Mathematical Physics: L. A. Takhtajan and P. G. Zograf
Mathematical Physics: L. A. Takhtajan and P. G. Zograf
Mathematical Physics: L. A. Takhtajan and P. G. Zograf
Communications
Mathematical
Physics
Springer-Verlag 1991
Introduction
The Atiyah-Singer index theorem for families of elliptic operators, which plays
an important role in modern mathematical physics, is of particular interest for
<3-operators on complex manifolds. Consider a holomorphic family p : 3C > B of
compact complex manifolds over a compact base B, and a holomorphic vector
bundle $ > 9C. The family d = {db}beB o 3-operators in the vector bundles
E\j > Xb (restrictions of $ over the fibers Xb = p~{b), b B) gives rise (in
the sense of iC-theory) to the index bundle ind G K(B) on B with fibers
kerf, cokerk over b e B. The Atiyah-Singer index theorem applied to this
special case states that
ch(ind d) = p* (ch td T).
(1)
Here ch denotes the Chern character, td T9 is the Todd class of the vertical
tangent (along the fibers of p : 9C B) bundle on S, and p* : H*(9C)
H*~imXb(B)
is the operation of "integration along the fibers" (see [1]).
In many applications the bundles and TV9 are Hermitian, so that each of
them carries the (canonical) unitary connection compatible with the holomorphic
400
1/2;
(2)
here || || is the ordinary ZAnorm in det ind d induced by the metrics in S and
TX9 d*db is the Laplace operator acting on sections of Eb -> Xb, and det d*/> is
its zeta function determinant regarded as a function on B. In [2] Quillen studied
in detail the family of all Cauchy-Riemann operators (i.e. holomorphic structures
in a Hermitian vector bundle E X on a compact Riemann surface X). In this
case B is an infinite dimensional complex affine space, and the curvature form of
the Hermitian line bundle (det ind d; || \\Q) on B appears to be equal, up to a
constant multiple, to the natural Kahler form on B.
A similar result was obtained somewhat later by Belavin and Knizhnik [3] (we
follow [4] in exposing of their result). Denote by B = Tg the Teichmller space
of compact Riemann surfaces of genus g, by 2 Zg -> Tg - the Teichmller
universal curve, by Tv2g -> 2g - the vertical line bundle of the fibration
p : 3g Tg9 and by dk - the family of 3-operators acting on /c-differentials on
Riemann surfaces (sections of (T~k3g)b Xb). Then for the first Chern form
of the determinant line bundle det ind dk endowed with Quillen's norm (2) the
following formula holds:
d (det ind~dk)= J (ch(Tv-k^g)
td(T^g))2,2,
(3)
fiber
where ( )2,2 denotes the (2,2)-component of a differential form on ^g9 and the
integration is taken over the fibers of &~g -> T g . As one can easily see, formula
(3) is a specification of (1) on the level of (1,1)-forms. Moreover, if we consider
a metric in Tv^~g which coincides on each fiber of Zg Tg with the Poincare
metric (i.e. Hermitian metric of constant curvature 1), then by a result of
Wolpert [5] formula (3) can be rewritten as
ci (det ind dk) =
^-^
Wp,
(4)
where p is the Weil-Petersson Kahler form on Tg. Such a form of local index
theorem is analogous to Quillen's original result. On the other hand, formula (4)
can be derived by the methods of Teichmller theory (see [6]) avoiding a difficult
heat kernel technique. The approach of [6] also allows us to prove a local index
theorem for families of 5-operators in stable bundles of rank n and degree k on
401
a compact Riemann surface [7] (which provides, for instance, an analytic proof
of projectivity of the corresponding moduli spaces for coprime n and k).
Quillen's local index theorem admits various deep generalizations (see [8-10]
for details), but in all of them the fibers of p : 9C > B are compact manifolds.
Otherwise considerable difficulties occur because of continuous spectrum of the
corresponding Laplace operators (along with purely technical difficulties there
is a problem how to define properly the determinant of the Laplace operator).
