Visible Lagrangians for Hitchin Systems and Pillowcase Covers

Locally at a smooth point b \in B'^\mathrm{reg} the subvariety B' \subset B is cut out by k=\mathrm{codim} B' many functions f_1,\dots,f_k\in \mathcal{O}_B. We can associate invariant vector fields X_i \in VM\: \rule[-3.5pt]{0.5pt}{11.5pt}\,{}_{B'} along the torus fibers so that \Omega(X_i, \cdot)= H^* \mathrm{d} f_i for i=1,\dots,k. Let m \in \mathcal{L}_b and Y \in T_mM, such that DH(Y) \in T_bB', then

Therefore, the connected components of the fiber of \mathcal{L} over b \in B'^\mathrm{reg} are integral submanifolds of the distribution V\mathcal{L}=\mathrm{span}(X_1, \dots, X_k). Hence the connected components of the fibers of \mathcal{L}_b \rightarrow B'^\mathrm{reg} are complex subtori. In particular, the subspace V_b\mathcal{L} \subset V_bM is rational with respect to the lattice \Lambda_b. By definition, this is equivalent to H_*T\mathcal{L}=TB' being rational with respect to the integral affine structure on B. Finally, by (Proof) we have V\mathcal{L} \subset (TB'^\mathrm{reg})^\perp \subset T^\vee B^\mathrm{reg} \cong VM. The first inclusion is an equality both being of rank k.

The original statement gives the isomorphism with values in H^0(\Sigma_{\underline{a}},\pi^*K_X^n). We have the isomorphism of sheaves K_{\Sigma_{\underline{a}}} \rightarrow \pi^*K_X^n given by \phi \mapsto \phi s_B\: \rule[-3.5pt]{0.5pt}{11.5pt}\,{}_U for \phi \in K_{\Sigma_{\underline{a}}}\: \rule[-3.5pt]{0.5pt}{11.5pt}\,{}_U. This is well defined by the following: Let (U,w) a coordinate disc centered at y \in \Sigma, such that \pi\colon U \rightarrow \pi(U), w \mapsto z=w^b with b \geq 1. Then s_B\: \rule[-3.5pt]{0.5pt}{11.5pt}\,{}_U= b w^{b-1} (\pi^* \mathrm{d} z)^{n-1}. Hence, \phi \cdot s_B(U)=f \mathrm{d} w \cdot b w^{b-1} (\pi^* \mathrm{d} z)^{n-1}=f (\pi^* \mathrm{d} z)^{n}, where we used that \pi^* \mathrm{d} z= b w^{b-1} \mathrm{d} w. Composing Baraglia’s isomorphism [Baraglia, Proposition 3.4] with the inverse of the above defines the asserted isomorphism t.=1

By Theorem 2.4, fibers of \mathcal{L} \rightarrow \mathcal{B}'^\mathrm{reg} are complex subtori in the Hitchin fiber. Hence, using a local section s\colon U \rightarrow \mathcal{L} on an open U \subset \mathcal{B}'^\mathrm{reg}, we can identify the fibers with an abelian subscheme of A \rightarrow \mathrm{Jac}(\Sigma/\mathcal{B}'^\mathrm{reg}). Condition (ii) is the family version of Theorem 2.4 reformulated by using the observations about the holomorphic symplectic form on \mathcal{M}_{\mathrm{GL}(n,\mathbb{C})} in the previous paragraph. Conversely, condition (ii) is equivalent to the restriction of the symplectic form of \mathcal{M}_{\mathrm{GL}(n,\mathbb{C})} to be zero on the tangent bundle to \mathcal{L}. Hence, \mathcal{L} is a Lagrangian.

