Arxiv:1912.02148V2 [Math.DG] 14 Jun 2021 Regular Foliations Appear Frequently in Diﬀerential Geometry

INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS

ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

Abstract. We give a new construction of the holonomy and fundamental groupoids of a singular foliation. In contrast with the existing construction of Androulidakis and Skandalis, our method proceeds by taking a quotient of an inﬁnite dimensional space of paths. This strategy is a direct extension of the classical construction for regular foliations and mirrors the integration of Lie algebroids via paths (per Crainic and Fernandes). In this way, we obtain a characterization of the holonomy and fundamental groupoids of a singular foliation that more clearly reﬂects the homotopic character of these invariants. As an application of our work, we prove that the constructions of the fundamental and holonomy groupoid of a foliation have functorial properties.

Contents Introduction 1 1. Foliations 3 2. Diﬀeology 4 3. -paths 8 4.F Comparison to Crainic-Fernandes Integration 15 5. Comparison to the Androulidakis-Skandalis construction 17 6. Functoriality 22 References 30

Introduction A regular foliation, is an integrable subbundle D, TM of the tangent bundle of a smooth manifold. The Frobenius theorem tells us→ that this is equivalent to the partition of M into immersed submanifolds which satisfies a certain local normal form. arXiv:1912.02148v2 [math.DG] 14 Jun 2021 Regular foliations appear frequently in differential geometry. For example: the fibers of a submersion or the orbits of some particularly nice Lie group actions. However, it is often the case that one must consider situations where the action is not nice or the map is not quite a submersion. This requires the consideration of distributions which may change in rank and therefore a suitable generalization of regular foliations. Along the lines of previous authors [1][8][13][17][22][25], we define a foliation to be a sheaf of Lie algebras which is also a submodule of the sheaf of vector fields satisfying some finiteness conditions (Definition 1.0.3). Singular foliations are closely related to the study of Lie algebroids. From a given Lie algebroid, one can construct a singular foliation by considering the sheaf 1 2 ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN of vector fields which arise as the image of the anchor map. On the other hand, given a leaf of a singular foliation, one can define a transitive Lie algebroid [2]. Two of the most useful tools for studying regular foliations are the associated fundamental groupoids and holonomy groupoids. These groupoids are invariants of the foliation which reflect the global properties of the leaves. When the foliation is regular it can be regarded as a Lie algebroid and both of these groupoids constitute integrations of this Lie algebroid structure. For the singular case, Androulidakis and Skandalis [1] constructed the holonomy groupoid. Although it corresponds with the standard object in the regular case, the construction bears little resemblance to the classical method for defining this invariant. The goal of this paper is to provide an alternative construction of the fundamental groupoid and holonomy groupoid of a singular foliation. Our method admits clear analogies with the classical constructions as well as with the integration of Lie algebroids per Crainic and Fernandes [6]. Another path-based approach to constructing the fundamental groupoid of a singular foliation appears in a paper [17] by Laurent-Gengoux, Lavau and Strobl and later in [18] by Laurent-Gengoux and Ryvkin. In [17] work, we see that if the foliation arises as the image of a vector bundle, then one can imitate the Crainic and Fernandes construction to obtain a topological groupoid which integrates each leaf-wise algebroid. In this paper we are faced with two distinct kinds of generalizations of manifolds: infinite dimensional mapping spaces and singular quotients. It is therefore convenient to choose a formalism of generalized manifold which encompasses both possibilities. The language of diffeology is suitable for this task without requiring much significant technical development. Therefore, this is the framework we will use for describing our manifold-like objects.

Constructing the holonomy groupoid. In Section 3.1, we explain our construction of the holonomy groupoid. We define an -path to be a time-dependent element X(t) of together with an integral curveFγ(t). Using the flow of X, one F can obtain the germ of a diffeomorphism from a transversal T0 through γ(0) to T1 through γ(1). However, there is ambiguity in this construction. So the ‘holonomy’ is actually the germ of an element of Diff(T0,T1) up to an equivalence relation. This equivalence relation is precisely the one used by Androlidakis and Zambon in [3]. The holonomy groupoid is therefore obtained by identifying -paths which are holonomic. F In Section 5 we show that our construction is diffeomorphic (as a diffeological groupoid) to the previous construction by Androulidakis and Skandalis. Although the proof involves diffeological spaces and is somewhat involved, it does not require more than a basic understanding of the subject of diffeology.

Constructing the fundamental groupoid. In Section 3.2 we construct the fundamental groupoid of a singular foliation. Taking inspiration from Crainic- Fernandes style integration, the key notion is that of -homotopy of -paths. The fundamental groupoid of is therefore deﬁned to beF the set of -pathsF up to -homotopy. It turns out thatF two -paths are holonomic wheneverF they are homotopic.F This way, one can view theF holonomy groupoid as a quotient of the fundamental groupoid, which correctly reﬂects the classical relationship between the two objects. INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 3

As we mentioned before, given a leaf L of a singular foliation one can naturally deﬁne an integrable Lie algebroid AL L. In Section 4, we show that (leaf-wise), the fundamental groupoid of a singular→ foliation is the same as the source simply connected integration of this algebroid. This clariﬁes the close connection between the notion of -homotopy and A-homotopy. F Functoriality. In the last section, we show that the construction of the fundamental groupoid is functorial in the same way that the path integration of a Lie algebroid is functorial. However, in order to state the result, we are forced to develop a suitably general notion of morphism of foliated manifolds. To do this, we use the notion of a comorphism of sheaves of modules previously discussed by Higgins and Mackenzie [14]. Acknowledgements. The authors would like to thank Marco Zambon for his advice throughout this project. We would also like to thank Rui Fernandes for his suggestions and comments. We would also like to thank Dan Christensen for his helpful remarks on an earlier draft. This work was partially funded by FWO EOS project G0H4518N and FWO Project G083118N.

1. Foliations ∞ Let CM denote the sheaf of smooth functions on a manifold M. In this text, a ∞ singular foliation on M is a well-behaved CM -submodule of the sheaf of vector fields X. The two key conditions are that should be locallyF finitely generated and involutive. Throughout, M will denoteF a fixed smooth manifold. ∞ Definition 1.0.1 (Locally Finitely Generated). A sheaf of CM -modules is said to be locally finitely generated if for all x M, there exists an open set x FU M, a natural number n 0 and a surjective∈ morphism of C∞(U)-modules: ∈ ⊆ ≥ M C∞(U)n (U) M F This definition is equivalent to the existence of a finite set of sections s1, . . . , sN (U) which generate (U) over C∞(U). ∈ F F M ∞ Definition 1.0.2 (Involutive). A CM submodule , X of the sheaf of vector fields on M is said to be involutive if for each open setF U→, the image of the inclusion (U) , X(U) is a Lie subalgebra. F → Definition 1.0.3 (Foliation). A (singular) foliation on a manifold M is a locally finitely generated and involutive C∞-submodule , X. A manifold equipped with such an object is called a foliated manifold. F → By Hermann [13], Stefan [22] and Sussman [23], singular foliations partition M into immersed submanifolds (called leaves) which are generated by the flows of vector fields inside of the foliation. The singular distribution obtained from the tangent spaces to the leaves is called the characteristic distribution of . Although we do not assume that is projective (i.e. the sheaf ofF sections of some vector bundle), we can still defineF something which resembles a vector bundle: ∞ Definition 1.0.4. Let be a CM -module and let x M be a point. We denote the ideal of smooth functionsF which vanish at x M by∈ I . The fiber of at x is: ∈ x F (M) A( ) := F F x I (M) xF 4 ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

∞ Since CM (M)/Ix is canonically isomorphic to R, the fiber of at x is a real vector space. The fiberspace of is the disjoint union of the fibers:F F G A( ) := A( ) F F x x∈M Many properties of can be determined from A( ). For instance, is finitely F F F generated in a neighborhood of x M if and only if A( )x is a finite dimensional vector space. Furthermore, if is∈ locally finitely generated,F then is projective if and only if the dimension of AF( ) is independent of x M. F F x ∈

2. Diffeology Since we are interested in considering time dependent sections of a singular foliation, it is important to make sense of the notion of a smooth map into (U). We will approach this problem by using the language of diﬀeology. F

2.1. Basics. We will give a brief overview of the main concepts from diffeology that we require. Diffeological spaces were first introduces by Souriau in [20][21]. For a more thorough treatment, we refer the reader to [16].

Definition 2.1.1 (Diffeological Space). A diffeology on a set X is an assignment to any d N and a non-empty open U Rd, a subset (U) XU of elements called plots∈ such that: ⊆ D ⊆ (a) if f XU is constant, then f (U); ∈ ∈ D (b) for all V Rn, f (U) and g C∞(V,U), then f g (V ); U⊆ ∈ D ∈ ◦ ∈ D (c) if f X and there exists an open cover Ui i∈I of U such that f Ui (U∈) for all i I, then f (U). { } | ∈ D i ∈ ∈ D A diffeological space is a pair (X, ) where X is a set and is a diffeology on X. Given diffeological spaces (X, D) and (Y, 0), a function Df : X Y is called smooth if f 0. The set ofD smooth functionsD from X to Y will→ be denoted C∞(X,Y ). ◦ D ⊆ D

Any smooth manifold is canonically a diﬀeological space and a function between smooth manifolds is smooth in the diﬀeological sense if and only if it is smooth in the traditional sense.

