Commun. Math. Phys.

149, 463-493 (1992)

Communications in

Mathematical
Physics

9 Springer-Verlag 1992

The Pfaftian Line Bundle

David Borthwick*
Department of Physics, Harvard University, Cambridge, MA 02138, USA
Received October 8, 1991; in revised form February 7, 1992

Abstract. We analyze the holomorphic Pfaffian line bundle defined over an infinite
dimensional isotropic Grassmannian manifold. Using the infinite dimensional
relative Pfaffian, we produce a Fock space structure on the space of holomorphic
sections of the dual of this bundle. On this Fock space, an explicit and rigorous
construction of the spin representations of the loop groups LO. is given. We also
discuss and prove some facts about the connection between the Pfaffian line bundle
over the Grassmannian and the Pfaffian line bundle of a Dirac operator.

1. Introduction
In this paper, we study the Pfaffian line bundle P F over the isotropic Grassmannian manifold of a Hilbert space. This line bundle, which was first defined in [21], is
a unique holomorphic square root of the determinant line bundle over the
Grassmannian. Here we will use the theory of the infinite dimensional relative
Pfaffian developed in [13] and [-16] to construct a Hilbert space f f out of the space
of holomorphic sections of the dual bundle of PF. If we use the space of squareintegrable wave functions on the circle for the underlying Hilbert space, then f f is
interpreted as the Fock space of a Majorana fermion on the circle (with half the
degrees of freedom of the Dirac Fock space, which arises from the corresponding
construction for the determinant line bundle). The physical interpretation of this
Pfaffian line bundle Fock space construction was speculated on in [26], and it
serves as an example of the Fock space functor described in [24].
The Fock space ~- is isomorphic to the completion of an exterior algebra, but
the Pfaffian line bundle approach reveals extra structure. The isotropic Grassmannian Grx(Yt) is a homogeneous space of the restricted orthogonal group Ores(3eg)
associated to a real structure on the Hilbert space Jr. We show that f f carries
a projective unitary representation of Ores(Yg), which is an analytic generalization
of the representations described algebraically in [14]. In analogy to the Borel-Weil
* Supported by a National Science Foundation Graduate Fellowship

464

D. Borthwick

theorem in finite dimensions, this representation is seen to come from an action of

Oros(~) on holomorphic sections of P F ' which covers its action on G r t ( ~ ) . By
embedding the loop groups LO, in Ores(~'~ we obtain the spin representations of
LO, for all n. These representations were pointed out in [21]. We show here how to
construct them explicitly using the relative Pfaffian as an analytic tool.
By applying the Pfaffian line bundle construction to the space J f @ ~ ' , where
ovf' is the dual of ~ , we can obtain by a pullback the determinant line bundle for
~ . The determinant line bundle construction gives rise to representations of LU,,
which were described in detail in [21]. Obtaining the determinant line bundle by
pulling back the Pfaffian line bundle corresponds, in terms of representations, to
the embedding LU, ~ L02, ~ Ordovf). The representations of LU, arising from
this embedding are exactly those obtained from the determinant line bundle
directly.
Freed [12] has shown how to define a Pfaffian line bundle ~ over the moduli
space of compact Riemann surfaces with spin structure. This ~r is a holomorphic
square root of the Quillen determinant line bundle over moduli space [22, 7, 8],
a structure which has been extremely important in string theory and conformal
field theory (see [1, 11], for example). The moduli space of Riemann surfaces was
connected to soliton theory by Krichever [18], a relation which was applied to
string theory in [19]. A version of the Krichever map linking the moduli space to
the Grassmannian and its determinant line bundle, which is the type of map we will
consider here, appeared in [23, 25], and has also had applications to physics. In
particular, this connection has been used to connect Virasoro algebras with the
geometry of Riemann surfaces [5, 6, 17], which leads to the unification of the
geometric and algebraic approaches to conformal field theory [2, 3, 15]. In our
case, we would like use the Krichever map to think of the isotropic Grassmannian
manifold as universal moduli space for once-punctured Riemann surfaces with spin
structure. We show that the line bundles ~ and P F are related by a pullback by the
Krichever map. ~( comes with a canonical hermitian structure and holomorphic
section, which are relevant to the physical interpretation [1, 11]. One would like to
relate these to PF, which has its own canonical hermitian structure. It is fairly clear
how to choose a section on P F which gives rise to the canonical section of X under
the pullback. Unfortunately, the canonical metrics on P F and 3ff do not coincide,
and we show that it is not possible to choose a metric on P F which pulls back
correctly.
The basic objects of our discussion, the Grassmannian manifolds and the
corresponding restricted unitary and orthogonal groups, are defined using the
Hilbert-Schmidt norm, following [21]. One might ask if the Hilbert-Schmidt
restriction could be relaxed to some other Schatten ideal I v, or to the case of
compact operators (which was the definition used in [23, 25]). This is a significant
question if we want to consider higher dimensions, because for the group
Map (X, G), where X is some d-dimensional manifold, the relevant operators lie in
the class Ia+l (see [10, 20, 21]). Unfortunately, not much along the lines of the
construction of this paper can be done for the class Iv, where p > 2. In [20], it was
shown how to modify the definition of the determinant line bundle to extend to the
p > 2 cases in such a way that D E T has a hermitian structure which depends on
regularized determinants. Because the transition functions for P F given below in
Sect. 4 involve only finite dimensional matrices, it is trivial to do the same for the
Pfaffian line bundle. The hermitian structure for the Pfaffian cases is just a positive
square root of the hermitian structure given by [20], so that no extra analysis is

Pfaffian Line Bundle

465

required. Our construction here relies, however, on an inner product, or at least

some metric structure, on the space of holomorphic sections of the line bundle.
Because the regularized determinants do not satisfy multiplicative relations, one
cannot use the methods presented here to obtain such structures. Furthermore, it is
shown in [20] that the groups Map(X, G) admit only trivial cyclic extensions for
d > 1, so that these cases would inherently behave very differently from the d = 1
case considered here.
This paper is organized as follows. In Sect. 2 we briefly review the definition of
the Grassmannian manifold over a Hilbert space. In Sect. 3 we describe the
construction of the determinant line bundle, and the formation of a Hilbert space
out of the space of holomorphic sections. Section 4 contains the analogous
constructions in the Pfaffian case. We define the line bundle P F by giving a trivialization such that the transition functions are Pfaffians of finite-dimensional matrices. In particular, this shows that the Pfaffian line bundle is holomorphic over
the restricted Grassmannian. In Sect. 5 we describe in detail the Fock space arising
from the space of holomorphic sections, and give an alternative definition of PF.
We discuss the construction of the spin representation in finite dimensions in Sect.
6, as a prelude to the infinite dimensional case. In Sect. 7 we deal with the infinite
dimensional case in detail, discussing the action of the restricted orthogonal group
on PF, and the corresponding representations on the Fock space. We briefly
describe some of the applications of these representations to loop groups in Sect. 8.
In Sect. 9 we discuss the relations between the Pfaffian line bundle over the
Grassmannian and the Pfaffian line bundle over moduli space and present some
results connecting the two.

2. The Grassmannian Manifold

We start with a separable, polarized infinite dimensional Hilbert space, W. By
polarized, we mean simply that W comes with a decomposition into closed, infinite
dimensional subspaces,
= ~ - G oug+ 9

(2.1)

This Hilbert space is to be thought of as a one-particle Hilbert space, with

subspaces consisting of negative and positive energy states, respectively. For the
Pfaffian case, we will interpret ~ f as the space of states for a single fermion moving
on a circle, which we identify with the half-densities on the circle, subject to
antiperiodic boundary conditions.
We define a subgroup of the unitary group of oug for whose elements the
off-diagonal parts are restricted to be Hilbert-Schmidt,
Ures(~) := {g~ U (2/{~): P-gP+, P+gP- ~I2(~r

(2.2)

where P+ are the orthogonal projections onto Jg+. Note that this restriction on
the off-diagonal parts of an element g implies that the diagonal parts, P+ gP+
and P_ g P-, must be Fredholm, through the condition that g is invertible. The
Grassmannian manifold is a homogeneous space of the restricted unitary group.
G r ( ~ ) := { W c Yg' : W = gogt~ for some g~ Ure=(~)}

"~ Ures(~)/U(~/tr

x U(~+)

(2.3)

466

D. Borthwick

G r ( ~ ) breaks up into connected components indexed by the index of the

Fredholm operator P_gP_ on ~f~_. If 9if_ and ~r are taken to have finite
dimensions m and n, respectively, the definition of the Grassmannian reduces
simply to Grm(~'+"). One can form a nested set of finite dimensional submanifolds
Gr(k)(gff) of Gr(oVf), with
2k-

G r t k ) ( ~ ) ---- U Grm(CZk),

(2.4)

m=l

and the union of these submanifolds is dense [21].

This definition of the Grassmannian of a Hilbert space is essentially that of
[21]. As noted in the introduction, works relating the Grassmannian and determinant bundle to dynamical systems [23, 25] have used a broader definition, restricting off-diagonal terms to be compact operators where we have required
Hilbert-Schmidt. The determinant and Pfaffian line bundles are still well-defined
and holomorphic in the compact case, but they are not homogeneous, and one
cannot introduce a Hilbert space structure on the space of holomorphic sections of
the dual bundle in the manner described below.
To show that Gr(oUf) has the structure of a complex manifold [21], we define
a set of coordinate charts as follows. Choose an orthonormal basis { ek} k%1 for
~ + , and a basis {ek} ~2-o~ for ~f_. In our example of half-densities on the circle,
we will take e+k = exp{ + i(k -89
Think of this as a canonical basis fixed by
our parametrization of the circle. For S a subset of the non-zero integers, let Ws be
the subset of ~r given by the span of {ek}k~S. We have Wse Gr(~,ug) if and only if
S e d , where

~r

~,card(7l_\S)< ~ } .