Using the section s, we can identify the Hitchin system over \mathcal{B}'^\mathrm{reg} with the abelian scheme \mathrm{Prym}(\Sigma/\mathcal{B}'^\mathrm{reg}) \rightarrow \mathcal{B}'^\mathrm{reg}. By Theorem 2.4, the fiber \mathcal{L}_b for b \in \mathcal{B}'^\mathrm{reg} is a complex subtorus that by assumption contains s(b). Hence, \mathcal{L} defines an abelian subscheme A \subset \mathrm{Prym}(\Sigma/\mathcal{B}'^\mathrm{reg}). We obtain an exact sequence of abelian schemes over \mathcal{B}'^\mathrm{reg}

where the quotient Q is again an abelian scheme over \mathcal{B}'^\mathrm{reg}. Dually, we obtain an exact sequence of abelian schemes

We can define a section s'= \delta \circ s\colon \mathcal{B}'^\mathrm{reg} \rightarrow \mathrm{Hit}^{-1}_{G^{\rm L}}(\mathcal{B}'^\mathrm{reg}) \subset \mathcal{M}_{G^{\rm L}}. Acting by Q^\vee on s'\colon \mathcal{B}'^\mathrm{reg} \rightarrow \mathcal{M}_{G^{\rm L}} defines a submanifold I \subset \mathrm{Hit}^{-1}(\mathcal{B}'^\mathrm{reg}) \subset \mathcal{M}_{G^{\rm L}}. We define \mathcal{I}=\overline{I}. We want to show that this is subintegrable system with regular locus I, i.e., the holomorphic symplectic form restricts to a non-degenerate form on I and \mathrm{Hit}\: \rule[-3.5pt]{0.5pt}{11.5pt}\,{}_{\mathcal{I}}: \mathcal{I} \rightarrow \mathcal{B}'^\mathrm{reg} is an integrable system. First note that for all b \in \mathcal{B}'^\mathrm{reg}

Hence, the restricted Hitchin map will lead the correct number of commuting Hamiltonian functions and the fibers are complex tori by definition. To show that it forms a subintegrable system, it suffices to show that the tangent space of \mathcal{I} at the section s' is a symplectic vector space with the restriction of the symplectic form. The argument at a general point follows by a translation along the fibers of \mathcal{I}. We identify the tangent spaces T_{s(b)}\mathcal{M}_G \cong T_{s'(b)}\mathcal{M}_{G^{\rm L}} using \delta. Then the vertical tangent spaces to V\mathcal{L} and V\mathcal{I} are complementary by definition. Denote by H'_b=Ds'(T_b\mathcal{B}'^\mathrm{reg}) the tangent space to the image of the section s' at b. Tangent vectors in H'_b pair to zero with the vertical tangent vectors in V_b\mathcal{L} as \mathcal{L} is Lagrangian. On the other hand, the symplectic form on T_{s'(b)}\mathcal{M}_{G^{\rm L}} is non-degenerate. Hence, it induces an isomorphism (H'_b)^\vee \cong V_b\mathcal{I} or equivalently the symplectic form restricts to a non-degenerate symplectic form on T_{s'(b)} \mathcal{I}=H'_b \oplus V_b \mathcal{I}.

Let \underline{a} \in \mathcal{B}'^\mathrm{reg} and denote by P=\mathrm{Prym}(\Sigma_{\underline{a}}). As in the previous proof we use the section s to obtain the exact sequences of abelian varieties

We use the symmetric Fourier–Mukai transform introduced in [Schnell]. More precisely, we are going to use Proposition 1.6 therein. We denote by \mathcal{P} \rightarrow P \times P^\vee the Poincaré bundle. Then the symmetric Fourier–Mukai transform is defined by

where \Delta_P=\mathrm{Hom}(\cdot,\omega_P[\dim P]) is the Serre duality functor. First we have to show that \mathcal{O}_A is a GV-sheaf. By [PareschiPopa, Definition 3.1], this means that the support loci

have codimension \geq i for all i. Let \xi \in P^\vee. We have

It is zero for i > \dim A. For i \leq \dim A, it is non-zero if and only if \mathcal{P}_\xi^{-1} \: \rule[-3.5pt]{0.5pt}{11.5pt}\,{}_A is trivial. Hence, if and only if \xi \in Q^\vee. Therefore, the support loci satisfy the dimension condition and \iota_*\mathcal{O}_A is a GV sheaf on P. In particular, it is WIT – that is \operatorname{SFM}_P(\mathcal{O}_A) is a sheaf – by [PareschiPopa, Proposition 3.2].

