The integration problem
for principal connections

Abstract.

In this paper we introduce the Integration Problem for principal connections. Just as a principal connection on a principal bundle $\phi:Q\rightarrow M$ may be used to split $TQ$ into horizontal and vertical subbundles, a discrete connection may be used to split $Q\times Q$ into horizontal and vertical submanifolds. All discrete connections induce a connection on the same principal bundle via a process known as the Lie or derivative functor. The Integration Problem consists of describing, for a principal connection ${\mathcal{A}}$, the set of all discrete connections whose associated connection is ${\mathcal{A}}$. Our first result is that for flat principal connections, the Integration Problem has a unique solution among the flat discrete connections. More broadly, under a fairly mild condition on the structure group $G$ of the principal bundle $\phi$, we prove that the existence part of the Integration Problem has a solution that needs not be unique. Last, we see that, when $G$ is abelian, given compatible continuous and discrete curvatures the Integration Problem has a unique solution constrained by those curvatures.

1. Introduction

Connections on bundles are very useful tools both in the applications of bundles to different geometric problems —including some areas of Physics— and, also, in the study of the geometry of the bundle itself. Let $\phi:Q\rightarrow M$ be a principal $G$-bundle. Roughly speaking, a principal connection is a way of providing the tangent spaces $T_{q}Q$ with subspaces that are complementary to (the tangent space to) the orbit of $G$ through $q$ in $Q$. Because of the applicability of this notion, a principal connection has been characterized in different ways, as we review later.

A discrete connection on $\phi$ is also a way of providing a complement to the $G$-orbits, but this time in $Q\times Q$ —instead of $TQ$— in some appropriate sense. Again, discrete connections can be characterized in different ways, to be recalled later in Section 2.

Principal connections on a principal $G$-bundle $\phi:Q\rightarrow M$ appear naturally in the study of symmetries of some continuous-time dynamical systems defined on $Q$, where some of the standard constructions require splitting the velocity vectors $v_{q}\in T_{q}Q$ into horizontal and vertical parts (see, for example, [CMR01]). Discrete connections were introduced in this context as a tool used to perform the same kind of construction for some discrete-time dynamical systems, where the role of $TQ$ is played by $Q\times Q$ —in a naive sense, velocities are replaced by pairs of nearby points (see [LMW05], [MMdDM06], and [FZ13]). Interestingly, discrete connections have also been considered in the context of Synthetic Differential Geometry as a kind of combinatorial analogue (on a first neighborhood of the diagonal of a manifold) of principal connections (see [Koc07], for instance).

It has always been known (Section 5.2 in [LMW05]) that there is a “derivation” process that can be used in a principal bundle $\phi$ to produce a principal connection ${\mathcal{A}}$ out of a discrete connection ${\mathcal{A}_{d}}$, both on $\phi$. Loosely speaking, given a discrete connection ${\mathcal{A}_{d}}$, the associated principal connection ${\mathcal{A}}$ is “its derivative”, while any discrete connection ${\mathcal{A}_{d}}$ whose derivative is ${\mathcal{A}}$ is called “an integral” of ${\mathcal{A}}$. Besides this construction, not much is known about the relationship between the two types of connection, other than the fact that both have very similar properties.

The purpose of this paper is to introduce and study aspects of the following questions, that we collectively call the Integration Problem.

Problem.

What can we say about the mapping from discrete connections to continuous connections on a principal bundle? More specifically:

1. Is it surjective? That is, given a connection ${\mathcal{A}}$ on a principal bundle, is there any discrete connection ${\mathcal{A}_{d}}$ on that bundle that maps into ${\mathcal{A}}$?

2. Is it injective? Maybe better, what is the structure of the set of discrete connections ${\mathcal{A}_{d}}$ that map to the same continuous connection ${\mathcal{A}}$?

3. If this mapping is not injective, are there any additional constraints that could be imposed in order to make it so?

In Section 2 we review three equivalent characterizations of connections and of discrete connections in principal bundles: using forms, horizontal liftings and splittings of certain sequences. We also study how a derivation functor relates the discrete to the continuous notions.

We focus our attention in Section 3 on a type of connection that has specially nice properties: the flat connections, both continuous and discrete. It is well known that flat principal connections on $\phi:Q\rightarrow M$ are equivalent to morphisms of Lie algebroids $TM\rightarrow TQ/G$ (see [Mac87]). It has been recently seen that flat discrete connections are equivalent to morphisms of local Lie groupoids $M\times M\rightarrow(Q\times Q)/G$ (see [FJZ22]). Then, the Integration Problem becomes an instance of Lie Theory for Lie groupoids and algebroids, especially Lie’s Second Theorem. Using a result of [CMS18], we easily prove in Theorem 3.18 that the Integration Problem has an essentially unique solution, meaning that, given a flat principal connection ${\mathcal{A}}$, there is an essentially unique flat discrete connection ${\mathcal{A}_{d}}$ (unique as a germ) that integrates ${\mathcal{A}}$.

Leaving the context of flat connections, in Section 4 we prove (Corollary 4.9) that arbitrary principal connections can be integrated provided that the principal bundle is equipped with a smooth retraction (see Chapter 4 of [AMS08]). Then we prove that such retractions do exist if, for instance, the structure group of the principal bundle is the Cartesian product of a compact Lie group and a vector space (Corollary 4.13). Thus, we have established that, under appropriate conditions, the existence part of the Integration Problem has a solution. Moreover, as we show in an example, this solution needs not be unique.

Last, in Section 5, we consider the special case where the structure group $G$ of the principal bundle is abelian. In this setting we see that the Integration Problem for arbitrary curvatures has a solution if and only if the curvatures of the principal and the discrete connections are related in a specific way. Then we prove that, when that relation occurs, the Integration Problem for connections whose curvatures satisfy that relationship admits an essentially unique solution. This result is compatible with the result from Section 3 where the two curvatures are chosen to be flat, which, indeed, satisfies the said relation. Therefore, we proved the existence and uniqueness of the solution to the Integration Problem when the structure group of the bundle is abelian.

The general Integration Problem remains open. The partial solutions that we provide in this paper might serve as indications that this is a rich and nuanced topic, worthy of further study.

Finally, we wish to thank A. Cabrera and I. Mărcuţ for discussing with us some of their results in [CMS18] as well as for their interest in our work.

Notation

Throughout the paper, if the Lie group $G$ acts on the manifold $X$ on the left we denote the action by $l^{X}$ and the lifted action on $TX$ by $l^{TX}$, so that, if $g\in G$ and $\delta x\in T_{x}X$, we have $l^{TX}_{g}(\delta x)\coloneqq T_{x}l_{g}^{Q}(\delta x)$. In this context, $\pi^{X,G}:X\rightarrow X/G$ is the quotient map and $\tau_{X}:TX\rightarrow X$ is the canonical projection.

2. Connections and discrete connections

In this section we recall some of the characterizations and basic properties of connections and discrete connections on a principal bundle.

2.1. Connections on principal bundles

As we mentioned in the Introduction, connections on principal bundles may be described in several equivalent ways. Next we review three such approaches that will suffice for the purpose of this paper. For more details, see [KN96] and [Mac87].

Let $\phi:Q\rightarrow M$ be a smooth principal (left) $G$-bundle and $\mathfrak{g}\coloneqq Lie(G)$; we denote the corresponding $G$-action on $Q$ by $l^{Q}$. A connection on $\phi$ is determined by a $\mathfrak{g}$-valued $1$-form ${\mathcal{A}}$ on $TQ$ satisfying ${\mathcal{A}}(\xi_{Q}(q))=\xi$ for all $q\in Q$ and $\xi\in\mathfrak{g}$, where

$$\xi_{Q}(q)\coloneqq\frac{d}{dt}\bigg{|}_{t=0}l^{Q}_{\exp(t\xi)}(q),$$

and ${\mathcal{A}}(l^{TQ}_{g}(v))=\operatorname{Ad}_{g}({\mathcal{A}}(v))$ for all $v\in TQ$ and $g\in G$, where $\operatorname{Ad}$ is the adjoint action of $G$ on $\mathfrak{g}$. Such ${\mathcal{A}}$ is called a connection $1$-form. The set of all connection $1$-forms on $\phi$ will be denoted by $\Sigma_{C}$.

Example 2.1.

Given a Lie group $G$ and a manifold $M$ we define $Q\coloneqq M\times G$. The induced left $G$-action on $Q$, defined by $l_{g}^{Q}(m,g^{\prime})\coloneqq(m,gg^{\prime})$ for $m\in M$ and $g,g^{\prime}\in G$, makes the projection onto the first component $p_{1}:Q\to M$ a principal $G$-bundle.

Given a connection form ${\mathcal{A}}$ on $p_{1}$ we may define the $1$-form $\omega\in\Omega^{1}(M,\mathfrak{g})$, which is often called the local expression of ${\mathcal{A}}$, by precomposition with the differential of the trivial section of $p_{1}$:

$$\omega_{m}(\delta m)\coloneqq{\mathcal{A}}_{(m,e)}(\delta m\oplus 0)\quad\text{for }m\in M\text{ and }\delta m\in T_{m}M.$$

Conversely, each form $\omega\in\Omega^{1}(M,\mathfrak{g})$ gives a connection ${\mathcal{A}}$ on $p_{1}$ defined by ${\mathcal{A}}(\delta m+\delta g)\coloneqq\omega(\delta m)+Tr^{G}_{g^{-1}}(\delta g)$ for $\delta m\in T_{m}M$ and $\delta g\in T_{g}G$, where $r^{G}_{h}$ denotes the right multiplication on $G$ by $h\in G$. These two constructions are reciprocal inverses.

Alternatively, a connection on $\phi$ may be described using a smooth morphism of vector bundles ${h}:\phi^{*}TM\rightarrow TQ$ that is $G$-equivariant for the $G$-actions $l^{TQ}$ and $l^{Q\times TM}_{g}(q,\delta m)\coloneqq(l^{Q}_{g}(q),\delta m)$, and $(\tau_{Q},T\phi)\circ{h}=id_{\phi^{*}TM}$. Such ${h}$ is called a horizontal lift. The set of all horizontal lifts ${h}$ on $\phi$ is denoted by $\Sigma_{H}$.

Given ${\mathcal{A}}\in\Sigma_{C}$, we define ${h}:\phi^{*}TM\rightarrow TQ$ by

(2.1) $${h}(q,\delta m)\coloneqq\delta q-({\mathcal{A}}(\delta q))_{Q}(q),$$

where $\delta q\in T_{q}Q$ is such that $T_{q}\phi(\delta q)=\delta m$. The following result is fairly straightforward to check.

Proposition 2.2.

The map $F_{CH}:\Sigma_{C}\rightarrow\Sigma_{H}$ where $F_{CH}({\mathcal{A}})\coloneqq{h}$ with ${h}$ given by (2.1) is well defined. Furthermore, $F_{CH}$ is a bijection.

Yet another description of connections on $\phi$ may be obtained by considering the splittings of the following exact sequence in the category of vector bundles over $M$, $Vb_{M}$, the subcategory of the category of vector bundles $Vb$ whose objects have base $M$ and morphisms over $id_{M}$. Consider the sequence

(2.2) $$\lx@xy@svg{\hbox{\raise 0.0pt\hbox{\kern 10.50002pt\hbox{\ignorespaces\ignorespaces\ignorespaces\hbox{\vtop{\kern 0.0pt\offinterlineskip\halign{\entry@#!@&&\entry@@#!@\cr&&&&\crcr}}}\ignorespaces{\hbox{\kern-10.50002pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{\{0\}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 34.50002pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 34.50002pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{\widetilde{\mathfrak{g}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 51.97017pt\raise 6.1111pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.75pt\hbox{$\scriptstyle{\phi_{1}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 70.05559pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 70.05559pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{(TQ)/G}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 122.19301pt\raise 6.1111pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.75pt\hbox{$\scriptstyle{\phi_{2}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 140.27843pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 140.27843pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{TM}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 188.30272pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 188.30272pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{\{0\}}}$}}}}}}}\ignorespaces}}}}\ignorespaces, \textstyle{{\{0\}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{{\widetilde{\mathfrak{g}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\phi_{1}} \textstyle{{(TQ)/G}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\phi_{2}} \textstyle{{TM}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{{\{0\}}} \textstyle{{\{0\}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{{\widetilde{\mathfrak{g}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\phi_{1}} \textstyle{{(TQ)/G}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\phi_{2}} \textstyle{{TM}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{{\{0\}}}$$

where $\widetilde{\mathfrak{g}}$ is the adjoint bundle $(Q\times\mathfrak{g})/G$ with $G$ acting by $l^{Q}$ and by $\operatorname{Ad}$; $\phi_{1}([q,\xi])\coloneqq[\xi_{Q}(q)]$ and $\phi_{2}([v_{q}])\coloneqq T_{q}\phi(v_{q})$. The sequence (2.2) is exact (Proposition 3.2.3 of [Mac05]) and is known as the Atiyah Sequence associated to the principal bundle $\phi$. It is well known that right splittings of (2.2) also determine principal connections on $\phi$. The set of all right splittings of (2.2) is denoted by $\Sigma_{R}$.

Given ${h}\in\Sigma_{H}$, we define $s:TM\rightarrow(TQ)/G$ by

(2.3) $$s(\delta m)\coloneqq[{h}(q,\delta m)]$$

for any $\delta m\in T_{\phi(q)}M$. Then, we have the following result.

Proposition 2.3.

The map $F_{HR}:\Sigma_{H}\rightarrow\Sigma_{R}$ where $F_{HR}({h})\coloneqq s$ with $s$ given by (2.3) is well defined. Furthermore, $F_{HR}$ is a bijection.

We define the map $F_{CR}:\Sigma_{C}\rightarrow\Sigma_{R}$ as $F_{CR}\coloneqq F_{HR}\circ F_{CH}$. Naturally, $F_{CR}$ is a bijection. All together, we have the following commutative diagram in the category of sets

2.2. Discrete connections on principal bundles

Roughly speaking, discrete connections on the principal $G$-bundle $\phi:Q\rightarrow M$ are similar to connections but substituting $TQ$ by $Q\times Q$ and $\mathfrak{g}$ by $G$. They also require an additional piece of information: while connections are globally defined, discrete connections may only exist semi-locally, as we see next. For more details, see [FZ13] and [FJZ22].

An open subset $\mathcal{U}\subset Q\times Q$ is said to be of $D$-type if it is $G\times G$-invariant and contains the diagonal $\Delta_{Q}$. If $\mathcal{U}$ is of $D$-type, it is said to be symmetric if $(q,q^{\prime})\in\mathcal{U}\iff(q^{\prime},q)\in\mathcal{U}$. When $\mathcal{U}$ is of $D$-type, we define $\mathcal{U}^{\prime\prime}\coloneqq(\phi\times\phi)(\mathcal{U})\subset M\times M$ and $\mathcal{U}^{\prime}\coloneqq(id_{Q}\times\phi)({\mathcal{U}})\subset Q\times M$. Both $\mathcal{U}^{\prime}$ and $\mathcal{U}^{\prime\prime}$ are open subsets.

One way of characterizing a discrete connection on $\phi$ is via a discrete connection form, as follows. Let $\mathcal{U}$ be of $D$-type; a smooth function ${\mathcal{A}_{d}}:\mathcal{U}\rightarrow G$ is a discrete connection form if ${\mathcal{A}_{d}}(q,q)=e$ for all $q\in Q$ and ${\mathcal{A}_{d}}(l^{Q}_{g}(q),l^{Q}_{g^{\prime}}(q^{\prime}))=g^{\prime}{\mathcal{A}_{d}}(q,q^{\prime})g^{-1}$ for all $(q,q^{\prime})\in\mathcal{U}$ and $g,g^{\prime}\in G$. The set of all discrete connection forms on $\phi$ with domain $\mathcal{U}$ is denoted by $\Sigma_{C}^{d}(\mathcal{U})$.

