Geometric Quantization of Poisson Manifolds via Symplectic Groupoids
Daniel Felipe Bermudez´ Montana˜
Advisor: Prof. Alexander Cardona Gu´ıo, PhD.
A Dissertation Presented in Candidacy for the Degree of Bachelor of Science in Mathematics
Universidad de los Andes Department of Mathematics Bogota,´ Colombia 2019
Abstract
The theory of Lie algebroids and Lie groupoids is a convenient framework for study- ing properties of Poisson Manifolds. In this work we approach the problem of geometric quantization of Poisson Manifolds using the theory of symplectic groupoids. We work the obstructions of geometric prequantization and show how they can be understood as Lie algebroid’s integrability obstructions. Furthermore, by examples, we explore the complete quantization scheme using polarisations and convolution algebras of Fell line bundles.
ii Acknowledgements
Words fall short for the the graditute I have for people I will mention. First of all I want to thank my advisors Alexander Cardona Guio and Andres Reyes Lega. Alexander Cardona, his advice and insighfull comments guided through this research. I am extremely thankfull and indebted to him for sharing his expertise and knowledge with me. Andres Reyes, as my advisor has tough me more than I could ever ever give him credit for here. By example he has shown me what a good scientist, educator and mentor should be.
To my parents, my fist advisors: Your love, support and caring have provided me with the moral and emotional support to confront every aspect of my life; I owe it all to you. To my brothers: the not that productive time I spent with you have always brought me a lot joy - thank you for allowing me the time to research and write during this work. I am also grateful to my best friend, her encouragement and affection motivated me when times got rough; I am lucky to know her.
At last, a special gratitude for those collegues and friends whose bewteen humor and thoughtful conversations have shaped me and this work.
iii For all those who have guided me through this work, for all those who distracted me from finishing it, but most of all for those who did both.
iv
Contents
Abstract ...... ii Acknowledgements ...... iii
1 Introduction 1
2 A First Look at Symplectic Groupoids 4 2.1 Lie Groupoids Basics ...... 4 2.2 Lie Algebroids Basics ...... 8 2.3 Integration Theory of Lie Algebroids ...... 14 2.4 Symplectic Groupoids ...... 18
3 Geometric Prequantization of Symplectic Groupoids 24 3.1 Prequantization of Symplectic Spaces ...... 24 3.2 The Groupoid and Algebroid Picture ...... 30 3.3 Quantization of Poisson Manifolds ...... 34
4 Quantization of Poisson Manifolds 39 4.1 Quantization of Symplectic Manifolds ...... 39 4.2 Quantization of Symplectic Groupoids ...... 44 4.3 Examples ...... 51
5 Outlook 52
Bibliography 53
v Chapter 1
Introduction
Loosely speaking quantization is a process for constructing a quantum theory for a given classical system. The generality of the procedure may range, depending on the definition of classical and quantum systems and the properties one would like to preserve from the former on the quantization process.
A first approach would be to define a classical system as a symplectic manifold, which is manifold together with a non-degenerate, closed 2-form. Within this approach, symplectic manifolds should be regarded as the phase space of some classical system. Similarly, the functions on the symplectic manifolds may be thought of as the observables of the system and the symplectic form as a geometric object that encodes the information of the Hamilton equations of motion[1].
A first generalisation of the classical picture is to focus on the observables of the system rather than on the system itself. It turns out that the space of functions of a symplectic manifold is naturally associated with a bivector field which determines an antisymmetric bracket on differentiable functions that satisfies the Jacobi identity. A manifold with such a bivector field is known as a Poisson Manifold and its space of functions as a Poisson Algebra. Although every symplectic manifold is a Poisson manifold, the converse is not true. It occurs that some Poisson manifolds are not symplectic. In spite of this, Poisson and symplectic structures are intimately related. For example, a Poisson manifold is foliated by immersed manifolds where the restriction of the Poisson bracket actually comes from a symplectic form. One usually refers to this as the foliation by symplectic leaves. Thus, a Poisson manifold can be regarded as a degenerate, or singular, generalisation of a symplectic space [2].
On the quantum counterpart a first notion of a Quantum system is given by a Hilbert space. This is a complex inner product vector space that is complete with respect to the metric induced by the inner product. In this description, due primarily to Dirac and von Neumann, the vectors in the Hilbert space are states of the quantum system, the self-adjoint operators are the observables of the system and the inner product assigns an expected value to each of these observables[3]. As in the classical picture, a generalisation of the Hilbert space description of quantum mechanics arises by focusing on the observables of the system. The important algebraic structure of bounded operators on the Hilbert space is that of a C∗-algebra, which is a complete, involutive, normed algebra with the
1 C∗-property, namely kaa∗k = kak2 for all a in the algebra.
A first approximation to a rigorous definition of quantization was proposed by Dirac. From his point of view, the classical system was given by a symplectic manifold, the quantum system by a Hilbert space, and quantization was a map from the space of smooth functions on a symplectic manifold to self-adjoint operators on a Hilbert space, in such a way that the Poisson bracket of two observable becomes the commutator of the corresponding quantum observables. The first procedure for constructing such a map out of the geometric information of the symplectic manifold was due to Konstant[4], Souriau[5] and Kirillov[6] on a process today known as Geometric Quantization. Their approximation focused on realising the symplectic form on a line bundle, in a way that the Poisson algebra could be represented by a subset of the derivations on such line bundle. On some important examples, the description using symplectic manifolds and Hilbert spaces is not suitable to a description of the systems. In these cases a more general theory of quantization aims to assign a C∗-algebra to a Poisson manifold. In this approach to quantization the Poisson manifold is regarded as a geometrical approximation to a noncommutative algebra depending on a parameter ~. Within the rigorousity of this assignment one can consider formal deformation quantization [7] or strict deformation quantization [8].
Symplectic groupoids were independently introduced by Karas¨ev[9] and Weinstein [10] as a tool to study and generalise diverse quantization procedures. A symplectic groupoid is a Lie groupoid with a compatible symplectic structure; the base manifold of a symplectic groupoid is a Poisson manifold and, if it exists, the symplectic groupoid of such Poisson manifold is unique up to covering. Thus, a symplectic groupoid can be thought as a symplectification of the base Poisson manifold. If such symplectification exists, the geometric quantization procedure developed for symplectic manifolds can be generalized to quantized Poisson manifolds. This generalization, mainly due to Weinstein, Xu [11], Hawkings[12] and Landsman[13], has the aim of constructing a C∗-algebra out of the symplectic groupoid of a Poisson Manifold. This process yields a broad class of C∗-algebras such as commutative C∗-algebras, C∗-algebras of locally compact groups [14], crossed product C∗-algebras [15], non-commutative Tori and foliation C∗-algebras[16].
The process of geometric quantization of Poisson manifolds via symplectic groupoids can be summarised in four steps: first, a symplectification of the Poisson Manifold to a symplectic groupoid; second, geometric prequantization of the symplectic groupoid; third, polarization of the prequantum bundle of the symplectic groupoid; fourth and last, construction of the convolution algebra of the polarized prequantum bundle. In Chapter 2 we will deal with the first step of this process, with this purpose in mind we introduce the reader to the general theory of Lie groups, Lie algebroids and Symplectic groupoids. We discuss the integration theory of Lie algebroids and use it to describe the relationship between Poisson manifolds and symplectic groupoids. The second step of the process is handled in the third chapter, through which we review the prequantization theory of symplectic manifolds and symplectic groupoids. Prequantization is the method of assigning a line bundle to a given symplectic manifold, which within the groupoid scenario will be considered as a central extension of a groupoid. We discuss the obstructions to existence and uniqueness of a prequantization from a cohomological perspective in the symplectic manifold case. Furthermore, we show how in the groupoid framework these obstructions can be reinterpreted as integrability obstructions. At last, we complete an exposition of 2 the quantization process in Chapter 4, where the polarization of symplectic manifolds and groupoids is reviewed and the convolution algebra is described. Due to the complications arising from the polarization process, the main examples of the quantization given in this chapter will refer mainly to nicely behaved objects, namely cotangent groupoids.
Throughout this project, the reader is expected to have a general knowledge on the theory of symplectic manifolds, Poisson Manifolds, Hilbert spaces and C∗-algebras. To obtain an overview of the subject we give some general references. For an introduction to the theory of symplectic manifolds and their relationship with classical mechanics we refer the reader to [1, 17]; for an introduction to the theory of Poisson manifolds refer to [2]; and, for an introduction to the theory of Hilbert spaces and C∗-algebras to [14]. The first chapters of any of these references will be more than enough to follow the theory developed in this work. For a general, concise and complete overview on quantization we encourage the reader to look at the prelude of [18].
3 Chapter 2
A First Look at Symplectic Groupoids
In this chapter we introduce the reader to the theory of Lie Symplectic groupoids. For this, we start by introducing the general theory of Lie groupoids and Lie algebroids, following [19]. In a sense, if groups represent abstract symmetries of an object, then groupoids represent abstract symmetry transformations between possibly many objects. A Lie groupoid is a groupoid with a smooth structure. As in the case of a Lie group, the smooth structure lets us work with an infinitesimal version of the object, namely a Lie algebroid. It is worth mentioning that, unlike the Lie group/algebra scenario, to a given Lie algebroid there might not be a groupoid integrating it. We will discuss the obstructions to the integrability of a Lie algebroid. For this we follow [20] and [21].
Given a Poisson manifold P one can associate to it a algebroid A(P ) which will describe many of its infinitesimal properties. For instance, the Poisson cohomology of P will agree with the algebroid cohomology of A(P ) and the foliation described by the anchor of A(P ) will correspond with the foliation by symplectic leaves of P . In case that such an algebroid is integrable, we will be able to find a groupoid with additional structure, that of a symplectic groupoid, which will enlighten the global properties of the Poisson manifold. The symplectic groupoid in discussion is unique up to covering and has as base the original Poisson manifold. Thus, a symplectic groupoid can be regarded as symplectification of a Poisson manifold. For a detailed discussion the reader can consult [2] and [19].
2.1 Lie Groupoids Basics
One of the motivations to use Lie groupoids in differential geometry is the fact that the smooth category behaves badly with respect to quotients. If M is a smooth manifold with an equivalence relation ∼, then M/ ∼ is usually not a smooth manifold. We introduce the following proposition (Theorem 8.3 of [22]).
Proposition 2.1.1. Let ∼ be an equivalence relation on a manifold M and denote R ⊂ M × M the graph of the equivalence relation. Then, the following are equivalent: i) M/ ∼ has a unique manifold structure such that π : M → M/ ∼ is a submersion. ii) R,→ M × M is a closed embedding and π : R → M is a surjective submersion.
4 In the case of a Lie group G acting smoothly on a manifold M the previous proposition asserts that a sufficient condition for M/G to have a smooth structure is for the action to be free and proper. Proposition 2.1.2. Let G be a Lie group acting smoothly on a manifold M. If the action is free and proper, then the quotient M/G has a unique differentiable structure such that π : M → M/G is a submersion. The above condition is rather strong, hence working with quotients in the smooth category is quite restrictive. A way to avoid non smooth quotients is to enlarge the objects in consideration in a way the information of the equivalence relation can be retrieved in a different setting to avoid working with M/ ∼.
The latter approach leads to the notion of Lie groupoids.
Definition 2.1.3. A Lie groupoid G ⇒ M is a pair of manifolds (G,M) together with following maps:
i) Surjective submersions s, t : G → M. These maps are called source and target respectively.
ii) A partial multiplication map defined on the set of composable pairs, G(2) = {(g, h) ∈ G × G|s(g) = t(h)} m : G(2) → G (g, h) → gh, which is smooth and associative, i.e. (gh)k = g(hk) whenever one of the sides on the equality makes sense.
iii) An identity embedding 1 : M → G
x → 1x,
which is smooth and satisfies s(1x) = t(1x) = x and 1xg = g, g1y = g (whenever the the multipliacation is defined).
iv) A smooth map i : G → G g 7→ g−1,
−1 −1 such that gg = 1t(g) and g g = 1s(g). This map is called the inverse. Remark 2.1.4. Note that G(2) is the fibered product over M of the maps s, t : G → M,
(2) G = Gt ×s G
G G s t M 5 since s and t are submersion the space G(2) has a manifold structure.
The manifold M is called the base of the Groupoid. Sometimes where there is possibility of confusion we will denote the base manifold of a groupoid G by G(0) instead of M. Elements of M are called points, while elements of G are called arrows. In some important examples G is not second-countable and Hausdorff, therefore we will not ask for this in the definition of a groupoid. However we will ask that the base space M and the fibers s−1(m) are Hausdorff-second countable manifolds.
Having defined the main objects of study in this work, it is necessary to define their morphisms. Notice that the definition of a groupoid is actually the definition of a (internal) category in the category of smooth manifolds. Thus, morphisms of a groupoid are functors within the category.
Definition 2.1.5. A groupoid morphism between G ⇒ M and H ⇒ N is a pair (F, f) of smooth maps F : G → H and f : M → N such that the following holds:
i) The pair commutes with multiplication, i.e. ϕ(mG(g, h)) = mH(ϕ(g), ϕ(h)). Equiva- lently, the following diagram commutes,
ϕ×ϕ G(2) H(2)
mG mH ϕ G H
ii) The pair commutes with the source and target maps, that is sH ◦ ϕ = ϕ0 ◦ sG and tH ◦ ϕ = ϕ0 ◦ tG. Equivalently, the following diagrams commute.
