LPTHE-0002

Cohomologies of Affine Jacobi Varieties

and Integrable Systems .
arXiv:math-ph/0001017v1 13 Jan 2000

a b 0
A. Nakayashiki and F.A. Smirnov

a
Graduate School of Mathematics, Kyushu University
Ropponmatsu 4-2-1, Fukuoka 810-8560, Japan

b
Laboratoire de Physique Théorique et Hautes Energies 1
Université Pierre et Marie Curie, Tour 16 1er étage, 4 place Jussieu
75252 Paris cedex 05-France

Abstract. We study the affine ring of the affine Jacobi variety of a hyper-elliptic
curve. The matrix construction of the affine hyper-elliptic Jacobi varieties due
to Mumford is used to calculate the character of the affine ring. By decomposing
the character we make several conjectures on the cohomology groups of the affine
hyper-elliptic Jacobi varieties. In the integrable system described by the family
of these affine hyper-elliptic Jacobi varieties, the affine ring is closely related to
the algebra of functions on the phase space, classical observables. We show that
the affine ring is generated by the highest cohomology group over the action of
the invariant vector fields on the Jacobi variety.

0 Membre du CNRS
1 Laboratoire associé au CNRS.

1
1 Introduction.
Our initial motivation is the study of integrable systems. Consider an integrable
system with 2n degrees of freedom, by definition it possesses n integrals in
involution. The levels of these integrals are n-dimensional tori. This is a general
description, but the particular examples of integrable models that we meet in
practice are much more special. Let us explain how they are organized.
The phase space M is embedded algebraically into the space RN . The
integrals are algebraic functions of coordinates in this space. This situation
allows complexification, the complexified phase space MC is an algebraic affine
variety embedded into CN . The levels of integrals in the complexified case allow
the following beautiful description. The systems that we consider are such that
with every one of them one can identify an algebraic curve X of genus n whose
moduli are defined by the integrals of motion. On the Jacobian J(X) of this
curve there is a particular divisor D (in this paper we consider the case when
this divisor coincides with the theta divisor, but more complicated situations
are possible). The level of integrals is isomorphic to the affine variety J(X) − D.
The real space RN ⊂ CN intersects with every level of integrals by a compact
real sub-torus of J(X) − D.
This structure explains why the methods of algebraic geometry are so impor-
tant in application to integrable models. Closest to the present paper account
of these methods is given in the Mumfords’s book [1].
Let us describe briefly the results of the present paper. We study the struc-
ture of the ring A of algebraic functions (observables) on the phase space of
certain integrable model. The curve X in our case is hyper-elliptic. As is clear
from the description given above this ring of algebraic functions is, roughly, a
product of the functions of integrals of motion by the affine ring of hyper-elliptic
Jacobian. The commuting vector fields defined by taking Poisson brackets with
the integrals of motion are acting on A. We shall show that by the action of
these vector-fields the ring A is generated from finite number of functions corre-
sponding to the highest nontrivial cohomology group of the affine Jacobian. We
conjecture the form of the cohomology groups in every degree and demonstrate
the consistence of our conjectures with the structure of the ring A.
Finally we would like to say that this relation to cohomology groups became
clear analyzing the results of papers [2] and [3] which deal with quantum in-
tegrable models. Very briefly the reason for that is as follows. The quantum
observables are in one-to-one correspondence with the classical ones. Consider a
matrix element of some observable between two eigen-functions of Hamiltonians.
An eigen-functions written in “coordinate” representation (for “coordinates” we
take the angles on the torus) must be considered as proportional to square-root
of the volume form on the torus. The matrix element is written as integral with
respect to “coordinates”, the product of two eigen-functions gives a volume form
on the torus, and the operator itself can be considered, at least semi-classically,
as a multiplier in front of this volume form, i.e. as coefficient of some differen-
tial top form on the torus which is the same as the form of one-half of maximal
dimension on the phase space. Further, those operators which correspond to

2
“exact form” have vanishing matrix elements. This is how the relation to the
cohomologies appears.
The paper, after the introduction, consists of five sections and six appendices
which contain technical details and some proofs.
In section 2 we recall the standard construction of the Jacobi variety which
is valid for any Riemann surface.
An algebraic construction of the affine Jacobi variety J(X) − Θ of a hyper-
elliptic curve X is reviewed in section 3 following the book [1]vol.II. This con-
struction is specific to hyper-elliptic curves or more generally spectral curves
[5].
In section 4 we study the affine ring of J(X) − Θ using the description in
section 3. The main ingredient here is the character of the affine ring. To be
precise we consider the ring A0 corresponding to the most degenerate curve
y 2 = z 2g+1 . The ring A and the affine ring Af of J(X) − Θ for a non-singular
X can be studied using A0 . It is important that A0 is a graded ring and the
character ch(A0 ) is defined. We calculate it by determining explicitly a C-
basis of A0 . The relation between A0 and Af for a non-singular X is given in
Appendix E.
A set of commuting vector fields acting on A is introduced in section 5. This
action descends to the quotients A0 and Af . The action of the vector fields
coincides with the action of invariant vector fields on J(X). With the help
of these vector fields we define the de Rham type complexes (C∗ , d), (C∗0 , d),
(C∗f , d) with the coefficients in A, A0 , Af respectively. The complex (C∗f , d)
is nothing but the algebraic de Rham complex of J(X) − Θ whose cohomology
groups are known to be isomorphic to the singular cohomology groups of J(X)−
Θ. What is interested for us is the cohomology groups of (C∗0 , d). We calculate
the q-Euler characteristic of (C∗0 , d) and show that it coincides with the quotient
of ch(A0 ) by the character ch(D) of the space of commuting vector fields. Then,
by the Euler-Poincaré principle, ch(A0 )/ch(D) is found to be expressible as the
alternating sum of the characters of cohomology groups of (C∗0 , d). Decomposing
independently the explicit formula of ch(A0 ) into the alternating sum, we make
conjectures on the cohomology groups of (C∗0 , d) which are formulated in the
next section.
In section six and in Appendix B we study the singular homology and co-
homology groups of J(X) − Θ. The Riemann bilinear relation plays an impor-
tant role here. We formulate conjectures on the cohomology groups of (C∗ , d),
(C∗0 , d), (C∗f , d).

Acknowledgements. This work was begun during the visit of one of the
authors (A.N.) to LPTHE of Université Paris VI and VII in 1998-1999. We
express our sincere gratitude to this institution for generous hospitality. A.N
thanks K. Cho for helpful discussions.

3
2 Hyper-elliptic curves and their Jacobians.
Consider the hyper-elliptic curve X of genus g described by the equation:

y 2 = f (z),

where

f (z) = z 2g+1 + f1 z 2g + · · · + f2g+1 . (1)

The hyper-elliptic involution σ is defined by

σ(z, y) = (z, −y).

The Riemann surface X can be realized as two-sheeted covering of the z-sphere

with the quadratic branch points which are zeros of the polynomial f (z) and
∞.
A basis of holomorphic differentials is given by:
dz
µj = z g−j , j = 1, · · · , g.
y
Choose a canonical homology basis of X: α1 , · · · , αg , β1 , · · · , βg . The basis
of normalized differentials is defined as
g
X
ωi = (M −1 )ij µj ,
j=1

where the matrix M consists of α-periods of holomorphic differentials µi :

Z
Mij = µi , i, j = 1 · · · , g. (2)
αj

The period matrix Z

Bij = ωj
βi

defines a point B in the Siegel upper half space:

Bij = Bji , Im(B) > 0.

The Jacobi variety of X is a g-dimensional complex torus:

Cg
J(X) = .
Zg + BZg
The Riemann theta function associated with J(X) is defined by
X
exp 2πi 12 t mBm + t mζ ,


θ(ζ) =
m∈Zg

4
where ζ ∈ Cg . The theta function satisfies

θ(ζ + m + Bn) = exp 2πi − 21 t nBn − t nζ

θ(ζ),

for m, n ∈ Zg .
Consider the symmetric product of X, the quotient of the product space by
the action of the symmetric group:

X(n) = X n /Sn .

