arXiv:1412.5230v2 [math.

DG] 27 Jan 2015

Normal Forms and Lie Groupoid Theory

Rui Loja Fernandes

Abstract. In these lectures I discuss the Linearization Theorem for Lie

groupoids, and its relation to the various classical linearization theorems
for submersions, foliations and group actions. In particular, I explain in
some detail the recent metric approach to this problem proposed in [6].
Mathematics Subject Classification (2010). Primary 53D17; Secondary
22A22.
Keywords. Normal form, linearization, Lie groupoid.

Lecture 1: Linearization and Normal Forms

In Differential Geometry one finds many different normal forms results which
share the same flavor. In the last few years we have come to realize that
there is more than a shared flavor to many of these results: they are actually
instances of the same general result. The result in question is a linearization
result for Lie groupoids, first conjectured by Alan Weinstein in [16, 17]. The
first complete proof of the linearization theorem was obtained by Nguyen
Tien Zung in [18]. Since then several clarifications and simplifications of the
proof, as well as more general versions of this result, were obtained (see [3,6]).
In these lectures notes we give an overview of the current status of the theory.
The point of view followed here, which was greatly influenced by an
ongoing collaboration with Matias del Hoyo [68], is that the linearization
theorem can be thought of as an Ehresmanns Theorem for a submersion
onto a stack. Hence, its proof should follow more or less the same steps as
the proof of the classical Ehresmanns Theorem, which can be reduced to
a simple argument using the exponential map of a metric that makes the
submersion Riemannian. Although I will not go at all into geometric stacks
(see the upcoming paper [8]), I will adhere to the metric approach introduced
in [6].
Let us recall the kind of linearization theorems that we have in mind.
The most basic is precisely the following version of Ehresmanns Theorem:

Supported in part by NSF grant DMS 1308472.

2 Rui Loja Fernandes

Theorem 1 (Ehresmann). Let : M N be a proper surjective submersion.

Then is locally trivial: for every y N there is a neighborhood y U N ,
a neighborhood 0 V Ty N , and diffeomorphism:

=
V 1 (y) // 1 (U ) M

pr

=

V // U

One can also assume that there is some extra geometric structure be-
having well with respect to the submersion, and then ask if one can achieve
linearization of both the submersion and the extra geometric structure.
For example, if one assumes that 2 (M ) is a closed 2-form such that the
pullback of to each fiber is non-degenerate, then one can show that is a
locally trivial symplectic fibration (see, e.g., [12]). We will come back to this
later, for now we recall another basic linearization theorem:
Theorem 2 (Reeb). Let F be a foliation of M and let L0 be a compact
leaf of F with finite holonomy. Then there exists a saturated neighborhood
L0 U M , a hol(L0 )-invariant neighborhood 0 V x0 (L0 ), and a
diffeomorphism:
h
=
f0
L V // U M
hol(L0 )

sending the linear foliation to F |U .

h
f0 L denotes the holonomy cover, a hol(L0 )-principal bundle,
Here, L
and the holonomy group hol(L0 ) acts on the normal space x0 (L0 ) via the
linear holonomy representation. By linear foliation we mean the quotient
h
f0 {t}, t x0 (L)}.
of the horizontal foliation {L
Notice that this result generalizes Ehresmanns Theorem, at least when
the fibers of the submersion are connected: any leaf of the foliation by the
fibers of has trivial holonomy so hol(L0 ) acts trivially on the transversal,
and then Reebs theorem immediately yields Ehresmanns Theorem. For this
reason, maybe it is not so surprising that the two results are related.
Let us turn to a third linearization result which, in general, looks to
be of a different nature from the previous results. It is a classical result
from Equivariant Geometry often referred to as the Slice Theorem (or Tube
Theorem):
Theorem 3 (Slice Theorem). Let K be a Lie group acting in a proper fashion
on M . Around any orbit Ox0 M the action can be linearized: there exist
K-invariant neighborhoods Ox0 U M and 0x0 V x0 (Ox0 ) and a
K-equivariant diffeomorphism:

=
K K x0 V // U M
 Normal Forms and Lie Groupoid Theory 3

Here Kx0 acts on the normal space x0 (Ox0 ) via the normal isotropy
representation. If the action is locally free then the orbits form a foliation, the
isotropy groups Kx are finite and hol(Ox ) is a quotient of Kx . Moreover, the
action of Kx on a slice descends to the linear holonomy action of hol(Ox ). The
slice theorem is then a special case of the Reeb stability theorem. However, in
general, the isotropy groups can have positive dimension and the two results
look apparently quite different.
Again, both in the case of foliations and in the case of group actions, we
could consider extra geometric structures (e.g., a metric or a symplectic form)
and ask for linearization taking into account this extra geometric structure.
One can find such linearization theorems in the literature (e.g., the local
normal form theorem for Hamiltonian actions [10]). Let us mention one such
recent result from Poisson geometry, due to Crainic and Marcut [4]:
Theorem 4 (Local normal form around symplectic leaves). Let (M, ) be a
Poisson manifold and let S M be a compact symplectic leaf. If the Poisson
homotopy bundle G y P S is a smooth compact manifold with vanishing
second de Rham cohomology group, then there is a neighborhood S U M ,
and a Poisson diffeomorphism:
: (U, |U ) (P G g, lin ).
We will not discuss here the various terms appearing in the statement
of this theorem, referring the reader to the original work [4]. However, it
should be clear that this result has the same flavor as the previous ones: some
compactness type assumption around a leaf/orbit leads to linearization or a
normal form of the geometric structure in a neighborhood of the leaf/orbit.
Although all these results have the same flavor, they do look quite dif-
ferent. Moreover, the proofs that one can find in the literature of these lin-
earization results are also very distinct. So it may come as a surprise that
they are actually just special cases of a very general linearization theorem.
In order to relate all these linearization theorems, and to understand the
significance of the assumptions one can find in their statements, one needs
a language where all these results fit into the same geometric setup. This
language exists and it is a generalization of the usual Lie theory from groups
to groupoids. We will recall it in the next lecture. After that, we will be in
shape to state the general linearization theorem and explain how the results
stated before are special instances of it.

