Unitary Matrix Models and Painlevé III

Masato Hisakado

Graduate School of Mathematical Sciences,

arXiv:hep-th/9609214v2 14 Nov 1996

University of Tokyo,
3-8-1 Komaba, Megro-ku, Tokyo, 113, Japan

February 1, 2008

Abstract
We discussed the full unitary matrix models from the view points of integrable equa-
tions and string equations. Coupling the Toda equations and the string equations, we
derive a special case of the Painlevé III equation. From the Virasoro constrains, we can
use the radial coordinate. The relation between t1 and t−1 is like the complex conjugate.

1
1 Introduction
Models of the symmetric unitary matrix model are solved exactly in the double scaling
limit, using orthogonal polynomials on a circle.[1] The partition function is the form
dU exp{− Nλ trV (U)}, where U is an N × N unitary matrix and trV (U) is some well
R

defined function of U. When V (U) is the self adjoint we call the model symmetric.[2]
The simplest case is given by V (U) = U + U † . This unitary models has been studied
in connection with the large-N approximation to QCD in two dimensions.(one-plaquette
model)[3] For this model “string equation” is the Painlevé II equation. “Discrete string
equation” is called the discrete Painlevé II equation.[4] When V (U) is the anti-self adjoint,
we call the model anti-symmetric model. The simplest case is given by V (U) = U − U † .
This is the theta term in the two-dimensional QCD.[5] It has the topological meaning.
The full non-reduced unitary model was first discussed in [6]. The full unitary model can
be embedded in the two-dimensional Toda Lattice hierarchy.
In this letter we shall try to reformulate the full unitary matrix model from the view
points of integrable equations and string equations. These two view points are closely
connected each other to describe this model. We unify these view points and clarify a
relation between these view points.
This letter is organized as follows. In the section 2 we present discrete string equa-
tions for the full unitary matrix model. Here we consider only the simplest case. From
the Virasoro constraints, a relation between times t1 and t−1 is like complex conjugate.
Because of this symmetry, we can use the radial coordinate. In the section 3 coupling
the Toda equation and the discrete string equations, we obtain the special Painlevé III
equation. In the section 4 we consider the reduced models, the symmetric and the anti-
symmetric model. From the symmetric and the anti-symmetric model we can obtain the
modified Volterra equation and the discrete nonlinear Schrödinger equation respectively.
We study the relation of the symmetric and the anti-symmetric model. In a special case,
we can transform the symmetric model into the anti-symmetric model. Using this map,
we can obtain Bäcklund transformation of the modified Volterra equation and the discrete
nonlinear Schrödinger equation. The last section is devoted to concluding remarks.

2 Unitary Matrix model

It is well known that the partition function τn of the unitary matrix model can be presented
as a product of norms of the biorthogonal polynomial system. Namely, let us introduce a
scalar product of the form

dµ(z)
I
< A, B >= A(z)B(z −1 ). (2.1)
2πiz

2
where
(tm z m + t−m z −m )}
X
dµ(z) = dz exp{− (2.2)
m>0

Let us define the system of the polynomials biorthogaonal with respect to this scalar
product
< Φn , Φ∗k >= hk δnk . (2.3)

Then, the partition function τn of the unitary matrix model is equal to the product of
hn ’s:
n−1
Y
τn = hk , τ0 = 1. (2.4)
k=0

The polynomials are normalized as follows (we should stress that superscript ‘*’ does not
mean the complex conjugation):

∗
Φn = z n + · · · + Sn−1 , Φ∗n = z n + · · · + Sn−1 ∗
, S−1 = S−1 ≡ 1. (2.5)

Now it is easy to show that these polynomials satisfy the following recurrent relations,