The Abel transformation defines the map

a
X(g) −→ J(X)

explicitly given by

g Z pk
X
wj = ωj + ∆,
k=1 ∞

where p1 , · · · , pg are points of X, ∆ is the Riemann characteristic corresponding

to the choice of ∞ for the reference point. In the present case ∆ is a half-period
because ∞ is a branch point [1].
The divisor Θ is the (g − 1)-dimensional subvariety of J(X) defined by

Θ = {w | θ(w) = 0}. (3)

The main subject of our study is the ring A of meromorphic functions on J(X)
with singularities only on Θ. The simple way to describe this ring is provided
by theta functions:
∞ 
[ Θk (w) 
A= , (4)
θ(w)k
k=0

where Θk is the space of theta functions of order k i.e. the space of regular
functions on Cg satisfying

θk (w + m + Bn) = exp 2kπi − 21 t nBn − t nw θk (w).

There are k g linearly independent theta functions of order k.

Let us discuss the geometric meaning of the ring A. It is well known that
with the help of theta functions one can embed the complex torus J(X) into
the complex projective space as a non singular algebraic subvariety. It can be
done, for example, using theta functions of third order:
1. 3g theta functions of third order define an embedding of J(X) into the
g
complex projective space P3 −1 ,
2. a set of homogeneous algebraic equations for these theta functions can be
written, which allows to describe this embedding as algebraic one.

5
 Now consider the functions
Θ3 (w)
.
θ(w)3
Obviously, with the help of these functions, we can embed the non-compact
g
variety J(X) − Θ into the complex affine space C3 −1 . Denote the coordinates
in this space by x1 , · · · , x3g −1 , the affine ring of J(X) − Θ is defined as the ring

C [x1 , · · · , x3g −1 ]/(gα ),

where (gα ) is the ideal generated by the polynomials {gα } such that {gα = 0}
defines the embedding. It is known that the affine ring is the characteristic of
the non-compact variety J(X)−Θ independent of a particular embedding of this
variety into affine space. Obviously the ring A defined above is isomorphic to
the affine ring. We remark that the above argument on the embedding J(X)−Θ
into an affine space is valid if (J(X), Θ) is replaced by any principally polarized
abelian variety.
Consider X(g) which is mapped to J(X) by the Abel map a. The Riemann
theorem says that
pj
g−1
XZ
θ(w) = 0 iff w= ω+∆
j=1 ∞

which allows to describe Θ in terms of the symmetric product. One easily argues
that the preimage of Θ under the Abel map is described as

D := D∞ ∪ D0 ,

where

D∞ = {(p1 , · · · , pg ) ∈ X(g)| pi = ∞ for some i}, (5)

D0 = {(p1 , · · · , pg ) ∈ X(g)| pi = σ(pj ) for some i 6= j }.

The Abel map is not one-to-one, and the compact varieties J(X) and X(g)
are not isomorphic. However, the affine varieties J(X) − Θ and X(g) − D are
isomorphic since the Abel map
a
X(g) − D −→ J(X) − Θ

is an isomorphism. In what follows we shall study the affine variety X(g) − D ≃

J(X) − Θ.

3 Affine model of hyperelliptic Jacobian.

Consider a traceless 2 × 2 matrix
 
a(z) b(z)
m(z) = ,
c(z) −a(z)

6
where the matrix elements are polynomials of the form:
a(z) = a 23 z g−1 + a 25 z g−2 + · · · + ag+ 12 , (6)
b(z) = z g + b1 z g−1 + · · · + bg ,
c(z) = z g+1 + c1 z g + c2 z g−1 + · · · + cg+1 .
Later we shall set b0 = c0 = 1. Consider the affine space C3g+1 with coordinates
a 32 , · · · , a g+1 , b1 , · · · , bg , c1 , · · · , cg+1 . Fix the determinant of m(z):
2

a2 (z) + b(z)c(z) = f (z), (7)

where the polynomial f (z) is the same as used above (1). Comparing each
coefficient of z i (i = 0, 1, · · · , 2g) of (7) one gets 2g + 1 different equations. In
fact the equations (7) define g-dimensional sub-variety of C3g+1 . This algebraic
variety is isomorphic to J(X) − Θ as shown in the book [1]. We shall briefly
recall the proof.
Consider a matrix m(z) satisfying (7). Take the zeros of b(z):
g
Y
b(z) = (z − zj )
j=1

and set
yj = a(zj ).
Obviously zj , yj satisfy the equation
yj2 = f (zj ),
which defines the curve X. So, we have constructed a point of X(g) for every
m(z) which satisfies the equations (7). Conversely, for a point (p1 , · · · , pg ) of
X(g), construct the matrix m(z) as
Y g Xg Y  z − zk 
b(z) = (z − zj ), a(z) = yj ,
j=1 j=1
zj − zk
k6=j
2
−a(z) + f (z)
c(z) = ,
b(z)
where zj = z(pj ) is the z-coordinate of pj . Considering the function b(z) as a
function on X(g) one finds that it has singularities when one of zj equals ∞.
The function a(z) is singular at zj = ∞ and also at the points where zi = zj
but yi = −yj . This is exactly the description of the variety D. The functions
a(z) and c(z) do not add new singularities. Thus we have the embedding of the
affine variety X(g) − D into the affine space:
X(g) − D ֒→ C3g+1 .
Therefore we can profit from the wonderful property of the hyper-elliptic Jaco-
bian: it allows an affine embedding into a space of very small dimension equal
to 3g + 1 (compare with 3g − 1 which we have for any Abelian variety). Actu-
ally, the space C3g+1 occurs foliated with generic leaves isomorphic to the affine
Jacobians.

7
4 Properties of affine ring.
Consider the free polynomial ring A:
A = C [a 32 , · · · , ag+ 12 , b1 , · · · , bg , c1 , · · · , cg+1 ].
On the ring A one can naturally introduce a grading. Prescribe the degree j to
any of generators aj , bj , cj and extend this definition to all monomials in A by
deg(xy) = deg(x) + deg(y).
Every monomial of the ring has positive degree (except for 1 whose degree equals
0). Thus, as a linear space, A splits into
M
A= A(p) ,
2p∈Z+

where A(j) is the subspace of degree j and Z+ = {0, 1, 2, · · · }. Define the

character of A by X
ch(A) = q p dim(A(p) ).
2p∈Z+

Since the ring A is freely generated by ap , bp , cp one easily finds

1
ch(A) =   2 , (8)
g + 12 ! [g]! [g + 1]!
where, for k ∈ Z+ ,
[k] = 1 − q k , k + 12 ! = 12 23 · · · k + 12 .
      
[k]! = [1] · · · [k],
This important formula allows to control the size of the ring A.
The relation of the ring A to the affine ring A is obvious. The latter is the
quotient of A by the ideal generated by the relations −det(m(z)) = f (z) where
the coefficients of f are considered fixed constants.
From the point of view of integrable models, it is more natural to see
f1 , · · · , f2g+1 as variables than complex numbers. If we assign degree j to the
variables fj , all the equations in (7) are homogeneous. Consider the polynomial
ring
F = C [f1 , · · · , f2g+1 ].
The ring F is graded and its character is
1
ch(F) = .
[2g + 1]!
The ring F acts on A, that is, f (z) acts by the multiplication of the left
hand side of (7). Consider the space A0 which consists of F-equivalence classes:
2g+1
X
A0 = A / (F× A), F× = Ffi .
i=1

8
Since F× A is a homogeneous ideal of A, A0 is a graded vector space:
M (p)
A0 = A0 .
2p∈Z+

One can consider the space A0 as a subspace of A taking a set of homo-

geneous representatives of the equivalence classes (being homogeneous they are
automatically of smallest possible degree). Consider any homogeneous x ∈ A.
One can write x as
2g+1
X
(0)
x=x + fi xi ,
i=1
(0)
where x ∈ A0 and xi is a homogeneous element in A satisfying deg xi =
deg x − i. Since the degree of xi is less than the degree of x, repeating the same
procedure for xi one arrives, by finite number of steps, at
(0)
X
x= hj xj , (9)
(0)
where xj ∈ A0 , hi ∈ F and the summation is finite. There is an F-linear map:
m
F ⊗C A0 −→ A,
which corresponds to multiplying the elements of A0 by elements from F and
taking linear combinations. The above reasoning shows that Im(m) = A. Hence
ch(A)
ch(A0 ) ≥ . (10)
ch(F)
The equality takes place iff Ker(m) = 0. We shall see that this is indeed the
case. Informally the equality Ker(m) = 0 is a manifestation of the fact that
the space C3g+1 is foliated into g-dimensional sub-varieties, the coordinates fj
describe transverse direction. The pure algebraic proof of this fact is given by
the following proposition.

