Prepared for submission to JHEP

From 2d Droplets to 2d Yang-Mills

Arghya Chattopadhyaya , Suvankar Duttab , Debangshu Mukherjeeb,c , Neetub

a Institute
of Mathematical Sciences, Homi Bhaba National Institute (HBNI)
IV Cross Road, Taramani, Chennai 600113, Tamil Nadu, India
b Indian
Institute of Science Education and Research Bhopal
arXiv:2010.11923v1 [hep-th] 22 Oct 2020

Bhopal Bypass, Bhopal 462066, India

c IndianInstitute of Science Education and Research Thiruvananthapuram
Vithura 695551, Kerala, India
E-mail: [email protected], [email protected],
[email protected], [email protected]

Abstract: We establish a connection between time evolution of free Fermi droplets

and partition function of generalised q-deformed Yang-Mills theories on Riemann
surfaces. Classical phases of (0 + 1) dimensional unitary matrix models can be
characterised by free Fermi droplets in two dimensions. We quantise these droplets
and find that the modes satisfy an abelian Kac-Moody algebra. The Hilbert spaces
H+ and H− associated with the upper and lower free Fermi surfaces of a droplet
admit a Young diagram basis in which the phase space Hamiltonian is diagonal with
eigenvalue, in the large N limit, equal to the quadratic Casimir of u(N ). We establish
an exact mapping between states in H± and geometries of droplets. In particular,
coherent states in H± correspond to classical deformation of upper and lower Fermi
surfaces. We prove that correlation between two coherent states in H± is equal to
the chiral and anti-chiral partition function of 2d Yang-Mills theory on a cylinder.
Using the fact that the full Hilbert space H+ ⊗ H− admits a composite basis, we
show that correlation between two classical droplet geometries is equal to the full
U (N ) Yang-Mills partition function on cylinder. We further establish a connection
between higher point correlators in H± and higher point correlators in 2d Yang-Mills
on Riemann surface. There are special states in H± whose transition amplitudes are
equal to the partition function of 2d q-deformed Yang-Mills and in general character
expansion of Villain action. We emphasise that the q-deformation in the Yang-Mills
side is related to special deformation of droplet geometries without deforming the
gauge group associated with the matrix model.
Contents
1 Introduction and summary 1

2 Unitary matrix model in (0 + 1) dimension 7

3 Quantisation of droplets 9
3.1 Quantisation and Kac-Moody algebra 10
3.2 The Hilbert space 12
3.3 Eigenstates 16

4 Evolution of classical droplets 17

4.1 Higher point functions 19
4.2 Connection to q-deformed theories 21
4.3 Joining two chiral sectors 22

5 Discussion 24

A Eigenstates of phase space Hamiltonian 25

B Reviewing u(N ) representations 26

B.1 Composite representations 26
B.2 Quadratic Casimirs of u(N ) and su(N ) 27

C Twisted surgery and higher point functions 28

D Villain action 29

1 Introduction and summary

The theory of random matrix integrals has achieved so many accolades that a very
few of the other toy models can ever reach. Random matrix integrals (in short
matrix models) have impacted both the realms of mathematics and physics almost
equally. Although, matrix models have not been very helpful in the context of critical
string theory or superstring theories but one of the most significant model, the c = 1
matrix model alias matrix quantum mechanics (MQM) did serve as an indispensable
tool to study lower dimensional bosonic string theories. Conventionally, the name
“c = 1 matrix model” is derived from the fact that the double scaling limit of this
one dimensional Hermitian matrix model represents a two-dimensional string theory

–1–
whose target space interpretation is that of a Liouville theory coupled with c = 1
matter. Early evidences of such connection were explored in [1, 2] and made robust
in the early 90’s by Gross and Klebanov [3]. For a more comprehensive overview, the
reader may refer to [4, 5]. Following the beautiful works laid down by [6, 7]1 , it was
shown in [14] that the same model can also describe a two dimensional Yang-Mills
theory with compactified spatial dimension. More interestingly [14] proved that the
same 2d Yang-Mills partition function can be rewritten as an one dimensional unitary
matrix model or a unitary matrix quantum mechanics (UMQM) for both U (N ) or
SU (N ) gauge groups2 . In [16] this observation was extended to include 2d Yang-
Mills defined on both cylinder and torus. One of the main aims of this paper is to
further investigate the correspondence between UMQM and 2d Yang-Mills defined
on higher genus Riemann surfaces.
2d Yang-Mills partition function with gauge group G on a Riemann surface Σg
with genus g can be written as [17]
X β
ZΣg = dim R2−2g e− 2 C2 (R) , (1.1)
R

where the sum is over all possible representations R of G. dim R is the dimension of
R. C2 (R) is the quadratic Casimir of R and β is a theory dependent constant. Due
to lack of propagating degrees of freedom in 2d Yang-Mills, the interesting objects
to studyH in this theory are the correlators of Wilson loops W (U ) = TrU where
U = P e γ A around different edges (or boundaries) of Σg . Following [18, 19] these
correlators are given by
X β
Kg,n ≡ hW (U1 ) · · · W (Un )i = dim R2−2g−n χR (U1 ) · · · χR (Un )e− 2 C2 (R) , (1.2)
R

where χR (Un ) is the character of the representation R evaluated on the holonomy

Un . One can then simply recover (1.1) from (1.2) by setting n to zero.
Starting with the partition function (1.1) and considering the gauge group G to
be U (N ) or SU (N ), [14] constructed the Hamiltonian by showing that the states
in the theory consist of interacting strings that wind around the circle. They also
observed that the Hamiltonian is equivalent to Das-Jevicki Hamiltonian [20, 21] of
the c = 1 matrix model with the condition that the spatial direction is compactified.
This provides an equivalence between 2d Yang-Mills and UMQM. In an alternate
approach, [16] showed that 2d Yang-Mills partition function on a cylinder or torus
is exactly same as one dimensional Kazakov and Migdal matrix model [22] with
eigenvalues distributed on a circle and can be interpreted as UMQM. Another way
to provide a free-fermionic description of 2d Yang-Mills is using quantum mechanics
on group manifold U (N ) [23]. This is arguably the most useful approach to the
1
See also [8–13] for more works on 2d Yang-Mills and string theory.
2
See [15] for connection between O(N ) or Sp(N ) 2d Yang-Mills theory and string theory.

–2–
problem since UMQM are nothing but a system of N free fermions, and therefore
the classical limit (N → ∞) can be best described in terms of phase space variables.
Contrary to the usual, in this paper we take a bottom-up approach to show the
equivalence between unitary matrix quantum mechanics and 2d Yang-Mills theory
and its variants via quantisation of phase space droplet. We explicitly construct the
partition function of 2d Yang-Mills theory on a generic Riemann surface with gauge
group U (N ) or q-deformed U (N ) from the evolution of free Fermi droplets in one
dimensional unitary matrix models. To be precise, we quantise the classical droplet
in the matrix model and construct the corresponding Hilbert space. The Hilbert
spaces H+ and H− associated with the upper and lower free Fermi surfaces of a
droplet admit a Young diagram basis in which the phase space Hamiltonian is diago-
nal with eigenvalue, in the large N limit, equal to the quadratic Casimir of u(N ). We
establish an exact mapping between the states in H± and the geometries of upper
and lower Fermi surfaces. In particular, coherent states in H± correspond to classi-
cal deformation of upper and lower Fermi surfaces. We then prove that correlation
between two coherent states in H± is equal to the chiral and anti-chiral partition
function of 2d Yang-Mills theory on a cylinder, establishing the fact that factorisa-
tion of 2d Yang-Mills in chiral and anti-chiral sectors is equivalent to independent
evolution of upper and lower Fermi surfaces. More generically we prove that chiral 2d
Yang-Mills partition function on a generic Riemann surface with n punctures (1.2)
have a rather simple interpretation in terms of correlations between coherent states
in H± . Using the fact that the full Hilbert space H+ ⊗ H− admits a composite basis,
we show that correlation between two classical geometries is equal to the full U (N )
Yang-Mills partition function on cylinder. There exists a special class of coherent
states in the Hilbert space - we call them q-deformed coherent states. q-deformed
coherent states are mapped to a particular geometries of Fermi surfaces - box shaped
surface. Correlation between these q-deformed coherent states is related to the char-
acter expansion of Villain action studied in [24]. Since the character expansion of
Villain action can be related to q-deformed Yang-Mills theory [25], our analysis gives
a dictionary between the correlators in UMQM Hilbert space and the heat kernel of
q-deformed Yang-Mills. We emphasise that the q-deformation in the Yang-Mills side
is related to special deformation of geometries without deforming the gauge group
associated with the matrix model. Therefore the droplet picture of unitary matrix
model is much more general and vivid. In some sense the geometry of droplet unifies
different versions of 2d Yang-Mills theories. A droplet contains more information
than it is expected.
We now elaborate our results in detail. The camaraderie between representation
theory and quantum field theory of free fermions, exploited earlier by [14, 23] is again
at the core of our current approach. We start with a generic (0 + 1) dimensional
unitary matrix model. The dynamics of eigenvalues can be described in terms of a
collective field ρ(t, θ) and its conjugate momentum π(t, θ) [21]. Classical dynamics

–3–
of collective fields also admits an equivalent description in terms of evolution of free
Fermi droplet [26]. This two pictures are related by the simple fact that solving
Hamilton’s equations for collective field and its conjugate momentum is equivalent
to finding upper and lower Fermi surfaces of a free Fermi droplet (denoted by p+ (t, θ)
and p− (t, θ) respectively in this paper). The advantage of the second picture is that
the equations of motion for p+ (t, θ) and p− (t, θ) are decoupled and hence their evo-
lution, except for the fact that p+ (t, θ) − p− (t, θ) ≥ 0 always as the difference is
equal to eigenvalue density up to a factor of 2π. The collective field theory Hamilto-
nian while written in terms of p± (t, θ) becomes diagonal (separable). Thus finding
eigenvalue configuration is equivalent to find the shape/geometry of the free Fermi
droplet. Unitary matrix quantum mechanics with zero potential admits a classical
solution p± = ± 21 . This solution corresponds to a uniform droplet and a constant
eigenvalue configuration. In this paper we quantise this classical solution/droplet.
However quantisation of other classical solutions can also be done in a similar way.
We list our main observations sequentially.

