surface via FrØchet reduction - arXiv · surface via FrØchet reduction ... holonomies and a...

Yang-Mills moduli spaces over an orientable closedsurface via Fréchet reduction

T. Diez‡ and J. Huebschmann§

September 17, 2019

Given a principal bundle on an orientable closed surface with compactconnected structure group, we endow the space of based gauge equiva-lence classes of smooth connections relative to smooth based gauge trans-formations with the structure of a Fréchet manifold. Using Wilson loopholonomies and a certain characteristic class determined by the topologyof the bundle, we then impose suitable constraints on that Fréchet manifoldthat single out the based gauge equivalence classes of central Yang-Millsconnections but do not directly involve the Yang-Mills equation. We alsoexplain how our theory yields the based and unbased gauge equivalenceclasses of all Yang-Mills connections and deduce the strati�ed symplecticstructure on the space of unbased gauge equivalence classes of centralYang-Mills connections. The crucial new technical tool is a slice analysis inthe Fréchet setting.Keywords: Yang-Mills connection on a closed Riemann surface, moduli space of smoothYang-Mills connections on a closed Riemann surface in the Fréchet setting, strati�edsymplectic structure on the moduli space of Yang-Mills connections on a closed Riemannsurface, moduli space of holomorphic vector bundles on a projective curve, Fréchet sliceanalysisMSC 2010: Primary: 58E15, 81T13; Secondary: 14D20, 53C07, 58D27

Contents

1 Introduction 22 The universal example 43 Topology of principal bundles over a closed oriented surface 64 Manifold structure on the space of based gauge equivalence classes of connections 85 Wilson loops 96 Preparing for the description of the Fréchet constraints 137 The constraints to be imposed on the based gauge orbit manifold 188 Reconstruction of the space of gauge equivalence classes of all Yang-Mills connections 249 Dependence on the Choices 2810 Strati�ed Symplectic Structure 2911 Fréchet Slices 3111.1 Principal Lie subgroups 3111.2 De�nition of a slice 3411.3 Slice for the action of a subgroup 35References 36

‡Max Planck Institute for Mathematics in the Sciences Inselstraße 22, 04103 Leipzig, Germany andInstitut für Theoretische Physik, Universität Leipzig, Postfach 100 920, 04009 Leipzig, [email protected]

§ USTL, UFR de Mathématiques, CNRS-UMR 8524, Labex CEMPI (ANR–11–LABX–0007–01), 59655Villeneuve d’Ascq Cedex, France. [email protected]

arX

iv:1

704.

0198

2v2

[m

ath.

DG

] 1

6 Se

p 20

19

Introduction 2

1 Introduction

We rework and extend, in the framework of Fréchet manifolds, the approach of Atiyahand Bott [AB82] to the moduli spaces of Yang-Mills connections on a closed Riemannsurface.

Let G be a compact connected Lie group and M a Riemannian manifold, and let� ∶ P → M be a principal G-bundle. Yang-Mills theory proceeds by constructingmoduli spaces of solutions of the relevant partial di�erential equations modulo gaugetransformations. In the case of a manifold M of arbitrary �nite dimension, smoothsolutions need not exist and, even if they exist, constructing solutions is a rather delicateendeavor. For example, Uhlenbeck established the local solvability in the Coulombgauge [Uhl82]. The main di�culty in Yang-Mills theory resides, perhaps, in the factthat the space of connections is too big, with too much unwanted gauge freedom orgauge ambiguity.

We will here concentrate on the case where the base manifold M is an orientableclosed (compact) Riemann surface, and we write this surface as Σ and denote by volΣ a(suitably normalized) volume form on Σ. The situation now greatly simpli�es. Indeed,the central Yang-Mills connections A on � (the Yang-Mills connections having centralcurvature) are characterized by the equation

curvA = −X� ⋅ volΣ, (1.1)

for a certain characteristic class X� in the center h of the Lie algebra g of G determinedby the topology of the bundle � . Concerning the sign, see Section 2 below. In afundamental paper [AB82], Atiyah and Bott showed that, in the case at hand, the Yang-Mills solution space decomposes into a disjoint union of solution spaces of central Yang-Mills connections for reductions of the structure group. Furthermore, they describedthe structure of such a central Yang-Mills solution space using in�nite dimensionaltechniques including symplectic reduction phrased in terms of suitable Sobolev spaces.They emphatically raised the issue of understanding the singularities of this kind ofspace. In the presence of singularities, the concept of a strati�ed symplectic space due to[SL91] satisfactorily isolates the singular structure of a space that arises by symplecticreduction in �nite dimensions. This approach has so far not been available for a spacewhich results from an in�nite-dimensional construction, such as the solution space ofcentral Yang-Mills connections. In particular, a rigorous theory of Poisson brackets inin�nite dimensions is lacking and functional analytic issues obscure the underlyinggeometry.

We circumvent these technical issues and reduce the analysis of the singularities ofthe moduli space of central Yang–Mills connections to a �nite-dimensional problem.As a new approach to the issue of gauge freedom, we perform reduction by stages,�rst relative to based gauge transformations and thereafter relative to the residualaction of the structure group. We deliberately write “reduction by stages” rather than“symplectic reduction by stages”, since there is no obvious momentum mapping forthe group of based gauge transformations. We therefore take the entire orbit space

Introduction 3

relative to the action, on the space of smooth connections, of the Fréchet Lie groupof based gauge transformations. We show that this orbit space is a Fréchet manifold.We then show that the Wilson loop mapping developed by the second-named authorin [Hue98] yields a smooth map from the Fréchet manifold of based gauge equivalenceclasses of connections to the product manifold G2� (technically the space of G-valuedhomomorphisms de�ned on the free group F on a family of 2� chosen generators ofthe fundamental group of Σ) where � denotes the genus of Σ. We then impose, interms of the holonomies around all contractible loops bounded by a disk, suitableconstraints on that Fréchet manifold which single out the based gauge equivalenceclasses of smooth central Yang-Mills connections. These constraints recover the centralcurvature condition in [AB82] and thereby yield a replacement for the Yang-Millsequation. We show that the Wilson loop mapping (more precisely, the map (7.4) below)yields a homeomorphism between the space of based gauge equivalence classes ofsmooth connections satisfying the appropriate constraint (equivalently: space of basedgauge equivalence classes of smooth central Yang–Mills connections) endowed with itsFréchet topology and a suitable space of homomorphisms into the structure group G,realized within the product manifold G2� . We also show that, relative to the G-orbitstrati�cations, the homeomorphism under discussion is an isomorphism of strati�edspaces which, on each stratum, restricts to a di�eomorphism. See Theorem 7.1 belowfor details.

A crucial step in the proof of Theorem 7.1 consists in establishing the fact that theWilson loop mapping from that space of based gauge equivalence classes to that space ofhomomorphisms into the structure group G is a local homeomorphism. The continuityof that Wilson loop mapping is a consequence of the smoothness of the Wilson loopmapping de�ned on the ambient space. A construction in [AB82, Section 6] yieldsthe inverse of the map under discussion, see Remark 7.3 below. While, on the onehand, the continuity of that inverse map is a consequence of Uhlenbeck compactness[Uhl82], that compactness theorem relies on Sobolev space techniques. We avoid thesetechniques and develop a proof that merely involves Fréchet space techniques. Theproof we give includes the statement of the compactness theorem for our Fréchet spacesituation over a surface. See Remark 7.4 and Lemma 7.5 below.

Our basic tool is a slice analysis technique for groups of smooth gauge transformationsin the Fréchet setting. This technique was developed in [ACM89] and [Sub86] and hasthereafter been extended by the �rst-named author in [Die13]. It is phrased within thedi�erential calculus due to Michal and Bastiani, see [Nee06] for further remarks concern-ing the di�erential calculus on locally convex spaces, and crucially involves the versionof the Nash–Moser inverse function theorem given in [Ham82] in the tame Fréchetsetting. Below we reproduce that result as Theorem 11.5. This di�erential calculus, inturn, has since been applied to global analytic problems in [Ham82], [Mil84], [Nee06]and elsewhere. The slice analysis technique enables us to reconstruct, in our framework,the strati�ed symplectic structure on such a Yang-Mills solution space developed previ-ously [Hue95b], [HJ94]. The upshot of the present paper is that the theory developedin [Hue95a], [Hue96], and [Hue98] – see [Hue01] for a leisurely introduction – can be

The universal example 4

built in terms of the appropriate spaces of smooth connections. Pushing a bit further,we endow the Fréchet manifold of based gauge equivalence classes of connectionswith a quasi-hamiltonian G-space structure relative to the structure group G, and theYang-Mills moduli space we are interested in then arises by reduction relative to G. SeeRemark 10.2 below for details. This observation, together with the idea of imposing suit-able constraints on the space of based gauge equivalence classes of smooth connections,justi�es, perhaps, the term Fréchet reduction in the title. The analysis of the topology ofthe Yang-Mills solution spaces in [AB82, Section 6] enables us to reconstruct, entirelyin terms of Fréchet topologies on spaces of smooth connections, the moduli space of allYang-Mills connections on the �xed bundle � and, furthermore, that of all Yang-Millsconnections relative to G over the surface Σ. See Section 8 for details. Thus, we canunderstand the Yang-Mills solution spaces over a closed surface in terms of spaces ofordinary smooth connections. This raises the issue whether we can understand theYang-Mills solution spaces over a general compact Riemannian manifold in terms ofspaces of smooth connections.

For a general gauge theory situation, an alternate model of the space of basedholonomies was developed by the second-named author in [Hue99]. That model yieldsa rigorous approach to lattice gauge theory. For the special case explored in the presentpaper, the alternate model comes down to the extended moduli space constructiondeveloped in [HJ94], [Hue95b], [Jef97], see Section 10 below. The construction in thepresent paper is somewhat a special case of that in [Hue99], but now in the frameworkof Fréchet manifolds.

In Section 2, following [AB82], we spell out the universal example in terms of the(necessarily unique) Schur cover of the fundamental group of Σ, tailored to our purposesand, in Section 3, we recall the requisite topological classi�cation of principal bundlesover Σ. In Section 4 we introduce the Fréchet manifold structure on the space of basedequivalence classes of connections. In Section 5 we explore the smoothness of theWilson loop mapping and in Section 6 we establish a technical result. In Section 7 wethen proceed to describe the constraints and to spell out and prove the main result,Theorem 7.1, and we complete the proof in Section 9. In Section 8 we explain how wecan recover, entirely in terms of Fréchet topologies on spaces of smooth connections,the moduli space of all Yang-Mills connections relative to the structure group G. InSection 10 we brie�y recall the resulting singular symplectic geometry, phrased in thelanguage of strati�ed symplectic spaces, and the �nal section, Section 11, is devoted tothe requisite Fréchet space technology.

2 The universal example

We maintain the notation G for a compact connected Lie group. Our base manifold isan orientable (real) closed (compact) connected surface Σ of genus ≥ 1. Our notationfor the di�erential forms on a smooth manifold M is (M, ⋅ ).

Ordinary Yang-Mills theory necessitates a choice of Riemannian metric on the struc-ture group and on the base manifold. Our approach does not involve such metrics

The universal example 5

except when we link it to ordinary Yang-Mills theory, and a choice of complex structureof Σ will play no role either, except possibly for the orientation it determines. As a sideremark we note that the Yang-Mills equation dA ∗ curvA = 0 makes sense for a principalbundle having as structure group a general Lie group when we interpret the operator ∗as having its values in the forms with values in the dual to the adjoint bundle.

Let � (≥ 1) denote the genus of Σ, let Q be a point of Σ, �xed throughout and takenhenceforth as base point, and let �1 = �1(Σ, Q) denote the fundamental group of Σ atthe point Q. Consider the standard presentation

⟨x1, y1,… , x� , y� ; r⟩, r = [x1, y1] ⋅… ⋅ [x� , y� ] (2.1)

of �1. Let u1, v1,… , u� , v� be a canonical system (in the sense of [ZVC80]) of (closed)curves in Σ that have Q as starting point; that is to say: Cutting Σ successively alongthese curves yields a planar disk e2; see, e.g., [ZVC80, §3.3.2] for details. We supposethings arranged in such a way that these curves represent the respective generatorsx1, y1,… , x� , y� of the fundamental group �1 = �1(Σ, Q) of Σ and that the boundary pathof the disk e2 yields the relator r . Thus the standard cell decomposition of Σ with asingle 2-cell e2 corresponding to r results.

We denote by F the free group on x1, y1,… , x� , y� and by N the normal closure of rin F . Consider the quotient group Γ = F /[F , N ]. The image [r] ∈ Γ of r generates thecentral subgroup ℤ⟨[r]⟩ = N /[F , N ] of Γ, and the resulting extension

0 ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ ℤ⟨[r]⟩ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Γ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ �1 ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ 1 (2.2)

is central. Since the transgression homomorphism H2(�1)→ ℤ⟨[r]⟩ is an isomorphism,the extension (2.2) is a maximal stem extension (Schur cover) and since, furthermore,the abelianization of �1 is a free abelian group, that maximal stem extension is uniqueup to within isomorphism [Gru70, §9.9 Theorem 5 p. 214]. Atiyah and Bott use theterminology “universal central extension” to refer to this situation [AB82, §6].