(2.5)

It is straightforward to check that any element of Gr(oUf) can in fact be written as

the graph of some operator a sI2( Ws, W~), and conversely the graph of any such
operator is an element of Gr(~g). Thus we have an open cover of G r ( ~ ) by sets

Us := { W ~ G r ( ~ ) : W = graph(a), for A~I2(Ws, Ws~)} ,

(2.6)

each of which is isomorphic to I 2 ( ~ + , J r - ) . MQreover, the change of coordinate

maps determined by these isomorphisms involve only determinants of finite dimensional submatrices. They are thus clearly holomorphic, which gives G r ( J ~ ) the
structure of a complex manifold. We will generally identify the set of subspaces Us
with the set of maps 12( Ws, W~), with no distinction of notation.

3. The Determinant Line Bundle

Over the finite dimensional Grassmannians, the determinant line bundle can be
defined by taking the bundle whose fiber over a subspace is the top exterior power
of that subspace. In the infinite dimensional case, we will think of a trivialization of
D E T as given formally by the maps
A~

A (1 +

A)ek,

(3.1)

k~S

where A~12(Ws, W~). This would be a proper definition in finite dimensions,

but the infinite wedge product is unfortunately ill-defined. However, suppose

Pfaffian Line Bundle

467

W e Us c~ UR is given by graph(A) in Us, and by graph(B) in UR. The sets R and

S can differ only by a finite number of elements, and on the subspace WR S the
maps A and B must agree. Furthermore, if we write W = 9 ~ f _ , then
index(P_ a P - ) -- card(S c~ Z , o ) - c a r d ( Z < o \ S )
= card(R c~ 2g~o) - card(Tl<o\R ) ,

(3.2)

which implies that c a r d ( S \ R ) = c a r d ( R \ S ) . Thus the formal expression (3.1) can

be interpreted as giving rise to the well-defined transition functions,

gRs(A):=

AkES\R(1 + A)ek
/~keR\S(1 + A)ek

det {(ej, Aek)} .

jeR\S
k~S\R

(3.3)

As a function of A, this determinant is non-zero precisely when A corresponds to an

element of Us c~ [JR. Because they involve only determinants of finite-dimensional
submatrices, the functions (3.3) are clearly holomorphic. The properties needed to
define transition functions, namely that gRSgSR = 1 and gRSgSTgTR = 1, follow
immediately from the expression of gRS as a ratio of wedge products. Therefore we
can take the set of gRs'S as the definition of a holomorphic line bundle D E T over
G r ( W ) . This definition does depend on the choice of basis for ~r As we mentioned
earlier, in the case of half densities on the circle, we think of the elements
exp { _+ i(k - 89 } as a canonical choice of basis determined by the parametrization
of the circle.
Let FD be the space of holomorphic sections the dual bundle D E T ' (the bundle
D E T itself has no non-zero holomorphic sections). We will create a Hilbert space
out of Fb, the topological dual of this space. We start by defining a map,
fl: D E T x D E T -o ~. For p, q ~ D E T , let fl(p, q) unless p, q both lie in DETIvs for
some S, and in this case set

fl(p, q) = Zp2q det (1 + A*Aq).

ws

(3.4)

The notation here is that p corresponds to (A,, 2p) under DETIvs ~ Us x C, and
similarly for q. The determinant over Ws is a Fredholm determinant, which is
well-defined because Ap and Aq are Hilbert-Schmidt.
Proposition 3.1. fl defines a map D E T x D E T ~ C, which is holomorphic in the

second variable, antiholomorphic in the first, and linear and antilinear on the
respective fibers.
Proof. We first need to show that fi is well-defined under the transition maps of the
bundle DET. To see this, suppose we have points p, q ~ D E T which lie above AR
and BR, respectively, in UR, and also above As and Bs in Us. We want to show that
det (1 + A~Bs) = gsR(AR)gsR(BR) det (1 + A*BR).
w~
wl

(3.5)

The fact that graph(As) = graph(AR), and that graph(Bs) = graph(Bg), leads to
some simple identities for various submatrices of the A's and B's. These can be used
to prove the following fact,
(1 + A*BsPs) = (Psc~R -- PSc~RA*Ps\RA'~PR\s + P s \ R A ~ P R \ s )
x (1 + A*BRPR)(Psc~R -- PR\sBsPs\RBRPRns

+ PR\sBsPs\R ) ,

(3.6)

468

D. Borthwick

where all these expressions are considered as operators on the full space ~ff. The
proof is easy, but somewhat tedious. Each factor in (3.6) is of the form 1 + (trace
class), so we can use the product rule when taking the determinant of each side. The
determinant of the left-hand side just gives the determinant over Ws in the
expression above, and the second term on the right-hand side gives the corresponding determinant over WR. That leaves two determinants of expressions which differ
from the identity only by operators of finite rank. It is easy to see that they reduce
to the finite dimensional determinants which appear in the transition functions
gsR(BR) and gsR(AR).
This shows that fl satisfies the correct transition law. The remaining step is to
prove holomorphicity. Note first that fl satisfies a hermitian property,

fl(q, p) = fl(p, q) ,

(3.7)

so that we really need only prove holomorphicity in the second variable. The
linearity on the fibers is obvious. Fix a point p e DET, which lies over some open set
Us. It suffices for us to prove that on each open set UR, the function

f ( B ) = fl(p, (B, 1)R)

(3.8)

is holomorphic, where (B, 1)Re UR X C refers to a point in DET[v,.

Clearly, if UR and Us lie in different connected components of G r ( ~ ) , then f
is identically zero, so we can assume that the two sets lie in the same component.
Over Us n UR, we have

f ( B ) = 2pgsa(B) det (1 + A * B s ) ,
(3.9)
ws
where Bs is the point in Us corresponding to B. The 9SR appears when we transform
the point (B, 1) z UR X C. We can rewrite this expression as follows,
f ( B ) = 2 v det {<(1 + Ap)ej,(1 + Bs)ek>} det { ( e i,Bek>}
j,k~S

jeS\R
keR\S

= )~p det {<(1 + Ap)e~, (1 + B)e,>} .

(3.10)

jeS
ksR

We now observe that this last expression in fact gives a well-defined, explicit
formula for f on all of U~. It is holomorphic in B because of the absolute
convergence of the expansion for the Fredholm determinant. []
Now that we have the map fl, we can easily define a hermitian structure on
DET. Given smooth sections of DET, Zl, z2, we simply take
<"~'1, "C2> ( W ) : = fl(~l(W), ~2(m)) 9

(3.11)

We have already remarked that fl satisfies a hermitian law. The smoothness of the
hermitian structure follows immediately from the holomorphicity of ft.
The map/3 provides more than a hermitian structure, however. With it, we also
obtain an inner product on the space F~. This is done as follows. An element of FD
can be thought of simply as a holomorphic map D E T ~ IE, which is linear on each
fiber. Clearly, we can regard /3 as being an element of FD | FD, wher FD is the
natural complex conjugate space to FD. Given two elements q, ~ ~ F~, we define the
pairing
(q, ~>p := (f/| 4)'/3.

(3.12)

Pfaffian Line Bundle

469

By
t7 we mean the element of F-'o which is the conjugate of t/, so that (f/| ~)~
--!
F 9 @ F ; . The pairing (3.12) is continuous, by the continuity of the evaluation map,
and it is hermitian by the hermitian property of ft.
It is also, in fact, positive definite. Given an index set S, we write Ps for the point
(A = 0, 2 = 1)s in DETv~. We can define an element 7s ~ F~ by evaluation at Ps,
?s(a) := tr(ps),

(3.13)

because ~ e FD is just a map DET ~ C. Note that such elements of F~ form an

orthonormal set,
<]~s, 7R ) B = fl(Ps, PR) = ORS.

(3.14)

We can also use/3 to define elements Zs e FD by

Zs(P) :=/3(Ps, P) .

(3.15)

For a general ~ ~ Fb, we see easily that

(~s, ~)~ = ~(Zs) 9

(3.16)

Proposition 10.1.5 of [21] showed that the algebraic span of the Zs is dense in Fo.
Therefore, if <?s, ~)a = 0 for every S e s r then ~ = 0. This implies that the
algebraic span of the 7s'S is dense in F~. The pairing is thus positive definite, and
the ?s'S form an orthonormal basis.
Definition 3.2. ~ o is the completion of F'D in the inner product < 9 9 )~.
This Hilbert space gives the fundamental representation of Ure~(~) [21]. We
will define the corresponding Hilbert space for the Pfaffian fine bundle in the next
section.

4. The Pfaffian Line Bundle

The space of half-densities on the circle has a natural complex conjugation, which
maps Yf+ to ~ _ and vice-versa. When Yf has such a real structure, we can define
a submanifold of Gr(Yg) over which the determinant line bundle has a holomorphic square root.
In general, by a real structure we mean a complex anti-linear map J on Jr such
that J 2 1 and < Jx, J y ) = ( x, y ). In addition, for a polarized Hilbert space we
will assume that J : ~ ~ ~ .
In the case of half-densities, J is the natural
complex conjugation, which takes ek to e-k. We shall always assume that we have
chosen our basis to behave this way under J. With such a complex structure, we
can define a symmetric bilinear form on ~r
=

(x, y ) : - < J x , y ) .

(4.1)

Using this form, we define the isotropic Grassmannian by

G r / ( ~ ) := { W ~ G r ( ~ ) :

J W = W }.

(4.2)

The condition that J W = W implies that Gr~(Jr ~) is a submanifold of the zero

index component of Gr(~r
The spaces Ws lie in Grt(Yf) if and only if S e d ~ ,
where
dr:= {Sed:

J S = S ~} .