Lecture 2: Lie Groupoids

In this Lecture we provide a quick introduction to Lie groupoids and Lie alge-
broids. We will focus mostly on some examples which have special relevance
to us. A more detailed discussion, along with proofs, can be found in [1]. Let
us start by recalling:
Definition 1. A groupoid is a small category where all arrows are invertible.
4 Rui Loja Fernandes

Let us spell out this definition. We have a set of objects M , and a set
of arrows G. For each arrow g G we can associate its source s(g) and its
target t(g), resulting in two maps s, t : G M . We also write g : x y
for an arrow with source x and target y.
For any pair of composable arrows we have a product or composition
map:
m : G(2) G, (g, h) 7 gh.
In general, we will denote by G(n) the set of n strings of composable arrows:
G(n) := {(g1 , . . . , gn ) : s(gi ) = t(gi+1 )}.
The multiplication satisfies the associativity property:
(gh)k = g(hk), (g, h, k) G(3) .
For each object x M there is an identity arrow 1x and the identity
property holds:
1t(g) g = g = g1s(g) , g G.
It gives rise to an identity section u : M G, x 7 1x .
For each arrow g G there is an inverse arrow g 1 G, for which the
inverse property holds:
gg 1 = 1t(g) , g 1 g = 1s(g) , g G.
This gives rise to the inverse map : G G, g 7 g 1 .
Definition 2. A morphism of groupoids is a functor F : G H.
This means that we have a map F : G H between the sets of arrows
and a map f : M N between the sets of objects, making the following
diagram commute:
F //
G H
s t s t
   
M // N
f

such that F (gh) = F (g)F (h) if g, h G are composable, and F (1x ) = 1f (x)
for all x M .
We are interested in groupoids and morphisms of groupoids in the
smooth category:
Definition 3. A Lie groupoid is a groupoid G M whose spaces of arrows
and objects are both manifolds, the structure maps s, t, u, m, i are all smooth
maps and such that s and t are submersions. A morphisms of Lie groupoids
is a morphism of groupoids for which the underlying map F : G H is
smooth.
Before we give some examples of Lie groupoids, let us list a few basic
properties.
the unit map u : M G is an embedding and the inverse : G G is
a diffeomorphism.
 Normal Forms and Lie Groupoid Theory 5

The source fibers are embedded submanifolds of G and right multipli-

cation by g : x y is a diffeomorphism between s-fibers
Rg : s1 (y) s1 (x), h 7 hg.
The target fibers are embedded submanifolds of G and left multiplica-
tion by g : x y is a diffeomorphism between t-fibers:
Lg : t1 (x) t1 (y), h 7 gh.
The isotropy group at x:
Gx = s1 (x) t1 (x).
is a Lie group.
The orbit through x:
Ox := t(s1 (x)) = {y M : g : x y}
is a regular immersed (possibly disconnected) submanifold.
The map t : s1 (x) Ox is a principal Gx -bundle.
The connected components of the orbits of a groupoid G M form a
(possibly singular) foliation of M . The set of orbits is called the orbit space
and denoted by M/G. The quotient topology makes the natural map : M
M/G into an open, continuous map. In general, there is no smooth structure
on M/G compatible with the quotient topology, so M/G is a singular space.
One may think of a groupoid as a kind of atlas for this orbit space.
There are several classes of Lie groupoids which will be important for
our purposes. A Lie groupoid G M is called source k-connected if the s-
fibers s1 (x) are k-connected for every x M . When k = 0 we say that G is a
s-connected groupoid, and when k = 1 we say that G is a s-simply connected
groupoid. We call the groupoid etale if dim G = dim M , which is equivalent
to requiring that the source or target map be a local diffeomorphism. The
map (s, t) : G M M is sometimes called the anchor of the groupoid and
a groupoid is called proper if the anchor (s, t) : G M M is a proper
map. In particular, when the source map is proper, we call the groupoid s-
proper. We will see later that proper and s-proper groupoids are, in some
sense, analogues of compact Lie groups.
Example 1. Perhaps the most elementary example of a Lie groupoid is the
unit groupoid M M of any manifold M : for each object x M , there is
exactly one arrow, namely the identity arrow. More Fgenerally, given F
an open
cover {Ui } of M one constructs a cover groupoid i,jI Ui Uj iI Ui :
we picture an arrow as (x, i) (x, j) if x Ui Uj . The structure maps
should be obvious. In both these examples, the orbit space coincides with
the original manifold M and the isotropy groups are all trivial. These are
all examples of proper, etale, Lie groupoids. If the open cover is not second
countable, then the spaces of arrows and objects are not second countable
manifolds. In Lie groupoid theory one sometimes allows manifolds which are
not second countable. However, we will always assume manifolds to be second
countable.
6 Rui Loja Fernandes

