Available at: http://www.ictp.

it/~puboff IC/2005/089
United Nations Educational Scientic and Cultural Organization
and
International Atomic Energy Agency
THE ABDUS SALAM INTERNATIONAL CENTRE FOR THEORETICAL PHYSICS
MORE ON DIFFERENTIAL CALCULI
ON BICROSSPRODUCTS
F. Ngakeu
Department of Mathematics and Computer Sciences, University of Douala,
Douala, Cameroon
and
The Abdus Salam International Centre for Theoretical Physics, Trieste, Italy.
Abstract
We extend a previous classication of dierentials and Cartan calculus on the bicrossproduct
quantum group k(M)kG to its dual Hopf algebra H = kMk(G). It turns out that the
usual bicovariant dierential calculi on kM and on k(G) extend naturally to H. We explicitly
work out the examples of kZ
2
k(Z
3
) and kZ
6
<k(S
3
).
MIRAMARE TRIESTE
September 2005
1. Introduction
There has been a lot of interest during the last two decades in noncommutative geometry.
Many approaches to this theory have been so far proposed, (see for instance, references [1, 2, 3]).
Central to all approaches is the notion of dierential calculus, which is of course the starting
point of any dierential geometry. At this level, the rst problem to address, in general, is that
of the classication of the (rst order) dierential calculi. This is still an open problem, even
for general Hopf algebras. However, in [4], the problem has been solved at least for bicovariant
dierential calculi on Hopf algebras with duals. The treatment for nite sets and nite groups
is completely solved via dierent methods and the results have been extensively exploited to
compute other geometric objects [5, 6, 7, 8, 9, 10].
Unlike the nite groups case, the geometry of nite bicrossproduct Hopf algebras is not yet
extensively developed, though it should be the rst less complicated and more computable case
after nite groups. One of the reasons may be the lack of explicit formulae for Cartan calculus.
The rst key results on dierential calculi on bicrossproducts appeared in [11, 12]. But this was
not constructive enough to allow more computations. In the recent paper [13] we give explicit
formulae for the Cartan calculus of bicovariant dierentials on bicrossproducts A = k(M)kG
associated to groups factorization X = GM and a eld k. The natural question from this was
the following: can one deduce the calculi on the dual Hopf algebra using the same mutual
actions of G and M leading to k(M)kG? The main purpose of this paper is to answer to
this question. Using similar methods as in [13], we show that analogous results for A obtained
there, hold also on H := A

. In fact we do not have just a deduction but a systematic and

independent treatment of dierentials on A

.
In reference [13], the classication of bicovariant dierentials on A = k(M)kG was refor-
mulated more explicitly as decomposition into conjugacy classes of the set
Z = v
1
t
1
v X, for v G, t M.
The result was that any irreducible bicovariant dierential calculus and its Cartan calculus on
A are described by a couple (C, /) where C Z is a conjugacy class in X and / kX is an
irreducible subrepresentation under a suitable action of the quantum double D(X).
For completeness, we now discuss the case of the dual Hopf algebra H = kMk(G), in relation
to results obtained in [13]. The outlines of this paper are as follows: Section 2 contains the
main results of classication of dierentials and Cartan calculus on H: Using the same actions
induced by the factorization X = GM, we nd that dierentials on H are completely described
by couples (C

, /

) where C

is one of the conjugacy classes in X which compose the set

Z = t
1
vt X, for v G, t M
2
and /

kX is an irreducible subrepresentation under a suitable action of the double D(k(X)).

The Cartan calculus given in this section 2 shows that all bicovariant dierential calculi (C

, /

)
are inner. Section 3 is devoted to the case of crossproduct algebras, i.e. when the right action
of G on M (say) is trivial. This includes the quantum double D(G) case. The main result here
is that all usual bicovariant dierential calculi on k(G) and on kM extend to bicovariant calculi
on H. Finally in Section 4, we demonstrate our theory on some explicit examples.
As preliminaries, we mention the following: We use the notations and conventions from [13].
Let X = GM be a nite group factorization and k a eld of characteristic zero. With the group
X are associated the bicrossproducts Hopf algebras H = kMk(G) and A H

= k(M)kG
generated respectively by the basis (t
v
) and (
t
v), for all t M, v G. The explicit
relations and main properties of X, H and H

are in [11, 12, 13] and we freely use them in

the sequel. In particular, if we denote by and the mutual actions of G and M dened by
su = (su)(su) GM, u G, s M as in [13], then the relations of H = kMk(G) are
(1) (s
u
)(t
v
) =
u,tv
(st
v
), (s
u
) =

xy=u
s
x
sx
y
(2) 1 =

u
e
u
, (s
u
) =
u,e
, S(s
u
) = (su)
1

(su)
1
Further general notions on bicrossproducts are in [14, 15]. See also [16, 17] and references
therein for denitions and some details about general quantum groups.
As mentioned in [4], the rst step in classifying bicovariant dierentials on H is to describe the
D(A)-modules and in particular the canonical or Schroedinger D(A)-representation on A
+
=
ker , namely the so-called quantum tangent spaces in A. It is exactly what we do in the next
section.
2. Classification of differentials on kMk(G)
In this Section we formulate, in terms of the set

Z, the classication of the irreducible sub-
representations L A
+
(quantum tangent spaces) under the canonical action of D(A) on A
+
.
Our result here completes the classication results on both forms of bicrossproducts H and H

associated with a group factorization. The technics are sometime the same as in [13]. Whenever
it is the case, some details of the calculus are omitted.
2.1. The D(k(M)kG)-representations.
Let us recall that a left D(A)-module means a compatible left module of A and right module of
H = A

(or left module of A

op
). This is true at least for any nite dimensional Hopf algebra.
Lemma 2.1. Let H be a nite dimensional Hopf algebra with invertible antipode. Then the
map : D(H

) D(H)
op
dened by (ha) = a h, h H, a H

is an Hopf algebras
isomorphism.
3
Proof. For the double we use the form D(A) = A
op
A built on A