Deﬁnition 2.1.2 ([16](I.46)). Suppose X and Y are diﬀeological spaces and f : X Y is smooth. Then f is called a subduction if for all plots φ: U Y and points→ → p U, there exists an open neighborhood p U 0 U and a plot φe: U 0 X such that:∈ ∈ ⊆ →

f φe = φ U 0 (1) ◦ | A local subduction is the case where the lift can arranged to pass through a speciﬁed point. That is, for all x X, given a plot φ: U Y such that φ(v) = f(x) for ∈ → some v U, there exists an open neighborhood U 0 U of v and a lift φe: U 0 X ∈ ⊆ → such that φe(v) = x and Equation (1) holds. If X and Y are smooth manifolds, then f : X Y is a local subduction if and only if it is a submersion. Now let us look at a few→ examples of diﬀeological spaces. INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 5

Example 2.1.1 (Intersection [16](I.22)). Suppose i i∈I is a set of diffeological structures on X indexed by I. Then one can define{D the} intersection diffeology ! \ \ (U) := (U) Di Di i∈I i∈I Example 2.1.2 (Products [16](I.55)). Suppose X and Y are diffeological spaces. Then the product diffeology on X Y is defined to be the intersection of all diffe- × ologies on X Y such that for all plots φ: U1 X and ψ : U2 Y the function φ ψ : U U× X Y is a plot. → → × 1 × 2 → × Example 2.1.3 (Mapping Spaces[16](I.13)). Let X and Y be diffeological spaces. Then C∞(X,Y ) is also a diffeological space. A function f : U C∞(X,Y ) is a plot if and only if the associated function U X Y given by (→u, x) f(u)(x) is smooth. × → 7→ Example 2.1.4 (Quotient Diffeology[16](I.50)). Suppose X is a diffeological space and is an equivalence relation on X. Let π : X X/ denote the quotient function.∼ Then the quotient diffeology on X/ is the→ intersection∼ of all diffeologies on X/ for which π : X X/ is diffeological∼ morphism. This∼ is equivalent to saying→ ∼ that the diffeology on X/ is the unique one which makes π a subduction. ∼ Example 2.1.5 (Subset Diffeology[16](I.33)). Suppose Y X and X is equipped with a diffeology and let ι: Y X be the inclusion map.⊆ Then we can define a → plot on Y to be any function f : Rn Y such that ι f is a plot on X. If X is a smooth manifold and Y,→ X is an immersed◦ submanifold. Then the diffeology coming from the smooth structure→ on Y agrees with the subset diffeology on Y if and only if Y is an initial submanifold. Example 2.1.6 (Sections of a vector bundle). Suppose E M is a smooth vector → bundle over a manifold M. Let ΓE denote the sheaf of smooth sections of E. ∞ We have an inclusion ΓE(M) C (M,E). Therefore, ΓE(M) inherits a subset diffeology from the mapping space⊆ diffeology on C∞(M,E). From now on, we will take the diffeology on the sections of a vector bundle to be implicit. 2.2. The coefficient diffeology. We wish to generalize the last example to sec- ∞ tions of a CM -module. The basis of the definition will be that we want the smallest diffeology with the following properties: It agrees with the diffeology on sections of a vector bundle, it makes module homomorphisms smooth, and is compatible with the restriction and gluing sheaf operations. To our knowledge this diffeology has not appeared before in the literature.

∞ Definition 2.2.1 (Coefficient Diffeology). Let be a CM -module. Given a subset of some Euclidean space U, a function φ: U F (M) is a plot if and only if it locally factors through the sections of a vector→ bundle. F In other words: φ is a plot, if and only if for all u0 U and p0 M, there exist open neighborhoods U 0 U and V M of u and p ∈together with:∈ ⊆ ⊆ 0 0 a vector bundle E V , • ∞→ a morphism of CM -modules T :ΓE(V ) (V ), • 0 → F and a smooth function φe: U ΓE(V ) • → 6 ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN such that the following diagram commutes:

ΓE(V ) ∃φe T

φ| 0 U 0 U (M) (V ) F F This is called the coefficient diffeology on (M). F The next lemma gives another characterization of the coefficient diffeology. Lemma 2.2.1. Suppose is a sheaf of C∞-modules. Let (M) be equipped with F M F the coefficient diffeology. Then, φ: U (M) is a plot if and only if for all u0 U → F 0 ∈ and p0 M there are open neighborhoods U U of u0 and V M of p0, together with ∈ ⊆ ⊆ a finite number of functions: • 1 n ∞ 0 c , . . . , c C (U V, R) ∈ × and elements X ,...,X (V ) • 1 n ∈ F such that: n X u U 0, p V φ(u) = ci(u, p)(X ) (2) ∀ ∈ ∈ p i p i=1 Proof. If we are given a φ as in the statement of the lemma. We want to show that i φ is smooth. Supose around u0 and p0 we have functions c and elements Xi as in the statement of the lemma. Then the following choices witness the fact that∈ F φ is smooth: E := Rn V is the trivial vector bundle of rank n • T :Γ (V ×) C∞(M) is the module homomorphism associated to the choice • E → of elements X1,...,Xn and φe is the smooth function: • 0 1 n φe: U ΓE(V ) φe(u)p := (c (u, p), . . . , c (u, p), p) → On the other hand, for any vector bundle E there exists another vector bundle W such that W E is trivial (Theorem 3.3 in [15]). Therefore, Definition 2.2.1 is unchanged if we⊕ require that E is the trivial bundle. So φ must locally factor through a trivial bundle. The functions ci n in Equation (2) correspond to the { }i=1 coefficients of φe relative to a choice of trivialization. The coefficient diffeology coincides with the mapping space diffeology when ∞ F is the sections of a vector bundle. It also makes all CM -module homomorphisms smooth. Although the definition of the coefficient diffeology is stated in terms of global sections, we can define the diffeology on sections over an arbitrary open subset similarly. In the following example, we see that when is a submodule of the sections of a vector bundle, then the coefficient diffeologyF may not coincide with the subset diffeology.

∂ Example 2.2.1. Let M = R and = f ∂t X(R) where f : R R is a smooth function such that f(x) = 0 for xF 0h andif ≤(t) > 0 when x > →0. Consider the ≤ INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 7 smooth function X : R (R) such that → F ∂ X(t)(x ) = f(x t2) 0 0 ∂x − x0 Then X fails to be smooth relative to the coefficient diffeology on . On the other hand, X is smooth as a time-dependent vector field, i.e. as a smoothF map R X(R). → ∞ From now on, all CM -modules shall be implicitly equipped with the coefficient diffeology.

2.3. Diffeology and the fiberspace. An important application of the coefficient diffeology is that it can be used to define a diffeology on the fiberspace. The contents of this subsection is not needed to understand the constructions in Section 3 where we construct the holonomy groupoid and fundamental groupoid of a singular foliation. However, it will be crucial for understanding Sections 4 and 6 where we study the relationships of these constructions to algebroid homotopy and their functoriality properties, respectively.

∞ Definition 2.3.1. Given a CM -module recall the fiberspace A( ) from Defini- tion 1.0.4. Since the fiber of at x is obtainedF from an equivalenceF relation on (M), the evaluation map isF the surjective function: F ev: (M) M A( ) F × → F (X, p) ev (X) 7→ p which sends (X, p) to the class of X in A( )p. The diffeology on A( ) is the unique diffeology which makes the evaluationF map a subduction. F Remark. There is a slight difference in notation here between the evaluation map that appears in [1] and the evaluation map just defined. Note that the evaluation map used by Androulidakis and Skandalis is only well defined for a foliation whereas the definition above makes sense for an arbitrary sheaf of modules. The diffeology on the fiberspace makes π : A( ) M into a local subduction. If we think of (M) M as a trivial vector bundleF → of infinite rank, the rescaling, addition, and zeroF section× operations on (M) M descend to smooth maps. In the terminology of Christensen and Wu[5]F the× set A( ) is a diffeological vector pseudo-bundle1. The pseudo- prefix is motivated by the factF that diffeological vector bundles as defined here need not be locally trivial. We will include diffeological vector pseudo-bundles here but we will not really need it until Section 6. Definition 2.3.2. A diffeological vector pseudo-bundle over a diffeological space X consists of a smooth functionπ : E X of diffeological spaces together with three smooth functions: →

E M E E (v, w) v+w, R E E (, v) v, M E x 0x × → 7→ × → 7→ · → 7→ which make each fiber of π into an R-vector space. A morphism of diffeological vector pseudo-bundles F : E W covering f : X Y is a smooth function such that π F = f π and F is fiber-wise→ linear. → ◦ ◦

1In some articles [4] the term vector space is also used for the same object 8 ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

In [2] it was observed that the restriction of the ﬁberspace to a leaf is a Lie algebroid. The next lemma tells us that this Lie algebroid structure agrees with the associated subspace diﬀeology.

Lemma 2.3.1. Suppose , XM is a foliation on M and L, M is a leaf. Then the set: F → → A( ) := π−1(L) A( ) F L ⊆ F is a Lie algebroid over L when equipped with the subset diffeology. Proof. Since the leaf of a singular foliation is an initial submanifold. It follows that the smooth structure on L is the subset diffeology relative to the inclusion L, M. This ensures that the projection π : A( ) L is smooth. → |L F L → We will now show that the subset diffeology on A( )L makes it a smooth manifold by exhibiting local trivializations. Suppose p LFis an arbitrary point in the leaf in question. Let X ,...,X be a set of sections∈ of (M) which represent a 1 n F basis for A( )p. Nakayama’s Lemma implies that X1,...,Xn are a minimal set of generators inF some open neighborhood U of p. Now consider the following smooth function: n φ: R U A( )U × → F X (c1, . . . , cn, q) ci ev (X ) 7→ q i i This function fits into a commuting square: C∞(U)n U (U) U M × F × ev ev

n φ R U A( )U × F Since X1,...,Xn generate (U), the top arrow is a subduction. The evaluation maps are also subductions.F Therefore φ is a surjective subduction. On the other hand, since the rank of A( ) is constant, the restriction F L n φ U∩L : R (U L) A( )U∩L | × ∩ → F is a bijective subduction (a diﬀeomorphism of diﬀeological spaces). To see the algebroid structure, let S : (M) ΓA(F)L (L) denote the function which sends an element X (M) to theF associated→ section of A( ) . The bracket on sections of ∈ F F L A( )L and the anchor map A( )L TL are uniquely determined by the following equations:F F →

[S(X),S(Y )] = S([X,Y ]F ) ρ(ev (X)) = X T L A(F) p p ∈ p 3. -paths F From now on we assume that is a singular foliation and that (M) is equipped with the coefficient diffeology. F F Definition 3.0.1. Given an open set U M. A smooth function: ⊆ X : [0, 1] (U) t X(t) → F 7→ is called a time-dependent element of (U). Given such a time-dependent element, the notation Φt : U U will denoteF the flow of X (when it exists). X → INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 9

