Sie sind auf Seite 1von 81

# The Jantzen sum formula at the

critical level

zur

## Dr. rer. nat

vorgelegt von
Johannes Kübel
aus Ulm
Als Dissertation genehmigt

## Prof. Catherine Meusburger, Ph. D.

Meinen Eltern
Zusammenfassung

Die vorliegende Arbeit beschäftigt sich mit der Darstellungstheorie einer symmetri-
sierbaren Kac-Moody Algebra. Hierbei beschränken wir uns auf die Höchstgewichts-
moduln, deren Struktur durch die BGG Kategorie O gegeben ist. Im Falle einer
affinen Kac-Moody Algebra unterscheidet man zwischen Darstellungen mit posi-
tivem, kritischem und negativem Level. Eine wichtige Rolle zur Bestimmung von
Charakteren einfacher Höchstgewichtsmoduln spielen hierbei die Vermamoduln im
positiven und negativen Level. Im kritischen Level ist es hilfreich, die restringierte
Kategorie O, eine gewisse Unterkategorie der üblichen Kategorie O, einzuführen.
Dabei werden die üblichen Vermamoduln durch restringierte Vermamoduln ersetzt,
die als maximale restringierte Quotienten der Vermamoduln auftauchen.
Die Jantzen-Filtrierung liefert ein gutes Hilfsmittel zur Berechnung von einfachen
Subquotienten in einem Vermamodul. Im Falle eines Vermamoduls im positiven
Level zeigen wir, dass die Struktur der Subquotienten dieser Filtrierung durch Ko-
effizienten gewisser Kazhdan-Lusztig Polynome gegeben ist, indem wir sie mit der
Andersen-Filtrierung im negativen Level in Verbindung bringen.
Im kritischen Level definieren wir analog eine Jantzen-Filtrierung auf den restrin-
gierten Vermamoduln und beweisen eine alternierende Summenformel für diese. Als
Korollar hiervon erhalten wir das ‘Linkage Principle’, welches schon von Arakawa
und Fiebig bewiesen wurde. Ein Spezialfall der Summenformel für subgenerische,
restringierte Vermamoduln erlaubt es uns, das Zentrum eines deformierten Blocks
der restringierten Kategorie O zu berechnen.

## Diese Arbeit beruht auf folgenden Teilpublikationen:

1. J. Kübel, Centers for the restricted category O at the critical level over affine
Kac-Moody algebras, preprint 2013, arXiv: 1301.6016, erscheint in Mathematische
Zeitschrift.

2. J. Kübel, Jantzen sum formula for restricted Verma modules over affine Kac-
Moody algebras at the critical level, preprint 2012, arXiv: 1212.0150, erscheint in
Pacific Journal of Mathematics.

## 3. J. Kübel, Tilting modules in category O and sheaves on moment graphs, Journal

of Algebra 371 (2012), 559–576.

## 4. J. Kübel, From Jantzen to Andersen filtration via tilting equivalence, Mathema-

tica Scandinavica 110 (2012), no. 2, 161–180.

i
Abstract

The present thesis deals with the representation theory of a symmetrizable Kac-
Moody algebra. We restrict ourselves to the highest weight modules whose struc-
ture is given by the BGG category O. In the case of an affine Kac-Moody algebra,
one distinguishes between representations of positive, negative and critical level. An
important role for the calculation of characters of simple highest weight modules is
played by the Verma modules in positive and negative level. At the critical level, it
is helpful to introduce the restricted category O, a certain subcategory of the usual
category O. The usual Verma modules are then replaced by restricted Verma mod-
ules that appear as maximal restricted quotients of aforementioned Verma modules.
The Jantzen filtration provides a good tool for studying simple subquotients of a
Verma module. In case of a Verma module in positive level, we show that the struc-
ture of subquotients of this filtration is given by the coefficients of certain Kazhdan-
Lusztig polynomials, by associating it with the Andersen filtration in negative level.
At the critical level, we define an analogue of the Jantzen filtration for restricted
Verma modules and prove an alternating sum formula for it. As a corollary from this
we get the ‘Linkage Principle’, which was already proven by Arakawa and Fiebig. A
special case of the sum formula for subgeneric restricted Verma modules allows us
to calculate the center of a deformed block of the restricted category O.

## This thesis is based on the following publications:

1. J. Kübel, Centers for the restricted category O at the critical level over affine
Kac-Moody algebras, preprint 2013, arXiv: 1301.6016, accepted in Mathematische
Zeitschrift.

2. J. Kübel, Jantzen sum formula for restricted Verma modules over affine Kac-
Moody algebras at the critical level, preprint 2012, arXiv: 1212.0150, to appear in
Pacific Journal of Mathematics.

## 3. J. Kübel, Tilting modules in category O and sheaves on moment graphs, Journal

of Algebra 371 (2012), 559–576.

## 4. J. Kübel, From Jantzen to Andersen filtration via tilting equivalence, Mathema-

tica Scandinavica 110 (2012), no. 2, 161–180.

ii
Danksagung

Mein Dank richtet sich an erster Stelle an Peter Fiebig, der mich schon während
des Studiums in Freiburg betreute und mich in das Themenfeld der vorliegenden
Dissertation einarbeitete. Seine stets freundliche Betreuung und Hilfsbereitschaft
schufen ein äußerst angenehmes Arbeitsumfeld in Erlangen. Weiter danke ich den
Mitarbeitern des Emmy-Noether-Zentrums, die den Alltag im mathematischen In-
stitut bereicherten. Allen voran möchte ich hierbei Narendiran nennen, der sowohl
mathematisch als auch als Freund meine Zeit in Freiburg während des Studiums
und in Erlangen und Nürnberg während der Promotion erheblich prägte. Belinda
Echtermeyer möchte ich für ihre hilfreiche Unterstützung zu administrativen Fragen
während der Promotionszeit danken. Auf zahlreichen Konferenzen konnte ich mich
mit mehreren Mathematikern austauschen. Dabei hatten Diskussionen mit Wolf-
gang Soergel und Geordie Williamson indirekten Einfluss auf diese Arbeit. Durch
meine Familie und Shaniah erfuhr ich stets Unterstützung während der letzten drei
Jahre, was ich sehr zu schätzen weiß. Ein Auslandsaufenthalt in Amsterdam wäre
ohne das Einverständnis durch Eric Opdam nicht möglich gewesen. Dafür und für
seine Betreuung in Amsterdam danke ich ihm herzlich. Schließlich möchte ich mich
beim DFG Schwerpunkt 1388 Darstellungstheorie für die finanzielle Unterstützung
meiner Promotion bedanken.

iii
Contents

Introduction vi

1 Kac-Moody algebras 1

2 Deformed Category O 4
2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1.1 Duality on OA . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.2 Deformed standard and costandard modules . . . . . . . . . . 5
2.1.3 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.4 Simple objects . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Projective objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Block decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Subgeneric blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.5 Tilting modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.6 Tilting equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6.1 The semi-regular bimodule . . . . . . . . . . . . . . . . . . . . 13
2.7 Description of blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

## 3 Andersen and Jantzen filtrations 18

3.1 Jantzen filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Andersen filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Connecting both filtrations . . . . . . . . . . . . . . . . . . . . . . . . 20

## 4 Combinatorics of modules with a Verma flag 24

4.1 Combinatorics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1.1 Moment graphs . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1.2 Sheaves on moment graphs . . . . . . . . . . . . . . . . . . . . 25
4.1.3 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.1.4 Sheaves on ordered moment graphs . . . . . . . . . . . . . . . 28
4.1.5 Braden-MacPherson sheaves . . . . . . . . . . . . . . . . . . . 30
4.2 The link to representation theory . . . . . . . . . . . . . . . . . . . . 31
4.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3.1 The affine case . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3.2 The finite case . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

iv
5 Critical representations of affine
Kac-Moody algebras 43
5.1 Affine Kac-Moody algebras . . . . . . . . . . . . . . . . . . . . . . . . 43
5.1.1 Affine roots, Weyl groups and bilinear forms . . . . . . . . . . 44
5.2 Restricted representations . . . . . . . . . . . . . . . . . . . . . . . . 45
5.2.1 Integral affine roots . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2.2 Restricted category O . . . . . . . . . . . . . . . . . . . . . . 46
5.2.3 Restricted projective objects . . . . . . . . . . . . . . . . . . . 48
5.2.4 The restricted block decomposition . . . . . . . . . . . . . . . 49
5.2.5 The generic and subgeneric cases . . . . . . . . . . . . . . . . 50
5.3 The center of OS,Λ
e for Λ critical . . . . . . . . . . . . . . . . . . . . . 51
5.3.1 The subgeneric case . . . . . . . . . . . . . . . . . . . . . . . . 53
5.3.2 The general case . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.4 Restricted Jantzen Sum Formula . . . . . . . . . . . . . . . . . . . . 59
5.4.1 The Shapovalov determinant . . . . . . . . . . . . . . . . . . . 60
5.4.2 Restricted Jantzen filtration . . . . . . . . . . . . . . . . . . . 61

v
Introduction

## Let g ⊃ b ⊃ h be a semisimple Lie algebra with a Borel and a Cartan subalgebra.

For λ ∈ h∗ denote by ∆(λ) the Verma module with highest weight λ. Let Cλ ⊂ h∗ be
a curve through λ such that the Verma module ∆(µ) is simple for each µ ∈ Cλ \{λ}
in a small neighborhood of λ. A non-zero contravariant form on a Verma module is
a certain symmetric bilinear form with its radical being the maximal submodule of
the Verma module. The Jantzen filtration on ∆(λ) is a decreasing filtration which
‘measures’ the degeneracy of a contravariant form on ∆(µ) as µ tends to λ in Cλ .
It has proven to be a powerful tool to study the structure of submodules in ∆(λ).
For example, the subquotients of a Jordan-Hölder series of ∆(λ) over a semisimple
Lie algebra of rank 2 is completely determined by the Jantzen filtration and its
properties (cf. [Jan79]). Amongst other questions regarding this filtration, Jantzen
conjectured the semisimplicity of the subquotients of the Jantzen filtration. Further
work in this direction showed that the affirmation of these conjectures implies that
the multiplicities of simple modules in the subquotients of Jantzen’s filtration are
given by coefficients of certain Kazhdan-Lusztig polynomials (cf. [GJ81]).

## The Beilinson-Bernstein localization functor transforms Verma modules to certain

D-modules on the flag variety corresponding to g ⊃ b. In [BB93] it is shown that the
localization functor and Riemann-Hilbert correspondence map the Jantzen filtration
to a weight filtration on the associated perverse sheaves. Since this filtration is easier
to manage by a sophisticated geometric machinery, Beilinson and Bernstein were able
to prove Jantzen’s conjectures and confirmed the Kazhdan-Lusztig algorithm for the
computation of multiplicities in consecutive subquotients. A different but more direct
approach, which translates highest weight modules over a semisimple Lie algebra to
geometric objects associated to a flag variety, is given in [Soe90]. As a mediator
between projective objects in the highest weight category O over g and perverse
sheaves on the Langlands dual flag variety of g ⊃ b, the so-called category of Soergel
bimodules provides a bridge between the representation theoretic and the geometric
picture. Although the definition of the category of Soergel bimodules is very basic
in the sense that it is just a special subcategory of bimodules over a commutative
ring, its structure, which is more of combinatorial nature, is highly complicated at
the same time. Soergel’s approach is used in [Soe08] to relate the dimensions of the
subquotients of the Andersen filtration to coefficients of certain Kazhdan-Lusztig
polynomials. The Andersen filtration is defined in a very similar way to Jantzen’s
filtration and was introduced in [And97] for modules over an algebraic group. Let K

vi
be a tilting module of category O and denote by ∇(λ) the dual of ∆(λ) according
to the usual duality functor on O. The composition pairing defines a bilinear map
Homg (∆(λ), K) × Homg (K, ∇(λ)) → Homg (∆(λ), ∇(λ))
which essentially has values in C after identifying Homg (∆(λ), ∇(λ)) ∼ = C. Let Cλ

again be a curve in h such that the pairing above is non-degenerate if we replace λ
by any µ ∈ Cλ \{λ} of a small neighborhood of λ. Then the Andersen filtration is a
decreasing filtration on the vector space Homg (∆(λ), K) that ‘counts’ the degree of
degeneracy of these forms as µ tends to λ in Cλ .
The relation between these two filtrations is given by the tilting functor which is a
self-equivalence of a certain subcategory of O. It is contravariant and sends Verma
modules to Verma modules and projective modules to tilting modules. Now, the
Jantzen filtration on a Verma module induces a filtration on the homomorphism
space between a projective object and the Verma module. Thus, the tilting functor
as a contravariant equivalence, induces an isomorphism between this Hom-space and
the space of homomorphisms between a Verma and a tilting module. In section 2.4
of [Soe08], Soergel raises the question whether it is possible to prove without using
Jantzen’s conjecture, that the isomorphism induced by the tilting functor identifies
the induced Jantzen filtration on the first Hom-space with the Andersen filtration on
the second. This was done in the author’s diploma thesis [Küb10] (cf. the published
version [Küb12a]) and with the main result of [Soe08] provides a Koszul dual proof
of the semisimplicity of subquotients in Jantzen’s filtration.

This thesis consists of three parts. The first part provides the preliminaries and
recalls the definition of a symmetrizable Kac-Moody algebra g in chapter 1 and
introduces the deformed category O over g with Borel subalgebra b and Cartan
subalgebra h in chapter 2. For this, let S be the symmetric algebra over the Cartan
subalgebra and denote by Se its localization at the maximal ideal generated by h.
The deformed category OSe is the full subcategory consisting of those g-S-bimodules
e
which decompose into deformed weight spaces and are locally finite over the Lie
algebra b ⊗C S.
e In chapter 2 we also repeat well-known properties about projective
objects in OSe, a deformed BGG-reciprocity, the center of OSe in the subgeneric case
and a deformed version of the aforementioned tilting functor (cf. [Fie03], [Fie06] and
[Soe98]). Furthermore, we adjust Soergel’s construction of tilting modules over the
complex numbers from [Soe98] to the deformed theory over S. e

The second part of this thesis should be seen as an extension of the author’s diploma
thesis and generalizes the situation discussed above to the case of a Kac-Moody al-
gebra with certain restrictions on the representations. It consists of chapters 3 and
4 and deals with the representation theory of a symmetrizable Kac-Moody algebra
outside the critical hyperplanes. The third chapter contains a straight forward gener-
alization of the relation between the Jantzen and Andersen filtration for Kac-Moody
algebras via the tilting functor, which we explained above. It contains more or less
the same ideas as the author’s diploma thesis (cf. [Küb12a]) where the semisimple
case was treated. Chapter 4 collects two applications of chapter 3 and constitutes

vii
the main results of the second part of this thesis. First we deal with the case when
g is of affine type. Let λ ∈ h∗ be a regular weight of positive level, i.e. ∆(λ) is con-
tained in a block of category O with a dominant weight, and let P (µ) be a projective
cover of the simple highest weight module L(µ). We assume that ∆(λ) and P (µ) lie
in the same block. Let ρ ∈ h∗ be an element that takes the value 1 on any coroot
associated to a simple root. Then the tilting functor maps ∆(λ) to ∆(−2ρ − λ) and
P (µ) to T (−2ρ − µ) where T (−2ρ − µ) denotes the indecomposable tilting module
with maximal weight −2ρ − µ. ∆(−2ρ − λ) and T (−2ρ − µ) lie in a block of negative
level and we consider the Andersen filtration on Homg (∆(−2ρ − λ), T (−2ρ − µ)).
Note that the choice of the blocks guarantees the existence of the projective and
the tilting module. Theorem 4.7 of [Fie12b] generalizes the geometric part of [Soe08]
to the affine case and implies that the dimensions of subquotients of the Andersen
filtration are given by coefficients of certain affine Kazhdan-Lusztig polynomials. We
can now use the main result of chapter 3 to express the multiplicities in consecutive
subquotients of the Jantzen filtration on ∆(λ) in terms of the Kazhdan-Lusztig al-
gorithm. In [Sha12], Shan uses a generalization of Beilinson-Bernstein localization
to prove an analogous result about the Jantzen filtration on a (parabolic) Verma
module over an affine Kac-Moody algebra in negative level. In this sense, the first
part of chapter 4 complements Shan’s result. However, our approach has the advan-
tage that it might give a chance to prove the mentioned semisimplicity with purely
algebraic methods using recent work of Elias and Williamson (cf. [EW12]). But it
also has the disadvantage that we do not know how to treat the parabolic case in
this way. The second part of chapter 4 deals with the case when g is semisimple. We
repeat the main result of chapter 3 but this time in a more combinatorial manner.
We use Fiebig’s description of a block of the deformed category O as combinatorial
sheaves on a moment graph associated to this block. The main objects in this com-
binatorial category are the so-called Braden-MacPherson sheaves which correspond
to deformed projective objects in OSe and are constructed using the induced Bruhat
order on the moment graph. We prove, without using the tilting functor, that tilting
modules in OSe correspond to Braden-MacPherson sheaves constructed along the
reversed Bruhat order on the graph and give a combinatorial replacement of the
tilting functor which provides an analogous connection between the Andersen and
Jantzen filtration. However, we do not yet know how to apply this to the case of an
infinite-dimensional Lie algebra without using the tilting functor.

While the second part of this thesis is about non-critical representations, the third
part focuses on representations over an untwisted affine Kac-Moody algebra at the
critical level. Instead of working with the category O, the suitable framework for
chapter 5 is the subcategory of modules on which homogeneous elements of the
Feigin-Frenkel center of non-zero degree act trivially. For these elements, this ac-
tion is only non-trivial at the critical level. We repeat the construction and some
properties of the so-called restricted category O from [AF12b] in chapter 5. Recall
that an affine Kac-Moody algebra b g can be constructed as a central extension of the
loop algebra of a simple complex Lie algebra g. We view g as a subalgebra of b g and
also get an embedding h ⊂ h of the corresponding Cartan subalgebras. Denote by
b

viii
Sb (resp. S) the symmetric algebra of b h (resp. h) and let Se be the localization of S
at the maximal ideal generated by h. In contrast to the first part of this thesis, S
denotes the symmetric algebra over the finite Cartan h and not over b h. The deformed
category OSe is then analogously defined. We risk a confusion with the notation of
the first part of this thesis for the purpose of better accordance with the notation
in [AF12b]. The action of the Feigin-Frenkel center on the category O is compatible
with deformations and we denote by OSe the deformed restricted category O. It de-
composes into blocks of the form OS,Λ e where Λ parameterizes the highest weights
of simple objects in this block. Our first aim is to calculate the center of OS,Λ e . The
embedding of g in b g and of the corresponding Cartan subalgebras brings about an
embedding of two types of root systems R ⊂ R; b the finite and the affine root system.
Denote by R(Λ) the roots of R that are integral on elements of Λ. Recall that the
center Z of OS,Λ
e is the endomorphism ring of the identity functor. The first result
of the third part of this thesis is an isomorphism

Z∼
Y
= {(zµ ) ∈ Se | zµ ≡ zα↓µ mod α∨ , α ∈ R(Λ)}
µ∈Λ

## where α ↓ · : Λ → Λ is a certain bijection defined by the dot-action of the affine Weyl

group. We will give a more precise definition of this bijection in the main body of
this thesis. Fiebig already calculated the center of a non-restricted block outside the
critical hyperplanes in his PhD thesis (cf. [Fie03]). More generally, he considered an
arbitrary symmetrizable Kac-Moody algebra. As in Fiebig’s thesis the calculation of
the center reduces to the question how the center acts on indecomposable projective
objects in OS,Λ
e . The restricted projective objects in the deformed and non-deformed
case are well studied in [AF12b] and we make intense use of the main results therein.
While Fiebig needs his result for the connection to sheaves on a moment graph,
which we mentioned above, we do not see an application for the critical center
yet. The second result in the third part of this thesis is a Jantzen sum formula for
restricted Verma modules which are maximal restricted quotients of ordinary Verma
modules under the action of the Feigin-Frenkel center. They play the role of standard
modules in the restricted category O. Let ∆(λ) be such a restricted Verma module
with highest weight λ. As for non-restricted Verma modules we define a Jantzen
filtration
∆(λ) = ∆(λ)0 ⊃ ∆(λ)1 ⊃ ...
and proof a Jantzen sum formula
!
X X X
ch∆(λ)i = ch∆(α ↓2i−1 λ) − ch∆(α ↓2i λ)

i>0 α∈R(λ)+ i>0

where α ↓ λ is the notation from above and where ch denotes the character of a
weight module. As an immediate corollary we get the linkage principle. It states that
the multiplicity [∆(λ) : L(µ)] of a simple module with highest weight µ in a Jordan-
Hölder series of ∆(λ) can only be unequal to zero, if µ lies in the orbit of λ under the
dot-action of the affine Weyl group associated to the affine roots that are integral on

ix
λ. The linkage principle was already proven in [AF12b] using the structure of OSe.
In particular, the authors first describe the indecomposable projective objects in
the restricted deformed category O and prove a restricted BGG-reciprocity to relate
[∆(λ) : L(µ)] to the multiplicity of ∆(λ) in a restricted Verma flag of a projective
cover of L(µ) in O. Then they reduce the problem to generic and subgeneric cases,
which are covered by [AF12a] and [Fre05]. Our approach still relies on the generic
and subgeneric cases but does not need the structure of OSe. In this sense our proof
of the linkage principle is independent of [AF12b] but still relies heavily on the main
results of [AF12a] and [Fre05]. Moreover, the above sum formula is analogous to
the one for baby Verma modules over the small quantum group associated to g (cf.
[AJS94], chapter 6). Thus it supports a conjecture of Lusztig that a certain category
of representations over the small quantum group resembles the restricted category
O over the affine Kac-Moody algebra at the critical level.

x
1 Kac-Moody algebras

## Based on [Kac90] and [Car05] we recall the structure of a symmetrizable Kac-Moody

algebra. Let A = (aij )i,j=1,...,n be a generalized Cartan matrix of rank l. Recall that A
is called symmetrizable if there is an invertible diagonal matrix D = diag(1 , ..., n )
and a symmetric matrix B = (bij ) such that A = DB. Let (h, Π, Π∨ ) be a minimal
realization of A. Thus, h is a C-vector space of dimension 2n − l, Π = {α1 , ..., αn } is
a linearly independent subset of h∗ and Π∨ = {α1∨ , ..., αn∨ } is a linearly independent
subset of h such that αj (αi∨ ) = aij for all i, j.
Let g be the Kac-Moody algebra associated to A, i.e, g is the quotient of the Lie
algebra we get from the Serre relations of a set of Chevalley generators corresponding
to A, and its unique maximal ideal that intersects h trivially. We call the set Π the
simple roots and Π∨ the simple coroots. The vector space h inherits the structure of
a Cartan subalgebra of g and we get a root space decomposition
M
g= gα
α∈h∗

## where gα = {x ∈ g | [x, h] = α(h)x ∀ h ∈ h}. We denote the corresponding root

system by R = {α ∈ h∗ | gα 6= 0, α 6= 0} and the simple roots determine the set of
positive roots of R which we denote by R+ . This implies a partial order ≤ on h∗ given
by λ ≤ µ for λ, µ ∈L h∗ if µ − λ can be L
written as a sum of positive roots. Thereby,
we can define n := α<0 gα and n+ := α>0 gα and get a triangular decomposition

g = n ⊕ h ⊕ n+ .

## We define b := h ⊕ n+ to be the Borel subalgebra of g associated to the set of simple

roots Π. Let {e1 , ..., en }∪{f1 , ...., fn } be a set of Chevalley generators of g. We denote
by σ : g → g the Chevalley involution given by

## For any root α ∈ R we define mult α := dimC gα to be the multiplicity

P of α. Denote
by ht α the height ofPan element α in the Z-span of Π, i.e., if j kj αj = α for
kj ∈ Z, then ht α = j kj ∈ Z. This definition allows us to introduce a Z-grading
on g, namely the principal grading g = i∈Z gi where g0 = h and for i 6= 0
L

M
gi = gα .
α∈R
ht α=i

1
Let h0 = i Cαi∨ be the subspace of h generated by the simple coroots and denote
P
by h00 a complementary subspace of h0 in h. If g is symmetrizable, we can define a
C-valued form (·|·) on h by setting

## (h0 |h00 ) = 0 for h0 , h00 ∈ h00 .

As usual, h·, ·i : h∗ ×h → C denotes the natural pairing of a vector space with its dual.
By induction on the degree of the principal grading of g, this form can be extended
to a form on g with the properties that (·|·) is a symmetric, non-degenerate bilinear
form which is also non-degenerate on h and invariant, i.e., ([x, y]|z) = (x|[y, z]) for

all x, y, z ∈ g. The non-degeneracy of (·|·) on h defines an isomorphism h → h∗ and
allows us to define a non-degenerate symmetric bilinear form on h∗ which we denote
by (·|·) again. Note that in case of a semisimple Lie algebra, (·|·) is the normalized
Killing form.