A in the double cross

product form [15, 17]. Namely with product
(ha)(g b) =

g
(2)
ha
(2)
b < Sa
(1)
, g
(1)
>< a
(3)
, g
(3)
> (3)
and the tensor product unit, counit and coproduct. Its antipode is
S(ha) = (1 Sa)(S
1
h1) = S
1
h
(2)
Sa
(2)
< a
(1)
, h
(1)
>< Sa
(3)
, h
(3)
>
h, g H, and a, b A.
That is bijective is obvious. If we denote by .
op
the opposite product then we have
((ha)(g b)) =

a
(2)
b g
(2)
h < Sa
(1)
, g
(1)
>< a
(3)
, g
(3)
>
=

a
(2)
b g
(2)
h < a
(1)
, Sg
(1)
>< a
(3)
, g
(3)
>
(ha).
op
(g b) = (a h).
op
(b g) = (b g).(a h)
=

a
(2)
b g
(2)
h < Sa
(1)
, g
(1)
>< a
(3)
, g
(3)
>
=

a
(2)
b g
(2)
h < a
(1)
, Sg
(1)
>< a
(3)
, g
(3)
>
= ((ha)(g b))
Next
( )(ha) = ( )(

h
(1)
a
(1)
h
(2)
a
(2)
) =

a
(1)
h
(1)
a
(2)
h
(2)
((ha)) = (hh) =

a
(1)
h
(1)
a
(2)
h
(2)
= ( )(ha)
Finally, since the antipode S
op
of D(H)
op
is the inverse of that of D(H), we have
S
op
((ha)) = S
op
(a h) = S
op
(h.
op
a) = S
op
(a).
op
S
op
(h) = (Sa 1).
op
(1 S
1
h)
S(ha) = ((1 Sa)(S
1
h1)) = (1 Sa).
op
(S
1
h1)
= (Sa 1).
op
(1 S
1
h) = S
op
((ha))
as expected.
This Lemma 2.1 tells us that left modules of D(H

) are right D(H)-modules, which were

classied in [13] in terms of bigraded G -M-modules. It follows from this remark that:
Proposition 2.2. [13, Prop. 3.1] The left modules of D(k(M)kG) are in one-one correspon-
dence with vector spaces W which are:
(i) G-graded right M-module such that [wt[ = t
1
[w[, for all t M, where [ [ denotes the
G-degree of a homogeneous element w W.
(ii) M-graded left G-module such that uw) = w)u
1
, for all u G, where ) denotes
the M-degree of a homogeneous element w W.
4
(iii) Bigraded by G, M together and mutually cross modules according to
wt) = t
1
w)(t
1
[w[)
1
, [uw[ = (w)u
1
)
1
[w[u
1
(iv) GM-bimodules according to
_
(t
1
[w[u)u
1
_
(wt) = (u
1
w)
_
t(t
1
w)u)
_
The corresponding action of the quantum double D(k(M)kG) is given by
(
s
u)w =
s,wu
1uw, w(t
v
) =
t
1
|w|,v
wt
u, v G and s, t M.
Next we show that D(A)-representations are D(k(X)) ones as follows: it was shown in [12]that
D(H) is the twisting of D(X), from which we deduce by suitable composition of maps that D(A)
is isomorphic to D(k(X)) as algebra. The isomorphism

: D(A) D(k(X)) being given by

(t
v

s
u) = u
1
(tv)
u
1
s
1
(tv)u
(4)
On the other hand, as for D(X) in [11, 13], The D(k(X)) representations are X-graded left
X
op
-modules W by the action and graduation [[.[[ obeying
[[xw[[ = x
1
[[w[[x
x X
op
for an homogeneous w W. Here the product in the right-hand side is that of X.
The action of D(k(X)) is given on such w W by
(x
y
)w =
y,||w||
xw (5)
the correspondence with the gradings and the actions in Proposition 2.2 is
[[w[[ =< w >
1
[w[, usw = (u
1
w)(s
1
[u
1
w[
1
)
1
(6)
We Compute explicitly the relations of D(A) and the Schroedinger action of D(A) on A,
namely
ab =

a
(1)
bSa
(2)
, bh =

< b
(1)
, h > b
(2)
a, b A, h A

, and we compare the results with the actions in Proposition 2.2. Then we
nd that the gradings, the G-M actions for the canonical representation of D(k(M)kG) on
W = k(M)kG are
[
t
v[ = tv, <
t
v >= t(tv)
1
(7)
(
t
v)s =
s
1
t
v, u(
t
v) =
tu
u

vu

1
(8)
where u

= (tv)
1
u
1
. The action in (6) is viewed as that of D(k(X)) on kX as follows:
Lemma 2.3. Let
vt
=
t
1
v
1 v,
vt
is a basis of A = k(M)kG labelled by vt X.
The action of D(k(X)) on A from (6) is

us

vt
=
us,t
1
vt

vt
, us
vt
=
(tu)
1
vtus
(9)
5
so that if we identify the vector spaces A with kX via
vt
vt then this action becomes that of
D(k(X)) on kX given by
[[vt[[ = t
1
vt, usvt = (tu)
1
vtus, us X
op
(10)
Proof. This is direct computation using (7) and (8). We just add the following remark: The
action in (10) can be viewed as right action of X (no longer X
op
) on kX and denoted similarly
vtus = (tu)
1
vtus, us X (11)
So, while manipulating (10) one must remind that
us(u

vt) = u

usvt
with the product of X in the right-hand side. This is equivalent in (11) to
(vtu

)us = vtu

us, u

, us X.
We can now express the projected canonical action of D(A) on A
+
= Ker A. By denition
it is given by
ab =

a
(1)
bSa
(2)
, bh =

< b
(1)
, h > b
(2)
< b, h > 1
b A
+
, a A, h A

. We denote by the corresponding action of D(k(X)) on A+. Then we

have the following
Lemma 2.4. The map : kX A
+
dened by
(vt) =
vt
(
vt
)1
is an intertwiner between the action of D(k(X)) on kX and the action , i.e.
(usvt) = us(vt), (
us
vt) =
us
(vt)
Proof.
us(vt) = us(
vt
(
vt
)1)
= us
vt
(us
vt
)1 (
vt
)(us1) + (
vt
)(us1)1
= us
vt
(us
vt
)1, (as us 1 = 1)
=
(tu)
1
vtus
(
(tu)
1
vtus
)1 = ((tu)
1
vtus) = (usvt)
Similarly for (
us
vt) =
us
(vt).
6
2.2. The quantum tangent spaces in k(M)kG.
Lemma 2.4 tells us that Subrepresentations of A
+
under the projected canonical action of D(A)
( or D(k(X))) above are obtained by projecting via the subrepresentations of kX under the
action in (10). Hence our classication of the quanum tangent spaces reduces to nd irreducible
subrepresentations of kX under the action (10). Here we use symbols with hat to compare
similar but completely dierent things in [13].
Let

