It is called the index of L and can be computed as the number of positive roots of the
polynomial P .n/ D .L ˝ Ln0 /, where L0 is an ample line bundle on A (see [16],
III.16). The index has the property
i.Ln / D i.L/ for n > 0:
We say that an element 2 Hom.A; B/Q is an isogeny if it is invertible in AbQ ,

or equivalently if n is a usual isogeny in Hom.A; B/ for some n > 0. Using the
yC
isomorphism (1.3.1), we can define the index of a symmetric isogeny 2 Hom.A; A/Q
by choosing n > 0 such that n D L and setting i./ D i.L/. For any isomorphism
W B ! A in AbQ we have
i. O / D i./:
1.4 Orbi-abelian varieties. We will use a natural extension of duality of abelian

varieties to some length-2 complexes of commutative group schemes. The setup is
similar to the duality of 1-motives considered in [4], [8] and can be generalized to
include them (see [23], Section 4.3).
pr
Let Gk denote the abelian category of commutative proper group schemes over k
(in characteristic zero every such group is a product of an abelian variety and a finite
group). There exists an exact duality functor
pr pr
D W D b .Gk /op ! D b .Gk /
such that D.A/ D Ay for an abelian variety A, D.G/ ' G Œ1 for a finite group scheme
G, and D 2 D Id (see Theorem 4.1.4 of [23]).
pr
An orbi-abelian variety is an object K 2 D b .Gk / with H i K D 0 for i 62 f1; 0g,
such that H 1 K is a finite group scheme. Every orbi-abelian variety can be represented
pr
by a complex of the form ŒG ! X (in degrees 1 and 0), where X 2 Gk and G is
a finite group scheme. With every orbi-abelian variety K we can associate an abelian
variety H 0 .K/0 and two finite groups: H 1 .K/ and 0 H 0 .K/. If K is an orbi-abelian
variety then D.K/ is also such and the corresponding abelian varieties are dual, while
the finite groups H 1 .K/ and 0 H 0 .K/ get interchanged and dualized (see [23],
Lemma 4.2.1).
1.5 Kernels and functors. Recall that there is a principle originating from noncom-
mutative geometry stating that “correct” functors between the derived categories of
coherent sheaves on smooth projective varieties are given by Fourier–Mukai functors

(see [17], [25] for some incarnations of this principle). Such functors have the form
ˆK W D b .X / ! D b .Y / W F 7! Rp2 .p1 F ˝L K/;
where K 2 D b .X Y /, p1 W X Y ! X and p2 W X Y ! Y are the projections.

We will refer to K as a kernel representing the functor ˆK .
The composition of functors corresponds to an operation on kernels that we call
convolution. Namely, for K 2 D b .X Y / and K 0 2 D b .Y Z/ we set
K BY K 0 D Rp13 .p12

K ˝L p23

K 0 / 2 D b .X Z/;
where pij are projections from X Y Z to the products of two of the factors. This
operation is associative and we have a natural isomorphism of functors
ˆKBY K 0 ' ˆK 0 B ˆK
(see e.g. Section 2.1 in [23]).
2 Representations of the Heisenberg groupoid

2.1 Symplectic setting for abelian varieties. This setting was introduced in [20] (see
also [22], 15.2). The basic “symplectic object” we work with is an abelian variety X
equipped with an isomorphism W X ! Xy such that O D .
Definition 2.1.1. We will call .X; / as above a symplectically selfdual abelian variety
(ss-abelian variety for short).
The corresponding analogue of a symplectic form is the skew-symmetric biexten-
sion E D .id / PX of X X obtained by the pull-back from the Poincaré line bundle
PX on X Xy (see [20]). Often we need to assume that there exists a biextension B
on X X such that E is obtained from B by antisymmetrization: E ' B ˝
B 1 ,
where
W X ! X is the permutation of factors.
Definition 2.1.2. The data .X; ; B/ as above is called an enhanced symplectically
selfdual abelian variety (or ess-abelian variety for short).
The main example of an ess-abelian variety is XA D A A,y where A is an abelian
y y
variety, and B D p14 P on X X D A A A A (here P D PA is the Poincaré line
y We will refer to this example as the standard ess-abelian variety
bundle on A A).
associated with A.
We consider ess-abelian varieties as analogs of symplectic vector spaces. In [20]
we introduced a natural analog of the Heisenberg group in this context. First, let us
consider the case XA D A A. y In this case for every point x D .a; / 2 XA .k/ we
have a functor
TxA D .˝P / B ta W D b .A/ ! D b .A/; (2.1.1)
210 A. Polishchuk
where ta W A ! A is the translation by a 2 A.k/ and P D P jAfg 2 Pic0 .A/ is the

y
line bundle on A corresponding to 2 A.k/. These functors compose according to the
canonical isomorphisms
TxA B TxA0 ' Bx;x 0 ˝ TxCx
A
0:
To express the algebraic dependence of these functors on x we can consider similar

functors on D b .A S/ depending on x 2 XA .S /, where S is a k-scheme, and observe
that the above isomorphisms still hold (and are compatible with pull-backs with respect
to base changes S 0 ! S). This motivates the following definition (see Definition 2.1
in [20]).
Definition 2.1.3. Let .X; B/ be an ess-abelian variety. The Heisenberg groupoid
H D H.X; B/ associated with .X; B/ is the stack BGm X equipped with the
following structure of a stack of Picard groupoids: for a k-scheme S define the functor
C W H.S/ H.S / ! H.S / by
.L; x/ C .L0 ; x 0 / D .L ˝ L0 ˝ Bx;x 0 ; x C x 0 /;
where L and L0 are line bundles over S and x; x 0 2 X.S /.
We intend to consider actions of H on fibered categories by generalizing the example
of the action of H.XA / on the fibered category S 7! D b .A S / given by the functors
TxA above (such that L 2 BGm .S / acts by tensoring with the pull-back of L to A S ).
2.2 Isotropic and Lagrangian pairs. Similarly to the case of the classical Heisenberg
group associated with a symplectic vector space we need Lagrangian subvarieties Y
in X to define representations of H. Below we generalize the setup of [20], Section 2,
slightly in that we allow the map Y ! X to have finite kernel.
Definition 2.2.1. A generalized isotropic pair .Y; ˛/ consists of a commutative group
scheme Y equipped with a homomorphism i W Y ! X , such that ker.i / is finite, and
of a line bundle ˛ over Y together with an isomorphism
ƒ.˛/ ' .i i / B (2.2.1)
of biextensions on Y Y inducing a cube structure on ˛. In the case when ker.i / is
trivial we say that .Y; ˛/ is an isotropic pair (see [20]).
The reason for this definition becomes clear in connection with the Heisenberg
groupoid. Namely, the data .Y; ˛/ as above determines a homofunctor of stacks of
Picard groupoids,
Y ! H W y 7! .˛y ; i.y//:
Note that if .Y; ˛/ is a generalized isotropic pair then the biextension .i i / E is
trivialized. Therefore, we can view .i id/ E as a biextension of Y ŒY ! X . By
Proposition 4.3.2 of [23], this biextension gives a morphism
Y ! DŒY ! X (2.2.2)
that fits into a morphism of exact triangles in the derived category D b .G pr /:
Y
i /X / ŒY ! X / Y Œ1
D.i/
DŒY ! X / Xy / D.Y / / DŒY ! X Œ1.
Conversely, if i W Y ! X is such that the composition

i D.i/
Y ! X ! Xy ! D.Y / (2.2.3)
is zero then Y extends to a generalized isotropic pair.

Definition 2.2.2. A generalized Lagrangian pair .Y; a/ is a generalized isotropic pair
such that (2.2.2) is an isomorphism in D b .G pr /. If ker.i / D 0 and Y is connected then
we say that .Y; ˛/ is a Lagrangian pair and Y is Lagrangian.
Lemma 2.2.3. (i) If .Y; ˛/ is a generalized isotropic pair then dim Y dim X=2. If
.Y; ˛/ is generalized Lagrangian then dim Y D dim X=2.
(ii) Let Y X be an isotropic abelian subvariety such that dim Y D dim X=2.
Then Y is Lagrangian.
(iii) Let .Y; ˛/ be a generalized Lagrangian pair. Then Y is connected if and only
if ker.i/ D 0 if and only if Y is Lagrangian.
Proof. (i) The first assertion follows from the fact that the composition
i 0 i O0
Y0 ! X ! Xy ! Y0
is zero and that i 0 D i jY0 has finite kernel. If .Y; ˛/ is generalized Lagrangian then we
obtain from the isomorphism (2.2.2) the equality
dim Y D dim H 0 DŒY ! X D dim H 0 ŒY ! X D dim X dim Y:
1
(ii) Since .Y / and X =Y are both abelian subvarieties in Xy , it follows that .Y / D
1
X =Y if and only if dim Y D dim X=Y .
(iii) This follows from the duality between the finite groups H 1 ŒY ! X D ker.i /
and 0 H 0 DŒY ! X ' 0 .Y / (see [23], Lemma 4.2.1).
Part (ii) of the above lemma shows that to determine whether an abelian subvariety
of X is Lagrangian we can work in the category of abelian varieties up to isogeny.
Examples 2.2.4. 1. Consider the standard ess-abelian variety XA D A A. y Then
an abelian subvariety Y XA , finite and surjective over A, gives rise to a morphism
212 A. Polishchuk
from A to Ay in the category AbQ of abelian varieties up to an isogeny, namely, the

morphism fY D pAy B pA1 , where pA W Y ! A and pAy W Y ! A are projections.
It is easy to see that the isotropy condition on Y is equivalent to the condition that
fY is symmetric, i.e., fOY D fY . By Lemma 2.2.3 (ii), this is equivalent to Y being
Lagrangian. Conversely, with every symmetric morphism f 2 Hom.A ! A/ y Q we
can associate its graph (which will be Lagrangian)
y
.f / WD f.Na; Nf .a// j a 2 Ag A A;
where N > 0 is an integer such that Nf 2 Hom.A; A/. y In this way we get a one-to-
one-correspondence between the set of Lagrangian subvarieties in XA , finite over A,
and the subspace Hom.A; A/ y C Hom.A; A/ y Q of symmetric homomorphisms.
Q
2. If Y X is a Lagrangian abelian subvariety then we can always choose a line
bundle ˛ such that .L; ˛/ is a Lagrangian pair. Indeed, the biextension BjY Y of
Y Y is symmetric, hence, it is of the form ƒ.˛/ for some line bundle ˛ on Y (see e.g.
Theorem 13.7 of [22]).
3. Let X D 0. Then a generalized Lagrangian .Y; ˛/ is a finite commutative group
scheme Y together with a central extension of Y by Gm (given by ˛), such that the
corresponding commutator pairing Y Y ! Gm induces an isomorphism Y ! Y .
In other words, this gives a finite Heisenberg group scheme.
Lemma 2.2.5. If .Y; ˛/ is a generalized isotropic pair such that dim Y D dim X=2
then i.Y0 / X is a Lagrangian abelian subvariety, where Y0 Y is the connected
component of 0 in Y .
y
Proof. By definition, the composition Y ! X
! X ! D.Y / is zero. Hence, the
same is true for the composition
! Xy ! Yb0 :

Y0 ! X
1
Since the natural morphisms Y0 ! i.Y0 / and i .Y0 / ! Yb0 are isogenies, we derive
1
that the composition

i.Y0 / ! X ! Xy ! i .Y0 /
is zero. By Lemma 2.2.3 (ii), this implies that i.Y0 / is Lagrangian.
Definition 2.2.6. Let us say that generalized Lagrangian pairs .Y; ˛/ and .Z; ˇ/ are
transversal if the images of Y and Z generate X , or equivalently if Y X Z is finite.
Lemma 2.2.7. (i) For any pair of Lagrangian subvarieties Y and Z in X there exists
a Lagrangian subvariety Q X transversal to both Y and Z.
(ii) Now assume in addition that X D XA (equipped with the standard symplectic
biextension) and Y D f0g A. y Let L be an ample line bundle on A. y Then for
almost all n 2 Z (i.e., for all except for a finite number) the Lagrangian subvariety
.nL / Ay A D XA (see Example 2.2.4.1) is transversal to Z.
Proof. (i) To prove the first assertion we can argue in the semisimple category AbQ of
abelian varieties up to isogeny (since we have natural analogs of the relevant notions
in AbQ ). In particular, in this category the condition of transversality of Q to Y and Z
becomes simply Q \ Y D Q \ Z D 0. We can assume that X ' Y ˚ Yy , where both
Y and Yy are Lagrangian in X and the self-duality of X restricts to the duality between
Y and Yy (see [20], Lemma 1.1). Let T D Y \ Z. We can write
Y D T ˚ P; X D T ˚ P ˚ Ty ˚ Py ; (2.2.4)
so that
Z D f.t; ./; 0; / j t 2 T; 2 Py g; (2.2.5)
where W Py ! P is a symmetric morphism. Then choosing a symmetric isomorphism
W Py ! P in AbQ we can set
Q D f.0; . C /./; ; / j 2 Ty ; 2 Py g:
y so
(ii) As in part (i) we consider the decompositions (2.2.4) in AbQ (with Y D A)
y
that (2.2.5) holds (still in AbQ ). Let us write the homomorphism L W Y ! Y in the
form
˛ ˇ
L D ;
ı
with ˛ 2 Hom.T; Ty /Q , ˇ 2 Hom.P; Ty /Q , 2 Hom.T; Py /Q and ı 2 Hom.P; Py /Q .
It is easy to see that the condition of transversality of .nL / with Z is equivalent to
the condition that nı is an isogeny (i.e., invertible in AbQ ). Note that some multiple
of ı is the symmetric homomorphism P ! Py corresponding to the restriction of L to
P . Since this restriction is still ample, it follows that ı is an isogeny. Thus, deg.nı /
is a polynomial in n with nonzero top degree coefficient deg.ı/, so deg.nı / ¤ 0
for almost all n 2 Z.
The construction of the following proposition is a straightforward generalization
of Lemma 15.4 of [22] (see also [20], Section 3). Recall that given a biextension L
of Y Z by Gm and a pair of homomorphisms A ! Y , B ! Z, together with
trivializations of the pull-backs of L to A Z and Y B, we obtain a pairing
A B ! Gm
measuring the difference between two induced trivializations of the pull-back of L to
A B (cf. [22], 10.4).
Proposition 2.2.8. Let .i W Y ! X; ˛/ and .j W Z ! X; ˇ/ be generalized Lagrangian
pairs such that Y X Z is finite. Then there is a natural central extension G of
K D Y X Z by Gm with the underlying line bundle ˇz ˝ ˛y1 over K. Moreover, G
is a Heisenberg group scheme and the corresponding commutator form K K ! Gm
is the pairing associated with the biextension .i j / E of Y Z and with natural
trivializations of its pull-backs to K Z and Y K.
214 A. Polishchuk
Proof. Pulling back the isomorphisms

ƒ.˛/ ' .i i / B; ƒ.ˇ/ ' .j j / B (2.2.6)
(see (2.2.1)) to K K we obtain an isomorphism ƒ.ˇz ˝ ˛y1 / ' OKK that gives
a central extension structure on the corresponding Gm -torsor G ! K. The isomor-
phisms (2.2.6) also induce trivializations of the pull-backs of E to Y Y and to Z Z.
Hence, we have natural trivializations of the pull-backs of E to K Z and Y K
whose difference gives a pairing
e W K K ! Gm :
Note that E induces a duality between Y and ŒY ! X ' ŒK ! Z and between Z
and ŒZ ! X ' ŒK ! Y . Thus, the dual of the exact triangle
K ! Z ! ŒK ! Z !
is the exact triangle
Y ! ŒK ! Y ! KŒ1 !
and we obtain an isomorphism K ' K . It is easy to see that it is given by the
pairing e. Furthermore, the same reasoning as in Lemma 15.4 of [22] shows that e is
the commutator form of the central extension G ! K.
2.3 Representations associated with Lagrangian pairs and intertwining functors.

Let .Y; ˛/ be an isotropic pair. Then the data .Y ! X; BjY X ; ˛/ is similar to the
twisting data considered in [23], Section 3.7, with the only difference that Y is not
a finite group scheme. In particular, we still have the 1-cocycle of Y with values in
Pic.X/ defined by
L D p1 ˛ 1 ˝ Bj1
Y X : (2.3.1)
Recall that in Section 2 of [20] we defined the category F .Y; ˛/ as the category of
L-twisted Y -equivariant objects in D b .X / (see Section 1.2). Explicitly, the objects of
F .Y; ˛/ are objects F 2 D b .X / equipped with isomorphisms
FxCy ' ˛y1 ˝ By;x
1
˝ Fx
on X Y , satisfying the natural cocycle condition on X Y Y .
For every k-scheme S we can define similarly the category F .Y; ˛/S as the category
of L-twisted Y -equivariant objects in D b .X S /, so that S 7! F .Y; ˛/S is a fibered
category with respect to the natural pull-back functors. The Heisenberg groupoid H
naturally acts on this fibered category. Namely, for every S we have an action of H.S /
on F .Y; ˛/S given by the functors
TL;x .F /x 0 D L ˝ Bx 0 ;x ˝ FxCx 0 ;
and these actions are compatible with the pull-backs with respect to morphisms S ! S 0 .
We also set Tx D TOS ;x for x 2 X.S /.
Remarks 2.3.1. 1. In the case when the projection X ! X=Y has a section s W X=Y !
X there is a natural equivalence F .Y; ˛/ ' D b .X=Y / obtained by restricting F 2
F .Y; ˛/ to s.X=Y / X . Furthermore, the functors TL;x are given by some kernels
on X=Y X=Y , so we get a kernel representation in the sense of [23], Section 3.2
(see also Example 2.3.2 below).
2. The above definition of F .Y; ˛/ is not quite adequate in general. For example,
it is not clear whether F .Y; ˛/ has a triangulated structure such that the embedding
F .Y; ˛/ ! D b .X / is an exact functor. If char.k/ D 0 then F .Y; ˛/ can be identified
with the (bounded) derived category of modules over some Azumaya algebra over X=Y
(see [20], Proposition 3.1). This identification uses the derived descent with respect to
finite flat morphisms (see Appendix in [20]1 , [7]). A more adequate replacement for
F .Y; ˛/ is the derived category of the abelian category F .Y; ˛/\Coh.X / D CohL Y .X /
(cf. Remark 2.3.5 below). However, we do not need this since we will mostly work
with objects of F .Y; ˛/ that are cohomologically pure.
Example 2.3.2. Consider the standard ess-abelian variety XA D A Ay with B D

p14 P and take Y D f0g Ay XA , ˛ D O. Then using the section A f0g XA
of the projection XA ! A we obtain an equivalence of the category F .Y; ˛/ with
D b .A/. One immediately checks that under this equivalence the functors Tx giving
the action of H.k/ on F .Y; ˛/ become precisely the functors TxA defined by (2.1.1).
We will denote by TxA 2 D b .A A/ the kernel giving the functor TxA . Explicitly, for
x D .a; / 2 A A,y
TxA D .ta ; idA / P : (2.3.2)
Remark 2.3.3. Given an object F 2 D b .A/ we can consider the subset YF of x 2 XA

such that TxA .F / ' F . Assuming that F is endosimple and Exti .F; F / D 0 for i < 0,
one can check that YF is an algebraic subgroup in XA and that we have a line bundle
˛ on YF and an isomorphism
TyA .F / ' ˛y ˝ F
over YF A (this can be done using results of [9]; the case of vector bundles is
considered in [14], Section 1). This easily implies that .YF ; ˛ 1 / is an isotropic pair.
In particular, dim YF dim A.
As in the classical case of the Heisenberg group, one expects that the representations
F .Y; ˛/ associated with different Lagrangian pairs .Y; ˛/ are equivalent. This can be
proved under some additional assumptions (due to the need to use the derived descent,
see Remark 2.3.1.2).
Theorem 2.3.4. Let .Y; ˛/ and .Z; ˇ/ be Lagrangian pairs for an ess-abelian variety
X. Assume that either char.k/ D 0 or Y \ Z D 0. Then there is an equivalence of
fibered categories F .Y; ˛/S ' F .Z; ˇ/S compatible with the H-action.
1
The assumption char.k/ D 0 was erroneously omitted in [20].
216 A. Polishchuk
Proof. In both cases the proof of Theorem 4.3 in [20] works. The assumption char.k/ D
0 allows one to use the derived descent, while in the case Y \Z D 0 the descent problem
does not arise.
Remark 2.3.5. In this paper we use only categories F .Y; ˛/ associated with La-
grangian pairs. However, the above equivalence can also be constructed for generalized
Lagrangian pairs and the extra assumptions in Theorem 2.3.4 can be removed if we
replace each category F .Y; ˛/ with the derived category of F .Y; ˛/ \ Coh.X / and
use Theorem 4.5.1 of [23].
2.4 Invariants of a generalized Lagrangian pair. Consider the representation of H

on the fibered category S 7! F .Y; ˛/S , associated with a Lagrangian pair .Y; ˛/. Now
let .j W Z ! X; ˇ/ be a generalized isotropic pair. Then we have a homofunctor
j W Z ! H W z 7! .ˇz ; j.z//
of Picard stacks, so it makes sense to consider the category F .Y; ˛/.Z;ˇ / of .Z; ˇ/-
invariants in F .Y; ˛/. By definition, the objects of this category are objects F 2
F .Y; ˛/ equipped with a collection of isomorphisms
.z/
Tj.z/ .FS / ! FS ;
for all k-schemes S and all z 2 Z.S /, where FS 2 F .Y; ˛/S is the pull-back of F ,
compatible with pull-backs with respect to base changes S ! S 0 and satisfying the
following cocycle condition:
Tj.z1 / .z2 /
Tj.z1 /Cj.z2 / .FS / / Tj.z / .FS /
1
.z1 /
.z1 Cz2 /
Tj.z1 Cz2 / .FS / / FS .
One can easily check that the collection ..z// is determined by the single element
.z un /, where z un 2 Z.Z/ is the tautological Z-point of Z. Thus, F .Y; ˛/.Z;ˇ / can
be also described as the category of objects F 2 F .Y; ˛/ equipped with isomorphisms
Fj.z/Cx ' ˇz1 ˝ Bx;j.z/
1
˝ Fx ;
over Z X, where z 2 Z, x 2 X , satisfying the natural cocycle condition over
Z Z X and the following compatibility for Fj.z/CxCy , where y 2 Y , z 2 Z:
Fj.z/CxCy / ˇz1 ˝ B 1 ˝ FxCy

xCy;j.z/

1
˛y1 ˝ By;j.z/Cx ˝ Fj.z/Cx / ˛y1 ˝ ˇz1 ˝ B 1 1
˝ By;x 1
˝ Bx;j.z/ ˝ Fx ,
y;j.z/
where the horizontal (resp., vertical) isomorphisms correspond to .Z; ˇ/-invariance

(resp., come from the definition of F .Y; ˛/).
Example 2.4.1. As in Example 2.3.2 let us consider the standard ess-abelian variety
X D XA and Y D f0g Ay XA , so that F .Y; O/ ' D b .A/. Then the skyscraper
sheaf k.0/ 2 D b .A/ is invariant with respect to .Y; O/. Now let F be a nontrivial
extension of k.0/ by itself, so F is isomorphic to the structure sheaf of a length 2
subscheme S A. Then the fact that the restriction of the Poincaré bundle to S Ay
is nontrivial implies that F is not .Y; O/-invariant, even though F ˝ P ' F for
y
every 2 A./. The Fourier dual of this example is a bundle V which is a nontrivial
extension of OA by itself. The bundle V is homogeneous, i.e., ta V ' V for every
a 2 A. However, it is not invariant with respect to A f0g XA (and ˇ D O)
according to our definition.
By Theorem 2.3.4, if .Y 0 ; ˛ 0 / is another Lagrangian pair then the categories
F .Y; ˛/.Z;ˇ / and F .Y 0 ; ˛ 0 /.Z;ˇ / are equivalent, provided either char.k/ D 0 or
Y \ Y 0 D 0. This observation will often allow us to reduce the study of F .Y; ˛/.Z;ˇ /
to the case when Y and Z are transversal.
Lemma 2.4.2. Let ˇ 0 D ˇ˝EjZfxg for some x 2 X . Then the functor Tx W F .Y; ˛/ !
F .Y; ˛/ induces an equivalence
0
t .Z; x/ W F .Y; ˛/.Z;ˇ / ! F .Y; ˛/.Z;ˇ / :
Proof. Let j0 W Z ! H W z 7! .ˇz0 ; j.z// be the homofunctor associated with ˇ 0 . Then

we have
Tj0 .z/ D Ej.z/;x ˝ Tj.z/ :
Hence, the structure of .Z; ˇ 0 /-invariance for F 2 F .Y; ˛/ induces isomorphisms
Tj.z/ Tx .F / ' Ej.z/;x ˝ Tx Tj.z/ .F / ' Tx Tj0 .z/ .F / ! Tx .F /;
i.e., the structure of .Z; ˇ/-invariance for Tx .F /.

We are going to show that under some technical assumptions the category
F .Y; ˛/.Z;ˇ / is generated by a single coherent sheaf in the sense of the following
definition.
Definition 2.4.3. Let T be a triangulated category, F T an additive (not necessarily
full) subcategory closed under shifts X 7! X Œi , i 2 Z. For objects F; F1 ; F2 2 F
we say that F is an F -extension of F2 by F1 if there exists a triangle
F1 ! F ! F2 ! F1 Œ1
in F which is an exact triangle in T . We say that F 2 F has an F -filtration with

successive quotients Q1 ; : : : ; Qn 2 F if there exists a collection of objects .Fij /,
0 i < j n, in F , such that Fi1;i D Qi for i D 1; : : : ; n, F0n D F , and for any
218 A. Polishchuk
i < j < k, Fik is an F -extension of Fj k by Fij . We say that an object F0 2 F is a t -

generator of F with respect to T if HomT .F0 ; F0 Œi / D 0 for i < 0, and every nonzero
object of F admits an F -filtration with successive quotients F0˚m1 Œk1 ; : : : ; F0˚mn Œkn
for some n > 0, where k1 > k2 > > kn .
Note that the above definition of a t-generator mimics the situation when F has a t -
structure with the heart consisting of direct sums of F0 , however, it uses the triangulated
structure of an ambient category T (since F does not have to be triangulated).
Lemma 2.4.4. Let F0 ¤ 0 be a t -generator of F with respect to T .

(i) Suppose A 2 F (resp., B 2 F ) has an F -filtration with successive quotients
˚m01 0
˚mp
F0˚m1 Œa1 ; : : : ; F0˚mn Œan (resp., F0 Œb1 ; : : : ; F0 Œbp )
such that a1 > > an (resp., b1 > > bp ), where mi ¤ 0, mj0 ¤ 0. Then
HomF .A; BŒan b1 / ¤ 0:
(ii) Let F 2 F be an object such that HomF .F; F Œi / D 0 for i < 0. Then
F ' F0˚m Œa in F .
(iii) Suppose F0 is an F -extension of B 2 F by A 2 F , where HomT .A; BŒi / D 0
for i 0. Then either A D 0 or B D 0.
Proof. (i) By definition, we have an exact triangle

A0 ! A ! F0˚mn Œan ! A0 Œ1
where A0 has an F -filtration with successive quotients of the form F0˚m Œi such that
i > an . Hence, HomT .A0 Œ1; F0 Œan / D 0 and the map
HomF .F0˚mn Œan ; F0 Œan / ! HomF .A; F0 Œan /
induced by is injective. In particular, HomF .A; F0 Œan / ¤ 0. Next, let us consider

an exact triangle
˚m01
F0 Œan ! BŒan b1 ! B 0 !
where B 0 has an F -filtration with successive quotients of the form F0˚m Œi , such that
i < an . It follows that HomT .A; B 0 Œ1/ D 0, so the map
˚m01
HomF .A; F0 Œan / ! HomF .A; BŒan b1 /
induced by is injective. Since the source space is non-zero, this implies the result.
(ii) We know that F has an F -filtration with successive quotients F0˚m1 Œk1 ; : : : ;
F0˚mn Œkn such that k1 > > kn . Applying (i) to A D B D F we deduce that
kn k1 , so in fact, n D 1.
(iii) Assume that A and B have F -filtrations as in (i). Then the result of (i) together
with our vanishing assumption implies that b1 < an . On the other hand, in order
to have a nonzero morphism A ! F0 (resp., F0 ! B) we should have an 0
(resp., b1 0). Thus, either HomT .A; F0 / D 0 or HomT .F0 ; B/ D 0. Assume that
HomT .A; F0 / D 0 (the second case is considered similarly). Then in T one has an
isomorphism B ' F0 ˚ AŒ1, so we obtain that
HomT .A; BŒ1/ D HomT .A; A/ ¤ 0;
a contradiction.
Theorem 2.4.5. Let .Y; ˛/ be a Lagrangian pair for an ess-abelian variety X, and let
.j W Z ! X; ˇ/ be a generalized Lagrangian pair for X . Assume that the following
condition is satisfied:
.?/ either Y and Z are transversal, or X D XA and Y D f0g A, y or char.k/ D 0.
Then there exists a coherent sheaf
SZ;ˇ D SZ;ˇ .Y; ˛/ 2 F .Y; ˛/.Z;ˇ / \ Coh.X /
which is a t -generator of F .Y; ˛/.Z;ˇ / with respect to F .Y; ˛/. Furthermore, SZ;ˇ is
an endosimple object of F .Y; ˛/.Z;ˇ / . Every coherent sheaf in F .Y; ˛/.Z;ˇ / \Coh.X /
is a direct sum of several copies of SZ;ˇ . In the case when Y and Z are transversal
we have
SZ;ˇ .Y; ˛/ D .sY;Z / .˛ 1 ˇ 1 ˝ .i j / B 1 /I ; (2.4.1)
where i W Y ! X is the embedding map, sY;Z W Y Z ! X is the natural map and I is
a maximal isotropic subgroup in ker.sY;Z / ' Y X Z with respect to the commutator
pairing from Proposition 2.2.8. We also have
dimk EndF .Y;˛/ .SZ;ˇ / D j0 .Z/j: (2.4.2)
Proof. Assume first that Y and Z are transversal and let us show the existence of an
endosimple t -generator in this case. Recall that an object F 2 F .Y; ˛/ is equipped
with an isomorphism
FxCy ' ˛y1 ˝ By;x 1
˝ Fx ;
where y 2 Y , x 2 X . On the other hand, the condition of .Z; ˇ/-invariance is an
isomorphism
Fj.z/Cx ' ˇz1 ˝ Bx;j.z/
1
˝ Fx ;
where z 2 Z, x 2 X . Combining these two isomorphisms we get an isomorphism
Fj.z/CxCy ' L.y;z/;x ˝ Fx ; (2.4.3)
over .Y Z/ X , where
L.y;z/;x D ˛y1 ˝ ˇz1 ˝ Bx;j.z/

1 1
˝ By;j.z/ 1
˝ By;x : (2.4.4)
220 A. Polishchuk
has a natural structure of a 1-cocycle for the action of Y Z on X induced by the

homomorphism sY;Z W Y Z ! X . Assume in addition that F is a coherent sheaf.
Then the compatibilities in the definition of F .Y; ˛/.Z;ˇ / reduce to the condition that
(2.4.3) gives F a structure of an L-twisted Y Z-equivariant coherent sheaf on X . By
Proposition 1.2.2, the category of such sheaves is equivalent to the category of weight
one representations of the central extension G of K D ker.sY;Z / by Gm given by the
restriction LjKf0g . We have an identification
W Y X Z ! K W .y; z/ 7! .y; z/
such that
.LjKf0g /y;z ' ˛y1 ˝ ˇz
1 1
˝ Bj.z/;j.z/ ' ˛y1 ˝ ˇj.z/ ;
where the last isomorphism comes from the isomorphism ƒ.ˇ/ ' .j j / B. Hence,
G can be identified with the Heisenberg group of Proposition 2.2.8. Let us define
SZ;ˇ 2 F .Y; ˛/.Z;ˇ / \Coh.X / as the sheaf corresponding to the Schrödinger represen-
tation of G. Note that it is an endosimple object of F .Y; ˛/.Z;ˇ / . The formula (2.4.1)
follows immediately from Lemma 1.2.3. For any object F 2 F .Y; ˛/.Z;ˇ / D b .X /
the triangles
n1 F ! F ! n F ! n1 F Œ1 (2.4.5)
(where are the truncation functors associated with the standard t -structure) automat-
ically belong to F .Y; ˛/.Z;ˇ / . Hence, the standard devissage of F into its cohomology
sheaves shows that SZ;ˇ is a t -generator of F .Y; ˛/.Z;ˇ / .
The existence of an endosimple t -generator of F .Y; ˛/.Z;ˇ / with respect to F .Y; ˛/
in the case when Y and Z are not necessarily transversal follows now from Theo-
rem 2.3.4. More precisely, if char.k/ D 0 then we can choose a Lagrangian pair
.Y 0 ; ˛ 0 /, transversal to Z (see Lemma 2.2.7 (i)) and use an equivalence of fibered cat-
egories F .Y; ˛/S ' F .Y 0 ; ˛ 0 /S compatible with the H-action. In the case when
Y D f0g Ay XA D X we can assume that Y 0 \ Y D 0 (see Lemma 2.2.7 (ii)), so
we again have an equivalence F .Y; ˛/S ' F .Y 0 ; ˛ 0 /S .
Let us prove that such a t -generator F0 of F .Y; ˛/.Z;ˇ / with respect to F .Y; ˛/
is necessarily cohomologically pure. Suppose F0 is not pure and consider the exact
triangle (2.4.5) with A D n1 F0 ¤ 0 and B D n F0 ¤ 0. Then Hom.A; BŒi / D
0 for i 0, hence, the same space of morphisms in F .Y; ˛/ also vanishes. But this
gives a contradiction with Lemma 2.4.4 (iii).
Thus, F0 can be shifted to become a coherent sheaf, and we set SZ;ˇ D F0 . The
fact that every F 2 F .Y; ˛/.Z;ˇ / \ Coh.X / is a direct sum of several copies of SZ;ˇ
follows from Lemma 2.4.4 (ii).
To prove (2.4.2) we can again assume that Y and Z are transversal. Let V be a
Schrödinger representation of the Heisenberg group scheme G considered above. By
Lemma 1.2.3, endomorphisms of F D SZ;ˇ can be identified with K-invariants in
endomorphisms of

sY;Z F ' V ˝ LjY Zf0g ;
i.e., with K-invariants in Endk .V / ˝ H 0 .Y Z; O/. But Endk .V / is isomorphic to

the space of functions on K (see (1.1.2)), so
End.SZ;ˇ / ' H 0 .K Y Z; O/K ' H 0 .K 0 .Z/; O/K :
The dimension of the latter space is j0 .Z/j.
Example 2.4.6. Recall that in the case when X is the standard ess-abelian variety
XA D A Ay and Y D 0 Ay we have an equivalence F .Y; O/ ' D b .A/. Consider the
Lagrangian subvariety Z D L A Ay associated with a symmetric homomorphism
2 Hom.A; A/ y C (see Example 2.2.4.1) and extend it in some way to a Lagrangian
Q
pair .Z; ˇ/. Then by Theorem 2.4.5, the sheaf E D SZ;ˇ 2 Coh.A/ is an endosimple
vector bundle. Furthermore, this bundle is semihomogeneous in the sense of [14],
and corresponds to the element det.E/= rk.E/ 2 NS.A/Q under the isomorphism
(1.3.1) (see [14], Proposition 7.7).
In the next proposition we study the relation between invariants of a generalized

Lagrangian j W Z ! X in F .Y; ˛/ and invariants of the corresponding Lagrangian
j.Z0 / X.
Proposition 2.4.7. Let .Y; ˛/ be a Lagrangian pair.

(i) Let .Z; N be a Lagrangian pair, transversal to Y , and let j W Z0 ! Z
x ˇ/ x be a
surjective homomorphism from an abelian variety Z0 with finite kernel C , so that we
N Consider the subgroup
have the generalized isotropic pair .Z0 ; ˇ/, where ˇ D j ˇ.

