ARTICLE IN PRESS

Advances in Mathematics 186 (2004) 363–395

http://www.elsevier.com/locate/aim

Poisson algebras associated to

quasi-Hopf algebras
Benjamin Enriquez
and Gilles Halbout
Institut de Recherche Mathématique Avancée de Strasbourg, UMR 7501 de l’Université Louis Pasteur et du
CNRS, 7, Rue R. Descartes, F-67084 Strasbourg, France
Received 25 November 2002; accepted 13 August 2003

Communicated by P. Etingof

Abstract

We define admissible quasi-Hopf quantized universal enveloping (QHQUE) algebras by

_-adic valuation conditions. We show that any QHQUE algebra is twist-equivalent to an
admissible one. We prove a related statement: any associator is twist-equivalent to a Lie
associator. We attach a quantized formal series algebra to each admissible QHQUE algebra
and study the resulting Poisson algebras.
r 2003 Elsevier Inc. All rights reserved.

0. Introduction

In [WX], Weinstein and Xu introduced a geometric counterpart of the R-matrix of

a quasi-triangular quantum group: they proved that if ðg; rÞ is a finite dimensional
quasi-triangular Lie bialgebra, then the dual group G
is equipped with a braiding
RWX AAutððG
Þ2 Þ with properties analogous to those of quantum R-matrices (in
particular, it is a set-theoretic solution of the quantum Yang–Baxter Equation). An
explicit relation to the theory of quantum groups was later given in [GH,EH,EGH]:
to a quasi-triangular QUE algebra ðU_ ðgÞ; m; RÞ quantizing ðg; rÞ; one associates
its quantized formal series algebra (QFSA) U_ ðgÞ0 CU_ ðgÞ; U_ ðgÞ0 is a flat deforma-
tion of the Hopf–Poisson algebra OG
¼ ðUðg
ÞÞ
of formal functions of G
: Then
%
one proves that AdðRÞ preserves U_ ðgÞ0#2 ; and AdðRÞj_¼0 coincides with the

Corresponding author.
E-mail addresses: enriquez@math.u-strasbg.fr (B. Enriquez), halbout@math.u-strasbg.fr
(G. Halbout).

0001-8708/$ - see front matter r 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.aim.2003.08.007
 ARTICLE IN PRESS
364 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

% %
automorphism RWX of O#2 #2
G
; moreover, r ¼ _ logðRÞj_¼0 is a function of OG
;


independent of a quantization of g ; which may be expressed universally in terms of
r; and RWX coincides with the ‘‘time one automorphism’’ of the Hamiltonian vector
field generated by r:
In this paper, we study the analogous problem in the case of quasi-quantum
groups (quasi-Hopf QUE algebras). The classical limit of a QHQUE algebra is a Lie
quasi-bialgebra (LQBA). V. Drinfeld proposed to attach Poisson–Lie ‘‘quasi-
groups’’ to each LQBA ([Dr4]). Axioms for Poisson–Lie quasi-groups are the quasi-
Hopf analogues of the Weinstein–Xu axioms.
A Poisson–Lie quasi-group is a Poisson manifold X ; together with a ‘‘product’’
mX
Poisson map X 2
! X ; a unit for this product eAX ; and Poisson automorphisms
FX AAutðX 3 Þ; F12;3;4
X ; F1;23;4
X and F1;2;34
X AAutðX 4 Þ; such that

mX 3 ðid  mX Þ ¼ mX 3 ðmX  idÞ 3 FX ;

ðmX  id  idÞ 3 F12;3;4

X ¼ FX 3 ðmX  id  idÞ;

ðid  mX  idÞ 3 F1;23;4

X ¼ FX 3 ðid  mX  idÞ; etc:

and F1;2;34
X 3 F12;3;4
X ¼ ðid  FX Þ 3 F1;23;4
X 3 ðFX  idÞ:

A twistor for the quasi-group ðX ; mX ; FX Þ is a collection of Poisson automorphisms

ð12Þ3;4 1ð23Þ;4 1ð23Þ;4 1;ð23Þ4
FX AAutðX 2 Þ; FX12;3 ; FX1;23 AAutðX 3 Þ; FX ; FX ; FX12;34 ; FX ; FX A
4
AutðX Þ such that

ðmX  idÞ 3 FX12;3 ¼ FX 3 ðmX  idÞ;

1ð23Þ;4
ððmX 3 ðid  mX ÞÞ  idÞ 3 FX ¼ FX 3 ððmX 3 ðid  mX ÞÞ  idÞ;

ð12Þ3;4 1ð23Þ;4
FX ¼ ðFX  idÞ 3 FX 3 ðFX  idÞ 1 ; etc:

A twistor replaces the quasi-group ðX ; mX ; FX Þ by ðX ; m0X ; F0X Þ with m0X ¼ mX 3 FX

and F0X ¼ ðFX1;23 Þ 1 3 ðFX  idÞ 1 3 FX 3 FX1;23 3 ðid  FX Þ:
It is useful to further require that the automorphisms FX ; FX are given by
Lagrangian bisections of a Karasev–Weinstein groupoid associated with X 3 ; X 2 :
Other axioms for Poisson–Lie quasi-groups were proposed in a differential-
geometric language in [Ban,KS].
We do not know a ‘‘geometric’’ construction of a twist-equivalence class of
ðX ; mX ; FX Þ associated to each Lie quasi-bialgebra, in the spirit of [WX]. Instead we
generalize the ‘‘construction of a QFS algebra and passage to Poisson geometry’’
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 365

part of the above discussion, and we derive from there a construction of triples
ðX ; mX ; FX Þ; in the case of Lie quasi-bialgebras with vanishing cobracket.
Let us describe the generalization of the ‘‘construction of a QFS algebra’’ part
(precise statements are in Section 1). We introduce the notion of an admissible quasi-
Hopf QUE algebra, and we associate a QFSA to such a QHQUE algebra. Each
QHQUE algebra can be made admissible after a suitable twist.
We generalize the ‘‘passage to Poisson geometry’’ part as follows. The reduction
modulo _ of the obtained QFS algebra is a quintuple ðA; m; P; D; jÞ* satisfying certain
b3
#
axioms; in particular expðVj* Þ is an automorphism of A ; and ðA; m; expðVj* ÞÞ
satisfies the axioms dual to those of ðX ; mX ; FX Þ:
When the Lie quasi-bialgebra arises from a metrized Lie algebra, admissible
QHQUE algebras quantizing it are given by Lie associators, and we obtain a quasi-
group ðX ; mX ; FX Þ using our construction. We also prove that its twist-equivalence
class does not depend on the choice of an associator.
Finally, we prove a related result: any associator is twist-equivalent to a unique
Lie associator.

1. Outline of results

Let K be a field of characteristic 0: Let ðU; mÞ be a topologically free K½½_-

algebra equipped with algebra morphisms
c U;
D : U-U # and e : U-K½½_

with ðe#idÞ 3 D ¼ ðid#eÞ 3 D ¼ id

such that the reduction of ðU; m; DÞ modulo _ is a universal enveloping algebra. Set
b
U 0 ¼ fxAUj for any tree P; dðPÞ ðxÞA_jPj U #jPj g

(see the definitions of a tree, dðPÞ ; and jPj in Section 2). We prove:

Theorem 1.1. U 0 is a topologically free K½½_-algebra. It is equipped with a complete

decreasing algebra filtration
b
ðU 0 ÞðnÞ ¼ fxAUj for any tree P; dðPÞ ðxÞA_n U #jPj g:

b
U 0 is stable under the multiplication m and the map D : U-U #2 induces a continuous
algebra morphism
0 , 1
% b X
DU 0 : U 0 -U 0#2 ¼ lim
’
@U 0#2 U 0ðpÞ #U 0ðqÞ A:
n p;qjpþq¼n
 ARTICLE IN PRESS
366 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

Set O :¼ U 0 =_U 0 : Then O is a complete commutative local ring and the reduction
modulo _ of DU 0 is a continuous ring morphism
0 , 1
%
X
DO : O-O#2 ¼ ’ lim
F
@O#2 OðpÞ #OðqÞ A;
n p;qjpþq¼n

where OðpÞ ¼ U 0ðpÞ =ð_U-U 0ðpÞ Þ:

Theorem 1.2. Let ðU; m; D; FÞ be a quasi-Hopf QUE algebra. Let g be the Lie algebra
of primitive elements of U=_U; so U=_U ¼ UðgÞ: Assume that
%
_ logðFÞAðU 0 Þ#3 : ð1:1Þ

Then there is a noncanonical isomorphism of filtered algebras U 0 =_U 0 -Sb ðgÞ; where
Sb ðgÞ is the formal series completion of the symmetric algebra S ðgÞ:

When ðU; m; D; FÞ satisfies the hypothesis (1.1), we say that it is admissible.

In that case, we say that U 0 is the quantized formal series algebra (QFSA)
corresponding to ðU; m; D; FÞ: Let us recall the notion of a twist of a quasi-
b
Hopf QUE algebra ðU; m; D; FÞ: This is an element F AðU #2 Þ ; such that
ðe#idÞðF Þ ¼ ðid#eÞðF Þ ¼ 1: It transforms ðU; m; D; FÞ into the quasi-Hopf algebra
ðU; m; F D; F FÞ; where
F
D ¼ AdðF Þ 3 D; and F
F ¼ ð1#F Þðid#DÞðF ÞFðD#idÞðF Þ 1 ðF #1Þ 1 :

Theorem 1.3.
(1) Let ðU; m; D; FÞ be an admissible quasi-Hopf QUE algebra. Let us say that a twist
%
F of U is admissible if _ logðF ÞAU 0#2 : Then the twisted quasi–Hopf algebra
ðU; m; D; FÞ is also admissible, and its QFSA coincides with U 0 :
F F

(2) Let ðU; m; D; FÞ be an arbitrary quasi-Hopf QUE algebra. There exists a twist F0
of U such that the twisted quasi–Hopf algebra ðU; m; F0 D; F0 FÞ is admissible.

Theorem 1.3 can be interpreted as follows. Let ðU; mÞ be a formal deformation of

b
a universal enveloping algebra. The set of twists of U is a subgroup T of ðU #2 Þ :
Denote by Q the set of all quasi-Hopf structures on ðU; mÞ; and by Qadm the subset of
admissible structures. If Q is nonempty, then Qadm is also nonempty, and all its
elements give rise to the same subalgebra U 0 CU (Theorem 1.3, (1)). Using U 0 ; we
then define the subgroup Tadm CT of admissible twists. We have a natural action of
T on Q; which restricts to an action of Tadm on Qadm : Theorem 1.3 (2) says that the
natural map

Qadm =Tadm -Q=T

 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 367

is surjective. Let us explain why it is not injective in general. Any QUE Hopf algebra
ðU; m; DÞ is admissible as a quasi-Hopf algebra. If uAU  and F ¼ ðu#uÞDðuÞ 1 ;
then ðU; m; F DÞ is a Hopf algebra. So ðU; m; DÞ and ðU; m; F DÞ are in the same class
of Q=T: These are also two elements of Qadm ; the corresponding QFS algebras are
U 0 and AdðuÞðU 0 Þ: In general, these algebras do not coincide, so ðU; m; DÞ and
ðU; m; F DÞ are not in the same class of Qadm =Tadm :
The following result is a refinement of Proposition 3.10 of [Dr2]. Let ðg; m; jÞ be a
V
pair of a Lie algebra ðg; mÞ and jA 3 ðgÞg : Then ðg; d ¼ 0; jÞ is a Lie bialgebra.

Proposition 1.4. There exists a series EðjÞAUðgÞ#3 ½½_; expressed in terms of ðm; jÞ
by universal acyclic expressions, such that ðUðgÞ½½_; m0 ; D0 ; EðjÞÞ is an admissible
quantization of ðg; m; d ¼ 0; jÞ:

This proposition is proved in Section 6.