Example 2. Another groupoid that one can associate to a manifold is the

pair groupoid M M M : an ordered pair (x, y) determines an arrow
x y. This is an example of transitive groupoid, i.e., a groupoid with
only one orbit. More generally, if K y P M is a principal K-bundle,
then K acts on the pair groupoid P P P by groupoid automorphisms
and the quotient P K P M is a transitive groupoid called the gauge
groupoid. Note that for any x M the isotropy group Gx is isomorphic to
K and the principal Gx -bundle t : s1 (x) M is isomorphic to the original
principal bundle. Conversely, it is easy to see that any transitive groupoid
G M is isomorphic to the gauge groupoid of any of the principal Gx -
bundles t : s1 (x) M . It is easy to check that that the gauge groupoid
associated with a principal K-bundle P M is proper if and only if K is a
compact Lie group. Moreover, it is s-proper (respectively, source k-connected)
if and only if P is compact (respectively, k-connected).

Example 3. To any surjective submersion : M N one can associate

the submersion groupoid M N M M . This is a subgroupoid of the pair
groupoid M M M , but it fails to be transitive if N has more than one
point: the orbits are the fibers of : M N , so the orbit space is precisely
N . The isotropy groups are all trivial. The submersion groupoid is always
proper and it is s-proper if and only if is a proper map.

Example 4. Let F be a foliation of a manifold M . We can associate to it the

fundamental groupoid 1 (F ) M , whose arrows correspond to foliated ho-
motopy classes of paths (relative to the end-points). Given such an arrow [],
the source and target maps are s([]) = (0) and t([]) = (1), while mul-
tiplication corresponds to concatenation of paths. One can show that 1 (F )
is indeed a manifold, but it may fail to be Hausdorff. In fact, this groupoid is
Hausdorff precisely when F has no vanishing cycles. In Lie groupoid theory,
one often allows the total space of a groupoid to be non-Hausdorff, while
M and the source/target fibers are always assumed to be Hausdorff. On the
other hand, we will always assume manifolds to be second countable.
Another groupoid one can associate to a manifold is the holonomy
groupoid Hol(F ) M , whose arrows correspond to holonomy classes of
paths. Again, one can show that Hol(F ) is a manifold, but it may fail to be
Hausdorff. In general, the fundamental groupoid and the holonomy groupoid
are distinct, but there is an obvious groupoid morphism F : 1 (F ) Hol(F ),
which to a homotopy class of a path associates the holonomy class of the path
(recall that the holonomy only depends on the homotopy class of the path).
This map is a local diffeomorphism.
It should be clear that the leaves of these groupoids coincide with the
leaves of F and that the isotropy groups of 1 (F ) (respectively, Hol(F ))
coincide with the fundamental groups (respectively, holonomy groups) of the
leaves. It is easy to check also that 1 (F ) is always s-connected. One can show
that 1 (F ) (respectively, Hol(F )) is s-proper if and only if the leaves of F
are compact and have finite fundamental group (respectively, finite holonomy
 Normal Forms and Lie Groupoid Theory 7

group). In general, it is not so easy to give a characterization in terms of F

of when these groupoids are proper.
Example 5. Let K y M be a smooth action of a Lie group K on a manifold
M . The associated action groupoid K M M has arrows the pairs (k, x)
K M , source/target maps given by s(k, x) = x and t(k, x) = kx, and
composition:
(k1 , y)(k2 , x) = (k1 k2 , x), if y = k1 x.
The isotropy groups of this action are the stabilizers Kx and the orbits coin-
cide with the orbits of the action. The action groupoid is proper (respectively,
s-proper) precisely when the action is proper (respectively, K is compact).
Moreover, this groupoid is source k-connected if and only if K is k-connected.
Lie groupoids, just like Lie groups, have associated infinitesimal objects
known as Lie algebroids:
Definition 4. A Lie algebroid is a vector bundle A M together with a Lie
bracket [ , ]A : (A) (A) (A) on the space of sections and a bundle
map A : A T M , such that the following Leibniz identity holds:
[, f ]A = f [, ]A + A (X)(f )Y, (1)
for all f C (M ) and , (A).
Given a Lie groupd G M the associated Lie algebroid is obtained as
follows. One lets A := ker dM s be the vector bundle whose fibers consists of
the tangent spaces to the s-fibers along the identity section. Then sections of
A can be identified with right-invariant vector fields on the Lie groupoid, so
the usual Lie bracket of vector fields induces a Lie bracket on the sections. The
anchor is obtained by restricting the differential of the target, i.e., A := dt|A .
There is a Lie theory for Lie groupoids/algebroids analogous to the
usual Lie theory for Lie groups/algebras. There is however one big difference:
Lies Third Theorem fails and there are examples of Lie algebroids which are
not associated with a Lie groupoid ( [2]). We shall not give any more details
about this correspondence since we will be working almost exclusively at the
level of Lie groupoids. We refer the reader to [1] for a detailed discussion of
Lie theory in the context of Lie groupoids and algebroids.