## Denote by si := sαi : h∗ → h∗ the fundamental reflection associated to αi ∈ Π, given

by
si (λ) = λ − hλ, αi∨ iαi for λ ∈ h∗ .
The Weyl group W ⊂ Gl(h∗ ) is the subgroup generated by the fundamental re-
flections si , (i = 1, ..., n). We denote the real roots by Rre := W({α1 , ..., αn }), the
orbit of the set of simple roots under the standard action of the Weyl group, and
the imaginary roots by Rim := R\Rre . For the positive real (resp. imaginary) roots
we write Rre,+ := Rre ∩ R+ (resp. Rim,+ := Rim ∩ R+ ). Note that (α|α) > 0 for
any real root α ∈ Rre but it can happen that this value is negative for an imag-
inary root. Moreover, a root α is imaginary if and only if (α|α) ≤ 0. For a real
root α ∈ Rre the vector space [gα , g−α ] is one-dimensional and we denote by α∨ ∈ h
its coroot which is uniquely defined by the properties α∨ ∈ [gα , g−α ] and hα, α∨ i = 2.

We choose an element ρ ∈ h∗ such that hρ, αi∨ i = 1 for any simple coroot αi∨ ∈ Π∨ .
Note that ρ is only uniquely defined in the finite-dimensional case. The action of W
on h∗ preserves the form (·|·) and allows us to view the form on h as a W-invariant
form with the induced action. However, for the representation theoretic point of
view it is more convenient to work with an action of W that is shifted by −ρ. For
w ∈ W, λ ∈ h∗ we define
w · λ := w(λ + ρ) − ρ
and call this the dot-action of W. We say g is indecomposable if it is associated to
an indecomposable, symmetrizable, generalized Cartan matrix A. In this case we
have

## Theorem 1.0.1 ([Car05], Proposition 15.14). If g is an indecomposable, symmetriz-

able Kac-Moody algebra, g is either of finite, affine or indefinite type, which are
defined by the following:

## 1. g is of finite type if A is positive definite,

2
2. g is of affine type if A is positive semidefinite of corank 1,

## 3. g is of indefinite type if A is neither positive definite nor positive semidefinite of

corank 1.

3
2 Deformed Category O

2.1 Definition
After giving an overview of the type of Lie algebra we are working with, we intro-
duce the representation theory of these objects. The main sources for this chapter
are [Fie03], [Soe08] and [Fie12a]. Let g be a symmetrizable Kac-Moody algebra.
We fix a Borel subalgebra b and a Cartan subalgebra h. We denote by S = S(h)
the symmetric algebra of the C-vector space h and call a unital, associative, com-
mutative, finitely generated and noetherian S-algebra a deformation algebra. Every
deformation algebra A has a structure map τ : S → A given by x 7→ x · 1A .
For an arbitrary Lie algebra a and A a deformation algebra we denote by U (a) the
universal enveloping algebra and define aA := a⊗C A, which has an induced structure
of a Lie algebra. We can identify the A-dual (hA )∗ of the extended Cartan subalgebra
with h∗ ⊗C A, which is the extension by scalars of the C-dual of the Cartan. Using
this identification, we view every element λ ∈ h∗ as an element λ ⊗ 1 ∈ h∗A .
In the following we discuss subcategories of g-mod-A, which is the category of g-
A-bimodules. By the commutativity of A we can equivalently view this category
as modules over the Lie algebra gA . We make use of both viewpoints. We call a
gA -module M locally bA -finite if for any m ∈ M the bA -submodule generated by m
is finitely generated over A. For λ ∈ h∗ we denote by

## the deformed λ-weight space of M , where we consider λ + τ as an element of h∗A . We

call M a weight module if M ∼
L
= λ∈h∗ Mλ .
Definition 2.1.1. The deformed category OA is the full subcategory of g-mod-A of
modules M , such that

## Remark 2.1.2. OA is an abelian category. Note that if A = k is a field, the structure

map τ : S → k is unequal to zero and the Lie algebra is split. Note that the definition
of the deformed category O involves only weights which are a sum of the canonical
weight τ and elements h∗ . This implies that the category Ok can be viewed as the
direct summand of the usual BGG category O over the split Lie algebra gk which

4
consists of modules with weights lying in the affine complex subspace τ + h∗ of the
C-vector space h∗k . In particular, OC is the usual BGG category O over g if the
S-structure on C is given by S  S/Sh.

2.1.1 Duality on OA
Recall the Chevalley automorphism σ : g → g from chapter 1. If M ∈ OA , we
define d(M ) := λ∈h∗ Mλ∗ where ∗ denotes the dual of an A-module. For f ∈ d(M ),
L
x ∈ g and m ∈ M we give d(M ) the structure of a g-module by setting (xf )(m) =
f (σ(x)m). One can check that this induces a contravariant functor d : OA → OA . For
M ∈ OA there is a natural homomorphism M → d(d(M )) which is an isomorphism
if M is free over the deformation algebra A with weight spaces of finite rank.

## 2.1.2 Deformed standard and costandard modules

Let λ ∈ h∗ and denote by τ |h the restriction of the structure map of a deformation
algebra A to the subspace h ⊂ S. We define Aλ to be the b-A-bimodule which is free
of rank one as a right A-module and on which an element b + h ∈ b according to the
decomposition b = [b, b]⊕h acts from the left by multiplication with (λ+τ |h )(h) ∈ A.
With this, the deformed Verma module with highest weight λ ∈ h∗ is given by

## ∆A (λ) = ∆A (λ + τ ) := U (g) ⊗U (b) Aλ .

Both, the left action of U (g) and the right action of A are given by multiplication.
Although the highest weight of ∆A (λ) is actually λ + τ , it is more practical in
the first and second part of this thesis to forget about the canonical weight τ in
the notation of the Verma module once we fixed the structure of a deformation
algebra. Nevertheless, in chapter 5 it will be important to distinguish between two
different S-structures on the same ring A. In this case we use the notation ∆A (λ+τ )
which emphasizes the dependence on the structure map τ . We denote by ∇A (λ) =
∇A (λ + τ ) := d(∆A (λ)) the deformed dual Verma module associated to λ ∈ h∗ which
we also call the deformed Nabla module.
Remark 2.1.3. ∆A (λ) and ∇A (λ) have a weight space decomposition, such that
∆A (λ)µ 6= 0 (resp. ∇A (λ)µ 6= 0) implies µ ≤ λ. Furthermore, since ∆A (λ) and
∇A (λ) are free over the deformation algebra A, we have d(∇A (λ)) ∼
= ∆A (λ).
Lemma 2.1.4. Let A be a deformation algebra and λ, µ ∈ h∗ .
(i) The restriction to the highest weight space induces an isomorphism
(
A if µ = λ,
HomOA (∆A (λ), ∇A (µ)) ∼=
0 otherwise.

## (ii) Ext1OA (∆A (λ), ∇A (µ)) = 0.

Proof. The proof of [Soe08], Proposition 2.12, for g a semisimple Lie algebra gener-
alizes to our case.

5
Definition 2.1.5. Let M ∈ OA . We say M has a deformed Verma (resp. Nabla)
flag if M has a finite filtration
M = Mn ⊃ Mn−1 ⊃ ... ⊃ M0 = 0
such that for every i ∈ {1, ..., n} the subquotient Mi /Mi−1 is isomorphic to a de-
formed Verma (resp. Nabla) module.
Remark 2.1.6. Note that any direct summand of a module with a deformed Verma
(resp. Nabla) flag has again a deformed Verma (resp. Nabla) flag. Further, any
module with a deformed Verma or Nabla flag is free over the deformation algebra
A.
Remark 2.1.7. If A = Q is a field, then ∆Q (λ) has a simple head LQ (λ) which is
the socle of ∇Q (λ) for any λ ∈ h∗ . If the structure map τ : S → Q maps any coroot
α∨ ∈ h to an element which does not lie in Z ⊂ Q, the category OQ is semisimple
and the map onto the simple head induces an isomorphism ∆Q (λ) ∼ = ∇Q (λ) (cf.
[Soe08], section 2.14).

2.1.3 Characters
Let Z[h∗ ] be the group ring of the abelian group h∗ . To define characters for modules
in OA we need a certain completion of this [ ∗ ] ⊂ Z[[h∗ ]] be the
P ring. For
λ
this, let Z[h

subring consisting of those formal sums λ∈h∗ aλ e ∈ Z[[h ]] with the property that
there exists a finite subset {λ1 , ...λn } ⊂ h∗ such that aλ 6= 0 implies λ ≤ λi for at
least one i ∈ {1, ..., n}.
Definition 2.1.8. If M ∈ OA has weights which are bounded from above and weight
spaces which are finitely generated and free over A, we define the character chM ∈
[
Z[h∗ ] of M by
X
chM = chA M = rkA Mλ · eλ
λ∈h∗

## where rkA denotes the rank of a free A-module.

For M ∈ OA a module with a deformed Verma (resp. Nabla) flag we denote by
(M : ∆A (λ)) ∈ Z (resp. (M : ∇A (λ))) the multiplicity of ∆A (λ) (resp. ∇A (λ)) in a
deformed Verma (resp. Nabla) flag of M .
Remark 2.1.9. The character is additive in the sense that the characters of the
external modules of a short exact sequence of free A-modules in OA sum up to
the character of the module in the middle. Hence, the numbers (M : ∆A (λ)) and
(M : ∇A (λ)) are independent of the choice of a Verma (resp. Nabla) flag.

## 2.1.4 Simple objects

For A → A0 a homomorphism of deformation algebras, we get a functor · ⊗A A0 :
OA → OA0 that obviously maps ∆A (λ) to ∆A0 (λ) and ∇A (λ) to ∇A0 (λ) for any
λ ∈ h∗ . Let A be a local deformation algebra with maximal ideal m ⊂ A and residue
field K = A/m.

6
Lemma 2.1.10 ([Fie03], Proposition 2.1). The functor · ⊗A K : OA → OK induces
a bijection between isomorphism classes of simple objects in OA and those in OK .
We denote by LA (λ) the simple module in OA with highest weight λ ∈ h∗ . As in
the case of a deformed Verma module the actual highest weight of LA (λ) is λ + τ
for τ the structure map of the deformation algebra. However, we will only use the
notation LA (λ + τ ) if we have to distinguish between two different S-structures.
Remark 2.1.11. Let Se := S(0) be the localization of S at the maximal ideal generated
by h ⊂ S. If Q := Quot(S) is the quotient field of S, then Q fulfills the assumption
of Remark 2.1.7 and we have ∆Q (λ) ∼ = LQ (λ) ∼= ∇Q (λ) for any weight λ ∈ h∗ .
Thus, for any localization Sp of Se at a prime ideal p ⊂ Se and Kp the residue field
of Sp , the restriction of the above isomorphism to the Sp -lattice ∆Sp (λ) → ∇Sp (λ)
is injective and induces the composition ∆Kp (λ)  LKp (λ) ,→ ∇Kp (λ) after base
change · ⊗Sp Kp .
Notation 1. If the deformation algebra is S/Sh ∼ = C we omit the subscripts of all
objects in O = OC we defined so far. For example, we write ∆(λ), ∇(λ) and L(λ)
instead of ∆C (λ), ∇C (λ) and LC (λ).
Definition 2.1.12 ([Soe98], Definition 4.1). For any simple object L in an abelian
category A and any object M ∈ A we define [M : L] ∈ N ∪ {∞} to be the supremum
over all finite filtrations of M of the multiplicity of L in these filtrations.

## 2.2 Projective objects

We recall the existence of projective covers in certain truncated subcategories of OA
where A is again a local deformation algebra. The proofs can be found in [Fie12a]
and [Fie03]. Denote by K the residue field of A. Since the category O over an infinite-
dimensional Lie algebra does not contain enough projective objects in general, we
need to introduce truncated subcategories of OA .
Definition 2.2.1. We call a subset J ⊂ h∗ an open and bounded subset if µ ≤ λ
and λ ∈ J implies µ ∈ J and if for any λ ∈ J the set {µ ∈ J | µ ≥ λ} is finite.
J
We denote by OA ⊂ OA the category of modules M ∈ OA such that Mλ 6= 0 implies
λ ∈ J.
J
P
The assignment M 7→ M/ λ∈J / U (g)Mλ induces a functor OA → OA that is left
J J
adjoint to the inclusion functor OA ⊂ OA . In OA the existence of projective objects
is guaranteed by
Theorem 2.2.2 ([Fie12a], Theorem 4.2, and [Fie03], Proposition 2.6). Let J ⊂ h∗
be an open and bounded subset and λ ∈ J . Then LA (λ) has a projective cover
J
PAJ (λ) in OA . Furthermore, PKJ (λ) ∼
= PAJ (λ) ⊗A K and for J 0 ⊂ J another open
and bounded subset with λ ∈ J 0 we have (PAJ (λ))J ∼
0 0
= PAJ (λ).
This theorem implies a one-to-one correspondence between projective isomorphism
J
classes of OA and OKJ via the functor · ⊗A K. The following lemma is called base
change and is an essential tool for the content of this thesis.

7
J
Lemma 2.2.3 ([Fie03], Proposition 2.4). Let P, M ∈ OA , P be projective and
0
A → A a homomorphism of (not necessarily local) deformation algebras. If P is
finitely generated, then the natural map

## HomOA (P, M ) ⊗A A0 −→ HomOA0 (P ⊗A A0 , M ⊗A A0 )

is an isomorphism.
Another important base change isomorphism is given in
Lemma 2.2.4 ([Soe08], Corollary 2.13). Let M, N ∈ OA where M has a Verma
and N a Nabla flag and A is a (not necessarily local) deformation algebra. Let
A → A0 be a homomorphism of deformation algebras. Then the space HomOA (M, N )
is projective over A and the natural map

## HomOA (M, N ) ⊗A A0 −→ HomOA0 (M ⊗A A0 , N ⊗A A0 )

is an isomorphism.
Proof. This is a direct consequence of Lemma 2.1.4.
Furthermore, we have a deformed version of the BGG-reciprocity:
Proposition 2.2.5 ([Fie03], Theorem 2.7). Let A be a local deformation algebra
with residue field K, J an open and bounded subset and λ ∈ J . Then PAJ (λ) admits
a deformed Verma flag and we have
(
[∆K (µ) : LK (λ)] if µ ∈ J ,
(PAJ (λ) : ∆A (µ)) =
0 otherwise.

## 2.3 Block decomposition

Let A be a local deformation algebra with residue field K. We define ∼A =∼K as the
equivalence relation on h∗ that is generated by tuples (λ, µ) with the property that
[∆K (λ) : LK (µ)] 6= 0. We call an equivalence class Λ ⊂ h∗ of ∼A an A-block, K-block
or just block, if it is clear which deformation algebra we are working with. Recall the
bilinear form (·|·) : g × g → C introduced in chapter 1. We also denote by (·|·) the
induced form on h∗ . For A a deformation algebra with structure map τ : S → A we
denote by (·|·)A : h∗A × h∗A → A the induced A-bilinear form. Kac and Kazhdan gave
a description of the equivalence classes under ∼A =∼K .
Theorem 2.3.1 ([KK79], Theorem 2). ∼A is generated by the tuples (λ, µ) ∈ h∗ ×h∗
that satisfy 2(λ + ρ + τ |α)K = n(α|α)K for n ∈ N and α ∈ R, such that λ − µ = nα.
We define the integral root system corresponding to an element λ ∈ h∗ (and τ : S →
A) by

## RA (λ) := {α ∈ R | 2(λ + ρ + τ |α)K ∈ Z(α|α)K }

8
and the integral Weyl group by

## WA (λ) := hsα | α ∈ RA (λ)i ⊂ W.

Let Λ be an A-block with λ, µ ∈ Λ. One easily sees that RA (λ) = RA (µ) and
therefore it is well-defined to write RA (Λ) := RA (λ). If the deformation algebra is
C = S/Sh we omit the subscript and write R(λ) (resp. W(λ)) instead of RC (λ)
(resp. WC (λ)).
For an A-block Λ ∈ h∗ / ∼A we denote by OA,Λ ⊂ OA the full subcategory of
modules M such that for each open and bounded subset J ⊂ h∗ and each morphism
ϕ : PAJ (λ) → M we have that ϕ 6= 0 implies λ ∈ Λ. Denote by MΛ the submodule
of a module M ∈ OA which is generated by all images PAJ (λ) → M where λ ∈ Λ
and J is an open and bounded subset.
Theorem 2.3.2 ([Fie03], Proposition 2.8). The functor
Q
OA,Λ −→ OA
Λ∈h∗ /∼A
L
(MΛ ) 7−→ MΛ
is an equivalence of categories.

## 2.4 Subgeneric blocks

Recall the localization Se of S at the maximal ideal generated by h and let Sp be a
localization of Se at a prime ideal p ⊂ S.
e We assume that p is of height one. Then
the functor · ⊗Se Sp : OSe → OSp yields a decomposition of certain blocks of OSe into
generic (i.e., those containing one element) and subgeneric (i.e., those containing
two elements) Sp -blocks:
Lemma 2.4.1 ([Fie06], Lemma 3). Let Λ ∈ h∗ / ∼Se such that Λ has a maximal or
a minimal element and p ⊂ Se a prime ideal.
(i) If α∨ ∈/ p for all roots α ∈ RSe(Λ), then Λ splits under ∼Sp into generic
equivalence classes of ∼Sp .
e ∨ for a root α ∈ R e(Λ), then Λ splits under ∼Sp into generic and
(ii) If p = Sα S
subgeneric equivalence classes of ∼Sp of the form {λ, sα · λ}.
We want to recall the description of the endomorphism ring of the deformed an-
tidominant projective cover in the subgeneric case due to [Fie03]. Let Λ ∈ h∗ / ∼Sp
be subgeneric and p = Sαe ∨ for α ∈ R, i.e., Λ = {λ, µ} with µ < sα · µ = λ. We then
have a non-split short exact sequence

## 0 → ∆Sp (λ) → PSp (µ) → ∆Sp (µ) → 0

and every endomorphism f : PSp (µ) → PSp (µ) stabilizes the subspace ∆Sp (λ) since
λ > µ. Thus f induces a commutative diagram

9
0 −−−→ ∆Sp (λ) −−−→ PSp (µ) −−−→ ∆Sp (µ) −−−→ 0
    
 f
 f f 
y λy y yµ y
0 −−−→ ∆Sp (λ) −−−→ PSp (µ) −−−→ ∆Sp (µ) −−−→ 0
Identifying fλ and fµ with the corresponding elements of Sp under isomorphisms
EndOSp (∆Sp (µ)) ∼
= Sp ∼
= EndOSp (∆Sp (λ)), we get a map

## χ : EndOSp (PSp (µ)) −→ Sp ⊕ Sp

f 7−→ (fλ , fµ ).
e ∨.
We write Sα instead of Sp for p = Sα
Proposition 2.4.2 ([Fie03], Corollary 3.5). Let Λ ∈ h∗ / ∼Sα . If Λ = {λ, µ} and
λ = sα · µ > µ, the map χ from above induces an isomorphism
EndOSα (PSα (µ)) ∼
= {(tλ , tµ ) ∈ Sα ⊕ Sα |tλ ≡ tµ mod α∨ } .
Definition 2.4.3. Let Λ ∈ h∗ / ∼A be an A-block for A a deformation algebra. We
call an element λ ∈ Λ dominant if λ ≥ λ0 for all λ0 ∈ Λ. If there is an element µ ∈ Λ
for which µ ≤ λ0 for all λ0 ∈ Λ we call µ antidominant.

## 2.5 Tilting modules

This chapter explains the construction of modules which admit both a Verma and
a Nabla flag. They were introduced and studied in [CI89] in the non-deformed,
finite-dimensional case. For the rest of this chapter we assume that the deformation
algebra A is local with maximal ideal m ⊂ A and residue field K = A/m. In the
author’s diploma thesis [Küb10], deformed tilting modules were defined in the case
of a semisimple Lie algebra using translation functors. In [Soe98] and [Soe99] Soergel
gives a construction of tilting modules over Kac-Moody algebras which avoids trans-
lation functors. Although Soergel works over a field, we are able to adopt most of his
arguments to the deformed case. But similarly to Fiebig’s construction of deformed
projective modules in [Fie06] for symmetrizable Kac-Moody algebras, it should also
be possible to construct deformed tilting modules using Fiebig’s deformed transla-
tion functors developed in [Fie03] in this case.
Definition 2.5.1. An object T ∈ OA is called a deformed tilting module if it has a
deformed Verma and a deformed Nabla flag.
Remark 2.5.2. Note that by definition, Verma and Nabla flags are finite filtrations.
This has the consequence that tilting modules might not exist in every block of OA
and is in contrast to [Soe98] where Verma and Nabla flags are allowed to be infinite.
Remark 2.5.3. Since the category of modules with a Verma (resp. Nabla) flag is
stable under taking direct sums and summands, the same is true for the category of
tilting modules.

10
Lemma 2.5.4. Let Λ ∈ h∗ / ∼A be an equivalence class containing an antidominant
weight and let M ∈ OA,Λ have a Verma flag. Then the following are equivalent:
(i) Ext1OA (∆A (λ), M ) = 0 for all λ ∈ h∗ ;

## (ii) M has a deformed Nabla flag.

Remark 2.5.5. If A is a field, the statement of the lemma is true if we assume the
weaker property for M being finitely generated instead of having a Verma flag.
Proof. That (i) follows from (ii) is Lemma 2.1.4.
The other implication is a modification of the proof in the field case (cf. [AF12b],
Proposition 4.1). For this we prove the dual statement. By our assumption on the
block Λ, d(M ) is finitely generated and has a Nabla flag. Thus, there is a filtration
0 = M0 ⊂ M1 ⊂ ... ⊂ Mn = d(M ) such that each subquotient Mi /Mi−1 of this
filtration is a quotient of a certain deformed Verma module ∆A (µi ). We can choose
this filtration such that µi ≤ µj with i 6= j, implies j < i.
If n = 1, there is µ ∈ Λ, such that ∆A (µ)  d(M ). But since d(M ) has a deformed
Nabla flag, a comparison of the characters shows that d(M ) ∼ = ∇A (µ) and thus
d(M ) ∼ = ∆A (µ). This implies that µ is the antidominant element of Λ. (If A is
a field, the additional assumption for d(M ) having a deformed Nabla flag is not
necessary here.)
Now the rest of the proof follows by induction on the length of the highest weight
filtration with the same arguments as in the field case (cf. for example [AF12b],
Proposition 4.1). The authors of [AF12b] work with restricted Verma modules but
the methods of their proof can be applied to our situation as well.
Next we prove the existence of indecomposable deformed tilting modules in a block
OA,Λ with an antidominant weight.
Theorem 2.5.6. Let Λ ∈ h∗ / ∼A be a block with an antidominant weight. For any
µ ∈ Λ there exists an, up to isomorphism unique, indecomposable deformed tilting
module TA (µ) ∈ OA,Λ which has a deformed Verma flag starting with ∆A (µ) ,→ T .
Proof. We give a deformed version of Soergel’s proof of Proposition 5.6 in [Soe98].
Note first that the A-modules Ext1OA (∆A (λ), ∆A (µ)) and also HomOA (∆A (λ), ∆A (µ))
are finitely generated and Ext1OA (∆A (λ), ∆A (µ)) = 0 unless λ < µ. As modules over
a local ring they have a minimal set of generators and the number of generators is
unique in the sense that the number of elements in any other minimal set of gener-
ators is the same.
Since Λ has an antidominant weight λ, there are only finitely many elements be-
low any weight µ ∈ Λ. If µ = λ, we can take ∆A (µ) = TA (µ). If µ 6= λ, we
can assume by induction on the number of elements below µ, that we already
found an object T >λ (µ) which fulfills the second condition of the theorem and
Ext1OA (∆A (ν), T >λ (µ)) = 0 for any ν ∈ Λ with ν 6= λ.
If in addition Ext1OA (∆A (λ), T >λ (µ)) = 0, we can apply Lemma 2.5.4 to show that
T >λ (µ) is a deformed tilting module. Since direct summands of modules with a
deformed Verma flag have a deformed Verma flag again and since we assume that

11
Verma flags are finite, we can take an indecomposable direct summand of T >λ (µ)
with a Verma flag that starts with ∆A (µ). This yields the claimed object TA (µ).
Otherwise, we have a minimal set {e1 , ..., en } of generators of Ext1OA (∆A (λ), T >λ (µ)).
Let e1 be represented by a short exact sequence
T >λ (µ) ,→ T  ∆A (λ).
Since Ext1OA (∆A (λ), ∆A (λ)) = 0 we get an exact sequence
Hom1OA (∆A (λ), ∆A (λ)) → Ext1OA (∆A (λ), T >λ (µ)) → Ext1OA (∆A (λ), T ) → 0
where the identity on the left maps to e1 . Thus Ext1OA (∆A (λ), T ) is generated by
the images of the elements {e2 , ..., en }. Applying the functor Ext1OA (∆A (ν), ·) to the
above short exact sequence for other ν ∈ Λ with ν 6= λ implies Ext1OA (∆A (ν), T ) = 0,
since Ext1OA (∆A (ν), ∆A (λ)) = Ext1OA (∆A (ν), T >λ (µ)) = 0. Thus we can use Lemma
2.5.4 and proceed in finitely many steps to get a tilting object T 0 with a deformed
Verma flag that starts with ∆A (µ). Again, we can take an indecomposable direct
summand of T 0 which contains ∆A (µ) in a Verma flag as the looked for tilting
module.
We now prove the uniqueness by tracing it back to the case where A is a field which
is already known by [Soe98]. Let T be a second indecomposable tilting module with
the same properties as TA (µ). We want to prove that both modules are isomorphic.
For this, we look at the following diagram
∆A (µ) ,→ TA (µ)  coker
k
∆A (µ) ,→ T  coker0
Since coker has a deformed Verma flag, we get ExtOA (coker, T ) = 0. Hence, we can
lift the identity on ∆A (µ) to get a map α : TA (µ) → T . With the same argument we
get a map β : T → TA (µ). Since all modules from the above diagram are free over
the deformation algebra, tensoring with · ⊗A K yields two maps αK = α ⊗ idK and
βK = β ⊗ idK which have the property that αK ◦ βK and βK ◦ αK induce the identity
on the submodule ∆K (µ). Hence, they are not nilpotent and therefore isomorphisms
by the Fitting lemma. Thus, by base change, Lemma 2.2.4, the maps α and β we
started with are inverse isomorphisms between T and TA (µ).
Notation 2. As in the case of Verma and Nabla modules we omit the subscript
for deformed tilting modules if the deformation algebra is C = S/Sh and proceed
similarly with indecomposable projective objects. Hence, we write T (λ) instead of
TC (λ) and P (µ) instead of PC (µ) whenever these objects exist.
Remark 2.5.7. Now assume that Λ is a block that contains a dominant but no
antidominant weight. Note that in this case the proof of the above theorem implies
that for ν, µ ∈ Λ we can also construct a unique indecomposable element TA≥ν (µ) in
OA with the properties
(i) Ext1OA (∆A (λ), TA≥ν (µ)) = 0 for all λ  ν,

(ii) TA≥ν (µ) has a deformed Verma flag that starts with ∆A (µ) ⊂ TA≥ν (µ).