The symmetric Fourier–Mukai transform on A has the property

where \mathbb{C}_0 is the skyscraper sheaf of length 1 at 0 \in P^\vee. Now [Schnell, Proposition 1.6] states that if a sheaf \mathcal{F} is GV on A and \iota_*\mathcal{F} is GV on P, then

By (?), this yields \operatorname{SFM}_P(\mathcal{O}_A)=\mathcal{O}_{Q^\vee}.

Tensoring with a line bundle preserves stability. Hence \mathcal{L} is well defined. The Hitchin fiber over \underline{a} \in \mathcal{B}'^\mathrm{reg} reflects the splitting of the differentials on \Sigma by the isomorphism

where \pi^*\mathrm{Jac}(X)[n] acts diagonally (see Mumford [Mumford]). In particular, we have an exact sequence of abelian schemes over \mathcal{B}'^\mathrm{reg}

As explained in Section ?, the holomorphic symplectic form on \mathrm{Hit}^{-1}\bigl(\mathcal{B}^{\mathrm{reg}}_{\mathrm{GL}(n,\mathbb{C})}\bigr) \subset \mathcal{M}_{\mathrm{GL}(n,\mathbb{C})} restricts to the Serre pairing between the tangent space of the base identified with H^0({\Sigma_{\underline{a}}},K_{\Sigma_{\underline{a}}}) by Proposition 3.1 and the tangent space to the fibers H^1({\Sigma_{\underline{a}}}, \mathcal{O}_{\Sigma_{\underline{a}}}). With respect to this pairing, we have

The subtorsor \mathcal{L} \subset \mathrm{Hit}^{-1}(\mathcal{B}'^\mathrm{reg}) that is defined by the action of \mathrm{Jac}(X) \times \mathcal{B}'^\mathrm{reg} on the Hitchin section has vertical tangent bundle H^1(X,\mathcal{O}_X) and the tangent space to the \mathrm{SL}(n,\mathbb{C}) Hitchin base are identified with H^0(\Sigma_{\underline{a}},K_{\Sigma_{\underline{a}}})^-. Therefore, \mathcal{L} is a visible Lagrangian.

By Theorem 3.4, the fiber-wise Fourier–Mukai transform of the structure sheaf of \mathcal{L} over \mathcal{B}'^\mathrm{reg} is supported on torsor over the abelian scheme that is dual to quotient in the exact sequence (Proof). That is the abelian scheme \mathrm{Prym}\bigl(\Sigma/ \mathcal{B}^{\mathrm{reg}}_{\mathrm{SL}(n,\mathbb{C})}\bigr). The closure of this locus is the moduli space of \mathrm{SL}(n,\mathbb{C})-Higgs bundles.

Let U \subset \mathcal{B}^{\mathrm{reg}}_{\mathrm{SL}(2,\mathbb{C})} open, such that there exist sections

such that \mathrm{Hit}^{-1}(U) \cap \mathcal{M}_{\mathrm{SL}(2,\mathbb{R})} = \bigsqcup_{i=1}^{6g-6} s_i(U). Then the orbit of each s_i(U) under tensoring with line bundles in \mathrm{Jac}(X) is Lagrangian by the previous proof. Hence, \mathcal{L} defines a visible Lagrangians.