Lecture 3: The Linearization Theorem

For a Lie groupoid G M a submanifold N M is called saturated if it is
a union of orbits. Our aim now is to state the linearization theorem which,
under appropriate assumptions, gives a normal form for the Lie groupoid in
a neighborhood of a saturated submanifold. First we will describe this local
normal form, which depends on some standard constructions in Lie groupoid
theory.
Let G M be a Lie groupoid. The tangent Lie groupoid T G T M
is obtained by applying the tangent functor: hence, the spaces of arrows and
objects are the tangent bundles to G and to M , the source and target maps
8 Rui Loja Fernandes

are the differentials ds, dt : T G T M , the multiplication is the differential

dm : (T G)2 T G(2) T G, etc.
Assume now that S M is a saturated submanifold. An important
special case to keep in mind is when S consists of a single orbit of G. Then
we can restrict the groupoid G to S:
GS := t1 (S) = s1 (S),
obtaining a Lie subgroupoid GS S of G M . If we apply the tangent
functor, we obtain a Lie subgroupoid T GS T S of T G T M .
Proposition 5. There is a short exact sequence of Lie groupoids:
1 // T GS // T G // (GS ) // 1

     
0 // T S // T M // (S) // 0

There is an alternative description of the groupoid (GS ) (S) which

sheds some light on its nature. For each arrow g : x y in GS one can
define the linear transformation:
Tg : x (S) y (S), [v] 7 [dg t(v)],
where v Tg G is such that dg s(v) = v. One checks that this map is inde-
pendent of the choice of lifting ve. Moreover, for any identity arrow one has
T1x =idx (S) and for any pair of composable arrows (g, h) G(2) one finds:
Tgh = Tg Th .
This means that GS S acts linearly on the normal bundle (S) S, and
one can form the action groupoid:
GS (S) (S).
The space of arrows of this groupoid is the fiber product GS S (S), with
source and target maps: s(g, v) = v and t(g, v) = Tg v. The product of two
composable arrows (g, v) and (h, w) is then given by:
(g, v)(h, w) = (gh, w).
One then checks that:
Proposition 6. The map F : (GS ) GS (S), vg 7 (g, [dg s(vg )]), is an
isomorphism of Lie groupoids.
Finally, let us observe that the restricted groupoid GS S sits (as the
zero section) inside the Lie groupoid (GS ) (S) as a Lie subgroupoid.
This justifies introducing the following definition:
Definition 5. For a saturated submanifold S of a Lie groupoid G M the
local linear model around S is the groupoid (GS ) (S).
 Normal Forms and Lie Groupoid Theory 9

Example 6 (Submersions). For the submersion groupoid G = M N M M

associated with a submersion : M N , a fiber S = 1 (z) is a saturated
submanifold (actually, an orbit) and we find that
GS := {(x, y) M M : (x) = (y) = z} = S S S.
is the pair groupoid. Notice that the normal bundle M (S) is naturally iso-
morphic to the trivial bundle S Tz N . It follows that the local linear model
is the groupoid:
S S Tz N S Tz N,
which is the direct product of the pair groupoid S S S and the identity
groupoid Tz N Tz N . More importantly, we can view this groupoid as the
submersion groupoid of the projection : S Tz N Tz N .
Example 7 (Foliations). Let F be a foliation of M and let L M be a
leaf. The normal bundle (F ) has a natural flat F -connection, which can be
described as follows. Given a vector field Y X(M ) let us write Y ((F ))
for the corresponding section of the normal bundle. Then, if X X(F ) is a
foliated vector field, one sets:
X Y := [X, Y ].
One checks that this definition is independent of the choice of representative,
and defines a connection : X(F ) ((F )) ((F )), called the Bott
connection. The Jacobi identity shows that this connection is flat.
Given a path : I L in some leaf L of F , parallel transport along
defines a linear map
: (0) (L) (1) (L).
Since the connection is flat, it is clear that this linear map only depends on the
homotopy class []. One obtains a linear action of the fundamental groupoid
1 (F ) on (F ). The restriction of 1 (F ) to the leaf L is the fundamental
groupoid of the leaf L and we obtain the linear model for 1 (F ) along the
leaf L as the action groupoid:
1 (L) (L) (L).
The fundamental groupoid 1 (L) is isomorphic to the gauge goupoid of the
universal covering space L e L, viewed as a principal 1 (L)-bundle. More-
over, (L) is isomorphic to the associated bundle L e (L,x) x (L). It follows
1
that the linear model coincides with the fundamental groupoid of the linear
foliation of (L) = Le (L,x) x (L).
1
The linear map only depends on the holonomy class of , since this
maps coincides with the linearization of the holonomy action along . For
this reason, there is a similar description of the linear model of the holonomy
groupoid Hol(F ) along the leaf L as an action groupoid:
Hol(F )L (L) (L).
Also, the holonomy groupoid is isomorphic to the gauge goupoid of the ho-
lonomy cover Le h L, viewed as a principal hol(L)-bundle and we can also
10 Rui Loja Fernandes