(4.3)

470

D, Borthwick

We also note that graph(A) lies in GrI(W), for A~I2(Ws, WXs), S~5~I, if and
only if

(x, ay) = - (Ax, y)

(4.4)

for all x, y e W. In other words, if A -- - JA* J. Such an operator is called skew,

a W,s, W~) the space of skew Hilbert-Schmidt operators from
and we denote by 12(
Ws to W We thus have a cover of G r i ( J f ) by open sets Vs ~- 1~2(Ws, W~). It is
easy to check that G r i ( ~ ) falls into two connected components, depending on
whether card(S ~ Z+) is odd or even.
The finite dimensional analog of the isotropic Grassmannian is easily described.
For a 2n-dimensional complex vector space with real structure, the space of
n-dimensional isotropic subspaces is just the homogeneous space O2,/U,, which
consists of two simply connected copies of SOa,/U,. These finite-dimensional
manifolds can be successively embedded to form a nested set of submanifolds of
Gr1(Yf) whose union is dense. The existence of a topological square root of the
determinant line bundle over SOE,/U, follows from the fact that the Chern class of
this line bundle is even. We can see this as follows. First of all, for n = 2,
S 0 4 / U 2 ~ CP 1. By writing out the transition functions, one sees immediately that
the determinant line bundle in this case is just L | L, where L is the tautological
bundle over C P x. Thus the Chern class is even in this case (and the square root is
obvious). Now consider the inclusion CEP ~ ~ S 0 2 n / U n. Because C P 1 and S 0 2 n / U n
are connected and simply connected, this inclusion induces an isomorphism in H 2.
Therefore, by naturality, the Chern class of the determinant line bundle over
SO2,/U, is always even.
To show the existence of a holomorphic square root, in the infinite dimensional
case, we turn to the transition functions. Over Gr,(oVf), with the open cover we
have just described, the restriction of the determinant line bundle is defined by the
same transition functions gRS as before, with S and R restricted to ~/~. Note,
however, that because of the condition that J W = W we have k E S \ R if and only
if - k ~ R \ S . Thus we can write,
ORs(A) = det {(ej, ek)}
j~R\S
k~S\R

det

{ ( e - i, Aek)}

j,k~S\R

det

{(e~, Ae~)} ,

(4.5)

j,k~S\R

where ( ' , -) is the bilinear form given by (4.1). This is the determinant of a skewsymmetric matrix. Because the determinant is non-zero when VR intersects Vs, the
skew matrices must be even dimensional in this case. These transition functions
therefore have holomorphic square roots, given by the Pfaffian.
Hence we will define a line bundle P F over Grl(3gf) which has transition
functions given by

hsR(A ) :=

Pf

{(ej, Aek)} .

(4.6)

j,k~S\R

For the moment, we allow an arbitrary choice of orientation for the Pfaftians in
these functions. We must check that the cocycle conditions, hRshsrhwR 1, are
=

Pfaffian Line Bundle

471

satisfied. Because the Pfaffian is the square root of the determinant, we know these
identities are satisfied up to sign. To check the sign, we note that the Pfaffian of
a matrix and the Pfaffian of its inverse appear in these expressions with opposite
orientations, where by opposite orientation we mean et, ^ . 9 9 /x et2. replaced by
e t 2 . / x . . . ^ et,. This involves a change of sign of ( - 1)'. It is a simple fact about
Pfaffians that when A is an invertible 2n-dimensional matrix,
Pf(A)Pf(A -1) = ( _ 1)',

(4.7)

so that these signs always work out correctly.

Theorem 4.1. As defined by the transition functions (4.6), P F is a holomorphic line
bundle over the isotropic Grassmannian. Furthermore, P F is a square-root of the
determinant line bundle in the sense that
PF | P F ~ DET]Gr,(~)

(4.8)

Proof The preceding discussion demonstrated the consistency of the transition

functions. Because they involve only finitely many variables and the finite dimensional Pfaffian function is holomorphic, they are clearly holomorphic functions on
the sets Vs. Because the square of the Pfaffian function is the determinant, the
second property follows from comparing the transition functions (4.6) to the
restrictions of the transition functions of D E T to G r , ( J t ~) (4.5). []
We still have not specified the orientation with which the Pfaffians in the
transition functions are to be defined, but we will show now that the condition (4.8)
determines P F up to a holomorphic isomorphism, so that any consistent choice of
orientation for the transition functions will give us an equivalent line bundle. Later,
we will fix a particular choice of orientations for convenience.
Theorem 4.2. P F is a unique holomorphic square-root of D E T over GrI(W).
Proof We will show in general that holomorphic square roots of line bundles over
G r , ( W ) are unique topologically and holomorphically. Because Gr,(oug) has
a collection of finite dimensional submanifolds whose union is dense, as we
mentioned above, a holomorphic line bundle on G r , ( W ) is specified completely by
its restrictions to these submanifolds. Thus it suffices to prove that holomorphic
square roots of holomorphic line bundles are unique on 02,,/U, for each n,
or rather SO2,,/U', since 02"/U,, consists of two copies of S02,,/U,,. Smooth
complex line bundles on SO2"/U" are classified completely by Chern classes in
H2(SO2,,/U', Z), which one can easily check to be isomorphic to Z for all n using
the fibration U'--* S02" ~ S02"/0". This means at least that square roots are
unique up to smooth isomorphism. Holomorphic line bundles are classified by
elements of H 1(SO 2,,/U,, (9 * ), where (9 * is the sheaf of non-vanishing holomorphic
functions. The short exact sequence of sheaves,
(4.9)
where (9 is the sheaf of holomorphic functions, gives us the cohomology sequence
HI(SO2"/U,, (9) ---,Hx( SO2JU., (9*) ~ H2(SO2"/U,, Z) --~ H2(SO2"/U., (9).
(4.10)

472

D. Borthwick

Because S 0 2 n / U n is a simply-connected K/ihler manifold, we can apply the results

of [9] to see that HI(SO2,/U,, (9)= HZ(SO2n/Un,(9)--0. Thus we have the
isomorphism Hl(SOzn/Un, (9*)~-H2(SOzn/Un, TZ). We conclude that the holomorphic square root of a line bundle is unique. []
We note briefly that the determinant line bundle can be obtained directly from
the Pfaffian line bundle construction. Suppose we are given a polarized Hilbert
space ~ f without any particular complex structure. Define a new Hilbert space
~ = J4"OJ4 ~'. We give this space the polarization X_ = ~ _ G~ut~'+,Sl+ =
~g+ | Jt ~'_. 2Y has a natural complex conjugation which is compatible with this
polarization, given by the canonical anti-linear m a p from a Hilbert space to its dual
(i.e., x ~ (x, 9 )). There is a natural embedding of G r ( ~ ) into Grt(s
and the
bundle D E T over Gr(Ng) is just given by the pullback of P F by this embedding.
Let Fp be the space of holomorphic sections of PF'. We construct an inner
product on this space just as we did in the determinant case. We will define a m a p
: P F x P F ~ ([;, which is in some sense the square root of the m a p fl we used
earlier. First, note that for BeI'2( Ws, W~), the m a p (1 - Ps)BPs on the full space
lies in I~ (Jr~ where Ps is the orthogonal projection on to the subspace Ws. This
is easy to check,
J((1 - Ps)BPs)*J = JPsB*(1 - Ps) J
= (1 - Ps) JB*JPs
= - (1 -- Ps)BPs.

(4.11)

The relative Pfaffian is defined for operators in I ~ ( ~ ) , and is a holomorphic square

root of the relative Fredholm determinant [13]. For the rest of this paper we will
use the notation
Pf(A, B ) : = Pf(PsA(1 -- Ps), (1 - Ps)BPs),
ws

(4.12)

where A e I~E(W is, Ws) and B e I"2( Ws, W~ ). Note that

P f ( J A J , B) 2 = det (1 + PsA*BPs)
ws
~"
= det (1 + A ' B ) ,
(4.13)
ws
which is the expression which appears in the definition of ft. It is clear that we
should define the m a p e as follows. We set c~(p, q) = 0 unless p, q are both in PFIvs
for some S, and in this case
~(p, q) = 2q Pf ( J Ap J, Aq) .
(4.14)
ws
Proposition 4.3. ct defines a map P F x P F --* ([;, which is holomorphic in the second
variable, antiholomorphic in the first, and linear and antilinear on the respective
fibers.
Proof. The proof follows almost entirely from the corresponding result for fi proven
in Proposition 3.1. The one point to check is the following. By taking the square
root of the corresponding equation for fl, we know that
Pf (JAs J, Bs) = + hRs(As)hRs(Bs) Pf (JAR J, BR) ,
ws
wR

(4.15)

Pfattian Line Bundle

473

where As ~ Vs corresponds to A R ~ V R , and likewise for the B's. We must check that
the sign is positive. To fix the sign, we let A = B. Then we have
Pf ( J a s J, As) = + [hRs(hs)l z Pf (JAR J, AR).
Ws
WR

The

Pfaffians

are

both

positive

by

continuity,

since

(4.16)
P f ( 0 , 0 ) = 1, and

P f ( J A J , A) z = det(1 + A ' A ) > 1. Thus the overall sign in Eq. (4.15) is (+), as

desired.

[]

This proposition tells us that a~/~e | Fe, and we can use it to define an inner
product ( ", 9)~ on F~,, exactly as in the determinant case.