• Quantisation of droplets : We quantise the classical droplet by imposing

equal time commutation relations on p± (θ, t). Expanding p± (θ, t) in Fourier
modes, we find that the modes in + and − sectors individually satisfy the
abelian Kac-Moody algebra and the modes in two different sectors commute.
Similar quantisation was studied by [27] in the context of Hermitian matrix
quantum mechanics. Writing the Hamiltonian in terms of these modes we find
that the Hamiltonian contains a usual quadratic piece (free part) as well as a
cubic piece (interaction terms). This Hamiltonian is exactly the same as the
one obtained by [12, 14, 23], with the cubic term being similar to the string
splitting-joining interaction.

• The Hilbert space H : We construct the Hilbert space associated with the
quantised droplets. Demanding area preserving time evolution of the droplet,
we find that the zero modes of + and − sectors are equal up to a sign. This
constraint along with the fact that the zero modes commute with the Hamilto-
nian suggests that the Hilbert space can be constructed upon a one parameter
family of ground state |s . We take the parameter s to be integer, such the
the phase space momentum is quantised. A generic state in the Hilbert space
is given by action of creation operators associated with p± on the ground state
|s .

• Since + and − sectors are decoupled, the excitation in + sector are isomorphic
to those in the − sector. The full Hamiltonian is also given by a direct sum of
H+ and H− : H = H+ + H− . As a result, the total Hilbert space is a direct
product of Hilbert spaces for + and − sectors : H = H+ ⊗ H− . Evolution of
states in H+ and H− are independent of each other.

–4–
• Mapping between H and droplet : We then define a mapping between a
state |ψ ∈ H± and geometries of upper and lower Fermi surfaces respectively
: |ψ → { ψ|p± (θ)|ψ }. The mapping is one-to-one as long as p± (θ) is a
single valued function of θ. The expectation value of p± in the ground state
is s|p± |s = ± 12 + Ns with zero dispersion. Therefore the ground state |s
corresponds to an overall shift of p± by an amount s/N over the classical shape.
Since the eigenvalue density is proportional to the difference between p+ and
p− , such constant shift in upper and lower Fermi surfaces does not change
the eigenvalue distribution. Expectation value of p± in a generic normalised
excited state is same as the expectation value in the ground state. However
the quantum dispersion ∆p± in a generic excited state is not zero and goes like
1/N . Therefore, these states are quantum excitations over the classical shape.
In [14] such excitations in 2d Yang-Mills were identified with the left and right
winding of strings around the circle. These excited states form a basis in the
Hilbert space for a given ground state |s . Since + and − sectors are decoupled
and isomorphic, we consider excitations in + sector only before combining the
two sectors.

• Coherent states in H+ : We also construct coherent states in the Hilbert

space H+ . Expectation value of p+ in a coherent state is non-zero and gives a
finite deformation of p+ over the ground state. Quantum dispersion of p+ in
a coherent state is zero. Hence we call such states classical. A classical state
preserves the quadratic nature of droplets : for a given θ, p+ is unique.

• Diagonalisation of the Hamiltonian : It turns out that the interaction

piece in the Hamiltonian (of either sectors) is not diagonal in the above basis.
We define a new basis in terms of representations of permutation group, in
which the Hamiltonian is diagonal [23]. In representation basis H± are spanned
by representations built out of fundamental or anti-fundamental respectively.
In N → ∞ limit the eigenvalue of the Hamiltonian H± in representation basis
is equal to the second Casimir of u(N ) representation. The Hamiltonian H± is
same as that of a U (N ) 2d Yang-Mills theory in chiral (anti-chiral) sector.

• Evolution of coherent states in H+ and the cylinder amplitude : We

next study the evolution of classical states in the Hilbert space. We show that
the transition probability of an initial coherent state to a final coherent state
in time T is same as the chiral partition function of a U (N ) 2d Yang-Mills
theory on a cylinder with two specified holonomies. In the droplet picture
such a propagator gives the evolution of upper Fermi surface from one classical
shape to the other. The initial and the final shapes are therefore mapped to
two holonomies in the 2d Yang-Mills side.

–5–
• Disc, sphere and torus : There exists a special coherent state which corre-
sponds to p+ (θ) ∼ δ(θ). The transition amplitude of a generic coherent state
to this special state is mapped to chiral disc partition function of 2d Yang-
Mills theory. If both the initial and final shapes are delta functions, then such
propagators are mapped to sphere partition function of the 2d Yang-Mills in
the chiral sector. Defining appropriate surgery in the space of coherent states
we also generate the torus partition function.

• Higher point correlators : We define a unique operator associated with a

coherent state. n-point correlations of these operators in a mixed ensemble
(weight factor depends on an integer g ≥ 0) in the Hilbert space H+ can be
related to a generic Yang-Mills amplitude (1.2) in a chiral sector. For example
transition amplitude between two coherent states, discussed above is same as
two point correlator in a mixed ensemble with g = 0.

• q -deformed theory : The correspondence between the correlators of coherent

states and heat kernel of 2d Yang-Mills can be naturally extended to q-deformed
Yang-Mills theory. We define a new class of coherent states as a function of q.
Projection of such states along representation basis gives q-deformed dimension
of the corresponding representations [23]. Considering the deformation param-
eter q = eigs and taking the double scaling limit gs → 0, N → ∞ keeping
gs N = λ fixed, the expectation value of p+ operator in such states is given by a
box function symmetrically distributed about θ = 0 with height 1/λ and width
λ. With the aid of these special class of coherent states we are able to reproduce
the sphere and disc partition function of q-deformed Yang-Mills. The sphere
partition function corresponds to evolution of a q-deformed coherent state to
itself. Evolution of a q-deformed coherent state to generic coherent state maps
to a disc partition function.

• Connection with Villain action : Going a step further, we also compute the
correlation between two different q-deformed coherent states. In the droplet
picture such a correlation corresponds to evolution of a box distribution to
another box distribution keeping the area preserved. These correlators do not
have any direct consequence in q-deformed theories. However, we show that
they appear in the context of character expansion of Villain action [24]. Inte-
grating two different U (N ) Villain actions (with two different parameters) over
U (N ) group manifold one obtains such amplitudes.

• Joining two chiral sectors : In order to recover the full 2d Yang-Mills

partition function, one has to consider the contribution of propagators coming
from states in both the Hilbert spaces H+ and H− . Tools for factorizing Yang-
Mills theories into chiral and anti-chiral sectors have been developed using

–6–
 the notion of composite representations. We show that the full Hilbert space
H+ ⊗ H− admits a composite representation basis. Expressing the evolution
amplitudes in this basis we recover the full partition of U (N ) Yang-Mills theory
in 2d.
We have structured the paper as follows. In section 2, we briefly discuss the
construction of classical phase space in matrix quantum mechanics. Quantisation of
classical phase space is given in section 3. The connection between droplet evolution
and partition functions of (q-deformed) 2d Yang Mills theory is discussed in section
4. Further in appendix A, we provide the eigenvalue analysis for the phase space
Hamiltonian. A brief review of composite representation and some comments on
irreducible representations of su(N ) and u(N ) appears in appendix B. In appendix
C, we elaborate on twisted surgery and explicitly write down an expression for 3-
point function (however the process can in principle to generalized to obtain general
n-point functions). Finally, in appendix D we discuss about the Villain action and
its character expansion.

2 Unitary matrix model in (0 + 1) dimension

Partition function for a unitary matrix model in (0 + 1) dimension is given by
Z Z  
2
Zt = [DU ] exp dt TrU̇ + W (U ) . (2.1)

Following the beautiful work by Jevicki and Sakita [21, 28] one can describe the
matrix model (2.1) in terms of a real collective bosonic field ρ(t, θ) (the eigenvalue
density) and its conjugate momentum π(t, θ). The corresponding Hamiltonian is
given by,
∂π(t, θ) π 2 ρ3 (t, θ)
Z  
1 ∂π(t, θ)
HB = dθ ρ(t, θ) + + W (θ)ρ(t, θ) (2.2)
2 ∂θ ∂θ 6
R
where dθW (θ)ρ(t, θ) = W (U (t)). The Hamilton’s equations for ρ(t, θ) and π(t, θ)
are given by,
∂t ρ(t, θ) + ∂θ (ρ(t, θ)v(t, θ)) = 0
1 π2 (2.3)
∂t v(t, θ) + ∂θ v(t, θ)2 + ∂θ ρ(t, θ)2 = −W 0 (θ)
2 2
where
v(t, θ) = ∂θ π(t, θ). (2.4)
These are coupled, non-linear partial differential equations and hence it is difficult
to find a solution in general. One can decouple these two equations by introducing
two new variable p± (t, θ)
p+ (t, θ) − p− (t, θ) p+ (t, θ) + p− (t, θ)
ρ(t, θ) = , and v(t, θ) = . (2.5)
2π 2