Recall that the main axioms for a quasi-Hopf algebra ðA; m; D; FÞ are that (a) F
measures the noncoassociativity of D; and (b) F satisfies the pentagon equation. By
analogy, we set:

* where
Definition 1.5. A quasi-Hopf Poisson algebra is a quintuple ðA; m0 ; P; D; jÞ;
* ðA; m0 Þ is a formal series algebra,
* P is a Poisson structure on A ‘‘vanishing at the origin’’ (i.e., such that ðPÞCmA ;
where mA is the maximal ideal of A),
* D : A-A# c A is a continuous Poisson algebra morphism, such that ðe#idÞ 3 D ¼
ðid#eÞ 3 D ¼ id; where e : A-A=mA ¼ K is the natural projection,
b3
#
* *
jAðm AÞ satisfies

*
ðid#DÞðDðaÞÞ ¼ j%ðD#idÞðDðaÞÞ%ð * aAA;
jÞ;

* 1;2;34 % j* 12;3;4 ¼ j
j * 2;3;4 %j
* 1;23;4 %j
* 1;2;3 ;

where we set f %g ¼ f þ g þ 12 Pð f ; gÞ þ ?; the Campbell–Baker–Hausdorff

(CBH) series of the Lie algebra ðA; PÞ:
Such a structure is the function algebra of a ‘‘formal Poisson–Lie quasi-group’’.
˜ # b2
If fAm *
A ; we define the twist of the quasi-Hopf Poisson algebra ðA; m0 ; P; D; jÞ
˜ f˜ f˜
* where
by f as the algebra ðA; m0 ; P; D; jÞ;
f˜ ˜
DðaÞ ¼ f%DðaÞ%ð f˜Þ; and

f˜
* ¼ f˜2;3 %f˜1;23 %j%ð
j * f˜12;3 Þ%ð f˜1;2 Þ;
˜ ˜
* is again a quasi-Hopf Poisson algebra.
then ðA; m0 ; P; f D; f jÞ
 ARTICLE IN PRESS
368 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

Remark 1.6. If L is any Artinian local K-ring with residue field K; set X ¼
HomK ðA; LÞ: Then X is the ‘‘Poisson–Lie quasi-group’’, in the sense of the
Introduction. Namely, D induces a product mX : X  X -X ; and expðVj* Þ;
expðVj* 12;3;4 Þ; etc., induce automorphisms FX ; F12;3;4
X ; etc., of X ; that satisfy the
b
quasi-group axioms (we denote by Vf the Hamiltonian derivation of A#k induced by
b
f AA#k ). Moreover, if f˜ is a twist of A; then expðV ˜ Þ; expðV ˜ 12;3 Þ; expðV ˜ ð12Þ3;4 Þ; etc.,
f f f
12;3 ð12Þ3;4
define a twistor ðFX ; FX ; FX ; yÞ of ðX ; mX ; FX Þ: Twisting A by f˜ corresponds
to twisting ðX ; mX ; FX Þ by ðFX ; FX12;3 ; yÞ:

* is a quasi-Hopf Poisson algebra, set g ¼ mA =ðmA Þ2 ;

Lemma 1.7. If ðA; m0 ; P; D; jÞ
then P induces a Lie bracket m on g; the map D D1;2 induces a linear map
* is an element j of L3 ðgÞ: Then ðg; m; d; jÞ is a
d : g-L2 ðgÞ; and the reduction of AltðjÞ
* by f˜ corresponds to twisting
Lie quasi-bialgebra. Moreover, twisting ðA; m0 ; P; D; jÞ
ðg; m; d; jÞ by

f :¼ ðAltð f˜Þ mod ðmA Þ2 #mA þ mA #ðmA Þ2 ÞAL2 ðgÞ:

Taking the reduction modulo _ of a QUE algebra over g induces a natural map

Qadm =Tadm -fquasi-Hopf Poisson algebra structures on Sb ðgÞg=twists:

To summarize, we have a diagram

quasi-Hopf poisson algebra

Q/T ← Qadm / Tadm → /twists
structures on S. (g)
〈

class ↓ ↓ red

{Lie quasi-bialgebra structures on (g,
)} / twists,

where class is the classical limit map described in [Dr2], and red is the map described
in Lemma 1.7. It is easy to see that this diagram commutes.
When U is a Hopf QUE algebra, it can be viewed as a quasi-Hopf algebra with
F ¼ 1; which is then admissible. The corresponding quasi-Hopf Poisson algebra is
the Hopf-Poisson structure on OG
¼ ðUðg
ÞÞ
; and j * ¼ 0:
Let ðg; m; d; jÞ be a Lie quasi-bialgebra. A lift ofðg; m; d; jÞ is a quasi-Hopf Poisson
algebra, whose reduction is ðg; m; d; jÞ: A general problem is to construct a lift for
any Lie quasi-bialgebra. We will not solve this problem, but we will give partial
existence and uniqueness results.
Assume that d ¼ 0: A Lie quasi-bialgebra is then the same as a triple ðg; m; jÞ of a
V
Lie algebra ðg; mÞ and jA 3 ðgÞg :

Theorem 1.8. (1) There exists a lift

ðSb ðgÞ; m0 ; Pg
; D0 ; jÞ

* ð1:2Þ
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 369

of ðg; m; d ¼ 0; jÞ: Here Pg
is the Kostant–Kirillov Poisson structure on g
and D0 is

the coproduct for which the elements of g are primitive.
(2) Any two lifts of ðg; m; d ¼ 0; jÞ of the form (1.2) are related by a g-invariant
twist.

Examples of Lie quasi-bialgebras with d ¼ 0 arise from metrized Lie algebras, i.e.,
pairs ðg; tg Þ of a Lie algebra g and tg AS 2 ðgÞg : Then j ¼ ½t1;2 2;3
g ; tg : Recall that a Lie
associator is a noncommutative formal series FðA; BÞ; such that log FðA; BÞ is a Lie
series ½A; Bþhigher degree terms, satisfying the pentagon and hexagon identities
(see [Dr3]).

Proposition 1.9. If F is a Lie associator, we may set j ¼ logðFÞðt%1;2

g ; t%g Þ; where t%g is
2;3 i; j

b b
the image of ti;g j in Sb ðgÞ#3 ; and we use the Poisson bracket of Sb ðgÞ#3 in the
g ; t%g Þ:
expression of logðFÞðt%1;2 2;3

We prove these results in Section 6. If now F is a general (non–Lie) associator,

g ; _tg ÞÞ is a quasi-Hopf QUE algebra, but it is admissible
ðUðgÞ½½_; m0 ; D0 ; Fð_t1;2 2;3

only when F is Lie ( for general g). According to Theorem 1.3 (2), it is twist-
equivalent to an admissible quasi-Hopf QUE algebra. We prove

Theorem 1.10. Any (non-Lie) associator is twist-equivalent to a unique Lie associator.

So the ‘‘concrete’’ version of the twist of Theorem 1.10 is an example of the twist
F0 of Theorem 1.3, (2).

2. Definition and properties of U 0

In this section, we prove Theorem 1.1. We first introduce the material for the
definition of U 0 : trees (a); the map dðPÞ (b); then we prove Theorem 1.1 in (c) and (d).

2.1. Binary complete planar rooted trees

Definition 2.1. An n-binary complete planar rooted tree (n-tree for short) is a set of
vertices and oriented edges satisfying the following conditions:
 each edge carries one of the labels fl; rg:
 if we set:

valency of a vertex ¼ ðcardðincoming edgesÞ; cardðoutgoing edgesÞÞ;

we have
3 there exists exactly one vertex with valency ð0; 2Þ (the root)
3 there exists exactly n vertices with valency ð1; 0Þ (the leaves)
 ARTICLE IN PRESS
370 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

3 all other vertices have valency ð1; 2Þ

3 if a vertex has valency ðx; 2Þ; then one of its outgoing edges has label l and the
other has label r:

Let us denote, for nX2;

Treen ¼ fn-binary complete planar rooted treesg:

By definition, Tree1 consists of one element (the tree with a root and one nonmarked
edge) and Tree0 consists of one element (the tree with a root and no edge). We will
write jPj ¼ n if P is a tree in Treen :

Definition 2.2 (Extracted trees). Let P be a binary complete planar rooted tree. Let
L be the set of its leaves and let L0 be a subset of L: We define the extracted subtree
PL0 as follows:
(1) P̃L0 is the set of all edges connecting the root with an element of L0 ;
(2) the vertices of P̃L0 all have valency ð0; 2Þ; ð1; 0Þ; ð1; 2Þ or ð1; 1Þ;
(3) PL0 is obtained from P̃L0 by replacing each maximal sequence of edges related by
a ð1; 1Þ vertex, by a single edge whose label is the label of the first edge of the
sequence.
Then PL0 is a jL0 j-binary complete planar rooted tree.

Definition 2.3 (Descendants of a tree). If we cut the tree P by removing its root and
the related edges, we get two trees P0 and P00 ; its left and right descendants.

In the same way, we define the left and right descendants of a vertex of P:
If P is a n-tree, there exists a unique bijection of the set of leaves with f1; y; ng;
such that for each vertex, the number attached to any leaf of its left descendant is
smaller than the number attached to any leaf of its right descendant.

b
2.2. Definition of DðPÞ ; dðPÞ : U-U #n

Let us place ourselves in the hypothesis of Theorem 1.1. For P0 (resp., P1 ; P2 ) the
only tree of Tree0 (resp., Tree1 ; Tree2 ), we set DðP0 Þ ¼ e (resp., DðP1 Þ ¼ id; DðP2 Þ ¼ D).
When P is a n-tree with descendants P0 and P00 ; we set
0 00
DðPÞ ¼ ðDðP Þ #DðP Þ Þ 3 D;

b
so DðPÞ is a linear map U-U #n :
b
We set dðPÞ ¼ ðid Z 3 eÞ#jPj 3 DðPÞ ; so dðPÞ is a linear map U-U #n :
In particular, dðP0 Þ ðxÞ ¼ eðxÞ; dðP1 Þ ðxÞ ¼ x eðxÞ1; and dðP2 Þ ðxÞ ¼ DðxÞ x#1
1#x þ eðxÞ1#1:
We use the notation dðiÞ ¼ DðPi Þ for i ¼ 0; 1; 2; and d ¼ dð2Þ :
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 371

We have also
0 00
dðPÞ ¼ ðdðP Þ #dðP Þ Þ 3 d:

2.3. Behavior of dðPÞ with respect to multiplication

If S ¼ fi1 ; y; ik g is a subset of f1; y; ng; where i1 oi2 o?oik ; the map x/xS is
b b
the linear map U #k -U #n ; defined by

x1 #?#xk /1#i1 1 #x1 #1#i2 i1 1 #x2 #?#1#ik ik 1 1 #xk #1#n ik 1 :

b
If S ¼ |; x/xS is the map K-U #n ; 1/1#n :

Proposition 2.4. For PATreen ; we have the identity

X 0 0 00 00
dðPÞ ðxyÞ ¼ ðdðS Þ ðxÞÞS ðdðS Þ ðyÞÞS ;
S0 ;S00 Cf1;y;ngj
S0 ,S00 ¼f1;y;ng

for any x; yAU:

This proposition is proved in Section 5.