We identify the Hitchin fibers with the abelian scheme \mathrm{Jac}(\Sigma) using the Hitchin section s_1. Fix \underline{a} \in U and define l_i = (s_1-s_i)(\underline{a}) \in \mathrm{Jac}(\Sigma_{\underline{a}}). Recall the exact sequence of abelian schemes (Proof). We have

where t_{l_i} denotes the translation by l_i on \mathrm{Jac}(\Sigma_{\underline{a}}). Let J=\mathrm{Jac}(\Sigma_{\underline{a}}) and \mathcal{P} \rightarrow J \times J the Poincaré bundle. Using the relation between tensor product and translations under Fourier–Mukai transform (see [Schnell]), we obtain

The right hand side defines a flat line bundle on \mathrm{Prym}(\Sigma_{\underline{a}}). The Fourier–Mukai of the structure sheaf \mathcal{O}_{\mathcal{L}_{\underline{a}}} is the direct sum of all these flat line bundles.

Starting from a pillowcase cover as above, we have the following diagram:

and want to show that there exist the dashed arrow p\colon \Sigma \rightarrow E, such that the diagram commutes. Away from the singularities of the flat structure associated with (X,q) a point on \Sigma corresponds to a choice of a (local) square of q. This corresponds to a choice of a local square root of \eta under the covering map to \mathbb{P}^1. Hence, there is a induced map p\colon \Sigma\setminus Z(\lambda) \rightarrow E \setminus Z(\omega) such that p^*\omega=\lambda. The map p uniquely extends to a map of Riemann surfaces \Sigma \rightarrow E. A zero x \in X of q of order k\geq 0 corresponds to a (k+2):1-branch point of \check{p}. Hence, the map can be extended to p\colon \Sigma \rightarrow E by gluing in a k:1 ramification point, if k is odd and a \frac{k}{2}:1 ramification point, if k is even, at y \in \pi^{-1}(x). By construction the relation of the differentials persists under this extension. Taking the quotient of E by two-torsion points we obtain a map (\Sigma,\lambda) \rightarrow E/E[2], branched over one point such that \lambda is the pullback of a differential on E. Hence, it is a parallelogram-tiled surface.

For the converse, we start with the configuration

The differential \lambda is anti-symmetric with respect to an involution J. In particular, J sends singularities of the flat structure to singularities of the same type, saddle connections to saddle connections and hence squares to squares. Hence, it descends to the elliptic curve (E,\omega) to the reflection in the two torsion points of E. The quotient is the pillowcase surface as illustrated in Figure ?. In particular, there is an induced map \check{p}\colon \Sigma/J=X \rightarrow \mathbb{P}^1, such that q=\check{p}^*\eta.

Assume that we know a single multifold uniform pillowcase cover X with simple zeros only of some genus g \geq 2. Let q_1, \dots, q_n be the corresponding quadratic differentials. Then we can obtain examples in infinitely many genera by taking unramified coverings f\colon \widehat X \rightarrow X in different degrees and the differentials \widehat q_i = f^* q_i.

We will provide explicit examples in the following.

The curves \Sigma_1 and \Sigma_2 are covers of \mathbb{P}^1 via \Sigma_i \rightarrow X \xrightarrow{f} \mathbb{P}^1, branched over three points. Both covers \Sigma_i \rightarrow \mathbb{P}^1 have a monodromy representation with elements in the symmetric group S_{2|G|}. The covers \Sigma_1 and \Sigma_2 are isomorphic if and only if the elements of the monodromy representation are conjugated in S_{2|G|}, which is easily checked not to be the case.

First, assume there exists a visible Lagrangian over \mathbb{C}\underline{a}. Let (\Sigma,\lambda) be the spectral curve to \underline{a}. Then, by Proposition 3.2, there is an exact sequence of abelian varieties

Here A is of codimension 1 and E is an elliptic curve. Under the assumption on \underline{a}, the spectral curve is \mathbb{Z}_n-Galois and s_B=\lambda^{n-1} (up to a constant). Hence, the map t of Proposition 3.1 becomes

In particular, the tangent vector X_n=a_n is mapped to \frac{\pi^*a_n}{\lambda^{n-1}}=\lambda. Hence, the differential of the map \psi can explicitly be written as