12
2.6 Tilting equivalence
The tilting equivalence was introduced by Arkhipov in [Ark97] as an algebraic ap-
proach to semi-infinite cohomology. In [Soe98] Soergel works on the tilting functor
for graded Lie algebras. In this context it is a contravariant self-equivalence of the
category of modules with a Verma flag which sends short exact sequences to short
exact sequences and projective modules to tilting modules. Soergel’s ideas were ex-
tended to Virasoro algebras in [IK06] and to Lie superalgebras in [Bru04]. A very
detailed overview of Soergel’s construction is given in [IK11], chapter 7.
For this chapter let K be a field of characteristic zero. We use the principal Z-grading
from chapter 1 to view the symmetrizable Kac-Moody algebra g over K as a graded
Lie algebra. With this grading we have h = g0 . For graded K-vector spaces V, W we
write M
grHomK (V, W ) = HomK (V, W )n
n∈Z
n
where HomK (V, W ) denotes the homogeneous homomorphisms between V and W
of degree n. For a, b two Z-graded Lie algebras and M, N two graded a-b-bimodules,
we have a left b-action defined by (bf )(m) = f (mb) and a right a-action defined
by (f a)(m) := f (am) for f ∈ grHomK (M, N ), m ∈ M , a ∈ a and b ∈ b. For
an a-subalgebra g, we denote by grHomg (M, N ) the subspace of grHomK (M, N ) of
homomorphisms which are compatible with the g-action.

## 2.6.1 The semi-regular bimodule

Before we introduce the semi-regular bimodule we first need to state the definition of
a semi-infinite character. In a more general setting, a semi-infinite character should
be a character of g0 whose differential coincides with the critical 2-cocycle of g. In
our setting, however, we can use a weaker definition due to [Soe98]:
Definition 2.6.1. We call an element γ ∈ HomK (g0 , K) a semi-infinite character
of g, if
(i) g is generated as a Lie algebra by the homogeneous pieces g1 , g0 and g−1 ,
(ii) γ([x, y]) = tr(ad(x) ◦ ad(y) : g0 → g0 ) for all x ∈ g1 and y ∈ g−1 .
Proposition 2.6.2 ([Soe98], Lemma 7.1). −2ρ is a semi-infinite character for g.
Note that for i ∈ Z, dimK gi < ∞ implies that the semi-infinite character is unique.
In the following we will omit the subscript of the tensor product in case it is over the
field K, i.e., we write ⊗ instead of ⊗K . Denote by U (n)~ := grHomK (U (n), K) the
n-n-bimodule where K is the one dimensional K-vector space concentrated in degree
zero. The left and right action of n are given as before and U (n)~ is Z≥0 graded. We
define
S−2ρ := U (n)~ ⊗ U (b)
as vector spaces. We want to give S−2ρ a left and a right g-action. Denote by K−2ρ
the one dimensional graded b-module on which an element b + h ∈ b, decomposed

13
with respect to b = [b, b] ⊕ h, acts by multiplication with −2ρ(h). The isomorphism

K−2ρ ⊗ U (b) → U (b), 1 ⊗ b 7→ b induces an isomorphism

grHomb (U (g), K−2ρ ⊗ U (b)) → grHomK (U (n), U (b))

of left n-modules. We can compose this isomorphism with the natural isomorphism

S−2ρ → grHomK (U (n), U (b))

and carry over the left g-module structure on grHomK (U (g), K−2ρ ⊗ U (b)) to get the
structure of a graded left g-module on S−2ρ .

## We also have an isomorphism of graded b-modules

U (n)~ ⊗ U (b) ← U (n)~ ⊗U (n) U (g)

and we use the right g-module structure on U (n)~ ⊗U (n) U (g) to give S−2ρ a right g-
module structure. Note that the left g-structure on S−2ρ depends on the semi-infinite
character, but the right g-structure does not. Now a long calculation, which can be
found in the proof of Theorem 1.3 in [Soe98], shows that the two g-structures are
compatible, such that S−2ρ is in fact a g-g-bimodule. Furthermore, the same theorem
implies that the inclusion of n-n-bimodules U (n)~ ,→ S−2ρ , f 7→ f ⊗ 1 induces an
isomorphism

U (g) ⊗U (n) U (n)~ → S−2ρ
of g-n-bimodules and also an isomorphism of n-g-bimodules

U (n)~ ⊗U (n) U (g) → S−2ρ . (2.1)
If M is a g-module with a Verma flag, then M is a finitely generated and free module
over U (n). Thus, we find a finite dimensional K-vector space E such that

M∼
= U (n) ⊗ E

## as n-modules. This implies that as n-modules we have

S−2ρ ⊗U (g) M ∼
= U (n)~ ⊗U (n) M ∼
= U (n)~ ⊗ E.

Now let
0→M →N →K→0
be a short exact sequence of modules with a Verma flag. Considered as n-modules,
the short exact sequence splits since all modules are free over U (n). Using the iso-
morphism (2.1), we get an isomorphism

S−2ρ ⊗U (g) X ∼
= U (n)~ ⊗U (n) X

## of n-g-bimodules where X = M, N, K. We conclude that the sequence

14
0 → U (n)~ ⊗U (n) M → U (n)~ ⊗U (n) N → U (n)~ ⊗U (n) K → 0
considered as a sequence of n-modules, splits as well. But this implies that the
sequence

## 0 → S−2ρ ⊗U (g) M → S−2ρ ⊗U (g) N → S−2ρ ⊗U (g) K → 0

of g-modules is exact.

Denote by M the category of modules with a Verma flag and by K the category
of modules that are finitely generated and free over U (n)~ . Then Soergel proves in
[Soe98], Theorem 2.1, that

S−2ρ ⊗U (g) · : M → K

induces an equivalence of categories which sends short exact sequences to short exact
sequences. For M a graded g-module, we denote by M ~ its graded dual with the
g-action givenLby (xf )(m) = −f (xm) for f ∈ M ~ , x ∈ g and mL ∈ M . In our case,
we have M = λ∈h∗ Mλ and the graded dual is given by M = λ∈h∗ Mλ∗ with the
~

## t := (S−2ρ ⊗U (g) ·)~ : M → Mopp

which sends short exact sequences to short exact sequences. Furthermore, for λ ∈ h∗
we have t(∆K (λ)) ∼= ∆K (−2ρ−λ). This and the Ext-vanishing property of projective
and tilting modules imply that projective modules are mapped to tilting modules
under t. In particular, t(PK (λ)) ∼
= TK (−2ρ − λ). We summarize the properties of the
tilting functor in
Theorem 2.6.3 ([Soe98], Corollary 2.3). The functor

## (S−2ρ ⊗U (g) ·)~ : M → Mopp

is an equivalence of categories that maps short exact sequences to short exact se-
quences, projective objects to tilting objects and ∆K (λ) to ∆K (−2ρ − λ) for any
λ ∈ h∗ .
As in [Fie06], section 2.6, one can extend the definition of the tilting functor to a de-
formed version. Let A be a deformation algebra and denote by MA the subcategory
of OA that consists of modules with a deformed Verma flag. If A is a localization of
S at a prime ideal that is stable under the map that sends h ∈ h to −h ∈ h or a
quotient of S that corresponds to a subvariety of h∗ which is stable under the dual
of this map, the tilting functor extends to an A-linear equivalence

tA = (S−2ρ ⊗U (g) ·)~ : MA → Mopp
A

where ~ is the similarly defined A-dual in the deformed setting. It has analogous
properties as the non-deformed tilting functor, i.e., it maps ∆A (λ) to ∆A (−2ρ − λ)

15
and PA (λ) to TA (−2ρ − λ).
Note that for A → A0 a homomorphism of deformation algebras that fulfill the sta-
bility condition under the map h 7→ −h from above, we have a natural isomorphism
of functors
tA0 ◦ (· ⊗A A0 ) ⇒ (· ⊗A A0 ) ◦ tA .
Thus the deformed version of the tilting module commutes with base change.
Remark 2.6.4. Recall that for A a local deformation algebra with residue field K, the
functor · ⊗A K : OA → OK induces a bijection between the isomorphism classes of
projective objects. Theorem 2.5.6 and the fact that the tilting functor interchanges
projective with tilting modules and commutes with base change, imply that · ⊗A K :
OA → OK induces an isomorphism between isomorphism classes of tilting modules
as well.

## 2.7 Description of blocks

We give a description of blocks outside the critical hyperplanes following [KK79]
and [KT00]. For this we work over the complex numbers C. Let Rim be the set of
imaginary and Rre the set of real roots of the symmetrizable Kac-Moody algebra g.
Definition 2.7.1. Let Λ ∈ h∗ / ∼C be a block. We say that Λ lies outside the critical
hyperplanes if and only if R(Λ) ⊂ Rre .
Remark 2.7.2. Since any homomorphism Se → A of local deformation algebras leads
to a refinement of the block decomposition under the functor OSe → OA , we get
S
that the S-block
e Λ decomposes into i Λi where Λi are A-blocks. Thus we have
RA (Λi ) ⊂ R(Λ) (cf. [Fie03], Corollary 2.10). Note that by definition, Λ is outside
/ Z(β|β) for λ ∈ Λ and β ∈ Rim .
the critical hyperplanes if and only if 2(λ + ρ|β) ∈
If Λ is a block outside the critical hyperplanes, the Kac-Kazhdan Theorem 2.3.1
implies that Λ = W(Λ) · λ for any λ ∈ Λ.
Following [Fie06], we say a block Λ outside the critical hyperplanes is of positive
level if it contains a dominant weight. We say Λ is of negative level if Λ contains an
antidominant weight. Note that Λ is of positive level if and only if Λ0 := −2ρ − Λ is
of negative level. Thus the tilting equivalence
t : M → Mopp
induces equivalences between blocks of negative and those of positive level. More-
over, a block Λ can be both, of positive and of negative level.

We want to give a more detailed description of the different types of blocks over an
affine Kac-Moody algebra. Recall that in this case the imaginary roots are given by
Rim = {nδ | n ∈ Z\{0}} where δ is the smallest positive, imaginary root. Let g be an
affine Kac-Moody algebra and Λ a block. Then Λ is outside the critical hyperplanes
if and only if δ ∈
/ R(Λ). Set
C := {λ ∈ h∗ | (δ|λ + ρ) 6= 0},

16
C + := {λ ∈ C | (α∨ |λ + ρ) ≥ 0 for any α ∈ R+ (Λ)},
C − := {λ ∈ C | (α∨ |λ + ρ) ≤ 0 for any α ∈ R+ (Λ)}.
The following lemma tells us which cases can appear outside the critical hyperplanes
in the affine case.

## Lemma 2.7.3 ([KT00], Lemma 2.10). Let Λ ⊂ C be a C-block and λ ∈ Λ. Assume

that Λ satisfies R(Λ) 6= 0.

## / Q, then |W(Λ) · λ ∩ C + | = |W(Λ) · λ ∩ C − | = 1, i.e., Λ = W(Λ) · λ

(i) If (δ|λ + ρ) ∈
has both, a dominant and an antidominant weight.

## (ii) If (δ|λ + ρ) ∈ Q>0 , then |W(Λ) · λ ∩ C + | = 1 and |W(Λ) · λ ∩ C − | = 0, i.e.,

Λ = W(Λ) · λ has a dominant but no antidominant weight.

## (iii) If (δ|λ + ρ) ∈ Q<0 , then |W(Λ) · λ ∩ C + | = 0 and |W(Λ) · λ ∩ C − | = 1, i.e.,

Λ = W(Λ) · λ has an antidominant but no dominant weight.

We will treat the case when Λ is of critical level, i.e., when (δ|λ + ρ) = 0 for any
λ ∈ Λ, separately in chapter 5. If g is of finite type or a semisimple Lie algebra, then
Rim = ∅ and all blocks have both, a dominant and an antidominant weight.

17
3 Andersen and Jantzen filtrations

In this chapter we generalize the main result of the author’s diploma thesis [Küb10]
(cf. the published version [Küb12a]) to the case of a symmetrizable Kac-Moody al-
gebra. Let g be a symmetrizable Kac-Moody algebra, Λ a block of positive level
and µ, λ ∈ Λ. Then the projective cover P (λ) of L(λ) exists and the Jantzen fil-
tration on the Verma module ∆(µ), which we define below, induces a filtration on
HomO (P (λ), ∆(µ)). Define Λ0 := −2ρ−Λ, λ0 := −2ρ−λ and µ0 := −2ρ−µ. Then Λ0
is of negative level and the indecomposable tilting object T (λ0 ) exists. There is a fil-
tration on the vector space HomO (∆(µ0 ), T (λ0 )) which is due to Andersen and whose
subquotients are related to Kazhdan-Lusztig polynomials (cf. [Soe08] and [Fie12b]).
The main result of this chapter is that the tilting equivalence t : MΛ → Mopp Λ0 ,
0 0
which maps ∆(µ) to ∆(µ ), P (λ) to T (λ ) and which induces an isomorphism

HomO (P (λ), ∆(µ)) → HomO (∆(µ0 ), T (λ0 )),

## 3.1 Jantzen filtration

The content of this section can be found in [Jan79] and [KK79]. Recall the anti-
involution σ : g → g which maps h ∈ h to −h and gα to g−α . According to [KK79],
the universal enveloping algebra U (g) has a decomposition U (g) = U (h) ⊕ (nU (g) +
U (g)n+ ) and we denote by β : U (g)  U (h) = S(h) the projection onto the first
summand. This induces an S(h)-bilinear form F : U (g) × U (g) → S(h) which is
defined by F (x, y) = β(σ(x)y). Viewed as a polynomial map on h∗ , F (x, y) can be
evaluated at any λ ∈ h∗ and since the Verma module ∆(λ) is free of rank one over
U (n), the map F descends to a bilinear form

## (·, ·)λ : ∆(λ) × ∆(λ) → C

(x, y) 7→ F (x, y)(λ)

which is contravariant with respect to σ, i.e., (x · v, w)λ = (v, σ(x) · w)λ for x ∈ U (g)
and v, w ∈ ∆(λ). Let Cλ be a curve in h∗ through λ such that ∆(µ) is simple for
µ ∈ Cλ \{λ} in a small neighborhood of λ. Intuitively, the Jantzen filtration

## ∆(λ) ⊃ ∆(λ)1 ⊃ ∆(λ)2 ⊃ ...

18
is counting the degree of degeneracy of the contravariant forms (·, ·)µ as µ tends
to λ along the curve Cλ . We want to formalize this viewpoint algebraically. Recall
the weight ρ ∈ h∗ with the property ρ(α∨ ) = 1 for all coroots associated to a
simple root α ∈ Π. Let λ ∈ h∗ and C[t] be the S-algebra with structure morphism
S := S(h) = C[h∗ ]  C[Cρ] ∼ = C[t] being the restriction of polynomials on h∗ to
the curve given by t 7→ tρ. The deformed Verma module ∆C[t] (λ) can be seen as a
family of Verma modules parameterized by the curve t 7→ λ + tρ. The choice of the
curve implies that for C(t) the field of fractions of C[t], the Verma module ∆C(t) (λ)
is irreducible as a g ⊗ C(t)-module. Now we can extend the contravariant form
(·, ·)λ : ∆(λ)×∆(λ) → C to a C(t)-bilinear form (·, ·)λC(t) : ∆C(t) (λ)×∆C(t) (λ) → C(t)
which has the property that it is non-degenerate, since ∆C(t) (λ) is simple, and that
(∆C[t] (λ), ∆C[t] (λ))λC(t) ⊂ C[t].
For n ∈ Z>0 we set
∆C[t] (λ)n := {m ∈ ∆C[t] (λ) | (m, ∆C[t] (λ))λC(t) ∈ tn C[t]}
which forms a descending filtration on ∆C[t] (λ) of g-C[t]-bimodules by the contravari-
ance of the form. Further, the specialization map C[t]  C[t]/tC[t] ∼ = C induces a
map ∆C[t] (λ)  ∆C (λ) = ∆(λ). We can now define the Jantzen filtration
∆(λ) = ∆(λ)0 ⊃ ∆(λ)1 ⊃ ∆(λ)2 ⊃ ...
by taking the images of ∆C[t] (λ)n in ∆(λ) under the specialization map. By defini-
tion, ∆(λ)1 coincides with the radical of (·, ·)λ on ∆(λ) and by the contravariance
of (·, ·)λ the radical is also the maximal submodule of ∆(λ). This is a first indica-
tion that the Jantzen filtration contains a lot of information about the submodule
structure of ∆(λ). For example, in [Jan79] it is shown that in case of a semi-simple
Lie algebra of rank two, the multiplicities [∆(λ) : L(µ)] are fully determined by this
filtration and the associated sum formula we will recall in chapter 5.

## Let P ∈ O be a projective object. Then the Jantzen filtration on ∆(λ) induces

a filtration of C-vector spaces on HomO (P, ∆(λ)). Since most of the constructions
of chapter 2 needed a local deformation algebra, we now work with the ring of
power series in one variable C[[t]] instead of the polynomial ring C[t]. Hence, the
structure map τ : S  C[t] ,→ C[[t]] factors through the localization S. e Using the
completion C[[t]] for the definition does not change the Jantzen filtration on ∆(λ).
Let PC[[t]] ∈ OC[[t]] be a projective C[[t]]-lift of P , which exists by Theorem 2.2.2. The
field of formal Laurent series C((t)) fulfills the properties of Remark 2.1.7. Hence,
we get an isomorphism ∆C((t)) (λ) ∼ = ∇C((t)) (λ) which restricts to an injective map
∆C[[t]] (λ) ,→ ∇C[[t]] (λ). Taking homomorphisms from the projective lift yields an
injective map
jC[[t]] : HomOC[[t]] (PC[[t]] , ∆C[[t]] ) ,→ HomOC[[t]] (PC[[t]] , ∇C[[t]] (λ))
of finitely generated free C[[t]]-modules of the same rank. We define a filtration on
HomOC[[t]] (PC[[t]] , ∆C[[t]] (λ)) by setting
−1
HomOC[[t]] (PC[[t]] , ∆C[[t]] (λ))(i) := jC[[t]] (ti · HomOC[[t]] (PC[[t]] , ∇C[[t]] (λ))).

19
The base change map π : HomOC[[t]] (PC[[t]] , ∆C[[t]] )  HomO (P, ∆(λ)) from Lemma
2.2.3 allows us to define a filtration on HomO (P, ∆(λ)) by taking the images of
HomOC[[t]] (PC[[t]] , ∆C[[t]] (λ))(i) under the map π. It is now easy to see that this filtra-
tion coincides with the filtration induced by the Jantzen filtration and, in particular,
that it does not depend on the choice of the projective lift we chose. (Cf. [Soe08],
section 4.6, for this paragraph.)