Φn+1 (z) = zΦn (z) + Sn z n Φ∗n (z −1 ),

Φ∗n+1 (z −1 ) = z −1 Φ∗n (z −1 ) + Sn∗ z −n Φn (z), (2.6)

and
hn+1
= 1 − Sn Sn∗ . (2.7)
hn
Note that hn , Sn , Sn∗ ,Φn (z) and Φ∗n depend parametrically on t1 , t2 , · · · , and t−1 , t−2 , · · · ,
but for convenience of notation we suppress this dependence. Hereafter we call t1 , t2 , · · · ,
and t−1 , t−2 , · · · , time variables.
Using (2.3) and integration by parts, we can obtain next relations:

dµ(z) ′
I
− V (z)Φn+1 (z)Φ∗n (z −1 )
2πiz
dµ(z) ∂Φn+1 (z) ∗ −1 dµ(z) ∂Φ∗n (z −1 ) dµ(z) Φn+1 (z)Φ∗n
I I I
= − Φn (z ) − Φn+1 (z) +
2πiz ∂z 2πiz ∂z 2πiz z
= (n + 1)(hn+1 − hn ), (2.8)

and
dµ(z) 2 ′
I
z V (z)Φ∗n+1 (z −1 )Φn (z)
2πiz
dµ(z) 2 ∂Φ∗n+1 (z −1 ) dµ(z) 2 ∗ −1 ∂Φn (z) dµ(z) ∗
I I I
= z Φn (z) + z Φn+1 (z ) + zΦn+1 (z −1 )Φn (z)
2πiz ∂z 2πiz ∂z 2πiz
= (n + 1)(hn+1 − hn ). (2.9)

(2.8) and (2.9) are string equations for the full unitary matrix model.

3
 If t1 and t−1 are free variables while t2 = t3 = · · · = 0 and t−2 = t−3 = · · · = 0, (2.8)
and (2.9) become

(n + 1)Sn Sn∗ = t−1 (Sn Sn+1

∗
+ Sn∗ Sn−1 )(1 − Sn Sn∗ ), (2.10)

(n + 1)Sn Sn∗ = t1 (Sn∗ Sn+1 + Sn Sn−1

∗
)(1 − Sn Sn∗ ). (2.11)

Next we introduce a useful relation. Using (2.3) and integration by parts, we can show

dµ(z) ′
I
zV (z)Φn (z)Φ( n)∗ (z −1 )
2πiz
dµ(z) ∂Φn (z) ∗ −1 dµ(z) ∂Φ∗ (z −1 )
I I
= z Φn (z ) + zΦn (z) n
2πiz ∂z 2πiz ∂z
= nhn − nhn = 0. (2.12)

This corresponds to the Virasoro constraint:[2]

∞
∂
L−cl
X
0 = ktk . (2.13)
k=−∞ ∂tn

This relation constrains a symmetry like complex conjugate between tk and t−k . It is
important in the next section. If we set that t1 and t−1 are free variables while t2 = t3 =
· · · = 0 and t−2 = t−3 = · · · = 0, from (2.12) we get

∗
t1 Sn Sn−1 = t−1 Sn∗ Sn−1 . (2.14)

Using (2.14), (2.10) and (2.11) can be written

(n + 1)Sn = (t1 Sn+1 + t−1 Sn−1 )(1 − Sn Sn∗ ), (2.15)

(n + 1)Sn∗ = (t−1 Sn+1

∗ ∗
+ t1 Sn−1 )(1 − Sn Sn∗ ). (2.16)

3 Toda equation and String equations

Using the orthogonal conditions, it is also possible to obtain the equations which describe
the time dependence of Φn (z) and Φ∗n (z). Namely, differentiating (2.3) with respect to
times t1 and t−1 gives the following evolution equations:

∂Φn (z) Sn hn
=− (Φn (z) − zΦn−1 ), (3.1)
∂t1 Sn−1 hn−1

∂Φn (z) hn
= Φn−1 (z), (3.2)
∂t−1 hn−1
∂Φ∗n (z −1 ) hn ∗
= Φn−1 (z −1 ), (3.3)
∂t1 hn−1

