Supported by Russian Foundation of Basic Research (grant no. 04-01-00190) .
<ph f="cmbx">Noncommutative geometry of foliations</ph>

Yuri A. Kordyukov

Institute of Mathematics, Russian Academy of Sciences, Ufa E-mail address : yuri@imat.rb.ru

Contents

1 Introduction

The starting point of noncommutative geometry is the passage from geometric spaces to algebras of functions on these spaces with the subsequent translation of basic analytic and geometric concepts and constructions on geometric spaces to the algebraic language and their extension to general noncommutative algebras. Such a procedure is well-known and was applied for a long time, for instance, in algebraic geometry, where it is related with the study of commutative algebras. It is also well known that the theory of C *   -algebras is a far-reaching generalization of the theory of topological spaces, and the theory of von Neumann algebras is a generalization of the classical measure and integration theory.
The main purpose of noncommutative differential geometry, which was initiated by Connes [39and is actively developing at present time (cf. the recent surveys [45, 46and the books [42, 81, 119in regard to different aspects of noncommutative geometry), consists in the extension of the methods described above to the analytic objects on geometric spaces and to the noncommutative algebras. Here the main attention is focused on that, first, a correct noncommutative generalization applied in the classical setting, that is, to an algebra of functions on a compact manifold should agree with its classical analogue, and, second, it should inherit nice algebraic and analytic properties of its classical analogue. Nevertheless, it should be noted that, as a rule, such noncommutative generalizations are quite nontrivial and have richer structure and essentially new features than their commutative analogues.
Noncommutative geometry lies on the border of functional analysis and differential geometry and is of great importance for these areas of mathematics.
On the one hand, the development of geometric methods in the operator theory and the theory of operator algebras allows to use fruitfully geometric intuition for the investigation of various problems of abstract functional analysis. On the other hand, there are many examples of singular geometric spaces, which are badly described from the point of view of classical measure theory and resist to the study by usual “commutative” methods of geometry, topology and analysis, but one can naturally associate to them a noncommutative algebra, which can be considered as an analogue of the algebra of (measurable, continuous, smooth and so on) functions on the given geometric object. Let us give some examples of such singular objects:
  • (1) Manifolds with singularities, for instance, manifolds with isolated conic singular points.
  • (2) Discrete spaces.
  • (3) The dual space to a group (discrete or a Lie group).
  • (4) Cantor sets.
  • (5) The orbit space of a group action on a manifold.
  • (6) The leaf space of a foliation.
Use of the notions and methods of noncommutative geometry for a noncommutative algebra, being an analogue of an algebra of functions on a singular geometric space, allows in many cases, first of all, just to define some reasonable analytic and geometric objects associated with the given space and pose sensible problems, that, in its turn, gives possibility to apply properties of these objects for getting an information about geometry of this space.
This paper contains a short exposition of the methods of noncommutative geometry as applied to the study of one of the fundamental examples of noncommutative geometry, namely, the leaf spaces of a foliation on a smooth manifold, or, that is the same, to the study of the transverse structure of foliations. Our purpose is to give a survey of the basic notions and methods of noncommutative geometry, to show how one can associate various objects of noncommutative geometry to foliations and how these noncommutative analogues of classical notions are related with classical objects on foliated manifolds, and to describe applications of the methods of noncommutative geometry to the study of geometry of foliations.
The author is grateful to N.I. Zhukova for useful remarks.

2 Background information on foliation theory

2.1 Foliations

In this Section, we recall the definition of a foliated manifold and some notions related to foliations (concerning to different aspects of the foliation theory see, for instance, [28, 29, 30, 74, 133, 136, 137, 153, 177).
Let M   be a smooth manifold of dimension n   . (Here and later on, “smooth” means “of class C   ”. We will always assume that all our objects under consideration are of class C   .)
Definition 2.1. (1) An atlas A = { ( U i , φ i ) }   , where φ i : U i M R n   , of the manifold M   is called an atlas of a foliation of dimension p   and codimension q   ( p n , p + q = n   ), if, for any i   and j   such that U i U j   , the coordinate transformation φ i j = φ i φ j 1 : φ j ( U i U j ) R p × R q φ i ( U i U j ) R p × R q   has the form φ i j ( x , y ) = ( α i j ( x , y ) , γ i j ( y ) ) , ( x , y ) φ j ( U i U j ) R p × R q .   (2) Two atlases of a dimension p   foliation are equivalent, if their union is again an atlas of a dimension p   foliation.
(3) A manifold M   endowed with an equivalence class   of atlases of a dimension p   foliation is called a manifolds with a dimension p   foliation (or a foliated manifold).
An equivalence class   of foliation atlases is also called a complete atlas of a foliation. We will also say that   is a foliation on M   .
A pair ( U , φ )   that belongs to the atlas of the foliation   and also the corresponding map φ : U R n   are called a foliated chart of the foliation   , and U   is called a foliated coordinate neighborhood.
Let φ : U M R n   be a foliated chart. The connected components of the set φ 1 ( R p × { y } ) , y R q   are called the plaques of the foliation   .
Plaques of   given by all possible foliated charts form a base of a topology on M   . This topology is called the leaf topology on M   . We will also denote by   the set M   endowed with the leaf topology. One can introduce the structure of a p   -dimensional manifold on   .
Connected components of the manifold   are called leaves of the foliation   . Leaves are (one-to-one) immersed p   -dimensional submanifolds in M   . For any x M   , there is a unique leaf, passing through x   . We will denote this leaf by L x   .
One can give the equivalent definition of a foliation, saying, that there is given a foliation   of dimension p   on a manifold M   of dimension n   , if M   is represented as a disjoint union of a family { L λ : λ }   of connected, (one-to-one) immersed submanifolds of dimension p   , and there is an atlas A = { ( U i , φ i ) }   of the manifold M   such that, for any i   and for any λ   , the connected components of the set L λ U i   are given by equations of the form y = const   .
Example 2.2. Let M   be a smooth manifold of dimension n   , B   a smooth manifold of dimension q   and π : M B   a submersion (that is, the differential d π x : T x M T π ( x ) B   is surjective for any x M   ). The connected components of the pre-images of points of B   under the map π   determine a codimension q   foliation on M   , which is called the foliation determined by the submersion π   . If, in addition, the pre-images π 1 ( b ) , b B ,   are connected, the foliation is called simple.
Example 2.3. Let X   be a nonsingular (that is, a non-vanishing) smooth vector field on a compact manifold M   . Then its phase curves form a codimension one foliation.
More generally, suppose that a connected Lie group G   acts smoothly on a smooth manifold M   , and, moreover, the dimension of the isotropy subgroup G x = { x G : g x = x }   does not depend on x M   . In particular, one can assume that the action is locally free, that means G x   is discrete for any x M   . Then the orbits of the Lie group G   action define a foliation on M   .
Example 2.4. Linear foliation on torus. Consider a vector field X ~   on R 2   given by X ~ = α x + β y   with constant α   and β   . Since X ~   is invariant under all translations, it determines a vector field X   on the two-dimensional torus T 2 = R 2 / Z 2   . The vector field X   determines a foliation   on T 2   . The leaves of   are the images of the parallel lines L ~ = { ( x 0 + t α , y 0 + t β ) : t R }   with the slope θ = β / α   under the projection R 2 T 2   .
In the case when θ   is rational, all leaves of   are closed and are circles, and the foliation   is determined by the fibers of a fibration T 2 S 1   . In the case when θ   is irrational, all leaves of   are everywhere dense in T 2   .
Example 2.5. Homogeneous foliations. Let G   be a Lie group and H G   its connected Lie subgroup. The family { g H : g G }   of right cosets of H   forms a foliation   on G   . If H   is a closed subgroup, then G / H   is a manifold and   is a foliation, whose leaves are given by the fibers of the fibration π : G G / H   .
Moreover, suppose that Γ G   is a discrete subgroup G   . Then the set M = Γ \ G   of left cosets of Γ   is a manifold of the same dimension as G   . If Γ   is cocompact in G   , M   is compact. In any case, because   is invariant under the left translations, and Γ   acts on the left, the foliation   is mapped by the map G M = Γ \ G   to a well-defined foliation Γ   on M   , which is often denoted by ( G , H , Γ )   and is called a locally homogeneous foliation. The leaf of Γ   through a point Γ g M   is diffeomorphic to H / ( g Γ g 1 H )   .
Example 2.6. The Reeb foliation. Let us describe a classical construction of a codimension 1 foliation on the three-dimensional sphere S 3   due to Reeb [152. Let D 2   denote the disk { ( x , y ) R 2 : x 2 + y 2 1 }   . We start with the foliation in the cylinder { ( x , y , z ) R 3 : x 2 + y 2 1 } = D 2 × R   , whose leaves are the boundary of the cylinder x 2 + y 2 = 1   and surfaces z = c + exp ( 1 / ( 1 x 2 y 2 ) )   with an arbitrary constant c   . Since this foliation is invariant under translations in z   , it is mapped by the standard projection D 2 × R D 2 × R / Z = D 2 × S 1   to a foliation R   on the solid torus D 2 × S 1   . Finally, observe that the standard three-dimensional sphere S 3 = { x = ( x 1 , x 2 , x 3 , x 4 ) R 4 : x 1 2 + x 2 2 + x 3 2 + x 4 2 = 1 }   is obtained by gluing along the boundary of two copies of D 2 × S 1   . More precisely, S 3 = S 1 3 S 2 3   , where S 1 3 = { x S 3 : x 1 2 + x 2 2 x 3 2 + x 4 2 } , S 2 3 = { x S 3 : x 1 2 + x 2 2 x 3 2 + x 4 2 } .   A diffeomorphism S 1 3 = D 2 × S 1   is given by x S 1 3 ( x 1 x 3 2 + x 4 2 , x 2 x 3 2 + x 4 2 , x 3 x 3 2 + x 4 2 , x 4 x 3 2 + x 4 2 ) D 2 × S 1 ,   where we consider S 1   as { ( x , y ) R 2 : x 2 + y 2 = 1 }   , and its inverse has the form ( y 1 , y 2 , y 3 , y 4 ) D 2 × S 1 ( y 1 | y | , y 2 | y | , y 3 | y | , y 4 | y | ) S 1 3 ,   where | y | 2 = y 1 2 + y 2 2 + y 3 2 + y 4 2   . Similar formulas can be written for S 2 3   .
Take the Reeb foliations R   on S 1 3   and S 2 3   . Since the boundaries S 1 3   and S 2 3   are compact leaves, there is a well-defined foliation on the union of S 1 3   and S 2 3   , that is, on the sphere S 3   , which is called the Reeb foliation on S 3   . The only compact leaf of the Reeb foliation is the common boundary S 1 3 = S 2 3   of S 1 3   and S 2 3   , diffeomorphic to the two-dimensional torus T 2   . The other leaves are diffeomorphic to the plane R 2   .
Example 2.7. Suspension. Let B   be a connected manifold and B ~   its universal cover equipped with the action of the fundamental group Γ = π 1 ( B )   by deck transformations. Suppose that there is given a homomorphism φ : Γ Diff ( F )   of Γ   to the group Diff ( F )   of diffeomorphisms of a smooth manifold F   . Define a manifold M = B ~ × Γ F   as the quotient of the manifold B ~ × F   by the action of Γ   given, for any γ Γ   , by γ ( b , f ) = ( γ b , φ ( γ ) f ) , ( b , f ) B ~ × F .   There is a natural foliation   on M   , whose leaves are the images of the sets B ~ × { f } , f F ,   under the projection B ~ × F M   . If, for any γ Γ , γ e   , the diffeomorphism φ ( γ )   has no fixed points, all leaves of   are diffeomorphic to B ~   .
There is defined the bundle π : M B : [ ( b , f ) ] b m o d Γ   such that the leaves of   are transverse to the fibers of π   . The bundle π : M B   is often said to be a flat foliated bundle.
A foliation   determines a subbundle F = T   of the tangent bundle T M   , called the tangent bundle to   . It consists of all vectors, tangent to the leaves of   . Denote by X ( M ) = C ( M , T M )   the Lie algebra of all smooth vector fields on M   with the Lie bracket and by X ( ) = C ( M , F )   the subspace of vector fields on M   , tangent to the leaves of   at each point. The subspace X ( )   is a subalgebra of the Lie algebra X ( M )   . Moreover, by the Frobenius theorem, a subbundle E   of T M   is the tangent bundle to a foliation if and only if it is involutive, that is, the space of sections of this bundle is a Lie subalgebra of the Lie algebra X ( M )   : for any X , Y C ( M , E )   we have [ X , Y ] C ( M , E )   .
Let us introduce the following objects:
  •   τ = T M / T   is the normal bundle to   ;
  •   P τ : T M τ   is the natural projection;
  •   N * = { ν T * M : ν , X = 0 for any X F }   is the conormal bundle to   .
Usually, we will denote by ( x , y ) I p × I q   ( I = ( 0 , 1 )   is the open interval) the local coordinates in a foliated chart φ : U I p × I q   and by ( x , y , ξ , η ) I p × I q × R p × R q   the local coordinates in the corresponding chart on T * M   .
Then the subset N * π 1 ( U ) = U 1   (here π : T * M M   is the bundle map) is given by the equation ξ = 0   . Therefore, φ   determines naturally a foliated chart φ n : U 1 I p × I q × R q   on N *   with the coordinates ( x , y , η )   .
Any q   -dimensional distribution Q T M   such that T M = F Q   is called a distribution transverse to the foliation or a connection on the foliated manifold ( M , )   . Any Riemannian metric g   on M   determines a transverse distribution H   , which is given by the orthogonal complement of F   with respect to the given metric: H = F = { X T M : g ( X , Y ) = 0 for any Y F }   .
Definition 2.8. A vector field V   on a foliated manifold ( M , )   is called an infinitesimal transformation of   if [ V , X ] X ( )   for any X X ( )   .
The set of all infinitesimal transformations of   is denoted by X ( M / )   .
If V X ( M / )   and T t : M M , t R   is the flow of the vector field V   , then the diffeomorphisms T t   are automorphisms of the foliated manifold ( M , )   , that is, they take each leaf of   to, possibly, another leaf.
Definition 2.9. A vector field V   on a foliated manifold ( M , )   is called projectable, if its normal component P τ ( V )   is locally the lift of a vector field on the local base.
In other words, a vector field V   on M   is projectable, if in any foliated chart with local coordinates ( x , y ) , x R p , y R q ,   it has the form V = i = 1 p f i ( x , y ) x i + j = 1 q g j ( y ) y j .   There is a natural action of the Lie algebra X ( )   on C ( M , τ )   . The action of X X ( )   on N C ( M , τ )   is given by θ ( X ) N = P τ [ X , N ~ ] ,   where N ~ X ( M )   is any vector field on M   such that P τ ( N ~ ) = N   . A vector field N X ( M )   is projectable if and only if its transverse component P τ ( N ) C ( M , τ )   is invariant under the X ( )   -action θ   . From here, one can easily see that a vector field on a foliated manifold is called projectable if and only if it is an infinitesimal transformation of the foliation.

2.2 Holonomy

Let ( M , )   be a foliated manifold. The holonomy map is a generalization to the case of foliations of the first return map (or the Poincaré map) for flows.
Definition 2.10. A smooth transversal is a compact q   -dimensional manifold T   , possibly with boundary, and an embedding i : T M   , whose image is everywhere transverse to the leaves of   : T i ( t ) i ( T ) T i ( t ) = T i ( t ) M   for any t T   .
We will identify a transversal T   with the image i ( T ) M   .
Definition 2.11. A transversal is complete, if it meets every leaf of the foliation.
Take an arbitrary continuous leafwise path γ   with endpoints γ ( 0 ) = x   and γ ( 1 ) = y   . (We will call a path γ : [ 0 , 1 ] M   leafwise, if its image γ ( [ 0 , 1 ] )   is contained entirely in one leaf of the foliation). Let T 0   and T 1   are smooth transversals such that x T 0   and y T 1   .
Choose a partition t 0 = 0 < t 1 < < t k = 1   of the segment [ 0.1 ]   such that for any i = 1 , , k   the curve γ ( [ t i 1 , t i ] )   is contained in some foliated chart U i   . Shrinking, if necessary, the neighborhoods U 1   and U 2   , one can assume that, for any plaque P 1   of U 1   , there is a unique plaque P 2   of U 2   , which meets P 1   . Shrinking, if necessary, the neighborhoods U 1   , U 2   and U 3   , one can assume that, for any plaque P 2   of U 2   , there is a unique plaque P 3   of U 3   , which meets P 2   and so on. After all, we get a family { U 1 , U 2 , , U k }   of foliated coordinate neighborhoods, which covers the curve γ ( [ 0 , 1 ] )   , such that, for any i = 1 , , k   and for any plaque P i 1   of U i 1   , there is a unique plaque P i   of U i   , which meets P i 1   . In particular, we get a one-to-one correspondence between the plaques of U 1   and the plaques of U k   .
The smooth transversal T 0   determines a parametrization of the plaques of U 1   near x   . Accordingly, a smooth transversal T 1   determines a parametrization of the plaques of U k   near y   . Taking into account one-to-one correspondence between the plaques of U 1   and U k   constructed above, we get a diffeomorphism H T 0 T 1 ( γ )   of some neighborhood of x   in T 0   to some neighborhood of y   in T 1   , which is called the holonomy map along the path γ   .
One can easily see that the germ of H T 0 T 1 ( γ )   at x   does not depend on the choice of a partition t 0 = 0 < t 1 < < t k = 1   of the segment [ 0.1 ]   and of a family of foliated coordinate neighborhoods { U 1 , U 2 , , U k }   . Moreover, the germ of H T 0 T 1 ( γ )   at x   is not changed, if we replace γ   by any other continuous leafwise path γ 1   with the initial point x   and the final point y   , which is homotopic to γ   in the class of continuous leafwise paths with the initial point x   and the final point y   .
There is a slightly different definition of holonomy [88. First, let us introduce some notions.
Definition 2.12. A map f : V M R q   is called distinguished, if in a neighborhood of any point in V   there exists a foliated chart ( U , φ )   such that the restriction of f   to U   has the form p r n q φ   , where p r n q : R n = R p × R q R q   is the natural projection.
Consider the set D   of germs of distinguished maps at various points of M   (or, briefly speaking, the set of distinguished germs). Let σ : D M   be the map, which associates to a distinguished germ at x M   the point x   .
Let ( U , φ )   be a foliated chart, P   its plaque. Consider the subset P ~   of D   , which consists of all germs of the corresponding distinguished map p r n q φ : U R q   in different points of P   . Sets of the form P ~   , determined by the plaques P   of all possible foliated charts, form a base of a topology on D   , which endows D   with the structure of a p   -dimensional manifold. This topology is called the leaf topology on D   . It can be easily checked [88that σ   is a covering map from the manifold D   to the manifold   .
Now consider a continuous leafwise path γ   with the initial point x   and the final point y   . Let π σ 1 ( x )   , and let γ ~   be the lift of γ   to D   with the initial point π   via the covering map σ   . The holonomy map associated with γ   is the map h γ : σ 1 ( x ) σ 1 ( y ) ,   which takes π σ 1 ( x )   to the final point of the path γ ~   .
The connection between the holonomy maps defined above is established as follows. Let T 0   and T 1   be smooth transversals such that x T 0   and y T 1   . Let π σ 1 ( x )   , and let f : U M R q   be a distinguished map defined in a neighborhood of x   . A choice of a distinguished map f   defined in a neighborhood of x   is equivalent to a choice of a local coordinate system f i   on T 0   defined in a neighborhood of x   . The diffeomorphism H T 0 T 1 ( γ )   allows to define a local coordinate system on T 1   defined in some neighborhood of y   , that, in its turn, gives a distinguished map f 1 : V M R q   defined in a neighborhood of y   . The germ of the distinguished map f 1   at y   coincides with h γ ( π ) σ 1 ( y )   .
If γ   is a closed leafwise path with the initial and final points x   , and T   is a smooth transversal such that x T   , then H T T ( γ )   is a local diffeomorphism of T   , which leaves x   fixed. The correspondence γ H T T ( γ )   defines a group homomorphism H T : π 1 ( L x , x ) Diff x ( T )   from the fundamental group π 1 ( L x , x )   of the leaf L x   to the group Diff x ( T )   of germs at x   of local diffeomorphisms of T   , which leave x   fixed. The image of the homomorphism H T   is called the holonomy group of the leaf L x   at x   .
The holonomy group of a leaf L   at a point x L   is independent modulo isomorphism of the choice of a transversal T   and x   . A leaf is said to have trivial holonomy, if its holonomy group is trivial.
Example 2.13. Let X   be a complete nonsingular vector field on a manifold M   of dimension n   , x 0   a (for simplicity, isolated) periodic point of the flow X t   of the given vector field and C   the corresponding closed phase curve. Let T   be an ( n 1 )   -dimensional submanifold of M   , passing through x 0   transverse to the vector X ( x 0 )   : T x M = T x T R X ( x 0 )   . For all x T   , closed enough to x 0   , there is the least t ( x ) > 0   such that the corresponding positive semi-trajectory of the flow { X t ( x ) : t > 0 }   meets T   : X t ( x ) ( x ) T   .
Thus, we get a local diffeomorphism φ T : x X t ( x ) ( x )   of T   , defined in a neighborhood of x 0   and taking x 0   to itself. This diffeomorphism is called the first return map (or the Poincaré map) along the curve C   .
If   is the foliation on M   given by the trajectories of X   , then the holonomy group of the leaf C   coincides with Z   , and the germ of φ T   at x 0   is a generator of this group.
Example 2.14. For the Reeb foliation of the three-dimensional sphere S 3   all noncompact leaves have trivial holonomy. The holonomy group of the compact leaf is isomorphic to Z 2   .
For any smooth transversal T   and for any x T   , there is a natural isomorphism of the tangent space T x T   with the normal space τ x   to   . Thus, the normal bundle τ   plays a role of the tangent bundle to the (germs of ) transversals to   . For any continuous leafwise path γ   with the initial point x   and the final point y   and for any smooth transversals T 0   and T 1   such that x T 0   and y T 1   , the differential of the holonomy map H T 0 T 1 ( γ )   at x   defines a linear map d H T 0 T 1 ( γ ) x : τ x τ y   . It is easy to check that this map is independent of the choice of transversals T 0   and T 1   . It is called the linear holonomy map and is denoted by d h γ : τ x τ y   . Taking the adjoint of d h γ   , one gets a linear map d h γ * : N * y N * x   .
Now we turn to another notion related with the holonomy, the notion of holonomy pseudogroup. First, recall the general definition of a pseudogroup.
Definition 2.15. A family Γ   , consisting of diffeomorphisms between open subsets of a manifold X   (or, in other words, of local diffeomorphisms of X   ) is called a pseudogroup on X   , if the following conditions hold:
  • (1) if Φ Γ   , then Φ 1 Γ   ;
  • (2) if Φ 1 : U U 1   and Φ 2 : U 1 U 2   belong to Γ   , then Φ 2 Φ 1 : U U 2   belongs to Γ   ;
  • (3) if Φ : U U 1   belongs to Γ   , then its restriction to any open subset V U   belongs to Γ   ;
  • (4) if a diffeomorphism Φ : U U 1   coincides on some neighborhood of each point in U   with an element of Γ   , then Φ Γ   ;
  • (5) the identity diffeomorphism belongs to Γ   .
Example 2.16. The set of all local diffeomorphisms of a manifold X   form a pseudogroup on X   . One can also consider pseudogroups, consisting of local diffeomorphisms of a manifold X   , which preserve a geometric structure, for instance, the pseudogroup of local isometries and so on.
Definition 2.17. Let ( M , )   be a smooth foliated manifold and X   the disjoint union of all smooth transversals to   . The holonomy pseudogroup of the foliation   is the pseudogroup Γ   , which consists of all local diffeomorphisms of X   , whose germ at any point coincides with the germ of the holonomy map along a leafwise path.
Definition 2.18. Let ( M , )   be a smooth foliated manifold and T   be a smooth transversal. The holonomy pseudogroup induced by the foliation   on T   is the pseudogroup Γ T   , which consists of all local diffeomorphisms of T   , whose germ at any point coincides with the germ of the holonomy map along a leafwise path.
There is a special class of smooth transversals given by good covers of the manifold M   .
Definition 2.19. A foliated chart φ : U M R p × R q   is called regular, if it admits an extension to a foliated chart φ ¯ : V R p × R q   such that U ¯ V   .
Definition 2.20. A cover of a manifold M   by foliated neighborhoods { U i }   is called good, if:
  • (1) Any chart ( U i , φ i )   is a regular foliated chart;
  • (2) If U ¯ i U ¯ j   , then U i U j   and the set U i U j   is connected. The same is true for the corresponding foliated neighborhoods V i   ;
  • (3) Each plaque of V i   meets at most one plaque of V j   . A plaque of U i   meets a plaque of U j   if and only if the intersection of the corresponding plaques of V i   and V j   is nonempty.
Good covers always exist.
Let U = { U i }   be a good cover for the foliation   , φ i : U i = I p × I q   . For any i   , put T i = φ i 1 ( { 0 } × I q ) .   Then T i   is a transversal, and T = T i   is a complete transversal. For y T i   denote by P i ( y )   the plaque of U i   , passing through y   . For any pair of indices i   and j   such that U i U j   , define T i j = { y T i : P i ( y ) U j } .   There is defined a transition function f i j : T i j T j i   given for y T i j   by the formula f i j ( y ) = y 1 ,   where y 1 T j i   corresponds to the unique plaque P j ( y 1 )   , for which P i ( y ) P j ( y 1 ) .   The holonomy pseudogroup Γ T   induced by   on T   coincides with the pseudogroup generated by the maps f i j   .
Definition 2.21 (cf., for instance, [92). A transverse structure on a foliation   is a structure on a complete transversal T   , invariant under the action of the holonomy pseudogroup Γ T   .
Using the notion of transverse structure, one can single out classes of foliations with specific transverse properties. For instance, if a complete transversal T   is equipped with a Riemannian metric, and the holonomy pseudogroup Γ T   consists of all local isometries of this Riemannian metric, we get a class of Riemannian foliations (see Section  2.5 ). Similarly, if a complete transversal T   is equipped with a symplectic structure, and the holonomy pseudogroup Γ T   consists of all local diffeomorphisms, preserving this symplectic structure, we get a class of symplectic foliations (see Section  2.6 ).
One can also consider Kaehler foliations, measurable foliations and so on.