## 3.2 Andersen filtration

The references for the second filtration we mentioned above are [And97] and [Soe08].
Let T ∈ O be a tilting module, λ ∈ h∗ and A a local deformation algebra. By
Theorem 2.5.6 and Remark 2.6.4 we can choose a lift TA of T in OA . Recall that we
have an isomorphism HomOA (∆A (λ), ∇A (λ)) ∼ = A by Lemma 2.1.4. Thus, we can
view the composition pairing

HomOA (∆A (λ), TA ) × HomOA (TA , ∇A (λ)) → HomOA (∆A (λ), ∇A (λ))
(ϕ, ψ) 7→ ψ◦ϕ

as an A-bilinear map with values in A. Let A be such that over the quotient field Q :=
Quot(A) the tilting module TQ decomposes into a direct sum of Verma modules.
Since all Verma and dual Verma modules are simple in OQ , this is a non-degenerate
pairing of finitely generated free A-modules of the same rank. Thus it induces an
injective map

## aA : HomOA (∆A (λ), TA ) ,→ HomOA (TA , ∇A (λ))∗

where ∗ denotes the A-dual of an A-module. Again, we take A = C[[t]] to be the
completion of the coordinate ring of Cρ ⊂ h∗ along the maximal ideal associated to
0 ∈ h∗ . We can then define a filtration

## HomOC[[t]] (∆C[[t]] (λ), TC[[t]] )(i) := a−1 i ∗

C[[t]] (t · HomOC[[t]] (TC[[t]] , ∇C[[t]] (λ)) ) (3.1)

Definition 3.2.1. The Andersen filtration on the C-vector space HomO (∆(λ), T ) is
defined as the image of the filtration from (3.1) under the map

## 3.3 Connecting both filtrations

Recall Se the localization of S at the maximal ideal generated by h. Let Λ ∈ h∗ / ∼C
be a block of positive level, i.e., Λ contains a dominant weight. For λ, µ ∈ Λ we
define λ0 := −2ρ − λ and µ0 := −2ρ − µ and for any local deformation algebra A we

20

denote by tA : HomOA (PA (µ), ∆A (λ)) → HomOA (∆A (λ0 ), TA (µ0 )) the isomorphism
induced by the deformed tilting functor tA . Note that PA (µ) and TA (µ0 ) exist since
µ is of positive and µ0 of negative level. Furthermore, we denote the injective maps
given in sections 3.1 and 3.2 by

## jA : HomOA (PA (µ), ∆A (λ)) ,→ HomOA (PA (µ), ∇A (λ))

aA : HomOA (∆A (λ0 ), TA (µ0 )) ,→ HomOA (TA (µ0 ), ∇A (λ0 ))∗

where denotes the dual of an A-module.
Theorem 3.3.1. Let λ, λ0 , µ, µ0 be as above. There is a natural isomorphism

γ = γSe : HomOSe (PSe(µ), ∇Se(λ)) → HomOSe (TSe(µ0 ), ∇Se(λ0 ))∗

## which makes the following diagram commutative

S
je
HomOSe (PSe(µ), ∆Se(λ)) −−−→ HomOSe (PSe(µ), ∇Se(λ))
 
t
y Se
γ e
yS (3.2)
aSe ∗
HomOSe (∆Se(λ0 ), TSe(µ0 )) −−− → HomOSe (TSe(µ0 ), ∇Se(λ0 ))

## Corollary 3.3.2. The isomorphism

t = tC : HomO (P (µ), ∆(λ)) → HomO (∆(λ0 ), T (µ0 ))

identifies the filtration induced by the Jantzen filtration on the left with the Andersen
filtration on the right.
Proof. (cf. [Küb12a], Corollary 4.4) Recall the S-structure on C[[t]] coming from
the inclusion Cρ ⊂ h∗ . The structure map factors through a map Se → C[[t]] of local
deformation algebras. Since the diagram in Theorem 3.3.1 is compatible with the
base change maps from Lemma 2.2.3 and Lemma 2.2.4, the theorem implies that
the filtrations on HomOC[[t]] (PC[[t]] (µ), ∆C[[t]] (λ)) and HomOC[[t]] (∆C[[t]] (λ0 ), TC[[t]] (µ0 ))
defined by jC[[t]] and aC[[t]] are identified by the isomorphism tC[[t]] . But then, again
by base change, the two filtrations on HomO (P (µ), ∆(λ)) and HomO (∆(λ0 ), T (µ0 ))
are identified by tC .
Sketch of proof for Theorem 3.3.1. We only give a sketch of the prove since it is
more or less the same as in the semi-simple case, which is Theorem 4.3 in [Küb12a].
Denote by P(S) e the set of all prime ideals of Se of height one and for p ∈ P(S)
e let
Sp be the localization of Se at p. Note that we have a decomposition in OSp :
n
PSe(µ) ⊗Se Sp ∼
M
= PSp (µi )
i=1

## for certain µiS∈ Λ with µi ≥ µ. Furthermore, the S-block

e Λ decomposes into Sp -
blocks Λ = i∈I Λi for some index set I. We have seen in Lemma 2.4.1 that if

21
α∨ ∈/ p for all α ∈ R(Λ), then Λi consists of only one element for all i ∈ I and if p
is generated by the coroot of α ∈ R(Λ), Λi consists of at most two elements.
We have a similar decomposition for TSe(µ0 ) ⊗Se Sp which is compatible with the one
for PSe(µ), i.e., we have
n n
∼ ∼
M M
0
TSe(µ ) ⊗Se Sp = tSp ( PSp (µi )) = TSp (µ0i )
i=1 i=1

## where µ0i := −2ρ − µi .

The homomorphisms jSe, aSe and tSe of diagram (3.2) respect the above decompositions
and base change isomorphisms. Thus, after applying · ⊗Se Sp , diagram (3.2) splits
into the direct sum of diagrams of the form

jSp
HomOSp (PSp (µ), ∆Sp (λ)) −−−→ HomOSp (PSp (µ), ∇Sp (λ))

tS
y p (3.3)
aSp
 ∗
HomOSp (∆Sp (λ0 ), TSp (µ0 )) −−−→ 0 0
HomOSp (TSp (µ ), ∇Sp (λ ))

Tensoring with the quotient field Q := Quot(S) of S over Sp , these diagrams can be
embedded for any p in a diagram of isomorphisms of Q-vector spaces. If Λi contains
only one element µi , then either µi 6= λ and all appearing maps are zero, or µi = λ
and all appearing maps are isomorphisms of Sp -modules. Hence, in this case we get
the claimed isomorphism γSp as the composition of jS−1 p
, tSp and aSp .

Now assume we have an isomorphism γSp for every prime ideal p of height one
which makes the diagram (3.3) commutative. Then each γSp is the restriction of the
isomorphism
−1 ∼ ∗
γQ = aQ ◦ tQ ◦ jQ : HomOQ (PQ (µ), ∇Q (λ)) → HomOQ (TQ (µ0 ), ∇Q (λ0 ))

since jSp , aSp and tSp are restrictions of the isomorphisms of Q-vector spaces jQ ,
aQ and tQ to the corresponding Sp -lattices. So all localized maps come from the
same isomorphisms over Q. But this implies that we get back the morphisms of the
theorem by intersecting over all prime ideals of height one. In particular, we get an
isomorphism
∼ ∗
γSe : HomOSe (PSe(µ), ∇Se(λ)) → HomOSe (TSe(µ0 ), ∇Se(λ0 ))

once we have isomorphisms γSp for every prime ideal p of height one. The last
isomorphism follows from the freeness of all S-modules
e appearing in diagram (3.2),
the fact that

## HomOSe (PSe(µ), ∇Se(λ)) ∼

\
= HomOSp (PSe(µ) ⊗Se Sp , ∇Sp (λ))
p∈P(S)
e

22
∗
and an analogous equation for HomOSe (TSe(µ0 ), ∇Se(λ0 )) .

The localized modules over Sp resemble the representation theory over the Lie alge-
bra sl2 and are therefore much easier to handle. If Λi has only one element, we have
seen, that we get the map γSp very easily. If Λi has two elements and λ, µi ∈ Λi ,
then there are four cases to be discovered:

1. λ = µi and λ is antidominant in Λi ,

2. λ = µi and λ is dominant in Λi ,

3. λ < µi ,

4. λ > µi .

Since all four cases work as in the finite dimensional case, we refer for these cases
to the proof of Theorem 4.3 in [Küb12a], which is the main result of the author’s
diploma thesis.
Remark 3.3.3. Note that if µ lies in a block of negative level, there is no projective
cover of L(µ) in O. However, for any ν ∈ h∗ we can construct the object P ≤ν (µ) as
in Theorem 2.2.2. Note further, that there is no tilting module with maximal weight
0
µ0 = −2ρ − µ in O either. But for ν 0 = −2ρ − ν we can take T ≥ν (µ0 ) from Remark
0
2.5.7. It is then easy to see that the tilting functor maps P ≤ν (µ) to T ≥ν (µ0 ) and
thus induces an isomorphism
0
HomO (P (µ)≤ν , ∆(λ)) → HomO (∆(λ0 ), T ≥ν (µ0 )).

Since there should also be deformed versions and base change results in this case,
the proof of Theorem 3.3.1 should imply that the filtration on the left, induced by
the Jantzen filtration, is identified with the analogous Andersen filtration on the
right.
Remark 3.3.4. The progress report [And12] extends the main result of [Küb12a] to
generalize the connection between Andersen’s and Jantzen’s filtration in case of a
generic quantum group over a semisimple Lie algebra. Instead of working with the
tilting functor, Andersen relates the homomorphism space between a tilting and a
Verma module to the space of homomorphisms between a tilting and a dual Verma
module via the twisting functor corresponding to the longest element of the Weyl
group. The pairing of section 3.2 can also be used to define an Andersen filtration
on the second space of homomorphisms. In this setting, the same arguments from
[Küb12a] can be used to show that the filtration induced by Jantzen’s filtration
on the first space of homomorphisms is identified with the alternative version of
the Andersen filtration on the second space via the twisting functor. However, as
Andersen mentions in the end of his report, these ideas fail in the case that the
parameter of the quantum group is a root of unity.

23
4 Combinatorics of modules with a
Verma flag

In this section we recall the connection between a block of deformed category O and
sheaves on the moment graph associated to this block as developed in [Fie08a]. We
then apply this to get a combinatorial viewpoint of Theorem 3.3.1 and, in particular,
give a purely combinatorial description in the case of a finite dimensional semisimple
Lie algebra. This allows us to view the objects of Theorem 3.3.1 as graded objects
and the isomorphism identifying the Andersen and Jantzen filtration as a homoge-
neous map of degree zero. In the case of an affine Kac-Moody algebra in positive level
we are able to apply the main result of chapter 3 to relate multiplicities in consec-
utive subquotients of Jantzen’s filtration to coefficients of certain Kazhdan-Lusztig
polynomials.

4.1 Combinatorics
We recall the theory of sheaves on a moment graph which is based on [BM01],
[Fie08a] and [Jan08].

## 4.1.1 Moment graphs

For us, a graph is given by a tuple (V, E) of vertices V and edges E. Thus, V is a
countable set and E ⊂ P(V) is a subset of the power set of V such that each element
of E consists of exactly two vertices. This definition implies that a graph has no
double edges.
Definition 4.1.1. Let V be a finite dimensional complex vector space. An unordered
moment graph G = (V, E, α) over V is a graph (V, E) which is equipped with a labeling
map α : E → P(V ) which associates to any edge E ∈ E a line αE := α(E) in V .
We call a subset Y ⊂ V ∪ E open if the following holds:
x ∈ Y ∩ V ⇒ {E ∈ E | x ∈ E} ⊂ Y.
Hence, Y is open if with any vertex x ∈ Y it contains all edges which are adjacent
to x. The system {Y ⊂ V ∩ E | Y open} forms a topology on the set V ∪ E and we use
this to view the moment graph G as a topological space while we ignore the labeling
map.

24
4.1.2 Sheaves on moment graphs
Let S := S(V ) be the symmetric algebra of the finite dimensional complex vector
space V and let G = (V, E, α) be a V -moment graph. Let A be either S, which we
view as a graded ring such that deg(V ) = 2, or a localization Sp of S at a prime
ideal p. If p is the maximal ideal of S associated to 0 ∈ V , we write Se := Sp as
before. For x ∈ V we define x0 := {x} ∪ {E ∈ E | x ∈ E} to be the fundamental open
set associated to x. Note that by definition each edge E ∈ E is open. In contrast
to topological sheaf theory we follow [Fie08a] and indicate the stalk of a sheaf on
a moment graph at a point x by putting x as a superscript instead of a subscript.
Since x0 is the smallest open subset of G containing x ∈ V, the stalk of a sheaf M
of A-modules on G is given by its sections of x0 , i.e., Mx = M(x0 ). The stalk of the
open point E ∈ E is clearly given by ME = M({E}). For {E} ⊂ x0 we denote the
restriction map by ρx,E : Mx → ME .

## Defining A(x0 ) := A, A({E}) := A/α(E)A and ρx,E : A(x0 )  A({E}) to be the

quotient map for x ∈ V and E ∈ E, we obtain a sheaf A = AA of rings on G which
we call the structure sheaf of G. As a sheaf on a topological space, we can take
the global sections Γ(AA ) = Γ(A) of the structure sheaf and get the global section
functor
Γ : AA −mod −→ Γ(AA )−mod
which has a left adjoint functor, namely the localization functor L : Γ(AA )−mod →
AA −mod. The aim of the rest of this section is to explain the relation between Γ
and L when we restrict both functors to a certain category of sheaves on a moment
graph, which we will connect to the representation theoretic side. For this, it is
convenient to switch from the sheaf-theoretic to a more combinatorial description
of sheaves on moment graphs. In [BM01], Proposition 1.1, the authors show that
an AA -module M is uniquely characterized by its stalks at the edges and vertices
of the moment graph and by the corresponding restriction maps between them. By
this means, we can view an AA -module as a tuple ({Mx }, {ME }, {ρx,E }) with the
properties

## Furthermore, we define the structure algebra of G over A by

Y
ZA = ZA (G) := {(zx ) ∈ A | zx ≡ zy mod αE ∀E = {x, y}}.
x∈V

The structure algebra can be identified with the global sections of the structure
sheaf, i.e., ZA (G) ∼
= Γ(AA ) and the global section functor can be identified with the
functor
Γ : AA −mod −→ ZA (G)−mod

25
given by
Y
Γ(M) = {(mx ) ∈ Mx | ρx,E (mx ) = ρy,E (my ) ∀E = {x, y} ∈ E}
x∈V

## for M ∈ AA -mod. It is this combinatorial version of the global section functor we

are working with in the rest of this chapter.

Let I ⊂ G be a subgraph, i.e., a subset which contains with any edge E = {x, y} also
both vertices x, y ∈ V. We further assume that any subgraph I is also a full subgraph,
i.e., if I contains two vertices x, y ∈ V such that there is an edge E = {x, y} in G,
then E is an edge of I. This definition implies that the subgraph I is already
determined by its set of vertices and we will use this to identify I with the set of
vertices of I. For M a sheaf of A-modules on G and I ⊂ G a subgraph, we define
the sections of M over I by
Y
M(I) = Γ(I, M) := {(mx ) ∈ Mx | ρx,E (mx ) = ρy,E (my ) ∀E = {x, y} ∈ I}
x∈I

and also

A | zx ≡ zy mod αE ∀E = {x, y} ∈ I} ∼
Y
ZA (I) := {(zx ) ∈ = AA (I).
x∈I

## For A, A0 two localizations of S such that there exists a homomorphism A → A0 of

S-algebras, we get a functor

## by setting (M⊗A A0 )x := Mx ⊗A A0 , (M⊗A A0 )E := ME ⊗A A0 and ρM⊗A

0
x,E = ρMx,E ⊗idA0
for M ∈ AA -mod. For A = S and A = Sp where p ⊂ S is a prime ideal, we define Gp
0

to be the graph we get from G by canceling all edges with a labeling which is invert-
ible in Sp . For M ∈ AS -mod, M ⊗S Sp can then be viewed as a sheaf of Sp -modules
on the reduced graph Gp and if Gp decomposes into a disjoint union of subgraphs,
the sheaf M ⊗S Sp decomposes into a direct sum of sheaves on the corresponding
connected components.

## For a subgraph I ⊂ G, the restriction map ZA = ZA (G) → ZA (I) is not surjective

in general. We denote by ZAI = ZA (G)I the image of this map. Furthermore, for an
A-sheaf M on G we define its support by supp(M) := {x ∈ V | Mx 6= 0}.
Definition 4.1.2. We define SHA (G) to be the full subcategory of AA -mod that
consists of sheaves M with finite support, and such that for each x ∈ V the A-module
Mx is finitely generated and torsion free.
Definition 4.1.3. We denote by ZA −modf the full subcategory of ZA -mod that
consists of modules M which are finitely generated and torsion free as A-modules
and for which there exists a finite subgraph I ⊂ G such that the action of ZA on M
factors through ZAI .

26
By [Fie08a], the global section functor restricts to a functor

## To define the combinatorial localization functor due to [Fie08a] we need a further

assumption on the graph G. For γ ∈ V and I ⊂ G a subgraph, we define Zγ (I) :=
ZSγ (I) where Sγ is the localization of S at the prime ideal generated by γ. For any
finite subgraph I ⊂ G the map S ,→ Sγ induces a map ZS (G)I ,→ ZS (I) ,→ Zγ (I).
This defines an injective map ZS (G)I ⊗S Sγ ,→ Zγ (I).

## Definition 4.1.4. We call a moment graph G quasi-finite if the map ZS (G)I ⊗S Sγ ,→

Zγ (I) is surjective for any finite subgraph I ⊂ G and any γ ∈ V \{0}.

## To associate a sheaf L(M ) to a ZS -module M in the following we need enough

‘local information’ of the module M at any vertex x ∈ V which is guaranteed by the
quasi-finiteness condition.

4.1.3 Localization
For the rest of this section, G will always be a quasi-finite moment graph. We also
omit the subscript of the structure sheaf in case A = S. Let Q = L Quot(S) be the
quotient field of S. We get a natural decomposition Z(G)I ⊗S Q ∼ = x∈I Q for any
finite subgraph I ⊂ G. Let M ∈ Z−modf and I be a finite subgraph of G such that
the action of Z on M factors over Z(G)I . Since M is torsion free over L
S, we get an
inclusion M ,→ M ⊗SPQ. We can decompose 1 ⊗ 1 ∈ Z(G)I ⊗S Q ∼ = x∈I Q into
idempotents 1 ⊗ 1 = x∈I ex and set
M
M x := ex M ⊂ M ⊗S Q = ey (M ⊗S Q).
y∈I

## We define L(M )x := M x as the stalks of the sheaf L(M ) and

M (E) := (ex + ey )M + αE ex M ⊂ ex (M ⊗S Q) ⊕ ey (M ⊗S Q)

for E = {x, y} ∈ E. The projections onto the stalks induce surjective maps πx :
M (E) → M x , m 7→ ex m and πy : M (E) → M y , m 7→ ey m. We use these to define
the stalk L(M )E by the push out diagram
x π
M (E) −−−→ L(M )x
 
πy ρx,E
y y
ρy,E
L(M )y −−−→ L(M )E .

This construction also works for any localization A of S and defines a functor L :
ZA −modf → SHA (G).

27
Theorem 4.1.5 ([Fie08a], Proposition 3.5). The functors Γ and L induce a pair of

SHA (G) o
Γ / ZA −modf
L

and the canonical maps Γ(M) → ΓLΓ(M) and LΓL(M ) → L(M ) are isomorphisms.
We denote by ZA −modloc the essential image of L and by SHA (G)glob the essential
image of Γ. Then the theorem implies that (L, Γ) is a pair of mutual inverse equiv-
alences between the categories ZA −modloc and SHA (G)glob . We set CA = CA (G) =
ZA −modloc and get a global (as ZA -modules) and a local (as sheaves on G) point of
view on the category CA .
Remark 4.1.6. Let A = S. Recall that we view S as a graded ring with the usual
grading doubled, i.e., deg(V ) = 2. We call an S-sheaf M on a quasi-finite moment
graph G graded if for each vertex x and each edge E the S-modules Mx and ME are
graded and each restriction map ρx,E : Mx → ME is homogeneous of Qdegree zero.
Note that Z = ZS (G) inherits a grading from S, where the product x∈V S in the
definition of Z is meant to be taken degree wise. As in [Jan08], chapter 2.19, one
can show that for M ∈ SHS (G)glob a graded S-sheaf, the Z-module Γ(M) is graded.
On the other hand, for M ∈ Z−modloc , there is a unique structure of a graded sheaf
on L(M ), such that the map M → L(M )x is a homogeneous map of degree zero for
any x ∈ V.
In case A = S we redefine the category C = CS as the respective category of graded
Z-modules. Thus, we view an object in C either as a graded Z-module or a graded
S-sheaf and both graded pictures are compatible with the functors Γ and L. Further-
more, all homomorphisms of graded S-modules are considered to be homogeneous
of degree zero.

## 4.1.4 Sheaves on ordered moment graphs

For what follows we need an ordering on the set of vertices of the moment graph G.
Definition 4.1.7. A(n) (ordered) moment graph G = (V, E, α, ≤) is an unordered
moment graph G = (V, E, α) with a partial order on the set of vertices such that
two vertices, which are joined by an edge, are comparable. This implies that the
underlying directed graph associated to this order has no cycles.
Let M ∈ ZA −modf and I ⊂ G be a full subgraph. We define
M
MI := M ∩ ex (M ⊗A Q)
x∈I

and M
M I := Image(M ,→ M ⊗A Q  ex (M ⊗A Q)).
x∈I

We recall some notions from [Fie08a] which are partly inspired by topological sheaf
theory and which depend on the order on the moment graph.

28
Definition 4.1.8. We call a subgraph I ⊂ G F-open if, identified with its set of
vertices, is downwardly closed, i.e., if y ∈ I and x ≤ y then x ∈ I.

Definition 4.1.9. We say that M ∈ ZA −modf admits a Verma flag if the mod-
ule M I is free (graded free in case A = S) for any F-open subgraph I ⊂ G. We
denote the full subcategory of Z−modf consisting of modules with a Verma flag by
ZA −modV erma .

## Definition 4.1.10. We call a moment graph G = (V, E, α, ≤) a GKM-graph if it

is quasi-finite and if for any pair E, E 0 ∈ E with E 6= E 0 and E ∩ E 0 6= ∅ we
have α(E) 6= α(E 0 ), i.e., the labels attached to edges containing the same vertex are
pairwise distinct.

## We denote by ZA −modref the category of modules M ∈ ZA −modf such that for

any F-open subgraph I the module M I is reflexive.

## ZA −modV erma ⊂ ZA −modref ⊂ ZA −modloc

and we can view each of these categories as respective categories of sheaves on the
moment graph. Therefore, we denote them by

## Definition 4.1.12. For G an ordered moment graph and x ∈ G a vertex, we define

Ux := {E ∈ E | E = {x, y}, y ∈ V, x ≤ y} and Dx := {E ∈ E | E = {x, y}, y ∈ V, x ≥
y}.

## Now let x ∈ V be a vertex of G and M ∈ SHA (G). We define

M
M[x] := ker((ρx,E )E∈Dx : Mx → ME ).
E∈Dx

\
Mx := ker(ρx,E )
E

## Definition 4.1.13. For x ∈ V a vertex of the moment graph G, we define the

skyscraper sheaf VA (x) as the sheaf on G given by VA (x)x = A and VA (x)y =
VA (x)E = 0 for any vertex y 6= x and any edge E ∈ E.

29
The global sections Γ(VA (x)) of the skyscraper sheaf can be identified with the ZA -
module A on which any tuple (zu )u∈V ∈ ZA acts by multiplication with zx .

## Following [Fie08a], chapter 4.1, we introduce an exact structure on the additive

category ZA −modf . We call a sequence A → B → C in ZA −modf exact if for any
F-open subgraph I ⊂ G, the sequence 0 → AI → B I → C I → 0 is short exact
in the category of abelian groups. By [Fie08a], Theorem 4.1, this induces an exact
structure (cf. [Qui73]) on ZA −modf . Any subcategory of ZA −modf , which is stable
under extensions, inherits an exact structure from ZA −modf . Since an extension
of two reflexive (resp. free) A-modules is reflexive (resp. free) over A, we get that
C V erma and C ref are exact categories. Recall that an object P in an exact category A
is called projective if the functor HomA (P, ·) : A → AB into the category of abelian
groups takes exact sequences to short exact sequences.
Proposition 4.1.14 ([Fie08a], Proposition 5.1). Let P ∈ CAref and P = L(P ). Then
P is projective if
(i) for any vertex x of G, the A-module Px is projective, and

(ii) for any edge E = {x, y} with y ≤ x, the restriction map ρy,E : Py → PE
induces an isomorphism Py /αE Py → PE .
The next definition is needed for a more detailed description of projective sheaves
in section 4.3.2.
Definition 4.1.15. A sheaf M ∈ SHA (G) is called F-flabby if for any F-open sub-
graph I ⊂ G the restriction map Γ(M) = M(G) → M(I) is surjective.
An element M ∈ ZA −modf is called F-flabby if for any F-open subgraph I we have
M I ∈ ZA −modloc , i.e., M I is ‘generated by local relations’.
Note that by [Fie08a], Proposition 4.2, a sheaf M ∈ SHA (G)glob is F-flabby if and
only if the corresponding ZA -module Γ(M) is flabby.

The next theorem classifies the indecomposable projective objects in C ref from the
local point of view. We also sketch the proof of their existence since it provides
an algorithm to construct them. Furthermore, we need to assume that the moment
graph is bounded from above, i.e., for any vertex x ∈ V of the moment graph G the
set {y ∈ V | y ≥ x} is finite. Recall the support supp(M) = {x ∈ V | Mx 6= 0} for M
a sheaf on G.
Theorem 4.1.16 ([FW08], Theorem 6.4, and [Jan08], section 3.5). (i) For any
vertex x ∈ V of the ordered moment graph G there exists a graded S-sheaf
B(v) ∈ CSref with the properties

## • B(v) is indecomposable (even as a non-graded sheaf ) and projective,

• supp(B(v)) ⊂ {x | x ≥ v} and B(v)v ∼
= S.

30
(ii) If P is a graded projective S-sheaf in CSref , then there are vertices v1 , ..., vn ∈ V,
integers l1 , ..., ln ∈ Z and an isomorphism of graded S-sheaves of degree zero

P∼
= B(v1 )[l1 ] ⊕ ... ⊕ B(vn )[ln ].

## (iii) In case A = Se and P is projective in CSref

e , there exist vertices v1 , ..., vn ∈ V
and an isomorphism of S-sheaves
e

P∼
= B(v1 ) ⊗S Se ⊕ ... ⊕ B(vn ) ⊗S S.
e

Sketch of proof. We give the algorithm to construct B(v) in the first part of the
theorem starting at the vertex v. The uniqueness and the other properties can be
found in [Fie08a], [Jan08] and [BM01].
Of course, we have to set B(v)v = S and B(v)x = 0 for x  v. Proposition 4.1.14
implies that if E = {x, y} is an edge with x ≤ y then B(v)E and ρx,E are already de-
termined by the stalk B(v)x and the label of E. Using induction, we can assume that
we already constructed all stalks and restriction maps of B(v) on the subgraph {< y}
for a given vertex y ∈ V. We only have to construct B(v)y and the L restriction maps
ρy,E : B(v)y → B(v)E for all E that end in y. We define B(v)δy ⊂ E∈Dy B(v)E to
be the image of the following composition of maps
M M ⊕ρx,E M
B(v)({< y}) ⊂ B(v)x −→ B(v)x −→ B(v)E
x<y {x,y}∈Dy E∈Dy

and define B(v)y to be the projective cover of B(v)δy in the category of graded S-
modules. A restriction map ρy,E : B(v)y → B(v)E for E = {x, y} with x ≤ y is then
defined as the composition B(v)y → B(v)δy → B(v)E .

