a

r
X
i
v
:
0
9
0
8
.
0
5
4
2
v
1


[
m
a
t
h
.
Q
A
]


4

A
u
g

2
0
0
9
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS
FRANCESCO COSTANTINO
Abstract. We show that Kauman brackets of colored framed graphs (also known as quan-
tum spin networks) can be renormalized to a Laurent polynomial with integer coecients by
multiplying it by a coecient which is a product of quantum factorials depending only on the
abstract combinatorial structure of the graph. Then we compare the shadow-state sums and
the state-sums based on R-matrices and Clebsch-Gordan symbols, reprove their equivalence and
comment on the integrality of the weight of the states. We also provide short proofs of most of
the standard identities satised by quantum 6j-symbols of Uq(sl
2
).
Contents
1. Introduction 1
1.1. Structure of the paper 2
1.2. Acnowledgements 3
2. Representations of U
q
(sl
2
) and graph invariants 3
2.1. Basic facts on U
q
(sl
2
) 3
2.2. Elementary operators 4
2.3. The state-sum computing G, col). 7
2.4. Invariants of framed trivalent graphs 8
2.5. Remarks on orientation and framings 10
3. Integrality 11
3.1. Examples and properties 12
4. Shadow state sums and integrality 13
4.1. Shadow state sums 14
4.2. Simplifying formulas 17
4.3. Examples and comments on integrality 17
4.4. Identities on 6j-symbols 18
5. R-matrices vs. 6j-symbols 20
References 22
1. Introduction
The family of U
q
(sl
2
)-quantum invariants of knotted objects in S
3
as knots, links and more in
general trivalent graphs, can be dened via the recoupling theory ([6]) as well as via the theory
of representations of the quantum group U
q
(sl
2
) ([12],[7]) and the so-called theory of shadows
([13], Chapter IX). These invariants are dened for framed objects (framed links or graphs, see
Denition 2.11) equipped with a coloring i.e. a map from the set of 1-dimensional strata of
the object satisng certain admissibility conditions (Denition 2.6), and take values in Q(q
1
2
). As
customary in the literature, we shall call these invariants the Kauman brackets of the colored,
1
2 COSTANTINO
framed graph G and denote them by G, col). In particular, if G is a framed knot k the set of
admissible colorings is the set of half-natural numbers (here we use the so-called spin notation)
and they coincide with the unreduced colored Jones polynomials of the knot: k, s) = J
2s+1
(k).
Although the denition of Kauman brackets via recoupling theory is simple and appealing,
the denition based on the theory of representations of U
q
(sl
2
) happens to be more useful for
our purposes. In Section 2 we will sketch the proof of the equivalence of the two denitions (the
relations have been already studied by S. Piunikhin [11]). Roughly speaking, the invariant of
a graph G equipped with a coloring col (which is a function of q) is the trace of a morphism
between representations of U
q
constructed by composing elementary morphisms associated to
a decomposition of a diagram of G into simple tangles. The only dierence with respect to the
general construction of quantum invariants via representations of quantum groups is that one
can get rid of the orientations of the strands since the very beginning: this can be achieved by
modifying the operators associated to maxima and minima and using the isomorphisms between
the representations of U
q
(sl
2
) and their duals.
It is known that, in general, G, col) is a rational function of the variable q
1
2
(there are various
notations in the literature, for instance our q
1
2
is A in [6]). If L is a framed link, it was shown by
T. Le ([9]) that, up to a factor of the form q

n
4
, L, n) is a Laurent polynomial in q
1
2
(actually
in [9] a much stronger result is proved which holds for general polynomial invariants issued from
quantum group representations).
On contrast it is well known that G, col) is not in general a Laurent polynomial if G is a trivalent
graph. The main result of the present paper is Theorem 3.3, where we show that multiplying
G, col) by an explicit balanced product of quantum factorials which depends only on the abstract
combinatorial structure of G, on col and on the framing on G, one gets a Laurent polynomial in
q
1
2
that we will denote G, col)). It turns out that G, col)) = G, col) if G is a 0-framed link.
This normalization was proposed and conjectured to be integral by S. Garoufalidis and R. Van der
Veen (in [5], where they also proved the integrality in the classical case when q = 1) in order
to dene generating function for classical spin networks evaluations. We hope that this result will
allow further development in that direction and in the understanding of the categorication of
U
q
(sl
2
)-quantum invariants for general knotted objects.
The last sections are almost independent on the preceding ones. In Section 4 we recall the deni-
tion of shadow-state sums to compute G, col)), give a new self-contained proof of the equivalence
(rst proved in [8]) between the shadow-state formulation and the R-matrix formulation of the
invariants, and comment on the non-integrality of the single shadow-state weights. Using shadow
state-sums we also provide short proofs of the most known identities for 6j-symbols (e.g. Racah,
Biedenharn-Elliot, orthogonality). In the last section we will quickly comment on the case when
G has non-empty boundary and on the algebraic meaning of the shadow-state sums with respect
to the state-sum based on R-matrices and Clebsch-Gordan symbols.
1.1. Structure of the paper. In Section 2 we will recall the basic facts we will need of the repre-
sentation theory of U
q
(sl
2
) and provide explicit formulas for the elementary morphisms associated
to each elementary tangle: if you are not interested in explicit formulas, jump to Subsection 2.3.
Then, in Section 3 we will dene G, col) and prove its integrality. Section 4 is almost independent
on the rst sections (basically it depends only on Lemma 3.10); there we explain how to com-
pute G, col)) via shadow state-sums and provide short proofs of some well known identities for
6j-symbols. In Section 5 we comment on the algebraic meaning of shadow-state-sums.
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 3
1.2. Acnowledgements. I wish to thank Fran cois Gueritaud, Roland Van der Veen and Vladimir
Turaev for the comments and suggestions they gave me. This work was supported by the French
ANR project ANR-08-JCJC-0114-01.
2. Representations of U
q
(sl
2
) and graph invariants
2.1. Basic facts on U
q
(sl
2
). Let q C, [n]
q
n
q
n
qq
1
and [n]!

n
j=1
[j], [0]! = 1. Let also

n
k

[n]!
[k]![nk]!
.
Denition 2.1. U
q
(sl
2
) is the algebra generated by E, F, K and K
1
with relations:
[E, F] =
K
2
K
2
q q
1
, KE = qEK, KF = q
1
FK, KK
1
= K
1
K = 1
Its Hopf algebra structure is given by:
(E) = E K +K
1
E, (F) = F K +K
1
F, (K) = K K
S(E) = qE, S(F) = q
1
F, S(K) = K
1
(E) = (F) = 0, (K) = 1
Remark 2.2. To make clear the relation with other notations note that our notation is coherent
with that of [7] after replacing their s with q; our q corresponds to q = A
2
in [1] and our E, F
respectively to X and Y .
Lemma 2.3. For each a
N
2
there is a simple representation V
a
of U
q
(sl
2
) of dimension 2a + 1
whose basis is e
j
m
, m = j, j + 1, . . . , j and on which the action of E, F, K is:
E(e
a
m
) = [a m][a +m+ 1]e
a
m+1
, F(e
a
m
) = e
a
m1
, K(e
a
m
) = q
m
e
a
m
Remark 2.4. In [7] a dierent base f
a
m
for V
a
is used. Namely in the base f
a
m
it holds:
E(f
a
m
) = [a +m+ 1]f
a
m+1
, F(f
a
m
) = [a m + 1]f
a
m1
, K(f
a
m
) = q
m
f
a
m
The change of basis is given by a diagonal matrix: f
a
m
=
a
m
e
a
m
with
a
m
=
[2a]!
[am]!
.
Recall that the dual (V
a
)

of V
a
is a a left representation via the action
X f(e) f(S(X) e), X U
q
(sl
2
), e V
a
, f (V
a
)

Let E
a
m
and F
a
m
be the dual bases in (V
a
)

respectively of e
a
m
and f
a
m
.
Proposition 2.5. The modules V
a
and (V
a
)

are isomorphic by the map D : (V

a
)

V
a
given
by D(E
a
m
) = [2a]!

