Monodromy theorems in the affine setting... · 2019. 2. 13. · This connection is equivariant...
Transcript of Monodromy theorems in the affine setting... · 2019. 2. 13. · This connection is equivariant...
Monodromy theorems
in the affine setting
A dissertation presented
by
Andrea Appel
to
the Department of Mathematics
In partial fulfillment of the requirements for the degree of
Doctor of Philosophy
in the field of
Mathematics
Northeastern University
College of Science
Boston, MA
June 21, 2013
Monodromy theorems
in the affine setting
by
Andrea Appel
Abstract of Dissertation
Submitted in partial fulfillment of the requirements
for the degree of Doctor of Philosophy in Mathematics
in the College of Science of Northeastern University
June 21, 2013
Abstract
My doctoral work revolves around a principle which first appeared in the
work of T. Kohno and V. Drinfeld and states, roughly speaking, that quan-
tum groups are natural receptacles for the monodromy representations of cer-
tain flat connections of Fuchsian type. It stems more specifically from the
monodromic interpretation of quantum Weyl groups obtained by V. Toledano
Laredo. This shows that these operators describe the monodromy of the
Casimir connection, a flat connection with logarithmic singularities on the
root hyperplanes of a semisimple Lie algebra g, and values in any finite di-
mensional g-modules. The main goal of this thesis is to extend these results
to the case of an affine Kac-Moody algebra g.
The method we follow is close to that developed by Toledano Laredo in the
semisimple case, and relies on the notion of a quasi–Coxeter quasitriangular
quasibialgebra (qCqtqba), which is informally a bialgebra carrying actions of a
given generalized braid group and Artin’s braid groups on the tensor products
of its modules. A cohomological rigidity result shows that there is at most one
such structure with prescribed local monodromies on the classical envelop-
ing algebra Ug[[~]]. It follows that the generalized braid group actions arising
from quantum Weyl groups and the monodromy of the Casimir connection are
5
6 ABSTRACT
equivalent, provided the quasi–Coxeter quasitriangular quasibialgebra struc-
ture responsible for the former can be transferred from U~g to Ug[[~]]. The
proof of this fact relies on a modification of the Etingof–Kazhdan quantization
functor, and yields an isomorphism between (appropriate completions of) U~g
and Ug[[~]] preserving a given chain of subalgebras.
Acknowledgements
I wish to express my deepest gratitude to my advisor Prof. Valerio Toledano
Laredo for the large doses of guidance, endless patience and encouragement he
has shown during my years at Northeastern, and for being a constant source
of inspiration to me.
I would like to thank Professors Pavel Etingof, Venkatraman Lakshmibai,
and Ben Webster for being part of the Dissertation Committee. I am par-
ticularly grateful to Pavel Etingof for his interest in this work and for many
enlightening discussions.
This work would not have been possible without the constant help of my
friend and ”academic brother” Sachin Gautam: for the innumerable discus-
sions, classes and seminars we had together, as well as beers, glasses of scotch
and movies (korean, obviously).
I am grateful to my teachers at Northeastern University, Profs. Marc
Levine, Alexander Martsinkovsky, Alex Suciu, Gordana Todorov, Jonathan
Weitsman, Jerzy Weyman, and, in particular, Andrei Zelevinsky for offering
several excellent courses and seminars over the years.
7
8 ACKNOWLEDGEMENTS
I would like to thank Tiffany Brown, Donika Kreste, Elizabeth Qudah,
and Logan Wangsgard for their constant help and support in my endless fight
against bureaucracy. I am also very grateful to Pam Mabrouk and Phil He,
for all the support that I received by the College of Science and the Provost
Office, especially during my last year.
These five year in Boston would not have been so enjoyable without all my
friends and colleagues, Federico, Jeremy, Jason, Giulia, Martina, Luis, Sasha,
Roberto, Giulio, Francesco, David, Martina, Thomas, Carl, Nate, Thomas,
Floran, Yaping, Gufang, and many others.
I owe special thanks to my friends and colleagues, my actual family in
Boston: Giorgia and Salvo. Five years with them, full of joy, sorrow and frus-
tration, cannot be summarized in few lines. Thanks.
I would like to thank my family, my parents, my brother and sister, my
little niece. And finally I would like to thank one more time, for her endless
love and patience, Valentina, to whom I dedicate this thesis.
Contents
Abstract 3
Acknowledgements 7
Contents 9
Chapter 1. Introduction 11
1. Monodromy theorems in the affine setting 11
2. Results - Part I 17
3. Results - Part II 22
4. Results - Part III 30
5. Towards a complete solution 36
6. Outline of the dissertation 38
Chapter 2. Quasi–Coxeter Categories 41
1. Diagrams and nested sets 41
2. Quasi–Coxeter categories 46
Chapter 3. An equivalence of quasi–Coxeter categories 63
1. Etingof-Kazhdan quantization 64
2. A relative Etingof–Kazhdan functor 78
3. Quantization of Verma modules 101
4. Universal relative Verma modules 106
9
10 CONTENTS
5. Chains of Manin triples 117
6. The equivalence of quasi–Coxeter categories 124
Chapter 4. A rigidity theorem for quasi–Coxeter categories 141
1. The Dynkin-Hochschild bicomplex 142
2. Dynkin–Hochschild complex of monoidal functors 145
3. Uniqueness of the relative twist 153
4. The main theorem 162
Appendix A. 181
1. The completion of Ug in category O 181
2. The isomorphism between Ug[[h]] and U~g 189
3. A braided Hopf algebra in DΦD(gD) 197
Bibliography 205
CHAPTER 1
Introduction
My doctoral work revolves around a principle which first appeared in the
work of T. Kohno and V. Drinfeld and states, roughly speaking, that quantum
groups are natural receptacles for the monodromy representations of certain
flat connections of Fuchsian type [Ko, D2]. It stems more specifically from
the monodromic interpretation of quantum Weyl groups obtained by Toledano
Laredo in [TL2, TL4]. This shows that these operators describe the mon-
odromy of the Casimir connection, a flat connection with logarithmic singu-
larities on the root hyperplanes of a semisimple Lie algebra g, and values in
any finite dimensional g-modules. The main goal of my doctoral work is to
extend these results to the case of an affine Kac-Moody algebra g.
1. Monodromy theorems in the affine setting
1.1. Motivation: the Knizhnik–Zamolodchikov equations. Around
1990, Kohno [Ko] and Drinfeld [D2] proved a rather astonishing result, now
known as the Kohno-Drinfeld Theorem. The theorem states that quantum
groups can be used to describe the monodromy of certain first order Fuchsian
PDE’s known as the Knizhnik-Zamolodchikov (KZ) equations. Given a sim-
ple Lie algebra g, a representation V of g and a positive integer n, the KZ
11
12 1. INTRODUCTION
equations are the following system of PDEs
∂Φ
∂zi= ~
∑j 6=i
Ωij
zi − zjΦ
where
(i) Φ is a function on the configuration space of n ordered points in C
Xn := (z1, . . . , zn) ∈ Cn | zi 6= zj
with values in V ⊗n,
(ii) Ω ∈ g⊗ g is the Casimir tensor,
(iii) ~ is a complex deformation parameter.
This system is integrable and equivariant under the natural Sn-action and
hence defines a one-parameter family of monodromy representations of Artin’s
braid group Bn = π1(Xn/Sn) on V ⊗n. The Kohno-Drinfeld theorem asserts
that this representation is equivalent to the R-matrix representation of Bn on
V⊗n arising from the quantum group U~g. Here V is a quantum deformation
of V , i.e., a U~g–module such that V/~V ' V .
1.2. The Casimir connection. In subsequent work, J. Millson and V.
Toledano Laredo [MTL, TL2], and independently C. De Concini (unpub-
lished) and G. Felder et al. [FMTV], constructed another flat connection
∇C , now known as the Casimir connection, which is described as follows. Let
h ⊂ g be a Cartan subalgebra, and hreg be the complement of the root hyper-
planes in h, that is
hreg = h \⋃α∈Φ
ker(α)
1. MONODROMY THEOREMS IN THE AFFINE SETTING 13
where Φ = α ⊂ h∗ is the set of roots of g. For a finite–dimensional g-module
V , ∇C is the following connection on the trivial vector bundle hreg×V → hreg:
∇C = d− ~2
∑α∈Φ+
dα
αCα
where
(i) the summation is over a chosen system of positive roots Φ+ ⊂ Φ;
(ii) Cα is the Casimir operator of the sl2-subalgebra of g corresponding
to the root α.
This connection is equivariant under the action of the Weyl group W of g, and
gives rise to a one-parameter family of monodromy representations on V of
the generalized braid group
Bg = π1(hreg/W )
If V is a quantum deformation of V , the quantum Weyl group operators
Ti ∈ U~g defined by G. Lusztig, A.N. Kirillov–N. Reshetikin and Y. Soibel-
man, [L, KR, S] define a representation of Bg on V . In this setting a Kohno–
Drinfeld theorem was obtained by V. Toledano Laredo, stating the equivalence
of the above two representations of Bg [TL1, TL2, TL3, TL4, TL5].
The proof of this theorem is inspired by Drinfeld’s proof of the Kohno–
Drinfeld theorem [D2]. It proceeds along the following lines.
14 1. INTRODUCTION
(i) The notion of quasi-Coxeter algebras is introduced. Informally speak-
ing, these are algebras which carry a natural Bg-action on their finite-
dimensional modules. They replace the role of Drinfeld’s quasitrian-
gular quasibialgebras with respect to Artin’s braid group Bn. Just
as quasitriangular quasi–Hopf algebras are related to the Deligne–
Mumford compactification of the configuration space of n points in
C, quasi–Coxeter algebras are related to the compactifications of hy-
perplane arrangements constructed by De Concini–Procesi [DCP1,
DCP2].
(ii) It is shown that the monodromy of the Casimir connection and the
quantum Weyl group representations of Bg arise from suitable quasi-
Coxeter algebra structures on Ug[[~]] and U~g respectively.
(iii) It is shown that the quasi-Coxeter structure on U~g can be cohomo-
logically transferred to one on Ug[[~]].
(iv) An appropriate deformation cohomology for quasi-Coxeter algebras
(called Dynkin diagram cohomology) is introduced. It shows however
that quasi-Coxeter algebras on Ug[[~]] are not rigid. To remedy this,
it is necessary to also take into account the quasitriangular quasib-
ialgebra structure on Ug[[~]].
(v) The notion of quasi-Coxeter algebras is extended to that of quasi-
Coxeter quasitriangular quasibialgebras, whose deformation theory
1. MONODROMY THEOREMS IN THE AFFINE SETTING 15
is controlled by a bicomplex, namely the Dynkin-Hochschild bicom-
plex. Finally, one proves the uniqueness of quasi-Coxeter quasitrian-
gular quasibialgebra structures on Ug[[~]].
1.3. Affine extension of rational Casimir connection. It is natural
to ask if it is possible to generalize the previous construction to the affine
setting. If l is a symmetrisable Kac-Moody algebra, and U~l is its quantized
universal enveloping algebra, the corresponding quantum Weyl group opera-
tors give an action of the braid group of type l on any integrable U~l–module.
In particular, if l = g is the affinization of a semi–simple Lie algebra g, we ob-
tain an action of the affine braid group Bg of type g on integrable g-modules.
The purpose of my thesis is to solve the following
Problem. Give a monodromic description of the quantum Weyl group
action of the generalized affine braid group Bg on integrable g-modules in
category O.
The strategy to attack this problem consists in adapting the machinery
developed by Toledano Laredo in [TL4] to the affine setting.
Two major tools involved in the proof of the main theorem in [TL4] are still
applicable to the affine case. The theory of quasi-Coxeter algebras refers to a
general connected graph and it is therefore possible to consider affine Dynkin
diagrams as well. In addition, the Kohno-Drinfeld theorem was extended
to all symmetrizable Kac-Moody algebras by P. Etingof and D. Kazhdan in
16 1. INTRODUCTION
[EK1, EK2, EK6].
We proceed along the following steps, according to the principle that the
quantum Weyl group action is encoded by an appropriate quasi-Coxeter struc-
ture.
(i) Show that the quantum Weyl group representation of Bg and the
R-matrix representation of Bn arise from a quasi-Coxeter quasitrian-
gular quasibialgebras structure on the quantum affine algebra U~g.
(ii) Show that this structure on U~g can be transferred to U g[[~]].
(iii) Prove the rigidity of quasi-Coxeter quasitriangular quasibialgebra
structures on U g[[~]].
(iv) Define an affine extension of the rational Casimir connection, whose
monodromy representations are encoded by a quasi–Coxeter struc-
ture on U g[[~]]
The main difficulty in carrying out the above program lies in step (ii).
This step is carried out in the semisimple case by relying on the vanishing of
Hochschild cohomology groups, which does not hold for affine Lie algebras. An
alternative solution which relies instead on the Etingof-Kazhdan quantization
functor is explained in Sections 2 and 3 of this chapter. The ongoing work on
the solution to steps (iii) and (iv) is presented in Sections 4 and 5, respectively.
The step (i) is trivial.
2. RESULTS - PART I 17
2. Results - Part I
In order to correctly state the problem in the infinite–dimensional setting,
we first need to rephrase the notion of transfer of the quasi–Coxeter structure
in a category theoric way.
We first outline the definition of quasi–Coxeter algebras, following [TL4].
We then give the description of their categorical counterpart.
2.1. Quasi-Coxeter algebras.
2.1.1. Let D be a connected diagram, that is an undirected graph, with-
out loops or multiple edges, SD(D) the collection of subdiagrams of D and
CSD(D) the collection of connected subdiagrams. The following combinato-
rial definitions are due to De Concini–Procesi [DCP1, DCP2]. Two elements
B,B′ ∈ CSD(D) are compatible if either B ⊆ B′, B′ ⊆ B or B ⊥ B′, where the
latter means that B and B′ are disjoint and that no two vertices α ∈ B,α′ ∈ B′
are connected by an edge of D. A maximal nested set on D is a maximal col-
lection of pairwise compatible elements in CSD(D). If D is the Dynkin diagram
of type An−1, this notion is equivalent to that of complete bracketing on the
non-associative monomial x1 . . . xn, by mapping the connected diagram [i, j]
to the bracketing x1 . . . xi−1(xi . . . xj+1)xj+2 . . . xn. We denote by Mns(D) the
collection of maximal nested sets on D and by ND the poset of nested sets on
D, ordered by reverse inclusion.
2.1.2. AD-algebra A is an algebra A endowed with a family of subalgebras
ABB∈CSD(D) satisfying
AB ⊆ AB′ if B ⊆ B′ and [AB, AB′ ] = 0 if B ⊥ B′
18 1. INTRODUCTION
For any B ∈ SD(D) with connected components Bk, we denote by AB the
subalgebra generated by ABk.
A naıve morphism of D–algebras Ψ : A → A′ is a homomorphism such
that Ψ(AB) ⊆ A′B for any B ∈ SD(D). This notion is too restrictive for appli-
cations however. The correct notion of morphism of D-algebras Ψ : A → A′
is a collection of homomorphisms ΨF : A→ A′ indexed by F ∈ Mns(D), such
that ΨF(AB) ⊂ A′B for any B ∈ F .
2.1.3. A D–bialgebra A is a D–algebra and a bialgebra with coproduct
∆ such that ∆(AB) ⊆ AB ⊗ AB. A D-quasitriangular quasibialgebra A is a
D-bialgebra in which every AB is equipped with an additional quasitriangular
quasibialgebra structure of the form (AB,∆, RB,ΦB), so that the category
Rep(AB) of AB-modules is a braided tensor category. Moreover, for each
inclusion B ⊆ B′, the restriction functor Rep(AB′) → Rep(AB) is endowed
with a tensor structure. The latter depends upon the additional choice of a
chain:
B′ = B0 ⊃ B1 ⊃ B2 · · · ⊃ Bn = B (2.1)
where |Bi| = |B′| − i. These tensor structures compose in the obvious way
under the concatenation of chains. For any F ∈ Mns(D), we denote by
JF∈ A⊗2 the structural twist defining the tensor structure on the forgetful
functor Rep(A)→ Vect = Rep(A∅).
2.1.4. A quasi-Coxeter quasitriangular quasibialgebra (qCqtqb) of type D
is a D-quasitriangular quasibialgebra with two additional data. First, for any
G,F ∈ Mns(D), we are given a gauge transformation ΦFG ∈ A relating the
2. RESULTS - PART I 19
corresponding twists, i.e.,
JG = Φ⊗2GF · JF ·∆(ΦGF)−1
These satisfy additional relations, in particular,
Orientation: for any F ,G ∈ Mns(D),
ΦFG = Φ−1GF
Transitivity: for any F ,G,H ∈ Mns(D),
ΦFH = ΦFG · ΦGH
In the following, we shall refer to the gauge transformation ΦFG as Casimir
associators.
We recall that a quasi–Coxeter algebra admits an equivalent description
in which the relations satisfied by the elements ΦFG are labeled by the two–
dimensional faces of the De Concini–Procesi associahedron, i.e., the regular
CW–complex whose poset of nonempty faces is ND [TL4, 3.8].
Assume now that D is labeled,i.e., an integer mij ∈ 2, 3, ...,∞ is given
for any pair αi, αj of distinct vertices of D, such that mij = mji and mij = 2 if
and only if αi and αj are orthogonal. The second piece of data is given by the
local monodromies Sαα∈D, which define the action of the generalized braid
group BD attached to D and its labeling. For any vertex α ∈ D, Sα is not
an element of Aα, but rather an element of the completion of Aα with respect
to a fixed full subcategory Cα ⊂ Rep(Aα). In [TL4], Cα is the subcategory of
20 1. INTRODUCTION
finite-dimensional representations. Moreover, the elements Sα and ΦG,F
satisfy the following
Braid relations: for any pair of vertices αi 6= αj ∈ D such that
2 < mij <∞, and for any G,F ∈ Mns(D) such that αi ∈ F , αj ∈ G,
Ad(ΦGF)(Sαi) · Sαj · · · = Sαj · Ad(ΦGF)(Sαi) · · ·
where the number of factors on both sides is equal to mij.
The local monodromies are not assumed to preserve the structural twists.
Rather, they satisfy the coproduct identity
∆JF (Sα) = (R21α )JF · (Sα ⊗ Sα)
for any F ∈ Mns(D) such that α ∈ F .
2.2. Categorical analogue of quasi–Coxeter algebras. It is clear
from the previous section that every ingredient in the definition of a quasi–
Coxeter quasitriangular quasibialgebra has a categorical interpretation. More-
over, the definition of the local monodromies seems more natural in the cat-
egorical setting. Implicitly, we are using this approach to describe a quasi–
Coxeter structure on the completion of a D–bialgebra with respect to a fixed
compatible subcategory of representations. A detailed description of quasi–
Coxeter categories is presented in Chapter 2.
A quasi–Coxeter braided monoidal category is to a quasi–Coxeter quasitri-
angular quasibialgebra what a braided monoidal category is to a quasitriangu-
lar quasibialgebra. Following this principle, we define a quasi–Coxeter braided
2. RESULTS - PART I 21
monoidal category of type D as a tuple
C = ((CB,⊗B,ΦB, βB), (FBB′ , JFBB′), ΦFG, Si)
where
• (CB,⊗B,ΦB, βB) are braided monoidal categories indexed by sub-
diagrams of D;
• (FBB′ , JFBB′) : CB′ → CB are monoidal functors, corresponding to
the inclusion B′ ⊂ B, with a tensor structure depending upon the
choice of a maximal nested set in Mns(B,B′);
• A collection of invertible endomorphisms Si ∈ End(Fi)i∈D and nat-
ural transformations
ΦFG ∈ Nat⊗((FBB′ , J
GBB′), (FBB′ , J
FBB′)
)×for any F ,G ∈ Mns(B,B′), satisfying analogous relations to those
described in the previous section.
We then obtain the description of morphisms of quasi–Coxeter structure
as the datum of
• for any B ⊆ D a functor HB : CB → C ′B;
• for any B ⊆ B′ and F ∈ Mns(B,B′), a natural transformation
CB′HB′
//
FBB′
C ′B′
F ′BB′
γFBB′
z
CBHB
// C ′B
22 1. INTRODUCTION
satisfying the obvious compatibility condition with respect to the quasi–Coxeter
braided monoidal structure.
Interestingly perhaps, these notions can be concisely rephrased in terms of a
2–functor from a combinatorially defined 2–category qC(D) to the 2–categories
Cat⊗ of tensor categories. The objects of qC(D) are the subdiagrams of the
Dynkin diagram D of B and, for two subdiagrams D′ ⊆ D′′, HomqC(D)(D′′, D′)
is the fundamental 1–groupoid of the De Concini–Procesi associahedron for the
quotient diagram D′′/D′.
3. Results - Part II
For a semisimple Lie algebra g, the transfer of structure ultimately rests on
the vanishing of the first and second Hochschild cohomology groups of Ug[[~]],
and in particular on the fact that U~g and Ug[[~]] are isomorphic as algebras,
a fact which does not hold for affine Kac–Moody algebras. Rather than the
cohomological methods of [TL4], we use instead the Etingof–Kazhdan (EK)
quantization functor [EK1, EK2, EK6], which yields a canonical isomor-
phism
ΨEK : U~g∼−→ Ug[[~]]
between the completions of U~g and Ug[[~]] with respect to category O.
Surprisingly perhaps, and despite its functorial construction, the isomor-
phism ΨEK does not preserve the inclusions of subalgebras
U~gD ⊆ U~g and UgD[[~]] ⊆ Ug[[~]]
determined by a subdiagram D of the Dynkin diagram of g, something which
is clearly required by the transfer of structure.
3. RESULTS - PART II 23
The isomorphism ΨEK has a categorical origin. Namely, it comes from
the equivalence between the category O of Ug[[~]] and U~g established by the
Etingof–Kazhdan functor. Similarly, the compatibility conditions with the
subalgebras gD can be expressed in terms of natural relations with the corre-
sponding restriction functors. It follows that the desired isomorphism needs
to be obtained from an equivalence of quasi–Coxeter categories, as discussed
in the previous section. Such equivalence is constructed in Chapter 3.
3.1. To outline our construction, which works more generally for an in-
clusion (gD, gD,−, gD,+) ⊂ (g, g−, g+) of Manin triples over a field k of char-
acteristic zero, recall first that the main steps of the EK construction are as
follows (cf. Ch.3, Section 1).
(i) One considers the Drinfeld category DΦ(g) of (deformation) equicon-
tinuous g–modules, with associativity constraints given by a fixed Lie
associator Φ over k. This category can be thought of as a topological
analogue of category O when g is the Manin triple associated to a
Kac–Moody algebra. It can equivalently be described as the category
of Drinfeld–Yetter modules over the Lie bialgebra g−.
(ii) One constructs a tensor functor F fromDΦ(g) to the category Vectk[[~]]
of topologically free k[[~]]–modules. The algebra of endomorphisms
H = End(F ) is then a topological bialgebra, i.e., it is endowed with
a coproduct ∆ mapping H to a completion of H ⊗H.
(iii) Inside H, one constructs a subalgebra U~g− such that ∆(U~g−) ⊂
U~g− ⊗ U~g−, and which is a quantization of Ug−. The quantum
group U~g is then defined as the quantum double of U~g−.
24 1. INTRODUCTION
(iv) By construction, Ug− acts and coacts on any F (V ), V ∈ DΦ(g),
so that the functor F lifts to F : DΦ(g) → Rep(U~g) where, by
definition, the latter is the category of Drinfeld–Yetter modules over
U~(g−).
(v) Finally, one proves that F is an equivalence of categories.
Since F is isomorphic to the forgetful functor f : DΦ(g) → Vectk[[~]] as
abelian functors, we obtain the following diagram
DΦ(g)
f
DΦ(g)F
//
F
Rep(U~g)
f~
ks
Vectk[[~]] Vectk[[~]] Vectk[[~]]
where f~ : Rep(U~g)→ Vectk[[~]] is the forgetful functor. The EK isomorphism
ΨEK is then given by the identifications
Ug[[~]] := End(f) ∼= End(F ) = End(f~ F ) ∼= End(f~) =: U~g
It is easy to show that in the Kac–Moody case the construction is compatible
with the weight decomposition. Therefore, by restriction of the above diagram
to the category O, we get the desired isomorphism.
3. RESULTS - PART II 25
3.2. Overlaying the above diagrams for an inclusion iD : gD → g of
Manin triples shows that constructing an isomorphism U~g∼−→ Ug[[~]] com-
patible with iD may be achieved by filling in the diagram
DΦ(g)i∗D
f
DΦ(g)Γ
F//
F
Rep(U~g)i∗D,~
yy
f~
DΦ(gD)fD
##
DΦ(gD)FD//
FD
##
Rep(U~gD)fD,~
%%
Vectk[[~]] Vectk[[~]] Vectk[[~]]
where fD, fD,~ are forgetful functors, FD the EK functor for gD, and iD,~ :
U~gD → U~g is the inclusion derived from the functoriality of the quantization.
To do so, we first construct a relative fiber functor, i.e., a (monoidal)
functor Γ on DΦ(g) whose target category is DΦ(gD) rather than Vectk[[~]], and
which is isomorphic as abelian functor to the restriction i∗D. We then show
the existence of a natural transformation between the composition FD Γ
and i∗D,~ F . Our constructions do not immediately yield a commutative
diagram, i.e., the two factorizations F ∼= FD Γ deduced from f = fD i∗Dand fh = fD,~ i∗D,~ do not coincide, but this can easily be adjusted by using
a different identification F ∼= f, which amounts to modifying the original EK
isomorphism.
3.3. The construction of the functor Γ (Ch.3 Section 2) is very much
inspired by [EK1]. The principle adopted by Etingof and Kazhdan is the
following. In a k-linear monoidal category C, a coalgebra structure on an
26 1. INTRODUCTION
object C ∈ Obj(C) induces a tensor structure on the Yoneda functor
hC = HomC(C,−) : C → Vectk
If C is braided and C1, C2 are coalgebra objects in C, then so is C1 ⊗ C2, and
there is therefore a canonical tensor structure on hC1⊗C2 .
If g is finite–dimensional, the polarization Ug 'M− ⊗M+, where M± are
the Verma modules Indgg∓ k, realizes Ug as the tensor product of two coalgebra
objects in DΦ(Ug[[~]]). This yields a tensor structure on the forgetful functor
hUg : DΦ(Ug)[[~]]→ Vectk[[~]]
Our starting point is to apply the same principle to the (abelian) restriction
functor i∗D : DΦ(Ug) → DΦ(UgD[[~]]). We therefore factorize Ug as a tensor
product of two coalgebra objects L−, N+ in the braided monoidal category of
(g, gD)–bimodules, with associator (Φ ·Φ−1D ), where Φ−1
D acts on the right. Just
as the modules M−,M+ are related to the decomposition g = g−⊕g+, L− and
N+ are related to the asymmetric decomposition
g = m− ⊕ p+
where m− = g− ∩ g⊥D and p+ = gD ⊕ m+. This factorization induces a tensor
structure on the functor Γ = hL−⊗N+ , canonically isomorphic to i∗D through
the right gD–action on N+. As in [EK1, Part II], this tensor structure can
also be defined in the infinite–dimensional case.
3. RESULTS - PART II 27
3.4. To construct a natural transformation making the following diagram
commute
DΦ(g)F
//
Γ
Rep(U~g)
v~
(iD)∗~
DΦ(gD)FD
// Rep(U~gD)
we remark, as suggested to us by P. Etingof, that a quantum analogue Γ~ of
Γ can be similarly defined using a quantum version L~−, N~+ of the modules
L−, N+. The functor Γ~ = HomU
EK~ g
(L~− ⊗ N~+,−) is naturally isomorphic
to (iD)∗~ as tensor functor, since there is no associator involved on this side.
Moreover, an identification
FD Γ ' Γ~ F
is readily obtained, provided one establishes isomorphisms of (UEK
~ g, UEK
~ gD)–
bimodules
FD F (L−) ' L~− and FD F (N+) ' N~+
3.5. While for M± it is easy to construct an isomorphism between F (M±)
and the quantum counterparts of M±, the proof for L−, N+ is more involved
and is presented in Ch.3, Section 4. It relies on a description of the quan-
tization functor F EK in terms of Prop categories (cf. [EK2, EG]) and the
28 1. INTRODUCTION
realization of L−, N+ as universal objects in a suitable colored Prop describ-
ing the inclusion of bialgebras gD ⊂ g. This yields in particular a relative
extension of the EK functor with input a pair of Lie bialgebras a, b which is
split, i.e., endowed with maps aip b such that p i = id.
3.6. The previous construction leads to the following
Theorem. Let g, gD be Manin triples with a finite N–grading on g−, gD,−
and iD : gD ⊆ g an inclusion of Manin triples compatible with the grading.
Then, there exists an algebra isomorphism
Ψ : UEK
~ g→ Ug[[~]]
restricting to ΨEKD on U
EK
~ gD, where the completion is given with respect to
Drinfeld–Yetter modules.
This result naturally extends to a chain of Manin triples of arbitrary length
(Ch.3, Section 5).
Theorem.
(i) For any chain of Manin triples
C : g0 ⊆ g1 ⊆ · · · ⊆ gn ⊆ g
there exists an isomorphism of algebras
ΨC : UEK
~ g→ Ug[[~]]
such that ΨC(UEK
~ gk) = Ugk[[~]] for any gk ∈ C.
3. RESULTS - PART II 29
(ii) Given two chains C,C′, the natural transformation
ΦCC′ := γ−1C γC′ ∈ Aut(F EK)
satisfies
Ad(ΦCC′)ΨC′ = ΨC
A straightforward computation shows that the natural transformations Φ
satisfy the properties of orientation and transitivity required by the Casimir
associators.
3.7. Given a maximal nested set F , by iteration of the previous pro-
cedure, we obtain an algebra isomorphism ΨF : U~g → Ug[[~]] such that
ΨF(U~gD) = UgD[[~]] for any D ∈ F . This isomorphism of Dg-algebras
transfers the quasi-Coxeter quasitriangular quasibialgebra structure of U~g to
Ug[[~]].
Theorem. Let g be a symmetrizable Kac–Moody algebra with a fixed Dg–
structure and U~g the corresponding Drinfeld–Jimbo quantum group with the
analogous Dg–structure. For any choice of a Lie associator Φ, there exists an
equivalence of quasi–Coxeter categories between
O := ((OintgB,⊗B,ΦB, σRB), (ΓBB′ , JBB
′
F ), ΦGF, Si,C)
and
O~ := ((OintU~gB,⊗B, id, σR~B), (Γ~BB′ , id), id, S~i )
30 1. INTRODUCTION
where ⊗B denotes the standard tensor product in OintgB
and
Si,C = si exp(h
2· Ci)
ΦB ≡ 1 mod ~2
RB = exp(~2
ΩD)
Alt2JBB′
F ≡ ~2
(rB′ − r21B′
2− rB − r21
B
2
)mod ~2
and ΦGF , JBB′
F are weight zero elements.
We note that Kac–Moody algebras of finite, affine and hyperbolic type are
easily endowed with a D–structure. This is canonical in finite and affine type
and depends upon an initial choice in the hyperbolic case. For an arbitrary
Kac–Moody is not always possible to define a D–structure, but it is still un-
clear how to characterize this special class of Lie algebras (cf. Ch.3, Section 6).
The results of Chapters 2 and 3 are contained in the preprint [ATL1].
4. Results - Part III
4.1. The quasi–Coxeter braided monoidal category naturally associated
to U~g admits an equivalent counterpart for Ug[[~]], as illustrated in the previ-
ous section. The next goal is to show that there is at most one quasi–Coxeter
braided monoidal category for the classical enveloping algebra Ug[[~]] with
prescribed local monodromies (Chapter 4). This result is needed to show that
the quasi–Coxeter braided monoidal category naturally attached to the quan-
tum affine algebra U~g is equivalent to the one constructed for Ug[[~]], which
underlies the monodromy of the KZ and Casimir connections of g.
4. RESULTS - PART III 31
4.2. In the semisimple case, a central role is played by the Dynkin–
Hochschild bicomplex of a D–bialgebra A, which controls the deformations
of quasi–Coxeter quasitriangular quasibialgebra structures on A. In Chap-
ter 4 Section 2, we extend the notion of the Dynkin–Hochschild bicomplex
to the case of monoidal D–categories, using the language of endomorphisms
algebra of monoidal functors and the corresponding cohomology theory. In
analogy with the semisimple case, this bicomplex controls the deformations of
the quasi–Coxeter structures on the D–category.