## 4.2 The link to representation theory

This section recalls the main result of [Fie08a] and explains how the deformed cate-
gory OSe is related to the moment graph picture. We fix a block Λ outside the critical
hyperplanes and recall the category MS,Λ e of modules in OS,Λ
flag. We want to associate a moment graph GΛ with the block Λ. For this, let us first
assume that Λ has an antidominant weight γ ∈ Λ. Recall the integral Weyl group
WSe(Λ) = W(Λ) and denote by Wγ the stabilizer of γ in W(Λ). Let V := W(Λ)/Wγ
be the set of vertices of the moment graph GΛ , which we identify with a set of mini-
mal representatives according to the length function given by the Bruhat order. We
connect two vertices x, y ∈ V with x 6= y by an edge E = {x, y} if there is a reflec-
tion sα ∈ W(Λ) associated to a root α ∈ R(Λ) such that x = sα y. Then we label
the edge E with the corresponding coroot α∨ ∈ h. These data define an unordered
moment graph over the vector space h. We now equip GΛ with the induced Bruhat
order on W(Λ)/Wγ to get an ordered moment graph GΛ = (V, E, α, ≤).
We can associate a second moment graph with Λ by defining the set of vertices to
be Λ and connect two elements λ, µ ∈ Λ by an edge E = {λ, µ} if there exists a

31
reflection sα ∈ W(Λ) such that λ = sα · µ and then label E with the associated
coroot α(E) = α∨ ∈ h. We now equip this moment graph with the ordering induced
by the ordering on h∗ which is defined through the set of positive roots.
Viewed as unordered moment graphs over h these two graphs are isomorphic, since
Λ∼ = W(Λ)/Wγ . However, the ordering on Λ in the second definition is in general a
refinement of the induced Bruhat order on the vertices of the first moment graph.
But the underlying directed graph is the same, i.e., if there is an edge E = {x, y} in
GΛ , then x ≤ y in the induced Bruhat order if and only if x · γ ≤ y · γ in the ordering
on Λ. Since the categories of sheaves on GΛ we are working with, only depend on the
underlying directed moment graph, we can use both definitions in the following. In
[Fie08a], Lemma 3.2, it is shown that GΛ is a GKM-graph. For the moment graph
GΛ and A a localization of S we define
VA (GΛ ) := ZA −modV erma
to be the category of ZA -modules with a Verma flag.

## If Λ contains a dominant weight λ ∈ Λ but no antidominant weight, we define GΛ

in the same way, but we reverse the order on the set of vertices. For an arbitrary
moment graph G we denote its tilted moment graph by t(G) and define it to be the
same unordered moment graph but with the order reversed. For A a localization
of S at a prime ideal and M ∈ ZA −modf , denote by DA (M ) := HomA (M, A) its
A-dual. It has a natural structure of a ZA -module. For z ∈ ZA , f ∈ D(M ) and
m ∈ M it is given by
(z.f )(m) := f (z.m).
In case A = S and M is a graded ZS -module, D = DS (M ) := i∈Z HomS (M, S)i is
L
again a graded ZS -module, where HomS (M, S)i are the graded homomorphisms of
degree i. In [Fie08a] it is shown that the functor DA : ZA −modf → (ZA −modf )opp
induces an equivalence
DA : VA (GΛ ) −→ VA (t(GΛ ))opp
on the corresponding sheaves with Verma flag. For ν ∈ Λ we define Λ≤ν := {µ ∈
Λ | µ ≤ ν} (resp. Λ≥ν := {µ ∈ Λ | µ ≥ ν}) and denote by GΛ≤ν (resp. GΛ≥ν ) the full
subgraph of GΛ with vertices corresponding to Λ≤ν (resp. Λ≥ν ).

## Now we assume again that Λ contains an antidominant weight λ ∈ Λ and let ν ∈ Λ.

{≤ν}
Let PSe≤ν (λ) = PSe (λ) be a projective cover of ∆Se(λ) in the truncated category
J
OS,Λ
e corresponding to the open and bounded subset J = {≤ ν}. For each pair
0
(ν 0 , ν) ∈ h∗ × h∗ with ν 0 ≥ ν we have a natural surjection PSe≤ν (λ)  PSe≤ν (λ). In the
category U (g)-S-Bimod
e we can define the inverse limit
PSe∞ (λ) := lim PSe≤ν (λ).
←−
Note that PSe∞ (λ) is not an object of OS,Λ
e in general. However, Fiebig proves in [Fie06]

## that for M ∈ M e the S-module Homg (P ∞ (λ), M ) carries a natural structure as

S,Λ
e
S
e S
e

32
a module over the categorical center Z(OS,Λ
e ). This structure is given by the action
of the center on the truncated projective objects PSe(λ)≤ν . In [Fie03] it is proven
that Z(OS,Λ
e ) coincides with the structure algebra ZS e(GΛ ) of the moment graph GΛ .
Hence, we get a functor VΛ := HomgSe (PSe∞ (λ), ·) : OS,Λ
e → ZSe(GΛ )−mod. Now the
main result of [Fie08a] is
Theorem 4.2.1 ([Fie08a], Theorem 7.1 and Proposition 7.2). The functor VΛ re-
stricts to an equivalence

V = VΛ : MS,Λ
e → VS
e(GΛ )

## of exact categories, where MS,Λ

e inherits the exact structure from OS,Λ
e . Further-
≤ν ≤ν ≤ν
more, if PSe (µ) ∈ MS,Λ
e is a projective cover of ∆Se(µ) in OS,Λ e (µ)) is
e , then V(PS
isomorphic to the global sections of the Braden-MacPherson sheaf B≤ν
e (x) ∈ VS
S e(GΛ )
≤ν
at the vertex given by µ = x · λ on the moment graph GΛ .
Remark 4.2.2. Let Λ contain a dominant weight. Then Fiebig’s proof of the above
Theorem implies that the composition of functors
opp
tSe opp VΛ0 De
MS,Λ
e −→ MS,Λ
e 0 −→ VSe(GΛ0 )opp −→
S
VSe(GΛ )
0
also defines an equivalence between MS,Λ e and VSe(GΛ ), where Λ := −2ρ − Λ and GΛ
is identified with t(GΛ0 ). In this case, Theorem 7.1 of [Fie08a] implies that indecom-
posable projective objects correspond to Braden-MacPherson sheaves. If Λ contains
both, a dominant and an antidominant weight, then Fiebig proved in [Fie06], Re-
mark 6, that there is a natural equivalence between the composition of the above
functors and the functor associated to the homomorphisms of the antidominant
projective.

4.3 Applications
4.3.1 The affine case
We first recall a hard Lefschetz theory on Braden-MacPherson sheaves (cf. [Fie12b]
and [Soe08]) before we give applications of the preceding chapters. As before, let
C[t] be the polynomial ring in one variable coming from the embedding Cρ ⊂ h∗
and denote by C[[t]] its completion with respect to the ideal generated by t. Let
Λ be a block of negative level with an antidominant weight γ ∈ Λ. We denote
by GΛ the moment graph associated to Λ and by Gt(Λ) the one associated to the
block t(Λ) = −2ρ − Λ of positive level. Recall that Gt(Λ) can be identified with the
same moment graph but with reversed order, which we denote by t(GΛ ). We further
assume that Λ is regular, i.e., the stabilizer Wγ of the antidominant weight γ ∈ Λ
or any other weight in Λ is trivial and we can identify Λ = W(Λ). Let x, y ∈ W(Λ)
and define µ = x · γ and λ = y · γ. Let ZS be the structure algebra of GΛ and
M ∈ ZS − modf . For x ∈ GΛ a vertex of GΛ recall the definitions
M x = Image(M ,→ M ⊗S Q → ex (M ⊗S Q))

33
and
Mx := M ∩ ex (M ⊗S Q).
We get a natural inclusion
Mx ,→ M x
of graded S-modules. Define B(x) := Γ(B(x)) to be the global sections of the
Braden-MacPherson sheaf associated to the vertex x on the tilted moment graph
t(GΛ ). We define µ0 := −2ρ − µ and λ0 := −2ρ − λ. By Theorem 4.2.1 we see that
Fiebig’s equivalence maps the projective cover PSe(µ0 ) to B(x)⊗S S,
e i.e., V(P e(µ0 )) =
S
B(x) ⊗S S,e which is viewed as an object in V(t(GΛ )). Further, the tilting module
TSe(µ) is mapped to DSe(B(x) ⊗S S)
e viewed as an object of V(GΛ ).

## Theorem 4.3.1 ([Fie08b], Theorem 6.1). Up to a degree-shift, B(x) is self-dual,

i.e.,
DS (B(x)) ∼
= B(x)[2l(x)]
where l(x) denotes the length of the element x ∈ W(Λ) according to the Bruhat
order.
We get a chain of isomorphisms

## HomOSe (∆Se(λ), TSe(µ)) ∼

= HomZ(GΛ ) (V∆Se(λ), VTSe(µ))

= HomZ(GΛ ) (VS (y), DS (B(x))) ⊗S S.
e

Thus, by self-duality we can naturally identify HomOSe (∆Se(λ), TSe(µ)) with the costalk
B(x)y ⊗S Se at y. Similarly, we can identify the S-module
e HomOSe (TSe(µ), ∇Se(λ))∗ with
the stalk B(x)y ⊗S Se (cf. the proof of Proposition 7.1 (3) in [Fie08b]). The pairing
we used to define the Andersen filtration led to an injective map

HomOSe (∆Se(λ), TSe(µ)) ⊗Se C[[t]] ,→ HomOSe (TSe(µ), ∇(λ))∗ ⊗Se C[[t]]

## B(x)y ⊗S Se ⊗Se C[[t]] ,→ B(x)y ⊗S Se ⊗Se C[[t]]

where C[[t]] is again the completion of the S-algebra C[t] from above. To define the
filtration on B(x)y ⊗S C that corresponds to the Andersen filtration under the above
identification, we can alternatively look at the map

## B(x)y ⊗S C[t] ,→ B(x)y ⊗S C[t]

and the Andersen filtration on the vector space HomO (T (µ), ∆(λ)) can be identi-
fied with the filtration on B(x)y ⊗S C which we get by first taking the preimages of
ti (B(x)y ⊗S C[t]) under the above inclusion and then taking the images in B(x)y ⊗S C
of this filtration under · ⊗C[t] C.

MacPherson sheaf which induce gradings on the vector spaces B(x)y ⊗S C and

34
B(x)y ⊗S C. The moment graph we associated to Λ corresponds to the one Fiebig
works with in [Fie12b]. Thus, Theorem 4.7 of loc. cit. (which is Theorem 4.4 in
[Soe08] in the finite case) shows that the cokernel of the inclusion of the graded
C[t]-modules B(x)y ⊗S C[t] ,→ B(x)y ⊗S C[t] fulfills the Lefschetz condition, i.e., the
resulting short exact sequence splits into the direct sum of short exact sequences of
the form
ti+1
C[t][−i − 2] ,→ C[t][i]  (C[t]/(ti+1 ))[i].
This implies, that the Andersen filtration on B(x)y ⊗S C and the grading filtration
on this graded vector space coincide. By [Fie12b], Theorem 4.6, the subquotients of
this filtration are given by coefficients of affine Kazhdan-Lusztig polynomials. Let
hy,x be the normalized Kazhdan-Lusztig polynomial associated to x, y ∈ W(Λ) in
i
the notation of [Soe97] and let Aµ,λ denote the i-th subquotient of the Andersen
filtration on HomO (∆(λ), T (µ)). Recall the definition µ = x · γ and λ = y · γ from
above. Then, in formulas, Fiebig’s theorem says
i
dimC Aµ,λ = hiy,x

## where the right hand side is the i-th coefficient of hy,x .

Now let ∆(λ0 )i /∆(λ0 )i+1 be the i-th subquotient of the Jantzen filtration. Since the
functor HomO (P (µ0 ), ·) : O → C−mod is exact, the dimension of the vector space
HomO (P (µ0 ), ∆(λ0 )i /∆(λ0 )i+1 ) coincides with the dimension of the i-th subquotient
of the filtration on the vector space HomO (P (µ0 ), ∆(λ0 )) induced by the Jantzen
filtration on ∆(λ0 ). Furthermore, we have

dimC HomO (P (µ0 ), ∆(λ0 )i /∆(λ0 )i+1 ) = [∆(λ0 )i /∆(λ0 )i+1 : L(µ0 )].

## Now Corollary 3.3.2 gives the following

Theorem 4.3.2. [∆(λ0 )i /∆(λ0 )i+1 : L(µ0 )] = hiy,x .
Remark 4.3.3. Note that we proved Theorem 4.3.2 only for regular weights of positive
level. This is due to the fact that Fiebig only considers the moment graph of a regular
block in Theorem 4.6 of [Fie12b]. However, since the proof of Fiebig’s theorem is just
a straight forward generalization to affine Kac-Moody algebras of the main result in
[Soe08], and since Soergel treats the singular case as well, it should not be difficult
to treat the singular affine case of Theorem 4.3.2 in a similar manner, but still only
in positive level.
Remark 4.3.4. In [Sha12] Shan generalizes the Beilinson-Bernstein localization of
[BB93] to affine Kac-Moody algebras and gets an analogous version of Theorem
4.3.2 for regular weights of negative level. Her methods are thus very different from
ours but her approach has the advantage of being able to treat the parabolic case as
well. In [SVV11] the authors consider a certain truncated subcategory of a block of
category O over an affine Kac-Moody algebra. In case the block is of negative level,
the subcategory corresponds to a truncated category O as we defined it above. If the
block is of positive level, the subcategory corresponds to a truncation from below,
i.e., only Verma modules with highest weight in a truncated subset of the block are

35
allowed to appear in a Verma flag. Further, the authors of loc. cit. state a possible
Koszul-duality result between truncated singular blocks of negative (resp. positive)
level and truncated parabolic blocks of positive (resp. negative) level. In this vague
context one could view Theorem 4.3.2 as a Koszul dual result to Shan’s approach.

## 4.3.2 The finite case

The content of this subsection can be found as published version in [Küb12b]. We
consider a semisimple Lie algebra g. Since we always have an antidominant weight
in each block Λ ∈ h∗ / ∼Se we do not need the tilting functor to define Fiebig’s equiv-
alence in Theorem 4.2.1. The aim of this section is to prove directly that tilting
modules correspond to Braden-MacPherson sheaves constructed along the reversed
order and repeat the result about the Andersen-Jantzen filtrations in a purely com-
binatorial manner without using Arkhipov’s tilting functor. We then show how this
fits in the Koszul-duality result of [BGS96].

## Since the set of blocks Λ is in one-to-one correspondence to their antidominant

weights λ ∈ Λ, we write OA,λ := OA,Λ . We fix an antidominant weight λ ∈ h∗ for
the rest of this chapter. In this case the structure functor is given by
f
e −→ ZS
V := HomOSe (PSe(λ), ·) : OS,λ e(GΛ )−mod

## V := HomOSe (PSe(λ), ·) : MS,λ

e −→ VS
e(GΛ )

where again GΛ denotes the moment graph associated to Λ. For this chapter we take
the definition of GΛ where the set of vertices V of GΛ is given by W(Λ)/Wλ and the
order is the induced Bruhat order. Note that with this convention, projective objects
in OSe correspond to Braden-MacPherson sheaves on GΛ . Let A be a localization of
S at a prime ideal and denote by Q the field of fractions of S.

e

## (i) If w ∈ V, then L(V∆A (w · λ)) ∼

= L(V∇A (w · λ)) ∼
= VA (w).
(ii) Let M ∈ OA,λ admit a Verma or a Nabla flag. Then VM is a free A-module
of finite rank.

## Proof. The proof is analogous to [Jan08] Proposition 4.13.

We are now going to prove that tilting modules correspond to Braden-MacPherson
sheaves associated to the moment graph with reversed order. For this we first repeat
some reformulations of [Jan08] from [Fie08a].

## Definition 4.3.6. A sheaf M on a moment graph G is generated by global sections

if the map Γ(M) → Mx is surjective for every x ∈ V. (Here we use the notation of
[Jan08].)

36
Due to [Jan08] an equivalent description of an F-projective sheaf P on the GKM-
graph G is then given by the properties

## 3. Any ρx,E with x ∈ V, E ∈ Ux induces an isomorphism Px /αE Px → PE of (graded)

A-modules (if A = S).

## P 3.14). Suppose that G is a GKM-graph.

Proposition 4.3.7 ([Jan08], Proposition
Let M ∈ ZA −modf and define eI := x∈I ex for any subgraph I ⊂ G. Suppose in
addition, that eI M is a reflexive A-module for any F-open subgraph I of G. Then
L(M ) is F-flabby on G and we have an isomorphism

eI M −→ Γ(I, L(M )).

## Following [Jan08], section 4.14, we construct certain submodules and quotients of a

module M ∈ OA . Let D be a subset of h∗ with the property: If λ ∈ D and µ ∈ h∗
with µ ≤ λ, then µ ∈ D. Set for M ∈ OA
X
OD M := U (gA )Mµ and M [D] := M/OD M.
µ∈D
/

## Proposition 4.3.8 ([Jan08], section 4.14.). (i) If M has a Verma flag, so do

OD M and M [D].

(ii)
 
∆A (λ) if λ ∈
/ D, 0 if λ ∈
/ D,
O ∆A (λ) ∼
D
= and ∆A (λ)[D] ∼
=
0 else ∆A (λ) else.

## We now change notation: For the moment graph GΛ = (V, E, α, ≤) associated to

the block OA,λ we call an F-open subgraph ↑-open and a subgraph which is F-open
according to the moment graph with the reversed order, we call ↓-open. A subgraph
I with set of vertices V 0 is ↑-open if and only if I c is ↓-open, where I c is the full
subgraph of G with vertices V c := V − V 0 . For I an ↓-open subgraph, set D equal to
c
the set of all ν ∈ h∗ , such that there exists x ∈ V c with ν ≤ x · λ. Set OI M := OD M
and M [I c ] = M [D].

## If not stated otherwise, we set A = Se and for p ⊂ Se a prime ideal we denote by Ap

the localization of A at p for the rest of this section.

## Lemma 4.3.9. Let M ∈ OA be a tilting module. There is a natural isomorphism

eI VM ∼
c
= V(d(OI M )).

37
Proof. (cf. the proof of Lemma 4.16 of [Jan08]) Let V 0 be the set of vertices of I.
Dualizing the short exact sequence
c
OI M ,→ M  M [I c ]

we get, by self-duality of M and the freeness of all modules over A, a short exact
sequence
c
d(OI M )  M ←- d(M [I c ]).
Then the exactness of V yields a short exact sequence
c
V(d(OI M ))  V(M ) ←- V(d(M [I c ])).
c
Denote by ϕ : V(M )  V(d(OI M )) the surjective map from the above short exact
sequence. Then ker(ϕ) ∼ = V(d(M [I c ])) and we want to show that ker(ϕ) coincides
with the set X := {f ∈ V(M ) | eI f = 0}. Since d(M [I c ]) ⊗A Q is the direct sum of
Verma modules of the form ∆Q (x·λ) ∼ = ∇Q (x·λ) with x ∈/ V 0 , we get ew f = 0 for any
c 0
f ∈ V(d(M [I ])) and w ∈ V . But this implies eI f = 0 which proves ker(ϕ) ⊂ X.
So let f ∈ X. Then the map f ⊗ idQ is zero on the direct summands of PA (λ) ⊗A Q
that are isomorphic to ∆Q (w · λ) with w ∈ V 0 . Thus the image of f ⊗ idQ lies in
d(M [I c ]) ⊗A Q. Now the short exact sequence
c
d(OI M )  M ←- d(M [I c ])

## splits over A and we conclude M ∩ d(M [I c ]) ⊗A Q = d(M [I c ]). Thus we have

f ∈ V(d(M [I c ])) and we get a natural isomorphism eI VM ∼
c
= V(d(OI M )).
Definition 4.3.10. A sheaf M on G is called ↓-flabby (resp. ↓-projective) if it is
F-flabby (resp. F-projective) according to the moment graph G with reversed order.
We call it ↑-flabby (resp. ↑-projective) if it is F-flabby (resp. F-projective) according
to the usual order on G.

## Proposition 4.3.11. Let M be a tilting module. Then L(VM ) is ↓-flabby.

Proof. For any ↓-open subgraph I we get that eI VM is a free A-module by Lemma
4.3.9. Now Proposition 4.3.7 tells us that L(VM ) is ↓-flabby.

Notation 3. For x ∈ V we denote by B↑ (x) the BMP-sheaf B(x) for the moment
graph with the original order. For the moment graph with reversed order we denote
the BMP-sheaf at the vertex x by B↓ (x).

In the following we want to show that the indecomposable deformed tilting module
TA (x · λ) with highest weight x · λ corresponds to B↓ (x) ⊗S A under L ◦ V (A = S).
e
For this we need some preparation.

## Lemma 4.3.12. ([Fie08a], Lemma 7.4.) Let M ∈ O admit a Verma flag, µ ∈ h∗

and k ∈ N. If g : (∆C (µ))k → M is a morphism which induces an injective map
gµ : (∆C (µ))kµ → Mµ on the µ-weight spaces, then g is injective.

38
Lemma 4.3.13. Let T ∈ OA,λ be a tilting module. Then for any w ∈ V the stalk
L(VT )w is free over A of rank r := (T : ∆A (w · λ)).
Proof. (cf. the proof of Lemma 4.18 of [Jan08]) We want to construct an isomor-

phism ew (VT ) = L(VT )w − → V(∇A (w · λ)r ). Using base change, Lemma 2.2.4, and
the fact that over the residue field C we have dimC HomOC (∆C (w · λ), T ⊗A C) =
dimC HomOC (T ⊗A C, ∇C (w · λ)) = (T ⊗A C : ∆C (w · λ)), we can deduce that
HomOA (∆A (w · λ), T ) is a free A-module of rank r. Now choose an A-basis f1 , ..., fr
of HomOA (∆A (w · λ), T ). Dualizing yields a basis df1 , ..., dfr of HomOA (T, ∇A (w · λ)).
Consider the map
f : ∆A (w · λ)r −→ T, (v1 , ..., vr ) 7→
P
fi (vi ).

## Dualizing this map yields by self-duality of T

df : T −→ ∇A (w · λ)r , v 7→
P
(dfi (v)).
Applying the functor V gives a map

## Vdf : VT −→ V∇A (w · λ)r

which factors through ew (VT ), since V∇A (w · λ)r has support {w}. So this implies
a well-defined map
(Vdf )w : ew VT −→ V∇A (w · λ)r
which is injective, since it becomes an isomorphism after applying · ⊗A Q. We now
want to prove that (Vdf )w is surjective. For this, by Nakayama’s lemma and by
exactness of V, it is enough to prove that
df ⊗ idC : T ⊗A C −→ ∇A (w · λ)r ⊗A C

## is surjective. But, by Lemma 4.3.12, f : ∆C (w · λ)r → T ⊗A C is injective and thus

we get the surjectivity of the dual map df : T ⊗A C → ∇C (w · λ)r . Since V∇A (w · λ)r
is free of rank r over A we get the result.
Theorem 4.3.14. Let T ∈ OA,λ be a tilting module (A = S).
e Then L(VT ) is
↓-projective as an A-sheaf on GΛ .
Proof. (cf. the proof of Proposition 4.20 of [Jan08]) By [Jan08] section 2.12.(A)
L(VT ) is generated by global sections and ↓-flabby by Proposition 4.3.11. Further-
more, the lemma above shows that (L(VT ))w is free over A for every w ∈ V. Hence,
we only have to prove that for any edge E := {x, y}, x > y and α(E) = α∨ where
x, y ∈ W(Λ)/Wλ and α∨ the coroot of a root α ∈ R(Λ), the map ρx,E : L(VT )x →
L(VT )E induces an isomorphism

L(VT )x /α∨ L(VT )x −→ L(VT )E .

