Technically, a smooth groupoid is a groupoid-valued stack on the site of smooth manifolds (sometimes called a “smooth stack”, but this is somewhat ambiguous), or equivalently on its dense subsite CartSp of just Cartesian spaces and smooth functions between them. This is equivalently an 1-truncated smooth ∞-groupoid.

For more see also at geometry of physics – smooth homotopy types.

Definition

Pre-smooth groupoids

For more on this see at geometry of physics – coordinate systems.

In view of the motivation for sheaves, cohomology and higher stacks and in direct analogy with the discussion at geometry of physics – smooth sets, just replacing sets by groupoids throughout, we set:

See at geometry of physics – homotopy types – groupoids for more on bare groupoids. Here we will freely assume familiarity with these.

The Yoneda lemma will turn this intuition into a theorem. For that we need to speak of homomorphisms of pre-smooth groupoids, hence of “smooth functors” betwen pre-smooth groupoids.

Definition

A homomorphism or smooth map between pre-smooth groupoids is a natural transformation between the presheaves that they are. We write

PreSmooth1Type≔Funct(CartSp op,Grpd) PreSmooth1Type \coloneqq Funct(CartSp^{op}, Grpd)

for the functor category, the category of pre-smooth groupoids, def. , regarded naturally as a Grpd-enriched category.

This means that for X,Y∈PreSmooth1TypeX,Y \in PreSmooth1Type two pre-smooth groupoids, then the bare hom-groupoid

PreSmooth1Type(X,Y) •∈Grpd PreSmooth1Type(X,Y)_\bullet \in Grpd

between them has

as objects the natural transformations f:X→Yf \colon X \to Y between XX and YY regarded as functors, hence collections {f(ℝ n)} n∈ℕ\{f(\mathbb{R}^n)\}_{n \in \mathbb{N}} of functors between groupoids of probes

f(ℝ n):X(ℝ n)⟶Y(ℝ n) f(\mathbb{R}^n) \colon X(\mathbb{R}^n) \longrightarrow Y(\mathbb{R}^n)

such that for every smooth function ϕ:ℝ n 1→ℝ n 2\phi \colon \mathbb{R}^{n_1} \to \mathbb{R}^{n_2} between abstract coordinate charts, these form a (strictly) commuting diagram with the probe-pullback functors of XX and YY:

X(ℝ n 1) ⟶f(ℝ n) Y(ℝ n 1) X(ϕ)↓ ↓ Y(ϕ) X(ℝ n 2) ⟶f(ℝ n) Y(ℝ n 2); \array{ X(\mathbb{R}^{n_1}) &\overset{f(\mathbb{R}^n)}{\longrightarrow}& Y(\mathbb{R}^{n_1}) \\ {}^{\mathllap{X(\phi)}}\downarrow && \downarrow^{\mathrlap{Y(\phi)}} \\ X(\mathbb{R}^{n_2}) &\overset{f(\mathbb{R}^n)}{\longrightarrow}& Y(\mathbb{R}^{n_2}) } \,;
as morphisms H:f→gH \colon f \rightarrow g the “modifications” of these, hence collections H(ℝ n):f(ℝ n)→g(ℝ n)H(\mathbb{R}^n) \colon f(\mathbb{R}^n)\to g(\mathbb{R}^n) of natural transformations, such that for every smooth function ϕ:ℝ n 1→ℝ n 2\phi \colon \mathbb{R}^{n_1} \to \mathbb{R}^{n_2} the whiskering of these satisfies

X(ℝ n 1) ↓ X(ϕ) ↗↘ f(ℝ n 2) X(ℝ n 2) ⇓ H(ℝ n 2) Y(ℝ n 2) ↘↗ g(ℝ n 1)= ↗↘ f(ℝ n 1) X(ℝ n 1) ⇓ H(ℝ n 1) Y(ℝ n 1) ↘↗ g(ℝ n 1) ↓ Y(ϕ) Y(ℝ n 2) \array{ X(\mathbb{R}^{n_1}) \\ \downarrow^{\mathrlap{X(\phi)}} \\ & \nearrow \searrow^{\mathrlap{f(\mathbb{R}^{n_2})}} \\ X(\mathbb{R}^{n_2}) &\Downarrow^{H(\mathbb{R}^{n_2})}& Y(\mathbb{R}^{n_2}) \\ & \searrow \nearrow_{\mathrlap{g(\mathbb{R}^{n_1})}} } \;\;\; = \;\;\; \array{ & \nearrow \searrow^{\mathrlap{f(\mathbb{R}^{n_1})}} \\ X(\mathbb{R}^{n_1}) &\Downarrow^{H(\mathbb{R}^{n_1})}& Y(\mathbb{R}^{n_1}) \\ & \searrow \nearrow_{\mathrlap{g(\mathbb{R}^{n_1})}} \\ && \downarrow^{\mathrlap{Y(\phi)}} \\ && Y(\mathbb{R}^{n_2}) }

The following is the Yoneda lemma in this context, and it says that this intuition in remark is fully correct:

Proposition

Every Cartesian space ℝ n\mathbb{R}^n defines a pre-smooth groupoid ℝ̲ n\underline{\mathbb{R}}^n, def. , by the assignment