4.3. Let g be a Kac–Moody algebra of affine type with extended Dynkin
diagram Dg. For any proper connected subdiagram D ⊂ Dg, let gD ⊂ lD ⊂ g
be the corresponding simple and Levi subalgebras, with decomposition lD =
gD ⊕ cD. Denote by
ΩD = xa ⊗ xa, CD = xa · xa and rgD =∑α0:
supp(α)⊆D
(α, α)
2· eα ∧ fα
where xaa, xaa are dual basis of gD with respect to (·, ·), the correspond-
ing invariant tensor, Casimir operator and standard solution of the modified
classical Yang–Baxter equation for gD respectively.
We then consider two quasi–Coxeter braided monoidal categories of type
D of the form:
(Oint
g [[~]], OintgD
[[~]],⊗, RD,ΦKZD , Si,c, Φk
(D;αi,αj), F(D;αi), J
k(D;αi)
)
32 1. INTRODUCTION
for k = 1, 2, where ⊗ is the standard tensor product in Og, (F(D;αi), Jk(D;αi)
)
are fiber functors,
Si,c = si · exp (~/2 · Ci) (4.1)
RD = exp(~/2 · ΩD) (4.2)
Alt2 J(D;αi) ≡~2
(rgD − rgD\αi
)mod ~2 (4.3)
and Φk(D;αi,αj)
, Jk(D;αi)are of weight 0, i.e., they properly extend to the corre-
sponding Levi’s subalgebras.
We are explicitly assuming that the braided monoidal structure on OintgD
is induced in both categories by the KZ equations for gD. This assumption
is not strictly necessary and can be easily dropped by Drinfeld’s uniqueness
theorem [D3, Prop. 3.5]. The proof of the rigidity property amounts to match
the relative twists Jk(D;αi) and the Casimir associator Φk
(D;αi,αj).
4.4. The procedure matching the relative twists relies on the essential
uniqueness (up to gauge transformations) of solutions of the equation
F (ΦVWZ) JV⊗W,Z (JV,W ⊗ idZ) = JV,W⊗Z (idV ⊗JW,Z) (ΦD)F (V )F (W )F (Z)
in HomUgD(F (V ) ⊗ F (W ) ⊗ F (Z), F (V ⊗ W ⊗ Z)), where F = FDgD and
V,W,Z ∈ Ointg .
4. RESULTS - PART III 33
In the semisimple case (cf. [TL3, Thm. 6.1]), it is possible to construct
elements u ∈ (Ug)gD [[~]] and λ ∈∧2 cD[[~]] such that
expλ · J2 = u ? J1
where u?J = ∆(u−1)·J ·u⊗u, i.e., the lack of uniqueness of such solutions (up
to gauge transformation) is measured by an element λ ∈∧2 cD. We illustrate
two cases in which is possible to eliminate λ and obtain the gauge equivalence.
When g is a finite type Lie algebra and |Dg \ D| = 1, the space cD is
one dimensional and λ = 0. Therefore, the twists J(D;αi) are uniquely deter-
mined up to gauge transformation. If we try to apply a similar result when
g is an affine Kac–Moody algebra, we can assert the uniqueness of the twists
J(D;αi) only for proper subdiagrams D ⊂ Dg. Indeed, when D = Dg and D′ is
a subdiagram satisfying |Dg\D′| = 1, dim cD′ = 2 and λ is not necessarily zero.
4.5. Alternatively, we can take into the account the behavior of the rel-
ative twists with respect to the Chevalley involution θ of g. It is easy to show
that if the relative twists satisfy the additional condition
Jθ = J21
then λ can be chosen to be zero and it is possible to determine a unique gauge
transformation u satisfying uθ = u.
34 1. INTRODUCTION
Once again, this condition cannot be directly applied to our case. First of
all, we have to make sense of the identity in the setting of category O, where
θ does not define an involution of the category. This can be easily rephrased
in terms of a natural compatibility of the monoidal functor with the duality
functor in category O.
More importantly, we have to make sure that the relative twists, described
in the previous section, satisfy (up to gauge equivalence) the compatibility con-
dition with the Chevalley involution. The solution to this problem presents
several difficulties. The current results are presented in Chapter 4 Section 3.
4.6. An easy computation shows that this is true for the Etingof–Kazhdan
twist in its finite–dimensional formulation i.e., as a tensor structure on the
functor hM+⊗M− . This is proved by identifying Jθ and J21 as tensor structures
on hM−⊗M+ given by the same formulae. This case is extremely special and it
can be shown, working modulo ~3, that this condition is not satisfied by the
twist JEK, in its universal form, or by the relative twist. We observe that in
both cases the symmetry between M+ and M− is broken.
We then observe that the condition Jθ = J21 can be weakened, assuming
there exists a natural transformation χ satisfying χθ = χ−1 and χ ? Jθ = J21.
Since the relative twist is gauge equivalent to a composition of twists of
Etingof–Kazhdan type, we reduced the problem to find a suitable χ for JEK.
4. RESULTS - PART III 35
This element is easily constructed in the finite–dimensional case, but its exis-
tence is not yet proven in the infinite–dimensional case.
We also have to remark here that the construction of the elements u and
λ above strongly relies on the assumption that the Hochschild cohomology of
the completion of Ug with respect to category O can be described in terms
of the exterior algebra of g, possibly completed. The result is well–know for
Ug [D3, 2.2], but it is not present in the literature for completed universal
enveloping algebras.
4.7. It remains to construct the natural transformation matching the
Casimir associators Φk(D;αi,αj)
and preserving the local monodromies. This
amounts to solving the following problem. Given a Cartan 2–cocycle in the
Dynkin complex, i.e., a special set of elements ϕ = ϕ(D;αi,αj) ⊂ hD, there
exists a Cartan 1–cochain a such that dDa = ϕ. The construction is recur-
sive and the initial choice a(αi,αi) = 0 allows to preserve the local monodromies.
The original proof of this fact in [TL4] is actually incomplete, since it is
implicitly assumed that the fundamental coweights λ∨i,D ⊂ hD do not de-
pends upon D or, in other words, that λ∨i,D′ ∈ Cλ∨i,D for D′ ⊂ D. We then
present a corrected proof of the semisimple case and we extend it to the affine
case. Since the proof strongly relies on the combinatorics of the Dynkin dia-
gram of g, it is actually necessary to go through the entire computation for
an affine Dynkin diagram. It is unclear at present whether such result holds
for an arbitrary symmetrizable Kac–Moody algebra. The results of Chapter 4
36 1. INTRODUCTION
are available in [ATL2].
5. Towards a complete solution
To complete the solution to Problem 1.3, we still need two steps, namely,
an affine extension of the rational Casimir connection, and the construction of
a monodromic quasi–Coxeter braided monoidal structure on the category of
integrable modules in category O of Ug[[~]].
5.1. The affine extension of the Casimir connection. A rational
Casimir connection was defined for any symmetrisable Kac–Moody algebra in
[FMTV]. It has the form
∇affC = d− ~
∑α>0
dα
α:Kα:
where α ranges over all positive (real and imaginary) roots, Kα is the truncated
Casimir operator corresponding to α, and
:Kα:=
mult(α)∑i=1
2e(i)−αe
(i)α
is its Wick ordered form.
This ordering makes the sum over α locally finite on a module in category
O, but breaks the equivariance of ∇affC with respect to the Weyl group of g.
If g is affine, this equivariance can however be restored by adding to ∇affC a
closed one–form A with values in the Cartan subalgebra h ⊂ g [TL6]. The
monodromy of the connection ∇affC + A then gives rise to a one–parameter
family of representations of the generalized affine braid group on V .
5. TOWARDS A COMPLETE SOLUTION 37
5.2. The Fusion operator. In the semisimple case, the quasi-Coxeter
quasitriangular quasibialgebra structure on Ug[[~]] underlying the monodromy
of the KZ and Casimir connections is obtained as follows [TL5]. The Dg–
algebra structure is the standard one. For any D ⊆ Dg, the quasitriangular
quasibialgebra structure on UgD[[~]] is given by the associator of the KZ equa-
tions for gD and the corresponding R–matrix. The gauge transformations ΦGF
and local monodromies are given by the De Concini–Procesi associators of the
Casimir connection, and its local monodromies. Finally, the twists JF are ob-
tained as suitable limits of a joint solution with prescribed asymptotics of the
system (cf. [FMTV])
dµJ =~2
∑α>0
dα
α
[∆(Kα)J − J(K(1)
α +K(2)α )]
+ (z1 ad(dµ(1)) + z2 ad(dµ(2)))J
dzJ = ~d(z1 − z2)
z1 − z2
Ω12J + (ad(µ(1))dz1 + ad(µ(2))dz2)J
where (z, µ) ∈ C2 × h and J has values in Ug[[~]]⊗2.
Following [Ka], we write
Kα =:Kα: + mult(α) · ν−1(α)
where ν : h → h∗ is the isomorphism induced by the non–degenerate bilinear
form on h. Since J is of weight zero, the above equations can be written with
the Wick ordered :Kα: instead of Kα, and therefore make sense for g. Moreover,
preliminary computations show that the construction of the joint solution J
carries over to affine Lie algebras and, in fact, arbitrary symmetrizable Kac–
Moody algebras.
38 1. INTRODUCTION
6. Outline of the dissertation
6.1. Chapter 2 is devoted to rephrasing the definition of quasi–Coxeter
quasitriangular quasibialgebra in categorical terms. This yields the notion
of a quasi–Coxeter category, which is to a generalized braid group B what
a braided tensor category is to Artin’s braid groups, and of a quasi–Coxeter
braided monoidal category. Both notions will be concisely rephrased in terms
of a 2–functor from a combinatorially defined 2–category qC(D) to the 2–
categories Cat,Cat⊗ of categories and monoidal categories respectively.
6.2. In Chapter 3, we establish an equivalence of quasi–Coxeter braided
monoidal categories for Ug[[~]] and U~g, where g is a Kac–Moody algebra
endowed with a D–algebra structure.
In Section 1, we start with a short review the construction of the Etingof–
Kazhdan quantization functor and the associated equivalence of categories,
following [EK1, EK6]. In Section 2, we modify this construction by using
generalized Verma modules L−, N+, and we obtain a relative fiber functor
Γ : DΦ(g) → DΦD(gD). In Section 3, we define the quantum generalized
Verma modules L~− and N~+. Using suitably defined Props we then show, in
Section 4 that these are isomorphic to the EK quantization of their classical
counterparts. In Section 5, we use these results to show that, for any given
chain of Manin triples ending in a given g, there exists a natural transformation
that preserves the given chain. Finally, in Section 6, we apply these results
to the case of a Kac–Moody algebra g and obtain the desired tranport of its
quasi–Coxeter quasitriangular quasibialgebra structure to the completion of
Ug[[~]] with respect to category O, integrable modules.
6. OUTLINE OF THE DISSERTATION 39
6.3. The goal of Chapter 4 is to show that, under appropriate hypotheses,
there is at most one quasi–Coxeter braided monoidal category for the classical
enveloping algebra Ug[[~]] with prescribed local monodromies (Thm 4.2).
In Section 1, we review a number of basic notions from [TL4], in particular
that of the Dynkin–Hochschild bicomplex of a D–bialgebra A which controls
the deformations of quasi–Coxeter quasitriangular quasibialgebra structures
on A. In Section 2, we extend the notion of the bicomplex to the case of
completion of algebras, adopting the language of endomorphisms algebra of
monoidal functors and the corresponding cohomology theory. In Section 3
we study invariant properties of the relative twist under the Chevalley invo-
lution of g and we prove a uniqueness property up to gauge transformation.
Section 4 contains the rigidity result, consisting in the construction of a natu-
ral transformation matching the Casimir associators and preserving the local
monodromies.
6.4. Finally, in Appendix 1 we discuss the isomorphism between the com-
pletion of the universal enveloping algebra described in [D5, EK6] and the
algebra of the endomorphism of the forgetful functor in category O. In Ap-
pendix 3 we use the relative twist to construct an Hopf algebra object in the
Drinfeld category of gD. The corresponding Radford algebra realizes a partial
quantization of Ug.
CHAPTER 2
Quasi–Coxeter Categories
This chapter is devoted to rephrasing the definition of quasi–Coxeter qua-
sitriangular quasibialgebra given in [TL4, Ch. 6] in categorical terms. This
yields the notion of a quasi–Coxeter category, which is to a generalized braid
group B what a braided tensor category is to Artin’s braid groups, and of
a quasi–Coxeter braided monoidal category. Interestingly, both notions can
be concisely rephrased in terms of a 2–functor from a combinatorially defined
2–category qC(D) to the 2–categories Cat,Cat⊗ of categories and monoidal cat-
egories respectively. The objects of qC(D) are the subdiagrams of the Dynkin
diagram D of B and, for two subdiagrams D′ ⊆ D′′, HomqC(D)(D′′, D′) is
the fundamental 1–groupoid of the De Concini–Procesi associahedron for the
quotient diagram D′′/D′ [DCP2, TL4].
1. Diagrams and nested sets
We review in this section a number of combinatorial notions associated to
a diagram D, in particular the definition of nested sets on D and of the De
Concini–Procesi associahedron of D following [DCP2] and [TL4, Section 2].
1.1. Nested sets on diagrams. By a diagram we shall mean an undi-
rected graph D with no multiple edges or loops. We denote the set of vertices
of D by V(D) and set |D| = |V(D)|. A subdiagram B ⊆ D is a full subgraph
of D, that is, a graph consisting of a subset V(B) of vertices of D, together
41
42 2. QUASI–COXETER CATEGORIES
with all edges of D joining any two elements of V(B). We will often abusively
identify such a B with its set of vertices and write i ∈ B to mean i ∈ V(B).
We denote by SD(D) the set of subdiagrams of D and by CSD(D) the set of
connected subdiagrams.1
The union B1 ∪ B2 of two subdiagrams B1, B2 ⊂ D is the subdiagram
having V(B1)∪ V(B2) as its set of vertices. Two subdiagrams B1, B2 ⊂ D are
orthogonal if V(B1)∩V(B2) = ∅ and no two vertices i ∈ B1, j ∈ B2 are joined
by an edge in D. B1 and B2 are compatible if either one contains the other or
they are orthogonal.
Definition. A nested set on a diagram D is a collection H of pairwise
compatible, connected subdiagrams of D which contains the empty set and
the connected components D1, . . . , Dr of D.
1.2. The De Concini–Procesi associahedron. LetND be the partially
ordered set of nested sets on D, ordered by reverse inclusion. ND has a unique
maximal element, consisting of the connected components Di of D and the
empty set. Its minimal elements are the maximal nested sets. We denote the
set of maximal nested sets on D by Mns(D). Every nested set H on D is
uniquely determined by a collection Hiri=1 of nested sets on the connected
components of D. We therefore obtain canonical identifications
ND =r∏i=1
NDi and Mns(D) =r∏i=1
Mns(Di)
1The definition of diagram, provided in [TL4, 2.1], did not allow empty graphs. InSection 2, we need to include the empty set in SD(D) and we are therefore removing thenon–emptiness condition.
1. DIAGRAMS AND NESTED SETS 43
The De Concini–Procesi associahedron AD is the regular CW–complex
whose poset of (nonempty) faces is ND. It easily follows from the definition
that
AD =r∏i=1
ADi
It can be realized as a convex polytope of dimension |D|−r. For any H ∈ ND,
we denote by dim(H) the dimension of the corresponding face in AD.
1.3. The rank function of ND. For any nested set H on D and B ∈ H,
we set iH(B) =⋃mi=1 Bi where the Bi’s are the maximal elements ofH properly
contained in B.
Definition. Set αBH = B \ iH(B). We denote by
n(B;H) = |αBH| and n(H) =∑B∈H
(n(B;H)− 1
)An element B ∈ H is called unsaturated if n(B;H) > 1.
Proposition. .
(i) For any nested set H ∈ ND,
n(H) = |D| − |H| = dim(H)
(ii) H is a maximal nested set if and only if n(B;H) = 1 for any B ∈ H.
(iii) Any maximal nested set is of cardinality |D|+ 1.
For any F ∈ Mns(D), B ∈ F , iF(B) denotes the maximal element in F
properly contained in B and αBF = B \ iF(B) consists of one vertex, denoted
αBF .
44 2. QUASI–COXETER CATEGORIES
For any F ∈ Mns(D), B ∈ F , we denote by FB ∈ Mns(B) the maximal
nested set induced by F on B.
1.4. Quotient diagrams. Let B ⊆ D be a subdiagram with connected
components B1, . . . , Bm.
Definition. The set of vertices of the quotient diagram D/B is V(D) \
V(B). Two vertices i 6= j of D/B are linked by an edge if and only if the
following holds in D
i 6⊥ j or i, j 6⊥ Bk for some k = 1, . . . ,m
For any connected subdiagram C ⊆ D not contained in B, we denote by
C ⊆ D/B the connected subdiagram with vertex set V (C) \ V (B).
1.5. Compatible subdiagrams of D/B.
Lemma. Let C1, C2 * B be two connected subdiagrams of D which are
compatible. Then
(i) C1, C2 are compatible unless C1 ⊥ C2 and C1, C2 ⊥Bi for some i.
(ii) If C1 is compatible with every Bi, then C1 and C2 are compatible.
In particular, if F is a nested set on D containing each Bi, then F = C,
where C runs over the elements of F such that C * B, is a nested set on
D/B.
Let now A be a connected subdiagram of D/B and denote by A ⊆ D the
connected sudbdiagram with vertex set
V (A) = V (A)⋃
i:Bi⊥V (A)
V (Bi)
1. DIAGRAMS AND NESTED SETS 45
Clearly, A1 ⊆ A2 or A1 ⊥ A2 imply A1 ⊆ A2 and A1 ⊥ A2 respectively, so the
lifting map A→ A preserves compatibility.
1.6. Nested sets on quotients. For any connected subdiagrams A ⊆
D/B and C ⊆ D, we have
A = A and C = C⋃
i:Bi⊥C
Bi
In particular, C = C if, and only if, C is compatible with B1, . . . , Bm and not
contained in B. The applications C → C and A→ A therefore yield a bijection
between the connected subdiagrams of D which are either orthogonal to or
strictly contain each Bi and the connected subdiagrams of D/B. This bijection
preserves compatibility and therefore induces an embedding ND/B → ND.
This yields an embedding
ND/B ×NB = ND/B ×(NB1 × · · · × NBm
)→ ND
with image the poset of nested sets on D containing each Bi. Similarly, for
any B ⊆ B′ ⊆ B′′, we obtain a map
∪ : NB′′/B′ ×NB′/B → NB′′/B
The map ∪ restricts to maximal nested sets. For any B ⊂ B′, we denote by
Mns(B′, B) the collection of maximal nested sets on B′/B. Therefore, for any
B ⊂ B′ ⊂ B′′, we obtain an embedding
∪ : Mns(B′′, B′)×Mns(B′, B)→ Mns(B′′, B)
46 2. QUASI–COXETER CATEGORIES
such that, for any F ∈ Mns(B′′, B′),G ∈ Mns(B′, B),
(F ∪ G)B′/B = G
1.7. Elementary and equivalent pairs.
Definition. An ordered pair (G,F) in Mns(D) is called elementary if G
and F differ by one element. A sequence H1, . . . ,Hm in Mns(D) is called
elementary if |Hi+1 \ Hi| = 1 for any i = 1, 2, . . . ,m− 1.
Definition. The support supp(F ,G) of an elementary pair in Mns(D) is
the unique unsaturated element of F ∩ G. The central support z supp(F ,G) is
the union of the maximal elements of F ∩G properly contained in supp(F ,G).
Thus
z supp(F ,G) = supp(F ,G) \ αsupp(F ,G)F∩G
Definition. Two elementary pairs (F ,G), (F ′,G ′) in Mns(D) are equiva-
lent if
supp(F ,G) = supp(F ′,G ′)
αsupp(F ,G)F = α
supp(F ′,G′)F ′ α
supp(F ,G)G = α
supp(F ′,G′)G′
2. Quasi–Coxeter categories
The goal of this section is to rephrase the notion of quasi–Coxeter quasi-
triangular quasibialgebra defined in [TL4] in terms of categories of represen-
tations.
2. QUASI–COXETER CATEGORIES 47
2.1. Algebras arising from fiber functors. We shall repeatedly need
the following elementary
Lemma. Consider the following situation
C
F **
H// D
G
α
w
A
where A, C,D are additive k–linear categories, F,G,H functors, and α is
an invertible transformation. If H is an equivalence of categories, the map
End(G) −→ End(F ) given by
gW 7→ Ad(α−1V )(gH(V ))
is an algebra isomorphism.
2.2. D–categories. Recall [TL4, Section 3] that, given a diagram D,
a D–algebra is a pair (A, ABB∈SD(D)), where A is a unital associative k–
algebra and ABB∈SD(D) is a collection of subalgebras indexed by SD(D) and
satisfying
AB ⊆ AB′ if B ⊆ B′ and [AB, AB′ ] = 0 if B ⊥ B′
with A∅ = k. Moreover, if B ∈ SD(D) has connected components Bk, then
AB is generated by the subalgebras ABk.
The following rephrases the notion of D–algebras in terms of their category
of representations.
Definition. A D–category C = (CB, FBB′) is the datum of
48 2. QUASI–COXETER CATEGORIES
• a collection of k–linear additive categories CBB⊆D
• for any pair of subdiagrams B ⊆ B′, an additive k–linear functor
FBB′ : CB′ → CB1
• for any B ⊂ B′, B′ ⊥ B′′, B′, B′′ ⊂ B′′′, a homomorphism of k–
algebras
η : End(FBB′)→ End(F(B∪B′′)B′′′)
satisfying the following properties
• For any B ⊆ D, FBB = idCB .
• For any B ⊆ B′ ⊆ B′′, FBB′ FB′B′′ = FBB′′ .
• For any B′ ⊥ B′′ and any B ⊂ B′ and any B′′′ ⊃ B′, B′′, the following
diagram of algebra homomorphisms commutes:
End(FBB′)id⊗1
tt
1⊗η++
End(FBB′)⊗ End(FB′B′′′)
**
End(FB(B∪B′′))⊗ End(F(B∪B′′)B′′′)
ss
End(FBB′′′)
Remark. We will usually think of C∅ as a base category and at the functors
F as forgetful functors. Then the family of algebras End(FB) defines, through
the morphisms α, a structure of D–algebra on End(FD). Conversely, every
D–algebra A admits such a description setting CB = RepAB for B 6= ∅ and
C∅ = Vectk, FBB′ = i∗B′B, where iB′B : AB ⊂ AB′ is the inclusion.
1We adopt the convention FB = F∅B .
2. QUASI–COXETER CATEGORIES 49
Remark. The conditions satisfied by the maps η imply that, for any B′ ⊥
B′′ and any B ⊂ B′ and B′′′ ⊃ B′, B′′, the images in End(FBB′′′) of the maps
End(FBB′)→ End(FBB′FB′B′′′) = End(FBB′′′)
End(FB(B∪B′′))→ End(FB(B∪B′′)F(B∪B′′)B′′′) = End(FBB′′′)
commute. In particular, given B′′′ =⊔rj=1Bj, with Bj ∈ SD(D) pairwise
orthogonal, the images in End(FB′′′) of the maps
End(FBj)→ End(FBjFBjB′′′) = End(F ′′′B )
pairwise commute. This condition replaces the D–algebra axiom
[AB′ , AB′′ ] = 0 ∀ B′ ⊥ B′′
which is equivalent to the condition that AB′ ⊂ AB′′
B , for any B ⊃ B′, B′′.
Remark. It may seem more natural to replace the equality of functors
FBB′ FB′B′′ = FBB′′ by the existence of invertible natural transformations
αB′
BB′′ : FBB′ FB′B′′ ⇒ FBB′′ for any B ⊆ B′ satisfying the associativity
constraints αB′
BB′′′ FBB′(αB′′
B′B′′′) = αB′′
BB′′′ (αB′
BB′′)FB′′B′′′ for any B ⊆ B′ ⊆
B′′ ⊆ B′′′. A simple coherence argument shows however that this leads to a
notion of D–category which is equivalent to the one given above.
Remark. The above definition of D–category may be rephrased as fol-
lows. Let I(D) be the category whose objects are subdiagrams B ⊆ D and
morphisms B′ → B the inclusions B ⊂ B′. Then a D–category is a functor
C : I(D) → Cat which is compatible with the disjoint union of orthogonal
subdiagrams.
50 2. QUASI–COXETER CATEGORIES
2.3. Strict morphisms of D–categories. The interpretation ofD–categories
in terms of I(D) suggests that a morphism of D–categories C, C ′ is one of the
corresponding functors
I(D)
C''
C′
::
Cat
This yields the following definition. For simplicity, we shall assume throughout
that C∅ = C ′∅.
Definition. A strict morphism of D–categories C, C ′ is the datum of
• for any B ⊆ D, a functor HB : CB → C ′B• for any B ⊆ B′, a natural transformation
CB′HB′
//
FBB′
C ′B′
F ′BB′
γBB′
z
CBHB
// C ′B
(2.1)
such that
• H∅ = id
• γBB = idHB
• For any B ⊆ B′ ⊆ B′′,
γBB′′ = γBB′ γB′B′′
2. QUASI–COXETER CATEGORIES 51
where is the composition of natural transformations defined by
CB′′ //
C ′B′′
y
CB′ //
C ′B′
y
CB // C ′B
(2.2)
The diagram (2.1), withB = ∅, induces an algebra homomorphism End(F ′B′)→
End(FB′) which, by (2.2) is compatible with the maps End(FB) → End(FB′)
and End(F ′B)→ End(F ′B′) for any B ⊂ B′. As pointed out in [TL4, 3.3], this
condition is too restrictive and will be weakened in the next paragraph.
2.4. Morphisms of D–categories.
Definition. A morphism of D–categories C, C ′, with C∅ = C ′∅, is the datum
of
• for any B ⊆ D a functor HB : CB → C ′B• for any B ⊆ B′ and F ∈ Mns(B,B′), a natural transformation
CB′HB′
//
FBB′
C ′B′
F ′BB′
γFBB′
z
CBHB
// C ′B
such that
• H∅ = id
• γFBB = idHB
52 2. QUASI–COXETER CATEGORIES
• for any B ⊆ B′ ⊆ B′′, F ∈ Mns(B,B′), G ∈ Mns(B′, B′′),
γFBB′ γGB′B′′ = γF∨GBB′′
Remark. For any F ∈ Mns(B′), the natural transformation γFB′ induces
an algebra homomorphism ΨFB′ : End(F ′B′)→ End(FB′) such that the following
diagram commutes for any B ∈ F
End(F ′B′)ΨFB′// End(FB′)
End(F ′B)ΨFBB//
OO
End(FB)
OO
In particular, the collection of homomorphisms ΨFD defines a morphism of
D–algebras End(F ′D)→ End(FD) in the sense of [TL4, 3.4].
Remark. The above definition may be rephrased as follows. Let M(D) be
the category with objects the subdiagramsB ⊆ D and morphisms Hom(B′, B) =
Mns(B′, B), with composition given by union and lift (cf. 1.6). There is a
forgetful functor M(D) → I(D) which is the identity on objects and maps
F ∈ Mns(B′, B) to the inclusion B ⊆ B′. Given two D–categories C, C ′ :
I(D) → Cat a morphism C → C ′ as defined above coincides with a morphism
of the functors M(D)→ Cat given by the composition
M(D) // I(D)C//
C′// Cat
2.5. Quasi–Coxeter categories.
2. QUASI–COXETER CATEGORIES 53
Definition. A labeling of the diagram D is the assignment of an integer
mij ∈ 2, 3, . . . ,∞ to any pair i, j of distinct vertices of D such that
mij = mji mij = 2
if and only if i ⊥ j.
Let D be a labeled diagram.
Definition. The generalized braid group of type D BD is the group gen-
erated by elements Si labeled by the vertices i ∈ D with relations
SiSj · · ·︸ ︷︷ ︸mij
= SjSi · · ·︸ ︷︷ ︸mij
for any i 6= j such that mij <∞. We shall also refer to BD as the braid group
corresponding to D.
Definition. A quasi–Coxeter category of type D
C = (CB, FBB′, ΦFG, Si)
is the datum of
• a D–category C = (CB, FBB′)
• for any F ,G in Mns(B,B′), a natural transformation
ΦFG ∈ Aut(FBB′)
• for any vertex i ∈ V(D), an element
Si ∈ Aut(Fi)
54 2. QUASI–COXETER CATEGORIES
satisfying the following conditions
• Orientation. For any F ,G ∈ Mns(B,B′),
ΦGF = Φ−1FG
• Transitivity. For any F ,G,H ∈ Mns(B,B′),
ΦFG ΦGH = ΦFH
• Factorization. The assignment
Φ : Mns(B,B′)2 → Aut(FB′B)
is compatible with the embedding
∪ : Mns(B,B′)×Mns(B′, B′′)→ Mns(B,B′′)
for any B′′ ⊂ B′ ⊂ B, i.e., the diagram
Mns(B,B′)2 ×Mns(B′, B′′)2
∪
Φ×Φ// Aut(FB′′B′)× Aut(FB′B)
Mns(B,B′′)2Φ
// Aut(FB′′B)
is commutative.
• Braid relations. For any pairs i, j of distinct vertices of B, such
that 2 < mij < ∞, and F ,G in Mns(B) such that i ∈ F , j ∈ G, the
following relations hold in End(FB)
Ad(ΦGF)(Si) · Sj · · · = Sj · Ad(ΦGF)(Si) · · ·
2. QUASI–COXETER CATEGORIES 55
where, by abuse of notation, we denote by Si its image in End(FB)
and the number of factors in each side equals mij.
The elements Si will be referred to as local monodromies.
Remark. The factorization property implies the support and forgetful
properties of [TL4, 3.12].
• Support. For any elementary pair (F ,G) in Mns(B,B′), let S =
supp(F ,G), Z = z supp(F ,G) ⊆ D. Then there are uniquely defined
maximal nested sets F , G ∈ Mns(Z, S) such that
ΦFG = idBZ ΦF G idB′S
where the expression above denotes the composition of natural trans-
formations
CB′
FBB′
FBB′
ΦGF+3
CB′
FSB′
CS
FZS
FZS
ΦGF+3=
CZFBZ
CB CB
• Forgetfulness. For any equivalent elementary pairs (F ,G), (F ′,G ′)
in Mns(B,B′)
ΦFG = ΦF ′G′
56 2. QUASI–COXETER CATEGORIES
Remark. To rephrase the above definition, consider the 2–category qC(D)
obtained by adding to M(D) a unique 2–isomorphism ϕBB′
GF : F → G for any
pair of 1–morphisms F ,G ∈ Mns(B′, B), with the compositions
ϕBB′
HG ϕBB′
GF = ϕBB′
HF and ϕBB′
F2G2 ϕB′B′′F1G1
= ϕBB′′
F2∪F1 G2∪G1
where F ,G,H ∈ Mns(B′, B), B ⊂ B′ ⊆ B′′ and F1,G1 ∈ Mns(B′′, B′), F2,G2 ∈
Mns(B′, B). There is a unique functor qC(D) → I(D) extending M(D) →
I(D), and a quasi–Coxeter category is the same as a 2–functor qC(D)→ Cat
fitting in a diagram
qC(D) //
##
Cat
I(D)
<<
Note that, for any B ⊂ B′, the category HomqC(D)(B′, B) is the 1–groupoid of
the De Concini–Procesi associahedron on B′/B [TL4].
2.6. Morphisms of quasi–Coxeter categories.
Definition. A morphism of quasi–Coxeter categories C, C ′ of type D is a
morphism (H, γ) of the underlying D–categories such that
• For any i ∈ B, the corresponding morphism Ψi : End(F ′i )→ End(Fi)
satisfies
Ψi(S′i) = Si
• For any elementary pair (F ,G) in Mns(B,B′),
HB(ΦFG) γFBB′ (Φ′GF)HB′ = γGBB′
2. QUASI–COXETER CATEGORIES 57
in Nat(F ′BB′ HB′ , HB FBB′), as in the diagram
C ′B′
Φ′FG//
55
HB′
CB′
ΦFG+3 γGrz
γF
C ′B55
HB
CB
Remark. Note that the above condition can be alternatively stated in
terms of morphisms ΨF as the identity
ΨG Ad(ΦGF) = Ad(Φ′GF) ΨF
2.7. Strict monoidal D–categories.
Definition. A strict monoidal D–category C = (CB, FBB′, JBB′)
is a D–category C = (CB, FBB′) where
• for any B ⊆ D, (CB,⊗B) is a strict monoidal category
• for any B ⊆ B′, the functor FBB′ is endowed with a tensor structure
JBB′
with the additional condition that, for every B ⊆ B′ ⊆ B′′, JBB′ JB′B′′ =
JBB′′ .