However, the methods of [6] work for families of -operators on non-compact
Riemann surfaces as well. The first examples of a local index theorem for
families with non-compact fibers were given in our papers [11,12]. Namely, we
have considered the case when B = Tg,n is the Teichmller space of Riemann
surfaces of type (g, n), i.e. of genus g with n punctures (cusps), 9C ^~gl is the
corresponding universal family (so that the fibers of the fibration p : 2%^n Tg>n
are Riemann surfaces of type (g, n)), and $ = T~k2g;n is the kth power of the
vertical line bundle on ^~^n (the case k = 0,1 was treated in [11], and the case
k > 2 in [12]). In this situation the Laplace operator k = d*dk associated with
the Poincare metric (i.e. complete Hermitian metric of constant curvature 1) on
a Riemann surface X of type (g, ) has n-fold continuous spectrum. To define a
regularized determinant of k consider the Selberg zeta function Z(s) which is
given for Re s > 1 by the absolutely convergent product
e-^^)9
(5)
{/} m=0
where runs over the set of all simple closed geodesies on X with respect
to the Poincare metric, and |^| is the length of /. The function Z(s) admits a
meromorphic continuation to the whole complex s-plane with a simple zero at
s = 1. For compact Riemann surfaces it was shown in [13] that the determinant
of k defined via its zeta function is equal, up to a constant multiplier depending
only on g and /c, to Z'(l) for k = 0,1 and Z(k) for k > 2. Similarly, for Riemann
surfaces of an arbitrary type (g, n) we define
JZ'(l),
* = 0,l,
k>2.
(6)
Using this definition, we_ can calculate the first Chern form of the determinant
line bundle k = detindfc on Tg,n endowed with Quillen's metric (2). The result
differs from (4) by an additional term in the right-hand side:
2
6/c -6/c + l
1
Q CUSp,
(7)
where wp is the Weil-Petersson Kahler form on Tg>n and CUSp is the symplectic
form of a new Kahler metric (, ) c u s p on TgjW(n 0).
We proceed with exact definitions. Let X be a Riemann surface of type (g, )
equipped with the Poincare metric and let be a torsion-free Fuchsian group
uniformizing X, i.e. X = \H, where H = {z e C | Im z > 0} is the upper halfplane. Denote by \, ..., n the set of non-conjugate parabolic subgroups in ,
and for every i = l
n f i x a n element oi e PSL(2,R) such that ^it = ^
where the group ^ is generated by the parabolic transformation z -> z + 1. The
402
v,
{>1
2)Q9
,v ~1(X)9
i = 1, . . . , n.
(8)
Each scalar product (,) f gives rise to a Kahler metric on Tg>w (see Sect. 2 below).
Their sum
1=1
{ = a(k) - J
(10)
* fiber
403
f(yz)y'(z)'(z) =f{z)9
(meZ),
(forms of weight (2/, 2m) correspond to tensors of type (/, m) on the Riemann
surface X = \H). Let (z)\dz\2 denote the Poincare metric y~2(dx2 + dy2) on
the upper half-plane H. We denote by ^f ^m the Hubert space of automorphic
forms of weight (2/, 2m) with the natural scalar product
(1.1)
\H
2\dx
dyj
dz
2 \dx
dy J
404
/(tz)(z)' = 4
i = 1, ...,
0,
t <-1,
= 0,
dime Of =
' = 1,
(1.2)
d*,
1
^ y
(1.3)
(s)2
(2s)
z-z'
(1.4)
1- z z
z z'
z z'
where F(a,b9c;z) is the hypergeometric function and (s) is the gamma function.
At s = 1 one has
zz
x
z-z'
Without writing an explicit expression for Qf* with t < 1 we give a simple
formula
z-z>
\z-z
(1.5)
405
ef(^^),
(1.6)
which admits term-by-term differentiation with respect to the variables z and z'.
The kernel Gf\z,zf) with z z'9 y e , admits a meromorphic continuation
in s to the entire complex s-plane and has the following Laurent expansion at
s = l:
(see [15, Theorem 2.3]). The kernel Gf\z,z') is called the Green's function of the
Laplacian A$ on the Riemann surface X = \H. One has (see [15, p. 161])
- 8^
= - t e m?