ℝ̲ n:ℝ k↦C ∞(ℝ k,ℝ n)∈Set↪Grpd \underline{\mathbb{R}}^n \colon \mathbb{R}^k \mapsto C^\infty(\mathbb{R}^k, \mathbb{R}^n) \in Set \hookrightarrow Grpd

where the set of smooth functions on the right is regarded as a groupoid with only identity morphisms. This construction constitutes a fully faithful functor

(−)̲:CartSp↪PreSmooth1Type \underline{(-)} \colon CartSp \hookrightarrow PreSmooth1Type

making CartSp a full subcategory of that of pre-smooth groupoids (the Yoneda embedding). Under this embedding for any pre-smooth groupoid X∈PreSmooth1TypeX\in PreSmooth1Type and any Cartesian space ℝ n\mathbb{R}^n, there is a natural equivalence of groupoids

PreSmooth1Type(ℝ̲ n,X)≃X(ℝ n) PreSmooth1Type(\underline{\mathbb{R}}^n, X) \simeq X(\mathbb{R}^n)

(the Yoneda lemma proper).

Example

For XX a smooth set (e.g. a smooth manifold), hence in particular a functor

X:CartSp op→Set X \colon CartSp^{op} \to Set

then its embedding into groupoids

X:CartSp op→Set↪Grpd X \colon CartSp^{op} \to Set \hookrightarrow Grpd

is a pre-smooth groupoid.

More generally:

Example

Let

𝒢 •={𝒢 1⟶t⟵i⟶s𝒢 0} \mathcal{G}_\bullet = \left\{ \mathcal{G}_1 \stackrel{\overset{s}{\longrightarrow}}{\stackrel{\overset{i}{\longleftarrow}}{\underset{t}{\longrightarrow}}} \mathcal{G}_0 \right\}

be an internal groupoid in smooth sets, hence a pair 𝒢 0,𝒢 1∈Smooth0Type\mathcal{G}_0, \mathcal{G}_1 \in Smooth0Type of smooth sets, equipped with source, target, identity homomorphisms of smooth sets between them, and equipped with a compatibly unital, associative and invertible composition map 𝒢 1×𝒢 0𝒢 1→𝒢 1\mathcal{G}_1 \underset{\mathcal{G}_0}{\times} \mathcal{G}_1 \to \mathcal{G}_1. By definition of smooth sets, this means that for every abstract coordinate system ℝ n\mathbb{R}^n then

𝒢(ℝ n) •={𝒢(ℝ n) 1⟶t(ℝ n)⟵i(ℝ n)⟶s(ℝ n)𝒢(ℝ n) 0} \mathcal{G}(\mathbb{R}^n)_\bullet = \left\{ \mathcal{G}(\mathbb{R}^n)_1 \stackrel{\overset{s(\mathbb{R}^n)}{\longrightarrow}}{\stackrel{\overset{i(\mathbb{R}^n)}{\longleftarrow}}{\underset{t(\mathbb{R}^n)}{\longrightarrow}}} \mathcal{G}(\mathbb{R}^n)_0 \right\}

is a bare groupoid. Moreover, this assignment is functorial and hence defines a pre-smooth groupoid

ℝ n↦𝒢(ℝ n) •. \mathbb{R}^n \mapsto \mathcal{G}(\mathbb{R}^n)_\bullet \,.

This exhibits sequence of full subcategory inclusions

Grp↪LieGrpd↪DiffGrpd↪Grpd(Smooth0Type)↪PreSmooth1Type Grp \hookrightarrow LieGrpd \hookrightarrow DiffGrpd \hookrightarrow Grpd(Smooth0Type) \hookrightarrow PreSmooth1Type

of internal groupoids in smooth sets into pre-smooth groupoids, and hence (by further restriction) also of diffeological groupoids, of Lie groupoids and of course of just bare groupoids, too.

Regarding a bare groupoid 𝒢 •\mathcal{G}_\bullet as a pre-smooth groupoid this way means to regard it as equipped with discrete smooth structure. It is given by the constant presheaf

ℝ n↦𝒢 • \mathbb{R}^n \mapsto \mathcal{G}_\bullet

exhibiting the fact that smooth functions from an ℝ n\mathbb{R}^n into a geometrically discrete space are constant on one of the points of this space, a situation which here is only refined by the fact that every morphism in the groupoid thus gives a homotopy/gauge transformation between the two smooth functions that are constant on the two endpoints of this morphism.

A particular case of this of special importance is this:

Example

For GG a Lie group, we write (BG) •∈Grpd(Smooth0Type)(\mathbf{B}G)_\bullet \in Grpd(Smooth0Type) for the Lie groupoid which

(BG) •=(G⟶⟵i⟶*) (\mathbf{B}G)_\bullet = \left( G \stackrel{\longrightarrow}{\stackrel{\overset{i}{\longleftarrow}}{\longrightarrow}} \ast \right)

whose composition is the product operation in the group (the groupoidal delooping of GG). The pre-smooth groupoid that this corresponds to under the embedding of example has groupoids of probes of the form

ℝ n↦B(C ∞(ℝ n,G) disc) \mathbb{R}^n \mapsto B \left(C^\infty(\mathbb{R}^n,G)_{disc}\right)

where on the right we have the homotopy 1-type whose fundamental group is that of smooth GG-valued functions on ℝ n\mathbb{R}^n, under pointwise mulitplication.

More specifically, the following class of examples plays a special role in the theory, as the encode what it takes for a pre-smooth groupoid to be a genuinely smooth groupoid.