Example 2.4.

Let $Q=M\times G$ and $p_{1}:Q\to M$ be the trivial bundle of Example 2.1. Let $\mathcal{U}^{\prime\prime}\subset M\times M$ be an open subset containing the diagonal $\Delta_{M}$ and $C:\mathcal{U}^{\prime\prime}\to G$ be a smooth function such that $C(m,m)=e$ for all $m\in M$. The $G$-valued function ${\mathcal{A}_{d}}$ with domain $(p_{1}\times p_{1})^{-1}(\mathcal{U}^{\prime\prime})$ defined by

(2.4) $${\mathcal{A}_{d}}\left((m_{0},g_{0}),(m_{1},g_{1})\right)\coloneqq g_{1}C(m_{0},m_{1})g_{0}^{-1}$$

is a discrete connection form on $p_{1}$, as one can readily check. Conversely, any connection form ${\mathcal{A}_{d}}$ defined on a $D$-type open subset $\mathcal{U}$ of $Q\times Q$ gives rise to a smooth function $C:(p_{1}\times p_{1})(\mathcal{U})\to G$ such that $C(m,m)=e$ for any $m\in M$ and (2.4) is valid.

An alternative way of characterizing a discrete connection on $\phi$ with domain $\mathcal{U}$ is using a discrete horizontal lift. A smooth map ${h_{d}}:\mathcal{U}^{\prime}\rightarrow Q\times Q$ is a discrete horizontal lift if ${h_{d}}(l^{Q}_{g}(q),m)=l^{Q\times Q}_{g}({h_{d}}(q,m))$ for all $(q,m)\in\mathcal{U}^{\prime}$ and $g\in G$, $(id_{Q}\times\phi)\circ{h_{d}}=id_{\mathcal{U}^{\prime}}$, and ${h_{d}}(q,\phi(q))=(q,q)$ for all $q\in Q$. The set of all discrete horizontal lifts on $\phi$ with domain $\mathcal{U}^{\prime}$ is denoted by $\Sigma_{H}^{d}(\mathcal{U})$. It is convenient to consider the second component of a discrete horizontal lift: $\overline{h_{d}}:Q\times M\rightarrow Q$ given by $\overline{h_{d}}\coloneqq p_{2}\circ{h_{d}}$.

Given ${\mathcal{A}_{d}}\in\Sigma_{C}^{d}(\mathcal{U})$ for $\mathcal{U}$ of $D$-type, we define ${h_{d}}:\mathcal{U}^{\prime}\rightarrow Q\times Q$ by

(2.5) $${h_{d}}(q,m)\coloneqq(q,l^{Q}_{{\mathcal{A}_{d}}(q,q^{\prime})^{-1}}(q^{\prime}))\quad\text{ for any }\quad q^{\prime}\in\phi^{-1}(m).$$

Proposition 2.5.

The map $F_{CH}^{d}:\Sigma_{C}^{d}(\mathcal{U})\rightarrow\Sigma_{H}^{d}(\mathcal{U})$ where $F_{CH}^{d}({\mathcal{A}_{d}})\coloneqq{h_{d}}$ with ${h_{d}}$ given by (2.5) is well defined. Furthermore, $F_{CH}^{d}$ is a bijection.

Proof.

It can be derived from Theorems 3.6 and 4.6 of [FZ13]. ∎

Just as in the continuous case, yet another description of discrete connections on $\phi$ is possible, in terms of right splittings of a certain sequence. Instead of working in the $Vb$ category as in the continuous case, here we consider the category ${Fbs}$ whose objects are smooth fiber bundles with a given global section and whose morphisms are smooth maps that commute with the bundle projections and map the given sections one to the other. ${Fbs}_{M}$ is the subcategory of ${Fbs}$ whose objects are fiber bundles over the manifold $M$ and whose morphisms are smooth maps over $id_{M}$.

Consider the sequence

(2.6)

where $\widetilde{G}$ is the conjugate bundle $(Q\times G)/G$ for $G$ acting on $Q$ by $l^{Q}$ and on itself by conjugation. All three spaces can be seen as fiber bundles over $M$ via the maps induced by the projection onto the first component. Each fiber bundle is equipped with the global section:

(2.7) $$\sigma_{\widetilde{G}}(m)\coloneqq[(q,e)],\quad\sigma_{(Q\times Q)/G}(m)\coloneqq[(q,q)],\quad\sigma_{M\times M}(m)\coloneqq(m,m),$$

where $q\in\phi^{-1}(m)$ is arbitrary. The maps $F_{1}$ and $F_{2}$ are defined by

(2.8) $$F_{1}([q,g])\coloneqq[(q,l^{Q}_{g}(q))]\quad\text{ and }\quad F_{2}([q,q^{\prime}])\coloneqq(\phi(q),\phi(q^{\prime})).$$

The sequence (2.6) is known as the Discrete Atiyah Sequence. It is easy to check that it is a sequence in the ${Fbs}_{M}$ category.

Let $(E,\sigma),(E^{\prime},\sigma^{\prime})\in{Fbs}_{M}$ and $\mathcal{V}\subset E^{\prime}$ be an open subset containing $\sigma^{\prime}(M)$; a map $H:\mathcal{V}\rightarrow E$ is a semi-local morphism if $\phi\circ H=\phi^{\prime}|_{\mathcal{V}}$ and $\sigma=H\circ\sigma^{\prime}$. Let $(E^{\prime\prime},\sigma^{\prime\prime})\xrightarrow{f_{1}}(E,\sigma)\xrightarrow{f_{2}}(E^{\prime},\sigma^{\prime})$ be a sequence in the ${Fbs}_{M}$ category. A semi-local morphism $H:\mathcal{V}\rightarrow E$ is a semi-local right splitting of the sequence if $f_{2}\circ H=id_{\mathcal{V}}$. If $\mathcal{U}$ is of $D$-type, the set of all semi-local right splittings of the Discrete Atiyah Sequence with domain $\mathcal{U}^{\prime\prime}$ is denoted by $\Sigma_{R}^{d}(\mathcal{U})$.

Let $\mathcal{U}$ be of $D$-type and ${h_{d}}\in\Sigma_{H}^{d}(\mathcal{U})$. We define $s_{d}:\mathcal{U}^{\prime\prime}\rightarrow(Q\times Q)/G$ by

(2.9) $$s_{d}(m,m^{\prime})\coloneqq[{h_{d}}(q,m^{\prime})]\quad\text{ for any }\quad q\in\phi^{-1}(m).$$

Proposition 2.6.

The map $F_{HR}^{d}:\Sigma_{H}^{d}(\mathcal{U})\rightarrow\Sigma_{R}^{d}(\mathcal{U})$ where $F_{HR}^{d}({h_{d}})\coloneqq s_{d}$ with $s_{d}$ given by (2.9) is well defined. Furthermore, $F_{HR}^{d}$ is a bijection.

Proof.

That $F_{HR}^{d}$ is well defined is a matter of routine checking. The second assertion is Proposition 3.15 in [FJZ22]. ∎

Last, if $\mathcal{U}$ is of $D$-type, we define $F_{CR}^{d}:\Sigma_{C}^{d}(\mathcal{U})\rightarrow\Sigma_{R}^{d}(\mathcal{U})$ as $F_{CR}^{d}\coloneqq F_{HR}^{d}\circ F_{CH}^{d}$ and, naturally, $F_{CR}^{d}$ is a bijection. We have the following commutative diagram in the category of sets

2.3. The derivation functor

It is well known that discrete connection forms on $\phi$ induce connection forms on $\phi$ (see Section 5.2 in [LMW05]): if ${\mathcal{A}_{d}}\in\Sigma_{C}^{d}(\mathcal{U})$, then ${\mathcal{A}}:TQ\rightarrow\mathfrak{g}$ defined by

(2.10) $${\mathcal{A}}(v_{q})\coloneqq D_{2}{\mathcal{A}_{d}}(q,q)(v_{q})$$

—where, as usual, $D_{2}$ is the differential restricted to the second component of the Cartesian product— is in $\Sigma_{C}$.

Proposition 2.7.

The map $F_{C}:\Sigma_{C}^{d}(\mathcal{U})\rightarrow\Sigma_{C}$ with $F_{C}({\mathcal{A}_{d}})\coloneqq{\mathcal{A}}$ for ${\mathcal{A}}$ given by (2.10) is well defined.

Example 2.8.

Let $Q\coloneqq M\times G$ and $p_{1}:Q\to M$ be the trivial bundle of Example 2.1. Let ${\mathcal{A}_{d}}$ be a discrete connection form on $p_{1}$ and ${\mathcal{A}}\coloneqq F_{C}({\mathcal{A}_{d}})$. If $N:M\times M\to G$ and $\eta\in\Omega^{1}(M,\mathfrak{g})$ are their associated local expressions from Examples 2.4 and 2.1 then

$$\eta(\delta m)=D_{2}N(m,m)(\delta m)$$

whenever $m\in M$ and $\delta m\in T_{m}M$.

Inspired by this idea and the Lie functor used in the Lie groupoid – Lie algebroid context (see [Mac05], p. 125) we introduce next a functor $V:{Fbs}_{M}\rightarrow Vb_{M}$ that will be used to assign to each “discrete object” a corresponding “continuous object”.

Let $(E,\psi,M,F)$ be a fiber bundle and $\sigma\in\Gamma(E)$, the set of all smooth global sections of $E$. As $\psi$ is a submersion, $\ker(T\psi)\subset TE$ is a vector bundle over $E$. Then, $V(E)\coloneqq\sigma^{*}\ker(T\psi)$ is a vector bundle over $M$ with rank $\dim(F)$. In fact, $p_{1}^{r}:V(E)\rightarrow M$ —the map induced by $p_{1}:M\times TE\rightarrow M$— is isomorphic to $\psi\circ\tau_{E}|_{\ker(T\psi)}:\ker(T\psi)|_{\sigma(M)}\rightarrow M$ as vector bundles, via the morphism $(m,\delta e)\mapsto\delta e$. Now, for any $f\in\hom_{{Fbs}_{M}}((E,\sigma),(E^{\prime},\sigma^{\prime}))$, we have that $Tf(\ker(T\psi))\subset\ker(T\psi^{\prime})$, so that $Tf(\ker(T\psi)|_{\sigma(M)})\subset\ker(T\psi^{\prime})|_{\sigma^{\prime}(M)}$. Thus, $Tf$ induces an element of $\hom_{Vb_{M}}(V(E),V(E^{\prime}))$ that we denote by $V(f)$. It is easy to check that $V$ is a well defined functor from ${Fbs}_{M}$ into $Vb_{M}$.

Notice that the definition of $V(f)$ only requires that $f$ be defined in an open neighborhood of $\sigma(M)$ in $E$, so that we can take $V(f)$ for any semi-local morphism $f$.

Example 2.9.

Let $f:X\rightarrow Y$ be a smooth map. Consider the fiber bundle $\psi\coloneqq p_{1}:X\times Y\rightarrow X$ with the section $\sigma_{f}(x)\coloneqq(x,f(x))$. Then, as $T_{(x,y)}\psi(\delta x,\delta y)=\delta x$, we have $\ker(T\psi)_{(x,y)}=\{0\}\oplus T_{y}Y$, so that

$$V((X\times Y,\sigma_{f}))=\sigma_{f}^{*}(\ker(T\psi))=\{(x,(0,\delta y)):\delta y\in T_{f(x)}Y\text{ and }x\in X\}.$$

As $(x,(0,\delta y))\mapsto(x,\delta y)$ defines an isomorphism in $Vb_{X}$, we have that

$$V((X\times Y,\sigma_{f}))\simeq f^{*}TY\quad\text{ in }\quad Vb_{X}.$$

If, in particular, $\phi:Q\rightarrow M$ is a principal $G$-bundle and choose $f\coloneqq\phi$, then $\sigma_{Q\times M}(q)\coloneqq\sigma_{f}(q)=(q,\phi(q))$ and

(2.11) $$V((Q\times M),\sigma_{(Q\times M)})\simeq\phi^{*}TM\quad\text{ in }\quad Vb_{Q}.$$

Example 2.10.

Here we consider two special cases of Example 2.9. If $X$ is any manifold and $f=id_{X}$, then $\sigma_{(X\times X)}(x)\coloneqq\sigma_{f}(x)=(x,x)$ and

(2.12) $$V((X\times X),\sigma_{(X\times X)})\simeq TX\quad\text{ in }\quad Vb_{X}.$$

If $Y\coloneqq G$ is a Lie group and $f:X\rightarrow G$ is the constant map $f(x)=e$, the identity element of $G$, then $\sigma_{X\times G}(x)\coloneqq\sigma_{f}(x)=(x,e)$ and

(2.13) $$V((X\times G),\sigma_{(X\times G)})\simeq X\times\mathfrak{g}\quad\text{ in }\quad Vb_{X},$$

where $\mathfrak{g}\coloneqq Lie(G)$.

Example 2.11.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle and $\mathcal{U}\subset Q\times Q$ be of $D$-type. Given ${\mathcal{A}_{d}}\in\Sigma_{C}^{d}(\mathcal{U})$. Define $\widetilde{{\mathcal{A}_{d}}}:Q\times Q\rightarrow Q\times G$ by $\widetilde{{\mathcal{A}_{d}}}(q,q^{\prime})\coloneqq(q,{\mathcal{A}_{d}}(q,q^{\prime}))$. As $\widetilde{{\mathcal{A}_{d}}}\in\hom_{{Fbs}_{Q}}(Q\times Q,Q\times G)$, using the characterizations provided by Example 2.10, we have $V(\widetilde{{\mathcal{A}_{d}}}):TQ\rightarrow\underline{\mathfrak{g}}$ (where $\underline{\mathfrak{g}}$ is the trivial bundle over $Q$ with fiber $\mathfrak{g}$) and, for $\delta q\in T_{q}Q$,

$$T_{(q,q)}\widetilde{{\mathcal{A}_{d}}}(0,\delta q)=(T_{(q,q)}p_{1}(0,\delta q),T_{(q,q)}{\mathcal{A}_{d}}(0,\delta q))=(0,D_{2}{\mathcal{A}_{d}}(q,q)(\delta q)),$$

and we conclude that

(2.14) $$V(\widetilde{{\mathcal{A}_{d}}})(\delta q)=D_{2}{\mathcal{A}_{d}}(q,q)(\delta q).$$

Comparing (2.14) with (2.10) we see that the assignment ${\mathcal{A}_{d}}\mapsto{\mathcal{A}}$ given by (2.10) can be seen as coming from the functor $V$. In the same spirit, we look at discrete horizontal lifts.

Example 2.12.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle and $\mathcal{U}\subset Q\times Q$ be of $D$-type. Given ${h_{d}}\in\Sigma_{H}^{d}(\mathcal{U})$, it follows immediately that ${h_{d}}\in\hom_{{Fbs}_{Q}}((Q\times M,\sigma_{Q\times M}),(Q\times Q,\sigma_{Q\times Q}))$. As, for $(q,\delta m)\in\phi^{*}(TM)$,

$$T_{(q,\phi(q))}{h_{d}}(0,\delta m)=D_{2}{h_{d}}(q,\phi(q))(\delta m),$$

the characterizations given in Examples 2.9 and 2.10 lead to

(2.15) $$V({h_{d}})(q,\delta m)=D_{2}\overline{h_{d}}(q,\phi(q))(\delta m).$$

In fact, there is the following result relating discrete and continuous horizontal lifts.