Closely related to the notion of a time-dependent element is the notion of an - path. This is the main object which we will use in our construction of the holonomyF and fundamental groupoids. Definition 3.0.2 ( -Path). An -path is a pair (X, p) where X is a time- dependent element ofF (M) and p isF an integral curve of X. The set of all -paths will be denoted by (F ). F The source of (PX,F p) is defined to be p(0) M while the target is p(1). The source and target maps define subductions s, t∈: ( ) M. P F → The next lemma tells us that the flow of a time-dependent element of is compatible with the coefficient diffeology on (M). F F t Lemma 3.0.1. Suppose (X, p0) is an -path and the flow ΦX of X exists for all t [0, 1]. Then F ∈ t (ΦX )∗ : X(M) X(M) t → satisfies (Φ )∗( ) = and the following function is smooth: X FM FM t [0, 1] (M) (M)(t, Y ) (Φ )∗Y × F → F 7→ X Proof. For this proof we use the theory of derivations on a vector bundle (sometimes called covariant differential operators) [19]. The general strategy of the proof is inspired by the similar proof (for constant vector fields) in [1] and [10]. The property that we wish to show is local in M. Therefore, we can assume without loss of generality that (M) is finitely generated. Fn n Suppose Xi i=1 (M) is a set of generators. Let R be the trivial vector { } ⊆ F n n bundle of rank n and let ei be the canonical basis. Define Ψ: R TM to { }i=1 → be the unique vector bundle morphism such that Ψ(ei) = Xi. Since X : [0, 1] (M) is smooth and finitely generated, for each 1 i, k n the definition of the→ Fcoefficient diffeology tellsF us that there exist smooth≤ functions≤ ak : [0, 1] C∞(M) such that: i → M n X k [X(t),Xi] = ai (t)Xk k=1 n Let (Dt,X(t)) be the unique time-dependent derivation on R such that: n t X k D (ei) = ai (t)ek k=1 t n n Let Φ(D,X) : R R denote the flow of this derivation. t → The flow ΦX : M M can be thought of as the flow of the time-dependent derivation ([X, ],X) and→ from the definition of D it follows that: · n e Γ(R ), Ψ D(e) = [X, Ψ(e)] (3) ∀ ∈ ◦ n We can now finish the proof. For any Y (M), there exists a lift Ye Γ(R ). By Equation (3), we have that: ∈ F ∈ t t (Φ )∗Y = Ψ (Φ )∗Ye X ◦ (D,X) t Since the image of Ψ is (M) it follows that (ΦX )∗Y (M). Furthermore, the F t ∈ F right hand side gives a factorization of t (ΦX )∗Y through a vector bundle. This t 7→ t implies that t (Φ )∗Y is smooth and therefore (t, Y ) (Φ )∗Y is smooth. 7→ X 7→ X 10ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

The set of -paths has a natural monoidal structure. Let us fix a reparameteriza- tion of the intervalF r : [0, 1] [0, 1] such that r is constant in an open neighborhood of the endpoints. → Definition 3.0.3. Suppose (X, p) and (Y, q) are -paths such that p(1) = q(0). Then we can define the concatenation: F (Y, q) (X, p) := (Y X, q p)

( 2r0(2t)X(r(2t)) t [0, 1 ] (Y X)(t) := ∈ 2 2r0(2t 1)Y (r(2t 1)) t ( 1 , 1] − − ∈ 2 ( p(r(2t)) t [0, 1 ] (q p)(t) := ∈ 2 q(r(2t 1)) t ( 1 , 1] − ∈ 2 Remark. Concatenation is compatible with flow in the following sense: Given an arbitrary -path (X, p), we know that the time-one flow of X exists in a neighborhood of p(0).F Furthermore, given two -paths (X, p) and (Y, q), then: F Φ1 = Φ1 Φ1 Y X Y ◦ X wherever the relevant flows exist. This means that the monoidal structure on - paths is compatible with the groupoid of germs of diffeomorphisms over M. F The goal for the rest of this section will be to define two equivalence relations on ( ), called holonomy and homotopy, for which the concatenation descends to anP honestF (diffeological) groupoid structure on the equivalence classes.

3.1. Holonomy. The notion of holonomy for classical (i.e. regular) foliations comes from constructing germs of diffeomorphisms between slices (i.e transverse submanifolds) out of paths tangent to the leaves. In the singular case, this approach is complicated by the increased ambiguity when choosing slices. Following the example of Androulidakis and Zambon [3], we correct this problem by operating modulo a subgroup of diffeomorphisms which control this ambiguity. Definition 3.1.1 (Slices). Suppose (M, ) is a foliated manifold. A slice through x M is an embedded submanifold S,F M, containing x, such that for all ∈ → y S, we have that TyM = TyS + TyLy where Ly is the leaf containing y and T ∈S T L = 0. x ∩ x x Now, for all x M, let us choose a slice Sx through x. The transversality property for slices∈ implies that each S inherits a foliation by restriction of x FSx the foliation on M to Sx[3]. Let DiffF (Sx,Sy) denote the set of germs of (foliation preserving) diffeomorphisms S S which map x to y. Finally, exp(I ) x → y xFSx ≤ Diff(Sx,Sx) denotes the subgroup of germs of diffeomorphisms generated by the flows of (possibly time dependent) elements of I . xFSx Definition 3.1.2. The holonomy transformation groupoid is the set:

G DiﬀF (Sx,Sy) HT := exp(Iy S ) x,y∈M F y Elements of HT are called holonomy transformations. INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 11

Sp(0) 1 Sp(1) φX (Sp(0)) y φ1 φ1 (y) Z ◦ X p(0) p(1)

Figure 1. A visualization the holonomy of an -path (X, p). The dashed lines represent integral curves of X whileF the dotted line represents an integral curve of Z.

Remark. The holonomy transformation groupoid is not a totally canonical object since it depends on the choice of a slice Sx at every point x M. When is regular, the foliation on each slice is trivial and the holonomy transformation∈ groupoidF can be identified with the groupoid of germs of diffeomorphisms between slices. Our next step is to define a function Hol: ( ) HT which will associate a holonomy transformation to every -path. ToP stateF the→ definition, we will need to borrow Lemma A.8 from [3]. F Lemma 3.1.1. Suppose S and S0 are two slices through x M. Then there exists a vector field X I and an open neighborhood U M of∈ x such that: ∈ xF ⊆ U Φ1 (S) = U S0 ∩ X ∩ Now suppose we are given an -path (X, p). Then, Lemma 3.0.1, implies that 1 F ΦX (Sp(0)) is a slice through p(1). Furthermore, Lemma 3.1.1, tells us that there exists a vector field Z I and a neighborhood U M of p(0) such that: ∈ p(1)F ⊆ Φ1 Φ1 (S U) S Z ◦ X p(0) ∩ ⊆ p(1) Since Φ1 Φ1 is a local diffeomorphism around p(0) it defines an element of Z ◦ X DiffF (Sp(0),Sp(1)) which represents an element of HT. This forms the basis for our definition of holonomy. Definition 3.1.3. Let (X, p) be a -path. Suppose Z is a time-dependent element of I such that the function F p(1)F Φ1 Φ1 : S S Z ◦ X |Sp(0) p(0) → p(1) is well defined in a neighborhood of p(0) S . The holonomy of (X, p), denoted ∈ p(0) Hol(X, p), is the equivalence class of Φ1 Φ1 in Z ◦ X DiffF (S ,S ) p(0) p(1) HT exp(I ) ⊆ p(1)FSp(1) The holonomy equivalence relation is the equivalence relation generated by the fibers of Hol. 12ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

Theorem 3.1.1. The function Hol: ( ) HT is well-deﬁned. P F → To prove this theorem, we will make use of another lemma from [3]:

Lemma 3.1.2. Suppose X : [0, 1] (M) is a time-dependent element of Ix 1 → F F such that ΦX (Sx) Sx. Then there exists another time-dependent element Y : [0, 1] I of I such⊆ that Φ1 = Φ1 . → xFSx xFSx X |Sx Y See Lemma A.6 in [3] for the proof.

Proof. (of Theorem 3.1.1) By Lemma 3.1.1 and Lemma 3.0.1, we know that for a given -path (X, p) it is always possible to find a Z Ip(1) which satisfies DefinitionF 3.1.3. Therefore, we only need to show the definition∈ F of Hol(X, p) does not depend on the choice of Z. 0 1 1 1 1 Suppose we have Z and Z such that Φ Φ and Φ 0 Φ both yield well- Z ◦ X Z ◦ X defined functions Sp(0) Sp(1). We are finished if we can show that the germ of → 1 1 1 1 −1 1 1 −1 (Φ 0 Φ ) (Φ Φ ) = Φ 0 (Φ ) : S S (4) Z ◦ X ◦ Z ◦ X Z ◦ Z p(1) → p(1) 1 1 −1 at p(1) is an element of exp(I ). Since Φ 0 (Φ ) is the composition of p(1)Fp(1) Z ◦ Z two flows, there exists a time dependent element Ze of (M) which takes values in 1 1 1 −1 F Ip(1) and such that Φ = Φ 0 (Φ ) . F Ze Z ◦ Z By Lemma 3.1.2, it follows that there exists a time-dependent element Zˆ of I such that: p(1)Fp(1) 1 1 1 1 −1 Φ = Φ = Φ 0 (Φ ) Zˆ Ze Z ◦ Z which shows that the function (4) is an element of exp(I ). p(1)Fp(1) The argument in the preceding proof can adapted to establish a sufficient condition for trivial holonomy. Lemma 3.1.3. Suppose (X, p) is an -path such that X is a time-dependent el- F ement of Ip(0) . Then the holonomy of (X, p) is trivial. That is: Hol(X, p) = 1 HT. F p(0) ∈ Proof. Let Z be a time-dependent element satisfying Definition 3.1.3. Then the holonomy of (X, p) is the equivalence class of Φ1 Φ1 in HT. Since both Z ◦ X |Sp(0) Z and X take values in I , we know that Φ1 Φ1 is the flow of some time p(0)F Z ◦ X dependent element Ze of Ip(0) . By Lemma 3.1.2 we can conclude that it is also the flow of a time dependent elementF of I . p(0)FSp(0) Theorem 3.1.2. The holonomy equivalence relation does not depend on the choice of slices used to construct HT.