Example

For XX a smooth manifold, let {U i→X} i∈I\{U_i \to X\}_{i \in I} be an open cover of XX. Its Cech groupoid is the Lie groupoid (diffeological groupoid) C({U i}) •C(\{U_i\})_\bullet whose

manifold of objects is C({U i}) 0≔∐i∈IU iC(\{U_i\})_0 \coloneqq \underset{i \in I}{\coprod} U_i is the disjoint union of all the charts of the cover;
manifold of morphisms is C({U i}) 1≔∐i,j∈IU i×XU jC(\{U_i\})_1 \coloneqq \underset{i,j \in I}{\coprod}U_i \underset{X}{\times} U_j is the disjoint union of all intersections of charts.

and whose source, target and identity maps are the evident inclusions. There is then a unique composition operation.

So a global point *→C({U i}) •\ast \to C(\{U_i\})_\bullet may be thought of as a pair (x,i)(x,i) of a point in the manifold XX and a label ii of a chart U iU_i that contains it, and there is is precisely one morphism between two such global point (x,i)→(y,j)(x,i)\to (y,j) whenever x=yx = y in XX and both U iU_i as well as U jU_j contain xx, hence one morphism for each point in an intersection of two patches. Composition of morphism is just re-remembering which intersections they sit in, the schematic picture of the Cech groupoid is this this:

C({U i}) •={ (x,j) ↗ ↘ (x,i) ⟶ (x,k)}. C(\{U_i\})_\bullet = \left\{ \array{ && (x,j) \\ & \nearrow && \searrow \\ (x,i) && \longrightarrow && (x,k) } \right\} \,.

Under the embedding of example there is an evident morphism of pre-smooth groupoids

C({U i})⟶X C(\{U_i\}) \longrightarrow X

from this Cech groupoid of the cover to the manifold that is being covered. This morphism simply forgets the information of which chart or intersection of charts a point is regarded to be in, and just remembers it as a point of XX.

((x,i)→(x,j))↦(x→id xx). ((x,i) \to (x,j)) \mapsto (x \stackrel{id_x}{\to} x) \,.

Proposition

Let XX a smooth manifold, {U i→X}\{U_i \to X\} an open cover and C({U i})C(\{U_i\}) the corresponding Cech groupoid, def. . Let GG be a Lie group and BG\mathbf{B}G its groupoidal delooping according to example .

Then the hom-groupoid PreSmooth1Type(C({U i}),BG)PreSmooth1Type(C(\{U_i\}), \mathbf{B}G) of maps from C({U i})C(\{U_i\}) to BG\mathbf{B}G, def. , has

as objects the Cech cocycles of degree 1 on XX relative to the cover and with values in GG; i.e. collections of smooth functions

g ij:U i×XU j⟶G g_{ i j} \;\colon\; U_i \underset{X}{\times} U_j \longrightarrow G

satisfying on each triple intersection the cocycle condition

g ijg jk=g ik g_{i j} g_{j k} = g_{i k}
as morphisms the coboundaries between such cocycles {g ij}\{g_{i j}\} and {g˜ jk}\{\tilde g_{j k}\}, hence collections of smooth functions

h i:U i⟶G h_{i} \colon U_i \longrightarrow G

such that on each intersection of charts

g ijh j=h ig˜ ij. g_{i j} h_j = h_i \tilde g_{i j} \,.

In particular the connected components of the Hom-groupoid is hence the Cech cohomology itself:

π 0PreSmooth1Type(C({U i}),BG)≃H Cech 1(X,{U i};G). \pi_0 PreSmooth1Type(C(\{U_i\}), \mathbf{B}G) \simeq H^1_{Cech}(X,\{U_i\}; G) \,.

Proof

It is a matter of unwinding the definitions that the statement holds disregarding smooth structure everywhere, hence after evaluating the morphisms of pre-smooth groupoids on ℝ 0\mathbb{R}^0. That the component functions that one finds this way all have to be smooth then follows componentwise with the Yoneda lemma for presheaves of sets on CartSpCartSp (as discussed at geometry of physics – smooth sets).

Gluing

The “bootstrap”-definition of pre-smooth groupoids above works as intended, by prop. , it just needs to be restricted now to something a little less general. The issue is that while this definition consistently identifies a smooth-structure-to-be by what its possible smooth probes are, it does not enforce yet the consistent gluing of probes:

for XX a pre-smooth groupoid, then given a probe of it of the form σ:ℝ n→X\sigma \colon \mathbb{R}^n \to X and given a covering {U i→ℝ n}\{U_i \to \mathbb{R}^n\} of the probe space by other probe spaces U i≃ℝ nU_i \simeq \mathbb{R}^n, then it should be possible to reconstruct σ\sigma from knowing its restrictions σ |U i:U i→X\sigma_{|U_i}\colon U_i \to X to these charts, and the information of how these are identified by gauge transformations on double overlaps.

Such a system of local data and of gauge identification on double overlaps is just what maps out of the Cech groupoid C({U i})C(\{U_i\}), def. , encode. This is shown by prop. for the case that XX is of the form (BG) •(\mathbf{B}G)_\bullet, example , but this is already the archetypical case.