Remark. The tensor structure JB induces on End(FB) a coproduct ∆B :
End(FB)→ End(F 2B), where F 2
B := ⊗ (FB FB), given by
gV V ∈CB 7→ ∆B(g)VW := Ad(JBVW )(gV⊗W )V,W∈CB
58 2. QUASI–COXETER CATEGORIES
Moreover, for any B ⊆ B′, End(FB) is a subbialgebra of End(FB′), i.e., the
following diagram is commutative
End(FB)
∆B// End(F 2
B)
End(FB′)∆B′
// End(F 2B′)
Remark. Note that a strict monoidal D–category can be thought of as
functor
C : I(D)→ Cat⊗0
where Cat⊗0 denotes the 2–category of strict monoidal category, with monoidal
functors and gauge transformations.
Definition. A morphism of strict monoidal D–categories is a natural
transformation of the corresponding 2–functors M(D) → Cat⊗0 , obtained by
composition with M(D)→ I(D).
2.8. Monoidal D–categories.
Definition. A monoidal D–category
C = ((CB,⊗B,ΦB), FBB′, JFBB′)
is the datum of
• A D–category ((CB, FBB′) such that each (CB,⊗B,ΦB) is a ten-
sor category, with C∅ a strict tensor category, i.e., Φ∅ = id.
• for any pair B ⊆ B′ and F ∈ Mns(B,B′), a tensor structure JBB′
F on
the functor FBB′ : CB′ → CB
2. QUASI–COXETER CATEGORIES 59
with the additional condition that, for any B ⊆ B′ ⊆ B′′, F ∈ Mns(B′′, B′),
G ∈ Mns(B′, B),
JGBB′ JFB′B′′ = JF∪GBB′′
Remark. The usual comparison with the algebra of endomorphisms leads
to a collection of bialgebras (End(FB),∆F , ε) endowed with multiple coprod-
ucts, indexed by Mns(B).
Remark. A monoidal D–category can be thought of as a functor M(D)→
Cat⊗ fitting in a diagram
M(D)
// Cat⊗
w
I(D) // Cat
Accordingly, a morphism of monoidal D–categories is one of the corresponding
functors.
M(D)
C))
C′
99
Cat⊗
2.9. Fibered monoidal D–categories. We shall often be concerned
with monoidal D–categories such that the underlying categories (CB,⊗B) are
strict, and the functors FBB′ : (CB′ ,⊗B′)→ (CB,⊗B) are tensor functors. This
may be described in terms of the category M(D) as follows. Let DCat⊗ be
the 2–category of Drinfeld categories, that is strict tensor categories (C,⊗)
endowed with an additional associativity constraint Φ making (C,⊗,Φ) a
monoidal category. There is a canonical forgetful 2–functor DCat⊗ → Cat⊗0 .
60 2. QUASI–COXETER CATEGORIES
We shall say that a monoidal D–category fibers over a strict monoidal D–
category if the corresponding functor M(D)→ Cat⊗ maps into DCat⊗ and fits
in a commutative diagram
M(D)
// DCat⊗
v~
I(D) // Cat⊗0
In this case, the coproduct ∆F on a bialgebra End(FB) is the twist of a reference
coassociative coproduct ∆0 on End(FD) such that ∆0 : End(FB)→ End(F 2B).
2.10. Braided monoidal D–categories.
Definition. A braided monoidal D–category
C = ((CB,⊗B,ΦB, βB), (FBB′ , JFBB′)
is the datum of
• a monoidal D–category ((CB,⊗B,ΦB), (FBB′ , JFBB′)
• for every B ⊆ D, a commutativity constraint βB in CB, defining a
braiding in (CB,⊗B,ΦB).
Remark. Note that the tensor functors (FBB′ , JFBB′) : CB′ → CB are not
assumed to map the commutativity constraint βB′ to βB.
Definition. A morphism of braided monoidal D–categories from C to C ′
is a morphism of the underlying monoidal D–categories such that the functors
HB : CB → C ′B are braided tensor functors.
2. QUASI–COXETER CATEGORIES 61
Remark. The fact that HB are braided tensor functors automatically
implies that
Ψ⊗2F ((RB)JF ) = (R′B)J ′F
in analogy with [TL4], where RB = (12) βB. We are assuming that C∅ = C ′∅is a symmetric strict tensor category.
2.11. Quasi–Coxeter braided monoidal categories.
Definition. A quasi–Coxeter braided monoidal category of type D
C = ((CB,⊗B,ΦB, βB), (FBB′ , JFBB′), ΦFG, Si)
is the datum of
• a quasi–Coxeter category of type D,
C = (CB, FBB′, ΦFG, Si)
• a braided monoidal D–category
C = ((CB,⊗B,ΦB, βB), (FBB′ , JBB′
F ))
satisfying the following conditions
• for any B ⊆ B′, and G,F ∈ Mns(B,B′), the natural transformation
ΦFG ∈ Aut(FBB′) determines an isomorphism of tensor functors from
(FBB′ , JGBB′) to (FBB′ , J
FBB′), that is for any V,W ∈ CB′ ,
(ΦGF)V⊗W (JFBB′)V,W = (JGBB′)V,W ((ΦGF)V ⊗ (ΦGF)W )
62 2. QUASI–COXETER CATEGORIES
• for any i ∈ D, the following holds:
∆i(Si) = (Ri)21Ji· (Si ⊗ Si)
A morphism of quasi–Coxeter braided monoidal categories of type D is a
morphism of the underlying quasi–Coxeter categories and braided monoidal
D–categories.
Remark. A quasi–Coxeter braided monoidal category of type D deter-
mines a 2–functor qC(D)→ Cat⊗ fitting in a diagram
qC(D)
// Cat⊗
w
I(D) // Cat
Note however that this functor does not entirely capture the quasi–Coxeter
braided monoidal category since it does not encode the commutativity con-
straints βB and automorphisms Si.
CHAPTER 3
An equivalence of quasi–Coxeter categories
In this chapter, we establish an equivalence of quasi–Coxeter braided monoidal
categories for Ug[[~]] and U~g, where g is a Kac–Moody algebra endowed with
a D–algebra structure. The construction is based on the equivalence, defined
by Etingof and Kazhdan in [EK6], between the categoryO for Ug[[~]] and U~g.
This amounts essentially to building a suitable tensor structure on the
restriction functors and natural transformations as described in 2.6. Our con-
struction works more generally for an inclusion (gD, gD,−, gD,+) ⊂ (g, g−, g+)
of Manin triples over a field k of characteristic zero, so we adopt this setting
until Section 6, where we apply the results to the Kac–Moody case.
In Section 1, we start with a short review the construction of the Etingof–
Kazhdan quantization functor and the associated equivalence of categories,
following [EK1, EK6]. In Section 2, we modify this construction by using
generalized Verma modules L−, N+, and we obtain a relative fiber functor
Γ : DΦ(g) → DΦD(gD). In Section 3, we define the quantum generalized
Verma modules L~− and N~+. Using suitably defined Props we then show, in
Section 4 that these are isomorphic to the EK quantization of their classical
counterparts. In Section 5, we use these results to show that, for any given
chain of Manin triples ending in a given g, there exists a natural transformation
63
64 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
that preserves the given chain. Finally, in Section 6, we apply these results
to the case of a Kac–Moody algebra g and obtain the desired tranport of its
quasi–Coxeter quasitriangular quasibialgebra structure to the completion of
Ug[[~]] with respect to category O, integrable modules.
1. Etingof-Kazhdan quantization
We review in this section the results obtained in [EK1, EK6]. More
specifically, we follow the quantization of Lie bialgebras given in [EK1, Part
II] and the case of generalized Kac–Moody algebras from [EK6].
1.1. Topological vector spaces. The use of topological vector spaces
is needed in order to deal with convergence issues related to duals of infinite
dimensional vector spaces and tensor product of such spaces.
Let k be a field of characteristic zero with the discrete topology and V a
topological vector space over k. The topology on V is linear if open subspaces
in V form a basis of neighborhoods of zero. Let V be endowed with a linear
topology and pV the natural map
pV : V → lim(V/U)
where the limit is taken over the open subspaces U ⊆ V . Then V is called
separated if pV is injective and complete if pV is surjective. Throughout this
section, we shall call topological vector space a linear, complete, separated
topological space.
1. ETINGOF-KAZHDAN QUANTIZATION 65
If U is an open subspace of a topological vector space V , then the quotient
V/U is discrete. It is then possible, given two topological vector spaces V and
W , to define the topological tensor product as
V ⊗W := limV/V ′ ⊗W/W ′
where the limit is take over open subspaces of V and W . We then denote by
Homk(V,W ) the topological vector space of continuous linear operators from
V to W equipped with the weak topology. Namely, a basis of neighborhoods
of zero in Homk(V,W ) is given by the collection of sets
Y (v1, . . . , vn,W1, . . . ,Wn) := f ∈ Homk(V,W ) | f(vi) ∈ Wi, i = 1, . . . , n
for any n ∈ N, vi ∈ V and Wi open subspace in W for all i = 1, . . . , n.In par-
ticular, if W = k with the discrete topology, the space V ∗ = Homk(V, k) has
a basis of neighborhoods of zero given by orthogonal complements of finite–
dimensional subspaces in V . When V is finite–dimensional, V ∗ coincides with
the linear dual and the weak topology coincides with the discrete topology.
The space of formal power series in ~ with coefficients in a topological
vector space V , V [[~]] = V ⊗k[[~]], is also a complete topological space with a
natural structure of a topological k[[~]]-module. A topological k[[~]]-module is
complete if it is isomorphic to V [[~]] for some complete V . The additive cat-
egory of complete k[[~]]-module, denoted A, where morphisms are continuous
k[[~]]-linear maps, has a natural symmetric monoidal structure. Namely, the
tensor product on A is defined to be the quotient of the tensor product V ⊗W
by the image of the operator ~ ⊗ 1 − 1 ⊗ ~. This tensor product will be still
66 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
denoted by ⊗. There is an extension of scalar functor from the category of
topological spaces to A, mapping V to V [[~]]. This functor respects the tensor
product, i.e., (V ⊗W )[[~]] is naturally isomorphic to V [[~]]⊗W [[~]].
1.2. Equicontinuous modules. Fix a topological Lie algebra g.
Definition. Let V be a topological vector space. We say that V is an
equicontinuous g-module if:
• the map πV : g → Endk V is a continuous homomorphism of topo-
logical Lie algebras;
• πV (g)g∈g is an equicontinuous family of linear operators,i.e., for
any open subspace U ⊆ V , there exists U ′ such that πV (g)U ′ ⊂ U
for all g ∈ g.
Clearly, a topological vector space with a trivial g-module structure is an
equicontinuous g-module. Moreover, given equicontinuous g-modules V,W,U ,
the tensor product V ⊗W has a natural structure of equicontinuous g-module
and (V ⊗W )⊗U is naturally identified with V ⊗(W ⊗U). The category of
equicontinuous g-modules is then a symmetric monoidal category, with braid-
ing defined by permutation of components. We denote this category by Repeq g.
1.3. Lie bialgebras and Manin triples. A Manin triple is the data of
a Lie algebra g with
• a nondegenerate invariant inner product 〈, 〉;
• isotropic Lie subalgebras g± ⊂ g;
such that
• g = g+ ⊕ g− as vector space;
1. ETINGOF-KAZHDAN QUANTIZATION 67
• the inner product defines an isomorphism g+ ' g∗−;
• the commutator of g is continuous with respect to the topology ob-
tained by putting the discrete and the weak topology on g−, g+ re-
spectively.
Under these assumptions, the commutator on g+ ' g∗− induces a cobracket
on g−, satisfying the cocycle condition [D1]. Therefore, g− is canonically
endowed with a Lie bialgebra structure. Notice that, in absolute generality,
g+ is only a topological Lie bialgebra, i.e., δ(g+) ⊂ g+⊗g+. The inner product
also gives rise to an isomorphism of vector spaces g− ' g∗∗− ' g∗+, where the
latter is the continuous dual, though this isomorphism does not respect the
topology. Conversely, every Lie bialgebra a defines a Manin triple (a⊕a∗, a, a∗).
1.4. Verma modules. In [EK1], Etingof and Kazhdan constructed two
main examples of equicontinuous g-modules in the case when g belongs to a
Manin triple (g, g+, g−). The modules M±, defined as
M+ = Indgg− k M− = Indg
g+k
are freely generated over U(g±) by a vector 1± such that g∓1±. Therefore, they
are naturally identified, as vector spaces, to U(g±) via x1± → x. The modules
M− and M∗+, with appropriate topologies, are equicontinuous g-modules.
The module M− is an equicontinuous g-module with respect to the discrete
topology. The topology on M+ comes, instead, from the identification of vector
spaces
M+ ' U(g+) =⋃n≥0
U(g+)n
68 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
where U(g+)n is the set of elements of degree at most n. The topology on
U(g+)n is defined through the linear isomorphism
ξn :n⊕j=0
Sjg+ → U(g+)n
where Sjg+ is considered as a topological subspace of (g⊗j− )∗, embedded with
the weak topology. Finally, U(g+) is equipped with the topology of the col-
imit. Namely, a set U ⊆ U(g+) is open if and only if U ∩ U(g+)n is open for
all n. With respect to the topology just described, the action of g on M+ is
continuous.
Consider now the vector space of continuous linear functionals on M+
M∗+ = Homk(M+, k) ' colim Homk(U(g+)n, k)
It is natural to put the discrete topology on U(g+)∗n, since, as a vector space,
U(g+)∗n 'n⊕j=0
Sjg∗+ 'n⊕j=0
Sjg− ' U(g−)n
We then consider on M∗+ the topology of the limit. This defines, in particular,
a filtration by subspaces (M∗+)n satisfying
0→ (M∗+)n →M∗
+ → (U(g+)n)∗ → 0
and such that M∗+ = limM∗
+/(M∗+)n. The topology of the limit on M∗
+ is, in
general, stronger than the weak topology of the dual. Since the action of g on
M+ is continuous, M∗+ has a natural structure of g–module. In particular, this
is an equicontinuous g–action.
1. ETINGOF-KAZHDAN QUANTIZATION 69
1.5. Drinfeld category. The natural embedding
g− ⊗ g∗− ⊂ Endk(g−)
induces a topology on g−⊗g∗− by restriction of the weak topology in Endk(g−).
With respect to this topology, the image of g− ⊗ g∗− is dense in Endk(g−) and
the topological completion g− ⊗ g∗− is identified with Endk(g−). Under this
identification, the identity operator defines an element r ∈ g−⊗g∗−.
Given two equicontinuous g–modules V,W , the map
πV ⊗ πW : g− ⊗ g∗− → Endk(V ⊗W )
naturally extends to a continuous map g−⊗g∗− → Endk(V ⊗W ). Therefore,
the Casimir operator
Ω = r + rop ∈ g−⊗g∗− ⊕ g∗−⊗g−
defines a continuous endomorphism of V ⊗W , ΩVW = (πV ⊗πW )(Ω), commut-
ing with the action of g.
Following [D2], it is possible to define a structure of braided monoidal
category on the category of deformed equicontinuous g–module, depending on
the choice of a Lie associator Φ, the bifunctor ⊗ and the Casimi operator Ω.
The commutativity constraint is explicitly defined by the formula
βVW = (12) e~2
ΩVW ∈ Homg(V ⊗W,W ⊗V )[[~]]
70 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
We denote this braided tensor category braided tensor category DΦ(Ug). The
category of equicontinuous g–modules is equivalent to the category of Yetter-
Drinfeld module over g−, YD(g−). The equivalence holds at the level of tensor
structure induced by the choice of an associator Φ,
DΦ(Ug) ' YDΦ(Ug−[[~]])
1.6. Verma modules. The modules M± are identified, as vector spaces,
with the enveloping universal algebras Ug±. Their comultiplications induce
the Ug–intertwiners i± : M± → M±⊗M±, mapping the vectors 1± to the g∓-
invariant vectors 1± ⊗ 1±.
For any f, g ∈ M∗+, consider the linear functional M+ → k defined by
v 7→ (f⊗g)(i+(v)). This functional defines a map i∗+ : M∗+⊗M∗
+ →M∗+, which
is continuous and extends to a morphism in Rep g[[~]], i∗+ : M∗+⊗M∗
+ → M∗+.
The pairs (M−, i−) and (M∗+, i∗+) form, respectively, a coalgebra and an algebra
object in DΦ(Ug).
For any V ∈ DΦ(Ug), the vector space Homg(M−,M∗+⊗V ) is naturally
isomorphic to V , as topological vector space, through the isomorphism f 7→
(1+ ⊗ 1)f(1−).
1.7. The fiber functor and the EK quantization. We will now recall
the main results from [EK1, EK2]. Where no confusion is possible, we will
abusively denote ⊗ by ⊗. Let then F be the functor
F : DΦ(Ug)→ A F (V ) = HomDΦ(Ug)(M−,M∗+ ⊗ V )
1. ETINGOF-KAZHDAN QUANTIZATION 71
There is a natural transformation
J ∈ Nat(⊗ (F F ), F ⊗)
defined, for any v ∈ F (V ), w ∈ F (W ), by
JVW (v ⊗ w) = (i∨+ ⊗ 1⊗ 1)A−1β−123 A(v ⊗ w)i−
where A is defined as a morphism
(V1 ⊗ V2)⊗ (V3 ⊗ V4)→ V1 ⊗ ((V2 ⊗ V3)⊗ V4)
by the action of (1⊗ Φ2,3,4)Φ1,2,34.
Theorem. The natural transformation J is invertible and defines a tensor
structure on the functor F .
The tensor functor (F, J) is called fiber functor. The algebra of endo-
morphisms of F is therefore naturally endowed with a topological bialgebra
structure, as described in the previous section.1
The object F (M−) ∈ A has a natural structure of Hopf algebra, defined
by the multiplication
m : F (M−)⊗ F (M−)→ F (M−) m(x, y) = (i∨+ ⊗ 1)Φ−1(1⊗ y)x
1By topological bialgebra we do not mean topological over k[[~]]. We are instead referringto the fact that the algebra End(F ) has a natural comultiplication ∆ : End(F )→ End(F 2),where End(F 2) can be interpreted as an appropriate completion of End(F )⊗2.
72 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
and the comultiplication
∆ : F (M−)→ F (M−)⊗ F (M−) ∆(x) = J−1(F (i−)(x))
The algebra F (M−) is naturally isomorphic as a vector space with M−[[~]] '
Ug−[[~]] and
Theorem. The algebra UEK
~ g− = F (M−) is a quantization of the algebra
Ug−.
In [EK2], it is shown that this construction defines a functor
QEK : LBA(k)→ QUE(K)
where LBA(k) denotes the category of Lie bialgebras over k and QUE(K) de-
notes the category of quantum universal enveloping algebras over K = k[[~]].
Another important result in [EK2] states the invertibility of the functor QEK.
The map
m− : UEK
~ g− → End(F ) m−(x)V (v) = (i∨+ ⊗ 1)Φ−1(1⊗ v)x
where V ∈ YDΦ(Ug−[[~]]) and v ∈ F (V ), is, indeed, an inclusion of Hopf
algebras. The map m− defines an action of UEK
~ g− on F (V ). Moreover, the
map
F (V )→ F (M−)⊗ F (V ) v 7→ RJ(1⊗ v)
where RJ denotes the twisted R–matrix, defines a coaction of UEK
~ g− on F (V )
compatible with the action, therefore
1. ETINGOF-KAZHDAN QUANTIZATION 73
Theorem. The fiber functor F : YDΦ(Ug−[[~]]) → A lifts to an equiva-
lence of braided tensor categories
F : YDΦ(Ug−[[~]])→ YD(UEK
~ g−)
1.8. Generalized Kac-Moody algebras. Denote by k a field of char-
acteristic zero. We recall definitions from [Ka] and [EK6]. Let A = (aij)i,j∈I
be an n× n symmetrizable matrix with entries in k, i.e. there exists a (fixed)
collection of nonzero numbers dii∈I such that diaij = djaji for all i, j ∈ I.
Let (h,Π,Π∨) be a realization of A. It means that h is a vector space of di-
mension 2n− rank(A), Π = α1, . . . , αn ⊂ h∗ and Π∨ = h1, . . . , hn ⊂ h are
linerly independent, and (αi, hj) = aji.
Definition. The Lie algebra g = g(A) is generated by h, ei, fii∈I with
defining relations
[h, h′] = 0 h, h′ ∈ h; [h, ei] = (αi, h)ei
[h, fi] = −(αi, h)fi; [ei, fj] = δijhi
There exists a unique maximal ideal r in g that intersects h trivially. Let
g := g/r. The algebra g is called generalized Kac-Moody algebra. The Lie alge-
bra g is graded by principal gradation deg(ei) = 1, deg(fi) = −1, deg(h) = 0,
and the homogenous component are all finite-dimentional.
Let us now choose a non–degenerate bilinear symmetric form on h such
that 〈h, hi〉 = d−1i (αi, h). Following [Ka], there exists a unique extension of
the form 〈, 〉 to an invariant symmetric bilinear form on g. For this extension,
74 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
one gets 〈ei, fj〉 = δijd−1i . The kernel of this form on g is r, therefore it de-
scends to a non–degenerate bilinear form on g.
Let n±, b± be the nilpotent and the Borel subalgebras of g, i.e., n± are
generated by ei, fi, respectively, and b± := n±⊕h. Since [n±, h] ⊂ n±, we
get Lie algebra maps ¯ : b± → h and we can consider the embeddings of Lie
subalgebras η± : b± → g⊕ h given by
η±(x) = (x,±x)
Define the inner product on g⊕ h by 〈, 〉g⊕h = 〈, 〉g − 〈, 〉h.
Proposition. The triple (g⊕ h, b+, b−) with inner product 〈−,−〉g⊕h and
embeddings η± is a graded Manin triple.
Under the embeddings η±, the Lie subalgebras b± are isotropic with re-
spect to 〈, 〉g⊕h. Since 〈, 〉g and 〈, 〉h are invariant symmetric non–degenerate
bilinear form, so is 〈, 〉g⊕h.
The proposition implies that g ⊕ h, b+, b− are Lie bialgebras. Moreover,
b∗+ ' bcop− as Lie bialgebras (where b∗+ :=⊕
(b+)∗n denotes the restricted dual
space and by cop we mean the opposite cocommutator). The Lie bialgebra
structures on b± are then described by the following formulas:
δ(h) = 0, h ∈ h ⊂ b±;
δ(ei) =di2
(ei ⊗ hi − hi ⊗ ei) =di2ei ∧ hi; δ(fi) =
di2fi ∧ hi
1. ETINGOF-KAZHDAN QUANTIZATION 75
The Lie subalgebra (0, h) | h ∈ h is therefore an ideal and a coideal in g⊕h.
Thus, the quotient g = (g⊕h)/h is also a Lie bialgebra with Lie subbialgebras
b± and the same cocommutator formulas.
1.9. Quantization of Kac–Moody algebras and category O. In
[EK6], Etingof and Kazhdan proved that, for any symmetrizable irreducible
Kac-Moody algebra g, the quantization UEK
~ g is isomorphic with the Drinfeld–
Jimbo quantum group U~g.
In particular, they construct an isomorphism of Hopf algebras U~b+ '
UEK
~ b+, inducing the identity on Uh[[~]], where b+ is the Borel subalgebra and
h is the Cartan subalgebra of g. Thanks to the compatibility with the doubling
operations
DUEK
~ b+ ' UEK
~ Db+
proved by Enriquez and Geer in [EG], the isomorphism for the Borel subalge-
bra induces an isomorphism U~g ' UEK
~ g.
Recall that the category O for g, denoted Og is defined to be the category
of all h–diagonalizable g–modules V , whose set of weights P(V ) belong to a
union of finitely many cones
D(λs) = λs +∑i
Z≥0αi λs ∈ h∗, s = 1, ..., r
76 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
and the weight subspaces are finite–dimensional. We denote by Og[[~]] the
category of deformation g–representations,i.e., representations of g on topo-
logically free k[[~]]–modules with the above properties (with weights in h∗[[~]]).
In a similar way, one defines the category OU~g: it is the category of U~g–
modules which are topologically free over k[[~]] and satisfy the same conditions
as in the classical case.
The morphism of Lie bialgebras
Db+ → g ' Db+/(h ' h∗)
gives rise to a pullback functors
Og → YD(Ub+) Og,Φ[[~]]→ YDΦ(Ub+[[~]])
where Og,Φ denotes the category Og with the tensor structure of the Drinfeld
category. Similarly, the morphism of Hopf algebras
DUEK
~ b+ → UEK
~ g ' U~g
gives rise to a pullback functor
OU~g → YD(UEK
~ b+)
Theorem. The equivalence F reduces to an equivalence of braided tensor
categories
FO : Og,Φ[[~]]→ OU~g
1. ETINGOF-KAZHDAN QUANTIZATION 77
which is isomorphic to the identity functor at the level of h–graded k[[~]]–
modules.
1.10. The isomorphism ΨEK. In [EK6], Etingof–Kazhdan showed that
the equivalence F induces an isomorphism of algebras
ΨEK : Ug[[~]]→ U~g
where
Ug = limUβ Uβ = Ug/Iβ, β ∈ NI
Iβ being the left ideal generated by elements of weight less or equal β (analo-
gously for U~g, cf. [EK6, Sec. 4] and [D5, 8.1]).
Proposition. The isomorphism ΨEK coincides with the isomorphism in-
duced by the equivalence FO, as explained in Section 2.1.
Proof. The identification of the two isomorphism is constructed in the
following way:
(a) First, we show that there is a canonical map
End(fO)→ CEnd(U)(Endg(U))
(b) There is a canonical multiplication in U , so that
(i) There is a canonical map
CEnd(U)(Endg(U))→ U
78 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
(ii) For every V ∈ O the action of Ug lifts to an action of U
Ug //
End(V )
U
<<
(c) It defines a map U → End(fO) and we have an isomorphism of alge-
bras
U ' End(fO)
A detailed proof of the equivalence is given in Appendix A Section 1.
If g is a semisimple Lie algebra, the equivalence of categories F leads to an
isomorphism of algebras
U(Db+)[[~]] ' DUEK
~ b+ =⇒ Ug[[~]] ' U~g
which is the identity modulo h. Toledano Laredo proved in [TL4, Prop. 3.5]
that such an isomorphism cannot be compatible with all the isomorphisms
Uslαi2 [[~]] ' U~slαi2 ∀i
where αi are the simple roots of g. This amounts to a simple proof that the
isomorphism ΨEK cannot be, in general, an isomorphism of D–algebras.
2. A relative Etingof–Kazhdan functor
2.1. In this section, we consider a split inclusion of Manin triples
iD : (gD, gD,+, gD,−) → (g, g+, g−)
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 79
We then define a relative version of the Verma modules M±, and use them to
prove the following
Theorem. There is a tensor functor
Γ : DΦ(Ug)→ DΦD(UgD)
canonically isomorphic, as abelian functor, to the restriction functor i∗D.
2.2. Split inclusions of Manin triples.
Definition. An embedding of Manin triples
i : (gD, gD,−, gD,+) −→ (g, g−, g+)
is a Lie algebra homomorphism i : gD → g preserving inner products, and
such that i(gD,±) ⊂ g±.
Denote the restriction of i to gD,± by i±. i± give rise to maps p± = i∗∓ :
g± → gD,±, defined via the identifications g± ' g∗∓ and gD,± ' g∗D,∓ by
〈p±(x), y〉D = 〈x, i∓(y)〉
for any x ∈ g± and y ∈ gD,∓. These map satisfy p± i± = idgD,± since, for any
x ∈ gD,±, y ∈ gD,∓,
〈p± i±(x)− x, y〉D = 〈i±(x), i∓(y)〉 − 〈x, y〉D = 0
This yields in particular a a direct sum decomposition g± = i(gD,±) ⊕ m±,
where
m± = Ker(p±) = g± ∩ i(gD)⊥
80 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
Definition. The embedding i : (gD, gD,−, gD,+) −→ (g, g−, g+) is called
split if the subspaces m± ⊂ g± are Lie subalgebras.
2.3. Split pairs of Lie bialgebras. For later use, we reformulate the
above notion in terms of bialgebras via the double construction.
Definition. A split pair of Lie bialgebras is the data of
• Lie bialgebras (a, [, ]a, δa) and (b, [, ]b, δb).
• Lie bialgebra morphisms i : a→ b and p : b→ a such that pi = ida.
Proposition. There is a one–to–one correspondence between split inclu-
sions of Manin triples and split pairs of Lie bialgebras. Specifically,
(i) If i : (gD, gD,−, gD,+) −→ (g, g−, g+) is a split inclusion of Manin
triples, then (gD,−, g−, i−, i∗+) is a split pair of Lie bialgebras.
(ii) Conversely, if (a, b, i, p) is a split pair of Lie bialgebras, then i⊕ p∗ :
(Da, a, a∗) −→ (Db, b, b∗) is a split inclusion of Manin triples.
2.4. Proof of (i) of Proposition 2.3. Given a split inclusion
i = i− ⊕ i+ : (gD, gD,−, gD,+) −→ (g, g−, g+)
we need to show that i− and i∗+ are Lie bialgebra morphisms. By assumption,
i− is a morphism of Lie algebras, and i∗+ one of coalgebras. Since i− = (i∗−)∗,
it suffices to show that p± = i∗∓ preserve Lie brackets.
We claim to this end that m± are ideals in g±. Since [m±,m±] ⊆ m± by
assumption, this amounts to showing that [i(gD,±),m±] ⊆ m±. This follows
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 81
from the fact that [i(gD,±),m±] ⊆ g±, and from
〈[i(gD,±),m±], i(gD,∓)〉 = 〈m±, [i(gD,±), i(gD,∓)]〉 ⊂ 〈m±, i(gD,±)〉+〈m±, i(gD,∓)〉
where the first term is zero since g± is isotropic, and the second one is zero by
definition of m±.
Let now X1, X2 ∈ g±, and write Xj = i±(xj) + yj, where xj ∈ gD,± and
yj ∈ m±. Since m± = Ker(p±) and p± i± = id, we have [p±(X1), p±(X2)] =
[x1, x2], while
p±[X1, X2] = p± (i±[x1, x2] + [i±x1, y2] + [y1, i±x2] + [y1, y2]) = [x1, x2]
where the last equality follows from the fact that m± is an ideal.
2.5. Proof of (ii) of Proposition 2.3. The bracket on Da is defined by
[a, φ] = ad∗(a)(φ)− ad∗(φ)(a) = −〈φ, [a,−]a〉+ 〈φ⊗ id, δa(a)〉
for any a ∈ a, φ ∈ a∗. Analogously for Db. Therefore, the equalities
〈p∗(φ)⊗ id, δb(i(a))〉 =〈φ⊗ id, (p⊗ id)(i⊗ i)δa(a)〉
=〈φ⊗ id, (id⊗i)δa(a)〉 = i(〈φ⊗ id, δa(a)〉)
and
〈p∗(φ), [i(a), b]b〉 = 〈φ, p([i(a), b]b)〉 = 〈φ, [a, p(b)]a〉
82 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
for all a ∈ a and b ∈ b, imply that the map i⊕ p∗ is a Lie algebra map. It also
respects the inner product, since for any a ∈ a, φ ∈ a∗,
〈p∗(φ), i(a)〉 = 〈φ, p i(a)〉 = 〈φ, a〉
Finally, m− = Ker(p) and m+ = Ker i∗ are clearly subalgebras.
2.6. Parabolic Lie subalgebras. Let
iD = i− ⊕ i+ : (gD, gD,−, gD,+)→ (g, g−, g+)
be a split embedding of Manin triples. We henceforth identify gD as a Lie
subalgebra of g with its induced inner product, and gD,± as subalgebras of g±
noting that, by Proposition 2.3, gD,− is a sub Lie bialgebra of g−.