2.4. Construction of U 0

Let us set

b
U 0 ¼ fxAUj for any tree P; dðPÞ ðxÞA_jPj U #jPj g:

Then U 0 is a topologically free K½½_-submodule of U: Moreover, if x; yAU 0 ; and P

is a tree, then
X 0
dðPÞ ð½x; yÞ ¼ ½dðPS Þ ðxÞS ; dðPS0 Þ ðyÞS ;
S;S0 Cf1;y;jPjg
S,S0 ¼f1;y;jPjg

the summand corresponding to a pair ðS; S0 Þ with S-S0 ¼ | is zero, and the _-adic
b
valuation of the other summands is XjSj þ jS0 jXjPj þ 1; so dðPÞ ð½x; yÞA_jPjþ1 U #jPj :
b
On the other hand, there exists zAU such that ½x; y ¼ _z; so dðPÞ ðzÞA_jPj U #jPj ; so
0 0 0 0
zAU and we get ½x; yA_U : It follows that U =_U is commutative. Let us set

U 0ðnÞ ¼ U 0 -_n U: ð2:3Þ

 ARTICLE IN PRESS
372 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

We have a decreasing filtration

U 0 ¼ U 0ð0Þ *U 0ð1Þ *U 0ð2Þ *?;

we have U 0ðnÞ C_n U; so U 0 is complete for the topology induced by this filtration.
This is an algebra filtration, i.e., U 0ðiÞ U 0ð j Þ CU 0ðiþj Þ : It induces an algebra filtration
on U 0 =_U 0 ;

U 0 =_U 0 *?*U 0ðiÞ =ðU 0ðiÞ -_U 0 Þ*?;

for which U 0 =_U 0 is complete. Moreover, the completed tensor product

0 , 1
X
% 0 ¼ lim @U 0 #
U 0 #U cU 0 c U 0ðqÞ A
U 0ðpÞ #
’F
n p;qjpþq¼n

identifies with

b jPjþjQj
c Uj8P; Q; ðdðPÞ #dðQÞ ÞðxÞA_jPjþjQj U #
lim ðfxAU #
’F
g=
n

b jPjþjQj
c Uj8P; Q; ðdðPÞ #dðQÞ ÞðxÞA_maxðn;jPjþjQjÞ U #
fxAU # gÞ:
If xAU 0 ; and P; Q are trees, with jPj; jQja0; then since dðPÞ ð1Þ ¼ dðQÞ ð1Þ ¼ 0;
we have

b
ðdðPÞ #dðQÞ ÞðDðxÞÞ ¼ ðdðPÞ #dðQÞ ÞðdðxÞÞ ¼ dðRÞ ðxÞA_jRj U #jRj

b
¼ _jPjþjQj U #jPjþjQj ;
where R is the tree whose left and right descendants are P and Q; so jRj ¼ jPj þ jQj:
On the other hand,

ðdðPÞ #eÞðDðxÞÞ ¼ dðPÞ ðxÞ#1A_jPj U #jPj :

ðe#dðPÞ ÞðDðxÞÞ ¼ 1#dðPÞ ðxÞA_jPj U #jPj ;

b
so DðxÞ satisfies ðdðPÞ #dðQÞ ÞðDðxÞÞA_jPjþjQj U #jPjþjQj for any pair of trees ðP; QÞ:
c U therefore induces an algebra morphism DU 0 : U 0 -U 0#2 %
D : U-U # ; whose
reduction modulo _ is a morphism of complete local rings
0 , 1
%
X
O-O#2 ¼ ’
lim
F
@O#2 OðpÞ #OðqÞ A;
n p;qjpþq¼n

where O ¼ U 0 =_U 0 and OðpÞ ¼ U 0ðpÞ =ðU 0ðpÞ -_U 0 Þ:

 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 373

3. Classical limit of U 0

We will prove Theorem 1.2 as follows. We first compare the various dðPÞ ; where P
is a n-tree (Proposition 3.1). Relations found between the dðPÞ imply that they have
_-adic valuation properties close to those of the Hopf case (Proposition 3.2). We then
prove Theorem 1.2.

3.1. Comparison of the various dðPÞ

b
Let P and P0 be n-trees. There exists an element FP;P0 AU #n ; such that DðPÞ ¼
ðP0 Þ
P;P0
AdðF Þ 3 D : The element F P;P0
is a product of images of F and F 1 by the
b b
various maps U #3 -U #n obtained by iteration of D: We have
0 0
FP ;P0 ¼ FP ;P FP;P0 ð3:4Þ

for any n-trees P0 ; P; P0 : For example,

ðid#DÞ 3 D ¼ AdðFÞ 3 ððD#idÞ 3 DÞ;

ðD#DÞ 3 D ¼ AdðF12;3;4 Þ 3 ððD#id#2 Þ 3 ðD#idÞ 3 DÞ; etc:

%
Proposition 3.1. Assume that _ logðFÞAðU 0 Þ#3 : Then there exists a sequence of
elements
X PP RSn % %
F PP0 RSn ¼ F1;a 0 #?#Fn;a PP0 RSn
AðU 0#n Þ#n ;
a

indexed by the triples ðR; S; nÞ; where R is a tree such that jRjon; S is a subset of
f1; y; ng with cardðSÞ ¼ jRj; and n is an integer X1; such that the equality
X X X
dðPÞ ¼ AdðFP;P0 Þ 3 dðP0 Þ þ
kjkon R a k tree SCf1;y;ng;
cardðSÞ¼k
X X PP0 RSn
ad_ ðF1;a PP0 RSn
Þ 3 ? 3 ad_ ðFn;a Þ 3 ðdðRÞ ÞS ð3:5Þ
nX1 a

holds. Here ad_ ðxÞðyÞ ¼ 1_ ½x; y:

Proof. Let us prove this statement by induction on n: When n ¼ 3; we find

dð1ð23ÞÞ ¼ AdðFÞdðð12Þ3Þ þ ðAdðFÞ 1Þðd1;2 þ d1;3 þ d2;3 þ dð1Þ1 þ dð1Þ2 þ dð1Þ3 Þ;

%
so the identity holds with F PP0 RSn ¼ n!1 ð_ log FÞ#n for all choices of ðR; S; nÞ; except
when jRj ¼ 0; in which case F PP0 RSn ¼ 0: Assume that the statement holds for any
 ARTICLE IN PRESS
374 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

pair of k-trees, kpn; and let us prove it for a pair ðP; P0 Þ of ðn þ 1Þ-trees. For k any
integer, let Pleft ðkÞ be the k-tree corresponding to

dðPleft ðkÞÞ ¼ ðd#id#k 2 Þ 3 ? 3 d:

Thanks to (3.4), we may assume that P0 ¼ Pleft ðn þ 1Þ and P is arbitrary. Let P0 and
0 00
P00 be the subtrees of P; such that jP0 j þ jP00 j ¼ n þ 1; and dðPÞ ¼ ðdðP Þ #dðP Þ Þ 3 d:
Let P1 and P2 be the n-trees such that
0 00 0 00
dðP1 Þ ¼ ðdðPleft ðk ÞÞ #dðP Þ Þ 3 d and dðP2 Þ ¼ ðdðPleft ðk ÞÞ #dðPleft ðk ÞÞ Þ 3 d

Assume that jP1 ja1: Using (3.4), we reduce the proof of (3.5) to the case of the pairs
ðP; P1 Þ; ðP1 ; P2 Þ and ðP2 ; P0 Þ: Then the induction hypothesis applied to the pair
ðP0 ; Pleft ðk0 ÞÞ; together with FP;P1 ¼ FP ;Pleft ðk Þ #1#k ; implies
0 0 00

X X X
dðPÞ ¼ AdðFP;P1 Þ 3 dðP1 Þ þ
kjkok0 R a k tree SCf1;y;k0 g;
cardðSÞ¼k
X X P0 Pleft ðk0 ÞSn
#1#k Þ?ad_ ðFn;a #1#k Þ
00 0 0 00
P Pleft ðk ÞSn
AdðFP;P1 Þ 3 ad_ ðF1;a
nX1 a
00 0
3 ððd ðRÞ
#dðP Þ Þ 3 dÞS;k þ1;y;nþ1 ;

which is (3.5) for ðP; P1 Þ: In the same way, one proves a similar identity relating P1
and P2 : Let us now prove the identity relating P2 and P0 : We have dðP2 Þ ¼
ðd#id#n 1 Þ 3 dðP2 Þ and dðP0 Þ ¼ ðd#id#n 1 Þ 3 dðP0 Þ ; where P02 and P00 are n-trees. We
0 0

have

FP2 ;P0 ¼ ðD#id#n 2 Þ 3 FP2 ;P0

0 0

so we get

dðP2 Þ ¼ AdðFP2 ;P0 Þ 3 dðP0 Þ

0 0 0
þ ðAdðFP2 ;P0 Þ AdððFP2 ;P0 Þ1;3;y;nþ1 ÞÞ 3 ðdðP0 Þ Þ1;3;y;nþ1
0 0 0
þ ðAdðFP2 ;P0 Þ AdððFP2 ;P0 Þ2;3;y;nþ1 ÞÞ 3 ðdðP0 Þ Þ2;3;y;nþ1
0
BX X X X
þ ðd#id#n 1 ÞB
@
kpn R a k tree SCf1;y;ng; nX1
cardðSÞ¼k
1
X P0 P00 Sn P0 P00 Sn C
 ad_ ðF1;a2 Þ?ad_ ðFn;a2 Þ 3 ðdðRÞ ÞS C
A:
a
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 375

% %
We have _ log FP2 ;P0 AU 0#nþ1 and _ log FP2 ;P0 AU 0#n
0 0
; this fact and the relations
ðd#id#n 1 Þðad_ ðx1 Þ?ad_ ðxn Þ 3 ðdðRÞ ÞS Þ

¼ ðad_ ðx12;y;nþ1
1 Þ 3 ? 3 ad_ ðx12;y;nþ1
n Þ ad_ ðx1;3;y;nþ1
1 Þ 3 ? 3 ad_ ðx1;3;y;nþ1
n Þ

ad_ ðx2;3;y;nþ1
1 Þ 3 ? 3 ad_ ðx2;3;y;nþ1
n ÞÞ 3 ðdðRÞ ÞSþ1

if 1eS; and
ðd#id#n 1 Þðad_ ðx1 Þ?ad_ ðxn Þ 3 ðdðRÞ ÞS Þ
0
¼ ad_ ðx12;y;nþ1
1 Þ 3 ? 3 ad_ ðx12;y;nþ1
n Þ 3 ððd#id#n 1 Þ 3 dðRÞ Þ1;2;S þ1

þ ðad_ ðx12;y;nþ1
1 Þ 3 ? 3 ad_ ðx12;y;nþ1
n Þ ad_ ðx1;3;y;nþ1
1 Þ 3 ? 3 ad_ ðx1;3;y;nþ1
n ÞÞ
0
3 ðdðRÞ Þ1;S þ1 þ ðad_ ðx12;y;nþ1
1 Þ 3 ? 3 ad_ ðx12;y;nþ1
n Þ ad_ ðx12;3;y;nþ1 Þ 3 ? 3 ad_ ðx2;3;y;nþ1
n ÞÞ
0
3 ðdðRÞ Þ2;S þ1

if S ¼ S0 ,f1g; where 1eS0 ; imply that dðP2 Þ AdðFP2 ;P0 Þ 3 dðP0 Þ has the desired form.
Let us now treat the case jP1 j ¼ 1: For this, we introduce the trees P3 and P4 ; such
that:
dðP3 Þ ¼ ðid#n 1 #dÞ 3 ðid#n 2 #dÞ 3 ? 3 d;

dðP4 Þ ¼ ðid#n 1 #dÞ 3 ðd#id#n 2 Þ 3 ðd#id#n 3 Þ 3 ? 3 ðd#idÞ 3 d:

We then prove the relation for the pair ðP; P3 Þ in the same way as for ðP1 ; P2 Þ (only
the right branch of the tree is changed); the relation for ðP3 ; P4 Þ in the same way as
for ðP2 ; P3 Þ (instead of composing a known relation by d#id#n 1 ; we compose it
with id#n 1 #d); and using the identity
dðP4 Þ ¼ ðd#id#n 1 Þ 3 ðid#n 2 #dÞ 3 ðd#id#n 3 Þ 3 ? 3 d;

we prove the relation for ðP4 ; PÞ in the same way as for ðP2 ; P3 Þ (composing a known
relation by d#id#n 1 ). &

3.2. Properties of dðPÞ

Proposition 3.2. Let n be an integer and xAU:

b
(1) Assume that for any tree R; such that jRjon; we have dðRÞ ðxÞA_jRj U #jRj : Then the
conditions
b
dðPÞ ðxÞA_n U #n ; ð3:6Þ

where P is an n-tree, are all equivalent.