Z be the set

Z = image( ), : X X, (vt) = [[vt[[ = t

1
vt
It is obvious that

Z is invariant under conjugation, and it is a partition into conjugacy classes
in X of its elements as follows

Z =
_
(
G
_
tM
(t
1
(
G
t)
where each C
X
=:

tM
(t
1
(
G
t) is a conjugacy class in X for a xed conjugacy class (
G
in G.
We note that for the semidirect case X = G>M, we have

Z G since t
1
vt = (t
1
v).e.
For each z

Z we set

J
z
= k
1
(z), the k-vector space generated by
1
(z) and denote
respectively by C
z
,G
z
the conjugacy class and the centralizer of z in X. Using the similar
technics as for the decomposition of kX into irreducibles under the action of D(X) in [11, 13]
we obtain the following result:
Lemma 2.5. [13, Lem. 3.1, 3.2 ]
(i) The space

J
z
is a G
z
-representation
(ii) M
Cz
=

yCz

J
y
kX is a subrepresentation under the action of D(k(X)) in (10) and
kX =

Cz
M
Cz
is the decomposition of kX into such subrepresentations.
(iii) The space M
Cz
decomposes into irreducibles as follows:
We let

J
z
=

J
1


J
2
...

J
n
be the decomposition into irreducibles under the action of G
z
from (10) and for each y (
z
we x y such that y = y
1
z y (
z
, then
M
i
=
yCz
( y

J
i
) M
Cz
, 1 i n
are irreducible subrepresentations under the action of D(k(X)). Moreover, M
Cz
=
i
M
i
is the
decomposition of M
Cz
into irreducibles.
Proof. (i) is immediate and for the (ii) and (iii), the essential steps are:
(a) The projection

Cz
=

yCz
(
y
)
of kX onto M
Cz
commutes with the action D(k(X)).
(b) The space y

J
i
does not depend on the choice of y.
7
(c) Let p
i
:

J
z


J
i
be the projection onto

J
i
with

J
l
Ker( p
i
) for l ,= i and

Q
i
=

yCz
( y) p
i
( y
1

y
)
Then

Q
i
denes a projection of M
Cz
onto M
i
and

Q
i
commutes with the action D(k(X)).
Those steps are achieved with litle care by the similar reasoning as in [11, 13].

One checks easily that the converse of the preceding Lemma is true, namely
Lemma 2.6. Let / kX be an irreducible representation under the action of D(k(X)) in
(10). Then as vector space, / is of the form
/=

yCz
( y/
0
)
for some conjugacy class C
z


Z and some irreducible subrepresentation /
0


J
z
under the
action of G
z
.
Proof. In fact we choose z such that /
0
:=
z
/ is nonzero. Then that /
0
is a G
z
-
representation is immediate and moreover /
0
is irreducible since by irreducibility of / as
D(k(X))-module and by Lemma 2.5 it cannot contain two distinct G
z
-modules.
We can now state the total description of the irreducible quantum tangent spaces in A =
k(M)kG as follows:
Theorem 2.7. The irreducible quantum tangent spaces L A
+
are all given by the following
cases.
(a) For a conjugacy class C
z
0
,= e and for each irreducible subrepresentation /
0

J
z
0
under the action of the centralizer G
z
0
of z
0
, we have an irreducible D(A)-module / =

zCz
0
( z/
0
) and an isomorphic irreducible quantum tangent space L = (/) A
+
.
(b) For z
0
= e we have

J
e
= kM, G
e
= X and for any irreducible subrepresentation
/
0
,= k.

sM
s under the action of X
op
(hence that of D(k(X))), we have the isomorphic
irreducible quantum tangent space L = (/
0
) A
+
.
Proof. If / =

zCz
0
( z/
0
) is an irreducible D(A)-representation of the unprojected action
and L = (/), then by the equivariance of from Lemma 2.4, the surjective map : /L
is a map of D(A)-representations: if it is one to one, then / and L are isomorphic. The unique
case where is not one to one is the case where / =

sM
s i.e. (/) = 0 which is the case
excluded.
Conversely if L A
+
is an irreducible D(k(X))-module then
1
(L) kX is a D(k(X))-
module containing

sM
s. Moreover if
1
(L) contains at least one other irreducible subrepresen-
tation /such that k.(

sM
s) /
1
(L) then /must be of the form /=

zCz
0
( z/
0
)
and by irreducibility of L, we have (/) = L
8
This ends the classication part of the irreducible quantum tangent spaces in A or irreducible
bicovariant dierential calculi in H. We recover immediately the well known cases of H = kM,
H = k(G) and H = kMk(G) [13] namely
Proposition 2.8. (i) The classication of irreducible bicovariant dierential calculi on kM
is given by nontrivial irreducible subrepresentations V kM under the right regular action
of M on itself. As quantum tangent space, V is isomorphic to (V ) k(M) where (t) =

t
1
t,e
1, t M.
(ii) The classication of irreducible bicovariant dierential calculi on k(G) is given by non-
trivial conjugacy classes C in G.
(iii) If X = GM then H = kMk(G). An irreducible bicovariant dierential calculus on
the tensor product algebra kMk(G) is the product of the calculi in (i) and in (ii). Precisely
the corresponding quantum tangent space is L = (C.V ) where C and V are as above.
3. Cartan calculus on kMk(G)
In this Section we give the explicit description of the dierential forms and the commutation
relations between functions and forms, from which we deduce that all the irreducible bico-
variant dierential calculi on H = kMk(G) are inner as it was the case for H

in [13]. This
description opens the way for jauge theory [18] and noncommutative Riemannian geometry [5]
of bicrossproducts as for the algebra of functions on groups in [10, 9].
We x a conjugacy class C of z
0