Therefore, \psi is given by D=\sum a_i y_i \mapsto \sum a_i\int_{y_0}^{y_i} \lambda up to the choice of a point y_0 \in \Sigma. Denote by \sigma a generator of the \mathbb{Z}_n action on \Sigma. We want show that the composition \psi \circ \mathrm{AP} with the Abel–Prym map

is a branched covering p\colon \Sigma \rightarrow E. Clearly, \mathrm{AP} identifies all ramification points of \pi\colon \Sigma \rightarrow X. If \Sigma is not hyperelliptic, it is easy to see that these are the only points on \Sigma that are identified. The spectral curve \Sigma is never hyperelliptic by [Baraglia, Lemma 4.1]. By definition, \sigma^*\lambda=\xi_n \lambda for some primitive n-th root of unity \xi_n. This implies that the differential of the composition \psi \circ \mathrm{AP} is (1-\xi) \lambda. Hence, the differential of the composition is injective away from the ramification points of \Sigma \rightarrow X. In particular, p=\psi \circ \mathrm{AP}\colon \Sigma \rightarrow E is a proper holomorphic map ramified at Z(\lambda) over 0 \in E. Furthermore, \lambda considered as abelian differential has an order n zero at each branch point and hence the points in Z(\lambda) are n+1:1 ramifications. In particular, the pullback of an abelian differential \omega on E has the same divisor as \lambda and we can find a specific \omega with p^*\omega=\lambda. Therefore, (\Sigma,\lambda) is parallelogram-tiled.

For the converse, let \underline{a}=(0,\dots,0,a_n) \in \mathcal{B}_{\mathrm{SL}(n,\mathbb{C})}^\mathrm{reg}, such that (\Sigma,\lambda) is parallelogram-tiled, i.e., there is a covering p\colon \Sigma \rightarrow E, such that \lambda=p^*\omega for some abelian differential \omega on E. The covering p induces a Norm map

The restriction to \mathrm{Prym}(\Sigma) defines the desired map

Let D=(n-1)y - \sigma y - \dots - \sigma^{n-1}y. Then D \in \mathrm{Prym}(\Sigma) and \mathrm{Nm}_E(D)= n \mathrm{Nm}_E(y). Hence, \psi surjects onto E being a divisible group. Its differential is the map (Proof). As \Sigma does not change, when multiplying q with a scalar, we can define an abelian subscheme \mathrm{ker}(\psi) \subset \mathrm{Prym}(\Sigma/\mathbb{C}^\times q) that satisfies the criterion of Proposition 3.2. To obtain a visible Lagrangian over \mathbb{C} q, we can act with the abelian scheme \mathrm{ker}(\psi) on any section of \mathrm{Hit} \: \rule[-3.5pt]{0.5pt}{11.5pt}\,{}_{\mathbb{C} \underline{a}}. To obtain a concrete example, we may choose the Hitchin section.

This is immediate from Theorem 6.1 and Lemma 4.3.

Hausel’s toy model is constructed as an elliptic surface as follows. Take \bigl(\mathbb{P}^1,p_1,p_2,p_3,p_4\bigr) and consider the elliptic curve E \rightarrow \mathbb{P}^1, which is the canonical cover of the pillowcase with involution \tau. Let further \tau' be the involution -1\colon \mathbb{C} \rightarrow \mathbb{C}. Consider M=E \times \mathbb{C}/(\tau \times \tau'). This orbifold has 4 \mathbb{Z}_2-points \widehat{p}_i \times 0. The projection to the second factor induces the map M \rightarrow \mathbb{C}, (x,y) \mapsto y^2 with generic fiber E. Blowing up the 4 orbifold points, we obtain a smooth surface – the toy model \mathcal{M}_{\mathrm{toy}}.