1
2m
q
m
e
a
m
. The inverse of D is D
1
(e
a
m
) = E
a
m

1
2m
q
m
[2a]!
.
Proof of 2.5. In [7], Theorem 2.13, it was proved that V
a
and (V
a
)

are isomorphic via D : (V

a
)

V
a
given in the basis F
a
m
by D(F
a
m
) = (q)
m
[am]![a+m]!
[2a]!
f
a
m
; hence since F
a
m
= (
a
m
)
1
E
a
m
we have
D(E
a
m
) =
[am]![a+m]!
[2a]!

a
m
(
a
m
)
1
(q)
m
e
a
m
= [2a]!(q)
m
e
a
m
. The last statement follows. 2.5
4 COSTANTINO
2.2. Elementary operators. In this subsection we recall how to associate morphisms between
U
q
(sl
2
) modules to the elementary graphs of Figure 2. Let G be a trivalent graph, E the set of
edges of G, V the set of vertices:
Denition 2.6 (Admissible coloring). An admissible coloring of a KTG G is a map col : E
N
2
(whose values are called colors) such that v V the following conditions are satised:
(1) a
v
+b
v
+c
v
N
(2) a
v
+b
v
c
v
, b
v
+c
v
a
v
, c
v
+a
v
b
v
where a
v
, b
v
, c
v
are the colors of the edges touching v.
Let now D be a diagram of G and let us x from now on a height function on R
2
which allows
us to chop D into elementary graphs as those shown in Figure 2. If G is equipped with a coloring
each subgraph inherits one.
The general construction of quantum invariants via representations of U
q
(sl
2
) ([12]) requires an
orientation on every 1-dimensional stratum of G, decomposes D it into elementary subgraphs and,
according to the color of each strand and its orientation, considers sutable morphisms between
tensor products of the representations associated to the oriented edges. In our case we want to do
the same but getting rid of the orientations. The rough idea to achieve this, is to orient all the
edges towards the bottom. This is possible for all the elementary subgraphs except for the two
extrema: in this case we will modify the standard construction by inserting a coupon representing
the isomorphism between a representation and its dual in order to allow sources on maxima and
sinks on minima. All the other operators we will dene coincide with the standard ones. Here
below we dene the operators associated to each elementary subgraph:
2.2.1. Half-twists. Let us start by the half twist operator H
a
: V
a
V
a
which is H
a

(

1)
2a
q
(a
2
+a)
id. We will later associate it to a vertical a-colored strand whose framing per-
forms a positive half twist with respect to the blackboard framing.
2.2.2. Cup and Cap. Let us now dene the
a
operator:
a
: C = V
a
V
a
V
a
. Let us
rst dene B
a
: V
0
V
j
(V
a
)

by B
a
(1)

u=a
u=a
e
a
u
E
a
u
, and T
a
: (V
a
)

V
a
V
0
by
T
a
(E
a
u
e
a
v
) =
u,v
1. We dene
a
as
a
(idD) B
a
(where D is the morphism of Proposition
2.5). Similarly, we dene
a
: V
a
V
a
V
0
= C as
a
T
a
(D
1
id). An explicit computation
in the bases e
a
u
of the above operators gives:
(1)
a
(e
a
u
e
a
v
) =
u,v

1
2u
q
u
[2a]!
e
0
0
(2)
a
(1) =
a

u=a
[2a]!

1
2u
q
u
e
a
u
e
a
u
2.2.3. R-matrix. We associate to each positive crossing the action of Drinfelds universal R-matrix.
Lemma 2.7. The morphism
a
b
R : V
a
V
b
V
b
V
a
given by the composition of Drinfelds
universal R-matrix with the ip of the coordinates in the basis e
a
u
e
b
v
is:
(3) R(e
a
u
e
b
v
) =

n0
[n]!(q q
1
)
n

a u
n

a +u +n
n

q
2uvn(uv)
n(n+1)
2
e
b
vn
e
a
u+n
where the sum is taken over all the n such that [u +n[ a and [v n[ b. We will denote
a
b
R
h,k
u,v
the coecient of R(e
a
u
e
b
v
) with respect to e
b
h
e
a
k
.
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 5
Proof of 2.7. We use the formulas provided in [7] (Corollary 2.32: recall that t = q

1
2
) in the basis
f
j
m
and the diagonal change of basis f
j
m
=
[2j]!
[jm]!
e
j
m
to compute the coecients of R in our base
and rewrite them using quantum binomials:
R(e
a
u
e
b
v
) =

n0
(q q
1
)
n
[n]!
[a +u +n]![b v +n]!
[a +u]![b v]!
[a u]!
[2a]!
[b v]!
[2b]!

[2a]![2b]!
[a u n]![b v +n]!
q
2uvn(uv)
n(n+1)
2
e
b
vn
e
a
u+n
=
=

n0
[n]!(q q
1
)
n

a u
n

a +u +n
n

q
2uvn(uv)
n(n+1)
2
e
b
vn
e
a
u+n
2.7
The morphism associated to a negative crossing whose upper strand is colored by a and whose
lower strand is colored by b is the inverse of
a
b
R and can be computed in terms of the one we just
gave and two extrema: (R

) (Id
a
Id
b

a
) (Id
a

a
b
RId
a
) (
a
Id
a
Id
a
). An explicit
formula is then computed out of 3, 2 and 1:
(4)
a
b
R

(e
b
v
e
a
u
) =

n0
[n]!(q
1
q)
n
q
2vu+n(uv)+
n(n+1)
2

a u
n

a +u +n
n

e
a
u+n
e
b
vn
where the sum is taken over all the n such that [u + n[ a and [v n[ b. Remark that
a
b
R

=
a
b
R
1
=
a
b
R[
qq
1 because
a
b
R

a
b
R = Id
a
Id
b
. We will denote
a
b
(R
1
)
h,k
u,v
the coecient
of R
1
(e
a
u
e
b
v
) with respect to e
b
h
e
a
k
.
2.2.4. Clebsch-Gordan theory. To dene the operators associated to the Y -shaped elementary
graph (the 7
th
from the left in Figure 2), we need to x once and for all morphisms between tensor
products of V
j
s. Recall that, by Clebsch-Gordan decomposition theorem, V
a
V
b
is isomorphic to
V
a+b
V
a+b1
. . . V
|ab|
. Hence by Schurs lemma, to x an element Y
a,b
c
Hom(V
c
, V
a
V
b
)
it it is sucient to x Y
a,b
c
(e
c
c
), and this can be done in a unique way (up to a scalar factor). Once
this is done, the image of e
c
u
will be computable by Y
a,b
c
(e
c
u
) = (F
cu
)(Y
a,b
c
(e
c
c
)). So suppose for
a moment we xed Y
a,b
c
(e
c
c
) for each pair (c, (a, b)):
Denition 2.8. The quantum Clebsch-Gordan coecient C
a,b,c
u,v,t
is the coecient in the sum:
Y
a,b
c
(e
c
t
) =

u+v=t
C
a,b,c
u,v,t
e
a
u
e
b
v
It is clear that C
a,b,c
u,v,t
is zero if u +v ,= t because (K)(e
a
u
e
b
v
) = q
u+v
e
a
u
e
b
v
and the weight
of a vector is preserved by a morphism. The co-product formula for F gives
(F)(e
a
u
e
b
v
) = q
v
e
a
u1
e
b
v
+q
u
e
a
u
e
b
v1
and so the equality Y
a,b
c
(Fe
c
t
) =

t
C
a,b,c
u,v,t
(F)(e
a
u
e
b
v
) implies the following recursion relation:
(5) C
a,b,c
u,v,t
= C
a,b,c
u+1,v,t+1
q
v
+C
a,b,c
u,v+1,t+1
q
u
which holds for every c > t c (with C
a,b,c
u,v,t
= 0 if [u[ > a or [v[ > b). Moreover, the following
recursion is a consequence of the equality 0 = Y
a,b
c
(E(e
c
c
)) and holds for u < a:
(6)
[a u]
[b +v]
q
v
C
a,b,c
u,v,c
+
[b v + 1]
[a +u + 1]
q
u1
C
a,b,c
u+1,v1,c
= 0
6 COSTANTINO
=
=

a
a a
b
b
b
c
c
c
P
c
a,b
(id
c

b
) Y
c,b
a
(
a
id
c
) Y
a,c
b
Figure 1. The rst equality is the denition of P
c
a,b
, the second is Lemma 2.10.
Therefore by equations 5 and 6, one can compute all the coecients C
a,b,c
u,v,t
out of C
a,b,c
a,ca,c
; this is
done in the following proposition (where an implicit choice of C
a,b,c
a,ca,c
is made):
Proposition 2.9.
C
a,b,c
u,v,t
=

1
cab
(1)
(bv)(ct)
q
(bv)(b+v+1)(au)(a+u+1)
2
[a +b c]![b +c a]![c +a b]!
[2c]!

