ITEP-TH-38/24

Multi-component Toda lattice hierarchy

T. Takebe∗ A. Zabrodin†

December 2024
arXiv:2412.20122v1 [nlin.SI] 28 Dec 2024

Dedicated to the memory of Masatoshi Noumi

Abstract

We give a detailed account of the N -component Toda lattice hierarchy. This

hierarchy is an extended version of the one introduced by Ueno and Takasaki.
Our version contains N discrete variables rather than one. We start from the Lax
formalism, deduce the bilinear relation for the wave functions from it and then,
based on the latter, prove existence of the tau-function. We also show how the
multi-component Toda lattice hierarchy is embedded into the universal hierarchy
which is basically the multi-component KP hierarchy. At last, we show how the
bilinear integral equation for the tau-function can be obtained using the free fermion
technique. An example of exact solutions (a multi-component analogue of one-
soliton solutions) is given.

Contents
1 Introduction 2

2 Lax formalism for the multi-component Toda lattice hierarchy 5

2.1 Lax and Zakharov-Shabat equations . . . . . . . . . . . . . . . . . . . . . 6
2.2 Gauge transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
∗
Takashi TAKEBE: Beijing Institute of Mathematical Sciences and Applications, No. 544, Hefangkou
Village, Huaibei Town, Huairou District, Beijing, 101408, People’s Republic of China; e-mail:
[email protected]
†
Anton ZABRODIN: Skolkovo Institute of Science and Technology, 143026, Moscow, Russia and
National Research University Higher School of Economics, 20 Myasnitskaya Ulitsa, Moscow 101000,
Russia, and NRC “Kurchatov institute”, Moscow, Russia; e-mail: [email protected]

1
3 Linearization: wave operators and wave functions 18
3.1 Wave operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Wave functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

4 The bilinear identity 30

5 The tau-function 34
5.1 Existence of the tau-function . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.2 Tau-function as a universal dependent variable . . . . . . . . . . . . . . . 44
5.3 Bilinear equations of the Hirota-Miwa type . . . . . . . . . . . . . . . . . 46

6 The multi-component Toda lattice from the universal hierarchy 47

6.1 The universal hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2 Specification to the multi-component Toda lattice hierarchy . . . . . . . 48

7 The multi-component Toda lattice hierarchy from free fermions 52

7.1 The multi-component fermions . . . . . . . . . . . . . . . . . . . . . . . . 52
7.2 The bilinear identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
7.3 An example of exact solution . . . . . . . . . . . . . . . . . . . . . . . . 58

8 Concluding remarks 62

Appendix: Non-abelian Toda lattice 64

Acknowledgments 66

References 66

1 Introduction
The Toda lattice [1, 2] is one of the most important integrable systems known at present.
The original version (sometimes called Toda chain) is

d2
ϕs (t) = eϕs (t)−ϕs−1 (t) − eϕs+1 (t)−ϕs (t) , (1.1)
dt2
where the field ϕs (t) is a function of one discrete space variable s ∈ Z and one time
variable t. There exists also a two-dimensional extension of the model,

∂2
ϕs (x, y) = eϕs (x,y)−ϕs−1 (x,y) − eϕs+1 (x,y)−ϕs (x,y) , (1.2)
∂x ∂y

2
with two space-time variables x and y and one discrete variable s ∈ Z. Integrability of
the latter model was proved by Mikhailov in [3]. In what follows by the Toda lattice we
understand its two-dimensional version.
Later it became clear that the Toda lattice equation can be embedded into an infinite
hierarchy of compatible nonlinear partial differential-difference equations. The theory of
the Toda lattice hierarchy was developed by Ueno and Takasaki in [4]. Starting from
the Lax and Zakharov-Shabat representations, they proved existence of the tau-function
which is a universal dependent variable of the hierarchy. The tau-function is a function
of infinitely many independent variables which are “times” parametrizing evolution along
different integrable flows. In terms of the tau-function, all equations of the hierarchy are
encoded in a generating bilinear integral equation.
It is known that integrable equations and their hierarchies admit multi-component
generalizations in which the dependent variables are matrices rather than numbers. For
example, the multi-component version of the Kadomtsev-Petviashvili (KP) hierarchy was
discussed in [5, 6, 7, 8]. An N-component generalization of the Toda lattice hierarchy
was suggested by Ueno and Takasaki in [4]. Equation (1.2) is generalized to the following
equation for an N × N matrix gs = gs (x, y):

∂y (∂x gs gs−1 ) = gs gs−1

−1
− gs+1gs−1 . (1.3)

This equation is known as the non-abelian two-dimensional Toda lattice. However, the
treatment in [4] was not complete (in particular, the tau-function of the multi-component
hierarchy was not discussed). The aim of our paper is to give a more comprehensive
account of the theory and suggest an extension of the system having N discrete variables
rather than one (as it was in [4]). In what follows we comment on differences between
[4] and our treatment as they occur in footnotes.
The independent variables of the N-component Toda hierarchy are 2N infinite sets
of “times”

t = {t1 , t2 , . . . , tN }, tα = {tα,1 , tα,2 , tα,3 , . . . }, α = 1, . . . , N

and
t̄ = {t̄1 , t̄2 , . . . , t̄N }, t̄α = {t̄α,1 , t̄α,2 , t̄α,3 , . . . }, α = 1, . . . , N
which are in general complex numbers. Besides, there are N discrete variables s =
{s1 , . . . , sN } which are integers. Equation (1.2) corresponds to the simplest case N = 1,
x = t1,1 , y = t̄1,1 .
Let us describe the contents of the paper in some detail.
We develop three different approaches which are shown to lead to the same result.
The starting point of one of them is the Lax representation in terms of difference Lax
operators L, L̄ and auxiliary difference operators U α , Ū α , P α , P̄ α with N × N matrix
coefficients. The logical structure of our consideration can be illustrated by the following
chain:

Lax representation −→ Zakharov-Shabat representation −→

−→ wave functions −→ tau-function.

3
Namely, starting from the Lax representation, we prove its equivalence to the Zakharov-
Shabat representation, then prove existence of so-called wave operators (which are some-
times called dressing operators) and use them to introduce N × N matrix wave functions
∗
Ψ(s, t, t̄; z), Ψ̄(s, t, t̄; z) (and the adjoint wave functions Ψ∗ (s, t, t̄; z), Ψ̄ (s, t, t̄; z)) de-
pending on all the times and on a complex spectral parameter z. The wave functions
satisfy a system of linear equations whose compatibility implies nonlinear equations of
the Toda lattice hierarchy and thus provide a linearization of it. We give a detailed
proof that the wave functions of the multi-component Toda lattice hierarchy satisfy the
integral bilinear identity
I I
∗ ′ ′ ′ ∗ ′
Ψ(s, t, t̄; z) Ψ (s , t , t̄ ; z) dz = Ψ̄(s, t, t̄; z) Ψ̄ (s′ , t′ , t̄ ; z) dz, (1.4)
C∞ C0

′
valid for all s, t, t̄, s′ , t′ , t̄ , where C∞ and C0 are contours encircling ∞ and 0 respectively
(Proposition 4.1). Next we prove that the identity (1.4) implies existence of a tau-function
which is an N × N matrix with matrix elements ταβ (s, t, t̄), α, β = 1, . . . , N (Theorem
5.1). The tau-function is shown to satisfy the following integral bilinear equation:
N I
′ ′ ′
X
δβγ
(−1) z sγ −sγ +δαγ +δβγ −2 eξ(tγ −tγ ,z) ταγ (s, t − [z −1 ]γ , t̄)τγβ (s′ , t′ + [z −1 ]γ , t̄ ) dz
γ=1 C∞

N I
X ′ ′
δαγ
= (−1) z sγ −sγ −2 eξ(t̄γ −t̄γ ,z)
γ=1 C∞

′
× ταγ (s + [1]γ , t, t̄ − [z −1 ]γ )τγβ (s′ − [1]γ , t′ , t̄ + [z −1 ]γ ) dz,
(1.5)
′
X
′ ′ k −1
which is valid for all s, t, t̄, s , t , t̄ (Theorem 5.2). Here ξ(tγ , z) = tγ,k z , t ± [z ]γ is
k≥1
the set t in which tγ,k is shifted by ± k1 z −k and s ± [1]γ is the set s in which sγ is shifted
by ±1. Equation (1.5) is the generating equation which encodes all differential-difference
equations of the hierarchy. They are obtained from it by expanding both sides in the
′
Taylor series in t − t′ , t̄ − t̄ . This approach is developed in Sections 2–5. In Section 5.3
we also obtain various bilinear equations of the Hirota-Miwa type as corollaries of (1.5).
The second approach developed in Section 6 is based on the universal hierarchy named
so in the recent paper [9]. It is basically the multi-component KP hierarchy in which
the discrete variables are allowed to take arbitrary complex values (but equations with
respect to them are difference). The tau-function of this hierarchy is a matrix-valued
function obeying a bilinear integral equation. We show how the N-component Toda
lattice hierarchy can be embedded into the 2N-component universal hierarchy and define
the tau-function of the former in terms of that of the latter. As a result, we obtain a
bilinear integral equation for the tau-function of the Toda lattice hierarchy introduced in
this way and show that it becomes equivalent to (1.5) after a simple redefinition of the
tau-function by multiplying its matrix elements by certain sign factors.
The third approach which is followed in Section 7 is based on the free fermion tech-
nique developed by the Kyoto school [10, 11, 12]. In this approach, the tau-function is
represented as a vacuum expectation value of certain fermionic operators. In order to

4
describe multi-component hierarchies, one should work with multi-component fermionic
operator fields ψ (α) (z), ψ ∗(α) (z), α = 1, . . . , N obeying the standard anticommutation
relations. The basic fact from which the bilinear equation for the tau-function follows is
the operator bilinear identity
N   X N  
X dz (γ) ∗(γ) dz (γ) ∗(γ)
res ψ (z)g ⊗ ψ (z)g = res gψ (z) ⊗ gψ (z) (1.6)
γ=1
z γ=1
z

characterizing the Clifford group elements g of the general form

N I I
X 
g = exp A(αβ) (z, w)ψ ∗(α) (z)ψ (β) (w)dzdw , (1.7)
α,β=1

where A(αβ) (z, w) is some function

 of two complex variables. (The operation res in
X
k
(1.6) is defined as res ak z dz = a−1 .) The bilinear equation for the tau-function is
k
obtained as a corollary of (1.6) after acting by both sides to a tensor product of certain
states from the fermionic Fock space and applying the (non-abelian) bosonization rules
[6]. This job is done in Section 7.2. As a result, we obtain the integral bilinear equation
for the tau-function defined as the expectation value
N
¯ t̄)
X
|s|(|s|−1)/2 J(t) −J(
ταβ (s, t, t̄) = (−1) s + [1]α − [1]β e ge s , |s| = sγ . (1.8)
γ=1

¯ t̄) are defined in (7.14). The bilinear equation obtained in this way
The operators J(t), J(
is the same as the one from Section 6. (For the one-component Toda lattice hierarchy
this approach was studied in [13], [14] and [15].) In Section 7.3 we give an example of
exact solution which is a multi-component analogue of the one-soliton solution to the
Toda lattice.
In Appendix we show that the non-abelian two-dimensional Toda lattice (1.3) is
indeed included in our multi-component Toda lattice hierarchy.

2 Lax formalism for the multi-component Toda lat-

tice hierarchy
In this section we define the N-component Toda lattice hierarchy in the Lax formalism,
essentially following §3.1 of [4] but there are differences from their formulation in the
following points:

• We replace Z × Z-matrices in [4] by difference operators.

• We introduce the set of N discrete variables {s1 , . . . , sN } ∈ ZN which we denote

by s. In [4] only one discrete variable s was introduced and the shift operator
corresponded to a matrix Λ acting on the space CZ ⊗ CN .

5
 The continuous independent variables of the multi-component Toda hierarchy are 2N
infinite sets of “times”

t = {t1 , t2 , . . . , tN }, tα = {tα,1 , tα,2 , tα,3 , . . . }, α = 1, . . . , N

and
t̄ = {t̄1 , t̄2 , . . . , t̄N }, t̄α = {t̄α,1 , t̄α,2 , t̄α,3 , . . . }, α = 1, . . . , N
which are in general complex numbers. The sets t and t̄ are often called “positive” and
“negative” times. Besides, there are discrete “zeroth times” s1 , . . . , sN which are integer
numbers. The set of “zeroth times” is denoted as s = {s1 , . . . , sN }.
We use the following notations:

• Diagonal matrix units Eα := (δµα δνα )µ,ν=1,...,N ∈ Mat(N × N, C).

• The unit matrix, 1N := (δαβ )α,β=1,...,N ,
• The shift operators e∂sα f (s) := f (s + [1]α ), where the notation s + [1]α means that
the α-th component is shifted by 1 and all other components remain unchanged.
We define the total shift operator e∂s by e∂s = e∂s1 +···+∂sN . The operator e∂s acts
as e∂s f (s) = f (s + 1), where s + 1 = {s1 + 1, . . . , sN + 1}.
• A difference operator A acting on an N × N-matrix function f (s) as
X
Af (s) = aj (s) f (s + j1), aj (s) ∈ Mat(N × N, C),
j∈Z

is expressed in the form X

A= aj (s) ej∂s .
j∈Z

For such an operator we define the non-negative shift part and the negative shift
part as X X
A≥0 := aj (s) ej∂s , A<0 := aj (s) ej∂s . (2.1)
j≥0 j<0

2.1 Lax and Zakharov-Shabat equations

In the framework of the Lax formalism, the multi-component Toda lattice hierarchy
is a system of partial differential equations for the following matrix-valued difference
operators1 :
∞
X
L(s) = bj (s) e(1−j)∂s , b0 (s) = 1N ,
j=0
(2.2)
∞
X
L̄(s) = b̄j (s) e(j−1)∂s , b̄0 (s) ∈ Mat(N × N, C) is invertible,
j=0

1
In [4] it was assumed that b̄0 (s) and ūα,0 (s) were expressed through a separately introduced matrix
w̃0 (s). We will show in Proposition 2.2 that this follows from the algebraic relations imposed below on
the difference operators. The operators Qα (s) and Q̄α (s) did not appear in [4].

6
 ∞
X
U α (s) = uα,j (s) e−j∂s , uα,0 (s) = Eα ,
j=0
(2.3)
∞
X
Ū α (s) = ūα,j (s) ej∂s , ūα,0 (s) ∈ Mat(N × N, C),
j=0
∞
X
Qα (s) = e−∂sα P α (s), P α (s) = pα,j (s) e−j∂s , pα,0 (s) = 1N ,
j=0
(2.4)
∞
X
Q̄α (s) = e−∂sα P̄ α (s), P̄ α (s) = p̄α,j (s) ej∂s , p̄α,0 (s) ∈ Mat(N × N, C).
j=0

Here α is an index from 1 to N and bj (s), b̄j (s), uα,j (s), ūα,j (s), pα,j (s) and p̄α,j (s) are
matrix-valued functions of size N × N depending on the variables s = {s1 , . . . , sN }, t
and t̄. (We do not write dependence on t and t̄ explicitly unless it is necessary.)
We require that these operators satisfy the following algebraic conditions: for any
α, β = 1, . . . , N it holds that

[L(s), U α (s)] = [L(s), Qα (s)] = [U (s), Qα (s)] = [Qα (s), Qβ (s)] = 0, (2.5)
N
X
U α (s)U β (s) = δαβ U β (s), U α (s) = 1N , (2.6)
α=1

N
Y
Qα (s) = L−1 (s), (2.7)
α=1

[L̄(s), Ū α (s)] = [L̄(s), Q̄α (s)] = [Ū (s), Q̄α (s)] = [Q̄α (s), Q̄β (s)] = 0, (2.8)
N
X
Ū α (s)Ū β (s) = δαβ Ū β (s), Ū α (s) = 1N , (2.9)
α=1

N
Y
Q̄α (s) = L̄(s), (2.10)
α=1

and
N N
−δαβ
X X
δαβ
Qα (s) U β (s)L (s) = Q̄α (s) Ū β (s)L̄ (s) (2.11)
β=1 β=1

for each α = 1, . . . , N.
Under these conditions, the Lax equations for the multi-component Toda lattice hi-
erarchy is the following system: for any indices α, β = 1, . . . , N, n = 1, 2, . . . and for any
operators A = L(s), L̄(s), U β (s), Ū β (s), Qβ (s), Q̄β (s) it holds

∂A ∂A
= [B α,n (s), A], = [B̄ α,n (s), A], (2.12)
∂tα,n ∂ t̄α,n
where
n
B α,n (s) := (Ln (s)U α (s))≥0 , B̄ α,n (s) := (L̄ (s)Ū α (s))<0 . (2.13)

7
Actually, discrete Lax equations for the shifts of sα by 1 are included in (2.5) and (2.8).
(See (2.24) and (2.25) below.) The meaning of condition (2.11) is related to this discrete
evolution. (See Remark 2.3.)

Lemma 2.1. The set of conditions (2.11) (α = 1, . . . , N) is equivalent to

N N N N
a −aβ
Y X Y X
Qaαα (s) aβ
U β (s)L (s) = Q̄αα (s) Ū β (s)L̄ (s). (2.14)
α=1 β=1 α=1 β=1

for all a = {a1 , . . . , aN } ∈ ZN .

Proof. As the s-variables are fixed in this lemma, we do not write them explicitly in the
proof.
Because of the commutativity (2.5) and the idempotence (2.6) we have:

N
Y N
Y
Qaαα a1 aN
Qbαα U 1 Lb1 + · · · + U N LbN



U 1L + · · · + U N L ×
α=1 α=1

N
Y
Qaαα +bα U 1 La1 +b1 + · · · + U N LaN +bN ,


=
α=1

and, in particular,
N N
!−1 N N
Y X Y X
Qaαα U βL aβ
= Q−a
α
α
U β L−aβ .
α=1 β=1 α=1 β=1

Similar equations hold also for L̄ and Ū α ’s because of (2.8) and (2.9). The equivalence
of (2.11) (α = 1, . . . , N) and (2.14) follow from these formulae.
Remark 2.1. The above defined system is an extended version of the multi-component
Toda lattice hierarchy introduced by Ueno and Takasaki in §3.1 of [4]. To recover their
(0) (0)
version we have only to restrict s-variables to the form s = s(0) + s1 = {s1 + s, . . . , sN +
(0) (0)
s} for a fixed s(0) = {s1 , . . . , sN } passing to the single variable s ∈ Z. In this case, there
are only shifts of the form s 7→ s + n1 (n ∈ Z). Then all functions f (s) are regarded as
functions f (s) of s only and the operator e∂s acts as the shift operator: e∂s f (s) = f (s+1).
The algebraic conditions (2.5), (2.6) for L and U α and conditions (2.8), (2.9) for L̄
and Ū α are those required by (3.1.6) in [4]. The differential equations (2.12) for L, L̄,
U and Ū are the system (3.1.8) in [4]. The operators Qα and Q̄α were not considered in
[4]. Note that for a of the form a = a1N condition (2.14) is trivial because of the other
algebraic conditions and we do not have to require (2.11).

Proposition 2.1. (i) The system (2.12) for L(s), L̄(s), U α (s) and Ū α (s) implies the
following compatibility conditions under the algebraic constraints (2.5), (2.6), (2.8) and

8
(2.9):
 
∂ ∂
− B α,m (s), − B β,n (s) = 0, (2.15)
∂tα,m ∂tβ,n
 
∂ ∂
− B̄ α,m (s), − B̄ β,n (s) = 0, (2.16)
∂ t̄α,m ∂ t̄β,n
 
∂ ∂
− B α,m (s), − B̄ β,n (s) = 0. (2.17)
∂tα,m ∂ t̄β,n
 
∂ m ∂ n
+ (L (s)U α (s))<0 , + (L (s)U β (s))<0 = 0, (2.18)
∂tα,m ∂tβ,n
 
∂ m ∂ n
+ (L̄ (s)Ū α (s))≥0 , + (L̄ (s)Ū β (s))≥0 = 0, (2.19)
∂ t̄α,m ∂ t̄β,n
 
∂ m ∂
+ (L (s)U α (s))<0 , − B̄ β,n (s) = 0, (2.20)
∂tα,m ∂ t̄β,n
 
∂ ∂ n
− B α,m (s), + (L̄ (s)Ū β (s))≥0 = 0, (2.21)
∂tα,m ∂ t̄β,n

(ii) Conversely, the system (2.15), (2.16) and (2.17) with the algebraic constraints
(2.5), (2.6), (2.8) and (2.9) implies the Lax representation (2.12) for L(s), L̄(s), U α (s)
and Ū α (s).

We call the system (2.15), (2.16) and (2.17) the Zakharov-Shabat representation of
the multi-component Toda lattice hierarchy.

Proof. As the s-variables are fixed in this proposition, we do not write them explicitly
in the proof, unless it is necessary.
(i) Derivation of the Zakharov-Shabat type equations from the Lax representation
is the same as that for the one-component system as shown in [4]. Indeed, applying
∂tα,m − [B α,m , ·] to Ln U β , we obtain
∂
(Ln U β ) = [B α,m , Ln U β ] = [B α,m , B β,n ] + [B α,m , (Ln U β )<0 ].
∂tα,m
from the Lax equation (2.12) for L and U . Exchanging (α, m) and (β, n), we also have
∂
(Lm U α ) = [B β,n , Lm U α ] = [Ln U β − (Ln U β )<0 , Lm U α ]
∂tβ,n
= −[(Ln U β )<0 , Lm U α ] = −[(Ln U β )<0 , B α,m ] − [(Ln U β )<0 , (Lm U α )<0 ],
because of (2.5) and (2.6). Subtracting the latter equation from the former, we have

∂B β,n ∂B α,m
− − [B α,m , B β,n ]
∂tα,m ∂tβ,n
∂(Lm U α )<0 ∂(Ln U β )<0
= − + [(Ln U β )<0 , (Lm U α )<0 ].
∂tβ,n ∂tα,m

9
The left-hand side is a difference operator with non-negative shifts, while the right-hand
side is a difference operator with negative shifts, which implies that the both sides are
zero. This implies (2.15) and (2.18).
The equations (2.16) and (2.19) are shown in the same way by changing “t” to “t̄”,
“L” to “L̄”, “B” to “B̄” and “U ” to “Ū ” in the above argument.
The equations
∂ ∂ n n
Lm U α = [B̄ β,n , Lm U α ], L̄ Ū β = [B α,m , L̄ Ū β ],
∂ t̄β,n ∂tα,m

follow from the Lax equations (2.12). Therefore, we have

∂B α,m ∂
− [B̄ β,n , B α,m ] = − (Lm U α )<0 + [B̄ β,n , (Lm U α )<0 ],
∂ t̄β,n ∂ t̄β,n
∂ B̄ β,n ∂ n n
− [B α,m , B̄ β,n ] = − (L̄ Ū β )≥0 + [B α,m , (L̄ Ū β )≥0 ].
∂tα,m ∂tα,m

Using these formulae, we can rewrite ∂t̄β,n B α,m − ∂tα,m B̄ β,n + [B α,m , B̄ β,n ] in two ways:

∂B α,m ∂ B̄ β,n
− + [B α,m , B̄ β,n ]
∂ t̄β,n ∂tα,m
∂ ∂ B̄ β,n
= − (Lm Uα )<0 − + [B̄ β,n , (Lm Uα )<0 ],
∂ t̄β,n ∂tα,m
∂B α,m ∂ n n
= + (L̄ Uβ )≥0 − [B α,m , (L̄ Uβ )≥0 ].
∂ t̄β,n ∂tα,m

Here the second line is a difference operator with negative shifts, while the third line
is a difference operator with non-negative shifts. Therefore all the expressions are zero,
which implies (2.17), (2.20) and (2.21).

