You are on page 1of 84

J. Scott Carter Daniel E. Flath

and

i~lasahicoSaito

Mathematical Notes 43

1

Introduction

2

Representations of U(sZ(2))

Basicdefinitions

7

Finite dimensional irreducible representations 7

Diagrammatics of U(sl(2)) invariant maps

12

The Temperley-Lieb algebra

15

Tensor products of irreducible representations

21

The 6j-symbols

27

Computations

43

h rccu~sio~lformula for the 61-symbols

63

Iteillh~hs

3

Quantum sZ(2)

67

Some finite dimensional representations

67

Representations of the braid groups

70

A finite dimensional quotient of C[B(n)]

74

The Jones-Wentzl projectors

80

The quantum Clebsch-Gordan theory 93

Quantum network evaluation

99

The quantum 6j-symbols - generic case

106

Diagrammatics of weight vectors (quantum case)

110

Twisting rules

111

Symmetries

123

Further identities among the quantum Gj-symbols

125

I

4

The Quantum Trace and Color Representations 127

The quantum trace

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

127

 

.

.

.

.

.

.

.

.

.

.

.

.

130

 

A bilinear form on tangle diagrams Color representations

The quantum Gj-symbol-

.

.

.

.

.

.

.

,

.

.

.

.

.

.

.

root of unity case .

.

.

.

.

.

.

.

.

.

133

139

5

The Turaev-Viro Invariant

 

The definition of the Turaev-Viro invariant

 

151

Epilogue

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

157

 

151

This book discusses the representation theory of classical and

quantum U(d(2))with an eye towards topological applications

of the latter. We use the Temperley-Lieb algebra and the Van-

I tum spin-networks to organize the computations. We define the

6j-symbols in the classical, quantum, and quantum-root-of-unity

cases, and use these computations to define the Turaev-Viro in-

variants of closed 3-dimensional manifolds. Our approach is ele-

References 160 mentary and fairly self-contained. We develop the spin-networks
i

I

I

from an algebraic point of view.

1 Introduction

These notes grew out of a series of seminars held at the University of South Alabama during 1993 that were enhanced by regular e- mail among the three of us. We became interested in quantum diagrammatic representation theory following visits from Ruth Lawrence and Lou Kauffman to Mobile.

We develop the Clebsch-Gordan theory and the recoupling the-

ory for representations of classical and quantum U(sl(2)).via the spin networks of Penrose 1271 and Kauffman [16]. In these the- ories, the finite dinlensio1la.1irreducible representa.tions are real-

i~edill spaces oi i~oluogcr~eoil:,l,oi> i~oi~ii,il\111 1\\(,

\ .I: I.II)I~~~TII

the quantum case the variables commute up to a factor of q; z.e. yx = qxy. The tensor product of two representations is decom- posed as a direct sum of irreducibles, and the coefficients of the various weight vectors are computed explicitly. In the quantum case, when the parameter is a root of unity, we only decompose the representations modulo those that have trace 0.

We use the spin networks to develop the theory in the classi- cal case for two reasons. First, they simplify and unify many of the tricky combinatorial facts. The simplification of the proofs is nowhere more apparent than in Theorem 2.7.14 where a plethora of identities is proven via diagram manipulations. Second, the spin networks are currently useful and quite popular in the quan- tum case (see for example [23], [18], [28]). One of our goals here is to explain the representation theory of quantum sl(2) in the spin network framework. We know of no better explanation than

-

--

--I-I-

=I 1I1

I

I.u

xu,II, I UIYI VJ-31 IVI,,"ba

IN L~UUU~IIVIV

on triangulations of a 3-manifold while the orthogonality condition

I and then to imitate the classical theory in the quantum case. can be interpreted as a Matveev [25] move on the dual 2-skeleton

I Here we give an overview.

determinant 1 over the complex numbers forms a group called The Turaev-Viro invariants were based on work of Kirillov
I

1

to run through the classical case (which should be more familiar),

The set

of

(2 by

2) matrices of

of a triangulation.

SL(2). The finite dimensional irreducible representations of SL(2) are well understood. In particular, it is known how to decompose

and Reshetikhin on the representation of quantum groups [19].

This work together with Reshetikhin-Turaev [29] formed a math-

I

I

I

I

ematicdy rigorous framework for the invariants of Witten [34].

Meanwhile Kauffman and Lins [18] gave a simple combinatoric

approach to the invariants based on the Kauffman bracket and

of the so-called Clebsch-Gordan theory. The finite dimensional the spin networks of Penrose [27]. Piunikhin [28] showed that the

representations of SL(2) are the same as those of U(sl(2)) which

is an algebra generated by symbols E, F and H subject to certain Some of Kauffman's contributions to the subject can also be relations.

Kauffman-Lins approach and the Turaev-Viro approach coincide.

the tensor product of two such representations into a direct sum

of irreducibles. In this decomposition one can compute explicitly

the image of weight vectors and such computations form the heart

found in the papers [14], [15], and [17]. A more traditional alge-

braic approach to quantum groups can be found in [30]; in partic-

ular, they discuss from the outset the Hopf-algebra structures.