Proposition 1. The set of elements

g g
i
Y Y
j
u 1+j ulg+1+k
k
,
2 2
j=1 k=1

where 
ap , p = half-integer
up =
bp , p = integer,
is a basis of A0 as a vector space, where i1 , · · · , ig are non-negative integers and
l1 , · · · , lg are 0 or 1.

The proof of Proposition 1 is given in Appendix A. Proposition 1 shows that

g g 1
1 Y 2 [2g + 1]! ch(A)
Y g+1+k

ch(A0 ) =  1+j  1+q 2 =  1 = , (11)
j=1 2
g + 2 ! [g]! [g + 1]! ch(F)
k=1

9
which means that Ker(m) = 0. We summarize this in the following:

Proposition 2. As an F module, A is a free module, A ≃ F ⊗C A0 . In other

words every element x ∈ A can be uniquely presented as a finite sum:
(0)
X
x= hj xj ,
(0)
where {xj } is a basis of the C-vector space A0 and hj ∈ F.

5 Poisson structure and cohomology groups.

The affine model of hyper-elliptic Jacobian is interesting for its application to
integrable models. The ring A that we introduced in the previous section can be
supplied with Poisson structure. This fact is also important because introducing
the Poisson structure is the first step towards the quantization. The Poisson
structure in question is described in r-matrix formalism as follows:

{m(z1 ) ⊗ I, I ⊗ m(z2 )} = [r(z1 , z2 ), m(z1 ) ⊗ I] − [r(z2 , z1 ), I ⊗ m(z2 )]. (12)

The r-matrix acting in C2 ⊗ C2 is

z2 1 3
⊗ σ 3 + σ + ⊗ σ − + σ − ⊗ σ + + z2 σ − ⊗ σ − ,


r(z1 , z2 ) = 2σ
z1 − z2
where σ 3 , σ ± are Pauli matrices.
The variables z1 , · · · , zg (zeros of b(z)) and yj = a(zj ) have dynamical mean-
ing of separated variables [4]. The Poisson brackets (12) imply the following
Poisson brackets for the separated variables:

{zi , yj } = δi,j zi .

The determinant f (z) of the matrix m(z) generates Poisson commutative

subalgebra:
{f (z1 ), f (z2 )} = 0.
It can be shown that the coefficients f1 , f2 , · · · , fg and f2g+1 belongs to the
center of Poisson algebra. The Poisson commutative coefficients fg+1 , · · · , f2g
are the integrals of motion. Introduce the commuting vector-fields

Di h = {fg+i , h}, i = 1, · · · g.

For completeness let us describe explicitely the action of these vector-fields on

m(z). Define
Xg
D(z) = z j−1 Dg+1−j .
j=1

Then the Poisson brackets (12) imply:

1
D(z1 )m(z2 ) = [m(z1 ), m(z2 )] − [σ − m(z1 )σ − , m(z2 )].
z1 − z2

10
 One can think of these commuting vector-fields as Dj = ∂τ∂ j where τj are
”times” corresponding to the integrals of motion fg+j . The ”times” τj are
coordinates on the Jacobi variety, they are related to w as follows

τ = 21 M w,

where M is the matrix defined in (2). We remark that Di here coincides with
−2Di in the Mumford’s book [1]vol.II. Earlier we have introduced a gradation on
the ring A. We can prescribe the degrees to the vector-fields Dj as deg (Dj ) =
j − 12 because it can be shown that:
1
Dj A(p) ⊂ A(p+j− 2 ) .

Consider the differential forms

fi1 ··· ik dτi1 ∧ · · · ∧ dτik , (13)

with fi1 ,··· ,ik ∈ A. These forms span the linear spaces Ck for k = 0, · · · , g. The
differential
X g
d= dτj Dj ,
j=1

acts from Ck to Ck+1 . As usual applying d we first apply the vector fields Dj
to the coefficients of the differential form and then take exterior product with
dτj . We have the complex
d d d d d
0 −→ C0 −→ C1 −→ · · · −→ Cg−1 −→ Cg −→ 0.

The k-th cohomology group of this complex is denoted by H k (C∗ ). Consider

the problem of grading of the spaces Cj . Clearly we have to prescribe the degree
to dτj as
deg(dτj ) = −j + 21
in order that d has degree zero.
Consider the spaces Ck0 spaned by (13) with fi1 ··· ik ∈ A0 . Since the ele-
ments of F are “constants” (commute with Di ), Di acts on A0 . So, we have
the complex C∗0 :
d d d d d
0 −→ C00 −→ C10 −→ · · · −→ Cg−1
0 −→ Cg0 −→ 0.

This complex is graded. One easily calculates that

 
1 2 2 g
ch(Cg−j
0 ) = q 2 (j −g ) ch (A0 ) , (14)
j

where the q-binomial coefficient is defined as

 
g [g]!
= .
j [j]! [g − j]!

11
The differential d respects the grading. In this case the q-Euler characteristic
can be introduced

χq (C∗0 ) = ch C00 − ch C10 + · · · + (−1)g ch (Cg0 ) ,

(15)

which possesses all the essential properties of the usual Euler characteristic.
Using the formula (14) one finds
1 2  2g−1 
χq (C∗0 ) = (−1)g q − 2 g 2 ! ch (A0 ) (16)
1 2 [2g + 1]! [ 12 ]
= (−1)g q − 2 g .
[g + 12 ] [g]! [g + 1]!

Consider the cohomology groups H k (C∗0 ). The vector spaces H k (C∗0 ) inherit
a grading from Cj0 . Then

χq (C∗0 ) = ch(H 0 (C∗0 )) − ch(H 1 (C∗0 )) + · · · + (−1)g ch(H g (C∗0 )).

The q-number in (16) has finite limit for q → 1:

(2g)! 
2g


2g
lim χq (C∗0 ) = (−1)g = (−1)g


g − g−1 . (17)
q→1 (g)! (g + 1)!
Certainly the fact that the q-Euler characteristic has a finite limit does not mean
that cohomology groups are finite-dimensional, but we believe that this is the
case. So, we put forward

Conjecture 1. The spaces H k (C∗0 ) are finite-dimensional.

More explicitly the cohomology groups will be discussed in the next section.
In the situation under consideration there is an important connection be-
tween the algebra A and the highest cohomology group H g (C∗0 ).

Proposition 3. Consider some homogeneous representatives of a basis of the

space H g (C∗0 ):
hα dτ1 ∧ · · · ∧ dτg .
Arbitrary x ∈ A0 can be presented in the form
X
x= Pα (D1 , · · · Dg )hα , (18)
α

where Pα (D1 , · · · Dg ) are polynomials in D1 , · · · , Dg with C-number coefficients.

Proof. For x ∈ A0 construct

Ω = x dτ1 ∧ · · · ∧ dτg ∈ Cg0 .

By the definition of cohomology group we have

Ω = Ω0 + dΩ′ , Ω0 ∈ H g (C∗0 ), Ω′ ∈ Cg−1

0 ,

12
which implies that X
x=h+ Di xi ,
with h such that Ω0 = h dτ1 ∧ · · · ∧ dτg , xi ∈ A0 . Apply the same procedure
to xi and go on along the same lines. The resulting representation (18) will be
achieved in finite number of steps for the reason of grading. QED.

Let us introduce the notation

D = C [D1 , · · · , Dg ].
We shall call the expressions of the type (18) the D-descendents of {hα }. The
interesting question concerning the formula (18) is whether such representation
is unique for any x. The answer is that it is not the case, and to understand
why it is so we have to return to the formula (16) which can be rewritten as
follows:
1 2 1
q − 2 g ch (A0 ) =   ch(H g (C∗0 )) −
g − 21 !
1 1
−  ch(H g−1 (C∗0 )) +   ch(H g−2 (C∗0 )) − · · · . (19)
g − 21 ! g − 12 !
Obviously, the first term in the RHS represents the character of the space of
all D-descendents of {hα } (recall that the degree of Dj equals j − 21 ). This
is equivalent to saying that the first term has the same character as the space
generated freely over D by H g (C∗0 ):
1 g ∗ g ∗ g ∗


1 ch(H (C0 )) = ch(D)ch(H (C0 )) = ch D ⊗C H (C0 ) .
 
g− 2 !
The existence of the second term of the RHS of (19) implies that, in A0 , there
are linear relations among D-descendents of {hα } and they are parametrized by
the second term. The third term explains that there are relations among linear
relations counted by the second term of the RHS of (19) and so on. This is
nothing but the usual argument of constructing a resolution of a module. In
the present case it is actually possible to construct a free resolution of A0 as a
D module assuming some conjectures. The construction of the free resolution
is given in Appendix F.
Combining Proposition 2 and Proposition 3 one arrives at

Proposition 4. Let {hα } be the same as in Proposition 3. Then every x ∈ A

can be presented as
X
x= Pα (D1 , · · · , Dg ) hα , (20)
α

where Pα (D1 , · · · , Dg ) are polynomials in D1 ,...,Dg with coefficients from F.