4
 ∂Φ∗n (z −1 ) S ∗ hn
= − ∗n (Φ∗ (z −1 ) − z −1 Φ∗n−1 ), (3.4)
∂t−1 Sn−1 hn−1 n
The compatibility condition gives the following nonlinear evolution equations:
∂Sn hn+1 ∂Sn hn+1
= −Sn+1 , = Sn−1 , (3.5)
∂t1 hn ∂t−1 hn
∂Sn∗ ∗ hn+1 ∂Sn∗ ∗ hn+1
= Sn+1 , = −Sn−1 , (3.6)
∂t1 hn ∂t−1 hn
∂hn ∗ ∂hn
= Sn Sn−1 hn , = Sn∗ Sn−1 hn , (3.7)
∂t1 ∂t−1
Here we define an , bn and b∗n :

hn+1
an ≡ 1 − Sn Sn∗ = , (3.8)
hn
∗
bn ≡ Sn Sn−1 , (3.9)

b∗n ≡ Sn∗ Sn−1 . (3.10)

Notice that from the definitions an , bn and b∗n satisfy the following identity:

bn b∗n = (1 − an )(1 − an−1 ). (3.11)

It can be shown using (2.14) that

t1 bn = t−1 b∗n . (3.12)

In terms of an , bn and b∗n , (3.5) and (3.6) become the two-dimensional Toda equations:

∂an ∂bn
= an (bn+1 − bn ), = an − an−1 , (3.13)
∂t1 ∂t−1
and
∂an ∂b∗n
= an (b∗n+1 − b∗n ), = an − an−1 . (3.14)
∂t−1 ∂t1
Using an , bn and b∗n , we rewrite (2.10) and (2.11)
n + 1 1 − an
= b∗n+1 + b∗n . (3.15)
t−1 an
and
n + 1 1 − an
= bn+1 + bn , (3.16)
t1 an
From (3.13) and (3.16) we eliminate bn+1 ,

1 n+1 ∂an
2bn = [ (1 − an ) − ]. (3.17)
an t1 ∂t1

5
In the same way, from (3.14) and (3.15) we eliminate b∗n+1 ,
1 n+1 ∂an
2b∗n = [ (1 − an ) − ]. (3.18)
an t−1 ∂t−1
Using (3.11) and (3.12), (3.13) and (3.14) can be written
∂bn t1 b2n
= (an − 1) + , (3.19)
∂t−1 t−1 1 − an
∂b∗n t−1 (b∗n )2
= (an − 1) + . (3.20)
∂t1 t1 1 − an
Using (3.17) and (3.19) to eliminate bn we obtain a second order ODE for an
∂ 2 an n + 1 ∂an n + 1 ∂an (n + 1)2 an − 1
= − − 2an (an − 1) +
∂t1 ∂t−1 t−1 an ∂t1 t1 an ∂t−1 2t1 t−1 an
1 ∂an ∂an 1 t1 1 ∂an 2
+ + ( ). (3.21)
an ∂t1 ∂t−1 2 t−1 (an − 1)an ∂t1
In the same way, we eliminate b∗n using (3.18) and (3.19) and obtain an ODE for an
∂ 2 an n + 1 ∂an n + 1 ∂an (n + 1)2 an − 1
= − − 2an (an − 1) +
∂t1 ∂t−1 t1 an ∂t−1 t−1 an ∂t1 2t1 t−1 an
1 ∂an ∂an 1 t−1 1 ∂an 2
+ + ( ). (3.22)
an ∂t1 ∂t−1 2 t1 (an − 1)an ∂t−1
The equality of (3.21) and (3.22) implies that
∂an ∂an
t1 = t−1 (3.23)
∂t1 ∂t−1
Also this constraint can be shown from (2.14), (3.13) and (3.14) directly. So an are
functions of the radial coordinate
x = t1 t−1 , (3.24)
only. Then from (3.21) and (3.22) we can obtain
∂ 2 an 1 1 1 ∂an 2 1 ∂an 2 (n + 1)2 an − 1
= ( + )( ) − − a (a
n n − 1) + . (3.25)
∂x2 2 an − 1 an ∂x x ∂x x 2x2 an
This is an expression of the Painlevé V equation (PV) with
(n + 1)2
αV = 0, βV = − , γV = 2, δV = 0. (3.26)
2
(3.25) is related to the usual one through
an
an −→ cn = . (3.27)
an − 1
(3.25) is the Painlevé III equation (P III) with (see [7])

αIII = 4(n + 1) βIII = −4n, γIII = 4, δIII = −4. (3.28)