Z, for each z C we x z such that z = z
1
z
0
z and x
an irreducible subrepresentation /=

zCz
0
( z/
0
) in Theorem 2.7. Then we choose a basis
(f
i
)
iI
of /
0
and set
f
iz
:= zf
i
To check that the vectors (f
iz
) form a basis of / with z as X-degree of f
iz
is straightforward.
We identify / with L as isomorphic vector spaces via . Then the dual space
1
of L is
equipped with the dual basis (e
iz
) of the basis (f
iz
) . We recall that the actions and on L
refer respectively to the action of D(A) and that of D(k(X)).
Lemma 3.1. Let <, > denote the dual pairing between L and
1
. Then
(i) < e
iz
, tf
jz
>,= 0 [f
jz
[ = (t[f
iz
[
1
)
1
(ii) < e
iz
, u
1
f
jz
>,= 0 < f
jz
>=< f
iz
> u
Proof. (i): From the hypothesis we can decompose f
jz
as
f
jz
=

lz

=iz

lz
(t
1
f
lz
) +
iz
(t
1
f
iz
)
where
iz
,= 0 and
lz
are some scalars. Since f
jz
is X-homogeneous, we have [[f
jz
[[ =
[[t
1
f
iz
[[. Equivalently
< f
jz
>
1
[f
jz
[ = t < f
iz
>
1
[f
iz
[t
1
9
taking the group inverse leads to
[f
jz
[
1
< f
jz
>= t[f
iz
[
1
< f
iz
> t
1
= (t[f
iz
[
1
).(t[f
iz
[
1
) < f
iz
> t
1
and the G-part of this is [f
jz
[
1
= (t[f
iz
[
1
) as expected.
(ii): is proved in the same manner.
Theorem 3.2. If we relabel the previous basis (f
iz
) simply as (f
a
)
aI
and (e
a
)
aI
the dual basis
of
1
(I is the set of indexes iz), then the dierential calculus on H = kMk(G) in Theorem 2.7
is explicitly dened by
(i) The right H-module of 1-forms

1
(H) =
1
H
(ii) The commutation relations between functions and forms are given by

u
.e
a
= e
a

|fa|
1
u
, s.e
a
= (s[f
a
[) e
a
s[f
a
[, s M, u G
where
us e
a
=

bI
< e
a
, usf
b
> e
b
, us X
op
(iii) The exterior dierentials are
d(
u
) =

aI
<
|fa|
, f
a
> e
a
(
|fa|
1
u

u
)
ds =

aI
<
|fa|
, f
a
> [(s[f
a
[) e
a
s[f
a
[ e
a
s]
where <
|fa|
, f
a
> is the pairing between k(X) and kX.
Proof. We rst of all note the following fact that we freely use in the proof and which is easily
obtained from the grading factorization [[w[[ = w)
1
[w[ and [[xw[[ = x
1
[[w[[x:
(12) [tf
a
[ = (t
1
[f
a
[
1
)
1
Then we build bicovariant calculi (
1
(H), d) on H following the usual prescription from [2]
as follows:

1
(H) =
1
H, he
a
= (h
(1)
e
a
) h
(2)
,
dh = (

1 id)(h
(1)
h
(2)
1 h), h H (13)
where

1 : H
+

1
is the projection adjoint to the injection
i : L A
+
, Thus:

u
e
a
=

xy=u
(
x
e
a
)
y
=

xG
(
x
e
a
)
x
1
u
,
10
using the isomorphism

from (4) we compute

x
e
a
=

b1
< e
b
,

(
x
)f
b
> e
b
=

sM

b1
< e
a
,
s
1
x
f
b
> e
b
=
x,|fa|
e
a
hence

u
e
a
=

xG

x,|fa|
e
a

x
1
u
= e
a
.
|fa|
1
u
.
Next let s M. In H, (s) =

vG
(s
v
) (sv 1) so that
s.e
a
=

vG
(s
v
)e
a
(sv)
On the other hand, if we change the basis (f
a
) to f

a
:= (sv)
1
f
a
then the dual basis changes
as e

a
= (sv) e
a
. Indeed for t M, setting f

a
:= t
1
f
a
and e

a
= t e
a
, the observation
e
a
= t
1
t e
a
leads to
< f
b
, e
a
>=

cI
< f
b
, < t e
a
, t
1
f
c
> e
c
>=

cI
< f
b
, e
c
>< t e
a
, t
1
f
c
>=< e

a
, f

b
> .
Using the basis (f

a
) and (12), we have
(s
v
)e
a
=

b1
< e
a
,

(s
v
)f

b
> e

b
=

b1,tM
< e
a
, (sv
t
1
(sv)
)f

b
> e

b
=

b1

sv,|f

b
|
< e
a
, (sv)f

b
> e

b
=
v,|fa|
(sv) e
a
(14)
hence
s.e
a
= (s[f
a
[) e
a
s[f
a
[
as stated.
For the exterior dierentials, let us mention that the map

1 : H
+

1
, as adjoint to the
injection i : L A
+
is in turn, via the identication /

= L adjoint to

: / kX A
+
with respect to the pairing between L and
1
i.e.

1( a) =

bI
< a,

(f
b
) > e
b
=

bI
<

( a), f
b
> e
b
where

: A

k(X)
is adjoint to

, a = a (a)1. A simple computation shows that

(
v

v,e
1) =
v.e

v,e

uG

u.e
, v G.
Using this last equation, we compute for u G
11
d(
u
) =

vy=u

1(
v

v,e
1)
y
=

vG,bI
<

(
v

v,e
1), f
b
> e
b

v
1
u
=

bI,vG
<
v.e
, f
b
> e
b

v
1
u

bI,vG
<
v.e
, f
b
> e
b

u
On the other hand, decomposing f
a
as f
a
=

vtX

vt
a
vt for some scalars
vt
a
, one nds that
<
v.e
, f
a
>=
v,|fa|
<
|fa|
, f
a
>, v G, a I
so that
d(
u
) =

bI
<
|f
b
|
, f
b
> e
b

|f
b
|
1
u

bI
<
|f
b
|
, f
b
> e
b

u
as expected. One can also write this as
df =

aI
c
a
e
a

a
(f), c
a
=<
|fa|
, f
a
>
where
a
= L
a
id, L
a
(f)(u) := f([f
a
[u), a I, u G.
Finally let s M H, we have in H
(s) =

vG
s
v
sv 1
hence
d(s) =

vG

1(s
v

v,e
1) sv
=

bI,vG
<

(s
v

v,e
1), f

b
> e

b
sv
where for any xed v G we set f

b
= (sv)
1
f
b
and then e

b
= (sv) e
b
. It is easy to compute

(s
v

v,e
1) =
v.(sv)
1
v,e

uG

u.e
so that
d(s) =

vG,bI
<
v.(sv)
1 , f

b
> e

b
sv

vG,bI
<
v.e
, f

b
> e

b
s.
It is obvious to see that

vG,bI
<
v.e
, f

b
> e

b
s =

vG,bI
<
v.e
, f
b
> e
b
s
=

bI
<
|f
b
|
, f
b
> e
b
s.
On the other hand we have
<
us
, svt >=
u,v

t,e
, us, vt X.
12
Using this last equality and again the decomposition f
a
=

vtX

vt
a
vt we show the following fact:
<
vt
, tf
a
>,= 0 =<
vt
, tf
a
>=
v,|fa|
<
v
, f
a
>, v G, t M
so that

vG,bI
<
v.(sv)
1 , f

b
> e

b
sv =

vG,bI

v,|f
b
|
<
v
, f
b
> e

b
sv
=

bI
<
|f
b
|
, f
b
> (s[f
b
[) e
b
s[f
b
[,
hence
d(s) =

bI
<
|f
b
|
, f
b
> (s[f
b
[) e
b
s[f
b
[

bI
<
|f
b
|
, f
b
> e
b
s
as stated. This ends the proof of Theorem 3.2.