The following summarizes the properties of the subspaces m± = g± ∩ g⊥D
and p± = m± ⊕ gD.
Proposition.
(i) m± is an ideal in g±, so that g± = m± o gD,±.
(ii) [gD,m±] ⊂ m±, so that p± = m± o gD are Lie subalgebras of g.
(iii) δ(m−) ⊂ m− ⊗ gD,− + gD,− ⊗m−, so that m− ⊆ g− is a coideal.
Proof. (i) was proved in 2.4. (ii) Since
〈[gD,m±], gD〉 = 〈m±, [gD, gD]〉 = 0
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 83
we have [gD,m±] ⊂ g⊥D = m− ⊕m+. Moreover,
〈[gD,m±],m±〉 = 〈gD, [m±,m±]〉 = 〈gD,m±〉 = 0
since m± is a subalgebra, and it follows that [gD,m±] ⊂ m±. (iii) is clear since
m− is the kernel of a Lie coalgebra map.
Remark. If the inclusion iD is compatible with a finite type N–grading,
then m+ ⊂ g+ is a coideal. Moreover, p± are Lie subbialgebras of g such
that the projection p± → gD is a morphism of bialgebras. Namely, a finite
type N–grading allows to define a Lie bialgebra structure on g, g+. We then
get a vector space decomposition g± = m± ⊕ gD,± and a Lie bialgebra map
g± → gD,±. It is also possible to define the Lie subalgebras
p± = m± ⊕ gD ⊂ g
If we assume the existence of a compatible grading on g and gD,i.e., preserved
by iD, then the natural maps
p± ⊂ g p± → gD
are morphisms of Lie bialgebras.
2.7. The relative Verma Modules.
Definition. Given a split embedding of Manin triples gD ⊂ g, and the
corresponding decomposition g = m− ⊕ p+, let L−, N+ be the relative Verma
modules defined by
L− = Indgp+
k and N+ = Indgm− k
84 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
Proposition. The g–modules L− and N∗+ are equicontinuous.
The description of the appropriate topologies on L− and N∗+, and the proof of
their equicontinuity will be carried out in 2.8–2.11.
2.8. Equicontinuity of L−. As vector spaces,
L− ' Um− ⊂ Ug−
so it is natural to equip L− with the discrete topology. The set of opera-
tors πL−(x)x∈g is then an equicontinuous family, and the continuity of πL−
reduces to checking that, for every element v ∈ L−, the set
Yv = b ∈ g+| b.v = 0
is a neighborhood of zero in g+. Since Um− embeds naturally in Ug− the proof
is identical to [EK1, Lemma 7.2]. We proceed by induction on the length of
v = ai1 . . . ain1−. If n = 0, then v = 1− and Yv = g+. If n > 1, then assume
v = ajw, with w = ai1 . . . ain−11− and Yw open in g+. For every x ∈ g+
x.v = x.(ajw) = [x, aj].w + (ajx).w
Call Z the subset of g+
Z = x ∈ g+| [x, aj] ∈ Yw
Z is open in g+, by continuity of bracket [, ], and clearly Z ∩ Yw ⊂ Yv.
2.9. Topology of N+. As vector spaces,
N+ = Indgm−k ' Up+ ' colimUnp+
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 85
where Unp+ denotes the standard filtration of Up+, so that
Unp+ 'n⊕
m=0
Smp+ =⊕i+j≤n
(Sig+ ⊗ SjgD,−
)We turn this isomorphism into an isomorphism of topological vector spaces,
by taking on Sig+ and SjgD,− the topologies induced by the embeddings
Sig+ → (g⊗i− )∗ and SjgD,− → g⊗jD,−
With respect to these topologies, Ump+ is closed inside Unp+ for m < n, and
we equip N+ with the direct limit topology. We shall need the following
Lemma. For any x ∈ g, the map πN+(x) : N+ → N+ is continuous.
Proof. We need to show that for any neighborhood of the origin U ⊂ N+,
there exists a neighborhood of zero U ′ ⊂ N+ such that πN+(x)U ′ ⊂ U . The
topology on N+ comes from the decomposition Up+ ' Ug+ ⊗ UgD,−, so that
an open neighborhood of zero in N+ has the form U ⊗UgD,−+Ug+⊗V , with
U open in Ug+ and V open in UgD,−. We apply the same procedure used in
[EK1, Lemma 7.3] to construct a set U ′ ⊗ UgD,−, with U ′ open in Ug+, such
that
πN+(x)(U ′ ⊗ UgD,−) ⊂ U ⊗ UgD,− ⊂ U ⊗ UgD,− + Ug+ ⊗ V
Since the topology on UgD,− is discrete, the set U ′⊗UgD,− is open in N+ and
the lemma is proved.
2.10. Topology of N∗+. As vector spaces,
N∗+ ' (Up+)∗ ' lim(Unp+)∗
86 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
Define a filtration (N∗+)n on N∗+ by
0→ (N∗+)n → (Up+)∗ → (Unp+)∗ → 0
so that N∗+ ⊃ (N∗+)0 ⊃ (N∗+)1 ⊃ · · · , and we get an isomorphism of vector
spaces
N∗+ ' limN∗+/(N∗+)n
Finally, we use the isomorphism to endow N∗+ with the inverse limit topology.
Lemma. πN∗+(x)x∈g is an equicontinuous family of operators.
Proof. Since p+ acts on N+ by multiplication,
p+(N∗+)n ⊂ (N∗+)n−1
If x ∈ m− and xi ∈ Up+ for i = 1, . . . , n, then in Ug,
xx1 · · · xn = x1 · · ·xnx−n∑i=0
x1 · · ·xi−1[xi, x]xi+1 · · ·xn
where [xi, x] ∈ g. Iterating shows that (x.f)(x1 · · ·xn) = 0 if f ∈ (N∗+)n,
so that x(N∗+)n ⊂ (N∗+)n. Then, for any neighborhood of zero of the form
U = (N∗+)n, it is enough to take U ′ = (N∗+)n+1 to get g(N∗+)n+1 ⊂ (N∗+)n.
2.11. Equicontinuity of N∗+.
Lemma. The map πN∗+ : g→ End(N∗+) is a continuous map.
Proof. Since g− is discrete, it is enough to check that, for any f ∈ N∗+
and n ∈ N, the subset
Y (f, n) = b ∈ g+| b.f ∈ (N∗+)n
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 87
is open in g+, i.e.
bi.f ∈ (N∗+)n for a.a. i ∈ I
Since f ∈ N∗+ ' limN∗+/(N∗+)n, we have f = fn where fn is the class of f
modulo (N∗+)n. Therefore bi.f ∈ (N∗+)n iff
(bi.f)n = bi.fn+1 = 0
Now, for any x1 · · ·xn ∈ Unp+, we have
bi.fn+1(x1 · · ·xn) = −fn+1(bix1 · · ·xn) = 0
for a.a. i ∈ I and the lemma is proved (it is enough to exclude the indices
corresponding to the generators involved in the expression of fn+1).
As a vector spaces, we can identify
p∗+ = g∗+ ⊕ g∗D,− ' g− ⊕ gD,+ = p−
We can give as a basis for p+ and p−
p+ ⊃ bii∈I , arr∈I(D) p− ⊃ aii∈I , brr∈I(D)
and obvious relations
(bi, aj) = δij (bi, br) = 0
(ar, aj) = 0 (ar, bs) = δrs
88 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
with i, j ∈ I, r, s ∈ I(D). We can then identify fn+1 with an element in
Un+1p−. Call Tn+1(f) the set of indices of all ai involved in the expression of
fn+1. Excluding these finite set of indices we get the result.
2.12. Coalgebra structure on L−, N+. Define g–module maps
i− : L− → L−⊗L− and i+ : N+ → N+⊗N+
by mapping 1∓ to 1∓⊗1∓. Note that, under the identification L− ' Um− and
N+ ' Up+, i∓ correspond to the coproduct on Um− and Up+ respectively.
Following [D3, Prop. 1.2], we consider the invertible element T ∈ (Ug⊗Ug)[[~]]
satisfying relations:
S⊗3(Φ321) · (T ⊗ 1) · (∆⊗ 1)(T ) = (1⊗ T )(1⊗∆)(T ) · Φ
T∆(S(a)) = (S ⊗ S)(∆(a))T
LetN∗+ be as before and f, g ∈ N∗+. Consider the linear functional in Homk(N+, k)
defined by
v 7→ (f ⊗ g)(T · i+(v))
This functional is continuous, so it belongs to N∗+ and allow us to define the
map
i∨+ ∈ Homk(N∗+ ⊗N∗+, N∗+)[[~]] , i∨+(f ⊗ g)(v) = (g ⊗ f)(T · i+(v))
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 89
This map is continuous and extends to a map from N∗+⊗N∗+ to N∗+. For any
a ∈ g, we have
i∨+(a(f ⊗ g))(v) = (f ⊗ g)((S ⊗ S)(∆(a))T · i+(v)) =
= (f ⊗ g)(T∆(S(a)) · i+(v)) =
= i∨+(f ⊗ g)(S(a).v) = (a.i∨+(f ⊗ g))(v)
and then i∨+ ∈ Homg(N∗+⊗N∗+, N∗+)[[~]].
The following shows that L− and N+ are coalgebra objects in the Drinfeld
categories of g–modules and (g, gD)–bimodules respectively.
Proposition. The following relations hold
(i) Φ(i− ⊗ 1)i− = (1⊗ i−)i−.
(ii) i∨+(1⊗ i∨+)Φ = i∨+(i∨+ ⊗ 1)S⊗3(Φ−1D )ρ
where (−)ρ denotes the right gD–action on N∗+.
Proof. We begin by showing that
Φ(1⊗3− ) = 1⊗3
− and Φ(1⊗3+ ) = ΦD(1⊗3
+ ) (2.1)
To prove the first identity, it is enough to notice that, since g+1− = 0 and
Ω =∑
(ai ⊗ bi + bi ⊗ ai),
Ωij(1⊗3− ) = 0
90 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
Then Φ(1⊗3− ) = 1⊗3
− . To prove the second one, we notice that m−1+ = 0 and
that we can rewrite
Ω =∑j∈ID
(aj⊗bj+bj⊗aj)+∑i∈I\ID
(ai⊗bi+bi⊗ai) = ΩD+∑i∈I\ID
(ai⊗bi+bi⊗ai)
where ajj∈ID is a basis of gD,− and bjj∈ID is the dual basis of gD,+. Then
Ωij(1⊗3+ ) = ΩD,ij(1
⊗3+ )
and, since for any element x ∈ gD, the right and the left gD-action coincide
on 1+, i.e. x.1+ = 1+.x, we have
Ωij(1⊗3+ ) = (1⊗3
+ )ΩD,ij
and consequently Φ(1⊗3+ ) = ΦD(1⊗3
+ ).
To prove (i), note that since the comultiplication in Um− is coassociative,
we have (i−⊗1)i− = (1⊗ i−)i−. We therefore have to show that Φ(i−⊗1)i− =
(1 ⊗ i−)i−. This is an obvious consequence of (2.1) and the fact that m− is
generated by 1−.
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 91
To prove (ii), consider v ∈ N+,
i∨+(1⊗ i∨+)(Φ(f ⊗ g ⊗ h))(v) =
= (h⊗ g ⊗ f)((S⊗3(Φ321) · (T ⊗ 1) · (∆⊗ 1)(T )) · (i+ ⊗ 1)i+(v)) =
= (h⊗ g ⊗ f)((1⊗ T )(1⊗∆)(T ) · Φ(i+ ⊗ 1)i+(v)) =
= (h⊗ g ⊗ f)((1⊗ T )(1⊗∆)(T )(1⊗ i+)i+(v)ΦD) =
= (S⊗3(ΦD)ρ(h⊗ g ⊗ f))((1⊗ T )(1⊗∆)(T )(1⊗ i+)i+(v)) =
= i∨+(i∨+ ⊗ 1)(S⊗3(Φ321D )ρ(f ⊗ g ⊗ h))(v) =
= i∨+(i∨+ ⊗ 1)S⊗3(Φ−1D )ρ(f ⊗ g ⊗ h)(v)
and (ii) is proved.
2.13. The fiber functor over gD. To any representation V [[~]] ∈ RepUg[[~]],
we can associate the k[[~]]–module
Γ(V ) = Homg(L−, N∗+⊗V )[[~]]
where Homg is the set of continuous homomorphisms, equipped with the weak
topology. The right gD–action on N∗+ endows Γ(V ) with the structure of a left
gD–module.
Proposition. The complete vector space Homg(L−, N∗+⊗V ) is isomorphic
to V as equicontinous gD–module. The isomorphism is given by
αV : f 7→ (1+ ⊗ 1)f(1−)
for any f ∈ Homg(L−, N∗+⊗V ).
92 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
Proof. By Frobenius reciprocity, we get an isomorphism
Homg(L−, N∗+⊗V ) ' Homp+(k, N∗+⊗V ) ' Homk(k, V ) ' V
given by the map
f 7→ (1+ ⊗ 1)f(1−)
For f ∈ Γ(V ) and x ∈ UgD, x.f ∈ Γ(V ) is defined by
x.f = (S(x)ρ ⊗ id) f
For any x ∈ UgD, we have
∑i,j
x(1)i fj ⊗ x(2)
i vj = ε(x)f(1−)
where ∆(x) =∑
i x(1)i ⊗ x
(2)i and f(1−) =
∑j fj ⊗ vj. Using the identity
1⊗ x =∑i
(S(x(1)i )⊗ 1) ·∆(x
(2)i )
holding in any Hopf algebra, we obtain
(1⊗ x)f(1−) =∑i
(S(x(1)i ε(x
(2)i ))⊗ 1)f(1−) = (S(x)⊗ 1)f(1−)
Finally, we have
x.αV (f) = 〈1+ ⊗ id, (1⊗ x)f(1−)〉 =
= 〈1+ ⊗ id, (S(x)⊗ 1)f(1−)〉 =
= 〈1+ ⊗ id, (S(x)ρ ⊗ 1)f(1−)〉 = αV (x.f)
Therefore, Γ(V ) is isomorphic to V [[~]] as equicontinuous gD-module.
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 93
2.14. For any continuous ϕ ∈ Homg(V, V′), define a map Γ(ϕ) : Γ(V )→
Γ(V ′) by
Γ(ϕ) : f 7→ (id⊗ϕ) f
This map is clearly continuous and for all x ∈ gD
Γ(ϕ)(x.f) = (S(x)ρ ⊗ ϕ) f = x.Γ(ϕ)(f)
then Γ(ϕ) ∈ HomgD(Γ(V ),Γ(V ′)).
Since the diagram
Γ(V )
αV
Γ(ϕ)// Γ(V ′)
αV ′
V [[~]]ϕ// V ′[[~]]
is commutative for all ϕ ∈ Homg(V, V′), we have a well–defined functor
Γ : Repeq Ug[[~]]→ Repeq UgD[[~]]
which is naturally isomorphic to the pullback functor induced by the inclusion
iD : gD → g via the natural transformation
αV : Γ(V ) ' i∗DV [[~]]
2.15. Tensor structure on Γ. Denote the tensor product in the cate-
gories DΦ(Ug), DΦD(UgD) by ⊗, and let B1234 and B′1234 be the associativity
constraints
B1234 : (V1 ⊗ V2)⊗ (V3 ⊗ V4)→ V1 ⊗ ((V2 ⊗ V3)⊗ V4)
94 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
and
B′1234 : (V1 ⊗ V2)⊗ (V3 ⊗ V4)→ (V1 ⊗ (V2 ⊗ V3))⊗ V4
For any v ∈ Γ(V ), w ∈ Γ(W ), define JVW (v ⊗ w) to be the composition
L−i−−→ L− ⊗ L−
v⊗w−−→ (N∗+ ⊗ V )⊗ (N∗+ ⊗W )A−→ N∗+ ⊗ ((V ⊗N∗+)⊗W )
β−132−−→ N∗+⊗ ((N∗+⊗ V )⊗W )
A′−→ (N∗+⊗N∗+)⊗ (V ⊗W )i∨+⊗1−−−→ N∗+⊗ (V ⊗W )
where the pair (A,A′) can be chosen to be (BN∗+,V,N∗+,W
, B−1N∗+,N
∗+,V,W
) or (B′N∗+,V,N∗+,W , B′−1N∗+,N
∗+,V,W
).
The map JVW (v ⊗ w) is clearly a continuous g-morphism from L− to N∗+ ⊗
(V ⊗W ), so we have a well-defined map
JVW : Γ(V )⊗ Γ(W )→ Γ(V ⊗W )
Proposition. The maps JVW are isomorphisms of gD–modules and define
a tensor structure on the functor Γ.
The proof of Proposition 2.15 is given in 2.16–2.19.
2.16. We easily observe that the maps JVW are compatible with the gD–
action. Indeed, the map i∨+ is compatible with the right gD–action on N∗+ and,
for any x ∈ gD,
x.JVW (v ⊗ w) = (Sρ(x)⊗ id)(i∨+ ⊗ id⊗ id)A(v ⊗ w)i− =
= (i∨+ ⊗ id⊗ id)(∆(S(x))ρ)12A(v ⊗ w)i− =
= (i∨+ ⊗ id⊗ id)A((S ⊗ S)(∆(x)))ρ)13(v ⊗ w)i− = JVW (x.(v ⊗ w))
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 95
where A = A′β−132 A.
JVW is an isomorphism, since it is an isomorphism modulo ~. Indeed,
JVW (v ⊗ w) ≡ (i∗+ ⊗ 1)(1⊗ s⊗ 1)(v ⊗ w)i− mod ~
Finally, to prove that JVW define a tensor structure on Γ, we need to show
that, for any V1, V2, V3 ∈ DΦ(Ug) the following diagram is commutative
(Γ(V1)⊗ Γ(V2))⊗ Γ(V3)
ΦD
J12⊗1// Γ(V1 ⊗ V2)⊗ Γ(V3)
J12,3// Γ((V1 ⊗ V2)⊗ V3)
Γ(Φ)
Γ(V1)⊗ (Γ(V2)⊗ Γ(V3))1⊗J23
// Γ(V1)⊗ Γ(V2 ⊗ V3)J1,23
// Γ(V1 ⊗ (V2 ⊗ V3))
where Jij denotes the map JVi,Vj and Jij,k the map JVi⊗Vj ,Vk .
2.17. For any vi ∈ Γ(Vi), i = 1, 2, 3, the map Γ(Φ)J12,3J12⊗1(v1⊗v2⊗v3)
is given by the composition
(1⊗ Φ)(i∗+ ⊗ 1⊗3)A4(1⊗ β1⊗2,N∗+⊗ 1)A3((i∗+ ⊗ 1)⊗ 1⊗3)(A2 ⊗ 1⊗ 1)
· (1⊗ βN∗+,1 ⊗ 1⊗3)(A1 ⊗ 1⊗ 1)(v1 ⊗ v2 ⊗ v3)(i− ⊗ 1)i−
where
A1 = BN∗+,1,N∗+,2
A3 = BN∗+,1⊗2,N∗+,3
A2 = B−1N∗+,N
∗+,1,2
A4 = B−1N∗+,N
∗+,1⊗2,3
96 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
illustrated by the diagram
L−
i−// L− ⊗ L−
i−⊗1
// (L− ⊗ L−)⊗ L−
v1⊗v2⊗v3// ((N∗+ ⊗ V1)⊗ (N∗+ ⊗ V2))⊗ (N∗+ ⊗ V3)
A1⊗1⊗1// (N∗+ ⊗ ((V1 ⊗N∗+)⊗ V2))⊗ (N∗+ ⊗ V3)
1⊗βN∗+
,1⊗1⊗3
// (N∗+ ⊗ ((N∗+ ⊗ V1)⊗ V2))⊗ (N∗+ ⊗ V3)A2⊗1⊗1
// ((N∗+ ⊗N∗+)⊗ (V1 ⊗ V2))⊗ (N∗+ ⊗ V3)
(i∗+⊗1)⊗1⊗3
// (N∗+ ⊗ (V1 ⊗ V2))⊗ (N∗+ ⊗ V3)A3
// N∗+ ⊗ (((V1 ⊗ V2)⊗N∗+)⊗ V3)
1⊗β1⊗2,N∗+⊗1
// N∗+ ⊗ ((N∗+ ⊗ (V1 ⊗ V2))⊗ V3)A4
// (N∗+ ⊗N∗+)⊗ ((V1 ⊗ V2)⊗ V3)
i∗+⊗1⊗3
// N∗+ ⊗ ((V1 ⊗ V2)⊗ V3)1⊗Φ
// N∗+ ⊗ (V1 ⊗ (V2 ⊗ V3))
By functoriality of associativity and commutativity isomorphisms, we have
A3(i∗+ ⊗ 1⊗4) = (i∗+ ⊗ 1⊗4)A5
where A5 = BN∗+⊗N∗+,12,N∗+,3,
(1⊗ β12,N∗+⊗ 1)(i∗+ ⊗ 1⊗4) = (i∗+ ⊗ 1⊗4)(1⊗2 ⊗ β12,N∗+
⊗ 1⊗2)
and
A4(i∗+ ⊗ 1⊗4) = (i∗+ ⊗ 1⊗4)A6
where A6 = B−1N∗+⊗N∗+,N∗+,1⊗2,3. Finally, we have
Γ(Φ)J12,3(J12 ⊗ 1)(v1 ⊗ v2 ⊗ v3)
= (1⊗3 ⊗ Φ123)((i∗+(i∗+ ⊗ 1))⊗ 1⊗3)A (v1 ⊗ v2 ⊗ v3)(i− ⊗ 1)i− (2.2)
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 97
where
A = A6(1⊗2 ⊗ β1⊗2,N∗+⊗ 1⊗2)A5(A2 ⊗ 1⊗2)(1⊗ βN∗+,1 ⊗ 1⊗3)(A1 ⊗ 1⊗ 1)
2.18. On the other hand, J1,2⊗3(1⊗ J23)ΦD(v1 ⊗ v2 ⊗ v3) corresponds to
the composition
(i∗+ ⊗ 1⊗3)A′4(1⊗ βN∗+,1 ⊗ 1⊗2)A′3(1⊗2 ⊗ i∗+ ⊗ 1⊗2)(1⊗ 1⊗ A′2)
(1⊗3 ⊗ β2,N∗+⊗ 1)(1⊗ 1⊗ A′1)ΦD(v1 ⊗ v2 ⊗ v3)(1⊗ i−)i−
where
A′1 = BN∗+,2,N∗+,3
A′3 = BN∗+,1,N∗+,2⊗3
A′2 = B−1N∗+,N
∗+,2,3
A′4 = B−1N∗+,N
∗+,1,2⊗3
illustrated by the diagram
L−
i−// L− ⊗ L−
1⊗i−// L− ⊗ (L− ⊗ L−)
ΦD(v1⊗v2⊗v3)// (N∗+ ⊗ V1)⊗ ((N∗+ ⊗ V2)⊗ (N∗+ ⊗ V3))
1⊗1⊗A′1// (N∗+ ⊗ V1)⊗ (N∗+ ⊗ ((V2 ⊗N∗+)⊗ V3))
1⊗3⊗β2,N∗+⊗1
// (N∗+ ⊗ V1)⊗ (N∗+ ⊗ ((N∗+ ⊗ V2)⊗ V3))1⊗1⊗A′2
// (N∗+ ⊗ V1)⊗ ((N∗+ ⊗N∗+)⊗ (V2 ⊗ V3))
1⊗2⊗i∗+⊗1⊗2
// (N∗+ ⊗ V1)⊗ (N∗+ ⊗ (V2 ⊗ V3))A′3
// N∗+ ⊗ ((V1 ⊗N∗+)⊗ (V2 ⊗ V3))
1⊗β1,N∗+⊗1
// N∗+ ⊗ ((N∗+ ⊗ V1)⊗ (V2 ⊗ V3))A′4
// (N∗+ ⊗N∗+)⊗ (V1 ⊗ (V2 ⊗ V3))
i∗+⊗1⊗3
// N∗+ ⊗ (V1 ⊗ (V2 ⊗ V3))
By functoriality of associativity and commutativity isomorphisms, we have
A′3(1⊗2 ⊗ i∗+ ⊗ 1⊗2) = (1⊗2 ⊗ i∗+ ⊗ 1⊗2)A′5
98 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
where A′5 = BN∗+,1,N∗+⊗N∗+,2⊗3,
(1⊗ β1,N∗+⊗ 1⊗2)(1⊗2 ⊗ i∗+ ⊗ 1⊗2) = (1⊗ i∗+ ⊗ 1⊗3)(1⊗ β1,N∗+⊗N∗+ ⊗ 1⊗2)
and
A′4(1⊗ i∗+ ⊗ 1⊗3) = (1⊗ i∗+ ⊗ 1⊗3)A′6
where A′6 = B−1N∗+,N
∗+⊗N∗+,1,2⊗3. Thus,
J1,23(1⊗ J23)ΦD(v1 ⊗ v2 ⊗ v3)
= (i∗+ ⊗ 1⊗3)((1⊗ i∗+)⊗ 1⊗3)B ΦD(v1 ⊗ v2 ⊗ v3)(1⊗ i−)i− (2.3)
where
B = A′6(1⊗ β1,N∗+⊗N∗+ ⊗ 1⊗2)A′5(1⊗2 ⊗ A′2)(1⊗3 ⊗ β2,N∗+⊗ 1)(1⊗ 1⊗ A′1)
2. A RELATIVE ETINGOF–KAZHDAN FUNCTOR 99
2.19. Comparing (2.2) and (2.3), we see that it suffices to show that the
outer arrows of the following form a commutative diagram.
L−
(i−⊗1)i−
xx
(1⊗i−)i−
&&
(L− ⊗ L−)⊗ L−
v1⊗v2⊗v3
L− ⊗ (L− ⊗ L−)
ΦD(v1⊗v2⊗v3)
((N∗+ ⊗ V1)⊗ (N∗+ ⊗ V2))⊗ (N∗+ ⊗ V3)
A
Φ// (N∗+ ⊗ V1)⊗ ((N∗+ ⊗ V2)⊗ (N∗+ ⊗ V3))
B
((N∗+ ⊗N∗+)⊗N∗+)⊗ ((V1 ⊗ V2)⊗ V3)
(i∗+(i∗+⊗1))⊗1⊗3
Φ⊗Φ// (N∗+ ⊗ (N∗+ ⊗N∗+))⊗ (V1 ⊗ (V2 ⊗ V3))
(i∗+(1⊗i∗+))⊗1⊗3
N∗+ ⊗ ((V1 ⊗ V2)⊗ V3)1⊗Φ
// N∗+ ⊗ (V1 ⊗ (V2 ⊗ V3))
Using the pentagon and the hexagon axiom, we can show that
(Φ⊗ Φ)A = BΦ
We have to show that
Γ(Φ)J12,3(J12 ⊗ 1)(v1 ⊗ v2 ⊗ v3) = J1,23(1⊗ J23)ΦD(v1 ⊗ v2 ⊗ v3)
in Homg(L−, N∗+ ⊗ (V1 ⊗ (V2 ⊗ V3))):
100 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
J1,23(id⊗J23)ΦD(v1 ⊗ v2 ⊗ v3) =
= (i∨+(id⊗i∨+)⊗ id⊗3)BΦD(v1 ⊗ v2 ⊗ v3)(id⊗i−)i−
= (i∨+(id⊗i∨+)⊗ id⊗3)BΦD(v1 ⊗ v2 ⊗ v3)Φ(i− ⊗ id)i−
= (i∨+(id⊗i∨+)⊗ id⊗3)BΦΦD(v1 ⊗ v2 ⊗ v3)(i− ⊗ id)i−
= (i∨+(id⊗i∨+)Φ⊗ Φ)AΦD(v1 ⊗ v2 ⊗ v3)(i− ⊗ id)i−
= (i∨+(id⊗i∨+)Φ⊗ Φ)(S⊗3(ΦD)ρ ⊗ id⊗3)A(v1 ⊗ v2 ⊗ v3)(i− ⊗ id)i−
= (i∨+(id⊗i∨+)ΦS⊗3(ΦD)ρ ⊗ Φ)A(v1 ⊗ v2 ⊗ v3)(i− ⊗ id)i−
= (i∨+(i∨+ ⊗ id)⊗ Φ)A(v1 ⊗ v2 ⊗ v3)(i− ⊗ id)i−
= Γ(Φ)J12,3(J12 ⊗ id)(v1 ⊗ v2 ⊗ v3)
where the second and seventh equalities follow from Proposition 2.12, the fifth
one from the definition of the gD–action on the modules Γ(Vi) and the others
from functoriality of the associator Φ. This complete the proof of Theorem
2.1.
2.20. 1–Jets of relative twists. The following is a straightforward ex-
tension of the computation of the 1–jet of the Etingof–Kazhdan twist given in
[EK1].
Proposition. Under the natural identification
αV : Γ(V )→ V [[~]]
3. QUANTIZATION OF VERMA MODULES 101
the relative twist JΓ satisfyies
αV⊗W JΓ (α−1V ⊗ α
−1W ) ≡ 1 +
~2
(r + r21D ) mod ~2
in End(V ⊗W )[[~]].
Proof. For v ∈ V,w ∈ W , let
α−1V (v)(1−) =
∑fi ⊗ vi α−1
W (w)(1−) =∑
gj ⊗ wj
in (N∗+ ⊗ V )p+ and (N∗+ ⊗W )p+ respectively. Then using
〈(1+ ⊗ 1)⊗2,Ω23
∑i,j
fi ⊗ vi ⊗ gj ⊗ wj〉 = −r(v ⊗ w)
and
〈(1+ ⊗ 1)⊗2,ΩD,23
∑i,j
fi ⊗ vi ⊗ gj ⊗ wj〉 = −ΩD(v ⊗ w)
where Ω = Ω + ΩD, we get
αV⊗W JΓ (α−1V ⊗ α
−1W )(v ⊗ w) ≡ v ⊗ w +
~2
(r + r21D )(v ⊗ w) mod ~2
because the definition of JΓ involves the braiding β′XY = β−1Y X .
Corollary. The relative twist JΓ satisfies
Alt2JΓ ≡~2
(r − r21
2− rD − r21
D
2
)mod ~2
3. Quantization of Verma modules
This section and the next contain results about the quantization of classical
Verma modules, which are required to construct the morphism of D–categories
102 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
between the representation theory of Ug[[~]] and that of U~g. In particular,
from now on, we will assume the existence of a finite N–grading on g−, which
induces on g = g−⊕g+ a Lie bialgebra structure and allows us to consider the
quantization of g through the Etingof–Kazhdan functor, UEK
~ g.
3.1. Quantum Verma Modules. Let (g, g−, g+) be a graded Manin
triple. This means that we have an isomorphism between g+ and the graded
dual of g− and that g+ is naturally endowed with a Lie bialgebra structure.
Because of the functoriality of the quantization defined by Etingof and Kazh-
dan in [EK2], in the category of Drinfeld-Yetter modules over UEK
~ g− we can
similarly define quantum Verma modules.
The standard inclusions of Lie bialgebras g± ⊂ g ' Dg− lift to UEK
~ g± ⊂
UEK
~ g ' DUEK
~ g−, and we can define the induced modules of the trivial repre-
sentation over UEK
~ g±
M~± = Ind
UEK
~ g
UEK~ g±
k[[~]]
Similarly, we have Hopf algebra maps UEK
~ p± ⊂ UEK
~ g and UEK
~ p± → UEK
~ gD,
and we can define induced modules
L~− = IndU
EK
~ g
UEK~ p+
k[[~]] N~+ = Ind
UEK
~ g
UEK~ p−
UEK
~ gD
We want to show that the equivalence F : YDΦ(Ug−)[[~]]→ YDU
EK~ g−
matches
these modules. We start proving the statement for M−,M∗+.