In other words, the condition that smooth probes of XX by coordinate charts ℝ n\mathbb{R}^n glue along covers {U i→ℝ n}\{U_i \to \mathbb{R}^n\} of these charts is the condition that the groupoid of smooth maps out of ℝ n\mathbb{R}^n itself

ℝ n⟶X \mathbb{R}^n \longrightarrow X

is equivalent to the groupoid of maps out of the Cech groupoid of any cover

C({U i})⟶X C(\{U_i\}) \longrightarrow X

and is so via the “restriction” map that takes the former and precomposes it with the canonical map C({U i})→ℝ nC(\{U_i\}) \to \mathbb{R}^n.

C({U i}) ⟶ X ↓ ↗ ℝ n. \array{ C(\{U_i\}) &\longrightarrow& X \\ \downarrow & \nearrow \\ \mathbb{R}^n } \,.

Definition

A pre-smooth groupoid X∈PreSmooth1TypeX \in PreSmooth1Type, def. , satisfies descent if for all n∈ℕn \in \mathbb{N} and for all differentiably good open covers {U i→ℝ n}\{U_i \to \mathbb{R}^n\} of the nn-dimensional abstract coordinate chart, the functor

X(ℝ n)≃PreSmooth1Type(ℝ n,X)⟶PreSmooth1Type(C({U i}),X) X(\mathbb{R}^n) \simeq PreSmooth1Type(\mathbb{R}^n, X) \longrightarrow PreSmooth1Type(C(\{U_i\}),X)

given by pre-composition with C({U i})→ℝ nC(\{U_i\}) \to \mathbb{R}^n, is an equivalence of groupoids.

Proposition

Let X∈PreSmooth0Type↪PreSmooth1TypeX \in PreSmooth0Type \hookrightarrow PreSmooth1Type be a pre-smooth groupoid which is really just a pre-smooth set, hence a presheaf on CartSpCartSp that takes values in groupoids with only identity morphisms

X:CartSp op⟶Set↪Grp. X \colon CartSp^{op} \longrightarrow Set \hookrightarrow Grp \,.

Then XX is a smooth groupoid in the sense of def. precisely if it is a smooth set, hence precisely if, as a presheaf, it satisfies the sheaf condition.

Example

In particular, for X∈SmoothMfd↪PreSmooth0Type↪PreSmooth1TypeX \in SmoothMfd \hookrightarrow PreSmooth0Type \hookrightarrow PreSmooth1Type a smooth manifold, it satisfies descent as a pre-smooth groupoid.

For example:

Proposition

For GG a Lie group, then the pre-smooth delooping groupoid (BG) •(\mathbf{B}G)_\bullet of example satisfies descent on CartSpCartSp, def. .

Proof

For {U i→ℝ n}\{U_i \to \mathbb{R}^n\} a differentiably good open cover, then by prop. the groupoid PreSmooth1Type(C({U i}),(BG) •)PreSmooth1Type(C(\{U_i\}),(\mathbf{B}G)_\bullet) is the groupoid of Cech 1-cocycles and coboundaries with coeffcients in GG on ℝ n\mathbb{R}^n relative to the cover. But this is equivalently the groupoid of GG-principal bundles on ℝ n\mathbb{R}^n. Now because the underlying topological space of ℝ n\mathbb{R}^n is contractible, all GG-principal bundles on it are equivalent to the trivial one. But this is evidently represented by the image of point under the map

B(C ∞(G) disc)≃PreSmooth1Type(ℝ n,(BG) •)⟶PreSmooth1Type(C({U i}),(BG) •). B (C^\infty(G)_{disc}) \simeq PreSmooth1Type(\mathbb{R}^n, (\mathbf{B}G)_\bullet) \longrightarrow PreSmooth1Type(C(\{U_i\}), (\mathbf{B}G)_\bullet) \,.

Therefore this is an essentially surjective functor of groupoids. Moreover, the automorphisms of the trivial GG-principal bundles are precisely the smooth GG-functions, hence this is also a fully faithful functor of groupoids. Accordingly it is an equivalence of groupoids.

The same argument however shows that on the larger site SmoothMfd this object does not satisfy descent. Put positively, this is the content of prop. . below.

Weak equivalences

While the morphisms of pre-smooth groupoids defined above correctly encode morphisms of smooth structures (by taking smooth probes compatibly to smooth probes), they are not sensitive enough yet to the required concept of equivalence. This is because smooth structure, being about existence of differentiation, is to be detected entirely locally, namely stalk-wise. If for instance XX is a smooth manifold, then its smooth structure is determined, around any of its points, by the smooth structure of an arbitrarily small open ball around that point.

Definition

For n∈ℕn \in \mathbb{N}, and 0<r<1∈ℝ0 \lt r \lt 1 \in \mathbb{R} let

ℝ n≃D r n↪ℝ n \mathbb{R}^n \simeq D^n_r \hookrightarrow \mathbb{R}^n

be the smooth function that regards the Cartesian space ℝ n\mathbb{R}^n as the standard nn-disk of radius rr around the origin in ℝ n\mathbb{R}^n.