0 0 Proof. Suppose we choose alternative slices Sx through each x M. Let HT denote the holonomy transformation groupoid relative to these choices∈ and Hol0 : ( ) HT0 be the resulting holonomy map. By Lemma 3.1.1, we know thatP forF each→ x M there exists a vector field Z I such that Φ1 defines a germ of a ∈ x ∈ xF Zx diffeomorphism from S to S0 . Then for each x, y M we get a bijection: x x ∈ −1 Diff(S ,S ) Diff(S0 ,S0 ) f Φ1 f (Φ1 ) x y → x y 7→ Zy ◦ ◦ Zx INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 13

1 Since each ΦZx preserves the foliation (Lemma 3.0.1), we obtain a bijection Ψ: HT 0 → HT . For each x M, let (Zx, x) be the constant -path. Then we have that for all (X, p) ( ):∈ F ∈ P F Hol0(X, p) = Hol0 (Z , p(1)) (X, p) ( Z , p(0)) = Ψ Hol(X, p) p(1) − p(0) ◦ where we apply Lemma 3.1.3 for the first equality. Therefore, the holonomy equiv- 0 alence relations relative to Hol and Hol are the same. Definition 3.1.4. The holonomy groupoid, denoted ( ), is defined to be the set of holonomy equivalence classes of ( ) equipped withH F the quotient diffeology. P F The set ( ) is a groupoid over M with product inherited from the concatenation operationH F on -paths. Following Remark 3, we know that this groupoid structure makes theF injection ( ) , HT a groupoid homomorphism. In this way, we can regard ( ) as a (set-theoretic)H F → subgroupoid of HT. H F 3.2. -Homotopy. To define the -homotopy equivalence relation, we need to understandF deformations of -paths.F F Definition 3.2.1. A variation of -paths is a pair (X, p) where X and p are smooth functions: F X : [0, 1]2 (M) p: [0, 1]2 M → F → such that for all s [0, 1]: 0 ∈ (X , p ) ( ) |[0,1]×{s0} |[0,1]×{s0} ∈ P F and p(0, s) does not depend on s. Given such a variation (X, p) we define the complement of (X, p) to be the two-parameter vector field: 2 d t,u (t, s) [0, 1] x MY (t, s) t,s := Φ (x) ΦX (x) X ∀ ∈ ∀ ∈ | du u=s t,s where ΦX denotes the flow of X in the t-direction.

1,1 ΦX (x) Y (t, s) 1,s x ΦX (x) X(t, s) 1,0 ΦX (x)

Figure 2. An illustration of the relationship between a variation and its complement.

Remark. Since p( , s) is an integral curve of X( , s) for each ﬁxed s, we know that · · Y (t, s) = d p(t, s). This implies that p(t, s) is a classical boundary-preserving |p(t,s) ds homotopy between the paths p(t, 0) and p(t, 1) if and only if Y (1, s) p(1,s) = 0 for all s [0, 1]. | ∈ 14ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

Lemma 3.2.1. Suppose (X, p) is a variation of -paths with complement Y . Then Y C∞([0, 1]2, (M)). F ∈ F t,s Proof. Let (ΦX )∗ : X(M) X(M) be the push-forward along the flow (in the t- direction) of X. The variation→ of parameters formula yields an expression for Y (t, s) t,s in terms of ΦX : Zt t,s u,s −1 dX Y (t, s) = (Φ )∗ (Φ ) (u, s)du (5) X X ∗ ds 0 t,s 2 By Lemma 3.0.1 we know that (ΦX )∗ defines a smooth function [0, 1] (M) (M). Furthermore, integrals of smooth functions [a, b] (M) are× F smooth→ sinceF we can integrate in each coefficient separately. Therefore,→ F we conclude that Equation (5) defines a smooth function Y : [0, 1]2 (M). → F Definition 3.2.2 (Homotopy). Suppose (X , p ) and (X , p ) are -paths. Then 0 0 1 1 F (X0, p0) is said to be -homotopic to (X1, p1) if and only if there exists a variation (X, p), with complementF Y such that (X, p) = (X , p ) for i = 0, 1 and |s=i i i Y (1, s) I s [0, 1]. ∈ p(1,s)F ∀ ∈ The diffeological space of all -homotopy classes is called the fundamental groupoid of and is denoted Π (F) F 1 F Remark. Another notion of homotopy for singular foliations appears in an article of Laurent-Gengoux, Lavau and Strobl[17] and more recently in [18] using the language of Q-manifolds. At the moment we do not know the precise relationship between these two definitions but we suspect that our notion of homotopy is essentially a ‘truncation’ of theirs.

Theorem 3.2.1. Suppose (X0, p0) and (X1, p1) are -homotopic, then they have the same holonomy. F

Proof. Let (X(t, s), p(t, s)) be an -homotopy from (X0, p0) to (X1, p1) with com- F t,s plement Y (t, s). For convenience, let q := p0(1) = p1(1). Let ΦX : M M denote t,s → the ﬂow of X in the t-direction and let ΨY denote the ﬂow of Y in the s-direction. By assumption we have that:

1,0 1 1,1 1 ΦX = ΦX0 and ΦX = ΦX1

1,s 1,s 1,0 On the other hand, we know that ΦX = ΨY ΦX and we can therefore conclude that: ◦ Ψ1,1 Φ1 = Φ1 Y ◦ X0 X1 1,1 By assumption, Y (1, s) q Iq and so ΨY exp(Iq ). Let (Y, q) denote the constant -path corresponding| ∈ F to Y . By Lemma∈ 3.1.3, F F Hol(X , p ) = Hol((Y, q) (X , p )) = Hol(X , p ) 0 0 0 0 1 1 INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 15

4. Comparison to Crainic-Fernandes Integration Our construction of the fundamental groupoid and the holonomy groupoid of a singular foliation bears a significant similarity to the construction of the integration of a Lie algebroid by Crainic and Fernandes [6]. In fact, the inspiration behind our construction of the fundamental groupoid came from the observation that the definition of algebroid homotopy used in the integration of Lie algebroids can be defined purely in terms of time-dependent sections, rather than paths. In this section we will make the analogy concrete by showing that the two constructions are equivalent when restricted to a single leaf. 4.1. A-homotopy. Let us recall the construction of the integration of a Lie algebroid via A-paths. Definition 4.1.1. Let π : A M be a Lie algebroid with anchor map ρ: A TM. An A-path is defined to be→ a smooth function a: [0, 1] A such that ρ→ a = d(π a)/ dt. → ◦ ◦ The set of all A-paths is denoted (A). It has a homotopy-like equivalence relation called A-homotopy2. P Definition 4.1.2. Let a and a be A-paths. A smooth map a(t, s): [0, 1] [0, 1] 0 1 × → A is called an A-homotopy from a0 to a1 if the following properties hold: ρ a = d(π a)/ dt • ◦ ◦ a(t, i) = ai(t) for i = 0, 1 • Any two parameter family of sections α(t, s) of A which extends a(t, s) has • the property that the unique solution, β(t, s), of the initial value problem: dα(t, s) dβ(t, s) = [α(t, s), β(t, s)] β(0, s) = 0 (6) ds − dt satisfies β(1, s) = 0. |(π◦a)(1,s) Lemma 4.1.1. Suppose α(t, s) is a two-parameter family of vector fields on M and t,s let Φα : M M denote the flow of α in the t-direction. Then the two parameter family of vector→ fields dΦt,s β(t, s) := α ds satisfies Equation (6). Remark. The proof of proceeding lemma is a direct calculation, which we omit. It is the main technical observation which shows that TM-homotopy coincides with classical endpoint preserving homotopies.

4.2. Leafwise Integration and Π1( ). To see the relationship between Lie algebroids and foliations, we return to theF topic of the fiberspace: π : A( ) M F → from Definition 2.3.1. In this section we will mildly abbreviate our notation for readability: A := A( ) and A := π−1(L) F L 2The definition of A-homotopy is slightly non-standard but one can see it is equivalent by examining the proof of Proposition 1.3 of [6]. 16ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

For an arbitrary leaf L of M, recall that AL is a Lie algebroid (Lemma 2.3.1). Even better, we learned from Debord [9] that AL is always an integrable Lie algebroid. Let Π ( ) denote the subgroupoid of Π ( ) of elements whose source is inside 1 F L 1 F of a given leaf L. We would like to compare Π1( )L with (AL). Indeed, it turns out that they are the same. F G

Theorem 4.2.1. The diﬀeological groupoids Π1( )L and (AL) are naturally diffeomorphic. In particular, Π ( ) is smooth. F G 1 F L Proof. Let G (A) := (A ) C∞([0, 1],A) P P L ⊆ L⊆M where the union is taken over all leaves L of . We equip (A) with the subset diffeology relative to C∞([0, 1],A). Two elementsF of (A) areP said to be A-homotopic P if they are AL-homotopic for some leaf L. Consider the smooth function Q: ( ) (A) P F → P

Q(X, γ)(t) := evγ(t)(X(t)) The theorem follows from two claims: (1) Q is a subduction. (2) Suppose (X , γ ) and (X , γ ) are -paths. Then (X , γ ) is -homotopic 0 0 1 1 F 0 0 F to (X1, γ1) if and only if Q(X0, γ0) is A-homotopic to Q(X1, γ1). Claim 1: We need to show that plots on (A) can be lifted to plots on ( ). Suppose V is a subset of some euclidean spaceP and 0 V . Let φ: V (AP) beF a plot. To prove the claim, we must show that there exists∈ an open neighborhood→ P W of 0 V together with a lift φe: W ( ) such that Q φe = φ V . The∈ smoothness of φ is equivalent→ to P theF smoothness of:◦ | a: V [0, 1] A a(v, t) := φ(v)(t) × → Let γ := π a: V [0, 1] M ◦ × → Since [0, 1] is compact, it follows that γ(W [0, 1]) is contained inside of the interior of a compact subset of M. Therefore, we× assume that (M) is finitely generated without loss of generality. F n Let Xi i=1 (M) be a set of generators. Using the coefficient diffeology on (M){ and} the compactness⊆ F of the interval, we conclude that there exists an open neighborhoodF W of 0 V together with smooth functions ci C∞ such ∈ { } ∈ W ×[0,1] that for all (w, t) W [0, 1] we have that: ∈ × X i a(w, t) = evπ◦a(w,t)(c (w, t)Xi)

Let φe(w) := (Xw, γw) where: X X (t) := ci(w, t)X γ (t) := π a(w, t) w i w ◦ Claim 2: ( ) Suppose (X , γ ) is -homotopic to (X , γ ). Let L be the ⇒ 0 0 F 1 1 unique leaf containing the image of γ1 and γ2. By assumption, there exists a INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 17 variation (X(t, s), γ(t, s)) with complement Y (t, s) such that Y (1, s) Iγ(1,s) . Lemma 4.1.1 implies that: ∈ F dX(t, s) dY (t, s) = [X(t, s),Y (t, s)] and Y (0, s) = 0 ds − dt