Corollary 3.3. All irreducible bicovariant dierential calculi on H = kMk(G) are inner,
precisely
dh = [, h], h H, =

b1
<
|f
b
|
, f
b
> e
b
Proof. It follows immediately from the denition of the dierentials in Theorem 3.2 above.
Proposition 3.4. The braiding induced on
1
(H)
1
(H) by the action of D(H

) is given
by
(e
a
e
b
) = (f
a
)[f
b
[) e
b
[f
b
[
1
e
a
Proof. By denition, we have
(e
a
e
b
) =

sM,uG
(s
u
)e
b
e
a
(
s
u).
From (14) we have
(s
u
)e
b
=
u,|f
b
|
(s[f
b
[) e
b
(15)
On the other hand, for any u G we consider the basis f

a
:= uf
a
with the dual basis
e

a
= u
1
e
a
as above and for any s M we have

(
s
u) =

vG
u
1

u
1
s
1
vu
so that
e
a
(
s
u) =

b1
< e
a
,

(
s
u)f

b
> e

b
=

b1
< e
a
,
f
b
,s
f
b
> u
1
e
b
=
s,fa
u
1
e
a
. (16)
13
Inserting (15) and (16) in the formula of the braiding above gives the result as stated.

Corollary 3.5. (i) The right-invariant 1-form is invariant under the action of H

given by
(16) in the sense
y = (y), y H

meaning that is invariant under the left coaction of H on

1
.
(ii) The 1-form generates d on higher degree forms via the graded commutator
d = ,
hence obeys the Maurer-Cartan relation: de
a
= , e
a
.
(iii) The form is closed and nontrivial in the rst noncommutative de Rham cohomology
H
1
(kMk(G)).
Proof. (i) is a direct computation from (16). (ii) comes out from Woronowicz theory[2]: A left
and right invariant 1-form (say) which obeys dh = [, h] h H generates d as stated.
Finally that is closed is obvious, is not exact because one checks easily that in any algebra
with an integral
_
such that
_
1 ,= 0 an invariant 1-form cannot be exact. H is a such algebra
with the integral dened by
_
(s
u
) =
s,e
.

We recover the well known results in cases H = k(G) and H = kM, with a basis of right
invariant 1-forms in each case.
Corollary 3.6. (i) H = k(G): as we said above, here /= kC where C is a certain conjugacy
class in G. The action in (10) becomes uv = u
1
vu. A basis of / = kC is (f
a
= a)
aC
and
we have <
|fa|
, f
a
>= 1, b e
a
= e
bab
1 , so that the results in Theorem 3.2 , Corollary 3.3 and
Proposition 3.4 for H = k(G) read

1
(k(G)) =< e
a
>
aC
k(G)
fe
a
= e
a
(L
a
f), df =

aC
e
a
(
a
f), f k(G), =

aC
e
a
(e
a
e
b
) = e
b
e
b
1
ab
a, b C, L
a
f(x) = f(ax), x G,
a
= L
a
id.
(ii) For H = kM, Let V kM be an irreducible subrepresentation under right regular action
of M on itself. Here /= V and we set (f
a
) a basis of / and (e
a
) the corresponding dual basis
of V

. We have [f
a
[ = e, a. The results in Theorem 3.2 , Corollary 3.3 and Proposition 3.4
for H = kM read

1
(kM) = V

kM
t.e
a
= t e
a
t, dt = (t 1) .t, t M, =

a
<
e
, f
a
> e
a
14
= , (e
a
e
b
) = e
b
e
a
4. Differential calculi on crossproducts kM<k(G)
In this section, we rst verify that the theory above t with the results on crossproducts in
[13] and then add a new result that the calculi on kM also extend naturally to kM<k(G). This
gives of course natural ways to construct dierential calculi on the double D(G).
4.1. Canonical calculi on kM<k(G).
Proposition 4.1. Let X = G>M be a semi-direct factorization. For any M-invariant conju-
gacy class C of G, when it exists, we set
/=

aC
k(

tM
(ta).t) kX (17)
Then
(i) The vector space / is isomorphic to an irreducible quantum tangent space in k(M)<kG,
precisely it is of the form
/=

zC
( z/
0
)
as above.
(ii) The dierential calculus dened on kM<k(G) by (/, C) restricts to the calculus dened
on k(G) by C as in Corollary 3.6.
We call the calculus (/, C), the canonical dierential calculus dened on kM<k(G) by the
conjugacy class C.
The proof which is similar to that of Proposition 6.1 in [13] leads to the following cartan
calculus:
We choose as canonical basis of / the set (f
a
)
aC
where
f
a
=

tM
(ta).t
and obtain from Theorem 3.2, Corollary 3.3 and Proposition 3.4 that:

u
e
a
= e
a

a
1
u
, se
a
= e
sas
1 s
d(
u
) =

aC
e
a
(
a
1
u

u
), ds =

aC
(e
sas
1 e
a
)s
=

aC
e
a
, (e
a
e
b
) = e
b
e
b
1
ab
The restriction of this on k(G) is the calculus in (i) of Corollary 3.6. This is what we expected
from [13], where the following was rather omited:
Proposition 4.2. Any irreducible dierential calculus dened on kM by an irreducible subrep-
resentation V (say) of kM, extends to the irreducible calculus on kM<k(G) dened by (e, V ).
We call it the canonical dierential calculus dened on kM<k(G) by V kM.
15
Proof. In the general theory of Theorem 2.7 we consider the conjugacy class e. Then we have

