Classication of bicovariant dierential calculi on

quantum groups (a representation-theoretic approach)

Pierre Baumann and Frdric Schmitt
U.F.R. de mathmatiques, Universit Louis Pasteur,
7 rue Ren Descartes, F-67084 Strasbourg Cedex, France
Abstract
The restricted dual of a quantized enveloping algebra can be viewed as the algebra
of functions on a quantum group. According to Woronowicz, there is a general notion
of bicovariant dierential calculus on such an algebra. We give a classication theorem
of these calculi. The proof uses the notion (due to Reshetikhin and Semenov-Tian-
Shansky) of factorizable quasi-triangular Hopf algebra and relies on results of Joseph
and Letzter. On the way, we also give a new formula for Rossos bilinear form.
Introduction
Let G be a semi-simple connected simply-connected complex Lie group, g its Lie algebra,
U
q
g the quantized enveloping algebra of g. U
q
g is a Hopf algebra. The associated quantum
group is an object of non-commutative geometry. According to a point of view due to
Woronowicz and developed by Faddeev, Reshetikhin and Takhtadzhyan [FRT], one may
view the restricted (Hopf) dual (U
q
g)
res
as the function algebra /
q
G on this quantum group.
In this way, the PeterWeyl theorem becomes a denition: the rational representations of
the quantum group are the nite dimensional representations of U
q
g.
In order to study the dierential geometry of quantum groups, Woronowicz [Wo] dened
the notion of bicovariant dierential calculus. As in the classical case, one needs only to dene
the dierential of functions at the unity point of the quantum group. If : /
q
G C(q)
is the augmentation map, this amounts to take the residual class of functions belonging to
ker modulo a right ideal 1 ker . In the classical case, one takes 1 = (ker )
2
. As for
quantum groups, it is more important to preserve the group structure than the innitesimal
structure, and one is led to select ideals 1as above by the requirement of a certain invariance
condition. In this article, we solve the classication problem for these ideals 1, and we give
a picture of what they look like.
We now compare our results with previous ones. Rosso [Ro3] showed how to use the
quasi-triangular structure of U
q
g in order to construct left covariant dierential calculi on
the quantum group. Modifying this construction, Juro [Ju] used the R-matrix in the natural
representation of U
q
g (and in the dual of it) so as to construct bicovariant dierential calculi:
he obtained particular cases (when M is the natural g-module or its dual) of our theorem2.
1
(In this spirit, see also [FP].) As regards classication results, Schmdgen and Schler have
classied the ideals 1 as above, but only when g is of classical type, and under restrictive
assumptions on 1. Most of the results in [SS1, SS2] are particular cases of our theorem 1.
For instance, the classication given in the theorem 2.1 corresponds (in the wording of our
theorem) to the ideals 1 constructed (up to a twisting character :
2X
/
2Q
C

, as
explained in the section 3.3) from the natural U
q
g-module or its dual.
Let us explain our proof and the contents of our article. Our proof relies on the quasi-
triangular structure of U
q
g. Since the formalism of R-matrices may be justied only for nite
dimensional Hopf algebras, we will employ the dual notion of co-quasi-triangular (c.q.t.) Hopf
algebra (see [LT]): the algebra /
q
G is c.q.t.. We use then a bilinear form on /
q
G, intro-
duced by Reshetikhin and Semenov-Tian-Shansky. As U
q
g is a factorizable quasi-triangular
Hopf algebra (in the terminology of [RS]), this pairing is non-degenerate and gives a linear
injection /
q
G U
q
g (/
q
G)

. The image of 1 under this map is nearly the annihilator

of a U
q
g-module. It is then easier to discuss what 1 may be. The denitions and the proofs
of these assertions are given in sections 1 and 2. In section 3, we present a contruction of
bicovariant dierential calculi valid for any factorizable c.q.t. Hopf algebra. Finally we link,
in the case of /
q
G, these constructions with our classication result.
Acknowledgements
Les deux auteurs remercient le Ministre Franais de lEnseignement Suprieur et de la
Recherche pour son soutien nancier (allocations de recherche). They also warmly and
sincerely thank Professor M. Rosso for explaining the subject, giving bibliographic references,
and supplying us with the leading idea.
Notations
Let A be a k-algebra. If M is an A-module, its annihilator is noted ann
A
M. If
m M and m

(the k-dual of M), we denote by

M
(m, m

) the matrix coecient

(A k, a m

, a m)).
For a Hopf algebra H, we will use Sweedlers notation for coproduct ((a) =

a
(1)

a
(2)
) and for coaction on comodules. The sum sign will generally be omitted. We will
denote the augmentation and the antipode of H by and S respectively.
Let H be a Hopf algebra, and H
res
be the restricted (Hopf) dual of H. A nite
dimensional left H-module M (with a basis (m
i
) and the dual basis (m

i
) of M

) can
be viewed as a right H
res
-comodule with structure map
R
: (M M H
res
, m

m
i

M
(m, m

i
) ).
2
1 Co-quasi-triangular Hopf algebras
1.1 Some denitions
Let H be a Hopf algebra over a eld k. A right crossed bimodule over H (in the sense of
Yetter [Ye]) is a k-vector space M, which is also a right H-module, a right H-comodule (with
structure map
R
: (M M H, m

m
(0)
m
(1)
)), both structures being compatible:

R
(m a) =

m
(0)
a
(2)
S(a
(1)
)m
(1)
a
(3)
(for m M, a H). When M and N are
right crossed bimodules over H, M N becomes a right crossed bimodule for the action
(mn) a = m a
(1)
n a
(2)
and the coaction
R
(mn) = (m
(0)
n
(0)
) m
(1)
n
(1)
.
There are two easy examples: we can endow H with the structures: a b = ab and

R
: (H H H, a a
(2)
S(a
(1)
)a
(3)
). Alternatively, we can put on H the structures
a b = S(b
(1)
)ab
(2)
(right adjoint action) and
R
: (H H H, a a
(1)
a
(2)
).
When is a bicovariant bimodule (see [Wo]), the space
L
of left coinvariants is a right
crossed bimodule over H. Conversely, any right crossed bimodule over H is the space of left
coinvariants of a bicovariant bimodule.
Finally (H still being a Hopf algebra), we endow the tensor product coalgebra H
res
H
with the product (f a)(gb) = g
(3)
, a
(3)
)g
(1)
, S(a
(1)
))(g
(2)
f a
(2)
b). We obtain a bialgebra,
called Drinfel

ds double of H and denoted by T(H). (Here H

res
is the standard dual of H,
the coproduct is not brought into its opposite.) When M is a right crossed bimodule over
H, it is a right T(H)-module for the actions: m (f 1) = f, m
(1)
)m
(0)
, m (1 a) = m a.
1.2 Denition of a co-quasi-triangular Hopf algebra
We give the denition of c.q.t. Hopf algebras, by now usual (see [LT] for historical notes):
Denition 1 A co-quasi-triangular Hopf algebra is a pair (/, ) where / is a Hopf algebra
and : / /
res
is a coalgebra morphism and algebra antimorphism such that we have
the Yang-Baxter equation (or rather the Baxter commutation relations): a
(1)
b
(1)
a
(2)
, b
(2)
) =
a
(1)
, b
(1)
)b
(2)
a
(2)
for all a, b /.
That is a coalgebra morphism and an algebra antimorphism gives us that for all a, b /,
a, b) = Sa, Sb). We call : / /

the map such that a, b) = b, Sa), for all a, b /.