(7)

z,w:z+w=ct
(1)
z
q
(zw)(c+t+1)
2

a +u +z
a +c b

b +v +w
b +c a

c t
z

where the sum is taken over all z, w N such that z + w = c t and all the arguments of the
factorials are non-negative integers.
Proof of 2.9. In [1], Lemma 3.6.10, using the basis g
a
u
= [a + u]!e
a
u
for V
a
, the following formula
was provided for the Clebsch Gordan coecients (with the same notation as above, where we are
rewriting the formula via q-binomials and correcting a missing factor of

1
(tc)
):
C
a,b,c
u,v,t
=

1
cab
(1)
(bv)(ct)
q
(bv)(b+v+1)(au)(a+u+1)
2
[c +t]![c t]!
[2c]!

(8)

z,w:z+w=ct
(1)
z
q
(zw)(c+t+1)
2

a +b c
a u z

a +u + z
z

b +v +w
w

To get our statement it is then sucient to multiply by

[a+u]![b+v]!
[c+t]!
, to single out of the factorials
the terms
[a+bc]![b+ca]![c+ab]!
[2c]!
and to pair the factorials in the denominators of the summands
so that their sums match a +u +z, b +v +w and c t. 2.9
In order to get invariants of unoriented graphs, the operators associated to the other elementary
graphs must constructed by moving a leg of the Y -shaped graph up or down (see Figure 1).
Hence we dene projectors P
c
a,b
: V
a
V
b
V
c
out of Y
a,b
c
as: P
c
a,b
(
a
Id
c
) (Id
a
Y
a,c
b
).
So letting P(e
a
u
e
b
v
) =

t
P
a,b,c
u,v,t
e
c
t
, the coecients can be explicitly computed as:
(9) P
a,b,c
u,v,t
= C
a,c,b
u,t,v

1
2u
q
u
[2a]!
One may also dene P
c
a,b
by pulling-up the right leg i.e. by setting P
c
a,b
(Id
c

b
)(Y
c,b
a
Id
b
)
(as in the r.h.s. of Figure 1), but as the following lemma shows, the two choices are equivalent:
Lemma 2.10. It holds:
C
a,c,b
u,t,v

1
2u
q
u
[2a]!
= C
c,b,a
t,v,u

1
2v
q
v
[2b]!
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 7
Proof of 2.10. Since both tensors express maps V
a
V
b
V
c
and the space of such maps is one
dimensional, it is sucient to prove the equality for a single vector. It is straightforward to check
them for t = c, recalling that u +v = t. 2.10
Similarly, we can dene non-smooth minima operators W
a,b,c
Hom(C, V
a
V
b
V
c
) by
pulling up the only lower leg in Y
a,b
c
, i.e. by setting W
a,b,c
(Y
a,b
c
Id
c
)
c
. So, letting:
W
a,b,c
(1)
a

u=a
b

v=b
c

t=c
W
a,b,c
u,v,t
e
a
u
e
b
v
e
c
t
the coecients are:
W
a,b,c
u,v,t
C
a,b,c
u,v,t

1
2t
q
t
[2c]! =

1
abc
(1)
(bv)2t
q
(bv)(b+v+1)(au)(a+u+1)
2

(10)
[a+bc]![b+ca]![c+ab]!

z,w:z+w=c+t
(1)
z
q
t+
(zw)(ct+1)
2

a +u +z
a +c b

b +v +w
b +c a

c +t
z

Finally, an invariant tensor in M

a,b,c
Hom(V
a
V
b
V
c
, C) can be dened by pulling down
the third leg of P
c
a,b
: M
a,b,c

c
(P
c
a,b
Id
c
). Explicitly, its coecients are:
(11) M
a,b,c
u,v,t
M
a,b,c
(e
a
u
e
b
v
e
c
t
) = P
a,b,c
u,v,t

1
2t
q
t
[2c]!
2.3. The state-sum computing G, col).
Denition 2.11 (KTG). A Knotted Trivalent Graph (KTG) is a nite trivalent graph G S
3
equipped with the germ of a smooth surface S S
3
such that S retracts on G.
Remark 2.12. Not that this is more than a fat graph as S is required to exist around all G
and not only around its vertices. Moreover we are not requiring S to be orientable, even though
this assumption can simplify the form of the renormalization constant (see 3.2).
In order to specify a framing S on a graph G we will only specify (via thin lines as in the leftmost
drawing of Figure 2) the edges around which it twists with respect to the blackboard framing in a
diagram of G, implicitly assuming that S will be lying horizontaly around G out of these twists.
Let now G be a KTG equipped with an admissible coloring col. Decompose G into elementary
subgraphs (i.e. those obtained by possibly adding some parallel vertical strands to those depicted
in Figure 2). Associated to each such subgraph there is a map of representations of U
q
(sl
2
) as
explained in the preceding subsection. The composition of all the morphisms associated to the
elementary subgraphs provides a morphisms G, col) : V
0
V
0
= C therefore a rational function
in q
1
4
(the
1
4
is only due to the H
j
operators: all the other operators provide terms in C(q
1
2
)). This
section is devoted to explain in detail how to compute this morphism via a state sum based on the
formulas provided in the preceding sections.
Let G be a closed KTG, E be the set of its edges, V the set of its vertices and col : E
N
2
an admissible coloring. Let also D be a diagram of G and for every e E, let g
e

N
2
be the
dierence between the framing of e in G and the blackboard framing on it (it is half integer
because the two framings may dier of an odd number of half twists). Let C, M, N be respectively
the set of crossings, maxima and minima in D (recall that we are xing a height function on R
2
to decompose D into elementary subgraphs). Then let f
1
, . . . f
n
be the connected components of
D (V M N C). Remark that each f
j
is a substrand of an edge of G therefore it inherits a
color which we will denote c
j
. To express G, col) as a state-sum, let us rst dene a state:
8 COSTANTINO
Denition 2.13 (States). A state is a map s : f
1
, . . . f
n

Z
2
such that [s(f
j
)[ c
j
, j =
1, . . . n and s(f
j
) is integer i c
j
is. Given a state s, we will call the value s(f
j
) the state of f
j
.
(Equivalently, a state is a choice of one vector e
cj
s(fj )
of the base of V
cj
for each component of
D (V M N C).)
The weight w(s) of a state s is the product of a factor w
s
(x) per each x crossing, vertex,
maximum and minimum of D. To dene these factors, in the following table use the letters a, b, c
for the colors of the strands and u, t, v, w for their states.
w
s
(
a
u v
) =
u,v

1
2u
q
u
[2a]!
w
s
(
a
u v
) =
v,u
[2a]!

