ITEP-TH-28/3

On the Virasoro constraints for torus knots

Oleg Dubinkin∗

Institute for Theoretical and Experimental Physics, 117259 Moscow, Russia

arXiv:1307.7909v1 [hep-th] 30 Jul 2013

and
Moscow Institute of Physics and Technology, 141700 Dolgoprudny, Russia

Abstract
We construct Virasoro algebra of differential operators for the Jones-Rosso matrix model.
These operators generate various relations between Wilson loops. Then we discuss the con-
structed operators and corresponding relations in the stability limit. A series of examples is
given.

1 Introduction
Matrix models provide a rich set of approaches to physical systems and quantities (see [6]
for review). Constructing a fundamental Virasoro algebra for considered matrix model is one
of the most useful methods. This algebra is very fruitful by itself, as it allows us to generate
various relations between quantities represented in this model. One can go further, for example,
constructing a spectral curve using Virasoro algebra. But the first step is to derive one, which
is the subject of this paper. We will study Jones-Rosso matrix model, that describes Wilson
loops along torus knots. To obtain Virasoro algebra we will use a standard procedure. First,
one should input a generating function for desirable quantities into matrix integral and obtain a
Ward identity for this new integral. The second step is to construct differential operators, that
can produce such Ward identities being applied to the matrix integral with generating function.
This procedure is not always possible, so it can be necessary to input several generating functions
in the integrand. We apply this technique to Jones-Rosso matrix integral and this paper results
in generators of Virasoro algebra for torus knots (27) and identities between Wilson loops,
produced by generators of this algebra.
This paper is organized as follows. Section 2 reviews the general algorithm for the Jones-
Rosso matrix integral of Chern-Simons partition function and obtains trivial differential equation
for it as an example. In section 3, we will discuss different approaches to Jones-Rosso matrix in-
tegral and obtain two different representations of Virasoro algebra that corresponds to different
generating functions. Section 4 discusses the so-called stability limit, where we will see decom-
position of Virasoro generators into three independent sets of differential operators. Finally,
section 5 describes a series of examples.
∗
dubinkin@itep.ru

1
2 Matrix integral and Ward identities for torus knot
Let us consider the next identity for Wilson loops in Chern-Simons theory with gauge group
G=SU(N), written for the case of torus knots [2]:
Z
1 u2 Y u·α u·α
WR (KP,Q ) = due− 2gs 4 sinh( ) sinh( )χR (eu ) (1)
Z(KP,Q ) α>0
2P 2Q

In this expression,
Z Y
u
− 2g
2
u·α u·α
Z(KP,Q ) = due s 4 sinh( ) sinh( ) (2)
2P 2Q
α>0

and χR (eu ) are the characters of G, gs = P κ+h

Q·2πi
, κ is the level of the theory, h is the dual Coxeter
number of G, u is an element in Λw ⊗ R and α is a set of positive simple roots of G. A torus
knot is specified by a pair of coprime integers P and Q. In this section we will work with the
integral (2) to develop the necessary technique and show the simplest example of deriving Ward
identity and the corresponding differential operator. Using a well-known fact that the positive
roots can be written in an orthonormal basis {ek }k=1,...,N as
αkl = ek − el , 1 6 k < l 6 N
P
and that u = k λk ek , one can rewrite the integral in a form, which is more appropriate for
further calculations:
Z Y
N λ2 Y
− i λk − λl λk − λl
Z(KQ,P ) = dλi e 2gs 4 sinh( ) sinh( ) (3)
2P 2Q
i=1 k<l

The following considerations are very similar to those in [4]. To obtain Ward identities, one
should perform the next substitution:
λ̃i = λi + ǫfi , λi = λ̃i − ǫfi (4)