Hence we have a, b) = b, Sa). We verify easily that takes its values in /
res
and (/, )
is a c.q.t. Hopf algebra.
If U is a Hopf algebra quasi-triangular for an R-matrix R
12
, then U
res
becomes a c.q.t.
Hopf algebra for the map given by: for a, b U
res
, (a), b) = b a, R
12
), and then
(a), b) = b a, R
1
21
). This follows from Drinfel

ds classical axioms. For instance, let H be

a nite dimensional Hopf algebra, and U = T(H): the dual vector space H H

of U is the
underlying space of the restricted dual of U. If (e
i
) is a basis for H, the canonical R-matrix is

(e

i
1)(1e
i
) UU. It corresponds to the maps : (HH

U, af (a)f 1)
and : (H H

U, b g g(1) S
1
(b)) (the antipode of a nite dimensional Hopf
algebra being invertible).
The category of left modules over a quasi-triangular Hopf algebra is braided. The trans-
lation in the present formalism is the:
3
Proposition 1 Let (/, ) be a c.q.t. Hopf algebra. If M is a right /-comodule, it becomes
a right crossed bimodule over / when endowed with the right module structure given by: for
m M and a /, m a = a, m
(1)
)m
(0)
. This extra structure is compatible with tensor
products of comodules and crossed bimodules.
Proof. Let
R
: (M M/, m m
(0)
m
(1)
) the structure map for M. Then we have:
m
(0)
a
(2)
S(a
(1)
)m
(1)
a
(3)
= m
(0)
a
(2)
, m
(1)
)S(a
(1)
)m
(2)
a
(3)
= m
(0)
S(a
(1)
)a
(2)
m
(1)
a
(3)
, m
(2)
)
= m
(0)
m
(1)
a, m
(2)
)
=
R
(m a).
The compatibility with tensor products is a consequence of being a coalgebra homomor-
phism.
We also note that the antipode of a c.q.t. Hopf algebra is always invertible, the square
of its transpose being an inner automorphism of the algebra /

(see [Dr2]).
Finally, when (/, ) is a c.q.t. Hopf algebra, we have the maps and , and Radford
[Ra] has shown that (im)(im) = (im)(im) is a sub-Hopf-algebra of /
res
. This was
shown in the early [RS]: there is a Hopf algebra structure (with invertible antipode) on the
tensor product coalgebra / / such that the map (/ / /
res
, a b b a) is a
coalgebra morphism and an algebra antimorphism.
Example. In the F.R.T. construction [FRT], one considers matrices L
+
and L

, whose
elements lie in im and im respectively. Then Faddeev, Reshetikhin and Takhtadzhyan
dened U
q
g to be the algebra (im)(im).
1.3 The maps I and J
We x in this subsection a c.q.t. Hopf algebra (/, ) over the eld k, and note the
associated map. We dene two maps I : (/ /
res
, a (a
(1)
) S(a
(2)
)) and J : (/
/
res
, a S(a
(1)
) (a
(2)
)). Equivalently, we may consider the pairing of two elements
a, b /: I(a), b) = J(b), a). (When / is the dual of a quasi-triangular Hopf algebra, this
pairing is a b, R
21
R
12
).) We have I = S J S and J = S I S.
We will now state an important property of the map I. /
res
is a left /
res
/
res
-module
for the law (x y) z = xz S(y). / is a right crossed bimodule over / for the structures:
a b = ab and
R
: (/ / /, a a
(2)
S(a
(1)
)a
(3)
), so / is a right T(/)-module. Let
: (T(/) /
res
/ /
res
/
res
, x b (b
(1)
)x
(1)
(b
(2)
)x
(2)
).
Proposition 2 In the set-up above, is an algebra antimorphism. If T(/) and a /,
then I(a ) = () I(a).
Proof. That is an antimorphism is already in [RS]. Then, as a consequence of the
Yang-Baxter equation, we may write, for x /
res
and a /, that S(a
(1)
)x, a
(2)
) =
4
x
(2)
, a
(1)
)x
(1)
S(a
(2)
)S(x
(3)
). Then we compute, for = x b T(/):
I(a ) = x, S(a
(1)
)a
(3)
) I(a
(2)
b)
= (b
(1)
) x, S(a
(1)
)a
(4)
) (a
(2)
) S(a
(3)
) S(b
(2)
)
= (b
(1)
) x
(1)
, S(a
(1)
)) SS(a
(2)
) x
(2)
, a
(4)
) S(a
(3)
) S(b
(2)
)
= (b
(1)
) x
(2)
, S(a
(2)
)) x
(1)
SS(a
(1)
) S(x
(3)
) x
(5)
, a
(3)
) x
(4)
S(a
(4)
) S(x
(6)
) S(b
(2)
)
= (b
(1)
) x
(2)
, S(a
(2)
)) x
(1)
SS(a
(1)
) x
(3)
, a
(3)
) S(a
(4)
) S(x
(4)
) S(b
(2)
)
= (b
(1)
) x
(1)
(a
(1)
) S(a
(2)
) S(x
(2)
) S(b
(2)
)
= () I(a).

We single out the particular case b = 1:

Proposition 3 We consider / and /
res
as left /
res
-modules for the adjoint action: if
x, y /
res
and a /, xa = x, S(a
(1)
)a
(3)
)a
(2)
and xy = x
(1)
y S(x
(2)
). Then I : / /
res
is a morphism of /
res
-modules.
Finally, we give the denition, originally due to Reshetikhin and Semenov-Tian-Shansky
[RS]:
Denition 2 One says that (/, ) is factorizable if the pairing (/ / k, (a, b)
I(a), b)) is non-degenerate.
Thus (/, ) is factorizable i the maps I and J are injective. It is possible to show that
(/, ) is factorizable i (/, ) is so.
1.4 A related construction
First, let U be a Hopf algebra. It is a left U-module for the adjoint action: xy = x
(1)
y S(x
(2)
).
We let F

(U) be the sum of all nite dimensional U-submodules of U. It is known [JL1] that
F

(U) is a subalgebra of U, a left coideal in U, and a U-submodule for the left adjoint action.
The multiplication in U denes a morphism of left U-modules F

(U) F

(U) F

(U). We
can then do the semi-direct product F

(U) U: we obtain an algebra. U U denoting the

ordinary tensor product algebra, there is an algebra morphism (F

(U) U UU, xy
xy
(1)
y
(2)
). We can make the same constructions on the right: we obtain an algebra F
r
(U). If
the antipode of U is invertible, the algebra morphism (UF
r
(U) UU, xy x
(1)
x
(2)
y)
has the same image as the previous one. Hence this image contains F

(U) F
r
(U) UU.
We take now a c.q.t. Hopf algebra (/, ), with , I and J as in the preceding subsection.
Let U = (im)(im) be the minimal sub-Hopf-algebra of /
res
in which and take their
values. We consider on / and /
res
the /
res
-module structures of proposition 3. By
restriction, / and /
res
are U-modules, and I : / /
res
is a morphism of U-modules. We
can see that I takes its values in U, which is a U-submodule of /
res
. Further, / is the sum
of its nite dimensional U-submodules, hence imI F

(U).
5
Proposition 4 Let (/, ) be a c.q.t. factorizable Hopf algebra, and I be the associated map.
Let U be the sub-Hopf-algebra (im)(im) /
res
. We suppose that imI = F

(U). Then I
induces a bijection between:
the set of right ideals 1 of /, which are subcomodules for the right coaction
R
: (/
//, a a
(2)
S(a
(1)
)a
(3)
).
the set of two-sided ideals 1 of F

(U), which are U-submodules for the adjoint action.

