GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA

SHEAVES

EVA MIRANDA AND FRANCISCO PRESAS

Abstract. In this article we develop tools to compute the Geometric Quanti-

zation of a symplectic manifold with respect to a regular Lagrangian foliation via
sheaf cohomology. The starting point is the definition of representation spaces
due to Kostant. We check that the associated sheaf cohomology apparatus sat-
ifies Mayer-Vietoris and Künneth formulae. As a consequence, new proofs of
classical results for fibrations are obtained. In the general case of Lagrangian fo-
liations, we compute Geometric Quantization with respect to almost any generic
regular Lagrangian foliation on a 2-torus.

Contents
1. Introduction 2
2. Geometric Quantization à la Kostant 4
3. Functorial properties of Geometric Quantization 7
3.1. A Mayer-Vietoris sequence 7
3.2. Mayer-Vietoris in closed domains 8
3.3. A Künneth formula 9
4. Some applications 16
4.1. The case of regular fibrations 16
4.2. Irrational foliation of the 2-torus. 20
5. Geometric Quantization of general foliations over the 2-torus 25
5.1. The topological classification 26
5.2. Geometric Quantization of the torus: The computation 30
References 33

Eva Miranda has been partially supported by the DGICYT/FEDER project MTM2009-07594:
Estructuras Geometricas: Deformaciones, Singularidades y Geometria Integral until December
2012. Her research will be partially supported by the project GEOMETRIA ALGEBRAICA,
SIMPLECTICA, ARITMETICA Y APLICACIONES with reference: MTM2012-38122-C03-01
starting in January 2013. Francisco Presas has has been partially supported by the DGI-
CYT/FEDER project MTM2010-19389. Both authors have been partially supported by ESF
network CAST, Contact and Symplectic Topology.
1
2 EVA MIRANDA AND FRANCISCO PRESAS

1. Introduction
Geometric Quantization attempts to create and understand a “(one-way) dic-
tionary” that “translates” classical systems into quantum systems.
In this way, a quantum system is associated to a classical system in which ob-
servables (smooth functions) become operators of a Hilbert space and the classical
Poisson bracket becomes the commutator of operators. We refer to the classi-
cal references of Kirillov [K94] and Woodhouse [W92] for a good introduction to
geometric quantization.
Grosso modo, following Dirac’s ideas a quantization is a linear mapping of the
Poisson algebra associated to functions on the manifold into the set of operators
on some (pre-)Hilbert space, having some additional properties.
The first step in the Geometric Quantization scheme, is the prequantization
which almost realizes the quantization scheme up to a condition (where the tech-
nical condition of “completeness” of commuting operators is not required).
The guinea pig for any model of Geometric Quantization is the cotangent bundle
of a manifold endowed with Liouville form. The model for prequantization of a
cotangent bundle was constructed by Segal [S60] in the 60’s. In trying to extend
this construction to general symplectic manifolds the integrality of the cohomology
class of the symplectic form is required.
Unlike the quantization by deformation approach in which the classical systems
can be seen as a limit of the deformation, in the geometric quantization approach,
the process only has partial memory of the information of the initial system. This
is why the choice of additional geometric structures (polarizations) plays an im-
portant role. The classical motivation for considering this polarization comes from
the distinction between “momentum” and “position” which is native from mathe-
matical formulation of mechanical systems in the cotangent bundle of a manifold.
A desired property is that the representation space does not depend on the
polarization and this property is satisfied in the case of considering Kähler polar-
izations.
Our point of view in this big endeavour is a modest one. We plan to construct
a “representation space” in the case of real polarizations.
The initial idea of quantization is to consider as initial representation space the
vector space of flat sections of a line bundle associated to the symplectic manifold
in the directions of the polarization. These flat sections exist just locally for a
generic leaf of a real polarization, making the quantization vector space usually
trivial. This is why Kostant’s approach to quantization [K75] proposes to replace
the set of flat sections along a polarization by cohomology groups with coefficients
in this sheaf as representation spaces. In this paper we will not discuss either the
(pre)Hilbert structure of this space nor the quantization rules.
An interesting related issue is the notion of Bohr-Sommerfeld leaves and its con-
tribution to Geometric Quantization. A leaf of the polarization is Bohr-Sommerfeld
if there are non-trivial flat sections of the bundle globally defined along it. The
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 3

characterization of Bohr-Sommerfeld leaves for real polarizations given by regular

fibrations is a well-known result by Guillemin and Sternberg ([GS83]). In par-
ticular, the set of Bohr-Sommerfeld leaves is discrete and is given by “action”
coordinates that are well-defined in a neighbourhood of a compact leaf.
Theorem 1.1 (Guillemin-Sternberg). If the polarization is a regular fibration with
compact leaves over a simply connected base B, then the Bohr-Sommerfeld set is
discrete and assuming that the zero-fiber is a Bohr-Sommerfeld leaf, the Bohr-
Sommerfeld set is given by BS = {p ∈ M, (f1 (p), . . . , fn (p)) ∈ Zn } where f1 , . . . , fn
are global action coordinates on B.
This result connects with Arnold-Liouville-Mineur theorem [A89], [D80] for
action-angle coordinates of integrable systems to Geometric Quantization. When
we consider toric manifolds, the base B may be identified with the image of the
moment map by the toric action (Delzant polytope).
In this case quantization is given by precisely the following theorem of Śniatycki
[Sni75]:
Theorem 1.2 (Śniatycki). If the leaf space B n is a Hausdorff manifold and the
natural projection π : M 2n → B n is a fibration with compact fibres, then all the
cohomology groups vanish except for degree half of the dimension of the manifold.
Furthermore, Q(M 2n ) = H n (M 2n , J ), and the dimension of H n (M 2n , J ) is the
number of Bohr-Sommerfeld leaves.
In this paper we will present a Geometric Quantization computation kit to deal
with this sheaf cohomology using a fine resolution approach. We will find applica-
tions to compute Geometric Quantization of general Lagrangian foliations which
are not in general a fibration. In particular, we will completely determine Geomet-
ric Quantization in the case of line fields on a torus via the classification of these
foliations up to topological conjugacy due to Denjoy and Kneser [D32, K24].
A different approach to compute Geometric Quantization is the one of Čech
cohomology. This approach was used in [H11] and [HM10] and turns out out to
be useful when we consider integrable systems with singularities. We will apply
our Geometric Quantization computation kit to reprove some results which were
obtained via the Čech approach like the theorem of Śniaticky [Sni75].

Organization of this paper:

In Section 2 we present the Kostant complex in this approach to Geometric

Quantization. Even if this paper focuses on the Geometric Quantization case,
when it comes to the scene we will also mention the corresponding results for
foliated cohomology, considered as a limit case with ω → 0. In Section 3 we prove
Mayer-Vietoris and Künneth formula for this complex. These results connect
directly with the work of Bertelson [B11] to prove Künneth formula for foliated
cohomology.
4 EVA MIRANDA AND FRANCISCO PRESAS

In Section 4 we present two simple applications of these functorial properties of

this complex to compute Geometric Quantization of a regular fibration and for the
irrational flow on the torus. In particular, we recover former results of Sniaticky
[Sni75] for Geometric Quantization and of El Kacimi-Alaoui for foliated cohomol-
ogy [K83]. In the last section, we use the classification of foliations on the 2-torus
up to topological equivalence due to Denjoy and Kneser [D32], [K24] together with
the Mayer-Vietoris argument to compute the Geometric Quantization associated
to any generic foliation of the 2-torus.

Acknowledgements:

We are thankful to Yael Karshon for raising the question that has led us to study
this problem. We are grateful to Mélanie Bertelson for sending us a copy of her
thesis that brought our attention to the study of a Künneth formula for foliated
cohomology and the reference of Grothendieck [G54] on a completed Künneth
formula. Last but not least, thanks to Romero Solha for reading a draft version
of this paper.

2. Geometric Quantization à la Kostant

Let (M 2n , ω) be a closed symplectic manifold of integer class and let L be an
associated prequantizable line bundle, i.e. an Hermitian complex line bundle1
equipped with a compatible connection whose curvature equals −iω.
A real polarization P over (M 2n , ω) is a foliation whose leaves are Lagrangian
with respect to ω. Given a hermitian vector bundle over (M 2n , ω) endowed with
a connection ∇, we define a section s : M → L to be flat along the polarization P
if the operator
∇P s : P ,→ T M → L
v → ∇v s
is the null operator. In other words, if the covariant derivative of the section
vanishes along the foliation directions.
One could consider the set of flat sections of this bundle to start constructing
a representation space for Geometric Quantization. In order to induce a metric
on the space of sections which can endow this space with a (pre)Hilbert structure,
we consider another line bundle, the so-called metaplectic correction. This line
bundle also has a physical meaning since it captures some Bohr-Sommerfeld leaves
in the Kähler case which are not captured otherwise [KMN10].