Let α(t, s) and β(t, s) denote the two parameter sections of AL represented by X and Y respectively. Since the algebroid bracket on AL comes from the bracket on , it follows that α(t, s) and β(t, s) satisfy Equation 6. Furthermore, since Y (1F, s) I it follows that β(1, s) = 0. Therefore π(X , γ ) and π(X , γ ) are ∈ γ(1,s)F 0 0 1 1 AL-homotopic. ( ) We will show the other direction in two parts. First, we will show that the fibers⇐ of π are contained in the -homotopy equivalence classes. Then we will show that an A -homotopy can be liftedF to a -homotopy between fibers. L F Suppose π(X0, γ0) = π(X1, γ1). In this case, γ0 = γ1. Let X(t, s) := sX1(t) + (1 s)X0(t). We claim that (X(t, s), γ) is a -homotopy. Let Y (t, s) denote the complement− of this variation. Recall EquationF 5 for Y : Z1 1,s u,s −1 Y (1, s) = (Φ )∗ (Φ ) (X (u) X (u))du X X ∗ 1 − 0 0 t,s Since (ΦX )∗ preserves the foliation (Lemma 3.0.1), and since (X1(u) X0(u)) γ(u) I . We can conclude that Y (1, s) I . Which proves that− (X , γ| ) and∈ γ(u)F ∈ γ(1)F 0 (X1, γ) are homotopic. Now we show that A homotopies can be lifted to -homotopies between fibers of L F π. Suppose a(t, s) is an AL homotopy between some AL-paths a0 and a1. Let γ(t, s) denote the base of this map. Using the argument from Claim 1, we can lift the map 2 2 a: [0, 1] AL to a map α: [0, 1] (M) (we can always lift the map locally and then patch→ together using a partition→ of F unity). Let β(t, s) be the complement of the associated variation (α, γ). Since we know a(t, s) is an A-homotopy, it follows that [β(1, s)] = 0 for all s. This implies that β(1, s) I . Therefore (X, γ) |Aγ (1,s) ∈ γ(1,s)F is a -homotopy. Furthermore π(X(t, 0), γ(t, 0)) = a and π(X(t, 1), γ(t, 1)) = a . F 0 1 This concludes Claim 2.

This shows that our construction of Π1( ) is (leaf-wise) equivalent to perform- ing a classical integration of A. An advantageF of our construction of Π ( ) over 1 F constructing the integration separately over each leaf is the fact that Π1( ) inherits a diﬀeological structure which glues the leaf-wise integrations together. F

5. Comparison to the Androulidakis-Skandalis construction A version of the holonomy groupoid originally appeared in a paper of Androuli- dakis and Skandalis [1]. Their construction does not bear much resemblance to ours. The main result of this section (Theorem 5.2.1) is the fact that the construction in Section 3.1 is equivalent (as a diffeological groupoid) to the Androulidakis-Skandalis construction. In order to prove this result, we will need to review the construction of Androulidakis and Skandalis. It is in this section that we will most use the language of diffeology. In addition to providing us with a ‘smooth sense’ in which our construction is the same as that of Androulidakis-Skandalis, the diffeology itself is ultimately crucial in our proof that the map we construct is a bijection. 18ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

5.1. Bisubmersions. Definition 5.1.1. A bisubmersion over (M, ) is a manifold P equipped with two submersions: F V t P s” U where U and V are open subsets of M. These maps must also be compatible with the foliation in the following sense: t−1( ) = Γ(ker dt) + Γ(ker ds) = s−1( ) (7) F F A bisection of P is defined to be a smooth map σ : U P such that s σ = 1U and t σ : U V is a diffeomorphism. → ◦ ◦ → A bisection σ is said to be through a point p P if p is in the image of σ. In such a case, we say that p P carries the diffeomorphism∈ t σ. In practice, we are mainly∈ interested in local bisections.◦ That is, bisections for which the domain U is small enough. Indeed, we can see that if we permit shrinking the open set U, one can always find a bisection through a given point p P . ∈ Example 5.1.1 (Path Holonomy Bisubmersion). Let x M and suppose X q ∈ { i}i=1 is a set of compactly supported elements of which represent a basis for Ax := q F /Ix . Then let t: R M M be the map defined as below: F F × → 1 t(c1, . . . , cq, y) ΦPq c X (y) 7→ i=1 i i and let s: Rq M M be the projection to the second component. Then there × → exists an open neighborhood P Rq M which contains (0,..., 0, x) such that: ⊆ × M V t P s U M ⊃ ⊆ is a bisubmersion. It is possible to compose bisections via a fiber product operation: Q P ×U2 Q P

U3 U2 U1 Let PHB be the disjoint union of all compositions of path holonomy bisubmersions. The manifold PHB also has a monoidal structure since given p P PHB and q Q PHB such that s(p) = t(q) we can define p q := (p, q) ∈P ⊆Q PHB. ∈ ⊆ · ∈ ×M ⊆ Definition 5.1.2. The holonomy groupoid of Androulidakis and Skandalis, denoted ( )AS is defined to be PHB/ where p q if and only if they carry the same diffeomorphism.H F The source and∼ target maps∼ of the groupoid are inherited from those of PHB. For the sake of clarity, we will denote -paths up to holonomy by ( )GV to distinguish it from ( )AS. F H F In section 3.1, weH F defined ( )GV by associating holonomy transformations to -paths. Before our work,H AndroulidakisF and Zambon [3] showed that it was possibleF to define holonomy transformations for ( )AS as well. We now briefly H F summarize their results. Suppose at each x M we have chosen a slice Sx and formed the corresponding groupoid of holonomy∈ transformations HT. Suppose INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 19 p PHB is a point in a bisubmersion P . Let f : U V be a locally defined ∈ → AZ diffeomorphism carried by p then holonomy of p, written Hol (p) is defined to be the class of f in HT. Theorem 5.1.1 (Androulidakis-Zambon). The map HolAZ : PHB HT is well → defined and the fibers of HolAS are precisely the fibers of the projection PHB ( )AS. → H F A consequence of this result is another characterization of the equivalence relation on PHB which leads to ( )AS. H F 5.2. Relating ( )AS and ( )GV . To understand the relationship between these two constructions,H F we beginH F by defining a map: Ψ:e PHB ( ). → P F q Suppose (c1, . . . cq, x) is a point in a path holonomy bisubmersion P R M with associated basis X q . Let: ⊆ × { i}i=1 q ! X t Ψ(e c1, . . . cq, x) := ciXi, ΦPq (x) . i=1 ciXi i=1 If p = (p , p ) PHB is an element of a bisubmersion P which comes as a product 1 2 ∈ P = P1 M P2. Then we define Ψe inductively as Ψ(e p1) Ψ(e p2) using the product of -paths× (Definition 3.0.3). F Theorem 5.2.1. The smooth map Ψ:e PHB ( ) descends to a diffeological groupoid isomorphism: → P F Ψ: ( )AS ( )GV H F → H F We will split the proof into a series of lemmas.

Lemma 5.2.1. The map Ψ:e PHB ( ) is compatible with holonomy transformations, i.e. it makes the following→ diagram P F commute:

PHB Ψe ( ) P F HolAZ Hol HT HT

Proof. First, observe that Ψe is compatible with the monoidal structures on PHB and ( ) by definition. Hence, we only need to show this diagram commutes for a setP ofF elements which generate PHB. Let p = (c1, . . . , cq, x) be a point inside of a path holonomy bisubmersion P q q ⊆ R M associated to some set of local generators Xi . There is a canonical × { }i=1 bisection σ(y) := (c1, . . . , cq, y) which contains p. By the definition of t: P M 1 → the diffeomorphism defined by σ(y) around x is precisely the flow f := ΦPq i=1 ciXi centered around x. Hence, the holonomy of p is the class of f in HT. Pq However, Ψ(e p) is defined to be the -path ( i=1 ciXi, γ(t)), where γ is the Pq F integral curve of i=1 ciXi starting at x. Therefore the holonomy of Ψ(e p) is also represented by f. From this lemma, we can conclude the following: 20ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

Corollary 5.2.1. The map Ψe descends to an injective homomorphism of diffeological groupoids Ψ: ( )AS ( )GV . H F → H F Proof. If p and p PHB represent the same element of ( )AS, then they must 1 2 ∈ H F have the same holonomy[3]. By the previous lemma, Ψ(e p1) must have the same holonomy of Ψ(e p2). Since Ψe respects the holonomy equivalence relation, it defines a map Ψ: ( )AS ( )GV on equivalence classes. Ψ is a homomorphism since H F → H F Ψe is compatible with the monoidal structures on PHB and ( ) that induce the product on ( )AS and ( )GV , respectively. Lastly, Ψ isP injectiveF since it fits into a commutingH F diagram:H F

( )AS Ψ ( )GV H F H F

HT HT Now that we know that Ψ is injective. Our next goal is to show that Ψ is a local subduction [16]. More concretely, plots centered on points in the image of Ψ locally factor through ( )AS. For that, we will need a lemma about lifting vector ﬁelds to bisubmersions.H F t s Lemma 5.2.2. Suppose U2 P U1 is a bisubmersion. Given a smooth n function X : V [0, 1] (U2) for V R such that X(0, t) = 0, there exists a smooth lift × → F ⊆ Xe : V [0, 1] Γ(ker ds) × → such that t∗Xe = X and Xe(0, t) = 0.