2.3 Transverse measures

Before we turn to the discussion of an analogue of the notion of measure on the leaf space of a foliation, we recall some basic facts, concerning to densities and integration of densities (cf., for instance, [31, 86).
Definition 2.22. Let L   be an n   -dimensional linear space and ( L )   the set of bases in L   . An α   -density on L   ( α R   ) is a function ρ : ( L ) C   such that, for any A = ( A i j ) G L ( n , C )   and e = ( e 1 , e 2 , , e n ) ( L )   , ρ ( e A ) = | det A | α ρ ( e ) ,   where ( e A ) i = j = 1 n e j A j i , i = 1 , 2 , , n   .
We will denote by | L | α   the space of all α   -densities on L   . For any vector bundle V   on M   , denote by | V | α   the associated bundle of α   -densities, | V | = | V | 1   .
For any smooth, compactly supported density ρ   on a smooth manifold M   there is a well-defined integral M ρ   , independent of the fact if M   is orientable or not. This fact allows to define a Hilbert space L 2 ( M )   , canonically associated with M   , which consists of square integrable half-densities on M   . The diffeomorphism group of M   acts on the space L 2 ( M )   by unitary transformations.
Definition 2.23. A (Borel) transversal to a foliation   is a Borel subset of M   , which intersects each leaf of the foliation in an at most countable set.
Definition 2.24. A transverse measure Λ   is a countably additive Radon measure, defined on the set of all transversals to the foliation.
Definition 2.25. A transverse measure Λ   is called holonomy invariant, if for any transversals B 1   and B 2   and any bijective Borel map φ : B 1 B 2   such that, for any x B 1   , the point φ ( x )   belongs to the leaf through the point x   , we have: Λ ( B 1 ) = Λ ( B 2 )   .
Example 2.26. Let us call by a transverse density any section of the bundle | τ |   . Since, for any smooth transversal T   , there is a canonical isomorphism T x T = τ x   , a continuous positive density ρ C ( M , | τ | )   determines a continuous positive density on T   , that determines a transverse measure. This transverse measure is holonomy invariant if and only if ρ   is invariant under the linear holonomy action.
Example 2.27. Any compact leaf L   of the foliation   defines a holonomy invariant transverse measure Λ   . For any transversal T   and for any set A T   , its measure Λ ( A )   equals the number of elements in A L   .
Let α C ( M , | T | )   be a smooth positive leafwise density on M   .
Starting from a transverse measure Λ   and the density α   , one can construct a Borel measure μ   on M   in the following way. Take a good cover { U i }   of M   by foliated coordinate neighborhoods with the corresponding coordinate maps φ i : U i I p × I q   and a partition of unity { ψ i }   subordinate to this cover. Consider the corresponding complete transversal T = i T i   , where T i = φ i 1 ( { 0 } × I q ) .   In any foliated chart ( U i , φ i )   , the transverse measure Λ   defines a measure Λ i   on T i   , and the smooth positive leafwise density α   defines a family { α i , y : y T i }   , where { α i , y }   is a smooth positive density on the plaque P i ( y )   . Observe that Λ   is holonomy invariant if and only if for any pair of indices i   and j   such that U i U j   , we have: f i j ( Λ i ) = Λ j   .
For any u C c ( M )   , put M u ( m ) d μ ( m ) = i T i P i ( y ) ψ i ( x , y ) u ( x , y ) d α i , y ( x ) d Λ i ( y ) .   One can show that this formula defines a measure μ   on M   , which is independent of the choice of a cover { U i }   and a partition of unity { ψ i }   .
A measure μ   on M   will be called holonomy invariant, if it is obtained by means of the above construction from a holonomy invariant transverse measure Λ   with some choice of a smooth positive leafwise density α   .
If we take in the above construction instead of the leafwise density u α   the restrictions to the leaves of an arbitrary differential p   -form ω   on M   , we obtain a well-defined functional C   on C c ( M , p T * M )   , called the Ruelle-Sullivan current [161, corresponding to Λ   : C , ω = i T i P i ( y ) ψ i ( x , y ) ω i , y ( x ) d Λ i ( y ) , ω C c ( M , p T * M ) ,   where ω i , y   is the restriction of ω   to the plaque P i ( y ) , y T i   .
A transverse measure Λ   is holonomy invariant, if and only if the corresponding Ruelle-Sullivan current C   is closed: C , d σ = 0 , σ C c ( M , p 1 T * M ) .  
Example 2.28. Suppose that a transverse measure Λ   is given by a smooth positive transverse density ρ C ( M , | τ | )   . Take a positive leafwise density α C ( M , | T | )   . Then the corresponding measure μ   on M   is given by the smooth positive density α ρ C ( M , | T M | )   , which corresponds to α   and ρ   under the canonical isomorphism | T M | = | T | | τ |   , defined by the short exact sequence 0 T T M τ 0   .
Example 2.29. Suppose that a holonomy invariant transverse measure Λ   is given by a compact leaf L   of the foliation   , and α C ( M , | T | )   is a smooth positive leafwise density on M   . Then the corresponding measure μ   on M   is the δ   -function along L   : M f ( x ) d μ ( x ) = L f ( x ) α ( x ) , f C c ( M ) .  
Example 2.30. Suppose that the foliation   is given by the orbits of a locally free action of a Lie group H   on the compact manifold M   and a smooth leafwise density α   is given by a fixed Haar measure d h   on H   . Then the corresponding measure μ   on M   is holonomy invariant if and only if it is invariant under the action of H   .

2.4 Connections

The infinitesimal expression of the holonomy on a foliated manifold is the canonical flat connection : X ( ) × C ( M , τ ) C ( M , τ )   in the normal bundle τ   , defined along the leaves of   (the Bott connection) [22, 23. It is given by
X N = θ ( X ) N = P τ [ X , N ~ ] , X X ( ) , N C ( M , τ ) , (2.1)
where N ~ C ( M , T M )   is any vector field on M   such that P τ ( N ~ ) = N   .
Thus, the restriction of τ   to any leaf of   is a flat vector bundle. The parallel transport in τ   along any leafwise path γ : x y   defined by   coincides with the linear holonomy map d h γ : τ x τ y   .
Definition 2.31. A connection : X ( M ) × C ( M , τ ) C ( M , τ )   in the normal bundle τ   is called adapted, if its restriction to X ( )   coincides with the Bott connection   .
One can construct an adapted connection, starting with an arbitrary Riemannian metric g M   on M   . Denote by g   the Levi-Civita connection, defined by g M   . An adapted connection   is given by
X N = P τ [ X , N ~ ] , X X ( ) , N C ( M , τ ) X N = P τ X g N ~ , X C ( M , F ) , N C ( M , τ ) , (2.2)
where N ~ C ( M , T M )   is any vector field such that P τ ( N ~ ) = N   . One can show that the adapted connection   described above has zero torsion.
The Bott connection   on τ   determines a connection ( ) *   on τ * = N *   by the formula
( ) X * ω ( N ) = X [ ω ( N ) ] ω ( X N ) , (2.3)
for any X T M , ω C ( M , τ * ) , N C ( M , τ )   .
Definition 2.32. An adapted connection   in the normal bundle τ   is called holonomy invariant, if, for any X X ( )   , Y X ( M )   and N C ( M , τ )   , we have ( θ ( X ) ) Y N = 0 ,   where, by definition, ( θ ( X ) ) Y N = θ ( X ) [ Y N ] θ ( X ) Y N Y [ θ Y N ] .   A holonomy invariant adapted connection in τ   is called a basic (or projectable) connection.
A fundamental property of basic connections is the fact that their curvature R   is a basic form, i.e. i X R = 0 , θ ( X ) R = 0   for any X X ( )   .
There are topological obstructions for the existence of basic connections for an arbitrary foliations found independently by Kamber and Tondeur and Molino (cf., for instance, [106, 134).

2.5 Riemannian foliations

Definition 2.33. A foliation ( M , )   is called Riemannian, if it has a transverse Riemannian structure. In other words, a foliation ( M , )   is called Riemannian, if there is a cover { U i }   of M   by foliated coordinate charts, φ i : U i I p × I q   , and Riemannian metrics g ( i ) ( y ) = α β g α β ( i ) ( y ) d y α d y β   , defined on the local bases I q   of   such that, for any coordinate transformation φ i j ( x , y ) = ( α i j ( x , y ) , γ i j ( y ) ) , ( x , y ) φ j ( U i U j ) ,   the map γ i j   preserves the metric on I q   , γ i j * ( g ( j ) ) = g ( i )   .
The class of Riemannian foliations was introduced in the papers of Reinhart [154, 155. There are several equivalent characterizations of Riemannian foliations.
Before we formulate the corresponding result (cf., for instance, [153,ChapterIV), we introduce some auxiliary notions.
Definition 2.34. A distribution on a Riemannian manifold is called totally geodesic, if every geodesic, which is tangent to the given distribution at some point, is tangent to it along the whole its length.
Definition 2.35. The second fundamental form of a distribution H   on a Riemannian manifold ( M , g )   is the tensor S   (which takes any vector X T x M   at a point x M   to a linear map S X : T x M T x M   ) given by
g ( S M N , X ) = 1 2 g ( M g N + N g M , X ) , g ( S M N , L ) = 0 , g ( S M X , Y ) = 0 , g ( S M X , N ) = g ( S M N , X ) , S X = 0 ,  
where X , Y X ( )   and L , M , N C ( M , F )   , g   denote the Levi-Civita connection determined by g   .
Definition 2.36. A map f : M B   of Riemannian manifolds M   and B   is called a Riemannian submersion, if the tangent map d f m : T m M T f ( m ) B   at any point m M   is surjective and induces an isometric map between the normal space T m M / T m f 1 ( f ( m ) )   to the level set f 1 ( f ( m ) )   of f   at m   and the tangent space T f ( m ) B   to B   at f ( m )   .
Theorem 2.37. A foliation ( M , )   is Riemannian if and only if there is a Riemannian metric g   on M   , satisfying any of the following equivalent conditions:
  • (1) The distribution H = F   is totally geodesic.
  • (2) The second fundamental form of H   vanishes.
  • (3) The induced metric g τ   on the normal bundle τ   is holonomy invariant: X g τ ( M , N ) = 0   for any X X ( )   and for any M , N C ( M , τ )   , where, by definition, X g τ ( M , N ) = X [ g τ ( M , N ) ] g τ ( X M , N ) g τ ( M , X N ) .  
  • (4) For any vector fields M   and N   , which are defined on an open set, are orthogonal to the leaves and are infinitesimal transformations of the foliation, and for any X X ( )   , we have X [ g ( M , N ) ] = 0 .  
  • (5) In any foliated chart φ : U I p × I q   with the local coordinates ( x , y )   , the restriction g H   of g   to H   is written in the form g H = α , β = 1 q g α β ( y ) θ α θ β ,   where θ α H *   is the 1-form, corresponding to the form d y α   under the isomorphism H * = T * R q   , and g α β ( y )   depend only on the transverse variables y R q   .
  • (6) ( M , )   locally has the structure of a Riemannian submersion, i. e., for any foliated chart φ : U I p × I q   , there exists a Riemannian metric on I q   such that the corresponding distinguished map p r n q φ : U I q   is a Riemannian submersion.
  • (7) The adapted connection   on the normal bundle τ   given by ( 2.2 ) is a (torsion-free) Riemannian connection: for any Y X ( M )   and M , N C ( M , τ )   Y [ g τ ( M , N ) ] = g τ ( Y M , N ) + g τ ( M , Y N ) .  
  • (8) The holonomy group of the adapted connection   on the normal bundle τ   given by ( 2.2 ) at any point preserves the metric.
Definition 2.38. Any Riemannian metric on M   , satisfying the equivalent conditions of Theorem  2.37 , is called bundle-like.
One can prove that a (torsion-free) Riemannian connection on the normal bundle τ   to a Riemannian foliation   is unique. It is uniquely determined by the transverse metric g τ   and is called the transverse Levi-Civita connection for   . Thus, the transverse Levi-Civita connection is an adapted connection.
Moreover, the transverse Levi-Civita connection turns out to be a holonomy invariant and, therefore, a basic connection. In particular, this shows the existence of a basic connection for any Riemannian foliation.
The existence of a bundle-like metric on a foliated manifold imposes strong restrictions on the geometry of the foliation. There are structure theorems for Riemannian foliations obtained by Molino (cf. [136, 135). Using these structure theorems, many questions, concerning to Riemannian foliations, can be reduced to the case of Lie foliations, that is, of foliations, whose transverse structure is modeled by a finite-dimensional Lie group (cf. Example  2.41 ).
Example 2.39. Any foliation defined by a submersion π : M B   is Riemannian.
Example 2.40. The orbits of a locally free isometric action of a Lie group on a Riemannian manifold define a Riemannian foliation.
On the other hand, flows, whose orbits form a Riemannian foliation, are called Riemannian flows. There are examples of Riemannian flows, which are not isometric. Concerning to Riemannian flows, see, for instance, [32, and also [137,AppendixA.
Example 2.41. Let M   be a smooth manifold, g   a real finite-dimensional Lie algebra and ω   an 1-form on M   with values in g   , satisfying the conditions:
  • (1) the map ω x : T x M g   is surjective for any x M   ;
  • (2) d ω + 1 2 [ ω , ω ] = 0   .
The distribution F x = ker ω x   is integrable and, therefore, defines a codimension q = dim g   foliation on M   . Such a foliation is called a Lie g   -foliation. The class of Lie foliations was introduced in [69.
Any Lie foliation is Riemannian. In the case g = R   , a Lie foliation is precisely a codimension one foliation given by a non-vanishing closed 1-form. Actually, it can be easily seen that a codimension one foliation is Riemannian if and only if it is given by a non-vanishing closed 1-form.
Example 2.42. A foliation   on a manifold M = B ~ × Γ F   , obtained by the suspension from a manifold B   and a homomorphism φ : Γ Diff ( F )   of the fundamental group Γ = π 1 ( B )   is Riemannian if and only if for any γ Γ   the diffeomorphism φ ( γ )   preserves a Riemannian metric on F   .
Let   be a transversely oriented Riemannian foliation and g   a bundle-like Riemannian metric. The induced metric on the normal bundle τ   yields the transverse volume form v τ C ( M , q τ * ) = C ( M , q N * )   , which is holonomy invariant and defines, therefore, a holonomy invariant transverse measure on   .

2.6 Symplectic foliations

Definition 2.43. A foliation ( M , )   is called symplectic, if it has a transverse symplectic structure. In other words, a foliation ( M , )   is called symplectic, if there is a cover { U i }   of M   by foliated coordinate charts, φ i : U i I p × I q   , and symplectic forms ω i   , defined on the local bases I q   of   such that, for any coordinate transformation φ i j ( x , y ) = ( α i j ( x , y ) , γ i j ( y ) ) , ( x , y ) φ j ( U i U j ) ,   the map γ i j   preserves the symplectic form, γ i j * ω j = ω i   .
Definition 2.44. A presymplectic manifold is a manifold equipped with a closed 2   -form of constant rank.
A transverse symplectic structure on ( M , )   uniquely determines a presymplectic structure ω   on M   such that T   coincides with the kernel of ω   . On the other hand, if ( M , ω )   is a presymplectic manifold, then the kernel of ω   determines an integrable distribution on M   and ω   induces a transverse symplectic structure on the corresponding foliation ( M , )   (cf. for instance, [16, 20, in [16symplectic foliations are called Hamiltonian). If ( M , )   is a symplectic foliation and ω   is the corresponding presymplectic structure on M   , then the q   -form q ω   defines the holonomy invariant transverse density | q ω | | q τ * |   , and, therefore, a holonomy invariant transverse measure, which can be naturally called the transverse Liouville measure.
Example 2.45. A foliation   on a manifold M = B ~ × Γ F   , obtained by the suspension from a manifold B   and a homomorphism φ : Γ Diff ( F )   of the fundamental group Γ = π 1 ( B )   is symplectic if and only if for any γ Γ   the diffeomorphism φ ( γ )   preserves a symplectic structure on F   .
Example 2.46. Recall (cf., for instance, [120) that a submanifold Σ   of a symplectic manifold X   is called coisotropic, if, for any σ Σ   , the skew-orthogonal complement ( T σ Σ )   of T σ Σ   is contained in T σ Σ   . If Σ   is a coisotropic submanifold, then the distribution ( T σ Σ )   is integrable, and the corresponding foliation Σ   is called the characteristic foliation of the coisotropic submanifold Σ   . It is well-known that there is a canonical symplectic structure on T σ Σ / ( T σ Σ )   , therefore, the foliation Σ   is symplectic. Moreover, if Σ   is simple, then the set Γ T * M × T * M   , which consists of all ( ν , ν ) T * M × T * M   such that ν   and ν   lie on the same leaf of Σ   is a canonical relation.
As shown in [20(see also [80), any presymplectic manifold can be obtained, using this construction, that is, as a coisotropic submanifold of a symplectic manifold. A particular example of the construction described above is the following one. For a foliated manifold ( M , )   , consider T * M   as a symplectic manifold with the standard symplectic structure.
Then N *   is a coisotropic submanifold in T * M   . The corresponding characteristic foliation N   is the natural lift of   to the conormal bundle and is called a horizontal (or linearized) foliation. Thus, the linearized foliation N   is symplectic. The coordinate chart φ n : N * I p × I q × R q   determined by a foliated coordinate chart φ   on M   (cf. Section  2.1 ) is a foliated chart for N   with plaques given by the level sets y = c o n s t , η = c o n s t   . Informally speaking, the leaf space N * / N   of N   can be considered as the cotangent bundle to the leaf space M /   of   .