Since α∨ L(VT )x ⊂ kerρx,E , we only have to prove kerρx,E ⊂ α∨ L(VT )x . For this it
is enough to show that

## kerρx,E ⊂ α∨ (L(VT )x ⊗A Ap ) = α∨ · ex (L(VT ) ⊗A Ap ) (4.1)

39
for every prime ideal p ⊂ A of height 1. For α∨ ∈
/ p, (4.1) follows since α∨ is invertible
in Ap . So we have to prove (4.1) for p = Aα∨ . Since ρx,E is a push-out map, we can
identify kerρx,E with the set ex {u|u ∈ VT (E), ey u = 0}. An element u ∈ VT (E) is
of the form u = (ex + ey )v + α∨ ex w for v, w ∈ VT . Thus, we have to show that

## Since x > y, we get TAp (y ·λ) ∼

= ∆Ap (y ·λ) and it is easy to see that in this subgeneric

case we also have TAp (x·λ) = PAp (y·λ). Now we can identify (ex +ey )(VT ⊗A Ap ) with
a direct sum of Ap -modules of the form M := HomgAp (PAp (y · λ), ∆Ap (y · λ)) ∼ = Ap
and N := HomgAp (PAp (y · λ), TAp (x · λ)). But by Proposition 2.4.2 we get N ∼ =

Ap (ex + ey ) + Ap α ex .
If f ∈ M , we get ex f = 0 which is equal to α∨ ex f . If f ∈ N , ey f = 0 implies
ex f ∈ α∨ Ap ex . But with our identification we get ex f ∈ α∨ ex (VT ⊗A Ap ).

## Corollary 4.3.15. We have L(VTA (w · λ)) ∼

= B↓ (w) ⊗S A for all w ∈ W(Λ).
Proof. The proof is essentially the same as the one for Theorem 4.22 of [Jan08] and
relies on the facts that L(VTA (w · λ)) is ↓-projective and indecomposable. Now the
description of the indecomposable ↓-projective sheaves by BMP-sheaves gives the
corollary.
Let w◦ ∈ W(Λ) be the longest element according to the Bruhat order. Then the
multiplication from the left

## is an isomorphism of moment graphs in the sense of [Lan12]. Thus, it induces a

pull-back functor ([Lan12], Definition 3.3)

## Lemma 4.3.17. We have w◦∗ (VA (t(GΛ ))) = VA (GΛ ).

Proof. Let M ∈ VA (t(GΛ )). Using Lemma 4.8 of [Fie08a], we have to show that
w◦∗ (M) is ↑-flabby and (w◦∗ (M))[x] is free over A for x ∈ V.
Let I be ↓-open. Then w◦ I is ↑-open and we get that

Γ(M) ∼
= Γ(w◦∗ (M)) → Γ(I, w◦∗ (M)) ∼
= Γ(w◦ I, M)

is surjective since M is flabby. As (w◦∗ (M))[x] = M[w◦ x] the claim follows. Note that
the definitions of (·)[x] and (·)[w◦ x] differ from each other in the sense that the first
underlies the graph t(GΛ ) and the second GΛ .

40
We get an equivalence of categories

## = B↓ (w◦ x) ⊗S A and w◦∗ (VA (x)) ∼

with the properties: w◦∗ (B↑ (x) ⊗S A) ∼ = VA (w◦ x).
Hence, using the self-duality of Braden-MacPherson sheaves and of the sky scraper
sheaves in this case, the composition

## is an equivalence with FA (B↑ (x) ⊗S A) ∼

= B↓ (w◦ x) ⊗S A and F (VA (x)) ∼
= VA (w◦ x)
which are isomorphisms of graded sheaves if A = S.

Now we can lift the functor FA via Fiebig’s equivalence to a functor tA on the
representation theoretic side such that the following diagram of functors commutes:

V
MA,λ −−−→ VA (GΛ )
 

tA y
F
y A
Vopp
Mopp
A,λ −−−→ VA (G)
opp

## Theorem 4.3.18. Let λ ∈ h∗ be antidominant, x, y ∈ W(Λ)/Wλ and w◦ ∈ W(Λ)

the longest element. Denote by A = Se the localization of S at 0. There exists an
isomorphism l = lA (x, y) which makes the diagram

j
HomOA (PA (x · λ), ∆A (y · λ)) −−−−→ HomOA (PA (x · λ), ∇A (y · λ))
 
 
yt yl
a
HomOA (∆A (w◦ y · λ), TA (w◦ x · λ)) −−−−→ (HomOA (TA (w◦ x · λ), ∇A (w◦ y · λ)))∗

## commutative. Here j = jAy·λ (PA (x · λ)) and a = aw A

◦ y·λ
(TA (w◦ x · λ)) denote the
inclusions defined in chapter 3 and t = tA denotes the isomorphism induced by the
functor tA from above.

Proof. The proof is essentially the same as the one for Theorem 3.3.1 (cf. Theorem
4.3. in [Küb12a]), where we proved the analogous result for tA the tilting functor.

Denote by tC : Homg (P (x · λ), ∆(y · λ)) → Homg (∆(w◦ y · λ), T (w◦ x · λ)) the isomor-
phism we get from tA ⊗A idC after base change. The next corollary now follows in
the same way as Corollary 3.3.2.

## Corollary 4.3.19. The isomorphism

tC : Homg (P (x · λ), ∆(y · λ)) → Homg (∆(w◦ y · λ), T (w◦ x · λ))

identifies the filtration induced by the Jantzen filtration with the Andersen filtration.

41
Now we consider C ∼
= S/Sh as a simple graded S-module living in degree 0. The
map

tC : Homg (P (x · λ), ∆(y · λ)) → Homg (∆(w◦ y · λ), T (w◦ x · λ))
can then be identified with

FS ⊗ idC : HomVS (GΛ ) (B↑ (x), VS (y)) ⊗S C → HomVS (GΛ ) (VS (w◦ y), B↓ (w◦ x)) ⊗S C

which is now an isomorphism of graded C-vector spaces. But using the proof of
Proposition 7.1. (3) in [Fie08b] this isomorphism becomes a graded isomorphism
between certain costalks of the Braden-MacPherson sheaves, namely an isomorphism

ϕ : B↑ (x)y ⊗S C → B↓ (w◦ x)w◦ y ⊗S C.

Remark 4.3.20. As before, the filtration on B↓ (w◦ x)w◦ y ⊗S C induced by the Ander-
sen filtration coincides with the grading filtration we get from the grading on the
the filtration on B↑ (x)y ⊗S C, which is induced by the Jantzen filtration, with the
filtration on B↓ (w◦ x)w◦ y ⊗S C, which is induced by the Andersen filtration, we get
that the Jantzen filtration coincides with the grading filtration coming from the
grading on L the Braden-MacPherson sheaf B↑ (x). Let λ ∈ Λ ⊂ h∗ be antidominant
and Pλ := P (w · λ) be a minimal projective generator of Oλ . Then there is an
equivalence of categories

## HomO (Pλ , ·) : Oλ −→ Mof − EndO (Pλ )

where Mof − EndO (Pλ ) denotes the category of finitely generated right modules over
EndO (Pλ ). In [BGS96] it is shown that E := EndO (Pλ ) is a non-negatively graded
algebra with E0 being semisimple. It turns out (see e.g. [Str03]) that this grading
coming from the geometric description of Oλ essentially coincides with the grading on
EndO (Pλ ) we get from the direct sum of Braden-MacPherson sheaves corresponding
to Pλ . For ∆(µ) ∈ Oλ , HomO (Pλ , ∆(µ)) is a graded E-module and the filtration
on HomO (Pλ , ∆(µ)) induced by the Jantzen filtration coincides with the grading
filtration on this space. But since E0 is semisimple and E non-negatively graded,
the subquotients of the grading filtration are semisimple and so are the Jantzen
filtration layers of ∆(µ).

42
5 Critical representations of affine
Kac-Moody algebras

In this chapter we work over an untwisted affine Kac-Moody algebra. In the preced-
ing chapters of this thesis we mainly focused on the non-critical representations. In
the critical case it is important to emphasize on the particular structure of an affine
Kac-Moody algebra as a central extension of the loop algebra of a finite-dimensional
simple complex Lie algebra. Distinguishing between the root systems corresponding
to the finite and the affine Lie algebra is very important for this chapter. For this
reason, we recall the construction of an affine Kac-Moody algebra here, although it
would also have made sense to place it in chapter 1. We redefine some notation from
chapter 2 as well to make it suitable for the purpose of this chapter. After having
recalled the restricted category O and, in particular, restricted Verma modules, we
will prove the main results of this chapter, which is the calculation of the center of
the deformed, restricted category O (Theorem 5.3.12) and a Jantzen sum formula
for restricted Verma modules (Theorem 5.4.1). The first result can be found in the
preprint [Küb13]. The result about the restricted Jantzen filtration is content of the
preprint [Küb12c].

## 5.1 Affine Kac-Moody algebras

References for this section are [Kac90] and [AF12b]. Let g be a simple Lie algebra
with a Borel subalgebra b and a Cartan subalgebra h. We denote by R the finite
root system with positive roots R+ and by Π the simple roots. Moreover, denote by
W the finite Weyl group and by κ : g × g → C the Killing form. We take a non-split
g of the loop algebra g ⊗C C[t, t−1 ]. As a vector space, e
central extension e g is the
−1
direct sum (g ⊗C C[t, t ]) ⊕ CK, where K is a central element.

Adding a derivation operator D with the property [D, ·] = t ∂t , we get the affine Kac-
Moody algebra b g associated to g. As a vector space, we have b g = (g ⊗C C[t, t−1 ]) ⊕
CK ⊕ CD and the Lie bracket is given by

g] = {0}
[K, b

[D, x ⊗ tn ] = nx ⊗ tn
[x ⊗ tn , y ⊗ tm ] = [x, y] ⊗ tm+n + nδm,−n κ(x, y)K

43
where x, y ∈ g and n, m ∈ Z. The Borel subalgebra of b
g corresponding to b ⊂ g is
given by
b = (g ⊗C tC[t] + b ⊗C C[t]) ⊕ CK ⊕ CD
b
g is
while the corresponding Cartan subalgebra of b

h = h ⊕ CK ⊕ CD.
b

g from chapter
In our new notation for this section the triangular decomposition of b
1 becomes
g=b
b n⊕b h⊕bn+
where again b n+ the direct sum of positive root
n is the direct sum of negative and b
g.
spaces of b

## 5.1.1 Affine roots, Weyl groups and bilinear forms

For a vector space V we denote by h·, ·i : V ∗ × V → C the natural pairing with its
dual space. The projection b h = h ⊕ CK ⊕ CD → h induces an embedding h∗ ⊂ b h∗ .
h∗ . We define two
By this embedding we can consider all finite roots as elements of b
h∗ by the relations
weights Λ◦ , δ ∈ b

## hδ, h ⊕ CKi = {0}

hδ, Di = 1
hΛ◦ , h ⊕ CDi = {0}
hΛ◦ , Ki = 1
h are given by R
g with respect to b
Then the roots of b bre ∪ R
b=R bim where

bre := {α + nδ | α ∈ R ⊂ R,
R b n ∈ Z}
are the real roots, and
bim := {nδ | n ∈ Z, n 6= 0}
R
b+ , i.e., the roots of b
the imaginary roots. The positive roots R b with respect to b
h are
then given by
b+ = R+ ∪ {α + nδ | α ∈ R, n > 0} ∪ {nδ|n > 0}.
R

Denote by θ the highest root of R. Then the set of simple affine roots is given by
b+ .
b = Π ∪ {−θ + δ} ⊂ R
Π

## We denote by W c ⊂ Gl(bh∗ ) the affine Weyl group of the root system R

b and identify
the finite Weyl group W with the subgroup of W c generated by the reflections sα :
∗ ∗
h → b
b
P to finite roots α ∈ R. Then W stabilizes the subspace
h corresponding
∗ ∗ 1
h ⊂ h . Let ρ := 2 α∈R+ α be the half sum of positive finite roots. We then set
b

ρ := ρ + h∨ Λ◦

44
where h∨ is the dual Coxeter number of g. Then hρ, α∨ i = 6 0 for all α ∈ Rbre and
hρ, Ki =
6 0 as well. Recall the ρ-shifted dot-action of W h∗ from chapter 1 defined
c on b
by
w · λ := w(λ + ρ) − ρ
where w ∈ W h∗ .
c and λ ∈ b
The Killing form κ on the simple Lie algebra g extends to a bilinear form (·|·) :
g×b
b g → C which is given by the following equations

## (K|g ⊗C C[t, t−1 ] ⊕ CK) = {0},

(D|g ⊗C C[t, t−1 ] ⊕ CD) = {0},
(K|d) = 1,

for x, y ∈ g, m, n ∈ Z. As in chapter 1 it induces an isomorphism ν : b
h→b h∗ which

coincides with the isomorphism h → h∗ induced by the Killing form when restricted
to the finite Cartan subalgebra, and which sends K to δ and D to Λ◦ . Thus, the
induced form on b h∗ is given by

(α|β) = κ(α, β)

## (Λ◦ |h∗ ⊕ CΛ◦ ) = {0},

(δ|h∗ ⊕ Cδ) = {0},
(Λ◦ |δ) = 1,
for α, β ∈ h∗ and κ the induced Killing form on h∗ . Note that this definition of the
bilinear form only coincides with the one from chapter 1 up to a non-zero scalar.

## 5.2 Restricted representations

Denote by Sb the symmetric algebra S(b h) over the affine Cartan subalgebra and write
S := S(h) for the symmetric algebra over the finite Cartan subalgebra. As in chapter
2 let A be a deformation algebra of the affine Kac-Moody algebra b g. If not otherwise
stated, we will always assume for the rest of this chapter that the structure map
τ : Sb → A of a deformation algebra A factors through the natural surjection Sb  S
induced by the embedding h∗ ⊂ b h∗ from above.
The deformed category OA defined in chapter 2 is always considered to be associated
to the affine data (b
g, b
b, b
h).

45
5.2.1 Integral affine roots
Let M ∈ OA and λ ∈ b h∗ . The central element K ∈ bg acts on the weight space Mλ by
λ(K) ∈ C. We denote by Mk the eigenspace of the action of K on M with eigenvalue
k ∈ C. If M = Mk we call M a module of level k. The category OA decomposes L into
the direct sum of subcategories of modules with equal level, i.e., OA = k∈C OA,k ,
where OA,k consists of those modules which are of level k.
For A a local deformation algebra with canonical weight τ : b hA → A and residue

h∗A × b
field K, we can extend the form (·|·) on h to an A-bilinear form (·|·)A : b
b h∗A → A.
Note that for the imaginary root δ ∈ R b we have (δ|δ) = 0. For λ ∈ bh∗ recall the set
of integral roots (with respect to λ and A) from chapter 2 which becomes

bA (λ) := {α ∈ R
R b | 2(λ + ρ + τ |α)K ∈ Z(α|α)K }

## in the new notation. The corresponding integral Weyl group is then

W
cA (λ) := hsα | α ∈ R bre i ⊂ W.
bA (λ) ∩ R c

## For an equivalence class Λ ∈ b h∗ / ∼A with λ ∈ Λ, we use the notation R bA (Λ) =

R
bA (λ) as before and omit the subscripts in this definition if A = C (or A = S, e
since RSe(Λ) = RC (Λ)). Furthermore, for λ, µ ∈ Λ, we have λ(K) = µ(K) and can
b b
therefore define the complex number λ(K) to be the level of Λ in a well-defined way.
We call the special value crit := −ρ(K) the critical level.

## h∗ / ∼A the following are equivalent:

Lemma 5.2.1 ([AF12a], Lemma 4.2). For Λ ∈ b

## (ii) We have λ + δ ∈ Λ for all λ ∈ Λ.

(iii) We have nδ ∈ R
bA (Λ) for all n 6= 0.

## 5.2.2 Restricted category O

We recall the action of the Feigin-Frenkel center on OA as explained in [AF12a] and
[AF12b]. Let A be a local deformation algebra with residue field K. In chapter 3.1
of [AF12b] the authors introduce an equivalence

T : OA −→ OA
defined by T := L(δ) ⊗C · where L(δ) is the one dimensional simple b
g-module with
−1
highest weight δ. The inverse functor of T is given by T = L(−δ) ⊗C ·. Denote by
T n the n-fold composition of T .

Lemma 5.2.2 ([AF12b], chapter 3.1). A block OA,Λ with Λ ∈ bh∗ / ∼A is preserved
by the functor T if and only if Λ is of level k(Λ) = crit.

46
Let V crit (g) be the universal affine vertex algebra of g at the critical level and denote
by z its center. On any M ∈ OA,crit the center z induces the structure of a module
over the polynomial ring (of infinite rank)
M
Z := Zn = C[p(i)
s , i = 1, ..., rkg, s ∈ Z].
n∈Z

## Any z ∈ Zn for n ∈ Z acts on OA as a natural transformation from T n to the identity

functor on OA . Hence, for any M ∈ OA we get a homomorphism z M : T n M → M .
This action respects base change, i.e., for A → A0 a homomorphism of deformation
algebras, the base change functor · ⊗A A0 : OA,crit → OA0 ,crit induces a natural
isomorphism z M ⊗A A ∼
0
= z M ⊗A idA0 .
Definition 5.2.3. Let M ∈ OA,crit . We call M restricted if for all n 6= 0 and all
z ∈ Zn , the homomorphism z M : T n M → M is zero. Then the restricted deformed
category OA,crit is the full subcategory of OA,crit of all restricted modules.
J J
h∗ open and bounded set OA,crit := OA,crit ∩ OA
For J ⊂ b . As in the non-restricted
J
case we get a truncation functor (·)J : OA,crit → OA,crit . Denote by Zn M the sub-
module of M generated by the subset {z M (m) | m ∈ T n M, z ∈ Zn } ⊂ M . The
functor (·)res : OA,crit → OA,crit given by
X
M 7→ M res := M/ Zn M
n∈Z\{0}

is well defined and left adjoint to the inclusion functor OA,crit ⊂ OA,crit .

## h∗ be of critical level. The restricted Verma module is

Definition 5.2.4. Let λ ∈ b
defined by
∆A (λ) := ∆A (λ)res .
As in the non-restricted case, we omit the subscript of the restricted Verma modules
if the deformation algebra is C ∼ = S/Sh and write ∆(λ) instead of ∆C (λ). We collect
some results which can be found in [AF12b] and which we need for the calculation
of the center of a critical restricted block.
Lemma 5.2.5. Restricted Verma modules respect base change, i.e., for any homo-
morphism A → A0 of deformation algebras we have ∆A (λ) ⊗A A0 ∼
= ∆A0 (λ).
Proof. The claim is obvious for non-restricted Verma modules. Since the action of
the center respects base change it is also true for restricted Verma modules.
h∗ be critical,
Lemma 5.2.6 ([AF12b], Lemma 3.8, Lemma 3.3, Lemma 3.4). Let λ ∈ b
J ⊂bh∗ open and bounded, A → A0 a homomorphism of local deformation algebras
and M ∈ OA . Then
h∗ the weight space ∆A (λ)µ is a free A-module with
(i) for any µ ∈ b

## rkA ∆A (λ)µ = dimK ∆K (λ)µ ,

47
(ii) (M res )J ∼
= (M J )res ,
(iii) the canonical map
(M ⊗A A0 )res → M res ⊗A A0
is an isomorphism.

Remark 5.2.7. Note that in [AF12b], Lemma 3.4, the isomorphism in part (iii) of
the above lemma is formulated by

(M ⊗A A0 )res → (M res ⊗A A0 )res .

But since M res ⊗A A0 is an object of OA0 already, we do not have to restrict again.

## 5.2.3 Restricted projective objects

We recall the construction of projective covers from [AF12b]. Let A be a local
deformation algebra with residue field K, J ⊂ b h∗ an open, bounded subset and let
λ ∈ J be of critical level. Recall the projective cover PAJ (λ)  LA (λ) from chapter
J
2. Define P A (λ) := PAJ (λ)res . We say a module M ∈ OA has a restricted Verma flag
if it has a finite filtration with subquotients isomorphic to restricted Verma modules.
As in the non-restricted case, we denote by (M : ∆A (λ)) the multiplicity of ∆A (λ)
as a subquotient in a restricted Verma flag of M .

Theorem 5.2.8 ([AF12b], Theorem 4.9, and [Fie12a], Theorem 4.3). Let A be a
local deformation algebra with residue field K, J an open and bounded subset of bh∗
J
and let µ ∈ J be of critical level. Then P A (µ) admits a restricted Verma flag and
we have (
J [∆K (λ) : LK (µ)], if λ ∈ J
(P A (µ) : ∆A (λ)) =
0, otherwise
J J
h∗ . Furthermore, P A (µ) is a projective cover of LA (µ) in OA .
for all λ ∈ b

case, we have

## Lemma 5.2.9. Let A → A0 be a homomorphism of local deformation algebras. Let

J
M, P ∈ OA and P be a finitely generated projective object. Then the canonical map

HomOA (P , M ) ⊗A A0 −→ HomOA0 (P ⊗A A0 , M ⊗A A0 )

is an isomorphism of A0 -modules.
J
Proof. By Theorem 5.2.8 we find a finitely generated projective object P ∈ OA , s.t.
res ∼ res J J
P = P . Since (·) is left adjoint to the inclusion functor OA ,→ OA and since
M ⊗A A0 is an object of OA0 , we get

HomOA (P res , M ) ⊗A A0 −→ HomOA (P, M ) ⊗A A0

48
and

HomOA0 ((P ⊗A A0 )res , M ⊗A A0 ) −→ HomOA0 (P ⊗A A0 , M ⊗A A0 )
By Lemma 5.2.6, we have (P ⊗A A0 )res ∼
= P res ⊗A A0 . Thus, the claim follows from
Lemma 2.2.3.
Remark 5.2.10. The Feigin-Frenkel conjecture claims that for λ, µ ∈ J the multi-
J
plicities (P A (µ) : ∆A (λ)) are given by certain periodic Kazhdan-Lusztig polynomials
evaluated at one. These values only depend on the relative position between λ and
µ, and if λ and µ are ‘far away’ from each other these polynomials are zero. Thus, if
J
the Feigin-Frenkel conjecture was true, the projective cover P A (µ) would stabilize
J
for J big enough and P A (µ) would also be a projective cover of LA (µ) in the bigger
non-truncated category OA for each big enough J .

## 5.2.4 The restricted block decomposition

We let bh∗crit := {λ ∈ bh∗ | λ(K) = crit} be the critical hyperplane. We want to define
an equivalence relation on b h∗crit . We write µ A λ if there exists an open and bounded
J
subset J ⊂ b h∗ such that LA (µ) appears as a subquotient of P A (λ). This defines a
partial order on b h∗ and we denote by ∼res b∗
A the equivalence relation on hcrit that is
res
generated by this order. We call the equivalence classes of ∼A restricted A-blocks
or just restricted blocks. For an equivalence class Λ ∈ b h∗crit / ∼res
A let O A,Λ ⊂ O A,crit
be the full subcategory of objects M , such that [M : LA (λ)] 6= 0 implies λ ∈ Λ.
Theorem 5.2.11 ([AF12b], Theorem 5.2). The functor
Q
OA,Λ −→ OA,crit
h∗crit /∼res
Λ∈b A L
(MΛ ) 7−→ MΛ
is an equivalence of categories.
We want to recall a more detailed description of the restricted critical blocks. Denote
h∗ → h∗ , λ 7→ λ the projection with respect to the decomposition b
by · : b h = h⊕
CD ⊕ CK and denote by Λ ⊂ h the image of a subset Λ ⊂ h . For Λ ∈ h∗crit / ∼res
∗ b∗ b
A
and λ ∈ Λ we define the finite integral root system (with respect to Λ and A) by
RA (λ) := {α ∈ R | 2(λ + ρ + τ |α)K ∈ Z(α|α)K }
and the finite integral Weyl group by
WA (λ) := hsα | α ∈ RA (Λ)i ⊂ W.
As usual we can write RA (Λ) = RA (λ) and WA (Λ) = WA (λ) and omit the subscripts
in case A = C or A = S.
e

h∗crit / ∼res
Lemma 5.2.12 ([AF12b], Lemma 5.3). Let Λ ∈ b A be a critical restricted
equivalence class. Then, for all λ ∈ Λ
Λ = WA (Λ) · λ.

49
5.2.5 The generic and subgeneric cases
For the calculation of the center we will need the description of the generic and
subgeneric equivalence classes (cf. [AF12a]). In the rest of this paper, Se will as usual
be the localization of S = S(h) at the maximal ideal generated by the finite Cartan
subalgebra h and if p ⊂ Se is a prime ideal, we denote by Sp the localization of Se at
p. Note that S is defined over the finite Cartan. We will work with the symmetric
algebra over the affine Cartan in the last section of this chapter.
Now let A be such a localization of S. e We need some more notation: Let Λ ∈
∗ res
hcrit / ∼A . For a root α ∈ RA (Λ) and λ ∈ Λ we define α ↓ λ (resp. α ↑ λ) to be the
b
element in the set {sα · λ, s−α+δ · λ} which is smaller (resp. larger) than or equal to λ.
We then define inductively α ↓n λ := α ↓ (α ↓n−1 λ) and α ↑n λ := α ↑ (α ↑n−1 λ).

h∗crit / ∼res
Definition 5.2.13. Let Λ ∈ b A be a critical restricted equivalence class. We
call Λ

## Similarly to Lemma 2.4.1, we get the following

Lemma 5.2.14 ([AF12b], Lemma 5.5). Let p ⊂ Se be a prime ideal of height one
h∗crit be an equivalence class for ∼res
and let Λ ⊂ b Sp .