ϕ ϕ G H G H
sG sH tG tH ϕ ϕ M 0 N M 0 N
Proposition 2.1.6. A groupoid morphism preserves identities and inverses, i.e. 1N ◦ ϕ = ϕ ◦ 1M and iH ◦ ϕ = ϕ ◦ iG. Proof. Let x ∈ M and g ∈ G such that s(g) = x. Then
ϕ(g) = ϕ(g1x) = ϕ(g)ϕ(1x),
symmetrically for every h such that t(h) = x it holds ϕ(h) = ϕ(1xh)ϕ(h). The previ- −1 ous equalities imply ϕ(1x) = 1ϕ(1x). Moreover, by the previous argument ϕ(g)ϕ(g ) = ϕ(1 ) = 1 and ϕ(g−1)ϕ(g) = 1 implying ϕ(g−1) = ϕ(g)−1. t(g) ϕ(1t(g)) ϕ(1s(g)) Along with the definition of a groupoid, there are some important notions which help to characterise its structure, for instance
Definition 2.1.7.
6 i) The orbit of x ∈ M is the set of points which are related by some arrow to x, i.e. −1 Ox = {t(s ({x}))} ⊂ M.
ii) The isotropy at x ∈ M is set of arrows which fix x, to be precise
−1 −1 Gx = {g ∈ G|s(g) = t(g) = x} = t ({x}) ∩ s ({x})
−1 It is worth noting that Gx is a pullback by a submersion, Gx = t ({x})t ×s G, thus Gx is a Lie group.
In the future, we will be interested in groupoids with some properties on the source fibers. Such properties will be denoted s-properties. For example, a groupoid with simply connected source fibers will be called s-simply connected, or in many occasions simply s-connected.
Examples Groupoids are sufficiently general objects to encompass a big class of important objects and examples in differential geometry. We present some of the examples that will be of interest later on. We follow [19], where greater details may be found.
Example 2.1.8. Manifolds: Let M be a manifold, by setting then s = t = Id, the set of multiplicative pairs is ∆(M) := {(m, m) ∈ ×M} and M ⇒ M is a groupoid with multiplicative structure defined by m · m = m.
Example 2.1.9. Lie Groups: Let G be a Lie group, consider the point as a base manifold M = {∗} and set s = t the maps to the point {∗}. Then G ⇒ {∗} is a groupoid with the is multiplicative structure of G.
Example 2.1.10. Pair Groupoid: Let M be a manifold. Consider G = M × M and G(0) = M. Define:
s(x, y) = y, t(x, y) = x, (x, y)(y, z) = (x, z), (x, y)−1 = (y, x), 1(x) = (x, x).
The previous maps define a groupoid structure on M × M ⇒ M. Example 2.1.11. Transformation Groupoids: Let α be an action of a Lie group G on a manifold M. Consider G = G n M := G × M and G(0) = M. Define:
s(g, x) = x t(g, x) = gx, (g, hx)(h, x) = (gh, x), (g, x)−1 = (x−1, gx), 1(x) = (e, x).
The previous maps define a groupoid structure on G n M ⇒ M. π Example 2.1.12. Gauge Groupoids: Let G P −→ M denote a principal bundle P with structure group G. Consider G = Gau(P ) := (P × P )/G, where G acts on P × P by the diagonal action and G(0) := M. Define:
s(g, x) = x t(g, x) = gx, (g, hx)(h, x) = (gh, x), (g, x)−1 = (x−1, gx), 1(x) = (e, x). 7 The previous maps define a groupoid structure on Gau(M) ⇒ M. Example 2.1.13. Cotangent Bundles of Lie Groups: Let G be a Lie group. Consider ∗ G = T G and G0 = g, where g is the Lie algebra of G. We will describe a groupoid structure ∗ ∗ on T G ⇒ g . Since G acts on itself by left translations Lg there exist a natural map, the pullback map ∗ ∗ µL : T G → g , µL(g, η) = η ◦ De(Lg). Notice the previous map trivialises the cotangent bundle T ∗G =∼ G × g∗. Similarly, for the right action there pullback map is µR(g, Y ) = η ◦ De(Rg). Define the groupoid structure by
s(g, η) = µL, t(g, η) = µR
(g, η) · (h, ξ) = (gh, η ◦ D(Rg ◦ Lh−1 )), −1 −1 (g, η) = (g , η ◦ DRg).
∗ Proposition 2.1.14. Consider the transformation groupoid Gng defined by the coadjoint action of G on g. The maps id : g → g and
∗ ∗ ϕ : G n g → T G;(g, η) 7→ (g, η ◦ D(Lg−1 )),
∗ ∗ ∗ ∗ define an isomorphism between the groupoids T G ⇒ g and G n g ⇒ g . Proof. Smoothness of the maps is clear. We will check it is a groupoid morphism. Recall ∗ Adgη = η ◦ D(Lg−1 ◦ Rg), thus
∗ t ◦ ϕ(g, η) = t[(g, η ◦ D(Lg−1 ))] = η ◦ D(Lg−1 ◦ Rg) = Adgη = t[(g, η)].
In the same fashion, s ◦ ϕ = ϕ0 ◦ s follows, proving property ii) of definition 2.1.5. To check property i):
ϕ[(g, η)(h, ξ)] = ϕ[(gh, η ◦ D(Rg ◦ Lh−1 ))] = [gh, η ◦ D(Rg ◦ Lh−1 ◦ L(gh)−1 )],
by the condition t(h, ξ) = s(g, η) it follows η ◦ D(Rg) = ξ ◦ D(Lh), implying
[gh, η ◦ D(Rg−1 ◦ Lh−1 ◦ L(gh)−1 )] = [gh, ξ ◦ L(gh)−1 )] = ϕ[(gh, ξ)] = ϕ[(g, η)(h, ξ)].
2.2 Lie Algebroids Basics
In Lie theory it is common to work with the Lie algebra of a group instead of the group itself. The Lie algebra is the infinitesimal object associated to a Lie group. This object bears most of the information of Lie group and is much easier to handle. Up to some connectedness and simply connectedness conditions, most of the information of the Lie group can be recovered from its Lie algebra via integration. In the case of Lie groupoids there is a similar procedure generalising that of Lie Groups: to every Lie groupoid one can assign a Lie Algebroid which is the infinitesimal object associated to it. However, this picture is not as well behaved as in the group case, mainly because integrability theory of Lie Algebroids is a more delicate issue here.
8 Definition 2.2.1. A Lie Algebroid on M is a vector bundle A → M with Lie bracket [·, ·] on Γ(A), together with a vector bundle map, called the anchor, # : A → TM
# A TM s.t [α, fβ] = f[α, β] + #α(f)β, M and the induced morphism in sections # : Γ(A) → X(M) is a Lie algebra morphism.
The anchor encodes a fair amount of geometric information of the Lie algebroid. For instance, the image of the anchor defines a distribution on M in the sense of Sussman [23].
Proposition 2.2.2. #(A) is in general not a vector bundle. However, it is an integrable distribution, i.e., for each x ∈ M there exists a maximal connected immersed submanifold Lx for which x ∈ Lx and TyLx = #(Ay) for all y ∈ Lx ⊂ M.
Proof. Following Theorem 4.2 of [23], it is enough to prove that for every m ∈ M there exist vector fields X1,...,Xk such that: 1 k (i) #x(Ax) = Span(X (m),...,X (m)), (ii) for every vector field X satisfying X = #α, α ∈ Γ(A), there exists > 0 such that ∞ i there are C functions fj , 1 ≤ i, j ≤ k, defined in (−, ) which satisfy
k i m X i j m [X,X ](ϕX (t)) = fj (t)X (ϕX (t)), j=1 m where ϕX (t) denotes the flow of the vector field X starting at m after time t.
To check the existence of such vector fields, let m ∈ M and consider a trivializing open ∞ set m ∈ Um for both A and TM. Thus Γ(A|U ) is finitely generated C -module with k ∞ generators α1, . . . , αk, and there exist fij ∈ C (M) such that
k [αi, αj] = fijαk.
We claim #α1,..., #αk are the desired set of vector fields. By definition #x(Ax) = Span(#α1,..., #αk) and since the anchor # is a Lie algebra morphism it holds
k k X k X k j [#αi, #αj] = #[αi, αj] = # fijαk = fij#α . j=1 j=1
Proposition 2.2.3. ker #x is a Lie algebra, called isotropy Lie algebra, under the bracket ˜ [αx, βx]ker #x = [˜α, β](x), whereα, ˜ β˜ ∈ Γ(A) satisfyα ˜(x) = αx, β˜(x) = βx.
9 Proof. Since the bracket [α, β] satisfies the Leibniz rule it depends just of the germ of α and β. Moreover, if α, β ∈ ker # it follows #α(f) = 0 for every f ∈ C∞(M), thus
[α, fβ] = f[α, β] + #α(f)β = f[α, β].
and the bracket is C∞(M) bi-linear when restricted to ker #. This implies the formula given in the proposition is well defined. It is clear from the Lie Algebroid axioms that this bracket defines a Lie algebra structure on ker #x.
Examples Just as in the case of Lie groupoids, Lie algebroids encompass a large class of objects and examples important in differential geometry. Example 2.2.4. Tangent Bundles: Let M be a manifold. Consider the tangent bundle TM → M, equipped with anchor # = Id and the vector field bracket the tangent bundle is a Lie Algebroid. Example 2.2.5. Lie Algebra: Let g be a Lie algebra considered as vector space, or equivalently as vector bundle over a point {∗}. By defining the anchor # as the map to the point {∗} and the bracket as the Lie Algebra bracket, the vector bundle g → {∗} is a a Lie algebroid. ρ˜ Example 2.2.6. Infinitesimal Action Algebroid: Let g −→ X(M) be a infinitesimal action (lie algebra homorphism) of a Lie algebra g. Define A = g n M := g × M → M. ∞ Since the bundle is trivial Γ(g n M) = C (M, g) and the bracket is uniquely determined if defined on constant sections α, β : M → g
#(α, x) =ρ ˜(α)|x [α, β](x) = [α(x), β(x)]g
To define the our next example we need a precise definition of Poisson manifold. Definition 2.2.7. A Poisson manifold (P, {·, ·}) is a Manifold together with a R-bilinear map C∞(P ) × C∞(P ) → C∞(P ), such that it satisfies the Jacobi identity, and is a derivation on each component. Example 2.2.8. Cotangent Space of a Poisson Manifold Let (P, {·, ·}) be a Poisson Manifold. The Poisson bivector Π ∈ Λ2TP defined by Π(df, dg) = {f, g}, and the induced map ∗ Π:˜ T P → TP ; df 7→ Π(df, ·) = Xf . Define the algebroid to be the vector bundle T ∗P → P , with anchorπ ˜ and bracket defined on df, dg by [df, dg] = d{f, g}. Many properties of the Poisson manifold P can be understood by means of the algebroid T ∗M. For example, the foliation on P associated to Π:˜ T ∗P → TP of proposition 2.2.2 is the foliation of P by symplectic leaves described in [2]. Another example is that where the Poisson cohomology of a Poisson manifold P corresponds to the algebroid cohomology of T ∗P , described in [2]. 10 Lie Algebroid Cohomology Definition 2.2.9. Let A → M be a Lie algebroid. An A-connection on a vector bundle E → M is a R-bilinear map ∇ : Γ(A) × Γ(E) → Γ(E) satisfying:
i)∇fαs = f∇αs
ii)∇αfs = f∇s + [ρ(α)(f)]s,
for all α ∈ Γ(A), s ∈ Γ(E) and f ∈ C∞(M):
Notice that a usual connection on a vector bundle E → M corresponds to a TM- connection. Just as in the case of usual TM-connections one can define curvature and torsion.
2 ∗ ∗ Definition 2.2.10. i) The curvature R∇ ∈ Γ(Λ A ⊗ E ⊗ E) of a A-connection ∇ is defined by R∇(α, β)s = ∇α∇βs − ∇β∇αs − ∇[α,β]s. 2 ∗ ii) The torsion of ∇ is T∇ ∈ Γ(Λ A ⊗ A) defined by
T∇(α, β) = ∇αβ − ∇βα − [α, β].
Definition 2.2.11. A representation of a Lie algebroid A → M is a vector bundle E → M together with a flat A-connection ∇, i.e. such that R∇ = 0. Example 2.2.12. Let (A → M) be a Lie algebroid and let L = M × R be the trivial line bundle on M, thus Γ(L) = C∞(M). Define the trivial A-connection ∇triv on L by
triv ∞ ∇α (f) = L#αf f ∈ C (M), α ∈ Γ(A).