RConsidering Rǫ an infinitesimal
R we can decompose integrals over each λi in ZP,Q R as follows:
f (λ)dλ = f (λ̃)dλ̃ + ǫ g(λ̃)dλ̃ + o(ǫ2 ). In the first order of ǫ this gives us: g(λ)dλ = 0.
Performing such procedure for each λi in ZP,Q one obtains the following Ward identity:
Z Y N N
!
X ∂fi λi fi fk − fl λk − λl fk − fl λk − λl
dλi ( − )+ coth( )+ coth( ) ×
∂λi gs 2Q 2Q 2P 2P
i=1 i=1
(5)
λ2 Y
− 2gi λk − λl λk − λl
×e s 4 sinh( ) sinh( )=0
2P 2Q
k<l

Now we put fi = λi and, to shorten the expression, we replace integrals with brackets (this
notation will be used throughout this paper):
DX
N
λ2i E D λk − λl λk − λl E D λk − λl λk − λl E
(1 − ) + coth( ) + coth( ) =0 (6)
gs 2Q 2Q 2P 2P
i=1

The resulting identity can be obtained by applying the following differential operator to ZP,Q :

b =N −P ∂ −Q ∂ ,
L b P,Q = 0.
LZ (7)
∂P ∂Q
The last equation simply states that ZP,Q is a homogeneous function of degree N with respect
to variables P and Q. This statement is easily verified by taking the integral.

2
3 Elements of Virasoro algebra
Now we will take a look at the observables in Chern-Simons theory - Wilson loops. In terms
of this matrix integral, Wilson loops for torus knots KQ,P are the averages of the characters of
G, which are Schur polynomials χR :
Z Y
N λ2 Y D E
1 − 2gi λk − λl λk − λl
WR (KP,Q ) = dλi e s 4 sinh( ) sinh( )χR (eλ ) = χR (eλ ) (8)
ZP,Q 2P 2Q
i=1 k<l

For further calculations it is convenient to perform the following change of variables [3]:

λi = P Q log xi (9)

The expression for a Wilson loop will then take the following form:
Z N
Y    
(P Q)N P Q(log xi )2 P +Q
WR (KP,Q ) = dxi exp − − (N − 1) + 1 log xi ×
ZP,Q RN 2ĝs 2
+ i=1
Y (10)
× (xPk − xPl )(xQ
k − x Q
l )χ R (x)
k<l

2πi
where ĝs = K+y . Let us replace the characters by their generating function:

∞
! N
X X X
χR (t)χR (x) = exp tk p k , where pk = xki (11)
R k=1 i=1

and the resulting integral will be:

X
W = χR (t)WR (KP,Q ) (12)
R

The next step consists of obtaining Ward identities for W , performing a change of variables as in
the previous section. Then we will construct differential polynomials with respect to { ∂t∂k } that
generate these identities. Applying them to W and fixing the infinite set of parameters {tk }, one
will obtain various relations between Wilson loops for the same knot in different representations.
Unfortunately, since logarithm can’t be represented as a power series of xi , that converges on
whole positive semi-axis, we can’t construct a term with log(xi ) in upcoming Ward identity
(15), using only differentials with respect to tk ’s. The most evident way to solve this problem
comprises inserting another generating function into the integral:
∞ N
!!
X X
k
exp lk xi log xi (13)
k=1 i=1

This functions also should be interpreted as Schur decomposition. From now on we will be
considering the following integral:
N Z N
Y    
f = (P Q) P Q(log xi )2
P +Q
W dxi exp − (N − 1) + 1 log xi ×
−
ZP,Q RN
+ i=1
2ĝs
2
∞
! ∞ N
!! (14)
Y Q Q
X X X
P P k
× (xk − xl )(xk − xl ) exp tk pk exp lk xi log xi
k<l k=1 k=1 i=1

3
Repeating the procedure described in the previous section, we obtain Ward identity:
N 
DX   E
∂f i P Q fi log xi P +Q fi
− − (N − 1) + 1 +
∂xi ĝs xi 2 xi
i=1
D X xQ−1 f − xQ−1 f E D X xP −1 f − xP −1 f E
k k l l k k l l
+ Q Q Q
+ P P − xP
+ (15)
k<l x k − x l k<l
x k l
! !
DX
∞ XN E DX ∞ XN E
+ ktk xik−1 fi + lk fi xik−1 (1 + k log xi ) =0
k=1 i=1 k=1 i=1