Proposition 2.13.

The map $F_{H}:\Sigma_{H}^{d}(\mathcal{U})\rightarrow\Sigma_{H}$ with $F_{H}({h_{d}})\coloneqq V({h_{d}})$ (computed in (2.15)) is well defined.

Last, we apply the $V$ functor to right splittings of (2.6) in an attempt to find right splittings of (2.2).

Example 2.14.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle. Consider the fiber bundle $\psi\coloneqq\savestack{\tmpbox}{\stretchto{\scaleto{\scalerel*[\widthof{\phi\circ p_{1}}]{\kern-0.6pt\bigwedge\kern-0.6pt}{\rule[-505.89pt]{4.30554pt}{505.89pt}}}{}}{0.5ex}}\stackon[1pt]{\phi\circ p_{1}}{\scalebox{-1.0}{\tmpbox}}:(Q\times Q)/G\rightarrow M$ with the section $\sigma_{(Q\times Q)/G}$ defined in (2.7). Then, as

$$\begin{split}T_{[(q,q^{\prime})]}\psi(T_{(q,q^{\prime})}\pi^{Q\times Q,G}(\delta q,\delta q^{\prime}))=&T_{(q,q^{\prime})}(\psi\circ\pi^{Q\times Q,G})(\delta q,\delta q^{\prime}))\\ =&T_{(q,q^{\prime})}(\phi\circ p_{1})(\delta q,\delta q^{\prime}))=T_{q}\phi(\delta q),\end{split}$$

we have that

$$\begin{split}\ker(T\psi)_{[(q,q^{\prime})]}=&\{T_{(q_{0},q_{0}^{\prime})}\pi^{Q\times Q,G}(\delta q_{0},\delta q_{0}^{\prime}):\delta q_{0}\in{\mathcal{V}}^{Q}_{q_{0}},\delta q_{0}^{\prime}\in T_{q_{0}^{\prime}}Q,[(q,q^{\prime})]=[(q_{0},q_{0}^{\prime})]\}\\ =&\{T_{(q_{0},q_{0}^{\prime})}\pi^{Q\times Q,G}(0,\delta q_{0}^{\prime\prime}):\delta q_{0}^{\prime\prime}\in T_{q_{0}^{\prime}}Q,[(q,q^{\prime})]=[(q_{0},q_{0}^{\prime})]\}.\end{split}$$

Consequently,

$$\begin{split}V((Q\times Q)/G)=&\{(m,T_{(q,q)}\pi^{Q\times Q,G}(0,\delta q^{\prime\prime})):\phi(q)=m,\delta q^{\prime\prime}\in T_{q}Q\}.\end{split}$$

In addition, mapping $(m,T_{(q,q)}\pi^{Q\times Q,G}(0,\delta q^{\prime\prime}))\mapsto\pi^{TQ,G}(\delta q^{\prime\prime})$, defines an isomorphism in $Vb_{M}$, so that

(2.16) $$V((Q\times Q)/G)\simeq(TQ)/G\quad\text{ in }\quad Vb_{M}.$$

Example 2.15.

Let $F_{2}$ be the map defined in (2.6) and (2.8). As $F_{2}$ is a morphism in the ${Fbs}_{M}$ category, we may compute $V(F_{2})$. Recalling the characterizations of $V(M\times M)$ and $V((Q\times Q)/G)$ given in the Examples 2.10 and 2.14, we have that, for any $\delta q^{\prime\prime}\in T_{q}Q$,

$$\begin{split}T_{[(q,q)]}F_{2}(T_{(q,q)}\pi^{Q\times Q,G}(0,\delta q^{\prime\prime}))=&T_{(q,q)}(F_{2}\circ\pi^{Q\times Q,G})(0,\delta q^{\prime\prime})\\ =&T_{(q,q)}(\phi\times\phi)(0,\delta q^{\prime\prime})=(0,T_{q}\phi(\delta q^{\prime\prime})).\end{split}$$

Taking into account (2.12) and (2.16), this leads to

$$V(F_{2})([\delta q^{\prime\prime}])=T_{q}\phi(\delta q^{\prime\prime}).$$

In other words,

$$V(F_{2})=\phi_{2}$$

for the map $\phi_{2}$ appearing in (2.2).

Remark 2.16.

It can also be seen that $V(\widetilde{G})=\widetilde{\mathfrak{g}}$ and $V(F_{1})=\phi_{1}$, so that the $V$ functor maps the Discrete Atiyah Sequence (2.8) into the Atiyah Sequence (2.2).

Proposition 2.17.

The map $F_{R}:\Sigma_{R}^{d}(\mathcal{U})\rightarrow\Sigma_{R}$ defined by $F_{R}(s_{d})\coloneqq V(s_{d})$ is well defined.

Proof.

We need to check that, for $s_{d}\in\Sigma_{R}^{d}(\mathcal{U})$, $s\coloneqq V(s_{d})$ is a right splitting of (2.2). Using the characterizations given by Examples 2.10, 2.14 and 2.15, we have that $s\in\hom_{Vb_{M}}(TM,(TQ)/G)$ and

$$\phi_{2}\circ s=V(F_{2})\circ V(s_{d})=V(F_{2}\circ s_{d})=V(id_{\mathcal{U}^{\prime\prime}})=id_{TM},$$

concluding the argument. ∎

We can summarize the different sets and maps that we have constructed so far via the following diagram in the category of sets,

(2.17) $$\begin{split}\lx@xy@svg{\hbox{\raise 0.0pt\hbox{\kern 16.4515pt\hbox{\ignorespaces\ignorespaces\ignorespaces\hbox{\vtop{\kern 0.0pt\offinterlineskip\halign{\entry@#!@&&\entry@@#!@\cr&&\\&&\\&&\\&&\crcr}}}\ignorespaces{\hbox{\kern-16.4515pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{\Sigma_{C}^{d}(\mathcal{U})}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 47.89162pt\raise 6.76944pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-2.40277pt\hbox{$\scriptstyle{F_{CH}^{d}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 97.24683pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 13.15602pt\raise-27.40831pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-2.40277pt\hbox{$\scriptstyle{F_{CR}^{d}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 40.64326pt\raise-29.49998pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces\ignorespaces\ignorespaces{}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern-14.6186pt\raise-60.01828pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.70833pt\hbox{$\scriptstyle{F_{C}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 0.0pt\raise-110.20326pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces{\hbox{\lx@xy@drawline@}}{\hbox{\kern 53.84917pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{}}$}}}}}}}{\hbox{\kern 97.24683pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{\Sigma_{H}^{d}(\mathcal{U})}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces{}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 82.91006pt\raise-27.40831pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-2.40277pt\hbox{$\scriptstyle{F_{HR}^{d}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 73.17009pt\raise-29.49998pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces\ignorespaces\ignorespaces{}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 114.05183pt\raise-60.01828pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.70833pt\hbox{$\scriptstyle{F_{H}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 114.05183pt\raise-110.20326pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces{\hbox{\lx@xy@drawline@}}{\hbox{\kern-3.0pt\raise-41.27774pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{}}$}}}}}}}{\hbox{\kern 40.4515pt\raise-41.27774pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{\Sigma_{R}^{d}(\mathcal{U})}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces{}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 56.84917pt\raise-61.28383pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.70833pt\hbox{$\scriptstyle{F_{R}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 56.84917pt\raise-71.45662pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces{\hbox{\lx@xy@drawline@}}{\hbox{\kern 111.05183pt\raise-41.27774pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{}}$}}}}}}}{\hbox{\kern-3.0pt\raise-81.28993pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{}}$}}}}}}}{\hbox{\kern 48.09042pt\raise-81.28993pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{\Sigma_{R}}}$}}}}}}}{\hbox{\kern 111.05183pt\raise-81.28993pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{}}$}}}}}}}{\hbox{\kern-8.8126pt\raise-120.03658pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{\Sigma_{C}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 47.89162pt\raise-126.11157pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.70833pt\hbox{$\scriptstyle{F_{CH}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 104.88573pt\raise-120.03658pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 13.55328pt\raise-94.58826pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.70833pt\hbox{$\scriptstyle{F_{CR}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 49.64232pt\raise-86.20326pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces{\hbox{\lx@xy@drawline@}}{\hbox{\kern 53.84917pt\raise-120.03658pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{}}$}}}}}}}{\hbox{\kern 104.88573pt\raise-120.03658pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{{\Sigma_{H}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces{}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 82.51924pt\raise-94.58826pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.70833pt\hbox{$\scriptstyle{F_{HR}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 64.0992pt\raise-86.20326pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}\ignorespaces\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces{\hbox{\lx@xy@drawline@}}\ignorespaces}}}}\ignorespaces\end{split} \textstyle{{\Sigma_{C}^{d}(\mathcal{U})}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F_{CH}^{d}} \scriptstyle{F_{CR}^{d}} \scriptstyle{F_{C}} \textstyle{{\Sigma_{H}^{d}(\mathcal{U})}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F_{HR}^{d}} \scriptstyle{F_{H}} \textstyle{{\Sigma_{R}^{d}(\mathcal{U})}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F_{R}} \textstyle{{\Sigma_{R}}} \textstyle{{\Sigma_{C}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F_{CH}} \scriptstyle{F_{CR}} \textstyle{{\Sigma_{H}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F_{HR}} \textstyle{{\Sigma_{C}^{d}(\mathcal{U})}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F_{CH}^{d}} \scriptstyle{F_{CR}^{d}} \scriptstyle{F_{C}} \textstyle{{\Sigma_{H}^{d}(\mathcal{U})}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F_{HR}^{d}} \scriptstyle{F_{H}} \textstyle{{\Sigma_{R}^{d}(\mathcal{U})}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F_{R}} \textstyle{{\Sigma_{R}}} \textstyle{{\Sigma_{C}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F_{CH}} \scriptstyle{F_{CR}} \textstyle{{\Sigma_{H}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{F_{HR}}$$

where the full arrows are bijections. Careful unraveling of the definitions leads to the following result.

Proposition 2.18.

Diagram (2.17) in the $Set$ category is commutative.

3. The flat case

In this section we review how flat connections and flat discrete connections may be viewed as splittings of sequences in the categories of Lie algebroids and local Lie groupoids. Then, using an integration result for morphisms in the category of Lie algebroids, we prove that the Integration Problem in the flat case admits a unique solution.

3.1. Lie groupoids and Lie algebroids

We start by reviewing the basic notions of local Lie groupoid and Lie algebroid, mostly to fix the notation. There are several slightly different notions of local Lie groupoid. We essentially follow the one introduced in p.40 of [CDW87], that is also used in [FJZ22].

Definition 3.1.

A local Lie groupoid consists of smooth manifolds $G$ and $M$ together with submersions $\alpha,\beta:G\rightarrow M$ as well as a diffeomorphism $i:G\rightarrow G$, and smooth maps $\epsilon:M\rightarrow G$ and $m:G_{m}\rightarrow G$, where $G_{m}\subset G_{2}\coloneqq G\tensor[_{{\beta}}]{\times}{{}_{{\alpha}}}G$ is an open subset, all subject to the conditions stated below. It is convenient to write $g_{1}g_{2}\coloneqq m(g_{1},g_{2})$ and $g^{-1}\coloneqq i(g)$.

1. $\alpha\circ\epsilon=id_{M}=\beta\circ\epsilon$.

2. For all $(g_{1},g_{2})\in G_{m}$, we have that $(g_{2}^{-1},g_{1}^{-1})\in G_{m}$ and $g_{2}^{-1}g_{1}^{-1}=(g_{1}g_{2})^{-1}$.

3. For all $g\in G$ we have that $(\epsilon(\alpha(g)),g),(g,\epsilon(\beta(g)))\in G_{m}$ and $\epsilon(\alpha(g))g=g=g\epsilon(\beta(g))$.

4. For all $g\in G$ we have that $(g,g^{-1}),(g^{-1},g)\in G_{m}$ and $gg^{-1}=\epsilon(\alpha(g))$ and $g^{-1}g=\epsilon(\beta(g))$.

5. If $(g_{1},g_{2}),(g_{2},g_{3}),(g_{1},g_{2}g_{3})\in G_{m}$, then $(g_{1}g_{2},g_{3})\in G_{m}$ and $(g_{1}g_{2})g_{3}=g_{1}(g_{2}g_{3})$.

We denote a local Lie groupoid as above by $G\rightrightarrows M$. If $G_{m}=G_{2}$ the local Lie groupoid is a Lie groupoid.

Definition 3.2.

Let $G\rightrightarrows M$ and $G^{\prime}\rightrightarrows M^{\prime}$ be local Lie groupoids. A smooth map $F:G\rightarrow G^{\prime}$ is a morphism of local Lie groupoids if

1. $F(\epsilon_{G}(M))\subset\epsilon_{G^{\prime}}(M^{\prime})$, and

2. for all $(g_{1},g_{2})\in G_{m}$, we have that $(F(g_{1}),F(g_{2}))\in G^{\prime}_{m}$ and $F(g_{1}g_{2})=F(g_{1})F(g_{2})$.

The local Lie groupoids together with their morphisms form a category that we denote by $lLGpd$. In what follows, we will be mostly interested in the subcategory $lLGpd_{M}$ of local Lie groupoids over a smooth manifold $M$ and whose morphisms cover $id_{M}$.

Example 3.3.

The Discrete Atiyah Sequence (2.6) is a sequence in the $lLGpd_{M}$ category. In fact, it is a sequence of Lie groupoids. Indeed, $M\times M$ is what is known as the pair groupoid and the structure of $(Q\times Q)/G$ is induced by that of the pair groupoid $Q\times Q$. The structure maps of $\widetilde{G}$ are $\alpha=\beta=$ the fiber bundle projection on $M$, $m((q,g_{0}),(q,g_{1}))\coloneqq(q,g_{0}g_{1})$, $\epsilon(m)\coloneqq\sigma_{\widetilde{G}}(m)$ for $\sigma_{\widetilde{G}}$ defined in (2.7) and $i(q,g)\coloneqq(q,g^{-1})$. For more details, see Section 5.1 in [FJZ22].

Example 3.4.

Let $\phi:Q\rightarrow Q/G$ be a principal $G$-bundle and $\mathcal{U}\subset Q\times Q$ be of symmetric $D$-type. Then we have the restriction of the Discrete Atiyah Sequence (2.6) to $\mathcal{U}$,

(3.1)

where the maps $F_{1}$ and $F_{2}$ are the appropriate restriction and co-restrictions of the maps defined in (2.8). Then, (3.1) is a sequence in the $lLGpd_{M}$ category. For more details, see Section 5.2 in [FJZ22].

Definition 3.5.

A Lie algebroid on the smooth manifold $M$ is triple $(A,[,],\rho)$ where $A\in Vb_{M}$, $[,]$ is a Lie bracket on $\Gamma(A)$ and $\rho\in\hom_{Vb_{M}}(A,TM)$ is the anchor map, satisfying

$$[X,fY]=(\rho(X)f)\,Y+f[X,Y],\quad\text{ for all }\quad X,Y\in\Gamma(A)\text{ and }f\in C^{\infty}(M).$$

Definition 3.6.