The curvature of the trivial connection is [LX , LY ] − L[X,Y ], which is known to vanish. Given a representation (E, ∇) of A we will define a chain complex with values on (E, ∇). Consider the chain complex Ω(A, E) with differential
k X d∇ω(α0, ··· , αk) = ∇αi ω(α0, ··· , αˆi, ··· , αk)+ i=0 X i+j + (−1) ω([αi, αj], ω0, ··· , αˆi, ··· , αˆj, ··· , αk). i In particular for the case of the trivial representation the previous complex is known as the algebroid deRham complex and the corresponding cohomology as the algebroid deRham cohomology of A. In the Poisson manifolds case we have the following result which exhibit relationship between algebroids and Poisson geometry, Lemma 2.1 of [11]: Theorem 2.2.13. Let P be a Poisson manifold and T ∗P its corresponding algebroid. Then the Poisson cohomology of P is isomorphic to the algebroid deRham cohomology of T ∗P . 11 Lie Algebroid of a Lie Groupoid Let G be a Lie group and consider the set of invariant vector fields under right translation Rh : G → G defined by Rh(g) = gh R Xinv(G) = {X ∈ X(G)|(DRh)g(Xg) = Xgh}. R One can show that Xinv(G) with the bracket [X,Y ] = XY − YX is a Lie algebra, moreover, R Xinv(G) → TeG X → Xe, is a vector space isomorphism. Thus the vector space TeG inherits the vector field bracket, in such a way one obtains the Lie algebra g of G. The previous construction can be generalised to Lie algebroids as follows. Let G be a Lie groupoid, notice that Rg is not a map from G to G but a map between fibers: −1 −1 Rg : s ({t(g)}) → s ({s(g)}) −1 −1 (right translation between fibers) (DRg)h : Ths ({t(g)}) → Thgs ({s(g)}). Therefore, we should consider vector fields tangent to the source fibers. Since s : G → M is a submersion, the level set theorem guarantees s−1({x}) is a embedded submanifold with s −1 tangent space Ty G := T s ({x}) = ker Dys. Thus, the space of vector fields tangent to s the fiber is X (G) := Γg(ker Ds), notice Ds is a fiber bundle because s has full rank ([24, Theorem 5.1]). We can define the set of invariant vector field to the fibers as follows s s Xinv(G) = {X ∈ X (G)|(DRg)Xh = Xgh}. s Proposition 2.2.14. Xinv(G) is closed under the vector field bracket, therefore the set of right invariant vector fields has a Lie algebra structure. Proof. Let X,Y ∈ Xinv(G), and let X|s−1(x) and Y |s−1(s) be the restrictions of the vector fields to the fiber s−1(x). It is clear Xs(G) defines a involutive distribution since it is the associated distribution of the foliation by fibers, thus [X|s−1(x),Y |s−1(x)] = [X,Y ]|s−1(x). Right invariance implies X|s−1(x) and Y |s−1(x) are Rg-related to X|s−1(xg) and Y |s−1(xg), re- spectively. Thus [X,Y ]|s−1(x) is Rg-related to [X,Y ]|s−1(xg)([24, proposition 8.2]), implying [X,Y ] ∈ Xs is right invariant. s In a similar way to the Lie group case, Xinv(G) is determined by its value at the identity. However, in the Lie groupoid case there are many identities, one for each point in M. In some sense we get a smooth collection of vector spaces one for every point in M, that is, a vector bundle. Let us construct an isomorphism analogous to the one of the Lie group case. For this purpose consider the restriction of T sG to M, namely the vector bundle A(G) = 1∗T sG → M, the fibered product of T sG → M over 1 : M → G, and a map s Γ(A) → Xinv(G) given by ∼ s R R Γ(A) −→ Xinv(G); α 7→ αG , αG (g) = (DRg)1t(g) (α(t(g))). 12 s Proposition 2.2.15. The map Γ(A) → Xinv(G) is a vector space isomorphism. Addition- ally, the bracket induced in Γ(A) from the vector field bracket satisfies the Leibinz rule [α, fβ] = f[α, β] + (Dt(α)(f))β , for all α, β ∈ Γ(A), f ∈ C∞(M). Proof. Linearity of the maps follows from linearity of the differential. To prove it is an isomorphism we show an explicit inverse Xinv(G) → Γ(A); X 7→ X|M , where X|M denotes the restriction of X to M. It is clear this map is linear. Moreover since R R1m |s−1(m) is the identity then (αG|M )(m) = α(DR1m )m(α(m)) = α(m). On the other hand, right invariance implies a vector field X ∈ Xinv(G) is completely determined by the values at some ”cross-section” of s, for instance is completely determined by the values at R M (seen as an embedded manifold of G). Thus (X|M )G = X, proving the given map is indeed an inverse. ∞ To prove the second statement we need to extend f ∈ C (M) to Rg-invariant function f˜ ∈ C∞ (G). Since the vector field bracket in X (G) satisfies the Leibinz rule the desired Rg inv conclusion will not depend of the extension. Consider the extension f˜ = f ◦ t, this is Rg-invariant map since f ◦ t(1m · g) = f(t(1m · g)) = f(t(1m)) = m. Now, R ˜ R ˜ R R R ˜ R R ˜ [α, fβ] = [αG , fβG ]|M = (f[αG , βG ] + (αG f)βG )|M = f[α, β] + (αG f)|M β. R ˜ To finish the proof we need to show (αG f)|M = Dt(α)(f). Indeed, for any m ∈ M, R ˜ [αG f](m) = [(DRg)(α)(f ◦ t)](m) = [α(m)](f ◦ t ◦ Rg) = [α(m)](f ◦ t) = [Dt(α)(f)](m), so the desired equality holds. Definition 2.2.16. The Lie algebroid of a Lie groupoid G, denoted by A(G), is the ∗ s vector bundle Ds : 1 T G → M with the bracket induced from the vector fields Xinv(G) and anchor Dt : 1∗T sG → TM. Example 2.2.17. Pair Groupoid: For the pair groupoid M × M the source fibers are s−1(x) = {(x, y) ∈ M × M|y ∈ M} which are diffeomorphic to the manifold M. Moreover Dt : T s(M × M) → TM defines an isomorphisms of vector bundles. We will show Dt : A(M × M) → TM is an isomorphisms of Lie algebroids. It is clear the anchor is preserved by the this isomorphism, indeed #TM ◦ Dt = idTM ◦ Dt = #A(M×M). Now, a vector field s on T (M × M) is by definition a collection of vector fields {Xy}y∈M , where Xy ∈ X(M), such a collection defines a right invariant vector field on T s(M × M) if and only if all the Xy are the same. From this description is clear that the bracket of invariant vector fields Xinv(M × M) is isomorphic to the bracket on X (M). Notice that in the case M is simply connected M × M ⇒ M is a s-connected Lie groupoid integrating TM → M. 13 Example 2.2.18. Lie Algebras: For a Lie group G considered a a groupoid over a point {∗}, the contruction of the lie algebroid yield the lie algebra g → {∗} considered as a vector bundle over a point. 2.3 Integration Theory of Lie Algebroids We described how to any Lie groupoid one can assign a Lie algebroid generalising the construction of the Lie algebra of a Lie group. It is natural to ask under which conditions, the Lie algebroid data integrates to groupoid data, that is, if the classical Lie theorems hold for Lie algebroids/groupoids. There are affirmative answers for generalisation of the first two Lie theorems. Definition 2.3.1. Let be A be a Lie algebroid, then A is integrable if there exists a Lie groupoid G such that A is the Lie algebroid associated to G. A Poisson manifold P is called integrable if the algebroid T ∗P is integrable. Theorem 2.3.2. (Lie I: Proposition 3.3 of [25]) If A is an integrable algebroid, then there exist a unique s-simply connected Lie groupoid integrating A. Theorem 2.3.3. (Lie II: Proposition 3.5 of [25]) Let φ : A → B be a morphism of integrable Lie algebroids, and G and H be the corresponding integrations. If G is s-simply connected, then there exists a unique morphism of Lie groupoids Φ : G → H integrating φ. However, there is a negative answer for the third Lie theorem, i.e. not every Lie algebroid integrates to a Lie groupoid. In this section we will briefly review the integration theory of Lie algebroids, the obstructions to integrability and the construction of the source simply connected Lie groupoid in the integrable case [26]. S-Connected Lie Groupoid of G To begin our discussion on the integrability theory of Lie groupoids we begin by reviewing the construction of the s-connected Lie groupoid of a given groupoid G. The space of G-paths is the space 2 P (G) = {γ : I → G|g ∈ C , g(0) = 1x, s(γ(t)) = x, }. The space P (G) endowed with with the C2-topology is a Banach (sub)manifold with tangent 2 s space at γ ∈ P (A) given by the Banach space Cγ (I,T G) of time dependent (fiber) vector fields over γ ([27] Theorem 1.11 and Proposition 1.13). The Monodromy groupoid Mon(G) of G is Mon(G) = P (G)/ ∼; γ ∼ η, where ∼ is the equivalence of C1-homotopies of path with the same ending and starting points. The groupoid structure is defined by s([γ]) = γ(0), t([γ]) = γ(1), ( γ(2t) t ∈ [0, 1/2] [η][γ](t) = η(2t − 1)η(1) t ∈ [1/2, 1]. The differential structure of Mon(G) is given in a similar fashion as the differential structure universal covering of a manifold [28]. In the case of a Lie group G the space 14 Mon(G) can be described in terms of homotopies of paths with values on the Lie algebra g, i.e. Mon(G) = Mon(g). Thu,s by starting with the Lie algebra one can describe Mon(g), at least as a topological space, in terms of the Lie algebra alone. One can then use some argument to endow Mon(g) with a differentiable structure coming from the Banach manifold structure on P (g) and show it is actually the s-connected groupoid integrating G ([29, Section 1.14]). The Lie algebroid case follows a similar idea, to begin we require to find the appropriate infinitesimal notion of G-path, this will be given by A-paths; in addition we require some notion of homotopy for A-paths. With these definitions one can define a topological groupoid G(A) of an algebroid A, an analogue of Mon(g). At last, one needs to find necessary and sufficient conditions on A such that G(A) has a differentiable structure, this will be done by considering G(A) as the leaf space of a foliation on a Banach manifold [26]. Definition 2.3.4. Let π : A → M be an Lie algebroid. A C1-curve a : I → A is an A-path if d #a(t) = γ(t), dt where π(a(t)) = γ is the base path. The space of A-paths is denoted by P (A). The following proposition shows that P (A) should be thought of as the algebroid ana- logue of P (G). Proposition 2.3.5. Let A be an integrable algebroid and G a integration of A. Then the map R R d D : P (G) → P (A); D γ(t) = (DR −1 ) γ(t), γ(t) γ(t) dt defines a homeomorphism. Proof. We will prove G-paths correspond to the groupoid morphism I × I → G, where I × I is the pair groupoid, and that A-paths corresponds to algebroid morphism TI → A. Since the map DR is just the derivation of morphism defined by the G-path, and since I × I is the simply connected groupoid integrating TI, then Lie II gives the desired isomorphism. To check a G-path is a groupoid morphism I × I → G, Let H : I × I → G be a groupoid morphism with base h : I → M, then H(t, ) = H(t, 0)H(0, ) = H(t, 0)H(, 0)−1 thus H is characterized by a path γ(t) = H(t, 0). Note s(H(t, 0)) = h(0) and γ(0) = H(0, 0) = 1h(0), that is γ is a G-path. To check an A-path is an algebroid morphism TI → A. Since TI has trivial bracket we only need to show the compatibility with the anchor. Since TI is trivializable a vector bundle morphism with base π(a(t)) = γ : I → A can be written as adt : TI → A with d a : I → A. Compatibility with the anchor is equivalent to the equality #a = dt(γ) = dt γ, i.e. the condition for a to be an A-path. The Weinstein Groupoid of an Lie Algebroid A-paths should be seen as infinitesimal analogues of G-paths. With this in mind let’s see how can we restate the notion of homotopy using just the algebroid data. 15 By a variation of A-paths we mean a family of A-paths a(t) = a(, t): I × I → A which is of class C2 on the parameter , and with the property that the base maps γ(t): I → M have fixed end-points. In the case A is integrable a variation of A-path corresponds to a family of G-paths g(t): I → G such that D(g) = a. In the groupoid scenario a homotopy between G-paths g0, g1 is a map H : I → P (G) with fixed starting and end-points, forgetting the differentiable structure for a second, one verifies that this is the same as a P (G)-path. Thus, heuristically, a homotopy of G-paths can be considered as a groupoid morphism (I × I) × (I × I) → G, where again (I × I) is the pair groupoid, while the outer product (I × I) × (I × I) carries a component-wise groupoid structure. In this picture, an A-path homotopy should be some subset of algebroid morphism h : TI × TI → A; h = adt + bd a, b : I × I → A. Now from the G-path picture b = DR(gt), where gt is a path in G defined by gt() = g(, t)([26, Proposition 1.3]). Thus, the condition that the end point are fixed can be restated as b(, 1) = 0. One can describe the conclusion of the previous discussion without the need of a groupoid G integrating A. If a : I × I → A is a variation of A-paths then the condition for adt + b : TI × TI → A to be an algebroid morphism for some b : I × I → A can be rewritten, by using an auxiliary TM-connection ∇ on A, as a differential equation on a and b. Proposition 2.3.6. Given a TM-connection in A and a : I × I → A an A-path variation, the condition for adt + bd to be a algebroid morphism is equivalent to the differential equation ∇∂t b − ∇∂ a = T∇(a, b), b(, 0), where T∇ denotes the torsion of the connection (see 2.2.10). The solution b of this equation always exist and is an A-path which does not depend on ∆ Proof. The first part of the proposition is proposition 1.14 in [30]. The second part is proposition 1.3 in [26]. We summarise the previous discussion in the following definition of A-path homotopy. Definition 2.3.7. Two A-paths a0, a1 : I → A are A-homotopic a0 ∼A a1, if there exist a variation a : I × I → A with the property that b given by proposition 2.3.6 satisfies b(, 1) = 0. The A-homotopy relation is an equivalence relation and one can consider the Weinstein groupoid G(A) ⇒ M of a Lie algebroid π : A → M defined by P (A)/ ∼A. This is a topological groupoid over M with the following structure maps: s([a]) = π(a(0)), t([a]) = π(a(1)), ( 2a2(2t) t ∈ [0, 1/2] [a1][a2] = 2a1(2t − 1) t ∈ [1/2, 1]. 