Ticl;ol.isll'q [23] d(\finition of tlic Rt~sl~ctiklii~l-~~~.aoviilval inlltb

Furthermore, the tensor product of three representations can

The comparison of these

rrrn~~,,l?riqt/)ro,~ll.clllfl

be decomposed in two natural ways.

two decomytosit ions

is sornctimcs c,rllctl

-p

I*

$1

tile rccou1)llllg t uell~c~erit:,,ir c. I,I~o\LI~ LJ~CG~'-~~~l~]~~l~,rlllese

1s of a colllbi~~ator~alnature. The I<auiinian-L~ns[IS] delin~t~o~l01

the Turaev-Viro invariants is defined similarly. Neither of these

combinatorial approaches relied on representation theory. How-

ever, the remarkable feature of quantum topology is that there are

close connections between algebra and topology that were hereto-

symbols satisfy two fundamental identities (orthogonality and the

Elliott-Biedenharn identity) that can be interpreted in terms of

the decomposition of the union of two tetrahedra. In the ~lli~tt-

Biedenharn identity the tetrahedra are glued along a single face

and recomposed as the union of three tetrahedra glued along an fore unimagined. The purpose of this paper is to explore these

l edge- For orthogonality the tetrahedra are glued along two faces, relations by examining the algebraic meaning of the diagrams and

It and the

symmetry of the 6j-symbols and their relationship to Here is our outline. Section 2 reviews the classical theory of

11 M~ahedrawas for the most part a mystery, until &-aev and of U(sZ(2)). There is nothing new here, but we

do

are computed in terms of the bracket expansion (at A = 1). In

Section 3 we mimic these constructions to obtain the quantum

Clebsch-Gordan and 6j-symbols. In Section 4 we will define the

recomposition is not simplicial.

by using diagrams to prove algebraic results.

I The

I

I/ I

Viro 1321 constructed 3-manifold invariants based on the analo-

gous theory for quantum sZ(2).

The

Gj-symbols are also satisfied by their

identities satisfied by the

quantum analogues. The

Elliott-Biedenharn identity corresponds to an Alexander [I] move

how the Clebsch-Gordan coefficients and the 6j-symbols

/'i

I

I!

I/

i

i

I :(/I

'

CL-8

I

/

11

111

j

$1

"

r

VUnLJLILUfiU

fl1lU

WUXIY 1 UM

tJJ->YMB(JLS

quantum trace and discuss the recoupling theory in the root of

unity case.

invariants and proves that the definition is independent of the triangulation by means of the Pachner Theorem [26].

and

R. Lawrence for the interesting conversations that we have had.

Section 5 reviews the definitions of the Turaev-Viro

Acknowledgments.