Proof. From the definition of the diffeology on (U2), we know that X can be Pq F written in the form X(a1, . . . , an, t) = i=1 ci(a1, . . . an, t)Yi for some constant elements Yi (U2) and smooth functions ci : V [0, 1] R. We can assume that ∈ F × → ci(0, t) = 0 for all i since otherwise we can just replace each ci with ci ci(0, t) if necessary. − From the bisubmersion condition we know that t−1( ) = Γ(ker t) + Γ(ker s). −1 −1 ∗ F Since t ( ) is defined to be (dt) (t (U2)), we know that dt: Γ(ker s) ∗ F ∞ F → t (U2) is a surjective morphism of C -modules. Therefore, we can choose Yei F P ∈ Γ(ker s) such that t∗Yei = Yi. Pn Let Xe(a1, . . . , an, t) := i=1 ci(a1, . . . an, t)Yei. From the definition of Yei, we obtain that Xe satisfies the claim of the lemma. The main application of this lemma is the following proposition which says, roughly, that plots centered on identity elements of ( )GV locally factor through Ψ. H F Proposition 5.2.1. Let x M. Suppose φ: V ( )GV is a plot such that ∈ GV → H F 0 V and φ(0) = 1x ( ) . Then there exists an open neighborhood of the ∈ 0 ∈ H F 0 AS origin, V V , and a plot ψ : V ( ) such that Ψ ψ = φ 0 . ⊆ → H F ◦ |V Proof. Suppose φ: V ( )GV is a plot as in the statement of the lemma. Let W V be a smaller neighborhood→ H F of the origin such that we can lift φ to a plot ⊆ |W INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 21

φe: W ( ). Let us write the components of φe in the following manner: → P F φe(a1, . . . , an)(t) = (X(a1, . . . an, t), γ(a1, . . . , an, t)) t s Now suppose U2 P U1 is a path holonomy bisubmersion such that the image of γ is contained in U2. Such a path holonomy bisubmersion exists since we can make the image of γ as small as we like by shrinking W . Furthermore, we can assume that the support of X(a1, . . . , an, t) is uniformly contained in U2. The previous lemma implies that there exists Xe : W [0, 1] ker ds X(P ) such × → ≤ that t∗Xe = X U2 and such that Xe(0, t) = 0. By possibly shrinking W , further, we | 1 can ensure that the flow Φ is well defined in some neighborhood of 0 U1 P . Xe { } × ⊆ Let ψe: W P PHB be defined by: → ⊆ 1 ψe(a1, . . . , an) = Φ (0,..., 0, γ(a1, . . . an, 0)) Xe(a1,...,an,·) Such a function defines a plot ψ : W ( )AS. We claim that Ψ ψ = φ . To → H F ◦ |W show this, we will show that Hol ψe = Hol φe. For each (a , . . . a ), let σ(a , . . . , a ): U P be the bisection defined by 1 n 1 n 1 → 1 id σ(a1, . . . , an) := Φ σ Xe(a1,...,an,·) Where σid is the canonical bisection of P . From the definition of Xe, it follows that the diffeomorphism carried by σ(a1, . . . , an) is precisely the flow of the time dependent vector field X . Furthermore, σ(a , . . . , a ) is a bisec- |{a1}×...×{an}×[0,1] 1 n tion through ψe(a1, . . . , an). This shows that the holonomy of ψe(a1, . . . , an) must be equal to the holonomy of φe(a1, . . . an) which completes the proof. We can use the special case of a plot centered on an identity, to show that the preceding proposition holds for any plot which is centered around an element in the image of Ψ. Corollary 5.2.2. Let x M. Suppose φ: V ( )GV is a plot such that 0 V and φ(0) Ψ( ( )AS) ∈ ( )GV . Then there→ existsH F an open neighborhood of∈ the 0∈ H F ⊆ H F 0 AS origin, V V , and a plot ψ : V ( ) such that Ψ ψ = φ 0 . ⊆ → H F ◦ |V Proof. Suppose φ: V ( )GV is a plot such that φ(0) = h Ψ( ( )AS) ( )GV . Let g (→)AS H Fbe an element such that Ψ(g) = h.∈ Furthermore,H F let⊆ Hσ beF a (locally defined)∈ H F bisection of ( )AS through g. Shrink V if necessary so that σ(t φ) is well defined. H F ◦ Then φ := Ψ σ(s φ)−1 φ is a well defined plot centered on a unit element of ◦ ◦ · ( )GV . Let ψ := V ( )AS be a plot such that Ψ ψ = φ. Such a plot exists H F → H F ◦ by Proposition 5.2.1. Let ψ := σ(s φ) ψ. Then: ◦ · Ψ ψ = Ψ (σ(s φ) ψ) ◦ ◦ ◦ · = (Ψ σ(s φ)) (Ψ ψ) ◦ ◦ · ◦ = (Ψ σ(s φ)) (Ψ σ(s φ)−1) φ ◦ ◦ · ◦ ◦ · = φ which concludes the proof. We can now complete the proof of Theorem 5.2.1. 22ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

Proof. (of Theorem 5.2.1) Let us now equip ( )GV and ( )AS with their re- spective D-topologies (II.8 in [16]). That is,H aF subset of H( F)GV or ( )AS is open if and only if its inverse image under every plot is open.H F Since ΨH isF smooth, it follows that it is continuous with respect to the D-topology. Furthermore, Corol- lary 5.2.2 showed us that plots on ( )GV which intersect elements in the image of Ψ, locally factor through Ψ. ByH II.20F from [16], this implies that Ψ is open. These topologies make both ( )GV and ( )AS source connected topological groupoids. Since Ψ is open, weH haveF that ΨH isF surjective [19]. This implies that all plots on ( )GV locally factor through ( )AS. Since Ψ is also injective, we H F H F conclude that it is a diﬀeomorphism of diﬀeological spaces.

6. Functoriality Our aim in this section is to understand the functorial properties of the construction of Π1( ) and ( ). The main technical difficulty in this section is that it is not generallyF possibleH F to push forward vector fields. A consequence of this is that it is usually difficult to define a function at the level of -paths. To circumvent this problem, we will require some technical development regardingF the relationship between a foliation and its fiberspace A( ). F F 6.1. Comorphisms of sheaves of modules. We need to develop a notion of morphism of foliated manifolds. Our model example is the case of projective foliations where a morphism of foliations will corresponds to a morphism of the associated Lie algebroids. Higgins and Mackenzie [14] observed that there is a duality between morphisms of vector bundles and comorphisms of modules. This observation will motivate our definition of a morphism of foliated manifolds.

∞ Definition 6.1.1. Suppose is a sheaf of CN -modules on N and f : M N is a F ! ∞→ smooth function. The pre-pullback f of along f is a presheaf of CM -modules such that for all open U M: F F ⊆ ! ∞ f (U) := C (U) C∞(N) (N) F M ⊗ N F Given u C∞(N), the action on C∞(U) is by the usual pullback operation: ∈ N M v v (u f) v C∞(U) 7→ · ◦ |U ∈ M The pullback f ! of along f is defined to be the sheafification of f ! . F F F Remark. The pre-pullback of is a separated presheaf. In particular, the canonical F sheafification map f ! f ! is injective and locally an isomorphism. Therefore, for any point p M Fthere → existsF an open neighborhood U such that: ∈ ! ∞ f (U) = C (U) C∞ (N) F M ⊗ N F If is locally finitely generated, then it follows that f ! is locally finitely generated.F Generally, it is easier to work with elements in theF image of the inclusion f ! , f ! since they have an explicit form in terms of the tensor product. FThe→ pullbackF we have just defined is a pullback in the world of modules, not singular foliations. In general, the pullback of a singular foliation (as a sheaf of modules) is only a module and not a singular foliation. It is a distinct operation from the inverse image f −1( ) and the pullback f ∗ operations seen in Androulidakis and Skandalis [1]. F F INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 23

Example 6.1.1. Suppose E N is a vector bundle over N and f : M N is → ∞ ! → smooth. Let ΓE denote the associated sheaf of CN -modules and f E := M N E denote the pullback vector bundle. Given an element: ×

X i ∞ u si C (U) C∞ ΓE(N) ⊗ ∈ M ⊗ N one can deﬁne an element σ of Γf !E(U): X σ(p) = p, ui(p)(s f)(p) i ◦

! This correspondence gives rise to a morphism of presheaves f ΓE Γf !E. Since it ! → is locally an isomorphism, the associated morphism of sheaves f ΓE Γf !E is an isomorphism. →

The pullback is a functorial operation. That is, given a morphism G: 1 2 ∞ ∞ F → F of CN -modules, then there is a canonical morphism of CM -modules: f !G: f ! f ! F1 → F2 which is uniquely determined by its behavior on the image of the pre-pullback: X X (f !G) ui s = ui G(s ) ⊗ i ⊗ i

Deﬁnition 6.1.2. Suppose M and N are sheaves of modules on manifolds M and N.A comorphism F F (F, f): FM → FN ! consists of a pair (F, f) where f : M N is a smooth map and F : M f N is ∞ → F → F a morphism of sheaves of CM -modules. Comorphisms are composed according to the following rule:

(F, f) (G, g) := (F f !G, f g) ◦ ◦ ◦ Example 6.1.2. Let E M and W N be vector bundles over M and N respectively. Suppose Ψ: E →W is a vector→ bundle morphism which covers f : M N. Let → →

Ψ∗ :Γ Γ ! E → f W ! be the associated pushforward map. If we identify Γf !W with f ΓW using the canonical isomorphism found in Example 6.1.1 then (Ψ∗, f):Γ Γ is a comorphism. E → W

In fact, it turns out that all comorphisms ΓE ΓW arises in this way. We will see how to construct the bundle map associated→ to a comorphism in the next section, where we will do it in a more general setting. The next example is a special case of the one just discussed.

! Example 6.1.3. Suppose f : M N is a smooth map. Then f XN is canonically isomorphic to the sheaf of sections→ of f !TN. The diﬀerential of f deﬁnes a morphism of vector bundles TM f !TN and hence a comorphism: →

(df∗, f): X X M → N 24ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

6.2. Comorphisms and the fiberspace. Higgins and Mackenzie [14] showed that there is a functor from projective locally finitely generated modules with comorphisms as maps to the category of vector bundles. In this section, we will show that one can extend this functor to arbitrary modules by allowing for diffeological vector bundles. The motivation for doing this is that working at the level of A( ) will permit us to circumvent the problem of pushing forward vector fields. AlthoughF one cannot push forward -paths, it makes sense to push forward a path in the fiberspace A( ). F F ∞ Let be a locally finitely generated CM -module. Recall the definition of the fiberspaceF of (Definition 1.0.4): F G (M) A( ) := A( )p A( )p := F F F F Ip (M) p∈M F Recall from Definition 2.3.1 that we equip the fiberspace with a quotient diffeology via the evaluation map: ev: M (M) A( )(p, X) ev (X) × F → F 7→ p The addition, multiplication and zero section operations on M (M) descend to smooth functions on A( ) which make it into a diffeological vector× F bundle over M. We begin with a lemmaF that tells us that the fiberspace construction plays well with pullbacks of modules.