Definition 4.4. ~-p is the completion of F'e in the inner product ( "," )~.
We obtain an orthonormal basis {Ts} for o~e, indexed by S ~ d i ,
way as for ~D.

in the same

5. The Fock Space

Define the positive energy Fock space ~ + as the completion of the full exterior
algebra of Jt~+,

n=l

in the inner product

(X t A . . . AXn,

Yl A , . . A ym)

:=

6,mdet{(xi, y j ) }

(5.1)

It is clear that the elements of the form es I ^ . . . /x % , for all sets of positive
integers, form an orthonormal basis for ~-+. An element S of d t is completely
specified by the subset S n Z + . If we denote the elements of S ~ Z + by
{Sl . . . . , s,}, then we have the obvious isomorphism p : J e ~ Y+ given by
P(7s) := % A . . . /x es, ,

(5.2)

where {Ts} is the orthonormal basis for fie. This isomorphism is singled out by our
choice of basis. There is no canonical way to relate the 7s under change of basis
of ~,~F.
Since an element of P F corresponds to an element of fie by the evaluation
map, we can ask how the isomorphism acts on the elements of PF. For example,
suppose we take a point (A, 2)s in PFlvs. Considering the point (A, 2)s as an
element of ~,~p, we can expand
(A, 2)s =

(7R,(A, 2)s)v?R

R~

Y~ v(pR,(A,,~)s)?R
REdt

~,
2has(A)Ta.
R e~41

(5.3)

474

D. Borthwick

Here and elsewhere, we will use the same notation (A, 2)s to refer to a point of P F
and a point in o~p via the evaluation map. We also adopt the convention that
has(A) = 0 if VR c~ Vs is empty. Equation (5.3) implies that
P((A,)Os) =)~

hRs(A)P(~g)

R ~,~

= ,~ ~

(5.4)

hgs(A)erl A . . . A er, 9

R E ,~ r

Let So denote the index set ;~_ ~ ~4,, and for convenience let 1Io denote Vso. W e
can define a map q5 from Vo -~ I~(d/f_, W+) to the completion of ^23tf+ by
q~(A) :=

1
(e-i, Ae_j)ei

(5.5)

^ ei ,

i,j~2Z+

This map is well-defined, because

II q~(A)LI ^2g+ = ~ II A II2 9

(5.6)

Because the space ~ + has an exterior algebra structure, we can take the exponential of an element. It in fact follows from the definition of the finite dimensional
Pfaffian that
e4~(A)=

Pf

vcZ+

l <~i,j<~n

{(e_~,,Ae_~j)}e~,^...^e~

(5.7)

Hence, if we orient the transition functions so that

h~so(A) =

Pf

{(e_,,,

Ae_,j)},

(S.8)

l~i,j<n

then the map p has a particularly nice form

p ( ( A , 2)So) = 2e ~

(5.9)

Because p is a Hilbert space isomorphism, this expression implies

e f (JAJ, B) = ( e ~

e~

(5.10)

,el_

This result, which is quite simple to prove directly, was first pointed out in [21].
We can develop a formula similar to (5.9) for points lying over an arbitrary
subspace Vs, provided we choose the appropriate orientations for transition
functions. First we introduce the action of a Clifford algebra on ~-+. Because of the
exterior algebra structure, we can let an element of ~ + act on ~ + by exterior
multiplication, or by interior multiplication. F o r j > 0, let Z~ be the operator which
acts as the sum of exterior multiplication by ej plus interior multiplication by e~.
Thus, if co is a form such that ej/x co # 0, then we have
Zj(CO) = ejAco,

zj(e~Aco) = co.

(5.11)

The )(s are clearly self-adjoint, and they satisfy the anticommutation relations,
{zj, zk} = 6jk.

(5.12)

Thus they form an infinite dimensional Clifford algebra which acts unitarily
on ~ + .

Pfaffian Line Bundle

475

Let aj be the element of U,,~(Scf) whose sole effect is to interchange the basis
elements ej and e_~. As above, we identify the index sets S s ~r with sets of integers
{sl . . . . . Sk} = S c~ ~E, arranged in increasing order. F o r A e Vo, we define a new
m a p A s: W s ~ W ~ b y
A s : = (as, 9 9 9 a J A ( a s ,

(5.13)

. . . ask).

This m a p A s lies in Vs. In fact, the pairing of A to A s gives an isomorphism

between Vo and Vs. N o w we can m a k e the following proposition.
Proposition 5.1. W e can c h o o s e o r i e n t a t i o n s f o r the t r a n s i t i o n f u n c t i o n s hRs so t h a t
f o r A ~ Vo, w e h a v e
p ( ( A s, 2)s) = 2;~s, 9 9 9 Zske r

(5.14)

w h e r e ( A s, 2)s r e p r e s e n t s a p o i n t in ~1, b y the e v a l u a t i o n map.

P r o o f . We have already noted that p has the general form
p ( ( A s, 2)s) = 2 ~, h R s ( A S ) e , l A . . . A e~, .
R

(5.15)

The function h R s ( A s) can be reduced as follows,

h ~ s ( A s) = __

Pf

{(ei, ASek)}

i, j e S \ R

= +

Pf

{(ei,(a,,...as~)A(a,,...a,~)ek)}

i,jeS\R

= _+

Pf

(5.16)

{(ei, Aek)},

i, j e S o k T

where T = (a~ 1 . . . a s ~ ) R (the a ' s act on ~r by interchanging j and - j ) . This last

expression is just the transition function h r s o ( A ) , so that
p ( ( A s, 2)s) = +_ ) , ~ h r s o ( A ) e r l A . . . ^ e , , .
T

(5.17)

The sign of course depends on the orientations we choose for the Pfaffians. N o w we
simply observe that T = (as1 9 9 9 (rs~)R implies that
% /x . . . A %

= ++_Zsl . . . Zs~(etl /x . . . ^ e t m ) .

(5.18)

Thus we have
P((AS, X)s)=

+ Zs, . . .

Zs~

hrso(A)(et,/x

.. . Aet=)

Te dr

= +- Zs, . . .

(5.19)

Zs~ e ~

We can fix the orientation so that the sign is positive as follows. Suppose that
S\R = {ql,...,qm},
with the q's in increasing order, q~ < . . .
< qm. Define
eRS = +__ 1 by the equation,
)~s~ 9 9 9 Z s ~ ( e , ^ 9 9 9 A e,~) = e g s % A . . . A e , .

(5.20)

The transition function hRs involves a Pfaffian over the vector space which is the
span of {eq . . . . . , eq=}. We now specify the orientation on this vector space by the
top form,
gRseqm A . . . A eql

(5.21)

476

D. Borthwick

where eRS is determined from Eq. (5.20). It is straightforward to check that these
choices of orientation imply the condition (5.14). []
The composition of the evaluation map P F -~ o~p with the map p : o~p ~ ~ +
gives us a map from P F to ~ + . The fiber of PF over any point of G r t ( ~ ) maps to
a ray in ~ + . Thus we obtain a map z from G r / ( ~ ) to the projective space P(o~+),
given by
"C(As) := [Zsl . . . Zsk er

(5.22)

As a projective space, P ( ~ + ) has a tautological holomorphic line bundle &o, whose

fiber over a ray is the ray itself.

Proposition 5.2. P F is the pullback by the map z of the tautological line bundle
&e over p ( o~ + ).
P r o o f It is simple to check that the trivialization with which we have defined P F is
recovered from the maps t/s: Vs ~ Y given by
tls: (A s, 2)s ~ (z(AS), p ( ( A s, 2)s)),

(5.23)

where p was given by (5.14). []

This proposition connects the definition of PF in Sect. 4 with the definition used
in [21].

6. The Representation in Finite Dimensions

Because the discussion of the restricted orthogonal group and its representations
on sections of P F ' in the following section is somewhat dense with analytical
details, we give here a brief discussion of the situation in finite dimensions. See [21]
for a more thorough treatment of the finite dimensional case.
The basic philosophy of the construction of irreducible representations on the
space of holomorphic sections of a homogeneous line bundle comes from the
Borel-Weil theorem [9]. As an example of this theorem, let X = ~2", and consider
the irreducible representation of Un on the k th exterior power A k x '. The ray
defined by the vector ~2:= o~1/x . . . /XO~kS A k x ', where {c@ is a basis for X ' , is
invariant under the subgroup Uk X U,,-k of U,. By considering complexifications,
we see that the orbit of f2 defines a holomorphic map U,,/Uk X U,,-k ~ P(Akx').
Associated to such a map is a complex line bundle L and a map A k X --* F(L'). This
line bundle L is just the determinant line bundle over Grk(X), and the map
/xk X ---, F( L ' ) is an isomorphism.
In Sect. 7, we essentially give the analog of this Borel-Weil construction for
a generalization of the infinite dimensional wedge representations which were
described algebraically in [14]. In finite dimensions this amounts to a construction
of the spin representation of S02,. Let X be a finite 2n-dimensional complex v e c t o r
space, with real structure J and decomposed into X_ | X+. Let PFx be the
Pfaffian line bundle over the even component of GrI(X). We will sketch the
construction of the spin representation on F, the even degree subspace of the full
exterior algebra A X+. Then we will show that the action on F, which is isomorphic to the space of holomorphic sections of PF3r comes from a holomorphic

Pfaffian Line Bundle

477

action on PFx. The extension of the representation to both components of the

Pfaffian line bundle is simple and will be done for the infinite dimensional case in
the next section.
Given g ~ S02,, let

with respect to the decomposition X _ @ X +, i.e. a is a m a p from X _ ~ X _, etc.

Let Ug be the subset of Vo consisting of the skew maps A : X _ --* X + for which
(a + bA) is invertible. Uo is just the subset of Vo whose image under the action of
g on G r I ( X ) still lies in Vo. This action takes A ~ Uo to

A~

(c + dA)(a + bA) -1

(6.2)

We will see in Sect. 7 that the elements o f F of the form e 4~(a), where A ~ Uo, span F.
Since we want the action of g on F to cover the action on GrI(X), by Proposition
5.1 we would like to set
g. e~(a) = #o(A)e ~(Ag),

(6.3)

where #0 is a numerical factor which depends holomorphically on A. For the

representation to be unitary, we see from the form (5.10) of the inner product on
F that we need to have
#g(A) 2 = det(a + b A ) .

(6.4)

This requirement is the key to the difference between the spin representation in
finite and infinite dimensions. For, in the finite dimensional case, Eq. (6.4) has two
solutions,

#o(A) 2 = +_ [ d e t a ] l / z P f ( a - l b , A ) .