If $(A_{1},[,]_{1},\rho_{1})$ and $(A_{2},[,]_{2},\rho_{2})$ are two Lie algebroids over $M$, a morphism of Lie algebroids is a map $F\in\hom_{Vb_{M}}(A_{1},A_{2})$ such that

1. $F([X,Y]_{1})=[F(X),F(Y)]_{2}$ for every $X,Y\in\Gamma(A_{1})$, and

2. $\rho_{2}\circ F=\rho_{1}$.

The Lie algebroids over a smooth manifold $M$ together with their morphisms form a category that we denote by $LAlgbd_{M}$.

Example 3.7.

The Atiyah Sequence (2.2) is a sequence in the $LAlgbd_{M}$ category. For the details, see Section 3.2 in [Mac05].

The classical Lie functor assigns all Lie groups and morphisms, the corresponding Lie algebras and morphisms, via a derivation process. This functor has been extended to the categories of (local) Lie groupoids and Lie algebroids (see [Mac05], p. 125). Because of this construction it is customary to say that the morphism of Lie algebroids obtained via the Lie functor from a morphism of Lie groupoids is “its derivative” and, conversely, call the latter “an integral” of the former. Notice that, for (transitive) Lie groupoids, forgetting the brackets and anchor map, the Lie functor becomes the $V$ functor introduced in Section 2.3. Also, it can be checked that applying the Lie functor from $lLGpd_{M}$ into $LAlgbd_{M}$ to the Discrete Atiyah Sequence (2.6) of a principal bundle gives its Atiyah Sequence (2.2).

The following existence and uniqueness result of A. Cabrera et al. is part of a version of Lie’s Second Theorem in the context of Lie groupoids. It is the key to solving the Integration Problem for flat connections.

Theorem 3.8.

Let $G_{1}\rightrightarrows M$ and $G_{2}\rightrightarrows M$ be local Lie groupoids with Lie algebroids $A_{1}$ and $A_{2}$ respectively, and let $f:A_{1}\rightarrow A_{2}$ be a Lie algebroid morphism. There exists a local Lie groupoid morphism $F:G_{1}\rightarrow G_{2}$ integrating $f$, and any two such integrations coincide around the unit section.

Proof.

Theorem 2.4 on p. 7 of [CMS18]. ∎

Remark 3.9.

The notion of local Lie groupoid used in [CMS18] is slightly different from the one that we are using in this paper. Still, as the authors prove in Proposition 2.1 of [CMS18] the germ of any local Lie groupoid in their sense can be represented by a local Lie groupoid in our sense. Also, the meaning of $F:G_{1}\rightarrow G_{2}$ being a “local Lie groupoid morphism” in [CMS18] is that there exist open subsets $U\subset G_{1}$ and $W\subset(G_{1})_{2}$ such that $\epsilon_{1}(M)\subset U$ and $\epsilon_{1}(M)_{2}\coloneqq(\epsilon_{1}(M))^{2}\cap(G_{1})_{2}\subset W$, that $F:U\rightarrow G_{2}$ is smooth, that $F(\epsilon_{1}(M))\subset\epsilon_{2}(M)$ and that for all $(g,g^{\prime})\in W$, $F(gg^{\prime})=F(g)F(g^{\prime})$.

Corollary 3.10.

Let $G_{1}\rightrightarrows M$ and $G_{2}\rightrightarrows M$ be Lie groupoids over $M$ with associated Lie algebroids $A_{1}\rightarrow M$ and $A_{2}\rightarrow M$. If $f:A_{1}\rightarrow A_{2}$ is a morphism of Lie algebroids, then, there are open subsets $U\subset G_{1}$ and $U_{m}\subset(G_{1})_{2}$ such that

1. $U$ with the structure maps induced by $G_{1}$ is a local Lie groupoid whose multiplication $m_{U}$ is defined on $U_{m}$.

2. There exists $F\in\hom_{lLGpd_{M}}(U,G_{2})$ such that $Lie(F)=f$.

Proof.

By Theorem 3.8, there are open subsets $V\subset G_{1}$ and $W\subset(G_{1})_{2}$ such that $\epsilon_{1}(M)\subset V$ and $\epsilon_{1}(M)_{2}\subset W$ and there exists a smooth map $F:V\rightarrow G_{2}$ such that $F(\epsilon_{1}(M))\subset\epsilon_{2}(M)$, and, for $(g,g^{\prime})\in W$, $F(gg^{\prime})=F(g)F(g^{\prime})$ (multiplications in $G_{1}$ and $G_{2}$). Furthermore, $Lie(F)=f$.

There is a claim in the proof of Proposition 2.1 in [CMS18] (but see Remark 3.19 later on) to the effect that the open subsets $U^{(2)}\coloneqq\left(U\times U\right)\cap(G_{1})_{2}$ for $U\subset G_{1}$ such that $\epsilon_{1}(M)\subset U$ form a basis of open neighborhoods of $\epsilon_{1}(M)_{2}\subset(G_{1})_{2}$. Using this claim applied to $W\subset(G_{1})_{2}$, we have that there is an open subset $V^{\prime}\subset G_{1}$ containing $\epsilon_{1}(M)$ and such that $\epsilon_{1}(M)_{2}\subset(V^{\prime})^{(2)}\subset W$.

Now let $U^{\prime}\coloneqq V\cap V^{\prime}$ and $U\coloneqq U^{\prime}\cap(U^{\prime})^{-1}$. Then, $U^{-1}\subset U$, so that it inherits a local Lie groupoid over $M$ structure from $G_{1}$ using Lemma 5.11 in [FJZ22]; notice that in this construction, the multiplication map $m_{U}$ is defined on $U_{m}\coloneqq U_{2}\cap m_{1}^{-1}(U)\subset U^{(2)}\subset W$, so that the map $F$ that integrates $f$ is, indeed, in $\hom_{lLGpd_{M}}(U,G_{2})$. ∎

3.2. Flat connections in the Lie context

Let ${\mathcal{A}}$ be a principal connection on the principal $G$-bundle $\phi:Q\rightarrow M$. If we let $s\coloneqq F_{CR}({\mathcal{A}})$ be the representation of the connection as a right splitting of the Atiyah Sequence (2.2), we see that, for any $X,Y\in\mathfrak{X}(M)$, $s([X,Y])-[s(X),s(Y)]\in\ker(\phi_{2})$. Due to the exactness of the Atiyah Sequence, there is a unique $b_{XY}\in\widetilde{\mathfrak{g}}$ such that $\phi_{1}(b_{XY})=s([X,Y])-[s(X),s(Y)]$.

Definition 3.11.

In the previous setting, the morphism of vector bundles over $M$, ${\mathcal{B}}:TM\oplus TM\rightarrow\widetilde{\mathfrak{g}}$ defined on vector fields by

$${\mathcal{B}}(X,Y)\coloneqq b_{XY}$$

is the curvature of ${\mathcal{A}}$. Also, ${\mathcal{A}}$ is said to be flat if ${\mathcal{B}}=0$.

Remark 3.12.

Definition 3.11 —Definition 5.3.8 in [Mac05]— is not the most common definition of this notion. Still, it is the most appropriate for our work. Furthermore, by Proposition 5.3.14 in [Mac05], it is equivalent to the standard one, as it appears in, for instance, [KN96].

Example 3.13.

Let $M$ be a manifold and $G$ be an abelian Lie group. Following Example 2.1, consider the principal $G$-bundle $p_{1}:Q\coloneqq M\times G\to M$. Given a connection form ${\mathcal{A}}$ on $p_{1}$ we define $\omega\in\Omega^{1}(M,\mathfrak{g})$ to be its local expression. It is straightforward to see that ${\mathcal{A}}$ is flat if and only if the differential form $\omega$ is closed.

Definition 3.11 states explicitly that the curvature of a principal connection ${\mathcal{A}}$ is the obstruction for its associated section $s$ to induce a morphism of Lie algebras between the corresponding spaces of sections. An immediate consequence is the following result.

Proposition 3.14.

Let ${\mathcal{A}}$ be a principal connection on the principal $G$-bundle $\phi:Q\rightarrow M$ and let $s\coloneqq F_{CR}({\mathcal{A}})$ be the representation of the connection as a right splitting of the Atiyah Sequence (2.2) in the $Vb_{M}$ category. Then $s$ is a morphism of Lie algebroids if and only if ${\mathcal{A}}$ is flat.

Proof.

By construction, $s\in\Sigma_{R}$ is a morphism of smooth vector bundles and it preserves the corresponding anchor maps. Then, as $\phi_{1}$ is injective, $s$ is a morphism of the Lie algebras of sections if and only if ${\mathcal{B}}$ vanishes. ∎

There is, also, a notion of curvature for discrete connections, as we recall next.

Definition 3.15.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle, $\mathcal{U}\subset Q\times Q$ be of $D$-type and ${\mathcal{A}_{d}}\in\Sigma_{C}^{d}(\mathcal{U})$. Let

(3.2) $$\mathcal{U}^{(3)}\coloneqq\{(q_{0},q_{1},q_{2})\in Q^{3}:(q_{i},q_{j})\in\mathcal{U}\text{ for all }0\leq i<j\leq 2\}.$$

We define the discrete curvature of ${\mathcal{A}_{d}}$ as ${\mathcal{B}_{d}}:\mathcal{U}^{(3)}\rightarrow G$ by

(3.3) $${\mathcal{B}_{d}}(q_{0},q_{1},q_{2})\coloneqq{\mathcal{A}_{d}}(q_{0},q_{2})^{-1}{\mathcal{A}_{d}}(q_{1},q_{2}){\mathcal{A}_{d}}(q_{0},q_{1}).$$

We say that ${\mathcal{A}_{d}}$ is flat if ${\mathcal{B}_{d}}=e$ on $\mathcal{U}^{(3)}$.

The following result shows that there is a bijection between the flat discrete connections and the right splittings of the Discrete Atiyah Sequence in the $lLGpd_{M}$ category.

Proposition 3.16.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle and $\mathcal{U}\subset Q\times Q$ be of symmetric $D$-type. Then, the bijection $F_{CR}^{d}:\Sigma_{C}^{d}(\mathcal{U})\rightarrow\Sigma_{R}^{d}(\mathcal{U})$ determines a bijection between the subset $\Sigma_{C}^{d,e}(\mathcal{U})\subset\Sigma_{C}^{d}(\mathcal{U})$ of flat discrete connections and the subset $\widetilde{\Sigma_{R}^{d}}(\mathcal{U})\subset\Sigma_{R}^{d}(\mathcal{U})$ of right splittings of extension (3.1) in the $lLGpd_{M}$ category.

Proof.

Proposition 5.30 in [FJZ22]. ∎

The derivation process preserves the flatness condition, as we see next.

Proposition 3.17.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle and ${\mathcal{A}_{d}}\in\Sigma_{C}^{d,e}(\mathcal{U})$ for a $D$-type open subset $\mathcal{U}\subset Q\times Q$. Then, $F_{C}({\mathcal{A}_{d}})\in\Sigma_{C}$ is a flat connection in $\phi$.

Proof.

As ${\mathcal{A}_{d}}\in\Sigma_{C}^{d,e}(\mathcal{U})$, by Proposition 3.16, $s_{d}\coloneqq F_{CR}^{d}({\mathcal{A}_{d}})\in\widetilde{\Sigma_{R}^{d}}(\mathcal{U})$. Then, is a diagram in the $lLGpd_{M}$ category and $F_{2}\circ s_{d}=id_{M\times M}$. Applying the Lie functor to the previous diagram and recalling Example 2.15 we obtain the diagram in the $LAlgbd_{M}$ category, where $s\coloneqq Lie(s_{d})=V(s_{d})$. Applying the same procedure to the identity, we obtain $\phi_{2}\circ s=id_{TM}$, proving that $s$ is a right splitting of (2.2) in the $LAlgbd_{M}$ category. Therefore by Proposition 3.14, ${\mathcal{A}}\coloneqq F_{RC}(s)$ is flat. Using Proposition 2.18 the statement follows because ${\mathcal{A}}=F_{RC}(s)=F_{RC}(V(s_{d}))=F_{RC}(F_{R}(s_{d}))=F_{RC}(F_{R}(F_{CR}^{d}({\mathcal{A}_{d}})))=F_{C}({\mathcal{A}_{d}})$. ∎

The next result puts together the interpretation of flat connections as morphisms of Lie groupoids and algebroids, turning the Integration Problem into one of integration of morphisms of Lie algebroids, that can be solved using Theorem 3.8.

Theorem 3.18.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle and ${\mathcal{A}}\in\Sigma_{C}$ that is flat. Then,

1. there is a subset $\mathcal{U}\subset Q\times Q$ of symmetric $D$-type and ${\mathcal{A}_{d}}\in\Sigma_{C}^{d}(\mathcal{U})$ flat that integrates ${\mathcal{A}}$, that is, ${\mathcal{A}}=F_{C}({\mathcal{A}_{d}})$.

2. If $\mathcal{U}_{1},\mathcal{U}_{2}$ are symmetric of $D$-type such that there are ${\mathcal{A}_{d}^{1}}\in\Sigma_{C}^{d}(\mathcal{U}_{1})$ and ${\mathcal{A}_{d}^{2}}\in\Sigma_{C}^{d}(\mathcal{U}_{2})$, both of which are flat and integrate ${\mathcal{A}}$, then there is a subset $\mathcal{U}\subset\mathcal{U}_{1}\cap\mathcal{U}_{2}$ of symmetric $D$-type set such that ${\mathcal{A}_{d}^{1}}$ and ${\mathcal{A}_{d}^{2}}$ agree on $\mathcal{U}$.

Proof.

Recall that $M\times M$ and $(Q\times Q)/G$ are Lie groupoids over $M$ whose corresponding Lie algebroids are $TM$ and $(TQ)/G$ (see Example 3.5.11 and Theorem 4.5.7 in [Mac05]).

Let $s\coloneqq F_{CR}({\mathcal{A}})\in\Sigma_{R}$; as ${\mathcal{A}}$ is flat, by Proposition 3.14, $s:TM\rightarrow(TQ)/G$ is a morphism of Lie algebroids over $M$. Then, by Corollary 3.10, there are open subsets $U\subset M\times M$ and $U_{m}\subset(M\times M)_{2}$ such that, with the structure induced by that of the Lie groupoid $M\times M$ and multiplication defined on $U_{m}$, $U$ is a local Lie groupoid over $M$ and there is $s_{d}\in\hom_{lLGpd_{M}}(U,(Q\times Q)/G)$ such that $Lie(s_{d})=s$, that is, $s_{d}$ integrates $s$.

Let $\mathcal{U}\coloneqq(\phi\times\phi)^{-1}(U)\subset Q\times Q$. It is easy to check that $\mathcal{U}$ is of symmetric $D$-type and $\mathcal{U}^{\prime\prime}=U$. Hence, $s_{d}\in\hom_{lLGpd_{M}}(U,(Q\times Q)/G)$ implies that ${\mathcal{A}_{d}}\coloneqq(F_{CR}^{d})^{-1}(s_{d})\in\Sigma_{C}^{d,e}(\mathcal{U})$. Following the arrows in diagram (2.17), we have that $F_{C}({\mathcal{A}_{d}})={\mathcal{A}}$, proving point (1) of the statement.