3.2. Quantization of M±. We denote by (M~+)∗ the U
EK
~ g–module
Homk(IndU
EK
~ g
UEK~ g−
k[[~]], k[[~]])
3. QUANTIZATION OF VERMA MODULES 103
Theorem. In the category of left UEK
~ g–modules,
(a) F (M−) 'M~−
(b) F (M∗+) ' (M~
+)∗
Proof. The Hopf algebra UEK
~ g− is constructed on the space F (M−) with
unit element u ∈ F (M−) defined by u(1−) = ε+⊗1−, where ε+ ∈M∗+ is defined
as ε+(x1+) = ε(x) for any x ∈ Ug+. Consequently, the action of UEK
~ g− on
u ∈ F (M−) is free, as multiplication with the unit element. The coaction
of UEK
~ g− on F (M−) is defined using the R-matrix associated to the braided
tensor functor F ,i.e.,
π∗M− : F (M−)→ F (M−)⊗ F (M−), π∗(x) = R(u⊗ x)
where x ∈ F (M−) and RVW ∈ EndU
EK~ g
(F (V ) ⊗ F (W )) is given by RVW =
σJ−1WV F (βVW )JVW , JV,WV,W∈YDUg−
being the tensor structure on F . It is
easy to show that J(u⊗ u)|1− = ε+ ⊗ 1− ⊗ 1−, and, since Ω(1− ⊗ 1−) = 0, we
have
R(u⊗ u) = u⊗ u
For a generic V ∈ YDUg− [[~]], the action of UEK
~ g∗− is defined as
F (M−)∗ ⊗ F (V )→ F (M−)∗ ⊗ F (M−)⊗ F (V )→ F (V )
This means, in particular that, for every φ ∈ I ⊂ UEK
~ g∗−, where I is the max-
imal ideal corresponding to u⊥, we have φ.u = 0. This proves (a).
104 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
The module M∗+ satisfies the following universal property: for any V in the
Drinfeld category of equicontinuous Ug-modules, we have
HomUg(V,M∗+) ' HomUg−(V, k)
Indeed, to any map of Ug-modules f : V → M∗+, we can associate f : V → k,
f(v) = 〈f(v), 1+〉. It is clear that f factors through V/g−.V . The equicon-
tinuity property is necessary to show the continuity of f with respect to the
topology on V .
Since F defines an equivalence of categories, we have
HomU
EK~ g
(F (V ), F (M∗+)) ' HomUg(V,M
∗+)[[~]] ' HomUg−(V, k)[[~]]
Using the natural isomorphism αV : F (V )→ V [[~]], defined by
αV (f) = 〈f(1−), 1+ ⊗ id〉
we obtain a map HomUg−(V, k)[[~]] → Homk(F (V ), k[[~]]). Consider now the
linear isomorphism α : UEK
~ g− → Ug−[[~]] and for any x ∈ Ug− consider the
g-intertwiner ψx : M− →M∗+ ⊗M− defined by ψx(1−) = ε+ ⊗ x1−. It is clear
that, if f(1−) = f(1) ⊗ f(2) in Swedler’s notation,
αV (ψx.f) = 〈(i∨+ ⊗ id)Φ−1(id⊗f)(ε+ ⊗ x.1−), 1+ ⊗ id〉
= 〈Φ−1(ε+ ⊗ id⊗ id)(id⊗∆(x))(id⊗f(1) ⊗ f(2)), (T ⊗ id)(1+ ⊗ 1+ id)〉
= 〈∆(x)(f(1) ⊗ f(2)), 1+ ⊗ id〉
= 〈f(1), 1+〉x.f(2)
= x.αV (f)
3. QUANTIZATION OF VERMA MODULES 105
using the fact that (ε⊗ 1⊗ 1)(Φ) = 1⊗2 and (ε⊗ 1)(T ) = 1. So, clearly, if φ ∈
HomUg−(V, k), then φ αV ∈ HomU
EK~ g−
(F (V ), k[[~]]). Then F (M∗+) satisfies
the universal property of Homk(IndU
EK
~ g
UEK~ g−
k[[~]], k[[~]]) and (b) is proved.
3.3. Quantization of relative Verma modules. The proof of (b) shows
that the linear functional F (M∗+) → k[[~]] is, in fact, the trivial deformation
of the functional M∗+ → k. These results extend to the relative case and hold
for the right gD–action on L−, N∗+.
Theorem. In the category YDU
EK~ g−
(a) F (L−) ' L~−
(b) F (N∗+) ' (N~+)∗
Moreover, as right UEK
~ gD–module
(c) FD(L−) ' L~−
(d) FD(N∗+) ' (N~+)∗
The proof of (a) and (b) amounts to constructing the morphisms
k[[~]]→ F (L−) F (N∗+)→ UEK
~ g∗D
equivariant under the action of UEK
~ p+ and UEK
~ p− respectively.
A direct construction along the lines of the proof of Theorem 3.2 is however
not straightforward. We prove this theorem in the next section by using a
description of the modules L−, N∗+ and their images through F via Prop
106 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
categories. These descriptions show that the classical intertwiners
k→ L− N∗+ → Ug∗D
satisfy the required properties and yield canonical identifications
F (L−) ' L~− F (N∗+) ' (N~+)∗
4. Universal relative Verma modules
In this section, we prove Theorem 3.3 by using suitable Prop (product-
permutation) categories compatible with the EK universal quantization functor
[EK2, EG].
4.1. Prop description of the EK quantization functor. We will
briefly review the construction of Etingof–Kazhdan in the setting of Prop
categories [EK2].
A Prop is a symmetric tensor category generated by one object. More
precisely, a cyclic category over S is the datum of
• a symmetric monoidal k–linear category (C,⊗) whose objects are
non–negative integers, such that [n] = [1]⊗n and the unit object is
[0]
• a bigraded set S =⋃m,n∈Z≥0
Snm of morphism of C, with
Snm ⊂ HomC([m], [n])
such that any morphism of C can be obtained from the morphisms in S and
permutation maps in HomC([m], [m]) by compositions, tensor products or lin-
ear combinations over k. We denote by FS the free cyclic category over S.
4. UNIVERSAL RELATIVE VERMA MODULES 107
Then there exists a unique symmetric tensor functor FS → C, and the follow-
ing holds (cf. [EK2])
Proposition. Let C be any cyclic category generated by a set S of mor-
phisms. Then C has the form FS/I, where I is a tensor ideal in FS.
Let N be a symmetric monoidal k–linear category, and X an object in
N . A linear algebraic structure of type C on X is a symmetric tensor functor
GX : C → N such that GX([1]) = X. A linear algebraic structure of type C on
X is a collection of morphisms between tensor powers of X satisfying certain
consistency relations.
We mainly consider the case of non–degenerate cyclic categories, i.e., sym-
metric tensor categories with injective maps k[Sn]→ HomC([n], [n]). We first
consider the Karoubian envelope of C obtained by formal addition to C of the
kernel of the idempotents in k[Sn] acting on [n]. Furthermore, we consider the
closure under inductive limits. In this category, denoted S(C), every object is
isomorphic to a direct sum of indecomposables, corresponding to irreducible
representations of Sn (cf. [EK2, EG]). In particular, in S(C), we can consider
the symmetric algebra
S[1] =⊕n≥0
Sn[1]
If N is closed under inductive limits, then any linear algebraic structure of
type C extends to an additive symmetric tensor functor
GX : S(C)→ N
We introduce the following fundamental Props .
108 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
• Lie bialgebras. In this case the set S consists of two elements of
bidegrees (2, 1), (1, 2), the universal commutator and cocommutator.
The category C = LBA is FS/I, where I is generated by the classi-
cal five relations.
• Hopf algebras. In this case, the set S consists of six elements of
bidegrees (2, 1), (1, 2), (0, 1), (1, 0), (1, 1, ), (1, 1), the universal prod-
uct, coproduct, unit, count, antipode, inverse antipode. The cate-
gory C = HA is FS/I, where I is generated by the classical four
relations.
The quantization functor described in Section 1 can be described in this
generality, as stated by the following (cf. [EK2, Thm.1.2])
Theorem. There exists a universal quantization functor Q : HA→ S(LBA).
Let g− be the canonical Lie dialgebra [1] in LBA with commutator µ and
cocommutator δ. Let Ug− := Sg− ∈ S(LBA) be the universal enveloping
algebra of g−. The construction of the Etingof–Kazhdan quantization functor
amounts to the introduction of a Hopf algebra structure on Ug−, which coin-
cides with the standard one modulo 〈δ〉, and yields the Lie bialgebra structure
on g− when considerd modulo 〈δ2〉. This Hopf algebra defines the object Q[1],
where [1] is the generating object in HA. The formulae used to defined the
Hopf structure coincide with those defined in [EK1, Part II] and described in
Section 1. In particular, they rely on the construction of the Verma modules
M− := Sg− M∗+ = Sg−
4. UNIVERSAL RELATIVE VERMA MODULES 109
realized in the category of Drinfeld–Yetter modules over g− as object of LBA
.
4.2. Props for split pairs of Lie bialgebras. Let (g−, gD,−) be a split
pair of Lie bialgebras, i.e., there are Lie bialgebra maps
gD,−i−→ g−
p−→ gD,−
such that p i = id. These maps induce an inclusion DgD,− ⊂ Dg− and
consequently an inclusion of Manin triple (gD, gD,−, gD,+) ⊂ (g, g−, g+), as
described in Section 2.6.
Definition. We denote by PLBA the Karoubian envelope of the multi-
colored Prop, whose class of objects is generated by the Lie bialgebra objects
[g−], [gD,−], related by the maps i : [gD,−]→ [g−] , p : [g−]→ [gD,−], such that
p i = id[gD,−].
The Karoubian envelope implies that [m−] := ker(p) ∈ PLBA.
Proposition. The multicolored Prop PLBA is endowed with a pair of
functors U,L
U,L : LBA→ PLBA U [1] := [g−], L[1] := [gD,−]
and natural transformations i, p, induced by the maps i, p in PLBA,
LBA
U
**
L
44PLBAp
i
KS
110 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
such that p i = id. Moreover, it satisfies the following universal property:
for any tensor category C, closed under kernels of projections, with the same
property as PLBA , there exists a unique tensor functor PLBA→ C such that
the following diagram commutes
LBA
U
**
L
44
$$
::PLBAp
i
KS
// C
4.3. Props for split pairs of Hopf algebras. We can analogously de-
fine suitable Prop categories corresponding to split pairs of Hopf algebras. In
particular, we consider the Prop PHA characterized by functors U~, L~ and
natural transformations p~, i~ satisfying
HA
U~
))
L~
55
$$
::PHAp~
i~
KS∃!
// C
where HA denotes the Prop category of Hopf algebras. These also satisfy
HAQEK
//
S(LBA)
PHAQPLBA
// S(PLBA)
where QPLBA is the extension of the Etingof–Kazhdan quantization func-
tor to PLBA, obtaine by the universal property described above with C =
S(PLBA).
4. UNIVERSAL RELATIVE VERMA MODULES 111
4.4. Props for parabolic Lie subalgebras. In order to describe the
module N∗+ it is necessary to deal with the Lie bialgebra object p− or, in other
words to introduce the double of gD,− and the Prop D⊕(LBA) [EG]. We then
introduce the multicolored Prop as a cofiber product of PLBA and D⊕(LBA)
over LBA .
Proposition. The multicolored Prop PLBAD is endowed with canonical
functors
D⊕(LBA)→ PLBAD ← PLBA
and satisfies the following universal property:
LBAdouble
//
D⊕(LBA)
PLBA
00
// PLBAD∃!
(( C
where double is the Prop map introduced in [EG].
In PLBAD we can consider the Lie bialgebra object [p−].
4.5. Props for parabolic Hopf subalgebras. Similarly, we introduce
the multicolored Prop PHAD , endowed with canonical functors (cf. [EG])
D⊗(HA)→ PHAD ← PHA
112 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
and satisfying an analogous universal property:
HAdouble
//
D⊗(HA)
PHA
00
// PHAD∃!
'' C
Moreover, we then have a canonical functor
QPLBAD : PHAD → S(PLBAD)
obtained applying such universal property with C = S(PLBAD) and satisfying
HA
LHAxx
double//
QEK
D⊗(HA)
Q2
vv
PHA
QPLBA
// PHAD
QPLBAD
S(LBA)
xx
S(double)// S(D⊕(LBA))
vv
S(PLBA) // S(PLBAD)
The commutativity of the square on the back is given by the compatibility of
the quantization functor with the doubling operations, proved in [EG].
4.6. Prop description of L−, N∗+. The modules L−, N
∗+ can be real-
ized in S(PLBAD). The module L− is constructed over the object Sm− ∈
4. UNIVERSAL RELATIVE VERMA MODULES 113
S(PLBA). The structure of Drinfeld-Yetter module over g− is determined in
the following way:
• the free action of the Lie algebra object m− is defined by the map
Sm− ⊗ Sm− → Sm−
given by Campbell-Hausdorff series, describing on Sm− the multipli-
cation in Um−.
• we define the action of gD,− to be trivial on 1→ Sm−.
• The actions of m−, gD,−, the relation
π ([, ]⊗ 1) = π (1⊗ π)− π (1⊗ π) σ12
and the map [, ] : gD,− ⊗m− → m− define the action of g−.
• We then impose the trivial coaction on 1 → Sm− and the compati-
bility condition between action and coaction
π∗ π = (1⊗ π)σ12(1⊗ π∗)− (1⊗ π)(δ ⊗ 1) + (µ⊗ 1)(1⊗ π∗)
determines the coaction for Sm−. The action defined is compatible
with [, ] : gD,− ⊗m− → m−
Similarly, the module N∗+ can be realized on the object Sp−, formally added
to S(PLBAD).
114 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
We determine the formulae for the action and the coaction of g− by direct
inspection of the action of g = g− ⊕ g+ on N+ in the category Vect. Namely,
the identification N∗+ = Sp− is clearly obtained through the invariant bilinear
form 〈−,−〉 and there are topological formulae expressing the action of g on
N+. Therefore we determine action and coaction on N∗+ in the following way:
• the g+–action on N+ = Sp+ = Sg+ ⊗ SgD,− is given by the free ac-
tion on the first factor Sg+ expressed by Campbell–Hausdorff series.
• the action of g− = m− ⊕ gD,− on the subspace SgD,− ⊂ Sp+ is given
by the trivial action of m− and the usual free action of gD,− by mul-
tiplication.
• The action of g− is then interpreted as a topological coaction of g+
and the aforementioned compatibility condition between action and
coaction allows to extend the formula for the topological g+ coaction
on the entire space Sp+.
• Through the invariant bilinear form 〈−,−〉, these formulae are car-
ried over N∗+ = Sp−, by switching, in particular, the bracket and the
topological cobracket on g+ with the cobracket and the bracket in
g−, respectively.
• The obtained formulae, describing the action and the coaction of
g− on N∗+, are well–defined in the category PLBAD and define the
requested structure of Drinfeld-Yetter module over g−.
4. UNIVERSAL RELATIVE VERMA MODULES 115
4.7. Proof of Theorem 3.3. The relative Verma module
N+ = Indgm− k ' Indg
p− UgD
satisfies
HomUg(N+, V ) ' HomUp−(UgD, V )
for every Ug-module V . We have a canonical map of p−-modules ρD : UgD →
N+ corresponding to the identity in the case V = N+. We get a map of
p−-modules ρ∗D : N∗+ → Ug∗D inducing an isomorphism
HomUg(V,N∗+) ' HomUp−(V, Ug∗D)
The morphism ρ∗D can indeed be thought as
Up− ⊗N∗+
// N∗+
ρ∗D
UgD ⊗ Ug∗D // Ug∗D
Assuming the existence of a suitable finite N–grading, a split pair of Lie
bialgebras (g−, gD,−), gives rise to a functor
PLBAD → Vect
Consider now the trivial split pair given by (gD,−, gD,−). We have a natural
transformation
PLBAD
(g−,gD,−)
))
(gD,−,gD,−)
55Vectp
116 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
where p naturally extends to the projection p− → gD.
The module U(gD)∗ is indeed the module N∗+ with respect to the trivial pair
(gD,−, gD,−). Consequently, the existence of the p−–intertwiner ρ∗D can be in-
terpreted as a simple consequence of the existence of natural transformation p.
The quantization functor QPLBAD extends the natural transformation p to
PHAD$$
::
// S(PLBAD)
(g−,gD,−)
))
(gD,−,gD,−)
55VectS(p)
and shows that
F (N∗+) ' (N~+)∗
Similarly, we can consider the natural transformation S(i) and the diagram
PHAD$$
::
// S(PLBAD)
(g−,gD,−)
))
(gD,−,gD,−)
55VectS(i)
KS
implying
F (L−) ' L~−
5. CHAINS OF MANIN TRIPLES 117
We can make analogous consideration for the right gD–action on L−, N∗+. This
leads to isomorphisms of right UEK
~ gD–modules
FD(N∗+) ' (N~+)∗ FD(L−) ' L~−
5. Chains of Manin triples
5.1. Chains of length 2. In Section 3, given an inclusion of Manin triples
iD : gD ⊆ g, we introduced the relative quantum Verma modules
L~− = IndU
EK
~ g
UEK~ p+
k[[h]] N~+ = Ind
UEK
~ g
UEK~ p−
UEK
~ gD
These modules allow to define the functor
Γ~ : D(UEK
~ g)→ D(UEK
~ gD)
by
Γ~(V) = HomU
EK~ g
(L~−, (N~+)∗ ⊗ V)
Lemma. The functor Γ~ is naturally tensor isomorphic to the restriction
functor (UEK
~ (i~D))∗.
Proof. The proof of the existence of the natural isomorphism as UEK
~ gD–
module is identical to that of Prop. 2.13. The isomorphism respects the tensor
structures, because there are only trivial associators involved.
We now prove the following
Theorem. Let g, gD be Manin triples with a finite Z–grading and iD :
gD ⊆ g an inclusion of Manin triples compatible with the grading. Then, there
118 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
exists an algebra isomorphism
Ψ : UEK
~ g→ Ug[[~]]
restricting to ΨEKD on U
EK
~ gD, where the completion is given with respect to
Drinfeld–Yetter modules.
Proof. In the previous section, we showed that the quantization of the
(UEK
~ g, UEK
~ gD)–modules N∗+, L− gives
F (N∗+)U
EK
~ g−−−→ (N~
+)∗U
EK
~ gD←−−−− F EK
D (N∗+)
F (L−)U
EK
~ g−−−→ L~−
UEK
~ gD←−−−− F EK
D (L−)
Recall that the standard natural transformations αV : F (V ) ' V [[~]], (αD)V :
FD(V ) ' V [[~]] give isomorphisms of right UgD[[~]]–modules
F (N∗+) ' N∗+[[~]] F (L−) ' L−[[~]]
and isomorphisms of Ug[[~]]–modules
F EK
D (N∗+) ' N∗+[[~]] F EK
D (L−) ' L−[[~]]
In particular, we get isomorphisms of right UEK
~ gD–modules
F EK
D F (N∗+) ' F EK
D (N∗+) ' (N~+)∗ F EK
D F (L−) ' F EK
D (L−) ' L~−
and isomorphisms of UEK
~ g–modules
F EK
D F (N∗+) ' F (N∗+) ' (N~+)∗ F EK
D F (L−) ' F (L−) ' L~−
5. CHAINS OF MANIN TRIPLES 119
We have a natural isomorphism through J :
HomU
EK~ g
(F (L−), F (N∗+)⊗ F (V )) ' Homg(L−, N∗+ ⊗ V )[[~]]
This is indeed an isomorphism of UgD[[~]]–modules, since, for x ∈ UgD, φ ∈
HomU
EK~ g
(F (L−), F (N∗+)⊗ F (V )), we have
x.φ := (F (x)⊗ id) φ J (F (x)⊗ id)) = F (x⊗ id) J
Quantizing both side and using the isomorphism F EKD F (N∗+) ' (N~
+)∗, we
obtain a natural transformation
γD : Γ~ F ' FD Γ
making the following diagram commutative
DΦ(Ug)F
//
Γ
D(U~g)
Γ~
γD
s
DΦD(UgD[[~]])FD
// D(U~gD)
Applying the construction above to the algebra of endomorphisms of the
fiber functor, we get the result.
5.2. Chains of arbitrary length. For any chain
C : 0 = g0 ⊆ g1 ⊆ · · · ⊆ gn−1 ⊆ gn = g
of inclusions of Manin triples, the natural transformations
γi,i+1 ∈ Nat⊗(Γ~i,i+1 Fi+1, Fi Γi,i+1)
120 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
where 0 ≤ i ≤ n − 1, Γ0,1 : DΦ(g1) → Vectk[[~]] is the EK fiber functor, and
F EK0 = id, yield a natural transformation
γC = γ0,1 · · · γn−1,n
∈ Nat⊗(Γ~0,1 · · · Γ~n−1,n Fn,Γ0,1 · · · Γn−1,n)
∼= Nat⊗((i∗0,n)~ Fn,Γ0,1 · · · Γn−1,n)
= Nat⊗(Fn,Γ0,1 · · · Γn−1,n)
where we used Γ~i,i+1∼= (i∗i,i+1)~, and the fact that the composition (i∗0,n)~ Fn
is the EK fiber functor for gn, which we denote by the same symbol as Fn.
This proves the following
Theorem.
(i) For any chain of Manin triples
C : g0 ⊆ g1 ⊆ · · · ⊆ gn ⊆ g
there exists an isomorphism of algebras
ΨC : UEK
~ g→ Ug[[~]]
such that ΨC(UEK
~ gi) = Ugi[[~]] for any gi ∈ C.
(ii) Given two chains C,C′, the natural transformation
ΦCC′ := γ−1C γC′ ∈ Aut(F EK)
satisfies
Ad(ΦCC′)ΨC′ = ΨC
5. CHAINS OF MANIN TRIPLES 121
The following proposition is clear.
Proposition. The natural transformations ΦCC′C,C′ satisfy the follow-
ing properties
(i) Orientation. Given two chains C,C′
ΦCC′ = Φ−1C′C
(ii) Transitivity. Given the chains C,C′,C′′
ΦCC′ ΦC′C′′ = ΦCC′′
(iii) Factorization. Given the chains
C,C′ : g0 ⊆ g0 ⊆ · · · ⊆ gn D,D′ : gn ⊆ · · · ⊆ gn+n′
Φ(C∪D)(C′∪D′) = ΦCC′ ΦDD′
5.3. Abelian Manin triples and central extensions. We will now
consider the following special case, that generalizes the role of Levi subalgebras
for Kac–Moody algebras.
Proposition. If g admits a Manin subtriple lD, obtained by a central
extension of gD, then the relative twists and the gauge transformations are
invariant under lD. In particular, the Etingof–Kazhdan constructions are in-
variant under abelian Manin subtriples.
Proof. For gD = 0, the statement reduces to prove that the Etingof–
Kazhdan functor preserves the action of an abelian Manin subtriple a ⊂ g
122 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
(cf. [EK6, Thm. 4.3], with a− = h). Under this assumption, the natural map
Ua− // UEK
~ g− := F (M−)
a // ψa : 1− 7→ ε⊗ a1−
defines an inclusion of bialgebras. For any V [[~]] ∈ DΦ(Ug), the natural
identification
αV : F (V )→ V [[~]]
is then an isomorphism of Ua–modules. This gives the following commutative
diagram
DΦ(Ug)F
//
&&
F
''
D(UEK
~ g)
ww
xx
D(Ua[[~]])
A
We can observe that the tensor restriction functor fits in an analogous dia-
gram. It is easy to show that the object ΓD(L−) is naturally a pointed Hopf
algebra in the category DΦD(UgD). We denote by DgD(ΓD(L−)) the category
of Drinfeld–Yetter modules over ΓD(L−) in the category DΦD(UgD). This cat-
egory is naturally equivalent to the category of Drinfeld–Yetter module over
the Radford’s product UgD,−[[~]]#ΓD(L−) and there is a natural identification
DΦ(Ug)
ΓD &&
// DgD(ΓD(L−))
ww
DΦD(UgD)
5. CHAINS OF MANIN TRIPLES 123
Moreover, there is a natural inclusion of bialgebras
U lD ⊂ UgD[[~]]#ΓD(L−)
and a natural U lD–module identification ΓD(V ) → V [[~]]. This originates
natural identifications
DΦ(Ug) //
&&
ΓD
%%
DgD(ΓD(L−))
xx
ww
D(U lD)
DΦD(UgD)
This proves that relative twists are invariant under lD. It is clear that the
Casimir operator ΩD ∈ (gD ⊗ gD)lD defines a braided tensor structure on
D(U lD[[~]]) which is preserved by the restriction functor induced by the inclu-
sion jD : gD ⊂ lD. Given the decomposition lD = gD o cD, the natural map
UcD → EndgD(j∗DV ) induces an action of UcD on FD(j∗DV ), commuting with
the action of UEK
~ gD. Therefore, we obtain a naturally commutative diagram
DΦD(U lD)˜FD//
j∗D
D(UEK
~ lD)
DΦD(UgD)FD// D(U
EK
~ gD)
where˜FD is the tensor functor induced by the composition FD j∗D. The
natural transformation γ automatically lifts to the level of lD, as showed in
124 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
the following diagram
DΦ(Ug)F
//
&&
Γ
D(UEK
~ g)
xx
DΦD(U lD)˜FD//
j∗D
xx
D(UEK
~ lD)
&&
DΦD(UgD)FD
// D(UEK
~ gD)
6. The equivalence of quasi–Coxeter categories
The following is the main result of this Chapter.
Theorem. Let g be a symmetrizable Kac–Moody algebra with a fixed Dg–
structure. Then the completion U~g is isomorphic to a quasi-Coxeter quasitri-
angular quasibialgebra of type Dg on the quasitriangular Dg–quasibialgebra
(Ug[[~]], UgD[[~]],∆0, ΦKZD , RKZ
D )
where the completion is taken with respect to the integrable modules in category
O.
6.1. D–structures on Kac–Moody algebras. Let A = (aij)i,j∈I be a
complex n×n matrix and g = g(A) the corresponding generalized Kac–Moody
algebra defined in Section 1. Let J be a nonempty subset of I. Consider the
submatrix of A defined by
AJ = (aij)i,j∈J
We recall the following proposition from [Ka, Ex.1.2]
6. THE EQUIVALENCE OF QUASI–COXETER CATEGORIES 125
Proposition. Let
ΠJ := αj | j ∈ J Π∨J := hj | j ∈ J
Let h′J be the subspace of h generated by Π∨J and
tJ =⋂j∈J
Kerαj = h ∈ h | 〈αj, h〉 = 0 ∀j ∈ J
Let h′′J be a supplementary subspace of h′J + tJ in h and let
hJ = h′J ⊕ h′′
J
Then,
(i) (hJ,ΠJ,Π∨J) is a realization of the generalized Cartan matrix AJ.
(ii) The subalgebra gJ ⊂ g, generated by ej, fjj∈J and hJ, is the Kac–
Moody algebra associated to the realization (hJ,ΠJ,Π∨J) of AJ.
Set
QJ =∑j∈J
Zαj ⊂ Q g = g(A) =⊕α∈Q
gα
Then,
(iii)
gJ = hJ ⊕⊕
α∈QJ\0
gα
Let A be a symmetrizable matrix with a fixed decomposition and (−|−) be the
standard normalized non–degenerate bilinear form on h. Then,
(iv) The restriction of (−|−) to hJ is non–degenerate.
126 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
Proof. Since dim(h′J ∩ tJ) = dim(z(gJ) = nJ − lJ, where nJ = |J| and
lJ = rank(AJ), it follows that
dim h′′J = nJ − lJ dim hJ = 2nJ − lJ
Moreover, by construction, the restriction of αjj∈J to hJ are linearly inde-
pendent. Indeed, since 〈∑cjαj, tJ〉 = 0 for all cj ∈ C,
〈∑j∈J
cjαj, hJ〉 = 0 =⇒ 〈∑j∈J
cjαj, h〉 = 0 =⇒ cj = 0
This proves (i). The proof of (ii) and (iii) is clear.
Assume now that A is irreducible and symmetrizable and there exists h ∈
hJ such that
(h|h′) = 0 ∀h′ ∈ hJ
In particular, (h|α∨j ) = 0 and h ∈ h′J ∩ tJ ⊂ h′J. Therefore, h =∑cjα
∨j and
(∑
cjα∨j |h′) =
∑cj(α
∨j |h′) = 〈
∑cjdjαj, h
′〉 = 0
Since the operators αj are linearly independent over hJ and dj 6= 0, we have
cj = 0 and h = 0. We conclude that (|) is non–degenerate on hJ and (iv) is
proved.
Remark. The derived algebra g′J = [gJ, gJ] is generated by ej, fj, hjj∈J,
where hj = [ej, fj]. Therefore, it does not depend of the choice of the subspace
h′′J. The assignment J 7→ g′J defines a structure that coincides with the one
provided in [TL4, 3.2.2].
6. THE EQUIVALENCE OF QUASI–COXETER CATEGORIES 127
Let now A be an irreducible, generalized Cartan matrix. Let Dg = D(A)
be the Dynkin diagram of g, that is, the connected graph having I as vertex
set and an edge between i and j if aij 6= 0. For any i ∈ I, let sli2 ⊂ g be the
three–dimensional subalgebra spanned by ei, fi, hi.
Any connected subdiagram D ⊆ Dg defines a subset JD ⊂ I. We would like
to use the assignement J 7→ gJ to define a Dg–algebra structure on g = g(A).
Remark. For any subset J of finite type, dim h′′J = nJ − lJ = 0 and
hJ = h′J. Therefore, if A is a generalized Cartan matrix of finite type, h′′J = 0
for any subset J ⊂ I. The Dg–algebra structure on g = g(A) is then uniquely
defined by the subalgebras sli2i∈I and the Cartan subalgebra is defined for
any subdiagram D ⊂ Dg by
hD = hi | i ∈ V(D)
If A is a generalized Cartan matrix of affine type, we obtain diagrammatic
Cartan subalgebras hD, where
hD =
hi | i ∈ V(D) if D ⊂ Dg
h if D = Dg
If A is an irreducible generalized Cartan matrix of hyperbolic type, i.e., every
submatrix is of finite or affine type, it is still possible to define a Dg–algebra
structure, depending upon the choice of the subspaces h′′J for |I \ J| = 1.
It is not always possible to define a Dg–algebra structure for a generic
matrix of order ≥ 3. In order to obtain a Dg–algebra structure on g = g(A),
128 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
we have to satisfy the following condition:
hJ ⊂ tJ⊥ ∩⋂J⊂J′
hJ′
Since tJ⊥ + tJ = h, we can always choose hJ ⊆ tJ⊥ .
Lemma. Assume given a Dg–algebra structure on g = g(A). Then for any
two subsets J′,J′′ ⊂ I,
corank(AJ′∩J′′) ≤ corank(AJ′) + corank(AJ′′)
In particular, if corank(AJ′) = corank(AJ′′) = 0, then corank(AJ′∩J′′) = 0.
Proof. The result is an immediate consequence of the estimate, given by
the construction,
dim(hJ′ ∩ hJ′′) ≤ |J′ ∩ J′′|+ (corank(AJ′) + corank(AJ′′))
and the constraint
hJ′∩J′′ ⊆ hJ′ ∩ hJ′′
Remark. Indeed, it is easy to show that the symmetric irreducible Cartan
matrix
A =
2 −1 0 0
−1 2 −2 0
0 −2 2 −1
0 0 −1 2
6. THE EQUIVALENCE OF QUASI–COXETER CATEGORIES 129
does not admit any Dg–algebra structure on g(A), since dim h23 = 3 and
dim h123 ∩ h234 = 2.
The previous condition on the corank is not sufficient to obtain a Dg–
algebra structure on g(A). Consider the symmetric Cartan matrix
A =
2 −2 0 0
−2 2 −1 0
0 −1 2 −1
0 0 −1 2
A clearly satisfies the above condition. Nonetheless, a suitable h′′12, complement
in h of (h′12 + t12), should satisfies:
h′′12 ⊂ h123 = h′123 and h′′12 ⊆ t4 = 〈h′12,−2α∨3 + α∨4 〉
that are clearly not compatible conditions. Therefore, there is no suitable
structure for A.
In the following, we will consider only symmetrizable Kac–Moody algebras
g that admit such a structure. It automatically defines an analogue structure
on UEK
~ g.
6.2. qCqtqba structure on U~g. Given a fixed Dg–structure on the
Kac–Moody algebra g, the quantum enveloping algebra U~g is naturally en-
dowed with a quasi–Coxeter quasitriangular quasibialgebra structure of type
Dg defined by
130 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
(i) Dg-algebra: for any D ∈ SD(Dg), let gD ⊂ g be the corresponding
Kac–Moody subalgebra. The Dg-algebra structure is given by the
subalgebras U~gD.