–7–
The equations for p± (t, θ) become

∂t p± (t, θ) + p± (t, θ)∂θ p± (t, θ) + W 0 (θ) = 0. (2.6)

The set of decoupled equations (2.6) governs the evolution of a free-Fermi droplet
in (p, θ) plane whose boundaries are given by p± (t, θ) [26]. To understand this in
detail, consider a system of N free Fermions (non-interacting) moving on S 1 under
a common potential W (θ). The single particle Hamiltonian is given by

p2
h(p, θ) = + W (θ). (2.7)
2
The Hamilton’s equations obtained from the single particle Hamiltonian (2.7) are
given by
dp dθ
= −W 0 (θ), = p. (2.8)
dt dt
Using these equations one can check that the boundaries of a droplet p → p± (t, θ)
follow equation (2.6). Therefore equations (2.6) determines classical evolution of
Fermi surface with time. The phase space Hamiltonian for such free Fermi system is
given by, Z  2 
1 p
Hp = dθ dp + W (θ) $(p, θ) (2.9)
2π 2
where $(p, θ) is the phase space density

$(p, θ) = Θ ((p+ (t, θ) − p)(p − p− (t, θ))) . (2.10)

There is a one to one correspondence between phase space variables and collective
field theory variables. Eigenvalue density and the corresponding momentum can be
obtained from phase space distribution by integrating over p
Z Z
1 1
ρ(t, θ) = dp $(p, θ), v(t, θ) = dp p $(p, θ). (2.11)
2π 2πρ
Thus the relations (2.5) serve as a dictionary between bosonic and fermionic (phase
space) variables. Integrating over p in (2.9) we have,

p+ (t, θ)3 p− (t, θ)3

Z   Z  
1 1
Hp = dθ + W (θ)p+ (t, θ) − dθ + W (θ)p− (t, θ) .
2π 6 2π 6
(2.12)

Using the dictionary (2.5) the Hamiltonian (2.9) reduces to Hamiltonian (2.2). Thus
we see that the matrix model (2.1) has two equivalent descriptions.
Solving the field theory equations of motion (2.3) is equivalent to solving for
upper and lower Fermi surfaces in phase space picture. In either case, one needs

–8–
to provide an initial data on a constant time slice in (t, θ) plane. After that the
problem reduces to a Cauchy problem. Existence of a unique solution depends on
the geometry of initial data curve3 .
For a unitary matrix model, phase space distribution has θ → −θ symmetry.
Using this fact it is possible to show that the phase space area covered by Fermi
surfaces is a constant of motion. Area covered by Fermi surfaces at a time t is given
by Z θ0
A(t) = (p+ (t, θ) − p− (t, θ)) dθ. (2.13)
−θ0

Using equation (2.6) and the fact that p± (t, θ) and W (θ) are symmetric functions of
θ it is easy to verify that
d
A(t) = 0. (2.14)
dt
Thus, phase space area is preserved during classical time evolution. One can nor-
malise the area to be unity : A(t) = 1. It is only the shape which changes during
evolution.

3 Quantisation of droplets
In order to regularise the total number of states in phase space we divide the phase
space into unit cells with volume ~ such that
Z
1
dpdθ$(p, θ) = N, with ~N = 1. (3.1)
2π~
The classical limit corresponds to N → ∞, ~ → 0 with ~N = 1. We also modify
our phase space Hamiltonian (2.12) accordingly and is given by
p+ (t, θ)3
Z  
1
Hp = dθ + W (θ)p+ (t, θ)
2π~ 6
(3.2)
p− (t, θ)3
Z  
1
− dθ + W (θ)p− (t, θ) .
2π~ 6
Our goal now is to find a simplectic form on the phase space [30, 31], so that the
Hamilton’s equation
ṗ± (t, θ) = {p± (t, θ), Hp } (3.3)
coincides with (2.6), where Hp is the total Hamiltonian (3.2). To achieve this goal
we introduce equal time Poisson brackets between p± (t, θ) and p± (t, θ0 )

{p± (t, θ), p± (t, θ0 )} = ±2π~δ 0 (θ − θ0 ) and {p+ (t, θ), p− (t, θ0 )} = 0. (3.4)

It is easy to check that using these Poisson brackets the equation (3.3) boils down
to (2.6).
3
See [29], for an example.

–9–
3.1 Quantisation and Kac-Moody algebra
(0)
Suppose p± (t, θ) are solutions of equation (2.6). Consider fluctuations p̃± (t, θ) about
such classical solutions
(0)
p± (t, θ) = p± (t, θ) + ~ p̃± (t, θ). (3.5)

The fluctuations p̃± (t, θ) satisfy

(0)
∂t p̃± (t, θ) + ∂θ (p± (t, θ)p̃± (t, θ)) + ~ p̃± (t, θ)∂θ p̃± (t, θ) = 0. (3.6)

This equation follows from the Hamiltonian

 3
~2 (0)
Z 
1 ~ 3 2
H̃p = dθ p̃+ (t, θ) + p+ (t, θ)p̃+ (t, θ)
2π~ 6 2
 3 (3.7)
~2 (0)
Z 
1 ~ 3 2
− dθ p̃− (t, θ) + p− (t, θ)p̃− (t, θ)
2π~ 6 2

with the following Poisson bracket relations

2π 0
{p̃± (t, θ), p̃± (t, θ0 )} = ± δ (θ − θ0 ) and {p̃+ (t, θ), p̃− (t, θ0 )} = 0. (3.8)
~
We also assume that the fluctuations preserve the total area of the droplet. This
implies that
Z π
dθ(p̃+ (t, θ) − p̃− (t, θ)) = 0. (3.9)
−π

The unitary matrix model (2.1) admits a minimum free energy classical con-
(0)
figuration given by circular droplet p± (t, θ) = ± 21 when W (θ) is constant. We
study quantum fluctuations about this solution. However our analysis can be fol-
lowed to study quantum fluctuations about other classical configurations as well. For
(0)
p± = ± 21 the Hamiltonian (3.7) is given by
π π
~2
Z Z
~
p̃2+ (t, θ) p̃2− (t, θ) p̃3+ (t, θ) − p̃3− (t, θ) dθ. (3.10)



H̃p = + dθ +
8π −π 12π −π

p̃± satisfies
1
∂t p̃± (t, θ) + ∂θ p̃± (t, θ) + ~ p̃± (t, θ)∂θ p̃± (t, θ) = 0. (3.11)
2
To quantise the above classical system we promote the Poisson brackets (3.8) to
commutation relations

[p̃± (t, θ), p̃± (t, θ0 )] = ±2πiδ 0 (θ − θ0 ) and [p̃+ (t, θ), p̃− (t, θ0 )] = 0. (3.12)

– 10 –
We decompose p̃± (t, θ) into Fourier modes
∞
X
p̃+ (t, θ) = a−n (t)einθ (3.13)
n=−∞

and
∞
X
p̃− (t, θ) = − bn (t)einθ . (3.14)
n=−∞

The constraint (3.9) implies that the zero-modes a0 and b0 are equal up to a sign

a0 = −b0 = π0 . (3.15)

It follows from the quantisation conditions (3.12) that the Fourier modes an and bn
satisfy u(1) Kac-Moody algebra

[am (t), an (t)] = mδm+n , [bm (t), bn (t)] = mδm+n , and [am (t), bn (t)] = 0. (3.16)

The Hamiltonian (3.10) in terms of these modes are given by

~X
H̃p = (an (t)a−n (t) + bn (t)b−n (t))
4 n
(3.17)
~2 X

+ am (t)an (t)a−(m+n) (t) + bm (t)bn (t)b−(m+n) (t)
6 m,n

The phase space Hamiltonian is not a free Hamiltonian, it contains a cubic inter-
action. Our next goal is to construct the Hilbert space for the system of quantised
droplet and set up a map between different states in the Hilbert space and shapes of
droplets (excitations).
Before constructing the Hilbert space we note that in ~ → 0 limit the quantum
excitations p̃± (t, θ) can be related to primary fields associated with a theory of free
bosons moving on a cylinder. We use the equations of motion (3.11) in ~ → 0 limit
and find that the time evolution for an (t) and bn (t) are given by

ak (t) = ak (0)eikt/2 , bk (t) = bk (0)eikt/2 . (3.18)

Therefore p̃± can be written as (we define an ≡ an (0) and bn ≡ bn (0))

X X it
p̃+ (t, θ) = an e−n(τ +iθ) and p̃− (t, θ) = bn e−n(τ −iθ) where τ = (3.19)
n n
2

Defining z = eτ +iθ and z̄ = eτ −iθ we see that p̃+ is a holomorphic function of z and
p̃− is an anti-holomorphic function of z̄
X X
p̃+ (z) = π0 + ak z −k and p̃− (z̄) = π0 + bk z̄ −k (3.20)
k6=0 k6=0

– 11 –
and the modes satisfy u(1) Kac-Moody algebra. Thus, in the N → ∞ limit p+ and
p− are related to holomorphic and anti-holomorphic conserved currents associated
with a free scalar CFT on a cylinder of radius one 4 .
The Hamiltonian H̃p also separates into two parts : H̃p = H+ + H− . Since the
Hamiltonian (3.17) does not depend on time explicitly, H± are given by (up to an
overall constant)

H0+
z }| {
~ 2 ~2 ~2 3
H+ = a0 − a0 + a0
4 24 6
~ X † ~2 X  †  (3.23)
+ (1 + 2~a0 ) ak ak + am+n am an + h.c.
2 k>0
2 m,n>0
| {z } | {z }
+ +
Hfree Hint

and similarly

~ 2 ~2 ~2 ~ X † ~2 X  † 
H− = b0 − b0 + b30 + (1 + 2~b0 ) bk bk + bm+n bm bn + h.c. .(3.24)
4 24 6 2 k>0
2 m,n>0

Both the Hamiltonians have two parts, a free part and an interacting part. This
Hamiltonian is similar to the Hamiltonian obtained by [14, 23, 26] for splitting-joining
of strings. Here the interaction piece is responsible for the interaction between the
boundary excitations.

