Introduction
We consider parabolic S L ( r ⋆ , C ) \mathrm{SL}(r_\star,\mathbb{C}) SL ( r ⋆ , C ) -Higgs bundles on a compact Riemann surface with fixed flag dimensions. The goal of this paper is to understand how the hyperkähler periods ( ∫ η ω 1 , ∫ η ω 2 , ∫ η ω 3 ) (\int_\eta\omega_1,\int_\eta\omega_2,\int_\eta\omega_3) ( ∫ η ω 1 , ∫ η ω 2 , ∫ η ω 3 ) of these Hitchin moduli spaces change as we vary the parabolic weights and the complex residues. We prove that, after canonically identifying homology classes, the three symplectic classes vary affine linearly with slopes given by a common characteristic class. We then compute these slopes for the four-dimensional moduli spaces of exceptional types E 6 , E 7 , E 8 E_6,E_7,E_8 E 6 , E 7 , E 8 .
Let C C C be a compact connected Riemann surface, D = { p 1 , … , p n } D=\{p_1,\ldots,p_n\} D = { p 1 , … , p n } a divisor of distinct marked points on C C C , and E E E a rank-r ⋆ r_\star r ⋆ bundle. Fix an identification det E ≅ O C \det E\cong\mathcal{O}_C det E ≅ O C . We consider parabolic S L ( r ⋆ , C ) \mathrm{SL}(r_\star,\mathbb{C}) SL ( r ⋆ , C ) -Higgs bundles with flag dimensions
0 < r i ( 0 ) < ⋯ < r i ( ℓ i ) = r ⋆ , i ∈ { 1 , … , n } . 0<r_i^{(0)}<\cdots<r_i^{(\ell_i)}=r_\star,\qquad i\in\{1,\ldots,n\}. 0 < r i ( 0 ) < ⋯ < r i ( ℓ i ) = r ⋆ , i ∈ { 1 , … , n } .
We denote the flags and the parabolic weights associated to them by
0 ⊂ F i ( 0 ) ⊂ ⋯ ⊂ F i ( ℓ i ) = E p i , α i ( 0 ) > ⋯ > α i ( ℓ i ) , 0\subset F_i^{(0)}\subset\cdots\subset F_i^{(\ell_i)}=E_{p_i},\qquad\alpha_i^{(0)}>\cdots>\alpha_i^{(\ell_i)}, 0 ⊂ F i ( 0 ) ⊂ ⋯ ⊂ F i ( ℓ i ) = E p i , α i ( 0 ) > ⋯ > α i ( ℓ i ) ,
where dim F i ( k ) = r i ( k ) \dim F_i^{(k)}=r_i^{(k)} dim F i ( k ) = r i ( k ) and α i ( 0 ) − α i ( ℓ i ) < 1 \alpha_i^{(0)}-\alpha_i^{(\ell_i)}<1 α i ( 0 ) − α i ( ℓ i ) < 1 . We allow partial flags, and denote the flag multiplicities m i ( k ) = r i ( k ) − r i ( k − 1 ) m_i^{(k)}=r_i^{(k)}-r_i^{(k-1)} m i ( k ) = r i ( k ) − r i ( k − 1 ) for k = 0 , … , ℓ i k=0,\ldots,\ell_i k = 0 , … , ℓ i , where r i ( − 1 ) = 0 r_i^{(-1)}=0 r i ( − 1 ) = 0 . The residue of the holomorphic Higgs field φ \varphi φ at p i p_i p i is required to preserve the flag and act on the associated graded space gr F i ( k ) = F i ( k ) / F i ( k − 1 ) \operatorname{gr}F_i^{(k)}=F_i^{(k)}/F_i^{(k-1)} gr F i ( k ) = F i ( k ) / F i ( k − 1 ) as scalar multiplication by σ i ( k ) ∈ C \sigma_i^{(k)}\in\mathbb{C} σ i ( k ) ∈ C . In other words, the complex residue res p i φ \operatorname{res}_{p_i}\varphi res p i φ has prescribed Levi part ⨁ k σ i ( k ) Id \bigoplus_k\sigma_i^{(k)}\operatorname{Id} ⨁ k σ i ( k ) Id . The parabolic weights and complex residues satisfy the trace-free conditions for each i ∈ { 1 , … , n } i\in\{1,\ldots,n\} i ∈ { 1 , … , n }
∑ k = 0 ℓ i m i ( k ) α i ( k ) = 0 , ∑ k = 0 ℓ i m i ( k ) σ i ( k ) = 0. \sum_{k=0}^{\ell_i}m_i^{(k)}\alpha_i^{(k)}=0,\qquad
\sum_{k=0}^{\ell_i}m_i^{(k)}\sigma_i^{(k)}=0. k = 0 ∑ ℓ i m i ( k ) α i ( k ) = 0 , k = 0 ∑ ℓ i m i ( k ) σ i ( k ) = 0.
We write M ( α , σ ) \mathcal{M}(\alpha,\sigma) M ( α , σ ) for the moduli space of parabolic S L ( r ⋆ , C ) \mathrm{SL}(r_\star,\mathbb{C}) SL ( r ⋆ , C ) -Higgs bundles with these weights and residues. We emphasize that only the Levi part of res p i φ \operatorname{res}_{p_i}\varphi res p i φ is prescribed, while the upper triangular blocks are allowed to vary. On its smooth stable locus, this moduli space carries a natural hyperkähler structure [4 ] , with Kähler forms ω 1 , ω 2 , ω 3 \omega_1,\omega_2,\omega_3 ω 1 , ω 2 , ω 3 . Given an oriented cycle η \eta η representing a class in H 2 ( M ( α , σ ) , Z ) H_2(\mathcal{M}(\alpha,\sigma),\mathbb{Z}) H 2 ( M ( α , σ ) , Z ) , define its period triple by
P η = ( ∫ η ω 1 , ∫ η ω 2 , ∫ η ω 3 ) . (1) P_\eta=\left(\int_\eta\omega_1,\int_\eta\omega_2,\int_\eta\omega_3\right). \tag*{(1)} P η = ( ∫ η ω 1 , ∫ η ω 2 , ∫ η ω 3 ) . (1)
To compare these triples as the weights and residues vary, we first need to identify homology classes in different moduli spaces.
Consider the parameter space Λ \Lambda Λ whose elements ν \nu ν encode pairs ( α , σ ) (\alpha,\sigma) ( α , σ ) satisfying the trace-free conditions above. We work over the generic parameter locus Λ 0 ⊂ Λ \Lambda_0 \subset\Lambda Λ 0 ⊂ Λ of Definition 5.3 so as to guarantee that M ν = M ( α , σ ) \mathcal{M}_\nu= \mathcal{M}(\alpha,\sigma) M ν = M ( α , σ ) is a smooth hyperkähler manifold. This Λ 0 \Lambda_0 Λ 0 is obtained by excluding the simultaneous weight and residue relations associated with a possible proper invariant subbundle. The first condition cuts out a real co-dimension one affine subspace inside Λ \Lambda Λ and the second a complex co-dimension one subspace. Under the co-primality assumption
gcd ( r ⋆ , { m i ( k ) } i , k ) = 1 , \operatorname{gcd}\left(r_{\star},\left\{m_i^{(k)}\right\}_{i,k}\right)=1, gcd ( r ⋆ , { m i ( k ) } i , k ) = 1 ,
these excluded loci form a finite union of real co-dimension three affine subspaces inside Λ \Lambda Λ . In particular, this implies that Λ 0 \Lambda_0 Λ 0 is simply connected.
In § 3 and § 4 , we assemble the fixed-parameter moduli spaces M ( α , σ ) \mathcal{M}(\alpha,\sigma) M ( α , σ ) into a smooth family M \mathcal{M} M with a submersion π : M → Λ 0 \pi:\mathcal{M}\to\Lambda_0 π : M → Λ 0 whose fibers are the spaces M ( α , σ ) \mathcal{M}(\alpha,\sigma) M ( α , σ ) . The required canonical identification of homology classes follows from the stronger result that π \pi π is in fact a fiber bundle. We prove this by showing that the family is locally topologically trivial over Λ 0 \Lambda_0 Λ 0 , which implies that its integral homology groups form a local system compatible with the inclusions of the fibers into the total space. Thus a path γ : [ 0 , 1 ] → Λ 0 \gamma:[0,1]\to\Lambda_0 γ : [ 0 , 1 ] → Λ 0 gives a parallel transport identification
PT γ : H 2 ( M γ ( 0 ) , Z ) ⟶ H 2 ( M γ ( 1 ) , Z ) \operatorname{PT}_{\gamma}:H_2(\mathcal{M}_{\gamma(0)},\mathbb{Z})\longrightarrow H_2(\mathcal{M}_{\gamma(1)},\mathbb{Z}) PT γ : H 2 ( M γ ( 0 ) , Z ) ⟶ H 2 ( M γ ( 1 ) , Z )
and dual parallel transport identification
PT γ ∗ : H 2 ( M γ ( 1 ) , R ) ⟶ H 2 ( M γ ( 0 ) , R ) . \operatorname{PT}_{\gamma}^{*}:H^2(\mathcal{M}_{\gamma(1)},\mathbb{R})\longrightarrow H^2(\mathcal{M}_{\gamma(0)},\mathbb{R}). PT γ ∗ : H 2 ( M γ ( 1 ) , R ) ⟶ H 2 ( M γ ( 0 ) , R ) .
Compatibility means that parallel transport commutes with the inclusions i ν : M ν ↪ M i_\nu:\mathcal{M}_\nu\hookrightarrow\mathcal{M} i ν : M ν ↪ M . Since Λ 0 \Lambda_0 Λ 0 is simply connected, these identifications depend only on the endpoints. The identifications of homology classes are canonical, even though the local trivializations used to construct them depend on choices.
To describe the variation of the three Kähler classes [ ω 1 , ν ] [\omega_{1,\nu}] [ ω 1 , ν ] , [ ω 2 , ν ] [\omega_{2,\nu}] [ ω 2 , ν ] , [ ω 3 , ν ] [\omega_{3,\nu}] [ ω 3 , ν ] for ν ∈ Λ 0 \nu\in\Lambda_0 ν ∈ Λ 0 , let B B B be the boundary torus constructed in § 3.1 , with Lie algebra b \mathfrak{b} b . This torus is a quotient of the gauge group, and has dimension d = ∑ i ℓ i d=\sum_i\ell_i d = ∑ i ℓ i . On the other hand, the real dimension of Λ 0 \Lambda_0 Λ 0 is 3 d 3d 3 d . Using the duality pairing and the normalization of (36) , we write the parameters as ν = ( ν 1 , ν 2 , ν 3 ) ∈ ( b ∗ ) ⊕ 3 \nu=(\nu_1,\nu_2,\nu_3)\in(\mathfrak{b}^{*})^{\oplus3} ν = ( ν 1 , ν 2 , ν 3 ) ∈ ( b ∗ ) ⊕ 3 , where
ν 1 = 2 π i α , ν 2 = 4 π i Re σ , ν 3 = 4 π i Im σ . \nu_1=2\pi i\alpha,\qquad\nu_2=4\pi i\operatorname{Re}\sigma,\qquad\nu_3=4\pi i\operatorname{Im}\sigma. ν 1 = 2 π i α , ν 2 = 4 π i Re σ , ν 3 = 4 π i Im σ .
Here the weights and residues are regarded as block scalar matrices.
The simplest case we could hope for is that there is a common class c ν ∈ H 2 ( M ν , R ) ⊗ b c_\nu\in H^2(\mathcal{M}_\nu,\mathbb{R})\otimes\mathfrak{b} c ν ∈ H 2 ( M ν , R ) ⊗ b , identified throughout Λ 0 \Lambda_0 Λ 0 via parallel transport, such that the Gauss–Manin directional derivative along a variation ν ˙ \dot{\nu} ν ˙ of ν ∈ Λ 0 \nu\in\Lambda_0 ν ∈ Λ 0 is
D ν ˙ [ ω s , ν ] = − ⟨ c ν , ν ˙ ⟩ D_{\dot{\nu}}[\omega_{s,\nu}]=-\langle c_\nu,\dot{\nu}\rangle D ν ˙ [ ω s , ν ] = − ⟨ c ν , ν ˙ ⟩
for all t t t . For a smooth path γ ( t ) : [ 0 , 1 ] → Λ 0 \gamma(t):[0,1]\to\Lambda_0 γ ( t ) : [ 0 , 1 ] → Λ 0 , we can integrate to obtain
PT γ ∗ [ ω s , γ ( 1 ) ] = [ ω s , γ ( 0 ) ] − ⟨ c γ ( 0 ) , γ s ( 1 ) − γ s ( 0 ) ⟩ , s = 1 , 2 , 3. (2) \operatorname{PT}_{\gamma}^{*}[\omega_{s,\gamma(1)}]=[\omega_{s,\gamma(0)}]-\langle c_{\gamma(0)},\gamma_s(1)-\gamma_s(0)\rangle,\qquad s=1,2,3. \tag*{(2)} PT γ ∗ [ ω s , γ ( 1 ) ] = [ ω s , γ ( 0 ) ] − ⟨ c γ ( 0 ) , γ s ( 1 ) − γ s ( 0 )⟩ , s = 1 , 2 , 3. (2)
Definition 5.5 proves precisely this formula. In particular, the same class governs all three variations.
There is a natural geometric candidate for the class c c c . Suppose that we can construct a principal B B B -bundle q : E → M q:\mathcal{E}\to\mathcal{M} q : E → M . Let θ \theta θ be a connection and let F θ F_\theta F θ denote its curvature, viewed as a b \mathfrak{b} b -valued two-form on M \mathcal{M} M . If i ν : M ν ↪ M i_\nu:\mathcal{M}_\nu\hookrightarrow\mathcal{M} i ν : M ν ↪ M is the inclusion, then the restricted curvature class
c ν : = i ν ∗ [ F θ ] c_\nu:= i_\nu^*[F_\theta] c ν := i ν ∗ [ F θ ]
is invariant under parallel transport. This gives a candidate class, but to identify it with the Gauss–Manin derivative of the symplectic classes we need additional geometric structure. Specifically, suppose E \mathcal{E} E carries three closed, B B B -invariant two-forms ϖ s \varpi_s ϖ s which satisfy the Hamiltonian identity
ι ζ b ϖ s = d ⟨ π ~ s , b ⟩ , b ∈ b , s = 1 , 2 , 3 , \iota_{\zeta_b}\varpi_s=d\langle\widetilde{\pi}_s,b\rangle,\qquad b\in\mathfrak{b},\quad s=1,2,3, ι ζ b ϖ s = d ⟨ π s , b ⟩ , b ∈ b , s = 1 , 2 , 3 ,
where π ~ = ( π ~ 1 , π ~ 2 , π ~ 3 ) \widetilde{\pi}=(\widetilde{\pi}_1,\widetilde{\pi}_2,\widetilde{\pi}_3) π = ( π 1 , π 2 , π 3 ) is the lifted parameter map E → Λ 0 \mathcal{E}\to\Lambda_0 E → Λ 0 and ζ b \zeta_b ζ b is the vector field generated by b b b . Furthermore, assume ϖ s \varpi_s ϖ s descends to ω s \omega_s ω s under the quotient by B B B . We can then show
ϖ s + d ⟨ θ , π ~ s ⟩ \varpi_s+d\langle\theta,\widetilde{\pi}_s\rangle ϖ s + d ⟨ θ , π s ⟩
descends to a closed two-form η s \eta_s η s on M \mathcal{M} M . Restriction to the fiber over ν \nu ν and calculating the exterior derivative yields
[ ω s , ν ] = i ν ∗ [ η s ] − ⟨ c ν , ν s ⟩ . [\omega_{s,\nu}]=i_\nu^*[\eta_s]-\langle c_\nu,\nu_s\rangle. [ ω s , ν ] = i ν ∗ [ η s ] − ⟨ c ν , ν s ⟩ .
Both i ν ∗ [ η s ] i_\nu^*[\eta_s] i ν ∗ [ η s ] and c ν c_\nu c ν are constant under parallel transport, so this identity gives (2) . Thus the Hamiltonian identities realize the curvature class c c c as the Gauss–Manin derivative, while compatibility of homology transport with the inclusions i ν i_\nu i ν makes that slope constant.
The process described above is almost identical to the mechanism behind the theorem of Duistermaat and Heckman [13 Theorem 1.1] . In its classical torus formulation, a Hamiltonian action of a compact torus G G G on a finite-dimensional symplectic manifold M M M , with moment map
μ : M ⟶ g ∗ , \mu:M\longrightarrow\mathfrak{g}^*, μ : M ⟶ g ∗ ,
gives affine variation of the reduced symplectic classes on a suitable region of regular moment-map values. In our setting, the analogous hyperkähler quotient involves an infinite-dimensional configuration space and an infinite-dimensional gauge group. Moreover, the reduced moduli spaces M ( α , σ ) \mathcal{M}(\alpha,\sigma) M ( α , σ ) are noncompact.
§ 2 develops a general framework for dealing with situations of this type. The first observation is that, under suitable hypotheses, partial reduction by a subgroup H H H leaves a finite-dimensional space with an action of the residual group B = G / H B=G/H B = G / H . The second observation is that the Duistermaat–Heckman proof applies once a geometric and a topological hypothesis are satisfied. The geometric hypothesis essentially is the existence of the principal bundle, closed forms, and Hamiltonian identities described above, while the topological hypothesis is the existence of a homology local system compatible with the inclusions into the total space.
Remark 1.1. As in the proof in [13 ] , the existence of a complete Ehresmann connection on the total space guarantees that the topological hypothesis is satisfied. We illustrate this through an example, treating the multi Taub-NUT family in § 2.2 . Proving the existence of a complete Ehresmann connection is highly non-trivial both generally and in particular for the case treated in this paper.
In § 3 , we produce the structures required by the geometric hypothesis of Definition 2.3 in the case of parabolic Higgs bundles. First we define the extended configuration space C \mathcal{C} C allowing varying boundary data, together with the appropriate gauge group G G G . Then we extend the usual L 2 L^2 L 2 Kähler forms to a triple of closed 2-forms ω ~ 1 , ω ~ 2 , ω ~ 3 \widetilde{\omega}_1,\widetilde{\omega}_2,\widetilde{\omega}_3 ω 1 , ω 2 , ω 3 on C \mathcal{C} C , and produce a map
μ = ( μ 1 , μ 2 , μ 3 ) : C ⟶ ( g ∗ ) ⊕ 3 \mu=(\mu_1,\mu_2,\mu_3):\mathcal{C}\longrightarrow(\mathfrak{g}^*)^{\oplus3} μ = ( μ 1 , μ 2 , μ 3 ) : C ⟶ ( g ∗ ) ⊕ 3
satisfying a Hamiltonian identity with respect to the ω ~ s \widetilde{\omega}_s ω s and the G G G -action. The moment map is obtained by recasting Hitchin’s equations using distributional derivatives and regarding them as bounded linear functionals on the Lie algebra g \mathfrak{g} g . Taking level sets, the interior part μ i n t \mu^{\mathrm{int}} μ int gives the usual Hitchin equations, and the boundary part μ b d r \mu^{\mathrm{bdr}} μ bdr gives the weight and residue prescriptions. We identify Λ ⊂ ( g ∗ ) ⊕ 3 \Lambda\subset(\mathfrak{g}^*)^{\oplus3} Λ ⊂ ( g ∗ ) ⊕ 3 using point-mass distributions supported on D D D .
The gauge group G G G contains a normal subgroup H H H such that G / H G/H G / H is a finite-dimensional boundary torus B B B . The forms ω ~ s \widetilde{\omega}_s ω s descend to closed forms ϖ s \varpi_s ϖ s on the partial quotient E = ( μ i n t ) − 1 ( 0 ) / H \mathcal{E}=(\mu^{\mathrm{int}})^{-1}(0)/H E = ( μ int ) − 1 ( 0 ) / H . The latter retains a residual action of B B B , and the boundary moment map descends to a moment map π ~ : E ⟶ Λ ⊂ ( g ∗ ) ⊕ 3 \widetilde{\pi}:\mathcal{E}\longrightarrow\Lambda\subset(\mathfrak{g}^*)^{\oplus3} π : E ⟶ Λ ⊂ ( g ∗ ) ⊕ 3 for the B B B -action which reads off the parameters. Taking one more quotient gives the moduli family π : M ⟶ Λ \pi:\mathcal{M}\longrightarrow\Lambda π : M ⟶ Λ .
The existence of these closed 2-forms ϖ s \varpi_s ϖ s and the Hamiltonian identity for π ~ \widetilde{\pi} π is precisely the geometric hypothesis of Definition 2.3. The difficulty in defining them is that the usual L 2 L^2 L 2 forms on C \mathcal{C} C have divergent integrals when evaluated on deformations with varying boundary data. To account for this, we define the truncated forms ω s ε \omega_s^\varepsilon ω s ε by integrating over the complement of small balls B ε ( p i ) B_\varepsilon(p_i) B ε ( p i ) in C \mathcal{C} C . Subtracting logarithmically divergent boundary terms log ( 1 / ε ) b s ( u , v ) \log(1/\varepsilon)b_s(u,v) log ( 1/ ε ) b s ( u , v ) obtained by pairing the parameter variations, we get a well-defined form
ω ~ s = lim ε → 0 ( ω s ε + log ( ε ) b s ) . \widetilde{\omega}_s=\lim_{\varepsilon\to0}\left(\omega_s^\varepsilon+\log(\varepsilon)b_s\right). ω s = ε → 0 lim ( ω s ε + log ( ε ) b s ) .
The fact that they descend to the forms ϖ s \varpi_s ϖ s on E \mathcal{E} E follows from the Hamiltonian identity for μ i n t \mu^{\mathrm{int}} μ int with respect to the H H H -action.
Remark 1.2. Although we only consider regular singularities in this paper, the distributional formulation suggests an extension to irregular singularities. In a local coordinate centered at p i p_i p i , the eigenvalues of the higher order singular terms A n z − n d z A_nz^{-n}\,dz A n z − n d z are prescribed by allowing Λ \Lambda Λ to contain Levi-central matrices times distributions of the form ∂ z n − 1 δ p i \partial_z^{n-1}\delta_{p_i} ∂ z n − 1 δ p i , after appropriately shrinking the gauge Lie algebra near the divisor.
Section 4 supplies the smoothness assumptions of the affine variation theorem. For fixed Sobolev regularity k ≥ 2 k\ge2 k ≥ 2 and weight δ > 0 \delta>0 δ > 0 , we consider the extended deformation complex for the action of G G G on M \mathcal{M} M , whose cohomology provides a tangent space model for M i r r \mathcal{M}^{\mathrm{irr}} M irr . We use Biquard–Boalch’s splitting of the fixed-parameter deformation complex [4 ] to produce a splitting for the extended deformation complex. This allows us to construct local coordinates on Coulomb slices in C \mathcal{C} C orthogonal to the gauge orbits, which in turn define smooth local charts on M i r r \mathcal{M}^{\mathrm{irr}} M irr . We then give local trivializations for q : E i r r → M i r r q : \mathcal{E}^{\mathrm{irr}} \to\mathcal{M}^{\mathrm{irr}} q : E irr → M irr , showing it is a principal bundle, and prove that π : M i r r → Λ \pi: \mathcal{M}^{\mathrm{irr}} \to\Lambda π : M irr → Λ is a submersion.
Let M δ \mathcal{M}_{\delta} M δ and E δ \mathcal{E}_{\delta} E δ be the spaces constructed for a given choice of Sobolev weight δ > 0 \delta> 0 δ > 0 . The smoothness arguments require the Sobolev weight to be smaller than the smallest gap between parabolic weights at each puncture, considered as elements of R / Z \mathbb{R}/\mathbb{Z} R / Z . To allow the parabolic weights to vary within the generic locus, we show that the families { M δ } δ , { E δ } δ \{\mathcal{M}_{\delta}\}_{\delta}, \{\mathcal{E}_{\delta}\}_{\delta} { M δ } δ , { E δ } δ form directed systems and consider the direct limits
M = lim δ → 0 M δ E = lim δ → 0 E δ . \mathcal{M} = \lim_{\delta\to0} \mathcal{M}_{\delta} \qquad\mathcal{E} = \lim_{\delta\to0} \mathcal{E}_{\delta}. M = δ → 0 lim M δ E = δ → 0 lim E δ .
Since the later sections use the algebraic description for Higgs bundles, we provide a bridge between the algebraic and analytic perspectives. The existence of harmonic metrics for stable parabolic Higgs bundles translates the algebraic moduli family into our analytic moduli family. To turn this into a diffeomorphism, we control the harmonic metrics along 1-parameter families of Higgs bundles with varying boundary data, following Kim and Wilkin’s proof in [29 ] for varying parabolic weights with fixed residues.
Section 5 establishes local triviality of the family π : M → Λ \pi: \mathcal{M} \to\Lambda π : M → Λ over Λ 0 ⊂ Λ \Lambda_{0} \subset\Lambda Λ 0 ⊂ Λ , where semistability implies stability. Λ 0 \Lambda_{0} Λ 0 is obtained from Λ \Lambda Λ by removing a codimension-3 subset. Due to the noncompactness of the fibers of π \pi π , a horizontal lift of a path in Λ 0 \Lambda_{0} Λ 0 could a-priori escape to infinity in finite time. We construct lifts locally for small open sets in Λ 0 \Lambda_{0} Λ 0 , and show that they give a well-defined local system for the homology groups H 2 ( M ν , C ) H_{2}(\mathcal{M}_{\nu}, \mathbb{C}) H 2 ( M ν , C ) . This proves the topological hypothesis of the affine variation theorem.
Working in the algebraic setting, we begin with fixed generic parabolic weights, only allowing the Higgs field residues to vary in the vector space V V V of possible residue data. Denote this family π α : M α a l g → V \pi_{\alpha} : \mathcal{M}^{\mathrm{alg}}_{\alpha} \to V π α : M α alg → V . We use the C × \mathbb{C}^{\times} C × -action to choose lifts of paths in V V V whose images under the Hitchin map Hit : M α a l g → B \operatorname{Hit} : \mathcal{M}^{\mathrm{alg}}_{\alpha} \to B Hit : M α alg → B stay bounded. Properness of the joint map ( π α , Hit ) (\pi_{\alpha}, \operatorname{Hit}) ( π α , Hit ) , established by Lee and Lee in [32 ] , ensures lifted paths remain in a compact set. We extend these local trivializations to the full generic parameter locus Λ 0 \Lambda_{0} Λ 0 by comparing stability conditions as the weights vary. The resulting integral homology local systems have trivial monodromy because Λ 0 \Lambda_{0} Λ 0 is simply connected. This completes the hypotheses needed for Definition 2.3, which we use to prove the statement for Higgs bundles Definition 5.5.
Section 6 calculates the period maps explicitly in the four-dimensional moduli spaces of exceptional types E 6 , E 7 , E 8 E_{6}, E_{7}, E_{8} E 6 , E 7 , E 8 over C = C P 1 \mathbb{C} = \mathbb{CP}^{1} C = CP 1 , extending the surjectivity result of [15 ] for the D 4 D_{4} D 4 case. Definition 2.3 states that the period map is affine-linear in ν ∈ Λ 0 \nu\in\Lambda_{0} ν ∈ Λ 0 . Choosing a basis for the cohomology of M ν = M ( α , σ ) \mathcal{M}_{\nu} = \mathcal{M}(\alpha,\sigma) M ν = M ( α , σ ) and transporting throughout Λ 0 \Lambda_{0} Λ 0 , we can express the period map as
P ( ν ) = A ν + b \mathcal{P}(\nu) = A\nu+ b P ( ν ) = A ν + b
for a matrix A A A whose entries are determined by the curvature class c = c ( E ) c = c(\mathcal{E}) c = c ( E ) , and a constant class b b b . To determine A A A , it suffices to work in a single reference fiber, since parallel transport commutes with restriction to fibers (see (13) ). We pick parabolic weights sufficiently small to admit an embedding from a star-shaped Nakajima quiver variety.
For a star-shaped quiver Q Q Q and dimension vector r ⃗ \vec{r} r determined by the flag dimensions, there is, for suitable α \alpha α , an embedding
T : X ( β , τ ) ↪ M ( α , σ ) \mathcal{T}:\mathcal{X}(\beta,\tau)\hookrightarrow\mathcal{M}(\alpha,\sigma) T : X ( β , τ ) ↪ M ( α , σ )
from a Nakajima quiver variety, where
β i ( k ) = α i ( k ) − α i ( k + 1 ) , τ i ( k ) = σ i ( k ) − σ i ( k + 1 ) . \beta_i^{(k)}=\alpha_i^{(k)}-\alpha_i^{(k+1)},\qquad\tau_i^{(k)}=\sigma_i^{(k)}-\sigma_i^{(k+1)}. β i ( k ) = α i ( k ) − α i ( k + 1 ) , τ i ( k ) = σ i ( k ) − σ i ( k + 1 ) .
This embedding was constructed by Godinho–Mandini in the strongly parabolic rank 2 case [19 ] , and generalized in [38 , 40 ] . Its image is the locus M 0 \mathcal{M}_0 M 0 of Higgs bundles with trivial underlying holomorphic structure. Its complement is a Hitchin section S p ≅ C S_p\cong\mathbb{C} S p ≅ C . The map T \mathcal{T} T intertwines the holomorphic symplectic forms up to a nonzero scalar depending on conventions. The quiver description determines the restriction of the boundary characteristic class to M 0 \mathcal{M}_0 M 0 . To determine the class on the full moduli space, we identify the homology class missing from this open subset.
The ALE topology of the quiver variety, pushed forward into M \mathcal{M} M , gives us n n n homology generators η j \eta_j η j with intersection matrix equal to the negative E n E_n E n Cartan matrix, all of which pair trivially with the class [ F ] [F] [ F ] represented by a smooth Hitchin fiber. In particular, the second Betti numbers are 7 , 8 , 9 7,8,9 7 , 8 , 9 for E 6 , E 7 , E 8 E_6,E_7,E_8 E 6 , E 7 , E 8 , respectively. We compute the characteristic class on the quiver generators using the determinants of the tautological flag bundles F i ( k ) F_i^{(k)} F i ( k ) . Under the quiver embedding, their Chern classes are related to the tautological vertex bundles for the quiver variety, whose pairings with the homology generators are well understood. In particular, we have ⟨ c ν , [ F ] ⟩ = 0 \langle c_\nu,[F]\rangle=0 ⟨ c ν , [ F ]⟩ = 0 , so the fiber periods ⟨ [ ω s , ν ] , [ F ] ⟩ \langle[\omega_{s,\nu}],[F]\rangle ⟨[ ω s , ν ] , [ F ]⟩ are constant in ν \nu ν .
Transporting the quiver generators η v \eta_v η v together with [ F ] [F] [ F ] gives an integral homology basis throughout Λ 0 \Lambda_0 Λ 0 . Here v v v runs over the vertices of the finite E n E_n E n diagram obtained by deleting a rank-one vertex from the affine quiver (see Figure 4 ). Let ν ˙ \dot{\nu} ν ˙ be a variation of ν \nu ν encoding a variation ( α ˙ , σ ˙ ) (\dot{\alpha},\dot{\sigma}) ( α ˙ , σ ˙ ) of ( α , σ ) (\alpha,\sigma) ( α , σ ) , with ( β ˙ , τ ˙ ) (\dot{\beta},\dot{\tau}) ( β ˙ , τ ˙ ) the first variation of the corresponding quiver parameters extended to the central vertex by
Figure 4. The exceptional-type flag arrangements as labeled star-shaped quivers. Deleting the circled vertex gives the corresponding finite Dynkin diagram.
β ˙ ⋆ = ∑ i α ˙ i ( ℓ i ) , τ ˙ ⋆ = ∑ i σ ˙ i ( ℓ i ) . \dot{\beta}_{\star}=\sum_i\dot{\alpha}_i^{(\ell_i)},\qquad\dot{\tau}_{\star}=\sum_i\dot{\sigma}_i^{(\ell_i)}. β ˙ ⋆ = i ∑ α ˙ i ( ℓ i ) , τ ˙ ⋆ = i ∑ σ ˙ i ( ℓ i ) .
Then Definition 6.6 states the period slopes
d P η v ( ν ˙ ) = ( − 4 π 2 d β ˙ v , − 8 π 2 Re d τ ˙ v , − 8 π 2 Im d τ ˙ v ) . dP_{\eta_v}(\dot{\nu})=\left(-4\pi^2d\dot{\beta}_v,-8\pi^2\operatorname{Re}d\dot{\tau}_v,-8\pi^2\operatorname{Im}d\dot{\tau}_v\right). d P η v ( ν ˙ ) = ( − 4 π 2 d β ˙ v , − 8 π 2 Re d τ ˙ v , − 8 π 2 Im d τ ˙ v ) .
This theorem tells us the matrix A \mathbf{A} A in the period formula P ( ν ) = A ν + b \mathcal{P}(\nu)=\mathbf{A}\nu+\mathbf{b} P ( ν ) = A ν + b . To find b \mathbf{b} b , we first calculate the η v \eta_v η v periods in the limit ( α , 0 ) → ( 0 , 0 ) (\alpha,0)\to(0,0) ( α , 0 ) → ( 0 , 0 ) using the U ( 1 ) U(1) U ( 1 ) -action on M ( α , 0 ) \mathcal{M}(\alpha,0) M ( α , 0 ) and its Morse–Bott moment map. Then, we work near a pair ( α , 0 ) (\alpha,0) ( α , 0 ) where the moduli space becomes an orbifold, applying the singular elliptic curve description of [21 ] to prove ⟨ [ ω 1 ] , [ F ] ⟩ = 4 π 2 \langle[\omega_1],[F]\rangle=4\pi^2 ⟨[ ω 1 ] , [ F ]⟩ = 4 π 2 and ⟨ [ Ω ] , [ F ] ⟩ = 0 \langle[\Omega],[F]\rangle=0 ⟨[ Ω ] , [ F ]⟩ = 0 .
Definition 6.9. Let ν ∈ Λ 0 \nu\in\Lambda_0 ν ∈ Λ 0 encode a weights and residues ( α , σ ) (\alpha,\sigma) ( α , σ ) , and define ( β , τ ) (\beta,\tau) ( β , τ ) as in Definition 6.2 . Let B \mathbb{B} B be the admissible basis for H 2 ( M ( α , σ ) , C ) H_2(\mathcal{M}(\alpha,\sigma),\mathbb{C}) H 2 ( M ( α , σ ) , C ) from § 6.3 , indexed by the affine Dynkin vertex set I ~ \widetilde{I} I with distinguished vertex a = v 1 ( 0 ) a=v_1^{(0)} a = v 1 ( 0 ) . For all v ∈ I ~ v\in\widetilde{I} v ∈ I ,
∫ η v ω 1 = 4 π 2 δ v a − 4 π 2 β v , ∫ η v Ω = − 8 π 2 τ v . (3) \int_{\eta_v}\omega_1=4\pi^2\delta_{va}-4\pi^2\beta_v,\qquad\int_{\eta_v}\Omega=-8\pi^2\tau_v. \tag*{(3)} ∫ η v ω 1 = 4 π 2 δ v a − 4 π 2 β v , ∫ η v Ω = − 8 π 2 τ v . (3)
Finally, Definition 6.11 makes this an almost-global surjectivity statement onto the space of allowable Torelli parameters in the corresponding family of ALG spaces.