1
2u
q
u
w
s
(
w
a b
u v
t
) =
a
b
R
t,w
u,v
(see Formula 3) w
s
(
w
a
b
u v
t
) =
a
b
(R
1
)
t,w
u,v
(see Formula 4)
w
s
( a b
c
u v
t
) = C
a,b,c
u,v,t
(see Formula 2.9) w
s
(
a b
c
u v
t
) = P
a,b,c
u,v,t
(see Formula 9)
w
s
(
a b c
u v t
) = W
a,b,c
u,v,t
(see Formula 2.2.4) w
s
(
a b
c
u v t
) = M
a,b,c
u,v,t
(see Formula 11)
Finally, to take into account the action of the half-twist operators H
j
on the edges of G, let
F(G, col)

eedges

1
4gecol(e)
q
2ge(col(e)
2
+col(e))
be the framing factor (note that it does not
depend on any state). The weight of a state s of G, col) is then dened as follows:
(12) w(s) = F(G, col)

Mmaxima
w
s
(M)

mminima
w
s
(m)

ccrossings
w
s
(c)

vvertices
w
s
(v)
The value of G, col) is then given by :
(13) G, col) =

sstates
w(s)
Remark 2.14. Note that, up to isotopy, one can always suppose that D contains only positive
crossing, maxima minima and vertices with 3 top legs.
2.4. Invariants of framed trivalent graphs. The state-sum dened in the preceding subsection
is just a reformulation of the standard recoupling theoretical invariants (and, in particular, is an
invariant of colored KTGs):
Proposition 2.15. The function G, col) C(q
1
4
) is an invariant up to isotopy of (G, col) and
coincides with the Kauman bracket G, col) of [6] after replacing A = q
1
2
.
Note that a way to prove that G, col) is an invariant consists to check Relations 1-13 of [12] (see
Figure 8) for the operators we provided. This is easy because the only dierence with respect to
the standard case of oriented graphs consists in the movements involving the creation/deletion of
a maximum or minimum. Therefore one only needs to check relations 1 4 (which are straightfor-
ward), 8 (which is easily computed on a highest weight vector), 9, 10 and 13 (which is a consequence
of Lemma 2.10). We will instead sketch a dierent approach which justies also the second state-
ment of the proposition.
Example 2.16. If L = L
1
, . . . L
n
is an un link with (possibly half-integral) framings g
1
, . . . g
n
and
colored by colors c
1
, . . . c
n
then:
L, col) =
n

i=1
(1)
2ci
[2c
i
+ 1](

1)
4gici
q
2gi(c
2
i
+ci)
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 9
a
a
a
a
a
a
a
a
a
b b
b
b
b
b
c
c
c
c

a
a
b
R
a
b
R

Y
a,b
c
H
a
W
a,b,c
M
a,b,c
P
c
a,b
Figure 2. The elementary graphs and the associated morphisms. In all the
drawings except the leftmost, the framing is the blackboard framing.
a
a
a
b
b
b
c
c
c
=

M
a,b,c
W
a,b,c
Figure 3. The operator associated to a colored theta graph.
In particular Proposition 2.15 holds in this case. Indeed it is sucient to prove it for the case of
an unknot colored by c
j
and with framing g
j
. In that case the value is the trace of the operator

cj
(Id H
2gj
cj
)
cj
which equals:
tr(
cj
(Id H
2gj
cj
)
cj
) = (

1)
4gici
q
2gi(c
2
i
+ci)
cj

u=cj

1
4u
q
2u
=
(

1)
4gici
q
2gi(c
2
i
+ci)
(1)
2cj
cj

u=cj
q
2u
= (1)
2cj
[2c
j
+ 1](

1)
4gici
q
2gi(c
2
i
+ci)
Remark 2.17. We allow half integer framings i.e. some component of L may be framed by a
Moebius band instead of an annulus. The standard case corresponds to g
i
N.
Proof of 2.15. We bound ourselves to sketch the idea of the proof. First remark that the operator
P
c
a,b
we dened coincides with the operator dened in [1] (Denition 3.6.7), up to operating the
diagonal change of base g
j
m
= [j + m]!e
j
m
. To prove this claim it is sucient to show that the
operators are equal on vectors of the form e
a
u
e
b
v
with u + v = c. This is proved by comparing
the explicit formulas given in [1] (Lemma 3.6.8) and our denition of P
c
a,b
.
Then the following results of [1] hold true in our case without changes: Theorem 3.7.1, Lemma
3.10.9, Theorem 3.12. These results are all the basic tools of skein theory which allow one to reduce
the computation of G, col) to that of a union of unlinks and in Example 2.16 we already saw that
in that case the invariants coincide. 2.15
Example 2.18 (The theta graph). Consider a theta graph as in Figure 3. Equip it with the
blackboard framing and color the edges by V
a
, V
b
and V
c
. Its invariant is then:
(14) (a, b, c)
a

m1=a
b

m2=b
c

m=c
W
a,b,c
m1,m2,m
M
a,b,c
m1,m2,m
10 COSTANTINO
By Proposition 2.15 the above formula equals the standard skein theoretical value:
(15) (a, b, c) = (1)
a+b+c
[a +b +c + 1]![a +b c]![a +c b]![b +c a]!
[2a]![2b]![2c]!
Remark 2.19. The above example shows that in general G, col) is not a Laurent polynomial:
consider for instance the case a = b = c = 1 in Formula 15.
2.5. Remarks on orientation and framings. Before concluding we stress that the machinery
we recalled applies to any framed, colored trivalent graph G and in particular to those whose
framing S is not an orientable surface. Of course, up to twisting around some edges it is always
possible to change the framing S of G in order to make it an orientable surface. Moreover, if D is a
diagram of G there is a framing S
D
induced on G simply by considering a surface containing G and
lying almost parallel to the projection plane. Pulling back the orientation of R
2
shows that S
D
is
orientable. The following is a converse: every orientable framing is (up to isotopy) the blackboard
framing of some diagram of G.
Lemma 2.20. If G is a framed graph and S is and orientable framing on G then there exists a
diagram D of G such that S coincides with the blackboard framing on G.
Proof of 2.20. The idea of the proof is to x a diagram D and count the number of half twists
of dierence on each edge of G between S
D
and S. The reduction mod 2 of these numbers forms
an explicit cochain in H
1
(G; Z
2
) which is null cohomologous because S and S
D
are orientable.
The coboundary reducing it to the 0 cochain corresponds to a nite number of moves as those in
Lemma 2.21 which change D and isotope G into a position such that the number of half twists of
dierence between S and S
D
is even on every edge. Then up to adding a suitable number of kinks
to each edge of G this dierence can be reduced to 0 everywhere. 2.20
Lemma 2.21 (Half-twist around a vertex). For every admissible 3-uple (a, b, c) it holds:
a b
c
=
a
b
R Y
a,b
c
= (H
1
b
H
1
a
) Y
b,a
c
(H
c
) =
a
b
c
Proof of 2.21. It is sucient to prove the equality for the highest weight vector e
c
c
V
c
; to do
this it is sucient to check that
a
b
R Y
a,b
c
(e
c
c
) and (H
1
b
H
1
a
) Y
b,a
c
(H
c
)(e
c
c
) have the same
coecient with respect to the element e
b
b
e
a
cb
of the basis e
b
j
e
a
i
, [i[ a, [j[ b of V
b
V
a
.
By formulas 3 and 2.9 this coecient is for the left hand side:

u+v=c
a
b
R
b,cb
u,v
C
a,b,c
u,v,c
=
a
b
R
b,cb
cb,b
C
a,b,c
cb,b,c
=
= q
2(cb)b

1
cab
(1)
0
q

1
2
(a(cb))(a+(cb)+1)
[2b]![a +c b]!
[2c]!
The coecient on the r.h.s. is (H
b
b
)
1
(H
a
a
)
1
C
b,a,c
b,cb,c
(H
c
c
) which equals:

1
2c2a2b
q
c
2
+c(a
2
+a)(b
2
+b)

1
cba
(1)
a(cb)
q
1
2
(a(cb))(a+(cb)+1)
[2b]![a +c b]!
[2c]!
=
=

1
cba
q
c
2
+c(a
2
+a)(b
2
+b)
q
1
2
(a(cb))(a+(cb)+1)
[2b]![a +c b]!
[2c]!
A straightforward computation shows that the two coecients are indeed equal. 2.21
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 11
3. Integrality
Let G be a closed KTG, E be the set of its edges, V the set of its vertices and col : E
N
2
an
admissible coloring. We will dene the Euler characteristic (e) of an edge e E as 1 if e touches
a vertex and 0 otherwise (some edges of G may be knots). Let also D be a diagram of G and for
every e E, let g
e

N
2
be the dierence between the framing of e in G and the blackboard framing
on it (it is half integer because the two framings may dier of an odd number of half twists). We
dene a renormalization for G, col) as follows:
G, col)) G, col)
F(G, col)
1

eE,(e)=1
[2col(e)]!

vV
[a
v
+b
v
c
v
]![b
v
+c
v
a
v
]![c
v
+a
v
b
v
]!
where by a
v
, b
v
, c
v
we denote the colors of the three edges surrounding v.
Remark 3.1. Except for the term F(G, col)
1
=