3.2 The Hilbert space

We now construct the Hilbert space for ‘+’ sector. There exists an isomorphic Hilbert
space for ‘−’ sector. We denote these Hilbert spaces by H± and the total Hilbert
space is therefore given by H = H+ ⊗ H− .
Since π0 commutes with all the an , bn and hence with H̃p , application of an ’s and
bn ’s cannot change the eigenvalue of π0 . Therefore the Hilbert space is constructed
4
From the above construction it is easy to see that the fluctuations p̃± can be related to conserved
currents associated with a free scalar on a cylinder of radius one. We define a scalar field on a
cylinder of radius one
X1
an z −n + bn z̄ −n


ϕ(z, z̄) = ϕ0 − iπ0 ln(z z̄) + i (3.21)
n
n6=0

such that
π0 X p̃+ (z) ¯ = π0 X p̃− (z̄)
i∂ϕ = + an z −n−1 = , and i∂ϕ + bn z̄ −n−1 = . (3.22)
z z z̄ z̄
n6=0 n6=0

– 12 –
upon a one parameter family of vacua |s, s ≡ |s where,

an |s = 0, bn |s = 0 for n > 0
(3.25)
and a0 |s = −b0 |s = π0 |s = s|s .

The Hilbert space constructed upon |s vacuum is denoted by H : s charged module

generated over the primary |s . We take s to be integer in order to get the phase
space momentum (2.11) quantised in the units of ~.
A generic excitation above the ground state is given by
∞
Y
kn
lm
|~k, ~l = a†n b†m |s . (3.26)
n,m=1

The ~k and ~l sectors correspond to excitations in upper and lower Fermi surfaces.
Since an and bn commute, generic excitation |~k, ~l ∈ H can be written as |~k ⊗ |~l .
The time evolution of the vectors in H± are governed by H± respectively and are
independent, except for the fact that p+ − p− ≥ 0. We first consider the states and
their evolution in H+ only. H− can be studied similarly5 . Later in section 4.3 we
combine the evolution of classical states in these two sectors and show how they are
related to correlation functions of 2d Yang-Mills theories on Riemann surfaces.
The excited states in H+ are orthogonal with the normalization
Y
k~0 |~k = z~k δ~kk~0 where z~k = kj !j kj (3.27)
j

and has π0 eigenvalue s. They also satisfy the completeness relation

X 1
|~k ~k| = IH+ (3.28)
z~k
~k

and hence form a basis in H+ . These states are particle like excitation above the
ground state |s . The excited states ~k in either sectors are eigenstates of the free
Hamiltonian
∞
!
± ~ ~ X
Hfree |k = (1 ± 2s~) nkn |~k , (3.29)
2 n=1

but not an eigenstate of the full Hamiltonian. The interaction Hamiltonian changes
|~k state to |~k 0 keeping the level fixed, i.e. n nkn = n nkn0 . The expectation value
P P

of p+ (t, θ) operator in |~k state is (1/2 + ~s)z~k with a non-zero quantum dispersion
∆p+ which goes as ~. Therefore |~k states are quantum excitations over the ground
state : ripples on Fermi surface. In [14] such excitations in 2d Yang-Mills were
5
One can study an entangled excitations of free Fermi droplets. We do not discuss such excita-
tions in this paper.

– 13 –
identified with the left and right winding of strings around the circle - excitation of
kn closed strings winding n times around the circle.
One can also define coherent state in H+
∞
!
+ †
X τ n an
|τ+ = exp |s . (3.30)
n=1
n~

The state |τ+ is not normalised, one can show that

!
X τ+a τ+b
τ+a |τ+b = exp . (3.31)
n
n~2

The coherent state |τ+ is an eigenstate of an (∀ n > 0) operator with eigenvalue

τn+ /~. A coherent state |τ+ can be expanded in |~k basis in the following way
X τ~+ Y  τ + km
k ~ + m
|τ+ = |k , where τ~k = . (3.32)
z~k m
~
~k

The expectation value of p+ operator in a coherent state |τ+ is given by

τ+ | p+2π(z) |τ+
 
1 s~ 1 X + n 1
ωτ+ (z) = = + + τ z + n . (3.33)
τ+ |τ+ 4π 2π 2π n>0 n z

Value of ωτ+ (z) on the unit circle (|z| = 1) in the complex z plane is given by,
1 s~
ωτ+ (θ) ≡ ωτ+ (z = eiθ ) = + + ω̃τ+ (θ)
4π 2π
1X + (3.34)
where ω̃τ+ (θ) = τ cos nθ.
π n>0 n

The quantum dispersion of p+ in a coherent state is zero. Therefore such states are
called classical.
We now define the following mapping between a state |ψ ∈ H+ and shape of
the upper Fermi surface
|ψ → { ψ|p+ (θ)|ψ }. (3.35)
This mapping maps the following three types of states in H+ to three different types
of shapes of the upper Fermi surface p+ .
• Expectation value of p+ in ground state is s|p+ |s = 12 + s~ with zero disper-
sion. Therefore the ground state |s corresponds to an overall shift of p+ by an
amount s/N over the classical value. Similar mapping exists in the − sector
as well and the ground state corresponds to the same shift in p− . Since the
eigenvalue density is proportional to the difference between p+ and p− , such
constant shift in upper and lower Fermi surfaces does not change the eigenvalue
distribution.

– 14 –
 • Expectation value of p+ in a generic normalised excited state is same as the
expectation value in the ground state with non-zero dispersion6 . Therefore
excited states correspond to O(~) ripples on p+ .

• The relation (3.34) defines a mapping between a coherent states |τ+ in H+ and
a classical deformation of droplet in ‘+’ sector over the ground state. Specifying
a coherent state is equivalent to specifying a O(1) deformation ωτ+ (θ) of p+ .
The mapping is one-to-one as long as a classical distribution ψ|p+ (θ)|ψ is a
single valued function of θ.

An important thing to note here is that classical deformations do not disturb the
quadratic profile of the droplets i.e for a given θ, there exists unique values of p± (θ).
There are other types of excitation, which destroy the quadratic profile of a droplet

Figure 1. Different Fermi surfaces.

- formation of folds. We shall discuss about fold states in a future work.

6
This is similar to the expectation of position or momentum operator of a simple harmonic
oscillator in the ground state or higher excited states.

– 15 –
3.3 Eigenstates
The basis states |~k in H+ are not eigenstates of the full Hamiltonian H+ . We
introduce a new basis in the Hilbert space H+ - Young diagram basis. We associate
a state |R+ in H+ for a given Young diagram R+ in the following way
X χR+ (~k)
|R+ = |~k (3.36)
z~k
~k

where χR+ (~k) is the character of the conjugacy class C(~k) of the permutation group
SK . |R+ corresponds to a Young diagram R+ whose rows have lengths l1 ≥ l2 ≥
l3 ≥ · · · ≥ lr ≥ 0, with r being the number of rows. Total number of boxes in R+ is
equal to the level of |~k states : i ni = n nkn . Following the normalization (3.27)
P P

of |~k we have
0
R+ |R+ = δR+ R+0 . (3.37)
Inverting the relation (3.36) we have
X
|~k = χR+ (~k)|R+ . (3.38)
R+

Thus |~k and |R+ are two equivalent basis of the Hilbert space H+ . The young
diagram basis also has interpretations in terms of fermionic excitations following
bosonisation [23, 32].
It turns out that (see appendix A) the Hamiltonian (3.23) is diagonal in the
Young diagram basis
h ~2 ~ 2 ~2 ~ ~2 i
3
H+ |R+ = s + s − s + (1 + 2s~) l(R+ ) + κR+ |R+ ≡ E(R+ , s)|R+
6 4 24 2 2
(3.39)
where
r
X r
X
l(R+ ) = li and κR+ = l(R+ ) + (li2 − 2ili ). (3.40)
i=1 i=1

Note that eigenvalues of the free and the interaction part of the Hamiltonian are
of the same order ~l(R+ ) ∼ ~2 κR+ in the large N limit. Therefore, the cubic part
(interaction) can not be treated as perturbation in the large N limit.
The quadratic Casimir C2 (R+ ) of u(N ) representation is given by

C2 (R+ ) = N l(R+ ) + κR+ . (3.41)

From equation (3.39) we see that in the limit N → ∞ the eigenvalue E(R+ , s) of
|R+ is equal to quadratic Casimir C2 (R+ ) for R+ representation

~2
E(R+ , s) = C2 (R+ ) + O(~). (3.42)
2

– 16 –
The Hilbert space, discussed above, is similar to the Hilbert space of a chiral sector of
U (N ) 2d Yang-Mills on Riemann surfaces7 . The states (3.26) in H+ , with all lm = 0
are the states of interacting strings that wind around the circle in one direction. The
Hamiltonian H+ is thus the Hamiltonian for the closed strings dual to the chiral
sector of two dimensional Yang-Mills [7, 14, 23, 33–35]. The evolution of the upper
Fermi surface p+ , therefore, maps to dynamics of string degrees of freedom dual to
the chiral sector of 2d Yang-Mills. The evolution of lower Fermi surface p− , in a
similar way, provides the other chiral sector.

