CONTACT DEGREE AND THE INDEX OF FOURIER

INTEGRAL OPERATORS

CHARLES EPSTEIN AND RICHARD B. MELROSE

Date: Wed May 27 15:55:55 1998 ; Run: May 27, 1998

Abstract. An elliptic Fourier integral operator of order 0, associated to a

homogeneous canonical diffeomorphism, on a compact manifold is Fredholm
on L2 . The index may be expressed as the sum of a term, which we call
the contact degree, associated to the canonical diffeomorphism and a term,
computable by the Atiyah-Singer theorem, associated to the symbol. The
contact degree is shown to be defined for any oriented-contact diffeomorphism
of a contact manifold and is then reduced to the index of a Dirac operator on
the mapping torus, also computable by the theorem of Atiyah and Singer. In
this case, of an operator on a fixed manifold, these results answer a question
of Weinstein in a manner consistent with a more general conjecture of Atiyah.

Introduction
Let X be a compact contact manifold, with oriented contact line bundle L ⊂
T ∗ X. Let M(X) be the contact mapping class group of X; that is the group of com-
ponents of the space of contact diffeomorphisms. We construct a homomorphism
which we call the contact degree
(1) c-deg : M(X) −→ Z.
This construction is based on the notion of a quantization of the contact structure
introduced by Boutet de Monvel and Guillemin [5]. In particular if the hyperplane
bundle, W = L◦ on X, is given an almost complex structure which is positive with
respect to the conformal symplectic structure and X is given a compatible partially
Hermitian metric then in [5] generalized Szegő projections are shown to exist. Al-
though the analysis in [5] is in terms of the Hermite calculus, these projections also
lie in the Heisenberg calculus discussed by Beals and Greiner [1] and Taylor [18],
and originally described by Dynin [6]. Extending an earlier idea of the second au-
thor [8] (for the integrable, that is CR, case) we introduce below the relative index
of two such generalized Szegő projections as the index of the Fredholm operator
which is their composite acting between their ranges:
(2) ind(S0 , S1 ) = ind(S1 S0 : Ran(S0 ) −→ Ran(S1 )).
We show below in Proposition 2 that this relative index faithfully labels the com-
ponents of the space of generalized Szegő projections. The action of the group of
contact diffeomorphisms, by conjugation, on the space of generalized Szegő projec-
tions induces the homomorphism (1),
(3) c-deg(φ) = ind(S, Sφ ), Sφ = (φ∗ )−1 Sφ∗ .
1
2 CHARLES EPSTEIN AND RICHARD MELROSE

Let Zφ be the mapping torus of φ, i.e. X × [−1, 1] with the ends identified by
φ. The contact structure on X gives Zφ a natural Spin-C structure. Let ðφ be the
associated Dirac operator.
Theorem 1. For any oriented contact diffeomorphism of a contact manifold the
contact degree is given by the index of the Dirac operator associated to the Spin-C
structure on the mapping torus, that is
c-deg(φ) = ind(ðφ ).
This is proved by first exhibiting the contact degree as the spectral flow of a
family of Dirac operators on the contact manifold. The space of Dirac operators
associated with partial Hermitian structures on the contact manifold is contractible.
Since these Dirac operators are elliptic and self-adjoint, the spectral flow along any
curve connecting one such Dirac operator to its φ-conjugate is well defined.
Theorem 2. For any contact diffeomorphism, c-deg(φ) is the spectral flow of the
curve of Dirac operators on the contact manifold associated with an isotopy from
any one partial Hermitian structure to its φ-conjugate.
Theorem 1 follows from this by a suspension argument.
The contact degree is directly related to a long-open question of Weinstein [20, 19]
asking for a geometric formula for the index of elliptic Fourier integral operators.
For such operators acting on a fixed manifold the theorem above provides an answer.
Namely, Zelditch (see [19], [21] and [22]) had observed that the properties of the
integral transformation studied by Guillemin [12] show the equality of the index of
the Fourier integral operator and the relative index of the Szegő projection on S ∗ X
and its φ conjugate. Combining these results we deduce
Theorem 3. If Y is a compact manifold and X = S ∗ Y is its cosphere bundle then
for any (oriented) contact diffeomorphism, φ, of X, i.e. a canonical diffeomorphism
of T ∗ X \ 0,
ind(Fφ ) = c-deg(φ)
where Fφ is a Fourier integral operator (see [13]) associated to φ and with ellip-
tic symbol corresponding to the positive trivialization of the Maslov bundle, hence
Fredholm on L2 (Y ).
An explicit cohomological formula for this index, which follows from the Atiyah-
Singer index theorem, is discussed in §8 below. We leave open, for the moment, the
part of Weinstein’s question concerning the index of elliptic Fourier integral oper-
ators between different manifolds. It is highly likely that methods closely related
to those used here can be applied to the general problem and provide an answer
in terms of the conjecture of Atiyah. Note that it is easily seen that our present
formula is consistent with that conjecture (see [19]).
The central result here, which is the passage from Proposition 4, essentially a
restatement of (3), to Theorem 2, is closely related to the homotopy argument for
the Toeplitz index given by Boutet de Monvel in [4]. Indeed his Toeplitz index
theorem can be proved in essentially the same way, and so made independent of
the Atiyah-Singer index theorem for pseudodifferential operators. This provides
an analytic alternative to the usual K-theory path from the theorem for Dirac
operators to the general case of pseudodifferential operators.
 CONTACT DEGREE 3

The authors would like to thank Ana Cannas da Silva, Yasha Eliashberg, Victor
Guillemin, Eric Leichtnam, Laszlo Lempert, Rafe Mazzeo, Tomasz Mrowka, Victor
Nistor, Paolo Piazza, Alan Weinstein and Steven Zelditch for helpful discussions
related to this work and Louis Boutet de Monvel and Sergiu Moroianu for useful
comments on the manuscript.

