The Atiyah Lie groupoid of a -principal bundle is the Lie groupoid whose objects are the fibers of the bundle, and whose morphisms are the -equivariant morphisms between the fibers. Schematically:
Its Lie algebroid is the Atiyah Lie algebroid of .
Both the Atiyah Lie groupoid and its Lie algebroid are used to characterize and are characterized by connections on .
As generally for every Lie algebroid, there are different Lie groupoids integrating the Atiyah Lie algebroid. We describe two of them.
The Aityah Lie algebroid of the principal bundle comes canonically with a morphism to the tangent Lie algebroid.
The simplest Lie integration of the tangent Lie algebroid is the codiscrete groupoid of . On the other hand, the universal integration is the fundamental groupoid (both coincide precisey if is a simply connected space).
Accordingly, there is a version of the Atiyah Lie groupoid over , and a richer version over .
For a Lie group and a -principal bundle, the Atiyah groupoid – also called the gauge groupoid or transport groupoid – of is the Lie groupoid with
The Atiyah groupoid sits in an sequence of groupoids
where
is the adjoint bundle of groups associated via the adjoint action of on itself; regarded as a smooth union of one-object groupoids coming from groups;
is the codiscrete groupoid of
the functor is the identity on objects and on morphisms given by the canonical identification , where again we use the diagonal action of on .
the functor is the unique one that is the identity on objects.
What is all of this stuff? I don't understand either or . —Toby
David Roberts: It’s to do with the diagonal action of on as opposed to the antidiagonal (if is abelian) or the action on only one factor. I agree that it’s a bad notation.
Toby: How well do you think it works now, with the notation suppressed and a note added in words? (For what it's worth, the diagonal action seems to me the only obvious thing to do here, although admittedly the others that you mention do exist.)
Todd: I personally believe it works well. A small note is that this construction can also be regarded as a tensor product, regarding the first factor as a right -module and the second a left module, where the actions are related by .
Toby: H'm, maybe we should write diagonal action? if there's something interesting to say about it.
Notice that a splitting (a section)
of the Atiyah groupoid is a trivialization of . On the other hand, locally on contractible we have with the fundamental groupoid of , and a splitting is still a trivialization over but indicates now that one may want to interpret it as giving rise to a flat connection.
Indeed, we have the sequence of of surjective and full functors of path categories
with the fundamental groupoid and the smooth path groupoid and may refine the Atiyah groupoid by pulling back along these.
Write therefore for the pullback
A splitting of the top row is now precisely a flat connection on .
If we pull back further to
then splittings of are precisely (not necessarily flat) connections on .
All this is more well known in terms of the Lie algebroid underlying the Atiyah Lie groupoid, i.e. the Atiyah Lie algebroid sequence
where
is the adjoint bundle of Lie algebras, associated via the adjoint action of on its Lie algebra;
is the Atiyah Lie algebroid
is the tangent Lie algebroid.
Indeed, a splitting of this sequence in the category of Lie algebroids is precisely again a flat connection on and integrates under Lie integration to the splitting of discussed above.
To get non-flat connections in the literature one often sees discussed splittings of the Atiyah Lie algebroid sequence in the category just of vector bundles. In that case one finds the curvature of the connection precisely as the obstruction to having a splitting even in Lie algebroids.
One can describe non-flat connections without leaving the context of Lie algebroids by passing to higher Lie algebroids, namely -algebroids.
…
The following is not in the literature. – Urs
We discuss the Atiyah Lie groupoid along the lines of nonabelian groupoid cohomology of the path n-groupoid of the base manifold .
Write for the delooping of . Write for the path ∞-groupoid of . For the present purpose of Atiyah 1-groupoids this may be taken to be simply the path 2-groupoid.
Much of what makes the following discussion somewhat less than straightforward is the fact already manifest in and discussed at nonabelian group cohomology: nonabelian degree 2-cocycles with coefficients in the (delooping of the) automorphism 2-group classify really action groupoids of actions on bundles with typical fiber , while the object that one wants to see classified (there: a group extension; here: the Atiyah Lie groupoid) is obtained from this only after forgetting the elements in these fibers, and just remembering the action.