Relation to Other Work
The standard treatment of moduli spaces of parabolic Higgs bundles fixes a marked Riemann surface ( C , D ) (C,D) ( C , D ) , flags at the markings, as well as parabolic weights and complex residues compatible with the flags. The question of constructing well-behaved moduli spaces of Higgs bundles where at least one of the above structures is allowed to vary has recently received considerable attention [9 , 28 , 10 , 2 , 27 ] .
The particular case of allowing the weights and residues to vary studied in the current manuscript is at least partially motivated by its intimate relationship to the Modularization conjecture:
Conjecture 1.3. Every A L G / A L G ⋆ \mathrm{ALG}/\mathrm{ALG}^{\star} ALG / ALG ⋆ hyperkähler gravitational instanton arises as the L 2 L^{2} L 2 -metric of a moduli space of possibly wild parabolic Higgs bundles.
In [15 ] , the authors computed the period triples of the 12-parameter family of S U ( 2 ) \mathrm{SU}(2) SU ( 2 ) -Hitchin moduli spaces on the four-punctured sphere and showed that the values realize an open dense subset of the admissible Torelli data of the 12-parameter family of A L G \mathrm{ALG} ALG -D 4 D_{4} D 4 gravitational instantons, thus essentially proving the Modularization conjecture modulo showing that the L 2 L^{2} L 2 -metrics have finite energy ∫ M ∣ R m ∣ 2 < ∞ \int_{\mathcal{M}}|\mathrm{Rm}|^{2}<\infty ∫ M ∣ Rm ∣ 2 < ∞ . The failure of actual surjectivity is because for particular values of the weight the associated moduli space in the literature becomes zero-dimensional and as a result does not correspond to an instanton; however, there should be a different four-dimensional Hitchin moduli space obtained by quotienting by a smaller gauge group which one can instead use.
The main application of the current paper is to establish the analogous result for E 6 , E 7 , E 8 E_{6},E_{7},E_{8} E 6 , E 7 , E 8 which, modulo showing that the associated L 2 L^{2} L 2 -metrics have finite energy, essentially settles the Modularization conjecture for all ALG metrics associated to moduli spaces of tame parabolic Higgs bundles. The analytic tools we develop are influenced by the ones appearing in [9 ] , which was the first paper carefully treating the moduli space problem when the complex residues are allowed to vary.
At this point it makes sense to explain how our treatment differs from [9 ] where the authors construct the framed moduli space where the residues are allowed to vary and only allow full flags. In our paper we only retain a determinant unitary framing which gives rise to the real torus bundle C \mathcal{C} C . We suspect that both constructions can lead to incomplete hyperkähler metrics. One example of this is the moduli space with extended Hitchin base C σ \mathcal{C}_{\sigma} C σ parameterizing q = − det φ = ( z 2 + σ ) d z 2 q=-\det\varphi=(z^{2}+\sigma)\mathrm{d}z^{2} q = − det φ = ( z 2 + σ ) d z 2 on C P 1 \mathbb{CP}^{1} CP 1 with an irregular singularity at infinity. The eigenvalues at infinity in coordinate z ~ = z − 1 \widetilde{z}=z^{-1} z = z − 1 are
λ ± = ± ( 1 z ~ 3 + σ 2 z ~ + O ( z ~ ) ) d z ~ . \lambda_{\pm}=\pm\left(\frac{1}{\widetilde{z}^{3}}+\frac{\sigma}{2\widetilde{z}}+O(\widetilde{z})\right)\mathrm{d}\widetilde{z}. λ ± = ± ( z 3 1 + 2 z σ + O ( z ) ) d z .
so one sees that changing σ \sigma σ corresponds to changing the residue of the eigenvalue associated to z = ∞ z=\infty z = ∞ , while leaving the higher-order polar part fixed. Gaiotto–Moore–Neitzke implicitly conjecture that this case should recover the well-known Ooguri–Vafa metric. Progress towards this perspective appears in the work of Tulli [Tul25] and Nackan [35 ] . We conjecture that after fixing a compact subset K ⊂ Λ × B K \subset\Lambda\times\mathcal{B} K ⊂ Λ × B of the product of the parameter space times the Hitchin base, its pre-image in E \mathcal{E} E should also admit a hyperkähler metric.
Weighted Sobolev Spaces
In this subsection we state our conventions for the weighted Sobolev spaces appearing in the paper and some estimates used explicitly or implicitly in various places.
Write C ∘ = C ∖ D C^\circ= C \setminus D C ∘ = C ∖ D , and choose disjoint coordinate disks centered at the punctures, with z = r e i θ z = r e^{i\theta} z = r e i θ and B ε ( p i ) = { ∣ z ∣ < ε } B_\varepsilon(p_i) = \{|z| < \varepsilon\} B ε ( p i ) = { ∣ z ∣ < ε } for 0 < ε ≤ 1 0 < \varepsilon\le1 0 < ε ≤ 1 . Write B ε ∘ ( p i ) = B ε ( p i ) ∖ { p i } B^\circ_\varepsilon(p_i) = B_\varepsilon(p_i) \setminus\{p_i\} B ε ∘ ( p i ) = B ε ( p i ) ∖ { p i } for the punctured disk. Fix once and for all a smooth metric g C g_C g C compatible with the complex structure on C C C . In a local coordinate z z z , write g C = κ ( z ) ∣ d z ∣ 2 g_C = \kappa(z)|dz|^2 g C = κ ( z ) ∣ d z ∣ 2 .
Let C ^ → C \widehat{C} \to C C → C be the real oriented blowup along D D D . Its b b b -vector fields are the smooth vector fields tangent to the boundary, locally generated by r ∂ r r\partial_r r ∂ r , ∂ θ \partial_\theta ∂ θ near the boundary and generating the usual tangent sheaf away from it. Together with smooth functions they generate a differential-operator algebra D b \mathcal{D}_b D b .
For a sheaf of sections S \mathcal{S} S of a vector bundle on C ^ \widehat{C} C with smooth fiber metric and connection, let H b q ( U , S ) H_b^q(U,\mathcal{S}) H b q ( U , S ) be the Sobolev completion of its smooth sections of finite norm over U ⊂ C ^ U \subset\widehat{C} U ⊂ C , using covariant derivatives of order up to q q q with respect to b b b -vector fields and the surface measure d vol g C d\operatorname{vol}_{g_C} d vol g C . Thus the measure near a puncture is κ r d r d θ \kappa r\,dr\,d\theta κ r d r d θ , rather than the cylindrical measure.
Choose positive functions ρ i \rho_i ρ i , smooth on C ∘ C^\circ C ∘ , equal to r r r on B ε ∘ ( p i ) B^\circ_\varepsilon(p_i) B ε ∘ ( p i ) and to 1 1 1 outside slightly larger disjoint coordinate disks, and set ρ = ∏ i ρ i \rho= \prod_i \rho_i ρ = ∏ i ρ i . For δ ∈ R \delta\in\mathbb{R} δ ∈ R , define
H δ q ( U , S ) = ρ δ + 1 H b q ( U , S ) , ∥ u ∥ H δ q = ∥ ρ − δ − 1 u ∥ H b q . (4) H_\delta^q(U,\mathcal{S}) = \rho^{\delta+1}H_b^q(U,\mathcal{S}), \qquad\|u\|_{H_\delta^q} = \|\rho^{-\delta-1}u\|_{H_b^q}. \tag*{(4)} H δ q ( U , S ) = ρ δ + 1 H b q ( U , S ) , ∥ u ∥ H δ q = ∥ ρ − δ − 1 u ∥ H b q . (4)
The exponent is chosen so that r a ∈ H δ q ( B ε ∘ ( p i ) ) r^a \in H_\delta^q(B^\circ_\varepsilon(p_i)) r a ∈ H δ q ( B ε ∘ ( p i )) if and only if a > δ a > \delta a > δ . We sometimes omit U , S U,\mathcal{S} U , S when the context is clear, and use the same notation for a puncture neighborhood and its preimage in C ^ \widehat{C} C . For sheaves of differential forms, we work with the b b b -cotangent bundle locally generated by d r / r , d θ dr/r,d\theta d r / r , d θ , and write Ω C ^ q \Omega_{\widehat{C}}^q Ω C q for its sheaf of smooth exterior q q q -forms.
We will use these spaces to define configuration space and gauge group in terms of model singular parts and decaying remainders. In a smooth local frame near p i p_i p i , the norm is equivalent to
∥ u ∥ H δ q 2 ∼ ∑ a + b ≤ q ∫ 0 1 ∫ 0 2 π r − 2 δ ∣ ( r ∂ r ) a ∂ θ b u ∣ 2 d r r d θ . (5) \|u\|_{H_\delta^q}^2 \sim\sum_{a+b\le q}\int_0^1\int_0^{2\pi} r^{-2\delta}\left|(r\partial_r)^a\partial_\theta^b u\right|^2\frac{dr}{r}\,d\theta. \tag*{(5)} ∥ u ∥ H δ q 2 ∼ a + b ≤ q ∑ ∫ 0 1 ∫ 0 2 π r − 2 δ ( r ∂ r ) a ∂ θ b u 2 r d r d θ . (5)
These spaces will be defined in terms of sections of vector bundles equipped with a background metric, giving the norm used above.
Much of our analytic construction is inspired by [4 ] , but we use b b b -derivatives in place of their weighted covariant derivatives. At order zero, our H δ 0 H_\delta^0 H δ 0 agrees with their L δ 2 L_\delta^2 L δ 2 on functions, while our coefficient convention on 1 1 1 -forms gives H δ 0 ( Ω C ^ 1 ) = L δ − 1 2 ( Ω 1 ) H_\delta^0(\Omega_{\widehat{C}}^1)=L_{\delta-1}^2(\Omega^1) H δ 0 ( Ω C 1 ) = L δ − 1 2 ( Ω 1 ) , with equivalent norms.
Useful Estimates
∑ a + b ≤ m sup B 0 1 ( p i ) r − δ ∣ ( r ∂ r ) a ∂ θ b u ∣ ≤ C ∥ u ∥ H δ q . (6) \sum_{a+b\le m}\sup_{B^1_0(p_i)}r^{-\delta}\left|(r\partial_r)^a\partial_\theta^b u\right|\le C\|u\|_{\mathcal{H}^q_\delta}. \tag*{(6)} a + b ≤ m ∑ B 0 1 ( p i ) sup r − δ ( r ∂ r ) a ∂ θ b u ≤ C ∥ u ∥ H δ q . (6)
H δ q ↪ H δ ′ q ′ . (7) \mathcal{H}^q_\delta\hookrightarrow\mathcal{H}^{q'}_{\delta'}. \tag*{(7)} H δ q ↪ H δ ′ q ′ . (7)
An operator P ∈ D b P\in\mathcal{D}_b P ∈ D b of order j ≤ q j\le q j ≤ q acts continuously from H δ q \mathcal{H}^q_\delta H δ q to H δ q − j \mathcal{H}^{q-j}_\delta H δ q − j .
∥ u v ∥ H δ + η k ≤ C ∥ u ∥ H δ k ∥ v ∥ H η k . (8) \|uv\|_{\mathcal{H}^k_{\delta+\eta}}\le C\|u\|_{\mathcal{H}^k_\delta}\|v\|_{\mathcal{H}^k_\eta}. \tag*{(8)} ∥ uv ∥ H δ + η k ≤ C ∥ u ∥ H δ k ∥ v ∥ H η k . (8)
Smooth coefficients on C ^ \widehat{C} C act as bounded multipliers.
∫ C ∘ ∣ ⟨ u , v ⟩ ∣ d V b ≤ ∥ u ∥ H δ 0 ∥ v ∥ H − δ 0 . (9) \int_{C^\circ}| \langle u,v\rangle|\mathrm{d}V_b\le\|u\|_{\mathcal{H}^0_\delta}\|v\|_{\mathcal{H}^0_{-\delta}}. \tag*{(9)} ∫ C ∘ ∣ ⟨ u , v ⟩ ∣ d V b ≤ ∥ u ∥ H δ 0 ∥ v ∥ H − δ 0 . (9)
∫ 0 r f ( s , θ ) d s s ∈ H δ q ( B 0 ∘ ( p i ) ) . (10) \int_0^r f(s,\theta)\frac{\mathrm{d}s}{s}\in\mathcal{H}^q_\delta(B^\circ_0(p_i)). \tag*{(10)} ∫ 0 r f ( s , θ ) s d s ∈ H δ q ( B 0 ∘ ( p i )) . (10)
⋆ d r r = d θ , ⋆ d θ = − d r r , ⋆ ( d r r ∧ d θ ) = 1 κ r 2 , ⋆ 1 = κ r 2 d r r ∧ d θ . (11) \star\frac{\mathrm{d}r}{r}=\mathrm{d}\theta,\qquad
\star\mathrm{d}\theta=-\frac{\mathrm{d}r}{r},\qquad
\star\left(\frac{\mathrm{d}r}{r}\wedge\mathrm{d}\theta\right)=\frac{1}{\kappa r^2},\qquad
\star1=\kappa r^2\frac{\mathrm{d}r}{r}\wedge\mathrm{d}\theta. \tag*{(11)} ⋆ r d r = d θ , ⋆ d θ = − r d r , ⋆ ( r d r ∧ d θ ) = κ r 2 1 , ⋆ 1 = κ r 2 r d r ∧ d θ . (11)
Acknowledgements
Acknowledgements. We thank Rafe Mazzeo, Andy Neitzke, Richard Wentworth, and Max Zimet for inspiring conversations over the years which refined the ideas presented in this paper. LF is partially supported by NSF DMS-2005258.
Use of AI. We used ChatGPT’s Astra 6 and DeepSeek V4.1 Flash to help digest existing literature, correct typographical mistakes, and write code for our figures. The ideas are our own. The authors are responsible for the mathematical content and the accuracy of the references.
A Generalized Duistermaat–Heckman Theorem
Consider a family { M α } α \{\mathcal{M}_\alpha\}_\alpha { M α } α of moduli spaces for α \alpha α in some parameter space Λ \Lambda Λ , with each M α \mathcal{M}_\alpha M α equipped with a symplectic form ω α \omega_\alpha ω α . The goal of this section is to describe how the cohomology class of ω α \omega_\alpha ω α depends on the choice of α \alpha α , which we do by adapting Duistermaat and Heckman’s theorem [13 Theorem 1.1] to have suitably weak hypotheses that can apply to noncompact quotients of infinite-dimensional configuration spaces. Namely, we give a geometric hypothesis and a topological hypothesis.
Assume the family of moduli spaces can be joined into a smooth family π : M → Λ \pi:\mathcal{M}\to\Lambda π : M → Λ , so that M α = π − 1 ( α ) \mathcal{M}_\alpha=\pi^{-1}(\alpha) M α = π − 1 ( α ) . Under ideal circumstances, the family arises from a single quotient of a single configuration space C \mathcal{C} C by an action G ↷ C G \curvearrowright\mathcal{C} G ↷ C which is Hamiltonian with respect to a symplectic form ω \omega ω on C \mathcal{C} C . In our convention, the moment map μ : C → g ∗ \mu:\mathcal{C}\to\mathfrak{g}^{*} μ : C → g ∗ satisfies
ι ζ ξ ω = d ⟨ μ , ξ ⟩ (12) \iota_{\zeta_\xi}\omega=d\langle\mu,\xi\rangle \tag*{(12)} ι ζ ξ ω = d ⟨ μ , ξ ⟩ (12)
where ζ ξ \zeta_\xi ζ ξ is any vector field on C \mathcal{C} C generated by some ξ ∈ g \xi\in\mathfrak{g} ξ ∈ g . In this situation, Λ ⊂ g ∗ \Lambda\subset\mathfrak{g}^{*} Λ ⊂ g ∗ and μ \mu μ descends to π \pi π after restricting to μ − 1 ( Λ ) \mu^{-1}(\Lambda) μ − 1 ( Λ ) and taking quotients. Also ideally, π : M → Λ \pi:\mathcal{M}\to\Lambda π : M → Λ admits a complete Ehresmann connection, allowing one to lift paths in Λ \Lambda Λ to paths in M \mathcal{M} M . This gives identifications between moduli spaces used to compare the classes [ ω α ] [\omega_\alpha] [ ω α ] . § 2.2 demonstrates this ideal situation in the well-studied multi-Taub-NUT family.
In the case of parabolic Higgs bundle moduli spaces, existence of such an ω \omega ω and completeness of an Ehresmann connection are substantial obstacles. This motivates our careful setup for Definition 2.3, which softens both hypotheses. The main features are local systems for homology, and a procedure for reducing to finite dimensions.
Affine Period Variation Theorem
Throughout this subsection, we work over F = R \mathbb{F}=\mathbb{R} F = R or F = C \mathbb{F}=\mathbb{C} F = C and take all Lie groups and Lie algebras to be respectively real or complex. We also assume all differential forms are F \mathbb{F} F -valued. We begin with a definition pertaining to the topological hypothesis of Definition 2.3.
Definition 2.1. Let π : M → Λ \pi:\mathcal{M}\to\Lambda π : M → Λ be a continuous family of smooth manifolds M α = π − 1 ( α ) \mathcal{M}_\alpha=\pi^{-1}(\alpha) M α = π − 1 ( α ) , with inclusions i α : M α ↪ M i_\alpha:\mathcal{M}_\alpha\hookrightarrow\mathcal{M} i α : M α ↪ M . A homology local system for this family consists of the vector spaces H 2 ( M α , F ) H_2(\mathcal{M}_\alpha,\mathbb{F}) H 2 ( M α , F ) and parallel transport isomorphisms
PT γ : H 2 ( M α ∗ , F ) ⟶ H 2 ( M α , F ) \operatorname{PT}_\gamma:H_2(\mathcal{M}_{\alpha_*},\mathbb{F})\longrightarrow H_2(\mathcal{M}_\alpha,\mathbb{F}) PT γ : H 2 ( M α ∗ , F ) ⟶ H 2 ( M α , F ) for paths γ \gamma γ in Λ \Lambda Λ from α ∗ \alpha_* α ∗ to α \alpha α . Transport depends only on homotopy relative to endpoints, respects concatenation, and is the identity on constant paths. The homology local system is compatible with π \pi π if, for every path γ \gamma γ in Λ \Lambda Λ , the following diagram commutes:
H 2 ( M α ∗ , F ) → PT γ H 2 ( M α , F ) ( i α ∗ ) ∗ ↓ ↓ ( i α ) ∗ H 2 ( M , F ) = H 2 ( M , F ) (13) \begin{CD}
H_2(\mathcal{M}_{\alpha_*},\mathbb{F}) @>{\operatorname{PT}_\gamma}>> H_2(\mathcal{M}_\alpha,\mathbb{F})\\
@V{(i_{\alpha_*})_*}VV @VV{(i_\alpha)_*}V\\
H_2(\mathcal{M},\mathbb{F}) @= H_2(\mathcal{M},\mathbb{F}) \tag*{(13)}
\end{CD} H 2 ( M α ∗ , F ) ( i α ∗ ) ∗ ↓ ⏐ H 2 ( M , F ) PT γ H 2 ( M α , F ) ↓ ⏐ ( i α ) ∗ H 2 ( M , F ) (13)
Thus a cycle and its transport are homologous in the total space M \mathcal{M} M . We write PT γ ∗ \operatorname{PT}_\gamma^* PT γ ∗ for the dual transport on degree-two cohomology.
For many moduli spaces of interest, one can take partial quotients to reduce from infinite dimensions to to finite dimensions. The first advantage of this reduction is that we can interpret the coefficients of the affine period map in terms of characteristic classes of a finite-dimensional principal bundle E E E over M \mathcal{M} M . The second advantage is that the geometric hypothesis of Definition 2.3 can be purely stated in finite-dimensional terms. Definition 2.2 below shows how the ideal situation described above automatically gives the structure needed for geometric hypothesis.
Let G ↷ ( M , ω ) G\curvearrowright(M,\omega) G ↷ ( M , ω ) with moment map μ \mu μ as in this section’s introduction. Fix a finite-dimensional parameter subspace Λ ⊆ ( g ∗ ) G \Lambda\subseteq(\mathfrak{g}^{*})^{G} Λ ⊆ ( g ∗ ) G , and a closed normal Lie subgroup H ⊆ G H\subseteq G H ⊆ G whose Lie algebra annihilates Λ \Lambda Λ , such that B = G / H B=G/H B = G / H is finite dimensional. Set E = μ − 1 ( Λ ) / H E=\mu^{-1}(\Lambda)/H E = μ − 1 ( Λ ) / H , and let H e f f H^{\mathrm{eff}} H eff denote the quotient of H H H by the kernel of its action on μ − 1 ( Λ ) \mu^{-1}(\Lambda) μ − 1 ( Λ ) . Suppose μ − 1 ( Λ ) \mu^{-1}(\Lambda) μ − 1 ( Λ ) is smooth and p : μ − 1 ( Λ ) ⟶ E p:\mu^{-1}(\Lambda)\longrightarrow E p : μ − 1 ( Λ ) ⟶ E is a smooth principal H e f f H^{\mathrm{eff}} H eff -bundle.
Lemma 2.2. In the above setting, the restriction of ω \omega ω to μ − 1 ( Λ ) \mu^{-1}(\Lambda) μ − 1 ( Λ ) descends to a closed B B B -invariant 2-form ϖ \varpi ϖ on the partial quotient E E E , and μ \mu μ descends to π ~ : E ⟶ Λ ⊂ b ∗ \widetilde{\pi}:E\longrightarrow\Lambda\subset\mathfrak{b}^{*} π : E ⟶ Λ ⊂ b ∗ such that
ι ζ b ϖ = d ⟨ π ~ , b ⟩ , b ∈ b . (14) \iota_{\zeta_b}\varpi=d\langle\widetilde{\pi},b\rangle,\qquad b\in\mathfrak{b}. \tag*{(14)} ι ζ b ϖ = d ⟨ π , b ⟩ , b ∈ b . (14) Moreover, π ~ − 1 ( α ) / B = μ − 1 ( α ) / G = M α \widetilde{\pi}^{-1}(\alpha)/B=\mu^{-1}(\alpha)/G=M_\alpha π − 1 ( α ) / B = μ − 1 ( α ) / G = M α .
Proof. Let j : μ − 1 ( Λ ) ↪ M j:\mu^{-1}(\Lambda)\hookrightarrow M j : μ − 1 ( Λ ) ↪ M . For ξ ∈ h \xi\in\mathfrak{h} ξ ∈ h , the function ⟨ μ , ξ ⟩ \langle\mu,\xi\rangle ⟨ μ , ξ ⟩ vanishes on μ − 1 ( Λ ) \mu^{-1}(\Lambda) μ − 1 ( Λ ) because h \mathfrak{h} h annihilates Λ \Lambda Λ , so ι ζ ξ j ∗ ω = 0 \iota_{\zeta_\xi}j^{*}\omega=0 ι ζ ξ j ∗ ω = 0 . Thus j ∗ ω j^{*}\omega j ∗ ω is H H H -basic and descends to a closed B B B -invariant form ϖ \varpi ϖ . Since μ \mu μ is constant on G G G -orbits in μ − 1 ( Λ ) \mu^{-1}(\Lambda) μ − 1 ( Λ ) , it descends to the map π ~ \widetilde{\pi} π . For a lift ξ ∈ g \xi\in\mathfrak{g} ξ ∈ g of b ∈ b b\in\mathfrak{b} b ∈ b ,
p ∗ ( ι ζ b ϖ ) = ι ζ ξ j ∗ ω = p ∗ d ⟨ π ~ , b ⟩ . (15) p^{*}(\iota_{\zeta_b}\varpi)=\iota_{\zeta_\xi}j^{*}\omega=p^{*}d\langle\widetilde{\pi},b\rangle. \tag*{(15)} p ∗ ( ι ζ b ϖ ) = ι ζ ξ j ∗ ω = p ∗ d ⟨ π , b ⟩ . (15)
The pairing is independent of the lift, again because h \mathfrak{h} h annihilates Λ \Lambda Λ . The quotient identification follows by taking subsequent quotients by H H H and then by G / H G/H G / H . □
The conclusion of the lemma is the setting in which we state Definition 2.3 . Having reduced, everything should be taken to be finite-dimensional over F = R \mathbb{F}=\mathbb{R} F = R or F = C \mathbb{F}=\mathbb{C} F = C .
Theorem 2.3. Let M \mathcal{M} M be a paracompact manifold, let B B B be a Lie group, and fix a coadjoint-invariant subspace Λ ⊂ ( b ∗ ) B \Lambda\subset(\mathfrak{b}^{*})^{B} Λ ⊂ ( b ∗ ) B . Suppose q : E ⟶ M q:E\longrightarrow\mathcal{M} q : E ⟶ M is a principal B B B -bundle, and π ~ : E ⟶ Λ \widetilde{\pi}:E\longrightarrow\Lambda π : E ⟶ Λ is a B B B -invariant map descending to a map π : M ⟶ Λ \pi:\mathcal{M}\longrightarrow\Lambda π : M ⟶ Λ .
Suppose there is an open region Λ 0 ⊆ Λ \Lambda_{0}\subseteq\Lambda Λ 0 ⊆ Λ such that the restrictions E 0 = π ~ − 1 ( Λ 0 ) E_{0}=\widetilde{\pi}^{-1}(\Lambda_{0}) E 0 = π − 1 ( Λ 0 ) and M 0 = π − 1 ( Λ 0 ) \mathcal{M}_{0}=\pi^{-1}(\Lambda_{0}) M 0 = π − 1 ( Λ 0 ) are smooth, and all the maps above are smooth after restriction. Assume further that π : M 0 ⟶ Λ 0 \pi:\mathcal{M}_{0}\longrightarrow\Lambda_{0} π : M 0 ⟶ Λ 0 is a submersion, and the fibers M α \mathcal{M}_{\alpha} M α have symplectic forms ω α \omega_{\alpha} ω α . Write i α i_{\alpha} i α , j α j_{\alpha} j α , and q α q_{\alpha} q α for the inclusions and submersion in the following diagram:
π ~ − 1 ( α ) → j α E 0 q α ↓ q α q ↓ q ↘ π ~ M α → i α M 0 → π Λ 0 (16) \begin{aligned}
\widetilde{\pi}^{-1}(\alpha)&\xrightarrow{\ j_{\alpha}\ }E_{0}\\
\mathllap{q_{\alpha}}\downarrow\phantom{q_{\alpha}}&\phantom{q}\downarrow\mathrlap{q}\qquad\searrow\widetilde{\pi}\\
\mathcal{M}_{\alpha}&\xrightarrow{\ i_{\alpha}\ }\mathcal{M}_{0}\xrightarrow{\ \pi\ }\Lambda_{0} \tag*{(16)}
\end{aligned} π − 1 ( α ) q α ↓ q α M α j α E 0 q ↓ q ↘ π i α M 0 π Λ 0 (16) We additionally assume the following:
ι ζ b ϖ = d ⟨ π ~ , b ⟩ , b ∈ b , (17) \iota_{\zeta_b}\varpi=d\langle\widetilde{\pi},b\rangle,\qquad b\in\mathfrak{b}, \tag*{(17)} ι ζ b ϖ = d ⟨ π , b ⟩ , b ∈ b , (17) where ζ b \zeta_b ζ b is the vector field generated by b b b . For each α ∈ Λ 0 \alpha\in\Lambda_{0} α ∈ Λ 0 , the restriction of ϖ \varpi ϖ to π ~ − 1 ( α ) \widetilde{\pi}^{-1}(\alpha) π − 1 ( α ) descends to ω α \omega_{\alpha} ω α .
Then there is a class c ∈ H 2 ( M 0 , F ) ⊗ F Λ ∗ c \in H^{2}(\mathcal{M}_{0},\mathbb{F}) \otimes_{\mathbb{F}} \Lambda^{*} c ∈ H 2 ( M 0 , F ) ⊗ F Λ ∗ such that, for every α 0 , α 1 ∈ Λ 0 \alpha_{0},\alpha_{1} \in\Lambda_{0} α 0 , α 1 ∈ Λ 0 and every path γ \gamma γ in Λ 0 \Lambda_{0} Λ 0 from α 0 \alpha_{0} α 0 to α 1 \alpha_{1} α 1 ,
PT γ ∗ [ ω α 1 ] = [ ω α 0 ] − ⟨ c α 0 , α 1 − α 0 ⟩ . (18) \operatorname{PT}_{\gamma}^{*}[\omega_{\alpha_{1}}]=[\omega_{\alpha_{0}}]-\langle c_{\alpha_{0}},\alpha_{1}-\alpha_{0}\rangle. \tag*{(18)} PT γ ∗ [ ω α 1 ] = [ ω α 0 ] − ⟨ c α 0 , α 1 − α 0 ⟩ . (18)
Here c c c is the curvature class of the principal B B B -bundle q : E 0 → M 0 q:E_{0}\to\mathcal{M}_{0} q : E 0 → M 0 and c α = i α ∗ c c_{\alpha}=i_{\alpha}^{*}c c α = i α ∗ c .
Proof. This proof is an adaptation of [13 ] . The strategy is to modify ϖ \varpi ϖ by an exact form so that it descends to M 0 \mathcal{M}_{0} M 0 . Choose a smooth principal connection θ \theta θ on q : E 0 → M 0 q:E_{0}\to\mathcal{M}_{0} q : E 0 → M 0 satisfying θ ( ζ b ) = b \theta(\zeta_{b})=b θ ( ζ b ) = b . Such a connection exists by standard partition of unity arguments.
Let r : b → Λ ∗ r:\mathfrak{b}\to\Lambda^{*} r : b → Λ ∗ be the evaluation map, and set L = r ∘ θ L=r\circ\theta L = r ∘ θ . Since Λ \Lambda Λ is coadjoint-invariant, L L L is a B B B -invariant one-form satisfying L ( ζ b ) = r ( b ) L(\zeta_{b})=r(b) L ( ζ b ) = r ( b ) . Cartan’s formula gives
ι ζ b d L = L ζ b L − d ( L ( ζ b ) ) = − d ( r ( b ) ) = 0. (19) \iota_{\zeta_{b}}dL=\mathcal{L}_{\zeta_{b}}L-d(L(\zeta_{b}))=-d(r(b))=0. \tag*{(19)} ι ζ b d L = L ζ b L − d ( L ( ζ b )) = − d ( r ( b )) = 0. (19)
Hence d L dL d L descends to a closed Λ ∗ \Lambda^{*} Λ ∗ -valued two-form Ω \Omega Ω on M 0 \mathcal{M}_{0} M 0 satisfying
q ∗ Ω = d L = d ( r ∘ θ ) = r ( F θ ) , (20) q^{*}\Omega=dL=d(r\circ\theta)=r(F_{\theta}), \tag*{(20)} q ∗ Ω = d L = d ( r ∘ θ ) = r ( F θ ) , (20)
where F θ = d θ + 1 2 [ θ ∧ θ ] F_{\theta}=d\theta+\frac{1}{2}[\theta\wedge\theta] F θ = d θ + 2 1 [ θ ∧ θ ] is the curvature of θ \theta θ , and the last equality uses the fact that r r r vanishes on Lie brackets by adjoint-invariance. Changing θ \theta θ changes L L L by a basic one-form and hence Ω \Omega Ω by an exact form. Thus the projected curvature class c ( E 0 ) = [ Ω ] c(E_{0})=[\Omega] c ( E 0 ) = [ Ω ] is independent of θ \theta θ .
The modified 2-form ϖ + d ⟨ L , π ~ ⟩ \varpi+d\langle L,\widetilde{\pi}\rangle ϖ + d ⟨ L , π ⟩ is closed and B B B -invariant. We also have
ι ζ b d ⟨ L , π ~ ⟩ = − d ⟨ L ( ζ b ) , π ~ ⟩ = − d ⟨ π ~ , b ⟩ = − ι ζ b ϖ , (21) \iota_{\zeta_{b}}d\langle L,\widetilde{\pi}\rangle=-d\langle L(\zeta_{b}),\widetilde{\pi}\rangle=-d\langle\widetilde{\pi},b\rangle=-\iota_{\zeta_{b}}\varpi, \tag*{(21)} ι ζ b d ⟨ L , π ⟩ = − d ⟨ L ( ζ b ) , π ⟩ = − d ⟨ π , b ⟩ = − ι ζ b ϖ , (21)
which implies ϖ + d ⟨ L , π ~ ⟩ \varpi+d\langle L,\widetilde{\pi}\rangle ϖ + d ⟨ L , π ⟩ is basic and descends to a closed two-form β \beta β on M 0 \mathcal{M}_{0} M 0 with
q ∗ β = ϖ + d ⟨ L , π ~ ⟩ . q^{*}\beta=\varpi+d\langle L,\widetilde{\pi}\rangle. q ∗ β = ϖ + d ⟨ L , π ⟩ .