16 Theorem 2.3.8. (Theorem 2.1 of [26]) Let A be Lie algebroid over M, then the We- instein groupoid G(A) is a s-simply connected topological groupoid. Moreover, whenever A is integrable, G(A) admits a differentiable structure which makes it the unique s-simply connected groupoid integrating A. The previous theorem is an analogue of Lie third theorem in the context of Lie algebroids and Lie groupoids. NOtice the Weinstein groupoid is always a topological groupoids, how- ever it may fail to be a LIe groupoid, In the following section we will explore the obstructions to endow the Weinstein Groupoid witha differentiable structure. Obstructions to Integrability: Monodromy We now come to the issue of finding obstructions to endow G(A) with a differential structure. The answer to this question is given by the monodromy groups which we will describe in this section. Let A → M be a Lie algebroid and let gx be the isotropy Lie algebra at x ∈ M, i.e. consider G(gx) the Lie group integrating gx, which by construction is subset of G(A)x, where the latter is the isotropy group of G(A) at x. Definition 2.3.9. The monodromy group Nx(A) at x ∈ M is the subgroup of G(gx) which consists of the equivalence classes [a] ∈ G(gx) of gx-paths that are trivial as A-paths. In practice there is a practical way to compute the monodromy groups Nx(A). By proposition 2.2.2 for every x ∈ M there exist leaves Lx such that algebroid A → M restricted to Lx is a Lie algebra bundle # 0 → gx → A|L −→ TL → 0. Choosing a linear splitting of the previous exact sequence δ : TL → AL and regarding it as 2 a connection one verifies its curvature Ω (L, gL) is given by Ωδ(X,Y ) := δ([X,Y ]) − [δ(X), δ(Y )]. Lemma 2.3.10. ([26, Lemma 3.6] ) If there is a splitting δ with the property that its curvature Ωδ is Z(gx)-valued, then Z ∗ 2 Nx(A) = γ Ωδ | (γ : S → L) ∈ π2(L, x) , S2 for all x ∈ L Remark 2.3.11. The integral is understood as a Z(gx)-valued integral. In general, even if Ωδ is not Z(gx)-valued the set of integrals defined in the proposition is a group P(A, x) containing Nx(A). Theorem 2.3.12. (Theorem 4.1 of [26]) A Lie algebroid A over M is integrable if and only if for all x ∈ M the groups Nx(A) are locally uniformly discrete in x ∈ M. Example 2.3.13. Consider a simply connected symplectic manifold (M, ω). Define the algebroid Aω = TM ⊕ω R as the algebroid A extended by a trivial line bundle with structure maps: #ω(a ⊕ λ) = #a, [X ⊕ f, Y ⊕ g]ω = [X,Y ] ⊕ LX g − LY f − ω(X,Y ). 17 This algebroid is transitive i.e. the anchor is surjective at every x ∈ M and there only one leaf, so the only obstruction to integrability is measured by whenever Nx(Aω) is discrete. To measure the obstruction we use lemma 2.3.10, consider the splitting θ(X) = X ⊗ 0, then Ωθ(X,Y ) = θ([X,Y ]) − [θ(X), θ(Y )] = [X,Y ] ⊕ 0 − [X,Y ] ⊕ (LX 0 − LY 0 − ω(X,Y )) = ω(X,Y ). Thus Aω is integrable if and only if the group of spherical periods defined as Z ∗ 2 P(M) = { γ ω | (γ : S → M) ∈ π2(M)}, S2 is discrete. Interesting enough this is the same condition (up to some scalar multiple) for the symplectic manifold (M, ω) to be prequantizable (see the Chapter 3). By the analysis in the previous example, in case of a Poisson manifold P the condition i) of 2.3.12 is equivalent to requiring that the group of periods P(L, x) of every symplectic leaf L of P to be discrete, i.e. every leaf symplectic leaf should be prequantizable. However due to condition ii) this is not a sufficient condition, examples of algebroids which fail to be integrable due to condition ii) can be found in [20]. We saw how when Lie algebroid T ∗P of a Poisson manifold P encoded much of the information of P . In the case this algebroid is integrable, the groupoid G(T ∗P ) will have a symplectic structure compatible with the groupoid multiplication in such a way that G(T ∗P ) will represent a symplectification of P . 2.4 Symplectic Groupoids For a groupoid G the space of composable pairs G(2) = {(g, h) ∈ G × G|t(h) = s(g)} has three “face” maps, pr1, pr2, m : G2 → G, pr1(g, h) = g, pr2(g, h) = h, m(g, h) = gh. These face maps are actually the face maps of the groupoid seen as a simplicial set by means of its nerve[19]. Definition 2.4.1. A differential form ω on G ⇒ M is called multiplicative if ∗ ∗ ∗ pr1ω − m ω + pr2ω = 0. Definition 2.4.2. A symplectic groupoid is a pair (G, ω) such that G is a Lie groupoid and ω ∈ Ω2(G) is a multiplicative symplectic form. Remark 2.4.3. Much of the theory to be developed do not require ω to be non-degenerate. A multiplicative closed 2-form will be called multiplicative pre-symplectic form, in this scenario a pair (G, ω) will be called a pre-symplectic groupoid. Lemma 2.4.4. A symplectic form ω in a groupoid G is multiplicative if and only if the graph of the multiplication (2) Γm = {(g, h, gh) ∈ G × G × G|(g, h) ∈ G }, 18 is a isotropic submanifold of G × G × G¯, where G¯ denotes the symplectic manifold (G, −ω). Proof. The map (2) G → Γm, (g, h) 7→ (g, h, gh), ∗ ∗ ∗ is a diffeormorphism such that ω × ω × ω¯ pulls back to pr1ω + pr2ω − m ω. Then Γm is ∗ ∗ ∗ isotropic if and only if pr1ω + pr2ω − m ω vanishes. Lemma 2.4.5. Let (G ⇒ M, ω) be a symplectic groupoid. Then: i) 1(M) is an isotropic submanifold, ii) Inversion is an anti-symplectomorphism iii) The s-fibers and t-fibers are symplectically orthogonal, i.e. (ker Ds)ω ⊂ (ker Dt) Proof. The proof of the 3 statements is very similar, we will only prove iii) and sketch the proof of the other assertions: −1 −1 −1 iii) Let vs and wt be tangent vector of s (x) and t (x), at some point g ∈ s (x). −1 Let γ be a curve on s (x) with tangent vector vs since (γ(t), 1x, γ(t)) ∈ Γm it follows −1 (vs, 0, vs) ∈ T Γm. In the same fashion considering a curve η ∈ t (x) with tangent vector wt one concludes (0, wt, wt) ∈ T Γm. Multiplicativity implies ω(0, vs) + ω(wt, 0) − ω(vs, wt) = 0, thus ω(vs, wt) = 0. i) Consider the curves γ, η in 1(M) with tangent vectors v, w ∈ Tm(1(M)). Then (γ, γ, γ), (η, η, η) ∈ Γm, infinitesimally multiplicativity yields the desired result. ii) Consider the curve γ() with tangent vector v at x ∈ M. Then (γ(), γ()−1, t(γ())) ∈ −1 Γm and (γ() , γ(), t(γ())) ∈ Γm, the infinitesimal version together with isotropy of Γm yields the result. It is clear that M, seen as a submanifold of G by the embedding 1 : M → G, and the source fiber s−1(x) are tranversal submanifolds of G implying −1 ∼ TxG = TxM ⊕ Txs (x) = TxM ⊕ Ax, for every x ∈ M. Thus T G =∼ TM ⊕ A and ω induces an isomorphism ω˜ : T G =∼ TM ⊕ A → T ∗G =∼ T ∗M ⊕ A∗. Since M is an isotropic submanifold, i.e. [˜ω(X)]|TM ≡ 0 for every X ∈ TM, it followsω ˜ restricts to a map ∗ R ω˜|TM : TM → A ; [˜ω(X)](α) = ω(X, αG ). In a similar way, since target fibers are also tranversal to M it follows TxG = TxM ⊕ −1 Txt (x). By 2.4.5 iii), source fibers and target fibers are symplectically orthogonal, i.e. 19 ω˜(α)|A ≡ 0 for every α ∈ A, thusω ˜ restricts to a map ∗ ω˜|A : Ax → T M. Lemma 2.4.6. The maps ∗ R ω˜|TM : TM → A ; [˜ω(X)](α) = ω(X, αG ), ∗ R ω˜|A : Am → T M; [˜ω(α)](X) = ω(αG ,X). are pointwise linear isomorphism. Proof. Consider v ∈ TM such thatω ˜|TM (v) = 0. Again since M is isotropic, it implies that ∗ ω˜(v) ≡ 0 on T G. Then non-degeneracy of ω implies v = 0, i.e.ω ˜|TM is injetive. By a similar argumentω ˜|A is injective. The fact thatω ˜ is an isomorphism from T G to T G together with dimension counting implies both maps are isomorphisms. Corollary 2.4.7. Let (G ⇒ M, ω) be a symplectic groupoid, then dim G = 2 dim M. With this results we prove the equivalence of the given definition of symplectic groupoids and another characterization of symplectic groupoids, historically the original one, which is useful in some occasions. Corollary 2.4.8. A symplectic form ω in a Lie groupoid G is multiplicative if and only if the graph of the multiplication Γm is a Langrangian submanifold of G × G × G¯. Proof. We already proved Γ is isotropic, to show Γm is Lagrangian we count dimensions. Recall G(2) is a pulback manifold over M, thus dim Γm = dim(G × G) − dim(M) = 2 dim(G) − dim(M) = 3 dim(M) Implying dim(G × G × G) = 2 dim(Γm). Corollary 2.4.9. The s-fibers and t-fibers are each other symplectic complement, i.e. (ker Ds)ω = (ker Dt) Proof. We already proved (ker Ds)ω ⊂ (ker Dt). Since s−1(x) =∼ t−1(x) via the inversion map, it follows they have the same dimension 1/2 dim(G). Counting dimensions dim(ker Ds)ω = 1/2 dim(G) = dim(ker Dt), it follows they are equal. Symplectic Groupoids and Poisson Manifolds The main property of symplectic groupoids is their close relationship with Poisson manifolds. We will state and prove the theorems that concretely establish the relationship. Theorem 2.4.10. Let (G ⇒ M, ω) be a symplectic groupoid. Then there exists a unique Poisson structure on M such that s : G → M is a Poisson map and t : G → M is an anti-Poisson map 20 Proof. Since s : G → M is a submersion, the pullback s∗ : C∞(M) → C∞(G); f 7→ f ◦ s, is an injective map. We will show s∗(C∞(M)) is invariant under the Poisson bracket on C∞(G), thus the restriction defines a Poisson bracket by ∗ −1 ∗ ∗ {f1, f2}M = (s ) {s f, s f}G. By definition the defined bracket makes s into a Poisson map and is clearly unique. We will call a function f ∈ ι∗(C∞(M)) basic. It is clear that basic fuctions are (locally) constant on fibers, thus its differential lie in the annihilator of the tangent space to the fibers i.e. df ∈ Ann(Ds). Now ω df ∈ Ann(ker Ds) ⇔ Xf ∈ (ker Ds) , where Xf is the Hamiltonian vector field of f. Thus lemma 2.4.5 implies f is basic ⇔ Xf ∈ ker Dt. Since ker Dt defines a involutive distribution (it is integrable) then we conclude that ∗ ∞ [Xs∗f ,Ys∗g] = s h for some h ∈ C (M). This fact together with the equality [Xf ,Xg] = X{f,g}G completes the proof. Thus, a symplectic groupoid G ⇒ M induces a Poisson structure on M such that the algebroid A(G) =∼ A(T ∗M). On the other hand, if one start with Poisson manifold P one can define a Lie Algebroid T ∗P , if the Poisson manifold is integrable we will see G(T ∗M) is canonnically a symplectic groupoid. Therefore there is a one to one correspondence between integrable Poisson manifolds and s-connected symplectic groupoids. To motivate this result we need to recall the construction of G(T ∗M): By definition ∗ ∗ ∗ G(T M) = P (T M)/ ∼T ∗M . We will endow P (T M) with a symplectic structure. Recall P (T ∗M) is Banach manifold with tangent space at a given by time dependent vector fields ∗ ∗ ∗ over a TaP (T M) = {U : I → T (T M)|U(t) ∈ Ta(t)(T M)}[27]. If ω0 is the canonical symplectic form on T ∗M a symplectic form on P (T ∗M) can be defined by Z 1 ω(U1,U2)a = ω0(U1(t),U2(t))dt. 0 Then, one can show that the equivalence relation of A-paths is actually given by symplectic reduction procedure[31], thus G(T ∗M) has a canonical symplectic structure. Details of this construction can be found in [21, 32], where the following proposition is proved (Theorem 4.6]): Theorem 2.4.11. Let M be a integrable Poisson manifold and let Σ(M) be the s-simply connected groupoid integrating M. Then Σ(M) admits a canonical multiplicative symplec- tic form ω, so that (Σ(M), ω) is a symplectic groupoid. Moreover, the Poisson bracket induced on M by Σ(M) is the original one. 21 Examples The following is a general construction for a general class of examples. Let G ⇒ M be a groupoid and