We all are grateful to L. Kauffman

Their visits to Mobile were supported by the University of south

~~~bama'sArts and Sciences Support and Development Fund.

Additional financial support was obtained from Alabama EPSC~R for funding of travel for the first named author and support of

a Conference in Knot Theory, Low Dimensional Topology, and

Quantum Groups in Mobile in 1994.

Murasugi, and R. Peele provided us with valuable information. Jim Stasheff read a preliminary version of the text and provided

us with many helpful comments. Cameron Gordon's past finanical

support of Masahico Saito was greatly appretiated. Finall!. ive all

C. Pillen, B. Kuripta, K.

grdtcfull) dcl\nowledgc tile suppo~taild patle~~ccL~~ALotll wlbes have shown to us over the years.

I

2

Representations of U(sl(2))

2.1 Definition.

Let C denote the complex numbers.

The

group SL(2) is defined to be

SL(~)={(;

\

\

;) :a,b,c,dEC,

ad-bc=l

where the law of composition is matrix multiplication.

sociated Lie algebra sZ(2) consists of the set of matrices of trace

The as-

0:

,,(,)={(;

i) :a,b,c,di(:,

a+d=n

I -

This is spanned by E =

(

;),.=(

a=

-

y2) . The Lie bmdei is c~mputed via [A, Bl = AB -

\r!.~]=! >-(liu1-1----l7

13-1.,oti,at[1:.1]=~~i. (

-

exponential function, exp : sl(2) -+ SL(2), which is defined by the

power series:

"

~XPQ= j=o

Q3

for Q E sZ(2). The function exp maps a trace 0 matrix to a matrix

with determinant 1.

The

2.2- Finite dimensional irreducible representations.

group SL(2) acts on the vector space of linear combinations of.

variables x and y by

I

i

i

1 11 11

'

I

i

t

AND QUANTUM6j-SYMBOLS

( ;)y=cxdy

where the action is extended

linearly.

This is

called the funda-

mental representation of SL(2).

 

More generally, define an action of SL(2) on and 1

the space of poly-

= (ax

One way to verify that this is a group action is to consider the embedding

.

.,

Y and where the sum is over all

geneous ~ol~nomialsof degree (r +s) in x and y into the tensor

product of (r+ s) copies of the fundamental represelltation space.

The tensor product V 8 W of representations V and ]/V illherits

an action via g(v @W) = gv 8gw where v E

I' denotes the fundamental representation space, then v@(T+s)is

V and w E W. ~h~~ if

also arepresentation space. Furthermore, the image of the space of

~olynomialsconsists of the subspace of tensors that

fixed under the ! action of the permutation group on the tensor factors of VB('+"), and this space is stable under the

action of SL(2).

denote the set of homogelleous poly-

Note

that v1I2 is the fundamental representation, and V0is the trivial

nomials of degree 2j = r + s where j E

It is customary to let ~j

).

representation, VO = C. The index j is sometimes called the spin

of the representation V3. The associated Lie algebra d(2) acts on V38s ~~UOWS:

E~~~~= dl

dt t=o

exp (tE)xTys = srTt1ys-',

F~~yS = rxT-I ySC1,

and

HSTyS= - r-s

2

xTyS.

A weight vector is an eigenvector under the action of H in any

representation; its eigenvalue is called its weight. For example

xTyS E V(T+S)I~is a weight vector of weight y.Observe that

the set of weights in ~j is {j,j-

the corresponding weight vectors form a basis for VJ.

and by definition

,-j),

2.2.1 well ~lzownXheorenl.

(See [g] or [33], hi-c~lm-

The representations of SL(2) On V3 are irreducible.

proof.1f W is an SL(2)-subrepresentation of V3, then W is also

invariant under the action of the algebra sl(2) that is given above.

Therefore, it is enough to show that the representation of sl(2) On

vj is irreducible. The matrix E acts by sending a weight vector to one of higher

weight F sends such to one of lower weight. Since the image

of anv non-zero vector under powers of E and F spans V37 this

representation is

Remark. In the sequel, it will be more convenient to work with

the universal enveloping algebra U(sl(2)). This is an algebra gen-

erated by symbols E, F, and H that are subject only to the rela- tions EF - FE = 2H, HE - EH = E, and HF - FH = -F.

The relations are motivated by the properties of the Lie bracket

irreducible.

iir

8

10

REPRESENTATIONSOF U(sl(2))

~rof U(sl(2))is deter- mined by assigning to E, F, and H operators on a vector space

that are subiect to

the relations above. The enveloping algebra

-- - ----e v is a vector in the

.,

-

L

--

--

-

-

representation V.

In the discussion of Section 3, the representations VJ of d(2)

and the enveloping algebra U(sl(2))will have quantum analogues.

There is a quantum analogue of the group SL(2), but we will not

use it to describe the representations.

e3,m = 5J+m 3-m.

Y

should be indexed by

The first subscript of e is the highest weight of the represen-

tatloll

<111dind~taft5 tht tl~rnel~slonof thc rcprt~.;crltdfioll 5pr7(t\

I) = LJ $ 1) wlllle the sccond ~~ldlcatestile icelght of the

vector. Note that j and m are both half-integers and that j + m

and j - m are integers. In this notation,

(dl111jl

and

Ee3,m = (j - m)eJ,m+l,

Fej,m = (j +m)eJ,m-17

Hel,m = meJ,,.

2. The vector v has weight j.

Then j

E

),

4 : VJ+ W such that $(x2"

the action of U(sl(2)).

and there is a unique linear map

= v and such that .11, commutes with

(b) Every finite

dimensional

irreducible

U(sl(2))is isomorphic to VJfor some j E

representation

).

of

Proof.

Let vo = v, and for r > 0 let v,

= FTv.

We assume by induction that v, has weight (j - r).

Since

Hv = jv and [H,F] = -F,

FHv, = (j - (r + l))~,+~;thus vT+l has weight (j - (r +1)).

we have HV,+~= HFv, = -Fur

+

Furthermore, we inductively assume that there are constants

using the relation [E,F] = 2H,

y, such that Ev, = y,v,-1.

By

we have that EV,+~= EFv, = 2HvT+ FEU, = (2(j - r) +YT)~T.

= r(2j-r+ 1).

Hence ?,+I

= 2(j-r)+yT; since yo = 0, we have y,

Yo\\,,

0 fol ,on~cI

I,c~c,l~l\c.If* I? fnrte tl~mr.nsio~~,~\and

vo, vl,

that v,

are eigenvectors for distlnct elgenvalues of H. Suppose

So

the subspace of W generated

= 0 and v,-1

# 0. Then Ev,

),

= 0 = r(2j - r + l)v,-1.