Proof. We shall prove the proposition by the induction on the degree of x. Since
A(p) = {0} for p < 0, the beginning of induction obviously holds. Suppose that

13
the proposition is true for all elements of degree less than deg(x). By Proposition
2, there exist xj such that
2g+1
X
x = x0 + fi xi , xj ∈ A0 ,
i=1

where deg(x) = deg(x0 ) and deg(xi ) = deg(x) − i < deg(x) for i > 0. By
Proposition 3, there exist polynomials Pα (D1 , · · · , Dg ) with the coefficients in
C such that
X 2g+1
X
x0 = Pα (D1 , · · · , Dg )hα + fi yi , yi ∈ A,
i=1

where deg yi = deg x0 − i. Since deg(yi ) < deg(x0 ), x can be written in the form
(20) by the induction hypothesis. QED.

Proposition 4 represents the most important result of this paper. The possi-
bility of presenting every algebraic function on the phase space of the integrable
model in the form (20) starting from finite number of functions {hα }, which are
representatives of the highest cohomology group, is important both in classical
and in quantum case. The description of null-vectors follows from the one given
above because Di commute with fi .

6 Conjectures on cohomology groups.

In the previous section we have seen that the cohomology group H g (C∗0 ) is
important for describing the algebra A. This cohomology group is rather exotic,
since the complex C∗0 corresponds to the case when the algebraic curve X is
singular, that is, y 2 = z 2g+1 . In this section we first discuss the relation between
H k (C∗0 ) and the singular cohomology groups of the non-singular affine Jacobi
variety J(X) − Θ. For a set of complex numbers f 0 = (f10 , · · · , f2g+1 0
) we set
2g+1
X
Af 0 = A ⊗F Cf 0 , Cf 0 = F/ F(fi − fi0 ),
i=1

0 2g+1
and f (z) = z In the case when all fi0 = 0, Af 0 = A0 .
+ f10 z 2g + · · ·+ f2g+1
0
.
2 0
If the curve X: y = f (z) is non-singular, Af 0 is isomorphic to the affine ring of
J(X) − Θ. Since d commutes with F, the complex (C∗ , d) induces the complex
(C∗f 0 , d), where
X
Ckf 0 = Ck ⊗F Cf 0 = Af 0 dτi1 ∧ · · · ∧ dτik .
i1 <···<ik

Recall that

g ∗ Cg g
Cgf 0
H (C ) = , H (C∗f 0 ) = .
dCg−1 dCg−1
f0

14
Thus, tensoring Cf 0 to the exact sequence
d
Cg−1 −→ Cg −→ H g (C∗ ) −→ 0,

we have
H g (C∗ ) ⊗F Cf 0 ≃ H g (C∗f 0 ),
for any f 0 . By Proposition 4 we have
X
H g (C∗ ) = FΩα ,
α

where {Ωα } are representatives of H g (C∗0 ) in Cg . In other words there is a

surjective map of F-modules:

F ⊗C H g (C∗0 ) −→ H g (C∗ ).

We conjecture that this map is in fact injective. In general we put forward the
following conjecture.

Conjecture 2. (1) H k (C∗ ) is a free F-module for any k.

(2) H k (C∗ ) ⊗F Cf 0 ≃ H k (C∗f 0 ) for any k and f 0 .

Notice that Conjecture 2 implies, in particular, that

H k (C∗ ) ≃ F ⊗C H k (C∗0 ).

It is known, by the algebraic de Rham theorem (cf. [6]), that, if X is non-

singular,
H k (C∗f 0 ) ≃ H k (J(X) − Θ, C),
where the RHS is the singular cohomology group of J(X) − Θ. Thus we have

Corollary of Conjecture 2.There is an isomorphism:

H k (C∗0 ) ≃ H k (C∗f 0 ),

for any f 0 . In particular, for any non-singular hyper-elliptic curve X,

H k (C∗0 ) ≃ H k (X(g) − D, C).

Notice that Conjecture 1 follows from Conjecture 2 because the singular

cohomology groups of a non-singular affine variety are finite-dimensional. We
shall comment more on Conjecture 2 later, for the moment let us concentrate
on the singular cohomology groups of X(g) − D for a non-singular X.
Consider the affine curve Xaff = X − {∞} and its symmetric powers:
n
Xaff (n) = Xaff /Sn .

15
Since Xaff is affine and connected,

H p (Xaff , C) = 0, p ≥ 2, dim H 0 (Xaff , C) = 1.

The cohomology group H 1 (Xaff , C) is 2g dimensional and it is generated by

dz
µj = z g−j , j = −g + 1, · · · , g,
y
in the algebraic de Rham cohomology description of H 1 (Xaff , C). On H 1 (Xaff , C)
there is a skew-symmetric bilinear form:
 Z p 
λ1 ◦ λ2 = resp=∞ λ1 (p) λ2 .

Canonical basis νj , j = −g + 1, · · · , νg , with respect to this form, is defined as

one satisfying
4
νi ◦ νj = δi+j,1 .
j−i
A particular example of such basis is given in Appendix B.
As in the case of the compact curve X, the cohomology groups of the sym-
metric products Xaff (n) is described as the Sn invariants, H ∗ (Xaff (n), C) ≃
H ∗ (Xaff
n
, C)Sn (cf. (1.2) in [7]). If we define
(1) (n)
µ̃i = µi + · · · + µi , (21)
k
(k)
µi = 1 ⊗ ···⊗ µ̆i ⊗ · · · ⊗ 1 ∈ H ∗ (Xaff , C)⊗n ,

then they generate the cohomology ring H ∗ (Xaff (n), C). Obviously

H 1 (Xaff (n), C) ≃ Cµ̃−g+1 ⊕ · · · ⊕ Cµ̃g ,

k
^
H k (Xaff (n), C) ≃ H 1 (Xaff (n), C). (22)

Recall that D = D0 ∪ D∞ . Obviously, Xaff (g) = X(g) − D∞ . Hence there is

a map from H k (Xaff (g), C) to H k (X(g) − D, C). In the Appendix B we prove
the following:

Proposition 5. Consider the natural map:

i′
H k (Xaff (g), C) −→ H k (X(g) − D, C).

The kernel of this map is described as follows:

Ker(i′ ) = ω ∧ H k−2 (Xaff (g), C),

where
g
1
P
ω= 4 (2k − 1) ν̃k ∧ ν̃−k+1 . (23)
k=1

16
We remark that ω does not depend on the choice of the canonical basis {νj }.
By Proposition 5 the map i′ induces an injective map:

W k := H k (Xaff (g), C)/ ω ∧ H k−2 (Xaff (g), C) ֒→ H k (X(g) − D, C). (24)

We can make W k a graded vector space by prescribing the degrees to differential

forms as
deg(µ̃j ) = −j + 21 , deg(ω) = 0.
From (22) one easily finds:

ch(W k ) = Rk − Rk−2 ,
 
1 2g
Rk = q 2 k(k−2g) .
k

Now we put forward the strong conjecture that the map (24) is in fact
surjective and the character of W k defined here coincides with the character of
H k (C∗0 ):

Conjecture 3. (1) W k ≃ H k (X(g) − D, C) for 0 ≤ k ≤ g.

(2) ch(W k ) = ch(H k (C∗0 )).