describe the normal bundle (L) as the associated bundle L e h hol(L,x) x (L).
The underlying foliation of this linear model is still the linear foliation of
(L). However, the linear model now depends on the germ of the foliation
around L, i.e., on the non-linear holonomy. Unlike the case of the fundamen-
tal groupoid, the knowledge of the Bott connection is not enough to build
this linear model.
Example 8 (Group actions). Let K be a Lie group that acts on a manifold
M and let Kx be the isotropy group of some x M . For each k Kx , the
map k : M M , y 7 ky, fixes x and maps the orbit Ox to itself. Hence,
dx k induces a linear action of Kx on the normal space x (Ox ), called the
normal isotropy representation. Using this representation, it is not hard to
check that we have a vector bundle isomorphism:
// K Kx x (Ox )
(Ox )


'' uu

Ox
where the action Kx y K x (Ox ) is given by k(g, v) := (gk 1 , kv). More-
over, one has an action of K on (Ox ), which under this isomorphism corre-
sponds to the action:
K y K Kx x (Ox ), k[(k , v)] = [(kk , v)].
Now consider the action Lie groupoid K M M . One checks that
the local linear model around the orbit Ox is just the action Lie groupoid
K (Ox ) (Ox ), which under the isomorphism above corresponds to the
action groupoid:
K (K Kx x (Ox )) K Kx x (Ox ).
As we have already mentioned above, the Linearization Theorem states
that, under appropriate conditions, the groupoid is locally isomorphic around
a saturated submanifold to its local model. In order to make precise the
expression locally isomorphic we introduce the following definition:
Definition 6. Let G M be a Lie groupoid and S M a saturated sub-
manifold. A groupoid neighborhood of GS S is a pair of open sets U S
and Ue GS such that U e U is a subgroupoid of G M . A groupoid
e e = GU .
neighborhood U U is said to be full if U
Our first version of the linearization theorem reads as follows:
Theorem 7 (Weak linearization). Let G M be a Lie groupoid with a 2-
metric (2) . Then G is weakly linearizable around any saturated submanifold
S M : there are groupoid neighborhoods U e U of GS S in G M and
e
V V of GS S in the local model (GS ) (S), and an isomorphism
of Lie groupoids:

(U e U)
= (Ve V ),
which is the identity on GS .
 Normal Forms and Lie Groupoid Theory 11

A 2-metric is a special Riemannian metric in the space of composable

arrows G(2) . We will discuss them in detail in the next lecture. For now
we remark that for a proper groupoid 2-metrics exist and, moreover, every
groupoid neighborhood contains a full groupoid neighborhood. Hence:
Corollary 8 (Linearization of proper groupoids). Let G M be a proper Lie
groupoid. Then G is linearizable around any saturated submanifold S M :
there exist open neighborhoods S U M and S V (S) and an
isomorphism of Lie groupoids

(GU U )
= ((GS )V V ),
which is the identity on GS .
This corollary does not yet yield the various linearization results stated
in Lecture 1. The reason is that the assumptions do not guarantee the ex-
istence of a saturated neighborhood S U M . This can be realized for
s-proper groupoids, since for such groupoids every neighborhood U of a sat-
urated embedded submanifold S, contains a saturated neighborhood of S.
Corollary 9 (Invariant linearization of s-proper groupoids). Let G M
be an s-proper Lie groupoid. Then G is invariantly linearizable around any
saturated submanifold S M : there exist saturated open neighborhoods S
U M and S V (S) and an isomorphism of Lie groupoids

(GU U )
= ((GS )V V ),
which is the identity on GS .
Example 9. For a proper submersion : M N the associated submersion
groupoid M N M M is s-proper. Using the description of the local
model that we gave before, it follows that for any fiber 1 (z) there is a
saturated open neighborhood 1 (z) U M where the submersion if
locally isomorphic to the trivial submersion 1 (z) V V , for some open
neighborhood 0 V Tz N . Hence, we recover the classical Ehresmanns
Theorem.
Example 10. Let F be a foliation of M whose leaves are compact with finite
holonomy. Then the holonomy groupoid Hol(F ) M is s-proper. Using
the description of the local model given before, it follows that for any leaf L
there is a saturated neighborhood L U M where the canonical foliation is
e h hol(L,x) V , for some hol(L, x)-
isomorphic to the linear foliation of (L) = L
invariant, neighborhood 0 V x (L). Hence, we recover the Local Reeb
Stability Theorem.
Example 11. Let K M M be an action of a compact Lie group. Then the
the action groupoid is s-proper and we obtain invariant linearization. From
the description of the local model, it follows that for any orbit Ox there is a
saturated open neighborhood Ox U M and a K-equivariant isomorphism
U K Kx V , where 0 V x (Ox ) is a Kx -invariant neighborhood. Hence,
12 Rui Loja Fernandes