1Givena symplectic form ω ∈ H 2 (M, R) with integer class [ω], the lift to H 2 (M, Z) is not
unique in general, we have a exact sequence T or(H 2 (M, Z)) −→ H 2 (M, Z) −→ H 2 (M, R), for
manifolds with non-vanishing T or(H 2 (M, Z)) this line bundle is not unique.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 5

Some authors (including the first author of this paper) who have been mainly
interested in the representation space itself have neglected this approach to quan-
tization using the metaplectic correction (see for instance [H11] [HM10]). In this
paper we will consider the metaplectic correction (which can be seen as a correction
in the line bundle) even if we do not consider in the current paper the endeavour of
completing the representation space to endow it with a Hilbert space. This effort
can be seen as a first step in this direction.
WeV say that P admits a metalinear correction if the complex determinant bundle
N = nC (P ⊗R C) admits a square root N 1/2 . This is equivalent to the evenness
of the first Chern class of N , c1 (N ). This can be easily proven using the isomor-
phism between Picard group of isomorphism classes of complex line bundles and
H 2 (M, Z). The group structure on the Picard group is an abelian group with group
operation the tensor product and therefore c1 (L ⊗ L0 ) = c1 (L) + c1 (L0 ). Thus P
admits a metaplectic correction if and only if there exists an element e ∈ H 2 (M, Z)
such that 2e = c1 (N ).
Let us prove that N admits a natural flat connection. This will be needed to
guarantee the existence of solutions of the associated equation of flat sections.
It was Weinstein [W71] who defined a natural flat connection along a Lagrangian
foliation using symplectic duality. This will be needed to define a canonical flat
connection on N .
Let us recall here Weinstein’s construction: The classical Bott connection ([B72])
can be defined as a partial connection associated to a foliation P. This allows to
define the covariant derivative of vector fields which belong to the normal bundle
of the foliation with respect to vector fields of the distribution.
Given X a vector field of the polarization, we denote by Y a vector field in the
normal bundle N . Let us denote by p : T M −→ N the projection on the normal
bundle. The Bott connection is defined as ∇B X (Y ) = p([X, Y ]) where Y is any
vector field on T M which projects to Y .
This Bott connection is flat. We now use the fact that symplectic duality estab-
lishes a natural isomorphism between the tangent and cotangent bundles of M to
induce a flat connection on P (given a Lagrangian leaf of the foliation the normal
bundle to the leaf can be easily identified with the cotangent bundle of the leaf).
In this way the bundle N acquires a canonical flat connection and so it does the
bundle N 1/2 . Let us call this connection 4B
The initial connection on the flat bundle ∇ together with the new connection
4B can be used to define a connection ∇0 on the bundle L⊗N 1/2 . We now consider
the equation,
∇0v s = 0, v ∈ P,
where s is a section of the bundle L⊗N 1/2 . The flatness of 4B and the Lagrangian
character of the polarization guarantee that this equation has a local solution.
As observed by Śniaticky in [Sni75], if the leaves have non-trivial fundamental
group then there exist no global flat section along the leaves of the polarization.
6 EVA MIRANDA AND FRANCISCO PRESAS

Thus, it makes sense to consider the sheaf of flat sections and consider as a first
candidate for Geometric Quantization the cohomology groups associated to this
sheaf. Let us formalize this idea: We denote by S be the sheaf of sections of the
line bundle L ⊗ N 1/2 ; and by J the sheaf of flat sections of the bundle L ⊗ N 1/2
along the real polarization P. Denote by H i (M 2n , J ) the i-th sheaf cohomology
group associated to the sheaf J . We define Geometric Quantization using these
cohomology groups,
Definition 2.1. The Geometric Quantization space of (M, ω) with respect to P is
defined as,
M
Q(M 2n , P) = H i (M 2n , J ).
i∈N

Remark 2.2. This cohomology with coefficients in the sheaf of flat sections admits
computations à la Čech and thus, a priori, H i (M 2n , J ) = 0 for i > 2n. Indeed,
as we will see now, this sheaf admits a fine resolution and we may even apply a
computation twisting the De Rham complex that computes foliated cohomology and
this yields H i (M 2n , J ) = 0 for i > n.

The Kostant complex

Vi
Let ΩiP denote the associated sheaves of sections of the vector bundles P, i.e.
i
^
ΩiP (U ) = Γ(U, P).

Let C be the sheaf of complex-valued functions that are locally constant along P.
Then, we consider the natural (fine) resolution
i d d d d
0 → C → Ω0P →
P
Ω1P →
P
Ω1P →
P
Ω2P →
P
··· ,
The differential operator dP is the restriction of the exterior differential along the
directions of the distribution. This is the standard resolution used to compute the
foliated (or tangential) cohomology of the foliation P (see for instance [K83]).
We can now use this fine resolution of C to construct a fine resolution of the
sheaf J . We just need to “twist” the previous resolution with the sheaf J . It
produces the following exact sequence,
i ∇0 ∇0
0 → J → S →P S ⊗ Ω1P →P S ⊗ Ω2P → · · ·
This is called the Kostant complex. The cohomology of this complex computes
exactly H i (M, P) and therefore computes Geometric Quantization.
In the next sections, we will develop some tools and techniques to compute the
cohomology of this complex.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 7

3. Functorial properties of Geometric Quantization

In this Section we provide some algebraic tools to compute the Geometric Quan-
tization associated to a real polarization.

3.1. A Mayer-Vietoris sequence. The classical Mayer-Vietoris theorem, proved

for singular cohomology, also works for the cohomology of any sheaf over quite
general topological spaces (see [Br97], pg. 94). For the sake of completeness we
are going to provide a proof for our sheaf J .
Take U and V a pair of open sets covering M . Then there is a sequence of
inclusions

M ← U t V ⇔ U ∩ V.

This induces a sequence as follows

r0
r / /
S ⊗ Ω∗P (M ) (S ⊗ Ω∗P (U )) ⊕ (S ⊗ Ω∗P (V )) / S ⊗ Ω∗P (U ∩ V )
r1

given by the restriction map r, r0 and r1 . Substracting the two morphisms on the
right side we get the following sequence,
(1)
r0 −r1
0 / S ⊗ Ω∗P (M )
r / (S ⊗ Ω∗P (U )) ⊕ (S ⊗ Ω∗P (V )) / S ⊗ Ω∗P (U ∩ V ) / 0

We have the following,

Theorem 3.1. The sequence (1) is exact.

Proof. The injectivity of the morphism r is obvious, and so it is the exactness at

(S ⊗ Ω∗P (U )) ⊕ (S ⊗ Ω∗P (V )). To check the surjectivity of r0 − r1 we take a partition
of the unity relative to the covering U, V given by functions χU and χV . Now for
any α ∈ S ⊗ Ω∗P (U ∩ V ), we define

(2) βU = χV · α ∈ S ⊗ Ω∗P (U ),
(3) βV = −χU · α ∈ S ⊗ Ω∗P (V ).

We immediately obtain that r0 (βU ) − r1 (βV ) = α. 

So after playing this game at all degrees, we obtain the following commutative
exact diagram,
8 EVA MIRANDA AND FRANCISCO PRESAS

0 0 0

  
r0 −r1
0 / S⊗ Ω0P (M )
r / (S ⊗ Ω0P (U )) ⊕ (S ⊗ Ω0P (V )) / S⊗ Ω0P (U ∩ V )0 / 0
∇P ∇P ∇P
  
r0 −r1
0 / S ⊗ Ω1P (M )
r / (S ⊗ Ω1P (U )) ⊕ (S ⊗ Ω1P (V )) / S ⊗ Ω1P (U ∩ V )0 / 0
∇P ∇P ∇P
  
.. .. ..
. . .

Observe that vertically we have the Kostant complex that computes cohomology
with coefficients in a sheaf. We apply the snake’s Lemma [A69] to prove the
following,
Corollary 3.2. The following sequence is exact

(4) 0 / H 0 (M, J ) / H 0 (U, J ) ⊕ H 0 (V, J ) / H 0 (U ∩ V, J )

eee eeeeee
e
eeeeee
eeeeeeeeeeeδ
e
reeeeee 1
H 1 (M, J ) / H (U, J ) ⊕ H 1 (V, J ) / H 1 (U ∩ V, J )
eeeeee
eeeeee
eeeeeeeeeee
eeeee
reeeeee
δ
H 2 (M, J )

We recall how to define the connecting operator δ which we need to make explicit
for future computations. We take a class a ∈ H i (U ∩ V, J ), represented by a form
α ∈ S ⊗ ΩiP (U ∩ V ). From the previous diagram we know that there are forms
(βU , βV ) ∈ (S ⊗ ΩiP (U )) ⊕ (S ⊗ ΩiP (V )), defined via the formulae (2) and (3).
Then we take the image under the vertical morphism to obtain (∇P βU , ∇P βV ) ∈
S ⊗ Ωi+1 i+1 i
P (U ) ⊕ S ⊗ ΩP (V ). Since a ∈ H (U ∩ V, J ), commutativity of the diagram
yields (r0 − r1 )(∇P βU , ∇P βV ) = 0. Hence there is a form γ ∈ S ⊗ Ωi+1
P (M ) such
that r(γ) = (∇P βU , ∇P βV ). Again by the commutativity of the diagram yields
∇P (γ) = 0 and thus a closed form defining a class in H i+1 (M, J ) by construction
δ([a]) = [γ] and this is how the connecting operator is defined. It is easy to check
that this definition does not depend on the form representing the class [a].

3.2. Mayer-Vietoris in closed domains. As usual the space of sections of the

sheaf J can be defined for closed sets C ⊂ M . Take the directed system of sections
J (Ai ) of open sets Ai containing the closed set C. It is directed by the restriction
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 9

morphism. Then, we define the space of sections over C as

J (C) = lim Ai ,
−→
the direct limit of the system. The proof of the previous Mayer–Vietoris result
can be easily adapted for the case of closed sets (see [BT82] for all the details). It
is stated as follows. Take C0 and C1 a pair of closed sets covering M . Then there
is, as in the previous Subsection, a sequence of inclusions
M ← C0 t C1 ⇔ C0 ∩ C1 .
Again, we obtain
r0
r / /
S ⊗ Ω∗P (M ) (S ⊗ Ω∗P (C0 )) ⊕ (S ⊗ Ω∗P (C1 )) / S ⊗ Ω∗P (C0 ∩ C1 )
r1

given by the restriction map r, r0 and r1 . Substracting the two morphisms on the
right side we get the following sequence,
(5)
r0 −r1
0 / S ⊗ Ω∗P (M )
r / (S ⊗ Ω∗P (C0 )) ⊕ (S ⊗ Ω∗P (C1 )) / S ⊗ Ω∗P (C0 ∩ C1 ) / 0
We have the following,
Theorem 3.3. The sequence (5) is exact and thus it induces a long exact sequence
in cohomology.
3.3. A Künneth formula. The classical Künneth formula also holds with great
generality for the cohomology of a sheaf ([Br97]). It works for the Geometric
Quantization scheme in a generalized form. Let (M1 , P1 ) and (M2 , P2 ) be a pair of
symplectic manifolds endowed with Lagrangian foliations. The natural cartesian
product for the foliations is Lagrangian with respect to the product symplectic
structure. The induced sheaf of flat sections associated to the product foliation
will be denoted J12 . Note that we use the pre quantum-bundle and metaplectic
corrections defined as pull-backs and tensor products of the ones defined over M1
and M2
There is a natural morphism
(6) Ψ : H ∗ (M1 , J1 ) ⊗ H ∗ (M2 , J2 ) → H ∗ (M1 × M2 , J12 )
induced by pull-back of the classes through the natural projections. Furthermore
we can prove the following Künneth formulae,
Theorem 3.4 (Künneth formula for Geometric Quantization). There is an iso-
morphism
M
H n (M1 × M2 , J12 ) = H p (M1 , J1 ) ⊗ H q (M2 , J2 ),
p+q=n

whenever the Geometric Quantization associated to (M1 , J1 ) has finite dimension,