A -principal bundle is classified by a morphism .
Putting a connection on give a parallel transport morphism into the inner automorphism 2-group of , which fits into a diagram
where we have added the morphism into the automorphism 2-group.
The action groupoid of the Atiyah Lie groupoid acting on the associated bundle , regarded as a groupoid over , is classified by : it is the homotopy fiber of , in that there is a homotopy pullback diagram
The Atiyah Lie groupoid itself is the corresponding quotient
Setup
Choose to model all ∞-Lie groupoids in terms of the local projective model structure on simplicial presheaves on the site CartSp of cartesian spaces.
All the Lie -groupoids appearing here are represented by Kan complex valued sheaves. They satisfy descent on contractible domains, hence on all representables, hence are fibrant in the local projective model structure.
As discussed in the subsection on cofibrant objects, a cofibrant replacement for the manifold is given by the Cech nerve of any good cover . So fix any such good cover of .
Accordingly, a cofibrant version of the path ∞-groupoid is given by the coend . This is the -groupoid generated in degree from
points in -fold intersections;
paths in -fold intersections
surfaces in -fold intersections,
etc.
The bundle
The bundle is, by definition, the one classified by the morphism
represented by a morphism
in .
We compute the homotopy pullback
as usual in terms of an ordinary pullback
along the representing cocycle of the universal bundle , which is the groupoid
whose morphisms are commuting trianfles in , with the projection to being the projection onto the horizontal component.
So the objects of are pairs consisting of a point in a patch , together with an element , and a morphism exists iff and is then given by
The homotopy fiber of
We compute now the homotopy pullback
for coming from the choice of any connection on , in terms of the ordinary pullback
along the representing cocycle of the generalized universal bundle of the automorphism 2-group .
The computation is (apart from the fact that it takes place in a smooth context which is taken care of by the abstract nonsense) entirely analogous as that of the 2-cocycles discussed at nonabelian group cohomology, the only difference being that the domain is here not a one-object groupoid that is the delooping of a group, but the genuine many-object groupoid .
As there, we ventually make use of the translation between strict 2-groups and crossed modules - as described at strict 2-group – in terms of crossed modules – in order to translate some of the diagrams to follow into formulas. We follow the convention LB, as described there, in order to do so.
In
an object is an automorphism
a morphism is a triangle
a 2-cell
is a 2-cell
in such that
Using this, we find the pullback 2-groupoid to be given as follows:
an object is pair consisting of a point and a patch containing it, and an automorphism
a morphism is, over a path in a diagram
where is the parallel transport of the chosen connection on along the path ;
and over a Cech morphism a diagram
a 2-cell over a surface is
and over a Cech 2-cell is
The identification with the groupoid
We now show that the homotopy fiber computed above is indeed the groupoid .
The main point is the realizaiton of the Atiyah Lie groupoid as a groupoid over .
A morphism in over a path is a fiber homomorphism . Since we have chose a connection on , we may always express this relative to the parallel transport of the connection along , which provides a reference frame in which to measure morpisms between fibers, and defines for given a group element defined by
When we express composition of fiber morphisms in terms of the compositiion of the group elements associated to them this way, we find an adjoint action of the parallel transport on the group elements:
We see that this is precisely the composition rule satisfied by the group elements labeled in the above description of the homotopy pullback:
(with labels in terms of the L B-convention-group – in terms of crossed modules](http://ncatlab.org/nlab/show/strict+2-group#InTermsOfCrossedModules))).
If and are two homotopica but different paths from to , then the parallel transport between them differs by the integrated curvature
The same morphism corresponds to two different group elements and depending on whether it is expressed with respect to the path or . The difference is precisely the integrated curvature
i.e.
This is precisely the relation imposed by the 2-cells in the homotopy pullback, as read off from the above diagrams.
In summary, this shows that in the homotopy pullback
if we forget the labels on the diagonal morphism, the resulting groupoid is the Atiyah groupoid over ;
the action of this on the labels, taken into account, is the action of the Atiyah Lie groupoid on .