j

= (r - 1)/2 E

by vo is spanned by the linearly

and this subspace is isomorphic to VJ.This proves (a).

independent vectors vo,

,v,-1,

(b) Let W denote a finite dimensional irreducible represertta-

tion of U(s1(2)),and let w be a non-zero eigenvector of H. Let the

integer r be such that E'w # 0 while E'+'w = 0. Then v = E'w

satisfies the

hypotheses of (a). Hence

W = VJ where Hv = jv.

2.2.3 Well Known Lemma. (a) Let W denote a finite 2.2.4 Theorem. Every finite dimensional representation space

for U(sZ(2))decomposes as a direct sum of irreducible representa-

dimensional representation space for the algebra U(sl(2)).Let W denote a non-zero vector that satisfies:

E

tions.

1.

Ev = 0.

I

Proof. See [33]or [8]for example.

111

The Pen-

rose spin networks facilitate the computation of U(sl(2))invariant maps via diagrammatic techniques. At the heart of the networks are three elementary maps U, n, and / that are defined in Sec- tion 2.3.1. Their relations are described in Lemma 2.3.2. The networks or spin-nets will consist of trivalent graphs embedded in the plane with non-negative half-integer labels on the edges.
P These labels will satisfy an admissibility condition at each vertex that will be made explicit as we continue the discussion.

i

2.3 Diagrammatics of U(sl(2))invariant maps.

I ,lit

2.3.1 Definition.

Consider the U(sI(2))invariant maps U :

170 +

~112,g, ~112, ,: v112 fg ~112+

vO, : v1/2 fg ~1124

I i lf 1/2fg171/2, and I : v1I2-+ v1I2that are defined on basis elements

I (and extended linearly) via

X(afgb)=b@a

where i = G.Finally

](a)= a

for

for

a,b€{x,y}

a E {x,y).

2.3.2 Lemma (Penrose [27]).

1. The maps U, n, X, and ( commute with the action of U(sl(2)).

2. The fundamental binor identity holds:

= I fg 1+ (U o n) : v1J28 v1I2 -+

v1I2 8 ~112:

3. (I fg n) 0 (U 8 1) = ( = (n 8 1) 0 (1 fg U) : v1I2-+ v1I2

under the identification of C ,g, v1I2= y1l2fg C = v1I2.

4.

(n 8

1) 0 (I fg 1) = (I @ n) 0

(1 @ 1) : (v1/2)@3

v1/2

where as before we identify C 8 '[fl/2 = ~1128 c = 1/1/2.

6.

/

/

flou:C-+C

is multiplication by -2.

Proof. Items (I), (2), (3) and (6) are elementary computations. Item (4) is a general property that holds for any bilinear. form n; similarly, item (5) follows for any "co-bilinear" form U.

2.3.3 Remarks. Penrose and I<auffman introduced these maps in a diagrammatic context. The domain of a map repre- sented by such a diagram appears at the bottom of the diagram,

I

(1

.-i

-

-z

,-

16 'THECLASSICALAND QUANTUM~~-SYMBOL,S

I

and

The key observation about the diagram algebra is that ~lallaf

hk H

-nL+

-

(

I

A -

I

21-L-1

I.

isotopies of arcs (that are properly and disjointly embedded ill

a rectangle) are generated by the topological moves that corre-

spend to (a) the relationship depicted in Lemma 2.3.2 (3) and (b)

example, in the illustla-

tloll above the Criticdlp01111~~el~le~el~tlllg111 Cdll 11~])~l~il~(/(I0'' "

:.,r c~lclldn~illgdisfant critical l>ointq.

Algebraically, this

The diagra7n algebra consists of formal linear combinations of cer- interchange represents the identity h1h3 = h3hl.

tain diagrams. The diagrams are generated by the diagralns rep- The correspondence between the Temperley-Lieb algebra and

resentillg I alld hk for k =

the algebra of diagrams shows that the dimension of TLn is the

and those representing h3 can be pushed up.

,2j

- 1 that are indicated above.

nth Catalan number, A(:), where n = 2j. This result follows

by establishing a one-to-one correspondence between the possible

diagrams and the collection of legitimate arrangements of n pairs

of parentheses (See also [16]).

Next we let the ground ring R denote ihe complex numbers,

be juxtaposed vertically to represent

the product of two of the elements. Having been so juxtaposed,

the product

angle. Two di%pms that are isotopic via an isotopy that keeps

the top and the bottom of the diagrams pointwise fixed represent

into a standard size rect-

two such diagrams can

is rescaled vertically to fit

the same element in the diagram algebra. For example, the prod-

uct

an ex~ressiollis at the top of a diagram; thus the bottom most 2.4.1 Lemma. Let j be a fixed element in

and choose matrix representations for the symbols U and n.

h2hlh3

is depicted below. Always the element on the left of

I'

I

18 THECLASSICALAND QUANTUM6j-sy~~o~~

REPRESENTATIONS

OF U(sz(2))

19

functions

roof. The proof depends on standard facts about the represen-

(~1/2)@2.1,(~1/2)@2.1

The image O(I) is the identity, and for k = 1,

We refer the reader to the excellent

text, [31] for details. Let n = 2j. Since dim (TLn) = Catalan(n)

we must prove that dim (B(TLn)) = (T)/(n +1)- The binor iden-

tity shows that B(TL,) = b(C[Cn]), where @ is the linear ex-

tension of the representation defined in the remark immediately

Proof. It is necessary to check that O respects the defining rela- above. We will establish that dim (@(C[xnI)) = (',") /(n + 1). The

representation 6 is decomposed as in [33]. Namely, as En repre-

ation theory of GL(2) and C,.

,2j

- 1, the gen-

erator hk is mapped to the composition Uon where

on the (k, k +1) factors of the tensor product as in Section 2.3.1.

these are acting

ill

I

I tions (1-5) of the Temperley-Lieb algebra. The calculatioll follows from the diagrammatics of Lemma 2.3.2 part (3). J-I

As we continue to discuss U(s1(2)), we will work with the

Temperley-Lieb algebra under this representation witllout explic-

Our justification for this notational

abuse is given in Theorem 2.4.3.

llill

I itly mentioning the map 8.

2.4.2 Lemma.

Let 6 =

-2.

fir

any j

E {0,1/2,1,3/2,. , .I,

i5

11

~J(~~I~oIIJO~~)/II~~~~(1 1 ,,,,l/c,/,o,j

qJ,J,,,,y2! ,,,,

2,

letters zlzto TL23 that 2s yzven by p(ak) = I + hk where ak zs the

transposition that interchanges

and k + 1.

Proof.

Clearly, the images (under p) of distant transpositions

, commute. Furthermore,

l1

/I1

p(ukokklok) = I+ hi +hxt~+hihi51 +hkil hk = p(okklokok*l1.