(6.5)

It is not possible to make a global choice of sign here for all of S02,, because
n1(SO2,) = 2~2. We can, however, define a projective representation which comes
from an honest unitary representation of Spin2,, the simply connected double
cover of S02,. The difference in infinite dimensions is that the square root of the
determinant of a becomes ill-defined. We have to define the numerical factor # in
a more involved way, resulting in a cyclic extension of the group rather than
a double cover. The fundamental group of the infinite dimensional group
SOres(~), which will be defined in the next section, is no longer Z2.
To continue the finite dimensional discussion, define the group Spin2, to consist
of pairs (g, #0), such that g ~ S02, and #0: Ug ~ (I: satisfies Eq. (6.4). We have seen
that Spin2, acts on F, but have not yet shown that this comes from an action on
PFx. Let i : PFx --* F be the inclusion, as described at the end of Sect. 5, consisting
of the composition of the evaluation m a p and p. We need to see that the action
preserves the image of i, and, moreover, that the resulting action on PFx is
holomorphic. Let ~ = (g, #g), and for index sets R, T, let U i T be the set of all B ~ VR
such that g maps the subspace graph(B) into VT. All that we need do is find
a holomorphic function fr on U~ T such that the action (6.3) corresponds under i to
9(B, 1)R = (g "B, fc(B))T,

(6.6)

478

D. Borthwick

where B ~ URoT. Suppose that B corresponds to a point C ~ U0. Then one can easily
check from (6.3) and the definition of i that we would have
~'(B, 1)R = (g'B, t~o(C)hTso(C~

(6.7)

Thus we will be finished if we can extend the holomorphic function,

f ( B ) = #o(C)hTso(C~

(6.8)

(C depends holomorphically on B), from uRoTc~ Uo to all of U~oT. Since the

complement of Uo is defined by the vanishing of a holomorphic function, namely
det(a + bA), in finite dimensions we can simply apply the Riemann extension
theorem.

7. The Restricted Orthogonal Group

We were able to realize the manifold G r ( f i ) as a homogeneous space for the
restricted unitary group, given by (2.2). We can do the same for the submanifold
G r i ( f i ) , using the restricted orthogonal group. By definition, any element of
G r ( f i ) has the form g f i - , for some g ~ Ures(fi). TO define an element of G r i ( f i ) ,
we require further that an operator g preserve the bilinear form on f i , i.e. that
g commute with the conjugation operator J. We thus define
Ores(fit~ := {ge Urr

= gJ } .

(7.1)

To write G r i ( f i ) as a homogeneous space, we need to know when two different

elements of O~s(Jt") act on 9~_ to give the same subspace. It is clear that
g f i _ = ghfi_ for h e U ( f i _ ) x U (fi+), since such an h preserves H_ and f i + . The
combination gh will be an element of O,r
however, only if [ J, hi = 0. This
condition is satisfied only if h is of the form (u, JuJ), for u e U ( f i _ ) . Therefore we
can write
Gr~(~ '~) = O ~ ( J f ) / U (fig_),

(7.2)

where U ( f i _ ) acts on the right by its embedding u ~ (u, JuJ).

The group O~es(fi) splits into two connected components. For any g ~ Oros(fi),
i n d e x ( ~ _ g P - ) = 0. The components are determined by whether the kernel and
cokernel of P_ gP_ have even or odd dimension. The identity component forms
a group, which we will label SOr~(fi). Because G r t ( f i ) is a homogeneous space
with respect to the orthogonal group, we would expect to have at least a projective
action of S O r ~ ( f i ) on the line bundle. We will define this projective action by
taking a central extension, S~"lffres(fi), of SOre~(fi). We will show that there is
a unitary representation of SOres(fi ) on the space ~-+, which comes from an action
on PF. At the end of this section we will show how to apply these results to the full
group, O ~ ( f i ) .
The first step is to understand the action of SOre~(fi) on Gr~(fi). Suppose
a point of G r ~ ( f i ) is represented by A ~ Vs. We want to find the point which
corresponds to g(graph(A)), for g ~ SO~e~(fi). If we assume that this point lies in
VR, then for some B ~ VR and some isomorphism q : Ws ~ WR, we must have

Pfaffian Line Bundle

479

Take S = R = So, where, as before, So is the set of all negative integers. If we let

g :=

(a;)
C

74,

with respect to the decomposition ~ _ 9 ~ + , then we see from Eq. (?.3) that
q = (a + bA), and that B = (c + d A ) ( a + bA) -1. Let Ug be the set of all A t Vo
such that (a + bA) is invertible. F o r A ~ Ug, define
A~

(c + d A ) ( a + bA) -1 ,

(7.5)

as in Sect. 6.
Because SOr~s(~) preserves the two connected c o m p o n e n t s of Gr1(J#), it will
also preserve the decomposition of ~ + into the subspaces of even and odd degree.
We will first consider the action on the space of even degree, which we will denote
by ~ ~en.
L e m m a 7.1. For any open set U c Vo, the elements o f the form e ~ta) for A ~ U span
~e~en.
Proof. We can assume that U is a n e i g h b o r h o o d of A = 0, because we can always
translate. That is, suppose U is a n e i g h b o r h o o d of the point Ao, and we want to
expand x. If we can expand
eC'(A~

= ~ ~.Ae*(A) ,
A

(7.6)

with sum over A's lying in a n e i g h b o r h o o d of A = 0, then we will have

X = Z )'A e(~176

(7.7)

with A + Ao lying in a n e i g h b o r h o o d of Ao.

Given a basis element % ^ . . . ^ er~ of ~ ~ " , with k even, we can always find
an A ~ Vo so that
(o(A) = tC(er,/x er2 + ' " " + e,~_, A e~k),

(7.8)

and with ~c sufficiently small we can find such a point in any n e i g h b o r h o o d of the
identity. Exponentiating gives
e *(A) = ~c(e,, ^ . . . /x e,k) + lower degree terms .
Thus we can recover all of the basis elements, and hence span all of . ~ ~r

[]

Because of this lemma, we can define the action for each g by specifying it on
elements of f f ~.**"of the form e *(A) for all A ~ U o. The action will preserve the inner
product, so that it will automatically extend to all of f f ....
+ . If the action is to cover
the action of "SO',~s(~vt~) on G r I ( J f ) , then we will have to have
9" er

= I~( A ) e4~(Ag) ,

(7.9)

where #o is some numerical factor depending holomorphically on A. The inner

p r o d u c t which the action must preserve is

(e $(A), e 4'(m )

= P f ( J A J, B ) .

(7.10)

480

D. Borthwick

If we transform A and B by g, then the inner product becomes

<er

er

= Pf (JA~

B e) .

(7.11)

2/g_

Recall that A ~ was defined so that

(')
Ag

=g

(7.12)

( a + b A ) -1

We can thus evaluate

( er

eC(Bg)>2 = det (1 + (A~

=det

B~

(a+bA)*-'

g*g

det(1 + A ' B )
= det(a + bB)(a + bA)*

(7.13)

This means we would like to choose the factor #o(A) so that

(#o(A)luo(B)) 2 = det(a + bB)(a + bA)* .

(7.14)

In the finite dimensional discussion of Sect. 6, this amounted to choosing one of

two square roots. In infinite dimensions it is not possible to choose any square root.
We are thus led to a more complicated procedure, which ends up giving a central
extension of SOres(J/f), as follows.
The reason (7.14) has no solution is because there is no infinite dimensional
Pfaffian analogous to the Fredholm determinant. We have only the relative
Pfaffian. It turns out that we should define the numerical factor as a function of two
variables,
#g(A, B):= Pf (A - B, (a + b A ) - l b ) .

(7.15)

~vf+

This relative Pfaffian is well-defined because both arguments are skew HilbertSchmidt operators. This is obvious for the first argument, A - B. The second
argument is Hilbert-Schmidt because b is Hilbert-Schmidt, and we see that it is
skew as follows. The fact that gJg* J = 1 implies that bJA* J is skew, and the
fact that A is skew implies that b A J B * J is also skew. The difference of these two
terms is
b(Ja*J - AJb*J) = bJ(a + bA)*J ,

(7.16)

so that b J ( a + bA)* J is skew,

b J ( a + bA)* J = - (a + bA) J b * J .

(7.17)

By applying (a + bA) -~ on the left and (a + b A ) * - ' on the right, we get

(a + b A ) - l b = - Jb*(a + bA) *-1 J ,

(7.18)

so that (a + b A ) - ~ b is skew.
With # depending on two variables, we will now try to replace the relation (7.9)
with something like
g" e 4,(A) := #o(B, A)e C~176 .

(7.19)

Pfaffian Line Bundle

481

Here B is an extra parameter, the choice of which will need to be included in our
extension of SOr~=(~f). We will also need to revise Eq. (7.14). Taking the square of
#o gives
#o(A, B) 2 = det(1 - (A - B)(a + bA)-Xb)

= det(1 - b(A - B)(a + b A ) - l ) ,

(7.20)

(we can switch the order of b and (A - B)(a + bA) -~ because both are HilbertSchmidt). We can simplify
1 - b ( A - - B)(a + bA) -1 = (a + bB)(a + bA) -~ ,

so that

#o(A, B) 2 --

det(a + bB)(a + bA) -~

(7.21)

We cannot extract the determinant of (a + bB) from this expression, but the
combination (a + bB~)(a + bB2)* does have a determinant. Therefore, we can
revise Eq. (7.14) by a constant factor which depends on the extra parameter,
(#o(A, Bz)#o(A, B1)) 2

det(a + bB~)(a + bB2)*

= det(a + bA)*(a + bA) "

(7.22)

We will take care of the extra factor (the denominator) through the following
extension of SOr,=(:gg).
Definition 7.2. SO ~e=(2/g) is the 9roup whose elements consist of triples
(9, A, 2)6SOre=(gff) x I~(~,~

ggY+)x 112x ,

(7.23)

such that A ~ Uo and

[2[ 2 = x / d e t ( a + bA)*(a + b a ) .

(7.24)

We identify two triples (g, A, 2) and (9, B, 2') when

2' = 2#o(B, A ) .

(7.25)

(gl, A1, 21)'(g2, A2, 22) = (g3, A3, 23),

(7.26)

The multiplication is 9iven by

where g3 = glg2, A3 ~ Uo3 ~ [-7o2is chosen so that A~ 2 e Uo,, and

23 =

2a22#o,(A1, A~

A3).

(7.27)

This extension is referred to as Spinr

~) in [21]. Because its definition is
somewhat involved, so we will explain where the various parts come from. The
requirement on the 12l given by Eq. (7.24) appears so that the denominator of Eq.
(7.22) will be cancelled off. The identification (7.25) simply equates two elements of
the group which for an action of the form (7.19) should be identical. The multiplication laws (7.26) and (7.27) also follows directly from the action (7.19).
We can easily see that SO ros(J4~ is a cyclic extension of SOre=(~ff). Suppose we
are given two elements lying over 9, which are represented by triples (9, A, 2) and
(9, B, 2'). The equivalence relation (7.25) tells us that
(g, B, 2') = (g, A, 2' #o(A , B)) .