2.7 Differential operators

Let ( M , )   be a compact foliated manifold and E   a smooth vector bundle on M   (unless otherwise is stated, we will assume that vector bundles under consideration are smooth and complex). We start with some general definitions, concerning to differential operators on M   .
Definition 2.47. A linear differential operator A   of order μ   , acting in C ( M , E )   , is called a tangential differential operator, if, in any foliated chart φ : U M I p × I q   and any trivialization of E   over it, A   is of the form
A = | α | μ a α ( x , y ) D x α , ( x , y ) I p × I q , (2.4)
where a α   are matrix-valued function on I p × I q   , D x = 1 i x   .
Definition 2.48. For a tangential differential operator A   given by ( 2.4 ) in some foliated chart φ : U M I p × I q   and a trivialization of E   over it, define its tangential (complete) symbol σ ( x , y , ξ ) = | α | μ a α ( x , y ) ξ α , ( x , y ) I p × I q , ξ R p ,   and its tangential principal symbol σ μ ( x , y , ξ ) = | α | = μ a α ( x , y ) ξ α , ( x , y ) I p × I q , ξ R p .  
The tangential principal symbol is invariantly defined as a section of the bundle ( π F * E )   on T *   (where π F : T * M   is the natural projection).
Definition 2.49. A tangential differential operator A   is called tangentially elliptic, if its tangential principal symbol σ μ   is invertible for ξ 0   .
Let D m ( M , E )   denote the set of all differential operators of order m   and D μ ( , E )   denote the set of all tangential differential operators of order μ   , acting in C ( M , E )   .
Introduce classes D m , μ ( M , , E )   , which are linearly generated by arbitrary compositions of tangential differential operators of order μ   and differential operators of order m   on M   . In other words, an operator A D m , μ ( M , , E )   is of the form A = α B α C α   , where B α D m ( M , E )   , C α D μ ( , E )   . If A 1 D m 1 , μ 1 ( M , , E )   , A 2 D m 2 , μ 2 ( M , , E )   , then A 1 A 2 D m 1 + m 2 , μ 1 + μ 2 ( M , , E )   and, if A D m , μ ( M , , E )   , then A * D m , μ ( M , , E )   . The classes D m , μ ( M , , E )   can be extended to classes Ψ m , μ ( M , , E )   , which contain, for instance, parametrices for elliptic operators of class D m , μ ( M , , E )   (cf. [113).
We will use the standard classes of pseudodifferential operators Ψ k ( M , E )   (for the theory of pseudodifferential operators see, for instance, [101, 175, 179, 167). Recall that a pseudodifferential operator on M   is a linear operator P : C ( M ) D ( M )   , which can be represented in a coordinate patch X R n   as P u ( x ) = e ( x y ) ξ p ( x , ξ ) u ( y ) d y d ξ , x X ,   where u C c ( X )   , p ( x , ξ ) S m ( X × R n )   is the complete symbol of P   .
Usually, we will assume that the complete symbol p   can be represented as an asymptotic sum p p m + p m 1 +   , where p k   is homogeneous of degree k   in ξ   for | ξ | > 1   . The principal symbol p m   of P   is well-defined as a section of the bundle ( π * E )   on T ~ * M = T * M \ { 0 }   , where π : T * M M   is the natural projection.
Definition 2.50. The transversal principal symbol σ P   of an operator P Ψ m ( M , E )   is the restriction of its principal symbol p m   to N ~ * = N * \ { 0 }   .
Definition 2.51. A operator P Ψ m ( M , E )   is called transversally elliptic, if its transversal principal symbol σ P ( ν )   is invertible for any ν N ~ *   .
Now suppose that a closed foliated manifold ( M , )   is equipped with a Riemannian metric g M   . Let H   be the orthogonal complement of F = T   .
Thus, there is a decomposition of T M   into the direct sum T M = F H   and the corresponding bigrading of the exterior power bundle Λ * T * M   :
Λ k T * M = i + j = k Λ i , j T * M , Λ i , j T * M = Λ i H * Λ j F * .   There is (cf., for instance, [15,Proposition10.1,[177) the corresponding decomposition of the de Rham differential d   into the sum of bigraded components of the form
d = d F + d H + θ . (2.5)
Here
  • (1) d F = d 0 , 1 : C ( M , Λ i , j T * M ) C ( M , Λ i , j + 1 T * M )   is the tangential de Rham differential, which is a first order tangentially elliptic operator, independent of the choice of g   ;
  • (2) d H = d 1 , 0 : C ( M , Λ i , j T * M ) C ( M , Λ i + 1 , j T * M )   is the transversal de Rham differential, which is a first order transversally elliptic operator;
  • (3) θ = d 2 , 1 : C ( M , Λ i , j T * M ) C ( M , Λ i + 2 , j 1 T * M )   is a zero order differential operator, which is the contraction operator by the 2-form θ   on M   with values in F   , θ C ( M , F Λ 2 τ * )   , given by θ ( X , Y ) = p F ( [ X , Y ] ) , X , Y C ( M , H ) ,   where P F : T M F   is the natural projection. In particular, θ   vanishes if and only if H   is integrable.
There is a similar decomposition for the adjoint:
δ = δ F + δ H + θ * , (2.6)
where δ F   , δ H   and θ *   are the adjoints of d F   , d H   and θ   in the Hilbert space L 2 ( M , Λ T * M )   accordingly.
The Laplace operator Δ g = d δ + δ d   of the metric g M   can be written in the form
Δ g = Δ F + Δ H + Δ 1 , 2 + K 1 + K 2 + K 3 , (2.7)
where
  •   Δ F = d F δ F + δ F d F D 0 , 2 ( M , , Λ T * M )   is the tangential Laplacian.
  •   Δ H = d H δ H + δ H d H D 2 , 0 ( M , , Λ T * M )   is the transversal Laplacian.
  •   Δ 1 , 2 = θ θ * + θ * θ D 0 , 0 ( M , , Λ T * M )   .
  •   K 1 = d F δ H + δ H d F + δ F d H + d H δ F D 1 , 0 ( M , , Λ T * M )   .
  •   K 2 = d F θ * + θ * d F + δ F θ + θ δ F D 0 , 0 ( M , , Λ T * M ) .  
  •   K 3 = d H θ * + θ * d H + δ H θ + θ δ H D 1 , 0 ( M , , Λ T * M )   .
We also introduce the first order differential operator D H = d H + d H *   in C ( M , H * )   , which is called the transverse signature operator.
A basic property of geometric operators on manifolds equipped with Riemannian foliation is that, if   is a Riemannian foliation and g M   is a bundle-like metric, then the operators d F δ H + δ H d F   and δ F d H + d H δ F   belong to D 0 , 1 ( M , , Λ T * M )   . In particular, K 1 D 0 , 1 ( M , , Λ T * M )   .

3 Operator algebras of foliations

In this Section, we will describe the noncommutative algebras associated with the leaf space of a foliation. First, we will define an algebra, consisting of very nice functions, on which all basic operators of analysis are defined, then, depending on a problem in question, we will complete this algebra and obtain an analogue of the algebra of measurable, continuous or smooth functions. The role of a “nice” algebra is played by the algebra C c ( G )   of smooth compactly supported functions on the holonomy groupoid G   of the foliation. Therefore, we start with the notion of holonomy groupoid of a foliation.

3.1 Holonomy groupoid

A foliation   defines an equivalence relation M × M   on M   : ( x , y )   if and only if x   and y   lie on the same leaf of the foliation   . Generally,   is not a smooth manifold, but one can resolve its singularity, constructing a smooth manifold G   , called the holonomy groupoid or the graph of the foliation, which “almost every where” coincides with   and which can be used in many cases as a substitution for   . The idea of the holonomy groupoid appeared in the papers of Ehresmann, Reeb and Thom (cf. [67, 176) and was completely realized by Winkelnkemper [184. First of all, we give the general definition of a groupoid (see [156, 137, 31, 126, 146for groupoids and related subjects).
Definition 3.1. We say that a set G   has the structure of a groupoid with the set of units G ( 0 )   , if there are defined maps
  •   Δ : G ( 0 ) G   (the diagonal map or the unit map);
  •   an involution i : G G   called the inversion and written as i ( γ ) = γ 1   ;
  •   a range map r : G G ( 0 )   and a source map s : G G ( 0 )   ;
  •   an associative multiplication m : ( γ , γ ) γ γ   defined on the set G ( 2 ) = { ( γ , γ ) G × G : r ( γ ) = s ( γ ) } ,  
satisfying the conditions
  •   r ( Δ ( x ) ) = s ( Δ ( x ) ) = x   and γ Δ ( s ( γ ) ) = γ   , Δ ( r ( γ ) ) γ = γ   ;
  •   r ( γ 1 ) = s ( γ )   and γ γ 1 = Δ ( r ( γ ) )   .
Alternatively, one can define a groupoid as a small category, where each morphism is an isomorphism.
It is convenient to think of an element γ G   as an arrow γ : x y   , going from x = s ( γ )   to y = r ( γ )   .
We will use the standard notation (for x , y G ( 0 )   ):
  •   G x = { γ G : r ( γ ) = x } = r 1 ( x )   ,
  •   G x = { γ G : s ( γ ) = x } = s 1 ( x )   ,
  •   G y x = { γ G : s ( γ ) = x , r ( γ ) = y }   .
Definition 3.2. A groupoid G   is called smooth (or a Lie groupoid), if G ( 0 )   , G   and G ( 2 )   are smooth manifolds, r   , s   , i   and m   are smooth maps, r   and s   are submersions, and Δ   is an immersion.
Example 3.3. Lie groups. A Lie group H   defines a smooth groupoid as follows: G = H   , G ( 0 )   consists of a single point, the maps i   and m   are given by the group operations in H   .
Example 3.4. Trivial groupoid. Let X   be an arbitrary set. Put G = X   , G ( 0 ) = X   , the maps s   and r   are the identity maps (that is, in other words, each element x G ( 0 ) = X   is identified with an unique element γ : x x   ).
Example 3.5. Equivalence relations. Any equivalence relation R X × X   defines a groupoid, if we put G ( 0 ) = X   , G = R   , the maps s : R X   and r : R X   are given by s ( x , y ) = y   , r ( x , y ) = x   . Thus, pairs ( x 1 , y 1 )   and ( x 2 , y 2 )   can be multiplied if and only if y 1 = x 2   , and in this case ( x 1 , y 1 ) ( x 2 , y 2 ) = ( x 1 , y 2 )   .
In the particular case of R = X × X   , we get a so called principal or pair groupoid.
Example 3.6. Group actions. Let a Lie group H   act smoothly from the left on a smooth manifold X   . The crossed product groupoid X H   is defined as follows: G ( 0 ) = X   , G = X × H   . The maps s : X × H X   and r : X × H X   have the form s ( x , h ) = h 1 x   , r ( x , h ) = x   . Thus, pairs ( x 1 , h 1 )   and ( x 2 , h 2 )   can be multiplied if and only if x 2 = h 1 1 x 1   , and in this case ( x 1 , h 1 ) ( x 2 , h 2 ) = ( x 1 , h 1 h 2 )   .
Example 3.7. Fundamental groupoid (cf. for instance, [169). Let X   be a topological space, G = Π ( X )   the set of homotopy classes of paths in X   with all possible endpoints. More precisely, if γ : [ 0 , 1 ] X   is a path from x = γ ( 0 )   to y = γ ( 1 )   , then we denote by [ γ ]   the homotopy class of γ   with fixed x   and y   . Define the groupoid Π ( X )   as the set of triples ( x , [ γ ] , y )   , where x , y X   , γ   is a path with the initial point x = γ ( 0 )   and the final point y = γ ( 1 )   , where the multiplication is given by the product of paths. The groupoid Π ( X )   is called the fundamental groupoid of X   .
Example 3.8. The Haefliger groupoid Γ n   [23, 89, 91. Let M   be a smooth manifold. A groupoid Γ M   consists of the germs of local diffeomorphisms of M   at arbitrary points of M   . ( Γ M ) ( 0 ) = M   . If γ Γ M   is the germ at x M   of a diffeomorphism f   from some neighborhood U   of x   on an open set f ( U )   , then s ( γ ) = x   , r ( γ ) = f ( x )   .
The multiplication in Γ M   is given by the composition of maps. If M = R n   , then the groupoid Γ M   is denoted by Γ n   .
The holonomy groupoid G = G ( M , )   of a foliated manifold ( M , )   is defined in the following way. Let h   be an equivalence relation on the set of continuous leafwise paths γ : [ 0 , 1 ] M   , setting γ 1 h γ 2   , if γ 1   and γ 2   have the same initial and final points and the same holonomy maps:
h γ 1 = h γ 2   . The holonomy groupoid G   is the set of h   -equivalence classes of leafwise paths. The set of units G ( 0 )   is a manifold M   . The multiplication in G   is given by the product of paths. The corresponding range and source maps s , r : G M   are given by s ( γ ) = γ ( 0 )   and r ( γ ) = γ ( 1 )   . Finally, the diagonal map Δ : M G   takes any x M   to the element in G   given by the constant path γ ( t ) = x , t [ 0 , 1 ]   . To simplify the notation, we will identify x M   with Δ ( x ) G   .
For any x M   the map s   maps G x   on the leaf L x   through x   . The group G x x   coincides with the holonomy group of L x   . The map s : G x L x   is the covering map associated with the group G x x   , called the holonomy covering. One can also introduce the holonomy groupoid G ( L )   of a leaf L   of   as the set of h   -equivalence classes of piecewise smooth paths in L   .
The holonomy groupoid G   has the structure of a smooth (in general, non-Hausdorff and non-paracompact) manifold of dimension 2 p + q   . Recall the construction of an atlas on G   [34.
Let φ : U I p × I q , φ : U I p × I q   be two foliated charts, π = p r n q φ : U R q   , π = p r n q φ : U R q   the corresponding distinguished maps. The foliated charts φ   , φ   are called compatible, if, for any m U   and m U   with π ( m ) = π ( m )   , there is a leafwise path γ   from m   to m   such that the corresponding holonomy map h γ   takes the germ π m   of π   at m   to the germ π m   of π   at m   .
For any pair of compatible foliated charts φ   and φ   denote by W ( φ , φ )   the subset in G   , consisting of all γ G   from s ( γ ) = m = φ 1 ( x , y ) U   to r ( γ ) = m = φ 1 ( x , y ) U   such that the corresponding holonomy map h γ   takes the germ π m   of the map π = p r n q φ   at m   to the germ π m   of the map π = p r n q φ   at m   . There is a coordinate map
Γ : W ( φ , φ ) I p × I p × I q , (3.1)
which takes each element γ W ( φ , φ )   such that s ( γ ) = m = ϰ 1 ( x , y )   , r ( γ ) = m = φ 1 ( x , y )   and h γ π m = π m   to the triple ( x , x , y ) I p × I p × I q   .
As shown in [34, the coordinate neighborhoods W ( φ , φ )   form an atlas of a ( 2 p + q )   -dimensional manifold (in general, non-Hausdorff and non-paracompact) on G   . Moreover, the groupoid G   is a smooth groupoid.
Non-Hausdorffness of the holonomy groupoid is related with the phenomenon of one-sided holonomy. The simplest example of a foliation with the non-Hausdorff holonomy groupoid is given by the trajectories of a nonsingular vector field on the plane, having a one-sided limit cycle. As shown in [184, the holonomy groupoid is Hausdorff if and only if the holonomy maps H T 0 T 1 ( γ 1 )   and H T 0 T 1 ( γ 2 )   along any leafwise paths γ 1   and γ 2   with the initial point x   and the final point y   , given by smooth transversals T 0   and T 1   , passing through x   and y   accordingly, coincide, if they coincide on some open subset U T 0   such that x U ¯   . In particular, the holonomy groupoid is Hausdorff, if the holonomy is trivial, or real analytic. Moreover, the holonomy groupoid of a Riemannian foliation is Hausdorff. In the sequel, we will always assume that G   is a Hausdorff manifold.
Example 3.9. If a simple foliation   is defined by a submersion π : M B   , then its holonomy groupoid G   consists of all ( x , y ) M × M   such that π ( x ) = π ( y )   , and, moreover, G ( 0 ) = M   , the maps s : G M   and r : G M   are given by s ( x , y ) = y   , r ( x , y ) = x   .
Example 3.10. If a foliation   is given by the orbits of a free smooth action of a connected Lie group H   on a manifold M   , then its holonomy groupoid coincides with the crossed product groupoid M H   .
Besides the holonomy groupoid, there are another groupoids, which can be associated with the foliation. First of all, it is the groupoid given by the equivalence relation on M   , setting points x   and y   to be equivalent, if they lie on the same leaf of the foliation (the coarse groupoid). As noted above, this groupoid is not smooth. One can also consider the fundamental groupoid of the foliation Π ( M , )   , which also consists of equivalence classes of leafwise paths, where two leafwise paths are called equivalent, if they are homotopic in the class of leafwise paths with fixed endpoints. The fundamental groupoid of the foliation Π ( M , )   is a smooth groupoid (cf., for instance, [149).
There is a foliation G   of dimension 2 p   on the holonomy groupoid G   . In any coordinate chart W ( φ , φ )   given by a pair of compatible foliated charts φ   and φ   , the leaves of G   are given by equations of the form y = c o n s t   . The leaf of G   through γ G   consists of all γ G   such that r ( γ )   and r ( γ )   lie on the same leaf of   and coincides with the holonomy groupoid of this leaf.
The holonomy group of a leaf of G   coincides with the holonomy group of the corresponding leaf of   . (The last statement corrects an erroneous one made in [184. This fact was noted, for instance, by Molino in his review of [184in Mathematical Reviews (see MR 85j:57043).) The differential of the map ( r , s ) : G M × M   maps isomorphically the tangent bundle T G   to G   to the bundle F F   on M × M   , therefore, there is a canonical isomorphism T G = r * F s * F   .
A distribution H   on M   transverse to   determines a distribution H G   on G   transverse to G   . For any X H y   , there is a unique vector X ^ T γ G   such that d s ( X ^ ) = d h γ 1 ( X )   and d r ( X ^ ) = X   , where d h γ : H x H y   is the linear holonomy map associated with γ   . The space H γ G   consists of all vectors of the form X ^ T γ G   for different X H y   . In any coordinate chart W ( φ , φ )   on G   , the tangent space T γ G   to G   at some γ   with the coordinates ( x , x , y )   consists of vectors of the form X x + X x   and the distribution H γ G   consists of vectors X x + X x + Y y   such that X x + Y y H ( x , y )   and X x + Y y H ( x , y )   .
Let g M   be a Riemannian metric on M   and H = F   . Then a Riemannian metric g G   on G   is defined as follows. All the components in T γ G = F y F x H γ G   are mutually orthogonal, and, by definition, g G   coincides with g M   on F y H γ G = F y H y = T y M   and with g F   on F x   .
If   is Riemannian and g M   is a bundle-like Riemannian metric, then g G   is bundle-like, and, therefore, G   is Riemannian. Moreover, in this case the maps s : G M   and r : G M   are Riemannian submersions and locally trivial fibrations. In particular, the holonomy coverings G x   of leaves of   are diffeomorphic [184.

3.2 The C *   -algebra of a foliation and noncommutative topology

In this Section, we will describe the construction of the C *   -algebra associated with an arbitrary foliation, which is an analogue of the algebra of continuous functions on the leaf space of the foliation.
Definition 3.11. (cf., for instance, [62, 141, 147, 174) A C *   -algebra is an involutive Banach algebra A   such that a * a = a 2 , a A .  
Example 3.12. The simplest example of a C *   -algebra is given by the algebra C 0 ( X )   of continuous functions on a locally compact Hausdorff topological space X   , vanishing at the infinity, which is endowed with operations of the pointwise addition and the multiplication, with the standard involution and with the uniform norm f = sup x X | f ( x ) | , f C 0 ( X ) .   The Gelfand-Naimark theorem allows to reconstruct uniquely from a commutative C *   -algebra A   the locally compact Hausdorff topological space X   such that A = C 0 ( X )   . More precisely, X   coincides with the set A ^   of all characters of the algebra A   , i. e., of all continuous homomorphisms A C   , endowed with the topology of pointwise convergence.
The previous example permits to consider an arbitrary C *   -algebra as the algebra of continuous functions on some virtual space. By this reason, the theory of C *   -algebras is often called as noncommutative topology.
Example 3.13. The algebra ( H )   of bounded operators in a Hilbert space H   equipped with the involution given by taking the adjoints and with the operator norm is a C *   -algebra.
By the second Gelfand-Naimark theorem, any C *   -algebra is isometrically *   -isomorphic to some norm closed *   -subalgebra of the algebra ( H )   for some Hilbert space H   .
There are two ways to define the C *   -algebras associated with a foliation.
The first way makes use of the auxiliary choices of a smooth Haar system, the second one requires no auxiliary choices and uses the language of half-densities.