(ii) Assume that the Zakharov-Shabat equations (2.15), (2.16) and (2.17) hold with
the algebraic conditions (2.5), (2.6), (2.8) and (2.9). We will show that the Lax equations
(2.12) for L, L̄, U α and Ū α follow from these assumptions.
Equation (2.15) implies

∂B β,n ∂B α,m
− [B α,m , B β,n ] = .
∂tα,m ∂tβ,n

Hence,
∂ ∂B α,m ∂
Ln U β − [B α,m , Ln U β ] = + (Ln U β )<0 − [B α,m , (Ln U β )<0 ]. (2.22)
∂tα,m ∂tβ,n ∂tα,m

The sum of the left-hand side of (2.22) for β = 1, . . . , N is equal to

n  
∂ X ∂
Ln − [B α,m , Ln ] = Ln−j L − [B α,m , L] Lj−1 (2.23)
∂tα,m j=1
∂tα,m

10
because of the condition (2.6). The right-hand side of (2.22) shows thatPthis expression is
∞
of order less than m for any n, where the order of a difference operator j=0 aj (s) e(K−j)∂s
is defined to be K if a0 (s) does not vanish.
Suppose that
∞
∂ X
L − [B α,m , L] = aj (s)e(K−j)∂s
∂tα,m j=0

does not vanish and is of order K (a0 (s) 6= 0). Since Ln−j = e(n−j)∂s + · · · and
e(n−j)∂s a0 (s) = a0 (s + (n − j)1) e(n−j)∂s , the right-hand side of (2.23) is of the form
n
!
X
a0 (s + (n − j)1) e(K+n−1)∂s + (sum of ãk (s) ek∂s , k < K + n − 1),
j=1

whose order grows to infinity when n → ∞, which contradicts the previous conclusion
that the order of (2.22) is bounded (< m). Therefore,
∂
L − [B α,m , L] = 0,
∂tα,m
which implies the Lax equation (2.12) for L and tα,m .
From the equation which we have just proved, it follows that the expression (2.23)
vanishes and the left-hand side of (2.22) is equal to
 
∂ n n n ∂
L U β − [B α,m , L U β ] = L U β − [B α,m , U β ] .
∂tα,m ∂tα,m
Recall that this is of order less than m for any n because of (2.22). However, if
∞
∂ X
U β − [B α,m , U β ] = uj (s)e(K−j)∂s
∂tα,m j=0

does not vanish and is of order K (u0 (s) 6= 0), the order of
 
n ∂
L U β − [B α,m , U β ] = u0 (s + n1)e(n+K)∂s + · · ·
∂tα,m
grows to infinity as n → ∞. Hence
∂
U β − [B α,m , U β ] = 0,
∂tα,m
which is (2.12) for U β and tα,m .
Similarly, the equation (2.17) implies
∂ ∂ B̄ α,m ∂(Ln U β )<0
Ln U β − [B̄ α,m , Ln U β ] = + − [B̄ α,m , (Ln U β )<0 ],
∂ t̄α,m ∂tβ,n ∂ t̄α,m
which means that the left-hand side is of negative order for any n. The same argument
as the argument for the Lax equations (2.12) for tα,m leads to the Lax equations for t̄α,m .
The Lax equations for L̄ and Ū β can be proved in the way parallel to the above proof
for L and U β .

11
 The algebraic conditions for Qα (s) and Q̄α (s) in (2.5) and (2.8) are equivalent to the
following conditions for P α (s) and P̄ α (s):

P α (s)L(s) = L(s + [1]α )P α (s), P α (s)U β (s) = U β (s + [1]α )P α (s), (2.24)

P̄ α (s)L̄(s) = L̄(s + [1]α )P̄ α (s), P̄ α (s)Ū β (s) = Ū β (s + [1]α )P̄ α (s), (2.25)
P β (s + [1]α )P α (s) = P α (s + [1]β )P β (s),
(2.26)
P̄ β (s + [1]α )P̄ α (s) = P̄ α (s + [1]β )P̄ β (s).

Remark 2.2. The role of the operators P α (s) and P̄ α (s) is the shift of the variable sα
by 1. They are analogues of operators P (s) in [16] and Pα,β (s, t, ∂) in [8].
Remark 2.3. We can interpret the condition (2.11) as follows. Let us rewrite it in terms
of the evolution operators P α (s) and P̄ α (s):
N N
−δαβ
X X
δαβ
P α (s) U β (s)L (s) = P̄ α (s) Ū β (s)L̄ (s). (2.27)
β=1 β=1

This is formally equivalent to

N N
!−1
−δαβ
X X
P −1
α (s)P̄ α (s) = U β (s)Lδαβ (s) Ū β (s)L̄ (s) . (2.28)
β=1 β=1

This means that P α and P̄ α are obtained by the Bruhat or Riemann-Hilbert type de-
composition of the right-hand side. (In general, both sides of (2.28) do not converge and
we need to use the form (2.27).)
This is a multiplicative analogue of the definitions (2.13) of B α,n and B̄ α,n which are
obtained from the additive Bruhat or Riemann-Hilbert type decomposition of a product
of L and U α or a product of L̄ and Ū α .

Generally, one can define operators P a (s) and P̄ a (s) for any a ∈ ZN which add a
to the s-variable in the L- and U α -operators as

P a (s)L(s) = L(s + a)P a (s), P a (s)U α (s) = U α (s + a)P a (s), (2.29)

P̄ a (s)L̄(s) = L̄(s + a)P̄ a (s), P̄ a (s)Ū α (s) = Ū α (s + a)P̄ a (s), (2.30)

and have the composition rules

P b (s + a)P a (s) = P a+b (s), P̄ b (s + a)P̄ a (s) = P̄ a+b (s). (2.31)

In fact, one has only to put

N N
PN PN
a
Y Y
P a (s) := e α=1 aα ∂sα
Qaαα (s), P̄ a (s) := e α=1 aα ∂sα
Q̄αα (s). (2.32)
α=1 α=1

12
It is easy to check that the Lax equations (2.12) for Qα (s) and Q̄α (s) imply

∂P a (s)
= B α,n (s + a)P a (s) − P a (s)B α,n (s), (2.33)
∂tα,n
∂P a (s)
= B̄ α,n (s + a)P a (s) − P a (s)B̄ α,n (s), (2.34)
∂ t̄α,n
∂ P̄ a (s)
= B α,n (s + a)P̄ a (s) − P̄ a (s)B α,n (s), (2.35)
∂tα,n
∂ P̄ a (s)
= B̄ α,n (s + a)P̄ a (s) − P̄ a (s)B̄ α,n (s). (2.36)
∂ t̄α,n

The following proposition provides some information2 about the leading coefficients
of the operators L̄, Ū α , P̄ α .

Proposition 2.2. There exists an invertible matrix-valued function w̃0 = w̃0 (s, t, t̄) such
that
b̄0 (s) = w̃0 (s)w̃0−1(s − 1),

ūα,0 (s) = w̃0 (s)Eα w̃0−1 (s), (2.37)

p̄α,0 (s) = w̃0 (s + [1]α )w̃0−1(s).

Proof. The existence of such w̃0 follows from the algebraic conditions (2.5)–(2.10) and
their corollaries (2.25), (2.26). From (2.26) we have, comparing the leading coefficients
in the second equation:

p̄α,0 (s + [1]β )p̄β,0 (s) = p̄β,0 (s + [1]α )p̄α,0 (s). (2.38)

It follows from this condition that there exists a matrix-valued function w̃0 (s) such that

p̄α,0 (s) = w̃0 (s + [1]α )w̃0−1 (s) (2.39)

(equation (2.38) is the compatibility condition for recursive relations w̃0 (s + [1]α ) =
p̄α,0 (s)w̃0 (s) which allow one to find values of the function w̃0 (s) in all points of the
s-lattice starting from some initial value w̃0 (0)). Plugging this into the second equation
of (2.25) and equating the leading coefficients, we have:

w̃0−1(s)ūβ,0 (s)w̃0 (s) = w̃0−1(s + [1]α )ūβ,0(s + [1]α )w̃0 (s + [1]α )

for all α, whence

ūβ,0 (s) = w̃0 (s)ūβ,0w̃0−1 (s), (2.40)
where ūβ,0 is an s-independent matrix. From commutativity of the operators Ū α it
follows that the matrices ūβ,0 for β = 1, . . . , N commute and can be simultaneously
N
X
diagonalized. The algebraic conditions (2.9) imply that ūβ,0 ūα,0 = δαβ ūβ,0, ūα,0 = 1N ,
α=1

2
In [4] it was assumed from the beginning that b̄0 (s) and ūα,0 (s) had the form (2.37).

13
hence the matrix ūβ,0 has N −1 zero eigenvalues and one eigenvalue equal to 1. Therefore,
if we choose the order of eigenvectors of ūβ,0’s appropriately, ūβ,0 = v̄Eβ v̄ −1 , where v̄ is
a non-degenerate s-independent matrix. The redefinition w̃0 → w̃0 v̄ −1 does not spoil the
relation (2.39) and allows one to put ūβ,0 = Eβ in (2.40) without loss of generality. At
last, from (2.10) we conclude that

b̄0 (s) = w̃0 (s)w̃0−1 (s − 1). (2.41)

Thus all the relations (2.37) are proved.

Remark 2.4. There is a freedom in the choice of the matrix w̃0 (s): it can be multiplied
from the right by an arbitrary invertible diagonal matrix which may depend on t, t̄ but
not on s. This fact will be essential below in section 2.2.

2.2 Gauge transformations

This subsection is an extended remark on the choice of the first coefficients in the opera-
tors L, U α , P α in (2.2)–(2.4) (b0 (s) = pα,0 (s) = 1N , uα,0 (s) = Eα ) and on the resulting
asymmetry in the definitions of the operators L, U α , P α and their bar-counterparts.
Let the coefficients b0 (s), uα,0 (s), pα,0 (s) of the operators L, U α , P α in (2.2)–(2.4)
be some matrix-valued functions of s, t, t̄. Suppose that b0 (s) and pα,0 (s) are invertible
matrices. Then the Lax equations (2.12) imply that they do not depend on t, t̄. Indeed,
from
∂L
= [(Lk U α )≥0 , L] = −[(Lk U α )<0 , L],
∂tα,k

∂L k
= [(L̄ Ū α )<0 , L],
∂ t̄α,k
we deduce, comparing the coefficients in front of e∂s , that
∂b0 (s) ∂b0 (s)
= = 0 for all α, k.
∂tα,k ∂ t̄α,k
The argument for uα,0 (s), pα,0 (s) is similar. To find the dependence of these functions on
s, we almost repeat the proof of Proposition 2.2, using the discrete Lax equations (2.24),
(2.26). From (2.26) we have, comparing the leading coefficients in the first equation:

pα,0 (s + [1]β )pβ,0 (s) = pβ,0 (s + [1]α )pα,0 (s). (2.42)

Similarly to the proof of Proposition 2.2, it follows from this condition that there exists
a matrix-valued function w0 (s) such that

pα,0 (s) = w0 (s + [1]α )w0−1(s).

Plugging this into (2.24) and equating the leading coefficients, we have:

w0−1(s)uβ,0 (s)w0 (s) = w0−1(s + [1]α )uβ,0(s + [1]α )w0 (s + [1]α )

for all α, whence

uβ,0 (s) = w0 (s)uβ,0w0−1 (s),

14
where uβ,0 is a constant (s-independent) matrix. The algebraic conditions (2.6) imply
N
X
that uβ,0 uα,0 = δαβ uβ,0, uα,0 = 1N , whence uβ,0 = vEβ v −1 , where v is a constant
α=1
non-degenerate matrix. At last, from (2.7) we conclude that
b0 (s) = w0 (s)w0−1 (s + 1).
Therefore, re-defining w0 (s) → w0 (s)v, we see that the “gauge transformation” A →
w0−1 (s)Aw0 (s), where A is any one of the operators L, U α , Qα , L̄, Ū α , Q̄α , allows us,
without loss of generality, to fix the coefficients b0 (s), uα,0 (s), pα,0 (s) to be
b0 (s) = pα,0 (s) = 1N , uα,0 (s) = Eα ,
as in (2.2)–(2.4). Since w0 (s) does not depend on t, this transformation preserves the
form of the Lax and Zakharov-Shabat equations.
One can also consider more general gauge transformations A 7→ A(G) = G−1 AG,
where the matrix G may depend also on t and t̄: G = G(s, t, t̄). In particular, the
transformed Lax operators become
L(G) (s) = G−1 (s)G(s + 1)e∂s + . . . ,
(2.43)
(G)
L̄ (s) = G −1
(s)w̃0 (s)w̃0−1 (s − 1)G(s − 1)e −∂s
+ ... ,
where w̃0 (s) is introduced by Proposition 2.2. In order to preserve the form of the Lax
and Zakharov-Shabat equations, the difference operators B α,k , B̄ α,k should transform as
follows:
(G)
B α,k 7→ B α,k = G−1 B α,k G − G−1 ∂tα,k G = ((L(G) )k U (G) −1
α )≥0 − G ∂tα,k G,
(2.44)
(G) −1 −1 (G) k (G) −1
B̄ α,k 7→ B̄ α,k = G B̄ α,k G − G ∂t̄α,k G = ((L̄ ) Ū α )<0 − G ∂t̄α,k G,
so that  
∂ (G) ∂
− B α,k = G−1 − B α,k G,
∂tα,k ∂tα,k
 
∂ (G) −1 ∂
− B̄ α,k =G − B̄ α,k G.
∂ t̄α,k ∂ t̄α,k
In particular, one may take G = w̃0 (s). Let us denote the operators and their coefficients
transformed in this way by adding prime: L 7→ L′ = w̃0−1 Lw̃0 , etc. Then the leading
′ ′ ′
coefficients of the operators L′ , L̄ , U ′α , Ū α , P ′α , P̄ α are, respectively:
b′0 (s) = w̃0−1 (s)w̃0 (s + 1), b̄′0 (s) = 1N ,

u′α,0 (s) = w̃0−1 (s)Eα w̃0 (s), ū′α,0 = Eα ,

p′α,0 (s) = w̃0−1 (s + [1]α )w̃0 (s), p̄′α,0 (s) = 1N .

Comparing the leading coefficients in the Lax type equation
′
∂ P̄ α (s) ′ ′
= B ′β,k (s + [1]α )P̄ α (s) − P̄ α (s)B ′β,k (s),
∂tβ,k

15
one finds that the left-hand side is 0 while the right-hand side yields

(B ′β,k (s + [1]α ))0 = (B ′β,k (s))0 ,

i.e., (B ′β,k (s))0 does not depend on s: (B ′β,k (s))0 = cβ,k (t, t̄) (by (A)0 we denote the
coefficient in front of e0∂s in the difference operator A). The transformation laws (2.44)
give in our case:
B ′β,k (s) = ((L′ )k U ′β )≥0 − w̃0−1 (s)∂tβ,k w̃0 (s),
(2.45)
′ ′ k ′ −1
B̄ β,k (s) = ((L̄ ) Ū β )<0 − w̃0 (s)∂t̄β,k w̃0 (s).
Therefore, taking the ( )0 -part of the first equation, we have:

((L′ )k U ′β )0 = w̃0−1(s)∂tβ,k w̃0 (s) + cβ,k (t, t̄) (2.46)

which means that

B ′β,k (s) = ((L′ )k U ′β )>0 + cβ,k (t, t̄).
Similar manipulations with the equation

∂P ′α (s) ′ ′
= B̄ β,k (s + [1]α )P ′α (s) − P ′α (s)B̄ β,k (s)
∂ t̄β,k

lead to
′ ′
((L̄ )k Ū β )0 = −w̃0−1 (s)∂tβ,k w̃0 (s) + w̃0−1 (s)c̄β,k (t, t̄)w̃0 (s) (2.47)
which means that
′ ′ ′
B̄ β,k (s) = ((L̄ )k Ū β )≤0 − w̃0−1 (s)c̄β,k (t, t̄)w̃0 (s)

with some s-independent matrix-valued function c̄β,k .

Some additional arguments allow one to put cβ,k (t, t̄) = c̄β,k (t, t̄) = 0 without any
loss of generality. Indeed, taking the ( )0 -part of the equation

∂ Ū α
= [B β,k , Ū α ],
∂tβ,k

we deduce that
[w̃0−1 (B β,k )0 w̃0 − w̃0−1 ∂tβ,k w̃0 , Eα ] = 0.
Together with (2.46) this means that

[cβ,k (t, t̄), Eα ] = 0 for all α, (2.48)

whence cβ,k (t, t̄) is a diagonal matrix. Let us rewrite (2.46) in the form

(B β,k )0 = (Lk U β )0 = ∂tβ,k w̃0 (s) · w̃0−1 (s) + w̃0 (s)cβ,k (t, t̄)w̃0−1 (s). (2.49)

Note that the second term in the right-hand Rside can be eliminated  by multiplying w̃0 (s)
tβ,k
by an s-independent diagonal matrix exp cβ,k (t, t̄) dtβ,k from the right. But this
is just the freedom left in the definition of the matrix w̃0 (s) by the formulae (2.2)–(2.4).

16
Therefore, we can put cβ,k (t, t̄) = 0 without loss of generality. Let us rewrite equation
(2.47) in the form

(B̄ β,k )0 = ((L̄)k Ū β )0 = −∂t̄β,k w̃0 (s) · w̃0−1(s) + c̄β,k (t, t̄) (2.50)

and prove that the choice cβ,k (t, t̄) = 0 implies that c̄β,k (t, t̄) = 0. This fact follows from
the Zakharov-Shabat equation (2.21). Its ( )0 -part is
h i
k
∂tα,m − (B α,m )0 , ∂t̄β,k + (L̄ Ū β )0 = 0

The general solution is

k
(B α,m )0 = ∂tα,m W · W −1 , (L̄ Ū β )0 = −∂t̄β,k W · W −1 ,

where W is an arbitrary matrix-valued function3 . According to (2.49) with cβ,k = 0, we

should put W = w̃0 (s), which means (after comparison with (2.50)) that c̄β,k (t, t̄) = 0.
Finally, we have:
′ ′ ′
B ′β,k = ((L′ )k U ′β )>0 , B̄ β,k = ((L̄ )k Ū β )≤0 . (2.51)

Therefore, we see that in this gauge the roles of the two Lax operators are exchanged.
This restores the symmetry in their definitions, which is broken in (2.2)–(2.4).
Remark 2.5. The gauge transformations of the one-component Toda lattice were dis-
cussed in [17]. They can be represent in the form
(G)
L 7→ L(G) := G−1 LG, L̄ 7→ L̄ := G−1 L̄G,
∂G (G) ∂G
B n 7→ B (G) −1
n := g B n G − G
−1
, B̄ n 7→ B̄ n := G−1 B̄ n G − G−1 ,
∂tn ∂ t̄n
   
∂ ∂ ∂ ∂ (G)
or G −1
− Bn G = − B (G)
n , G −1
− B̄ n G = − B̄ n .
∂tn ∂tn ∂ t̄n ∂ t̄n
1/2
The choice G = w̃0 (s) corresponds to the symmetric gauge, in which the leading co-
(sym)
efficients of the two Lax operators L(sym) , L̄ become equal up to a shift of s by
1:
(s) −1/2 1/2
L(sym) = b0 (s)e∂s + . . . = w̃0 (s)w̃0 (s + 1)e∂s + . . . ,

(sym) (s) −1/2 1/2

L̄ = b0 (s − 1)e−∂s + . . . = w̃0 (s − 1)w̃0 (s)e−∂s + . . .
(here w̃0 (s) is a scalar function, so we can change the order in the product). The B-
operators in the symmetric gauge look as follows:
(sym)
Bk = ((L(sym) )k )>0 + 21 ((L(sym) )k )0 ,
(2.52)
(sym) (sym) k 1 (sym) k
B̄ k = ((L̄ ) )<0 + 2
((L̄ ) )0 .

In the gauge with G = w̃0 (s) the roles of L and L̄ are exchanged. See [17] for details.
3
We use the fact that the general solution of the matrix equation [∂x − Bx , ∂y − By ] = 0 is Bx =
∂x W · W −1 , By = ∂y W · W −1

17
 An analogue of the symmetric gauge exists in the multi-component case, too. Taking
1/2
G = w̃0 (s), we have:
−1/2 1/2
L(sym) (s) = w̃0 (s)w̃0 (s + 1)e∂s + . . . ,
(2.53)
(sym) 1/2 −1/2
L̄ (s) = w̃0 (s)w̃0 (s − 1)e−∂s + . . . .

Because of the non-commutativity of matrices, the leading coefficients are not related in
(sym)
the same simple way as in the one-component case. The B (sym) - and B̄ -operators in
the symmetric gauge are expressed as follows:
(sym) (sym) 1/2 −1/2
B β,k = ((L(sym) )k U β )>0 + ∂tβ,k w̃0 · w̃0 ,
(2.54)
(sym) (sym) k (sym) −1/2 1/2
B̄ β,k = ((L̄ ) Ū β )<0 − w̃0 · ∂t̄β,k w̃0 .
1/2 −1/2 (sym)
However, since, for example, ∂tβ,k w̃0 6= 21 w̃0 ∂tβ,k w̄0 , the B (sym) - and B̄ -operators
do not have such a simple description as in the one-component case.