The subintegrable system \mathcal{I}_q associated to \mathcal{B}'=\mathbb{C} q for a pillowcase cover (X,q) is an elliptic fibration over \mathbb{C}^\times q \subset \mathcal{B}'. The modulus of the elliptic curve is constant and determined by the four points p_1,\dots,p_4 on the pillowcase. As for Hausel’s toy model, the monodromy around 0 \in \mathbb{C} q is given by -1. In fact, if we scale our quadratic differential with e^{i\phi}, the abelian differential is multiplied by e^{i\frac12 \phi}. This defines an isomorphism between M\setminus{\{y=0\}} and \mathcal{I}_q \: \rule[-3.5pt]{0.5pt}{11.5pt}\,{}_{\mathbb{C}^\times q}.

Recall the square of coverings of Figure ?. The morphism \Theta is given by a Hecke modified pullback along \check{p}. Let (\mathcal{E},\Phi) \in \mathcal{M}_{\underline{\alpha}}. First define

This is a meromorphic Higgs bundle on X with \mathrm{tr}(\Phi')=0. By assumption of q having simple zeros only all points in \check{p}^{-1}(D) are ramification points of \check{p}, which are 2:1 or 3:1. Let R_i \in \mathrm{Div}^+(X) be the divisor that has weight 1 at the branch points that are (i+1):1 and R=R_1 +R_2. The pullback of the quasi-parabolic structure defines a quasi-parabolic structure on \mathcal{E}' at R given by \mathcal{E}'_{x,1}:=\check{p}^*\mathcal{E}_{\check{p}x,1} \subset \mathcal{E}_{x}. Now we define a Hecke modification

Then the map \Theta is defined as \Theta\colon (\mathcal{E},\Phi) \mapsto (\hat{\mathcal{E}},\hat{\Phi}).

Well-definedness: First, we have to show that \Theta is well defined. We have

and \mathrm{div}(q)=R_2. Hence, the determinant of E' computes to

Therefore, the determinant of \hat{\mathcal{E}} is \det\bigl(\hat{\mathcal{E}}\bigr)=\det(\mathcal{E}')(-R_1-R_2)=\mathcal{O}_X.

To show that the Higgs field \hat{\Phi} is holomorphic we do a local computation. Locally at each y \in D, we can find a frame s_1, s_2 of \mathcal{E} adapted to the parabolic structure such that

and \mathcal{E}_{y,1}=\langle s_1\: \rule[-3.5pt]{0.5pt}{11.5pt}\,{}_{y} \rangle. Hence, at all ramification points x \in \check{p}^{-1}D the pullback has the form

where w is a coordinate centered at x, such that \check{p}\colon w \mapsto z=w^k. By assumption, k=2,3. The pullback quasi-parabolic structure is given by \mathcal{E}'_{x,1}=\langle \check{p}^*s_1 \: \rule[-3.5pt]{0.5pt}{11.5pt}\,{}_y\rangle. Now, it is an easy computation to see that the Hecke modification modifies the Higgs field to

Hence, indeed (\hat{\mathcal{E}},\hat{\Phi}) defines a \mathrm{SL}(2,\mathbb{C})-Higgs bundle. Its polystability will follow from the existence of a solution to Hitchin’s equation at the end of the proof.

Hitchin sections: Now, we want to apply this morphism to a point in the Hitchin section. We identify the Hitchin base of \mathcal{M}_{\underline{\alpha}} as \{ c \eta \mid c \in \mathbb{C}\} for a fixed quadratic differential \eta with simple poles at D. Denote by q=\check{p}^*\eta its pullback. Then a point (\mathcal{E},\Phi) in the Hitchin section of \mathcal{M}_{\underline{\alpha}} is given by

After pullback and tensoring, we obtain

Here \check{p}^*1 \in \check{p}^*\mathcal{O}(2) \otimes \check{p}^*K_{\mathbb{P}^1} has a zero of order 1 at each 2:1 ramification point and a zero of order 2 at each 3:1 ramification point. The pullback quasi-parabolic structure at \check{p}^{-1}D is given by the second coordinate with respect to the splitting. Hence, the Hecke modification yields

as asserted.