Z 0 D ker.X ! Xy ! Z
0 / X:
Then we have a natural duality
x
W Z 0 =Z
! C :
and an equivalence

ˆ W F .Y; ˛/.Z0 ;ˇ / \ Coh.X /
! C - rep;
such that for F 2 F .Y; ˛/.Z0 ;ˇ / and z 0 2 Z 0 one has
x ˝ ˆ.F /:
ˆ.Tz 0 F / ' .z 0 mod Z/ (2.4.6)
Furthermore, ˆ.SZ; x ˇN / is isomorphic to the trivial one-dimensional representation

of C .
(ii) Let .j W Z ! X; ˇ/ be a generalized Lagrangian pair and let Z0 be the con-
nected component of zero in Z. Assume that the condition .?/ from Theorem 2.4.5
is satisfied. Consider the finite group scheme … D 0 .j.Z// D j.Z/=j.Z0 / as a
222 A. Polishchuk
subgroup of X=j.Z0 /. Assume first that … is reduced. Then there exists a …-coset
…0 X=j.Z0 / such that
M ˚NZ
SZ;ˇ ' TxN Sj.Z0 /;ˇN (2.4.7)
N
x2…0
in F .Y; ˛/, where ˇN is chosen in such a way that .j.Z0 /; ˇ/

N is a Lagrangian pair, and
j0 .Z/j1=2
NZ D
j0 .j.Z//j1=2
(NZ is always an integer). If … is not reduced then there is still a …-coset …0

X=j.Z0 / such that the composition factors of SZ;ˇ in F .Y; ˛/.Z0 ;ˇ / \ Coh.X / (which
is a finite length abelian category) are .TxN Sj.Z0 /;ˇN /, xN 2 …0 , each with multiplicity
j…j
NZ #….k/
.
Proof. (i) As in the proof of Theorem 2.4.5 the line bundle L on .Y Z0 / X given
by (2.4.4) has a structure of a 1-cocycle for the (transitive) action of Y Z0 on X .
Furthermore, this 1-cocycle is the pull-back of a similar cocycle L x on .Y Z/
x X.
Therefore, the central extension G of K D ker.sY;Z0 / by Gm is the pull-back of
the similar central extension Gx of Kx D ker.s x / under the natural homomorphism
Y;Z
x
K ! K. Note that the exact sequence
x !0
0 ! C ! Z0 ! Z (2.4.8)
gives rise to an exact sequence
0 ! C ! K ! Kx ! 0
Thus, we have an exact sequence of groups

x ! 1;
1!C !G!G
x is a finite Heisenberg group. Thus, as in the
where C is a central subgroup in G and G
proof of Theorem 2.4.5 we obtain an equivalence
F .Y; ˛/.Z0 ;ˇ / \ Coh.X / ' G- rep1 :

x is a Heisenberg group, the desired equivalence ˆ follows from Lemma 1.2.1.
Since G
Note that the natural functor
x N
F .Y; ˛/.Z;ˇ / ! F .Y; ˛/.Z0 ;ˇ /
corresponds to the restriction under the homomorphism G ! G. x This implies that

ˆ.SZ;x ˇN / corresponds to the trivial one-dimensional representation of C .
Now considering the dual of the exact sequence (2.4.8) we get (using the fact that
x is Lagrangian) the exact sequence
Z
x !Z
C ! X=Z 0 !
which gives an isomorphism of Z 0 =Z x C !

x with C . Thus, the pairing W Z 0 =Z
Gm is obtained as the canonical pairing associated with the biextension .id j / EN of
X=Zx Z0 , where EN is the biextension of X=Zx Z x induced by E. Explicitly, let
t .C / W OX=ZC
x
N x
! .id j / EjX=ZC
and
x WO 0 x N
t .Z 0 =Z/ Z =ZZ0 ! .id j / EjZ 0 =ZZ
x 0
be natural trivializations of restricted biextensions. Then
t .C /jZ 0 =ZC
x
x 0 x :
D t .Z 0 =Z/jZ =ZC
The subgroup Z 0 X acts on X by translations, and we have an isomorphism of

1-cocycles of Y Z0 with values in Pic.X /
L.y;z0 /;xCz 0 ' L.y;z0 /;x ˝ Bz1 1

0 ;j.z / ˝ By;z 0 :
0
x we obtain an isomor-
Using the trivialization of .id j / EjZ 0 Zx induced by t .Z 0 =Z/
phism
1 W L.y;z0 /;xCz 0
1
! L.y;z0 /;x ˝ Bj.z0 /;z
1
0 ˝ By;z 0
1
(2.4.9)

! L.y;z0 /;x ˝ ByCj.z 0 /;z
0:
Thus, for fixed z 0 2 Z 0 we are in the situation of Proposition 1.2.2 (iii) with H D
Y Z0 , f W X ! X the translation by z 0 , M D BjXfz 0 g , the isomorphism f D
given by (2.4.9) and the fixed point x D 0. Note that the functor Tz 0 on F .Y; ˛/ sends
F to tz0 F ˝ M , which is exactly the functor f;M considered in Proposition 1.2.2 (iii).
Let Gz 0 be the central extension of K by Gm with the underlying line bundle LjKfz 0 g .
The diagram (1.2.11) gives in our case an isomorphism of G-representations
fib0 .Tz 0 F / ' ˛ fib0 .F /
for F 2 F .Y; ˛/.Z0 ;ˇ / , where the automorphism ˛ D 1 B N W G ! G (identical

on Gm G) is the composition of isomorphisms of central extensions N W G ! Gz 0
and 1 W Gz 0 ! G defined as follows. The isomorphism N W G ! Gz 0 is obtained by
specializing (2.4.9) to .y; z0 / 2 K and x D 0, and using the trivialization of B0;z 0 .
On the other hand, choosing .y 0 ; z00 / such that y 0 C j.z00 / D z 0 we obtain another
224 A. Polishchuk
isomorphism 1 W Gz 0 ! G given by the isomorphism

L.y;z0 /;y 0 Cj.z00 / D L.y;z0 /;0 ˝ By1 1
0 Cj.z 0 /;j.z / ˝ By;y 0 Cj.z 0 /
0
0 0
' L.y;z0 /;0 ˝ By1 0

1
0
1 1
0 ;j.z / ˝ Bj.z /;j.z 0 / ˝ By 0 ;y ˝ By;j.z 0 /
0 0
' L.y;z0 /;0/ ˝ By1 1

0 ;yCj.z / ˝ ByCj.z /;j.z 0 /
0 0 0
' L.y;z0 /;0/ ;

where we used the symmetry of BjY Y and of BjZ x Z x together with trivializations of
By ;0 and of B0;j.z0 / ' Bj.z0 /;0 . One can easily see from this that the restriction of
0 0 0
˛ D 1 B N to C Gm G is given by .c; / 7! .c; z 0 .c//, where .z 0 / W C ! Gm

is the character corresponding to z 0 mod Z. x
(ii) First, let us consider the case when Y and Z are transversal. Set Z x D j.Z0 /
x
X. By Lemma 2.2.5, Z is Lagrangian, so we can choose ˇ, so that .Z; N N is a
x ˇ/
Lagrangian pair. Now we have two completions of Z0 to an isotropic pair: .Z0 ; ˇjZ0 /
and .Z0 ; j ˇ/.N Hence, ˇjZ ' j ˇN ˝ for some 2 Z 0 . Let us choose x 2 X such
0
that z0 ' Ej.z0 /;x and set
ˇ 0 D ˇ ˝ E 1 jZfxg
Then ˇ 0 jZ ' j ˇ. N On the other hand, by Lemma 2.4.2, we have an equivalence
0
0/
t .Z; x/ W F .Y; ˛/.Z;ˇ / ! F .Y; ˛/.Z;ˇ
induced by Tx . This equivalence sends SZ;ˇ to Tx .SZ;ˇ / ' SZ;ˇ 0 . Thus, it is enough
to prove our statement with ˇ replaced by ˇ 0 . In other words, we can assume that
N
ˇjZ0 ' j ˇ.
Let G ! K D ker.sY;Z / (resp., G0 ! K0 D ker.sY;Z0 /) be the central extensions
by Gm appearing in the proof of Theorem 2.4.5 (resp., in the proof of (i)), so that we
have equivalences
F .Y; ˛/.Z;ˇ / \ Coh.X / ' G- rep1 F .Y; ˛/.Z0 ;ˇ / \ Coh.X / ' G0 - rep1 :
and
(2.4.10)
We have a commutative diagram of groups in which the horizontal arrows are injective:
G0 /G

K0 / K.
Under the equivalences (2.4.10) the natural functor

F .Y; ˛/.Z;ˇ / \ Coh.X / ! F .Y; ˛/.Z0 ;ˇ / \ Coh.X /
corresponds to the restriction functor
G- rep1 ! G0 - rep1 ' C - rep;
where C D ker.j jZ0 /.
We claim that the subgroup Z 0 X considered in (i) coincides with j.Z/. Indeed,
using the definition of Z 0 and the fact that j W Z ! X is generalized Lagrangian we
obtain a morphism of exact triangles:
Z /X / D.Z/ / ZŒ1
id

Z0 /X /Z
0 / Z 0 Œ1.
Since H 0 D.Z/ ' Z 0 , this implies the surjectivity of the map Z ! Z 0 , induced by j ,
which proves our claim.
Assume that … ' C is reduced. Then the restriction of the Schrödinger repre-
sentation VG ofLG to G0 corresponds under the equivalence G0 - rep1 ' C - rep to a
representation 2C ˚m . Let
M
VG D .VG /
2C
be the decomposition of VG , viewed as a representation of C , into isotypic components.

It is well known that all .VG / have the same dimension (for example, this can be
checked by embedding C into a maximal isotropic subgroup of K), hence
dim VG
dim.VG / D :
j…j
Using Lemma 1.2.1, we see that
m dim VGx D dim.VG / ;

x D G0 =C . Thus, we obtain the
where VGx is the Schrödinger representation of G
following formula for the multiplicities:
dim VG
m D : (2.4.11)
j…j dim VGx
x ˇN corresponds to the trivial representation of C , the isomorphism (2.4.7)

Since SZ;
will follow now from (2.4.6), once we show that m D NZ , i.e.,
dim VG
D j…j1=2 j0 .Z/j1=2
dim VGx
(note that … D j.Z/=j.Z0 / D 0 .j.Z//). Equivalently, we have to check that

deg.sY;Z /
D j…j j0 .Z/j:
deg.sY;j.Z0 / /
226 A. Polishchuk
Consider the commutative diagram of isogenies
Y Z0 /Y Z
id j sY;Z
sY;j.Z0 /
Y j.Z0 / / X,
where the top horizontal arrow is an embedding of the connected component of zero.
Note that
deg.j W Z0 ! j.Z0 // D jC j D j…j (2.4.12)
due to duality between … and C . Hence, this diagram leads to the following equality
of degrees
deg.sY;Z /
deg.sY;Z0 / D D j…j deg.sY;j.Z0 / /
j0 .Z/j
which gives the desired identity.
In the case when … is not reduced the representation of C corresponding to VG jG0
has a composition series, where the multiplicity m of a character 2 C .k/ is given
by a formula similar to (2.4.11) but with j…j replaced by #….k/. This leads to the factor
j…j
#….k/
in the multiplicities of the composition series of SZ;ˇ in F .Y; ˛/.Z0 ;ˇ / \Coh.X /.
Now let us consider the case when Y and Z are not necessarily transversal. As in
the proof of Theorem 2.4.5, using Lemma 2.2.7 we choose a Lagrangian pair .Y 0 ; ˛ 0 /
transversal to Z, such that we have an equivalence W F .Y; ˛/S ' F .Y 0 ; ˛ 0 /S com-
patible with H-action. Then induces an equivalence
F .Y; ˛/.Z0 ;ˇ / ' F .Y 0 ; ˛ 0 /.Z0 ;ˇ /

N
(and similar equivalences for .Z; ˇ/-invariants and for .j.Z0 /; ˇ/-invariants). As we
have seen in the proof of Theorem 2.4.5, .Sj.Z0 /;ˇN .Y; ˛// is cohomologically pure,
so changing by Œm for appropriate m 2 Z we can assume that induces an
equivalence
N N
F .Y; ˛/.j.Z0 /;ˇ / \ Coh.X / ' F .Y 0 ; ˛ 0 /.j.Z0 /;ˇ / \ Coh.X /:
We claim that in this case also induces an equivalence
F .Y; ˛/.Z0 ;ˇ / \ Coh.X / ' F .Y 0 ; ˛ 0 /.Z0 ;ˇ / \ Coh.X /: (2.4.13)
Indeed, we know that all simple objects of the finite length abelian category
F .Y 0 ; ˛ 0 /.Z0 ;ˇ / \ Coh.X /
are of the form Tx S, where S D Sj.Z0 /;ˇN .Y 0 ; ˛ 0 /. But 1 .Tx S / D Tx 1 .S / is

a coherent sheaf. Hence, 1 sends F .Y 0 ; ˛ 0 /.Z0 ;ˇ / \ Coh.X / to coherent sheaves.
Since the subcategory F .Y 0 ; ˛ 0 /.Z0 ;ˇ / D b .X / is compatible with the devissage with

respect to the standard t-structure, this easily implies that sends F .Y; ˛/.Z0 ;ˇ / \
Coh.X/ to coherent sheaves and our claim follows. Using the equivalence (2.4.13),
we can transfer the decomposition of SZ;ˇ .Y 0 ; ˛ 0 / in the category F .Y 0 ; ˛ 0 /.Z0 ;ˇ / \
Coh.X/ to that of SZ;ˇ .Y; ˛/ in the category F .Y; ˛/.Z0 ;ˇ / \ Coh.X /.
Corollary 2.4.8. Let .j W Z ! X; ˇ/ be a generalized Lagrangian pair. Assume that
char.k/ D 0. Then
MN
SZ;ˇ ' Sj.Z0 /;ˇi
iD1
for some line bundles ˇ1 ; : : : ; ˇN on j.Z0 / such that .j.Z0 /; ˇi / are Lagrangian pairs.
Proof. This follows from Proposition 2.4.7 (ii) together with Lemma 2.4.2.
Definition 2.4.9. Let X D XA and Y D f0g A, y so that we have F .Y; O/ ' D b .A/
(see Example 2.3.2). A coherent sheaf F on A is called a gLI-sheaf if there exists a
generalized Lagrangian pair .Z; ˇ/ such that F is .Z; ˇ/-invariant. We say that F is
an LI-sheaf if Z can be chosen to be a subvariety in XA .
We can show that for an LI-sheaf the subvariety Z in the above definition can be
recovered as the stabilizer with respect to the action of H. Here is a slightly more
general result.
Proposition 2.4.10. Let .Y; ˛/ and .Z; ˇ/ be Lagrangian pairs for an ess-abelian
variety X. Assume that the condition .?/ from Theorem 2.4.5 is satisfied and consider
the generating object SZ;ˇ D SZ;ˇ .Y; ˛/ 2 F .Y; ˛/.Z;ˇ / \ Coh.X /. Then the subset
of points x 2 X such that Tx .SZ;ˇ / ' SZ;ˇ coincides with Z.
Proof. The same argument as in the proof of Theorem 2.4.5 (based on Lemma 2.2.7
and Theorem 2.3.4) shows that it is enough to consider the case when Y and Z
are transversal. In this case SZ;ˇ is a vector bundle (see Theorem 2.4.5). Suppose
Tx .SZ;ˇ / ' SZ;ˇ for some x 2 X . By definition of .Z; ˇ/-invariance, we have an
isomorphism
ˇz ˝ Tz .p SZ;ˇ / ' p SZ;ˇ
in F .Y; ˛/Z D b .X Z/, where z D z un 2 Z.Z/ is the universal point and
p W X Z ! X is the projection. This leads to isomorphisms
ˇz ˝ Tz Tx .p SZ;ˇ / ' ˇz ˝ Tx Tz .p SZ;ˇ / ' p SZ;ˇ
in F .Y; ˛/Z . Since the commutator in the Heisenberg groupoid is given by the biex-
tension E, we deduce an isomorphism
Ez;x ˝ p SZ;ˇ ' p SZ;ˇ
on Z X. Restricting to Z f0g we get a trivialization of the line bundle Ez;x on Z.
Since Z is Lagrangian, this implies that x 2 Z.
228 A. Polishchuk
Corollary 2.4.11. Keep the assumptions of Proposition 2.4.10. Suppose that SZ;ˇ is
invariant with respect to another Lagrangian pair .Z 0 ; ˇ 0 /. Then Z 0 D Z and ˇ 0 D ˇ.
Proof. Proposition 2.4.10 implies that Z 0 Z, hence Z 0 D Z (since dim Z D
dim Z 0 ). By Lemma 2.2.7, we can assume that Y is transversal to Z, so that SZ;ˇ 2
F .Y; ˛/ is a vector bundle (see Theorem 2.4.5). Now the isomorphism
ˇz ˝ Tz .SZ;ˇ / ' ˇz0 ˝ Tz .SZ;ˇ /
on Z X leads to
ˇz1 ˝ ˇz0 ˝ SZ;ˇ ' SZ;ˇ :
Restricting to Z f0g we deduce the triviality of ˇ 1 ˝ ˇ 0 .
The formula for SZ;ˇ .Y; ˛/ from Theorem 2.4.5 has the following analog in the
non-transversal case.
Proposition 2.4.12. Let .Y; ˛/ be a Lagrangian pair, and let .j W Z ! X; ˇ/ be a
generalized Lagrangian pair. Assume that the condition .?/ from Theorem 2.4.5 holds.
Assume in addition that the pull-backs of ˛ and ˇ to the connected component of zero
in Y X Z are isomorphic. Then SZ;ˇ .Y; ˛/ is a direct summand in
.sY;Z / .˛ 1 ˇ 1 ˝ .i j / B 1 /;
where i W Y ! X is the embedding map.

Proof. We can factor the homomorphism sY;Z W Y Z ! X as a composition Y
q
Z ! X 0 ! X , where q is surjective and is injective. Now as in the proof of
Theorem 2.4.5, we obtain a structure of 1-cocycle with respect to the action of Y Z
on X 0 on the line bundle L on .Y Z/ X 0 given by the restriction of (2.4.4).
Furthermore, we see that for an L-twisted Y Z-equivariant sheaf F on X 0 one has
F 2 F .Y; ˛/.Z;ˇ / . The coherent sheaf
S D q .LjY Zf0g / D q .˛ 1 ˇ 1 ˝ .i j / B 1 /
on X 0 has a natural L-twisted Y Z-equivariant structure coming from the structure

of a 1-cocycle on L. Thus, by Theorem 2.4.5, SZ;ˇ .Y; ˛/ is a direct summand of S
provided S ¤ 0.
Now we observe that for y C j.z/ D 0, where y 2 Y , z 2 Z, we have an
isomorphism
1
By;j.z/ ' ˛01 ˝ ˛y ˝ ˛y :
Hence, we obtain the isomorphism of the restriction of L to ker.sY;Z / f0g with
˛y ˝ ˇz1 . Under the isomorphism Y X Z ! ker.sY;Z / W .y; z/ 7! .y; z/ this
line bundle corresponds to the difference between the restrictions of ˛ and ˇ. Thus, if
A is the connected component of zero in ker.sY;Z /, then our assumption on ˛ and ˇ
implies that LjAf0g ' O, hence S ¤ 0.
3 Functors associated with Lagrangian correspondences

3.1 Lagrangian correspondences for ess-abelian varieties
Definition 3.1.1. Let .X; BX / and .Y; BY / be ess-abelian varieties. A generalized
Lagrangian correspondence (g-Lag-correspondence for short) from X to Y is a gen-
eralized Lagrangian pair .L; ˛/ for .X Y; BX1 BY /. In the case when L is a
subvariety of X Y (i.e., .L; ˛/ is a Lagrangian pair), we say that we have a La-
grangian correspondence. Note that with every g-Lag-correspondence .L; ˛/ from X
to Y one can associate the opposite g-Lag-correspondence .
.L/; ˛ 1 / from Y to X ,
where
W X Y ! Y X is a natural isomorphism.
Example 3.1.2. Let f 2 Hom.X; Y /Q be an isomorphism between X and Y in the
category AbQ , so that we have an equality
X D fO B Y B f; (3.1.1)
where X W X ! Xy and Y W Y ! Yy are the symplectic self-dualities (in this situation

we say that f is symplectic). Then similarly to Example 2.2.4.1, we consider the graph
of f
L.f / D f.N x; Nf .x// j x 2 X g X Y;
where N > 0 is an integer such that Nf 2 Hom.X; Y /. The equality (3.1.1) implies
an isomorphism
2
.Nf Nf / EY ' EXN ' .N idX N idX / EX
of biextensions of X X . Let pX W L.f / ! X , pY W L.f / ! Y and W X ! L.f /

be the natural maps. Then the previous isomorphism can be rewritten as
. / .pY pY / EY ' . / .pX pX / EX :
Hence, .pY pY / EY ' .pX pX / EX , i.e., L.f / is isotropic. Since dim L.f / D
dim X D dim.X Y /=2, by Lemma 2.2.3 (ii), it is Lagrangian. Conversely, it is easy
to see that all Lagrangian abelian subvarieties L X Y , finite over X , are obtained
by the above construction (in particular, they are automatically finite over Y ).
If .L; ˛/ is a g-Lag-correspondence from X to Y and .M; ˇ/ is a g-Lag-correspon-
dence from Y to Z then we can try to define the composition .M B L; ˇ B ˛/ by setting
M B L D L Y M and defining ˇ B ˛ as the tensor product of the pull-backs of ˛ and
ˇ to L Y M . Below we will give a sufficient condition for .M B L; ˇ B ˛/ to be a
g-Lag-correspondence from X to Z (see Corollary 3.1.4).
It is convenient to consider the following more general setup. Let .X; ; B/ be an
ess-abelian variety. Assume that I X is an isotropic abelian subvariety so that the

composition I ! X ! Xy ! Iy is zero. In other words, we have I I ? , where
I ? is the kernel of the composition X ! Xy ! Iy. Then we can define the reduced
230 A. Polishchuk
ss-abelian variety .Xx ; /,

N where Xx D I ? =I and N is induced by . Moreover, if we
x on
assume that the restriction BjI I ? is trivial then there is an induced biextension B
x x x x
X X , so that .X ; B/ is an ess-abelian variety.
In this situation one can start with a generalized Lagrangian pair .Y; ˛/ for X and
try to define the corresponding Lagrangian pair for Xx .
Proposition 3.1.3. Let .Y; ˛/ be a generalized Lagrangian pair for X such that the
natural map Y ! X=I ? is surjective. Set Yx D Y X I ? and let ˛N be the pull-back
of ˛ to Yx . Then .Yx ; ˛/
N is a generalized Lagrangian pair for Xx .
Proof. By assumption we have an exact sequence of commutative groups
0 ! Yx ! Y ! X=I ? ! 0;
where the map Y ! X=I ? is the composition of the natural maps Y ! X and
X ! X=I ? . Dualizing and using the fact that .Y; ˛/ is generalized Largangian we
obtain an exact triangle in D b .G pr /
I ! ŒY ! X ! D.Yx / !
where the first arrow is the composition I ! X ! ŒY ! X . By the octahedron
axiom we obtain also an exact triangle
Y ! X=I ! D.Yx / ! :
In other words, D.Yx / is represented by the complex ŒY ! X=I . It follows that the
map Y ! X=I has finite kernel. Now the cartesian square
Yx /Y

Xx / X=I
leads to the exact triangle

Yx ! Xx ! D.Yx / ! :
It is easy to deduce from this that the pair .Yx ; ˛/
N is generalized Lagrangian.
Corollary 3.1.4. Let .L; ˛/ (resp., .M; ˇ/) be a g-Lag-correspondence from X to Y
(resp., from Y to Z). Assume that the natural map L M ! Y is surjective. Then
.M B L; ˇ B ˛/ is a g-Lag-correspondence from X to Z.
Proof. Consider the ess-abelian variety
.X Y Y Z; BX1 BY BY1 BZ /
and the isotropic subvariety I D 0X .Y / 0Z , where W Y ! Y Y is the
diagonal embedding. To get the result we apply Proposition 3.1.3 to the generalized
Lagrangian pair .L M; ˛ ˇ/ for this ess-abelian variety.
This result allows us to make the following

Definition 3.1.5. (i) For an ess-abelian variety .X; BX / we denote by Lag.X / the
set of isomorphism classes of g-Lag-correspondences .L; ˛/ from X to X such that
the projections p1 ; p2 W L ! X are surjective. The composition of correspondences
makes Lag.X/ into a monoid, where the unit corresponds to L D .X / X X
(and trivial ˛).
(ii) For an ss-abelian variety X we denote by U.X; Q/ the group of symplectic
automorphisms of X in AbQ (see Example 3.1.2).
Lemma 3.1.6. Associating with .i W L ! X; ˛/ 2 Lag.X / the morphism in End.X /Q
given by the correspondence i.L0 / X X gives a surjective homomorphism of
monoids
W Lag.X / ! U.X; Q/:
Proof. Since the projection p1 B i W L0 ! X is an isogeny there exists a morphism
W X ! L0 such that
i..x// D .nx; a.x//
for some integer n > 0 and some element a 2 End.X /. By definition, .L/ D
a=n 2 End.X/Q . Now let .j W M ! X; ˇ/ be another element of Lag.X /, and let
W X ! M0 be a morphism such that
j. .x// D .mx; b.x//
for some integer m > 0 and some element b 2 End.X /, and so .M / D b=m. Then
we have a morphism
X ! M B L D L X M W x 7! .mnx; ma.x/; ba.x//
which factors through the connected component of zero in M B L. Thus,
ba
.M B L/ D D .M / B .L/:
mn
The homomorphism is surjective since the map g 7! L.g/ gives its set-theoretic
section (see Example 3.1.2).
Later we will need the following simple result about the composition of correspon-
dences. For a g-Lag-correspondence L ! X Y let us set
q.L/ D deg.L ! X / (3.1.2)
with the convention that this is zero if the projection L ! X is not an isogeny.
Lemma 3.1.7. For g-Lag-correspondences L ! X Y and M ! Y Z such that
the map L M ! Y is surjective one has
q.M B L/ D q.L/ q.M /:
232 A. Polishchuk
Proof. The projection L Y M ! X factors as the composition L Y M ! L ! X ,

so
q.M B L/ D q.L/ deg.L Y M ! L/:
Now the cartesian square
L Y M /M

L /Y
shows that deg.L Y M ! L/ D deg.M ! Y / D q.M /.
3.2 LI-kernels and functors. Recall that with every abelian variety A we associate
the standard ess-abelian variety XA D A Ay equipped with a symplectic biextension
EA D BA ˝
BA1 , where BA D p14
P . The corresponding Heisenberg groupoid
acts on the category D .A/ (see Example 2.3.2).
b
Definition 3.2.1. (i) Let A and B be abelian varieties. Given a g-Lag-correspondence

.L; ˛/ from XA to XB we can consider the setup of Theorem 2.4.5 for the Heisenberg
groupoid associated with .XA XB ; BA1 BB /, acting on the category D b .AB/ that
can be identified with F .Y; ˛/ for Y D f0g Ay f0g By XA XB and ˛ D O (see
Remark 2.3.1.1). Therefore, we have the t -generator SL;˛ 2 Coh.A B/ of .L; ˛/-
invariant objects in D b .A B/ (see Theorem 2.4.5). We call SL;˛ the gLI-kernel
defined by .L; ˛/ and denote the corresponding gLI-functor by
ˆL;˛ WD ˆSL;˛ W D b .A/ ! D b .B/:
In the case when L is a subvariety of XA XB we call SL;˛ (resp., ˆL;˛ ) the LI-kernel
(resp., LI-functor) defined by .L; ˛/
(ii) We say that a g-Lag-correspondence L ! XA XB D A Ay B By is
nondegenerate if the projection pAB W L ! A B is surjective.
For a g-Lag-correspondence L ! XA XB we will denote projections to products
y etc.
of factors in XA XB as pA W L ! A, pAAy W L ! A A,
Example 3.2.2. Let L.g/ be the Lagrangian correspondence from XA to itself associ-
ated with an element g 2 U.XA ; Q/ (see Example 3.1.2). Let us write

a b
gD ;
c d
where a 2 Hom.A; A/Q , b 2 Hom.A;y A/Q , c 2 Hom.A; A/ y Q , d 2 Hom.A;y A/

y Q.
Then the correspondence L.g/ is nondegenerate if and only if the map in AbQ
A Ay ! A A W .x; / 7! .x; ax C b/
is an isomorphism. Equivalently, b should be invertible in AbQ .
The proof of the following lemma is straightforward and is left to the reader.
Lemma 3.2.3. Under the natural identification of D b .A B/ with F .Y; O/, where
Y D f0g Ay f0g By XA XB one has for K 2 D b .A B/
A 1
T.a;;b; / .K/ D t.a;b/ K ˝ .P P / D .Ta; / BA K BB Tb;
B
;
y b 2 B, 2 B.
where a 2 A, 2 A, y Here TxA are the kernels (2.3.2) and .T A /1 D
a;
Pa; ˝ Ta; .
A
This lemma allows to rewrite the condition of .L; ˛/-invariance for K 2 D b .AB/
as an isomorphism

K ' ˛l ˝ .TpA y.l/ /1 BA K BB TpB y .l/ ;
AA BB
for l 2 L, or equivalently,
TpA y .l/ BA K ' ˛l ˝ K BB TpBy .l/ : (3.2.1)
AA BB
Hence, the gLI-functor associated with .L; ˛/ satisfies the following “intertwining”
isomorphisms involving Heisenberg groupoids actions on D b .A/ and D b .B/:
ˆL;˛ B TpAAy.l/ ' ˛l ˝ TpB By .l/ B ˆL;˛
for l 2 L.
In the case when both projections L ! XA and L ! XB are surjective, i.e., .L; ˛/
is an element of Lag.XA / (see Definition 3.1.5 (i)), we can use (3.2.1) to move the
kernels of the form TxA through K. We record this observation for future use in the
next lemma.
Lemma 3.2.4. Let .L; ˛/ be a g-Lag-correspondence in Lag.XA / and K 2 D b .AA/
an .L; ˛/-invariant kernel. For any x 2 XA there exists x 0 2 XA and x 00 2 XA such
that
K BA TxA D TxA0 BA K and TxA BA K D K BA TxA00
in D b .A A/.
From Theorem 2.4.5 and Proposition 2.4.12 we get the following (almost) explicit
formulas for SL;˛ .
Lemma 3.2.5. (i) If .L; ˛/ is a nondegenerate g-Lag-correspondence from XA to XB
then SL;˛ is a vector bundle on A B given by
I
SL;˛ ' pAB ˛ 1 ˝ pAAyP 1 ˝ pB By P ;
where pAAy and pB By are projections from L to XA D A Ay and XB D B B, y

respectively; I is a Lagrangian subgroup in G WD ker.pAB / with respect to the central
extension of G associated with ˛jG .
234 A. Polishchuk
(ii) Let .L; ˛/ be an arbitrary g-Lag-correspondence from XA to XB . Assume that

the restriction of ˛ to the connected component of zero in ker.pAB / is trivial. Then
SL;˛ 2 Coh.A B/ is a direct summand in

pAB ˛ 1 ˝ pAAyP 1 ˝ pB By P :
Proof. (i) Applying Theorem 2.4.5 to the generalized Lagrangians Z D L ! XA XB

and Y D f0g Ay f0g By XA XB , we obtain that SL;˛ as an object of F .Y; O/
is given by
I
SL;˛ ' q ˛.l/1 ˝ PpA .l/; ˝ Pp1 B .l/;
;
where .l; ; / 2 L Ay B,
y and q is the map
L Ay By ! A Ay B By W .l; ; / 7! .pA .l/; pAy.l/ C ; pB .l/; pBy .l/ C /:
The identification of F .Y; O/ with D b .A B/ is given by the restriction to A f0g

B f0g XA XB (see Example 2.3.2). Since pAB W L ! A B is surjective, the
map q is also surjective. Therefore, we can use the base change formula to get the
required expression for SL;˛ as an object of D b .A B/.
(ii) This follows from Proposition 2.4.12 by a similar argument. Note that we can
still use the base change formula since the image of q is transversal to the subvariety
A f0g B f0g XA XB .
Examples 3.2.6. 1. According to [18] any equivalence between D b .A/ and D b .B/
appears as an LI-functor associated with the graph L.f / of a symplectic isomorphism
f W XA ' XB (see Example 3.1.2). More precisely, to construct such an equivalence
one uses the natural equivalence of D b .A/ with the representation F .Y; ˛/ of the
Heisenberg groupoid H.XB / of XB associated with the Lagrangian pair .Y; ˛/ that
corresponds to f0g Ay under the isomorphism f . Now Theorem 4.3 of [20] gives
an equivalence of F .Y; ˛/ with D b .B/. Note that the above construction depends on
a choice of an extension of f to an equivalence of Heisenberg groupoids H.XA / '
H.XB /, and the obtained equivalence D b .A/ ' D b .B/ is compatible with the action
of these groupoids. This implies that the corresponding kernel K on A B belongs to
D b .A B/.L.f /;ˇ / , where .L.f /; ˇ/ is some Lagrangian correspondence extending
the graph L.f /. Furthermore, by Proposition 3.2 of [18], K is cohomologically pure.
Hence, by Theorem 2.4.5, K is a direct sum of several copies of SL.f /;ˇ , which implies
that K ' SL.f /;ˇ (since the corresponding functor is an equivalence). For example,
the Fourier–Mukai transform W D b .A/ ! D b .A/ y is associated with the Lagrangian
XA XAy D A A y AA
y y Note
consisting of .x; ; ; x/, where x 2 A, 2 A.
that in the case A D B autoequivalences corresponding to symplectic automorphisms
of XA were also considered in [19] and [15].
2. Let f W A ! B be a homomorphism. Then with f we can associate a Lagrangian
correspondence from XA to XB by setting L D A B, y where the map L ! B is
induced by f and the map L ! Ay is induced by fO. In this case we can take ˛ to
be trivial. Then L-invariants in D b .A B/ are generated by the structure sheaf of

the graph of f . The corresponding LI-functor D b .A/ ! D b .B/ is the derived push-
forward Rf . The functor corresponding to the opposite Lagrangian is the pull-back
Lf W D b .B/ ! D b .A/.
The following proposition shows that the adjoint functors to gLI-functors are also
gLI-functors.
Proposition 3.2.7. Let .L; ˛/ be a g-Lag-correspondence from XA to XB , and consider
the permutation maps

W A B ! B A and
X W XA XB ! XB XA
Then the objects SL;˛ and
S
X .L/;˛1 in D b .A B/ are dual up to a shift. Hence,
the functors ˆL;˛ W D b .A/ ! D b .B/ and ˆ
L;˛1 W D b .B/ ! D b .A/ are adjoint
up to a shift.
Proof. Consider the duality functor
D W D b .A B/ ! D b .A B/ W K 7! R Hom.K; O/:
We have
T.a;;b; / .D.K// ' D.T.a;;b; / .K//
for .a; ; b; / 2 XA XB . On the other hand, for K 0 2 D b .B A/ we have
T.a;;b; / .
K 0 / '
.T.b; ;a;/ K 0 /:
Combining these isomorphisms we obtain
Tu .D.
K 0 // ' D.
.T
X .u/ K 0 //;
where u 2 XA XB . This shows that D.
S
X .L/;˛1 / has an .L; ˛/-invariant
structure.
Our main goal is to calculate convolutions of gLI-kernels (and hence compositions
of gLI-functors). We will need the following technical result for this.
Lemma 3.2.8. Let X be a commutative group scheme equipped with homomorphisms
f W X ! A, g W X ! A, y where A is an abelian variety. Assume that both f and g are
surjective with finite kernel. Assume also that L is a line bundle on X such that
ƒ.L/ ' .f g/ PA : (3.2.2)
Then the restriction of L to every connected component of X is nondegenerate of the
same index iL , so H i .X; L/ D 0 for i ¤ iL . Furthermore,
.dim H iL .X; L//2 D j ker.f /j j ker.g/j
and
iL D i.fy0 B g0 /; where f0 D f jX0 ; g0 D gjX0
(here we use the notation of Section 1.3).
236 A. Polishchuk
Proof. Set G D ker.g/, H D ker.f /, and let G0 D G \ X0 , H0 D H \ X0 . Then

we have exact sequences
g0
0 ! G0 ! X0 ! Ay ! 0;
f0
0 ! H0 ! X0 ! A ! 0:
Also, we have 0 .X / ' G=G0 ' H=H0 . The isomorphism (3.2.2) implies that for
every point x 2 H one has tx L ' L. Since H surjects onto 0 .X /, this implies
that the restrictions of L to all connected components of X are obtained from LjX0 by
translations. Hence,
H .X; L/ ' H .X0 ; LjX0 /˚j0 .X/j : (3.2.3)
On the other hand, the isomorphism (3.2.2) induces an isomorphism
ƒ.LjX0 / ' .f0 g0 / PA :
c0 associated with LjX0 is equal to the

Therefore, the homomorphism W X0 ! X
composition
f0 b0 c
g
X0 ! A ! X 0:
Therefore, the group K WD K.LjX0 / D ker./ fits into an exact sequence
0 ! H0 ! K ! G0 ! 0:
Hence, K is finite, i.e., the line bundle LjX0 is nondegenerate of some index i D iL .
It remains to observe that
jH j jGj
jKj D jH0 j jG0 j D ;
j0 .X /j2
and so from (3.2.3) we obtain
dim H i .X; L/ D j0 .X /j jKj1=2 D jH j1=2 jGj1=2 :
The formula for the index iL follows from the fact that L D fy0 B g0 .
In the next proposition we calculate convolutions of gLI-kernels under some tech-
nical nondegeneracy assumptions.
Proposition 3.2.9. (i) Let .L; ˛/ (resp., .M; ˇ/) be a g-Lag-correspondence from XA
to XB (resp., from XB to XC ). Assume that the natural homomorphism L M ! XB
(restricting to the given ones on L and M ) is surjective. Then for an .L; ˛/-invariant
object F 2 D b .AB/ and an .M; ˇ/-invariant object G 2 D b .B C / the convolution
F BB G (see Section 1.5) can be equipped with an .M B L; ˇ B ˛/-invariance structure.
(ii) Assume in addition that the g-Lag-correspondences L, M and M B L are

nondegenerate. Then one has an isomorphism in D b .A C /
SL;˛ BB SM;ˇ ' SM BL;ˇ B˛ Œ.L; M /;
and hence an isomorphism of functors
ˆM;ˇ B ˆL;˛ ' ˆM BL;ˇ B˛ Œ.L; M /:
Here the integer .L; M / 0 is defined as follows. Let us set
F D ker.L B M ! A C; .l; m/ 7! .pA .l/; pC .m///:
Consider the maps

y
f W F ! B and g W F ! B;
where f is the restriction by the projection to B and g is the restriction of the map
L B M ! By W .l; m/ 7! pBy .l/ pBy .m/:
Set f0 D f jF0 and g0 D gjF0 . Then both f0 and g0 are isogenies, fy0 Bg0 is symmetric
and we set
.L; M / D i.fy0 B g0 /: (3.2.4)
Proof. (i) For F 2 D b .A B/, G 2 D b .B C / let us set

F ˘ G D p12 F ˝L p23

G 2 F b .A B C /:
We have

.t.a;b/ F / ˘ .t.b;c/ G / ' t.a;b;c/ .F ˘ G /
and
.F ˝ .P P // ˘ .G ˝ .P P // D .F ˘ G / ˝ .P OB P /;
where .a; b; c/ 2 A B C and .; ; / 2 Ay By Cy . Hence, for .l; m/ 2 L XB M ,

F 2 D b .A B/.L;˛/ and G 2 D b .B C /.M;ˇ / we obtain an isomorphism
1
F ˘ G ' Tl .F / ˘ Tm .G / ' .TpA y .l/ / BA tb.l;m/ .F ˘ G / BC TpCy .m/ ;
AA CC
where b.l; m/ D pB .l/ D pB .m/. Consider the extension of the canonical map
L XB M ! XA XC to a map i W L XB M ! XA B XC XA XB XC
with the B-component given by b.l; m/. We can view in this way .M B L; ˇ B ˛/ as a
generalized isotropic pair for XA XB XC equipped with the (degenerate) biextension
BA1 OXB BC . The above calculation shows that F ˘ G has an .M B L; ˇ B ˛/-
invariance structure. This immediately leads to the required invariance structure on the
push-forward of F ˘ G to D b .A C /, i.e., on F BB G .
238 A. Polishchuk
(ii) Let pAB W L ! A B, pBC W M ! B C and pAC W L B M ! A C denote

the natural projections. Recall that
SzL;˛ WD pAB .˛ 1 ˝ pAAyP 1 ˝ pB By P / ' VL ˝ SL;˛ ;
SzM;ˇ WD pBC .ˇ 1 ˝ pB By P 1 ˝ pC Cy P / ' VM ˝ SM;ˇ ;
where VL and VM are vector spaces of ranks j ker.pAB /j1=2 and j ker.pBC /j1=2 , re-
spectively. Also, SM BL;ˇ B˛ is a vector bundle of rank j ker.pAC /j1=2 . Thus, by part (i)
and Theorem 2.4.5, it suffices to check that
SzL;˛ BB SzM;ˇ ' V Œi.fy0 B g0
where V is a vector bundle on A C of rank
j ker.pAC /j1=2 j ker.pAB /j1=2 j ker.pBC /j1=2 :
Using the commutative diagram with cartesian squares
L B MN
q qq NNN
qqq NNN
qqq NNN
q NNN
xqqq &
L CN A MI
vv NNN pp II
vvv NNN
N ppppp II
II
vv NN p p II (3.2.5)
vv NNN p pp II
{vv & wp $
L HH A B NC M
HH ppp N NNN uu
HH p NNN uu
H pp uu
pAB HHH
ppppp NNN
N uuu pBC
H# xpp N' zuu
AB B C
and applying base change and projection formulas we find an isomorphism
SzL;˛ BB SzM;ˇ

' pAB ˛ 1 ˝ pAAyP 1 ˝ pB By P BB pBC ˇ 1 ˝ pB By P 1 ˝ pC Cy P (3.2.6)
0
' pAC .L/;
0
where pAC W L B M ! A C is the projection and
L D ˛ 1 ˝ ˇ 1 ˝ pAAyP 1 ˝ pC Cy P 1 ˝ pB By By .idB ı/ P
with ı W By By ! By given by ı.1 ; 2 / D 1 2 . It is easy to see that
ƒ.L/.l1 ;m1 /;.l2 ;m2 / ' Pp1

A .l2 /;p y.l1 /
˝ PpC .m2 /;pCy .m1 / ˝ PpB .l2 /;pBy .l1 /pBy .m1 / :
A
0
Let us consider the fiber F D ker.pAC W L B M ! A C /. Then the above formula
specializes to
ƒ.L/jF F ' .f g/ P :
The diagram (3.2.5) implies that the projection L B M ! A B C is surjective
with a finite kernel G, where
jGj D deg.pAB / deg.pBC / D j ker.pAB /j j ker.pBC /j:
Hence, the projection f W F ! B is also surjective with the kernel isomorphic to G.