(ii) Quasitriangular quasibialgebra: the universalR-matrices R~,D,
with trivial associators ΦD = 1⊗3 and structural twists FF = 1⊗2.
(iii) Quasi-Coxeter: the local monodromies are the quantum Weyl group
elements S~i i∈I. The Casimir associators ΦGF are trivial.
We transfer this qCqtqba structure on Ug[[~]]. More precisely, we define
an equivalence of quasi–Coxeter categories between the representation theories
of U~g and Ug[[~]].
6.3. Gauge transformations for g(A). For any D ⊂ Dg, the inclusion
gD ⊂ g, defined in the previous section, lifts to an inclusion of Manin triples
gD ⊕ hD ⊂ g⊕ h
We denote by gD = (gD ⊕ hD, bD,+, bD,−) the Manin triple attached to gD, for
any D ⊆ Dg.
Theorem. There exists an equivalence of braided Dg–monoidal categories
from
((DΦB(U gB[[~]]),⊗B,ΦB, σRB), (ΓBB′ , JBB′
F ))
to
((D(U~gB),⊗B, id, σR~B), (Γ~BB′ , id))
given by (FB, γFBB′).
6. THE EQUIVALENCE OF QUASI–COXETER CATEGORIES 131
Proof. The natural transformations γBB′ , B ⊆ B′ ⊆ Dg constructed in
Section 5, define, by vertical composition, a natural transformation
γCBB′ ∈ Nat⊗(Γ~BB′ FB′ , FB ΓBB′)
for any chain of maximal length
C : B = C0 ⊂ C1 ⊂ · · · ⊂ Cr = B′
Any chain of maximal length defines uniquely a maximal nested set FC ∈
Mns(B,B′), but this is not a one to one correspondence. For example, for
D = A3, the maximal nested set
F = α1, α3, α1, α2, α3
corresponds to two different chains of maximal length
C1 : α1 ⊂ α1 t α3 ⊂ A3 C2 : α3 ⊂ α1 t α3 ⊂ A3
In order to prove that the natural transformations γ define a morphism of
braided Dg–monoidal categories, we need to prove that the transformation
γCBB′ depend only on the maximal nested set corresponding to C.
132 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
In particular, we have to prove that, for any B1 ⊥ B2 in I(D), the con-
struction of the fiber functor
CB1tB2
FB1,B1tB2
##
FB2,B1tB2
CB1
FB1 ##
CB2
FB2
C∅
is independent of the choice of the chain. In our case,
CB1tB2 = D(U gB1 [[~]]⊗ U gB2 [[~]])
and the braided tensor structure is given by product of the braided tensor
structures on
CB1 = DΦB1(U gB1 [[~]]) CB2 = DΦB2
(U gB2 [[~]])
Similarly, the tensor structure on the forgetful functor
CB1tB2 → CBi i = 1, 2
is obtained killing the tensor structure on CBi , i = 1, 2, i.e., applying the tensor
structure on CBi → C∅. In particular, the tensor structure on FB1 FB1,B1tB2
and FB2 FB2,B1tB2 coincide, since [gB1 , gB2 ] = 0.
Analogously we have an equality of natural transformation
γB1 γB1,B1tB2 = γB2 γB2,B1tB2
6. THE EQUIVALENCE OF QUASI–COXETER CATEGORIES 133
Therefore, for any maximal nested set F ∈ Mns(B,B′), it is well defined a
natural transformation
γFBB′ ∈ Nat⊗(Γ~BB′ FB′ , FB ΓBB′)
so that the data (FB, γFBB′) define an isomorphism of D–categories from
DΦB(U gB[[~]]) to D(U~gB).
6.4. Extension to Levi subalgebras. In analogy with [TL4, Thm. 9.1],
we want to show that the relative twists and the Casimir associators are weight
zero elements. This corresponds to show that the corresponding tensor func-
tors Γ and the natural transformations γ lift to the level of Levi subalgebras:
gD ⊂ lD = nD,+ ⊕ h⊕ nD,− ⊂ g
Proposition. The relative twists and the Casimir associators are weight
zero elements.
Proof. For D = ∅, the statement reduces to prove that the Etingof–
Kazhdan functor preserves the h–action [EK6, Thm. 4.3]. The result is a
consequence of Proposition 5.3 applied to Levi subalgebras.
6.5. Reduction to category Oint. The Etingof–Kazhdan functor gives
rise, by restriction, to an equivalence of categories
F : Og[[~]]→ OU~g
134 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
We will show now that this equivalence can be further restricted to integrable
modules in category O, i.e., modules in category O with a locally nilpotent
action of the elements ei, fii∈I (respectively Ei, Fi).
Proposition. The Etingof–Kazhdan functor restricts to an equivalence of
braided tensor categories
F : Ointg [[~]]→ Oint
U~g
which is isomorphic to the identity functor at the level of h–graded k[[~]]–
modules.
Proof. Let V ∈ Ointg . Then, the elements ei, fi for i ∈ I act nilpotently
on V . Then, by [Ka], for all λ ∈ P(V ), there exist p, q ∈ Z≥0 such that
t ∈ Z | λ+ tαi ∈ P(V ) = [−p, q]
Since the Cartan subalgebra h is not deformed by the quantization, the functor
F preserves the weight decomposition. In U~g, for any h ∈ h and i ∈ I, we
have
[h,Ei] = αi(h)Ei
Therefore the action of the Ei’s on V is locally nilpotent. The action of the
Fi’s is always locally nilpotent, since
P(V ) ⊂r⋃s=1
D(λs)
The result follows.
6. THE EQUIVALENCE OF QUASI–COXETER CATEGORIES 135
Corollary. .
(i) There exists an equivalence of braided Dg–monoidal categories be-
tween
O := ((OintgB,⊗B,ΦB, σRB), (ΓBB′ , JBB
′
F ))
and
O~ := ((OintU~gB,⊗B, id, σR~B), (Γ~BB′ , id))
(ii) There exists an isomorphism of Dg–algebras
ΨF : U~g→ Ug[[~]]
such that ΨF(U~gDi) = UgDi [[~]] for any Di ∈ F , where the comple-
tion is taken with respect to the integrable modules in category O.
6.6. Quasi–Coxeter structure. The previous equivalence of braided
Dg–monoidal categories induces on
O = ((OintgB
[[~]],⊗B,ΦB, σRB), (ΓBB′ , JBB′
F ))
a structure of quasi–Coxeter category of tipe Dg, given by the Casimir asso-
ciators ΦGF ∈ Nat⊗(ΓF ,ΓG) and the local monodromies Si ∈ End(Γi) defined
for any G,F ∈ Mns(B,B′) and i ∈ I(D) by
FB(ΦGF) = (γFBB′)−1 γGBB′ Si = ΨEK
i (S~i )
136 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
where ΨEKi : U~sl
i2 → Usli2[[~]] is the isomorphism induced at the sli2 level by
the Etingof–Kazhdan functor.
Proposition. The equivalence of braided Dg–monoidal categories O →
O~ induces a structure of quasi–Coxeter category on O.
Proof. In order to prove the proposition, we have to prove the compatibil-
ity relations of the elements ΦGF , Si with the underlying structure of braided
Dg– monoidal category on O.
The element Si’s satisfy the relation
∆F(Si) = (Ri)21F · (Si ⊗ Si)
since ΨF is given by an isomorphism of braided monoidal D–categories and
therefore
ΨF((R~i )F) = (Ri)F
Similarly, the braid relations are easily satisfied, since
Ad(ΦGF)ΨF = ΨG
The elements ΦFG defined above satisfy all the required properties:
(i) Orientation For any elementary pair (F ,G) in Mns(B,B′)
FB(ΦFG) = (γFBB′)−1 γGBB′ =
(FB(ΦGF)
)−1
6. THE EQUIVALENCE OF QUASI–COXETER CATEGORIES 137
(ii) Coherence For any F ,G,H ∈ Mns(B,B′)
FB(ΦFG) = (γFBB′)−1γHBB′ (γHBB′)
−1 γGBB′ =
=FB(ΦFH) HB(ΦHG)
This property implies the coherence.
(iii) Factorization. Clear by construction.
Finally, the elements ΦGF satisfy
∆(ΦGF) JF = JG Φ⊗2GF
because they are given by composition of invertible natural tensor transforma-
tions.
6.7. Normalized isomorphisms. In the completion Usli2[[~]] with re-
spect to category O integrable modules, there are preferred element Si,C
Si,C = si exp(~2Ci)
where
si = exp(ei) exp(−fi) exp(ei) Ci =(αi, αi)
2(eifi + fiei +
1
2h2i )
Proposition. There exists an equivalence of quasi–Coxeter categories of
type Dg between
O := ((OintgB,⊗B,ΦB, σRB), (ΓBB′ , JBB
′
F ), ΦGF, Si,C)
138 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
and
O~ := ((OintU~gB,⊗B, id, σR~B), (Γ~BB′ , id), id, S~i )
Proof. Using the result of Proposition 6.6, it is enough to prove that the
natural transformation γi
Ointi
Fi//
f
Ointi,~
f~~~
γipx
A
can be modified in such a way that the induced isomorphism at the level of
endomorphism algebras U~sli2 → Usli2[[~]] maps S~i to Si,C . The natural trans-
formation used in Corollary 6.5 induces the Etingof–Kazhdan isomorphism
ΨEK
i : U~sli2 → Usli2[[~]]
which is the identity mod ~ and the identity on the Cartan subalgebra. As
above, we denote by Si the element ΨEKi (S~i ). Then Si ≡ si mod ~ and, by
[TL4, Proposition 8.1, Lemma 8.4], we have
S2i = S2
i,C Si = Ad(x)(Si,C)
on the integrable modules in category O, for x = (Si,C · S−1i )
12 . Therefore, the
modified isomorphism
Ψi := Ad(x) ΨEK
i
6. THE EQUIVALENCE OF QUASI–COXETER CATEGORIES 139
maps S~i to Si,C . Moreover, Ψi correspond with the natural transformation
given by the composition of γi with x ∈ Usli2[[~]] = End(f)
Ointi
f
Ointi
F//
f
Ointi,~
f~
γiks
xks
A A A
The result follows substituting γi with x γi in Proposition 6.6.
6.8. The main theorem. We now state in more details the main theorem
of this Chapter and summarize the proof outlined in the previous results.
Theorem. Let g be a symmetrizable Kac–Moody algebra with a fixed Dg–
structure and U~g the corresponding Drinfeld–Jimbo quantum group with the
analogous Dg–structure. For any choice of a Lie associator Φ, there exists an
equivalence of quasi–Coxeter categories between
O := ((OintgB,⊗B,ΦB, σRB), (ΓBB′ , JBB
′
F ), ΦGF, Si,C)
and
O~ := ((OintU~gB
,⊗B, id, σR~B), (Γ~BB′ , id), id, S~i )
where ⊗B denotes the standard tensor product in OintgB
and
Si,C = si exp(h
2· Ci)
ΦB = 1 mod ~2
140 3. AN EQUIVALENCE OF QUASI–COXETER CATEGORIES
RB = exp(~2
ΩD)
Alt2JBB′
F =~2
(rB′ − r21B′
2− rB − r21
B
2
)and ΦGF , JBB
′F are weight zero elements.
Proof. The existence of an equivalence is a consequence of the construc-
tions of Section 5 and proved in Theorem 6.3 and Proposition 6.6, 6.7, con-
cerning the local monodromies Si,C .
The properties of associators ΦB and R–matrices RB are direct conse-
quences of the construction in Section 1,2. The relation satisfied by the relative
twists JBB′
F is proven by a simple application of Proposition 2.20 and Corollary
2.20. It is easy to check that the 1–jet of the twist JBB′
F differs from the 1–jet
of the twist JBB′(as defined in Section 2) by a symmetric element that cancels
out computing the alternator. Therefore, Corollary 2.20 holds for JBB′
F as well.
Finally, as previously explained, the weight zero property of the relative
twists JBB′
F and the Casimir associators ΦGF is proved in Proposition 5.3, 6.4.
This complete the proof of Theorem 6.8.
CHAPTER 4
A rigidity theorem for quasi–Coxeter categories
The goal of this chapter is to show that, under appropriate hypotheses,
there is at most one quasi–Coxeter braided monoidal category on the classi-
cal enveloping algebra Ug[[~]] with prescribed local monodromies (Thm 4.2).
This result is needed to show that the quasi–Coxeter braided monoidal cate-
gory naturally attached to the quantum affine algebra U~g is equivalent to the
one constructed for Ug[[~]], which underlies the monodromy of the rational
KZ and Casimir connections of g.
In Section 1, we review a number of basic notions from [TL4], in particular
that of the Dynkin–Hochschild bicomplex of a D–bialgebra A which controls
the deformations of quasi–Coxeter quasitriangular quasibialgebra structures
on A. In Section 2, we extend the notion of the bicomplex to the case of com-
pletion of algebras, using the language of endomorphisms algebra of monoidal
functors and the corresponding cohomology theory. In Section 3 we study in-
variant properties of the relative twist under the Chevalley involution of g and
we prove a uniqueness property up to gauge transformation. Section 4 contains
the rigidity result, consisting in the construction of a natural transformation
matching the Casimir associators and preserving the local monodromies.
141
142 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
1. The Dynkin-Hochschild bicomplex
In this section, we briefly recall the definition of D–bialgebras and the con-
struction of the Dynkin–Hochschild complex introduced in [TL4].
1.1. The Dynkin complex of a D–algebra. We begin by defining the
category of coefficients of the Dynkin complex of D–algebra A (2.2).
Definition. A D–bimodule over A is an A–bimodule M , with left and
right actions denoted by am and ma respectively, endowed with a family of
subspaces MD1 indexed by the connected subdiagrams D1 ⊆ D of D such that
the following properties hold
• for any D1 ⊆ D,
AD1 MD1 ⊆MD1 and MD1 AD1 ⊆MD1
• For any pair D2 ⊆ D1 ⊆ D,
MD2 ⊆MD1
• For any pair of orthogonal subdiagrams D1, D2 of D, aD1 ∈ AD1 and
mD2 ∈MD2 ,
aD1mD2 = mD2aD1
A morphism of D–bimodules M,N over A is an A–bimodule map T : M → N
such that T (MD) ⊆ ND for all D ⊆ D.
Clearly, A is a D–bimodule over itself. We denote by BimodD(A) the
abelian subcategory of Bimod(A) consisting of D–bimodules over A. If M ∈
1. THE DYNKIN-HOCHSCHILD BICOMPLEX 143
BimodD(A), and D1 ⊆ D is a subdiagram, we set
MD1 = m ∈M | am = ma for any a ∈ ADi1
where Di1 are the connected components of D1. In particular, if D1, D2 ⊆ D
are orthogonal, and D1 is connected, then MD1 ⊆MD2 .
Let M ∈ BimodD(A). For any integer 0 ≤ p ≤ n = |D|, set
Cp(A;M) =⊕
α⊆D1⊆D,|α|=p
MD1\αD1
where the sum ranges over all connected subdiagrams D1 of D and ordered
subsets α = α1, . . . , αp of cardinality p of D1 and MD1\αD1
= (MD1)D1\α. We
denote the component of m ∈ Cp(A;M) along MD1\αD1
by m(D1;α).
Definition. The group of Dynkin p–cochains on A with coefficients in M
is the subspace CDp(A;M) ⊂ Cp(A;M) of elements m such that
m(D1;σα) = (−1)σm(D1;α)
where, for any σ ∈ Sp, σα1, . . . , αp = ασ(1), . . . , ασ(p).
Note that
CD0(A;A) =⊕D1⊆D
Z(AD1) and CDn(A;M) ∼= MD
For 1 ≤ p ≤ n− 1, define a map dpD : Cp(A;M)→ Cp+1(A;M) by
dDm(D1;α) =
p+1∑i=1
(−1)i−1
(m(D1;α\αi) −m(
D1\αiα\αi
;α\αi)
)
144 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
where α = α1, . . . , αp+1, D1\αiα\αij
is the connected component of D1 \ αi con-
taining α \ αi if one such exists and the empty set otherwise, and we set
m(∅;−) = 0. For p = 0, define d0D : C0(A;M)→ C1(A;M) by
d0Dm(D1;αi) = mD1 −mD1\αi
where mD1\αi is the sum of mD2 with D2 ranging over the connected com-
ponents of D1 \ αi. Finally, set dnD = 0. It is easy to see that the Dynkin
differential dD is well–defined and that it leaves CD∗(A;M) invariant.
Theorem. (CD∗(A;M), dD) is a complex. The cohomology groups
HDp(A;M) = Ker(dpD)/ Im(dp−1D )
for p = 0, . . . , |D| are called the Dynkin diagram cohomology groups of A with
coefficients in M .
1.2. The Dynkin–Hochschild bicomplex of a D–bialgebra. Let D
be a connected diagram and A a D–bialgebra (the algebraic counterpart of
strict D–monoidal categories 2.7, cf. [TL4, 6.1]). By combining the Dynkin
complex of A with the cobar complexes of its subalgebras AB, we review the
definition of a bicomplex which controls the deformations of quasi–Coxeter
quasibialgebra structures on A introduced in [TL4, 7.1].
Let A be a bialgebra and C∗(A) the cobar complex of A, regarded as a
coalgebra. If C ⊆ A is a sub–bialgebra, let
Cn(A)C = a ∈ Cn(A)| [a,∆(n)(c)] = 0 for any c ∈ C
2. DYNKIN–HOCHSCHILD COMPLEX OF MONOIDAL FUNCTORS 145
where ∆(n) : C → C⊗n is the nth iterated coproduct, be the submodule of
C–invariants. It is easy to check that C∗(A)C is a subcomplex of C∗(A).
Assume now that A is a D–bialgebra. For p ∈ N and 0 ≤ q ≤ |D|, let
CDq(A;A⊗p) ⊂⊕
α⊆B⊆D,|α|=q
(A⊗pB )B\α
be the group of Dynkin q–cochains with values in the D–bimodule A⊗p over
A. The Dynkin differential dDD defines a vertical differential CDq(A;A⊗p)→
CDq+1(A;A⊗p) while the Hochschild differential dH defines a horizontal dif-
ferential CDq(A;A⊗p) → CDq(A;A⊗(p+1)). A straightforward computation
yields the following.
Theorem. One has dDD dH = dH dDD. The corresponding cohomology
of the bicomplex CDq(A;A⊗p) is called the Dynkin–Hochschild cohomology of
the D–bialgebra A.
2. Dynkin–Hochschild complex of monoidal functors
We use the language of monoidal functor and cohomology of monoidal
functor to reformulate the definition of the Dynkin–Hochschild bicomplex in a
more general setting. This approach allows to consider the Dynkin–Hochschild
cohomology for completions of bialgebras with respect to a fixed tensor sub-
category.
2.1. Hochschild complex of monoidal functors. Let F : C → C ′ be
a monoidal functor of associative monoidal k–linear categories with associated
146 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
tensor structure J . The nth tensor power of F is the functor
F⊗n : Cn → C ′ F⊗n(X1, . . . , Xn) =n⊗i=1
F (Xi)
Through repeated applications of J , we obtain a canonical isomorphism
F⊗n(X1, · · · , Xn) ' F (n⊗i=1
Xi)
The k–algebra of endomorphisms End(F⊗n) form the cosimplicial complex of
algebras
EndC′(1) //// End(F ) ////// End(F⊗2) //
//////
End(F⊗3) · · ·
with face homomorphisms din : End(F⊗n) → End(F⊗n+1), i = 0, . . . , n + 1,
defined as follows:
(d00f)X : F (X)→ F (X)⊗ 1→ F (X)⊗ 1→ F (X)
(d10f)X : F (X)→ 1⊗ F (X)→ 1⊗ F (X)→ F (X)
for X ∈ C, f ∈ EndC′ 1, and
(dinf)X1,...,Xn+1 =
id⊗fX2,...,Xn+1 i = 0
Ad(JXiXi+1)fX1,...,Xi⊗Xi+1,...Xn+1 1 ≤ i ≤ n
fX1,...,Xn ⊗ id i = n+ 1
for f ∈ End(F⊗n), Xi ∈ C, i = 1, . . . , n + 1. The Hochschild differential is
defined in the usual way:
dn =n+1∑i=0
(−1)idin : End(F⊗n)→ End(F⊗n+1)
2. DYNKIN–HOCHSCHILD COMPLEX OF MONOIDAL FUNCTORS 147
The degeneration homomorphisms sin : End(F⊗n) → End(F⊗n−1) are de-
fined, for i = 1, . . . , n, by
(sinf)X1,...,Xn−1 := fX1,...,Xi−1,1,Xi,...,Xn−1
The morphisms sin, din satisfy the standard relations
djn+1din = din+1d
j−1n i < j
sjnsin+1 = sins
j+1n+1 i ≤ j
sjn+1din =
din−1s
j−1n i < j
id i = j, j + 1
di−1n−1s
jn i > j + 1
2.2. Composition of monoidal functor and Hochschild subcom-
plexes. Let F : C → C ′ and G : C ′ → C ′′ be monoidal functors of associative
monoidal categories with associated tensor structures J and K, respectively.
For any n ∈ N there are well– defined homomorphisms
End(F⊗n)⊗ End(G⊗n)→ End((G F )⊗n)
mapping the tensor α ⊗ β into α β ∈ End((G F )⊗n) defined to be the
composition α β = Ad(K)(G(α)) β
⊗ni=1 GF (Xi)
βF//⊗n
i=1GF (Xi)
//⊗n
i=1 GF (Xi)
G(⊗n
i=1 F (Xi))G(α)// G(
⊗ni=1 F (Xi))
OO
The functor G F is naturally endowed with a tensor structure induced by J
and K. A quick computation proves the following
148 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
Proposition.
(i) The composition maps are compatible with the Hochschild differential
maps, i.e., the diagram
End(F⊗n)⊗ End(G⊗n)
dFn⊗dGn
n// End((G F )n)
dGFn
End(F⊗n+1)⊗ End(G⊗n+1)n+1// End((G F )⊗n+1)
is commutative for any n ∈ N.
(ii) The natural maps
End(F⊗n)→ End((G F )⊗n) End(G⊗n)→ End((G F )⊗n)
are well–defined morphisms of the associated cochain complexes.
Remark. If F = G = id, then the composition law described above
coincides with the multiplication in End(id⊗n).
2.3. Cohomology of monoidal functors. Let C be a k–linear abelian
monoidal category and F : C → Vect a faithful, exact, associative monoidal
functor. Then we can consider on End(F ) the initial topology with basis of
neighborhood of zero the kernels of the maps End(F )→ End(F (X)). We get
canonical isomorphisms
End(F ) ' limX∈C
End(F (X))
2. DYNKIN–HOCHSCHILD COMPLEX OF MONOIDAL FUNCTORS 149
and
End(F⊗n) ' End(F )⊗n
where ⊗ denotes the completion of the tensor product with respect to the
topology of the inverse limit. This identification allows in particular to look
at the canonical map
∆ : End(F )→ End(F⊗2) ' End(F )⊗End(F )
as a topological coproduct on End(F ).
Definition. The cohomology of the cochain complex (End(F⊗•), d) is
called additive cohomology of the monoidal functor F and denoted H•(F ).
The external multiplication
End(F⊗n)⊗ End(F⊗m)→ End(F⊗n+m)
defined in the obvious way on the complex End(F⊗•) induces a structure of
graded ring on H•(F ). This follows from the relations
dn+mi =
dni (α) · β i ≤ n
α · dmn−i(β) i > ndnn+1(α) · β = α · dm0 (β)
The differential graded algebra (End(F⊗•),⊗, d) is homotopically commutative
[D]. Thus the cohomology Hn(F ) is a skew–commutative graded algebra.
Moreover, the homotopy of the commutativity defines on H•(F ) a structure of
Lie superalgebra J·, ·K : Hn(F )⊗Hm(F )→ Hn+m−1(F ), which coincides on the
150 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
subalgebra∧• H1(F ) with the Schouten bracket
Ju1∧· · ·∧un, v1∧· · ·∧vmK =∑i,j
(−1)i+j[ui, vj]∧u1∧· · ·∧ui∧. . . un∧v1∧· · ·∧vj∧· · ·∧vm
2.4. Cohomology of the forgetful functor in category O. Let g be
a Lie algebra over a field k, with char k = 0, and F : RepUg → Vect be the
forgetful functor. Tannaka–Krein duality gives a canonical isomorphism Ug '
End(F ). The cochain complex End(F⊗•) coincides with the co–Hochschild
complex of Ug and H1(F ) = Prim(Ug) = g. Since dH(g⊗n) = 0, we obtain
a map from T •g → H•(F ) and we denote by µ the restriction to the exterior
algebra
µ :•∧g→ H•(F )
It is well–known, in this case, that the map µ is an isomorphism [D3].
As mentioned above, for an arbitrary monoidal functor F : C → C ′ we have
a canonical inclusion
µ :•∧H1(F )→ H•(F )
observing that dH(H1(F )⊗n) = 0. For an arbitrary monoidal functor, the mor-
phism µ is not surjective. Easy counterexamples are given by the identity func-
tor on RepUg (cocommutative case) or the forgetful functor RepH → Vect
where H is the Sweedler’s Hopf algebra (non cocommutative). In both cases
we have H1(F ) = 0 and H•(F ) 6= 0.
Our main interest lies in the forgetful functor F : Og → Vect, where g is
a Kac–Moody algebra and Og is the corresponding category O. In this case,
2. DYNKIN–HOCHSCHILD COMPLEX OF MONOIDAL FUNCTORS 151
we assume that there exists an appropriate completion of the exterior algebra
that extends µ to an isomorphism.
Conjecture. Let g be a Kac–Moody algebra and F : Og → Vect be the
forgetful functor from the category O, then
H•(F ) '∧•
g H1(F )h ' h
2.5. Dynkin complex of D–categories. Let C = (CB, FBB′)B,B′∈SD(D)
be a D–category. We recall that, by definition, for any B ⊂ B′ ⊂ B′′, we are
given maps
End(FBB′)→ End(FBB′′) End(FB′B′′)→ End(FBB′′)
and, for any B′ ⊥ B′′, B′, B′′ ⊂ B′′′, and B ⊂ B′,
End(FBB′)→ End(F(B∪B′′)B′′′)
For any integer 0 ≤ p ≤ n = |D|, we set
Cp(C) =⊕
α⊆B∈Csd(D),|α|=p
End(F(B,α))
where F(B;α) := F(B\α)B.
Definition. The group of Dynkin p–cochains on C is the subspace CDp(C) ⊂
Cp(C) of elements m such that
f(B;σα) = (−1)σf(B;α)
where, for any σSp, σα1, . . . , αp = ασ(1), . . . , ασ(p).
152 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
Note that
CD0(C) =⊕B⊂D
End(idCB) CDn(C) = End(FD)
For 1 ≤ p ≤ n− 1, we define the map dpD : CDp(C)→ CDp+1(C) by
(dDf)(B;α) =
p+1∑i=1
(−1)i−1(f(B;α\αi) − f(
B\αiα\αi
;α\αi)
)as in Section 1.1. By abuse of notation, by f(B;α\αi) we mean its image under
the map
End(F(B;α\αi))→ End(F(B;α))
and by f(B\αiα\αi
;α\αi)its image under the map
End(F(B\αiα\αi
;α\αi))→ End(F(B\αi;α\αi))→ End(F(B;α))
For p = 0, define d0D : CD0(C)→ CD1(C) by
d0Df(B;α) = fB − fB\α
where fB\α is the sum over B′, the connected components of B \ α, of fB′ .
Finally, we set dnD = 0. The Dynkin differential is well–defined and preserves
CD∗(C).
Proposition. (CD∗(C); dD) is a cochain complex.
2.6. Dynkin-Hochschild bicomplex for associative D–monoidal cat-
egories. We now combine the Dynkin complex for a D–category and the
3. UNIQUENESS OF THE RELATIVE TWIST 153
Hochschild complex for monoidal functors to define a suitable bicomplex for as-
sociative D–monoidal categories, which controls the deformation of the quasi–
Coxeter monoidal structure.
For p ∈ N and 0 ≤ q ≤ |D|, we denote by
CDp(Cq) ⊂⊕
α⊆B∈Csd(D),|α|=p
End(F⊗q(B,α))
the group of Dynkin p–cochains for Cq. The Dynkin differential dD defines a
vertical differential
dpD : CDp(Cq)→ CDp+1(Cq)
while the Hochschild differential dH defines a horizontal differential
dqH : CDp(Cq)→ CDp(Cq+1)
A simple computation yields the following.
Proposition. One has dD dH = dH dD.
The corresponding cohomology of the bicomplex CDq(Cp) is called the
Dynkin–Hochschild cohomology of the monoidal D–category C. As in [TL4,
7.2], the DynkinHochschild bicomplex of C controls the formal, one- parameter
deformations of the trivial quasi-Coxeter monoidal structure on C.
3. Uniqueness of the relative twist
3.1. Chevalley invariance. We denote by θ the Chevalley involution on
g and the induced morphism on the Manin triple attached to g. In this section
154 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
we will study the following invariant property of a twist under θ:
Jθ = J21 (3.1)
Proposition. The Etingof–Kazhdan twist JEKfd , defined in [EK1, Part I],
satisfies
(JEK
fd )θ = (JEK
fd )21
Proof. The twist JEKfd defines a tensor structure on the functor hM+⊗M−
and it can be identified with the action on V ⊗W of the element J ∈ Ug⊗2[[~]]
J := (ϕ⊗ ϕ)−1(Φ−11,2,34 · Φ234 · e
~2
Ω23 · Φ−1324 · Φ1,3,24)(1+ ⊗ 1− ⊗ 1+ ⊗ 1−)
where ϕ : Ug∼→ M+ ⊗M−. The involution θ defines an autoequivalence of
DΦ(g) and a natural identification
hM+⊗M− θ∗ = hMθ+⊗Mθ
−
Using the isomorphism M θ± ' M∓ and the relations Ωθ = Ω, Φθ = Φ, we
observe that the twist induced on the functor hM+⊗M− θ∗, denoted (JEK)θ, is
given by the same formula defining JEK on the functor hM−⊗M+ . It corresponds
to the action on V ⊗W of the element
Jθ := (ψ ⊗ ψ)−1(Φ−11,2,34 · Φ234 · e
~2
Ω23 · Φ−1324 · Φ1,3,24)(1− ⊗ 1+ ⊗ 1− ⊗ 1+)
3. UNIQUENESS OF THE RELATIVE TWIST 155
where ψ : Ug∼→ M− ⊗M+. It is easy to check that ϕ ψ−1 : M− ⊗M+ →
M+ ⊗M− corresponds to the permutation σ12. We then obtain
J21 =(ψ−1 ⊗ ψ−1)σ14σ23(Φ−11,2,34 · Φ234 · e
~2
Ω23 · Φ−1324 · Φ1,3,24)(1+ ⊗ 1− ⊗ 1+ ⊗ 1−) =
=(ψ−1 ⊗ ψ−1)(Φ−14,3,21 · Φ321 · e
~2
Ω23 · Φ−1231 · Φ4,2,31)(1− ⊗ 1+ ⊗ 1− ⊗ 1+) =
=(ψ−1 ⊗ ψ−1)(Φ12,3,4 · Φ−1123 · e
~2
Ω23 · Φ132 · Φ13,2,4)(1− ⊗ 1+ ⊗ 1− ⊗ 1+) = Jθ
where the last equivalence follows by the pentagon axiom. Alternatively, we
can also notice that J21 can be thought as the twist of the functor F EK with
respect to the opposite tensor product, naturally identified with hM−⊗M+ .
The Etingof–Kazhdan functor F EK can be naturally identified with hM+⊗M− ,
in the finite dimensional setting, via the natural transformation χ : F EK →
hM+⊗M−
χV (v) := (ev ⊗ id) Φ−1 (id⊗v)
The identification preserves the tensor structure, but it does not satisfy χθ = χ.
We therefore conclude that the twist JEK does not satisfy 3.1. The following
lemma shows that we are allowed to weaken the requirement.
Lemma. Let u ∈ End(F )[[~]] an invertible natural transformation of the
tensor functor (F, J) satisfying
u ? Jθ = J21
where u ? J := ∆(u−1) · J · u ⊗ u. If uθ = u−1, J is gauge equivalent to a
solution of 3.1.