To restrict this identity to the fiber over α \alpha α , consider the commutative square
π ~ − 1 ( α ) → j α E 0 q α ↓ ↓ q M α → i α M 0 . (22) \begin{CD}
\widetilde{\pi}^{-1}(\alpha) @>{j_{\alpha}}>> E_{0}\\
@V{q_{\alpha}}VV @VV{q}V\\
\mathcal{M}_{\alpha} @>{i_{\alpha}}>> \mathcal{M}_{0}. \tag*{(22)}
\end{CD} π − 1 ( α ) q α ↓ ⏐ M α j α i α E 0 ↓ ⏐ q M 0 . (22)
The restriction of d ⟨ L , π ~ ⟩ d\langle L,\widetilde{\pi}\rangle d ⟨ L , π ⟩ to π ~ − 1 ( α ) \widetilde{\pi}^{-1}(\alpha) π − 1 ( α ) is j α ∗ ⟨ d L , α ⟩ j_{\alpha}^{*}\langle dL,\alpha\rangle j α ∗ ⟨ d L , α ⟩ , and by the geometric hypothesis j α ∗ ϖ = q α ∗ ω α j_{\alpha}^{*}\varpi=q_{\alpha}^{*}\omega_{\alpha} j α ∗ ϖ = q α ∗ ω α . Thus
j α ∗ q ∗ β = q α ∗ ω α + j α ∗ ⟨ d L , α ⟩ = q α ∗ ω α + j α ∗ q ∗ ⟨ Ω , α ⟩ . (23) j_{\alpha}^{*}q^{*}\beta=q_{\alpha}^{*}\omega_{\alpha}+j_{\alpha}^{*}\langle dL,\alpha\rangle=q_{\alpha}^{*}\omega_{\alpha}+j_{\alpha}^{*}q^{*}\langle\Omega,\alpha\rangle. \tag*{(23)} j α ∗ q ∗ β = q α ∗ ω α + j α ∗ ⟨ d L , α ⟩ = q α ∗ ω α + j α ∗ q ∗ ⟨ Ω , α ⟩ . (23)
The diagram (22) says j α ∗ q ∗ = q α ∗ i α ∗ j_{\alpha}^{*}q^{*}=q_{\alpha}^{*}i_{\alpha}^{*} j α ∗ q ∗ = q α ∗ i α ∗ . Since q α q_{\alpha} q α is a surjective submersion, q α ∗ q_{\alpha}^{*} q α ∗ is injective, so the above implies
ω α = i α ∗ β − ⟨ i α ∗ Ω , α ⟩ . (24) \omega_{\alpha}=i_{\alpha}^{*}\beta-\langle i_{\alpha}^{*}\Omega,\alpha\rangle. \tag*{(24)} ω α = i α ∗ β − ⟨ i α ∗ Ω , α ⟩ . (24)
Let c = c ( E 0 ) c=c(E_{0}) c = c ( E 0 ) be the curvature class, and write c α = i α ∗ c = i α ∗ [ Ω ] c_{\alpha}=i_{\alpha}^{*}c=i_{\alpha}^{*}[\Omega] c α = i α ∗ c = i α ∗ [ Ω ] . By the topological hypothesis, parallel transport commutes with the family of maps ( i α ) ∗ : H 2 ( M α , F ) → H 2 ( M 0 , F ) (i_{\alpha})_{*}:H_{2}(\mathcal{M}_{\alpha},\mathbb{F})\to H_{2}(\mathcal{M}_{0},\mathbb{F}) ( i α ) ∗ : H 2 ( M α , F ) → H 2 ( M 0 , F ) (see (13) ), and duality gives the analogous statement for cohomology. For a path γ \gamma γ from α 0 \alpha_{0} α 0 to α 1 \alpha_{1} α 1 in Λ 0 \Lambda_{0} Λ 0 ,
PT γ ∗ i α 1 ∗ [ β ] = i α 0 ∗ [ β ] , PT γ ∗ c α 1 = c α 0 . (25) \operatorname{PT}_{\gamma}^{*}i_{\alpha_{1}}^{*}[\beta]=i_{\alpha_{0}}^{*}[\beta],\qquad\operatorname{PT}_{\gamma}^{*}c_{\alpha_{1}}=c_{\alpha_{0}}. \tag*{(25)} PT γ ∗ i α 1 ∗ [ β ] = i α 0 ∗ [ β ] , PT γ ∗ c α 1 = c α 0 . (25)
Applying these identities to (24) gives
PT γ ∗ [ ω α 1 ] = ι α 0 ∗ [ β ] − ⟨ c α 0 , α 1 ⟩ = [ ω α 0 ] + ⟨ c α 0 , α 0 ⟩ − ⟨ c α 0 , α 1 ⟩ = [ ω α 0 ] − ⟨ c α 0 , α 1 − α 0 ⟩ , \operatorname{PT}_{\gamma}^{*}[\omega_{\alpha_1}] = \iota_{\alpha_0}^{*}[\beta] - \langle c_{\alpha_0},\alpha_1\rangle
= [\omega_{\alpha_0}] + \langle c_{\alpha_0},\alpha_0\rangle- \langle c_{\alpha_0},\alpha_1\rangle
= [\omega_{\alpha_0}] - \langle c_{\alpha_0},\alpha_1-\alpha_0\rangle, PT γ ∗ [ ω α 1 ] = ι α 0 ∗ [ β ] − ⟨ c α 0 , α 1 ⟩ = [ ω α 0 ] + ⟨ c α 0 , α 0 ⟩ − ⟨ c α 0 , α 1 ⟩ = [ ω α 0 ] − ⟨ c α 0 , α 1 − α 0 ⟩ ,
as required. □
A stronger, sufficient condition for the topological hypothesis is the existence of a complete Ehresmann connection. An Ehresmann connection on π : M 0 → Λ 0 \pi:\mathcal{M}_0\to\Lambda_0 π : M 0 → Λ 0 consists of horizontal tangent spaces complementary to ker d π \ker d\pi ker d π , giving uniquely-defined lifts of vector fields on Λ 0 \Lambda_0 Λ 0 to M 0 \mathcal{M}_0 M 0 . It is complete if horizontal lifts of complete vector fields remain complete.
Corollary 2.4. Under the geometric hypothesis of Definition 2.3, suppose π : M 0 → Λ 0 \pi:\mathcal{M}_0\to\Lambda_0 π : M 0 → Λ 0 admits a smooth and complete Ehresmann connection. Then π \pi π is a fiber bundle and satisfies the topological hypothesis of the theorem.
Proof. Fix a convex neighborhood U U U of a parameter α ∗ \alpha_* α ∗ . Horizontal transport along straight segments from α ∗ \alpha_* α ∗ gives a smooth trivialization over U U U . By homotopy invariance, the induced transport on homology depends only on the homotopy class of the path relative to its endpoints, so the fiber homology groups form a local system. Along any path, horizontal transport gives a homotopy between the inclusion of the initial fiber and the transported inclusion, so the induced homology transport is compatible with the fiber inclusions. □
Multi-Taub–NUT Example
The multi-Taub–NUT spaces are the ALF-A k A_k A k gravitational instantons given by the Gibbons–Hawking ansatz. They are noncompact and hence outside the scope of Duistermaat–Heckman’s original proof, but their periods can be directly shown to vary affine-linearly with the relative positions of their k + 1 k+1 k + 1 centers. We recall their hyperkähler quotient construction (see [18 Section 3] for details), which is already finite dimensional, so no partial reduction is needed. We apply Definition 2.3 by proving completeness of a natural Ehresmann connection.
Fix k ≥ 0 k\ge0 k ≥ 0 and λ > 0 \lambda>0 λ > 0 , and let
E = H k + 1 × ( R 3 × S 1 ) , B = T k + 1 , E=\mathbb{H}^{k+1}\times(\mathbb{R}^3\times S^1),\qquad B=T^{k+1}, E = H k + 1 × ( R 3 × S 1 ) , B = T k + 1 ,
with quaternion coordinates q j q_j q j ; Cartesian coordinates y = ( y 1 , y 2 , y 3 ) \mathbf{y}=(y_1,y_2,y_3) y = ( y 1 , y 2 , y 3 ) , and τ ∈ R / 2 π Z \tau\in\mathbb{R}/2\pi\mathbb{Z} τ ∈ R /2 π Z . Equip E E E with the product flat hyperkähler metric g E g_E g E , using λ ∣ d y ∣ 2 + λ − 1 d τ 2 \lambda\lvert d\mathbf{y}\rvert^2+\lambda^{-1}d\tau^2 λ ∣ d y ∣ 2 + λ − 1 d τ 2 on R 3 × S 1 \mathbb{R}^3\times S^1 R 3 × S 1 . Use J 1 , J 2 , J 3 J_1, J_2, J_3 J 1 , J 2 , J 3 given by left multiplication by i , j , k \mathrm{i},\mathrm{j},\mathrm{k} i , j , k on the quaternion factors, and choose the structure on R 3 × S 1 \mathbb{R}^3 \times S^1 R 3 × S 1 so that ι ∂ τ ϖ s = d y s \iota_{\partial_\tau}\varpi_s = \mathrm{d}y_s ι ∂ τ ϖ s = d y s . The j j j th factor of B B B acts by ( q j , τ ) ↦ ( q j e − i θ , τ + θ ) (q_j,\tau) \mapsto(q_j e^{-\mathrm{i}\theta},\tau+\theta) ( q j , τ ) ↦ ( q j e − i θ , τ + θ ) , fixing the other coordinates. The hyperkähler moment map is given by the triple
π ~ = ( π ~ 1 , π ~ 2 , π ~ 3 ) : E ⟶ ( b ∗ ) ⊕ 3 , π ~ s = ( π ~ s ( 0 ) , … , π ~ s ( k ) ) , (27) \widetilde{\pi} = (\widetilde{\pi}_1,\widetilde{\pi}_2,\widetilde{\pi}_3) : E \longrightarrow(\mathfrak{b}^*)^{\oplus3}, \qquad\widetilde{\pi}_s = (\widetilde{\pi}_s^{(0)},\ldots,\widetilde{\pi}_s^{(k)}), \tag*{(27)} π = ( π 1 , π 2 , π 3 ) : E ⟶ ( b ∗ ) ⊕ 3 , π s = ( π s ( 0 ) , … , π s ( k ) ) , (27)
whose components satisfy
π ~ 1 ( j ) i + π ~ 2 ( j ) j + π ~ 3 ( j ) k = 1 2 q j i q j ‾ + y , \widetilde{\pi}_1^{(j)}\mathrm{i}+\widetilde{\pi}_2^{(j)}\mathrm{j}+\widetilde{\pi}_3^{(j)}\mathrm{k}=\frac{1}{2}q_j\mathrm{i}\overline{q_j}+\mathbf{y}, π 1 ( j ) i + π 2 ( j ) j + π 3 ( j ) k = 2 1 q j i q j + y ,
where we identify y = y 1 i + y 2 j + y 3 k ∈ H \mathbf{y}=y_1\mathrm{i}+y_2\mathrm{j}+y_3\mathrm{k}\in\mathbb{H} y = y 1 i + y 2 j + y 3 k ∈ H .
Write α = ( α 1 , α 2 , α 3 ) \boldsymbol{\alpha}=(\alpha_1,\alpha_2,\alpha_3) α = ( α 1 , α 2 , α 3 ) with α s = ( α s ( 0 ) , … , α s ( k ) ) \alpha_s=(\alpha_s^{(0)},\ldots,\alpha_s^{(k)}) α s = ( α s ( 0 ) , … , α s ( k ) ) , and set α ( j ) = ( α 1 ( j ) , α 2 ( j ) , α 3 ( j ) ) \boldsymbol{\alpha}^{(j)}=(\alpha_1^{(j)},\alpha_2^{(j)},\alpha_3^{(j)}) α ( j ) = ( α 1 ( j ) , α 2 ( j ) , α 3 ( j ) ) . Over
Λ 0 = { α ∈ ( b ∗ ) ⊕ 3 : α ( j ) ≠ α ( ℓ ) for j ≠ ℓ } , \Lambda_0=\left\{\boldsymbol{\alpha}\in(\mathfrak{b}^*)^{\oplus3}:\boldsymbol{\alpha}^{(j)}\ne\boldsymbol{\alpha}^{(\ell)}\text{ for }j\ne\ell\right\}, Λ 0 = { α ∈ ( b ∗ ) ⊕ 3 : α ( j ) = α ( ℓ ) for j = ℓ } ,
the action on E 0 = π ~ − 1 ( Λ 0 ) E_0=\widetilde{\pi}^{-1}(\Lambda_0) E 0 = π − 1 ( Λ 0 ) is free, giving a principal bundle q : E 0 → M 0 = E 0 / B q:E_0\to\mathcal{M}_0=E_0/B q : E 0 → M 0 = E 0 / B . The fibers of π : M 0 → Λ 0 \pi:\mathcal{M}_0\to\Lambda_0 π : M 0 → Λ 0 are the multi–Taub–NUT spaces, with Gibbons–Hawking centers α ( j ) ∈ R 3 \boldsymbol{\alpha}^{(j)}\in\mathbb{R}^3 α ( j ) ∈ R 3 , the points over which the circle fibers collapse.
Lifting vectors in Λ \Lambda Λ to vectors in E 0 E_0 E 0 requires a smooth choice of right inverses to d π ~ \mathrm{d}\widetilde{\pi} d π . At a point x ∈ E 0 x\in E_0 x ∈ E 0 , consider the operator D D † DD^\dagger D D † where D = d π ~ x : T x E → ( b ∗ ) ⊕ 3 D=\mathrm{d}\widetilde{\pi}_x:T_xE\to(\mathfrak{b}^*)^{\oplus3} D = d π x : T x E → ( b ∗ ) ⊕ 3 . If D D † DD^\dagger D D † is invertible, then D † ( D D † ) − 1 D^\dagger(DD^\dagger)^{-1} D † ( D D † ) − 1 gives a natural choice of right inverse. As x x x varies, this defines an Ehresmann connection. To bound the lifts and prove completeness, we must bound the eigenvalues of D D † DD^\dagger D D † below over compact sets in Λ \Lambda Λ . Let u j u_j u j generate the j j j th S 1 S^1 S 1 factor of B B B . The moment-map identity d π ~ s ( j ) = ι u j ϖ s = g E ( J s u j , − ) \mathrm{d}\widetilde{\pi}_s^{(j)}=\iota_{u_j}\varpi_s=g_E(J_su_j,-) d π s ( j ) = ι u j ϖ s = g E ( J s u j , − ) gives
( D D † ) ( s , j ) , ( r , ℓ ) = g E ( J s u j , J r u ℓ ) = δ s r A j ℓ , (DD^\dagger)_{(s,j),(r,\ell)}=g_E(J_su_j,J_ru_\ell)=\delta_{sr}A_{j\ell}, ( D D † ) ( s , j ) , ( r , ℓ ) = g E ( J s u j , J r u ℓ ) = δ sr A j ℓ ,
A j ℓ = g E ( u j , u ℓ ) = δ j ℓ ∣ q j ∣ 2 + λ − 1 , A_{j\ell}=g_E(u_j,u_\ell)=\delta_{j\ell}|q_j|^2+\lambda^{-1}, A j ℓ = g E ( u j , u ℓ ) = δ j ℓ ∣ q j ∣ 2 + λ − 1 ,
which shows D D † DD^\dagger D D † consists of three diagonal blocks, each equal to A = ( A j ℓ ) A=(A_{j\ell}) A = ( A j ℓ ) .
We now find a lower bound on the eigenvalues of A A A , uniform over an arbitrary compact K ⊂ Λ 0 K\subset\Lambda_0 K ⊂ Λ 0 . There is a constant δ > 0 \delta>0 δ > 0 such that ∣ α ( j ) − α ( ℓ ) ∣ ≥ δ |\boldsymbol{\alpha}^{(j)}-\boldsymbol{\alpha}^{(\ell)}|\ge\delta ∣ α ( j ) − α ( ℓ ) ∣ ≥ δ for j ≠ ℓ j\ne\ell j = ℓ throughout K K K . Since ∣ q j ∣ 2 = 2 ∣ α ( j ) − y ∣ |q_j|^2=2|\boldsymbol{\alpha}^{(j)}-\mathbf{y}| ∣ q j ∣ 2 = 2∣ α ( j ) − y ∣ , at most one ∣ q j ∣ 2 |q_j|^2 ∣ q j ∣ 2 can be smaller than δ \delta δ . Relabel the coordinates so that ∣ q 0 ∣ |q_0| ∣ q 0 ∣ is smallest. Then,
v T A v = ∑ j = 0 k ∣ q j ∣ 2 v j 2 + λ − 1 ( ∑ j = 0 k v j ) 2 ≥ δ ∑ j = 1 k v j 2 + λ − 1 ( ∑ j = 0 k v j ) 2 . (28) v^TAv=\sum_{j=0}^{k}|q_j|^2v_j^2+\lambda^{-1}\left(\sum_{j=0}^{k}v_j\right)^2\ge\delta\sum_{j=1}^{k}v_j^2+\lambda^{-1}\left(\sum_{j=0}^{k}v_j\right)^2. \tag*{(28)} v T A v = j = 0 ∑ k ∣ q j ∣ 2 v j 2 + λ − 1 ( j = 0 ∑ k v j ) 2 ≥ δ j = 1 ∑ k v j 2 + λ − 1 ( j = 0 ∑ k v j ) 2 . (28)
The right hand side is positive definite in v v v , so A ≥ c K Id A\ge c_K\operatorname{Id} A ≥ c K Id on π ~ − 1 ( K ) \widetilde{\pi}^{-1}(K) π − 1 ( K ) for some c K > 0 c_K>0 c K > 0 . With this estimate, Grönwall’s inequality shows that all paths in Λ \Lambda Λ have well-defined lifts, which proves completeness. By Definition 2.4 and Definition 2.5, Definition 2.3 gives affine period variation with respect to α \boldsymbol{\alpha} α .
Universal Geometric Structure
In this section we describe the geometric construction of the moduli family π : M → Λ \pi:\mathcal{M}\to\Lambda π : M → Λ of parabolic Higgs bundles and its natural principal torus bundle E \mathcal{E} E . We endow E \mathcal{E} E with an invariant triple of extended Hitchin forms, and recover the moduli spaces M ( α , σ ) \mathcal{M}(\boldsymbol{\alpha},\sigma) M ( α , σ ) together with their kä̈hler forms through Hamiltonian reduction. We then show that our construction satisfies the geometric hypothesis of Definition 2.3.
As in the introduction, we fix a compact connected Riemann surface C C C , a divisor of distinct marked points D = { p 1 , … , p n } D = \{p_1,\ldots,p_n\} D = { p 1 , … , p n } , and a rank-r ⋆ r_{\star} r ⋆ bundle E E E over C C C , together with a fixed identification det E ≅ O C \det E \cong\mathcal{O}_C det E ≅ O C . We denote C ∘ = C ∖ D C^\circ= C \setminus D C ∘ = C ∖ D and E ∘ E^\circ E ∘ the restriction to C ∘ C^\circ C ∘ . We also fix parabolic filtrations over each point in D D D , and a background metric h 0 h_0 h 0 on E E E .
The Gauge Group
To describe the boundary gauge transformations, consider the associated graded space of the parabolic flag
gr F i = ⨁ k = 0 ℓ i gr F i ( k ) . \operatorname{gr} F_i = \bigoplus_{k=0}^{\ell_i} \operatorname{gr} F_i^{(k)}. gr F i = k = 0 ⨁ ℓ i gr F i ( k ) .
where gr F i ( k ) = F i ( k ) / F i ( k − 1 ) \operatorname{gr} F_i^{(k)} = F_i^{(k)}/F_i^{(k-1)} gr F i ( k ) = F i ( k ) / F i ( k − 1 ) . Recall r i ( k ) = dim F i ( k ) r_i^{(k)} = \dim F_i^{(k)} r i ( k ) = dim F i ( k ) and m i ( k ) = dim gr F i ( k ) m_i^{(k)} = \dim\operatorname{gr} F_i^{(k)} m i ( k ) = dim gr F i ( k ) . Consider the complex Levi group
L i = { g ∈ ∏ k = 0 ℓ i GL ( gr F i ( k ) ) ∣ det ( g ) = 1 } . L_i = \left\{g \in\prod_{k=0}^{\ell_i} \operatorname{GL}(\operatorname{gr} F_i^{(k)}) \mathrel{\Big|} \det(g)=1\right\}. L i = { g ∈ k = 0 ∏ ℓ i GL ( gr F i ( k ) ) det ( g ) = 1 } .
and the real compact Levi group
K i = { g ∈ ∏ k = 0 ℓ i U ( gr F i ( k ) ) ∣ det ( g ) = 1 } . K_i = \left\{g \in\prod_{k=0}^{\ell_i} \operatorname{U}(\operatorname{gr} F_i^{(k)}) \mathrel{\Big|} \det(g)=1\right\}. K i = { g ∈ k = 0 ∏ ℓ i U ( gr F i ( k ) ) det ( g ) = 1 } .
We denote the respective Lie algebras by l i \mathfrak l_i l i and k i \mathfrak k_i k i . Note the latter has center
z ( k i ) = { ⨁ k = 0 ℓ i a i ( k ) Id gr F i ( k ) : a i ( k ) ∈ i R , ∑ k = 0 ℓ i m i ( k ) a i ( k ) = 0 } . (29) \mathfrak z(\mathfrak k_i)=\left\{\bigoplus_{k=0}^{\ell_i}a_i^{(k)}\operatorname{Id}_{\operatorname{gr} F_i^{(k)}}:a_i^{(k)}\in i\mathbb{R},\ \sum_{k=0}^{\ell_i}m_i^{(k)}a_i^{(k)}=0\right\}. \tag*{(29)} z ( k i ) = { k = 0 ⨁ ℓ i a i ( k ) Id gr F i ( k ) : a i ( k ) ∈ i R , k = 0 ∑ ℓ i m i ( k ) a i ( k ) = 0 } . (29)
Taking block determinants gives a homomorphism from K i K_i K i onto the i i i th boundary torus
T i = { ( u 0 , … , u ℓ i ) ∈ U ( 1 ) ℓ i + 1 ∣ ∏ k = 0 ℓ i u k = 1 } , T_i=\left\{(u_0,\ldots,u_{\ell_i})\in\operatorname{U}(1)^{\ell_i+1}\mathrel{\Big|}\prod_{k=0}^{\ell_i}u_k=1\right\}, T i = { ( u 0 , … , u ℓ i ) ∈ U ( 1 ) ℓ i + 1 k = 0 ∏ ℓ i u k = 1 } ,
and taking traces identifies z ( k i ) \mathfrak z(\mathfrak k_i) z ( k i ) with the Lie algebra of T i T_i T i .
We now specify the gauge transformations allowed at a puncture. Full details will be given in § 4 . Choose once and for all orthonormal trivializations
E ∣ B 1 ( p i ) ≅ B 1 ( p i ) × gr F i , E|_{B_1(p_i)}\cong B_1(p_i)\times\operatorname{gr} F_i, E ∣ B 1 ( p i ) ≅ B 1 ( p i ) × gr F i ,
which we call the analytic model frames , since they are needed to define our function spaces. Fix an integer k ≥ 2 k\geq2 k ≥ 2 and a sufficiently small positive weight δ \delta δ . See our weighted Sobolev space conventions in § 1.2 . We use the h 0 h_0 h 0 -unitary gauge group
G = { g ∈ H l o c k + 1 ( C ∘ , SU ( E ) ) ∣ g − g i ∈ H δ k + 1 on B 1 ∘ ( p i ) for some g i ∈ K i } . (30) G=\left\{g\in H^{k+1}_{\mathrm{loc}}(C^\circ,\operatorname{SU}(E))\mathrel{\Big|}g-g_i\in H^{k+1}_{\delta}\text{ on }B_1^\circ(p_i)\text{ for some }g_i\in K_i\right\}. \tag*{(30)} G = { g ∈ H loc k + 1 ( C ∘ , SU ( E )) g − g i ∈ H δ k + 1 on B 1 ∘ ( p i ) for some g i ∈ K i } . (30)
The Lie algebra g \mathfrak g g consists of sections ξ \xi ξ with limits ξ i = lim z → p i ξ ( z ) ∈ k i \xi_i=\lim_{z\to p_i}\xi(z)\in\mathfrak k_i ξ i = lim z → p i ξ ( z ) ∈ k i and ξ − ξ i ∈ H δ k + 1 \xi-\xi_i\in H^{k+1}_{\delta} ξ − ξ i ∈ H δ k + 1 on each punctured disk B 1 ∘ ( p i ) B_1^\circ(p_i) B 1 ∘ ( p i ) , so elements can be extended to C C C . For a i ∈ z ( t i ) a_i \in\mathfrak{z}(\mathfrak{t}_i) a i ∈ z ( t i ) , the point distribution a i δ p i a_i\delta_{p_i} a i δ p i given by
⟨ a i δ p i , ξ ⟩ = ⟨ a i , ξ i ⟩ h 0 \langle a_i\delta_{p_i},\xi\rangle=\langle a_i,\xi_i\rangle_{h_0} ⟨ a i δ p i , ξ ⟩ = ⟨ a i , ξ i ⟩ h 0
defines an element of g ∗ \mathfrak{g}^* g ∗ . We use this to identify ⨁ i z ( t i ) ⊂ g ∗ \bigoplus_i\mathfrak{z}(\mathfrak{t}_i)\subset\mathfrak{g}^* ⨁ i z ( t i ) ⊂ g ∗ .
For g ∈ G g\in G g ∈ G , let g i ( k ) g_i^{(k)} g i ( k ) denote the induced actions of g i g_i g i on the associated graded gr F i ( k ) \operatorname{gr}F_i^{(k)} gr F i ( k ) . Let C r ⋆ ⊂ G C_{r_\star}\subset G C r ⋆ ⊂ G be the subgroup of r ⋆ r_\star r ⋆ th roots of unity (as scalar endomorphisms). Then
H = C r ⋆ ⋅ { g ∈ G | det g i ( k ) = 1 for every i , k } (31) H=C_{r_\star}\cdot\left\{g\in G\ \middle|\ \det g_i^{(k)}=1\text{ for every }i,k\right\} \tag*{(31)} H = C r ⋆ ⋅ { g ∈ G det g i ( k ) = 1 for every i , k } (31)
is a subgroup of G G G , and the quotient B : = G / H B:=G/H B := G / H is the effective boundary torus ( T 1 × ⋯ × T n ) / C r ⋆ (T_1\times\cdots\times T_n)/C_{r_\star} ( T 1 × ⋯ × T n ) / C r ⋆ . The Lie algebra h \mathfrak{h} h of H H H satisfies Ann ( h ) = ⨁ i z ( t i ) \operatorname{Ann}(\mathfrak{h})=\bigoplus_i\mathfrak{z}(\mathfrak{t}_i) Ann ( h ) = ⨁ i z ( t i ) under the identification ⨁ i z ( t i ) ⊂ g ∗ \bigoplus_i\mathfrak{z}(\mathfrak{t}_i)\subset\mathfrak{g}^* ⨁ i z ( t i ) ⊂ g ∗ above. The short exact sequence
0 ⟶ b ∗ ⟶ g ∗ ⟶ h ∗ ⟶ 0 0\longrightarrow\mathfrak{b}^*\longrightarrow\mathfrak{g}^*\longrightarrow\mathfrak{h}^*\longrightarrow0 0 ⟶ b ∗ ⟶ g ∗ ⟶ h ∗ ⟶ 0
therefore shows ⨁ i z ( t i ) = b ∗ ⊂ g ∗ \bigoplus_i\mathfrak{z}(\mathfrak{t}_i)=\mathfrak{b}^*\subset\mathfrak{g}^* ⨁ i z ( t i ) = b ∗ ⊂ g ∗ .
Distribution-Valued Moment Maps
We define the extended configuration space of Higgs bundles, where the parabolic weights and Levi-central residues are allowed to vary within an admissible parameter locus to be specified later. Write α i \alpha_i α i and σ i \sigma_i σ i for the block scalar matrices holding the weights and residues, respectively, so that i α i ∈ z ( t i ) i\alpha_i\in\mathfrak{z}(\mathfrak{t}_i) i α i ∈ z ( t i ) and σ i ∈ z ( l i ) \sigma_i\in\mathfrak{z}(\mathfrak{l}_i) σ i ∈ z ( l i ) . Let C \mathcal{C} C consist of H l o c k H^k_{\mathrm{loc}} H loc k pairs ( A , Φ ) (A,\Phi) ( A , Φ ) with fixed determinant, which can locally be written as
A = i α i d θ + H δ k , Φ = − 2 i ( Re σ i d θ + Im σ i d r r ) + H δ k (32) A=i\alpha_i\,d\theta+\mathcal{H}^k_\delta,\qquad
\Phi=-2i\left(\operatorname{Re}\sigma_i\,d\theta+\operatorname{Im}\sigma_i\,\frac{dr}{r}\right)+\mathcal{H}^k_\delta \tag*{(32)} A = i α i d θ + H δ k , Φ = − 2 i ( Re σ i d θ + Im σ i r d r ) + H δ k (32)
for some admissible α , σ \alpha,\sigma α , σ . Our Sobolev space convention for 1-forms means each remainder has the form a d r / r + b d θ a\,dr/r+b\,d\theta a d r / r + b d θ for functions a , b ∈ H δ k ( B ϵ o ( p i ) ) a,b\in H^k_\delta(B_\epsilon^o(p_i)) a , b ∈ H δ k ( B ϵ o ( p i )) .
We shall interpret Hitchin’s equations in terms of distributions on C C C , so that they can be packaged together with weight and residue specifications in a moment map. For a locally integrable 1-form β \beta β on U ⊂ C U\subset C U ⊂ C , its distributional derivative d β d\beta d β pairs with a smooth compactly supported test section ξ ∈ C c ∞ ( U , s u ( E , h 0 ) ) \xi\in C^\infty_c(U,\mathfrak{su}(E,h_0)) ξ ∈ C c ∞ ( U , su ( E , h 0 )) by
⟨ d β , ξ ⟩ = ∫ U ⟨ β ∧ d ξ ⟩ h 0 . \langle d\beta,\xi\rangle=\int_U\langle\beta\wedge d\xi\rangle_{h_0}. ⟨ d β , ξ ⟩ = ∫ U ⟨ β ∧ d ξ ⟩ h 0 .
This agrees with the usual derivative away from D D D , but also retains contributions at the punctures.
Definition 3.1. The distributional moment map is the triple μ = ( μ 1 , μ 2 , μ 3 ) : C → ( g ∗ ) 3 \mu=(\mu_1,\mu_2,\mu_3):\mathcal{C}\to(\mathfrak{g}^*)^3 μ = ( μ 1 , μ 2 , μ 3 ) : C → ( g ∗ ) 3 defined by
μ 1 = F A − 1 2 [ Φ ∧ Φ ] , μ 2 = − d A Φ , μ 3 = − d A ( ⋆ Φ ) , (33) \mu_1=F_A-\frac{1}{2}[\Phi\wedge\Phi],\qquad
\mu_2=-d_A\Phi,\qquad
\mu_3=-d_A(\star\Phi), \tag*{(33)} μ 1 = F A − 2 1 [ Φ ∧ Φ ] , μ 2 = − d A Φ , μ 3 = − d A ( ⋆ Φ ) , (33)
interpreted as distributional 2-forms on C C C . The boundary moment map μ b d r : C → ( b ∗ ) 3 \mu^{\mathrm{bdr}}:\mathcal{C}\to(\mathfrak{b}^*)^3 μ bdr : C → ( b ∗ ) 3 is the part of μ \mu μ supported on D D D . The remainder μ i n t = μ − μ b d r \mu^{\mathrm{int}}=\mu-\mu^{\mathrm{bdr}} μ int = μ − μ bdr is the interior moment map .
The interior part μ i n t \mu^{\mathrm{int}} μ int maps into H δ k − 1 ( C ∘ , Ω C 2 ⊗ s u ( E , h 0 ) ) H^{k-1}_\delta(C^\circ,\Omega_C^2\otimes\mathfrak{su}(E,h_0)) H δ k − 1 ( C ∘ , Ω C 2 ⊗ su ( E , h 0 )) , whose elements give bounded functionals on g \mathfrak{g} g by the weighted Cauchy–Schwarz inequality (9) . The name “moment map” is justified below. First, we describe μ b d r \mu^{\mathrm{bdr}} μ bdr explicitly.
Lemma 3.2. For ( A , Φ ) ∈ C (A,\Phi)\in\mathcal{C} ( A , Φ ) ∈ C with weights α \alpha α and residues σ \sigma σ , the boundary moment map is
μ b d r ( A , Φ ) = ( 2 π i ∑ i α i δ p i , 4 π i ∑ i Re σ i δ p i , 4 π i ∑ i Im σ i δ p i ) . (34) \mu^{\mathrm{bdr}}(A,\Phi)=\left(2\pi i\sum_i\alpha_i\delta_{p_i},4\pi i\sum_i\operatorname{Re}\sigma_i\delta_{p_i},4\pi i\sum_i\operatorname{Im}\sigma_i\delta_{p_i}\right). \tag*{(34)} μ bdr ( A , Φ ) = ( 2 π i i ∑ α i δ p i , 4 π i i ∑ Re σ i δ p i , 4 π i i ∑ Im σ i δ p i ) . (34)
Proof. Under distributional differentiation, we have
d ( d θ ) = 2 π δ p i , d ( d r r ) = 0. (35) \mathrm{d}(\mathrm{d}\theta)=2\pi\delta_{p_i},\qquad\mathrm{d}\left(\frac{\mathrm{d}r}{r}\right)=0. \tag*{(35)} d ( d θ ) = 2 π δ p i , d ( r d r ) = 0. (35)
Applying these identities to the local models in (32) gives the part of μ \mu μ supported on D D D . The derivative terms in μ 1 , μ 2 , μ 3 \mu_1,\mu_2,\mu_3 μ 1 , μ 2 , μ 3 are respectively d A \mathrm{d}A d A , − d Φ -\mathrm{d}\Phi − d Φ , and − d ( ⋆ Φ ) -\mathrm{d}(\star\Phi) − d ( ⋆ Φ ) . Using ⋆ ( d r / r ) = d θ \star(\mathrm{d}r/r)=\mathrm{d}\theta ⋆ ( d r / r ) = d θ and ⋆ d θ = − d r / r \star\mathrm{d}\theta=-\mathrm{d}r/r ⋆ d θ = − d r / r , the leading-order d θ \mathrm{d}\theta d θ coefficients of A A A , − Φ -\Phi − Φ , and − ⋆ Φ -\star\Phi − ⋆ Φ are respectively i α i i\alpha_i i α i , 2 i Re σ i 2i\operatorname{Re}\sigma_i 2 i Re σ i , and 2 i Im σ i 2i\operatorname{Im}\sigma_i 2 i Im σ i . The nonlinear terms [ A ∧ A ] [A\wedge A] [ A ∧ A ] , [ Φ ∧ Φ ] [\Phi\wedge\Phi] [ Φ ∧ Φ ] , [ A ∧ ⋆ Φ ] [A\wedge\star\Phi] [ A ∧ ⋆ Φ ] , and [ A ∧ Φ ] [A\wedge\Phi] [ A ∧ Φ ] in Hitchin’s equations have size O ( r δ − 2 ) O(r^{\delta-2}) O ( r δ − 2 ) and are therefore integrable,1 with no point-mass contribution. □ \square □
In light of the lemma, we define ν = ( ν 1 , ν 2 , ν 3 ) \nu=(\nu_1,\nu_2,\nu_3) ν = ( ν 1 , ν 2 , ν 3 ) where
ν 1 , i = 2 π i α i , ν 2 , i = 4 π i Re σ i , ν 3 , i = 4 π i Im σ i , ν s = ∑ i ν s , i δ p i . (36) \nu_{1,i}=2\pi i\alpha_i,\qquad\nu_{2,i}=4\pi i\operatorname{Re}\sigma_i,\qquad\nu_{3,i}=4\pi i\operatorname{Im}\sigma_i,\qquad\nu_s=\sum_i\nu_{s,i}\delta_{p_i}. \tag*{(36)} ν 1 , i = 2 π i α i , ν 2 , i = 4 π i Re σ i , ν 3 , i = 4 π i Im σ i , ν s = i ∑ ν s , i δ p i . (36)
Definition 3.3. We call the distributions ν s ∈ b ∗ ⊂ g ∗ \nu_s\in\mathfrak{b}^*\subset\mathfrak{g}^* ν s ∈ b ∗ ⊂ g ∗ the boundary parameters or simply parameters . The parameter space is the subset Λ ⊂ ( b ∗ ) 3 \Lambda\subset(\mathfrak{b}^*)^3 Λ ⊂ ( b ∗ ) 3 of those ν \nu ν whose parabolic weights remain strictly ordered.
We choose C \mathcal{C} C to exclude parabolic weight collisions, so that the boundary moment map is μ b d r : C → Λ \mu^{\mathrm{bdr}}:\mathcal{C}\to\Lambda μ bdr : C → Λ . Note this is G G G -invariant by the Levi-central condition, and hence so is μ \mu μ .
Definition 3.4 (Extended parabolic moduli space). Let Z = { ( A , Φ ) ∈ C : μ i n t ( A , Φ ) = 0 } \mathcal{Z}=\{(A,\Phi)\in\mathcal{C}:\mu^{\mathrm{int}}(A,\Phi)=0\} Z = {( A , Φ ) ∈ C : μ int ( A , Φ ) = 0 } be the solution space of Hitchin’s equations on C \mathcal{C} C . The extended parabolic moduli space is M = Z / G \mathcal{M}=\mathcal{Z}/G M = Z / G . Using the subgroup H ⊂ G H\subset G H ⊂ G from (31) , define the partial quotient E = Z / H \mathcal{E}=\mathcal{Z}/H E = Z / H , which comes equipped with a quotient map q : E → M q:\mathcal{E}\to\mathcal{M} q : E → M and a residual action of the effective boundary torus B B B .