3.2.1 Definitions, using a Haar system.

In this Section we give the definition of the C *   -algebras associated with an arbitrary smooth groupoid G   . In fact, the assumption of smoothness of a groupoid is not essential here, and all the definitions can be generalized to the case of topological groupoids (cf., for instance, [156).
We will only consider Hausdorff groupoids. For the definition of the C *   -algebra of a foliation in the case when the holonomy groupoid is non-Hausdorff, cf. [37(and also [57, 58, 60).
Definition 3.14. A smooth Haar system on a smooth groupoid G   is a family of positive Radon measures { ν x : x G ( 0 ) }   on G   , satisfying the following conditions:
  • (1) The support of the measure ν x   coincides with G x   , and ν x   is a smooth measure on G x   .
  • (2) The family { ν x : x G ( 0 ) }   is left-invariant, that is, for any continuous function f C ( G x ) , f 0 ,   and for any γ G , s ( γ ) = x , r ( γ ) = y ,   we have G y f ( γ 1 ) d ν y ( γ 1 ) = G x f ( γ γ 1 ) d ν x ( γ 1 ) .  
  • (3) The family { ν x : x G ( 0 ) }   is smooth, that is, for any φ C c ( G )   the function x G ( 0 ) G x φ ( γ ) d ν x ( γ )   is a smooth function on G ( 0 )   .
For a compact foliated manifold ( M , )   , there is a distinguished class of smooth Haar systems { ν x : x G ( 0 ) }   on the holonomy groupoid G   of   given by smooth positive leafwise densities α C ( M , | T | )   . The positive Radon measure ν x   on G x , x M ,   is given as the lift of the density α   via the holonomy covering s : G x M   . In the sequel, we will assume that a Haar system on G   constructed in such a way is fixed.
Let G   be a smooth groupoid, G ( 0 ) = M   and { ν x : x M }   a smooth Haar system. Introduce the structure of an involutive algebra on C c ( G )   by
k 1 * k 2 ( γ ) = G x k 1 ( γ 1 ) k 2 ( γ 1 1 γ ) d ν x ( γ 1 ) , γ G x ,
k * ( γ ) = k ( γ 1 ) ¯ , γ G .
For any x M   , there is a natural representation of C c ( G )   in the Hilbert space L 2 ( G x , ν x )   given, for k C c ( G )   and ζ L 2 ( G x , ν x )   , by R x ( k ) ζ ( γ ) = G x k ( γ 1 γ 1 ) ζ ( γ 1 ) d ν x ( γ 1 ) , r ( γ ) = x .   The completion of the involutive algebra C c ( G )   in the norm k = sup x R x ( k )   is called the reduced C *   -algebra of the groupoid G   and denoted by C r * ( G )   .
There is also defined the full C *   -algebra of the groupoid C * ( G )   , which is the completion of C c ( G )   in the norm k max = sup π ( k ) ,   where supremum is taken over the set of all *   -representations π   of the algebra C c ( G )   in Hilbert spaces.
Any k C c ( G )   defines a bounded operator R ( k )   in C c ( M )   . For any u C c ( M )   , we have R ( k ) u ( x ) = G x k ( γ ) u ( s ( γ ) ) d ν x ( γ ) , x M .   The correspondence k R ( k )   defines a representation of the algebra C c ( G )   in C c ( M )   .
Example 3.15. Lie groups. If a groupoid G   is given by a Lie group H   , then a smooth Haar system on G   is given by a left-invariant Haar measure d h   on H   , and the multiplication in C c ( G )   is the convolution operation defined for any u , v C c ( H )   by ( u * v ) ( g ) = H u ( h ) v ( h 1 g ) d h , g H ,   the involution is given by u * ( g ) = u ( g 1 ) ¯ , u C c ( H ) ,   and the operator algebras C r * ( G )   and C * ( G )   are the group C *   -algebras C r * ( H )   and C * ( H )   (cf., for instance, [147).
Example 3.16. Trivial groupoid. Let X   be a smooth manifold.
Put G = X   , G ( 0 ) = X   , the maps s   and r   are the identity maps. The operator algebras C r * ( G )   and C * ( G )   coincide with the commutative C *   -algebra C 0 ( X )   .
Example 3.17. Principal groupoid. Let X   be a smooth manifold.
Put G ( 0 ) = X   , G = X × X   . The maps s : X × X X   and r : X × X X   are given by s ( x , y ) = y   , r ( x , y ) = x   . If we choose a Haar system on G   , taking as ν x   a fixed smooth measure μ   in each G x = X   , then the operations in C c ( G )   are given by
( k 1 * k 2 ) ( x , y ) = X k 1 ( x , z ) k 2 ( z , y ) d μ ( z ) , ( x , y ) X × X ,
k * ( x , y ) = k ( y , x ) ¯ , ( x , y ) X × X ,
where k , k 1 , k 2 C c ( G )   . Thus, elements of C c ( G )   can be considered as the kernels of integral operators in C ( X )   . The representation k R x ( k )   takes each k C c ( G ) C ( X × X )   to the integral operator in L 2 ( G x , ν x ) = L 2 ( X , μ )   with the integral kernel k   : R x ( k ) u ( y ) = X k ( y , z ) u ( z ) d μ ( z ) , u L 2 ( X , μ ) .   Finally, the operator algebras C r * ( G )   and C * ( G )   coincide with the algebra C 0 ( X ) K ( L 2 ( X , μ ) )   .
Example 3.18. Group actions. Let a Lie group H   act smoothly from the left on a smooth manifold X   . Consider the corresponding crossed product groupoid G = X H   . Then G x = H   for any x X   and a smooth Haar system on G = X H   is given by a left-invariant Haar measure d h   on H   . The multiplication in C c ( G )   is defined for any u , v C c ( X × H )   by ( u * v ) ( x , g ) = H u ( x , h ) v ( h 1 x , h 1 g ) d h , ( x , g ) X × H ,   the involution is defined for u C c ( X × H )   by u * ( x , g ) = u ( g 1 x , g 1 ) ¯ , ( x , g ) X × H .   The operator algebras C r * ( G )   and C * ( G )   associated with the crossed product groupoid G = X H   coincide with the crossed products C 0 ( X ) r H   and C 0 ( X ) H   of the algebra C 0 ( X )   by the group H   with respect to the induced action of H   on C 0 ( X )   (cf., for instance, [147).
If the group H   is discrete, elements of C c ( G )   are families { a γ C ( X ) : γ H }   such that a γ 0   for finitely many elements γ   . It is convenient to write them as a = γ H a γ U γ   . The multiplication in C c ( G )   is written as ( a γ 1 U γ 1 ) ( b γ 2 U γ 2 ) = ( a γ 1 T γ 1 ( b γ 2 ) ) U γ 1 γ 2 ,   where T γ   denotes the operator in C 0 ( X )   induced by the action of γ H   : T γ f ( x ) = f ( γ 1 x ) , x X , f C 0 ( X ) .   The involution in the algebra C c ( G )   is given by ( a γ U γ ) * = a ¯ γ U γ 1 .  
Let G   be the holonomy groupoid of a foliation   on a compact manifold M   . Elements of the algebra C c ( G )   can be considered as families of the kernels of integral operators along the leaves of the foliation (more precisely, on the holonomy coverings G x   ). Namely, each k C c ( G )   corresponds to the family { R x ( k ) : x M }   , where R x ( k )   is the integral operator in L 2 ( G x , ν x )   given by the integral kernel K ( γ 1 , γ 2 ) = k ( γ 1 1 γ 2 ) , γ 1 , γ 2 G x .   The product of elements k 1   and k 2   from C c ( G )   corresponds to the composition of integral operators { R x ( k 1 ) R x ( k 2 ) : x M }   . Finally, the representation R   means the natural action of such families of leafwise integral operators in L 2 ( M )   . In this case, the algebra C r * ( G )   will be often called the reduced C *   -algebra of the foliation and denoted by C * ( M , )   .
Example 3.19. Consider the simplest example of a foliation, given by the linear foliation on the torus. Thus, suppose that M = T 2 = R 2 / Z 2   is the two-dimensional torus and a foliation θ   is given by the trajectories of the vector field X = x + θ y ,   where θ R   is a fixed irrational number.
Since this foliation is given by the orbits of a free group action of R   on T 2   , its holonomy groupoid coincides with the crossed product groupoid T 2 R   . Thus, G = T 2 × R   , G ( 0 ) = T 2   , s ( x , y , t ) = ( x t , y θ t )   , r ( x , y , t ) = ( x , y )   , ( x , y ) T 2 , t R ,   and the multiplication is given by ( x 1 , y 1 , t 1 ) ( x 2 , y 2 , t 2 ) = ( x 1 , y 1 , t 1 + t 2 ) ,   if x 2 = x 1 t 1 , y 2 = y 1 θ t 1   .
The C *   -algebra C r * ( G )   of the linear foliation on T 2   coincides with the crossed product C ( T 2 ) r R   . Therefore, the product k 1 * k 2   of k 1 , k 2 C c ( T 2 × R ) C ( T 2 ) r R   is given by
( k 1 * k 2 ) ( x , y , t ) = k 1 ( x 1 , y 1 , t 1 ) k 2 ( x t 1 , y θ t 1 , t t 1 ) d t 1 , ( x , y ) T 2 , t R ,  
and, for any k C c ( T 2 × R )   , k * ( x , y , t ) = k ( x t , y θ t , t ) ¯ , ( x , y ) T 2 , t R .   For any k C c ( T 2 × R )   and for any ( x , y ) T 2   , the operator R ( x , y ) ( k )   in L 2 ( G ( x , y ) ) = L 2 ( R )   has the form: for any u L 2 ( R )   R ( x , y ) ( k ) u ( t ) = k ( x t 1 , y θ t 1 , t t 1 ) u ( t 1 ) d t 1 , t R .   Finally, for any k C c ( T 2 × R )   , the corresponding operator R ( k )   in L 2 ( T 2 )   is given by R ( k ) u ( x , y ) = k ( x , y , t ) u ( x t , y θ t ) d t , ( x , y ) T 2 .   If θ   is rational, then the linear foliation on T 2   is given by the orbits of a free group action of S 1   on T 2   , and its holonomy groupoid coincides with the crossed product groupoid G = T 2 S 1   .
We will also need the twisted version of the notion of the C *   -algebra of a foliation ( M , )   , with coefficients in a Hermitian vector bundle E   on M   . Denote by C c ( G , ( E ) )   the space of smooth, compactly supported sections of the vector bundle ( s * E ) * r * E   on G   . In other words, the value of k C c ( G , ( E ) )   at any γ G   is a linear map k ( γ ) : E s ( γ ) E r ( γ ) .   The structure of an involutive algebra on C c ( G , ( E ) )   is defined by analogous formulas
k 1 * k 2 ( γ ) = G x k 1 ( γ 1 ) k 2 ( γ 1 1 γ ) d ν x ( γ 1 ) , γ G , r ( γ ) = x ,
k * ( γ ) = k ( γ 1 ) * , γ G .
with the only difference that the product k 1 ( γ 1 ) k 2 ( γ 1 1 γ )   means here the composition of the linear maps k 2 ( γ 1 1 γ ) : E s ( γ ) E s ( γ 1 )   and k 1 ( γ 1 ) : E s ( γ 1 ) E r ( γ 1 )   and k ( γ 1 ) *   means the adjoint of k ( γ 1 ) : E r ( γ ) E s ( γ ) .   Consider the vector bundle E ~ = s * ( E )   on G   . Let E ~ x   be the restriction of the bundle E ~   to G x   , and L 2 ( G x , E ~ x )   the Hilbert space of L 2   -sections of the bundle E ~ x   , determined by the fixed Hermitian structure on E ~ x   and the measure ν x   . For any x M   , we introduce a representation R x   of the algebra C c ( G , ( E ) )   in L 2 ( G x , E ~ x )   , given, for k C c ( G , ( E ) )   , by
R x ( k ) ζ ( γ ) = G x k ( γ 1 γ 1 ) ζ ( γ 1 ) d ν x ( γ 1 ) , ζ L 2 ( G x , E ~ x ) . (3.2)
The completion of C c ( G , ( E ) )   in the norm k = sup x R x ( k )   is called the reduced (twisted) C *   -algebra of the foliation with coefficients in E   and denoted by C r * ( G , E )   or C * ( M , , E )   . Denote by C * ( G , E )   the full C *   -algebra of the foliation, defined in the same way as in the case of trivial coefficients.
Any k C c ( G , ( E ) )   defines an operator R E ( k )   in C ( M , E )   . For any u C ( M , E )   , we have
R E ( k ) u ( x ) = G x k ( γ ) u ( s ( γ ) ) d ν x ( γ ) , x M . (3.3)
The correspondence k R E ( k )   determines a representation of the algebra C c ( G , ( E ) )   in L 2 ( M , E )   .
Let us give some facts, which relate the structure of the reduced C *   -algebra of a foliation C * ( M , )   with the topology of   (for more details cf. [68, 99).
Theorem 3.20 ([68). Let ( M , )   be a foliated manifold.
  • (1) The C *   -algebra C * ( M , )   is simple if and only if   is minimal, i.e. every its leaf is dense in M   .
  • (2) The C *   -algebra C * ( M , )   is primitive if and only if   is (topologically) transitive, i.e. it has a leaf, dense in M   .
  • (3) If   is amenable in the sense that C * ( M , ) = C * ( G )   , the C *   -algebra C * ( M , )   has a representation, consisting of compact operators if and only if   has a compact leaf.
In the work [68, a description of the space of primitive ideals of the C *   -algebra C * ( M , )   is also given.

3.2.2 Definition, using half-densities

In this Section, we will give the definitions of the operator algebras associated with a foliated manifold, which make no choice of a Haar system.
For this, we will use the language of half-densities.
Let ( M , )   be a compact foliated manifold. Consider the vector bundle of leafwise half-densities | T | 1 / 2   on M   . Pull back | T | 1 / 2   to the vector bundles s * ( | T | 1 / 2 )   and r * ( | T | 1 / 2 )   on the holonomy groupoid G   , using the source map s   and the range map r   . Define a vector bundle | T G | 1 / 2   on G   as | T G | 1 / 2 = r * ( | T | 1 / 2 ) s * ( | T | 1 / 2 ) .   The bundle | T G | 1 / 2   is naturally identified with the bundle of leafwise half-densities on the foliated manifold ( G , G )   .
The structure of an involutive algebra on C c ( G , | T G | 1 / 2 )   is defined as
σ 1 * σ 2 ( γ ) = γ 1 γ 2 = γ σ 1 ( γ 1 ) σ 2 ( γ 2 ) , γ G ,
σ * ( γ ) = σ ( γ 1 ) ¯ , γ G ,
where σ , σ 1 , σ 2 C c ( G , | T G | 1 / 2 )   . The formula for σ 1 * σ 2   should be interpreted in the following way. If we write γ : x y , γ 1 : z y   and γ 2 : x z   , then
σ 1 ( γ 1 ) σ 2 ( γ 2 ) | T y | 1 / 2 | T z | 1 / 2 | T z | 1 / 2 | T x | 1 / 2
= | T y | 1 / 2 | T z | 1 | T x | 1 / 2 ,
and, integrating the | T z | 1   -component σ 1 ( γ 1 ) σ 2 ( γ 2 )   with respect to z M   , we get a well-defined section of the bundle r * ( | T | 1 / 2 ) s * ( | T | 1 / 2 ) = | T G | 1 / 2 .   In the case of nontrivial coefficients, taking values in an Hermitian vector bundle E   on M   , it is necessary to consider the space C c ( G , ( E ) | T G | 1 / 2 )   , where the structure of an involutive algebra is defined by the same formulas.
The formula ( 3.3 ) can be rewritten in the language of half-densities as follows. For any σ C c ( G , ( E ) | T G | 1 / 2 )   and u C ( M , E | T | 1 / 2 )   , the element R E ( σ ) u   of C ( M , E | T | 1 / 2 )   is given by R E ( σ ) u ( x ) = G x σ ( γ ) s * u ( γ ) , x M .   This formula should be interpreted as follows. We have s * u C c ( G , s * ( E | T | 1 / 2 ) )   and, hence, σ s * u C c ( G , r * E r * ( | T | 1 / 2 ) s * ( | T | ) )   . The integration of the component in s * ( | T | )   over G x   , i.e. with a fixed r ( γ ) = x M   , gives a well-defined section R E ( σ ) u   of E | T | 1 / 2   on M   .

3.3 Von Neumann algebras and noncommutative measure theory

The initial datum of noncommutative measure theory is a pair ( , φ )   , consisting of a von Neumann algebra   and a weight φ   on   .
Definition 3.21. A von Neumann algebra is an involutive subalgebra of the algebra ( H )   of bounded operators in a Hilbert space H   , closed in the weak operator topology.
Definition 3.22. A weight on a von Neumann algebra   is a function φ   , defined on the set +   of positive elements of   , with values in R ¯ + = [ 0 . + ]   , satisfying the conditions
φ ( a + b ) = φ ( a ) + φ ( b ) , a , b + , φ ( α a ) = α φ ( a ) , α R + , a + .  
A weight on a von Neumann algebra   is called a trace, if φ ( a * a ) = φ ( a a * ) , a + .  
Definition 3.23. A weight φ   on a von Neumann algebra   is called
  • (1) faithful, if, for any a +   , the identity φ ( a ) = 0   implies a = 0   ;
  • (2) normal, if, for any bounded increasing net { a α }   of elements from +   with the least upper bound a   , the following identity holds: φ ( a ) = sup α φ ( a α ) .  
  • (3) semifinite, if the linear span of the set { x + : φ ( x ) < }   is σ   -weakly dense in   .
Every von Neumann algebra has a faithful, normal, semifinite weight.
Example 3.24. The trace functional t r   on the von Neumann algebra ( )   of bounded linear operators in a Hilbert space   is a faithful, normal, semifinite trace.
Example 3.25. If a measure space X   is endowed with a σ   -finite measure μ   , then elements of L ( X , μ )   considered as multiplication operators in the Hilbert space L 2 ( X , μ )   form a von Neumann algebra.
Moreover, the identity φ ( f ) = X f ( x ) d μ ( x ) , f L ( X , μ ) ,   defines a faithful, normal, semifinite trace on L ( X , μ )   .
The foundations of the noncommutative integration theory for foliations were laid by Connes in [34. Let ( M , )   be a compact foliated manifold. Let ρ   be a strictly positive continuous transverse density, α   a strictly positive smooth leafwise density, and ν = s * α   the corresponding smooth Haar system.
The measures ρ   and α   can be combined to construct a measure μ   on M   .
Finally, the measure μ   and the Haar system ν   define a measure m   on G   : G f ( γ ) d m ( γ ) = M ( G x f ( γ ) d ν x ( γ ) ) d μ ( x ) , f C c ( G ) .   In the previous Section we defined the representation R x   of the involutive algebra C c ( G )   in the Hilbert space L 2 ( G x , ν x )   for any x M   . Consider a representation R   of the algebra C c ( G )   in L 2 ( G , m ) = M L 2 ( G x , ν x ) d μ ( x )   defined as the direct integral of the representations R x   : R = M R x d μ ( x ) .  
Definition 3.26. The von Neumann algebra W * ( M , )   of the foliation   is defined as the closure of the image of C c ( G )   under the representation R   in the weak operator topology of ( L 2 ( G , m ) )   .
Since the definition of the von Neumann algebra W * ( M , )   depends only on the class of the measure m   (i.e. the family of all sets of m   -measure zero), it is easy to see that this definition does not depend on the choice of ρ   and α   .
If we assume that the union of all leaves with nontrivial holonomy has measure zero, elements of W * ( M , )   can be considered as measurable families { T l : l M / }   , where T l   is a bounded operator in L 2 ( l )   for each leaf l   of   (see more details in [34, 33, 42).
A holonomy invariant measure μ   on M   defines a normal semi-finite trace t r μ   on the von Neumann algebra W * ( M , )   . For any bounded measurable function k   on G   , the value t r μ ( k )   is finite and is given by t r μ ( k ) = M k ( x ) d μ ( x ) .   The paper [34gives a description of weights on the von Neumann algebra W * ( M , )   . As explained in [33, the construction of [34can be interpreted as a correspondence between weights on W * ( M , )   and operator-valued densities on the leaf space M /   .
Example 3.27. Consider the linear foliation θ   on the torus T 2   , θ R   is a fixed irrational number. The Lebesgue measure μ = d x d y   on T 2   is a holonomy invariant measure. The value of the corresponding faithful normal semi-finite trace t r μ   on k C c ( T 2 × R ) W * ( T 2 , θ )   is given by t r μ ( k ) = T 2 k ( x , y , 0 ) d x d y .  
A von Neumann algebra is called a factor if its center consists of operators of the form λ I , λ C   .
Theorem 3.28. Let ( M , )   be a foliated manifold. A von Neumann algebra W * ( M , )   is a factor if and only if the foliation is ergodic, that is, any bounded measurable function, constant along the leaves of the foliation   , is constant on M   .
It is known that von Neumann algebras are classified in three classes:
type I, II and III. Any von Neumann algebra   is canonically represented as a direct sum I I I I I I   of von Neumann algebras, where I   , I I   and I I I   are von Neumann algebras of type I, II and III accordingly.
Theorem 3.29. Let ( M , )   be a foliated manifold. The von Neumann algebra W * ( M , )   is of:
  • (1) type I if and only if the leaf space is isomorphic to the standard Borel space.
  • (2) type II if and only if there is a holonomy invariant transverse measure and the algebra is not of type I.
  • (3) type III if and only if there is no holonomy invariant transverse measure.