let A → M be its associated algebroid. Define the cotangent groupoid ∗ ∗ T G ⇒ A with the the following structure maps: The source map is defined by pulling G s back the one-forms on T to the algebroid A = T G|M i.e. [s(α)](Y ) = α(DRg(Y )), [t(α)](Y ) = α(DLg(X − D1(#X))), ∗ ∗ ∗ where α ∈ Tg G, X ∈ At(g) and Y ∈ As(g). For α ∈ Tg G and β ∈ Th G the multiplication is given by α • β(T m(η, ξ)) = α(η) + β(ξ), where η ∈ TgG, ξ ∈ ThG and T m : TgG ×ThG → TghG is the differential of the multiplication ∗ map. At last the multiplicative symplectic form on T G is ωcan the canonical symplectic form on a cotangent bundle. The details concerning that this is indeed a symplectic groupoid can be found in [33]. Remark 2.4.12. In the previous section we saw a symplectic groupoid induces a Poisson structure on its base Theorem 2.4.10. In the example T ∗G → A∗ this Poisson structure can be described as follows: Let A → M be a Lie algebroid. Then A∗ (dual as a vector bundle) is a Poisson Manifold. Indeed consider the subsets Γ(b A) ⊂ C∞(A∗) and Cb∞(M) ⊂ C∞(A∗) given by, ∗ Γ(A) = {sˆ : A → R|sˆ(a) = aπ(a)(s ◦ π(a)) for s ∈ Γ(A)}, C∞(M) = {fˆ : A∗ → R|fˆ = π∗f for f ∈ C∞(M)}. One can define a Poisson bracket by, {f,ˆ gˆ} = 0 {s,ˆ tˆ} = [ds, t] {s,ˆ fˆ} = ρ\(s)f. It is worth noting that, although we motivated this Poisson structure by theorem 2.4.10, it is independent of the existence of groupoid and actually can be defined for any Lie algebroid A → M (c.f. [33]). In Chapter 4 we will focus in the geometric quantization of this family of Poisson struc- tures, that is, integrable Poisson structures for which the integrating symplectic groupoid is a cotangent groupoid. Some particular examples of the previous construction are: Example 2.4.13. Trivial Poisson Bracket: Let M be a simply connected manifold. Consider M → M as a lie algebroid with zero dimensional fibers, this has integrable ∗ groupoid M ⇒ M. Now the cotangent groupoid is T M ⇒ M where s = t = π and the multiplication is given by addition on fibers. Moreover, the Poisson structure on M is the trivial Poisson structure. Example 2.4.14. Cotangent Bundles: Let N be a simply comnnected manifold. Con- sider the algebroid TN → N, this algebroid has integrable groupoid the pair groupoid ∗ ∗ ∗ N × N ⇒ N. Now the cotangent groupoid is T N × T N ⇒ T N with the pair groupoid structure. Moreover the Poisson structure on T ∗N is the one given by the symplectic structure on T ∗N. Example 2.4.15. Lie Algebras: Consider the algebroid g → {∗}, this algebroid has integrable groupoid G → {∗}, where G is the simply connected groupoid integrating g. The 22 ∗ ∗ contangent groupoid in this example T G ⇒ g is the groupoid of example 2.1.13 with its the natural symplectic form. Moreover, the Poisson structure on g∗ is the Kirillov-Poisson structure of g∗. 23 Chapter 3 Geometric Prequantization of Symplectic Groupoids The first step in the geometric quantization procedure is to construct a prequantization line bundle, or equivalently a principal S1-bundle over the symplectic manifold in consideration, together with a connection whose curvature is the symplectic form. The existence of such a bundle is equivalent to the symplectic form being integral, this is the prequantization condition for a symplectic manifold. If such a bundle exists it describes a Lie groupoid central extension of the pair groupoid of the symplectic manifold. In turn, such groupoid central extension describes an algebroid central extension. In the general description of prequantization of Lie algebroids and Lie groupoids one starts with the infinitesimal data given by an algebroid extension, and find conditions to in- tegrate such extension to a Lie groupoid extension. In the case such an algebroid extension is the one of the geometric quantization for symplectic manifolds, the integrability obstruc- tion is the same as the prequantization condition. Our main interest in this description will be when the considered algebroid A(P ) comes from a Poisson manifold P , if the Poisson manifold is integrable, then one would seek to prequantize the symplectic groupoid G(P ) integrating A(P ) by finding a suitable extension. 3.1 Prequantization of Symplectic Spaces Principal S1-Bundles We begin this section by introducing some of main objects in geometric quantization. Definition 3.1.1. A principal S1-bundle is a fiber bundle L → M together with a smooth (left) action S1 × L → L such that: i) The action of S1 preserves the fibers of L, i.e. S1 ·π−1(m) ⊂ π−1(m) for every m ∈ M, ii) S1 acts freely and transitively on fibers, and iii) L → M is locally trivial, i.e. there exists an open covering {Uα}α∈Λ together with a −1 1 collection of equivariant diffeomorphisms ηα : π (Uα) → Uα × S such that the following 24 diagram commutes −1 ηα π (Uα) Uα × G π pr1 Uα Definition 3.1.2. A base preserving S1-bundle isomorphism between (L → M) and (L0 → M) is a S1-equivariant diffeomorphism from L to L0. 1 Let L → M be a S bundle and {Uα, ηα}α∈Λ a trivialization for L → M, if Uβ is another trivializing open set such that Uα ∩ Uβ 6= ∅ we can consider the following diagram: η−1 1 β −1 ηα 1 (Uα ∩ Uβ) × S π (Uα ∩ Uβ) (Uα ∩ Uβ) × S pr 1 π pr1 Uα ∩ Uβ −1 1 1 The mapg ˜αβ := ηα ◦ηβ : Uα ∩Uβ ×S → Uα ∩Uβ ×S should preserve the first component, 0 1 thusg ˜αβ(m, λ) = (m, λ ). Moreover, since the action of S is transitive and free on fibers 1 0 there exist a unique gαβ(m) ∈ S such that λ = [gαβ(m)]λ. Therefore, the covering (Uα, ηα) 1 induces a set of transition functions gαβ : Uα ∩ Uβ → S which satisfy: i)gαα = 1, ii)gαβgβα = 1, iii)gαβgβγgγα = 1. (3.1) For a good covering {Uα}, a covering such that every intersection Uα1 ∩ · · · ∩ Uαn is contractible, the transition functions are actually an example of a element the first Chechˇ 1 1 1 1 cohomology Hˇ (M,S ). Actually for a good covering {Uα} an element in Hˇ (M,S ) can 1 be defined as a collection of functions cαβ : Uα ∩Uβ → S satisfying (3.1). For the definition of higher Chechˇ cohomology groups Hˆ •(M,A) of a manifold M with coefficients on an abelian group A, and some of its properties we refer to the reader to [34]. On the other ˇ 1 1 1 hand given a element [c] ∈ H (M,S ) one can find a S -bundle L[c] such that the transition functions defines the same element [c]. Fixing a good covering {Uα} the element [c] can be 1 represented by a collection of functions cαβ : Uα ∩Uβ → S satisfying (3.1), then the bundle L[c] can be constructed as a 1 L[c] = ( Uα × S )/ ∼, α∈ where (p, λ) ∼ (q, µ) if and only if p = q and [cαβ(p)] · λ = µ. The conditions imposed in (3.1) assure the quotient is a manifold with a principal S1-bundle structure. Thus we have sketched the proof pf Theorem 1.1.1 of [35]: 1 Proposition 3.1.3. Let M be a manifold and denote by PicS1 (M) the group of S -Bundles over M, known as the Picard group, then ∼ ˇ 1 1 PicS1 (M) = H (M,S ). Connections and Curvature on S1-Bundles For our purposes we will require more geometrical insight of S1-bundles, namely the notion of connection and curvature. Since the action of S1 on fibers is free and transitive it estab- 25 1 −1 lishes a diffeomorphism between S and the fiber Lm = π (m) for every fiber. Moreover 1 ∼ we have an identification of the tangent spaces TeS = R and TpV := ker dπ, given by the derivative of the action, explicitly d #: R → T V ; #(X, p) = (eiXt · p)| = X#. p dt t=0 p With this identification we have the following exact sequence of vector bundles, TV TL Dπ TM. In the above sequence the vertical space TV is naturally embedded in TL, however there is not a canonical way of choosing a “horizontal” complement for TV inside TL. Such a complement can be chosen with help of a connection form. Definition 3.1.4. A connection form θ on a principal S1-bundle L → M is a element of Ω1(L) such that: ∗ ∗ i) Rλθ = θ, where Rλ denotes the action of λ on L, # 1 ii) θ(X ) = X, for every X ∈ TeS . Proposition 3.1.5. The map # ◦ θ : TL → TV is a splitting of the exact sequence TV TL Dπ TM. Proof. We just need to check (# ◦ θ) ◦ d(ιp) = IdTV , where ιp is the inclusion of the fiber π−1(p) in L. Take some Y˜ ∈ TV represented by Y˜ = Y # with Y ∈ g, then # # # (# ◦ θ) ◦ d(ιp)(Y ) = #[θ(Y )] = #(Y ) = Y . Now, one important use of the identification TL = TV ⊕ TM is that it lets us lift (uniquely) vector fields X ∈ X(M) to vector fields H(X) ∈ X(L) by the requirement that the vertical component of H(X) be zero. Proposition 3.1.6. Let θ be a connection form and X ∈ X(M) then there exists a unique vector field H(X) ∈ X(L) characterised by the following properties properties: i) dπ(H(X)) = X, ii) θ(H(X)) = 0. This vector field is called the horizontal lift of X by θ. Proof. For the existence, consider some Y ∈ L such that dπ(Y ) = X (such a Y exists because π is a submersion). One checks by direct computation Y − θ(Y )# satisfies the given conditions. To prove it is unique notice that, by the splitting TL = TV ⊕ TH, a vector Y ∈ TL that vanishing both on θ and dπ should be zero. Since the difference of two vectors H1(X), H2(H) satisfying the previous conditions vanishes both under θ and under dπ, it follows H1(X) = H2(H). 26 Definition 3.1.7. Let L → M be a principal S1-bundle and θ ∈ Ω(L, g) be a connection one form on L. The curvature Ω of θ is element of Ω2(L) defined by Ω = dθ. Having introduced our basic objects of study we are ready start our discussion on pre- quantization. Prequantum Bundles and Prequantum Hilbert Spaces The idea of geometric quantization (at least in its beginning) is to construct a Hilbert space together with a representation of the Poisson algebra of smooth functions on a symplectic manifold on this Hilbert space. The Hilbert space will turn out to be some completition of a space of the sections of a line bundle. The first step in a geometric quantization procedure is to construct such a prequantization line bundle, or equivalently a principal S1-bundle, with a connection whose curvature is the symplectic form. In this section we review facts about the existence and uniqueness of such a bundle. Definition 3.1.8. A symplectic manifold (M, ω) is prequantizable if there exists a princi- pal S1-bundle L −→π M with connection form θ such that dθ = π∗ω. Such a pair (L → M, θ) is called a prequantum bundle. Example 3.1.9. Let M be a manifold and T ∗M its the cotangent bundle. The cotangent bundle is always a symplectic manifold with symplectic form ω = dθ0, where θ0 is the Liouville form. Now, (T ∗M, ω) is prequantizable by a trivial prequantum bundle (S1 × ∗ π 1 T M −→ M, dθ0 + πS1 ∗ θL), where θL is the Maurer-Cartan form on S defined by θL(Y ) = 1 DRλ−1 (Y ) for Y ∈ TλS . The above example enlightens the role of the connection. In a physical system the cotangent bundle T ∗M is the phase space for a configuration space M, the coordinates on this space can be classified on position coordinates qi, given by coordinates in M, and ∗ (local) momentum coordinates pj, given by coordinates along the fibers of T M. The i Liouville form with respect to those coordinates is θL = pidq , thus the Liouville form can be used to differentiate along position coordinates. In the general case of a symplectic manifold the connection one form plays the role of the Liouville form, that is, a first approach of choosing how to differentiate along the position coordinates. Back to our main topic, given a prequantum bundle (L → M, θ) for (M, ω) we can consider the vector space of smooth S1-invariant functions ∞ S1 C ¯ C (L)0 := {h : L → |f(λ · p) = λf(p) and π(Suppf) is compact.