Note that we have omitted generating functions inside the brackets to make the expression
shorter. This notation with skipped generating functions will be used throughout this paper.
Now we have to choose actual form of the deformations fi ’s. The most common way to do
so is to put fi = xnP
i
Q+1
. The resulting identities can be obtained by applying the following
f
differential operators to W :
PQ ∂
∂ X ∂ ∂
cnP Q = (nP Q − (P + Q)(N − 1))
M − +P +
∂tnP Q ĝs ∂lnP Q ∂taP ∂tbP
a+b=nQ
∞ ∞   (16)
X ∂ ∂ X ∂ X ∂ ∂
+Q + ktk + lk +k
∂taQ ∂tbQ ∂tnP Q+k ∂tnP Q+k ∂lnP Q+k
a+b=nP k=1 k=1

This operators form the Virasoro algebra. Applying them to W f and defining two infinite sets
of parameters {tk } and {lk }, one can generate identities between WilsonPloops in different
N k
representations and averages of the form hχR1 (pk )χR2 (p̃k )i, where pk = i=1 xi and p̃k =
PN k
i=1 xi log(xi ). For example, if we put all tk ’s and lk ’s equal to zero, we will obtain relations
P nP Q
between average h xi log(xi )i and Wilson loops, taken in different representations with the
same number of boxes in corresponding Young tableau. For more details, see section 5.
There is a way to rewrite this matrix model without the logarithmic term. This will allow
us to obtain Virasoro algebra using only one generating function, which is our final goal in this
section. Let us consider the following simple identity of integrals (as in [7], appx. B):
Z ∞   I
1 d (log x)2 (d+1)2 /2 1
√ dx x exp =q = dz z d θq (z) (17)
2πg 0 2g 2π |z|=1
P k 2 /2 z k is Jacobi theta function. This allows us to rewrite
where q = eg and θq (z) = ∞ k=−∞ q
the partition function and Wilson loops (10) in the following way:
I N
Y Y
(2πĝs P Q)N/2 −( P +Q (N −1)+1)
WR (KP,Q ) = dzi zi 2
θq (zi ) (zkP −zlP )(zkQ −zlQ )χR (z) (18)
ZP,Q |z|=1 i=1 k<l

with q = exp( PĝQ

s
). We will make a small shift of coupling constant ĝs → ĝs −ǫ, where 0 < ǫ << 1.
Since Re ĝs = 0 and P , Q ∈ Z we get |q| < 1. Also note that |zi | = 1. As usual, we will work
with the modified integral, where characters are replaced with their generating function:
I N ∞
!
(2πĝs P Q)N/2 Y −( P +Q (N −1)+1) Y Q Q
X
W = dzi zi 2 θq (zi ) (zkP − zlP )(zk − zl ) exp tk p k
ZP,Q |z|=1 i=1 k<l k=−∞
(19)