M1 is compact and M2 admits a good covering.
10 EVA MIRANDA AND FRANCISCO PRESAS

The isomorphism is defined by the inverse of the map Ψ above and the core of
the proof will be to show its existence.
Remark 3.5. In this paper the condition of good covering will be the one as stated
in [Br97]. This condition is automatically fulfilled in particular if M2 has finite
topology or it is compact. The compactness condition on M1 can indeed be relaxed
to the following one: M1 is a submanifold of a compact manifold. What we will
do is to use Künneth formula for closed balls and use compactness.
Remark 3.6. In [Br97] a more general formula is given where a torsion of a com-
plex is present. Whenever the complex is torsionless, we obtain a clean Künneth
formula. One way to look at this theorem as a torsionless case of a generalized
Künneth formula à la Grothendieck [G54].
Many authors have studied more general conditions under which a Künneth
formula holds for sheaf cohomology with infinite dimension. See for instance the
works of Kaup [K67] and Grothendieck [G54] where a Künneth formula is given in
terms of completion of the tensor product:
M
H n (M1 × M2 , J12 ) = H p (M1 , J1 )⊗H
ˆ q (M2 , J2 ).
p+q=n
This approach using nuclear spaces has also been adopted by Bertelson in [B11]
for foliated cohomology. In particular, Bertelson obtains a similar result to our
Künneth formula for foliated cohomology under similar hypothesis (imposing al-
ready compactness of one of the factor manifolds). This foliated cohomology result
can be reinterpreted as the “zero limit”case of Geometric Quantization.
The proof follows very closely the de Rham cohomology case. We start by
proving the following proposition which automatically implies the Künneth formula
in case one of the factors is an open set with the property that the leaves restricted
to the domain U are contractible and the leaf space is also contractible. We will
call these open sets cotangent balls. Its compactification, whenever the leaves are
still contractible, will be called a closed cotangent ball.2
Proposition 3.7. Let U be a closed cotangent ball and M a compact manifold
with finite dimensional Geometric Quantization. The following equality holds,
H n (M × U, J10 ) = H n (M, J ) ⊗ H 0 (U, J0 ),
Before proving this proposition, we need the following lemma,
Lemma 3.8. Given an element α ∈ S ⊗ΩpP (M ×U ) which is closed by the Kostant
differential ∇P , we can always find another element α0 in the same cohomology
class in the Kostant complex such that α0 ∈ S(M × U ) ⊗ ΩpP (M ).
2Noticethat for this we need to extend the definitions of Geometric Quantization to that of
manifolds with boundary. For classical references see [G93] and [TZ99]. However, it follows the
definitions provided in Subsection 3.2.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 11

Proof. Assume that we are in the case M × U , for a cotangent ball U ⊂ R2 . The
proof can be adapted for higher dimensions of U just by sophisticating the notation.
Denote PM the polarization in M and (LM , ∇M ) the quantization bundle over M .
In the same way, the cotangent ball has as quantization bundle the restriction of
the quantization bundle on R2 which we denote by LU with connection ∇U . The
product connection will be denoted by ∇ = ∇M
∇U .
Let us take a system of coordinates over U in which we can trivialize the pre-
quantization bundle in the horizontal directions starting with a parallel section
along the vertical axis3. With respect to this trivialization the connection on the
quantization bundle LU is written,
∇U = d − iφ(x, y)dy.
Now, take an element α ∈ S ⊗ ΩpP (M × U ). It can be written as,
(7) α = α̂ + β ∧ dy,
where α̂ ∈ S ⊗ ΩpP (M )
and β ∈ S ⊗ Ωp−1
P (M ).
Observe that α is closed in the Kostant complex and therefore ∇P α = 0. Recall
that U is a cotangent ball, therefore
U = {(x, y) ∈ R2 : −1 ≤ x ≤ 2 , h1 (x) ≤ y ≤ h2 (x)}.
h1 (x)+h2 (x)
Take h3 (x) = 2
. Define γ ∈ S ⊗ Ωp−1
P (M ) satisfying the differential
equation
∂γ
− iφ(x, y)γ = β,
∂y
with the initial conditions γ(p, x, h3 (x)) = 0. This is a first order linear ordinary
differential equation with “parameter” p ∈ M and fixed initial conditions. Thus
it has a unique solution.
It is simple to check that the following equality holds ∇γ = ∇M γ + β ∧ dy.
Thus, replacing this expression in (7), we obtain
α = (α̂ − ∇M γ) + ∇γ = α0 + ∇γ,
we conclude by observing that α0 = α̂ − ∇M γ ∈ S(M × U ) ⊗ ΩpP (M ) and ∇γ is
obviously exact. 
We now recall a few facts about orthogonal series of functions in interpolation
theory. Let f : S 1 → C a continuous function over the circle. Denote by fˆ : Z → C
its Fourier coefficients. We have the following relation between the decay of the
coefficients and the smoothness of the function.
Theorem 3.9 (Theorem 3.2.9 and Proposition 3.2.12 in [Gr09]). For any s ∈ Z+ ,
the following two statements hold:
3This has been known in the literature of Geometric Quantization as the existence of a “triv-
ializing section” for neighbourhoods in the case of regular foliations.
12 EVA MIRANDA AND FRANCISCO PRESAS

h2

h3

h1

|fˆ(k)|
(1) If f is a function of C s -class then we have limk→±∞ 1+|k|s
= 0,
|fˆ(k)|
(2) If limk→±∞ 1+|k| s+1 = 0 then we have f is a function of C s -class.

Proof of Proposition 3.7.

The parametrized Poincaré Lemma immediately implies that for the cotangent
balls we have H i (U, J0 ) = 0 for i > 0 and H 0 (U, J0 ) = C ∞ (Rn , C). Assume that
we are in the product case M × U , for a cotangent ball U ⊂ R2 . The proof can be
adapted for higher dimensions of U just by sophisticating the notation.
Recall the natural morphism (12), in particular, produces a map

∆ : H n (M, J ) ⊗ H 0 (U, J0 ) → H n (M × U, J10 ).

Let us check that this map is injective. Consider a pair of elements (a, f ) with
a ∈ S(M × U ) ⊗ ΩnP (M ) satisfying that ∇M a = 0 and f ∈ Ω0P (U ) = C ∞ [0, 1] such
that there exists an element b ∈ ΩPn−1 (M × U ) satisfying that a · f = ∇b (the pair
(a, f ) goes to the zero class). Since f 6= 0, there is a number x0 ∈ [0, 1] such that
f (x0 ) 6= 0. Recall that U is a cotangent ball, therefore

U = {(x, y) ∈ R2 : −1 ≤ x ≤ 2 , h1 (x) ≤ y ≤ h2 (x)}.

h1 (x)+h2 (x)
Take h3 (x) = 2
and consider the inclusion map

i0 : M → M × U
p → (p, x0 , h3 (x0 )).

Define the element b̃ = i∗0 b. Since the covariant derivative commutes with the
restriction, We obtain ∇M b̃ = i∗0 ∇b. This yields ∇M ( f (x1 0 ) b̃) = a and hence a is
exact and [a] = 0. This proves the injectivity.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 13

We will now check the surjectivity of this mapping 4. Select an element A ∈

H n (M × U, J10 ). By Lemma 3.8, there is a representative α of the class A such
that α ∈ S(M × U ) ⊗ ΩnP (M ). Recall that ∇P α = 0 implies that, for coordinates
(x, y) ∈ U ⊂ R2 , we have
(8) ∇ ∂ α = 0.
∂y

Consider the inclusion,

ih : M × [−1, 1] → M × U
(p, x) → (p, x, h3 (x))
and the pull-back form obtained via ih , α̃ = i∗h α. Recall that, by equation (8), the
form α̃ completely determines α. The form α̃ can be understood as a 1-parametric
family α̃t (p) = α̃(p, t), for t ∈ [−1, 1], with α̃t a closed element of ΩnP (M ). Thus,
there is a family α̃t0 , for t ∈ [−2, 2] satisfying:
(1) α̃t0 = α̃t , for t ∈ [−1, 1],
(2) α̃t0 = 0 for |t| ≥ 3/2,
(3) ∇M α̃t0 = 0.
This defines an element α̃0 ∈ S(M ×[−2, 2])⊗ΩnP (M ) and by applying the differen-
0 n
T form α ∈ S(M ×V )⊗ΩP (M ),
tial equation (8), we can extend it to a unique closed
where the closed domain V ⊃ U satisfies that V (R × {0}) = [−2, 2].
Fix a point p ∈ M . Define the inclusion map ep : [−2, 2] → M × [−2, 2]. Now
we obtain the map
α̃p = e∗p α̃ : [−2, 2] → S(M × [−2, 2])p ⊗ ΩnP (M )p ' S(M )p ⊗ ΩnP (M )
t → α̃(p, t).
Recall that α̃p extends to a smooth section α̃p : S 1 = R/(4Z) → S(M ×
[−2, 2])p ⊗ ΩnP (M )p . The Fourier coefficients of α̃p are computed as,
Z 2
(9) α̃p (m) = α̃p (t) · eiπmt/2 dt,
−2

that are obviously smooth in p ∈ M . This equation defines elements α̃(m) ∈

S(M )p ⊗ ΩnP (M ) that are closed since the covariant derivative commutes with the
integration in formula (9).
Since the space M is compact and the variation of parameters is continuous,
Theorem 3.9 yields the following inequality for any s > 0,
|α̃p (m)| ≤ C · |m|s ,

4In the case of foliated cohomology surjectivity of this map was not well sorted in works prior
to the work of Bertelson [P85]. In our opinion a complete proof of this fact for the limit case of
foliated cohomology is only achieved in [B11].
14 EVA MIRANDA AND FRANCISCO PRESAS

where C > 0 is a constant that does not depend on p ∈ M ( but which may depend
on s). So, it can be rewritten as,
(10) |α̃(m)|C 0 ≤ C · |m|s ,
Recall that
S(M ) ⊗ ΩnP (M )
H n (M, J1 ) = .
Im∇(S(M ) ⊗ Ωn−1
P (M ))

The topology of H n (M, J1 ) is defined as the vector space quotient topology, in-
duced out of the C 0 -norm in ΩnP (M ). Therefore, by using the equation (10) and
the decrease of the norm in the projection we obtain
(11) |[α̃(m)]| ≤ |α̃(m)|C 0 ≤ C · |ms |,
where [α̃(m)] is the class represented by the element α̃(m) ∈ S(M ) ⊗ ΩnP (M ) in
H n (M, J1 ).
Now, we can check that H n (M, J ) ⊗ H 0 (V, J0 ) = C ∞ ([−2, 2], H n (M, J )), i.e.
the smooth maps from the interval to the finite vector space H n (M, J ) (we are
using the differential equation (8) to uniquely extend the section from M × [−2, 2]
to M × V ). Once this identification is done, we can prove surjectivity of ∆ in the
following way: Define the formal series,
α̂ = Σm∈Z [α̃(m)]eiπmt/2 .
we can use again Theorem 3.9 in combination with the inequality (11) to conclude
that α̂ is smooth. i.e. α̂ ∈ C ∞ ([−2, 2], H n (M, J )). Denote its restriction by
α̂0 ∈ C ∞ ([−1, 1], H n (M, J )). The following equality holds ∆α = α̂0 and this
proves surjectivity of ∆.
Note that we have worked out the proof for 2−dimensional cotangent balls, for
the case of cotangent balls U having dimension 2k, we can still apply Poincaré
Lemma to make sure that the y-directions in the ball are constant. In this case we
need to work with Fourier coefficients in the space of maps C ∞ (Tk , H n (M, J )). All
the arguments go through; the key point in the proof being the finite dimensionality
of the space H n (M, J1 ). 
We now proceed with the proof of Künneth for the general case.