J
e
= kM and G
e
= X. The action of X on kM becomes tus = ts. Hence the irreducible
subrepresentations / kM under the action of G
e
= X are subrepesentations V kM under
the regular right action of M. By Theorem 2.7 if V =

tM
t i.e. the trivial representation, then
it denes a nul calculus on kM<k(G) as (V ) = 0. Otherwise, V / denes an irreducile
nonzero calculus on kM<k(G). Furthermore, let (f
a
) be a basis of / = V with dual basis
(e
a
). Here [f
a
[ = e =< f
a
> so that the Cartan calculus for the dierential calculus (e, V ) on
kM<k(G) from Theorem 3.2, Corollary 3.3 and Proposition 3.4 reads

v
e
a
= e
a

v
, t.e
a
= t e
a
t, t M, v G
d(
v
) = 0, dt = (t 1) t, =

a
<
e
, f
a
> e
a
= , braiding : (e
a
e
b
) = e
b
e
a
.
The restriction of this calculus on kM is manifestly the calculus dened on kM by V as in
Corollary 3.6.
We then deduce the special case when M = G with the left adjoint action in G, namely:
Corollary 4.3. (i) The irreducible bicovariant dierential calculi on the quantum double D(G)
are dened by conjugacy classes in G and irreducible subrepresentations of the isotropy groups.
(ii) Any nontrivial conjugacy class in G denes both canonical irreducible calculi on k(G)
and on D(G) such that the rst calculus extends to the second.
(iii) Any nontrivial irreducible subrepresentation of kG under the right regular action of G
on itself denes both canonical irreducible calculi on kG and on D(G) such that the rst calculus
extends to the second.
The dierential calculus on D(G) given in Corollary 4.3-(iii) corresponds under the twisting
theorem [19] to the calculus dened on the tensor product algebra kGk(G) by /= k.e.V as
in Proposition 2.8. Hence the geometrical properties of D(G) associated with this calculus are
similar to those of kGk(G). For instance, one checks easily that D(G) is a quantum principal
bundle with this nonuniversal calculus as it was already the case with universal calculus in [18].
Explicitly it is a trivial kG-bundle over the quantum homogeneous space k(G), dened by the
projection
D(G) kGk(G)

i
k.G
where (a
b
) =
b,e
a, i(a) = a 1, for a, b G. The Maurer Cartans 1-form w(g) = g
1
dg
denes also a trivial connection on D(G).
Finally for general bicrossproduct case, we expect a method to deduce calculi on the bi-
crossproduct kMk(G) (say) from those on its dual bicrossproduct k(M)kG. This seems
not obvious even for the case of D(G) with D(G)

where we have at least a link between Z and

16

Z namely the map

C : Z

Z,

C(v
1
t
1
v) = v
1
t
1
v
where v is v G viewed as element of G.e X.
4.2. The case of the crossproduct kM<k(G) with G abelian.
It is known (see [13] and[17]) that when G is abelian, the algebras kG and k(

G) are isomorphic
via Fourier transform. Here

G is the group of characters of G. Similarly k(G) and k.

G are
isomorphic. Explicitly an element f k(G) is viewed as

f =

G,uG
1
[G[
(u
1
)f(u) k

G
while

G is viewed as

uG
(u)
u
k(G)
This induces Hopf algebras (Fourier) isomorphisms
T : kM<k(G) k.M<

G, T

: k(M<

G) k(M)kG
dened by
T(s
u
) =

G
1
[G[
(u
1
)s., T
1
(s.) =

uG
(u)s
u
and
T

(
t.
) =

uG
1
[G[
(u
1
)
t
u, T
1
(
t
v) =

G
(v)
t.
t, s M, u G,

G, where we denote an element (t, ) of M<

G by t. using factorization
notation. The action of M on

G is
(s)(u) = (su)
Hence the cross product kM<k(G) becomes the group algebra k.M<

G. Then the subrepre-
sentations / kX that dene the quantum tangent spaces associated to calculi on kM<k(G)
in the general theory in Theorem 2.7 should correspond to the subrepresentations V that dene
the quantum tangent spaces associated to calculi on k.M<

G in Proposition 2.8. Namely we
have the following result.
Proposition 4.4. In the semidirect case X = G>M with G abelian, There is one to one
correspondence between irreducible subrepresentations / of kX under the action of D(k(X))
in (10) and the irreducible subrepresentations V of k.M<

G under the regular right action of
M<

G on itself.
Proof. (i): We start from a given irreducible subrepresentation V k.M<

G. By Proposi-
tion 2.8, V denes an irreducible bicovariant dierential calculus on k.M<

G whose quantum
tangent space L (k(M<

G))
+
is given by L =

(V ) where

(t.) =
(t.)
1
t.,e
1, (t.) M<

G.
17
As subvector space of (k(M)kG))
+
, L is viewed as L = T

(V ). Next we consider the

D(k(X))-modules morphism : kX (k(M)kG))
+
and set
k

1 /=
1
(L)
as decomposition into submodules with

1 =

sM
s. That / is irreducible is immediate since L
is and : /L is an isomorphism of D(k(X))-modules. Furthermore we have
Lemma 4.5. If V is generated as M<

G-module by w
0
=:

tM,

G

t.
t., (say), then /
is generated as D(k(X))-module by w

0
=:

tM,

G

vG
1
|G|

t.
(t
1
v)v.t, where
t.
are some
scalars.
Indeed what we want is that w

0
is nonzero element of / since the latter is irreducible and
hence generated by each of its nonzero elements. That w

0
is nonzero is obvious from w
0
is
nonzero. To check that w

0
/, it is enough to show that T

(w
0
) = (w

0
), and this is
straightfoward using the property

vG
(t
1
v) = [G[
,e
, t M.
(ii): Conversely, suppose we are given an irreducible subrepresentation / of kX under the
action in (10) which becomes
usvt = v.ts, [[vt[[ = t
1
v.
We associate to / a subrepresention V k.M<