For X∈PreSmooth1TypeX \in PreSmooth1Type we write

(D n) *X≔lim⟶ rX(D r n)∈Grpd (D^n)^* X \coloneqq {\underset{\longrightarrow_r}{\lim}} X(D^n_r) \in Grpd

for the colimit (in the 1-category of groupoids) of the restrictions of its groupoids of plots along the inclusion of these open balls – the nn-stalk of XX. This extends to a functor

(D n) *:PreSmooth1Type⟶Grpd (D^n)^* \colon PreSmooth1Type \longrightarrow Grpd

This means that objects in (D n) *X(D^n)^\ast X are equivalence classes of pairs (r,x r)(r,x_r) where 0<r<10 \lt r \lt 1 and where x r∈X(D r n)x_r \in X(D^n_r), with two such pairs being equivalent (r 1,x r 1)∼(r 2,x r 2)(r_1, x_{r_1})\sim (r_2, x_{r_2}) precisely if there is r 0<r 1,r 2r_0 \lt r_1,r_2 such that x r 1x_{r_1} becomes equal to x r 2x_{r_2} after restriction to D r 0 nD^n_{r_0}.

Definition

A morphism f:X⟶Yf \colon X \longrightarrow Y of pre-smooth groupoids, def. , is called a local weak equivalence if for every n∈ℕn \in \mathbb{N} the nn-stalk, def. , is an equivalence of groupoids

((D n) *f):(D n) *X⟶(D n) *Y. ((D^n)^* f) \colon (D^n)^* X \stackrel{}{\longrightarrow} (D^n)^* Y \,.

We write X⟶≃YX\stackrel{\simeq}{\longrightarrow} Y for local weak equivalences of pre-smooth groupoids. We will mostly just say weak equivalence for short.

Proposition

For XX a smooth manifold and {U i→X}\{U_i \to X\} an open cover for it, then the canonical morphism from the corresponding Cech groupoid to XX, def. , is a local weak equivalence in the sense of def. .

Proof

The nn-stalk of the smooth manifold XX regarded as a presheaf is the set of equivalence classes of maps from open pointed nn-disks into it, where two such are identified if they coincide on some small joint sub-disk of their domain. We may call this the set of germs of XX (but beware that this terminology is typically used for something a little bit more restrictive, namely for the case that nn is the dimension of XX and that all maps from the disks into XX are required to be embeddings).

On the other hand the nn-stalk of C({U i})C(\{U_i\}) is the groupoid whose set of objects is the set of germs, in this sense, of the disjoint union ∐iU i\underset{i}{\coprod} U_i, and whose set of morphisms is the set of germs of the disjoint union ∐i,jU i×XU j\underset{i,j}{\coprod} U_i \underset{X}{\times} U_j.

But now since the cover is by open subsets, it follows that for every (x,i,j)∈∐i,jU i×XU j(x,i,j) \in \underset{i,j}{\coprod} U_i \underset{X}{\times} U_j, then every germ of objects [g][g] around (x,i)(x,i) has a representative gg that factors through this double intersection charts: g:D r n→U i×XU j→∐iU ug \colon D^n_r \to U_i \underset{X}{\times} U_j \to \underset{i}{\coprod} U_u. And similarly for (x,j)(x,j).

This means that the groupoid of nn-stalks is a disjoint union of groupoids, one for each germ of XX, all whose components are groupoids in which there is a unique morphism between any two objects, which are copies of this germ regarded as sitting in one of the charts of the cover. This means that each of these connected components is equivalent to the point.

Now the canonical cover projection sends each of these connected components to the germ that it corresponds to. Hence this is a an equivalence of groupoids.

Hypercovers

Definition

A morphism p:Y⟶Xp \colon Y \longrightarrow X of pre-smooth groupoids is called a split hypercover if

YY is
1. degreewise a coproduct of Cartesian spaces;
2. such that the degenerate elements split off as a dijoint summand.
pp is a weak equivalence, def. .

Proposition

For XX a smooth manifold and {U i→X}\{U_i \to X\} an open cover, then the canonical projection C({U i})→XC(\{U_i\}) \to X from the corresponding Cech groupoid, def. , is a split hypercover precisely if the cover is differentiably good.

Proof

For every cover the map is a weak equivalence, by prop. .

For the Cech groupoid the condition of cofibrancy in def. means that every non-empty finite intersection of patches is diffeomorphic to a Cartesian space, hence to an open ball. This is precisely the definition of differentialbly good open cover.

The (2,1)(2,1)-category of smooth groupoids

The Grpd-enriched category of genuine smooth groupoids is that obtained from that of pre-smooth groupoids, def. by “universally turning local weak equivalences, def. , into actual homotopy equivalences”. This is stated formally by def. below, but for many applications in practice certain concrete presentations of what this means concretely are well sufficient, one of these we state below in prop. .

Proposition

Let X,A∈PreSmooth1TypeX,A \in PreSmooth1Type such that AA satisfies descent, def. . Let Y→XY \to X be a split hypercover of XX, def. .

Then there is an equivalence of groupoids

Smooth1Type(X,A)≃PreSmooth1Type(Y,A) Smooth1Type(X,A) \simeq PreSmooth1Type(Y,A)

between the hom-groupoid of smooth groupoids from XX to AA, and that of pre-smooth groupoids, def. , from YY to AA.

Such statements follow with model structures on simplicial presheaves after embedding the present situation in the more general context of smooth infinity-groupoids. See there for more.

Proposition

Let XX be a smooth manifold, regarded as a smooth 0-groupoid, and let GG be a Lie group, with smooth delooping groupoid BG\mathbf{B}G (example , remark ).

Then

Smooth1Type(X,BG)≃GBund(X) Smooth1Type(X,\mathbf{B}G) \simeq G Bund(X)

is equivalently the groupoid of GG-principal bundles on XX.