 ARTICLE IN PRESS
376 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

b
(2) Assume that for any tree R; such that jRjon; we have dðRÞ ðxÞA_jRjþ1 U #jRj : Then
the elements

1 ðPÞ
d ðxÞ mod _ AUðgÞ#n ;
_n

where P is an n-tree, are all equal and belong to ðg#n ÞSn ¼ S n ðgÞ:

Proof. Let us prove (1). We have dðPÞ ¼ ðid Z 3 eÞ#jPj 3 dðPÞ ; where Z : K½½_-U is
the unit map of U; so
X X X X X
dðPÞ ¼ AdðFP;P0 Þ 3 dðP0 Þ þ
kjkon R a k tree SCf1;y;ng; nX1 a
cardðSÞ¼k

ðid Z 3 eÞ#n 3 ad_ ðF1;a

PP0 RSn PP0 RSn
Þ 3 ? 3 ad_ ðFn;a Þ 3 ðdðRÞ ÞS :

Then (1) follows from:

Lemma 3.3. Let S be a subset of f1; y; ng (we will write jSj instead of cardðSÞ) and
b
let U0 be the kernel of the counit of U. Let xA_jSj ðU0 Þ#jSj and F1 ; y; Fn be elements of
%
ðU 0 Þ#n : Then
b
ðid Z 3 eÞ#n ðad_ ðF1 Þ?ad_ ðFn ÞðxS ÞÞA_n ðU0 Þ#n :

%
Proof of Lemma. Each element F AðU 0 Þ#n is uniquely expressed as a sum F ¼
P
SAPðf1;y;ngÞ FS ; where FS belongs to the image of

% %
ðU00 Þ#jSj -ðU 0 Þ#n ;

f /f S ;

Pðf1; y; ngÞ is the set of subsets of f1; y; ng; and U00 is the kernel of the counit of
U 0 : Then

ðid Z 3 eÞ#n ðad_ ðF1 Þ?ad_ ðFn ÞðxS ÞÞ

X
¼ ðid Z 3 eÞ#n ðad_ ððF1 ÞS1 Þ?ad_ ððFn ÞSn ÞðxS ÞÞ:
S1 ;y;Sn APðf1;y;ngÞ

The summands corresponding to ðS1 ; y; Sn Þ such that S1 ,?Sn ,Saf1; y; ng are

b
all zero. Moreover, each ðFa ÞSa can be expressed as ð fa ÞSa ; where fa A_jSa j ðU0 Þ#jSa j :
The lemma then follows from the statement:
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 377

b b 0
Statement 3.4. If S; S0 Cf1; y; ng; xA_jSj ðU0 Þ#jSj ; yA_jS j ðU0 Þ#jS j ; then 1_½x; y can
0

b
be expressed as zS,S ; where zA_jS,S j ðU0 Þ#jS,S j :
0 0 0

Proof. If S-S0 ¼ |; then ½x; y ¼ 0; so the statement holds. If S-S0 a|; then the
_-adic valuation of 1_ ½x; y is X 1 þ jSj þ jS0 jXjSj þ jS0 j jS-S0 j ¼ jS,S0 j: &

Let us now prove property (2). The above arguments immediately imply that the
ð_1n dðPÞ ðxÞ mod _Þ; jPj ¼ n; are all equal. This defines an element Sn ðxÞAUðgÞ#n :
b
If jPj ¼ n; we have ðid#k #d#id#n k 1 Þ 3 dðPÞ ðxÞA_nþ1 U #nþ1 ; so if d0 : UðgÞ-
UðgÞ#UðgÞ is defined by d0 ðxÞ ¼ D0 ðxÞ x#1 1#x þ eðxÞ1#1; D0 being the
coproduct of UðgÞ; then ðid#k #d0 #id#n k 1 ÞðSn ðxÞÞ ¼ 0; so

Sn ðxÞAg#n : ð3:7Þ

Let us denote by si;iþ1 the permutation of the factors i and i þ 1 in a tensor power.
For i ¼ 1; y; n 1; let us compute ðsi;iþ1 idÞðSn ðxÞÞ: Let P0 be a ðn 1Þ-tree and
let P be the n-tree such that dðPÞ ¼ ðid#i 1 #d#id#n i 1 Þ 3 dðP Þ : Then
0

 
1
ðsi;iþ1 idÞðSn ðxÞÞ ¼ ðid#i 1 #ðd2;1 dÞ#id#n i 1 Þ 3 dðP Þ ðxÞ mod _ :
0

_
0 b
By assumption, dðP Þ ðxÞA_n U #n 1 ; moreover, d2;1 d ¼ D2;1 D; so ðd2;1 dÞðUÞC
c UÞ; therefore
_ðU #
b
ðid#i 1 #ðd2;1 dÞ#id#n i 1 Þ 3 dðP Þ ðxÞA_nþ1 U #n ;
0

it follows that ðsi;iþ1 idÞðSn ðxÞÞ ¼ 0; therefore Sn ðxÞ is a symmetric tensor of

UðgÞ#n : Together with (3.7), this gives Sn ðxÞAðg#n ÞSn : This ends the proof of
Proposition 3.2. &

3.3. Flatness of U 0 (proof of Theorem 1.2)

Let us set
b
U 00ðnÞ ¼ fxAU 0 jdðPÞ ðxÞA_jPjþ1 U #jPj if jPjpn 1g:

Then by Proposition 2.4, we have a decreasing algebra filtration

U 0 ¼ U 00ð0Þ *U 00ð1Þ *U 00ð2Þ *?*_U 0 : ð3:8Þ

We have U 00ðnÞ *U 0ðnÞ þ _U 0 (we will see later that this is an equality). We derive
from (3.8) a decreasing filtration
O ¼ O00ð0Þ *O00ð1Þ *O00ð2Þ *?;
 ARTICLE IN PRESS
378 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

where O ¼ U 0 =_U 0 and O00ðnÞ ¼ U 00ðnÞ =_U 0 : We have

\
O00ðnÞ ¼ f0g;
nX0
T
this means that nX0 U ¼ _U 0 ; which is proved as follows: if x belongs to
00ðnÞ

-nX0 U 00ðnÞ ; then eðxÞ ¼ Oð_Þ; x eðxÞ1 ¼ Oð_2 Þ; so x ¼ _y; where yAU: Moreover,
djPj ðyÞ ¼ Oð_jPj Þ for any P; so yAU 0 :
The fact that O is complete for this filtration will follow from its identification
with the filtration O*O0ð1Þ *? (this will be established in Proposition 3.6), where
O0ðiÞ ¼ U 0ðiÞ =_U 0 -U 0ðiÞ and U 0ðiÞ is defined in (2.3). We first prove:

Proposition 3.5. Set gr c nX0 O00ðnÞ =O00ðnþ1Þ : Then there is a unique linear map
b 00 ðOÞ ¼ "
00 n
ln : grn ðOÞ-S ðgÞ; taking the class of x to the common value of all
1 1 ðPÞ
n!ð_n d ðxÞ mod _Þ; where P is a n-tree. The resulting map l : gr b 00 ðOÞ-Sb ðgÞ is an
isomorphism of graded complete algebras.

Proof. In Proposition 3.2, we constructed a map U 00ðnÞ -Sn ðgÞ; by x/ common

value of n!1 ð_1n dðPÞ ðxÞ mod _Þ for all n-trees P: The subspace U 00ðnþ1Þ CU 00ðnÞ is clearly
contained in the kernel of this map, so we obtain a map

ln : U 00ðnÞ =U 00ðnþ1Þ ¼ O00ðnÞ =O00ðnþ1Þ -S n ðgÞ:

Let us prove that l ¼ " c nX1 ln is a morphism of algebras. If xAU 00ðnÞ and yAU 00ðmÞ ;
Proposition 2.4 implies that if R is any ðn þ mÞ-tree, we have
X 0 00
dðPÞ ðxyÞ ¼ dðRS0 Þ ðxÞS dðRS00 Þ ðyÞS :
S0 ;S00 Cf1;y;nþmgj
S0 ,S00 ¼f1;y;nþmg

The _-adic valuation of the term corresponding to ðS0 ; S00 Þ is XjS0 j þ jS00 j if jS0 jXn
and jS00 jXm; and XjS0 j þ jS00 j þ 1 otherwise, so the only contributions to
1
ð_nþm dðRÞ ðxyÞ mod _Þ are those of the pairs ðS0 ; S00 Þ such that S0 -S00 ¼ |: Then:

1 ðRÞ
d ðxyÞ mod _
_nþm
X  
1 ðRS0 Þ 1 ðRS00 Þ
¼ d ðxÞ mod _ d ðyÞ mod _
S0 ;S00 Cf1;y;nþmgj
_n _m
jS0 j¼n;jS00 j¼m;
S0 -S00 ¼|
X 0 00
¼ ðn!ln ðxÞS Þðm!lm ðyÞS Þ
S0 ;S00 Cf1;y;nþmgj
jS0 j¼n;jS00 j¼m;
S0 -S00 ¼|

¼ ðn þ mÞ!ln ðxÞlm ðyÞ;

 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 379

because the map

S  ðgÞ-ðTðgÞ; shuffle productÞ;
X
x1 ?xn / xsð1Þ #?#xsðnÞ
sASn

is an algebra morphism. Therefore lnþm ðxyÞ ¼ ln ðxÞlm ðyÞ: Let us prove that ln is
injective. If xAU 00ðnÞ is such that ð_1n dðPÞ ðxÞ mod _Þ ¼ 0 for any n-tree P; then
xAU 00ðnþ1Þ ; so its class in O00ðnÞ =O00ðnþ1Þ ¼ U 00ðnÞ =U 00ðnþ1Þ is zero. So each ln is injective,
so l is injective.
To prove that l is surjective, it suffices to prove that l1 is surjective. Let us fix xAg:
We will construct a sequence xn AU; nX0 such that eðxn Þ ¼ 0; ð1_ xn mod _Þ ¼ x;
xnþ1 Axn þ _n U for any nX1; and if P is any tree such that jPjpn;
b b
dðPÞ ðxn ÞA_jPj U #jPj (this last condition implies that dðQÞ ðxn ÞA_n U #jQj for jQjXn).
0
Then the limit x̃ ¼ limn-N ðxn Þ exists, belongs to U ; satisfies eðx̃Þ ¼ 0 and
ð1_ d1 ðx̃Þ mod _Þ ¼ x; so its class in U 00ð1Þ =U 00ð2Þ is a preimage of x:
Let us now construct the sequence ðxn ÞnX0 : We fix a linear map g-fyAUjeðyÞ ¼
0g; y/y; % such that for any yAg; ðy% mod _Þ ¼ y: We set x1 ¼ _x: % Let us construct
b
xnþ1 knowing xn : By Proposition 3.2, if Q is any ðn þ 1Þ-tree, d ðxn ÞA_n U #nþ1 ; and
ðQÞ

1 ðQÞ
ð_n d ðxn Þ mod _Þ is an element of S ðgÞ; independent of Q: Let us write this
nþ1

element as
X X X
yasð1Þ ?yasðnþ1Þ ; where ya1 #?#yanþ1 Ag#nþ1 :
sASnþ1 a a

Then we set

_n X X
xnþ1 ¼ xn y% asð1Þ ?y% asðnþ1Þ : &
ðn þ 1Þ! sAS a
nþ1

We now prove:

Proposition 3.6. (1) For any nX0; U 00ðnÞ ¼ U 0ðnÞ þ _U 0 ;

(2) The filtrations O ¼ O0ð0Þ *O0ð1Þ *? and O ¼ O00ð0Þ *O00ð1Þ *? coincide, and O is
complete and separated for this filtration.

Proof. Let us prove (1). We have to show that U 00ðnÞ CU 0ðnÞ þ _U 0 : Let xAU 00ðnÞ :
b
We have dðPÞ ðxÞA_jPjþ1 U #jPj for jPjpn 1; and for P an n-tree,
ðPÞ
ð_1n d ðxÞ mod _ÞAS n ðgÞ and is independent of P: Write this element of Sn ðgÞ as
P P a
a ysð1Þ #?ysðnÞ :
a
sASn
In Proposition 3.5, we construct a linear map g-U 0 -_U; x/x̃; such that eðx̃Þ ¼ 0
and ð1_ x̃ mod _Þ ¼ x:
 ARTICLE IN PRESS
380 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

P P
Set fn ¼ n!1 sASn a ỹ asð1Þ ?ỹsðnÞ
a
: Then each ỹia belongs to U 0 -_U; so
fn AU 0 -_n U ¼ U : Moreover, x fn belongs to U 00ðnþ1Þ : Iterating this procedure,
0ðnÞ

we construct elements fnþ1 ; fnþ2 ; y; where each fk belongs to U 0ðkÞ : The series
P 0
T 00ðkÞ
kXn fk converges in U ; denote by f its sum, then x f belongs to kXn U ¼
_U : So U CU þ _U : The inverse inclusion is obvious. This proves (1). Then
0 00ðnÞ 0ðnÞ 0

(1) immediately implies that for any n; O0ðnÞ ¼ O00ðnÞ : We already know that O is
complete and separated for O ¼ O0ð0Þ *O0ð1Þ *?; which proves (2). &

Proof of Theorem 1.2 (End). O is a complete local ring, and we have a ring
isomorphism grb
grðOÞ- Sb ðgÞ: Then any lift g-O0ð1Þ of O0ð1Þ -O0ð1Þ =O0ð2Þ ¼ g yields a
continuous ring morphism m : Sb ðgÞ-O: The associated graded of m is the identity, so
m is an isomorphism. So O is noncanonically isomorphic to Sb ðgÞ: &