Finally, (It hk) 0 (Ithk) = 1 $ 2hk +6hk = 1.0

Remark. The binor identity shows that the 1~omomorphismp is

-a factor of the representation

: CZ3-+ AU~((V~/~)@~~)where the

-

permutation group acts on (V~/~)@~Jby permuting tensor factors.

2.4.3 Theorem.

In case 6 = -2,

the representation, B of TLz3

on (V~/~)@~J,is faithful for evert, i E -!0.1/2.1.3/?

1

(v'l2)@" = $T(m)d(T)

(1) the index T ranges over all 2-row Young frames with n

(i.e.

T = (r,s), n = r +s, and 0 L r

I s);

(2)the sul13$a~d\ITrr is the irrc~I,,cil,lc rel)resentatioll of

lllclt

corresponds to the toung lldllle 1 ;

(3) the exponent d(T) is a positive integer that, incidentally, is

equal to the di~nensionof the representation of GL(2) correspond-

ing to the Young frame T.

jyow c[xn] is a semi-simple algebra because the group En is

finite. The Wedderburn theory of semi-simple algebras [g], applied

to C[Cn], asserts that as algebras

I

- @(C[Cn])% $~Mat(n~X n~)

where nT

matrices. It follows that

= dim WT and Mat(- x -) denotes the algebra

dim (j(C[C,]))=

n;.

square

For T = (r, n - r) with 0 5 r _< Ln/2] (where 1.1 denotes the

greatest integer function), we have

This is the number of ways of filling in the n boxes in the Young

frame T with the integers 1,2,

increase across both rows and increase down all columns. The Young frame has n - r boxes on the top row and r boxes on the

in such a way that numbers

,n

bottom row.

The proof will follow from the following interesting combina-

torial identity for Catalan numbers:

Let

G(r, n - r)

= n(,,,-,)

denote the number of legitimate

filling.; of the Young frame nxilh I, I~n~c>inri f 11~1101 lor?] J.(IU. ;I nd

n - 1. boxes on the top. For a two ron. rectangular array,

so in fact G(r, r) is the rth Catalan number. that

We wish to show

Each term in the sum on the right is the square of the number

of ways of Wgg in a smaller

[n/2j, we consider the Young frame (n, n) to be decomposed as the union

of a frame (s, n - s) and its mirror image. For example,

Young frame.

For s

I

.

A filling of the frame (s, n - s) with the integers 1, .,n to-

ther with a filling of its mirror image with the integers n $

yields a filling of the rectangular frame. Therefore, the

m on the right is no larger than G(n, n).

On the other hand, let a filling of the rectangular frame be

iven. Then consider the subset of the rectangular array that

ontains the numbers 1, .,n. This subset is convex and forms

a smaller frame of type (s,n - s). Thus we have a filling of it

and a filling of its mirror image. Therefore, G(n, n) is no larger

than the sum on the right. This proves the combinatorial identity.

Consequently, the representation is faithful.

,2n

Re-

call that if V and W are spaces on which the group SL(2) acts,

then there is an action given on the tensor product by g(v @ tu) =

2.5 Tensor products of irreducible representations.

911

~IL)

\1i

element .Y in the assoriatctl 1,ic alge111.a. d(2).acts

on tensor protluc~svia the Leibniz rule, .Y(u LI ru) = St?)I , iu $

v @ X(w) since the action is determined by differentiation. Notice

that if v and w are

weight vectors, then so is v @ w, and its weight

is the sum of the weights of v and w. Recall that V3 is isomor-

phic to a sub-representation of the 2j-fold tensor product of the

, fundamental representation via the map

(P3 : XI . ' ' ' ' x2j ++

1

-

Cp

xu(1) @ . ' @ Xu(2j)

2.5.1 The projectors. The projection of (V1/')@'j onto the

image (Pj(Vj) can be written in terms of the Temperley-Lieb ele-

ments as the map

Observe that +2j

0+2j

= 42, So that this map is indeed a projec-

In

3-5, the quantum analogues of these projectors are de-

2.5.2 Definition.

Let

I

u= u : c

~112a ~112. H~~~~~

defined

C -+ (v"~)@~("-~),define 5 to be the composition

The map fi is defined dudy. Let

C (1,

,n}.

Define

2.5.3 Lemma.

proof. The proof follows by induction.

2.5.4 Definition.

j

E

{a + b,a + b -

Suppose that a, b E

{0,1/2,1,3/2,. Let

,la

- bl + 1,la - bl). Such a triple of

half-integers (a,b,j) is said to be admissible. Notice that admis-

sibility is a symmetric condition in a, b, and j. Define an U(sl(2))

invariant map

: (~'/2)@23- (~1/2)@2a

(~1/2)@~~

where ,1

is the identity map on the m-fold tensor power of v1l2-

The map (pa@pa)oY

ab

0q5~-where

de is the isomorphism of

3

ve with the symmetric tensors while pe(xl@ -@~2e)= 51 -

for 1 = j,a, or b - is called the Clebsch-Gordan map: V3 -+

' 'x2e

2.5.5 Theorem.

There is a direct sum decomposition

24 THECLASSICALAND QUANTUM6j-SYMBOLS

where the sum is taken over all j such that (a,b,j) is admissible.

Furthermore, if (a,b,j) is an admissible triple, then any U(sl(2))

invariant map Vj + Va @ Vb is a scalar multiple

of pa @ pb o

v

0 q5j. Finally,

Proof.

The map

y:"is U(sl(2))invariant because it is the

composition of lT(s/(2))invariant maps

for

(Pa @ 1

(

,'

(4. (x'J)))/ follows by cornputatiol~using

and

G. TIle follnu~~l

Lemma 2.5.3; thus

y

ab

# 0 for (a,6,j) admissible. The tensor

3

product Va€3vbhas dimension (20+ 1)(2b+ 1) while the image of V3 has dimension (2j + 1). Since

x(2j +1) = (2a +1)(2b+1)

3

(3 )

P (

isomorphic to v'.

is the only space of v

8 v

Consequently every U(sl(2))invariant

V3 --+

Va 8 Vbmust be a multiple of this map.

is

(

Example. Consider v~/~@v~/~.According to Theorem 2.5.5,

this tensor product decomposes as the direct sum of V0 and V1.

The map

112,112

coincides with U while

0

and finally

112,112

yI

(~xY)=~I~(x@Y+YBx).

2.5.6 The Clebsch-Gordan coefficients.

weight vector xj+tyj-t in ~j

V

ab

o $; :

of

Let ej,t denote the

weight t. We have maps pa @ pb o

vj -+ Va 4 vbprovided that (a, b, j) are admissible.

in the sum

8 pb (y3

c:;t:ea,U (A (e3,t)l) = u+v=~

8 eblv.

2.5 -7 Lemma, The Clebsch-Gordan coeficients satisfy the fol-

lowing recursion relation

(j + t + l)ct:%= (a+ U + l)~:;b:,v,t+l t (b+v + l)'::b:l,t+~.

Z,w:~+w=3-t

(jtt)!(j- t)!(a+b- j)!

(a+u +~)!(b+v+w)! z!w!(a- u - z)!(b- v - w)!'

1

11

I

E

I

11.