G such that T

(V ) = (/) as follows:
Lemma 4.6. If /is generated as D(k(X))- module by m
0
=

uG,tM

ut
ut, then V is generated
as M<

G-module by m

0
=

tM,

G

uG

ut
(t
1
u
1
)t.
Indeed, m

0
is nonzero since m
0
is nonzero and (t
1
u
1
) is nonzero by denition of

G.
Then we consider the M<

G-module V generated by m

0
. What we need to conclude is to show
that for any s. M<

G we have
T

(m

0
.s) = (m
0
)
for a certain D(k(X)). We compute:
T

(m

0
.s) =

t,u

ut
((ts)
1
u)
(ts)
1 u

t,u

ut

ts,e
(u)1
using the properties
1
(u) = (u
1
) and

G
(u) =
u,e
[G[. Then we show that the element
=: ((ts)
1
u)s D(k(X)) obeys T

(m

0
.s) = (m
0
).
5. Examples
5.1. Dierential calculi on kZ
2
k(Z
3
).
This is a semidirect case with G abelian as in the preceding Section. The calculi on its dual
algebra, namely k(Z
2
)<kZ
3
are in [13].
18
Here X = S
3
factorizes into G = Z
3
= e, u, u
2
and M = Z
2
= e, s where s = (12), u = (123)
are permutations. The right action of G on M is trivial while the left action of M on G is dened
by s = (u, u
2
) ( a permutation). Here

G = Z
3
= e,
0
,
2
0
where
0
(u) = q =: e
2i
3
. Of course
we will identify
0
with u Z
3
. Hence H = kZ
2
k(Z
3
) is isomorphic to k.Z
2
<Z
3
= kS
3
. We
determine the irreducible bicovariant dierential calculi on kZ
2
k(Z
3
) and demonstrate that
they correspond to irreducible subrepresentations of kS
3
under the right regular action of S
3
on
itself.
First of all, we recall that these subrepresentations of kS
3
are given by: S
3
= e, s, u, u
2
, us, u
2
s :
(i) One trivial subrepresention of dimension 1
V
0
=< e + s + u + u
2
+ us + u
2
s >
k
(ii) One nontrivial subrepresentation of dimension 1
V
1
=< e + u + u
2
s us u
2
s >
k
(ii) One irreducible subrepresentation of dimension 2
V
2
=< e + qu
2
+ q
2
u, s + qu
2
s + q
2
us >
k
Of course there is one other 2-dimensional irreducible subrepresentation
V

2
=< e + qu + q
2
u
2
, s + qus + q
2
u
2
s >
k
which is isomorphic to V
2
. Hence if we associate V
0
to the nul dierential calculus then we expect from general theory of Theorem 2.7 two more
irreducible bicovariant calculi: One of dim 1 and one of dim 2.
Indeed here the set

Z = Z
3
splits into two conjugacy classes in S
3
: C
1
= e and C
2
= u, u
2
.
(i) For C
1
= e, G
e
= Z
3
,

J
e
= kZ
2
is decomposed as G
e
-modules as kZ
2
= k(e+s)k(es).
The rst factor /
0
=: k(e + s) corresponds to the nul calculus and /
1
= k(e s) denes one
dimensional bicovariant calculus with the Cartan relations
f
1
= e s,
1
= /

1
H

v
e
1
= e
1

v
, se
1
= e
1
s, d(
v
) = 0, ds = 2e
1
s, v Z
3
,
braiding : (e
1
e
1
) = e
1
e
1
.
(ii) For C
2
= u, u
2
, choose z
0
= u, then G
u
= Z
3
and the decomposition of

J
u
under
G
u
action is

J
u
= ku kus, which leads to the irreducible D(k(S
3
))-subrepresentation /
2
=
ku kus =< f
1
, f
2
>
k
kS
3
, where f
1
= u, f
2
= us. The representation /
2
denes 2-
dimensional dierential calculus on H with Cartan relations

e
e
1
= e
1

u
2,
u
e
1
= e
1

e
,
u
2e
1
= e
1

e
e
2
= e
2

u
,
u
e
2
= e
2

u
2,
u
2e
2
= e
2

e
, se
1
= e
2
s, se
2
= e
1
s
d(
e
) = e
1
(
u
2
e
), d(
u
) = e
1
(
e

u
), d(
u
2 ) = e
1
(
u

u
2), ds = (e
2
e
1
)s
braiding : (e
a
e
b
) = e
b
e
a
Finally, to check that these two calculi match with the bicovariant dierential calculi dened
on kS
3
respectively by V
1
and V
2
, one can turn them to calculi on kS
3
using the isomorphism
19
T above, but it is easy and it is enough to check the correspondence between the generator m
i
of /
i
and w
i
of V
i
(i = 1, 2) given in Lemma 4.6 above. Namely
m
i
=

t,v

vt
vt w
i
=

tZ
2
,v,Z
3

vt
(t
1
v
1
)t.
Indeed, from this formula, for m
1
= e s /
1
, we obtain w
1
= e +u +u
2
s u
2
s us V
1
and for m
1
= u /
2
we obtain w
2
= e + q
2
u + qu
2
V
2
as expected.
5.2. Dierentials and Cartan calculus on kZ
6
<k(S
3
). This is an intermediary case be-
tween the tensor product algebra kZ
6
k(S
3
) and kZ
6
k(S
3
) mentioned in [13] and will also
demonstrate our general theory in Section 4. We take k = C and set Z
6
= t
k
, 0 k 5
i.e. it is generated by t with the unit element t
0
= e., and S
3
= e, b, b
2
, a, ab, ab
2
where
a = (12) and b = (123). The left action of Z
6
on S
3
in the factorization X = S
3
>Z
6
is given
by t = (b, b
2
) (ab, ab
2
) while the right action of S
3
is trivial. Here

Z = S
3
consists of 3
conjugacy classes C = e, C = b, b
2
and C = a, ab, ab
2
. We recall the action from (10),
which becomes: [[vt[[ = t
1
vt, usvm = (tu)
1
v(tu).ts, us X
op
.
We denote the quntum tangent space L rather by / (without hat), and we set L
u
(f)(x) =
f(ux),
u
= L
u
id, x, u S
3
, f C(S
3
). Using these data, we nd the following irreducible
bicovariant dierential calculi and associated cartan calculus for H kZ
6
<k(S
3
):
(i): C = e, /= C.f
q
, where q
6
= 1, f
q
= e + qt + q
2
t
2
+ q
3
t
3
+ q
4
t
4
+ q
5
t
5
. For q = 1
we have the nul dierential calculus.
For q ,= 1, let e
q
be the dual of f
q
. Then the Cartan calculus is given by:
f C(S
3
), t
i
Z
6
, fe
q
= e
q
f, t
i
e
q
= q
i
e
q
t
i
, df = 0, d(t
i
) = (q
i
1)e
q
t
i
.
Braiding: (e
q
e
q
) = e
q
e
q
. dh = [, h], h H where = e
q
.
(ii) For C = b, b
2
, we have /= C. < f
0
, f
1
> where f
0
and f
1
are :
f
0
= b + q
4
bt
2
+ q
2
bt
4
+ q
5
b
2
t + q
3
b
2
t
3
+ qb
2
t
5
f
1
= b
2
+ q
4
b
2
t
2
+ q
2
b
2
t
4
+ q
5
bt + q
3
bt
3
+ qbt
5
, q
6
= 1.
Let (e
0
, e
1
) be the dual basis to (f
0
, f
1
).
1
= /