(7.28)

482

D. Borthwick

Using Eqs. (7.22) and (7.24), we see that the scalar factor relating the two elements
lies on the unit circle,
)~,

i~o(A, B)

= 1,

(7.29)

so that our extension is cyclic. We will compute the Lie algebra cocyle corresponding to this extension in the next section.
Theorem 7.3. For an element ~ = (9, A, 2)e

SOres(~), and for B~ Uo, let

~" e ~ m = 2#o(A, B)e r

(7.30)

This action extends linearly to define a unitary action of S'-O~( Jg) on ~ 2 Cn.
Proof First we check that Eq. (7.30) is compatible with the multiplication laws
(7.26) and (7.27). For j = 1 , 2 , 3 ,
let ~j=(gj, Aj, 2 j ) e ' ~ r ~ ( ~ ) , such that
~1~2 :

43. Given B e Uo~ such that Bg~e Uol, we have

~1 "(~z "e~'tm) = ~ "(22go~(A2, B)e r
= 21~2#0~(A1, Bo2)No2(A2, B)er

= 23( Po,(A1,B~
#o~(Ax, A~

B) )
A3) cO(n"3)"

(7.31)

We need to show that the factor is parentheses is equal to pg3(A3, B). The square of
the numerator is given by

(#o1(A1, Bo2)#oz(A2, B)) 2 = det(at + blBO2)(al + b l A t ) -1

x det(a2 + bzB)(a2 + b2A2) -1

(7.32)

Now, since

(al + blB~

+ beB) = al(a2 + beB) + b1(c2 + d2 B)

= a3 + b3B,

(7.33)

we can write Eq. (7.32) as

(#or(A1, Bo2)#o2(A2, B)) 2 = det(a3 + b3B)(az + bzA2)-1(al + blAa) -1 . (7.34)

Similarly, we can write

(#o~(A1,A302)po~(A 2, A3))/=det(a3+b3A3)(az+b2Az)-l(al

+ blA1)-i
(7.35)

Taking the ratio of Eqs. (7.34) and (7.35), we have

B)

(#o~(Al_,_B~)l~o2(A

,
,~z = det(a3 + b3A3)(a 3 + b3A3) -1
]'/01( A 1, A g~
3 )~02( A 2, a-3- ) / /

(7.36)

Pfaffian Line Bundle

483

so that
#gl(A1, Bg2)#g2(A2, B)

#o~(A1,A ~

"A A3)

= ___#03(A3, B ) .

(7.37)

To fix the sign, we simply let B = A3, and see that both sides are equal to 1. This
proves that
41 "(42" e o(n)) = 43" e 0(n) 9

(7.38)

The next step is to check that the action preserves the inner product. For
4 = (O, A, 2), we have
( 4 " e *(B1), r

*(B2)) = 12l 2#o(A, B1)#o(A,

B2)<e r

e *(B~)) .

(7.39)

As usual we first work with the square,

<e *(B]), eO(B~)> 2 = det(1 +

B~*B~

1 3"
By the definition of

B ~ we have
B9

=9

a + bB)-i ,

(7.41)

so that we get

<eg~(n~176

+ b B 1 )1 " [ 9 ( 1 2 ) ( a + b B 2 ) -1]

det(1 + B*B2)
det(a + bB1)*(a + bB2)

(eO(B~),e4a(B2)>2
det(a +

bB~)*(a + bB2)

(7.42)

Putting this together with Eqs. (7.22) and (7.24), we find that
( { ' e 4,(n'), 4" eO(B2)>2 = ___( e *(n'), e4'(B2)>2 .

(7.43)

If B1 is set equal to B2, we see that the sign is positive.

We have now checked the behavior of the action (7.30) for elements of the form
e *(B), where B lies in an open set of Vo. Lemma 7.1 tells us, however, that these
elements span all of ~ gr Since the action on these elements preserves the inner
product, it clearly extends to all of ~ g~". []
The initial form of the action of SO r ~ ( ~ ) on
was defined so as to
correspond to an action of SO r ~ ( ~ ) on P F ..... the Pfaffian line bundle restricted
to the even component of Gr1(~4~ We will now show how SOr,s(~/g) acts on PF ....
holomorphically. Given 4 = (9, A, 2), let U~ r denote the set of all B e VR such that
9(graph(B)) lies in Vr. For each R and T, we need to find a holomorphic function
fr on U RT such that
~'(B, 1)R =

(S',f~(B))r,

(7.44)

484

D. Borthwick

where B'e VT is the point g(graph(B)). From Theorem 7.3 we know already that
such a function exists when R = T = So, and is given by 2#g(A, B) in this case.
Lemma 7.4. Given R and T, and for a fixed ~ = (g,A, 2)e SOres(~'~), there is

a holomorphic function f~ on U~ T such that

9(B, 1)R = ~(B',f~(B))T,

(7.45)

for Be UgR T .
Proof. Suppose that Be U~ r corresponds to a point C e Ug. Using Eq. (7.30) and
the transition functions for PF, we have
~'(B, 1)a = ~'(C, hsoa(B))so
= (C a, 2pg(A, C)hsoR(B))so)
= (B', 2pg(A, C)hTso(Cg)hsoR(B))so).

(7.46)

The lemma will be proven if we can find an extension to all of UaoT of the function
fo (B) = 2/~g(A, C)hTso(Cg)hsoa(B))so),

(7.47)

which is well-defined and holomorphic on U 9R T c~ Ug. For, if such an extension

exists, it is unique, and by continuity the action of ~ on elements of P F must obey
the formula (7.45).
The extension of fo can in fact be written out explicitly. Let
g =

(7.48)

with respect to the decomposition Wa@ W ~

WR --* Wr, etc. Choose some D e U oRr , and let

W r O W~, e.g. art maps

car

dRr

f ( B ) = ~c Pf (D - B, (aRt + bRrD)-lbRr) .
w,

(7.49)

Because the set U~ r is defined precisely by the condition that (aRt + barD) be
invertible, this function is holomorphic on all of U~ r. For the moment, ~cand D are
arbitrary.
Because ~ preserves the inner product of ~ ven, we know that
Pfw,(JBJ, B)
] f o ( B ) [ 2 __

(7.50)

Pfw~(JB'J, B')"

By manipulations similar to those done to obtain Eq. (7.22) for #o, we have
]f(B)]2 = ][2 ~ detwR(aRT + bRTB)(aRT + bRTB)* 11/2

L detw,(aaT + bRTD)(aRT + bRrD)*

"

(7.51)

The arguments used in Eq. (7.42) can be repeated to give

Pfw~(JBJ, B)

det (aRT + bRTB)(aRT + bRTB)* = Pfw~(JB'J, B') "

~w,

(7.52)

Pfaffian Line Bundle

485

Combining Eqs. (7.51) and (7.52), we have

If(B)[ 2 =

[~:f~
2
detwR(aRT + bRTD)(aRT + bRrD)*

(7.53)

Because there are no non-constant holomorphic functions on Grt ( ~ ) , if the norms

of two holomorphic functions differ by a constant, the functions themselves differ
only by a constant. Thus, by choosing ~c appropriately, we have f = f o on
UgR T hUg. []
Theorem 7.5. SO~es(Yt~ acts on the even component of PF. The action is holomorphic
and linear on each fiber.
Proof The form of the action is given by (7.45). Lemma 7.4 tells us that the
functions f~ and holomorphic, so the action is holomorphic. []

We have now worked out the results for the even components of O~s(Yg) and
PF. We can apply them to the full spaces by means of a simple trick. We enlarge the
Hilbert space Yt~ slightly, and embed PF in the even component of the Pfaffian line
bundle for the new Hilbert space. Let
~'~:= ~ O C

2,

(7.54)

which we decompose as ~ + = Y~+ q) G. We define the complex structure J for

d~ by

Y(x, ,~):= ( Jx, 2),

(7.55)

for (x, 2) 9 Y#+ @ C. Starting with the pair Y~, d, we construct a line bundle t;F and
the corresponding Fock space ~-e -~ Y+.
Let 7 denote the representation of Ore~(Yt~) on 1122given by

?(g):= { I O
1 ) 101)"
0 f iotherwise.
Of g l i er s i n~ theidentity
(
~
)cOmpOnent
;
0
0

(7.56)

This gives us a homomorphism

O~e~(~) ~ SOr~s(~),

(7.57)

g ~ 0 := (g, 7(g)) 9

(7.58)

which maps

Using this map we make the following definition.

Definition 7.6. The extension ()res(~) is the pullback to O~es(~ ) of the extension
SO~es(Jt~) by the map (7.58).
This map (7.58) also defines an embedding of Grl(~f'~) in the even component of
G r / ( ~ ) , by
gJVt~~ 0~@.
(7.59)

486

D. Borthwick

There is a corresponding map of index sets S ~-~ S,

g:= ~(S, +),
(S,-),

ifdim(Sr~2~+) is even;
i f d i m ( S c ~ Z + ) is o d d ,

(7.60)

where the + and - refer to a basis { ~ + ) for C 2. We can use this to describe the
embedding of P F in P~F.... very simply.
(A, 2)s ~-* (A, 2)g,

(7.61)

where the map A is extended from Ws to Wg by zero. Recall that the elements of P F
can be identified with elements of the Hilbert space ~ + . The embedding (7.61) leads
to the following map ~ + ~ ~-~"",
gOeven ~
(l)od d ~

gOeven ,
~ + /X (Dod d .

(7.62)

This map is a Hilbert space isomorphism.