On the other hand, the subgroup H D ker.g/ F can be identified with
.L XB M / \ F D ker.pAC W M B L ! A C /;
where we view M BL D LXB M as a subgroup in LB M . Recall that by assumption,

pAC is surjective with finite kernel, so g W F ! By is surjective with finite kernel. Now
Lemma 3.2.8 implies that pAC 0 y
.L/ is of the form V Œi.f0 Bg0 for some vector bundle
V of rank
jGj1=2 jH j1=2 D j ker.pAC /j1=2 j ker.pAB /j1=2 j ker.pBC /j1=2 ;
as required.
To get rid of the nondegeneracy assumption in Proposition 3.2.9 we will use twisting
by autoequivalences. Namely, we will use the fact that in the case when L D L.f /
XA XA is the graph of a symplectic automorphism f W XA ! XA the corresponding
functor ˆL.f /;˛ is an equivalence (see Example 3.2.6.1).
Lemma 3.2.10. Let .L; ˛/ be a g-Lag-correspondence from XA to XB , and let

f W XA ! XA (resp., g W XB ! XB ) be a symplectic automorphism. Let also
.L.f /; ˛f / (resp., .L.g/; ˛g /) be a Lagrangian correspondence from XA to XA (resp.
from XB to XB ) extending the graph of f (resp., g). Then
SL.f /;˛f BA SL;˛ BB SL.g/;˛g ' SL.g/BLBL.f /;˛g B˛B˛f Œi
for some i 2 Z, and

L.g/ B L B L.f / D .f 1 g/.L/: (3.2.7)
Proof. Note that (3.2.7) follows immediately from the definition of the composition of
correspondences. Let us set for brevity Sf D SL.f /;˛f , Sg D SL.g/;˛g , S D SL;˛ and
S 0 D SL.g/BLBL.f /;˛g B˛B˛f . Let Sf 1 denote the kernel of the inverse autoequivalence
to the one defined by Sf , so that
Sf 1 BA Sf ' OA :
240 A. Polishchuk
Then Sf 1 has an .L.f 1 /;

˛f1 /-invariance structure, where
W L.f 1 / ! L.f /
is the isomorphism induced by the permutation of factors in XA XA , so Sf 1 is isomor-
phic to SL.f 1 /;
˛1 Œm for some m 2 Z. Similarly, for Sg 1 D SL.g 1 /;
˛g1 Œn,
f
where n 2 Z, we have
Sg BB Sg 1 ' OB :
Now by Proposition 3.2.9 (i), the object
P WD Sf BA S BB Sg 2 D b .A B/
has a structure of an object of D b .A A/.L.g/BLBL.f /;˛g B˛B˛f / . Similarly the object
Q WD Sf 1 BA S 0 BB Sg 1
has a structure of an object of D b .A B/.L;˛/ . Let Œa; b (resp., Œs; t ) be the co-
homological amplitude of P (resp., Q). Then by Theorem 2.4.5, we have an exact
triangle
<b P ! P ! V ˝ S 0 Œb ! ;
where V is a vector space and <b P is a successive extension of S 0 Œi with i < b.
Applying the convolution with Sf 1 on the left and with Sg 1 on the right we obtain
an exact triangle
Sf 1 BA <b P BB Sg 1 ! S ! V ˝ QŒb ! ;
where Sf 1 BA <b P BB Sg 1 is a successive extension of QŒi with i < b. It follows
that
H bCt .S / ' V ˝ H t .Q/ ¤ 0:
A similar argument shows that H aCs .S 0 / ¤ 0. Hence, b C t D a C s D 0 which
implies that b D a and s D t D a. Thus, we have
P ' V ˝ S 0 Œa and Q ' W ˝ S Œa
for some vector spaces V and W . Therefore, we obtain
S ' Sf 1 BA P BB Sg 1 ' V ˝ QŒa ' V ˝ W ˝ S;
which implies that V and W are one-dimensional.
Now we are ready to prove our main result about the convolution of gLI-kernels.
Theorem 3.2.11. Let .L; ˛/ (resp., .M; ˇ/) be a g-Lag-correspondence from XA to
XB (resp., from XB to XC ). Assume that the natural homomorphism L M ! XB is
surjective. Then one has an isomorphism in D b .A C /,
SL;˛ BB SM;ˇ ' SM BL;ˇ B˛ Œi for some i 2 Z;
and hence an isomorphism of functors
ˆM;ˇ B ˆL;˛ ' ˆM BL;ˇ B˛ Œi :
Proof. The idea is to reduce to the case when the correspondences L, M and M B L
are nondegenerate, considered in Proposition 3.2.9 (ii). Let A W Ay ! A, B W By ! B
and C W Cy ! C be symmetric isogenies associated with some ample line bundles
y By and Cy . For any n 2 Z let us define the symplectic automorphism fA .n/ of
on A,
XA D A Ay by
idA nA
fA .n/ D :
0 idAy
Similarly, we define symplectic automorphisms fB .n/ of XB and fC .n/ of XC .
We claim that for all n except for a finite number of values the correspondence
.fA .n/ fB .n//.L/ is nondegenerate. Indeed, by definition, nondegeneracy means
transversality to the Lagrangian subvariety f0g Ay f0g By XA XB . Thus, we
need n such that L is transversal to
.fA .n/ fB .n//.f0g Ay f0g B/

y D .n.A B //;
and our claim follows from Lemma 2.2.7 (ii). Similarly, for generic n the correspon-
dences .fB .n/ fC .n//.M / and .fA .n/ fC .n//.M B L/ are nondegenerate. Let
us choose such n and extend fA D fA .n/, fB D fB .n/ and fC D fC .n/ to some
Lagrangian correspondences .L.fA /; ˛1 /, .L.fB /; ˛2 / and .L.fC /; ˛3 /. Let also SfA ,
Sf 1 , etc., be the kernels giving the corresponding autoequivalences and their inverses.
A
By Lemma 3.2.10, we have
SL ' SfA BA S.fA fB /.L/ BB Sf 1 Œa

B
and
SM ' SfB BB S.fB fC /.M / BC Sf 1 Œb
C
for some a; b 2 Z, where we abbreviate SL;˛ to SL , etc. Therefore,
SL BB SM ' SfA BA S.fA fB /.L/ BB S.fB fC /.M / BC Sf 1 Œa C b:

C
Now Proposition 3.2.9 (ii) allows to compute the convolution of the kernels in the
middle, so we obtain
SL BB SM ' SfA BA S.fA fC /.M BL/ BC Sf 1 Œc

C
for some c 2 Z. By Lemma 3.2.10, the right-hand side is isomorphic to SM BL Œi for

some i 2 Z.
Corollary 3.2.12. Let .L; ˛/ and .M; ˇ/ be generalized Lagrangian pairs for the
standard ess-abelian variety XA . Assume that L and M are transversal, and let
G be the corresponding Heisenberg extension of L XA M (see Proposition 2.2.8),
with the underlying Gm -torsor given by the tensor product of the pull-backs of ˛ 1

and ˇ. Then HomD b .A/ .SL;˛ ; SM;ˇ / is concentrated in one degree and is an irreducible
representation of G of weight one.

242 A. Polishchuk
Proof. We have
Hom .SL;˛ ; SM;ˇ / ' H .A; SL;˛
_
˝L SM;ˇ /:
Now we interpret .L; ˛/ and .M; ˇ/ as correspondences from 0 to XA and use Propo-
_
sition 3.2.7 that says that SL;˛ is isomorphic to a shift of the LI-kernel correspond-
1
ing to .L; ˛ /, viewed as a correspondence from XA to 0. Now we can compute
_
H .A; SL;˛ ˝ SM;ˇ / by applying Theorem 3.2.11. Note that the composition of cor-
respondences in our case is L XA M , and ˇ B ˛ 1 is exactly the underlying Gm -torsor
of G. Thus, .M B L; x ˇ B ˛ 1 /-invariance structure on a vector space can be viewed
as a weight-1 representation of the Heisenberg extension G (see Example 2.2.4.3 and
Section 1.1).
Remark 3.2.13. Corollary 3.2.12 generalizes the well known result that for a nonde-
generate line bundle L on A the cohomology H .A; L/ is concentrated in one degree
and is an irreducible weight-1 representation of the Mumford’s theta group attached
to L (see [16], Section 16 and 23). Note that actions of some natural groups on the
cohomology of vector bundles on A were also considered by Umemura [26].

For g D ac db 2 U.XA ; Q/ let us denote b D b.g/ 2 Hom.A; y A/Q . Let us
consider the subset U 0 U.XA ; Q/ consisting of g such that b.g/ is invertible. Note
that the group U.XA ; Q/ is completely determined by the algebra R D End.A/˝Q and
the Rosati involution on it (with respect to some polarization). Namely, it consists
of automorphisms of the free rank-2 module over R preserving the standard skew-
Hermitian form ..x1 ; y1 /; .x2 ; y2 // D .x1 /y2 .x2 /y1 . It follows that U.XA ; Q/
can be identified with the group of Q-points in a connected algebraic group UXA (see
[19], Section 9, and [21], Section 4, for a more detailed study of the group UXA ,
which is denoted there by SL2;A;Q ). Since U.XA ; Q/ is Zariski dense in UXA (see
[1], 18.3) and since the invertibility of b.g/ is a Zariski open condition, we deduce that
the subset U 0 U.XA ; Q/ is big in the following sense: for any triple of elements
g1 ; g2 ; g3 2 U.XA ; Q/ the intersection .U 0 /1 \ U 0 g1 \ U 0 g2 \ U 0 g3 is non-empty
(this notion goes back to [27], IV. 42 while the term is due to D. Kazhdan). The
importance of this condition is due to the fact that a 2-cocycle of U.XA ; Q/ is uniquely
determined by its restriction to U 0 U 0 (see [21], Lemma 4.2).
Recall that we denote by Lag.XA / the monoid of g-Lag-correspondences .L; ˛/
from XA to XA such that both projections L ! XA are surjective. We have a surjective
homomorphism W Lag.XA / ! U.XA ; Q/ (see Lemma 3.1.6). Note that a g-Lag-
correspondence L is nondegenerate if and only if .L/ 2 U 0 (see Example 3.2.2).
Proposition 3.2.9 and Theorem 3.2.11 lead to the following computation of the con-
volution of gLI-kernels (and hence the composition of gLI-functors) associated with
g-Lag-correspondences from Lag.XA /.
Theorem 3.2.14. (i) For any pair of g-Lag-correspondences .L; ˛/; .M; ˇ/ 2 Lag.XA /
we have
SL;˛ BA SM;ˇ ' SM BL;ˇ B˛ Œ..M /; .L//
for some integer ..L/; .M //. Hence, we also have an isomorphism of functors
ˆM;ˇ B ˆL;˛ ' ˆM BL;ˇ B˛ Œ..L/; .M //:
(ii) The map .g1 ; g2 / is a 2-cocycle of U.XA ; Q/ with values in Z. We have
.g1 ; g2 / D i.b.g1 /1 b.g1 g2 /b.g2 /1 /
whenever b.g1 /, b.g2 / and b.g1 g2 / are invertible. Here i./ denotes the index of a
symmetric isogeny (see Section 1.3).
Proof. (i) Theorem 3.2.11 implies that
SL;˛ BA SM;ˇ ' SM BL;ˇ B˛ Œ.L; M /
for some integer .L; M /. We have to prove that .L; M / depends only on .L/ and
.M /. Let g D .L/ and let us equip L.g/ XA XA with a line bundle ˛g so that
.L.g/; ˛g / is a Lagrangian correspondence. By Proposition 2.4.7 (ii), SL;˛ is obtained
by successive extensions from objects of the form Tx .SL.g/;˛g / where x 2 XAA .
Using Lemmas 3.2.3 and 3.2.4, we can rewrite such objects as Tx 0 BA SL.g/;˛g with x 0 2
XA . This immediately implies that SL;˛ BA SM;ˇ is obtained by successive extensions
from objects of the form Tx 0 BA SL.g/;˛g BA SM;ˇ , hence .L; M / D .L.g/; M /. A
similar argument shows that .L; M / depends only on .M /.
(ii) The fact that .; / is a 2-cocycle follows from the definition. Let us show how
to rewrite the formula (3.2.4) in the required form for L D L.g2 / and M D L.g1 /,
where g1 ; g2 2 U 0 are such that g1 g2 2 U 0 . Let

˛i ˇi
gi D ; i D 1; 2;
i ıi
where ˇ1 and ˇ2 are invertible. For i D 1; 2 we have isomorphisms in AbQ
i W A Ay ! L.gi / W .a; / 7! .a; ; ˛i a C ˇi ; i a C ıi /:
Recall that .M; L/ is the index of the symmetric isogeny fy0 g0 W F0 ! F0 , where
p17
F D ker.L.g2 / A L.g1 / ! A A/ and f0 and g0 were defined in Proposi-
tion 3.2.9 (ii). Here we view L.g2 / A L.g1 / as a subvariety in XA XA XA XA D
A Ay (the last product has 8 factors) and denote by p17 the corresponding
projection to A A. It is easy to check that there is an isomorphism in AbQ
W Ay ! F0 W 7! .0; ; ˇ2 ; ı2 /; .ˇ2 ; ; 0; . 1 ˇ2 ı1 //;
where D ˇ11 ˛1 ˇ2 . Furthermore, we have
f0 ..// D ˇ2 and g0 ..// D .ı2 C /;

244 A. Polishchuk
Hence,
O fOg D ˇO2 .ı2 C / D ˇO2 .ı2 C ˇ11 ˛1 ˇ2 /
and we obtain
.g1 ; g2 / D i.ˇO2 .ı2 C ˇ11 ˛1 ˇ2 // D i..ı2 C ˇ11 ˛1 ˇ2 /ˇ21 / D i.ı2 ˇ21 C ˇ11 ˛1 /
which is equivalent to the desired formula.
Remark 3.2.15. In the case when g1 ; g2 2 U.XA ; Q/ \ End.XA /, our formula for
.g1 ; g2 / agrees with that of Orlov [18], Section 4, due to the standard formula for the
index of a line bundle (see Section 1.3).
3.3 Central extensions related to LI-endofunctors. From now on we assume that

char.k/ D 0.
Definition 3.3.1. Let us denote by KER.A; A/ the set of isomorphism classes of objects
of D b .A A/. We equip it with a semiring structure by taking the direct sum ˚ as
addition and the convolution BA (see Section 1.5) as multiplication.
Note that we have a homomorphism of semirings
KER.A; A/op ! Fun.D b .A/; D b .A// W K 7! ˆK ;
where ˆK is the Fourier–Mukai functor given by the kernel K (see Section 1.5) and
Fun.; / is the set of isomorphism classes of exact functors.
Definition 3.3.2. (i) Let us say that a cohomologically pure kernel K 2 KER.A; A/
is an LI-kernel (resp., weak LI-kernel) if there exists a Lagrangian correspondence
L
.L; ˛/ in Lag.XA / such that K is .L; ˛/-invariant (resp., K D N iD1 Ki , where Ki
are .L; ˛i /-invariant for some Lagrangian correspondences .L; ˛i / with common L).
(ii) Let KERLI .A; A/ (resp., KERwLI .A; A/ KER.A; A/; resp., KERLI ˚ .A; A/
KER.A; A/) denote the subset consisting of LI-kernels (resp., weak LI-kernels;
resp., finite direct sums of LI-kernels). By Theorem 3.2.14 (i) and Corollary 2.4.8,
KERLI ˚ .A; A/ is a subsemiring in KER.A; A/, while KER .A; A/ is a submonoid in
wLI
the mutliplicative monoid .KER.A; A/; BA /.

(iii) Let us consider the following equivalence relation on KERLI .A; A/ (resp.,
KERwLI .A; A/): K H K 0 if there exists a point x 2 XAA such that K 0 ' Tx .K/
LN
(resp., there exists a direct sum decomposition K D iD1 Ki and a collection of
L
points xi 2 XAA , i D 1; : : : ; N , such that K 0 ' N iD1 xi .Ki /.) By Lemmas 3.2.3
T
and 3.2.4, this is equivalent to the existence of x 0 2 XA and x 00 2 XA such that
K 0 ' TxA0 BA K ' K BA TxA00 :
This easily implies that the set of equivalence classes
KERLI .A; A/ WD KERLI .A; A/= H ' KERwLI .A; A/= H
inherits the monoid structure (with respect to the convolution BA ).
We have a central submonoid Z KERLI .A; A/ consisting of kernels of the form

˚m
OA
Œn, where A A A is the diagonal. The convolution with an element in Z is
a very simple operation:
˚m
K BA O Œn ' K ˚m Œn:
We will use the notation m K WD K ˚m . Note that Z is isomorphic to N Z, so
the localization of Z with respect to the multiplicative set N Z gives the group
ZQ D Q Z.
Let us denote by KERLI .A; A/Q the localization of KERLI .A; A/ with respect to
the central multiplicative set N Z.
Lemma 3.3.3. The monoid KERLI .A; A/Q is a group.
Proof. Let .L; ˛/ be a Lagrangian correspondence from XA to XA , such that both pro-
jections L ! XA are surjective. Consider the transposed Lagrangian correspondence
.
L; ˛ 1 / and the composed g-Lag-correspondence .Z; ˇ/ D .L B
L; ˛ B ˛ 1 /. Let
j W Z ! XA XA be the natural map, and let Z0 be the connected component of zero
in Z. Since we have the diagonal map L ! Z, we obtain that j.Z0 / is equal to the
diagonal X XA XA . Therefore, by Theorem 3.2.14 and Proposition 2.4.7 (ii), we
obtain
SL;˛ BA S
L;˛1 D N SX ;O Œi D N OA Œi
in KERLI .A; A/ with N 2 N and i 2 Z. This shows that SL;˛ has a right inverse in
KERLI .A; A/Q . Exchanging the roles of L and
L we also obtain the existence of a
left inverse. Since every element of KERLI .A; A/ is of the form n SL;˛ with n 2 N ,
we see that it is invertible in KERLI .A; A/Q .
The first two parts of the following theorem summarize most of the picture discussed
above.
Theorem 3.3.4. Assume that char.k/ D 0.
(i) The semiring KERLI ˚ .A; A/ has as an additive basis the kernels of the form
SL.g/;˛ Œn, where L.g/ is the Lagrangian correspondence from XA to XA associated
with g 2 U.XA ; Q/ (see Example 3.1.2), ˛ is a line bundle on L.g/ such that .L.g/; ˛/
is a Lagrangian pair, and n is an integer. The multiplication in KERLI ˚ .A; A/ has the
form
M
N
SL.g2 /;˛2 Œn2 BA SL.g1 /;˛1 Œn1 D SL.g1 g2 /;ˇi Œn1 C n2 C .g1 ; g2 / (3.3.1)
iD1
with .; / as in Theorem 3.2.14 and some line bundles ˇ1 ; : : : ; ˇN on L.g1 g2 /, such
that .L.g1 g2 /; ˇi / are Lagrangian pairs.
(ii) For each g 2 U.XA ; Q/ let us choose a line bundle ˛g in such a way that
.L.g/; ˛g / is a Lagrangian pair. Then the class S.g/ of the kernel SL.g/;˛g 2
KERLI .A; A/ in KERLI .A; A/ depends only on g. For g1 ; g2 2 U.XA ; Q/ one has the
following equality in KERLI .A; A/:
S.g2 / BA S.g1 / D N.g1 ; g2 / S.g1 g2 /Œ.g1 ; g2 /; (3.3.2)
246 A. Polishchuk
where N.g1 ; g2 / 2 N . The map g 7! S.g/ extends to an isomorphism of groups

b .XA ; Q/ ! KERLI .A; A/op ;
U (3.3.3)
Q
where Ub .XA ; Q/ is the central extension of U.XA ; Q/ by ZQ D Q Z associated

with the 2-cocycle .N.g1 ; g2 /; .g1 ; g2 //.
(iii) For g1 ; g2 2 U.XA ; Q/ one has
q.L.g1 //1=2 q.L.g2 //1=2

N.g1 ; g2 / D 2 N : (3.3.4)
q.L.g1 g2 //1=2
with q.L/ given by (3.1.2).
Proof. The assertion about the additive basis in KERLI ˚ .A; A/ follows from Theo-
rem 2.4.5 and Corollary 2.4.11. The equation (3.3.1) follows from Theorem 3.2.14
together with Corollary 2.4.8.
The fact that the class of SL.g/;˛g in KERLI .A; A/ does not depend on the choice
of ˛g follows immediately from Lemma 2.4.2. Furthermore, together with Corol-
lary 2.4.11 this implies that the map .g; m; n/ ! S.g/˚m Œn is a bijection between
U.XA ; Q/ N Z and KERLI .A; A/.
Applying Proposition 2.4.7 (ii) to Z D L.g1 / B L.g2 / we get
M ˚NZ
SL.g1 /BL.g2 /;˛g1 B˛g2 ' Tx .SL.g1 g2 /;˛g1 g2 / :
x2…0
Recall that here
j…0 j D j0 .j.Z//j D d WD deg.Z0 ! j.Z0 //

j0 .Z/j1=2
(see (2.4.12)) and NZ D d 1=2
. Hence, in KERLI .A; A/ we have
SL.g1 /BL.g2 /;˛g1 B˛g2 D j0 .Z/j1=2 d 1=2 S.g1 g2 /:
By Theorem 3.2.14, this implies (3.3.2) with
N.g1 ; g2 / D j0 .Z/j1=2 d 1=2 :
This gives a homomorphism (3.3.3) from the central extension U b .XA ; Q/, which is
easily seen to be an isomorphism from the above identification of the set KERLI .A; A/.
It remains to prove the formula (3.3.4). Note that by Lemma 3.1.7, we have
q.Z/ D q.L.g1 // q.L.g2 //:
Thus, (3.3.4) can be rewritten as

q.Z/ deg.Z ! XA /
j0 .Z/j d D D :
q.j.Z0 // deg.j.Z0 / ! XA /
But this can be checked by comparing the degrees in the commutative diagram of
isogenies
Z0 /Z

j.Z0 / /X
as in the end of proof of Proposition 2.4.7 (ii).

Corollary 3.3.5. Let KERLI .A; A/R be the push-out of the central extension sequence
1 ! Q ! KERLI .A; A/Q ! KERLI .A; A/Q =Q ! 1
with respect to the embedding Q ! R , and let Uz .X; Q/ be the central extension
of U.X; Q/ by Z associated with the 2-cocycle .; /. Then the map g 7! S.g/
q.L.g//1=2 extends to a homomorphism
Uz .X; Q/ ! KERLI .A; A/R :
The 2-cocycle N.; / given by (3.3.4) gives in general a nontrivial cohomology

class in H 2 .U.XA ; Q/; Q / as the following result shows.
Proposition 3.3.6. Let E be an elliptic curve with complex multiplication by
p
D, where D < 1 and jDj is square-free. Then the cohomology class in
H 2 .U.XE ; Q/; Q / given by N.; / is nontrivial.
p
Proof. Let Q R be the subgroup of x 2 R such that x 2 2 Q . We have a
homomorphism
p
qN 1=2 W U.XE ; Q/ ! Q =Q W g 7! q.L.g//1=2 mod Q
p
that we can view as an element of H 1 .U.XE ; Q/; Q =Q /. Now the class of N.; /
is equal to ı.q 1=2 / where
p
ı W H 1 .U.XE ; Q/; Q =Q / ! H 2 .U.XE ; Q/; Q /
is the coboundary map coming from the exact sequence

p p
1 ! Q ! Q ! Q =Q ! 1:
p it is enough to prove that qN cannot be lifted to a homomorphism U.XE ; Q/ !

1=2
Thus,
Q . We are p going to prove that it cannot even be p lifted to a homomorphism
U.XE ; Q/ ! Q =˙1. Indeed, using the isomorphism Q =˙1 ! Q W x 7! x 2
we see that this is equivalent to asserting that the homomorphism
qN W U.XE ; Q/ ! Q =.Q /2 W g 7! q.L.g// mod.Q /2

248 A. Polishchuk
cannot be lifted to a homomorphism to Q .

Note that the group U.XE ; Q/ p in our case is isomorphic to the unitary group
U D U2 .K; f /, where K D Q. D/ and f is a skew-Hermitian form of index 1
(we use the terminology of [6]). Let us consider the natural embedding
aN
W K =Q ! K W a 7! :
a
It is known that the image of the determinant map det W U ! K is contained in
.K =Q / (see [5], Theorem 3). Hence, we have a unique homomorphism W U !
K =Q such that det.g/ D ..g//. We claim that
q.g/
N D Nm..g//;
where Nm W K =Q ! Q =.Q /2 is induced by the norm homomorphism Nm W K !

Q . Indeed, it is known that in our case the special unitary group SU U coincides
with the normal subgroup T U generated by unitary transvections (see §5 in [6]), and
the latter subgroup coincides with SL2 .Q/ U . It is easy to see that for g 2 SL2 .Q/
the degree of the projection p12 W L.g/ ! E 2 is a square. Thus, qN is trivial on SU and
we should have
q.g/
N D ..g//
for some homomorphism W K =Q ! Q =.Q /2 . It remains to check that our
statement holds for diagonal matrices
1
a 0
ga D
0 aN
where a 2 K (since .ga / D a). In other words, we claim that in this case
2
q.L.g
N a // Nm.a/ mod.Q / :
It is enough to check this in the case when a 2 End.E/ K. Then the Lagrangian
correspondence L.ga / can be described as
L.ga / D f.ax; y; x; ay/

N j x; y 2 Eg E 2 E 2 ;
so q.L.ga // D deg.a/ D Nm.a/.

Finally, suppose we have a homomorphism W U ! Q such that Nm..g// D
.g/ mod.Q /2 . Let us specialize this equality to g D gpD . Then we should have
.gpD / D mod.Q /2 :
But factors through U=ŒU; U and gpD projects to an element of finite order in
U=ŒU; U . Therefore, .gpD / should be an element of finite order in Q , i.e.,
.gpD / D ˙1 which is a contradiction.
References
[1] A. Borel, Linear algebraic groups. 2nd edition, Springer-Verlag, New York 1991.
[2] T. Bridgeland, Derived categories of coherent sheaves. In International Congress of Math-
ematicians, Vol. II, Eur. Math. Soc., Zürich 2006, 563–582.
[3] T. Bridgeland, Stability conditions on K3 surfaces. Duke Math. J. 141 (2008), no. 2,
241–291.
[4] P. Deligne, Théorie de Hodge, III. Inst. Hautes Études Sci. Publ. Math. 44 (1974), 5–78.
[5] J. Dieudonné, On the structure of unitary groups. II. Amer. J. Math. 75 (1953), 665–678.
[6] J. Dieudonné, La géometrie des groupes classiques. Troisième édition, Springer-Verlag,
Berlin 1971.
[7] A. D. Elagin, Descent theory for derived categories. Russian Math. Surveys 64 (2009),
no. 4, 748–749.
[8] G. Laumon, Transformation de Fourier generalisée. Preprint, arXiv:alg-geom/9603004.
[9] M. Inaba, Toward a definition of moduli of complexes of coherent sheaves on a projective
scheme. J. Math. Kyoto Univ. 42 (2002), no. 2, 317–329.
[10] M. Kontsevich, Y. Soibelman, Notes on A-infinity algebras, A-infinity categories and non-
commutative geometry. In Homological mirror symmetry, Lecture Notes in Phys. 757,
Springer-Verlag, Berlin 2009, 153–219.
[11] Yu. I. Manin, Real Multiplication and noncommutative geometry (ein Alterstraum). In The
legacy of Niels Henrik Abel, Springer-Verlag, Berlin 2004, 685–727.
[12] L. Moret-Bailly, Pinceaux de variétés abéliennes. Astérisque 129 (1985).
[13] S. Mukai, Duality between D.X / and D.XO / with its application to Picard sheaves. Nagoya
Math. J. 81 (1981), 153–175.
[14] S. Mukai, Semi-homogeneous vector bundles on an abelian variety. J. Math. Kyoto Univ.
18 (1978), 239–272.
[15] S. Mukai, Abelian variety and spin representation. In Hodge theory and algebraic geometry
(Sapporo, 1994) (in Japanese); English transl. Univ. of Warwick preprint, 1998.
[16] D. Mumford, Abelian varieties. Tata Institute of Fundamental Research Studies in Math.
5, Oxford University Press, Oxford 1970.
[17] D. Orlov, Equivalences of derived categories and K3 surfaces. J. Math. Sci. (New York) 84
(1997), no. 5, 1361–1381.
[18] D. Orlov, Derived categories of coherent sheaves on abelian varieties and equivalences
between them. Izv. Math. 66 (2002), no. 3, 569–594.
[19] A. Polishchuk, Biextensions, Weil representation on derived categories and theta-functions.
Harvard Ph.D. thesis, 1996.
[20] A. Polishchuk, Symplectic biextensions and a generalization of the Fourier–Mukai trans-
form. Math. Research Lett. 3 (1996), 813–828.
[21] A. Polishchuk, Analogue of Weil representation for abelian schemes. J. Reine Angew. Math.
543 (2002), 1–37.
250 A. Polishchuk
[22] A. Polishchuk, Abelian varieties, theta functions and the Fourier transform. Cambridge
Tracts in Math. 153, Cambridge University Press, Cambridge 2003.
[23] A. Polishchuk, Kernel algebras and generalized Fourier–Mukai transforms. J. Noncommut.
Geom. 5 (2011), 153–251.
[24] A. Polishchuk, A. Schwarz, Categories of holomorphic vector bundles on noncommutative
two-tori. Comm. Math. Phys. 236 (2003), 135–159.
[25] B. Toën, The homotopy theory of dg-categories and derived Morita theory. Invent. Math.
167 (2007), 615–667.
[26] H. Umemura, On a certain type of vector bundles over an abelian variety. Nagoya Math. J.
64 (1976), 31–45.
[27] A. Weil, Sur certains groupes d’opérateurs unitaires. Acta Math. 111 (1964), 143–211.
Generic vanishing filtrations and perverse objects
in derived categories of coherent sheaves
Mihnea Popa
1 Introduction
This is a partly expository paper, intended to explain a very natural relationship between
two a priori distinct notions appearing in the literature: generic vanishing in the context
of vanishing theorems and birational geometry ([21], [22], [15], [24], [39], [41]), and
perverse coherent sheaves in the context of derived categories ([30], [7], [4], [46]).
Criteria for checking either condition are provided and applied in geometric situations.
A few new results, and especially new proofs, are included.
Let X and Y be noetherian schemes of finite type over a field, with D.X / and D.Y /
denoting their bounded derived categories of coherent sheaves. If X is proper and P
is a perfect object in D.X Y /, we have an integral (or Fourier–Mukai-type) functor
L
RˆP W D.X / ! D.Y /; RˆP ./ WD RpY .pX ./ ˝ P /:
Based on [39] and [41] one can introduce a generic vanishing (GV) filtration on the
derived category D.X /, with respect to this functor. Let dim X D d and dim Y D g.
Definition. For every integer m, define the full subcategory of D.X /
GVm .X/ WD fA 2 D.X / j codim Supp Ri ˆP A i C m for all i > 0g:
We say that an object in GVm .X / satisfies generic vanishing with index m with respect
to P . (The definition depends of course on P , and should rather be GVPm .X /, but since
there will be no danger of confusion I will avoid this to simplify the notation.)
For dimension reasons we have
GVg .X / D GVgC1 .X / D D RˆP1 .D0 .Y //:
On the other hand, the negative indices can go indefinitely, as for k 0 we have that
A 2 GVk .X/ is equivalent to AŒ1 2 GVk1 .X /, i.e.
GVk .X / D GV0 .X /Œk; for all k 0:

The author was partially supported by the NSF grant DMS-0758253 and a Sloan Fellowship.
252 M. Popa
(This is not the case for k < 0.) In conclusion the integral functor RˆP induces a
generic vanishing filtration on D.X / given by
RˆP1 .D0 .Y // D
D GVg .X / GV1 .X / GV0 .X /
DW GV.X / GV1 .X / GVd .X / GVd 1 .X /
In §2 I compare this with a cohomological support loci filtration appearing in vanishing

theorems.
By definition, the images of objects in the central component GV.X / correspond
via RˆP to the negative component of a perverse t -structure on D.Y / introduced by
Bezrukavnikov [7] (following Deligne; cf. also [4]) and Kashiwara [30],1 Grothendieck
dual to the standard t -structure. I include in §3 a self-contained proof of the existence
of this t -structure, following methods in [30] and [41], and identify the preimage
of its positive part in D.X / via RˆP as well. Perverse sheaves correspond via this
functor to what will be called geometric GV objects, and furthermore via duality to
objects satisfying the Weak Index Theorem with index d (WITd ) with respect to a
related kernel; see Theorem 3.8. (The standard example of such an object is the
canonical bundle !X of a variety X with generically finite Albanese map, according to
a fundamental theorem of Green–Lazarsfeld [21].) In case RˆP is a Fourier–Mukai
equivalence, this means that the subcategories of geometric GV objects and WITd
objects in D.X/ are hearts of natural t -structures.
The main point of the paper is that there are explicit cohomological criteria for
detecting all of the individual components of the generic vanishing filtration, or the
abelian category of perverse coherent sheaves and the components of a refined filtration
on it. It is shown in §4 and §5 how GVm .X / can be characterized either in terms of the
vanishing of certain hypercohomology (for m 0), or in terms of commutative algebra
(for m > 0). While quite basic theoretically, these criteria have concrete geometric
applications which provided the initial motivation for undertaking this study. Below is
a brief explanation.
I characterize in §4 the negative part of the GV filtration, when Y is projective, by
the vanishing of suitable hypercohomology groups. This is a slight generalization of
the main technical results in [39]. These were in turn inspired by the approach to the
Green–Lazarsfeld results for the canonical bundle in Hacon’s paper [24], where derived
category techniques made their first appearance in the study of generic vanishing. The
proof given here clarifies the treatment in [39]: after applying Grothendieck Duality to
1
This is one of many perverse t -structures, each depending on the choice of a perversity function. The
main result of [7] and [30] is to describe all functions that define such t -structures. Here the notion of
perversity corresponds only to this particular t-structure (cf. Definition 3.4), which is very geometric in
flavor, but only a sliver of the algebraic picture. It is an interesting question whether there exist natural
geometric objects satisfying generic vanishing conditions associated to other perversity functions, especially
since a few such functions have recently appeared in the study of Donaldson–Thomas invariants [6], [44],
[45] or stability conditions [33]. A different t -structure, roughly speaking associated to generic vanishing
for D-modules, will appear in [42].
Generic vanishing filtrations and perverse objects in derived categories 253
the integral functor, it emphasizes a distinct local part having to do with Kashiwara’s
characterization of the perverse t -structure on D.Y /, and a global one (for Y projective)
which is a simple cohomological characterization of perverse sheaves.
Section 5 deals with the positive part of the filtration. It contains results that were
announced (and proved for the correspondence between a smooth projective variety X
and Pic0 .X/) in [41]. One introduces a filtration on the abelian category of perverse
coherent sheaves given by a variant of Serre’s condition Sk . Due to a criterion of
Auslander–Bridger [3], this corresponds via RˆP to the GVm filtration with m > 0,
and in the case of finite homological dimension with a filtration by syzygy conditions
as well. In this last case, one can apply the Evans–Griffith Syzygy Theorem to bound
the rank of the dual of a non-locally free perverse sheaf. Significantly, this rank is
sometimes a standard invariant of X , like its holomorphic Euler characteristic .!X /
in the case of the Fourier–Mukai transform of !X . The homological commutative
algebra needed here, as well as in §3, is reviewed in an Appendix in §7.
In §6 I present geometric applications of the two main criteria characterizing com-
ponents of the GV filtration. The hypercohomology vanishing in Theorem 4.1 can
often be checked in practice by reducing it to standard Kodaira–Nakano-type vanish-
ing theorems. On projective varieties this accounts for essentially all known extensions
of the generic vanishing theorems of [21] (cf. §6.1 and §6.2 – as briefly explained there,
this whole circle of ideas will be generalized using M. Saito’s vanishing theorems for
mixed Hodge modules in the upcoming [42].) It can also be applied in connection
with problems related to moduli spaces of vector bundles (cf. §6.5) and to cohomol-
ogy classes on abelian varieties (cf. §6.4). On the other hand, the syzygy criterion in
Theorem 5.3 is applied in conjunction with the Evans–Griffith theorem as explained in
the previous paragraph, in situations when the cohomological support loci are known
to be small, to the study of irregular varieties (cf. §6.3).
At the moment there is an obvious disconnect between the theoretical results in
§3–5, which work in a general setting, and the applications in §6 and elsewhere, which
are almost all in the context of the integral functor induced by a universal line bundle
on X Pic0 .X /, for a smooth projective X . While this is not very restrictive from
the point of view of birational geometry, it is natural to expect generic vanishing
phenomena (or, equivalently, interesting perverse sheaves) on other moduli spaces,
and also on some spaces with singularities. I give a few such examples in §6.5, but
full results in this direction are still to be discovered, the main obstruction being the
current poor understanding of moduli spaces of sheaves on varieties of dimension three
or higher.2 Hence overall this material has two rather distinct aspects: characterizing
filtrations defined by support conditions in a formal setting on one hand, and applying
this to the concrete geometric study of generic vanishing for the Picard variety (or
other parameter spaces to a lesser extent) on the other. Work in the two directions has
been done by somewhat disjoint groups of people. I hope this note indicates that the
2
More precisely, from the present perspective one needs a good understanding of the Fourier–Mukai
transforms of their determinant line bundles.
254 M. Popa
results and methods involved often overlap or are extremely similar, and will serve
as an introduction to the geometric aspects for those more algebraically inclined and
viceversa.
Acknowledgements. Much of the material here has been worked out in articles or
discussions with G. Pareschi, so this paper should be considered at least partially
joint with him. I also thank Dima Arinkin, David Ben-Zvi, Iustin Coandă, Daniel
Huybrechts, Robert Lazarsfeld, Mircea Mustaţă and Christian Schnell for very useful
conversations. It was David Ben-Zvi who first pointed out [7] and Daniel Huybrechts
who first pointed out [30], both guessing that our work should have a connection with
these papers. Some of the material was presented at a workshop at University of
Michigan in May 2009. Special thanks are due to the organizers, especially Mircea
Mustaţă, for the opportunity and for an extremely pleasant week. Finally, I thankYujiro
Kawamata and Yukinobo Toda for the stimulating Tokyo conference in January 2011,
an outcome of which is the present volume.
2 Preliminaries and examples

Integral functors. The technical results will be proved for noetherian schemes of
finite type over a field.3 Given any such X , we denote by D.X / the bounded derived
category of coherent sheaves on X . For any other Y of this type, and P a perfect object
in D.X Y / (or more generally of finite tor-dimension over X and Y ), if X is proper
we have an integral functor
L
RˆP W D.X / ! D.Y /; RˆP ./ WD RpY .pX ./ ˝ P /:
If Y is also proper, we have the analogous

L
R‰P W D.Y / ! D.X /; R‰P ./ WD RpX .pY ./ ˝ P /:
According to standard terminology, an object A in D.X / is said to satisfy WITj , i.e.

the Weak Index Theorem with index j , if Ri ˆP A D 0 for all i ¤ j .
The dual of an object A in D.X / is A_ WD R Hom.A; OX /. When X is Cohen–
Macaulay, we also write RA WD R Hom.A; !X /, where !X is the dualizing sheaf of
X. When X is in addition projective, Grothendieck duality takes the following form
(cf. [39], Lemma 2.2):
Lemma 2.1. Assume that X is Cohen–Macaulay and projective, of dimension d . Then
for any A in D.X /, the Fourier–Mukai and duality functors satisfy the exchange formula
.RˆP A/_ Š RˆP _ .RA/Œd :

3
Note that all of the results in §3–6, except for those in §6 involving the Evans–Griffith syzygy theorem,
hold more generally for schemes of finite type over the spectrum of a ring with dualizing complex. The reader
will also note that many of the results hold, with appropriate formulations, for complex analytic spaces.
t-structures. Recall the following well-known

Definition 2.2. Let D bea triangulated category. A t -structure on D is a pair of full
subcategories D0 ; D0 , with Dn WD D0 Œn and Dn WD D0 Œn, satisfying
(a) D0 D1 and D1 D0 .