Relative to the embedding ℤ⟨[r]⟩ → ℝ of abelian groups given by the assignmentto [r] of 2� ∈ ℝ, similarly as in [AB82, §6 p. 560], let Γℝ denote the 1-dimensional Liegroup characterized by the requirement that

0 ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ ℤ⟨[r]⟩ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Γ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ �1 ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ 1⏐⏐↓

⏐⏐↓

‖‖‖0 ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ ℝ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Γℝ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ �1 ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ 1

(2.3)

be a commutative diagram of central extensions. Consider a principal U(1)-bundle�MΣ ∶ MΣ → Σ on Σ having Chern class 1, cf., e.g., [Hir95, I.4 p. 58 �] for the precisemeaning of having Chern class 1. Similarly as in [AB82, §6], we realize (2.2) in terms of�MΣ as follows (in [AB82, §6], the compact 3-manifold MΣ is written as Q): Choose asmooth connection form

!MΣ ∶ TMΣ ⟶ iℝ = Lie(U(1)) ⊆ Lie(GL(1,ℂ)) = ℂ (2.4)

Topology of principal bundles over a closed oriented surface 6

on �MΣ having non-degenerate curvature form and write the curvature form as −2�ivolΣ ∈2(Σ, iℝ). This de�nes the volume form volΣ on Σ and �xes an orientation of Σand, when we orient e2 consistently with Σ, �xes an orientation of the closed path[u1, v1] ⋅… ⋅ [u� , v� ], the “boundary path” of e2, as well. Unlike the approach in [AB82],in our setting, there is no need to choose a Riemannian metric on Σ and to arrange forvolΣ to be the volume form associated to that Riemannian metric. We can, of course,choose a (positive) complex structure on Σ that is compatible with the volume formvolΣ (as a symplectic structure) and work with the resulting Kähler metric on Σ. Thenthe 2-form volΣ is the corresponding normalized Riemannian volume form on Σ, andthe orientation of Σ coincides with the orientation arising from the complex structure.This reconciles our approach with that in [AB82] and shows that the sign in the aboveexpression for the curvature form is consistent with standard Chern-Weil theory, cf.[Che95].

The group Γ is isomorphic to the fundamental group ofMΣ. To make the identi�cationof Γ with the fundamental group of MΣ explicit, we choose a pre-image QMΣ ∈ MΣ ofthe chosen base point Q of Σ and, thereafter, closed lifts u1,MΣ , v1,MΣ , . . . , u� ,MΣ , v� ,MΣ ,to MΣ, of the canonical curves having QMΣ ∈ MΣ as starting point. Though this is notstrictly necessary, we note that, since, up to gauge transformations, the curvaturedetermines the connection only up to a member of H1(Σ,U(1)), we can rechoose thesmooth principal U(1)-connection form !MΣ in such a way that its holonomies withrespect to QMΣ and the canonical curves in Σ are trivial. See also Remark 7.6 below.Such a connection is unique up to gauge transformations. The respective horizontallifts u1,MΣ , v1,MΣ , . . . , u� ,MΣ , v� ,MΣ in MΣ having QMΣ as starting point are then closed, andwe can take their based homotopy classes as generators of �1(MΣ, QMΣ) ≅ Γ.

We take the universal covering projection ℝ → U(1) to be given by the associationt ↦ eit (t ∈ ℝ). The U(1)-bundle projection �MΣ ∶ MΣ → Σ lifts to a principal ℝ-bundleprojection �MΣ

∶ MΣ → Σ de�ned on the universal covering manifold MΣ of MΣ havingas base the universal covering manifold Σ of Σ, and the ℝ-action and Γ-action on MΣcombine to a Γℝ-action on MΣ turning the resulting projection �MΣ

∶ MΣ → Σ into aprincipal bundle having Γℝ as its structure group. The connection form !MΣ on �MΣ de-termines a connection form !MΣ

∶ TMΣ → ℝ on �MΣ∶ MΣ → Σ and hence a connection

having curvature form −2�volΣ ∈ 2(Σ,ℝ), and this connection is necessarily central.Here and henceforth we refer to a connection as being central when the values of itscurvature form lie in the center of the structure group. The universal covering projec-tion ℝ → U(1) induces the map Lie(Γℝ) = ℝ → iℝ = Lie(U(1)) given by t ↦ it (t ∈ ℝ)and, by construction, the connection and curvature forms on �MΣ

and �MΣ correspondvia that induced map.

3 Topology of principal bundles over a closed oriented surface

To unveil, in the Fréchet setting, the structure of the moduli space of Yang-Millsconnections on a �xed principal G-bundle over Σ we exploit a certain topologicalcharacteristic class that lies in the center of the Lie algebra of G, cf. [AB82, §6 p. 560].

Topology of principal bundles over a closed oriented surface 7

For intelligibility and for later reference, we brie�y recall how this class arises and howit relates to the topological classi�cation of principal G-bundles on Σ.

By structure theory, the compact connected Lie group G is generated by its maximalsemi-simple subgroup S = [G, G] and the connected component H of the identity ofthe center of G and, accordingly, we write G as G = H ×D S, for the �nite discretecentral subgroup D = H ∩ S of G. The injection H → G induces an isomorphismH = H /D → G/S onto the abelianized group GAb = G/S, and the group H is a torusgroup.

Let h denote the Lie algebra of H (and of H ). The assignment to ' ∈ Hom(U(1), H )of the induced principal H -bundle �MΣ ×' H ∶ MΣ ×' H → Σ topologically classi�esprincipal H -bundles on Σ. When we assign to a member ' of Hom(U(1), H ), viewedas a 1-parameter subgroup of H , its tangent vector X' ∈ h at the origin (so that'(eit) = exp(tX')), we obtain an isomorphism

Hom(U(1), H )⟶ ker(exp∶ h → H ) = �1(H ) (3.1)

of discrete abelian groups. Observing that Poincaré duality with respect to the orien-tation of Σ chosen in Section 2 above identi�es H2(�1, �1(H )) canonically with �1(H )reconciles this description with the ordinary topological classi�cation of principalH -bundles on Σ. In this language, our reference bundle �MΣ ∶ MΣ → Σ introduced inSection 2 above has characteristic class X�MΣ = 2�i ∈ iℝ = Lie(U(1)).

Consider a principalG-bundle � ∶ P → Σ onΣ. The bundle projection factors throughthe induced map from the orbit manifold P/S to Σ, necessarily a smooth principal H -bundle, and we denote this bundle by �H ∶ P/S → Σ and by X� ∈ h the single topologicalcharacteristic class of �H . This yields the requisite topological characteristic class forthe bundle � , cf. [AB82, §6 p. 560].

Consider the relator map

r ∶ Hom(F , G)⟶ G, r(� ) = [� (x1), � (y1)] ⋅ ⋯ ⋅ [� (x� ), � (y� )] ∈ G. (3.2)

The injection Hom(Γ, G) ⊆ Hom(F , G) induced by the surjection F → Γ identi�es acertain subspace of Hom(Γ, G) with the subspace r−1(exp(X� )) of Hom(F , G), and wedenote that subspace ofHom(Γ, G) byHomX� (Γ, G). Viewed as a subspace ofHom(F , G),the space HomX� (Γ, G) is necessarily compact.

Let HomX� (Γℝ, G) denote the space of homomorphisms � from Γℝ to G that havethe property that � (t[r]) = exp(tX� ) (t ∈ ℝ). Given � ∈ HomX� (Γ, G), with a slightabuse of the notation � , setting � (t[r]) = exp(tX� ) (t ∈ ℝ), we obtain an extension of� to a homomorphism � ∶ Γℝ ⟶ G, uniquely determined by � and X� , that is, therestriction HomX� (Γℝ, G)→ HomX� (Γ, G) is a bijection. Hence, given � ∈ HomX� (Γ, G),the associated principal G-bundle �� ∶ MΣ ×� G → Σ on Σ is de�ned.Proposition 3.1: The assignment to � ∈ HomX� (Γ, G) of �� yields a bijection between theconnected components of HomX� (Γ, G) and the topological types of principal G-bundles onΣ having X� ∈ h as its corresponding characteristic class, the number of components beinggiven by the order |�1(S)| of �1(S).

Manifold structure on the space of based gauge eqivalence classes of connections 8

The reasoning in [AB82, Section 6] yields a proof of this proposition. Su�ce itto note the following: Let S = G/H ≅ S/D. The bundle � is topologically classi�edby its characteristic class in H2(Σ, �1(G)) and, likewise, the principal H -bundle �H istopologically classi�ed by its characteristic class in H2(Σ, �1(H )). Under the inducedhomomorphism H2(Σ, �1(G))→ H2(Σ, �1(H )), the characteristic class of � goes to thecharacteristic class of �H . Poincaré duality relative to the fundamental class [Σ] identi�esthat homomorphism with the induced homomorphism �1(G)→ �1(H ), and the inducedhomomorphism �1(S)→ �1(G) identi�es that kernel with �1(S). We leave the details tothe reader.

We denote by HomX� (Γ, G)� ⊆ HomX� (Γ, G) the connected component that corre-sponds to the principal G-bundle � .

4 Fréchet manifold structure on the space of based gauge eqiv-alence classes of connections

Consider a principal G-bundle � ∶ P → M on a smooth compact connected manifold M .Let g denote the Lie algebra of G. We use the standard formalism with structure groupG acting on P from the right etc. We denote by ad� ∶ P ×G g → M the (in�nitesimal)adjoint bundle associated to � and write the graded object of di�erential forms on Mwith values in ad� as ∗(M, ad� ).

Consider the space � of smooth G-connections on � ∶ P → M . The Atiyah se-quence [Ati57] associated with � , spelled out here for the total spaces, has the form

0⟶ P ×G g ⟶ (TP )/G ⟶ TM ⟶ 0, (4.1)

and we realize the space � as that of vector bundle sections TM → (TP )/G for (4.1).Thus the points of � are in bijective correspondence with the sections of an a�nebundle over M and in this way carry a natural tame Fréchet manifold structure modeledon the vector space 1(M, ad� ) [ACMM86].

Let � denote the group of smooth gauge transformations of � . We realize this group asthe group Γ∞(Ad� ) of smooth sections of the associated adjoint bundle Ad� ∶ P ×G G →M , endowed with the pointwise group structure. In this manner, � acquires thestructure of a tame Fréchet Lie group [ACMM86].

As usual, we identify the Lie algebra of � with the Lie algebra gau� = Γ∞(ad� ) ofsmooth sections of the adjoint bundle ad� onM , endowed with the pointwise Lie bracket.This Lie algebra acquires the structure of a tame Fréchet Lie algebra and, relative to theFréchet structures, the exponential map exp∶ gau� → � is a local di�eomorphismat 0, see [CM85, Section IV]. In these topologies, the standard left � -action on � istame smooth, proper and admits slices at every point [ACM89, Die13]. In particularthe orbit space � ⧵� is strati�ed by tame Fréchet manifolds.

Choose a base point QP for the total space P of � ∶ P → M and let Q = � (QP ) ∈ M .

Wilson loops 9

The evaluation map evQP ∶ � → G characterized by the identity

�(Q) = [QP , evQP (�)] ∈ P ×G G,

as � ranges over � (the group of sections of Ad� ), is a morphism of Lie groups. Thegroup of smooth based gauge transformations associated to the data is the kernel ofevQP . This group is a normal, locally exponential Lie subgroup of � [Nee06, Prop.IV.3.4]. While evQP depends on the choice of QP , the kernel of evQP is independent ofthe particular choice of QP in � −1(Q) and depends only on Q; we therefore denote thegroup of based gauge transformations by � ,Q .

Relative to � , the group � ,Q has �nite codimension equal to the dimension of G.Theorem 11.4 below entails that the obvious surjection � → � /� ,Q is a smooth rightprincipal � ,Q-bundle; we refer to this kind of situation by the phrase “� ,Q is a principalLie subgroup of � ”.

Proposition 11.9 below will say that slices for the � -action on � yield slices for the� ,Q-action on � and thus implies the following:Proposition 4.1: Relative to the Fréchet manifold structures, the � ,Q-action on �is smooth and free; furthermore, it admits slices at every point. Hence the orbit space� ,Q = � ,Q⧵� acquires a smooth Fréchet manifold structure, and the canonical projection� → � ,Q is a smooth principal � ,Q-bundle.

5 Wilson loops

Consider a principal G-bundle � ∶ P → M on a smooth connected manifold M , notnecessarily compact. Choose a base point QP of P and consider a smooth closed pathu∶ [0, 1]→ M in M starting at the point � (QP ) of M . With respect to QP and a smoothconnection A on � ∶ P → M , we denote by Holu,QP (A) ∈ G the holonomy of A along u.For later reference, we recall a description, cf. [Bry95, Appendix to Lecture 3], [KN63,Proof of Proposition II.3.1 p. 69].

Let uP ∶ [0, 1] → P be a lift of u having QP as its starting point, and let !A denotethe connection form associated to A. The solution uG ∶ [0, 1]→ G of the di�erentialequation

!A((uPuG) ) = 0, uG(0) = e, (5.1)

then yield the desired horizontal lift uA∶ [0, 1]→ P of u as uA = uPuG .Now (uPuG) = uPuG + uP uG , the sum being evaluated, for given t ∈ [0, 1], in the

vector space TuP (t)uG (t)P , and

!A(uPuG + uP uG) = !A(uPuG) + !A(uP uG) = Ad−1uG!A(uP ) + u−1G uG , (5.2)

whence (5.1) is equivalent to

uGu−1G = −!A(uP ), uG(0) = e. (5.3)

Wilson loops 10

Then the identityQP Holu,QP (A) = uA(1) = uP (1)uG(1) (5.4)

characterizes Holu,QP (A) ∈ G.The following observation is a Fréchet version of [Hue98, Theorem 2.1].