What does it mean? The divisor D consists of D∞ and D0 . The forms from
W k describe the part of cohomology groups with singularities on D∞ only. Our
conjecture is that this part exhausts the whole space of the cohomology groups,
i.e. that adding exact forms one can move singularities of any form from D0
to D∞ . This is a strong statement which is rather difficult to prove. The first
non-trivial case is g = 3 for which we were able to prove Conjecture 3. The
details of it will be published elsewhere.
For k = 1 we can prove Conjecture 3 (1) for any g. The proof is given in
Appendix D.
Let us now present a simple calculation which shows that Conjectures 2, 3
are consistent with the calculation of q-Euler characteristic of C∗0 . Indeed

ch(W 0 ) − ch(W 1 ) + ch(W 2 ) − · · · + (−1)g ch(W g )

=(R0 ) − (R1 ) + (R2 − R0 ) − (R3 − R1 ) + · · · + (−1)g (Rg − Rg−2 )
=(−1)g (Rg − Rg−1 ) = χq (C∗0 ).

This calculation was actually the starting point for Conjectures 2, 3. Certainly it
does not prove anything, but it shows remarkable consistence between different
calculations performed in this paper.
In order to understand the magic of the hyper-elliptic case it is instructive
to compare it with the case of an Abelian variety in generic when the divisor
Θ is non-singular (which rarely the case for Jacobians of algebraic curves). As

17
explained in the Appendix C in the latter case the following can be proven:
W k ≃ H k (J − Θ, C), k ≤ g − 1,
W g ֒→ H g (J − Θ, C).
Actually for g ≥ 3 the space H g (J − Θ, , C) is bigger than W g , the difference of
dimensions being
(2g)!
dim H g (J − Θ, C) − dim W g = g! − .
g!(g + 1)!
We would conjecture that the equality H g (J − Θ, C) ≃ W g specifies hyper-
elliptic Jacobians.

7 Appendix A. Proof of Proposition 1

We have to determine the basis of A0 . Recall that
2g+1
X
A0 ≃ A/ fj A.
j=1

Write the equations f1 = · · · = f2g+1 = 0 explicitly:

X X
ck + b i cj + ai+ 21 aj+ 12 = 0, 1 ≤ k ≤ g + 1, (25)
i+j=k,j6=k i+j=k−1
X X
b i cj + ai+ 12 aj+ 21 = 0, g + 2 ≤ k ≤ 2g + 1. (26)
i+j=k i+j=k−1

From (25) ck (1 ≤ k ≤ g + 1) can be solved by bi , aj . It is sometimes convenient

to use the following notation;
m
um
g
1
+1
· · · ug+g 1 = [−m1 , · · · , −mg ],
2 2

B0 = C [u1 , u 23 , · · · , u g+1 ].
2

We use both this bracket notation and the uj notation to denote a monomial.
The ring B0 is considered as a coefficient in the sequel.
Let us first prove

Proposition A. In the ring A0 any monomial [−m1 , · · · , −mg ] can be written

as a linear combination of monomials of the form [−n1 , · · · , −ng ], n1 , · · · , ng =
0, 1 with the coefficient in B0 .

Proof. Define the degree of [−m1 , · · · , −mg ] as that of the monomial:

g
X g+1+k
deg[−m1 , · · · , −mg ] = mk .
2
k=1

We define a total order on the set of monomials {[−m1 , · · · , −mg ]} by the

following rule. Let P = [−m1 , · · · , −mg ], P ′ = [−m′1 , · · · , −m′g ].

18
1. If deg(P ) < deg(P ′ ), then P < P ′ .
2. If deg(P ) = deg(P ′ ), compare P and P ′ by the lexicographical order from
the left.
Notice that the product of two elements P = [−m1 , · · · , −mg ] and P ′ =
[−m′1 , · · · , −m′g ] is expressed as

P P ′ = [−(m1 + m′1 ), · · · , −(mg + m′g )].

The following property obviously holds.

Lemma A 1. For monomials P1 , P2 , P3 , if P1 < P2 then P1 P3 < P2 P3 .

From (25)

ck = −uk + · · · , 1 ≤ k ≤ g + 1,

where · · · part does not contain uk . Then from (26)

u2k = · · · , g + 2 ≤ k ≤ 2g + 1. (27)
2

The next lemma describes what kind of monomials appear in the right hand
side of (27).

Lemma A2. The right hand side of (27) is a linear combination of elements
of the form ul , ul um , ul um un with the coefficients in B0 .

Proof. It is sufficient to show that the term like xui1 · · · uir , r ≥ 4, x ∈ B0 does
not appear in the expression. Since (25), (26) are homogeneous, if x 6= 0 and
homogeneous, then

deg(xui1 · · · uir ) = k,

where we take into account the degree of x. Since i1 , · · · , ir ≥ g/2 + 1 and

2g + 1 ≥ k,
g 
deg(xui1 · · · uir ) ≥ deg(ui1 · · · uir ) = i1 + · · · + ir ≥ 4 + 1 > k.
2
Thus x = 0. QED.

Lemma A3. In each of the cases in Lemma A2 we have the following state-
ments, where x is a homogeneous element in B0 .
1. If u2k/2 = xul + · · · , then u2k/2 > ul .

2. If u2k/2 = xul um + · · · , then u2k/2 > ul um .

3. If u2k/2 = xul um un + · · · , then u2k/2 > ul um un .

19
Proof. 1. Since deg(u2k/2 ) = k ≥ g + 2 > g + 1/2 ≥ l = deg(ul ),the claim follows.
2. If k > l + m, there is nothing to be proved. Suppose that k = l + m. Then
l < k < m. Thus comparing by the lexicographical order we have u2k/2 > ul um .
The statement of 3 is similarly proved. QED.

Starting from any element P = [−m1 , · · · , −mg ] we shall show that P can
be reduced to the desired form. If some mj ≥ 2, then rewrite it using (27).
By Lemma A3 every term in the resulting expression is less than P . Repeating
this procedure we finally arrive at the linear combinations of [−n1 , · · · , −ng ],
n1 , · · · , ng = 0, 1 with the coefficients in B0 . Thus Proposition A is proved.
QED.

By Proposition A we have
g g  1
1 2 [2g + 2]! ch(A)
Y Y g+1+k

ch(A0 ) ≤  1+j  1+q 2 =  = . (28)
j=1 2 k=1
g + 12 ! [g]! [g + 1]! ch(F)

Thus from (10) we conclude

ch(A)
ch(A0 ) =
ch(F)
which completes the proof of Proposition 1. QED.

8 Appendix B. Proof of Proposition 5.

We define W k by the LHS of (24):
W k = H k (Xaff (g), C)/ ω ∧ H k−2 (Xaff (g), C) .

We first show that the map

i′ : H k (Xaff (g), C) −→ H k (X(g) − D, C)
satisfies
i′ ω ∧ H k−2 (Xaff (g), C) = 0

and thereby it induces the map

i′ : W k −→ H k (X(g) − D, C).
Next we shall construct a subspace Wk of the homology group Hk (X(g) − D, C)
such that the pairing between Wk and i′ (W k ) is non-degenerate. This proves
Proposition 5.
Let us study the properties of the differential form ω defined in (23). Con-
sider some differentials λj from H 1 (Xaff , C), j = 1, · · · , k − 2, and construct the
g-form:
 
Ω = d κ̃ ∧ λ̃1 ∧ · · · ∧ λ̃k−2 (29)

20
where the one form κ̃ is given by
 
1 yi −yj dzj
X
κ̃ = κ(ij) , κ(ij) = 4 zi −zj
dzi
yi + yj
i<j

The form under d in RHS of (29) belongs to Cfk−1 , that is, it has singularity on
the divisor D = D0 ∪ D∞ , but after the differential is applied the singularities
on D0 disappear. Indeed, one easily shows that
g  
(j) (i) (i) (j)
dκ(ij) = 1
P
4 (2k − 1) νk ν−k+1 − νk ν−k+1 , (30)
k=1

where νj are defined as

dz
νj = qj (z) , j = −(g − 1), · · · , g − 1, g,
y
and qj is the following polynomial of degree g − j:
!
y1 z1−j √
qj (z) = resp1 =∞ d z1 .
z1 − z

The differentials νj are normalized at infinity as

√
νj ∼ −2 z −j + O(z −g−1 ) d z for p → ∞.