(b) HomD .A; B/ D 0, for all A 2 D0 and B 2 D1 .
(c) For any X 2 D there exists a triangle
A ! X ! B ! AŒ1
with A 2 D0 and B 2 D1 .
The category D0 \ D0 is called the heart (or core) of the t -structure; it is an
is of course, fora scheme X as above, the standard
abelian category. The main example
t-structure on D.X / given by D0 .X /; D0 .X / . Its heart is Coh.X /. Another
important t-structure will appear in §3.
Cohomological support loci. Generic vanishing conditions were originally given

in terms of cohomological support loci, so it is natural to compare the definition of
a GVm -object A in the Introduction with the condition that the i -th cohomological
support locus of A has codimension i C m. For any y 2 Y , let Py D Liy P in
D.Xy / D D.X/, where iy W Xy D X fyg ,! X Y is the inclusion.
Definition 2.3. Given an object A in D.X /, the i -th cohomological support locus of
A with respect to P is
L
VPi .A/ WD fy 2 Y j H i .Xy ; A ˝ Py / ¤ 0g:
Although Supp Ri ˆP A and VPi .A/ are in general different, they carry the same
numerical information in the following sense:
Lemma 2.4. For every m 2 Z, the following conditions are equivalent:
(1) A is a GVm -object with respect to P .
(2) codim VPi .A/ i C m for all i .
Proof. For the sake of completeness I reproduce the proof of Lemma 3.6 in [39]. Since
by cohomology and base change (for the hypercohomology of bounded complexes –
cf. [23], 7.7, especially 7.7.4, and Remarque 7.7.12 (ii)) we have that Supp Ri ˆP A
VPi .A/, it is enough to prove that (1) implies (2). The proof is by descending induction
L
on i. There certainly exists an integer s such that H j .X; A ˝ Py / D 0 for any j > s
and for any y 2 Y . Then, by base change, Supp Rs ˆP A D VPs .A/. The induction
256 M. Popa
step is as follows: assume that there is a component Vx of VPi .A/ of codimension

less than i C m. Since (1) holds, the generic point of Vx cannot be contained in
Supp Ri ˆP A and so, again by base change, we have that Vx VPiC1 .A/. This implies
that codim VPiC1 .A/ < i C m, which contradicts the inductive hypothesis.
Examples. Here are some basic examples related to the generic vanishing filtration.
Many more examples will appear in §6.
(1) The pioneering result on generic vanishing is the following theorem of Green–
Lazarsfeld, [21], Theorem 1: if X is a smooth projective variety and the general fiber of
its Albanese map has dimension k, then !X 2 GVk .X /. This is of course with respect
to the integral functor RˆP W D.X / ! D.Pic0 .X // given by a normalized Poincaré
bundle on X Pic0 .X /. If k D 0 we will see below criteria for when !X 2 GVm .X /
for some m > 0.
(2) Let C be a smooth projective curve of genus g 2, and P a Poincaré bundle
on C Pic0 .C /. It is a simple exercise to check that if L is a line bundle of degree d
on C , then V i .L/ D ; for i 2 and
V 1 .L/ Š W2g2d .C / Pic2g2d .C /;
the image of the Abel–Jacobi map of the .2g 2 d /-th symmetric product of C .
This has dimension 2g 2 d for d g 1. Hence the restriction of the generic
vanishing filtration to Pic.C / looks as follows:
GVg .C / GV1 .C / GV0 .C / GV1 .C /
where GVm .C / WD GVm .C / \ Pic.C / is equal to the whole of Pic.C / for m D 1

and to Picg1Cm .C / for m 0.
As a side remark, note that one can similarly introduce more refined loci
Vp1 .L/ D f˛ 2 Pic0 .C / j h1 .C; L ˝ ˛/ pg:
As we vary d and p, these correspond to all Brill–Noether loci Wdr .C / (see e.g. [2],
Chapter IV, §3), which can be quite complicated. An understanding of the pieces of
the generic vanishing filtration on D.C /, and of suitable refinements, could then be
seen as a broad generalization of Brill–Noether theory to arbitrary coherent sheaves, or
even objects in the derived category. The same goes for D.X / for higher dimensional
X, where even the line bundle picture is quite mysterious.
3 Comparison with a Bezrukavnikov–Kashiwara perverse

t-structure
Let Y be a Cohen–Macaulay scheme of finite type over a field. Bezrukavnikov [7]
and Kashiwara [30] have introduced a t -structure on D.Y / which corresponds to the
standard t -structure via taking derived duals, as part of a general procedure of defin-
ing perverse t -structures on the bounded derived category of coherent sheaves.4 The
exposition here is closer to that of [30]. Explicitly, define
p
D0 .Y / WD fA 2 D.Y / j codim Supp H i A i for all i 0g
and
p
D0 .Y / WD fA 2 D.Y / j HZi .A/ D 0; for all Z X closed with codim Z > ig:
Here HZi ./ denotes local hypercohomology with support in Z. The following was
proved in [30] (Proposition 4.3) in the smooth case (cf. also Lemma 5 of [7], in a more
general setting):

Proposition 3.1. (1) The pair p D0 .Y /; p D0 .Y / is a t -structure on D.Y /.

(2) We have R D0 .Y / D p D0 .Y / and R D0 .Y / D p D0 .Y /.
Regarding (2), the proof

in [30] shows theinclusions from left to right. Together
with the fact that the pair p D0 .Y /, p D0 .Y / is a t -structure (cf. [30], Theorem 3.5),
this formally implies that they are both equalities. To keep the exposition self-contained,
I give below direct proofs for both equalities in (2) for Y as above, using arguments
from [41], §2, and [30], §4, plus some consequences of duality for local cohomology
listed in Corollary 7.2. It implies that (1) also holds, by transfer of the standard t -
structure via duality. The two equalities are treated separately in the following two
propositions, stated slightly more generally.

Proposition 3.2. For every k 0 we have R Dk .Y / D p Dk .Y /, i.e. for
A 2 D.Y /,
codim Supp H i A i k for all i > 0 ” Ri A D 0 for all i < k:
Proof. Since the condition codim Supp.H i A/ i k is non-trivial only for i k,

by replacing A by AŒk we can reduce to the case k D 0.
Start first with A such that codim Supp.H i A/ i for all i . We have a spectral
sequence
E2pq WD Ext p .H q A; !Y / H) Rpq A:
In other words, to converge to Ri A, we start with the E2 -terms Ext iCj .H j A; !Y /.
Since codim Supp H j A j for all j , by Corollary 7.2 (a) we get Ext iCj .H j A; !Y / D
0 for all i < 0, which by the spectral sequence implies that Ri A D 0 for i < 0.
Now start with A such that Ri A D 0 for i < 0. Since R is an involution on
D.Y /, we have a spectral sequence
E2pq WD Ext p .Rq A; !Y / H) H pq A:

4
In fact Bezrukavnikov works under slightly more general, and Kashiwara under slightly more restrictive
hypotheses.
258 M. Popa
In other words to converge to H i A, we start with the E2 -terms Ext iCj .Rj A; !Y /.
But Corollary 7.2 (b) implies that for all j we have
codim Supp Ext iCj .Rj A; !Y / i C j:
Since by hypothesis we must have j 0, this implies that the codimension of the
support of all E2 -terms is at least i . But then chasing through the spectral sequence
this immediately implies that codim Supp H i A i .

Proposition 3.3. For every k 0 we have R Dk .Y / D p Dk .Y /, i.e. for
A 2 D.Y /,
HZi .A/ D 0; for all Z X closed with codim Z > i k
”
Ri A D 0; for all i > k:
Proof. One can again reduce to the case k D 0 by shifting by k. Let A be such that
Ri A D 0 for i > 0 and let B D RA. For any closed Z, we use the derived local
duality isomorphism
RZ .A/ Š R Hom.B; RZ .!Y //:
There is a (double) spectral sequence computing the right-hand side, namely
Ext p .H i B; HZj .!Y // H) RpiCj Hom.B; RZ .!Y //:
Now H i B D 0 for i > 0, while HZj .!Y / D 0 for j < codim Z by Corollary 7.2 (a).
This implies that HZi .A/ D 0 for i < codim Z as well.
Let now A be such that HZj .A/ D 0 for any closed Z with j < codim Z. By
[30], Proposition 4.6, this is equivalent to the fact that A can be represented by a
bounded complex F of locally free OY -modules in non-negative degrees. But then
we have a spectral sequence with E2 -terms Ext iCj .F j ; !Y / converging to Ri A.
Now j 0, hence Ext iCj .F j ; !Y / D 0 for i > 0, since the F j are flat. This implies
that Ri A D 0 for i > 0.
Definition 3.4. Coherent perverse t -structures were defined and studied in general in
[7] (Theorem 1) and [30] (Theorem 5.9), by means of perversity (or support) functions.
The particular t -structure in Proposition 3.1 corresponds to the perversity function
p W f0; : : : ; gg ! Z; p.m/ D g m;
where g D dim Y (or equivalently p 0 W Y ! Z, p 0 .y/ D dim OY;y ). It is further
studied via rigid dualizing complexes in a more general context in [46], where it is
called the rigid perverse t -structure.
A perverse coherent sheaf on Y is an object in
the heart of the t -structure p D0 .Y /; p D0 .Y / . We denote these by
Per.Y / D p D0 .Y / \ p D0 .Y /:
By Proposition 3.1 we have simply that Per.Y / D R.Coh.Y //.
Comparison via integral functors. Let now in addition X be a Cohen–Macaulay

scheme, projective over k, and let P be a perfect object in D.X Y /. The goal is to
use the above discussion after applying the functor RˆP to objects in D.X /. When Y
is projective, the next result is the equivalence of (1) and (3) in [39], Theorem 3.7. In
general it was stated (in the smooth case) in [41], Theorem 5.2 – the proof was given
there only in a special case, but the method is similar. Let
Q WD P _ ˝ pY !Y :
Corollary 3.5. Let A 2 D.X / and k 0. Then
A 2 GVk .X / ” RˆQ .RA/ 2 Dd k .Y /:
Proof. First, it is clear by definition that for k 0 we have
A 2 GVk .X / ” RˆP A 2 p Dk .Y /:
This result is then simply a reinterpretation of the equivalence in Proposition 3.2 applied
to RˆP A, via duality. Indeed, by Lemma 2.1 we have that RˆQ .RA/ 2 Dd k .Y /
if and only if .RˆQ_ A/_ 2 Dk .Y /. But the projection formula implies
.RˆQ_ A/_ Š R.RˆP A/:
Example 3.6 (The Fourier transform of the canonical bundle). In the situation of
Example (1) in §2, we saw that !X 2 GVk .X / (with respect to a Poincaré bundle
on X Pic0 .X/). On the other hand, by dimension reasons we have Ri ˆP _ OX D 0
for i > d . This means that when X is of maximal Albanese dimension (i.e. k D 0),
!X 2 GV.X/ and equivalently by Corollary 3.5, RˆP _ OX is a sheaf supported in
degree d (see Theorem 6.1 below for a more general statement). In other words,
RˆP !X is a perverse sheaf on Pic0 .X /.
In concrete geometric applications it is often the case as in the example above that
P is a sheaf, flat over Y , and RA 2 D0 .X / (usually a sheaf). By basic properties
of push-forwards we then have Ri ˆQ .RA/ D 0 for all i > d , as d is the dimension
of the fibers of the second projection. Among these objects, those that are GV deserve
a special name, as they are intimately related to perverse sheaves in D.Y / via RˆP .
Definition 3.7. A geometric GV -object on X is an object A 2 GV.X / such that
Ri ˆQ .RA/ D 0 for all i > d:
The conclusion of the discussion in this section can be summarized in the following
result. With respect to [39] I am also adding the new observation that under reasonable
assumptions (for example when X and Y are smooth and P lives in non-negative
degrees) the GV-objects in non-negative degrees are always geometric.5 The picture
will be expanded in the next two sections.
5
I thank C. Schnell for help with this issue.
260 M. Popa
Theorem 3.8. Let X , Y and P be as above. For A 2 D.X /, the following are
equivalent:
(1) A is a geometric GV-object.

(2) RˆP A is a perverse coherent sheaf.
(3) RA satisfies WITd with respect to Q D P _ ˝ pY !Y .
Assuming in addition that P ˝ pX !X 2 p D0 .X Y / (i.e. it can be represented by a

complex of locally free sheaves in non-negative degrees),

RˆP GV.X / \ D0 .X / Per.Y /
i.e. GV-objects in non-negative degrees are geometric.

Proof. By definition we have RˆP1 p D0 .Y / D GV.X /. On the other hand, by
Corollary 3.5, this is equivalent to Ri ˆQ .RA/ D 0 for all i < d . For the other half,
note that the identity in Proposition 3.3 applied to RˆP A implies after shifting by k
that
RˆP1 p D0 .Y / D fA 2 D.X / j R.RˆP A/ 2 D0 .Y /g:
But note that the formula at the end of the proof of Corollary 3.5, combined with
Lemma 2.1, implies
RˆQ .RA/ Š R.RˆP A/Œd
which gives
p
RˆP1 D0 .Y / D fA 2 D.X / j Ri ˆQ .RA/ D 0 for i > d g:
Finally, the last statement in the theorem uses part (2) of Proposition 3.9 below.

Proposition 3.9. (1) If P 2 D0 .X Y /, then RˆP p D0 .X / Dd .Y /.

(2) If P ˝ pX !X 2 p D0 .X Y /, then RˆP D0 .X / p D0 .Y /.
Proof. (1) Let A 2 p D0 .X /. We have a spectral sequence
E2pq WD Rp ˆP .H q A/ H) RpCq ˆP A:
By definition we have dim Supp H q A d q for all q. Since P is supported only in

negative degrees, we get by basic properties of push-forwards that Rp ˆP .H q A/ D 0
for all p > d q, which implies the result.
(2) Let A 2 D0 .X /. By Proposition 3.2 we have RA 2 p D0 .X /. Since
L
!Y ˝ !Y_ Š OY , by Proposition 3.3 we have Q 2 D0 .X Y /. By (1) we then have
RˆQ .RA/ 2 Dd .Y /. But we noted in the proof of the theorem that RˆQ .RA/ Š
R.RˆP A/Œd . We deduce R.RˆP A/ 2 D0 .Y /, which by Proposition 3.3 is
equivalent to RˆP A 2 p D0 .Y /.
Remark 3.10 (Gorenstein assumption). When Y is Gorenstein, the right-hand side

condition in Corollary 3.5, the definition of a geometric GV-object, and condition (3)
in Theorem 3.8 can all be phrased in terms of the simpler looking RˆP _ .RA/ instead
of RˆQ .RA/. Since this will always be the case in applications, we will use it freely
in what follows.
Example 3.11. With respect to the last statement in Theorem 3.8, it is not the case that
the preimage in D.X / of the category of perverse sheaves is equal to GV.X /\D0 .X /.
This does not happen even in the most favorable case: let X be an abelian variety,
Y D Xy its dual, and P a Poincaré bundle on X Xy . Consider L a nondegenerate
line bundle on X (i.e. with .L/ ¤ 0) satisfying WIT1 , so that L1 satisfies WITg1 .
Consider A D LŒ1 62 D.X /0 . Then RˆP A is a vector bundle supported in degree 0,
so A 2 GV.X/. On the other hand RˆP _ .RA/ Š Rg1 ˆP _ L1 Œg is supported
in degree g, so RˆP A is perverse.
An immediate consequence of Proposition 3.1 and its avatar Theorem 3.8 is extra
structure on the category of geometric objects satisfying generic vanishing with respect
to Fourier–Mukai equivalences.
Corollary 3.12. If RˆP is an equivalence, the geometric GV-objects with respect to
P (or dually the WITd -objects with respect to Q) form the heart of a t -structure on
D.X/, whose negative half consists of the category of GV-objects.
It is tempting to wonder whether this still holds under weaker assumptions on RˆP .
4 Hypercohomology vanishing characterization of Per.Y /

and of GVm .X / with m Ä 0
The main result of this section is a characterization of the negative part of the generic
vanishing filtration. It is a slightly more general version of [39], Theorem 3.7 and
Theorem A, in turn following Hacon’s approach to generic vanishing for the canonical
bundle in [24]. The proof given below follows a strategy different from that in the
papers above, emphasizing the connection with Proposition 3.1 and the fact that the
projectivity of Y is needed only for a simple cohomological characterization of perverse
sheaves. By a sufficiently positive ample line bundle I always mean a sufficiently high
power of an ample line bundle, and use the notation L 0.
Theorem 4.1. Let X and Y be Cohen–Macaulay schemes of finite type over k, of
dimension d and g respectively with X projective. Let P 2 D.X Y / be a perfect
object. For A 2 D.X / and k 0, the following are equivalent:
(1) A 2 GVk .X /.
(2) codimY VPi .A/ i k for all i .
(3) RˆQ .RA/ 2 Dd k .Y /.

262 M. Popa
If in addition Y is also projective, they are equivalent to the hypercohomology

vanishing
L
(4) H i .X; A ˝ R‰P Œg .L1 // D 0 for all i > k and any L 0 on Y .
Proof. The equivalence between (1) and (2) is Lemma 2.4. The equivalence between
(1) and (3) is Corollary 3.5, based on Proposition 3.2. So we only need to prove the
equivalence between (1) and (4) when Y is projective. We have by definition that
A 2 GVk .X/ if and only if .RˆP A/Œk 2 p D0 .Y /. It is a standard consequence of
the derived projection formula and Leray isomorphism that
L L
H j .Y; RˆP A ˝ L1 / Š H j .X; A ˝ R‰P .L1 // Š H j g .X; A ˝ R‰P Œg .L1 //
(cf. [39], Lemma 2.1). Hence all we need to show is the vanishing
H j .Y; RˆP A ˝ L1 / D 0 for all j > g C k and any L 0 on Y ;
which is a consequence of Corollary 3.5, and of Lemma 4.2 which describes perverse
sheaves cohomologically via a standard Serre vanishing argument.
Lemma 4.2. On a Cohen–Macaulay scheme Y , projective over k and of dimension g,
Per.Y / D fA 2 D.Y / j H j .Y; A ˝ L1 / D 0 for all j ¤ g and any L 0 on Y g:
More precisely:
• A 2 p D0 .Y / if and only if H j .Y; A ˝ L1 / D 0 for all j > g and any L 0
on Y ;
• A 2 p D0 .Y / if and only if H j .Y; A ˝ L1 / D 0 for all j < g and any L 0
on Y .
Proof. It is enough to prove the two statements at the end. Note first that A 2 p D0 .Y /
is equivalent to RA 2 D0 .Y / by Proposition 3.2. Now by Grothendieck–Serre
duality we have
H j .Y; A ˝ L1 / Š H gj .Y; RA ˝ L/ :
We have a spectral sequence
E2pq WD H p .Y; Rq A ˝ L/ H) H pCq .Y; RA ˝ L/:
By Serre vanishing, for L 0 we have E2pq D 0 for all p > 0 and all q, so the
spectral sequence degenerates and
H gj .Y; RA ˝ L/ Š H 0 .Y; Rgj A ˝ L/:
Again by Serre’s theorem, this is nonzero for L 0 if and only if Rgj A ¤ 0, hence
the assertion. The proof of the assertion for A 2 p D0 .Y / is completely analogous,
using Proposition 3.3.
The same argument shows that RˆQ .RA/ 2 Dd k .Y / (i.e. A is geometric) if
and only if
L
H i .X; A ˝ R‰P Œg .L1 // D 0
for all i < k and all L sufficiently positive on Y . In particular one can check the
perversity of the Fourier–Mukai transform via vanishing on X , which is crucial for the
applications in §7.
Corollary 4.3. For A 2 D.X / we have

L
RˆP A 2 Per.Y / ” H i .X; A˝R‰P Œg .L1 // D 0; for all i ¤ 0 and all L 0.
Remark 4.4. (1) The path pursued in [24] and [39] is to prove instead the equivalence
between (3) and (4) in Theorem 4.1, still based on Serre vanishing. This equivalence is
the natural extension to integral functors of a basic degeneration of the Leray spectral
sequence used in the proof of Grauert–Riemenschneider-type theorems. Concretely,
let f W X ! Y be a morphism of smooth projective varieties, and consider P WD O as
a sheaf on X Y , where X Y is the graph of f . Hence P induces the integral
functor RˆP D Rf , and R‰P is the adjoint Lf . Consider A and B D RA
objects in D.X/. A routine calculation shows that the equivalence of (3) and (4) in
Theorem 4.1 applied to A is the same as the well-known statement (individually for
each i ):
Ri f B D 0 ” H i .X; B ˝ f L/ D 0 for all L 0 on Y .
For instance, say B D !X and f is generically finite. Then for any ample L on Y ,
f L is big and nef, so H i .X; !X ˝ f L/ D 0 by Kawamata–Viehweg vanishing.
We get that Ri f !X D 0 for all i > 0, which is of course Grauert–Riemenschneider
vanishing (in the projective case).
(2) It is interesting to note that the vanishing condition in (4) is of a different
nature from standard vanishing theorems. For instance, when X and Y are dual
abelian varieties and LRˆP is the standard Fourier–Mukai functor, Mukai showed that
L R‰P Œg .L1 / Š L, where L W Y ! X is the standard isogeny associated to
L (see the proof of Theorem 6.1 below for more details). This suggests that, at least
when P is a locally free sheaf, R‰P Œg .L1 / should be interpreted as a positive vector
bundle, but which is less and less so as L becomes more positive.
5 Commutative algebra filtration on Per.Y /, describing

GVm .X / with m > 0
This section is concerned with a characterization of the positive part of the generic
vanishing filtration, extending the results in §3 of [41]. Note to begin with that the
subcategories GVm .X / with m > 0 are not obtained simply by shifting GV0 .X /, as
264 M. Popa
in the negative case. The main result, Theorem 5.3, is at this stage merely a matter of
notation and of navigating through results in other sections and in the Appendix.
Let Y be a Cohen–Macaulay scheme of finite type over a field, not necessarily
projective. The characterization of perverse coherent sheaves in Proposition 3.1 gives
Per.Y / D fA 2 D.Y / j RA is a sheaf in degree 0g:
We can naturally consider a filtration on Per.Y / according to the singularities of the

sheaf RA. To this end, define for m 0 (see Definition 7.4):
0
Per m .Y / WD fA 2 Per.Y / j RA satisfies Sm g:
We get a filtration
flocally free sheavesg Per 1 .Y / D D Per g .Y /

Per 1 .Y / Per 0 .Y / D Per.Y /:
If we restrict to the subcategory Per fhd .Y / Per.Y / consisting of objects such that
RA has finite homological dimension (so all of Per.Y / if Y is smooth), then we have
1 .Y / D D Per g .Y /:
fmaximal Cohen–Macaulay sheavesg D Per fhd fhd
(Note that in general F is a maximal Cohen–Macaulay sheaf if and only if RF is a

maximal Cohen–Macaulay sheaf; see [9], Theorem 3.3.10.) On the other hand, by the
Auslander–Bridger criterion in Proposition 7.5
m .Y / D fA 2 Per .Y / j RA is an m-th syzygy sheafg:

Per fhd fhd
Note that this last thing holds for the entire Per m .Y / if Y is Gorenstein in codimension
less than or equal to one. The Evans–Griffith Syzygy Theorem 7.6 can be rephrased
as follows.
Corollary 5.1. Let m > 0 be an integer, and let A be a perverse sheaf in Per fhd
m .Y /
which is not a locally free sheaf. Then rank.RA/ m.
The equivalence of (b) and (c) in Proposition 7.5 gives a characterization of objects
in Per m .Y /.
Lemma 5.2. For A 2 Per.Y / and m > 0,
A 2 Per m .Y / ” codim Supp H i A i C m for all i > 0:
This last condition corresponds to the GVm -piece of the generic vanishing filtration,
in the case of an integral functor RˆP W D.X / ! D.Y /. Recall that if A is a geometric
1
GV-object in D.X /, then RˆP A is perverse in D.Y /, so RA WD RˆQ .RA/Œd is
a sheaf.
Theorem 5.3. Let X and Y be Cohen–Macaulay schemes of finite type over k, with X
projective. Fix a kernel P 2 D.X Y /, and let A be a geometric GV-object in D.X /,
with respect to P . Let m > 0 be an integer. Then the following are equivalent:
(1) A 2 GVm .X /.
(2) RˆP A 2 Per m .Y /.
1 0
(3) RA satisfies Sm .
If these conditions are satisfied and in addition RˆP A 2 Per fhd
m .Y / or Y is Goren-
stein in codimension less than or equal to one, then they are also equivalent to
1
(4) RA is an m-th syzygy sheaf.
Proof. The equivalence of (1) and (2) is the content of Lemma 5.2. The equivalence of
(2) and (3) follows by definition and Lemma 2.1. In the Gorenstein or finite homological
dimension case, the equivalence of (3) and (4) is the Auslander–Bridger criterion quoted
in Proposition 7.5.
Definition 5.4 (Generic vanishing index). Let A be an object in D.X /. The generic
vanishing index of A (with respect to P ) is the integer
gv.A/ WD minfcodim Supp Ri ˆP A i g D minfcodim VPi .A/ i g:

i>0 i>0
(The last equality holds due to Lemma 2.4.) If Supp Ri ˆP A D ; for all i > 0, we
declare gv.A/ D 1. By definition A 2 GVm .X / if and only if gv.A/ m.
Theorem 5.3 and Corollary 5.1 imply then the following useful
Corollary 5.5. If A is a geometric GV-object with gv.A/ < 1, then rank.RA/ 1
gv.A/.
6 Geometric applications
The characterizations of GVm -objects (or of perverse objects and the syzygy filtration
on them) given in §4 and §5 can often be checked in practice. From a derived category
point of view, one obtains nontrivial concrete examples of perverse coherent sheaves.
This produces a number of different geometric applications, some described in what
follows. The general literature on applications of generic vanishing theorems to bira-
tional geometry is very extensive, a small sampling being given by [21], [19], [11],
[12], [13], [14], [25], [39], [41]. For instance, some of the most striking applications
include results of Ein–Lazarsfeld [19] on singularities of theta divisors, and of Chen–
Hacon [11] on a conjecture of Kollár on characterizing abelian varieties (cf. also [37]
for a proof using the interpretation in §6.2 as well) and [14] on Ueno’s conjecture on
the Albanese map of varieties with Kodaira dimension equal to 0.
266 M. Popa
6.1 Generic vanishing theorems. The derived category approach to generic vanish-
ing theorems was pioneered by Hacon [24]. The work described here is mostly taken
from [39], and grew out of trying to extend Hacon’s approach and the Green–Lazarsfeld
results [21].
In what follows let X be a smooth projective complex variety of dimension d , with
Albanese map a W X ! A. Let P be a Poincaré line bundle on X Pic0 .X /, and
consider as usual
RˆP W D.X / ! D.Pic0 .X //:
Every GV condition will be considered with respect to this functor. For a Q-divisor L
on X, we define L to be .LjF /, the Iitaka dimension along the generic fiber F of a,
if .L/ 0, and 0 if .L/ D 1.
Theorem 6.1 ([39], Theorem B). Let L be a line bundle and D an effective Q-divisor
on X such that L D is nef. If the dimension of the Albanese image a.X / is d k,
then !X ˝ L ˝ J.D/ belongs to GV.k.LD/ / .X /, where J.D/ is the multiplier
ideal sheaf associated to D.6 In particular, if L is a nef line bundle, then !X ˝ L
belongs to GV.kL / .X /.
The simplest instance of this (explaining also the terminology “generic vanishing”)
is the following:
Corollary 6.2. Let X be a smooth projective variety, and L a nef line bundle on X.
Assume that either one of the following holds:
(1) X is of maximal Albanese dimension (i.e. k D 0).
(2) .L/ 0 and LjF is big, where F is the generic fiber of a.
Then !X ˝ L belongs to GV.X /. In particular
H i .X; !X ˝ L ˝ P / D 0 for all i > 0 and P 2 Pic0 .X / general
and consequently .!X ˝ L/ 0.

Corollary 6.3. Under the hypotheses of Corollary 6.2, RˆP .!X ˝ L/ is a perverse
coherent sheaf on Pic0 .X /.
A more precise statement for !X is given in Corollary 6.11.
I only sketch the proof of Corollary 6.2, under the hypothesis (1), as an example
of the use of Theorem 4.1. This contains all the key ideas needed for Theorem 6.1,
the rest involving a standard extension to multiplier ideal sheaves, the additivity of
the Iitaka dimension, and extensions of the Kawamata–Viehweg vanishing theorem
(for full details cf. [39], §5). The main idea of the proof is due to Hacon [24], with
6
Multiplier ideals are treated for example in Chapter 9 of [32]. Here are some easily understood examples:
when D is integral one has J.D/ D OX .D/, while when D is a simple normal crossings Q-divisor one
has J.D/ D OX .ŒD/.
a refinement from [39] that allows for bypassing Hodge-theoretic results, hence the
extension to twists by arbitrary nef line bundles.
Let L be a nef line bundle on X . It is enough to show that !X ˝L satisfies condition
(4) in Theorem 4.1. Let M be an ample line bundle on Ay Š Pic0 .X /, and assume for
simplicity that it is symmetric, i.e. .1Ay/ M Š M . We consider the two different
Fourier transforms RM D R0 M D pA .p yM ˝P / (on A), where P is a Poincaré
A
bundle on A Ay so that P Š .a id y/ P , and R‰P Œg .M 1 / D Rg ‰P .M 1 / DW
1 (on X). These are both locally free sheaves. One can check with a little care that
M 1
A
1 Š a R .M / Š a .R M / :
M 1 g 1 0 _
On the other hand, by [34], 3.11, the vector bundle R0 M has the property:

M .R0 M / Š H 0 .M / ˝ M 1 :
Here M W Ay ! A is the standard isogeny induced by M . We consider then the fiber

product X 0 WD X A Ay induced by a and M :
X0 /X
b a
M
y
A / A.
It follows that
1
M 1 Š
a .R0 M /_ Š b .H 0 .M / ˝ M / Š H 0 .M / ˝ b M: (1)
What we want is the vanishing
1
H i .X; !X ˝ L ˝ M 1 / D 0 for all i > 0:
(Note that this will work for any ample line bundle M , the condition M 0 being
required only for the equivalence in Theorem 4.1 to hold.) Since , like M , is étale,
1
it is enough to prove this after pull-back to X 0 , so for H i .X 0 ; !X 0 ˝ L ˝ M 1 /.
But by .1/ we see that this amounts to the vanishing
H i .X 0 ; !X 0 ˝
L ˝ bM / D 0 for all i > 0:
Now L is nef and b M is big and nef (as the pull-back of an ample line bundle by
a generically finite map), so this follows from Kawamata–Viehweg vanishing.
A completely similar approach, replacing at the end Kawamata–Viehweg by other
standard vanishing theorems, proves the following results for higher direct images of
canonical bundles and for bundles of holomorphic forms.
268 M. Popa
Theorem 6.4. Let f W Y ! X be a morphism, with X, Y smooth projective varieties.

Let L be a nef line bundle on f .Y / (reduced image of f ). If the dimension of f .Y / is
d and that of its image via the Albanese map of X is d k, then Rj f !Y ˝ L is a
GV.kL / -sheaf on X for any j .
Theorem 6.5. Let X be a smooth projective variety of dimension d , with Albanese
image of dimension d k. Denote by f the maximal dimension of a fiber of a, and
consider l WD maxfk; f 1g. Then:
j
(1) X belongs to GV.d j Cl/ .X / for all j .
j
(2) codimAy V i .X / maxfi C j d l; d i j lg for all i and all j .
j
(3) If L is a nef line bundle on X and a is finite, then X ˝L belongs to GV.d j / .X /
for all j .
Theorem 6.4 is again due to Hacon [24] in the case L D OX . Another statement of a
j
slightly different flavor for X , not covered by these methods, can be found in [21]. For
the first statement one uses Kollár’s vanishing and torsion-freeness theorems for higher
direct images of canonical bundles, while for the second Nakano and Bogomolov–
Sommese vanishing. Details and further results along these lines can be found in [39],
§5. Note that Theorem 6.4 produces other natural examples of perverse sheaves on
Picard varieties.
Corollary 6.6. Let X be a smooth projective variety and a W X ! A its Albanese map,
and let L be any nef line bundle on a.X /. Then RˆP .Ri a !X ˝ L/ are perverse
sheaves on Pic0 .X / for all i , where RˆP is the standard Fourier–Mukai functor on
D.A/.
Remark 6.7. All of these results, suitably interpreted, hold more generally for mor-
phisms X ! A, with A an abelian variety and Ay replacing Pic0 .X /. This allows in
particular for a more general statement of Theorem 6.4, where X can be singular. It
is also interesting to note that the methods presented here give an algebraic proof of
generic vanishing in characteristic 0 (as well as in positive characteristic in some cases),
which had traditionally been approached via transcendental methods. (For details cf.
[39], end of §5 and Remark 6.7.)
Generic vanishing and perverse coherent sheaves via mixed Hodge modules. The
upcoming [42] generalizes the results above, when L D OX , using M. Saito’s theory
of mixed Hodge modules (see e.g. [43]). The main point is that every mixed Hodge
module on an abelian variety gives rise to a collection of perverse coherent sheaves
y namely the graded pieces of the filtration on the underlying D-module. This
on A,
is again an application of the criterion in §4, together with the vanishing theorem
[43], Proposition 2.33, for the graded pieces of the de Rham complex associated to a
D-module of this type.
Theorem 6.8 ([42]). Let A be a complex abelian variety, and M a mixed Hodge
module on A with underlying filtered D-module .M; F /. Then for each integer k, the
transform RˆP .Gr F y
k M/ is a perverse coherent sheaf on A.
This provides new information, via the Decomposition Theorem, even when applied
to the push forward to A D Alb.X / of the Hodge module QX H
Œd associated to the
trivial variation of Hodge structures. For s 2 N, let As D fy 2 A j dim a1 .y/ sg,
and define the defect of a to be

ı.a/ D max 2s dim X C dim As ;
s
where the maximum is over all values of s up to the maximal dimension of a fiber of a.
Theorem 6.9 ([42]). Let X be a smooth complex projective variety of dimension d .

j
Then X is a GV.d j Cı.a// -sheaf for each j .
For instance, if the Albanese map is semi-small (so that in particular that X is of
maximal Albanese dimension) then ı.a/ D 0, and one obtains that X satisfies the
generic Nakano vanishing theorem: for every i; j 2 N with i C j > d ,
j
codim V i .X / i C j d:
In [42] we explain how Theorem 6.9, or its proof, imply all of the standard generic
vanishing theorems mentioned above for L D OX . We provide further applications
to cohomological support loci for local systems, as well as to constructing perverse
coherent sheaves with respect to a Bezrukavnikov–Kashiwara t -structure different from
the one appearing in this paper.
6.2 Vanishing of higher direct images. The equivalence between (1) and (3) in
Theorem 4.1 says that once the (geometric) GV-condition is established for an object
A, the corresponding integral transform of the Grothendieck dual object is, up to shift,
a sheaf. Again in the setting of RˆP W D.X / ! D.Pic0 .X //, here is the main instance
of this:
Corollary 6.10. Let X be a smooth projective variety of dimension d and Albanese

dimension d k. Then
Ri ˆP _ OX .Š Ri p2 .P _ // D 0 for i 62 Œd k; d :
In particular, if X is of maximal Albanese dimension, then OX satisfies WITd , so that

its Fourier–Mukai transform is a sheaf
c
OX Š RˆP _ OX Œd :
270 M. Popa
Although with the methods presented in this paper this statement is an immediate
consequence of generic vanishing, its history follows somewhat the opposite direction:
a more precise version was first proved when X D A is an abelian variety by Mumford
[35], §13. It was then conjectured to be true in general by Green–Lazarsfeld [22],
Problem 6.2. This was proved by Hacon [24] and Pareschi [36], both showing that
it further implies generic vanishing. (A similar statement holds for compact Kähler
manifolds, cf. [41].) The existence of the sheaf O cX (or equivalently the perversity of
RˆP !X ) has significant consequences via the results in §5, as we will see in §6.3.
In the reverse direction, the non-vanishing of higher direct images implies a more
precise version of Corollary 6.3 in the case of the canonical bundle.
Corollary 6.11. The object RˆP !X is a perverse sheaf if and only if X is of maximal
Albanese dimension.
Proof. The if part follows from Corollary 6.3. Assume now that RˆP !X is perverse.
We have seen that this is equivalent to the fact that Ri ˆP _ OX D 0 for i ¤ d . We
claim on the other hand that a result of Kollár implies that if the Albanese map a of X
has generic fiber of dimension k, then Rd k ˆP _ OX ¤ 0. Indeed, by Theorem 3.1 of
[31] one has a decomposition in the derived category
L
Ra !X Š kiD0 Ri a !X Œi :
This decomposition implies in particular that H d .X; !X / Š H d k .A; Rk a !X /,

as the support of all Ri a !X is at most d k dimensional. Hence we must have
Rk a !X ¤ 0. But then the claim follows by a standard application of Grothendieck
duality Lemma 2.1.
Finally, the extra results listed in §6.1 produce other examples of WITd -objects, so
via Fourier–Mukai duals of perverse sheaves. For instance:
• The same statements as in Corollary 6.10 hold replacing OX by L1 for any nef
line bundle L on X.
• Under the hypotheses of Theorem 6.4, Aj WD R.Rj f !X ˝ L/ satisfies WITd
j D RˆP Aj Œd are all sheaves.
for each j , so A
6.3 Bounding the holomorphic Euler characteristic and applications to irregular

varieties. The syzygy filtration in §5, combined with the Evans–Griffith theorem, can
be applied to bound the holomorphic Euler characteristic of irregular varieties (i.e.
those with h1 .OX / ¤ 0, or equivalently b1 .X / ¤ 0) of maximal Albanese dimension.
c
Starting with such a variety X , we consider the sheaf OX as in Corollary 6.10, and find
the suitable Per m .Y / to which its dual RˆP !X belongs according to the existence of
nontrivial irregular fibrations of X .
Corollary 6.12. If X is a smooth projective complex variety of maximal Albanese
dimension:
(1) RˆP !X 2 Per gv.!X / .Pic0 .X //.

(2) .!X / gv.!X /.
Proof. This follows immediately from Theorem 5.3 and Corollary 5.5 applied to !X .
The only thing that needs mention is that gv.!X / < 1, as 0 2 VPd .!X / with d D
dim X.
The key point is that according to [22], gv.!X / is bounded in terms of the fibrations
of X over lower dimensional varieties of maximal Albanese dimension (those that
cannot be fibered further should be considered the building blocks in the study of
irregular varieties).
Corollary 6.13. Let X be of maximal Albanese dimension, with q D h1 .X; OX / and
d D dim X. If X is not fibered over any nontrivial normal projective variety of maximal
Albanese dimension satisfying q.Yz / dim Y q d m C 1 for any smooth model
Yz , then
RˆP !X 2 Per m .Pic0 .X // and .!X / m:
Proof. One applies Corollary 6.12, noting that the hypothesis implies gv.!X / m.
Indeed, an argument based on [22], Theorem 0.1, says that X is fibered over a normal
projective variety Y of lower dimension such that any smooth model Yz is of maximal
Albanese dimension, and
q.X / dim X .q.Yz / dim Y / gv.!X /:
The argument is described in the proof of [41], Theorem B.
The most significant instance of this result is an extension to arbitrary dimension of
a classical result on surfaces due to Castelnuovo and de Franchis. I only state a slightly
weaker version here for simplicity. Note that it holds in the Kähler case as well.
Theorem 6.14 ([41], Theorem A). Let X be an irregular smooth projective complex
variety. If X does not admit any surjective morphism with connected fibers onto a
normal projective variety Y with 0 < dim Y < dim X and with any smooth model Yz
of maximal Albanese dimension, then
.!X / q.X / dim X:
Along similar lines, arguments based on Theorem 5.3 lead to classification results.
For instance, they are crucial in another recent extension to arbitrary dimension due
to Barja–Lahoz–Naranjo–Pareschi of a well-known fact on the bicanonical map of
surfaces of general type.
Theorem 6.15 ([5], Corollary B). Under the same assumptions as in Theorem 6.14, if
in addition X is of general type and dim X < q.X /, the linear system j2KX j induces
a birational map unless X is birational to a theta divisor in a principally polarized
abelian variety.
272 M. Popa
6.4 Perverse sheaves coming from special subvarieties of abelian varieties. Let
.A; ‚/ be an indecomposable principally polarized abelian variety (ppav) of dimension
g, and let P be a Poincaré bundle on A A. y For subvarieties X A, the question
whether RˆP .X .‚// is a perverse sheaf on Ay is related to a beautiful geometric
problem. First, note that some of the most widely studied special subvarieties of
ppav’s satisfy this property:
• If .J.C /; ‚/ is the polarized Jacobian of a curve of genus g 2, and for any
1 d g 1 we denote by Wd the image of the d -th symmetric product of C in J.C /
via an Abel–Jacobi map, then Wd .‚/ is a geometric GV-sheaf, i.e. RˆP .X .‚// is
perverse (cf. [38], Proposition 4.4).
• If X P4 is a smooth cubic hypersurface, .J.X /; ‚/ is its intermediate Jacobian
(a five dimensional ppav which is not the Jacobian of a curve), and F J.X / is
the Fano surface parametrizing lines on X , then F .‚/ is a geometric GV-sheaf, i.e.
RˆP .F .‚// is perverse (cf. [28], Theorem 1.2).
Note also that further results along these lines in the case of Prym varieties can be
found in [10].
What the Wd ’s and F have in common is that they are subvarieties whose coho-
gd
mology classes are minimal (i.e. not divisible in H .A; Z/). Indeed, ŒWd D .gd /Š
3
and ŒF D 3Š by results of Poincaré and Clemens–Griffiths respectively. An problem
stemming from work of Beauville and Ran in low dimensions, and suggested in gen-
eral by Debarre [16], is to show that these are in fact the only subvarieties of ppav’s
representing minimal cohomology classes. In [40], Conjecture A, it is proposed that
this should also be equivalent to the fact that RˆP .X .‚// is a perverse sheaf. One
direction is known: in [40], Theorem B, it is shown that if X is a nondegenerate closed
reduced subscheme of pure dimension d of a ppav .A; ‚/ of dimension g and X .‚/
gd
is a GV-sheaf, then X is Cohen–Macaulay and ŒX D .gd /Š
. The main ingredient is
the criterion provided by Theorem 4.1 above.
6.5 Moduli spaces of vector bundles. The most natural higher rank analogue of
generic vanishing involving Pic0 .X / is to consider the (singular) moduli space MX .r/
of semistable rank r vector bundles with trivial Chern classes on a smooth projective X .
Some bounds on the dimension of cohomological support loci in MX .r/ are given by
Arapura [1], §7, but a more thorough understanding remains a very interesting problem.
(Note however that the structure of these loci is very well understood in [1], by means
of non-abelian Hodge theory.)
On the other hand, generic vanishing statements can be considered whenever M is
a fine moduli space of objects on X, with a universal object E on X M inducing the
functor
RÊ W D.X / ! D.M /:
In other words, the problem is to study the variation of the cohomology of the objects
parametrized by M . In practice, in order to apply Theorem 4.1 the main difficulty to be
b
overcome is a good understanding of the transform A1 D R‰E .A1 /Œdim M , with
A a very positive line bundle on M . Few concrete examples seem to be known: besides
the case of curves, there are only sparse examples of generic vanishing phenomena on
moduli spaces in higher dimension, coming from constructions of Yoshioka on K3
and abelian surfaces and of Bridgeland–Maciocia on threefolds with fibrations by such
surfaces. I finish with a very brief discussion of these.
Curves. The issue whether RÊ !X is a perverse sheaf on moduli spaces of higher
rank vector bundles on a curve X is not so interesting, as it is easy to check by hand.
However, Theorem 4.1 is useful when applied to bundles other than !X . For instance,
when applied to higher rank stable bundles, it links the entire indeterminacy loci of
determinant line bundles on these moduli spaces to a well-known construction of Ray-
naud, as first considered by Hein [27]. This is of a somewhat different flavor from
the main direction of this note, so I will refrain from including details. The interested
reader can consult §7 in [39].
Some surfaces and threefolds. Consider first X to be a complex abelian or K3 surface.