Gf)M
= M
'
(L7)
where (z9zr) is the so-called Schiffer_kernel. It is defined as a symmetric bidifferential of the second kind o n x with a double pole of biresidue 1 at the
diagonal z = z1 and the property
v.p. (z, zf) (z') dx1 dy' = 0
\H
for every e . The Schiffer kernel does not depend on a marking of X (i.e.
on a choice of generators of the Fuchsian group ). Moreover, the following
formula holds [15, p. 160]:
g
(1.8)
/=
= 2v 4 I (z),
i = 1, ..., g;
COJ(Z')
dzr = j9
i, j = 1, ..., g ,
406
= / j( ')
dz
i, 7 = 1, ..., g,
>
the period matrix of the marked Riemann surface X. It has a symmetric positive
definite imaginary part Im = (Im i ; ) (which enters in (1.8)) with the property
Im l 7
/ coiCOj,
1, . . . , g .
ij=
x
The Green's function of the Laplacian Af with { < 1 is defined simply as
Gf\z,zf);
'
- Q{f](z,zf)).
(1.10)
oz'
y> oo
k=-oo
zy
dz'l('
{z-z>\z-z>
k=oo
407
y>oo
near each cuspzy, j = 1, ..., n (see [15,16]). For any , v e " 1 1 set / ^ =
(Jo + 5)" 1 (v) ^ The result below will play an important role in the proof
of Theorem 1 of Sect. 4.
Lemma 2. The function f has the following asymptotics near the cusp z\ of the
group :
c(0
fv (, z) = + exponentially decreasing terms as y oo,
t = 3 / ^i(s
= 1, ..., n.
Proof. Since , v G ""1'1, one has = y2, v = j; 2 for some cusp forms ,
e 2 , and hence the function v ^f is exponentially decreasing at the cusps
zi, . . . , zn. Let
2\
408
{z e F \ lm( z)
<
\H
=1
G " u (X),
G 2(X).
x
ljl
t h e r e e x i s t s a u n i q u e d i f f e o m o r p h i s m f :H-+H
dz
satisfying t h e Beltrami e q u a t i o n
dz
409
neighbourhoods of two points [X] and [X] the Bers coordinates transform complex analytically. The differential of this coordinate change at the point [X] 6 Tg>M
is a linear map D : ~1(X) -> ~l(X),
(-. - =
\d
dvJ
, 2)v,
i=
(f)*()=of(-j-\
-i-J
T,
=0
(2.1)
410
for any G " 1 * 1 ^ ) (see [17]). For the second variation of the following
formula was obtained in [5]:
d2
d\ d&2
= 2 =0
= \ (0 + i ) - 1 (v) = \Q
v
, v "1-1 (X).
/
1
(2.2)
field - on Tgj in a neighbourhood of the point [X], are given by the formulas:
_i
_ i -
'
(2.3)
L p y = 0,
U j = 1, ..., g.
V (Im)^
( 2 . 4 )
dlogdetlm=
x
>
^logdetlm
7
,
B(f(z), f(zf))^P=0
1 1
d z
(z)d-Pr (z') = 0
d z
411
( 1 5 )
= 0,
(, v ) = - \ (Ao + \)~X{v) =
-\fv- jefm'(X)
of linear
d
LA
~T
=0
((f
=0
to ^'m(X).
Ldi = 5* + 1 ,
Ldt = 0,
d^d^+9
.. (2.1)
These formulas will be used in the next section. Here we prove the following
Lemma 3. The metrics (,),-, i = 1, ..., n, on Tg>n are Khlerian.
Proof. We must show that
( v , )i = (, )i
(2.8)
^ - (v,>i = A ^ ^ , ^ ) ,
2)v(I),
x x
2) + -d(dEi(
2)) = 0,
(2.9)
412
By means of this formula and formula (2.3) we obtain from (2.9) integrating by
parts
/ - (v,)i = - (Ao + \r{Q-d{dEit
d
J
2)))' v
since d(v) = 0 for any v ~ (X) and since Beltrami differentials and v
are rapidly decreasing at the cusps. The obtained formula is obviously symmetric
with respect to and v, which proves (2.8) and the lemma.