we recover the slice theorem for actions of compact groups. For a general
proper action, the results above only give weak linearization, which does not
allow to deduce the slice theorem. However, due to the particular structure of
the action groupoid, every orbit has a saturated neighborhood and one has a
uniform bound for the injectivity radius of the 2-metric. This gives invariant
linearization and leads to the Slice Theorem for any proper action.
Example 12. Let (M, ) be a Poisson manifold. The cotangent bundle T M
has a natural Lie algebroid structure with anchor : T M T M given by
contraction by and Lie bracket on sections (i.e., 1-forms):
[, ] := L() L() d(, ).
In general, this Lie algebroid fails to be integrable. However, under the as-
sumptions of the local normal form theorem stated in Lecture 1, this groupoid
is integrable, and then its source 1-connected integration is an s-proper Lie
groupoid whose orbits are the symplectic leaves. This groupoid can then be
linearized around a symplectic leaf, but this linearization does not yet yield
the local canonical form for the Poisson structure.
It turns out that the source 1-connected integration is a symplectic Lie
groupoid, i.e., there is a symplectic structure on its space of arrows which is
compatible with multiplication. One can apply a Moser type trick to further
bring the symplectic structure on the local normal form to a canonical form,
which then yields the canonical form of the Poisson structure. The details of
this approach, which differ from the original proof of the canonical form due
to Crainic and Marcut, can be found in [4].

Lecture 4: Groupoid Metrics and Linearization

Let us recall that a submersion : (M, ) N is called a Riemannian
submersion if the fibers are equidistant. The base N gets an induced metric
for which the linear maps dx : (ker dx ) T(x) N , x M , are
all isometries. More generally, a (possibly singular) foliation in a Riemannian
manifold M is called a Riemannian foliation if the leaves are equidistant. This
is equivalent to the following property: any geodesic which is perpendicular
to one leaf at some point stays perpendicular to all leaves that it intersects.
A simple proof of Ereshmans Theorem can be obtained by choosing a
metric on the total space of the submersion : M N , that makes it into
a Riemannian submersion. A partition of unit argument shows that this is
always possible. Then the linearization map is just the exponential map of
the normal bundle of a fiber, which maps onto a saturated neighborhood of
the fiber, provided the submersion is proper. We will see that a proof similar
in spirit also works for the general linearization theorem (Theorem 7).
Let G M be a Lie groupoid. Like any category, G has a simplicial
model:
////// (n) ////// //// (2) //// (1) //// G(0) [ // G(0) /G(1) ]
... //// G // //// G // G
 Normal Forms and Lie Groupoid Theory 13

where, for each n N, the face maps i : G(n) G(n1) , i = 0, . . . , n and the
degeneracy maps i : G(n) G(n+1) , i = 1, . . . , n, are defined as follows: for
a n-string of composable arrows the i-th face map associates the (n1)-string
of composable arrows obtained by omitting the i-object:
g1 gn
! g1 gi gi+1 gn
xx xx xx yy xx
i =

while the i-th degeneracy map inserts into a n-string of composable arrows
an identity at the i-th entry:
1xi

g1 gn
! g1 gn
xx xx xx  xx
i =

For a Lie groupoid there is, additionally, a natural action of Sn+1 on G(n) : for
a string of n-composable arrows (g1 , . . . , gn ) choose (n+1) arrows (h0 , . . . , hn ),
all with the same source, so that:

h0 hn
h1 h2 hn1

xxyy g1 xx~~ g2
(( xx gn

Then the Sn+1 -action on the arrows (h0 , . . . , hn ) by permutation gives a well
defined Sn+1 -action on G(n) . Notice that this action permutes the face maps
i , since there are maps i : Sn+1 Sn such that:
i = i () (i) .
Definition 7. A n-metric (n N) on a groupoid G M is a Riemannian
metric (n) on G(n) which is Sn+1 -invariant and for which all the face maps
i : G(n) G(n1) are Riemmanian submersions.
Actually, it is enough in this definition to ask that one of the face maps
is a Riemannian submersion: the assumption that the action of Sn+1 is by
isometries implies that if one face map is a Riemannian submersion then all
the face maps are Riemannian submersions.
For any n 1, the metrics induced on G(n1) by the different face maps
i : G(n) G(n1) coincide, giving a well defined metric (n1) , which is a
(n 1)-metric. Obviously, one can repeat this process, so that a n-metric on
G(n) determines a k-metric on G(k) for all 0 k n.
Example 13 (0-metrics). When n = 0 we adopt the convention that (0) is a
metric on M = G(0) which makes the orbit space a Riemannian foliation and
is invariant under the action of G on the normal space to the orbits, i.e., such
that each arrow Tg acts by isometries on (O). Such 0-metrics were studied
by Pflaum, Posthuma and Tang in [15], in the case of proper groupoids. One
can think of such metrics as determining a metric on the (possibly singular)
14 Rui Loja Fernandes

orbit space M/G. Indeed, it is proved in [15] that the distance on the orbit
space determined by (0) enjoys some nice properties.
Example 14 (1-metrics). A 1-metric is just a metric (1) on the space of arrows
G = G(1) for which the source and target maps are Riemannian submersions
and inversion is an isometry. These metrics were studied by Gallego et al.
in [9]. A 1-metric induces a 0-metric (0) on M = G(0) , for which the orbit
foliation is Riemannian.
Example 15 (2-metrics). A 2-metric is a metric (2) in the space of com-
posable arrows G(2) , which is invariant under the S3 -action generated by the
involution (g1 , g2 ) 7 (g21 , g11 ) and the 3-cycle (g1 , g2 ) 7 ((g1 g2 )1 , g1 ), and
for which multiplication is a Riemannian submersion. Hence, a 2-metric on
G(2) induces a 1-metric on G(1) . These metrics were introduced in [6].
Notice that the first 3 stages of the nerve
1
// s //
G(2) m //
// G // M
2 t