4
 PN k
where pk = i=1 zi . Note that summation in generating function is taken over all integers
including negative values. This means that we are actually using two generating functions for
positive and negative values of k. Performing transformation of variables, as in the previous
case:
z̃i = zi + ǫzinP Q+1 , zi = z̃i − ǫz̃inP Q+1 (20)
we obtain the following identity for W integral in the first order of ǫ:
D X N E DX N P∞ k 2 /2 z k E
P +Q nP Q nP Q k=−∞ kq i
nP Q − (N − 1) zi + zi +
2 θq (zi )
i=1 i=1
!
D X z Q(nP +1) − z Q(nP +1) E D X z P (nQ+1) − z P (nQ+1) E D X∞ N
X E
+ Q k
Q
l
Q
+ P k l
P − zP
+ ktk zinP Q+k =0
k<l zk − zl k<l
zk l k=−∞ i=1
(21)
The second term in the above expression can be transformed as follows:
DX P∞ k 2 /2 z k E DX zi θq (zi )′zi E D X nP Q+1 E
N N N
nP Q k=−∞ kq i
zi = zinP Q = zi (log θq (zi ))′zi (22)
θq (zi ) θq (zi )
i=1 i=1 i=1
We will use the next representation for Jacobi theta function, that will help us to rewrite the
logarithmic term in a more convenient way:
∞
Y zi 1
θq (zi ) = (1 − q k )(1 + q k √ )(1 + q k √ ) (23)
q zi q
k=1
∞      !′
X  
k zi k 1
zi (log θq (zi ))′zi = zi k
log 1 − q + log 1 + q √ + log 1 + q √ =
q zi q
k=1 zi
! (24)
X∞ 1
q k− 2 1 1
= zi −
k− 21 z 1
i 1 + zi q −(k− 2 )
k=1 1 + zi q
Now, using that |q| < 1 and |zi | = 1, we can decompose these two fractions into the Laurent
power series:
∞ ∞ ∞
!
X X 1 X 1 1
(−1)l+1 (zi q k− 2 )l − (−1)l+1 ( q k− 2 )l =
zi
k=1 l=1 l=1
−1   ∞  √  ∞  √ |l| (25)
X l X l X
l 1−q l q l+H(l) q l
= (−1) √ zi − (−1) zi = (−1) zi − 1
q 1−q 1−q
l=−∞ l=1 l=−∞

where H(l) is the Heaviside step function with H(0) = 0. Finally, we can write the second term
in the Ward identity in a more appropriate form:
DX N P∞ k 2 /2 z k E D X
∞  √ |l| E
nP Q k=−∞ kq q
zi i
= (−1)l+H(l) zinP Q+l − zinP Q (26)
θq (zi ) 1−q
i=1 l=−∞

Thus we can finally write down differential operators corresponding to Ward identity (21):
X∞  √ |l|
b ∂ l+H(l) q ∂
LnP Q = (nP Q − (P + Q)(N − 1) − 1) + (−1) +
∂tnP Q 1−q ∂tnP Q+l
l=−∞
∞ (27)
X ∂ ∂ X ∂ ∂ X ∂
+P +Q + ktk
∂taP ∂tbP ∂taQ ∂tbQ ∂tnP Q+k
a+b=nQ a+b=nP k=−∞

5
 P
As mentioned already, we have used two generating functions. First one is exp ( ∞ k=1 tk pk )
which
P is the generating
 function for the characters of G. The second generating function is
−1
exp k=−∞ tk pk . It can be represented as a Schur decomposition, but in this case, variables
PN k
pk = i=1 zi in Schur polynomials have negative values of k’s. We denote Schur polynomials
with negative k’s as χR (p− ) and the Schur polynomials with positive k’s as χR (p+ ). Considering
the integral (19), it is easy to see that hχR (p− )iP,Q = hχR (p+ )i−P,−Q = hχR (p+ )iP,Q . Thus the
integral (19) can be written as:
XX
W = χR (t+ )χR′ (t− )hχR (p+ )χR′ (p− )i (28)
R R′
After we apply the differential operator (27) to the expression above and put all tk ’s equal to
zero, we obtain the relations of the form:
X X X
AR hχR (p− )i + BR hχR (p+ )i = (AR + BR )hχR (p+ )i = 0 (29)
R R R
where AR and BR are some constants.