3.4 C *   -modules and vector bundles

A noncommutative generalization of the notion of vector bundle (as a topological object) is the notion of Hilbert C *   -module. Hilbert C *   -modules are also natural generalizations of Hilbert spaces, which arise if we replace the field of scalars C   to an arbitrary C *   -algebra. The theory of Hilbert C *   -modules appeared in the papers [145, 157and has found many applications in the theory of operator algebras and its applications (see more detailed expositions of basic facts of this theory in [118, 127, 104, 183).
Definition 3.30. [145, 157Let B   be a C *   -algebra. A pre-Hilbert B   -module is a right B   -module X   equipped with a sesquilinear map (linear in the second argument) , X : X × X B   , satisfying the following conditions:
  • (1) x , x X 0   for any x X   ;
  • (2) x , x X = 0   if and only if x = 0   ;
  • (3) y , x X = x , y X *   for any x , y X   ;
  • (4) x , y b X = x , y X b   for any x , y X , b B   .
The map , X   is called a B   -valued inner product.
Let X   be a pre-Hilbert B   -module. It can be shown that the formula x X = x , x X 1 / 2   defines a norm on X   . If X   is complete in the norm X   , then X   is called a Hilbert C *   -module. In a general case, the action of B   and the inner product on X   are extended to its completion X ~   , making X ~   into a Hilbert C *   -module.
Example 3.31. If E   is a Hermitian vector bundle on a compact manifold X   , then the space of its continuous sections C ( X , E )   is a Hilbert module over the algebra C ( X )   of continuous functions on X   . The action of C ( X )   on C ( X , E )   is given by ( a s ) ( x ) = a ( x ) s ( x ) , x X ,   and the inner product s 1 , s 2 ( x ) = s 1 ( x ) , s 2 ( x ) E x , x X .  
Definition 3.32. A vector bundle E   on a foliated manifold ( M , )   is called holonomy equivariant, if there is given a representation T   of the holonomy groupoid G   of the foliation   in the fibers of E   , that is, for any γ G , γ : x y   , there is defined a linear operator T ( γ ) : E x E y   such that T ( γ 1 γ 2 ) = T ( γ 1 ) T ( γ 2 )   for any γ 1 , γ 2 G   with r ( γ 2 ) = s ( γ 1 )   .
A Hermitian vector bundle E   on a foliated manifold ( M , )   is called holonomy equivariant, if it is a holonomy equivariant vector bundle and the representation T   is unitary: T ( γ 1 ) = T ( γ ) *   for any γ G   .
For any holonomy equivariant vector bundle E M   , the action of the groupoid G   on E   defines a horizontal foliation E   on E   of the same dimension as the foliation   . The leaf of E   through a point v E   consists of all points of the form T ( γ ) 1 ( v )   with γ G   , r ( γ ) = π ( v )   . Thus, any holonomy equivariant vector bundle is foliated in the sense of [107.
Definition 3.33. A vector bundle p : P M   is called foliated, if there is a foliation ¯   on P   of the same dimension as   such that its leaves are transversal to the fibers of p   and are mapped by p   to the leaves of   .
Equivalently, one can say that a foliated vector bundle is a vector bundle P   on M   , which has a flat connection in the space C ( M , P )   defined along the leaves of   , that is, an operator : X ( ) × C ( M , P ) C ( M , P ) ,   satisfying, for any f C ( M ) , X X ( ) , s C ( M , P )   , the standard conditions f X = f X , X ( f s ) = ( X f ) s + f X s ,   and also the flatness condition [ X , Y ] = [ X , Y ] , X , Y X ( ) .   The parallel transport along leafwise paths associated with the connection   defines an action of the fundamental groupoid of the foliation Π ( M , )   in the fibers of the foliated vector bundle P   . In general, the parallel transport may depend on the holonomy of the corresponding path, therefore, this action does not necessarily pull down to an action of the holonomy groupoid in the fibers of P   , and, therefore, a foliated vector bundle is not necessarily holonomy equivariant.
Let E   be a holonomy equivariant vector bundle. The holonomy groupoid G E   of the horizontal foliation ( E , E )   is described as follows:
G E = r * ( E ) = { ( γ , v ) G × E : r ( γ ) = π ( v ) } ,   ( G E ) ( 0 ) = E   , the source map s : G E E   is given by s ( γ , v ) = T ( γ ) 1 ( v )   , the range map r : G E E   by r ( γ , v ) = v   and the composition has the form ( γ , v ) ( γ , v ) = ( γ γ , v )   , where v = T ( γ ) 1 ( v )   .
Example 3.34 (Normal bundle). The normal bundle τ x = T x M / T x , x M ,   is a holonomy equivariant vector bundle, if it equipped with the action of the holonomy groupoid G   by the linear holonomy map d h γ : τ x τ y , γ : x y   . The corresponding partial flat connection defined along the leaves of   is the Bott connection   (cf. ( 2.1 )). The normal bundle τ   is a holonomy equivariant Hermitian vector bundle, if   is a Riemannian foliation.
Example 3.35 (Transverse cotangent bundle). The dual example to the previous one is given by the conormal bundle N *   equipped with the action of the holonomy groupoid G   by the linear holonomy map ( d h γ * ) 1 : N x * N y *   for γ : x y   . The corresponding flat connection defined along the leaves of   is the connection ( ) *   , dual to the Bott connection (cf. ( 2.3 )).
The associated groupoid, which will be denoted by G N   , is described as follows:
G N = { ( γ , η ) G × N * : r ( γ ) = π ( η ) }   with the source map s : G N N * , s ( γ , η ) = d h γ * ( η )   , the range map r : G N N * , r ( γ , η ) = η   and the composition ( γ , η ) ( γ , η ) = ( γ γ , η )   defined in the case when η = d h γ * ( η )   . The projection π : N * M   induces a map π G : G N G   by the formula π G ( γ , η ) = γ , ( γ , η ) G N .   The corresponding horizontal foliation coincides with the linearized foliation N   on N *   (see the symplectic description of this foliation in Example  2.46 ), which can be described in the following way (cf.
also an invariant definition in [136). The leaf L ~ η   of N   through a point η N *   consists of all points of the form d h γ * ( η )   with γ G   such that r ( γ ) = π ( η )   . It is diffeomorphic to the holonomy covering G x   of the leaf L x , x = π ( η )   of   through x   . Each leaf has trivial holonomy, and the holonomy groupoid of the linearized foliation N   coincides with G N   .
Let us also note that the canonical relation Γ   in T * M   associated with N *   (cf. Example  2.46 ) is given by the (one-to-one) immersion ( r , s ) : G N T * M × T * M .  
For any holonomy equivariant vector bundle E   on a foliated manifold ( M , )   , there is defined a pre-Hilbert C c ( G )   -module   . As a linear space,   coincides with C c ( G , r * E )   . The module structure on   is introduced in the following way: the action of f C c ( G )   on s   is given by ( s * f ) ( γ ) = G y s ( γ ) f ( γ 1 γ ) d ν y ( γ ) , γ G y ,   and the inner product on   with values in C c ( G )   is given by s 1 , s 2 ( γ ) = G y s 1 ( γ 1 ) , s 2 ( γ 1 γ ) E s ( γ ) d ν y ( γ ) , s 1 , s 2 .   The completion of   in the norm s = R ( s , s ) 1 / 2   defines a C *   -Hilbert module over C r * ( G )   , that we denote by   . It is equipped with a C r * ( G )   -sesquilinear form ,   with values in C r * ( G )   , which is the extension by continuity of the sesquilinear form on   . The C *   -Hilbert module   can be considered as a noncommutative analogue of the algebra of continuous sections of the bundle E   considered as a bundle on the leaf space M /   .
There is also a left action of C c ( G )   on   given by
( f * s ) ( γ ) = G y f ( γ ) T ( γ ) [ s ( γ 1 γ ) ] d ν y ( γ ) , γ G y , (3.4)
where f C c ( G )   and s   . Unlike the right action, the left action does not extend to an action of C r * ( G )   by bounded endomorphisms of the C *   -Hilbert module   over C r * ( G )   . As shown in [42(see also [38), for any f C c ( G )   the formula ( 3.4 ) defines an endomorphism λ ( f )   of the C *   -Hilbert module   with the adjoint, given by ( λ ( f ) * s ) ( γ ) = G y f ( γ ) T ( γ ) [ s ( γ 1 γ ) ] d ν y ( γ ) , γ G y ,   where f ( γ ) = f ¯ ( γ 1 ) Δ ( γ )   and Δ ( γ ) : E r ( γ ) E r ( γ )   is the linear operator given by Δ ( γ ) = ( T ( γ ) 1 ) * T ( γ ) 1 .   Thus, if the Hermitian structure on E   is not holonomy invariant, the representation λ   is unbounded with respect to the C *   -norm on C r * ( G )   .
Nevertheless, it can be shown that the homomorphism λ   is a densely defined closable homomorphism of C *   -algebras. Hence, the domain of the closure λ ¯   of the homomorphism λ   endowed with the graph norm x λ = x + λ ( x )   is a Banach algebra B   , dense in the C *   -algebra C r * ( G )   . Thus, the C *   -Hilbert module   is a B   C r * ( G )   -bimodule.
In particular, if we take E   to be the holonomy equivariant bundle j N *   , we get the C *   -Hilbert module of transverse differential j   -forms Ω j   over the C *   -algebra C r * ( G )   as the completion of the pre-Hilbert C c ( G , | T G | 1 / 2 )   -module Ω j = C c ( G , r * j N * | T G | 1 / 2 )   . There is a product Ω j × Ω k Ω j + k   , compatible with the bimodule structure:
( ω ω 1 ) ( γ ) = G y ω ( γ 1 ) γ 1 ω 1 ( γ 1 1 γ ) , ω Ω j , ω 1 Ω k . (3.5)
For any holonomy equivariant vector bundle E   on a foliated manifold ( M , )   , there is defined a natural representation of C c ( G )   in the space of sections of this bundle. It is obtained as the restriction of the representation R E   of the algebra C c ( G , ( E ) )   to C c ( G )   , which is embedded in C c ( G , ( E ) )   by means of the homomorphism k C c ( G ) k T C c ( G , ( E ) ) .   If { ν x : x G ( 0 ) }   is a smooth Haar system on G   , then this representation, which we will also denote by R E   , is defined as follows. For any u C ( M , E )   , the section R E ( k ) u C ( M , E )   is given by
R E ( k ) u ( x ) = G x k ( γ ) T ( γ ) [ u ( s ( γ ) ) ] d ν x ( γ ) , x M . (3.6)
If E   is a holonomy equivariant Hermitian vector bundle on M   , then R E   is a *   -representation. In this case, we define a C *   -algebra C E * ( G )   as the closure of R E ( C c ( G ) )   in the uniform topology of ( L 2 ( M , E ) )   .
By Theorem 2.1 in [68, there is an estimate
k C r * ( G , E ) R E ( k ) , k C c ( G , ( E ) ) , (3.7)
therefore, the reduced C *   -algebra C r * ( G , E )   is the quotient of the algebra C E * ( G )   , and there is a natural projection π : C E * ( G ) C r * ( G , E )   .

4 Noncommutative topology

4.1 K   -theory and K   -homology

One of the main tools for the investigation of topological spaces is the topological K   -theory (see [1, 108, 132). Recall that the group K 0 ( X )   associated with a compact topological space X   is generated by stable equivalence classes of locally trivial finite-dimensional complex vector bundles on X   .
Equivalently, elements of K 0 ( X )   are described as stable isomorphism classes of finitely generated projective modules over the algebra C ( X )   or as equivalence classes of projections in the matrix algebra over C ( X )   (see below).
In this Section, we briefly recall the definition of the K   -theory for C *   -algebras, the noncommutative analogue of the topological K   -theory, and also the definition of the K   -homology groups (see, for instance, [19, 42, 98, 108, 160, 183for further information).
Let A   be a unital C *   -algebra. Denote by M n ( A )   the algebra of n × n   -matrices with elements from A   . We will assume that M n ( A )   is embedded into M n + 1 ( A )   by the map X ( X 0 0 0 )   . Let M ( A ) = lim M n ( A )   .
The group K 0 ( A )   is defined as the set of homotopy equivalence classes of projections ( p 2 = p = p *   ) in M ( A )   equipped with the direct sum operation p 1 p 2 = ( p 1 0 0 p 2 ) .   Denote by G L n ( A )   the group of invertible n × n   -matrices with elements from A   . We will assume that G L n ( A )   is embedded into G L n + 1 ( A )   by the map X ( X 0 0 1 )   . Let G L ( A ) = lim G L n ( A )   . The group K 1 ( A )   is defined as the set of homotopy equivalence classes of unitary matrices ( u * u = u u * = 1   ) in G L ( A )   , equipped with the direct sum operation.
If A   has no unit and A +   is the algebra, obtained by adding the unit to the algebra A   , then we have the homomorphism i : C A + : λ λ 1   , inducing a homomorphism i * : K 0 ( C ) K 0 ( A + )   , and K 0 ( A )   is defined as the kernel of this homomorphism. Moreover, by definition, K 1 ( A ) = K 1 ( A + )   .
For an arbitrary algebra A   over C   , there are defined the groups K 0 ( A )   and K 1 ( A )   of the algebraic K   -theory (see, for instance, [7, 130). The group K 0 ( A )   is defined similarly to the group K 0 ( A )   of the topological K   -theory, using idempotents ( e 2 = e   ) in M ( A )   instead of projections. The group K 1 ( A )   is defined as the quotient of the group G L ( A )   by the commutant [ G L ( A ) , G L ( A ) ]   .
The construction of the K   -homology for noncommutative algebras is based on the notion of Fredholm module. This notion is a functional-analytic abstraction of the notion of elliptic pseudodifferential operator on a compact manifold. The idea that abstract elliptic operators can be naturally considered as elements of the K   -homology groups, suggested by Atiyah [2, was realized by Kasparov (see [110). Recall that a Hilbert space H   is called Z 2   -graded, if there is given its decomposition into a direct sum of Hilbert subspaces H = H 0 H 1   . Equivalently, a Z 2   -grading on H   is defined by a self-adjoint operator γ ( H )   such that γ 2 = 1   . Given the decomposition H = H + H   , the operator γ   has the matrix ( 1 0 0 1 ) .  
Definition 4.1. A Fredholm module (or a K   -cycle) over a C *   -algebra A   is a pair ( H , F )   , where
  • (1) H   is a Hilbert space, equipped with a *   -representation ρ   of the algebra A   ;
  • (2) F   is a bounded operator in H   such that, for any a A   , the operators ( F 2 1 ) ρ ( a )   , ( F F * ) ρ ( a )   and [ F , ρ ( a ) ]   are compact in H   .
A Fredholm module ( H , F )   is called even, if the Hilbert space H   is endowed with a Z 2   -grading γ   , the operators ρ ( a )   are even, γ ρ ( a ) = ρ ( a ) γ   , and the operator F   is odd, γ F = F γ   . In the opposite case, it is called odd.
Recall that, for any p 1   , the Schatten class p ( )   consists of all compact operators T   in a Hilbert space   such that the operator | T | p   is a trace class operator. Let μ 1 ( T ) μ 2 ( T )   be the characteristic numbers ( s   -numbers) of a compact operator T   in   , that is, the eigenvalues of the operator | T | = T * T   , taken with multiplicities. Then T p ( ) t r | T | p = n = 1 | μ n ( T ) | p < .  
Definition 4.2. A Fredholm module ( H , F )   is called p   -summable, if ( F 2 1 ) ρ ( a )   , ( F F * ) ρ ( a )   and [ F , ρ ( a ) ]   belong to p ( H )   .
The homology groups K 0 ( A )   ( K 1 ( A )   ) are defined as the sets of homotopy equivalence classes of even (resp. odd) Fredholm modules over A   . The direct sum operation defines the structure of an abelian group on K 0 ( A )   and K 1 ( A )   .
Example 4.3 ([39). Let M   be a compact manifold, E   a Hermitian vector bundle on M   . Then the pair ( H , F )   , where:
  •   H = L 2 ( M , E )   ;
  •   F Ψ 0 ( M , E )   is an elliptic operator with the principal symbol σ F   such that σ F 2 = 1 , σ F * = σ F   (for instance, one can take an operator of the form D ( 1 + D 2 ) 1 / 2   with some self-adjoint elliptic operator D Ψ m ( M , E ) , m > 0   )
is a Fredholm module over C ( M )   .
Example 4.4 ([39). Let ( M , )   be a compact foliated manifold, E   a holonomy equivariant Hermitian vector bundle on M   . Then the pair ( H , F )   , where:
  •   H = L 2 ( M , E )   ;
  •   F Ψ 0 ( M , E )   is a transversally elliptic operator with the holonomy invariant (see Definition  7.7 ) transversal principal symbol σ F   such that σ F 2 = 1 , σ F * = σ F   .
is a Fredholm module over C c ( G )   .
A Fredholm module ( H , F )   defines an index map ind : K * ( A ) Z   .
In the even case, with respect to the decomposition H = H + H   given by the Z 2   -grading of H   , the operator F   takes the form
F = ( 0 F F + 0 ) , F ± : H H ± . (4.1)
For any projection e M q ( A )   , the operator e ( F + 1 ) e   , acting from e ( H + C q )   to e ( H C q )   , is Fredholm, its index depends only on the homotopy class of e   . Therefore, a map ind : K 0 ( A ) Z   is defined as
ind [ e ] = ind e ( F + 1 ) e . (4.2)
In the odd case, for a unitary matrix U G L q ( A )   , the operator ( P 1 ) U ( P 1 )   , where P = 1 + F 2   , is a Fredholm operator. Moreover, the index of the operator ( P 1 ) U ( P 1 )   depends only on the homotopy class of U   .
Therefore, the map ind : K 1 ( A ) Z   is given by
ind [ U ] = ind ( P 1 ) U ( P 1 ) . (4.3)
In both cases, both in the even one, and in the odd one, the map ind   depends only on the class defined by the Fredholm module ( H , F )   in the K   -homology group.
Computation of the K   -theory for the C *   -algebra of a foliation is not an easy problem. In the case of the linear foliation on T 2   , the computation of the K   -theory of the C *   -algebra of the foliation can be reduced to the computation of the K   -theory of the C *   -algebra A θ   (see Example  4.10 ) and was given in [150. For arbitrary foliations on T 2   and S 3   , the computation of the K   -theory of the C *   -algebra of the foliation was done in [178.
In [8(see also [55) a geometrical construction of elements of the group K ( C * ( M , ) )   was suggested. For any topological groupoid C   , one can define the classifying space B C   , which is constructed, using a modification of the classical Milnor's construction of the classifying space of a group [89, 129. In particular, the classifying space B G   of the holonomy groupoid of a foliation is defined (see [37,Chapter9). The normal bundle τ   on M   defines a vector bundle on B G   , which is also denoted by τ   . The geometric group K * , τ ( B G )   is defined as the relative K   -theory group K * , τ ( B G ) = K * ( B ( τ ) , S ( τ ) ) ,   where B ( τ )   is the unit ball subbundle of the bundle τ   and S ( τ )   is the unit sphere bundle. In [8(cf. also [55), a map (the topological index) μ : K * , τ ( B G ) K ( C * ( M , ) )   was constructed by means of topological constructions. The Baum-Connes conjecture claims that μ   is an isomorphism. We refer the reader to the papers [93, 124, 125, 142, 143, 170, 171, 172, 178, 182, 181for various aspects of the Baum-Connes conjecture for foliations and related computations of the K   -theory for C *   -algebras of foliations.

4.2 Strong Morita equivalence

The notion of isometric *   -isomorphism between C *   -algebras is a natural analogue of the notion of homeomorphism of topological spaces. However, there is a wider equivalence relation for C *   -algebras, strong Morita equivalence, introduced by Rieffel [157, 159, which preserves many invariants of C *   -algebras, for instance, the K-theory, the space of irreducible representations, the cyclic cohomology, and coincides with the isomorphism on the class of commutative C *   -algebras. The isometric *   -isomorphism of C *   -algebras of foliations is related with the isomorphism of the corresponding holonomy groupoids, while their strong Morita equivalence is related with the isomorphism of the corresponding leaf spaces. Therefore, in many respects the notion of strong Morita equivalence is a more adequate notion for the study of transverse geometric structures on foliated manifolds.
Definition 4.5. Let A   and B   be C *   -algebras. An A   - B   -equivalence bimodule is an A   - B   -bimodule X   , endowed with A   -valued and B   -valued inner products , A   and , B   accordingly, such that X   is a right Hilbert B   -module and a left Hilbert A   -module with respect to these inner products, and, moreover,
  • (1) x , y A z = x y , z B   for any x , y , z X   ;
  • (2) The set X , X A   generates a dense subset in A   , and the set X , X B   generates a dense subset in B   .
We call algebras A   and B   strongly Morita equivalent, if there is an A   - B   -equivalence bimodule.
It is not difficult to show that the strong Morita equivalence is an equivalence relation.
For any linear space L   , denote by L ~   the conjugate complex linear space, which coincides with L   as a set and has the same addition operation, but the multiplication by scalars is given by the formula λ x ~ = ( λ ¯ x )   . If X   is an A   - B   -equivalence bimodule, then X ~   is endowed with the structure of a B   - A   -equivalence bimodule. For instance, b x ~ a = ( a * x b * ) .  
Theorem 4.6. [157Let X   be an A   - B   -equivalence bimodule. Then the map E X B E   defines an equivalence of the category of Hermitian B   -modules and the category of Hermitian A   -module with the inverse, given by the map F F B X ~   .
In particular, Theorem  4.6 implies that two commutative C *   -algebras are strongly Morita equivalent if and only if they are isomorphic.
The following theorem relates the notion of strong Morita equivalence with the notion of stable equivalence.
Theorem 4.7. [26Let A   and B   are C *   -algebras with countable approximate units. Then these algebras are strongly Morita equivalent if and only if they are stably equivalent, i.e. A K = A K   , where K   denotes the algebra of compact operators in a separable Hilbert space.
Example 4.8. If   is a simple foliation given by a bundle M B   , then the C *   -algebra C * ( M , )   is strongly Morita equivalent to the C *   -algebra C 0 ( B )   .
Example 4.9. Consider a compact foliated manifold ( M , )   . As usual, let G   denote the holonomy groupoid of   . For any subsets A , B M   , denote G B A = { γ G : r ( γ ) A , s ( γ ) B } .   In particular, G T M = { γ G : s ( γ ) T } .   If T   is a transversal, then G T T   is a submanifold and a subgroupoid in G   . Let C r * ( G T T )   be the reduced C *   -algebra of this groupoid.
As shown in [99, if T   is a complete transversal, then the algebras C r * ( G )   and C r * ( G T T )   are strongly Morita equivalent. In particular, this implies that C r * ( G ) = K C r * ( G T T ) .  
Example 4.10. Consider the linear foliation θ   on the two-dimensional torus T 2   , where θ R   is a fixed irrational number. If we choose the transversal T   given by the equation y = 0   , then the leaf space of the foliation θ   is identified with the orbit space of the Z   -action on the circle S 1 = R / Z   generated by the rotation R θ ( x ) = x + θ m o d 1 , x S 1 .   Elements of the algebra C c ( G T T )   are determined by matrices a ( i , j )   , where the indices ( i , j )   are arbitrary pairs of elements i   and j   of T   , lying on the same leaf of   , that is, on the same orbit of the Z   -action R θ   . Since in this case the leafwise equivalence relation on the transversal is given by a free group action, the algebra C * ( G T T ) = A θ   coincides with the crossed product C ( S 1 ) Z   of the algebra C ( S 1 )   by the group Z   with respect to the Z   -action R θ   on C ( S 1 )   . Therefore (cf. Example  3.18 ), every element of A θ   is given by a power series a = n Z a n U n , a n C ( S 1 ) ,   the multiplication is given by ( a U n ) ( b U m ) = a ( b R n θ 1 ) U n + m   and the involution by ( a U n ) * = a ¯ U n .   The algebra C ( S 1 )   is generated by the function V   on S 1   defined as V ( x ) = e 2 π i x , x S 1 .   Hence, the algebra A θ   is generated by two elements U   and V   , satisfying the relation V U = λ U V , λ = e 2 π i θ .   Thus, for example, a general element of C c ( G T T )   can be represented as a power series a = ( n , m ) Z 2 a n m U n V m ,   where a n m S ( Z 2 )   is a rapidly decreasing sequence (that is, for any natural k   we have sup ( n , m ) Z 2 ( | n | + | m | ) k | a n m | <   ).
Since, in the commutative case ( θ = 0   ), the above description defines the algebra of smooth functions on the two-dimensional torus, the algebra A θ   is called the algebra of continuous functions on a noncommutative torus T θ 2   , C c ( G T T ) = C ( T θ 2 )   , A θ = C ( T θ 2 )   .
The algebra A θ   was introduced in the paper [158(see also [35) and has found many applications in mathematics and physics (cf., for instance, a survey [112).
The C *   -algebra C r * ( G )   of the linear foliation on T 2   is strongly Morita equivalent to A θ   (cf. Example  4.9 ).
An important property of the groupoid G T T   associated with a complete transversal T   is that it is an etale groupoid (cf. [59).
Definition 4.11. A smooth groupoid G   is called etale, if its source map s : G G ( 0 )   is a local diffeomorphism.
One can introduce an equivalence relation for groupoids, similar to the strong Morita equivalence relation for C *   -algebras [92, 140(cf. also [59).
Informally speaking, two groupoids are equivalent, if they have the same orbit spaces and, therefore, the same transverse geometry. It is proved in [140that, if groupoids G   and H   are equivalent, then its reduced C *   -algebras are strongly Morita equivalent.
By [59, a smooth groupoid is equivalent to an etale one if and only if all its isotropy groups G x x   are discrete. In the latter case, such a groupoid is called a foliation groupoid. Examples of groupoids with these property are given by the holonomy groupoid G ( M , )   and the fundamental groupoid Π ( M , )   of a foliation ( M , )   . The holonomy map defines a groupoid morphism hol : Π ( M , ) G ( M , ) ,   which is identical on G ( 0 ) = M   (a morphism over M   ).
Any foliation groupoid G   defines a foliation   on G ( 0 ) = M   . G   is said to be an integration of   . The next theorem provides a precise formulation of a well-known principle, which says that the holonomy groupoid G ( M , )   and the fundamental groupoid Π ( M , )   of a foliation ( M , )   are extreme examples of groupoids, integrating   .
A groupoid G   is said to be s   -connected, if its s   -fibers G x   are connected.
Theorem 4.12. [59Let ( M , )   be a foliated manifold. For any s   -connected groupoid G   , integrating   , there is a natural representation of the holonomy morphism hol   as the composition of morphisms h G   , hol G   over M   hol : Π ( M , ) h G G hol G G ( M , ) .   The maps h G   and hol G   are surjective local diffeomorphisms.
Finally, G   is s   -simply connected (i.e. has simply connected s   -fibers G x   ) if and only if the morphism h G   is an isomorphism.