The differentials νj for j = 1, · · · , g are holomorphic and νj for j = −g +1, · · · , 0

are of the second kind. It is easy to verify that
4
νi ◦ νj = δi+j,1 .
j−i
This means, in particular, that, for two cycles γ1 , γ2 on X
g
Z R 
dκ(12) = 14
P R R R
(2k − 1) γ1 ν−k+1 γ2 νk − γ2 ν−k+1 γ1 νk
γ1 ×γ2 k=1
= γ1 ◦ γ2 , (31)

due to Riemann bilinear relation, where γ1 ◦ γ2 is the intersection number. This

is an important property of dκ(ij) .
The equation (30) means that
 
Ω = d κ̃ ∧ λ̃1 ∧ · · · ∧ λ̃k−2 = −ω ∧ λ̃1 ∧ · · · ∧ λ̃k−2

where we have to remind that

g
1 P
ω= 4 (2k − 1) ν̃k ∧ ν̃−k+1 .
k=1

21
  
This proves that i′ ω ∧ H k−2 (Xaff (g), C) = 0.
Consider the homology groups of Xaff (g). Taking dual to the relation (22)
we obtain a similar relation for the homology groups:
k
^
Hk (Xaff (g), C) ≃ H1 (Xaff (g), C).

The first homology group H1 (Xaff (g), C) is isomorphic to H1 (Xaff , C). To

have an element δ̃ from H1 (Xaff (g), C) one takes a cycle δ from H1 (Xaff , C) and
symmetrizes it over g copies of Xaff . In other words, if we fix a point p0 in Xaff ,
then

δ̃ = δ (1) + · · · + δ (g) ,
⊗(i−1) ⊗(g−i) g
δ (i) = p0 ⊗ · · · ⊗ δ ⊗ · · · ⊗ p0 ∈ H0 ⊗ H 1 ⊗ H0 ֒→ H1 (Xaff , C),

where Hj = Hj (Xaff , C).

There is an obvious embedding X(g) − D into Xaff (g). It induces a map
between the homology groups
i
Hk (X(g) − D, C) −→ Hk (Xaff (g), C). (32)

The meaning of this map is simple: every cycle on X(g) − D is at the same time
a cycle on Xaff (g). There are two subtleties:
1. Nontrivial cycle on X(g) − D can be trivial on Xaff (g) i.e. the map (32) can
have kernel.
2. On Xaff (g) there are cycles that intersect with D0 which means that they are
not cycles on X(g) − D, so, the map (32) can have cokernel.
Let us study the image of the map (32). A k-cycle from Hk (Xaff (g), Z) is a
linear combination of elements of the form:

∆ = δ̃1 ∧ · · · ∧ δ̃k ,

where δj ∈ H1 (Xaff , Z). The product

∆′ = δ1 × · · · × δk × p0 × · · · × p0
g g
defines an element of Hk (Xaff , Z). Let π be the projection map Xaff −→
g
Xaff (g) and π∗ the induced map on the homology groups, π∗ : H k (X aff , Z) −→
Hk (Xaff (g), Z). Then ∆ = k! kg π∗ (∆′ ) in Hk (Xaff (g), Z).

Thus the cycle ∆ belongs to Im(i) if ∆′ does not intersect π −1 (D). Recall
that D = D0 ∪ D∞ . By construction ∆′ has no intersection with π −1 (D∞ ).
One easily realizes that ∆′ does not intersect with π −1 (D0 ) iff

δi ◦ σ(δj ) = 0 ∀ i, j.

It is rather obvious property of the hyper-elliptic involution that

δi ◦ σ(δj ) = −δi ◦ δj .

22
Hence we come to the following

Proposition B. The Im(i) contains linear combination of cycles

∆ = δ̃1 ∧ · · · ∧ δ̃k , δ1 , · · · , δk ∈ H1 (Xaff , Z),

such that
δi ◦ δj = 0 ∀ i, j.

Let Wk denote the space obtained as C-linear span of cycles in Hk (X(g) − D, Z)

corresponding to ∆’s in this proposition.
Take a canonical cycles αi , βj and set

Ai = α̃i , Ai+g = β̃i , 1 ≤ i ≤ g.

Define

VZ = H1 (Xaff , Z) = ⊕2g
i=1 ZAi , VC = H1 (Xaff , C) = ⊕2g
i=1 CAi .

The symplectic form on VC is defined by

Ai ◦ Aj = ±δj,i±g .

By definition

i(Wk ) = SpanC (Uk ), Uk = {γ1 ∧ · · · ∧ γk |γ1 , · · · , γk ∈ VZ , γi ◦ γj = 0 ∀i, j}.

Define

W̃k = SpanC (Ũk ), Ũk = {γ1 ∧ · · · ∧ γk |γ1 , · · · , γk ∈ VC , γi ◦ γj = 0 ∀i, j}.

Consider the map

ϕk : ∧k VC −→ ∧k−2 VC ,
X
ϕk (γ1 ∧ · · · ∧ γk ) = (−1)i+j−1 (γi ◦ γj )γ{ij} , (33)
i<j

where γ{ij} is obtained from γ1 ∧ · · · ∧ γk removing γi and γj . It is known

that, for k ≤ g, ϕk is surjective and its kernel Kerϕk is isomorphic to the k-th
fundamental irreducible representation of Sp(2g, C) (cf. Theorem 17.5 [9]). In
particular

dk := dim Ker(ϕk ) = 2g 2g



k − k−2 .

The following lemma can be easily proved.

Lemma B.Suppose that k ≤ g. Then

i(Wk ) = W̃k = Ker(ϕk ).

23
 As a consequence of the lemma one has in particular

dim Wk ≥ dim i(Wk ) = 2g 2g

k − k−2 . (34)

Let us show that Wk and i′ (W k ) pairs completely. The pairings

< , >1 : Hk (X(g) − D, C) ⊗ H k (X(g) − D, C) −→ C (35)

and

< , >2 : ∧k H1 (Xaff (g), C) ⊗ ∧k H 1 (Xaff (g), C) −→ C (36)

are related by

< γ, i′ (η) >1 =< i(γ), η >2 ,

for γ ∈ Hk (X(g) − D, C), η ∈ ∧k H 1 (Xaff (g), C). The pairing (36) is given by
the integral:

< γ̃ ∧ · · · ∧ γ̃k , η̃1 ∧ · · · ∧ η̃k >2

Z 1 Z
g


= η̃1 ∧ · · · ∧ η̃k = k! k det( ηj )1≤i,j≤k .
γ̃1 ∧···∧γ̃k γi

By (31) we have

< i(Wk ), ω ∧k−2 H 1 (Xaff (g), C) >2 = 0.

Thus the pairing (35), (36) induce pairings

< , >1 : Wk ⊗ i′ (W k ) −→ C (37)

k
< , >2 : i(Wk ) ⊗ W −→ C. (38)

Since (36) is non-degenerate and dim i(Wk ) = dim W k , the pairing (38) is
non-degenerate. It easily follows from this that the pairing (37) is also non-
degenerate. Thus we have proved Proposition 5. QED.

Corollary B. We have Wk ≃ i(Wk ). In particular

dim Wk = 2g 2g



k − k−2 .

9 Appendix C. The case of generic Abelian va-

riety.
Let (J, Θ) be a principally polarized Abelian variety such that Θ is non-singular.
Then

24
Proposition C. The dimensions of cohomology groups of J − Θ are given by
   
k 2g 2g
dimH (J − Θ, C) = − , k ≤ g − 1,
k k−2
   
2g 2g (2g)!
= − + g! − , k = g,
g g−2 g!(g + 1)!
= 0, k > g.

Proof. Consider the inclusions Θ ⊂ J ⊂ (X, J) and the induced homology exact
sequence:

· · · −→ Hk (Θ, C) −→ Hk (J, C) −→ Hk (J, Θ) −→ Hk−1 (Θ, C) −→ · · · .

Taking the dual sequence of this and using the Poincare-Lefschetz duality we
get

· · · −→ Hk−1 (Θ, C) −→ Hk (J − Θ, C) −→ Hk (J, C) −→

−→ Hk−2 (Θ, C) −→ · · · . (39)

Since J − Θ is affine
Hk (J − Θ, C) = 0, k > g.
Then we have

Hk (Θ, C) ≃ Hk+2 (J, C), k ≥ g, Hk (Θ, C) ≃ Hk (J, C), k ≤ g − 2.