Lemma 5.1: Given a smooth loop u∶ [0, 1]→ M in M having starting point � (QP ), thehorizontal lift map

Horu,QP ∶ � ⟶ Γ∞(u∗� ), A⟼ uA, (5.5)

with respect to QP is smooth and, at a smooth connection A, the derivative is given by

TAHoru,QP ∶ 1(M, ad� )⟶ C∞([0, 1], g), # ↦ f# ,

f# (t) =t

∫0

#uA(� )(uA(� ))d� , t ∈ [0, 1].(5.6)

To guide the reader through (5.6), we note that, for any � ∈ [0, 1], the notation uA(� )refers to a point of the total space P of � and uA(� ) to a tangent direction at uA(� ).

The idea that underlies the proof we are about to give is the same as that for theproof of [Hue98, Theorem 2.1]. However, we here consider the horizontal lift as amap � → Γ∞(u∗� ) into the space of smooth sections of the induced bundle u∗� ratherthan as a map of the kind � × [0, 1] → P . Moreover, we exploit a very natural andwell-established notion of smoothness instead of the ad hoc concept of a map that is“smooth on every �nite-dimensional submanifold” used in [Hue98].

Proof. Let u∶ [0, 1]→ M be a smooth loop in M having starting point � (QP ), let A bea smooth connection on � , and let uA denote the horizontal lift of u with respect to Aand starting at QP .

The induced bundle u∗� trivializes. We take the connection A as reference connection,use the horizontal lift uA of u relative to A to trivialize the bundle u∗� and arguehenceforth in terms of the total space [0, 1] × G. The trivialization identi�es the spaceΓ∞(u∗� ) of smooth sections of u∗� with the space C∞([0, 1], G) of smooth G-valued mapson [0, 1]. By construction, in the chosen trivialization, the original path u amounts tothe identity path [0, 1] → [0, 1]. Accordingly, given # ∈ 1(M, ad� ), the componentuA+# ∶ [0, 1]→ G into G of the horizontal lift [0, 1]→ [0, 1] × G of the identity path of[0, 1] determines the horizontal lift of u relative to A + # . Thus the horizontal lift maptakes the form

Horu,QP ∶ 1(M, ad� )⟶ C∞([0, 1], G), # ↦ uA+# . (5.7)

In particular, uA is the trivial path at the neutral element e of G.We use the variable t as coordinate on [0, 1]. Let # ∈ 1(M, ad� ). This 1-form,

restricted to the bundle u∗� , now takes, along the factor [0, 1] of the decomposition[0, 1] × G of the total space, the form ℎ#dt for some smooth g-valued function ℎ# on[0, 1] uniquely determined by # , and the path uA+# is the solution of the di�erential

Wilson loops 11

equationuA+#u−1A+# = ℎ# . (5.8)

The sign here is consistent with (5.3). Indeed, in our realization of connections as vectorbundle sections for (4.1), the connectionA+# on � has connection form!A−# ∶ TP → g,where we identify ad� -valued 1-forms on M with G-equivariant 1-forms TP → g asusual. The right-hand side of equation (5.8) depends continuously on the parameter# ∈ 1(M, ad� ). The continuous dependence of solutions of di�erential equations onparameters [Die69, 10.7.1] now implies that the horizontal lift map (5.7) is a continuousmap.

Let s denote a real variable. Di�erentiating the identity

uA+s#u−1A+s# = sℎ# (5.9)

with respect to s yields the identity

))s (uA+s# )u

−1A+s# + uA+s# )

)s (u−1A+s# ) = ℎ# .

However, the path uA is constant whence, for s = 0, the di�erential equation reduces to

))s|||s=0(uA+s# ) = ℎ# .

Interchanging the order of di�erentiation, we write this equation as

ddt

))s|||s=0(uA+s# ) = ℎ# .

Consequently

))s|||s=0(uA+s# )(t) =

t

∫0

ℎ# (� )d� =t

∫0

#uA(� )(uA(� ))d� .

The associated Magnus series provides more insight: Since (5.7) is a continuous map,the horizontal lift with respect to a connection near A lies in an appropriate tubularneighborhood of uA. Hence, when # is su�ciently close to zero, we can write the pathuA+# as

uA+# = exp(WA+# )

for some path WA+# ∶ [0, 1]→ g. Using the familiar identity

ddt (exp(Y (t)))exp(Y (t))

−1 =ead(Y (t)) − 1ad(Y (t))

Y (t),

Wilson loops 12

letting Y = WA+# , we rewrite equation (5.8) as

ead(Y ) − 1ad(Y )

Y = ℎ# . (5.10)

The �rst two terms of a Magnus expansion Y (t) = ∑j≥1 Y (j)(t) of the solution Y of (5.10)read

Y (1)(t) =t

∫0

d� ℎ# (� ), (5.11)

Y (2)(t) = −12

t

∫0

d�1

�1

∫0

d�2 [ℎ# (�1), ℎ# (�2)], (5.12)

and the higher order terms involve higher order commutators of ℎ# evaluated at di�erenttimes. The expansion yields explicit expressions for the derivatives of the horizontallift map (5.7), since every order in the series equates to the order of di�erentiation. Inparticular, this argument con�rms that (5.7) is smooth and that the expression (5.6)yields the derivative.

We return to our surface Σ and maintain the choice of base point Q and of a canonicalsystem u1, v1,… , u� , v� of closed curves in Σ whose homotopy classes generate �1(Σ, Q),cf. Section 2. Consider a principal G-bundle � ∶ P → Σ on Σ and choose a base pointQP of P over the base point Q of Σ. Henceforth the base point QP of P remains �xed.Given a smooth connection A on � , for 1 ≤ j ≤ � , let uA,j and vA,j denote the horizontallifts, in the total space P of � , of uj and vj , respectively, with reference to A and QP . TheWilson loop mapping

�∶ � ⟶ G2� (5.13)

relative to QP and u1, v1,… , u� , v� assigns to a smooth connection A on � the point

�(A) = (Holu1,QP (A),Holv1,QP (A),… ,Holu� ,QP (A),Holv� ,QP (A)) (5.14)

ofG2� [Hue98, (2.6)]. This map depends on the choices made to carry out its construction.In Section 9 below we explore the dependence on these choices.

For q ∈ G, we denote the induced operation of left translation by

Lq ∶ g = TeG ⟶ TqG.

The subsequent result is the Fréchet version of [Hue98, Theorem 2.7], spelled out forthe special case where the base manifold is a closed surface.Theorem 5.2: In the Fréchet topologies, the Wilson loop mapping �∶ � → G2� , cf.(5.13) above, is smooth. At a smooth connection A, the value (5.14) under � being writtenas

�(A) = (a1, b1,… , a� , b� ) ∈ G2� ,

Preparing for the description of the Fréchet constraints 13

the tangent map

TA�∶ TA� ⟶ T�(A)G2� = Ta1G × Tb1G ×⋯ × Ta�G × Tb�G (5.15)

of � sends # ∈ 1(Σ, ad� ) = TA� to

⎛⎜⎜⎝La1 ∫

uA,1

#, Lb1 ∫vA,1

#,… , La� ∫uA,�

#, Lb� ∫vA,�

#⎞⎟⎟⎠, (5.16)

by construction a vector in Ta1G × Tb1G ×⋯ × Ta�G × Tb�G.

Proof. This is a consequence of Lemma 5.1 since, given a smooth closed path u∶ [0, 1]→Σ starting at the base point Q, evaluation at 1 is a smooth map Γ∞(u∗� )→ PQ ≃ G, andthe derivative thereof also comes down to evaluation at 1.

6 Preparing for the description of the Fréchet constraints

To explain the idea as to how the constraints arise, consider a 2-form � on Σ such that,with respect to the volume form volΣ on Σ chosen in Section 2 above, ∫Δ � = ∫Δ volΣ forany disk Δ in Σ. Then ∫Δ(� − volΣ) = 0, for any disk Δ in Σ whence

� = volΣ + d�

for some 1-form � on Σ. Now

∫)Δ

� = ∫Δ

d� = ∫Δ

(� − volΣ) = 0,

for any disk Δ in Σ. However, a 1-form whose integral over any closed contractiblepath vanishes is an exact form. Hence d� = 0 and so � = volΣ. In other words, thefunctionals given by integration over arbitrary disks separate points of 2(Σ,ℝ).

We return to our principal G-bundle � ∶ P → Σ on Σ and maintain the choice ofbase point QP of P over the base point Q of Σ. For convenience, we slightly vary theclassical construction that reduces a principal bundle with connection to the holonomybundle at a chosen base point of the total space, cf., e. g., the reduction theorem [KN63,II.7.1]. Recall the choice of a principal U(1)-connection form !MΣ ∶ TMΣ → iℝ on�MΣ ∶ MΣ → Σ and its lift !MΣ

∶ TMΣ → ℝ to a principal Γℝ-connection form on theprincipal Γℝ-bundle �MΣ

∶ MΣ → Σ and, furthermore, the choice of a base point QMΣ

of MΣ over Q. Let A be a smooth connection on � , and let Phor denote the space ofpaths u∶ [0, 1]→ MΣ in MΣ that start at QMΣ

and are horizontal relative to A, and letev∶ Phor → MΣ denote the evaluation map which sends a path in Phor to its end point.This map is surjective. Indeed, since the principal Γℝ-bundle �MΣ

∶ MΣ → Σ coincideswith its holonomy bundle at QMΣ

with respect to the connection form !MΣ, any point T


of MΣ is the end point of a smooth horizontal path joining QMΣto T .

We de�ne the pre-reduction map �QMΣ ,QP ,A∶ Phor → P of A relative to QMΣand QP as

follows: Given a member u∶ [0, 1]→ MΣ of Phor, that is, a path in MΣ that is horizontalrelative to !MΣ

and starts at the point QMΣof MΣ, let u be the path in Σ obtained by

projecting u into Σ, and let uP ∶ [0, 1]→ P be the unique lift of u that is horizontal forA and has starting point QP ; de�ne the value �QMΣ ,QP ,A(u) to be the end point of uP . Wesay that the pre-reduction map is de�ned on MΣ when it factors through the evaluationmap ev∶ Phor → MΣ; we then write the resulting map as �QMΣ ,QP ,A∶ MΣ → P .

Given a disk Δ in Σ, let aΔ = ∫Δ volΣ denote the area of Δ with respect to volΣ. Recallthe topological characteristic class X� ∈ h of � , see Section 3.Lemma 6.1: Consider a smooth connection A on � ∶ P → Σ. The following are equivalent:

(i) The values of the curvature form curvA of A lie in the center h of the Lie algebra gof G in such a way that curvA = −X� ⋅ volΣ.

(ii) The values of the curvature form curvA of A lie in the center h of the Lie algebra gof G, and the curvature form curvA of A, being accordingly viewed as an ordinaryh-valued 2-form on Σ, satis�es the identity

∫Δ

curvA = −aΔX�⎛⎜⎜⎝= − ∫

Δ

volΣ X�⎞⎟⎟⎠∈ h (6.1)

whenever Δ ⊆ Σ is a disk in Σ.

(iii) For every smooth contractible loop u in Σ starting at some point q of Σ and boundedby a disk Δ in Σ,

Holu,qP (A) = exp(aΔX� ) ∈ H, (6.2)

for any choice of pre-image qP ∈ P of q ∈ Σ.

(iv) The pre-reduction map of A relative to QMΣand QP is de�ned on MΣ and yields a

morphism

Γℝ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ MΣ

�MΣ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Σ

�A⏐⏐↓ �QMΣ ,QP ,A

⏐⏐↓

‖‖‖G ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ P

�←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Σ

(6.3)

of principal bundles with connection in such a way that the following hold:

(a) The homomorphism �(A)∶ F → G which arises as the value of A under theWilson loop mapping (5.13) coincides with the composite

F ⟶ Γ⟶ Γℝ�A⟶ G,

the unlabeled arrows being the obvious maps.


(b) The homomorphism �A satis�es the identity �A(t[r]) = exp(tX� ), for t ∈ ℝ.

Remark 6.2. Since the genus � of Σ is (supposed to be) positive, given a contractible loopu in Σ bounded by a disk, that disk is uniquely determined by u (up to reparametrization).In this sense, the right-hand side of (6.2) depends only on u.

Proof of Lemma 6.1. In view of the observation spelled out at the beginning of thepresent section, the equivalence of (i) and (ii) is immediate. It is also immediate that (ii)implies (iii). We now show that (iii) implies (i).

When G is semisimple, the topological characteristic class X� of the bundle � is zero,and the constraints (6.2) simplify to

Holu,qP (A) = {e} whenever u is a contractible loop in Σ, (6.4)

for a choice q ∈ Σ of starting point of u and of a pre-image qP ∈ P of q.Consider now the case where the center of G has dimension ≥ 1 and, as before, let h

denote the center of the Lie algebra g of G. The group G/H = S ≅ S/D is semisimple,and the bundle projection � ∶ P → Σ factors through the induced map from the orbitmanifold P/H to Σ, necessarily a principal S-bundle, and we denote this bundle by�S ∶ P/H → Σ.

Consider a smooth G-connection A on � that satis�es the constraints (6.2), andlet AS denote the S-connection it induces on �S . Relative to �S , the correspondingconstraints (6.2) say that, given a smooth contractible loop u in Σ, for a choice q ∈ Σ ofstarting point of u and of a pre-image qH ∈ P/H of q ∈ Σ,

Holu,qH (AS) = e ∈ S.

Hence the connection AS on �S is �at. Consequently the curvature form curvA ∈2(Σ, ad� ) of A is central in the sense that the values of curvA lie in the center h of g.

Relative to the decomposition g = h ⊕ s of the Lie algebra g into its center h andits semisimple constituent s = [g, g], the connection form !A of A decomposes as!A = !h + !s and, accordingly, the curvature form curvA of A decomposes as

curvA = d!h + d!s + 12[!s, !s]. (6.5)

Since the values of the curvature form curvA lie in the center h of g, the constituentd!s + 1

2[!s, !s] is zero.Consider a contractible loop u∶ [0, 1] → Σ in Σ and a disk Δ in Σ that bounds u.