Theorem 7.7. The action Of Orcs(~ff) on G r I ( ~ ) is covered by a holomorphic action

of Orr
on PF, and this gives rise to a unitary representation of O~r
on

Proof Given the embeddings and isomorphisms defined above, this is a straightforward generalization of Theorem 7.3 and Theorem 7.5. []

8. Loop Groups
Let ~ ( " ) be the space J4~ | C", where ~ is the Hilbert space of square-integrable
half-densities on S ~. An element of oug(") can be thought of as a vector valued
function

f(O) = (f~ (0) . . . . , f , ( 0 ) ) ,

(8.1)

such that f(O + 2~) = - f ( O ) . This space has a natural basis given by e~j with
k = l . . . . . n and je7z+, where ejk corresponds to the function for which
f(O) = 6tkexp( _+ i(j -- 89 ~ ( " ) has its natural complex conjugation

J f(O) := f(O) .

(8.2)

We can construct an isomorphism ~(~") -~ oct~ by mapping e~j ~ e+_t,{j_l)+k),

which clearly preserves the action of J on the two spaces.
The group LO, of smooth loops in O, acts naturally on J4~{"). For 7~
L O , , f e ~r~"), we set
(Tf)z(0) := ~ 7ij(O)fj(O).
J
Because 7 is real-valued, we clearly have JyJ = 7.

(8.3)

Proposition 8.1. Equation (8.3) defines an embedding LO, ~ Orcs(~ (")) (and hence

in O~cs(~)).
Proof We have already noted that J?J = 7- For the remainder of the proof, the
corresponding result for unitary matrices done in [21] applies directly. []

Pfaffian Line Bundle

487

The pullback through this embedding of the extension 0 ~ ( ~ ) gives us cyclic

extensions L O , of LO,. To determine what these extensions are we first compute
the Lie algebra cocycle of the extension 0ro~(~). The Lie algebra o~(~r
of
O~,~(~) is given by
or~s(W) -- { q e ~ ( W ) :

q* = - r/,Jr/J = rl, P+r/P_

and P_r/P+ are Hilbert-Schmidt} .

It is easy to verify that an element of orodJg ) must have the form

( v w)
r~ =

-- w*

JvJ

'

with respect to the decomposition W_ G W+,

I~(~+, ~_).

where v * = -

v, and we

To compute the Lie algebra cocycle, we need to find a cross section of the
extension 0ro~(~) over a neighborhood of the identity. For our neighborhood, we
will use the set U consisting of all g e Ores(W) for which a is invertible, when g is
written in the usual form (7.4). The cross section U ~ Oros(OVg) is
g ~ 0 := (9, 0, [det a * a ] 1/4).

(8.4)

We are looking for as m a p c : U x U ~ C such that

0102 = c(gl, 92)03,
when gig2

(8.5)

g3. This m a p is given by

21

c ( g t , g 2 ) = I d e t a ~ a l deta*a2 ]t/4
deta~a3
Pf(-c2a
W+

,a;lbx).

(8.6)

Given such a map c, the general formula for the Lie algebra cocyle of the
extension is

co(ql, rlz) = Di D2c(rh, ~I2) -- Di D2c(r/2, r/1) ,

(8.7)

where D~Dac: o r , s ( ~ ) x Ores(OVf)~ C is the mixed second partial derivative of c at

the identity. We compute that for r/a,/12 E Ores(~),

c(e"', e "~) =

Pf(w*, Wl)
+ (9(r/2, r/2)
[det(1 - w ' w 1 - w ' w 2 ) ] 1/4
exp

f - 1

= 1 + ~ T r ( w * w 2 - W*Wx) + . . . .

...

(8.8)

Thus we read off that

1

co(r/a, I/2) = ~ T r ( w a w2 -- w ~ w l ) .

(8.9)

488

D. Borthwick

We can think of this cocycle as being the Chern class of the unit circle bundle in PF,
which is identified with Or~(~)/U(J4Q) [21].
%
The extensions LO, will be completely determined by their Lie algebra
cocycles, which we can now specify.
Proposition 8.2. The extension L'O, given by the pullback Of Or~( ~Ut~) corresponds to

the Lie algebra cocycle

i 2~

s(~,/~) = U~ ! <~(0), ~'(O))so.dO,

(8.10)

where ~, fle Lso,, and ( X, Y)~o, = - tr X Y .

Proof Let the Fourier decompositions of c~ and fl be
a(O) = ~ e(k)eikO,
kE2g
fl(0) :

E fl (k)eikO"
ks7]

(8.11)

We simply compute the induced cocycle

s(~,/~) = co(~, ~ )
I
= ~ Tr (t/* P_ a~ -- t/~P_ t/~)
.~+

= 2k,~>O tr(a(-k-~+l)*fl(-k-"+l) -- fl(-k-m+l)*a(-k-,,+l))

1

2 E ktr(o~(k)fl(-k)--fl (k)O~(-k))

k>O

2 ~ ktra(k)fl(-k)
keZ
9

=~

2~

! (a(O),fi'(O))~o dO.

[]

Theorem 8.3. The loop groups, LO, act on o~p by irreducible projective unitary

representations, corresponding to the cyclic extensions defined above.

Proof The only aspect of this theorem which has does not follow immediately
from our previous results is the irreducibility of the representations 9 The proof or
irreducibility can be taken virtually unchanged from the corresponding proof for
LU, in [21J, so we will not repeat the arguments here. []
We mentioned in Sect. 4 that the determinant line bundle can be pulled back
from the Pfaffian line bundle defined over a larger Hilbert space f = W @ ~'~'.
The embedding of the two line bundles corresponds to an embedding of U,es(Jf) in
Ores(:g)- The subgroup LU, of U r e s ( ~ ) maps to LO2n under this embedding,

Pfaffian Line Bundle

489

precisely by the canonical map U, ~ 02,. Thus, we see that if we apply the above
results for LOznt o L U , via the embedding U, ~ O2,, we obtain the same extensions
and projective representations which were obtained in [21] using the determinant
line bundle construction.

9. The Pfaffian Line Bundle Over Moduli Space

The determinant line bundle of Quillen [22], which involves the determinant lines
of Dirac operators, has been studied extensively. Its differential geometry [7] and
holomorphic structure [8] have been worked out. As mentioned in the introduction, the construction has proven extremely useful in string theory and conformal
field theory. Freed has also given a construction of a Pfaffian line bundle which is
a square root of the Quillen determinant line bundle [12]. We will attempt here to
relate his Pfaffian line bundle to our construction. We start by outlining his
construction briefly.
Let J//o be the moduli space of Riemann surfaces of genus g together with
a choice of spin structure. An element ofJ/g0 is thus a pair (Z, S), with S a holomorphic line bundle on 2: such that
8 2 "~ T * 2 : 1 ' ~

(9.1)

Given such a pair, we define the Dirac operator D to be the { o p e r a t o r acting on

sections of S,
D := ~-s: Y2~~

--* ~o, 1(S) .

(9.2)

Note that because of the relation (9.1), we have a natural bilinear pairing

~2o,o(s ) | ~ o , l ( s ) ~ ~ ,

(9.3)

given by integration over 2:. This pairing will play the role of the bilinear form ( ' , 9)
on At, and its existence is the reason for requiring S to be a spin bundle. The papers
dealing with the determinant line bundle and the Krichever map use a more
general moduli space, involving an arbitrary line bundle [2, 5, 6, 15, 25].
It is a special case of the results of [8] that the complex (9.2) varies holomorphically over J/lg. Given a point m e J/L0, we can choose a finite-dimension subspace
V of the bundle ~2o, 0(S) which varies holomorphically and for which Ker D c V at
the point m. Then V necessarily contains Ker D in some neighborhood U of m.
Over this neighborhood, we define the holomorphic line bundle
~z"U : =

A V' ,

where /x designates the highest exterior power.

These locally defined line bundles can be patched together as follows. Suppose
that V1 c V2 are finite dimensional subbundles of ~2~176 defined on open sets
U1 c U2. Let X = V2/Va. By construction, D has no kernel when restricted to X.
Via the pairing (9.3), we see that D corresponds to an element of C~De X ' | X',
given by
coD(a, fl) = S ~ | Dfi .
Z

(9.4)

490

D. Borthwick

We see through integration by parts that coo is skew, and so belongs to

n o := 1 0 ) 5 ,

A2X

'.

Let
(9.5)

I ',

where 2r = rank X (we know that the rank of X is even because D : X ~ X ' is both
invertible and skew-symmetric), n o gives us an isomorphism
g(Cvl ~ 3((v2 ,

(9.6)

s ~ s/x nD 9

(9.7)

defined over U~, which is

Because the complex (9.2) varies holomorphically, the form n varies holomorphically over U1. Thus the patching maps defined by (9.7) are holomorphic, and the
result is a holomorphic line bundle oU. A canonical holomorphic section, which
vanishes where D has a kernel, is easily obtained. We can define nDe ^ V' for any
subbundle V, just as above, and we define the section locally by this form. The line
bundle Y and its canonical section are the square roots of the determinant line
bundle and its holomorphic section.
To make contact with the Pfaffian over a Grassmannian manifold, we must
enlarge the moduli space to specify a coordinate patch and describe a variant of the
Krichever map [18]. Krichever's original map was from a moduli space to the
space of solutions to the KdV equations. The equivalent map to the Grassmannian,
which we use, was introduced in [23, 25]. Let M/0 be the space consisting of a pair
(Z, S) ~ J/go together with a local coordinate patch z on Z. By local coordinate patch
we mean an invertible holomorphic map from an open set of 2; to an open
neighborhood of the unit disk in IlL M/0 can be given the structure of an infinite
dimensional complex manifold [4]. Denote by rc the natural projection ~/r o ~ Jr
Let ~ be the Hilbert space we have mentioned previously, the space of
square-integrable function on S 1 with anti-symmetric boundary conditions. For
each point (S, S, z) E Jgo, we can define a map from n ~ ~
~ o~, as follows. In our
coordinate patch, x / ~ gives a trivialization of S. For and co e n ~ ~
we can find
a function f o n an open neighborhood of the unit disk in 1[; so that locally co has the
form
e) = f , , f ~ .

(9.8)

We define the map n ~ 1 7 6 --* Jr" to be

o~ ~

c~,o(O) = f ( ei~

I~ .

(9.9)

Note that for v , o ~ e n ~ 1 7 6

the product v|
can be regarded as a form in
nl,~
so that ~-(v | o~) is a volume form on Z. Let O be the disk Iz] < 1. By
Stokes' theorem we have
S ~(~|
~'- 0

~ 4~(O)c~(O)dO
00

= (q~v, qSo)xe
(recall that (x, y ) ~ = ( Jx, y >~).