1. The analytic construction

For the treatment of (generalized) Szegő projectors and related ideas we refer
to the book of Boutet de Monvel and Guillemin [5] and references therein and for
the ‘Heisenberg algebra’ of a contact manifold to that of Beals and Greiner [1] (see
also the book of Taylor [18]). The properties of the Heisenberg algebra and the
extended Heisenberg algebra will be discussed more fully in a forthcoming paper of
G. Mendoza and the present authors, see also [8] and [16].
Let Ψ0He (X) be the Heisenberg algebra, of ‘parabolic’ pseudodifferential operators
0
associated to the contact structure on X. This has a natural ideal, IHe (X) ⊂
0
ΨHe (X) consisting of those elements with full symbols which vanish in the lower
half of the cotangent bundle. The (non-commutative) principal symbol map for the
Heisenberg calculus gives a short exact sequence of algebras
−1 0
(4) 0 −→ IHe (X) −→ IHe (X) −→ S(W̃ ) −→ 0.
Here W̃ is a vector bundle isomorphic to T ∗ X/L. The product on the Schwartz
space S(W̃ ) is local to each fiber and is given there by the usual ‘pseudodifferential’
product which can be written in terms of the differential of a contact form
a#b = eidα(D) a ⊗ b

Diag
.
This is isomorphic to the operator product on S(R2n ) acting as kernels of operators
on Rn .
The choice of a positive almost complex structure and admissible metric on
T ∗ X/L induces harmonic oscillators on the fibers of W̃ . The ground state is a
positive Schwartz function on Rn . Let s ∈ S(W̃ ) be the projection onto it. The
set of these projections, for different choices and corresponding ground states, is
simply connected. Boutet de Monvel and Guillemin show that s can be lifted
0
to a projection in IHe (X), a generalized Szegő projection, or ‘quantized contact
structure.’ We recall and slightly extend their result as
Proposition 1. For any smooth isotopy of ground-state projections st , t ∈ I, aris-
ing from an isotopy of partial Hermitian structures, there is a smooth isotopy St of
generalized Szegő projections with Heisenberg symbols st . Two projections with the
same symbol differ by an operator of negative order. One can be deformed, through
Szegő projections, to differ from the other by a finite rank self-adjoint projection.
Proof. A global quantization map for the Heisenberg calculus is given in [10, 16]
and this shows immediately that the smooth curve st can be lifted to a smooth
curve S̃t in the Heisenberg calculus. These operators satisfy
−1
S̃t2 − S̃t = Et (1) ∈ IHe (X) ⊂ Ψ−1
H (X).
This also shows the uniqueness of the projection up to terms of negative order. The
spectrum of S̃t is discrete outside 0 and 1. Furthermore S̃t can be replaced by its
self-adjoint part so the spectrum is real. Using Cauchy’s theorem to replace S̃t by
the projection onto the span of the eigenspaces for eigenvalues larger than a ∈ (0, 1),
4 CHARLES EPSTEIN AND RICHARD MELROSE

for some a not in the spectrum, gives a locally smooth family of generalized Szegő
projections near any t. The existence of a global representation easily shows that
this integral is in the Heisenberg calculus.
Thus, we may cover the parameter interval, I by subintervals, Ij , j = 0, . . . , m
such that only consecutive intervals intersect and on each interval we have a self-
adjoint projection, Stj , t ∈ Ij , which differs from S̃t by a compact operator. From
the construction, on each overlap, Ij ∩ Ij+1 , either Stj < Stj+1 or vice versa, with
the differences smoothing operators. As both the nullspaces and ranges of these
projectors are infinite dimensional we can modify the projectors, working from left
to right, by adding or subtracting finite rank, self-adjoint, smoothing projections so
that after finitely many modifications we obtain a smooth family of projections, St ,
t ∈ I. The difference St0 − St is a smooth family of Heisenberg operators of negative
order. This proves the first claim.
Consider two generalized Szegő projections S0 , S1 with the same symbol. They
differ by a negative order, hence compact, operator so S1 − S0 = A + B where A
is a finite rank smoothing operator and B is a self-adjoint operator with norm less
than 1/10 as an operator on L2 (X). It follows that the family S1 − tB, t ∈ [0, 1],
consists of operators with real spectrum never equal to 1/2. Projecting this family
onto the part of the spectrum greater than 1/2 therefore gives a smooth isotopy
of projections from S1 to S00 where S0 and S00 differ by an operator of finite rank.
Let R = Ran S00 ∩ Ran S0 and N = Null S00 ∩ Null S0 ; both are subspaces of finite
codimension in their respective factors, invariant under both projections. In fact

S0 R⊕N = S00 R⊕N .

 

Thus R ⊕ N is a subspace of finite codimension and its orthocomplement,

 K isan
invariant subspace for both S0 and S00 . Suppose that

dim Ran S  0
0 K ≥ dim Ran S0 K .
0
It is an elementary fact that we can deform S0 K through self-adjoint projections
to a subprojection of S0 K . This completes the proof of the second claim.

For any two such projections S, S 0 (possibly with different choices for symbols
s) the composite S 0 S is Fredholm as a mapping from the range of S to the range
of S 0 . This follows from the fact that, since all ground states are positive, the inner
product of the ground states of any two harmonic oscillators on Rn cannot be zero.
Thus the composite symbol s0 s is an isomorphism from the range of s to that of s0 ,
see [22]. We define the relative index as

ind(S, S 0 ) = ind(S 0 S : Ran S −→ Ran S 0 ).

From the discussion in the proof of Proposition 1 above it follows that this index
equals the difference of the ranks when the second projection is deformed, through
projections, to be equal to the first up to finite rank. In particular the relative
index assumes all integral values and its vanishing is equivalent to the existence
of an isotopy between the two projections. Gathering these conclusions we have
shown

Proposition 2. The set of connected components of the space of generalized Szegő

projectors on a compact oriented contact manifold is isomorphic to Z with the nat-
ural Z action given by the relative index; this Z action induces the homomorphism
c-deg .
 CONTACT DEGREE 5

It also follows that the relative index satisfies the cocycle condition
(5) ind(S1 , S2 ) + ind(S2 , S3 ) = ind(S1 , S3 ).
Louis Boutet de Monvel has pointed out that this cocycle condition is also just
the additivity of the index. Namely the left side is the index of the composite
S3 S2 ◦ S2 S1 = S3 S2 S1 from Ran(S1 ) to Ran(S3 ). Now S2 may be deformed to
S1 , through a family of Szegő projections up to smoothing errors. Under this
deformation S3 St0 S1 remains Fredholm as a map from Ran(S1 ) to Ran(S3 ), and
so has constant index. At the endpoint, S3 S12 = S3 S1 , so the index is equal to
ind(S1 , S3 ).
Using somewhat different, though related, techniques the first author had ear-
lier defined this relative index for a pair of Szegő projectors induced by a pair of
embeddable, strictly pseudoconvex CR–structures with the given underlying con-
tact structure, see [8]. In that paper the opposite sign convention is used in the
definition of ind(S, S 0 ).
0
Now if φ is a contact diffeomorphism of X and S ∈ IHe (X) is any choice of
∗ −1 ∗
generalized Szegő projection then Sφ = (φ ) Sφ is another generalized Szegő
projection, associated to the image under φ of the Hermitian structure. If S 0 is
another choice of generalized Szegő projection then the conjugation invariance of
the index shows that ind(Sφ0 , Sφ ) = ind(S 0 , S). The cocycle condition then shows
that the ‘contact degree’ defined by
c-deg(φ) = ind(S, Sφ )
is independent of the choice of S. The stability of the index shows that it is an
homotopy invariant of φ, hence defined on M(X). If ψ is a second contact dif-
feomorphism then Sφ◦ψ = (Sψ )φ . In view of (5) the contact degree is therefore
multiplicative,

c-deg(φ ◦ ψ) = ind(S, Sφ◦ψ ) = ind(S, Sψ ) + ind(Sψ , (Sψ )φ ) = c-deg(ψ) + c-deg(φ).