3 Linearization: wave operators and wave functions

3.1 Wave operators

Here we derive the linear problems associated with the multi-component Toda lattice
hierarchy defined in Section 2.1. The first half of the derivation is almost the same as
the one given by Ueno and Takasaki, but some details of the proof of the existence of the
wave matrix functions are added, correcting inaccurate arguments in §3.2 of [4].
Proposition 3.1. (i) Let (L, L̄, U α , Ū α , Qα , Q̄α )α=1,...,N be a solution of the N-component
Toda lattice hierarchy. Then there exist operators W (s, t, t̄) and W̄ (s, t, t̄) of the form
∂s )
W (s, t, t̄) = Ŵ (s, t, t̄) diagα (eξ(tα ,e ),
ˆ (s, t, t̄) diag (eξ(t̄α ,e−∂s ) ),
W̄ (s, t, t̄) = W̄ α
∞
X
Ŵ (s, t, t̄) = wj (s, t, t̄) e−j∂s , w0 (s, t, t̄) = 1N , (3.1)
j=0

∞
ˆ (s, t, t̄) =
X
W̄ w̄j (s, t, t̄) ej∂s , w̄0 (s, t, t̄) ∈ GL(N, C),
j=0

where  
∞ a1
..
X
ξ(tα , e∂s ) = tα,n en∂s , diagα (aα ) =  . ,
 
n=1 aN (3.2)
wj (s, t, t̄) = (wj,αβ (s, t, t̄))α,β=1,...,N ∈ Mat(N × N, C),
w̄j (s, t, t̄) = (w̄j,αβ (s, t, t̄))α,β=1,...,N ∈ Mat(N × N, C),

18
satisfying the following linear equations:

L(s)W (s) = W (s)e∂s , U α (s)W (s) = W (s)Eα , (3.3)

L̄(s)W̄ (s) = W̄ (s)e−∂s , Ū α (s)W̄ (s) = W̄ (s)Eα , (3.4)
∂W (s) ∂W (s)
= B α,n (s)W (s), = B̄ α,n (s)W (s), (3.5)
∂tα,n ∂ t̄α,n
∂ W̄ (s) ∂ W̄ (s)
= B α,n (s)W̄ (s), = B̄ α,n (s)W̄ (s), (3.6)
∂tα,n ∂ t̄α,n

and
Qα (s)W (s) = W (s)e−∂sα , Q̄α (s)W̄ (s) = W̄ (s)e−∂sα . (3.7)
ˆ (s), equations (3.3) and (3.4) are equivalent to the following
In terms of Ŵ (s) and W̄
equations:

L(s)Ŵ (s) = Ŵ (s)e∂s , U α (s)Ŵ (s) = Ŵ (s)Eα , (3.8)

ˆ (s) = W̄
L̄(s)W̄ ˆ (s)e−∂s , ˆ (s) = W̄
Ū α (s)W̄ ˆ (s)E , (3.9)
α

and equations (3.5), (3.6) are equivalent to

∂ Ŵ (s) ∂ Ŵ (s)
= B α,n (s)Ŵ (s) − Ŵ (s)en∂s Eα , = B̄ α,n (s)Ŵ (s), (3.10)
∂tα,n ∂ t̄α,n
ˆ (s)
∂ W̄ ˆ (s)
∂ W̄
ˆ (s),
= B α,n W̄ ˆ (s) − W̄
= B̄ α,n (s)W̄ ˆ (s)e−n∂s E . (3.11)
α
∂tα,n ∂ t̄α,n

ˆ are rewritten as
Relations (3.7) in terms of Ŵ and W̄

Qα (s)Ŵ (s) = Ŵ (s)e−∂sα , Q̄α (s)W̄ ˆ (s)e−∂sα ,

ˆ (s) = W̄ (3.12)

or, in terms of P α ,

P α (s)Ŵ (s) = Ŵ (s + [1]α ), P̄ α (s)W̄ ˆ (s + [1] ).

ˆ (s) = W̄ (3.13)
α

ˆ (s) are unique up to multiplication of matrix difference

The operators Ŵ (s) and W̄
operators of the form
∞
X
Ŵ (s, t, t̄) 7→ Ŵ (s, t, t̄) cj e−j∂s , c0 = 1 N ,
j=0

∞
ˆ (s, t, t̄) 7→ W̄
ˆ (s, t, t̄)
X
W̄ c̄j ej∂s , c̄0 is invertible,
j=0

where cj and c̄j are diagonal matrices4 which do not depend on t, t̄ and s.
4
In Theorem 3.3 of [4] it is not stated that cj and c̄j should be diagonal.

19
 ˆ (s, t, t̄) of the form (3.1) are solutions of differ-
(ii) Conversely, if Ŵ (s, t, t̄) and W̄
ential equations (3.10) and (3.11) for certain difference operators B α,n and B̄ α,n , then
the operators L, L̄, U α , Ū α , Qα , Q̄α defined by (3.8), (3.9) and (3.12), or, equivalently,
by
−1 −1
L(s) := Ŵ (s)e∂s Ŵ (s), U α (s) := Ŵ (s)Eα Ŵ (s),
ˆ (s)e−∂s W̄
L̄(s) := W̄ ˆ −1 (s), ˆ (s)E W̄
Ū α (s) := W̄ ˆ −1 (s), (3.14)
α

−1 ˆ (s)e−∂sα Ŵ −1 (s),
Qα (s) := Ŵ (s)e−∂sα Ŵ (s), Q̄α (s) := W̄
solve (2.12) and also satisfy (2.13) and relations ( 2.5, 2.6, 2.7, 2.8, 2.9, 2.10).

We call W and W̄ wave operators. They are also sometimes called dressing operators
because the expression L(s) = W (s)e∂s W −1 (s) equivalent to the first equation in (3.3)
is interpreted as a “dressing” of the “bare” shift operator e∂s by the wave operator W .

ˆ ) follows immediately
Proof. (i) The equivalence of equations for (W , W̄ ) and (Ŵ , W̄
from e∂s Eα = Eα e∂s .
We begin the proof assuming that each sequence of the s-variables is of the form
(0) (0)
s = s(0) +s1 = {s1 +s, . . . , sN +s} for a fixed s(0) and consider all functions as functions
of the single variable s. We denote L(s, t, t̄), Ŵ (s, t, t̄), ... by L(s, t, t̄), Ŵ (s, t, t̄), ...
respectively.
The proof of statement (i) goes in a few steps.

ˆ (s) = W̄
1. Find Ŵ 0 (s) = Ŵ (s, 0, 0) satisfying (3.8) and W̄ ˆ (s, 0, 0) satisfying (3.9)
0
at t = t̄ = 0.
ˆ (s) to construct
2. Solve the differential equations with initial values Ŵ 0 (s) and W̄ 0
ˆ (s, t, t̄).
Ŵ (s, t, t̄) and W̄
ˆ (s, t, t̄) satisfy (3.8), (3.9), (3.10) and (3.11).
3. Show that these Ŵ (s, t, t̄) and W̄

4. Show the uniqueness up to multiplication of constant diagonal matrix difference

operators.

5. Restore the dependence of the s-variables s = {s1 , . . . , sN } to prove the existence

ˆ (s, t, t̄) satisfying (3.12) or, equivalently, (3.13).
of Ŵ (s, t, t̄) and W̄

ˆ (s): Recall that L(s, 0, 0) has the form

Step 1. Construction of Ŵ 0 (s) and W̄ 0

∞
X ∞
X
L(s, 0, 0) = b0,j (s)e(1−j)∂s := bj (s, t = 0, t̄ = 0)e(1−j)∂s ,
j=0 j=0

Hence, at t = t̄ = 0
∞
X
Ŵ 0 (s) = w0,j (s)e−j∂s
j=0

20
should satisfy
j
∞
! ∞
X X X
(1−j)∂s
b0,k (s) w0,j−k (s + 1 − k) e = w0,j (s)e(1−j)∂s
j=0 k=0 j=0

by the first equation of (3.8), which means that

j
X
b0,k (s) w0,j−k (s + 1 − k) = w0,j (s)
k=0

for each integer j ≥ 1. It is equivalent to the following system of difference equations for
{w0,j (s)}j=0,1,...:
j
X
w0,j (s + 1) − w0,j (s) = − b0,k (s) w0,j−k (s + 1 − k),
k=1

which can be solved recursively with respect to j, starting from w0,0 (s) = 1N .
Note
P∞that such an operator Ŵ 0 (s) is unique up to multiplication of an operator of the
−j∂s
form j=0 cj e from the right, where cj is a constant matrix of size N ×N and c0 = 1N .
In fact, if Ŵ 0,1 (s) and Ŵ 0,2 (s) satisfy L(s, 0, 0) Ŵ 0,i (s) = Ŵ 0,i (s) e∂s (i = 1, 2), then
−1 −1
Ŵ 0,1 (s)Ŵ 0,2 (s) e∂P
s
= e∂s Ŵ 0,1 (s)Ŵ 0,2 (s). An operator commutes with e∂s , if and only
if it has the form cj e−j∂s with constant cj , so
∞
!
X
Ŵ 0,2 (s) = Ŵ 0,1 (s) cj e−j∂s .
j=0

It follows c0 = 1N from the condition w0,0 (s) = 1N .

We construct Ŵ 0 (s) satisfying the condition U α (s, 0, 0)Ŵ 0 (s) = Ŵ 0 (s)Eα in (3.8),
making use of this ambiguity5 .
Let us take any Ŵ 0 (s) satisfying L(s, 0, 0)Ŵ 0 (s) = Ŵ 0 (s)e∂s and call it Ŵ 0,0 (s) =
P∞ −j∂s −1
j=0 w0,0,j (s)e . By the adjoint action of Ŵ 0,0 (s) to (2.5) and (2.6) we obtain the
(0) −1
following conditions for Ũ α (s) := Ŵ 0,0 (s) U α (s, t = 0, t̄ = 0) Ŵ 0,0 (s):
(0)
[e∂s , Ũ α (s)] = 0,
N (3.15)
(0) (0) (0) X (0)
Ũ α (s)Ũ β (s) = δαβ Ũ β (s), Ũ α (s) = 1N .
α=1
(0) (0) P∞ (0) −j∂s (0)
Since Ũ α (s) commutes with e∂s , it has a form Ũ α (s) = j=0 ũα,j e , where ũα,j is
(0)
a constant N × N-matrix. Moreover, since uα,0 = Eα and w0,0,0 (s) = 1N , ũα,0 = Eα .
(0) (0)
Namely, Ũ α = Ũ α (s) can be expanded as
∞
(0) X (0)
Ũ α = Eα + ũα,j e−j∂s . (3.16)
j=1

5
Here we refine the proof of Theorem 3.3 in [4]. As the equation for W (3) in that proof is degenerate,
it is not obvious that it has a solution.

21
Assume that we have Ŵ 0,k (s) (k ≥ 0) which satisfies
∞
(k) (k) X (k)
U α (s, 0, 0)Ŵ 0,k (s) = Ŵ 0,k (s)Ũ α , Ũ α = Eα + ũα,j e−j∂s , (3.17)
j=k+1

(k)
where ũα,j ∈ Mat(N × N, C). Indeed the above chosen Ŵ 0,0 (s) satisfies this condition
for k = 0.
By the same argument as that for Ŵ 0,0 (s) we can show that
∞
(k) X (k)
Ũ α = Eα + ũα,j e−j∂s
j=k+1

in (3.17) satisfies algebraic equations (3.15) with the index (k) instead of (0). The
(k)
2 (k)
equation Ũ α = Ũ α is expanded as
(k) (k) (k)
Eα + (Eα ũα,k+1 + ũα,k+1Eα )e−(k+1)∂s + · · · = Eα + ũα,k+1e−(k+1)∂s + · · · .

Hence,
(k) (k) (k)
Eα ũα,k+1 + ũα,k+1 Eα = ũα,k+1 ,
(k) (k) (k)
which means 0 = (ũα,k+1)ij (i 6= α, j 6= α, i 6= j) and 2(ũα,k+1)αα = (ũα,k+1)αα , namely,
(k)
(ũα,k+1)ij = 0 if (i 6= α and j 6= α), or i = j = α. (3.18)
(k) (k) (k) (k)
The commutativity equation Ũ α Ũ β = Ũ β Ũ α (α 6= β) is expanded as
(k) (k) (k) (k)
(ũα,k+1Eβ + Eα ũβ,k+1)e−(k+1)∂s + · · · = (ũβ,k+1Eα + Eβ ũα,k+1)e−(k+1)∂s + · · · ,

in particular,
(k) (k) (k) (k)
ũα,k+1 Eβ + Eα ũβ,k+1 = ũβ,k+1Eα + Eβ ũα,k+1 .
Together with (3.18) this implies
(k) (k) (k) (k)
(ũα,k+1)αβ + (ũβ,k+1)αβ = (ũα,k+1)βα + (ũβ,k+1)βα = 0. (3.19)
PN (k)
(This is also a consequence of the condition α=1 Ũ α = 1N in (3.15).)
If we can find an operator W̃ k,k+1 such that
∞
!
(k) X
Ũ α W̃ k,k+1 = W̃ k,k+1 Eα + ũ˜α,j e−j∂s , (3.20)
j=k+2

the operator Ŵ 0,k+1(s) satisfying (3.17) for k + 1 is obtained as

Ŵ 0,k+1 (s) = Ŵ 0,k (s)W̃ k,k+1

from Ŵ 0,k (s).

22
 As such an operator W̃ k,k+1, we take
(k) (k)
W̃ k,k+1 = 1N + w ′ e−(k+1)∂s , (w ′ )αβ := (ũβ,k+1)αβ = −(ũα,k+1 )αβ . (3.21)

It is easy to see that this W̃ k,k+1 satisfies (3.20) due to (3.18) and (3.19).
Note that multiplication Ŵ 0,k+1 (s) = Ŵ 0,k (s)W̃ k,k+1 by W̃ k,k+1 of the form (3.21)
does not change the coefficients ŵ0,k,j (s) (j = 1, . . . , k) in the expansion
∞
X
Ŵ 0,k (s) = 1N + ŵ0,k,j (s) e−j∂s .
j=1

Hence the sequence {Ŵ 0,k (s)}k constructed in this way has a limit

Ŵ 0 (s) := lim Ŵ 0,k (s), (3.22)

k→∞

which satisfies the second equation in (3.8) for t = t̄ = 0, U α (s, 0, 0)Ŵ 0 (s) = Ŵ 0 (s)Eα .
The first equation L(s, 0, 0)Ŵ 0 (s) = Ŵ 0 (s) e∂s of (3.8) is kept unchanged in the above
procedure, as we have already mentioned.
The operator W̄ ˆ (s) satisfying (3.9) at t = t̄ = 0 is constructed in the same way.
0
The first term w̄0,0 (s) in
∞
ˆ (s) =
X
W̄ w̄0,j (s) ej∂s
j=0

is w̃0 (s, t = 0, t̄ = 0), where w̃0 (s, t, t̄) is the matrix introduced in Proposition 2.2.
ˆ (s): The system
Step 2. Solving the differential equations for Ŵ (s) and W̄

∂ Ŵ (s) ∂ Ŵ (s)
= −(Ln (s)U α (s))<0 Ŵ (s), = B̄ α,n (s)Ŵ (s) (3.23)
∂tα,n ∂ t̄α,n
is compatible because of (2.18), (2.20) and (2.16). Actually this is equivalent to the
system (3.10), if Ŵ (s) satisfies (3.8). Let Ŵ (s) be the unique solution of (3.23) with
the initial value Ŵ (s, t = 0, t̄ = 0) = Ŵ 0 (s).
Since the coefficients in the right-hand sides of the equations in (3.23) are difference
operators with negative shifts, the solution Ŵ is of the form
∞
X
Ŵ (s, t, t̄) = wj (s, t, t̄)e−j∂s ,
j=0

with w0 (s, t, t̄) = 1N .

Similarly, the system
ˆ (s)
∂ W̄ ˆ (s)
∂ W̄
ˆ (s),
= B α,n (s)W̄
n ˆ (s),
= −(L̄ (s)Ū α (s))≥0 W̄ (3.24)
∂tα,n ∂ t̄α,n

is compatible due to (2.15), (2.19) and (2.21) and equivalent to (3.11), if W̄ ˆ satisfies
ˆ (s) with the initial value W̄
(3.9). We take its solution W̄ ˆ (s, t = 0, t̄ = 0) = W̄
ˆ (s).
0

23
 Since the coefficients in the right-hand sides of the equations in (3.24) are difference
ˆ (s) is of the form
operators with non-negative shifts, the solution W̄
∞
ˆ (s, t, t̄) =
X
W̄ w̄j (s, t, t̄) ej∂s ,
j=0

with w̄0 (s, t = 0, t̄ = 0) = w̃0 (s, t = 0, t̄ = 0).

Step 3. The proof that Ŵ and W̄ˆ constructed above satisfy equations (3.8), (3.9),
(3.10) and (3.11): As we have mentioned in Step 2, equation (3.10) is a consequence
of (3.8) and (3.23), while equation (3.11) is a consequence of (3.9) and (3.24). So, it
is enough to show (3.8) and (3.9). Since we do not touch s here, we do not write it
explicitly in the formulae below.
The first Lax equation in (2.12) for L and the first equation of (3.23) imply
∂
(LŴ − Ŵ e∂s ) = [B α,n , L]Ŵ − L(Ln U α )<0 Ŵ − (Ln U α )<0 Ŵ e∂s
∂tα,n
= B α,n LŴ − L(Ln U α )Ŵ − (Ln U α )<0 Ŵ e∂s

= B α,n LŴ − (Ln U α )LŴ − (Ln U α )<0 Ŵ e∂s

= −(Ln U α )<0 (LŴ − Ŵ e∂s )

because L and U α commute. Similarly, the second Lax equation in (2.12) and the second
equation of (3.23) imply
∂
(LŴ − Ŵ e∂s ) = [B̄ α,n , L]Ŵ + LB̄ α,n Ŵ − B̄ α,n Ŵ e∂s
∂ t̄α,n
= B̄ α,n (LŴ − Ŵ e∂s ).
Therefore, LŴ − Ŵ e∂s satisfies the system (3.23) instead of Ŵ with the initial value
(LŴ − Ŵ e∂s )|t=t̄=0 = L(0, 0)Ŵ 0 − Ŵ 0 e∂s = 0. Since the solution of the Cauchy
problem for the compatible system is unique, LŴ − Ŵ e∂s = 0 for all t and t̄. Thus we
have proved the first equation of (3.8).
The proof of the second equation of (3.8) is similar: the first Lax equation (2.12) for
U α and the first equation of (3.23) imply
∂
(U α Ŵ − Ŵ Eα ) = [B α,n , U α ]Ŵ − U α (Ln U α )<0 Ŵ − (Ln U α )<0 Ŵ Eα
∂tα,n
= B α,n U α Ŵ − U α (Ln U α )Ŵ − (Ln U α )<0 Ŵ Eα

= B α,n U α Ŵ − (Ln U α )U α Ŵ − (Ln U α )<0 Ŵ Eα

= −(Ln U α )<0 (U α Ŵ − Ŵ Eα ),
and the second Lax equation (2.12) for U α and the second equation of (3.23) imply
∂
(U α Ŵ − Ŵ Eα ) = [B̄ α,n , U α ]Ŵ + U α B̄ α,n Ŵ − B̄ α,n Ŵ Eα
∂ t̄α,n
= B̄ α,n (U α Ŵ − Ŵ Eα ).

24
Together with (U α Ŵ − Ŵ Eα )|t=t̄=0 = U α (0, 0)Ŵ 0 − Ŵ 0 Eα = 0 these equations mean
that U α Ŵ − Ŵ Eα = 0.
ˆ can be proved in the same way.
Equations (3.9) for W̄

Step 4. Check the uniqueness of Ŵ and W̄ ˆ : Assume that there are two operators
Ŵ 1 and Ŵ 2 satisfying the conditions (3.8) and (3.10), or equivalently, (3.8) and (3.23).
Here again we do not write s explicitly.
The first equation of (3.8) implies
∞
∂s −1 −1 −1 X
e (Ŵ 1 Ŵ 2 ) = (Ŵ 1 Ŵ 2 )e∂s , i.e., Ŵ 1 Ŵ 2 = cj (t, t̄)e−j∂s ,
j=0

where cj (t, t̄) is an N × N-matrix independent of s. It follows from the second equation
of (3.8) that
−1 −1
Eα (Ŵ 1 Ŵ 2 ) = (Ŵ 1 Ŵ 2 )Eα i.e., cj (t, t̄) Eα = Eα cj (t, t̄) for any α and j.
Therefore each cj (t, t̄) is a diagonal matrix.
By the differential equations (3.23),
∂ −1 −1 ∂ Ŵ 1 −1 −1 ∂ Ŵ 2
(Ŵ 1 Ŵ 2 ) = −Ŵ 1 Ŵ 1 Ŵ 2 + Ŵ 1
∂tα,n ∂tα,n ∂tα,n
−1 −1
= −Ŵ 1 (−(Ln U α )<0 )Ŵ 2 + Ŵ 1 (−(Ln U α )<0 )Ŵ 2 = 0,
and
∂ −1 −1 ∂ Ŵ 1 −1 −1 ∂ Ŵ 2
(Ŵ 1 Ŵ 2 ) = −Ŵ 1 Ŵ 1 Ŵ 2 + Ŵ 1
∂ t̄α,n ∂ t̄α,n ∂ t̄α,n
−1 −1
= Ŵ 1 B̄ α,n Ŵ 2 − Ŵ 1 B̄ α,n Ŵ 2 = 0,
which means that cj (t, t̄) are constant. Thus we have proved that the ambiguity of
Ŵ (t, t̄) is of the form
∞
X
Ŵ (t, t̄) 7→ Ŵ (t, t̄) cj e−j∂s , (3.25)
j=0

where each cj is a constant diagonal matrix. Because of the normalisation w0 (s, t, t̄) = 1N ,
c0 should be an identity matrix, 1N .
ˆ (t, t̄) is of the form
Similarly, the ambiguity of W̄
∞
ˆ (t, t̄) 7→ W̄
ˆ (t, t̄)
X
W̄ c̄j ej∂s ,
j=0

where each c̄j is a constant diagonal matrix and c̄0 is invertible.

Step 5. Now we restore the variables s and show that we can modify Ŵ (s, t, t̄) =
ˆ (s, t, t̄) = W̄
Ŵ (s(0) + s1, t, t̄) and W̄ ˆ (s(0) + s1, t, t̄) obtained above so that they satisfy
(3.12).

25
 First, note that any s ∈ ZN can be uniquely decomposed as s = (s1 , . . . , sN ) =
(0) (0) (0) (0)
s(0) + s1, where s(0) = (s1 , . . . , sN ) satisfies s1 + · · · + sN ∈ {0, . . . , N − 1} and s ∈ Z.
In fact, one has only to take s := the integer part of (s1 + . . . + sN )/N and s(0) := s − s1.
Thus any s is included in a uniquely defined sequence {s(0) + s1}s∈Z .
We have already obtained Ŵ (s(0) + s1, t, t̄) and W̄ ˆ (s(0) + s1, t, t̄) satisfying (3.8),
(3.9), (3.10) and (3.11) for each s(0) and s. As we have noted above, this means that
we have Ŵ (s) and W̄ ˆ (s) for each s. Let us denote these temporary wave operators as
temp ˆ temp (s, t, t̄) respectively.
Ŵ (s, t, t̄) and W̄
temp
Using the operator Ŵ (s, t, t̄), we modify the P a -operators (in particular, the
P α -operators) as follows:
temp temp
P̃ a (s) = (Ŵ (s + a))−1 P a (s)Ŵ (s). (3.26)
It follows from equations (2.24) that
P̃ a (s)e∂s = e∂s P̃ a (s), P̃ a (s)Eα , = Eα P̃ a (s) (3.27)
which means that the coefficients p̃a,j (s, t, t̄) in the expansion
∞
X
P̃ a (s) = p̃a,j (s, t, t̄)e−j∂s (3.28)
j=0

are diagonal matrices invariant under translation of the s-variable: s 7→ s + 1.

Differentiating (3.26) by tα,n , we have
∂ P̃ a (s)
= en∂s Eα P̃ a (s) − P̃ a (s)en∂s Eα = 0, (3.29)
∂tα,n
temp
due to the differential equations (3.10) for Ŵ (s), (2.33) for P a (s) and commutativity
(3.27). Similarly, the equation
∂ P̃ a (s)
=0 (3.30)
∂ t̄α,n
follows from (3.10), (2.34) and (3.27). Namely, P̃ a (s) does not depend on t and t̄. The
composition rule
P̃ b (s + a)P̃ a (s) = P̃ a+b (s) (3.31)
is a consequence of the corresponding formula (2.31).
Lemma 3.1. P̃ a+s1 (s(0) ) = P̃ a (s(0) ) for any a ∈ ZN , s ∈ Z and s(0) .