26 THECLASSICALAND QUANTUM6j-SYMBOLS

Proof. The recursion relation is found by applying F to both

sides of the equation that defines the Clebsch-Gordan coefficient.

The closed form is determined by solving the recursion using the

2.5.8 Diagrammatics for weight vectors.

By definition of

4, the weight vector $t(ej,t) is the image under -kzj of

m

n

where m = j +t, n = j - t.

We represent x (resp. y) by a white (resp. black) vertex with

a string coming out from the top: 6 (resp. ). Then the neight

These conventions have been known to physicists (see 1161).

It is convenient to introduce similar diagrams for dual vectors.

Consider the dual vector space (v1l2)*. We represent the dual

basis vectors x* and y* of tkis dual space diagrammatically by

7 aid 7 , respectively. For parallel strings representing tensor

products of the fundamental representation, putting one of these

dots on the top of the strings algebraically means that we take

the values of the pairing among vectors and dual vectors. In par-

ticular, X means the pairing between x and x* which is equal to

one. We have, then, that the dual to

I

1

REPRESENTATIONS OF U(~1(2))

a+u

a-u

a~61-vb-3

Y=----T

j+t

j-t

27

From the closed formula given in Lemma 2.5.7 for the Clebsch-

Gordan coefficients, one cannot easily see the symmetry properties

"P< of the coefficients under replacing ej,m with ej,-,. However, this

symmetry and the probabilistic nature of the coefficients is more

apparent in the network evaluation.

Here we consider the space of U(sl(2))

invariant maps vk -+ Va @I Vb@I VC.We will construct such maps

2.6 The 6j-symbols.

THECLASSICALAND QUANTUMSYMBOL^

REPRESENTATIONS OF U(SZ(~))

in two different ways. First, consider the composition (a, b, n), and (n, e, k) form admissible triples, form bases for the

I1li

i

'

I

I

/11 /I

I

I

II~I11 If

it

I

/I

\ \

k

N

( I2a €3 ','--

f

vector space of U(sZ(2)) invariant linear maps vk-+ va@vb@VC. fiere pabc = pa @ pb @ pc is the tensor product of the multiplication maps and d3 : ~j -+ (v1I2)B2j sends a a homogeneous polynomial

in x and y to a symmetric tensor.

Proof. The triple tensor product Va @ vb@ VC decom-

(v1I2)@2k

(v112 ~2a@(

poses as ($,Vn) €3 VC = $,(Vn €3 VC) where the direct sum

is taken over all n such that (a, b, n) is admissible, by Theo-

Second, consider the composition rem 2.5.5. For each such n, (Vn @ Vc) contains at most one copy

of vk.Thus h~m~(,~(~))(v~,Va@vb@VC)decomposes as a direct

112)826 @ (~.l/~)@2c

for various values of j.

y.'

I.

I&

ab sum of the 1-dimensional spaces h~m~(,~(~))(v~,vn€3 VC).Sim-

ilarly, it decomposes as a direct sum of the 1-dimensional spaces

h~mu(~i(z))(V~,Va €3 V3 ).

Define the Gj-symbol to be the coefficient

=( y

y

nc

k

812C)o

2.6.2

{

Definition.

] in the following equation.

for various values of n.

The values of j and n are restricted so that (b, c, j), (a, j, k),

(a, b, n), and (n, c, k) all form admissible

'

'

Lrlples.

Alternatively,

if one of these triples is not admissible, then we may declare the

corresponding map

is admissible, then so are the triples obtained by permuting a, b,

V to be the zero man. (Recall that if fn

,------"&.-"IA

h

;\

I

Pa @ Pb @ PC'

2.6.1

Lemma.

The sets

aVbA

)

a\

b\/c)

as the indices j and n range in such a way that (6, c, j), (a, j,k),

By convention, {

abn

}

= 0 if any of the triples (0,c,j),

(a,j, k), (a, 0, n),jn, c, k) is not admissible.

//

I/1

nlI

,

I

30

In the 'paces

hom~(~l(2))(Vk7Va @ Vb @ Vc), we have the two

bases that are defined by these trees, and the ~i-~~~b~lis the

change Of basis matrix. For example, consider the case when

=

= =

= l/2- One can compute direct]y from the definitions that the possible values for j and n are 0 and 1, and that:

and

Ji'

S('rfioil 2 fi

{:;::;; {:;:;;: }

] = -lP,

Recall that a spin-net is an embedding in the plane of a graph

with edges labeled by non-negative half-integers in which each

vertex has valence3. The three edges coincident at a vertex must

= 1, form an admissible triple. The half-integer labels represent the

spin carried by an edge. When we need to emphasize the number

of strings represented by an edge (and hence the number of tensor

{ti; :;:P}= 3/47

factors of the fundamental representation carried by an edge), we

will label the edges with natural numbers that are twice the half- {;;; ;:; } = 1/2. integers. The suffers from this minor inconsistency, but

we have found that the meaning of the labels is clear within the

coll~c'xt111 ~til:t h !i I-

\'

I 1l1~'ll

-

ur~yivc A

IYIII

5i1 o

:,,(,t

I,(,(!

( c,,,

,

we need t~ define an u(sz(~))invariant

map

(V112)@2a@ (v1'2)@2b -+ as follows:

(v112)B2J,for admissible triples (a,b, j)

j

= +2J

ab

( /a+j-b 8

a+b--j

n

8

/j+b-a ) 0 (42,

+2b)

.

The compositi~nirj 0

U(sz(2)) invariant.

3

ab

0 (+a 8 mb) :

va 8 vb ,vj is also

2.6.3

computations.

~h~ elegance of the

addition, we assume that the embedding is in general Po-

sition with respect to a fixed height function: Thus each vertex

appears at a distinct level, the critical points on each edge are

non-degenerate, and these critical points are at distinct levels from

the vertices. Furthermore, some edges may be marked with 'ym-

metrizers: +. More precisely, we include valence 2 vertices in

which the two incoming edges have the same hbel.

~h~ principal results of the current section are the orthogo-

nality and the EEott-Biedenharn identities that are satisfied

spin-net notation that we have developed so far is it facili- the ~j-~~~b~l~.We will give proofs of these relationships (and

tates Otherwise tedious calculations.

tage in that the calculations are performed in the tensor power,

There is

slight

disadvan- others) that are simply manipulations of diagrams. To this end

we state a diagrammatic lemma.

32

THECLASSICALAND QUANTUM6j-SYMBOLS

The following relationships hold among the

U(sZ(2)) invariant maps )\ , y, 6,Inl and fi. (Here we iden-