Thus we have defined the homomorphism (1).

2. Extended Heisenberg calculus

In the homotopy arguments below we use the ‘extended Heisenberg algebra.’ We
note here some of its properties; for more details see [10]. We use both the algebra,
ΨZ,Z
eH (X; E) of operators of integral double orders and the more general spaces with
0
real orders Ψm,meH (X; E), all well defined acting on sections of any vector bundle,
E, over the contact manifold. Two closely related algebras, (one non-classical and
one not complete) were introduced by Taylor [18]; the use of his algebras would
suffice, with some additional complexity in the arguments, for our purposes here.
In fact it is likely that the homotopies could be controlled in the ‘standard’ algebra
of pseudodifferential operators of type 21 , much as is done by Boutet de Monvel in
[4] but this would involve careful examination of commutators.
The main property of ΨZ,Z eH (X; E) is that it contains both the traditional algebra
Z
Ψ (X; E), defined without any reference to the contact structure, and the Heisen-
berg algebra ΨZH (X; E) which has parabolic homogeneity and for the properties of
which we refer to [1], [18]. Of the two indices, the first is ‘traditional’ order and
6 CHARLES EPSTEIN AND RICHARD MELROSE

the second the Heisenberg order:

(6) Ψk (X; E) ⊂ Ψk,2k m m,m
eH (X; E), k ∈ R, ΨH (X; E) ⊂ ΨeH (X; E), m ∈ R.

As explained in [10], all three ‘full symbol algebras,’ the quotients by the ideal
of smoothing operators, can be identified with non-commutative products on the
spaces of Laurent series ‘at infinity’ for different compactifications of T ∗ X. The stan-
dard calculus requires the fiber-wise radial compactification of T ∗ X; the Heisenberg
calculus uses the parabolic compactification defined by the contact line bundle. For
the extended Heisenberg calculus we begin with the radial compactification and
parabolically blow up the submanifold of the boundary defined by the contact line
bundle; this is the eH-compactification. We denote the new boundary face (which
has two components) by BL and the lift of the radial boundary by BS . The inclu-
sions in (6) correspond to the natural maps between the three compactifications.
The algebra ΨeH (X; E) is a bi-graded algebra of operators on C ∞ (X; E) :
0 0 0 0
Ψk,l k ,l k+k ,l+l
eH (X; E) ◦ ΨeH (X; E) = ΨeH (X; E), k, l, k 0 , l0 ∈ R.
In fact, ΨZ,Z
eH (X; E) is a smooth completion of the subalgebra generated by the
two subalgebras in (6) and its symbolic properties strongly reflect this. The ‘stan-
dard symbol’ map extends to a homomorphism giving a short exact sequence,
σ
0 −→ Ψk−1,l
eH (X; E) −→ Ψk,l S ∞ k l
eH (X; E) −→ C (BS ; Hom(E) ⊗ GS ⊗ GL ) −→ 0.
The bundle GS is the inverse of the conormal bundle to BS and GL is the inverse
of the conormal bundle to its boundary. It may be identified with the restriction
to BL ∩ BS of the inverse of the conormal bundle to BL . The composition rule for
this symbol map is given by point-wise multiplication of functions.
The ‘Heisenberg symbol’ is a little more complicated to describe since it is non-
commutative. The second boundary hypersurface, BL , of the eH-compactification
of T ∗ X can be identified with two copies of the radial compactification of the
dual bundle W ∗ to the hyperplane bundle W ⊂ T X which is the annihilator of the
contact line bundle. The interior of each component of BL is naturally a symplectic
vector bundle over X and this means that the ‘isotropic’ pseudodifferential algebra
of operators on Rn , which is a noncommutative product on C ∞ (B2n ) depending
only on the symplectic structure of R2n ,→ B2n can be transferred naturally, and
smoothly, to the fibers of BL . In this sense the Heisenberg algebra defines a short
exact sequence,
σ
0 −→ Ψk,l−1
eH (X; E) −→ Ψk,l H ∞ k l
eH (X; E) −→ C (BL ; Hom(E) ⊗ GS ⊗ GL ) −→ 0.

Here, since BL has two components (one corresponding to the positive direction
of L and one to the negative direction) the symbol consists of two operators at
each point of X. Thus the Heisenberg symbol defines two smooth families of model
operators in the isotropic calculus; the composition rule for this symbol comes from
the composition of these operators. The bundle GL has a trivialization along the
fibers of BL , the constant sections of which commute with the product.
Jointly these two symbol maps capture both orders, and satisfy only the nat-
ural compatibility condition of equality at the corner. In particular the elements
of Ψ0,0 Z,Z 2
eH (X; E) are precisely the elements of ΨeH (X; E) which are L -bounded and
−1,−1
the subspace ΨeH (X; E) is the null space of the joint symbol and consists pre-
2
cisely of the elements of ΨZ,Z
eH (X; E) which are compact on L (X; E). Note that
k,l
the ellipticity of an element of A ∈ ΨeH (X; E) means the invertibility of this joint
 CONTACT DEGREE 7

symbol; this is equivalent to the ellipiticity in the usual sense for σS (A) but the
model operators defined by the Heisenberg symbol must be invertible on each fiber.
The ellipticity of σS implies, via the compatibility condition, the microlocal ellip-
ticity of the Heisenberg symbol. Its invertibility on L2 is therefore equivalent to
its invertibility in the isotropic algebra, and its inverse is automatically a model
operator.
The usual symbol of an element A ∈ Ψk (X; E) can be interpreted as a section
over S ∗ X, of the bundle Hom(E) ⊗ Gk , where G is the inverse conormal bundle to
S ∗ X viewed as the boundary of the radial compactification of T ∗ X. The bundle G
pulls back to the boundary BS ∪ BL of the extended Heisenberg compactification
as GS G2L and any section of Gk pulls back to a section of GkS G2k L which is in the
center of the symbol algebra on the Heisenberg face. Thus the usual symbol, σk (A)
determines both the standard symbol and the Heisenberg symbol. If ξ is a vector
field transversal to the hyperplane bundle and positive on the contact line then
σL (ξ) is a section of G2L which is positive on the positive side of BL and in the
1
center of the algebra. Thus |σL (ξ)| 2 can be used to trivialize GL .
For a CR manifold with Hermitian structure, ∂¯b and ∂¯b∗ are both operators of
order 1 in the Heisenberg, and hence in the extended Heisenberg, algebra. Their
symbols are creation and annihilation differentials. Acting on functions the Heisen-
berg symbol of b = ∂¯b∗ ∂¯b is the harmonic oscillator shifted by a constant. On the
positive side this makes the lowest eigenvalue 0; on the negative side the symbol is
strictly positive (as an operator). The Szegő projector is an element of Ψ−∞,0
H (X)
with Heisenberg symbol the projection onto the ground state of this harmonic os-
cillator on the positive side; since it is in the ideal corresponding to the positive
direction of its symbol vanishes identically on the negative side. Boutet de Mon-
vel and Guillemin in [5] construct analogues of these objects for a general contact
manifold (with oriented contact line), see Proposition 3.