This bijection preserves dimensions, codimensions, and the inclusion ordering in both sets.
Proof. By assumption, I : / F

(U) is a U-module isomorphism. We adopt the notations

of the proposition 2. / is a T(/)-module, and U / is (the underlying space of) a sub-
Hopf-algebra of T(/), so we will view / as a right U/-module: 1 / acts on / by right
multiplication, U
op
1 acts on / by the left adjoint action. The injectivity of I implies that
imJ U separates the points of /: hence the sub-U /-modules of / are the right ideals
which are subcomodules for the right coaction
R
.
On the other hand, we let E be the image of the morphism (F

(U) U UU, xy
xy
(1)
y
(2)
). U is a UU-module, so is an E-module, and F

(U) is a sub-E-module of U. E
contains F

(U) F
r
(U), with S(F
r
(U)) = F

(U). Therefore, the sub-E-modules of F

(U) are
the two-sided ideals 1 which are U-submodules for the adjoint action.
Now the proposition is a consequence of the proposition 2: writing as the composition
(F

(U) /
res
/
res
/
res
, x y xy
(1)
y
(2)
) (/
res
/ F

(U) /
res
, x a
I(a
(1)
) (a
(2)
)x), and using the assumption imI = F

(U), we can see that E is the image of

U / through .
2 The case of the quantum coordinate algebra
2.1 Notations
In this section, we study the preceding constructions in the case where / is the algebra /
q
G
of regular functions on a quantum group.
Let g be a nite dimensional semi-simple split Lie algebra, h a splitting Cartan subalgebra,

1
, . . . ,

a basis for the root system,

1
, . . . ,

h the inverse roots, P h

and Q h

the weight and the root lattices. The choice of an invariant (under Weyl group
action) scalar product ([) allows us to identify h and h

, with
i
= d
i

i
, d
i
=
(
i
|
i
)
2
. We
choose the normalization of ([) so that ( [ ) Z whenever and belong to P. We
denote by half the sum of the positive roots, by P
+
the set of dominant weights, and by
w
0
the longest element in the Weyl group.
We now choose the following version of U
q
g: this is a C(q)-algebra (q is generic) gener-
ated by E
i
, F
i
and K

( P). The relations are the usual ones among which: K

E
i
=
q
(|
i
)
E
i
K

, K

F
i
= q
(|
i
)
F
i
K

, E
i
F
j
F
j
E
i
=
ij
K

i
K

i
q
d
i q
d
i
. The coproduct is given by:
K

= K

, E
i
= E
i
1+K

i
E
i
, F
i
= 1F
i
+F
i
K

i
. We note S the antipode
of U
q
g. If one chooses a dominant weight and a character :
P
/
2Q
C

, one knows
how to construct a simple nite dimensional U
q
g-modules, in which there is a highest weight
6
vector m

such that K

= ( mod 2Q)q
(|)
m

. We note L

() such a U
q
g-module ;
when is the trivial character, we simply write L(), and then L

() = L() L

(0).
The matrix coecients of the representation L() span a linear subspace C() of the
restricted dual of U
q
g, and we let /
q
G =

P
+
C(). This is a Hopf subalgebra of (U
q
g)
res
.
The elements of /
q
G separate the points of U
q
g [JL1], so that there is an inclusion of U
q
g
into the dual of /
q
G, actually into the restricted dual of /
q
G. We note S the antipode of
/
q
G, which is the restriction to /
q
G of the transpose of the antipode of U
q
g.
There is an R-matrix for U
q
g [Dr1, Ta, Ga]. We choose the R-matrix with the structure

(diagonal part)(polynomial in F) (polynomial in E). If a and b belong to /

q
G, the
number R
12
, b a) C(q) is well-dened (thanks to the weight graduation of U
q
g and
of any nite dimensional U
q
g-module), and we can dene , : /
q
G (/
q
G)

such that
R
12
, b a) = (a), b) = (b), S(a)). (/
q
G, ) and (/
q
G, ) are c.q.t. Hopf algebras, im()
and im() are the sub-Hopf-algebras U

U
0
and U
0
U
+
of U
q
g (/
q
G)
res
respectively, and
U
q
g is the sub-Hopf-algebra (im)(im) = (im)(im) of (/
q
G)
res
.
2.2 Factorizability of /
q
G
Let (/
q
G, ) be the c.q.t. algebra presented above, and be the associated map. For all the
section, we endow /
q
G and U
q
g with the left adjoint action of U
q
g, as in the section 1.4:
in particular, the map I : /
q
G F

(U
q
g) is a morphism of left U
q
g-modules. Joseph and
Letzter [JL1, JL2] have studied the structure of F

(U
q
g), and we need the following results:
If P
+
, K
2
generates a nite dimensional U
q
g-submodule of U
q
g, and F

(U
q
g) =

P
+
(U
q
g K
2
).
Each block U
q
g K
2
contains a unique one-dimensional U
q
g-submodule; it denes a
unique (up to scalars) element z

of the center of U
q
g.
F

(U
q
g) (/
q
G)

separates the points of /

q
G.
The next assertion has been stated in [RS]:
Proposition 5 (/
q
G, ) is a factorizable c.q.t. Hopf algebra, and imI = F

(U
q
g).
Proof. Let P
+
, L() the standard U
q
g-module, m

a highest weight vector, m

w
0

a
lowest weight vector, (m
i
) a basis for L() composed of weight vectors, (m

i
) the dual basis.
We have:
The matrix element
L()
(m
w
0

, m

w
0

) is the linear form on U

q
g given by (in the trian-
gular decomposition U
+
U
0
U

of U
q
g): EK

F (E)q
(w
0
|)
(F).
On this element, takes the value K
w
0

and the value K

w
0

.
The image by (respectively ) of the matrix element
L()
(m
i
, m

w
0

) (respectively

L()
(m
w
0

, m

i
)) is zero if i ,= w
0
.
7
So we have:
I(
L()
(m
w
0

, m

w
0

)) = ((
L()
(m
w
0

, m

w
0

))
(1)
) S(((
L()
(m
w
0

, m

w
0

))
(2)
))
=

(
L()
(m
i
, m

w
0

)) S((
L()
(m
w
0

, m

i
)))
= (
L()
(m
w
0

, m

w
0

)) S((
L()
(m
w
0

, m

w
0

)))
= K
2w
0

.
Hence imI is a U
q
g-submodule of F

(U
q
g) which contains all the K
2w
0

( P
+
), so imI =
F

(U
q
g). We now want to show that J is injective. If b ker J, then for all a /
q
G,
I(a), b) = J(b), a) = 0, so b is null when viewed as a linear form on imI = F

(U
q
g). Then
b = 0, because F

(U
q
g) separates the points of /
q
G. Finally, owing to the formula J = SIS
and to the invertibility of S, I is also injective. This concludes the proof of the proposition.