(i) If α∨ ∈
/ p for all α ∈ R, then Λ is generic.

(ii) If α∨ ∈ p for some α ∈ R, then RSp (Λ) ⊂ {α, −α} and Λ is either generic or
subgeneric.

We finish this chapter with the main results of [AF12b] which in our setting becomes

h∗crit / ∼res
Theorem 5.2.15 ([AF12b], Theorem 5.6). Let Λ ∈ b b∗
A , λ ∈ Λ and J ⊂ hcrit
an open and bounded subset.

## (i) Suppose that Λ is generic. Then Λ consists of one element and

J
P A (λ) ∼
= ∆A (λ)

if λ ∈ J .

(ii) Suppose that Λ is subgeneric and that Λ = {λ, sα · λ} for some α ∈ R. Then
there is a non-split short exact sequence
J
0 → ∆A (α ↑ λ) → P A (λ) → ∆A (λ) → 0

## if J contains λ and α ↑ λ, and a short exact sequence

0 → LA (α ↓ λ) → ∆A (λ) → LA (λ) → 0.

50
(iii) If Λ is subgeneric with RA (Λ) = {±α} and λ ∈ Λ, we have

Λ = {..., α ↓2 λ, α ↓ λ, λ, α ↑ λ, α ↑2 λ, ...}.

The second short exact sequence in (ii) follows from the first one and the restricted
BGGH-reciprocity.
Remark 5.2.16. Note that the first part of the theorem implies that for λ generic
and A = K a field, the restricted Verma module ∆K (λ) is simple. This fact was
already proven in [Fre05], Theorem 4.8.

## 5.3 The center of OS,Λ

e for Λ critical

The content of this chapter can be found in the preprint [Küb13]. It mainly follows
the ideas of the PhD thesis [Fie01] and its published version [Fie03] and should be
seen as a supplement to the first part of [Fie03].
Recall that the center of a category is the ring of endo-transformations of the identity
functor. Let A be a deformation algebra and fix a critical equivalence class Λ ∈
h∗crit / ∼res
b b∗
A . For an open, bounded subset J ⊂ hcrit denote by ZA (Λ, J ) the center of
J
OA,Λ and by ZA (Λ) the center of OA,Λ . We first consider the case A = S. e Thus, we
have
J
ZSe(Λ, J ) := Z(OS,Λ
e ) = End(idOJ ).
S,Λ
e

We want to describe the center by observing its action on projective objects. Since
in general we only have enough projective objects in the truncated categories, we
have to express the center of OS,Λ
e as a limit of the centers ZA (Λ, J ) which runs
over open and bounded subsets J ⊂ b h∗ . The main result is then
crit

## Theorem 5.3.1. Let Se be the localization of S at the maximal ideal generated by h

h∗crit / ∼res
and Λ ∈ b e . Then we have an isomorphism of rings
S
( )

ZSe(Λ) ∼
Y
= (zµ )µ∈Λ ∈ Se zµ ≡ zα↓µ (mod α∨ ) ∀ α ∈ R(Λ) .

µ∈Λ

We use Lemma 5.2.9 and a localization process to split the problem into generic and
subgeneric situations. Following [Fie03], we first relate the center to the endomor-
phism rings of a generating set of projective objects.
Remark 5.3.2. Let A be a localization of Se at a prime ideal and Γ an equiva-
J
lence class under ∼res
A . The block O A,Γ is generated by the set of projective covers
J
{P A (λ)}λ∈Γ∩J . By the same arguments as given in [Fie03], chapter 3.1, we get that
evaluating on indecomposable projective objects induces an injective map
Y J
ZA (Γ, J ) ,→ EndOA (P A (µ))
µ∈Γ∩J

51
and the image of this map is given by the subset

(

Y J
(zµ )µ∈Γ∩J ∈ EndOA (P A (µ)) zµ ◦ f = f ◦ zλ

µ∈Γ∩J
)
J J
∀ λ, µ ∈ Γ ∩ J , f ∈ HomOA (P A (λ), P A (µ)) .

## Let A → A0 be a homomorphism of local deformation algebras. The base change,

Lemma 5.2.9, for the endomorphism rings of restricted projective objects induces a
map
ZA (Γ, J ) −→ ZA0 (Γ, J )
(cf. the proof of Proposition 3.1 in [Fie03]).
For another open, bounded subset J 0 ⊂ b h∗ with J ⊂ J 0 the restriction of the iden-
J J0
tity functor to the subcategory OA,Γ ⊂ OA,Γ induces a map ZA (Γ, J 0 ) → ZA (Γ, J ).
This defines a directed system. Since for every finitely generated module M of OA,Γ
J
there exists an open and bounded subset such that M lies in OA,Γ , and since the
center is already uniquely defined by its action on the finitely generated objects, we
have
ZA (Γ) ∼
= lim Z (Γ, J ).
←− A
The base change maps ZA (Γ, J ) → ZA0 (Γ, J ) then induce a base change map
ZA (Γ) → ZA0 (Γ) (cf. [Fie03]), by the universal property of inverse limits.

## Lemma 5.3.3. Let A be a localization of Se at a prime ideal. The evaluation on

restricted Verma modules induces an injective map

## EndOA (∆A (µ)) ∼

Y Y
ZA (Γ) ,→ = A
µ∈Γ µ∈Γ

Proof. (cf. [Fie01], proof of Proposition 5.10) Let Q := Q(A) be the quotient field
of A. Then by Theorem 5.2.15 (i) and Remark 5.2.16, all Verma modules ∆Q (λ) are
projective and irreducible and we have HomOQ (∆Q (µ), ∆Q (λ)) = 0 for µ 6= λ. But
by the description of the center from above and by base change for the center we
get a commutative diagram

## EndOA (∆A (µ)) ∼

Q Q
ZA (Γ) −−−→ µ∈Γ = µ∈Γ A
 
 
y y

ZQ (Γ) −−−→ µ∈Γ EndOQ (∆Q (µ)) ∼
Q Q
= µ∈Γ Q

where the lower horizontal is an isomorphism and the verticals are injective. But
then the upper horizontal is injective as well.

52
Remark 5.3.4. We want to describe the image of the map
Y
ZSe(Γ) ,→ S.
e
µ∈Γ

## The strategy to describe this embedding is to localize all appearing modules at

prime ideals p of height one. So let p ⊂ Se be such an ideal. If αQ ∈/ p for all

α ∈ R(Γ) = RSe(Γ), all ∆Sp (µ) are projective and we get ZSp (Γ) → µ∈Γ Sp which
is the generic situation. We will deal with the subgeneric case in the next chapter.

## 5.3.1 The subgeneric case

Let Sα be the localization of Se at the prime ideal generated by α∨ . We fix an
h∗crit under ∼res
equivalence class Γ ⊂ b Sα , which is not generic. Let λ ∈ Γ. Then, Lemma
5.2.14 and Theorem 5.2.15 (iii) imply Γ = W cα · λ = {..., α ↓ λ, λ, α ↑ λ, α ↑2 λ, ...},
where Wcα ⊂ W c is the affine subgroup generated by the reflections sα+nδ with n ∈ Z.
Recall that by Theorem 5.2.15 we have a short exact sequence
J
∆Sα (α ↑ λ) ,→ P Sα (λ)  ∆Sα (λ)
J J 0
if α ↑ λ, λ ∈ J . This implies that for J 0 ⊃ J we have P Sα (λ) ∼
= P Sα (λ) in OSα .
Thus, for any µ ∈ Γ we will always assume that the open, bounded subset Jµ we

are working with is big enough, such that we can write P Sα (µ) = P Sα (µ).

## Lemma 5.3.5. Restriction to the restricted Verma module ∆Sα (α ↑ µ) ,→ P Sα (µ)

for every µ ∈ Γ induces a surjective map
Q Q
µ∈Γ EndOSα (P Sα (µ))  µ∈Γ EndOSα (∆Sα (µ))
(fµ ) 7→ (fµ |∆Sα (α↑µ) ).

Proof. Since EndOSα (∆Sα (µ)) = Sα · id∆Sα (µ) , every endomorphism of the restricted
Verma module lifts to an endomorphism of P Sα (α ↓ µ).
Q Q
Identifying µ∈Γ EndOSα (∆Sα (µ)) with µ∈Γ Sα and using the naturality of the ac-
tion of the center, we get a commutative diagram


ZSα (Γ)   v /
Q

66 µ∈Γ

Q (
µ∈Γ EndOSα (P Sα (µ)).

The aim is now to describe the image of the composition of the down- with the
up-going arrow in this diagram.

53
Proposition 5.3.6. Let λ, µ ∈ Γ, then HomOSα (P Sα (µ), P Sα (λ)) is a free Sα -module
and we have

1 if µ = α ↓ λ or µ = α ↑ λ,

rkSα HomOSα (P Sα (µ), P Sα (λ)) = 2 if µ = λ,

0 otherwise.

## Proof. By our assumption, Γ is subgeneric, i.e., Γ contains exactly two elements.

HomOSα (P Sα (µ), P Sα (λ)) is a finitely generated Sα -module. Let K be the residue
field and Q the quotient field of Sα . By Lemma 5.2.9 we get

## HomOSα (P Sα (µ), P Sα (λ)) ⊗Sα K ∼

= HomOK (P K (µ), P K (λ))

and

## HomOSα (P Sα (µ), P Sα (λ)) ⊗Sα Q ∼

= HomOQ (P Sα (µ) ⊗ Q, P Sα (λ) ⊗ Q).

But by [Fie13], chapter 6.2, the dimension of the right hand side of the first equality
is given by

1 if µ = α ↓ λ or µ = α ↑ λ,

dimK HomOK (P K (µ), P K (λ)) = 2 if µ = λ,

0 otherwise.

## Since we have a decomposition P Sα (µ) ⊗Sα Q ∼

= ∆Q (α ↑ µ) ⊕ ∆Q (µ) and P Sα (λ) ⊗Sα
Q∼= ∆Q (α ↑ λ) ⊕ ∆Q (λ) into simple Verma modules we also get

1 if µ = α ↓ λ or µ = α ↑ λ,

dimQ (HomOSα (P Sα (µ), P Sα (λ)) ⊗Sα Q) = 2 if µ = λ,

0 otherwise.

But then by [AF12b], Lemma 3.7, HomOSα (P Sα (µ), P Sα (λ)) is a free Sα -module with

1 if µ = α ↓ λ or µ = α ↑ λ,

rkSα HomOSα (P Sα (µ), P Sα (λ)) = 2 if µ = λ,

0 otherwise.

## For fixed λ ∈ Γ we want to give bases for the following Sα -modules:

EndOSα (P Sα (λ)),

## HomOSα (P Sα (λ), P Sα (α ↑ λ)),

HomOSα (P Sα (λ), P Sα (α ↓ λ)).

54
We follow the notation of [Fie13]. Clearly, we can take the identity in EndOSα (P Sα (λ))
as the first basis element. Over the residue field K we have a composition

## P K (λ)  ∆K (λ) ,→ ∆K (α ↑ λ).

We can lift this composition via base change to a map P Sα (λ) → ∆Sα (α ↑ λ) which
is unequal to 0. Composing this map with the inclusion ∆Sα (α ↑ λ) ,→ P Sα (λ) yields
an endomorphism nλ : P Sα (λ) → P Sα (λ) which is unequal to 0 and, after applying
· ⊗Sα K, corresponds to the composition

## λ : P K (λ)  ∆K (λ) ,→ ∆K (α ↑ λ) ,→ P K (λ).

nK

Since the identity and nλ are linearly independent, {id, nλ } ⊂ EndOSα (P Sα (λ)) is a
basis.
Taking the map P Sα (λ) → ∆Sα (α ↑ λ) from above and the projectivity of P Sα (λ)
we get a morphism bλ : P Sα (λ) → P Sα (α ↑ λ) as the dotted arrow in the diagram

P Sα (λ) / P Sα (α ↑ λ)

& 
∆Sα (α ↑ λ).
This morphism is unequal to 0 after applying · ⊗Sα K. We conclude that bλ is a basis
of HomOSα (P Sα (λ), P Sα (α ↑ λ)).
Finally, we have the composition

## aλ : P Sα (λ)  ∆Sα (λ) ,→ P Sα (α ↓ λ)

which is unequal to 0 after base change · ⊗Sα K with the residue field. Thus it is a
basis of HomOSα (P Sα (λ), P Sα (α ↓ λ)).
Remark 5.3.7. Since α ↑ λ ≥ λ, the endomorphism nλ : P Sα (λ) → P Sα (λ) restricts
to an endomorphism ∆Sα (α ↑ λ) → ∆Sα (α ↑ λ) and thus induces an endomorphism
on the cokernel ∆Sα (λ) of the inclusion ∆Sα (α ↑ λ) ,→ P Sα (λ).
Proposition 5.3.8. Up to an invertible element of Sα , nλ induces the map α∨ · id
on ∆Sα (α ↑ λ) and the zero map on ∆Sα (λ).
Before we prove this proposition we need a result which we will state in the next
chapter a second time, but with a slightly different proof.
Recall the Shapovalov form for Kac-Moody algebras introduced in chapter 3. It
induces a contravariant form (·, ·)Sα on ∆Sα (µ) for any µ ∈ Γ. As in chapter 3 we
define a filtration on ∆Sα (µ) by setting

## ∆K (µ)i := im(∆Sα (µ)i ,→ ∆Sα (µ)  ∆K (µ))

55
where ∆Sα (µ)  ∆K (µ) is the map induced by · ⊗Sα K. We will recall the Jantzen
filtration on a restricted Verma module in general in the next section. Recall that a
restricted Sα -block is either subgeneric or generic. But we assumed for this section
that µ lies in a subgeneric Sα -block.

## Lemma 5.3.9. The Jantzen filtration on ∆K (µ) is given by

∆K (µ) ⊃ LK (α ↓ µ) ⊃ 0.

Proof. We give a proof which is adapted from [Jan79], chapter 5.14, to our situation.
Since, by construction, ∆K (µ)1 coincides with the maximal submodule of ∆K (µ), we
get, by Theorem 5.2.15, that ∆K (µ)1 ∼ = LK (α ↓ µ). We have to prove ∆K (µ)2 = 0.
Let m ∈ ∆K (µ)1α↓µ be a generator of highest weight α ↓ µ. We just have to prove
m∈ / ∆K (µ)2 . Let us assume m ∈ ∆K (µ)2 . Thus, there is an element m0 ∈ ∆Sα (µ)2
such that m0 7→ m under specialization ∆Sα (µ)  ∆K (µ). Since for ν > α ↓ µ we
have ∆K (µ)1ν = 0, we conclude

## ∆Sα (µ)1ν ⊂ α∨ · ∆Sα (µ)ν

and
∆Sα (µ)2ν ⊂ (α∨ )2 · ∆Sα (µ)ν .
The generalized Casimir operator C can be split into a sum C = C1 +C2 of operators
L the following properties: For γ a weight of ∆(µ), we have C1 (∆Sα (µ)γ ) ⊂
with
η>γ ∆Sα (µ)η , C2 (∆Sα (µ)γ ) ⊂ ∆Sα (µ)γ and C2 acts on ∆Sα (µ)γ by multiplication
with (γ + τ + ρ|γ + τ + ρ)Sα − (ρ|ρ)Sα ∈ Sα . Since C commutes with the b
g-action, we
get that C acts on ∆Sα (µ) by multiplication with (µ + τ + ρ|µ + τ + ρ)Sα − (ρ|ρ)Sα
where τ is the canonical weight associated to Sα . Thus, applying C to m0 yields

## Cm0 ∈ ((α ↓ µ + τ + ρ|α ↓ µ + τ + ρ)Sα − (ρ|ρ)Sα )m0 + (α∨ )2 · ∆Sα (µ).

Therefore, by W-invariance
c of (·|·)Sα and a little calculation we get

## Since µ − α ↓ µ is either equal to nα or n(−α + δ) for n > 0 an integer and since

(δ|τ )Sα = 0, we get (µ − α ↓ µ|τ )Sα = kα∨ with k ∈ C\{0}. But then, specializing
α∨ 7→ 0 yields km = 0 which is a contradiction. Thus m ∈/ ∆K (µ)2 .
Proof of Proposition 5.3.8. Over the residue field K, we get the following diagram
of short exact sequences in the horizontals

56

∆K (α ↑ λ)  / P K (λ) // ∆K (λ)


∆K (λ)
_
y·id x·id

∆K (α _ ↑ λ)

  

∆K (α ↑ λ)  / P K (λ) // ∆K (λ)
where the composition nK λ in the middle is induced by nλ : P Sα (λ) → P Sα (λ) and
x, y ∈ K are the scalars induced by the induced maps on the restricted Verma
modules. Since the composition ∆K (α ↑ λ) ,→ P K (λ)  ∆K (λ) is zero, we get that
both scalars, x and y, are zero.
Over Sα the composition

## P Sα (λ) → ∆Sα (α ↑ λ) ,→ P Sα (λ)  ∆Sα (λ)

is zero, so nλ induces the zero map on ∆Sα (λ). But the composition ∆Sα (α ↑ λ) ,→
P Sα (λ) → ∆Sα (α ↑ λ) is unequal to zero since otherwise we would have a factoriza-
tion over the cokernel, in formulas

∆Sα (α ↑ λ)  / P Sα (λ) / ∆Sα (αO ↑ λ)
∃!
&&
∆Sα (λ).

But ∆Sα (λ) → ∆Sα (α ↑ λ) is the zero map, because it is zero after applying · ⊗Sα Q,
while P Sα (λ) → ∆Sα (α ↑ λ) is unequal to zero. Now, after multiplying nλ with an
appropriate invertible element of Sα , nλ induces (α∨ )n · id on ∆Sα (α ↑ λ). We want
to show n = 1.
Assume n ≥ 2. Then im(nλ ) ⊂ ∆Sα (α ↑ λ)2 ⊂ P Sα (λ) by definition of the Jantzen
2
filtration. But after base change · ⊗Sα K, we have im(nK λ ) ⊂ ∆Sα (α ↑ λ) ⊗Sα K.
2
Since ∆Sα (α ↑ λ) ⊗Sα K is the second step of the Jantzen filtration on ∆K (α ↑ λ)
which is zero by Lemma 5.3.9, this contradicts nK λ 6= 0 and we conclude n = 1.

Let us fix a short exact sequence ∆Sα (α ↑ λ) ,→ P Sα (λ)  ∆Sα (λ) and let nλ :
P Sα (λ) → P Sα (λ) be the map from above, normalized, s.t. it induces α∨ · id on
∆Sα (α ↑ λ) and 0 on ∆Sα (λ).
Lemma 5.3.10. (i) We have nλ ◦ aα↑λ = aα↑λ ◦ (α∨ · id − nα↑λ ).

## (ii) We have (α∨ · id − nα↑λ ) ◦ bλ = bλ ◦ nλ .

Proof. (i) This part of the lemma is clear by definition of the map nλ and the
effect it has on restricted Verma modules.

57
(ii) Applying · ⊗Sα Q, identifies

P Sα (λ) / P Sα (α ↑ λ)
nλ α∨ ·id−nα↑λ
 
P Sα (λ) / P Sα (α ↑ λ)

## with the diagram

∆Q (λ) ⊕ ∆Q (α ↑ λ) / ∆Q (α ↑2 λ) ⊕ ∆Q (α ↑ λ)
g1 g2
 
∆Q (λ) ⊕ ∆Q (α ↑ λ) / ∆Q (α ↑2 λ) ⊕ ∆Q (α ↑ λ)

## where g1 , g2 are both given by the matrix

 
0 0
.
0 (α∨ ·)
But this diagram certainly commutes. Then the diagram above over Sα com-
mutes as well.

By the description of the center in Remark 5.3.2, and by what we have discovered
above, the image of the inclusion
Y
ZSα (Γ) ,→ EndOSα (P Sα (λ))
λ∈Γ

is generated by the tuples (idP Sα (µ) )µ∈Γ and the set of tuples {(δλµ )µ∈Γ }λ∈Γ where

nλ if µ = λ,

µ
δλ = α∨ · id − nα↑λ if µ = α ↑ λ,

0 otherwise.

## EndOSα (∆Sα (µ)) ∼

Y Y Y
φ: EndOSα (P Sα (µ)) → = Sα
µ∈Γ µ∈Γ µ∈Γ

are φ((idP Sα (µ) )µ∈Γ ) = (1)µ∈Γ and φ((δλµ )µ∈Γ ) = (κµλ )µ∈Γ with
(
µ α∨ if µ = α ↑ λ
κλ = . As a conclusion we have
0 else
Proposition 5.3.11.
( )

ZSα (Γ) ∼
Y
= (zµ )µ∈Γ ∈ Sα zµ ≡ zα↑µ (mod α∨ ) .

µ∈Γ

58
5.3.2 The general case
In this chapter we want to collect our results in the subgeneric and generic cases to
prove the main theorem. Let Λ be an equivalence class under ∼res e .
S

Theorem 5.3.12.
( )
Y
ZSe(Λ) ∼
= (zµ )µ∈Λ ∈ Se zµ ≡ zα↓µ (mod α∨ ) ∀α ∈ R(Λ) .

µ∈Λ

## Proof. For p ⊂ Se a prime ideal of height one and for Q = Quot(S),

e we have a base
change map
ZSe(Λ) ,→ ZSe(Λ) ⊗Se Sp ⊂ ZQ (Λ) ∼
Y
= Q.
µ∈Λ
S
Note that Λ splits into a disjoint union Λi of generic and subgeneric Sp -blocks and
we get a decomposition

ZSe(Λ) ⊗Se Sp ∼
Y
= ZSp (Λi ).
i

We also have \Y
ZSe(Λ) = ZSp (Λi )
p∈P i
Q
in the Q-vector space µ∈Λ Q (cf. the proof of Proposition 5.11 in [Fie01]). Here, P
denotes the set of all prime ideals of Se of height one. If α∨ ∈
/ p for all α ∈ R(Λ), then

Q restricted deformed Verma modules are projective and we have ZSe(Λ) ⊗Se Sp =
all
µ∈Λ Sp .
If p is generated by α∨ for α ∈ R(Λ), ZSe(Λ) ⊗Se Sp decomposes into the product
of
Q modules of the form described in Proposition 5.3.11 and in modules of the form
µ∈Λi Sp , if Λi is generic. Putting these together in the above intersection proves
the claim.

## 5.4 Restricted Jantzen Sum Formula

In [AJS94], chapter 6, the authors establish a Jantzen sum formula for baby Verma
modules over a small quantum group. It relates the sum of the characters of the
Jantzen filtration to an alternating sum of characters of baby Verma modules with
smaller highest weights. We deduce a similar formula for the restricted Verma mod-
ules at the critical level. We state the main result of this section in

## h∗crit . There is a filtration

Theorem 5.4.1. Let λ ∈ b

## with the properties

59
(i) ∆(λ)1 is the maximal submodule of ∆(λ)

(ii) !
X X X
ch∆(λ)i = ch∆(α ↓2i−1 λ) − ch∆(α ↓2i λ) .

i>0 α∈R(λ)+ i>0

Note that the sum is taken over all finite, positive, integral roots α ∈ R(λ)+ .

For this chapter we need an alternative definition of restricted Verma modules. Let
λ∈b h∗ and τ : Sb → A be a deformation algebra which is an integral domain. Denote
by Q(A) its field of fractions and assume that both structure maps factor through the
restriction map Sb → S. This implies that τ (c) = τ (d) = 0. We define ∆− A (λ+τ ) to be
the submodule of ∆A (λ + τ ) which is generated by the images of all homomorphisms
∆A (λ − nδ + τ ) → ∆A (λ + τ ) for n ∈ N>0 . Since τ (c) = τ (d) = 0, we have (τ |δ) = 0
and by Theorem 2.3.1 there is an injective map ∆Q(A) (λ−nδ +τ ) ,→ ∆Q(A) (λ+τ ) for
every n > 0 and λ critical. But by our assumption on A, this also induces an injective
map ∆A (λ − nδ + τ ) ,→ ∆A (λ + τ ). If λ is non-critical, we get ∆− A (λ + τ ) = {0}.
The restricted Verma module can now be described as the quotient

∆A (λ + τ ) = ∆A (λ + τ )/∆−
A (λ + τ ).