Setfor, v- u (A),
dv)
The metric (,)CUSp is obviously invariant under the action of the Teichmller
modular group Modg>n.
Finally, let us observe that the scalar products (,v) and (,v), , i = 1,..., n,
for , v " 1 ' 1 ^ ) can be expressed in terms of the values of the Rankin L-series
associated with the cusp forms = y~2, y~2v e 2(X). Indeed, let
k}kL\9 {^}^i> i = 1, ..., n, be the Fourier coefficients of the cusp forms and
of weight 4 at the cusps z\,..., zn (see Sect. 1). Then for Res > 1 (see [19]),
i( ,
\H
if
JJ
oo
V
J7f.
-1
'
) d x2d y
y
oo 1
^ JJ
o o
9+2
413
where
is the Rankin L-series for the cusp forms and at the cusp z, , i = 1,..., . It
is absolutely convergent for Re s > 4 and has a simple pole at s = 4. Therefore,
from (2.10) it follows that
(,v)i = 5 L(,v; 5);
i = 1, ..., n.
(2.11)
Moreover, since
Res \s=Ei(z, s) =
2(2g-2
+ n)'
'
(4^3 ( * -
+ ^)
R e s
m=0
where *f runs over the set of all simple closed geodesies on X (with respect to the
Poincare metric), and |/| is the length of . The function Z(s) has a meromorphic
continuation to the whole s-plane with a simple zero at s = 1. For the logarithmic
derivative of Z(s) one has
\H
x
'
hyperbolic
where the sum is taken over all hyperbolic elements of the Fuchsian group
uniformizing X; Res > 1. This formula can be derived from the definition of
Z(s) by means of the Selberg transform; see [16] for details.
As it follows from the Teichmller theory (see Sect. 2), the value of the Selberg
zeta function Z(s) at a fixed point 5 with Res > 1 is a smooth function on Tg>n.
The next lemma gives an expression for the first derivatives of Z(s) with respect
to coordinates on TgjM.
Lemma 3. For any e ~>(X) and Res > 1 the following formula holds
(3.2)
414
where
- <0))M*) = ^ -
(Gf (z,z') -
Qf\z,z'))\z,=z
(here and in what follows a dash on an operator means that it acts on the variable zf).
Proof. First of all, ddf(Gf^ QP)\D is a smooth automorphic form of weight
4 for the Fuchsian group , and the integral in the right-hand side of (3.2) is
convergent because decreases rapidly near the cusps zu ..., zn of the group .
Further differentiating both sides of (3.1) and taking (2.1) into account we obtain
that for R e s > 1,
L
2^I "(^
lgZ(s)
-UN
\H
. hyperbolic
,yz'))
yer,
\H
y parabolic
where
Q?](f(z)J(z'))
=0
and
GT(f(z)J(zf))
LG?\z,z') =
=0
s
(G stands for the resolvent of the Laplace operator Ao on the Riemann surface
X = \H). Denote by Gs4(z,zf) the resolvent kernel of the Laplace operator
on the Riemann surface i\H; for Res > 1,
yen
From the definition of the resolvent and formula (2.7) for LAo it follows that
zf) = - v.p. JJ
? W ) ^7 Q?)(z",z')dx"dy", z z',
, z") {L,)"GSj{z\ zf)
-^
>) dx df9
yz>,
415
and
LGV>(z,z') = -.j>.JJ
\H
z yz\ y e ,
\H
/f
where {LA0) means that the differential operator L0 acts in the variable z".
Using now the above expressions for LQf\ LGSj, LGf^ and a simple formula
(GSti(z,z')-Q?\z,z')),
i=
ye,
y parabolic
-Mi
(z)dxdy
d2
dzdz'
z'=z
\H
\H
-
1
(z) dx dy
\H
zz'
z'=z
x L \Gf\z,z') - Qf\z,z') -
US
er,\
_
L [[
~ -2s ds\HJJ dzdz
x (Gf\z,z')~Qf\z,z')-
(z
yer,
y parabolic
=Z
(z) dx dy
z'z
hyperbolic
(in this calculation we reversed the order of integration and applied the Hilbert
identity to the resolvent kernels). Now let us integrate the last formula over the
416
d2
-o
s>oo
y hyperbolic
uniformly in z e H, we get
(z) dx dy.