On Z \mathcal{Z} Z , the total moment map equals its boundary part and induces maps π : M → Λ \pi:\mathcal{M}\to\Lambda π : M → Λ and π ~ : E → Λ \widetilde{\pi}:\mathcal{E}\to\Lambda π : E → Λ . In § 4 we will prove that, on the irreducible loci, the quotients are smooth, q q q is a principal B B B -bundle, and π \pi π is a submersion.
This fits the setup for Definition 2.3. The next subsection establishes the geometric hypothesis.
Closed Forms on the Partial Quotient
We now extend the usual Kähler forms on the fixed-parameter moduli spaces to closed forms on the partial quotient E = Z / H \mathcal{E}=\mathcal{Z}/H E = Z / H , and to prove the Hamiltonian identity (17) . We in fact construct the extended 2-forms on the full configuration space C \mathcal{C} C , and show they descend to the desired forms.
Figure 1. The partial quotient and the extended parabolic moduli space.
If one evaluates the usual L 2 L^2 L 2 pairings on Higgs bundle deformations with varying boundary data, the integral diverges. We address this by subtracting off a logarithmic term coming from the boundary pairing to get an integrable expression. On tri-symplectic leaves, the resulting forms agree with the usual L 2 L^2 L 2 pairings. Definition 2.3 does not require nondegeneracy, and indeed we suspect that our forms degenerate on certain deformations varying boundary data.
Set C ε = C ∖ ⋃ i = 1 n B ε ( p i ) C_\varepsilon= C \setminus\bigcup_{i=1}^{n} B_\varepsilon(p_i) C ε = C ∖ ⋃ i = 1 n B ε ( p i ) for ε > 0 \varepsilon> 0 ε > 0 , and let u = ( A ˙ , Φ ˙ ) u=(\dot A,\dot\Phi) u = ( A ˙ , Φ ˙ ) and w = ( A ˙ ′ , Φ ˙ ′ ) w=(\dot A',\dot\Phi') w = ( A ˙ ′ , Φ ˙ ′ ) in T ( A , Φ ) C T_{(A,\Phi)}\mathcal{C} T ( A , Φ ) C be Higgs bundle deformations. Define the truncated Hitchin forms by
ω 1 ε ( u , w ) = ∫ C ε ⟨ A ˙ ∧ A ˙ ′ ⟩ h 0 − ⟨ Φ ˙ ∧ Φ ˙ ′ ⟩ h 0 , \omega_1^\varepsilon(u,w)=\int_{C_\varepsilon}\langle\dot A\wedge\dot A'\rangle_{h_0}-\langle\dot\Phi\wedge\dot\Phi'\rangle_{h_0}, ω 1 ε ( u , w ) = ∫ C ε ⟨ A ˙ ∧ A ˙ ′ ⟩ h 0 − ⟨ Φ ˙ ∧ Φ ˙ ′ ⟩ h 0 ,
ω 2 ε ( u , w ) = ∫ C ε ⟨ A ˙ ′ ∧ Φ ˙ ⟩ h 0 − ⟨ A ˙ ∧ Φ ˙ ′ ⟩ h 0 , \omega_2^\varepsilon(u,w)=\int_{C_\varepsilon}\langle\dot A'\wedge\dot\Phi\rangle_{h_0}-\langle\dot A\wedge\dot\Phi'\rangle_{h_0}, ω 2 ε ( u , w ) = ∫ C ε ⟨ A ˙ ′ ∧ Φ ˙ ⟩ h 0 − ⟨ A ˙ ∧ Φ ˙ ′ ⟩ h 0 ,
ω 3 ε ( u , w ) = ∫ C ε ⟨ ⋆ A ˙ ∧ Φ ˙ ′ ⟩ h 0 − ⟨ ⋆ A ˙ ′ ∧ Φ ˙ ⟩ h 0 . (37) \omega_3^\varepsilon(u,w)=\int_{C_\varepsilon}\langle\star\dot A\wedge\dot\Phi'\rangle_{h_0}-\langle\star\dot A'\wedge\dot\Phi\rangle_{h_0}. \tag*{(37)} ω 3 ε ( u , w ) = ∫ C ε ⟨ ⋆ A ˙ ∧ Φ ˙ ′ ⟩ h 0 − ⟨ ⋆ A ˙ ′ ∧ Φ ˙ ⟩ h 0 . (37)
As it stands, the forms ω s ε \omega_s^\varepsilon ω s ε may diverge as ε → 0 \varepsilon\to0 ε → 0 , so we must isolate the divergent parts.
Lemma 3.5. Let ν ˙ s , i ( u ) = d μ s , i b d r ( u ) \dot\nu_{s,i}(u)=d\mu_{s,i}^{\mathrm{bdr}}(u) ν ˙ s , i ( u ) = d μ s , i bdr ( u ) and ν ˙ s , i ( w ) = d μ s , i b d r ( w ) \dot\nu_{s,i}(w)=d\mu_{s,i}^{\mathrm{bdr}}(w) ν ˙ s , i ( w ) = d μ s , i bdr ( w ) be the parameter variation along u u u and w w w . For ( s , j , k ) (s,j,k) ( s , j , k ) a cycle of ( 1 , 2 , 3 ) (1,2,3) ( 1 , 2 , 3 ) , near p i p_i p i the integrand defining ω s ε ( u , w ) \omega_s^\varepsilon(u,w) ω s ε ( u , w ) is
1 4 π 2 ( ⟨ ν ˙ j , i ( u ) , ν ˙ k , i ( w ) ⟩ h 0 − ⟨ ν ˙ j , i ( w ) , ν ˙ k , i ( u ) ⟩ h 0 ) r − 1 d r ∧ d θ + H δ k . (38) \frac{1}{4\pi^2}\left(\langle\dot\nu_{j,i}(u),\dot\nu_{k,i}(w)\rangle_{h_0}-\langle\dot\nu_{j,i}(w),\dot\nu_{k,i}(u)\rangle_{h_0}\right)r^{-1}dr\wedge d\theta+H_\delta^k. \tag*{(38)} 4 π 2 1 ( ⟨ ν ˙ j , i ( u ) , ν ˙ k , i ( w ) ⟩ h 0 − ⟨ ν ˙ j , i ( w ) , ν ˙ k , i ( u ) ⟩ h 0 ) r − 1 d r ∧ d θ + H δ k . (38)
Proof. Near p i p_i p i we write
A ˙ = A ˙ i , r d r + A ˙ i , θ d θ , Φ ˙ = Φ ˙ i , r d r + Φ ˙ i , θ d θ . \dot A=\dot A_{i,r}dr+\dot A_{i,\theta}d\theta,\qquad\dot\Phi=\dot\Phi_{i,r}dr+\dot\Phi_{i,\theta}d\theta. A ˙ = A ˙ i , r d r + A ˙ i , θ d θ , Φ ˙ = Φ ˙ i , r d r + Φ ˙ i , θ d θ .
To highlight the hyperkähler symmetry between connection and Higgs field, we set
( u 0 , i , u 1 , i , u 2 , i , u 3 , i ) = ( − r A ˙ i , r , A ˙ i , θ , − Φ ˙ i , θ , − r Φ ˙ i , r ) (u_{0,i},u_{1,i},u_{2,i},u_{3,i})=(-r\dot A_{i,r},\dot A_{i,\theta},-\dot\Phi_{i,\theta},-r\dot\Phi_{i,r}) ( u 0 , i , u 1 , i , u 2 , i , u 3 , i ) = ( − r A ˙ i , r , A ˙ i , θ , − Φ ˙ i , θ , − r Φ ˙ i , r )
and similarly for w w w . Then for ( s , j , k ) (s,j,k) ( s , j , k ) a cycle of ( 1 , 2 , 3 ) (1,2,3) ( 1 , 2 , 3 ) , the integrand of ω s ε \omega_s^\varepsilon ω s ε near p i p_i p i is
( ⟨ u s , i , w 0 , i ⟩ h 0 − ⟨ w s , i , u 0 , i ⟩ h 0 + ⟨ u j , i , w k , i ⟩ h 0 − ⟨ w j , i , u k , i ⟩ h 0 ) r − 1 d r ∧ d θ , (39) \left(\langle u_{s,i},w_{0,i}\rangle_{h_0}-\langle w_{s,i},u_{0,i}\rangle_{h_0}+\langle u_{j,i},w_{k,i}\rangle_{h_0}-\langle w_{j,i},u_{k,i}\rangle_{h_0}\right)r^{-1}dr\wedge d\theta, \tag*{(39)} ( ⟨ u s , i , w 0 , i ⟩ h 0 − ⟨ w s , i , u 0 , i ⟩ h 0 + ⟨ u j , i , w k , i ⟩ h 0 − ⟨ w j , i , u k , i ⟩ h 0 ) r − 1 d r ∧ d θ , (39)
By definition of C \mathcal{C} C , the remainders u s , i − ν ˙ s , i ( u ) 2 π u_{s,i}-\frac{\dot\nu_{s,i}(u)}{2\pi} u s , i − 2 π ν ˙ s , i ( u ) and w s , i − ν ˙ s , i ( w ) 2 π w_{s,i}-\frac{\dot\nu_{s,i}(w)}{2\pi} w s , i − 2 π ν ˙ s , i ( w ) belong to H δ k \mathcal{H}_\delta^k H δ k for s = 1 , 2 , 3 s=1,2,3 s = 1 , 2 , 3 , and u 0 , i , w 0 , i ∈ H δ k u_{0,i},w_{0,i}\in\mathcal{H}_\delta^k u 0 , i , w 0 , i ∈ H δ k as well. Thus the expression in (38) indeed captures the leading order of the integrand. Multiplying these remainders against the constant matrices ν ˙ s , i ( u ) \dot\nu_{s,i}(u) ν ˙ s , i ( u ) , ν ˙ s , i ( w ) \dot\nu_{s,i}(w) ν ˙ s , i ( w ) preserves their regularity, and the Sobolev multiplication estimate (8) controls the products of the remainders with each other.
Integrating (38) over the annuli ε ≤ r ≤ 1 \varepsilon\le r \le1 ε ≤ r ≤ 1 gives − log ( ε ) b s ( u , w ) -\log(\varepsilon)b_s(u,w) − log ( ε ) b s ( u , w ) , where
b s ( u , w ) = 1 2 π ∑ i ( ⟨ ν ˙ j , i ( u ) , ν ˙ k , i ( w ) ⟩ h 0 − ⟨ ν ˙ j , i ( w ) , ν ˙ k , i ( u ) ⟩ h 0 ) . (40) b_s(u,w)=\frac{1}{2\pi}\sum_i\left(\langle\dot{\nu}_{j,i}(u),\dot{\nu}_{k,i}(w)\rangle_{h_0}-\langle\dot{\nu}_{j,i}(w),\dot{\nu}_{k,i}(u)\rangle_{h_0}\right). \tag*{(40)} b s ( u , w ) = 2 π 1 i ∑ ( ⟨ ν ˙ j , i ( u ) , ν ˙ k , i ( w ) ⟩ h 0 − ⟨ ν ˙ j , i ( w ) , ν ˙ k , i ( u ) ⟩ h 0 ) . (40)
Subtracting this term leaves an integrand belonging to H δ k \mathcal{H}_\delta^k H δ k .
Definition 3.6. For s = 1 , 2 , 3 s=1,2,3 s = 1 , 2 , 3 , the extended Hitchin form is
ω ~ s = lim ε → 0 ( ω s ε + log ( ε ) b s ) . (41) \widetilde{\omega}_s=\lim_{\varepsilon\to0}\left(\omega_s^\varepsilon+\log(\varepsilon)b_s\right). \tag*{(41)} ω s = ε → 0 lim ( ω s ε + log ( ε ) b s ) . (41)
By weighted Cauchy–Schwarz (9) , we have
∣ [ ( ω s ε + log ( ε ) b s ) − ( ω s ε ′ + log ( ε ′ ) b s ) ] ( u , w ) ∣ ≤ C δ ε δ (42) \left|\left[(\omega_s^\varepsilon+\log(\varepsilon)b_s)-(\omega_s^{\varepsilon'}+\log(\varepsilon')b_s)\right](u,w)\right|\le C\delta\varepsilon^\delta \tag*{(42)} [ ( ω s ε + log ( ε ) b s ) − ( ω s ε ′ + log ( ε ′ ) b s ) ] ( u , w ) ≤ C δ ε δ (42)
for C C C uniform in ( A , Φ ) (A,\Phi) ( A , Φ ) , 0 < ε ′ ≤ ε 0<\varepsilon'\le\varepsilon 0 < ε ′ ≤ ε , and u , w u,w u , w in bounded subsets of the tangent space. This proves convergence. The forms ω s ε \omega_s^\varepsilon ω s ε and b s b_s b s are translation-invariant, so their limit ω ~ s \widetilde{\omega}_s ω s is smooth and closed. The resulting forms do not depend on the analytic frames, but they do depend on the local coordinates ( r , θ ) (r,\theta) ( r , θ ) near the punctures.
It remains to check that these forms descend to the quotient Z / H \mathcal{Z}/H Z / H and that the residual torus action is Hamiltonian. We do this by showing that the distributional expressions (33) are moment maps for the extended Hitchin forms.
For ξ ∈ g \xi\in\mathfrak{g} ξ ∈ g , the infinitesimal action on ( A , Φ ) (A,\Phi) ( A , Φ ) is ζ ξ = ( − d A ξ , [ ξ , Φ ] ) \zeta_\xi=(-d_A\xi,[\xi,\Phi]) ζ ξ = ( − d A ξ , [ ξ , Φ ]) . We can now state the Hamiltonian identity:
Lemma 3.7. On the configuration space C \mathcal{C} C , the extended moment maps satisfy
ι ζ ξ ω ~ s = d ⟨ μ s , ξ ⟩ . (43) \iota_{\zeta_\xi}\widetilde{\omega}_s=d\langle\mu_s,\xi\rangle. \tag*{(43)} ι ζ ξ ω s = d ⟨ μ s , ξ ⟩ . (43)
Proof. It suffices to prove the identity for the limit of the truncated forms ω s ε \omega_s^\varepsilon ω s ε , as ζ ξ = ( − d A ξ , [ ξ , Φ ] ) \zeta_\xi=(-d_A\xi,[\xi,\Phi]) ζ ξ = ( − d A ξ , [ ξ , Φ ]) has no boundary variation and therefore b s ( ζ ξ , w ) b_s(\zeta_\xi,w) b s ( ζ ξ , w ) from (40) vanishes. Let w = ( A ˙ , Φ ˙ ) ∈ T ( A , Φ ) C w=(\dot{A},\dot{\Phi})\in T_{(A,\Phi)}\mathcal{C} w = ( A ˙ , Φ ˙ ) ∈ T ( A , Φ ) C . The linearizations on C ∘ \mathcal{C}^{\circ} C ∘ are
D μ 1 ( w ) = d A A ˙ − [ Φ ∧ Φ ˙ ] , D\mu_1(w)=d_A\dot{A}-[\Phi\wedge\dot{\Phi}], D μ 1 ( w ) = d A A ˙ − [ Φ ∧ Φ ˙ ] ,
D μ 2 ( w ) = − d A Φ ˙ − [ A ˙ ∧ Φ ] , D\mu_2(w)=-d_A\dot{\Phi}-[\dot{A}\wedge\Phi], D μ 2 ( w ) = − d A Φ ˙ − [ A ˙ ∧ Φ ] ,
D μ 3 ( w ) = − d A ( ⋆ Φ ˙ ) − [ A ˙ ∧ ⋆ Φ ] . (44) D\mu_3(w)=-d_A(\star\dot{\Phi})-[\dot{A}\wedge\star\Phi]. \tag*{(44)} D μ 3 ( w ) = − d A ( ⋆ Φ ˙ ) − [ A ˙ ∧ ⋆ Φ ] . (44)
Substituting ζ ξ \zeta_\xi ζ ξ into (37) and integrating by parts gives
ω 1 ε ( ζ ξ , w ) = ∫ C ε ⟨ ξ , D μ 1 ( w ) ∣ C ∘ ⟩ h 0 − ∫ ∂ C ε ⟨ ξ , A ˙ ⟩ h 0 , \omega_1^\varepsilon(\zeta_\xi,w)=\int_{\mathcal{C}_\varepsilon}\langle\xi,D\mu_1(w)|_{\mathcal{C}^{\circ}}\rangle_{h_0}-\int_{\partial\mathcal{C}_\varepsilon}\langle\xi,\dot{A}\rangle_{h_0}, ω 1 ε ( ζ ξ , w ) = ∫ C ε ⟨ ξ , D μ 1 ( w ) ∣ C ∘ ⟩ h 0 − ∫ ∂ C ε ⟨ ξ , A ˙ ⟩ h 0 ,
ω 2 ε ( ζ ξ , w ) = ∫ C ε ⟨ ξ , D μ 2 ( w ) ∣ C ∘ ⟩ h 0 + ∫ ∂ C ε ⟨ ξ , Φ ˙ ⟩ h 0 , \omega_2^\varepsilon(\zeta_\xi,w)=\int_{\mathcal{C}_\varepsilon}\langle\xi,D\mu_2(w)|_{\mathcal{C}^{\circ}}\rangle_{h_0}+\int_{\partial\mathcal{C}_\varepsilon}\langle\xi,\dot{\Phi}\rangle_{h_0}, ω 2 ε ( ζ ξ , w ) = ∫ C ε ⟨ ξ , D μ 2 ( w ) ∣ C ∘ ⟩ h 0 + ∫ ∂ C ε ⟨ ξ , Φ ˙ ⟩ h 0 ,
ω 3 ε ( ζ ξ , w ) = ∫ C ε ⟨ ξ , D μ 3 ( w ) ∣ C ∘ ⟩ h 0 + ∫ ∂ C ε ⟨ ξ , ⋆ Φ ˙ ⟩ h 0 . (45) \omega_3^\varepsilon(\zeta_\xi,w)=\int_{\mathcal{C}_\varepsilon}\langle\xi,D\mu_3(w)|_{\mathcal{C}^{\circ}}\rangle_{h_0}+\int_{\partial\mathcal{C}_\varepsilon}\langle\xi,\star\dot{\Phi}\rangle_{h_0}. \tag*{(45)} ω 3 ε ( ζ ξ , w ) = ∫ C ε ⟨ ξ , D μ 3 ( w ) ∣ C ∘ ⟩ h 0 + ∫ ∂ C ε ⟨ ξ , ⋆ Φ ˙ ⟩ h 0 . (45)
Near π i \pi_i π i , write
u 1 , i = A ˙ i , θ , u 2 , i = − Φ ˙ i , θ , u 3 , i = − r Φ ˙ i , r u_{1,i}=\dot{A}_{i,\theta},\qquad u_{2,i}=-\dot{\Phi}_{i,\theta},\qquad u_{3,i}=-r\dot{\Phi}_{i,r} u 1 , i = A ˙ i , θ , u 2 , i = − Φ ˙ i , θ , u 3 , i = − r Φ ˙ i , r
as in (39) . The integrals over C ε C_\varepsilon C ε in (45) converge absolutely as ε → 0 \varepsilon\to0 ε → 0 , matching their respective D μ s i n t ( w ) D\mu_s^{\mathrm{int}}(w) D μ s int ( w ) . Since the inner boundary is oriented by − d θ -d\theta − d θ , the three boundary integrals become
∑ i ∫ 0 2 π ⟨ ξ ( ε , θ ) , u s , i ( ε , θ ) ⟩ h 0 d θ (46) \sum_i\int_0^{2\pi}\langle\xi(\varepsilon,\theta),u_{s,i}(\varepsilon,\theta)\rangle_{h_0}\,d\theta \tag*{(46)} i ∑ ∫ 0 2 π ⟨ ξ ( ε , θ ) , u s , i ( ε , θ ) ⟩ h 0 d θ (46)
for the respective s = 1 , 2 , 3 s=1,2,3 s = 1 , 2 , 3 . These integrals read off the leading-order coefficients ∑ i ⟨ ξ i , ν ˙ s , i ( w ) ⟩ h 0 \sum_i\langle\xi_i,\dot{\nu}_{s,i}(w)\rangle_{h_0} ∑ i ⟨ ξ i , ν ˙ s , i ( w ) ⟩ h 0 with an O ( ε δ ) O(\varepsilon^\delta) O ( ε δ ) correction, giving ⟨ D μ s b d r ( w ) , ξ ⟩ \langle D\mu_s^{\mathrm{bdr}}(w),\xi\rangle ⟨ D μ s bdr ( w ) , ξ ⟩ . This proves the identity. □
Since μ i n t = 0 \mu^{\mathrm{int}}=0 μ int = 0 gives Hitchin’s equations and μ b d = ν \mu^{\mathrm{bd}}=\nu μ bd = ν fixes the boundary parameters, Hamiltonian reduction for the extended moment map recovers the fixed-parameter moduli space
M ν = ( μ 1 − 1 ( ν 1 ) ∩ μ 2 − 1 ( ν 2 ) ∩ μ 3 − 1 ( ν 3 ) ) / G . (47) \mathcal{M}_{\nu}=\left(\mu_1^{-1}(\nu_1)\cap\mu_2^{-1}(\nu_2)\cap\mu_3^{-1}(\nu_3)\right)/G. \tag*{(47)} M ν = ( μ 1 − 1 ( ν 1 ) ∩ μ 2 − 1 ( ν 2 ) ∩ μ 3 − 1 ( ν 3 ) ) / G . (47)
As usual, the combination μ 2 + i μ 3 = 0 \mu_2+\mathrm{i}\mu_3=0 μ 2 + i μ 3 = 0 gives the holomorphicity condition 2 ∂ ˉ A φ = 0 2\bar{\partial}_A\varphi=0 2 ∂ ˉ A φ = 0 , which is now packaged together with the Levi-central specification for the residues of Φ \Phi Φ . The remaining μ 1 = 0 \mu_1=0 μ 1 = 0 gives F A + [ Φ , Φ ] = 0 F_A+[\Phi,\Phi]=0 F A + [ Φ , Φ ] = 0 together with the parabolic weight specification.
The forms ω ~ s \widetilde{\omega}_s ω s are G G G -invariant, and the Hamiltonian identity of Definition 3.7 implies they descend to closed forms ϖ 1 , ϖ 2 , ϖ 3 \varpi_1,\varpi_2,\varpi_3 ϖ 1 , ϖ 2 , ϖ 3 on E = Z / H \mathcal{E}=\mathcal{Z}/H E = Z / H . Moreover, μ b d r \mu^{\mathrm{bdr}} μ bdr descends to a triple π ~ \widetilde{\pi} π of moment maps for the torus action B ⊂ E B\subset\mathcal{E} B ⊂ E . Upon restriction to a fiber π ~ − 1 ( ν ) \widetilde{\pi}^{-1}(\nu) π − 1 ( ν ) , the forms ϖ s \varpi_s ϖ s agree with the usual L 2 L^2 L 2 Hitchin forms. This proves the geometric hypothesis of Definition 2.3.
Smooth Structure of the Moduli Family
In this section, we rigorously construct the quotients in § 3 and show that their irreducible loci are smooth. We also show q : E i r r → M i r r q:\mathcal{E}^{\mathrm{irr}}\to\mathcal{M}^{\mathrm{irr}} q : E irr → M irr is a principal B B B -bundle, and π : M i r r → Λ \pi:\mathcal{M}^{\mathrm{irr}}\to\Lambda π : M irr → Λ is a submersion. We then relate our analytic construction to the algebraic moduli family using harmonic metrics.
Throughout this section we restrict to Higgs bundles whose parabolic weights are strictly decreasing, so that
0 < α i ( k ) − α i ( k ′ ) < 1 for k < k ′ . (48) 0<\alpha_i^{(k)}-\alpha_i^{(k')}<1\qquad\text{for }k<k'. \tag*{(48)} 0 < α i ( k ) − α i ( k ′ ) < 1 for k < k ′ . (48)
We use ν = ( ν 1 , ν 2 , ν 3 ) \nu=(\nu_1,\nu_2,\nu_3) ν = ( ν 1 , ν 2 , ν 3 ) to encode the parabolic weights and residue parameters ( α , σ ) (\alpha,\sigma) ( α , σ ) as in (36) , and we use Λ \Lambda Λ for the space of parameters.
Extended Deformation Complex
We first define the analytic configuration space and restate the gauge group’s definition, and show the action is well-defined. We consider the extended deformation complex describing the tangent space of M \mathcal{M} M , and create a splitting using Biquard–Boalch’s inverse to the gauge Laplacian [4 ] . We then construct local slices with smooth coordinates near irreducible Higgs bundles in M \mathcal{M} M . We use the local slices to establish smoothness for M i r r \mathcal{M}^{\mathrm{irr}} M irr , to show that q : E i r r → M i r r q:\mathcal{E}^{\mathrm{irr}}\to\mathcal{M}^{\mathrm{irr}} q : E irr → M irr is a principal B B B -bundle, and to show the parameter map π : M i r r → Λ \pi:\mathcal{M}^{\mathrm{irr}}\to\Lambda π : M irr → Λ is a submersion. Finally, we use direct limits to allow δ → 0 \delta\to0 δ → 0 .
Fix the SL ( r ⋆ , C ) \operatorname{SL}(r_\star,\mathbb{C}) SL ( r ⋆ , C ) -bundle E → C E\to C E → C with its parabolic filtrations, background metric h 0 h_0 h 0 and analytic model frames, with h 0 h_0 h 0 normalized so that the trivialization det E ≅ O C \det E\cong\mathcal{O}_C det E ≅ O C has unit norm in h 0 h_0 h 0 . The configuration space and gauge action below are defined using H δ k \mathcal{H}^k_\delta H δ k Sobolev completions for any given k ≥ 2 k\ge2 k ≥ 2 and δ > 0 \delta>0 δ > 0 , though we will eventually specify an upper bound on δ \delta δ for later results.
Let A l o c k ( E , h 0 ) \mathcal{A}^k_{\mathrm{loc}}(E,h_0) A loc k ( E , h 0 ) be the affine space of H l o c k \mathcal{H}^k_{\mathrm{loc}} H loc k unitary connections inducing the fixed connection on det E \det E det E . We define the configuration space
C ⊆ A l o c k ( E , h 0 ) × H l o c k ( C ∘ , Ω C ‾ 1 ⊗ s u ( E , h 0 ) ) \mathcal{C}\subseteq\mathcal{A}^k_{\mathrm{loc}}(E,h_0)\times\mathcal{H}^k_{\mathrm{loc}}(C^\circ,\Omega^1_{\overline{C}}\otimes\mathfrak{su}(E,h_0)) C ⊆ A loc k ( E , h 0 ) × H loc k ( C ∘ , Ω C 1 ⊗ su ( E , h 0 ))
consisting of pairs ( A , Φ ) (A,\Phi) ( A , Φ ) such that, on each B 1 o ( p i ) B^o_1(p_i) B 1 o ( p i ) ,
A − ν 1 , i 2 π d θ ∈ H δ k ( B 1 o ( p i ) , Ω C ^ 1 ⊗ s u ( E , h 0 ) ) A-\frac{\nu_{1,i}}{2\pi}d\theta\in\mathcal{H}^k_\delta(B^o_1(p_i),\Omega^1_{\widehat{C}}\otimes\mathfrak{su}(E,h_0)) A − 2 π ν 1 , i d θ ∈ H δ k ( B 1 o ( p i ) , Ω C 1 ⊗ su ( E , h 0 ))
Φ + 1 2 π ( ν 3 , i d r r + ν 2 , i d θ ) ∈ H δ k ( B 1 o ( p i ) , Ω C ^ 1 ⊗ s u ( E , h 0 ) ) \Phi+\frac{1}{2\pi}\left(\nu_{3,i}\frac{dr}{r}+\nu_{2,i}d\theta\right)\in\mathcal{H}^k_\delta(B^o_1(p_i),\Omega^1_{\widehat{C}}\otimes\mathfrak{su}(E,h_0)) Φ + 2 π 1 ( ν 3 , i r d r + ν 2 , i d θ ) ∈ H δ k ( B 1 o ( p i ) , Ω C 1 ⊗ su ( E , h 0 ))
for some parameter triple ν \nu ν in the admissible locus Λ \Lambda Λ .
As before, we define the gauge group
G = { g ∈ H l o c k + 1 ( C ∘ , SU ( E , h 0 ) ) ∣ g − g i ∈ H δ k + 1 on B 1 o ( p i ) for some g i ∈ K i } G=\left\{g\in\mathcal{H}^{k+1}_{\mathrm{loc}}(C^\circ,\operatorname{SU}(E,h_0))\mathrel{\Big|}g-g_i\in\mathcal{H}^{k+1}_\delta\text{ on }B^o_1(p_i)\text{ for some }g_i\in K_i\right\} G = { g ∈ H loc k + 1 ( C ∘ , SU ( E , h 0 )) g − g i ∈ H δ k + 1 on B 1 o ( p i ) for some g i ∈ K i }
where K i K_i K i is the compact Levi group defined in § 3 . Multiplication is smooth by the weighted Sobolev multiplication estimate (8) . Since g g g is unitary, g − 1 = g † g^{-1}=g^\dagger g − 1 = g † , and g − 1 − g i − 1 = ( g − g i ) † ∈ H δ k + 1 g^{-1}-g_i^{-1}=(g-g_i)^\dagger\in\mathcal{H}^{k+1}_\delta g − 1 − g i − 1 = ( g − g i ) † ∈ H δ k + 1 , so inversion is smooth. The Lie algebra of G G G is
g = { ξ ∈ H l o c k + 1 ( C ∘ , s u ( E , h 0 ) ) ∣ ξ − ξ i ∈ H δ k + 1 on B 1 o ( p i ) for some ξ i ∈ k i } . \mathfrak{g}=\left\{\xi\in\mathcal{H}^{k+1}_{\mathrm{loc}}(C^\circ,\mathfrak{su}(E,h_0))\mathrel{\Big|}\xi-\xi_i\in\mathcal{H}^{k+1}_\delta\text{ on }B^o_1(p_i)\text{ for some }\xi_i\in\mathfrak{k}_i\right\}. g = { ξ ∈ H loc k + 1 ( C ∘ , su ( E , h 0 )) ξ − ξ i ∈ H δ k + 1 on B 1 o ( p i ) for some ξ i ∈ k i } .
The asymptotic model of ( A , Φ ) (A,\Phi) ( A , Φ ) near p i p_i p i is the leading order part
A m o d = ν 1 , i 2 π d θ , Φ m o d = − 1 2 π ( ν 3 , i d r r + ν 2 , i d θ ) , (49) A_{\mathrm{mod}}=\frac{\nu_{1,i}}{2\pi}d\theta,\qquad\Phi_{\mathrm{mod}}=-\frac{1}{2\pi}\left(\nu_{3,i}\frac{dr}{r}+\nu_{2,i}d\theta\right), \tag*{(49)} A mod = 2 π ν 1 , i d θ , Φ mod = − 2 π 1 ( ν 3 , i r d r + ν 2 , i d θ ) , (49)
and the asymptotic model of g g g and ξ \xi ξ are the limits g i ∈ K i g_i\in K_i g i ∈ K i and ξ i ∈ k i \xi_i\in\mathfrak{k}_i ξ i ∈ k i .
Near p i p_i p i , we use the Sobolev norm (5) on functions. We apply this to sections of E E E using h 0 h_0 h 0 and the model frames. For differential forms, we work in the b-cotangent frame d r / r dr/r d r / r , d θ d\theta d θ . For compactly supported deformations u = ( A ˙ , Φ ˙ ) u=(\dot{A},\dot{\Phi}) u = ( A ˙ , Φ ˙ ) , w = ( A ˙ ′ , Φ ˙ ′ ) w=(\dot{A}',\dot{\Phi}') w = ( A ˙ ′ , Φ ˙ ′ ) , the pairing is
( u , w ) = − ∫ C ⟨ ⋆ A ˙ ∧ A ˙ ′ ⟩ h 0 + ⟨ ⋆ Φ ˙ ∧ Φ ˙ ′ ⟩ h 0 . (u,w)=-\int_C\langle\star\dot{A}\wedge\dot{A}'\rangle_{h_0}+\langle\star\dot{\Phi}\wedge\dot{\Phi}'\rangle_{h_0}. ( u , w ) = − ∫ C ⟨ ⋆ A ˙ ∧ A ˙ ′ ⟩ h 0 + ⟨ ⋆ Φ ˙ ∧ Φ ˙ ′ ⟩ h 0 .
The Hodge star uses the metric g C g_C g C , which, in a holomorphic coordinate z = r e i θ z=re^{i\theta} z = r e i θ , has the form
g C = κ d z d z ‾ = κ ( d r 2 + r 2 d θ 2 ) , dvol = κ r d r d θ . g_C=\kappa\,dz\,d\overline{z}=\kappa(dr^2+r^2d\theta^2),\qquad\operatorname{dvol}=\kappa r\,dr\,d\theta. g C = κ d z d z = κ ( d r 2 + r 2 d θ 2 ) , dvol = κ r d r d θ .
See (11) for Hodge star formulas. These pairings extend from the spaces of compactly supported sections to the Sobolev completions, and are used to define formal adjoints below.
Lemma 4.1. The gauge group G G G acts smoothly on C \mathcal{C} C , preserving the weight and residue parameters.
Proof. Away from the punctures, this follows from ordinary Sobolev multiplication. On B 0 1 ( p i ) B^1_0(p_i) B 0 1 ( p i ) , write A = A m o d + a A=A_{\mathrm{mod}}+a A = A mod + a , Φ = Φ m o d + ψ \Phi=\Phi_{\mathrm{mod}}+\psi Φ = Φ mod + ψ , and g = g i + u g=g_i+u g = g i + u , where a , ψ ∈ H δ k a,\psi\in H^k_\delta a , ψ ∈ H δ k and u ∈ H δ k + 1 u\in H^{k+1}_\delta u ∈ H δ k + 1 . Since g i g_i g i is constant and commutes with A m o d A_{\mathrm{mod}} A mod and Φ m o d \Phi_{\mathrm{mod}} Φ mod ,
g ⋅ A − A m o d = g a g − 1 + ( [ u , A m o d ] − d u ) g − 1 , g\cdot A-A_{\mathrm{mod}}=gag^{-1}+\left([u,A_{\mathrm{mod}}]-du\right)g^{-1}, g ⋅ A − A mod = g a g − 1 + ( [ u , A mod ] − d u ) g − 1 ,
g ⋅ Φ − Φ m o d = g ψ g − 1 + [ u , Φ m o d ] g − 1 . g\cdot\Phi-\Phi_{\mathrm{mod}}=g\psi g^{-1}+[u,\Phi_{\mathrm{mod}}]g^{-1}. g ⋅ Φ − Φ mod = g ψ g − 1 + [ u , Φ mod ] g − 1 .