4 Evolution of classical droplets

In quantum mechanics all the information of a system is stored in the propagator :
the transition amplitude from an initial state |i at ti = 0 to a final state |f at time
tf = T , given by
i
Ki→f (T ) = f |e ~ HT |i . (4.1)

The Euclidean version of this amplitude is obtained by replacing T → −iT

1
Ki→f (T ) = f |e− ~ HT |i . (4.2)

We define S+ : a set of all classical droplet configurations over the ground state in
H+ . There exists a one-to-one mapping between S+ and the coherent states of H+ .
Our goal is to compute the propagators8 in S+ .
In order to calculate the transition amplitude between coherent states |τ+a and
|τ+b in S+ we first expand a coherent state in the Young diagram basis
 
X X χR+ (~k)
|τ+ =  τ~k+  |R+ . (4.4)
R
z ~k
+ ~k

The term inside the parenthesis can be simplified using equation (3.34). τk+ is given
by Z π
+
τk = dθ ω̃τ+ (θ)eikθ . (4.5)
−π

7
The complete Hamiltonian can be obtained by considering excitations in both p+ and p− sectors.
8
If we take the initial state to be a representation |R , it will remain in the same state as |R is
an eigenstate of H+ . A transition between initial state |~k and and final state |~l is given by
X
K(~k → ~l; T ) = χR (~k)χR (~l)e− 2 E(R)T .
~
(4.3)
R

This amplitude is zero if the levels of |~k and |~l are different.

– 17 –
Since ω̃τ+ (θ) defines a distribution of θi in N → ∞ limit, we can write
N
1 X 1
ω̃τ+ (θ) = lim δ(θ − θi ) − . (4.6)
N →∞ N 2π
i=1

Hence, we have
τk+ X ikθi
= e . (4.7)
~ i

This equation provides a map between infinite dimensional τ + = {τ1+ , τ2+ , · · · } space
and the set of angular variables {θ1 , · · · , θN } in N → ∞ limit. This mapping is one
to one - a coherent state |τ+ ≡ {τn+ } corresponds to a unique point {θi } in θ space.
Using (4.7) we can write
X χR+ (~k)
τ~k+ = TrR+ U (∞) ≡ sR+ (τ+ ), (4.8)
z~k
~k

where sR+ (τ+ ) is Schur polynomial9 . Thus we have

X
|τ+ = sR+ (τ+ )|R+ . (4.9)
R+

From the closed string point of view sR+ (τ+ ) is the wave function of chiral winding
string states in |R+ basis [35]. A coherent state in H+ captures information of
winding states in the chiral sector of closed string theory. Thus a classical shape of
upper Fermi surface has a correspondence with winding states in the chiral sector of
closed strings dual to 2d Yang-Mills.
We now consider the transition amplitude between two coherent states |τ+a and
|τ+b in time interval T
1
K+ (a → b, T ) = τ+b |e− ~ H+ T |τ+a . (4.10)

Using relation (4.9) we can express this transition amplitude as a sum over represen-
tations
X ~
K+ (a → b, T ) = sR+ (τ+a )sR (τ+b )e− 2 C2 (R+ )T . (4.11)
R+

The transition amplitude between two coherent states in time T is same as the chiral
partition function of 2d Yang-Mills theory on a cylinder with holonomies specified at
the two circular ends [7, 12, 13, 34]. The droplet profiles of the initial and the final
coherent states are mapped to these two holonomies. We denote such an amplitude
by C+ (a, b, T ).
9
In equation (4.8), Schur polynomial sR+ is a function of θi s. Since equation (4.7) maps τi+
space to θi space we write sR+ as a function of τ+ .

– 18 –
 There is a special point in τ space, τ ∗ : τi+ = 1, ∀i. Droplet profile for such a
coherent state is given by
1 ~s
ωτ ∗ (θ) = − + + δ(θ). (4.12)
4π 2π
Schur polynomial for such a distribution is equal to the dimension of the represen-
tation R+ , denoted by dim R+ . If the initial (or final) configuration corresponds to
this particular configuration then C+ (τ ∗ → b, T ) is given by
X ~
C+ (τ ∗ , b, T ) = sR+ (τ b )dimR+ e− 2 C2 (R+ )T . (4.13)
R+

The delta function distribution of ωτ (θ) at one end of the cylinder is equivalent to
shrinking the radius of that end to zero. As a result the corresponding amplitude
becomes a disk amplitude D+ (b, T ) ≡ C+ (τ ∗ , b, T ). For both τ+a = τ+b = τ ∗ , we get
X
(dimR+ )2 e− 2 C2 (R+ )T .
~
C+ (τ ∗ , τ ∗ , T ) = (4.14)
R+

This is the chiral partition function of 2d Yang-Mills on a sphere.

We next define a surgery in τ+ space to generate higher point function. Consider
a cylinder C+ (a, b, A1 ) and a disk D+ (c, A2 ). To join the outer edge of the disk with
one side (say the b side) of the cylinder we first set τ+c = τ+b∗ . Then sum over all
possible coherent states |τ+b . Since τ+ space is isomorphic to θ space, the integration
in τ+ space can be replaced by an integration in θ space with the Vandermonde factor
θ −θ
V (θi ) = i<j sin2 i 2 j :
Q

θi − θj
Z Z Y
[dτ+ ]sR+ (τ )s 0
R+ (τ ) → [dθ] sin2 sR+ (θ)sR+0 (θ) = δR+ R+0 . (4.15)
i<j
2

Therefore we get,
Z
[dτ+b ]C+ (a, b, T1 )D+ (b, T2 ) = D+ (a, T1 + T2 ). (4.16)

Following the surgery one can also find a torus partition function of 2d Yang-Mills
in the chiral sector
Z
Z = dτ+a C+ (a, a, A) = Tre−T H+ . (4.17)

4.1 Higher point functions

In 2d Yang-Mills theory one can define higher point functions : partition function
with more than two holonomies. Such partitions are mapped to a Riemann surface
with more than two punctures.

– 19 –
 We define an operator O+ (τ+ ) associated with a coherent state |τ+ in H+
X χR+ (τ+ )
O+ (τ ) = |R+ R+ |. (4.18)
R+
dimR+

This operator has the properties

(a) O+ (τ ∗ ) = I, (b) [O+ (τ+a ), O+ (τ+b )] = 0,

(4.19)
(c) [O+ (τ+a ), H] = 0, (d) O+ (τ+ )|τ ∗ = |τ+ .

Following (3.34), the operator (4.18) defines a correspondence between classical shape
ω̃(τ+ ) and operators O+ (τ )
p̂+ (z)
ω̃(τ+ ) = τ ∗ |O+ (τ+ ) O+ (τ+ )|τ ∗ . (4.20)
2π z=eiθ

We introduce a density matrix for a mixed ensemble Eg+ in H+ for g = 0, 1, 2, · · ·

X
(dimR+ )2−2g e− 2 C2 (R+ )T |R+ R+ |.
~
ρ+g = (4.21)
R+

The partition function in Eg+ is given by

X
(dimR+ )2−2g e− 2 C2 (R+ )T .
~
Zg+ = Trρ+ g = (4.22)
R+

Ensemble average of any operator A is defined as

hAig = Tr(Aρg ).

Therefore the ensemble average of a product of n operators O+ (τ+a1 ), · · · , O+ (τ+an )

is given by

hO+ (τ+a1 ) · · · O+ (τ+an )ig = Tr O+ (τ+a1 ) · · · O+ (τ+an )ρ+

g
n
X Y (4.23)
(dimR+ )2−2g−n e− 2 C2 (R+ )T
~
= sR+ (τ+ai ).
R+ i=1

For g = 0, n = 1 and n = 2 we get back our disc and cylinder amplitude respectively.
n = 0 gives the partition function of 2d Yang-Mills on a generic Riemann surface in
the chiral sector. The operator O+ (τ+ ), which corresponds to a classical distribution
of droplet, creates a puncture on the Riemann surface. Three (and higher) point
correlators in H+ can also be thought of as evolution of initial droplet to a final one
in the presence of some external disturbances in between. Such external disturbances
can be mathematically expressed as local twists. For example, a 3-point correlator
can be interpreted as twisted surgery between two cylinders. See appendix C for
details.

– 20 –
4.2 Connection to q -deformed theories
There exists another interesting class of coherent states given by [32]
"∞ #
q
X γ n2 − γ − n2
|τ+ = exp n
−n
a†n |s (4.24)
n(q 2 − q 2)
n=1

where q is a deformation parameter such that 0 < q < 1 and γ = q N . In the limit
q → 1 we find that
"∞ #
q
X 1
lim |τ+ = exp an |s = |τ ∗ .
†
(4.25)
q→1
n=1
n~

The q-deformed state |τ+q has the property

Y [hi − hj ]
R+ |τ+q = dimq R+ = (4.26)
1≤i<j≤N
[j − i]

where dimq R+ is q-deformed dimension of R+ . Therefore the q-deformed state can

be written as
X
|τ+q = (dimq R+ ) |R+ . (4.27)
R+

Using the state droplet mapping one can find the shape of droplet for a q-
deformed coherent state. We take the deformation parameter q = eigs and then
consider a double scaling limit gs → 0, N → ∞ keeping gs N = λ fixed. In this limit
ωτ+q (θ) is given by10

λ2
 
1 ~s 1 2
ω (θ) = −
q
τ+ + + Θ −θ . (4.29)
4π 2π λ 4
The q → 1 limit corresponds to λ → 0. In this limit, the theta function approaches
to δ(θ), as expected.
The Lorentzian amplitude (4.1) from a |τ+q state to a |τ+b state is given by11
1
X
K(τ+q → τ+b , T ) = sR+ (τ+b )dimq R+ q 2 C2 (R+ ) where ~T = gs . (4.30)
R+