Proof. By definition (2.32), P 1 (s) = e∂s N

Q
α=1 Qα (s). Because of the condition (2.7), the
right-hand side is equal to e∂s L−1 (s). Therefore, by (3.26)
temp temp
P̃ 1 (s) = (Ŵ (s + 1))−1 P 1 (s)Ŵ (s)
temp temp
= (Ŵ (s + 1))−1 e∂s L−1 (s)Ŵ (s)
temp temp
= e∂s (Ŵ (s))−1 L−1 (s)Ŵ (s) = e∂s e−∂s = 1N .
The statement of the lemma follows from this equation and the composition rule (3.31).

26
 temp
Let us modify the operators Ŵ (s). The new wave operators Ŵ (s) are defined
by
temp temp
Ŵ (s) := Ŵ (s) P̃ s (0) = P s (0) Ŵ (0), (3.32)
where 0 = {0, . . . , 0}. (The second equality follows from the definition (3.26).) This
modification does not spoil equations (3.8) and (3.10), as P̃ s (0) is an operator of the form
(3.28) whose coefficients are diagonal matrices and constant with respect to t ((3.29)), t̄
((3.30)) and s (Lemma 3.1).
It follows from the definition (3.26) of P̃ a (s) and (2.31) that for b ∈ ZN
temp
Ŵ (s + b) = P s+b (0))Ŵ (0)
temp
= P b (s)P s (0)Ŵ (0) = P b (s)Ŵ (s).
In particular, for b = [1]α , we have
Ŵ (s + [1]α ) = P α (s)Ŵ (s),
which is the first equation in (3.13).
Thus we have obtained the desired operators Ŵ (s). The assertion about the unique-
temp
ness follows from that of Ŵ (s), (3.25), and (3.13).
ˆ (s) is proved in a similar way.
The existence of W̄

The converse statement (ii) follows immediately from definitions (3.14) and their
derivatives, once B α,n (s) and B̄ α,n (s) are expressed in terms of L(s), L̄(s), U α (s) and
Ū α (s) as in (2.13).
Both of the operators B (s) and B̄ (s) are expressed in two ways by Ŵ and W̄ ˆ
α,n α,n
because of (3.10) and (3.11) as follows:
ˆ (s)
∂ W̄
B α,n (s) = ˆ −1 (s) = Ŵ (s)en∂s E Ŵ −1 (s) + ∂ Ŵ (s) Ŵ −1 (s),
W̄ α
∂tα,n ∂tα,n

∂ Ŵ (s) −1 ˆ
B̄ α,n (s) = Ŵ (s) = W̄ ˆ −1 (s) + ∂ W̄ (s) W̄
ˆ (s)e−n∂s E W̄ ˆ −1 (s).
α
∂ t̄α,n ∂ t̄α,n
The first equality in the first equation means that B α,n (s) is a difference operator with
non-negative shifts. Hence the latter half of the same equation implies B α,n (s) =
−1 ˆ (s)e−n∂s E Ŵ −1 (s)) follows from
(Ŵ (s)en∂s Eα Ŵ (s))≥0 . Similarly, B̄ α,n (s) = (W̄ α <0
the second equation. Thus equations (2.13) are proved.
Remark 3.1. Comparing the coefficients of e−∂s in the expansion of the first equation of
(3.9) by using b̄0 (s) = w̃0 (s) w̃0−1(s − 1) (Proposition 2.2), we have
w̃0 (s) w̃0−1(s − 1) w̄0 (s − 1) = w̄0 (s), i.e., w̃0−1 (s − 1) w̄0 (s − 1) = w̃0−1 (s) w̄0 (s),
which means that the matrix c̄˜(s) := w̃0−1 (s) w̄0(s) is invariant under the shift s 7→
s + a1 (a ∈ Z). Similarly, it follows from the second equation in (3.9) and ūα,0 (s) =
w̃0 (s)Eα w̃0−1(s), (Proposition 2.2) that
w̃0−1 (s) w̄0(s) Eα = Eα w̃0−1 (s) w̄0 (s)

27
for any α = 1, . . . , N. Therefore c̄˜(s) is a diagonal matrix. The second equation of (3.13)
and p̄α,0 (s) = w̃0 (s + [1]α ) w̃0−1(s) (Proposition 2.2) implies
w̃0−1 (s + [1]α ) w̄0 (s + [1]α ) = w̃0−1(s) w̄0 (s).
Hence c̄˜(s) is invariant under any shift s 7→ s + a (a ∈ ZN ), i.e., it does not depend
on s. Thus we have shown that w̄0 (s) and w̃0 (s) are related as w̄0 (s) = w̃0 (s) c̄˜(t, t̄),
where c̄˜(t, t̄) is a diagonal matrix independent of s. However, it can depend on t and t̄
non-trivially. So, w̄0 (s) and w̃0 (s) are closely related but, strictly speaking, are not the
same. This is just the ambiguity in the definition of w̃0 (s) (see Remark 2.4). One may
also say that the freedom in the definition of w̃0 is partially fixed in the w̄0 . Actually,
once we have found w̄0 (s), we can replace w̃0 (s) in all equations in Section 2 with w̄0 (s).
This does not change those equations at all.
A useful corollary of Proposition 3.1 is the following statement.
Proposition 3.2. The following relations hold:
N
X
b̄0 (s) = w̄0 (s)w̄0−1 (s − 1) = ∂t̄1 w1 (s), ∂t̄1 ≡ ∂t̄µ,1 , (3.33)
µ=1

N
X
b1 (s) = ∂t1 w̄0 (s)w̄0−1(s), ∂t1 ≡ ∂tµ,1 . (3.34)
µ=1

Proof. These relations follow immediately from equations (3.10) and (3.11) for the wave
operators Ŵ , W̄ˆ . They are obtained from them by summing over α from 1 to N and
restriction to the highest coefficients.

3.2 Wave functions

The system of linear equations (3.3), (3.4), (3.5), (3.6) and (3.7) can be rewritten as a
linear system for the matrix wave functions defined by the wave matrices as
Ψ(s, t, t̄; z) = (Ψαβ (s, t, t̄; z))α,β := W (s, t, t̄) diagα (z sα )
∞
X
wj (s, t, t̄) diagα z sα −j eξ(tα ,z) ,


=
j=0
(3.35)
Ψ̄(s, t, t̄; z) = (Ψ̄αβ (s, t, t̄; z))α,β := W̄ (s, t, t̄) diagα (z sα )
∞
−1 )
X
w̄j (s, t, t̄) diagα z sα +j eξ(t̄α ,z


= .
j=0

In other words, the matrix elements of the wave functions are

∞
X
sβ ξ(tβ ,z)
Ψαβ (s, t, t̄; z) = wαβ (s, t, t̄; z) z e , wαβ (s, t, t̄; z) := wj,αβ (s, t, t̄) z −j ,
j=0
(3.36)
∞
,z −1 )
X
Ψ̄αβ (s, t, t̄; z) = w̄αβ (s, t, t̄; z) z sβ eξ(t̄β , w̄αβ (s, t, t̄; z) := w̄j,αβ (s, t, t̄) z j .
j=0

28
The linear systems for Ψ and Ψ̄ are as follows:

L(s)Ψ(s; z) = z Ψ(s; z), U α (s)Ψ(s; z) = Ψ(s; z)Eα , (3.37)

L̄(s)Ψ̄(s; z) = z −1 Ψ̄(s; z), Ū α (s)Ψ̄(s; z) = Ψ̄(s)Eα , (3.38)
∂Ψ(s; z) ∂Ψ(s; z)
= B α,n (s)Ψ(s; z), = B̄ α,n (s)Ψ(s; z), (3.39)
∂tα,n ∂ t̄α,n
∂ Ψ̄(s; z) ∂ Ψ̄(s; z)
= B α,n (s)Ψ̄(s; z), = B̄ α,n (s)Ψ̄(s; z), (3.40)
∂tα,n ∂ t̄α,n
and
Qα (s)Ψ(s; z) = Ψ(s; z) diagγ (z −δαγ ) Q̄α (s)Ψ̄(s; z) = Ψ̄(s; z) diagγ (z −δαγ ). (3.41)

The adjoint wave functions are defined by the adjoint actions of wave matrices6 :

∗
T
Ψ∗ (s, t, t̄; z) = (Ψ∗αβ (s, t, t̄; z))α,β = W −1 (s − 1, t, t̄) diagα (z −sα )
∞
X
diagα z −sα −j e−ξ(tα ,z) wj∗ (s, t, t̄),


=
j=0

∗ T (3.42)
∗ −1
Ψ̄ (s, t, t̄; z) = (Ψ̄∗αβ (s, t, t̄; z))α,β = W̄ (s − 1, t, t̄) diagα (z −sα
)
∞
−1 )
X
diagα z −sα +j e−ξ(t̄α ,z w̄j∗ (s, t, t̄).


=
j=0

ˆ as
Here we expand the inverses of Ŵ and W̄
∞
−1 X
Ŵ (s, t, t̄) = e−j∂s wj∗ (s + 1, t, t̄), w0∗(s, t, t̄) = 1N ,
j=0
(3.43)
∞
ˆ −1 (s, t, t̄) =
X
W̄ ej∂s w̄j∗ (s + 1, t, t̄), w̄0∗ (s, t, t̄) ∈ GL(N, C),
j=0

where
wj∗ (s, t, t̄) = (wj,αβ
∗
(s, t, t̄))α,β=1,...,N ∈ Mat(N × N, C),
(3.44)
w̄j∗ (s, t, t̄) = (w̄j,αβ
∗
(s, t, t̄))α,β=1,...,N ∈ Mat(N × N, C).
The matrix elements of the adjoint wave functions are
∞
X
Ψ∗αβ (s, t, t̄; z) = ∗
wαβ (s, t, t̄; z) z −sα e−ξ(tα ,z) , ∗
wαβ (s, t, t̄; z) := ∗
wj,αβ (s, t, t̄) z −j ,
j=0

∞
−1 )
X
Ψ̄∗αβ (s, t, t̄; z) = w̄αβ
∗
(s, t, t̄; z) z −sα e−ξ(t̄α ,z ∗
, w̄αβ (s, t, t̄; z) := ∗
w̄j,αβ (s, t, t̄) z j .
j=0
(3.45)
6
The formal adjoint operator A∗ of A = en∂sα ◦ a(s) is defined by A∗ = a(s)T e−n∂sα , where (·)T is
the transposed matrix.

29
4 The bilinear identity
The wave operators W and W̄ are characterized by a bilinear identity satisfied by the
wave functions and the adjoint wave functions.

Proposition 4.1. (i) The wave functions Ψ(s, t, t̄; z), Ψ̄(s, t, t̄; z) and the adjoint wave
∗
functions Ψ∗ (s, t, t̄; z), Ψ̄ (s, t, t̄; z) of the multi-component Toda lattice hierarchy satisfy
the following bilinear identity:
I I
∗ ′ ′ ′ ∗ ′
Ψ(s, t, t̄; z) Ψ (s , t , t̄ ; z) dz = Ψ̄(s, t, t̄; z) Ψ̄ (s′ , t′ , t̄ ; z) dz, (4.1)
C∞ C0

where C∞ is a circle around ∞ and C0 is a small circle around 0. This identity holds for
′
all s, s′ , t, t′ , t̄ and t̄ .
(ii) Conversely, assume that matrix-valued functions of the form
∞
X
wj (s, t, t̄) diagα z sα −j eξ(tα ,z) ,


Ψ(s, t, t̄; z) =
j=0

∞
−1 )
X
w̄j (s, t, t̄) diagα z sα +j eξ(t̄α ,z


Ψ̄(s, t, t̄; z) = ,
j=0
(4.2)
∞
X
Ψ∗ (s, t, t̄; z) = diagα z −sα −j e−ξ(tα ,z) wj∗ (s, t, t̄),


j=0

∞
∗ −1 )
X
diagα z −sα +j e−ξ(t̄α ,z w̄j∗ (s, t, t̄)


Ψ̄ (s, t, t̄; z) =
j=0

satisfy the bilinear identity (4.1). (In (4.2) wj , w̄j , wj∗ , w̄j∗ are N × N matrices and
w0 = w0∗ = 1N , w̄0 , w̄0∗ ∈ GL(N, C).) Then they are wave functions and adjoint wave
functions of the multi-component Toda lattice hierarchy. Namely, the functions wj , w̄j ,
wj∗ and w̄j∗ in (4.2) are the coefficients of the wave matrices and their inverse matrices
in the expansions (3.1) and (3.43) and the sextet (L, L̄, U α , Ū α , Qα , Q̄α )α=1,...,N defined
by (3.14) is a solution of the N-component Toda lattice hierarchy.

In terms of matrix elements the bilinear identity (4.1) acquires the form

N I
′ ′ ′
X
z sγ −sγ eξ(tγ −tγ ,z) wαγ (s, t, t̄; z) wγβ
∗
(s′ , t′ , t̄ ; z) dz
γ=1 C∞

N I
X ′ ′ −1 ) ′
= z sγ −sγ eξ(t̄γ −t̄γ ,z ∗
w̄αγ (s, t, t̄; z) w̄γβ (s′ , t′ , t̄ ; z) dz. (4.3)
γ=1 C0

Proof. (i) Assume that Ψ is a matrix wave function of the multi-component Toda lattice
hierarchy and Ψ∗ is a corresponding adjoint wave function.

30
 First, let us regard the left-hand side and the right-hand side of (4.1) as functions of
′
t and t̄ with parameters s, s′ , t′ and t̄ . Then, both of them (T = the right-hand side or
the left-hand side) satisfy the same linear system of differential equations,
∂ ∂
T (t, t̄) = B α,n T (t, t̄), T (t, t̄) = B̄ α,n T (t, t̄),
∂tα,n ∂ t̄α,n
because of (3.39) (for the left-hand side) and (3.40) (for the right-hand side). Since
this system is compatible by the Zakharov-Shabat equations (2.15), (2.16) and (2.17),
′
its solution T (t, t̄) is determined by the initial value, T (t′ , t̄ ). Hence, in order to prove
′
(4.1), i.e. (4.3), it is sufficient to prove the identity with t = t′ and t̄ = t̄ , namely,

N I
′
X
z sγ −sγ wαγ (s, t, t̄; z) wγβ
∗
(s′ , t, t̄; z) dz
γ=1 C∞

N I
′
X
= z sγ −sγ w̄αγ (s, t, t̄; z) w̄γβ
∗
(s′ , t, t̄; z) dz. (4.4)
γ=1 C0

Substituting the Taylor expansions (3.36) and (3.45) of wαγ (s, t, t̄; z), w̄γβ (s, t, t̄; z),
∗ ∗
wαγ (s, t, t̄; z) and w̄γβ (s, t, t̄; z), we have

N
X X
∗
wj,αγ (s, t, t̄) wk,γβ (s′ , t, t̄)
γ=1 j,k≥0
j+k=sγ −s′γ +1

N
X X
∗
= w̄j,αγ (s, t, t̄) w̄k,γβ (s′ , t, t̄). (4.5)
γ=1 j,k≥0
j+k=−sγ +s′γ −1

On the other hand, the (α, β)-element of the left hand side of the condition (2.14) is

N N
!
Y X  P −1

Qµ (s)aµ U ν (s)Laν (s) = Ŵ (s) e− δ aδ ∂sδ diagγ (eaγ ∂s ) Ŵ (s)
αβ
µ=1 ν=1 αβ

N X
∞
X P
= wj,αγ (s, t, t̄) e−j∂s e− δ aδ ∂sδ aγ ∂s
e e−k∂s wk,γβ
∗
(s + 1, t, t̄)
γ=1 j,k=0
 
N X
∞
X X ∗
 −(n−a )∂ − P a ∂
= 
 w j,αγ (s, t, t̄) w k,γβ (s − (n − aγ − 1)1 − a, t, t̄) e

γ s
e δ δ sδ
γ=1 n=0 j,k≥0
j+k=n
 
N ∞
X X  X
∗
 −n∂ − P a ∂
= 
 wj,αγ (s, t, t̄) wk,γβ (s − (n − 1)1 − a, t, t̄)
e
s
e δ δ sδ ,
γ=1 n=−aγ j,k≥0
j+k=n+aγ

31
where a = {a1 , . . . , aN } ∈ ZN . As there are no non-negative j and k such
X that j + k =
n + aγ , if n < −aγ , we may replace the sum over n in the last line by . In this way
n∈Z
we obtain

N N
!
Y X
Qµ (s)aµ U ν (s)Laν (s)
µ=1 ν=1 αβ
 
N
XX  X
∗
 −n∂ − P a ∂
= 
 wj,αγ (s, t, t̄) wk,γβ e
(s − (n − 1)1 − a, t, t̄) s
e δ δ sδ . (4.6)
n∈Z γ=1 j,k≥0
j+k=n+aγ

Similarly, the (α, β)-element of the right-hand side of (2.14) is

N N
!
−aν
Y X
Q̄µ (s)aµ Ū ν (s)L̄ (s)
µ=1 ν=1 αβ
 
N
XX  X
∗
 m∂ − P a ∂
= 
 w̄j,αγ (s, t, t̄) w̄k,γβ e e
(s + (m + 1)1 − a, t, t̄) s δ δ sδ . (4.7)
m∈Z γ=1 j,k≥0
j+k=m−aγ

Comparing the coefficients in (4.6) and (4.7), we have

N
X X
∗
wj,αγ (s, t, t̄) wk,γβ (s − (n − 1)1 − a, t, t̄)
γ=1 j,k≥0
j+k=n+aγ

N
X X
∗
= w̄j,αγ (s, t, t̄) w̄k,γβ (s − (n − 1)1 − a, t, t̄). (4.8)
γ=1 j,k≥0
j+k=−n−aγ

If we take n and a such that s′ = s − (n − 1)1 − a, then sγ − s′γ + 1 = n + aγ . Hence

the equation (4.8) gives the bilinear identity (4.5), which is equivalent to (4.4). Thus we
have proved the bilinear identity (4.3), i.e. (4.1).

(ii) Assume that the bilinear identity (4.1), which is equivalent to (4.3), holds. Setting
′
t′ = t, t̄ = t̄ and s′ = s + (1 − n)1 (n ∈ Z>0 ) in (4.3), we have

N I
X
z n−1 wαγ (s, t, t̄; z) wγβ
∗
(s + (1 − n)1, t, t̄; z) dz
γ=1 C∞

N I
′
X
= z n−1 w̄αγ (s, t, t̄; z) w̄γβ
∗
(s + (1 − n)1, t′ , t̄ ; z) dz.
γ=1 C∞

32
The integrand in the right-hand side is a series of z with non-negative powers. Hence we
obtain
XN X
∗
wj,αγ (s, t, t̄) wk,γβ (s + (1 − n)1, t, t̄) = 0. (4.9)
γ=1 j,k≥0
j+k=−n

On the other hand, the (α, β)-element of the product of matrix difference operators
∞ ∞
X
−j∂s ∗ X
Ŵ (s, t, t̄) = wj (s, t, t̄) e and Ŵ (s, t, t̄) = e−k∂s wk∗ (s + 1, t, t̄) is
j=0 k=0

 
∞
X N
X X
∗
 −n∂

 wj,αγ (s, t, t̄) wk,γβ (s + (1 − n)1, t, t̄)
e
s
. (4.10)
n=0 γ=1 j,k≥0
j+k=−n

Each coefficient of e−n∂s for non-zero n vanishes because of (4.9). Therefore the matrix
∗
difference operator Ŵ (s, t, t̄) Ŵ (s, t, t̄) acts just by multiplication by a matrix. Actu-
ally, since w0 = w0∗ = 1N by assumption, the product should be 1N . Namely, the inverse
of the operator Ŵ (s, t, t̄) defined as (3.1) from the coefficients of Ψ has the expression
as in (3.43).
′
If we set t′ = t, t̄ = t̄ and s′ = s + (1 + n)1 (n ∈ Z>0 ) in (4.3), then we obtain
N
X X
∗
0= w̄j,αγ (s, t, t̄) w̄k,γβ (s + (1 + n)1, t, t̄). (4.11)
γ=1 j,k≥0
j+k=n

by a similar argument as above.

The (α, β)-element of the product of matrix difference operators
∞
ˆ (s, t, t̄) =
X
W̄ w̄j (s, t, t̄) ej∂s
j=0

and ∞
∗ X
Ŵ (s, t, t̄) = ek∂s w̄k∗ (s + 1, t, t̄)
k=0

is  
∞
X N
X X
∗
 n∂

 w̄j,αγ (s, t, t̄) w̄k,γβ (s + (1 + n)1, t, t̄)
e .
s

n=0 γ=1 j,k≥0

j+k=n

This and (4.11) imply that the matrix difference operator W̄ ˆ (s, t, t̄) W̄
ˆ ∗ (s, t, t̄) is a
multiplication operator by a matrix, whose (α, β)-element is equal to
N
X
∗
w̄j,αγ (s, t, t̄) w̄k,γβ (s + 1, t, t̄)
γ=1

33
 ′
by the above computation. The bilinear identity (4.3) with t′ = t, t̄ = t̄ and s = s + 1
yields:
N
X N
X
∗ ∗
w0,αγ (s, t, t̄) w0,γβ (s + 1, t, t̄) = w̄0,αγ (s, t, t̄) w̄0,γβ (s + 1, t, t̄).
γ=1 γ=1

ˆ (s, t, t̄) W̄
Its right-hand side is the (α, β)-element of W̄ ˆ ∗ (s, t, t̄), as we have just shown
and its left-hand side is the (α, β)-element of the coefficient of e0 ∂s in (4.10). As (4.10)
−1 ˆ (s, t, t̄) W̄
ˆ ∗ (s, t, t̄) = 1 .
is nothing but Ŵ (s, t, t̄) Ŵ (s, t, t̄) = 1N , we obtain W̄ N
ˆ
Namely, the inverse of the operator W̄ (s, t, t̄) defined as (3.1) from the coefficients of Ψ̄
has the expression as in (3.43).

A similar consideration with s′ = s + (1 − n)1 (n ∈ Z) in the bilinear identity (4.3)

leads to the bilinear identity for matrix difference operators:
′ −1 ′
W (s, t, t̄) W −1 (s, t′ , t̄ ) = W̄ (s, t, t̄) W̄ (s, t′ , t̄ ). (4.12)
′
Differentiating this equation by tα,n and then setting t′ = t, t̄ = t̄, we have

∂ Ŵ −1 −1
ˆ
∂ W̄
(s, t, t̄)Ŵ (s, t, t̄) + Ŵ (s, t, t̄) e∂s Eα Ŵ (s, t, t̄) = ˆ −1 (s, t, t̄).
(s, t, t̄)W̄
∂tα,n ∂tα,n

The left-hand side is a sum of ek∂s (k ≤ n) with matrix coefficients and the right-hand
side is a sum of ek∂s (k ≥ 0) with matrix coefficients. Thus we obtain the first equations
in (3.10) and (3.11) by the standard argument.
The second equations in (3.10) and (3.11) are proved similarly by differentiating (4.12)
by t̄α,n .
The Lax equations (2.12) are direct consequences of the definitions (3.14) of L, L̄,
U α , Ū α , Qα , and Q̄α (α = 1, . . . , N) and the linear equations (3.10) and (3.11) for Ŵ
and W̄ˆ .

The algebraic conditions (2.5), (2.6), (2.7), (2.8), (2.9) and (2.10) are trivially satisfied
because of (3.14).
The remaining algebraic constraint (2.14) has already been shown to be equivalent to
′
the bilinear identity with s′ = s − (n − 1)1 − a (n ∈ Z), t′ = t and t̄ = t̄ in the proof of
the statement (i) of the proposition.
Hence the sextet (L, L̄, U α , Ū α , Qα , Q̄α )α=1,...,N is a solution of the multi-component
Toda lattice hierarchy.