Remark 3.7. When U is Hopf and g is finite-dimensional, U 0 =_U 0 identifies

canonically with OG
¼ ðUðg
ÞÞ
; where g
is the dual Lie bialgebra of g
(see [Dr1,Ga]). The natural projection Tðg
Þ-Uðg
Þ and the identification Tðg
Þ
¼
TbðgÞ (where TbðgÞ means the degree completion) induce an injection U 0 =_U 0 ¼
OG
¼ ðUðg
ÞÞ
+TbðgÞ: The map U 0 =_U 0 +TbðgÞ can be interpreted simply as
follows. For any xAU 0 ; we have ð_1n dn ðxÞ mod _ÞAg#n : Then U 0 =_U 0 +TbðgÞ takes
the class of xAU 0 to the sequence ð_1n dn ðxÞ mod _ÞnX0 :
On the quasi-Hopf case, we have no canonical embedding U 0 =_U 0 +TbðgÞ because
the various ð_1n dðPÞ ðxÞ mod _Þ do not necessarily coincide for all the n-trees P: This is
related to the fact that one cannot expect a Hopf pairing Uðg
Þ#ðU 0 =_U 0 Þ-K since
g
is no longer a Lie algebra, so Uðg
Þ does not make sense.
On the other hand, Theorem 1.2 can be interpreted as follows: in the Hopf case,
the exponential induces an isomorphism of formal schemes g
-G
; so U 0 =_U 0
identifies noncanonically with Og
¼ Sb ðgÞ: In the quasi-Hopf case, although there is
B
no formal group G
; we still have an isomorphism U 0 =_U 0 ! Sb ðgÞ:

4. Twists

4.1. Admissible twists

b
If ðU; m; D; FÞ is an arbitrary QHQUE algebra, we will call a twist F AðU #2 Þ
%
admissible if _ logðF ÞAðU 0 Þ#2 :

Proposition 4.1. Let ðU; m; D; FÞ be an admissible quasi-Hopf algebra and F an

admissible twist. Then the twisted quasi-Hopf algebra ðU; m; F D; F FÞ is admissible.
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 381

e mod _
Proof. Let e0 : U 0 -K be the composed map U 0 ! K½½_

! K; where e is the counit
0 #2 %
ð3Þ
%
map. Let m_ ¼ Kerðe0 Þ: We set m_ ¼ Kerðe#3 Þ: We have m_ ¼ m_ #ðU Þ þ
ð3Þ

0 % %
#2
%
U #m_ #U 0
þ ðU Þ #m
0 % _:
ð3Þ
When a; b are in ðm_ Þ2 ; the CBH series a%b ¼ a þ b þ ½a; b_ þ ? converges in
%
ðU 0 Þ#3 ; where ½ ; _ ¼ 1_ ½ ; : Indeed, ½m_ ; m_ _ Cm_ ; so
ð3Þ ð3Þ ð3Þ

ð3Þ ð3Þ ð3Þ

½ðm_ Þ2 ; ½y; ðm_ Þ2 _ ; _ Cðm_ Þnþ2 ;
P
where n is the number of ½ ; _ involved. Finally, a series nX0 fn ; where
% % %
converges in ðU Þ : indeed, m_ C_ðU Þ ; so ðm_ Þ C_n ðU 0 Þ#3 :
0 #3 0 #3
ð3Þ ð3Þ ð3Þ n
fn Aðm_ Þn ;
ð3Þ
Both f :¼ _ logðF Þ and _ logðFÞ belong to ðm_ Þ2 : Then we have

_ logðF FÞ ¼ f 1;2 %f 12;3 %ð_ logðFÞÞ%ð f 1;23 Þ%ð f 2;3 Þ:

% %
Since U 0#3 is stable under %; we have _ logðF FÞAU 0#3 : So ðU; m; F D; F FÞ is
admissible. &

Let us now prove

Proposition 4.2. Under the hypothesis of Proposition 4.1, the QFS algebra UF0
corresponding to ðU; m; F D; F FÞ coincides with the QFS algebra U 0 corresponding to
ðU; m; D; FÞ:

We will first prove the following lemma:

Lemma 4.3. Let P be an n-tree. Then

ðPÞ
X X X
dF ¼ dðPÞ þ
kpn R a k tree SCf1;y;ngj
cardðSÞ¼k
X X S;P
ad_ ð f1;a S;P
Þ 3 ? 3 ad_ ð fn;a Þ 3 ðdðRÞ ÞS ; ð4:9Þ
nX1 a

P % %
where for each n; S;P
a f1;a #?#fn;a AðU
S;P 0#n
Þ#n :

Remark 4.4. One can prove that in the right-hand side of (4.9), the contribution of
all terms with k ¼ n is ðAdðF ðPÞ Þ idÞ 3 dðPÞ where F ðPÞ is the product of F I;J (I; J
subsets of f1; y; ng; such that maxðIÞominðJÞ) and their inverses such that
ðPÞ
DF ¼ AdðF ðPÞ Þ 3 DðPÞ :
 ARTICLE IN PRESS
382 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

Proof of the Lemma. Eq. (4.9) may be proved by induction on jPj: Let us prove it for
the unique tree P such that jPj ¼ 2:

ð2Þ
X 1
dF ¼ dð2Þ þ ad_ ð f Þn ðdð2Þ ðxÞ þ dð1Þ ðxÞ1 þ dð1Þ ðxÞ2 Þ;
nX1
n!

where (1) and (2) are the 1- and 2-trees. Assume that (4.9) is proved when jPj ¼ n:
Let P0 be an ðn þ 1Þ-tree. Then for some iAf1; y; ng; we have
ðP0 Þ
dF ¼ ðid#i 1 #dF #id#n i Þ 3 dF ;
ðPÞ ð2Þ

where jP0 j ¼ n: Then:

ðP0 Þ ðP0 Þ b ðP0 Þ ic

dF ¼ ðid#i 1 #DF #id#n i Þ 3 dF
ðPÞ þ1
ðdF Þ1;y;i;y;nþ1 ðdF Þ1;y;iþ1;y;nþ1
0
B ðP0 Þ X X X
¼ ðid#i 1 #DF #id#n i Þ 3 B
@d þ
kpn R a k tree SCf1;y;ngj
cardðSÞ¼k
1
X X S;P0
C
S;P0
 ad_ ð f1;a Þ 3 ? 3 ad_ ð fn;a Þ 3 ðdðRÞ ÞS C
A
nX1 a

b ic
þ1
ð?Þ1?i;y;nþ1 ð?Þ1;y;iþ1;y;nþ1

0 b 0 ic
þ1
¼ AdðF i;iþ1
Þ 3 dðPÞ þ ðdðP Þ Þ1;y;i;y;nþ1 þ ðdðP Þ Þ1;y;iþ1;y;nþ1
X X X X X S;P0 1;y;fi;iþ1g;y;nþ1
þ ad_ ðð f1;a Þ Þ
kpn R a k tree SCf1;y;ngj nX1 a
cardðSÞ¼k

Þ 3 ð1#i 1 #D#1#n i Þ 3 ðdðRÞ ÞS
0
S;P 1;y;fi;iþ1g;y;nþ1
3 ad_ ðð fn;a Þ

b ic
þ1
ð?Þ1;y;i;y;nþ1 ð?Þ1;y;iþ1;y;nþ1 ;

this has the desired form because:

0 b 0 ic
þ1
ðAdðF i;iþ1 Þ 1Þ 3 ðdðPÞ þ ðdðP Þ Þ1;y;i;y;nþ1 þ ðdðP Þ Þ1;y;iþ1;y;nþ1 Þ
X 1 0 b 0 ic
þ1
¼ ad_ ð f i;iþ1 Þn ðdðPÞ þ ðdðP Þ Þ1;y;i;y;nþ1 þ ðdðP Þ Þ1;y;iþ1;y;nþ1 Þ:
nX1
n!

This proves (4.9). &

 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 383

Proof of Proposition 4.2 (End). One repeats the proof of Proposition 3.2 to prove
ðPÞ b
that if xAU 0 ; then we have dF ðxÞA_jPj U #jPj for any tree P: So U 0 CUF0 : Since
ðU; m; D; FÞ is the twist by F 1 of ðU; m; F D; F FÞ; and _ logðF 1 Þ ¼
%
0 #2 %
_ logðF ÞAðU Þ CðUF0 Þ#2 ; F 1 is admissible for ðU; m; F D; F FÞ; so we have also
UF0 CU 0 ; so UF0 ¼ U : 0
&

4.2. Twisting any algebra into an admissible algebra

Proposition 4.5. Let ðU; m; D; FÞ be a quasi-Hopf algebra. There exists a twist F0 such
that the twisted quasi-Hopf algebra ðU; m; F0 D; F0 FÞ is admissible.

Proof. We construct F0 as a convergent infinite product F0 ¼ ?Fn ?F2 ; where

b
Fn A1 þ _n 1 U #2 ; and the Fn ’s have the following property: if F%n ¼ Fn Fn 1 ?F2 ; if
b jPj
Fn ¼ F%n F; and dðPÞ
n : U-U
#
is the map corresponding to a tree P and to Dn ¼
AdðF%n Þ 3 D; then we have

b
jPjþjQjþjRj #
ðdðPÞ ðQÞ ðRÞ
n #dn #dn Þð_ logðFn ÞÞA_ U jPjþjQjþjRj

for any trees P; Q; R such that jPj þ jQj þ jRjpn:

Assume that we have constructed F1 ; y; Fn ; and let us construct Fnþ1 :
The argument of Proposition 3.2 shows that for any integers ðn1 ; n2 ; n3 Þ such that
n1 þ n2 þ n3 ¼ n þ 1; and any trees P; Q; R such that jPj ¼ n1 ; jQj ¼ n2 ; jRj ¼ n3 ;

1 ðPÞ ðQÞ ðRÞ
ðd #dn #dn Þð_ logðFn ÞÞ mod _ ASn1 ðgÞ#S n2 ðgÞ#S n3 ðgÞ;
_n n

and is independent of the trees P; Q; R: The direct sum of these elements is an

#3
element j% n of S ðgÞ ; homogeneous of degree n þ 1: Since Fn satisfies the pentagon
equation

ðid#id#Dn ÞðFn Þ 1 ð1#Fn Þðid#Dn #idÞðFn ÞðFn #1ÞðDn #id#idÞðFn Þ 1 ¼ 1;

j_n :¼ _ logðFn Þ satisfies the equation

ð ðid#id#Dn Þðj_n ÞÞ%ð1#j_n Þ%ððid#Dn #idÞðj_n ÞÞ%

ðj_n #1Þ%ð ðDn #id#idÞðj_n ÞÞ ¼ 0; ð4:10Þ

where we set
a%b ¼ a þ b þ 12 ½a; b_ þ ?
 ARTICLE IN PRESS
384 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

(the CBH series for the Lie bracket ½ ; _ ). The left-hand side of (4.10) is equal to
ð Dn #id#id þ id#Dn #id id#id#Dn Þðj_n Þ

þ ð1#j_n Þ þ ðj_n #1Þ þ brackets: ð4:11Þ

Let ðn1 ; n2 ; n3 ; n4 Þ be integers such that n1 þ ? þ n4 ¼ n þ 1: Let P; Q; R; S be trees

such that jPj ¼ n1 ; y; jSj ¼ n4 : Let us apply dðPÞ n #?#dn
ðSÞ
to (4.11). On the one
hand,
_ _
ðdðPÞ ðQÞ ðRÞ ðSÞ ðP,QÞ
n #dn #dn #dn ÞðDn #id#idÞðjn Þ ¼ ðdn #dðRÞ ðSÞ
n #dn Þðjn Þ;

where P,Q is the tree with left descendant P and right descendant Q: Therefore

1 ðPÞ ðQÞ ðRÞ ðSÞ _
ðd #d #d #d ÞðDn #id#idÞðj Þ mod _ ¼ ðD0 #id#idÞðj
% n Þn1 ;n2 ;n3 ;n4 ;
_n n n n n n

where the index ðn1 ; y; n4 Þ means the component in #4i¼1 Sni ðgÞ: In the same way,
ðdðPÞ ðQÞ ðRÞ ðSÞ
n #dn #dn #dn Þðð4:11Þ without bracketsÞ ¼ dðj
% n Þn1 ;n2 ;n3 ;n4 ;

where d : S  ðgÞ#2 -S  ðgÞ#3 is the co-Hochschild cohomology differential.