For a coherent sheaf E on X, the Mukai vector of E is
v.E/ WD rk.E/ C c1 .E/ C ..E/ rk.E//ŒX 2 H ev .X; Z/;
where is 0 if X is abelian and 1 if X is K3. Given a polarization H on X and a vector

v 2 H ev .X; Z/, we consider the moduli space MH .v/ of sheaves E with v.E/ D v,
stable with respect to H . If the Mukai vector v is primitive and isotropic, and H is
general, the moduli space is M D MH .v/ is smooth, projective and fine, and it is in
fact again an abelian or K3 surface (cf. e.g. [47]). The universal object E on X M
induces an equivalence of derived categories RÊ W D.X / ! D.M /. Yoshioka gives
many examples that amount to satisfying:
The Mukai vector v is primitive and isotropic, and the structure sheaf
.
/ OX .Š !X / satisfies WIT2 with respect to RÊ _ (or equivalently is GV
with respect to RÊ by Theorem 4.1).
• Let .X; H / be a polarized K3 surface such that Pic.X / D Z H , with H 2 D 2n.
Let k > 0 be an integer such that kH is very ample. Consider v D k 2 n C kH C ŒX .
It is shown in [49], Lemma 2.4, that under these assumptions .
/ is satisfied.7
• Let .X; H / be a polarized abelian surface with Pic.X / D ZH . Write H 2 D 2r0 k,
with .r0 ; k/ D 1. Consider the Mukai vector v0 WD r0 C c1 .H / C kŒX . By [48],
Theorem 2.3, and the preceding remark, in this situation .
/ is again satisfied by OX
(among many other examples).
A Calabi–Yau fibration is a morphism
W X ! S of smooth projective varieties,
with connected fibers, such that KX C D 0 for all curves C contained in fibers of
.
7
It is also shown in §2 of loc. cit. that in fact X Š M .
274 M. Popa
Assuming that X is a threefold, it is an elliptic, abelian surface, or K3-fibration (in

the sense that the nonsingular fibers are of this type). Say
is flat, and consider a
polarization H on X , and Y an irreducible component of the relative moduli space
M H;P .X=S/ of sheaves on X (over S), semistable with respect to H , and with fixed
Hilbert polynomial P . The choice of P induces on every smooth fiber Xs invariants
which are equivalent to the choice of a Mukai vector v 2 H ev .Xs ; Z/ as above. Assum-
ing that Y is also a threefold, and fine, Bridgeland and Maciocia (cf. [8], Theorem 1.2)
proved that it is smooth, and the induced morphism
O W Y ! S is a Calabi–Yau fibration
of the same type as
. If E is a universal sheaf on X Y , then RÊ W D.X / ! D.Y /
is an equivalence. Now if the first half of (
) is satisfied for each smooth fiber of
, it
is proved in [8], §7, that the moduli space M H;P .X=S / does have a fine component
Y which is a threefold, so the above applies. The same method based on Theorem 4.1
leads to the following:
Proposition 6.16 ([39], Proposition 7.7). Let X be a smooth projective threefold with
a smooth Calabi–Yau fibration
W X ! S of relative dimension two. Let H be
a polarization on X and P a Hilbert polynomial, and assume that condition .
/ is
satisfied for each fiber of
. Consider a fine three-dimensional moduli space component
Y M H;P .X=S /, and let E be a universal sheaf on X Y . Then !X is a GV1 -sheaf
with respect to E. In particular
H i .X; !X ˝ E/ D 0; for all i > 1 and all E 2 Y general:
This can be extended to the case of singular fibers, involving a slightly technical
condition which should nevertheless be often satisfied; see [39], Remark 7.8. A similar
statement holds for threefold elliptic fibrations.
7 Appendix: some homological commutative algebra

A useful technical point is that over Cohen–Macaulay rings one can avoid finite homo-
logical dimension hypotheses in statements of Auslander–Buchsbaum-type, by involv-
ing the canonical module. The key fact is the following consequence of Grothendieck
duality for local cohomology.
Lemma 7.1 ([9], Corollary 3.5.11). Let .R; m/ be a local Cohen–Macaulay ring of
dimension n with canonical module !R , and let M be a finitely generated module over
R of depth t and dimension d . Then:
(a) ExtiR .M; !R / D 0 for i < n d and i > n t .
(b) ExtiR .M; !R / ¤ 0 for i D n d and i D n t .
(c) dim ExtiR .M; !R / n i for all i 0.
Corollary 7.2. Let F be a coherent sheaf on a Cohen–Macaulay scheme of finite type
over a regular local ring, or a complex analytic space X . Then:
(a) Ext i .F ; !X / D 0 for all i < codim Supp F and Ext i .F ; !X / ¤ 0 for i D
codim Supp F .
(b) codim Supp Ext i .F ; !X / i for all i .
(c) If F is locally free, then Ext i .F ; !X / D 0 for all i > 0.
Let now X be a noetherian scheme of finite type over a field, or a complex analytic
space.
Definition 7.3. Let F be a coherent sheaf on X . Then F is called a k-th syzygy sheaf
if locally there exists an exact sequence
0 ! F ! Ek ! ! E1 ! G ! 0 (2)
with Ej locally free for all j . It is well known for example that if X is normal, then
1-st syzygy sheaf is equivalent to torsion-free, and 2-nd syzygy sheaf is equivalent to
reflexive. Every coherent sheaf is declared to be a 0-th syzygy sheaf. A locally free
one is declared to be an 1-syzygy sheaf.
Following [3] and [20], we consider a variant of Serre’s condition Sk .
Definition 7.4. A coherent sheaf F on X satisfies property Sk0 if for all x in the support
of F we have:
depth Fx minfk; dim OX;x g:
Proposition 7.5. Let X be Cohen–Macaulay, and let F be a coherent sheaf on X .
Consider the following conditions:
(a) F is a k-th syzygy sheaf.

(b) codim Supp Ext i .F ; !X / i C k; for all i > 0:
(c) F satisfies Sk0 .
Then (b) is equivalent to (c), and if in addition F has finite homological dimension
or X is Gorenstein in codimension less than or equal to one, they are both equivalent
to (a).
Proof. When F is of finite homological dimension or X is Gorenstein in codimension
less than or equal to one, the equivalence between (a) and (c) is a well-known result
of Auslander–Bridger, [3], Theorem 4.25. The equivalence between (b) and (c) is
also standard in the case of finite homological dimension, when it follows from the
Auslander–Buchsbaum formula (cf. [29], Proposition 1.1.6 (ii)). We note here that a
variation of the usual argument, involving Lemma 7.1, proves it in general. Consider
first x 2 Supp Ext i .F ; !X / for some i > 0. Then by Lemma 7.1 (a) we get that
i dim OX;x depth Fx . Combined with Sk0 , this implies dim OX;x i C k. On
276 M. Popa
the other hand, consider x 2 Supp F . We need to show that dim OX;x depth Fx
maxfdim OX;x k; 0g. But by Lemma 7.1 (a) and (b), we have
dim OX;x depth Fx D maxfp j ExtpOX;x .Fx ; !X;x / ¤ 0g:
But for any such p we have x 2 Supp Ext p .F ; !X /. Hence either the maximum is 0
and we’re done, or it is positive and for all such p > 0 we have dim OX;x p C k.
Finally, a key result here is the Syzygy Theorem of Evans–Griffith.
Theorem 7.6 ([20], Corollary 1.7). Let X be a Cohen–Macaulay scheme over a field,
and F a k-th syzygy sheaf of finite homological dimension on X which is not locally
free. Then rank.F / k.
References
[1] D. Arapura, Geometry of cohomology support loci II: integrability of Hitchin’s map.
Preprint, arXiv:alg-geom/9701014.
[2] E. Arbarello, M. Cornalba, P. A. Griffiths, and J. Harris, Geometry of algebraic curves.
Grundlehren Math. WIss. 267, Springer-Verlag, Berlin 1985.
[3] M. Auslander and M. Bridger, Stable module theory. Mem. Amer. Math. Soc. 94, Amer.
Math. Soc., Providence, R.I., 1969.
[4] D. Arinkin and R. Bezrukavnikov, Perverse coherent sheaves. Mosc. Math. J. 10 (2010),
no .1, 3–29.
[5] M. A. Barja, M. Lahoz, J. C. Naranjo, and G. Pareschi, On the bicanonical map of irregular
varieties. J. Algebraic Geom. 21 (2012), no. 3, 445–471.
[6] A. Bayer, Polynomial Bridgeland stability conditions and the large volume limit. Geom.
Topol. 13 (2009), no. 4, 2389–2425.
[7] R. Bezrukavnikov, Perverse coherent sheaves (after Deligne). Preprint,
arXiv:math/0005152 [math.AG].
[8] T. Bridgeland and A. Maciocia, Fourier-Mukai transforms for K3 and elliptic fibrations. J.
Algebraic Geom. 11 (2002), no. 4, 629–657.
[9] W. Bruns and J. Herzog, Cohen-Macaulay rings. Cambridge Stud. Adv. Math. 39, Cam-
bridge University Press, Cambridge 1993.
[10] S. Casalaina-Martin, M. Lahoz, and F. Viviani, Cohomological support loci for Abel-Prym
curves. Matematiche (Catania) 63 (2008), no. 1, 205–222.
[11] J. A. Chen and Ch. Hacon, Characterization of abelian varieties. Invent. Math. 143 (2001)
no. 2, 435–447.
[12] J.A. Chen and Ch. Hacon, Pluricanonical maps of varieties of maximalAlbanese dimension.
Math. Ann. 320 (2001), no. 2, 367–380.
[13] J. A. Chen and Ch. Hacon, On algebraic fiber spaces over varieties of maximal Albanese
dimension. Duke Math. J. 111 (2002), no. 1, 159–175.
[14] J. A. Chen and Ch. Hacon, Kodaira dimension of irregular varieties. Invent. Math. 186
(2011), no. 3, 481–500.
[15] H. Clemens and Ch. Hacon, Deformations of flat line bundles and their metrics. Amer. J.
Math. 124 no.4 (2002), 769–815.
[16] O. Debarre, Minimal cohomology classes and Jacobians. J. Algebraic. Geom. 4 (1995),
321–335.
[17] P. Deligne and L. Illusie, Relèvements modulo p 2 et décomposition du complexe de de
Rham. Invent. Math. 89 (1987), 247–270.
[18] J.-M. Drezet and M. S. Narasimhan, Groupe de Picard des variétés des modules de fibrés
semi-stables sur les courbes algébriques. Invent. Math. 97 (1989), 53–94.
[19] L. Ein and R. Lazarsfeld, Singularieties of theta divisors and the birational geometry of
irregular varieties. J. Amer. Math. Soc. 10 (1997), 243–258.
[20] E. G. Evans and P. Griffith, The syzygy problem. Ann. of Math. 114 (1981), 323–333.
[21] M. Green and R. Lazarsfeld, Deformation theory, generic vanishing theorems, and some
conjectures of Enriques, Catanese and Beauville. Invent. Math. 90 (1987), 389–407.
[22] M. Green and R. Lazarsfeld, Higher obstructions to deforming cohomology groups of line
bundles. J. Amer. Math. Soc. 1 (1991), no. 4, 87–103.
[23] A. Grothendieck and J. Dieudonné, Eléments de Géométrie Algébrique, III, Étude coho-
mologique des faisceaux coherents. Inst. Hautes Études Sci. Publ. Math. 11 (1961) and 17
(1963).
[24] Ch. Hacon, A derived category approach to generic vanishing. J. Reine Angew. Math. 575
(2004), 173–187.
[25] C. Hacon and R. Pardini, On the birational geometry of varieties of maximal Albanese
dimension. J. Reine Angew. Math. 546 (2002), 177–199.
[26] R. Hartshorne, Residues and duality. Lecture Notes in Math. 20, Springer-Verlag, Berlin
1966.
[27] G. Hein, Raynaud’s vector bundles and base points of the generalized Theta divisor. Math.
Z. 257 (2007), no. 3, 597–611.
[28] A. Höring, M -regularity of the Fano surface. C. R. Math. Acad. Sci. Paris 344 (2007),
691–696.
[29] D. Huybrechts and M. Lehn, The geometry of moduli spaces of sheaves. Aspects Math.
E31, Vieweg, Braunschweig 1997.
[30] M. Kashiwara, T -structures on the derived category of holonomic D-modules and coherent
O-modules. Mosc. Math. J. 4 (2004), no. 4, 847–868.
[31] J. Kollár, Higher direct images of dualizing sheaves II. Ann. of Math. 124 (1986), 171–202.
[32] R. Lazarsfeld, Positivity in algebraic geometry I & II. Ergeb. Math. Grenzgeb. 48 & 49,
Springer-Verlag, Berlin 2004.
[33] S. Meinhardt, Stability conditions on generic complex tori. Internat. J. Math. 23 (2012),
no. 5, 1250035.
[34] S. Mukai, Duality between D.X / and D.Xy / with its application to Picard sheaves. Nagoya
Math. J. 81 (1981), 153–175.
278 M. Popa
[35] D. Mumford, Abelian varieties. Second edition, Tata Inst. Fund. Res. Studies in Math. 5,
Oxford University Press, Oxford 1974.
[36] G. Pareschi, Generic vanishing, Gaussian maps, and Fourier-Mukai transform. Preprint,
[37] G. Pareschi, Basic results on irregular varieties via Fourier-Mukai methods. In Current
developments in algebraic geometry, Math. Sci. Res. Inst. Publ. 59, Cambridge University
Press, Cambridge 2012, 379–403.
[38] G. Pareschi and M. Popa, Regularity on abelian varieties I. J. Amer. Math. Soc. 16 (2003),
285–302.
[39] G. Pareschi and M. Popa, GV-sheaves, Fourier-Mukai transform, and generic vanishing.
Amer. J. Math. 133 (2011), no. 1, 235–271.
[40] G. Pareschi and M. Popa, Generic vanishing and minimal cohomology classes on abelian
varieties. Math. Ann. 340 (2008), no. 1, 209–222.
[41] G. Pareschi and M. Popa, Strong generic vanishing and a higher dimensional Castelnuovo-
de Franchis inequality. Duke Math. J. 150 (2009), no. 2, 269–285.
[42] M. Popa and C. Schnell. Generic vanishing theory via mixed Hodge modules. Preprint,
[43] M. Saito, Mixed Hodge modules. Publ. Res. Inst. Math. Sci. 26 (1990), no. 2, 221–333.
[44] Y. Toda, Limit stable objects on Calabi-Yau 3-folds. Duke Math. J. 149 (2009), no. 1,
157–208.
[45] Y. Toda, Curve counting theories via stable objects I. DT/PT correspondence. J. Amer. Math.
Soc. 23 (2010), no. 4, 1119–1157.
[46] A.Yekutieli and J. J. Zhang, Dualizing complexes and perverse sheaves on noncommutative
ring schemes. Selecta Math. 12 (2006), 137–177.
[47] K.Yoshioka, Chamber structure of polarizations and the moduli of stable sheaves on a ruled
surface. Internat. J. Math. 7 (1996), 411–431.
[48] K.Yoshioka, Some examples of isomorphisms induced by Fourier-Mukai functors. Preprint,
[49] K. Yoshioka, Stability and the Fourier-Mukai transform I. Math. Z. 245 (2003), 657–665.
The fundamental group is not a derived invariant
Christian Schnell
This short paper is based on my talk at the conference Derived Categories Tokyo 2011,
and I am very grateful to Yujiro Kawamata and Yukinobu Toda for inviting me to the
conference. I also thank Lev Borisov for suggesting the example to me in the first
place, and Andrei Căldăraru and Sukhendu Mehrotra for useful discussions.
Some time after giving the talk, I discovered that Anthony Bak [2] already has a
preprint on the arXiv in which he obtains the same result (that is to say, Theorem 3.1
below). In fact, his proof is more concrete, and therefore much more useful for doing
calculations with the derived equivalence. My apology for nevertheless writing this
note is that the proof given here is different and, by relying on the theorem of Bridgeland
and Maciocia [4], a little bit shorter than Bak’s.
1 Introduction
For a smooth complex projective variety X, we denote by D b .X / D D b Coh.X / the
bounded derived category of coherent sheaves on X . Recall that two smooth projective
varieties X are Y are said to be derived equivalent if D b .X / ' D b .Y / as C-linear
triangulated categories. We sometimes write X Y to indicate that X and Y are
derived equivalent.
From the work of Bondal, Orlov, Căldăraru, Kawamata, and others, it is known
that many of the basic invariants of algebraic varieties are preserved under derived
equivalence. These include the dimension, the Kodaira dimension, the canonical ring,
and the order of the canonical class. It has also been conjectured that the Hodge
structure on the cohomology with rational coefficients is a derived invariant, in the
sense that if X Y , then one should have H k .X; Q/ ' H k .Y; Q/ as rational Hodge
structures, for every k 2 Z. In particular, it is expected that the Hodge numbers
p
hp;q .X / D dimC H p;q .X / D dimC H q X; X
are invariant under derived equivalences.

In joint work with Mihnea Popa [10], we showed that if X and Y are derived
equivalent, then H 1 .X; Q/ ' H 1 .Y; Q/ as rational Hodge structures; in geometric

This work was supported by the World Premier International Research Center Initiative (WPI Initiative),
MEXT, Japan, and by NSF grant DMS-1100606.
280 C. Schnell
terms, this means that the two Picard varieties Pic0 .X / and Pic0 .Y / are isogenous
abelian varieties. Ignoring the choice of base point,

H 1 .X; Q/ ' HomZ 1 .X /; Q ;
and so our result naturally leads to the question whether the fundamental group 1 .X /
is itself a derived invariant. The point of this paper is to show that this is not the case.
More precisely, I will describe an example of a simply connected Calabi–Yau
threefold X, with a nontrivial free action by a finite group G, such that the quotient
X=G is derived equivalent to X. Since 1 .X=G/ D G, while 1 .X / D f1g, this
means that neither the fundamental group, nor the property of being simply connected,
are preserved under derived equivalence.
2 A related problem
Before continuing, I should point out that this result is connected to a larger question
raised by Daniel Huybrechts and Marc Nieper-Wisskirchen, about derived equivalences
of varieties with trivial first Chern class. To set up some notation, suppose that X is
a smooth projective variety whose first Chern class c1 .X / is zero as an element of
H 2 .X; R/. By Yau’s theorem, X admits a Ricci-flat Kähler metric; by studying the
holonomy of this metric, Bogomolov and Beauville [3] have shown that a finite étale
cover X 0 ! X decomposes into a finite product
Y Y
X0 ' A Yi Zj
i j
with A an abelian variety, each Yi a simply connected Calabi–Yau manifold of dimen-

sion at least three, and each Zj a holomorphic symplectic manifold. Huybrechts and
Nieper-Wisskirchen ask whether the structure of this decomposition is invariant under
derived equivalences (see [7], Question 0.2).
In the special case of Calabi–Yau threefolds, the question becomes the following:
Suppose that X is a simply connected Calabi–Yau threefold, and that Y X. Because
of the example in this paper, we cannot expect Y to be simply connected. On the other
hand, the first Chern class of Y is also trivial, and so a finite étale cover of Y must be
of one of the following three types:
(1) A simply connected Calabi–Yau threefold.
(2) An abelian threefold.
(3) The product of an elliptic curve and a K3-surface.
Note that there are examples of finite quotients of abelian threefolds (or products of
an elliptic curve and a K3-surface) with trivial canonical bundle and zero first Betti
number; a partial classification may be found in [9]. Nevertheless, it seems likely
The fundamental group is not a derived invariant 281
that such varieties cannot be derived equivalent to a simply connected Calabi–Yau

threefold. A proof of this would be a useful step towards answering the general question
of Huybrechts and Nieper-Wisskirchen.
3 The example
Let us now turn to the description of the example, which was suggested to me by
Lev Borisov. The Calabi–Yau threefold in question is one of a class of such varieties
constructed by Mark Gross and Sorin Popescu [6], and has to do with .1; 8/-polarized
abelian surfaces.
We shall begin by recalling their construction. Let .A; L/ be a .1; 8/-polarized
abelian surface. In other words, suppose that A is an abelian surface, and that L is an
ample line bundle on A such that the isogeny
'L W A ! Pic0 .A/; a 7! ta .L/ ˝ L1 ;
has kernel isomorphic to Z8 Z8 .1 One can show that L is then automatically very
ample with h0 .L/ D 8; by the Riemann–Roch theorem, it follows that we have 8 D
.L/ D L2 =2, which gives L2 D 16. The line bundle therefore embeds A as a surface
of degree 16 into P 7 , and it is possible to choose the coordinates on the projective space
in such a way that the action of G D Z8 Z8 on P 7 is given by the formulas
.x0 W x1 W W x6 W x7 / D .x1 W x2 W W x7 W x0 /;

.x0 W x1 W W x6 W x7 / D x0 W x1 W W 6 x6 W 7 x7 :
Here and denote the two natural generators of the group G, and is a primitive
eighth root of unity.
The idea of Gross and Popescu is to look at quadrics in P 7 that contain the image
of A. Provided that the pair .A; L/ is general in moduli, they show that the space of
such quadrics has dimension four, and that it is generated by the four quadrics f , f ,
2 f , and 3 f , where
f D y1 y3 .x02 C x42 / y22 .x1 x7 C x3 x5 / C .y12 C y32 /x2 x6 ;
and y 2 P 2 is a general point. The intersection
V8;y D Z.f / \ Z.f / \ Z. 2 /f \ Z. 3 f /
is then a threefold on which the group G acts freely. Conversely, if we assume that
y 2 P 2 is chosen sufficiently general, V8;y will be a complete intersection of dimension
three which is smooth except for 64 ordinary double points, the G-orbit of the point
.0 W y1 W y2 W y3 W 0 W y3 W y2 W y1 /. There is always a one-dimensional family
˚2
1
In general, a polarization L is of type .m; n/ if the kernel of 'L is isomorphic to the group Zm Zn
for positive integers mjn.
282 C. Schnell
of .1; 8/-polarized abelian surfaces contained in V8;y , and every member of the family
passes through the 64 distinguished points.
Gross and Popescu discovered that V8;y admits two small resolutions V8;y 1
and V8;y
2
,
both Calabi–Yau threefolds. The original abelian surface A is a Weil divisor on V8;y
that is not Cartier; blowing up A produces a small resolution V8;y 2
! V8;y . Using the
Lefschetz theorem and adjunction, one can easily show that V8;y is a simply connected
2
Calabi–Yau threefold. The 64 exceptional curves can be flopped simultaneously to

produce another simply connected Calabi–Yau threefold V8;y 1
, and Gross and Popescu
compute that
h1;1 .V8;y
1
/ D h1;2 .V8;y
1
/ D 2:
In fact, the Picard group of V8;y
1
is generated (modulo torsion) by the classes of two
divisors: the strict transform A of the original abelian surface, and the preimage H of
a hyperplane section. They satisfy
H 3 D 16; H 2 A D 16; H A2 D 0; A3 D 0:
The reader can find a concise summary of all the properties of the Calabi–Yau
threefolds V8;y
1
and V8;y
2
in [5]. We shall only list those that are needed below.
(1) The linear system jAj is one-dimensional, and the resulting morphism
p W V8;y
1
! P1
is an abelian surface fibration with exactly 64 sections. The images of these

sections are the 64 exceptional curves of the flop.
(2) Every smooth fiber of p is a .1; 8/-polarized abelian surface, with polarization
induced by the restriction of the line bundle OX .H /. The intersection with the
images of the 64 sections is precisely the kernel of the polarization.
(3) There are exactly eight singular fibers, each of them an elliptic translation scroll.
Such a scroll is obtained from an elliptic normal curve E in P 7 by fixing a point
e 2 E, and letting Te .E/ be the union of all lines through x and x Ce, for x 2 E.
It is not hard to see that Te .E/ is singular precisely along the elliptic curve E,
and that E P 1 is a resolution of singularities.
(4) The group G acts freely on V8;y 1
, and the 64 sections form a single G-orbit.
On each smooth fiber, the action of G is the natural action by the kernel of the
polarization; on each singular fiber, the action of G is the natural action by the
group of 8-torsion points of the elliptic curve.
Now let X D V8;y 1
. Since G D Z8 Z8 acts freely on X , the quotient X=G is
again a smooth projective variety with trivial canonical bundle. The following theorem
is the main result of this paper.
Theorem 3.1. The two varieties X and X=G are derived equivalent.
Since X is simply connected, while the quotient X=G has fundamental group
isomorphic to G, it follows that the fundamental group is not invariant under derived
equivalences.
Note 3.2. For reasons coming from physics, Mark Gross and Simone Pavanelli [5]
conjecture that the quotient of X by one of the two Z8 -factors of G should be the
mirror manifold of X , and that X=G should be the mirror of the mirror. Homological
mirror symmetry would therefore predict that D b .X / ' D b .X=G/.
4 Proof of the theorem

We now describe one possible proof of Theorem 3.1. An earlier proof, more concrete
but also slightly longer, may be found in the preprint by Anthony Bak [2].
To explain the basic idea, let us consider one of the smooth fibers A of the morphism
p W X ! P 1 ; it is a .1; 8/-polarized abelian surface, with polarization L induced by
the restriction of H . The group G acts on A, and the 64 points in the kernel of the
isogeny
'L W A ! Pic0 .A/
form a single G-orbit. Consequently, 'L gives rise to an isomorphism A=G ' Pic0 .A/.
Moreover, once we choose one of the 64 points in the kernel as the unit element on
A, there is a well-defined Poincaré line bundle on A Pic0 .A/, and the associated
Fourier–Mukai transform induces an equivalence D b .A/ ' D b .Pic0 .A//. In our
proof, we shall generalize these observations by (1) using the theorem of Bridgeland
and Maciocia to prove that X is derived equivalent to the compactified relative Picard
scheme M , and (2) showing that M is actually isomorphic to X=G.
We begin by introducing the space M . Let s0 be one of the 64 sections of p W X !
P . The general fiber of p is an abelian surface, and the eight singular fibers are
1
elliptic translation scrolls, and therefore reduced and irreducible. Thus it makes sense
to consider the compactified relative Picard scheme
M D Pic 0 .X=P 1 /
defined by Altman and Kleiman [1]. For any smooth fiber A of X ! P 1 , the cor-
responding fiber of M ! P 1 is also smooth and isomorphic to Pic0 .A/. In general,
the compactified relative Picard scheme may be singular, and may fail to be a fine
moduli space because the ambiguity in normalizing the Poincaré bundle can prevent
the existence of a universal sheaf. But in our case, everything works out nicely.
Lemma 4.1. M is a nonsingular projective Calabi–Yau threefold. Moreover, a uni-
versal sheaf exists on M X, and induces an equivalence D b .M / ' D b .X /.
Proof. To begin with, Sawon [11], Lemma 8, has shown that M is projective, because it
is an irreducible component of Simpson’s moduli space of stable rank-one torsion-free
sheaves on X. Next, the existence of a section s0 implies that there is a universal sheaf
284 C. Schnell
on M P 1 X M X . Indeed, because X is nonsingular, the image of s0 has to

lie inside the smooth locus of p, and so we can apply [1], Theorem 3.4, to obtain the
existence of a universal sheaf. In particular, M is a fine moduli space.
The theorem of Bridgeland and Maciocia [4], Theorem 1.2, now allows us to con-
clude that M is also a nonsingular Calabi–Yau threefold, and that the universal sheaf
induces an equivalence between the derived categories of M and X .
The remainder of the proof consists in showing that M is, in fact, isomorphic to the
quotient X=G. Our argument is based on the observation, explained above, that the
smooth fibers of X=G and M are canonically isomorphic. The main issue is to extend
this isomorphism to the singular fibers.
We begin by constructing a rational map from X=G to M , using the universal
property of the fine moduli space M . The idea is the following: Let .A; L/ be a
polarized abelian variety, and let PA denote the normalized Poincaré bundle on A
Pic0 .A/. The pullback of PA under the morphism
id 'L W A A ! A Pic0 .A/
satisfies
.id 'L / PA ' m L ˝ pr 1 L1 ˝ pr 2 L1 ; (1)
where m W A A ! A is the multiplication on A. This allows us to describe the
morphism 'L to the moduli space Pic0 .A/ in terms of a line bundle on A A.
To extend this construction to X , let B P 1 be the complement of the eight singular
values of p, and set U D p 1 .B/. Then p W U ! B is smooth, and our choice of
section s0 makes it into a group scheme over B. We also denote the multiplication
morphism by m W U B U ! U .
Lemma 4.2. There is a morphism f W U=G ! M , commuting with the projections

to P 1 , whose restriction to any smooth fiber agrees with the natural isomorphism
A=G ! Pic0 .A/ induced by the .1; 8/-polarization OX .H /jA .
Proof. We begin by constructing a morphism U ! M whose restriction to every

smooth fiber A agrees with the natural morphism 'L W A ! Pic0 .A/ induced by the
polarization L D OX .H /jA . By the universal property of M , it suffices to construct
a line bundle on the product U B U whose restriction to A A is isomorphic to the
pullback of the Poincaré bundle PA . But clearly
m OX .H / ˝ pr 1 OX .H / ˝ pr 2 OX .H /
is such a line bundle, by virtue of (1). The same identity shows that this line bundle is
invariant under the action of G on the second factor of U B U , and therefore descends
to a line bundle on U B .U=G/ whose restriction to A .A=G/ equals the pullback of
PA . The universal property of M now gives us the desired morphism f W U=G ! M .
In particular, f is an isomorphism onto its image, and so X=G and M are birational.
This is already sufficient to conclude that 1 .M / ' G; but in fact, we can use the
geometry of both varieties to show that they must be isomorphic.
Lemma 4.3. f W U=G ! M extends to an isomorphism X=G ' M .
Proof. We consider f as a birational map from X=G to M . Since both are smooth
Calabi–Yau threefolds, any birational map between X=G and M is either an isomor-
phism, or a composition of flops [8], Theorem 5.3. The second possibility is easily
ruled out: Indeed, since f is an isomorphism over U , the exceptional locus is con-
tained in the eight singular fibers; moreover, it must be a union of rational curves [8],
Lemma 4.3. Now each singular fiber is an elliptic translation scroll, which means that
the only rational curves on it are the one-dimensional family of lines on the scroll. Since
these lines cover the singular fiber, which is a divisor in X=G, they cannot be flopped.
Consequently, the birational map f must extend to an isomorphism X=G ' M .
Note 4.4. Following the argument in [5], Lemma 1.2, one can show more generally
that X=G does not admit any flops relative to P 1 at all.
References
[1] A. B. Altman and S. L. Kleiman, Compactifying the Picard scheme. II. Amer. J. Math. 101
(1979), no. 1, 10–41.
[2] A. Bak, The spectral construction for a .1; 8/-polarized family of abelian varieties. Preprint,
[3] A. Beauville, Variétés Kähleriennes dont la première classe de Chern est nulle. J. Differential
Geom. 18 (1983), no. 4, 755–782.
[4] T. Bridgeland and A. Maciocia, Fourier-Mukai transforms for K3 and elliptic fibrations.
J. Algebraic Geom. 11 (2002), no. 4, 629–657.
[5] M. Gross and S. Pavanelli, A Calabi-Yau threefold with Brauer group .Z=8Z/2 . Proc. Amer.
Math. Soc. 136 (2008), no. 1, 1–9.
[6] M. Gross and S. Popescu, Equations of .1; d /-polarized abelian surfaces. Math. Ann. 310
(1998), no. 2, 333–377.
[7] D. Huybrechts and M. Nieper-Wisskirchen, Remarks on derived equivalences of Ricci-flat
manifolds. Math. Z. 267 (2011), no. 3–4, 939–963.
[8] Y. Kawamata, Crepant blowing-up of 3-dimensional canonical singularities and its applica-
tion to degenerations of surfaces. Ann. of Math. (2) 127 (1988), no. 1, 93–163.
[9] K. Oguiso and J. Sakurai, Calabi-Yau threefolds of quotient type. Asian J. Math. 5 (2001),
no. 1, 43–77.
[10] M. Popa and C. Schnell, Derived invariance of the number of holomorphic 1-forms and
vector fields. Ann. Sci. Éc. Norm. Supér. (4) 44 (2011), no. 3, 527–536.
[11] J. Sawon, Twisted Fourier-Mukai transforms for holomorphic symplectic four-folds. Adv.
Math. 218 (2008), no. 3, 828–864.
Introduction and open problems of Donaldson–Thomas
theory
Yukinobu Toda
1 Introduction
The Donaldson–Thomas (DT) invariant is roughly speaking a counting invariant of
holomorphic vector bundles on a Calabi–Yau 3-fold, which was introduced in [31] in
order to introduce a holomorphic analogue of the Casson invariants on real 3-manifolds.
Since then it has drawn much attention, especially its connection with Gromov–Witten
(GW) invariants, Bridgeland stability conditions and wall-crossing phenomena in the
derived category. The purpose of this article is to give an introduction to DT theory, a
survey of its recent developments, and discuss some open problems.
The DT theory requires a compactification of the moduli space of holomorphic
vector bundles on a Calabi–Yau 3-fold. Under a perturbation of (almost) complex
structures, that moduli space may be a finite number of points, hence we may count
them. However at this moment, such a differential geometric argument is not available
due to the difficulties of singularities of degenerated vector bundles. On the other hand,
there is a good algebro geometric way in constructing a compactification of the moduli
space of vector bundles via (semi)stable coherent sheaves. Instead of a perturbation of
complex structures, we can construct the zero dimensional virtual fundamental cycle
on the moduli space of stable sheaves. The DT invariant is defined to be the integration
of the virtual cycle. The invariant can be also defined using Behrend’s micro local
function, which is useful in several applications.
If we focus on the rank one DT theory, then the moduli spaces defining the invari-
ants are classical Hilbert schemes of curves, and the associated invariants correspond
to curve counting invariants on Calabi–Yau 3-folds. In this case, the DT invariants
are expected to be related to the GW invariants by Maulik–Nekrasov–Okounkov–
Pandharipande (MNOP) [27].
By its definition, the DT theory depends on the choice of a stability condition on the
category of coherent sheaves. Classically a stability condition is given by the choice of
an ample divisor on a Calabi–Yau 3-fold. The behavior of DT invariants under change
of stability conditions is described by the wall-crossing formula, whose general theory
is established by Joyce–Song [21] and Kontsevich–Soibelman [24]. In rank one DT
theory, there is no interesting wall-crossing phenomena under change of ample divisors.
However we can extend the theory to counting stable objects in the derived category
w.r.t. Bridgeland’s stability conditions [10], (or their generalization,) and observe wall-
crossing phenomena for rank one invariants in the derived category [28], [37], [1].
It turns out that the above derived category wall-crossing is relevant to the MNOP
conjecture. We will review these developments.
288 Y. Toda
The organization of this paper is as follows. In Section 2, we recall the definition

of DT invariants. In Section 3 we focus on rank one theory and discuss the MNOP
conjecture. In Section 4, we review the recent results on the generating series of rank
one DT type invariants. In Section 5, we discuss some open problems related to the
DT theory.
2 Donaldson–Thomas theory
In this section, we recall the definition of Donaldson–Thomas (DT) invariants and their
properties. In what follows, X is a smooth projective Calabi–Yau 3-fold over C, i.e.
^
3
TX_ Š OX ; H 1 .X; OX / D 0;
e.g. a quintic 3-fold X PC4 .
2.1 (Semi)stable sheaves. The DT invariant is constructed via the moduli space of
holomorphic vector bundles on X . A holomorphic vector bundle is naturally interpreted
as a locally free sheaf on X , and the moduli space may be compactified by adding non-
locally free sheaves. In general the moduli space of sheaves is non-separated, non-finite
type, even if we fix a numerical class of sheaves. By choosing a stability condition, we
can construct a ‘good’ moduli space of sheaves on X . Classically a stability condition
is given by the choice of an ample line bundle on X.
Let L be an ample line bundle on X and E a coherent sheaf on X . Recall that the
Hilbert polynomial of E is defined by
.E ˝ L˝n / D ad nd C ad 1 nd 1 C ;
where ad ¤ 0 and d is the dimension of the support of E. Note that by the Riemann–
Roch theorem, the Hilbert polynomial only depends on the Chern character of E, i.e.
there is a map
L; .n/ W H .X; Q/ ! QŒn;
such that we have
.E ˝ L˝n / D L;ch.E / .n/:
Below we write L;E .n/ D L;ch.E / .n/ for simplicity. The reduced Hilbert polynomial
of E is defined by
NL;E .n/ :D L;E .n/=ad :
Definition 2.1. A coherent sheaf E on X is L-(semi)stable if the following conditions

hold:
Introduction and open problems of Donaldson–Thomas theory 289
• E is a pure sheaf, i.e. there is no subsheaf E 0 E with dim Supp.E 0 / <

dim Supp.E/.
• For any subsheaf E 0 E, we have the inequality
NL;E 0 .n/ < ./NL;E .n/; n 0:
2.2 Moduli theory of stable sheaves. Let X and L be as in the previous subsection.
We also fix a numerical class
v 2 H .X; Q/:
The moduli theory of L-semistable sheaves is based on the following functor.
Definition 2.2. We define the functor
ML .v/ W .Sch=C/ ! .Set/;
which sends a C-scheme T to the set of isomorphism classes of T -flat families of
coherent sheaves E 2 Coh.X T /, such that for each closed point t 2 T the restriction
E t :D EjXt is L-semistable with ch.E t / D v. For two families E1 , E2 as above, they
are isomorphic if there is a line bundle LT on T such that
E1 Š E2 ˝ T LT ;
where T is the projection X T ! T .
The moduli space of L-semistable sheaves E on X satisfying ch.E/ D v is defined
to be the C-scheme representing the functor ML .v/. However the functor ML .v/ is
not representable by a scheme when there is an L-semistable sheaf E with ch.E/ D v,
which is not L-stable. (We call such a sheaf strictly L-semistable.) Note that when E
is L-stable, we have (cf. [16], Corollary 1.2.8)
Aut.E/ D C ; (1)
but (1) is not satisfied when E is strictly L-semistable. The existence of non-trivial
automorphisms prevent the moduli functor to be representable. In this case, the correct
framework for the moduli problem is not a functor as above but a 2-functor from
the category of C-scheme to the 2-category of groupoids, involving the information
of the automorphism groups of semistable sheaves. The resulting moduli theory is
represented by an Artin stack of finite type over C.
Let us focus on L-stable sheaves. If we take a numerical class v to be primitive in
some sense, then we can see that there is no strictly L-semistable sheaf of numerical
type v. One of such a criterion is given by (cf. [35], Lemma 3.4)
g:c:d:fL;v .n/ W n 2 Zg D 1: (2)
In fact in this case, the following theorem holds (cf. [16]).
290 Y. Toda
Theorem 2.3. Suppose that (2) holds. Then there is no strictly L-semistable sheaf E
with ch.E/ D v. There is a projective scheme ML .v/ and a universal sheaf
E 2 Coh.X ML .v//;
which induces a functorial isomorphism,

Hom.T; ML .v//
! ML .v/.T /;
for any C-scheme T .
The space ML .v/ is a moduli space of L-stable sheaf E with ch.E/ D v.
2.3 Perfect obstruction theory. Let ML .v/ be the moduli space of L-stable sheaves
as in Theorem 2.3. The local structure of ML .v/ can be captured by the Ext-groups
of stable sheaves. Let T be an affine C-scheme and Tx its square zero extension, i.e.
there is an ideal J OTx satisfying
OTx =J Š OT ; J 2 D 0:
Note that J is an OT -module since J 2 D 0.