z'=z
y hyperbolic
Since
dz dzf
II
y parabolic
where the sum is taken over all parabolic elements of the group . We have
yer,
z'=z
parabolic
=r^
n
oo
-
-
z'z
(0)
where
k=oo
kO
From formula (1.4) for the kernel Qf\z,z') it is not difficult to deduce that s(z)
is a bounded function on H, depending only on y = Imz. On the other hand,
417
(3.4)
where
(see Sect. 1). The first derivatives of Z(s) on Tg?n for integer 5 = 2, 3, ... can
be expressed in a similar way in terms of the corresponding Green's functions.
Namely, we have
Lemma 4. For any integer k > 1 and ~{^(X)
^ - ^ - ^ ' ( G H O - H ^ .
Proo/ First we will prove that for / = k< 1, Res> 1,
| fl = JdQ+WlG&V - Q^l%
(3.5)
418
We have
= J((Qd' + f!d^d(<(Gf
- f ))\D
- Q<f> )| D ) = 0 ,
because for any e ~lfl(X) one has d() = 0; this proves (3.5). Now starting
from Lemma 3 with s = k + 1 and applying formula (3.5) k times we arrive at
the assertion of the lemma.
dv
= - Tr (((-v + (1 - k) (LV)-)I
+ (Ld^k)lk
(L-v~dlk))P^k) ,
419
\(Z (l))-^\\
k=L
In the next theorem we compute the first Chern form c\(k, || \\Q) of the Hermitian
holomorphic line bundle (k, \\ ' \\Q) on Tg,n.
Theorem 1. For k > 1,
ci(k,
||
||) =
^ 2
w p - ^ C U S p,
(4.3)
where wp *s the Weil-Peter sson Khler form and c u s p is the Khler form of the
metric (,)CUsP.
Proof Since
/c>2,
\ g l
2^
ds dv
6m2 + 6 m + 1 I,I I
.\
\h'>> I
III
,
n
y\
\A*>v/cuspj
/A
(A \
v+ v
12
9
where for m = 0 Z(m + 1) should be replaced by Z ; (l). In the main, the proof of
this formula follows the proofs of Theorem 2 in [6] and Theorem 2 in [11]. By
Lemma 4,
where
-logZ(m+l) = -
{ m)
(R - \D
U (R(-m)
(4.5)
420
O
Ql
dz'
z
(Z Z)
'
(
(z-zV\z-z>
therefore
(4.7)
Because the variation of the kernel dmd'Q{\~n{) contributes a finite amount
to (4.5) the same is true for the kernel dmd'G^m\ which allows us to vary it
outside the diagonal D, and then we can pass to the limit as z' z. The kernel
dmdfG{]~m) for m > 0 is the kernel of the operator -md*_mAzld!m
from Jf""1'1
1/n
to JtT . As it is shown in [6, Sect. 4.5],
: = - Tr((-v J
For m 0 we have
uu
CJ
==
2,
where 2 is the Schiffer kernel (see (1.7)). Therefore using (1.8) we obtain
U{dd'Gf)\D
x
L-J\B(Z,Z')
(z)dxdy
z'z
^iCjL,
where we have also used (2.4) and the fact that LVB = 0 (see Sect. 2). By formula
(2.3) the last integral vanishes identically and we obtain that
(dd'Gf]) \D = - -^r-
log det Im .
(4.9)
7 , i = l , ...,n}
and
Cy
= F n { z e H
421
\D~d-l~d(o +
dz
\r{v)dxdy
d z
dz
where fV = (zlo + \)~(v) (see Lemma 1). For the integral I\ we have, due to
-invariance of the integrand, that
lim Y
-
z =x + v^y.
Using Lemma 1 and 2 we obtain that
*"
_^(,v)cusp.