completely determine the remaining G(n) , for n 3. Hence, one should expect
that n-metrics, for n 3, are determined by their 2-metrics. In fact, one has
the following properties, whose proof can be found in [7, 8]:
there is at most one 3-metric inducing a given 2-metric and every 3-
metric has a unique extension to an n-metric for every n 3.
there are examples of groupoids which admit an n-metric, but do not
admit a n + 1-metric, for n = 0, 1, 2.
uniqueness fails in low degrees: one can have, e.g., two different 2-metrics
on G(2) inducing the same 1-metric on G(1) .
The geometric realization of the nerve of a groupoid G M is usu-
ally denoted by BG, can be seen as the classifying space of principal G-
bundles (see [11]). Two Morita equivalent groupoids G1 M1 and G2 M2
(see [5] or [13]), give rise to homotopy equivalent spaces BG1 and BG2 . An
alternative point of view, is to think of BG as a geometric stack with atlas
G M , and two atlas represent the same stack if they are Morita equivalent
groupoids. The following result shows that one may think of a n-metric as a
metric in BG:
Proposition 10 ( [7, 8]). If G M and H N are Morita equivalent
groupoids, then G admits a n-metric if and only if H admits a n-metric.
In fact, it is shown in [7, 8] that it is possible to transport a n-metric
via a Morita equivalence. This constructions depends on some choices, but
the transversal component of the n-metric is preserved. We refer to those
references for a proof and more details.
Since a n-metric determines a k-metric, for all 0 k n, a necessary
condition for a groupoid G M to admit a n-metric is that the orbit foliation
can be made Riemannian. This places already some strong restrictions on the
 Normal Forms and Lie Groupoid Theory 15

class of groupoids admitting a n-metric. So one may wonder when can one
find such metrics. One important result in this direction is that any proper
groupoid admits such a metric:
Theorem 11. A proper Lie groupoid G M admits n-metrics for all n 0.
Proof. The proof uses the following trick, called in [6] the gauge trick. For
each n, consider the manifold G[n] Gn of n-tuples of arrows with the same
source, and the map
(n) : G[n+1] G(n) , (h0 , h1 . . . , hn ) 7 (h0 h11 , h1 h1 1
2 , . . . , hn1 hn ).

The fibers of (n) coincide with the orbits of the right-multiplication action,
f f h0
[n+1]
G xG o
o oo
(h0 , . . . , hn ) k = (h0 k, . . . , hn k). rrrr k
xxrr
hn
and this action is free and proper, hence defining a principal G-bundle. The
strategy is to define a metric on G[n+1] in a such way that (n) becomes a
Riemannian submersion, and that the resulting metric on G(n) is a n-metric.
The group Sn+1 acts on the manifold G[n+1] by permuting its coordi-
nates, and this action covers the action Sn+1 y G(n) , so the map (n) is
Sn+1 equivariant. On the other hand, there are (n + 1) left groupoid actions
G y G[n+1] , each consisting in left multiplication on a given coordinate.
oo ff h0 ff h0 ff h0
k

o o o o
k o o ... o o
rrr rrr rr
r
xxrrr xxrrr k xxrrr
hn hn oo hn
These left actions commute with the above right action and cover (n + 1)-
principal actions G y G(n) , with projection the face maps i : G(n)
G(n1) .
Now for a proper groupoid one can use averaging to construct a metric
on G which is invariant under the left action of G on itself by left translations.
The product metric on G[n+1] is invariant both under the (n+1) left G-actions
above and the Sn+1 -action. In general, it will not be invariant under the right
G-action G y G[n+1] G(n) , but we can average it to obtain a new metric
which is invariant under all actions. It follows that the resulting metric on
G(n) is an n-metric. 
Remark 12. The maps (n) : G[n+1] G(n) , n = 0, 1, . . . that appear in
this proof form the simplicial model for the universal principal G-bundle
: EG BG. This bundle plays a key role in many different constructions
associated with the groupoid (see [7]).
Still, there are many examples of groupoids, which are not necessarily
proper, but admit a n-metric. Some are given in the next set of examples.
16 Rui Loja Fernandes