This amplitude is same as the disc amplitude of a q-deformed Yang-Mills theory

[36]. This is also similar to the character expansion of a “lesser known” generalized
10
Here we use the following definition of q-deformation

[x] = q x/2 − q −x/2 . (4.28)

11
We consider Lorentzian amplitude since we have taken q = eigs .

– 21 –
U (N ) Villain action (D.3) (up to a normalization factor) [24, 25]. We have given a
detailed discussion on Villain action in appendix D. In H+ , one can also consider an
evolution of a coherent state to another coherent state : |τ+q1 → |τ+q2 . In the droplet
picture such an evolution corresponds to a box distribution evolving to another box
distribution keeping the area preserved. The amplitude for such evolution is given
by
1 1
C2 (R+ ) C2 (R+ )
X
dimq1 R+ dimq2 R+ q12 q22 . (4.31)
R+

Such a propagator appears when we glue two Villain actions with different ’t Hooft
coupling λ1 and λ2 such that λ1 + λ2 = T . The amplitude (4.31) can be thought of
as gluing two different q-deformed disc over the edges. In (4.31) if we take q1 = 1
(or q2 = 1) we get
1
C2 (R+ )
X
dimR+ dimq2 R+ q22 . (4.32)
R+

This is the transition amplitude between |τ+q2 and |τ ∗ . In the large N limit these
amplitudes give rise to a mixed Riemann-Hilbert problem [37]. Such Riemann-Hilbert
problems appear in different contexts both in physics and mathematics. In physics,
for example, they appear in open topological-A theory amplitudes on some spe-
cific Calabi-Yau caps [38] and thus in the context of black hole microstate count-
ing in type II string theory owing to the Ooguri-Strominger-Vafa (OSV) conjecture
[39]. In mathematics, the Riemann-Hilbert problem associated with the q-deformed
Plancherel growth belong to the same class [40].
We see that the q-deformation in the Yang-Mills side is related to special de-
formation of droplet geometries without deforming the gauge group associated with
the matrix model. Thus the geometry of droplet unifies different versions of 2d
Yang-Mills theories. A droplet contains more information than it is expected.

4.3 Joining two chiral sectors

The Hilbert space H+ captures only one chiral sector of the Yang-Mills theory on a
circle [7]. The propagator defined in (4.11), gives only the chiral partition function of
the 2d Yang-Mills theory on cylinder. In order to recover the full partition function,
one has to consider the contribution of propagators coming from states in both the
Hilbert spaces H+ and H− . Tools for factorizing Yang-Mills theories into chiral and
anti-chiral sectors has been developed and discussed previously [7, 35, 41] using the
notion of composite representations (we elaborate on this further in appendix B). A
composite representation R+ R̄− can be built from the Young diagrams of R+ and
R− by drawing the diagram corresponding to R+ in a standard way on top right and
the Young diagram of R− as anti-boxes in the bottom left corner turning it upside

– 22 –
down. The remaining rows have zero boxes such that the total number of rows in
the composite diagram is N . Note that such a procedure makes sense where each
Young diagram R+ and R− has less than N2 rows. However, in N → ∞ limit one can
extrapolate this procedure for any R+ and R− . The representations corresponding to
the Young diagram of R+ R̄− (as described above) forms a basis for U (N ). Another
basis for U (N ) is given by R+ ⊗ R− . The relation between the two basis is given by
!
X X R R
|R+ ⊗ |R̄− = NS++S 0 NS−−S 0 |S+ S̄− (4.33)
S+ ,S− S0

and the inverse relation

!
|R0 |
X X S S
|S+ S̄− = (−1) NR++ R0 NR−− R̄0 |R+ ⊗ |R̄− . (4.34)
R+ ,R− R0

A generic coherent state for a 2d droplet is simply the tensor product of states
for the upper and lower Fermi surfaces and is given by
X
|τ a ≡ |τ+a , τ−a = sR+ (τ+a )sR̄− (τ−a )|R+ ⊗ |R̄−
R+ ,R̄−
 ! 
X X X R R
=  NS++S 0 NS−−S 0 sR+ (τ+a )sR̄− (τ−a ) |S+ S̄− (4.35)
S+ ,S− R+ ,R− S0
!
X X
= sS 0 (τ+a )sS 0 (τ−a ) sS+ (τ+a )sS− (τ−a )|S+ S̄− .
S+ ,S− S0

Here we have used the fact that sR̄− (τ−a ) = sR− (τ−a ) up to a sign. Using the definition
of coherent state one can show that the sum over S 0 is given by
!
a
X τ+n a
X τ−n
sS 0 (τ+a )sS 0 (τ−a ) = (τ+a |τ−b ) = exp 2
. (4.36)
S 0 n>0
n~

This eventually gives us

X
|τ a ≡ |τ+a , τ−a = (τ+a |τ−b ) sS+ (τ+a )sS− (τ−a )|S+ S̄− . (4.37)
S+ ,S−

Now, for a composite representation S ≡ S+ S̄− , we define a composite Schur function

sS (τ+a , τ−a ) as

sS (τ a ) ≡ sS+ S̄− (τ+a , τ−a ) = (τ+a |τ−b )sS+ (τ+a )sS− (τ−a ) (4.38)

and hence |τ a in (4.37) can be written as

X
|τ a = sS (τ a )|S . (4.39)
S

– 23 –
Using (4.34) one can show that
1 ~
e− ~ HT |S = e− 2 C2 (S)T |S (4.40)

Finally, evolution of |τ a state to |τ b state in time T is given by,

1
X ~
τ b |e− ~ H |τ a = sS (τ a )sS (τ b )e− 2 C2 (S)T . (4.41)
S

This is the partition function of 2d Yang-Mills theory on cylinder specified by two

holonomies corresponding to τ a and τ b . As a special case one can check that when
both τ+a = τ−a = τ ∗ then we have (up to an overall constant)

sS (τ a ) = dimS+ dimS̄− = dimS (4.42)

and we get back the disc amplitude. When all the |τ states are special i.e. τ+a =
τ−a = τ+b = τ−b = τ ∗ we get back the sphere partition function of 2d Yang-Mills theory.

5 Discussion
In this paper we show the equivalence between unitary matrix quantum mechanics
and 2d Yang-Mills theory and its variants via quantisation of phase space droplet.
We explicitly construct the partition function of 2d Yang-Mills theory on a generic
Riemann surface with gauge group U (N ) or q-deformed U (N ) from the evolution
of free Fermi droplets in one dimensional unitary matrix models. We note that the
q-deformation in the Yang-Mills side is related to the evolution of a particular types
of geometries of droplets without deforming the gauge group associated with the
matrix model. In that sense the droplet picture is more universal. The Hamiltonian
as appearing in (3.17) also appears in earlier literature [27, 42]. However, they ap-
pear in the context of 2d bosonic string theory where the interesting cubic term can
be interpreted as the splitting-joining interaction. In the current scenario we have
arrived at the same by quantising the fluctuations over a droplet in two-dimensional
phase space that captures the eigenvalue distribution of (0 + 1) dimensional unitary
matrix model described by the action (2.1). Thus, our procedure gives a bottom-up
approach of arriving at (3.17). In our work we have considered the potential W (θ)
to be constant. It would be interesting to derive the phase space Hamiltonian for
a generic potential and understand its meaning in the context of 2d Yang-Mills and
string theory. Starting with a sufficiently generic action, quantisation of these fluctu-
ations follows a Kac-Moody algebra. Our methodology, however has one restriction.
The classical configuration of the Fermi surface that we start with has a quadratic
profile. The small fluctuations introduced on this Fermi sea are “small and shal-
low enough” so as not to destroy the quadratic profile. This ensures that a θ =
constant line intersects the Fermi surface exactly twice validating (2.5). Of course

– 24 –
initial states with non-quadratic profiles are interesting in their own right but we
will postpone such discussion to future works. It must however be noted that folds
will generically form under time evolution even if we start with an unfolded Fermi
surface where the collective field theory describing those non-interacting fermions is
non-relativistic [43–45]. Usually the fold states have non-zero quantum dispersion.
Given a starting profile for the Fermi surface (say at t = 0) [43] explicitly calculated
such fold formation times (tf ). We consider the evolution of coherent states (3.30)
constructed in section 3.2 such that the evolution keeps the quadraticity of the Fermi
surface at all times.

Acknowledgments
We would like to thank Suresh Govindarajan for fruitful discussions. We also ac-
knowledged the illuminating discussion with Nabamita Banerjee, Suhas Gangad-
haraiah, Arnab Rudra, Ashoke Sen. The work of SD is supported by the grant no.
EMR/2016/006294 and MTR/2019/000390 from the SERB, Government of India.
SD also acknowledges the Simons Associateship of the Abdus Salam ICTP, Trieste,
Italy. We also thank all the medical and non-medical workers who are working tire-
lessly in these troubled times. Finally, we are grateful to people of India for their
unconditional support towards researches in basic sciences.