6
4 Symmetric and Anti-symmetric model
In this section we consider reduced unitary matrix models. The following reductions of
the time variables tk leads to the symmetric and the anti-symmetric model:

tk = t−k = t+
k k = 1, 2, · · · , (symmetric model) (4.1)

and
tk = −t−k = t−
k k = 1, 2, · · · , (anti − symmetric model) (4.2)
If t+ + + ∗
1 are free variables while t2 = t3 = · · · = 0, from (2.14) Sn = Sn . From (2.12) and
(2.14) the string equation becomes

(n + 1)Sn = t+ 2
1 (Sn+1 + Sn−1 )(1 − Sn ). (4.3)

This is called the discrete Painlevé II (dP II) equation. Appropriate continuous limit
of (4.3) yields the Painlevé II (P II) equation. From (3.5) and (3.6) we can obtain the
modified Volterra equation:[6]
∂Sn
= −(1 − Sn2 )(Sn+1 − Sn−1 ). (4.4)
∂t+
1

Appropriate continuous limit of (4.4) yields the modified KdV equation.

(4.3) and (4.4) can be written in the form
1 n+1 ∂Sn
2Sn+1 = 2
( + Sn − + ), (4.5)
1 − Sn t1 ∂t1
and
1 n+1 ∂Sn
2Sn−1 = 2
( + Sn + + ), (4.6)
1 − Sn t1 ∂t1
Writing (4.6) as
1 n+2 ∂Sn+1
2Sn = ( + Sn+1 + ), (4.7)
2
1 − Sn+1 t1 ∂t+
1

and using (4.5) to eliminate Sn+1 we obtain a second order ODE for Sn .
∂ 2 Sn Sn ∂Sn 2 1 ∂Sn (n + 1)2 Sn
= − (
2 ∂t+
) − + + + − 4Sn (1 − Sn2 ). (4.8)
∂t+21 1 − S n 1 t1 ∂t1 t+2
1 1 − S 2
n

It is important to keep in mind that the relevant function is 1 − Sn2 = an . an satisfies

∂ 2 an 1 1 1 ∂an 2 1 ∂an 2(n + 1)2 an − 1
= ( + )( ) − − 8a (a
n n − 1) + . (4.9)
∂t+21 2 an − 1 an ∂t+
1 t+ +
1 ∂t1 t+2
1 an
If we set x = (t+ 2
1 ) , (4.9) is the same as (3.25). Then, we obtain (4.9) which is the special
case of Painlevé III. In conclusion, coupling the modified Volterra and the dP II, we can
obtain the P III. The double limit
t+2
1
n → ∞, t+
1 → ∞, = O(1), (4.10)
n

7
maps P III(4.9) to P II. Clearly, this kind of limit can be discussed independently of the
connection with the modified Volterra and the modified KdV equation.
Next we consider the anti-symmetric model. If t− − −
1 are free variables while t2 = t3 =
· · · = 0, from (2.14)
∗
Sn Sn−1 = Sn∗ Sn−1 . (4.11)

From (2.12) and (2.14) the string equations become

(n + 1)Sn = t− ∗
1 (−Sn+1 + Sn−1 )(1 − Sn Sn ),

(n + 1)Sn∗ = t− ∗ ∗ ∗
1 (Sn+1 − Sn−1 )(1 − Sn Sn ). (4.12)

On the other hand, from (3.5) and (3.6) we obtain the discrete nonlinear Schrödinger
(NLS) equation:[8]

∂Sn
= −(1 − Sn Sn∗ )(Sn+1 + Sn−1 ),
∂t−
1
∂Sn∗
− = (1 − Sn Sn∗ )(Sn+1
∗ ∗
+ Sn−1 ). (4.13)
∂t1
Using the same method in the symmetric model case we can obtain the P III. Coupling
the discrete NLS and the string equation, we can obtain P III.
Through the transformation

+
z → iz it−
1 → t1 , (4.14)

the anti-symmetric model is transformed into the symmetric model. Then we get the
Bäcklund transformation from the discrete NLS to the modified Volterra equation:

Sn −→ (i)n+1 Sn . (4.15)

However we restrict t−
1 to a real number, we can not transform the anti-symmetric model
into the symmetric model.
We change variables an → un = ln an . Then (3.13) and (3.14) become

∂ 2 un
= eun+1 − 2eun + eun−1 . (4.16)
∂t1 ∂t−1

In the anti-symmetric model from (2.5) and (4.11) we can get

Sn = Sn∗ , an = 1 − Sn2 , (n = odd),

Sn = −Sn∗ , an = 1 + Sn2 , (n = even). (4.17)

In the case that t−

1 is real, we can see the oscillation of an . This phenomenon can be seen
only in the anti-symmetric model.