Proof

By prop. the object (BG) •(\mathbf{B}G)_\bullet satisfies descent on CartSp. Choose {U i→X}\{U_i \to X\} a differentiably good open cover. By prop. the correspoding Cech groupoid projection C({U i})→XC(\{U_i\}) \to X is a split hypercover, def. . Hence, by prop. , there is an equivalence of groupoids

Smooth1Type(X,BG)≃PreSmooth1Type(C({U i}),(BG) •). Smooth1Type(X,\mathbf{B}G) \simeq PreSmooth1Type(C(\{U_i\}), (\mathbf{B}G)_\bullet) \,.

The groupoid on the right is, by prop. , the groupoid of Cech 1-cocycles and coboundaries with values in GG relative to a good open cover. This is equivalently the groupoid of GG-principal bundles.

Properties

Embedding into smooth ∞\infty-groupoids

By generalizing here groupoids to general Kan complexes and equivalences of groupoids to homotopy equivalences of Kan complexes, one obtains smooth ∞-stacks or smooth ∞-groupoids, which we write

\;\;\; Smooth∞Grpd ≔Sh (∞,1)(CartSp)≃L lheFunc(CartSp op,KanCplx)\coloneqq Sh_{(\infty,1)}(CartSp) \simeq L_{lhe} Func(CartSp^{op}, KanCplx) .

We often write H≔\mathbf{H} \coloneqq Smooth∞Grpd for short.

By the (∞,1)-Yoneda lemma there is a sequence of faithful inclusions

\;\;\; SmoothMfd ↪\hookrightarrow SmoothGrpd ↪\hookrightarrow Smooth∞Grpd.

This induces a corresponding inclusion of simplicial objects and hence of groupoid objects

LieGrpd↪Grpd ∞(SmoothMfd)↪Grpd ∞(Smooth∞Grpd). LieGrpd \hookrightarrow Grpd_\infty(SmoothMfd) \hookrightarrow Grpd_\infty(Smooth\infty Grpd) \,.

For 𝒢 •∈Grpd ∞(H)\mathcal{G}_\bullet \in Grpd_\infty(\mathbf{H}) a groupoid object we write

𝒢 0→𝒢≔lim⟶ n𝒢 n \mathcal{G}_0 \to \mathcal{G} \coloneqq \underset{\longrightarrow}{\lim}_{n} \mathcal{G}_n

for its (∞,1)-colimiting cocone, hence 𝒢∈H\mathcal{G} \in \mathbf{H} (without subscript decoration) denotes the realization of 𝒢 •\mathcal{G}_\bullet as a single object in H\mathbf{H}.

By the Giraud-Rezk-Lurie axioms of the (∞,1)-topos H\mathbf{H} this morphism 𝒢 0→𝒢\mathcal{G}_0 \to \mathcal{G} is a 1-epimorphism – an atlas – and its construction establishes is an equivalence of (∞,1)-categories Grpd ∞(H)≃H 1epi Δ 1Grpd_\infty(\mathbf{H}) \simeq \mathbf{H}^{\Delta^1}_{1epi}, hence morphisms 𝒢 •→𝒦 •\mathcal{G}_\bullet \to \mathcal{K}_\bullet in Grpd ∞(H)Grpd_\infty(\mathbf{H}) are equivalently diagrams in H\mathbf{H} of the form

𝒢 0 → 𝒦 0 ↓ ⇙ ↓ 𝒢 → 𝒦. \array{ \mathcal{G}_0 &\to& \mathcal{K}_0 \\ \downarrow &\swArrow& \downarrow \\ \mathcal{G} &\to& \mathcal{K} } \,.

This is evidently more constrained that just morphisms

𝒢→𝒦 \mathcal{G} \to \mathcal{K}

by themselves. The latter are the “generalized” or Morita morphisms between the groupoid objects 𝒢 •\mathcal{G}_\bullet, 𝒦 •\mathcal{K}_\bullet. These can be modeled as 𝒢 •\mathcal{G}_\bullet-𝒦 •\mathcal{K}_\bullet-bibundles.

Examples

Zoo of concepts subsumed by smooth groupoids

Proposition

The canonical identification yields a full subcategory

Smooth0Type↪Smooth1Type. Smooth0Type \hookrightarrow Smooth1Type \,.

Every Lie groupoid presents a smooth groupoids. Those of this form are also called differentiable stacks.

A 0-truncated smooth groupoid is equivalently a smooth space.

For GG a smooth group, its delooping BG\mathbf{B}G is a smooth groupoid, the moduli stack of smooth GG-principal bundles.

Action groupoids

Example

For XX a smooth space and GG a smooth group and

ρ:X×G→X\rho : X \times G \to X

an action then the action groupoid

X//G∈Smooth1Type X // G \in Smooth1Type

X//G=(X×G⟶p 1⟶ρX) X // G = \left( X \times G \stackrel{\overset{\rho}{\longrightarrow}}{\underset{p_1}{\longrightarrow}} X \right)

is a smooth groupoid.

Moduli spaces of gauge fields

Groupoid of electromagnetic field configurations

The mathematical concept of smooth groupoid is well familiar, in slight disguise, in the physics of basic gauge theory. Here we make this explicit for basic electromagnetism. For more exposition and details along these lines see (Eggersson 14).