5 The tau-function

5.1 Existence of the tau-function

The aim of this subsection is to show that the bilinear identity for the wave functions
for the multicomponent Toda lattice hierarchy implies existence of the tau-function. Let

34
Ψ(s, t, t̄; z), Ψ̄(s, t, t̄; z) be the N × N matrix wave functions for the N-component Toda
∗
hierarchy and Ψ∗ (s, t, t̄; z), Ψ̄ (s, t, t̄; z) the adjoint wave functions. The bilinear relation
for the wave functions has the form (4.1) (see Proposition 4.1). Changing the integration
variable z in the right-hand side as z → z −1 , we can write it in the form
N I N I
′ ′
X X
Ψαγ (s, t, t̄; z)Ψ∗γβ (s′ , t′ , t̄ ; z)dz = Ψ̄αγ (s, t, t̄; z −1 )Ψ̄∗γβ (s′ , t′ , t̄ ; z −1 )z −2 dz.
γ=1 C∞ γ=1 C∞
(5.1)
′ ′ ′
This identity is valid for all α, β, t, t , t̄, t̄ , s, s .
This subsection is devoted to the proof of the following theorem:

Theorem 5.1. The bilinear
identity ( 5.1) implies that there exists an N × N matrix-
valued function ταβ (s, t, t̄) α,β=1,...,N such that the diagonal elements ταα (s, t, t̄) =: τ (s, t, t̄)
are all the same and such that the wave functions and adjoint wave functions are expressed
through it as

ταβ (s, t − [z −1 ]β , t̄)

Ψαβ (s, t, t̄; z) = z sβ +δαβ −1 eξ(tβ ,z) ,
τ (s, t, t̄)

ταβ (s, t + [z −1 ]α , t̄)

Ψ∗αβ (s, t, t̄; z) = (−1)δαβ −1 z −sα +δαβ −1 e−ξ(tα ,z) ,
τ (s, t, t̄)
(5.2)
−1
ταβ (s + [1]β , t, t̄ − [z ]β )
Ψ̄αβ (s, t, t̄; z −1 ) = (−1)δαβ −1 z −sβ eξ(t̄β ,z) ,
τ (s, t, t̄)

ταβ (s − [1]α , t, t̄ + [z −1 ]α )
Ψ̄∗αβ (s, t, t̄; z −1 ) = z sα e−ξ(t̄α ,z) ,
τ (s, t, t̄)

where
z −j
t ± [z −1 ]γ


αj
= tα,j ± δαγ . (5.3)
j
and
s ± [1]α = {s1 , . . . , sα−1 , sα ± 1, sα+1 , . . . , sN }. (5.4)


The matrix-valued function ταβ (s, t, t̄) α,β=1,...,N
is called the tau-function of the N-
component Toda lattice hierarchy.
Now we proceed to the proof of Theorem 5.1. Let us represent the wave functions in
the form
Ψαβ (s, t, t̄; z) = z sβ eξ(tβ ,z) wαβ (s, t, t̄; z),

Ψ∗αβ (s, t, t̄; z) = z −sα e−ξ(tα ,z) wαβ

∗
(s, t, t̄; z),
(5.5)
−1 −sβ ξ(t̄β ,z) −1
Ψ̄αβ (s, t, t̄; z ) = z e w̄αβ (s, t, t̄; z ),

Ψ̄∗αβ (s, t, t̄; z −1 ) = z sα e−ξ(t̄α ,z)w̄αβ

∗
(s, t, t̄; z −1 ),

35
where all the w-functions are assumed to be regular as z → ∞. Substituting this into
(5.1), we write the bilinear identity for the wave functions in the form
N I
′ ′
X ′
z sγ −sγ eξ(tγ −tγ ,z) wαγ (s, t, t̄; z)wγβ
∗
(s′ , t′ , t̄ ; z)dz
γ=1 C∞

(5.6)
N I
X ′ ′ ′
= z sγ −sγ −2 eξ(t̄γ −t̄γ ,z) w̄αγ (s, t, t̄; z −1 )w̄γβ
∗
(s′ , t′ , t̄ ; z −1 )dz.
γ=1 C∞

The w-functions are normalized as

w(s, t, t̄; ∞) = w ∗ (s, t, t̄; ∞) = 1N ,

∗
i.e., their non-diagonal elements vanish and wαα (s, t, t̄; ∞) = wαα (s, t, t̄; ∞) = 1, α =
1, . . . , N by (3.1) and (3.43). Taking this into account, we introduce the functions w̃αβ ,
∗
w̃αβ by extracting the vanishing z-dependent factor explicitly7 :

wαβ (s, t, t̄; z) = z δαβ −1 w̃αβ (s, t, t̄; z), ∗

wαβ (s, t, t̄; z) = z δαβ −1 w̃αβ
∗
(s, t, t̄; z).

′
Let us first set s′ = s, t̄ = t̄ in (5.6). Then the right-hand side vanishes and we get
N I
′
X
z δαγ +δβγ −2 eξ(tγ −tγ ,z) w̃αγ (s, t, t̄; z)w̃γβ
∗
(s, t′ , t̄; z −1 )dz = 0. (5.7)
γ=1 C∞

To simplify the notation, we temporally will not write the arguments s, t̄ explicitly since
they are fixed in (5.7). Set t − t′ = [a−1 ]µ , then
′
 a δγµ
eξ(tγ −tγ ,z) = .
a−z
Putting µ = α or µ = β, we have from (5.7) for α 6= β:
a
I
z −1 ∗
w̃αα (t, z)w̃αβ (t − [a−1 ]α , z)dz
C∞ a − z
I (5.8)
+ z −1 w̃αβ (t, z)w̃ββ
∗
(t − [a−1 ]α , z)dz = 0,
C∞

a
I
z −1 ∗
w̃αβ (t, z)w̃ββ (t − [a−1 ]β , z)dz
C∞ a−z
I (5.9)
−1 ∗ −1
+ z w̃αα (t, z)w̃αβ (t − [a ]β , z)dz = 0
C∞

and
a
I
∗
dz w̃αα (t, z)w̃αα (t − [a−1 ]α , z)dz = 0 (5.10)
C∞ a−z
7
The functions w̃αβ = w̃αβ (s, t, t̄; z) introduced here should not be mixed with w̃0 from Proposi-
tion 2.2.

36
for β = α, where we write only non-vanishing terms of the sum over γ. The residue
calculus applied to (5.8), (5.9) and (5.10) yields8 , respectively:
∗
w̃αα (t, a)w̃αβ (t − [a−1 ]α , a) = −w̃αβ
∗
(t, ∞), (5.11)
∗
w̃αβ (t, a)w̃ββ (t − [a−1 ]β , a) = −w̃αβ
∗
(t − [a−1 ]β , ∞), (5.12)
∗
w̃αα (t, a)w̃αα (t − [a−1 ]α , a) = 1. (5.13)
Tending a → ∞ in (5.11), (5.12), we get
∗
w̃αβ (t, ∞) = −wαβ (t, ∞) for α 6= β. (5.14)

Using (5.13), we see from (5.11), (5.12) that

w̃αβ (t, a) = w̃αβ (t − [a−1 ]β , ∞)w̃ββ (t, a),

(5.15)
∗ w̃αβ (t + [a−1 ]α , ∞)
w̃αβ (t, a) =− .
w̃αα (t + [a−1 ]α , a)

Next, we set t − t′ = [a−1 ]α + [b−1 ]α in (5.7). At β = α the residue calculus yields

∗
w̃αα (t, a)w̃αα (t − [a−1 ]α − [b−1 ]α , a) = w̃αα (t, b)w̃αα
∗
(t − [a−1 ]α − [b−1 ]α , b).

Taking into account (5.13), we can write this relation in the form

w̃αα (t, a) w̃αα (t, b)

−1
= . (5.16)
w̃αα (t − [b ]α , a) w̃αα (t − [a−1 ]α , b)

As is proven in [10], it follows from this relation that there exists a function ταα (t) such
that
ταα (t − [a−1 ]α )
w̃αα (t, a) = wαα (t, a) = . (5.17)
ταα (t)
In the next step, we set t − t′ = [a−1 ]α + [b−1 ]β with α 6= β in (5.7). The residue calculus
yields
∗
w̃αα (t, a)w̃αβ (t − [a−1 ]α − [b−1 ]β , a) = −w̃αβ (t, b)w̃ββ
∗
(t − [a−1 ]α − [b−1 ]β , b),

or, taking into account (5.13), (5.15),

w̃αα (t, a) w̃ββ (t, b)

−1
= . (5.18)
w̃αα (t − [b ]β , a) w̃ββ (t − [a−1 ]α , b)

Note that at β = α we get (5.16).

Lemma 5.1. The condition ( 5.18) implies that we can choose ταα in ( 5.17) which does
not depend on the index α: ταα (t) =: τ (t).
8
When calculating the residues one should take into account that the point a lies outside of the
a
H
contour C∞ , so, shrinking the contour to infinity, we have C∞ z −1 a−z f (z)dz = +2πif (a) rather than
−2πif (a) (for functions f (z) regular in some neighborhood of infinity).

37
Proof. In the proof we follow [18]. Denote fα (t, z) = log wαα (t, z), then (5.18) acquires
the form
fα (t − [b−1 ]β , a) − fα (t, a) = fβ (t − [a−1 ]α , b) − fβ (t, b). (5.19)
Let us introduce the differential operator
X
∂α (z) = ∂z − z −k−1 ∂tα,k
k≥0

and apply ∂α (a) to the both sides of (5.19). We get

X
∂α (a)fα (t − [b−1 ]β , a) − ∂α (a)fα (t, a) = a−k−1 ∂tα,k fβ (t, b).
k≥0

Multiply both sides of this equality by ai and take the residue:

rα,i (t) = rα,i (t − [b−1 ]β ) − ∂tα,i fβ (t, b), (5.20)

where rα,i (t) = resz (z i ∂α (z)fα (t, z)) and the residue is defined as coefficient at z −1 .
Writing the same equality with indices γ, j instead of α, i, applying ∂tγ,j to the former
and ∂tα,i to the latter and subtracting one from the other, we get:

∂tγ,j rα,i (t − [b−1 ]β ) − ∂tα,i rγ,j (t − [b−1 ]β ) = ∂tγ,j rα,i (t) − ∂tα,i rγ,j (t).

This means that ∂tγ,j rα,i (t) − ∂tα,i rγ,j (t) = constant by the following lemma.
Lemma 5.2. A function G of t = {t1 , t2 , t3 , . . .} obeying the relation G(t − [a−1 ]) = G(t)
identically in a does not depend on t.

Proof. In the proof we follow [10]. We have:

 X 
−k ˜
a−k pk (−∂˜t )G(t),
X
−1
G(t − [a ]) = exp − a ∂k G(t) =
k≥1 k≥0

where the polynomials pk (x) (x = {x1 , x2 , . . .} =: (xn )n ) are defined by the generating
function X  X
exp xk a−k = pk (x)a−k
k≥1 k≥0

and ∂˜t = (∂˜n )n = (n−1 ∂tn )n . (The polynomials pk (x) are the Schur polynomials associ-
ated with one-row Young diagrams, see for example [10], §§2.3–2.4.) Note that p0 (x) = 1.
Therefore, we have the condition

a−k pk (−∂˜t )G(t) = 0

X

k≥1

for all a, hence

pk (−∂˜t )G(t) = 0 for all k ≥ 1. (5.21)
It follows from this condition that ∂tk G(t) = 0 for all k ≥ 1. This can be shown by
induction. At k = 1 we have ∂t1 G(t) = 0 because p1 (x) = x1 . Assume that (5.21) is true
for k = 1, . . . , k0 . Note that

pk0 +1 (x) = xk0 +1 + (polynomial of x1 , . . . , xk0 ).

38
Hence equation (5.21) for k = k0 + 1 has the form
 1 
∂k +1 + (polynomial of ∂t1 , . . . , ∂tk0 ) G(t) = 0.
k0 + 1 0
By the induction assumption only the first term in the left-hand side survives and gives
∂tk0 +1 G(t) = 0.

From the definition of rα,i it follows that the constant ∂tγ,j rα,i (t) − ∂tα,i rγ,j (t) is zero.
Therefore,
∂tγ,j rα,i (t) = ∂tα,i rγ,j (t),
which implies the existence of a function τ (t) such that rα,i (t) = ∂tα,i log τ (t). From
(5.20) we then see that
 
∂tα,i fβ (t, b) = ∂tα,i log τ (t − [b−1 ]β ) − log τ (t) .

Integrating, we get
τ (t − [z −1 ]α )
wαα (t, z) = c(z) ,
τ (t)
where the constant c(z) can be eliminated by multiplying the tau-function by exponent
of a linear form in the times.

As it follows from the lemma, we can write (5.17) in the form

τ (t − [a−1 ]α )
w̃αα (t, a) = , (5.22)
τ (t)
where the function τ (t) already does not depend on the index α. Plugging (5.22) into
(5.15), we get the equations
τ (t − [a−1 ]β )
w̃αβ (t, a) = w̃αβ (t − [a−1 ]β , ∞) ,
τ (t)
(5.23)
−1
∗ τ (t + [a ]α )
w̃αβ (t, a) = (−1)δαβ −1 w̃αβ (t + [a−1 ]α ) .
τ (t)
In the original notation, they read:
τ (s, t − [a−1 ]β , t̄)
w̃αβ (s, t, t̄; a) = w̃αβ (s, t − [a−1 ]β , t̄; ∞) ,
τ (s, t, t̄)
(5.24)
−1
∗ τ (s, t + [a ]α , t̄)
w̃αβ (s, t, t̄; a) = (−1)δαβ −1 w̃αβ (s, t + [a−1 ]α , t̄; ∞) ,
τ (s, t, t̄)
These formulae are of the form (5.2), i.e.,
ταβ (s, t − [a−1 ]β , t̄)
w̃αβ (s, t, t̄; a) = ,
τ (s, t, t̄)
(5.25)
∗ ταβ (s, t + [a−1 ]α , t̄)
w̃αβ (s, t, t̄; a) = (−1)δαβ −1
τ (s, t, t̄)

39
with
ταβ (s, t, t̄) = w̃αβ (s, t, t̄, ∞)τ (s, t, t̄). (5.26)


Note that the matrix-valued function ταβ (s, t, t̄) α,β=1,...,N is defined up to multiplication
by an arbitrary scalar function of s and t̄.
Let us now put s′ = s + [1]α + [1]β , t′ = t in (5.6). In this case the left-hand side of
(5.6) vanishes and we arrive at the relation
N I
X ′ ′
z δαγ +δβγ −2 eξ(t̄γ −t̄γ ,z) w̄αγ (s, t, t̄; z −1 )w̄γβ
∗
(s + [1]α + [1]β , t, t̄ ; z −1 )dz = 0. (5.27)
γ=1 C∞

The form of this equation is very similar to (5.7). The only difference is the normalization
of the w̄-functions. To make equation (5.27) closer to (5.7), we rewrite it in an equivalent
form:
N I ∗ ′
X − [1]α , t, t̄; z −1 ) w̄γβ
δαγ +δβγ −2 ξ(t̄γ −t̄′γ ,z) w̄αγ (s
(s + [1]β , t, t̄ ; z −1 )
z e ∗ ′ dz = 0, (5.28)
γ=1 C∞ w̄αα (s − [1]α , t, t̄; 0) w̄ββ (s + [1]β , t, t̄ ; 0)

in which it can be analyzed in the same way as (5.7). (The functions

w̄αα (s − [1]α , t, t̄; z −1 )

˜αα (s, t, t̄; z −1 ) =
w̄ ,
w̄αα (s − [1]α , t, t̄; 0)
∗ ′
∗ ′ w̄ββ (s + [1]β , t, t̄ ; z −1 )
w̄˜ββ (s, t, t̄ ; z −1 ) = ∗ ′
w̄ββ (s + [1]β , t, t̄ ; 0)
are normalized in the same way as the diagonal elements of the w̃-functions in (5.7).)
The conclusion is that there exists a function τ̄ (s, t, t̄) such that

w̄αβ (s, t, t̄; a−1 ) τ̄αβ (s + [1]α , t, t̄ − [a−1 ]β )

= ,
w̄αα (s, t, t̄; 0) τ̄ (s + [1]α , t, t̄)
(5.29)
∗
w̄αβ (s, t, t̄; a−1 ) τ̄αβ (s − [1]β , t, t̄ + [a−1 ]α )
= (−1)δαβ −1
w̄ββ (s, t, t̄; 0) τ̄ (s − [1]β , t, t̄)

with
w̄αβ (s − [1]α , t, t̄; 0)
τ̄αβ (s, t, t̄) = τ̄ (s, t, t̄). (5.30)
w̄αβ (s − [1]α , t, t̄; 0)
These equations do not yet fix the dependence of τ̄ on s. We can fix this dependence by
setting
τ̄ (s + [1]α , t, t̄)
w̄αα (s, t, t̄; 0) = ,
τ̄ (s, t, t̄)
(5.31)
∗ τ̄ (s − [1] α , t, t̄)
w̄αα (s, t, t̄; 0) = .
τ̄ (s, t, t̄)
The consistency of this definition follows from the relation
∗
w̄αα (s, t, t̄; 0)w̄αα (s + [1]α , t, t̄; 0) = 1 (5.32)

40
 ′
which is a corollary of (5.6) (one should put β = α, t′ = t, t̄ = t̄, s′ = s + [1]α and

calcu-
late the residues at infinity). Note that the matrix-valued function τ̄αβ (s, t, t̄) α,β=1,...,N
is defined up to multiplication by an arbitrary scalar function of t.
It remains to connect the functions τ and τ̄ . To this end, we set s′ = s + [1]β ,
′
t − t′ = [a−1 ]α , t̄ − t̄ = [b−1 ]β in (5.6). Calculating the residues, we obtain:
∗
w̃αα (s, t, t̄, a)w̃αβ (s + [1]β , t − [a−1 ]α , t̄ − [b−1 ]β , a)
(5.33)
−1 ∗ −1 −1 −1
= w̄αβ (s, t, t̄, b )w̄ββ (s + [1]β , t − [a ]α , t̄ − [b ]β , b )

or, in terms of the tau-functions,

τ (s, t − [a−1 ]α , t̄) ταβ (s + [1]β , t, t̄ − [b−1 ]β )
τ (s, t, t̄) τ (s + [1]β , t − [a−1 ]α , t̄ − [b−1 ]β )
(5.34)
−1 −1
τ̄αβ (s + [1]α , t, t̄ − [b ]β ) τ̄ (s, t − [a ]α , t̄)
= (−1)δαβ −1 .
τ̄ (s, t, t̄) τ̄ (s + [1]β , t − [a−1 ]α , t̄ − [b−1 ]β )
Putting here a = b = ∞, we arrive at the relation
τ̄αβ (s + [1]αβ , t, t̄) τ̄ (s, t, t̄)
(−1)δαβ −1 = =: f (s, t, t̄), (5.35)
ταβ (s, t, t̄) τ (s, t, t̄)
where [1]αβ = [1]α −[1]β and a function f is the same for all α, β. In terms of the function
f relation (5.34) acquires the form
f (s + [1]β , t, t̄ − [b−1 ]β )f (s, t − [a−1 ]α , t̄)
= 1. (5.36)
f (s + [1]β , t − [a−1 ]α , t̄ − [b−1 ]β )f (s, t, t̄)
Putting here b = ∞, we have the relation
f (s + [1]β , t, t̄) f (s + [1]β , t − [a−1 ]α , t̄)
= =: gβ (s, t, t̄) (5.37)
f (s, t, t̄) f (s, t − [a−1 ]α , t̄)
which means the following condition for the function gβ (s, t, t̄):

gβ (s, t − [a−1 ]α , t̄) = gβ (s, t, t̄) for all α and a. (5.38)

The condition (5.38) makes it possible to apply Lemma 5.2 to the function gβ (s, t, t̄),
namely, gβ (s, t, t̄) does not depend on t.
Substituting equation (5.37) back to (5.36), we get:

f (s, t, t̄ − [b−1 ]β )f (s, t − [a−1 ]α , t̄)

= 1. (5.39)
f (s, t − [a−1 ]α , t̄ − [b−1 ]β )f (s, t, t̄)

Lemma 5.3. If f (s, t, t̄) satisfies ( 5.39) for any a, b, α, β, then there exist functions
F (s, t), F̄ (s, t̄) such that f (s, t, t̄) = F (s, t)F̄ (s, t̄).

Proof. Since the following argument does not depend on the number of components, we
take N = 1 in this proof for simplicity of the notation. We also omit the argument s of
the functions since it is the same for all of them.

41
 Taking logarithm of (5.39), we have:

log f (t − [a−1 ], t̄) + log f (t, t̄ − [b−1 ])

− log f (t, t̄) − log f (t − [a−1 ], t̄ − [b−1 ]) = 0. (5.40)

Let us expand the different terms in this equation in a series. For example, the first term
in this equation is expressed as
P∞ ∞
a−i ∂˜i
a−k pk (−∂˜t ) log f (t, t̄),
X
log f (t − [a−1 ], t̄) = e− i=1 log f (t, t̄) =
k=0

where the polynomials pk (x) (the Schur polynomials) and the symbol ∂˜t are defined in
the proof of Lemma 5.2. Hence, (5.40) is expanded as
∞ X
∞
a−k b−l pk (−∂˜t ) pl (−∂˜t̄ ) log f (t, t̄) = 0,
X

k=1 l=1

which means that

pk (−∂˜t ) pl (−∂˜t̄ ) log f (t, t̄) = 0 (5.41)
for all k ≥ 1, l ≥ 1.
From this condition we will show that
∂ ∂
log f = 0 (5.42)
∂tk ∂ t̄l
for any k ≥ 1, l ≥ 1 by induction as follows. The equation (5.41) for k = l = 1 is
∂t1 ∂t̄1 log f = 0, as p1 (x) = x1 . Assume that (5.42) is true for k = 1, l = 1, . . . , l0 . Note
that
pl0 +1 (x) = xl0 +1 + (polynomial of x1 , . . . , xl0 ).
Hence, equation (5.41) for k = 1, l = l0 + 1 has the form
  
∂ 1 ∂ ∂ ∂
+ polynomial of ,..., = 0.
∂t1 l0 + 1 ∂ t̄l0 +1 ∂ t̄1 ∂ t̄l0

By the induction assumption only the first term in the left hand side survives and gives
∂t1 ∂t̄l0 +1 log f = 0. Thus we have (5.42) for k = 1 and all l. Fixing l and applying the
induction for k in the same way, we can prove (5.42) for any k.
Assume that log f (t, t̄) is expanded into a power series:
X n̄
log f (t, t̄) = cnn̄ tn t̄ .
n,n̄∈{0,1,2,... }∞

Then equation (5.42) for k and l means

X n̄−e
nk n̄l cnn̄ tn−ek t̄ l = 0,
n,n̄∈{0,1,2,... }∞

42
where ek = (δnk )n . Hence cnn̄ = 0 unless either nk or n̄l is zero. Therefore, cnn̄ = 0
unless n = 0 or n̄ = 0:
X X n̄
log f (t, t̄) = c00 + cn0 tn + c0n̄ t̄
n∈{0,1,2,... }∞ n̄∈{0,1,2,... }∞

= log f (t, 0) + log f (0, t̄) − log f (0, 0),

so the factorization of f follows by setting, for example, F (t) = f (t, 0)/f (0, 0) and
F̄ (t̄) = f (0, t̄).