It is easy to check that, for k ≤ g, the dual map of

Hk (J, C) −→ Hk−2 (Θ, C)

is given by wedging the fundamental class [Θ] of Θ:

[Θ]∧ : H k−2 (Θ, C) ≃ H k−2 (J, C) −→ H k (J, C). (40)

Using the representation theory of sl2 as in the proof of the hard Lefschetz
theorem (c.f.[6]), the map (40) is injective for k ≤ g. Thus by (39) the following
exact sequences hold:
[Θ]∧
0 → H k−2 (Θ, C) −→ H k (J, C) → H k (J − Θ, C) → 0, k < g,
[Θ]∧
0 → H g−2 (Θ, C) −→ H g (J, C) → H g (J − Θ, C) →
→ H g−1 (Θ, C) → H g+1 (J, C) → 0.

Proposition C follows from these exact sequences and the fact

χ(Θ) = (−1)g−1 g!.

QED.

25
 By Proposition F 3 the fundamental class of Θ coincides with ω in Proposi-
tion 5 in the hyper-elliptic case. Thus, if we define W k in a similar formula to
(24), we have

W k ≃ H k (J − Θ, C), k ≤ g − 1,
W g ֒→ H g (J − Θ, C).

10 Appendix D. The proof of Conjecture 3 for

k=1
Notice that

H k (Xaff (g), C) ≃ ∧k H 1 (X, C) ≃ ∧k H 1 (J(X), C),

and X(g) − D ≃ J(X) − Θ. In particular W 1 ≃ H 1 (J(X), C).

Proposition D. For any principally polarized Abelian variety (J, Θ) such that
Θ is irreducible we have the isomorphism

H 1 (J, C) ≃ H 1 (J − Θ, C).

Proof. Following [8] we shall use the following notations:

O(nΘ) : the sheaf of meromorphic functions on J which have poles only on Θ
of order at most n,
O(∗Θ) : the sheaf of meromorphic functions on J which have poles only on Θ,
Ωk (nΘ) : the sheaf of meromorphic k-forms on J which have poles only on Θ
of order at most n,
Ωk (∗Θ) : the sheaf of meromorphic k-forms on J which have poles only on Θ,
Φk (nΘ) : the sheaf of closed meromorphic k-forms on J which have poles only
on Θ of order at most n,
Φk (∗Θ) : the sheaf of closed meromorphic k-forms on J which have poles only
on Θ,
Rk (nΘ) = Φk (nΘ)/d Ωk−1 ((n − 1)Θ) , Rk (∗Θ) = Φk (∗Θ)/d Ωk−1 (∗Θ) .

In particular
Ω0 (nΘ) = O(nΘ), Ω0 (∗Θ) = O(∗Θ).
We first recall the description of H 1 (J, C) in terms of the differentials of the
first and second kinds. Consider the sheaf exact sequence:
d

0 −→ C −→ O(∗Θ) −→ d O(∗Θ) −→ 0.

26
Since
H k (J, O(∗Θ)) = 0, k ≥ 1,
we have

H 1 (J, C) ≃ H 0 (J, dO(∗Θ))/dH 0 (J, O(∗Θ)). (41)

The numerator in the right hand side of (41) is nothing but the space of differ-
ential one forms of the first and the second kinds on J and the denominator is
the space of globally exact meromorphic one forms.
On the other hand, by the algebraic de Rham theorem, the first cohomology
group of the affine variety J − Θ is described as

H 1 (J − Θ, C) ≃ H 0 (J, Φ1 (∗Θ))/dH 0 (J, O(∗Θ)). (42)

Comparing (41) and (42) what we have to prove is

H 0 (J, dO(∗Θ)) ≃ H 0 (J, Φ1 (∗Θ)). (43)

Consider the exact sequence

0 −→ dO(∗Θ) −→ Φ1 (∗Θ) −→ R1 (∗Θ) −→ 0.

The cohomology sequence of this is

0 −→ H 0 (J, dO(∗Θ)) −→ H 0 (Φ1 (∗Θ)) −→ H 0 (R1 (∗Θ)) −→ · · · .

From this what should be proved is that the map

H 0 (Φ1 (∗Θ)) −→ H 0 (R1 (∗Θ))

is a 0-map. To study this map we refer the lemma from [8].

Lemma D.(Lemma 8 [8])

(1) R1 (∗Θ) ≃ CΘ , where CΘ is the constant sheaf on Θ and the isomorphism
is given by
dθ
[ ] ←− [1Θ ]
θ
at any stalk.
(2) R1 (∗Θ) ≃ R1 (nΘ), n = 1, 2, · · · .

In the proof of Lemma D (1) we use our assumption that Θ is irreducible.

Using this lemma we reduce the problem from ”∗Θ” to ”nΘ” with finite n.
Consider the exact sequence

0 −→ dO(nΘ) −→ Φ1 ((n + 1)Θ) −→ R1 (∗Θ) −→ 0, n = 0, 1, · · · . (44)

27
From the cohomology sequence of it we have the map

H 0 (J, R1 (∗Θ)) −→ H 1 (J, dO(nΘ))

which we denote by πn . Let us prove

Kerπn = 0, n = 0, 1, 2, · · · .

To this end we study H 1 (J, dO(nΘ)). Using the exact sequence

d
0 −→ C −→ O(nΘ) −→ dO(nΘ) −→ 0,

we easily have

H 1 (J, dO(nΘ)) ≃ H 2 (J, C), n ≥ 1, (45)

1 2
H (J, dO) ֒→ H (J, C). (46)

The natural maps

dO(nΘ) −→ dO((n+ 1)Θ), Φ1 (nΘ) −→ Φ1 ((n+ 1)Θ), R1 (nΘ) ≃ R1 ((n+ 1)Θ),

and the sequence (44) induce a commutative diagram of cohomology groups. It

follows from this commutative diagram and (45), (46) that Kerπ0 = 0 implies
Kerπn = 0 for n ≥ 1. Now Kerπ0 = 0 follows from

H 0 (J, dO) ≃ H 0 (J, Ω1 ), H 0 (J, Φ1 (Θ)) ≃ H 0 (J, Ω1 ).

The second isomorphism follows from the fact that a meromorphic function on
J which has poles only on Θ of order at most one is a constant. Thus the
Proposition D is proved. QED.

11 Appendix E.
0
Recall that, for a set of complex numbers f 0 = (f10 , · · · , f2g+1 ), the ring Af 0 is
defined by
A
Af 0 = A ⊗F Cf 0 = P2g+1 ,
i=1 A(fi − fi0 )

where fi is the coefficient of f (z). If all fi0 = 0, then Af 0 = A0 . The ring A0 is

graded while Af 0 is not graded unless all fi = 0. Instead Af 0 is a filtered ring
for any f 0 . Let Af 0 (n) be the set of elements of Af 0 represented by ⊕k≤n A(n) .
Then

Af 0 = ∪n≥0 Af 0 (n), Af 0 (0) = C ⊂ Af 0 12 ⊂ Af 0 (1) ⊂ · · · .

Consider the graded ring associated with this filtration:

Af 0 (n)
grAf 0 = ⊕grn Af 0 , grn Af 0 = .
Af 0 (n − 12 )

28
Since deg(fi0 ) = 0, grAf 0 becomes a quotient of A0 . In other words there is a
surjective ring homomorphism
A0 −→ grAf 0 . (47)

We shall prove that this map is injective.

Proposition E. There is an isomorphism of graded rings:

A0 ≃ grAf 0 .

Proof. Notice that the map (47) respects the grading and it is surjective at each
grade. By Proposition 2, A ≃ F ⊗C A0 as a C-vector space. Thus we have
A0 ≃ Af 0 as a C-vector space for any f 0 . In particular the basis of A0 given in
(n)
Proposition 1 is also a basis of Af 0 . Denote by {xj } the basis of the degree
(n) (n)
n part A0 . Let us prove that {xj } are linearly independent in grAf 0 by
the induction on n. For n = 0 the statement is obvious. We assume that the
(m)
statement is true for all m satisfying m < n. This means that {xj } is a basis
(m)
of the degree m part of grAf 0 for all m < n. In particular {xj |m < n} is a
basis of Af 0 (n − 21 ). Suppose that the relation
(n)
X
αj xj ∈ Af 0 (n − 12 ).
j

(m)
holds, where some αj 6= 0. Then this means that {xj |m ≤ n} are linearly
(k)
dependent in Af 0 . This contradicts the fact that {xj } is a basis of Af 0 . Thus
(n)
{xj } are linearly independent in grAf 0 . QED.