We have seen in The electromagnetic field strength that a configuration of the electromagnetic field on ℝ 4\mathbb{R}^4 is given by a differential 1-form A∈Ω 1(ℝ 4)A \in \Omega^1(\mathbb{R}^4), the “electromagnetic potential”. It describes a field configuration with field strength Lorentz tensor (with respect to the canonical coordinates (t,x 1,x 2,x 3)≔(x i) i=1 4=(t, x^1, x^2, x^3) \coloneqq (x^i)_{i = 1}^4 = on ℝ 4\mathbb{R}^4)

F =dA =E 1dx 1∧dt+E 1dx 1∧dt+E 1dx 1∧dt+ +B 1dx 2∧dx 3+B 2dx 3∧dx 1+B 3dx 1∧dx 2 \begin{aligned} F & = \mathbf{d}A \\ & = E_1 \mathbf{d}x^1 \wedge \mathbf{d}t + E_1 \mathbf{d}x^1 \wedge \mathbf{d}t + E_1 \mathbf{d}x^1 \wedge \mathbf{d}t + \\ & + B_1 \mathbf{d}x^2 \wedge \mathbf{d}x^3 + B_2 \mathbf{d}x^3 \wedge \mathbf{d}x^1 + B_3 \mathbf{d}x^1 \wedge \mathbf{d}x^2 \end{aligned}

with electric field strength (E i∈C ∞(ℝ 4)) i=1 3(E_i \in C^\infty(\mathbb{R}^4))_{i = 1}^3 and magnetic field strength (B i∈C ∞(ℝ 4)) i=1 3(B_i \in C^\infty(\mathbb{R}^4))_{i = 1}^3 given that is given by the de Rham differential of AA:

F=dA. \begin{aligned} F = \mathbf{d}A \end{aligned} \,.

After Maxwell it was thought that F∈Ω 2(ℝ 4)F \in \Omega^2(\mathbb{R}^4) alone genuinely reflects the configuration of the electromagnetic field. But with the discovery of quantum mechanics it became clear that it is indeed the potential AA itself that reflects the configuration of the electromagnetic field: in the presence of magnetic flux or other topoligical constraints, there can be different A,A′A, A' with the same F=dA=dA′F = \mathbf{d}A = \mathbf{d}A' which nevertheless describe experimentally distinguishable electromagnetic field configurations. (Distinguishable by the Aharonov-Bohm effect and also to some extent by Dirac charge quantization; this is discussed at Circle-principal connections below.)

However, not all different gauge potentials describe different physics. The actual configuration space of electromagnetism on a spacetime XX is finer than Ω cl 2(X)\mathbf{\Omega}^2_{cl}(X) but coarser than Ω 1(X)\mathbf{\Omega}^1(X). And it is not quite a smooth space itself, but a smooth groupoid:

one finds that two electromagnetic potentials A,A′∈Ω 1(ℝ 4)A, A' \in \Omega^1(\mathbb{R}^4) for which there is a function λ∈C ∞(ℝ 4)\lambda \in C^\infty(\mathbb{R}^4) such that

A′=A+dλ A' = A + \mathbf{d}\lambda

represent different but equivalent field configurations. One says that λ\lambda induces a gauge transformation from AA to A′A'. We write λ:A→≃A′\lambda \colon A \stackrel{\simeq}{\to} A' to reflect this.

So the configuration space of electromagnetism does not just have points and coordinate systems. But it is also equipped with the information of a space of gauge transformations between any two coordinate systems laid out in it (which may be empty).

To see what the structure of such a smooth gauge groupoid should be, notice that the above defines an action of smooth functions λ\lambda on smooth 11-forms AA,

Definition

For any U∈U \in CartSp, Write Ω vert 1(X×U)\Omega^1_{vert}(X \times U) for the set of differential 1-forms on X×UX \times U with no components along UU, and write C ∞(X×U,U(1))C^\infty(X \times U , U(1)) for the group of circle group valued smooth functions. There is an action of this group on the 1-forms

ρ U:Ω vert 1(X×U)×C ∞(X×U,U(1))→Ω vert 1(X×U) \rho_U \colon \Omega^1_{vert}(X \times U) \times C^\infty(X \times U, U(1)) \to \Omega^1_{vert}(X \times U)

given by

ρ U:(A,λ)↦A+d Xλ. \rho_U \colon (A,\lambda) \mapsto A + \mathbf{d}_X \lambda \,.

Given such an action of a discrete group on a set, we might be demoted to form the quotient set Ω vert 1(X×U)/C ∞(X×U,U(1))\Omega^1_{vert}(X\times U)/C^\infty(X \times U, U(1)). This set contains the gauge equivalence classes of UU-parameterized collections of electromagnetic gauge fields on XX. But it turns out that this is too little information to correctly capture physics. For that instead we need to remember not just that two gauge fields are equivalent, but how they are equivalent. That is, we for λ\lambda a gauge transformation from A 1A_1 to A 2A_2, we should have an equivalence λ:A 1→≃A 2\lambda \colon A_1 \stackrel{\simeq}{\to} A_2.