Choose a starting point q of u in Σ with w(0) = q and a pre-image qP ∈ P of q. Sincethe loop u is contractible, the summand !s of the decomposition !A = !h + !s does notcontribute to the G-holonomy of u relative to A and qP .

To justify this claim, consider the restriction of � to the disk Δ. This restrictiontrivializes, and we �x a smooth G-equivariant embedding Δ × G → P that covers theembedding Δ → Σ in such a way that the embedding sends the point (q, e) of Δ × Gto the point qP of P . Consider the restrictions �h ∈ 1(Δ, h) and �s ∈ 1(Δ, s), to Δ, of


the respective 1-forms !h and �s. The restriction, to Δ, of the connection form !A of Anow coincides with the sum �A = �h + �s ∈ 1(Δ, g). This sum yields a connection formrelative to the covering group H × S of G = H ×D S.

The closed path u now amounts to the boundary path u∶ [0, 1] → Δ of Δ havingstarting point q ∈ Δ, and the holonomy with respect to A and (qP , e), in H × S, of uis the end point of the path (uH , uS)∶ [0, 1] → H × S that solves the two di�erentialequations

uHu−1H = −�h(u), uH (0) = e ∈ H,uSu−1S = −�s(u), uS(0) = e ∈ S.

Since �s arises from a �at connection and since the path u is contractible, the pathuS is closed. Consequently that holonomy, in H × S, of u is the end point of the path(uH , e)∶ [0, 1]→ H × S.

As in Section 5, let uP ∶ [0, 1]→ P denote a closed lift of u having the point qP as itsstarting point. We can take the composite of the boundary path u∶ [0, 1] → Δ withthe embedding Δ→ Δ × G → P but this is not strictly necessary. Since the summand!s of !A = !h + !s does not contribute to the G-holonomy of u relative to A and qP , inview of (5.3),

Holu,qP (A) = exp⎛⎜⎜⎝− ∫uP

!h

⎞⎟⎟⎠. (6.6)

Since

exp⎛⎜⎜⎝− ∫uP

!h

⎞⎟⎟⎠= exp

⎛⎜⎜⎝− ∫Δ

d!h

⎞⎟⎟⎠= exp

⎛⎜⎜⎝− ∫Δ

curvA⎞⎟⎟⎠, (6.7)

the constraint (6.2) implies that

exp⎛⎜⎜⎝− ∫Δ

curvA⎞⎟⎟⎠= exp(aΔX� ) ∈ H whenever Δ is a disk in Σ. (6.8)

Consider the h-valued 2-form � = curvA+X� ⋅volΣ on Σ. This 2-form has the propertythat exp (∫Δ �) = e ∈ H , for any disk Δ in Σ. Consequently ∫Δ � ∈ �1(H ) ⊆ h, for anydisk Δ in Σ. Now �x a disk Δ. Within Δ, consider a descending sequence {Δ"}" ofdisks whose intersection is a single point. In a suitable coordinate patch, we realizethe disks Δ" as ordinary disks having radius " ∈ ℝ. Now, on the one hand, the mapf ∶ I → �1(H ) given by f (") = ∫Δ" � is continuous and hence constant, and on the otherhand, lim"→0 f (") = lim"→0 ∫Δ" � = 0, whence ∫Δ � = 0. Since Δ is an arbitrary disk, weconclude � = 0, that is, curvA = −X� ⋅ volΣ as asserted.

We now show that (i) implies (iv). Consider �rst the special case where the group Gis abelian, i.e., in our notation, coincides with the compact connected abelian subgroupH of G. Thus � ∶ P → Σ is momentarily a principal H -bundle with an H -connection A


having curvature curvA = −X� ⋅ volΣ.Recall from Section 3 above that the characteristic class X� determines the corre-

sponding homomorphism �� ∶ U(1)→ H via the identity �� (eit) = exp(tX� ) (t ∈ [0, 2�]).Let �A∶ PA → Σ denote the principal H -bundle associated to our reference U(1)-bundle�MΣ ∶ MΣ → Σ via �� , and denote the resulting smooth map from MΣ to PA by �. Thecommutative diagram

U(1) ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ MΣ�MΣ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Σ

��⏐⏐↓ �

⏐⏐↓

‖‖‖H ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ PA

�A←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Σ

(6.9)

displays the resulting bundle map. Our reference connection form !MΣ ∶ TMΣ → iℝ on�MΣ ∶ MΣ → Σ induces an H -connection form on �A having curvature form −X� ⋅ volΣ,and we denote this connection form by (�� )∗!MΣ ∶ TPA → iℝ. The bundles � and �A aretopologically equivalent. Hence there is an H -equivariant di�eomorphism Φ∶ P → PAcovering the identity of Σ that induces an isomorphism �A → � of principal H -bundles.The H -connection form (�� )∗!MΣ induces, via Φ, an H -connection form !� ,Φ∶ TP → iℝon � having, likewise, curvature form −X� ⋅ volΣ. Since this is the curvature form of Aas well, for a suitable bundle automorphism (gauge transformation) � ∶ P → P , pullingback the connection form !A of A via � yields !� ,Φ as !� ,Φ = � ∗!A. By construction, �has the form of a pre-reduction map, now de�ned on MΣ rather than on MΣ, and Φ and� send horizontal paths to horizontal paths. The composite �◦Φ◦� yields the morphism

U(1) ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ MΣ�MΣ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Σ

��⏐⏐↓ �◦Φ◦�

⏐⏐↓

‖‖‖H ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ P

�←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Σ

(6.10)

of principal bundles with connection. Claim (iv) now boils down to the observationthat the composite of �◦Φ◦� with the universal covering projection MΣ → MΣ admitsthe description spelled out as pre-reduction map.

Now we consider the case of a general principal G-bundle � ∶ P → Σ with a G-connection A whose curvature form reads curvA = −X� ⋅ volΣ. Maintain the notationH = G/S ≅ H /D and �H ∶ P/S → Σ introduced in Section 3 above and let A denote theinduced H -connection on the principal H -bundle �H . Let QP/S denote the image of QPin P/S. The above argument shows that, with a slight abuse of the notation �QMΣ ,QP/S ,A,the U(1)-bundle �MΣ ∶ MΣ → Σ being endowed with the reference connection form!MΣ ∶ TMΣ → iℝ, the expression for the pre-reduction map of A relative to QMΣ andQP/S yields a morphism

U(1) ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ MΣ�MΣ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Σ

��H⏐⏐↓ �QMΣ ,QP/S ,A

⏐⏐↓

‖‖‖H ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ P/S

�H←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Σ

(6.11)

The constraints to be imposed on the based gauge orbit manifold 18

of principal bundles with connection. Since the summand !s in the decomposition!A = !h+!s of the connection form!A ofA is the connection form of a �at S-connectionAs on the principal S-bundle P → P/S, the expression for the pre-reduction map of Arelative to QMΣ

and QP yields the unique lift MΣ → P of �QMΣ ,QP/S ,A∶ MΣ → P/S withrespect to the �at connection As and the choice of base points QMΣ

of MΣ and QP of P .The commutative diagram

MΣ

�QMΣ ,QP ,A←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ P⏐⏐↓

⏐⏐↓

MΣ

�QMΣ ,QP/S ,A←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ P/S

(6.12)

displays the situation.Let ur denote the “boundary path” ∏ ujvju−1j v−1j of the 2-cell e2 of Σ; its horizontal lift

ur in MΣ joins QMΣto QMΣ

[r] ∈ MΣ. Let uP,r denote the horizontal lift in P of ur relativeto A having starting point QP . Then QP Holur ,QP (A) coincides with the end point uP,r (1)of uP,r . By construction,

QP �A([r]) = �QMΣ ,QP ,A(QMΣ[r]) = uP,r (1) = QP Holur ,QP (A). (6.13)

By (iii), Holur ,QP (A) = exp(X� ) whence �A([r]) = exp(X� ). Since the homomorphism �Aand the surjective homomorphism Γℝ → U(1) associated with the universal coveringprojection MΣ → MΣ render the diagram

Γℝ�A←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ G

⏐⏐↓

⏐⏐↓

U(1)��H←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ H

commutative, the homomorphism �A satis�es the identity �A(t[r]) = exp(tX� ), for t ∈ ℝ.It is immediate that (iv) implies (i). This completes the proof.

7 The constraints to be imposed on the based gauge orbit mani-fold

In the Atiyah–Bott approach, the moduli space arises by symplectic reduction withrespect to the group of gauge transformations, the vector space 2(Σ, ad� ) being viewedas the dual of the Lie algebra of in�nitesimal gauge transformations, and the map whichassigns to a connection its curvature being interpreted as a momentum mapping. Inthe Fréchet setting, the dual of an in�nite-dimensional vector space is a delicate notion.Below we substitute for that momentum mapping a suitable in�nite family of smoothG-equivariant G-valued maps on the orbit manifold � ,Q = � ,Q⧵� , and we refer tothese maps as constraints. The Wilson loop mapping �, cf. (5.13), is invariant under the


action of the group of based gauge transformations and hence induces a smooth map

�♭∶ � ,Q ⟶ G2� (7.1)

between Fréchet manifolds. With a slight abuse of terminology, we refer to �♭ as Wilsonloop mapping as well. Via evaluation, the choice of generators for the presentation(2.1) of �1(Σ, Q) induces an obvious di�eomorphism from Hom(F , G) onto G2� , and weidentify the two manifolds. The constraints will single out a subspace such that therestriction of the Wilson loop mapping �♭ to that subspace is a homeomorphism ontothe subspaceHomX� (Γ, G)� ofHom(F , G) ≅ G2� introduced at the end of Section 3 above.The subspace of � ,Q thus cut out by the constraints will recover the space of basedgauge equivalence classes of smooth central Yang-Mills connections on � .

For a smooth connection A on � ∶ P → Σ, we denote by [A] its class in the orbitmanifold � ,Q = � ,Q⧵� . Consider the maps

cu,Δ,QP ∶ � ,Q ⟶ G, [A]⟼ Holu,QP (A) exp(aΔX� ), (7.2)

as u ranges over contractible loops in Σ having the base point Q of Σ as starting pointand bounded by disks Δ in Σ. Lemma 5.1 entails that these maps are smooth. Let

� ,Q ⊆ � ,Q

be the subspace of � ,Q cut out by the constraints

cu,Δ,QP ([A]) = e ∈ G. (7.3)

Lemma 6.1 entails that � ,Q recovers the space of smooth based gauge equivalenceclasses of smooth central Yang–Mills connections on � and that the values of therestriction of the Wilson loop mapping (7.1) to � ,Q lie in HomX� (Γ, G)� . We endow� ,Q with the topology induced from the Fréchet topology of � ,Q and refer to thistopology as the Fréchet topology on � ,Q .Theorem 7.1: The space � ,Q is non-empty, the values of the restriction of the Wilsonloop mapping �♭, cf. (7.1), to the subspace � ,Q of � ,Q lie in HomX� (Γ, G)� , and theresulting map

�♯∶ � ,Q ⟶ HomX� (Γ, G)� , (7.4)

de�ned on � ,Q endowed with its Fréchet topology, is aG-equivariant homeomorphism.Furthermore, abstractly, the space � ,Q depends only on the bundle � but not on thechoices made to carry out its construction. Finally, with respect to the G-orbit strati�cationson both sides, �♯ is an isomorphism of strati�ed spaces which, on each stratum, restricts toa di�eomorphism onto the corresponding stratum of the target space.

We split the proof of Theorem 7.1 into Lemmata 7.2 and 7.5. In Section 9 below, weestablish the “Furthermore” statement claiming the independence of the choices.Lemma 7.2: The map �♯∶ � ,Q → HomX� (Γ, G)� given as (7.4) above is a bijection.More precisely: For every member � of HomX� (Γ, G)� , there is a connection A on � rep-


resenting a point of � ,Q , unique up to based gauge transformations, such that the

composite Γ → Γℝ�A⟶ G, with the injection Γ → Γℝ, of the constituent �A∶ Γℝ → G

in the associated pre-reduction map (6.3) of A relative to QMΣand QP in Lemma 6.1 (iv)

coincides with � ∶ Γ→ G.

Proof. Given � ∈ HomX� (Γ, G), via the extended homomorphism � ∶ Γℝ → G givenby � (t[r]) = exp(tX� ), for t ∈ ℝ, the Γℝ-connection form !MΣ

∶ TMΣ → ℝ on theprincipal reference Γℝ-bundle �MΣ

∶ MΣ → Σ and the Maurer-Cartan form of G yielda G-connection form !� ∶ TP� → g on the associated principal G-bundle �� ∶ P� ∶=MΣ ×� G → Σ. Below we sometimes write the resulting smooth principal G-connectionon �� as A� . By construction, the derivative d� ∶ ℝ → g of � ∶ Γℝ → G is given byd� (2� ) = X� ∈ g. Since curv!MΣ = −2�volΣ ∈ 2(Σ,ℝ), we conclude that curv!� =−X� ⋅ volΣ ∈ 2(Σ, h).

In view of the de�nition of the connected component HomX� (Γ, G)� of HomX� (Γ, G),cf. Section 3 above, the bundles � and �� have the same topological type, that is, areisomorphic. Since the principal G-bundles � and �� are isomorphic, there exists a(vertical) di�eomorphism Φ∶ P → P� that induces an isomorphism � → �� of principalG-bundles. Let QP � = [QMΣ

, e] ∈ P� . A suitable gauge transformation of �� carries Φ(QP )to QP � . Hence we can arrange for the di�eomorphism Φ to be based in the sense thatΦ(QP ) = QP � .