The Lie brackets involving u u u belong to H δ k + 1 H^{k+1}_\delta H δ k + 1 , and r ∂ r u , ∂ θ u ∈ H δ k r\partial_r u,\partial_\theta u\in H^k_\delta r ∂ r u , ∂ θ u ∈ H δ k . The weighted multiplication estimate (8) shows products of the remainders lie in H δ k H^k_\delta H δ k , with smooth dependence on g , a , ψ , ν g,a,\psi,\nu g , a , ψ , ν . Since g i g_i g i is Levi-central, the action preserves the boundary data and decay conditions. □
Let C ν ∗ = ( μ b d r ) − 1 ( ν ∗ ) \mathcal{C}_{\nu_*}=(\mu^{\mathrm{bdr}})^{-1}(\nu_*) C ν ∗ = ( μ bdr ) − 1 ( ν ∗ ) be the fixed parameter configuration space . The tangent space is
T x C ν ∗ = ker d μ x b d r = H δ k ( C ∘ , ( Ω C ^ 1 ⊕ Ω C ^ 1 ) ⊗ s u ( E , h 0 ) ) . (50) T_x\mathcal{C}_{\nu_*}=\ker d\mu_x^{\mathrm{bdr}}=H^k_\delta\left(C^\circ,\left(\Omega^1_{\widehat C}\oplus\Omega^1_{\widehat C}\right)\otimes\mathfrak{su}(E,h_0)\right). \tag*{(50)} T x C ν ∗ = ker d μ x bdr = H δ k ( C ∘ , ( Ω C 1 ⊕ Ω C 1 ) ⊗ su ( E , h 0 ) ) . (50)
Fix an irreducible solution x = ( A ∗ , Φ ∗ ) x=(A_*,\Phi_*) x = ( A ∗ , Φ ∗ ) with μ ( x ) = ν ∗ \mu(x)=\nu_* μ ( x ) = ν ∗ . The usual fixed-parameter deformation complex is
0 ⟶ g → d 0 T x C ν ∗ → d 1 H δ − 2 k − 1 ( C ∘ , Ω C ^ 2 ⊗ s u ( E , h 0 ) ) ⊕ 3 ⟶ 0 , (51) 0\longrightarrow\mathfrak{g}\xrightarrow{d_0}T_x\mathcal{C}_{\nu_*}\xrightarrow{d_1}H^{k-1}_{\delta-2}\left(C^\circ,\Omega^2_{\widehat C}\otimes\mathfrak{su}(E,h_0)\right)^{\oplus3}\longrightarrow0, \tag*{(51)} 0 ⟶ g d 0 T x C ν ∗ d 1 H δ − 2 k − 1 ( C ∘ , Ω C 2 ⊗ su ( E , h 0 ) ) ⊕ 3 ⟶ 0 , (51)
with d 0 ξ = ( − d A ∗ ξ , [ ξ , Φ ∗ ] ) d_0\xi=(-d_{A_*}\xi,[\xi,\Phi_*]) d 0 ξ = ( − d A ∗ ξ , [ ξ , Φ ∗ ]) the infinitesimal action and d 1 d_1 d 1 the restriction of d μ i n t d\mu^{\mathrm{int}} d μ int . The quotient ker d 1 / im d 0 \ker d_1/\operatorname{im}d_0 ker d 1 / im d 0 is the tangent space to the fixed-parameter moduli space M ν ∗ \mathcal{M}_{\nu_*} M ν ∗ . The adjoint d 0 ∗ d_0^* d 0 ∗ satisfies
( d 0 ξ , u ) = ( ξ , d 0 ∗ u ) , (d_0\xi,u)=(\xi,d_0^*u), ( d 0 ξ , u ) = ( ξ , d 0 ∗ u ) ,
and maps T x C ν ∗ T_x\mathcal{C}_{\nu_*} T x C ν ∗ to H δ − 2 k − 1 ( C ∘ , s u ( E , h 0 ) ) H^{k-1}_{\delta-2}(C^\circ,\mathfrak{su}(E,h_0)) H δ − 2 k − 1 ( C ∘ , su ( E , h 0 )) . The following lemma is due to Biquard and Boalch:
Lemma 4.2. Fix an admissible parameter ν ∗ ∈ Λ \nu_*\in\Lambda ν ∗ ∈ Λ with parabolic weights α i ( k ) \alpha_i^{(k)} α i ( k ) . Choose δ > 0 \delta>0 δ > 0 satisfying
δ < min i , k ( α i ( k ) − α i ( k + 1 ) ) , δ < min i ( 1 − α i ( 0 ) + α i ( ℓ i ) ) . (52) \delta<\min_{i,k}\left(\alpha_i^{(k)}-\alpha_i^{(k+1)}\right),\qquad\delta<\min_i\left(1-\alpha_i^{(0)}+\alpha_i^{(\ell_i)}\right). \tag*{(52)} δ < i , k min ( α i ( k ) − α i ( k + 1 ) ) , δ < i min ( 1 − α i ( 0 ) + α i ( ℓ i ) ) . (52) Then the gauge Laplacian
L ( A ∗ , Φ ∗ ) = d 0 ∗ d 0 : g ⟶ H δ − 2 k − 1 ( C ∘ , s u ( E , h 0 ) ) (53) \mathcal{L}_{(A_*,\Phi_*)}=d_0^*d_0:\mathfrak{g}\longrightarrow H^{k-1}_{\delta-2}(C^\circ,\mathfrak{su}(E,h_0)) \tag*{(53)} L ( A ∗ , Φ ∗ ) = d 0 ∗ d 0 : g ⟶ H δ − 2 k − 1 ( C ∘ , su ( E , h 0 )) (53) is an isomorphism at any irreducible $(A_,\Phi_ )\in\mu^{-1}(\nu_*).
Proof. In the tame case, [4 Lemma 5.1] is our situation. Their gauge Lie algebra is equal to ours, with equivalent norms for sufficiently small δ > 0 \delta>0 δ > 0 . The gauge Laplacian has the same domain and codomain.2 □
Remark 4.3. We emphasize that the condition on δ \delta δ depends on ν ∗ \nu_* ν ∗ . The quantities α i ( k ) − α i ( k + 1 ) \alpha_i^{(k)}-\alpha_i^{(k+1)} α i ( k ) − α i ( k + 1 ) and 1 − α i ( 0 ) + α i ( ℓ i ) 1-\alpha_i^{(0)}+\alpha_i^{(\ell_i)} 1 − α i ( 0 ) + α i ( ℓ i ) in (52) are the gaps between parabolic weights in R / Z \mathbb{R}/\mathbb{Z} R / Z , and the condition (48) defining Λ \Lambda Λ says these gaps are positive.
The inverse of the gauge Laplacian allows one to project deformations in T x C ν ∗ T_x\mathcal{C}_{\nu_*} T x C ν ∗ to the orthogonal complement of gauge orbits, placing them in Coulomb gauge . Biquard–Boalch also provide a useful right inverse for d 1 d_1 d 1 :
Lemma 4.4. Let
S x = ker ( d 0 ∗ : T x C ν ∗ ⟶ H δ − 2 k − 1 ( C ∘ , s u ( E , h 0 ) ) ) . S_x = \ker(d_0^*: T_x\mathcal{C}_{\nu_*} \longrightarrow\mathcal{H}^{k-1}_{\delta-2}(C^\circ,\mathfrak{su}(E,h_0))). S x = ker ( d 0 ∗ : T x C ν ∗ ⟶ H δ − 2 k − 1 ( C ∘ , su ( E , h 0 ))) . be the fixed-parameter Coulomb space at x x x . Under the hypotheses of Definition 4.2, the map
d 1 ∣ S x : S x ⟶ H δ − 1 k − 1 ( C ∘ , Ω C 2 ⊗ s u ( E , h 0 ) ) ⊕ 3 d_1|_{S_x}: S_x \longrightarrow\mathcal{H}^{k-1}_{\delta-1}(C^\circ,\Omega_C^2\otimes\mathfrak{su}(E,h_0))^{\oplus3} d 1 ∣ S x : S x ⟶ H δ − 1 k − 1 ( C ∘ , Ω C 2 ⊗ su ( E , h 0 ) ) ⊕ 3 admits a bounded right inverse G G G .
Now we define the extended deformation complex for the universal configuration space. We begin by enlarging the codomain of d 1 d_1 d 1 to include boundary variations:
I x = H δ k − 1 ( C ∘ , Ω C 2 ⊗ s u ( E , h 0 ) ) ⊕ 3 + T ν ∗ Λ ⊂ ( g ∗ ) ⊕ 3 . (54) \mathcal{I}_x = \mathcal{H}^{k-1}_{\delta}(C^\circ,\Omega_C^2\otimes\mathfrak{su}(E,h_0))^{\oplus3} + T_{\nu_*}\Lambda\subset(\mathfrak{g}^*)^{\oplus3}. \tag*{(54)} I x = H δ k − 1 ( C ∘ , Ω C 2 ⊗ su ( E , h 0 ) ) ⊕ 3 + T ν ∗ Λ ⊂ ( g ∗ ) ⊕ 3 . (54)
As before, the second summand consists of distributions supported on D D D , so that pairing I x × g ⊕ 3 → C \mathcal{I}_x \times\mathfrak{g}^{\oplus3} \to\mathbb{C} I x × g ⊕ 3 → C is defined using integration over C C C . We equip this space with the Banach norm
∥ η + ν ∥ I x = ∥ η ∥ H δ k − 1 + ∥ ν ∥ , \|\eta+\nu\|_{\mathcal{I}_x}=\|\eta\|_{\mathcal{H}^{k-1}_{\delta}}+\|\nu\|, ∥ η + ν ∥ I x = ∥ η ∥ H δ k − 1 + ∥ ν ∥ ,
using any fixed norm on ( b ∗ ) ⊕ 3 (\mathfrak{b}^*)^{\oplus3} ( b ∗ ) ⊕ 3 .
Going forward, d 0 d_0 d 0 shall refer to the extended deformation complex
0 ⟶ g → d 0 T x C → d μ x I x ⟶ 0 (55) 0 \longrightarrow\mathfrak{g} \xrightarrow{d_0} T_x\mathcal{C} \xrightarrow{d\mu_x} \mathcal{I}_x \longrightarrow0 \tag*{(55)} 0 ⟶ g d 0 T x C d μ x I x ⟶ 0 (55)
where d μ = d μ i n t + d μ b d r d\mu=d\mu^{\mathrm{int}}+d\mu^{\mathrm{bdr}} d μ = d μ int + d μ bdr is the linearization of the extended moment map. Surjectivity is shown below. We define the extended Coulomb space by
S ~ x = ker d 0 ∗ = { u ∈ T x C : ( d 0 ξ , u ) = 0 for every ξ ∈ g } . (56) \widetilde{S}_x=\ker d_0^*=\{u\in T_x\mathcal{C}:(d_0\xi,u)=0\text{ for every }\xi\in\mathfrak{g}\}. \tag*{(56)} S x = ker d 0 ∗ = { u ∈ T x C : ( d 0 ξ , u ) = 0 for every ξ ∈ g } . (56)
Although u u u need not have finite L 2 L^2 L 2 norm, its pairing with d 0 ξ d_0\xi d 0 ξ converges because of the decay condition on ξ \xi ξ on the off-diagonal blocks in the model frame. Under the hypotheses of Definition 4.2, the orthogonal projection
Π x = Id − d 0 L x − 1 d 0 ∗ : T x C ⟶ S ~ (57) \Pi_x=\operatorname{Id}-d_0\mathcal{L}_x^{-1}d_0^*:T_x\mathcal{C}\longrightarrow\widetilde{S} \tag*{(57)} Π x = Id − d 0 L x − 1 d 0 ∗ : T x C ⟶ S (57)
gives the extended Coulomb splitting
T x C = im d 0 ⊕ S ~ x , S ~ x ∩ T x C ν ∗ = S x . (58) T_x\mathcal{C}=\operatorname{im}d_0\oplus\widetilde{S}_x,\qquad\widetilde{S}_x\cap T_x\mathcal{C}_{\nu_*}=S_x. \tag*{(58)} T x C = im d 0 ⊕ S x , S x ∩ T x C ν ∗ = S x . (58)
Proposition 4.5. Under the hypotheses of Definition 4.2, the restriction of the linearized moment map to the extended Coulomb space
d μ x : S ~ x ⟶ I x d\mu_x:\widetilde{S}_x\longrightarrow\mathcal{I}_x d μ x : S x ⟶ I x admits a bounded right inverse G ~ \widetilde{G} G . In particular, im d μ x = I x \operatorname{im}d\mu_x=\mathcal{I}_x im d μ x = I x .
Proof. Given a vector in ν ˙ ∈ T ν ∗ Λ \dot{\nu} \in T_{\nu_*}\Lambda ν ˙ ∈ T ν ∗ Λ , one can construct a lift ( A ˙ , Φ ˙ ) ∈ T x C (\dot{A},\dot{\Phi}) \in T_x\mathcal{C} ( A ˙ , Φ ˙ ) ∈ T x C using bump functions and the asymptotic model (49) . Fixing the bump functions defines a continuous right inverse m : T ν ∗ Λ ⟶ T x C m:T_{\nu_*}\Lambda\longrightarrow T_x\mathcal{C} m : T ν ∗ Λ ⟶ T x C for the unrestricted d μ x d\mu_x d μ x . Composing with the projection Π x \Pi_x Π x from (57) gives
m ~ = Π x ∘ m : T ν ∗ Λ ⟶ S ~ x . \widetilde{m}=\Pi_x\circ m:T_{\nu_*}\Lambda\longrightarrow\widetilde{\mathcal{S}}_x. m = Π x ∘ m : T ν ∗ Λ ⟶ S x .
We still have d μ x b d r ∘ m ~ = Id d\mu_x^{\mathrm{bdr}}\circ\widetilde{m}=\operatorname{Id} d μ x bdr ∘ m = Id , since the term d 0 L x − 1 d 0 ∗ d_0\mathcal{L}_x^{-1}d_0^* d 0 L x − 1 d 0 ∗ in (57) is killed by d μ x b d r d\mu_x^{\mathrm{bdr}} d μ x bdr .
Let f + ν ˙ ∈ I x f+\dot{\nu}\in\mathcal{I}_x f + ν ˙ ∈ I x , where the summands respect the decomposition (54) . The lift m ~ ( ν ˙ ) \widetilde{m}(\dot{\nu}) m ( ν ˙ ) captures the boundary variation, but need not infinitesimally preserve Hitchin’s equations. However, m ~ ( ν ˙ ) − G ∘ d μ x i n t ( m ~ ( ν ˙ ) ) \widetilde{m}(\dot{\nu})-\mathcal{G}\circ d\mu_x^{\mathrm{int}}(\widetilde{m}(\dot{\nu})) m ( ν ˙ ) − G ∘ d μ x int ( m ( ν ˙ )) does:
d μ x 0 i n t ( m ~ ( ν ˙ ) − G ∘ d μ x i n t ( m ~ ( ν ˙ ) ) ) = d μ x 0 i n t ( m ~ ( ν ˙ ) ) − d μ x 0 i n t ( m ~ ( ν ˙ ) ) = 0. (59) d\mu_{x_0}^{\mathrm{int}}\left(\widetilde{m}(\dot{\nu})-\mathcal{G}\circ d\mu_x^{\mathrm{int}}(\widetilde{m}(\dot{\nu}))\right)=d\mu_{x_0}^{\mathrm{int}}(\widetilde{m}(\dot{\nu}))-d\mu_{x_0}^{\mathrm{int}}(\widetilde{m}(\dot{\nu}))=0. \tag*{(59)} d μ x 0 int ( m ( ν ˙ ) − G ∘ d μ x int ( m ( ν ˙ )) ) = d μ x 0 int ( m ( ν ˙ )) − d μ x 0 int ( m ( ν ˙ )) = 0. (59)
Here we use G \mathcal{G} G from Definition 4.4, recalling that d 1 d_1 d 1 is the restriction of d μ i n t d\mu^{\mathrm{int}} d μ int . Define
G ~ ( f + ν ˙ ) = G ( f ) + m ~ ( ν ˙ ) − G ∘ d μ x i n t ( m ~ ( ν ˙ ) ) ∈ S ~ x . (60) \widetilde{\mathcal{G}}(f+\dot{\nu})=\mathcal{G}(f)+\widetilde{m}(\dot{\nu})-\mathcal{G}\circ d\mu_x^{\mathrm{int}}(\widetilde{m}(\dot{\nu}))\in\widetilde{\mathcal{S}}_x. \tag*{(60)} G ( f + ν ˙ ) = G ( f ) + m ( ν ˙ ) − G ∘ d μ x int ( m ( ν ˙ )) ∈ S x . (60)
Since G ( f ) \mathcal{G}(f) G ( f ) is tangent to C ν ∗ = ( μ b d r ) − 1 ( ν ∗ ) \mathcal{C}_{\nu_*}=(\mu^{\mathrm{bdr}})^{-1}(\nu_*) C ν ∗ = ( μ bdr ) − 1 ( ν ∗ ) , we have
d μ x ∘ G ~ ( f + ν ˙ ) = ( d μ i n t + d μ b d r ) ( G ( f ) + m ~ ( ν ˙ ) − G ∘ d μ x i n t ( m ~ ( ν ˙ ) ) ) = d μ i n t ∘ G ( f ) + d μ b d r ( m ~ ( ν ˙ ) − G ∘ d μ x i n t ( m ~ ( ν ˙ ) ) ) = f + ν ˙ , (61) \begin{aligned}
d\mu_x\circ\widetilde{\mathcal{G}}(f+\dot{\nu})=(d\mu^{\mathrm{int}}+d\mu^{\mathrm{bdr}})(\mathcal{G}(f)+\widetilde{m}(\dot{\nu})-\mathcal{G}\circ d\mu_x^{\mathrm{int}}(\widetilde{m}(\dot{\nu}))) \\
&=d\mu^{\mathrm{int}}\circ\mathcal{G}(f)+d\mu^{\mathrm{bdr}}(\widetilde{m}(\dot{\nu})-\mathcal{G}\circ d\mu_x^{\mathrm{int}}(\widetilde{m}(\dot{\nu}))) \\
&=f+\dot{\nu}, \tag*{(61)}
\end{aligned} d μ x ∘ G ( f + ν ˙ ) = ( d μ int + d μ bdr ) ( G ( f ) + m ( ν ˙ ) − G ∘ d μ x int ( m ( ν ˙ ))) = d μ int ∘ G ( f ) + d μ bdr ( m ( ν ˙ ) − G ∘ d μ x int ( m ( ν ˙ ))) = f + ν ˙ , (61)
where the second equality used (59) , and the third used d μ x i n t ∘ G = Id d\mu_x^{\mathrm{int}}\circ\mathcal{G}=\operatorname{Id} d μ x int ∘ G = Id . □
Local Slices
When working with the universal moduli space M \mathcal{M} M , care must be taken to handle the dependence of δ \delta δ on ν ∈ Λ \nu\in\Lambda ν ∈ Λ . To address this, we fix 0 < δ ≪ 1 0<\delta\ll1 0 < δ ≪ 1 and define Λ δ ⊂ Λ \Lambda_\delta\subset\Lambda Λ δ ⊂ Λ consisting of those ν \nu ν whose parabolic weight gaps in R / Z \mathbb{R}/\mathbb{Z} R / Z are all greater than δ \delta δ (see Definition 4.3). Let C δ i r r ⊂ ( μ b d r ) − 1 ( Λ δ ) \mathcal{C}^{\mathrm{irr}}_\delta\subset(\mu^{\mathrm{bdr}})^{-1}(\Lambda_\delta) C δ irr ⊂ ( μ bdr ) − 1 ( Λ δ ) be the configuration space of irreducible Higgs bundles with parameters in Λ δ \Lambda_\delta Λ δ . Recall the subgroup H ⊂ G H\subset G H ⊂ G defined in (31) and the effective boundary torus B = G / H B=G/H B = G / H . Set E δ i r r = C δ i r r / H \mathcal{E}^{\mathrm{irr}}_\delta=\mathcal{C}^{\mathrm{irr}}_\delta/H E δ irr = C δ irr / H and M δ i r r = C δ i r r / G \mathcal{M}^{\mathrm{irr}}_\delta=\mathcal{C}^{\mathrm{irr}}_\delta/G M δ irr = C δ irr / G .
Proposition 4.6. Fix 0 < δ ≪ 1 0<\delta\ll1 0 < δ ≪ 1 and let Λ δ \Lambda_\delta Λ δ be as above. The extended moduli space of irreducible Higgs bundles M δ i r r \mathcal{M}^{\mathrm{irr}}_\delta M δ irr has a smooth structure, and the moment map descends to a submersion π : M δ i r r → Λ δ \pi:\mathcal{M}^{\mathrm{irr}}_\delta\to\Lambda_\delta π : M δ irr → Λ δ . For the partial quotient E δ = μ − 1 ( Λ δ ) / H \mathcal{E}_\delta=\mu^{-1}(\Lambda_\delta)/H E δ = μ − 1 ( Λ δ ) / H , the restriction q : E δ i r r → M δ i r r q:\mathcal{E}^{\mathrm{irr}}_\delta\to\mathcal{M}^{\mathrm{irr}}_\delta q : E δ irr → M δ irr is a principal B B B -bundle.
Proof. Let Z δ = ( μ i n t ) − 1 ( 0 ) \mathcal{Z}_\delta=(\mu^{\mathrm{int}})^{-1}(0) Z δ = ( μ int ) − 1 ( 0 ) be the solution space to Hitchin’s equations, and let x ∗ = ( A ∗ , Φ ∗ ) ∈ Z δ x_*=(A_*,\Phi_*)\in\mathcal{Z}_\delta x ∗ = ( A ∗ , Φ ∗ ) ∈ Z δ be an irreducible solution satisfying μ ( x ∗ ) = ν ∗ ∈ Λ δ \mu(x_*)=\nu_*\in\Lambda_\delta μ ( x ∗ ) = ν ∗ ∈ Λ δ . Consider the affine Coulomb slice C = x ∗ + S ~ x ∗ ⊂ C δ \mathcal{C}=x_*+\widetilde{\mathcal{S}}_{x_*}\subset\mathcal{C}_\delta C = x ∗ + S x ∗ ⊂ C δ , where S ~ x ∗ = ker d 0 ∗ ⊂ T x ∗ C \widetilde{\mathcal{S}}_{x_*}=\ker d_0^*\subset T_{x_*}\mathcal{C} S x ∗ = ker d 0 ∗ ⊂ T x ∗ C is the extended Coulomb space from (56) . At the point x ∗ x_* x ∗ , the slice C \mathcal{C} C is orthogonal to the G G G -orbit (see Figure 2 ).
Figure 2. The affine Coulomb slice C \mathcal{C} C is orthogonal to the gauge orbit at x ∗ x_* x ∗ . On the right, C \mathcal{C} C is shown as a plane containing a neighborhood U U U of x ∗ x_* x ∗ with C ∩ Z δ \mathcal{C} \cap\mathcal{Z}_{\delta} C ∩ Z δ passing through.
The tangent space of the intersection C ∩ μ − 1 ( ν ∗ ) \mathcal{C}\cap\mu^{-1}(\nu_*) C ∩ μ − 1 ( ν ∗ ) at x ∗ x_* x ∗ is the space of fixed-parameter harmonic deformations
H = S ~ x ∗ ∩ ker d μ x ∗ ⊂ T x ∗ C ν ∗ , \mathcal{H}=\widetilde{\mathcal{S}}_{x_*}\cap\ker d\mu_{x_*}\subset T_{x_*}\mathcal{C}_{\nu_*}, H = S x ∗ ∩ ker d μ x ∗ ⊂ T x ∗ C ν ∗ ,
which is finite dimensional by [4 Lemma 5.2] . The right inverse G ~ \widetilde{\mathcal{G}} G of d μ x ∗ d\mu_{x_*} d μ x ∗ from Definition 4.5 gives the splitting
S ~ x ∗ = H ⊕ G ~ ( I x ∗ ) . (62) \widetilde{\mathcal{S}}_{x_*}=\mathcal{H}\oplus\widetilde{\mathcal{G}}(\mathcal{I}_{x_*}). \tag*{(62)} S x ∗ = H ⊕ G ( I x ∗ ) . (62)
The Banach submersion theorem therefore gives local coordinates ( a , b ) (a,b) ( a , b ) on a neighborhood U ⊂ C U \subset\mathcal{C} U ⊂ C of x ∗ x_* x ∗ , where a ∈ H a \in\mathcal{H} a ∈ H , b ∈ I x ∗ b \in\mathcal{I}_{x_*} b ∈ I x ∗ , and μ ( a , b ) = ν ∗ + b \mu(a,b)=\nu_*+b μ ( a , b ) = ν ∗ + b . Restricting to U ∩ Z δ U \cap\mathcal{Z}_{\delta} U ∩ Z δ , where the interior moment map vanishes, we obtain coordinates ( a , b ) (a,b) ( a , b ) for a ∈ H a \in\mathcal{H} a ∈ H and b ∈ T ν ∗ Λ δ ⊂ I x ∗ b \in T_{\nu_*}\Lambda_{\delta} \subset\mathcal{I}_{x_*} b ∈ T ν ∗ Λ δ ⊂ I x ∗ sufficiently small. These coordinates satisfy μ ( a , b ) = ν ∗ + b ∈ Λ δ \mu(a,b)=\nu_*+b \in\Lambda_{\delta} μ ( a , b ) = ν ∗ + b ∈ Λ δ .
Claim 4.6.1. Let p ′ p' p ′ be the map taking U ∩ Z δ U \cap\mathcal{Z}_{\delta} U ∩ Z δ to its image V ⊆ M δ V \subseteq\mathcal{M}_{\delta} V ⊆ M δ under the quotient map p : Z δ → M δ p:\mathcal{Z}_{\delta} \to\mathcal{M}_{\delta} p : Z δ → M δ . After shrinking U U U and V V V if needed, p ′ p' p ′ is a homeomorphism.
Proof of Claim 4.6.1. To invert p ′ p' p ′ , we show that every orbit near [ x ∗ ] [x_*] [ x ∗ ] meets the slice U ∩ Z δ U \cap\mathcal{Z}_{\delta} U ∩ Z δ in exactly one point. For simplicity we now work with the effective gauge group G e f f = G / C r ∗ G^{\mathrm{eff}}=G/C_{r_*} G eff = G / C r ∗ (recall C r ∗ C_{r_*} C r ∗ is the r ∗ r_* r ∗ th roots of unity). Consider the map
N x : g → H δ − 2 k − 1 ( C ∘ , s u ( E , h 0 ) ) , ξ ↦ d 0 ∗ ( e ξ ⋅ x − x ∗ ) (63) \mathcal{N}_x:\mathfrak{g}\to\mathcal{H}^{k-1}_{\delta-2}(C^\circ,\mathfrak{su}(E,h_0)),\qquad\xi\mapsto d_0^*(e^\xi\cdot x-x_*) \tag*{(63)} N x : g → H δ − 2 k − 1 ( C ∘ , su ( E , h 0 )) , ξ ↦ d 0 ∗ ( e ξ ⋅ x − x ∗ ) (63)
for x ∈ C δ x \in\mathcal{C}_{\delta} x ∈ C δ . Since C = x ∗ + ker d 0 ∗ \mathcal{C}=x_*+\ker d_0^* C = x ∗ + ker d 0 ∗ , a solution to N x ( ξ ) = 0 \mathcal{N}_x(\xi)=0 N x ( ξ ) = 0 produces a gauge transformation e ξ e^\xi e ξ taking x x x into C \mathcal{C} C . The linearization d N x ∗ d\mathcal{N}_{x_*} d N x ∗ at ξ = 0 \xi=0 ξ = 0 is the gauge Laplacian, which is an isomorphism by Definition 4.2. The implicit function theorem provides a unique smooth family of gauge corrections e ξ ( x ) e^{\xi(x)} e ξ ( x ) as x x x varies in a neighborhood V ~ ⊂ C δ \widetilde{V}\subset\mathcal{C}_{\delta} V ⊂ C δ of x ∗ x_* x ∗ . Shrink the neighborhoods so that U ⊂ V ~ U\subset\widetilde{V} U ⊂ V , p ( V ~ ∩ Z δ ) = V p(\widetilde{V}\cap\mathcal{Z}_{\delta})=V p ( V ∩ Z δ ) = V , and e ξ ( x ) ⋅ x ∈ U e^{\xi(x)}\cdot x\in U e ξ ( x ) ⋅ x ∈ U for every x ∈ V ~ x\in\widetilde{V} x ∈ V . Since G G G preserves Z δ \mathcal{Z}_{\delta} Z δ , this shows every x ∈ V ~ ∩ Z δ x\in\widetilde{V}\cap\mathcal{Z}_{\delta} x ∈ V ∩ Z δ has a conjugate point in U ∩ Z δ U\cap\mathcal{Z}_{\delta} U ∩ Z δ .
For the proposed x ↦ e ξ ( x ) ⋅ x x\mapsto e^{\xi(x)}\cdot x x ↦ e ξ ( x ) ⋅ x to descend to an inverse V → U ∩ Z δ V\to U\cap\mathcal{Z}_{\delta} V → U ∩ Z δ of p ′ p' p ′ , we must prove invariance on group orbits. Concretely, this can fail if a pair x , y ∈ V ~ x,y\in\widetilde{V} x , y ∈ V is equivalent under a gauge transformation far from the identity. Allowing ourselves to shrink U U U , V ~ \widetilde{V} V , V V V further, we must show that if y n = g n ⋅ x n y_n=g_n\cdot x_n y n = g n ⋅ x n for sequences x n , y n → x ∗ x_n,y_n\to x_* x n , y n → x ∗ and g n ∈ G e f f g_n\in G^{\mathrm{eff}} g n ∈ G eff , then g n → 1 g_n\to1 g n → 1 . In the fixed-parameter setting, Biquard [3 Lemma 1.7] obtains the necessary weighted Sobolev bounds to prove g n → 1 g_n\to1 g n → 1 . To apply this to our case, we note that the parabolic weight gaps are uniformly bounded below by δ \delta δ , and the leading order Levi parts of x n , y n x_n,y_n x n , y n are constrained to a small compact set near ν ∗ \nu_* ν ∗ , so the Sobolev bounds of [3 ] still hold. This proves the claim. □3
The map p ′ p' p ′ takes the coordinates ( a , b ) (a,b) ( a , b ) on U ∩ Z δ U\cap Z_\delta U ∩ Z δ to coordinates on V V V , and the transition maps are smooth since ξ ( x ) \xi(x) ξ ( x ) varies smoothly. The map π : M δ → Λ δ \pi:\mathcal{M}_\delta\to\Lambda_\delta π : M δ → Λ δ descends from μ \mu μ , so we have π ∘ h = μ ∣ U ∩ Z δ \pi\circ h=\mu|_{U\cap Z_\delta} π ∘ h = μ ∣ U ∩ Z δ and π ( a , b ) = ν ∗ + b \pi(a,b)=\nu_*+b π ( a , b ) = ν ∗ + b . Thus π \pi π is a submersion at x ∗ x_* x ∗ .
To show q : E δ i r r → M i r r q:\mathcal{E}^{\mathrm{irr}}_\delta\to\mathcal{M}^{\mathrm{irr}} q : E δ irr → M irr is a principal B B B -bundle, recall the neighborhood V ~ ⊂ C δ \widetilde{V}\subset\mathcal{C}_\delta V ⊂ C δ of x ∗ x_* x ∗ from the proof of the claim. The implicit function theorem argument used to produce gauge corrections ξ ( x ) \xi(x) ξ ( x ) gives the homeomorphism
O × ( U ∩ Z δ ) → V ~ ∩ Z δ , ( g , y ) ↦ g ⋅ y O\times(U\cap Z_\delta)\to\widetilde{V}\cap Z_\delta,\qquad(g,y)\mapsto g\cdot y O × ( U ∩ Z δ ) → V ∩ Z δ , ( g , y ) ↦ g ⋅ y
where O ⊂ G e f f O\subset G^{\mathrm{eff}} O ⊂ G eff is a neighborhood of the identity. Since C r ∗ ⊂ H C_{r_*}\subset H C r ∗ ⊂ H , the quotient of O × ( U ∩ Z δ ) O\times(U\cap Z_\delta) O × ( U ∩ Z δ ) by H H H -equivalence gives
q − 1 ( V ) ≃ B × V . q^{-1}(V)\simeq B\times V. q − 1 ( V ) ≃ B × V .
The B B B -action is multiplication on the first factor, so q : E δ i r r → M i r r q:\mathcal{E}^{\mathrm{irr}}_\delta\to\mathcal{M}^{\mathrm{irr}} q : E δ irr → M irr is a principal B B B -bundle.
We now construct the extended moduli family M \mathcal{M} M over all of Λ \Lambda Λ by varying δ \delta δ , with the Sobolev order k ≥ 2 k\ge2 k ≥ 2 fixed. We write C δ \mathcal{C}_\delta C δ for the configuration space and G δ G_\delta G δ , H δ H_\delta H δ , B δ B_\delta B δ for the groups at weight δ \delta δ .
Proposition 4.7. As δ → 0 \delta\to0 δ → 0 , the families M δ \mathcal{M}_\delta M δ , E δ \mathcal{E}_\delta E δ , and B δ B_\delta B δ form directed systems. Denote their direct limits
M = lim δ → 0 M δ , E = lim δ → 0 E δ , B = lim δ → 0 B δ . (64) \mathcal{M}=\lim_{\delta\to0}\mathcal{M}_\delta,\qquad\mathcal{E}=\lim_{\delta\to0}\mathcal{E}_\delta,\qquad B=\lim_{\delta\to0}B_\delta. \tag*{(64)} M = δ → 0 lim M δ , E = δ → 0 lim E δ , B = δ → 0 lim B δ . (64)
The irreducible locus M i r r \mathcal{M}^{\mathrm{irr}} M irr has a smooth structure, and π : M i r r → Λ \pi:\mathcal{M}^{\mathrm{irr}}\to\Lambda π : M irr → Λ is a submersion. The preimage E i r r = q − 1 ( M i r r ) \mathcal{E}^{\mathrm{irr}}=q^{-1}(\mathcal{M}^{\mathrm{irr}}) E irr = q − 1 ( M irr ) is smooth, and q : E i r r → M i r r q:\mathcal{E}^{\mathrm{irr}}\to\mathcal{M}^{\mathrm{irr}} q : E irr → M irr is a principal B B B -bundle.
Proof. For 0 < δ ′ < δ 0<\delta'<\delta 0 < δ ′ < δ , the Sobolev inclusions H δ k ⊂ H δ ′ k \mathcal{H}^k_\delta\subset\mathcal{H}^k_{\delta'} H δ k ⊂ H δ ′ k , H δ k + 1 ⊂ H δ ′ k + 1 \mathcal{H}^{k+1}_\delta\subset\mathcal{H}^{k+1}_{\delta'} H δ k + 1 ⊂ H δ ′ k + 1 from (7) and Λ δ ⊂ Λ δ ′ \Lambda_\delta\subset\Lambda_{\delta'} Λ δ ⊂ Λ δ ′ give inclusions of the configuration spaces, solution spaces, and gauge groups, inducing maps on the quotients. Every B δ = G δ / H δ B_\delta=G_\delta/H_\delta B δ = G δ / H δ is isomorphic to the same boundary torus ( T 1 × ⋯ × T n ) / C r ∗ (T_1\times\cdots\times T_n)/C_{r_*} ( T 1 × ⋯ × T n ) / C r ∗ .