b
On the other hand, if a1 and a2 AU #4 are such that

b
infðjPjþ?þjSj;nÞ #
ðdðPÞ ðSÞ
n #?#dn Þðai ÞA_ U 4

for any trees ðP; y; SÞ; then if ðP; y; SÞ are such that jPj þ ? þ jSj ¼ n; we have

1 b
ðdðPÞ #?#d ðSÞ
Þ ½a ;
1 2a  A_nþ1 U #n ;
n n
_

one proves this in the same way as the commutativity of U 0 =_U 0 (see Theorem 1.1).
Then the relation _1n ðdðPÞ ðSÞ
n #?#dn Þð4:11Þj_¼0 ¼ 0 yields dðj % n Þ ¼ 0:
This relation implies that

% n ¼ dð f%n Þ þ ln ;
j

where f%n AS  ðgÞ#2 and ln AL3 ðgÞ: Moreover, fn and ln both have degree n þ 1: This
implies that ln ¼ 0: Let fn AðUðgÞ#2 Þpnþ1 be a preimage of f%n by the projection

ðUðgÞ#2 Þpnþ1 -ðUðgÞ#2 Þpnþ1 =ðUðgÞ#2 Þpn ¼ ðS  ðgÞ#2 Þnþ1

(where the indices n and pn mean ‘‘homogeneous part of degree n’’ and ‘‘part of
b b
degree pn’’). Let fn_ AU #2 be a preimage of fn by the projection U #2 -
b b
U #2 =_U #2 ¼ UðgÞ#2 : Set Fnþ1 ¼ expð_n 1 fn Þ: We may assume that _n fn A
% b
ðUðF%n Þ0 Þ#2 ; where UðF%n Þ0 ¼ fxAUjdðPÞ
n ðxÞA_
infðn;jPjÞ #
U jPj g: Then Fnþ1 ¼ Fnþ1 Fn :
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 385

If P; Q; R are such that jPj þ jQj þ jRj ¼ n þ 1; then

b
nþ1 #
ðdðPÞ ðQÞ ðRÞ
n #dn #dn Þð_ logðFnþ1 ÞÞA_ U nþ1 :

Then according to Lemma 4.3,

ðPÞ ðQÞ ðRÞ
ðdnþ1 #dnþ1 #dnþ1 dðPÞ ðQÞ ðRÞ
n #dn #dn Þð_logðFnþ1 ÞÞ

ðPÞ
has _-adic valuation 4jPj þ jQj þ jRj when jPj þ jQj þ jRjpn þ 1: So ðdnþ1 #
ðQÞ ðRÞ b
dnþ1 #dnþ1 Þð_ logðFnþ1 ÞÞA_jPjþjQjþjRj U #jPjþjQjþjRj whenever jPj þ jQj þ jRjp
n þ 1: &

5. Proof of Proposition 2.4

We work by induction on n: The statement is obvious when n ¼ 0; 1: For n ¼ 2;

we get

dð2Þ ðxyÞ ¼ dð2Þ ðxÞdð2Þ ðyÞ þ dð2Þ ðxÞðdð1Þ ðyÞ1 þ dð1Þ ðyÞ2 þ dð0Þ ðyÞ| Þ

þ ðdð1Þ ðxÞ1 þ dð1Þ ðyÞ1 þ dð0Þ ðyÞ| Þdð2Þ ðyÞ

þ dð1Þ ðxÞ1 dð2Þ ðyÞ2 þ dð1Þ ðxÞ2 dð2Þ ðyÞ1 ; ð5:12Þ

so the statement also holds.

Assume that the statement is proved when P is a n-tree. Let P% be a ðn þ 1Þ-tree.
There exists an integer kAf0; y; n 1g; such that P% may be viewed as the glueing of
the 2-tree on the kth leaf of a n-tree P: Then we have
%
dðPÞ ¼ ðid#k #dð2Þ #id#n k 1 Þ 3 dðPÞ :

Let us assume, for instance, that k ¼ n 1: If n is an integer, set

Sn ¼ fðS0 ; S00 ÞjS0 ; S00 Cf1; y; ng and S0 ,S00 ¼ f1; y; ngg:

Then

Sn ¼ ffng;| ðSn 1 Þ,f|;fng ðSn 1 Þ,ffng;fng ðSn 1 Þðdisjoint unionÞ;

where fa;b ðS0 ; S00 Þ ¼ ðS0 ,a; S00 ,bÞ: By hypothesis, we have
X
dðPÞ ðxyÞ ¼ dðPS1 Þ ðxÞS1 dðPS2 Þ ðyÞS2 ;
ðS1 ;S2 ÞASn
 ARTICLE IN PRESS
386 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

therefore
X
dðPS0 ,fng Þ ðxÞS ,fng dðPS00 Þ ðyÞS
0 00
dðPÞ ðxyÞ ¼
ðS0 ;S00 ÞASn 1
0 00
,fng
þ dðPS0 Þ ðxÞS dðPS00 ,fng Þ ðyÞS

þ dðPS0 ,fng Þ ðxÞS ,fng dðPS00 ,fng Þ ðyÞS

0 00
,fng
:

Applying id#n 1 #dð2Þ to this identity and using (5.12) and the identities

ðid#k #dð1Þ #id#jPj k 1 Þ 3 dðPÞ ¼ dðPÞ ;

ðid#k #dð0Þ #id#jPj k 1 Þ 3 dðPÞ ¼ 0;

%
we get dðPÞ ðxyÞ
X
ðððid#jS j #dð2Þ Þ 3 dðPS0 ,fng Þ ÞðxÞS ,fn;nþ1g dðPS00 Þ ðyÞS
0 0 00
¼
ðS0 ;S00 ÞASn 1
0 00
,fn;nþ1g
þ dðPS0 Þ ðxÞS ððid#jS j #dð2Þ Þ 3 dðPS00 ,fng Þ ÞðyÞS
00

þ ððid#jS j #dð2Þ Þ 3 dðPS0 ,fng Þ ÞðxÞS ,fn;nþ1g ððid#jS j #dð2Þ Þ 3 dðPS00 ,fng Þ ÞðyÞS
0 00
0 00 ,fn;nþ1g

0 00 00
þððid#jS j #dð2Þ Þ 3 dðPS0 ,fng Þ ÞðxÞS ,fn;nþ1g ðdðPS00 ,fng Þ ðyÞS ,fng ,fnþ1g
0
þ dðPS00 ,fng Þ ðyÞS Þ
0 0 00
þðdðPS0 ,fng Þ ðxÞS ,fng þ dðPS0 ,fng Þ ðxÞS ,fnþ1g Þððid#jS j #dð2Þ Þ 3 dðPS00 ,fng Þ ÞðyÞS ,fn;nþ1g
00

þ dðPS0 ,fng Þ ðxÞS ,fng dðPS00 ,fng Þ ðyÞS

0 00
,fnþ1g
þ dðPS0 ,fng Þ ðxÞS ,fnþ1g dðPS00 ,fng Þ ðyÞS
0 00
,fng
Þ:

%
So we get dðPÞ ðxyÞ
X % %
ðdðPS0 ,fn;nþ1g Þ ðxÞS ,fn;nþ1g dðPS00 Þ ðyÞS
0 00
¼
ðS0 ;S00 ÞASn 1
% 0 % 00
,fn;nþ1g
þ dðPS0 Þ ðxÞS dðPS00 ,fn;nþ1g Þ ðyÞS
% %
þ dðPS0 ,fn;nþ1g Þ ðxÞS ,fn;nþ1g dðPS00 ,fn;nþ1g Þ ðyÞS
0 00
,fn;nþ1g

% % %
þ dðPS0 ,fn;nþ1g Þ ðxÞS ,fn;nþ1g ðdðPS00 ,fng Þ ðyÞS
0 00
,fng 00
,fnþ1g
þ dðPS00 ,fnþ1g Þ ðyÞS Þ
% % %
þ ðdðPS0 ,fng Þ ðxÞS ,fng þ dðPS0 ,fnþ1g Þ ðxÞS ,fnþ1g ÞdðPS00 ,fn;nþ1g Þ ðyÞS
0 0 00
,fn;nþ1g

% 0 % 00 % 0 % 00
þdðPS0 ,fng Þ ðxÞS ,fng dðPS00 ,fnþ1g Þ ðyÞS ,fnþ1g
þ dðPS0 ,fnþ1g Þ ðxÞS ,fnþ1g dðPS00 ,fng Þ ðyÞS ,fng
Þ:
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 387

We have

Snþ1 ¼ ffn;nþ1g;fn;nþ1g ðSn 1 Þ,ffn;nþ1g;fng ðSn 1 Þ,ffn;nþ1g;fnþ1g ðSn 1 Þ

, ffn;nþ1g;| ðSn 1 Þ,ffng;fn;nþ1g ðSn 1 Þ,ffnþ1g;fn;nþ1g ðSn 1 Þ

, f|;fn;nþ1g ðSn 1 Þ,ffng;fnþ1g ðSn 1 Þ,ffnþ1g;fng ðSn 1 Þ ðdisjoint unionÞ;

where we recall that fa;b ðS0 ; S00 Þ ¼ ðS0 ,a; S00 ,bÞ: So we get
%
X %0 % 00
dðPÞ ðxyÞ ¼ dðPS% 0 Þ ðxÞjS j dðPS% 00 Þ ðyÞjS j :
% 0 ;S% 00 ÞASnþ1
ðS

The proof is the same for a general kAf0; y; n 1g: This establishes the induction.

6. Proofs of Proposition 1.4, Theorem 1.8 and Proposition 1.9

Proof of Proposition 1.4. According to [Dr2], Proposition 3.10, there exists a series
E0 ðjÞAUðgÞ#3 ½½_; expressed in terms of ðm; jÞ by universal acyclic expressions (and
therefore invariant), such that E0 ðjÞ ¼ 1 þ Oð_2 Þ; and E0 ðjÞ satisfies the pentagon
identity. Then ðUðgÞ½½_; m0 ; D0 ; E0 ðjÞÞ is a quasi-Hopf algebra. By Theorem 1.3(2),
there exists a twist F AUðgÞ#2 ½½_ ; such that ðUðgÞ½½_; m0 ; F D0 ; F E0 ðjÞÞ is
admissible.
E0 ðjÞ gives rise to a collection of invariant elements E0 ðjÞp1 ;p2 ;p3 ;n A#2i¼1 S pi ðgÞ;
defined by the condition that the image of E0 ðjÞ by the symmetrization map
P
UðgÞ#3 ½½_-S ðgÞ½½_ is nX0;p1 ;p2 ;p3 X0 _ E ðjÞp1 ;p2 ;p3 ;n : F is then expressed using
n 0

only the E0p1 ;p2 ;p3 ;n ; the Lie bracket and the symmetric group operations on the g#n :
So F is invariant and defined by universal acyclic expressions. Therefore F D0 ¼ D0 :
EðjÞ :¼ F E0 ðjÞ is then expressed by universal acyclic expressions, and defines an
admissible quantization of ðg; m; d ¼ 0; jÞ: &

%
Proof of Theorem 1.8(1). We have then EðjÞAðUðgÞ½½_0 Þ#3 : Since the coproduct
is D0 ; UðgÞ½½_0 is the complete subalgebra of UðgÞ½½_ generated by _g; so it is
a flat deformation of Sb ðgÞ with Kostant–Kirillov Poisson structure. We then set
j* :¼ EðjÞ modulo _: &

%
* 2 be the elements of Sb ðgÞ#3 such that
* 1; j
Proof of Theorem 1.8(2). Let j

ðSb ðgÞ; m0 ; Pg
; D0 ; j

* iÞ

*1 j
are quasi-Hopf Poisson algebras. Let C be the lowest degree component of j * 2:
Then the degree k of C is X4: Taking the degree k part of the difference of the
 ARTICLE IN PRESS
388 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

pentagon identities for j* 1 and j * 2 ; we find dðCÞ ¼ 0; where d : S ðgÞ#3 -S ðgÞ#4 is
the co-Hochschild differential. So AltðCÞAL3 ðgÞ; and since AltðCÞ also has degree
X4; AltðCÞ ¼ 0: If Cp1 ;p2 ;p3 is the component of C in #3i¼1 Spi ðgÞ then we may define
inductively BAðS ðgÞ#2 Þg ; homogeneous of degree k; such that dðBÞ ¼ C; as follows.
We set B0;k ¼ B1;k 1 ¼ 0; B2;k 2 ¼ 12ðid#mÞðC1;1;k 2 Þ; and
1
Biþ1;k i 1 ¼ ðid#mÞ½Ci;1;k i 1 þ ððid#dÞðBi;k i ÞÞi;1;k i 1 ;
iþ1
where Bi; j is the component of B in S i ðgÞ#S j ðgÞ and m is the product of S  ðgÞ:
Applying the twist B to the quasi-Hopf Poisson algebra ðSb ðgÞ; m0 ; Pg
; D0 ; j * 1Þ
amounts to replacing j* 1 by j* 01 ; such that j* 01 j
* 2 has valuation Xk þ 1: Applying
successive twists, we obtain the result. &

Proof of Proposition 1.9. According to [Dr3],

ðUðgÞ½½_; m0 ; D0 ; e_tg =2 ; Fð_t1;2

g ; _tg ÞÞ
2;3

is a quasi-triangular quasi-Hopf algebra. One checks that since F is Lie, it is

admissible; then the reduction modulo _ of the corresponding QFS algebra is the
quasi-Hopf Poisson algebra of Proposition 1.9. &

Remark 6.1. In the proof of Theorem 1.8(2), we cannot use Theorem A of [Dr2]
because we do not know that the twist constructed there is admissible.