Compatibility with pullback by p: We show that this map extends the spectral correspondence on the regular locus. First we showed above that the Hitchin section of \mathcal{M}_{\underline{\alpha}} is mapped on the Hitchin section of \mathcal{M}_{\mathrm{SL}(2,\mathbb{C})}(X). It is easy to see that the eigen-line bundle on E of the Hitchin section of \mathcal{M}_{\underline{\alpha}} is \check{\pi}^*\mathcal{O}(-3). Similarly, the eigen-line bundle on \Sigma of the Hitchin section of \mathcal{M}_{\mathrm{SL}(2,\mathbb{C})}(X) is \pi^*K_X^{-\frac12}. We use these two Hitchin sections to identify the fibers with the corresponding abelian varieties. Then (\mathcal{E},\Phi) \in \mathcal{M}_{\underline{\alpha}} corresponds to an element L_1 \in \mathrm{Jac}(E). Similarly, \bigl(\hat{\mathcal{E}},\hat{\Phi}\bigr) corresponds to an element L_2 \in \mathrm{Prym}(\Sigma_q). We need to show that L_2=p^*L_1. From the spectral correspondence, we have an exact sequence

Tensoring with \mathcal{O}(3) and pulling back, we obtain

Here we used the commutativity of diagram in Figure ?. Finally, twisting by K_X^{-\frac12} we see that the line bundle associated to \pi^*(\mathcal{E}',\Phi') through the spectral correspondence is p^*L_1. The (pullback of the) Hecke modification \hat{\mathcal{E}} \rightarrow \mathcal{E} can potentially change this line bundle by twisting it with a divisor supported at \pi^{-1}R. However, on the regular locus of \mathcal{M}_{\underline{\alpha}} the quasi-parabolic structure is uniquely determined through the Higgs field at each p \in D (see [FMSW, Proposition 8.1]). Hence, we can compute the eigen-line bundle and the quasi-parabolic structure of \pi^*(\mathcal{E}',\Phi') with respect to the pullback of the local frame of (?). Then it is easy to see that the eigen-line bundle p^*L_1 \otimes K_x^{-\frac12} descends to a subbundle of \pi^*(\hat{\mathcal{E}},\hat{\Phi}). Hence p^*L_1=L_2.

Solutions to Hitchin equation: Finally, we will show that a hermitian metric h on {(\mathcal{E},\Phi) \in \mathcal{M}_{\underline{\alpha}}} that solves the Hitchin equation is transformed to a solution to the Hitchin equation for \bigl(\hat{\mathcal{E}},\hat{\Phi}\bigr) \in \mathcal{M}_{\mathrm{SL}(2,\mathbb{C})}(X). First there is a section of \mathcal{O}(4) with divisors D. Promoting this section to have norm 1 defines a singular hermitian metric on \mathcal{O}(4) locally given by \vert z \vert^{-2} at y \in D. This induces a singular hermitian metric on \mathcal{O}(1) and hence on \mathcal{O}(3). The latter is given by \vert z \vert ^{-\frac32} at y \in D and will be denoted by h_{\mathcal{O}(3)}. Similarly, q \in H^0\bigl(X,K_X^2\bigr) induces a singular hermitian metric h_{-\frac12 K_X} on K_X^{-\frac 12} that is smooth away from Z(q) and given by \vert w \vert^{\frac12} locally at x \in Z(q). It is easy to see that singular hermitian metrics defined in this way are automatically flat.

We need to extend the morphism \Theta to hermitian Higgs bundles (\mathcal{E},\Phi,h). In the first step, we use the singular hermitian metrics defined above to obtain a hermitian metric on \mathcal{E}' by

This hermitian metric is holomorphic on X\setminus R. The hermitian metric h' pulls back to a hermitian metric \hat{h} on \mathcal{E} through the Hecke modification \hat{\mathcal{E}} \rightarrow \mathcal{E} a priori holomorphic only on X \setminus R.