Recall that we have gβ (s, t, t̄) = gβ (s, t̄). According to Lemma 5.3, we can write

F (s + [1]β , t) F̄ (s, t̄)gβ (s, t̄)

=
F (s, t) F̄ (s + [1]β , t̄)

where F is the function introduced by Lemma 5.3. Since the right-hand side does not
depend on t, the same must be true for the left-hand side. Let us denote the function
in the left-hand side by hβ (s). Then we obtain a recursion relation F (s + [1]β , t) =
hβ (s)F (s, t). Therefore, F (s, t) can be expressed as a product of several factors of the
form hβ (s′ ) (they depend on s) and F (0, t) for any s. Thus we conclude that the function
F factorizes into a product of a function of s and a function of t. Since the function
of s can be included into F̄ (s, t̄), we are free to consider F as a function only of t.
Therefore, f (s, t, t̄) can be represented as a product of a function of t and a function of
s, t̄: f (s, t, t̄) = F (t)F̄ (s, t̄). Thus the functions τ and τ̄ are connected by the relation

τ̄ (s, t, t̄) = F (t)F̄ (s, t̄)τ (s, t, t̄).

However, the possible factors F (t)F̄ (s, t̄) just reflect the freedom in the choice of the
functions τ and τ̄ mentioned above. That is why we can put F (t) = F̄ (s, t̄) = 1 without
loss of generality. Then from (5.35) we see that the relation between the tau-functions
looks as follows:
τ̄αβ (s, t, t̄) = (−1)δαβ −1 ταβ (s + [1]βα , t, t̄). (5.43)
Therefore, we have:

−1 δαβ −1 ταβ (s + [1]β , t, t̄ − [z −1 ]β )

w̄αβ (s, t, t̄; z ) = (−1) ,
τ (s, t, t̄)
(5.44)
∗ ταβ (s − [1]α , t, t̄ + [z −1 ]α )
w̄αβ (s, t, t̄; z −1 ) = ,
τ (s, t, t̄)

which agrees with (5.2), (5.5). This concludes the proof of existence of the tau-function
for the multi-component Toda hierarchy.

Theorem 5.2. The tau-function of the multi-component Toda lattice hierarchy satisfies

43
the matrix bilinear equation
N I
′ ′ ′
X
δβγ
(−1) z sγ −sγ +δαγ +δβγ −2 eξ(tγ −tγ ,z) ταγ (s, t − [z −1 ]γ , t̄)τγβ (s′ , t′ + [z −1 ]γ , t̄ ) dz
γ=1 C∞

N I
X ′ ′
δαγ
= (−1) z sγ −sγ −2 eξ(t̄γ −t̄γ ,z)
γ=1 C∞

′
× ταγ (s + [1]γ , t, t̄ − [z −1 ]γ )τγβ (s′ − [1]γ , t′ , t̄ + [z −1 ]γ ) dz
(5.45)
′
valid for all s, t, t̄, s′ , t′ , t̄ .

Proof. This equation is obtained by substitution of the relations (5.2) into the bilinear
identity (5.1) for the wave functions.
N
X X 
The simplest solution of equation (5.45) is ταβ (s, t, t̄) = δαβ exp ktγ,k t̄γ,k .
k≥1 γ=1

5.2 Tau-function as a universal dependent variable

The tau-function plays the role of the universal dependent variable of the hierarchy
meaning that the coefficients of the Lax operators and of the operators U α , Ū α , P α , P̄ α
can be expressed through the tau-function. Indeed, the coefficients of the wave functions
(and of the wave operators) can be expressed through the tau-function by expanding
the formulae (5.2) in inverse powers of z, then the coefficients of all the operators that
participate in the Lax formalism are obtained after “dressing” with the help of the wave
operators.
In particular, we have from (5.2):
ταβ (s + [1]β , t, t̄)
(w̄0 (s, t, t̄))αβ = (−1)δαβ −1 . (5.46)
τ (s, t, t̄)
It can be also deduced from (5.2) that the inverse matrix is expressed through the tau-
function as follows:
ταβ (s + 1 − [1]α , t, t̄)
(w̄0−1(s, t, t̄))αβ = . (5.47)
τ (s + 1, t, t̄)
′
To see this, let us choose s′ = s + 1, t′ = t, t̄ = t̄ in (5.45), then it becomes the identity
N
X
(−1)δαγ −1 ταγ (s + [1]γ , t, t̄)τγβ (s + 1 − [1]γ , t, t̄) = δαβ τ (s, t, t̄)τ (s + 1, t, t̄)
γ=1

which means that for solutions of the Toda lattice hierarchy the right-hand side of (5.47)
is indeed the inverse of (5.46). Therefore, the leading coefficient b̄0 (s) of L̄(s) is expressed
through the tau-function by the formula
N
X ταγ (s + [1]γ , t, t̄)τγβ (s − [1]γ , t, t̄)
(b̄0 (s, t, t̄))αβ = (−1)δαγ −1 . (5.48)
γ=1
τ 2 (s, t, t̄)

44
 −1
From the “dressing relation” L(s) = Ŵ (s)e∂s Ŵ (s) we have

L(s) = 1N e∂s + w1 (s) + w1∗ (s + 1) + . . . ,

whence the coefficient b1 (s) is given by

b1 (s) = w1 (s) + w1∗(s + 1) = w1 (s) − w1 (s + 1)

−1
since from Ŵ (s)Ŵ (s) = 1N it follows that w1∗ (s) = −w1 (s). The coefficient w1 (s)
can be found by expanding the first equation in (5.2) in inverse powers of z up to z −1 :

ταβ (s, t, t̄)  

(w1 (s, t, t̄))αβ = − δαβ ∂tα,1 log τ (s, t, t̄) + 1 . (5.49)
τ (s, t, t̄)

For b1 (s) we then obtain:

τ (s + 1, t, t̄) ταβ (s, t, t̄) ταβ (s + 1, t, t̄)

(b1 (s, t, t̄))αβ = δαβ ∂tα,1 log + − . (5.50)
τ (s, t, t̄) τ (s, t, t̄) τ (s + 1, t, t̄)
In the diagonal elements the last two terms cancel and we are left with the expression
similar to the one known in the one-component case.
Equations (3.33), (3.34) from Proposition 3.2 provide alternative expressions for the
b̄0 (s) and b1 (s) which are often more convenient than the ones obtained above. It is
instructive to give here another proof of Proposition 3.2 which is based on the bilinear
equation (5.45).
To show (3.33), we differentiate both sides of (5.45) with respect to t̄1 (i.e., apply the
N
′
X
differential operator ∂t̄µ,1 ) and put s′ = s, t′ = t, t̄ = t̄ after that. In this way we
µ=1
get the relations
X
(−1)δαγ −1 ταγ (s + [1]γ )τγβ (s − [1]γ ) = ∂t̄1 ταβ (s)τ (s) − ∂t̄1 τ (s)ταβ (s)
γ

for α 6= β and
X
(−1)δαγ −1 ταγ (s + [1]γ )τγα (s − [1]γ ) = ∂t̄1 τ (s) ∂tα,1 τ (s) − ∂t̄1 ∂tα,1 τ (s) τ (s)
γ

for β = α (we do not indicate the dependence on t, t̄ explicitly). Together with (5.46),
(5.47) and (5.49) they are equivalent to (3.33).
To show (3.34), we express its right-hand side through the tau-function:
 
X
−1
X
δαγ −1 ταγ (s + [1]γ ) τγβ (s + 1 − [1]γ )
(∂t1 w̄0 (s))αγ (w̄0 (s))γβ = (−1) ∂t1
γ γ
τ (s) τ (s + 1)

X ∂t1 ταγ (s + [1]γ ) τγβ (s + 1 − [1]γ ) ∂t τ (s)

= (−1)δαγ −1 − δαβ 1 .
γ
τ (s)τ (s + 1) τ (s)
(5.51)

45
To transform the sum, we differentiate both sides of (5.45) with respect to t1 (i.e., apply
N
′
X
the differential operator ∂tµ,1 ) and put s′ = s + 1, t′ = t, t̄ = t̄ after that. We get
µ=1
the relations
X
(−1)δαγ −1 ∂t1 ταγ (s + [1]γ ) τγβ (s + 1 − [1]γ ) = ταβ (s)τ (s + 1) − τ (s)ταβ (s + 1)
γ

for α 6= β and
X
(−1)δαγ −1 ταγ (s + [1]γ )τγα (s + 1 − [1]γ )
γ

= ∂t̄α,1 τ (s + 1) τ (s) − ∂t̄α,1 τ (s) τ (s + 1) + ∂t1 τ (s) τ (s + 1)

for β = α. Together with (5.50) they mean that the right-hand side of (5.51) is indeed
equal to b1 (s).
Other coefficients of the operators which participate in the Lax formalism can be
found in a similar way. However, for higher coefficients the explicit formulae become
rather bulky.

5.3 Bilinear equations of the Hirota-Miwa type

In this subsection we derive some bilinear equations for the tau-function which follow
′
from the generating bilinear equation (5.45). Choosing s − s′ , t − t′ , t̄ − t̄ in some
special ways, one can obtain corollaries of (5.45) which are known as equations of the
Hirota-Miwa type.
′
First of all we note that if s′ ≤ s, which means s′γ ≤ sγ for all γ, and t̄ = t̄,
then the right-hand side of (5.45) vanishes while the left-hand side becomes the integral
bilinear equation for the tau-function of the multi-component modified KP hierarchy
in the independent variables t. In this way the latter is realized as a subhierarchy of
the multi-component Toda lattice. Some bilinear equations which are corollaries of the
integral one are listed in [8, 19].
′
We proceed with the choice β = α, s′ = s, t − t′ = [a−1 ]α , t̄ − t̄ = [b−1 ]α . In this case
(5.45) takes the form
a
I
dz τ (s, t − [z −1 ]α , t̄)τ (s, t − [a−1 ]α + [z −1 ]α , t̄ − [b−1 ]α )
C∞ a − z
b
I
= dzz −2 τ (s + [1]α , t, t̄ − [z −1 ]α )τ (s − [1]α , t − [a−1 ]α , t̄ − [b−1 ]α + [z −1 ]α ),
C∞ b − z
where we write only non-vanishing terms in the sum over γ. The integrals can be cal-
culated by means of taking residues at the poles outside the contour C∞ , including the
non-zero residue at infinity in the left-hand side. As a result, after the shifts t → t+[a−1 ]α ,
t̄ → t̄ + [b−1 ]α we obtain the equation
τ (s, t, t̄ + [b−1 ]α )τ (s, t + [a−1 ]α , t̄) − τ (s, t, t̄)τ (s, t + [a−1 ]α , t̄ + [b−1 ]α )
(5.52)
−1 −1 −1
= (ab) τ (s + [1]α , t + [a ]α , t̄)τ (s − [1]α , t, t̄ + [b ]α ),

46
which is the bilinear equation of the one-component Toda lattice for each α-th component.
The other equations given below mix different components. If β 6= α, we obtain in the
same way:
ταβ (s, t + [a−1 ]α , t̄ + [b−1 ]α )τ (s, t, t̄) − ταβ (s, t + [a−1 ]α , t̄)τ (s, t, t̄ + [b−1 ]α )
(5.53)
= b−1 ταβ (s − [1]α , t, t̄ + [b−1 ]α )τ (s + [1]α , t + [a−1 ]α , t̄) (β 6= α).
′
Our next choice is β 6= α, s′ = s, t − t′ = [a−1 ]α , t̄ − t̄ = [b−1 ]β . In this case the
residue calculus yields:
ταβ (s, t + [a−1 ]α , t̄)τ (s, t, t̄ + [b−1 ]β ) − ταβ (s, t + [a−1 ]α , t̄ + [b−1 ]β )τ (s, t, t̄)
(5.54)
−1 −1 −1
= b ταβ (s + [1]β , t + [a ]α , t̄)τ (s − [1]β , t, t̄ + [b ]β ) (β 6= α).
′
Next, we choose β 6= α, s′ = s + [1]α + [1]β , t − t′ = [a−1 ]α , t̄ − t̄ = [b−1 ]β and obtain
the equation
ταβ (s + [1]β , t + [a−1 ]α , t̄)τ (s + [1]α , t, t̄ + [b−1 ]β )

− ταβ (s + [1]β , t, t̄)τ (s + [1]α , t + [a−1 ]α , t̄ + [b−1 ]β )

= a−1 ταβ (s + [1]α + [1]β , t + [a−1 ]α , t̄)τ (s, t, t̄ + [b−1 ]β ) (β 6= α).

(5.55)
′ ′ −1 ′ −1
In a similar way, the choice β 6= α, s = s + [1]α , t − t = [a ]α , t̄ − t̄ = [b ]β leads to
the equation
a−1 ταβ (s + [1]α , t + [a−1 ]α , t̄)τ (s, t, t̄ + [b−1 ]β )

− ταβ (s, t + [a−1 ]α , t̄ + [b−1 ]β )τ (s + [1]α , t, t̄)

(5.56)
= b−1 ταβ (s + [1]β , t + [a−1 ]α , t̄)τ (s + [1]α − [1]β , t, t̄ + [b−1 ]β )

− ταβ (s, t, t̄)τ (s + [1]α , t + [a−1 ]α , t̄ + [b−1 ]β ) (β 6= α).

′
At last, we put β 6= α, s′ = s + [1]µ , t′ = t, t̄ = t̄, µ 6= α, β. The residue calculus
yields:
ταβ (s + [1]µ , t, t̄)τ (s, t, t̄) − ταβ (s, t, t̄)τ (s + [1]µ , t, t̄)
(5.57)
= ταµ (s + [1]µ , t, t̄)τµβ (s, t, t̄).

6 The multi-component Toda lattice from the uni-

versal hierarchy

6.1 The universal hierarchy

Let us start from the M-component KP hierarchy [5, 6, 7, 8] in which the additional
integrable flows are allowed to take arbitrary complex values. In [9] it is called the

47
universal hierarchy. The independent variables are M infinite sets of (in general complex)
“times”

t = {t1 , t2 , . . . , tM }, tα = {tα,1 , tα,2 , tα,3 , . . . }, α = 1, . . . , M

and M additional variables r1 , . . . , rM such that

M
X
rα = 0. (6.1)
α=1

We denote by r the set {r1 , . . . , rM } and use the already introduced notation

r + [1]α = {r1 , . . . , rα +1, . . . , rM }, r + [1]αβ = r + [1]α − [1]β . (6.2)

In general we treat r1 , . . . , rM as complex variables, as in [9]. If they are restricted to be

integers, the hierarchy coincides with the one considered in [5]–[8].
In the bilinear formalism, the dependent variable is the tau-function τ (r, t). The
universal hierarchy is the infinite set of bilinear equations for the tau-function which are
encoded in the basic bilinear relation
M I
′ ′
X
−1 ′
ǫαγ (r)ǫβγ (r ) dz z rγ −rγ +δαγ +δβγ −2 eξ(tγ −tγ , z)
C∞
γ=1 (6.3)

× τ r + [1]αγ , t − [z −1 ]γ τ r′ + [1]γβ , t′ + [z −1 ]γ = 0

valid for any α, β, t, t′ and r, r ′ such that r − r′ ∈ ZM (and subject to the constraint
(6.1)). In (6.3) 
 X 

 exp −iπ rµ , α<γ



 α<µ≤γ

ǫαγ (r) = 1, α=γ (6.4)


  X 
 − exp iπ rµ , α>γ



γ<µ≤α

(see [9]). The contour C∞ is a big circle around infinity. It is easy to see that the
equation (6.3) depends only on the differences rα − rα′ which are integers. Different
bilinear relations for the tau-function which follow from (6.3) for special choices of r − r ′
and t − t′ are given in [8].

6.2 Specification to the multi-component Toda lattice hierarchy

Let us show that M = 2N-component universal hierarchy contains the N-component
Toda hierarchy [4].
Let the set of 2N indices in the universal hierarchy be {1, 2, . . . , N, 1̄, 2̄, . . . , N̄}. We
identify the time variables tα of the Toda lattice with tα of the universal hierarchy and
t̄α with tᾱ for α = 1, . . . , N. Besides, we set rα = −rᾱ =: sα . In this section we consider
the s′α s as complex numbers such that sα − s′α are integers. By t we denote the set
of times t = {t1 , t2 , . . . , tN , t̄1 , t̄2 , . . . , t̄N } and by s we denote the 2N-component set

48
s = {s1 , s2 , . . . , sN , −s1 , −s2 , . . . , −sN }. Consider the tau-function τ (s + r, t) of the 2N-
component universal hierarchy, where the variables r are subject to the constraint (6.1).
The tau-function of the N-component Toda hierarchy is the N × N matrix with matrix
elements
ταγ (s, t, t̄) = τ (s + [1]αγ , t), α, γ = 1, . . . , N. (6.5)
It is easy to see that
τ (s + [1]αγ̄ , t) = ταγ (s + [1]γ , t, t̄),

τ (s + [1]ᾱγ , t) = ταγ (s − [1]α , t, t̄), (6.6)

τ (s + [1]ᾱγ̄ , t) = ταγ (s + [1]γα , t, t̄).

Taking into account (6.6), we can represent the bilinear equation (6.3) for α ∈
{1, . . . , N}, β ∈ {1̄, . . . , N̄} in the form
N I
′ ′
X
ǫαγ (s)ǫ−1
β̄γ
(s′ ) dz z sγ −sγ +δαγ −2 eξ(tγ −tγ ,z)
γ=1 C∞

′
×ταγ (s, t − [z −1 ]γ , t̄)τγβ (s′ + [1]β , t′ + [z −1 ]γ , t̄ )
(6.7)
N I
X ′ ′
−1 ′
+ ǫαγ̄ (s)ǫβ̄γ̄ (s ) dz z sγ −sγ +δβγ −2 eξ(t̄γ −t̄γ ,z)
γ=1 C∞

′
×ταγ (s + [1]γ , t, t̄ − [z −1 ]γ )τγβ (s′ + [1]βγ , t′ , t̄ + [z −1 ]γ ) = 0.
′
This equation holds for all t, t′ , t̄, t̄ , s, s′ such that s − s′ ∈ ZN .
Regarding the set {1, 2, . . . , N, 1̄, 2̄, . . . , N̄} as the ordered set, it is not difficult to
express the ǫ-factors as functions of s:
  X 

 exp −iπ sµ , α ≤ γ,

α<µ≤γ


ǫαγ (s) =   (6.8)
 X
 − exp iπ sµ , α > γ,



γ<µ≤α

  X 

 exp iπ sµ , α ≤ γ,

α<µ≤γ


ǫᾱγ̄ (s) =   (6.9)
 X
− exp −iπ s , α > γ,

µ



γ<µ≤α
  X X 

 exp −iπ s µ + iπ s µ , α ≤ γ,

γ<µ≤N 1≤µ≤α


ǫαγ̄ (s) =   (6.10)
 X X
 exp −iπ sµ + iπ sµ , α > γ,



α<µ≤N 1≤µ≤γ

49
   X X 

 − exp iπ sµ − iπ sµ , α ≤ γ,

γ<µ≤N 1≤µ≤α


ǫᾱγ (s) =   (6.11)
 X X
− exp iπ s − iπ s µ , α > γ.

µ



α<µ≤N 1≤µ≤γ

Proposition 6.1. Equation ( 6.7) is equivalent to the following equation:

N I
′ ′
X
ǫαγ ǫ0N (s)ǫβγ (s)ǫ−1 ′
βγ (s ) z sγ −sγ +δαγ −2 eξ(tγ −tγ ,z)
γ=1 C∞

′
×ταγ (s, t − [z −1 ]γ , t̄)τγβ (s′ + [1]β , t′ + [z −1 ]γ , t̄ ) dz
(6.12)
N I
X ′ ′
= ǫβγ ǫ0N (s′ )ǫβγ (s)ǫ−1 ′
βγ (s ) z sγ −sγ +δβγ −2 eξ(t̄γ −t̄γ ,z)
γ=1 C∞

′
×ταγ (s + [1]γ , t, t̄ − [z −1 ]γ )τγβ (s′ + [1]βγ , t′ , t̄ + [z −1 ]γ ) dz,

where ǫαγ are sign factors such that ǫαγ = 1 for α ≤ γ and ǫαγ = −1 for α > γ and
 X 
ǫ0N (s) = exp −iπ sµ .
1≤µ≤N

Clearly, the products of ǫ-factors here depend only on s − s′ and are just signs ±1 for
s − s′ ∈ ZN .

Proof. Equation (6.12) is obtained by plugging (6.8)–(6.11) into (6.7). Let us present
some details of the calculation which transforms the ǫ-factors in (6.7) to those in (6.12).
For brevity, we write the γ-th terms of the sums in (6.7) as
−1 ′
ǫαγ (s)ǫβ̄γ (s )Iαβγ + ǫαγ̄ (s)ǫ−1
β̄γ̄
(s′ )Iαβγ
′
.

Suppose first that α ≤ β ≤ γ. Plugging here (6.8)–(6.11), we represent this as

 X X X 
−ǫαβ exp −iπ sµ − iπ s′µ + iπ s′µ Iαβγ
α<µ≤γ γ<µ≤N 1≤µ≤β

 X X X 
+ǫβγ exp −iπ sµ + iπ sµ − iπ s′µ Iαβγ
′
,
γ<µ≤N 1≤µ≤α β<µ≤γ

or
 X X X X 
−ǫαβ exp −iπ sµ + iπ sµ + iπ (sµ − s′µ ) − iπ (sµ − s′µ ) Iαβγ
α<µ≤N 1≤µ≤β γ<µ≤N 1≤µ≤β

 X X X 
+ǫβγ exp −iπ sµ + iπ sµ + iπ (sµ − s′µ ) ′
Iαβγ .
β<µ≤N 1≤µ≤α β<µ≤γ
(6.13)

50
We note that
 X X   X X 
exp −iπ sµ + iπ sµ = exp −iπ sµ + iπ sµ =: Aαβ (s).
α<µ≤N 1≤µ≤β β<µ≤N 1≤µ≤α

Recall also that sµ − s′µ are integers, so (6.13) can be written as

h  X X 
Aαβ (s) −ǫαβ exp iπ (sµ − s′µ ) + iπ (sµ − s′µ ) Iαβγ
γ<µ≤N 1≤µ≤β

 X  i
+ ǫβγ exp iπ (sµ − s′µ ) Iαβγ
′
,
β<µ≤γ

which is
h i
Aαβ (s) −ǫαβ ǫ0N (s)ǫ−1
0N (s ′
)ǫβγ (s)ǫ−1 ′
βγ (s )I αβγ + ǫ ǫ
βγ βγ (s)ǫ−1 ′ ′
βγ (s )I αβγ .

Similar calculations show that in the other cases, α ≤ γ ≤ β and γ ≤ α ≤ β, one obtains
the same result. Therefore, Aαβ (s) is an inessential common multiplier and we arrive at
(6.12).