Given ${\mathcal{A}_{d}^{j}}\in\Sigma_{C}^{d}(\mathcal{U}_{j})$, $j=1,2$, both flat and integrating ${\mathcal{A}}$, we define $s_{d}^{j}\coloneqq F_{CR}^{d}({\mathcal{A}_{d}^{j}})\in\widetilde{\Sigma_{R}^{d}}(\mathcal{U}_{j})$ and $s\coloneqq F_{CR}({\mathcal{A}})\in\Sigma_{R}$; as ${\mathcal{A}}$ is flat, by Proposition 3.14, $s:TM\rightarrow(TQ)/G$ is a morphism of Lie algebroids over $M$ and $s=Lie(s_{d}^{j})$ for $j=1,2$. Then, by Theorem 3.8, there is an open subset $V\subset\mathcal{U}_{1}^{\prime\prime}\cap\mathcal{U}_{2}^{\prime\prime}\subset M\times M$ containing $\Delta_{M}$ such that $s_{d}^{1}|_{V}=s_{d}^{2}|_{V}$. Defining $V^{\prime}\coloneqq V\cap V^{-1}$ (the inverse in the pair groupoid structure of $M\times M$), we see that $\Delta_{M}\subset V^{\prime}\subset V$, so that $\mathcal{U}\coloneqq(\phi\times\phi)^{-1}(V^{\prime})$ is symmetric of $D$-type and $\mathcal{U}^{\prime\prime}=V^{\prime}$. Last, $s_{d}^{1}|_{V}=s_{d}^{2}|_{V}$ implies that ${\mathcal{A}_{d}^{1}}|_{\mathcal{U}}={\mathcal{A}_{d}^{2}}|_{\mathcal{U}}$, concluding the proof of point (2) in the statement. ∎

Remark 3.19.

Corollary 3.10, referenced in the proof of Theorem 3.18 above, relies on a claim made in the proof of Proposition 2.1 of [CMS18]. Unfortunately, the proof of that claim is only slightly hinted at in the paper. We provide an alternative, more detailed, explanation of the part of the claim that is needed in the proof of Corollary 3.10 in Proposition A.2.

4. Existence of solution in the general case

In this section we abandon the special case of flat connections and return to the arbitrary curvature case. We will be able to integrate arbitrary principal connections provided that we have the additional data of a smooth retraction. Then we state sufficient conditions on the structure group of the principal bundle under which these retractions do exist.

4.1. Smooth retractions

Following [AMS08, Definition 4.1.1] we introduce the following notion.

Definition 4.1.

Let $X$ be a smooth manifold and denote by $Z_{X}$ the image of its zero section in the tangent bundle. A smooth retraction on $X$ is a smooth map $R:\mathcal{Z}\rightarrow X$, where $\mathcal{Z}\subset TX$ is an open neighborhood of $Z_{X}$, such that, if we denote $R_{x}\coloneqq R|_{T_{x}X\cap\mathcal{Z}}$ for any $x\in X$, the following properties are valid.

1. $R_{x}(0_{x})=x$ for all $x\in X$, where $0_{x}$ is the null vector in $T_{x}X$.

2. Identifying canonically $T_{0_{x}}T_{x}X\simeq T_{x}X$, we have $T_{0_{x}}R_{x}=id_{T_{x}X}$.

Remark 4.2.

Condition (2) in Definition 4.1 is equivalent to requiring that

(4.1) $$\frac{d}{dt}\bigg{|}_{t=0}R_{x}(tv_{x})=v_{x}\in T_{x}X\quad\text{ for all }\quad v_{x}\in T_{x}X\text{ and }x\in X.$$

Example 4.3.

Let $(X,g)$ be a Riemannian manifold. Then the exponential map $\exp_{g}:\mathcal{Z}_{g}\rightarrow X$ is defined in an open neighborhood $\mathcal{Z}_{g}$ of $Z_{X}\subset TX$; notice that when $g$ is a complete metric $\mathcal{Z}_{g}=TX$. Then, $R\coloneqq\exp_{g}$ is a smooth retraction. Indeed, $\exp_{g}(0_{x})=x$ and, for any $v_{x}\in T_{x}X$,

$$\begin{split}\frac{d}{dt}\bigg{|}_{t=0}\exp_{g}|_{x}(tv_{x})=T_{0_{x}}(\exp_{g}|_{x})(v_{x})=v_{x}.\end{split}$$

The smoothness of $\exp_{g}$ and the computation of $T(\exp_{g}|_{x})$ are in Proposition 5.19 of [Lee18].

If $R:\mathcal{Z}\rightarrow X$ is a smooth retraction, we define its associated extended retraction $\widetilde{R}:\mathcal{Z}\rightarrow X\times X$ by $\widetilde{R}(v_{x})\coloneqq(x,R(v_{x}))$.

Before we tackle an important property of $\widetilde{R}$, we recall a well known fact about $T_{0_{x}}TX$ for $x\in X$. Fix $x\in X$ and define two vector subspaces $H_{x},V_{x}\subset T_{0_{x}}TX$ by

(4.2) $$H_{x}\coloneqq T_{0_{x}}Z_{X}\quad\text{ and }\quad V_{x}\coloneqq\left\{\frac{d}{dt}\bigg{|}_{t=0}t\,\delta x:\delta x\in T_{x}X\right\}=\ker(T_{0_{x}}\tau_{X}).$$

Notice that $T_{x}X\simeq H_{x}$ (via $\gamma^{\prime}(0)\in T_{x}X\mapsto\frac{d}{dt}\big{|}_{t=0}0_{\gamma(t)}\in H_{x}$) and $T_{x}X\simeq V_{x}$ (via $\delta x\in T_{x}X\mapsto\frac{d}{dt}\big{|}_{t=0}t\,\delta x$). It is easy to see that $T_{0_{x}}TX=H_{x}\oplus V_{x}$.

Now we prove that the extended retraction $\widetilde{R}$ is a diffeomorphism between open neighborhoods of $Z_{X}$ and $\Delta_{X}$ in $TX$ and $X\times X$ respectively.

Proposition 4.4.

$\widetilde{R}$ maps the zero section $Z_{X}\subset TX$ bijectively onto the diagonal $\Delta_{X}\subset X\times X$. Furthermore, there are open subsets $V_{R}\subset TX$ and $W_{R}\subset X\times X$ containing $Z_{X}$ and $\Delta_{X}$ respectively, such that $\widetilde{R}|_{V_{R}}^{W_{R}}:V_{R}\rightarrow W_{R}$ is a diffeomorphism.

Proof.

The first assertion is immediate from point (1) in Definition 4.1. The idea for the second assertion is as follows: it is easy to see by direct computation using the Inverse Function Theorem (or invoking Proposition 2.1 in [BLnMdD22] together with the fact that $\widetilde{R}$ is a discretization map as in Definition 2.2 of the same paper) that $\widetilde{R}$ is a local diffeomorphism in a neighborhood of $Z_{X}$. Then, the existence of the open subsets $V_{R}$ and $W_{R}$ follows using Theorem 1 in [CP07]. ∎

From (1) in Definition 4.1 we see that $(\widetilde{R}|_{V_{R}}^{W_{R}})^{-1}(x,x)=0_{x}$ for all $x\in X$. In what follows, whenever we write $\widetilde{R}^{-1}$, we mean $(\widetilde{R}|_{V_{R}}^{W_{R}})^{-1}$.

4.2. Discrete connections from retractions and connections

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle and ${\mathcal{A}}$ be a connection on $\phi$. Given a smooth retraction $R$ on $Q$, we want to see how to “discretize” ${\mathcal{A}}$ using $R$ in order to construct a discrete connection ${\mathcal{A}_{d}}$ on $\phi$ such that, ideally, it satisfies $F_{C}({\mathcal{A}_{d}})={\mathcal{A}}$. It is natural that, in order to do so, we may need to impose some additional conditions to $R$.

The first step is the construction of a smooth retraction on $M$ using $R$ and ${\mathcal{A}}$.

As we know, the horizontal distribution of ${\mathcal{A}}$, $Hor_{\mathcal{A}}\subset TQ$, is a $G$-invariant distribution on $Q$ that satisfies $TQ=\mathcal{V}\oplus Hor_{\mathcal{A}}$, where $\mathcal{V}\coloneqq\ker(T\phi)$ is the vertical bundle of $\phi$. In particular, it is a regular submanifold ($Hor_{\mathcal{A}}={\mathcal{A}}^{-1}(\{0\})$ and $0\in\mathfrak{g}$ is a regular value of ${\mathcal{A}}$).

Assume that $R$ is $G$-equivariant for the actions $l^{TQ}$ and $l^{Q}$. As $\phi\circ R$ is $G$-invariant, it induces a smooth map $R_{G}:(TQ)/G\rightarrow M$ and we have the following commutative diagram

where $i$ is the natural inclusion and $\bar{R}$ is the restriction of $R_{G}$ to $Hor_{\mathcal{A}}/G$.

The connection ${\mathcal{A}}$ allows us to define an isomorphism of vector bundles over $M$ (see Section 2.4 in [CMR01]):

$$\alpha_{\mathcal{A}}:(TQ)/G\rightarrow TM\oplus\widetilde{\mathfrak{g}}\quad\text{ by }\quad\alpha_{\mathcal{A}}([v_{q}]):=(T_{q}\phi(v_{q}),[(q,{\mathcal{A}}(v_{q}))]),$$

where $\tilde{}\mathfrak{g}\coloneqq(Q\times\mathfrak{g})/G$ is the adjoint bundle (where $G$ acts by the principal bundle action on $Q$ and the adjoint representation on $\mathfrak{g}$). Notice that $\alpha_{\mathcal{A}}(Hor_{\mathcal{A}}/G)=TM\oplus 0_{M}$. Using this new isomorphism we have the following commutative diagram in the category of manifolds:

where we have defined

(4.3) $$\check{R}:TM\rightarrow M\quad\text{ by }\quad\check{R}\coloneqq\bar{R}\circ(\alpha_{\mathcal{A}}|_{Hor_{\mathcal{A}}/G})^{-1}\circ i_{0}.$$

More explicitly, if $\delta m\in T_{m}M$ and $q\in Q|_{m}$,

(4.4) $$\check{R}(\delta m)\coloneqq\bar{R}((\alpha_{\mathcal{A}}|_{Hor_{\mathcal{A}}/G})^{-1}(\delta m,0))=R_{G}([{h}(q,\delta m)])=\phi(R({h}(q,\delta m))),$$

where ${h}=F_{CH}({\mathcal{A}})$ is the horizontal lift map associated to the connection ${\mathcal{A}}$.

Lemma 4.5.

If $R$ is a $G$-equivariant smooth retraction on $Q$, the map $\check{R}$ defined by (4.3) is a smooth retraction on $M$.

Proof.

Being a composition of smooth maps, $\check{R}$ is a smooth map. Also, for any $m\in M$, and $q\in Q|_{m}$,

$$\check{R}(0_{m})=\phi(R(({h}(q,0_{m}))))=\phi(R(0_{q}))=\phi(q)=m.$$

Last, for $\delta m\in T_{m}M$ and $q\in Q|_{m}$,

$$\begin{split}\frac{d}{dt}\bigg{|}_{t=0}\check{R}(t\,\delta m)=&\frac{d}{dt}\bigg{|}_{t=0}\phi(R({h}(q,t\,\delta m)))=\frac{d}{dt}\bigg{|}_{t=0}\phi(R(t\,{h}(q,\delta m)))\\ =&T_{q}\phi\bigg{(}\frac{d}{dt}\bigg{|}_{t=0}R(t\,{h}(q,\delta m))\bigg{)}=T_{q}\phi({h}(q,\delta m))=\delta m.\end{split}$$

∎

As before, let ${\mathcal{A}}\in\Sigma_{C}$ and ${h}\coloneqq F_{CH}({\mathcal{A}})\in\Sigma_{H}$. Define

(4.5) $${h_{d}}:\mathcal{U}^{\prime}\rightarrow Q\times Q\quad\text{ by }\quad{h_{d}}(q,m)\coloneqq\widetilde{R}\big{(}{h}(q,{(\widetilde{\check{R}})}^{-1}(\phi(q),m))\big{)},$$

for

(4.6) $$\mathcal{U}\coloneqq(\phi\times\phi)^{-1}(W_{\check{R}})\quad\text{ and }\quad\mathcal{U}^{\prime}\coloneqq(id_{Q}\times\phi)(\mathcal{U}),$$

where $W_{\check{R}}\subset M\times M$ is the open subset constructed by Proposition 4.4 for $\check{R}$.

Remark 4.6.

For ${h_{d}}$ to be well defined, we need to check that, for $(q,m)\in\mathcal{U}^{\prime}$, $(q,{\widetilde{\check{R}}}^{-1}(\phi(q),m))\in\phi^{*}TM$, that is, we have to check that $\phi(q)=\tau_{M}({\widetilde{\check{R}}}^{-1}(\phi(q),m))$. This is true because for $\delta m\in TM$, such that $\delta m={\widetilde{\check{R}}}^{-1}(\phi(q),m)$, we have $(\tau_{M}(\delta m),\check{R}(\delta m))=\widetilde{\check{R}}(\delta m)=(\phi(q),m)$, so that $\tau_{M}({\widetilde{\check{R}}}^{-1}(\phi(q),m))=\tau_{M}(\delta m)=\phi(q)$.

Proposition 4.7.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle, ${\mathcal{A}}$ be a connection on $\phi$ and $R$ be a $G$-equivariant smooth retraction on $Q$. Then,

1. the set $\mathcal{U}$ defined in (4.6) is of $D$-type, and

2. the map ${h_{d}}$ defined by (4.5) is in $\Sigma_{H}^{d}(\mathcal{U})$.

Proof.

By Proposition 4.4, $\Delta_{M}\subset W_{\check{R}}$ and $W_{\check{R}}\subset M\times M$ is open. Thus, by the continuity of $\phi$, so are $\mathcal{U}$ and $\mathcal{U}^{\prime}=(id_{Q}\times\phi)(\mathcal{U})=(\phi\times id_{M})^{-1}(W_{\check{R}})$. By the $G$-invariance of $\phi$, $\mathcal{U}$ is $G\times G$-invariant. As $\Delta_{M}\subset W_{\check{R}}$, we have $\Delta_{Q}\subset\mathcal{U}$ and we conclude that point (1) is true. Point (2) essentially follows by unraveling the definitions in order to check that ${h_{d}}$ is a discrete connection with domain $\mathcal{U}^{\prime}$. ∎

Theorem 4.8.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle, $R$ be a $G$-equivariant smooth retraction on $Q$ and ${h}$ a horizontal lift on $\phi$. Define ${h_{d}}\in\Sigma_{H}^{d}(\mathcal{U})$ by (4.5) and (4.6). Then, $F_{H}({h_{d}})={h}$, that is, ${h_{d}}$ integrates ${h}$.

Proof.