Let f be a morphism
f W T ! ML .v/;
which corresponds to a T -flat family of L-stable sheaves ET 2 Coh.X T /. The

deformation theory of sheaves concerns the extension of the morphism f to a morphism
from Tx ,
fN W Tx ! ML .v/:
It is well known that there is an obstruction class
ob 2 ExtXT
2
.ET ; ET ˝OT J /; (3)
such that an extension fN exists if and only if ob D 0. Moreover in this case, the set of
possible extensions fN is a torsor over ExtXT
1
.E; E ˝OT J /.
The above deformation theory can be described by the notion of perfect obstruction
theory [6].
Definition 2.4. For a C-scheme M , a perfect obstruction theory is given by the data
.E ; /; W E ! LM ;
where LM , E and are as follows.

• LM is a truncated cotangent complex of M . If M is embedded into a smooth

scheme A with defining ideal I OA , then LM is given by
LM D .I =I 2 ! A jM /:
Here A jM is located in degree zero.
• E is a perfect complex on M of perfect amplitude Œ1; 0, i.e. locally quasi-

isomorphic to a two term complex of vector bundles on M concentrated on
Œ1; 0.
• is a morphism in the derived category of coherent sheaves on M such that

H 0 ./ is an isomorphism and H 1 ./ is surjective.
Let W E ! LM be a morphism in the derived category, and E satisfies the first

and the second condition in Definition 2.4. Let T Tx be a square zero extension as
before. We have the distinguished triangle
˛
J ! LTx jT ! LT ! J Œ1:
Then for a morphism f W T ! M , we have the element
of 2 Ext1T .f E ; J /; (4)
given by the composition

f ˛
f E ! f LM ! LT
! J Œ1:
Then by Theorem 4.5 of [6], we have the following criterion for the morphism
W E ! LM to be a perfect obstruction theory.
Proposition 2.5. Let E be a perfect complex on a C-scheme M of perfect amplitude

Œ1; 0. Then a morphism W E ! LM gives a perfect obstruction theory if and only
if for any morphism f W T ! M as above, f extends to fN W Tx ! M iff of D 0 and
the set of extensions fN forms a torsor over HomT .f E ; J /.
Next we recall the Atiyah class. For a C-scheme M and a coherent sheaf E on M ,
its Atiyah class is a morphism
L
E ! E ˝ M Œ1;
L
given by applying p2 .p1 . / ˝ E/, where is given by the distinguished triangle

I =I2 ! OM M =I2 ! OM ! I =I2 Œ1:
292 Y. Toda
Here is a diagonal in M M , and pi are the projections from M M onto

corresponding factors. In Subsection 2.2 of [17], the lift of the Atiyah class
L
aE W E ! E ˝ LM Œ1
is constructed and called truncated Atiyah class. Namely the composition aE and the
natural morphism LM Œ1 ! M Œ1 coincides with the usual Atiyah class of E.
When M D ML .v/, then the perfect obstruction theory on M can be constructed by
the composition of the truncated Atiyah class of the universal sheaf E 2 Coh.X M /

and the natural morphism LXM ! M LM for the projection M W X M ! M :
aE L L

aE;M W E ! E ˝ LXM Œ1 ! E ˝ M LM :
We have the isomorphisms

L

Hom.E; E ˝ M LM Œ1/ Š Hom.R Hom.E; E/; M LM Œ1/
Š Hom.M Š R Hom.E; E/; LM Œ1/
Š Hom.R HomM .E; E/Œ2; LM /:
Here R HomM is the right derived functor of M Hom. Therefore aE;M corresponds
to a morphism
R HomM .E; E/Œ2 ! LM : (5)
The complex of the LHS is not concentrated on Œ1; 0, but on Œ2; 1. In order to
construct a complex concentrated on Œ1; 0, we assume that rank.v/ 1 for simplicity.
We have the morphisms
tr W R HomM .E; E/ ! RM OXM ;

id W RM OXM ! R HomM .E; E/:
Here tr is the trace morphism and id is the natural identity morphism. The composition
tr B id is equal to rank.v/ id, hence we have the splitting
R HomM .E; E/ Š R HomM .E; E/0 ˚ RM OXM :
Hence the morphism (5) induces the morphism
E;M W R HomM .E; E/0 Œ2 ! LM : (6)
The following proposition is due to Theorem 4.1 of [17].
Proposition 2.6. The morphism E;M is a perfect obstruction theory on M .

Proof. We just give an outline of the proof. For the details see [17], Theorem 4.1. Let
T be an affine C-scheme and f W T ! M a morphism. Let ET 2 Coh.X T / be the
pull-back of E via f . Then by the base change, we have
H i .Lf R HomM .E; E/0 / Š ExtX

i
T .ET ; ET /0 ; (7)
T .ET ; ET /0 is the kernel of the trace map,

i
where ExtX
tr W ExtXT
i
.ET ; ET / ! H i .OXT /: (8)
In particular if T D Spec C, then both sides of (8) are one dimensional by the Calabi–
Yau assumption of X and the fact that the endomorphism algebra of a stable sheaf is
always one dimensional. (Cf. [16], Corollary 1.2.8.) Hence we have
i
ExtXT .ET ; ET /0 D 0; i ¤ 1; 2;
showing that R HomM .E; E/Œ2 is a perfect complex with perfect amplitude Œ1; 0.
It is enough to check the criterion in Proposition 2.5. Let T Tx be a square zero
extension with OTx =J Š OT . For the class of in (4) we have
of 2 ExtXT
1
.Lf R Hom.E; E/0 Œ2; J /;
Š ExtXT
2
.ET ; ET ˝ J /:
In [17], Corollary 3.4, it is shown that the class of coincides with the obstruction class
(3). Therefore the condition in Proposition 2.5 follows from the standard deformation
theory of sheaves.
Using the assumption that X is a Calabi–Yau 3-fold, the perfect obstruction theory
(6) has a certain stronger property. By definition, an obstruction theory E ! LM on
a scheme M is symmetric if there is an isomorphism in the derived category,
W E

! .E /_ Œ1;
such that we have
_ D Œ1:
In particular, a symmetric perfect obstruction theory E ! LM satisfies rank.E / D 0.

By using the Serre duality, the perfect obstruction theory (6) is shown to be symmetric
(cf. [4]).
2.4 Virtual class. Let M be a C-scheme and W E ! LM a perfect obstruction

theory. Given the above data, we can construct the virtual fundamental cycle [6]
ŒM vir 2 Arank.E / .M /:
294 Y. Toda
We explain the construction of ŒM vir when E is globally quasi-isomorphic to a two

term complex of vector bundles E 1 ! E 0 , and is given by a morphism of com-
plexes,
E 1 / E0
1 0

I =I 2 / A jM;
in the notation of the previous subsection. This assumption is not necessary in the
construction of [6], but such a representative can always be chosen when M is a
projective.
The mapping cone of is given by the complex
0 ! E 1 ! E 0 ˚ I =I 2 ! A jM ! 0:
The condition for the perfect obstruction theory implies that the above sequence is right
exact. We take the associated abelian cones

0 ! TA jM ! E0 M C.I =I 2 / ! E1 ;
where Ei D .E i /_ and C.I =I 2 / is given by
C.I =I 2 / D pecOM Sym .I =I 2 /:
The normal cone CM=A is a subcone of C.I =I 2 /,

L d C1

CM=A D pecOM d 0 I =I
d
C.I =I 2 /:
Then the map

factors through E0 M CM=A (cf. [6]),
TA jM E0 M CM=A E0 M C.I =I 2 /;
hence we have the associated quotient cone
C vir :D .E0 M CM=A /=TA jM E1 :
Let s W M ! E1 be the zero section. The virtual cycle is defined to be the Gysin
pull-back
ŒM vir D s Š C vir :
Suppose that M D ML .v/ and E ! LM is a symmetric perfect obstruction theory

constructed in Proposition 2.6. Since rank.E / D 0, we have the associated zero
dimensional virtual cycle
ŒML .v/vir 2 A0 .ML .v//:

Definition 2.7. We define the Donaldson–Thomas (DT) invariant DTL .v/ to be
DTL .v/ D degŒML .v/vir 2 Z:
The following result follows from a general theory of virtual cycles [6].
Theorem 2.8. The invariant DTL .v/ is invariant under deformations of X . Namely
for a flat family of Calabi–Yau 3-folds X ! C over a smooth pointed curve 0 2 C
and a relatively ample line bundle L on X, suppose that .X0 ; L0 / is isomorphic to
.X; L/. Here for t 2 C , we denote X t :D 1 .t / and L t :D LjX t . Then for
v 2 .C; R QX /, we have
DTL t .v t / D DTL .v/:
2.5 Behrend function. Another (and easier) way to define the DT invariant is to use
Behrend’s microlocal function [4]. In order to discuss this, we see that the moduli space
ML .v/ has a certain particular local description due to the Calabi–Yau assumption of X .
Theorem 2.9 ([21], Theorem 5.2). For any p 2 ML .v/, there is an analytic open
neighborhood p 2 U ML .v/ a complex manifold V and a holomorphic function
f W V ! C such that there is an isomorphism of complex analytic spaces:
U Š fdf D 0g V:
Proof. We give an outline of the proof. For the details see [21], Theorem 5.2. Let
E 2 Coh.X/ be the stable sheaf representing p 2 ML .v/. First by applying Seidel–
Thomas twists associated to sufficiently ample line bundles, we may assume that E
is a vector bundle on X (cf. [21], Corollary 8.5). Then the analytic local structure of
ML .v/ can be identified with the local deformation space of E as a holomorphic vector
bundle.
Let
@N W E ! E ˝ X
0;1
;
N
be the @-connection which determines a holomorphic structure of E. Here X 0;1
the
sheaf of .0; 1/-forms on X . Then giving a deformation of E is equivalent to giving a
deformation of @N up to gauge equivalence. This is equivalent to giving
A 2 A0;1 .X; End.E//;
where A0;1 .X; End.E// is the space of End.E/-valued .0; 1/ forms, satisfying
.@N C A/2 D 0; (9)
up to gauge equivalence. The equation (9) is equivalent to

N C A ^ A D 0:
@A
296 Y. Toda
Let CS be the holomorphic Chern Simons function,
CS W A0;1 .X; End.E// ! C;
defined by
Z
1N 1
CS.A/ D @A ^ A C A ^ A ^ A ^ X ;
X 2 3
where X is a no-where vanishing holomorphic 3-form on X (cf. [31]). Then A 2
A0;1 .X; End.E// satisfies the equation (9) if and only if A is a critical locus of the
function CS. Therefore the local moduli space of E is written as
fd CS D 0g=G;
where G is the group of isomorphisms of E as a C 1 -vector bundle, i.e. the local moduli
space of objects in Coh.X / is written as a critical locus up to gauge equivalence.
However A0;1 .X; End.E// is an infinite dimensional vector space, and we need to
find suitable finite dimensional vector subspace of A0;1 .X; End.E//. This is worked
out in [21] (Theorem 5.2) by using Hodge theory. Namely the space of harmonic
forms W on A0;1 .X; End.E// is finite dimensional, satisfying W Š Ext 1 .E; E/, and
we restrict CS to W . The desired complex manifold V is obtained as an open subset
of W . For details see [21], Theorem 5.2.
For p 2 ML .v/, let p 2 U ML .v/ and f W V ! C be as in Theorem 2.9. We
set .p/ 2 Z to be
.p/ D .1/dim V .1 .Mp .f ///: (10)
Here Mp .f / is the Milnor fiber of f at p, which is defined as follows. Let p 2 V 0 V

be an analytic neighborhood with a norm k k on V 0 . Then for 0 < "
ı
1, the
topological type of the space
fz 2 V 0 W kz pk ı; f .z/ D f .p/ C "g; (11)
does not depend on " and ı. The Milnor fiber Mp .f / is defined to be the topological
space (11).
In [4], the map
W ML .v/ ! Z
is proved to be independent of choices and a well-defined constructible function on

ML .v/. Furthermore Behrend [4] proves the following result.
Theorem 2.10. We have
Z X
DTL .v/ D d D m. 1 .m//:
ML .v/ m2Z
Example 2.11. Suppose that ML .v/ is a connected non-singular variety. Then we

have .p/ D .1/dim ML .v/ for any p 2 ML .v/ by the definition of . Hence we have
DTL .v/ D .1/dim ML .v/ .ML .v//:
Remark 2.12. More precisely, Behrend [4] shows the following: for any scheme C-
scheme M , there is a canonical constructible function W M ! Z such that if M
admits a symmetric perfect obstruction theory, then we have
Z
degŒM D
vir
d:
M
If M is analytically locally written as a critical locus of some holomorphic function,

then is written in terms of Euler characteristic of the Milnor fiber as in (10).
3 Rank one theory

In this section, we give a survey of rank one DT theory and its relation to Gromov–
Witten (GW) theory.
3.1 Rank one DT invariants. Let X be a Calabi–Yau 3-fold and L an ample line
bundle on X. We set v 2 H .X; Q/ to be
v D .1; 0; ˇ; n/ 2 H 0 ˚ H 2 ˚ H 4 ˚ H 6 : (12)
In this case, the moduli space ML .v/ is isomorphic to the Hilbert scheme of curves on
X by the following lemma.
Lemma 3.1. If v is given by (12) , then we have the isomorphism
ML .v/ Š Hilbn .X; ˇ/; (13)
where the RHS is the Hilbert scheme of subschemes C X satisfying dim C 1 and
ŒC D ˇ; .OC / D n: (14)
Here H2 .X; Q/ is identified with H 4 .X; Q/ via Poincaré duality.

Proof. Let I 2 Coh.X / be an L-stable sheaf. By the definition of stability, I is a
torsion free rank one sheaf. We have the exact sequence of sheaves
0 ! I ! I __ ! T ! 0:
The sheaf I __ is a reflexive rank one sheaf on X , hence it is a line bundle on X . By

the assumption, c1 .I / D c1 .I __ / D 0, hence I __ is isomorphic to OX . Also the
embedding I ,! OX is an isomorphism in codimension one, hence T is isomorphic to
298 Y. Toda
OC for some subscheme C X with dim C 1, and I is the ideal sheaf of C . The
condition ch.I / D v is equivalent to the condition (14).
Conversely let C X be a subscheme with dim C 1 and satisfying (14). Then
the ideal sheaf IC is obviously L-stable. The above argument implies the isomorphism
(13).
By Lemma 3.1, the rank one DT invariant is given by

Z
In;ˇ D degŒHilbn .X; ˇ/ D
vir
d;
Hilbn .X;ˇ /
where is the Behrend function on Hilbn .X; ˇ/. Note that the invariant In;ˇ does not
depend on L, and it counts subschemes C X satisfying dim C 1 and (14).
We consider the generating series
X
DTˇ .X / :D In;ˇ q n ;
n
X
DT.X / :D DTˇ .X /t ˇ :
ˇ
It is easy to see that In;ˇ D 0 for n

0, hence the series DTˇ .X / is the Laurent
series of q. As the following example implies, the above series may admit nice product
expansion formula.
Example 3.2. (i) When ˇ D 0, then In;0 counts zero dimensional subschemes in
X with length n. In this case, the series DT0 .X / is computed by Li [26], Behrend–
Fantechi [7] and Levine–Pandharipande [25]. The result is
DT0 .X / D M.q/.X/ ;
where M.q/ is the MacMahon function:

Y 1
M.q/ D :
.1 q k /k
k1
(ii) Let f W X ! Y be a birational contraction which contracts a .1; 1/-curve

C , i.e.
C Š P 1; NC =X Š OP 1 .1/ ˚ OP 1 .1/:
Then the series DTmŒC .X / is computed by Behrend–Bryan [5],

X Y
DTmŒC .X /t m D M.q/.X/ .1 .q/m t /m :
m0 m1
(iii) Let S be a smooth projective del Pezzo surface over C. Take general elements
f 2 .S; OS .4KS //; g 2 .S; OS .6KS //:
We construct a Calabi–Yau 3-fold X with an elliptic fibration,
W X ! S;
by the defining equation
y 2 D x 3 C f x C g;
in the projective bundle
Proj SymS .OS ˚ OS .2KS / ˚ OS .3KS // ! S:
Here x and y are local sections of OS .2KS / and OS .3KS / respectively. A Calabi–
Yau 3-fold X constructed in this way is called a Weierstrass model. Let F X be a
fiber of W X ! S. The series DTmŒF .X / is computed in [33] (Theorem 6.9):
X Y
DTmŒF .X /t m D M.q/.X/ .1 .q/j t m /j.X/ .1 t m /.S/ :
m0 m1;j 1
As the above example shows, there is a factor M.q/.X/ D DT0 .X / in the product
expansions of DT.X /. The factor M.q/.X/ is interpreted to be a contribution of zero
dimensional subschemes, which should be ignored if we are only interested in curves
on X. The following reduced DT series are regarded to be curve counting theories on
X via DT invariants:
DT0ˇ .X / :D DTˇ .X /=M.q/.X/ ;

DT0 .X / :D DT.X /=M.q/.X/ :
3.2 GW/DT correspondence. As we observed in the previous subsection, the rank

one DT invariants have to do with curve counting invariants on X . On the other hand,
there is another curve counting theory called Gromov–Witten (GW) invariants. In this
subsection, we review the conjecture by Maulik–Nekrasov–Okounkov–Pandharipande
(MNOP) on an equivalence between rank one DT theory and GW theory.
First we briefly recall the definition of GW invariants. Let C be a connected reduced
curve with at worst nodal singularities and f an algebraic map,
f W C ! X: (15)
The above map f is called a stable map if the group of automorphisms of C which
commutes with f is a finite group. For g 0 and ˇ 2 H2 .X; Z/, the space
x g .X; ˇ/
M (16)
300 Y. Toda
is the moduli space of stable maps (15) satisfying g.C / D g and f ŒC D ˇ. Here
g. / is the arithmetic genus. The moduli space (16) is Deligne–Mumford stack of
finite type over C (cf. [23], [13]). We have the commutative diagram
f
C / X;

x g .X; ˇ/
M
where W C ! M x g .X; ˇ/ is the universal curve and f is a universal morphism.

By [6], there is a perfect obstruction theory on M x g .X; ˇ/. Its construction consists
of two steps: first we consider the forgetting morphism,
x g .X; ˇ/ ! Mg ;
hW M
sending .C; f / to C . Here Mg is the moduli stack of connected reduced curves with
at worst nodal singularities of arithmetic genus g, which is an Artin stack locally of
finite type over C. By the deformation theory of maps, we can construct the relative
obstruction theory:
.R f TX /_ ! LMx g .X;ˇ /=Mg :
As for the absolute theory, we consider the commutative diagram
h LMg / E / .R f TX /_ / h LMg Œ1;
id id

h LMg / LMx .X;ˇ / / LMx .X;ˇ /=M / h LMg Œ1.
g g g
Here the top and the bottom sequences are distinguished triangles. By the smoothness
of Mg and the stability, the morphism

W E ! LMx g .X;ˇ /
is an absolute perfect obstruction theory on Mx g .X; ˇ/.

By the Riemann–Roch theorem and the Calabi–Yau condition, it is easy to check
that rank.E / D 0. Hence there is a zero dimensional virtual cycle
x g .X; ˇ/vir 2 A0 .M
ŒM x g .X; ˇ//:
The GW invariant is defined by
Ng;ˇ
GW x g .X; ˇ/vir 2 Q:
:D degŒM
x g .X; ˇ/
Note that the GW invariant is not an integer in general, since the moduli space M
is not necessary a scheme but a Deligne–Mumford stack.
Let GW.X/ be the generating series,
X
GW.X / D Ng;ˇ
GW 2g2 ˇ
t :
g;ˇ
The above series is conjectured to be equivalent to the reduced DT series DT0 .X / by

MNOP [27].
Conjecture 3.3. (i) The series DT0ˇ .X / is the Laurent expansion of a rational function
of q, invariant under q $ 1=q.
(ii) We have the formula
DT0 .X / D exp GW.X /
after the variable change q D e i .

For instance, in the case of Example 3.2 (ii), the series DT0ŒC .X / is given by
DT0ŒC .X / D q 2q 2 C 3q 3
q
D ; (17)
.1 C q/2
which is a rational function of q, invariant under q $ 1=q.
3.3 Stable pair theory. The notion of stable pairs and their counting invariants are
introduced by Pandharipande–Thomas [28] in order to give a geometric understanding
of the reduced DT theory. By definition, a stable pair consists of the data
.F; s/; s W OX ! F;
where F is a pure one dimensional sheaf and s is a morphism which is surjective in

dimension one. For instance if C X is a smooth curve and D C is a divisor on
it, then F D OC .D/ and
s W OX OC ,! OC .D/;
determines a stable pair.

For n 2 Z and ˇ 2 H2 .X; Z/, the space
Pn .X; ˇ/;
is the moduli space of stable pairs .F; s/ satisfying (cf. [29])
ŒF D ˇ; .F / D n:
302 Y. Toda
Similarly to the DT theory, the moduli space Pn .X; ˇ/ is a projective scheme with a
universal stable pair,
OXPn .X;ˇ / ! F : (18)
In fact, the space Pn .X; ˇ/ can be interpreted as the moduli space of two term complexes
in the derived category
s
I D .OX ! F / 2 D b .Coh.X //; (19)
and the universal stable pair (18) is the universal object in the derived category. As we
discussed in the previous section, we need to choose a stability condition in order to
construct a ‘good’ moduli space. Indeed, the two term complex (19) can be regarded as
stable objects in the derived category w.r.t. a certain stability condition in D b Coh.X /.
See [1], [38].
As in the DT theory, the Atiyah class of the universal object (18) induces a perfect
obstruction theory and the zero dimensional virtual fundamental cycle on Pn .X; ˇ/.
The Pandharipande–Thomas (PT) invariant is defined as follows.
Definition 3.4. We define the invariant Pn;ˇ 2 Z to be

Z
Pn;ˇ :D degŒPn .X; ˇ/ D vir
P d;
Pn .X;ˇ /
where P is the Behrend function on Pn .X; ˇ/.
Similarly to the DT theory, the generating series PTˇ .X / and PT.X / are defined
in the following way:
X
PTˇ .X / D Pn;ˇ q n ;
n
X
PT.X / D PTˇ .X /t ˇ :
ˇ
In [28], the following conjecture is proposed.
Conjecture 3.5. We have the following formula:
DT0 .X / D PT.X /:
Example 3.6. Let P 1 Š C X be as in Example 3.2 (ii). Then a stable pair .F; s/
contributes to the invariant Pn;ŒC if and only if
F D OC .n 1/; 0 ¤ s 2 H 0 .OC .n 1//;

which corresponds to a closed point in P .H 0 .OC .n 1/// Š P n1 . The invariant

Pn;ŒC is calculated as
Pn;ŒC D .1/n1 .P n1 /

D .1/n1 n:
The series PTŒC .X / is
PTŒC .X / D q 2q 2 C 3q 3
q
D ;
.1 C q/2
which coincides with DT0ŒC .X / by (17).
4 Results for Euler characteristic invariants

Let X be a Calabi–Yau 3-fold as in the previous section. In this section, we review
the author’s results in [38], [39], [32], [34] on the properties of the generating series
of Euler characteristic versions of DT or PT invariants. Namely in the notation of the
previous section, instead of taking the weighted Euler characteristic of the Behrend
function, we take the naive Euler characteristic of the moduli spaces,
Iyn;ˇ :D .In .X; ˇ//;

Pyn;ˇ :D .Pn .X; ˇ//: (20)
The generating series DT.X c /, PT.X c / are defined in a similar way. If In .X; ˇ/,
Pn .X; ˇ/ are non-singular, then the above invariants coincide with DTn;ˇ , PTn;ˇ up
to sign respectively. The Euler characteristic invariants share similar properties to the
virtual invariants In;ˇ , Pn;ˇ , and showing results on these invariants provide good tests
for the conjectures on DT or PT invariants. We remark that the results discussed in
this section can be improved to involve the Behrend function by using the results by
Joyce–Song [21] and Behrend–Getzler’s announced result [8]. For the details see [38],
Section 8, in the arXiv version.
4.1 DT/PT, rationality. The main results in [38], [39] are summarized as follows.
(See also [30], [12].)
Theorem 4.1 ([38], [39]). For n 2 Z and ˇ 2 H2 .X; Z/, there are invariants
Nyn;ˇ 2 Q; y n;ˇ 2 Q;
L
satisfying that
• for ˇ ¤ 0, there is d 2 Z>0 such that Nyn;ˇ D Nym;ˇ for n ˙m modulo d ,
304 Y. Toda
y n;ˇ D L
• L y n;ˇ , which are zero for jnj 0, depending on ˇ,
such that we have the following product expansion formulas:

Y X
c /D
PT.X exp nNyn;ˇ q n t ˇ y n;ˇ q n t ˇ ;
L (21)
n>0;ˇ >0 n;ˇ
Y
c /D
DT.X exp nNyn;0 q n PT.X
c /:
n>0
Here ˇ > 0 means that ˇ is a homology class of an effective one cycle on X . The
y n;ˇ are defined in the following way. Let L be an ample line
invariants Nyn;ˇ and L
bundle on X. The invariant Nyn;ˇ is an Euler characteristic version of DT type counting
invariant of one dimensional L-semistable sheaves F on X satisfying
ch.F / D .0; 0; ˇ; n/ 2 H 0 ˚ H 2 ˚ H 4 ˚ H 6 ; (22)
where we regard ˇ 2 H2 .X; Z/ as an element of H 4 .X; Z/ via Poincaré duality. More

precisely, suppose that n and L ˇ are coprime. Then v D .0; 0; ˇ; n/ satisfies the
condition (2), and there is a projective fine moduli space,
ML .0; 0; ˇ; n/; (23)
parameterizing L-stable sheaves F satisfying (22), as in Theorem 2.3. In this case,

Nyn;ˇ is given by
Nyn;ˇ D .ML .0; 0; ˇ; n// 2 Z:
When n and L ˇ are not coprime, then the condition (2) is not satisfied and there may
be strictly L-semistable sheaves F satisfying (22). In this case, there is no fine moduli
space (23), and it is much more difficult to define Nyn;ˇ . Instead of working with the
classical moduli space, we should work with moduli stack of L-semistable sheaves,
denoted by
ML .0; 0; ˇ; n/; (24)
which is known to be an Artin stack of finite type over C. Then the stack (24) determines
an element of the Hall algebra of coherent sheaves on X (cf. Theorem 5.2 of [18]).
Roughly speaking, the invariant Nyn;ˇ is defined by taking the Euler characteristic of
the ‘logarithm’ of (24) in the Hall algebra. For details on the construction of Nyn;ˇ ,
see [20], [38]. Also see Section 4 of [33]. A priori, Nyn;ˇ is defined using an ample line
bundle L. However it can be shown that Nyn;ˇ is independent of L. (Cf. Theorem 6.16
in [21].)
y n;ˇ . It essentially counts objects in the bounded

Let us discuss the invariant L
derived category of coherent sheaves D b Coh.X /. More precisely, let us consider the
subcategory
AX :D hOX ; Coh1 .X /Œ1iex D b Coh.X /;
the smallest extension closed subcategory which contains OX and
Coh1 .X / :D fE 2 Coh.X / W dim Supp.E/ 1g:
The above subcategory AX is shown to be an abelian category (cf. [38], Lemma 3.5).
Let E be an object in AX satisfying
ch.E/ D .1; 0; ˇ; n/ 2 H 0 ˚ H 2 ˚ H 4 ˚ H 6 : (25)
y n;ˇ if the following conditions hold:

Then the object E contributes to the invariant L
• For any surjection E F Œ1 in AX with F 2 Coh1 .X /, we have .F / 0.
• For any injection F Œ1 ,! E in AX with F 2 Coh1 .X /, we have .F / 0.
The above objects are shown to be self dual, i.e. E 2 AX satisfies the above conditions
if and only its derived dual
R Hom.E; OX / 2 AX ; (26)
also satisfies the above conditions.

The above conditions can be interpreted to be a stability condition on objects in AX
satisfying (25). (Cf. Subsection 5.3 in [33].) If we denote this stability condition by ,
then we have the moduli space
M
.1; 0; ˇ; n/ (27)
which parameterizes -semistable objects E 2 AX satisfying (25). (I.e. the moduli

space of objects E 2 AX satisfying (25) and the above conditions. See [37], [39].) If
there is no strictly -semistable object E 2 AX satisfying (25), then M
.1; 0; ˇ; n/
y n;ˇ is defined by
is proven to be an algebraic space of finite type over C. In this case, L
y n;ˇ D .M
.1; 0; ˇ; n//:
L
If there are strictly -semistable objects, then the moduli space (27) is not an algebraic
space but an Artin stack. In this case, similarly to Nyn;ˇ , the invariant L y n;ˇ can be
defined by taking the logarithm of the stack (27) in the Hall algebra of AX , and its
‘Euler characteristic’. For details see [39] (Definition 4.1), [33] (Definition 5.5). The
property L y n;ˇ D L y n;ˇ is a consequence of the self duality (26).
The idea of the proof of Theorem 4.1 is as follows. As we mentioned, the invariant
y n;ˇ is interpreted to be counting semistable objects in the abelian category AX . If
L
306 Y. Toda
we change stability conditions on AX , then the corresponding counting invariants

may change. The behavior of these invariants under change of stability conditions is
described by wall-crossing formula established Joyce [19]. Furthermore we are able to
find stability conditions on AX so that the corresponding counting invariants coincide
with Iyn;ˇ or Pyn;ˇ . The formulas in Theorem 4.1 are consequence of wall-crossing
formula.
As a corollary of Theorem 4.1, we have the following.
Corollary 4.2. We have the equality
c 0 .X / D PT
DT c ˇ .X /;
ˇ
c 0 .X / and PT
and both of DT c ˇ .X / are Laurent expansions of a rational function of q,
ˇ
invariant under q $ 1=q.
Remark 4.3. It should be possible to improve Theorem 4.1 to involve the Behrend
function, using [21] and assuming the announced result [8]. Then the result cor-
responding to Corollary 4.2 shows Conjecture 3.3 (i) and Conjecture 3.5. We also
remark that Bridgeland [12] shows the result of Theorem 4.1 involving the Behrend
function, without using [8], in a way different from ours.
4.2 Flop formula. In this subsection, we review the result in [32] on a comparison
theorem of generating series of DT type invariants under birational transformations of
Calabi–Yau 3-folds.
For a Calabi–Yau 3-fold X , a projective birational contraction
fWX !Y (28)
is called a flopping contraction if the following conditions hold:

• f is an isomorphism in codimension one.
• Y has only Gorenstein singularities and we have KX D f KY .
• The relative Picard number of f is one.
Given a flopping contraction (28), we can uniquely construct the diagram of a flop,

X A_ _ _ _ _ _ _/ X
AA
AA
A (29)
f
AA f
Y.
The morphism f is again a flopping contraction, but the birational map is not an
isomorphism. It is well known that any birational map between Calabi–Yau 3-folds
are decomposed into a finite sequence of flops.
Example 4.4. For instance if Y has only one ordinarily double point p 2 Y , i.e.
analytic locally written as
p D 0 2 .xy C zw D 0/ C 4 ;
then the diagram (29) can be constructed by taking the blowing up of the ideals .x; z/
to construct f and .x; w/ to construct f . In this case, both of f and f contract
smooth rational curves C X, C X whose normal bundles are O.1/ ˚ O.1/.
Let W X Ü X be a flop as in the diagram (29). Then Bridgeland [9] shows that
there is an equivalence of derived categories,

ˆ W D b Coh.X /
! D b Coh.X /:
In the paper [32], using the above equivalence and wall-crossing formula, the author
gave a formula relating DT type invariants under flops. We set the generating series
c
DT.X=Y / to be
X
c
DT.X=Y / :D Iyn;ˇ q n t ˇ :
f ˇ D0
The following result is proved in [32].

Theorem 4.5 ([32], Theorem 1.3). We have the formulas
c
DT.X=Y c =Y /;
/ D i B DT.X
c /
DT.X c /
DT.X
D :
c
DT.X=Y / c =Y /
DT.X
Here and i are variable changes .ˇ; n/ D . ˇ; n/ and i.ˇ; n/ D .ˇ; n/.
Example 4.6. In the situation of Example 4.4, a computation similar to (and easier
than) Example 3.2 (ii) shows that
Y
c
DT.X=Y / D M.q/.X/ .1 C q m t ŒC /m :
m1
Noting .X/ D .X / and ŒC D ŒC , the result of Theorem 4.5 yields
Y
c /D
DT.X c /:
.1 C q m t ŒC /m .1 C q m t ŒC /m DT.X
m1
4.3 Stable pairs on local K3 surfaces. Let S be a projective K3 surface over C,
KS D OS ; H 1 .S; OS / D 0:
308 Y. Toda
Let X be the total space of the canonical line bundle (i.e. trivial line bundle) on S ,
X D S C:
The space X is a non-compact Calabi–Yau 3-fold. For the data
n 2 Z; ˇ 2 H2 .X; Z/;
the invariant Pyn;ˇ can be defined similarly to (20), by taking the Euler characteristic
of the moduli space of stable pairs. (However the moduli space is no longer compact.)
In [34], the author proved a formula which relates the invariants Pyn;ˇ to counting
invariants of semistable sheaves on the fibers of the second projection,
W X D S C ! C: (30)
The result together with a certain multiple cover formula of the latter invariants lead to
an Euler characteristic version of Katz–Klemm–Vafa prediction [see [22], Section 6)
of curve counting invariants on K3 surfaces. In this subsection, we review the main
result of [34].
The invariants counting semistable sheaves on the fibers (30) are defined similarly
to Nyn;ˇ in Theorem 4.1. Let us define the abelian subcategory
Coh .X / Coh.X /
to be the category of E 2 Coh.X / supported on (possibly more than one) fibers of .

For an object E 2 Coh .X / and an ample line bundle L on S , we can construct
its Hilbert polynomial
.E ˝ p L˝n / 2 QŒn;
and its reduced Hilbert polynomial as in Subsection 2.1. Then we have the notion of
L-(semi)stability on Coh .X /, as in Definition 2.1.
The Mukai vector is a map
v W Coh .X / ! H .X; Z/;
given by
p
v.E/ D ch.p E/ tdS 2 H .S; Z/:
Here p is the first projection,
p W X D S C ! S:
Given a numerical class
v D .r; ˇ; n/ 2 Z ˚ H 2 .S; Z/ ˚ Z;
we consider the moduli space of L-semistable sheaves E satisfying v.E/ D v, denoted

by
ML .r; ˇ; n/:
If v satisfies the condition (2), then the above moduli space is a projective fine moduli
scheme. In this case, J.r; ˇ; n/ is defined by
J.r; ˇ; n/ :D .ML .r; ˇ; n//:
The moduli space ML .v/ is known to be a holomorphic symplectic manifold of di-

mension .v; v/ C 2. Here for vi D .ri ; ˇi ; ni /, .v1 ; v2 / is Mukai inner product,
.v1 ; v2 / D ˇ1 ˇ2 r1 n2 r2 n1 :
Moreover ML .v/ is deformation equivalent to the Hilbert scheme of points in S

(cf. [40]), hence
.ML .v// D .Hilb.v;v/=2C1 .S //: (31)
Here .Hilbn .S // is the Hilbert scheme of n-points in S . The RHS of (31) is calculated
by the GRottsche’s formula [14],
X Y 1
.Hilbn .S //q n D : (32)
n0 n1
.1 q n /24
If the condition (2) is not satisfied, the definition of J.r; ˇ; n/ is much more complicated.
In this case, similarly to Nyn;ˇ , we consider the stack
ML .r; ˇ; n/;
parameterizing L-semistable sheaves E 2 Coh .X / with Mukai vector .r; ˇ; n/. The
above stack is an Artin stack of finite type over C, with possibly complicated stabilizers.
The invariant
J.r; ˇ; n/ 2 Q
is defined by taking the ‘logarithm’ in the Hall algebra of Coh .X /, and taking its
Euler characteristic. For details see Definition 4.20 in [34]. For r D 0, the definition
of J.0; ˇ; n/ implies that
J.0; ˇ; n/ D Nyn;ˇ : (33)
Here we have naturally identified H2 .X; Z/ with H 2 .S; Z/. Similarly to Nyn;ˇ , the
invariant J.r; ˇ; n/ does not depend on the choice of L. The following is the main
theorem of [34].
310 Y. Toda
Theorem 4.7 ([34])). We have the following product expansion formula:

Y
c
PT.X/ D exp .n C 2r/J.r; ˇ; r C n/t ˇ q n
r0;ˇ >0;n0
Y (34)
exp .n C 2r/J.r; ˇ; r C n/t ˇ q n :
r>0;ˇ >0;n>0
As we discussed in the previous section, the LHS of (34) is related to curve counting
invariants on X . On the other hand, the invariants J.r; ˇ; n/ in the RHS of (34) have
the following properties:
• If .r; ˇ; n/ 2 H .X; Z/ is primitive and algebraic, then (cf. [34], Equation (65))
2 =2rnC1
J.r; ˇ; n/ D .Hilbˇ .S //: (35)
The RHS of (35) is computed by (32).

• If g 2 Aut.H .S; Z// is an isomorphism of the lattice H .S; Z/ preserving the
weight two Hodge structure on it, we have (cf. [34], Theorem 4.21)
J.gv/ D J.v/: (36)
Here the weight two Hodge structure on H .S; C/ is given by
H 2;0 D H 2;0 ; H 0;2 D H 0;2 ;

H 1;1 D H 0;0 ˚ H 1;1 ˚ H 2;2 :
When .r; ˇ; n/ is not primitive, we conjecture that J.r; ˇ; n/ is related to the Euler
characteristic of the Hilbert scheme of points in S via multiple cover formula. The
details will be discussed in Subsection 5.2 below.
5 Open problems
In this section, we discuss several open problems in Donaldson–Thomas theory.
5.1 Multiple cover formula. The result of Theorem 4.1, involving the Behrend func-
tion, is now available by the work of Bridgeland [12]. Namely there are invariants
Nn;ˇ 2 Q; Ln;ˇ 2 Q;
satisfying
• for ˇ ¤ 0, there is d 2 Z>0 such that Nn;ˇ D Nm;ˇ for n ˙m modulo d ,
• Ln;ˇ D Ln;ˇ , which are zero for jnj 0, depending on ˇ,
such that we have the following product expansion formulas:

Y X
PT.X / D exp .1/n1 nNn;ˇ q n t ˇ Ln;ˇ q n t ˇ ; (37)
n>0;ˇ >0 n;ˇ
Y
DT.X / D exp .1/ n1
nNn;0 q n
PT.X /:
n>0
In particular, Conjecture 3.3 (i) and Conjecture 3.5 are now true.
However if we believe GW/DT correspondence in Conjecture 3.3 (ii), the series
PT.X/ is expected to satisfy a stronger rationality property, as discussed in [28]. This
is formulated as the conjectural Gopakumar–Vafa formula of PT.X /:
Conjecture 5.1. There are integers
nˇg 2 Z; for g 0; ˇ 2 H2 .X; Z/;
such that we have
1 1 2g2
YY ˇ Y Y kCg nˇ 2g2
g1k ˇ .1/
PT.X/ D .1 .q/ t /
j ˇ j n0
.1 .q/ t / g k :
ˇ >0 j D1 gD1 kD0
(38)
Using the formula (37), the above conjecture is reduced to showing a certain multiple
cover formula of the invariant Nn;ˇ . Namely we have the following:
Theorem 5.2 ([33], Theorem 6.4). For any .n; ˇ/, Conjecture 5.1 is equivalent to the
formula
X 1
Nn;ˇ D N1;ˇ=k : (39)
k2
k1;kj.n;ˇ /
In that case, we have

nˇ0 D N1;ˇ ;
and there is a way to write down nˇg for g 1 in terms of Ln;ˇ .