3. Dirac operators
We use the conventions for Dirac operators from [15]. A discussion of odd and
even-dimensional Dirac operators, and the relationship between them as in the
Atiyah-Patodi-Singer index theorem, can be found there or in [14].
Suppose an almost complex structure has been chosen on the hyperplane bundle,
W of the contact manifold and that this structure is positive with respect to the
conformal symplectic structure it inherits. The choice of a contact form therefore
defines an Hermitian metric on W ; an admissible metric on X is one which restricts
to this and gives the contact form length one. The complex structure gives a
reduction of the structure bundle of X (of dimension 2n + 1) to U (n). It therefore
defines a Spin-C structure on X. The Dirac operator associated to this structure
can be taken to act on the exterior bundle of W † , which is the complex bundle
which is the dual to the (0, 1)–part of C ⊗ W. If ∇ is the Levi-Civita connection
projected onto W then the Dirac operator is of the following form in terms of the
decomposition of Λ∗ W † into odd and even parts:

− 1i ∇ξ ∂¯b∗ + ∂¯b
 
ð=
∂b∗ + ∂¯b
¯ 1
i ∇ξ
8 CHARLES EPSTEIN AND RICHARD MELROSE

in the CR case. Similarly in the general case

− i ∇ξ D 0
 1 
ð=
D0 1
i ∇ξ
where D has the same symbol as ∂¯b + ∂¯b in the integrable case and ξ, as above, is
0 ∗

the unit normal to the hyperplane field.

Let Zφ denote the mapping torus defined by a contact diffeomorphism φ of X :
(7) Zφ = X × [0, 1]/(x, 0) ' (φ(x), 1).
Since φ is a contact transformation the lift of the hyperplane bundle W to X × [0, 1]
projects to a subbundle, Wφ , of T Zφ of codimension 2. This has a well defined
conformal symplectic structure. The parameter direction determines a trivial line
bundle, M, in T Zφ as does the orthocomplement to Wφ ⊕ M. Thus T Zφ = Wφ ⊕
R2 . On the R2 –factor we use a “constant” almost complex structure. Using the
conformal symplectic structure we may choose a positive almost complex structure
on Wφ . This induces the canonical Spin-C structure of Zφ and hence defines a Dirac
operator ðφ .

4. Resolution
If S is a generalized Szegő projection for a pseudo-Hermitian structure on a com-
pact contact manifold then, essentially following Boutet de Monvel and Guillemin
[5], we consider the notion of a resolution of S. By this we shall mean a Heisenberg
pseudodifferential operator, B, of order 1 acting on C ∞ (X; Λ∗ W † ) which defines an
acyclic graded complex with respect to form degree, has symbol that of the formal
∂¯b operator, i.e. the annihilation complex, and is such that
(8) Null(B (0) ) = Ran(S), Null(B (j) ) = Ran(B (j−1) ), 1 ≤ j < n,
Ran(B (n−1) ) ⊕ Ran(T ) = C ∞ (X; Λn W † ), (dim X = 2n + 1).
Here T is a generalized Szegő projector for the negative of the contact structure. In
fact the last condition is superfluous; given the exactness conditions the projector
onto the orthocomplement of the range of B (n−1) is automatically a generalized
Szegő projector for the canonical line bundle.
In [5] Boutet de Monvel and Guillemin construct a resolution, which is a classical
pseudodifferential operator, for their generalized Szegő projections in the Hermite
calculus. However, the construction in the appendix to [5] is easily adapted to the
Heisenberg calculus. Clearly B ∈ Ψ1H (X; Λ∗ W † ) is determined, by the requirement
that its symbol be the annihilation complex, up to a term of order 0. In the inte-
grable case we may take B = ∂¯b and then B 2 = 0. In the general case the choice
of symbol ensures only that B 2 ∈ Ψ1H (X; Λ∗ W † ). However, B can be modified by
the addition of lower order terms to give a resolution in the sense of (8).
Proposition 3. Every generalized Szegő projection has a resolution; any two such
resolutions are smoothly isotopic and any smooth family of Szegő projections has a
smooth family of resolutions.
Proof. The existence of such a resolution follows from the methods of the appen-
dix to [5]. See in particular Theorem 5.9 and the remarks following, where it is
noted that the finite-dimensional homology of the resolving complex may be taken
to be trivial. To translate the construction there to the framework of the extended
Heisenberg calculus consider a differential operator B, on the contact manifold
 CONTACT DEGREE 9