Proof. of Theorem 3.4

Denote by P0 the vertical foliation in R2m , with coordinates (x1 , . . . xm , y1 , . . . , ym ),
whose leaves are defined by the equations
xi = ci , ci ∈ R, i = 1, . . . , m.
With respect to the standard symplectic structure the foliation P0 is standard.
Choose a primitive 1-form as the connection form for the quantization line bundle
L ⊗ N 1/2 over R2n and denote by J0 the associated sheaf of flat sections.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 15

The rest of the proof is standard and follows step by step pg.49 in [BT82]. We
outline the main ideas. There is a natural morphism
(12) Ψ : H ∗ (M1 , J1 ) ⊗ H ∗ (M2 , J2 ) → H ∗ (M1 × M2 , J12 )
given by the pull-back of forms in each component. Now we take U and V open
sets of M1 and we tensor with the fixed vector space H n−p (M2 , J2 ) the associated
Mayer-Vietoris sequence to obtain
· · · → H p (U ∪ V, J1 ) ⊗ H n−p (M2 , J2 )
→ (H p (U, J1 ) ⊗ H n−p (M2 , J2 )) ⊕ (H p (V, J1 ) ⊗ H n−p (M2 , J2 ))
→ H p (U ∩ V, J1 ) ⊗ H n−p (M2 , J2 ) → · · · .
Summing up for p = 0, . . . , n, we obtain the exact sequence
M n
··· → H p (U ∪ V, J1 ) ⊗ H n−p (M2 , J2 )
p=0
n
M
→ (H p (U, J1 ) ⊗ H n−p (M2 , J2 )) ⊕ (H p (V, J1 ) ⊗ H n−p (M2 , J2 ))
p=0
Mn
→ H p (U ∩ V, J1 ) ⊗ H n−p (M2 , J2 ) → · · ·
p=0

to which we apply the morphisms (12) to obtain the following commutative exact
diagram
.. ..
. .
↓ ↓
Ln Ψ
p=0 H p (U ∪ V, J1 ) ⊗ H n−p (M2 , J2 ) → H n ((U ∪ V ) × M2 , J12 )
↓ ↓
Ln p (U, J Ψ
p=0 (H 1) ⊗ H n−p (M2 , J2 )) ⊕ (H p (V, J1 ) ⊗ H n−p (M2 , J2 )) → H n (U × M2 , J12 ) ⊕ H n (V × M2 , J12 )
↓ ↓
Ln Ψ
p=0 H p (U ∩ V, J1 ) ⊗ H n−p (M2 , J2 ) → H n ((U ∩ V ) × M2 , J12 )
↓ ↓
Ln+1 Ψ
p=0 H p (U ∪ V, J1 ) ⊗ H n+1−p (M2 , J2 ) → H n+1 ((U ∪ V ) × M2 , J12 )
↓ ↓
.. ..
. .

The commutativity is obvious except for the block

Ln p (U Ψ
p=0 H ∩ V, J1 ) ⊗ H n−p (M2 , J2 ) → H n ((U ∩ V ) × M2 , J12 )
↓δ ↓δ
Ln+1 Ψ
p=0 H p (U ∪ V, J1 ) ⊗ H n+1−p (M2 , J2 ) → H n+1 ((U ∪ V ) × M2 , J12 )
that we check as follows. Let α ⊗ β ∈ H p (U ∩ V, J1 ) ⊗ H n−p (M2 , J2 ). Denote the
projections to the factors of the cartesian product M1 ×M2 as π1 and π2 . Therefore
Ψ ◦ δ(α ⊗ β) = π1∗ (δα) ∧ π2∗ (β)
16 EVA MIRANDA AND FRANCISCO PRESAS

and
δ ◦ Ψ(α ⊗ β) = δ(π1∗ (α) ∧ π2∗ (β)).
We have to check that they are equal and a simple computation shows that it is
true, just recalling that the form β is closed.5
We now conclude the proof of Künneth formula in the same way Bott and Tu
do it for the De Rham cohomology case (page 50 of [BT82]).
Observe that because of lemma 3.7, Künneth formula holds for U and V from
the previous argument, Künneth formula also holds for U ∩ V , thus from the Five
Lemma it also holds for U ∪ V . The Künneth formula now follows by induction
on the cardinality of a good cover (or a finite covering in the compact case). This
concludes the proof of the theorem. 

4. Some applications
In this section we apply Mayer-Vietoris and Künneth formula to compute Geo-
metric Quantization of regular fibrations and the irrational flow on the torus.
4.1. The case of regular fibrations. As an application of the previous formal-
ism, we now give a simple proof of the quantization of a real polarization given by
a regular fibration. This is exactly the case studied by Śniaticky in [Sni75]. Let us
point out that a different proof was obtained by Hamilton in [H11] where a Cêch
approach was used to deal with the general toric case (thus real polarizations given
by integrable also admitting elliptic singularities).6
In order to do this, we will apply a Künneth argument that will allow to prove
the result by recursion on the following Lemma which addresses the 2-dimensional
case.
Lemma 4.1. Fix the domain W = (−, )×S 1 endowed with a symplectic structure
of integer class ω and consider as real polarization the vertical foliation by circles.
Then,
(1) In case there are no Bohr-Sommerfeld leaves them the Geometric Quanti-
zation is zero.
(2) If there is one Bohr-Sommerfeld leaf, then the Geometric Quantization is
given by H 0 (W, J ) = 0 and H 1 (W, J ) = C.
Proof. Take coordinates (x, θ) ∈ (−, ) × S 1 . Recall
R x that ω = f (x, θ)dx ∧ dθ, for
1
some f (x, θ) > 0. Fix S = R/Z. The form λ = ( 0 f (t, θ)dt)dθ is a primitive for
ω in the open set (−, ) × (0, 1). So λ = h(x, θ)dθ, where dh dx
> 0. Let us take a
system of coordinates over U in which we can trivialize the prequantization bundle
in the horizontal directions starting with a parallel section along the vertical axis.
5For
this we can follow exactly the same argument of [BT82] (page 50) which consists in
picking a partition of unity subordinated to the open sets U and V .
6A different proof of this result seen from the Poisson perspective is a joint work of the first
author of this paper with Mark Hamilton [HM12].
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 17

In this bundle trivialization, the connection 1-form for the prequantizable bundle
is expressed as
∇ = d − 2πih(x, θ)dθ.
Let us denote by s0 this trivializing section. Any parallel section s : W →
L ' W × C can be expressed in terms of this trivializing section by means of the
following formula,
Rθ
2πih(x,s)ds
(13) s(x, θ) = f (x, 1/2)e 1/2 s0 .
From now on, and for the sake of simplicity, we will identify the section sU (x0 , θ0 )
in a given neighbourhood with a function f (x0 , θ0 ) via this trivializing section.
Now we take a covering of the domain W by the pair of open sets
U = (−, ) × (−0.1, 0.6).

V = (−, ) × (0.4, 1.1).

The intersection is the domain
U ∩ V = (−, ) × ((−0.1, 0.1) ∪ (0.4, 0.6)) = W1 ∪ W2 .
Lemma 3.7 entails that H 1 (D, J ) = 0 and H 0 (D, J ) = C ∞ ((−, ), C) for any of
the domains D ∈ {U, V, W1 , W2 } because all of them are cotangent balls.

1
W1
0.9

U
0.6
W2
0.4

V
0.1 W1
0

−ε +ε
Figure 1. Picture of the Mayer-Vietoris covering.
18 EVA MIRANDA AND FRANCISCO PRESAS

We apply the Mayer-Vietoris sequence (3.2) to the pair U, V to obtain

r1 −r2
0 / C ∞ ((−, ), C) ⊕ C ∞ ((−, ), C) / C ∞ ((−, ), C) ⊕ C ∞ ((−, ), C)
ccccc
ccccccccccccccc
ccc
cccccccc
ccccccccccccccc
c
q /0 /0
H 1 (M, J )
Observe that H 0 (W, J ) = 0 because there are no global flat sections since the
space of closed non Bohr-Sommerfeld orbits is dense and over them any global
section has to vanish.
Let us now compute H 1 (W, J ): the exactness of the sequence yields
(C ∞ ((−, ), C) ⊕ C ∞ ((−, ), C))
H 1 (W, J ) =
Im(r1 − r2 )
So we just need to examine the image of r1 − r2 . In order to do that, Take
sections sU ∈ H 0 (U, J ), sV ∈ H 0 (V, J ), sW1 ∈ H 0 (W1 , J ) and sW2 ∈ H 0 (W2 , J ).
From formula (13)it is easily seen that the restrictions of the sections sU (x, 0.5),
sV (x, 0.5), sW2 (x, 0.5) and sW1 (x, 0) completely determine the sections. Also ob-
serve that we are identifying the sections with functions with the help of a local
trivializing section s0 . Thus the space of global flat sections of these domains is
C ∞ ((−, ), C). Now we want to determine the image of the morphism r1 − r2 .
Choose an element (sW1 , sW2 ) of the space H 0 (W1 , J ) ⊕ H 0 (W2 , J ). We look for
sections sU and sV such that the image under r1 − r2 are the given pair. For this
it is necessary to impose
(14) sW1 (x, 0.5) = sU (x, 0.5) − sV (x, 0.5).
Using the equation (13) we obtain the following condition
R0 R1
2πih(x,s)ds 2πih(x,s)ds
(15) sW2 (x, 0) = sU (x, 0.5)e 0.5 − sV (x, 0.5)e 0.5 .
The equations (16) and (17) provide a family of systems of two equations and
two variables depending on the parameter x ∈ (−, ). In order for them to have
solution for every value of the pair (sW1 , sW2 ), the following conditions needs to
hold,
1 −1
R0
2πih(x,s)ds
R1 6= 0.
e 0.5 −e 0.5 2πih(x,s)ds
Observe that this is not satisfied exactly for the values of x lying in a Bohr-
Sommerfeld orbit.
Summing up, this implies,
(1) In the case there are no Bohr-Sommerfeld leaves the image of r1 − r2 is
surjective and therefore H 1 (W, J ) = 0.
(2) In the case there is one Bohr-Sommerfeld leaf, the sections sW1 and sW2
are related by one equation; the image of r1 − r2 has codimension 1 and
H 1 (W, J ) = C.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 19

and this finishes the proof of the Lemma. 