8
 Here we consider the continuum limit near the anti-symmetric model. We are in-
terested in Sn2 = ǫgn , ǫ → 0 and n → ∞. We assume t1 = −t−1 + 2ǫ/n and define
gn+1 − gn = ǫgn′ .
Then the continuum limit yields

u ≡ un = −un+1 + ǫ(±gn′ + gn2 ) + O(ǫ2 ), (4.18)

where ± corresponds to n =odd and n =even respectively. So in the continuous limit

(4.16) becomes well known the 1D sinh Gordon equation

∂2u
= −2 sinh 2u, (4.19)
∂t1 ∂t−1
where t1 = −t−1 . Here we introduce the radial coordinate
q
r= −t1 t−1 . (4.20)

u obeys an ODE of the form

d2 u 1 du
+ = 2 sinh 2u. (4.21)
dr 2 r dr
This is the P III with

αIII = 0 βIII = 0, γIII = 1, δIII = −1. (4.22)

This equation is obtained from the 2 states Toda field equation, too.[9] Because of the
oscillation of an , in the continuous limit un looks like having 2 states.
At last we consider the relation between the symmetric and anti-symmetric model
from the determinant form. The partition function of the symmetric model is

τN+ = detij Ii−j (t+

1 ), (4.23)

where Im is the modified Bessel function of order m.[10] In the same way, we can calculate
the partition function of the anti-symmetric model:

τN− = detij Ji−j (t−

1 ), (4.24)

where Jm is the Bessel function of order m. (4.14) is also the transformation between the
Bessel and the modified Bessel function. (4.17) comes from the oscillation of the Bessel
function.

9
5 Concluding remarks
We try to reformulate the full unitary matrix model from the view points of integrable
equations and string equations. Coupling the Toda equation and the string equations,
we obtain the P III equation. Because of the Virasoro constraint, t1 and t−1 have the
symmetry. This symmetry is like complex conjugate. Then we can use the radial coordi-
nate. This PIII also describe the phase transition between the week and strong coupling
region. Next we consider the relation among the symmetric, anti-symmetric model and
the P III equation. If t−
1 is a purely imaginary number, the anti-symmetric model can be
transformed into the anti-symmetric model. Using this map we construct the Bäcklund
transformation from the discrete nonlinear Schrödinger equation to the modified Volterra
equation . This map is also the transformation between the Bessel and the modified Bessel
function. If we restrict t−
1 to a real number, the symmetric and the anti-symmetric are
different.

References
[1] V.Periwal and D.Schevitz, Phys.Rev,Lett.64(1990)1326.

[2] M.Bowick, A.Morozov and D.Schevitz, Nucl.Phys.B354(1991)496.

[3] D.Gross and E.Witten, Phys.Rev.D.21(1980)446.

[4] B.Gramaticos, A.Ramani and V.Papageorigou, Phys.Rev.Lett. 67(1991)1825.

[5] T.G.Koács, E.T.Tomoulis,Z.Schram, Nucl.Phys.B454(1995)45.

[6] S.Kharchev, A.Marshakov, A.Mironov, A.Orlov and A.Zabrodin,

Nucl.Phys.B366(1991)569.

[7] K.Okamoto, Proc.Jpn.Acad. 56(1980)367.

[8] S.Kharchev, A.Mironov and Zhedanov, Faces of Relativistic Toda Chain, hep-
th9606144.

[9] K.Takasaki and T.Nakatsu, Isomonodromic Deformations and Supersymmetric

Gauge Theories, hep-th9603069.

[10] J.-M.Drouffe and J.-B.Zuber, Phys.Rep.102(1983)104.

10