Pulling back the connection from !� on �� along Φ yields a G-connection !� ,Φ on� . The curvature of !� ,Φ still equals −X� ⋅ volΣ. Another choice Φ∶ P → P� of baseddi�eomorphism inducing an isomorphism � → �� of principal G-bundles yields a G-connection form !� ,Φ on � in such a way that !� ,Φ and !� ,Φ are based gauge equivalent.Lemma 6.1 now implies the claim.

Remark 7.3. The proof of Lemma 7.2 essentially involves the argument for [AB82, Propo-sition 6.16], with the connection !MΣ

on �MΣ∶ MΣ → Σ substituted for the harmonic

Yang-Mills connection in [AB82, § 6]. As noted in the introduction, the novelty ofour approach resides in the characterization of the space � ,Q merely in termsof constraints imposed on the Fréchet manifold of smooth based gauge equivalenceclasses of all smooth G-connections on � .

Lemma 7.2 entails in particular that the subspace � ,Q of � ,Q cut out by theconstraints (7.3) is non-empty.Remark 7.4. By Theorem 5.2, the Wilson loop mapping (5.13) is smooth, and hence theWilson loop mapping (7.1) is smooth. Consequently �♯∶ � ,Q → HomX� (Γ, G)� , cf.(7.4), is a continuous bijection onto the compact subspace HomX� (Γ, G)� of the smoothcompact manifold Hom(F , G). By the strong Uhlenbeck compactness theorem [Uhl82],for any sequence {Aj} of smooth Yang-Mills connections with constant curvature (infact, the hypothesis that the curvature be uniformly L2-bounded su�ces), there exists asubsequence {Ai} of smooth Yang-Mills connections and a sequence {�i} of smoothgauge transformations such that the sequence {�i ⋅ Ai} converges uniformly in the C∞-topology with all derivatives, that is, in the Fréchet topology, to a smooth connection A.


Hence, in the Fréchet topology, the space � ,Q is as well compact. Consequentlythe map �♯ is a homeomorphism as asserted. However, the Fréchet slice analysis in theproof of Lemma 7.5 below avoids that theorem. On the other hand, that slice analysisrecovers the statement of the Uhlenbeck compactness theorem in the case at hand.Lemma 7.5: Themap �♯∶ � ,Q → HomX� (Γ, G)� given as (7.4) above is aG-equivarianthomeomorphism that is compatible with the G-orbit strati�cations on both sides.

Proof. First we show that the bijection �♯ is a local homeomorphism.Consider a smooth connection A on � that represents a point of � ,Q . Since the

connection A is central, the operator dA of covariant derivative turns ∗(Σ, ad� ) into acochain complex. We denote the resulting cohomology by H∗

A and we use the notation

Z 0A = Z

0A(Σ, ad� ) = ker(dA∶ 0(Σ, ad� )→ 1(Σ, ad� )) (7.5)

Z 1A = Z

1A(Σ, ad� ) = ker(dA∶ 1(Σ, ad� )→ 2(Σ, ad� )) (7.6)

B1A = B1A(Σ, ad� ) = im(dA∶ 0(Σ, ad� )→ 1(Σ, ad� )) (7.7)

B2A = B2A(Σ, ad� ) = im(dA∶ 1(Σ, ad� )→ 2(Σ, ad� )). (7.8)

In particular, the cokernel of dA∶ 1(Σ, ad� )→ 2(Σ, ad� ) yields the associated secondcohomology group H2

A. We choose a complement 2A ⊆ 2(Σ, ad� ) for B2A ⊆ 2(Σ, ad� )

such that the direct sum decomposition 2(Σ, ad� ) = B2A ⊕ 2A is a homeomorphism

in the Fréchet topology. Then the projection 2(Σ, ad� )→ H2A, restricted to 2

A, is anisomorphism of �nite-dimensional vector spaces.

Let SA ⊆ � be a Fréchet slice for the action of the group � of gauge transformationson � . Then

TA� = B1A ⊕ TA(SA). (7.9)

Consider the smooth map

ℎ∶ SA → B2A, A′ ↦ curvA′ mod 2

A, A′ ∈ SA. (7.10)

The derivative TA′ℎ∶ TA′SA → B2A at the point A′ of SA is the composite

TA′SAdA′ |⟶ 2 proj

⟶ B2A. (7.11)

By construction, the derivative TAℎ = dA at the point A is surjective. Our aim is toconclude that, for an open neighborhood U of A in SA, the pre-image ℎ−1(0) ∩ U is asmooth manifold, necessarily �nite-dimensional, having as tangent space at A the �nite-dimensional vector space Z 1

A ∩TASA ≅ H1A. To this end, we use the Nash–Moser theorem

to show that ℎ is a submersion at A. Hence we need to verify that the derivative TA′ℎis surjective in an open neighborhood of A in SA and that the family of right inverses istame smooth.

Let A + � be a point of � , where � ∈ 1(Σ, ad� ), and suppose that A + � ∈ SA.


Consider the operators

L� ∶ 1(Σ, ad� )⟶ 2(Σ, ad� ), � ↦ dA+�� = dA� + [� ∧ �],T� = L�◦L∗0 = dA+� d

∗A∶ 2(Σ, ad� )⟶ 2(Σ, ad� ).

The operator T� satis�es the identity

T� (�) = △A(�) + [� ∧ d∗A(�)] (� ∈ 2(Σ, ad� )).

In particular, T0 is the covariant Laplacian on 2(Σ, ad� ) relative to A and so is an ellipticoperator of index zero. Since the symbol map is continuous, for small � , the deformedoperator T� is still elliptic and has index zero. We claim that the induced operator

B2Ainj⟶ 2(Σ, ad� )

T�⟶ 2(Σ, ad� )proj⟶ B2A (7.12)

is invertible. In view of the Fredholm alternative, since T� has index zero, it is enoughto show that ker T� = H2

A. The identity (dA d∗A�, �)L2 = ‖d∗A�‖2L2 shows that H2A = ker T0 =

ker dA d∗A coincides with the kernel of d∗A. Hence ker T0 ⊆ ker T� . The converse inclusionfollows from the upper semi-continuity of the dimension of the kernel of an ellipticoperator [H07, Corollary 19.1.6], that is, dim ker T� ≤ dim ker T0 for small � . Hence theinduced operator (7.12) is invertible and, with a slight abuse of notation, we denotethis operator on B2A by T� as well. The inverse operator T −1

� arises from the Green’soperator and, by [Ham82, II.3.3.3 Theorem p. 158], is therefore tame.

Since L�◦(L∗0◦T −1� ) = IdB2A for small � , we obtain the tame family {L∗0◦T −1

� }� of rightinverses for the family {TA+�ℎ}� . Hence the Nash–Moser inverse function theoremimplies that ℎ is a submersion nearA. In other words, there exists an open neighborhoodU ofA in SA such that the pre-image ℎ−1(0)∩U is a smooth manifoldMA having as tangentspace at A the �nite-dimensional vector space Z 1

A ∩TASA ≅ H1A. Since the canonical map

2A → H2

A is a linear isomorphism, the manifold MA contains connections A′ ∈ SA thatare C∞-close to A and have curvature map ∗ curvA′ ∶ P → g constant on the holonomybundle of A.

The action of the group � of gauge transformations on � restricts to a smoothaction of the stabilizer subgroup ZA of A on MA. Moreover, in a ZA-neighborhood ofA in MA, the projection to � ,Q is a di�eomorphism onto a smooth �nite-dimensionalZA-submanifold MA,Q of � ,Q . The restriction to MA,Q of the Wilson loop mapping�♭∶ � ,Q → G2� given as (7.1) above yields a smooth map

�♮∶ MA,Q → Hom(F , G) ≅ G2� (7.13)

having injective tangent map

T[A]�♮∶ T[A]MA,Q → T�(A)Hom(F , G)

at the point [A] and, up to left translation by �(A) ∈ Hom(F , G) ≅ G2� , that tangent


map takes the formT[A]�♮∶ T[A]MA,Q → g2� . (7.14)

We shall shortly justify the injectivity claim. Passing, if need be, to a smaller opensubmanifold of MA,Q containing the point [A], we can arrange for the image M�(A),Q ∶=�♮(MA,Q) ⊆ Hom(F , G) to be an ordinary (embedded) submanifold of Hom(F , G).

Since the connection A is central, the homomorphism �A = �(A)∶ F → G descendsto a homomorphism �1 → G/H ≅ S and hence induces, via the adjoint action of G,a �1-module structure on g. We write the resulting �1-module as gA or g�A . Then thecellular cochain complex of Σ with local coe�cients is de�ned and takes the form

g�Ad�A←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ g2��A

d�A←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ g�A . (7.15)

Twisted integration yields a morphism of cochain complexes

0(Σ, ad� )dA←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ 1(Σ, ad� )

dA←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ 2(Σ, ad� )⏐⏐↓ ∫

⏐⏐↓

⏐⏐↓

g�Ad�A←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ g2��A

d�A←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ g�A

(7.16)

inducing an isomorphism on cohomology, in particular an isomorphism H1A(Σ, ad� )→

H1(Σ, g�A). Given � ∶ Γ→ G such that the value � ([r]) lies in H (the connected compo-nent of the center of G), we use the notation

Z 0� = Z

0� (Σ, g) = ker(d� ∶ C

0(Σ, g)→ C1(Σ, g)) (7.17)Z 1� = Z

1� (Σ, g) = ker(d� ∶ C

1(Σ, g)→ C2(Σ, g)) (7.18)B1� = B

1� (Σ, g) = im(d� ∶ C

0(Σ, g)→ C1(Σ, g)) (7.19)B2� = B

2� (Σ, g) = im(d� ∶ C

1(Σ, g)→ C2(Σ, g)). (7.20)

By Theorem 5.2, up to left translation by the value �(A) ∈ Hom(F , G) ≅ G2� , thetwisted integration mapping ∫ ∶ 1(Σ, ad� )→ g2��A is the derivative (5.15) of the Wilsonloop mapping � at the point A of � . Hence the image of 1

A = T[A]MA,Q underT[A]�♮∶ T[A]MA,Q → g2� is a linear subspace of g2��A which lies in the subspace Z 1

�A ⊆g2��A of 1-cocycles relative to d�A ∶ g2��A → g�A and maps, under the projection fromZ 1�A to the cohomology group H1(Σ, g�A), isomorphically onto that cohomology group.

Consequently, T[A]�♮ is injective and thus �♮ is a di�eomorphism MA,Q → M�(A),Q ontoits image M�(A),Q , at least in a neighborhood of [A] in MA,Q .

The evaluation map evQP ∶ � → G identi�es the stabilizer ZA of Awith the stabilizerZ�A ⊆ G of the value �A = �(A) ∈ M�(A),Q . Thus the di�eomorphism �♮∶ MA,Q → M�(A),Qextends to a di�eomorphism

�♮∶ G ×ZA MA,Q → G ×Z�A M�(A),Q ⊆ Hom(F , G) ≅ G2� . (7.21)

Reconstruction of the space of gauge eqivalence classes of all Yang-Millsconnections 24

Consequently the restriction

(G ×ZA MA,Q) ∩ � ,Q → (G ×Z�A M�(A),Q) ∩ HomX� (Γ, G)� (7.22)

is a homeomorphism. By construction, this homeomorphism is the restriction, to theintersection (G ×ZA MA,Q) ∩ � ,Q , of the map �♯ given as (7.4) above.

Since our reasoning works for any point [A] of � ,Q , we conclude that �♯ is ahomeomorphism. Moreover, the above Fréchet slice analysis shows that �♯ is compatiblewith the orbit type strati�cations on both sides and that, on each stratum, �♯ restrictsto a di�eomorphism onto the corresponding stratum of the target space.

Remark 7.6. Let G = U(1) and endow our reference U(1)-bundle �MΣ ∶ MΣ → Σ in-troduced in Section 3 above with a connection A whose curvature form curvA equals−2�ivolΣ, that is, represents a point of �MΣ ,Q . Under these circumstances, thepre-reduction map of A relative to QMΣ

and QMΣ induces a gauge transformation of �MΣ

that identi�es the reference connection chosen in Section 3 with A. In particular, thespace �MΣ ,Q reduces to a point. This recovers the fact that a complex line bundleon Σ has a harmonic connection that is unique up to gauge transformations.

8 Fréchet-reconstruction of the space of gauge eqivalenceclasses of all Yang-Mills connections

In this section we reconstruct the space of gauge equivalence classes of Yang-Millsconnections on Σ within our Fréchet setting. Our procedure parallels the correspond-ing classical description of the topological decomposition of the space of Yang-Millsconnections given in [AB82, Section 6].