}. This set will be called the space of compactly supported sections and will be denoted by Γ(L). It is called the space of sections because it is isomorphic to the space of section of the associated vector bundle to L[4]. We will study further the relationship between principal bundle and associated line bundles later in the chapter. Since the pullback π∗f of a function f ∈ C∞(M) is S1 invariant it follows that (π∗f)·h ∈ ∞ G ∞ G ∞ ∞ G C (L)0 , i.e. the space C (L)0 is naturally a C (M)-module. We can endow C (L)0 27 with an inner product, Z hϕ, ψi = ϕ(m)ψ(m)ωm. M Remark 3.1.10. Notice ϕ(λ · p)ψ(λ · p) = λλϕ¯ (p)ψ(p) = ϕ(p)ψ(p), thus ϕ(·)ψ(·) defines a function on M, which can be integrated by using the symplectic volume form ωn = ω∧· · ·∧ω. ∞ G The completion of the inner product space (C (L)0 , h·, ·i) is the desired prequantum Hilbert space Hω. Although not entirely important for the future discussions, for sake of completeness, we turn to the task of constructing a representation of the Poisson algebra ∞ ∞ C (M), i.e. a linear map Q : C (M) → End(H), such that [Qf ,Qg] = Q{f,g}. Define ∞ Q : C (M) → End(Hω) Q(f)(ϕ) = H(Xf )ϕ − f · ϕ, where Xf is the Hamiltonian vector field of f. ∞ Proposition 3.1.11. (Lemma 3.9 of [36]) The assignment Q : C (M) → End(Hω) is a morphism of Lie algebras. Example 3.1.12. In example 3.1.9 we saw that T ∗R was prequantizable by a trivial bundle 1 ∗ π ∗ ∞ ∗ (S × T R −→ M, dθ0 + π θL). Thus the space of sections is Γ(L) = C (T R, C) and the inner product is the usual inner product of C-valued functions. We conclude the prequantum Hilbert space is L2(R2). Prequantum Line Bundles: Existence and Uniqueness In the previous section we constructed a Hilbert space together with a representation of the Poisson algebra C∞(M) of a prequantizable symplectic manifold (M, ω). In this section we will deal with the problem of existence and uniqueness of prequantum S1-bundles. To begin with we will introduce the classification problem of S1-bundles over a fixed manifold M. By the discussion in section 3.1, S1-bundles over a manifold M are charac- terized by their transition functions, i.e. by an element in the first Cechˇ cohomolgy group exp Hˇ 1(M,S1). The exact sequence of groups 2πiZ → R −−→ S1, induces a long exact sequence in Chechˇ cohomology, · · · → Hˇ 1(M, C) → Hˇ 1(M,S1) → Hˇ 2(M, 2πiZ) → Hˇ 2(M, C) → Hˇ 2(M,S1) ··· Moreover, since H•(M, C) admits partition of unity it follows Hn(M, C) = 0 for n > 0 [34]. These two facts imply: Proposition 3.1.13. Let M be a manifold, then Hˆ 1(M,S1) =∼ Hˆ 2(M, Z). Given a connection on a S1-principal bundle, and considering the equivalence between the deRham cohomology and the real Chechˇ cohomology H2dR(M) =∼ Hˇ (M, R), one can realize the isomorphism map as the chern class of the curvature. In the case of a pre- quantizable space this implies the necessary condition ω ∈ H2(M, 2πiZ), where the latter 28 cohomology group is the image in the deRham cohomology of Hˇ 2(M, 2πi) under the afore- mentioned isomorphism. A detailed discussion of the previous paragraph can be found on [34], an explicit construction of the objects and isomorphisms mentioned can be found in [4]. On the other hand, if ω ∈ H2(M, 2πiZ) then one can associate to it a S1-bundle π : L → M via the isomorphism of proposition 3.1.13. A theorem by Weil (Proposition 2.1.1 of [4]) asserts such a bundle can always be endowed with a connection θ such that dθ = π∗ω. Theorem 3.1.14. A symplectic manifold (M, ω) is prequatizable if and only if ω ∈ H2(M, 2πiZ). In practice, given the pairing between deRham cohomology and singular (co)homology given by Z n R n R n ∗ H (M, ) ⊗ HdR(M) → ; [(δ : ∆ → M), θ] 7→ δ θ, ∆n one can check ω ∈ H2(M, 2πiZ) if and only if the integral above has values in 2πZ([37, Theorem 18.14]). Then the theorem 3.1.14 can be restated: Proposition 3.1.15. A is prequatizable if and only if P(M) ⊂ 2πZ. This is, up to a constant, exactly the condition for the algebroid Aω of example §2.3.13 to be integrable. The previous theorem concludes the discussion concerning existence of prequantum bun- dles over a symplectic manifold (M, ω). The discussion of uniqueness is a bit more subtle, and goes as follows. We already saw that given the curvature ω ∈ HdR(M, 2πiZ) there is one S1-bundle, up to S1-bundle isomorphism, with the curvature ω. However if we want to distinguish between prequantization (L → M, θ) and (L0 → M, θ0) we also require to distinguish the S1-bundle up to connection preserving S1-bundle isomorphism, i.e. a bun- dle diffeomorphism τ : L → L0 such that τ ∗θ0 = θ. Given a loop γ : S1 → M, using the connection one can lift the vector field it defines to a vector field in L over γ, then the integral curve of this vector field defines a horizotal lift ofγ ˜θ : L → M. Sinceγ ˜(0) andγ ˜(1) belongs to the same fiber their difference is an element Q(γ) ∈ S1 called the holonomy of γ. It turns out this holonomy classifies S1-bundle with connection ([4, 1.12.3]). Now, for null-homotopic γ it holds Z Z Q(γ) = exp(−i θ) = exp(i ω) = 1, γ ∆ where ∆ is a surface with boundary γ. Since ω is integral then null-homotopic curves have fixed holonomy. We arrive to the conclusion: Proposition 3.1.16. Let (M, ω) be a simply connected symplectic manifold. Then there exists, up to connection preserving S1-bundle morphism, a unique prequantization (L → M, θ) of (M, ω). 29 The Kostant Picture of Prequantization We end this section with a remark about the vector bundle picture of prequantization, also known as the Kostant picture. What we have described so far is the principal bundle de- scription of prequantization, also known as the Souriau picture. There is another description making use of line bundles instead of S1-principal bundles, in this description the prequan- tization bundle is given by a Hermitian, line bundle with an compatible affine connection. A Hermitian line bundle (L → M, h·, ·i) is a one dimensional complex vector bundle with and fiberwise Hermitian inner product, and an affine connection is a TM-connection ∇ on L → M, such that X(hs, ti) = hs, ∇xt, i + h∇xs, ti, for all s, t ∈ γ(L) and X ∈ X (M). The two pictures are related by the relationship between principal bundles and associated bundles. Given an S1-bundle L → M and the natural representation χ : S1 → GL(C) = C×, λ 7→ λ, 1 one can compose the transition functions {gαβ : Uα → S } of L → M to obtain transition function {χ ◦ gαβ : Uα → C}. In the same way as described in section 3.1, these transition functions can be used to construct a fiber bundle with typical fiber C. The described bundle is known as the associated bundle A(L, C, χ) of L with respect to χ. This bundle can 1 also be described as the quotient L × C/ ∼, where (s, λ) ∼ (s · µ, χµ−1 λ) for s ∈ L, µ ∈ S , λ ∈ C. It is worth mentioning that in this description the set of sections of the line bundle is isomorphic to the space we called the sections of the prequantum S1-bundle. For a detailed discussion about the equivalence of objects in the principal bundle description and in the associated bundle description we refer to the reader to [4]. 3.2 The Groupoid and Algebroid Picture In this section we will generalise the previous discussion from prequantization of symplectic manifolds to prequantization of symplectic groupoids. By using the integration theory of Chapter 2, we will recover the prequantization uniqueness and existence conditions showing how prequantizability can be understood as an integrability problem. Together with the symplification of Poisson manifolds discussed in Chapter 2 this will allow us to generalise the prequantization procedure to (integrable) Poisson manifolds. Prequantum Symplectic Groupoid Extension We now describe how groupoids and algebroids appear in the picture. Suppose (L → M, θ) is the prequantization of a simply connected symplectic manifold M. We will show how the prequantum bundle induces a central Lie groupoid extension. 1 Definition 3.2.1. Let G ⇒ M be a groupoid and let S := S ×M endowed with a fiberwise groupoid structure. A central S1-Lie groupoid extension of G is a exact sequence of groupoids morphism over M, S −→ι G˜ −→Gπ , such that ι is an embedding, π a submersion and ι(γ)g = gι(γ) for every g ∈ G and γ ∈ S. 30 The sequence is exact in the sense that the kernel of the morphism π ker π = {g ∈ G|˜ π(g) = 1x for some x ∈ M} is SM It is worth noting that for every g ∈ ker π sG˜(g) = tG˜(g) = sG(π(g)) = x for some x ∈ M, therefore the kernel of a groupoid morphism has the structure of an Lie group bundle with base M. Proposition 3.2.2. Let S → G˜ −→Gπ , be a central extension of Lie groupoid then π : G˜ → G is a S1-bundle over G . 1 Proof. By denoting λm the element (λ, m) of S. One can define the action of S on G˜ by λ · g = ι(λt(g))g, where the latter is the multiplication in the groupoid G˜. This action preserves fibers since π(λ) = [π ◦ ι(λt(g))]π(g) = 1t(g)π(g) = π(g). 0 ˜ 0 0−1 0 −1 Furthermore, if g, g ∈ G satisfy π(g) = π(g ) then π(gg ) = π(g)π(g ) = 1t(g). Hence, 0−1 0−1 0 gg ∈ ker π and there exist some λt(g) such that gg = λt(g), i.e. g = λ · g . Since ι is injective such λ is unique if we fix the fiber. We conclude the action of G is free and transitive on fibers. Back to the prequantization picture, let (L → M, θ) be a prequantization of a simply connected symplectic manifold (M, ω). Associated to the S1-bundle (L → M) there is a gauge groupoid Gau(L) §2.1.12. Since L projects to M in an S1-invariant fashion there is a map to the pair grupoid, π × π : Gau(L) → M × M, π × π([p, q]) = (π(q), π(q)). This map is a groupoid morphism and its kernel is given by equivalence classes [p, q] such that π(p) = π(q). An equivalence class is uniquely determined by a π(p) ∈ M and a 1 λ[p,q] ∈ S such that p = λ[p,q] · q. For another pair of representatives it holds λ[t·p,t·q]t · q = t · p = tλ[p,q] · q = λ[p,q]t · q, then effectiveness of the action on the fibers implies λ[p,q] = λ[t·p,t·q], i.e. λ[p,q] is well defined. We conclude ker π =∼ S1 ×M, and that the embedding into Gau(L) is given by λ 7→ [λ·p, q]. Moreover, since [λ · p, p][p, q] = [λ · p, q] = [λ · p, λ · q][λ · q, q] = [p, q][λ · p, p] we have proved: Proposition 3.2.3. The sequence S −→ι Gau(P ) −−−→π×π M × M, is central S1-extension of Lie groupoids. The correspondence between S1-extensions of the pair groupoid and S1-bundles is even more profound, for instance the geometric information of the symplectic form ω and the connection θ can be incorporated in the previous S1-extension as multiplicative forms. 31 Proposition 3.2.4. There is a one to one correspondence between S1−bundles over M and S1 extensions of M × M, S −→ι Gau(L) −−−→π×π M × M. Moreover, given a S1 bundle L there are one to one correspondences between: i) Closed 2-forms ω on M and closed multiplicative 2 formsω ˜ on M × M ii) Connections on L and multiplicative 1-forms θ˜ on Gau(L), ∗ ∗ ˜ iii) the relation πP ω = dθ is equivalent to π ω˜ = dθ. Proof. In this case multiplicativity of 1-forms on Ω(M × M) can be rewritten as ωm[(X,Y )] = ωm[(X,Z)]. − ωm[(Y,Z)] For higher forms a similar relation holds component-wise. In what follows we will prove statement for 1-forms which generalize directly to k-forms. Recall s(x, y) = x and t(x, y) = y. Define a bijection between k-forms Ωk(M) and multi- plicative k-forms Ωm(M × M) by k ∗ ∗ Ω (M) 7→ Ωm(M × M), ω 7→ s ω − t ω k ∗ Ωm(M × M) 7→ Ω(M), ωm 7→ ιxωm, ∗ where ιx : M → M × M is defined by ιx(y) = (y, x) for x ∈ M. Notice that ιxωm(X) = ∗ ∗ ωm(X, 0) = ιyωm(X), so the above map does ιxω not depend on x. Now since (s ◦ ιx)∗(X,Y ) = (X, 0) and (t ◦ ιx)∗(X,Y ) = (Y, 0) it follows ∗ ∗ (s ◦ ιx) ωm[(X,Y )] − (t ◦ ιx) ωm[(X,Y )] = ωm[(X, 0)] − ωm[(Y, 0)], ∗ ∗ multiplicativity implies (s ◦ ιx) ωm − (t ◦ ιx) ωm = ωm. Likewise, since (s ◦ ιx)∗ = IdTM and (t ◦ ιx)∗ = 0, then ∗ ∗ ∗ ∗ ι (s ω − t ω) = (s ◦ ιx) ω = ω. Concluding the proof that the given maps establish a bijection. i) This is the previous discussion in the particular case of 2-forms. ii) Multiplicative 1-forms on L × L are in bijection with 1-forms on L. Moreover a one form θ ∈ Ω(L) is S1-invariant if and only if factors through the quotient to a one form θ˜ in Gau(L) = (L × L)/S1. iii) It is clear from the discussions in ii) and the given map between multiplicative forms on M × M and forms on M. One can generalize the previous picture to other symplectic groupoids leading to the general definition of prequantization. Definition 3.2.5. Let G ⇒ M be a groupoid and ω a multiplicative 2-form on G.A prequantization of (G, ω) is a central extension of Lie groupoids (R → G ⇒ M) together with a multiplicative 1-form which is a connection form for the principal bundle G˜ → G that satisfies π∗ω = dθ In the case of integrable Poisson manifolds one says one has a definition for prequanti- zability via its s-connected symplectic groupoid. 32 Definition 3.2.6. A integrable Poisson manifold is called prequantizable if its symplectic groupoid G(P ) is prequantizable. Algebroid Extension Associated to a Groupoid Extension Having described the relationship between the usual prequantization and the groupoid pic- ture we now turn to describe the infinitesimal counterpart of the situation. Just as in the case of groupoids the description is focused around extensions. Definition 3.2.7. A Lie algebroid extension of A by R is short exact sequence of algebroids ι π RM −→ A˜ −→ A, where RM is the trivial line bundle. Such an extension is called central if [˜α, ι(f)] = L#αf, ∞ for allα ˜ ∈ A˜, f ∈ Γ(RM ) = C (M). Just as in the case of Lie algebras, extensions of Lie algebroids are classified by the a second cohomology class of the deRham algebroid cohomology. Given a cocycle c ∈ H2(A) one can extend the algebroid A using the cocycle, explicitly, R → Ac → A, where Ac = A ⊕ R is endowed with the bracket, [α ⊕ f, β ⊕ g]c = [α, β] ⊕ (L#αg − L#βf + c(α, β)), and anchor #c = # ◦ prA. Proposition 3.2.8. The Lie algebroid central extensions of A by R are classified by the deRham algebroid cohomology H2(A). 