(4.10)
i=l
Let us now proceed with the integral h First of all, for m = 0 we have h = 0
since R^\D is holomorphic. In this case combining formulas (4.5), (4.7), (4.9), and
(4.10), we obtain that
d2 . _
d2 , . .,
. 1
logZ'(l) = r ^ logdet Im + _ (,
( // v)
v ) - - (, v) c u s p ,
which proves (4.4) for m = 0. For m > 1 we observe first that
s *-w> -
422
y-
IJ h
The boundary integral h can be calculated in the same way as I\. We have
lim V
Y-
dx,
z = x
= y- 2m / 7 (
k=co
1 '(z)
C 1
(z + fc)fc z (z) + -z + -T,
z2
and the Stirling formula. Taking now Lemma 2 into account we conclude that
/3=0.
In order to calculate the integral U we observe that
d
The kernel
m)
C7 OZ
= I P-m\ (see
formula (1.2)) in the space ^f"" ' , where / is the identity operator, and P_m,i is
(-m){z
423
m
onto ~ (X).
,} = 2 m + l
^_
Therefore
2m
= U = \ JJ (-y- P-mz,Z)
x
P)
(2m+l)m
+
(4.11)
+
where we used formula (2.2) and the formula
which follows from the equality Ao(Ao + \)~ = / \ (o + j)'1. Combining the
results of computations (formulas (4.7), (4.8), (4.10), (4.11)) we obtain that
8 2
- logZ(m + 1) = Tr((-v/ + (L^_ w )Jli(L,3! w ) d dv
mQ-{L
6m2 + 6m + 1 7 x
. x
W v) - - (, v ) c u s p .
Finally setting m = fc 1 and taking Lemma 5 into account we arrive at the
assertion of the theorem, i.e. formula (4.3).
5. Concluding Remarks
Here we will calculate the cuspidal defect
? = cx(k9 || || ) - J (ch(T-k^g,n)
fiber
(5.1)
Moreover,
K)ir
for any , v ~1'(X),
Sect. 2).
424
Proof. By definition,
k
rg<n)
td(Tv<g,n)
where is the curvature form of the Poincare metric in the line bundle
3~gn (see Sect. 2). Using (2.5) we have
fiber
X
/
2 /
VQ - -7T
z
Jx
ifv)Q
(5.3)
%v),
~
where we also used Lemma 2 and Green's formula to make sure that the last
integral in (5.3) vanishes identically. Rewriting (5.3) in the form
2
J \ 2 "J "
fib
fiber
1
2
(cf. Corollary 5.11 in [5]) and taking (4.3) into account we obtain (5.1). Formula
(5.2) follows now from (5.1) and (2.11).
Note that the cuspidal defect
4 0 ) = dimind, - J (ch(T-krg/)
td(Tg/))1
fiber
J
Finally we present some algebraic geometry consequences of Theorem 1. First
of all, because all bundles and metrics on g?n considered here are invariant under
the action of the Teichmller modular group Mod g s n , formula (4.3) holds also on
the moduli space J^n = T g j n / Mod g>n (in the sense of orbifolds). Consider the
universal curve ^ g = Jgi\ and denote by the relative dualizing sheaf on ^ g ?
i.e. the line bundle dual to the vertical tangent bundle (along fibers of projection
p : ^ g -> Jfg) on # g . Further, let us denote by [ W p], [ c u s p ] e # 2 ( ^ g , I R ) = 1R2
the cohomology classes of the closed (1, l)-forms wp and c u s p on # g . Theorem
425
1 means that for the first Chern class c\(k) of the line bundle k on (g the
following formula holds:
C\{k) =
^2
(5.3)
is the
= p* (fa)
on ^ g (here fa denotes the determinant line bundle detind2 on Jig). This
formula together with Mumford's isomorphism fa = \3 on Mg [14] and with
the fact that \ = p*(fa) yields the isomorphism
2=\3.
(5.4)
(5-5)
1
4
i.e. ~2 [COWP], T [ c u s p ] G H2(^g9 ) are integral cohomology classes. In particular,
it follows from (5.5) that for any compact Riemann surface X of genus g > 2,
imbedded into ^g as a fiber a projection p : (g ^ g , we have
4
2 / c u s p = ^2 / wp = 2g - 2.