Now, the classical Śniaticky theorem becomes a simple consequence of this
Lemma and Künneth formula. Let F be a Lagrangian foliation on a manifold
(M, ω). We say that a torus Tn , that is a leaf of the foliation, admits a cotan-
gent neighbourhood if there is a neighborhood U of Tn and a symplectic chart
φ : U → V ⊂ Tn × Rn such that
• The Lagrangian foliation becomes the foliation given by leaves Tn × {p}.
• There are (x1 , . . . , xn ) global affine coordinates in T2 and (y1 , . . . , yn ) the
standard coordinates of Rn , such that the symplectic form ωφ = φ∗ ω can
be written as
ωφ (x1 , . . . , xn , y1 , . . . , yn ) = Σni=1 dxi ∧ dyi .
As an example, any compact fiber of an integrable systems admits a cotangent
neighbourhood (the fiber is a Liouville torus and the foliation is semilocally a
fibration by tori [A89, D80]). We can now prove the following theorem which is a
reformulation of Śniaticky’s theorem.
Theorem 4.2 (Śniaticky). Let (M 2n , ω) be a symplectic manifold, N a smooth
manifold and π : M → N a fibration by tori. Assume that all the fibers admit
cotangent neighborhoods. Moreover assume that the number of Bohr-Sommerfeld
fibers is finite and equal to k. Then,
(1) H i (M, J ) = 0, for all i 6= n.
(2) H n (M, J ) = Ck .
(3) Q(M, P) = Ck .
Proof. Take a locally finite covering {Vi }i∈N of N in such a way that the image
by π of each Bohr-Sommerfeld leaf is covered by a single open set. Denote by
Ui = π −1 (Vi ) that is a locally finite covering of M . Without any loss of generality
we may assume that the first k indices correspond to the open sets containing the
Bohr-Sommerfeld orbits.
We claim that Q(Uj , P) = 0 for any j > k and Q(Ui , P) = C for i ≤ k. This
is because the neighborhood (Ui , Ji ) trivializes, using the cotangent neighborhood
trivialization, as:
n
Y
(Ui , .Ji ) = ((−, ) × S 1 , J˜),
l=1
where the sheaf J˜ is the one associated to the Lagrangian foliation defined in
the statement of Lemma 4.1. We are in the hypothesis of Künneth Theorem,
Theorem 3.4, and we can apply induction and Künneth formula to prove that the
Quantization of the open sets Ui is the one required.
Last but not least, we need to perform a Mayer-Vietoris
Sj argument to glue these
neighbourhoods. To that end, define Mj = i=1 Ui ; the Mayer-Vietoris sequence
yields,
20 EVA MIRANDA AND FRANCISCO PRESAS

(1) H i (Mk , J ) = 0, for all i 6= n.

(2) H n (Mk , J ) = Ck .
(3) Q(Mk , P) = Ck .
Moreover, the Mayer-Vietoris sequence also yields that we obtain Q(Mj , P) =
Q(Ml , P), for any j, l ≥ k. This concludes the proof. 

4.2. Irrational foliation of the 2-torus. Let T2 = R2 /Z2 be the 2-torus with
∂ ∂
coordinates (x, θ). Define the vector field Xη = η ∂x + ∂θ , with η ∈ R. This
vector field descends to a vector field in the quotient torus that we still denote
Xη . Denote by Pη the associated foliation in T2 . If η is an irrational number,
the foliation η is named the irrational foliation of the torus with slope η. It is
well-known [D32, K24] that any foliation of the torus without periodic orbits is
topologically conjugated to Pη for some irrational η. So the next result is close
to compute the Geometric Quantization of the 2-torus polarized by any regular
foliation without periodic orbits. The metalinear bundle N is trivial in this case.

Theorem 4.3. Let (T2 , ω) be the 2-torus with the standard symplectic structure
ω = pdx ∧ dθ of area p ∈ N and let Pη the irrational foliation of slope η in this
manifold. Then
(1) The Geometric Quantization space is infinite dimensional.
(2) The foliated cohomology space Q(T2 , J ) is infinite dimensional if the ir-
rationality measure of η is infinite (i.e. it is a Liouville number). If the
irrationality measure is finite then Q(T2 , J ) = C C.
L

Remark 4.4. It is interesting to point out here as it was done by Heitsch in

[H75] in the case η is not a Liouville number, there is a nice interpretation of the
elements of H 1 (M, J ) in the limit case of foliated cohomology as the associated
infinitesimal deformations of a differentiable family of foliations.

Proof. Take coordinates (x, θ) ∈ S 1 × S 1 . Recall that ω = kdx ∧ dθ, for some
k > 0. Let λη = dθ − η1 dx be the 1-form defining the Lagrangian foliation.
In the case of foliated cohomology α = 0 and therefore the parallel transport
equation implies that
s(x, 1/2) = s(x + k/η, 1/2),
for all k ∈ Z. Therefore the section is constant in a dense set, because of the
irrationality of η, and so it is constant. Thus H 0 (T2 , J ) = C. For the Geometric
Quantization case, assume that there is a section σ ∈ H 0 (T2 , J ). Therefore, since
the prequantizable bundle is topologically non-trivial we have that there is a point
p ∈ T 2 such that σ(p) = 0. The parallel transport along the leaf of the foliation
containing p allows to conclude that the section vanishes along the leaf. Since, the
leaf is dense we obtain that σ = 0. So, we have that H 0 (T2 , J ) = 0 as claimed.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 21

As in previous Sections, we cover T2 by two open sets U = S 1 × (−0.1, 0.6) and

V = S 1 × (0.4, 1.1). The intersections are given by
W1 = S 1 × (−0.1, 0.1),
W2 = S 1 × (0.4, 0.6).
To compute the Geometric Quantization of (T2 , Pη ) we use the Mayer-Vietoris
sequence applied to U and V . As we have already proved in Section 3, H 0 (S 1 ×
(−δ, δ), J ) = C ∞ (S 1 , C) and H 1 (S 1 ×(−δ, δ), J ) = 0. Thus the sequence becomes,
r1 −r2
0 / C ∞ (S 1 , C) ⊕ C ∞ (S 1 , C) / C ∞ (S 1 , C) ⊕ C ∞ (S 1 , C)
dd
ddddddd
dddd ddddddddd
d
ddddddd
r dddddd
d /0 /0
H 1 (T2 , J )

Now we have sections sU (x, θ), sV (x, θ), s1 (x, θ), s2 (x, θ) over U , V , W1 and
W2 respectively. By the parallel transport equation these sections are completely
determined by the restrictions sU (x, 0.5), sV (x, 0.5), s1 (x, 0), s2 (x, 0.5). We want
to solve the equations,

(16) s2 (x, 0.5) = sU (x, 0.5) − sV (x, 0.5).

Using equation (13) the following condition is obtained,
(17) s1 (x, 0) = sU (x, 0) − sV (x, 0).
Let us start by solving the foliated cohomology case. In that case the sections
are functions and the parallel transport implies just that the functions are constant
along each leaf. The equation (17) reads
η η
(18) s1 (x, 0) = sU (x + , 0.5) − sV (x − , 0.5).
2 2
We denote w(x) = sW1 (x, 0), ŵ(x) = sW2 (x, 0.5), u(x) = sU (x, 0.5) and v(x) =
sV (x, 0.5). Since they are smooth functions over the circle, they admit a Fourier
series representing them. The coefficients associated to the previous 4 functions
are wk , ŵk , uk and vk respectively. By Theorem 3.9 the decay of the Fourier
coefficients of uk and vk satisfies for any fixed positive integer q > 0 the following
relations,
|uk |
(19) lim = 0,
k→±∞ 1 + |k|q

|vk |
(20) lim = 0.
k→±∞ 1 + |k|q
22 EVA MIRANDA AND FRANCISCO PRESAS

The equations (16) and (18) imply the following set of equations in the Fourier
coefficients:      
ŵk 1 −1 uk
= · , k ∈ Z.
wk eπikη −e−πikη vk
Since the number η is irrational the equation has a unique solution expressed as
−e−πikη 1
     
uk 1 ŵk
(21) = , k ∈ Z − {0}.
vk −eπikη 1 eπikη − e−πikη wk
In particular , when ŵ0 6= w0 there is no solution.
Thus, we will assume from now on ŵ0 = w0 in order to have a solution.
If ν ≥ 2, the irrationality measure of η, is finite (i. e. it is not a Liouville
number), we obtain
p 1
(22) |η − | ≥ ν ,
k k
for any pair (p, k) with k large enough. From there we obtain
1
|kη − p| ≥ .
k ν+1
Therefore, we have
1 1 1
(23) −πikη
= 2πikη ≤ ≤ k ν+1 ,
eπikη −e |e − 1| |kη − p|
the first inequality comes from the inequality
|e2πit − e2πis | ≤ |t − s|.
We easily obtain from the equations (21), (19), (20) and the inequality (23) the
following limits, for any q ≥ 0:
|ŵk |
lim = 0,
k→±∞ 1 + |k|q

|wk |
lim = 0.
k→±∞ 1 + |k|q

Now, we have bounded the decay of the coefficients of the solutions u(x) and v(x).
We are in the hypothesis of the Theorem 3.9 to conclude that u and v are smooth
functions on the circle. Therefore going back to the long exact sequence, the map
r1 − r2 is surjective when w0 = ŵ0 yielding H 1 (T2 , J ) = C.
Now for the case of η being a Liouville number we may assume that, there exists
a sequence of pairs of positive integers {ps , ks }s∈Z∗ such that
ps 1
|η − | < s,
ks k
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 23

satisfying that ks is strictly increasing. From this equation we obtain arguing as

in the previous case that
1 kss−1
(24) > .
|e2πikη − 1| 2π
Define the constant sequence a(0)k = 1. Now, we define the functions ŵ = 0 and
w with Fourier coefficients
(
a(0)k
kss−1
, if ∃s ∈ Z+ , such that ks = l,
wl =
0, otherwise.
Notice that because of Theorem 3.9, the function w is smooth. However, by
combining the equation (21) and the inequality (24) we obtain
1
|vks | ≥ .
2π
Therefore, the function v, solving equations (16) and (18) for the selected input
data ŵ and w, is not smooth according to Theorem 3.9 since its Fourier coefficients
do not converge to zero for k large.
We have shown that the map r1 − r2 is not surjective. Therefore the cokernel of
the map is not zero. To show that the cokernel is infinite dimensional we define
the family of sequences a(r) defined as follows:

1, k 6= r,
a(r)k =
2, k = r
Then, we choose input data ŵ = 0 and w a smooth function with the following
prescribed Fourier coefficients,
(
a(r)k
kss−1
, if ∃s ∈ Z+ , such that ks = l,
wl =
0, otherwise.
The linear expansion of the set of choices conforms an infinite dimensional subspace
of functions and so it shows that the cokernel of r1 − r2 is infinite dimensional.
This implies that the dimension of H 1 (T2 , J ) is infinite.
In the case of Geometric Quantization, we assume that ω = pdx ∧ dθ, for a
fixed positive integer p. We fix two different trivialization sections σU and σV
over the regions U and V . Without loss of generality we assume that the circle
S 1 × {0} = S0 is a Bohr-Sommerfeld submanifold (for the horizontal polarization)
with fixed global section over it. They are defined by the condition that they
coincide over S0 with the Bohr-Sommerfeld section and that they are defined by
parallel vertical transport (upwards for σU and downwards for σV ). They satisfy
the gluing equation
(25) σU (x, 0.5) = σV (x, 0.5)e2πipx .
24 EVA MIRANDA AND FRANCISCO PRESAS

With respect to any of these two trivializing sections the connection of the pre-
quantizable bundle is written as
(26) ∇ = d + 2πipθdx.
Use the following notation
(1) sU (x, θ) = fU (x, θ) · σU (x, θ),
(2) sV (x, θ) = fV (x, θ) · σU (x, θ),
(3) s2 (x, θ) = f2 (x, θ) · σU (x, θ),
(4) s1 (x, θ) = f1 (x, θ) · σU (x, θ).
Dividing by σU (x, 0.5) on both sides of equation (16) we obtain
f2 (x, 0.5) = fU (x, 0.5) − fV (x, 0.5).
Now, by using the connection formula (26) over U we obtain, by parallel transport
along the leaves, the following formula
fU (x + ηθ, θ) = fU (x, 0) · e2πipθ/η .
Thus, we obtain
(27) fU (x, 0) = fU (x + 0.5η, 0.5) · e−πipθ/η .
Realize that
sV (x, θ) sV (x, θ) σV (x, θ)
fV (x, θ) = = = f˜V (x, θ)e−2πipx ,
σU (x, θ) σV (x, θ) σU (x, θ)

where f˜V (x, θ) = σsVV (x,θ)

(x,θ)
. It satisfies the equation, by a computation analogous to
the one providing (27),

(28) f˜V (x, 0) = f˜V (x − 0.5η, 0.5) · eπip/η .

Therefore, after quotienting by σU , equation (17) becomes,
f1 (x, 0) = fU (x, 0) − fV (x, 0),
that is, by substituting in the previous equations
(29) f1 (x, 0) = fU (x + 0.5η, 0.5) · e−πip/η − fV (x − 0.5η, 0.5) · eπip/η · e2πipx ,
We have smooth functions:
ŵ = f2 (x, 0.5),
w = f1 (x, 0),
u = fU (x, 0.5),
v = fV (x, 0.5).
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 25

They completely recover the initial four sections sU , sV , s1 and s2 . Expanding the
Fourier coefficients of them and substituting them into the equations (16) and (29)
we obtain the sequence of systems of equations

wˆk = uk − vk
−πi(kη+p/η) ,k ∈ Z
wk = e uk − eπi(kη+p/η) vk−p

The system is not linearly dependent since η + 1/η is irrational. Substituting the
first equation in the second and simplifying we obtain

(30) wk − e−πi(kη+p/η) ŵk = e−πi(kη+p/η) vk − eπi(kη+p/η) vk−p , k ∈ Z.

Let us compute a particular case. Assume that ŵ = 0 and w = 0. Therefore
wk = δ0k . This immediately implies that
vq = 0,
for q ∼6= 0 mod p. For the multiples of p, we can choose any v0 ∈ C. Substituting
in (30) for k = 0, we obtain the value of v−p . Using again equation (30) for k > 0,
it is simple to check that
|vk·p | = |v0 |,
for k > 0. Substituting for negative values we obtain
|vk·p | = |v−p |,
for k < −1. Therefore, at least one of the two limits of vk does not converge to
zero. This implies by Theorem 3.9 that any of the solutions (one for each choice
of v0 ) of equation (30) produces a function v that is not smooth.
The previous choice of w and ŵ can be slightly perturbed to produce examples
of pairs (w, ŵ) such that the associated solution (u, v) is not a smooth pair. This
shows that the dimension of the space H 1 (T2 , J ) is infinite. 
Remark 4.5. Most certainly a similar argument yields the infinite dimensionality
of the Geometric Quantization space for the case of a general symplectic form. It
is just a matter of complicating the formulae, but the basic non-decaying behavior
of the Fourier coefficients is kept.

5. Geometric Quantization of general foliations over the 2-torus

In this section we deal with the case of general real polarizations given by a
non–singular flow flow on the torus. For this we first recover the topological
classification and then apply the previous functorial properties to compute its
Geometric Quantization. In the way, we also obtain some results for foliated
cohomology associated to this foliation.
26 EVA MIRANDA AND FRANCISCO PRESAS

5.1. The topological classification. We recall here the topological Denjoy-

Kneser [D32, K24, Re62] classification, up to topological equivalence, of line fields
F on the torus. Let us point out that the classifications that we use here is the
C 0 classification and not the C 1 classification. For instance the so-called Denjoy
foliation (see for instance [A90])7 is not diffeomorphic to the irrational flow. We
follow the ideas and notation as presented in [Re62]. A line field of the torus is
completely determined by a map LF : T2 → RP1 , i.e. we are trivializing the tan-
gent bundle and taking its projectivization so each line becomes a point. If the
line field is oriented then the maps lifts to L̂F : T2 → S 1 .
Recall that the standard obstruction theory establishes that the homotopy type
of the line field, as a distribution, is determined by an obstruction class provided
by
λ ∈ H 1 (T2 , π1 (RP1 )) = H 1 (T2 , Z).
Fix a basis of H1 (T2 , Z) = Z2 provided by two elements represented as loops γ1
and γ2 that intersect transversely at the point (0, 0) ∈ T2 = R2 /Z2 . The maps
Lγi : S 1 → S 1 , i = 1, 2, have integer degrees d1 and d2 and λ([γi ]) = di . Therefore
the homotopy classes of line fields are represented by a pair of integers that are the
degrees of the restrictions of the map L to a positive basis of H1 (T2 , Z). Moreover,
it is obvious that the distribution is orientable if the degrees are even. Realize that
the integers depend on the trivialization chosen on the torus, but for each fixed
trivialization the classes run over all the possible pairs of integers.
The following lemma holds,
Lemma 5.1. A regular foliation F over the torus T2 always admits a standard
metaplectic correction.
Proof. The real line bundles F over T2 are completely classified by the first Stieffel-
Withney class w1 (F) ∈ H 1 (T2 , Z2 ) = (Z2 )2 . Therefore, there exist 4 of them. We
have the following exact sequence
·2
0 → Z → Z → Z2 →, 0
that induces a long exact sequence
b i
· · · H 1 (T2 , Z) → H 1 (T2 , Z2 ) → H 2 (T2 , Z) → H 2 (T2 , Z) · · · ,
where the map b is the Bockstein morphism that maps the first Stieffel-Whitney
class of a real line bundle w1 (F) to the first Chern class of its complexification
c1 (F ⊗R C). The metaplectic correction gives a choice of square root of F ⊗ C.
So, we just need to have c1 (F ⊗R C) ∈ H 2 (T2 , Z) = Z of even degree. But since
i = 2·id is injective, we obtain c1 (F ⊗R C) = 0. Therefore we have Λ1 (L⊗R C) = C
and the trivial bundle can be chosen as bundle of half-forms, i.e. the square root
7TheDenjoy example has been a key example in the theory of foliations. For instance, it was
used by Paul Schweitzer to give a counterexample to the Seifert conjecture for closed orbits for
C 1 flows on S3 and indeed on any three dimensional manifold (see [S74]).
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 27