There is another way to present this result. Rosso [Ro1] introduced a bilinear non-
degenerate ad-invariant form on U
q
g, that Caldero [Ca] writes (U
q
gU
q
g C(q
1/2
), (x, y)
(x), S
1
(y))), where : U
q
g (U
q
g)

. Rossos non-degeneracy result is that is injec-

tive; Calderos theorem states that maps F

(U
q
g) onto /
q
G (U
q
g)
res
. The triangular
behaviour of Rossos form gives us that (K
2w
0

) =
L()
(m
w
0

, m

w
0

). The ad-invariance
of Rossos form can be translated for : when we restrict to F

(U
q
g) and /
q
G, is a
morphism of U
q
g-modules for the adjoint structures. Now I : F

(U
q
g) F

(U
q
g) and
I : /
q
G /
q
G are morphisms of U
q
g-modules and x the respective generators K
2w
0

and
L()
(m
w
0

, m

w
0

) of these modules. (The fact that

L()
(m
w
0

, m

w
0

) generates the U
q
g-
submodule C() of /
q
G is equivalent to the fact that m

w
0

m
w
0

generates the U
q
g-module
L()

L().) So we conclude that and I are mutually inverse isomorphisms, and that I is
a bijection between C() and U
q
g K
2w
0

. The analysis also shows the amusing side-result:

Proposition 6 If x F

(U
q
g), y U
q
g, then the Rosso form on (x, y) is given by
I
1
(x), S
1
(y)) where I : (/
q
G F

(U
q
g), a a id
U
q
g
, R
21
R
12
)) is related to the univer-
sal R-matrix and S is the antipode of U
q
g.
Remarks. 1. It is also possible to give an heuristic proof of this result, using the canonical
R-matrix for Drinfel

ds double and using Rossos formula for his form [Ro2].

2. In the preceding discussion, we were lying a bit. Calderos map does not give ex-
actly Rossos bilinear form, but our formula connecting I and Rossos form is correct
as stated. In our notations, Calderos map is the inverse of the map (/
q
G
F

(U
q
g), a (a
(1)
) S(a
(2)
)).
Later, we will need to know the relations between the central elements z

dened above.
To this aim, we recall Drinfel

ds construction of the center of U

q
g [Dr2]. Let P
+
and
t /
q
G be the quantum trace in L(): for x U
q
g, t, x) = Tr
L()
(K
2
x). t is an invariant
element for the adjoint action of U
q
g in /
q
G, so I(t) is central, and belongs to U
q
g K
2w
0

.
We choose the normalization of z
w
0

by letting z
w
0

= I(t). We then have a Mackey-like

theorem (which is implicit in [Dr2] and in the thesis of Caldero, chapter II, 1.2):
8
Proposition 7 Let c

be the fusion coecients for g: L() L()

L(). Then
z

.
Proof. Let P
+
. We compute J(
L()
(m

, m

)) = K
2
(with the help of the formulas
J = S I S and S(
L()
(m

, m

)) =
L(w
0
)
(m

, m

)). Now let P

+
and let t be the
quantum trace in L(). Let be the Harish-Chandra morphism from the center of U
q
g to
U
0
[Ro1]. We want to compute (z
w
0

) on +. (Evaluation on + means the algebra

homomorphism (U
0
C(q), K

q
(|+)
).) The result will be the image of z
w
0

by the
central character of L(). So it is I(t),
L()
(m

, m

)) = J
L()
(m

, m

), t) = K
2
, t) =
Tr
L()
(K
2
K
2
) = Tr
L()
(K
2(+)
). Thus (z
w
0

) equals the sum of K

2
for in the set of
weights (with multiplicities) of L(). We then use the fact that is an injective algebra
homomorphism.
We note R the Grothendieck ring of the category of nite dimensional U
q
g-modules
whose components are modules L(), without any twisting character :
P
/
2Q
C

. Let
Z(U
q
g) the center of U
q
g, and Z[P] the group algebra of P (with standard Z-basis denoted by
(e

)
P
). The map (R /
q
G, [M] Tr
M
(K
2
)) is a ring homomorphism. If a, b /
q
G
are such that I(a) belongs to the center of U
q
g, then I(ab) = I(a) I(b). As a consequence,
the map (R Z(U
q
g), [M] I(Tr
M
(K
2
))) is a ring homomorphism. This shows again
the statement in proposition 7, and we can paraphrase the above proof by saying that the
following diagram is commutative:
R /
q
G Z(U
q
g)
Z[P] U
0
? ?
-
-
-
I
ch
Here ch : R Z[P] is the ring homomorphism which maps a module to its formal character,
and the bottom arrow is the map (Z[P] U
0
, e

K
2
).
2.3 A technical result on the representation ring
We have just introduced a Grothendieck ring R: by the classical results of Lusztig and Rosso,
R is naturally isomorphic to the representation ring of g. The elements [L()] ( P
+
) form
a Z-basis of R and a Q-basis of R
Z
Q.
Proposition 8 Let P
+
. Then the ideal of R
Z
Q generated by the elements [L(+)]
( P
+
) is the whole algebra R
Z
Q.
The proof of this proposition can be skipped without any drawbacks.

Before we give it, we

have to state an elementary lemma:

Indeed the proof presented here is very inecient. P. Polo showed us a much shorter and easier proof,
which we reproduce with his permission as a note added in proof at the end of this article.
9
Lemma Let (
(1)
, . . . ,
(k)
) (C

)
k
be such that their image in (
C
/
Z
)

are all dierent, and

let (P
(1)
, . . . , P
(k)
) (C[X
1
, . . . , X

])
k
. If

i
P
(i)
(n
1
, . . . , n

) exp(2i

j
n
j

(i)
j
) = 0 holds
for all (n
1
, . . . , n

) N

, then the polynomials P

(1)
,. . . , P
(k)
are all equal to zero.
For = 1, this lemma states linear independance of elementary solutions of a linear dierence
equation. The general proof is by induction on .
Proof of proposition 8 In this proof, we are in a classical context and we do not identify
h and h

. Let R h

and R

h be the direct and inverse root systems, (

) the
canonical bijection between R and R

, Q(R

) h the root lattice. P = P(R) h

is
still the weight lattice; we denote by

1
, . . . ,

the set of inverse simple roots, and by

1
, . . . ,

the set of fundamental weights. R

and R dene Q-structures on h and h

,
and we can dene h
R
and h
C
. The Weyl group W operates on h and h

, and the ane Weyl

group W
a
= W Q(R

) operates on h. Let Z[P] be the Z-algebra of the group P, Z[P]

W
be the set of elements which are invariant under Weyl group action, ch : (R

Z[P]
W
) be the
ring isomorphism formal character. Finally, we denote by (w) = 1 the determinant of
an element w of the Weyl group.
For h
C
, let ev