2 Proof. Let c ∈ H (A), we will prove Ac is indeed and algebroid. The only non clear step in this assertion is checking that the bracket [·, ·]c satisfies the Leibniz rule: [α ⊕ f, h(β ⊕ g)]c = [α, hβ] ⊕ (L#αhg − L#hβf + c(α, hβ)) = (h[α, β] + (#α)hβ) ⊕ (hL#αg + gL#αh − hL#βf + hc(α, β)) = h ([α, β] ⊕ (L#αg + L#βf + c(α, β))) + (#αh)β ⊕ gL#αh = h([α ⊕ f, β ⊕ g]) + [#c(α ⊕ f)h](β ⊕ g). 2 Now, we check the extension depends on the cohomology class. Let dAb ∈ H (A) be a boundary term, we will show it induces an isomorphic extension. Indeed if dAb is a boundary term then, dAb(α, β) = L#αb(β) − L#βb(α) − b([α, β]), for some b ∈ H1(A). A straightforward (but tedious) calculation shows the map ϕ(α⊕f) = α ⊕ (f − b(α)) is an isomorphism of extensions. 33 At last, we will show that an extension R → A˜ → A induces a cocycle c ∈ H2(A). Since the previous extension is in particular a vector bundle extension, there exist a linear splitting σ : A → A˜. Again a straighforward (and also tedious) computation shows, c(α, β) = σ([α, β]) − [σ(α), σ(β)], is a 2-cocycle. Since the Lie algebroid A(G) is a sub-bundle of the tangent bundle T G any closed multilplicative 2-form on G induces (by restriction) a 2-cocycle cω on A cω(α, β) = ω(α, β). Thus given a closed multiplicative 2-form on a groupoid G, one can consider the algebroid Acω given by the extension of A(G) by the cocycle cω. In the case the groupoid is prequan- tizable, we will show the algebroid Acω is integrable. Proposition 3.2.9. Let (G, ω) be a symplectic groupoid, then any prequantization ˜ groupoid (G → G ⇒ M) of (G, ω) integrates Acω . Proof. Differentiating the extension given by the prequantization (G˜, θ˜) we obtain an ex- tension R → A˜ −−→Dπ A, where A˜ = A(G˜). Moreover, the one form θ˜ defines (by restriction) ˜ ∗ ∗ a one cocycle lθ on A, and the condition π ω = dθ restricts to π cω = dA˜lθ . ˜ ∼ ˜ We claim A = Acω , implying G integrates Acω . Define the map Θ: A˜ → A ⊕ R, α 7→ (dπ(α), l(α)). This map is an isomorphism. Indeed, since G˜ is a S1-bunde over G then T G˜ = T G ⊕ R, where the isomorphism is given by the splitting defined by the connection one form. Recall A˜ = T sG˜, because the groupoid multiplication commutes with the S1-action it implies the splitting restricts to T sG =∼ T sG ⊕ R, this restriction is Θ. 3.3 Quantization of Poisson Manifolds Lets recall the panorama, given a symplectic manifold (G, ω) one can associate to it the symplectic groupoid (M × M, ω˜ = ω × ω), which in turn describes a Lie algebroid extension R → Acω → TM. In the case (M, ω) is prequantizable, say by (P, θ), the previous algebroid extension can be integrated to a groupoid extension S1 → Gau(P ) −→π M ×M, which carries an invariant multiplicative form θ˜ such that π∗ω˜ = dθ˜. Moreover, we saw that the necessary and sufficient condition for (M, ω) to be prequantizable is equivalent to the integrability condition of the algebroid extension Aω. We will see how we can recover the prequantum data from the integration of the Lie algebroid extension concluding that, in the general scenario, prequantization can be seen as an integration problem. Integrability of Algebroid Extensions In this section we will use the theory of integration of Lie algebroids introduced in chapter 2 to solve the integrability problem of extensions of integrable Lie algebroids by 2 cocycles. 34 For this, let A be a algebroid and c ∈ H2(A) be a 2-cocycle, we will find a explicit description for Ac-paths and Ac-homotopies in terms of A-paths and A-homotopies. By definition a Ac-path over γ : I → M is a path ac : I → A ⊕ R; ac = (a, f); for a : I → A, f : I → R, d satisfying dt γ = #ac = #a. That is a Ac-path is a couple (a, f) where a : I → A is an A-path and f : I → R is a function. Now, the condition for a variation of Ac-paths (a, f): I × I → A ⊕ R, to be a Ac-homotopy is the condition that unique (b, h): I × I → A ⊕ R making (a, f)dt + (b, h)d : TI × TI → A ⊕ R into a algebroid morphism satisfies (b(, 1), h(, 1)) = (0, 0). Thus it breaks into two con- ditions: first b(, 1) = 1, i.e. a should be an A-homotopy; and second, h(, 1) = 0 for every ∈ [0, 1]. The second condition can be computed explicitly, recall that given a TM- connection on an algebroid the condition that (a, f)dt + (b, h)d : TI × TI → A ⊕ R is an algebroid morphism can be rewritten as a differential equation. In this scenario considering the connection (∇, ∂), where ∂ is the canonical flat connection in R, the condition in this component becomes d d h − f = c(a, b), h(, 0) = 0. dt d Integrating we find 0 Z t df h(, t0) = − c(f, h)dt; 0 d Z 1 df Z 1 h(, 1) = 0 ⇒ dt = c(a, b)dt. 0 d 0 Thus, the condition for (a, f) to be an Ac-homotopy is equivalent to a being a A-homotopy and f : I × I → R to satisfy the previous equation. Proposition 3.3.1. Any Ac-path (a, f) is Ac-homotopic to a path (a, r) where a is an A-path and r ∈ R. Moreover, any two such paths (a0, ro) and (a1, r1) are homotopic if a ∼ a as A-paths and 0 1 Z r1 − r0 = c(a, b)ddt. I2 R 1 Proof. Consider the variation of Ac-path between (a, f) and (a, 0 f) given by Z 1 f = f + (1 − ) fdt. 0 This variation is a A-homotopy, indeed Z 1 df Z 1 Z 1 Z 1 Z 1 dt = (f − fdt)dt = fdt − fdt = 0, 0 d 0 0 0 0 since c(a, a) = 0 the above equality is the condition to be a Ac-homotopy. 35 R Now if r1 − r0 = I2 c(a, b) then it is clear f = r0 + (1 − )r1 is a homotopy between the paths. Now, if A → M is the Lie algebroid of a s-connected groupoid G we will describe a 2 ∗ s topological extension of G by the Weinstein groupoid of Ac. Since c ∈ Γ(Λ (1 T G)) one x 2 −1 can extend this to a 2-form ωc ∈ Ω (s (x)) by x ωc (Xg,Yg) = ct(g)(DRg−1 (Xg),DRg−1 (Yg)). And define the period bundle by Z ∗ x 2 −1 −1 P(c) → M, Px(c) := γ ωc | (γ : S → s (x)) ∈ π2(s (x)) . S2 One can also define the bundle c R S(c) → M, Sx = /Px(c). Notice that the period bundle is fiberwise countable, thus if it is a smooth embedded sub-bundle of R then necessarily the fiber is a discrete subgroup. Moreover this implies S is a smooth (trivial) bundle of abelian groups with fiber S1. Example 3.3.2. Let (M, ω) be a simply connected symplectic manifold. The algebroid extension corresponding to the symplectic groupoid M × M is the algebroid AωTM ⊕ω R described in example §1.4.13. Since TM is integrable by M × M then TM-homotopies are given by differentiating homotopies h : I × I → M × M which are characterised by homotopies on M. Because M is simply connected any two paths with the same starting point and ending point are homotopic. Summarizing, G(TM) is the the set of differentials of paths I → M where two paths are homotopic if they have the same starting and ending points. Now Ac-paths (γ, r0), (γ, r1), where γ ∼ η i.e. γ(0) = η(0) = x and γ(1) = η(1) = x, are Ac-homotopic if and only if Z ω = r0 − r1, r0 − r1 ∈ P(M), ∇ where ∂∇ = γ − η. If the periods satisfies P(ω) ⊂ 2πZ then they describe a line bundle M × R/P(ω). The previous example can be taken to a general setting. In general by the from of the A-paths on A ⊕ R the monodromy groups of G will be given by the a direct product the monodromy groups of G and the periods bundle. Since the groupoid G is integrable the condition of A⊕R being integrable is equivalent to the monodromy groups being uniformly discrete. Now if the bundle of periods is an embedded sub-bundle of R × M it would imply the monodromy groups are, indeed, uniformly discrete and 2.3.8 would imply A ⊕ R is integrable. We have sketched: Proposition 3.3.3. Let G be a s-connected Lie groupoid with algebroid A. Suppose c ∈ 2 H (A) is a 2-cocycle and let R → Ac → A be the extension induced by such cocycle. Then there exist a extension of topological groupoids S(c) → G(Ac) → G. 36 This extension is a central extension of Lie groupoids if and only if P(c) is a smooth sub- bundle of R. From Integrability to Prequantizability: Existence and Uniqueness Back to the prequantizability problem, just as in the case of symplectic manifolds, the existence of S1-extensions implies the existence of a 1-form turning it into a prequantum extension. For this we use the auxiliary result by Crainic (Theorem 4.2 of [38]): Theorem 3.3.4. Let G be a s-simply connected Le groupoid, let ω ∈ Ω2(G) be a closed 2 multiplicative 2-form, and let cω ∈ H (A) be the algebroid 2-cocycle induced by restriction of ω. Then there exist a multiplicative 1-form θ satisfying dθ = ω if and only if cω is an exact algebroid cocycle. To show the extension of proposition 3.3.3 is a prequantum extension we will prove ∗ 2 ∗ the 2-cocycle π (c) ∈ H (Ac) coming from π ω ∈ G(Ac) is exact. Consider the algebroid Aω = A ⊕ R, then the projection on the second component lc(α, λ) = λ is a one form on A ⊕ R which satisfies dA(lc)((α, f), (β, g)) = L#αlc(β, g) + L#βlc(α, f) − lc([(α, f), (β, g)]) = L#αg + L#βf − L#αlcg − L#βf + c(α, β) = π∗c((α, f), (β, g)) Theorem 3.3.5. Let (G, ω) be a s-connected symplectic groupoid with algebroid A. Then G is prequantizable if and only if the bundle of periods P(c) is a smooth sub-bundle of R, where C ∈ H2(A) is the 2-cocycle induced by ω. Moreover the prequantization is unique. Proof. One on hand by proposition 3.2.9 the algebroid extension R → Ac → A is integrable, thus theorem 3.3.3 implies the bundle of periods is smooth. On the other hand, if the bundle of periods is smooth then theorem 3.3.3 implies the extension A → Ac → R is integrable to ∗ S(c) → G(Ac) → G. Moreover, since π (c) = dA(l) theorem 3.3.4 assures the existence of a ∗ 1-form θ such that π ω = dθ. Notice the groupoid G(Ac) extending G is unique given G is s-simply connected. In addition the 1-form θ described is unique up to boundary terms. For a integrable Poisson manifold P , we consider the s-simply connected symplectic groupoid integrating P and carry the discussion of prequantizability with this groupoid. We end the prequantization discussion with an example. ∗ Example 3.3.6. Cotangent groupoid: Consider the cotangent groupoid (T G, ωcan). Since there exist θ ∈ Ω(G) such that dθ = ω then the algebroid cocycle cω is trivial, up to boundary terms. Proposition 3.2 implies the algebroid extension A0(G) = A(G) ⊕ R has no twist in the bracket. From example ?? the groupoid T ∗G × S1 integrates the algebroid A0(G). It worth noting that in this case Z Z ω = dθ = 0, S2 S2 ∗ and the periods of bundle is trivial. Thus the set of A0-paths is isomorphic to T G × R which is exactly the s-connected groupoid of G ×S1. This implies that in the case the period bundle is trivial one should consider the groupoid T ∗G × S1 instead of the one given in the theorem 3.3.5. Now following the symplectic manifold case the prequantization of T ∗G is 37 ∗ 1 ∗ 1 (T G × S , dθ0 + π θL), where θL is the Maurer-cartan form on S . The multiplicativity of this connection follows from the multiplicativity of dθ0 and the multiplicativity of θL. Fell Line bundles We end this chapter by discussing the Line bundle picture on groupoid S1-extensions. We saw groupoids S1-extension are in particular S1-bundle over G, which itself had a multiplicative structure. The concept that formalises the multiplicative structure in the Kostant picture is that of a Fell line bundle: Definition 3.3.7. Suppose π : L → G is a line bundle over a groupoid G. Let (2) ∗ ∗ L := pr1L ⊗ pr2L, thus a generic fiber of L(2) is finite linear combination of elements of the set {s ⊗ t ∈ L ⊗ L|(π(s), π(t)) ∈ G(2)}. We say L → G is a Fell line bundle over G if there is an continuous multiplication bundle morphism m : L(2) → L, and an involution s 7→ s∗ such that: i) π(m(s, t)) = π(s)π(t), ii) π(b∗) = π(s)−1, iii) m(s, t)∗ = m(s∗, t∗), for all s, t ∈ L. We will normally suppress the map m, and simply write st in place of m(s, t). We will show the associated bundle A = (L × C)/ ∼ to a S1-central extension L → G of G is a Fell line bundle. For this consider the multiplication and involution [s, λ][t, µ] := [st, λµ], [s, λ]∗ = [s−1, λ¯], for all s, t ∈ L and λ, µ ∈ C. It is easy to check that these operation endow A → G with a Fell line bundle structure, in fact all the properties of a Fell line bundle follow from the properties of groupoid structure on L. We only have to check the maps are well defined. Let [sz, zλ¯ ] be another representative of the class [s, λ], where z ∈ S1, then [sz, zλ¯ ][t, µ] = [szt, zλµ¯ ] = [stz, zλµ¯ ] = [st, λµ], where we used the fact that L → G is a central S1-extension. A similar calculation shows the involution is also well defined. Now the multiplicativity of the connection on a S1 extension is translated in an affine connection ∇ on L → M satisfying the Leibinz rule ∇Dm(X)st = [∇Dpr1(X)s]t + s[∇Dpr2(X)t], for any X ∈ T G(2) and s ⊗ t ∈ Γ(L(2)). 