7. Associators and Lie associators

In this section, we state precisely and prove Theorem 1.10.

7.1. Statement of the result

Let Tn ; nX2; be the algebra with generators ti; j ; 1p1ajpn; and relations
t ¼ ti; j ;
j;i

½ti; j þ ti;k ; t j;k  ¼ 0 when i; j; k are all distinct;

½ti; j ; tk;l  ¼ 0 when i; j; k; l are all distinct:

tn is defined as the Lie algebra with the same generators and relations. Then Tn ¼
Uðtn Þ: (tn is introduced in [Dr3]; Tn is called the ‘‘algebra of infinitesimal chord
diagrams’’ in [BN].)
When npm and ðI1 ; y; In Þ is a collection of disjoint subsets of f1; y; mg; there is
P
a unique algebra morphism Tn -Tm taking ti; j to a;b
aAIi ;bAIj t : We call it an
insertion-coproduct morphism and denote it by x/xI1 ;y;In : In particular, we have
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 389

an action of Sn on Tn : Let us attribute degree 1 to each generator ti; j ; this defines

gradings on the algebra Tn and on the Lie algebra tn : We denote by c
Tn and btn their
completions for this grading. Then ðc 
Tn Þ is the preimage of K by the natural
Tn -K; and the exponential is a bijection ðc
projection c Tn Þ0 -1 þ ðc Tn Þ0 (where
c c
ðTn Þ ¼ KerðTn -KÞ). We have an exact sequence
0

Tn Þ0 -ðc
1-1 þ ðc Tn Þ -K -1:

An associator is an element F of 1 þ ðc
Tn Þ0 ; satisfying the pentagon equation

F1;2;34 F12;3;4 ¼ F2;3;4 F1;23;4 F1;2;3 ; ð7:13Þ

the hexagon equations

t1;3 þt2;3 t1;3 t2;3
e 2 ¼ F3;1;2 e 2 ðF1;3;2 Þ 1 e 2 F1;2;3

and
t1;2 þt1;3 t1;3 t1;2
e 2 ¼ ðF2;3;1 Þ 1 e 2 F2;1;3 e 2 ðF1;2;3 Þ 1

and AltðFÞ ¼ 18 ½t1;2 ; t2;3 þ terms of degree 42: We denote by Assoc the set of
associators. If F satisfies the duality condition F3;2;1 ¼ F 1 ; then both hexagon
equations are equivalent. We denote by Assoc0 the subset of all FAAssoc satisfying
the duality condition. If F A1 þ ðc T2 Þ0 and FA1 þ ðc
T3 Þ0 ; the twist of F by F is
F
F ¼ F 2;3 F 1;23 FðF 1;2 F 12;3 Þ 1 :

This defines an action of 1 þ ðc T2 Þ0 on 1 þ ðc

T3 Þ0 ; which preserves Pent ¼ fFA1 þ
c3 Þ jF satisfies (7.13)g; Assoc and Assoc0 (Pent and Assoc are preserved because F
ðT 0
has the form f ðt1;2 Þ; f A1 þ tK½½t; so the ‘‘twisted R-matrix’’ F R ¼ F 2;1 RF 1 ¼
f ðt2;1 Þet =2 f ðt1;2 Þ 1 ¼ et =2 : Assoc0 is preserved because each F is such that
1;2 1;2

F ¼ F 2;1 :) We denote by Assoc0Lie ; AssocLie and PentLie the subsets of all F in Assoc;
Assoc0 and Pent; such that logðFÞAbt3 :

Theorem 7.1. There is exactly one element of PentLie (resp., AssocLie ; Assoc0Lie ) in
c2 Þ on Pent (resp., Assoc; Assoc0 ). The isotropy
each orbit of the action of 1 þ ðT 0
1;2
group of each element of Pent is felt jlAKgC1 þ ðT c2 Þ :
0

7.2. Proof of Theorem 7.1

The arguments are the same in all three cases, so we treat the case of Assoc:
 ARTICLE IN PRESS
390 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

P
Let F belong to Assoc: Set F ¼ 1 þ i40 Fi ; where Fi is the degree i component
of F: Let d be the co-Hochschild differential,

d : Tn -Tnþ1

X
n
x/ ð 1Þiþ1 x1;y;fi;iþ1g;y;nþ1 x2;3;y;nþ1 þ ð 1Þn x1;2;y;n :
i¼1

Then dðF2 Þ ¼ 0; and AltðF2 Þ ¼ 18 ½t1;2 ; t2;3 : Computation shows that this implies
that for some lAK; we have F2 ¼ 18 ½t1;2 ; t2;3  þ ldððt1;2 Þ2 Þ: We construct F A1 þ
ðc
TÞ0 ; such that F FAAssocLie ; as an infinite product F ¼ ?Fn ?F2 ; where Fi A1 þ
ðc
T2 Þ (the index Xi means the part of degree Xi). If we set F2 ¼ 1 þ lðt1;2 Þ2 ; then
Xi
logð FÞAt3 þ ðc
F2
T3 ÞX3 : Assume that we have found F3 ; y; Fn 1 ; such that
F%n 1 %
logð FÞAt3 þ ðcT3 ÞXn ; where F%n 1 ¼ Fn 1 ?F2 : Then jðn 1Þ :¼ logðFn 1 FÞ satisfies

ðjðn 1Þ Þ1;2;34 %ðjðn 1Þ Þ12;3;4 ¼ ðjðn 1Þ Þ2;3;4 %ðjðn 1Þ Þ1;23;4 %ðjðn 1Þ Þ1;2;3 ;

ðn 1Þ
where % is the CBH product in ðc
T3 Þ0 : Let jn be the degree n part of jðn 1Þ :
ðn 1Þ
Then we get dðjn ÞAt4 : We now use the following statement, which will be proved
in the next subsection.

Proposition 7.2. If gAT3 is such that dðgÞAt4 ; then there exists bAT2 ; such that
g þ dðbÞAt3 : If g has degree n; one can choose b of degree n:

ðn 1Þ
It follows that there exists bAT2 of degree n; such that jn dðbÞAt3 : Set
F%n
ðnÞ
Fn ¼ 1 þ b; then j ¼ logð FÞ is such that j Aj ðnÞ ðn 1Þ
dðbÞ þ ðcT3 ÞXnþ1 ; so
ðnÞ c
j At3 þ ðT3 ÞXnþ1 : Moreover, the product F ¼ ?Fn ?F2 is convergent, and F F
then satisfies logðF FÞAbt3 : This proves the existence of F ; such that F FAAssocLie :
Let us now prove the uniqueness of an element of AssocLie ; twist-equivalent to
FAAssoc: This follows from:

Proposition 7.3. Let F0 and F00 be elements of AssocLie ; and let F belong to 1 þ ðc
T2 Þ0 :
F 0 00 lt1;2 00 0
Then F ¼ F if and only if there exists lAK such that F ¼ e and F ¼ F :

Proof of Proposition 7.3. Since t1;2 þ t1;3 þ t2;3 is central in c

T3 ; we have Fl F0 ¼ F0
lt
when Fl ¼ e ; for any lAK: Conversely, let Fi be the degree i part of F : Then for
0
some l0 AK; we have F1 ¼ l0 t: Replacing F by F 0 ¼ FF l0 ; we get F F0 ¼ F00 ; and
0 0
F 1 has valuation X2 ( for the degree in t). Assume that F 1a0 and let n be its
valuation. Let Fn0 be the degree n part of F 0 : Then dðFn0 ÞAt3 : On the other hand,
Fn0 ¼ mðt1;2 Þn ; where mAK f0g: Now dððt1;2 Þn ÞAT3 ¼ Uðt3 Þ has degree pn
for the filtration of Uðt3 Þ; and its symbol in S n ðt3 Þ ¼ grn ðUðt3 ÞÞ is
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 391

Pv 1  n  0 0 P  n  1;2 n00 1;3 n n00

n0 ¼1 ðt1;3 Þn ðt2;3 Þn n n 1
n0 n00 ¼1 n00 ðt Þ ðt Þ : this is the image of a nonzero
element in Sn ðKt1;2 "Kt1;3 "Kt2;3 Þ under the injection S n ð"1piojp3 Kti; j Þ+S n ðt3 Þ;
so it is nonzero. So F 0 a1 leads to a contradiction. So F ¼ Fl0 ; therefore
F00 ¼ F0 : &

Note that we have proved the analogue of Proposition 7.2, where the indices of
T3 ; t4 ; etc., are shifted by 1:

7.3. Decomposition of t3 and proof of Proposition 7.2

To end the proof of the first part of Theorem 7.1, it remains to prove Proposition
7.2. For this, we construct a decomposition of tn : For i ¼ 1; y; n; there is a unique
algebra morphism ei : Tn -Tn 1 ; taking ti; j to 0 for any jai; and taking tj;k to
tj li ð j Þ;k li ðkÞ if j; kai; where li ð j Þ ¼ 0 if joi and ¼ 1 if j4i: Then ei induces a Lie
T
algebra morphism e* i : tn -tn 1 : Set *tn ¼ ni¼1 Kerð*ei Þ: Then we have

Lemma 7.4.
n
M M
tn ¼ ð*tk ÞI ;
k¼0 IAPk ðf1;y;ngÞ

where Pk ðf1; y; ngÞ is the set of subsets of f1; y; ng of cardinal k; and ð*tk ÞI is the
image of t*k under tk -tn ; x/xi1 ;y;ik ; where I ¼ fi1 ; y; ik g:

Proof. Let F be the free Lie algebra with generators t̃i; j ; where 1piojpn: It is
graded by G :¼ Nfði;j Þj1piojpng : the degree of t̃i; j is the vector di; j ; whose ði0 ; j 0 Þ
coordinate is dði;j Þ;ði0 ;j0 Þ : For kAG; we denote by Fk the part of F of degree k: Let
% % %
p : F-tn be the canonical projection. Since the defining ideal of tn is graded, we have
M
tn ¼ pðFk Þ: ð7:14Þ
kAG %
%

On the other hand, one checks that t*n ¼ "kAG* pðFk Þ; where G * is the set of maps
% P% P
k : fði; j Þj1piojpng-N; such that for each i; j j j4i kði; j Þ þ j j joi kð j; iÞa0:
Define a map l : G-Pðf1; y; ngÞ as follows (Pðf1; y; ngÞ is the set P of subsets of
f1; y; ngÞ: l takes the map k : fði; j Þj1piojpng-N to fij j j j4i kði; j Þ þ
P
j j joi kð j; iÞa0g: Then for each IAPðf1; y; ngÞ; ðt*jIj ÞI identifies with
"kAl 1 ðIÞ pðFk Þ: Comparing with (7.14), we get
% %
M
tn ¼ ð*tjIj ÞI : &
IAPðf1;y;ngÞ
 ARTICLE IN PRESS
392 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

When n ¼ 3; we get t3 ¼ Kt1;2 "Kt1;3 "Kt2;3 "*t3 : On the other hand, the fact that
the insertion-coproduct maps take tn to tm implies that d : Tn -Tnþ1 is compatible
with the filtrations induced by the identification Tn ¼ Uðtn Þ; Tnþ1 ¼ Uðtnþ1 Þ: The
associated graded map is

gr ðdÞ : S  ðtn Þ-S  ðtnþ1 Þ:

Proposition 7.2 now follows from:

Lemma 7.5. When kX2; the cohomology of the complex

grk ðdÞ grk ðdÞ

S k ðt2 Þ

! Sk ðt3 Þ

! Sk ðt4 Þ

vanishes.