: (Z[P] C) be the ring morphism which sends a basic element e

( P) to exp(2i, )), where exp is the complex exponential. This extends to an algebra
morphism ev

: (C[P] C). If P
+
, let f

be the map (h
C
C, ev

(ch L())).
We rst assert that given any (x
1
, . . . , x

) C

, there exists h
C
such that for all
i 1, . . . , , f

i
() = x
i
. We view C[P] as the coordinate ring of the ane variety (C

,
and we view an element =

i
(
i
C) as the point (e
2i
1
, . . . , e
2i

) (C

. By
the Nullstellensatz, it is sucient to prove that the elements (ch L(
i
) x
i
e
0
) (i = 1, . . . , )
generate a proper ideal in C[P]. This is already true in C[P]
W
by [Bo], ch. VI, 3, Thorme 1.
The case of C[P] is given by a standard trick: let : (C[P] C[P]
W
) be the projection onto
the trivial homogeneous component in C[P] for the action of W; is a morphism of C[P]
W
-
modules, and thus a relation

Q
i
(ch L(
i
) x
i
e
0
) = 1 in C[P] would give a relation

i
(ch L(
i
) x
i
e
0
) = 1 in C[P]
W
, which is impossible.
We now want to prove a formula for the character f

() = ev

(ch L()). We rst remark

that f

is invariant under the action of the ane Weyl group W

a
in h
C
. If the real part
Re() of lies in an open alcove of h
R
, our formula will just be Weyls character formula:
f

() =

wW
(w) exp(2iw, + ))

wW
(w) exp(2iw, ))
.
Writing the denominator as a product over the positive roots:
exp(2i, ))

R,0
(1 exp(2i, ))),
we can see that it is a non-zero complex number. In the general case, we let T = R [
Re(, )) Z: this is a closed symmetric subset of R ([Bo], ch. VI, 1, Dnition 4), thus T
is a root system in the vector space V
1
h

R
that it spans ([Bo], ch. VI, 1, Proposition 23).
The stabilizer of in W
a
is generated by the reections across the ane hyperplanes in
which Re() lies ([Bo], ch. V, 3, Proposition 2), thus W
1
:= w W [ w Q(R

)
is precisely the subgroup generated by reections along

( T), and its restriction

10
to V
1
is the Weyl group of T. Let be half the sum of the inverse positive roots of T:
=
1
2

T,0

. In restriction to V
1
, is the sum of the fundamental weights of the root
system T

of V

1
. Let h be a small real parameter: Re() + h then lies in an open alcove
of h
R
and we can compute (with a small piece of abuse):
f

() = lim
h0
f

( + h)
= lim
h0

wW/W
1

w
1
W
1
(ww
1
) exp(2iw, + )) exp(2ihw
1
, w
1
( + )))

wW/W
1

w
1
W
1
(ww
1
) exp(2iw, )) exp(2ihw
1
, w
1
))
.
In the sums, we x w
W
/
W
1
and compute the sums on w
1
: in the numerator for instance,
we have an alternating sum of exp(2ihw
1
, w
1
( + ))) where w
1
( + ) P(R) has to
be projected on V
1
, as in [Bo], ch. VI, 1, Proposition 28. The formula (valid in the group
algebra of the weight lattice of T

):

w
1
W
1
(w
1
)e
w
1

= e

T,0
(1 e

) then gives:
f

() =

wW/W
1
(w) exp(2iw, + ))

T,0

, w
1
( + ))

wW/W
1
(w) exp(2iw, ))

T,0

, w
1
)
.
As + and are regular, neither of the products occurring here can be zero. (We will see
soon that the denominator cannot be zero.)
We now prove that the ideal of R
Z
C generated by the elements [L( + )] ( P
+
)
is the whole algebra R
Z
C. We consider again [Bo], ch. VI, 3, Thorme 1: this time,
the isomorphism : C[X
1
, . . . , X

] C[P]
W
is given by (X
i
) = ch L(
i
). Composing
with the isomorphism ch : R Z[P]
W
, we can see that R
Z
C is a polynomial algebra
over C. We suppose by the way of contradiction that the elements [L( + )] ( P
+
) all
belong to some maximal ideal of R
Z
C. Then, by the Nullstellensatz, there exists a point
(x
1
, . . . , x

) C

such that for all P

+
,
1
(ch L( + ))(x
1
, . . . , x

) = 0. We can nd
h
C
such that f

i
() = x
i
(i = 1, . . . , ): then f
+
() = 0 for all P
+
. We next use
the formula:
f
+
() (denominator) =

wW/W
1
(w) exp(2iw, + + ))

T,0

, w
1
( + + )),
and write =

n
i

i
, where (n
i
) N

are any integers. The w (w

W
/
W
1
) are all
distinct modulo Q(R

), and the expressions

T,0

, w
1
( + + )) are non-zero
polynomials in (n
1
, . . . , n

) (they never vanish indeed). Then the above lemma states that
the right-hand side cannot vanish for all (n
i
) N

. This proves rstly that the denominator

is not null, and secondly that f
+
P
n
i

i
() cannot vanish for all (n
i
) N

. We have reached
a contradiction.
To go down to the case of R
Z
Q is then easy: we have shown that we can express in
R
Z
C the unity as a nite sum 1 =

x
i
[L(
i
)][L(
i
)], where
i
P
+
,
i
+ P
+
and
x
i
C. As the structure constants of R
Z
C are integer-valued, this system, viewed as
linear equations in (x
i
), has a solution in C, so has a solution in Q.
11
2.4 Classication of some ideals of F

(U
q
g)
In order to achieve our classication of ideals 1 /
q
G in the next section, we must study
the ideals 1 F

(U
q
g) which are stable by the adjoint action of U
q
g. The analysis requires
the use of the subalgebra V of U
q
g generated by F

(U
q
g) and by the elements K
2
( P
+
).
Joseph and Letzter [JL1] have shown that V is the subalgebra generated by the elements
E
i
, F
i
K

i
and K
2
( P). As it is such a big subalgebra of U
q
g, its representation theory
is similar to that of U
q
g. We will describe it in the next subsection, but in the following
proof, we need to know that the annihilator of a nite dimensional V-module is homogeneous
with respect to the Q-graduation of V.
Proposition 9 The following two properties for a subspace 1 F

(U
q
g) are equivalent:
1. 1 is the annihilator in F

(U
q
g) of a nite dimensional V-module;
2. 1 is a nite codimensional two-sided ideal of F

(U
q
g) and a U
q
g-submodule of F

(U
q
g)
for the left adjoint action.
Proof. We rst show that (1) (2). If M is a nite dimensional V-module, its annihilator in
V is a nite codimensional two-sided ideal of V, and is homogeneous w.r.t. the Q-graduation
of V. It is then easy to see that ann
V
M is a U
q
g-submodule of V for the left adjoint action.
The annihilator 1 = (ann
V
M) F