5 Noncommutative differential topology

5.1 Cyclic cohomology

In this Section, we give the definition of the cyclic cohomology, playing the role of a noncommutative analogue of the de Rham homology of topological spaces (concerning to cyclic cohomology, see the books [25, 42, 109, 123and the bibliography therein). In the commutative case, the definition of the de Rham cohomology requires an additional structure on a topological space in question, for instance, the structure of a smooth manifold. In the noncommutative case, this shows up in the fact that cyclic cocycles are usually defined not on the C *   -algebra, an analogue of the algebra of continuous functions on the corresponding geometrical object, but on some subalgebra, consisting of “smooth” functions. We postpone the consideration of a noncommutative analogue of the notion of the algebra of smooth functions on a smooth manifold, the notion of smooth algebra, till the next Section, and now we turn to the definition of the cyclic cohomology for an arbitrary algebra.
Let A   be an algebra over C   . Consider the complex ( C * ( A , A * ) , b )   , where:
  •   C k ( A , A * )   is the space of ( k + 1 )   -linear forms on A   , k N   ;
  •   For ψ C k ( A , A * )   , its coboundary b ψ C k + 1 ( A , A * )   is given by
    b ψ ( a 0 , , a k + 1 ) = ( 1 ) j ψ ( a 0 , , a j a j + 1 , , a k + 1 )
    + ( 1 ) k + 1 ψ ( a k + 1 a 0 , , a k ) , a 0 , a 1 , , a k + 1 A .
The cohomology of this complex is called the Hochschild cohomology of the algebra A   with coefficients in the bimodule A *   and are denoted by H H ( A )   .
Let C λ k ( A )   be the subspace of C k ( A , A * )   , which consists of all ψ C k ( A , A * )   , satisfying the cyclicity condition
ψ ( a 1 , , a k , a 0 ) = ( 1 ) k ψ ( a 0 , a 1 , , a k ) , a 0 , a 1 , , a k A . (5.1)
The differential b   takes C λ k ( A )   to C λ k + 1 ( A )   , and the cyclic cohomology H C * ( A )   of the algebra A   is defined as the cohomology of the complex ( C λ * ( A ) , b )   .
Example 5.1. For k = 0   , H C 0 ( A )   is the linear space of all trace functionals on A   . By this reason, cyclic k   -cocycles on A   are called k   -traces on A   (for k > 0   , higher traces).
Example 5.2. For A = C   , H C n ( C ) = 0   , if n   is odd, and H C n ( C ) = C   , if n   is even. A nontrivial cocycle φ C λ n ( C )   is given by φ ( a 0 , a 1 , , a n ) = a 0 a 1 a n , a 0 , a 1 , , a n C .  
Equivalently, the cyclic cohomology can be described, using a ( b , B )   -bicomplex.
Define an operator B : C k ( A , A * ) C k 1 ( A , A * )   as B = A B 0 ,   where, for any a 0 , a 1 , , a k 1 A   ,
A ψ ( a 0 , , a k 1 ) = j = 0 k 1 ( 1 ) ( k 1 ) j ψ ( a j , a j + 1 , a k 1 , a 0 , a 1 , , a j 1 ) , B 0 ψ ( a 0 , , a k 1 ) = ψ ( 1 , a 0 , , a k 1 ) ( 1 ) k ψ ( a 0 , , a k 1 , 1 ) .  
One has that B 2 = 0   , b B = B b   .
Consider the following double complex: C n , m = C n m ( A , A * ) , n , m Z ,   with the differentials d 1 : C n , m C n + 1 , m   and d 2 : C n , m C n , m + 1   given by d 1 ψ = ( n m + 1 ) b ψ , d 2 ψ = 1 n m B ψ , ψ C n , m .   For any q N   , consider the complex ( F q C , d )   , where ( F q C ) p = m q , n + m = p C n , m , p N , d = d 1 + d 2 .   Then one has an isomorphism H C n ( A ) = H p ( F q C ) , n = p 2 q .   More precisely, this isomorphism associates to any ψ H C n ( A )   a cocycle φ H p ( F q C )   for some p   and q   with n = p 2 q   , which has the only nonzero component φ p , q = ( 1 ) [ n / 2 ] ψ .   In particular, any cocycle in the complex F q C   is cohomologic to a cocycle of the form described above.
One can also the periodic cyclic cohomology H P ev ( A )   and H P odd ( A )   by taking the inductive limit of the groups H C k ( A )   , k 0   , with respect to the periodicity operator S : H C k ( A ) H C k + 2 ( A )   given as the cup product with the generator in H C 2 ( C )   . In terms of the ( b , B )   -bicomplex, the periodic cyclic cohomology is described as the cohomology of the complex C ev ( A ) b + B C odd ( A ) b + B C ev ( A ) ,   where C ev / odd ( A ) = k even / odd C k ( A ) .  
Example 5.3. Let A   be the algebra C ( M )   of smooth functions on an n   -dimensional compact manifold M   and let D k ( M )   denote the space of k   -dimensional de Rham currents on M   . Any C D k ( M )   determines a Hochschild cochain on C ( M )   as
ψ C ( f 0 , f 1 , , f k ) = C , f 0 d f 1 d f k , f 0 , f 1 , , f k C ( M ) . (5.2)
This cochain satisfies the condition B ψ C = k ψ d t C   , where d t   is the de Rham boundary for currents. Thus, the map
D ev / odd ( M ) C = ( C k ) φ C = ( 1 k ! ψ C k ) C ev / odd ( C ( M ) ) (5.3)
induces a homomorphism from the de Rham homology groups H ev / odd ( M )   to the cyclic periodic cohomology groups H P ev / odd ( C ( M ) )   .
This homomorphism is an isomorphism, if we restrict ourselves by the cohomology of continuous cyclic cochains [42.
Example 5.4. Another important example of cyclic cocycles is given by normalized cocycles on a discrete group Γ   . Recall (cf, for instance, [82) that a (homogeneous) k   -cocycle on Γ   is a map h : Γ k + 1 C   , satisfying the identities
h ( γ γ 0 , , γ γ k ) = h ( γ 0 , , γ k ) , γ , γ 0 , , γ k Γ ; i = 0 k + 1 ( 1 ) i h ( γ 0 , , γ i 1 , γ i + 1 , , γ k + 1 ) = 0 , γ 0 , , γ k + 1 Γ .  
One can associate to any homogeneous k   -cocycle h   a (nonhomogeneous) k   -cocycle c Z k ( Γ , C )   by the formula c ( γ 1 , , γ k ) = h ( e , γ 1 , γ 1 γ 2 , , γ 1 γ k ) .   It can be easily checked that c   satisfies the following condition
c ( γ 1 , γ 2 , , γ k ) + i = 0 k 1 ( 1 ) i + 1 c ( γ 0 , , γ i 1 , γ i γ i + 1 , γ i + 2 , , γ k ) + ( 1 ) k + 1 c ( γ 0 , γ 1 , , γ k 1 ) = 0 .  
A cocycle c Z k ( Γ , C )   is said to be normalized (in the sense of Connes), if c ( γ 1 , γ 2 , , γ k )   equals zero in the case when, either γ i = e   for some i   , or γ 1 γ k = e   . The group ring C Γ   consists of all functions f : Γ C   with finite support. The multiplication in C Γ   is given by the convolution operation f 1 * f 2 ( γ ) = γ 1 γ 2 = γ f 1 ( γ 1 ) f 2 ( γ 2 ) , γ Γ .   A normalized cocycle c Z k ( Γ , C )   determines a cyclic k   -cocycle τ c   on C Γ   by the formula τ c ( f 0 , , f k ) = γ 0 γ k = e f 0 ( γ 0 ) f k ( γ k ) c ( γ 1 , , γ k ) ,   where f j C Γ   for j = 0 , 1 , k   .
In the commutative case, the K   -theory and the cohomology of a compact topological space M   are related by the Chern character ch : K * ( M ) H * ( M , R ) ,   which provides an isomorphism K * ( M ) Q H * ( M , Q )   (cf., for instance, [132).
If M   is a smooth manifold, then there is an explicit construction of the Chern character (the Chern-Weil construction, cf., for instance, [131, 111), that makes use of differential forms, currents, connections and curvature.
More precisely, if E   is a smooth vector bundle on M   , then the Chern character ch ( E ) H ev ( M , R )   of the corresponding class [ E ]   in K 0 ( M )   is represented by the closed differential form ch ( E ) = T r exp ( 2 2 π i )   for any connection : C ( M , E ) C ( M , E T * X )   in the bundle E   .
In the odd case (see [9, 72), if U C ( M , U N ( C ) )   , then the Chern character ch ( U ) H odd ( M , R )   of the corresponding class [ U ]   in K 1 ( M )   is given by the cohomology class of the differential form ch ( U ) = k = 0 + ( 1 ) k k ! ( 2 k + 1 ) ! T r ( U 1 d U ) 2 k + 1 .   Any closed de Rham current C D k ( M )   defines, for an even k   , a map φ C : K 0 ( M ) C   by the formula
φ C , E = C , ch ( E ) , E K 0 ( M ) , (5.4)
and, for an odd k   . a map φ C : K 1 ( M ) C   by the formula
[ φ C ] , [ U ] = 1 2 i π C , ch ( U ) , U C ( M , U N ( C ) ) . (5.5)
The noncommutative analogue of the Chern-Weil construction consists in the construction of a pairing between H C * ( A )   and K * ( A )   for an arbitrary algebra A   , generalizing the maps ( 5.4 ) and ( 5.5 ). The pairing between H C ev ( A )   and K 0 ( A )   is defined as follows. For any cocycle φ = ( φ 2 k )   in C ev ( A )   and for any projection e   in M q ( A )   put
[ φ ] , [ e ] = k 0 ( 1 ) k ( 2 k ) ! k ! φ 2 k # T r ( e 1 2 , e , , e ) , (5.6)
where φ 2 k # T r   is the ( 2 k + 1 )   -linear map on M q ( A ) = M q ( C ) A   , given by
φ 2 k # T r ( μ 0 a 0 , , μ 2 k a 2 k ) = T r ( μ 0 μ 2 k ) φ 2 k ( a 0 , , a 2 k ) , (5.7)
for any μ j M q ( C )   and a j A   .
The pairing between H C odd ( A )   and K 1 ( A )   is given by
[ φ ] , [ U ] = 1 2 i π k 0 ( 1 ) k k ! φ 2 k + 1 # T r ( U 1 1 , U 1 , , U 1 1 , U 1 ) , (5.8)
where φ = ( φ 2 k + 1 ) C odd ( A )   and U   is a unitary matrix in G L q ( A )   .
Now consider a p   -summable Fredholm module ( H , F )   over an algebra A   . The index maps ( 4.2 ) and ( 4.3 ) are computed in terms of the pairing of K * ( A )   with some cyclic cohomology class τ = ch * ( H , F ) H C n ( A )   , called the Chern character of the Fredholm module ( H , F )   . First of all, any Fredholm module can be replaced by an equivalent one, where one has F 2 = 1   . Under this condition, in the odd case, τ   is given by
τ ( a 0 , a 1 , , a n ) = λ n t r ( a 0 [ F , a 1 ] [ F , a n ] ) , a 0 , a 1 , , a n A , (5.9)
where n   is odd, n > p   , λ n   are some constants, depending only on n   .
In the even case, τ   is given by
τ ( a 0 , a 1 , , a n ) = λ n t r ( γ a 0 [ F , a 1 ] [ F , a n ] ) , a 0 , a 1 , , a n A , (5.10)
where n   is even, n > p   , γ   is the Z 2   -grading in the space H   , λ n   are some constants, depending only on n   .

5.2 Smooth algebras

In this Section, we give general facts about smooth subalgebras of C *   -algebras, being noncommutative analogues of the algebra of smooth functions on a smooth manifold. Suppose that A   is a C *   -algebra and A +   is the algebra obtained by adjoining the unit to A   . Suppose that A   is a *   -subalgebra of the algebra A   and A +   is the algebra obtained by adjoining the unit to A  
Definition 5.5. We say that A   is a smooth subalgebra of the algebra A   , if the following conditions hold:
  • (1) A   is a dense *   -subalgebra of the C *   -algebra A   ;
  • (2) A   is stable under the holomorphic functional calculus, that is, for any a A +   and for any function f   , holomorphic in a neighborhood of the spectrum of a   (considered as an element of the algebra A +   ) f ( a ) A +   .
Suppose that A   is a dense *   -subalgebra of the algebra A   , endowed with the structure of a Fréchet algebra whose topology is finer than the topology induced by the topology of A   . A necessary and sufficient condition for A 0   to be a smooth subalgebra is given by the spectral invariance (cf. [166,Lemma1.2):
  •   A + G L ( A + ) = G L ( A + )   , where G L ( A + )   and G L ( A + )   denote the group of invertibles in A +   and A +   respectively.
This fact remains true in the case when A   is a locally multiplicatively convex (i.e its topology is given by a countable family of submultiplicative seminorms) Fréchet algebra such that the group G L ( A ~ )   of invertibles is open [166,Lemma1.2.
One of the most important properties of smooth subalgebras consists in the following fact — an analogue of smoothing in the operator K   -theory (cf.
[36,Sect.VI.3, [21).
Theorem 5.6. If A   is a smooth subalgebra in a C *   -algebra A   , then inclusion A A   induces an isomorphism in K   -theory K ( A ) = K ( A )   .
For any p   -summable Fredholm module ( H , F )   over an algebra A   , the algebra, which consists of all operators T   in the uniform closure A ¯   of A   in ( )   , satisfying the condition [ F , T ] p ( )   , is a smooth subalgebra of the C *   -algebra A ¯   . Similarly, every spectral triple ( A , , D )   — a noncommutative analogue of the Riemannian structure (cf. Section  6.2 ) defines a smooth subalgebra of the C *   -algebra A ¯   (see Section  8.1 ). However, in many cases the construction of an appropriate smooth algebra is nontrivial and makes use of specific properties of the problem in question.

5.3 Transverse differential

Let ( M , )   be a foliated manifold, H   an arbitrary distribution on M   , transverse to F = T   . In this Section, following the exposition of [42, 165, we define the transverse differentiation, which is a linear map d H : Ω 0 = C c ( G , | T G | 1 / 2 ) Ω 1 = C c ( G , r * N * | T G | 1 / 2 ) ,   satisfying the condition d H ( k 1 * k 2 ) = d H k 1 * k 2 + k 1 * d H k 2 , k 1 , k 2 C c ( G , | T G | 1 / 2 ) .   The differentiation d H   depends on the choice of a horizontal distribution H   .
For any f C c ( G )   , define d H f C c ( G , r * N * )   by d H f ( X ) = d f ( X ^ ) , X ( r * τ ) γ = H y , γ : x y ,   where X ^ H γ G T γ G   is a unique vector such that d s ( X ^ ) = d h γ 1 ( X )   and d r ( X ^ ) = X   .
For an arbitrary smooth leafwise density λ C c ( M , | T | )   define a 1-form k ( λ ) C ( M , H * ) = C ( M , N * )   as follows. Take an arbitrary point m M   . In a foliated chart φ : U I p × I q   defined in a neighborhood of m   ( φ ( m ) = ( x 0 , y 0 )   ), λ   can be written as λ = f ( x , y ) | d x | , ( x , y ) I p × I q .   Then k ( λ ) = d H f + i = 1 p x i d H x i ,   where, for any X = i = 1 p X i x i X ( )   and for any ω = j = 1 q ω j d y j C ( M , H * )   , the Lie derivative X ω C ( M , H * )   is given by X ω = i = 1 p j = 1 q X i ω j x i d y j .   One can give a slightly different description of k ( λ )   . For any X H m   , let X ~   be an arbitrary projectable vector field, that coincides with X   at m   : X ~ ( x , y ) = i = 1 p X i ( x , y ) x i + j = 1 q Y j ( y ) y j .   Put
k ( λ ) ( X ) = i = 1 p X i ( x 0 , y 0 ) f x i ( x 0 , y 0 ) + j = 1 q Y j ( y 0 ) f x j ( x 0 , y 0 ) + i = 1 p X i x i ( x 0 , y 0 ) f ( x 0 , y 0 ) .  
It can be checked that this definition is independent of the choice of a foliated chart φ : U I p × I q   and an extension X ~   .
If M   is Riemannian, λ   is given by the induced leafwise Riemannian volume form and H = F   , then k ( λ )   coincides with the mean curvature 1-form of   (cf., for instance, [177). In Section  2.7 , we have defined the transversal de Rham differential d H   , acting from C c ( M , j N * )   to C c ( M , j + 1 N * )   . The transverse distribution H   naturally defines a transverse distribution H G = r * H   on the foliated manifold ( G , G )   (cf. Section  3.1 ) and the corresponding transversal de Rham differential d H : C c ( G ) C c ( G , r * N * )   (see ( 2.5 )).
An arbitrary leafwise half-density ρ C ( M , | T | 1 / 2 )   can be written as ρ = f | λ | 1 / 2   with f C ( M )   and λ C ( M , | T | )   . Then d H ρ C c ( M , N * | T | 1 / 2 )   is defined as d H ρ = ( d H f ) | λ | 1 / 2 + 1 2 f | λ | 1 / 2 k ( λ ) .   Any f C c ( G , | T G | 1 / 2 )   can be written as f = u s * ( ρ ) r * ( ρ )   , where u C c ( G )   and ρ C c ( M , | T | 1 / 2 )   . The element d H f C c ( G , r * N * | T G | 1 / 2 )   is defined as d H f = d H u s * ( ρ ) r * ( ρ ) + u s * ( d H ρ ) r * ( ρ ) + u s * ( ρ ) r * ( d H ρ ) .   Finally, the operator d H   has a unique extension to a differentiation of the differential graded algebra Ω = C c ( G , r * N * | T G | 1 / 2 )   . By definition, for any f C c ( G , | T G | 1 / 2 )   and ω C c ( M , N * )   , one has that d H ( f r * ω ) = ( d H f ) r * ω + f r * ( d H ω ) .  