A result by Frenkel and Gaitsgory (cf. [FG06]) shows that for any critical weight
λ∈b h∗ and n < 0 there is a surjective map Zn → Hombg (∆(λ + nδ), ∆(λ)). Thus, the
old definition of restricted Verma modules in the preceding section coincides with
this one.

## 5.4.1 The Shapovalov determinant

Recall the Shapovalov form F : U (b g) × U (b
g) → S(bh) with F (x, y) = β(σ(x), y)
from chapter 3. Recall that for u, x, y ∈ U (g) we have F (σ(u)x, y) = F (x, uy). For
b
η ∈ NR b+ we denote by Fη the restriction of F to the weight space U (b n)−η . Recall
∼ b∗
the isomorphism ν : h → h induced by the bilinear form (·|·) on h and define
b b
hα := ν −1 (α) for any root α ∈ R.
b

## Theorem 5.4.2 ([KK79], Theorem 1). The determinant of

n)−η × U (b
Fη : U (b n)−η → S(b
h)

## is, up to multiplication with a non-zero complex number, given by the formula

∞  mult(α)·P(η−nα)
Y Y (α|α)
detFη = hα + ρ(hα ) − n
2
b+ n=1
α∈R

## where P is Kostant’s partition function and mult(α) := dimC (b

gα ).

60
We equip the polynomial ring C[t] in one variable with two different structures of
a deformation algebra. The first one is given by the map τ1 : Sb  C[t], where τ1
is induced by the inclusion of the line Cρ ⊂ b h∗ . The second S-module
b structure
∗ ∗
τ2 : S  C[t] is given by the inclusion Cρ ⊂ h . Recall that ρ ∈ h which implies
b b
that τ2 factors through the restriction map Sb  S while τ1 does not factor through
this map in general. For a more intuitive notation, we follow [Jan79] and define
∆C[t] (λ + tρ) := ∆C[t] (λ + τ1 ) and ∆C[t] (λ + tρ) := ∆C[t] (λ + τ2 ).
Note that for λ ∈ bh∗crit critical and since τ2 (c) = τ2 (d) = tρ(c) = 0, we can construct
the restricted Verma module ∆C[t] (λ + tρ). Let C(t) be the quotient field of C[t].

## h∗crit . Then ∆C(t) (λ + tρ) = ∆C[t] (λ + tρ) ⊗C[t] C(t) is simple.

Lemma 5.4.3. Let λ ∈ b

Proof. If we prove that RC(t) (λ) = ∅, the lemma follows from Theorem 5.2.15 and
Remark 5.2.16. But since (ρ|α) 6= 0 for all α ∈ R+ we get 2(λ + ρ + tρ|α)C(t) ∈/
Z(α|α)C(t) ⊂ C(t) for all α ∈ R+ .
The Shapovalov form induces symmetric, contravariant bilinear forms on ∆Sb(λ + 0 )
and ∆S (λ + ) where we denote by 0 ∈ b h∗Sb the canonical weight induced by b
h ,→ Sb
and by  ∈ b h∗S its composition with Sb  S. Moreover, it induces contravariant
forms on all Verma modules ∆C[t] (λ + tρ), ∆C[t] (λ + tρ), ∆C[t] (λ + tρ), ∆(λ) and
∆(λ) we have to deal with in the rest of this paper. The contravariance of the forms
implies for ∆(λ) and ∆(λ) that the radicals of the forms coincide with the maximal
submodules of ∆(λ) and ∆(λ).

## 5.4.2 Restricted Jantzen filtration

Denote by (·, ·) the contravariant form on ∆C[t] (λ + tρ) induced by the Shapovalov
form. As in the previous section we define a filtration on ∆C[t] (λ + tρ) by

## ∆(λ)i := im(∆C[t] (λ + tρ)i ,→ ∆C[t] (λ + tρ)  ∆(λ))

where the second map is specialization t 7→ 0. In the same way, we get the Jantzen
filtration on ∆(λ) as it is defined in chapter 3 using the deformed Verma module
∆C[t] (λ + tρ).

## Notation 4. Let µ ≤ λ. We denote the determinants of the contravariant bilinear

forms on the µ-weight spaces ∆C[t] (λ + tρ)µ , ∆C[t] (λ + tρ)µ and ∆C[t] (λ + tρ)µ by
Dλ+tρ (µ + tρ), Dλ+tρ (µ + tρ) and Dλ+tρ (µ + tρ).

## For a polynomial P ∈ C[t] denote by ordt (P ) the integer n ∈ N with tn | P but

tn+1 - P .

61
Lemma 5.4.4 ([Jan79], Lemma 5.1). For the Jantzen filtrations of the µ-weight
spaces of the non-restricted and restricted Verma modules we have the formulas
X
dimC ∆(λ)iµ = ordt (Dλ+tρ (µ + tρ))
i>0

and X
dimC ∆(λ)iµ = ordt (Dλ+tρ (µ + tρ)).
i>0

First, we want to recall the Jantzen filtration of a restricted Verma module with a
highest weight, which has critical level and is subgeneric. The next proposition is
Lemma 5.3.9. Here, however, we call it a proposition since it has a higher impact
on the main result of this section. We give an alternative proof which only works in
type A.
Proposition 5.4.5. Let λ ∈ b h∗crit be subgeneric, i.e., R(λ) = {±α} for a finite
+
positive root α ∈ R and α ↓ λ 6= λ. Then the Jantzen filtration of ∆(λ) is
∆(λ) ⊃ L(α ↓ λ) ⊃ 0
and we have the alternating sum formula
X
ch∆(λ)i = chL(α ↓ λ) = ch∆(α ↓ λ) − ch∆(α ↓2 λ) + ch∆(α ↓3 λ) − ...
i>0

Proof. The statement of the first part is identical with Lemma 5.3.9 and is proved
there already. However, we give an alternative proof in case g is of type A.
From Theorem 5.2.15 we see that L(α ↓ λ) is the maximal submodule of ∆(λ). By
the definition of the Jantzen filtration we conclude that ∆(λ)1 coincides with the
radical of the contravariant form on ∆(λ) induced by the Shapovalov form. Thus,
∆(λ)1 ∼= L(α ↓ λ).
We are left to prove ∆(λ)2 = {0}. For β ∈ R b+ and n ∈ N set µβ,n := λ − α ↓
λ − nβ. Then, by Theorem 5.4.2, the polynomial Dλ+tρ (α ↓ λ + tρ) ∈ C[t] is, up to
multiplication with a non-zero complex number, given by the product
∞  mult(β)·P(µβ,n )
Y Y (β|β)
(α ↓ λ)(hβ ) + tρ(hβ ) + ρ(hβ ) − n
2
+
β∈R
b n=1

## and similarly Dλ+tρ (α ↓ λ + tρ) ∈ C[t] is given by

∞  mult(β)·P(µβ,n )
Y Y (β|β)
(α ↓ λ)(hβ ) + tρ(hβ ) + ρ(hβ ) − n .
2
+
β∈R
b n=1

## b+ if and only if β = mδ for m ∈ N\{0}. But for

Note that ρ(hβ ) = 0 for β ∈ R
β = mδ with m ∈ N\{0} we have P(λ − α ↓ λ − nβ) = 0 (here we use that g is of
type A). Now we conclude
ordt (Dλ+tρ (α ↓ λ + tρ)) = ordt (Dλ+tρ (α ↓ λ + tρ)).

62
But since α ↓ λ does not appear as a weight in the submodule ∆(λ)− ⊂ ∆(λ), we
have
Dλ+tρ (α ↓ λ + tρ) = Dλ+tρ (α ↓ λ + tρ).
Thus, by Lemma 5.4.4, we have
X X
dimC ∆(λ)iα↓λ = dimC ∆(λ)iα↓λ .
i>0 i>0

## From the non-restricted Jantzen filtration of [KK79], one knows

Pthat the character
of the simple module L(α ↓ λ) only appears once in the sum i>0 ch∆(λ)i . Since
1
P weight of i∆(λ) the character 2of L(α ↓ λ) can also only appear
α ↓ λ is a maximal
once in the sum i>0 ch∆(λ) which implies ∆(λ) 6= L(α ↓ λ) and thus ∆(λ)2 = 0.
The second part of the proposition follows inductively from Theorem 5.2.15 since
chL(α ↓ λ) = ch∆(α ↓ λ) − chL(α ↓2 λ).

hS → S induced by b
Recall the canonical weight  : b h ⊂ Sb  S and for ν ≥ 0 denote
by D−ρ ( − ρ − ν) the determinant of the contravariant form on ∆S ( − ρ)−ρ−ν . Let
φ : S  C[t] be the algebra homomorphism given by φ(H) := (λ + ρ)(H) + tρ(H)
for all H ∈ h. If p ∈ S is a prime element and a ∈ S, we denote by ordp (a) the
integer n ∈ N such that pn | a but pn+1 - a. By [Jan79], chapter 5.6, we get for a ∈ S
X
ordt (φ(a)) = ordp (a)ordt (φ(p)) (5.1)
p

where p runs over all classes of associated prime elements of S. As in Lemma 5.4.3
we see that, for the quotient field Q = Q(S) of S, the restricted Verma module
∆Q ( − ρ) ∼
= ∆S ( − ρ) ⊗S Q is simple. We conclude that D−ρ ( − ρ − ν) 6= 0 and
also φ(D−ρ ( − ρ − ν)) 6= 0 for all ν ∈ NRb+ . Combining equation (5.1) with Lemma
5.4.4, we get
X X X
ch∆(λ)n = eλ ordt (φ(p)) ordp (D−ρ ( − ρ − ν))e−ν . (5.2)
n>0 p b+
ν∈NR

We are now able to prove the general case. We follow [Jan79], chapter 5.7.

## Proof of Theorem 5.4.1. If λ ∈ b h∗crit fulfills hλ+ρ, α∨ i ∈

/ Z\{0} for any finite positive
+
root α ∈ R , then λ is a generic weight and ∆(λ) is simple, by Theorem 5.2.15 and
Remark 5.2.16. The evaluation of the polynomial D−ρ ( − ρ − ν) ∈ S at λ + ρ
for ν ∈ NR b+ can be viewed as the determinant of the contravariant form on the
weight space ∆(λ)λ−ν induced by the Shapovalov form. Since the weight spaces are
orthogonal to each other according to the contravariant form, D−ρ (−ρ−ν)(λ+ρ) is
unequal to zero for all ν ∈ NR b+ . Otherwise we could construct a proper submodule
of ∆(λ), which would be a contradiction. D−ρ ( − ρ − ν) decomposes into a product
of linear factors and it follows that all prime divisors of D−ρ ( − ρ − ν) are of the

63
form α∨ − r where α ∈ R+ and r ∈ Z\{0}.
[
For α ∈ R+ and r ∈ Z we define να,r ∈ Z[bh∗ ] by
X
να,r = ordα∨ −r (D−ρ ( − ρ − η))e−η .
b+
η∈NR

Since hρ, α∨ i =
6 0 for any α ∈ R+ , the restriction of α∨ −r to the curve (λ+ρ)+Cρ ⊂
h∗ is unequal to zero, i.e., in formulas we have
b

φ(α∨ − r) = hλ + ρ, α∨ i − r + thρ, α∨ i =
6 0.

## If r 6= hλ + ρ, α∨ i, then ordt (hλ + ρ, α∨ i − r + thρ, α∨ i) = 0. But for r = hλ + ρ, α∨ i

we have ordt (hλ + ρ, α∨ i − r + thρ, α∨ i) = 1. Now α∨ − hλ + ρ, α∨ i can only be a
prime divisor of D−ρ ( − ρ − η) if α ∈ R(λ)+ . Applying formula (5.2) we conclude
X X
ch∆(λ)i = να,hλ+ρ,α∨ i eλ . (5.3)
i>0 α∈R(λ)+

Let α ∈ R(λ)+ . Perturbing the weight λ in the hyperplane that contains λ and is par-
h∗crit such
allel to the reflection hyperplane corresponding to α, we find a weight µ ∈ b
∨ ∨ ∨ +
that hµ + ρ, α i = n = hλ + ρ, α i and hµ + ρ, β i ∈
/ Z for all β ∈ R(λ) \{α}. Thus,
we found a subgeneric weight µ with R(µ) = {±α} and να,hµ+ρ,α∨ i = να,hλ+ρ,α∨ i . But
by Proposition 5.4.5, the Jantzen filtration of ∆(µ) is given by

∆(µ) ⊃ L(α ↓ µ) ⊃ 0.

## We conclude by equation (5.3):

X
να,n eµ = chL(α ↓ µ) = ch∆(α ↓2i−1 µ) − ch∆(α ↓2i µ) .

i>0

Now the choice of µ implies that eλ−µ ch ∆(α ↓n µ) = ch ∆(α ↓n λ). Thus, we
conclude X
να,n eλ = ch∆(α ↓2i−1 λ) − ch∆(α ↓2i λ) .

i>0

Since we can apply this to any root β ∈ R(λ)+ we can use equation (5.3) once more
to get the formula in Theorem 5.4.1.
As a consequence of Theorem 5.4.1 we get the linkage principle for restricted Verma
modules at the critical level in the same way as in [AJS94], chapter 6, or [KL97],
Theorem 10.3. The linkage principle was already proven in [AF12b] by using re-
stricted projective objects of the restricted category O over the Lie algebra b
g. Our
proof, however, avoids the rather complicated construction of restricted projective
objects.
h∗crit and µ ∈ b
Corollary 5.4.6 ([AF12b], Theorem 5.1). Let λ ∈ b h∗ . Then [∆(λ) :
L(µ)] 6= 0 implies µ ∈ W(λ)
c · λ and µ ≤ λ.

64
Proof. The statement is obvious for λ = µ and it is also clear that [∆(λ) : L(µ)] 6= 0
implies µ ≤ λ. We use induction on λ − µ and assume µ < λ. If [∆(λ) : L(µ)] 6=
0, then also [∆(λ)1 : L(µ)] 6= 0 since ∆(λ)1 ⊂ ∆(λ) is the maximal submodule.
But then the restricted Jantzen sum formula implies that L(µ) has to appear as
a subquotient in some ∆(α ↓n λ) where α ∈ R(λ)+ and n > 0. Our induction
c ↓n λ) · (α ↓n λ) and the definition of α ↓ λ yields
assumption implies µ ∈ W(α
c ↓n λ) · (α ↓n λ) = W(λ)
W(α c · λ.
Remark 5.4.7. As mentioned above, a baby Verma module over a small quantum
group or a modular Lie algebra in positive characteristic has a Jantzen filtration with
an analogous sum formula as the one of Theorem 5.4.1 (cf. [AJS94], chapter 6, or
[KL97], Theorem 10.1). A conjecture of [Lus90] and [Lus91] states a much stronger
relation between critical representations of affine Kac-Moody algebras and quantum
groups at roots of unity and claims that the restricted category O resembles the
category considered in [AJS94]. By this means, it makes sense to conjecture that
restricted projective objects are also tilting modules in O. The dual of a restricted
Verma module ∆(λ) is again restricted (cf. [AF12b]) and we denote it by ∇(λ). Let
P ∈ O be projective. Since we conjecture that P is also tilting, we can define an
Andersen filtration on the space HomO (∆(λ), P ) in a very similar way as in chapter
3 by a deformed version of the composition pairing

## where all Hom-spaces are meant to be taken in O. In accordance to [AK01] we

denote the Andersen filtration by Fλ (P )j and conjecture the equation

## dimC Fλ (P )j = dimC HomO (∆(λ)j , P )

where ∆(λ)j denotes the Jantzen filtration on ∆(λ) which we defined above. This
conjecture is motivated by Theorem 3.8 of [AK01] where an analogous result is
proven in the case of G1 T -modules and small quantum groups at roots of unity.

65
Bibliography

[And97] Henning H. Andersen, Filtrations and tilting modules, Ann. Sci. École Norm. Sup. 30
(1997), no. 3, 353–366.
[And03] , Twisted Verma modules and their quantized analogues, Combinatorial and geo-
metric representation theory, Contemp. Math., vol. 325, Amer. Math. Soc., Providence,
RI, 2003, pp. 1–10.
[AJS94] Henning H. Andersen, Jens C. Jantzen, and Wolfgang Soergel, Representations of quan-
tum groups at a p-th root of unity and of semisimple groups in characteristic p: inde-
pendence of p, Astérisque 220 (1994).
[AK01] Henning H. Andersen and Masaharu Kaneda, Filtrations on G1 T -modules, Proc. London
Math. Soc. 82 (2001), no. 3, 614–646.
[And12] Troels B. Andersen, Filtrations of representations of quantum groups using deformation
theory, progress report, Aarhus (2012).
[AF12a] Tomoyuki Arakawa and Peter Fiebig, On the restricted Verma modules at the critical
level, Trans. Amer. Math. Soc. 364 (2012), no. 9, 4683–4712.
[AF12b] , The linkage principle for restricted critical level representations of affine Kac-
Moody algebras, Compos. Math. 148 (2012), no. 6, 1787–1810.
[Ark97] Sergey Arkhipov, Semi-infinite cohomology of associative algebras and bar duality, In-
ternat. Math. Res. Notices 17 (1997), 833–863.
[BB93] Alexander A. Beilinson and Joseph N. Bernstein, A proof of Jantzen conjectures, I. M.
Gelfand Seminar, Adv. Soviet Math., 16, Part 1, AMS (1993), 1–50.
[BGS96] Alexander A. Beilinson, Victor Ginzburg, and Wolfgang Soergel, Koszul duality patterns
in representation theory, J. Amer. Math. Soc. 9 (1996), no. 2, 473–527.
[BM01] Tom Braden and Robert MacPherson, From moment graphs to intersection cohomology,
Math. Ann. 321 (2001), no. 3, 533–551.
[Bru04] Jonathan Brundan, Tilting modules for Lie superalgebras, Comm. Algebra 32 (2004),
no. 6, 2251–2268.
[Car05] Roger Carter, Lie algebras of finite and affine type, Cambridge Studies in Advanced
Mathematics, 96., Cambridge University Press, Cambridge, 2005.
[CI89] David H. Collingwod and Ronald S. Irving, A decomosition theorem for certain self-dual
modules in the category O, Duke math. J. 58 (1989), no. 1, 89–102.
[EW12] Ben Elias and Geordie Williamson, The Hodge theory of Soergel bimodules,
arXiv:1212.0791 (2012).
[Fie01] Peter Fiebig, Zentren deformierter Darstellungskategorien und Verschiebungsfunktoren,
PhD thesis, Freiburg (2001).

66
[Fie03] , Centers and translation functors for the category O over Kac-Moody algebras,
Math. Z. 243 (2003), no. 4, 689–717.
[Fie06] , The combinatorics of category O over symmetrizable Kac-Moody algebras,
Transform. Groups 11 (2006), no. 1, 29–49.
[Fie08a] , Sheaves on moment graphs and a localization of Verma flags, Adv. Math. 217
(2008), no. 2, 683–712.
[Fie08b] , The combinatorics of Coxeter categories, Trans. Amer. Math. Soc. 360 (2008),
no. 8, 4211–4233.
[Fie12a] , On the restricted projective objects in the affine category O at the critical level,
Algebraic groups and quantum groups, Contemp. Math., vol. 565, Amer. Math. Soc.,
Providence, RI, 2012, pp. 55–70.
[Fie12b] , An upper bound on the exceptional characteristics for Lusztig’s character for-
mula, J. Reine Angew. Math. 673 (2012), 1–31.
[Fie13] , On the subgeneric restricted blocks of affine category O at the critical level, Sym-
metries, Integrable Systems and Representations, Springer Proceedings in Mathematics
and Statistics, vol. 40, Springer, 2013, pp. 65–84.
[FW08] Peter Fiebig and Geordie Williamson, Parity sheaves, moment graphs and the p-smooth
locus of Schubert varieties, Ann. Inst. Fourier, to appear (2008).
[Fre05] Edward Frenkel, Wakimoto modules, opers and the center at the critical level, Adv. Math.
195 (2005), no. 2, 297–404.
[FG06] Edward Frenkel and Dennis Gaitsgory, Local geometric Langlands correspondence and
affine Kac-Moody algebras, Algebraic geometry and number theory, Progr. Math.,
vol. 253, Birkhäuser Boston, Boston, MA, 2006, pp. 69–260.
[GJ81] Ofer Gabber and Antony Joseph, Towards the Kazhdan-Lusztig conjecture, Ann. Sci.
École Norm. Sup. 4e série 14 (1981), no. 3, 261–302.
[IK06] Kenji Iohara and Yoshiyuki Koga, Tilting equivalence for superconformal algebras, Math.
Scand. 99 (2006), no. 1, 17–52.
[IK11] , Representation theory of the Virasoro algebra, Springer Monographs in Mathe-
matics, Springer-Verlag London, Ltd., London, 2011.
[Jan79] Jens Carsten Jantzen, Moduln mit einem höchsten Gewicht, Lecture Notes in Mathe-
matics, vol. 750, Springer, Berlin, Heidelberg, New York, 1979.
[Jan08] , Moment graphs and representations, Summer school on geometric methods in
representation theory, Grenoble, France (2008).
[Kac90] Victor Kac, Infinite-dimensional Lie algebras. Third edition, Cambridge University
Press, Cambridge, 1990.
[KK79] Victor G. Kac and David A. Kazhdan, Structure of representations with highest weight
of infinite-dimensional Lie algebras, Adv. in Math. 34 (1979), no. 1, 97–108.
[KT00] Masaki Kashiwara and Toshiyuki Tanisaki, Characters of irreducible modules with non-
critical highest weights over affine Lie algebras, Representations and quantizations
(Shanghai, China, 1998), China High. Educ. Press, Beijing, 2000, pp. 275–296.
[Küb10] Johannes Kübel, Eine kombinatorische Version der Jantzen-Filtrierung, Diplomarbeit,
Freiburg (2010).
[Küb12a] , From Jantzen to Andersen filtration via tilting equivalence, Math. Scand. 110
(2012), no. 2, 161–180.

67
[Küb12b] , Tilting modules in category O and sheaves on moment graphs, J. Algebra 371
(2012), 559–576.
[Küb12c] , Jantzen sum formula for restricted Verma modules over affine Kac-Moody al-
gebras at the critical level, arXiv: 1212.0150, to appear in Pacific J. Math. (2012).
[Küb13] , Centers for the restricted category O at the critical level over affine Kac-Moody
algebras, arXiv: 1301.6016, submitted to Math. Z. (2013).
[KL97] Shrawan Kumar and Gail Letzter, Shapovalov determinant for restricted and quantized
restricted enveloping algebras, Pacific J. Math. 179 (1997), no. 1, 123–161.
[Lan12] Martina Lanini, Kazhdan-Lusztig combinatorics in the moment graph setting, J. Algebra
370 (2012), 152–170.
[Lus90] George Lusztig, On quantum groups, J. Algebra 131 (1990), no. 2, 466–475.
[Lus91] , Intersection cohomology methods in representation theory, Proceedings of the
International Congress of Mathematicians, Vol. I, II (Kyoto, 1990), Math. Soc. Japan,
Tokyo, 1991, pp. 155–174.
[Qui73] Daniel Quillen, Higher algebraic K-theory. I, Algebraic K-theory, I: Higher K-theories
(Proc. Conf., Battelle Memorial Inst., Seattle, Wash., 1972, 1972), Vol. 341, Lecture
Notes in Math., Springer, Berlin, 1973, pp. 85–147.
[SVV11] Peng Shan, Michela Varagnolo, and Eric Vasserot, Koszul duality of affine Kac-Moody
algebras and cyclotomic rational double affine Hecke algebras, arXiv: 1107.0146v1 (2011).
[Sha12] Peng Shan, Graded decomposition matrices of v-Schur algebras via Jantzen filtration,
Represent. Theory 16 (2012), 212–269.
[Soe90] Wolfgang Soergel, Kategorie O, perverse Garben und Moduln über den Koinvarianten
zur Weylgruppe, Journal of the AMS 3 (1990), no. 2, 421–445.
[Soe97] , Kazhdan-Lusztig-Polynome und eine Kombinatorik für Kipp-Moduln, Repre-
sent. Theory 1 (1997), 37–68.
[Soe98] , Character formulas for tilting modules over Kac-Moody algebras, Represent.
Theory 2 (1998), 432–448.
[Soe99] , Character formulas for tilting modules over quantum groups at roots of one,
Current developments in mathematics (Cambridge, MA, 1997), Int. Press, Boston, MA,
1999, pp. 161–172.
[Soe08] , Andersen-Filtration and hard Lefschetz, Geom. Funct. Anal. 17 (2008), no. 6,
2066–2089.
[Str03] Catharina Stroppel, Category O: gradings and translation functors, J. Algebra 268
(2003), no. 1, 301–326.

68