There is an equivalent form of equation (6.12) which is more symmetric with respect
to α, β:
N I
′ ′
X
−1 ′
ǫβγ ǫ0N (s)ǫαγ (s)ǫβγ (s ) z sγ −sγ +δαγ −2 eξ(tγ −tγ ,z)
γ=1 C∞

′
×ταγ (s, t − [z −1 ]γ , t̄)τγβ (s′ + [1]β , t′ + [z −1 ]γ , t̄ ) dz
(6.14)
N I
X ′ ′
= ǫαγ ǫαγ (s)ǫ−1 ′ ′
βγ (s )ǫ0N (s ) z sγ −sγ +δβγ −2 eξ(t̄γ −t̄γ ,z)
γ=1 C∞

′
×ταγ (s + [1]γ , t, t̄ − [z −1 ]γ )τγβ (s′ + [1]βγ , t′ , t̄ + [z −1 ]γ ) dz.

The bilinear equation (6.3) contains also three other types of equations which cor-
respond to the choices α, β ∈ {1, . . . , N}, α, β ∈ {1̄, . . . , N̄ } and α ∈ {1̄, . . . , N̄},
β ∈ {1, . . . , N}. A thorough analysis shows that they are equivalent to (6.14) and
can be obtained from it by shifting some of s- and s′ -variables by ±1. For example, the
equation which corresponds to the choice α, β ∈ {1, . . . , N} has the form
N I
′ ′
X
−1 ′
ǫαγ (s)ǫβγ (s ) z sγ −sγ +δαγ +δβγ −2 eξ(tγ −tγ ,z)
γ=1 C∞

′
×ταγ (s, t − [z −1 ]γ , t̄)τγβ (s′ , t′ + [z −1 ]γ , t̄ ) dz
(6.15)
N I
X ′ ′
= − ǫαγ ǫβγ ǫαγ (s)ǫ0N (s)ǫ−1 ′ −1 ′
βγ (s )ǫ0N (s ) z sγ −sγ −2 eξ(t̄γ −t̄γ ,z)
γ=1 C∞

′
×ταγ (s + [1]γ , t, t̄ − [z −1 ]γ )τγβ (s′ − [1]γ , t′ , t̄ + [z −1 ]γ ) dz.

51
 ′
It can be obtained from (6.14) by the substitution s′ → s′ − [1]β . If we set t̄ = t̄, the
right-hand side vanishes for s′µ ≤ sµ and we get the bilinear equation for the N-component
modified KP hierarchy given in [19].
Remark 6.1. For N = 1 equation (6.14) becomes the standard bilinear equation
I
′ ′ ′
z s−s −1 eξ(t−t ,z) τsToda (t − [z −1 ], t̄)τsToda ′
′ +1 (t + [z
−1
], t̄ ) dz
C∞
I (6.16)
′ ′
s′ −s−1
= z eξ(t̄−t̄ ,z) τs+1
Toda
(t, t̄ − [z −1 ])τsToda
′ (t′ , t̄ + [z −1 ]) dz
C∞

πi
s(s−1)
for the tau-function τsToda (t, t̄) = e 2 τ11 (s, t, t̄) of the Toda lattice hierarchy [4].

It remains to clarify how the tau-function from this section is related to the tau-
function introduced in section 5, because the bilinear equations for them look different.
This matter is clarified by the following proposition.
(sec.5) (sec.6)
Proposition 6.2. The tau-functions ταβ (s, t, t̄) and ταβ (s, t, t̄) differ by a simple
sign factor:
(sec.5) (sec.6)
ταβ (s, t, t̄) = (−1)|s|(|s|−1)/2 ǫαβ (s) ταβ (s, t, t̄), (6.17)
N
X
where |s| = sµ . After this substitution the bilinear equations ( 5.45) and ( 6.15) become
µ=1
identical.

The proof consists in a direct verification.

7 The multi-component Toda lattice hierarchy from

free fermions
In this section we show how the multi-component Toda lattice hierarchy can be obtained
in the framework of the free fermion technique developed by the Kyoto school. In this
approach the variables sα are assumed to be integers again, as before §6.

7.1 The multi-component fermions

The Clifford algebra A is generated by the creation-annihilation N-component free
fermionic operators:
D E
(α) ∗(α)
A := ψj , ψj α = 1, . . . , N, j ∈ Z .

These operators satisfy the standard anti-commutation relations:

(α) ∗(β) (α) (β) ∗(α) ∗(β)
[ψj , ψk ]+ = δαβ δjk , [ψj , ψk ]+ = [ψj , ψk ]+ = 0.

52
 (α) ∗(α)
The generating functions of ψj and ψj ,
X (α) X ∗(α) −j
ψ (α) (z) = ψj z j , ψ ∗(α) (z) = ψj z ,
j∈Z j∈Z

are called the free fermionic fields.

Remark 7.1. The algebra of N-component fermionic operators is in fact isomorphic to
the algebra of one-component operators ψj , ψj∗ . The isomorphism is established by the
(α) ∗(α) ∗
map ψj 7→ ψN j+α , ψj 7→ ψN j+α . See, for example, [11], §4. However, in practice it is
(α) ∗(α)
more convenient to deal with the operators ψj , ψj rather than ψj , ψj∗ .
The Fock space F and the dual Fock space F ∗ are generated as A-modules by the
vacuum state |0i and the dual vacuum state h0| that satisfy the conditions
(α) ∗(α)
ψj |0i = 0 (j < 0), ψj |0i = 0 (j ≥ 0),
(7.1)
(α) ∗(α)
h0| ψj = 0 (j ≥ 0), h0| ψj = 0 (j < 0),
(α) ∗(α) (α)
so ψj with j < 0 and ψj with j ≥ 0 are annihilation operators while ψj with j ≥ 0
∗(α)
and ψj with j < 0 are creation operators. In other words,

F = A |0i ∼
= A/AWann , |0i = 1 mod AWann , (7.2)
F ∗ = h0| A ∼
= Wcre A\A, h0| = 1 mod Wcre A, (7.3)

where
(α) ∗(α) (α) ∗(α)
M M M M
Wann := Cψj ⊕ Cψj , Wcre := Cψj ⊕ Cψj .
j<0,α j≥0,α j≥0,α j<0,α

A pairing F ∗ ⊗A F −→ C, hu| ⊗ |vi 7→ hu|vi which satisfies

(hu|a)|vi = hu|(|avi), (hu| ∈ F ∗ , a ∈ A, |vi ∈ F )

is determined by the normalization condition h0|0i = 1. To simplify the notation, we

denote h0|a|0i = hai for a ∈ A. In particular,
D E D E
(α) ∗(α)
h1i = 1, ψj = ψj = 0,
D E D E D E
(α) (β) ∗(α) ∗(β) (α) ∗(β)
ψj ψk = ψj ψk = 0, ψj ψk = δαβ δjk θ(j < 0),

where θ(P ) is the boolean characteristic function: θ(P ) = 1 when P is true and θ(P ) = 0
when P is false.
(α) ∗(β)
The normal ordered product of two fermion operators φ1 , φ2 (= ψj or ψk ) is
defined by
•
φ φ • = φ1 φ2 − hφ1 φ2 i .
• 1 2•
(7.4)
In other words, the normal ordering means moving annihilation operators to the right
and creation operators to the left, changing the sign any time when two fermion operators
are permuted.

53
 For an N-tuple of integers s = {s1 , s2 , . . . , sN } ∈ ZN , a vector |si in F and a vector
hs| in F ∗ are defined by

|si = Ψ∗(N ) ∗(2) ∗(1)

sN · · · Ψs2 Ψs1 |0i , hs| = h0| Ψ(1) (2) (N )
s 1 Ψs 2 · · · Ψs N ,

where
 (α) (α)
 ∗(α) ∗(α)
ψs−1 . . . ψ0
 (s > 0), ψ0 . . . ψs−1
 (s > 0),
Ψ∗(α)
s = 1 (s = 0), Ψ(α)
s = 1 (s = 0),
 ∗(α)
 ∗(α)  (α)
 (α)
ψs . . . ψ−1 (s < 0), ψ−1 . . . ψs (s < 0).
(β) ∗(β)
The commutation relations of these operators with ψj and ψj (α 6= β) are
(β) (β) (β) (β)
Ψ∗(α)
s ψj = (−1)s ψj Ψ∗(α)
s , Ψ(α)
s ψj = (−1)s ψj Ψ∗(α)
s ,
(7.5)
∗(β) ∗(β) ∗(β) ∗(β)
Ψ∗(α)
s ψj = (−1) s
ψj Ψ∗(α)
s , Ψ(α)
s ψj = (−1) s
ψj Ψ∗(α)
s .

Therefore, if α 6= β,

Ψ∗(α) ∗(β)
sα Ψsβ = (−1)
sα sβ ∗(β) ∗(α)
Ψs β Ψs α , Ψ(α) (β)
sα Ψsβ = (−1)
sα sβ (β) (α)
Ψs β Ψs α , (7.6)

and (cf. (7.1)) 

 0 (s′ ≥ sα ),
(α)
hs| ψs′ =
ǫα (s) hs − [1]α | (s′ = sα − 1),



 0 (s′ < sα ),
∗(α)
hs| ψs′ =
ǫα (s) hs + [1]α | (s′ = sα ),


 (7.7)
′
 0 (s < sα ),
(α)
ψs′ |si =
ǫα (s) |s + [1]α i (s′ = sα ),



 0 (s′ ≥ sα ),
∗(α)
ψs′ |si =
ǫα (s) |s − [1]α i (s′ = sα − 1),


where the notation s ± [1]α is introduced in (5.4) and the sign factor ǫα (s) is

ǫα (s) = (−1)sα+1 +···+sN . (7.8)

The current operator J (α) (z) is defined by

(α)
X
J (α) (z) = •• ψ (α) (z)ψ ∗(α) (z) •• = Jk z −k ,
k∈Z

which is a generating function of the operators

(α)
X (α) ∗(α)
Jk = •
ψ ψj+k •• .
• j
(7.9)
j∈Z

54
 (α)
(The normal ordering here is essential only for J0 .)
(α)
The commutation relations for the operators Jk are as follows:
(α) (β) (α) (α) ∗(β) ∗(α)
[Jk , ψj ] = δαβ ψj−k , [Jk , ψj ] = −δαβ ψj+k ,
(α) (β)
[Jk , Jl ] = kδαβ δk+l,0.

Hence,
(α) (α)
[Jk , ψ (β) (z)] = z k δαβ ψ (α) (z), [Jk , ψ ∗(β) (z)] = −z k δαβ ψ ∗(α) (z), (7.10)

and
(α) (α)
[Jk , Ψs(β) ] = [Jk , Ψs∗(β) ] = 0 (α 6= β). (7.11)
As a consequence of (7.7) and (7.9) we have
(α)
hs| J−n = 0, Jn(α) |si = 0 (n ≥ 1). (7.12)
(α)
The operators J0 =: Qα are charge operators:

hs| Qα = hs| sα , Qα |si = sα |si . (7.13)

In order to describe the multi-component Toda lattice in the fermionic language, we

introduce 2N infinite sets of independent continuous variables

t = {t1 , t2 , . . . , tN }, tα = {tα,1 , tα,2 , tα,3 , . . . },

t̄ = {t̄1 , t̄2 , . . . , t̄N }, t̄α = {t̄α,1 , t̄α,2 , t̄α,3 , . . . }

which will be the time variables of the Toda lattice as before. The evolution is induced
by the following operators:
N X N X
(α) ¯ t̄) = (α)
X X
J(t) = tα,k Jk , J( t̄α,k J−k . (7.14)
α=1 k≥1 α=1 k≥1

¯ t̄) are as follows:

The commutation relations for J(t) and J(

[J(t), ψ (α) (z)] = ξ(tα , z)ψ (α) (z), [J(t), ψ ∗(α) (z)] = −ξ(tα , z)ψ ∗(α) (z),
¯ t̄), ψ (α) (z)] = ξ(t̄α , z −1 )ψ (α) (z),
[J( ¯ t̄), ψ ∗(α) (z)] = −ξ(t̄α , z −1 )ψ ∗(α) (z).
[J(

Therefore,

eJ(t) ψ (α) (z) = eξ(tα ,z) ψ (α) (z) eJ(t) , eJ(t) ψ ∗(α) (z) = e−ξ(tα ,z) ψ ∗(α) (z) eJ(t) ,
¯ −1 ) ¯ ¯ −1 ) ¯
(7.15)
eJ(t̄) ψ (α) (z) = eξ(t̄α ,z ψ (α) (z) eJ(t̄) , eJ(t̄) ψ ∗(α) (z) = e−ξ(t̄α ,z ψ ∗(α) (z) eJ(t̄) .

As is shown in [10] (equation (2.6.5)) and [11] (equation (1.21)), we have the following
formulae for the one-component case:
−1 ]) −1 ])
hs| ψ(z) = z s−1 hs − 1| e−J([z , hs| ψ ∗ (z) = z −s hs + 1| eJ([z

55
(for details of the proof, see [12], Lemma 5.3), which are often referred to as bosonization
rules. In a similar manner, one can prove the formulae
¯ ¯
ψ ∗ (z) |si = z −s+1 e−J([z]) |s − 1i , ψ(z) |si = z s eJ([z]) |s + 1i .
Using commutation relations (7.5), (7.6) and (7.11), we can deduce multi-component
analogues of the bosonization rules:
−1 ]
hs| ψ (α) (z) = ǫα (s)z sα −1 hs − [1]α | e−J([z α)
,
−1 ]
hs| ψ ∗(α) (z) = ǫα (s)z −sα hs + [1]α | eJ([z α)
,
(7.16)
(α) ¯ α)
sα J([z]
ψ (z) |s̄i = ǫα (s̄)z e |s̄ + [1]α i ,
¯
ψ ∗(α) (z) |s̄i = ǫα (s̄)z −sα +1 e−J([z]α ) |s̄ − [1]α i ,
where the sign factor ǫα (s) is defined in (7.8).
Let g be a general element of the Clifford group whose typical form is
!
X X (αβ) ∗(α) (β)
g = exp Ajk ψj ψk (7.17)
α,β j,k

(αβ)
with some infinite matrix Ajk . The tau-function τ (s, s̄, t, t̄; g) is defined as the expec-
tation value
¯
τ (s, s̄, t, t̄; g) = hs| eJ(t) ge−J(t̄) |s̄i . (7.18)
Note that it is non-zero only if |s| = |s̄|.

7.2 The bilinear identity

An important property of the Clifford group elements is the following operator bilinear
identity. (See, for example, §2 and §4 of [11].)
Proposition 7.1. Let g be a Clifford group element of the form ( 7.17). Then it satisfies
the operator bilinear identity
N X N X
(γ) ∗(γ) (γ) ∗(γ)
X X
ψj g ⊗ ψj g = gψj ⊗ gψj . (7.19)
γ=1 j∈Z γ=1 j∈Z

Both sides of this identity should be understood as acting to arbitrary states U ⊗ V

from the fermionic Fock space to the right and U ′ ⊗ V ′ to the left:
N X N X
(γ) ∗(γ) (γ) ∗(γ)
X X
′ ′
U ψj g U V ψj g V = U ′ gψj U V ′ gψj V . (7.20)
γ=1 j∈Z γ=1 j∈Z

Using the free fermionic fields ψ (α) (z) and ψ ∗(α) (z), we can rewrite the operator bilinear
identity in the following form:
N   X N  
X dz (γ) ∗(γ) dz (γ) ∗(γ)
res ψ (z)g ⊗ ψ (z)g = res gψ (z) ⊗ gψ (z) (7.21)
γ=1
z γ=1
z

56
 X 
(the operation res is defined as res ak z k dz = a−1 ).
k

Theorem 7.1. Let g be a Clifford group element of the form ( 7.17). Then the function

ταβ (s, t, t̄) = (−1)|s|(|s|−1)/2 τ (s + [1]α − [1]β , s, t, t̄; g)

(7.22)
|s|(|s|−1)/2 J(t) ¯ t̄)
−J(
= (−1) hs + [1]α − [1]β | e ge |si

is the tau-function of the multi-component Toda lattice hierarchy, i.e., it satisfies the
integral bilinear equation ( 6.15).
′ ¯
Proof. Putting the identity (7.21) between the states hs| eJ(t) ⊗ hs′ | eJ(t ) and e−J(t̄) |s̄i ⊗
¯ ′
e−J(t̄ ) |s̄′ i, we obtain, using (7.15):
N I
X dz ξ(tγ −t′γ ,z) ¯ ′ ¯ ′
e s ψ (γ) (z)eJ(t) ge−J(t̄) s̄ s′ ψ ∗(γ) (z)eJ(t ) ge−J(t̄ ) s̄′
γ=1 C∞ z
(7.23)
N I
X dz ξ(t̄γ −t̄′γ ,z −1 ) ¯ ′ ¯ ′
= e s eJ(t) ge−J(t̄) ψ (γ) (z) s̄ s′ eJ(t ) ge−J(t̄ ) ψ ∗(γ) (z) s̄′
γ=1 C0 z

Application of the bosonization rules (7.16) yields:

N
dz sγ −s′γ −1 ξ(tγ −t′γ ,z)
X I
′
ǫγ (s)ǫγ (s ) z e
γ=1 C∞ z

−1 ] ¯ ′ −1 ] ¯ ′
× s − [1]γ eJ(t−[z γ)
ge−J(t̄) s̄ s′ + [1]γ eJ(t +[z γ)
ge−J(t̄ ) s̄′
(7.24)
N
dz s̄γ −s̄′γ +1 ξ(t̄γ −t̄′γ ,z −1 )
X I
= ǫγ (s̄)ǫγ (s̄′ ) z e
γ=1 C0 z

¯ ′ ¯ ′
× s eJ(t) ge−J(t̄−[z]γ ) s̄ + [1]γ s′ eJ(t ) ge−J(t̄ +[z]γ ) s̄′ − [1]γ

Using the definition of the tau-function (7.18), we can rewrite this in the form
N
dz sγ −s′γ −1 ξ(tγ −t′γ ,z)
X I
′
ǫγ (s)ǫγ (s ) z e
γ=1 C∞ z
′
× τ (s − [1]γ , s̄, t − [z −1 ]γ , t̄; g) τ (s′ + [1]γ , s̄′ , t′ + [z −1 ]γ , t̄ ; g)
(7.25)
N
dz s̄γ −s̄′γ +1 ξ(t̄γ −t̄′γ ,z −1)
X I
= ǫγ (s̄)ǫγ (s̄′ ) z e
γ=1 C0 z
′
× τ (s, s̄ + [1]γ , t, t̄ − [z]γ ; g) τ (s′ , s̄′ − [1]γ , t′ , t̄ + [z]γ ; g).

57
Now let us shift s → s + [1]α , s′ → s′ − [1]β in (7.25) and put s̄ = s after that. In the
notation defined by (7.22) equation (7.25) acquires the form
N I
′ ′
X
′
ǫγ (s + [1]α )ǫγ (s − [1]β )ǫ0N (s) z sγ −sγ +δαγ +δβγ −2 eξ(tγ −tγ ,z)
γ=1 C∞

′
× ταγ (s, t − [z −1 ]γ , t̄) τγβ (s′ , t′ + [z −1 ]γ , t̄ ) dz
(7.26)
N I
X ′ ′
=− ǫγ (s)ǫγ (s′ )ǫ0N (s′ ) z sγ −sγ −2 eξ(t̄γ −t̄γ ,z)
γ=1 C∞

′
× ταγ (s + [1]γ , t, t̄ − [z −1 ]γ ) τγβ (s′ − [1]γ , t′ , t̄ + [z −1 ]γ ) dz,

where we have changed the integration variable z → z −1 in the right-hand side (the
orientation of the contour has been changed accordingly).
Let us compare this equation with (6.15). The only difference is in the sign factors in
front of the integrals. However, a simple verification shows that the sign factors in the
left-hand side are

ǫγ (s + [1]α ) = ǫαγ (s)ǫα (s), ǫγ (s′ − [1]β ) = ǫβγ (s′ )ǫβ (s′ ),

while the sign factors in the right-hand side are

ǫγ (s) = ǫαγ ǫαγ (s)ǫα (s), ǫγ (s′ ) = ǫβγ ǫβγ (s′ )ǫβ (s′ ).

Extracting the common multipliers ǫα (s)ǫβ (s′ ) and taking into account that sγ , s′γ are
integers, we see that equation (7.26) is identical to (6.15). Therefore, the function
ταβ (s, t, t̄) is indeed the tau-function of the multi-component Toda lattice hierarchy.

Let us note that the tau-function introduced in Section 5 and the one constructed
here from fermions differ by a sign factor:

τ (sec.5) (s, t, t̄) = ǫαβ (s)τ (sec.7) (s, t, t̄). (7.27)

This relation should be taken into account in dealing with an example of exact solution
of the non-abelian Toda lattice equation in the appendix.

7.3 An example of exact solution

Here we present a simple example of exact solution which is analogous to the well-known
one-soliton solution of the one-component Toda lattice.
Let us take the Clifford group element of the form
N
X 
g= ×
×
exp Aµν ψ ∗(µ) (q)ψ (ν) (p) ×
×
, (7.28)
µ,ν=1

58
where Aµν is some N × N matrix and the normal ordering ×× (. . .) ×× means that the ψ-
operators are moved to the right while ψ ∗ -operators are moved to the left (with the
minus sign factor appearing each time when two neighboring operators are permuted).
Expanding the exponential function, we have:
X
g =1+ Aµν ψ ∗(µ) (q)ψ (ν) (p)
µ,ν
(7.29)
1 XX
+ Aµ1 ,ν1 Aµ2 ,ν2 ψ ∗(µ1 ) (q)ψ ∗(µ2 ) (q)ψ (ν2 ) (p)ψ (ν1 ) (p) + . . . .
2! µ ,ν µ ,ν
1 1 2 2

In order to find the tau-function

¯
ταβ (s + [1]β , t, t̄) = s + [1]α eJ(t) ge−J(t̄) s + [1]β

explicitly, we use the Wick’s theorem and the pair correlation functions

psµ q 1−sµ
s ψ ∗(µ) (q)ψ (ν) (p) s = δµν ,
q−p
(7.30)
s + [1]α ψ ∗(µ) (q)ψ (ν) (p) s + [1]β = ǫα (s)ǫβ (s)δµβ δνα psα q −sβ , α 6= β.

In fact we will deal with the slightly modified tau-function9

X X 
′
ταβ (s + [1]β , t, t̄) = exp ktγ,k t̄γ,k ταβ (s + [1]β , t, t̄)
γ k≥1
(7.31)
¯ ¯
= s + [1]α eJ(t̄) eJ(t) ge−J(t) e−J(t̄) s + [1]β .

Let us first consider the case β = α. Using the Wick’s theorem and the expansion
¯ ¯
(7.29), we find for τ ′ (s, t, t̄) = s eJ(t̄) eJ(t) ge−J(t) e−J(t̄) s :
 
Aν1 ν1 (s, t, t̄) Aν1 ν2 (s, t, t̄) . . . Aν1 νk (s, t, t̄)
 
 
 Aν ν (s, t, t̄) Aν ν (s, t, t̄) . . . Aν ν (s, t, t̄) 
N
X 1  q k X  2 1 2 2 2 k 
τ ′ (s, t, t̄) = 1 + det 
 
k! q−p ν ,...,ν . . .

. . . . .
.
 
k=1 1 k 
 . . . 

 
Aνk ν1 (s, t, t̄) Aνk ν2 (s, t, t̄) . . . Aνk νk (s, t, t̄)
 
Aν1 ν1 (s, t, t̄) Aν1 ν2 (s, t, t̄)
X 1 X
=1+ Aνν (s, t, t̄) + det  + ... ,
2! ν ,ν
ν 1 2 Aν2 ν1 (s, t, t̄) Aν2 ν2 (s, t, t̄)
(7.32)
where
Aµν (s, t, t̄) = Aµν q −sµ psν eηµν (t,t̄,p,q) (7.33)
9
This modification was introduced in (1.3.32) of [4] for the one-component case and was interpreted
¯
as τ ′ (s, t, t̄) = hs|Ad(eJ(t) eJ(t̄) )g|si in (2.2) of [14].