which replaces ∂¯b in the CR case. The symbol sequence of this operator is exact,
in fact its symbolic properties are the same as in the CR case. In particular it is
of order 1 in the Heisenberg sense and the Heisenberg symbol of B0 , the action
on functions, is the annihilation operator with null space exactly spanned by s,
the symbol of the Szegő projector. Similarly the symbol acting on maximal degree
forms is the annihilator of the symbol of the ‘dual’ Szegő projector. In interme-
diate form degrees the complex defined by B is elliptic in the Heisenberg sense,
and hence subelliptic. It follows that there are projections in each form degree,
Si ∈ Ψ0H (X; Λ∗ W † ), in the Heisenberg calculus with symbols, si , which are the
orthogonal projections onto the ranges of the symbols in the creation complex.
In fact for a smooth family of structures these projections can be chosen smoothly
in all degrees, with the given family in degree 0. Modifying the original choice of B
to
(9) Bi0 (t) = Si+1 (t)Bi (t)(Id −Si (t))
gives a smooth resolution, in the Heisenberg calculus, up to smoothing errors. Thus
the only remaining step is to pass from a complex modulo smoothing errors to an
actual complex.
This is done, for a single operator (and in the G-equivariant case) in [5]. Thus
we may assume that the smooth family S(t) of generalized Szegő projectors has
a smooth family B 0 (t) of resolutions up to smoothing errors; we may assume that
B 0 (0) is a resolution. If necessary, first replace B00 (t) by B000 (t) = B00 (t)(Id −S(t)),
from which it differs by a smoothing family. Then Ran(S(t)) ⊂ Null(B 00 (t)) has
finite codimension for each t. Since 0 is isolated in the spectrum of (B000 (t))∗ B000 (t)
we can use the argument in the proof of Proposition 1 to construct a smooth fam-
ily of orthogonal projections, S0 (t), differing from S(t) by a finite rank smoothing
family, such that Null(B000 (t)) ⊂ Ran(S0 (t)) for each t. Now consider B0000 (t) =
B000 (t)(I − S0 (t)) then Null(B0000 (t)) = Ran(S0 (t)). Adding a smooth family of fi-
nite rank smoothing operators to B0000 (t) we obtain a smooth family, B0 (t) with
Null(B0 (t)) = Ran(S(t)). Since B0 (t)(B0 (t))∗ , being isospectral to (B0 (t))∗ B0 (t),
has 0 isolated in its spectrum, the orthogonal projections onto the ranges of the
B0 (t) form a smooth family of projections. This allows the resolution to be extended
step by step in terms of form degree, just as in [5].
The existence of a smooth isotopy between any two resolutions of the same
projection follows from this argument and the existence of a symbolic isotopy.
Suppose that S0 and S1 are two generalized Szegő projections with S0 > S1 in
the sense that S0 S1 = S1 S0 = S1 ; set k = ind(S0 , S1 ) ≥ 0. Thus K0 = Ran(S0 )
Ran(S1 ) ⊂ C ∞ (X) is a vector space of dimension k and S0 = S1 + π0 , with π0 the
orthogonal projection onto K0 . Consider how a resolution for S0 may be obtained
from a resolution for S1 . Namely, the resolution, B1 , for S1 may be decomposed
into a resolution, B0 , for S0 and a finite dimensional complex. Set
(0)
B1 = B0 (Id −π0 ) = (Id −π1 )B0
(0)
where π1 is the orthogonal projection onto K1 = B0 K0 , which has dimension
k. If n > 1, choose a subspace K2 ⊂ C ∞ (X; Λ2 W † ) in the complement to the
(1) (1)
null space of B (2) and set B0 = B1 + E1 , where E1 is a smoothing operator of
(0)
rank k which is an isomorphism of K1 onto K2 and annihilates the range of B0 .
Proceeding in this manner one constructs a complex, B0 which is a resolution of S0 .
10 CHARLES EPSTEIN AND RICHARD MELROSE

(n−1)
The generalized Szegő projector, T0 , onto the complement of the range of B0
for this complex is such that if n is odd then T0 > T1 , with Ran(T1 ) ⊂ Ran(T0 ) a
subspace of codimension k and if n is even instead T0 < T1 with Ran(T0 ) ⊂ Ran(T1 )
a subspace of codimension k.
With the resolution B0 constructed from B1 in this way, consider the two oper-
ators
(10) Di = −Si + Bi + Bi∗ + (−1)n Ti , i = 0, 1.
These are self-adjoint operators in the Heisenberg calculus for the contact manifold.
Both act on C ∞ (X; Λ∗ W † ). They are operators of order 1; in fact they differ by a
finite rank smoothing operator. Whilst not elliptic they are Fredholm as operators
on the same domain, namely the anisotropic Sobolev space which is the completion
of C ∞ (X; Λ∗ W † ) with respect to the norm
(11) kuk2 = kuk2L2 + kBuk2L2 + kB ∗ uk2L2 .
This Hilbert space is not compactly embedded in L2 , since on the image of S it is
quasi-isometric to it. Since we shall use it for some time, we shall denote this Hilbert
space H. Thus the linear homotopy between these operators, Dt = (1 − t)D0 + tD1 ,
consists of unbounded self-adjoint Fredholm operators with domain H in L2 , and
hence has discrete spectrum near 0. Furthermore, both D0 and D1 are invertible,
essentially by construction. Thus the spectral flow, at 0, along this family is well
defined.
Lemma 1. The spectral flow of the family Dt on t ∈ [0, 1] at 0 is ind(S0 , S1 )
(assumed positive).
Proof. The spectral flow is that of the finite dimensional family obtained by pro-
jection onto
M
K= Kj ,
j

since this is invariant for the family and outside it the family is constant.
First take n odd. Then, restricted to K the two complexes are
(1)
0 0 B 0
K0 −→ K1 −→ K2 · · · −→ Kn , S0 = Id on K0 , T0 = Id on Kn
(0) (n−1)
B1 0 1 B
K0 −→ K1 −→ K2 · · · −→ Kn , S1 = 0 on K0 , T1 = 0 on Kn .
(j) (j)
Here the B0 are isomorphisms for j odd and the B1 are isomorphisms for j
even. It follows that the eigenvalues of D0 = −S0 + B0 + B0∗ − T0 consist of −1
with multiplicity 2k, k = ind(S0 , S1 ), on K0 ⊕ Kn and on the remaining space of
dimension (n − 1)k the eigenvalues come in pairs with opposite signs. On the other
hand D1 has all (n + 1)k eigenvalues occurring in pairs with opposite signs. Thus
the net flow of eigenvalues across 0 for the curve from t = 0 to t = 1 is k.
For n even, the complexes are again equal off K, on which they take the form
(1) (n−1)
0 0 B 0 B
K0 −→ K1 −→ K2 · · · −→ Kn , S0 = Id on K0 , T0 = 0 on Kn
(0)
B
1 0 0
K0 −→ K1 −→ K2 · · · −→ Kn , S1 = 0 on K0 , T1 = Id on Kn ,
with the same invertibility properties. Thus the eigenvalues of D0 = −S0 + B0 + B0∗
on K consist of −1 with multiplicity k on K0 with the remaining nk eigenvalues
 CONTACT DEGREE 11

occurring in pairs with opposite signs. Similarly D1 = B1 + B1∗ + T1 on K has nk

eigenvalues occurring in pairs with opposite signs, together with 1 of multiplicity k
on Kn . Thus, again the net flow of eigenvalues is k.