156 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
Proof. We first prove that the condition uθ = u−1 is equivalent to the
existence of an invertible element v ∈ End(F )[[~]] such that u = v−1vθ. One
implication is trivial.
We have u ≡ 1 mod ~ and uθ = u−1. Denote u = 1 +∑
n≥0 un~n, so that
uθ1 = −u1. The element v1 := −u1/2 satisfies
vθ1 − v1 = u1
Consider the natural transformation
u(2) := (1 + v1~) · u · (1 + vθ1~)−1
Then
u(2) ≡ 1 mod ~2 (u(2))θ = (u(2))−1
Assume we constructed u(n) satisfying u(n) ≡ 1 mod ~n and (u(n))θ = (u(n))−1.
Similarly, consider the element vn = −u(n)n /2 and
u(n+1) := (1 + vn~n) · u(n) · (1 + vθn~n)−1
Then u(n+1) satisfies u(n+1) ≡ 1 mod ~n+1 and (u(n+1))θ = (u(n+1))−1. By
induction, we obtain
u =∏n≥0
(1 + vn~n)−1 ·∏n≥0
(1 + vn~n)θ
and we denote by v the natural transformation∏
n≥0(1 + vn~n). The factor-
ization is uniquely defined up to a natural transformation w ∈ End(F )[[~]]
3. UNIQUENESS OF THE RELATIVE TWIST 157
invariant under θ.
Finally, the relation u ? Jθ = J21 and the factorization u = v−1vθ imply
(v ? J)θ = (v ? J)21
An easy computation on the 2–jet of JΓ (cf. 2.15) shows that the relative
twist does not satisfy, in the finite dimensional setting, the condition 3.1. We
observe that the natural transformation constructed in 5.1 gives rise to a gauge
transformation relating JΓ and the composition of JEK and (JEKD )−1. We get
the following
Corollary. In the finite–dimensional setting, there exists a natural trans-
formation u ∈ End(Γ) such that
(u ? JΓ)θ = (u ? JΓ)21
where (u ? J)V⊗W := u−1V⊗WJVWuV ⊗ uW for any V,W ∈ DΦ(g).
3.2. Uniqueness property. We now prove a uniqueness property of the
relative twist. By 2.4 it is easy to check that
H2S(Γ) := H2((
∧H1(F ))lD , Jr − rD, ·K) = (
∧2lD)lD =
∧2cD
Theorem. Let Ji, i = 1, 2 be two tensor structures on the functor Γ :
Oint(g)→ Oint(lD) such that Ji ≡ 1⊗2 + ~fi mod ~ and
Alt2Ji ≡ 1⊗2 +~2
(r − r21
2− rD − r21
D
2
)mod ~2 i = 1, 2
158 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
Then there exist
u ∈ End(Γ) λ ∈ H2S(Γ)
such that
expλ · J2 = u ? J1
In the finite dimensional case, under the additional assumption that
Jθi = J21i
we have λ = 0 and u is uniquely determined by the property uθ = u.
Proof. Denote
r =r − r21
2rD =
rD − r21D
2
and assume
Alt2fi =1
2(r − rD) + νi νi ∈ H2
S(Γ)
Then there exist gi ∈ End(Γ) such that
fi − Alt2fi = dHgi
Therefore, we can replace Ji with
exp(−~νi)(1− ~gi) ? Ji
and assume that fi = 1/2(r − rD). In the particular case of the relative twist
JΓ we have
f − Alt2f =1
4(Ω− ΩD) =
1
4dH(m(rD)−m(r))
3. UNIQUENESS OF THE RELATIVE TWIST 159
where m(r) is the element obtained by multiplication of the components of
the r–matrix.
The additional assumption Jθi = J21i implies
dHgθi + νθi = dHgi + ν21
i ⇒ dHgθi = dHgi, νθi = ν21
i
since dHgi is symmetric and (r− rD)θ = −(r− rD) = (r− rD)21. Therefore, we
have νi = 0, since θ acts as the identity on∧2cD, and we may assume gθi = gi,
replacing gi with 12(gi + gθi ).
We will now construct inductively two sequences
vn ∈ End(Γ) µn ∈ H2S(Γ)
such that
J2 ≡ exp(λn)(un ? J1) mod ~n+1
where
un =n∏i=1
(1 + ~nvn) λn =n∑i=1
~iµi
Since we reduced both twists to the form
Ji ≡ 1⊗2 +~2
(r − rD) mod ~2
we may assume v1 = 0, µ1 = 0. Therefore, we assume elements vk, µk defined
for k = 1, . . . , n, n ≥ 1. Denote J ′1 the element exp(λn)(un ? J1) so that
J2 = J ′1 + ~n+1η mod ~n+2 η ∈ End(Γ⊗2)
160 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
We clearly have ΦJ ′1= ΦD = ΦJ2 . Computing mod ~n+2 we obtain
1⊗ η − (∆⊗ id)(η) + (id⊗∆)(η)− η ⊗ 1 = 0
i.e., dHη = 0. Moreover, if Jθi = J21i , we get ηθ = η21. Then, there exist
v ∈ End(Γ) and µ ∈ H2(Γ)= (∧2H1(F ))gD such that η = dHv + µ and
vθ = v µθ = µ21= −µ
Therefore, we set vn+1 = −v and
J ′′1 =(1 + ~n+1vn+1) ? J ′1 =
=(1 + ~n+1vn+1) ? (exp(λn) · (un ? J1)) =
= exp(λn) ·((1 + ~n+1vn+1) · un
)? J1
since clearly vn+1 and λn commute. We now have
J2 ≡ exp(−~n+1µ) · J ′′1 mod ~n+2
So we set µn+1 = −µ and it remains to show that µ lies in H2S(Γ).
Consider the truncation of J2 mod ~n+2
J(n+2)2 = 1⊗2 +
~2
(r − rD) +n+1∑i=2
~ifi
Then we have
ΦJ
(n+2)2
≡ ΦD + ~n+2ξ mod ~n+3
3. UNIQUENESS OF THE RELATIVE TWIST 161
for some ξ ∈ End(Γ⊗3). Since J(n+2)2 is the truncation of a relative twist mod-
ulo ~n+3, we have dHξ = 0 and Alt3ξ = 0. Similarly, we can consider the
truncation of J ′′1 modulo ~n+2, J ′′1(n+2), and denote by ξ′′ the corresponding
error. Then it satisfies dHξ′′ = 0 and Alt3ξ
′′ = 0 (cf. [TL3, 5.4]).
Now, since J ′′1(n+2) = J
(n+2)2 + ~n+1µ mod ~n+2, denote by f = 1
2(r − rD)
and we obtain
ξ′′ − ξ = f 23(µ12 + µ13) + µ23(f 12 + f 13)− f 12(µ13 + µ23)− µ12(f 13 + f 23)
then we get Jf, µK = 0. Then we have
µ = [[f, x]] + y for some x ∈ H1(F )lD = cD ⊂ h, y ∈ H2S(Γ)
Since f is a weight zero element, we have Jf, xK = − ad(x)f = 0. We conclude
that
µ ∈ H2S(Γ) = H2((
∧H1(F ))lD , Jr − rD, ·K)'
∧2cD[[~]]
Since we are assuming Jθi = J21i , we get µθ = −µ and µ = 0.
Let u ∈ End(Γ) be such that
u ? J1 = J1
with uθ = u. Assume u ≡ 1 mod ~n and write u ≡ 1 + ~nun mod ~n+1, with
uθn = un. Taking the coefficient of ~n+1 we get dHun = 0, implying un ∈ h,
since it is of weight zero. Then we get un = 0, θ acting on h by −1.
162 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
4. The main theorem
4.1. The D–bialgebra structure on Ug. Let A = (aij)i,j∈I be an irre-
ducible, symmetrizable, generalised Cartan matrix of affine type and g = g(A)
the corresponding affine Kac–Moody algebra [Ka]. Let Dg = D(A) be the
Dynkin diagram of g, that is the connected graph having I as its vertex set
and an edge between i and j if aij 6= 0. For any i ∈ I, let sli2 ⊆ g be the three–
dimensional subalgebra spanned by ei, fi, hi. In this case, any proper connected
subdiagram of Dg is a Dynkin diagram of finite type. Therefore, for any proper
subdiagram D ⊂ Dg, we consider the subalgebra gD spanned by ei, fi, hii∈D.
For D = Dg, we consider the entire algebra g. Then (Ug, UgD) is a Dg-
algebra over C.
4.2. The rigidity theorem. Extend the bilinear form (·, ·) on h to a
non–degenerate, symmetric, bilinear, ad–invariant form on g. For any proper
connected subdiagram D ⊂ Dg, let gD ⊂ lD ⊂ g be the corresponding simple
and Levi subalgebras. Denote by
ΩD = xa ⊗ xa, CD = xa · xa and rgD =∑α0:
supp(α)⊆D
(α, α)
2· eα ∧ fα
where xaa, xaa are dual basis of gD with respect to (·, ·), the correspond-
ing invariant tensor, Casimir operator and standard solution of the modified
classical Yang–Baxter equation for gD respectively. Abbreviate slαi2 , Ωαi and
Cαi to sli2, Ωi and Ci respectively and let si be the triple exponentials
si = exp(ei) exp(−fi) exp(ei)
4. THE MAIN THEOREM 163
where ei = eαi , fi = fαi are a fixed choice of simple root vectors.
In the next sections 4.3-4.5, we outline the proof of the following
Theorem. Up to equivalence, there exists a unique quasi-Coxeter braided
monoidal D–category of type Dg of the form
(Oint
g [[~]], OintgD
[[~]],⊗, RD,ΦD, Si,c, Φ(D;αi,αj), F(D;αi), J(D;αi))
where ⊗ is the standard tensor product in Og, (F(D;αi), J(D;αi)) are fiber func-
tors,
Si,c = si · exp (~/2 · Ci) (4.1)
RD = exp(~/2 · ΩD) (4.2)
Alt2 J(D;αi) =~2
(rgD − rgD\αi
)mod ~2 (4.3)
and Φ(D;αi,αj), J(D;αi) are of weight 0, i.e., they properly extend to the corre-
sponding Levi’s subalgebras.
4.3. Proof of Theorem 4.2. Let
(Oint
g [[~]], OintgD
[[~]],⊗, RD,ΦkD, Si,c, Φk
(D;αi,αj), F(D;αi), J
k(D;αi)
)
k = 1, 2 be two quasi–Coxeter braided monoidal categories.
By Drinfeld’s uniqueness, [TL3, TL4] and the uniqueness of the twist, we
may assume that
Φ2D = Φ1
D J2(D;αi)
= J1(D;αi)
164 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
and that
Φ1(D;αi,αj)
= Φ2(D;αi,αj)
+ ~nϕ(D;αi,αj) mod ~n+1
for some ϕ(D;αi,αj) ∈ End(FD\αi,αj).
Let αi 6= αj ∈ D ⊆ Dg and F ,G ∈ Mns(Dg) such that
F \ G = D\αiαjG \ F = D\αjαi
Subtracting the equations
JF ·∆(Φ1(D;αi,αj)
) = Φ2(D;αi,αj)
⊗2 · JG
JF ·∆(Φ2(D;αi,αj)
) = Φ1(D;αi,αj)
⊗2 · JG
and equating the coefficients of ~n+1, we find
∆(ϕ(D;αi,αj))− ϕ(D;αi,αj) ⊗ 1− 1⊗ ϕ(D;αi,αj) = 0
so that ϕ(D;αi,αj) is a primitive element of End(FD). Since ϕ(D;αi,αj) is also of
weight 0, we find that
ϕ(D;αi,αj) ∈ hD
Moreover, since Φa(D;αi,αj)
satisfy the generalized pentagon identities corre-
sponding to the 2-faces of the De Concini–Procesi associahedron, we also find
dDϕ(D;αi,αj) = 0
We are then reduced to prove the following
4. THE MAIN THEOREM 165
Proposition. Let ϕ = ϕ(D;αi,αj) be a 2–cocycle in the Dynkin complex
of End(F ) such that
ϕ(D;αi,αj) ∈ hD
for any αi 6= αj ∈ D ⊆ Dg. Then, there exists a Dynkin 1–cochain a =
a(D;αi) such that
a(D;αi) ∈ hD and dDa = ϕ
The element a may be chosen such that a(αi;αi) = 0 for all αi and is then
unique with this additional property.
The element a = 1−~na(D;αi)αi∈D⊆Dg defines the twist matching the two
qCqtqba structures. Notice that, since a(αi;αi) = 0, then (Si,c)a = Si,c for all i.
The theorem is proved.
4.4. Proof of Prop. 4.3: simple case. Let us first give a proof of
Prop.4.3 for the simple case.1
Let then g be a simple complex Lie algebra of rank l, Dg be the Dynkin di-
agram associated to g with vertices α1, . . . , αl. For any i ∈ [1, l], let slαi2 ⊂ g
be the three dimensional subalgebra spanned by ei, fi, hi. Then the collec-
tion Uslαi2 defines a Dg–algebra structure on Ug.
Let hD ⊂ h be the subalgebra generated by α∨i | αi ∈ D, where D ⊂ Dg,
and λ∨i,D be the fundamental coweights associated to hD, for any D ⊂ Dg.
1the original proof in [TL4] is incomplete, since it is implicitly assumed that λ∨i,D does
not depends upon D or, in other words, that λ∨i,D′ ∈ Cλ∨i,D for D′ ⊂ D.
166 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
We wish to solve the equation ϕ = dDa. In components, this reads
ϕ(D;αi,αj) = a(D;αj) − a(D\αiαj
;αj)− a(D;αi) + a
(D\αjαi
;αi)(4.4)
for any connected subdiagram D ⊆ Dg and αi 6= αj, αi, αj ∈ D. The assump-
tions ϕ(D;αi,αj), a(D;αi) ∈ hD and the fact that ϕ, a lie in the Dynkin complex
of g imply that
ϕ(D;αi,αj) ∈ Cλ∨i,D ⊕ Cλ∨j,D and a(D;αk) ∈ Cλ∨k,D for k = i, j
respectively, where λ∨k,D is the fundamental coweight in hD dual to αk|hD .
Projecting (4.4) on λ∨i,D and λ∨j,D we therefore find that it is equivalent to
a(D;αi) = ai(D\αjαi
;αi)− ai
(D\αiαj
;αj)− ϕi(D;αi,αj)
a(D;αj) = aj(D\αiαj
;αj)− aj
(D\αjαi
;αi)+ ϕj(D;αi,αj)
where
• by (−)i we denote the component along λ∨i,D of the projection
pi,D : hD =⊕αj∈D
Cλ∨j,D −→ Cλ∨i,D
• for any subdiagram D′ ⊂ D, we look at
a(D′;αi) ∈ hD′ ⊂ hD
as an element of the bigger subspace, since in general
λ∨i,D′ 6∈ Cλ∨i,D λ∨i,D′ ∈⊕αk∈D
λ∨k,D
4. THE MAIN THEOREM 167
• we are identifying a(D;αi), a(D,αj) with their components along λ∨i,D, λ∨j,D
respectively.
Thus, since ϕ(D;αi,αj) = −ϕ(D;αj ,αi), ϕ = dDa iff
a(D;αi) = ai(D\αjαi
;αi)− ai
(D\αiαj
;αj)− ϕi(D;αi,αj)
(4.5)
for all D and 1 ≤ i 6= j ≤ n with αi, αj ∈ D. Induction on the cardinality of
D readily shows that the above equations possess at most one solution once
the values of a(αi;αi) are fixed. To prove that one solution exists, assume that
a(D;αi) have been constructed for all D with at most m vertices in such a way
that the equations (4.5) hold for all such D. We claim that (4.5) may be
used to define a(D;αi) for all D with |D| = m + 1 in a consistent way, i.e.,
independently of j 6= i such that αj ∈ D. This amounts to showing that for
all such D, and distinct vertices αi, αj, αk ∈ D, one has
ai(D\αjαi
;αi)− ai
(D\αiαj
;αj)− ϕi(D;αi,αj)
= ai(D\αkαi
;αi)− ai
(D\αiαk
;αk)− ϕi(D;αi,αk)
(4.6)
To see this, consider the (D;αi, αj, αk) component of dDϕ i.e., the sum(ϕ(D;αj ,αk) − ϕ(
D\αiαj,αk
;αj ,αk)
)−(ϕ(D;αi,αk) − ϕ
(D\αjαi,αk
;αi,αk)
)+
(ϕ(D;αi,αj) − ϕ(
D\αkαi,αj
;αi,αj)
)
168 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
Since dDϕ = 0 we get, by projecting on λ∨i,D and using ϕi(D;αj ,αk) = 0,
ϕi(D;αi,αj)= ϕi(D;αi,αk) − ϕi
(D\αjαi,αk
;αi,αk)+ ϕi
(D\αiαj,αk
;αj ,αk)+ ϕi
(D\αkαi,αj
;αi,αj)
so that (4.6) holds iff
ai(D\αjαi
;αi)− ai
(D\αiαj
;αj)− ai
(D\αkαi
;αi)+ ai
(D\αiαk
;αk)=
= −ϕi(D\αjαi,αk
;αi,αk)+ ϕi
(D\αiαj,αk
;αj ,αk)+ ϕi
(D\αkαi,αj
;αi,αj)
(4.7)
ai(D\αiαk
;αk)− ai
(D\αiαj
;αj)− ϕi
(D\αiαj,αk
;αj ,αk)=
= ai(D\αkαi
;αi)+ ϕi
(D\αkαi,αj
;αi,αj)− ai
(D\αjαi
;αi)− ϕi
(D\αjαi,αk
;αi,αk)
(4.8)
The inductive assumption yields the following relation for any distinct αi, αj, αk ∈
D
ϕ(D\αjαi,αk
;αi,αk)= a
(D\αjαi,αk
;αk)− aD\αjαi,αk
\αiαk
;αk
− a
(D\αjαi,αk
;αi)+ aD\αjαi,αk
\αkαi
;αi
(4.9)
We consider five separate cases.
4.4.1. At least two connected component among D\αjαi,αk , D\αkαi,αj ,D\αiαj ,αk
are empty. We may assume D\αjαi,αk = ∅, D\αkαi,αj = ∅. This case cannot arise
since the first condition implies that any path in D from αi to αk must pass
through αj before it reaches αk while the second one implies that the portion
of this path linking αi to αj must first pass through αk.
4. THE MAIN THEOREM 169
4.4.2. D\αjαi,αk = ∅, D\αkαi,αj 6= ∅, D\αiαj ,αk 6= ∅. In this case,
ϕ(D\αjαi,αk
;αi,αk)= 0
and (4.7) reads, using (4.9),
ai(D\αiαk
;αk)− ai
(D\αiαj
;αj)− ai
(D\αiαj,αk
;αk)
+ aiD\αiαj,αk\αj
αk;αk
+ ai(D\αiαj,αk
;αj)− aiD\αiαj,αk
\αkαj
;αj
=
= ai(D\αkαi
;αi)+ ai
(D\αkαi,αj
;αj)− aiD\αkαi,αj
\αiαj
;αj
− ai
(D\αkαi,αj
;αi)+ aiD\αkαi,αj
\αjαi
;αi
− ai
(D\αjαi
;αi)
This equation holds since:
• D\αkαi = D\αkαi,αj , so a(D\αkαi
;αi)= a
(D\αkαi,αj
;αi)
• D\αiαk = D\αiαj ,αk , so a(D\αiαk
;αk)= a
(D\αiαj,αk
;αk)
• D\αiαj = D\αiαj ,αk , so a(D\αiαj
;αj)= a
(D\αiαj,αk
;αj)
• Since D\αjαi,αk = ∅, then
D\αjαi= D\αj ,αkαi
= D\αkαi,αj
\αjαi =⇒ a
(D\αjαi
;αi)= aD\αkαi,αj
\αjαi
;αi
• Similarly,
D\αkαi,αj
\αiαj = D\αi,αkαj
= D\αiαj,αk
\αkαj =⇒ aD\αkαi,αj
\αiαj
;αj
= aD\αiαj,αk\αk
αj
170 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
Condition (4.7) reduces to:
ai(D\αkαi,αj
;αj)= aiD\αiαj,αk
\αjαk
;αk
that is
ai(D\αkαj
;αj)= ai
(D\αjαk
;αk)
We claim that for all distinct i, j, k ∈ I,
ai(D\αkαj
;αj)= 0
that is equivalent to:
λ∨j,D\αkαj
∈ Cλ∨j,D ⊕ Cλ∨k,D (4.10)
Indeed, if i ∈ Ikj := r ∈ I | αr ∈ D\αkαj , i 6= j, then
〈αi, λ∨j,D\αkαj
〉 = 0
by definition of λ∨j,D\αkαj
. On the other hand, if i 6∈ Ikj , then, by orthogonality,
we have:
∀s ∈ Ikj 〈αi, α∨s 〉 = 0
Consequently, since λ∨j,D\αkαj
∈ hD\αkαj
, we have
〈αi, λ∨j,D\αkαj
〉 = 0
and (4.10) is proved.
4.4.3. D\αjαi,αk 6= ∅, D\αkαi,αj = ∅, D\αiαj ,αk 6= ∅. This case reduces to the
previous one under the interchange αj ↔ αk.
4. THE MAIN THEOREM 171
4.4.4. D\αjαi,αk 6= ∅, D\αkαi,αj 6= ∅, D\αiαj ,αk = ∅. In this case,
ϕ(D\αkαi,αj
;αi,αk)= 0
and (4.7) reads, using (4.9),
ai(D\αiαk
;αk)− ai
(D\αiαj
;αj)=
= ai(D\αkαi
;αi)+ ai
(D\αkαi,αj
;αj)− aiD\αkαi,αj
\αiαj
;αj
− ai
(D\αkαi,αj
;αi)
+aiD\αkαi,αj\αj
αi;αi
− ai
(D\αjαi
;αi)− ai
(D\αjαi,αk
;αk)− aiD\αjαi,αk
\αiαk
;αk
+ai
(D\αjαi,αk
;αi)− aiD\αjαi,αk
\αkαi
;αi
This equation holds, since:
• D\αjαi = D\αjαi,αk , so a(D\αjαi
;αi)= a
(D\αjαi,αk
;αi)
• D\αkαi = D\αkαi,αj , so a(D\αkαi
;αi)= a
(D\αkαi,αj
;αi)
• Since D\αiαj ,αk = ∅, then
D\αiαk= D\αi,αjαk
= D\αjαi,αk
\αiαk =⇒ a
(D\αiαk
;αk)= aD\αjαi,αk
\αiαk
;αk
D\αiαj= D\αi,αkαj
= D\αkαi,αj
\αiαj =⇒ a
(D\αiαj
;αj)= aD\αkαi,αj
\αiαj
;αj
• Similarly,
D\αjαi,αk
\αkαi = D\αj ,αkαi
= D\αkαi,αj
\αjαi
172 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
implies
aD\αjαi,αk\αk
αi;αi
= aD\αkαi,αj\αj
αi;αi
As before, it remains to check that:
ai(D\αkαi,αj
;αj)= ai
(D\αkαj
;αj)= 0
ai(D\αjαi,αk
;αk)= ai
(D\αjαk
;αk)= 0
that it is true by (4.10).
4.4.5. D\αjαi,αk 6= ∅, D\αkαi,αj 6= ∅, D\αiαj ,αk 6= ∅. It is interesting to observe
that this case can occur only if there is a cycle in D passing through αi, αj, αk
that is not our case, since g is simple. Instead, in the affine case g, it can
happen only if g = An and D = Dg.
Even in this case, anyway, the consistency relation holds. Indeed, (4.7) reads,
4. THE MAIN THEOREM 173
using (4.9),
ai(D\αiαk
;αk)− ai
(D\αiαj
;αj)− ai
(D\αiαj,αk
;αk)+ aiD\αiαj,αk
\αjαk
;αk
+ai
(D\αiαj,αk
;αj)− aiD\αiαj,αk
\αkαj
;αj
=
= ai(D\αkαi
;αi)+ ai
(D\αkαi,αj
;αj)− aiD\αkαi,αj
\αiαj
;αj
− ai
(D\αkαi,αj
;αi)
+aiD\αkαi,αj\αj
αi;αi
− ai
(D\αjαi
;αi)− ai
(D\αjαi,αk
;αk)+ aiD\αjαi,αk
\αiαk
;αk
+ai
(D\αjαi,αk
;αi)− aiD\αjαi,αk
\αkαi
;αi
This equation holds, since:
• D\αjαi = D\αjαi,αk , so a(D\αjαi
;αi)= a
(D\αjαi,αk
;αi)
• D\αiαj = D\αiαj ,αk , so a(D\αiαj
;αj)= a
(D\αiαj,αk
;αj)
• D\αkαi = D\αkαi,αj , so a(D\αkαi
;αi)= a
(D\αkαi,αj
;αi)
• D\αiαk = D\αiαj ,αk , so a(D\αiαk
;αk)= a
(D\αiαj,αk
;αk)
• Similarly,
D\αjαi,αk
\αkαi = D\αj ,αkαi
= D\αkαi,αj
\αjαi
implies
aD\αjαi,αk\αk
αi;αi
= aD\αkαi,αj\αj
αi;αi
174 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
then
D\αiαj,αk
\αkαj = D\αi,αkαj
= D\αkαi,αj
\αiαj
implies
aD\αiαj,αk\αk
αj;αj
= aD\αkαi,αj\αi
αj;αj
and finally
D\αjαi,αk
\αiαk = D\αi,αjαk
= D\αiαj,αk
\αjαk
implies
aD\αjαi,αk\αi
αk;αk
= aD\αiαj,αk\αj
αk;αk
As before it remains to check that:
ai(D\αkαi,αj
;αj)= ai
(D\αkαj
;αj)= 0
ai(D\αjαi,αk
;αk)= ai
(D\αjαk
;αk)= 0
that is true, by (4.10).
This concludes the proof of Proposition 4.3
4.5. Proof of Prop. 4.3: Affine case. We will now discuss the previ-
ous proof using the combinatorics of the Dynkin diagrams of affine type.
Let g be a simple complex Lie algebra of rank l and g be the affine Lie
algebra of type g. We denote by Dg the Dynkin diagram associated to g with
4. THE MAIN THEOREM 175
vertices α0, α1, . . . , αl and by h = h ⊕ Cc ⊕ Cd ⊂ g the Cartan subalgebra
generated by α∨0 , α∨1 , . . . , α∨l , α∨l+1 where
α∨0 =c
(θ|θ)− θ∨ ⊗ 1 α∨l+1 = d
Let B = α0, α1, . . . , αl, αl+1 be a basis for h∗, where αl+1 = c∗ is defined by
∀i ∈ 0, 1, . . . , l, l + 1 〈αl+1, α∨i 〉 = δ0i
We denote by λ∨0 , λ∨1 , . . . , λ∨l , λ∨l+1 the fundamental coweights of h. for any
subdiagram D ⊆ Dg is given a subalgebra
hD =
〈α∨i | αi ∈ D〉 if D 6= Dg
h if D = Dg
As in the previous section, we aim at proving the following
Proposition. Let ϕ = ϕ(D;αi,αj) be a 2–cocycle in the Dynkin complex
of U g such that
ϕ(D;αi,αj) ∈ hD
for any αi 6= αj ∈ D ⊆ Dg. Then, there exists a Dynkin 1–cochain a =
a(D;αi) such that
a(D;αi) ∈ hD and dDa = ϕ
The element a may be chosen such that a(αi;αi) = 0 for all αi and is then
unique (up to a coboundary) with this additional property.
176 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
Proof. We will follow the same outline of Prop. 4.3. Indeed, as before,
the equation ϕ = dDa, in components, reads
ϕ(D;αi,αj) = a(D;αj) − a(D\αiαj
;αj)− a(D;αi) + a
(D\αjαi
;αi)(4.11)
for any connected subdiagram D ⊆ Dg and i < j such that αi, αj ∈ D. If D is
a proper subdiagram, then gD is of finite type and the discussion is the same
that before.
If D = Dg is the Dynkin diagram of g, then ϕ(D;αi,αj) ∈ Cλ∨i ⊕ Cλ∨j ⊕ Cλ∨l+1
a(D;αi) ∈ Cλ∨i ⊕ Cλ∨l+1
since λ∨l+1 = c. Componentwise (4.11) reads
ϕi(D;αi,αj)= ai
(D\αjαi
;αi)− ai
(D\αiαj
;αj)− ai(D;αi)
(4.12)
ϕj(D;αi,αj)= aj
(D\αjαi
;αi)− ai
(D\αiαj
;αj)+ aj(D;αj)
(4.13)
ϕl+1(D;αi,αj)
= al+1(D;αj)
− al+1
(D\αiαj
;αj)− al+1
(D;αi)+ al+1
(D\αjαi
;αi)(4.14)
It is useful to notice that, if D = Dg, (4.10) becomes
λ∨j,D\αkαj
∈ Cλ∨j,D ⊕ Cλ∨k,D ⊕ Cλ∨l+1,D (4.15)
then the consistency of (4.11) along λ∨i,D, λ∨j,D is granted by the proof of (4.3).
Moreover, it is easy to see that, if α0 6∈ D\αkαj , then
λ∨j,D\αkαj
∈ Cλ∨j,D ⊕ Cλ∨k,D (4.16)
4. THE MAIN THEOREM 177
or, in other words,
al+1
(D\α0αj
;αj)= 0 (4.17)
Indeed, if
λ∨j,D\αkαj
=l+1∑i=0
yij,kα∨i =
l+1∑i=0
xij,kλ∨i,D
we have xl+1j,k = y0
j,k.
It remains to prove the consistency of (4.11) along λ∨l+1,D. Up to a coboundary,
we may assume al+1(D;α0) = 0. It is enough to observe that the element a′ =
a′(D′;αi) defined by
a′(D′;αi) =
0 if D′ 6= Dg
δ0ic if D′ = Dg
is indeed a coboundary. Assume then al+1(D;α0) = 0. By (4.17), for each j 6= 0,
ϕl+1(D;α0,αj)
= al+1(D;αj)
+ al+1
(D\αjα0
;α0)(4.18)
Using this relation in (4.14), we get
ϕl+1(D;αi,αj)
= ϕl+1(D;α0,αj)
− al+1
(D\αjα0
;α0)− al+1
(D\αiαj
;αj)
−ϕl+1(D;α0,αi)
+ al+1
(D\αiα0
;α0)+ al+1
(D\αjαi
;αi)
(4.19)
To see this, consider the (D;α0, αi, αj) component of dDϕ i.e., the sum(ϕ(D;αi,αj) − ϕ(
D\α0αi,αj
;αi,αj)
)−(ϕ(D;α0,αj) − ϕ(
D\αkα0,αj
;α0,αj)
)+
(ϕ(D;α0,αi) − ϕ(
D\αjα0,α1
;α0,α1)
)
178 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
Since dDϕ = 0 and ϕl+1
(D\α0αi,αj
;αi,αj)= 0 by (4.16), we get, by projecting on
λ∨l+1,D,
ϕl+1
(D\αjα0,αi
;α0,αi)− ϕl+1
(D\αiα0,αj
;α0,αj)= al+1
(D\αiα0
;α0)+ al+1
(D\αjαi
;αi)
−al+1
(D\αjα0
;α0)− al+1
(D\αiαj
;αj)
(4.20)
Recall that a = a(D′;αi) is determined by induction on D and the relation
al+1(D;αi)
= ϕl+1(D;α0,αj)
− al+1
(D\αjα0
;α0)(4.21)
gives the recursion formula for the component along λ∨l+1. In particular, in-
duction on D gives us the following relation:
ϕ(D\αjα0,αi
;α0,αi)= a
(D\αjα0,αi
;αi)− aD\αjα0,αi
\α0αi
;αi
−a
(D\αjα0,αi
;α0)+ aD\αjα0,αi
\αiα0
;αi
(4.22)
Projecting along λ∨l+1,D, by (4.16),
ϕl+1
(D\αjα0,αi
;α0,αi)= al+1
(D\αjα0,αi
;αi)− al+1
(D\αjα0,αi
;α0)+ al+1D\αjα0,αi
\αiα0
;αi
(4.23)
We consider four separate cases.
4.5.1. D\αjα0,αi = ∅ and D\αiα0,αj = ∅. This case cannot arise since the first
condition implies that any path in D from α0 to αi must pass through αj
before it reaches αi while the second one implies that the portion of this path
linking α0 to αj must first pass through αi.