4 Stability limit
Now let us consider the stability limit, i.e.:
P
P → ∞, Q → ∞, while = Const. (30)
Q
f integral (14) and cor-
In this and the following sections we will carry out calculations for W
responding M cnP Q operators (16), but all of the following conclusions can be easily rewritten
for W integral and L b operators (27). First, our Ward identity (15) with fi = xnP Q+1 must be
i
resolved in all orders of P and Q. Thus we end up with three independent equations:
DX
N E DX
N
PQ E
nP QxnP
i
Q
− xnP
i
Q
log x i =0
ĝs
i=1 i=1
DXN
P +Q E D X xQ+nP Q − xQ+nP Q E D X xP +nP Q − xP +nP Q E
(N − 1)xnP
i
Q
) = Q k
Q
l
Q
+ P k l
P − xP
(31)
2 x − x x k l
i=1 k<l k l k<l
! !
DX∞ N
X E DX ∞ X N E
ktk xik+nP Q + lk xik+nP Q(1 + k log xi ) =0
k=0 i=1 k=0 i=1
And, correspondingly, every single Virasoro algebra generator factorizes into three simple oper-
ators:
∂ PQ ∂
ên = nP Q −
∂tnP Q ĝs ∂lnP Q
∂ X ∂ ∂ X ∂ ∂
fˆn = −(P + Q)(N − 1) +P +Q (32)
∂tnP Q ∂taQ ∂tbQ ∂taP ∂tbP
a+b=nQ a+b=nP
X∞ X∞  
∂ ∂ ∂
ĥn = ktk + lk +k
∂tnP Q+k ∂tnP Q+k ∂lnP Q+k
k=0 k=0
Thus, in this limit, we got much simpler expressions for Ward identities and corresponding
differential operators. As we will see, this simplifies things greatly when we want to get actual
relations for Wilson loops.

6
5 Trefoil example
Now let us consider a trefoil knot and generate relations between its Wilson loops, or HOMFLY
polynomials, written in different representations. To do so, one should consider generalized
matrix integral (14) and put P = 2, Q = 3. This matrix integral can be written as:
XX
f=
W χR (tk )χR′ (lk )hχR (pk )χR′ (p̃k )i (33)
R R′
PN PN
where pk = i=1 xki and p̃k = i=1 xki log(xi ). We are interested in hχR (pk )i as they appear
to be WR (KP,Q ), this is why we should put all lk ’s and tk ’s equal to zero – that will leave us
only separate averages of χR (pk ) and χR (p̃k ) in (33) and put generating functions to 1 in Ward
identities. Now let us write an exact form of n = 1 element of Virasoro algebra for trefoil:
X ∂ ∂
L b 6 = (6 − 5(N − 1)) ∂ − 6 ∂ + 2
b nP Q = L +
∂t6 ĝs ∂l6 ∂t2a ∂t2b
a+b=3
X ∂ ∂ ∞
X ∞
X   (34)
∂ ∂ ∂
+3 + ktk + lk +k
∂t3a ∂t3b ∂t6+k ∂t6+k ∂l6+k
a+b=2 k=0 k=0

f and put all {tk } and {lk } equal to zero,

Next, one should apply this differential polynomial to W
as was discussed above and take, for example, simplest case of 2 × 2 matrix:
b6 W
L f |t =l =0,∀k = 0 (35)
k k

m
N
6 X 6
h xi log xi i = 5W(6) − 5W(5,1) + 2W(4,1,1) − 3W(2,1,1,1,1) +
ĝs (36)
i=1
+3W(1,1,1,1,1,1) + 3W(4,2) − W(3,3) − W(2,2,2) + 3W(2,2,1,1) − 2W(3,2,1)
P 6
In a considered case, N = 2 and the average of xi log(xi ) will be:
 
37ĝs  
XN
6 2
h xi log xi i = 48πĝs exp 7 − 11eĝs − 13e2ĝs + 17e5ĝs (37)
24
i=1
P
It is easy to see that in LHS of equation (37) we always have a term proportional to h xnPi
Q
log(xi )i.
That means one can easily rewrite this identity as the identity between Wilson loops in different
representations for two different knots if it is possible to choose such n’s that the products of n,
P and Q for this knots are equal. For example, we can compare Wilson loops for trefoil knot
with n = 2 and for torus knot with P = 4, Q = 3, n = 1, but the result is too overwhelming to
present it here.
In the stability limit, as was discussed above, we get two following identities (the third
P
one vanishes as we put tk ’s and lk ’s equal to zero) for Wilson loops, when Q = 23 :

6 X 6
N DX
N E
h xi log xi i = W6 − W5,1 + W4,1,1 − W3,1,1,1 + W2,1,1,1,1 − W1,1,1,1,1,1 ≡ 6 x6i (38)
ĝs
i=1 i=1

W(3,3) + 3W(4,2) + W(2,2,2) − 4W(3,2,1) − 3W(4,1,1) + 3W(2,2,1,1) + 7W(3,1,1,1) −

(39)
−10W(2,1,1,1,1) + 10W(1,1,1,1,1,1) = 0
This identities correspond to ê1 and fˆ1 operators in (32).