1
P
eE
4gecol(e)

eE
q
ge(col(e)
2
+col(e))
, the
normalization factor depends only on the combinatorial structure of the abstract graph G and not
on its embedding in S
3
. The factor F(G, col) encodes only how much the framing of G is twisted
with respect to the blackboard framing and does not inuence the integrality of the invariant, so
we could omit it from our normalization. The reason why we keep it is that it may shift powers of
q of the form q
n
4
, n N present in G, col) so that only to half integral powers of q are present
in G, col)).
Remark 3.2. Of course if the framing on G is the blackboard framing of its diagram then the
factor F(G, col) is 1, and, by Lemma 2.20 this situation can be achieved up to isotopy i the
framing is an orientable surface. Hence, if one restricts to graphs whose framings are orientable
surfaces then one may drop the factor F(G, col)
1
from the above denition and the following
Theorem still holds with no changes in its wording.
Theorem 3.3 (Integrality of the renormalized Kauman brackets).
G, col)) Z[q

1
2
]
Moreover, << G, col >> is divisible in Z[q

1
2
] by [2n + 1] for each n color of an edge of G.
Proof of 3.3. Up to isotopy, we can suppose that the diagram D of G is the closure of a (1, 1)-
tangle G

whose boundary strands are colored by n and also (by small isotopies around vertices and
crossings) that D contains only maxima, minima, positive crossings and vertices with 3 top legs.
The factor F(G, col)
1
in our normalization simplies with the factor F(G, col) in the weight of
each state (see formula 12), so that we may suppose from now on that all the framings g
e
are 0 (i.e.
that G is equipped with the blackboard framing). By Schurs lemma the morphism represented by
G

is Id
V
n. We claim that
Q
eE,(e)=1
[2col(e)]!
Q
vV
[av+bvcv]![bv+cvav]![cv+avbv]!
belongs to Z[q

1
2
]; this will
conclude because G, col)) = (1)
2n
[2n + 1].
To prove our claim let us dene renormalized operators associated to each maximum, mini-
mum, crossing and vertex of D equipped with a state as follows:
NH
j
H
j
(N
a
)
u,v

u,v
1
[2a]!
u v
a
(N
a
)
u,v

u,v
[2a]!
v u
a
(
a
b
NR)
t,w
u,v
:=
a
b
R
t,w
u,v
(NW)
a,b,c
u,v,t

W
a,b,c
u,v,t
[a +b c]![b +c a]![c +a b]!
Since the morphism represented by G

is diagonal, the only non-zero weight states are those

where the states of the top and bottom strand of G

are equal. Therefore, if in formula 13 one xes

12 COSTANTINO
the same state u on the top and bottom strand of G

and replaces each weight by its normalized

version dened above, the result will be:
G

, col)) = G

, col)

eE
([2col(e)]!)
(cap(e)cup(e))

vV
[a
v
+b
v
c
v
]![b
v
+c
v
a
v
]![c
v
+a
v
b
v
]!
where for each edge, cup(e) (resp. cap(e)) are the number of minima (resp. maxima) on e. One
sees that the above formula gives the same normalization factor as in the claim by remarking that,
since by our hypothesis on D all the vertices have 3 top legs, for each edge e dierent from the top
strand it holds (e) = cap(e) cup(e), and that for the top strand cap(e) cup(e) = 0.
To prove the claim remark now that that each normalized operator belongs to Z[

1][q

1
2
].
This is straightforward because of Lemma 3.5 and Formulas 3 and 2.2.4. We are then left to show
that actually all the coecients are non-imaginary. Let us remark that in the state-sum for G, col,
since each edge with (e) = 1 has two endpoints and each NW
a,b,c
belongs to

1
abc
Z[q

1
2
],
the product of the factors

1
col(e)
coming from these vertices is

1
2col(e)
. Similarly the
product of the factors

1
2ui
coming from the cups and caps on e is

1
2col(e)(cap(e)cup(e))
=

1
2col(e)
(because each u
i
is half integer i col(e
i
) is) and this cancels with the previous
imaginary phase. 3.3
Corollary 3.4. If G, col is a colored KTG and a
1
, . . . a
k
are pairwise co-prime colors of some
edges of G then G, col)) is divisible by

j
[2a
j
+ 1] in Z[q

1
2
].
Lemma 3.5. Let a
1
, . . . a
s
be integers and let the q-multinomial be dened as

a
1
+ +a
s
a
1
a
2
. . . a
s1
a
s

[a
1
+ +a
s
]!
[a
1
]! [a
s
]!
Then the q-multinomial is a Laurent polynomial with positive, integer coecients.
Proof of 3.5. If s = 2 the statement is a direct consequence of the fact that, if yx = q
2
xy are two
skew-commuting variables, then:
(x +y)
n
=
n

j=0
q
n(n+1)
2

j(j+1)
2

(nj)(nj+1)
2

n
j

x
j
y
nj
which is easily proved by induction. The general case follows by induction on s by remarking that:

a
1
+ +a
s
a
1
a
2
. . . a
s1
a
s

a
1
+ +a
s
(a
1
+a
2
) . . . a
s1
a
s

a
1
+a
2
a
1
a
2

To prove the last statement, remak that we just proved that each state in the state-sum 3.5
3.1. Examples and properties. The following examples can be proved by re-normalizing the
formulas provided in [10] for the standard skein invariants of tetrahedra and -graphs.
Example 3.6 (Unknot).
(16)
a
)) = (1)
2a
[2a + 1]
Example 3.7 (Theta graph).
(17)
a
b
c
)) = (1)
a+b+c

a +b +c + 1
a +b c, b +c a, c +a b, 1

INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 13

Example 3.8 (The tetrahedron or symmetric 6j-symbol).
(18)
a
b c
d
e f
)) =
z=MinQj

z=MaxTi
(1)
z

z + 1
z T
1
, z T
2
, z T
3
, z T
4
, Q
1
z, Q
2
z, Q
3
z, 1

where T
1
= a +b +c, T
2
= a + e + f, T
3
= d + b + f, T
4
= d + e + c, Q
1
= a +b +d +e, Q
2
=
a +c +d +f, Q
3
= b +c +e +f.
From now on we will denote respectively
a
,
a
b
c
and
a
b c
d
e f
the values provided respectively
in Formulas 16, 17 and 18 and often drop the )): all the invariants we will be dealing with will
be the normalized version unless explicitly stated the contrary.
Example 3.9 (The crossed tetrahedron).
(19)
a
b
c
d
e
f
=
a
b
c
d e
f
= a
b
c d
e
f
=

1
2(f+ceb)
q
f
2
+f+c
2
+cb
2
be
2
e
e
f a
b
c d
The following lemma can be easily proved by starting from the analogous statements for the
standard skein theoretical normalization of the invariants:
Lemma 3.10 (Properties of the renormalized invariant). The following are some of the properties
of G, col)):
(1) (Erasing 0-colored strand) If G

is obtained from (G, col) by deleting a 0-colored edge, then

G

, col

)) = G, col)).
(2) (Connected sum) If G = G
1
#G
2
along an edge colored by a, then

a
a
G
1
G
2 )) =
1
(1)
2a
[2a + 1]
a a
G
1
G
2 ))
(3) (Whitehead move) If G and G

dier by a Whitehead move then:

a
b
c
d
j
)) =

i
i
a
b i
c
d j
a
i
d
b
i
c

a
b c
d
i ))
where i ranges over the all the admissible values.
(4) (Fusion rule) In particular, applying the preceding formula to the case j = 0 one has:

a b
)) =

i
i
a
i
c

a
a
b
b
i ))
4. Shadow state sums and integrality
In this section we will rst provide a so-called shadow-state sum formula for the invariants
G, col)) and give some examples. Then we will discuss some integrality issues and in Subsection
?? will provide short proofs of known identities on 6j-symbols.
14 COSTANTINO
4.1. Shadow state sums. Let G, col be a xed colored graph, D R
2
be a diagram of it and
V, E be the sets of vertices and edges of G, and C, F the sets of crossings and edges of D (each edge
of D is a sub-arc of one of G therefore it inherits the coloring from col). Let the regions r
0
, . . . r
m
of D be the connected components of R
2
D with r
0
the unbounded one; we will denote by R the
set of regions and with a slight imprecision we will say that a region contains an edge of D or a
crossing if its closure does.
Denition 4.1 (Shadow-state). A shadow-state s is a map s : R F
N
2
such that s(f) equals
the color of the edge of G containing f, s(r
0
) = 0 and whenever two regions r
i
and r
j
contain an
edge f of D then s(r
i
), s(r
j
), s(f) form an admissible three-uple.
Given a shadow-state s, we can dene its weight as a product of factors coming from the local
building blocks of D i.e. the regions, the edges of D, the vertices of G and the crossings. To dene
these factors explicitly, in the following we will denote by a, b, c the colors of the edges of G (or
of D) and by u, v, t, w the shadow-states of the regions and will use the examples given in Section
3.1.
(1) If r is a region whose shadow-state is u then w
s
(r)
u
(r)
(where (r) is the Euler
characteristic of r).
(2) If f is an edge of D colored by a and u, v are the shadow-states of the regions containing it
then w
s
(f)
u
a
v
(f)
where (f) = 0 if f is a closed component of D and 1 otherwise.
(3) If v is a vertex of G colored by a, b, c and u, v, t are the shadow-states of the regions
containing it then w
s
(v)
a
b c
u
v t
.
(4) If c is a crossing between two edges of G colored by a, b and u, v, t, w are the shadow-states
of the regions surrounding c then w
s
(c)
a
u
v
b
t
w
.
From now on, to avoid a cumbersum notation, given a shadow-state s we will not explicit the
colors of the edges of each graph providing the weight of the local building blocks of D as they are
completely specied by the states of the regions and the colors of the edges of G surrounding the
block. Then we may dene the weight of the shadow-state s as:
(20) w(s) =

rR
(r)

fF
(f)

vV

cC
Then, since the set of shadow-states of D is easily seen to be nite, we may dene the shadow-state
sum of (G, col) as
shs(G, col)

sshadow states
w(s)
As the following theorem says, the shadow state-sums provide a dierent approach to the compu-
tation of G, col)):
Theorem 4.2. It holds: G, col)) = F(G, col)shs(G, col), where (as in the preceding sections)
F(G, col)

eedges

1
4gecol(e)
q
2ge(col(e)
2
+col(e))
.
The original denition of the shadow state sums and proof of the above result (but for the
standard normalization of the invariants) is due to Kirillov and Reshetikhin ([8]) and was later
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 15
generalized to general shadows by Turaev ([13]). We used this formulation to extend the denition
of colored Jones polynomials to the case of graphs and links in connected sums of copies of S
2
S
1
([3]) and to prove a version of the generalized volume conjecture for an innite family of hyperbolic
links called fundamental hyperbolic links ([2]). These links were already studied in [4] for their
remarkable topological and geometrical properties.
Proof of 4.2. Multiplying by the factor F(G, col)
1
we can reduce to the case when the framing
of G is the blackboard framing. Let D be a diagram of G; we can add to G some 0-colored
edges cutting the regions of D (except r
0
) into discs (this changes G and D but not the value
of the resulting invariant); for each region we will need (r) 1 such arcs. Fix also a maximal
connected sub-tree T in D and let o R
2
be a 0-colored unknot bounding a disc containing D; it
is clear that G, col)) = (G, col)
0
)). Let also A be the trivalent graph dened as follows:
A (N(T) D) N(T) (where N(T) is the regular neighborhood of T in R
2
). The idea of
the proof is to apply a sequence of fusion rules and inverse connected sums in order to express
G, col)) as a

coli
c(col
i
)A, col
i
)) for some colorings col
i
of A and coecients c(col
i
); then to
show that each summand c(col
i
)A, col
i
)) is the weight of exactly one shadow-state. The unknot o
is isotopic to N(T) and, while following the isotopy, at isolated moments it will cross some edges
of D (N(T) D) (but no vertices or crossings because they are all contained in N(T)). Let us
choose the isotopy so that every edge of D (N(T) D) is crossed exactly once (this can be done
since each region is a disc and T is a maximal connected sub-tree of D). We say that u enters a
region r if during the isotopy a subarc of u not contained in r crosses an edge contained in r. We
claim that, since T, is connected each region r
i
, i = 1, . . . n will be entered by o exactly once
during the isotopy. Indeed if u enters twice a region r
i
, let , the subarcs of o in r
i
, connecting
them by an arc we may produce two unknots whose connected sum is u. Since T is connected and
contains all the vertices and crossings of D, one of the two discs bounded by these unknots cannot
contain not contain vertices and so and cross the same edge of r
i
, against our hypothesis on
the isotopy.
We interpret each crossing moment as a fusion rule
so that the isotopy of u progressively erases each
arc of D(N(T) D) exactly when entering a region
containing that arc, and the sum is taken over all
admissible u:
a b
=

u
u
a
u
b
a b
u
At the end of this isotopy, since for each i n u entered r
i
only once, all the components of
N(T) r
i
(which are arcs) will be colored by the same colors u
i
(see Figure 4 for an example of
this construction in the case of a planar graph). Therefore each summand in the nal expression
will be associated to a shadow-state s given by s(r
0
) = 0, s(r
i
) = u
i
. Moreover, the other edges of
A (i.e. those of N(T) D) are included in those of G and therefore inherit the coloring col. Then
we proved the following equality:
G, col)) =

u1,...un

rR
u
i
fF\T
1
A, col u
1
, . . . u
n
))
where the colors of the edges of the graphs are specied by the u
i
s and col. Remark that the
summation range is exactly the set of shadow-states because the colors of the arcs of N(T) r
i
are all u
i
and the admissibility conditions for a three-uple of colors around an edge are satised
at every moment we apply the fusion rule. Moreover, in the above formula we already got part of
the weights of each shadow-state (i.e. those of the regions and of all the edges out of T). We are
16 COSTANTINO
Figure 4. On the left a planar graph G to which we apply the construction of
the proof of Theorem 4.2. In the middle the graph A constructed by shrinking
on a maximal subtree of G an unknot bounding a disk containing G (the dotted
parts are left just for reference). On the right the nal union of planar tetrahedra
(in this case there are no crossings).
left to prove that what is missing equals A, col u
1
, . . . u
n
)), i.e. we claim the following:
A, col(u
1
, . . . u
n
))) =

fT
1

vV

cC
where in each factor the colors are specied by the combinatorics of D and by the state u
1
, . . . u
n

col on R F. To prove this, remark that:

a
b
u
)) =

i
i
a
b
i

i
a
b
u
)) =
a
b
c
1

a a
b b
u u
))
where the rst equality is a fusion rule and the second is the inverse of a connected sum. Applying
this identity on all the edges of T we split A, col u
1
, . . . , u
n
)) into the product of the graphs
remaining in the neighborhoods of the crossings and vertices (which are respectively and
) divided by the product of the s corresponding to the edges of T. This proves the
claim and completes the proof when all the regions are discs. If in the beginning we added some
0-colored edge to G to cut D into a diagram D

whose regions are discs, then it is clear that every

shadow state s

of D

can be lifted to a unique shadow state s of D: indeed the compatibility

conditions around a 0-colored edge force the states of the neighboring regions to be the same.
Moreover, since the only dierence between D

and D is given by the presence of the 0-colored

edge, it holds:
w(s) = w(s

fF0
u
u
0
1
= w(s

fF0
u
1
= w(s

rR
u
(r)1
This concludes the proof. 4.2
The formula given by Theorem 4.2 is often re-written by means of the so-called gleams:
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 17
Denition 4.3 (Gleam). The gleam of a diagramD of G is a map
g : R
Z
2
which on a region r
i
equals the sum over all the sectors
of crossings contained in r
i
of
1
2
times the local contributions of
the crossing determined according to the pattern on the right.
+ +

Remark 4.4. Do not confuse a shadow state with the gleam. In general for a given diagram D
there is a unique gleam but many dierent shadow-states.
Then one can re-state Theorem 4.2 as follows:
Corollary 4.5. Under the same hypotheses as Theorem 4.2, it holds:
(21) G, col)) = F(G, col)

rR

1
4g(r)u
q
2g(r)(u
2
+u)
u
(r)

fF
(f)

V C
Proof of 4.5. Using Example 3.9 one can rewrite the factors coming from crossings in term of
tetrahedra multiplied by extra factors of the form