The obvious surjection can∶ F → Γ de�ned on the free group F on the chosengenerators x1, y1,… , x� , y� of the fundamental group �1(Σ, Q) extends to the surjectivehomomorphism

ℝ × F ⟶ Γℝ, (t, w)⟼ t[r] can(w), t ∈ ℝ, w ∈ F . (8.1)

This surjective homomorphism, in turn, induces an injection ofHom(Γℝ, G) intoHom(ℝ×F , G). The restriction to the central copy of ℝ in Γℝ generated by [r] of a homomor-phism '∶ ℝ × F → G is a 1-parameter subgroup of G and therefore of the kindt ↦ exp(tX') (t ∈ ℝ) for some uniquely determined member X' of g. Hence the assign-ment to ' ∈ Hom(ℝ × F , G) of (X' , '(x1), '(y1),… , '(x� ), '(y� )) injects Hom(ℝ × F , G)into g × G2� and hence yields an injection

Hom(Γℝ, G)⟶ g × G2� . (8.2)

We endow Hom(Γℝ, G) with the induced topology.Let X ∈ g. We denote by GX ⊆ G the stabilizer of X under the adjoint action,

necessarily a closed connected subgroup of G, cf. [GS84, §32 Theorem 32.16 p. 259], and


we denote by HomX (Γℝ, G) the space, if non-empty, of homomorphisms � from Γℝ to Ghaving the property that � (t[r]) = exp(tX ). Since the copy of ℝ in Γℝ generated by [r]is central, the values of any ' ∈ HomX (Γℝ, G) lie in GX , that is, the canonical injectionmap from HomX (Γℝ, GX ) to HomX (Γℝ, G) is a homeomorphism. Given an adjoint orbit ⊆ g, we denote by Hom(Γℝ, G) the space of homomorphisms � from Γℝ to G havingthe property that � (t[r]) = exp(tY ), for some Y ∈ . The space Hom(Γℝ, G) is thetotal space of a �ber bundle over having compact �ber, the �ber over Y ∈ beingthe compact space HomY (Γℝ, GY ), and so Hom(Γℝ, G) is therefore itself compact.

We return to our principal G-bundle � ∶ P → Σ. Given an adjoint orbit , wedenote by Hom(Γℝ, G)� the subspace, if non-empty, of Hom(Γℝ, G) that consists ofthe homomorphisms '∶ Γℝ → G in Hom(Γℝ, G) which have the property that theassociated principal G-bundle �M ×' G is topologically equivalent to � . When consistsof a single point X , we use the notation HomX (Γℝ, G)� . We then denote by Hom(Γℝ, G)�the disjoint union of the spaces of the kind Hom(Γℝ, G)� , where ranges over theadjoint orbits such that Hom(Γℝ, G)� is non-empty. In terms of the notation thusestablished, the restriction mapping

HomX� (Γℝ, G)⟶ HomX� (Γ, G) (8.3)

yields a homeomorphism from HomX� (Γℝ, G)� onto the space HomX� (Γ, G)� used inTheorem 7.1 above.

Recall from the proof of [AB82, Theorem 6.7] that Yang–Mills connections on �arise from central Yang–Mills connections on certain subbundles. Indeed, considera connection A on � that satis�es the Yang–Mills equation. Then the G-equivariantmap ∗ curvA ∶ P → g is constant along horizontal paths. Let X =∗ curvA(QP ) andPA,X = {p ∈ P ∶∗ curvA(p) = X}. The bundle projection � restricts to a principalGX -bundle projection �X ∶ PA,X → Σ, and the connection A restricts to a connectionAX on �X having the property that curvAX = −X ⋅ volΣ.

For X ∈ g, denote by � ,Q,X the subspace, if non-empty, of � ,Q that consists ofthe points [A] characterized as follows: There exists a reduction of structure group�X ∶ PX → Σ of � to GX that is based in the sense that, relative to the associatedinjection of PX into the total space P of � , the base point QP of P lies in PX , suchthat some connection A representing the class [A] restricts to a connection AX on �Xhaving the property that curvAX = −X ⋅ volΣ. The residual G-action then correspondsto the adjoint action on g, i.e., x ∈ G maps � ,Q,X to � ,Q,AdxX . This leads usto de�ne, for a given adjoint orbit , the G-invariant subspace � ,Q, of � ,Q tobe the union of the spaces � ,Q,X for X ∈ . The above argument shows that thespace � ,Q ⊆ � ,Q of based gauge equivalence classes of Yang–Mills connectionson � is the disjoint union of the spaces of the kind � ,Q,, where ranges overthose adjoint orbits such that � ,Q, is non-empty. The space � ,Q is our Fréchetversion of the space of based gauge equivalence classes of Yang-Mills connections on � .Likewise we take the Fréchet version G,Σ,Q of the space of based gauge equivalenceclasses of all Yang-Mills connections over Σ relative to G to be the disjoint union of thespaces of the kind � ,Q , where � ranges over all topological types of G-bundles on


Σ. By construction, each connected component of � ,Q is characterized by Fréchetconstraints, and the same is true of G,Σ,Q . In this sense, the spaces � ,Q andG,Σ,Q are entirely characterized by Fréchet constraints. We say that a space of thekind � ,Q, and, likewise, one the kind Hom(Γℝ, G)� is de�ned when its de�ningcondition is non-vacuously satis�ed.Theorem 8.1: Given an adjoint orbit in g, the space � ,Q, is de�ned if and only ifthe space Hom(Γℝ, G)� is de�ned and, when this happens to be the case, the Wilson loopmapping �♭∶ � ,Q → G2� , cf. (7.1), induces a G-equivariant homeomorphism

� ,Q, ⟶ Hom(Γℝ, G)� . (8.4)

Hence the Wilson loop mapping �♭ induces a G-equivariant homeomorphism

� ,Q ⟶ Hom(Γℝ, G)� . (8.5)

Furthermore, as � ranges over all topological types of principal G-bundles on Σ, the Wilsonloop mapping �♭ induces a homeomorphism

G,Σ,Q ⟶ Hom(Γℝ, G). (8.6)

Moreover, abstractly the spaces � ,Q and G,Σ,Q do not depend on the choicesmade to carry out their construction. Finally, each of the above homeomorphisms iscompatible with the G-orbit type strati�cations and is, hence, an isomorphism of strati�edspaces.

Proof. ConsiderX ∈ g such thatHomX (Γℝ, G) is non-empty and choose � ∈ HomX (Γℝ, G).We denote by HomX (Γℝ, G)� the connected component of HomX (Γℝ, G) that contains� . Since [r] is central in Γℝ, the values of any ' ∈ HomX (Γℝ, G)� lie in GX whence thecanonical injection from HomX (Γℝ, GX )� to HomX (Γℝ, G)� is a homeomorphism. Theassociated bundle

�� = �M ×� GX ∶ P� = M ×� GX ⟶ Σ (8.7)

is a principal GX -bundle, and the canonical restriction map

HomX (Γℝ, GX )� ⟶ HomX (Γ, GX )�� (8.8)

is a homeomorphism. We take the point Q� = [(QMΣ, e)] as the base point of P� . By

Theorem 7.1,�♯∶ �� ,Q ⟶ HomX (Γ, GX )�� (8.9)

is a G-equivariant homeomorphism.Extension of the structure group GX to G yields the principal G-bundle

�� ,G ∶ P� ×GX G → Σ (8.10)

on Σ. The action �� ,G ×�� ,G → �� ,G of the group �� ,G of smooth gauge transforma-tions of �� ,G on the space �� ,G of smooth connections on �� ,G induces a G-equivariant


homeomorphism

G ×GX �� ,Q ⟶ �� ,G ,,Q .

Likewise, the conjugation action

G × Hom(Γ, G)⟶ Hom(Γ, G), (x, � )↦ x� , x� (y) = x� (y)x−1, x ∈ G, y ∈ Γ, � ∈ Hom(Γ, G),

of G on Hom(Γ, G) induces a G-equivariant homeomorphism

G ×GX HomX (Γ, GX )�� ⟶ Hom(Γ, G)�� ,G .

Consequently, the Wilson loop mapping induces a G-equivariant homeomorphism

�� ,G ,Q, ⟶ Hom(Γℝ, G)�� ,G

of the kind (8.4).The rest of the proof is straightforward. We leave the details to the reader.

Remark 8.2. Theorem 8.1 is a based version of [AB82, Theorem 6.7], but phrased in ourFréchet setting. We can paraphrase Theorem 8.1 by saying that our Fréchet constructionprecisely recovers the Atiyah–Bott spaces of based gauge equivalence classes of Yang-Mills connections.

To spell out what Theorems 7.1 and 8.1 entail for the ordinary moduli spaces ofYang-Mills connections, we use the notation

� = G⧵� ,Q (8.11)� = G⧵� ,Q (8.12)

G,Σ = G⧵G,Σ,Q (8.13)

and

RepX� (Γℝ, G)� = G⧵HomX� (Γℝ, G)� (≅ G⧵HomX� (Γ, G)� ) (8.14)

Rep(Γℝ, G)� = G⧵Hom(Γℝ, G)� (8.15)Rep(Γℝ, G) = G⧵Hom(Γℝ, G). (8.16)

Plainly, the homeomorphism �♯, cf. (7.4), induces homeomorphisms

� ⟶ RepX� (Γℝ, G)� , (8.17)

� ⟶ Rep(Γℝ, G)� , (8.18)G,Σ ⟶ Rep(Γℝ, G), (8.19)

and these homeomorphism preserve the strati�cations on both sides. In view of Theo-rems 7.1 and 8.1, the spaces � , � , and G,Σ do not depend on the choicesmade to construct the Wilson loop mapping. The orbit spaces � , � , and

Dependence on the Choices 28

G,Σ recover, respectively, the moduli space of gauge equivalence classes of cen-tral Yang-Mills connections on � , that of Yang-Mills connections on � , and that of allYang-Mills connections on Σ relative to G.

9 Dependence on the Choices

The Wilson loop mapping �♭, cf. (7.1), and the resulting map �♯, cf. (7.4), dependon the choices of the point QP of the total space P of � and the canonical systemu1, v1,… , u� , v� of curves in Σ. Let u1, v1,… , u� , v� be another canonical system ofcurves in Σ that start at Q. Their based homotopy classes x1, y1,… , x � , y� constitutea system of generators of the fundamental group �1 = �1(Σ, Q) of Σ at Q. De�ne therelator by r = [x1, y1] ⋅ ⋯ ⋅ [x � , y�]. The association

(u1, v1,… , u� , v� )⟶ (u1, v1,… , u� , v� )

induces an automorphism of Γ and one of �1 that make the diagram

0 ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ ℤ⟨[r]⟩ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Γ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ �1 ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ 1‖‖‖

⏐⏐↓

⏐⏐↓

0 ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ ℤ⟨[r]⟩ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ Γ ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ �1 ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ 1

(9.1)

commutative. Since the closed paths [u1, v1] ⋅… ⋅ [u� , v� ] and [u1, v1] ⋅… ⋅ [u� , v� ] acquirethe same orientation, the two members [r] and [r] of Γ coincide. This is a consequenceof the fact that, up to within isomorphism, the group Γ is the uniquely determinedSchur cover of �1 = �1(Σ, Q). Hence the identity symbol in (9.1) makes sense. Thatautomorphism, in turn, determines a member of the mapping class group of Σ. Theseobservations yield a proof of the following.Proposition 9.1: In the situation ofTheorem 7.1 above, the base pointQP of the total spaceP of � being �xed, two isomorphisms � ,Q → HomX� (Γ, G)� of strati�ed spaces thatarise from twoWilson loopmappings of the kind (7.4) associated to distinct choices of canon-ical systems of curves in Σ di�er by a homeomorphism HomX� (Γ, G)� → HomX� (Γ, G)� ,necessarily an automorphism of strati�ed spaces, induced by an automorphism of Γ thatrepresents a member of the mapping class group of Σ.Proposition 9.2: In the situation of Theorem 7.1 above, the base point Q ∈ Σ being �xed,two isomorphisms of strati�ed spaces of the kind (7.4) that arise from the Wilson loopmappings associated to the same choice of canonical system of curves in Σ but one relativeto QP and the other one relative to QP ⋅ x for some x ∈ G di�er by the homeomorphismHomX� (Γ, G)� → HomX� (Γ, G)� induced by conjugation with x ∈ G, necessarily anautomorphism of strati�ed spaces.Proposition 9.3: In the situation of Theorem 7.1 above, with a new choice of basepoints Q′ of Σ and Q′

MΣof MΣ, a choice of a homotopy class of paths in MΣ from QMΣ

to Q′MΣ

induces an isomorphism of strati�ed spaces from HomX� (�1(MΣ, QMΣ), G)� onto

Stratified Symplectic Structure 29

Hom(�1(MΣ, Q′MΣ), G)� .

The three propositions imply the independence statement in Theorem 7.1 above aswell as that in Theorem 8.1 above.

10 Stratified Symplectic Structure

Recall that, by de�nition, the algebra C∞(Hom(Γℝ, G)� ) of Whitney smooth functions onHom(Γℝ, G)� (relative to the embedding into Hom(F , G)) is the algebra of continuousfunctions onHom(Γℝ, G)� that are restrictions of smooth functions on the ambient spaceHom(F , G) ≅ G2� ; cf. [Whi34], [Whi57]. Let C∞(Rep(Γℝ, G)� ) = C∞(Hom(Γℝ, G)� )Gbe the algebra of G-invariant functions in C∞(Hom(Γℝ, G)� ), viewed as an algebraof continuous functions on the orbit space Rep(Γℝ, G)� in an obvious manner. Thealgebras C∞(Hom(Γℝ, G)� ) and C∞(Rep(Γℝ, G)� ) are what has come to be known as asmooth structure on Hom(Γℝ, G)� and Rep(Γℝ, G)� , respectively.

Recall that a di�erential space is a topological space T together with a subalgebraC of the algebra of continuous functions on T subject to certain conditions [Sik72]. Asmooth structure on a topological space T is an algebra of functions that turns T into adi�erential space, subject to suitable additional conditions depending on context.

As � ,Q is a Fréchet manifold, it acquires a natural algebra C∞(� ,Q) of smoothfunctions on � ,Q . Furthermore, since the Wilson loop mapping �∶ � ,Q → Hom(F , G),cf. (7.1), is smooth, the image �∗(C∞(Hom(F , G))) ⊆ C∞(� ,Q) yields a smooth structureon � ,Q as well, but not every smooth function on B� ,Q arise as the pull back, underthe Wilson loop mapping, of a smooth function on Hom(F , G). When we consider thespace � ,Q , the situation changes:Theorem 10.1: The smooth structure on � ,Q that arises via restriction of func-tions in C∞(� ,Q) coincides with the smooth structure that arises via restriction of func-tions in �∗(C∞(Hom(F , G))). In other words, the Wilson loop mapping �♯∶ � ,Q →HomX� (Γ, G)� , cf. (7.4), is an isomorphism of smooth spaces.