(9.10)

Pfaffian Line Bundle

491

Proposition

9.1. For rh = (X, S, z)~ J/{g, let W~a be the subspace of fig which is the
closure of the set,
{f~'~:f=

~ for some c o ~ F ( S I z - o ) } ,

where F(SI2-o) is the space of holomorphic sections of S over Z assignment rh ~ W,~ defines a map

(9.11)

O. Then the

k: ~//{0~ Gr1(J4~),

(9.12)

which is injective and holomorphic.

Proof The fact that k defines an injective map to the Grassmannian was proven in
[-25] (in greater generality), and the holomorphicity of this map has been discussed
in [5, 15]. Therefore we will only demonstrate that the image of k lies in Grl (W) for
our particular moduli space. Fix rh and let W = W,~. We will show that the
projection P_ : W-~ W_ has index zero, and that W is isotropic with respect to

(.,-).
To prove the first, let ,!;o be the patch where the coordinate z is defined, and let
2;~ denote the set X\{z = 0}. For any Riemann surface we have an exact sequence,
0 ~ H~

S) ~ r(Sl2o) | r(Slz=)-~

r(Slzo~Zo)-~ H~(Z, S) ~

O,

(9.13)

where F denotes the space of holomorphic sections, and H * ( Z , S) is the cohomology of Z with values in the sheaf of holomorphic sections of S. From the Dolbeault
theorem, we have

H~

S) ~- KerD,

H i ( Z , S) ~- C o k D .

(9.14)

It is also clear that

r(Sl~o) ~ ~ ,
r(Sl~)
r(Slzo~)

~ w an ,
~ ~an

(9.15)

where the superscript an refers to the subspaces of analytic functions. The map
W an

(~ o ~

-'-* ~ a n ,

(9.16)

appearing in the exact sequence above takes ( f g ) ~ - * f - g . Its kernel is

KerP_: W"-~ ~,
which is just K e r P _ , since all functions in this kernel are
analytic. Similarly, the cokernel of the map (9.16) is just C o k P _ . From the
sequence (9.13) we thus obtain the exact sequence
0 ~ Ker D ~ W ~ ~ _ --* Cok D --* 0 .

(9.17)

From this exact sequence, we read off that P_ is Fredholm with index zero, since
the Dirac operator D has index zero. We note that the arguments which show that
the complex (9.2) varies holomorphically over ~'o can be applied to see that (9.17)
varies holomorphically over Jr
To prove the isotropy of W, we use integration by parts and Stokes' theorem.
Suppose that f, g ~ W, and let f = ~b~,9 = ~b~,. We have already noted that
( f g) =

~ 8-(v| co).

(9.18)

2-0

Since v and co are holomorphic, this is equal to 0. Hence, W is isotropic.

[]

492

D. Borthwick

This type of Krichever map is essential to the Fock space functor construction
of [24]. We use it to give the relationship between the two Pfaffian line bundles in
the following theorem. The corresponding result for the determinant line bundle is
well known [-5, 15].

Theorem 9.2. With ~z the projection J/gg--* d/do, and k the map defined in Proposition
9.1, we have the holomorphic isomorphism
~z*X ~ k * P F ' .

(9.19)

Proof The proof is easy if we use a slightly different definition of the Pfaffian line
bundle. Given a point W e G r 1 ( ~ ) , we can choose a finite dimensional subspace
V c W such that Ker P_ c V in a neighborhood U of W. Then we define P F
locally by
P F v := U x A V' .

(9.20)

These definitions are patched together using the bilinear form ( ' , -) and the m a p
P_, which is skew with respect to the pairing, in exactly the same way we patched
S using the pairing (9.3) and the Dirac operator. It is easy to check that this
definition is equivalent to the one given in Sect. 4. The parallel to the construction
of ~ff is obvious. All that is needed to complete the isomorphism is the exact
sequence (9.17), which varies holomorphically over ~ g . []
This theorem relates the two bundles PF and ~ . The latter bundle has
a canonical holomorphic section, which is thought of as the Pfaffian of the Dirac
operator. Its defining property is that it vanishes precisely when the Dirac operator
has zero eigenvalues, and the order of vanishing is equal to half the number of zero
modes (the zero modes come in pairs because D is always skew symmetric). It is
easy to pick out the section of P F ' which has this property when pulled back by k,
because zero modes of D correspond to elements of W ~ ~ + . The holomorphic
section YSo of PF', in the notation of the first few sections, pulls back to the
canonical section of oU. This is very natural, because in the Fock space interpretation of the space of holomorphic sections of PF', 7so corresponds to the vacuum
state.
Unfortunately, the situation with regard to metrics is not so clear. It would be
nice to be able to relate the metrics on the line bundles oU over the moduli spaces
for different genera to a single metric coming from PF, but this is impossible. The
metric on ~ is independent of any local coordinate patches. By choosing an
arbitrarily small coordinate patch on a higher genus surface, we can bring the
corresponding point in G r 1 ( ~ ) arbitrarily close to the point ~ _ , which corresponds to the sphere. Thus no smooth hermitian structure on PF can be pulled back
to give the canonical metrics on S over moduli spaces for different genera.

Acknowledgements. I wish to thank Arthur Jaffe and Andrzej Lesniewski for many helpful
discussions and much encouragement.
References
1. Alvarez-Gaum6, L., Moore, G., Vafa, C.: Theta functions, modular invariance, and strings.
Commun. Math. Phys. 106, 140 (1986)
2. Alvarez-Gaum6, L., Gomez, C., Moore, G., Vafa, C.: Strings in the operator formalism. Nucl.
Phys. B303, 455-521 (1988)

Pfaffian Line Bundle

493

3. Alvarez-Gaum6, L., Gomez, C., Reina, C.: Loop groups, Grassmannians, and string theory.
Commun. Math. Phys. 117, 1 36 (1988)
4. Arbarello, E., DeConcini, C., Griffiths, P.A., Harris, J,: Geometry of algebraic curves. Vols. I,
I1. Berlin, Heidelberg, New York: Springer 1985
5. Arbarello, E., DeConcini, C., Kac, V.G., Procesi, C.: Moduli Spaces of Curves and Representation Theory. Commun. Math. Phys. 117, 1-36 (1988)
6. Beilinson, A.A., Schechtman, V.V.: Determinant bundles and Virasoro algebras. Commun.
Math. Phys. 118, 651-701 (1988)
7. Bismut, J.-M., Freed, D.: The analysis of elliptic families I. Commun. Math. Phys. 106,
159-176 (1986); The analysis of elliptic families II. Commun. Math. Phys. 107, 103-163 (1986)
8. Bismut, J.-M., Gillet, H., Soule, C.: Analytic torsion and holomorphic determinant bundles
I-III. Commun. Math. Phys. 115, 49-78, 79 126, 301-351 (1988)
9. Bott, R.: Homogeneous vector bundles. Ann. Math. 57, 203 248 (1957)
10. Connes, A.: Non-commutative differential geometry. Publ. Math. I.H.E.S. 62, 41 (1984)
11. Freed, D.: Determinants, torsion, and strings. Commun. Math. Phys. 107, 483-513 (1986)
12. Freed, D.: On determinant line bundles. In: Mathematical Aspects of String Theory. San
Diego, 1986
13. Jaffe, A., Lesniewski, A., Weitsman, J.: Pfaffians on Hilbert space. J. Funct. Anal. 83, 348-363
(1989)
14. Kac, V.G., Peterson, D.H.: Spin and wedge representations of infinite dimensional Lie
algebras and groups. Proc. Natl. Acad. Sci. USA 78, 3308-3312 (1981)
15. Kawamoto, N., Namikawa, Y., Tsuchiya, A., Yamada, Y.: Geometric realization of conformal
theory on Riemann surfaces. Commun. Math. Phys. 116, 247 308 (1988)
16. Klimek, S., Lesniewski, A.: Pfaffians on Banach spaces. J. Funct. Anal. 101 (1991)
17. Kontsevich, M.L.: Virasoro algebra and Teichmfiller spaces. Funct. Anal. Appl. 21, 78-79
(1987)
18. Krichever, I.M.: Integration of non-linear equations by methods of algebraic geometry.
Funct. Anal. Appl. 11, 15-3l (1977) (Russian), 12 26 (English); Methods of algebraic geometry
in the theory of non linear equations. Usp. Math. Nauk 32, 183-208 (1977); Russ. Math. Surv.
32, 185-214 (1977)
19. Krichever, I.M., Novikov, S.P.: Virasoro type algebras, Riemann surfaces, and structures of
soliton theory. Funct. Anal. Appl. 21, 46~63 (1987); Virasoro type algebras, Riemann surfaces,
and strings in Minkowsky space. Funct. Anal. Appl. 21, 294-307 (1987); Virasoro type
algebras, energy-momentum tensor, and decomposition operators on Riemann surfaces.
Funct. Anal. Appl. 23, 19-32 (1989)
20. Mickelsson, J., Rajeev, S.G.: Current algebras in d + 1 dimensions and determinant bundles
over infinite dimensional Grassmanians. Commun. Math. Phys. 116, 365-400 (1988)
21. Pressley, A., Segal, G.: Loop Groups. London, New York: Oxford University Press 1986
22. Quillen, D.: Determinants of Cauchy-Riemann operators over a Riemann surface. Funct.
Anal. Appl. 19, 37~41 (1985)
23. Sato, M.: Soliton equations as dynamical systems on infinite dimensional Grassmann manifolds. RIMS Kokyuroku 439, 30-40 (1981)
24. Segal, G.: The definition of conformal field theory, unpublished manuscript
25. Segal, G., Wilson, G.: Loop groups and equations of KdV type. Publ. Math. I.H.E.S. 61, 5 64
(1985)
26. Witten, E.: Quantum field theory, Grassmannians, and algebraic curves. Commun. Math.
Phys. 113, 529 600 (1988)
Communicated by A. Jaffe