To prove injectivity of M δ → M δ ′ \mathcal{M}_\delta\to\mathcal{M}_{\delta'} M δ → M δ ′ , we must show that any g ∈ G δ ′ g\in G_{\delta'} g ∈ G δ ′ relating two points in C δ \mathcal{C}_\delta C δ in fact belongs to G δ G_\delta G δ . Denote their connection 1-forms by A A A and B = g ⋅ A = g A g − 1 − ( d g ) g − 1 B=g\cdot A=gAg^{-1}-(dg)g^{-1} B = g ⋅ A = g A g − 1 − ( d g ) g − 1 . Rearranging to d g = g A − B g dg=gA-Bg d g = g A − B g and comparing the r − 1 d r r^{-1}dr r − 1 d r and d θ d\theta d θ coefficients gives
r ∂ r g = g A r − B r g , ∂ θ g = g A θ − B θ g . (65) r\partial_rg=gA_r-B_rg,\qquad\partial_\theta g=gA_\theta-B_\theta g. \tag*{(65)} r ∂ r g = g A r − B r g , ∂ θ g = g A θ − B θ g . (65)
Since g i g_i g i commutes with the common leading-order coefficient i α i i\alpha_i i α i , we have r ∂ r ( g − g i ) ∈ H δ k r\partial_r(g-g_i)\in\mathcal{H}^k_\delta r ∂ r ( g − g i ) ∈ H δ k and ∂ θ ( g − g i ) ∈ H δ k \partial_\theta(g-g_i)\in\mathcal{H}^k_\delta ∂ θ ( g − g i ) ∈ H δ k . By the fundamental theorem of calculus,
g ( r , θ ) − g i = ∫ 0 r ( g A r − B r g ) ( s , θ ) d s s . (66) g(r,\theta)-g_i=\int_0^r(gA_r-B_rg)(s,\theta)\frac{ds}{s}. \tag*{(66)} g ( r , θ ) − g i = ∫ 0 r ( g A r − B r g ) ( s , θ ) s d s . (66)
Now A r , B r ∈ H δ k A_r,B_r\in\mathcal{H}^k_\delta A r , B r ∈ H δ k , and the integration preserves their decay weight by the estimate (9), so g − g i ∈ H δ k g-g_i\in\mathcal{H}^k_\delta g − g i ∈ H δ k . Combining, we get g − g i ∈ H δ k + 1 = G δ g-g_i\in\mathcal{H}^{k+1}_\delta=G_\delta g − g i ∈ H δ k + 1 = G δ as required. The same argument applies to H δ H_\delta H δ . The compositions M δ → π Λ δ ↪ Λ \mathcal{M}_\delta\xrightarrow{\pi}\Lambda_\delta\hookrightarrow\Lambda M δ π Λ δ ↪ Λ induce π : M → Λ \pi:\mathcal{M}\to\Lambda π : M → Λ , and similarly the maps E δ → q M δ ↪ M \mathcal{E}_\delta\xrightarrow{q}\mathcal{M}_\delta\hookrightarrow\mathcal{M} E δ q M δ ↪ M induce q : E → M q:\mathcal{E}\to\mathcal{M} q : E → M . On the irreducible locus, the local charts from Definition 4.6 are compatible as the sufficiently small Sobolev weight varies (elliptic regularity identifies the fixed-parameter harmonic deformations), so they define a smooth structure on M i r r \mathcal{M}^{\mathrm{irr}} M irr compatible with each inclusion M δ ↪ M \mathcal{M}_{\delta}\hookrightarrow\mathcal{M} M δ ↪ M . In particular, π : M i r r → Λ \pi:\mathcal{M}^{\mathrm{irr}}\to\Lambda π : M irr → Λ is a submersion, and E i r r → q M i r r \mathcal{E}^{\mathrm{irr}}\xrightarrow{q}\mathcal{M}^{\mathrm{irr}} E irr q M irr is a smooth principal B B B -bundle. □
The Algebraic Moduli Family
For the homology transport of § 5 and the characteristic class computation of § 6 , use will the algebraic description of parabolic Higgs bundles to study the family M → Λ \mathcal{M}\to\Lambda M → Λ . The existence of harmonic metrics adapted to the parabolic structure gives the translation from the algebraic construction to the analytic construction of the Higgs bundle moduli spaces when the boundary data is held fixed. We must show that the family of harmonic metrics—and therefore the conversions to the unitary formulation—depend smoothly on the parameters. In this subsection, we use h b g h_{\mathrm{bg}} h bg for the background metric from before to distinguish from a family of metrics h t h_t h t .
Fix a surface C C C , a divisor D D D , and a degree 0 0 0 bundle E → C E\to C E → C with fixed determinant structure. The algebraic data of a Higgs bundle is a holomorphic structure ∂ ˉ E \bar{\partial}_E ∂ ˉ E , a holomorphic Higgs field φ ∈ H 0 ( C , End ( E ) ⊗ K C ( D ) ) \varphi\in H^0(C,\operatorname{End}(E)\otimes K_C(D)) φ ∈ H 0 ( C , End ( E ) ⊗ K C ( D )) , and a parabolic filtration F \mathcal{F} F over the marked points in C C C . The Higgs field residues are required to preserve the filtration and act on gr F i ( k ) \operatorname{gr}\mathcal{F}_i^{(k)} gr F i ( k ) by σ i ( k ) \sigma_i^{(k)} σ i ( k ) . To define the moduli space M ( α , σ ) \mathcal{M}(\alpha,\sigma) M ( α , σ ) , one fixes the flag dimensions and associated parabolic weights ( α i ( k ) ) i , k (\alpha_i^{(k)})_{i,k} ( α i ( k ) ) i , k , and considers the space of stable Higgs bundles up to smooth complex gauge transformations in SL ( E ) = C ∞ ( C , SL ( r ⋆ , C ) ) \operatorname{SL}(E)=C^\infty(C,\operatorname{SL}(r_{\star},\mathbb{C})) SL ( E ) = C ∞ ( C , SL ( r ⋆ , C )) . Unlike the unitary formulation, the flag F \mathcal{F} F is part of the varying moduli data and the gauge group can change the flag. Stability here is given in terms of the parabolic degree of φ \varphi φ -invariant subbundles, where parabolic degree is the sum of the regular degree and all the induced parabolic weights.
Consider a local family ( ∂ ˉ E t , φ t , F t ) (\bar{\partial}_{E_t},\varphi_t,\mathcal{F}_t) ( ∂ ˉ E t , φ t , F t ) , t ∈ ( − ε , ε ) t\in(-\varepsilon,\varepsilon) t ∈ ( − ε , ε ) of stable parabolic Higgs bundles with varying parabolic weights α ( t ) \alpha(t) α ( t ) and residue blocks σ ( t ) \sigma(t) σ ( t ) . For each t t t , stability implies there exists a harmonic metric h t h_t h t solving Hitchin’s equations and adapted to the parabolic structure (see [39 ] ), which is unique up to unitary equivalence. To translate to the unitary description with background metric h b g h_{\mathrm{bg}} h bg , let B t B_t B t be the matrix satisfying h t ( v , w ) = h b g ( B t v , w ) h_t(v,w)=h_{\mathrm{bg}}(B_tv,w) h t ( v , w ) = h bg ( B t v , w ) , and let B t 1 / 2 B_t^{1/2} B t 1/2 be the unique positive Hermitian root with ( B t 1 / 2 ) † B t 1 / 2 = B t (B_t^{1/2})^\dagger B_t^{1/2}=B_t ( B t 1/2 ) † B t 1/2 = B t . Such a B t 1 / 2 B_t^{1/2} B t 1/2 is unique, and gives an isometry ( E t , h t ) → ( E , h b g ) (E_t,h_t)\to(E,h_{\mathrm{bg}}) ( E t , h t ) → ( E , h bg ) . Define ( A t , Φ t ) (A_t,\Phi_t) ( A t , Φ t ) by
d A t = B t 1 / 2 D h t B t − 1 / 2 , Φ t = B t ( φ t † h t − φ t ) B t − 1 . \mathrm{d}A_t=B_t^{1/2}D_{h_t}B_t^{-1/2},\qquad\Phi_t=B_t(\varphi_t^{\dagger_{h_t}}-\varphi_t)B_t^{-1}. d A t = B t 1/2 D h t B t − 1/2 , Φ t = B t ( φ t † h t − φ t ) B t − 1 .
Equation not extracted
(67)
Since h t h_t h t unique up to unitary transformations, this produces a well-defined element of M \mathcal{M} M . Indeed, in light of our comparison to Biquard–Boalch’s function spaces in Definition 4.2, the weighted correspondence of [4 Theorem 6.1 and Lemma 7.5] gives harmonic representatives in C δ \mathcal{C}_{\delta} C δ for sufficiently small δ > 0 \delta>0 δ > 0 , hence the need for the direct limit in Definition 4.7.
To prove smooth dependence of ( A t , Φ t ) (A_t,\Phi_t) ( A t , Φ t ) on t t t near $0$, we give a modification of [29 Theorem 4.1] . Choose smoothly varying gauge transformations to hold the flags fixed, as in [40 Section 3] . In a trivialization respecting the flag, define the model metrics h m o d , t = ⨁ k r 2 α i ( k ) ( t ) Id h_{\mathrm{mod},t}=\bigoplus_k r^{2\alpha_i^{(k)}(t)}\operatorname{Id} h mod , t = ⨁ k r 2 α i ( k ) ( t ) Id near p i p_i p i , which are adapted to the parabolic structure for all $t. Choose bump functions χ i \chi_i χ i supported near each puncture, and define S t S_t S t by
S t = ⨁ k r χ i ( r ) ( α i ( k ) ( t ) − α i ( k ) ( 0 ) ) Id . S_t=\bigoplus_k r^{\chi_i(r)(\alpha_i^{(k)}(t)-\alpha_i^{(k)}(0))}\operatorname{Id}. S t = k ⨁ r χ i ( r ) ( α i ( k ) ( t ) − α i ( k ) ( 0 )) Id .
These transformations take the harmonic metric h 0 h_0 h 0 to nearby approximate solutions h t a p p h_t^{\mathrm{app}} h t app by setting
h t a p p ( v , w ) = h 0 ( S t v , S t w ) . h_t^{\mathrm{app}}(v,w)=h_0(S_tv,S_tw). h t app ( v , w ) = h 0 ( S t v , S t w ) .
Note h 0 a p p = h 0 h_0^{\mathrm{app}}=h_0 h 0 app = h 0 , and each h t h_t h t is adapted to α ( t ) \alpha(t) α ( t ) . Set h t , u = h t a p p u h_{t,u}=h_t^{\mathrm{app}\,u} h t , u = h t app u for a u ∈ g u\in\mathfrak{g} u ∈ g . Shrinking ϵ \epsilon ϵ if needed, choose a common sufficiently small δ > 0 \delta>0 δ > 0 for all t ∈ ( − ϵ , ϵ ) t\in(-\epsilon,\epsilon) t ∈ ( − ϵ , ϵ ) . Using positive Hermitian roots as before to pick unique isometries ( E , h t , u ) → ( E , h b g ) (E,h_{t,u})\to(E,h_{\mathrm{bg}}) ( E , h t , u ) → ( E , h bg ) , we define
K t : g ⟶ H δ − 2 k − 1 ( C ∞ , i s u ( E , h b g ) ) , K t ( u ) = i ⋆ ( F h t , u + [ φ t † h t , u ∧ φ t ] ) . (68) \mathcal{K}_t:\mathfrak{g}\longrightarrow H^{k-1}_{\delta-2}(C^\infty,\mathfrak{isu}(E,h_{\mathrm{bg}})),\qquad\mathcal{K}_t(u)=\mathrm{i}\star\left(F_{h_{t,u}}+[\varphi_{t}^{\dagger h_{t,u}}\wedge\varphi_t]\right). \tag*{(68)} K t : g ⟶ H δ − 2 k − 1 ( C ∞ , isu ( E , h bg )) , K t ( u ) = i ⋆ ( F h t , u + [ φ t † h t , u ∧ φ t ] ) . (68)
By weighted Sobolev multiplication (8) and the definition of g \mathfrak{g} g , this K \mathcal{K} K is smooth and satisfies K ( t ∗ , 0 ) = 0 \mathcal{K}(t_\ast,0)=0 K ( t ∗ , 0 ) = 0 . The linearization of K 0 \mathcal{K}_0 K 0 at u = 0 u=0 u = 0 is a multiple of the gauge Laplacian, which is invertible by Definition 4.2. The implicit function theorem therefore gives a smooth correction u ( t ) u(t) u ( t ) ; uniqueness with fixed determinant identifies h t , u ( t ) h_{t,u(t)} h t , u ( t ) with h t h_t h t .
We therefore have continuous bijections between the algebraic moduli family M a l g \mathcal{M}^{\mathrm{alg}} M alg and the analytic family M \mathcal{M} M . On the smooth stable locus, the linearization of this correspondence identifies the tangent spaces computed by the respective deformation complexes, so the correspondence is a local diffeomorphism.
Homology Transport and Period Variation
We now establish a homology local system for π : M → Λ \pi:\mathcal{M}\to\Lambda π : M → Λ by locally trivializing the family over generic parameters, proving the topological hypothesis of Definition 2.3. We first fix generic parabolic weights and vary the Higgs field residues parameters, using the algebraic description from § 4 . We then allow the weights and residues parameters to vary together.
Variation of Higgs Field Residues
For a fixed flag arrangement, let
V = { σ = ( σ i ( k ) ) : ∑ k m i ( k ) σ i ( k ) = 0 for every i } . V=\left\{\sigma=(\sigma_i^{(k)}):\sum_k m_i^{(k)}\sigma_i^{(k)}=0\text{ for every }i\right\}. V = { σ = ( σ i ( k ) ) : k ∑ m i ( k ) σ i ( k ) = 0 for every i } .
be the vector space of Levi-central residue parameters. Choose parabolic weights α \alpha α for which semistability equals stability throughout this residue family, and suppose the moduli space is nonempty for at least one σ ∈ V \sigma\in V σ ∈ V . These hypotheses hold for the exceptional-type flag arrangements below. More generally, a sufficient condition for the existence of such weights is gcd ( r ⋆ , deg E , { m i , k ( k ) } i , k ) = 1 \gcd(r_\star,\deg E,\{m_{i,k}^{(k)}\}_{i,k})=1 g cd( r ⋆ , deg E , { m i , k ( k ) } i , k ) = 1 , by a similar argument as in [6 Proposition 3.2] . We continue to assume deg E = 0 \deg E=0 deg E = 0 , so this condition reduces to gcd ( r ⋆ , { m i , k ( k ) } i , k ) = 1 \gcd(r_\star,\{m_{i,k}^{(k)}\}_{i,k})=1 g cd( r ⋆ , { m i , k ( k ) } i , k ) = 1 . The local trivialization argument below also applies to other fixed degrees, provided the corresponding moduli family is smooth and its joint Hitchin map is proper.
Write the algebraic moduli family as
π α : M α a l g ⟶ V , π α − 1 ( σ ) = M ( α , σ ) . \pi_\alpha:\mathcal{M}^{\mathrm{alg}}_\alpha\longrightarrow V,\qquad\pi_\alpha^{-1}(\sigma)=\mathcal{M}(\alpha,\sigma). π α : M α alg ⟶ V , π α − 1 ( σ ) = M ( α , σ ) .
Using the trace-free deformation complex gives the fixed-determinant version of the smooth family in [33 ] . The Hitchin fibration extends to the full family. Let
B j = H 0 ( C , K C ( D ) ⊗ j ) , B = ⨁ j = 2 r ∗ B j , B_j = H^0(C,K_C(D)^{\otimes j}), \qquad\mathcal{B} = \bigoplus_{j=2}^{r_\ast} B_j, B j = H 0 ( C , K C ( D ) ⊗ j ) , B = j = 2 ⨁ r ∗ B j ,
be the Hitchin base, and define
Hit : M α a l g ⟶ B , Hit ( ∂ ‾ E , φ , F ) = ( tr ( ∧ 2 φ ) , … , tr ( ∧ r ∗ φ ) ) . (69) \operatorname{Hit}:\mathcal{M}^{\mathrm{alg}}_\alpha\longrightarrow\mathcal{B}, \qquad\operatorname{Hit}(\overline{\partial}_E,\varphi,\mathcal{F}) = \left(\operatorname{tr}(\wedge^2\varphi),\ldots,\operatorname{tr}(\wedge^{r_\ast}\varphi)\right). \tag*{(69)} Hit : M α alg ⟶ B , Hit ( ∂ E , φ , F ) = ( tr ( ∧ 2 φ ) , … , tr ( ∧ r ∗ φ ) ) . (69)
Write a j ( φ ) a_j(\varphi) a j ( φ ) for the higher traces tr ( ∧ j φ ) ∈ B j \operatorname{tr}(\wedge^j\varphi)\in B_j tr ( ∧ j φ ) ∈ B j . The joint map
( π α , Hit ) : M α a l g ⟶ V × B (\pi_\alpha,\operatorname{Hit}):\mathcal{M}^{\mathrm{alg}}_\alpha\longrightarrow V\times\mathcal{B} ( π α , Hit ) : M α alg ⟶ V × B
is proper by the argument in the proof of [32 Proposition 4.19] , restricted to the closed locus with fixed determinant and tr φ = 0 \operatorname{tr}\varphi=0 tr φ = 0 .
Following [15 ] , we consider the R > 0 \mathbb{R}_{>0} R > 0 -actions on the family and on V × B V\times\mathcal{B} V × B given by
λ ⋅ ( ∂ ‾ E , φ , F ) = ( ∂ ‾ E , λ φ , F ) , λ ⋅ ( σ , ( a j ) j ) = ( λ σ , ( λ j a j ) j ) . (70) \lambda\cdot(\overline{\partial}_E,\varphi,\mathcal{F})=(\overline{\partial}_E,\lambda\varphi,\mathcal{F}),\qquad
\lambda\cdot(\sigma,(a_j)_j)=(\lambda\sigma,(\lambda^j a_j)_j). \tag*{(70)} λ ⋅ ( ∂ E , φ , F ) = ( ∂ E , λ φ , F ) , λ ⋅ ( σ , ( a j ) j ) = ( λσ , ( λ j a j ) j ) . (70)
The first action preserves stability, and ( π α , Hit ) (\pi_\alpha,\operatorname{Hit}) ( π α , Hit ) is equivariant for these actions. Fix a point x ∈ M α a l g x\in\mathcal{M}^{\mathrm{alg}}_\alpha x ∈ M α alg and a sequence λ n → 0 \lambda_n\to0 λ n → 0 . The images of λ n ⋅ x \lambda_n\cdot x λ n ⋅ x under ( π α , Hit ) (\pi_\alpha,\operatorname{Hit}) ( π α , Hit ) converge to ( 0 , 0 ) (0,0) ( 0 , 0 ) , so they lie in a compact subset of V × B V\times\mathcal{B} V × B . Properness gives a convergent subsequence of λ n ⋅ x \lambda_n\cdot x λ n ⋅ x , with limit in π α − 1 ( 0 ) \pi_\alpha^{-1}(0) π α − 1 ( 0 ) . Since π α \pi_\alpha π α is a submersion, its image contains a neighborhood of zero. Scaling then shows that π α \pi_\alpha π α is surjective.
To transport homology, we seek horizontal lifts of paths in V V V whose flows identify the fibers of π α \pi_\alpha π α . Choosing a right inverse of d π α d\pi_\alpha d π α lifts vectors in V V V locally, but the resulting path could escape to infinity in finite time. Since the fibers of ( π α , Hit ) (\pi_\alpha,\operatorname{Hit}) ( π α , Hit ) are compact, we only need to prevent escape along the Hitchin base. We address this by making a different choice of lift away from a compact set in M α a l g \mathcal{M}^{\mathrm{alg}}_\alpha M α alg which preserves a certain norm on B \mathcal{B} B .
Theorem 5.1. For fixed generic ordered weights as above, the residue family π α : M α a l g → V \pi_\alpha:\mathcal{M}^{\mathrm{alg}}_\alpha\to V π α : M α alg → V is locally trivial with smooth trivializations. Its integral homology local systems have trivial monodromy.
Proof. Choose norms on B j B_j B j , set L = ∏ j = 2 r ∗ j L=\prod_{j=2}^{r_\ast}j L = ∏ j = 2 r ∗ j , and define
ρ : M α a l g ⟶ R ≥ 0 , ρ ( x ) = ∑ j = 2 r ∗ ∥ a j ( x ) ∥ 2 L / j . (71) \rho:\mathcal{M}^{\mathrm{alg}}_\alpha\longrightarrow\mathbb{R}_{\geq0},\qquad
\rho(x)=\sum_{j=2}^{r_\ast}\lVert a_j(x)\rVert^{2L/j}. \tag*{(71)} ρ : M α alg ⟶ R ≥ 0 , ρ ( x ) = j = 2 ∑ r ∗ ∥ a j ( x ) ∥ 2 L / j . (71)
This function factors through Hit \operatorname{Hit} Hit , say ρ = ρ ‾ ∘ Hit \rho=\overline{\rho}\circ\operatorname{Hit} ρ = ρ ∘ Hit . It is smooth and semi-invariant of degree 2 L 2L 2 L , and ( π α , ρ ) (\pi_\alpha,\rho) ( π α , ρ ) is proper because a bound on ρ \rho ρ bounds every a j a_j a j . Let ζ 1 \zeta_1 ζ 1 be the vector field on M α a l g \mathcal{M}^{\mathrm{alg}}_\alpha M α alg generated by the action of 1 ∈ Lie ( R > 0 ) 1\in\operatorname{Lie}(\mathbb{R}_{>0}) 1 ∈ Lie ( R > 0 ) . On π α − 1 ( 0 ) \pi_\alpha^{-1}(0) π α − 1 ( 0 ) ,
d π α ( ζ 1 ) = 0 , d ρ ( ζ 1 ) = 2 L ρ . d\pi_\alpha(\zeta_1)=0,\qquad d\rho(\zeta_1)=2L\rho. d π α ( ζ 1 ) = 0 , d ρ ( ζ 1 ) = 2 L ρ .
Since π α \pi_\alpha π α is a submersion, these identities show that ( π α , ρ ) (\pi_\alpha,\rho) ( π α , ρ ) is a submersion on π α − 1 ( 0 ) ∖ ρ − 1 ( 0 ) \pi_\alpha^{-1}(0)\setminus\rho^{-1}(0) π α − 1 ( 0 ) ∖ ρ − 1 ( 0 ) .
Let Z Z Z be the critical locus of ( π α , ρ ) (\pi_\alpha,\rho) ( π α , ρ ) . We claim ρ \rho ρ is bounded on Z ∩ π α − 1 ( K ) Z\cap\pi_\alpha^{-1}(K) Z ∩ π α − 1 ( K ) for any compact K ⊂ V K\subset V K ⊂ V . For suppose x n x_n x n is a sequence in Z ∩ π α − 1 ( K ) Z\cap\pi_\alpha^{-1}(K) Z ∩ π α − 1 ( K ) with ρ ( x n ) → ∞ \rho(x_n)\to\infty ρ ( x n ) → ∞ , and set λ n = ρ ( x n ) − 1 / ( 2 L ) \lambda_n = \rho(x_n)^{-1/(2L)} λ n = ρ ( x n ) − 1/ ( 2 L ) . Semi-invariance gives ρ ( λ n ⋅ x n ) = 1 \rho(\lambda_n \cdot x_n) = 1 ρ ( λ n ⋅ x n ) = 1 and by equivariance π α ( λ n ⋅ x n ) = λ n π α ( x n ) → 0 \pi_\alpha(\lambda_n \cdot x_n) = \lambda_n\pi_\alpha(x_n) \to0 π α ( λ n ⋅ x n ) = λ n π α ( x n ) → 0 . The R > 0 \mathbb{R}_{>0} R > 0 -action is smooth and takes critical points to critical points, so properness of ( π α , ρ ) (\pi_\alpha,\rho) ( π α , ρ ) gives a subsequence converging to a critical point in π α − 1 ( 0 ) ∩ ρ − 1 ( 1 ) \pi_\alpha^{-1}(0) \cap\rho^{-1}(1) π α − 1 ( 0 ) ∩ ρ − 1 ( 1 ) . Since Z Z Z is closed, this contradicts the previous paragraph.
Fix σ ∗ ∈ V \sigma_* \in V σ ∗ ∈ V and a neighborhood U U U of σ ∗ \sigma_* σ ∗ with compact closure. The above says that for R > 0 R > 0 R > 0 large enough, ( π α , ρ ) (\pi_\alpha,\rho) ( π α , ρ ) is a submersion on π α − 1 ( U ‾ ) ∩ { ρ > R } \pi_\alpha^{-1}(\overline{U}) \cap\{\rho> R\} π α − 1 ( U ) ∩ { ρ > R } . Choose a smooth right inverse r 0 r_0 r 0 to d π α d\pi_\alpha d π α on all of π α − 1 ( U ) \pi_\alpha^{-1}(U) π α − 1 ( U ) , and on π α − 1 ( U ) ∩ { ρ > R } \pi_\alpha^{-1}(U) \cap\{\rho> R\} π α − 1 ( U ) ∩ { ρ > R } choose a smooth right inverse r 1 r_1 r 1 of ( d π α , d ρ ) (d\pi_\alpha,d\rho) ( d π α , d ρ ) . Use cutoff functions to glue r 0 r_0 r 0 and r 1 ∣ T V × 0 r_1|_{TV \times0} r 1 ∣ T V × 0 to get a smooth family of sections s x : T π α ( x ) V → T x M α a l g s_x : T_{\pi_\alpha(x)}V \to T_xM_\alpha^{\mathrm{alg}} s x : T π α ( x ) V → T x M α alg of d π α d\pi_\alpha d π α satisfying
d ( π α ) x ∘ s x = id , d ρ x ∘ s x = 0 if ρ ( x ) ≥ R + 1. (72) d(\pi_\alpha)_x \circ s_x = \operatorname{id}, \qquad d\rho_x \circ s_x = 0 \quad\text{if } \rho(x) \ge R+1. \tag*{(72)} d ( π α ) x ∘ s x = id , d ρ x ∘ s x = 0 if ρ ( x ) ≥ R + 1. (72)
Figure 3. The lift of a path σ ( t ) \sigma(t) σ ( t ) in U ⊂ V U \subset V U ⊂ V to a path x ( t ) x(t) x ( t ) , which must remain in the region { ρ ≤ R } \{\rho\le R\} { ρ ≤ R } . This prevents finite-time escape along B \mathcal{B} B .
Let σ ( t ) \sigma(t) σ ( t ) be a smooth path in U U U defined for t ∈ [ 0 , 1 ] t \in[0,1] t ∈ [ 0 , 1 ] . A lift x ( t ) x(t) x ( t ) must solve x ˙ ( t ) = s x ( t ) ( σ ˙ ( t ) ) \dot{x}(t) = s_{x(t)}(\dot{\sigma}(t)) x ˙ ( t ) = s x ( t ) ( σ ˙ ( t )) . Short-time existence is automatic, and by construction, ρ ( x ( t ) ) \rho(x(t)) ρ ( x ( t )) stays bounded where x ( t ) x(t) x ( t ) is defined. By properness of ( π α , ρ ) (\pi_\alpha,\rho) ( π α , ρ ) , the lift x ( t ) x(t) x ( t ) therefore stays in compact set, and can therefore be extended until t = 1 t = 1 t = 1 . For convex U U U , lifting the radial vector field out of σ ∗ \sigma_* σ ∗ gives a smooth trivialization π α − 1 ( U ) ≅ π α − 1 ( σ ∗ ) × U \pi_\alpha^{-1}(U) \cong\pi_\alpha^{-1}(\sigma_*) \times U π α − 1 ( U ) ≅ π α − 1 ( σ ∗ ) × U .
The diffeomorphisms used for transport along a path depend on the choice of horizontal lift, but their maps on homology depend only on the endpoints of the path. The resulting integral homology local systems have trivial monodromy because V V V is contractible. □
For the upcoming period calculations, we need the transported class of a smooth Hitchin fiber to have a complex submanifold representative. We obtain such representatives by choosing the local trivialization to preserve a holomorphic level near the reference fiber.
Corollary 5.2. In the setting of Definition 5.1, let F ⊂ M ( α , σ ∗ ) F \subset\mathcal{M}(\alpha,\sigma_*) F ⊂ M ( α , σ ∗ ) be a smooth compact regular Hitchin fiber. Over a sufficiently small neighborhood U U U of σ ∗ \sigma_* σ ∗ , the smooth trivialization
can be chosen to take F F F to a smooth compact complex submanifold F σ ⊂ M ( α , σ ) F_{\sigma} \subset\mathcal{M}(\alpha,\sigma) F σ ⊂ M ( α , σ ) for every σ ∈ U \sigma\in U σ ∈ U . In particular, the transported class [ F ] [F] [ F ] is represented by F σ F_{\sigma} F σ .
Proof. Write b = Hit ( F ) b=\operatorname{Hit}(F) b = Hit ( F ) and let Hit α , σ ∗ \operatorname{Hit}_{\alpha,\sigma_*} Hit α , σ ∗ denote the restriction of Hit \operatorname{Hit} Hit to M ( α , σ ∗ ) \mathcal{M}(\alpha,\sigma_*) M ( α , σ ∗ ) . By [33 Theorem 5.4] , adapted to the fixed-determinant setting, the image of Hit α , σ ∗ \operatorname{Hit}_{\alpha,\sigma_*} Hit α , σ ∗ is smooth and has complex dimension dim C F = 1 2 dim C M ( α , σ ∗ ) \dim_{\mathbb{C}} F=\frac{1}{2}\dim_{\mathbb{C}}\mathcal{M}(\alpha,\sigma_*) dim C F = 2 1 dim C M ( α , σ ∗ ) . Choose a linear projection h : B → T b im ( Hit α , σ ∗ ) h:\mathcal{B}\to T_b\operatorname{im}(\operatorname{Hit}_{\alpha,\sigma_*}) h : B → T b im ( Hit α , σ ∗ ) . Regularity of F F F and properness of ( π α , Hit ) (\pi_{\alpha},\operatorname{Hit}) ( π α , Hit ) give neighborhoods U U U of σ ∗ \sigma_* σ ∗ and W W W of b b b on which F = ( h ∘ Hit ) − 1 ( h ( b ) ) \mathcal{F}=(h\circ\operatorname{Hit})^{-1}(h(b)) F = ( h ∘ Hit ) − 1 ( h ( b )) meets the fibers of π α \pi_{\alpha} π α transversely in complex submanifolds of dimension dim C F \dim_{\mathbb{C}}F dim C F .
Choose the horizontal lifts from Definition 5.1 tangent to F \mathcal{F} F on π α − 1 ( U ) ∩ Hit − 1 ( W ) \pi_{\alpha}^{-1}(U)\cap\operatorname{Hit}^{-1}(W) π α − 1 ( U ) ∩ Hit − 1 ( W ) (say, using another bump function around sufficiently small W W W ). Since F F F is compact, for σ \sigma σ sufficiently close to σ ∗ \sigma_* σ ∗ , the parallel transport of F F F from σ ∗ \sigma_* σ ∗ to σ \sigma σ remains in Hit − 1 ( W ) \operatorname{Hit}^{-1}(W) Hit − 1 ( W ) . The image F σ F_{\sigma} F σ is a smooth compact submanifold of F ∩ π α − 1 ( σ ) ∩ Hit − 1 ( W ) \mathcal{F}\cap\pi_{\alpha}^{-1}(\sigma)\cap\operatorname{Hit}^{-1}(W) F ∩ π α − 1 ( σ ) ∩ Hit − 1 ( W ) with the same real dimension. Since F σ F_{\sigma} F σ has no boundary, it is both closed and open in F ∩ π α − 1 ( σ ) ∩ Hit − 1 ( W ) \mathcal{F}\cap\pi_{\alpha}^{-1}(\sigma)\cap\operatorname{Hit}^{-1}(W) F ∩ π α − 1 ( σ ) ∩ Hit − 1 ( W ) , hence a complex submanifold. □
Global Period Variation
We now wish to vary the full triple ν = ( ν 1 , ν 2 , ν 3 ) ∈ Λ \nu=(\nu_1,\nu_2,\nu_3)\in\Lambda ν = ( ν 1 , ν 2 , ν 3 ) ∈ Λ under the correspondence (36) between ν \nu ν and ( α , σ ) (\alpha,\sigma) ( α , σ ) . Since the family π α : M α a l g → V \pi_{\alpha}:\mathcal{M}^{\mathrm{alg}}_{\alpha}\to V π α : M α alg → V only depends on the stability chamber W W W of α \alpha α , our topological transport extends to compact subsets of W × V W\times V W × V . To allow the weights to cross chamber walls, we exclude the simultaneous weight and residue relations associated with possible stable factors, following Nakajima’s definition of genericity [36 Section 2] .
Let g g g be the genus of C C C and n = ∣ D ∣ n=|D| n = ∣ D ∣ . If ( E , φ ) (E,\varphi) ( E , φ ) is strictly semistable, choose a proper stable Jordan–Hölder factor S S S . Write d = deg S d=\deg S d = deg S , s = rk ( S ) s=\operatorname{rk}(S) s = rk ( S ) , and s i ( k ) s_i^{(k)} s i ( k ) for its graded multiplicities. Since S S S is stable, its deformation theory gives the necessary bound
( 2 − 2 g − n ) s 2 + ∑ i , k ( s i ( k ) ) 2 ≤ 2. (73) (2-2g-n)s^2+\sum_{i,k}\left(s_i^{(k)}\right)^2\le2. \tag*{(73)} ( 2 − 2 g − n ) s 2 + i , k ∑ ( s i ( k ) ) 2 ≤ 2. (73)
This bound is automatic when g ≥ 1 g\ge1 g ≥ 1 , but can impose an additional restriction when g = 0 g=0 g = 0 . Since S S S is a proper factor of E E E , its numerical data also satisfy
0 < s < r ⋆ , 0 ≤ s i ( k ) ≤ m i ( k ) , ∑ k s i ( k ) = s for every i . (74) 0<s<r_{\star},\qquad0\le s_i^{(k)}\le m_i^{(k)},\qquad\sum_k s_i^{(k)}=s\quad\text{for every }i. \tag*{(74)} 0 < s < r ⋆ , 0 ≤ s i ( k ) ≤ m i ( k ) , k ∑ s i ( k ) = s for every i . (74)
The factor S S S has parabolic degree zero, and tr ( φ ∣ S ) \operatorname{tr}(\varphi|_S) tr ( φ ∣ S ) is a meromorphic one-form on C C C . The residue theorem therefore gives the simultaneous relations
d + ∑ i , k s i ( k ) α i ( k ) = 0 , ∑ i , k s i ( k ) σ i ( k ) = 0. (75) d+\sum_{i,k}s_i^{(k)}\alpha_i^{(k)}=0,\qquad\sum_{i,k}s_i^{(k)}\sigma_i^{(k)}=0. \tag*{(75)} d + i , k ∑ s i ( k ) α i ( k ) = 0 , i , k ∑ s i ( k ) σ i ( k ) = 0. (75)
Definition 5.3 (Generic parameter locus). A parameter ν ∈ Λ \nu\in\Lambda ν ∈ Λ is generic if the two relations (75) do not hold simultaneously for any integers d d d , s s s , ( s i ( k ) ) (s_i^{(k)}) ( s i ( k ) ) satisfying (74) and (73) . We denote the generic parameter locus by Λ 0 ⊂ Λ \Lambda_0\subset\Lambda Λ 0 ⊂ Λ .