12 Appendix F. Construction of a free resolu-

tion of A0
Recall that
D = C [D1 , · · · , Dg ]
is the polynomial ring generated by the commuting vector fields D1 , · · · , Dg .
Then A, A0 and Af 0 are D modules. We shall construct a free D-resolution
of A0 assuming Conjecture 2 and 3. To avoid the notational confusion we shall
describe the space W k using the abstract vector space V of dimension 2g with
a basis vi , ξi (1 ≤ i ≤ g):

V = ⊕gi=1 Cvi ⊕gi=1 Cξi .

Set
g
X
ω= vi ∧ ξi ∈ ∧2 V
i=1

29
and define
∧k V
Wk = . (48)
ω ∧k−2 V
Assign degrees to the basis elements by

deg(vi ) = −(i − 21 ), deg(ξi ) = i − 21 .

Then W k defined by (48) has the same character as W k defined in Section 6.

This justifies the use of the same symbol. Consider the free D-module D ⊗C W k
generated by W k . We shall construct an exact sequence of D-modules of the
form
d d d
0 ← A0 dτ1 ∧ · · · ∧ dτg ← D ⊗C W g ← D ⊗C W g−1 ← · · · ← D ⊗C W 0 ← 0.

Define the map

d
D ⊗C ∧k V −→ D ⊗C ∧k+1 V,

by
g
X
d(P ⊗ vI ∧ ξJ ) = Di P ⊗ vi ∧ vI ∧ ξJ ,
i=1

where for I = (i1 , · · · , ir ), vI = vi1 ∧· · ·∧vir etc., P ∈ D and Di P is the product

of Di and P in D. Since the map d commutes with the map taking the wedge
with Q ⊗ ω for any Q ∈ D:
g
X
d(QP ⊗ ω ∧ vI ∧ ξJ ) = QDi P ⊗ ω ∧ vi ∧ vI ∧ ξJ = (Q ⊗ ω) ∧ d(P ⊗ vI ∧ ξJ ),
i=1

it induces a map
d
D ⊗C W k −→ D ⊗C W k+1 .
It is easy to check that the map d satisfies d2 = 0.

Proposition F 1. The complex

d d d
0←−D ⊗C W g ←− D ⊗C W g−1 ←− · · · ←− D ⊗C W 0 ←− 0 (49)

is exact at D ⊗C W k except k = g.

Proof. Notice that the following two facts:

1. The following complex is exact at D ⊗ ∧k V except k 6= g:
d d d
0 ←− D ⊗C ∧g V −→ · · · ←− D ⊗C V ←− D ←− 0. (50)

2. The map ω∧ : ∧k V −→ ∧k+2 V is injective for k ≤ g − 2.

30
 The property 1 is the well known property of the Koszul complex. The
property 2 is also well known and easily proved using the representation theory
of sl2 .
Now suppose that x ∈ D ⊗C ∧k V , k < g and dx = ω ∧ y for some y ∈
D ⊗C ∧k−2 V . Then ω ∧ dy = 0 and thus dy = 0 by the property 2. Then y = dz
for some z by the property 1. Thus we have d(x − ω ∧ z) = 0 and x = dw + ω ∧ z
for some w again by the property 1. QED.
Next we shall define a map from D ⊗C W g to A0 dτ1 ∧ · · · ∧ dτg . To this end
we need to identify W k in this section and that of Section 6, which is defined as
the quotient of the cohomology groups of a Jacobi variety. For this purpose we
describe the cohomology groups of a Jacobi variety in terms of theta functions.
Define
∂
ζi (w) = Di log θ(w) = log θ(w).
∂τi
Then, for each i, the differential dζi (w) defines a meromorphic differential
form on J(X) which has double poles on Θ and which is locally exact. This
means that dζi is a second kind differential on J(X). By (41) first and second
kinds differential one forms define elements of H 1 (J(X), C). The pairing with
H1 (J(X), C) is given by integration. The following proposition can be easily
proved by calculating periods.

Proposition F 2.The first and the second kinds differentials dτ1 ,...,dτg ,
dζ1 ,...,dζg give a basis of the cohomology group H 1 (J(X), C).

Thus we can identify V with H 1 (J(X), C) ≃ H 1 (Xaff (g), C) by

vi = dτi , ξi = dζi .

The next proposition can be easily proved by calculating integrals over two
cycles in a similar way to [1](vol.I, p188).

Proposition F 3. Let ω be defined in Proposition 5. Then we have

g
1 X
√ dτi ∧ dζi = ω
2π −1 i=1

as elements in ∧2 H 1 (J(X), C). Moreover ω represents the fundamental class of

the theta divisor Θ in H 2 (J(X), C).

From this proposition it is possible to identify W k in this section and that

in the previous sections. Define the map

∧g V → Af 0 dτ1 ∧ · · · ∧ dτg ,
vI ∧ ξJ 7→ dτI ∧ dζJ ,

31
where for I = (i1 , · · · , ir ), dτI = dτi1 ∧ · · · ∧ dτir etc. and f 0 is any set of
complex numbers such that y 2 = f 0 (z) defines a non-singular curve X.
This map extends to the map of D modules
D ⊗C ∧g V → Af 0 dτ1 ∧ · · · ∧ dτg ,
in
P the KfollowingKmanner. Let P ∈ D and consider P ⊗ (vI ∧ ξJ ). Write dζJ =
K FJ dτK , FJ ∈ Af 0 . Then we define
X
P ⊗ (vI ∧ ξJ ) −→ P (FJK )dτI ∧ dτK .

Since, as a meromorphic differential form on J(X),

g g
X X ∂ 2 log θ(w)
dτi ∧ dζi = dτi ∧ dτj = 0,
i=1 i,j=1
∂τi ∂τj

the above map induces the map

ev
D ⊗C W g −→ Af 0 dτ1 ∧ · · · ∧ dτg .
Denote by (D ⊗C W g )n the subspace of elements with degree n and by evn
the restriction of ev to (D ⊗C W g )n . By the definition, for x ∈ (D ⊗C W g )n ,
evn (x) ∈ Af 0 (n + g 2 /2). By Proposition E, there is a natural isomorphism
A0 ≃ grAf 0 (see Appendix E for the filtration of Af 0 ). Composing the map
(k)
evn with the natural projection map Af 0 (k) → grk Af 0 ≃ A0 we obtain the
map, which we denote by evn too,
ev
(D ⊗C W g )n −→
n
(A0 dτ1 ∧ · · · ∧ dτg )n ,
where the RHS means the degree n subspace. Taking the sum of evn we finally
have the map
ev = ⊕n evn : D ⊗C W g −→A0 dτ1 ∧ · · · ∧ dτg ,
which we also denote by the symbol ev.
If we assume Conjecture 2 and 3, H k (C∗0 ) ≃ W k . If this holds, then ev is
surjective by Proposition 3.

Lemma F. Suppose that Conjecture 2 and 3 are true. Then the kernel of ev is
given by
Ker(ev) = d D ⊗ W g−1 .

Proof. It is easy to check that

Ker(ev) ⊃ d D ⊗ W g−1 .

Since ev is surjective and

 D ⊗C W g 
ch(A0 ) = ch
,
d D ⊗C W g−1

32
the claim of the lemma follows. QED.
We summarize the result as

Theorem F. Suppose that Conjecture 2, 3 are true. Then the following complex
gives a resolution of A0 dτ1 ∧ · · · ∧ dτg as a D-module:
ev d d
0←−A0 dτ1 ∧ · · · ∧ dτg ←− D ⊗C W g ←− · · · ←− D ⊗C W 0 ←− 0.

References
[1] D. Mumford, Tata Lectures on Theta, vol. I and II, Birkhäuser,
Boston (1983)
[2] O. Babelon, D. Bernard, F.A. Smirnov, Comm. Math. Phys. 186
(1997) 601
[3] F.A. Smirnov, J. Phys. A: Math. Gen. 31 (1998) 8953
[4] E.K. Sklyanin, Separation of Variables, Progress in Theoretical
Physics Supplement 118 (1995) 35
[5] A. Beauville, Acta Math. 164 (1990) 211
[6] P. Griffiths and J. Harris, Principles of Algebraic Geometry, A
Wiley-Interscience publication (1978)
[7] I.G. Macdonald, Topology 1 (1962) 319
[8] M. Atiyah and W.V.D. Hodge, Ann. of Math. 62 (1955) 56
[9] W. Fulton and J. Harris, Representation Theory, Springer, New
York (1991)

33