Let $\widehat{{h}}\coloneqq F_{H}({h_{d}})$, we want to prove that $\widehat{{h}}={h}$. Fix $(q,\delta m)\in\phi^{*}TM$. Then, (see Proposition 2.13),

$$\begin{split}\widehat{{h}}(q,\delta m)=&(D_{2}\overline{h_{d}})(q,\phi(q))(\delta m)=(D_{2}(R({h}(q,\widetilde{\check{R}}^{-1}(\phi(q_{0}),m_{1})))))\big{|}_{q_{0}=q,m_{1}=\phi(q)}(\delta m).\end{split}$$

Thus, if $\gamma:(-\epsilon,\epsilon)\rightarrow M$ is a smooth curve satisfying $\gamma(0)=\phi(q)$ and $\gamma^{\prime}(0)=\delta m$, we have

(4.7) $$\widehat{{h}}(q,\delta m)=\frac{d}{dt}\bigg{|}_{t=0}R({h}(q,\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t)))).$$

We notice that, as $id_{W_{\check{R}}}=\widetilde{\check{R}}\circ\widetilde{\check{R}}^{-1}$, we have

$$id_{T_{(\phi(q),\phi(q))}(M\times M)}=T_{0_{\phi(q)}}\widetilde{\check{R}}\circ T_{(\phi(q),\phi(q))}\widetilde{\check{R}}^{-1}.$$

Evaluating the previous identity at $(0,\delta m)$ we obtain

$$\begin{split}(0,\delta m)=&T_{0_{\phi(q)}}\widetilde{\check{R}}(T_{(\phi(q),\phi(q))}\widetilde{\check{R}}^{-1}(0,\delta m))\\ =&T_{0_{\phi(q)}}\widetilde{\check{R}}\bigg{(}T_{(\phi(q),\phi(q))}\widetilde{\check{R}}^{-1}\bigg{(}\frac{d}{dt}\bigg{|}_{t=0}(\phi(q),\gamma(t))\bigg{)}\bigg{)}\\ =&T_{0_{\phi(q)}}\widetilde{\check{R}}\bigg{(}\frac{d}{dt}\bigg{|}_{t=0}\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))\bigg{)}\end{split}$$

Writing $\delta m(t)\coloneqq\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))$, we have $(\phi(q),\gamma(t))=\widetilde{\check{R}}(\delta m(t))=(\tau_{M}(\delta m(t)),\check{R}(\delta m(t)))$, so that $\delta m(t)\in T_{\phi(q)}M$ for all $t$. As a consequence, $\frac{d}{dt}\big{|}_{t=0}\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))\in V_{\phi(q)}\subset T_{0_{\phi(q)}}T_{\phi(q)}M$ (see (4.2)). Then,

(4.8) $$\begin{split}(0,\delta m)=&T_{0_{\phi(q)}}\widetilde{\check{R}}\bigg{(}\frac{d}{dt}\bigg{|}_{t=0}\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))\bigg{)}\\ =&\bigg{(}T_{0_{\phi(q)}}\tau_{M}\bigg{(}\frac{d}{dt}\bigg{|}_{t=0}\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))\bigg{)},T_{0_{\phi(q)}}\check{R}\bigg{(}\frac{d}{dt}\bigg{|}_{t=0}\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))\bigg{)}\bigg{)}\\ =&\bigg{(}0,T_{0_{\phi(q)}}\check{R}_{\phi(q)}\bigg{(}\frac{d}{dt}\bigg{|}_{t=0}\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))\bigg{)}\bigg{)}\end{split}$$

Now, by condition (2) in Definition 4.1, $T_{0_{\phi(q)}}\check{R}_{\phi(q)}=id_{T_{\phi(q)}}M$, but notice that this last identity assumes the identification $T_{0_{\phi(q)}}T_{\phi(q)}M$ with $T_{\phi(q)}M$. If we don’t make this identification, from (4.8) we obtain

$$\frac{d}{dt}\bigg{|}_{t=0}t\ \delta m=\frac{d}{dt}\bigg{|}_{t=0}\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t)).$$

Now, we have

$$\begin{split}\frac{d}{dt}\bigg{|}_{t=0}{h}(q,(\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))))=&T_{(q,\phi(q))}{h}\bigg{(}\frac{d}{dt}\bigg{|}_{t=0}(q,(\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))))\bigg{)}\\ =&T_{(q,\phi(q))}{h}\bigg{(}0_{q},\frac{d}{dt}\bigg{|}_{t=0}(\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t)))\bigg{)}\\ =&T_{(q,\phi(q))}{h}\bigg{(}0_{q},\frac{d}{dt}\bigg{|}_{t=0}t\ \delta m\bigg{)}=\frac{d}{dt}\bigg{|}_{t=0}{h}(q,t\ \delta m)\\ =&\frac{d}{dt}\bigg{|}_{t=0}t\ {h}(q,\delta m)\in V_{q}\subset T_{0_{q}}T_{q}Q,\end{split}$$

where, in the last equality, we used the fact that ${h}$ is (fiberwise) a linear map.

Finally, using (4.7) and, later, (4.1),

$$\begin{split}\widehat{{h}}(q,\delta m)=&\frac{d}{dt}\bigg{|}_{t=0}R({h}(q,\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))))\\ =&T_{0_{q}}R\bigg{(}\frac{d}{dt}\bigg{|}_{t=0}{h}(q,(\widetilde{\check{R}}^{-1}(\phi(q),\gamma(t))))\bigg{)}=T_{0_{q}}R\bigg{(}\frac{d}{dt}\bigg{|}_{t=0}t\ {h}(q,\delta m)\bigg{)}\\ =&\frac{d}{dt}\bigg{|}_{t=0}R(t\ {h}(q,\delta m))=\frac{d}{dt}\bigg{|}_{t=0}R_{q}(t\ {h}(q,\delta m))={h}(q,\delta m).\end{split}$$

∎

Corollary 4.9.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle, $R$ be a $G$-equivariant smooth retraction on $Q$, and ${\mathcal{A}}\in\Sigma_{C}$. If ${h_{d}}\in\Sigma_{H}^{d}(\mathcal{U})$ is the one defined by Theorem 4.8 using ${h}\coloneqq F_{CH}({\mathcal{A}})$, then ${\mathcal{A}_{d}}\coloneqq(F_{CH}^{d})^{-1}({h_{d}})$ integrates ${\mathcal{A}}$, that is, $F_{C}({\mathcal{A}_{d}})={\mathcal{A}}$.

Proof.

By the commutativity of diagram (2.17) and Theorem 4.8, we have

$$F_{C}({\mathcal{A}_{d}})=F_{C}((F_{CH}^{d})^{-1}({h_{d}}))=F_{CH}^{-1}(F_{H}({h_{d}}))=F_{CH}^{-1}({h})={\mathcal{A}}.$$

∎

Using Theorem 4.8 or Corollary 4.9, the Integration Problem has a solution, provided that a $G$-equivariant smooth retraction on the principal $G$-bundle $\phi$ is available. The next result provides a context where such retractions do exist.

Theorem 4.10.

Let $\phi:Q\rightarrow M$ be a principal $G$-bundle and $\left\langle{},{}\right\rangle_{Q}$ be a $G$-invariant Riemannian metric on $Q$. Then, for any connection ${\mathcal{A}}$ on $\phi$, there is, at least, one discrete connection ${\mathcal{A}_{d}}$ on $\phi$ that integrates ${\mathcal{A}}$, that is, it satisfies $F_{C}({\mathcal{A}_{d}})={\mathcal{A}}$.

Proof.

As seen in Example 4.3, $\left\langle{},{}\right\rangle_{Q}$ induces a smooth retraction $R$ on $Q$; explicitly, $R(\delta q)=\exp_{\left\langle{},{}\right\rangle_{Q}}(\delta q)$. As $\left\langle{},{}\right\rangle_{Q}$ is $G$-invariant, its geodesics are mapped to geodesics by the $G$-action $l^{Q}$; that is, $\exp_{\left\langle{},{}\right\rangle_{Q}}\circ l^{TQ}_{g}=l^{Q}_{g}\circ\exp_{\left\langle{},{}\right\rangle_{Q}}$, and it follows that $R$ is $G$-equivariant. If ${\mathcal{A}}$ is any connection on $\phi$, the existence of a discrete connection ${\mathcal{A}_{d}}$ on $\phi$ such that $F_{C}({\mathcal{A}_{d}})={\mathcal{A}}$ is guaranteed by Corollary 4.9. ∎

Finally, the following result, based on a classical result on the existence of invariant metrics on Lie groups, gives a sufficient condition on the structure group of the principal bundle to ensure the existence of invariant Riemannian metrics.

Proposition 4.11.

Let $G$ be a connected Lie group that is the Cartesian product of a compact Lie group and a vector space (with its additive group structure). Then, any principal $G$-bundle has a $G$-invariant Riemannian metric defined in its total space.

Proof.

For $G$ as in the statement, let $\phi:Q\rightarrow M$ be a principal $G$-bundle. Pick a principal connection ${\mathcal{A}}$ on $\phi$ and a Riemannian metric $\left\langle{},{}\right\rangle^{M}$ on $M$. The horizontal distribution associated to ${\mathcal{A}}$, $Hor_{\mathcal{A}}\subset TQ$, is a $G$-invariant regular distribution satisfying $TQ=\mathcal{V}\oplus Hor_{\mathcal{A}}$, where $\mathcal{V}\coloneqq\ker(T\phi)$ is the vertical bundle of $\phi$. Let $Hor_{\mathcal{A}}(q)$ be the fiber of $Hor_{\mathcal{A}}$ over $q\in Q$; it is immediate that $T_{q}\phi|Hor_{\mathcal{A}}(q):Hor_{\mathcal{A}}(q)\rightarrow T_{\phi(q)}M$ is an isomorphism of vector spaces. For $\delta q,\delta q^{\prime}\in Hor_{\mathcal{A}}(q)$ define

$$\left\langle{\delta q},{\delta q^{\prime}}\right\rangle^{Hor_{\mathcal{A}}}\coloneqq\left\langle{T_{q}\phi(\delta q)},{T_{q}\phi(\delta q^{\prime})}\right\rangle^{M}.$$

It is easy to check that $\left\langle{},{}\right\rangle^{Hor_{\mathcal{A}}}$ is a $G$-invariant Riemannian metric on the vector bundle $Hor_{\mathcal{A}}\rightarrow Q$.

Next, we want to find a $G$-invariant Riemannian metric on $\mathcal{V}$. As vector bundles over $Q$, $Q\times\mathfrak{g}\simeq\mathcal{V}$ with the isomorphism given by $(q,\xi)\mapsto\xi_{Q}(q)$. The $G$-action $l^{TQ}$ induces a $G$-action on $\mathcal{V}$:

$$l^{TQ}_{g}(\xi_{Q}(q))=(\operatorname{Ad}_{g}(\xi))_{Q}(l^{Q}_{g}(q))\quad\text{ for all }\quad q\in Q,\ \xi\in\mathfrak{g},\ g\in G.$$

Using the previous isomorphism, this action induces the $G$-action on $Q\times\mathfrak{g}$:

$$l^{Q\times\mathfrak{g}}_{g}(q,\xi)\coloneqq(l^{Q}_{g}(q),\operatorname{Ad}_{g}(\xi))\quad\text{ for all }\quad q\in Q,\ \xi\in\mathfrak{g},\ g\in G.$$

Thus, in order to find a $G$-invariant metric on $\mathcal{V}$, we need to find such a metric on $Q\times\mathfrak{g}$ that is invariant for $l^{Q\times\mathfrak{g}}$. This last task we can do if we find an $\operatorname{Ad}$-invariant inner product on $G$. It is well known that this last problem is equivalent to that of finding a bi-invariant metric on $G$. That those metrics exist is equivalent (see [Mil76, Lemma 7.5]) to $G$ being connected and isomorphic to the Cartesian product of a compact Lie group with an additive vector group. As these conditions are satisfied in the statement, there is an $\operatorname{Ad}$-invariant metric in $\mathfrak{g}$ which produces a $G$-invariant Riemannian metric $\left\langle{},{}\right\rangle^{\mathcal{V}}$ on the vector bundle $\mathcal{V}\rightarrow Q$.

Finally, as $TQ=\mathcal{V}\oplus Hor_{\mathcal{A}}$, defining $\left\langle{},{}\right\rangle^{Q}\coloneqq\left\langle{},{}\right\rangle^{\mathcal{V}}\oplus\left\langle{},{}\right\rangle^{Hor_{\mathcal{A}}}$ (making the two subbundles orthogonal) we have that $\left\langle{},{}\right\rangle^{Q}$ is a $G$-invariant Riemannian metric on $Q$, as required. ∎

Remark 4.12.

With a minor alteration in the proof, the result of Proposition 4.11 remains valid for structure groups that are compact but not necessarily connected. In this case, the existence result follows from [AB15, Proposition 2.24] instead of [Mil76].

Corollary 4.13.

If $G$ is a Lie group that is either

1. compact, or

2. a Cartesian product of a compact and connected Lie group with a vector space,

then the Integration Problem on any principal $G$-bundle has a solution, that is, for any ${\mathcal{A}}\in\Sigma_{C}$, there is a $D$-type set $\mathcal{U}$ and ${\mathcal{A}_{d}}\in\Sigma_{C}^{d}(\mathcal{U})$ such that $F_{C}({\mathcal{A}_{d}})={\mathcal{A}}$.

The previous results show that, in many circumstances, integrals of principal connections do exist. Still, those integrals are far from unique, as we see next.

Example 4.14.

Consider the principal bundle $p_{1}:\mathbb{R}^{2}\rightarrow\mathbb{R}$ with structure group $\mathbb{R}$ acting on $\mathbb{R}^{2}$ by translation on the second component. The $1$-form ${\mathcal{A}}\coloneqq dy$ defines a principal connection on $p_{1}$. On the other hand, let $f:\mathbb{R}^{2}\rightarrow\mathbb{R}$ be any smooth function; define $C:\mathbb{R}^{2}\rightarrow\mathbb{R}$ by $C(x_{0},x_{1})\coloneqq(x_{1}-x_{0})^{2}f(x_{0},x_{1})$. By Example 2.4 ${\mathcal{A}_{d}^{f}}((x_{0},y_{0}),(x_{1},y_{1}))\coloneqq y_{1}-y_{0}+C(x_{0},x_{1})$ is a discrete connection on $p_{1}$ with domain $\mathcal{U}\coloneqq\mathbb{R}^{2}\times\mathbb{R}^{2}$. Then,

$$\begin{split}F_{C}({\mathcal{A}_{d}^{f}})(\dot{x}\partial_{x}+\dot{y}\partial_{y})=&D_{2}{\mathcal{A}_{d}^{f}}((x,y),(x,y))(\dot{x}\partial_{x}+\dot{y}\partial_{y})=\frac{d}{dt}\bigg{|}_{t=0}(t\dot{y}+C(x,x+t\dot{x}))\\ =&\dot{y}+\frac{d}{dt}\bigg{|}_{t=0}((t\dot{x})^{2}f(x,x+t\dot{x}))=\dot{y},\end{split}$$

so that $F_{C}({\mathcal{A}_{d}^{f}})=dy={\mathcal{A}}$. Thus, we have a (different) integral of ${\mathcal{A}}$ for each different smooth function $f$.

5. The abelian structure group case

In this section, we specialize the Integration Problem to the case where the structure group of the principal bundle is abelian. We see first that the Integration Problem for arbitrary curvatures has a solution if and only if the curvatures of the continuous and discrete connections are related in a specific way. Then, we prove that when that relation is realized, the Integration Problem for connections whose curvatures satisfy that relationship admits a unique solution.

Throughout this section $G$ denotes a connected abelian Lie group and, in keeping with the usual tradition, the product in $G$ is written with the additive notation. Let us also fix a principal $G$-bundle $\phi:Q\rightarrow M$ until the end of this section. Unless stated otherwise, both continuous and discrete connections are assumed to be on $\phi$.

Proposition 5.1.

For any ${\mathcal{A}}\in\Sigma_{C}$ there is a $D$-type set $\mathcal{U}$ and ${\mathcal{A}_{d}}\in\Sigma_{C}^{d}(\mathcal{U})$ such that $F_{C}({\mathcal{A}_{d}})={\mathcal{A}}$.

Proof.