(U
q
g) of M in F

(U
q
g) thus satises the property (2).
Conversely, let 1 F

(U
q
g) satisfying the property (2). We consider the left regular
F

(U
q
g)-module M =
F

(U
q
g)
/
1
. 1 is its annihilator, so it is sucient to show that M
extends to a V-module. We thus want to show that the elements K
2
F

(U
q
g) ( P
+
)
map to invertible operators in End(M).
1. M is a nite dimensional algebra, and is also a left U
q
g-module (for the adjoint action).
The multiplication in M denes a morphism of left U
q
g-modules: MM M. Thus
the Q-graduation of M (dened by the structure of U
q
g-module) is an algebra grading.
2. We x P
+
. We can write M = M
0
M

(as C(q)-vector space) where K

2
acts
nilpotently on M
0
and inversibly on M

(Fittings decomposition). M
0
and M

are
stable by the commutant of K
2
in End(M), so are right ideals of M. If x F

(U
q
g)
is homogeneous w.r.t. the Q-graduation of F

(U
q
g), x commutes (up to a scalar) with
K
2
, so M
0
and M

are stable by left multiplication by x. Thus M

0
and M

are also
left ideals of M.
3. We now show that M
0
and M

are U
q
g-submodules of M.
(a) Let e
1
, . . . , e
k
be the set of central idempotents in M. The elements K

( P)
of U
q
g act on M (by the adjoint action) as algebra automorphisms, so permute the
elements of the set e
1
, . . . , e
k
. Hence for each , there exists an integer n 1
such that K
n
xes each e
i
. Since M is, as a U
q
g-module, a direct sum of modules
L() (without any twisting character ), and since q is generic, we conclude that
e
1
,. . . , e
k
are xed by the adjoint action of the elements K

.
12
(b) Let e be a central idempotent in M. e is of weight zero. We consider the q-
exponential exp
q
(ad E
i
) =

n0
q
d
i
n(n1)/2
ad E
n
i
[n]
i
!
(i 1, . . . , xed).
Then exp
q
(ad E
i
) is a well dened operator in M. The formula (E
n
i
) =

n
k=0

n
k

i
q
d
i
(nk)k
E
nk
i
K
k

i
E
k
i
enables us to see that exp
q
(ad E
i
)(e) is an idempo-
tent which we write e+x. Then 2ex+x
2
= x, x(12e) = x
2
, x = x(12e)
2
= x
3
.
The weights of the Q-homogeneous components of x belong to n
i
[ n 1; so
the weights of the Q-homogeneous components of x
3
belong to n
i
[ n 3,
and the homogeneous component of x of weight
i
is null. We obtain that
(ad E
i
)(e) = 0. Similarly, (ad F
i
)(e) = 0 for all i 1, . . . , .
(c) M
0
and M

are ideals in M generated by central idempotents e

0
and e

respec-
tively. (a) and (b) show that e
0
and e

dene the trivial U

q
g-module. Hence for
x M
0
and u U
q
g, u x = u (xe
0
) = (u
(1)
x)(u
(2)
e
0
) = (u
(1)
x)(u
(2)
)e
0
=
(u x)e
0
M
0
. The same holds for M

.
4. We rst consider the case g = sl
2
. We choose naturally = the fundamental weight,
and write M
0
=
L
0/
1
and M

=
L
/
1
. The points 2 and 3 show that L
0
and L

are two-sided ideals and left U

q
g-submodules of F

(U
q
g). By denition of the Fitting
decomposition, there exists an integer n 0 such that K
2n
L

. Hence for all

integers m n, we have K
2m
L

, and thus z
m
L

. Let n
0
0 be the
smallest integer such that for all m n
0
, z
m
L

. The proposition 7 and the

ClebschGordan theorem show that if n 1, z
(n+1)
+ z
(n1)
= z

z
n
. Thus n
0
has
to be equal to zero. So 1 = z
0
L

, M

= M, and K
2
acts inversibly on M.
5. The general case is solved in the same way. We consider the decomposition of the
point 2 and write M
0
=
L
0/
1
and M

=
L
/
1
. L
0
and L

are two-sided ideals

and left U
q
g-submodules of F

(U
q
g), and there exists an integer n 0 such that
K
2n
L

. If P
+
, then K
2(n+)
L

, and thus z
n+
L

. Let be the
Q-algebra morphism (R
Z
Q Z(U
q
g), [M] I(Tr
M
(K
2
))) considered at the end
of section 2.2. Then
1
(L

) is an ideal of R
Z
Q, which contains all the elements
[L(w
0
n + )] ( P
+
). Thus
1
(L

) = R
Z
Q by the proposition 8, and so
1 = ([L(0)]) L

, M

= M, and K
2
acts inversibly on M.

Remark. This result is a particular case of the proposition 8.4.13 in [Jo]. Accordingly, its
proof is shorter than the one of Josephs theorem, and does not require the knowledge of the
inclusions between Verma modules, nor the use of Gel

fandKirillov dimensions.
2.5 Classication of some right ideals of /
q
G
The notations /
q
G, U
q
g, V have the same meaning as in sections 2.1 and 2.4. The map
I : (/
q
G

F

(U
q
g)) was introduced in section 1.3.
We now specify the structure of the nite dimensional V-modules: they are completely
reducible; each U
q
g-module L

() (with P
+
, :
P
/
2Q
C

) is (by restriction) a simple

V-module; the V-modules L

() and L

() are isomorphic i = and the characters ,

13
restrict to the same character
2P
/
2Q
C

. The simple nite dimensional V-modules

will be denoted by L

() with P
+
and :
2P
/
2Q
C

a character. We nally
remark (see [JL1]) that a simple nite dimensional V-module is still simple as a F

(U
q
g)-
module. Consequently, if (M
i
) is a nite family of non-isomorphic nite dimensional simple
V-modules, the natural ring homomorphism F

(U
q
g)

End M
i
is surjective.
Theorem 1 1. Let 1 be a nite codimensional right ideal of /
q
G, which is a subcomodule
of /
q
G w.r.t. the right coaction
R
: (/
q
G /
q
G/
q
G, a a
(2)
S(a
(1)
)a
(3)
). Then
there exists a nite dimensional V-module M such that 1 = I
1
(ann
F

(U
q
g)
M).
2. If M is a nite dimensional V-module, then I
1
(ann
F

(U
q
g)
M) is a nite codimensional
right ideal of /
q
G, stable by the right coaction
R
.
3. If M and N are nite dimensional V-modules, then I
1
(ann
F

(U
q
g)
M) = I
1
(ann
F

(U
q
g)
N)
i M and N have the same irreducible components.
4. I
1
(ann
F

(U
q
g)
M) is included in the augmentation ideal of /
q
G i M contains the trivial
V-module.
Proof. (1) and (2) are consequences of the propositions 4 and 9. Let M and N be two
nite dimensional V-modules having the same annihilator in F

(U
q
g). Then ann
F

(U
q
g)
M =
ann
F

(U
q
g)
(M N). Let M
1
, . . . , M
k
(respectively M
1
, . . . , M
n
) be the distinct irreducible
components of M (respectively MN). Then we have:
F