Suppose now that S and S 0 are any two generalized Szegő projections, for con-
venience ordered so that ind(S, S 0 ) ≥ 0. Let Dt , t ∈ [−1, 1] be a curve in the
Heisenberg calculus chosen as follows. For t ∈ [−1, 0] let St be an isotopy of gener-
alized Szegő projections with S−1 = S and such that S0 > S1 = S 0 ; that is S0 and
S 0 commute with S0 S 0 = S 0 . Take any resolution B−1 for S = S−1 and deform it
as an isotopy of resolutions, Bt for St , t ∈ [−1, 0]. Let B = B1 be any resolution of
S 0 = S1 . Let B 21 be a resolution of S1 constructed, as above, from the resolution
B0 of S0 . Let Bt , t ∈ [ 21 , 1] be an isotopy of resolutions of S1 . Now, define
(
St + Bt + Bt∗ + (−1)n−1 Tt , t ∈ [−1, 0] ∪ [ 21 , 1]
(12) D̃t =
(1 − 2t)D0 + 2tD 12 , t ∈ (0, 12 ).

Thus, D̃t is a (continuous) curve of self-adjoint Fredholm operators, given by el-

ements of the Heisenberg calculus, it is invertible for t = −1 and t = 1, so the
spectral flow across 0 is well defined. The domains here vary with t, always being
the space Ht , given by (11), for the varying resolution; in fact it only depends on
the Hermitian structure.
Proposition 4. The spectral flow, across 0 of the curve of self-adjoint operators
D̃t is ind(S, S 0 ).
Proof. The family is invertible for t ∈ [−1, 0] and again for t ∈ [ 12 , 1]. Thus the
spectral flow is simply that of the family, linear in t ∈ [0, 12 ], which is ind(S0 , S1 ) =
ind(S, S 0 ) by Lemma 1.

5. Proof of Theorem 2
Now, suppose that S is any generalized Szegő projector on the compact contact
manifold X. Let S 0 = Sφ = (φ∗ )−1 Sφ∗ be the conjugate projector, where φ is
a contact diffeomorphism. The family D̃t considered in Proposition 4 then has
spectral flow c-deg(φ); for simplicity we shall relabel the parameter so that it runs
over [0, 1]. Furthermore, if we choose the resolution of Sφ to be the φ-conjugate of
the resolution for S then D0 , associated to S, and D1 , associated to Sφ are conjugate
operators. The spectral flow on the curve is then well defined independently of the
point at which it is measured in (−1, 1), since in this interval the spectrum remains
discrete. We now further deform this family; we do this in such a manner that the
ends remain conjugate, and hence isospectral, so the spectral flow does not change
provided the family remains self-adjoint and Fredholm.
Choose a classical self-adjoint pseudodifferential operator of order 0, M, on X
which acts on each of the Λj W † and which has symbol that of Id on the positive
contact direction and − Id on the negative. Let Mt be any homotopy, with the
same properties at each point, from M = M0 to M1 = Mφ , its conjugate under
φ. Written out in terms of the odd-even decomposition of Λ∗ W † , the operator D̃t
takes the form
−St0 Bt∗ + Bt
 0
−St + Tt0 Bt∗ + Bt
  
(13) (n odd), (n even).
Bt∗ + Bt −Tt0 Bt∗ + Bt 0
12 CHARLES EPSTEIN AND RICHARD MELROSE

Here, St0 and Tt0 are homotopies between S, Sφ and T, Tφ respectively, which are
projections, up to smoothing operators, associated to the ground state of the har-
monic oscillator of the same Hermitian structure as Bt . The deformed family
−Mt − St0 Bt∗ + Bt −Mt − St0 + Tt0 Bt∗ + Bt
   
(n odd), (n even)
Bt∗ + Bt Mt − Tt0 Bt∗ + Bt Mt
remains Fredholm on Ht and lies in the extended Heisenberg calculus, for  ≥ 0.
Indeed, the symbol is independent of  in the classical region and in the Heisenberg
region, on which Mt acts as the constant given by its symbol, the noncommutative
symbol is invertible provided the diagonal term is invertible on the ground state.
This is true for all  > 0. Setting  = 1, consider the further deformed family
−Mt − δSt0 Bt∗ + Bt −Mt − δSt0 + δTt0 Bt∗ + Bt
   
(n odd), (n even).
Bt∗ + Bt Mt − δTt0 Bt∗ + Bt Mt
The same reasoning shows that this remains Fredholm for δ ∈ [0, 1].
Up to this point the operators have had domains Ht , given by (11). Consider
now a family of self-adjoint elliptic pseudodifferential operators Mt (s) of order s
with Mt (0) = Mt and Mt (1) = 1i ∇tξ . Here ξ is the vector normal to the hyperplane
bundle which is positive on the positive contact direction and ∇t is the connection
for the Hermitian structure associated to Bt . We further demand that M0 (s) and
M1 (s) are always φ-conjugate and that the symbol of Mt (s) is equal to that of
±|σ1 (ξ)|s on the positive and negative contact directions respectively. Clearly these
conditions are consistent, and we may even insist that Mt (s) is graded of degree
0. At δ = 0 the form of the previous family no longer depends on the parity of n.
Starting from this consider the homotopy of operators
−Mt (s) Bt∗ + Bt
 
(14) .
Bt∗ + Bt Mt (s)
Since Mt (s) is a classical pseudodifferential operator of order s this remains an
element of the extended Heisenberg calculus or order (1, 1), until s = 21 . For s ∈
[0, 21 ] it is Fredholm on the Sobolev spaces defined similarly to (11), namely as the
completion of C ∞ (X; Λ∗ W † ) with respect to the norm
(15) kuk2 = kuk2L2 + kMt (s)uk2 + kBt uk2L2 + kBt∗ uk2L2 .
These are closely related to the natural Sobolev spaces associated to elliptic oper-
ators of this double order (1, 1).
Lemma 2. For 0 ≤ s ≤ 21 the Hilbert space obtained from C ∞ (X; Λ∗ W † ) by the
completion with respect to the norm (15), for any partial Hermitian structure and
associated resolution Bt , is the Sobolev space H1,1 (X; Λ∗ W † ) ⊂ L2 (X; Λ∗ W † ) de-
fined by elliptic operators of order (1, 1) plus the range of (1 + ∆)s/2 (S + T ) acting
on L2 (X; Λ∗ W † ) where ∆ is the Laplacian of some (full) metric; for s = 21 this
summand can be dropped.
In the range s ∈ [ 12 , 1] the family is genuinely elliptic as an extended Heisenberg
operator of order (2s, 1) and so is Fredholm on the Sobolev space (independent of
t) associated with operators of this order. Finally, at s = 1, this reduces to the
standard Sobolev space H 1 (X; Λ∗ W † ).
At s = 1, the family (14) has a classical symbol, which is elliptic in the classical
sense, is self-adjoint and the end points are conjugate. The spectral flow of the fam-
ily, already shown to be c-deg(φ), depends only on the principal symbol, provided
 CONTACT DEGREE 13

the end-points are held conjugate. Since the symbol of the family at s = 1 is the
same as that of the family of Dirac operators associated to the partial Hermitian
structure, Theorem 2 is proved.