Substituting (5.5) into (5.3), we obtain the formula
{k) = (6k2 - 6/c + l)ci(Ai) + ^ ^
(),
Cl
(5.6)
on ^ g , because according to Harer's result for g > 3 the Picard group Pic(^ g ) =
2
) is isomorphic t o Z Z [20]. The isomorphism (5.6) is analogous to
g 9
Mumford's isomorphism k = \h2~k+{
on Jig (see [14]).
References
1. Atiyah, M.F., Singer, I.M.: The index of elliptic operators. IV. Ann. Math. (2), 93, 119-138 (1971)
2. Quillen, D.: Determinants of Cauchy-Riemann operators over a Riemann surface. Funk. Anal, i
Prilozen. 19 (1), 37-41 (1985) (in Russian); Funct. Anal. Appl. 19, 31-34 (1985)
3. Belavin, A.A., Knizhnik, V.G.: Complex geometry and the theory of quantum strings. JETP 91,
364-390 (1986) (in Russian); Sov. Phys. JETP 64, 214-228 (1986)
426
4. Bost, J.B., Jolicoeur, T.: A holomorphy property and the critical dimension in string theory from
the index theorem. Phys. Lett. 147 B, 273-276 (1986)
5. Wolpert, S.: Chern forms and the Riemann tensor for the moduli space of curves. Inv. Math. 85
(1), 119-145 (1986)
6. Zograf, P., Takhtajan, L.: A local index theorem for families of d-operators on Riemann surface.
Usp. Mat. Nauk 42 (6), 169-190 (1987) (in Russian); Russ. Math. Surv. 42 (6), 169-190 (1987)
7. Zograf, P., Takhtajan, L.: On the geometry of moduli spaces of vector bundles over a Riemann
surface. Izv. Akad. Nauk SSSR, ser. Mat. 53 (4), 753-770 (1989) (in Russian); Math. USSR Izv.
35(1), 83-100(1990)
8. Bismut, J.-M., Freed, D.: The analysis of elliptic families. I. Metrics and connections on determinant bundles. Commun. Math. Phys. 106, 159-176 (1986); II. Dirac operators, eta invariants,
and the holonomy theorem. Commun. Math. Phys. 109, 103-163 (1986)
9. Gillet, H., Soule, C : Direct images of hermitian vector bundles. Bull. Am. Math. Soc. (N.S.) 15,
209-212 (1986)
10. Bismut, J.-M., Gillet, H., Soule, C.: Analytic torsion and holomorphic determinant bundles. IIII.
Commun. Math. Phys. 115, 49-78, 78-126, 301-351 (1988)
11. Takhtajan, L., Zograf, P.: The Selberg zeta function and a new Kahler metric on the moduli
space of punctured Riemann surfaces. J. Geom. Phys. 5 (4), (1988)
12. Zograf, P.: A local index theorem for families of d-operators on punctured Riemann surfaces.
LOMI preprint E-14-89, Leningrad 1989
13. Doker, E., Phong, D.: On determinants of laplacians on Riemann surfaces. Commun. Math.
Phys. 104, 537-545 (1986)
14. Mumford, D.: Stability of projective varietes. Lns. Math. 24, 39-110 (1977)
15. Fay, J.: Fourier coefficients of the resolvent for a Fuchsian group. J. Reine Angew. Math. 293/294,
143-203 (1977)
16. Venkov, A.: Spectral theory of automorphic functions. Trudy Mat. Inst. Steklov. 153 (1981) (in
Russian); Proc. Steklov Inst. Math. 153 (4) (1982)
17. Ahlfors, L.: Some remarks on Teichmller's space of Riemann surfaces. Ann. Math. 74, 171-191
(1961)
18. Rauch, H.: A transcendental view of the space of algebraic Riemann surfaces. Bull. Am. Math.
Soc. 71, 1-39 (1965)
19. Rankin, R.A.: Contributions to the theory of Ramanujan's function () andf similar arithmetic
functions. III. Proc. Cambridge Phil. Soc. 35, 351-356, 357-372 (1939)
20. Harer, J.: The second homology group of the mapping class group of an orientable surface. Inv.
Math. 72, 221-239 (1983)
Communicated by A. Jaffe