tify C 8 V and V for any U(sl(2))space V.)

1. (In 8;)0(68 l,)=~n=(fi~ln)o(ln 86).

2.6.4

Lemma.

Proof. Part 1 follows from Lemma 2.3.2 part 3, and induction since the cancellation of a U with a n can occur regardless of the tensor factors on which those maps are acting. Part 2 follows by induction and part 5 of Lemma 2.3.2. Part 3 follows similarly using part 4 of Lernma 2.3.2. Part 4 and part 5 are proved using

2.6.5

Remark. Consider the collection of proper embeddings

of two and three valent graphs in a rectangle whose edges are parallel to the coordinate axes in the plane. The free end points

of the edges of the graph are embedded in the top and bottom

edges of the rectangle. If two such embedded graphs are isotopic via an isotopy that keeps the boundary fixed, then there is an

isotopy between them that can be decomposed as a sequence

moves that are the diagrammatic descriptions of items 1 through

5 in Lemma 2.6.4. The valence two vertices are represented in

the Lemma by the projectors, and the valence three vertices are represented by the Clebsh-Gordan maps.

To find such a nice isotopy, one replaces a given isotopy by

one that is in general position with respect to the height fuliction defined on the rectangle. The existence of the generic isotopy is

guaranteed by a transversality argument, and a similar transver-

of

,,i~,d~

CIII~I(\II~(~(~(~~~~t:~~~~~~.:II,>

l,O~olj\

IIII~~ ~II!II~~-

1

,

,

of pieces each of which is of the diagrammatic form specified.

(a, b, n),

(c,k,n), (a, b, m), and (c, k, m) are admissible triples. Then then 6j-symbols satisfy the following relation:

2.6.6

Theorem

(Orthogonality).

Suppose that

Proof. Define

~h~~ we have the following calculation.

Therefore,

THECLASSICALAND QUANTUMSYMBOLS

REPRESENTATIONSOF U(sz(2))

39

. This complex is depicted in Figure 1. Each face of the Zdimensional complex arises as the cartesian product of an edge of the tree and a unit interval; the interval factor is thought of as a time parameter in the deformation between the two trees. The vertex of the complex occurs as the edge labeled b passes through the lower junction of three edges. The 2-dimensional complex has I vertex, 4 edges, and 6 faces; thus it is the dual complex to a

tetrahedron as indicated in the figure.

Let us associate to this complex a 6j-symbol. Then consider the orthogonality conditions and the Elliott-Biedenharn identities. Either side of each equation can be similarly thought of as a 2-

tllllle~l~lo~ldlcell co~iil)lcx.1'01

C~~III~~C,111 I'I~LIIC'2 ulle \icil: O[ t11e

orthogonality relation is depicted both as a movie description and as a 2-dimensional complex; the Zcomplex is the "time elapsed" version of the accompanying movie. The sucessive stills of the

movie differ by a 6j-symbol and these symbols are associated to the vertices of the 2-dimensional complex on the right of the fig- ure. The trees in the stills are the trees that represent the maps in 2.6.8 Associating the 6j-symbol to the dual skeleton of the orthogonality relation. Figure 3 depicts the other side of the orthogonality relation, and the 2-dimensional complex here differs that occurs when the branch labeled by b'is moved from the from the previous one by one of the Matveev moves [25, 16, 321. Simaarly, Figure 4 depicts in a movie fashion the three trees that

right side of the tree

to the left side of the tree

cessive stills in the movie differ by a 6j-symbol. In Figure 5 the stills in the movie are the trees on the other side of the Elliott- Biedenharn identity. The 2-complex in either figure represents the

THECLASSICALAND QUANTUM6j-SYMBOLS

Figure 1: A movie of a 6j-symbol and the Matveev complex

Figure 2: A movie of Gj-symbols and ort.hogonality (left hancl side)

Figure 3: A movie of 6j-symbols and orthogonality (right hand side)

Figure 4: A movie of 6j-symbols and the Elliott-Biedenharn iden-

tity (left hand side)

dual to the union of tetrahedra - two tetrahedra glued along a

face for Figure 4 and three tetrahedra glued along an edge for

Figure 5. These two complexes are related by a deformation that

is also one of the Matveev moves [25, 16, 321.

Thus the identities expressed among the 6j-symbols are dia-

grammatically expressed as deformations between 2-dimensional

cell complexes. So the use of 6j-symbols in the construction of

the Turaev-Viro invariant appears quite natural in the context of

the representations and their diagrammatic realizations.

However, there are two obstacles to overcome before the defini-

tion of the Turaev-Viro invariant can be made. First, the invariant

is a sum over representations, and since we have irreducible rep-

resentations for all non-negative half-integers j, such a sum would

be infinite, so we cannot use the classical theory of U(sl(2)) to

obtain an invariant. Second, the Gj-symbols that we have defined

(IJJ

1101

r1111I(~~~~II~(~CI~~II,) llllll~~t

III(I

1 IIII,

t ill,\

( ,111~10~I)(>

associated to tetrahedra in any meaningful way. We will overcome

the first obstacle in Section 4, by passing to the representations of

U,(s1(2)) for q a root of unity. Professor Biedenharn informs us

that this is the physicist's notion of renormalization since we are

converting an infinite sum to a finite sum. We will overcome the

second obstacle by normalizing the 6j-symbols and by showing

that the normalized versions possess the desired symmetr'y while

still satisfying orthogonality and Elliott-Biedenharn relations.

2.7 Computations. We now express the 6+symbols in terms

of evaluations of certain spin-nets. In particular, we determine

some of the symmetry properties of the 6j-symbols in the current

Figure 5: A movie of 6j-symbols and the Elliott-Biedenharn iden- tity (right hand side) normalization, and we find a normalization that has full tetrahe-

dral symmetry.

44 THECLASSICALAND QUANTUM 63-SYMBOLS

2.7.1 Topological invariance. Let a spin-net be given. This