A Eigenstates of phase space Hamiltonian

We will demonstrate explicitly that the state |R+ as defined in (3.36) is an exact
eigenstate of the Hamiltonian (3.17). (3.17) consists of two families of decoupled
modes an and bn . For the sake of brevity we will only look at the action of the an
modes on |R+ . It is easy to see from (3.25) that, the action of the part of the
Hamiltonian denoted as H0 in (3.23) on the Kac-Moody primary |s is given by
 2
~ 2 ~2 3

~
H0 |s = − s + s + s |s (A.1)
24 4 6
Using (3.16) and noting a−n = a†n , one can easily show [an , (a†m )p ] = pn(a†n )p−1 δn,m .
This further gives us the identity

a−n an |~k, s = a†n an |~k, s = n kn |~k, s . (A.2)

Thus, the identity above along with (3.36) leads to

~ X
Hfree |R+ = (1 + 2~a0 ) a†n an |R+
2 n>0
(A.3)
~ X
= (1 + 2~s) nkn |R+
2 n>0

– 25 –
 P
where l(R) = n>0 nkn is the total number of boxes associated to the Young diagram
corresponding to the representation R. Thus, the ”mostly zero vector” ~k can be
interpreted as the cycle numbers of the Young diagram. The cubic part of the
Hamiltonian (3.23) denoted by Hint , has a significantly more complicated action on
|R+ . Action of Hint on R+ is given by
" #
~2 X
Hint |R+ = (am an a†m+n + a†m a†n am+n ) |R+
2 m,n>0
! (A.4)
N
~2 X
= l(R) + (li2 − 2ili ) |R+
2 i=1

where li happens to be number of boxes in each row of the corresponding Young

diagram of the representation R.

B Reviewing u(N ) representations

Any representation of the u(N ) = [su(N ) × u(1)] /ZN algebra can be represented by
a pair (R, Q) comprising of a standard Young diagram R and a number Q, coming
from the u(1) generator of the algebra. However the u(N ) representations can also
be written in terms of composite Young diagrams, specially in the N → ∞ limit,
which we discuss in the following sections.

B.1 Composite representations

As shown in [7], a naive grouping of Young diagrams in terms of their total number
of boxes under-counts the all possible u(∞) representations exactly by a factor of
half. One can easily circumvent this problem by taking recourse to the notion of
composite representations.

Figure 2. An example of a composite representation for N = 9. The red coloured border in

the rightmost figure represents the boundary of the conjugate Young diagram corresponding to
representation R− .

– 26 –
 Given two representations R+ and R− , a composite representation R+ R̄− is
given by a Young diagram with boxes corresponding to R+ placed on top right in a
standard way and the boxes corresponding to R− placed upside down in the bottom
left as anti-boxes. The total number of rows in a Young diagram corresponding to a
composite representation is always N such that if the number of rows r+ in R+ and
r− in R− do not add up to N , then N − (r+ + r− ) number of rows have zero length,
in between R+ and R− as shown in the centre diagram of Fig. 2. Note that this
procedure makes sense only in the large N limit where none of the representations
R+ and R− has more than N/2 number of rows. In the N → ∞ limit, summing over
all possible representations of u(N ) is indeed equivalent to summing over all possible
composite Young diagrams.
This way of representing a composite diagram was used to demonstrate the
factorization of u(N ) Yang-Mills theory into a chiral and anti-chiral sector in [41].
They further showed that this is equivalent to the composite representations of [7]
where the conjugate Young diagram of R− is drawn first and then the diagram of
R+ is attached to it, on the right as depicted in the rightmost diagram of Fig. 2.

B.2 Quadratic Casimirs of u(N ) and su(N )

For any semisimple Lie algebra G with generators denoted by TRa the quadratic
Casimir C2 (R) can be read off from the formula [46, 47]
X
[K T a , T b ]−1 TRa TRb ,


C2 (R)1dimR = (B.1)
a,b

with K T a , T b being the Killing form where “ ” symbolizes the fundamental rep-
resentation. Interestingly one can also check that quadratic Casimir is the same for a
representation and its conjugate. For the case of G ≡ su(N ), with a ∈ [1, N 2 −1], one
can simplify the quadratic Casimir by characterising the irreducible representations
R of su(N ) by a standard Young diagram with row lengths li where (i ∈ [1, N − 1])
with the condition ∞ ≥ l1 ≥ l2 ≥ · · · ≥ lN −1 ≥ 0. In that case, (B.1) simply reads
N −1 N −1
l(R)2 X X
C2 (R) = N l(R) + κR − ; l(R) = li ; κR = l(R) + (li2 − 2ili ). (B.2)
N i=1 i=1

Since u(N ) = [su(N ) × u(1)] /ZN , the eigenvalue Q of the u(1) generator, appropri-


ately called the charge, is constrained to be equal to l(R) mod N . Again resorting
to (B.1), the quadratic Casimir for u(N ) algebra can be written as

Q2
C2 (R, Q) = C2 (R) + = N l(R) + κR + 2sl(R) + N s2 , (B.3)
N
where the second equivalence stems from the relation Q = l(R)+N s with s ∈ Z which
is just a simple manifestation of the fact that Q = l(R) mod N . We have chosen this

– 27 –
particular basis of the u(N ) in this paper because of its apparently simple relation
with the su(N ) irreducible representation R. Interestingly one can go a step further
with the u(N ) representations (R, Q) and introduce coupled representations [7, 48]
or the extended Young diagram R which may or may not have negative number of
boxes (also termed as “anti-boxes” [41]) as discussed in the previous section.
The origin of these exotic Young diagrams relies on the clever rewriting of (B.3) as

C2 (R, Q) = N l(R) + κR
N N
X X 2
l(R) = l¯i ; κR = l(R) + (li − 2ili ) (B.4)
i=1 i=1
li = li + s; lN = s; s ∈ Z.

Therefore one can define an extended Young diagram R with boxes in the ith row
being li with the condition that ∞ ≥ l1 ≥ l2 ≥ · · · ≥ lN ≥ −∞. Hence the charge is
now simply the total number of boxes of R as Q = N
P
i=1 li . Therefore the quadratic
Casimir of u(N ) now reduces to the relation
N
X 2
C2 (R) = N Q + κR ; κR = Q + (li − 2ili ). (B.5)
i=1

C Twisted surgery and higher point functions

We consider time evolution of a classical shape |τ a to |τ in time T1 - such evolution
is given by a cylinder amplitude C(τ a , τ, T1 ). We then consider another amplitude -
transition from a classical shape |τ to |τ b from T1 time to T2 , C(τ, τ b , T2 − T1 ). If
we glue these two amplitudes along τ circle and integrate over all possible τ states
we get an amplitude for transition from |τ a shape to |τ b shape in time T2 . However
one can glue these two cylinders along τ circle after giving a local twist. The twists
is given in θ plane. A |τ state has a image in θ plane. We glue θi point of the
first cylinder with θi point of the second cylinder after giving a local twist φci and
then we integrate over all θi points i.e. |τ state. We call this φ ~ c twisted amplitude
~ c , τ b ). Thus we have
P(τ a , φ
Z
0
X
a ~ ~
b
P(τ , φ, τ ) = [dθ]W (θi ) sR (τ a )sR (θi + φci )sR (θi )sR (τ b )e− 2 (C2 (R)T1 +C2 (R )(T2 −T(C.1)
1 ))
.
RR0

We use the identity

sR (φci )
Z
[dθ]W (θi )sR (θi + φci )sR0 (θi ) = δRR0 (C.2)
dimR
Note that in the large N limit a local twist {φci } corresponds to a distribution σ(φc )
and hence there exists a corresponding coherent state |τ c . Therefore we write

– 28 –
 ~ c ) → sR (τ c ) and denote the above amplitude by P(τ a , τ c , τ b , T ). Thus we have
sR (φ
X sR (τ a )sR (τ b )sR (τ c ) ~
P(τ a , τ c , τ b , T ) = e− 2 C2 (R)T2 . (C.3)
R
dimR

As a consistency check, if we set the twist parameters φci = 0 i.e. |τ c = |τ ∗ , we

get back the cylinder amplitude as expected. Therefore from the point of view of
evolution of droplet, one may think that an initial shape |τ a evolves to a final shape
~ c at some intermediate time 0 < T1 < T2 .
|τ b in time T2 with an external twist φ

D Villain action
The abelian U (1) Villain action is just a Jacobi theta function
∞
1
− (θ+2πl)2
X
exp(−SV (θ)) = e g2 ; l ∈ Z. (D.1)
l=−∞

It was first introduced in [49] as an approximation to the Hamiltonian of a 2d planar

classical magnet and is well-known to be used in the study of planar Heisenberg
model [50]. Owing to the analogy between 2d planar model and four-dimensional
abelian gauge theory [51], it also appears in the study of lattice gauge theories and
therefore its appearance in the context of 2d Yang-Mills theory is not that much of
a surprise. Generalization of Villain’s action with an U (N ) gauge group is given by
∞
N X
Y N
exp(−SV ({θi }) = exp[− (θi + 2πli )2 ] (D.2)
i=1 li =−∞
λ

where θi ’s are the invariant angles of U ∈ U (N ). As the form of (D.2) suggests, one
can expand this action in the U (N ) character basis following [24] as

exp[−SV ] X
= cR TrR (U ) (D.3)
Z R

where TrR (U ) is the unitary group characters and Z is the two-dimensional lattice
gauge theory partition function with generalized Villain’s action. The coefficients cR
can then be found using the orthogonality of characters. After a little rearrangement
which can be written as
P 2
Y  1 − q 2(lj −lk ) 
−j+1)
cR = q j (l j
2(j−k)
(q = e−λ/4N ) (D.4)
j>k
1 − q

where the integers li are related to the number of boxes in a Young diagram corre-
sponding to the representation R of U (N ). In [25] it was shown that the character

– 29 –
expansion (D.3) with coefficients given by (D.4) reduces to the q-deformed disc am-
plitude of 2D Yang-Mills theory (4.30). Generalization of Villain’s action to U (N )
also leads to the heat kernel (1.2) as pointed out in [6].
Another interesting aspect of writing the character expansion of Villain action is
the gluing of two Yang-Mills theories with different q-deformations. If one considers
generalized Villain’s action with ’t Hooft couplings λ1 and λ2 then the integral
Z
exp[−SV (U, λ1 )] exp[−SV (U, λ2 )] X
dU = cR (q1 )cR (q2 ) (D.5)
Z(λ1 ) Z(λ2 ) R

is equivalent to gluing two disc amplitudes with deformations q1 and q2 respectively.