= C. < e
0
, e
1
>. We have the following
Cartan calculus:

1
(H) =
1
H, f C(S
3
), fe
0
= e
0
L
b
(f), fe
1
= e
1
L
b
2(f), t
2j
e
i
= q
2j
e
i
t
2j
, t
2j+1
e
i
=
q
2j+1
e
i+1
t
2j+1
, i = 0, 1 Z
2
, 0 j 5.
Direntials: df = e
0

b
(f) + e
1

b
2(f), d(t
j
) = (q
j
1)(e
0
+ e
1
)t
j
Braiding: (e
i
e
j
) = e
j
e
i
. d = , , (H), where = e
0
e
1
.
(iii) For C = a, ab, ab
2
, we have /= C. < f
0
, f
1
, f
2
> where :
f
0
= a + q
5
at + q
4
at
2
+ q
3
at
3
+ q
2
at
4
+ qat
5
f
1
= ab
2
+ q
5
abt + q
4
ab
2
t
2
+ q
3
abt
3
+ q
2
ab
2
t
4
+ qabt
5
,
20
f
2
= ab + q
5
ab
2
t + q
4
abt
2
+ q
3
ab
2
t
3
+ q
2
abt
4
+ qab
2
t
5
, q
6
= 1.
Commutation relations and exterior dierentials:
f C(S
3
), fe
0
= e
0
L
a
(f), fe
1
= e
1
L
ab
2(f), fe
2
= e
2
L
ab
(f), t
2j
e
i
= q
2j
e
i
t
2j
, t
2j+1
e
i
=
q
2j+1
e
i
t
2j+1
, i = 0, 1, 2 Z
3
, 0 j 2.
df = e
0

a
(f) + e
1

ab
2 (f)) + e
2

ab
(f), d(t
j
) = (q
j
1)(e
0
+ e
1
+ e
2
)t
j
Braiding: (e
i
e
0
) = e
0
e
i
, (e
i
e
1
) = e
1
e
1i
. (e
i
e
2
) = e
2
e
2i
,
d = , , (H), where = e
0
e
1
e
2
.
For q = 1 in both cases (ii) and (iii), we have the canonical dierential calculus induced on
H by the conjugacy class C from Proposition 4.1.
Applying the theory above to kS
3
<k(S
3
) with conjugation action of S
3
on itself, one recovers
the results obtained on D(S
3
) in [13]. These explicit computations of Cartan calculus for all
irreducible bicovariant dierential calculi on the algebras H open the way for gauge theory
and gravity on H.
Acknowledgements. Part of the main results, examples and the writing up were done during
my visit to the Abdus Salam International Centre for Theoretical Physics, Trieste, Italy, in
summer 2005. I thank the sta there for the invitation. I also thank Shahn Majid (QMUL) for
interesting suggestions. Financial support from ICTP and IMU-CDE is acknowledged.
21
References
[1] A. Connes, Noncommutative Geometry, Academic press, New York.
[2] S.L Woronowicz, Dierential calculus on compact matrix pseudogroups (Quantum Groups), Commun. Math.
Phys. 122: 125-170 (1989).
[3] M. D. Violette, Derivations et Calcul dierentiel non commutatif, C.R. Acad. Sci. Paris, 307, Serie I, 1988:
P:403-408.
[4] S. Majid, Classication of bicovariant dierential calculi, J. Geom. Phys., 25:119-140 (1998).
[5] S. Majid, Riemannian geometry of quantum groups and nite groups with nonuniversal dierentials, Com-
mun. Math. Phys. 225:131-170, (2002).
[6] S. Majid, Classication of dierentials on quantum doubles and nite noncommutative
geometry, Lect. Notes Pure Appl. Math. 239(2004) 167-188, Marcel Dekker
[7] A. Sitarz, Noncommutative geometry and gauge theory on discrete groups, J. Geom. Phys. 15:123-136 (1995)
[8] K. Bresser, F. Muller Hoissen, A. Dimakis, A. Sitarz, Noncommutative geometry of nite groups, J. Phys.
A 29: 2705-2735 (1996).
[9] Ngakeu, S. Majid and D. Lambert, Noncommutative Riemannian geometry of the alternating group A4, J.
Geom. Phys., 42:259-282 (2002).
[10] S. Majid and E. Raineri, Electromagnetism and gauge theory on the permutation group S3, J. Geom. Phys.,
44 (2002) 129-155.
[11] E.J. Beggs and S. Majid, Quasitriangular and dierential structures on bicrossproduct Hopf algebras, J.
Algebra 219: 682-727 (1999).
[12] E.J. Beggs, J.D. Gould and S. Majid, Finite group factorizations and braiding, J. Algebra 181:112-151 (1996).
[13] F. Ngakeu, S. Majid and J-P. Ezin, Cartan calculus for quantum dierentials on
bicrossproducts, Acta Applicandae Mathematicae, 84:193-236, 2004.
[14] M. Takeuchi, Matched Pairs of Groups and Bismash Products of Hopf Algebras, Commun. Alg., 9:841, (1981).
[15] S. Majid, Physics for algebraists: Non-commutative and non-cocommutative Hopf algebras by a bicrossprod-
uct construction, J. Algebra, 130:1764 (1990).
[16] V.G. Drinfeld, Quantum groups, in A. Gleason, ed., Proceedings of the ICM, pp. 798820, AMS (1987).
[17] S. Majid. Foundations of Quantum Group Theory, Cambridge Univ. Press (1995).
[18] T. Brzezi nski, S. Majid, Quantum group gauge theory on quantum spaces, Commun. Math. Phys. 157:591-638
(1993), erratum 167(1995)235.
[19] S. Majid and R. Oeckl, Twisting of quantum dierentials and the Planck scale Hopf algebra, Commun. Math.
Phys., 205:617-655 (1999)
22