38 Chapter 4 Quantization of Poisson Manifolds The last step in a geometric quantization procedure is to cut down the space of sections of the prequantum line bundle constructed in the chapter 3. This is done by using a polarisation on the symplectic manifold, in the good case scenario the reduced space of sections can be described as the space of sections of another line bundle. Afterwards, with the use of half-densities we will define an inner product in this new space of sections in order to complete it to the desired quantum Hilbert space of the symplectic manifold. Through the general discussion the example of the quantization of the cotangent bundle will motivate and illustrate the main concepts and constructions. 4.1 Quantization of Symplectic Manifolds Again, as in the previous chapter, we begin our discussion for the case of symplectic mani- folds and work our way to the general notion for symplectic groupoids. Until now we have constructed a prequantum Hilbert space out of a prequantum line bundle. However this Hilbert space is not the right one; in some sense it is to big to represent the quantization of the symplectic manifold (M, ω). Considering the symplectic manifold (M, ω) as the phase space of some physical system the quantum Hilbert space has information of both the con- figuration variables and the momenta variables. Since in quantum mechanics the Hilbert space has information just of the configuration variables the prequantum Hilbert space is twice as big as the desired Hilbert space. Thus we need a way of choosing configuration and momenta variables on the manifold (M, ω), this is done though a polarisation. In the case of a good enough polarisation we will discuss how to construct the quantum Hilbert space of a symplectic manifold (M, ω) out of the prequantum data and the polarisation data. polarisation of Symplectic Manifolds The need of a polarisation can be motivated by the quantization of the prototypical symplec- tic space T ∗Rn =∼ R2n of example §2.3.13. The prequantum Hilbert space of this symplectic manifold is L2(R2). However, T ∗R is the phase space of a free particle in one dimension, and it is known that the corresponding quantum space of this space should be L2(R), not L2(R2). As mentioned before, the intuitive idea is that the quantum Hilbert space should only contain information of the configuration variables, however the prequantum space has the information of both configuration and momenta variables. Thus we need to halve the variables of the prequantum Line bundle in order to get the right quantum Hilbert space. 39 Definition 4.1.1. A Real polarisation of a symplectic manifold (M, ω) is an involutive Lagrangian subbundle F of TM. Since a real polarisation is involutive, Frobenius theorem asserts that there exists a foliation FM of M such that T FM = Fx, thus a real polarisation induces a foliation by Lagrangian manifolds. On the other hand, it is easily seen that the tangent bundle to the leafs of a foliation by Lagrangian submanifolds is a involutive lagrangian subbundle, that is, a real polarisation. Thus we have proved, Proposition 4.1.2. There is a one to one correspondence between real polarisations and foliations by Lagrangian submanifolds. ∗ Example 4.1.3. Let M be a manifold and let (T M, ωcan) be the cotangent space with the canonical symplectic structure. Now T ∗M −→π M has a natural Lagrangian foliation by the fibers of the cotangent bundle, this foliation induces a real polarisation known as the −1 vertical polarisation. Since the tangent space to the fiber (x, p) ∈ π (x) is ker(Dπ)(x,p) the polarisation can be described as F = ker(Dπ): T (T ∗M → TM). It is worth mentioning that real polarisations might not exist for a general symplectic manifold. There is a more general definition of complex polarisation, this is a complex involutive Lagrangian sub-bundle of the complexified tangent bundle TCM. Although this kind of polarisations are more general there and allow to consider more examples even in this case there examples of symplectic manifolds that admit no polarisations whatsoever, real or complex, see [39]. We will be interested in well behaved examples where there exist real polarisations, thus we will not be interested in complex polarisation and will not discuss them further. Definition 4.1.4. A section s ∈ Γ(L) of a line bundle (L → M, ∇) is said to be polarised if ∇X s = 0 for all X ∈ ΓF. Remark 4.1.5. In the Souriau picture a section ϕ ∈ Γ(L) = C∞(L, C)S1 of a S1-bundle with connection (L → M, θ) is said to be polarised if Hθ(X)s = 0 for all X ∈ Γ(F). The space of polarised sections will be denoted by ΓF (L). ∗ Example 4.1.6. Consider (T M, ωcan) the cotangent space with canonical symplectic 1 ∗ π ∗ structure, (S × T M −→ M, dθ0 + π θL) the prequantum bundle, and F the vertical po- larisation of T ∗M. In this scenario the space of sections is C∞(T ∗M) and H(X)ϕ = Xϕ. Since a vector field on the vertical polarisation is a vector field tangent to the fibers, we conclude that the space of polarised sections is the same as the space of functions constant on fibers π∗C∞(T ∗M) = {ϕ ∈ C∞(T ∗M)|ϕ = π∗f, for some f ∈ C∞(M)}. Fibrating polarisations and Line Bundles on Leaf Spaces Naively the desired quantum Hilbert space would be a completition of ΓF (L) with respect to some inner product. However there is no clear description space of polarised sections, thus there is no clear way of how to endow ΓF (L) with a natural inner product. In what follows we will deal with a good polarisation scenario. In this scenario the space of polarised sections can be realised as the space of sections of some other line bundle. 40 Definition 4.1.7. A real polarisation is said to be fibrating if: i) the leaf space M/F of the foliaton defined by the real polarisation is a manifold and that the quotient map π : M → M/F is a submersion, ii) each leaf of the polarisation is simply connected. For clearness of the discussion in what follows we will deal with the vector bundle description of prequantization, that is, Hermitian line bundles. Suppose (M, ω) is a prequantizable symplectic manifold and (L → M, θ) is a prequantum Line bundle of M, and suppose F is a fibrating real polarisation. Let Λ be a leaf of the polarisation, since Λ is Lagrangian and the curvature of θ is ω it follows that (L|Λ → Λ, θ|Λ) is a flat line bundle over a simply connected space. One can define parallel section, i.e. H(X)ϕ = 0, on the L|Λ as follows: given any path γ on Λ and a starting point l ∈ L|Λ one can use the connection θ to horizontaly lift γ to a path on L|Λ. Since Λ is simply connected the lifting does not depend on the path, the difference between two paths will be given by the holonomy of the bundle which in this case is trivial, the parallel section ϕ is well defined. Notice that the set of parallel sections is parametrized by the initial value at a point, thus it is a one dimensional complex vector space. One can consider a such a one dimensional space for each leaf moreover the local triviality of the foliation implies such an assignment is locally trivial. Thus one can consider a line bundle LF → M/F over the leaf space. Proposition 4.1.8. The space of section on Γ(LF ) is isomorphic to the space ΓF (L) of polarised sections on L. Proof. A section s ∈ Γ(LF ) is by definition a transversaly smooth collection {sy}y ∈ M/F of parallel sections each over the restriction of L on Λx. The smoothness of such a collection imply that they define a smooth section on L. On the other hand, the restriction of a polarised section s ∈ ΓF (L) on a leaf Λx is by definition a parallel section on L|Λx . Thus the restriction of s to each leaf is a transversally smooth family of parallel sections on {Λx}, that is, a section of Γ(LF ) Remark 4.1.9. Notice that the property that each leaf of the polarisation was a simply connected manifold is not necessary for the previous discussion. One could simply ask that the holonomy of the connection restricted to each leaf be trivial in order to carry with the construction presented here. Intermezzo: Half-Densities We will associate a Hilbert space to LF , however as a preparatory for this purpose we will quickly discuss about densities and half-densities. Densities are geometric objects that under change of coordinates change as the absolute value of the Jacobian matrix of the change of variables. Thus densities satisfy the change of coordinates rule necessary to define integration over them. On an oriented manifold densities can be identified with positive top forms on M, thus densities can be regarded as generalisations of volume forms. 41 Let FC(M) → M be the complexified frame bundle over a n-dimensional manifold M, this is the principal GL(n, C) bundle over M associated to the complexified tangent bundle T CM → M. Since there is a natural GL(n, C) representation on C by GL(N, C) → C∗, α(T ) = | det T |−α, one can consider an associated bundle with respect to this representation. Definition 4.1.10. For α > 0, the bundle of α-densities |Λ|α(M) is the line bundle over M associated to (F (M), C, | det |α). An α-density is a section of |Λ|α(M). We will be interested in 1-densities, which will be simply called densities, and 1/2- densities which will be referred as half-densities. In the principal bundle description a α-density is a map ρ : TM → C satisfying the equivariance condition −α ρx(T v1, . . . T vn) = | det(T ) |ρx(V1,...,Vn), n for any T ∈ GL(n, R) and any frame v1, . . . vn ∈ (TxM) . In particular under a change of coordinates from (Uα, φα) to (Uβ, φβ) given by gαβ :(Uα ∩ Uβ) → (Uα ∩ Uβ) a density changes as ρα(Jv1, . . . Jvn) = | det J| · ρβ(v1, . . . vn) where J = Dgαβ is the Jacobi matrix of the change of coordinates gαβ. Due to the transformation rule of densities the integral of a density ρ ∈ |Λ|M supported on the chart (Uα, φα) Z Z −1 ρ = ψα ◦ ρ ◦ φα dµ, U φα(Uα) R is well defined. In the above expression tα : |Λ|M|Uα → φ(Uα)× is the trivialisation of the density bundle induced from the chart (Uα, φα) (recall the density bundle is an associated bundle of F (M)) and dµ is the Lebesgue measure. Thus, using partitions of unity {bα} subject to chart covering {Uα, φα} of M one can define an integral for any compactly supported density ρ ∈ |Λ|cM by Z X Z ρ = bα ρ. Uα By the Riezs representation theorem (c.f [40, Section 7.2]) such an integral defines a measure dρ on M which is locally absolutely continuous with respect to the Lebesgue measure. By construction, any other measure η arising from another density is absolutely contin- uous with respect to the one described above, thus Z Z dρ ρ = η dη dρ where dη is the Radon-Nikodym derivative. The local form of the integral and the fun- dρ damental theorem of calculus shows that locally dη is given by a smooth function. Since smoothness is a local property it implies that fixing a a density ρ any other density can be written as ρ = dρ η for some smooth function dρ ∈ C∞(M). The previous discussion yields dη η. c the following proposition. 42 Proposition 4.1.11. The bundle |Λ|cM is a trivialisable bundle, with the trivialisation dρ given by η 7→ dη for some fixed density ρ. Given that a density ρ induces a measure on M one can talk about Lp(M, |Λ|) the space of p-integrable functions with respect to the measure dρ. Notice this space does not depend on the density since any two densities are absolutely continuous with respect to each other. It is worth noting that in the case the manifold M has just one chart the measure dρ is mutually absolutely continuous to the Lebesgue measure, implying L2(M, |Λ|) =∼ L2(M). To end this section we will discuss how half-densities can be multiplied to define a densities. Proposition 4.1.12. The multiplication map α β α+β |Λ| M ⊗ |Λ| M → |Λ| ρ1 ⊗ ρ2 7→ ρ1ρ¯2 induces a bundle isomorphism. α β Proof. If ρ1 ∈ |Λ| M and ρ2 ∈ |Λ| M, then