Now, we start with a polystable Higgs bundle (\mathcal{E},\Phi) \in \mathcal{M}_{\alpha} and let h be a solution to Hitchin’s equation. By the flatness of the hermitian metrics h_{\mathcal{O}(3)} and h_{-\frac12 K_X}, the resulting hermitian metric h' on \mathcal{E} will still be a solution to Hitchin’s equation wherever it is smooth. By definition, the Hecke modification \hat{\mathcal{E}} \rightarrow \mathcal{E} is an isomorphism on X \setminus R and hence the induced metric \hat{h} is a solution to Hitchin’s equation on this locus. To show that it defines a solution to Hitchin’s equation on X, we are left with showing that it extends smoothly over R.

To do so, we compute the local description of \hat{h} at x \in R. The metric h is adapted to the parabolic structure. Hence, at y \in D we can find a local frame s_1, s_2 of \mathcal{E} such that firstly

secondly the Higgs field is given by (?) and thirdly the quasi-parabolic structure is the ascending flag \langle s_1 \rangle \subset \langle s_1,s_2 \rangle. We have to consider two cases depending on the ramification index of the fiber over y \in D. We will only give the details for y \in D of ramification index 2, i.e., \check{p}^{-1}y is made up from 3:1 ramification points.

In this case, the compatible parabolic weight are \underline{\alpha}_y=\bigl(\frac34, \frac14\bigr), so that h=\mathrm{diag}\bigl( \vert z \vert^{\frac{4}{3}}, \vert z \vert^{\frac{2}{3}}\bigr). Tensoring with (\mathcal{O}(3),h_{\mathcal{O}(3)}), pulling back and then tensoring with \bigl(K_X^{-\frac12},h_{-\frac12 K}\bigr), we obtain a local description for h' at x \in Z(q) with respect to the induced frame

Here w is local coordinate at x \in \check{p}^{-1}y, such that \check{p}\colon w \mapsto w^3=z. Finally, with respect to the frame s_1, s_2 the Hecke modification \hat{\mathcal{E}} \rightarrow \mathcal{E} is given by \mathrm{diag}( 1, w) and hence the induced metric on \hat{\mathcal{E}} is indeed smooth at x \in R of ramification index 2. The case of ramification index 1 follows along the same lines. Hence, \hat{h} defines a smooth solution to the Hitchin equation for the Higgs bundle \bigl(\hat{\mathcal{E}},\hat{\Phi}\bigr).

In particular, for a polystable Higgs bundle (\mathcal{E},\Phi) \in \mathcal{M}_{\underline{\alpha}} the image \Theta(\mathcal{E},\Phi) is polystable and hence indeed \Theta defines a map of moduli spaces

This finishes the proof.

By Theorem 6.4, \Theta maps solutions to the Hitchin equation on \bigl(\mathbb{P}^1,D\bigr) to solutions to the Hitchin equation on X. Hence, it is holomorphic not only with respect to the holomorphic structure I, but also with respect to the holomorphic structure J and K on the moduli spaces \mathcal{M}_{\underline{\alpha}}\bigl(\mathbb{P}^1,D\bigr) and \mathcal{M}_{\mathrm{SL}(2,\mathbb{C})}(X). In particular, its image is a hyperholomorphic subvariety. However, Theorem 6.4 (i) and (ii) shows that \Theta restricts to an isomorphism from the regular locus of \mathcal{M}_{\underline{\alpha}} to the torsor \mathcal{I}' \subset \mathcal{M}_{\mathrm{SL}(2,\mathbb{C})}(X) obtained by acting with the abelian scheme E \subset \mathrm{Prym}(\Sigma/\mathbb{C}^\times q ) over \mathbb{C}^\times q on the Hitchin section. Hence, \overline{\mathcal{I}}' the image of \Theta is a hyperholomorphic subvariety. In particular, its image \mathcal{I}_q under the quotient map \delta\colon \mathcal{M}_{\mathrm{SL}(2,\mathbb{C})}(X) \rightarrow \mathcal{M}_{\mathrm{PGL}(2,\mathbb{C})}(X) is hyperholomorphic.