Definition

The action groupoid

Ω vert 1(X×U)//C ∞(X×U,U(1))≔(Ω vert 1(X×U)×C ∞(X×U,U(1))⟶s=p 1⟶t=ρΩ vert 1(X×U)) \Omega^1_{vert}(X\times U)// C^\infty(X \times U, U(1)) \coloneqq ( \Omega^1_{vert}(X\times U) \times C^\infty(X \times U, U(1)) \stackrel{\overset{t = \rho}{\longrightarrow}}{\underset{s = p_1}{\longrightarrow}} \Omega^1_{vert}(X\times U) )

is the groupoid whose

objects are UU-parameterized collections of gauge potentials A∈Ω vert 1(X×U)A \in \Omega^1_{vert}(X \times U);
morphisms are pairs (A,λ)(A,\lambda) with AA an object and λ∈C ∞(X×U,U(1))\lambda \in C^\infty(X \times U, U(1)), with domain AA and codomain A+d XλA + \mathbf{d}_X \lambda;
composition is given by multiplication of the λ\lambda-labels in the group C ∞(X×U,U(1))C^\infty(X \times U, U(1)).

This is the discrete gauge groupoid for UU-parameterized collections of fields. It refines the gauge group, which is recoverd as its fundamental group:

Proposition

Let A 0≔0∈Ω vert 1(X×U)A_0 \coloneqq 0 \in \Omega^1_{vert}(X \times U) be the trivial gauge field. Then its automorphism group in the gauge groupoid of def. is the group of circle-group valued functions on UU:

π 1(Ω vert 1(X×U)//C ∞(X×U,U(1)),A 0)=Aut(A 0)≃C ∞(U,U(1)). \pi_1(\Omega^1_{vert}(X\times U)// C^\infty(X \times U, U(1)), A_0) = Aut(A_0) \simeq C^\infty(U,U(1)) \,.

Proof

By definition, an automorphism of A 0A_0 is given by a function λ∈C ∞(X×U,U(1))\lambda \in C^\infty(X \times U, U(1)) such that A 0=A 0+d XλA_0 = A_0 + \mathbf{d}_X \lambda. This is the case precisely if d Xλ=0\mathbf{d}_X \lambda = 0, hence if λ\lambda is contant along XX. But a function on X×UX \times U which is constant along XX is canonically identified with a function on just UU.

All this data in in fact natural in UU. Recall that Ω vert 1(X×U)=Ω 1(X)(U)\Omega^1_{vert}(X \times U) = \mathbf{\Omega}^1(X)(U) is the set of UU-charts of the smooth moduli space Ω 1(X)\mathbf{\Omega}^1(X) of smooth 1-forms on XX. Similarly C ∞(X×U)=C ∞(X)(U)C^\infty(X \times U) = \mathbf{C}^\infty(X)(U).

Proposition

There is a homomorphism of smooth spaces (def. )

ρ:Ω 1(X)×C ∞(X)→Ω 1(X) \rho \colon \mathbf{\Omega}^1(X)\times \mathbf{C}^\infty(X) \to \mathbf{\Omega}^1(X)

from the product smooth space, def. , of the smooth moduli spaces of 1-forms and 0-forms on XX, def. , to that of smooth functions, def. , whose component over U∈U \in CartSp is the action

ρ U:(A,λ)↦A+dλ \rho_U \colon (A,\lambda) \mapsto A + \mathbf{d}\lambda

of def. .

We then also want to consider a smooth action groupoid.

Definition

Write

Ω 1(X)//C ∞(X,U(1)):CartSp op→Grpd \mathbf{\Omega}^1(X) // \mathbf{C}^\infty(X, U(1)) : CartSp^{op} \to Grpd

for the contravariant functor from coordinate systems to the category of groupoids, which assigns to U∈CartSpU \in CartSp the discrete action groupoid of UU-collections of gauge fields of def. .

U↦Ω vert 1(X×U)//C ∞(X×U,U(1)). U \mapsto \Omega^1_{vert}(X \times U) // C^\infty(X\times U, U(1)) \,.

Such a structure presheaf of groupoids is a common joint generalization of the notion of discrete groupoids and smooth spaces. We call them smooth groupoids. This is what we turn to in Smooth groupoids

References

Discussion of diffeological groups (such as diffeomorphism groups and quantomorphism groups) goes back to

Jean-Marie Souriau, Groupes différentiels, in Differential Geometrical Methods in Mathematical Physics (Proc. Conf., Aix-en-Provence/Salamanca, 1979), Lecture Notes in Math. 836, Springer, Berlin, (1980), pp. 91–128. (MathScinet)

Exposition of the concept of smooth groupoids motivated from basic gauge theory is in

Discussion of smooth stacks as target spaces for sigma-model quantum field theories is in

Tony Pantev, Eric Sharpe, String compactifications on Calabi-Yau stacks, Nucl.Phys. B733 (2006) 233-296, (arXiv:hep-th/0502044)
Tony Pantev, Eric Sharpe, Gauged linear sigma-models for gerbes (and other toric stacks), (arXiv:hep-th/0502053)

S. Hellerman, Eric Sharpe, Sums over topological sectors and quantization of Fayet-Iliopoulos parameters, (arXiv:1012.5999)

Discussion of geometric Langlands duality in terms of 2d sigma-models on stacks (moduli stacks of Higgs bundles over a given algebraic curve) is in

Edward Witten, section 6 of Mirror Symmetry, Hitchin’s Equations, And Langlands Duality (arXiv:0802.0999)

Last revised on August 1, 2022 at 18:26:17. See the history of this page for a list of all contributions to it.