Proof. This is a consequence of the Fréchet slice analysis in the proof of Lemma 7.5.

Accordingly, we take the smooth structure C∞(� ,Q) on � ,Q to be thealgebra of continuous functions on � ,Q that arise from ordinary smooth functionson Hom(F , G) via pull back under the Wilson loop mapping. Moreover, we take thesmooth structure C∞(� ) on � to be the algebra

C∞(� ) = C∞(� ,Q)G

of G-invariant functions, viewed as an algebra of continuous functions on � =G⧵� ,Q in the obvious way. By construction, the homeomorphism �♯, cf. (7.4),induces isomorphisms

C∞(� ,Q)⟶ C∞(Hom(Γℝ, G)� ) (10.1)C∞(� )⟶ C∞(Rep(Γℝ, G)� ) (10.2)

Stratified Symplectic Structure 30

of real algebras.The extended moduli space construction developed in [Hue95b], [Jef97], [HJ94], see

[Hue01] for a leisurely introduction and more references, yields the strati�ed symplecticstructure

(C∞(Rep(Γℝ, G)� ), { ⋅ , ⋅ })

on Rep(Γℝ, G)� determined by the data. On each stratum of Rep� (Γ, G), that strati�edsymplectic structure comes down to a symplectic structure. On the other hand, thesymplectic structure that was constructed in [AB82] from the familiar expression

!(�, �) = ∫Σ

� ∧ �, �, � ∈ 1(Σ, ad� ) = TA(� ), A ∈ � ,

by in�nite dimensional symplectic reduction involving Sobolev space techniques in-duces a symplectic structure on the top stratum of � ; here the exterior product �∧�of the two ad� -valued 1-forms � and � on Σ is evaluated in terms of the inner producton g. With respect to these symplectic structures, the Wilson loop mapping induces asymplectomorphism from the top stratum of Rep� (Γ, G) onto the top stratum of � .Suitably interpreted, the Wilson loop mapping also induces a symplectomorphism fromeach stratum of Rep� (Γ, G) onto the corresponding stratum of � .Remark 10.2. We can even push a bit further: In [AMM98], the authors reworked theextended moduli space construction and introduced the terminology quasi-hamiltonianG-space. Via the Wilson loop mapping (7.1), the 2-form on Hom(F , G) constructed in[HJ94], [Hue95b], [Jef97], pulls back to a 2-form on � ,Q but, beware, this 2-form is notclosed. We can then view the composite

r◦�♭∶ � ,Q ⟶ G

as a Lie group valued momentum mapping with respect to the pulled back 2-formon � ,Q . Then � ,Q together with the G-action on � ,Q and the map r◦�♭ is a quasi-hamiltonian G-space except that on � ,Q (as opposed toHom(F , G)) the non-degeneracycondition is to be interpreted appropriately. Reduction with respect to the structuregroup G then yields the moduli space of central Yang-Mills connections as a strati�edsymplectic space. Suitably interpreted, we can view � ,Q as the reduced space relativeto the group of based gauge transformations. This is, perhaps, not strictly correct– we did not make precise the momentum mapping for the group of based gaugetransformations – and the purported space that arises by reduction relative to the groupof based gauge transformations would have to be a proper subspace of � ,Q . However,to use some physics terminology, we are interested only in what happens on shell, thatis, on the subspace � ,Q , and we can interpret Theorem 7.1 by saying that, on shell,the Fréchet manifold � ,Q has the appropriate behavior as if it did arise by reductionrelative to the group of based gauge transformations.

Fréchet Slices 31

11 Fréchet Slices

Our references for terminology and notation in the framework of in�nite-dimensionaldi�erential geometry are [Ham82] for the Fréchet case and [Nee06] for the locallyconvex setting. Thus we freely use the terms smooth manifold, Lie group, Fréchetstructure etc. without further explanation.

11.1 Principal Lie subgroups

In �nite dimensions, for any Lie subgroup H ⊆ G, the space of left cosets G/H carriesthe structure of a smooth manifold that turns the natural projection � ∶ G → G/H intoa right principal H -bundle. Recall that the local di�eomorphism

�∶ g = k ⊕ h → G, (X, Y )↦ exp(X ) exp(Y ), X ∈ k, Y ∈ h, (11.1)

plays an important role in the construction of charts on G/H . This concept needs somere�nement for general locally convex manifolds since in general � fails to be a localdi�eomorphism (and the exponential map need not exist in the �rst place).Proposition 11.1 ([GN, Proposition 7.1.21]): Let G be a Lie group and H ⊆ G asubgroup. Given a Lie group structure onH , the statements (1) and (2) below are equivalent:

(1) The group H is a splitting Lie subgroup, and there exists a unique smooth structureon the space of left cosets G/H such that the canonical projection � ∶ G → G/H de�nes asmooth principal H -bundle structure. In this case, a map f ∶ G/H → M to some manifoldM is smooth if and only if the induced map f = f ◦� ∶ G → M is smooth.

(2) The inclusion �∶ H → G is a morphism of Lie groups, and there exist an opensubset V containing 0 in some locally convex space k and a smooth map � ∶ V → G with� (0) = e such that

�∶ V × H → G, (X, y)↦ � (X )y, X ∈ V , y ∈ H, (11.2)

is a di�eomorphism onto an open subset of G. In this case, �(V × H ) is a tube around Hin G.

Direct inspection of the proof of that proposition in [GN] shows the following.Complement 11.2: The equivalence of (1) and (2) in Proposition 11.1 is in particular validin the category of tame smooth Fréchet manifolds and tame smooth maps.

Given a Lie group G and a subgroup H , we refer to H as a principal Lie subgroupwhen one (and hence both) of the conditions in Proposition 11.1 are met. In [GN]such a subgroup H is termed a split Lie subgroup but, as the requirement of being asplitting submanifold is not enough to ensure the properties in the above propositionwe prefer the terminology principal Lie subgroup. When the Lie group G admits anexponential map, a local product structure around the identity already su�ces sincethe local product structure can then be extended to the whole subgroup.

The subsequent proposition is a consequence of [GN, Proposition 7.1.24]. Since the

Fréchet Slices 32

�nal version of this book is not yet available, for ease of exposition, we spell out acomplete statement.Proposition 11.3: Let G be a Lie group with smooth exponential map and H ⊆ G asplitting Lie subgroup. Denote a complement of h in g by k, and let V ⊆ k be an openneighborhood of 0 in k. If the map

�∶ V × H → G, (X, y)↦ exp(X )y, X ∈ V , y ∈ H, (11.3)

is a local di�eomorphism at (0, e), then H is a principal Lie subgroup.

Proof. Since � is a local di�eomorphism at (0, e) there exists a neighborhood UH ⊆ H ofe in H such that the restriction ��V×UH is a di�eomorphism onto an open neighborhoodUG ⊆ G of e in G (after shrinking V if need be). Because of the identity

�(X, qy) = �(X, q)y, X ∈ V , q, y ∈ H,

the map � is a local di�eomorphism at every point of V × H .Thus, to apply Proposition 11.1, it su�ces to show that � is injective. Since UH is

open in H , there exists an open neighborhood WG of e in G such that UH = H ∩WG . Byshrinking V further, we can arrange for the set exp(−V ) exp(V ) to lie completely in WG .Now let (X, a) and (Y , b) be two points of V × H having the same image under �. Then

exp(−Y ) exp(X ) = ba−1. (11.4)

The left-hand side lies in WG by assumption and the right-hand side is an element ofH whence both expressions are contained in UH . On the other hand, � is bijective onV × UH , and hence the calculation

�(Y , ba−1) = �(Y , b)a−1 = �(X, a)a−1 = �(X, e)

establishes the desired result X = Y and a = b.

Since the derivative �′0,e ∶ k × h → g of � at the point (0, e) of V × H is just the directsum isomorphism k ⊕ h → g, the inverse function theorem yields the necessary localdi�eomorphism so that Proposition 11.3 applies to Banach Lie groups; thus all splittingLie subgroups of a Banach Lie group are principal.

Exploiting the Nash–Moser theorem, we obtain the following important consequencefor tame Fréchet Lie groups.Theorem 11.4: Let G be a tame Fréchet Lie group with tame smooth exponential mapwhich is a local di�eomorphism close to 0. Every splitting Lie subgroup H ⊆ G of �nitedimension or �nite codimension is a principal Lie subgroup.

Proof. Denote the complement of h in g by k, and let V ⊆ k be an open neighborhoodof 0. In view of Proposition 11.3 above, we have to show that

�∶ V × H → G, (X, y)↦ exp(X )y, X ∈ V , y ∈ H, (11.5)

Fréchet Slices 33

is a local di�eomorphism at (0, e).The derivative of � at the point (X, e) of V × H is given by:

�′(X,e)∶ k × h → Texp(X )G, (Y , Z )↦ exp′X Y + (Lexp(X ))′eZ, Y ∈ k, Z ∈ h. (11.6)

In the case of a �nite-dimensional subgroup H , apply Proposition 11.6 below to

(exp−1 ◦�)′(X,e)∶ k × h → g, (Y , Z )↦ Y + (exp−1 ◦Lexp(X ))′eZ, Y ∈ k, Z ∈ h, (11.7)

and conclude that �′(Y ,e) is invertible with tame inverse for every Y ∈ V near X . By H -translation the same statement holds true at every point (Y , ℎ) near (X, e). By the versionof the Nash–Moser inverse function given in [Die13, Theorem 3.4.2], we conclude that� is a local di�eomorphism at (0, e).

A similar reasoning, applied to the map

(Lexp(−X )◦�)′(X,e)∶ k × h → g, (Y , Z )↦ (Lexp(−X )◦ exp)′XY + Z, Y ∈ k, Z ∈ h, (11.8)

establishes the case of �nite codimension.

For intelligibility, we now recall the tame Fréchet setting and the Nash–Moser inversefunction theorem, cf. [Ham82, Die13]. A Fréchet space is called graded if it carries adistinguished increasing fundamental system of seminorms ‖ ⋅ ‖k . If the growth of thesequence k ↦ ‖ ⋅ ‖k is controlled by the exponential map, then the graded Fréchet spaceis called tame (see [Ham82, De�nition II.1.3.2] for the exact statement). A smooth mapf between graded Fréchet spaces is r-tame smooth if f and all its derivatives satisfy alocal estimate of the form

‖f (x)‖k ≲ 1 + ‖x‖k+r . (11.9)

Roughly speaking, this means that f has a ‘maximal loss of r derivatives’.Theorem 11.5 ([Ham82, section III.1]): Let X and Y be tame Fréchet spaces, let U be anopen subset of X , and let f ∶ U → Y be tame smooth. Suppose that the derivative f ′ hasa tame smooth family Ψf of inverses, that is, Ψf ∶ U × Y → X is a tame smooth map,and the family Ψf

x ∶ Y → X is inverse to f ′x for every x ∈ U . Then the map f is locallybijective and the inverse is a tame smooth map.Proposition 11.6 ([Sub85, pp. 47ff]): Let A and X be tame Fréchet spaces, and letE ⊆ X and F ⊆ X be closed subspaces. Suppose that E is �nite-dimensional. Moreover,let Φ∶ A × (E × F )→ X be a tame smooth family of linear maps which decomposes into

Φa(u, v) = 'a(u) + v, a ∈ A, u ∈ E, v ∈ F , (11.10)

where '∶ A×E → X is a tame smooth family of injective, linear maps. If, for some a0 ∈ A,the partial map Φa0 ∶ E × F → X is a linear and topological isomorphism E ⊕ F → Xthen there is an open neighborhood U ⊆ A of a0 in A such that Φa is a bijection for everya ∈ U in such a way that the inverses constitute a tame smooth map U × X → E × F .

Fréchet Slices 34

11.2 De�nition of a slice

In �nite dimensions, a smooth retraction and a tubular neighborhood are equivalent tothe existence of a slice, see [OR04, Theorem 2.3.26]. However, the proof relies on theinverse function theorem; indeed, this theorem entails that an in�nitesimal splittinggenerates a local product decomposition. This argument does not carry over to arbitraryin�nite-dimensional manifolds. For this reason and for later reference, we now makeprecise the notion of slice we subsequently use; this is the strongest concept of a slicewe could possibly think of.

De�nition 11.7. Let Υ∶ G×M → M be a smooth action of a Lie groupG on a manifoldM . Let q be a point of M and suppose that the stabilizer Gq ⊆ G at q ∈ M is a principalLie subgroup. A slice at q ∈ M is a submanifold S of M that contains q in such a waythat the following conditions are met:

(S1) The G-closure G ⋅ S of S is an open neighborhood of the orbit G ⋅ q in M and S isclosed in G ⋅ S.

(S2) The submanifold S is closed under the induced action of Gq in the sense thatGq ⋅ S ⊆ S.

(S3) Any x ∈ G such that (x ⋅ S) ∩ S ≠ ∅ necessarily lies in Gq.

(S4) The smooth submersion G → G/Gq admits a local section � ∶ U → G de�nedon an open neighborhood U ⊆ G/Gq of the identity coset in such a way that themap

� S ∶ U × S → M, (x, y)↦ � (x) ⋅ y, x ∈ U , y ∈ S,

is a di�eomorphism onto an open neighborhood V of q in M .