of the trivial line bundle is itself. Therefore the prequantizable bundle remains the
same after the metaplectic correction and we call that choice the standard one. 
Now assume that there is a closed leaf of the line field L. It represents an
element [L] of H1 (M, Z). It is obvious that λ([L]) = 0. Any other closed leaf
L0 represents the same homology class since it cannot intersect L (since different
orbits are disjoint) and by Poincaré-Bendixon Theorem cannot be null-homologous.
Therefore, there are always homology classes which are not represented by closed
leaves.
For any homology class A ∈ H1 (M, Z), we have the following result that is
proved using Sard’s lemma.
Lemma 5.2. Fix A ∈ H 2 (M, Z). There exists an embedded smooth loop γ : S 1 →
T2 representing the class such that the number of tangencies of γ and F is finite.
Consequently, we can minimize that number obtaining the following,
Definition 5.3. A minimal contact curve γ for a class A ∈ H1 (M, Z), not repre-
sented by closed curves, is a smooth curve that minimizes the number of tangencies
with F.
We obtain
Corollary 5.4. The leaf L0 of L at a tangency point t0 of a curve of minimal
contact do not cross the curve.
Proof. We can choose a small chart around x0 such that t0 = (0, 0) ∈ R2 , the
distribution L is locally given by the equation { y=const } and the curve γ :
(−, ) → R2 is written as
γ(t) = (t, f (t)),
0
with f (0) = f (0) = 0 and moreover we assume, by hypothesis, that f (t) is
increasing in a neighborhood of t = 0. Then, it is simple to locally perturb
(see figure 2) f to a new g such that γ̂(t) = (t, g(t)) has no tangency points in
the neighborhood of (0, 0). This implies that γ̂ has less tangencies than γ and
therefore the initial γ was not a minimal contact curve. 
We say that a tangency of a minimal contact curve γ is positive if the map
Lγ : S 1 → S 1 is increasing at the tangency point, it is negative if it is decreasing.
Assume that the homology class representing the closed leaves is the class (0, 1) ∈
Z2 = H1 (T2 , Z). This can be always satisfied by composing the T2 = R2 /Z2
torus by an element in SL(2, Z). The minimal contact curve can be assumed to
lie in the class (1, 0). Now, any open leaf is diffeomorphic to the real line. It is
known that the semiline (semiorbit) of the open leaf is asymptotically tangent to a
closed orbit. Therefore, denoting by γ : R → T2 a parametrization of the line and
γ̄ = (γ̄1 , γ̄2 ) : R → R2 the canonical lift to the universal cover, then the following
limit exists
lim γ̄2 (t),
t→∞
28 EVA MIRANDA AND FRANCISCO PRESAS

γ γ
L0 L0

Figure 2. Deformation of γ such that γ̂(t) = (t, g(t)) has no tan-

gency points.

and it is either +∞ or −∞. We say that the semiorbit is positive if the limit is
+∞ and negative otherwise.
We obtain,
Theorem 5.5 (Proposition 3 in [Re62]). Let Γ of minimal contact have µ+ positive
tangencies and µ− negative tangencies. Then the leaves of the foliation can be
described as follows:
(1) There are µ+ regions bounded by a pair of (possibly not distinct) closed
leaves, such that all the leaves interior to this region are open and have
both semiorbits negative.
(2) There are µ− regions with both semiorbits positive.
(3) There are some other regions (at most numerable in number) in which the
two semiorbits of a given orbit have opposite signs.
(4) If the complement of these regions is not the whole space, it is composed
entirely of closed leaves.
(5) If the complement is the whole space either all the leaves are closed, or all
are dense.
A foliation is called generic if the linear monodromy of the closed leaves is non-
degenerate (not the identity), i.e. the linearized Poincaré return map is of the
form p(t) = λt, with λ > 0 and different from 1. This, in particular, implies
that the closed leaves are isolated and stable under C 1 -small perturbations. The
Theorem 5.5 restricts to this particular case as follows. The regions described in
the Theorem behave as follows for a generic foliation:
• The regions described in the points (1), (2) and (3) are finite in number,
• the rest of the regions do not exist.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 29

There is a local linearization Theorem for the neighborhood of a closed leaf with
non-degenerate linear monodromy.
Proposition 5.6. Let U ' S 1 × (−, ) a neighborhood of a non-degenerate closed
leaf F ' S 1 ×{0} of a foliation L, then there is a smaller neighborhood F ⊂ V ⊆ U
and a diffeomorphism φ : V → S 1 × (−δ, δ), for some δ > 0, such that,
(1) φ−1 (S 1 × {0}) = F,
(2) fixing coordinates (θ, r) ∈ S 1 × (−δ, δ), we have φ∗ (ker{dr + λ · r · dθ}) = L
for some λ 6= 0.
Proof. By shrinking U if necessary, we may assume that L can be expressed as
ker α = ker{f · dr + g · dθ} for some f, g functions on S 1 × (−, ) with f > 0. So
we assume that the foliation is expressed by the kernel of the form (after suitable
re-scaling)
(31) α = dr + Hdθ,
with H : S 1 × (−, ) → R, satisfying that H(θ, 0) = 0. Moreover, there exists a
change of coordinates r̂ = r̂(r) such that r̂(0) = 0 satisfying that the foliation is
expressed as the kernel of a new form
(32) α̂ = dr̂ + Ĥdθ,
where the function Ĥ satisfies the same properties as the previous smooth function
H. The only difference being that the Poincaré’s return map associated to the
transverse segment m : (−r̂0 , r̂0 ) → (−r̂1 , r̂1 ) defined for some small r̂0 > 0 and
r̂1 > 0 satisfies that is purely linear m(r̂) = c · r̂, for some c > 0 and c 6= 1. This is
proved by using the non-degeneracy condition and the classical Fatou’s Lemma on
the linearization of contracting germs of diffeomorphisms in the real line [AR95].
Thus we may assume, without loss of generality, that the Poincaré’s return map
is linear in the initial coordinates provided by the equation (32). We can restrict
ourselves to the domain [0, 1) × (−δ, δ) ⊂ S 1 × (−δ, δ) with coordinates (θ, r). We
can easily change coordinates in that domain to (θ, r) = φ(θ, R) = (θ, r(θ, R))
in such a way that r(0, R) = R satisfying that ker φ∗ α = dR. This change of
coordinates is given by the solution of the differential equation
∂r
+ H(r, θ) = 0,
∂θ
with initial value r(R, 0) = R. This provides a unique diffeomorphism φ that is well
defined over [0, 1)×(−, ) for some small  > 0. We change the coordinates by the
diffeomorphism (R, θ) = Φ(ρ, θ) = (R(ρ, θ), θ) with R(ρ, θ) = ρeλθ . This implies
that αλ = ker Φ∗ (dR) = dρ + λρdθ that is well-defined in (θ, ρ) ∈ [0, 1) × (−0 , 0 )
for some 0 > 0. By adjusting λ > 0, we can make the parallel transport along
the foliation coincide with the Poincaré’s return map. If this is the case the chart
Φ ◦ φ smoothly extends to a diffeomorphism in S 1 × (−0 , 0 ). This completes the
proof. 
30 EVA MIRANDA AND FRANCISCO PRESAS

From now on, the previously constructed neighborhood V of a periodic orbit

will be called a non-degenerate annulus. A neighborhood of the zero section of
the cotangent bundle T ∗ S 1 with the vertical Lagrangian foliation will be called a
cotangent annulus. Recall that its Geometric Quantization has been computed in
Lemma 4.1.

5.2. Geometric Quantization of the torus: The computation. We can sum

up the discussion above in the following,
Corollary 5.7. Let L be a generic foliation of the torus, with N , N > 0, closed
leaves. Then there exists a finite covering by open sets {Vj }j=N
j=1
+1
such that:
(1) Vj Tis diffeomorphic to S 1 × (0, 1),
(2) Vj T Vk = ∅, if j − k 6= ±1 mod N ,
(3) Vj Vj+1 is diffeomorphic to a cotangent annulus.
(4) If j ≤ N , Vj is diffeomorphic to a a non-degenerate annulus.
(5) VN +1 is a cotangent annulus.
This result is immediate from all the previous considerations (see Figure 3).

V2 VN-2 VN

{
{
{

...
{

V3
{

V1 VN-1

Figure 3. Covering by open sets {Ui } of a torus with a generic

foliation. The red curves are the leaves.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 31

In order to compute the Geometric Quantization and the foliated cohomology

of a non-degenerate foliation of the torus, we just need to apply Mayer-Vietoris
and the following,
Lemma 5.8. Let V = S 1 × (−1, 1) be a non-degenerate annulus with standard
foliation provided by
ker α = ker{dr + λ · r · dθ}.
Then for any symplectic form (even in the foliated cohomology case with ω = 0),
• If the closed leaf is Bohr-Sommerfeld, then H 0 (V, J ) = C, otherwise H 0 (V, J ) =
0.
• H 1 (V, J ) = C.
Remark 5.9. Recall that in foliated cohomology case, any closed leaf is Bohr-
Sommerfeld. Also observe that the constant λ may be always assumed to be negative
(possibly after composition with an orientation reversing diffeomorphism (θ →
−θ)).
Proof. If the leaf is Bohr-Sommerfeld then the sheaf admits a parallel section over
it sB : S 1 → L. Fix a non-compact leaf determined by the unique point pθ = (θ, 1)
in which it touches the boundary. The bundle L is topologically trivial so it admits
a connection ∇ = d + α. Since F is Bohr-Sommerfeld we may assume that α|F = 0
and thus
α = rβ,
for some 1-form β. Fix a value s(pθ ) ∈ L, by parallel transport it uniquely extends
to a parallel section over the whole leaf. Therefore if we fix any section s(pθ ) :
S 1 × {1} → L, it extends to a unique parallel section ŝ(pθ ) : S 1 × (0, 1] → L. We
want to check how it extends to the boundary S 1 × {0}. Fix a value θ0 ∈ S 1 . We
want to compute the following limit
(33) lim ŝ(θ0 , r),
r→0

since the invariance of the section along the leaves of the section allows us to reduce
the existence of the
lim ŝ(θ, r),
(θ,r)→(θ0 ,0)

to the proposed limit along the curve (θ0 , r).