∞ Lemma 6.2.1. Suppose is a CN -module and f : M N is a smooth map. Then there is a morphism:F → q : A(f ! ) A( ) f F → F of diffeological vector bundles covering f : M N. → Proof. Observe that for all p M, there is a C∞-bilinear function: ∈ N C∞(M) (N) A( ) (u, X) u(p) ev (X) M × FN → FN f(p) 7→ · f(p) This induces a smooth function: Q : M (f ! )(M) A( ) f × F → F which is uniquely determined by its behavior on elements in the pre-pullback: X X Q p, ui X := ui(p) ev (X ) f ⊗ i · p i ! Since Qf (p, Y ) = 0 for all Y Ip(f N ), we get that Qf descends to a diffeological vector bundle map: ∈ F q : A(f ! ) A( ) f F → F Example 6.2.1. If is the sheaf of sections of a vector bundle E N. Then we have a natural identification:F → A(f ! ) = f !E := M E F ×N Under this identification, q : f !E E is projection to the second component. f → INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 25

Deﬁnition 6.2.1. Suppose (F, f): is a comorphism of modules. Let: FM → FN F : A( ) A(f ! ) ev (X) ev (F (X)) FM → FN p 7→ p Then we deﬁne: A(F, f) := q F : A( ) A(f ! ) A( ) f ◦ FM → FN → FN where qf is as in Lemma 6.2.1.

Example 6.2.2. Suppose M = ΓE and N = ΓW are sheaves of sections of a vector bundle. Then A(ΓF ) M is canonicallyF isomorphic to E M and E → → A(ΓW ) N is canonically isomorphic to W N. Using these identifications, we get vector→ bundle morphism → A(F, f): E W → Furthermore, if we apply the reverse construction seen in Example 6.1.1, we will recover the comorphism (F, f). It is straightforward to check that the construction of the fiberspace A( ) and the diffeological vector bundle map A(F, f) yields a functor A from the categoryF of manifolds equipped with sheaves of modules to the category of diffeological vector bundles over manifolds.

6.3. Functoriality of Π ( ). We now specialize our discussion to foliations rather 1 F than general modules. In this section, ιM : M , XM is a foliation on M and ι : , X is a foliation on N. F → N FN → N Deﬁnition 6.3.1. A comorphism (F, f): M N is a morphism of foliated manifolds if it satisﬁes the following properties:F → F F is compatible with the inclusions. That is, the following diagram com- • mutes: F f ∗ FM FN ∗ ιM f ι (8) df∗ ∗ XM f XN

F is compatible with the brackets. That is, given X1,X2 M (M) such • that: ∈ F X X F (X ) = ui Y F (X ) = vj Y 1 ⊗ i 2 ⊗ j i=1 j=1 we have that: X X X F ([X ,X ]) = uivj [Y ,Y ] + X (vj) Y + X (ui) Y (9) 1 2 ⊗ i j 1 ⊗ j 2 ⊗ i i,j j i Example 6.3.1. Suppose and are projective foliations. Then a comor- FM FN phism (F, f): M N is a morphism of foliated manifolds if and only if the associated vectorF bundle→ F homomorphism: A(F, f): A( ) A( ) FM → FN is a morphism of Lie algebroids. In this setting, Diagram (8) is equivalent to compatibility with the anchor maps and Equation (9) is equivalent to the usual compatibility with the Lie brackets (e.g. compare to Equation (2.18) in [7].) 26ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

Example 6.3.2. Suppose 1 and 2 are two foliations on M. Then a morphism of foliated manifolds (F, Id ):F F exists if and only if is a submodule of . M F1 → F2 F1 F2 If such a morphism exists, it is unique since Diagram 8 implies that F : 1 2 is the submodule inclusion map. F → F Example 6.3.3. Suppose f is a submersion and = f −1( ). That is, is FM FN FM the foliation generated by projectable vector ﬁelds X XM such that df(X) N . By choosing local sections of f, we can conclude that:∈ ∈ F f !ι : f ! f !X N FN → N is injective. Therefore, if we are given a morphism of foliated manifolds (F, f): = f −1( ) FM FN → FN ! then Diagram (8) implies that F : f is just df∗ F . FM → FN | M Example 6.3.4. Suppose (F, f): = is a morphism of foliated manifolds FM FN and f : M N is a submersion. Then Diagram (8) implies that M is a submodule of f −1( →). Therefore, (F, f) factors uniquely into a compositionF FN f −1( ) FN (F1,IdM ) (F2,f)

(F,f) FM FN That is (F, f) is a composition of Example 6.3.3 with Example 6.3.2. Consequently, when f is a submersion, a morphism of foliated manifolds M N exists if and only if , f −1( ). Furthermore, if such a morphismF exists→ then F it is unique. FM → FN The bracket condition in the deﬁnition of morphism of foliated manifolds is not implied by compatiblity with the inclusion and a morphism of foliated manifolds (F, f) is not always uniquely determined by f. The following two examples demon- strate these facts.

2 2 Example 6.3.5. Let M = R with M = XM and take N = R with the foliation: F = x∂ , y∂ , x∂ , y∂ FN h x y y xi ! Now take f : M N to be the zero map f(x, y) = 0 and let F : M f N be such that: → F → F F (∂ ) = 1 x∂ F (∂ ) = 1 y∂ x ⊗ y y ⊗ x Then this pair (F, f) deﬁnes a comorphism of modules and satisﬁes compatibility with the inclusion. However, it does not satisfy compatibility with the bracket.

Example 6.3.6. Let M = R with the full foliation and N = R2 with the foliation given by the free module, N := R C∞ F h i N with the inclusion ι : X R y∂ x∂ N FN → N 7→ x − y Now let f : R R2 be the constant map which sends every point to zero. Notice → ! ! ! that the pullback of the inclusion f ιN : f N f XN is not injective. For example, 1 R is a non-trivial element of f ! but:F → ⊗ FN (f !ι )(1 R) = 1 i (R) = 1 (y∂ x∂ ) = (y f) ∂ (x f) ∂ = 0 N ⊗ ⊗ N ⊗ x − y ◦ ⊗ x − ◦ ⊗ y INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 27

Indeed, given an arbitrary constant C R, we can deﬁne a module homomorphism: ∈ F C : X f ∗ F C (∂ ) = C R R → FN t ⊗ For each choice of C,(F C , f) is a morphism of foliated manifolds. Therefore, we have exhibited a family of distinct morphisms with the same base map.

Suppose (F, f): M N is a morphism of foliated manifolds. Then Dia- gram (8) implies thatF f preserves→ F the characteristic distribution associated to each foliation. In particular, for each leaf L, M, there is a unique leaf f∗L, N containing f(L). From now on, we will use→ this push-forward notation to refer→ to the induced function at the level of leaf-spaces.

Theorem 6.3.1. Suppose we are given a morphism (F, f): M N of foliated manifolds. Then there exists a unique morphism of diﬀeologicalF groupoids→ F Π (F, f):Π ( ) Π ( ) 1 1 FM → 1 FN such that the restriction to each leaf Π (F, f) :Π ( ) Π ( ) 1 L 1 FM L → 1 FN f∗L integrates the Lie algebroid morphism A(F, f) : A( ) A( ) L FM L → FN f∗L Proof. The uniqueness follows from the fact that the source ﬁbers of Π1( M ) are simply connected and Lie’s theorems for Lie algebroids. We only need toF concern ourselves with constructing Π1(F, f). Recall the set of A( )-paths discussed in the proof of Theorem 4.2.1. FM G (A( )) := (A( ) ) C∞([0, 1],A( )) P FM P FM L ⊆ FM L

There is a similar set of A( N )-paths, (A( N )). By post-composition of A(F, f), we obtain a smooth function:F P F Φ: (A( )) (A( )) P FM → P FN The deﬁnition of morphism of foliated manifolds implies that A(F, f) is a Lie algebroid morphism when restricted to each leaf. Therefore A(F, f) respects the algebroid homotopy equivalence relation. By Claim 1 and Claim 2 in the proof of Theorem 4.2.1, Φ descends to a smooth function: Φ:Π ( ) Π ( ) 1 FM → 1 FN On each leaf, Φ has been constructed in exactly the same manner as the classical proof of Lie’s second theorem for groupoids. Therefore, we take Π1(F, f) = Φ.

−1 Example 6.3.7. Suppose f is a surjective submersion and M = f ( N ). We call an -path (X, γ) projectable if: F F FM X i X(t) = c (t)Xi

i where each c (t): [0, 1] R is a smooth function and Xi M is a projectable → ∈ F vector ﬁeld along f. Let f ( M ) ( M ) denote the set of all M -paths which are projectable along f. P F ⊆ P F F 28ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN

Since is generated by projectable vector fields, it follows that any -path FM FM Y is M -homotopic to a projectable M -path XY . Consequently, the map Φ from TheoremF 6.3.1 arises from a functionF at the level of projectable -paths: FM ( ) ( )(X, γ) (df(X), γ) Pf FM → P FN 7→ The construction of the homomorphism in Theorem 6.3.1 yields a functor from the category of foliated manifolds to the category of diffeological groupoids. The compatibility of Π1 with composition of morphisms follows from the usual functoriality of the construction of the universal integration of a Lie algebroid. We warn the reader that Π1(F, f) does not always descend to a homomorphism at the level of the associated holonomy groupoids. By this we mean that there does not always exist a morphism of diffeological groupoids ( M ) ( N ) which completes the following diagram: H F → H F

Π (F,f) Π ( ) 1 Π ( ) 1 FM 1 FN (10)

( ) ( ) H FM H FN This is already known to occur in the case of regular foliations. For example: Example 6.3.8. Suppose N S1 is a non-trivial line bundle equipped with a flat connection (i.e. the Möbiusband).→ We think of N as a foliated manifolds by taking 1 N to be the vector fields tangent to the connection. Let M = S be equipped Fwith the full foliation. Both foliations are regular and the inclusion f : M, N of the zero section preserves the distributions so it induces a morphism of foliated→ manifolds (F, f): . FM → FN The isotropy groups of Π1( M ) are Z and the induced map Π1( M ) Π1( N ) is the inclusion of the full subgroupoidF at the zero section. F → F The isotropy groups of the holonomy groupoid ( ) are trivial whereas the H FM isotropy groups of ( N ) are Z2 on the zero section. Since there is no way to complete the followingH F diagram of group homomorphisms

1 Z Z Z

1 Z2 there cannot exist a groupoid homomorphism ( M ) ( N ) which completes Diagram (10). H F → H F

2 Example 6.3.9. Suppose M = R . Let R XM (M) be the usual counterclockwise 3 ∈ rotation vector ﬁeld. Now let N := R and let X XN (N) be the vector ﬁeld: ∈ ∂ X := R z ⊕ − ∂z We take M and N to be the foliations generated by R and X, respectively. Now let f be theF inclusion:F f : M N (x, y) (x, y, 0) → 7→ ∗ Let F : M f N be such that F (R) = 1 X. Then (F, f) is a morphism of foliatedF manifolds.→ F The smooth function Π ( ⊗) , Π ( ) is the inclusion of the 1 FM → 1 FN INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 29 full subgroupoid over the z = 0 plane. In particular, it is an isomorphism at the level of isotropy groups. If we compute the isotropy group of the holonomy of M at the origin, we can see that it is isomorphic to S1. On the other hand, the isotropyF group of ( ) at the H FN origin is R. Since there does not exist a non-trivial homomorphism of Lie groups 1 S R, we conclude that (F, f) does not induce a morphism ( M ) ( N ). → H F → H F We conclude with a theorem that says that we can indeed complete Diagram (10) when f is a submersion.