Only finitely many integers d d d can occur in the first relation. Under the coprimality condition (76) , the excluded locus is a finite union of intersections of Λ \Lambda Λ with real codimension-3 3 3 affine subspaces, so Λ 0 \Lambda_0 Λ 0 is connected.
Corollary 5.4. Assume the parabolic flag multiplicities satisfy
gcd ( r ⋆ , { m i ( k ) } i , k ) = 1 , (76) \gcd\left(r_{\star},\left\{m_i^{(k)}\right\}_{i,k}\right)=1, \tag*{(76)} g cd( r ⋆ , { m i ( k ) } i , k ) = 1 , (76) and assume M 0 : = π − 1 ( Λ 0 ) \mathcal{M}_0 := \pi^{-1}(\Lambda_0) M 0 := π − 1 ( Λ 0 ) is nonempty. Then the family π : M 0 → Λ 0 \pi: \mathcal{M}_0 \to\Lambda_0 π : M 0 → Λ 0 is locally topologically trivial. Its integral homology local systems have trivial monodromy.
Proof. Fix ( α ∗ , σ ∗ ) ∈ Λ 0 (\alpha_{\ast},\sigma_{\ast}) \in\Lambda_0 ( α ∗ , σ ∗ ) ∈ Λ 0 and write Λ = P × V \Lambda= P \times V Λ = P × V where P P P is the space of parabolic weights and V V V is the space of Levi-central Higgs field residues. Consider the pairs of (real and complex) hyperplanes in P P P and V V V defined by the equations in (75) , indexed by numerical data satisfying (73) and (74) . Choose a neighborhood W × U ⊂ Λ 0 W \times U \subset\Lambda_0 W × U ⊂ Λ 0 of ( α ∗ , σ ∗ ) (\alpha_{\ast},\sigma_{\ast}) ( α ∗ , σ ∗ ) where W ⊂ P W \subset P W ⊂ P is convex and only meets any hyperplanes which already pass through α ∗ \alpha_{\ast} α ∗ , and where U ⊂ V U \subset V U ⊂ V remains disjoint from the corresponding hyperplanes in V V V . Then by construction, for σ ∈ U \sigma\in U σ ∈ U the stable Higgs bundle representatives are independent of α ∈ W \alpha\in W α ∈ W .
By the hypothesis (76) on the GCD, there exists a generic rational α ′ ∈ W \alpha' \in W α ′ ∈ W so that (shrinking U U U if needed) Definition 5.1 gives the local trivialization of the family M α ′ a l g \mathcal{M}^{\mathrm{alg}}_{\alpha'} M α ′ alg over { α ′ } × U \{\alpha'\} \times U { α ′ } × U , which extends to W × U W \times U W × U since stability is preserved. The harmonic metrics of § 4 gives the same trivialization for the analytic family over W × U W \times U W × U .
To identify the homology globally, we observe that Λ 0 \Lambda_0 Λ 0 is simply connected: Λ \Lambda Λ is convex, and the excluded simultaneous relations form a finite union of affine subspaces of real codimension 3. Thus, the homology local systems have trivial monodromy.
We can now apply the affine variation theorem to the family over Λ 0 \Lambda_0 Λ 0 .
Theorem 5.5 (Affine variation of periods). Let Λ 0 ⊂ Λ \Lambda_0 \subset\Lambda Λ 0 ⊂ Λ be the generic parameter locus. Assume the coprimality condition (76) and that M 0 = π − 1 ( Λ 0 ) \mathcal{M}_0 = \pi^{-1}(\Lambda_0) M 0 = π − 1 ( Λ 0 ) is nonempty. Fix ν ∗ ∈ Λ 0 \nu_{\ast} \in\Lambda_0 ν ∗ ∈ Λ 0 and identify cohomology throughout Λ 0 \Lambda_0 Λ 0 by parallel transport. Let E 0 = π ~ − 1 ( Λ 0 ) \mathcal{E}_0 = \widetilde{\pi}^{-1}(\Lambda_0) E 0 = π − 1 ( Λ 0 ) , let i ν ∗ : M ν ∗ ↪ M 0 i_{\nu_{\ast}} : \mathcal{M}_{\nu_{\ast}} \hookrightarrow\mathcal{M}_0 i ν ∗ : M ν ∗ ↪ M 0 be the fiber inclusion, and set c = i ν ∗ ∗ ( E 0 ) ∈ H 2 ( M ν ∗ , C ) c = i_{\nu_{\ast}}^*(\mathcal{E}_0) \in H_2(\mathcal{M}_{\nu_{\ast}},\mathbb{C}) c = i ν ∗ ∗ ( E 0 ) ∈ H 2 ( M ν ∗ , C ) . Then, for every ν ∈ Λ 0 \nu\in\Lambda_0 ν ∈ Λ 0 ,
[ ω s , ν ] − [ ω s , ν ∗ ] = − ⟨ c , ν s − ν ∗ , s ⟩ , s = 1 , 2 , 3. (77) [\omega_{s,\nu}] - [\omega_{s,\nu_{\ast}}] = -\langle c,\nu_s-\nu_{\ast,s}\rangle,\qquad s=1,2,3. \tag*{(77)} [ ω s , ν ] − [ ω s , ν ∗ ] = − ⟨ c , ν s − ν ∗ , s ⟩ , s = 1 , 2 , 3. (77) Proof. By Definition 4.7, q : E 0 → M 0 q : \mathcal{E}_0 \to\mathcal{M}_0 q : E 0 → M 0 is a principal B B B -bundle. The forms constructed in § 3 and the homology transport of Definition 5.4 satisfy the hypotheses of Definition 2.3, which gives the formula.
Characteristic Classes and Periods
In this section, we specialize to three families of parabolic Hitchin moduli spaces of exceptional-type (E 6 E_6 E 6 , E 7 E_7 E 7 , E 8 E_8 E 8 ) over C = C P 1 C = \mathbb{CP}^1 C = CP 1 with three marked points D = { p 1 , p 2 , p 3 } D = \{p_1,p_2,p_3\} D = { p 1 , p 2 , p 3 } . The flag dimensions are given by the corresponding affine quivers in Figure 4 , and the multiplicities are
Table 77.
Due to the repetition, we write m i = m i ( k ) m_i=m_i^{(k)} m i = m i ( k ) . By Definition 5.5, the period formula has the form A ν + b \mathbf{A}\nu+\mathbf{b} A ν + b after chooses a cohomology basis, identified throughout by parallel transport. The matrix A \mathbf{A} A is determined by the characteristic class c = c ( E ) c=c(\mathcal{E}) c = c ( E ) restricted to any given fiber, and b \mathbf{b} b is a constant cohomology class corresponding to the limit ν → 0 \nu\to0 ν → 0 . We work at convenient choices of ν \nu ν to carry out these calculations, relying on Groechenig’s orbifold parameters for homology information [21 ] , and the quiver embedding T : X ↪ M T:\mathcal{X}\hookrightarrow\mathcal{M} T : X ↪ M for evaluating the class c c c .
Our period calculation is motivated by the Modularization Conjecture : every ALG or ALG∗ ^{*} ∗ gravitational instanton should arise as a Hitchin moduli space with its natural hyper-kähler metric [15 Conjecture 1.2] . For the exceptional types E 6 E_6 E 6 , E 7 E_7 E 7 , E 8 E_8 E 8 , this requires both the appropriate ALG asymptotics and realization of the corresponding Torelli parameters. The asymptotic analysis is the subject of upcoming work of Dimakis–Fredrickson–Mazzeo–Weiss.
Homology Basis and Intersection Pairing
For each family of type E 6 E_6 E 6 , E 7 E_7 E 7 , E 8 E_8 E 8 , the integral homology H 2 ( M ( α , σ ) , Z ) H_2(\mathcal{M}(\alpha,\sigma),\mathbb{Z}) H 2 ( M ( α , σ ) , Z ) has a basis of sphere classes corresponding to the vertices of the affine diagram (see Figure 4 ). Let I ~ \widetilde{I} I be the vertex set, with central vertex v ⋆ v_{\star} v ⋆ and remaining vertices v i ( k ) v_i^{(k)} v i ( k ) , matching our flag indexing convention. At σ = 0 \sigma=0 σ = 0 and suitable generic α \alpha α near a certain orbifold weight discussed below, Groechenig’s identification with the resolution of T ∗ T / Z r ⋆ T^{*}T/\mathbb{Z}_{r_{\star}} T ∗ T / Z r ⋆ for an elliptic curve T T T shows that the nilpotent fiber of the Hitchin map has Kodaira type I V ∗ IV^{*} I V ∗ , I I I ∗ III^{*} I I I ∗ , I I ∗ II^{*} I I ∗ , respectively [21 Theorem 4.1] .4 The reduced nilpotent fiber is a deformation retract [23 Corollary 1.3.6] , so its components give a homology basis { [ S v ] ∣ v ∈ I ~ } \{[S_v]\mid v\in\widetilde{I}\} {[ S v ] ∣ v ∈ I } . Write S ⋆ = S v ⋆ S_{\star}=S_{v_{\star}} S ⋆ = S v ⋆ and S i ( k ) = S v i ( k ) S_i^{(k)}=S_{v_i^{(k)}} S i ( k ) = S v i ( k ) . The homology local system of Definition 5.4 lets us transport these classes over Λ 0 \Lambda_0 Λ 0 .
The intersection matrix is the negative affine Cartan matrix − C ~ -\widetilde{C} − C , namely the adjacency matrix minus 2 Id 2\operatorname{Id} 2 Id , and has a 1 1 1 -dimensional kernel spanned by a weighted sum of the classes corresponding to the diagram vertices. Specializing a smooth Hitchin fiber F F F to the nilpotent fiber gives the following relation:
[ F ] = r ⋆ [ S ⋆ ] + ∑ i = 1 3 ∑ k = 0 ℓ i − 1 r i ( k ) [ S i ( k ) ] . (78) [F]=r_{\star}[S_{\star}]+\sum_{i=1}^{3}\sum_{k=0}^{\ell_i-1}r_i^{(k)}[S_i^{(k)}]. \tag*{(78)} [ F ] = r ⋆ [ S ⋆ ] + i = 1 ∑ 3 k = 0 ∑ ℓ i − 1 r i ( k ) [ S i ( k ) ] . (78)
Transporting these classes throughout Λ 0 \Lambda_0 Λ 0 , the fiber relation holds everywhere. We call [ F ] [F] [ F ] the fiber class of M ( α , σ ) \mathcal{M}(\alpha,\sigma) M ( α , σ ) for any ( α , σ ) (\alpha,\sigma) ( α , σ ) .
For v ∈ I ~ v\in\widetilde{I} v ∈ I , write r v r_v r v for the corresponding flag dimension r ⋆ r_{\star} r ⋆ or r i ( k ) r_i^{(k)} r i ( k ) . The next definition follows [15 ] .
Definition 6.1. An admissible basis B = ( b v ) v ∈ I ~ \mathcal{B}=(b_v)_{v\in\widetilde{I}} B = ( b v ) v ∈ I for M ( α , σ ) \mathcal{M}(\alpha,\sigma) M ( α , σ ) is an ordered integral basis satisfying
b v ⋅ b w = − C ~ v w , ∑ v ∈ I ~ r v b v = [ F ] , (79) b_v\cdot b_w=-\widetilde{C}_{vw},\qquad\sum_{v\in\widetilde{I}}r_vb_v=[F], \tag*{(79)} b v ⋅ b w = − C v w , v ∈ I ∑ r v b v = [ F ] , (79) where C ~ \widetilde{C} C is the affine E ~ n \widetilde{E}_n E n Cartan matrix. We write C C C for the finite E n E_n E n Cartan matrix.
The classes [ S v ] [S_v] [ S v ] above form an admissible basis, and the affine Weyl group W ( E ~ n ) W(\widetilde{E}_n) W ( E n ) and diagram automorphisms act on the set of admissible bases. For later calculations, we will construct a different admissible basis extending a homology basis for an embedded quiver variety.
Definition 6.2. For a pair ( α , σ ) (\alpha,\sigma) ( α , σ ) , the corresponding quiver parameters are
β i ( k ) = α i ( k ) − α i ( k + 1 ) , τ i ( k ) = σ i ( k ) − σ i ( k + 1 ) (80) \beta_i^{(k)}=\alpha_i^{(k)}-\alpha_i^{(k+1)},\qquad\tau_i^{(k)}=\sigma_i^{(k)}-\sigma_i^{(k+1)} \tag*{(80)} β i ( k ) = α i ( k ) − α i ( k + 1 ) , τ i ( k ) = σ i ( k ) − σ i ( k + 1 ) (80) for k = 0 , … , ℓ i − 1 k=0,\ldots,\ell_i-1 k = 0 , … , ℓ i − 1 . The extension to the central node also includes
β ⋆ = ∑ i α i ( ℓ i ) , τ ⋆ = ∑ i σ i ( ℓ i ) . (81) \beta_{\star}=\sum_i\alpha_i^{(\ell_i)},\qquad\tau_{\star}=\sum_i\sigma_i^{(\ell_i)}. \tag*{(81)} β ⋆ = i ∑ α i ( ℓ i ) , τ ⋆ = i ∑ σ i ( ℓ i ) . (81) We also write β v i ( k ) = β i ( k ) \beta_{v_i^{(k)}}=\beta_i^{(k)} β v i ( k ) = β i ( k ) , β v ⋆ = β ⋆ \beta_{v_{\star}}=\beta_{\star} β v ⋆ = β ⋆ , and similarly for τ \tau τ . The extension is defined to satisfy the trace-free condition
∑ v ∈ I ~ r v β v = 0 , ∑ v ∈ I ~ r v τ v = 0. (82) \sum_{v\in\widetilde{I}}r_v\beta_v=0,\qquad\sum_{v\in\widetilde{I}}r_v\tau_v=0. \tag*{(82)} v ∈ I ∑ r v β v = 0 , v ∈ I ∑ r v τ v = 0. (82)
Choose a small generic α \alpha α satisfying
∑ i = 1 3 ∑ k = 0 ℓ i − 1 r i ( k ) β i ( k ) < 1. (83) \sum_{i=1}^{3}\sum_{k=0}^{\ell_i-1}r_i^{(k)}\beta_i^{(k)}<1. \tag*{(83)} i = 1 ∑ 3 k = 0 ∑ ℓ i − 1 r i ( k ) β i ( k ) < 1. (83)
For such α \alpha α , there is an embedding T : X ( β , τ ) ↪ M ( α , σ ) \mathcal{T}:\mathcal{X}(\beta,\tau)\hookrightarrow\mathcal{M}(\alpha,\sigma) T : X ( β , τ ) ↪ M ( α , σ ) from a star-shaped Nakajima quiver variety with underlying graph the affine Dynkin diagram [40 ] , whose image is the open locus M 0 \mathcal{M}_0 M 0 consisting of holomorphically trivial Higgs bundle.
Proposition 6.3. For any ( α , σ ) (\alpha,\sigma) ( α , σ ) satisfying (83), the inclusion j : M 0 ( α , σ ) ↪ M ( α , σ ) j:\mathcal{M}_0(\alpha,\sigma)\hookrightarrow\mathcal{M}(\alpha,\sigma) j : M 0 ( α , σ ) ↪ M ( α , σ ) gives
H 2 ( M , Z ) = j ∗ H 2 ( M 0 , Z ) ⊕ Z [ F ] . H_2(\mathcal{M},\mathbb{Z})=j_*H_2(\mathcal{M}_0,\mathbb{Z})\oplus\mathbb{Z}[F]. H 2 ( M , Z ) = j ∗ H 2 ( M 0 , Z ) ⊕ Z [ F ] .
Evaluation on [ F ] [F] [ F ] gives an isomorphism
H 2 ( M , Z ) ⟶ H 2 ( M 0 , Z ) ⊕ Z , u ⟼ ( j ∗ u , ⟨ u , [ F ] ⟩ ) . H^2(\mathcal{M},\mathbb{Z}) \longrightarrow H^2(\mathcal{M}_0,\mathbb{Z}) \oplus\mathbb{Z}, \qquad u \longmapsto(j^*u,\langle u,[F]\rangle). H 2 ( M , Z ) ⟶ H 2 ( M 0 , Z ) ⊕ Z , u ⟼ ( j ∗ u , ⟨ u , [ F ]⟩) .
Proof. Using parallel transport, it suffices to prove the statement for σ = 0 \sigma=0 σ = 0 . The deformation retract to the nilpotent fiber, which is a tree of 2-spheres, shows H 1 ( M , Z ) = 0 H_1(\mathcal{M},\mathbb{Z})=0 H 1 ( M , Z ) = 0 . Since the fiber relation (78) has a coefficient equal to 1, [ F ] [F] [ F ] is primitive. The affine Cartan pairing therefore has radical Z [ F ] \mathbb{Z}[F] Z [ F ] . The quotient by Z [ F ] \mathbb{Z}[F] Z [ F ] has intersection pairing given by the negative Cartan matrix − C -\mathrm{C} − C . The quiver variety has an integral basis with the same intersection pairing [31 , 36 ] .
Inclusion preserves intersection numbers, and the negative definiteness of the intersection pairing implies j ∗ j_* j ∗ is injective. Thus, the composition
P : H 2 ( M 0 , Z ) ⟶ H 2 ( M , Z ) ⟶ H 2 ( M , Z ) / Z [ F ] (84) P:H_2(\mathcal{M}_0,\mathbb{Z})\longrightarrow H_2(\mathcal{M},\mathbb{Z})\longrightarrow H_2(\mathcal{M},\mathbb{Z})/\mathbb{Z}[F] \tag*{(84)} P : H 2 ( M 0 , Z ) ⟶ H 2 ( M , Z ) ⟶ H 2 ( M , Z ) / Z [ F ] (84)
remains injective because Z [ F ] \mathbb{Z}[F] Z [ F ] is the radical of the pairing. Pick bases on H 2 ( M 0 , Z ) H_2(\mathcal{M}_0,\mathbb{Z}) H 2 ( M 0 , Z ) and H 2 ( M , Z ) / Z [ F ] H_2(\mathcal{M},\mathbb{Z})/\mathbb{Z}[F] H 2 ( M , Z ) / Z [ F ] whose intersection matrices are − C -\mathrm{C} − C . Since P P P preserves the pairing, its matrix with respect to these bases satisfies P T ( − C ) P = − C P^{T}(-\mathrm{C})P=-\mathrm{C} P T ( − C ) P = − C . Taking determinants gives ( det P ) 2 = 1 (\det P)^2=1 ( det P ) 2 = 1 , so P P P is an integral isomorphism. This gives the decomposition. The cohomology isomorphism follows from H 1 ( M , Z ) = 0 H_1(\mathcal{M},\mathbb{Z})=0 H 1 ( M , Z ) = 0 and the universal coefficient theorem. □
Figure 5. The nilpotent fiber and smooth Hitchin fibers in type E 6 E_6 E 6 . The image of the quiver variety homology basis together with a smooth Hitchin fiber generate H 2 ( M , Z ) H_2(\mathcal{M},\mathbb{Z}) H 2 ( M , Z ) .
The Boundary Characteristic Class
Keep ν \nu ν and the corresponding ( α , 0 ) (\alpha,0) ( α , 0 ) from above. We now calculate
c = i ν ∗ c ( E ) ∈ H 2 ( M , C ) ⊗ C b C . c=i_\nu^*c(\mathcal{E})\in H^2(\mathcal{M},\mathbb{C})\otimes_{\mathbb{C}}\mathfrak{b}_{\mathbb{C}}. c = i ν ∗ c ( E ) ∈ H 2 ( M , C ) ⊗ C b C .
where i ν : π − 1 ( ν ) = M ( α , 0 ) ↪ M i_\nu:\pi^{-1}(\nu)=\mathcal{M}(\alpha,0)\hookrightarrow\mathcal{M} i ν : π − 1 ( ν ) = M ( α , 0 ) ↪ M . By Definition 6.3, we can determine c c c by its restriction to the quiver variety X \mathcal{X} X and its evaluation on [ F ] [F] [ F ] . We define line bundles on M \mathcal{M} M relating to Nakajima’s vertex bundles on X \mathcal{X} X . The latter defines a dual basis for the homology basis for X \mathcal{X} X described above.
For each i , k i,k i , k , a gauge transformation g ∈ G g\in G g ∈ G acts on the line det ( gr ( k ) F i ) \det(\operatorname{gr}^{(k)}F_i) det ( gr ( k ) F i ) via multiplication by det ( g i ( k ) ) \det(g_i^{(k)}) det ( g i ( k ) ) , where g i = lim z → p i g g_i=\lim_{z\to p_i}g g i = lim z → p i g and g i ( k ) g_i^{(k)} g i ( k ) is the induced action on gr F i ( k ) \operatorname{gr}F_i^{(k)} gr F i ( k ) . As the Higgs bundle varies, these lines give a G G G -equivariant line bundle L i ( k ) L_i^{(k)} L i ( k ) on the configuration space. The ineffective subgroup C r ⋆ C_{r_\star} C r ⋆ acts trivially on the r ⋆ r_\star r ⋆ th tensor power of L i ( k ) L_i^{(k)} L i ( k ) , which therefore descends to M \mathcal{M} M . We write
λ i ( k ) = 1 r ⋆ c 1 ( ( L i ( k ) ) ⊗ r ⋆ ) ∈ H 2 ( M , Q ) . (85) \lambda_i^{(k)}=\frac{1}{r_\star}c_1\left((L_i^{(k)})^{\otimes r_\star}\right)\in H^2(\mathcal{M},\mathbb{Q}). \tag*{(85)} λ i ( k ) = r ⋆ 1 c 1 ( ( L i ( k ) ) ⊗ r ⋆ ) ∈ H 2 ( M , Q ) . (85)
To express c c c in terms of these classes, choose a principal B B B -connection Θ \Theta Θ on E → M \mathcal{E}\to\mathcal{M} E → M and restrict it to our reference fiber π − 1 ( ν ) \pi^{-1}(\nu) π − 1 ( ν ) . Reading off the scalar on each block gives an identification
b ≅ { ( a i ( k ) ) ∈ ⨁ i , k i R : ∑ k = 0 ℓ i m i a i ( k ) = 0 for each i } . \mathfrak{b}\cong\left\{(a_i^{(k)})\in\bigoplus_{i,k}\mathrm{i}\mathbb{R}:\sum_{k=0}^{\ell_i}m_i a_i^{(k)}=0\text{ for each }i\right\}. b ≅ ⎩ ⎨ ⎧ ( a i ( k ) ) ∈ i , k ⨁ i R : k = 0 ∑ ℓ i m i a i ( k ) = 0 for each i ⎭ ⎬ ⎫ .
under which the curvature 2-form F Θ F_\Theta F Θ can be decomposed into i R \mathrm{i}\mathbb{R} i R -valued components F Θ , i ( k ) F_{\Theta,i}^{(k)} F Θ , i ( k ) . The action of G G G on ( L i ( k ) ) ⊗ r ⋆ (L_i^{(k)})^{\otimes r_\star} ( L i ( k ) ) ⊗ r ⋆ has weight r ⋆ m i r_\star m_i r ⋆ m i , so the induced connection from Θ \Theta Θ is d − r ⋆ m i Θ i ( k ) \mathrm{d}-r_\star m_i\Theta_i^{(k)} d − r ⋆ m i Θ i ( k ) (the minus sign comes from our left-action convention). The Chern class of a bundle V V V is c 1 ( V ) = 1 2 π i [ tr ( F V ) ] c_1(V)=\frac{1}{2\pi\mathrm{i}}[\operatorname{tr}(F_V)] c 1 ( V ) = 2 π i 1 [ tr ( F V )] , so
λ i ( k ) = m i 2 π i [ F Θ , i ( k ) ] , \lambda_i^{(k)}=\frac{m_i}{2\pi\mathrm{i}}[F_{\Theta,i}^{(k)}], λ i ( k ) = 2 π i m i [ F Θ , i ( k ) ] ,
and
c = ( [ F Θ , i ( k ) ] ) i , k = ( 2 π i m i λ i ( k ) ) i , k . (86) c=\left([F_{\Theta,i}^{(k)}]\right)_{i,k}=\left(\frac{2\pi\mathrm{i}}{m_i}\lambda_i^{(k)}\right)_{i,k}. \tag*{(86)} c = ( [ F Θ , i ( k ) ] ) i , k = ( m i 2 π i λ i ( k ) ) i , k . (86)
The fixed determinant of E E E gives the relation ∑ k λ i ( k ) = 0 \sum_k\lambda_i^{(k)}=0 ∑ k λ i ( k ) = 0 .
On the quiver side, consider Nakajima’s tautological vertex bundles R v \mathcal{R}_v R v over X \mathcal{X} X [36 Section 2] . We normalize the collection of vertex bundles by R a ≅ O \mathcal{R}_a\cong\mathcal{O} R a ≅ O . Write t v = c 1 ( det R v ) t_v=c_1(\det\mathcal{R}_v) t v = c 1 ( det R v ) , with t a = 0 t_a=0 t a = 0 .
The lift of T \mathcal{T} T to configuration spaces identifies the non-central vertex spaces with the flag levels F i ( k ) F_i^{(k)} F i ( k ) [40 ] . As before, although the ineffective group C r ⋆ C_{r_\star} C r ⋆ may act nontrivially on their determinant bundles, the r ⋆ r_\star r ⋆ th powers descend. We define
q i ( k ) = 1 r ⋆ c 1 ( ( det R i ( k ) ) ⊗ r ⋆ ) . q_i^{(k)}=\frac{1}{r_\star}c_1\left((\det\mathcal{R}_i^{(k)})^{\otimes r_\star}\right). q i ( k ) = r ⋆ 1 c 1 ( ( det R i ( k ) ) ⊗ r ⋆ ) .
Lemma 6.4. Under this renormalization, Nakajima’s vertex bundle Chern classes q i ( k ) ∈ H 2 ( X , Q ) q_i^{(k)}\in H^2(\mathcal{X},\mathbb{Q}) q i ( k ) ∈ H 2 ( X , Q ) and our block determinant line bundle Chern classes λ i ( k ) ∈ H 2 ( M , Q ) \lambda_i^{(k)}\in H^2(\mathcal{M},\mathbb{Q}) λ i ( k ) ∈ H 2 ( M , Q ) satisfy
T ∗ λ i ( k ) = q i ( k ) − q i ( k − 1 ) (87) \mathcal{T}^\ast\lambda_i^{(k)}=q_i^{(k)}-q_i^{(k-1)} \tag*{(87)} T ∗ λ i ( k ) = q i ( k ) − q i ( k − 1 ) (87) where q i ( − 1 ) q_i^{(-1)} q i ( − 1 ) and q i ( ℓ i ) q_i^{(\ell_i)} q i ( ℓ i ) are understood to be 0 0 0 .
Proof. The exact sequence
0 ⟶ F i ( k − 1 ) ⟶ F i ( k ) ⟶ gr F i ( k ) ⟶ 0 0\longrightarrow F_i^{(k-1)}\longrightarrow F_i^{(k)}\longrightarrow\operatorname{gr}F_i^{(k)}\longrightarrow0 0 ⟶ F i ( k − 1 ) ⟶ F i ( k ) ⟶ gr F i ( k ) ⟶ 0
gives the relation
c 1 ( L i ( k ) ) = c 1 ( det ( gr F i ( k ) ) ) = c 1 ( det R i ( k ) ) − c 1 ( det R i ( k − 1 ) ) . c_1(L_i^{(k)})=c_1\left(\det(\operatorname{gr}F_i^{(k)})\right)=c_1\left(\det\mathcal{R}_i^{(k)}\right)-c_1\left(\det\mathcal{R}_i^{(k-1)}\right). c 1 ( L i ( k ) ) = c 1 ( det ( gr F i ( k ) ) ) = c 1 ( det R i ( k ) ) − c 1 ( det R i ( k − 1 ) ) .
Dividing by r ⋆ r_\star r ⋆ proves the lemma. □ \square □
Nakajima produces an integral basis e w ∈ H 2 ( X , Z ) e_w \in H_2(\mathcal{X},\mathbb{Z}) e w ∈ H 2 ( X , Z ) indexed by the set I I I obtained by deleting a rank-1 vertex from I ~ \widetilde{I} I (see Figure 6 ). We will use I = I ~ ∖ { v 1 ( 0 ) } I=\widetilde{I}\setminus\{v_1^{(0)}\} I = I ∖ { v 1 ( 0 ) } . The intersection pairing in this basis is given by the negative Cartan matrix as before. In Nakajima’s normalization R v 1 ( 0 ) ≅ O \mathcal{R}_{v_1^{(0)}}\cong\mathcal{O} R v 1 ( 0 ) ≅ O , the Chern classes of the remaining R v \mathcal{R}_v R v form a dual basis { t v } v ∈ I \{t_v\}_{v\in I} { t v } v ∈ I [37 ] . Our normalization q v = t v − ( r v / r ⋆ ) t ⋆ q_v=t_v-(r_v/r_\star)t_\star q v = t v − ( r v / r ⋆ ) t ⋆ only change on the pairings with e ⋆ e_\star e ⋆ . Writing e i ( k ) = e v i ( k ) e_i^{(k)}=e_{v_i}^{(k)} e i ( k ) = e v i ( k ) , the classes duality pairing for the q i ( k ) q_i^{(k)} q i ( k ) is
Figure 6. Deleting the grey rank-1 vertex from the affine E 7 E_7 E 7 quiver gives the intersection configuration of seven 2-spheres.
⟨ q i ( k ) , e j ( h ) ⟩ = δ i j δ k h , ⟨ q i ( k ) , e ⋆ ⟩ = − r i ( k ) r ⋆ . (88) \left\langle q_i^{(k)},e_j^{(h)}\right\rangle=\delta_{ij}\delta_{kh},\qquad\left\langle q_i^{(k)},e_\star\right\rangle=-\frac{r_i^{(k)}}{r_\star}. \tag*{(88)} ⟨ q i ( k ) , e j ( h ) ⟩ = δ ij δ k h , ⟨ q i ( k ) , e ⋆ ⟩ = − r ⋆ r i ( k ) . (88)
Here 0 ≤ k < ℓ i 0\leq k<\ell_i 0 ≤ k < ℓ i and v j ( h ) ∈ I ∖ { ⋆ } v_j^{(h)}\in I\setminus\{\star\} v j ( h ) ∈ I ∖ { ⋆ } , so the omitted vertex v 1 ( 0 ) v_1^{(0)} v 1 ( 0 ) is excluded. Thus Definition 6.4 gives
⟨ T ⋆ λ i ( k ) , e j ( h ) ⟩ = δ i j ( δ k h − δ k − 1 , h ) , \left\langle\mathcal{T}^\star\lambda_i^{(k)},e_j^{(h)}\right\rangle=\delta_{ij}\left(\delta_{kh}-\delta_{k-1,h}\right), ⟨ T ⋆ λ i ( k ) , e j ( h ) ⟩ = δ ij ( δ k h − δ k − 1 , h ) ,
⟨ T ⋆ λ i ( k ) , e ⋆ ⟩ = δ k , ℓ i − m i ( k ) r ⋆ . (89) \left\langle\mathcal{T}^\star\lambda_i^{(k)},e_\star\right\rangle=\delta_{k,\ell_i}-\frac{m_i^{(k)}}{r_\star}. \tag*{(89)} ⟨ T ⋆ λ i ( k ) , e ⋆ ⟩ = δ k , ℓ i − r ⋆ m i ( k ) . (89)
Here 0 ≤ k ≤ ℓ i 0\leq k\leq\ell_i 0 ≤ k ≤ ℓ i and v j ( h ) ∈ I ∖ { ⋆ } v_j^{(h)}\in I\setminus\{\star\} v j ( h ) ∈ I ∖ { ⋆ } . By (86) , this gives the pairing between c = ι v ⋆ c ( E ) c=\iota_v^\star c(\mathcal{E}) c = ι v ⋆ c ( E ) and H 2 ( M 0 , Z ) ⊂ H 2 ( M , Z ‾ ) H_2(\mathcal{M}_0,\mathbb{Z})\subset H_2(\mathcal{M},\overline{\mathbb{Z}}) H 2 ( M 0 , Z ) ⊂ H 2 ( M , Z ) . It remains to pair c c c with a smooth Hitchin fiber F F F .
Lemma 6.5. The boundary characteristic class c = ι v ⋆ c ( E ) c=\iota_v^\star c(\mathcal{E}) c = ι v ⋆ c ( E ) satisfies ⟨ c , [ F ] ⟩ = 0 \langle c,[F]\rangle=0 ⟨ c , [ F ]⟩ = 0 .
Proof. Fix a parameter ν \nu ν corresponding to a pair ( α , σ ) (\alpha,\sigma) ( α , σ ) , and let ν ′ \nu' ν ′ correspond to a pair ( α , σ ′ ) (\alpha,\sigma') ( α , σ ′ ) for σ ′ \sigma' σ ′ close to σ \sigma σ . Write F F F for a smooth Hitchin fiber for ν \nu ν . By Definition 5.2, the transport of [ F ] [F] [ F ] from ν \nu ν to ν ′ \nu' ν ′ can be represented by a smooth complex curve F σ ′ F_{\sigma'} F σ ′ . This implies the holomorphic symplectic form Ω ν ′ = ω 2 , ν ′ + i ω 3 , ν ′ \Omega_{\nu'}=\omega_{2,\nu'}+i\omega_{3,\nu'} Ω ν ′ = ω 2 , ν ′ + i ω 3 , ν ′ restricts to zero on F σ ′ F_{\sigma'} F σ ′ , so Definition 5.5 gives
0 = ∫ F σ ′ Ω ν ′ − ∫ F Ω ν = − ⟨ ( ν 2 ′ − ν 2 ) + i ( ν 3 ′ − ν 3 ) , c ( F ) ⟩ . (90) 0=\int_{F_{\sigma'}}\Omega_{\nu'}-\int_F\Omega_\nu=-\left\langle(\nu'_2-\nu_2)+i(\nu'_3-\nu_3),c(F)\right\rangle. \tag*{(90)} 0 = ∫ F σ ′ Ω ν ′ − ∫ F Ω ν = − ⟨ ( ν 2 ′ − ν 2 ) + i ( ν 3 ′ − ν 3 ) , c ( F ) ⟩ . (90)
As σ ′ \sigma' σ ′ varies near σ \sigma σ , this forces c ( F ) c(F) c ( F ) to pair trivially with every element of b ∗ ⊗ C \mathfrak{b}^*\otimes\mathbb{C} b ∗ ⊗ C , so c ( F ) = 0 c(F)=0 c ( F ) = 0 . □ \square □
The affine variation formula and Definition 6.5 implies that throughout Λ 0 \Lambda_0 Λ 0 ,
( ∫ F ω 1 , ∫ F Ω ) = ( A , 0 ) (91) \left(\int_F \omega_1,\int_F \Omega\right)=(A,0) \tag*{(91)} ( ∫ F ω 1 , ∫ F Ω ) = ( A , 0 ) (91)
for a constant A > 0 A>0 A > 0 . Indeed, at the reference parameter F F F is a compact complex curve, so its volume with respect to ω 1 \omega_1 ω 1 is positive and its holomorphic symplectic period vanishes.