5.4 Transverse fundamental class of a foliation

In this Section, we describe, following [38(cf. also [42), the simplest construction of a cyclic cocycle on the algebra C c ( G , | T G | 1 / 2 )   associated with a foliated manifold, and, namely, the construction of the transverse fundamental class.
There is a general construction of cyclic cocycles on an arbitrary algebra due to Connes [39. Let us call by a cycle of dimension n   a triple ( Ω , d , )   , where Ω = j = 0 n Ω j   is a graded algebra over C   , d : Ω Ω   is a graded differentiation of degree 1   such that d 2 = 0   , and : Ω n C   is a graded trace. Thus, the maps d   and   satisfy the following conditions:
  • (1) d ( ω ω ) = d ω ω + ( 1 ) | ω | ω d ω   ;
  • (2) d 2 = 0   ;
  • (3) ω 2 ω 1 = ( 1 ) | ω 1 | | ω 2 | ω 1 ω 2   ;
  • (4) d ω = 0   for ω Ω n 1   .
Let A   be an algebra over C   . A cycle over the algebra A   is a cycle ( Ω , d , )   along with a homomorphism ρ : A Ω 0   .
For any cycle ( A ρ Ω , d , )   over A   , one can define its character as τ ( a 0 , a 1 , , a n ) = ρ ( a 0 ) d ( ρ ( a 1 ) ) d ( ρ ( a n ) ) .   One can easily check that τ   is a cyclic cocycle on A   . Moreover, any cyclic cocycle is the character of some cycle.
Now let ( M , )   be a manifold equipped with a transversally oriented foliation, H   an arbitrary distribution on M   , transverse to F = T   . Consider the differential graded algebra ( Ω = C c ( G , r * N * | T G | 1 / 2 ) , d H )   .
There is a closed graded trace τ   on ( Ω , d H )   defined by τ ( ω ) = M ω | M ,   where ω Ω q = C c ( G , r * q N * | T G | 1 / 2 )   . Here ω | M   denotes the restriction of ω   to M   . This is a section of the bundle q N * | T |   on M   , and its integral over M   is well-defined, using the transverse orientation of the foliation.
A problem is that, since H   is not integrable, it is not true, in general, that d H 2 = 0   . Using ( 2.5 ) and calculating the ( 0 , 2 )   -component in the operator d 2   , we get d H 2 = ( d F θ + θ d F ) .   The operator ( d F θ + θ d F )   is a tangential differential operator. Therefore, it can be written as R T * M ( k )   (cf. ( 3.3 )), where k C c ( G , ( T * M ) )   is given by the exterior product with a vector-valued distribution Θ D ( G , r * 2 N * | T G | 1 / 2 )   supported in G ( 0 )   . One can show that d H 2 ω = Θ ω ω Θ , ω Ω .   Moreover, d H Θ = 0   . Since the action of d H 2   is given by the commutator with some 2-form, one can canonically construct a new differential graded algebra ( Ω ~ , d ~ H )   with d ~ H 2 = 0   (see [39, 42). The algebra Ω ~   consists of 2 × 2   -matrices ω = { ω i j }   with elements from Ω   . An element ω Ω ~   has degree k   , if ω 11 Ω k   , ω 12 , ω 21 Ω k 1   and ω 22 Ω k 2   . The product in Ω ~   is given by ω ω = [ ω 11 ω 12 ω 21 ω 22 ] [ 1 0 0 Θ ] [ ω 11 ω 12 ω 21 ω 22 ] ,   the differential by d ~ H ω = [ d H ω 11 d H ω 12 d H ω 21 d H ω 22 ] + [ 0 Θ 1 0 ] ω + ( 1 ) | ω | ω [ 0 1 Θ 0 ] ,   and the closed graded trace τ ~ : Ω ~ q C   is defined as τ ~ ( [ ω 11 ω 12 ω 21 ω 22 ] ) = τ ( ω 11 ) ( 1 ) q τ ( ω 22 Θ ) .   Finally, the homomorphism ρ ~ : C c ( G , | T G | 1 / 2 ) Ω ~ 0   is given by ρ ~ ( k ) = [ k 0 0 0 ] .   One can check that the triple ( Ω ~ , d ~ H , τ ~ )   is a cycle over C c ( G , | T G | 1 / 2 )   .
This cycle is called the fundamental cycle of the transversally oriented foliation ( M , )   . The character of this cycle defines a cyclic cocycle φ H   on C c ( G , | T G | 1 / 2 )   . The cocycle φ H   depends on the auxiliary choice of H   , but the corresponding class in the cyclic cohomology is independent of this choice. The class [ M / ] H C q ( C c ( G , | T G | 1 / 2 ) )   defined by the cocycle φ H   is called the transverse fundamental class of the foliation ( M , )   .
Let C c ( G , | T G | 1 / 2 ) +   denote the algebra C c ( G , | T G | 1 / 2 )   with the adjoined unit. For an even q   , let us extend the cycle ( Ω ~ , d H , τ ~ )   to a cycle over C c ( G , | T G | 1 / 2 ) +   , putting τ ~ ( Θ q / 2 ) = 0   . In [76, a formula for a cocycle in the ( b , B )   -bicomplex of the algebra C c ( G , | T G | 1 / 2 ) +   , which gives the transverse fundamental class of the foliation ( M , )   , is obtained: the formula χ r ( k 0 , k 1 , , k r ) = ( 1 ) q r 2 ( q + r 2 ) ! i 0 + + i r = q r 2 M k 0 Θ i 0 d H k 1 d H k r Θ i r ,   where r = q , q 2 ,   , k 0 C c ( G , | T G | 1 / 2 ) +   , k 1 , , k r C c ( G , | T G | 1 / 2 )   , defines a cocycle in the ( b , B )   -bicomplex of the algebra C c ( G , | T G | 1 / 2 ) +   .
The class defined by χ   in H C q ( C c ( G , | T G | 1 / 2 ) )   coincides with [ M / ]   .
The pairing with [ M / ] H C q ( C c ( G , | T G | 1 / 2 ) )   defines an additive map K ( C c ( G , | T G | 1 / 2 ) ) C   — “integration over the transverse fundamental class”. An important problem is the question of topological invariance of this map, that is, the question of the possibility to extend it to an additive map from K ( C r * ( G ) )   to C   . A standard method of solving this problem consists in constructing a smooth subalgebra   in C r * ( G )   , which contains the algebra C c ( G , | T G | 1 / 2 )   , such that the cyclic cocycle φ H   on C c ( G , | T G | 1 / 2 )   , which defines the transverse fundamental class [ M / ]   , extends by continuity to a cyclic cocycle on   and defines, in that way, a map K ( ) = K ( C r * ( G ) ) C   . One managed to do this for so called para-Riemannian foliations (cf.
Section  6.3 ). An example of a para-Riemannian foliation is given by the lift of   to a foliation V   in the total space P   of the fibration π : P M   , whose fiber P x   at x M   is the space of all Euclidean metrics on the vector space τ x = T x M / T x   . Thus, one can define a map K ( C * ( P , V ) ) C   . Since the leaves of the fibration P   are connected spin manifolds of nonpositive curvature, one can construct an injective map K ( C r * ( G ) ) K ( C * ( P , V ) )   , that allows to prove the existence of the map K ( C r * ( G ) ) C   , given by the transverse fundamental class, for an arbitrary foliation.
For geometrical consequences of this construction, see [38.

5.5 Cyclic cocycle defined by the Godbillon-Vey class

Consider a smooth compact manifold M   equipped with a transversally oriented codimension one foliation   . The Godbillon-Vey class is a 3-dimensional cohomology class G V H 3 ( M , R )   . It is the simplest example of secondary characteristic classes of the foliation. Recall its definition. Since   is transversally oriented, it is globally defined by a nonvanishing smooth 1-form ω   (that is, for any x M   , we have ker ω x = T x   ). It follows from the Frobenius theorem that there exists a 1-form α   on M   such that d ω = α ω   .
One can check that the 3-form α d α   is closed, and its cohomology class does not depend on the choice ω   and α   . This class G V H 3 ( M , R )   is called the Godbillon-Vey class of   .
Let T   be a complete smooth transversal given by a good cover of M   by regular foliated charts. Thus, T   is an oriented one-dimensional manifold.
In this Section, we construct a cyclic cocycle on the algebra C c ( G T T )   , corresponding to the Godbillon-Vey class. This is done in several steps.
To start with, we recall the construction of the Godbillon-Vey class as a secondary characteristic class associated with the cohomology H * ( W 1 , R )   of the Lie algebra W 1 = R [ [ x ] ] x   of formal vector field on R   .
Recall (cf., for instance, [82, 70), that the cohomology of a Lie algebra g   are defined as the cohomology of the complex ( C * ( g ) , d )   , where:
  C q ( g )   is the space of (continuous) skew-symmetric q   -linear functionals on g   ;   for any c C q ( g )   , the cochain d c C q + 1 ( g )   is defined by
d c ( g 1 , , g q + 1 ) = 1 i < j q + 1 ( 1 ) i + j c ( [ g i , g j ] , g 1 , , g ^ i , , g ^ j , , g q + 1 ) , g 1 , , g q + 1 g .  
For cochains on W 1   , the continuity means the dependence only on finite order jets of its arguments. The cohomology H * ( W 1 , R )   were computed by Gelfand and Fuchs (see, for instance, the book [70and references therein).
They are finite-dimensional and the only nontrivial groups are H 0 ( W 1 , R ) = R 1   and H 3 ( W 1 , R ) = R g v   , where g v ( p 1 x , p 2 x , p 3 x ) = | p 1 ( 0 ) p 2 ( 0 ) p 3 ( 0 ) p 1 ( 0 ) p 2 ( 0 ) p 3 ( 0 ) p 1 ( 0 ) p 2 ( 0 ) p 3 ( 0 ) | , p 1 , p 2 , p 3 R [ [ x ] ] .   Consider the bundle F + k T T   of positively oriented frames of order k   on T   and the bundle F + T = lim F + k T   of positively oriented frames of infinite order on T   . By definition, a positively oriented frame r   of order k   at x T   is the k   -jet at 0 R   of an orientation preserving diffeomorphism f   , which maps a neighborhood of 0   in R   on some neighborhood of x = f ( 0 )   in T   . If y : U R   is a local coordinate on T   , defined in a neighborhood U   of x   , then the numbers y 0 = y ( x ) , y 1 = d ( y f ) d t ( 0 ) , , y k = d k ( y f ) d t k ( 0 ) ,   are coordinates of the frame r   , and, moreover, y 1 > 0   .
There is defined a natural action of the pseudogroup Γ + ( T )   of orientation preserving local diffeomorphisms of T   on F + T   . Let Ω * ( F + T ) Γ + ( T )   denote the space of differential forms on F + T   , invariant under the action of Γ + ( T )   .
There is a natural isomorphism of differential algebras J : C * ( W 1 ) Ω * ( F + T ) Γ + ( T )   defined in the following way. First of all, let v W 1   , and let h t   be any one-parameter group of local diffeomorphisms of R   such that v   is the   -jet of the vector field d h t d t | t = 0   . Then define a Γ + ( T )   -invariant vector field on F + T   , whose value at r = j 0 f J + ( T )   is given by v ( r ) = j 0 ( d ( f h t ) d t | t = 0 )   .
For any c C q ( W 1 )   , put J ( c ) ( v 1 ( r ) , , v q ( r ) ) = c ( v 1 , , v q ) .   One can check that this isomorphism takes the cocycle g v C 3 ( W 1 , R )   to the three-form g v = 1 y 1 3 d y d y 1 d y 2 Ω 3 ( F + 2 T ) Γ + ( T ) .   Consider the bundle F ( M / )   on M   , which consists of infinite order jets of distinguished maps. There is a natural map F ( M / ) F + T / Γ + ( T )   .
Using this map and Γ + ( T )   -invariance of g v Ω 3 ( F + 2 T )   , one can pull back g v   to a closed form g v ( ) Ω 3 ( F ( M / ) )   . Since the fibers of the fibration F 2 ( M / ) M   are contractible, H 3 ( F ( M / ) , R ) = H 3 ( M , R )   , and the cohomology class in H 3 ( M , R )   , corresponding under this isomorphism to the cohomology class of g v ( )   in H 3 ( F ( M / ) , R )   , coincides with the Godbillion-Vey class of   .
To construct the cyclic cocycle on C c ( G T T )   , corresponding to the Godbillion-Vey class, one uses a Van Est type theorem (cf. [91). This theorem states the existence of an embedding
Ω * ( F + T ) Γ + ( T ) C * ( G T T , Ω * ( G T T ) ) , (5.11)
where C * ( G T T , Ω * ( G T T ) )   denotes the space of cochains on the groupoid G T T   with values in the space Ω * ( G T T )   of differential forms on G T T   , which is a homomorphism of complexes, inducing an isomorphism in the cohomology.
Let us describe the cocycle from C * ( G T T , Ω * ( G T T ) )   , corresponding to the form g v Ω 3 ( F + 2 T ) Γ + ( T )   (cf. [91). Let ρ   be an arbitrary smooth positive density on T   . Define a homomorphism δ : G T T R + *   , called the modular homomorphism, setting δ ( γ ) = ρ y d h γ ( ρ x ) , γ G T T , γ : x y ,   where the map d h γ : | τ x | | τ y |   is induced by the linear holonomy map, and also a homomorphism = log δ : G T T R   .
The formula c ( γ 1 , γ 2 ) = ( γ 2 ) d ( γ 1 ) ( γ 1 ) d ( γ 2 ) , γ 1 , γ 2 G T T ,   defines a 2-cocycle on G T T   with values in the space of 1-forms on G T T   . This cocycle, called the Bott-Thurston cocycle, corresponds to g v Ω 3 ( F + 2 T ) Γ + ( T )   under the isomorphism given by the embedding ( 5.11 ).
Finally, the Bott-Thurston cocycle c   defines a cyclic 2-cocycle ψ   on C c ( G T T )   by the formula (cf. Example  5.4 ): ψ ( k 0 , k 1 , k 2 ) = γ 0 γ 1 γ 2 T k 0 ( γ 0 ) k 1 ( γ 1 ) k 2 ( γ 2 ) c ( γ 1 , γ 2 ) , k 0 , k 1 , k 2 C c ( G T T ) .   This is the cyclic cocycle, corresponding to the Godbillon-Vey class of   .
There is another description of cyclic cocycle, corresponding to the Godbillion-Vey class of   , which relates it with invariants of the von Neumann algebra of this foliation [38. The following formula defines a cyclic 1   -cocycle on C c ( G T T )   : τ ( k 0 , k 1 ) = G T T k 0 ( γ 1 ) d k 1 ( γ ) , k 0 , k 1 C c ( G T T ) .   The isomorphism H C 1 ( C c ( G T T ) ) = H C 1 ( C c ( G ) )   , defined by the strong Morita equivalence (cf. Example  4.9 ), associates the class of τ   in the group H C 1 ( C c ( G T T ) )   to the transverse fundamental class of the foliation ( M , )   in H C 1 ( C c ( G ) )   .
The fixed smooth positive density ρ   on T   defines a faithful normal semi-finite weight φ ρ   on the von Neumann algebra W * ( G T T )   of the groupoid G T T   . For any k C c ( G T T )   , the value of the weight φ ρ   is given by φ ρ ( k ) = T k ( x ) ρ ( x ) .   Consider the one-parameter group σ t   of automorphisms of the von Neumann algebra W * ( G T T )   given by σ t ( k ) ( γ ) = δ ( γ ) i t k ( γ ) , k C c ( G T T ) , t R .   This group is the group of modular automorphisms, associated with the weight, by the Tomita-Takesaki theory [173.
The importance of the group of modular automorphisms is explained, in particular, by the following its characterization [173: a one-parameter group of *   -automorphisms σ t   of a von Neumann algebra M   is the group of modular automorphisms, associated with a weight ω   , if and only if ω   satisfies the Kubo-Martin-Schwinger condition with respect to σ t   , that is, there is an analytic in the strip Im z ( 0 , 1 )   and continuous in its closure function f   such that, for any a , b M , t R   , f ( t ) = ω ( σ t ( a ) b ) , f ( t + i ) = ω ( b σ t ( a ) ) .   Following Connes [38, we call by a 1   -trace on a Banach algebra B   a bilinear functional φ   , defined on a dense subalgebra A B   , such that
  • (1) φ   is a cyclic cocycle on A   ;
  • (2) for any a 1 A   , there is a constant C > 0   such that | φ ( a 0 , a 1 ) | C a 0 , a 0 A ,  
and a 2   -trace on B   a trilinear functional φ   , defined on a dense subalgebra A B   , such that
  • (1) φ   is a cyclic cocycle on A   ;
  • (2) for any a 1 , a 2 A   , there is a constant C > 0   such that | φ ( x 0 , a 1 x 1 , a 2 ) φ ( x 0 a 1 , x 1 , a 2 ) | C a 1 a 2 , x 0 , x 1 A .  
The formula τ ˙ ( k 0 , k 1 ) = lim t 0 1 t ( τ ( σ t ( k 0 ) , σ t ( k 1 ) ) τ ( k 0 , k 1 ) ) , k 0 , k 1 C c ( G T T )   defines a 1-trace on C r * ( G T T )   with the domain C c ( G T T )   , invariant under the action of the automorphism group σ t   .
For any 1   -trace φ   on C *   -algebra A   , invariant under an action of an one-parameter automorphism group α t   with the generator D   , such that the space dom φ dom D   is dense in A   , a 2   -trace χ = i D φ   on C r * ( G T T )   (an analogue of the contraction) is defined by χ ( a 0 , a 1 , a 2 ) = φ ( D ( a 2 ) a 0 , a 1 ) φ ( a 0 D ( a 1 ) , a 2 ) , a 0 , a 1 , a 2 dom φ dom D .  
Theorem 5.7. Let ( M , )   be a manifold with a transversally oriented codimension one foliation, T   a complete smooth transversal, ρ   a smooth positive density on T   . Then the cyclic cocycle ψ H C 2 ( C c ( G T T ) )   , corresponding to the Godbillon-Vey class of   , coincides with i D τ ˙   .
As a consequence of this statement, Connes obtained the following geometrical fact. First, recall that one can naturally associate to any von Neumann algebra M   an action of the multiplicative group R + *   (called the flow of weights [56) on some commutative von Neumann algebra, the center of the crossed product M R   of M   by R   relative to the action of R   on M   given by the modular automorphism group σ t   .
Theorem 5.8. [38Let ( M , )   be a manifold with a transversally oriented codimension one foliation. If the Godbillon-Vey class G V H 3 ( M , R )   differs from zero, then the flow of weights of the von Neumann algebra of the foliation   has a finite invariant measure.
In particular, this implies the following, previous result.
Theorem 5.9. [103Let ( M , )   be a manifold with a transversally oriented codimension one foliation. If the Godbillon-Vey class G V H 3 ( M , R )   differs from zero, then the von Neumann algebra of the foliation   has the non trivial type I I I   component.
In [138, 139, one gives another construction of the cyclic cocycle, associated with the Godbillon-Vey class, as a cyclic cocycle on the C *   -algebra of foliation C r * ( G )   , in the case of a foliated S 1   -bundle, that is, when the foliation   is constructed, using the suspension construction for a homomorphism φ : Γ = π 1 ( B ) Diff ( S 1 )   .

5.6 General constructions of cyclic cocycles

The secondary characteristic classes of foliations are given by the characteristic homomorphism [17, 24, 90 χ : H * ( W q ; O ( q ) ) H * ( M , R ) .   defined for any codimension q   foliation   on a smooth manifold M   , where H * ( W q ; O ( q ) )   denotes the relative cohomology of the Lie algebra W q   of formal vector fields in R q   . A basic property of the secondary characteristic classes consists in their functoriality: if a smooth map f : N M   is transverse to a transversely oriented foliation   and f *   is the foliation on N   induced by f   (by definition, the leaves of f *   are the connected components of the pre-images of the leaves of   under the map f   ), then f * ( χ ( α ) ) = χ f * ( α ) , α H * ( W q ; O ( q ) ) .   The classifying space B Γ q   of the groupoid Γ q   classifies codimension q   foliations on a given manifold M   in the sense that any foliation   on M   defines a map M B Γ q   , and, moreover, in the case when M   is compact, a homotopy class of maps M B Γ q   corresponds to a concordance class of foliations on M   [89.
By the functoriality of the characteristic homomorphism, it suffices to know it for the universal foliation on B Γ q   , that gives the universal characteristic homomorphism χ : H * ( W q ; O ( q ) ) H * ( B Γ q , R ) .   For any complete transversal T   , the universal characteristic homomorphism χ   is represented as a composition χ : H * ( W q ; O ( q ) ) H * ( B G T T , R ) H * ( B Γ q , R ) .   Since the groupoids G T T   and G   are equivalent, H * ( B G T T , R ) = H * ( B G , R )   , that defines a map H * ( W q ; O ( q ) ) H * ( B G , R ) .   The relative cohomology H * ( W q ; O ( q ) )   of the Lie algebra W q   are calculated as follows (cf., for instance, [73). Consider the complex W O q = Λ ( u 1 , u 3 , , u m ) P ( c 1 , c 2 , , c q ) q   where m   is the largest odd number, less than q + 1   , Λ ( u 1 , u 3 , , u m )   denotes the exterior algebra with generators u 1 , u 3 , , u m   , deg u i = 2 i 1   , P ( c 1 , c 2 , , c q )   denotes the polynomial algebra with generators c 1 , c 2 , , c q   , deg c i = 2 i   and P ( c 1 , c 2 , , c q ) q   denotes the quotient of P ( c 1 , c 2 , , c q )   by the ideal generated by the monomials of degree greater than 2 q   . The differential in W O q   is defined by d ( u i 1 ) = 1 c i   , d ( 1 c i ) = 0   . There is a homomorphism of the complex W O q   to the complex C * ( W q ; O ( q ) )   of relative cochains on the algebra W q   , which induces an isomorphism in the cohomology H * ( W O q ) = H * ( W q ; O ( q ) )   . For any i = 1 , 2 , , q   , the cohomology class [ 1 c i ] H 2 i ( W O q )   , defined by 1 c i   , is mapped by the characteristic homomorphism χ   to the i   -th Pontryagin class p i   of the normal bundle τ   . The Godbillon-Vey class G V H 3 ( M , R )   of a codimension one foliation on a compact manifold M   is the image of [ u 1 c 1 ] H 3 ( W O 1 )   under the characteristic homomorphism χ   .
Connes [42,Section2 δ ,Theorem14andRemarkb)constructed a natural map Φ * : H * ( B G , R ) H P * ( C c ( G , | T G | 1 / 2 ) )   for any oriented etale groupoid G   . The constructions of the transverse fundamental class of a foliation and of the cyclic cocycle associated with the Godbillon-Vey class are special cases of this general construction.
In [77, Gorokhovsky generalized the above construction of the cyclic cocycle associated with the Godbillon-Vey class, which uses the group of modular automorphisms, to the case of arbitrary secondary characteristic classes.
This construction makes an essential use of the cyclic cohomology theory for Hopf algebras developed in the paper [51(cf. Section  8.3 ).
For the calculations of the cyclic cohomology of C *   -algebras of etale groupoids, cf. [27, 42, 57, 58.