(U
q
g)
/
ann
F

(U
q
g)
(M)

k
i=1
End M
i
and:
F

(U
q
g)
/
ann
F

(U
q
g)
(MN)

n
i=1
End M
i
,
and so k = n: all the irreducible components of N appear in M. (3) follows. (4) can be
proved in a similar way, using the fact that the augmentation ideal of /
q
G is the inverse
image by I of the annihilator of the trivial V-module.
3 Dierential calculi on quantum groups
3.1 Woronowiczs denition
Let / be a Hopf algebra, be a bicovariant bimodule and d : / be a linear map. We
say that (, d) is a bicovariant dierential calculus on / if d is a derivation, a morphism of
two-sided comodules and if the image of d generates the left /-module . The dimension of
the space
L
of left coinvariants will be supposed to be nite.
When (, d) is a dierential calculus over /, we note d
L
the map (/
L
, a S(a
(1)
)
d(a
(2)
)). The subspace 1 = ker d
L
ker is a nite codimensional right ideal of /, and
a subcomodule for the right coadjoint coaction
R
: (a a
(2)
S(a
(1)
)a
(3)
). As shown by
Woronowicz, the subspace 1 determines (up to isomorphism) the bicovariant dierential
calculus (, d): we call it the ideal associated to (, d).
14
Geometrically, / must be viewed as the algebra of functions over a group G, is the
space of 1-forms on G,
L
is the space of left-G-invariant 1-forms on G, identied with the
cotangent space at the unity point of G, and d
L
maps a function on G to its dierential at
the unity point.
3.2 A construction of bicovariant dierential calculi
Let / be a c.q.t. Hopf algebra over the eld k, and let , be the associated maps.
We take a nite dimensional right /-comodule M. We note (m
i
) a basis of M, (m

i
)
the dual basis, and R
ij
the elements of / such that
R
(m
i
) =

m
j
R
ji
. Then R
ji
=

R
jk
R
ki
and (R
ji
) =
ji
(Kroneckers symbol). Also, M is a left /

-module, and the

R
ji
(viewed as linear forms on /

) are the matrix coecients

M
(m
i
, m

j
) of this module.
Since (/, ) is c.q.t., M becomes a right crossed bimodule over / for the action m
i
a =

(a), R
ji
)m
j
(proposition 1). M

is a right comodule over / too, for the coaction

R
(m

i
) =

j
S(R
ij
). Using the fact that (/, ) is a c.q.t. Hopf algebra, we may
endow M

with the structure of a right crossed bimodule over / for the action m

i
a =

(a), S(R
ij
))m

j
. Then, by making the tensor product, we obtain that End(M) MM

is a right crossed bimodule.

We denote by the bicovariant bimodule associated to this right crossed bimodule
End(M). As a vector space, is just the tensor product / M M

. On the basic el-

ements, the structure maps are:
b (a m
i
m

j
) = ba m
i
m

j
(a m
i
m

j
) b =

ab
(1)
(b
(2)
), R
ki
)m
k
(b
(3)
), S(R
j
))m

L
(a m
i
m

j
) = a
(1)
a
(2)
m
i
m

R
(a m
i
m

j
) =

a
(1)
m
k
m

a
(2)
R
ki
S(R
j
).
It follows that the canonical element X =

1 m
i
m

i
of is left and right coinvariant.
The linear map d : (/ , a X a a X) is then a derivation and a morphism of
two-sided comodules.
Theorem 2 1. If (/, ) is a factorizable c.q.t. Hopf algebra and if M is a simple nite
dimensional non-trivial /-comodule, then the above construction gives a bicovariant
dierential calculus d : (/ /End(M)).
2. Its associated ideal is 1 = I
1
(ann
A
(k M)), where k is the trivial /

-module.
Proof. We rst compute for a /:
d(a) =

a
(1)
I(a
(2)
), R
k
) m
k
m

a
(1)
a
(2)
,
k
) m
k
m

a
(1)
I(a
(2)
), R
k

k
) m
k
m

and so:
d
L
(a) =

I(a (a)), R
k
)m
k
m

J(R
k

k
), a)m
k
m

.
15
The R
ji
are the matrix coecients
M
(m
i
, m

j
) of the /

-module M, which is irreducible

and non-trivial. Thus, by the Jacobson density theorem, the (dimM)
2
+1 elements 1, R
ji

are linearly independant in /. The (dimM)

2
linear forms J(R
k

k
) are then linearly
independant in /

, and the formula for d

L
(a) shows that d
L
maps / onto
L
= End(M).
(1) is proved. The same formula shows that 1 is the set of elements a in the augmentation
ideal of / such that I(a) is orthogonal to all the matrix coecients R
k
of the /

-module
M. Thus 1 = ker I
1
(ann
A
M) = I
1
(ann
A
(k M)). We have shown (2).
If we consider now a nite family (M
i
) of non-trivial non-isomorphic nite dimensional
simple right /-comodules, we can do the direct sum of such constructions. If (/, ) is
factorizable, then the map d : (/

(/ End M
i
)) is a bicovariant dierential calculus.
The associated ideal is I
1
(ann
A
(k

M
i
)).
3.3 The link with the classication theorem
We are now gathering the pieces of our patchwork. According to the statements in section 3.1,
the theorem 1 yields a complete classication of bicovariant dierential calculi on /
q
G.
Morally, they are all given by the construction described in section 3.2.
Proposition 10 Let U
q
g and /
q
G be the objects dened in section 2.1. If the root and
the weight lattices for g are equal, all the bicovariant dierential calculi on /
q
G can be
constructed by the method described in section 3.2.
Proof. The results in section 2.5 tell us that an ideal 1associated to a bicovariant dierential
calculus on /
q
G is a subspace I
1
(ann
F

(U
q
g)
M), where M is a V-module containing the trivial
V-module. Let M
1
, . . . , M
n
be the distinct non-trivial irreducible components of M. The
assumption on g gives us that the M
i
are modules L(
i
) (without any twisting character),
and so can be considered as non-trivial non-isomorphic simple right /
q
G-comodules. The
construction of section 3.2 for this family of comodules leads to a bicovariant dierential
calculus whose associated ideal is the inverse image by I of the annihilator of the (/
q
G)

-
module C(q)

M
i
. It is 1, and the proposition is proved.
In the remainder of this section, we will discuss what happens when the root and the
weight lattices dier. Up to the end of this article, we consider this case. There exist
non-trivial characters :
2P
/
2Q
C

, and for any weight , we can look at the ideal

1 = I
1
(ann
F

(U
q
g)
(C(q) L

())), and at the associated bicovariant dierential calculus.