6. Proof of Theorem 3
Guillemin in [12] considers the push-forward of functions on the coball bundle of
a compact manifold, Y, to its base. Grauert in [11] had shown that the ball bundle
can always be given a complex structure in which the zero section, which can be
identified with the manifold, is totally real; it then looks like a tubular neighborhood
of X in T ∗ X. If the boundary defining function is chosen appropriately, and the
tube Ω is taken to be small enough, then it is shown in [12] that the operator
of integration over the fibers (with respect to an appropriate fiber volume form)
restricted to holomorphic functions
F : u ∈ C ∞ (Ω); ∂u ¯ = 0 −→ C ∞ (Y )


is Fredholm. In fact it is shown in [9] to be an isomorphism. This space of holo-

morphic functions in the tubular domain is canonically isomorphic to the space
of smooth CR functions on the boundary, X = S ∗ Y of Ω. This space is also the
range, on C ∞ (X), of the Szegő projector. Thus F can be replaced by the analo-
gous integral operator G over the spherical fibers of X. Guillemin analyses G by
showing that G ◦ G∗ is an elliptic pseudodifferential operator with positive symbol.
The analytic part of his argument, which relies only on the composition formula
for Fourier integral operators with complex phase, can be applied almost without
change to show that if φ is a contact diffeomorphism of X then G ◦ φ∗ ◦ G∗ is an
elliptic Fourier integral operator, associated with φ and defining the positive trivi-
alization of its Maslov bundle. Since G is Fredholm from the range of S to C ∞ (Y ),
this shows that
ind(Gφ∗ G∗ ) = ind((φ∗ )−1 Sφ∗ S) : Ran(S) −→ Ran(Sφ ) = ind(S, Sφ ).
This proves Theorem 3.

7. The mapping torus

As already noted in the Introduction, Theorem 1 follows from Theorem 2 by a
suspension argument. Closely related results can be found in [2]. For completeness
we include an essentially analytic approach to the proof that if ðφ is the Spin-C
Dirac operator on the mapping torus for the contact diffeomorphism φ, and ð0t is
the homotopy, for t ∈ [0, 1], of Dirac operators for an isotopy of partial Hermitian
structures from one on W to its φ image, then
(16) ind(ðφ ) = SF(ð0t ),
where SF denotes the spectral flow of the family.
Denote the spectral flow by k as before. Let Dθ = −id/dθ be the standard self-
adjoint translation-invariant differential operator on the circle of length 2π. The
family Dθ (t) = Dθ − kt has spectral flow −k for t ∈ [0, 1]. If Lt is the flat complex
line bundle over the circle with global sections the functions on R satisfying the
compatibility condition
u(t0 + 2πr) = eiktr u(t0 ), t0 ∈ R, r ∈ Z
14 CHARLES EPSTEIN AND RICHARD MELROSE

then, naturally, L1 ≡ L0 , and iDθ (t) may be considered as the twisted exterior
differential for Lt , with dθ used to trivialize 1-forms. Thus Dθ (t) becomes a family
of differential operators on the fibers of the bundle L over the torus Sθ × S0t where
the second circle has length one. The spectral flow of this family is −k.
The mapping torus for the contact diffeomorphism φ is a fiber bundle over S0t
with fiber X. By assumption the Dirac operators ð0t , acting on the exterior algebra
of the hyperplane bundle W † with varying almost complex structure, has spectral
flow k. Thus the direct sum, A(t), of these two families has spectral flow 0 as a
family over S0t .
It follows from the results of [17] that the image of the family in K 1 (S0t ) is 0 and
that the combined family has a spectral section. That is, there is a smooth family
of projections, P (t) which are of the form
 
P00 (t) P01 (t)
P (t) =
P10 (t) P11 (t)
where P00 (t) and P11 (t) are pseudodifferential operators on the fibers of the two
factors and the off-diagonal terms are smoothing operators between the two fac-
tors; the crucial property is that there exists a constant R such that if eλ (t) is an
eigenfunction of A(t) with eigenvalue λ then
P (t)eλ = eλ if λ > R,
P (t)eλ = 0 if λ < −R.
Now consider the positive Dirac operator on the mapping torus. This can be
written G(t)(∂t + ð0t ) where G(t) is Clifford multiplication by dt. Thus its index is
the same as that of the operator
∂
+ ð0t : H 1 (Zφ ; Λ∗ W † ) −→ L2 (Zφ ; Λ∗ W † ).
∂t
Consider the direct sum operator ∂t + A(t). In the decomposition
A(t) = P (t)A(t)P (t) + (Id −P (t))A(t)(Id −P (t)) + A0 (t)

the term A0 (t) is a smoothing operator. It is therefore compact on H 1 (Zφ ; Λ∗ W † ),

so can be dropped without changing the index, similarly any constant term can be
dropped and therefore

ind(∂t + A(t)) = ind(∂t + A+ (t) − A− (t)),

A+ (t) = P (t)A(t)P (t) + RP (t),
− A− (t) = (Id −P (t))A(t)(Id −P (t)) + R(Id −P (t)).
Thus, A+ (t), A− (t) > 0 commute. It follows that the index of ∂t + A(t) is zero.
Thus the index of ðφ is − ind(∂t + Dθ (t)). Essentially by construction this latter
index is −k, so (16) is proved and hence so is Theorem 1.

8. Cohomological formula
The Atiyah–Singer theorem gives the following formula for the index, (see for
example [14])
c
(17) ind(ðφ ) = e 2 Â(Zφ )[Zφ ].
 CONTACT DEGREE 15

Here c = c1 (T 1,0 Zφ ) and  is the total –class. As noted in §3 the tangent bundle
splits so
c1 (T 1,0 Zφ ) = c1 (Wφ1,0 ), Â(Zφ ) = Â(Wφ ).

If X is 3–dimensional then Â(Wφ ) = 1 and therefore we find the simpler formula:

1
(18) ind(ðφ ) = c1 (Wφ1,0 ) ∧ c1 (Wφ1,0 )[Zφ ], if dim X = 3.
4
This formula can be made somewhat more explicit by choosing a Chern–Weil
representative for c1 . To do this we choose an Hermitian metric h on the fibers of
Wφ1,0 . We can lift this metric up to X × R and set

Wt1,0 = Wφ1,0 X×{t} , ht = hX×{t} .