1
2u
q
(u
2
+u)
for each of the 4 sectors around
a crossing. To conclude, collect these factors according to the region containing the corresponding
sectors and compare with Denition 4.3. 4.5
4.2. Simplifying formulas. Both formulas 21 and 20 are far from being optimal: indeed most of
the factors in the state-sum can be discarded from the very beginning. Instead of giving a general
theorem for doing this let us show why this happens through some examples. We will say that an
edge, vertex or crossing is external if it is contained in r
0
and a region is external if its closure
contains an external edge. Then the following simplications can be operated:
(1) If an external region contains two distinct external edges whose colors are dierent, then
G, col)) = 0
(2) If an external region r contains external edges f
1
, . . . , f
k
whose colors are all c, then, for ev-
ery shadow-state s on D, the total contribution coming fromrf
1
, . . . f
k
is
c
(r)
P
i
(fi)
(3) If an external vertex v is the endpoint of a non-external edge f then their contribution
simplify because
a
b c
c
0 a
=
a
b
c
(beware: if f is composed of two external vertices only
one of them can be simplied with f)
(4) For the same reason, if D contains a sequence
a b r
0
of n Z 0 half-twists
separating r
0
from a region r, then, in each summand of formula 21 the total contribution
of the crossings and internal edges of the twist (excepted the initial and nal ones) is
a
b
u
where u is the state of r and a, b are the colors of the strands (beware: the power of q
coming from the gleams do not simplify).
4.3. Examples and comments on integrality. According to Theorem 3.3, G, col)) is a Lau-
rent polynomial, and this is proved is by showing that the weight of each state in the state-sum
expressing it via R-matrices and Clebsch-Gordan symbols is a Laurent polynomial. Surprisingly
enough, this is not true for shadow-state sums: the weight of a single shadow-state may be a rational
function, but the poles of these functions will cancel out when summing on all the shadow-states.
We will now clarify this by explict examples.
18 COSTANTINO
4.3.1. Complicated formulas for unlinks. Consider the n-colored unnormalized Jones polynomials
of the following unlink:
J
n
(
1
2
1
2
-1
) =

1
2n
q
4(n
2
+n)
2n

u,v=0

1
2(u+v)
q
(u
2
+u)+(v
2
+v)
u v
n
n u
n
n v
n
n
u
n
n
v
In the picture we included the gleams of the regions to help the reader recovering formula 21. Of
course this is a very complicated way of re-writing (1)
2n
[2n+1]
2
q
2(n
2
+n)
, but what is interesting
is that the single states are again not Laurent polynomials: for instance when n = u = v = 1 the
weight is
[3]([5]1)
[2][4]
.
4.3.2. A more complicated link example. Fix a, b N and consider the n-colored unnormalized
Jones polynomials of the following link:
J
n
(
a
b
) =
2n

u,v=0

1
2(a+1)u+2(b+1)v4n(a+b+1)
q
g(u,v)
u v
n
n u
n
n v
n
n
u
n
n
v
where a box with a Z stands for a sequence of a half twists and g(u, v) (a +1)(u
2
+
u) + (b + 1)(v
2
+v) 2(a +b + 1)(n
2
+n). To get the above formula, before applying 21, remark
that up to isotopy of the diagram, the region r
0
can choosed freely. Therefore, it is better to pick
r
0
as the region touching the two boxes contemporaneously. Again, the summands are not Laurent
polynomials but the sum is (take for instance a = b = n = u = v = 1). More surprisingly, by
Theorem 3.3, for every a, b 0 the resulting invariant will be divisible by [2n + 1] in Z[q, q
1
].
4.3.3. Some planar graphs. If G is a planar graph equipped with the blackboard framing, then
the gleam of its regions are 0 and so by Corollary 4.5 G, col)) has a simple expression. In the
example of Figure 4, if all the edges of G are colored by n N then
(22) G, n)) =
2n

u,v=0,|uv|<n<u+v
u v
n
n n
u
n n
2
n
n n
v
n n
2
n
n n
u
n v
2
n
n
u
3
n
n
v
3
u
n
v
where u, v are the shadow-states of the two internal regions. If for instance in Formula 22 one
puts n = 1, and considers the shadow-state with u = v = 1 then its weight is
[3]
2
([4]!([5]1))
6
([4]!)
7
=

[3]
2
([5]1)
6
[4]!
/ Z[q].
4.4. Identities on 6j-symbols. Shadow state formulas provide a straightforward way to re-prove
standard identities on 6j-symbols. (The normalization we are using here for the symbols is that of
Example 3.8).
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 19
4.4.1. Normalizations of 6-symbols. It holds:
(23)
b,0
a c
=

u
u
a
a b
c
c u
a
c
u
where u ranges between [a c[ and a + c. This is proved by applying Formula 21 to:
a
b
c
and recalling that the invariant of a union of two unlinked graphs connected by a single arc is zero
unless the color of the arc is 0, in which case the invariant is just the product of the invariants of
the graphs. Similarly, it holds:
(24)
a
b
c

1
4b
q
2(b
2
+b)
=

1
2(u+a2c)
q
u
2
+u+a
2
+a2(c
2
+c))
u
a
b c
u
b c
b
c
u
where u ranges between [b c[ and b +c. This is proved by applying Formula 21 to
a b c
and
to the same framed graph after undoing the kink on the b-colored edge.
4.4.2. Orthogonality relation. A direct corollary of Formula 21 is the well-known orthogonality
relation:

b,d
a
b
c
e
f
b
b
=

1
2(db)
q
d
2
+db
2
b
u
a
b c
f
u e a
c d
f
e u
a
e
u
c
f
u
where u ranges over all the admissible colorings of the union of the two tetrahedral graphs on the
right. To prove it, just apply formula 21 to the following two isotopic graphs and simplify the
common factors:
a
b
c
d
e f

a
b
c
d
e
f
4.4.3. Racah identity. It holds:
(25)

1
2(a+bc)
q
a
2
+a+b
2
+bc
2
c
a
b c
d
e f
=

1
2(u+fed)
q
u
2
+u+f
2
+fd
2
de
2
e
u
a
e f
b
d u a
c b
e
u d
a
b
u
b
d
u
where u ranges over all the admissible colorings of the union of tetrahedra on the right. Indeed
it is sucient to apply Formula 21 to the following two isotopic graphs and simplify the common
20 COSTANTINO
factors:
a
b c
d
e
f

a
b c
d
e
f
4.4.4. Biedenharn-Elliot identity. Another direct corollary of Formula 21 is the well-known Biedenharn-
Elliot identity:
(26)
a
b c
d
e f g
h e
c
d i
c
e
d
=

u
u
a
e f
g
u h d
b f
u
g i a
b c
i
h u
a
h
u
b
i
u
f
g
u
where u ranges over all the admissible colorings of the union of tetrahedra on the right. Indeed
it is sucient to apply Formula 21 to the following two isotopic graphs and simplify the common
factors:
a
b
c
d
e
f
g h
i

a
b
c
d
e
f
g h
i
5. R-matrices vs. 6j-symbols
In the preceding sections we showed to compute invariants of colored graphs by means of two
dierent state-sums. Although for practical computation shadow state-sums turn out to be easier
to deal with, state-sums based on R-matrices and Clebsch-Gordan symbols allow one to prove
integrality results. In this section we will compare the state-sums when applied to graphs with
boundary. So let a (n, m)-KTG be a framed graph embedded in a square box which contains only
3-valent vertices (inside the box) and n (resp. m) 1-valent vertices on the bottom (resp. top) edge
of the box. A typical example is a framed (n, m)-tangle. Let as before E, V be the set of edges
and 3-valent vertices of G, G

G box and, once chosen a diagram D of G, let F, C be

the set of edges, and crossings of D. Let also b
1
, . . . b
n
be the bottom (univalent) vertices of G
and t
1
, . . . t
m
the top vertices. The denition of admissible coloring of a (n, m) is the same as the
standard one, but in this case, a second coloring is needed to get a numerical invariant out of G,
namely a coloring on G.
Denition 5.1 (-colorings for (n, m)-KTGs). Let G, col be a colored (n, m)-KTG; a -coloring
for G is a map col

: G
Z
2
such that if i
k
(resp. j
k
) is the color of the edge containing b
k
(resp.
t
k
) then [col

(b
k
)[ i
k
and col

(b
k
) i
k
Z.
Equivalently a -coloring is a choice of a vector in V
i1
V
in
of the form e
i1
col