Similarly to Nyn;ˇ , the invariant Nn;ˇ is defined by using the stack (24) and its
‘logarithm’ in the Hall algebra. Instead of taking the Euler characteristic, we take the
integration of the Behrend function on (24) in order to define Nn;ˇ . The invariant
Nn;ˇ is the generalized DT invariant introduced by Joyce–Song [21], and known to be
independent of the choice of L and deformations of X .
The formula (39) is stronger than Kontsevich–Soibelman’s integrality conjecture
[24]. More precisely, let .n; ˇ/ be the invariant defined by the formula
X 1
Nn;ˇ D .n=k; ˇ=k/:
k2
k1;kj.n;ˇ /
312 Y. Toda
Then the integrality conjecture in [24] claims that .n; ˇ/ should be an integer. On
the other hand, the conjectural formula (39) claims that .n; ˇ/ is independent of n,
which in particular implies that
.n; ˇ/ D .1; ˇ/ D N1;ˇ 2 Z:
It is an open problem to show the formula (39). Even in the case that n and ˇ are
coprime, the formula (39) seems to be difficult to prove. As a summary, we propose
the following problem:
Problem 5.3. Prove the formula (39).
5.2 Computation of J.r; ˇ; n/. Let S be a smooth projective K3 surface over C,

and X D S C as in Subsection 4.3. In Theorem 4.7 we have established a formula
relating stable pair invariants and invariants counting semistable sheaves on the fibers
of X D S C ! C. The latter invariants were denoted by J.r; ˇ; n/. In [34] we
proposed a conjecture (Conjecture 1.3) relating J.r; ˇ; n/ to the Euler characteristic of
the Hilbert scheme of points in S .
As we discussed in (33), the invariant J.0; ˇ; n/ coincides with the Euler charac-
teristic invariant Nyn;ˇ . It is expected that the invariant Nyn;ˇ satisfies a multiple cover
formula similar to (39). Namely we expect that
X 1 y
Nyn;ˇ D N1;ˇ=k :
k2
k1;kj.n;ˇ /
Combined with (33), (35) and (36), we proposed the following conjecture in [34]:
Conjecture 5.4 ([34], Conjecture 1.3). If v 2 H .S; Z/ is an algebraic class, we have
the formula
X 1
J.v/ D .Hilb.v=k;v=k/=2C1 .S //: (40)
k2
k1;kjv
If we assume the above conjecture, then by Theorem 4.7, the generating series
c
PT.X/ is written as
Y ˇ 2 =2r.nCr/C1 .S//
c
PT.X/ D .1 y ˇ z n /.nC2r/.Hilb
r0;ˇ >0;n0
Y ˇ 2 =2r.nCr/C1 .S//
.1 y ˇ z n /.nC2r/.Hilb : (41)
r>0;ˇ >0;n>0
As we discussed in Section 6 of [34], the above formula is interpreted to be an Euler

characteristic version of Katz–Klemm–Vafa’s conjecture (see Section 6 in [22]).
Problem 5.5. Prove the formula (40).
5.3 Construction of DT type invariants w.r.t. Bridgeland stability. Let X be a

smooth projective Calabi–Yau 3-fold over C. In Subsection 4.1, we explained that the
stability conditions on the abelian subcategory AX D b Coh.X / plays an important
role in proving Theorem 4.1. This was enough since AX is closed under taking the
derived dual. However in order to see the effect of the invariants under the group of
autoequivalences of D b Coh.X /, (or derived equivalences D b Coh.X / Š D b Coh.Y /
for another variety Y ,) we need to generalize the notion of stability conditions to the
derived category D b Coh.X / and study the corresponding DT type invariants.
The notion of stability conditions on triangulated categories is introduced by Bridge-
land [10]. For a triangulated category D, a stability condition on it consists of the data
D .Z; A/;
where Z W K.D/ ! C is a group homomorphism called central charge, and A D

is the heart of a bounded t-structure, which satisfies the following axioms.
• For any non-zero E 2 A, we have
Z.E/ 2 H :D fr exp.i / W r > 0; 0 < 1g:
An object E 2 A is called Z-(semi)stable if for any subobject 0 ¤ F ¨ E, we

have arg Z.F / < ./ arg Z.E/.
• For any object E 2 A, there is a filtration (called Harder–Narasimhan filtration)
0 D E0 E1 EN D E;
such that each Fi D Ei =Ei1 is Z-semistable with arg Z.Fi / > arg Z.FiC1 /
for all i .
When X is a smooth projective variety, then Bridgeland shows that the set of ‘good’
stability conditions Stab.X / forms a complex manifold. So far, several examples have
been studies (cf. [11], [36], [2]).
The most interesting case is when X is a smooth projective Calabi–Yau 3-fold. In
this case, the space Stab.X / is expected to be related to the stringy KRahler moduli space
of X, that is the moduli space of complex structures of a mirror manifold of X . So the
following problem seems to be very important:
Problem 5.6. (i) For 2 Stab.X / and v 2 H .X; Q/, construct the invariant
DT .v/ 2 Q;
counting -semistable objects E 2 D b Coh.X / with ch.E/ D v.

(ii) Study how DT .v/ changes under change of 2 Stab.X /.
314 Y. Toda
5.4 Construction of Bridgeland stability conditions on 3-folds. The most serious

issue on Problem 5.6 is that we do not know whether Stab.X / is non-empty or not.
In fact when X is a smooth projective Calabi–Yau 3-fold, there is no known example
of a Bridgeland stability condition on D b Coh.X /. In [3], A. Bayer and E. Macri and
the author proposed a candidate of Bridgeland stability condition on D b Coh.X / for
any projective 3-fold X. It led to a conjectural Bogomolov–Gieseker type inequality
including higher Chern characters. This inequality seems to be a very difficult, but an
important, problem to solve. Below we discuss the conjecture we proposed in [3].
Let X be a smooth projective 3-fold, and take an element
B C i ! 2 H 2 .X; C/
such that ! is ample. From the idea of string theory, there should exist a stability
condition on X of the form .ZB;! ; AB;! /, where ZB;! W K.X / ! C is given by
Z
ZB;! .E/ D e .BCi!/ ch.E/
X

D e 3 .E/ C 1 ch
ch e 2 .E/ ! 1 ch
e 1 .E/ ! 2 C i ch e 0 .E/ ! 3 :
2 6
e
Here ch.E/ is the twisted Chern character,
e
ch.E/ D e B ch.E/ 2 H .X; Q/:
We propose the following conjecture:

Conjecture 5.7. There is the heart of a bounded t-structure AB;! D b Coh.X / such
that we have
.ZB;! ; AB;! / 2 Stab.X /:
In [3], we constructed a candidate of the heart AB;! in the following way. First we
have the (twisted) slope function,
B;! W Coh.X / n f0g ! Q n f0g;
given by
e 1 .E/ ! 2
ch
B;! .E/ D :
e 0 .E/
ch
The above slope function satisfies the following weak seesaw property: for any exact
sequence of sheaves 0 ! F ! E ! G ! 0 with F; G ¤ 0, we have
B;! .F / B;! .E/ B;! .G/; or (42)

B;! .F / B;! .E/ B;! .G/:
We say that E 2 Coh.X / is B;! -semistable if for any exact sequence of sheaves
0 ! F ! E ! G ! 0 as above, the inequality (42) is satisfied. We set TB;! , FB;!
to be
TB;! :D hE W E is B;! -semistable with B;! .E/ > 0i;

FB;! :D hE W E is B;! -semistable with B;! .E/ 0i:
Here h i is the smallest extension closed subcategory which contains . The pair
of subcategories .TB;! ; FB;! / forms a torsion pair (cf. [15]), and we define BB;!
D b Coh.X/ to be the corresponding tilting:
BB;! :D hFB;! Œ1; TB;! i:
The category BB;! is the heart of a bounded t-structure on D b Coh.X /, hence in

particular it is an abelian category.
Next we define the slope function on BB;! ,
B;! W BB;! n f0g ! Q [ f1g;
given by
Im ZB;! .E/
B;! .E/ D :
e 1 .E/ ! 2
ch
In [3], Lemma 3.2.1, we proved that B;! satisfies the weak seesaw property as in the
case of B;! on Coh.X /. In particular, we can define B;! -stability on BB;! .
0 0
We set TB;! ; FB;! BB;! to be
0
TB;! :D hE W E is B;! -semistable with B;! .E/ > 0i;
0
FB;! :D hE W E is B;! -semistable with B;! .E/ 0i:
The above subcategories form a torsion pair on BB;! . We define the abelian subcategory
AB;! D b Coh.X / to be
0 0
AB;! :D hFB;! Œ1; TB;! i:
In [3] we conjectured (Conjecture 3.2.6) that .ZB;! ; AB;! / is a Bridgeland stability

condition. This conjecture is shown to be equivalent to the following conjecture (cf.
Corollary 5.2.4 in [3]):
Conjecture 5.8 ([3], Conjecture 3.2.7). For any B;! -semistable object E 2 BB;!
with B;! .E/ D 0, we have the inequality
e 3 .E/ < 1 ch
ch e 1 .E/ ! 2 :
2
316 Y. Toda
p we checked the above conjecture in some cases, e.g. X D P , B D 0 and

3
In [3],
! < 3 3. The above conjecture seems to be difficult to solve, but we would like to
3
propose:
Problem 5.9. Prove Conjecture 5.8.
The above problem is the first step toward Problem 5.6.
References
[1] A. Bayer, Polynomial Bridgeland stability conditions and the large volume limit.
Geom. Topol. 13 (2009), 2389–2425.
[2] A. Bayer and E. Macri, The space of stability conditions on the local projective plane. Duke
Math. J. 160 (2011), no. 2, 263–322.
[3] A. Bayer, E. Macri, and Y. Toda, Bridgeland stability conditions on 3-folds I: Bogomolov-
Gieseker type inequalities. J. Algebraic Geom., to appear; preprint arXiv:1103.5010
[math.AG].
[4] K. Behrend, Donaldson-Thomas invariants via microlocal geometry. Ann. of Math. 170
(2009), 1307–1338.
[5] K. Behrend and J. Bryan, Super-rigid Donaldson-Thomas invariants. Math. Res. Lett. 14
(2007), 559–571.
[6] K. Behrend and B. Fantechi, The intrinsic normal cone. Invent. Math. 128 (1997), 45–88.
[7] K. Behrend and B. Fantechi, Symmetric obstruction theories and Hilbert schemes of points
on threefolds. Algebra Number Theory 2 (2008), 313–345.
[8] K. Behrend and E. Getzler, Chern-Simons functional. In preparation.
[9] T. Bridgeland, Flops and derived categories. Invent. Math. 147 (2002), 613–632.
[10] T. Bridgeland, Stability conditions on triangulated categories. Ann. of Math. 166 (2007),
317–345.
[11] T. Bridgeland, Stability conditions on K3 surfaces. Duke Math. J. 141 (2008), 241–291.
[12] T. Bridgeland, Hall algebras and curve-counting invariants. J. Amer. Math. Soc. 24 (2011),
969–998.
[13] W. Fulton and R. Pandharipande, Notes on stable maps and quantum cohomology. In
Algebraic geometry – Santa Cruz 1995, Proc. Sympos. Pure Math. 62, Part 2, Amer. Math.
Soc., Providence, R.I., 1997, 45–96.
[14] L. GRottsche, The Betti numbers of the Hilbert scheme of points on a smooth projective
surface. Math. Ann. 286 (1990), 193–207.
[15] D. Happel, I. Reiten, and S. O. Smalø, Tilting in abelian categories and quasitilted algebras.
Mem. Amer. Math. Soc. 120, Amer. Math. Soc., Providence, R.I., 1996.
[16] D. Huybrechts and M. Lehn, Geometry of moduli spaces of sheaves. Aspects Math. E31,
Vieweg, Braunschweig 1997.
[17] D. Huybrechts and R. P. Thomas, Deformation-obstruction theory for complexes via Atiyah
and Kodaira-Spencer classes. Math. Ann. 346 (2010), no. 3, 545–569.
[18] D. Joyce, Configurations in abelian categories II. Ringel-Hall algebras. Adv. Math. 210
(2007), no. 2, 635–706.
[19] D. Joyce, Holomorphic generating functions for invariants counting coherent sheaves on
Calabi-Yau 3-folds. Geom. Topol. 11 (2007), 667–725.
[20] D. Joyce, Configurations in abelian categories IV. Invariants and changing stability condi-
tions. Adv. Math. 217 (2008), no. 1, 125–204.
[21] D. Joyce and Y. Song, A theory of generalized Donaldson-Thomas invariants. Mem. Amer.
Math. Soc. 217 (2012), no. 1020.
[22] S. Katz, A. Klemm, and C. Vafa, M-theory, topological strings and spinning black holes.
Adv. Theor. Math. Phys. 3 (1999), 1445–1537.
[23] M. Kontsevich, Enumeration of rational curves via torus actions. In The moduli space of
curves, Progr. Math. 129, Birkhäuser, Boston, MA, 1995, 335–368.
[24] M. Kontsevich and Y. Soibelman, Stability structures, motivic Donaldson-Thomas invari-
ants and cluster transformations. Preprint, arXiv:0811.2435 [math.AG].
[25] M. Levine and R. Pandharipande, Algebraic cobordism revisited. Invent. Math. 176 (2009),
63–130.
[26] J. Li, Zero dimensional Donaldson-Thomas invariants of threefolds. Geom. Topol. 10
(2006), 2117–2171.
[27] D. Maulik, N. Nekrasov, A. Okounkov, and R. Pandharipande, Gromov-Witten theory and
Donaldson-Thomas theory. I. Compos. Math. 142 (2006), 1263–1285.
[28] R. Pandharipande and R. P. Thomas, Curve counting via stable pairs in the derived category.
Invent. Math. 178 (2009), 407–447.
[29] J. Le Potier, Systèmes cohérents et structures de niveau. Astérisque 214 (1993).
[30] J. Stoppa and R. P. Thomas, Hilbert schemes and stable pairs: GIT and derived category
wall crossings. Bull. Soc. Math. France 139 (2011), no. 3, 297–339.
[31] R. P. Thomas, A holomorphic Casson invariant for Calabi-Yau 3-folds and bundles on
K3-fibrations. J. Differential. Geom. 54 (2000), 367–438.
[32] Y. Toda, Curve counting theories via stable objects II. DT/ncDT/flop formula. Preprint,
[33] Y. Toda, Stability conditions and curve counting invariants on Calabi-Yau 3-folds. Kyoto J.
Math. 52 (2012), no. 1, 1–50.
[34] Y. Toda, Stable pairs on local K3 surfaces. J. Differential Geom., to appear; preprint
[35] Y. Toda, Fourier-Mukai transforms and canonical divisors. Compos. Math. 142 (2006),
962–982.
[36] Y. Toda, Stability conditions and crepant small resolutions. Trans. Amer. Math. Soc. 360
(2008), 6149–6178.
[37] Y. Toda, Limit stable objects on Calabi-Yau 3-folds. Duke Math. J. 149 (2009), 157–208.
[38] Y. Toda, Curve counting theories via stable objects I: DT/PT correspondence. J. Amer.
Math. Soc. 23 (2010), 1119–1157.
318 Y. Toda
[39] Y. Toda, Generating functions of stable pair invariants via wall-crossings in derived cate-
gories. In New developments in algebraic geometry, integrable systems and mirror symme-
try (RIMS, Kyoto, 2008), Adv. Stud. Pure Math. 59, Mathematical Society of Japan, Tokyo
2010, 389–434.
[40] K. Yoshioka, Some examples of Mukai’s reflections on K3 surfaces. J. Reine Angew. Math.
515 (1999), 97–123.
Notes on formal deformations of abelian categories
Michel Van den Bergh
1 Introduction
In these notes we provide the foundation for the deformation theoretic parts of [16],
[17]. In [17] we construct non-commutative analogues of quadrics and in [16] we
define non-commutative P 1 -bundles over commutative varieties. A notable special
case of the latter are non-commutative analogues of Hirzebruch surfaces.
Indeed [16] contains a proof that any formal deformation of a Hirzebruch surface
(in a suitable sense) is given by a non-commutative Hirzebruch surface. Similarly
the original (privately circulated) version of [17] contains a proof that any formal
deformation of a quadric is a non-commutative quadric (see [15], §11.2, for a sketch).
I deleted this proof when I first put the paper on the arXiv (8 years after it was written)
since I was unhappy with the deformation theoretic setup that was used.
Meanwhile a satisfactory infinitesimal deformation theory for abelian categories
has been developed in [11], [12]. In the noetherian setting (which is sufficient for the
applications we have in mind) the passage from the infinitesimal context to the formal
context is an application of Jouanolou’s results in [10]. Nonetheless Jounalou’s exposé
is written for a different purpose so some translation is necessary. After several (not
very satisfactory) attempts to rewrite the deformation theoretic parts of [16], [17] using
Jouanolou’s language of “AR-J-adic systems” I decided that it was better to write a
self contained paper on formal deformations of noetherian abelian categories, which
resulted in the current paper. On the purely mathematical level there is very little
originality in what we will do. Besides Jouanolou’s exposé we have also borrowed
from [3] (which basically discusses trivial deformations) and from §5 of [8] (which
discusses formal schemes). On the expository level we deviate from the aforementioned
references by systematically using Pro-objects instead of adic objects. Pro-objects form
a co-Grothendieck category so in particular they have very well behaved inverse limits.
We now give a more detailed exposition of our setup. Let R be a commutative

noetherian ring and let J be an ideal in R.
Definition 1.1. Let C be a noetherian R-linear abelian category. The completion Cy of

C is the full subcategory of Pro.C / consisting of the pro-objects M over C such that
M=MJ n 2 C for all n and such that the canonical map M ! lim M=MJ n is an
n
isomorphism.

The author is a director of research at the FWO.
320 M. Van den Bergh
We reproduce Jouanolou’s proof (recast in our language) that Cy is a noetherian

abelian category (see Proposition 2.2.5 below).
There is an obvious exact functor ˆ W C ! Cy W M 7! lim M=MJ n and we
n
say that C is complete if this functor is an equivalence. Roughly speaking an .R; J /-
deformation of an R=J -linear abelian category E will be a complete R-linear category
F together with an equivalence E Š FR=J where FR=J is the full subcategory of F
consisting of objects annihilated by J . To make this definition work one has to impose
certain flatness conditions. See §2.1 and §3 for more details.
So to understand formal deformations of abelian categories we have to under-
stand completion. We first note that completion extends to functors. We show that if
.T i /i2Z W C ! D is a @-functor between noetherian abelian R-linear categories then
this functor extends to a @-functor .Ty i /i2Z W Cy ! D
y (see Theorem 2.3.1). This is a
slight improvement over [10] as Jouanolou imposes some extra conditions on T which
seem to be superfluous.
We use the good behaviour of @-functors to study Ext-groups. Let C t be the full
subcategory of C consisting of the objects which are annihilated by some power of J .
Then we define the completed Ext-groups between objects M; N 2 Cy as follows:
‘ Extiy .M; N / D Ext iPro.C t / .M; N /: (1.1)

C
An alternative point of view to this definition is that we consider the full subcategory
Dc .C/ of D.Pro.C t // of complexes whose cohomology lies in Cy . Thus Dc .C/ has a
t -structure whose heart is Cy . Then the completed Ext-groups for M; N 2 Cy may be
reinterpreted as
‘ ExtCy .M; N / D HomDc .C/ .M; N Œn/:
In the case that C is the category of torsion l-adic constructible sheaves it would be
interesting to compare this derived category to the standard derived category of l-adic
sheaves [4], [5], [7].
Obviously ‘ ExtCy .; / is a @-functor in both arguments but apart from this we
don’t have anything to say about it. However in the event that CR=J (the objects in C
annihilated by J ) has finitely generated Ext-groups over R=J and C is “formally flat”
(see §2.4) then we have the expected formula
‘ Extiy .M; N / D lim lim Ext iC l .M=MJ l ; N=NJ k /:

C ! R=J
k lk
An important theorem in algebraic geometry is Grothendieck’s existence theorem [8].

This theorem extends to the current setting (see also [3]). In §4 we introduce the notion
of a strongly ample sequence.1 By definition a sequence .O.n//n2Z of objects in a
noetherian abelian category E is strongly ample if the following conditions hold
1
We could have used ample sequences [14] but to have good behaviour of higher Ext-groups it is conve-
nient to use a slightly stronger notion.
Notes on formal deformations of abelian categories 321
(A1) For all M 2 E and for all n there is an epimorphism ˚tiD1 O.ni / ! M with
ni n.
(A2) For all M 2 E and for all i > 0 one has ExtiE .O.n/; M / D 0 for n 0.
A strongly ample sequence .O.n//n2Z in E is ample in the sense of [14]. Hence using
the methods of [2] or [14] one obtains E Š gr.A/=f:l: if E is Hom-finite, where A is
the noetherian Z-algebra ˚ij HomE .O.j /; O.i //.
The following is our version of Grothendieck’s existence theorem.
Proposition 1.2 (see Proposition 4.1). Assume that R is J -adically complete. Let E be
an Ext-finite R-linear noetherian category with a strongly ample sequence .O.n//n .
Then
(1) E is complete;
(2) if in addition E is flat (see §2.1) then we have for M; N 2 E:
ExtiE .M; N / D ‘ ExtiE .M; N /:
The property for a sequence to be strongly ample lifts well under deformations.
Theorem 1.3 (an extract of Theorem 5.4). Let D be an R-deformation of an Ext-finite
flat R=J -linear noetherian abelian category C and .O.n//n be a sequence of R-flat
objects in D. Then .O.n/=O.n/J /n is strongly ample in C if and only if .O.n//n is
strongly ample in D.
Many algebraic varieties (e.g. Del Pezzo surfaces) have a strongly ample sequence
consisting of exceptional objects. Such a sequence can then be lifted to any deformation
(see §5). This idea is basically due to Bondal and Polishchuk and is described explicitly
in [15], §11.2. It was used to define non-commutative quadrics in [17] and indirectly in
the classification of non-commutative Hirzebruch surfaces in [16]. See also the recent
paper [6].
Let us also mention that a very complete treatment of deformations of algebraic
varieties as ringed spaces (including their derived categories) over kŒŒt has been given
in [9].
Acknowledgement. We thank the referee for his careful reading of the manuscript and
for pointing out that the proof of Proposition 4.1 was not complete.
2 Completion of abelian categories

2.1 Base extension. We recall briefly some notions from [12]. Throughout R will
be a commutative noetherian ring and mod.R/ is its category of finitely generated
modules.
Let C be an R-linear abelian category. Then we have bifunctors ˝R W C

mod.R/ ! C and HomR .; / W mod.R/B C ! C defined in the usual way.
These functors may be derived in their mod.R/-argument to yield bi-delta-functors
i .; /, Ext R .; /. An object M 2 C is R-flat if M ˝R is an exact functor,
Tor R i
i .M; / D 0 for i > 0.

or equivalently if Tor R
By definition (see [12], §3) C is R-flat if Tor R i
i or equivalently Ext R is effaceable
in its C -argument for i > 0. This implies that Tor i and Ext R are universal @-functors
R i
in both arguments.
If f W R ! S is a morphism of commutative rings and C is an R-linear abelian
category then CS denotes the (abelian) category of objects in C equipped with an S-
action. We usually refer to objects in CS as .S; C /-objects and if S is graded then we
also talk about graded .S; C /-objects. If f is surjective then CS identifies with the full
subcategory of C given by the objects annihilated by ker f . If R is noetherian and S
is module finite over R then the inclusion functor CS ! C has right and left adjoints
given respectively by HomR .S; / and ˝R S .
2.2 Completion of noetherian abelian categories. Below we refer to a pair .R; J /

where R is a commutative noetherian ring and J R is an ideal as a J -adic noetherian
ring. Below .R; J / is a J -adic noetherian ring. We put Rn D R=J n and we denote
the J -adic completion of R by R.y This is also a noetherian ring. Using a slight abuse
of notation we denote the extended ideal J Ry by J .
Recall that an abelian category C is said to be noetherian if it is essentially small
and all objects are noetherian. Below C is an R-linear noetherian category.
If D is an essentially small abelian category then the category Pro.D/ of pro-
objects over D is the category whose objects are filtered inverse systems .M˛ /˛ and
whose Hom-sets are given by
HomPro.D/ ..M˛ /˛ ; .Nˇ /ˇ / D lim lim HomD .M˛ ; Nˇ /: (2.1)
!
ˇ ˛
In particular if we identify D with the one-object inverse systems in Pro.D/ then

.M˛ /˛ D lim M˛ in Pro.D/.
˛
Lemma 2.2.1 ([1], §I.8). Assume that D is an essentially small abelian category.
Then Pro.D/opp is a Grothendieck category and in particular Pro.D/ has exact filtered
inverse limits and enough projectives. The natural functor D ! Pro.D/ is fully faithful
exact and its essential image is closed under extensions. If D 2 D then HomD .; D/
sends inverse limits to direct limits (in other words D is co-finitely presented).
Definition 2.2.2. The completion Cy of C is the full subcategory of Pro.C / consisting
of the objects M such that M=MJ n 2 C for all n and such that the canonical map
M ! lim M=MJ n is an isomorphism.
n
It is easy to see that Cy is a R-linear
y category. To study objects in Cy we need to
consider filtrations. By definition a filtration on an object M of an R-linear category D
is a descending chain of subobjects M D F0 M F1 M . The associated

graded objects gr F M is the Z-graded object over D defined by the formal direct sum
L
n Fn M=FnC1 M . By FJ we denote the J -adic filtration. I.e. FJ;i M D MJ for
i
i 0 and FJ;i M D M for i 0.

We say that the filtration F is adapted to J if .Fi M /J FiC1 M (see [10], §4.2).
In that case gr F M is a graded .gr J R; DR=J /-object.
Lemma 2.2.3. If M is a noetherian object in an R-linear abelian category D and S
is a positively graded noetherian R-algebra such that S0 D R then M ˝R S is a
noetherian graded .S; D/-object.
Proof. This follows from a variant of Hilbert’s basis theorem. See e.g. Theorem 5.1.4
in [10] and Lemma 4.2.4 in [10].
Lemma 2.2.4 (compare with [10], Theorem 4.2.6). Assume that K 2 Pro.C / is
equipped with a J -adapted filtration F such that
(1) lim K=Fn K D K.
n
(2) gr F K is a noetherian graded .gr J R; CR=J /-object.
y
Then K 2 C.
Proof. We follow somewhat the idea of [10], Lemma 4.2.7. For any r 0 define
Fi.r/ K D KJ ir \ Fi1 K C Fi K (with J ir D R for r i ). Then we have
def
Fi K F .r/ K Fi1 K and .Fi.r/ K/J FiC1 .r/
K Fi K. In other words gr .r/
F K D
L .r/ i
FiC1 K=FiC1 K is an ascending chain of graded .gr J R; CR=J /-subobjects of gr F K
which must be stationary. Thus there is an r such that for all i,
KJ ir \ Fi1 K C Fi K D KJ ir1 \ Fi1 K C Fi K D D Fi1 K;
and in particular Fi1 K KJ ir C Fi K. Iterating this inclusion and renumbering

we get that there exists an r such that
Fi K KJ ir C Fj K
for all j i. Fix i and choose generators f1 ; : : : ; fp for J ir . Then we get diagrams
for j i :
0O (2.2)
.fi /i
.K=Fj K/p / K=Fj K / K=.Fj K C J ir K/ /0
O O
K / K=Fi K.
Using exactness of filtered inverse limits we get from (2.2)
0O
.fi /i
Kp /K / lim K=.Fj K C J ir K/ /0
j O
K / K=Fi K
and hence from the upper exact sequence we obtain
K=KJ ir D lim K=.Fj K C J ir K/:

j
In other words the identity map K ! K induces a map K=Fi K ! K=KJ ir which
yields Fi K KJ ir .
The fact that gr F K is noetherian implies easily that it has left bounded grading.
Since C is closed under extensions inside Pro.C / it follows that K=Fi K 2 C for
all i . Furthermore since C is an abelian subcategory of Pro.C/ it is also closed under
˝R M for M 2 mod.R/.
Hence K=KJ i D .K=FiCr K/ ˝R R=J i 2 C. Furthermore since the J -adic
filtration and the F -filtration are cofinal we also get lim K=KJ n D lim K=Fn K D
n n
K. This shows that indeed K 2 Cy .
Proposition 2.2.5 (compare with [10], Theorem 5.2.3). Cy is a noetherian abelian
subcategory of Pro.C /.
Proof. We first prove that Cy is an abelian subcategory of Pro.C /. It is obviously closed

under cokernels (using the exactness of lim and right exactness of ˝R R=J n ) so we

must prove it is closed under kernels.
Let
0!K!M !N
be an exact sequence in Pro.C / with M , N 2 C. y We must prove K 2 C. y Put
def
Fi K D MJ i \K KJ i . This is a filtration on K which is adapted to J . Furthermore
we have exact sequences
0 ! K=Fn K ! M=MJ n ! N=NJ n : (2.3)
By exactness of filtered inverse limits we deduce K D lim K=Fn K. Furthermore we

n
obtain exact sequences
0 ! gr F K ! gr FJ M ! gr FJ N:
Since M=MJ 2 C it follows from Lemma 2.2.3 that gr FJ M is a noetherian graded

.gr J R; CR=J / object. Hence gr F K is also a noetherian graded .gr J R; CR=J /-object.
By Lemma 2.2.4 we conclude K 2 Cy .
It remains to show that Cy is noetherian. Since any object M in Cy satisfies M D
lim M=J n M and the category of Z-indexed inverse systems over C is essentially
n
small it follows that Cy is essentially small as well. Thus it remains to show that any
M 2 Cy is noetherian.
Let N ,! M be a subobject of M in Cy . Put Fn N D N \ MJ n . Then N=Fn N is
the image of N=NJ n ! M=MJ n and so it lies in C . Furthermore taking the inverse
limits of the maps
N=NJ n N=Fn N ,! M=MJ n
and using exactness of filtered inverse limits we get N D lim N=Fn N . Now assume
n
that we have inclusions N1 N2 M in Cy such that if we equip N1 , N2 with the
filtrations induced from the J -adic filtration on M then the map gr F N1 ! gr F N2
is an isomorphism. We claim that then necessarily N1 D N2 . Indeed from the five
lemma we obtain N1 =Fi N1 D N2 =Fi N2 . It then suffices to take inverse limits.
Now let M .r/ M be an ascending chain of subobjects and equip them with
the filtrations induced from the J -adic filtration on M . As indicated above gr FJ M
is a noetherian graded object over .gr J R; CR=J / and hence the chain .gr F M .r/ /r is
stationary. By the discussion in the previous paragraph the chain .M .r/ /r is stationary
as well.
We may compare our definition of Cy with the notion of J -adic inverse systems.
Definition 2.2.6 (see [10], §3.1). Let C be an R-linear noetherian abelian category.
The category of J -adic inverse systems CL over C is defined as the full subcategory
of inverse systems .Mn ; n / over C such that Mn J n D 0 and such that the transition
maps n W Mn ! Mn1 induce isomorphisms Mn =Mn J n1 ! Mn1 .
Proposition 2.2.7. The functor
† W Cy ! CL W M 7! .M=MJ n /n
is an equivalence of categories. Its inverse is given by
‰ W CL ! Cy W .Nn /n 7! lim Nn :

n
Proof. We first show that ‰ is well defined. Let .Nn /n 2 CL and let N be its inverse
limit in Pro.C /. Using exactness of filtered inverse limits in Pro.C/ we get
N=NJ i D .lim Nn / ˝R R=J i D lim.Nn =Nn J i / D Ni :

n n
Thus we have indeed N D lim Ni D lim N=NJ i . From this reasoning we also get
i i
†‰.Ni /i D .Ni /i .
The fact that ‰† is the identity follows from the definitions.
The following easy result motivates the definition of Cy .

y
Proposition 2.2.8. One has mod.R/O D mod.R/.
Proof. In the proof it is necessary to distinguish between inverse limits in Mod.R/ and
Pro.mod.R//. Therefore we will temporarily denote the latter by p lim.

Let pRy be the object of mod.R/ygiven by p lim R=J n . Its endomorphism ring
n
is equal to lim R=J n D R. y It suffices to prove that pR
y is a projective generator of
n
mod.R/O.
We first show that pR y is projective. Let M 2 mod.R/O. Then
y M / D lim HomR=J n .R=J n ; M=MJ n /

HomPro.mod.R// .pR;

n
D lim M=MJ n :

n
Hence we must prove that M 7! lim M=MJ n is exact. Now let

n
0!K!M !N !0
be an exact sequence in mod.R/O. By Lemma 2.2.9 below we have that
0 ! K=KJ n ! M=MJ n ! N=NJ n ! 0
is exact up to essentially zero systems. From this one easily deduces that its inverse
limit is exact.
Now we prove that pR y is a generator. Let M be a object of mod.R/O. Choose R=J -
generators for M=JM 2 mod.R=J / and lift those to R=J n -generators for M=MJ n 2
mod.R=J n /. By Nakayama we get compatible epimorphisms .R=J n /t ! M=MJ n
for some fixed t.
Taking inverse limits we obtain an epimorphism pR yt ! M and we are done.
The following lemma was used.

Lemma 2.2.9. Assume that D is an R-linear abelian category and M N is an in-
clusion of noetherian objects in D. Then these objects satisfy the Artin–Rees condition
in the sense that there exists an r such that for all n we have NJ nCr \ M MJ n .
L
Proof. This is proved in the standard way. Let R zD
n0 J be the Rees ring of R.
n
z
L The igraded ring R is finitely generated over R and it follows from Lemma
L 2.2.3 that
NJ is a noetherian graded object over . z D/. Hence so is C D
R; NJ i
\ M.
i i
L
Inside C we have an ascending chain of subobjects C .r/ D i NJ \ MJ
i ir
(with J D R for p 0) which must be stationary. Hence for a certain r we have for
p
any i: NJ i \ MJ ir D NJ i \ MJ ir1 D D NJ i \ M . Putting i D n C r

yields MJ n NJ nCr \ MJ n D NJ nCr \ M .
There is a canonical functor
ˆ W C ! Cy W M 7! lim.M=MJ n / (2.4)

n
Proposition 2.2.10. The functor ˆ introduced above is exact. It induces an equivalence
CR=J n Š .Cy /R=J n :
Proof. Exactness is a consequence of Lemma 2.2.9. It is similar to the proof of exact-

ness of M 7! lim M=MJ n in the proof of Proposition 2.2.8.
n
The second statement is a tautology when written out formally.
Lemma 2.2.11. Let ˆ W C ! D be a functor between R-linear noetherian abelian

categories which induces equivalences ˆn W CR=J n ! DR=J n . Then
ŷ W Cy ! D
y W M 7! lim ˆn .M=MJ n /

n
is an equivalence.
Proof. It is easy to check that ŷ is well defined and that its inverse is given by
ŷ 1 .N / D lim ˆ1 .N=NJ n /.
n n
Definition 2.2.12. A noetherian R-linear abelian category C is complete if the functor

ˆ W C ! Cy is an equivalence.
Proposition 2.2.13. Cy is complete.
Proof. This follows from the second statement of Proposition 2.2.10 combined with
Lemma 2.2.11.
For completeness let us recall the following result
Lemma 2.2.14 (Nakayama). Let M 2 Cy be such that MJ D M . Then M D 0.
Proof. We have M D lim M=MJ n D 0 since M D MJ D MJ 2 D .

n
2.3 Functors. Now we consider functors. Let T D .T i /i be a @-functor between R-

linear noetherian abelian categories C and D. We extend T to a @-functor Ty commuting
with filtered inverse limits between Pro.C / and Pro.D/.
In this section we prove the following strengthening of Proposition 5.3.1 in [10].
Theorem 2.3.1. The functor Ty sends Cy to D.

y
Proof. This is a variant of [10], Proposition 5.3.1. For the convenience of the reader
we adapt the proof in loc. cit. to our setting.
We need some rudiments from the foundation of the theory of spectral sequences.
In its abstract form a spectral sequence over an abelian category E is a sequence of

complexes E D .Er ; dr /r1 together with isomorphisms H .Er ; d / Š .ErC1 ; 0/. If

the terms of the complexes Er carry a grading then we assume that dr is homogeneous.
Note that this setup is shifted with respect to the usual indexing of spectral sequences.
This is more convenient for filtered objects.
Starting from spectral sequence E we may construct subobjects
0 D B1n Brn Zrn Z1n D E1n
with Ern D Zrn =Brn . The subobjects Brn ; Zrn are constructed recursively using the
following exact sequences.
drn
nC1
0 ! ZrC1
n
=Brn ! Zrn =Brn ! ZrnC1 =BrnC1 ! ZrnC1 =BrC1 ! 0: (2.5)
If Z1n
D lim Zrn , B1 n
D lim Brn exist then we say that E converges to E1 n
D
r !
r
Z1 =B1 . The graded object E1 is called the limit of the spectral sequence.
n n
The spectral sequence is said to degenerate at Ern0 if drn D 0, drn1 D 0 for r r0 .

In that case it follows from (2.5) that BrC1
n
D Brn , ZrC1
n
D Zrn for r r0 and thus
E1 exists and is equal to Er0 .
n n
If .T n /n W E ! F is a @-functor between abelian categories and X 2 E is an

object equipped with a descending filtration FkC1 X Fk X indexed by Z then the
method of exact couples yields a spectral sequence starting with E1n D T n .gr F X /.
def L
Here T n .gr F X / D k T .Fk X=FkC1 X / is viewed as a Z-graded object over F
n
(through the k-index). Hence this spectral sequence lives in the abelian category of
Z-graded objects over F . The expressions for Zrn and Brn are
M
Zrn D ker.T n .Fk X=FkC1 X / ! T nC1 .FkC1 X=FkCr X //;
k
M
Brn D im.T n1 .FkrC1 X=Fk X / ! T n .Fk X=FkC1 X //:
k
We now make a number of hypotheses.

(1) E, F have all limits and filtered limits are exact.
(2) .T n /n commutes with filtered limits.

(3) We have X D Fk X for k
0.
(4) X is complete. I.e. X D lim X=Fk X .
k
We note that limits and colimits on graded objects can be computed degreewise. Hence
Z1
n
exists and is equal to
M
Z1n
D ker.T n .Fk X=FkC1 X / ! T nC1 .FkC1 X //:
k
Similarly B1
n
exists and is equal to
M
B1
n
D im.T n1 .X=Fk X / ! T n .Fk X=FkC1 X //:
k
It is now well-known and an easy verification that

M im.T n .Fk X / ! T n X /
Z1
n
=B1
n
D :
im.T n .FkC1 X / ! T n X /
k
In other words if we equip T n X with the filtration Fk .T n X / D im.T n .Fk X / !