4. THE MAIN THEOREM 179
4.5.2. D\αjα0,αi 6= ∅ and D\αiα0,αj = ∅. In this case,
ϕ(D\αiα0,αj
;α0,αj)= 0
and (4.20) reads
al+1
(D\αjα0,αi
;αi)− al+1
(D\αjα0,αi
;α0)+ al+1D\αjα0,αi
\αiα0
;α0
=
= al+1
(D\αiα0
;α0)+ al+1
(D\αjαi
;αi)− al+1
(D\αjα0
;α0)− al+1
(D\αiαj
;αj)
(4.24)
This equation holds since
• D\αjαi = D\αjα0,αi , so al+1
(D\αjαi
;αi)= al+1
(D\αjα0,αi
;αi)
• D\αjα0 = D\αjα0,αi , so al+1
(D\αjα0
;α0)= al+1
(D\αjα0,αi
;α0)
• Since D\αiα0,αj = ∅,
D\αiα0= D\αi,αjα0
= D\αjα0,αi
\αiα0 =⇒ al+1
(D\αiα0
;α0)= al+1
(D\αjα0,αi
\αiα0
;α0)
So, we need to verify
al+1
(D\αiαj
;αj)= 0
Since D\αiα0,αj = ∅, then α0 6∈ D\αiαj and, by (4.17), we can conclude.
4.5.3. D\αjα0,αi = ∅ and D\αiα0,αj 6= ∅. This case reduces to the previous one
under the interchange αj ↔ αk.
180 4. A RIGIDITY THEOREM FOR QUASI–COXETER CATEGORIES
4.5.4. D\αjα0,αi = ∅ and D\αiα0,αj 6= ∅. This case is possible when α0 is a
vertex in D of degree at least 2 i.e., only for g = sln. Equation (4.20) reads:
al+1
(D\αiα0
;α0)+ al+1
(D\αjαi
;αi)− al+1
(D\αjα0
;α0)− al+1
(D\αiαj
;αj)=
= al+1
(D\αjα0,αi
;αi)− al+1
(D\αjα0,αi
;α0)+ al+1D\αjα0,αi
\αiα0
;α0
− al+1
(D\αiα0,αj
;αj)+ al+1
(D\αiα0,αj
;α0)− al+1D\αiα0,αj
\αjα0
;α0
(4.25)
This equation holds since
• D\αjαi = D\αjα0,αi , so al+1
(D\αjαi
;αi)= al+1
(D\αjα0,αi
;αi)
• D\αjα0 = D\αjα0,αi , so al+1
(D\αjα0
;α0)= al+1
(D\αjα0,αi
;α0)
• D\αiαj = D\αiα0,αj , so al+1
(D\αiαj
;αj)= al+1
(D\αiα0,αj
;αj)
• D\αiα0 = D\αiα0,αj , so al+1
(D\αiα0
;α0)= al+1
(D\αiα0,αj
;α0)
• we have
D\αjα0,αi
\αiα0 = D\αi,αjα0
= D\αiα0,αj
\αjα0 =⇒ aD\αjα0,αi
\αiα0
;α0
= aD\αiα0,αj\αj
α0;α0
This concludes the proof of Proposition 4.5 and the proof of the main
theorem for the affine case.
APPENDIX A
In this Appendix, we provide a detailed proof of two results used in Chapter
3. Namely, in Section 1, we prove that the algebra Ug, described by Drinfeld
in [D5, 8.1] and by Etingof–Kazhdan in [EK6], is isomorphic to the algebra of
endomorphisms of the forgetful functor with respect to the category O of g. In
Section 2, we prove that the isomorphism ΨEK between Ug[[~]] and its quantum
counterpart given in [EK6] corresponds, under the previous identification,
with the natural isomorphism induced by the equivalence of categoriesO[[~]] '
O~.
Finally, in Section 3, we prove, in analogy with the results of [EK1, Part
II], that the relative fiber functor Γ : DΦ(g) → DΦD(gD) gives rise naturally
to a braided Hopf algebra object in DΦD(gD). This result is used in Chapter
3 Section 5 to study the behavior of the functor and its tensor structure with
respect to abelian extensions of gD.
1. The completion of Ug in category O
Let Q be the root lattice of g. Consider the modules
Uβ := Ug/Iβ
181
182 A
where Iβ ⊂ Ug is the left ideal generated by all elements of weight ≤ β. Define
Ug := limβ∈Q
Uβ
It is easy to prove that there is an injection
limβ
colimγ HomUg(Uβ, Uγ) −→ HomUg(Ug, Ug)
and we want to prove that, in this case, this is an isomorphism obtained by
the identification
colimγ HomUg(Uγ, Uβ) ' HomUg(Ug, Uβ)
1.1. The Ug-module structure of Uβ. The module Uβ is associated to
the short exact sequence
0 // Iβiβ// Ug
pβ// Uβ // 0
In particular, Uβ is a cyclic module generated by an element 1β such that
Iβ.1β = 0.
Lemma. For each v ∈ Uβ, there exists a weight β′ = β′(v) ∈ Q, β′ ≤ β,
such that Iβ′ .v = 0
Proof. We can assume v = a1...an.1β for some ai ∈ Ug. Proceed by
induction. If n = 0, then v = 1β and β′ = β. Now assume v = a.w, where
w = a1...an.1β and there exists β′ = β′(w) such that Iβ′ .w = 0.
1. THE COMPLETION OF Ug IN CATEGORY O 183
(i) if a ∈ (Ug)µ, µ ≥ 0, then for each element x of weight λ ≤ β′ (i.e.
for each generator of Iβ′) we have
x.v = [x, a].w + a.x.w = [x, a].w
and [x, a] ∈ (Ug)λ+µ. Now, if we choose λ ≤ β′ − µ, we have [x, a] ∈
Iβ′ , so [x, a].w = 0. We conclude that, for β′′ := β′ − µ ≤ β′,
Iβ′′ .v = 0.
(ii) if a ∈ (Ug)µ, µ ≤ 0, for each generator x of Iβ′ of weight λ, the
commutator [x, a] has weight λ + µ ≤ β′, since λ ≤ β′ and µ ≤ 0.
Then Iβ′ .v = 0.
For general a ∈ Ug, use the weight decomposition
a ∈ Ug =⊕β∈Q
(Ug)β ⇒ v = a.w =m∑i=1
aµi .w =m∑i=1
vi
where aµi ∈ (Ug)µi . Then there exist β1...βm such that Iβi .vi = 0. Choosing
β′′ := minβi, i = 1..m (i.e. β′′ =∑
α∈∆ min(βi, α)α), we have
Iβ′′ .v =( m⋂i=1
Iβi).v = 0
As a consequence we can say that the action of Ug on Uβ, for every element
v, factors through Uβ′(v), β′(v) ≤ β, i.e. for every element u ∈ Ug u.v = uβ′ .v,
in the sense that the action of u on v is the same that the action of any
representative of the class of u in Uβ′ , i.e. uβ′ = pβ′(u).
1.2. The Ug-module structure of Ug. Define an action a : Ug⊗ Ug→
Ug by u.xβ = u.xβ. Since the maps πβ′β are maps of Ug-modules, the
184 A
map Ug⊗ Ug→ Ug is well-defined, because for every β′ ≤ β,
u.xβ = u.πβ′β(xβ′) = πβ′β(u.xβ′)
and it’s obviously an action of Ug-modules.
1.3. The multiplication map in Ug. Using the action
Ug⊗ Uβ → Uβ
for every β ∈ Q, we can define a bilinear map:
U ⊗ Uβ // Uβ
x⊗ v // xω(v).v
This map lifts up to a multiplication in Ug
Ug⊗ Ugm
// Ug
x⊗ y // xω(yβ).yβ
where ω(yβ) is a weight such that Iω(yβ).yβ = 0. In order to show that this
map is well-defined, we need to prove that, for every β′ ≤ β,
xω(yβ).yβ = πβ′β(xω(yβ′ ).yβ′)
Indeed, we have yβ′ ∈ Uβ′ and Iω(yβ′ ).yβ′ = 0. Then
0 = πβ′β(Iω(yβ′ ).yβ′) = Iω(yβ′ )
.πβ′β(yβ′) = Iω(yβ′ ).yβ
1. THE COMPLETION OF Ug IN CATEGORY O 185
So, ω(yβ′) ≤ ω(yβ). In particular,
xω(yβ) = πω(yβ′ )ω(πβ)(xω(yβ))
Consequently, their action on yβ is the same, i.e.
xω(yβ).yβ = xω(yβ′ ).yβ = xω(yβ′ )
.πβ′β(yβ′) = πβ′β(xω(yβ′ ).yβ′)
So, we have an algebra structure on Ug.
1.4. Compatibility with the Ug-action. We have a map i : Ug→ Ug,
sending each element u in the collection of its image in Uβ for all β ∈ Q,
i(u) := uββ∈Q. Since
Ker(i) =⋂β∈Q
Iβ = 0
i is an injection. It’s very easy to show that the multiplication map is com-
patible with the Ug-action in the following way:
(a) By definition, the action of Ug on Ug is given by the formula
a(u,x) = u.x = uβ.x = m(i(u),x)
so that the following diagram commutes:
Ug⊗ Uga
//
i⊗1 %%
Ug
Ug⊗ Ug
m
OO
186 A
(b) Since the injection map is a map of Ug-modules, we have a commu-
tative diagram
Ug⊗ Ugm
//
1⊗i
Ug
i
Ug⊗ Uga
// Ug
Then
Ug⊗ Ug
m
Ug⊗ Ug
a
i⊗1oo Ug⊗ Ug
1⊗ioo
m
Ug Ug Ugi
oo
is commutative and Ug ⊂ Ug is a subalgebra.
1.5. Topology on Ug. We consider the modules Uβ with the discrete
topology and we equip Ug = limUβ with the topology induced by limit. In
particular, a basis for the open sets is given by the sets U(β1... βn, u1... un) for
n ≥ 0, βi ∈ Q, ui ∈ Uβi , where
U(β1..βn, u1...un) := x ∈ Ug | xβi = ui ∀i
1. THE COMPLETION OF Ug IN CATEGORY O 187
Let β := minβi, i = 1... n. Then, by surjectivity of pβ, there exists an
element u ∈ Ug such that uβi = ui.
Ug
pβ
Uβπββ1
~~
πββn
!!
Uβ1 · · · Uβn
Then we have proved that Ug is dense in Ug.
For any element u ∈ Ug, we can find ui ∈ Ug such that
u = limi→∞
ui
In particular, if φ ∈ EndUg(Ug), then, for any u, x ∈ Ug, we have
φ(u.x) = φ((limiui).x) = lim
iφ(ui.x) = (lim
iui).φ(x) = u.φ(x)
and
EndUg(Ug) ' EndUg(Ug) ' Ug
1.6. HomUg(Ug, Uβ). The map
Ug⊗ Uβ // Uβ
is in fact an action of the algebra Ug on the module Uβ. Using the same
density argument as before, we have
HomUg(Ug, Uβ) ' HomUg(Ug, Uβ)
188 A
Consequently, for any map φ ∈ HomUg(Ug, Uβ), we have
φ(u) = u.φ(1)
for any u ∈ Ug. In particular, φ is completely determined by φ(1) ∈ Uβ. By
the previous discussion, we know that there exists a weight ω = ω(φ(1)) such
that Iω.φ(1) = 0. Then the action of u on φ(1) factors through Uω, i.e.
φ(u) = u.φ(1) = uω.φ(1) =: φ(uω)
and
Ugφ//
πω
Uβ
Uω
φ
>>
where φ ∈ HomUg(Uω, Uβ). It’s clear that we can associate to φ a uniquely
determined class [φ] ∈ colimγ HomUg(Uγ, Uβ) and consequently
colimγ HomUg(Uγ, Uβ) ' HomUg(Ug, Uβ)
and
limβ
colimγ HomUg(Uγ, Uβ) ' limβ
HomUg(Ug, Uβ) ' HomUg(Ug, Ug)
Notice that the same argument can be used to show that, for any module
V ∈ O, we have
HomUg(Ug, V ) ' HomUg(Ug, V )
2. THE ISOMORPHISM BETWEEN Ug[[h]] AND U~g 189
and consequently for any map f ∈ HomUg(Ug, V ) there exists β ∈ Q and
fβ ∈ HomUg(Uβ, V ) such that
Ugf
//
πβ
V
Uβ
fβ
??
is commutative.
2. The isomorphism between Ug[[h]] and U~g
The equivalence of categories between O[[h]] ' Oh, established by the
Etingof–Kazhdan functor, carries an isomorphism
HomUg(Uγ, Uβ)[[h]] ' HomUhg(Uhγ , U
hβ )
and then
limβ colimγ HomUg(Uγ, Uβ)[[h]]'//
o
limβ colimγ HomUhg(Uhγ , U
hβ )
o
EndUg(Ug)[[h]]
o
EndUhg(Uhg)
o
Ug[[h]]'
// Uhg
On the other hand, we can consider the forgetful functors
F : O[[h]]→ A Fh : Oh → A
190 A
and the associated completions End(F ), End(Fh). Using the equivalence be-
tween |Co[[h]] and Oh we have a commutative diagram
O[[h]]Φ//
F
""
OhFh
A
that induces a natural isomorphism
End(F ) ' End(Fh)
We want to show that this construction is equivalent to the previous one,
in particular we want to show that there exists an algebra isomorphism
Ug[[h]]'−→ End(F )
2.1. The map φ. First of all, let us construct a map φ : Ug[[h]] →
End(F).
Lemma. For any v ∈ V there exists β ∈ Q such that Iβ.v = 0
Proof. Assume first v ∈ Vλ. Set µ := minλi|λ ∈ D(λi). For any
x ∈ (Ug)λ′ where λ′ ≤ µ(v) := µ− λ, we have x.v = 0. Then Iµ(v).v = 0.
In general, v =∑vλ, where vλ ∈ Vλ. Set µ(v) := minλ∈P (V )µ(vλ). Then
Iµ(v).v = 0
For each V ∈ O we have a Ug-action given by, for any u ∈ Ug,
u.v = uµ(v).v
2. THE ISOMORPHISM BETWEEN Ug[[h]] AND U~g 191
where by uµ(v) we mean any representative of the class uµ(v) ∈ Uµ(v) (to be
precise, any representative of the classes uβ ∈ Uβ with β ≤ µ(v)).
In particular, for any V ∈ O we have a map
φV : Ug→ End(V )
Define φ(u) := φV (u).
2.2. φ is well-defined. We need to show that for any V,W ∈ O and
f ∈ HomUg(V,W ) we have
Vf//
φV (u)
W
φW (u)
Vf// W
i.e. for any v ∈ V
f(uµ(v).v) = uµ(f(v)).f(v)
Since f ∈ HomUg(V,W ), then
0 = f(Iµ(v).v) = Iµ(v).f(v)
so µ(v) ≤ µ(f(v)) and the action of uµ(v) is equivalent to that one of uµ(f(v)).
So,
uµ(f(v)).f(v) = uµ(v).f(v) = f(uµ(v).v)
Then we have a well-defined map
φ : Ug[[h]]→ End(F )
192 A
2.3. φ is injective. Recall that for any β ∈ Q Uβ ∈ O. In particular, for
any v ∈ Uβ
φβ(u)(v) = uβ.v
Then, if φ(u) = 0, we have uβ = 0 for all β ∈ Q, and u = 0.
2.4. φ is surjective. For any f ∈ End(F ), β′ ≤ β, we have a commuta-
tive diagram
Uβ′πβ′β
//
fβ′
Uβ
fβ
Uβ′πβ′β
// Uβ
By universal property of limits in the category Vec(k), there exists a map
f ∈ Homk(Ug, Ug) such that the diagram
Ugπβ
//
πβ′
f
Uβ
fβ
Uβ′
fβ′
πβ′β
??
Ugπβ
//
πβ′
Uβ
Uβ′
πβ′β
??
is commutative.
2.5. Commutativity in O. Recall that HomUg(Ug, V ) ' HomUg(Ug, V ).
Then for any g ∈ HomUg(Ug, V ), there exists µ := µ(g, V ) ∈ Q and gµ ∈
2. THE ISOMORPHISM BETWEEN Ug[[h]] AND U~g 193
HomUg(Uµ, V ) such that
Ugg
//
πµ
V
Uµ
gµ
??
Consequently, the following diagram
Ugg
//
πµ
f
V
fV
Uµ
fµ
gµ
??
Ugg
//
πµ
V
Uµ
gµ
??
is commutative and then, for any V ∈ O and g ∈ HomUg(Ug, V ), the diagram
Ugg//
f
V
fV
Ugg// V
is commutative. This fact, in particular, is true for Uβ, β ∈ Q.
2.6. Commutativity for EndUg(Ug). Suppose now g ∈ HomUg(U , U) '
limβ HomUg(U , Uβ). It means that for any β, there exists gβ ∈ HomUg(Ug, Uβ)
194 A
such that the diagram
Ugg//
gβ
Ug
πβ
Uβ
By density, there exists β′ ≤ β and gβ′β ∈ HomUg(Uβ′ , Uβ) such that the
diagram
Ugg//
πβ′
gβ
Ug
πβ
Uβ′gβ′β
// Uβ
is commutative. Then we have a box
Ugg
//
f
πβ
Ugπβ′
f
Uβ′gβ′β
//
fβ′
Uβ
fβ
Ugg//
πβ′
Ugπβ
Uβ′gβ′β
// Uβ
where every face is commutative except one behind, i.e. we have
πβ g f = πβ f g
It means that the maps f g and g f define the same endomorphism of Ug
as vector space. Then, for any g ∈ HomUgg(Ug, Ug) we have a commutative
2. THE ISOMORPHISM BETWEEN Ug[[h]] AND U~g 195
diagram
Ugg//
f
Ug
f
Ugg// Ug
2.7. The inverse map φ−1. Notice that, for each element y ∈ Ug, the
map
my : Ug→ Ug|my(x) = x · y
is in HomUg(Ug, Ug). Indeed, for any u ∈ Ug,
my(u.x) = my(i(u) · x) = i(u) · x · y = u.my(x)
and, for any u ∈ Ug,
f(u) = f(mu(1)) = mu(f(1)) = u · f(1)
Then we have defined a map
φ−1 : End(F )→ Ug | φ−1(f) = f(1)
and we have also proved the surjectivity of φ. indeed, for any v ∈ V , take the
map g ∈ HomUg(Ug, V ) defined by g(1) = v. We have a commutative diagram
Ugg//
f
V
fV
Ugg// V
i.e.
fV (v) = fV (g(1)) = g(f(1)) = f(1).g(1) = f(1).v
196 A
It is equivalent to say that
f = φ(f(1))
We can conclude that the diagram
Ug[[h]]'
//
o
Uhg
o
End(F )'// End(Fh)
is commutative.
3. A BRAIDED HOPF ALGEBRA IN DΦD (gD) 197
3. A braided Hopf algebra in DΦD(gD)
3.1. This section is entirely dedicated to the proof of the following
Theorem.
(i) The object Γ(L−) has a structure of braided Hopf algebra in DΦD(gD).
(ii) On every object Γ(V ) there are well-defined actions and coalitions of
Γ(L−), so that the functor Γ naturally factors as
DΦ(g)
Γ
Γ
**
DΦD,UgD(Γ(L−)) ' DΦD(Γ(L−)]UgD,−)
fDtt
DΦD(gD)
where Γ(L−)]UgD,− denotes the Radford algebra attached to Γ(L−).
Theorem 3.1 is used in Chapter 3 Section 5 to study the behavior of the
relative twists with respect to abelian extensions of gD.
3.2. Let (Γ, JΓ) : DΦ(g)→ DΦD(gD) be the monoidal functor introduced
in Chapter 3. Following [EK1], for every V ∈ DΦ(g) we can define a map
µV : Γ(L−)⊗ Γ(V )→ Γ(V ) µV (x⊗ v) = (i∨+ ⊗ id)Φ−1(id⊗v)x
Proposition. For every V ∈ DΦ(g), the map µV is a morphism of gD–
modules. Moreover, it satisfies
µV (id⊗µV )ΦD = µV (µL− ⊗ id)
198 A
Therefore, (Γ(L−), µ) is an associative algebra object in DΦD(gD) and acts on
the image of the functor Γ.
Proof. Let ∆(a) = a1 ⊗ a2 for a ∈ UgD. Then
µV (a1.x⊗ a2.v) = (i∨+ ⊗ id)Φ−1(id⊗S(a2)ρ ⊗ id)(id⊗v)(S(a1)ρ ⊗ id)x
= (i∨+ ⊗ id)Φ−1(S(a1)ρ ⊗ S(a2)ρ ⊗ id)(id⊗v)x
= (i∨+ ⊗ id)(S(a1)ρ ⊗ S(a2)ρ ⊗ id)Φ−1(id⊗v)x
= (S(a)ρ ⊗ id)(i∨+ ⊗ id)Φ−1(id⊗v)x
= a.µV (x⊗ v)
The second statement is also easily proved. Let x, y ∈ Γ(L−), v ∈ Γ(V ) and
ΦD =∑Pk ⊗Qk ⊗Rk. Then
µV (id⊗µV )ΦD(x⊗ y ⊗ v) =
= (i∨+ ⊗ id)Φ−1(id⊗i∨+ ⊗ id)(id⊗Φ−1)(id⊗ id⊗Rk.v)(id⊗Qk.y)(Pk.x)
= (i∨+ ⊗ id)(id⊗i∨+ ⊗ id)Φ−11,23,4Φ−1
234ΦρD,123(id⊗ id⊗v)(id⊗y)x
= (i∨+ ⊗ id)(id⊗i∨+ ⊗ id)Φ123ΦρD,123Φ−1
12,3,4Φ−11,2,34(id⊗ id⊗v)(id⊗y)x
= (i∨+ ⊗ id)(i∨+ ⊗ id⊗ id)Φ−112,3,4Φ−1
1,2,34(id⊗ id⊗v)(id⊗y)x
= (i∨+ ⊗ id)Φ−1(id⊗ id⊗v)(i∨+ ⊗ id)Φ−1(id⊗y)x
= µV (µL− ⊗ id)(x⊗ y ⊗ v)
where the equalities are obtained by applications of the pentagon axiom and
the associativity of i∨+.
3. A BRAIDED HOPF ALGEBRA IN DΦD (gD) 199
3.3. The tensor structure on Γ and the canonical coalgebra structure on
L− define the map
δ : Γ(L−)→ Γ(L−)⊗ Γ(L−) δ = J−1Γ Γ(i−)
Proposition. The map δ is a morphism of gD–modules and it satisfies
ΦD(δ ⊗ id)δ = (id⊗δ)δ
Therefore, (Γ(L−), δ) is a coassociative algebra object in DΦD(gD).
Proof. Let a ∈ gD. The maps JΓ and Γ(i−) are both morphisms of
gD–modules, therefore we have
δ(a.x) = ∆(a).δ(x)
Moreover,
ΦD(δ ⊗ id) = ΦDJ−1Γ,1,2(Γ(i−)⊗ id)J−1
Γ Γ(i−)
= ΦDJ−1Γ,1,2J
−1Γ,12,3Γ(i− ⊗ id)Γ(i−)
= J−1Γ,2,3J
−1Γ,1,23Γ(Φ)Γ(i− ⊗ id)Γ(i−)
= J−1Γ,2,3(id⊗Γ(i−))J−1
Γ Γ(i−)
= (id⊗δ)δ
3.4. We now want to study the action of Γ(L−) on the tensor product.
200 A
Proposition. The object (Γ(L−), µ, δ) is a bialgebra object in DΦD(gD).
The action on the tensor product is compatible with the coproduct and the
tensor structure JΓ, i.e.,
µV⊗W (id⊗JΓ) = JΓ(µV ⊗ µW )βD,23(δ ⊗ id⊗ id)
as morphisms from Γ(L−)⊗ Γ(V )⊗ Γ(W ) to Γ(V ⊗W )
Proof. We ignored the action of the associator ΦD since we proved that it
is compatible with µV and JΓ. The equality above corresponds to the identity
in Γ(V ⊗W )
(i∨+ ⊗ id⊗ id)(id⊗i∨+ ⊗ id⊗ id)β34(id⊗v ⊗ w)(id⊗i−)x =
= (i∨+ ⊗ id⊗ id)β23(i∨+ ⊗ id⊗i∨+ ⊗ id)(id⊗b.v ⊗ id⊗w)(x1 ⊗ ai.x2)i−
where v ∈ Γ(V ), w ∈ Γ(W ), x ∈ Γ(L−), δ(x) = x1 ⊗ x2 and exp(−~ΩD/2) =
ai⊗ bi. It is more convenient, in this case, to follow a pictorial proof. We have
to show that the following diagrams are equivalent:
v w
x
x1 ai.x2
bi.v w=
3. A BRAIDED HOPF ALGEBRA IN DΦD (gD) 201
The relation satisfied by δ is represented by
δ(x)x
=
Since the action of gD on the objects Γ(V ) is given by right action on N∗+, we
represent the braiding β−1D as
ai.x2 = x2
•bi.v v
=
We notice that we are allowed to move the black and the white bullet along
the lines, since the commute with the left action of g. Clearly, the RHS corre-
sponds to
x1 x2
•v w
x1 x2
•v w
x1 x2
wv• = =
202 A
We now use the fact that the map i∨+ satisfies
i∨+ β (β−1D )ρ = i∨+
i.e.,
•
=
•
=
Finally we get
x1 x2
v w=
v w
x
3.5. The tensor functor (Γ, JΓ) induces a natural braiding on the sub-
category generated in DΦD(gD) b the objects Γ(V ), for any V ∈ DΦ(g). The
braiding is given by the usual formula
βJΓ:= J−1
Γ,21 Γ(β) JΓ
3. A BRAIDED HOPF ALGEBRA IN DΦD (gD) 203
where β = σ · R is the braiding in DΦ(g). Clearly, βJΓis a morphism of gD–
module and so is the relative R–matrix RD := σβJΓ.
Let now u ∈ Γ(L−) be the unit element. For every V ∈ DΦ(g) we define
the trivial comodule structure on Γ(V ) by
ηV : Γ(V )→ Γ(L−)⊗ Γ(V ) ηV (v) = u⊗ v
The maps η are clearly morphisms of gD–modules and they satisfy
ΦD(η ⊗ id)ηV = (id⊗ηV )ηV
where η := ηL− and η(u) = u⊗ u = δ(u). Consider
RD,V ∈ EndUgD(Γ(L−),Γ(V ))
and define the following map:
δV : Γ(V )→ Γ(L−)⊗ Γ(V ) δV = RD,V ηV
Proposition. The maps δV define on any Γ(V ), V ∈ Dg(Φ), a structure
of comodule over Γ(L−).
Proof. Clearly we have that the relative R–matrix satisfies
RD,1,23 = Φ−1D,231RD,13ΦD,213RD,12Φ−1
D,123
RD,12,3 = ΦD,312RD,13Φ−1D,132RD,23ΦD,123
204 A
and therefore
RD,12ΦD,312RD,13Φ−1D,132RD,23ΦD,123 = ΦD,321RD,23Φ−1
D,231RD,13ΦD,213RD,12
Moreover, we have the obvious relations
RD,1,23(id⊗ηV ) = (id⊗ηV )RD RD,12,3(ηV ⊗ id) = (ηV ⊗ id)RD
for V ∈ DΦ(g). Then
(id⊗δV )δV = RD,23(id⊗ηV )RDηV
= RD,23RD,1,23(id⊗ηV )ηV
= ΦDRD,12RD,12,3Φ−1D (id⊗ηV )ηV
= ΦDRD,12RD,12,3(η ⊗ id)ηV
= ΦDRD,12(η ⊗ id)RDηV
= ΦD(δ ⊗ id)δ
This concludes the proof of the Theorem 3.1.
Bibliography
[ATL1] A. Appel, V. Toledano Laredo, Quasi–Coxeter categories and a relative Etingof–
Kazhdan quantization functor. Available at arXiv:1212.6720.
[ATL2] A. Appel, V. Toledano Laredo, A rigidity theorem for quasi–Coxeter quasitrian-
gular quasibialgebra structure on affine Kac–Moody algebras. Preliminary version
available at www.math.neu.edu/aappel/Andrea Appel/Preliminary.html
[CP] V. Chari, A. Pressley, A guide to quantum groups. Cambridge University Press,
Cambridge, 1994.
[D] A. A. Davydov, Monoidal categories. Algebra, 5. J. Math. Sci. (New York) 88
(1998), no. 4, 457-519.
[DCP1] C. De Concini, C. Procesi, Wonderful models of subspace arrangements. Selecta
Math. (N. S.) 1 (1995), no. 3, 459-494.
[DCP2] , Hyperplane arrangements and holonomy equations. Selecta Math. (N. S.) 1
(1995), no. 3, 495-535.
[D1] V. G. Drinfeld, Quantum groups. Proceedings of the International Congress of
Mathematicians, Vol. 1,2 (Berkeley, Calif., 1986), 798–820, Amer. Math. Soc.,
Providence, RI, 1987.
[D2] , Quasi Hopf algebras and Knizhnik–Zamolodchikov equations. Problems of
modern quantum field theory (Alushta, 1989), 1-13, Res. Rep. Phys., Springer,
Berlin, 1989.
[D3] , Quasi–Hopf Algebras, Leningrad Math. J. 1 (1990), no. 6, 1419-1457.
[D4] , On quasitriangular quasi-Hopf algebras and on a group that is closely con-
nected with Gal(Q/Q). Leningrad Math. J. 2 (1991), no. 4, 829-860.
[D5] , On some unsolved problems in quantum group theory. Quantum groups
(Leningrad, 1990), 1-8, Lecture Notes in Math., 1510, Springer, Berlin, 1992.
205
206 BIBLIOGRAPHY
[EG] B. Enriquez, N. Geer, Compatibility of quantization functors of Lie bialgebras with
duality and doubling operations, Selecta Math. (N.S.) 15 (2009), no. 1, 1-59.
[EK1] P. Etingof, D. Kazhdan, Quantization of Lie bialgebras. I. Selecta Math. (N.S.) 2
(1996), no. 1, 1-41.
[EK2] , Quantization of Lie bialgebras. II. Selecta Math. (N.S.) 4 (1998), no. 2,
213-231.
[EK6] , Quantization of Lie bialgebras. VI. Quantization of generalized Kac–Moody
algebras, Transform. Groups 13 (2008), no. 3-4, 527-539.
[FMTV] G. Felder, Y. Markov, V. Tarasov, A. Varchenko, Differential Equations Compat-
ible with KZ Equations, Math. Phys. Anal. Geom. 3 (2000), no. 2, 139-177.
[Ka] V. Kac, Infinite-dimensional Lie algebras. Third edition. Cambridge University
Press, Cambridge, 1990.
[KR] A. N. Kirillov, N. Reshetikin, q–Weyl group and a multiplicative formula for uni-
versal R–matrices, Comm. Math. Phys. 134 (1990), no. 2, 421-431.
[Ko] T. Kohno, Monodromy representations of braid groups and Yang-Baxter equations,
Ann. Inst. Fourier (Grenoble) 37 (1987), no. 4, 139-160.
[L] G. Lusztig, Introduction to quantum groups. Progress in Mathematics, 110.
Birkhauser Boston, Inc., Boston, MA, 1993.
[MTL] J. Millson, V. Toledano Laredo, Casimir operator and monodromy representations
of generalized braid groups, Transformation Groups 10 (2005), no. 2, 217-254.
[S] Y. Soibelman, Algebra of functions on a compact quantum group and its represen-
tations, Leningrad Math. J. 2 (1991), no. 1, 161-178.
[TL1] V. Toledano Laredo, Flat connections and quantum groups, Acta Appl. Math. 73
(2002), no. 1-2, 155-173.
[TL2] , A Kohno–Drinfeld theorem for quantum Weyl groups, Duke Math. J. 112
(2002), no. 3, 421-451.
[TL3] , Cohomological construction of relative twists, Adv. Math. 210 (2007), no. 1,
375-403.
BIBLIOGRAPHY 207
[TL4] , Quasi–Coxeter algebras, Dynkin diagram cohomology and quantum Weyl
groups, Int. Math. Res. Pap. IMRP 2008, art. ID rpn009, 167 pp.
[TL5] , Quasi–Coxeter quasitriangular quasibialgebras and the Casimir connection.
(forthcoming)
[TL6] , Affine extension of rational Casimir connection. (forthcoming)