(b1)

e
in
col

(bn)
and a vector in V
j1
V
jm
of the form e
j1
col

(t1)
e
jm
col

(tm)
. Given a (n, m)-KTG
equipped with a coloring col and a -coloring col

, one can compute G, col col

)) exactly as in
Formula 13: it is sucient to restrict the set of admissible states to those such that the state of the
boundary edges coincide with col

. Then G represents a morphism Z(G, col) : V

i1
V
in

V
j1
V
jm
and G, col col

)) is an entry in the matrix expressing Z(G, col) in the bases

formed by tensor products of basis elements.
The integrality result 3.3 still holds true (the idea of the proof is exactly the same):
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 21
Theorem 5.2 (Integrality: case with boundary). The following belongs to Z[q

1
2
]:
G, col col

)) G, col col

)
F(G, col)

eE
[2col(e)]!

vV
[a
v
+b
v
c
v
]![a
v
+c
v
b
v
]![c
v
+b
v
a
v
]!

m
k=1

1
j
k

n
k=1

1
i
k
where E

is the set of all the edges of G which do not intersect G

+
and F(G, col) is dened as in
the preceding sections.
What is interesting is that one can re-compute the invariant of (G, col) also via shadow-state
sums and Clebsch-Gordan symbols. To explain this, we will use the following:
Denition 5.3. Given a nite sequence j
1
, . . . , j
m
a Bratteli sequence associated to it is a sequence
s
0
, s
1
, . . . s
m
such that s
0
= 0 and for each 0 k m1 the three-uple s
k
, j
k
, s
k
+1 is admissible.
It is not dicult to realize that the set of Bratteli sequences associated to j
1
, . . . , j
m
is in bijection
with the set of irreducible submodules of V
j1
V
jm
and that the submodule V (s) corresponding
to a Bratteli sequence s = (s
0
, . . . s
m
) is isomorphic to V
sm
. Moreover, using the Clebsch-Gordan
symbols dened in Subsection 2.2.4, we may x explicit maps (s) : V
j1
V
jm
V
sm
and
i(s) : V
sm
V
j1
V
jm
. Graphically, (s) =
s
2
sm
j
1
j
2 j
3
jm


and i(s) =
sn
s
2
j
1
j
2
j
3
jm


.
More explicitly, to construct (s), apply the Clebsch-Gordan morphism P
s1
j1,j2
to the rst two
factors of V
j1
V
jm
(the isomorphism is xed by the choice of Clebsch-Gordan projectors
made in Subsection 2.2.4). The result is in V
s2
V
j3
V
jm
and composing iteratively with
with P
s
k+1
s
k
,j
k+1
one gets the seeked map (s) : V
j1
V
jm
V
sm
. Similarly, i(s) : V
sm

V
j1
V
jm
is dened by composing recursively from right to left the Clebsch-Gordan morphisms
of Subsection 2.2.4 V
sm
V
sm1
V
jm
V
sm2
V
jm2
V
jm
. . . V
j1
V
jm
.
Let G G box viewed as a framed graph in R
2
by embedding the box in R
2
with the
blackboard framing around its boundary. Let s
+
1
, . . . s
+
m
and s

1
, . . . s

m
be Brattelli sequences
associated to j
1
, . . . , j
m
and i
1
, . . . , i
n
and suppose that s
+
m
= s

n
x. We can extend col to a
coloring col s

s
+
of G: the color of the edge in the top (resp. bottom) edge of the box bounded
by j
k
and j
k+1
(resp. i
k
and i
k+1
) is s
+
k
(resp. s

k
), the color of the left edge of the box is 0 and
that of the right edge x (see Figure 5).
Theorem 5.4 (Shadow state-sums vs. R-matrices and Clebsch-Gordan symbols). Let (s

, s
+
)
be the coecient such that the morphisms (s
+
) Z(G, col) i(s

) : V
x
V
x
and (s

, s
+
)Id
V
x
coincide. Then:
(s

, s
+
)
x
= G, col s

s
+
))
and
Z(G, col) =

s
+
n1

t=1
st
it
st
s
t+1
m1

t=1
st
jt
st
s
t+1
G, col s

s
+
))
sn
s
+
2
s

2
s

n
s
+
m
i
1
i
2 i
3
in
j
1
j
2
j
3
jm




.
Proof of 5.4. The value on the right hand side is the invariant of the colored graph (G, cols
+
s

)
depicted on the right of Figure 5. But, as shown in the picture, G is the closure of the graph
representing (s
+
) Z(G, col) i(s

).
22 COSTANTINO
s
+
1
s
+
1
s

1
s

1
s
+
m1
s
+
m1
s

n1
s

n1
x
x
0
0
G
G
G




Figure 5. Using a pair of Bratteli sequences it is possible to close (G, col) to a
colored closed KTG (in the middle).
The second statement follows by applying n 1 times the fusion rule on the bottom legs of G
and m1 times on the top legs to get:
G


=

s
+

n1
t=1
st
it
st
s
t+1

m1
t=1
st
jt
st
s
t+1
s
+
2
s

2
s

n
s
+
m
i
1
i
2 i
3
in
j
1
j
2
j
3
jm
G

where s

and s
+
range over all Bratteli sequences associated to i
1
, . . . , i
n
and j
1
, . . . , j
n
respectively
and G

is the (1, 1) colored KTG obtained by opening G, col s

+
s

along the right edge of the

box. 5.4
Example 5.5 (R-matrices vs 6j-symbols). It holds:
a
b
R
t,w
u,v
=
a+b

c=|ab|
C
b,a,c
t,w,v+u

1
2(cab)
q
c
2
+ca
2
ab
2
b
c
a
b
c
P
a,b,c
u,v,u+v
Indeed it is sucient to apply the preceding theorem to (G, col) being a crossing.
References
1. J. S. Carter, D. E. Flath, M. Saito, The classical and quantum 6j-symbols, Math. Notes, Princeton
University Press (1995).
2. F. Costantino, 6j-symbols hyperbolic structures and the Volume Conjecture, Geom. Topol. 11, (2007),
1831-1853.
3. F. Costantino, Colored Jones invariants of links in #S
2
S
1
and the Volume Conjecture, J. London Math.
Soc. (2) 76, (2007), no 1, 1-15.
4. F. Costantino, D. P. Thurston, 3-manifolds eciently bound 4-manifolds, J. Topology (1), 3, (2008).
5. S. Garoufalidis, R. Van der Veen, Asymptotics of classical spin networks, www.arXiv.org/0902.3113
(2009).
6. L. Kauffman, S. Lins, Temperley-Lieb recoupling theory and invariants of 3-manifolds., Annals of Mathe-
matics Studies, 134. Princeton University Press, Princeton, NJ, 1994.
INTEGRALITY OF KAUFFMAN BRACKETS OF TRIVALENT GRAPHS 23
7. R. Kirby, P. Melvin, The 3-manifold invariants of Witten and Reshetikhin-Turaev for sl(2, C), Invent. Math.
105, 473-545 (1991).
8. A.N. Kirillov, N.Yu Reshetikhin, Representations of the algebra Uq(sl
2
), q-orthogonal polynomials and
invariants of links, Innite dimensional Lie Algebras and Groups (V.G. Kac, ed.), Advanced Ser. In Math.
Phys., Vol. 7, 1988, 285-339.
9. T.Q. Le, Integrality and symmetry of quantum link invariants, Duke Math. J. 102 (2000), no. 2, 273306.
10. G. Masbaum , P. Vogel, Three-valent graphs and the Kauman bracket, Pac. J. Math. Math. 164, N. 2
(1994).
11. S. Piunikhin, Turaev-Viro and Kauman-Lins invariants for 3-manifolds coincide. J. Knot Theory Ramica-
tions 1 (1992), no. 2, 105135.
12. N.Yu. Reshetikhin, V. G. Turaev, Ribbon graphs and their invariants derived from quantum groups.
Comm. Math. Phys. 127 (1990), no. 1, 126.
13. V. G. Turaev, Quantum invariants of knots and 3-manifolds, de Gruyter Studies in Mathematics, Vol. 18,
Walter de Gruyter & Co., Berlin, 1994.
Institut de Recherche Mathematique Avancee, Rue Rene Descartes 7, 67084 Strasbourg, France
E-mail address: costanti@math.u-strasbg.fr