Theorem 6.3.2. Suppose (F, f): M N is a morphism of foliated manifolds and f : M N is a submersion.F There→ exists F a diﬀeological groupoid homomorphism → Hol(F, f): ( ) ( ) H FM → H FN which makes the following diagram commute:

Π (F,f) Π ( ) 1 Π ( ) 1 FM 1 FN

Hol(F,f) ( ) ( ) H FM H FN Proof. Let q :Π ( ) ( ) q :Π ( ) ( ) M 1 FM → H FM N 1 FN → H FM be the natural quotient maps. We need to show that for all g Π ( ) such that ∈ 1 FM qM (g) is an identity, we have that qN Π1(F, f)(g) is an identity element. Recall the pullback foliation f −1( ◦) which is generated by C∞-linear combina- FN M tions of projectable vector fields X such that df(X) N . Since f is a submersion, ∈ F −1 Example 6.3.4 told us that (F, f) factors uniquely through f ( N ). That is, there exist comorphisms: F (F , f): f −1( ) (F , Id ): f −1( ) 1 FN → FN 2 M FM → FN such that (F, f) = (F , f) (F , Id ) 1 ◦ 2 M Therefore, we can split the proof into two cases. One case is where f = IdM and one case where = f −1( ). FM FN Suppose f = IdM . Then the condition of being compatible with the inclusion implies that F : is just the inclusion of a submodule of vector fields. In FM → FN particular, Π1(F, f) sends the M homotopy class of (X, γ) to the N homotopy class of (X, γ). If g Π ( )F and q (g) is an identity element, thereF must exist ∈ 1 FM M an M -path (X, γ) which represents g and a transversal τ to the leaves of M such thatF F Φ1 = Id . X |τ τ Since a submanifold transverse to must contain a submanifold transverse to FM N , it follows that the holonomy of (X, γ) relative to N is also trivial. F For the other case we take = f −1( ). GivenF a transversal τ to , then FM FM FM f(τ) is a transversal to N . Both τ and f(τ) inherit foliations from M and N respectively. From the definitionF of the pullback foliation, it follows that:F F f : τ f(τ) |τ → 30ALFONSO GARMENDIA, UNIVERSITAT¨ POTSDAM AND JOEL VILLATORO, KU LEUVEN is a foliation preserving diffeomorphism. Recall the discussion of projectable -paths from Example 6.3.7. Suppose FM g Π1( M ) is represented by a projectable M -path (X, γ) and qM (g) is an identity.∈ F Given any transversal τ through γ(0) weF have that f Φ1 = Φ1 ◦ X |τ df(X)|f(τ) Since f τ : τ f(τ) is a foliation preserving diffeomorphism, it follows that the holonomy| of (→X, γ) is trivial if and only if the holonomy of (df(X), f γ) is trivial. ◦ Remark. According to Example 6.3.4, the condition in Theorem 6.3.2 that (F, f) is a morphism of foliated manifolds is equivalent to the claim that M is contained in f −1( ). This result is an improvement on theorems that appearF in preprints FN [26] and [12]. However, by decomposing the problem into the cases f = IdM and −1 M = f ( N ), one can mostly recover Theorem 6.3.2 as a corollary of these papers.F F

Example 6.3.10. Let π : P (M, M ) be a surjective submersion with connected → F −1 fibers. On P , take the foliation P := π ( M ). In a recent preprint [11], it is shown that: F F −1 ( P ) ∼= π ( M ) := P π,t ( M ) s,π P H F H−F1 × H F × The canonical projection map π ( M ) ( M ) is Hol(dπ∗, π) from Theo- rem 6.3.2. H F → H F If we are given a Hausdorff Morita equivalence N P M as defined in [11], ← → we get a pair of morphisms foliated manifolds M P N and a commuting diagram: F ← F → F Π ( ) Π ( ) Π ( ) 1 FN 1 FP 1 FM

( ) ( ) ( ) H FN H FP H FM

On the bottom row, the diffeological groupoid morphisms are weak equivalences (per [11]). On the top row, the horizontal arrows are diffeological groupoid fibra- tions which are not necessarily weak equivalences. If one wishes to obtain a weak equivalence at the top level, then one needs to put additional conditions on the submersions N P M. To see which conditions are needed, we refer the reader to Theorem 1.2← in [24].→

References [1] Iakovos Androulidakis and Georges Skandalis. The holonomy groupoid of a singular foliation. Reine Angew. Math., 626:1–37, 2009. [2] Iakovos Androulidakis and Marco Zambon. Smoothness of holonomy covers for singular foliations and essential isotropy. Math. Z., 275(3-4):921–951, 2013. [3] Iakovos Androulidakis and Marco Zambon. Holonomy transformations for singular foliations. Adv. Math., 256:348–397, 2014. [4] J. Daniel Christensen and Enxin Wu. Tangent spaces and tangent bundles for diffeological spaces. Cah. Topol. Géom.Différ.Catég., 57(1):3–50, 2016. [5] J. Daniel Christensen and Enxin Wu. Exterior bundles in diffeology, 2020. [6] Marius Crainic and Rui Loja Fernandes. Integrability of Lie brackets. Ann. of Math. (2), 157(2):575–620, 2003. INTEGRATION OF SINGULAR FOLIATIONS VIA PATHS 31

[7] Marius Crainic and Rui Loja Fernandes. Lectures on integrability of Lie brackets. In Lectures on Poisson geometry, volume 17 of Geom. Topol. Monogr., pages 1–107. Geom. Topol. Publ., Coventry, 2011. [8] Claire Debord. Holonomy groupoids of singular foliations. J. Differential Geom., 58(3):467– 500, 2001. [9] Claire Debord. Longitudinal smoothness of the holonomy groupoid. C. R. Math. Acad. Sci. Paris, 351(15-16):613–616, 2013. [10] Alfonso Garmendia and Ori Yudilevich. On the inner automorphisms of a singular foliation. Math. Z., Nov 2018. https://doi.org/10.1007/s00209-018-2212-0. [11] Alfonso Garmendia and Marco Zambon. Hausdorff morita equivalence of singular foliations. Ann. Global. Anal. Geom., 55(1):99–132, Feb 2019. [12] Alfonso Garmendia and Marco Zambon. Quotients of singular foliations and Lie 2-group actions, 2019. arXiv:1904.08890. [13] Robert Hermann. The differential geometry of foliations. II. J. Math. Mech., 11:303–315, 1962. [14] Philip J. Higgins and Kirill C. H. Mackenzie. Duality for base-changing morphisms of vector bundles, modules, Lie algebroids and poisson structures. Math. Proc. Cambridge Philos. Soc., 114(3):471–488, 1993. [15] Morris W. Hirsch. Differential topology, volume 33 of Grad. Texts Math. Springer-Verlag, New York, 1994. Corrected reprint of the 1976 original. [16] Patrick Iglesias-Zemmour. Diffeology, volume 185 of Math. Surveys Monog. American Math- ematical Society, Providence, RI, 2013. [17] Camille Laurent-Gengoux, Sylvain Lavau, and Thomas Strobl. The universal Lie ∞-algebroid of a singular foliation. arXiv:1806.00475, 2018. [18] Camille Laurent-Gengoux and Leonid Ryvkin. The holonomy of a singular leaf, 2019. arXiv:1912.05286. [19] Kirill C. H. Mackenzie. General theory of Lie groupoids and Lie algebroids, volume 213 of London Math. Soc. Lecture Note Ser. Cambridge University Press, Cambridge, 2005. [20] J.-M. Souriau. Groupes diff’erentiels. In Differential geometrical methods in mathematical physics (Proc. Conf., Aix-en-Provence/Salamanca, 1979), volume 836 of Lecture Notes in Math., pages 91–128. Springer, Berlin-New York, 1980. [21] J.-M. Souriau. Groupes diff’erentiels de physique math’ematique. In South Rhone seminar on geometry, II (Lyon, 1983), Travaux en Cours, pages 73–119. Hermann, Paris, 1984. [22] P. Stefan. Accessible sets, orbits, and foliations with singularities. Proc. London Math. Soc. (3), 29:699–713, 1974. [23] HéctorJ. Sussmann. Orbits of families of vector fields and integrability of distributions. Trans. Amer. Math. Soc., 180:171–188, 1973. [24] Joel Villatoro. Poisson manifolds and their associated stacks. Lett. Math. Phys., 108(3):897– 926, Mar 2018. [25] Roy Wang. On Integrable Systems Rigidity for PDEs with Symmetry. PhD thesis, Utrecht University, 12 2017. [26] Marco Zambon. Singular subalgebroids. arXiv:1805.02480, 05 2018.

KU Leuven Departement Wiskunde, Celestijnenlaan 200B, 3001 Leuven, Belgium and Institut fur¨ Mathematik, Haus 9, Karl-Liebknecht-Straße 24, 14476 Potsdam, Germany