Slope of Period Variation
We now use the characteristic class to compute period variation in an admissible homology basis extending the image of Nakajima’s basis under T ∗ \mathcal{T}_* T ∗ . Call the deleted vertex a = v 1 ( 0 ) ∈ I ~ ∖ I a=v_1^{(0)}\in\widetilde{I}\setminus I a = v 1 ( 0 ) ∈ I ∖ I , and set η v = T ∗ e v \eta_v=\mathcal{T}_*e_v η v = T ∗ e v for v ∈ I v\in I v ∈ I . Since r a = 1 r_a=1 r a = 1 , we can complete these classes by
η a = [ F ] − ∑ v ∈ I r v η v . (92) \eta_a=[F]-\sum_{v\in I}r_v\eta_v. \tag*{(92)} η a = [ F ] − v ∈ I ∑ r v η v . (92)
By Definition 6.3, B = ( η v ) v ∈ I ~ \mathcal{B}=(\eta_v)_{v\in\widetilde{I}} B = ( η v ) v ∈ I is an integral basis, and the intersection submatrix with η a \eta_a η a removed is − C -C − C . Since [ F ] [F] [ F ] is in the radical of the pairing, the fiber relation (78) determines the remaining pairings and gives the negative affine Cartan matrix − C ~ -\widetilde{C} − C . Thus B \mathcal{B} B is admissible in the sense of Definition 6.1 . We transport this basis throughout Λ 0 \Lambda_0 Λ 0 using Definition 5.4.
For a deformation ( α ˙ , σ ˙ ) (\dot{\alpha},\dot{\sigma}) ( α ˙ , σ ˙ ) of ( α , σ ) (\alpha,\sigma) ( α , σ ) , consider the variation of quiver parameters extended to the central node:
β ˙ i ( k ) = α ˙ i ( k ) − α ˙ i ( k + 1 ) , β ˙ ∗ = ∑ i α ˙ i ( ℓ i ) , τ ˙ i ( k ) = σ ˙ i ( k ) − σ ˙ i ( k + 1 ) , τ ˙ ∗ = ∑ i σ ˙ i ( ℓ i ) . (93) \begin{aligned}
\dot{\beta}_i^{(k)}&=\dot{\alpha}_i^{(k)}-\dot{\alpha}_i^{(k+1)}, & \dot{\beta}_*&=\sum_i\dot{\alpha}_i^{(\ell_i)},\\
\dot{\tau}_i^{(k)}&=\dot{\sigma}_i^{(k)}-\dot{\sigma}_i^{(k+1)}, & \dot{\tau}_*&=\sum_i\dot{\sigma}_i^{(\ell_i)}. \tag*{(93)}
\end{aligned} β ˙ i ( k ) τ ˙ i ( k ) = α ˙ i ( k ) − α ˙ i ( k + 1 ) , = σ ˙ i ( k ) − σ ˙ i ( k + 1 ) , β ˙ ∗ τ ˙ ∗ = i ∑ α ˙ i ( ℓ i ) , = i ∑ σ ˙ i ( ℓ i ) . (93)
As in Definition 6.2 , we write β ˙ v i ( k ) = β ˙ i ( k ) \dot{\beta}_{v_i^{(k)}}=\dot{\beta}_i^{(k)} β ˙ v i ( k ) = β ˙ i ( k ) , β ˙ v ∗ = β ˙ ∗ \dot{\beta}_{v_*}=\dot{\beta}_* β ˙ v ∗ = β ˙ ∗ , and similarly for τ ˙ \dot{\tau} τ ˙ . We still have
∑ v ∈ I ~ r v β ˙ v = 0 , ∑ v ∈ I ~ r v τ ˙ v = 0. (94) \sum_{v\in\widetilde{I}}r_v\dot{\beta}_v=0,\qquad\sum_{v\in\widetilde{I}}r_v\dot{\tau}_v=0. \tag*{(94)} v ∈ I ∑ r v β ˙ v = 0 , v ∈ I ∑ r v τ ˙ v = 0. (94)
Although these numbers come from the quiver embedding, the following proposition applies throughout Λ 0 \Lambda_0 Λ 0 by parallel transport.
Proposition 6.6. Let D D D denote the Gauss–Manin connection on the degree-two cohomology bundle over Λ 0 \Lambda_0 Λ 0 . Let ν ˙ ∈ T ν Λ 0 \dot{\nu}\in T_\nu\Lambda_0 ν ˙ ∈ T ν Λ 0 correspond to variations ( α ˙ , σ ˙ ) (\dot{\alpha},\dot{\sigma}) ( α ˙ , σ ˙ ) of the weights and residues, and let ( β ˙ , τ ˙ ) (\dot{\beta},\dot{\tau}) ( β ˙ , τ ˙ ) be the variations in (93) . Let Ω = ω 2 + i ω 3 \Omega=\omega_2+i\omega_3 Ω = ω 2 + i ω 3 . In the admissible homology basis B \mathcal{B} B ,
⟨ D ν ˙ [ ω 1 ] , η v ⟩ = − 4 π 2 β ˙ v , ⟨ D ν ˙ [ Ω ] , η v ⟩ = − 8 π 2 τ ˙ v (95) \langle D_{\dot{\nu}}[\omega_1],\eta_v\rangle=-4\pi^2\dot{\beta}_v,\qquad\langle D_{\dot{\nu}}[\Omega],\eta_v\rangle=-8\pi^2\dot{\tau}_v \tag*{(95)} ⟨ D ν ˙ [ ω 1 ] , η v ⟩ = − 4 π 2 β ˙ v , ⟨ D ν ˙ [ Ω ] , η v ⟩ = − 8 π 2 τ ˙ v (95) for all vertices v ∈ I ~ v\in\widetilde{I} v ∈ I . Moreover,
⟨ D ν ˙ [ ω 1 ] , [ F ] ⟩ = 0 , ⟨ D ν ˙ [ Ω ] , [ F ] ⟩ = 0. (96) \langle D_{\dot{\nu}}[\omega_1],[F]\rangle=0,\qquad\langle D_{\dot{\nu}}[\Omega],[F]\rangle=0. \tag*{(96)} ⟨ D ν ˙ [ ω 1 ] , [ F ]⟩ = 0 , ⟨ D ν ˙ [ Ω ] , [ F ]⟩ = 0. (96)
Proof. By (77) , the characteristic class c c c is constant under transport and determines the slopes throughout Λ 0 \Lambda_0 Λ 0 by D ν ˙ [ ω s ] = − ⟨ ν ˙ s , c ⟩ D_{\dot{\nu}}[\omega_s]=-\langle\dot{\nu}_s,c\rangle D ν ˙ [ ω s ] = − ⟨ ν ˙ s , c ⟩ . It therefore suffices to use the pairings computed at the reference parameter used above. The variation u = ν ˙ s u=\dot{\nu}_s u = ν ˙ s decomposes into u i = ⨁ k i u i ( k ) u_i = \bigoplus_k i u_i^{(k)} u i = ⨁ k i u i ( k ) Id with the trace-free condition ∑ k m i u i ( k ) = 0 \sum_k m_i u_i^{(k)} = 0 ∑ k m i u i ( k ) = 0 . The pairing formulas (89) give
⟨ D i [ ω s ] , η v i ( k ) ⟩ = − 2 π ( u i ( k ) − u i ( k + 1 ) ) , ⟨ D i [ ω s ] , η v ⋆ ⟩ = − 2 π ∑ i u i ( ℓ i ) . (97) \begin{aligned}
\langle D_i[\omega_s],\eta_{v_i^{(k)}}\rangle&= -2\pi\left(u_i^{(k)}-u_i^{(k+1)}\right),\\
\langle D_i[\omega_s],\eta_{v_\star}\rangle&= -2\pi\sum_i u_i^{(\ell_i)}. \tag*{(97)}
\end{aligned} ⟨ D i [ ω s ] , η v i ( k ) ⟩ ⟨ D i [ ω s ] , η v ⋆ ⟩ = − 2 π ( u i ( k ) − u i ( k + 1 ) ) , = − 2 π i ∑ u i ( ℓ i ) . (97)
The normalization (36) identifies u i ( k ) u_i^{(k)} u i ( k ) with 2 π α ˙ i ( k ) 2\pi\dot{\alpha}_i^{(k)} 2 π α ˙ i ( k ) , 4 π Re σ ˙ i ( k ) 4\pi\operatorname{Re}\dot{\sigma}_i^{(k)} 4 π Re σ ˙ i ( k ) , and 4 π Im σ ˙ i ( k ) 4\pi\operatorname{Im}\dot{\sigma}_i^{(k)} 4 π Im σ ˙ i ( k ) for s = 1 , 2 , 3 s=1,2,3 s = 1 , 2 , 3 , respectively, giving the desired formulas. The pairings with [ F ] [F] [ F ] follow from (92) and (94) . □
Explicit Period Formula
The numbers β i ( k ) \beta_i^{(k)} β i ( k ) and τ i ( k ) \tau_i^{(k)} τ i ( k ) give coordinates on Λ 0 \Lambda_0 Λ 0 . In a fixed cohomology basis, the affine period map has the form A ν + b \mathbf{A}\nu+\mathbf{b} A ν + b , where A \mathbf{A} A encodes the characteristic classes calculated above, and b \mathbf{b} b is a fixed cohomology class representing the limit as β → 0 \beta\to0 β → 0 , τ → 0 \tau\to0 τ → 0 . It suffices to calculate the period for any fixed reference parameter. First, we study the limit β → 0 \beta\to0 β → 0 directly with σ = 0 \sigma=0 σ = 0 by comparing to the quiver variety, whose period goes to 0 0 0 in the limit. Taking the limit along such a ray is particularly convenient, since the complex period [ ω 2 ] + i [ ω 3 ] [\omega_2]+\mathrm{i}[\omega_3] [ ω 2 ] + i [ ω 3 ] vanishes. To evaluate [ ω 1 ] [\omega_1] [ ω 1 ] on the additional homology class, we choose a different reference parameter to evaluate the period on the Hitchin fiber class [ F ] [F] [ F ] , which is constant throughout Λ 0 \Lambda_0 Λ 0 by (91) .
Set σ = 0 \sigma=0 σ = 0 , and choose a small generic α \alpha α for which the quiver embedding T \mathcal{T} T exists and stable parabolic bundles with zero Higgs field exist. Such weights exist by [30 Proposition 4.4] and the embedding T \mathcal{T} T . For all t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] , the second cohomology of M 0 ( t α , 0 ) \mathcal{M}_0(t\alpha,0) M 0 ( t α , 0 ) is generated by the spheres in the nilpotent fiber. Under the U ( 1 ) \mathrm{U}(1) U ( 1 ) -action λ ⋅ ( ∂ ˉ E , φ , F ) = ( ∂ ˉ E , λ φ , F ) \lambda\cdot(\bar{\partial}_E,\varphi,\mathcal{F})=(\bar{\partial}_E,\lambda\varphi,\mathcal{F}) λ ⋅ ( ∂ ˉ E , φ , F ) = ( ∂ ˉ E , λ φ , F ) , the central sphere is pointwise fixed, and each remaining sphere has two fixed points [26 Section 2] .
For t ∈ ( 0 , 1 ] t\in(0,1] t ∈ ( 0 , 1 ] , write ω 1 , t \omega_{1,t} ω 1 , t for the Kähler form of M ( t α , 0 ) \mathcal{M}(t\alpha,0) M ( t α , 0 ) . The U ( 1 ) \mathrm{U}(1) U ( 1 ) -action is Hamiltonian for ω 1 , t \omega_{1,t} ω 1 , t , and admits a moment map Ψ t : M ( t α , 0 ) → R \Psi_t:\mathcal{M}(t\alpha,0)\to\mathbb{R} Ψ t : M ( t α , 0 ) → R given by
Ψ t ( ∂ ˉ E , φ , F ) = i 2 ∫ C tr ( φ † h t ∧ φ ) . (98) \Psi_t(\bar{\partial}_E,\varphi,\mathcal{F})=\frac{\mathrm{i}}{2}\int_C\operatorname{tr}(\varphi^{\dagger h_t}\wedge\varphi). \tag*{(98)} Ψ t ( ∂ ˉ E , φ , F ) = 2 i ∫ C tr ( φ † h t ∧ φ ) . (98)
Here h t h_t h t is the harmonic metric for ( ∂ ˉ E , φ , F ) (\bar{\partial}_E,\varphi,\mathcal{F}) ( ∂ ˉ E , φ , F ) , adapted to t α t\alpha t α . This is a Morse–Bott function by [GGM04].
Lemma 6.7. Let α \alpha α be the parabolic weights chosen above, and let η v ∈ H 2 ( M 0 , C ) \eta_v\in H_2(\mathcal{M}_0,\mathbb{C}) η v ∈ H 2 ( M 0 , C ) for v ∈ I v\in I v ∈ I be the classes described in § 6.3 . Using parallel transport to identify homology,
lim t → 0 ⟨ [ ω 1 , t ] , η v ⟩ = lim t → 0 ∫ η v ω 1 , t = 0. \lim_{t\to0}\langle[\omega_{1,t}],\eta_v\rangle=\lim_{t\to0}\int_{\eta_v}\omega_{1,t}=0. t → 0 lim ⟨[ ω 1 , t ] , η v ⟩ = t → 0 lim ∫ η v ω 1 , t = 0.
Proof. Our choice of α \alpha α contains the φ = 0 \varphi=0 φ = 0 locus as the U ( 1 ) \mathrm{U}(1) U ( 1 ) -fixed sphere, which we address last. The remaining spheres each have two fixed points. At these points, [GGM04, Proposition 3.4] gives a splitting as parabolic holomorphic bundles
E = ⨁ j = 1 r ⋆ E j , φ j : = φ ∣ E j : E j ⟶ E j + 1 ⊗ K ( D ) , 1 ≤ j < r ⋆ . E=\bigoplus_{j=1}^{r_\star}E_j,\qquad\varphi_j:=\left.\varphi\right|_{E_j}:E_j\longrightarrow E_{j+1}\otimes K(D),\qquad1\leq j<r_\star. E = j = 1 ⨁ r ⋆ E j , φ j := φ ∣ E j : E j ⟶ E j + 1 ⊗ K ( D ) , 1 ≤ j < r ⋆ .
Here zero summands are allowed and φ r ⋆ = 0 \varphi_{r_\star}=0 φ r ⋆ = 0 . Consider the action of U ( 1 ) \mathrm{U}(1) U ( 1 ) on E = ⨁ j E j E=\bigoplus_j E_j E = ⨁ j E j given by multiplication by λ j \lambda^j λ j on E j E_j E j . This action sends φ \varphi φ to λ φ \lambda\varphi λ φ , so uniqueness of the harmonic metric implies that it preserves h t h_t h t . The summands E j E_j E j have distinct weights under this action and are therefore orthogonal for h t h_t h t .
This decomposition lets us evaluate Φ \Phi Φ using the curvature of each summand. Define the j j j th component
Ψ j , t = i 2 ∫ C ∖ D tr ( φ j † h t ∧ φ j ) , Ψ 0 , t = Ψ r ⋆ , t = 0. (99) \Psi_{j,t}=\frac{\mathrm{i}}{2}\int_{C\setminus D}\operatorname{tr}\left(\varphi_j^{\dagger h_t}\wedge\varphi_j\right),\qquad\Psi_{0,t}=\Psi_{r_\star,t}=0. \tag*{(99)} Ψ j , t = 2 i ∫ C ∖ D tr ( φ j † h t ∧ φ j ) , Ψ 0 , t = Ψ r ⋆ , t = 0. (99)
Since the splitting is orthogonal and the determinant metric is flat, the component of Hitchin’s equation on E j E_j E j is
F j , t + φ j † h t ∧ φ j + φ j − 1 ∧ φ j − 1 † h t = 0 , F_{j,t}+\varphi_j^{\dagger h_t}\wedge\varphi_j+\varphi_{j-1}\wedge\varphi_{j-1}^{\dagger h_t}=0, F j , t + φ j † h t ∧ φ j + φ j − 1 ∧ φ j − 1 † h t = 0 ,
where φ 0 = φ r ⋆ = 0 \varphi_0=\varphi_{r_\star}=0 φ 0 = φ r ⋆ = 0 . Taking traces,
tr F j , t = tr ( φ j − 1 † h t ∧ φ j − 1 ) − tr ( φ j † h t ∧ φ j ) . \operatorname{tr} F_{j,t}=\operatorname{tr}\left(\varphi_{j-1}^{\dagger h_t}\wedge\varphi_{j-1}\right)-\operatorname{tr}\left(\varphi_j^{\dagger h_t}\wedge\varphi_j\right). tr F j , t = tr ( φ j − 1 † h t ∧ φ j − 1 ) − tr ( φ j † h t ∧ φ j ) .
Substituting (99) and applying parabolic Chern–Weil theory yields
π pardeg t α ( E j ) = i 2 ∫ C ∖ D tr F j , t = Ψ j − 1 , t − Ψ j , t . (100) \pi\operatorname{pardeg}_{t\alpha}(E_j)=\frac{\mathrm{i}}{2}\int_{C\setminus D}\operatorname{tr}F_{j,t}=\Psi_{j-1,t}-\Psi_{j,t}. \tag*{(100)} π pardeg t α ( E j ) = 2 i ∫ C ∖ D tr F j , t = Ψ j − 1 , t − Ψ j , t . (100)
The parabolic degree is in terms of the induced parabolic structure on E j E_j E j . Taking the weighted sum and telescoping,
π ∑ j = 1 r ⋆ j pardeg t α ( E j ) = ∑ j = 1 r ⋆ j ( Ψ j − 1 , t − Ψ j , t ) = ∑ j = 1 r ⋆ − 1 Ψ j , t = Ψ t . (101) \pi\sum_{j=1}^{r_\star}j\operatorname{pardeg}_{t\alpha}(E_j)=\sum_{j=1}^{r_\star}j(\Psi_{j-1,t}-\Psi_{j,t})=\sum_{j=1}^{r_\star-1}\Psi_{j,t}=\Psi_t. \tag*{(101)} π j = 1 ∑ r ⋆ j pardeg t α ( E j ) = j = 1 ∑ r ⋆ j ( Ψ j − 1 , t − Ψ j , t ) = j = 1 ∑ r ⋆ − 1 Ψ j , t = Ψ t . (101)
Since E E E is holomorphically trivial, each E j E_j E j must have topological degree 0 0 0 . Therefore the parabolic degrees are just the sums of the weights, so Ψ t ∈ O ( t ) \Psi_t\in O(t) Ψ t ∈ O ( t ) at the fixed point.
Let x − , x + x_-,x_+ x − , x + be the U ( 1 ) \mathrm{U}(1) U ( 1 ) -fixed points of a sphere S S S ordered by increasing Ψ t \Psi_t Ψ t . Let d S d_S d S be the order of the stabilizer of S S S . Integrating in action-angle coordinates gives
∫ S ω 1 , t = 2 π d S ( Ψ t ( x + ) − Ψ t ( x − ) ) ∈ O ( t ) . \int_S\omega_{1,t}=\frac{2\pi}{d_S}\left(\Psi_t(x_+)-\Psi_t(x_-)\right)\in O(t). ∫ S ω 1 , t = d S 2 π ( Ψ t ( x + ) − Ψ t ( x − ) ) ∈ O ( t ) .
It remains to address the U ( 1 ) \mathrm{U}(1) U ( 1 ) -fixed sphere, where φ = 0 \varphi=0 φ = 0 and Hitchin’s equations reduce to the flat unitary connection equation. Jeffrey’s small-weight comparison theorem [25 Theorem 6.6] identifies these moduli spaces, for sufficiently small t t t , with a fixed symplectic manifold whose symplectic form is multiplied by t t t . Hence the sphere’s volume with respect to ω 1 , t \omega_{1,t} ω 1 , t tends to zero linearly in t t t . □
To compute the remaining fiber period ⟨ [ ω 1 ] , [ F ] ⟩ \langle[\omega_1],[F]\rangle ⟨[ ω 1 ] , [ F ]⟩ , we use the orbifold weights associated with the elliptic quotients of orders 3 3 3 , 4 4 4 , 6 6 6 for types E 6 E_6 E 6 , E 7 E_7 E 7 , E 8 E_8 E 8 , respectively (see [21 Section 4] ). Here σ = 0 \sigma=0 σ = 0 and, translating to our degree-zero and trace-free convention, the parabolic weights are
Table 102.
Lemma 6.8. The Hitchin fiber class satisfies
⟨ [ ω 1 ] , [ F ] ⟩ = 4 π 2 , ⟨ [ Ω ] , [ F ] ⟩ = 0. \langle[\omega_1],[F]\rangle= 4\pi^2,\qquad\langle[\Omega],[F]\rangle= 0. ⟨[ ω 1 ] , [ F ]⟩ = 4 π 2 , ⟨[ Ω ] , [ F ]⟩ = 0. Proof. The complex period was shown to be zero in (91) . For the real period, let T → C T \to C T → C be the elliptic cover and T ∨ = Pic 0 ( T ) T^\vee= \operatorname{Pic}\nolimits_0(T) T ∨ = Pic 0 ( T ) . As the parameter approaches the orbifold parameter in [21 Section 4] , the exterior spheres collapse to zero volume and the central sphere survives as S ⋆ = T ∨ / Z r ⋆ S_\star= T^\vee/\mathbb{Z}_{r_\star} S ⋆ = T ∨ / Z r ⋆ . The harmonic bundles descend from orthogonal sums of flat line bundles on T T T , so the induced L 2 L^2 L 2 metric is the flat quotient metric whose pullback to T ∨ T^\vee T ∨ has volume 4 π 2 4\pi^2 4 π 2 .
These orbifold parameters are not in Λ 0 \Lambda_0 Λ 0 , but the points in a smooth Hitchin fiber F F F remain stable here, so smoothness of M i r r \mathcal{M}^{\mathrm{irr}} M irr identifies the periods with the limits from nearby admissible parameters. Thus, the fiber relation (78) shows F F F has volume 4 π 2 4\pi^2 4 π 2 as well. □
Theorem 6.9. Let ν ∈ Λ 0 \nu\in\Lambda_0 ν ∈ Λ 0 encode weights and residues ( α , σ ) (\alpha,\sigma) ( α , σ ) , and define ( β , τ ) (\beta,\tau) ( β , τ ) as in Definition 6.2 . Let B \mathcal{B} B be the admissible basis for H 2 ( M ( α , σ ) , C ) H_2(\mathcal{M}(\alpha,\sigma),\mathbb{C}) H 2 ( M ( α , σ ) , C ) from § 6.3 , indexed by the affine Dynkin vertex set I ~ \widetilde{I} I with distinguished vertex a = v 0 ( 0 ) a=v_0^{(0)} a = v 0 ( 0 ) . For all v ∈ I ~ v \in\widetilde{I} v ∈ I ,
∫ η v ω 1 = 4 π 2 δ v a − 4 π 2 β v , ∫ η v Ω = − 8 π 2 τ v . (102) \int_{\eta_v}\omega_1 = 4\pi^2\delta_{va} - 4\pi^2\beta_v,\qquad\int_{\eta_v}\Omega= -8\pi^2\tau_v. \tag*{(102)} ∫ η v ω 1 = 4 π 2 δ v a − 4 π 2 β v , ∫ η v Ω = − 8 π 2 τ v . (102) Proof. We know that for τ = 0 \tau= 0 τ = 0 the complex period vanishes, as the nilpotent fiber generates degree-2 homology and consists of complex curves. Combining the period slopes of Definition 6.6 with the reference periods of Definition 6.7 and Definition 6.8 proves the theorem. □
Remark 6.10. Type D 4 D_4 D 4 parabolic Hitchin moduli spaces parameterize rank 2 bundle with four marked points and full flags. Its period calculation is already contained in [15 ] , so we do not cover it in depth here. Nevertheless, our same construction applies and the analogous formulas hold.
The affine E ~ n \widetilde{E}_n E n diagrams have n + 1 n+1 n + 1 vertices. The real numbers β v \beta_v β v and complex numbers τ v \tau_v τ v for v ∈ I ~ v \in\widetilde{I} v ∈ I are only subject to the trace-free constraints (82) , so β \beta β has n n n real degrees of freedom and τ \tau τ has n n n complex degrees of freedom. Thus the linearized period map has full real rank 3 n 3n 3 n . Thus, the marked period map is a local affine diffeomorphism onto an open subset of the space of period triples with fiber period ( 4 π 2 , 0 , 0 ) (4\pi^2,0,0) ( 4 π 2 , 0 , 0 ) .
Using Definition 6.9, local surjectivity can almost be upgraded to global surjectivity. Following the strategy of [15 Section 9] , we vary the admissible marking by the affine Weyl group and use its action to move the real periods into the image of the allowed weight domain. This potentially excludes a real codimension-2 subset of allowable parameters.
Theorem 6.11. In the E 6 E_6 E 6 , E 7 E_7 E 7 , E 8 E_8 E 8 cases, as α , σ \alpha,\sigma α , σ vary, the classes [ ω 1 ] , [ ω 2 ] , [ ω 3 ] [\omega_1],[\omega_2],[\omega_3] [ ω 1 ] , [ ω 2 ] , [ ω 3 ] nearly exhaust the allowable Torelli parameters in the corresponding family of ALG spaces. Namely, the image excludes a set of real codimension at least 2 2 2 .
For parameters admitting the quiver embedding (see (83) ), the induced map T ∗ : H 2 ( X , C ) ↪ H 2 ( M , C ) \mathcal T^*:H_2(\mathcal X,\mathbb C)\hookrightarrow H_2(\mathcal M,\mathbb C) T ∗ : H 2 ( X , C ) ↪ H 2 ( M , C ) is injective, and its image has integral basis { η v ∣ v ∈ I } \{\eta_v\mid v\in I\} { η v ∣ v ∈ I } . By Definition 6.9, the pulled-back Kähler forms have the same homology pairings as the forms ω s q u i v \omega_s^{\mathrm{quiv}} ω s quiv on the quiver variety [31 ] , up to a normalization factor.
Corollary 6.14. For the exceptional types E 6 , E 7 , E 8 E_6,E_7,E_8 E 6 , E 7 , E 8 , let ( α , σ ) (\alpha,\sigma) ( α , σ ) be a pair admitting the quiver embedding T : X ( β , τ ) ↪ M ( α , σ ) \mathcal T:\mathcal X(\beta,\tau)\hookrightarrow\mathcal M(\alpha,\sigma) T : X ( β , τ ) ↪ M ( α , σ ) . In the normalization and hyperkähler conventions of [40 ] , we have
T ∗ [ ω s ] = 2 π [ ω s q u i v ] , s = 1 , 2 , 3. \mathcal T^*[\omega_s]=2\pi[\omega_s^{\mathrm{quiv}}],\qquad s=1,2,3. T ∗ [ ω s ] = 2 π [ ω s quiv ] , s = 1 , 2 , 3.
Appendix A. Quasi-Hamiltonian Reduction
One of our ideas along the way to proving affine period variation for parabolic Higgs bundles was to pass through the nonabelian Hodge correspondence to the Betti moduli space, which admits a construction as a quasi-Hamiltonian quotient [41 ] . At that time, we proved a corollary of Definition 2.3 which applies to group-valued moment maps. Restricting to a central logarithmic chart gives the necessary geometric hypothesis, and the topological hypothesis would follow from a similar Laplacian estimate as in § 2.2 , which became quite a challenge in this case. Here, we present just the corollary.
Let G G G be a finite-dimensional Lie group with an invariant nondegenerate symmetric bilinear form ( , ) (\, ,\, ) ( , ) on g \mathfrak{g} g over F = R \mathbb{F} = \mathbb{R} F = R or C \mathbb{C} C . Write θ L \theta_{\mathrm{L}} θ L and θ R \theta_{\mathrm{R}} θ R for the Maurer–Cartan forms and χ = 1 12 ( θ L , [ θ L , θ L ] ) \chi= \frac{1}{12}(\theta_{\mathrm{L}},[\theta_{\mathrm{L}},\theta_{\mathrm{L}}]) χ = 12 1 ( θ L , [ θ L , θ L ]) for the Cartan three-form. A quasi-Hamiltonian G G G -space ( M , ω , μ ) (M,\omega,\mu) ( M , ω , μ ) has an invariant two-form ω \omega ω and an equivariant map μ : M → G \mu: M \to G μ : M → G satisfying
d ω = − μ ∗ χ , \mathrm{d}\omega= -\mu^{*}\chi, d ω = − μ ∗ χ ,
ι ζ ξ ω = 1 2 μ ∗ ( θ L + θ R , ξ ) , \iota_{\zeta_{\xi}}\omega= \frac{1}{2}\mu^{*}(\theta_{\mathrm{L}}+\theta_{\mathrm{R}},\xi), ι ζ ξ ω = 2 1 μ ∗ ( θ L + θ R , ξ ) ,
ker ω x = { ζ ξ ( x ) : ( Ad μ ( x ) + 1 ) ξ = 0 } . (103) \ker\omega_{x} = \{\zeta_{\xi}(x) : (\operatorname{Ad}_{\mu(x)}+1)\xi= 0\}. \tag*{(103)} ker ω x = { ζ ξ ( x ) : ( Ad μ ( x ) + 1 ) ξ = 0 } . (103)
These are the conventions of [1 Definition 2.2] . In the complex case we use holomorphic data as in [41 Definition 2.11] , with smooth transport on the underlying manifolds.
Fix z ∗ ∈ Z ( G ) z_{*} \in Z(G) z ∗ ∈ Z ( G ) and a subspace Λ ⊆ g G \Lambda\subseteq\mathfrak{g}^{G} Λ ⊆ g G , which we identify with a subspace of ( g ∗ ) G (\mathfrak{g}^{*})^{G} ( g ∗ ) G using the bilinear form. Choose an open region Λ 0 ⊆ Λ \Lambda_{0} \subseteq\Lambda Λ 0 ⊆ Λ such that
e ∗ : Λ 0 ⟶ Z ( G ) , α ⟼ z ∗ exp ( α ) e_{*} : \Lambda_{0} \longrightarrow Z(G), \qquad\alpha\longmapsto z_{*}\exp(\alpha) e ∗ : Λ 0 ⟶ Z ( G ) , α ⟼ z ∗ exp ( α )
is a diffeomorphism onto its image, which we call C 0 C_{0} C 0 . Set E 0 = μ − 1 ( C 0 ) E_{0} = \mu^{-1}(C_{0}) E 0 = μ − 1 ( C 0 ) and π ~ = e ∗ − 1 ∘ μ \widetilde{\pi} = e_{*}^{-1} \circ\mu π = e ∗ − 1 ∘ μ . Let G e f f G^{\mathrm{eff}} G eff be the quotient by the kernel of the action on E 0 E_{0} E 0 , and assume Λ \Lambda Λ annihilates the ineffective Lie algebra, so its elements define covectors on g e f f \mathfrak{g}^{\mathrm{eff}} g eff .
Theorem A.1. Suppose E 0 E_{0} E 0 is smooth and q : E 0 → M 0 = E 0 / G q : E_{0} \to\mathcal{M}_{0} = E_{0}/G q : E 0 → M 0 = E 0 / G is a smooth principal G e f f G^{\mathrm{eff}} G eff -bundle over a paracompact manifold. Suppose the induced map π : M 0 → Λ 0 \pi: \mathcal{M}_{0} \to\Lambda_{0} π : M 0 → Λ 0 is a smooth surjective submersion with symplectic reductions M α = μ − 1 ( e ∗ ( α ) ) / G \mathcal{M}_{\alpha} = \mu^{-1}(e_{*}(\alpha))/G M α = μ − 1 ( e ∗ ( α )) / G as fibers, and admits a compatible homology local system in the sense of Definition 2.1 . Let c = c ( E 0 ) ∈ H 2 ( M 0 , F ) ⊗ Λ ∗ \mathsf{c} = \mathsf{c}(E_{0}) \in H^{2}(\mathcal{M}_{0},\mathbb{F}) \otimes\Lambda^{*} c = c ( E 0 ) ∈ H 2 ( M 0 , F ) ⊗ Λ ∗ be the curvature class defined in the proof of Definition 2.3, and write c α ∗ = ι α ∗ ∗ c \mathsf{c}_{\alpha_{*}} = \iota_{\alpha_{*}}^{*}\mathsf{c} c α ∗ = ι α ∗ ∗ c for its restriction under the inclusion ι α ∗ : M α ∗ ↪ M 0 \iota_{\alpha_{*}} : \mathcal{M}_{\alpha_{*}} \hookrightarrow\mathcal{M}_{0} ι α ∗ : M α ∗ ↪ M 0 . Then, for every path γ \gamma γ from α ∗ \alpha_{*} α ∗ to α \alpha α in Λ 0 \Lambda_{0} Λ 0 ,
PT γ ∗ [ ω α ] = [ ω α ∗ ] − ⟨ c α ∗ , α − α ∗ ⟩ . (104) \operatorname{PT}_{\gamma}^{*}[\omega_{\alpha}] = [\omega_{\alpha_{*}}] - \langle\mathsf{c}_{\alpha_{*}},\alpha-\alpha_{*}\rangle. \tag*{(104)} PT γ ∗ [ ω α ] = [ ω α ∗ ] − ⟨ c α ∗ , α − α ∗ ⟩ . (104)
Proof. Since C 0 ⊂ Z ( G ) C_{0} \subset Z(G) C 0 ⊂ Z ( G ) , we have e ∗ ∗ θ L = e ∗ ∗ θ R = d α e_{*}^{*}\theta_{\mathrm{L}} = e_{*}^{*}\theta_{\mathrm{R}} = \mathrm{d}\alpha e ∗ ∗ θ L = e ∗ ∗ θ R = d α and e ∗ ∗ χ = 0 e_{*}^{*}\chi= 0 e ∗ ∗ χ = 0 . For the inclusion j : E 0 ↪ M j : E_{0} \hookrightarrow M j : E 0 ↪ M , the first two identities in (103) therefore give
d ( j ∗ ω ) = 0 , ι ζ ξ j ∗ ω = d ( π ~ , ξ ) ∀ ξ ∈ g . (105) \mathrm{d}(j^{*}\omega)=0,\qquad\iota_{\zeta_{\xi}}j^{*}\omega= \mathrm{d}(\widetilde{\pi},\xi)\quad\forall\xi\in\mathfrak{g}. \tag*{(105)} d ( j ∗ ω ) = 0 , ι ζ ξ j ∗ ω = d ( π , ξ ) ∀ ξ ∈ g . (105)
The pairing ( π ~ , ξ ) (\widetilde{\pi},\xi) ( π , ξ ) depends only on the image of ξ \xi ξ in g e f f \mathfrak{g}^{\mathrm{eff}} g eff because Λ \Lambda Λ annihilates the ineffective Lie algebra. Thus ϖ = j ∗ ω \varpi= j^{*}\omega ϖ = j ∗ ω satisfies the geometric hypothesis of Definition 2.3 with B = G e f f B = G^{\mathrm{eff}} B = G eff , and its restrictions descend to the reduced forms ω α \omega_{\alpha} ω α . This proves the theorem. □
References
Email address: pdimakis@umd.edu
Email address: lfredric@uoregon.edu
Email address: aryayae@outlook.com