A Lie group action Υ∶ G ×M → M is said to admit a slice at q ∈ M if the stabilizer Gqof q in G is a principal Lie subgroup of G and there is a slice at q ∈ M .

Proposition 11.8: Let Υ∶ G ×M → M be a smooth Lie group action admitting a slice Sat the point q of M . The following hold:

(1) Given y ∈ S, the stabilizer Gy of y is a subgroup of the stabilizer Gq of q. That is,the point q has the maximal stabilizer of the whole slice.

(2) The point q has a neighborhood V in M such that any point of V has a stabilizerthat is conjugate to a subgroup of Gq.

Proof. The �rst claim follows directly from property (S3).Let V be a neighborhood of the kind characterized in (S4). Since every v ∈ V is

obtained by translation of the slice S along the orbit, for some x ∈ G, the point x ⋅ v liesin S. Applying the �rst statement together with the equivariance of stabilizer subgroupswe conclude that xGvx−1 ⊆ Gq for some x ∈ G.

Fréchet Slices 35

11.3 Slice for the action of a subgroup

Let (M,G) be a G-manifold, and let q be a point of M having the property that theaction admits a slice at q. In this section we discuss how one can construct a slice at qfor the induced action of a subgroup H of G.Proposition 11.9: Let M be a manifold with a smooth G-action Υ admitting a slice atq ∈ M . Let H ⊆ G be a normal Lie subgroup. If the induced action of H on M is freeand the product GqH is a principal Lie subgroup of G, then there exists a slice at q forthe H -action as well.

In the gauge theory context in Section 4 above, we apply this proposition with Gbeing the group � of all gauge transformations and H the subgroup � ,Q of based gaugetransformations with respect to the chosen base point Q of the manifold written inSection 4 as M (but presently the notation M refers to a G-manifold). It is a standardfact that � ,Q acts freely on the space of connections and that it is a normal subgroupof � . By Theorem 11.4, � ,Q is a principal Lie subgroup of � . Furthermore, given asmooth connection A on � , the stabilizer � ,A of A in � is �nite-dimensional, and thegroup � ,A� ,Q has �nite codimension in � (indeed, dim(G) − dim(� ,A)) and hence, byTheorem 11.4, the group � ,A� ,Q is a principal Lie subgroup of � .

Proof. Let S be a slice at q ∈ M for the G-action. The group GqH is a principal Liesubgroup of G by assumption. Hence there is a closed subspace tq of g, an openneighborhood V of 0 in tq, and a smooth map � ∶ V → G with � (0) = e such that themap

�∶ V × GqH → G, (X, xy)↦ � (X ) xy, x ∈ Gq, y ∈ H,

is a di�eomorphism onto an open subset of G. Since H is a normal subgroup of G,

H� (V ) = � (V )H = �(V , H ).

We claim that S = � (V ) ⋅ S is a slice for the H -action. Indeed, S is a submanifold of Mcontaining q since, in view of Proposition 11.8 (1), for any point y of S, the stabilizerGy of y is a subgroup of Gq. Consequently the restriction of the action Υ to � (V ) × Sis a di�eomorphism onto S; we note that � (X ) lies in Gq only for the trivial elementX = 0. Furthermore, all de�ning properties of a slice are met:(S1) The H -closure H ⋅ S = H� (V ) ⋅ S = �(V , H ) ⋅ S is open in M since G ⋅ S ⊆ M is openand since V is an open subset of tq . Furthermore, since V × {e} is closed in V × H , thespace S = �(V , e) ⋅ S is closed in H ⋅ S = �(V , H ) ⋅ S.(S2) The manifold S is clearly invariant under the trivial stabilizer Hq = {e}.(S3) Let x ∈ H and y ∈ S such that x ⋅ y lies in S again. We have to show x = e. Bythe construction of S there are X, X ′ ∈ V and s, s′ ∈ S with x� (X ) ⋅ s = � (X ′) ⋅ s′. Nowproperty (S3) for the G-slice S implies that, for some w ∈ Gq,

x� (X ) = � (X )� (X )−1ℎ� (X ) = � (X ′)w.

References 36

Since H is a normal subgroup of G, the element x ∶= � (X )−1x� (X ) lies in H again andhence

�(X, e) = � (X ) = � (X ′)wy−1 = �(X ′, wy−1).

The map �∶ V × GqH → G is injective and thus yields X = X ′ and w = e, x = e.Consequently x = e.(S4) Since S is a slice, there is a smooth map � ∶ U → G de�ned on an open neighbor-hood U of 0 in t × h such that the maps

� S ∶ U × S → M, (X, y)↦ � (X ) ⋅ y, X ∈ U , y ∈ S,� ∶ U × Gq → G, (X, y)↦ � (X )y, X ∈ U , y ∈ Gq,

are di�eomorphisms onto open neighborhoods WM of q in M and WG of e in G, respec-tively. Shrink V (and hence the slice S) and choose an open and conjugation invariantsubset VH of H such that �(V , VH ) ⊆ �(U , e). Then the map �∶ V × VH ↦ U de�nedby � = �−1◦� is a di�eomorphism onto an open subset of U .

Finally, the map

�∶ VH × S → M, (v, � (X ) ⋅ y)↦ v� (X ) ⋅ y, v ∈ VH , X ∈ V , y ∈ S,

is a di�eomorphism onto an open neighborhood of q in M . Indeed, this map can bewritten as the composite

VH × S VH × V × S U × S WMId×�−1 �×Id � S

made explicit by the association

(v, � (X ) ⋅ y) (v, X , y) (�−1◦�(v, X ), y) �(v, X ) ⋅ y,

where v ∈ VH , X ∈ V , y ∈ S.

Acknowledgements

We are indebted to B. Tumpach and A. Waldron for discussion. We gratefully ac-knowledge support by the CNRS, by the Labex CEMPI (ANR-11-LABX-0007-01), by theGerman National Academic Foundation (DAAD), and by the Max Planck Institute forMathematics in the Sciences (Leipzig).

References

[AB82] Michael F. Atiyah and Raoul Bott. The Yang-Mills equations over Riemannsurfaces. Phil. Trans. R. Soc. London A, 308:523–615, 1982.

[ACM89] Maria C. Abbati, Renzo Cirelli, and Alessandro Manià. The orbit space of

References 37

the action of gauge transformation group on connections. J. Geom. Phys.,6(4):537–557, 1989.

[ACMM86] Maria C. Abbati, Renzo Cirelli, Alessandro Manià, and Peter Michor.Smoothness of the action of the gauge transformation group on connec-tions. J. Math. Phys., 27(10):2469–2474, 1986.

[AMM98] Anton Alekseev, Anton Malkin, and Eckhard Meinrenken. Lie group valuedmoment maps. J. Di�erential Geom., 48(3):445–495, 1998.

[Ati57] Michael F. Atiyah. Complex analytic connections in �bre bundles. Trans.Amer. Math. Soc., 85:181–207, 1957.

[Bry95] Robert L. Bryant. An introduction to Lie groups and symplectic geometry.In Geometry and quantum �eld theory (Park City, UT, 1991), volume 1 ofIAS/Park City Math. Ser., pages 5–181. Amer. Math. Soc., Providence, RI,1995.

[Che95] Shiing-Shen Chern. Complex manifolds without potential theory (with anappendix on the geometry of characteristic classes). Universitext. Springer-Verlag, New York, second edition, 1995.

[CM85] Renzo Cirelli and Alessandro Manià. The group of gauge transformationsas a Schwartz-Lie group. J. Math. Phys., 26(12):3036–3041, 1985.

[Die69] Jean A. E. Dieudonné. Foundations of modern analysis. Academic Press, NewYork-London, 1969. Enlarged and corrected printing, Pure and AppliedMathematics, Vol. 10-I.

[Die13] Tobias Diez. Slice theorem for Fréchet group actions and covariant symplectic�eld theory, Master Thesis–Leipzig University. Leipzig University, 2013.ARXIV:1405.2249.

[GN] Helge Glöckner and Karl-Hermann Neeb. In�nite-dimensional Lie Groups:Geometry and Topology. Springer-Verlag New York Inc. In preparation.

[Gru70] Karl W. Gruenberg. Cohomological topics in group theory. Lecture Notes inMathematics, Vol. 143. Springer-Verlag, Berlin-New York, 1970.

[GS84] Victor Guillemin and Shlomo Sternberg. Symplectic techniques in physics.Cambridge University Press, Cambridge, 1984.

[H07] Lars Hörmander. The analysis of linear partial di�erential operators. III:Pseudo-di�erential operators. Classics in Mathematics. Springer-Verlag,Berlin, 2007. Reprint of the 1994 edition.

[Ham82] Richard S. Hamilton. The inverse function theorem of Nash and Moser.Bull. Amer. Math. Soc. (N.S.), 7(1):65–222, 1982.

References 38

[Hir95] Friedrich Hirzebruch. Topological methods in algebraic geometry. Classicsin Mathematics. Springer-Verlag, Berlin, 1995. Translated from the Germanand Appendix One by R. L. E. Schwarzenberger; with a preface to the thirdEnglish edition by the author and R. L. E. Schwarzenberger; Appendix Twoby A. Borel. Reprint of the 1978 edition.

[HJ94] Johannes Huebschmann and Lisa C. Je�rey. Group cohomology construc-tion of symplectic forms on certain moduli spaces. Internat. Math. Res.Notices, (6):245–249, 1994.

[Hue95a] Johannes Huebschmann. The singularities of Yang-Mills connections forbundles on a surface. I. The local model. Math. Z., 220(4):595–609, 1995.DG-GA/9411006.

[Hue95b] Johannes Huebschmann. Symplectic and Poisson structures of certainmoduli spaces. I. Duke Math. J., 80(3):737–756, 1995. HEP-TH/9311212.

[Hue96] Johannes Huebschmann. The singularities of Yang-Mills connections forbundles on a surface. II. The strati�cation. Math. Z., 221(1):83–92, 1996.DG-GA/9411007.

[Hue98] Johannes Huebschmann. Smooth structures on certain moduli spacesfor bundles on a surface. J. Pure Appl. Algebra, 126(1-3):183–221, 1998.DG-GA/9411008.

[Hue99] Johannes Huebschmann. Extended moduli spaces, the Kan construction,and lattice gauge theory. Topology, 38(3):555–596, 1999. DG-GA/9505005,DG-GA/9505006.

[Hue01] Johannes Huebschmann. Singularities and Poisson geometry of cer-tain representation spaces. In Quantization of singular symplectic quo-tients, volume 198 of Progr. Math., pages 119–135. Birkhäuser, Basel, 2001.MATH.DG/0012184.

[Jef97] Lisa C. Je�rey. Symplectic forms on moduli spaces of �at connections on2-manifolds. In Geometric topology (Athens, GA, 1993), volume 2 of AMS/IPStud. Adv. Math., pages 268–281. Amer. Math. Soc., Providence, RI, 1997.

[KN63] Shoshichi Kobayashi and Katsumi Nomizu. Foundations of di�erentialgeometry. Vol I. Interscience Publishers, a division of John Wiley & Sons,New York-London, 1963.

[Mil84] John Milnor. Remarks on in�nite dimensional Lie groups. In Bryce S.DeWitt and R. Stora, editors, Relativity, Groups and Topology II: Les HouchesSummer School Proceedings, pages 1007–1059. Elsevier Science Ltd, Septem-ber 1984.

References 39

[Nee06] Karl-Hermann Neeb. Towards a Lie theory of locally convex groups.Japanese Journal of Mathematics, 1(2):291–468, 2006.

[OR04] Juan-Pablo Ortega and Tudor S. Ratiu. Momentum maps and Hamiltonianreduction, volume 222 of Progress in Mathematics. Birkhäuser Boston, Inc.,Boston, MA, 2004.

[Sik72] Roman Sikorski. Wst‘ep do geometrii różniczkowej. Państwowe

Wydawnictwo Naukowe, Warsaw, 1972. Biblioteka Matematyczna, Tom42. [Mathematics Library, Vol. 42].

[SL91] Reyer Sjamaar and Eugene Lerman. Strati�ed symplectic spaces and re-duction. Ann. of Math. (2), 134(2):375–422, 1991.

[Sub85] T. N. Subramaniam. Slices for the actions of smooth tame Lie groups (Fréchet,gauge, di�eomorphism, Nash-Moser inverse function theorem). ProQuestLLC, Ann Arbor, MI, 1985. Thesis (Ph.D.)–Brandeis University.

[Sub86] T. N. Subramaniam. Slices for actions of in�nite-dimensional groups.In Di�erential analysis in in�nite-dimensional spaces (Albany, N.Y., 1983),volume 54 of Contemp. Math., pages 65–77. Amer. Math. Soc., Providence,RI, 1986.

[Uhl82] Karen K. Uhlenbeck. Connections with Lp bounds on curvature. Comm.Math. Phys., 83(1):31–42, 1982.

[Whi34] Hassler Whitney. Analytic extensions of di�erentiable functions de�nedin closed sets. Trans. Amer. Math. Soc., 36(1):63–89, 1934.

[Whi57] Hassler Whitney. Elementary structure of real algebraic varieties. Ann. ofMath. (2), 66:545–556, 1957.

[ZVC80] Heiner Zieschang, Elmar Vogt, and Hans-Dieter Coldewey. Surfaces andplanar discontinuous groups, volume 835 of Lecture Notes in Mathematics.Springer-Verlag, Berlin, 1980. Translated from the German by John Stillwell.

surface via FrØchet reduction - arXiv · surface via FrØchet reduction ... holonomies and a...

Documents

Transcript of surface via FrØchet reduction - arXiv · surface via FrØchet reduction ... holonomies and a...