7
6 Discussion
We obtained the Virasoro constraints for two different cases. Resulting Virasoro generators are
obviously connected. Such connection can be seen after applying them to Jones-Rosso integral
with corresponding generating function(s) and putting all additional parameters equal to zero.
Namely, we can make a following statement:
 √ |l| !  
X∞
q ∂ ∂  P Q ∂ 
 f 
(−1)l+H(l) − W =− W (40)
1−q ∂tnP Q+l ∂tnP Q  ĝs ∂lnP Q 
l=−∞ tk =0,∀k tk =lk =0,∀k

Perhaps it would look better if we rewrite this equality in terms of averages:

D X
∞  √ |l| E DP Q X
N E
q
(−1) l+H(l)
xnP
i
Q+l
− xnP
i
Q
=− xnP
i
Q
log(xi ) (41)
1−q ĝs
l=−∞ i=1

Note that in the LHS we have replaced zi with xi . This is a valid operation, as was discussed
in section 3 (see (17)). However, such relations are yet to be discussed. Direct application of
Virasoro generators gives us identities betweenPWilson loops, or HOMFLY polynomials. These
identities, contain either an undesirable term h xnP
i
Q
log(xi )i or infinite sum of Wilson loops in
different representations that prevents us from getting closed recurrent relations. On the other
hand, the logarithmic term can be directly computed, for example in N = 2 case:


 
X nP Q ĝs (P −Q)2
+2n(P Q−P −Q)
log(xi )i = 8ĝs2 πe − eĝs (P + Q − 2nP Q + 1)+
4 PQ
h xi
i (42)

ĝs (1+n(P +Q)) ĝs nQ ĝs nP
+e (P + Q + 2nP Q) + e (P − Q − 2nP Q) + e (Q − P − 2nP Q)

or it can be used to compare compositions of colored Wilson loops for two different knots, with
equal products of n, P and Q. The obtained Virasoro generators are also useful for studying
these relations in the so-called stability limit, where they are significantly simplified.
Though there is a working formula for HOMFLY polynomials or Wilson loops for all torus
knots, the relations between these objects are not fully understood and this example of Virasoro
algebra can give some idea of general structure of such algebras for all knots.

Aknowledgements
Author is indebted to A. Gorsky for suggesting this problem and numerous fruitful discus-
sions. I am grateful to A. Milekhin and Sh. Shakirov for useful comments and valuable insights.
This work was supported in part by grant of Ministry of Education and Science of the Russian
Federation under contract 8207, by grant RFBR-12-02-00284 and by grant NSh-3349.2012.2.

References
[1] E. Witten, ”Quantum Field Theory and the Jones Polynomial,” Commun. Math. Phys.
121, 351 (1989).

[2] M.Rosso and V.F.R.Jones, J. Knot Theory Ramifications, 2 (1993) 97-112 X.-S.Lin and
H.Zheng, Trans. Amer. Math. Soc. 362 (2010) 1-18 math/0601267

8
[3] A. Brini, B. Eynard, M. Marino, ”Torus knots and mirror symmetry, arXiv:1105.2012.

[4] A. Mironov and A. Morozov, ”On the origin of Virasoro constraints in matrix models:
Lagrangian approach”, Phys. Lett. 252B (1990) 47.

[5] A. Morozov, ”Unitary Integrals and Related Matrix Models”, Theor.Math.Phys. 162 (2010)
1-33, arXiv:0906.3518

[6] A. Morozov. ”Faces of matrix models”, JETP Lett. 95 (2012) 586-593 (2012)
arXiv:1204.3953

[7] M. Aganagic and S. Shakirov, ”Knot Homology from Refined Chern-Simons Theory”, (2011)
arXiv:1105.5117.