Indeed, the computation of the previous limit (33) is ensured in two steps. First,
fix a point (θ0 , r0 ), we will compute the following limit
(34) lim ŝ(θ0 , mn (r0 )),
n→+∞

where mn is the n-th iterate of the Poincaré’s return map. The key point is to
check the following inequality
lim ||ŝ(θ0 , mn+1 (r0 )) − ŝ(θ0 , mn (r0 ))|| ≤ Ce−n·λ
n→∞
32 EVA MIRANDA AND FRANCISCO PRESAS

that holds in the case in which S 1 × {0} is Bohr-Sommerfeld. In any other case,
the limit does not exist and we get that H 0 (V, J ) = 0. The previous inequality
implies the existence of the limit (34).
Secondly we study the section ŝ(θ0 , r) restricted to the interval {θ0 }×[r0 , m1 (r0 )].
It is completely determined by the value at r0 and by the condition that the limit
(34) has to be independent of the choice of r ∈ [r0 , m1 (r0 )]. This shows the
existence of the limit (33). Therefore each element of H 0 (V, J ) can be recovered
out of a parallel section at S 1 × {0}. More precisely, we have recovered a unique
parallel section over S 1 × [0, 1], but an analogous argument recovers a section over
S 1 × [−1, 0] and the two glue together to provide a global smooth section. This
shows that H 0 (V, J ) = C if the closed leaf is Bohr-Sommerfeld and H 0 (V, J ) = 0
otherwise.
To compute the group H 1 (V, J ), we start by an element of Ω1 (V, J ) and we
study under which conditions is exact. The restriction to any non-compact leaf
is exact because of the standard (parametric) Poincaré lemma applies to the real
line. In the case of the closed leaf the condition to be exact amounts to be in the
kernel of an linear operator Ω(V, J ) → C, i.e. the integral along the leaf S 1 × {0}.
Let us detail that operator. Fix an element γ ∈ Ω1 (V, J ). We want to construct a
section s : S 1 × {0} → L. We fix an arbitrary (non-zero) value v at L(0,0) . There
is a unique section in [0, 1) × {0} such that:
• s(0, 0) = v,
• dF s(θ, 0) = γ(θ, 0), ∀θ ∈ (0, 1).
Consider
P : Ω1 (V, J ) → L(0,0) ' C
γ → s(1, 0) − s(0, 0)
Observe that γ is exact if and only if P (γ) = 0. Therefore H 1 (V, J ) = ImP = C.
This concludes the proof.

Lemma 4.1 and 5.8 compute the Geometric Quantization provided by all the
types of open sets appearing in the Corollary 5.7. Therefore, the computation of
the Geometric Quantization becomes a simple task
Corollary 5.10. Let F be a non-degenerate regular foliation over the torus with
N > 0 closed leaves. Assume that 0 ≤ b ≤ N of them are Bohr-Sommerfeld.
Then,
• H 0 (T2 , J ) = C if b = N and a parallel transport condition (to be described
in the proof ) is fulfilled. H 0 (T2 , J ) = 0 otherwise.
LN ∞
• H 1 (T2 , J ) = 1 b(i)
) CN , where b(i) = 1 if the i-th
L
i=1 (C (S , C)/(C)
closed leaf is Bohr-Sommerfeld; b(i) = 0 otherwise.
• The same holds for the foliated cohomology case.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 33

Proof. We recall the Mayer-Vietoris Lemma (Lemma 3.2). It states that for two
open domains U and V we have the following exact sequence:
0 / H 0 (M, J ) / H 0 (U, J ) ⊕ H 0 (V, J ) / H 0 (U ∩ V, J )
eee eeeeee
e
eeeeee
eeeeeeeeeee
e
r eeeee 1
e
H 1 (M, J ) / H (U, J ) ⊕ H 1 (V, J ) / H 1 (U ∩ V, J )
Away fromSthe open sets defined in Corollary 5.7, we define the sequence of open
sets Uj = jk=1 Vk . It is clear that UN +1 = T2 . It is simple to check that
j
M M
1
H (Uj+1 , J ) = (C ∞ (S 1 , C)/(C)b(i+1) ) Cj+1 , j = 1, . . . , n − 1,
i=1
S
by sequentially applying the Mayer-Vietoris sequence to the sets Uj+1 = Uj Vj+1 .
All the instances follow the same pattern and a recursive argument applies. The
exception is the last one, being T2 = UN +1 = UN VN +1 , in which the algebraic
S
computation changes since VN +1 is a cotangent annulus. 
Notice that the condition on closed leaves not to be Bohr-Sommerfeld is a generic
one.
Thus the Quantization space depends only on the topology of the foliation.
Concretely, just in the number of closed orbits. In the general case, it depends
on the symplectic geometry of the foliation, i.e. the number of Bohr-Sommerfeld
leaves is invariant under symplectic diffeomorphisms. In a future work [MP12] we
plan to use this idea to define the Quantization of a general Hamiltonian on a torus,
since the Quantization spaces remain unchanged by the flow of the Hamiltonian
and construct explicit isomorphisms of the Quantization spaces.

References
[A69] M. F. Atiyah; I. G. Macdonald: Introduction to Commutative Algebra. Oxford 1969,
Addison Wesley Publishing Company.
[A89] V. I. Arnold, Mathematical methods of classical mechanics, Springer Graduate texts in
mathematics, 60, Second edition, 1989, Springer-Verlag, New York.
[A90] K. Athanassopoulos, The flow near nontrivial minimal sets on 2-manifolds. Math. Proc.
Cambridge Philos. Soc. 108 (1990), no. 3, 569-573.
[AR95] P. Ahern, J-P Rosay, Entire functions in the classification of differentiable germs tangent
to the identity, in one and two variables, Transactions of the AMS, 347 (1995), n. 2.
[B11] M. Bertelson, Remarks on a Künneth formula for foliated de Rham cohomology. Pacific
J. Math. 252 (2011), no. 2, 257–274.
[B00] M. Bertelson, Foliations associated to regular Poisson structures, Ph.D. Thesis, Stanford
University (June 2000).
[B72] R.Bott, Lectures on Characteristic Classes and Foliations, Lecture Notes in Math., vol.
279, Springer-Verlag, 1972, pp. 1-80.
[BT82] R. Bott, L.W. Tu,. Differential forms in algebraic topology. Graduate Texts in Mathe-
matics, 82. Springer-Verlag, New York–Berlin, 1982. xiv+331 pp.
34 EVA MIRANDA AND FRANCISCO PRESAS

[Br97] G. E. Bredon, Sheaf theory. Second edition. Graduate Texts in Mathematics, 170.
Springer-Verlag, New York, 1997. xii+502 pp.
[D32] A. Denjoy, Sur les courbes définies par les équations différentielles à la surface du tore,
J. Math. Pures Appl. (9), 11 : 4 (1932) pp. 333-375.
[D80] J.J Duistermaat, On global action-angle coordinates, Communications on pure and applied
mathematics, 23, (1980), 687-706.
[G93] P. Gilkey, On the index of geometrical operators for Riemannian manifolds with boundary.
Adv. Math., 102 (1993), pp. 129-183.
[GS83] V. Guillemin and S. Sternberg. The Gelfand-Cetlin system and quantization of the com-
plex flag manifolds. J. Funct. Anal., 52(1):106–128, 1983.
[Gr09] L. Grafakos, Classical Fourier Analysis, 2nd edition, Springer, 2008.
[G54] A. Grothendieck, Séminaire Schwartz de la Faculté des Sciences de Paris, 1953/1954.
Produits tensoriels topologiques d’espaces vectoriels topologiques. Espaces vectoriels
topologiques nucléaires. Applications. Exposé 24, Secrétariat mathématique, 11 rue Pierre
Curie, Paris, 1954. iii+144 pp.
[H75] J.Heitsch, A cohomology for foliated manifolds Commentarii Mathematici Helvetici,
50(1975), 197-218.
[H11] M. Hamilton, “Locally toric manifolds and singular Bohr-Sommerfeld leaves,” Mem.
Amer. Math. Soc. 207 (2010), no. 971, vi+60 pp. ISBN: 978-0-8218-4714-5.
[HM10] M. Hamilton and E. Miranda, Geometric quantization of integrable systems with hyper-
bolic singularities. Ann. Inst. Fourier (Grenoble) 60 (2010), no. 1, 51–85.
[HM12] M. Hamilton and E. Miranda, Geometric quantization of Poisson manifolds. work in
progress.
[K67] L. Kaup, Eine Künnethformel für Fréchetgarben. Math. Z. 97 1967 158–168.
[K94] A. A. Kirillov, Geometric Quantization, from the book Dynamical Systems IV, Springer-
Verlag, (1994).
[KMN10] William D. Kirwin, José M. Mourão, João P. Nunes, Degeneration of Kaehler struc-
tures and half-form quantization of toric varieties, arXiv:1011.3363, 2010.
[K24] H. Kneser, Hellmuth, Reguläe Kurvenscharen auf den Ringflächen. Math. Ann. 91 (1924),
no. 1-2, 135-154.
[K75] B. Kostant, “On the Definition of Quantization,” Geometrie Symplectique et Physique
Mathematique, Coll. CNRS, No. 237, Paris, (1975), 187-210.
[K83] A. El Kacimi-Alaoui, Sur la cohomologie feuilletée. Compositio Math. 49 (1983), no. 2,
195-215.
[MP12] E. Miranda, F. Presas, Geometric Quantization of general Lagrangian foliations., work
in progress.
[P85] M.Puta, Some remarks on the cohomology of a real foliated manifold. Rend. Mat. (7) 5
(1985), no. 1-2, 189-201.
[Re62] B. Reinhart, Line elements on the torus. Amer. J. Math. 81 1959 617–631.
[Ra77] J. Rawnsley, ” On the Cohomology Groups of a Polarization and Diagonal quantization”,
Transaction of the American Mathematical Society, Vol 230 (1977) pp 235-255.
[S60] I. E. Segal, Quantization of nonlinear systems. J. Math. Phys. I, 468-488 (1960).
[Sni75] J. Śniatycki, On Cohomology Groups Appearing in Geometric Quantization, Differential
Geometric Methods in Mathematical Physics I (1975).
[S74] P. Schweitzer, Counterexamples to the Seifert conjecture and opening closed leaves of
foliations. Ann. of Math. (2) 100 (1974), 386-400.
[TZ99] Y. Tian, W. Zhang, Quantization formula for symplectic manifolds with boundary. Geom.
Funct. Anal., 9 (1999), pp. 596-640.
 GEOMETRIC QUANTIZATION OF REAL POLARIZATIONS VIA SHEAVES 35

[W71] A. Weinstein, Symplectic manifolds and their Lagrangian submanifolds Adv. Math. 6
(1971), 329-346.
[W92] N.M.J. Woodhouse, Geometric quantization, Second edition. Oxford Mathematical
Monographs, Oxford Science Publications, The Clarendon Press, Oxford University Press,
New York, 1992.

Eva Miranda, Departament de Matemàtica Aplicada I, Universitat Politècnica

de Catalunya, EPSEB, Avinguda del Doctor Marañón, 44-50, 08028, Barcelona,
Spain, e-mail: [email protected]

Francisco Presas, ICMAT, UAM-UCM-UC3M, Nicolás Cabrera, n 13-15, Campus

Cantoblanco UAM, 28049 Madrid e-mail: [email protected]