Any connected abelian Lie group can be expressed as a Cartesian product of a torus and a vector space — see C. Procesi’s book [Pro07, §4.2]— and then Corollary 4.13 yields that the Integration Problem has a solution. ∎

The following two lemmas show, both in the continuous and in the discrete case, that the difference between connections on a principal bundle give rise to connections on the trivial bundle.

Lemma 5.2.

Let ${\mathcal{A}},{\mathcal{A}}^{\prime}:TQ\to\mathfrak{g}$ be two connection forms on $\phi$. There exists a unique $\epsilon\in\Omega^{1}(M,\mathfrak{g})$ such that $\epsilon\circ T\phi={\mathcal{A}}-{\mathcal{A}}^{\prime}$.

Proof.

Suppose that $q,q^{\prime}\in Q$, $\delta q\in T_{q}Q$ and $\delta q^{\prime}\in T_{q^{\prime}}Q$ are such that $\phi(q)=\phi(q^{\prime})$ and $T_{q}\phi(\delta q)=T_{q^{\prime}}\phi(\delta q^{\prime})$. Let $g\in G$ be such that $q^{\prime}=l_{g}^{Q}(q)$. Consider $v\coloneqq\delta q^{\prime}-Tl^{Q}_{g}(\delta q)\in T_{q^{\prime}}Q$ and observe that

$$T_{q^{\prime}}\phi(v)=T_{q^{\prime}}\phi(\delta q^{\prime})-T(\phi\circ l^{Q}_{g})_{q}(\delta q)=T_{q^{\prime}}\phi(\delta q^{\prime})-T_{q}\phi(\delta q)=0,$$

that is, $v$ is vertical. Letting then $\xi\in\mathfrak{g}$ be such that $v=\xi_{Q}(q^{\prime})$ we see that

(5.1) $${\mathcal{A}}(v)=\xi={\mathcal{A}}^{\prime}(v).$$

As the adjoint action is trivial in an abelian group ${\mathcal{A}}(l^{TQ}_{g}(\delta q))=\operatorname{Ad}_{g}({\mathcal{A}}(\delta q))={\mathcal{A}}(\delta q)$. Remembering that $Tl^{Q}_{g}=l^{TQ}_{g}$ we obtain that

(5.2) $${\mathcal{A}}(\delta q^{\prime})-{\mathcal{A}}(\delta q)={\mathcal{A}}\left(\delta q^{\prime}-l^{TQ}_{g}(\delta q)\right)={\mathcal{A}}(v).$$

Now, this very equation (5.2) holds for ${\mathcal{A}}^{\prime}$, and putting this together with (5.1) we see that ${\mathcal{A}}(\delta q^{\prime})-{\mathcal{A}}(\delta q)={\mathcal{A}}^{\prime}(\delta q^{\prime})-{\mathcal{A}}^{\prime}(\delta q).$ Rearranging, we obtain

(5.3) $$\left({\mathcal{A}}-{\mathcal{A}}^{\prime}\right)(\delta q)=\left({\mathcal{A}}-{\mathcal{A}}^{\prime}\right)(\delta q^{\prime}).$$

Let us define for each open set $V$ in $M$ and section $\sigma:V\to Q$ of $\phi$ the function $\epsilon^{\sigma}:TV\to\mathfrak{g}$ by $\epsilon^{\sigma}=\left({\mathcal{A}}-{\mathcal{A}}^{\prime}\right)\circ T\sigma$. Thanks to (5.3), if $\sigma:V\to Q$ and $\sigma:V^{\prime}\to Q$ are two such sections then the functions $\epsilon^{\sigma}$ and $\epsilon^{\sigma^{\prime}}$ coincide in the intersections of their domains. This gives the desired function $\epsilon:TM\to\mathfrak{g}$, and it is evidently linear at each point of $M$. The uniqueness is a consequence of the surjectivity of $T\phi$. ∎

Lemma 5.3.

Let $\mathcal{U}\subset Q\times Q$ be of $D$-type and ${\mathcal{A}_{d}}$, ${\mathcal{A}_{d}}^{\prime}$ two discrete connection forms with domain $\mathcal{U}$. There exists a unique smooth function $\zeta:\mathcal{U}^{\prime\prime}\coloneqq(\phi\times\phi)(\mathcal{U})\to G$ such that $\zeta\circ(\phi\times\phi)\mid_{\mathcal{U}}={\mathcal{A}_{d}}-{\mathcal{A}_{d}}^{\prime}$ and $\zeta(m,m)=0$ for every $m\in M$.

Moreover, if ${\mathcal{A}_{d}}$ and ${\mathcal{A}_{d}}^{\prime}$ have the same discrete curvature then the discrete connection form ${\mathcal{A}_{d}^{\zeta}}$ on $M\times G$ determined by $\zeta$ as in Example 2.4 is flat.

Proof.

The difference $\alpha\coloneqq{\mathcal{A}_{d}}-{\mathcal{A}_{d}}^{\prime}:\mathcal{U}\to G$ satisfies that if $(q,q^{\prime})\in\mathcal{U}$ and $g,g^{\prime}\in G$ then

$$\alpha(l^{Q}_{g}(q),l^{Q}_{g^{\prime}}(q^{\prime})) ={\mathcal{A}_{d}}(l^{Q}_{g}(q),l^{Q}_{g^{\prime}}(q^{\prime}))-{\mathcal{A}_{d}}^{\prime}(l^{Q}_{g}(q),l^{Q}_{g^{\prime}}(q^{\prime}))$$

$$=g^{\prime}+{\mathcal{A}_{d}}(q,q^{\prime})-g-(g^{\prime}+{\mathcal{A}_{d}}(q,q^{\prime})-g)$$

$$=\alpha(q,q^{\prime}).$$

As a consequence of this invariance $\alpha$ factors through the quotient $\mathcal{U}^{\prime\prime}$, which yields the desired $\zeta$. The uniqueness follows from the surjectivity of $(\phi\times\phi)|_{\mathcal{U}}$.

Denote by ${\mathcal{A}_{d}^{\zeta}}$ the discrete connection form on $M\times G$ determined by $\zeta$ as in Example 2.4. If $(q_{0},q_{1},q_{2})\in\mathcal{U}^{(3)}$ and $g_{0},g_{1},g_{2}\in G$, we write $m_{i}\coloneqq\phi(q_{i})$ for $i=0,1,2$ and see that the curvature of ${\mathcal{A}_{d}^{\zeta}}$ evaluated in $\left((m_{0},g_{0}),(m_{1},g_{1}),(m_{2},g_{2})\right)$ is

$${\mathcal{A}_{d}^{\zeta}}((m_{0},g_{0}),(m_{1},g_{1}))-{\mathcal{A}_{d}^{\zeta}}((m_{0},g_{0}),(m_{2},g_{2}))+{\mathcal{A}_{d}^{\zeta}}((m_{1},g_{1}),(m_{2},g_{2}))$$

$$=g_{1}+\zeta(m_{0},m_{1})-g_{0}-g_{2}-\zeta(m_{0},m_{2})+g_{0}+g_{2}+\zeta(m_{1},m_{2})-g_{1}$$

$$=\alpha(q_{0},q_{1})-\alpha(q_{0},q_{2})+\alpha(q_{1},q_{2})=0.$$

This, of course, proves that ${\mathcal{A}_{d}^{\zeta}}$ is flat. ∎

We next see that the uniqueness in the Integration Problem, whose solution we already know exists thanks to Proposition 5.1, can be controlled through the curvature.

Proposition 5.4.

Let $\mathcal{U}\subset Q\times Q$ be of $D$-type and let ${\mathcal{A}_{d}}$, ${\mathcal{A}_{d}}^{\prime}$ be two discrete connection forms on $\phi$ with domain $\mathcal{U}$. If ${\mathcal{A}_{d}}$ and ${\mathcal{A}_{d}}^{\prime}$ have the same discrete curvature and $F_{C}({\mathcal{A}_{d}})=F_{C}({\mathcal{A}_{d}}^{\prime})$ then there exists a symmetric open subset of $\mathcal{W}$ of $D$-type on which ${\mathcal{A}_{d}}$ and ${\mathcal{A}_{d}}^{\prime}$ agree.

Proof.

Write $\alpha\coloneqq{\mathcal{A}_{d}}-{\mathcal{A}_{d}}^{\prime}$ and, using Lemma 5.3, let $\mathcal{U}^{\prime\prime}\coloneqq(\phi\times\phi)(\mathcal{U})$ and $\zeta:\mathcal{U}^{\prime\prime}\to G$ such that $\zeta\circ(\phi\times\phi)\mid_{\mathcal{U}}={\mathcal{A}_{d}}-{\mathcal{A}_{d}}^{\prime}$. By virtue of Example 2.4, $\zeta$ determines a discrete connection form ${\mathcal{A}_{d}^{\zeta}}$ on $p_{1}:M\times G\rightarrow M$ defined on the open subset of $D$-type

$$\mathcal{U}_{0}\coloneqq(p_{1}\times p_{1})^{-1}(\mathcal{U}^{\prime\prime})\subset(M\times G)\times(M\times G).$$

Given $q\in Q$ and $\delta q\in T_{q}Q$ we have

$$\begin{split}D_{2}\zeta(\phi(q),&\phi(q))(T\phi(\delta q))=T_{(\phi(q),\phi(q))}\zeta(0,T\phi(\delta q))=T_{\left(q,q\right)}\alpha(0,\delta q)\\ &=T_{\left(q,q\right)}({\mathcal{A}_{d}}-{\mathcal{A}_{d}}^{\prime})(0,\delta q)=F_{C}({\mathcal{A}_{d}})(\delta q)-F_{C}({\mathcal{A}_{d}}^{\prime})(\delta q)=0.\end{split}$$

Thus, recalling Example 2.8, ${\mathcal{A}_{d}^{\zeta}}$ integrates the trivial connection in $M\times G$. Thanks to the second part of Lemma 5.3, the curvature of ${\mathcal{A}_{d}^{\zeta}}$ is zero. Consider now the discrete connection form ${\mathcal{A}_{d}^{0}}$ on $M\times G$ defined by ${\mathcal{A}_{d}^{0}}\left((m_{1},g_{1}),(m_{2},g_{2})\right)=g_{2}-g_{1}$ for any $m_{1},m_{2}$ in $M$ and $g_{1},g_{2}$ in $G$. It is straightforward to check that ${\mathcal{A}_{d}^{0}}$ is a flat connection form that integrates the trivial continuous connection. Then we may replace $\mathcal{U}_{0}$ by its symmetrization and apply Theorem 3.18 to find a symmetric open set $\mathcal{U}_{1}\subset\mathcal{U}_{0}$ of $D$-type such that the restrictions to $\mathcal{U}_{1}$ of ${\mathcal{A}_{d}^{\zeta}}$ and ${\mathcal{A}_{d}^{0}}$ coincide. Let $\mathcal{W}\coloneqq((\phi,0)\times(\phi,0))^{-1}(\mathcal{U}_{1})$. Then, if $(q_{0},q_{1})\in\mathcal{W}$, we have $\left((\phi(q_{0}),0),(\phi(q_{1}),0)\right)\in\mathcal{U}_{1}$ and, then,

$$\alpha(q_{0},q_{1}) =\zeta((\phi(q_{0}),\phi(q_{1}))={\mathcal{A}_{d}^{\zeta}}\left((\phi(q_{0}),0),(\phi(q_{1}),0)\right)$$

$$={\mathcal{A}_{d}^{0}}\left((\phi(q_{0}),0),(\phi(q_{1}),0)\right)=0,$$

which is tantamount to what we wanted to see. ∎

In what remains of this section we see that fixing any discrete curvature ${\mathcal{B}_{d}}$ and a compatible continuous curvature $\mathcal{R}({\mathcal{B}_{d}})$, any connection ${\mathcal{A}}$ with curvature $\mathcal{R}({\mathcal{B}_{d}})$ can be integrated to a unique discrete connection ${\mathcal{A}_{d}}$ whose curvature is ${\mathcal{B}_{d}}$.

By Proposition 3.17, we know that the derivation process preserves the flatness of connections. The next result, somehow, extends this result to the non-flat case, when $G$ is abelian.

Proposition 5.5.

Let $\mathcal{U}\subset Q\times Q$ be of $D$-type and ${\mathcal{A}_{d}}$, ${\mathcal{A}_{d}}^{\prime}$ be two discrete connection forms on $\phi$ with domain $\mathcal{U}$. If the discrete curvatures of ${\mathcal{A}_{d}}$ and of ${\mathcal{A}_{d}}^{\prime}$ are equal then their derived connection forms $F_{C}({\mathcal{A}_{d}})$ and $F_{C}({\mathcal{A}_{d}}^{\prime})$ have the same curvature.

Proof.

Let $\alpha\coloneqq{\mathcal{A}_{d}}-{\mathcal{A}_{d}}^{\prime}:\mathcal{U}\to G$ and, using Lemma 5.3, let $\zeta:\mathcal{U}^{\prime\prime}\coloneqq(\phi\times\phi)(\mathcal{U})\to G$ be the function such that $\zeta\circ(\phi\times\phi)\mid_{\mathcal{U}}=\alpha$. We know from this same lemma that ${\mathcal{A}_{d}^{\zeta}}$ is flat, so that, by Proposition 3.17, the associated connection $F_{C}({\mathcal{A}_{d}^{\zeta}})$ is also flat. Let $\eta$ be the form in $\Omega^{1}(M,\mathfrak{g})$ that determines the connection $F_{C}({\mathcal{A}_{d}^{\zeta}})$ as in Example 2.1: the fact that $F_{C}({\mathcal{A}_{d}^{\zeta}})$ is flat implies, thanks to Example 3.13, that $d\eta=0$. Now, given $q\in Q$ and $\delta q\in T_{q}Q$, using Example 2.8,

$$\eta_{\phi(q)}(T\phi(\delta q)) =T_{\left(\phi(q),\phi(q)\right)}\zeta\left(0,T\phi(\delta q)\right)$$

$$=T_{\left(q,q\right)}\alpha(0,\delta q)=F_{C}({\mathcal{A}_{d}})(\delta q)-F_{C}({\mathcal{A}_{d}}^{\prime})(\delta q)$$

and, then, $\phi^{\star}\eta=F_{C}({\mathcal{A}_{d}})-F_{C}({\mathcal{A}_{d}}^{\prime})$. It follows that

$$d\left(F_{C}({\mathcal{A}_{d}})-F_{C}({\mathcal{A}_{d}}^{\prime})\right)=d\phi^{\star}\eta=\phi^{\star}d\eta=0,$$

and thus $dF_{C}({\mathcal{A}_{d}})=dF_{C}({\mathcal{A}_{d}}^{\prime})$, as desired. ∎

It follows from Proposition 5.5 that given a discrete curvature form ${\mathcal{B}_{d}}$ on an open set of $D$-type $\mathcal{U}\subset Q\times Q$ —this is, the curvature form of some discrete connection with domain $\mathcal{U}$— the continuous curvature form ${\mathcal{B}}$ obtained by choosing a discrete connection ${\mathcal{A}_{d}}$ with curvature ${\mathcal{B}_{d}}$ and computing the curvature of $F_{C}({\mathcal{A}_{d}})$ is uniquely determined. We write $\mathcal{R}({\mathcal{B}_{d}})\coloneqq{\mathcal{B}}$.

Theorem 5.6.