Example 16. The unit groupoid M M obviously admits n-metrics. The

pair groupoid M M M and, more generally, the submersion groupoid
M N M M associated with a submersion : M N , are proper so
also admit n-metrics. For example, if is a metric on M which makes a
Riemannian submersion, then one can take (0) = , (1) = p1 +p2 pN N ,
(2) = p1 + p2 + p3 2pN N , etc.
Example 17. Let F be a foliation of M . Then Hol(F ) and 1 (F ) admit a n-
metric if and only if F can be made into a Riemannian foliation. We already
know that if these groupoids admit a n-metric, then the underlying foliation,
i.e. F , must be Riemannian. Conversely, if F is Riemannian then Hol(F ) is
a proper groupoid (see, e.g., [14]), so it carries a n-metric. Since 1 (F ) is a
covering of Hol(F ), it also admits a n-metric. Not that, in general, 1 (F )
does not need to be proper (e.g., if the fundamental group of some leaf is not
finite).
Example 18. Any Lie group admits a n-metric. More generally, the action Lie
groupoid associated with any isometric action of a Lie group on a Riemannian
manifold admits a n-metric. This can be shown using a gauge trick, similar
to the one used in the proof of existence of a n-metric for a proper groupoid
sketched above (see [6]). Note that an isometric action does not need not be
proper (e.g., if some isotropy group is not compact).
Finally, we can sketch the proof of the main Linearization Theorem,
which we restate now in the following way:
Theorem 13. Let G M be a Lie groupoid endowed with a 2-metric (2) ,
and let S M be a saturated embedded submanifold. Then the exponential
map defines a weak linearization of G around S.
Proof. One choses a neighborhood S V (S) where the exponential map
of (0) is a diffeomorphism onto its image, and set
Ve = (ds)1 (V ) (dt)1 (V ) Domain of exp (1) (GS ).

One shows that Ve V is a groupoid neighborhood of GS S in the linear

model (GS ) (S) and that we have a commutative diagram:
exp(2)
Ve V Ve // G(2)

   exp(1)   
Ve // G

  exp(0)  
V // M
It follows that the exponential maps of (1) and (0) give the desired weak
linearization, i.e., a groupoid isomorphism
exp
(Ve V )
= (exp(Ve ) exp(V )).

 Normal Forms and Lie Groupoid Theory 17

Acknowledgments
I would specially like to thank Matias del Hoyo, for much of the work reported
in these notes rests upon our ongoing collaboration [68]. He and my two PhD
students, Daan Michiels and Joel Villatoro, made several very usual com-
ments on a preliminary version of these notes, that helped improving them.
I would also like to thank Marius Crainic, Ioan Marcut, David Martinez-
Torres and Ivan Struchiner, for many fruitful discussions which helped shape
my view on the linearization problem. Finally, thanks go also to the organiz-
ers of the XXXIII WGMP, for the invitation to deliver these lectures and for
a wonderful stay in Bialowieza.

References
[1] M. Crainic and R.L. Fernandes; Lectures on integrability of Lie brackets; Geom.
Topol. Monogr. 17 (2011), 1-107.
[2] M. Crainic and R. L. Fernandes; Integrability of Lie brackets; Ann. of Math. (2)
157 (2003), 575-620.
[3] M. Crainic and I. Struchiner; On the Linearization Theorem for proper Lie
groupoids; Ann. Scient. Ec. Norm. Sup. 4e serie, 46 (2013), 723-746.
[4] M. Crainic and I. Marcut; A Normal Form Theorem around Symplectic Leaves;
J. Differential Geom. 92 (2012), no. 3, 417-461.
[5] M. del Hoyo; Lie groupoids and their orbispaces; Portugal. Math. 70 (2013),
161-209.
[6] M. del Hoyo and R.L. Fernandes; Riemannian metrics on Lie groupoids;
preprint arXiv:1404.5989.
[7] M. del Hoyo and R.L. Fernandes; Simplicial metrics on Lie groupoids; Work
in progress.
[8] M. del Hoyo and R.L. Fernandes; A stacky version of Ehresmanns Theorem;
Work in progress.
[9] E. Gallego, L. Gualandri, G. Hector and A. Reventos; Groupodes Rieman-
niens; Publ. Mat. 33 (1989), no. 3, 417422.
[10] V. Guillemin and S. Sternberg; Symplectic techniques in physics, Cambridge
Univ. Press, Cambridge, 1984.
[11] A. Haefliger, Homotopy and integrability, in ManifoldsAmsterdam 1970 (Proc.
Nuffic Summer School), Lecture Notes in Mathematics, Vol. 197 197, Berlin,
New York: Springer-Verlag, 133163.
[12] D. McDuff and D. Salamon; Introduction to symplectic topology, 2nd edition,
Oxford Mathematical Monographs, Oxford University Press, New York, 1998.
[13] I. Moerdijk and J. Mrcun; Introduction to foliations and Lie groupoids; Cam-
bridge Stud. Adv. Math. 91, Cambridge University Press, Cambridge, 2003.
[14] P. Molino; Riemannian foliations, Progress in mathematics vol. 73, Birkhauser,
Basel, 1988.
[15] M. Pflaum, H. Poshtuma and X. Tang; Geometry of orbit spaces of proper Lie
groupoids; to appear in J. Reine Angew. Math.. Preprint arXiv:1101.0180.
18 Rui Loja Fernandes

[16] A. Weinstein; Linearization problems for Lie algebroids and Lie groupoids,
Lett. Math. Phys. 52 (2000), 93-102.
[17] A. Weinstein; Linearization of regular proper groupoids, J. Inst. Math. Jussieu.
1 (2002), 493-511.
[18] N.T. Zung; Proper groupoids and momentum maps: linearization, affinity and
convexity, Ann. Scient. Ec. Norm. Sup. 4e serie, 39 (2006), 841-869.

Rui Loja Fernandes

Department of Mathematics
University of Illinois at Urbana-Champaign
1409 W. Green Street
Urbana, IL 61801, USA
e-mail: [email protected]