Proof. We have
!
k
M M
Sk ðt3 Þ ¼ S k a Kti; j #Sa ð*t3 Þ: ð7:15Þ
a¼0 1piojp3

Let xASk ðt3 Þ; and let ðxa Þa¼0;y;k be its components in the decomposition (7.15).
We have
4
S  ðt4 Þ ¼ S  ð*t4 Þ# # 3 Þ# # S ðt2 Þ#S ðt3
S ð*t1;i;j  *1;i  2;3;4
Þ:
2piojp4 i¼2

We denote by p the projection

p : S  ðt4 Þ-*t1;3;4
3 #S  ðt2;3;4
3 Þ;

which is the tensor product of: the identity on the last factor, the projection to degree
1 on the factor S  ð*t1;3;4
3 Þ; and the projection to degree 0 in all other factors. We also
*
denote by m : t3 #S  ðt2;3;4
1;3;4
Þ-S  ðt3 Þ the map induced by the identifications
3
*t Ct Ct3 ; t Ct3 followed by the product map in S  ðt3 Þ: We denote by
1;3;4 1;3;4 2;3;4
3 3 3
d1 ; d2 ; d3 the maps T3 -T4 defined by

d1 ðxÞ ¼ x12;3;4 x1;3;4 x2;3;4 ;

d2 ðxÞ ¼ x1;23;4 x1;2;4 x1;3;4 ;

d3 ðxÞ ¼ x1;2;34 x1;2;3 x1;2;4 ;

so d ¼ d1 d2 þ d3 : The maps di are compatible with the filtrations of T3

and T4 ; we denote by grk ðdi Þ the corresponding graded maps, so
 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 393

grk ðdÞ ¼ grk ðd1 Þ grk ðd2 Þ þ grk ðd3 Þ: Then if we set
X
x1 ¼ ðt1;2 Þa ðt1;3 Þb ðt2;3 Þc #ea;b;c ;
a;b;cjaþbþc¼k 1

where ea;b;c A*t3 ; we have

!
X
k
k
m 3 p 3 gr ðd1 ÞðxÞ ¼ axa ðt2;3 Þk 1 e0;0;k 1 :
a¼0

On the other hand, let us define the i-degree of an element of ð*tjIj ÞI to be 1 if iAI and
P
0 if ieI: Then the i-degree of #ICf1;y;ng SaI ðð*tjIj ÞI ÞCS  ðtn Þ is IjiAI aI : If x is
homogeneous for the 1-degree, then so is grk ðd2 ÞðxÞ; and 1-degreeðgrk ðd2 ÞðxÞÞ ¼
1-degreeðxÞ: On the other hand, the elements of S  ðt4 Þ whose 1-degree is a1 are in
the kernel of p: It follows that

m 3 p 3 grk ðd2 Þðxa Þ ¼ 0 if aa1;

and p 3 grk ðd2 Þðx1 Þ ¼ ðe0;0;k 1 Þ1;3;4 ½ðt2;4 þ t3;4 Þk 1 ðt3;4 Þk 1 ; so

m 3 p 3 grk ðd2 Þðx1 Þ ¼ e0;0;k 1 ½ðt1;3 þ t2;3 Þk 1 ðt2;3 Þk 1 :

Finally, p 3 grk ðd3 ÞðxÞ ¼ 0: If x is such that grk ðdÞðxÞ ¼ 0; we have

m 3 p 3 grk ðdÞðxÞ ¼ 0; so
X
aX0
axa ¼ e0;0;k 1 ðt1;3 þ t2;3 Þk 1 :

Looking at degrees in the decomposition (7.15), we get xa ¼ 0 for aX2; and x1 ¼

e0;0;k 1 ðt1;3 þ t2;3 Þk 1 : Using the projection p0 : S  ðt4 Þ-*t1;2;4
3 #S  ðt1;2;3
3 Þ; we get in the
1;2 1;3 k 1 1;2 1;3 k 1
same way x1 ¼ ek 1;0;0 ðt þ t Þ : Now ek 1;0;0 ðt þ t Þ ¼ e0;0;k 1 ðt1;3 þ
2;3 k 1
t Þ implies ek 1;0;0 ¼ e0;0;k 1 ¼ 0 so x1 ¼ 0: Therefore xAS ð"1piojp3 Kti; j Þ:
k

Let us set x ¼ Sðt1;2 ; t1;3 ; t2;3 Þ; where S is a homogeneous polynomial of degree k of

K½u; v; w: Since dðxÞ ¼ 0; we have

Sðt1;3 þ t2;3 ; t1;4 þ t2;4 ; t3;4 Þ Sðt1;2 þ t1;3 ; t1;4 ; t2;4 þ t3;4 Þ

þ Sðt1;2 ; t1;3 þ t1;4 ; t2;3 þ t2;4 Þ ¼ Sðt2;3 ; t2;4 ; t3;4 Þ þ Sðt1;2 ; t1;3 ; t2;3 Þ

(equality in S  ð"1piojp4 Kti; j Þ).

Applying @t@1;2 3 @t@3;4 to this equality, we get

ð@u @w SÞðt1;2 þ t1;3 ; t1;4 ; t2;4 þ t3;4 Þ ¼ 0;

 ARTICLE IN PRESS
394 B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395

therefore @u @w S ¼ 0: We have therefore

Sðu; v; wÞ ¼ Pðu; vÞ þ Qðv; wÞ;

where P and Q are homogeneous polynomials of degree k: Moreover, dðxÞ ¼ 0; so

½Pðt1;2 ; t1;3 þ t1;4 Þ Pðt1;2 þ t1;3 ; t1;4 Þ Pðt1;2 ; t1;3 Þ

þ ½Qðt1;4 þ t2;4 ; t3;4 Þ Qðt1;4 ; t2;4 þ t3;4 Þ Qðt2;4 ; t3;4 Þ

þ ½Pðt1;3 þ t2;3 ; t1;4 þ t2;4 Þ þ Qðt1;3 þ t1;4 ; t2;3 þ t2;4 Þ

Pðt2;4 ; t2;4 Þ Qðt1;3 ; t2;3 Þ ¼ 0: ð7:16Þ

Write this as an identity

Bðt1;2 ; t1;3 ; t1;4 Þ þ Cðt1;4 ; t2;4 ; t3;4 Þ þ Aðt2;3 ; t1;4 ; t1;3 ; t2;4 Þ ¼ 0:

Then A (resp., B; C) is independent of t2;3 (resp., t1;2 ; t3;4 ). Let us now determine P
and Q: Since Bðt1;2 ; t1;3 ; t1;4 Þ ¼ Bð0; t1;3 ; t1;4 Þ; we have Pðu; v þ wÞ Pðu þ v; wÞ
Pðu; vÞ ¼ Pð0; v þ wÞ Pðv; wÞ Pð0; vÞ: Therefore ðdP̃Þðu; v; wÞ ¼ 0; where
P̃ðu; vÞ ¼ Pðu; vÞ Pð0; vÞ and d is the co-Hochschild differential of polynomials in
one variable. The corresponding cohomology is zero, so there exists a polynomial P; %
such that
% þ vÞ PðuÞ
Pðu; vÞ Pð0; vÞ ¼ Pðu % %
PðvÞ:

We conclude that Pðu; vÞ has the form

% þ vÞ PðuÞ
Pðu; vÞ ¼ Pðu % RðvÞ ð7:17Þ

where P% and R are polynomials in one variable of degree k; since Pðu; vÞ is

homogeneous of degree k; we can assume that P% and R are monomials of degree k:
In the same way, since Cðt1;4 ; t2;4 ; t3;4 Þ ¼ Cðt1;4 ; t2;4 ; 0Þ; we have Qðu þ v; wÞ
Qðu; v þ wÞ Qðv; wÞ ¼ Qðu þ v; 0Þ Qðu; vÞ Qðv; 0Þ; so ðdQ̃Þðu; v; wÞ ¼ 0; where
Q̃ðu; vÞ ¼ Qðu; vÞ Qðu; 0Þ: So Qðu; vÞ has the form
% þ vÞ QðvÞ
Qðu; vÞ ¼ Qðu % SðuÞ; ð7:18Þ

where Q% and S are polynomials in one variable of degree k; which can be assumed to
be monomials of degree k: We have therefore

x ¼ P% 1;23 þ Q% 12;3 P% 1;2 Q% 2;3 T 1;3 ;

where P% ¼ Pðt% 1;2 Þ; Q% ¼ Qðt

% 1;2 Þ and T ¼ ðR þ SÞðt1;2 Þ: So x ¼ dðQÞ
% þ ðP% þ QÞ % 1;23
%
ðP% þ QÞ 1;2
T : Set a ¼ P% þ Q;
1;3 % we have dðyÞ ¼ 0; where y ¼ a 1;23
a T 1;3 ;
1;2

applying e1 to dðyÞ ¼ 0; we get T 2;3 T 2;4 ¼ 0; so T ¼ 0: We then get a12;34

 ARTICLE IN PRESS
B. Enriquez, G. Halbout / Advances in Mathematics 186 (2004) 363–395 395

a12;3 a2;34 þ a2;3 ¼ 0: Applying e3 3 e2 to this identity, we get a1;4 ¼ 0: Finally P% ¼

Q;% so x ¼ dðQÞ;% which proves the lemma. &

7.4. Isotropy groups

Proposition 7.3 can be generalized to the case of a pair of elements of PentLie ; and
it implies that the isotropy group of each element of PentLie is the additive group
1;2
felt ; lAKg: Let F be an element of Pent: There exists an element FLie of PentLie in
the orbit of F: So the isotropy groups of F and FLie are conjugate. Since 1 þ ðc
T2 Þ0 is
lt1;2
commutative, the isotropy group of F is fe ; lAKg:

Acknowledgments

B.E. thanks Y. Kosmann-Schwarzbach for communicating to him [Dr4], where

the question of the construction of quasi-groups is raised, as well as for discussions
on this question in 1995. Both authors would also like to thank P. Etingof,
F. Gavarini, Y. Kosmann-Schwarzbach and P. Xu for discussions.

References

[Ban] M. Bangoura, Quasi-groupes de Lie-Poisson, C.R. Acad. Sci. Paris Sér. I Math. 319 (9) (1994)
975–978.
[BN] D. Bar-Natan, On associators and the Grothendieck-Teichmüller group I, Selecta Math. (N.S.)
4 (2) (1998) 183–212.
[Dr1] V. Drinfeld, Quantum groups, Proceedings of the International Congress of Mathematicians.
Vol. 1, Number 2, American Mathematical Society, Providence, RI, 1987, pp. 798–820.
[Dr2] V. Drinfeld, Quasi-Hopf algebras, Leningrad Math. J. 1 (6) (1990) 1419–1457.
[Dr3] V. Drinfeld, On quasitriangular quasi-Hopf algebras and on a group closely connected with
%
GalðQ=QÞ; Leningrad Math. J. 2 (4) (1991) 829–860.
[Dr4] V. Drinfeld, Letter to Y. Kosmann-Schwarzbach, 1991.
[EGH] B. Enriquez, F. Gavarini, G. Halbout, On the unicity of braidings of quasitriangular Lie
bialgebras, IMRN, to appear.
[EH] B. Enriquez, G. Halbout, An _-adic valuation property of universal R-matrices, J. Algebra 261 (2)
(2003) 434–447.
[Ga] F. Gavarini, The quantum duality principle, Ann. Inst. Fourier 52 (3) (2002) 809–834.
[GH] F. Gavarini, G. Halbout, Tressages des groupes de Poisson formels à dual quasitriangulaire,
J. Pure Appl. Algebra 161 (3) (2001) 295–307.
[KS] Y. Kosmann-Schwarzbach, Poisson–Lie groups and beyond, J. Math. Sci. 82 (6) (1996)
3807–3813.
[WX] A. Weinstein, P. Xu, Classical solutions of the quantum Yang–Baxter equation, Comm. Math.
Phys. 148 (1992) 309–343.