It cannot be constructed by the method of the theorem 2, since L

() is not a right /
q
G-
comodule. However, one may notice that the main trick in the construction of section 3.2
consisted in using two dierent R-matrices, namely R
12
and R
1
21
. R
12
was used to endow
the /
q
G-comodule L() with the structure of a right crossed bimodule over /
q
G, and R
1
21
turned the /
q
G-comodule L()

into a right crossed bimodule over /

q
G. The tensor product
of these right crossed bimodules then gave the bicovariant dierential calculus associated to
I
1
(ann
F

(U
q
g)
(C(q) L())). When one uses the small freedom allowed in the choice of the
R-matrix of U
q
g (see [Ga]), one can make similar constructions for the bicovariant dierential
calculi associated with some of the ideals I
1
(ann
F

(U
q
g)
(C(q) L

())). We will not write

16
all the details, but point out that this is the way followed by Schmdgen and Schler for the
construction described in [SS1], theorem 2.2.
As an example, we now describe explicitely the bicovariant dierential calculus associated
with the ideal I
1
(ann
F

(U
q
g)
(C(q) L

(0))). Let (
P
/
Q
)

be the group of characters :

P
/
Q
C

. If is such a character, it extends to a one-dimensional representation

of /
q
G
by letting

(
L()
(m, m

)) = ( mod Q)m

, m), and this gives an inclusion of the group

(
P
/
Q
)

into the center of (/

q
G)
res
. Since (

id)
R
: /
q
G C(q)/
q
G is given by (x

(x)1), we can see that the kernel of

is a one-codimensional two-sided ideal of /
q
G, stable
by the right coaction
R
. If is non-trivial, the ideal 1 = ker ker

denes a bicovariant
dierential calculus on /
q
G. Putting : (
2P
/
2Q
C

, 2 mod 2Q ( mod Q)), we can

check that 1 = I
1
(ann
F

(U
q
g)
(C(q)L

(0))). This construction gives all the one-dimensional

dierential calculi on /
q
G (generalizing the result of [SS1], remark 4 after the theorem 2.2).
Finally, let X be an intermediate lattice between P and Q. The matrix coecients of
the irreducible representations of U
q
g whose highest weight belongs to X span a subalgebra
/
q
G
X
/
q
G. These algebras /
q
G
X
are factorizable c.q.t. Hopf algebras. For instance,
/
q
G
Q
is the algebra of functions on the quantum adjoint group, and /
q
G /
q
G
P
is the
algebra of functions on the quantum simply-connected group. Our arguments in the sec-
tion 2.5 show that the indecomposable bicovariant dierential calculi on /
q
G
X
are classied
by ideals 1 = /
q
G
X
I
1
(ann
F

(U
q
g)
(C(q) L

())), where :
2X
/
2Q
C

is a character
(extended arbitrarily to a character of the group
2P
/
2Q
). Thus the twisted bicovariant
dierential calculi are non-local, their appearance depending of the choice of X. The bico-
variant dierential calculi seem localized at the central elements of G
X
, that is to say, at the
xed points of G
X
under the adjoint action.
References
[Bo] Bourbaki N.: Groupes et algbres de Lie, chapitres 4, 5 et 6. Paris: Masson 1981
[Ca] Caldero P.: Elments ad-nis de certains groupes quantiques. C. R. Acad. Sci.
Paris 316, 327329 (1993)
[Dr1] Drinfel

d V. G.: Quantum groups. In: Proceedings of the International Congress of

Mathematicians Berkeley 1986, pp. 798820. Providence: American Mathematical
Society 1987
[Dr2] Drinfel

d V. G.: On almost cocommutative Hopf algebras. Leningrad Math. J. 1,

321342 (1990)
[FP] Faddeev L. D., Pyatov P. N.: The dierential calculus on quantum linear groups.
In: Dobrushin R. L., Minlos R. A., Shubin M. A., Vershik A. M. (eds.) Contempo-
rary Mathematical Physics (Berezin memorial volume). Amer. Math. Soc. Transl.
series 2 175, pp. 3547. Providence: American Mathematical Society 1996
[FRT] Faddeev L. D., Reshetikhin N. Yu., Takhtadzhyan L. A.: Quantization of Lie groups
and Lie algebras. Leningrad Math. J. 1, 193225 (1990)
17
[Ga] Gaitsgory D.: Existence and uniqueness of the R-matrix in quantum groups. J.
Algebra 176, 653666 (1995)
[Jo] Joseph A.: Quantum groups and their primitive ideals (Ergebnisse der Mathematik
und ihrer Grenzgebiete, 3. Folge, Band 29). Berlin, Heidelberg: Springer-Verlag
1995
[JL1] Joseph A., Letzter G.: Local niteness of the adjoint action for quantized envelop-
ing algebras. J. Algebra 153, 289318 (1992)
[JL2] Joseph A., Letzter G.: Separation of variables for quantized enveloping algebras.
Amer. J. Math 116, 127177 (1994)
[Ju] Juro B.: Dierential calculus on quantized simple Lie groups. Lett. Math. Phys.
22, 177186 (1991)
[LT] Larson R. G., Towber J.: Two dual classes of bialgebras related to the concepts
of quantum group and quantum Lie algebra. Comm. Algebra 19, 32953345
(1991)
[Ra] Radford D. E.: Minimal quasi-triangular Hopf algebras. J. Algebra 157, 285315
(1993)
[RS] Reshetikhin N. Yu., Semenov-Tian-Shansky M. A.: Quantum R-matrices and fac-
torization problems. J. Geom. Phys. 5, 533550 (1988)
[Ro1] Rosso M.: Analogues de la forme de Killing et du thorme dHarish-Chandra pour
les groupes quantiques. Ann. Sci. Ecole Norm. Sup. 23, 445467 (1990)
[Ro2] Rosso M.: Certaines formes bilinaires sur les groupes quantiques et une conjecture
de Schechtman et Varchenko. C. R. Acad. Sci. Paris 314, 58 (1992)
[Ro3] Rosso M.: Algbres enveloppantes quanties, groupes quantiques compacts de
matrices et calcul direntiel non commutatif. Duke Math. J. 61, 1140 (1990)
[SS1] Schmdgen K., Schler A.: Classication of bicovariant dierential calculi on quan-
tum groups of type A, B, C and D. Commun. Math. Phys. 167, 635670 (1995)
[SS2] Schmdgen K., Schler A.: Classication of bicovariant dierential calculi on quan-
tum groups. Commun. Math. Phys. 170, 315335 (1995)
[Ta] Tanisaki T.: Killing forms, Harish-Chandra isomorphisms, and universal R-
matrices for quantum algebras. In: Innite analysis part B. Proceedings, Kyoto
1991, pp. 941961. Singapore: World Scientic Publishing 1992
[Wo] Woronowicz S. L.: Dierential calculus on compact matrix pseudogroups (quantum
groups). Commun. Math. Phys. 122, 125170 (1989)
[Ye] Yetter D. N.: Quantum groups and representations of monoidal categories. Math.
Proc. Camb. Phil. Soc. 108, 261290 (1990)
18
Note added in proof. P. Polo kindly communicated the following simple proof of Propo-
sition 8. By the formal character isomorphism, ( Z[P]. Let Z[P]
W
Z[P] the subring of
W-invariant elements. Z[P] is a module of nite type over the noetherian ring Z[P]
W
, hence
one can choose a nite generating set (e

i
)
1in
from the family (e

)
P
. Take a weight
such that all +
i
are dominant. Let P
+
. Then there exist some a
i
Z[P]
W
such that
e

i
a
i
e

i
, hence 1 =

i
a
i
e
++
i
. Multiplying this by e

and making the alternating

sum over the Weyl group, one obtains that
ch L(0) =

i
a
i
ch L( + +
i
).
This concludes the proof.
Thanks are also due to A. Joseph for some useful comments about this work.
19