5.7 Index of tangentially elliptic operators

In this Section, we briefly describe applications of the above methods to the index theory for tangentially elliptic operators.
Let D   be a tangentially elliptic operator on a compact foliated manifold ( M , )   . The restrictions of D   to the leaves of   define a family ( D l ) l M /   , where D l   is an elliptic differential operator on a leaf l   of   . It turns out that the families ( P K e r D l ) l M /   and ( P K e r D l * ) l M /   , which consist of the orthogonal projections to K e r D l   and K e r D l *   in L 2 ( l )   accordingly, define elements of the von Neumann algebra of the foliation W * ( M , )   . Suppose that   has a holonomy invariant transverse measure Λ   . Then a normal semi-finite trace t r Λ   on W * ( M , )   is defined. It is proved in [34that the dimensions
dim Λ ( ( K e r D l ) l M / ) = t r Λ ( ( P K e r D l ) l M / ) , dim Λ ( ( K e r D l * ) l M / ) = t r Λ ( ( P K e r D l * ) l M / ) ,  
are finite, and, therefore, the index of D   is defined by ind Λ ( D ) = dim Λ ( ( K e r D l ) l M / ) dim Λ ( ( K e r D l * ) l M / ) .   Like in [5, the tangential principal symbol σ D   of D   defines an element of K 1 ( T * )   . Using the Thom isomorphism (for simplicity, assume that the bundle T   is orientable), one can define the Chern character ch ( σ D )   as an element of the rational cohomology group H * ( M , Q )   of the manifold M   . Let Td ( T * ) H * ( M , Q )   denote the Todd class of the cotangent bundle T *   to   .
Theorem 5.10. [34One has the formula ind Λ ( D ) = ( 1 ) p ( p + 1 ) / 2 C , ch ( σ D ) Td ( T * ) ,   where C   is the Ruelle-Sullivan current, corresponding to the transverse measure Λ   .
Theorem  5.10 is completely similar to the Atiyah-Singer index theorem in cohomological form [4with the only difference that one uses here the pairing with the Ruelle-Sullivan current C   instead of the integration over the compact manifold in the right-hand side of the Atiyah-Singer formula.
An odd version of Theorem  5.10 is proved in the paper [64(see also. [63). It is related with the index theory for Toeplitz operators.
In the paper [55(see also [37), a K   -theoretic version of the index theorem for tangentially elliptic operators on an arbitrary compact foliated manifold ( M , )   is proved. Let D   be a tangentially elliptic operator on a compact manifold M   . One can construct an analytical index I n d a ( D ) K 0 ( C * ( G ) )   of the operator ( D l ) l M /   , using operator constructions ([34, 37), and, starting with the class [ σ D ] K 1 ( T * )   defined by the tangential principal symbol σ D   of the operator D   , one can construct its topological index I n d t ( D ) K 0 ( C * ( G ) )   .
Theorem 5.11. [55, 37For any tangentially elliptic operator D   on a compact foliated manifold ( M , )   , one has the formula I n d a ( D ) = I n d t ( D ) .  
If   is given by the fibers of a fibration M B   , then K 0 ( C * ( G ) ) = K 0 ( B )   , and Theorem  5.11 is reduced to the Atiyah-Singer theorem for families of elliptic operators [5.
In [95(see also [96), Heitsch and Lazarov proved an analogue of the Atiyah-Bott-Lefschetz fixed point formula [3in the setting of Theorem  5.10 .
More precisely, they consider a compact foliated manifold ( M , )   , equipped with a holonomy invariant transverse measure Λ   , and a diffeomorphism f : M M   , which takes each leaf of the foliation to itself. They assume additionally that the fixed point sets of f   are submanifolds, transverse to the foliation. Under these assumptions, the Lefschetz theorem proved in [95states the equality of the alternating sum of the traces of the action of f   on the L 2   spaces of harmonic forms along the leaves of the foliation (an appropriate trace is given by t r Λ   ) with the average with respect to Λ   of local contributions of the fixed point sets.
In [12, an equivariant generalization of Theorem  5.11 to the case when there is a compact Lie group action, taking each leaf of the foliation   to itself, is obtained. As a consequence, the author extended the Lefschetz theorem proved in [95to the case of an arbitrary tangentially elliptic complex when the diffeomorphism f : M M   is included to a compact Lie group action, taking each leaf of   to itself.
Finally, in [42,ChapterIII,Section7. γ ,Corollary13, a cyclic version of the index theorem for tangentially elliptic operators is given.
Theorem 5.12. Let D   be a tangentially elliptic operator on a compact manifold with a transversely oriented foliation ( M , )   , I n d a ( D )   its analytic index. Then, for any ω H * ( B G )   , we have Φ * ( ω ) , I n d a ( D ) = ω , c h τ ( σ D ) .  
Here Φ * : H * ( B G ) H P * ( C c ( G , | T G | 1 / 2 ) )   is the map introduced in Section  5.6 , c h τ ( σ D )   denotes the twisted Chern character c h τ ( σ D ) = T d ( τ ) 1 c h ( σ D ) ,   where c h : K * ( B G ) H * ( B G , Q )   is the Chern character.
Let us also mention the papers [139, 94, 97, 78, 79, concerning to generalizations of the local index theorem for families of elliptic operators [18to the case of tangentially elliptic operators on a foliated manifold, and the papers [11, 13, 14, concerning to cyclic versions of the Lefschetz formula for diffeomorphisms, which take each leaf of a foliation to itself.

6 Noncommutative differential geometry

In this Section, we will describe noncommutative analogues of two of the most important transverse geometric structures for foliated manifolds:
symplectic and Riemannian ones.

6.1 Noncommutative symplectic geometry

Based on the ideas of the deformation theory of Gerstenhaber [71, Ping Xu [186and Block and Getzler [20introduced a noncommutative analogue of the Poisson bracket. Namely, they defined a Poisson structure on an algebra A   as a Hochschild 2   -cocycle P Z 2 ( A , A )   such that P P   as a Hochschild 3   -coboundary, P P B 3 ( A , A )   . In other words, a Poisson structure on A   is given by a linear map P : A A A   such that
( δ P ) ( a 1 , a 2 , a 3 ) a 1 P ( a 2 , a 3 ) P ( a 1 a 2 , a 3 ) + P ( a 1 , a 2 a 3 ) P ( a 1 , a 2 ) a 3 = 0 , (6.1)
and there is a 2   -cochain P 1 : A A A   such that
(6.2) P P ( a 1 , a 2 , a 3 ) P ( a 1 , P ( a 2 , a 3 ) ) P ( P ( a 1 , a 2 ) , a 3 ) = a 1 P 1 ( a 2 , a 3 ) P 1 ( a 1 a 2 , a 3 ) + P 1 ( a 1 , a 2 a 3 ) P 1 ( a 1 , a 2 ) a 3 .
The identity ( 6.1 ) is an analogue of the Jacobi identity for a Poisson bracket, and the identity ( 6.2 ) is an analogue of the Leibniz rule.
A connection of this definition with the deformation theory consists in the following fact. Let us call by a formal deformation of an algebra A   an associative multiplication on the vector space A [ [ ν ] ] = C [ [ ν ] ] A   over the field C [ [ ν ] ]   of formal complex series such that the induced multiplication on A = A [ [ ν ] ] / ν A [ [ ν ] ]   coincides with the multiplication in A   . Such a deformation is described by a cochain m ( a 1 , a 2 ) = i = 0 ν i m i ( a 1 , a 2 ) , a 1 , a 2 A [ [ ν ] ] ,   and, moreover,
  • (1) m m ( a 1 , a 2 , a 3 ) = m ( a 1 , m ( a 2 , a 3 ) ) m ( m ( a 1 , a 2 ) , a 3 ) = 0 ;  
  • (2) m 0 ( a 1 , a 2 ) = a 1 a 2 .  
It easily follows from here that m 1   defines a Poisson structure on A   .
Block and Getzler [20defined a Poisson structure on the operator algebra C c ( G , | T G | 1 / 2 )   of a symplectic foliation   in the case when the normal bundle τ   to   has a basic connection   (recall that a basic connection on τ   is a holonomy invariant adapted connection).
Consider an arbitrary symplectic foliation   . Let ω   be the corresponding closed 2   -form of constant rank. It defines a nondegenerate holonomy invariant 2   -form on the normal bundle τ   , which gives a bundle isomorphism of τ   and τ * = N *   and, therefore, defines a 2   -form Λ   on N *   .
Choose an arbitrary distribution H   on M   , transverse to   . There is defined the transverse differential d H : Ω 0 = C c ( G , | T G | 1 / 2 ) Ω 1 = C c ( G , r * N * | T G | 1 / 2 ) .   The exterior product of differential forms gives a map Ω 1 × Ω 1 Ω 2   (cf. ( 3.5 )). Finally, the pairing with the 2   -form Λ   defines a map Ω 2 Ω 0 : s Λ , s ,   where Λ , s ( γ ) = Λ s ( γ ) , s ( γ ) .   A Poisson bracket P ( k 1 , k 2 )   of k 1 , k 2 C c ( G , | T G | 1 / 2 )   is defined as
P ( k 1 , k 2 ) = Λ , d H k 1 d H k 2 . (6.3)
It is not difficult to check that P   satisfies ( 6.1 ).
A construction of a 2   -cochain P 1 : A A A   , which satisfies ( 6.2 ), is given in [20only in the case when   has a basic connection.

6.2 Noncommutative Riemannian spaces

According to [50, 43, the initial datum of the noncommutative Riemannian geometry is a spectral triple.
Definition 6.1. A spectral triple is a set ( A , , D )   , where:
  • (1) A   is an involutive algebra;
  • (2)   is a Hilbert space, equipped with a *   -representation of A   ;
  • (3) D   is an (unbounded) self-adjoint operator in   such that 1. for any a A   , the operator a ( D i ) 1   is a compact operator in   ; 2. D   almost commutes with elements of A   in the sense that [ D , a ]   is bounded for any a A   .
A spectral triple is called even, if   is endowed with a Z 2   -grading γ ( )   , γ = γ *   , γ 2 = 1   , and, moreover, γ D = D γ   and γ a = a γ   for any a A   . In the opposite case, a spectral triple is called odd.
Spectral triples were considered for the first time in the paper [6, where they were called unbounded Fredholm modules. A spectral triple ( A , , D )   defines a Fredholm module ( , F )   over A   , where F = D ( I + D 2 ) 1 / 2   [6.
On the contrary, any Fredholm module ( , F )   over A   is homotopic to a Fredholm module determined by a spectral triple. In a sense, the operator F   is connected with measurement of angles and is responsible for the conformal structure, whereas | D |   is connected with measurement of lengths.
Definition 6.2. A spectral triple ( A , , D )   is called p   -summable (or p   -dimensional), if, for any a A   , the operator a ( D i ) 1   is an element of the Schatten class p ( )   .
A spectral triple ( A , , D )   is called finite-dimensional, if it is p   -summable for some p   .
The greatest lower bound of all p   's, for which a finite-dimensional spectral triple is p   -summable, is called the dimension of the spectral triple.
As shown in [6, the dimension of the spectral triple ( A , , D )   coincides with the dimension of the corresponding Fredholm module ( , F )   , F = D ( I + D 2 ) 1 / 2   , over A   .
The classical Riemannian geometry is described by the spectral triple ( A , , D )   , associated with a compact Riemannian manifold ( M , g )   :
  • (1) An involutive algebra A   is the algebra C ( M )   of smooth functions on M   ;
  • (2) A Hilbert space   is the space L 2 ( M , Λ * T * M )   , on which the algebra A   acts by multiplication;
  • (3) An operator D   is the signature operator d + d *   .
Let us show that this spectral triple contains a basic geometric information on the Riemannian manifold ( M , g )   . First of all, it is finite-dimensional, and the dimension of this spectral triple coincides with the dimension of M   .
This fact is an immediate consequence of the Weyl asymptotic formula for eigenvalues of self-adjoint elliptic operators on a compact manifold.
For any T ( )   , denote by δ i ( T )   the i   -th iterated commutator with | D |   . Consider the space O P 0   , consisting of all T ( )   such that δ i ( T ) ( )   for any i N   . Then O P 0 C ( M )   coincides with C ( M )   . This allows to reconstruct the smooth structure of M   , based on its topological structure and the spectral triple ( A , , D )   (observe that here one can take as A   any involutive algebra, which consists of Lipschitz functions and is dense in C ( M )   ).
Based on the spectral triple, one can compute the geodesic distance d ( x , y )   between any two points x , y M   by the formula (cf. [41) d ( x , y ) = sup { | f ( x ) f ( y ) | : f A , [ D , f ] 1 } .   Then, starting from the triple ( A , , D )   , one can reconstruct the Riemannian volume form d ν   , given in local coordinates by d ν = det g d x   . For this, one uses a trace Tr ω   , introduced by Dixmier in [61as an example of a nonstandard trace on the algebra ( )   . Before we describe the Dixmier trace, we introduce some auxiliary notions. Consider the ideal 1 + ( )   in the algebra of compact operators K ( )   , which consists of all T K ( )   such that sup N N 1 ln N n = 1 N μ n ( T ) < ,   where μ 1 ( T ) μ 2 ( T )   are the singular numbers of T   . For any invariant mean ω   on the amenable group of upper triangular 2 × 2   -matrices, Dixmier constructed a linear form lim ω   on the space ( N )   of bounded sequences, which coincides with the limit functional lim   on the subspace of convergent sequences. The trace Tr ω   on 1 + ( )   is defined for a positive operator T 1 + ( )   as Tr ω ( T ) = lim ω 1 ln N n = 1 N μ n ( T ) .   Let M   be a compact manifold, E   a vector bundle on M   and P Ψ m ( M , E )   a classical pseudodifferential operator. Thus, in any local coordinate system, its complete symbol p   can be represented as an asymptotic sum p p m + p m 1 +   , where p l ( x , ξ )   is a homogeneous function of degree l   in ξ   . As shown in [40, the Dixmier trace Tr ω ( P )   does not depend on the choice of ω   and coincides with the value of the residue trace τ ( P )   , introduced by Wodzicki [185and Guillemin [83. The residue trace τ   is defined as follows.
For P Ψ * ( M , E )   , its residue form ρ P   is defined in local coordinates as ρ P = ( | ξ | = 1 T r p n ( x , ξ ) d ξ ) | d x | .   The density ρ P   turns out to be independent of the choice of a local coordinate system and, therefore, gives a well-defined density on M   . The integral of the density ρ P   over M   is the residue trace τ ( P )   of P   :
τ ( P ) = ( 2 π ) n M ρ P = ( 2 π ) n S * M T r p n ( x , ξ ) d x d ξ . (6.4)
Wodzicki [185showed that τ   is a unique trace on the algebra Ψ * ( M , E )   of classical pseudodifferential operators of arbitrary order.
According to [40(cf. also [81), any P Ψ n ( M , E )   ( n = dim M   ) belongs to the ideal 1 + ( L 2 ( M , E ) )   and, for any invariant mean ω   , Tr ω ( P ) = τ ( P ) .   The above results imply the formula M f d ν = c ( n ) Tr ω ( f | D | n ) , f A ,   where c ( n ) = 2 ( n [ n / 2 ] ) π n / 2 Γ ( n 2 + 1 )   . Thus, the Dixmier trace Tr ω   can be considered as a proper noncommutative generalization of the integral.
Finally, the Egorov theorem for pseudodifferential operators allows to describe the geodesic flow on the cotangent bundle T * M   in terms of the given spectral triple (cf. Section  8.4 ).
Example 6.3. Let us give examples of spectral triples ( A , , D )   associated with the noncommutative torus T θ 2   (see [44). These triples are parametrized by a complex number τ   with I m τ > 0   .
Put A = A θ = { a = ( n , m ) Z 2 a n m U n V m : a n m S ( Z 2 ) } .   Define a canonical normalized trace τ 0   on A θ   as τ 0 ( a ) = a 00 , a A θ .   Let L 2 ( A θ , τ 0 )   be the Hilbert space, which is the completion of A θ   in the inner product ( a , b ) = τ 0 ( b * a ) , a , b A θ   . The Hilbert space   is defined as the sum of two copies of L 2 ( A θ , τ 0 )   , equipped with the grading, given by γ = ( 1 0 0 1 )   .
The representation ρ   of A θ   in   is given by the left multiplication, that is, for any a A θ   ρ ( a ) = ( λ ( a ) 0 0 λ ( a ) ) ,   where the operator λ ( a )   is given on A θ L 2 ( A θ , τ 0 )   by λ ( a ) b = a b , b A θ .   Introduce the differentiations δ 1   and δ 2   on the algebra A θ   by δ 1 ( U ) = 2 π i U , δ 1 ( V ) = 0 ; δ 2 ( U ) = 0 , δ 2 ( V ) = 2 π i V .   The operator D   depends explicitly on τ   and has the form D = ( 0 δ 1 + τ δ 2 δ 1 τ ¯ δ 2 0 ) .   The triples constructed above are two-dimensional smooth spectral triples.
Finally, we describe the simplest example of a spectral triple associated with a closed manifold M   , equipped with a Riemannian foliation   . Fix a bundle-like metric g M   on M   . Let H = F   be the orthogonal complement of F = T   .
Define a triple ( A , , D )   as follows:
  • (1) A = C c ( G )   ;
  • (2) = L 2 ( M , Λ * H * )   is the space of transverse differential forms, endowed with the natural action R Λ * H *   of A   ;
  • (3) D = d H + d H *   is the transverse signature operator (cf. Section  2.7 ).
Theorem 6.4. [113 The spectral triple ( A , , D )   is a finite-dimensional spectral triple of dimension q = codim   .
More general examples of spectral triples given by transversally elliptic operators on foliated manifolds are defined later in Section  8 . Before we turn to the study of properties of spectral triples associated with Riemannian foliations, we need to have an appropriate pseudodifferential calculus, that is a subject of Section  7 .

6.3 Geometry of para-Riemannian foliations

Definition 6.5. A foliation   on a manifold M   is called para-Riemannian, if there is an integrable distribution V   on M   , which contains the tangent bundle T   to   such that there are holonomy invariant Euclidean structures in the fibers of the bundles T M / V   and V / T   .
If V   is an integrable distribution on M   , which defines a para-Riemannian structure, and V   is the corresponding foliation on M   , V   , then V   is Riemannian, and the restriction of   to every leaf of V   is a Riemannian foliation.
An interest in studying of para-Riemannian foliations is related with the fact that, in some cases, the study of arbitrary foliations can be reduced to the case of para-Riemannian foliations. A basic observation consists in the following. Let ( M , )   be an arbitrary foliated manifold. Consider the bundle P   over M   , whose fiber P x   at x M   is the space of all Euclidean metrics on the vector space τ x = T x M / T x   . There is a natural lift of   to a foliation V   on P   , moreover, this foliation is para-Riemannian. For the first time, this construction was used by Connes [38in order to extend the map K 0 ( C c ( G , | T G | 1 / 2 ) ) C   given by the transverse fundamental class of an arbitrary foliation ( M , )   to a map K 0 ( C r * ( G ) ) C   (cf. Section  5.4 ).
Connes constructed such an extension for ( P , V )   , using the para-Riemannian condition, and then, using the Thom isomorphism, he showed that the map K 0 ( C r * ( P , V ) ) C   given by the transverse fundamental class of ( P , V )   defines the desired extension K 0 ( C r * ( G ) ) C   for the initial foliation ( M , )   .
In [100, Hilsum and Skandalis constructed a Fredholm module associated with an arbitrary para-Riemannian foliation. To do this, they made use of transversally hypoelliptic operators and pseudodifferential operators of type ( ρ , δ )   .
In [50, Connes and Moscovici described a spectral triple associated with a para-Riemannian foliations. More precisely, they considered the closely related setting (strongly Morita equivalent), when there is a manifold equipped with an action of a discrete group, preserving a triangular structure.
Let us describe the construction of Connes and Moscovici. Let W   be an oriented smooth manifold, and Γ   a group of diffeomorphisms of W   . Consider the bundle π : P ( W ) W   , whose fiber P x ( W )   at x W   is the space of all Euclidean metrics on the vector space T x W   . Thus, a point p P ( W )   is given by a point x W   and a nondegenerate quadratic form on T x W   .
Let F + ( W )   be the bundle of positive frames in the bundle W   , whose fiber F x ( W )   at x W   is the space of orientation preserving, linear isomorphisms R n T x W   . Equivalently, the bundle P ( W )   can be described as the quotient of the bundle F + ( W )   by the fiberwise action of the subgroup S O ( n ) G L ( n , R )   , P ( W ) = F + ( W ) / S O ( n )   . We will use the natural invariant Riemannian metric on the symmetric space G L + ( n , R ) / O ( n )   , given by the matrix Hilbert-Schmidt norm on the tangent space to G L ( n , R ) / S O ( n )   , which is identified with the space of symmetric n × n   -matrices. If we transfer this metric to the fibers P x   of the fibration P ( W ) = P   , we get a Euclidean structure on the vertical distribution V T P   . The normal space N p = T p P / V p   is naturally identified with the space T x W , x = π ( p )   .
Thus, the quadratic form on