network is a graph with its edges labeled by non-negative half- integers that has only 2-valent and 3-valent vertices. Two edges incident at a 2-valent vertex must have the same label, and if edges with labels a, b, and j are incident at a vertex, then the triple (a, b, j) is admissible. Choose an embedding of the spin-net

into a rectangle such that the endpoints of free edges (if any) are on the top and bottom of the rectangle. Suppose that the labels

.,a,, and those that

of the edges that appear on the top are al,

appear on the bottom are bl,

from left to right once the rectangle has been embedded in the .

plane. The embedding of the spin-net allows us to define a map

,b,

where these labels are read

jectors

are associated to the 2-valent vertices and maps

and )\ are associated to the 3-valent vertices. At the bottom of

the rectangle the vbs are mapped via q5 into the tensor powers of the fundamental representation, and at the top the tensor powers - are projected back onto the Vas. Some care has to be taken with regard to the indices along the edges as we indicated in the defi- nition given in Section 2.6.3. Finally, the embedding of the graph

' is to be in general position with respect to the height function on

2.7.2 Lemma.

class (rel.

does not depend on the isotopy

boundary) of the embedding of the given graph in the

The map

KEYKESLN'l.A'l.lVNS

OF U (S1(L))

45

Proof. This follows from the remarks 2.6.5.

2.7.3 Lemma.

given by

The value of fi: (v~/~)@"@ (v~/~)@~-i C is

in(- l)#{"'"k'~)

if

if

~j=fjforsome j=

for all

1,

j = 1,

{xj, fi} = {x, y}

1 where xj,q E {x, y} for all j =

.,n.

n,

, n

Proof. This follows by induction, let us exemplify the formula. In case n = 2, the non-zero values of n are as follows:

2

2.7.4

Lemma.

For u + v

= j, consider the map,pj o

(Pa @ q5b : Va @ vb-i

Vj. We have

Proof. Recall that e,,, = xa+"y

eb,, holds, and that the image under 4 of a weight vector, ej,t, is a symmetrized version of it.

, that a similar formula for

THE CLASSICALAND QUANTUM6j-syMBOLS

Proof. Part (1) is a direct calculation that can be achieved by

evaluating

(2)follows from (1) and Lemma 2.7.2 by rotating the vertex of the

for (2) 180'. Parts (3) and (4) follow by

right side of the diagram

either side on the highest

weight vector, 4J(22~). part

on the tensor

products of appropriate weight vectors.

2.7.6 Theorem.

where

is a Kronecker 6 function, A, = (-1)2~(2j+ 11,

@(a,b, k) =

REPRESEN lAllUlV3 Up

u

\J1\"J/

proof. Since the given composition maps an irreducible repre-

sentation VJ into vk,it is O when k # j. And when k = j it is a

constant multiple of the identity. In Theorem 2.5.5 we computed

4a(xa+uya-u)4 4b(~~+~€3 yb-')))

= x2~

u+u=1

( -])'-i'

ia+b-~ia+b-~(-I)~-u

(c,

--

+ 1, -

?)'((L t

71)'(h t

(2a)!(Lb)I

!))I

.-

((afb- j)!)2(a+b+j+l)!(a+j-b)!(b+j-a)! (2~+ l)!(at b - j)!

(a+b - j)!(a+j - b).( 1 b+j - a)!(b+j +a +I)! (2a)!(2b)!(2jt I)!

~h~ next to the last equality is a combinatoric identity. The

proof that follows was indicated to us by Rhodes Peele. Consider

a + b + j

+ 1) such that the value f(a +j

-

b + 1) is

every element of f ({a + j - b + 2,. - ., 2j + 1))- We count the

elements of the set B in two ways-

50

THECLASSICALAND QUANTUM6j-SYMBOLS

There are ('+,'tl)possible choices for the image f ({1,

,2j+

I)), and each such choice can be arranged in (a+j - b)!(b +j -a)!

distinct ways while f ({2j

in (a +b - j)! distinct ways. Thus

,a

+b +j +1)) can be arranged

Alternatively, B = UeBe where Be = {f E B

f

+ I), so #B = xe#Be.

the set f -'({l,.

,!))

a+b-j'

There are (e-(a+j-b)

1) =

) possibilities for

since the inverse image must exclude the

the

in the closed interval [I, a +j - b]. Each such set can be

: f(a+ j-bf

integers in the closed interval [a+j - b +1,2j+11 and include

integers

arranged in l! distinct ways. Furthermore, f -'({l+

,a

+b +

j +1))is determined by f

in (u + b +j - e)! distinct ways. So

,!))

that

and this can be arranged

Now let C = a + u, and let j

= u + v, we have

(a t j - b)!(b +j - a)!(a +b - j)!.

The required identity

follows by rearranging the factors

in the

above equation. This completes the proof.

The spin-net version of Theorem 2.7.6 is indicated in

Corol-

lary 2.7.7. Here we have labeled the spin-nets with the number of

strings (= number of tensor factors of v'f2involved). The loop

closure of 'Czj gives the value Azj = (-1)~j(2j + 1) which is the

denominator of the right hand side. The value AZj gives the spe-

cial case when a = 0 of Theorem 2.7.6. The closed network in the

numerator on the right has the value @(a,b, j) - this is why the

function is named theta.

2.7.7 Corollary. Let (a, b,j) denote an admissible triple of half-

integers.

Then we have equality between the following spin-nets.

2.7.8 Lemma.

The 6j symbols possess the following symmetry

Proof. Embed the spin-net

labeled m and p attached to the top edge while the edges labeled

By Lemma 2.7.2, the eval-

uation of the spin-net remains the same when the cross bar of

s and t are attached to the bottom.

in a rectangle with the edges

S

52 THECLASSICALAND QUANTUM6j-SYMBOLS

the mHis rotated clockwise 180' while the end points of the

s

t

boundary remain fixed.

Having performed such a rotation, recouple; the 6j-symbol

I

that appears is

, where a sum is being taken over

u. Then rotate the (now vertkal) cross bar 180' counterclockwise. We obtain,

,If

u

Thus the coefficients are equal since pm @ppo

04,@I& forms

Vm@I VP.

a ba.sis for the set of U(sZ(2))invariant maps VS@Vt i

Proof. Any of these constants is the coefficient Z in the ecluation

That there is such a constant Z follows since the space of U(sZ(2)) invariant maps vk -+ Vm@ VP is 1-dimensional and is spanned

by pm @ pp o

corresponding diagrams.

Y

mP

k

o 4k. We leave it to the reader to draw the

1

f

2.7.10 Lemma. Let TET(a, b, c, d,e, f) denote

spin-net depicted below:

Then,

the value of the

Proof. The proof follows by recoupling and then applying Corol-

2.7I.,- and LCII~LI;~2.7.9(~f. [!(i].;). I Sagrammatic proof is shown below.

- -

n