Let $\mathcal{U}\subset Q\times Q$ be of $D$-type and let ${\mathcal{B}_{d}}$ be the discrete curvature of a discrete connection with domain $\mathcal{U}$. Given a continuous connection ${\mathcal{A}}$ with curvature $\mathcal{R}({\mathcal{B}_{d}})$ there exists an open set of $D$-type $\mathcal{U}_{0}\subset\mathcal{U}$ and a discrete connection ${\mathcal{A}_{d}}$ with domain $\mathcal{U}_{0}$ that integrates ${\mathcal{A}}$ and has curvature ${\mathcal{B}_{d}}$.

Proof.

Let $n\coloneqq\dim H_{\mathrm{dR}}^{1}(M,\mathfrak{g})$ and $\eta_{1},\ldots,\eta_{n}\in\Omega^{1}(M,\mathfrak{g})$ be closed differential forms whose classes in $H_{\mathrm{dR}}^{1}(M,\mathfrak{g})$ form a basis.

Let ${\mathcal{A}_{d}}^{\prime}$ be a discrete connection with domain $\mathcal{U}$ and curvature ${\mathcal{B}_{d}}$, and write ${\mathcal{A}}^{\prime}\coloneqq F_{C}({\mathcal{A}_{d}}^{\prime})$. Following Lemma 5.2, we let $\epsilon\in\Omega^{1}(M,\mathfrak{g})$ be such that $\epsilon\circ T\phi={\mathcal{A}}-{\mathcal{A}}^{\prime}$. Since ${\mathcal{A}}$ and ${\mathcal{A}}^{\prime}$ have the same curvature we have that $d({\mathcal{A}}-{\mathcal{A}}^{\prime})=0$, and therefore the fact that $T\phi:TQ\to TM$ is surjective implies that $\epsilon$ is a closed differential form. There exist then $c_{1},\ldots,c_{n}\in\mathbb{R}$ and $f:M\to\mathfrak{g}$ such that

$$\epsilon=df+c_{1}\eta_{1}+\cdots+c_{n}\eta_{n},$$

where the value of $df$ at each $\delta m\in T_{m}M$ is the element of $\mathfrak{g}$ that identifies with $T_{m}f(\delta m)\in T_{f(m)}\mathfrak{g}$ through the differential of the left multiplication by $f(m)$. Let $i\in\left\{1,\dots,n\right\}$. Thanks to Example 3.13 we may interpret $\eta_{i}\in\Omega^{1}(M,\mathfrak{g})$ as a flat connection ${\mathcal{A}}_{i}$ on $p_{1}:M\times G\rightarrow M$, and use Theorem 3.18 to find a subset $\mathcal{U}_{i}$ of $(M\times G)\times(M\times G)$ of $D$-type and a flat discrete connection ${\mathcal{A}_{d}^{i}}$ in $p_{1}:M\times G\rightarrow M$ defined on $\mathcal{U}_{i}$ that integrates ${\mathcal{A}}_{i}$. If $\mathcal{U}_{i}^{\prime\prime}\coloneqq(p_{1}\times p_{1})(\mathcal{U}_{i})$ then we have the local expression of ${\mathcal{A}_{d}^{i}}$, $N_{i}:\mathcal{U}_{i}^{\prime\prime}\to G$ (Example 2.4). Evidently, $N_{i}(m,m)=0$ for all $m\in M$; by virtue of Example 2.8 we have that $T_{\left(m,m\right)}N_{i}(0,\delta m)=\eta_{i}(\delta m)$ whenever $m\in M$ and $\delta m\in T_{m}M$; moreover, using the flatness of ${\mathcal{A}_{d}^{i}}$ it is straightforward to see that for any $m_{0},m_{1},m_{2}\in M$

(5.4) $$0=N_{i}(m_{0},m_{1})-N_{i}(m_{0},m_{2})+N_{i}(m_{1},m_{2}).$$

Consider the function $H:M\times M$ defined by $H(m_{1},m_{2})\coloneqq\exp_{G}\left(f(m_{2})-f(m_{1})\right)$. Set $\mathcal{U}_{0}^{\prime\prime}\coloneqq\bigcap_{i=1}^{n}\mathcal{U}_{i}^{\prime\prime}$ and define $\zeta:\mathcal{U}_{0}^{\prime\prime}\to G$ by $\zeta\coloneqq H+\sum c_{i}N_{i}$. We first observe that $\zeta(m,m)=0$ for every $m\in M$, and second that the map associated to $\zeta$ through the derivation functor is $\epsilon$: indeed, if $m\in M$ and $\delta m\in T_{m}M$ then

$$T_{\left(m,m\right)}\zeta(0,\delta m) =T_{\left(m,m\right)}H(0,\delta m)+\sum c_{i}T_{\left(m,m\right)}N_{i}(0,\delta m)$$

$$=T_{m}f(\delta m)+\sum c_{i}\eta_{i}(\delta m)=\epsilon(\delta m).$$

Let $\mathcal{U}_{0}\coloneqq(\phi\times\phi)^{-1}(\mathcal{U}^{\prime\prime}_{0})\cap\mathcal{U}$ and $\alpha\coloneqq\zeta\circ(\phi\times\phi)\mid_{\mathcal{U}_{0}}$; if $m=\phi(q)$ and $\delta m=d\phi(\delta q)$ for $q\in Q$ and $\delta q\in T_{q}Q$ the calculation above means that

$$T_{\left(q,q\right)}\alpha(0,\delta q)=T_{\left(m,m\right)}\zeta(0,\delta m)=\epsilon(\delta m)={\mathcal{A}}(\delta q)-{\mathcal{A}}^{\prime}(\delta q).$$

Let now ${\mathcal{A}_{d}}\coloneqq{\mathcal{A}_{d}}^{\prime}+\alpha$. Evidently ${\mathcal{A}_{d}}$ is a connection, for its value along the diagonal is null and

$${\mathcal{A}_{d}}(l^{Q}_{g}(q),l^{Q}_{g^{\prime}}(q^{\prime})) ={\mathcal{A}_{d}}^{\prime}(l^{Q}_{g}(q),l^{Q}_{g^{\prime}}(q^{\prime}))+\alpha(l^{Q}_{g}(q),l^{Q}_{g^{\prime}}(q^{\prime}))$$

$$=g^{\prime}+{\mathcal{A}_{d}}^{\prime}(q,q^{\prime})-g+\alpha(l^{Q}_{g}(q),l^{Q}_{g^{\prime}}(q^{\prime}))$$

$$=g^{\prime}+{\mathcal{A}_{d}}(q,q^{\prime})-g$$

for any $g,g^{\prime}\in G$ and $(q,q^{\prime})\in\mathcal{U}_{0}$. For any $q\in Q$ and $\delta q\in T_{q}Q$ we have

$$F_{C}({\mathcal{A}_{d}})(\delta q) =T_{\left(q,q\right)}{\mathcal{A}_{d}}^{\prime}(0,\delta q)+T_{\left(q,q\right)}\alpha(0,\delta q)$$

$$={\mathcal{A}}^{\prime}(\delta q)+{\mathcal{A}}(\delta q)-{\mathcal{A}}^{\prime}(\delta q)={\mathcal{A}}(\delta q).$$

It is easy to see that the curvatures ${\mathcal{B}_{d}}$ and ${\mathcal{B}_{d}}^{\prime}$ of ${\mathcal{A}_{d}}$ and ${\mathcal{A}_{d}}^{\prime}$ are related as follows. For any $(q_{0},q_{1},q_{2})\in\mathcal{U}_{0}^{(3)}$,

$${\mathcal{B}_{d}}(q_{0},q_{1},q_{2})={\mathcal{B}_{d}}^{\prime}(q_{0},q_{1},q_{2})+\alpha(q_{0},q_{1})-\alpha(q_{0},q_{2})+\alpha(q_{1},q_{2}).$$

If for any such $(q_{0},q_{1},q_{2})\in\mathcal{U}_{0}^{(3)}$ we define $m_{j}\coloneqq\phi(q_{j})$ for $0\leq j\leq 2$, we have that

$$\alpha(q_{0},q_{1})-\alpha(q_{0},q_{2})+\alpha(q_{1},q_{2})$$

$$=\zeta(m_{0},m_{1})-\zeta(m_{0},m_{2})+\zeta(m_{1},m_{2})$$

$$\!\begin{multlined}=H(m_{0},m_{1})-H(m_{0},m_{2})+H(m_{1},m_{2})\\ +\sum c_{i}\left(N_{i}(m_{0},m_{1})-N_{i}(m_{0},m_{2})+N_{i}(m_{1},m_{2})\right).\end{multlined}=H(m_{0},m_{1})-H(m_{0},m_{2})+H(m_{1},m_{2})\\ +\sum c_{i}\left(N_{i}(m_{0},m_{1})-N_{i}(m_{0},m_{2})+N_{i}(m_{1},m_{2})\right).$$

The first three of these terms add up to zero — this follows from the additivity of the exponential map in abelian groups — and the other ones vanish in view of (5.4). It follows that ${\mathcal{B}_{d}}={\mathcal{B}_{d}}^{\prime}$, finishing the proof. ∎

Corollary 5.7.

Let $\mathcal{U}\subset Q\times Q$ be of $D$-type, ${\mathcal{B}_{d}}$ be the discrete curvature of a discrete connection with domain $\mathcal{U}$ and ${\mathcal{A}}$ be a continuous connection. There exists an open subset $\mathcal{V}\subset\mathcal{U}$ of $D$-type and a discrete connection ${\mathcal{A}_{d}}$ with domain $\mathcal{V}$ and curvature ${\mathcal{B}_{d}}$ that integrates ${\mathcal{A}}$ if and only if the curvature of ${\mathcal{A}}$ is $\mathcal{R}({\mathcal{B}_{d}})$.

In case these conditions hold and ${\mathcal{A}_{d}}$ and ${\mathcal{A}_{d}}^{\prime}$ are two such discrete connections there exists a symmetric open subset of $\mathcal{V}$ of $D$-type on which ${\mathcal{A}_{d}}$ and ${\mathcal{A}_{d}}^{\prime}$ agree.

Proof.

The necessity is a consequence of the fact that $\mathcal{R}({\mathcal{B}_{d}})$ is well defined —see Proposition 5.5—, the sufficiency follows from Theorem 5.6 above, and, finally, the uniqueness is guaranteed by Proposition 5.4. ∎

Appendix A A basis of neighborhoods

Theorem 3.18 —one of our main results— relies on a particularization of the following claim made in the proof of Proposition 2. 1 in [CMS18]:

open sets of the form $U^{(k)}=G^{(k)}\cap U^{k}$, where $U\subset G_{\tau}\cap G_{\iota}$ is an open neighborhood of $M$, form a basis of neighborhoods of $M$ in $G^{(k)}$.

In response to our request, I. Mărcuţ generously provided guidelines for a more detailed explanation, which we further elaborate below: the conclusions are formulated as Proposition A.2.

Let $X$ be a smooth manifold. Consider the manifold $P\coloneqq X\times X$ and, for $j=1,2$, the fiber bundle $P^{j}$ over $X$ given by the map $p_{j}:P\to X$ defined by $p_{j}(x_{1},x_{2})\coloneqq x_{j}$. Our main proposition concerns the fiber products $P_{2}\coloneqq P^{2}\prescript{}{p_{2}}{\times}_{p_{1}}P^{1}$ and $(\Delta_{X})_{2}\coloneqq\Delta_{X}\prescript{}{p_{2}}{\times}_{p_{1}}\Delta_{X}\subset P_{2}$.

Let $g$ be a Riemannian metric on $X$, and define $E_{g}^{1},E_{g}^{2}:TX\to P^{j}$ by $E_{g}^{1}(v_{x})\coloneqq(x,\exp_{g}(x))$ and $E_{g}^{2}(v_{x})\coloneqq(\exp_{g}(x),x)$. The following claims can be straightforwardly verified:

1. There exists an open neighborhood $\mathcal{E}_{g}\subset TX$ of the zero section $0_{X}$ such that if $LP^{j}\coloneqq E^{j}_{g}(\mathcal{E}_{g})$ then $E^{j}_{g}:\mathcal{E}_{g}\to LP^{j}$ is a fiber-preserving diffeomorphism whenever $j=1,2$.

2. The map $E_{g}:TX\times TX\to P^{2}\times P^{1}$ defined by $E_{g}(v_{x},w_{y})\coloneqq(E^{2}_{g}(v_{x}),E^{1}_{g}(w_{y}))$ restricts to a diffeomorphism from $\left(\mathcal{E}_{g}\times\mathcal{E}_{g}\right)\cap(TX\oplus TX)$ onto $\left(LP^{2}\times LP^{1}\right)\cap P_{2}$, that are open neighborhoods of $0_{X}\oplus 0_{X}$ and of $(\Delta_{X})_{2}$, respectively.

Lemma A.1.

Let $\mathcal{V}\to X$ be a vector bundle with a bundle metric and write $\mathcal{V}_{\rho}=\left\{v_{x}\in\mathcal{V}_{x}:\lVert v_{x}\rVert<\rho(x)\text{ and }x\in X\right\}$ for each real-valued function $\rho$ on $X$. The family $\left\{\mathcal{V}_{\rho}:\text{$\rho:X\to\mathbb{R}_{>0}$ continuous}\right\}$ is a basis of open neighborhoods of the zero section of $\mathcal{V}$.

Proof.

This can be proved using Theorem 2.1.3 on p. 56 of [Jos17]. ∎

Proposition A.2.

Let $W\subset P_{2}$ be an open neighborhood of $(\Delta_{X})_{2}$. There exists an open neighborhood $U\subset P$ of $\Delta_{X}$ such that $U^{(2)}\coloneqq(U\times U)\cap P_{2}$ is contained in $W$ and contains $(\Delta_{X})_{2}$.

Proof.

We may assume that $W\subset LP^{2}\times LP^{1}$ and define $W^{\prime}\coloneqq E_{g}^{-1}(W)$, which is contained in $\left(\mathcal{E}_{g}\times\mathcal{E}_{g}\right)\cap(TX\oplus TX)$ and contains $0_{X}\times 0_{X}$.

The Riemannian metric $g$ on $X$ provides a bundle metric $\langle\cdot,\cdot\rangle$ on $TX$ and, then, the law $\langle(v_{x},w_{x}),(v_{x}^{\prime},w_{x}^{\prime})\rangle\coloneqq\langle v_{x},v^{\prime}_{x}\rangle+\langle w_{x},w^{\prime}_{x}\rangle$ endows $TX\oplus TX\to X$ with a bundle metric. By Lemma A.1 there exists $\rho:X\to\mathbb{R}_{>0}$ such that $(TX\oplus TX)_{\rho}\subset W^{\prime}$. Let $\rho^{\prime}\coloneqq{\tfrac{1}{\sqrt{2}}\rho}$ and consider $(TX)_{\rho^{\prime}}$. By definition, $(TX)_{\rho^{\prime}}\subset TX$ is an open neighborhood of $0_{X}$; on the other hand, a short computation shows that

(A.1) if $$v_{x},v_{x}^{\prime}\in(TX)_{\rho^{\prime}} (v_{x},v_{x}^{\prime})\in(TX\oplus TX)_{\rho}\subset W^{\prime}$$ then .

Define $U^{j}\coloneqq E^{j}_{g}((TX)_{\rho^{\prime}})\subset P^{j}$ for $j=1,2$ and $U\coloneqq U^{2}\cap U^{1}$. It is immediate that $U$ is an open neighborhood of $\Delta_{X}$ in $P$, and one can see using (A.1) that $U^{(2)}\coloneqq\left(U\times U\right)\cap P_{2}$, that evidently contains $(\Delta_{X})_{2}$, is contained in $W$. ∎