T n X/ then E1 n
Š gr F T n X . Note that the conditions also imply that T n X is complete
for this filtration. Indeed
lim T n X=Fk T n X D lim coker.T n .Fk X / ! T n X /

k k
D coker.T n .lim Fk X / ! T n X /

k
D T X:
n
Now revert to the notations in the statement of the proposition. We apply the pre-
vious discussion with E D Pro.C /, F D Pro.D/ and X D M . We equip M
with the J -adic filtration. By the above discussion we get a spectral sequence E
with E1n D T n .gr FJ M / which converges to gr F Ty n .M /. The terms occurring in
this spectral sequence are graded .gr J R; DR=J / objects. The limit is a priori only a
.gr J R; .Pro D/R=J /-object.
By Lemma 2.2.3 gr J M is a noetherian .gr J R; CR=J / object. Hence by Lem-
ma 2.3.2 below T n .gr J M / is a noetherian graded .gr J R; DR=J /-object. Hence the
ascending chain Brn must be stationary. By (2.5) we obtain drn D 0 for r 0. Hence E
degenerates at Ern for r 0. It follows that E1n
D gr F Ty n .M / is a noetherian graded
.gr J R; DR=J /-object.
Since we had already shown that Ty n .M / is complete we conclude by Lemma 2.2.4.
Lemma 2.3.2. Let S0 be a commutative noetherian ring and let S be a finitely generated
commutative positively graded S0 -algebra whose part of degree zero is S0 . Let .T i /i
be a @-functor between noetherian abelian S0 -linear categories E, F . Then for any
def L
noetherian graded .S; E/-object N and for any i we have that T i .N / D n T i .Nn /
is a noetherian graded .S; F /-object.
Proof. We perform induction on the minimal number of generators d of S as S0 -

algebra. If d D 0 then S D S0 and hence N is concentrated in a finite number of
degrees. In this case T i .N / is obviously noetherian.
Now assume d > 0 and pick a homogeneous generator t of S over S0 of strictly
positive degree. Then N can be written as an extension
0 ! N 0 ! N ! N 00 ! 0
where N 0 is annihilated by some power of t and N 00 is t -torsion free. The object N 0

can itself be written as a repeated extension of objects annihilated by t . Thus it suffices
to treat the cases where N is annihilated by t and where N is t -torsion free.
If N is annihilated by t then T i .N / is noetherian by induction (since N is now an
S=tS -module and S=tS has one generator less than S). Hence we assume that N is
t-torsion free. From the exact sequence
t
0 ! N ! N ! N=N t ! 0
we obtain an injection
T i .N /=T i .N /t ,! T i .N=N t /:
By induction T i .N=N t / is noetherian and hence so is T i .N /=T i .N /t . From this one

easily obtains that T i .N / itself is noetherian.
2.4 Formal flatness. The notations R, J , C are as above. For use below it would be
convenient to assume that C if flat. Unfortunately even if C is R-flat then there seems
to be no a priori reason for Cy to be flat (although we do not know an explicit counter
example). To work around this issue we make the following definition
Definition 2.4.1. The R-linear category C is formally flat if the categories CR=J n are
R=J n -flat for all n.
Since CyR=J n D CR=J n it immediately follows that if C is formally flat then so is C.

y
The following proposition yields a different characterization of formal flatness.
Proposition 2.4.2. Let C t be the full subcategory of objects in C that are annihilated
by some power of J . This is naturally an R-linear category. Then C is formally flat if
and only if C t is flat.
Proof. Assume first that C is formally flat. Take M 2 C t , N 2 mod.R/. We must

i .M; N / is effaceable in its first argument in C t for i > 0. We assume
prove that Tor R
M 2 CR=J n .
By dimension shifting in N we may reduce to the case i D 1. Take an exact
sequence in mod.R/
0 ! N0 ! P ! N ! 0
with P projective. Then we get an exact sequence
0
1 .M; N / ! M ˝R N ! M ˝R P ! M ˝R N ! 0:
0 ! Tor R (2.6)
Let F be the filtration on N 0 induced from the J -adic filtration on P . From
0 ! N 0 =Fl N 0 ! P =PJ l ! N=NJ l ! 0
we obtain an exact sequence (for l n)
l
0 ! Tor R=J
1 .M; N=NJ l / ! M ˝R=J l N 0 =Fl N 0 ! M ˝R=J l P =PJ l
! M ˝R=J l N=NJ l ! 0:
Combining these two sequences we see that there is a map
R=J l
1 .M; N / ! Tor 1
Tor R .M; N=NJ l / (2.7)
natural in M (for l n).
By the Artin–Rees condition (Lemma 2.2.9) we may take an l such that Fl N 0
0 n
N J . Then from (2.6) we obtain an exact sequence
0 0
0 ! Tor R
1 .M; N / ! M ˝R=J l N =Fl N ! M ˝R=J l P =PJ
l
! M ˝R=J l N=NJ l ! 0
and thus we have deduced that for l large (2.7) is an isomorphism
R=J l
1 .M; N / D Tor 1
Tor R .M; N=NJ l /: (2.8)
l
Since CR=J l is flat Tor R=J
1 .M; N=NJ l / is effaceable in its first argument in CR=J l
0
by an epimorphism M ! M . Thus we get a commutative diagram
0 / Tor R=J l .M 0 ; N=NJ l /
1 .M ; N /
Tor R 1

l
Tor R
1 .M; N/ Tor R=J
1 .M; N=NJ l /
with the right most map being zero. It follows that the left most map is also zero. Thus
1 .M; N / is effaceable in M in C t and hence C t is flat.
Tor R
Conversely assume C t is flat. Since C t;R=J n D CR=J n and flatness is stable under
base change ([12], Proposition 4.8) we conclude that CR=J n is flat.
By Lemma 2.2.11 the inclusion of abelian categories C t ! C yields an equivalence
Cyt Š Cy :
Hence when C is formally flat we may always reduce to the case that C is flat.
Flatness on the level of objects does not present any pitfalls as the following propo-
sition shows.
Proposition 2.4.3. Let M 2 Cy . Then M is R-flat if and only if M=MJ n is R=J n -flat
for all n.
Proof. We consider the non-obvious direction. Assume that M 2 Cy is such that all
M=MJ n are flat. We need to prove that M ˝R is exact.
Consider an exact sequence in mod.R/.
0 ! K ! L ! N ! 0:
We have to show that
0 ! M ˝R K ! M ˝R L ! M ˝R N ! 0
is exact. After tensoring with R=J n it is sufficient to show that
0 ! M ˝R K=KJ n ! M ˝R L=LJ n ! M ˝R N=NJ n ! 0
is exact up to essentially zero systems. This is the same sequence as
0 ! M=MJ n ˝R=J n K=KJ n ! M=MJ n ˝R=J n L=LJ n

! M=MJ n ˝R=J n N=NJ n ! 0:
Hence by flatness of M=MJ n it is sufficient that
0 ! K=KJ n ! L=LJ n ! N=NJ n ! 0
is exact up to essentially zero systems. This follows from the Artin–Rees condition
(see Lemma 2.2.9).
2.5 Ext-groups. Now we discuss Ext-groups, by which we always meanYoneda Ext-

groups. We keep the notations from the previous section. Thus .R; J / is a noetherian
J -adic ring and C is an R-linear noetherian abelian category.
We make the following definition
Definition 2.5.1. Let M; N 2 Cy . Then the completed Ext-groups between M , N are

defined by
‘ Ext iy .M; N / D Ext iPro.C t / .M; N /:

C
It is clear that ‘ Extiy .M; N / is a @-functor in both arguments. Apart from this nice
C
property we don’t know if completed Ext-groups are meaningful objects in general. To
get better control we will assume that C is formally flat and we impose an additional
finiteness condition
Definition 2.5.2. An R-linear abelian category D is Ext-finite if for all objects M; N 2
D we have that ExtiD .M; N / is a finitely generated R-module for all i .
The following will be the main result of this section.
Proposition 2.5.3. Assume that C is formally flat and that CR=J is Ext-finite. Then
y for M; N 2 Cy and furthermore
‘ Extiy .M; N / 2 mod.R/
C
‘ Extiy .M; N / D lim Ext Cy .M; N=NJ k / (2.9)

C
l
D lim lim Ext iC l .M=MJ l ; N=NJ k /: (2.10)

! R=J
k lk
If M is in addition R-flat then

‘ Extiy .M; N / D lim Ext iC k .M=MJ k ; N=NJ k /: (2.11)
C R=J
k
The proof is a series of lemmas.

Lemma 2.5.4. Assume that C is formally flat and that CR=J is R=J -Ext finite then
CR=J n is R=J n -Ext-finite for all n.
Proof. We need to prove that for M; N 2 CR=J n we have that ExtiCR=J n .M; N / is a
finitely generated R=J n -module. By filtering M we may assume M 2 CR=J . Then
we conclude by a change of rings spectral sequence (which depends on flatness, see
Proposition 4.7 of [12])
E2pq W Ext pCR=J .M; Ext qC .R=J; N // ) ExtpCq
C .M; N /:
R=J n n
R=J
The following lemma is proved in a similar way.

Lemma 2.5.5. Assume that C is formally flat and that CR=J is Ext-finite. Let M 2 Cy
and N 2 CR=J n . Then Ext iPro.C t / .M; N / is a finitely generated R=J n -module for
all i . Furthermore we have
ExtiPro.C t / .M; N / D lim Ext iC l .M=MJ l ; N / (2.12)
! R=J
l
and
ExtiPro.C t / .M; N / D Ext iy .M; N /: (2.13)
C
Finally if M is flat over R then
ExtiPro.C t / .M; N / D Ext iCR=J n .M=MJ n ; N /: (2.14)
Proof. If C t is flat then so is Pro.C t / (see Proposition 3.6 of [12] for the dual statement).
Now we use the spectral sequence (for l n)
E2pq .l/ W Ext pPro.C t / .Tor R pCq

q .M; R=J /; N / ) Ext Pro.C t / .M; N /
l
(2.15)
R=J l
which may derived in a similar way as [12], Proposition 4.7 (the existence depends on
flatness of Pro.C t /). The formation of Pro-objects commutes with certain base changes
(see [12], Proposition 4.5, for the dual statement) and in particular Pro.C t /R=J l D
Pro.CR=J l /.
Since Tor R .M; R=J l / lies in Cy and is annihilated by J l it lies in CR=J l . For
q
an object K 2 CR=J l we have Ext iPro.C l / .K; N / D Ext iC l .K; N / (see [12],
R=J R=J
Proposition 2.14). Hence the spectral sequence (2.15) becomes
E2pq .l/ W Ext pC .Tor R pCq

q .M; R=J /; N / ) Ext Pro.C t / .M; N /:
l
(2.16)
R=J l
To prove finite generation we put l D n and invoke Lemma 2.5.4.

To prove (2.12) we note that for l l 0 there are maps of spectral sequence E.l/ !
E.l 0 / which are given by the compositions
Ext pC .Tor R l p
q .M; R=J /; N / ! Ext C 0
.Tor R
q .M; R=J /; N /
l
R=J l R=J l
0
! ExtpC 0
.Tor R
q .M; R=J /; N /:
l
R=J l
It follows from Lemma 2.5.6 below that E2pq .l/ ! E2pq .l 0 / is zero for q > 0 and
l
l 0 (taking into account that M is a noetherian object in Cy ). Taking a direct limit
over l of (2.16) we find that indeed
lim ExtpC l .M=MJ l ; N / D Ext pCq .M; N /:

! R=J
Pro.C t /
l
The claim (2.13) follows from Lemma 2.5.7 below together with (2.12). The claim
(2.14) follows from the degeneration of the spectral sequence (2.16).
We have used the next two lemmas.
Lemma 2.5.6. Let D be an R-linear abelian category and assume that M 2 D is a

i .M; R=J /n is an essentially zero system for i > 0.
noetherian object. Then Tor R n
Proof. We first replace D with the smallest full abelian subcategory of D containing M .
This is a noetherian abelian category and being a full subcategory of D it is still R-
linear. Then we replace D by its category of Ind-objects. Then D becomes a locally
noetherian Grothendieck category and M is still a noetherian object in D. In particular
ExtiD .M; / commutes with filtered colimits.
As a Grothendieck category D has enough injectives. Hence we have to show that

for an arbitrary injective object E 2 D we have
lim HomD .Tor R

i .M; R=J /; E/ D 0:
n
!
n
Now we have
lim HomD .Tor R

i .M; R=J /; E/ D lim Ext D .M; Hom R .R=J ; E//
n i n
! !
n n
D ExtiD .M; lim HomR .R=J n ; E//:

!
n
Thus we have to show that F D lim HomR .R=J n ; E/ is injective. In a locally noethe-
!n
rian Grothendieck category we can test this on inclusions of noetherian objects. Hence
let K ,! M be such an inclusion. We need to prove that lim Hom.M=MJ n ; E/ !
!n
lim Hom.K=KJ n ; E/ is an epimorphism, or equivalently that the kernel of K=KJ n !
!n
M=MJ n is an essentially zero system. This follows from the Artin–Rees condition
(Lemma 2.2.9).
Lemma 2.5.7. Let D be a noetherian R-linear category and assume that N 2 D is

annihilated by J n . Let M 2 D. Then we have
ExtiD .M; N / D lim ExtiD l .M=MJ l ; N /:

! R=J
l
Proof. From the Artin–Rees condition (Lemma 2.2.9) we know that . ˝R R=J l /l
is exact up to essentially zero systems. From this we obtain in the usual way that
M 7! lim ExtiD l .M=MJ l ; N / is a @-functor with values in R-modules. To show
!l R=J
that this @-functor coincides with ExtiD .M; N / it is sufficient to prove this for i D 0
and to show that any element of lim Ext iD l .M=MJ l ; N / is effaceable for i > 0.
!l R=J
The case i D 0 is trivial so assume that a 2 Ext iD l .M=MJ l ; N / represents an
R=J
element aN of lim ExtiD l .M=MJ l ; N / for i > 0.
!l R=J
There exists an epimorphism T ! M=MJ l which effaces a for some T 2 DR=J l .
Let M 0 be the pullback of T for the map M ! M=MJ l . Then the epimorphism
M 0 =M 0 J l ! M=MJ l factors through T and hence effaces a. This means that
M 0 ! M effaces a. N
Lemma 2.5.8. Assume that C is formally flat and that CR=J is Ext-finite and let
y Then
M; N 2 C.
y
lim ExtiPro.C t / .M; N=NJ n / 2 mod.R/O .Š mod.R//:

n
Proof. This follows by applying Theorem 2.3.1 to the @-functor
ExtiPro.C t / .M; /i W C t ! mod.R/ t :
To construct this functor we use Lemma 2.5.5.
Proof of Proposition 2.5.3. Let P be a projective resolution of M in Pro.C t /. Since

C t is flat by Proposition 2.4.2 the Pm are R-flat (see Proposition 3.4 in [12] for the
dual version). We compute
ExtiPro.C t / .M; N / D H i .HomPro.C t / .P ; lim N=NJ n //

n
D H i .lim HomPro.C t / .P ; N=NJ n //:

n
We need to exchange H i and lim . This is possible if the terms of the inverse system
n
of complexes HomPro.C t / .P ; N=NJ n / as well as its cohomology satisfy the Mittag-
Leffler condition. For the terms this follows from the projectivity of Pm . For the
cohomology which is equal to ExtiPro.C t / .M; N=NJ n / we invoke Lemma 2.5.8 together
with Lemma 2.5.9 below.
Assuming this we now obtain
ExtiPro.C t / .M; N / D lim H i .HomPro.C t / .P ; N=NJ n //

n
D lim ExtiPro.C t / .M; N=NJ n /:

n
This implies (2.9) via (2.13). Furthermore we obtain (2.10) via (2.14). Finally we
obtain (2.11) from (2.13).
Lemma 2.5.9. Let .Un /n be an inverse system in an R-linear noetherian abelian

y Then
category D such that Un J n D 0 and such that the pro-object .Un /n lies in D.
.Un /n satisfies the Mittag-Leffler condition.
y it follows that U=UJ n 2 DR=J n .
Proof. Let U be the pro-object .Un /n . As U lies in D
Define Cn , Kn 2 DR=J n as the kernel and cokernel of the natural maps.
U=UJ n ! Un :
Taking inverse limits in Pro.mod.R// we see that .Kn /n and .Cn /n are zero pro-objects,
or equivalently they are essentially zero systems. From this one easily deduces that
.Un /n satisfies the Mittag-Leffler condition.
2.6 The complete derived category. We use the same notations as above. In particu-
lar .R; J / is a J -adic noetherian ring and C is an R-linear noetherian abelian category.
We define Dc .C/ as the full subcategory of D.Pro.C t // of complexes whose coho-
mology lies in Cy . Thus Dc .C / has a t -structure whose heart is Cy . Then the completed
Ext-groups for M; N 2 Cy may be reinterpreted as
‘ ExtCy .M; N / D HomDc .C/ .M; N Œn/:
In the case that C is the category of torsion l-adic constructible sheaves it would be
interesting to compare this derived category to the standard derived category of l-adic
sheaves [4], [5], [7].
3 Formal deformations of abelian categories

Let R be a noetherian J -adic ring and C a flat R=J -linear noetherian abelian category.
Then we define an R-deformation of C to be a formally flat complete R-linear abelian
category D together with an equivalence DR=J Š C . It follows from the above
discussion that D is specified up to equivalence by specifying the flat R=J n -linear
categories Dn D DR=J n together with the equivalences (isomorphisms in this case)
Dm D Dn;R=J m for n m and D1 Š C .
Denote by Def R .C / the class of R-deformations of C. This is a 2-groupoid. The
observations in the previous paragraph may be used to construct a 2-equivalence
Def R .C / Š 3- lim Def R=J n .C/:

n
We leave it to the interested reader to formalize this statement. It will not be used in
this form.
4 Ampleness
We define what we mean by a strongly ample sequence. This is stronger than strictly
necessary but easier to work with.
Let E an noetherian abelian category. For us a sequence .O.n//n2Z of objects in
E is strongly ample if the following conditions hold.
(A1) For all M 2 E and for all n there is an epimorphism ˚tiD1 O.ni / ! M with
ni n.
(A2) For all M 2 E and for all i > 0 one has ExtiE .O.n/; M / D 0 for n 0.
A strongly ample sequence .O.n//n2Z in E is ample in the sense of [14]. Hence using
the methods of [2] or [14] one obtains E Š qgr.A/ if E is Hom-finite, where A is the
noetherian Z-algebra ˚ij HomE .O.j /; O.i //.
Below we fix a complete noetherian J -adic ring R. The following is a version of
Grothendieck’s existence theorem [8].
Proposition 4.1. Let E be an Ext-finite R-linear noetherian category with a strongly

ample sequence .O.n//n . Then
(1) E is complete;
(2) if in addition E is flat then we have for M; N 2 E:
Ext iE .M; N / D ‘ ExtiE .M; N /: (4.1)
Proof. Step 1. We first claim that E satisfies Nakayama’s lemma. This would follow
from Lemma 2.2.14 once we knew E is complete but we are not there yet.
Let M 2 E be such that MJ D M . Pick generators for a1 ; : : : ; an for J and
consider the corresponding epimorphisms
.ai /i
M ˚n ! M ! 0:
Applying HomE .O.m/; / for m large we get an epimorphism
.ai /i
HomE .O.m/; M /˚n ! HomE .O.m/; M / ! 0
and thus by Nakayama’s lemma for R and Ext-finiteness
HomE .O.m/; M / D 0
for large m. It then follows from (A1) that M D 0.
Step 2. Let M be an object in E. We claim that for i > 0 and for large m we have
ExtiE .O.m/; MJ n / D 0 (4.2)
for all n and furthermore for large m we also have
HomE .O.m/; MJ n / D HomE .O.m/; M /J n (4.3)
for all n.
Put S D R˚J ˚J 2 ˚ . According to Lemma 2.2.3 N D M ˚MJ ˚MJ 2 ˚
is a noetherian graded .S; E/-object. We now apply Lemma 4.2 below with Tmi D
ExtE .O.m/; / using (A2).
Step 3. Now we show that the functor
ˆ W E ! Ey
is an equivalence. We first check fully faithfulness. If M , N 2 E then since ˆ.M / is
the pro-object M=MJ n and similarly for N we deduce from (2.1) we have
HomEy .ˆ.M /; ˆ.N // D lim lim HomER=J l .M=MJ l ; N=NJ k /
!
k l
D lim HomE .M; N=NJ k /:

k
Thus we have to show that the natural map

HomE .M; N / ! lim HomE .M; N=NJ n / (4.4)

n
is an isomorphism. Using left exactness of lim and (A1) we immediately reduce to

M D O.m/ where m may be chosen arbitrarily large.
We first observe that (4.4) must be a monomorphism. Suppose on the contrary that
there is an f W M ! N such that K D im f NJ n for all n. Then by Lemma 2.2.9
we have K D KJ . Hence K D 0 by Step 1 which implies f D 0.
Now we prove that (4.4) is an epimorphism when M D O.m/ for m large.
Suppose we are given a compatible system of maps fn W M ! N=NJ n . By (4.2) the
0
fn may be lifted to maps fn0 W M ! N such that im.fn0 fnC1 / NJ n . Using (4.2)
0 0
this implies fn fnC1 2 HomE .M; N /J . Since HomE .M; N / 2 mod.R/ the limit
n
f D limn fn0 exists. It has the property that the image of f in HomE .M; N=NJ n / is
equal to fn . This proves what we want.
Now we prove essential surjectivity. Let M 2 E.y We will construct an epimorphism
M
t
ˆ.O.ni // ! M ! 0 (4.5)
iD1
for suitable ni . Repeating this construction for the kernel we obtain a presentation
M
s
f M
t
ˆ.O.nj0 //
! ˆ.O.ni // ! M ! 0:
j D1 iD1
By fully faithfulness (already proved) we have f D ˆ.g/. By exactness of ˆ (see

Proposition 2.2.10) we obtain M Š ˆ.coker.g//.
We now proceed with constructing (4.5). We have MJ n =MJ nC1 2 E. We claim
that for i > 0 and for all n we have
ExtiE .O.m/; MJ n =MJ nC1 / D 0 (4.6)
for mL 0. This follows L from Lemma 4.2 to be shown below by considering
S D n J n =J nC1 , N D n MJ n =MJ nC1 (which is noetherian by Lemma 2.2.3)
and Tmi D ExtiE .O.m/; /.
Take m large enough so that the indicated vanishing holds. By (A1) we may choose
an epimorphism
Mt
O.ni / ! M=JM ! 0
iD1
with ni m. Using (4.6) for i D 1 this epimorphism may be lifted to compatible
epimorphisms
M t
O.ni / ! M=J n M ! 0
iD1
and hence to compatible epimorphisms in
M
t
O.ni /=J n O.ni / ! M=J n M ! 0:
iD1
Taking the inverse limit in Pro.E/ we get an epimorphism as in (4.5), finishing the
proof of essential surjectivity.
Step 4. Finally we prove (4.1). To show that ‘ Ext D Ext in E it is sufficient to
show that ‘ Hom D Hom and furthermore that ‘ Ext is effaceable in its first argument.
The fact that ‘ Hom D Hom is the fact that (4.4) is an isomorphism. So let us show
that ‘ Ext is effaceable.
Let N be an object in E. If m is large and i > 0 then it follows from (4.2) that
Ext E .O.m/; N=NJ n / D 0 for all n. Hence ‘ ExtiE .O.m/; N / D 0 by (2.9). Now
i
L
let M 2 E. To efface ‘ ExtiE .M; N / we take an epimorphism tiD1 O.ni / ! M
with the ni sufficiently large. This finishes the proof.
Lemma 4.2. Let S0 be a commutative noetherian ring and let S be a finitely generated
positively commutative graded S0 -algebra whose part of degree zero is S0 . Let .Tmi /i
be a sequence, indexed by m, of @-functors between noetherian abelian S0 -linear
categories E, F such that for all M 2 E and for all i > 0 one has Tmi .M / D 0 for
m 0. Then for any noetherian graded .S; E/-object N and for any i > 0 we have
that Tmi .N / D 0 for m 0. Furthermore if N is generated in degree zero (by the
S -action) then so is Tm0 .N / for m 0.
Proof. The proof is very similar to the proof of Lemma 2.3.2. We perform induction
on the minimal number of generators d of S as S0 -algebra. We first prove that for
i > 0 one has Tmi .N / D 0 for m 0.
If d D 0 then S D S0 and hence N is concentrated in a finite number of degrees.
In this case there is nothing to prove.
Now assume d > 0 and pick a homogeneous generator t of S over S0 of strictly
positive degree. Then N can be written as an extension
0 ! N 0 ! N ! N 00 ! 0
where N 0 is annihilated by some power of t and N 00 is t -torsion free. The object N 0

can itself be written as a repeated extension of objects annihilated by t . Thus it suffices
to treat the cases where N is annihilated by t and where N is t -torsion free.
If N is annihilated by t then Tmi .N / D 0 for m 0 by induction (since N is now
an S=tS-module and S=tS has one generator less than S ). Hence we assume that N
is t -torsion free. From the exact sequence
t
0 ! N ! N ! N=N t ! 0
and the fact that by induction Tmi .N=N t / D 0 for m 0 we obtain for m 0 a
surjection
t
Tmi .N / ! Tmi .N / ! 0:
From the fact that t has strictly positive degree we obtain Tmi .N / D 0.
To prove the last claim assume that N is generated in degree zero. We perform
again induction on d . If d D 0 then N must be concentrated in degree zero and there
is nothing to prove. If d ¤ 0 then we pick t as above.
By the vanishing part of the lemma which has already been proved we see that Tm0
sends a right exact sequence to a right exact sequence for m 0. Thus for m 0 the
following sequence is exact
t
Tm0 .N / ! Tm0 .N / ! Tm0 .N=N t / ! 0:
We also have an exact sequence

t
S ˝S0 Tm0 .N0 / ! S ˝S0 Tm0 .N0 / ! S=S t ˝S0 Tm0 .N0 / ! 0
and by induction the canonical map S=tS ˝S0 Tm0 .N0 / ! Tm0 .N=N t / is a epimor-
phism.
Put C D coker.S ˝S0 Tm0 .N0 / ! Tm0 .N //. Using a bit of diagram chasing we
obtain C D C t and hence C D 0 since t has strictly positive degree.
5 Lifting and base change

The usual lifting results for infinitesimal deformations generalize without difficulty to
formal deformations. We collect some results here which were used in [16], [17].
As usual R is a J -adic noetherian ring and we assume that D is an R-deformation
of a noetherian Ext-finite R=J -linear flat abelian category C . To simplify the notation
we assume C D DR=J .
First we discuss some properties which are useful for lifting objects.
Proposition 5.1. Let M 2 C be a flat object such that Ext iC .M; M ˝R=J J n =J nC1 / D
0 for i D 1; 2 and n 1. Then there exists a unique R-flat object (up to non-unique
isomorphism) M x 2 D such that M
x =M x J Š M.
Proof. This follows in a straightforward way from the infinitesimal lifting criterion for
objects (see [13], Theorem A, for the dual version).
Proposition 5.2. Let Mx;N x 2 D be flat objects and put M

x =M
x J D M, N
x =N
xJ D N.
Assume that for all X in mod.R=J / we have ExtC .M; N ˝R=J X / D 0 for a certain
i
x;N
i > 0. Then we have ‘ ExtiD .M x ˝R X / D 0 for all X 2 mod.R/.
Proof. This follows in a straightforward way from Proposition 6.13 of [12].

Proposition 5.3 (“base change”). Let M x;N x 2 D be flat objects and put Mx =M
x J D M,
x x
N =N J D N . Assume that for all X in mod.R=J / we have ExtC .M; N ˝R=J 1
X/ D 0. Then HomD .M x;Nx / is R-flat and furthermore for all X in mod.R/ we have
HomD .M x;Nx ˝R X / D HomD .M x;N x / ˝R X .
Proof. This is routine. Choose a short exact sequence
0!Y !P !X !0
with P a finitely generate projective. We then get an exact sequence

x;N
0 ! HomD .M x ˝R Y / ! HomD .M
x;N
x ˝R P /
x;N
! HomD .M x ˝R X / ! Ext1 .M
x;N
x ˝R Y /:
D
x;N
By the Proposition 5.2 and the hypotheses we get Ext1D .M x ˝R Y / D 0. We then
obtain a commutative diagram
0 / HomD .M
x;N x ˝R Y / / HomD .M
x;N
x ˝R P / / HomD .M
x;N x ˝R X/ /0
O O
˛Y ˛X
x;N
HomD .M x / ˝R Y / HomD .M
x;N
x / ˝R P / HomD .M
x;N
x / ˝R X / 0.
We obtain that ˛X is an epimorphism for all X . But then ˛Y is an epimorphism from

which we then deduce that ˛X is an isomorphism. But then ˛Y is an isomorphism and
hence the lower right exact sequence is in fact exact. Thus HomR .M x;N
x / is R-flat,
finishing the proof.
Next we show that ampleness lifts well.
Theorem 5.4. Let O.n/n be a sequence of R-flat objects in D. Then
(1) .O.n/=O.n/J /n is strongly ample in C if and only if O.n/n is strongly ample in
D.
(2) If .O.n/=O.n/J /n is strongly ample then D is flat (instead of just formally flat).
y
(3) D is Ext-finite as R-linear category.
Proof. Property (2) follows immediately from (A1) and (1). To prove (1) we note that
since the O.n/ are flat we have
ExtiD .O.n/; M / D ExtiC .O.n/=O.n/J; M / (5.1)
for M 2 C. From this it is easy to see that if O.n/n is strongly ample in C then
.O.n/=O.n/J /n is strongly ample in C . So our main task is to prove the converse.
So assume that .O.n/=O.n/J /n is strongly ample. We will first show that O.n/n
is strongly ample if we replace Ext by ‘ Ext in the definition of strongly ample.
We claim that for m large we have
ExtiD .O.m/; MJ n =MJ nC1 / D 0 for all n. (5.2)
This is proved in the same way as (4.6).

In particular we find that for m large one has
ExtiD .O.m/; M=MJ n / D 0 for all n.
Taking inverse limits and using Proposition 2.5.3 we find ‘ ExtiD .O.m/; M / D 0 for
all i > 0. Hence this proves (A2) for ‘ Ext.
We now prove (A1) (which does not involve any Ext). We first find m0 such
that (5.2) holds for i D 1 and m m0 . Using (A1) for C we find that there is an
def L
epimorphism F D tiD1 O.ni / ! M=MJ with ni m0 . Using (5.2) for i D 1
may lift this map to a compatible series of maps F ! M=MJ n . Taking the inverse
limit yields a map F ! M . By Nakayama’s lemma (see Lemma 2.2.14) it follows
that this must be an epimorphism.
So we have proved ampleness with ‘ Ext replacing Ext. Now we claim that in
fact ‘ Ext D Ext in D. This is proved in the same way as Step 4 of the proof of
Proposition 4.1 (using condition (A1) which was already proved).
Property (3) follows from Proposition 2.5.3.
References
[1] M. Artin, A. Grothendieck, and J. L. Verdier, Théorie des topos et cohomologie étale des
schémas, SGA4, Tome 2. Lecture Notes in Math. 269, Springer-Verlag, Berlin 1972.
[2] M. Artin and J. Zhang, Noncommutative projective schemes. Adv. in Math. 109 (1994),
no. 2, 228–287.
[3] M. Artin and J. Zhang, Abstract Hilbert schemes. I. Algebr. Represent. Theory 4 (2001),
no. 4, 305–394.
[4] K. A. Behrend, Derived l-adic categories for algebraic stacks. Mem. Amer. Math. Soc. 163
(2003), no. 774.
[5] A. Beilinson, J. Bernstein, and P. Deligne, Faisceaux pervers. Astérisque 100, Soc. Math.
France, Paris 1983.
[6] O. De Deken and W. Lowen, Abelian and derived deformations in the presence of Z-
generating geometric helices. J. Noncommut. Geom. 5 (2011), no. 4, 477–505.
[7] T. Ekedahl, On the adic formalism. In The Grothendieck Festschrift, vol. 2, Birkhäuser,
Boston, MA, 1990, 197–218.
[8] A. Grothendieck, Éléments de géométrie algébrique. III. Étude cohomologique des fais-
ceaux cohérents. I. Inst. Hautes Études Sci. Publ. Math. 11 (1961).
[9] D. Huybrechts, E. Macri, and P. Stellari, Formal deformations and their categorical general
fibre. Comment. Math. Helv. 86 (2011), no. 1, 41–71.
[10] J. P. Jouanolou, Systèmes projectifs J -adiques. In Cohomologie l-adique et fonctions L,

SGA5 (Berlin), Lecture Notes in Math. 589, Springer-Verlag, Berlin 1977.
[11] W. Lowen and M. Van den Bergh, Hochschild cohomology of abelian categories and ringed
spaces. Adv. Math. 198 (2005), no. 1, 172–221.
[12] W. Lowen and M. Van den Bergh, Deformation theory of abelian categories. Trans. Amer.
Math. Soc. 358 (2006), no. 12, 5441–5483.
[13] W. Lowen, Obstruction theory for objects in abelian and derived categories. Comm. Algebra
33 (2005), no. 9, 3195–3223.
[14] A. Polishchuk, Noncommutative proj and coherent algebras. Math. Res. Lett. 12 (2005),
no. 1, 63–74.
[15] J. T. Stafford and M.Van den Bergh, Noncommutative curves and noncommutative surfaces.
Bull. Amer. Math. Soc. (N.S.) 38 (2001), no. 2, 171–216.
[16] M. Van den Bergh, Non-commutative P 1 -bundles over commutative schemes. Trans. Amer.
Math. Soc. 364 (2012), 6279–6313.
[17] M. Van den Bergh, Non-commutative quadrics. Internat. Math. Res. Notices 2011 (2011),
no. 17, 3983–4026.
List of Contributors
Marcello Bernardara, Institut de Mathématiques de Toulouse, Université Paul Sabatier,

118, route de Narbonne, 31062 Toulouse CEDEX 9, France
mbernard@math.univ-toulouse.fr
Michele Bolognesi, IRMAR, Université de Rennes 1, 263 Av. Général Leclerc, 35042
Rennes CEDEX, France
michele.bolognesi@univ-rennes1.fr
Alberto Canonaco, Dipartimento di Matematica “F. Casorati”, Università degli Studi
di Pavia, Via Ferrata 1, 27100 Pavia, Italy
alberto.canonaco@unipv.it
Sabin Cautis, Department of Mathematics, University of Southern California, Los An-
geles, CA 90089-2532, U.S.A.
cautis@usc.edu
Akira Ishii, Department of Mathematics, Graduate School of Science, Hiroshima Uni-
versity, 1-3-1 Kagamiyama, Higashi-Hiroshima, 739-8526, Japan
akira@math.sci.hiroshima-u.ac.jp
Dmitry Kaledin, Department of Algebraic Geometry, Steklov Mathematical Institute,
119991 Moscow, Russia
kaledin@mi.ras.ru
Bernhard Keller, Université Paris Diderot – Paris 7, Institut universitaire de France,
U.F.R. de mathématiques, Institut de mathématiques de Jussieu, U.M.R. 7586 du CNRS,
Case 7012, Bâtiment Chevaleret, 75205 Paris, France
keller@math.jussieu.fr
Izuru Mori, Department of Mathematics, Shizuoka University, 836 Ohya Suruga-ku,
Shizuoka, 422-8529, Japan
simouri@ipc.shizuoka.ac.jp
Alexander Polishchuk, Department of Mathematics, University of Oregon, Eugene, OR
97403, U.S.A.
apolish@uoregon.edu
Mihnea Popa, Department of Mathematics, University of Illinois at Chicago, 851 S.
Morgan Street, Chicago, IL 60607, U.S.A.
mpopa@math.uic.edu
Christian Schnell, Kavli Institute for the Physics and Mathematics of the Universe, The
University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa-shi, Chiba 277-8583, Japan
christian.schnell@ipmu.jp
346 List of Contributors
Paolo Stellari, Dipartimento di Matematica “F. Enriques”, Università degli Studi di

Milano, Via Cesare Saldini 50, 20133 Milano, Italy
paolo.stellari@unimi.it
Yukinobu Toda, Kavli Institute for the Physics and Mathematics of the Universe, The
University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa, 277-8583, Japan
yukinobu.toda@ipmu.jp
Kazushi Ueda, Department of Mathematics, Graduate School of Science, Osaka Uni-
versity, Machikaneyama 1-1, Toyonaka, Osaka, 560-0043, Japan
kazushi@math.sci.osaka-u.ac.jp
Michel Van den Bergh, Departement WNI, Hasselt University, Agoralaan, 3590 Diepen-
beek, Belgium
michel.vandenbergh@uhasselt.be

Derived Categories in Algebraic Geometry - Yujiro Kawamata

Uploaded by

Document Information

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

Derived Categories in Algebraic Geometry - Yujiro Kawamata

Uploaded by

Copyright:

Available Formats

EMS Series of Congress Reports

Yujiro Kawamata Department of Mathematics

© 2012 European Mathematical Society

Typeset using the authors’ TEX files: I. Zimmermann, Freiburg

Categorical representability and intermediate Jacobians of Fano threefolds

Fourier–Mukai functors: a survey

Flops and about: a guide

A note on derived categories of Fermat varieties

Homology of infinite loop spaces

Cluster algebras and derived categories

Some derived equivalences between noncommutative schemes and algebras

Lagrangian-invariant sheaves and functors for abelian varieties

Generic vanishing filtrations and perverse objects in derived categories

The fundamental group is not a derived invariant

Introduction and open problems of Donaldson–Thomas theory

Notes on formal deformations of abelian categories

List of Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

Bolognesi treats the conjectural correspondence by Orlov between the semi-orthogonal

Tokyo, November 2012 Yujiro Kawamata

Notation. Any triangulated category is assumed to be essentially small. Given a

2 Categorical and classical representabilities for smooth

Definition 2.1 ([17], [18]). A semiorthogonal decomposition of T is a sequence of

• HomT .Ai ; Aj / D 0 for all i > j and any Ai in Ai and Aj in Aj ;

Such a decomposition will be written

Definition 2.2 ([16]). An object E of T is called exceptional if HomT .E; E/ D K, and

T D hA? ; Ai D hA;? Ai;

In [46], Kuznetsov suggest a definition for which the resolution of singularities Dz

geometrical resolution of singularities Xz ! X , whenever X has rational singulari-

Notice that any fully faithful functor F W Db .X / ! Db .Y / between the derived

Remark 2.8. If X and Yi are smooth projective and

Db .X / D hDb .Y1 /; : : : ; Db .Yk /i;

and the Riemann–Roch Theorem gives an isomorphism of Q-vector spaces

Proposition 2.9 ([61]). Let X be smooth projective and Z  X a smooth subvariety

Db .Xz / D h" Db .X /; Db .Z/1 ; : : : ; Db .Z/d 1 i;

where Db .Z/i is equivalent to Db .Z/ for all i D 1; : : : ; d 1.

z W A1Q ./ ! AiQ .X /:

The variety X is rationally representable if AiQ .X / is rationally representable for

Definition 2.13. Let A be an abelian variety. A group homomorphism g W AiZ .X / ! A

where I.z/ in Corr.T / is the composition of the correspondences z 2 Corr.T; X / and

equivalence is the usual Künneth component. For example, if K D C, we require

h.X/ Š 1 ˚ h1 .X / ˚ L˚b ˚ .h1 .J /.1// ˚ .L2 /˚b ˚ h5 .X / ˚ L3 ;

3 Interactions between categorical and classical

Conjecture 1.1. Let X and Y be smooth projective varieties and ˆ W Db .Y / ! Db .X /

(?) X is a smooth projective rationally representable threefold with h1 .X / D 0 and

Db .X / D hDb .1 /; : : : ; Db .k /; E1 ; : : : ; El i:

Corollary 3.10. Suppose X satisfies .\/, is a smooth projective curve of positive

1) Fano varieties of index 2: X is a smooth cubic [21].

3) Conic bundles: X is a standard conic bundle X ! P 2 degenerating along a

4) Del Pezzo fibrations: X ! P 1 non-rational of degree four [1].

Notice that if X is a smooth cubic threefold, the equivalence class of a notable

Remark 3.12 (Threefolds with X < 0 categorically representable

Db .X / D hDb .1 /; : : : ; Db .k /; E1 ; : : : ; El i;

with Ei exceptional objects.

4 Categorical representability and rationality: further

4.1 Surfaces. If X is a smooth projective rational surface, then it is categorically

threefolds which are categorically representable in codimension 2 or non-rational ones

1) If X is rational, is X categorically representable in codimension 2?

where Ei are exceptional objects and a complete intersection of a quadric and a

W Xz ! X its blow-up. The exceptional locus ˛ W Q ,! Xz of

where AzX is a categorical resolution of AX , as in [49], Lemma 5.8. The representability

where ˆ W Db ./ ! Db .Xz / is fully faithful. Now as in Lemma 5.3 of [49], if we

Db .Xz / D hOXz .3h C D/; OXz .2h C D/; B; OXz i: (4.2)

zX are equivalent, following exactly the same path of

where Ei are exceptional objects and AS is the orthogonal complement in Db .S / of

Proposition 2.9 ([61]). Let X be smooth projective and Z X a smooth subvariety