59
and
ηµν (t, t̄, p, q) = ξ(tµ , p) + ξ(t̄µ , p−1 ) − ξ(tν , q) − ξ(t̄ν , q −1 ).
To see what it is, we recall the following well-known lemma:

Lemma 7.1. For an N × N matrix M, det(1N + M) is 1 plus the sum of all diagonal
minors of M (the finite-dimensional Fredholm determinant):
 
Mν1 ν1 Mν1 ν2 . . . Mν1 νk
N  Mν ν Mν ν . . . Mν ν 
X 1 X 2 1 2 2 2 k 
det (1N + M) = 1 + det  .. . . . (7.34)

k! ν ,...,ν . . . .
N ×N
k=1
 . . . . 
1 k
Mνk ν1 Mνk ν2 . . . Mνk νk

Therefore, comparing this with (7.32), we conclude that

 
′ q
τ (s, t, t̄) = det 1N + A(s, t, t̄) . (7.35)
N ×N q−p

At N = 1 this formula gives the one-soliton tau-function of the one-component Toda

lattice.
In the case α 6= β the calculation is slightly more involved. We have:
N −1
′
X 1 (αβ)
ταβ (s + [1]β , t, t̄) = ǫα (s)ǫβ (s) D (s, t, t̄), (7.36)
k=1
k! k

where
(αβ)
X
Dk (s, t, t̄) = Aµ1 ν1 eηµ1 ν1 (t,t̄,p,q) . . . Aµk νk eηµk νk (t,t̄,p,q)
µ1 ,...,µk
ν1 ,...,νk
(7.37)
× s ψs∗(α)
α
ψ ∗(µ1 ) (q) . . . ψ ∗(µk ) (q)ψ (νk ) (p) . . . ψ (ν1 ) (p)ψs(β)
β
s
(here it is assumed that sα , sβ ≥ 0). Applying the Wick’s theorem, we can represent the
multi-point correlation function as determinant of the two-point ones:
(αβ)
X
Dk (s, t, t̄) = Aµ1 ν1 eηµ1 ν1 (t,t̄,p,q) . . . Aµk νk eηµk νk (t,t̄,p,q)psα q −sβ det C(s, {µi }, {νi }).
µ1 ,...,µk
ν1 ,...,νk

(7.38)
Here C(s, {µi }, {νi }) is the (k + 1) × (k + 1) matrix
 
0 δαν1 δαν2 ... δανk
 
 
 δβµ
 1 s ψ ∗(µ1 ) (q)ψ (ν1 ) (p) s s ψ ∗(µ1 ) (q)ψ (ν2 ) (p) s ... s ψ ∗(µ1 )
(q)ψ (p) s  (νk )

 
 
 δβµ
 2 s ψ ∗(µ2 ) (q)ψ (ν1 ) (p) s s ψ ∗(µ2 ) (q)ψ (ν2 ) (p) s . . . s ψ ∗(µ2 ) (q)ψ (νk ) (p) s 
.
 
 
 .. .. .. .. .. 
 . . . . . 
 
 
δβµk s ψ ∗(µk ) (q)ψ (ν1 ) (p) s s ψ ∗(µk ) (q)ψ (ν2 ) (p) s ... s ψ ∗(µk ) (νk )
(q)ψ (p) s

60
Expanding the determinant in the first row and the first column, we get:
 q k−1 Xk  
(αβ) a+b+1
Dk (s, t, t̄) = (−1) Aµ1 ν1 (s,t,t̄) . . . Aµk νk (s,t,t̄)δµa β δνb α det δµi νj
q−p a,b=1
1≤i,j≤k
i6=a,j6=b

After some transformations we obtain the following result:

 
Aβα Aν1 α Aν2 α ... Aνk−1 α
 Aβν Aν1 ν1 Aν2 ν1 ... Aνk−1 ν1 
 q k−1 X  1 
(αβ)  Aβν A A ... Aνk−1 ν2
Dk = −k det   (7.39)
ν ν ν

2 1 2 2 ν2
q−p .
.. .
.. .
.. .. ..
.
ν1 ,...,νk−1
 
 . 
Aβνk−1 Aν1 νk−1 Aν2 νk−1 . . . Aνk−1 νk−1

(to save the space, here we do not indicate the dependence on s, t, t̄). To see what it is,
we need the following lemma.

Lemma 7.2. For an N × N matrix M, the αβ-minor of the matrix 1N + M with α 6= β,

(1N + M)α̂β̂ = (−1)α+β det (1N + M)µν ,

1≤µ,ν≤N
µ6=α,ν6=β

is expressed in terms of minors of the matrix M in the following way:

 
Mβα Mν1 α Mν2 α . . . Mνk−1 α
 Mβν Mν1 ν1 Mν2 ν1 . . . Mνk−1 ν1 
N −1 1
X 1 X 
 Mβν M M . . . Mνk−1 ν2

(1N +M)α̂β̂ = − det  ν1 ν2 ν2 ν2

2
(k−1)! . . . ..

.. .. .. ..
.
ν1 ,...,νk−1
 
k=1  . 
Mβνk−1 Mν1 νk−1 Mν2 νk−1 . . . Mνk−1 νk−1
 
  Mβα Mβν1 Mβν2
X Mβα Mβν 1 X
= −Mβα − det − det  Mν1 α Mν1 ν1 Mν1 ν2  − . . . .
Mνα Mνν 2! ν ,ν
ν 1 2 Mν2 α Mν2 ν1 Mν2 ν2
(7.40)

Proof. We assume that α 6= β. By permutation of rows and columns of the matrix

1N + M in which the α-th row and the β-th column are removed we can represent it as
a block matrix of the form
m(β)
 
Mβα
=: K (αβ) ,
m̂(α) 1N −2 + M̃N −2

where m(β) is the (N − 2)-dimensional row vector with components (m(β) )µ = Mβµ ,
µ 6= α, β, m̂(α) is the (N −2)-dimensional column vector with components (m̂(α) )µ = Mµα ,
µ 6= α, β and M̃N −2 is the (N − 2) × (N − 2) square matrix which is obtained from the
matrix M by removing α-th and β-th columns and rows. By counting the number of
necessary permutations of rows and columns it is easy to see that

(1N + M)α̂β̂ = − det K (αβ) .

61
Writing Mβα = 1 + (Mβα − 1) =: 1 + M̃βα with M̃βα = Mβα − 1, we bring the matrix
K (αβ) to the form 1N −1 + M̃N −1 , so its determinant can be represented as the sum of
diagonal minors of the matrix

Mβα −1 m(β)
 
M̃N −1 =
m̂(α) M̃N −2

as is stated in Lemma 7.1. An easy calculation shows that in this way one obtains
equation (7.40).

Therefore, the sum in the right-hand side of equation (7.36) is nothing else than
q
the αβ-minor of the matrix 1N + A(s, t, t̄) (up to a common multiplier). More
q−p
precisely, we have:

′ q−p  q 
ταβ (s + [1]β ), t, t̄) = ǫα (s)ǫβ (s) 1N + A(s, t, t̄)
q q−p α̂β̂

α+β q−p  q 
= (−1) ǫα (s)ǫβ (s) det δµν + Aµν (s, t, t̄) , α 6= β.
q 1≤µ,ν≤N
µ6=α,ν6=β
q−p
(7.41)
Together with equation (7.35) this gives a multi-component analogue of the one-soliton
solution to the Toda lattice.

8 Concluding remarks
In this paper we have introduced an extension of the N-component Toda lattice hierarchy.
This hierarchy contains N discrete variables s = {s1 , . . . , sN } rather than one, as it goes
in the version suggested by Ueno and Takasaki in 1984 [4]. Simultaneously, we have
refined some arguments from [4].
We have obtained the multi-component Toda lattice hierarchy in three different ways,
deducing it from different starting points.
One of them is the Lax formalism whose main ingredients are two Lax operators L, L̄
and auxiliary operators U α , Ū α , P α , P̄ α , α = 1, . . . , N, which are realized as difference
operators with N ×N matrix coefficients. These operators are subject to certain algebraic
relations. Their evolution in the time variables is given by the Lax equations (or discrete
Lax equations for evolution in the discrete variables s1 , . . . , sN ). We have presented a
detailed proof that the Lax representation is equivalent to the system of Zakharov-Shabat
(or zero curvature) equations. Next, we have proved the existence of the so-called wave
operators W , W̄ from which all other operators of the Lax formalism are obtained
by “dressing”. With the help of the wave operators, one can introduce matrix wave
functions which obey an infinite system of linear equations. Compatibility conditions for
this system is just the Zahkarov-Shabat equations. The wave functions, together with
their adjoint functions, are shown to obey a fundamental integral bilinear identity. In
its turn, this identity implies the existence of the matrix tau-function which is the most
fundamental dependent variable of the hierarchy. The tau-function is shown to satisfy

62
the integral bilinear equation which is a sort of generating equation for equations of the
hierarchy.
The alternative starting point is the multi-component KP hierarchy, which is es-
sentially equivalent to the so-called universal hierarchy [9]. We have shown that the
N-component Toda lattice hierarchy can be embedded into the 2N-component univer-
sal hierarchy. Namely, we have shown that under certain conditions the integral bilin-
ear equation for the latter becomes the integral bilinear equation for the former (more
precisely, to identify them, some simple redefinition of the tau-functions consisting in
multiplying them by some sign factors is necessary).
Last but not least, there is an approach based on the quantum field theory of free
fermions, which was developed in early 1980’s by Kyoto school. For multi-component
hierarchies one should deal with multi-component fermions. In this formalism, the tau-
functions are defined as expectation values of certain operators constructed from free
fermions (Clifford group elements); and the integral bilinear equation for the tau-function
is a corollary of the bilinear identity for fermionic operators which is a characteristic
property of Clifford group elements. In order to deduce this corollary, one needs certain
relations between fermi- and bose-operators which are often referred to as bosonization
rules. In this paper, we have implemented this program. As a result, we have obtained
the integral bilinear equation for the tau-function of the multi-component Toda lattice
hierarchy which turns out to be the same as the ones obtained in the framework of the
other two approaches.
We hope that the present paper provides a complete treatment of the subject. Let us
list possible directions for further work.
One of them is the problem of defining multi-component analogues of the Toda lattices
with constraints of types B and C introduced and studied in [20, 21] and [22] respectively.
The former hierarchy is equivalent to the so-called large BKP hierarchy (see, e.g., [23]
and §7.4 of the book [24]).
Another direction for further work is studying the dispersionless limit of the multi-
component Toda lattice hierarchy (see [25] for the case N = 1). In the dispersionless limit
one should re-scale the independent variables as tα,k → tα,k /~, t̄α,k → t̄α,k /~, sα → sα /~
and consider solutions (tau-functions) that have an essential singularity at ~ = 0 and
have the form 1
τ (s/~, t/~, t̄/~) = e ~2 F (s,t,t̄,~)
as ~ → 0, where F is a smooth function of s and t, t̄ having a regular expansion in
~ as ~ → 0. The function F = F (s, t, t̄, 0) is sometimes called the dispersionless tau-
function (although the ~ → 0 limit of the tau-function itself does not exist). The bilinear
equations for the tau-function τ (s, t, t̄) lead to non-linear equations for the F -function.
It would be interesting to write these equations explicitly and compare with the ones
obtained in [7, 26].
It seems to be an intriguing problem to study the N-component Toda lattice in the
limit N → ∞ and to see whether any new phenomena arise in this limit.
Finally, it would be desirable to obtain multi-component analogues of multi-soliton
solutions to the Toda lattice in an explicit form. The natural framework for this is the
free fermion technique. The example of one-soliton solution is given in this paper in

63
Section 7.3.

Appendix: Non-abelian Toda lattice

In the appendix we consider the non-abelian Toda lattice which provides an explicit ex-
ample of the general construction of Section 2. Namely, we derive the non-abelian Toda
lattice equation from the Zakharov-Shabat representation of the matrix Toda lattice hi-
erarchy which can be regarded as a subhierarchy of the multi-component one discussed in
Section 2. In the matrix Toda hierarchy, the independent variables s, tk , t̄k are introduced
by assigning the variables sα , tα,k , t̄α,k the following values:
(0) (0)
sα = s(0)
α + s, tα,k = tα,k + tk , t̄α,k = t̄α,k + t̄k ,
(0) (0) (0)
where sα , tα,k , t̄α,k are some fixed parameters, so the corresponding vector fields are

N
X N
X
∂tk = ∂tα,k , ∂t̄k = ∂t̄α,k .
α=1 α=1

Accordingly, the operators U α , Ū α , P α , P̄ α do not take part in the construction and we

are left with the two Lax operators with matrix coefficients
∞
X
L(s) = bj (s)e(1−j)∂s , b0 (s) = 1N ,
j=0
(A1)
∞
X
L̄(s) = b̄j (s)e(j−1)∂s , b̄0 (s) = g(s)g −1(s − 1),
j=0

where g(s) is an N × N invertible matrix (which was denoted by w̃0 in Section 2). The
dependent variables (in particular, g(s)) are regarded as functions of s, tk , t̄k .
The Lax operators satisfy the Lax equations

∂tk L(s) = [B k (s), L(s)], ∂tk L̄(s) = [B k (s), L̄(s)],

(A2)
∂t̄k L(s) = [B̄ k (s), L(s)], ∂t̄k L̄(s) = [B̄ k (s), L̄(s)],

where
k
B k (s) = (Lk (s))≥0 , B̄ k (s) = (L̄ (s))<0 .
Compatibility conditions for the Lax equations are expressed as Zakharov-Shabat equa-
tions. We will derive the first nontrivial equation of the hierarchy from the Zakharov-
Shabat equation
[∂tm − B m (s), ∂t̄n − B̄ n (s)] = 0 (A3)
at m = n = 1. In what follows we put t1 = t, t̄1 = t̄.
We have:
B 1 = 1N e∂s + b1 (s), B̄ 1 = b̄0 (s)e−∂s . (A4)

64
Plugging this into the Zakharov-Shabat equation, we obtain the system of equations

 ∂t̄ b1 (s) = b̄0 (s) − b̄0 (s + 1),
(A5)
∂t b̄0 (s) = b1 (s)b̄0 (s) − b̄0 (s)b1 (s − 1).


Substituting b̄0 (s) = g(s)g −1(s − 1), we represent the second equation in the form

g −1(s)h(s)g(s) = g −1 (s − 1)h(s − 1)g(s − 1), (A6)

where
h(s) = ∂t g(s) g −1(s) − b1 (s). (A7)
Therefore, g −1(s)h(s)g(s) = h0 does not depend on s. We assume that it does not depend
also on the times, so h0 is a constant matrix. From (A6) we have:

h(s) = g(s)h0 g −1(s). (A8)

Plugging this into the first equation in (A5), we obtain:

∂t̄ (∂t g(s) g −1(s)) = g(s)g −1(s − 1) − g(s + 1)g −1 (s) + ∂t̄ (g(s)h0g −1 (s)). (A9)

The simple redefinition g(s) → g(s)eth0 kills the last term in the right-hand side, so we
can put h(s) = 0 without loss of generality. In this way we obtain the equation of the
non-abelian Toda lattice:

∂t̄ (∂t g(s) g −1(s)) = g(s)g −1(s − 1) − g(s + 1)g −1(s). (A10)

ˆ and the Zakharov-Shabat equation

Actually, using the linear equations (3.11) for W̄
(A3) for n = m = 1, we can easily show that g(s) = w̄0 (s) satisfies this equation. In fact,
the above redefinition g(s) → g(s)eth0 corresponds to w̄0 (s) = w̃0 (s) c̄˜(t, t̄) in Remark 3.1.
The tau-function provides bilinearization of the non-abelian Toda lattice equation
(A10). In terms of the tau-function from Section 5 we have:

ταβ (s1 + [1]β )

gαβ (s) = (−1)δαβ −1 ,
τ (s1)
(A11)
ταβ ((s + 1)1 − [1]α )
(g −1 (s))αβ =
τ ((s + 1)1)

and (A10) follows from (3.33) and (3.34). Note that the tau-functions here can be
′
substituted by the modified tau-functions ταβ defined in (7.31).
Finally, let us give an explicit example of exact solutions to (A10) based on the one
considered in Section 7.3. Taking into account the difference between the tau-functions
introduced in Sections 5 and 7 (see (7.27)), we write:
 q 
τ ′ (s1) = det δµν + Aµν (s, t, t̄) , (A12)
1≤µ,ν≤N q−p

65
 q 1−δβµ pδβν
  
det δµν + Aµν (s, t, t̄) , α = β,


q−p


 1≤µ,ν≤N
′
ταβ (s1 + [1]β ) = (A13)
α+β+1 q − p q
  
 (−1) det δµν + Aµν (s, t, t̄) ,


q q−p

1≤µ,ν≤N
µ6=α,ν6=β

where
−1 −q −1 )t̄
Aµν (s, t, t̄) = (p/q)s e(p−q)t+(p Aµν .
For example, at N = 2 we have the solution
 
1+pB11 +qB22 +pq det B (p − q)B12
1
g(s) =  
1 + qtrB + q 2 det B
(p − q)B21 1+qB11 +pB22 +pq det B

(12 + pB)(12 + q B̃)

= ,
det(12 + qB)
 (A14)

1+qB11 +pB22 +pq det B (p − q)B12
1
g −1(s) =  
1 + ptrB + p2 det B
(p − q)B21 1+pB11 +qB22 +pq det B

(12 + qB)(12 + pB̃)

= ,
det(12 + pB)
(A15)
where
1  p s (p−q)t+(p−1 −q−1 )t̄ T
B = B(s, t, t̄) = e A
q−p q
and B̃ is the matrix of minors of B (B̃11 = B22 , B̃22 = B11 , B̃12 = −B21 , B̃21 = −B12 ).

Acknowledgments
We thank S. Kakei and V. Prokofev for discussions. The work of A.Z. has been supported
in part within the framework of the HSE University Basic Research Program (section 5)
and within the state assignment of NRC “Kurchatov institute” (section 6).

We dedicate this paper to the memory of Masatoshi Noumi, who passed away on 20
November 2024. His interest, comments and encouragements to our works were invaluable
for us.

References
[1] M. Toda, Vibration of a chain with a non-linear interaction, J. Phys. Soc. Japan 22
(1967) 431–436.

66
 [2] M. Toda, Hisenkei koushi rikigaku (in Japanese; Nonlinear lattice dynamics),
Iwanami (1978) (English translation: Theory of nonlinear lattices, Springer-Verlag,
Berlin, Heidelberg, New York, 1981).

[3] A.V. Mikhailov, Integrability of the two-dimensional generalization of Toda chain,

JETP Letters 30 (1979) 443–448.

[4] K. Ueno and K. Takasaki, Toda lattice hierarchy, Adv. Studies in Pure Math. 4
(1984) 1–95.

[5] E. Date, M. Jimbo, M. Kashiwara and T. Miwa, Transformation groups for soliton
equations III, J. Phys. Soc. Japan 50 (1981) 3806–3812.

[6] V. Kac and J. van de Leur, The n-component KP hierarchy and representation
theory, in: A.S. Fokas, V.E. Zakharov (Eds.), Important Developments in Soliton
Theory, Springer-Verlag, Berlin, Heidelberg, 1993.

[7] K. Takasaki and T. Takebe, Universal Whitham hierarchy, dispersionless Hirota

equations and multicomponent KP hierarchy, Physica D 235 (2007) 109–125.

[8] L.-P. Teo, The multi-component KP hierarchy: differential Fay identities and Lax
equations, J. Phys. A: Math. Theor. 44 (2011) 225201.

[9] I. Krichever and A. Zabrodin, Quasi-periodic solutions of the universal hierarchy, to

be published in Annales Henri Poincaré, arXiv:2308.12187.

[10] E. Date, M. Jimbo, M. Kashiwara and T. Miwa, Transformation groups for soliton
equations, in: M. Jimbo, T. Miwa (Eds.), Nonlinear Integrable Systems – Classical
and Quantum, World Scientific, 1983, pp. 39–120.

[11] M. Jimbo and T. Miwa, Solitons and infinite dimensional Lie algebras, Publ. Res.
Inst. Math. Sci. Kyoto 19 (1983) 943–1001.

[12] E. Date, M. Jimbo and T. Miwa, Soliton no suuri (in Japanese; Mathematics of
solitons), Iwanami (1993) (English translation: Solitons, Cambridge Tracts in Math-
ematics 135, Cambridge University Press (1999)).

[13] S. Kharchev, A. Marshakov, A. Mironov, A. Orlov and A. Zabrodin, Matrix Models

among Integrable Theories: Forced Hierarchies and Operator Formalism, Nuclear
Physics B366 (1991) 569–601.

[14] T. Takebe, Representation theoretical meaning of the initial value problem for the
Toda lattice hierarchy I, Lett. Math. Phys. 21 (1991), 77–84; II, Publ. RIMS 27
(1991), 491–503.

[15] J.-L. Gervais and Y. Matsuo, W-geometries, Physics Letters B274 (1992) 309–316.

[16] T. Takebe, A note on the modified KP hierarchy and its (yet another) dispersionless
limit, Lett. Math. Phys. 59 (2002), 157–172.

[17] T. Takebe, Toda Lattice Hierarchy and Conservation Laws, Commun. Math. Phys.
129 (1990) 281–318.

67
[18] L.A. Dickey, On τ -functions of Zakharov-Shabat and other matrix hierarchies of inte-
grable equations, in: Fokas, A.S., Gelfand, I.M. (eds), Algebraic Aspects of Integrable
Systems, Progress in Nonlinear Differential Equations and Their Applications, vol.
26 (1997), Birkhauser Boston, arXiv:hep-th/9502063.

[19] A. Zabrodin, On matrix modified Kadomtsev-Petviashvili hierarchy, Teor. Mat. Fys.

199 (2019) 343–356 (English translation: Theor. Math. Phys. 199 (2019) 771–783),
arXiv:1802.02797.

[20] I. Krichever, A. Zabrodin, Toda lattice with constraint of type B, Physica D, 453
(2023) 133827.

[21] V. Prokofev, A. Zabrodin, Tau-function of the B-Toda hierarchy, Teor. Mat. Fys.
217:2 (2023) 299–316, (English translation: Theor. Math. Phys. 217:2 (2023)
1673—1688).

[22] I. Krichever, A. Zabrodin, Constrained Toda hierarchy and turning points of the
Ruijsenaars-Schneider model, Letters in Mathematical Physics 112 23 (2022),
arXiv:2109.05240

[23] W.C. Guan, S. Wang, W.J. Rui, J.P. Cheng, Lax structure and tau function for large
BKP hierarchy, preprint arXiv:2404.09815.

[24] J. Harnad and F. Balogh, Tau functions and their applications, Cambridge Mono-
graphs on Mathematical Physics, Cambridge University Press, 2021.

[25] K. Takasaki and T. Takebe, Integrable hierarchies and dispersionless limit, Rev.
Math. Phys. 7 (1995), 743—808.

[26] A. Zabrodin, Dispersionless version of the multicomponent KP hierarchy revisited,

Physica D 467 (2024) 134286, arXiv:2404.10406.

68