 

Clearly h1 = φ∗ (h0 ), conversely given a hermitian metric, g for W 1,0 → X we can

define h so that h0 = g. Let c1 (Ω) be the Chern–Weil representative given by this
choice of metric, this form can be decomposed as
c1 (Ω) = αt + βt ∧ dt
where i∂t αt = i∂t βt = 0. By restriction to the fibers of Zφ over the circle with
parameter t, αt must be the Chern–Weil representative of c1 (Wt1,0 ) with respect to
the metric ht . From type considerations it follows that
Z Z1
(19) c1 (Wφ1,0 ) ∧ c1 (Wφ1,0 )[Zφ ] =2 αt ∧ βt ∧ dt.
X 0

For a smooth family of forms, γt on X × R with i∂t γt = 0 we set

γ̇t = i∂t dγt and dX γt = dγt − dt ∧ γ̇t .
Since c1 (Ω) is closed and dX αt = 0 we deduce that α̇t + dX βt = 0. Then defining
Rt
Bt = 0 βs ds, it follows that dBt = α0 − αt . Using Stokes’ formula and these
relations we find
Z Z1
1,0 1,0
c1 (Wφ ) ∧ c1 (Wφ )[Zφ ] = − 2α1 ∧ B1 [X] − 2 α̇t ∧ βt ∧ dt
X 0
Z Z1
= − 2α1 ∧ B1 [X] + 2 dβt ∧ Bt ∧ dt
X 0
Z Z1
= − 2α1 ∧ B1 [X] + 2 d[βt ∧ Bt ∧ dt] + βt ∧ dBt ∧ dt
X 0
Z Z1
= − 2α1 ∧ B1 [X] + 2 βt ∧ (α0 − αt ) ∧ dt.
X 0

Combining this with (19) we deduce that

1
(20) c1 (Wφ1,0 ) ∧ c1 (Wφ1,0 )[Zφ ] = [α0 − α1 ] ∧ B1 [X].
2
16 CHARLES EPSTEIN AND RICHARD MELROSE

Note that as dB1 = α0 −α1 we have found an expression for c-deg(φ) as a secondary
characteristic class on X.
If W 1,0 is a flat bundle then we may pick h0 to have zero curvature and therefore
α1 = φ∗ (α0 ) = 0. Thus we have:
Corollary 1. If X is a contact 3–manifold such that the contact field is a flat
bundle then c-deg(φ)=0 for all φ ∈ M(X).
As shown in [7] the hypothesis is satisfied if X is the unit circle bundle in a
holomorphic line bundle over a Riemann surface with W 1,0 the induced S 1 –invariant
CR–structure. This includes the unit sphere in C2 with the induced CR–structure
or quotients of the three dimensional Heisenberg group by cocompact lattices. The
latter case was considered by Zelditch [22]. For application to Weinstein’s question
we observe that the contact field defined on the boundary of the Grauert tube of
a surface is always a trivial plane field. Thus we conclude that c-deg(φ) = 0 for
φ a contact transformation of the boundary of the Grauert tube over a surface.
On a 3–manifold a contact class determines an orientation, namely that of θ ∧
dθ. If the surface has genus zero then the two contact structures, the one of the
Grauert tube and the one on the boundary of a strictly pseudoconvex disk bundle
in the canonical bundle, are isotopic. If the genus is greater than 1 then these
two contact structures cannot be diffeomorphically equivalent as they belong to
opposite orientation classes.

References
[1] Richard Beals and Peter Greiner, Calculus on Heisenberg manifolds, Annals of Mathematics
Studies, vol. 119, Princeton University Press, Princeton, NJ, 1988.
[2] B. Booß-Bavnbek and K.P. Wojciechowski, Elliptic boundary problems for Dirac operators,
Birkhäuser, Boston, 1993.
[3] J. Brüning and V.W. Guillemin (Editors), Fourier integral operators, Springer-Verlag, Berlin,
Heidelberg, New York, Tokyo, 1994.
[4] L. Boutet de Monvel, On the index of Toeplitz operators of several complex variables, Invent.
Math. 50 (1979), 249–272.
[5] L. Boutet de Monvel and V. Guillemin, The spectral theory of Toeplitz operators, Ann. of
Math. Studies, vol. 99, Princeton University Press, 1981.
[6] A. Dynin, Pseudodifferential operators on the Heisenberg group, Dokl. Akad. Nauk SSSR 225
(1975), no. 6, 1245–1248.
[7] C.L. Epstein, CR-structures on three dimensional circle bundles, Invent. Math. 109 (1992),
351–403.
[8] , A relative index on the space of embeddable CR-structures, I and II, Ann. of Math
147 (1998), 1–59, 61–91.
[9] C.L Epstein and R.B. Melrose, Shrinking tubes and the ∂-neumann problem, Preprint.
[10] C.L. Epstein, R.B. Melrose, and G. Mendoza, Homology and the Heisenberg algebra, In
preparation.
[11] H. Grauert, On Levi’s problem and the imbedding of real analytic manifolds, Annals of Math.
68 (1958), 460–472.
[12] V.W. Guillemin, Toeplitz operators in n dimensions, Integral Equations and Operator Theory
7 (1984), 145–205.
[13] L. Hörmander, Fourier integral operators, I, Acta Math. 127 (1971), 79–183, See also [3].
[14] H. B. Lawson Jr. and M.-L. Michelsohn, Spin geometry, Princeton Mathematical Series,
vol. 38, Princeton University Press, Princeton, NJ, 1989.
[15] R.B. Melrose, The Atiyah-Patodi-Singer index theorem, A K Peters, Wellesley, Mass, 1993.
[16] , Homology and the Heisenberg algebra, Seminaire: Equation aux dérivées partielles,
Ecole Polytechnique, Mai 1997.
[17] R.B. Melrose and P. Piazza, An index theorem for families of Dirac operators on odd-
dimensional manifolds with boundary, To appear in Jour. Diff. Geom.
 CONTACT DEGREE 17

[18] M.E. Taylor, Noncommutative microlocal analysis. I, Mem. Amer. Math. Soc., vol. 313, AMS,
1984.
[19] A. Weinstein, Some questions about the index of quantized contact transformations, To ap-
pear in RIMS Kokyuroku.
[20] , Fourier integral operators, quantization, and the spectrum of a Riemannian mani-
fold, Géométrie Symplectique et Physique Mathématique, Colloque Internationale de Centre
National de la Recherche Scientifique, no. 237, 1976, pp. 289–298.
[21] S. Zelditch, An index problem on Zoll surfaces, Unpublished manuscript.
[22] , Index and dynamics of quantized contact transformations, Ann. Inst. Fourier 47
(1997), no. 1, 305–363.

Department of Mathematics, University of Pennsylvania

E-mail address: cle@math.upenn.edu

Department of Mathematics, MIT

E-mail address: rbm@math.mit.edu