Using (D.4) and following [25], the above expression can be reduced to
1 1
C2 (R) C2 (R)
X
dimq1 R dimq2 R q12 q22 . (D.6)
R

References
[1] A. Polyakov, Quantum geometry of bosonic strings, Physics Letters B 103 (1981)
207 .
[2] A. Polyakov, Quantum geometry of fermionic strings, Physics Letters B 103 (1981)
211 .
[3] D. J. Gross and I. R. Klebanov, Vortices and the non-singlet sector of the c = 1
matrix model, Nuclear Physics B 354 (1991) 459 .
[4] I. R. Klebanov, String theory in two-dimensions, in Spring School on String Theory
and Quantum Gravity (to be followed by Workshop), pp. 30–101, 7, 1991,
hep-th/9108019.
[5] S. Mukhi, Topological matrix models, Liouville matrix model and c = 1 string theory,
in IPM String School and Workshop 2003 Caspian Sea, Iran, September 29-October
9, 2003, 2003, hep-th/0310287.
[6] P. Menotti and E. Onofri, The Action of SU(N ) Lattice Gauge Theory in Terms of
the Heat Kernel on the Group Manifold, Nucl. Phys. B 190 (1981) 288.
[7] D. J. Gross and W. Taylor, Two-dimensional QCD is a string theory, Nucl. Phys. B
400 (1993) 181 [hep-th/9301068].
[8] D. Boulatov, Infinite tension strings at d ¿ 1, Mod. Phys. Lett. A 8 (1993) 557
[hep-th/9211064].
[9] J. Baez and W. Taylor, Strings and two-dimensional QCD for finite N, Nucl. Phys.
B 426 (1994) 53 [hep-th/9401041].
[10] S. Cordes, G. W. Moore and S. Ramgoolam, Large N 2-D Yang-Mills theory and
topological string theory, Commun. Math. Phys. 185 (1997) 543 [hep-th/9402107].

– 30 –
[11] P. Horava, Topological strings and QCD in two-dimensions, in NATO Advanced
Research Workshop on New Developments in String Theory, Conformal Models and
Topological Field Theory, 11, 1993, hep-th/9311156.
[12] D. J. Gross, Two-dimensional QCD as a string theory, Nucl. Phys. B 400 (1993)
161 [hep-th/9212149].
[13] D. J. Gross and W. Taylor, Twists and Wilson loops in the string theory of
two-dimensional QCD, Nucl. Phys. B 403 (1993) 395 [hep-th/9303046].
[14] J. A. Minahan and A. P. Polychronakos, Equivalence of two-dimensional QCD and
the C = 1 matrix model, Phys. Lett. B 312 (1993) 155 [hep-th/9303153].
[15] S. Ramgoolam, Comment on two-dimensional O(N) and Sp(N) Yang-Mills theories
as string theories, Nucl. Phys. B 418 (1994) 30 [hep-th/9307085].
[16] M. Caselle, A. D’Adda, L. Magnea and S. Panzeri, Two-dimensional QCD is a
one-dimensional Kazakov-Migdal model, Nucl. Phys. B 416 (1994) 751
[hep-th/9304015].
[17] S. Cordes, G. W. Moore and S. Ramgoolam, Lectures on 2-d Yang-Mills theory,
equivariant cohomology and topological field theories, Nucl. Phys. B Proc. Suppl. 41
(1995) 184 [hep-th/9411210].
[18] G. Blau, Matthiasand Thompson, Quantum yang-mills theory on arbitrary surfaces,
International Journal of Modern Physics A 07 (1992) 3781
[https://doi.org/10.1142/S0217751X9200168X].
[19] E. Witten, On quantum gauge theories in two-dimensions, Commun. Math. Phys.
141 (1991) 153.
[20] S. R. Das and A. Jevicki, String Field Theory and physical interpretation of D = 1
strings, Modern Physics Letters A 05 (1990) 1639.
[21] A. Jevicki and B. Sakita, The quantum collective field method and its application to
the planar limit, Nuclear Physics B 165 (1980) 511 .
[22] V. Kazakov and A. A. Migdal, Induced QCD at large N, Nucl. Phys. B 397 (1993)
214 [hep-th/9206015].
[23] M. R. Douglas, Conformal field theory techniques in large N Yang-Mills theory, in
NATO Advanced Research Workshop on New Developments in String Theory,
Conformal Models and Topological Field Theory, 5, 1993, hep-th/9311130.
[24] E. Onofri, SU(N ) Lattice Gauge Theory With Villain’s Action, Nuovo Cim. A 66
(1981) 293.
[25] M. Romo and M. Tierz, Unitary chern-simons matrix model and the villain lattice
action, Physical Review D 86 (2012) .
[26] J. Polchinski, Classical limit of (1 + 1)-dimensional string theory, Nucl. Phys. B362
(1991) 125.

– 31 –
[27] A. Jevicki, Introduction to path integrals, matrix models and strings, in 9th Chris
Engelbrecht Summer School in Theoretical Physics: Field Theory, Topology and
Condensed Matter Physics, pp. 55–98, 11, 1996.
[28] A. Jevicki and B. Sakita, Collective field approach to the large-N limit: Euclidean
field theories, Nuclear Physics B 185 (1981) 89 .
[29] P. Basu, B. Ezhuthachan and S. R. Wadia, Plasma balls/kinks as solitons of large N
confining gauge theories, JHEP 01 (2007) 003 [hep-th/0610257].
[30] L. Grant, L. Maoz, J. Marsano, K. Papadodimas and V. S. Rychkov, Minisuperspace
quantization of ‘Bubbling AdS’ and free fermion droplets, JHEP 08 (2005) 025
[hep-th/0505079].
[31] L. Maoz and V. S. Rychkov, Geometry quantization from supergravity: The Case of
‘Bubbling AdS’, JHEP 08 (2005) 096 [hep-th/0508059].
[32] M. Marino, Chern-Simons theory, matrix models, and topological strings, Int. Ser.
Monogr. Phys. 131 (2005) 1.
[33] J. A. Minahan and A. P. Polychronakos, Classical solutions for two-dimensional
QCD on the sphere, Nucl. Phys. B 422 (1994) 172 [hep-th/9309119].
[34] D. J. Gross and A. Matytsin, Some properties of large N two-dimensional
Yang-Mills theory, Nucl. Phys. B437 (1995) 541 [hep-th/9410054].
[35] W. Donnelly and G. Wong, Entanglement branes in a two-dimensional string theory,
JHEP 09 (2017) 097 [1610.01719].
[36] X. Arsiwalla, R. Boels, M. Marino and A. Sinkovics, Phase transitions in q-deformed
2-D Yang-Mills theory and topological strings, Phys. Rev. D73 (2006) 026005
[hep-th/0509002].
[37] A. Polyanin and A. Manzhirov, Handbook of Integral Equations, Second Edition,
Updated, Revised and Extended. Chapman and Hall, CRC Press, 2008.
[38] J. Bryan and R. Pandharipande, The Local Gromov-Witten theory of curves, J. Am.
Math. Soc. 21 (2008) 101 [math/0411037].
[39] H. Ooguri, A. Strominger and C. Vafa, Black hole attractors and the topological
string, Physical Review D 70 (2004) .
[40] E. Strahov, A differential Model for the Deformation of the Plancherel Growth
Process, arXiv e-prints (2007) arXiv:0706.3292 [0706.3292].
[41] M. Aganagic, A. Neitzke and C. Vafa, BPS microstates and the open topological
string wave function, Adv. Theor. Math. Phys. 10 (2006) 603 [hep-th/0504054].
[42] M. Natsuume and J. Polchinski, Gravitational scattering in the c = 1 matrix model,
Nucl. Phys. B 424 (1994) 137 [hep-th/9402156].
[43] S. R. Das and S. D. Mathur, Folds, bosonization and nontriviality of the classical
limit of 2-D string theory, Phys. Lett. B 365 (1996) 79 [hep-th/9507141].

– 32 –
[44] S. Alexandrov, Matrix quantum mechanics and two-dimensional string theory in
nontrivial backgrounds, ph.d thesis, 9, 2003.
[45] S. R. Das, D-branes in 2-d string theory and classical limits, in 3rd International
Symposium on Quantum Theory and Symmetries, pp. 218–233, 1, 2004,
hep-th/0401067, DOI.
[46] P. D. Francesco, P. Mathieu and D. Sénéchal, Conformal Field Theory, Springer .
[47] W. Fulton and J. Harris, Representation Theory: A First Course, Graduate texts in
mathematics. Springer, 1991.
[48] S. G. Naculich and H. J. Schnitzer, Level-rank duality of the U(N) WZW model,
Chern-Simons theory, and 2-D qYM theory, JHEP 06 (2007) 023 [hep-th/0703089].
[49] Villain, J., Theory of one- and two-dimensional magnets with an easy magnetization
plane. ii. the planar, classical, two-dimensional magnet, J. Phys. France 36 (1975)
581.
[50] L. Susskind, Lattice models of quark confinement at high temperature, Phys. Rev. D
20 (1979) 2610.
[51] J. B. Kogut, An introduction to lattice gauge theory and spin systems, Rev. Mod.
Phys. 51 (1979) 659.

– 33 –